Professional Documents
Culture Documents
Edited by
R. Gous
University of KwaZulu-Natal
Pietermaritzburg
South Africa
T. Morris
University of Reading
Reading
UK
and
C. Fisher
Consultant
Midlothian
Scotland
CABI is a trading name of CAB International
A catalogue record for this book is available from the British Library,
London, UK.
A catalogue record for this book is available from the Library of Congress,
Washington, DC.
ISBN-10: 1–84593–070–3
ISBN-13: 978–1-84593–070–7
v
vi Contents
E.T. Berhe, Animal and Poultry Science, School of Agricultural Sciences and
Agribusiness, University of KwaZulu-Natal, Private Bag X01, Scottsville 3209,
South Africa.
O.A. Blanco, Animal and Poultry Science, School of Agricultural Sciences and
Agribusiness, University of KwaZulu-Natal, Private Bag X01, Scottsville 3209,
South Africa.
J. Dijkstra, Animal Nutrition Group, Wageningen Institute of Animal Sciences,
Wageningen University, Marijkeweg 40, 6709 PG Wageningen, The
Netherlands.
A.B. Doeschl-Wilson, Sygen International, Scottish Agricultural College, Bush
Estates, Penicuik, Edinburgh, EH26 0PH, UK.
G.C. Emmans, Animal Nutrition and Health Department, Scottish Agricultural
College, West Mains Road, Edinburgh, EH9 3JG, UK.
N.S. Ferguson, Maple Leaf Foods Agresearch, 150 Research Lane, Guelph,
Ontario, Canada, N1G 4T2.
J. France, Centre for Nutrition Modelling, Department of Animal and Poultry
Science, University of Guelph, Guelph, Ontario, Canada, N1G 2W1.
D.M. Green, University of Oxford, Department of Zoology, South Parks Road,
Oxford, OX1 3PS, UK.
R.M. Gous, Animal and Poultry Science, School of Agricultural Sciences and
Agribusiness, University of KwaZulu-Natal, Private Bag X01, Scottsville 3209,
South Africa.
S.A. Johnston, Animal and Poultry Science, School of Agricultural Sciences and
Agribusiness, University of KwaZulu-Natal, Private Bag X01, Scottsville 3209,
South Africa.
B.P. Kinghorn, Sygen International, Scottish Agricultural College, Bush Estates,
Penicuik, Edinburgh, EH26 0PH, UK.
P.W. Knap, PIC International Group, Ratsteich 31, D-24837 Schleswig, Germany;
and Sygen International, Scottish Agricultural College, Bush Estates, Penicuik,
Edinburgh, EH26 0PH, UK.
vii
viii Contributors
ix
x Preface
systems is an area that is developing quite rapidly at this time. On the more
general questions of animal modelling, many different philosophies and
approaches are viable and the reader will find a reflection of this diversity
in these pages. The day-to-day application of modelling in management
decision-making is still some way off but the progress towards this ideal is
reflected in this book.
Professor Trevor Morris led the way in the application of quantitative
methods in poultry production and nutrition. Mainly using statistical
methods but always with a clear view of the underlying mechanisms, for
many years he showed the benefits of combining the results of different
experiments into a set of simple and applicable quantitative rules. Whilst
this is some way from modern, computer-based mechanistic modelling,
these earlier ideas undoubtedly showed the way and, in particular, trained
and motivated many students who later became modellers. Professor
Morris is now retired and this volume is warmly dedicated to him in
recognition and appreciation of his work and contribution. It will be an
appropriate testament to his work if this volume encourages some young
animal scientists to see how the application of modelling techniques and
ideas can enhance their own work.
R.M. Gous
Acknowledgements
Aviagen
Degussa
EFG Software (Natal)
Elsevier Publishing
Maple Leaf Foods
University of KwaZulu-Natal
World’s Poultry Science Association
xi
This page intentionally left blank
1 An Introduction to Modelling in
the Animal Sciences
T.R. MORRIS
School of Agriculture, Policy and Development, The University of Reading,
New Agriculture Building, PO Box 237, Reading, RG6 6AR, UK
trevor@trevormorris.wanadoo.co.uk
Scientists and engineers use models to represent parts of what they regard
as the real world; to help them to convey to others an understanding of the
way in which things work and, sometimes, to help them to make
predictions about the consequences of alternative courses of action.
Some models are pictorial and none the less useful for that. We could
draw a diagram of the digestive system of a pig that would be helpful in
explaining how the animal converts its food into components that can be
absorbed, leaving a residue to be excreted. Note that the diagram does not
need to look much like the real guts of a pig and, indeed, a colour
photograph of an alimentary tract, although more ‘accurate’ than a simple
line drawing, is actually less suitable for our purpose. A single photograph
would not reveal the teeth, the salivary glands, the bile duct, the pancreas
and the hepatic portal vein. For similar reasons, an accurate scale map of
the London underground railway system, showing all its tracks, stations and
platforms might be valuable to a maintenance engineer but is not helpful to
the visitor trying to find his or her way around London. The map which
does appear on the walls of London underground stations is a much
simplified diagram, showing the relationships between stations on a given
line (but not their real geographical locations) and the interconnections
between lines, which are distinguished by the use of colour codes. This
model is the work of an electrical engineer who knew how to draw a circuit
diagram and applied his skill to the problem of making the planning of a
journey as simple as possible. What these examples tell us is that models
have to be sufficient for their purpose and, therefore, that we must define
the purpose carefully before setting out to construct a model.
A diagram of a pig’s digestive system and a map of an underground
system are examples of models which represent something in the real
world, but they do not involve equations. Wordsworth (1798) wrote that
‘poetry is emotion recollected in tranquillity’ and we might argue that
obeyed everywhere. However, there are many cases where we are not yet
in a position to build a causal model from fundamental components and
we must therefore, for the time being, rely on empiricism in those cases.
Two examples will illustrate this point. We know that changes in photo-
period have marked effects on the rate of sexual development in pullets
and we can trace changes in the concentration of gonadotrophic hormones
in the blood, following an increase or decrease in photoperiod. We are
confident that the brain responds to photoperiod by altering the flow of a
releasing hormone from the hypothalamus to the pituitary gland, which in
turn adjusts the flow of gonadotrophins to the ovary. However, this
knowledge does not put us in a position to write equations representing
the effect of a stated series of photoperiods during rearing upon the age at
which the pullet will lay her first egg. We can, however, provide quite
robust empirical equations to predict age at first egg for any specified
pattern of photoperiod applied during rearing (Lewis et al., 2002, 2003). It
is perhaps going too far to say that this particular gap between theory and
empiricism will never be bridged, but it seems unlikely that a full
mechanistic model would have any better predictive capability than the
present empirical one; and so it may be that further effort to quantify the
mechanism is not justified. A second example is the response to added
copper in a diet. A small amount of copper (about 6 mg/kg diet) is needed
for metabolic purposes, but this level is supplied by almost all natural diets.
If copper sulphate is added to raise copper levels to 50–100 mg/kg, growth
rate of pigs and baby chicks is enhanced. This effect is attributed to
modification of the gut flora in a manner that is beneficial to the host
animal. At higher concentrations, copper begins to be toxic and growth is
depressed. All these effects, copper deficiency at very low levels, copper as
a growth promoter in the medium range and copper toxicity at high doses,
can be demonstrated and quantified by appropriate trials, but this does not
lead to a theory connecting the cause to the effects in a mechanistic model.
On the other hand, we could produce a well researched response curve,
which could be reliably used to predict the effect of adding copper
sulphate to pig and poultry diets.
There are many more examples available where we have no immediate
prospect of being able to develop a good mechanistic model, even though
we believe that we understand a good deal about the mechanisms involved.
Empirical modelling is therefore not to be despised if it is the best available
tool for solving a particular problem. What is not acceptable is the use of
empirical modelling where others have already developed a mechanistic
model capable of resolving the particular question being approached
empirically. I believe that the modelling of growth and development in
pigs and poultry has now reached a sufficiently advanced state that
empirical models in this area can no longer be justified.
Introduction to Modelling in Animal Sciences 5
References
Armsby, H.P. (1917) Nutrition of Farm Animals. Macmillan, New York.
Baldwin, R.L. and Hanigan, M.D. (1990) Biological and physical systems: animal
sciences. In: Jones, J.G.W. and Street, P.R. (eds) Systems Theory Applied to
Agriculture and the Food Chain. Elsevier Science, Barking, UK, pp. 1–21.
Brody, S. (1945) Bioenergetics and Growth. Hafner, New York.
Chiba, L.I. (1999) Feeding systems for pigs. In: Theodoru, M.K. and France, J.
(eds) Feeding Systems and Feed Evaluation Models. CAB International,
Wallingford, UK, pp. 181–209.
Fawcett, R.H. (1973) Towards a dynamic production function. Journal of
Agricultural Economics 20, 543–549.
Fisher, C., Morris, T.R. and Jennings, R.C. (1973) A model for the description and
prediction of the responses of laying hens to amino acid intake. British Poultry
Science 14, 469–484.
Fisher, C. and Wilson, B.M. (1974) Response to dietary energy concentration by
growing chickens. In: Morris, T.R. and Freeman, B.M. (eds) Energy
Requirements of Poultry. Constable, Edinburgh, UK, pp. 151–184.
Kellner, O. (1909) The Scientific Feeding of Animals. Translated by Goodwin, W.
Duckworth, London.
Leeson, S. and Summers, J.D. (1999) Feeding systems for poultry. In: Theodoru,
M.K. and France, J. (eds) Feeding Systems and Feed Evaluation Models. CAB
International, Wallingford, UK, pp. 211–237.
Lewis, P.D., Morris, T.R. and Perry, G.C. (2002) A model for predicting the age at
sexual maturity for growing pullets of layer strains given a single change in
photoperiod. Journal of Agricultural Science, Cambridge 138, 441–448.
Lewis, P.D., Morris, T.R. and Perry, G.C. (2003) Effect of two opposing changes in
photoperiod upon age at first egg in layer-hybrid pullets. Journal of Agricultural
Science, Cambridge 140, 373–379.
McDonald, M.W. and Morris, T.R. (1985) Quantitative review of amino acid intakes
for young laying pullets. British Poultry Science 26, 253–264.
Morris, T.R. (1968) The effect of dietary energy level on the voluntary calorie
intake of laying birds. British Poultry Science 9, 285–295.
Morris, T.R. (1969) Nutrient density and the laying hen. In: Swan, H. and Lewis,
D. (eds) Proceedings of the Third Nutrition Conference for Feed Manufacturers.
University of Nottingham, Nottingham, UK, pp. 103–114.
Morris, T.R. (1983) The interpretation of response data from animal feeding trials.
In: Haresign, W. (ed.) Recent Advances in Animal Nutrition – 1983. Butterworths,
London, pp. 2–23.
Thornley, J.H.M. and France, J. (1984) Role of modelling in animal research and
extension work. In: Baldwin, R.L. and Bywater, A.C. (eds) Modelling Ruminant
Nutrition, Digestion and Metabolism; Proceedings of Second International Workshop.
University of California Press, Davis, California, pp. 4–9.
Whittemore, C.T. and Fawcett, R.H. (1974) Model responses of the growing pig to
the dietary intake of energy and protein. Animal Production 19, 221–231.
Whittemore, C.T. and Fawcett, R.H. (1976) Theoretical aspects of a flexible model
to simulate protein and lipid growth in pigs. Animal Production 22, 87–96.
Wordsworth, W. (1798) Preface. In: Wordsworth, W. and Coleridge, S.T. (eds)
Lyrical Ballads. Longmans, London.
2 Scientific Progress and
Mathematical Modelling:
Different Approaches to
Modelling Animal Systems
J. FRANCE1 AND J. DIJKSTRA2
1Centre for Nutrition Modelling, Department of Animal and Poultry Science,
University of Guelph, Guelph, Ontario, Canada, N1G 2W1;
2Animal Nutrition Group, Wageningen Institute of Animal Sciences,
Introduction
A general understanding of science influences the scientific questions that
are asked, the choice of problems for scientific investigation and also how
these are attacked. A more widespread understanding of this topic might
enable a greater contribution to be made for the same effort. This chapter
attempts to describe what science is, how it progresses, the role and practice
of mathematical modelling and different approaches to modelling animal
systems. This is done with particular reference to animal science, with
examples from poultry and pigs. It represents, of course, a personal view.
Organizational hierarchy
Biology, including pig and poultry science, is notable for its many
organizational levels. It is the different levels of organization that give rise
to the rich diversity of the biological world. For animal science, a typical
Scientific Progress and Mathematical Modelling 9
Teleonomic modelling
modelling, though this role might expand. It has hardly been applied to
problems in animal physiology though it has found some application in
plant and crop modelling (Thornley and Johnson, 1989).
Empirical modelling
Empirical models are models in which experimental data are used directly to
quantify relationships, and are based at a single level (e.g. the whole animal)
in the organizational hierarchy discussed above. Empirical modelling is
concerned with using models to describe data by accounting for inherent
variation in the data. Thus, an empirical model sets out principally to
describe, and is based on observation and experiment and not necessarily on
any preconceived biological theory. The approach derives from the
philosophy of empiricism and adheres to the methodology of statistics.
Empirical models are often curve-fitting exercises. As an example,
consider modelling voluntary feed intake in a growing pig. An empirical
approach to this problem would be to take a data set and fit a linear
multiple regression equation, possibly relating intake to liveweight,
liveweight gain and some measure of diet quality.
We note that level i behaviour (intake) is described in terms of level i
attributes (liveweight, liveweight gain and diet quality). As this type of
model is principally concerned with prediction, direct biological meaning
usually cannot be ascribed to the equation parameters and the model
suggests little about the mechanisms of voluntary feed intake. If the model
fits the data well, the equation could be extremely useful though it is
specific to the particular conditions under which the data were obtained,
and so the range of its predictive ability is limited.
Mechanistic modelling
Model evaluation
Mathematical Approaches
At this point in our discussion, it is important to give a correct picture of the
nature of mathematics. Mathematics is often seen as a kind of tool, as the
handmaiden of science and technology. This view fails to acknowledge or
reflect the potential role of mathematics in science and technology as an
integral part of the basic logic underlying the previewing and developmental
imagination which drives these vital disciplines. The use of the word tool to
describe mathematics is, we submit, pejorative. Tools operate on materials in a
coercive way by cutting, piercing, smashing, etc. Mathematics is used in a
12 J. France and J. Dijkstra
Regression paradigm
V(Y X1 , X2 , ..., X q ) = σ 2 .
as the regression equation. The parameters 1 ,2 , ..., q are the partial
regression coefficients.
It is convenient to write Eqn 2.2 in the form:
˜ + ( X − x ) + ( X − x ) + ... + ( X − x ),
E(Y X1 , X2 , ..., X q ) = 0 1 1 1 2 2 2 q q q
where the –x i ’s are computed from the n observations (y1, x11, x21, …, xq1),
n
(y2, x12, x22, …, xq2), …, (yn, x1n, x2n, …, xqn) as, e.g. x1 = ∑ x1 j n. The sum
j=1
of squares of the yj’s from their expectations is therefore:
∑ [ yj ],
n 2
S(˜0 , 1 , ..., q ) = ˜ − ( x − x ) − ( x − x ) − ... − ( x − x )
− 0 1 1j 1 2 2j 2 q qj q
j =1
˜
and the least squares estimates of the parameters β0 , β1 , β2 , ..., β q are the
solutions of the normal equations:
∂S ∂S ∂S ∂S
= = = ... = = 0.
˜
∂0 ∂1 ∂2 ∂ q
Parameter β0 can be determined knowing β̃0. A linear multiple regression
model is linear in the parameters 1 , 2 , ..., q. A non-linear model that
can be transformed into a form which is linear in the parameters (e.g. by
taking natural logarithms) is said to be intrinsically linear. Draper and
Smith (1998) is recommended reading on regression methods.
Many of the models applied in pig and poultry science are systems of
linked regression equations, e.g. current feed evaluation systems.
LP paradigm
where Z is the objective function and the Xj’s are decision variables. The cj’s,
aij’s and bi’s (bi 0) are generally referred to as costs, technological
14 J. France and J. Dijkstra
Rate:state formalism
Differential equations are central to the sciences and act as the cornerstone of
applied mathematics. It is often claimed that Sir Isaac Newton’s great
discovery was that they provide the key to the ‘system of the world’. They
arise within biology in the construction of dynamic, deterministic, mechanistic
models. There is a mathematically standard way of representing such models
called the rate:state formalism. The system under investigation is defined at
time t by q state variables: X1, X2, …, Xq. These variables represent properties
or attributes of the system, such as visceral protein mass, quantity of substrate,
etc. The model then comprises q first order differential equations which
describe how the state variables change with time:
dX i
= f i ( X1 , X2 , ..., X q; S); i = 1, 2, ..., q, (2.3)
dt
where S denotes a set of parameters, and the function fi gives the rate of
change of the state variable Xi.
The function fi comprises terms which represent the rates of processes
(with dimensions of state variable per unit time), and these rates can be
calculated from the values of the state variables alone, with of course the
values of any parameters and constants. In this type of mathematical
modelling, the differential equations are constructed by direct application of
scientific law based on the Cartesian doctrine of causal determinism (e.g. the
law of mass conservation, the first law of thermodynamics) or by application
of a continuity equation derived from more fundamental scientific laws. The
Scientific Progress and Mathematical Modelling 15
where Xi(0) denotes the initial (zero time) value of Xi. Integral equations
arise, not only as the converse of differential equations, but also in their
own right. For example, the response of a system sometimes depends not
just on the state of the system per se but also on the form of the input.
Input P and output U might then be related by the convolution (or
Faltung) integral:
t
∫0
U ( t) = P( x)W ( t − x)dx = P( t) * W ( t),
16 J. France and J. Dijkstra
where x is a dummy variable ranging over the time interval zero to the
present time t during which the input has occurred, and W is a weighting
function which weights past values of the input to the present value of the
output. The symbol * denotes the convolution operator. Integral equations
are much less common in biology than differential equations though they
occur as convolution integrals in areas such as tracer kinetics. Further
discussion of these issues can be found in Thornley and France (2006).
Application of the rate:state formalism is illustrated with reference to
coccidiosis, an intestinal disease in chickens caused by protozoan parasites
of the genus Eimeria. The life cycle of E. tenella, a typical species that
invades the caecum, is depicted in Fig. 2.3.
Fig. 2.3. Life cycle of E. tenella: (a) sporulated oocyst; (b) sporozoite being liberated from
oocyst and sporocyst; (c) sporozoite; (d) trophozoite parasitizing an epithelial cell; (e) early
schizont; (f) mature first-generation schizont; (g) first-generation merozoite parasitizing another
epithelial cell; (h, i) second-generation schizonts; (j) rupture of second-generation schizont; (k)
second-generation merozoite may parasitize other epithelial cells (l) for a third asexual cycle, or
may parasitize an epithelial cell (m) to become a female gametocyte (q); merozoite parasitizing
an epithelial cell (n) and becoming a male gametocyte (o); (p) liberated microgametes unite with
macrogamete (r), which develops into oocyst (s) and is liberated in the faeces by host (t),
sporulation (u) of oocyst occurs in outside environment (source: Reid, 1984).
Scientific Progress and Mathematical Modelling 17
The main features of this complex life cycle are: (i) the exogenous
development of newly excreted oocysts in the litter to become infectious
sporulated oocysts; and (ii) programming of the parasite to undergo
controlled replication within the intestinal mucosa, with a time delay
between each stage. Endogenous development occurs from the eight
sporozoites released from each oocyst which then undergo a maximum of
three cycles of sexual divison (schizogony) with known multiplication rates.
Each of the three generations of schizonts contains a different but
relatively constant number of protozoan forms known as merozoites.
Merozoites released from third generation schizonts give rise to the sexual
phases of the cycle, forming either male microgametes or female
macrogametes. Fertilization of the macrogametes results in zygotes which,
after the development of a protective wall, are released as unsporulated
oocysts. A much simplified version of this life cycle is shown in Fig. 2.4.
The system as represented is defined at time t by five state variables: X1,
X2, …, X5. These variables represent the number of oocysts per bird in the
litter (X1), the number of oocysts inside a single bird (X2), the number of
sporozoites inside a bird (X3), the number of schizonts inside a bird (X4),
and the number of zygotes inside a bird (X5). The model then comprises
five first order differential equations given by Eqn 2.3 with q = 5. The solid
arrows between boxes represent flows (per unit time) between the different
stages of the life cycle included in the model. Time delays are incorporated
to allow for stages not explicitly or inadequately represented. The model
can be solved using an appropriate set of parameter values (S) to give values
of the state variables over time, and to simulate the effects of intervention
strategies such as the use of vaccine oocysts in the feed (Fig. 2.4).
A second application of the formalism is demonstrated by considering
the synthesis of milk fat and lactose, two of the principal constituents of
Oocysts Ingested
Zygotes, Time
in litter, oocysts,
X5 X1 X2
Time
delay Vaccine
oocysts
Fig. 2.4. Simplified representation of the life cycle of E. tenella for use as a model (from
Parry et al., 1992).
18 J. France and J. Dijkstra
Fig. 2.5. Biochemical pathways involved in the synthesis of milk fat and lactose in the
mammary gland of the lactating sow (source: J.P. Cant, personal communication).
Scientific Progress and Mathematical Modelling 19
(X4) and lactose (X5). The model then comprises five first order differential
equations given by Eqn 2.3 with q = 5, as for the simplified representation
of the life cycle of E. tenella (Fig. 2.4). The solid arrows between boxes
represent flows (mols per unit time) between the state variables of the
model. An acetyl CoA transaction is incorporated to generate ATP for
metabolic transactions via its oxidation. The model can be used to simulate
a complete lactation and to help develop practical feeding strategies.
These two applications illustrate the power of differentials and the
rate:state formalism in providing quantitative, dynamic descriptions of
biological life cycles and biochemical pathways, which are central to pig
and poultry science.
Conclusions
Fig. 2.6. Simplified milk fat and lactose synthesis for use as a model (adapted from
Pettigrew et al., 1992).
20 J. France and J. Dijkstra
that is of value in its own right, that value is greatly enhanced if these
refinements can be related to the interaction between observations
resulting from experiments and from simulations. Modelling increases the
efficiency and effectiveness of experiments with animals and enhances
progress in understanding and controlling pig and poultry production.
Biological research, if it is to remain truly relevant, must be
undertaken at several levels of generality, e.g. cell, tissue or organ, whole
organism, population. There is much more to biology than just molecular
science. Hopefully, the molecular chauvinism that seems to have
dominated biological research thinking (and hence funding) for much of
the last quarter century is finally at an end. This chapter has identified
different modelling approaches, i.e. teleonomic, empirical and mechanistic
modelling, and different mathematical paradigms drawn from different
branches of mathematics. No approach or paradigm is advocated as being
universally superior; no one has a monopoly on wisdom. It is noteworthy
and pleasing that papers on pig and poultry modelling were read at the
present workshop and also formed a significant part of a recently held 5-
yearly farm animal modellers workshop (Kebreab et al., 2006). It is, after
all, a truism that those modelling pig nutrition have things to learn from
their counterparts working, for example, in poultry nutrition, and vice
versa. Thus scientific pluralism, not just across animal species but also
across levels of generality and types of modelling, should be a pillar for
future development of the activity of pig and poultry modelling.
Peering into a crystal ball and attempting to foretell what lies ahead is
usually a futile task. To quote Baldwin (2000): ‘previewing the future is an
equivocal process’. We think it sufficient to conclude by saying that a future
focus for pig and poultry modelling based on scientific pluralism, with
emphasis on solving biological problems rather than applying mathematical
techniques, offers a fruitful way ahead.
Acknowledgement
We thank Dr John Thornley for many useful discussions on this topic over
a number of years.
References
Anon (1960) Working models in medicine. Journal of the American Medical Association
174, 407–408.
Baldwin, R.L. (2000) Introduction: history and future of modelling nutrient
utilization in farm animals. In: McNamara, J.P., France, J. and Beever, D.E.
(eds) Modelling Nutrient Utilization in Farm Animals. CABI Publishing,
Wallingford, UK, pp. 1–9.
Dijkstra, J., Kebreab, E., Kwakkel, R.P. and France, J. (2006) Development of a
dynamic model of Ca and P flows in layers. In: Kebreab, E., Dijkstra, J., Gerrits,
W.J.J., Bannink, A. and France, J. (eds) Nutrient Digestion and Utilization in Farm
Animals: Modelling Approaches. CABI Publishing, Wallingford, UK, pp. 192–210.
Scientific Progress and Mathematical Modelling 21
Draper, N.R. and Smith, H. (1998) Applied Regression Analysis, 3rd edn. Wiley, New
York.
Haldane, J.B.S. (1927) Science and theology as art forms. In: Possible Worlds. Chatto
& Windus, London.
Halas, V., Dijkstra, J., Babinszky, L., Verstegen, M.W.A. and Gerrits, W.J.J. (2004)
Modelling of nutrient partitioning in growing pigs to predict their anatomical
body composition. 1. Model description. British Journal of Nutrition 92,
707–723.
Kebreab, E., Dijkstra, J., Gerrits, W.J.J., Bannink, A. and France, J. (eds) (2006)
Nutrient Digestion and Utilization in Farm Animals: Modelling Approaches. CAB
International, Wallingford, UK, viii+480 pp.
Kuhn, T.S. (1963) The Structure of Scientific Revolutions. University Press, Chicago,
Illinois.
Mach, E. (1942) The Science of Mechanics, 9th edn. Open Court, LaSalle, Illinois.
Monod, J. (1975) Chance and Necessity. Collin metabolism of lactating sows. Journal of
Animal Science 70, 3742–3761.
Popper, K.R. (1968) The Logic of Scientific Discovery. Hutchinson, London.
Popper, K.R. (1969) Conjecture and Refutations. The Growth of Scientific Knowledge, 3rd
edn. Routledge & Kegan Paul, London.
Reid, W.M. (1984) Coccidiosis. In: Hofstad, M.S. (ed.) Diseases of Poultry, 7th edn.
Iowa State University Press, Ames, Iowa, pp. 784–846.
Rosenbluth, A. and Weiner, N. (1945) The role of models in science. Philosophical
Science 12, 316–321.
Thornley, J.H.M. and France, J. (2006) Mathematical Models in Agriculture, 2nd edn.
CAB International, Wallingford, UK, 886 pp.
Thornley, J.H.M. and Johnson, I.R. (1989) Plant and Crop Modelling. Oxford
University Press, Oxford, UK.
Wilson, E.O. (1978) On Human Nature. Harvard University Press, Cambridge,
Massachusetts.
3 Basic Concepts Describing
Animal Growth and Feed Intake
N.S. FERGUSON
Maple Leaf Foods Agresearch, 150 Research Lane, Guelph, Ontario,
Canada, N1G 4T2
fergusns@mapleleaf.ca
Introduction
Animal
Maintenance
Growth Gut
Capacity
Requirements C
R Feed O
E N
S S
O Physical T
U R
Environment A
R
C I
E N
S Social T
Desired Intake Environment Constrained
Potential Growth S
Intake and Growth
Nutrient Nutritional
Carcass Manure
Requirements Management
Grading
Fig. 3.1. Framework of the processes involved in modelling growth and feed intake
and the subsequent commercial application (after Emmans and Oldham, 1988).
1. The animal will always attempt to achieve its potential rate of growth
which is defined by its current state and genetic potential;
2. The amount of feed eaten will be the lesser of what the diet can offer to
achieve potential growth and the capacity of the gut, within the constraint
of maintaining heat balance;
3. Health status and stocking density are possible constraints on potential
growth;
4. Predicted responses are of the average individual.
With an accurate description of the genotype, the potential growth rate of
the animal may be predicted. When the nutritional and environmental
inputs are inadequate the animal will fail to achieve its potential growth; the
extent to which it is constrained will be defined by a set of rules governing
the partitioning of nutrients according to the most limiting factor (e.g. amino
acid, energy, disease challenge, gut capacity, maintaining heat balance, etc.).
The corollary to this suggests that the nutrient and environmental inputs
required to achieve potential growth can be determined. Based on this
approach, an adequate description of the genotype, the feed, health status,
physical and social environment are required.
24 N.S. Ferguson
Animal Description
The equations above indicate that the rate at which an animal grows
will depend almost entirely on its current state and two inherent
characteristics, B and Pm (Taylor, 1980). Potential protein deposition (pPD)
will only be realized if the animal is able to ingest sufficient quantities of
energy and the first limiting amino acid, and if the environment is
sufficiently cool to allow the animal to lose the subsequent heat produced.
Otherwise, actual protein deposition rate (PD) will be lower than pPD.
Examples of estimated constants for different sexes and strains of pigs
derived from the literature and experiments are shown in Table 3.1.
Fat growth
Empty body weight gains (EBWTg) for each day will be calculated from
the sum of the four components, after all other constraints or conditions
have been met, including environmental
EBWTg = PD + LD + AD + WD (g/day).
the empty body weight gains are added to the empty body weight at the
start of the day to give the empty body weight at the end of the day.
To translate empty into total body weights, gut fill has to be considered.
Gut fill is determined from the equation of Whittemore (1998):
Gut fill = 1.05 + 0.05 (0.008 crude fibre 1000 – 40) (g/day),
where crude fibre (CF) is a dietary input value, the daily live weight gains
(ADG) are calculated as:
ADG = EBWTg Gut fill (g/day).
Body weight at any given time (BWTt ) becomes:
BWTt = BWT(t-1) + ADG/1000 (kg/day).
Similarly for each body component, their weights equal the sum of their
starting weight and growth rate for each day. Whether the animal is able to
achieve its potential growth rate each day is dependent on the feed being
offered, the health status and on the environment in which it is housed.
28 N.S. Ferguson
Table 3.2. Profiles of the amino acid coefficients (expressed as mg/g protein) used for the various
constituents of protein requirements (after Green and Whittemore, 2003 and de Lange, 2004).
Lys M+C Thr Trp Ile Leu P+T His Val
maximum feed intake limits based on animal size (Black et al., 1986)
irrespective of the actual dietary constituents. Kyriazakis and Emmans
(1995) and Whittemore et al. (2001a, 2003) have proposed the use of water
holding capacity as a means of estimating the bulk constraint of a diet.
Unfortunately estimates of water holding capacity for diets are not readily
available and therefore an alternative approach is proposed using the IOM
content (indigestible component) of the diet to predict the bulk density of
the diet (BLKDN):
BLKDN = 0.36 + (0.857 + form) IOM,
where IOM = OM – digOM (kg/kg)
OM = organic matter (1-ash)
DigOM = digestible organic matter (Noblet et al., 2004)
form = physical form of the feed where with pellets: form = 0.0;
crumbles: form = 0.01; mash: form = 0.02.
The maximum feed intake on any one day or constrained feed intake (cFI)
is determined by the following equation, which also incorporates a
rudimentary adjustment for breed differences in appetite capacity:
cFI = ((26.78 + (171.34 Pt) + (2.3316 Pt2) )/ BULKDN)
AdjustBlk (g/day),
where AdjustBlk = appetite factor depending on genotype.
The equation considers both the size effect of the animal and the
indigestible component of the diet, in determining a constrained intake. In
this case, the contribution of size is a quadratic function of protein weight.
The desired feed intake (dFI) of the pig in a thermal neutral environment
would be the larger of dFIe and dFIp while the actual daily feed intake
(aFI) would be the lesser of dFI and cFI. For a perfectly balanced diet,
dFIe would equal dFIp. The decision-making process to determine aFI is
illustrated in Fig. 3.2.
There are three possible pathways to consider, each with their own
consequences on protein and fat growth assuming the physical
environment is not limiting.
Feed bulk
dFIe dFIp
greater
Actual FI
Compensatory responses
The work by Kyriazakis and Emmans (1991), de Greef (1992), Tsaras et al.
(1998) and Ferguson and Theeruth (2002), provides substantial
experimental support for the previously discussed compensatory growth
theory. These authors found that when young pigs were made fatter, by
eating a poor quality diet (low CP:ME ratio), and then were placed on a
high protein diet, they deposited fat at a much slower rate, than pigs that
were leaner. These data corroborate the idea of a preferred level of fatness
and the desire of the pig to maintain this level (Fig. 3.3).
The implication of this compensation is that pigs will adjust their
voluntary intake to maintain pPD but at the expense of LD once the
causative factor has been removed. It is therefore possible for negative LD
to occur simultaneously with a positive PD, provided there is enough fat to
lose. A minimum amount of body fat is essential for sustaining life, and is
often expressed in terms of a minimum body lipid:protein ratio (L:Pmin).
Wellock et al. (2003a) assumed a L:Pmin ratio of 0.1 but this was not
substantiated. In practice, it is highly improbable that pPD will continue
unabated when the ratio of body lipid:protein decreases below 0.3, even in
young pigs who already have a low L:P ratio (Stamataris et al., 1991).
Physical Environment
feed intake it is necessary to compare the daily heat production (THP) with
the maximum (THLmax) and minimum (THLmin) daily heat loss limits. If
THP falls within these upper and lower bounds then growth and feed
intake are unaffected by the thermal environment. However, should THP
extend beyond these boundaries, there will be changes in predicted feed
intake and body composition (Fig. 3.4).
Any external stimuli that can affect the temperature the animal actually
‘feels’ at its level will influence the rate of heat loss from the animal
(Whittemore, 1983). This includes factors such as ventilation, the type of floor
material and the insulation of the house. Therefore, ambient temperature
needs to be adjusted accordingly, to reflect an ‘effective’ temperature (Te).
Total heat loss (THL) is the sum of the non-evaporative (or sensible, SHL)
and evaporative heat loss (EHL) components. Therefore, to determine
THLmax and THLmin the minimum and maximum amounts of SHL and
EHL, respectively, have to be determined.
Desired
Lipid weight (kg)
Improve diet
THLmax
THLmin
Restrict Remove
intake restriction
Temperature
Protein weight (kg)
Fig. 3.3. The compensatory responses in fat Fig. 3.4. The relationship between
growth when animals are made either fatter temperature and total heat loss, and the
or leaner than their preferred or desired level maximum (THLmax) and minimum (THLmin)
of fatness. bounds within which growth and feed intake
are unaffected.
36 N.S. Ferguson
SHLmax
Sensible heat loss (SHL)
Evaporative heat loss (EHL)
SHLmin
EHLmax
EHLmin
Temperature Temperature
Fig. 3.5. Relationship between evaporative Fig. 3.6. Relationship between sensible
heat loss and temperature. (non-evaporative) heat loss and temperature.
Basic Concepts Describing Animal Growth and Feed Intake 37
The amount of heat lost through SHL depends on the slope of the line
in Fig. 3.6 (HLslope), the temperature difference between the animal and its
immediate surroundings, and the surface area from which the heat is lost.
Therefore, SHL can generally be defined as:
SHL = HLslope (Body Temperature – Te) BWT0.67 (kJ/day),
where HLslope = 48
Te = effective temperature.
A minimum amount of heat can be stored by allowing body temperature to
increase from 38 to 40.5 when temperatures exceed the upper bound of the
thermoneutral range. If the animal is hot or cold then certain anatomical
and behavioural changes occur, resulting in the core body temperature
either rising (40.5°C, SHLmax) or remaining constant at 38°C (SHLmin),
respectively. The SHL component contributes very little towards heat
production at high temperatures, because sensible heat loss depends on the
difference in temperature between the environment and the surface of the
pig (Mount, 1975). At an effective temperature of 40.5°C, SHL will be zero.
Social Environment
Stocking density
Recent evidence suggests that stress associated with high stocking density
results in a reduction in protein growth, irrespective of feed intake, such that
the amount of feed consumed is driven by the lower PD requirement and not
Basic Concepts Describing Animal Growth and Feed Intake 39
vice versa (Chapple, 1993; Baker and Johnson, 1999; Morgan et al., 1999;
Matteri et al., 2000; Ferguson et al., 2001). Based on these findings and within
the current context, it is reasonable to assume that stocking density exerts its
effect by reducing protein growth through a lower rate of maturing. When
the amount of space per pig declines below a minimum value then the rate of
maturity will be reduced accordingly. The function used to implement this
effect is based on the surface area of the pig and the amount of space available
(kg0.67/m2) (StDen). The adjusted rate of maturity (Badj) is calculated as:
Badj = B (1 – ((StDen –25) /50)) (/day),
where StDen = (pigs/pen BWT0.67) / pen size (kg0.67/m2).
Disease affects the growth performance and feed intake of growing pigs,
the response depending on the source and severity of infection (Baker and
Johnson, 1999; Greiner et al., 2000; Escobar et al., 2002). The possible
effect that health status has on growth under commercial conditions is
illustrated in Fig. 3.7. Although animals may show no signs of clinical
disease, the sub-clinical disease challenge can reduce performance.
The main effects of sub-clinical disease appear to be an increased
maintenance requirement, reduced nutrient digestibility, reduced protein
growth and feed intake (Black et al., 1999; Knap, 2000a). How these are
mediated is not clear. Within the proposed modelling framework, the
question becomes what factors, be they animal or dietary, are likely to be
affected by disease or health status, and how are they adjusted in a simple
Optimal
8 days longer to market
130 Good
At 110 kg Poor
110
Live weight (kg)
90
18 days longer to market
70
50
30
10
50 75 100 125 150 175 200 225
Age (days)
Fig. 3.7. Effects of sub-clinical disease challenge on changes in live weight over time. The
health status of the pig is indicated by Optimal, Good and Poor.
40 N.S. Ferguson
but meaningful way. Logic would dictate that of the animal parameters,
mature protein weight is unlikely to be affected, except under a severe
disease challenge, but the rate at which it achieves this will be adversely
affected. With little evidence to support or disprove this theory, it is
proposed that a proportional reduction in B, relative to the degree of
disease challenge or health status, be adopted to account for the reduction
in protein deposition and subsequent feed intake. In addition, maintenance
protein and energy requirements will be increased, while activity levels will
be reduced.
1.2
1.0
Health coefficients
0.8
0.6
0.4
0.2
0
0 5 10 15 20 25 30 35 40
Weeks after weaning
Fig. 3.8. Health profile illustrating the changes in health and well-being status post weaning.
Basic Concepts Describing Animal Growth and Feed Intake 41
10.0
7.5
2.5
0.0
–2.5
–5.0
–7.5
–10.0
21 35 49 63 77 91 105 119 133 147 164
Age (days)
Fig. 3.9. Comparison of the deviations from actual live weight over time between predicted
results, which include and exclude health status adjustments.
Commercial example
Comparative data from a number of different commercial production units
within Canada were used. The main cause of the differences in
performance between units was the health and well-being status of the
pigs. The data were separated according to the performances of the best
(High) and worst (Low) health status units, and were compared against the
model predictions. To simulate the differences in health conditions, the
health status coefficients of the Low health units were assumed to be 0.9
42 N.S. Ferguson
times the High health units over the live weight range of 6 to 108 kg. The
growth rate and feed intake results are shown in Figs 3.10a and b.
The similarity between actual versus predicted suggests that the
proposed approach does not produce unrealistic results and is sufficiently
sensitive to differentiate between animals differing in health status.
However, it would be inappropriate and too simplistic to assume that the
process is valid as a means of quantifying the complex effects of specific
health challenges. Nevertheless, the results indicate that it is a reasonable
attempt to incorporate the adverse effects of reduced health status on
growth and feed intake.
Critical to any model is the need to test the underlying theories that drive
or control the modelling process. This is not an easy exercise, as an
(a)
900
800
Growth rate (g/day)
700
600
500
400
300
High health Low health High health Low health
(6–29 kg) (6–29 kg) (26–108 kg) (26–108 kg)
(b) 2300
2000
Feed Intake (g/day)
1700
1400
1100
800
500
High health Low health High health Low health
(6–29 kg) (6–29 kg) (26–108 kg) (26–108 kg)
Fig. 3.10. Comparison of (a) growth rates and (b) feed intakes, between Actual ( ) and
Predicted (䊐), for High versus Low health status producers.
Basic Concepts Describing Animal Growth and Feed Intake 43
accurate prediction under one set of circumstances does not mean that it is
‘valid’ (Black, 1995). However, confidence in the validity of the model may
be gained when accurate predictions, for a number of diverse
circumstances, are consistently obtained. Unfortunately, there is no specific
way in which models can be validated other than by comparing the
similarities in the model predictions with experimental outcomes. The
three key components of the proposed model that need to be tested are
the growth and intake responses when energy is limiting, when protein
(amino acid) is limiting and when ambient temperatures change.
Nursery pigs
To evaluate the effect of Effective Energy content in young nursery pigs,
data from a recent trial (Ferguson, 2005 unpublished results) were used.
Five energy levels, ranging from 13.00 to 13.82 MJ/kg, were fed to pigs
between 12 and 25 kg live weight. The animal definition parameters used
were: B = 0.0142, Pm = 33.0kg, LPRm = 2.30, WPRm = 3.40, APRm =
0.20. The results of the comparison between actual and predicted growth
rates and feed intakes are shown in Figs 3.11a and b. Given that young
pigs (< 30 kg) cannot eat sufficient to satisfy their high relative potential
growth irrespective of the nutrient density of the diet (Whittemore et al.,
2001a), this comparison, in many respects, evaluates the ability of the
model to define the gut capacity and the effects of this constraint on
predicted performance. From Figs 3.11a and b, the results would suggest
that the method proposed in this chapter is sufficiently robust to predict
feed intake and performance within the 95% confidence interval of actual
results.
The higher predicted feed intake value at 13.00 MJ/kg was the result of
the reduced efficiency of protein utilization associated with a low
ME:digestible CP ratio (65 MJ/kg). As the minimum ratio, below which the
efficiency of protein utilization will decline, is 73 MJ/kg (Kyriazakis and
Emmans, 1992b) there will an increased demand for dietary protein to
compensate for the reduced efficiency of utilization. The increased
requirement for protein has resulted in an increase in voluntary feed
intake. Therefore, the slightly higher predicted ADFI at 13.00 MJ/kg would
suggest either an overestimation of the gut capacity or that the equation
predicting efficiency of protein utilization is underestimating the efficiency
in low ME:dCP diets. However, as the overestimation was within the 95%
confidence limits, there is no justification for rejecting the status quo.
Grower-Finisher pigs
The data from Campbell et al. (1985) were used to compare the responses
to energy intake in grower pigs between 48 and 90 kg live weight. The
parameters used to describe the castrate genotype were: B = 0.0125, Pm
44 N.S. Ferguson
(a)
10
8
6
4
2
0
–2
–4
–6
–8
–10
13.00 13.23 13.48 13.72 13.82
Effective energy content (MJ/kg)
Fig. 3.11. Comparison of the deviations in (a) growth rates (ADG) and (b) feed intakes
(ADFI), between Actual and Predicted in response to Effective energy in nursery pigs. (—)
95% Confidence interval.
= 28.0 kg, LPRm = 4.0, WPRm = 3.00, APRm = 0.20. The lysine:energy
ratio was reported to be in excess of that required for maximum protein
deposition on ad libitum feed intake, therefore energy was always the most
limiting nutrient. A summary of the differences between the predicted
outcomes from the model and actual data is shown in Figs 3.12a,b
and c.
The most noticeable differences (> 1 standard deviation) were
confined to PD and LD at very low energy intakes (55% of ad libitum
intakes) where the model overestimated PD and underestimated LD.
These differences are due, in part, to the difficulty in establishing the
relationship between protein and fat deposition when energy is severely
limiting (restricted intakes) and secondly, in the inadequate description of
the environmental conditions given in the scientific paper. Without a
Basic Concepts Describing Animal Growth and Feed Intake 45
(a)
1000
800
ADG (g/day)
600
400
200
20 25 30 35 40 45
DE Intake (MJ/day)
Fig. 3.12a. Comparison of actual (䊏) versus predicted (—) responses in growth rate (ADG)
to digestible energy intake (DE Intake), in pigs grown from 48 to 90 kg live weight. Data from
Campbell et al. (1985). Bars represent 1 standard deviation.
(b)
4.2
4.0
3.8
3.6
Feed:Gain
3.4
3.2
3.0
2.8
2.6
2.4
20 25 30 35 40 45
DE Intake (MJ/day)
Fig. 3.12b. Comparison of actual (䊏) versus predicted (—) responses in feed:gain ratio to
digestible energy intake (DE Intake), in pigs grown from 48 to 90 kg live weight. Data from
Campbell et al. (1985). Bars represent 1 standard deviation.
(c)
400
350
300
PD and LD (g/d)
250
200
150
100
50
0
20 25 30 35 40 45
DE Intake (MJ/d)
Fig. 3.12c. Comparison of actual versus predicted responses in protein deposition (PD) (䊏, —)
and lipid deposition (LD) (䉱, ....) to digestible energy intake (DE Intake), in pigs grown from 48
to 90 kg live weight. Data from Campbell et al. (1985). Bars represent 1 standard deviation.
(a)
800
700
ADG (g/day)
600
500
400
300
0.5 0.7 0.9 1.1 1.3 1.5 1.7
Total lysine content (%)
(b)
1100
1000
Feed Intake (g/day)
900
800
700
600
0.5 0.7 0.9 1.1 1.3 1.5 1.7
Total lysine content (%)
(c)
2.25
2.00
FCR (g/g)
1.75
1.50
1.25
1.00
0.5 0.7 0.9 1.1 1.3 1.5 1.7
Total lysine content (%)
Fig. 3.13. (a) Comparison of responses in growth rates (ADG) to total lysine content in the diet.
(b) Comparison of responses in feed intake to total lysine content in the diet. (c) Comparison of
responses in feed:gain (FCR) to total lysine content in the diet. Gatel Expt A, Actual (䊏),
Predicted (—); Gatel Expt B, Actual (䉱), Predicted (....); Kyriazakis Actual (䊉), Predicted (– –).
Bars represent 1 standard deviation.
48 N.S. Ferguson
To test the effects of the thermal environment, the data from Rinaldo and
Le Dividich (1991) were selected. This study examined the performance of
growing pigs between 10 and 30 kg live weight when kept in one of four
different temperatures, 12, 18.5, 25 and 31.5°C. The animal description
parameters used were: B = 0.0150, Pm = 35 kg, LPRm = 3.0, WPRm =
3.3, APRm = 0.20. The differences between actual and predicted are
summarized in Fig. 3.14a and b. The only significant difference between
the model predicted and the reported results was in daily growth rates
(ADG) at 12°C, which in turn affected the feed:gain ratio. The very low
ADG observed by Rinaldo and Le Dividich (1991) is contrary to what is
found in most published literature on growth responses at low
temperatures, where ADG remains constant at low temperatures (Nienaber
et al., 1987; Ferguson and Gous, 1997; Wellock et al., 2003b). No
explanation is given for the reduced growth rate reported in the
publication by Rinaldo and Le Dividich (1991).
Conclusions
The theory of growth and feed intake described in this chapter is based on
the proposition that an animal eats to grow to its potential, and if anything
prevents it attaining this growth, then it will grow according to what it has
eaten. Foremost, therefore, is the need to predict desired feed intake and
how the subsequent interactions with the animal and its environment
influence the actual voluntary feed intake. Energy that is available for
maintenance and productive purposes, after removing the heat increment
(b)
(a) 3
ADG and feed intake (g/day)
1400
1200 2.5
Feed:gain
1000 2
800
1.5
600
400 1
8 12 16 20 24 28 32 8 12 16 20 24 28 32
Ambient temperature (°C) Ambient temperature (°C)
Fig. 3.14a. Comparison of actual versus Fig. 3.14b. Comparison of actual versus
predicted responses in growth rate (ADG) predicted responses in feed:gain (䊏, —) to
(䉱, ....) and feed intake (䊏, —) to ambient ambient temperature, in pigs grown from 10
temperature, in pigs grown from 10 to 30 kg to 30 kg live weight. Data from Rinaldo and
live weight. Data from Rinaldo and Le Dividich Le Dividich (1991). Bars represent 1
(1991). Bars represent 1 standard deviation. standard deviation.
Basic Concepts Describing Animal Growth and Feed Intake 49
References
AmiPig (2000) Ileal standardized digestibility of amino acids in feedstuffs for pigs.
AFZ, Ajinomoto Eurolysine, Aventis Animal Nutrition, INRA and ITCF.
Association Francaise de Zootechnie, Paris.
Baker, D.H. and Johnson, R.W. (1999) Disease, stress, cytokines and amino acid
needs of pigs. Pig News and Information 20, 123N–124N.
Black, J.L. (1995) The testing and evaluation of models. In: Moughan, P.J.,
Verstegen, M.W.A. and Visser-Reyneveld, M.I. (eds) Modelling Growth in the Pig.
EAAP publication no.78, Wageningen Pers, Netherlands, pp. 23–31.
Black, J.L., Campbell, R.G., Williams, I.H., James, K.J. and Davies, G.T. (1986)
Simulation of energy and amino acid utilisation in the pig. Research and
Development in Agriculture 3, 121–145.
Black, J.L., Bray, H.J. and Giles, L.R. (1999) The thermal and infectious
environment. In: Kyriazakis, I. (ed.) A Quantitative Biology of the Pig. CABI
Publishing, Wallingford, UK, pp. 71–97.
Boisen, S., Hvelplund, T. and Weisbjerg, M.R. (2000) Ideal amino acid profiles as a
basis for feed protein evaluation. Livestock Production Science 64, 239–251.
Bridges, T.C., Turner, L.W., Smith, E.M., Stahly, T.S. and Louwer, O.J. (1986) A
mathematical procedure for estimating animal growth and body composition.
Transactions of the American Association of Agricultural Engineers 29, 1342–1347.
Brody, S. (1945) Bioenergetics and Growth. Hafner Publishing Company, New York.
Bruce, J.M. and Clark, J.J. (1979) Models of heat production and critical
temperature for growing pigs. Animal Production 28, 353–369.
Campbell R.G., Taverner, M.R. and Curic, D.M. (1985) The effects of sex and
energy intake between 48 and 90 kg live weight on protein deposition in
growing pigs. Animal Production 40, 497–503.
50 N.S. Ferguson
Ferguson, N.S., Gous, R.M. and Emmans, G.C. (1994) Preferred components for
the construction of a new simulation model of growth, feed intake and nutrient
requirements of growing pigs. South African Journal of Animal Science 24, 10–17.
Ferguson, N.S., Arnold, G.A., Lavers, G. and Gous, R.M. (2000a) The response of
growing pigs to amino acids as influenced by environmental temperature. 1
Threonine. Animal Science 70, 287–297.
Ferguson, N.S., Arnold, G.A., Lavers, G. and Gous, R.M. (2000b) The response of
growing pigs to amino acids as influenced by environmental temperature. 2
Lysine. Animal Science 70, 299–306.
Ferguson, N.S., Lavers, G. and Gous, R.M. (2001) The effect of stocking density on
the responses of growing pigs to dietary lysine. Animal Science 73, 459–469.
France, J., Dijkstra, J., Thornley, J.H.M. and Dhanoa, M.S. (1996) A simple but
flexible growth function. Growth, Development and Aging 60, 71–83.
Gatel, F., Buron, G. and Fekete, J. (1992) Total amino acid requirements of weaned
piglets 8 to 25 kg live weight given diets based on wheat and soya-bean meal
fortified with free amino acids. Animal Production 54, 281–287.
Gompertz, B. (1825) On the nature of the function expressive of the law of human
morality and on a new method of determining the value of life contingencies.
Philosophical Transactions of the Royal Society, 513–585.
Green, D.M. and Whittemore, C. T. (2003) Architecture of a harmonized model of
the growing pig for the determination of dietary net energy and protein
requirements and of excretions into the environment. (IMS Pig). Animal Science
77, 113–130.
Greiner, L.L., Stahly, T.S. and Stabel, T.J. (2000) Quantitative relationship of
systemic virus concentration on growth and immune response in pigs. Journal
of Animal Science 78, 2690–2695.
Hyun, Y., Ellis, M., Riskowski, G. and Johnson, R.W. (1998) Growth performance
of pigs subjected to multiple concurrent environmental stressors. Journal of
Animal Science 76, 721–727.
Knap, P.W. (1999) Simulation of growth in pigs: evaluation of a model to relate
thermoregulation to body protein and lipid content and deposition. Animal
Production 68, 655–679.
Knap, P.W. (2000a) Variation in maintenance requirements of growing pigs in
relation to body composition. A simulation study. PhD thesis. University of
Wageningen, Netherlands.
Knap, P.W. (2000b) Time trends of Gompertz growth parameters in “meat-type”
pigs. Animal Science 70, 39–49.
Knap, P.W., Roehe, R., Kolstad, K., Pomar, C. and Luiting, P. (2003)
Characterization of pig genotypes for growth modeling. Journal of Animal
Science 81, (E. Suppl 2) E187–195.
Kornegay, E.T. and Notter, D.R. (1984) Effects of floor space and number of pigs
per pen on performance. Pig News and Information 5, 23N–33N.
Kyriazakis, I. and Emmans, G.C. (1991) Diet selection in pigs: Dietary choices made
by growing pigs following a period of underfeeding with protein. Animal
Production 52, 337–346.
Kyriazakis, I. and Emmans, G.C. (1992a) The effects of varying protein and energy
intakes on the growth and body composition of pigs. 1. The effects of energy
intake at constant, high protein intake. Animal Production 68, 603–613.
Kyriazakis, I. and Emmans, G.C. (1992b) The effects of varying protein and energy
intakes on the growth and body composition of pigs. 2. The effects of varying
both energy and protein intake. Animal Production 68, 615–625.
52 N.S. Ferguson
Kyriazakis, I. and Emmans, G.C. (1995) The voluntary feed intake of pigs given
feeds based on wheatbran, dried citrus pulp and grass meal in relation to
measurements of feed bulk. British Journal of Nutrition 73, 191–207.
Kyriazakis, I. and Emmans, G.C. (1999) Voluntary food intake and diet selection.
In: Kyriazakis, I. (ed.) A Quantitative Biology of the Pig. CABI Publishing,
Wallingford, UK, pp. 229–248.
Kyriazakis, I., Emmans, G.C. and Whittemore, C.T. (1990) Diet selection in pigs:
Choices made by growing pigs given foods of different protein concentrations.
Animal Production 51, 189–199.
Kyriazakis, I., Stamataris, C., Emmans, G.C. and Whittemore, C.T. (1991) The
effects of food protein on the performance of pigs previously given food with
low or moderate protein contents. Animal Production 52, 165–173.
Matteri, R.L., Carroll, J.A. and Dyer, C.J. (2000) Neuroendocrine responses to
stress. In: Moberg, G.P. and Mench J.A. (eds) The Biology of Animal Stress. Basic
Principles and Implications for Animal Welfare. CABI Publishing, Wallingford, UK,
pp. 43–76.
Morgan, C.A., Nielsen, B.L., Lawrence, A.B. and Mendl, M.T. (1999) Describing
the social environment and its effects on food intake and growth. In:
Kyriazakis, I. (ed.) A Quantitative Biology of the Pig. CABI Publishing,
Wallingford, UK, pp. 99–125.
Moughan, P.J. (1999) Protein metabolism in the growing pig. In: Kyriazakis, I. (ed.)
A Quantitative Biology of the Pig. CABI Publishing, Wallingford, UK,
pp. 299–331.
Moughan, P.J., Smith, W.C. and Pearson, G. (1987) Description and validation of a
model simulating growth in the pig (20–90kg liveweight). New Zealand Journal
of Agricultural Research 30, 481–489.
Moughan, P.J., Smith, W.C. and Stevens, E.V.J. (1990) Allometric growth of
chemical body components and several organs in the pig (20–90 kg live
weight). New Zealand Journal of Agricultural Research 33, 77–83.
Mount, L.E. (1975) The assessment of thermal environment in relation to pig
production. Livestock Production Science 2, 381–392.
Nienaber, J.A., Hahn, G.L. and Yen, J.T. (1987) Thermal environmental effects on
growing-finishing swine. 1. Growth, food intake and heat production.
Transactions of the American Association of Agricultural Engineers 30, 1772–1775.
Noblet, J., Sève, B. and Jondreville, C. (2004) Nutritional values for pigs. In:
Sauvant, D., Perez, J.M. and Tran, G. (eds) Tables of Composition and Nutritional
Value of Feed Materials. Wageningen Academic Publishers, Wageningen,
Netherlands, pp. 25–37.
Parks, J.R. (1982) A Theory of Feeding and Growth of Animals. Springer, Berlin.
Pomar, C., Harris, D.L. and Minvielle, F. (1991) Computer simulation model of
swine production systems: I. Modeling the growth of young pigs. Journal of
Animal Science 69, 1468–1488.
Rinaldo, D. and Le Dividich, J. (1991) Assessment of optimal temperature for
performance and chemical body composition of growing pigs. Livestock
Production Science 29, 61–75.
Robertson, T.B. (1923) The Chemical Basis of Growth and Senescence. Monographs of
Experimental Biology. J.B. Lipincott, Philadelphia.
Schinckel, A.P. (1999) Describing the pig. In: Kyriazakis, I. (ed.) A Quantitative
Biology of the Pig. CABI Publishing, Wallingford, UK.
Schinckel, A.P. and de Lange, C.F.M. (1996) Characterization of growth parameters
needed as inputs for pig growth models. Journal of Animal Science 74, 2021–2036.
Basic Concepts Describing Animal Growth and Feed Intake 53
Stamataris, C., Kyriazakis, I. and Emmans, G.C. (1991) The performance and body
composition of young pigs following a period of growth retardation by food
restriction. Animal Production 53, 373–381.
Taylor, St.C.S. (1980) Genetic size-scaling rules in animal growth. Animal Production
30, 161–165.
Tsaras, L.N., Kyriazakis, I. and Emmans, G.C. (1998) The prediction of the
voluntary food intake of pigs on poor quality foods. Animal Science 66,
713–723.
Von Bertalanffy, L. (1957) Quantitative laws for metabolism and growth. Quarterly
Reviews of Biology 32, 217–231.
Webster, A.J.F. (1993) Energy partitioning, tissue growth and appetite control.
Proceedings of the Nutrition Society 52, 69–76.
Wellock, I.J., Emmans, G.C. and Kyriazakis, I. (2003a) Modelling the effects of
thermal environment and dietary composition on pig performance: model
logic and concepts. Animal Science 77, 255–266.
Wellock, I.J., Emmans, G.C. and Kyriazakis, I. (2003b) Modelling the effects of
thermal environment and dietary composition on pig performance: model
testing and evaluation. Animal Science 77, 267–276.
Wellock, I.J., Emmans, G.C. and Kyriazakis, I. (2004) Describing and predicting
potential growth in the pig. Animal Science 78, 379–388.
Whittemore, C.T. (1983) Development of recommended energy and protein
allowances for growing pigs. Agricultural Systems 11, 159–186.
Whittemore, C.T. (1998) The Science and Practice of Pig Production, 2nd edn.
Longman Scientific & Technical, Essex, UK.
Whittemore, C.T., Green, D.M. and Knap, P.W. (2001a) Technical review of the
energy and protein requirements of growing pigs: food intake. Animal Science
73, 3–17.
Whittemore, C.T., Green, D.M. and Knap, P.W. (2001b) Technical review of the
energy and protein requirements of growing pigs: energy. Animal Science 73,
199–215.
Whittemore, C.T., Green, D.M. and Knap, P.W. (2001c) Technical review of the
energy and protein requirements of growing pigs: protein. Animal Science 73,
363–374.
Whittemore, C.T., Emmans, G.C. and Kyriazakis, I. (2003) The relationship
between live weight and the intake of bulky food in pigs. Animal Science 76,
89–100.
4 The Effects of Social Stressors on
the Performance of Growing Pigs
I.J. WELLOCK, G.C. EMMANS AND I. KYRIAZAKIS
Animal Nutrition and Health Department, Scottish Agricultural College,
West Mains Road, Edinburgh, EH9 3JG, UK
Ian.Wellock@sac.ac.uk
Introduction
Different published approaches to pig simulation modelling have advanced
our understanding of pig performance under a wide range of
environmental conditions. They range from the first relatively simple
attempt to model pig growth by Whittemore and Fawcett (1974, 1976),
where predictions were based upon empirical equations, to more recent
and elaborate attempts such as those made by Black et al. (1986),
Whittemore and Green (2002) and Pomar et al. (2003). These latter models
contain various combinations of empirical and mechanistic equations
usually with an underlying biological basis. Attempts to predict feed intake,
although still not universal, are more frequent in recent modelling attempts
(e.g. Black et al., 1986; Bridges et al., 1992; Ferguson et al., 1994) and more
factors have been considered and introduced as model inputs. Stressors in
the physical environment, such as ambient temperature, humidity, air
velocity and floor type have been comprehensively modelled (e.g. Bruce
and Clark, 1979; Black et al., 1986; Wellock et al., 2003a,b) allowing
predictions of performance under varying conditions to be made. Factors
which may act as social stressors, which include group size (N), space
allowance (SPA, m2/BW0.67), feeder space allowance (FSA, feeders/pig), and
mixing on the other hand, have been largely ignored. This is mainly due to
a lack of quantitative data on which to build models and a lack of
understanding of how such stressors affect performance. Effects of the
infectious environment are yet to be included in a systematic way.
The objective of this chapter is to describe how the effects of social
stressors on the performance of growing pigs can be quantified and to
show how these relationships, including variation between genotypes in
their ability to cope (AB), can be incorporated into a more general pig
growth model. The consequences of introducing individual variation into
the model are investigated and the difficulties of estimating parameter
values, with particular regard to AB, are described. Finally, the practical
implications of AB in relation to production, welfare and genetic selection
are discussed along with potential future model developments.
Table 4.1. Equations from Kornegay and Notter (1984) and Turner et al., (2003)
relating group size (N) to average daily gain (ADG, kg), average daily feed intake
(ADFI, kg) and feed conversion ratio (FCR).
Kornegay and Notter (1984) Turner et al. (2003)
Weaner period (7.6 to 21.1 kg)
ADG = 0.4178 – 0.0037N (R2 = 0.97) ADG = 0.416 – 0.00036N (R2 = 0.97)
ADFI = 0.8317 – 0.092N (R2 = 0.97) ADFI = 0.618 – 0.00051N (R2 = 0.98)
FCR = 1.9535 – 0.0051N (R2 = 0.94) FCR = 1.650 + 0.00004N (R2 = 0.96)
Grower period (26.6 to 53.5 kg)
ADG = 0.6407 – 0.0019N (R2 = 0.43) ADG = 0.654 – 0.00048N (R2 = 0.90)
ADFI = 1.5950 – 0.0025N (R2 = 0.87) ADFI = 1.790 – 0.00005N (R2 = 0.90)
FCR = 2.4974 + 0.0037N (R2 = 0.94) FCR = 2.750 + 0.00179N (R2 = 0.97)
Finisher period (44.1 to 92.3 kg)
ADG = 0.7497 – 0.0012N (R2 = 0.82) ADG = 0.715 – 0.00009N (R2 = 0.99)
ADFI = 2.3748 + 0.0032N (R2 = 0.92) ADFI = 2.340 + 0.00033N (R2 = 0.84)
FCR = 3.2182 + 0.0060N (R2 = 0.72) FCR = 3.329 + 0.00104N (R2 = 0.97)
56 I.J. Wellock et al.
Developing a Model
To develop a mechanistic model for predicting the effects of social stressors
on pig performance it is necessary to do the following three things: (i)
determine the mechanism by which social stressors affect performance, this
is needed to integrate social stressor effects into a model in a mechanistic
way; (ii) quantify the effects of the individual social stressors; and (iii)
integrate these social stressor effects into an overall growth model. These
three steps are discussed below in turn.
Space allowance
Table 4.2. Parameter values for the conceptual equations relating the major social stressors
to pig performance estimated from experimental data in the literature (reproduced from
Wellock et al., 2003c, Table 1).
Equationa Parameter 1 Parameter 2 Parameter 3
1 RSPA = b1 + g1 ln (SPA) b1 = 168.49 g1 = 21.48
(9.62)b (2.65)b –
2 RN = b2 – g2 ln (N) b2 = 100c g2 = 3.6971
(0.69)b –
3 EN = (x1.(N-1)).EMaint x1 = 0.0075 – –
5 FRmax = (g3.BW1) / (1000.WHC) g3 = 2.85 – –
7 RMix = b3 – g4.BW – ((g5.BW).ln.(t)) b3 = 100c g4 = 0.6 g5 = 0.18
8 EMix = (x2 – (x3.ln.(t))).EMaint x2 = 1.15 x3 = 0.050 –
aR
SPA, RN and RMix represent the relative daily gain as a percentage of maximal performance
in relation to space allowance (SPA, m2/BW0.67), group size (N) and mixing (Mix),
respectively. EMaint, EN and EMix represent the energy expenditure (MJ/d) due to maintenance,
group size and mixing, respectively. FRmax is the maximal feeding rate (kg/min) and WHC is
the water holding capacity (kg/kg) of the food used as a measure of its bulk.
b Values in parenthesis are standard errors.
c Denotes fixed parameter values.
Group size
1
SPAmin
RSPA
SPAcrit
0
SPA, m2/BW0.67
Fig. 4.1. Effect of space allowance on the relative daily gain (RSPA) in relation to that recorded
at SPA > SPAcrit. SPAcrit represents the point below which space becomes limiting resulting in
a depression in performance and SPAmin represents the minimum amount of space required
for a given animal. Both SPAcrit and SPAmin are affected by body weight and genotype.
performance as N increases (e.g. Wolter et al., 2001; Hyun and Ellis, 2001).
Increasing N by a fixed number has a greater influence on smaller groups
than larger ones, because the social hierarchy of small groups is disrupted
to a greater degree than that of large groups that appear to lack social
structure (Arey and Edwards, 1998; Turner et al., 2001). A logarithmic
form is used to represent the relationship (Fig. 4.2):
RN = b2 – g2.ln.(N) (4.2)
RN is the daily gain as a proportion of that of a singly housed counterpart.
The value of the constant b2 is set to 100 and the value of g2 is assumed to
differ between breeds (see below). Calculated parameter values are given
in Table 4.2.
Although the evidence for greater activity in larger groups is equivocal,
the literature does indicate such a trend (e.g. Petherick et al., 1989; Turner
et al., 2002). Consequently, energy expenditure is increased as N increases.
This increase, EN, is calculated as a proportion of maintenance energy
requirements (Emaint, MJ/day) and included in the calculation of daily
energy requirements. It is assumed that EN will not increase indefinitely
and so a proposed maximum is set at Nm. When N < Nm:
EN = (x1.(N1)).Emaint (4.3)
The value of x1 will differ between genotypes as discussed later. When
NNm, Nm replaces N in Eqn 4.3. To account for a twofold increase in
activity as N increases to Nm, a value of 0.0075 was assigned to x1. A value
of 21 was assigned to Nm.
RN
Group size (N)
Fig. 4.2. Effect of increasing group size (N) on relative daily gain (RN) in relation to that of a
singly housed counterpart.
reached when all of the pigs in the group can no longer satisfy their
desired feed intake (FId, kg/day), due to increased pig competition and is
dependent upon N, FId, maximum feeding rate (FRmax kg/min), and the
number of minutes in the day, 1440.
FSAcrit = (FId/(1440.FRmax)) N (4.4)
FRmax depends upon aspects of mouth capacity, feed composition and
method of feed presentation. It can be calculated as:
FRmax = (g3.BW1.0)/(1000.WHC) (4.5)
where WHC (kg/kg) is the water holding capacity of the feed, used as a
measure of feed bulk. It is assumed that g3 is unaffected by genotype.
When FSA < FSAcrit, then the constrained feed intake (FIc, kg/pig), is
calculated as:
FIc = (1440.FSA.FRmax)/N (4.6)
If troughs are used FSA is calculated as the number of pigs able to feed
simultaneously so that FSA = trough width/ j.BWk. The values assumed for
j and k are 0.064 and 0.33, respectively (Petherick, 1983) and these
estimate the width of the pig at the shoulders.
Mixing
Generally results indicate that mixing is a transient stressor and that, given
sufficient time, there are no noticeable effects of mixing on performance in
the longer run (Spoolder et al., 2000; Heetkamp et al., 2002). There is
however, an initial decline in performance immediately after mixing most
likely due to the increased frequency of antagonistic encounters (D’Eath,
2002) associated with establishing a new stable social structure. Mixing
depresses performance to a greater extent in larger animals due to the
increased ferocity of their fighting (Spoolder et al., 2000) before returning
to normal values. The mixing effect is described as:
Effects of Social Stressors on Performance of Growing Pigs 61
Genetic differences
tmix
ADG (kg/day)
ADGdep
Non-mix
Mix
Time (days)
Fig. 4.3. Effect of mixing on average daily gain (ADG). The extent of depression in ADG
(ADGdep) and the time taken (tmix) to return to non-stressed levels of ADG is determined by
both body weight and genotype.
62 I.J. Wellock et al.
SPAcrit
105
100
95
RSPA
90
85
AB = 10
80
AB = 5
75 AB = 1
70
0.2 0.4 0.6 0.8 1.0
Area (m2/pig)
Fig. 4.4. The effect of genotypic differences in ability to cope with social stressors (AB) on
relative daily gain (RSPA) at differing space allowances. SPAcrit represents the points at which
space becomes limiting.
The model framework used as the starting point is that described and
tested by Wellock et al. (2003a,b). Information is needed about the pig, its
diet and the social and physical environments in which it is kept. No
additional inputs are required to describe either the thermal or dietary
environment. In addition to the three genetic parameters used to predict
potential growth [protein weight at maturity (Pm, kg), the lipid to protein
ratio at maturity (Lm/Pm, kg/kg) and a growth rate parameter (B, d1)], the
parameter, AB, discussed above, is required to describe the pig’s ability to
cope with social stressors. Additional inputs to describe the social
environment are pen area, the number of pigs per pen (N), the number of
individual feeders or the trough length, and the occurrence or not of
mixing. Up to two mixing events are allowed during a run and the weight
at which each mixing occurs is required. The model can be run either to a
final BW (BWf, kg) or for a given period (t, days).
Effects of Social Stressors on Performance of Growing Pigs 63
Table 4.3. Scaling factors (z) for the appropriate parameters to account for variation in ability
to copea (AB) with social stressors (adapted from Wellock et al., 2003c).
Space allowance Group size Mixing
zb1 = 1.2 – 0.04AB zg2 = 1.5 – 0.1AB zg4 = 1.2 – 0.04AB
zg1 = 1.5 – 0.1AB zx1 = 1.1 – 0.02AB zg5 = 0.977 + 0.066AB
zx2 = 1.025 – 0.005AB
zx3 = 0.9 + 0.02AB
aAn AB value of 5 represents the mean and therefore all parameters are multiplied
by a scaling factor of 1 when AB = 5. Values were chosen to represent deviations of
approximately 1% from the mean performance per unit change in AB.
It has been shown in studies with pigs (Hyun et al., 1998a,b) that the effects
of stressors, at intensities expected under commercial conditions, are likely
to be additive rather than multiplicative, antagonistic or synergistic. It is
therefore assumed that, within the bounds of the model, the effects of
multiple stressors on maximum relative daily gain of the stressed animal,
Rs, are additive and are predicted by summing the effects of the individual
stressors.
Rs = Rp.((100 – ((100 – RSPA) + (100 – RN) + (100 – Rmix))) / 100) (4.9)
Here Rp = ADGp / BW and is the pig’s potential relative daily gain. ADGp is
dependent upon the genotype and the current state of the pig. It is
assumed in the model that social stressors lead to a down-regulation in
lean tissue growth, i.e. a decrease in the animal’s ability to attain its
potential. This is equivalent to lowering the growth rate parameter, B,
which in turn leads to a decrease in the ADGp that the pig is able to
achieve. Consequently, Rs is calculated on a daily basis and used to
calculate the modified growth rate parameter, Bs = Rs B, from which
the maximum daily gain of the stressed animal is predicted, ADGs,
replacing ADGp in the model.
As it is assumed that animals eat to attain their potential, a decrease in
ADGp necessarily leads to a decrease in FId. Consequently, the desired feed
intake of the stressed animal is predicted directly from the animal’s
depressed growth potential. Predictions of FId, actual intake and gain are
then made taking account of any changes in energy requirements due to
increases in activity, EN and Emix, and possible constraints on intake due to
limiting FSA, feed composition and the climatic environment.
between the response of the average individual and the mean response of
the population, which is an average of all individuals (Fisher et al., 1973;
Emmans and Fisher, 1986). These differences may prove important when
predicting nutrient requirements (Fisher et al., 1973; Curnow, 1986),
optimizing pig production systems (Pomar et al., 2003) and devising animal
breeding strategies (Knap, 1995).
In order to account for differences between individuals in a group,
between-animal variation was introduced into the model. In addition to
accounting for differences in growth potential, as in the stochastic pig
growth models of Knap (2000) and Pomar et al. (2003), variation in initial
state as described by initial BW (BW0) and ability to cope when exposed to
stressors were also included. Variation in growth potential was generated
by creating variation around the population means of each of the genetic
parameters describing growth, Pm, Lm/Pm and B (Ferguson et al., 1997).
Individual variation in BW0 is generated from the assigned genotype
mean, (mBW0, kg) and standard deviation (sBW0, kg) using the simulated
genetic parameters of the individual to correlate BW0 with potential
growth. By this means individuals in the group with the greatest potential
will tend to have the highest BW0 as would be expected from non-limiting
growth. For further details see Wellock et al. (2004).
It is expected that within a population or group the social
environment (i.e. position within the social hierarchy) affects an
individual’s ability to cope (Muir and Schinckel, 2002) and that pigs
classified as dominant tend to outperform their subordinates (e.g. Hessing
et al., 1994; D’Eath, 2002). There is also evidence that social dominance is
positively correlated to BW in pigs (Drickamer et al., 1999; D’Eath, 2002).
Taken together these results suggest that the larger pigs within a group
tend to be dominant and better able to cope when conditions are sub-
optimal, i.e. when pigs are exposed to social stressors. Consequently, it is
assumed in the model that there is a positive correlation between BW0 and
AB. Individual values for AB (ABi) are generated around the assigned
genotype mean (mAB) and standard deviation (sAB) of AB, whilst being
positively correlated to BW0.
ABi = mAB + b4.((BW0i / mBW0).(sAB.(mBW0 / sBW0))) ± residuali (4.10)
The parameter b4 determines the degree of correlation between BW0 and
AB and is set equal to one. The residuali is drawn at random taking
account of sAB. Within a population, AB is not directly correlated to
leanness. However, leaner animals will tend to have lower AB values due to
the positive correlations between Lm/Pm and BW0 and BW0 and AB.
performance, ‘good’ and ‘poor’. Simulations for pigs grown from 80 to 100
kg and either mixed or not on day 1 are shown in Table 4.4. As expected,
mixing led to a decrease in performance, with pigs having the poorest
ability to cope (AB = 1) displaying the largest decrease. For example, good
genotype pigs with AB values of 10, 5 and 1 were predicted to show a
decrease in ADG of 9, 20 and 35%, respectively, compared to the non-
mixed counterparts. Interestingly, poor genotype pigs with high (AB = 10)
and average (AB = 5) abilities to cope were both predicted to outperform
the good genotype pigs with a low ability to cope (AB = 1), reaching 100
kg 6 and 3 days earlier, respectively. This implies that pigs with the highest
potential for growth do not always outperform others and that an animal’s
response to stressors, i.e. its ability to cope, may be as important as its
growth potential, particularly when raised under commercial conditions.
Table 4.4. The effects of pig potential (ADGp) and ability to cope (AB) when exposed to
stressors on the time taken (t) to reach 100 kg, average daily gain (ADG), average daily feed
intake (ADFI) and feed conversion ratio (FCR) from a starting weight of 80 kg. The effect of
mixing pigs on day 1 of the simulation is also shown for the AB = 10 genotype.
ADGp AB Mix t (days) ADG (kg/day) ADFI (kg/day) FCR (kg/kg)
Gooda 10 No 22 0.93 2.51 1.58
10 Yes 24 0.84 2.37 2.82
5 Yes 28 0.74 2.22 3.00
1 Yes 34 0.60 2.01 3.35
Poorb 10 No 26 0.80 2.66 3.33
10 Yes 28 0.74 2.52 3.41
5 Yes 31 0.65 2.34 3.60
1 Yes 38 0.53 2.09 3.94
a Genetic growth parameters: B = 0.016 day1, Pm = 32 kg, Lm/Pm = 1.2 kg/kg.
b Genetic growth parameters: B = 0.011 day1, Pm = 30 kg, Lm/Pm = 2.0 kg/kg.
66 I.J. Wellock et al.
(a) (b)
1.7 2.50
1.6 2.48
ADFI (kg/day)
ADFI (kg/day)
1.5
2.46
1.4
2.44
1.3
Fig. 4.5. Predicted effect of environmental stressors on the average daily feed intake (ADFI)
response of the average individual and population mean; (a) the effect of increasing group
size on the ADFI of pigs from 20 (± 2 kg) to 50 kg; (b) the effect of decreasing space
allowance on the ADFI of 100 kg (± 10 kg) pigs over a simulation period of one day.
Table 4.5. Effect of variation in initial BW (sBW0, kg) and ability to cope (sAB) on the
variation in the time taken (t, days) to reach 100 kg from a mean BW0 of 60 kg. P and L are
the protein and lipid content, respectively. Mixing occurred at 75 kg and pigs were given a
space allowance of 0.7 m2/pig throughout (adapted from Wellock et al., 2004).
sBW0a sAB t (days) ADG (kg/day) ADFI (kg/day) P (kg) L (kg)
0.00 0.00 50.7 (3.93)b 0.801 (0.060)b 2.17 (0.104)b 17.78 (0.352)b 15.59 (1.534)b
2.07 0.00 50.8 (5.48) 0.804 (0.057) 2.17 (0.107) 17.78 (0.333) 15.59 (1.445)
4.19 0.00 51.5 (7.64) 0.802 (0.056) 2.17 (0.119) 17.78 (0.357) 15.59 (1.592)
5.74 0.00 51.5 (9.42) 0.801 (0.052) 2.16 (0.121) 17.78 (0.358) 15.58 (1.544)
7.87 0.00 52.2 (11.59) 0.801 (0.053) 2.16 (0.131) 17.77 (0.346) 15.63 (1.500)
10.04 0.00 52.2 (14.23) 0.795 (0.054) 2.15 (0.132) 17.78 (0.338) 15.60 (1.506)
12.32 0.00 54.1 (17.42) 0.792 (0.054) 2.14 (0.148) 17.78 (0.352) 15368 (1.548)
0.00 0.47 50.8 (4.06) 0.799 (0.061) 2.17 (0.105) 17.76 (0.350) 15.70 (1.548)
0.00 0.94 50.8 (4.16) 0.798 (0.063) 2.17 (0.108) 17.79 (0.359) 15.54 (1.580)
0.00 1.45 50.8 (4.71) 0.799 (0.072) 2.17 (0.117) 17.79 (0.342) 15.61 (1.495)
0.00 1.93 50.8 (5.14) 0.799 (0.079) 2.17 (0.128) 17.77 (0.349) 15.64 (1.579)
0.00 2.53 50.8 (5.52) 0.802 (0.087) 2.17 (0.140) 17.79 (0.354) 15.53 (1.576)
5.77 1.40 52.1 (11.48) 0.801 (0.071) 2.17 (0.161) 17.77 (0.331) 15.63 (1.435)
12.18 2.42 54.2 (21.05) 0.799 (0.080) 2.15 (0.222) 17.79 (0.338) 15.46 (1.479)
a Simulated values.
b Result of variation in growth potential only.
Frequency
Frequency
80 80
60 60
40 40
20 20
0 0
20 30 40 50 60 70 80 90 100 20 30 40 50 60 70 80 90 100
Time (days) Time (days)
Frequency
80 80
60 60
40 40
20 20
0 0
20 30 40 50 60 70 80 90 100 20 30 40 50 60 70 80 90 100
Time (days) Time (days)
Fig. 4.6. The effect of variation in initial BW (sBW0) and ability to cope with social stressors
(sAB) on the time taken to reach 100 kg from an initial mean BW of 60 kg (N = 500); (a) sAB
= 0, sBW0 = 0, i.e. variation in growth potential only; (b) sAB = 1.5, sBW0 = 0; (c) sAB = 0,
sBW0 = 6 kg, and (d) sAB = 6, sBW0 = 1.5 kg. Mixing occurred at 75 kg and pigs were given
a space allowance of 0.7 m2/pig throughout.
Schinckel (2002) with quail and Muir (1996) and Muir and Craig (1998)
with poultry also demonstrated that selection for desirable associate effects
within a group may be a means to select animals which are better adapted
to their rearing environment. Any genetic correlation between AB and the
growth parameters that can be evaluated could be included in the model by
incorporating the co-variation between the identified parameters and AB.
Quantifying the variation in AB may improve the rate of breeding for
a better ability to cope, as the amount of heritable variation determines the
degree of selection pressure able to be applied. If a parameter such as AB
was included in a selection index then individual pigs with both the
greatest growth potential and best ability to cope could be selected for. For
example, animal ‘a’ shown in Fig. 4.7 may be a better breeding prospect
than animal ‘b’ as it has the desirable properties of having a low Lm/Pm
value and high AB unlike animal ‘b’, which has a high Lm/Pm and low AB.
This would result in benefits for both welfare and production. If increased
growth rate and ability to cope are antagonistic, then trying to increase pig
performance achieved under excellent conditions, i.e. improving potential
Effects of Social Stressors on Performance of Growing Pigs 69
8
a
7
AB 4
b
1
0.0 0.5 1.0 1.5 2.0 2.5
Lm/Pm (kg/kg)
Fig. 4.7. The simulated correlation between leanness, as represented by the lipid to protein
ratio at maturity (Lm/Pm), and ability to cope when exposed to environmental stressors (AB).
Five hundred individuals were simulated with a population mean (± SD) of 5 (1) for AB and 1.2
(0.18) Lm/Pm. Animal ‘a’ may be a better breeding prospect than animal ‘b’ as it has the
desirable properties of having a low Lm/Pm value and high AB, unlike animal ‘b’ which has a
high Lm/Pm and low AB.
alone, may not prove to be the best selection strategy. It is likely that
improvements in the growth potential of the animals and in the
environment, particularly better biosecurity and vaccination, are required
in addition to improving pigs’ ability to cope.
It was assumed in the model that all animals were in good health and free
from exposure to infectious stressors throughout. Any response to
infectious stressors, such as an increase in resource requirements to cope
with the consequences of infection, acquire and express an immune
response, a change in the efficiency of energy utilization or a voluntary
reduction in feed intake (anorexia), all of which would result in a decrease
in performance, were ignored. In reality of course, pigs are continuously
exposed to many different kinds and intensities of infectious stressors.
These include pathogens and other harmful environmental components
that may trigger tissue injury or further infection, such as bites and
scratches from other individuals in the same pen.
The incorporation of infectious stressors into simulation models is an
important next step in the attempt to predict commercial pig performance
accurately. To include the effects of infectious stressors into a model in a
systematic way it is necessary first to do a number of things. The metabolic
load imposed by infectious stressors, i.e. increased nutrient requirements,
and the extent to which performance is decreased need to be quantified.
How animals allocate resources when exposed to infectious stressors, e.g.
cope with a pathogen challenge, needs investigating and the biological
mechanism responsible for the decrease in performance needs elucidating.
Two possible mechanisms may lead to the decrease in pig performance
observed when pigs are exposed to disease. These are either a decrease in
the pigs’ ability to attain their potential, as suggested by Schinckel et al.
(2003), or a direct decrease in appetite as suggested by Kyriazakis (2003).
There is also likely to be between-animal variation in immune response
and resilience, i.e. differences in the ability of individual pigs to cope and
perform during exposure to pathogens (see Kyriazakis and Sandberg,
Chapter 7, this volume). This should be accounted for in any future
modelling attempt, along with any potential interactions between stress
and disease susceptibility when such information becomes available.
Conclusion
Despite their importance, few attempts have been made to quantify the
effects of social stressors on pig performance and incorporate these effects
into a pig growth simulation model. Here we describe how the effects of
the major social stressors, i.e. group size, space allowance, mixing and
feeder space allowance, can be described by conceptual equations based on
the biology of the animal, quantified and incorporated into a more general
pig growth model. The adapted model allows the performance of both
individuals and populations of growing pigs differing in initial state,
growth potential and ability to cope with social stress when raised under
given dietary, physical and social environmental conditions to be explored
and, at least in principle, predicted.
72 I.J. Wellock et al.
Among the main outcomes of the model simulations are the following:
(i) allowance for population variance is crucial in making decisions as there
may be differences between the response of the population and the
average individual; (ii) improving management to minimize stressors and
decrease variation in initial state is an important factor in decreasing the
heterogeneity of a group, particularly in commercial production systems
where payment is based upon uniformity; and (iii) if growth rate and
ability to cope when exposed to social stressors are antagonistic, trying to
improve pig performance by increasing growth potential alone may not be
the best selection strategy.
Acknowledgements
References
Arey, D.S. and Edwards, S.A. (1998) Factors influencing aggression between sows
after mixing and the consequences for welfare and production. Livestock
Production Science 56, 61–70.
Beilharz, R.G. and Cox, D.F. (1967) Social dominance in swine. Animal Behaviour
15, 117–122.
Black, J.L. (2002) Pig growth models future. Journal of Animal Science 80 (Suppl. 1),
174 (Abstr.).
Black, J.L., Campbell, R.G., Williams, I.H., James, K.J. and Davies, G.T. (1986)
Simulation of energy and amino acid utilisation in the pig. Research and
Development in Agriculture 3, 121–145.
Bridges, T.C., Turner, L.W., Stahly, T.S., Usry, J.L. and Loewer, O.J. (1992)
Modelling the physiological growth of swine. Part I: Model logic and growth
concepts. Transactions of the American Association of Agricultural Engineers 35,
1019–1028.
Bruce, J.M. and Clark, J.J. (1979) Models of heat production and critical
temperature for growing pigs. Animal Production 28, 353–369.
Chapple, R.P. (1993) Effect of stocking arrangement on pig performance. In:
Batterham, E.S. (ed.) Manipulating Pig Production IV. Australian Pig Science
Association, Victoria, pp. 87–97.
Curnow, R.N. (1986) The statistical approach to nutrient requirements. In: Fisher,
C. and Boorman, K.N. (eds) Nutrient Requirements of Poultry and Nutritional
Research. Butterworths, London, pp. 79–89.
D’Eath, R.B. (2002) Individual aggressiveness measured in a resident-intruder test
predicts the persistence of aggressive behaviour and weight gain of young pigs
after mixing. Applied Animal Behaviour Science 77, 267–283.
de Greef, K.H., Kanis, E., Van der Belt, H., Groen, A.F. and Schakel, J. (2003)
Breeding for improved welfare of pigs, a conceptual framework. Proceedings of
the 7th World Congress on Genetics Applied in Livestock Production 32, 39–42.
Effects of Social Stressors on Performance of Growing Pigs 73
Drickamer, L.C., Arthur, R.D. and Rosenthal, T.L. (1999) Predictors of social
dominance and aggression in gilts. Applied Animal Behaviour Science 63,
121–129.
Edmonds, M.S., Arenston, B.E. and Mente, G.A. (1998) Effect of protein levels and
space allocation on performance of growing-finishing pigs. Journal of Animal
Science 76, 814–821.
Edwards, S.A., Armsby, A.W. and Spechter, H.H. (1988) Effects of floor area
allowance on performance of growing pigs kept on fully slatted floors. Animal
Production 46, 453–459.
Elsasser, T.H., Klasing, K.C., Filipov, N. and Thompson, F. (2000) Metabolic
consequences of stress. In: Moberg, G.P. and Mench, J.A. (eds) The Biology of
Animal Stress. Basic Principles and Implications for Animal Welfare. CABI
Publishing, Wallingford, UK, pp. 77–110.
Emmans, G.C. and Fisher, C. (1986) Problems in nutritional theory. In: Fisher, C.
and Boorman, K.N. (eds) Nutrient Requirements of Poultry and Nutritional
Research. Butterworths, London, pp. 9–39.
Ferguson, N.S., Gous, R.M. and Emmans, G.C. (1994) Preferred components for
the construction of a new simulation model of growth, feed intake and nutrient
requirements of growing pigs. South African Journal of Animal Science 24, 10–17.
Ferguson, N.S., Gous, R.M. and Emmans, G.C. (1997) Predicting the effects of
animal variation on growth and feed intake in growing pigs using simulation
modelling. Animal Science 64, 513–522.
Ferguson, N.S., Lavers, G. and Gous, R.M. (2001) The effect of stocking density on
the response of growing pigs to dietary lysine. Animal Science 73, 459–469.
Fisher, C., Morris, T.R. and Jennings, R.G. (1973) A model for the description and
prediction of the response of laying hens to amino acid intake. British Poultry
Science 14, 469–484.
Gonyou, H.W., Chapple, R.P. and Frank, G.R. (1992) Productivity, time budgets
and social aspects of eating in pigs penned in groups of five or individually.
Applied Animal Behaviour Science 34, 291–301.
Grandin, T. (1994) Solving livestock handling problems. Veterinary Medicine 89,
989–998.
Griffing, B. (1966) Selection in reference to biological groups I. Individual and
group selection applied to populations of unordered groups. Australian Journal
of Biological Science 20, 127–139.
Hamilton, D.N., Ellis, M., Wolter, B.F., Schinckel, A.P. and Wilson, E.R. (2003) The
growth performance of the progeny of two swine sire lines reared under
different floor space allowances. Journal of Animal Science 81, 1126–1135.
Heetkamp, M.J.W., Schrama, J.W., De Jong, L., Swinkels, J.W.G.M., Schouten,
W.G.P. and Bosch, M.W. (1995) Energy metabolism in young pigs affected by
mixing. Journal of Animal Science 73, 3562–3569.
Heetkamp, M.J.W., Schrama, J.W., Schouten, W.G.P. and Swinkels, J.W.G.M. (2002)
Energy metabolism in young pigs as affected by establishment of new groups
prior to transport. Journal of Animal Physiology 86, 144–152.
Hessing, M.J.C., Schouten, W.G.P., Wiepkema, P.R. and Tielen, M.J.M. (1994)
Implications of individual behavioural characteristics on performance in pigs.
Livestock Production Science 40, 187–196.
Hyun, Y. and Ellis, M. (2001) Effect of group size and feeder type on growth
performance and feeding patterns in growing pigs. Journal of Animal Science 79,
803–810.
Hyun, Y., Ellis, M., Riskowski, G. and Johnson, R.W. (1998a) Growth performance
74 I.J. Wellock et al.
Introduction
Feed Predicted
specification performance
Optimizer
Fig. 5.1. Flow of information in optimizing the feeding programme of a broiler chicken.
criteria mentioned above. The broiler growth model allows for multiple
harvesting from one flock and calculates revenues from any mixture of
whole-bird sales or processing. Typical economic variables are included,
although these may be readily customized to fit with individual enterprises.
Growth models developed in different laboratories may be included in
optimization schemes of the sort described above. The key to this approach
clearly lies in the ability of the broiler growth model to reflect accurately
the performance expected under commercial conditions. However, the
performance of the average individual in the flock, subjected to average
conditions in the broiler house and being fed the average food formulated
by the nutritionist, without considering the variation inherent in genotype,
environment and food, cannot accurately reflect the variation that would
result in commercial conditions. For this, a stochastic model of growth and
food intake is required.
(growth rate) of each animal at a time, but because each bird is growing
and food intake is increasing, variation in growth rate is also introduced
within each animal over time (Gous and Morris, 1985), making the need
for a population model of broiler growth even more important (or critical)
than in the case of laying hens, which are in a relatively steady state. A
population model should be built by simulating the responses of many
individuals, and not directly as a group, so that the accepted assumptions
about marginal efficiencies are not compromised.
But it is not only the variation between individuals in their response to
a given feed or environment that controls the variation in the response of a
population of birds to a feeding programme in a given environment.
Variation also exists in the environmental conditions to which the birds are
subjected, as well as the composition of the feed used. Each of these
sources of variation will be addressed in the following review, but emphasis
will be placed on the effect of variation at the level of the genotype on the
response, as this aspect has been used in numerous exercises by the
authors to illustrate the difference in response between an individual and a
population of broilers.
Describing an individual
Describing a population
contents with the use of the ‘Poultry Growth Model’ of Emmans (1981),
and then integrating the bivariate probability distribution for different
degrees of variation in nutrient content over the response surface to
determine the value of reducing the variance in the feed. This approach,
although interesting, does not simulate the day-to-day variation in feed
nutrient content to which broilers are subjected in practice, where nutrient
composition varies continuously throughout the rearing period. A more
useful technique would be to predict the response of each bird in the
population to random variations in feed quality on each day of the
growing period, thereby obtaining a more realistic assessment of the effect
of variation in feed quality. This would require a stochastic approach to the
definition of the composition of the feed offered to the broilers, as well as
to the broilers themselves.
Such an approach is a useful tool for determining the upper limit to
the value of variance reduction, but does not address the issue of further
systematic variation in nutrient content brought about by separation
during road, rail, auger, blower or chain transportation. Such variation
tends to separate the fine from the coarse particles, particularly when the
feed is conveyed in a mash form, or when the quality of pellets is poor,
resulting in a high proportion of fines. This problem does not appear to be
addressed in the literature, so the variation introduced in this way is
unknown, but could be considerable. This variation is likely to be
systematic, in that separation would take place along the length of the
feeder lines, with fines being left behind as the coarser particles move to
the end of the line. If the fines consisted predominantly of major and
minor minerals and vitamins, bone development could be seen to worsen
at the far end of a broiler house compared with the end at which the feed
is introduced. The effect of such nutrient separation could be modelled in
the same way as are the systematic changes that take place in the
environment within a broiler house, the birds along the length of the
house being subjected to changes in temperature, humidity and air quality.
But it is not only the chemical separation of the food that takes place that
influences performance along the length of a broiler house; the physical
nature of the feed may also be affected.
Pelleted feeds are known to improve feed conversion efficiency when
compared with mash feeds (Jensen et al., 1962), this being due to the
shorter period of time spent by the broilers consuming pellets, resulting in
lower energy expenditure. The physical nature of the feed therefore has
an effect on performance and, because of the abrasive nature of some
feeding equipment used in broiler houses, the amount of time spent
eating, and hence the energy expended in eating, may well increase as the
distance of the feed trough from the hopper increases.
If the performance of broilers subjected to variation in the chemical
and physical nature of the feed offered to them could be accurately
simulated, a cost benefit analysis could be conducted to determine to what
extent the reduction in such variability is worth pursuing. Only with the
aid of stochastic programming is such an analysis possible.
Modelling Populations for Purposes of Optimization 85
the required mean, and the distribution of values about the mean reflects
the required standard deviation. The parameter values for each individual
may be generated by means of random numbers. A random number is a
number chosen as if by chance from some specified distribution such that
the selection of a large set of these numbers reproduces that distribution.
The subject of random number generation and testing is reviewed
extensively by Knuth (1997) and Hellekalek (2004). Many non-uniform
random number generators are available, even in such accessible
programmes as Excel and Minitab, whilst the source code for generating
these numbers in C++ is available free on the Internet. These
programmes generate random data from a normal distribution, given the
mean and standard deviation of the variate. It is also possible
simultaneously to generate values of correlated variates.
(m1.5d) 4 10 4
m 10 25 10
(m +1.5d) 4 10 4
88 R.M. Gous and E.T. Berhe
36
–12 –20
–24
–40
–36
–20–10 0 10 20–20 –10 0 10 20 –20–10 0 10 20–20 –10 0 10 20
Parameter value (deviation from mean %) Parameter value (deviation from mean %)
Fig. 5.2. The relative effect on food intake in male and female broilers fed a lysine-limiting
feed containing 9 (A, C) or 16 g lysine/kg (B, D) from 7 to 21 days (left) or 7 (E, G) or 11 g
lysine/kg (F, H) from 22 to 35 days (right) when the means of each of six genetic parameters
were increased or decreased by 0.05, 0.1, 0.15 or 0.20 whilst holding the five remaining
genetic parameters constant (B = .-䉲.-; Fr = ..-䊊..-; LPRm = ..䊏..; MLG = --䊉--; Pm = —ⵧ—;
and Wo = —䉮—).
40 40
CV = 0.0 CV = 0.05
30 30
Frequency
Frequency
20 20
10 10
0 0
1400 1600 1800 2000 2200 2400 2600 1400 1600 1800 2000 2200 2400 2600
Body weight (g) Body weight (g)
40 40
CV = 0.10 CV = 0.15
30 30
Frequency
Frequency
20 20
10 10
0 0
1400 1600 1800 2000 2200 2400 2600 1400 1600 1800 2000 2200 2400 2600
Body weight (g) Body weight (g)
Fig. 5.3. Effect of increasing the coefficient of variation (CV) of feathering rate on the
frequency distribution of body weights at 35 d of age in a simulated group of male broilers
given feeds designed to enable them to achieve their potential growth rate.
intake on feeds with the lowest lysine contents (Gous and Morris, 1985;
Burnham et al., 1992) is evident only at the lower environmental
temperatures. The efficiency of utilization of lysine for body protein
growth remains the same at all temperatures on the marginally deficient
feeds, so protein gain is the same for a given lysine intake; it is only on
feeds with the highest lysine contents, and at the higher environmental
temperatures, that protein growth rate is curtailed due to the rapid
feathering rate, resulting in a separation of the maximum protein growth
rates at high dietary lysine contents.
100 A 6
A
80 4
60 B 2
Protein gain (g/d)
Food intake (g/d)
120 9
B
100 6
80 3
160 C
12
140 9 C
120 6
100 3
4 6 8 10 12 14 16 400 600 800 1000 1200 1400 1600 1800
Lysine content (g/kg) Lysine intake (mg/bird day)
Fig. 5.4. The effect of variation in feathering rate (FR) on food intake (g/day) and body
protein gain (g/day) of female broilers from 22 to 35 days of age, fed lysine-limiting feeds at
environmental temperatures of 29°C (A), 25°C (B) and 21°C (C). Coefficients of variation
used were 0.0 (䊊—䊊), 0.05 (䉭---䉭), 0.10 (䊐-..-䊐) and 0.15 (*-.-*).
92 R.M. Gous and E.T. Berhe
Table 5.2. Optimum 35-day performance, and optimum lysine contents in feeds, of an
individual and a population of broiler females, where margin/m2 annum was maximized at five
coefficients of variation (CV) of feathering rate, whilst the CVs of W0, Pm, LPRm and MLG
remained the same, as given in the text. Two feeding programmes were used: fixed amounts
of each feed, or a fixed number of days on each feed. Populations of 100 birds were
generated afresh for each optimization.
CV of feathering rate
Indiv. 0.00 0.05 0.10 0.15 0.25
Liveweight, g/bird 1968 1969 1966 1885 1775 1630
Food intake, g/bird 2844 2852 2883 2728 2565 2344
Breast meat, g 344 345 343 329 307 279
Abdominal fat, g 47 46 49 41 37 31
Cost of feeding, relative 100 101 101 97 91 84
Margin over feeding cost 100 100 98 95 90 82
g/bird grower and the remainder finisher. The CVs of five of the genotypic
parameters were held constant (W0 = 0.1, Pm = 0.1, B* = 0.06, LPRm =
0.06 and MLG = 0.1) whilst FR was varied. A population of 100
individuals was generated afresh for each optimization.
Only one obvious trend emerged from this exercise, namely that the
optimum lysine content in the grower feed increased with variation in FR.
Interestingly, this happened also when a fixed number of days on each
feed was used in place of the fixed amount of each food (Table 5.2). The
inconsistent variation in optimum lysine contents (less when using a fixed
number of days in the feeding programme) probably reflects the variation
in the responses to different populations that were simulated for each
optimization. Generally, though, optimum lysine contents were the same in
the starter feeds, higher in the grower feeds, and lower in the finisher
feeds for populations than for the mean individual in the population.
The relatively small differences in amino acid contents of the three
feeds required to optimize performance of the mean individual and that of
the population are of more than passing interest, considering that the
mean performance of the population is much reduced when variation in
FR is high. Two precedents can be found to substantiate this observation.
The first is in Wethli and Morris (1978), where it was demonstrated that
the daily tryptophan required by a flock of laying hens does not decrease
Modelling Populations for Purposes of Optimization 93
during the first laying year, despite a decrease in mean rate of egg output.
The second is that when pigs are housed at high stocking rates, causing
reduced feed intakes and growth rates, efficiency of lysine utilization is
unaffected by the stress, and optimum biological performance is obtained
on feeds with lysine contents the same as those which optimize the
performance of individually housed pigs growing close to their potential
(B.A. Theeruth and R.M. Gous, 2005, unpublished results). It is not
unusual, therefore, for the optimum feeds to be unaffected by the reduced
mean performance of the animals in the population.
Conclusions
Two conclusions may be reached from these exercises, both of which relate
to the non-linear effect of variation in FR, which both increases the
variation in response within a population and decreases its mean
performance. The first is that, where a fixed feeding programme is being
used, the optimum amino acid content in the feeds used in the programme
differs only marginally for a population and for the mean individual in the
population; whereas, if proprietary feeds of fixed composition are being
used, almost twice as much of the starter and grower feeds is needed to
optimize the performance of a population of broilers compared with that
required for the mean individual in the population. The second conclusion
94 R.M. Gous and E.T. Berhe
References
Aerts, J.M., Berckmans, D., Saevels, P., Decuypere, E. and Buyse, J. (2000)
Modelling the static and dynamic responses of total heat production of broiler
chickens to step changes in air temperature and light intensity. British Poultry
Science 41, 651–659.
Al Homidan, A., Robertson, J.F. and Petchey, A.M. (1997) Effect of temperature,
litter and light intensity on ammonia and dust production and broiler
performance. British Poultry Science 38, S5–S6.
Al Homidan, A., Robertson, J.F. and Petchey, A.M. (1998) Effect of environmental
factors on ammonia and dust production and broiler performance. British
Poultry Science 39, S9–S10.
Berhe, E.T. (2003) Introducing stochasticity into a model of food intake and growth
of broilers. MScAgric thesis. University of KwaZulu-Natal, Pietermaritzburg,
South Africa.
Brody, S. (1945) Bioenergetics and Growth. Reinhold, New York.
Boshouwers, F.M.G., Develaar, F.G., Landman, W.J.M., Nicaise, E and Van Den
Bos, J. (1996) Vertical temperature profiles at bird level in broiler houses.
British Poultry Science 37, 55–62.
Burnham, D., Emmans, G.C. and Gous, R.M. (1992) Isoleucine responses in broiler
chickens. Interactions with leucine and valine. British Poultry Science 33, 71–87.
Modelling Populations for Purposes of Optimization 95
Cahaner, A., Druyan, S. and Deeb, N. (2003) Improving broiler meat production,
especially in hot climates, by genes that reduce or eliminate feather coverage.
British Poultry Science 44, S22–S23.
Dale, N. (1996) Variation in feed ingredient quality: oilseed meals. Animal Feed
Science Technology 59, 129–135.
Deming, W.E. (1986) Out of the Crisis. MIT Center for Advanced Engineering Study,
Cambridge, Massachusetts.
Duncan, M.S. (1988) Problems of dealing with raw ingredient variability. In:
Haresign, W. and Cole, D.J.A. (eds) Recent Advances in Animal Nutrition.
Butterworths, Boston, Massachusetts, pp. 3–11.
EFG Software Natal (1995) Advanced Computer Software for the Animal Industry.
http://www.efgsoftware.com/ (accessed February 2005).
Emmans, G.C. (1981) A model of the growth and feed intake of ad libitum fed
animals, particularly poultry. In: Hillyer, G.M., Whittemore, C.T. and Gunn,
R.G. (eds) Computers in Animal Production. Occasional Publication No.5. British
Society of Animal Production, Edinburgh, UK, pp. 103–110.
Emmans, G.C. (1987) Growth, body composition, and feed intake. World’s Poultry
Science Journal 43, 208–227.
Emmans, G.C. and Fisher, C. (1986) Problems of nutritional theory. In: Fisher, C.
and Boorman, K.N. (eds) Nutritional Requirements and Nutritional Theory.
Butterworths, London, pp. 9–39.
Emmans, G.C. and Oldham, J.D. (1988) Modelling of growth and nutrition in
different species. In: Karver, S. and Van Arendonk, J.A.M. (eds) Modelling of
Livestock Production Systems. Kluwer Academic, Dordrecht, Netherlands,
pp. 13–21.
Fawcett, R.H., Webster, M., Thornton, P.K., Roan, S.W. and Morgan, C.A. (1992)
Predicting the response to variation in diet composition. In: Recent Advances in
Animal Nutrition. Butterworths, Boston, Massachusetts, pp. 137–158.
Fisher, C., Morris, T.R. and Jennings, R.C. (1973) A model for the description and
prediction of the response of laying hens to amino acid intake. British Poultry
Science 14, 469–484.
Gous, R.M. and Morris, T.R. (1985) Evaluation of a diet dilution technique for
measuring the response of broiler chickens to increasing concentrations of
lysine. British Poultry Science 26, 147–161.
Gous, R.M., Pym, R.A.E., Mannion, P. and Wu, J.X. (1996) An evaluation of the
parameters of the Gompertz growth equation that describe the growth of eight
strains of broiler. In: Balnave, D. (ed.) Australian Poultry Science Symposium, Vol.
8. University of Sydney, Sydney, New South Wales, pp. 174–177.
Gous, R.M., Moran, E.T. Jr, Stilborn, H.R., Bradford, G.D. and Emmans, G.C.
(1999) Evaluation of the parameters needed to describe the overall growth, the
chemical growth and the growth of feathers and breast muscles of broilers.
Poultry Science 78, 812–821.
Hancock, C.E., Bradford, G.D., Emmans, G.C. and Gous, R.M. (1995) The
evaluation of the growth parameters of six strains of commercial broiler
chickens. British Poultry Science 36, 247–264.
Hellekalek, P. (2004) PLab. Theory and Practice of Random Number Generation.
http://random.mat.sbg.ac.at/ (accessed February 2005).
Hughes, R.J. and Choct, M. (1999) Chemical and physical characteristics of grains
related to variability in energy and amino acid availability in poultry. Australian
Journal of Agricultural Research 50, 689–701.
Jensen, L.S., Merrill, L.H., Reddy, C.V. and McGinnis, J. (1962) Observations on
96 R.M. Gous and E.T. Berhe
eating patterns and rate of food passage of birds fed pelleted and unpelleted
diets. Poultry Science 41, 1414–1419.
Knap, P.W. (1995) Aspects of stochasticity: variation between animals. In: Moughan,
P.J., Verstegen, M.W.A. and Visser-Reyneveld, M.I. (eds) Modelling Growth in the
Pig. EAAP publication no 78, Wageningen Agricultural University,
Wageningen, Netherlands, pp.165–172.
Knuth, D.E. (1997) The Art of Computer Programming: Seminumerical Algorithms, Vol.
2, 3rd edn. Addison-Wesley, Reading, Massachusetts, USA.
Mack, S., Hohler, D. and Pack, M. (2000) Evaluation of dose-response data and
implications for commercial formulation of broiler diets. In: Balnave, D. (ed.)
Australian Poultry Science Symposium, Vol. 12. University of Sydney, Sydney, New
South Wales, pp. 82–87.
McCoy, R.A., Behnke, K.C., Hancock, J.D. and McEllhiney, R.R. (1994) Effect of
mixing uniformity on broiler chick performance. Poultry Science 73, 443–451.
Metayer, J.P., Grosjean, F. and Casting, J. (1993) Study of variability in French
cereals. Animal Feed Science and Technology 43, 87–108.
Mitchell, M.A. (1985) Effects of air velocity on convective and radiant heat transfer
from domestic fowls at environmental temperatures 20°C and 30°C. British
Poultry Science 26, 413–423.
Morris, M.D. (1991) Factorial sampling plans for preliminary computational
experiments. Technometrics 33, 161–174.
Parmar, R.S., Diehl, K.C., Collins, E.R. and Hulet, R.M. (1992) Simulation of a
turkey house environment. Agricultural Systems 38, 425–445.
Reece, F.N. and Lott, B.D. (1982) The effect of environmental temperature on
sensible and latent heat production of broiler chickens. Poultry Science 61,
1590–1593.
Roush, W.B., Cravener, T.L. and Zhang, F. (1996) Computer formulation
observations and caveats. Journal of Applied Poultry Research 5, 116–125.
Senkoylu, N. and Dale, N. (1999) Sunflower meal in poultry diets: a review. World
Poultry Science Journal 55, 153–174.
Singh, P. and Trehan, P.K. (2002) Inheritance of rate of feathering and feather
density score and its relationship with body weight in broiler chicken. SARAS
Journal of Livestock and Poultry Production 18, 41–47.
Taylor, St. C.S. (1980) Genetic size scaling rules in animal growth. Animal Production
30, 161–165.
Van Beek G. and Beeking, F.F.E. (1995) A simple steady state model of the
distribution of vertical temperature in broiler houses without internal air
circulation. British Poultry Science 36, 341–356.
Wathes, C.M. and Clark, J.A. (1981) Sensible heat transfer from the fowl: radiative
and convective heat losses from a flock of broiler chickens. British Poultry Science
22, 185–196.
Weaver, W.D., Jr and Meijerhof, R. (1991) The effect of different levels of relative
humidity and air movement on litter conditions, ammonia levels, growth and
carcass quality for broiler chickens. Poultry Science 70, 746–755.
Wethli, E. and Morris, T.R. (1978) Effects of age on the tryptophan requirement of
laying hens. British Poultry Science 19, 559–565.
Whittemore, C.T. and Fawcett, R.H. (1976) Theoretical aspects of a flexible model
to simulate protein and lipid growth in pigs. Animal Production 22, 87–96.
Xin, H., Berry, I.L., Tabler, G.T. and Barton, T.L. (1994) Temperature and
humidity profiles of broiler houses with experimental, conventional and
tunnel ventilation systems. Applied Engineering in Agriculture 10, 535–542.
6 Advancements in Empirical
Models for Prediction and
Prescription
W.B. ROUSH
USDA-ARS Poultry Research Unit, Mississippi State, MS 39762, USA
BRoush@msa-msstate.ars.usda.gov
From the beginning of efforts to model systems – long before the computer
era – linearity has dominated computation. This is not because anyone
believed that the systems of interest were truly linear, but simply for reasons of
computational tractability.
(Simon, 1990)
Nothing in Nature is random … A thing appears random only through the
incompleteness of our knowledge.
(Spinoza; quoted by Peitgen et al., 1992a, p. 319)
Introduction
Brown and Rothery (1994) define mechanistic models as concerned with the
‘nuts and bolts’ of biological processes and the way in which the
component parts fit together. They attempt to describe the observations in
terms of fundamental postulates about the biological processes. Theodorou
and France (2000) have noted that mechanistic models are constructed by
looking at the structure of the whole system, dividing it into its key
components and analysing the behaviour of the whole system in terms of
its individual components and their interactions with one another.
In contrast, empirical models provide quantitative descriptions of
patterns in the observations without attempting to describe the underlying
processes or mechanisms involved. In a sense, all models are empirical,
differing only in the degree of resolution and level of complexity.
Nevertheless, the broad distinction between mechanistic and empirical
models is a useful one (Brown and Rothery, 1994).
Theodorou and France (2000) have noted that the accuracy of
prediction of animal response using mechanistic models currently may be
lower than that achieved by the empirical methods used in practical
application. However, these research models are very useful in evaluating
the adequacy of current knowledge and data, identifying those areas
where research should be focused.
Empirical models are commonly referred to as black boxes. That is,
their mathematical workings are not transparent. Mechanistic models, at
the other extreme, are white boxes (or at least off-white boxes). The goal is
to have mechanistic models in which the workings are transparent. Artificial
neural networks (to be discussed later) are the ultimate empirical black box.
Recent research has suggested a merger between empirical and
mechanistic models. The merging of the two methods of modelling results
in a grey-box model. The grey-box model is intended to blend the better of two
worlds: knowledge-based modelling and black-box modelling (Oussar and
Dreyfus, 2001).
Two of the tenets of classical science are order and precision. However
nature is not always orderly and precise. There is much variability and
imprecision. Casti (1994) comments:
… one of the great challenges to both science and philosophy is to provide a
rational, coherent account of the perceived uncertainty surrounding the events
of daily life. Classical probability theory offers one such approach but is riddled
with many well-known epistemological flaws and paradoxes. The theories of
fuzzy sets, satisfying and possibilities represent recent attempts to rectify some
of the deficiencies in the classical methods. Each of these newer schemes has at
its heart the basic fact that randomness is only one face of the mask of
uncertainty.
Advancements in Empirical Models for Prediction and Prescription 99
The stochastic and fuzzy concepts change the picture of the accurate
and precise answer of a deterministic model to a model that produces
answers that are based on probability and possibility. This presents a
challenge to the researcher in classical modelling and decision making. As
Zimmerman (1992) points out:
… until the 1960s, uncertainty, vagueness, and inexactness were features with
rather negative meanings. Nobody wanted to be called a ‘vague decision
maker’; a scientist that could not make precise and definite statements was not
regarded as a ‘true’ scientist, and uncertainty was considered to be something
disturbing that should, if at all possible, be avoided in models, theories, and
statements. The only theory that dealt with uncertainty was probability theory,
and this – predominately in its frequentative interpretation – was restricted to
situations in which the law of large numbers was valid and uncertainty could
be attributed to randomness.
Recent discoveries in nonlinear dynamics (chaos theory) further
complicate the matter of uncertainty by calling into question the nature of
randomness (Peitgen et al., 1992a).
red? From the fuzzy logic point of view the nanometres of light
representing red would be full membership in a red set and the
nanometres of light representing yellow would be full membership in
the yellow set. Nanometres representing colours between red and yellow
would have partial memberships in the red and yellow sets. Half way
between the red and yellow are the nanometres representing orange
which would have 0.5 membership in the red set and 0.5 membership in
the yellow set. Other imprecise concepts such as hot and cold, light and
heavy, short and tall, etc. can be represented by fuzzy logic in a similar
manner. Roush et al. (1989) and Roush and Cravener (1990) used fuzzy
logic to describe the imprecise concept of stress in a caged laying
situation. Fuzzy sets have been applied to the imprecision of human
nutrition and nutritional requirements by Wirsam and Uthus (1996),
Wirsam et al. (1997) and Gedrich et al. (1999).
There have been interesting fuzzy logic extensions to modelling of
control systems (Kosko, 1992) and simulations of social interaction of fish,
using fuzzy cognitive maps (Dickerson and Kosko, 1997).
Aristotle (c. 330 BC) pointed out that the ‘whole is greater than the sum of
its parts’. This is very evident in agricultural systems in which the response
of the organism results from the interaction of numerous inputs. Biological
and environmental inputs and the resulting outputs are not necessarily
additive or linear.
Sir Isaac Newton, early in his education, was a student of Aristotle’s
philosophy. However, Newton’s views of the world changed as he was later
influenced by the works of René Descartes and other mechanical
philosophers. The mechanical philosophers, in contrast to Aristotle, viewed
the world as composed entirely of particles of matter in motion and held
that all phenomena of nature result from their mechanical interaction
(Encyclopædia Britannica, 15th edn).
As a result, Newton and Descartes were advocates of a universe that
operates like clockwork, where everything is very orderly and mechanical.
The assumption was that if enough is known about a system (universe),
there is nothing that cannot be predicted for that system (universe). Under
this philosophy, the universe was a gigantic complicated clockwork
mechanism. The logic makes perfect sense that, if we understand each and
every part of the machine, then we can predict how the machine will act
and react. By taking apart the clock and studying each gear, an
understanding of the workings of the clock can be developed. Hence the
development of a reductionist approach to science.
Modern biology has inherited the reductionist approach. The living
organism is viewed as a complex biochemical machine. Examination is
made of the organs, tissues, cells and even the molecules in an effort to
define the mechanisms of life. Although the mechanistic view is desirable, it
Advancements in Empirical Models for Prediction and Prescription 101
has been recognized that the ideal model involves, in addition to defining
mechanisms, the inclusion of stochastic and dynamic concepts. Models
need to contain all three elements. The current focus has been mainly on
mechanistic models. There is a caveat that the exclusion of the dynamic
and stochastic elements in modelling promotes an illusion of precision in
attaining answers.
Prediction models
Factorial statistical models are known by most students who have taken a
graduate course in statistics. Usually the factorial model is analysed
104 W.B. Roush
Mixture designs are a type of response surface design that have application
to problems in the animal sciences. In the general mixture problem, the
Advancements in Empirical Models for Prediction and Prescription 105
Gous and Swatson (2000) have used mixture experiments to study the
ability of the broiler to choose from three protein sources the combination
of ingredients that would maximize biological performance. Roush et al.
(2004) suggested that the optimal time to feed broilers the starter, grower
and finisher feeds could be viewed as a mixture problem with the objective
of finding the optimal proportion of time to feed each diet.
decrease. There will be a point at which the errors will start to increase
which is the point at which over-training has started to occur. That is the
point at which the neural network is saved. At this point, validation is made
of the new neural network on data that are independent of the training
and testing data sets. The second procedure for reducing over-training is
to use statistical Jack-Knife and bootstrap procedures, where the
experimental data are re-sampled in the process of development of the
neural network. In this way there is not a need for the test set.
The development of neural networks can incorporate training, testing
(to avoid over-training) and validation to make robust models. For
example Roush et al. (1996b) developed an artificial neural network to
predict the presence or absence of ascites in broilers. The neural network
was a three layer back propagation neural network with an input of 15
physiological variables. After developing the neural network with training
and test sets, the neural network predictive ability was validated with two
data sets that were not involved in the training. The neural network
accurately identified two false positives and one false positive in the first
and second evaluation data sets, respectively. The birds identified as false
positives were actually determined to be in the developmental stages of
ascites.
There are many different types of neural networks. These different
types can be generally classified as supervised and unsupervised networks.
In supervised learning, the neural network learns from an example. With
unsupervised learning, the neural network examines the data to define
clusters of information. The neural network is used to associate data,
classify data, transform data into a different representation and to model
data (Zupan and Gasteiger, 1993). The commercial neural network
package NeuralShell 2 (Ward Systems Group, 1996) contains 16 different
types of neural network, which include the following:
1. Backpropagation neural networks. This neural network is the standard.
Usually three layers are sufficient. The layers are the input, hidden and
output layers. Each layer is linked only to the previous layer.
2. Jump connection neural network. This type of backpropagation
network has every layer connected to every previous layer.
3. Recurrent network. This is a type of backpropagation neural network.
There is feedback to previous layers. These networks are often used for
time series data. Regular feed forward neural networks respond to a given
input with the same output each time. A recurrent network may respond
to the same input pattern differently from time to time, depending upon
the input patterns previously presented to it. The recurrent network
builds a long term memory based on the patterns presented.
4. Kohonen architecture. This is an unsupervised neural network. It is
able to learn without being shown correct output patterns. The use of this
type of network is for clustering problems. The network is able to separate
data into a specified number of groups or categories.
5. Probabalistic neural network (PNN). This is a powerful neural network
Advancements in Empirical Models for Prediction and Prescription 107
Genetic algorithms
Genetic algorithms are search procedures that use the principles of natural
selection and genetics. The genetic algorithm was first developed by John
H. Holland in the 1960s. The search procedure is usually looking for an
optimum condition. The model to be optimized can be a formula or even a
neural network in which the maximum, minimum or a particular value is
required. The genetic algorithm works particularly well with problems that
are ‘not well behaved’. That is, situations where it may be difficult to find
the global optimum.
Commercial neural networks and genetic algorithms are available that
can be incorporated into a spreadsheet. The setup for the genetic
algorithm is based on an objective equation and constraints similar to the
setup of a linear program.
Prescription Models
Linear programming
Since its inception in 1947, linear programming has been the workhorse of
decision-making algorithms. Numerous texts and applications have been
written about its use. Linear programs have been used for blending (e.g.
petroleum products), mixes (e.g. investments and budgeting), scheduling
(e.g. production to satisfy customer demand, production capacity, and
storage limitations), assignment (e.g. workers to tasks) and
transportation/dispatching (e.g. routing of pick up and deliveries). In the
animal sciences, the term ‘linear programming’ is considered by many as
synonymous with the mixing problem of feed formulation. The Sadia
company, the largest broiler producer in Brazil, used linear programming
and other operations research methods effectively to improve decision
making about production and product distribution in their business
(Taube-Netto, 1996).
The linear program consists of an objective equation and constraint
equations. For example in a feed formulation problem the objective is to
minimize the cost of ingredients subject to meeting the nutritional
constraints. The following is an example:
Objective equation:
Minimize cost: 0.08 Maize + 0.20 Soybean
Constraint equations:
87 Maize + 488 Soybean ≥ 230 (protein constraint)
Maize + Soybean = 1 (amount constraint)
where 0.08 and 0.20 represent the price ($/kg) of maize and soybean and
87 and 488 represent the protein contents (g/kg) of these ingredients. The
objective is to minimize the cost of the diet, subject to the constraints that
the protein supplied by the maize and soybean together must be ≥ 230
g/kg diet (the requirement of the animal) and the fractional amounts of
maize and soybean must total to 1.
Mathematically there are certain assumptions made about linear
programming (Render and Stair, 1982; Roush et al., 1996a):
Advancements in Empirical Models for Prediction and Prescription 109
9 + 16 =7 (6.1)
and
9+16 = 5 (6.2)
110 W.B. Roush
Quadratic programming
Miller et al. (1986) and Pesti et al. (1986) combined broiler growth
equations obtained using response surface methodology with quadratic
programming. The result was that they were able to demonstrate that a
quadratic programming model would provide a method of ration
formulation that would take into account the productivity of the broiler.
They did this by defining a quadratic objective as the growth response to
intake of protein and energy. Live weight, transformed to feed input space,
was maximized subject to a given cost per bird and other common
constraints in linear programming of a feed mix. Linear programming
(LP) and quadratic programming (QP) results were compared. Pesti et al.
(1986) reported the energy concentrations of the diets were similar by both
methods of formulation (LP = 13.62 and QP = 13.232 MJ/kg). However,
Advancements in Empirical Models for Prediction and Prescription 111
the protein content was much higher when the QP method was used (LP
= 217 and QP = 244 g/kg). Miller et al. (1986) reported that analysis using
QP indicated that a leading broiler firm could have improved economic
efficiency by increasing protein density and (slightly) reducing energy
density of broiler finisher feed. Further if this was applied industry wide,
the savings would be US$120 million per year.
Decision analysis
Conclusion
The study of chaos, its implications and how it occurs, is a hot topic in
mathematics and physics. The dynamic results are not tricks.
Developments in empirical modelling are constantly expanding.
Nonlinear mathematics, Artificial Intelligence and a relatively new field,
Artificial Life (Levy, 1992), are areas where, in the opinion of the author,
there will be important melding of empirical and mechanistic modelling.
References
Aristotle (c.330 BC) Metaphysica 1045a, 10 f.
Beveridge, W.I.B. (1957) The Art of Scientific Investigation. Random House, New York.
Box, G.E.P. and Draper, N.R. (1987) Empirical Model-Building and Response Surfaces.
John Wiley, New York.
Box, G.E.P. and Wilson, K.B. (1951) On the experimental attainment of optimum
conditions (with discussion). Journal of the Royal Statistical Society B 13, 1–45.
Brown, D. and Rothery P. (1994) Models in Biology: Mathematics, Statistics and
Computing. John Wiley, New York.
Casti, J.L. (1989) Alternate Realities: Mathematical Models of Nature and Man. John
Wiley, New York.
Casti, J.L. (1994) Complexification: Explaining a Paradoxical World through the Science of
Surprise. Harper Collins, New York.
Cochran, W.G. and Cox, G.M. (1957) Experimental Designs, 2nd edn. John Wiley,
New York.
Cornell, J. (2002) Experiments with Mixtures: Designs, Models, and the Analysis of Mixture
Data, 3rd edn. John Wiley, New York.
Cross, S., Harrison, R.F. and Kennedy, R.L. (1995) Introduction to neural
networks. Lancet 346, 1075–1079.
Degn, H., Holden, A.V. and Olsen, L. F. (eds) (1986) Chaos in Biological Systems.
NATO ASI Series, Series A: Life Sciences Vol. 138. Plenum Press, New York.
Deming, W.E. (2000) Out of the Crisis. MIT Press, Cambridge, Massachusetts.
Dickerson, J.A. and Kosko, B. (1997) Virtual worlds in fuzzy cognitive maps. In:
Kosko, B. (ed.) Fuzzy Engineering. Prentice-Hall, Upper Saddle River, New
Jersey.
Encyclopædia Britannica, 15th edn, s.v. “Newton, Sir Isaac”.
Advancements in Empirical Models for Prediction and Prescription 113
Myers, R.H. and Montgomery, D.C. (2002) Response Surface Methodology, 2nd edn.
John Wiley, New York.
Nott, H. and Combs, G.F. (1967) Data processing of ingredient composition data.
Feedstuffs 14, 21.
Oberstone, J. (1990) Management Science: Concepts, Insights and Applications. West
Publishing, New York.
Oltjen, J.W. and Owens, F.N. (1987) Beef cattle feed intake and growth: Empirical
Bayes derivation of the Kalman filter applied to a nonlinear dynamic model.
Journal of Animal Science 65, 1362–1370.
Oussar, Y. and Dreyfus, G. (2001) How to be a gray box: dynamic semi-physical
modelling. Neural Networks 14, 1161–1172.
Parks, J.R. (1982) A Theory of Feeding and Growth of Animals. Springer, Berlin.
Peitgen, H-O., Jürgens, H. and Saupe, D. (1992a) Fractals for the Classroom. Part One:
Introduction to Fractals and Chaos. Springer, New York.
Peitgen, H-O., Jürgens, H. and Saupe, D. (1992b) Fractals for the Classroom. Part Two:
Complex Systems and Mandelbrot Set. Springer, New York.
Pesti, G.M., Arraes, R.A. and Miller, B.R. (1986) Use of quadratic growth response
to dietary energy and protein concentrations in least-cost feed formulation.
Poultry Science 65, 1040–1051.
Piepel, G.F. and Cornell, J.A. (1985) Models for mixture experiments when the
response depends on the total amount. Technometrics 27, 219–227.
Rehman, T. and Romero, C. (1984) Multiple-criteria decision making techniques
and their role in livestock ration formulation. Agricultural Systems 15, 23–49.
Rehman, T. and Romero, C. (1987) Goal programming with penalty functions and
livestock ration formulation. Agricultural Systems 23, 117–132.
Render, B. and Stair, R.M. (1982) Quantitative Analysis for Management. Allyn and
Bacon, Boston, Massachusetts.
Roush, W.B. (1983) An investigation of the protein levels for broiler starter and
finisher rations and the time of ration change by response surface
methodology. Poultry Science 62, 110–116.
Roush, W.B. (1986) A decision analysis approach to the determination of
population density in laying cages. World’s Poultry Science Journal 42, 26–31.
Roush, W.B. (2001a) Stochastic programming: Implications for precision feeding.
Feedstuffs July 2, 14–15.
Roush, W.B. (2001b) An operations research approach to leveraging new
technology opportunities. Journal of Applied Poultry Research 10, 293–302.
Roush, W.B. and Cravener, T.L. (1990) Evaluation of colony size and cage space for
laying hens (Gallus domesticus) using fuzzy decision analysis. Poultry Science 69,
1480–1484.
Roush, W.B. and Cravener, T.L. (1992) Choosing a commercial laying hen strain
based on multicriteria decision analysis. Journal of Applied Poultry Research 1,
403–409.
Roush, W.B. and Wideman, R.F., Jr (2000) Evaluation of broiler growth velocity
and acceleration in relation to pulmonary hypertension syndrome. Poultry
Science 79, 180–191.
Roush, W.B., Petersen, R.G. and Arscott, G.H. (1979) An application of response
surface methodology to research in poultry nutrition. Poultry Science 58,
1504–1513.
Roush, W.B., Mylet, M., Rosenberger, J.L. and Derr, J. (1986) Investigation of
calcium and available phosphorus requirements for laying hens by response
surface methodology. Poultry Science 65, 964–970.
Advancements in Empirical Models for Prediction and Prescription 115
Roush, W.B., Bock, R.G. and Marszalek, M.A. (1989) Evaluation of crowding of
laying hens (Gallus domesticus) using fuzzy set decision analysis. Applied Animal
Behaviour Science 23, 155–163.
Roush, W.B., Tomiyama, K., Garnaoui, K.H., D’Alfonso, T.H. and Cravener, T.L.
(1992) Kalman filter and an example of its use to detect changes in poultry
production responses. Computers and Electronics in Agriculture 6, 347–356.
Roush, W.B., Barbato, G.F. and Cravener, T.L. (1994) A nonlinear dynamical
(chaos) approach to the analysis of broiler growth. Poultry Science 73,
1183–1195.
Roush, W.B., Cravener, T.L. and Zhang, F. (1996a) Computer formulation
observations and caveats. Journal of Applied Poultry Research 5, 116–125.
Roush, W.B., Kochera Kerby, Y., Cravener, T.L. and Wideman, R.F., Jr (1996b)
Artificial neural network prediction of ascites in broilers. Poultry Science 75,
1479–1487.
Roush, W.B., Boykin, D. and Branton, S.L. (2004) Optimization of phase feeding of
starter, grower and finisher diets for male broilers by mixture experimental
design: Forty-eight day production period. Poultry Science 83, 1264–1275.
Simon, H.A. (1990) Prediction and prescription in systems modelling. Operations
Research 38, 7–14.
Smith, A.F.M. and Cook, D.G. (1980) Straight lines with a change-point: a Bayesian
analysis of some renal transplant data. Applied Statistics 29, 180–189.
Smith, A.F.M. and West, M. (1983) Monitoring renal transplants: An application of
the multiprocess Kalman filter. Biometrics 39, 867–878.
Snee, R.D. (1971) Design and analysis of mixture experiments. Journal of Quality
Control 3, 159–169.
Solé, R. and Goodwin, B. (2000) Signs of Life: How Complexity Pervades Biology. Basic
Books, New York.
Taube-Netto, M. (1996) Integrated planning for poultry production at Sadia.
Interfaces 26, 38–53.
Theodorou, M.K. and France, J. (eds) (2000) Feeding Systems and Feed Evaluation
Models. CABI Publishing, New York.
Trimble, I.M., West, M., Knapp, M.S., Pownall, R. and Smith, A.F.M. (1983)
Detection of renal allograft rejection by computer. British Medical Journal 286,
1695–1699.
Tsoukalas, L.H. and Uhrig, R.E. (1997) Fuzzy and Neural Approaches in Engineering.
John Wiley, New York.
Waddel, D.G. and Sell, J.L. (1964) Effects of dietary calcium and phosphorus on
the utilization of dietary iron by the chick. Poultry Science 43, 1249–1257.
Ward Systems Group (1996) NeuroShell 2. Ward Systems Group, Frederick, MD,
USA.
Williams, G.P. (1997) Chaos Theory Tamed. Joseph Henry, Washington, DC.
Wirsam, B. and Uthus, E.O. (1996) The use of fuzzy logic in nutrition. Journal of
Nutrition 126 (9 Suppl.), 2337S–2341S.
Wirsam, B., Hahn, A., Uthus, E.O. and Leitzmann, C. (1997) Fuzzy sets and fuzzy
decision making in nutrition. European Journal of Clinical Nutrition 51, 286–296.
Yoshida, M., Hizikuro, S., Hoshii, H. and Morimoto, H. (1962) Effect of dietary
protein and energy levels on the growth rate, feed efficiency and carcass
composition of chicks. Agricultural and Biological Chemistry 26, 640–647.
Yoshida, M., Hoshii, H. and Morimoto, H. (1968) Breed difference in the pattern
of growth response to the change in the dietary energy and protein levels.
Japanese Poultry Science 5, 211–225.
116 W.B. Roush
Yoshida, M., Hoshii, H. and Morimoto, H. (1969) Age difference in the pattern of
growth response to the change in the dietary energy and protein levels.
Japanese Poultry Science 6, 7–21.
Zadeh, L.A. (1965) Fuzzy sets. Information and Control 8, 338–353.
Zhang, F. and Roush, W.B. (2002) Multiple-objective (goal) programming model
for feed formulation: an example for reducing nutrient variation. Poultry
Science 81, 182–192.
Zimmerman, H.J. (1992) In: Zadeh, L.A. and Kacprzyk, J. (eds) Fuzzy Logic for the
Management of Uncertainty. Wiley, New York.
Zimmerman, H.J. (1996) Fuzzy Set Theory and its Applications, 3rd edn. Kluwer
Academic, Boston, Massachusetts.
Zupan, J. and Gasteiger, J. (1993) Neural Networks for Chemists: An Introduction.
VCH, New York.
7 The Problem of Predicting the
Partitioning of Scarce Resources
during Sickness and Health
in Pigs
I. KYRIAZAKIS AND F.B. SANDBERG
Animal Nutrition and Health Department, Scottish Agricultural College
West Mains Road, Edinburgh, EH9 3JG, UK
Ilias.Kyriazakis@sac.ac.uk
Introduction
Pigs are often faced with the problem of partitioning one or more scarce
food resources. This situation may arise when they are offered
intentionally restricted amounts of food, but also when they are offered ad
libitum access to a food. In the latter case intake of scarce resources may
arise when pigs are given access to a poor quality food (e.g. bulky food,
Whittemore et al., 2001) or when environmental stressors unintentionally
constrain food intake (e.g. high environmental temperature, Wellock et al.,
2003a). The voluntary reduction of food intake that accompanies
subclinical infection (anorexia, Kyriazakis et al., 1998) is a special case that
may lead to intake of scarce resources.
We are interested in the problem of predicting the partitioning of
absorbed scarce protein (and energy) to protein (PR) and lipid (LR)
retention in healthy and ‘diseased’ growing pigs. Quantitative solutions to
this problem in healthy pigs have been evolving for over 30 years. For this
reason, we will start with a historical approach to resolving the issue. We
will then discuss the current solutions to it offered in the literature and aim
to reach a solution that appears able to predict satisfactorily protein and
lipid retention in healthy growing pigs. The preferred solution will form
the basis of a framework that will be developed to account for the
partitioning of absorbed scarce resources in pigs challenged by pathogens.
This part of the framework will mainly have a heuristic value, as it will be
pointing towards issues that need to be resolved in order to be able to
predict adequately protein and lipid retention in pigs challenged by
pathogens.
Table 7.1. The current solutions to the problem of predicting the partitioning of scarce
resources. The solutions have been grouped together according to the key characteristic they
invoke to resolve the problem.
Solutions Key characteristic
the energy to protein ratio of the food. It calls for the values of four
parameters in order to solve the problem; each of these values are stated to
be affected by genotype, state, liveweight and nutritional history. The
practical consequence of the solution is that each experiment needs to be
carried out across all of these factors, possibly in all combinations, in order
for the rule to apply in any given case. An enormous amount of
information is called for, and for this reason this rule will not be discussed
any further. The same criticism applies to the rule proposed by van Milgen
et al. (2000), as they estimate that 21 parameters are required in order to
predict protein and lipid retention. The information required by this rule
for any particular genotype, existing in the future, is unlikely ever to be
available. Sandberg et al. (2005a,b) have provided recently a more detailed
criticism on the deficiencies of the three rules and why they cannot have a
general application to the prediction of PR and LR.
The three solutions that survive qualitative testing against
experimental evidence (i.e. Black et al., 1986 (and its derivative by NRC,
1998); Kyriazakis and Emmans, 1992a,b and de Greef and Verstegen,
1995) identify the marginal response in protein retention to protein and
energy intakes as the key variable to solving the problem. Black et al. (1986)
proposed that:
PR = b . (MEI – (c . MEm)) g/day (7.1)
where MEI is metabolizable energy intake, c is a constant and MEm is the
metabolizable energy requirement for maintenance. On the other hand,
Kyriazakis and Emmans (1992a,b) propose that:
PR = ep. (IP – IPm) g/day (7.2)
where IP is the ideal protein intake and IPm the ideal protein requirements
for maintenance. The solution offered by de Greef and Verstegen (1995)
has been shown by Emmans and Kyriazakis (1997) to be algebraically
equivalent to that offered by Kyriazakis and Emmans (1992b) and has a
relatively high information requirement. For these reasons, their solution
will not be considered further here. The question then is whether either of
the above two key parameters, the marginal response in PR to energy
supply (b) and the marginal response in PR to protein supply (ep) are
affected by the pig, the environment in which it is kept and the composition
of the food it is offered. These factors will be considered in turn below.
The solution of Black et al. (1986) was developed for energy limiting foods,
i.e. when MEI is less than required for PRmax. An implicit assumption was
that for such foods the value of b would be independent of food composition,
i.e. it would attain its maximum value. Below we review the effects of
liveweight, genotype including sex, and environmental temperature on the
marginal response in protein retention to energy supply on such foods.
120 I. Kyriazakis and F.B. Sandberg
210
190
Protein retention, PR (g/day)
170
150
130
110
90
70
5 7 9 11 13 15 17 19 21
ME intake (MJ/day)
Fig. 7.1. The response in protein retention (PR) to metabolizable energy intake (MEI) above
maintenance of pigs of four different live weights (Quiniou et al., 1995): these were 45 kg
(–䊉–), 65 kg (- -䊊- -), 80 kg (–䊏–) and 94 kg (- -ⵧ- -). The four levels of MEI were achieved by
four levels of feeding, all of which provided a constant high supply of protein. A common
slope has been fitted for the relationship between PR and MEI.
Partitioning of Scarce Resources during Sickness and Health 121
In their proposals for the nutrient requirements for swine, NRC (1998)
suggested that the marginal response to energy intake, on protein-
adequate foods, falls as the temperature increases. The argument follows
from the experiment of Close et al. (1978), whose data are plotted in Fig.
7.2.
Although it is difficult to be certain that the foods used in the
experiment were limiting in energy, the data are far from persuasive that
the response varied with temperature. This is consistent with the view of
Black et al. (1986) and Wellock et al. (2003a), who propose that
environmental temperature has no effect on the marginal response in PR
to energy intake on protein adequate foods.
Unlike the previous solution, the rule proposed by Kyriazakis and Emmans
(1992a,b) is intended to apply across protein- and energy-limiting foods.
According to this rule, such foods are defined according to their ratio of
the metabolizable energy content (MEC, MJ/kg) to the digestible crude
122 I. Kyriazakis and F.B. Sandberg
300
250
200
PR (kJ/kg0.75)
150
100
50
0
0 500 1000 1500 2000 2500
MEI (kJ/kg0.75.day)
Fig. 7.2. The response in protein retention (PR kJ/kg0.75.day) to metabolizable energy intake
(MEI kJ/kg0.75.day) for pigs given different allowances of the same food at five different
temperatures from Close et al. (1978); 10°C (䊉), 15°C (䊊), 20°C (䉲), 25°C (䉮), and 30°C (䊏).
The regression line for all the data is PR = 0.147 (0.0062).MEI – 44.07 (7.58).
protein content (DCPC, kg/kg) of the food, R (MJ ME/kg DCP). Foods with
a value of R > 72.55 are defined as being protein limiting and the
marginal response in PR to protein supply on such foods, ep, is assumed to
attain its maximum value (ep)max. Foods that are defined as energy limiting
have an ep that varies according to R:
ep = µ.R (7.3)
The above rule makes the distinction between the protein and energy
dependent phases of the marginal response in PR to protein supply. The
marginal response in PR when both the energy and the protein allowances
offered to pigs are varied is shown in Fig. 7.3.
Thus, liveweight, genotype and environmental temperature have the
potential to affect both (ep)max in protein limiting foods and ep in energy
limiting foods.
PRmax
E2
E1
ep
Fig. 7.3. The predicted rates of protein retention of a pig given access to feeds of different
protein contents at different levels of feed (and hence energy intake). Feeding levels E1 and
E2 constrain protein retention by being energy limiting, whereas level E3 is protein limiting
and hence allows the animal to reach its maximum rate of protein retention, PRmax.
1985a; Mohn et al., 2000; and de Lange et al., 2001) conclude that the
maximum ep, around 0.75, is not affected by liveweight. The only
experiment where a liveweight effect has been reported is that of Campbell
and Dunkin (1983) who found a high response in N retention to N intake
for pigs between 1.8 and 6.5 kg liveweight. Their value of (ep)max is closer to
0.90 and it is perhaps a reflection of the difficulty in measuring N balance
in very small pigs. Emmans and Kyriazakis (1996) have measured the ep for
small (12 kg) and large (72 kg) pigs given access to both protein- and
energy-limiting foods. The mean values of ep are not different between the
two different liveweights on any of the foods used.
Kyriazakis et al. (1994) used entire male Large White Landrace and pure
bred Chinese Meishan pigs to investigate the effect of genotype on the
relationship between food composition and ep in foods of varying
energy:protein ratios, R. The value of ep was found to be directly proportional
up to a maximum value of R. The overall constant of proportionality, µ, was
0.0108 and did not differ significantly between the two breeds. The
maximum value of ep was also similar between the two breeds.
Other experiments by de Greef et al. (1992) and Fuller et al. (1995) also
conclude that the maximum ep does not differ between different genotypes.
There is also evidence that different sexes of pigs use a limiting protein
supply with similar efficiency, ep. Campbell et al. (1984, 1985b) found no
difference in maximum ep between entire males and females. The data of
Batterham et al. (1990) shown in Fig. 7.4 lend strong support to this
conclusion.
It would, therefore, appear safe to conclude from the above
experiments that the marginal response in PR to protein supply does not
differ between different genotypes, including sexes.
124 I. Kyriazakis and F.B. Sandberg
10
0
0 2 4 6 8 10 12 14 16 18
Ileal digestible lysine intake (g/day)
Fig. 7.4. Response in lysine retention (Rlys g/day) to ileal digestible lysine intake (Ilys g/day):
the regression line is Rlys = 0.763.(Ilys – 1.245) for the combined data of male (䊉) and female
(䊊) Large White (20–45 kg) pigs used by Batterham et al. (1990). Separate plateaux for
males and females are shown.
The data of the experiment by Ferguson and Gous (1997), who grew pigs
fed ad libitum from 13 to 30 kg on food with 93–230 g crude protein/kg at
18, 22, 26 and 30°C, are reproduced in Fig. 7.5.
The data clearly show that the marginal response in protein retention
was not affected by environmental temperature. They are consistent with
the findings of Berschauer et al. (1983) and Campbell and Taverner (1988)
who conclude that temperature does not affect ep.
140
120
80
60
40
20
50 100 150 200 250 300 350 400
Crude protein intake (g/day)
Fig. 7.5. The response in protein retention to crude protein intake for pigs fed ad libitum
foods that were limiting in protein at four different temperatures: 18°C (䊉), 22°C (䊊),
26°C (䉲), and 30°C (䉮) as found by Ferguson and Gous (1997). The solid line is described
by PR = 0.525 (CPI – 4.92) until the plateau of 117.4 g/day is reached.
limiting for both genotypes. The same would apply for male and female
pigs, despite the fact that females have the propensity to be fatter than
males of the same protein weight when fed ad libitum. Here, it would be
important to emphasize that we need to account for potential differences
between the genotypes (or sexes) in their maintenance requirements. This
has not always been taken into account in the interpretation of
experiments that have measured PR responses to different levels of
protein and/or energy intake (e.g. Quiniou et al., 1996, see above).
A second implication is that both frameworks are capable of making
predictions across a wide range of conditions. These include conditions
where the pig is depositing protein at the expense of lipid retention. Such
conditions may arise in weaned pigs consuming small amounts of a
relatively high protein content food (Kyriazakis and Emmans, 1992a,b).
Many current solutions in the literature are still unable to predict this (e.g.
de Lange, 1995; Whittemore, 1995; Green and Whittemore, 2003).
The framework that predicts PR as a function of protein intake does so
on the basis of ideal protein. It is, therefore, an implicit assumption in this
framework that the marginal response in PR will be the same for all amino
acids, when they are first limiting. This assumption has recently been
challenged by Heger et al. (2002, 2003), but there is considerable
uncertainty over their estimates of maximum efficiency of amino acid
retention (with values ranging from 1.17 to 0.66 for different amino acids).
126 I. Kyriazakis and F.B. Sandberg
For the time being it may be safe to assume a single overall efficiency for all
amino acids, whichever is the first limiting (Sandberg et al., 2005b).
Whilst the position that the marginal response in protein retention
may only be affected by food composition may be attractive from a
nutritional modelling point of view, it can be viewed as having unattractive
perspectives (Luiting and Knap, 2005). Animal breeders, interested in
genetic variation between individuals, have criticized it on the basis that it
does not ‘accommodate a genotype-specific drive towards body fatness as
well as a drive towards protein deposition’. The task is always to keep the
framework variables to the necessary minimum in order to lead to
tractable solutions and make predictions for populations of pigs. This
becomes more important when the framework is broadened to account, for
example, for the partitioning of nutrients during disease.
When a pig that has not been previously exposed to a pathogen, i.e. an
immunologically naive pig, encounters the pathogen for the first time, it
may require nutrients for functions that will enable it to cope with the
challenge. Such functions may include the innate immune response, which
is one of the first lines of defence to pathogens, and the repair and
replenishment of damaged or lost tissue, such as blood plasma or cells.
Eventually, the pig will be expected to develop an acquired immune
response towards the pathogen, and nutrient resources will need to be
directed towards the maintenance of this function. When the pig is re-
exposed to the same kind of pathogen, the main additional resource
requirement would be due to the function of acquired immunity, as
investment towards this function would minimize the potential pathogen-
associated damage to the host.
When nutrient resources are scarce, the challenged pig can be seen as
having the problem of allocating these resources between its various
functions. These arise from the exposure to the pathogen, but they also
include the ‘normal’ functions of a healthy pig, such as maintenance and
growth. In this chapter, we concentrate upon the problem of allocating
scarce protein during exposure to pathogens. This is because: (i) current
evidence suggests that protein is often the first limiting resource in
challenged pigs; (ii) many components of the immune response are highly
proteinaceous (Houdijk et al., 2001); and (iii) as a consequence, most
evidence in the literature is in relation to the partitioning of protein or
amino acids during challenge. There is some limited evidence that energy
may become a limiting resource during pathogen challenge, as a
consequence of the increased energy requirements due to fever and the
requirements of the immune response. It is generally accepted, however,
that these requirements are relatively small for growing animals (chickens,
Klasing et al., 1987; mice, Demas et al., 1997; pigs, van Heugten et al., 1996).
Partitioning of Scarce Resources during Sickness and Health 127
Food
intake
(g/day)
Sub- Clinical
clinical
0
0 Pathogen dose or level
Fig. 7.6. A proposed schematic description of the effect of dose or level of parasites on the
daily rate of food intake by the host, reproduced from Kyriazakis et al. (1998). Food intake is
reduced once a threshold pathogen dose (or level) is reached which leads to sub-clinical
disease. Reductions in food intake become very severe at higher pathogen doses or levels
that lead to clinical disease.
128 I. Kyriazakis and F.B. Sandberg
Parameter k would be pathogen specific and may also reflect the virulence
of the pathogen. The exponential form of the relationship is consistent
with the view that small pathogen loads have little effect on hosts and that
an already weakened system would suffer more damage from increases in
its challenge (i.e. further exposure to pathogens). It should be emphasized
that there would be a (Ploss)max value equivalent to the amount of damage
that may lead to death.
The prediction of partitioning of scarce protein in naive pigs during
exposure to pathogens appears to be straightforward. Scarce protein
would be expected to be prioritized towards the functions of maintenance
and innate immune response; any protein above this should be used for
growth, after the protein needs for repair have been accounted for. The
parameterization of Eqns (7.4) and (7.5) should be relatively
straightforward to estimate for specific pathogens, providing that during
the course of the measurement the animal does not develop an acquired
immune response to the pathogen. As the development of the acquired
immune response may be very rapid, current information presented in the
literature where PR is measured over a number of days usually includes
both the phase of innate and acquired immune responses. This makes
interpretation of current experiments difficult (see below).
Maximum
damage
Actual
damage
Fig. 7.7. The proposed relationship between the actual damage caused by a pathogen load
to a host in relation to the level of investment in immunity (see Eqn 7.9 in text). The maximum
damage caused by the pathogen load is achieved when there is no investment in immunity.
200
180 PRmax
160
Protein retention (g/day)
140
120
100
80
60
40
20
0
0 50 100 150 200 250
Ideal protein intake (g/day)
Fig. 7.8. Predictions of protein retention in relation to different intakes of ideal protein for an
uninfected (solid line) and an immune pig challenged by a pathogen (dotted line). The increase
in the intercept for the challenged pig is a reflection of the costs associated with the pathogen,
when there is no investment of protein intake in immunity. Both healthy and challenged pigs
are expected to attain their genetically defined maximum protein retention, PRmax.
132 I. Kyriazakis and F.B. Sandberg
The above framework was developed on the basis of ideal protein intake.
This assumes that the composition of the protein retained as either the
immune response or growth is similar. In Table 7.2 the amino acid
composition of some proteinaceous components of the immune response is
compared to the composition of pig body protein, which forms the basis of
the ideal protein system (ARC, 1981). The difference in amino acid
composition between these two body components is striking. Such
differences will be expected to have a significant effect on the prediction of
protein retention, if the contribution of the immune response to the
overall protein retention is relatively high (Wang and Fuller, 1989). In this
case, the framework will need to be modified to make predictions on the
basis of individual amino acid responses. Protein retention would then be
reconstituted on the basis of individual amino acid retention. This solution
will have exceedingly high information requirements and, therefore,
parameterization of the framework will be exceedingly difficult. An added
complication would also arise from the fact that stimulation by different
pathogens invokes different effector mechanisms of the immune response
(Dong and Flavell, 2001). These different effector mechanisms may have
different amino acid compositions (Table 7.2), which in turn may be
utilized by different efficiencies.
An alternative, but indirect solution to the problem that arises from
the above would be to retain the ideal protein system, but assume that
the efficiency with which protein is utilized for the purposes of the
immune response is modified. This would be in order to account for the
different ‘ideal’ protein composition of the two body components. This
seems to be a less onerous task than the above and experiments that can
be designed to contribute towards the parameterization of this solution
can be envisaged.
Partitioning of Scarce Resources during Sickness and Health 133
Table 7.2. A summary of the amino acid composition of different effector proteins associated
with the immune response, in relation to the reference pig body protein that is normally used
for calculating the biological value of a food (ideal protein).
Amino acid composition (g/kg protein)a
Average of
Pig protein 7 APPs IgA IgE MCP Mucin
PL0
Pathogen load
Fig. 7.9. The proposed relationship between the level of pathogen challenge and the
partitioning ratio of scarce resources (denoted by p) between growth and immune functions,
once the level of challenge has exceeded a threshold, PL0, in a susceptible pig (dotted line)
and resistant pigs (solid line and dashed lines). The two resistant pigs differ in the rate of
increase in the value of p for a given change in pathogen load.
the animal starts to acquire immunity depends, within certain limits, upon
its pathogen load. This is assumed to be a reflection of the PL required to
stimulate the acquired immune response (see above). However, the
duration of the phase of acquisition of immunity is assumed to be
independent of pathogen load (Steel et al., 1980; Houdijk et al., 2005), and
only dependent on pathogen type. Duration of this phase can be very short
in certain pathogens, such as bacterial (Turner et al., 2002a,b) and viral
(Zijlstra et al., 1997) challenges.
The above models also assume that the duration of the acquisition of
immunity is independent of host nutrition. Sandberg et al. (2006) reviewed
the literature on this issue and concluded that whilst this variable may be
affected by food composition (food protein content in particular), no
general relationship could be proposed between the two. As such, it was
suggested that until experiments that address this issue are performed, the
above simple assumption should be retained. A simulation of how
pathogen load changes within the host as a function of time based on the
above assumptions is shown on Fig. 7.10. The change in pathogen load,
assuming normal acquisition and expression of immunity, is consistent with
experimental findings for bacterial (e.g. Kelly et al., 1996) and viral (e.g.
Bocharov, 1998; McDermott et al., 2004) pathogen challenges.
The effect of nutrition on pathogen load as shown in Fig. 7.10 is the
outcome of the proposed framework.
Future Directions
Time
Fig. 7.10. A prediction of the change in pathogen load (arbitrary units) over time of animals
given access to different levels of protein supply low (dashed line), medium (solid line) and
high (dotted line). No difference is observed in the early stage of infection whilst the animal
acquires immunity; however, once expression of acquired immunity commences, its level of
expression is proposed to be affected by the level of resource supply.
136 I. Kyriazakis and F.B. Sandberg
Acknowledgements
This work was in part funded by the Biotechnology and Biological Science
Research Council of the UK and PIC/Sygen. SAC receives support from the
Scottish Executive, Environment and Rural Affairs Department. We are
grateful to our colleagues Dimitris Vagenas, Andrea Doeschl-Wilson and
Will Brindle for comments on earlier versions of the manuscript and to
everyone in the Animal Nutrition and Health Department for their support
in this activity.
References
Anderson, R.M. (1994) Mathematical studies of parasitic infection and immunity.
Science 264, 1884–1886.
ARC (1981) The Nutrient Requirements of Pigs. Agricultural Research Council,
Commonwealth Agricultural Bureaux, Slough, UK.
Partitioning of Scarce Resources during Sickness and Health 137
2. Growth rate, including protein and fat deposition. British Journal of Nutrition
40, 423–431.
Coop, R.L. and Kyriazakis, I. (1999) Nutrition-parasite interaction. Veterinary
Parasitology 84, 187–204.
de Greef, K.H. and Verstegen, M.W.A. (1995) Evaluation of a concept on energy
partitioning in growing pigs. In: Moughan, P.J., Verstegen, M.W.A., Visser-
Reyneveld, M.I. (eds) Modelling Growth in the Pig. EAAP Publication No. 78.
Wageningen Press, Wageningen, Netherlands, pp. 137–149.
de Greef, K.H., Kemp, B. and Verstegen, M.W.A. (1992) Performance and body
composition of fattening pigs of two strains during protein deficiency and
subsequent realimentation. Livestock Production Science 30, 141–153.
de Lange, C.F.M. (1995) Framework for a simplified model to demonstrate principles
of nutrient partitioning for growth in the pig. In: Moughan, P.J., Verstegen,
M.W.A., Visser-Reyneveld, M.I. (eds) Modelling Growth in the Pig. EAAP
Publication No. 78. Wageningen Press, Wageningen, Netherlands, pp. 71–85.
de Lange, C.F.M., Gillis, A.M. and Simpson, G.J. (2001) Influence of threonine
intake on whole-body protein deposition and threonine utilisation in growing
pigs fed purified diets. Journal of Animal Science 79, 3087–3095.
Demas, G.E., Chefer, V., Talan, M.I. and Nelson, R.J. (1997) Metabolic cost of
mounting an antigen-stimulated immune response in adult and aged
C57BL/6J mice. American Journal of Physiology 273, R1631–R1637.
Dong, C. and Flavell, R.A. (2001) Th1 and Th2 cells. Current Opinion in Hematology
8, 47–51.
Dunkin, A.C. and Black, J.L. (1985) The relationship between energy intake and
nitrogen balance in the growing pig. In: Moe, P.W., Tyrrell, H.F. and Reynolds,
P.J. (eds) Energy Metabolism of Farm Animals: Proceedings of the 10th Symposium.
EAAP Publication No 32. Rowman and Littlefield, Totowa, N, pp. 110–113.
Emmans, G.C. and Kyriazakis, I. (1996) The effect of live weight on the efficiency
of using ideal protein for protein retention. Proceedings of the British Society of
Animal Science, pp 13.
Emmans, G.C. and Kyriazakis, I. (1997) Models of pigs growth: problems and
proposed solutions. Livestock Production Science 51, 119–129.
Emmans, G.C. and Kyriazakis, I. (2001) Consequences of genetic change in farm
animals on food intake and feeding behavior. Proceedings of the Nutrition Society
60, 115–125.
Ferguson, N.S. and Gous, R.M. (1997) The influence of heat production on
voluntary food intake in growing pigs given protein-deficient diets. Animal
Science 64, 365–378.
Fisher, C., Morris, T.R. and Jennings, R.C. (1973) A model for the description and
prediction of the response of laying hens to amino acid intake. British Poultry
Science 14, 469–484.
Fuller, M.F. and Crofts, R.M.J. (1977) The protein-sparing effect of carbohydrate.
1. Nitrogen retention of growing pigs in relation to diet. British Journal of
Nutrition 38, 479–488.
Fuller, M.F., Franklin, M.F., McWilliam, R. and Pennie, K. (1995) The responses of
growing pigs, of different sex and genotype, to dietary energy and protein.
Animal Science 60, 291–298.
Green, D.M. and Whittemore, C.T. (2003) Architecture of a harmonized model of
the growing pig for the determination of dietary net energy and protein
requirements and of excretions into the environment (IMS Pig). Animal Science
77, 113–130.
Partitioning of Scarce Resources during Sickness and Health 139
Greer, A.W., Stankiewicz, M., Jay, N.P., McNulty, R.W. and Sykes, A.R. (2005) The
effect of concurrent corticosteriod induced immuno-suppression and infection
with the intestinal parasite Trichostrongylus colubriformis on food intake and
utilisation in both immunologically naïve and competent sheep. Animal Science
80, 89–99.
Haile, A., Anindo, D.O., Tembley, S., Mukasa-Mugerwa, E., Tibbo, M., Yami, A.,
Baker, R.L. and Rege, J.E.O. (2004) Effects of dietary protein supplementation
and infection with gastrointestinal nematode parasites on some nutritional and
metabolic parameters in Ethiopian Menz and Horro sheep Livestock Production
Science 91,183–195.
Heger, J., van Phung, T. and Krizova, L. (2002) Efficiency of amino acid utilisation
in the growing pig at sub-optimal levels of intake: lysine, threonine, sulphur
amino acids and tryptophan. Journal of Animal Physiology and Animal Nutrition
86, 153–165.
Heger, J., van Phung, T., Krizova, L., Sustala, M. and Simecek, K. (2003) Efficiency
of amino acid utilisation in the growing pig at sub-optimal levels of intake:
branched chain-chain amino acids, histidine and phenylalanine + tyrosine.
Journal of Animal Physiology and Animal Nutrition 87, 52–65.
Houdijk, J.G.M. and Athanasiadou, S.L. (2003) Direct and indirect effects of host
nutrition on ruminant gastrointestinal nematodes. In: t’Mannetje, L., Ramirez-
Aviles, L., Sandoval-Castro, C.A. and Ku-vera, J.C. (eds) Proceedings of the VI
International Symposium on the Nutrition of Herbivores: Matching Herbivore Nutrition
to Ecosystems Biodiversity. Merida, Mexico.
Houdijk, J.G.M., Jessop, N.S. and Kyriazakis, I. (2001) Nutrient partitioning
between reproductive and immune functions in animals. Proceedings of the
Nutrition Society 60, 515–525.
Houdijk, J.G.M., Anderson, D.H. and Kyriazakis, I. (2005) Induction of
experimental sub-clinical post-weaning colibacillosis in pigs. Proceedings of the
British Society of Animal Science 47, 84.
Kelly, B.P., Furney, S.K., Jessen, M.T. and Orme, I.M. (1996) Low-dose aerosol
infection model for testing drugs for efficacy against Mycobacterium tuberculosis.
Antimicrobial Agents and Chemotherapy 40, 2809–2812.
Klasing, K.C., Laurin, D.E., Peng, R.K. and Fry, M. (1987) Immunologically
mediated growth depression in chicks: Influence of feed intake, corticosterone
and interleukin-1. Journal of Nutrition 117, 1629–1637.
Knap, P.W. and Bishop, S.C. (2000) Relationships between genetic change and
infectious disease in domestic livestock. Proceedings of the British Society of Animal
Science: Occasional Publication 27, 65–80.
Kreukniet, M.B. and van der Zijpp, A.J. (1989) Effects of different doses of sheep
erythrocytes on the humoral immune response of chicken lines selected for
high or low antibody response. Poultry Science 69, 608–614.
Kyriazakis, I. and Emmans, G.C. (1992a) The effects of varying protein and energy
intakes on the growth and body-composition of pigs. 1. The effects of energy-
intake at constant, high protein-intake. British Journal of Nutrition 68, 603–613.
Kyriazakis, I. and Emmans, G.C. (1992b) The effects of varying protein and energy
intakes on the growth and body-composition of pigs. 2. The effects of varying
both energy and protein-intake. British Journal of Nutrition 68, 615–625.
Kyriazakis, I., Dotas, D. and Emmans, G.C. (1994) The effect of breed on the
relationship between feed composition and the efficiency of protein utilisation
in pigs. British Journal of Nutrition 71, 849–859.
140 I. Kyriazakis and F.B. Sandberg
Kyriazakis, I., Emmans, G.C. and with the technical assistance of Anderson, D.H.
(1995) Do breeds of pig differ in the efficiency with which they use a limiting
protein supply? British Journal of Nutrition 74, 183–195.
Kyriazakis, I., Tolkamp, B.J. and Hutchings, M.R. (1998) Towards a functional
explanation for the occurrence of anorexia during parasitic infections. Animal
Behaviour 56, 265–274.
Le Jambre, L.F. (1995) Relationship of blood loss to worm numbers, biomass and
egg production in Haemonchus infected sheep. International Journal of
Parasitology 25, 269–273.
Louie, K., Vlassoff, A. and Mackay, A. (2005) Nematode parasites of sheep:
extension of a simple model to include host variability. Parasitology 130,
437–446.
Luiting, P. and Knap, P.W. (2005) Comparison of pig growth models – the genetic
point of view. Recent Advances in Pig and Poultry Modelling. Kwazulu-Natal,
South Africa.
McDermott, A.B., Mitchen, J., Piaskowski, S., Souza, I., Yant, L.J., Stephan, J.,
Furlott, J. and Watkins, D.I. (2004) Repeated low-dose mucosal simian
immunodeficiency virus SIVmac239 challenge results in the same viral and
immunological kinetics as high-dose challenge: a model for the evaluation of
vaccine efficacy in nonhuman primates. Journal of Virology 78, 3140–3144.
Mohn, S., Gillis, A.M., Moughan, P.J. and de Lange, C.F.M. (2000) Influence of
dietary lysine and energy intakes on body protein deposition and lysine
utilisation in the growing pig. Journal of Animal Science 78, 1510–1519.
Nowak, M.A., Bonhoeffer, S., Hill, A.M., Boehme, R., Thomas, H.C. and McDade,
H. (1996) Viral dynamics in hepatitis B virus infection. Proceedings of the
National Academy of Sciences 93, 4398–4402.
NRC (1998) Nutrient Requirements of Swine, 10th edn. National Academy Press,
Washington, DC.
Olesen, I., Groen, A.F. and Gjerde, B. (2000) Definition of animal breeding goals
for sustainable production systems. Journal of Animal Science 78, 570–582.
Pomar, C., Kyriazakis, I., Emmans, G.C. and Knap, P.W. (2003) Modeling
stochasticity: Dealing with population rather than individual pigs. Journal of
Animal Science 81, E178–E186.
Quiniou, N., Noblet, J., van Milgen, J. and Dourmad, J-Y. (1995) Effect of energy
intake on performance, nutrient and tissue gain and protein and energy
utilisation in growing boars. Animal Science 61, 133–143.
Quiniou, N., Dourmad, J-Y. and Noblet, J. (1996) Effect of energy intake on the
performance of different types of pig from 45 – 100 kg body weight. 1. Protein
and lipid deposition. Animal Science 63, 277–288.
Rao, S.V.R., Praharaj, N.K., Reddy, M.R. and Panda, A.K. (2003) Interaction
between genotype and dietary concentrations of methionine for immune
function in commercial broilers. British Poultry Science 44, 104–112.
Reeds, P.J., Fjeld, C.R. and Jahoor, F. (1994) Do the differences between the amino
acid compositions of acute-phase and muscle proteins have a bearing on
nitrogen loss in traumatic states? Journal of Nutrition 124, 906–910.
Sandberg, F.B., Emmans, G.C. and Kyriazakis, I. (2005a) Partitioning of limiting
protein and energy in the growing pig: description of the problem, possible
rules and their qualitative evaluation. British Journal of Nutrition 93, 205–212.
Sandberg, F.B., Emmans, G.C. and Kyriazakis, I. (2005b) Partitioning of limiting
protein and energy in the growing pig: testing quantitative rules against
experimental data. British Journal of Nutrition 93, 213–224.
Partitioning of Scarce Resources during Sickness and Health 141
Sandberg, F.B., Emmans, G.C. and Kyriazakis, I. (2006) A model for predicting
food intake of growing animals during exposure to pathogens. Journal of
Animal Science (in press).
Stamataris, C., Kyriazakis, I. and Emmans, G.C. (1991) The performance and
body-composition of young pigs following a period of growth retardation by
food restriction. Animal Production 53, 373–381.
Steel, J.W., Symons, L.E.A. and Jones, W.O. (1980) Effects of level of larval intake
on the productivity and physiological and metabolic responses of lambs
infected with Trichostrongylus colubriformis. Australian Journal of Agricultural
Research 31, 821–838.
Stewart, T.B., Hale, O.M. and Johnson J.C., Jr (1969) Failure of parasitised gilt and
barrow pigs on different planes of nutrition to respond alike to a superimposed
infection with Strongyloides ransomi. The Journal of Parasitology 55, 1055–1062.
Takhar, B.S. and Farrell, D.J. (1979) Energy and nitrogen metabolism of chickens
infected with either Eimeria acervulina or Eimeria tenella. British Poultry Science
20, 197–211.
Taylor-Robinson, A.W. (2000) Increased production of acute-phase proteins
corresponds to the peak parasitemia of primary malaria infection. Parasitology
International 48, 297–301.
Turner, J.L., Dritz, S.S., Higgins, J.J., Herkelman, K.L. and Minton, J.E. (2002a)
Effect of a Quillaja saponaria extract on growth performance and immune
function of weanling pigs challenged with Salmonella typhimurium. Journal of
Animal Science 80, 1939–1946.
Turner, J.L., Dritz, S.S., Higgins, J.J. and Minton, J.E. (2002b) Effects of
Ascophyllum nodosum extract on growth performance and immune function on
young pigs challenged with Salmonella typhimurium. Journal of Animal Science 80,
1947–1953.
Van Dam, J.T.P., Schrama, J.W., Vreden, A., Verstegen, M.W.A., Wensing, T., van
der Heide, D. and Zwart, A. (1998) The effect of previous growth retardation
on energy and nitrogen metabolism of goats infected with Trypanosoma vivax.
British Journal of Nutrition 77, 427–441.
van Heugten, E., Coffey, M.T. and Spears, J.W. (1996) Effects of immune challenge,
dietary energy density, and source of energy on performance and immunity in
weanling pigs. Journal of Animal Science 74, 2431–2440.
van Houtert, M.F.J., Barger, I.A., Steel, J.W., Windon, R.G. and Emery, D.L. (1995)
Effects of dietary protein intake on responses of young sheep to infection with
Trichostrongylus colubriformis. Veterinary Parasitology 56, 163–180.
van Milgen, J. and Noblet, J. (1999) Energy partitioning in growing pigs: the use of
a multivariate model as an alternative for the factorial analysis. Journal of
Animal Science 77, 2154–2162.
van Milgen, J., Quiniou, N. and Noblet, J. (2000) Modelling the relation between
energy intake and protein and lipid deposition in growing pigs. Animal Science
71, 119–1130.
Waller, P.J. (1997) Nematode parasite control of livestock in the Tropics/Subtropics:
the need for novel approaches. International Journal of Parasitology 27,
1193–1201.
Wang, T.C. and Fuller, M.F. (1989) The optimum dietary amino acid pattern for
growing pigs. 1. Experiments by amino acid deletion. British Journal of Nutrition
62, 77–89.
Webel, D.M., Johnson, R.W. and Baker, D.H. (1998a) Lipopolysaccharide-induced
reductions in body weight gain and feed intake do not reduce the efficiency of
142 I. Kyriazakis and F.B. Sandberg
Introduction
Systems of nutrient and energy utilization are widely used in the animal feed
industry. Existing energy systems such as digestible energy (DE),
metabolizable energy (ME) and net energy (NE) attribute a single energy
value to a feed but differ in the way that energy losses are accounted for.
Although these systems are simple to use and relatively robust, they have
been criticized because they cannot account for interactions between the feed
and the animal. This has led to the development of more mechanistic models
that can be used for feed evaluation or that predict the response of an
animal to a changing nutrient supply. The objective of this chapter is to
describe different aspects of energy loss and energy utilization in animals and
to challenge this knowledge with empirical and mechanistic approaches to
feed evaluation. Mechanistic models of animal nutrition undoubtedly
provide a more solid theoretical basis for feed evaluation. However, they
only explain part of the efficiency with which nutrients are used for different
productive purposes. Moreover, although ‘scientific truth’ may be a reason to
push for a more mechanistic approach to feed evaluation systems, there are
also reasons to be somewhat conservative. Past history, established
knowledge and the use of an established reference base can be expected to
play major roles in the acceptance or failure of future energy systems.
Gross energy
There are different ways to express the energy value of a feed. In one of its
more basic forms, it can be represented by its gross energy (GE) value,
digesting fibre than growing pigs, due to the four-times greater retention
time of digesta in the gastro-intestinal tract. Similarly, finishing pigs
typically digested fibre better than did growing pigs (Le Goff et al., 2002).
This means that, in contrast to GE, the DE value of a diet is not a
characteristic of the diet itself but is influenced by the animal.
The metabolizable energy (ME) value of a diet accounts for energy losses
that occur from fermentation gases and in the urine. Energy losses as
methane and hydrogen are relatively small for diets fed to growing pigs
(typically less than 0.5% of DE). Noblet et al. (2002) assumed that energy
losses in fermentation gases correspond to approximately 0.67 kJ/g of
fermented dietary fibre. Energy losses in the urine are of the order of 3.5%
of DE. Most of this energy is lost as urea (or uric acid in birds), which
originates from amino acid catabolism in the liver. Amino acids given in
excess of requirements will be deaminated and result in additional urinary
energy loss. There is a relatively straightforward relation between the
nitrogen and energy contents in the urine. In the INRA tables (Noblet et
al., 2002), the following equation is used for pigs:
(energy in urine; kJ/kg DM intake) = 192 + 31 × (nitrogen in urine; g/kg
DM intake)
Under the assumption that approximately half of the digestible nitrogen is
excreted in the urine, the equation above can easily be converted into a
relation between urinary energy losses and dietary protein content.
Heat production
All metabolizable energy that is not retained by the animal is lost as heat.
Heat production can be measured through calorimetry, whereas energy
retention can be measured using the serial slaughter technique. Indirect
calorimetry is based on the measurement of gas exchanges between the
animal and its environment. When nutrients are oxidized, animals
consume oxygen and produce carbon dioxide and methane. These gas
exchanges and the nitrogen excretion originating from protein catabolism,
combined with the stoichiometry of carbohydrate, protein, and lipid
oxidation allow calculation of heat production (Brouwer, 1965).
Calorimetry has the advantage over the serial slaughter technique that it
can be used to measure energy balance over successive short periods of
time. We have further refined this technique in order to obtain estimates of
different components of heat production related to fasting, physical activity
and the thermic effect of feeding (van Milgen et al., 1997). The energy and
nitrogen balance techniques typically give higher retention values than the
comparative slaughter technique (Quiniou et al., 1995) and are thought to
overestimate actual lipid and protein retention.
146 J. van Milgen
Total heat production is not to be confused with the heat increment (or
thermic effect of feeding). As indicated above, total heat production is the
difference between ME intake and retained energy. The heat increment is
the change in heat production associated with a change in ME intake.
Energy Systems
with biological reality and the following examples will illustrate this. As
indicated earlier, dietary protein affects the ME value of a diet because
amino acids that are not deposited are deaminated, and urea is excreted in
the urine. In other words, it is not the dietary protein itself that will
determine its ME value, as it will also depend on what fraction of that
protein the animal retains. A similar example can be given for the NE
system. The NE requirement is based on a requirement for maintenance
(FHP) and energy retention. The maintenance requirement is essentially
an ATP requirement. If glucose and lipid are compared, it can be shown
that their potential to produce ATP is similar (~ 74 kJ/ATP produced; van
Milgen, 2002). When these nutrients are used for maintenance, glucose
and lipid should therefore be attributed similar energy efficiencies.
However, when these nutrients are used for lipid deposition, large
differences occur. The biochemical efficiency of using glucose for lipid
deposition is close to 81%, whereas the efficiency of using dietary lipid for
lipid deposition is around 98%. This indicates that the final utilization will
eventually determine the energy value.
The fact that there are interactions between the animal and its diet
implies that classical energy systems are theoretically incorrect. This
problem can only be overcome if we acknowledge that there is no such
thing as ‘an energy value of a diet’ expressed in such a way that it can be
related to a requirement. The only energy value that would be an attribute
of the diet itself would be its GE value. However, this is of no use when
expressing a requirement.
Mathematical models are ideally suited to account for interactions
between the diet and the animal. In the past and also recently, several
models have been published that address this issue. It is beyond the scope
of this paper to compare these models, but Luiting and Knap (Chapter 13,
this volume) deal with this in some detail. This chapter will focus only on
how recent models deal with nutrient and energy transactions and to what
extent this mechanistic approach is more useful than classical systems of
nutrient evaluation.
Metabolism includes several pathways that are designed to cope with the
short-term dynamics of nutrient supply and requirement, or that deal with
nutrient transfers between different organs. Apart from the physiological
cost of using these pathways, there is also a biochemical cost related to the
nutrient transformations. Few, if any, of the published mechanistic whole
animal models include the examples that will be developed here. These
examples are not given to criticize these models, but to illustrate that
energy metabolism in animals is a very complex process, which goes far
beyond the approach employed in most models.
Although it may seem evident not to include glycogen in models that
are based on empirical estimates of efficiency (as the energy loss is
included, empirically, in the efficiency estimate and glycogen stores vary
little from day to day), this is less so for models that are based on
biochemistry. Depending on the site of glycogen storage and utilization
(muscle or liver), ATP synthesis from glucose is stoichiometrically 3 to 6%
less efficient if this glucose is stored as glycogen before being oxidized.
Type IIb muscle fibres have a high glycogen storage capacity. As their
mitochondrial oxidative capacity is limited these fibres, when solicited,
produce ATP mainly through glycolysis and yield lactate as an end-
product. Although lactate can be used directly by some tissues (e.g. heart
muscle but also skeletal muscle itself), it can also be used by the liver to
regenerate glucose through gluconeogenesis. In the transformation of
glucose to lactate (in muscle) and back to glucose (in the liver), a total of
four ATP will be lost, hence an energy loss of more than 10%.
Another example is the temporary storage of energy as lipid (see also
Baldwin, 1995 for a discussion of this topic). Type I (red) muscle fibres are
used to provide sustained muscular work (e.g. for maintaining posture).
Although these oxidative fibres can use both lipid and glucose as fuel, most
of the energy storage occurs as lipid. The temporary energy storage of
glucose as lipid is rather inefficient and requires 30% more energy than the
direct utilization of glucose.
Stoichiometry explains only part of the observed heat increment. Table 8.1
shows some results of an experiment in which either starch, maize gluten
Flow Models, Energy Transactions and Feed Systems 151
meal (an unbalanced protein source), casein or lipid was added to a basal
diet limiting in lysine supply (van Milgen et al., 2001). Also the coefficients
of the ME and NE equations described above can be used to evaluate the
contribution of stoichiometry to observed energy efficiencies.
The stoichiometric efficiency of using glucose for lipid varies between
81 and 84%, depending on whether the ATP synthesis during lipogenesis
from glucose is accounted for or not (Baldwin, 1995; van Milgen, 2002).
For starch, the NE/ME ratio from the equations used above is 0.82,
whereas van Milgen et al. (2001) observed a value of 0.84 (Table 8.1). This
suggests that the heat increment of glucose metabolism not related to
stoichiometry (e.g. intake, hydrolysis of starch, absorption, maintaining
blood glucose) is relatively small or that this cost is accounted for
elsewhere, for example in the maintenance energy expenditure.
Larger differences occur when comparing the stoichiometric and
experimental efficiencies for lipid metabolism. The theoretical efficiency of
lipid deposition from dietary lipid is very high. The only energy loss is due
to the re-activation of fatty acids to acyl-CoA during re-esterification.
Assuming that dietary lipids are hydrolysed and re-esterified twice, the
energy loss does not exceed 3%. However, the NE/ME ratio of the
equations above suggests that more than 10% of the energy in dietary lipid
is lost when it is used for lipid deposition. The results in Table 8.1 also
indicate that the observed efficiency is much lower than the stoichiometric
efficiency. The relatively low efficiency of using dietary lipid may be due to
the oxidation of dietary lipids (e.g. for ATP synthesis) combined with de
Table 8.1. Utilization of energy by growing pigs (van Milgen et al., 2001).
Maize gluten
Starch meal Casein Lipid
nutrients, excretion of faeces and urine, and costs of protein and lipid
deposition, all expressed on an ATP basis. As such, the model resembles the
concepts used in the effective energy system (Emmans, 1994), although the
units of expression are different. The model of Halas et al. (2004) also relies
on the stoichiometry of nutrient transactions. However, rather than
estimating efficiency values for different processes through calibration,
energy expenditures were attributed a priori to different physiological
processes. Energy costs for absorption and transport were attributed to
different nutrients. Protein synthesis was assumed to cost 4 ATP per
peptide, whereas hydrolysis of a peptide bond also required 1 ATP. Because
the model includes four different protein pools (muscle, skin–backfat,
organs and bone), each with their own turnover characteristics, differences
in body protein composition will result in differences in ATP requirement
and, hence, energy expenditure. Other costs specifically accounted for
include urea synthesis and bone mineralization.
Service functions
Kidney (Na+ transport) 6–7
Heart 9–11
Nervous tissues 15–20
Respiration 6–7
Repair functions
Protein re-synthesis 10–15
Lipid re-synthesis 1–2
Na+ transport (membrane potential) 20–25
154 J. van Milgen
included (e.g. the cost of protein turnover discussed above), we will have to
rely to a large extent on empirical estimates of maintenance. From a
biological and physiological point of view, it is very difficult to define and
measure maintenance unambiguously. Theoretically, maintenance corre-
sponds to a situation in which energy retention equals zero (i.e. energy
intake equals heat production) but, for a growing animal, this corresponds
to a non-physiological situation. Moreover, zero energy retention may
theoretically be achieved while depositing protein and catabolizing body
lipid and therefore does not correspond to maintaining a constant body
weight.
The FHP is closely related to the maintenance energy requirement and
serves as the maintenance energy requirement in NE systems (Noblet et al.,
1994). During fasting, animals mobilize body reserves in order to supply
energy for vital body functions. Measured values for FHP after a 1-day
fasting period range from 700 to 800 kJ/((kg BW)0.60•d) in growing pigs
offered feed close to ad libitum prior to fasting. Genotype (or leanness)
appears to have an important impact on FHP with lower estimates for
obese Meishan barrows and higher estimates for lean Piétrain boars (van
Milgen et al., 1998). Part of the difference between genotypes seems to be
due to differences in body composition. In particular, the viscera make an
important contribution to FHP. This is consistent with the observation that
the previous plane of nutrition, and thus the viscera mass, affects FHP
(Koong et al., 1982) and justifies the use of different protein pools to
explain energy expenditure (Halas et al., 2004).
Part of the differences observed in FHP may be due to differences in
the body reserves that are mobilized during fasting. It is likely that during
fasting, the animal first mobilizes glycogen reserves, followed by the
mobilization of labile protein stores (visceral proteins) and lipids.
Consequently, the length of the fasting period may affect FHP. During
prolonged fasting, the animal is likely to adapt its metabolism so as to
minimize energy expenditure. It is for this reason that in our laboratory
we measure FHP after a 1-day fasting period so that metabolism still bears
some resemblance to a producing animal.
The energy cost of physical activity can be an important source of
variation between animals. Energy expenditure per hour of standing
appears at least fourfold greater in pigs than in most other domestic
species (Noblet et al., 1993). Different techniques exist to measure physical
activity including measurement of standing duration, motion detection,
and force detection. Heat production due to physical activity is estimated
from a statistical relation between variation in heat production and
variation in recorded physical activity. Our current estimate (based on
force detection) is approximately 200 kJ/((kg BW)0.60 •d), which
corresponds to 3 h of standing per day (van Milgen et al., 2001); this value
corresponds to approximately 20% of the maintenance energy
requirement. Physical activity appears rather variable between animals and
may be affected by feeding level, type of diet, and genotype (Susenbeth
and Menke, 1991; Schrama et al., 1996). Because of its contribution to heat
Flow Models, Energy Transactions and Feed Systems 155
equilibrium
nutrient concentration
Fig. 8.1. Homeostatic regulation of amino acid catabolism (solid line) and anabolism (dashed
line) of carcass proteins. At equilibrium nutrient concentration (the solid dots), anabolism and
catabolism are regulated by a time-dependent homeorhetic process (Lovatto and Sauvant,
2003).
158 J. van Milgen
References
Armstrong, D.G. (1969) Cell bioenergetics and energy metabolism. In: Lenkeit, W.,
Breirem, K. and Crasemann, E. (eds) Handbuch der Tierernährung. Parey,
Hamburg, Germany, pp. 385–414.
Baldwin, R.L. (1995) Modeling Ruminant Digestion and Metabolism. Chapman and
Hall, London.
Birkett, S. and de Lange, K. (2001a) A computational framework for a nutrient
flow representation of energy utilisation by growing monogastric animals.
British Journal of Nutrition 86, 661–674.
Birkett, S. and de Lange, K. (2001b) Calibration of a nutrient flow model of energy
utilisation by growing pigs. British Journal of Nutrition 86, 675–689.
Black, J.L., Campbell, R.G., Williams, I.H., James, K.J. and Davies, G.T. (1986)
Simulation of energy and amino acid utilisation in the pig. Research and
Development in Agriculture 3, 121–145.
Boisen, S. (2000) A simple nutrient-based production model for the growing pig.
In: McNamara, J.P., France, J. and Beever, D.E. (eds) Modelling Nutrient
Utilisation in Farm Animals. CAB International, Wallingford, UK, pp. 183–196.
Boisen, S. and Verstegen, M.W.A. (2000) Developments in the measurement of the
energy content of feed and energy utilisation in animals. In: Moughan, P.J.,
Verstegen, M.W.A. and Visser-Reyneveld, M.I. (eds) Feed Evaluation. Principles
and Practice. Wageningen Pers, Wageningen, Netherlands, pp. 57–76.
Brouwer, E. (1965) Report of sub-committee on constants and factors. In: Blaxter,
K.L. (ed.) Energy Metabolism. Proceedings of the 3rd Symposium held at Troon,
160 J. van Milgen
Introduction
Specification of the animal’s genetic potential for growth, efficiency and
robustness to environmental stressors is the key to successful livestock
production and breeding. The current practice in breeding is to record
performance traits on individuals and use regression to disentangle the
genetic from the environmental influence (Fig. 9.1, top). Traits considered
are typically related to growth, feed efficiency and, more recently,
robustness to various stressors. This widely used technique has several
shortcomings.
First, as noted by Bourdon (1998), the obtained estimated breeding
values (EBVs) do not represent the ‘true genetic potential’ (the maximum
performance given optimal conditions), because the data were measured
under sub-optimal conditions. Second, regression models are designed to
fit a specific data set rather than to represent the underlying biological
processes. This usually results in useful statistics (EBVs) appropriate for
the prevailing conditions. However, it provides a narrow scope of use; in
particular, empirical models assume simple linear relationships between
(combinations of) individual model components, which can cause problems
when extrapolating to conditions not covered in the data. In order to
avoid unexpected poor performance in environments that differ from the
data conditions, the covariances required for the EBVs need to be re-
estimated whenever the production conditions change.
Illustrating the genotype-phenotype relationship is Fig. 9.1, with
phenotypic performance at one end of the spectrum and the genotype at
the opposite end; mechanistic animal growth models represent this
relationship in the reverse direction to the mathematical models currently
Genes Genotype
specification Mathematical methods Measurable traits
Simulation model
Genetic potential
environmental, Phenotypic
for growth, Potential × dietary and performance
energy efficiency performance physiological
and robustness conditions
Model Model
input output
Knap et al. (2003) proposed two methods for estimating genotypic model
parameters: (i) direct measurements of the underlying traits represented
by the model parameters, combined with controlled environmental
conditions; and (ii) serial measurement of performance traits and fitting
the data to the mechanistic growth model itself (Model inversion).
are possible for a given set of input parameters, the inversion process
cannot produce unique values for the input parameters. All mechanistic
pig growth models fall into this category. Stochastic models, for example,
which integrate a random component, cannot be inverted by this method.
Most pig growth models are a composite of several algebraic equations
with an intrinsic hierarchical structure. This prevents a straightforward
reversion of the multiple modelling steps. It is therefore desirable to
represent the entire model by one system of equations with the model
input parameters as unknowns and the observable model outputs assumed
as known. The inversion process then corresponds to solving the system of
equations for the genotypic parameters for given values of predictable
performance traits. Exact solutions of a system of algebraic equations
require that the system is exactly determined, i.e. that the number of
variables to be solved for equals the number of equations.
Most pig growth models represent the genotype of the pig by two to five
parameters. This implies that for model inversion, the model should be
Evaluating Animal Genotypes through Model Inversion 169
into play when the ambient temperature is outside the pig’s comfort zone
(Bruce and Clark, 1979; Black et al., 1986; Knap, 2000b; Wellock et al.,
2003a), and abrupt changes in the rates of lipid retention under
malnutrition (Green and Whittemore, 2003; Wellock et al., 2003a).
Since many of these criteria depend on intermediate model results and
are therefore not known a priori, the representation of the model by a
system of equations suited for algebraic model inversion can become
problematic. Model inversion may thus require prior assumptions about
the conditions for which the modelled responses divide. These
assumptions should be based on the conditions from which the data are
obtained, and need to be re-validated for the calculated genotypic values
once the inversion has been carried out.
Implemented thresholds
The Implicit Function Theorem states the conditions for the existence of a
unique solution, but does not provide an explicit expression of the
solution. In fact, only few non-linear systems of equations can be solved by
Evaluating Animal Genotypes through Model Inversion 171
Under the assumption that the body protein and lipid mass of the pig at
time t0 are known and the required dietary information is available, the
hierarchical model was reformulated into systems of two equations, from
which unique values for LPmin and Prmax were obtained. The derivation
was built upon the assumptions of the original research project, i.e. that
information about feed intake, diet composition, body weight and backfat
depth were available and that initial body composition could be estimated
from initial body weight.
A system of equations with unique solutions for LPmin and Prmax could
only be obtained if the two constraints C1 and C2 below were satisfied. The
final Eqns 9.1a and 9.1b contain LPmin and Prmax as the only unknowns. All
other components are expressed in terms of initial and final body weight
(W0 and W1, respectively) or final backfat depth (BF1).
0.95 (W1 – W0) / dt = (0.001 × ((1.189 – E) × PRmax + B(W0))
Table 9.1. Constants of the model. Except for the dietary constants, DProt, DLys and DDE,
which were specified in Whittemore et al. (2003), all other constants are taken from de Lange
(1995).
Constants Explanation Estimated value
Aaa Post absorptive efficiency for utilizing amino acids and
dietary protein for growth 0.85
Dprot Dietary protein content (g/kg) 194
Dlys Dietary lysine content (g/kg) 11.4
Ep Gross energy content of protein (kJ/g) 23.6
DDE Dietary energy content (kJ/g) 14.5
LysBalP Lysine content of balanced protein (%) 7
Epe Energy cost of available protein excretion (kJ/g) 12.1
Epd Energy cost for protein deposition (kJ/g) 43.9
Eld Energy cost for lipid deposition (kJ/g) 52.8
Evaluating Animal Genotypes through Model Inversion 173
Table 9.2. Weekly least square means for daily food intake, body weight and
backfat depth used for the inversion of the model. The values are the results of the
statistical models for ‘Landrace type’ pigs, published in Green et al. (2003).
Week Daily food intake (kg) Body weight (kg) Backfat depth (mm)
Table 9.3. Weekly estimates of the genotypic parameters PRmax and LPmin from algebraic
model inversion (columns 2 and 3) and inversion through optimization (columns 4 and 5).
RMSD is the root of the mean of the squared deviations of predicted body weight and backfat
depths from the observations (Table 9.2) using the optimized parameter values.
Week PRmax LPmin (alg. PRmax LPmin RMSD
(alg. inversion) inversion) (optimization) (optimization) (optimization)
Implicit Function Theorem for the existence of a unique solution with the
weekly least square means were satisfied. The numerical solver fzero of Matlab
(Matlab 6.5, 2002) was used to calculate values for PRmax and LPmin as
solutions of the corresponding systems of equations. The weekly solutions are
in Table 9.3 (columns 1 to 3). All solutions satisfied constraints C1 and C2.
The value for PRmax obtained through inversion of the de Lange
model ranged between 114 and 171 g/day (Table 9.3), which agrees with
estimates provided in the literature, i.e. 99 g/day (SE 4 g/day) to 212 g/day
(SE 29 g/day) (Knap, 2000a). Literature estimates for the lipid to protein
174 A.B. Doeschl-Wilson et al.
ratio at maturity vary between 0.97 (SE 0.28) and 5.16 (SE 0.50). In
comparison with these values, the range between 0.53 and 1.22 for the
estimates for the minimal lipid to protein ratio obtained from the algebraic
model inversion also appears realistic. In the last row of Table 9.3,
solutions corresponding to a time span of 12 weeks are shown. The
calculated values are unrealistic, demonstrating that algebraic model
inversion fails if observations are far apart.
Although the solutions of PRmax and LPmin appear realistic, they are
not, as assumed, constant over time. Over the 12 weeks, PRmax and LPmin
both increase systematically. This apparent time trend contradicts the
assumption that genotypic parameters are time invariant. Agreement of
the model predictions for body weight and backfat depths with the
observed least square means is only possible if the genotypic parameters
are allowed to vary over time, which indicates that the growth trends for
body weight and backfat depth predicted by the model differ from the
observed ones. This suspicion is confirmed in Fig. 9.2, which plots those
growth curves corresponding to fixed values of the genotypic parameters
together with the least square means from the data. For the parameter
combinations considered, the predicted backfat depth curves differ
strongly from the data. Sensitivity analysis indicates that no combination of
genotypic parameters can produce a trend similar to the data.
(a)
110
observed LSM
100 algebraic inversion, week 12
algebraic inversion, week 6
inversion through optimization, weeks 1–12
90 inversion through optimization, week 12
70
60
50
40
30
0 1 2 3 4 5 6 7 8 9 10 11 12 13
Week
(b)
16
observed LSM
algebraic inversion, week 1
14 algebraic inversion, week 12
inversion through optimization, weeks 1–12
inversion through optimization, week 12
Backfat depth (mm)
12
10
4
0 1 2 3 4 5 6 7 8 9 10 11 12 13
Week
Fig. 9.2. (a) Predicted body weight for different solutions for the genotypic parameters of the
model. The estimates of the genotypic parameters used to generate these plots are
presented in Table 9.3. (b) Predicted backfat depth for different solutions for the genotypic
parameters of the model. The estimates of the genotypic parameters used to generate these
plots are presented in Table 9.3.
176 A.B. Doeschl-Wilson et al.
Algebraic model inversion relies upon the assumption that a unique set of
model input parameters exists for which the model predictions equal the
data. Uncertainties and errors in the model itself or in the data are
ignored. Also, if the number of parameters to be determined through
inversion exceeds the number of parameters for which a unique solution
of the inverse model exists, the algebraic approach forces the modeller to
restrict the number of unknowns by guessing the values of some
parameters. This restriction may lead to poor estimates for the remaining
parameters, since they are dependent on these guesses.
By treating the estimation of genotypic parameters by model inversion
as an optimization problem, various sources of uncertainty can be taken
into account. Instead of determining the set of genotypic parameters for
which the model predictions exactly match the data, inversion through
optimization determines the set of parameters for which the discrepancy
between predictions and data is minimal.
Common problems
The model uses four traits to describe the pig’s intrinsic capacity for
growth and energy utilization. The genetic potentials for protein and lipid
growth are modelled by Gompertz functions and full allometry between
body protein and lipid is assumed. The description of the genotype’s
growth potential then requires three independent parameters: mature
protein mass (Pmat), the ratio between mature lipid and protein mass
(LPmat) and the standardized rate parameter B* = BGomp × Pmat0.27.
The resource-demanding processes other than protein and lipid
growth are categorized as ‘maintenance processes’. The current version of
the model explicitly deals with the energy requirements for protein
turnover and thermoregulation, which depend on body composition. The
requirements for all other maintenance processes, including physiological
service functions, are combined into a single parameter (MEmaint)
calculated as a simple function of metabolic body weight: (MEm0 + b.BW)
× BW0.75, where MEm0 is the fourth genotype-specific model parameter.
The phenotypic value for each of the physiological traits can be de-
Evaluating Animal Genotypes through Model Inversion 179
composed according to
Pij = µ + Ai + PEi + eij (9.2)
where Pij is the phenotype of animal i on day j, µ is the population mean for
the trait, and Ai, PEi and eij are its additive genetic and permanent
environmental deviations, and its random environmental deviation on day j.
The genetic deviation is determined according to
A i = rannorA i × h2 × σ P (9.3)
with rannorPEi also being a random deviation drawn from the standard
Normal distribution. The random environmental variation is modelled
according to
e ij = Ri j × σ e (9.5)
Estimates for population means, CVs and heritabilities for the four
underlying traits, produced by four optimization runs with different
random number sequences and different starting points for the
Evaluating Animal Genotypes through Model Inversion 181
100000 0.30
0.25
10000
Runtime (days)
0.20
CV of solutions
1000
runtime
0.15
100
0.10
10
0.05
CV of soutions
0 0.00
0 10000 20000 30000 40000 50000
Population size
Fig. 9.3. Runtime for the optimization process and average coefficient of variation (CV) for
the calculated covariances and heritabilities for the predicted performance traits as affected
by population size.
Table 9.4. Genetic correlations (upper triangle of the white area), heritabilities (on
the diagonal) and phenotypic correlations (lower triangle of the white area),
together with phenotypic means (grey area) for a PIC Sire Line as predicted from
data analysis and from the growth model combined with optimization routines from
four different simulations.
Table 9.5. Estimates of the genetic parameters of the growth model corresponding to a PIC
Sire Line obtained from four simulations.
Simulation results Parameter Pmat LPmat B* MEm0
derived from data of a genotype, bred from the same sireline as above but
with a different damline. Differences between the parameter sets A and B
versus the sets C and D (higher values for Pmat and LPmat) are clearly
reflected in the associated growth curves (upper versus lower curves in Fig.
9.4). If genetic correlations together with heritabilities, and repeated
measurements from various growth stages, were simultaneously available
for a genotype (which is not the case here), the most appropriate
parameter combination could be chosen according to the model fit to
empirically established growth curves. Alternatively, phenotypic growth
trends could be directly included in the optimization criteria using
dynamic control of the objective functions (as pioneered by Kinghorn et al.,
2002).
Conclusion
1.2
1.0
ADG (kg/day)
0.8
Empirical estimates
Model, parameter set A
0.6
Model, parameter set B
Model, parameter set C
Model, parameter set D
0.4
60 80 100 120
Body weight (kg)
Fig. 9.4. Average daily gain relative to body weight as predicted from empirical studies for a
crossbred type and as predicted for the pure bred type by the growth model according to the
parameter sets A to D specified in Table 9.5. The crossbred was produced from the same
sireline as that used for model inversion, but from a different damline.
Evaluating Animal Genotypes through Model Inversion 185
is crucial for accurate model predictions and thus for the appropriate use
of growth models. Growth models also give a window on to underlying
physiological traits that are intrinsic drivers of observed phenotype,
whatever environment it is expressed in. Their approach can be used for
making genetic evaluation of animals for these underlying traits. There
are prospects that this will give better gains – through an integration of
different observed traits in a more intelligent manner than the linear
statistical approach that conventional selection index theory in animal
breeding involves, but also through a better choice of observed traits to
measure (including choices of ages and diets), leading to a more accurate
evaluation of both underlying physiological traits, and observed traits as
expressed in different environments, and thus of the genetic potential.
Combined with phenotypic performance data, which are relatively easy
to measure, model inversion has been identified as a promising tool to
derive the desired specification of the genotype intrinsic physiological
traits. The case studies presented here demonstrate that conclusive
estimates for the genotypic specifications are only possible if: (i) the growth
model simulates the physiological mechanisms of pig growth sufficiently
accurately that observed growth trends can be reproduced; and (ii) if
sufficient data expressing the genetic potential for growth and energy
utilization are available.
Acknowledgements
References
Baldwin, R.L. (1976) Principles of modelling animal systems. Proceedings of the
NewZealand Society of Animal Production 36, 128–139.
Black, J.L. (1995) The testing and evaluation of models. In: Moughan, P.J.,
Verstegen, M.W.A. and Visser-Reyneveld, M.I. (eds) Modelling Growth in the Pig.
EAAP publication no.78, Wageningen Pers, Netherlands, pp. 23–31.
Black, J.L., Campbell, R.G., Williams, I.H., James, K.J. and Davies, G.T. (1986)
Simulation of energy and amino acid utilisation in the pig. Research and
Development in Agriculture 3, 121–145.
Bourdon, R.M. (1998) Shortcomings of current genetic evaluation systems. Journal
of Animal Science 76, 2308–2323.
186 A.B. Doeschl-Wilson et al.
Bruce, J.M. and Clark, J.J. (1979) Models of heat production and critical
temperature for growing pigs. Animal Production 28, 353–369.
de Lange, C.F.M. (1995) Framework for a simplified model to demonstrate
principles of nutrient partitioning for growth in the pig. In: Moughan, P.J.,
Verstegen, M.W.A. and Visser-Reyneveld, M.I. (eds) Modelling Growth in the Pig.
EAAP publication no.78, Wageningen Pers, Netherlands, pp. 151–163.
Emmans, G.C. (1988) Genetic components of potential and actual growth. In:
Land, R.B., Bulfield, G. and Hill, W.G. (eds) Animal Breeding Opportunities.
British Society of Animal Production, Edinburgh, UK, pp. 153–181.
Ferguson, N.S. and Gous, R.M. (1993) Evaluation of pig genotypes. 1. Theoretical
aspects of measuring genetic parameters. Animal Production 56, 233–243.
Green, D., Brotherstone, S., Schofield, C.P. and Whittemore, C.T. (2003) Food
intake and live growth performance of pigs measured automatically and
continuously from 25 to 115 kg live weight. Journal of the Science of Food and
Agriculture 83, 1150–1155.
Green, D.M. and Whittemore, C.T. (2003) Architecture of a harmonized model of
the growing pig for the determination of dietary net energy and protein
requirements and of excretions into the environment. (IMS Pig). Animal Science
77, 113–130.
Kinghorn, B.P., Meszaros, S.A. and Vagg, R.D. (2002) Dynamic tactical decision
systems for animal breeding. In: Proceedings of the 7th World Congress on Genetics
Applied to Livestock Production 33, 179–186.
Knap, P.W. (1999) Simulation of growth in pigs: evaluation of a model to relate
thermoregulation to body protein and lipid content and deposition. Animal
Science 68, 655–679.
Knap, P.W. (2000a) Time trends of Gompertz growth parameters in “meat type”
pigs. Animal Science 70, 39–49.
Knap, P.W. (2000b) Stochastic simulation of growth in pigs: relations between body
composition and maintenance requirements as mediated through protein
turnover and thermoregulation. Animal Science 71, 11–30.
Knap, P.W., Roehe, R., Kolstad, K., Pomar, C. and Luiting, P. (2003)
Characterization of pig genotypes for growth modelling. Journal of Animal
Science 81 (E-suppl. 2), E187-E195. www.asas.org/symposia/03esupp2/jas2593.pdf.
Krantz, S.G. (1951) The Implicit Function Theorem: History, Theory, and Applications.
Birkhäuser, Boston, Massachusetts.
Landgraf, S., Roehe, R., Susenbeth, A., Baulain, U., Knap, P.W., Looft, H., Plastow,
G.S. and Kalm, E. (2002) Biological growth model as a new selection strategy
for improvement of feed efficiency in swine. Veterinarija ir Zootechnica 18,
84–89.
Matlab, 6.5. (2002) The Language of Technical Computing 1984–2002. The
MathWorks, Inc.
Mayer, D.G., Kinghorn, B.P. and Archer, A.A. (2005) Differential evolution – an
easy and efficient evolutionary algorithm for model optimisation. Agricultural
Systems 83, 315–328.
Parks, J.R. (1982) A Theory of Feeding and Growth of Animals. Springer, Berlin.
Pomar, C., Kyriazakis, I., Emmans, G.C. and Knap, P.W. (2003) Modeling
stochasticity: dealing with populations rather than individual pigs. Journal of
Animal Science 81 (E-suppl. 2), E178–E186.
Price, K. and Storn R. (1997) Differential evolution. Dr. Dobb’s Journal, April, 18, 24
and 78.
Roehe, R., Lorenzo Bermejo, J., Landgraf, S., Knap, P.W. and Kalm, E. (2002)
Evaluating Animal Genotypes through Model Inversion 187
Introduction
Homeothermic animals dissipate heat in the form of both sensible (H) and
latent (λE) heat (Mount, 1979). The heat flow density from the animal to
the environment (expressed normally in W m2) depends on the structure
190 O.A. Blanco and R.M. Gous
and physiological state of the animal (body size, feather cover, respiratory
rate, etc.) and on the thermal gradient (in the case of H) or the vapour
pressure gradient (∆eV) (in the case of λE) between the surface of the
animal and the surrounding environment (Mount, 1979). The level of
micro-environmental radiation affects H, whilst air velocity (u) affects both
λE and H (Monteith and Unsworth, 1990) (see Fig. 10.1).
According to the peculiarities of heat exchange with its surroundings,
a farm animal can be modelled as a system with two interfaces, i.e. the
‘body surface’ (Ib) and the ‘surface of the anterior respiratory tract’ (Ir)
Fig. 10.1. Partition of heat dissipated by a chicken. The circular detail shows the
components of the heat exchange between breathed air and mucous surface of the upper
respiratory tract, where the double-headed arrow indicates the movement of the circulating
air. λEf: dissipation of latent heat from feathered areas; Hf: dissipation of sensible heat from
feathered areas; rHRf: thermal resistance of the boundary layer of feathered areas to
convective and radiant heat transfer; rvf: resistance of the boundary layer of feathered areas
to water vapour transfer; λEa: dissipation of latent heat from bare appendages; Ha:
dissipation of sensible heat from bare appendages; rHRa: thermal resistance of the boundary
layer of bare appendages to convective and radiant heat transfer; rva: resistance of the
boundary layer of bare appendages to water vapour transfer; λEr: dissipation of latent heat
from the surface of the anterior respiratory tract (interface Ir); Hr: dissipation of sensible heat
from interface Ir; rHRr: thermal resistance of the boundary layer of interface Ir to latent and
radiant heat transfer; rvr: resistance of the boundary layer of the interface Ir to water vapour
transfer; Rni: micro-environmental isothermal net radiation; u: micro-environmental air
velocity; C: thermogenic core; Sk: skin; F: feather coat; BL: Boundary-layer (note: due to
space restrictions, it was impossible to represent rva, rHRa, eVa and Ta on the wattles) (based
on Campbell and Norman, 1998 and Blanco et al., 2004a,b).
Micro-environmental Conditions in Simulation Models 191
(Blanco, 2004a,b). A thin layer of air over the surface of each of these
interfaces, called the boundary-layer (BL), imposes resistance to heat and
water vapour loss to the environment (Monteith and Unsworth, 1990) (Fig.
10.1). The micro-environmental variables may impose dissimilar demands
on each of the interfaces due to the different values of boundary-layer
resistances (Bakken, 1981; Blanco et al., 2004a,b), the various processes
controlling the heat exchange at each interface, and the anatomical and
physiological differences between respiratory tract and body surface.
rvf.) The value of rvs (i.e. rva and rvf) is inversely related to u, whilst rvr is
inversely proportional to the respiratory rate (RR) (McArthur, 1981;
McArthur, 1987; Turnpenny et al., 2000a,b).
⎫
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎬
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎭
Fig. 10.2. Diagram showing the components of the action of the micro-environment
and the consequent micro-environmental effect or animal response. rHR: Thermal resistance
of the boundary layer to radiant and convective heat transfer; rv: resistance of the boundary
layer to water vapour transfer. Note that BL is clearly included in the environment.
and for the water vapour content of the air. Their empirical nature is the
result of the inclusion of multipliers experimentally derived. Considering the
importance of the physical and thermoregulatory characteristics of the
broiler on the thermal relationship between bird and environment, as well as
the variation of such properties according to factors such as maturity
(Poczopko, 1981), breed (Özkan et al., 2002), and adaptation at genetic,
phenotypic and epigenetic levels (Nichelman and Tzschentke, 2002), the
usefulness of these parameters is very limited, unless new experimental
multipliers are calculated for each new case. Therefore, these discomfort
indices do not fulfil the first criterion of being a suitable parameter.
In addition, the inductive approach used in the construction of the
parameters included in this first group neither allows for understanding the
animal–micro-environment interaction, nor for analysing the role of each
variable as part of such a relationship. This is basically because they have
been conceived for purposes other than the creation of models for the
analysis of the action and effect of micro-environmental conditions on the
heat balance of a broiler. They were originally used for classifying animals as
heat stressed and non-heat stressed, according to experimentally determined
comfort ranges, constituting a safer way of controlling the environment and
a moderately accurate way of predicting production results. We have not
found them useful for our model, but do not discard the possibility that such
parameters could be applied in models approaching the relationship
between animal and micro-environment from another viewpoint. Examples
of such parameters are the Discomfort Index (Tselepidaki et al., 1995), the
Thermal Hygrometric Index (Esmay, 1978; Tao and Xin, 2003) and the
Wind Chill Factor (Siple and Passel, cited by Starr, 1981).
A second group of parameters, which we term a family of equivalent
temperatures, has been identified. These parameters, derived following
the criteria proposed by Monteith (1974) and Mount (1979), include
environmental apparent equivalent temperature (θ*e) (Mount, 1974),
environmental temperature (Te) (McArthur, 1981), effective temperature
(Tef) (Monteith and Unsworth, 1990), equivalent temperature (Monteith
and Unsworth, 1990) and operative temperature (Campbell and Norman,
1998). Such parameters are characterized by the deductive approach used
in their construction.
The equations for the parameters of the second group combine the
value of two or more micro-environmental variables with the factors
accounting for the physical properties of the animal. These parameters
have two advantages over those from the former groups:
1. They have not been developed inductively; hence they do not have an
empirical character. Therefore, the three first characteristics of an ideal
parameter are, at least, partially fulfilled.
2. The inclusion of the thermal resistances of the boundary layer in some
of these parameters results in a thermal equivalent of the strain imposed
by the immediate surroundings on the bird, and accounts therefore for the
action of the environment on the animal.
198 O.A. Blanco and R.M. Gous
Tef
⎧
⎨
T⎩ air
Rni
Heat flux density (W m–2)
Fig. 10.3. Temperature vs heat flux density diagram for the graphical determination of Tef.
Tair: air temperature, Tef: effective temperature, Rni: isothermal net radiation, ζ: Radiation
increment (after Monteith and Unsworth, 1990).
200 O.A. Blanco and R.M. Gous
u1>u2>u3
Tef3
Temperature (°C)
rHR1<rHR2<rHR3
3
Tef2
Tef1<Tef2<Tef3
2
Tef1
1
Tair
Rni
Fig. 10.4. Temperature vs heat flux density diagram for the graphical determination of Tef
considering three different u, with the same Tair and Rni. Tair: air temperature, Tef: effective
temperature, Rni: Isothermal net radiation, u: air velocity, rHR: thermal resistance to radiant
and convective heat transfer through the boundary layer (after Monteith and Unsworth,
1990).
Tef(n-ins)
Temperature (°C)
Tef(ins)
Tair
Rni(ins) Rni(n-ins)
Fig. 10.5. Comparison between Tef values reached in two comparable hypothetical houses
with and without insulated roofs in the same season, same air temperature (Tair) and same
air velocity. Rni(ins): isothermal net radiation loaded on chickens under insulated roof; Rni(n-ins):
isothermal net radiation loaded on chickens under non-insulated roof; Tef(ins): effective
temperature for chickens under insulated roof; Tef(n-ins): effective temperature for chickens
under non-insulated roof (after Monteith and Unsworth, 1990 and Blanco et al., 2003).
Micro-environmental Conditions in Simulation Models 201
environmental conditions in a broiler house: for the same value of Tair and
u, a broiler exposed to higher Rni, would experience higher Tef.
Limitations of Tef
The most evident limitation of Tef (Monteith and Unsworth, 1990) is its
inability to account for the action of the micro-environmental vapour
pressure. In order to include such a variable, Monteith and Unsworth
(1990) returned to the concept of apparent equivalent temperature (Teq)
introduced by Monteith (1974). As in the case of the equation developed in
1974, this apparent equivalent temperature (here symbolized with Te*) is
obtained after the addition of a humidity increment (ξ) to Tair, as shown in
Eqn 10.5.
Te* = Tair + ξ= Tair + e (Tair) (γ*)–1 (10.5)
where e(Tair) is the vapour pressure of the micro-environmental air (kPa) at
a given Tair (°C), and γ* is the psychrometric constant in its apparent form
(kPa K1). The mathematical expression for the apparent psychometric
constant γ* is γ* = γ [ rv (rHR)1], where γ is the psychrometric constant
(=0.066 kPa K1). The form γ* has the advantage of accounting for the
thermal resistance for convective and radiant heat exchange through BL
(rHR) and the resistance to water vapour transfer through BL (rv) in a same
term.
Monteith and Unsworth (1990) obtained an environmental apparent
equivalent temperature (Ter*), after combining Te* and a radiant
increment (ζ). The mathematical expression of Ter*, comparable to that for
environmental apparent equivalent temperature (θ*e) reported by
Monteith (1974), is presented here as Eqn 10.6.
Ter* = Te* + rHR (ρcp)–1 Rni = Tair + eVair rHR (rv γ*)1 + rHR (ρcp)–1 Rni (10.6)
where Ter* is the environmental apparent equivalent temperature (°C).
As in the case of θ*e (Monteith, 1974), in steady state conditions, Ter*
can be applied to both interfaces of the system, i.e. Ib and Ir. In the
particular case of Ir, it should be remembered that the value of the
resistance to water vapour transfer through its boundary, rvr, depends on
the respiratory rate (RR). McArthur (1981) published an empirical
equation based on experimental data by Hutchinson (1954), reproduced
here as Eqn 10.7.
rvr = [(9.1 × 105 RR) + 9.0 × 103]–1 (10.7)
Problems appear when attempting to create a unique value of Ter* to
represent the conditions of both interfaces. Since these two interfaces act in
parallel, and the conditions affecting each surface are substantially
202 O.A. Blanco and R.M. Gous
Estimation of
latent heat
Re dissipation from
estimation respiratory
of rvr interface
(λEr)
1 s cycle
Tb RR Estimation
of S
(S=H-λEr)
Information
used in the OUTPUT Recalculation Recalculation
next cycle of Tb of RR of Tb
estimations RR
Fig. 10.6. Diagram of a simulation model for estimating the effect of micro-environmental
conditions on the heat balance of a broiler, including an alternative approach for estimating
the latent heat dissipation from the respiratory tract. The estimations are performed once
every second, as indicated by the circular arrow; Tef: Effective temperature estimated for a
particular broiler under given environmental conditions; TNT: particular value of Tef at which
the animal is under least thermoregulatory effort; H: sensible heat loss; HE: Heat excess; λEr:
latent heat loss from the respiratory tract; S: heat storage; Tb: body temperature; RR:
Respiratory rate; rvr: resistance of the boundary-layer of interface Ir to water vapour transfer.
Micro-environmental Conditions in Simulation Models 205
Conclusions
The authors of this chapter consider that the way in which micro-
environmental conditions are usually accounted for in simulation models
needs to be revised.
In the first place, and bearing in mind the differentiation of the
environment directly surrounding the animals, micro-environmental
conditions should always be considered in models strongly influenced by
the heat exchange between animal and environment as, for example, those
dealing with heat balance and energy utilization.
Secondly, considering the importance of each of the micro-
environmental variables on the heat exchange between animal and
environment, models should at least consider Tair, eVair, u and Rni in order
to produce balanced estimations.
A third point to take into account: since the BL of the animal plays a
fundamental role in the relationship with its surroundings, its
characteristics should be considered when accounting for the
environmental conditions. The description of the micro-environmental
conditions in the context of the BL allows for the estimation of what we
have termed micro-environmental action. This can be a starting point for
further modelling procedures in order to estimate the consequent effect of
those micro-environmental conditions on the heat balance of the chicken.
Regarding mathematical expressions that account for the micro-
environmental action, two groups of parameters have been identified in
the literature. Among the parameters belonging to the group of equivalent
temperatures, effective temperature (Tef) (Monteith and Unsworth, 1990)
best suited the needs of the authors of this chapter. Although most of the
parameters of this group adequately account for the micro-environmental
action by considering the properties of the boundary layer, they need to be
adapted to the following peculiarities of the thermoregulatory processes of
non-sweating animals such as poultry:
1. Water vapour dissipation is important from both the body surface and
the upper respiratory surface.
2. Both body surface and mucous membranes of the upper respiratory tract
are separate interfaces of a system tract (Ib and Ir, respectively).
3. Because different factors control heat and vapour exchange, the action of
the environment on those two interfaces cannot be included in the same
parameter.
4. As RR is directly related to the value of Tb, which is a product of the
action of the environmental conditions to be specified by the parameter,
the formulation of a parameter accounting for the action of the micro-
environment in both interfaces is infeasible.
An alternative approach in dynamic models would be to estimate the latent
heat loss from the respiratory tract as a consequence of the effect of the
micro-environmental conditions on the sensible heat dissipation. By doing
so, the introduction of a respiratory component on a parameter accounting
206 O.A. Blanco and R.M. Gous
for the action of the micro-environment would not be necessary, since the
latent heat loss from the respiratory tract could be estimated by
considering the value of the resulting respiratory rate and the gradient of
vapour pressure between the circulating air and the surface of the anterior
respiratory tract.
Finally, we strongly believe that encouraging the interaction between
animal scientists and environmental physiologists would be a fruitful
means of improving the way in which the environmental conditions are
specified in models dealing with the energy balance of poultry.
References
Bakken, G.S. (1981) A two-dimensional operative temperature model for thermal
energy management by animals. Journal of Thermal Biology 6, 23–30.
Barnas, G.M. and Rautenburg, W. (1987) Temperature control. In: Seller, T.J. (ed.)
Bird Respiration, Vol. I. CRC Press, Boca Raton, Florida, pp. 131–153.
Bernstein, M.H. (1971) Cutaneous and respiratory evaporation in the painted
quail, Excalfactoria chinensis, during ontogeny of thermoregulation. Comparative
Biochemistry and Physiology (Part A) 38, 611–617.
Blanco, O.A., Gous, R.M. and Savage, M.J. (2003) Biophysical basis for controlling
the effect of microenvironment on the heat balance of broilers. In: Proceedings
of the 22nd Scientific Day WPSA Southern African Branch, Pretoria, pp. 123–138.
Blanco, O.A., Gous, R.M. and Savage, M.J. (2004a) A simulation model for estimating
the effect of warm micro-environmental conditions on the heat balance of a
broiler. In: Proceedings of the XII World Poultry Congress, Istanbul, Turkey.
Blanco, O.A., Gous, R.M. and Savage, M.J. (2004b) Modelling the physiological
responses of a broiler exposed to micro-environmental conditions above
thermoneutrality. In: Proceedings of the 23rd Scientific Day, World Poultry Science
Association Southern African Branch, Pretoria, pp. 10–23.
Campbell, G.S. and Norman, J.M. (1998) Environmental Biophysics. Springer, New
York.
Cena, K. (1974) Radiative heat loss from animals and man. In: Monteith, J.L. and
Mount, L.E. (eds) Heat Loss from Animals and Man: Assessment and Control.
Butterworths, London, pp. 33–58.
Charles, D.R. (1974) The definition and measurement of the climatic environment
in poultry houses. In: Morris, T.R. and Freeman, B.M. (eds) Energy
Requirements for Poultry. British Poultry Science, Edinburgh, UK, pp. 73–77.
Charles, D.R. (1994) Comparative climatic requirements. In: Wathes, C.M. and
Charles, D.R. (eds) Livestock Housing. CAB International, Wallingford, UK,
pp. 3–24.
Charles, D.R., Elson, H.A. and Haywood, M.P.S. (1994) Poultry housing. In:
Wathes, C.M. and Charles, D.R. (eds) Livestock Housing. CAB International,
Wallingford, UK, pp. 249–272.
Czaric, M. and Tyson, B.L. (1990) Temperature Variations in a High Density, Curtain-
sided Laying House. ASAE Paper No. 90–4059. ASAE, Michigan.
Esmay, M.L. (1978) Principles of Animal Environment. AVI Publishing, Westport,
Virginia.
Hutchinson, J.C.D. (1954) Evaporative cooling in fowls. The Journal of Agricultural
Sciences (Cambridge), 45, 48–59.
Micro-environmental Conditions in Simulation Models 207
Wathes, C.M. and Clark, J.A. (1981a) Sensible heat transfer from the fowl:
boundary-layer resistance of a model fowl. British Poultry Science 22, 161–173.
Wathes, C.M. and Clark, J.A. (1981b) Sensible heat transfer from the fowl: thermal
resistance of the pelt. British Poultry Science 22, 175–185.
Xin, H., Berry, I.L., Tabler, G.T. and Barton, T.L. (1994) Temperature and
humidity profiles of broiler houses with experimental conventional and tunnel
ventilation systems. Transactions of the ASAE 10, 535–542.
Zhou, W.T. and Yamamoto, S. (1997) Effects of environmental temperature and
heat production due to food intake on abdominal temperature, shank skin
temperature and respiration rate of broilers. British Poultry Science 38, 107–114.
11 Using Physiological Models to
Define Environmental Control
Strategies
M.A. MITCHELL
Roslin Institute, Roslin, Midlothian, EH25 9PS, UK
malcolm.mitchell5@btinternet.com
Introduction
systematic errors and simplicity. A given model is usually fitted and the
optimal parameters determined by minimization of an objective function,
usually the sum of squared errors (Garfinkel and Fegley, 1984).
Models of all types are commonly applied in animal biology and
agriculture and are employed to describe a wide range of processes
including growth and the influence thereon of genetics, nutrition and
husbandry (Wellock et al., 2003a, 2004a); reproductive performance; the
effects of environmental and other stressors upon animal production and
well-being (Wellock et al., 2003b, 2004b); behavioural and physiological
needs; factors influencing animal health; toxicity and environmental
hazard and management and risk extrapolation. The outputs of such
models may provide the basis for the setting of standards, optimum
practices and nutritional strategies, operational control systems and
practical guidelines and legislation. In the current review attention is
focused particularly upon the thermal environments to which livestock are
exposed and the modelling of animal responses in the context of
environmental specification and control.
health and welfare. Clearly the heat exchange between an animal and its
immediate environment and the thermoregulatory responses to the
imposed conditions are the major consideration when determining
optimum temperature and humidity ranges in addition to predictions of
effects upon production parameters. Thus, in cattle, dynamic models of
responses to thermal loads have been developed based upon fractal
analysis of tympanic temperature responses that allow identification of the
environmental temperature at which heat stress is apparent (Hahn, 1999).
The fractal dimensions for body temperature may be employed to
categorize thermal stress (environmental temperature) from slight through
moderate to severe. Measurements of respiration rate changes were used
to further refine this ‘stress’ model. Dynamic responses provide the basis
for evaluating stress thresholds and differentiation of stress levels in cattle
and therefore can be used as criteria for environmental management
(Korthals et al., 1997; Hahn, 1999). In pigs, stochastic models have been
developed and evaluated that predict the effects of the thermal
environment upon food intake, daily weight gain, food conversion
efficiency and metabolic heat production (Wellock et al., 2003a). The model
indicated that within the apparent zone of thermal comfort there are no
temperature effects upon the production variables. A heat balance model
of the pig proposed by Fiahlo et al. (2004) and derived from first principles
of heat and mass transfer, simulates heat loss, heat balance and deep body
temperature responses in animals exposed to different thermal
environments. It is suggested that this model might be integrated into a
more comprehensive physiological model of the pig for precise definition
of optimum production environments and limits for imposed thermal
loads. Important advances in understanding the relationships between
animals and their thermal environments can be made through entirely
theoretical models of heat exchange which cannot always be evaluated by
comparison with experimental or empirical data. An idiosyncratic,
although useful, example has been presented by Phillips and Heath (2001)
in which the heat loss from ‘Dumbo’ (© Walt Disney Company) has been
modelled. Using fundamental principles, these authors have demonstrated
that Dumbo may often be at risk of losing more heat than he can produce
at rest and therefore will be at risk of hypothermia. The large pinnae area
associated with this character may have evolved to favour heat loss during
the hypermetabolism of flying!
Mathematical models describing both evaporative and non-evaporative
heat losses from livestock in relation to the thermal environment have been
developed (e.g. Ehrlemark, 1993). Such models provide valuable predictive
data based on the thermal properties of the animals but without reference
to the internal mechanisms of the thermoregulatory systems. The model
outputs can underpin determination of the thermoneutral zone and upper
critical temperature and ultimately may be employed as the basis for control
systems or algorithms. Thermal balance models for livestock, including
cattle and sheep (outdoors) and pigs and chickens (indoors), have also been
described by Turnpenny et al. (2000a,b). These mathematical models,
Using Physiological Models to Define Environmental Control 213
Thermal Comfort
The term ‘thermal comfort’ has evolved in the discipline of human thermal
physiology. The ‘Zone of Thermal Comfort’ in man is defined as
the range of ambient temperatures, associated with specified mean radiant
temperature, humidity and air movement within which a human in specified
clothing expresses indifference to the thermal environment for an indefinite
period.
(Commission for Thermal Physiology of the International Union of
Physiological Sciences, 2003)
214 M.A. Mitchell
It is abundantly clear that the heat and mass exchange between an animal
and its immediate environment is dependent not only upon the dry bulb
temperature but upon the water vapour pressure or density, air movement
and radiant temperature. Whilst it is possible to model the effects of each
variable upon animal performance, physiological and metabolic responses
and thermal comfort, it is often desirable to develop a single integrated
index of imposed thermal load that will allow prediction of the biological
response. Various indices have been proposed for man and other animals
with a particular focus upon the prediction of the risk of heat stress or cold
stress. The situation may be summarized thus:
Ta is only one of several physical factors determining heat exchange with the
environment which occurs via ‘dry’ (conduction, radiant, convection) and ‘wet’
(evaporation) mechanisms. In addition to depending upon Ta each mechanism
also depends upon one or more of the following physical factors, humidity, air
velocity, barometric pressure, contact with housing structures and material e.g.
bedding/litter and effective radiant field.
(Romanovsky et al., 2002)
The age, physiological status, acclimation state, diet and food intake and
behavioural freedom of the animals will all contribute to the integrated
heat exchange of the animals and thus to any assessment of desirable
thermoneutral conditions.
216 M.A. Mitchell
Severe
Moderate
Mild
psychrometric constant (Eqn 11.2) and describes the total heat exchange
between a wetted surface and the environment (Eqn 11.1).
θ*app. = T + ( e/γ* ) (11.1)
where θ*app = AET
T = absolute temperature (K)
e = water vapour pressure (mbar)
γ* = corrected psychrometric constant (mbar/K)
γ* = γ (rv/rh) (11.2)
where rv = the resistance to water vapour transfer (s/m)
rh = the resistance to heat transfer (s/m).
AET may be calculated from the dry bulb temperature and the relative
humidity (RH) alone (Eqn 11.3).
10
(30.5905 −8.2⋅Log 10 ( K )+ 0.0024804 K − 3142.31 / K ) ⋅Φ (11.3)
AET = T +
(
0.93 ⋅ 0.0006363601K + 0.472 )
where
T = Observed temperature (°C)
K = T corrected to Kelvin (°C + 273.16)
Φ = Observed relative humidity (RH/100).
The derivation and application of AET is perhaps best described
graphically (Fig. 11.2). If a sample of air at 18°C (t) and 10 mbar vapour
pressure (e) is represented by point X then the line YXZ with a slope of -γ
yields the wet bulb temperature Y (12°C) and the equivalent temperature Z
(33.3°C). The line QX gives the dew point temperature (7.1°C) and the line
XP gives the saturation vapour pressure (20.6 mbar). If water vapour was
condensed adiabatically from a sample of saturated air (point Y) then the
temperature and water vapour pressure changes are described by the line
YX. When all the water vapour has condensed (e = 0) then t = Z
(Equivalent Temperature). In the calculation of AET the corrected
psychrometric constant replaces γ as described above.
Using the AET approach, the combinations of temperatures and
humidities that produce equivalent biological effects were determined. The
relationship between change in deep body temperature and AET is
presented in Fig. 11.3. It is clear that the response to thermal load is
similar to the hypothetical curve presented in Fig. 11.1 and as such allows
definition of temperature-humidity combinations imposing mild, moderate
and severe physiological stress. Similar response patterns were observed
for changes in blood pH and pCO2 which are used to assess homeostatic or
thermoregulatory effort.
Physiological stress response modelling, exploiting the concept of AET,
has thus allowed identification of ‘safe’, ‘alert’ and ‘danger’ combinations of
temperature and humidity which equate to mild, moderate and severe
220 M.A. Mitchell
30
Q
10 X
Slope
–γ
M Z
0
0 10 20 30
Temperature (°C)
Fig. 11.2. The curve PQ gives saturation vapour pressure as a function of ambient
temperature. Point X represents a sample of air at 18°C and 10 mbar vapour pressure. The
slope of the line YZ, passing through X, corresponds to the psychometric constant. Other
symbols are explained in the text.
Change in body temperature (°C)
0
0 20 40 60 80 100
AET (°C)
Fig. 11.3. Relationship between change in deep body temperature and AET.
100
80
20
0
10 15 20 25 30 35 40
Dry bulb temperature (°C)
Fig. 11.4. ‘Thermal Comfort Zones’ for broiler transport. Safe limit AET = 40°C; danger limit
AET = 65°C or greater.
Conclusions
15
10
0
265 270 275 280 285 290
Fig. 11.5. Relationship between change in deep body temperature in wetted birds and
environmental temperature (K).
Using Physiological Models to Define Environmental Control 223
Acknowledgements
The work described in this review was supported by DEFRA. The author is
greatly indebted to Peter Kettlewell and Richard Hunter for their efforts
and inputs in all of the research projects in this area and to all the staff
involved at the Roslin and Silsoe Research Institutes for their technical
assistance, which ensured the successful completion of the modelling
research programme.
References
Abeyesinghe, S.M., Nicol, C.J., Wathes, C.M. and Randall, J.M. (2001a)
Development of a raceway method to assess aversion of domestic fowl to
concurrent stressors. Behavioural Processes 56, 175–194.
Abeyesinghe, S.M., Wathes, C.M., Nicol, C.J. and Randall, J.M. (2001b) The
aversion of broiler chickens to concurrent vibrational and thermal stressors.
Applied Animal Behaviour Science 73, 199–215.
Aerts, J.M., Wathes, C.M. and Berckmans, D. (2003) Dynamic data based modeling
224 M.A. Mitchell
Gates, R.S., Chao, K. and Sigrimis, N. (2001) Identifying design parameters for
fuzzy control of staged ventilation control systems. Computers and Electronics in
Agriculture 31, 61–74.
Gershenfeld, N. (1999) The Nature of Mathematical Modeling. Cambridge University
Press, Cambridge, UK, pp. 1–22.
Hahn, G.L. (1999) Dynamic responses of cattle to thermal heat loads. Journal of
Animal Science 77 (supplement 2), 10–20.
Hubbard, K.G., Stooksbury, D.E., Hahn, G.L. and Mader, T.L. (1999) A
climatological perspective on feedlot cattle performance and mortality related
to the temperature-humidity index. Journal of Production Agriculture 12,
650–653.
Hulzebosch, J. (2004) What affects the climate in poultry houses. World Poultry 20,
36–38.
Hunter, R.R., Mitchell, M.A. and Matheu, C. (1997) Distribution of ‘dead on
arrivals’ within the bio-load on commercial broiler transporters: correlation
with climate conditions and ventilation regimen. British Poultry Science 38,
Supplement, S7–S9.
Hunter, R.R., Mitchell, M.A. and Matheau, C. (2001) Mortality of broiler chickens
in transit – correlation with the thermal micro-environment. In: Stowell, R.R.
(ed.) Proceedings of the 6th International Livestock Environment Symposium,
Louisville, Kentucky, USA, 21–23 May 2001. American Society of Agricultural
Engineers, pp. 542–549.
Jones, B.W. (2002) Capabilities and limitations of thermal models for use in
thermal comfort standards. Energy and Buildings 34, 653–659.
Kaynakli, O. and Kilic, M. (2005) Investigation of indoor thermal comfort under
transient conditions. Building and Environment 40, 165–174.
Kettlewell, P.J. and Mitchell, M.A. (2001a) Comfortable ride: concept 2000 provides
climate control during poultry transport. Resource Engineering and Technology for
a Sustainable World 8, 13–14.
Kettlewell, P.J. and Mitchell, M.A. (2001b) Mechanical ventilation: improving the
welfare of broiler chickens in transit. Journal of the Royal Agricultural Society of
England 162, 175–184.
Kettlewell, P.J., Mitchell, M.A. and Meeks, I.M. (1997) An implantable radio-
telemetry system for remote monitoring of heart rate and deep body
temperature in poultry. Computers and Electronics in Agriculture 17, 161–175.
Kettlewell, P.J., Hoxey, R.P., Hampson, C.J., Green, N.R., Veale, B.M. and Mitchell,
M.A. (2001a) Design and operation of a prototype mechanical ventilation
system for livestock transport vehicles. Journal of Agricultural Engineering
Research 79, 429–439.
Kettlewell, P.J., Hoxey, R.P., Hampson, C.J., Green, N.R., Veale, B.M. and Mitchell,
M.A. (2001b) Heat and moisture generation of livestock during transportation.
In: Stowell, R.R. (ed.) Proceedings of the 6th International Livestock Environment
Symposium, Louisville, Kentucky, USA, 21–23 May, 2001. American Society of
Agricultural Engineers, pp. 519–526.
Kjerulfjensen, P., Nishi, Y., Fanger, P.O. and Gagge, A.P. (1975) Investigation on
man’s thermal comfort + physiological response. Journal of the American Society
of Heating Refrigerating and Air-Conditioning Engineers 17, 65–68.
Korthals, R.L., Chen, Y.R., Hahn, G.L. and Eigenberg, R.A. (1997) Calculation of
fractal dimension from cattle thermoregulatory responses. Journal of Thermal
Biology 22, 285–293.
MacCaluim, J.M., Abeyesinghe, S.M., White, R.P. and Wathes, C.M. (2003) A
226 M.A. Mitchell
Introduction
The processes of ovulation and oviposition in domestic hens have been the
subjects of much interest and of numerous investigations by poultry
scientists over the past century. The associated temporal relationships
make an intriguing study, both for the modeller aiming to predict rate of
lay and for the physiologist attempting to integrate the hormonal and
neurological pathways. Although tremendous progress has been made in
recent years in understanding the hormonal control, mainly due to the
advent of radioimmunoassay techniques, some aspects of the hen’s
reproductive system continue to elude scientific explanation.
Domestic hens lay their eggs in sequences; a sequence being made up
of a number of consecutive daily ovipositions followed by one or more
pause days, when no egg is laid. The first egg of a sequence is usually laid
early in the morning, roughly 8 h after the onset of darkness. Each
subsequent oviposition occurs progressively later in the day, at intervals
slightly longer than 24 h, on successive days. The difference in times of day
at which eggs are laid on consecutive days is known as the lag, as is the
corresponding delay in the ovulation cycle (Fraps, 1955). Total lag is a
measure of the proportion of the day used for egg laying and can be
regarded as an indication of the length of the open period for luteinizing
hormone (LH) release and hence ovulation (Lillpers and Wilhelmson,
1993). The reason why hens lay eggs in sequences is that the ovulatory
cycle is due to the interaction of two asynchronous physiological systems
(Etches, 1984; Johnson, 1984). These are the maturation of the largest
ovarian follicle and a circadian rhythm that restricts the preovulatory surge
of LH to a limited portion of the day. Ovulation occurs when the
maturation of a follicle coincides with a certain phase of the circadian
rhythm.
Fig. 12.1. The observed (shaded columns) and predicted (blank columns) mean lag for eight
sequence lengths (data from Etches and Schoch, 1984).
where a1 is a constant and S1 is the time of day when the cycle is completed
and the regulator function is reinitiated. The events leading up to
ovulation take place some hours beforehand, but the inclusion of the
parameter S1 shifts the curve so that R3(t) exceeds a threshold value from
about 07:00 to early afternoon, not during the night. Thus the curve given
by Eqn 12.1 also represents the period of the day during which ovulation
may take place. The values of the parameters 1, 2 and S1 and of the
constant a1 used to predict ovulation sequences of varying lengths are
given in Table 12.1.
The value of a2 depends on the values allocated to parameters 1 and
2 and may be calculated from the function
a2 = (a1(a1 .e (λ1 . 24) )) / (1e (λ2 . 24)) (12.2)
when the light:dark cycle is of 24 h duration.
The second equation in the ovulation model of Etches and Schoch
(1984) represents the maturation of the largest ovarian follicle (Fraps’
excitation hormone concentration) and takes the form of a Gompertz
function:
G(t) = b1 .eb2 eb3 (tS2) (12.3)
where S2 represents the time taken after the previous ovulation to
reinitiate G(t). This interval is necessary because the follicle is not sensitive
to gonadotrophic stimulation for a number of hours after the previous
ovulation. The values for the parameters b1, b2, b3 and S2 to predict
sequences of different lengths are shown in Table 12.1.
Follicle maturation refers to the competence of the follicle to ovulate in
response to an LH surge, rather than to an increase in weight or diameter
of the follicle. This phase is brought about by a number of physiological
Modelling Egg Production in Laying Hens 233
Table 12.1. The values of the parameters and constants used by Etches and Schoch (1984),
University of Guelph data, to predict the times of ovulation and oviposition.
Seq. length a1 1 2 S1 b1 b2 b3 S2
2 2.175 0.14 0.25 8.0 0.75 4.5 0.22 17.0
3 2.175 0.14 0.25 7.5 0.75 5.1 0.27 17.0
4 2.175 0.14 0.25 6.75 0.75 5.2 0.30 17.0
5 2.175 0.15 0.285 6.5 0.75 5.15 0.315 17.5
6 2.175 0.15 0.285 6.5 0.76 5.05 0.32 17.5
7 2.175 0.15 0.285 6.5 0.78 5.2 0.325 17.5
8 2.175 0.15 0.285 6.5 0.78 5.15 0.33 17.5
9 2.175 0.15 0.285 6.5 0.785 5.17 0.334 17.5
Fig. 12.2. An illustration of the asynchronous cycle of ovulation for a three-egg sequence;
ovulation occurs when the regulator concentration 1R3(t) (solid line) intersects with follicle
maturation G(t) (dotted line).
Table 12.2. Values assigned to the Gompertz function parameters for each of the seven
ovulatory cycle parameters.
Parameter B M C A
λ1 8 0.66667 0.020456 0.132475
λ2 8 0.66667 0.071596 0.223662
S1 8 0.66667 3.068414 9.128776
b1 8 0.66667 0.071596 0.723662
b2 6.148166 0.708292 21.097792 0.760415
b3 8 0.66667 0.233199 0.134213
S2 8 0.66667 3.596213 17.080953
236 S.A. Johnston and R.M. Gous
Fig. 12.3. The curvilinear relationship between b3 and ovulation rate, showing the fitted
Gompertz function (—) and the given parameter values ().
Fig. 12.4. The values for λ2 given by Etches and Schoch (1984) (䉬) and the Gompertz
function for predicting λ2 from ovulation rate (—).
The starting point for any layer model aiming to predict flock egg
production is to estimate the mean age at first egg. This is because a hen’s
chronological age determines her ovulation rate and yolk weight, as well as
the incidence of internal ovulations, soft shells and double-yolked eggs.
Sequence length and rate of lay are therefore also influenced by the mean
age at sexual maturity. Although other factors do play a role, light is the
Modelling Egg Production in Laying Hens 237
Fig. 12.5. Theoretical distributions of ages at first egg for Hy-Line Brown pullets
photostimulated at 12 weeks (hatched columns), 15 weeks (blank columns) and 18 weeks
(solid columns).
238 S.A. Johnston and R.M. Gous
Fig. 12.6. Curvilinear relationships between ICL and time from first egg for two hens.
Fig. 12.7. The rates of lay for two hens, using the internal cycle lengths shown in Fig. 12.6.
The ovulatory cycle parameter S1 (from Eqn 12.1) determines the time
when the circadian rhythm for the regulator concentration function is
reinitiated. Because of the phase-setting effect of sunset, S1 needs to be
linked to the time that the lights are turned off in the poultry house, so
that predicted ovulation and oviposition times are relative to the onset of
darkness. The predicted mean time of lay on 15- to 16-h photoperiods is
about 13–14 h after sunset, as reported by Lillpers (1991) and Patterson
(1997), although as the duration of the scotoperiod increases, the interval
between dusk and the mean oviposition time increases (Etches, 1996).
The open period for LH release is thought to commence about 2-h
after sunset and to last for about 8–9 h (Williams and Sharp, 1978). The
preovulatory surge of LH can take place at any time during the open
period. Between 4 and 6 h after the LH peak, ovulation occurs. There is a
corresponding open period of about 8–9 h during which ovulation can
take place. Presumably the earliest ovulations at the start of a sequence
take place about 6–8 h after the onset of darkness. In a population model
each hen may be allocated a unique value for the sunset-ovulation interval
by creating a normal distribution, using a mean of 7 h and a coefficient of
variation of 5%. Suggested temporal relationships between the onset of
darkness, LH release and ovulation are shown in Fig. 12.8.
The method used to link S1 to the onset of darkness is explained in
detail by Johnston (2004). Briefly, the ovulatory cycle parameter S1 is itself
Lights off
Open period for LH release
Time of day
Fig. 12.8. An illustration of the temporal relationships between lights off (21:00), the start of
the open period for LH release (23:00) and the start of the open period for ovulation (04:00).
Modelling Egg Production in Laying Hens 241
An ovulation needs to occur the day before the hen lays its first egg. At
sexual maturity the time from first egg is zero, and this determines the
bird’s initial internal cycle length (Eqn 12.9), which in turn determines the
ovulation rate (using Eqns 12.7 and 12.8). A normal distribution of lag
values with, for example, a mean of 8.5 h and a coefficient of variation of
2%, introduces further variation. The ovulation rate establishes the time of
each consecutive ovulation by using the appropriate values for the seven
ovulatory cycle parameters. In a flock of hens, the mean ovulation rate at
onset of lay will be determined by the distribution of ages at first egg and
also by the initial sequence lengths, or individual ovulation rates. The
predicted flock ovulation rate over a laying cycle for a hypothetical flock of
birds is illustrated in Fig. 12.9. In this example the ovulation rate reaches a
peak 9 weeks after the onset of lay.
Fig. 12.9. The predicted ovulation rate for a theoretical flock of 100 Hy-Line Silver hens.
242 S.A. Johnston and R.M. Gous
In theory, the ovulation rate and the rate of lay are identical unless internal
ovulations occur, or the production of double-yolked or soft-shelled eggs
causes interruptions to egg sequences. Johnston (2004) found that 8% of
double yolks and 18% of soft shells were associated with pauses in egg
sequences, and these percentages increased with earlier photostimulation.
The time that an egg is laid is determined by the time of the associated
ovulation (Gilbert and Wood-Gush, 1971), which means that, except for
the terminal egg of a sequence, the ovulation times may be used to
estimate oviposition times. Oviposition normally occurs about half an hour
before the next ovulation, with a range of 7–75 min (Warren and Scott,
1935; Fraps, 1955). A normal distribution of oviposition-ovulation intervals
may be produced using a mean of 30 min and a coefficient of variation of
30%, so that the range of intervals is from 3 to 57 min. In a population,
each hen may therefore be allocated a different value for the interval and
this value may change daily. This interval needs to be subtracted from the
predicted ovulation time to give the estimated time of the associated
oviposition.
Estimation of the oviposition time for the last egg of a sequence needs
to be handled in a different way, because there is no associated ovulation.
The relationship between the lag for the last two eggs of sequences and
sequence length is illustrated in Fig. 12.10.
Fig. 12.10. The relationship between sequence length and lag (the oviposition interval minus
24 h) for the last two eggs of a sequence. Data from Morris, unpublished (×) and Johnston,
2004 (). The solid line shows the fitted linear-by-linear function.
Modelling Egg Production in Laying Hens 243
Shorter sequences tend to have longer intervals between the last two
ovipositions. The large variation in lag for the longer egg sequences is due
to the fact that the sample sizes are very small.
A fitted linear-by-linear equation of the form
y = 1.75 – 0.9 / ( 1 – 1.969 . x ) (12.11)
predicts the lag between the last two ovipositions of a sequence (y, h) from
ovulation rate (x) which, in the population model, determines sequence
length. Consequently for an ovulation rate of 0.75 (a three-ovulation
sequence), the lag for the last two eggs is estimated to be 03 h 33 min.
The predicted distribution of oviposition times for a theoretical flock of
100 Hy-Line Silver hens is illustrated in Fig. 12.11. The distribution is
bimodal, because the shorter sequences that come with advancing age lead
to an increasing number of late afternoon ovipositions (Foster, 1968). The
mean time of lay is 10:28, 12 h 58 min after the sunset signal at 21:30.
Prior to the rupture of the follicle and the release of the ovum, the follicle is
usually engulfed by the infundibulum so that ovulation occurs directly into
the infundibulum. This process is under hormonal and neurological control
and its effectiveness may therefore be impaired at puberty and with ageing
(Gilbert, 1972). Internal laying occurs when, for some reason, an ovum is
not grasped by the infundibulum and therefore remains in the body cavity.
The yolk material is usually resorbed within 24 h (Sturkie, 1955).
Fig. 12.11. Frequency distribution of the predicted times of lay for a theoretical flock of 100
Hy-Line Silver hens for 500 days, with sunset at 21:30.
244 S.A. Johnston and R.M. Gous
5
% Internal ovulations
0
12 17 22 27 32 37 42 47 52 57 62 67 72
Age (weeks)
Fig. 12.12. The relationship between expected per cent internal ovulations and hen age for
Hy-Line Silver (solid line) and Brown (dotted line) birds.
Modelling Egg Production in Laying Hens 245
Double-yolked eggs
Fig. 12.13. An illustration of the relationship between expected per cent double yolks and
hen age for Hy-Line birds.
246 S.A. Johnston and R.M. Gous
Fig. 12.14. The distribution of double yolks in a theoretical flock of 100 Hy-Line Silver hens
photostimulated at 18 weeks of age.
hens laid two double-yolked eggs. Only 23% of the population produced
double yolks. This proportion is substantially lower than anticipated, the
model having identified 36 of the 100 hens as potential multiple ovulators,
but the figure is presumably influenced by the relatively short period
allocated to the production of double yolks. These 37 double-yolked eggs
accounted for 0.11% of the total eggs laid to 70 weeks of age.
Soft-shelled eggs
factors, it is useful to have some value for the purpose of testing the model.
Many hens lay normal-shelled eggs throughout their laying year. The
population model therefore needs to set apart a proportion of the hens
that will randomly produce soft-shelled eggs.
Two line-plus-exponential functions, derived from experimental data
(Johnston, 2004), may be used to predict the percentage of soft shells for
the Silver and Brown birds, respectively:
y = 0.8006 + 1202.6 . (0.952622 x) + 0.0036895 . x (12.15)
y = 0.8134 + 714.6 . (0.953762 x) + 0.0037948 . x (12.16)
where y = % soft shells and x = hen age, in days. The curves given by
these functions are shown in Fig. 12.15. Hy-Line Silver birds that are
photostimulated at 12 weeks and come into lay at about 15 weeks of age
may be expected to produce 6.9% soft shells in the first week of lay. At 18
weeks the frequency will be reduced to 2.3%.
A hypothetical distribution of soft shells within a population of 70-
week-old Hy-Line Silver hens photostimulated at 18 weeks of age is shown
in Fig. 12.16. Three birds laid one soft-shelled egg each and one bird laid
eight. A total of 141 soft shells (0.43% of total eggs) was produced by 36%
of the hens; 64 of the 100 hens did not lay any soft-shelled eggs. Fig. 12.17
illustrates how the per cent soft shells changed during the course of the
laying year for the same flock. Because the pullets were photostimulated at
the recommended age, there were very few soft shells at onset of lay. An
increasing number were produced towards the end of the productive
period.
Fig. 12.15. An illustration of the relationship between expected per cent soft shells and hen
age for Hy-Line Silver (solid line) and Brown (dotted line) birds.
248 S.A. Johnston and R.M. Gous
Fig. 12.16. The distribution of soft shells in a theoretical flock of 100 Hy-Line Silver hens
photostimulated at 18 weeks of age.
Fig. 12.17. The per cent soft shells produced over a full laying cycle, by a theoretical flock of
100 Hy-Line Silver hens photostimulated at 18 weeks.
Modelling Egg Production in Laying Hens 249
where y = yolk weight, in grams and x = hen age, in days. The parameters
of both functions need to be defined for each genotype, because strains
differ both in their initial yolk weight and in their rate of increase in yolk
weight with advancing hen age. In addition, the values allocated to each of
the parameters may need to be adjusted from time to time in line with
genetic advancements, since the component weights are likely to be
affected. The relationship between yolk weight and hen age, for three
commercially available genotypes, is illustrated in Fig. 12.18.
It is interesting to note that, although the eggs from Hy-Line Brown
hens are heavier, they have a smaller proportion of yolk than those from
Hy-Line Silver hens. Some suggested values of the parameters are given in
Table 12.4.
The next step is to predict albumen weight from yolk weight, using
Eqn 12.19 and substituting the relevant values for the two parameters for
the specific genotype. Subsequently, shell weight may be estimated from the
weight of the egg contents (yolk plus albumen) using Eqn 12.20. Finally, egg
weight is calculated as the sum of yolk, albumen and shell weights. The
suggested values for the allometric function parameters for the same three
strains of bird are summarized in Table 12.5. The method used to derive the
values for the parameters is explained by Johnston (2004).
Fig. 12.18. The curvilinear relationship between yolk weight and hen age, for Amber-Link
(bold line), Hy-Line Silver (solid line) and Hy-Line Brown (dotted line) hens.
Table 12.4. Values of the parameters derived for the logistic (Amber-Link) and Gompertz
(Hy-Line Silver and Brown) functions used to predict yolk weight from age.
Strain A C B M
Amber-Link 224.7 243.2 0.01268 116.4
Hy-Line Silver 51107 51123 0.01771 370.1
Hy-Line Brown 101.1 116.0 0.01972 15.36
Modelling Egg Production in Laying Hens 251
Table 12.5. Estimates of the parameter values in the allometric functions used to predict
albumen and shell weight.
Parameter
a1 (albumen) b1 (albumen) a2 (shell) b2 (shell)
Hy-Line Silver 9.473515 0.5044 0.138207 0.9180
Hy-Line Brown 10.8906 0.5020 0.133187 0.9310
Amber-Link 10.9900 0.4491 0.33875 0.6896
Fig. 12.19. An illustration of how predicted mean yolk percentage increases with age, for a
hypothetical flock of 100 Hy-Line Silver hens.
252 S.A. Johnston and R.M. Gous
Fig. 12.20. An illustration of how predicted mean albumen percentage decreases with age
for a hypothetical flock of 100 Hy-Line Silver hens.
Fig. 12.21. A scatter diagram illustrating the positive relationship between albumen weight
and egg weight (r2 = 0.99), for a hypothetical flock at 163 days of age.
Modelling Egg Production in Laying Hens 253
The mean sequence lengths for two theoretical flocks, both photostimulated
at 18 weeks, one consisting of 100 Hy-Line Silver hens and the other of 100
Hy-Line Brown hens, are shown in Fig. 12.22. It may be seen that the
predicted mean sequence length for both strains is initially short at onset of
lay, owing to the quadratic-by-linear function used to predict changes in
internal cycle length. The maximum mean sequence lengths are 86.7 and
76.8 days for the Hy-Line Silver and Hy-Line Brown flocks, respectively.
The trends simulated by the model are similar to those observed in practice
and reported by Johnston (2004) and Lewis and Perry (1991). As expected,
the model predicts that the Silver strain has slightly longer mean sequence
lengths than the Brown strain for most of the laying cycle.
The predicted sequence lengths per bird over an extended laying
period may be imported into the Sequence Analyzer program (Zuidhof et
al., 1999). This software calculates (as well as other variables) the mean
sequence length, the number of sequences, the prime sequence length and
the mean pause length for each bird, and is useful for comparing the
performances of individuals or groups of birds subjected to different
treatments. The performance indicators for the same Hy-Line Silver and
Brown flocks discussed above are summarized in Table 12.6. The Silver
strain tends to have longer prime sequences and to lay fewer but longer
sequences than the Brown strain. These factors contribute to the higher
rate of lay seen in the Hy-Line Silver hens. The large standard deviations
indicate the huge amount of variation between individuals in terms of
sequence lengths.
Fig. 12.22. Predicted mean sequence lengths for 100 Hy-Line Silver (solid line) and 100
Hy-Line Brown (dotted line) hens.
254 S.A. Johnston and R.M. Gous
Table 12.6. Summary of mean sequence characteristics (± SD) for the theoretical flocks.
Prime sequence Number of Sequence
Strain length sequences length Pause length
Hy-Line Silver 82.76 (±60.20) 27.41 (±9.15) 13.60 (±5.56) 1.02 (±0.03)
Hy-Line Brown 72.70 (±61.69) 28.47 (±9.82) 13.02 (±5.90) 1.02 (±0.03)
Discussion
whether or not ovulation takes place. If young hens have fewer follicles in
the hierarchy at onset of lay, the result would be shorter sequences.
Another benefit to be derived from modelling the hierarchy is that yolk
weight may be predicted with a greater degree of accuracy. The follicles
may be allowed to continue to accumulate yolk up to the precise time of
ovulation. In this way the yolk of the first egg of a sequence is likely to be
heavier than in subsequent eggs.
It would be useful to know why some birds have inter-sequence pauses
longer than one day. These may well be due to inadequate maintenance of
the follicular hierarchy, but there may also be times during the laying cycle
when the preovulatory surge of LH does not proceed as expected. Stresses
that are imposed on the hens over a number of days, such as the
withdrawal of feed, cause apoptosis and subsequent cell proliferation in the
tissues of the anterior pituitary, with corresponding changes to the LH
secretions (Chowdhury and Yoshimura, 2002). Short-term stresses may
also have the ability to prevent the secretion of LH. The current model
presumes that the circadian rhythm of LH release continues without
interruption. A method of blocking the rhythm of the regulator substance
can be found such that, for one day or a number of days, ovulation will not
take place. This will allow the model to respond to a limiting environment
by altering rates of ovulation and lay. Accordingly, pauses longer than one
day can be introduced into sequences. Increasing the incidence of internal
ovulations or allowing soft-shelled and double-yolked eggs to disrupt egg
sequences will also result in longer inter-sequence pauses, as long as two or
more disturbances occur on consecutive days. A well-designed trial,
focusing on the poor producers, could provide valuable information on the
underlying reasons for their low rate of lay.
The proportional changes in yolk, albumen and shell with increasing
age have been observed and accepted by poultry scientists for more than
half a century (Romanoff and Romanoff, 1949; Harms and Hussein, 1993).
It is important to be able to model these changes successfully, so that
increases in egg weight with advancing hen age are brought about by
increases in the three components occurring at different rates. The
parameters in the allometric functions need to be carefully defined for
each genotype, since strains differ in the weights and proportions of yolk,
albumen and shell and this may entail different nutritional requirements.
Moreover, it is necessary to reassess the information every few years in
order to keep pace with genetic advancements.
A great many factors are taken into account by the layer model
presented in this chapter. The mean age at first egg is calculated from the
lighting programme applied during rearing. Ovulation times are predicted
for a theoretical flock of 100 birds over a laying year. The flock ovulation
rate is seen to increase rapidly to reach a peak about 8 weeks from onset of
lay, before declining in a linear fashion. Oviposition times are predicted
from the associated ovulation times. For all but the last egg of a sequence,
oviposition occurs roughly half an hour before the associated ovulation.
The lag between the penultimate and ultimate eggs of a sequence is
256 S.A. Johnston and R.M. Gous
References
Adams, C.J. and Bell, D.D. (1980) Predicting poultry egg production. Poultry Science
59, 937–938.
Chowdhury, V.S. and Yoshimura, Y. (2002) Cell proliferation and apoptosis in the
anterior pituitary of chicken during inhibition and resumption of laying.
General and Comparative Endocrinology 125, 132–141.
Christmas, R.B. and Harms, R.H. (1982) Incidence of double yolked eggs in the
initial stages of lay as affected by strain and season of the year. Poultry Science
61, 1290–1292.
Emmans, G.C. and Fisher, C. (1986) Problems in nutritional theory. In: Fisher, C.
and Boorman, K.N. (eds) Nutrient Requirements of Poultry and Nutritional
Research. Butterworths, London, pp. 9–39.
Etches, R.J. (1984) Maturation of ovarian follicles. In: Cunningham, F.J., Lake, P.E.
and Hewitt, D. (eds) Reproductive Biology of Poultry. Longman Group, Harlow,
UK, pp. 51–73.
Etches, R.J. (1996) Reproduction in Poultry. Cambridge University Press, Cambridge,
UK.
Etches, R.J. and Schoch, J.P. (1984) A mathematical representation of the ovulatory
cycle of the domestic hen. British Poultry Science 25, 65–76.
Foster, W.H. (1968) The effect of light-dark cycles of abnormal lengths upon egg
production. British Poultry Science 9, 273–284.
Fraps, R.M. (1955) Egg production and fertility in poultry. In: Hammond, J. (ed.)
Progress in the Physiology of Farm Animals. Butterworths, London, pp. 661–740.
Gilbert, A.B. (1972) The activity of the ovary in relation to egg production. In:
Freeman, B.M. and Lake, P.E. (eds) Egg Formation and Production. British
Poultry Science, Edinburgh, UK, pp. 3–17.
258 S.A. Johnston and R.M. Gous
Gilbert, A.B. and Wood-Gush, D.G.M. (1971) Ovulatory and ovipository cycles. In:
Bell, D.J. and Freeman, B.M. (eds) Physiology and Biochemistry of the Domestic
Fowl, Vol. 3. Academic Press, London, pp. 1353–1378.
Hansen, K.K., Kittok, R.J., Sarath, G., Toombs, C.F., Caceres, N. and Beck, M.M.
(2003) Estrogen receptor-α populations change with age in commercial laying
hens. Poultry Science 82, 1624–1629.
Harms, R.H. and Hussein, S.M. (1993) Variations in yolk: albumen ratio in hen
eggs from commercial flocks. Journal of Applied Poultry Research 2, 166–170.
Hussein, S.M., Harms, R.H. and Janky, D.M. (1993) Research Note: Effect of age
on the yolk to albumen ratio in chicken eggs. Poultry Science 72, 594–597.
Johnson, A.L. (1984) Interactions of progesterone and luteinizing hormone leading
to ovulation in the domestic hen. In: Cunningham, F.J., Lake, P.E. and Hewitt,
D. (eds) Reproductive Biology of Poultry. Longman Group, Harlow, UK,
pp.133–143.
Johnson, A.L. (1996) The avian ovarian hierarchy: a balance between follicle
differentiation and atresia. Poultry and Avian Biology Reviews 7, 99–110.
Johnson, P.A., Dickerman, R.W. and Bahr, J.M. (1986) Decreased granulosa cell
luteinizing hormone sensitivity and altered thecal estradiol concentration in
the aged hen, Gallus domesticus. Biology of Reproduction 35, 641–646.
Johnston, S.A. (1993) Simulation modelling as a means of predicting the
performance of laying hens. MSc Agric thesis, University of Natal, South Africa.
Johnston, S.A. (2004) A stochastic model to predict annual egg production of a
flock of laying hens. PhD thesis, University of KwaZulu-Natal, South Africa.
Johnston, S.A. and Gous, R.M. (2003) An improved mathematical model of the
ovulatory cycle of the laying hen. British Poultry Science 44, 752–760.
Lewis, P.D., Morris, T.R. and Perry, G.C. (2002) A model for predicting the age at
sexual maturity for growing pullets of layer strains given a single change in
photoperiod. Journal of Agricultural Science 138, 441–458.
Lewis, P.D. and Perry, G.C. (1991) Oviposition time: correlations with age, egg
weight and shell weight. British Poultry Science 32, 1135–1136 (Abstr.).
Lewis, P.D., Perry, G.C. and Morris, T.R. (1997) Effect of size and timing of
photoperiod increase on age at first egg and subsequent performance of two
breeds of laying hen. British Poultry Science 38, 142–150.
Lillpers, K. (1991) Genetic variation in the time of oviposition in the laying hen.
British Poultry Science 32, 303–312.
Lillpers, K. and Wilhelmson, M. (1993) Genetic and phenotypic parameters for
oviposition pattern traits in three selection lines of laying hens. British Poultry
Science 34, 297–308.
Liou, S.S., Cogburn, L.A and Biellier, H.V. (1987) Photoperiodic regulation of
plasma melatonin levels in the laying chicken (Gallus domesticus). General and
Comparative Endocrinology 67, 221–226.
Menaker, M., Hudson, D.J. and Takahashi, J.S. (1981) Neural and endocrine
components of circadian clocks in birds. In: Follett, B.K. and Follett, D.E. (eds)
Biological Clocks in Seasonal Reproductive Cycles. Wright, Bristol, UK,
pp. 171–183.
Patterson, P.H. (1997) The relationship of oviposition time and egg characteristics
to the daily light:dark cycle. Journal of Applied Poultry Research 6, 381–390.
Robinson, F.E., Renema, R.A., Oosterhoff, H.H., Zuidhof, M.J. and Wilson, J.L.
(2001) Carcass traits, ovarian morphology and egg laying characteristics in
early versus late maturing strains of commercial egg-type hens. Poultry Science
80, 37–46.
Modelling Egg Production in Laying Hens 259
Romanoff, A.L. and Romanoff, A.J. (1949) The Avian Egg. John Wiley, New York.
Siopes, T.D. and Wilson, W.O. (1980) A circadian rhythm in photosensitivity as the
basis for the testicular response of Japanese quail to intermittent light. Poultry
Science 59, 868–873.
Sturkie, P.D. (1955) Absorption of egg yolk in the body cavity of the hen. Poultry
Science 34, 736–737.
Tilly, J.L., Kowalski, K.I. and Johnson, A.L. (1991) Stage of ovarian follicular
development associated with the initiation of steroidogenic competence in
avian granulosa cells. Biology of Reproduction 44, 305–314.
Turek, F.W., Penev, P., Zhang, Y., Van Reeth, O., Takahashi, J.S. and Zee, P. (1995)
Alterations in the circadian system in advanced age. In: Chadwick, D.J. and
Ackrill, K. (eds) Circadian Clocks and Their Adjustment. Wiley, Chichester, UK,
pp. 212–234.
van Middelkoop, J.H. (1972) The relationship between ovulation interval of White
Plymouth Rock pullets and the laying of abnormal eggs. Archiv für
Geflügelkunde 6, 223–230.
Warren, D.C. and Scott, H.M. (1935) The time factor in egg formation. Poultry
Science 14, 195–207.
Williams, J.B. and Sharp, P.J. (1978) Control of the preovulatory surge of
luteinizing hormone in the hen (Gallus domesticus): the role of progesterone
and androgens. Journal of Endocrinology 77, 57–65.
Zakaria, A.H. (1999) Ovarian follicular growth in laying hens in relation to
sequence length and egg position in various sequence lengths. Archiv für
Geflügelkunde 63, 264–269.
Zuidhof, M.J., Bignell, D. and Robinson, F.E. (1999) Egg Production and Sequence
Analyzer Software. Alberta Poultry Research Centre, Alberta Agriculture, Food
and Rural Development, University of Alberta, Canada.
13 Comparison of Pig Growth Models
– the Genetic Point of View
P. LUITING AND P.W. KNAP
PIC International Group, Ratsteich 31, D-24837 Schleswig, Germany
pieter.knap@pic.com
Introduction
All the models discussed here distinguish three processes that require
energy: protein deposition (PD), body maintenance, and lipid deposition
(LD), all three on a daily basis. The driving factor is PD; it is predicted
first, after which it determines all other factors in the simulation. The
metabolizable energy (ME) requirements for body maintenance (MEm) are
predicted from metabolic body weight or from body protein mass: MEm =
α × BW or MEm = α × P, where α and are constants. It follows that at
known energy intake (MEI), LD is predicted as the item that balances the
requirements of the law of conservation of energy:
MEI = MEm + PD × EP / kP + LD × EL / kL →
MEI - MEm − (PD × EP / kP ) (13.1)
LD =
EL / kL
where LD = lipid deposition (kg per day), MEI = ME intake (MJ per day),
MEm = ME required for maintenance (MJ per day), PD = protein
deposition (kg per day), EP and EL = combustion energy contents of body
protein and lipid, respectively (MJ per kg), and kP and kL = efficiencies of
use of ME for PD and LD, respectively.
All models assume EP, EL, kP, kL, α and to be constant, although they
vary somewhat with regard to their values. Genetic variation in these
parameters is presumed absent.
All models calculate water and ash deposition from PD, using empirical
equations with constant parameter values. Empty body weight gain (BWG)
is calculated by summation of the four components; adjustment for a
certain amount of gut fill leads to daily growth:
All models define an initial status with the starting values for body
protein, lipid, ash and water mass. The predicted deposition values are
added to the current values of body component mass, and next day’s
calculation is performed. This process is repeated until the desired end
status is reached. The models vary somewhat with regard to the water-
and ash-related empirical equations and/or their parameter values, with
regard to the magnitude of gut fill, and with regard to the initial body
component mass assumptions. Genetic variation is presumed absent
here.
262 P. Luiting and P.W. Knap
PD
ep = maxep
Pm DCPI
Fig. 13.1. Protein deposition (PD) in relation to digestible crude protein intake (DCPI). Pm
denotes the protein requirement for body maintenance, eP is the material efficiency of protein
utilization. In this range of DCPI, eP attains its maximum value maxeP.
Comparison of Pig Growth Models 263
PD
maxPD
ep < maxep
ep = maxep
Pm DCPI
Fig. 13.2. An extension of Fig. 13.1. PD has an upper limit (maxPD), reached at DCPI =
desired DCPI. At higher DCPI levels, eP is lower than its maximum value.
264 P. Luiting and P.W. Knap
PD LD
maxPD
LD
minLD maxPD
PD
b1
Fig. 13.3. Protein (PD) and lipid deposition (LD) in relation to ME intake (MEI). For PD <
maxPD, PD = f(MEI) = b1 × (MEI – b0). Above that level (i.e. above MEI = minMEImaxPD and
LD = minLDmaxPD), all surplus energy is used for LD.
Comparison of Pig Growth Models 265
PD
maxPD
PD =
f(MEI)
eP < maxeP
eP < maxeP
eP = maxeP
Pm DCPI
Fig. 13.4. An extension of Fig. 13.2. At insufficient ME supply, PD has an upper limit that
depends on ME intake (MEI) and eP is lower than its maximum value.
Model Comparison
The main issue from the above section is: body protein deposition (PD)
takes place at a low material efficiency when it is limited either by the
genetic maximum to PD or by insufficient energy supply. The question is
then how the various models parameterize these relations. We deal next
with maximum protein deposition, with insufficient energy supply, and
give a brief comparison of how some of these models describe voluntary
food intake.
maxPD maxPD
Thorbek (1975)
eqn (8)
Fig. 13.5. Various functions to describe the course of maxPD in relation to body weight (BW)
or body protein mass (P) as a proportion of its mature value (Pmat).
Comparison of Pig Growth Models 267
All models described here use a function f(MEI) that describes the
reduction of PD for MEI < minMEImaxPD. This function is not always
explicit, but may follow implicitly from other, explicitly defined, relations.
Table 13.1 introduces these structures. Method 1 explicitly defines a value
for minLD/PD, and implicitly describes f(MEI) as a linear regression line
of PD on MEI with b0 = MEm. Methods 2 and 3 define f(MEI) explicitly
but differently. Methods 4 and 5 explicitly define a function for eP and a
value for minL/P, respectively, leading implicitly to a function PD =
f(MEI).
Method 1 was published earliest, and some subsequent methods give,
in addition to their explicit element f(MEI), also their own implicit value
for minLD/PD, for comparison purposes. We include minLD/PD in Table
13.1 for the same reason, although it is not required for a comparison of
these five methods when f(MEI) has already been described.
The general form of f(MEI) is PD = b1 × (MEI – b0), where b0 is an x-
intercept and b1 represents the reduction of PD at MEI < minMEImaxPD.
The general form of minLD/PD is derived as follows. Substituting f(MEI)
into Eqn 13.1 gives:
1 − b1 × EP / kP ⎛ MEm − b0 × b1 × EP / kP ⎞
= × ⎜ MEI − ⎟
EL / kL ⎝ 1 − b1 × EP / kP ⎠
so that
Table 13.1. Methods for the reduction of protein deposition at MEI < minMEImaxPD.
Method minLD/PD PD = f(MEI) Third function
Whittemore explicit definition implicit description
1
Moughan constant → Lin(MEI)
Black implicit description explicit definition
2
De Greef nonLin(MEI) ← Lin(MEI)
3 Van Milgen implicit description ← explicit definition
nonLin(MEI) Quad(MEI)
Emmans and implicit description ← implicit description ← explicit definition
4
Kyriazakis nonLin(MEI, DCPI) Lin (MEI), nonLin(DCPI) eP = Lin(MEI/DCPI)
5 De Lange implicit description ← implicit description ← explicit definition
nonLin(MEI) Lin(MEI, L, P) minL/P = constant
Lin(): linear function. nonLin(): nonlinear function. Quad(): quadratic function.
References in the text.
268 P. Luiting and P.W. Knap
1 − b1 × EP / kP ⎛ MEm − b 0 × b1 × EP / kP ⎞
× ⎜ MEI − ⎟
EL / kL ⎝ 1 − b1 × EP / kP ⎠
minLD / PD = =
b1 × (MEI − b0 )
(13.11a)
MEm − b 0 × b1 × EP / kP
MEI −
1 − b1 × EP / kP 1 − b1 × EP / kP
= ×
b1 × EL / kL MEI − b0
1 − b1 × EP / kP ⎛ MEm − b 0 1 ⎞
minLD / PD = × ⎜1 − × ⎟ (13.11b)
b1 × EL / kL ⎝ 1 − b1 × EP / kP MEI − b0 ⎠
Table 13.2a. The forms of regression coefficientsa b0 and b1 for the methods in Table 13.1.
The forms of margminLD/PD and minLD/PD are in Table 13.2b.
Method b0 b1
Whittemore MEm 1
1
Moughan EP / kP + minLD / PD × EL / k L
Black
2 g1 g2
De Greef
g2 × (MEI − MEm − g1) =
3 Van Milgen MEm −4 × maxPD
= × (MEI − MEm − g1)
g12
1
5 De Lange MEm + (minL/P × P – L) × EL / kL
EP / kP + minL / P × EL / k L
1 = minLD/PD g1
1− g2 × EP / kP ⎛ MEm − g1 1 ⎞
2 margminLD / PD × ⎜1− × ⎟
g2 × EL / kL ⎝ 1− g2 × EP / kP MEI − g1 ⎠
⎛ ⎞
1 ⎜ 1 ⎟
3 ×⎜ − EP / kP ⎟ margminLD/PD
−4 × maxPD
EL / kL ⎜ × (MEI − MEm − g1) ⎟
⎝ g12 ⎠
⎛ ⎞
1 ⎛1 DCPI ⎞ ⎜ MEm 1 ⎟
4 × × − EP / kP ⎟ margminLD / PD × ⎜1− × ⎟
EL / kL ⎜⎝ (DCPI − Pm ) ⎠ ⎜ 1− × (DCPI − Pm ) × E / k MEI ⎟
⎝ DCPI
P P
⎠
margminLD / PD ×
269
270 P. Luiting and P.W. Knap
b1 =
−4 × maxPD
g12
(
× MEI − MEm − g1 ) (13.15)
5. De Lange (1995)
De Lange assumed that an intrinsic minimum amount of lipid mass (L)
must accompany each unit of protein mass (P), hence that there is an
intrinsic minimum lipid to protein ratio in the body (minL/P), which is
genetically determined. Therefore, minLD can be written as minLD =
minL/P × (P + PD) – L. The equivalent of Eqn 13.12 is then:
PD =
1
EP / kP + minL / P × EL / kL
(
× MEI − [MEm + (minL / P × P − L) × EL / kL ]) (13.17)
1
So b0 =MEm +(minL / P × P – L) × EL / kL, and b1 =
EP / kP + minL / P × EL / kL
Two of these models predict voluntary food intake, and they do so from
the combination of maintenance requirements and desired PD and LD.
Desired LD is defined in both cases via an additional equation. The forms
of these equations (and their difference between the two models) are
similar to the associated definitions of maxPD, as follows.
Emmans (1988) used again the derivative of the Gompertz growth
function to describe desired LD. This equation has two genetic parameters:
the rate parameter BGomp (equal to the one for maxPD in Eqn 13.9), and
mature body lipid mass Lmature.
desired LD = BGomp L ln(Lmature / L) (13.18)
Black (1988) used again the derivative of a modified Richards growth
function to describe the desired increase of net energy (NE) in the body:
desired NE = k × (BW + S)a × [(NEmature – NE)/NEmature]
which gives
desiredNE − maxPD × EP
desiredLD = (13.19)
EL
Comparison of Pig Growth Models 273
Apart from several empirical relations, the models described here contain
three truly mechanistic elements in the form of Eqns 13.1, 13.2 and 13.3.
All genetic factors enter this mechanistic system through eP in Eqn 13.3:
maximum protein deposition (maxPD) and protein deposition at
insufficient energy supply.
The differences and similarities between the various models with
regard to these genetic factors are summarized and discussed below.
The genetic aspects of the five discussed methods are summarized in Table
13.4 and Fig. 13.6.
De Greef and Verstegen (1992) showed that minLD/PD decreases with
decreasing MEI. This makes method 1 (with a single constant value for the
genetic parameter minLD/PD, g1 in Table 13.4) unsuitable.
PD PD
high L/P
maxPD maxPD
3 4
1
L/P > minL/P
high
DCPI
low
1 DCPI
3
4
2 L/P = minL/P
Fig. 13.6. Methods to describe protein deposition with insufficient ME supply, as in Table 13.3.
Left: (1) Whittemore, Moughan; (2) Black, De Greef; (3) Van Milgen and Noblet; (4) Emmans
and Kyriazakis. Right: De Lange. References and explanation of symbols in the text.
Comparison of Pig Growth Models 275
on MEI and is not constant, and b0 MEm. There are two more reasons
why this method is attractive.
First, b0 depends on P and L in such a way that b0 approaches MEm
when [minL/P × P – L] approaches zero, i.e. when L/P approaches minL/P.
This is illustrated in Fig. 13.7. When a pig that used to be fed at a high
MEI level (so that it is on a high PD metabolism, and has a high L/P level)
is suddenly brought to a low MEI level (for example, MEI = MEm as in
Fig. 13.7), it will not directly reduce its PD metabolism to zero, but will
catabolize body lipid to support its protein retention. In Fig. 13.6 this
initial situation is represented by the base of the arrow at the upper
regression line with a high L/P level (b0 < MEm). When this low MEI level
continues, the resulting lipid catabolism reduces L/P until it has reached
minL/P (through subsequently lower regression lines in Fig. 13.6), and PD
metabolism follows the arrow to approach zero as represented by the
regression line with b0 = MEm. Black et al. (1986) describe a similar effect
of b0 approaching MEm during a long period (100 days) of insufficient
energy supply. But they had to add yet another empirical equation to
method 2 to quantify this effect; method 5 does not require that.
It also follows that methods that set b0 = MEm (methods 1 and 3 in
Table 13.4) implicitly assume a steady state situation where metabolism has
been adapted to a low MEI level, another reason why they are less suitable.
Second, the concept of minL/P in method 5 is more elegant (because more
mechanistic) than the empirical approach of the linear regression
parameters in method 2. It operates on the same level as Emmans’s (1988,
1997) system of potential body protein mass and desired body lipid mass,
PD LD
maxPD
L/P = minL/P
B
⎫
⎬minLDmaxPD
PD A⎭
LD
0 MEm A B C MEI
⎭
⎬
⎫
minMEI maxPD
Fig. 13.7. An extension of Fig. 13.3. The PD system is the same as in the right-hand plot of
Fig. 13.6. Each PD line has an associated LD line which changes to a steeper slope in the
point minMEImaxPD where PD and LD reach maxPD and minLDmaxPD, respectively. See the
text for scenarios A, B and C.
Comparison of Pig Growth Models 277
0.23
⎛L ⎞
with = 1.46 × ⎜ mature ⎟ as an ‘auxiliary variable’ (Emmans, 1997).
⎝ Pmature ⎠
MEI than minMEImaxPD. Both would lead the pig to realize a datapoint (as in
Fig. 13.7) beyond the maxPD level; Campbell et al. (1983, 1985), Campbell
and Taverner (1988) and Dunkin and Black (1987) have shown that this
actually happens. Option (ii) can more appropriately be expressed as:
voluntary MEI − (MEm + maxPD × EP / kP + minLDmaxPD × EL / kL )
extra LD =
EL / kL
(13.22)
It follows that whereas the trajectory for MEI < minMEImaxPD is controlled by
a genetic drive for lipid deposition (expressed as minL/P), the trajectory for
MEI minMEImaxPD is controlled by a genetic drive for protein deposition
(expressed as maxPD) and a genetic drive for either lipid deposition or
‘luxury’ ME intake (which would lead to extra LD through Eqn 13.22).
Final remarks
Conclusions
genotypes, item (i) will always be hard to meet, and not necessarily as a
shortcoming of the models. For the description of maxPD, method 4 of
Table 13.3 seems the most appropriate method to meet the other criteria.
With respect to feature (ii) it is best combined with method 5 of Table 13.4
for the description of PD at MEI < minMEImaxPD. An additional advantage
of this combination of methods is that it provides a smooth and consistent
connection with Taylor’s genetic size scaling principles.
References
ARC (1981) The Nutrient Requirements of Pigs. Agricultural Research Council,
Commonwealth Agricultural Bureaux, Slough, UK.
Black, J.L. (1988) Animal growth and its regulation. Journal of Animal Science 66
(Suppl. 3), 1–22.
Black, J.L., Campbell, R.G., Williams, I.H., James, K.J. and Davies, G.T (1986)
Simulation of energy and amino acid utilisation in the pig. Research and
Development in Agriculture 3, 121–145.
Campbell, R.G. and Taverner, M.R (1988) Genotype and sex effects on the
relationship between energy intake and protein deposition in growing pigs.
Journal of Animal Science 66, 676–686.
Campbell, R.G., Taverner, M.R. and Curic, D.M (1983) The influence of feeding
level from 20 to 45 kg liveweight on the performance and body composition of
female and entire male pigs. Animal Production 36, 193–199.
Campbell, R.G., Taverner, M.R. and Curic, D.M (1985) Effects of sex and energy
intake between 48 and 90 kg liveweight on protein deposition in growing pigs.
Animal Production 40, 497–503.
Close, W.H. and Fowler, V.R (1982) Energy requirements of pigs. In: Haresign, W.
(ed.) Recent Advances in Animal Nutrition. Butterworths, London, UK, pp. 1–16.
De Greef, K.H. (1992) Prediction of production: nutrition induced tissue
partitioning in growing pigs. PhD dissertation, Wageningen Agricultural
University, Wageningen, Netherlands.
De Greef, K.H. and Verstegen, M.W.A. (1992) Partitioning of protein and lipid
deposition in the body of growing pigs. Livestock Production Science 35, 317–328.
De Lange, C.F.M. (1995) Framework for a simplified model to demonstrate
principles of nutrient partitioning for growth in the pig. In: Moughan, P.J.,
Verstegen, M.W.A. and Visser-Reyneveld, M.I. (eds) Modelling Growth in the Pig.
Wageningen Pers, Wageningen, Netherlands, pp. 71–85.
Dunkin, A.C. and Black, J.L. (1987) The relationship between energy intake and
nitrogen balance in the growing pig. In: Moe, P.W., Tyrrell, H.F. and Reynolds,
P.J. (eds) Energy Metabolism in Farm Animals. EAAP publication 32. Rowman and
Littlefield, Totowa, New Jersey, pp. 110–113.
Emmans, G.C. (1988) Genetic components of potential and actual growth. In:
Land, R.B., Bulfield, G. and Hill, W.G. (eds) Animal Breeding Opportunities.
BSAP occasional publication 12, British Society for Animal Production,
Edinburgh, UK, pp. 153–181.
Emmans, G.C. (1997) A method to predict the food intake of domestic animals from
birth to maturity as a function of time. Journal of Theoretical Biology 186, 189–199.
Emmans, G.C. and Kyriazakis, I. (1995) A general method for predicting the
weight of water in the empty bodies of pigs. Animal Science 61, 103–108.
Emmans, G.C. and Kyriazakis, I. (1997) Models of pig growth: problems and
proposed solutions. Livestock Production Science 51, 119–129.
280 P. Luiting and P.W. Knap
Ferguson, N.S., Gous, R.M. and Emmans, G.C. (1994) Preferred components for
the construction of a new simulation model of growth, feed intake and nutrient
requirements of growing pigs. South African Journal of Animal Science 24, 10–17.
Knap, P.W. (2000) Variation in maintenance requirements of growing pigs in
relation to body composition; a simulation study. PhD dissertation,
Wageningen Agricultural University, Wageningen, Netherlands.
Kyriazakis, I. and Emmans, G.C. (1992) The effects of varying protein and energy
intakes on the growth and body composition of pigs. 2: The effects of varying
both energy and protein intake. British Journal of Nutrition 68, 615–625.
Luiting, P. (1990) Genetic variation of energy partitioning in laying hens: causes of
variation in residual feed consumption. World’s Poultry Science Journal 46, 133–152.
Moughan, P.J. and Smith, W.C. (1984) Prediction of dietary protein quality based
on a model of the digestion and metabolism of nitrogen in the growing pig.
New Zealand Journal of Agricultural Research 27, 501–507.
Moughan, P.J. and Verstegen, M.W.A. (1988) The modelling of growth in the pig.
Netherlands Journal of Agricultural Research 36, 145–166.
Neter, J., Wasserman, W. and Kutner, M. (1985) Applied Linear Statistical Models.
Irwin, Homewood, Illinois.
NRC (1996) Nutrient Requirements of Beef Cattle. National Academy Press,
Washington, DC.
Pomar, C., Harris, D.L. and Minvielle, F. (1991) Computer simulation of swine
production systems. 1: Modeling the growth of young pigs. Journal of Animal
Science 69, 1468–1488.
Thorbek, G. (1975) Studies on energy metabolism in growing pigs. 2: Protein and fat gain in
growing pigs fed different feed compounds. Efficiency of utilization of metabolizable energy
for growth. Report 424, Statens husdyrbrugsforsøg, Copenhagen, Denmark.
TMV (1991) Technisch model varkensvoeding: informatiemodel. Research report P1.66,
Research Institute for Pig Husbandry, Rosmalen, Netherlands.
Van Milgen, J. and Noblet, J. (1999) Energy partitioning in growing pigs: the use of
a multivariate model as an alternative for the factorial analysis. Journal of
Animal Science 77, 2154–2162.
Van Milgen, J., Quiniou, N. and Noblet, J. (2000) Modelling the relation between
energy intake and protein and lipid deposition in growing pigs. Animal Science
71, 119–130.
Walker, B. and Young, B.A. (1993) Prediction of protein accretion, support costs
and lipid accretion in the growing female pig and dry sow. Agricultural Systems
42, 343–358.
Whittemore, C.T. and Fawcett, R.H. (1976) Theoretical aspects of a flexible model
to simulate protein and lipid growth in pigs. Animal Production 22, 87–96.
Whittemore, C.T. (1995) Modelling the requirement of the young growing pig for
dietary protein. Agricultural Systems 47, 415–425.
Appendix 13.1
He defines the ratio between the two linear regression coefficients as the
marginal minLD/PD ratio and assumes that this parameter is a genetically
determined constant, which makes minLD/PD dependent on MEI:
t × (MEI − v) t MEI − v MEI − v (13.A2)
minLD / PD = = × = margminLD / PD ×
r × (MEI − s) r MEI − s MEI − s
However, because of the law of conservation of energy, the linear
regression of LD on MEI can be written in terms of the parameters of the
linear regression of PD on MEI; De Greef does not mention this:
MEI − MEm − (PD × E / kP )
LD = P =
EL / kL
MEI − MEm − (r × (MEI − s) × EP / kP )
= =
EL / kL (13.A3)
1- r × EP / kP ⎛ MEm − r × s × EP / kP ⎞
= × ⎜ MEI − ⎟=
EL / kL ⎝ 1- r × EP / kP ⎠
= t × (MEI − v)
This shows that t is a function of r, and that v is a function of r, s and MEm,
which automatically makes both t and v constants too. As a consequence,
the ratio between t and r (i.e. marg minLD/PD) is a constant with value (1 –
r × EP / kP) / (r × EL / kL).
Hence given that r = b1 and s = b0, substitution in Eqn 13.A2 shows
that De Greef ’s minLD/PD equals the one of Black (see Table 13.2):
MEI − v
minLD / PD = margminLD / PD × =
MEI − s
MEm − r × s × EP / kP
MEI −
1 − r × EP / kP 1 − r × EP / kP
= × =
r × EL / kL MEI − s
MEm − b0 × b1 × EP / kP
MEI −
1 − b1 × EP / kP 1 − b1 × EP / kP
= × =
b1 × EL / kL MEI − b 0
⎛ MEm − b0 × b1 × EP / kP ⎞
MEI − b0 + b0 − ⎜ ⎟
1 − b1 × EP / kP ⎝ 1 − b1 × EP / kP ⎠
= × =
b1 × EL / kL MEI − b0
⎛ b0 − b0 × b1 × EP / kP − MEm + b0 × b1 × EP / kP ⎞
⎜
1 − b1 × EP / kP ⎜ 1 − b1 × EP / kP ⎟
= × 1+ ⎟=
b1 × EL / kL ⎜ MEI − b0 ⎟
⎜ ⎟
⎝ ⎠
1 − b1 × EP / kP ⎛ MEm − b0 1 ⎞
= × ⎜1 − × ⎟
b1 × EL / kL ⎝ 1 − b1 × EP / kP MEI − b0 ⎠ (13.A4)
14 Mechanistic Modelling at the
Metabolic Level: a Model of
Metabolism in the Sow as an
Example
J.P. MCNAMARA
Department of Animal Sciences, Washington State University
PO Box 646351, Pullman WA 99164-6351, USA
mcnamara@wsu.edu
Introduction
Models are representations of reality. The fields of nutrition, metabolism
and biomedicine have used models to aid in research and its application
since before World War II (see Black et al., 1986; Baldwin, 1995; NRC,
1998, 2001). A model may also be defined as an ordered way of describing
knowledge of some real system. Much research into nutrition of farm
animals since the early 1900s has been used, directly or ‘by default’, to
build, evaluate and improve models of nutrient requirements. An example
would be the Net Energy System, first fully described by Lofgreen and
Garrett (1968) as applied to beef cattle nutrition. Since then, this model
system and others derived from it has been used to determine energy
requirements of many animal species, and to determine energy values of
foods and feeds (NRC, 1998, 2001).
An early model in energy metabolism was the linear equation of fasting
heat production = 0 + 70 kcal/kg BW0.75. Energeticists such as Brody,
Kleiber and Blaxter asked questions about the commonality of energy use
by organisms. This model was derived from empirical data and spurred
decades of discussion, experimentation and further more detailed and
mechanistic model building, encompassing new information as it became
available. These lines of research led to practical empirical models used in
food production and human nutrition as well as to mechanistic research
into the control of energy metabolism in animals. It was the effort of
describing increasingly complex chemical knowledge in strict mathematical
terms, in an iterative fashion as more knowledge became available, and
applying the knowledge in the real world as appropriate, that has lead to
great efficiencies and improvements in nutrition.
Billions of US dollars, represented as conservation of resources and
decreased costs of raising feed for livestock and of the labour for feeding of
livestock, have been saved by application of these nutritional models (NRC,
1998, 2001). If we assume that application of practical models and
resultant changes in ration formulation and feeding has gained a cost
saving of $0.50 per market pig, and an average of 100 million pigs killed in
the US each year for the last 20 years, that is a saving of about US$1
billion. If we make a liberal estimate of 30 states with two pig nutritionists
per state at a salary of $40,000 and each had a total of $100,000 in total
research support for 20 years, that is about $168 million. Not a bad return
on investment. And this is only for pigs.
Our ability to describe metabolic transactions, and their resultant effect
on nutrient requirements is critical to raising food-producing animals in
efficient ways around the globe. As our knowledge of variation in genetic
and environmental situations continues to increase, this author takes the
philosophy that it is only through continuing to develop models of
increasing complexity, ever grounded in validated research data, that we
will continue to improve our true knowledge, wisdom and their
application to feeding the world.
Another use of such models is in teaching. The more complex models
have great utility for demonstrating to students how the complex system
works. In addition, when used in a teaching programme which also
exposes students to practical models used in production or to daily
nutrient requirements, the natural comparison between the research
models and the practical models can help them to see the direct
connections between the specific biological mechanisms and the practical
application of them.
One final set of reasons to continue to expand research in models has
already been given by one much more capable than myself, so I will close
the introduction with a list of reasons for using research models quoted
directly from Chapter 1 of Modelling Ruminant Digestion and Metabolism
(Baldwin, 1995):
Objectives in research modelling:
(a) Integration of existing concepts and data in a format compatible with
quantitative and dynamic analyses.
(b) Reduction of conceptual difficulties in analyses of interactions among elements
of complex systems.
(c) Evaluation of concepts and data for adequacy in both the quantitative and
dynamic domains.
(d) Evaluation of alternative hypotheses for probable adequacy when current
concepts are found to be inadequate, and identification of critical experiments
and measurements.
(e) Estimation of parameter values not directly measurable and interpretation of
new data.
284 J.P. McNamara
Lactation provides a challenge not only to the female, but to the modeller
as well. Key challenges are the interaction of the complex of organs
involved in the adaptation and sustaining of metabolism to the new
physiological state, and the range and speed of change in early lactation.
One goal for the continued improvement of detail and accuracy is to
improve our quantitative understanding of control of metabolic systems.
Pregnancy and early lactation are times of metabolic stress met with a
coordinated response from the hormonal and neural systems (McNamara,
1994). In order to manage the changes in nutrient flows and
interconversions in the animal there is a complex and redundant system of
control factors, better known as hormones and neurotransmitters
(McNamara, 1998; McNamara and Boyd, 1998). Adipose tissue, as an
energy storage reservoir and, as we now understand, an endocrine organ
(Mohamed-Ali et al., 1998), adapts to support fetal growth and lactation
(McNamara et al., 1985; McNamara and Pettigrew, 1994, 2002a,b; Parmley
and McNamara, 1996). These systems are inextricably linked and excellent
reductionist experiments have identified key elements of each subsystem.
However, experiments that describe equation forms and parameter values
for the control exerted on flux are more limited in availability, primarily
because of the cost and complexity of conducting such experiments. A
more detailed analysis of quantifying metabolic control is given in
McNamara and Boyd (1998).
It is not always easy or affordable to do research with sufficient
repeated measures to estimate such rapidly changing fluxes, so some
Mechanistic Modelling at the Metabolic Level 285
and Baldwin, 2000; Phillips et al., 2003) would increase energy for
maintenance by about 8.4 MJ/day (see Baldwin, 1995 for calculations and
stoichiometry). Over 100 days, that is an error of 840 MJ or about 28 kg of
adipose tissue. In a lactating sow, then, we may err by 170 MJ or over 5 kg
of body fat in predicting energy use.
cAMP mRNA
Hormone* or
Nutrient
*K1
Substrate
Amino
Initial acids
substrate(s) enzyme
>>> Pathway >>>
Vm,Ks*
*K2
product
Fig. 14.1. Conceptual model of control of flux through a reaction, indicating types of genetic
control over flux. Conceptual flow of genetic mechanisms of control of flux. The asterisk
indicates states (proteins, nutrients) that are genetically controlled and may have a
measurable heritability. Each state (box) and rate (arrow) can be measured to determine
appropriate parameters. Large dashes indicate a mechanistic state (with components), small
dashes indicate a lower level of organization to the main objective.
288 J.P. McNamara
10
V = Vmax/(1 + (Km/[S])Θ)
9
6
Rate of Reaction
Vm = 10, Km = 3
3
Vm = 15, Km = 3
Vm = 10, Km = 1
2
Theta = 3
0
1 1.8 2.6 3.4 4.2 5 5.8
Substrate Concentration
Fig. 14.2. Example of genetic control over substrate saturation (Michaelis-Menten) reactions.
A typical example of a reaction with a maximal velocity and substrate sensitivity. In an
aggregated pathway model, the Vmax is the genetic component representing total enzyme
concentration, cell number, or tissue mass or protein. The substrate sensitivity constant is
also genetically controlled and may also represent enzyme concentration. The theta value on
the (Km/[S]) function indicates sensitivity to substrate, which is also genetically controlled.
This may represent sensitivity to hormonal or substrate/product allosteric control.
100.0
90.0
80.0
70.0
Percent bound
60.0
50.0
40.0 Normal
Hi M
30.0
Lo K
20.0
Hi Theta
10.0
0.0 2
8
4
5
1
2.
2.
3.
3.
4.
4.
5.
5.
1.
1.
Concentration of Hormone
Fig. 14.3. Example of genetic control over hormone-receptor binding reactions. A typical
example of a hormone-receptor reaction with a maximal binding of receptor [R] and
dissociation constant [K] in response to hormone [H] concentration. The maximal binding of
hormone receptor is the genetic component, as is the dissociation constant. The hormone
concentration is also genetic, but would be represented with another model showing
hormone synthesis and degradation rates. Changing either [R] or K by genetic selection will
change hormone-receptor binding and, possibly, the end result in flux in the cell (by altering
allosteric control of the reaction given in Fig. 14.2, for example).
kinetics of this system, and there are data in adipose tissue on the response
of cAMP and lipolysis to concentrations of catecholamines (McNamara et
al., 1992; McNeel and Mersmann, 1999 and references therein).
Expanding on Eqn 14.5 may yield:
UTsFa = (vTsFa /(1 + (MTsFa /Clh)))* (Qts**0.67)*
(14.6)
(1((QsTs/Qts)**thTsFa))
VTsFa = v1TsFa*(cAMP/rcAMP) (14.7)
cAMP = KcAMP*[B2RecNE] (14.8)
[B2RecNE] = MB2Rec /(1 +(Kd/[NE]θ)) (14.9)
such that the maximal velocity of lipolysis (VTsFa; let us say this represents
total active hormone sensitive lipase) is a function of cAMP concentration
(cAMP) in relation to a reference concentration (rcAMP); cAMP con-
centration is a function of per cent beta-2-adrenergic receptor bound
([B2RecNE]); and bound receptor is in turn a function of maximal binding
(MB2Rec), dissociation constant (Kd) and norepinephrine concentration
([NE]θ), raised to a power (sigmoidal function, as in Fig. 14.3). Further
homeorhetic control (Bauman and Vernon, 1993; McNamara and Boyd,
1998) on the beta-adrenergic response may be exerted by lactational
hormone to increase per cent bound during lactation:
MB2Rec = MB2Rec * (Chl/rCHl) (14.10)
This treatment is basically the same as that proposed in the cow model of
Baldwin (Baldwin, 1995; Baldwin et al., 1987a,b,c), but with more points of
control exerted on the final pathway flux, as in reality. A point to note is
that in fact in any given condition (nutritional status, day of lactation), the
actual pathway flux described in the new model may be exactly the same as
in the old model, but with a fuller description of control.
Another example of descriptions of control within this model of flux is
that of glucose utilization. It is glucose around which the major regulatory
processes of the body have evolved. The brain and central nervous system
require glucose. Although the mass amount of glucose required by the
brain is small compared to that used by the mammary gland or other
organs, the regulatory mechanisms invoked as glucose availability changes
has a major affect on metabolic rates in other organs. Recent work in mice
shows that altering glucose use by food restriction, or by gene insertion for
proteins such as the insulin dependent glucose transporter (GLUT4) result
in changes in transcription of thousands of genes (Fu et al., 2004).
Carbohydrate in the body which is metabolized for energy (or to make
fat or lactose) eventually is converted to triose phosphates or glucose, or is
used through the same metabolic pathways so, for simplicities sake we can
aggregate a lot of this. So changes in glucose concentration or pool size
(Gl) in the body are summed as:
dQGldt = PAaGl + PAbGl + PPaGl + P6TsFa – UGlTs – U6FaTm
(14.11)
– U6FaTs – UGlCd – UGlGc – U6GlTs – UGlLm – UglTm
Mechanistic Modelling at the Metabolic Level 295
0.016
0.012
0.01
0.008
0.006
0.004
0.002
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Day of Lactation
0.8
0.7
Gluconeogenesis, moles amino acid /d
0.6
0.5
0.4
0.3
0.2
0.1
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Day of Lactation
Fig. 14.4. Simulations of glucose use in a model of metabolism in the sow. (a) Shows the
concentrations of blood glucose during lactation and (b) shows rates of gluconeogenesis
from amino acids. The different lines indicate alterations in either intake or in demand by
mammary gland (McNamara and Pettigrew, 2002a,b). This demonstrates the behaviour and
sensitivity of the model. Further work will then link glucose concentrations and downstream
endocrine signals to reproductive functions (Fig. 14.6).
Mechanistic Modelling at the Metabolic Level 297
Using this information, we easily went back to test the effects in the sow
model described herein. We had known that this model also over-
predicted the accumulation of body fat on low-fat, but not high-fat diets
(McNamara and Pettigrew, 2002a,b), and wondered whether correcting
the ATP yield could prevent this in the sow model. Fig. 14.5 shows that the
answer is ‘It depends’. Reducing the ATP yield by 16% corrected the error
on the low fat diet, but then did not describe enough fat on the high fat
diet. This indicates that there are other errors remaining. Certainly some
of these are related to the parameters for glucose use and body fat
synthesis, which are genetically regulated. Construction of more refined
models including genetic regulation will help us define and correct further
errors.
A question that can now be asked is: ‘Can we select for, or by other
means change, the genetic control of body fat synthesis to improve
efficiency of milk synthesis (without reducing the health of the cow)?’
Complex mechanistic dynamic models, integrating ‘genetics’ into ‘kinetics’,
can help us to predict the potential outcomes of such decisions. These
simple examples given here just touch on the complexity of genetic
changes for ‘practical traits’ at the organism level. It is one thing to do a
‘knockout’ to determine what happens if a protein is not made. But it is a
30
Hi F Obs = 29.4
29
28
Body Fat, kg
27
LoFnorm
26 HiFnorm
LoFHiPO Lo F Obs = 26.3
25
HiFHiPO
24
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Day of Lactation
Fig. 14.5. Effect of altering ATP yield on use of body fat in lactation of sows. If the current
newest estimates of ATP yield from NADH2 and FADH2 of 2.5 and 1.5 are correct, this will
help explain the over-prediction of body fat. The sow model over-predicts the accumulation of
body fat in situations of high fat intake. However, the error does not seem to be in the
stoichiometry of the adipose tissue equations (McNamara and Pettigrew, 2002b). If instead,
the original older estimates of ATP yield were too high, then less glucose and fat would be
oxidized to supply the ATP demand, and more body fat would accumulate. An example of
how a properly constructed model can help to point out errors in knowledge, which can then
be corrected.
Mechanistic Modelling at the Metabolic Level 299
different problem to describe and predict what may happen due to more
practical genetic changes. That is our charge, however.
Differences in milk production, body fat and body protein are genetic
and inherited. Rates of amino acid, fatty acid and glucose use in the major
organs are also genetically controlled. Empirical applied data (feed intake,
litter growth) can be collected and compared to a ‘theoretical’ model to ask
such questions as: ‘What would we expect the range of milk production to
be if this number were used as the Vmax for lactose synthesis (or body
protein synthesis)?’ Here again, our quantitative knowledge of
biochemistry allows us to ask such questions. What factors control gene
transcription in the mammary gland? Conceptual and quantitative models
will be of great assistance here to codify our knowledge and set hypotheses.
Hypothalamus
Leptin
Oestrogen
Adipose Pituitary
Glucose
NH3 Ovulated
Amino Egg(s)
Acids
Fetus(es)
Implantation Corpus
Body Luteum
Protein
Fig. 14.6. Description of a model of nutrient control of reproduction in the sow. Flux rates of
several nutrients, and secondary hormonal responses, have direct effects on follicular
development, ovulation and uterine environment. Why do we not yet have a mechanistic
model of nutrient/reproduction interactions, given the wealth of information we have? Here is
presented a flow diagram of one such model. Equation forms will need to be derived and
parameters estimated. The model then becomes a framework for identifying inadequacies in
our knowledge.
When one stands on the shoulders of giants, one is humbled and grateful
that the horizon is closer in view. While the objective of eventually
understanding everything that happens in an animal may be far beyond
the horizon, progress is being made and our true purpose of feeding the
world efficiently throughout many geographical areas and for many
centuries is becoming closer. A quantitative and systems approach is the
only way to continue upon this path. Reductionistic research, especially
into the true functionality of the genome and physiological control of
metabolism will be essential to improvement. In addition, continued efforts
in simple, economical and effective chemical methodology for defining
feedstuffs is a must. But over-riding all, is the integration of knowledge,
data, and concepts, into an organized representation of reality: a model
that will achieve our goal. Research can help us overcome politics and
other human frailties with a system that can ensure adequate food
supply for all. That is our ultimate goal, and societal, industrial and
governmental support for the modelling approach to research is critical to
achieve it.
References
Baldwin, R.L. (1995) Modelling Ruminant Digestion and Metabolism. Chapman & Hall,
New York, pp. 469–518.
Baldwin, R.L., France, J. and Gill, M. (1987a) Metabolism of the lactating cow. I.
Animal elements of a mechanistic model. Journal of Dairy Research 54, 74–105.
Baldwin, R.L., France, J. and Gill, M. (1987b) Metabolism of the lactating cow. II.
Rumen elements of a mechanistic model. Journal of Dairy Research 54, 106–132.
Baldwin, R.L., France, J., Beever, D.E., Gill, M. and Thornley, J.H.M. (1987c)
Metabolism of the lactating cow. III. Properties of mechanistic models suitable
for evaluation of energetic relationships and factors involved in the partition of
nutrients. Journal of Dairy Research 54, 133–145.
Bastianelli, D., Sauvant, D. and Rerat, A. (1996) Mathematical modeling of
digestion and nutrient absorption in pigs. Journal of Animal Science 74,
1873–1887.
Bauman, D.E and Vernon, R.G. (1993) Effects of exogenous somatotropin on
lactation. Annual Review of Nutrition 13, 437–461.
Black, J.L., Campbell, R.G., Williams, I.H., James, K.J. and Davies, G.T. (1986)
Simulation of energy and amino acid utilization in the pig. Research and
Development in Agriculture 3, 121–145.
Bosch, F., Pujol, A. and Valera, A. (1999) Transgenic mice in the analysis of
metabolic regulation. Annual Review of Nutrition 18, 207–232.
302 J.P. McNamara
Collins, J.C., Harris, A., Linsenmeier, R.A. and Olds, S. (2005) Testing hypotheses
of learning at the VaNTH bioengineering education center: applications to
nutrition and metabolism. Chapter 14. In Hargrove, J.L. and Berdanier, C.D.
(eds) Mathematical Modeling in Nutrition and Toxicology. Mathematical Biology
Press, Athens, Georgia, pp. 235–244.
Daniel, P.B., Walker, W.H. and Habener, J.F. (1999) Cyclic AMP signaling and gene
regulation. Annual Review of Nutrition 18, 353–383.
Drackley, J.K. (1999) Biology of dairy cows during the transition period: the final
frontier. Journal of Dairy Science 82, 2259–2273.
Farmer, C. and Palin, M.F. (2005) Exogenous prolactin stimulates mammary
development and alters expression of prolactin-related genes in prepubertal
gilts. Journal of Animal Science 83, 825–832.
Fu, C., Xi, L., Wu, Y., McCarter, R., Richardson, A., Hickey, M. and Han, E.S.
(2004) Hepatic genes altered in expression by food restriction are not
influenced by the low plasma glucose level in young male GLUT4 transgenic
mice. Journal of Nutrition 134, 2965–2974.
Garcia-Bojalil, C.M., Staples, C.R., Risco, C.A., Savio, J.D. and Thatcher, W.W.
(1998) Protein degradability and calcium salts of long-chain fatty acids in the
diets of lactating dairy cows: reproductive responses. Journal of Dairy Science 81,
1385–1395.
Girard, J., Ferre, P. and Foufelle, F. (1997) Mechanism by which carbohydrates
regulate expression of genes for glycolytic and lipogenic enzymes. Annual
Review of Nutrition 17, 325–352.
Hanigan, M.D., Bateman, H.G., Fadel, J.G. and McNamara, J.P. (2005) Metabolic
models of ruminant metabolism: recent improvements and current status.
Journal of Dairy Science 88 (in press).
Hargrove, J.L. (2004) Kinetic basis for modeling gene expression: a historical
overview. In: Hargrove, J.L. and Berdanier, C.D. (eds) Mathematical Modeling
in Nutrition and Toxicology. Mathematical Biology Press, Athens, Georgia,
pp. 265–280.
Hausman, G.J. and Hausman, D.B. (1993) Endocrine regulation of porcine adipose
tissue development: cellular and metabolic aspects. In: Hollis, G.R. (ed.) Growth
of the Pig. CAB International, Wallingford, UK, pp. 49–73.
Hirooka, H., de Koning, D.J., Harlizius, B., van Arendonk, J.A., Rattink, A.P.,
Groenen, M.A., Brascamp, E.W. and Bovenhuis, H. (2001) A whole-genome
scan for quantitative trait loci affecting teat number in pigs. Journal of Animal
Science 79, 2320–2326.
Hurley, W.L., Wang, H., Bryson, J.M. and Shennan. D.B. (2000) Lysine uptake by
mammary gland tissue from lactating sows. Journal of Animal Science 78,
391–395.
Jiang, Z.H. and Gibson, J.P. (1999a) Genetic polymorphisms in the leptin gene and
their association with fatness in four pig breeds. Mammalian Genome 10,
191–193.
Jiang, Z.H. and Gibson, J.P. (1999b) Bi-PASA genotyping of a new polymorphism in
the APOB gene shows no evidence for an association with fatness in pigs.
Animal Genetics 30, 54–56.
Kanis, E., DeGreef, K.H., Hiemstra, A. and Van Arendonk, J.A.M. (2005) Breeding
for societally important traits in pigs. Journal of Animal Science 83, 948–957.
Lofgreen, G.P. and Garrett, W.N. (1968) A system for expressing net energy
requirements and feed values for growing and finishing beef cattle. Journal of
Animal Science 27, 793–806.
Mechanistic Modelling at the Metabolic Level 303
McNamara, J.P., Phillips, G.J., Citron, T.L. and Sage, J. S. (2006) Challenge and
improvement of a model of dairy cattle metabolism to describe changes in
metabolism in cows in early lactation. In: International Workshop on Modelling
Nutrient Utilization. Proceedings of 6th International Workshop on Modelling in Farm
Animals. CAB International, Wallingford, UK (in press).
McNeel, R.L. and Mersmann, H.J. (1999) Distribution and quantification of beta1-,
beta2-, and beta3-adrenergic receptor subtype transcripts in porcine tissues.
Journal of Animal Science 77, 611–621.
Mohamed-Ali, V., Pinkney, J.H. and Coppack, S.W. (1998) Adipose tissue as an
endocrine and paracrine organ. International Journal of Obesity and Related
Metabolic Disorders 22, 1145–1158.
National Research Council (1998) Nutrient Requirements of Swine, 10th revised edn.
National Academy Press, Washington, DC.
National Research Council (2001) Nutrient Requirements of Dairy Cattle, 7th revised
edn. National Academy Press, Washington, DC.
Overton, T.R. and Waldron, M.R. (2004) Nutritional management of transition
dairy cows: strategies to optimize metabolic health. Journal of Dairy Science 87,
E105–E119.
Parmley, K.L.S. and McNamara, J.P. (1996) Lipid metabolism in adipose tissue of
pigs fed varying amounts of energy. Journal of Nutrition 126, 1644–1656.
Pettigrew, J.E., Gill, M., France, J. and Close, W.H. (1992a) A mathematical
integration of energy and amino acid metabolism. Journal of Animal Science 70,
3742–3761.
Pettigrew, J.E., Gill, M., France, J. and Close, W.H. (1992b) Evaluation of a
mathematical model of lactating sow metabolism. Journal of Animal Science 70,
3762–3773.
Phillips, G.J., Citron, T.L., Sage, J.S., Cummins, K.A., Cecava, M.J. and McNamara,
J.P. (2003) Adaptations in body muscle and fat in transition dairy cattle fed
differing amounts of protein and methionine hydroxy analog. Journal of Dairy
Science 86, 3634–3647.
Quintanilla, R., Milan, D. and Bidanel, J.P. (2002) A further look at quantitative
trait loci affecting growth and fatness in a cross between Meishan and Large
White pig populations. Genetic Selection and Evolution 34, 193–210.
Reynolds, C.K., Aikman, P.C., Lupoli, B., Humphries, D.J. and Beever, D.E. (2003)
Splanchnic metabolism of dairy cows during the transition from late gestation
through early lactation. Journal of Dairy Science 86, 1201–1217.
Roehe, R. (1999) Genetic determination of individual birth weight and its
association with sow productivity traits using Bayesian analyses. Journal of
Animal Science 77, 330–343.
Staples, C.R., Thatcher, W.W. and Clark, J.H. (1990) Relationship between ovarian
activity and energy status during the early postpartum period of high
producing dairy cows. Journal of Dairy Science 73, 938–947.
Tepperman, J.and Tepperman, H.M. (1970) Gluconeogenesis, lipogenesis and the
Sherringtonian metaphor. Federation Proceedings 20, 1284–1293.
Westwood, T., Lean, I.J. and Garvin, J.K. (2002) Factors influencing fertility of
Holstein dairy cows: a multivariate description. Journal of Dairy Science 85,
3225–3237.
15 The Place of Models in the New
Technologies of Production
Systems
D.M. GREEN1 AND D.J. PARSONS2
1Universityof Oxford, Department of Zoology, South Parks Road, Oxford,
OX1 3PS, UK; 2Cranfield University, Silsoe, Bedford, MK45 4HS, UK.
darren.green@zoo.ox.ac.uk
Introduction
The problem
Farmers aim to produce livestock within the tight targets imposed on them
by their customers (increasingly, supermarkets) and to do so at a profit.
Both the consumer and the government, however, are increasingly being
influenced by concerns other than price alone, including issues such as
environmental damage and pollution, human and animal health and
animal welfare (Fox and Sachs, 2003). The role of science and technology
in agriculture is to enable the farmer both to increase profit and to address
the further concerns of the consumer.
At the core of an IMS lies a model, which provides two benefits. First, it
allows forward predictions to be made of the impact of management
decisions: the best course of action can then be automatically determined
by the system. Secondly, IMS involves considerable automated monitoring
of livestock and its environment therefore the model provides a powerful
diagnostic tool to detect when livestock performance deviates from that
expected and to alert the stockperson to the potential problem.
There are few examples in the literature of fully functioning IMS
systems, but various authors have made steps in this direction. Of these,
automated feeding and weighing systems will be discussed below.
Halachmi et al. (1998) discuss an automatic feeding system for cows, which
allows a substantial collection of data, but this is not integrated with a
model or control system. Pietersma et al. (2001) developed a decision
support system to analyse lactation data for cattle and to examine
individual performance, but again without a model or control system.
Many models of pig and poultry nutrition exist, ranging in scope from
those describing the nutrition of a single animal up to those describing the
whole farm and its economy. But aside from the examples described
below, these models have not yet been integrated into on-line livestock
production systems.
The general approach shared by the IMS systems described below is
that of model-based predictive control (IMS Pigs1; IMS Poultry2; Aerts et al.,
2003a,b). This is a two-stage process: first, continuously collected data are
fed back into an adaptive (growth) model; second, forward predictions of
this model are used to identify the management regime required to satisfy
the control objectives. How the various components of a generic IMS
system interact is shown in Fig. 15.1. The goal of IMS is to produce a
system that is entirely on-line and operates in real-time without human
intervention (in normal operation: problems would require intervention
when detected). The systems currently in development achieve this to
varying degrees.
IMS brings together many of the subjects that have been discussed
elsewhere in this book. Different approaches can be used to model feed
intake, nutrient and energy usage, growth and feed intake: approaches
from detailed modelling at the metabolic level (e.g. Green and
Whittemore, 2003) through to data-based ‘black box’ approaches (e.g.
Aerts et al., 2003a,b). Regardless of the approach used, in practice no
animal will ever perform exactly as predicted by a model, due to
consequences of the social, disease and thermal environment (see
Wellock et al., Chapter 4, this volume) and genetic effects. These
unknown parameters must be estimated from the data before the model
can be used for accurate prediction. Identifying these parameters is a
form of model inversion problem (see Doeschl-Wilson et al., Chapter 9,
this volume).
This chapter addresses the components and concepts of IMS systems.
First, the various pressures and objectives of IMS are addressed, then the
equipment and data sources available on the farm. The models themselves
Models in the New Technologies of Production Systems 307
Previous
status
Diet and
Run model
environment
Predict
status now
Review and
amend model
Measure Diet and
status now environment
Amend diet
Run model
and environment
Predict Future
future status control targets
are then discussed: modelling strategies used for IMS, methods for
optimizing models in response to the collected data, and how model
predictions can be used to make control decisions. Finally, some IMS
systems already developed and in development are discussed, before
considering likely future developments.
low margin per animal sold or, on the other, a lower level of production of
meat of higher value.
2. Environmental objectives. In England, over 70% of nitrates and 40% of
phosphates in water are derived from agricultural land (DEFRA, 2002).
The increasing concern within the EU has led to the designation of nitrate
vulnerable zones and legislation to minimize nitrogen emissions. IMS will
be able to assist here as emissions targets, along with production targets,
can be embedded in the systems. A typical target might be to minimize
nitrogen emissions per animal produced. A model is required to solve this
problem: feed excess dietary protein and growth will be fast, but wasteful
of nitrogen; feed too little dietary protein and growth will be slow, with the
animal contributing to pollution over a longer timescale.
3. Health and welfare objectives. It is not anticipated that automated
systems will ever replace the experienced stockperson. However, where
time is limited, the IMS system will provide a useful extra tool to alert the
stockperson to health and welfare issues. Indeed, where animals are
continuously monitored in terms of growth rate and feed intake, an IMS
system may alert the farmer to a problem before any clinical signs appear.
For example, an animal with low intake and with both low predicted and
low observed growth would be categorized as suffering from low appetite.
Alternatively, an animal with normal intake and normal predicted growth,
but low observed growth (i.e. a high feed conversion ratio) would be
designated a sick animal.
Multiple objectives. The optimum for control is much more difficult to
determine where there are multiple control objectives. A discussion of the
complexities of multiple criteria decision making is beyond the scope of this
chapter. It may, however, be simpler to consider additional objectives as
constraints on the system; this may be appropriate where, for example, there
are government-imposed limits on emissions associated with production.
The IMS systems IMS Pigs and IMS Poultry and that of Aerts et al.
(2003a,b) were not equipped with full economic models of production,
having specific growth targets already set, in order to test the ability of
control to achieve such targets. For the poultry systems, the target was
weight at a particular age; for pigs, as well as growth rate or final size, an
additional objective was specified: level of fatness at the end of the trial.
This is an important target in a species where payment by the
slaughterhouse may be based on carcass quality.
Equipment
Fig. 15.2. Screenshot from a running VIA system showing identification of three body regions.
for poultry (de Wet et al., 2003). VIA analysis of pigs from alternative
viewing angles (side and rear views) is also currently under investigation.
Such data may provide measurement of condition, which is particularly
important for sows (Schofield, 2004): but note again the concern over
camera placement.
Automatic (conventional) weighing systems have long been available
for both pigs and poultry. For example, Williams et al. (1996) report on the
effectiveness of weighing pigs through automatic forelegs-only weighing
platforms associated with feeders. For pigs, electronic tagging allows
collection of weight data for individual animals, whereas for poultry,
average weights of the birds visiting the weigher are obtained.
Pig fatness is generally determined by backfat depth at the P2 site (in
the UK). Though such measurements would be valuable input into IMS
systems for pigs, providing an added dimension of data for model
adaptation, its manual measurement can cause distress and is labour
intensive. Ultrasonic P2 measurement remains currently the most practical
means of easily determining carcass composition, amongst a field including
ultrasound (Newcom et al., 2002) and alternatives such as CT scanning
(Kolstad, 2001) and MRI (Mitchell et al., 2001), whose use is restricted to
research.
Feeding equipment is at once both a potential sensor and a controller.
Systems exist for controlling and measuring feed intake and blend for both
pigs and poultry (Filmer, 2001; Ellis and Hyun, 2005), though whether
feeding other than ad libitum is possible will depend upon housing
conditions.
Models in the New Technologies of Production Systems 311
Modelling strategies
system to make control decisions at the beginning of the trials, before data
had been collected.
The IMS Poultry programme (Frost et al., 2003; Stacey et al., 2004) uses
a semi-mechanistic model of poultry growth: a compromise between a
complex, slow, mechanistic model, with extensive a priori assumptions
concerning growth built in, and a quick-running empirical model, which
would be of limited use early in a trial. This model was built from the
principles established by Emmans (1981, 1987, 1989, 1994), Emmans and
Fisher (1986) and Hancock et al. (1995). Also for poultry, Aerts et al.
(2003a) chose an empirical approach, and used a recursive linear
regression relationship to relate cumulative feed intake to mass, using a
limited number of past measurements to predict future weight; they found
a time-window of 5 days to give good forward prediction (Aerts et al.,
2003b).
Model adaptation
Environmental, health, and genetic effects will mean that at run time, the
model will produce output that deviates to a greater or lesser degree from
the observed data. Therefore, the model is continuously updated by
reparameterization from these data. Model parameterization can be
considered as a function minimization problem, where the function to be
minimized is a loss function describing the poorness of fit of the model;
minimization by least squares or negative log-likelihood are the two main
approaches here. Strategies of function minimization are many and varied
(Press et al., 1992), and which is the most appropriate depends upon the
nature of the model response surface (in however many dimensions there
are parameters to be optimized). The surface can have single or multiple
minima, be continuous or discontinuous, differentiable or nondifferentiable
(Fig. 15.3). The most simple and foolproof method is the grid search (in one
dimension, linear search): this method searches through the whole of
parameter space, with a given precision, to find the minimum value.
However, such a brute force approach requires an extraordinarily large
amount of processing time, especially where the number of dimensions
(model parameters) is not small.
Several more ‘intelligent’ approaches exist, which can be divided
generally into methods that only require evaluation of the model/function,
and those that in addition require evaluation of the gradient (Press et al.,
1992). The latter methods are less appropriate for models where output
shows plateaux and discontinuities; such output is possibly more likely
encountered with complex mechanistic models. Gradient-following
approaches include the Newton-Raphson and quasi-Newton methods (Gill et
al., 1981), as well as the Marquardt-Levenberg algorithm (Marquardt, 1963).
The revised simplex method (Nelder and Mead, 1965) does not require
calculation of gradients, but does follow the descent of the model response
surface and so can become trapped at local minima. The methods of
Models in the New Technologies of Production Systems 313
Output variable
Input variable
Stacey et al. (2004) draw attention to the question of how much data it
is necessary to collect before model adaptation should be performed.
Though this period depends critically on the confidence level used, a
certain minimal period must elapse before statistically significant changes
in animal state can be detected. Kearney et al. (2004), on automatic
weighing of cattle, recommend collection of data for up to 2 months in
order to quantify growth rate. Stacey et al. (2004) decided upon a period of
14 days for growing chickens; this was chosen as a compromise between
being able to respond correctly and promptly to differences in animal
growth rate, and responding falsely to measurement errors. For pigs,
White et al. (2004) noted that for platform weighers, a period of 4–13 days
is required before changes in pig state can be detected with 95%
confidence; and for VIA systems, 8–10 days. Precisely quantifying growth
rate would take longer. Aerts et al. (2003a,b) chose a window of 5 days of
collected data on which to base forward predictions.
In the IMS Poultry, IMS Pigs and Aerts et al. (2003a,b) growth models, a
single dimension of data, live weight, was available for model optimization.
With VIA (White et al., 2004) or with automatic fat depth measurements
(Frost et al., 2004), there is the potential of extra dimensions of data in
future systems. In this case, there will be greater potential to include more
parameters and preserve model robustness.
Control
Controllable elements
Objectives
In the IMS Poultry project, both control of feed intake and control of
feed blend (varying dietary protein content between two extremes) were
possible. With each control variable able to be specified for each day of the
trial, the dimensions of the problem were at least 30. With this large
number of dimensions, a genetic algorithm (Holland, 1975; Goldberg,
1989) was found to be robust, and operated at an acceptable speed (Stacey
et al., 2004). The system of Aerts et al. (2003a,b), also studying poultry,
considered control by restricted feed intake alone. These authors fitted a
single control variable – the amount of food to be fed in the succeeding 24
hours – to follow a defined target trajectory for growth.
Rationing of feed was not possible in the IMS Pigs study. But here, pigs
were housed in relatively small groups of 12 and the feed blend could be
varied at the individual feeder level (again, dietary protein content varied
between two extremes). This control variable was divided into four control
periods, as tests showed that little improvement in control would be gained
by using shorter periods (Parsons et al., 2004). The number of dimensions
here being considerably smaller than in Stacey et al. (2004), the revised
simplex algorithm was found to be suitable for optimization, as was also
used for model adaptation in this study.
In the case of blending feed as a control variable, both the IMS Pigs
and IMS Poultry projects considered not the blend itself, but the rate of
change of the blend, constrained within narrow bounds. This prevented
large step changes or quick changes in the blend of the diets, which might
be poorly tolerated.
periods. Where an animal was eating more than an average amount, for
the purposes of future prediction (and therefore control) this behaviour
was expected to continue for a period, after which average feed intake was
then assumed. The problem of feed intake prediction was not an issue for
the growth model of Aerts et al. (2003a,b), who fed specific feed rations to
groups of chickens subject to control.
The IMS Pigs project investigated the performance of a novel system for
pig production and pollution control (Green et al., 2004; Parsons et al.,
2004). This system is work in development and not all components are
online and fully automated. Nevertheless, the scientific proof of principle
has been shown. The main data stream for IMS Pigs was a VIA system
(Marchant et al., 1999; Schofield et al., 1999) which provided automatic live
weight estimates on a daily basis for individual animals. It was
complemented with manual P2 backfat measurements and manual
measurements of live weight. Feeding was ad libitum but the crude protein
content of the feed was varied between 13 and 19% on a per-pen basis
(manually, at the instruction of the IMS controller). Each of the 12 pens
housed 12 pigs during each trial, pairs of pens sharing one of six
controlled-environment rooms at ADAS Terrington, Norfolk, UK.
A revised simplex method (Nelder and Mead, 1965) was used to fit two
parameters of a mechanistic model of pig growth (Green and Whittemore,
2003, 2005) to the data collected from the VIA system. In effect, one
parameter specified the efficiency of growth and the other defined the
partitioning of dietary nutrients between protein and lipid deposition.
Weekly forward predictions from the fitted models were used to determine
the crude protein content to be blended for each pen in order to best
satisfy the control objectives. Fitting of model output to the control
objectives was also carried out using the revised simplex approach.
In the trials, two sets of objective targets were specified: targets for
weight gain (50 or 60 kg gain during the growth period) and for P2
backfat depth at the slaughter point (12 or 16 mm). Results from one of
the trials are shown in Table 15.2. Model fit was good before optimization,
with the means of the final observed and predicted weights differing by
only 6 kg. Optimization reduced this to less than 1 kg, even where
Models in the New Technologies of Production Systems 319
Table 15.2. Growth targets in IMS Pigs. Growth targets for each pen,
with mean final deviation from targets. Standard errors are shown in
brackets.
Target Deviation
Weight gain (kg) (approximate) 50 2.1 (2.4)
50 2.3 (0.9)
60 –5.8 (1.5)
60 2.0 (2.4)
Final P2 backfat depth (mm) 12 –0.9 (0.53)
12 0.2 (0.60)
16 –2.1 (0.72)
16 –2.4 (0.68)
optimization was performed only on the early part of the collected data.
This shows the optimized models to have good predictive power.
Control of fat depth was partially successful. The lower target was
achieved with good accuracy, but the pigs set the higher target ended the
trial 2–2.5 mm thinner than specified. However, reference pigs fed only on
the low protein diet were similarly below target, showing that achievement
of this target was beyond the capability of the system. Weight gain control
was achieved within 3 kg of the targets on three of the four pens. The
fourth pen finished below target by nearly 6 kg. This was unavoidable
because these pigs, although on target until the penultimate week of the
trial, then showed an unexplained reduction in growth rate. Where such
changes, possibly caused by a sub-clinical illness, occur early in the trial,
there is time for the IMS system to take corrective action. In this case,
there was no time remaining on trial for such a severe departure from the
target to be corrected. Nevertheless, for healthy pigs, this study provides
proof of principle of control of pig weight through control of diet on ad
libitum feeding. If rationed feeding is implemented, greater control of both
weight and fatness can be expected.
The IMS Poultry project uses, as does the IMS Pigs project, the IMS
structure shown in Fig. 15.1 (Frost et al., 2003; Stacey et al., 2004).
Compared with the individual animals studied in IMS Pigs the commercial-
sized scale here was much larger: each of the eight houses on the trial farm
held 30,000–40,000 birds. Mean live weight for each house was provided
by automatic ‘Flockman’ weigher systems (http://www.flockman.com), as
with IMS Pigs providing daily weight data. Again, as with IMS Pigs, the
main controller was diet blend: the birds were usually fed ad libitum, but
the composition of the food was varied to give a range of protein content.
Rationed feed intake was, however, available in some trials.
320 D.M. Green and D.J. Parsons
Future Prospects
Further progress in IMS systems for pigs and poultry are to some extent
determined by the development of sensor systems. For sows, some
progress has been made in the development of VIA systems (Schofield,
2004). However, the changes during pregnancy in the sow are small
compared with those of the growing pig, being more a matter of shape
than size, and this provides more of a challenge for the VIA system
(Doeschl-Wilson et al., 2005). Side-view VIA is more appropriate for
evaluating the sow, but needs more awkward placement of cameras than
the overhead placement used for viewing the growing pig. Amongst other
uses for VIA is the potential to track animal movements (Tillett et al.,
1997), thereby providing a measure of activity and its concomitant energy
usage.
Another data source that may prove useful in IMS is sound.
Vocalizations are known to be an indicator of welfare in both pigs and
poultry (Manteuffel et al., 2004) and progress has been made in
recognition of vocalization types with neural networks (Chedad et al.,
2001). For reviews of recent developments in the detection of chemicals
relating to health and welfare, nuisance, or pollution see Persaud (2001)
and Frost et al. (1997).
The IMS systems described above are all indoor systems. IMS outdoors
will provide more obstacles. VIA systems would have to be more robust
and contend with more varied lighting conditions than are found indoors.
Modelling of the thermal environment and its effects on energy use would
be considerably more difficult than in the controlled conditions of indoor
housing.
Models in the New Technologies of Production Systems 321
Conclusions
Acknowledgements
Endnotes
1. Integrated Management Systems for pig nutrition control and pollution reduction.
LINK Sustainable Livestock Production programme funded project; UK
Department of Environment, Food, and Rural Affairs (formerly MAFF)
http://www.defra.gov.uk/science/Link/Agriculture/SLP/
Key references: Parsons et al. (2004); Green and Whittemore (2003,
2005); White et al. (2004).
2. Integrated management systems to enhance efficiency and pollution control in
poultry production. As above.
Key references: Frost et al. (2003); Stacey et al. (2004).
References
Aerts, J.-M., Wathes, C.M. and Berckmans, D. (2003a) Dynamic data-based
modelling of heat production and growth of broiler chickens: development of
an integrated management system. Biosystems Engineering 84, 257–266.
Aerts, J.-M., Lippens, M., De Groote, G., Buyse, J., Decuypere, E., Vranken, E. and
Berckmans, D. (2003b) Recursive prediction of broiler growth response to
food intake by using a time-variant parameter estimation method. Poultry
Science 82, 40–49.
Chedad, A., Moshou, D., Aerts, J.-M., van Hirtum, A., Ramon, H. and Berckmans,
D. (2001) Recognition system for pig cough based on probabilistic neural
networks. Journal of Agricultural Engineering Research 79, 449–457.
DEFRA (2002) Nitrates – Reducing Water Pollution from Agriculture. http://www.defra.
gov.uk/environment/water/quality/nitrate/
de Wet, L., Vranken, E., Chedad, A., Aerts, J.-M., Ceunen, J. and Berckmans, D.
(2003) Computer-assisted image analysis to quantify daily growth rates of
broiler chickens. British Poultry Science 44, 524–532.
322 D.M. Green and D.J. Parsons
Doeschl, A.B., Green, D.M., Whittemore, C.T., Schofield, C.P., Fisher, A.V. and
Knap, P.W. (2004) The relationship between the body shape of living pigs and
their carcass morphology and composition. Animal Science 79, 73–83.
Doeschl-Wilson, A.B., Green, D.M., Fisher, A.V., Carroll, S.M., Schofield, C.P. and
Whittemore, C.T. (2005) The relationship between body dimensions of living
pigs and their carcass composition. Meat Science 70, 229–240.
Ellis, M. and Hyun, Y. (2005) Feed Intake Behavior and Implications for Energy Utilization.
http://www.traill.uiuc.edu/porknet/paperDisplay.cfm?Type=paper&ContentID=63
Emmans, G.C. (1981) A model of the growth and feed intake of ad libitum fed
animals, particularly poultry. In: Hillyer, G.M., Whittemore, C.T. and Gunn,
R.G. (eds) Computers in Animal Production, Occasional Publication No. 5. British
Society of Animal Production, Thames Ditton, UK, pp. 103–110.
Emmans, G.C. (1987) Growth body composition and feed intake. World Poultry
Science 3, 208–227.
Emmans, G.C. (1989) The growth of turkeys. In: Nixey, C. and Grey, T.C. (eds)
Recent Advances in Turkey Science. Butterworths, London, pp. 135–166.
Emmans, G.C. (1994) Effective energy: a concept of energy utilisation applied
across species. British Journal of Nutrition 71, 801–821.
Emmans, G.C. and Fisher, C. (1986) Problems in nutritional theory. In: Fisher, C.
and Boorman, K.N. (eds) Nutritional Requirements of Poultry and Nutritional
Research. Butterworths, London, pp. 9–39.
Filmer, D. (2001) Nutritional management of meat poultry. In: Wathes, C.M., Frost,
A.R., Gordon, F. and Wood, J.D. (eds) Integrated Management Systems for
Livestock. Occasional Publication No. 28. British Society of Animal Science,
Edinburgh, UK, pp. 133–146.
Fox, L. and Sachs, J. (2003) The Meatrix http://www.themeatrix.com/
Frost, A.R. (2001) An overview of integrated management systems for sustainable
livestock production. In: Wathes, C.M., Frost, A.R., Gordon, F. and Wood, J.D.
(eds) Integrated Management Systems for Livestock. Occasional Publication No. 28.
British Society of Animal Science, Edinburgh, UK, pp. 45–50.
Frost, A.R., Schofield, C.P., Beaulah, S.A., Mottram, T.T., Lines, J.A. and Wathes,
C.M. (1997) A review of livestock monitoring and the need for integrated
systems. Computers and Electronics in Agriculture 17, 139–159.
Frost, A.R., Parsons, D.J., Stacey, K.F., Robertson, A.P., Welch, S.K., Filmer, D. and
Fothergill, A. (2003) Progress towards the development of an integrated
management system for broiler chicken production. Computers and Electronics in
Agriculture 39, 227–240.
Frost, A.R., French, A.P., Tillett, R.D., Pridmore, T.P. and Welch, S.K. (2004) A
vision guided robot for tracking a live, loosely constrained pig. Computers and
Electronics in Agriculture 44, 93–106.
Gill, P.E., Murray, W. and Wright, M.H. (1981) Practical Optimization. Academic
Press, New York.
Goldberg, D.E. (1989) Genetic Algorithms in Search, Optimization and Machine Learning.
Addison-Wesley, Reading, MA, USA.
Green, D.M. and Whittemore, C.T. (2003) Architecture of a harmonised model of
the growing pig for the determination of dietary net energy and protein
requirements and of excretions into the environment (IMS Pig). Animal Science
77, 113–130.
Green, D.M. and Whittemore, C.T. (2005) Calibration and sensitivity analysis of a
model of the growing pig for weight gain and composition. Agricultural Systems
84, 279–295.
Models in the New Technologies of Production Systems 323
Green, D.M., Parsons, D.J., Schofield, C.P. and Whittemore, C.T. (2004) Real-time
control of pig growth through an integrated management system (IMS).
Proceedings of the British Society of Animal Science, p. 26.
Halachmi, I., Edan, Y., Maltz, E., Peiper, U.M., Moallem, U. and Brukental, I.
(1998) A real-time control system for individual dairy cow food intake.
Computers and Electronics in Agriculture 20, 131–144.
Hancock, C.E., Bradford, G.D., Emmans, G.C. and Gous, R.M. (1995) The
evaluation of the growth parameters of six strains of commercial broiler
chickens. British Poultry Science 36, 247–264.
Holland, J.H. (1975) Adaptation in Natural and Artificial Systems. University of
Michigan Press, Ann Arbor, Michigan.
Kearney, G.A., Knee, B.W., Graham, J.F. and Knott, S.A. (2004) The length of test
required to measure liveweight change when testing for feed efficiency in
cattle. Australian Journal of Experimental Agriculture 44, 411–414.
Kirkpatrick, S., Gelatt, C.D. and Vecchi, M.P. (1983) Optimization by simulated
annealing. Science 220, 671–680.
Kolstad, K. (2001) Fat deposition and distribution measured by computer tomography
in three genetic groups of pigs. Livestock Production Science 67, 281–292.
Manteuffel, G., Puppe, B. and Schön, P.C. (2004) Vocalization of farm animals as a
measure of welfare. Applied Animal Behaviour Science 88, 163–182.
Marchant, J.A., Schofield, C.P. and White, R.P. (1999) Pig growth and conformation
monitoring using image analysis. Animal Science 68, 141–150.
Marquardt, D.W. (1963) An algorithm for least squares estimation of parameters.
Journal of the Society of Industrial and Applied Mathematics 11, 431–441.
Mitchell, A.D., Scholz, A.M., Wange, P.C. and Song, H. (2001) Body composition
analysis of the pig by magnetic resonance imaging. Journal of Animal Science 79,
1800–1813.
Nelder, J.A. and Mead, R. (1965) A simplex method for function minimization.
Computer Journal 7, 308–313.
Newcom, D.W., Bass, T.J. and Lampe, E.F. (2002) Prediction of intramuscular fat
percentage in life swine using real-time ultrasound. Journal of Animal Science
80, 3046–3052.
Parsons, D.J. (1992) Performance of constrained forms of the simplex method for nonlinear
optimization. IED Technical Note TN 003, Silsoe Research Institute, Silsoe, UK.
Parsons, D.J., Schofield, C.P., Green, D.M., Whittemore, C.T., Carroll, S. and Kay,
R. (2004) Real-time control of pig growth through an integrated management system
(IMS). Institute of Agricultural Engineers, Leuven, Belgium.
Persaud, K.C. (2001) Biosensors for livestock production. In: Wathes, C.M., Frost,
A.R., Gordon, F. and Wood, J.D. (eds) Integrated Management Systems for
Livestock. Occasional Publication No. 28. British Society of Animal Science,
Edinburgh, UK, pp. 37–44.
Pietersma, D., Lacroix, R., Lefebvre, D., Block, E. and Wade, K.M. (2001) A case-
acquisition and decision-support system for the analysis of group-average
lactation curves. Journal of Dairy Science 84, 730–739.
Press, W.H., Teukolsky, S.A., Vetterling, W.T. and Flannery, B.P. (1992) Numerical
Recipes in C. Cambridge University Press, Cambridge, UK.
Schinckel, A.P. and de Lange, C.F.M. (1996) Characterization of growth parameters
needed as inputs for pig growth models. Journal of Animal Science 74, 2021–2036.
Schofield, C.P. (2004) Integrated management systems for pig nutrition control and
pollution reduction. DEFRA LINK project, technical report. Silsoe Research
Institute, UK.
324 D.M. Green and D.J. Parsons
Schofield, C.P., Marchant, J.A., White, R.P., Brandl, N. and Wilson, M. (1999)
Monitoring pig growth using a prototype imaging system. Journal of
Agricultural Engineering Research 72, 205–210.
Stacey, K.F., Parsons, D.J., Frost, A.R., Fisher, C., Filmer, D. and Fothergill, A.
(2004) An automatic growth and nutrition control system for broiler
production. Biosystems Engineering 89, 363–371.
Tillett, R.D., Onyango, C.M. and Marchant, J.A. (1997) Using model-based image
processing to track animal movements. Computers and Electronics in Agriculture
17, 249–261.
Tillett, R., McFarlane, N. and Lines, J. (2000) Estimating dimensions of free-
swimming fish using 3D point distribution models. Computer Vision and Image
Understanding 79, 123–141.
Wellock, I.J., Emmans, G.C. and Kyriazakis, I. (2003a) Modelling the effects of
thermal environment and dietary composition on pig performance: model
logic and concepts. Animal Science 77, 255–266.
Wellock, I.J., Emmans, G.C. and Kyriazakis, I. (2003b) Predicting the consequences
of social stressors on pig food intake and performance. Journal of Animal Science
81, 2995–3007.
White, R.P., Schofield, C.P., Green, D.M., Parsons, D.J. and Whittemore, C.T.
(2004) The effectiveness of a visual image analysis (VIA) system for monitoring
the performance of growing/finishing pigs. Animal Science 78, 409–418.
Whittemore, C.T. (1998) Simulation modelling. In: The Science and Practice of Pig
Production. Blackwell Science, Oxford, UK.
Whittemore, C.T., Green, D.M. and Schofield, C.P. (2001a) Nutrition management
of growing pigs. In: Wathes, C.M., Frost, A.R., Gordon, F. and Wood, J.D.
(eds) Integrated management systems for livestock. Occasional Publication No. 28.
British Society of Animal Science, Edinburgh, UK, pp. 89–95.
Whittemore, C.T., Green, D.M. and Knap, P.W. (2001b) Technical review of the
energy and protein requirements of growing pigs: food intake. Animal Science
73, 3–17.
Whittemore, C.T., Green, D.M. and Knap, P.W. (2001c) Technical review of the
energy and protein requirements of growing pigs: energy. Animal Science 73,
199–215.
Whittemore, C.T., Green, D.M. and Knap, P.W. (2001d) Technical review of the
energy and protein requirements of growing pigs: protein. Animal Science 73,
363–374.
Whittemore, E.C., Emmans, G.C. and Kyriazakis, I. (2003) The problem of
predicting food intake during the period of adaptation to a new food: a model.
British Journal of Nutrition 89, 383–398.
Williams, S.R.O., Moore, G.A. and Currie, E. (1996) Automatic weighing of pigs fed
ad libitum. Journal of Agricultural Engineering Research 64, 1–10.
Wu, J.H., Tillett, R., McFarlane, N., Ju, X.Y., Siebert, J.P. and Schofield, P. (2004)
Extracting the three-dimensional shape of live pigs using stereo
photogrammetry. Computers and Electronics in Agriculture 44, 203–222.
Index
325
326 Index
model optimization
application of 2,10, 14, 22, 79, 98, algorithm 176
165, 263 criteria in model inversion 176
Bristol–Reading 237 pig production strategies 64
deterministic 14, 99, 171, 209, 260, optimum (optimal)
311 amino acid content in feed 77–92
dynamic 14, 101, 169, 202, 209, feeding programme 77–82, 93
212, 285, 298 nutrient density 77
economic 2,308 production environment 212
empirical 4,9–10, 15, 20, 97–103, thermal micro-environment
210–213, 216
163, 230, 282, 311–314
ovarian follicle 229, 232
examples 1
oviposition time 240–243, 255–256
inversion 165
ovulation
by optimization 167
internal 236, 244–245, 254–257
laying hen rate 230, 234–243, 254–256
linear multiple regression 12 ovulatory cycle 229–235, 240–241, 254
metabolic flux 286, 289
mixture 104
Reading 79 partitioning
sow 297–299 nutrients 23, 155, 318
of nutrients during disease 126
stochastic 64, 79, 97–101, 109, 168,
of scarce food resources 117
178, 212, 230, 236, 311
immune pigs 129
teleonomic 9
naïve pigs 129
thermal balance 212
pathogen
see also requirement response challenge 117, 126–136
modelling virulence of 129
progress in 19 phase-shift 241
physiological response 217 photoperiodic control 230
mortality 83, 218, 222 piglets 46, 51
pigs per pen 62
pollution control 318
neutral detergent fibre (NDF) 144 populations 71, 79, 92, 126, 311
nitrogen correlated distributions 85
balance techniques 145 Predicted Mean Vote (PMV) 214
prime sequence length 238, 253–254
excretion 145, 147
programming
retention 29, 147
chance constrained 99, 109–110
see also protein
compromise 14
nursery pigs 43, 44
goal 14, 110
nutrient linear 2,12, 77, 83, 99, 108–110
absorption 10 parametric 14
content, variation in 83–84, 109 quadratic 110
density 33, 43, 77–78 separable 14
partitioning 15, 23, 126, 130 protein
requirements 3,22, 71, 109, 118, deaminated 29, 145, 148, 152
257, 283 deposition rate 25, 152, 261
specifications 2 maximum 265
utilization 10, 79, 297, 306, 311 losses
nutritional status 294 endogenous gut 29
objective function 13–14, 78, 184, 210 integument 29, 30
operations research 12, 103–108, 110 maintenance protein 30
330 Index