reviews reviews -1 TREE TREE vol. 8, 8, no. no. 6, 6, June June 1993 1993 Life colLife history history theory theory and and behavioral Gehavioraleecol- ogy, ogy, two two branchesin branches in the the study study Ofadaptaof adupta- Dynamic DynamicModelsof Models of Behavior: Behavior:An An Extension Extensionof of LifeHistoryTheory life Hook Theory tion, tion, have have relied relied extensivelyon extensively on mathmathematical models, but have tended to ematical models, 6ut have tended to employ different types Ofmodels,and employ different types of models, and different currencies currencies Of of fitness. fitness. Recently, Recently, aa new approachbasedon dynamic, In the 1970s it was observed that new approach based on dynamic, statestateIn the 1970s it was observed that variablemous variable modelshasbeenincreasinglyaphas 6een increasingly ap this life history optimization prinprinplied dapta- this life history optimization plied to to the the study study ot of behavioral behavioral aadaptaciple could be understood in terms tions_In fact, this approachamounts to a ciple could be understood in terms tions. In fact, this approach amounts to a of dynamic programming, as forof dynamic programming, as forunificationof unification of life life history history tkeorg theory and and be6e- mulated mulated by by ællmana. Bellman4. Dynamic Dynamic kavioral ecology,to the extent that the haviorul ecology, to the extent that the programming is programming is an an algorithmic algorithmic line separatingthe two fields is virtually line separating the two fields is virtually method that allows one to determethod that allows one to deterobliterated o6literated. Dynamic models (usually (usually mine the optimal control strategy mine the optimal control strategy solvedby computer) tan yield both Enersolvedby computer) can yield 60th gener- for for aa dynamical dynamical system. system. Dynamical Dynamical al al principles principles and and testable,quantitative testa6le, quantitative or or systems are characterized by their systems are characterized by their qualitative to qualitative predictions predictions about about specificbespecific be- states. I will use the Colin W. Clark Cohn W. Clark studies, one example will studies; one example will be dedescribed in some detail, Finally I will scribed in some detail. Finally I will outline outline the the main main advantages advantages of of the the method, and discuss its principal method, and discuss its principal limitations. limitations. under Foraging under risk of predation years ago ago itit was was that foraging and that foraging and antipredator behavior could could not not be be encompassed by a single model, states. I will use the symbol X(t) to encompassed by a single model, havioraland life history phenomena. fitness havioral and rife history phenomena. denote a system's state X at time t. because the denote a system’s state Xat time t. because the corresponding fitness currencies were incommensurable. The state variable X may have The state variable X may have currencies were incommensurable. Life Life history history theory theory analyses analyses the the one one or or many many components. In In the the Indeed, Indeed, the the currency currency problem problem was was sometimes listed as a maior limibiological setting, state variables trade-offs that organisms face in sometimes listed as a major limitrade-offs that organisms face in biological setting, state variables making of individual individual organisms organisms may may inin- tation tation ofof optimization optimization theory theory inin making decisions decisions that that affect affect their their of evolutionary biol*y'_ I clude En)dy mass, somatic energy fecundity and mortality schedules, evolutionary biology7. I Will will use use fecundity and mortality schedules. clude body mass, somatic energy two very simple rncxiels to show Most such trade-offs can be classireserves, other reMost such trade-offs can be classi- reserves, other biochemical re- two very simple models to show how easily the currency problem fied as how easily the currency problem fied as being concerned concerned either either with with serves, serves, territory territory size, size, number number of of can resolved8. The illusoffspring being cared for, and other current versus future reprcKiuction, offspring being cared for, and other can be resolved6. The models illuscurrent versus future reproduction, trate dynamic mcxieling and the biologically relevant factors;. or with size versus of offor with size versus number of off- biologically relevant factors5. trate dynamic modeling and the use of state variables. These state variables change over spring'. But many important spring’. But many important be- These state variables change over use of state variables. First, consider havioral First, consider aa continuously continuously havioral decisions decisions involved involved with with time. time, in in ways ways that that are are affected affected by by reprcxiucing foraging, foraging, predator predator avoidance, avoidance, territerri- the the organism's organism’s Ekhavior. behavior. Behavior Behavior reproducing animal animal that that forages forages for a certain of also affects the organism's reprotorial for a certain period of time time and and torial defense, defense, migration, migration, social social bebe- also affects the organism’s reprohavior, then lays lays aa clutch clutch ofof eggs eggs deterdeterductive output, output, i.e. i.e. its its fitness. fitness. then havior, etc., etc., do do not not readily readily fit fit into into ductive mined by her foraging success. The Finally, the Finally, the the state state XX may may be affected affected mined by her foraging success. The the classical classical life life history history framework. framework. Yet these decisions also affect surby extemal environmental events events prcress process isis repeated repeated for for aa total total of of TT Yet these decisions also affect sur- by external environmental (or as long as the animal not entirely under the organism's Vival and reproduction, and have periods (or as long as the animal vival and reproduction, and have not entirely under the organism’s survives). control. therefore survives). In In any any given given time time peritxi period therefore been subiect subject to to the the same same control. the animal can one of two Traditional life history theory selective forces as conventional Traditional life history theory the animal can choose one of two selective forces as conventional does not allow for state variables. foraging life does not allow for state variables. foraging habitats. habitats, habitat habitat ii Iring being life history history traits. traits. characterized by two parameters: By using state variables in dynamic The concept of reprcxiuctive of reproductive By using state variables in dynamic characterized by two parameters: qj The concept value decision models, models, itit is is possible possible to to (average (average fecundity fecundity per per foraging foraging value isis central central to to life life history history thethe- decision achieve an extension and unificabout) and (predation risk per ory23. Reproductive value is an orbout) and pj (predation risk per ory2m3.Reproductive value is an or- achieve an extension and unification ganism's tion of of both life life history history theory theory and and foraging foraging bout). bout). ganism’s expected future future reprtxiucreproduck*havioral This Assume tive success (rx»ssibly discounted, behavioral ecology. This approach Assume that that @,<< & and and g, p, << gr p2, tive success (possibly discounted, i.e. habitat 2 is more is now increasingly used in in the case of a growing popuis now being increasingly used in i.e. habitat 2 is more productive in the case of a growing populationl the study study of of behavioral behavioral adaptadapt- and and more more risky risky than than habitat habitat l.I. lation) not not including including current current reprorepro- the Which is the optimal habitat? ations. Behavioral ecologists and duction. An optimal life history ations. Behavioral ecologists and Which is the optimal habitat? duction. An optimal life history To profile other evolutionary evolutionary biologists biologists nc»w now To Ergin begin with, with, consider consider the the optioptiprofile is is one one that that at at each each age age maxmax- other imizes need toto be as as familiar familiar with with the the mization mization problem problem inin the the final final imizes the the sum sum ofof current current reprorepro- need duction elements of of dynamic dynamic modeling as as v*riod period T. T. Here Here only only aa single single dededuction plus plus reprcxductive reproductive value. value. elements The The central central idea idea of of life life history history thethe- they they have have been been with with life life history history Cision cision isis involved. involved. Expected reproreproductive ory, ory, then, then, isis that that there there must must exist exist aa theory. theory. Fortunately, Fortunately, the the technique technique ductive Output output from from habitat habitat ii isis trade-off trade-off in in the the sense sense that that aa high high of of dynamic dynamic programming programming isis easy easy equal equal to to (I —g) - p> "and qjand the the optimal optimal level level of of current current repnxiuctive reproductive effort effort to to learns; learn6; graduate graduate students students of of bibi- decision decision maximizes maximizes this this quantity quantity. Let Ology results ology With with limited limited mathematical mathematical Let us us denote denote this this maximum maximum exexresults in in aa decrease decrease in in reproducreproducpected in background are now routinely tive value — the •cost of reproducbackground are now routinely pected reproduction in pericxå period rby T by tive value - the ‘cost of reproduclearning tion•. tion’. Optimization Optimization of of this this trade-off, trade-off, learning to to construct construct and and analyse analyse FIT): through through natural natural selection, selection, is is what what dynamic dynamic behavioral behavioral mcxiels. models. II will F(T)=~=~(I-Pj)@j determines (1) determines the the variety variety of of life life history history will first first explain explain the the logic logic of of dy namic programming, using two strategies observed in nature. namic programming, using two strategies observed in nature. simple simple models of of foraging foraging under under Since Since we we are are assuming assuming that that the the nunumerical values of the parameters risk of predation. II will then briefly Colin Clark is at Irstitute of risk of predation. will then briefly merical values of the parameters $,, Cohn Clark is at the Institute of Applied discuss Mathematics.University of discuss recent recent applications applications ofof the the pj are are given, given, Eqn Eqn I1 specifies specifies the the Mathematics, University of British British Colwmbia, Columbia, numerical value of FIT). This value methcxi to Vancouver. method to SEECific specific behavioral numerical value of F(T). This value Vancouver, K, BC, CanadaVET Canada V6T IZ2 122. 0 1993, Elsevier Science Publishers Ltd Ltd •;u'Kl (UK) Not Not many many years widely thought widely thought antipredator 205 reviews reviews TREEvol. TREE vol. e8, no. no. 6, 6, June June 1993 1993 represents represents expected expected reprcxduction reproduction at at the the start start of of period period T, T, assuming assuming that the optimal habitat that the optimal habitat is is choæn. chosen. Let Let the the optimal optimal habitat habitat in in period period TT T). be denoted denoted by by i"i*(T). Next, Next, what what is is the the optimal optimal habitat habitat for for earlier earlier time time periods t?t? For For t:t = T— T- II the analog to Eqn 1 isis FIT-I) F(T-1) =ya;(l-~J$j+F(T)l (2) Understanding Understanding Eqn Eqn 22 is is the the crux crux of dynamic programming. The of dynamic programming. The arguargument ment is is simplicity simplicity itself. itself. Consider Consider the the decision decision in in period period TT — - l.1. IfIf habitat habitat ability ability ii is is chosen, chosen, then then the the probprob- of of surviving surviving the the period period is is (l( 1 —g), - pi), and and total total expected expected reprorepro(conditional (conditional on on surviving surviving period T— T- l)1) isis then then (qi + exsrcted expected reproduction inin period T) = & ++ FIF(T). T). Expected total total reproduction reproduction at at the the start start of of period period T— T - II is is therefore therefore equal to '1 + equal to (1 - ,u.)[$~ + F(T)]. The The Optimal optimal habitat habitat ilT— /*(T - l)1) in in grricxi period T— T- I1 maximizes maximizes this this expression, expression, as as stated stated in in Eqn Eqn 2. 2. The The value value F( T - l)1) represents represents maximum maximum expected expected total total reproduction over the reproduction over the last last two two periods. Exactly Exactly the the same same argument argument shows shows that, that, for for any any tt << TT duction duction F(t)=m;(Wj)[4j ++F(t+l)l 13) (3) reservesat reserves at the the Erginning beginning of of period period t.t. AIM) let denote Also let R[X(T)] denote reproreproof ductive output output inin the the final final period of its its simplicity. simplicity, the the model model can can be be ductive T. We assume that R is an increasexpressed T. We assume that R is an increasexpressed in in the the standard standard life life ing history ing function, function, with with aa threshold threshold XXcrit history notation. notation. For For pedagogical such that purvx:ses I prefer not to do so purposes I prefer not to do so such that here.) here.) AA numerical ) R[X(T)] = 0 ifif X(T) <xe„t X,, numerical example example (see (see Box Box I1) illustrates the iterative solution of illustrates the iterative solution of Eqns The optimal optimal foraging foraging strategy strategy isis Eqns I1 and and 3. 3. For For tt near near the the terter- The that which maximizes mina! time T. the risky habitat that which maximizes expected expected minal time T, the risky habitat terminal 2) is optimal. but for ts T—5 reproduction. (i* = 2) is optimal, but for t I T - 5 terminal reproduction the Reproductive value value now now Ercomes becomes the safer safer habitat habitat (i• (i” == I)I) becomes becomes state dependent: optimal. The prediction is that optimal. The prediction is that state dependent: older older animals animals will will accept accept greater greater ftx.n F(x,O predation predation risks risks than than younger younger aniani—max EIRIX(T// x) (4) =maxE{RIX(T)]/X(t)=x) (4) mals. The explanation for this lies mals. The explanation for this lies Here E denotes mathematical ex. in value In subin reproductive value F(t). In sub- Here E denotes mathematical exjecting jecting itself itself to to predation predation risk, risk, the the grctation pectation (relative (relative to to the the random random animal tx)th its current and predation event) and animal places both its current and predation event) and the the prefix prefix future future reprcxduction reproduction at at risk. risk, while while •max' ‘max’ refers refers toto the the use use ofof the the optioptigaining gaining only only aa m.tential potential increase increase inin mal mal strategy. strategy. Equation Equation 44 leads leads imimcurrent The current reproduction. The larger larger mediately mediately toto the the dynamic dynamic programprogramming the the future future reproduction (reproduce (reproducming algorithm• algorithm: tive value). the more important it is tive value), the more important it is to to protect protect this this 'asset'. ‘asset’. Since Since reproreproF(x,T)= R(x) (5) ductive ductive value value isis aa decreasing decreasing funcfunction ftx,t) tion ofof time time tt Ifewer (fewer reproductive episodes episodes remain remain as as tt increases increases totowards wards T). T), older older animals animals can can afford afford greater risks. greater risks. Similar The The first first iteration iteration t= T— T- 1gives gives Similar effects effects ofof temporal temporal reducreductions tions in in reprcxfuctive reproductive value value have have F(x,T -1) been discussed, discussed, for for example. example, reregarding parental investment• garding parental investment9. =max(l-pi)R(x+g,) (7) the the standard standard life life history history optimizaoptimization principle. (In tion principle. (In fact. fact, because i=1,2 Equation Equation 33 is is the the dynamic dynamic propro- Gro.th Growth model still still lacks aa mcxiel. model. This This equation equation has has three three imim- state state variable. variable, and and may may be mismisgxytant portant features: features: first. first, itit isis clcxely closely leadingly leadingly simplistic simplistic. Let Let us us therethererelated related to to traditional traditional life life history history fore fore consider consider aa different different situation. situation, still involving foraging theory; second, it provides a practheory; second, it provides a prac- still involving foraging under under risk risk Of of predation. Assume now that tical algorithm for computing the tical algorithm for computing the predation. Assume now that the the optimal optimal decision decision ilo i*(t) for for all all times times forager forager reprcxiuces reproduces once once only, only, atat t,r; and and third. third, similar similar equations equations can can the the end end TT ofof the the foraging foraging season. output be be obtained, obtained, by by analogous analogous arguargu- Reprcxiuctive Reproductive output isis deterdetermined by somatic ments, for much more complex ments, much more complex be- mined by somatic reserves reserves atat time time TT. Somatic havioral havioral models. Somatic reserves reserves are are increased increased To To see see the the connection connection with with life as as aa result result of of foraging. foraging, and and are are also also history required for for metaix»lism metabolism throughthroughhistory theory, theory, observe observe that that at) F(t) required represents the the animal's animal’s total total exex- out out the the season. season. By By foraging foraging inin the the more productive habitat, the reprcxiuction from time more productive habitat, the ani• anipected reproduction from time period tt through through to to the the final final mal mal can can increase increase itsits rate rate Of of somatic period r,T, assuming assuming that that the the animal animal growth, growth, but but at at an an increased increased risk risk of of employs employs the the optimal optimal habitat habitat at at predation. predation. We We imagine imagine the the same same all all times. times. This This isis exactly exactly how how two two habitats habitats as as before. before, but but the the par• parare now Fisher Fisher defined defined reproductive reproductive value. value. ameters ameters are now Equation Equation 33 therefore therefore asserts that that g, gi = net net increase inin rærves reserves per per the the optimal optimal behavioral behavioral decision decision atat period time time tt isis that that which which maximizes maximizes the the risk predation risk per Pi = sum sum Of of expected current current reproreproperiod duction duction (l(1 — - pi)+i plus plus exsrcted expected Clearly future i. Clearly We we now now need need aa State state future reprcxiuction reproduction II(1 — - &)flt + I1). Again, Again, this this agrees agrees exactly exactly with with variable. variable. Let Let Xin X(t) denote denote somatic gram ming equation gramming equation for for the the present present 206 The The forqoing foregoing Here Here the the trade-off growth growth and and mortality mortality between between is is explicit. explicit. From From the the conditions conditions g,< g, and itit follows follows from from Eqn Eqn 77that that there there isisaa reserves reserves level level rXT-, >> with with the the propprop- erty erty that that .* I T-l 2ifx<x 2 if x<x* = lifx>x 1 if x T-l (8) > xi-, Animals Animals with with low low reserves reserves atat time time T— T- I1 should should employ employ the the risky. risky, prtxiuctive habitat, but productive habitat, but those those with with high high reserves reserves should should use use the the safer, habitat. safer, less-productive less-productive habitat. AA similar similar situation situation applies applies for for earlier earlier time time periods t.t. The The level level which which the the optimal optimal isis an an increasing increasing x’r atat habitat habitat switches switches function function of of time time t(Fig_ t (Fig. 1). I). In In this this model reprtx*uctive reproductive value value nx,0 flx,t) isis kx)th both time time and and state state dede- pendent. The The terms terms Of of the the trade-off growth and between growth and predation predation risk risk change change over over time, time, and and depend on on reviews reviews TREE vol. 8, no. 6, June '993 TREE vol. 8, no. 6, June 1993 the animal's current state. The the animal’s current state. The same asset-protection principle same asset-protection principle used to explain the predictions of used to explain the predictions of also applies also applies for constant for constant in t (because x, F(x,t) is decreasing in t (because older animals have less time left in older animals have less time left in which to grow); as the asset value which to grow); as the asset value Flx,n decreases, optimal behavior flx,t) decreases, optimal behavior favors more risky choices. Similarly favors more risky choices. Similarly is an increasing function of x, f+f, t) is an increasing function of X, for fixed t, implying that larger for fixed t, implying that larger animals should adopt less risky animals should adopt less risky choices. These two considerations choices. These two considerations showthat the switchingline x; has show that the switching line xi has gxysitive slogr (Fig. l). positive slope (Fig. I). the previous model the previous model here. For example, here. For example, x, is decreasing Appncatims Applications The simple The simple described theoretical theoretical models models were to acquaint described above were to acquaintof reader with the mechanics the reader with the mechanics of state-variable modeling of behavior state-variable modeling of behavior and life history. In recent years and life history. In recent years this technique has been used in this technique has been used in many empirical studies. Examples many empirical studies. Examples include include: the • oviposition behavior of the para- l oviposition behavior of the parasitic wasp Nasonia vitripennis1° • allocation in breeding tree l food allocation in breeding tree swallows (Tachycineta bicolon (D. swallows (Tachycineta bicolorj (D. Winkler and F. Adler, unpublished) Winkler and F. Adler, unpublished) • caching in Carolina chickl food caching in Carolina chickadees (Parus carolinensis) I adees (Parus carolinensis) ’ ’ • diel vertical migration of juvenile l die) vertical migration of juvenile sockeye salmon (Oncorhynchus sockeye salmon ( Oncorhynchus nerka) 12 sitic wasp Nasonia vitripennis10 eling * 300 t greatly extends our eling approach greatly extends our ability ability to to understand understand evolutionary evolutionary which G.C. Williams adaptation, which G.C. Williams has called •a phenomenon of has called ‘a phenomenon of pervasive vasive importance importance in in biology•20_ biology’20. Xcrit Example: termral decHneIn Example: temporal decline in the mass mass at fledging of seabirds at fledging of seabirds A common phenomenon in the 0 A common phenomenon in the life history of birds Erlife history of birds beto the family Alcidæ longing to the family Alcidae (murres, puffins, auks, etc.) is a de(murres, puffins, auks, etc.) is a decline, cline, as as the the breeding breeding season season proprogresses, in the average mass at gresses, in the average mass at which nestlings fledge. Considerwhich nestlings fledge. Considerable variation cxcurs within able variation occurs within species, in the age and mass species, in the age and mass at at fledging, fledging, and and even even greater greater variavariation different tion occurs between different sgkcies14, but the pattern species r4, but the pattern of of temtemporal decline of average mass at poral decline of average mass at fledging fledging is is widespread within within many many populations. A standard explanation attributes A standard explanation attributes this this decline decline in in fledging fledging mass mass to to aa corresgXJnding decline in the availcorresponding decline in the availability ability of of food, food, but but little little ifif any any evidence suEmrts this claim. An evidence supports this claim. An alternative explanation 14 hygfthalternative explanationI hypothesizes that the phenomenon conesizes that the phenomenon constitutes a to stitutes a behavioral adaptation to iuvenile juvenile longing For the I 20 40 60 Time, t 60 Time, t Fig. Optimal habitat choice for a Fig. 1. Optimal habitat choice for a model of foraging risk Ofpredation. The Shadedarea is a region Of under risk of predation. The shaded area is a region of zero fitness: fora«rs whose reserves x at time r lie in zero fitness: foragers whose reserves x at time t lie in this this regionare region are unableto unable to growlargeenotÆhto grow large enough to repmreproduce duce. Otherwise, foragers ShoulduseHa should use H2 (risky habitat' habitat) H, Isafer habitat' if their x at time t lie in or H, (safer habitat) if their reserves x at time t lie in indicated region. Typical growth Curvesare Shown the indicated region. Typical growth curves are shown lines. parameters; g, as dotted lines. IModel (Model parameters: p, = 0.02, 0.02, & = 0.04, g, = 2, g2 = 5, X&, = 100) the the trade-off Ertween between growth rates and mortality relative relative risks in growth rates and mortality risks in the the nest nest and and at at sea. sea. Typically Typically the the nest is a safer environment for nest is a safer environment for young birds, but growth rates are young birds, but growth rates are higher at sea. higher at sea. A A simple simple dynamic dynamic programming programming (Box 2) was developed to model (Box 2) was developed to test whether this hw)thesis test whether this hypothesis was was reasonable. The model is similar to reasonable. The model is similar to the the growth growth model of of Eqns Eqns 4-7 above, except that above, except that the the nestling nestling growth growth rate rate isis assumed assumed to to be on on Of nerka) ’ 2 • territorial bequeathal in Ameril territorial bequeathal in American red squirrels (Tamiasciurus can red squirrels ( Tamiasciurus h udsonicus) hudsonicus) I3 • fledging behavior of common l fledging behavior of common murres aalge)14 murres (Uria ( CJriaaalge)14 •0 parent—offspring conflict and parent-offspring conflict and nestling mass decline in dovekies nestling mass decline in dovekies (Alle alle'15 (Ale alle) I5 • anaerobic diving l anaerobic diving of of Westem Western grebes (Aechmorphus cxcidengrebes (Aechmorphus occidentalis 16 talis) ’ 6 • host seeking in mc%quitoeS17 l host seeking in mosquitoesI • vegetative versus reprcxiuctive l vegetative versus reproductive growth in plants' 819 growth in plants’a,‘9 and flexibility of the The generality method and flexibility of the are attested state-variable method are attested by the variety of species and types by the variety of species and types of behavior encompassed by these of behavior encompassed by these publications. These studies have publications. These studies have led to new insights and hypotheses led to new insights and hypotheses regarding the adaptive significance regarding the adaptive significance of many Erhavioral and life history of many behavioral and life history traits. Indeed, the distinction traits. Indeed, the distinction between life history and kkhavioral tween life history and behavioral decisions becomes largely blurred decisions becomes largely blurredIn in the state-variable framework. in the state-variable framework. In my opinion, the state-variable my opinion, the state-variable mod- A1-gag 2 The generality state-variable 2.7 2 11.0 025, 207 s review TREE vol. 6, 8, no. no. 6, 6. June June 1993 TREE vol. 1993 current with current body mass: mass: g, g, = gl(x), g,(x), with g,(x) declining at high (parents g,(x) declining at high x (parents are are unable unable to to provision provision nestlings nestlings at at a rate sufficient to overcome ina rate sufficient to overcome increasing costs). creasing metabolic metabolic costs). Also, Also, only habitat only a a single single ontogenetic ontogenetic habitat switch (fledging) switch (fledging) is is allowed. allowed. The rncxiel indeed The model indeed predicts predicts aa ternEX)ral decline in fledging mass", temporal decline in fledging massr4, a result that can explained a result that can be explained in in terms terms of of the the asset-protection asset-protection prinprinciple referred referred to light ciple to earlier. earlier. First, First, light birds should should remain remain in birds in the the nest, nest, where growth growth rates where rates are are higher, higher, and and mortality risks lower, at sea. mortality lower, than than at sea. As nestlings As nestlings grow grow larger, larger, however, however, their growth rate in their growth rate in the the nest nest dedecreases and and eventually creases eventually falls falls below the potential at sea. the potential growth growth rate rate at sea. The trade-off and The trade-off between between safety safety and low growth low growth rate rate in in the the nest nest then then becomes crucial. becomes crucial. given A nestling that that grows grows to to aa given mass x earlier has has greater greater reproreproductive value value than ductive than another another nestling nestling that reaches reaches xx later later on, that on, because because the former former nestling nestling has the has greater greater for growth to the potential for growth prior prior to the end of the the season. Having a greater end season. Having greater reproductive asset to to protect, protect, the reproductive early grower grower will be more more careful predation risk, and about predation and will thus leave the nest nest than be less apt to leave than later-growing nestling. the later-growing nestling. Given that the the later bird fledges that fledges on a certain date date at at a a certain mass x, tain certain mass x, the the bird that reaches x at an earlier bird that reaches x at an earlier date will Will benefit benefit by remaining remaining in date has grown larger the nest until it has grown larger than x. In other Other words, nestlings nestlings than that fledge fledge early will Will do so at at larger larger that mass levels levels than mass than later later fledgers. fledgers. Among different different nestlings nestlings in Among in aa given colony, colony, variation variation in mass mass may may given result result from from variation variation 'Of not particular not very very sensitive sensitive to to a a particular that variabehavioral decision, decision, so so that variain behavior would in behavior would be selecselectively neutral. tively neutral. The The evolutionary evolutionary implications implications of of are space. In practice, stochastic environments environments are exexspace. In practice, the the data data show show stochastic considerable a tremely subtle, subtle, however however. ReproReproconsiderable spread around around a tremely ductive fitness, is mean, negatively fledging mean, negatively sloped, fledging ductive value, value, or or fitness, is usually usually defined boundary. AA more more complex complex model defined in in terms terms of of expected expected (aver(average) success, age and maturation incorporating age and maturation age) reproductive success, but but this this is in the effects account effects could could perhaps account is known known to to be be incorrect incorrect in the case case for some environments, for some of of this this variation, variation, but but of of •coarse-grained' ‘coarse-grained’ environments, in such such aa model has has not not yet yet been in which which the the timescale timescale of of fluctufluctuations is comparable to generation developed. ations is comparable to generation An extension of fledging times21•22.Although An extension of the the fledging times2im22. Although state-variable state-variable rncxiel has used to study the readily include model has been used to study the models readily include random random enenphenomenon of nestling-mass devironmental variations, phenomenon of nestling-mass de- vironmental variations, they they still still cline in rely cline in seabirds". seabirds15. This This extended extended rely on on expected reproduction as as conmodel treats treats parent—offspring parent-offspring con- the the criterion criterion Of of fitrBs. fitness. These These terms of dynamic evolutiontherefore appropriate flict in terms of dynamic evolutionmodels are are therefore appropriate to fluctuations, i.e. ary games ary games. to fine-grained fine-grained fluctuations, i.e. fluctuations that take place on fluctuations that take place on aa that is Advantages and and nmIUti0% limit&Ions timescale that is short short relative relative to to The The state-variable state-variable approach to to an an individual's individual’s life span. span, How How to to modeling life history, with modeling tRhavior behavior and and life history, modify the the approach approach to to deal deal with like has coarse-grained like any any modeling modeling framework, framework, has coarse-grained environments environments is is not not both advantages and limitations. yet clear. Indeed, evolution cannot both advantages and limitations. yet clear. Indeed, evolution cannot Its most important fully maximizing Its most advantages are are be a a fully maximizing process under these the followings: the following? under these circumstances' circumstances7f2’. Nature is complex. Nature complex. ItIt is therefore therefore •l Model parameters have direct parameters have sometimes mistakenly thought that sometimes mistakenly thought that biological meaning, meaning, and and can often often of nature must be equally models of nature must be equally be measured measured experimentally. complex. knew betcomplex. Einstein, who who knew •l Model predictions quantitatpredictions are are quantitatter, said that theories should be as ter, said that theories should be as ive ive and and subje•ct subject to to experimental simple as but not simple as possible, but not too testing. However, qualitative pretesting. However, qualitative pre- simple. What is meant by simple' simple. What is meant by ‘simple’ dictions dictions and and general general principles principles can can depends in part on in part on current current techalso be useful also be useful in in broadening Our our depends nology_ Life history theory theory was denology. understanding of adaptation. understanding adaptation. veloped veloped long long before before the the advent advent of of •l Multiple choices, and Multiple behavioral choices, and automatic computation. The automatic computation. modthe associated ass.xiated trade-offs, can the trade-offs, can be em extension to em extension to state-variable state-variable studied in a single studied single model. is a consequence of the models consequence •l Constraints Constraints on on behavior, behavior, and and on on computer revolution. (Dynamic computer revolution. (Dynamic state variables, are a natural comstate variables, are a natural com- programming programming is itself a computercomputerponent Of of the the model. intensive technique.) But it is well intensive technique.) But well •l The implications Of The implications of environmenenvironmenknown that dynamic optimization known that dynamic optimization tal fluctuations tal fluctuations (both deterministic deterministic models with with more more than than aa few few state state and stochastic) can be analysed analysed. variables variables quickly quickly encounter encounter the the State-variable mcxiels realistic 'curse State-variable models with with realistic ‘curse of dimensionality•«, dimensionality’4*6, which stcxhastic not precan overwhelm latest cornput• stochastic components do not overwhelm the latest computdict a unique behavioral repering capacity. A similar toire, but rather predict behavioral similar problem arises arises in senvariation within a given population, sitivity testing, which can become variation within population, testing, which become resulting from environmentally indaunting resulting from environmentally daunting for a complex complex model induced duced variation variation in in individual individual states. states. volving numerous numerous parameters. parameters. But From this point view, at least as I hope point of view, hope to have have demonstrated, demonstrated, part Of inevitably endynamic of the variability inevitably dynamic models Of of behavior behavior need countered in field studies can be not be highly complex in order to seen to have signifimake useful, predictions. seen to have adaptive adaptive make useful, testable testable predictions. cance, rather than simply being Simple cance, rather than simply being Simple dynamic dynamic models often gendisregarded disregarded as meaningless meaningless statis- erate erate predictions predictions that that differ radtical noise in observations. On ically from tical noise in the the observations. On ically from equilibrium equilibrium models. In In the other Other hand, model may may Other hand, a model other cases no nondynamic nondynamic analog analog sometimes indicate is may be possible at sometimes indicate that that fitness fitness is at all. e ab ?oiidr uutnx:r in the in in hatch hatch date, date, or or in in nest nest growth growth rate. rate. Our Our simple simple model predicts that all nestlings in model predicts that all nestlings in a given colony will fledge along a given colony will fledge along a a single single 'fledging ‘fledging boundary boundary’ in in t—x t-x tions tions ing TREE TREE vol. 8.8,no. no. 6,6,June June reviews reviews 1993 The The models described inin this this productively, about the the meaning meaning Of of brief brief survey survey have have all all concemed concerned inin- adaptation adaptation. dividual dividual optimization. optimization, and and itit might might thought that the technique be thought that the technique isis References limited limited to to this this case. case. Fortunately. Fortunately, I I Leswls. Lessels, CM. C.M. (1991) in Behavioral this this isis not not so; so; aa variety variety ofof dynamic dynamic Ecology: An Evolutionary Approach game-theoretic, or game-theoretic, or ES ESS, models (Krebs, 1.R. and Davies, N.B., eds), pp. 32-68, Blackwell have have tren been developed15f24. Dynamic Dynamic 2 Fisher. RA R.A. ( 1930) The Genetic& Theory ESSmodels ESS models can can be be extremely extremely comcom- 2of Fisher, Natural Selection, Clarendon plex unless ingenuity is applied n plex unless ingenuity is applied n 3 Steams Steams, sc S.C. (I 992) The The Evolutionon.ife Evolution of Life the the choice choice ofof simplifying simplifying assumpassumpHistories, Oxford University University Press 44 eellman. tions. Bellman, RR. (1957) Dynamic Programming, tions. For For example. example, inin aa two-particitwo-participant pant game game between dominant dominant and and Princeton University Press subordinate, the the priority priority Of of decision cision can can be be asumed asumed toto trlong belong toto the the dominant dominant15. In In aa sense. sense, the the limitations limitations ofof dp dy namic may namic modeling may be be aa blessing blessing in in disguise, disguise, ifif they they cause cause scientists scientists to to think think profoundly profoundly atX)ut about their their syssysterns, tems, rather rather than than relying relying on on incomincomprehensible prehensible computer computer simulations simulations. I 1 believe that that state-variable state-variable modeling isis helping helping evolutionary evolutionary biolobioloeling gists gists to to think think differently. differently, and and more more Virugs Viruses have have DD. and 5 McFarland. McFarland, and Houston. Houston, KI. A.1. ( 198 I ) Quantitative Ethology: The State Space Approach, Pitman 6 Man«l. Mangel, M. andCIaå.C.W. and Clark, C.W. '1988' ( 1988) Dynamic Modeling in Behavioral Ecology, Princeton Press Princeton University University Press 7 Lewontin, R.C. (1987) in The Latest on the Best: Essays on the Evolution of Optimality (Dupti, J., ed.), pp. 151-150. 151-159, MIT MIT Press Press 'H, 8 McNamara. McNamara, J.M. and and Houston, A.I. A.I. (1986) Am Am. Nat Nat. 127.358-378 127,358-378 9 Trivers, R.L. (1974) Am. Zoo/. 14, 249-264 10 10 Mangel, M. (I 987) /. Math. Biol 25, l-22 1-22 6een assumed assumed toto play aa uptake2 and and methcxis methods for for measuremeasureuptake2 ment of grazing3, were developed. ment of grazing'. were In this this In perspective, recent work work recent showing that that viruses viruses may may be ix»th both showing research inin marine marine microbial microbial ecolecolresearch active inin cell cell lysis'. lysis5, represents represents active discovery of yet another piece discovery of yet another piece F. Thingstad, M.Heu Heldal, G. G. Bratbak and I. Dundas are at the Dept of Microbiology and Plant Physiology, University of Bergen, jahnebakken 5, N-5020 Bergen, l_meßity lahneba"en 5. Norway. F, Q 1993, Elsevier Science Publishers C seiÄ Ltd (UK) 271-290 131,271-290 13 Price, Price, K.K. (1992) Bull. Math. Biol. 54, 335-354 14 Ydent*rg.R.C. Ydenberg, R.C. (1989) Ecology70,70. 1494-1506 15 Clan. Clark, C.W..ndYdenkx•rg. C.W. and Ydenberg, R.C. (1990) Evol. Ecol. 4, 3 12-325 16 R.C, 16 Ydenberg, R.C. and and Clark. Clark, C.W. C.W. (1989) I. Theor Theor. Biol Bioi. 139,437-449 17 17 Roitberg, B.D. and and Friend, Friend, W.G. (1991) Bull. Math Math. Biol. 54,40 I-4 I 2 l.I.and I8 Kozlowski, Kozlowski, and Wiegert, Wiegert, R.G. R.G. 11987) (1987) EVO/ Evol. Ecol. I 1,23 l-244 l.1.and 19 Kozlowski. Kozlowski, and Ziölko. Ziblko, M. M. (19881 Theor. pop. fop. Biol. 34, 34, II&l29 20 and 20 Williams. Williams, G.C G.C. (1966) Adaptation Adaptation and Natural *lection, Princeton Naturalselection, Princeton University University Press 2 1 Levins, RR. (I 1968) Evolution Evolution inin Changing Changing Environments, Princeton University University Press Environments. Princeton Press 22 Yoshimura, J. and Clark, C.W. ( 199 I ) Evat Evol. Ecoi. 5,5, 173-192 23 23 Yoshimura. Yoshimura, ).and and Clark. Clark, C.W C.W., eds Adaptation Stochastic Environments, Adaptation inin Sttxhastic Environments, Springer un (in press press) I Sprin«r 24 24 Houston. Houston, A1. A.I. and and McNamara. McNamara, 1.M. (1987) j. Theor. Biol. 129:37-68 129: 57-68 1. T.F. hingstad, M. T.F.TThingstad, M.Heldal,G. Heldal,G.Bratbak Bratbakand andl.I.Dundas Dundas one uses uses the the allegory allegory ofof aa iigjigIfIf one saw puzzle, the last two decades saw puzzle, the last two ofof ogy have have been been charKterized characterized by by atattempts toto fitfit together together emerging emerging tempts pieces ofof knowledge knowledge toto form form aa pie picpieces ture of how elements are cycled in ture of how elements are cycled in the microbial food web: who are the microbial web: who are the trophic trophic partners partners inin the the food the web, how how active active are are they they and and how how web. this ecosystem ecosystem controlled controlled by by exexisis this ternal and and by by intemal internal fm-tors? factors? The The ternal very existence existence ofof many many ofof its its very pieces remained unknown until pieces remained unknown until new techniques techniques such such as as fluoresfluoresnew cence microscopy’, , measurement measurement cence of DNA DNA synthesis synthesis by by thymidine thymidine Of Nar. Nat. AreViruses annersin Are VirusesImportant ImportantP Partners in PelagicFood Pelagic F Webs? rather negligiblerok negligible roleas aspartnersin partners in micromicrorather 6iaI foodweb we6dynamics. dynamics. However, recent bial discoveries suggest that tke the rate rateofofvirally virally discoveries suggestthat induced lgsis of marine marine mkrobia/ microbial popuinducedlysis lations may be 6esignifiant. significant. 7%is, Tclis,inin turn. turn, lationsmay may have important consequences for the may haveimportant'onsequen«s tke developing conceptual frameworkof framework of tke the devebpingconceptual microbial foodVeb we6. il1 I Lueas, Lucas, J.R. and and waiter. Walter, L.R. L.R. ( I99 1) Anim. Behav Behav. 41.579-601 4 1,579-60 I 12 12 Clark. Clark, C.W C.W. and Levy, D.A. (1988) Am. numerous numerous the water water mass', mass4, and and inin the has toto be be fitted fitted has the the that that somewhere inin inin somewhere the very center of this puzzle. InIn the very center of this puzzle. addition to to •classical' ‘classical’ questions questions rereaddition lated toto element element cycling cycling and and food lated web dynamics. dynamics, the the virus virus story story adds adds web new aspects to the puzzle, such as new to the puzzle, such as the potential exchange of genetic the EX'tential exchange Of genetic information between microbial information microbial species and and the the potential impact impact ofof species viruses on the diversity of the mimiviruses on the diversity of the crobial population. With this develcrobial 'xvulation. With this development, •phage ‘phage ecology ecology’ isischanging changing opment. from aa field field mainly mainly concerned concerned With with from the sanitary sanitary aspects Of of survival survival Of of the enteric viruses in natural environenteric viruses in natural environments? toto aa discipline discipline inin the the mainmainments6 stream Ofof research research inin the the field field of general general Of microbial ecology ecology Of of microbial aquatic ecosystems. aquatic ecosystems _ The existence existence Of of Viral viral activity activity and and phage-host systems inin aquatic aquatic phage—host systems ecosystems where where the the hosts hosts are are ininecosystems digenous has has been been known known for for more more digenous than 30 30 years', years7. Until Until recently, recently, how howthan ever, attention attention toto the the role role ofof virusvirusever, es inin the the marine marine environment environment was was es hampered by by aageneral general acceptance acceptance hampered of arguments arguments for for aa low low probability probability Of of infection. infection. Viruses Viruses were were known known Of from laboratory studies to be host from latX)ratory studies to host specific, which, which, combined combined with with relarelatively sparse sparse populations populations Of of hosts hosts tively (c.10” ml-’ for for bacteria; bacteria; 103—104 103-IO4ml-l ml-’ (c. 100 ml-' for small eukaryotes) and the asfor small eukaryotes) and the as. sumption of generally very diverse sumption of generally very diverse communities, seemed seemed toto make make the the communities, probability ofof virus—host virus-host encounter encounter probability very small. small. Thus, Thus, there there was was aa relarelavery tive lack of interest in early work tive lack Of interest in early work suggesting a high number of free suggesting a high number of free virusesa. viruses'. 209