PERSPECTIVES

with Evans’ suggestions. However, these issues have been considered in detail by Interpreting phenotypicvariation in many researchers examining re- productive effortl2,‘” and morphological patterns in animalsl4. Misinterpretations James S. Coleman vidual plants on which these conclusions of the functional role of phenotypic vari- Kelly D.M. McConnaughay are based have generally been conducted ation in physiological and vegetative at common points in time or at common characteristics of plants can arise when David D. Ackerly plant ages. the relationships between phenotypic ex- pression, and growth Plant ecologists and evolutionary Assessing phenotypic variation: rate are not considered. Box 1 shows biologists frequently examine patterns of what we often miss four hypothetical scenarios that illustrate phenotypic variation across variable en- G.C. Evans7 argued over 20 years ago possible relationships among develop- vironments or genetic identities. that comparing plant traits as a function mental patterns and growth rates, and Too often, we ignore the fact that most of plant age may lead to serious com- the potential for contrasting interprct- phenotypic traits change throughout plications in interpreting data. He noted ations of experimental results. Through- growth and development of individual that many phenotypic traits of plants out, ‘treatment’ differences among plants plants, and that rates of growth and change dramatically over the course of refer to environmental, genetic or inter- development are highly variable. Plants plant growth and development - a specific causes of variation. growing in different environments are phenomenon he referred to as ‘ontogen- In the first example in Box 1, the pat- likely to grow at different rates, and will etic drift’. For example, total in tern of change in phenotypic expression be of different sizes and stages of different plant parts, such as , stem throughout growth and development is development at a particular age. When and , clearly increases in concert fixed, but since growth and developmental we compare plants as a function of plant with overall growth; additionally, the pro rates vary, phenotypic expression varies size or developmental stage, as well as a portional distribution of biomass among at any point in time. Thus, phenotypes function of age, we broaden our these parts is rarely constant for extended differ when compared at a common time, understanding of phenotypic variation periods7. In some herbaceous plants, the but are identical when compared at a between plants. ratio of to shoot biomass is initially common size or developmental stage. very high because of early root growth This example has tremendous bear- and establishment in the soil, but then ing on studies testing optimal partition- James Coleman is at the Dept of Biology, Syracuse University, Syracuse, NY 13244, USA; drops rapidly over the course of the first ing models3 of plant growth. A great deal Kelly McConnaughay is at the Dept of few weeks of growths. Physiological traits of support for these models has been Biology, Bradley University, Peoria, can also differ markedly between indi- garnered from studies in which the pro- IL 61625, USA; viduals of different sizes, or at different portional allocation of biomass to leaves, David Ackerly is at the Dept of Organismic stages of ontogenysJO. shoots and/or roots was compared as and Evolutionary Biology,Harvard The absence of appreciable change a function of plant age]“. For example, University, Cambridge,MA 02138, USA. during ontogeny is probably the excep- shade-grown plants often appear to invest tion rather than the rule for traits related a greater proportion of their resources to resource acquisition, allocation and into area than do those grown in nvironmentally-induced phenotypic partitioningrJ1. Furthermore, plants grow- high-light (e.g. have a higher leaf area Evariation () in ing in different environments often grow ratiorJ6JT). However, allocation patterns plants is often considered to be a func- at different rates, and thus will be of dif- change throughout ontogeny, and shade- tional response that maximizes fitness in ferent sizes and developmental stages at grown plants usually have slower growth variable environments. The assessment a given age. Thus, plants of a particular rates than those in high-light. Conse- of variability in plant morphology (e.g. age can frequently be ontogenetically dis- quently, observed differences in leaf area biomass partitioning) and physiology similar. The interpretation of variation in ratio between low- and high-light plants (e.g. nutrient use efficiency) is critical many phenotypic traits will, therefore, de- are confounded with differences in growth in testing ecological and evolutionary pend on whether comparisons are made rates’. Is increased leaf area ratio in low- models that make predictions about the as a function of plant age, size or devel- light plants the result of functional adjust- functional or adaptive nature of pheno- opmental stage (Box 1). For example, ments as predicted by optimal partitioning typic plasticity (e.g. plant strategy theory’, Evans strongly suggested that environ- models, or rather a consequence of slow the resource-ratio hypothesis*; optimal mentally-induced changes in traits rep- growth rates and delayed development? partitioning models”; and evolutionary resenting any aspect of plant biomass If increases in the relative production of interpretations of phenotypic plasticityb). partitioning need to be examined as a leaf area in low-light represents func- For example, differences in the partition- function of plant size in order to draw tional adjustments in allocation, as opti- ing of biomass to leaves between plants any conclusions as to the functional sig- mal partitioning models predict, then a of the same species grown in high versus nificance of the phenotypic variation. plant grown under low light should have low light has been cited as evidence for a greater leaf area ratio than a high-light optimal partitionings. Functional adjustment or ontogenetic drift? plant when both plants are compared iit Differences among species in the Although Evans warned that func- the same size’. amount of morphological plasticity tional interpretations of phenotypic vari- When leaf area ratio has been com- exhibited in response to varying resource ation due to genotype or environment pared among same-sized plants, the dif- availability have been used to support can be compromised when comparisons ferences in allocation patterns due to the notion that plasticity may be an adap are made at a single common age, very light environment often disappear, even tation to patterns of resource availability few of the data used to support models of when the leaf area ratio varies as pre- in the habitat in which a particular species plant strategy or optimal partitioning have dicted by optimal partitioning models evolved”. The comparisons among indi- been interpreted in a manner congruent when comparisons are made at a common Box 1. Same-age versus same-size (stage) comparisons of plant phenotypes

Variable growth rate Same age Same size comparison (stage) comparison In this case, growth rates and phenotypic expression throughout growth and development are vari- able, but the pattern of change in phenotypic expression throughout growth and development is fixed. Variable phenotypic expression throughout growth and Phenotypes differ when compared at a common time, but are identical when compared at a common development size or developmental stage. This situation is fairly common (Refs 17-25). i.// I/ Fixed pattern of change in phenotypic expression throughout L I*Plant age Plant size growth and development

Variable growth rate

Variable phenotypic expression z In these cases, growth rates, phenotypic expression throughout growth and development, and the throughout growth and e 2 pattern of change in phenotypic expression throughout growth and development, are all variable. development a Variability at both levels can result in any of three distinct scenarios. In the first scenario, pheno Variable pattern of change in types that do not differ at a common age may be found to differ as a function of plant size or de- phenotypic exprass~on throughout g velopmental stage7.20. In the second case, comparisons at a common time reveal phenotypic dif- growth and development e B ferences, but comparisons at a common size or developmental stage reveal that variation in the 2 a trait actually occurs in the opposite direction than would be inferred by comparisons at a common time7. In the third scenario, qualitative interpretations do not vary as a function of the axis used for comparison, but quantitative interpretations of phenotypic expression can be markedly different for comparisons at a common plant age versus comparisons at 8common plant size or developmental stagem3.2’,28

Whenever plant growth rates are variable (as a function of environmental conditions or genetic identity) and phenotypic expression changes throughout growth and development, comparisons of phenotypes at a common plant age may differ from comparisons of phenotypes as a function of plant size or developmental stage. Each graph above compares the effects of two environments or two genotypes on the expression of a phenotypic trait. The first graph in each pair represents corn mon age comparisons, and the second graph represents common size (or developmental stage) comparisons; the treatment or genotype represented by the solid line grows faster than that represented by the dashed line. Comparisons at a common age will be identical to those at a common size (stage) whenever growth rates and/or phenotypic expression throughout development are constant. Since these conditions are rarely met, it is critical to consider which axis is more appropriate for a given question (see Box 2).

plant ageIT-2”.Furthermore, other com- Abutilon theophrasti, did not vary as a lar age, but the difference between treat- monly measured traits such as root:shoot function of differing light treatments for ments is greatly reduced when they are ratio (Refs 21-24) and %N (an estimate plants of the same age. However, they compared at a common plant size. For of nitrogen use efficiency25) also exhibit found that low-light plants had consider- example, Poorter and Pothmannz7 showed marked size-dependent changes during ably greater height and leaf number, rela- that the overall qualitative interpretations growth. This does not mean that all tive to plant weight, than high-light plants. of interspecific differences in a number of ‘treatment’-induced differences in a trait Thus, they point out that if they had not traits such as specific leaf area or specific that may be revealed by comparisons at compared at common plant weights, they growth rate were not affected by the a common plant age are an illusion or might have incorrectly concluded that choice of plant age or plant size as the irrelevant. For example, differences in light treatments did not affect the allo- basis for comparison. However, the quan- leaf area ratios at a given plant age may cation of resources to stem elongation titative magnitude of differences among determine the outcome of competitive and leaf production, whereas the com- species did change depending upon the interactions among plant+, but these parison at a common weight revealed a choice of comparison. This result has differences may not result from ‘treat- very clear effect. Similar cases have been been echoed in other studies on environ- merit’-induced changes in the patterns reported by Evans7 with regard to stem mental effects on specific leaf areala, leaf of leaf area production. Rather, they can weight ratio and total leaf weight for weight allocationl7, and on the production arise from developmental or allometric Impatiens parviflora grown in differing of leaf area (Ref. 28; in this study, a relationships that result in plants of differ- light regimes. developmental index was substituted for ent sizes having different leaf area ratios. In the second example, an even more biomass). Thus, quantitative interpret- disturbing scenario is depicted. Here, ations of results will be affected by What if there are changes in both growth the patterns of variation in growth rate whether comparisons are made on the rate and the pattern of change in pheno- and in the size-dependence of a trait basis of age or size even though qualitat- typic expression during development? result in qualitatively different patterns ive ones will not. In fact, it is only when The last three possible scenarios in for comparisons at a common age versus a trait is known to show no ontogenetic Box 1 depict situations where treatments a common size. Thus, a trait exhibits a drift or when growth and developmental alter both growth rates and the pattern greater value in one treatment, when rates are invariant that one may be cer- of change of the phenotypic trait through- compared at a common size, but a lower tain that the interpretation of a ‘treat- out growth and development. In the first value when compared at any point in merit’-induced change in a trait will not of these, shifts in the relationship be- time. For example, Evans7 showed that depend on whether that trait was com- tween a particular trait and plant size changes in leaf weight as a function of pared as a function of age or size. are masked by compensatory changes certain light environments in Impatiens in growth rate such that the trait does parviflora often followed this pattern. Experimental design and data not vary among plants of the same age. In the last example, comparisons of a analysis However, ‘treatment’-induced differences trait at a common age or size will pro- Allocate replicates over time in the trait become evident as a function vide similar qualitative results, but will Currently, most studies of phenotypic of plant size. For example, Rice and differ in the magnitude of response. Thus, variation are designed to compare pheno- Bazzaz?” showed that the traits of height plants in a slow-growing treatment may types at a single or few point(s) in time, and leaf number of an annual plant, have a lower value of a trait at a particu- usually including a large end-of-season

188 TREE “01. 9, no. 5 May lY.94 PERSPECTIVES harvest. Incorporating variation in plant growth and development in such studies Box 2. Some alternative standards for comparing phenotypic traits will entail a shift in experimental design. Single phenotypic traits To examine explicitly the pattern of Individual phenotypic traits may vary In relation to the age. size or developmental stage of a plant. change in phenotypic expression over As a result, the interpretation of environmental or genetic variation may differ depending upon whether growth and development, we must collect comparisons are conducted among individuals of the same age, srze or stage. The choice of one of these data more often, perhaps using fewer standards depends critically upon the question addressed in the study, and the three are brrefly compared below: replicates. For example, if a researcher has enough material for 20 replicates of Standard Uses Chronological Comparisons at a common point in time are important in relation to ‘real each treatment, one must decide whether time’ processes, such as plant-plant and plant-animal interachons, and for to harvest all 20 replicates at the end of the analysis of seasonal patterns, such as reproductrve output in relation to the season, to harvest one plant every the length of the growing season. Examples include leaf mtrogen levels In week for 20 weeks, or some compromise herbivory studies, flowering loads and pollinator actrvity. and total fitness in between the two. More frequent harvests annual plants susceptible to end-of-season frosts. facilitate the comparison of treatments Growth For traits that exhibit size-dependent changes during growth, especially those throughout growth and development as directly related to biomass accumulation, comparisons at a common size (e.g. well as through time (i.e. throughout the total plant biomass) are important. Examples include proportional biomass season), thus allowing one to distinguish allocation (see also below), whole plant nitrogen uptake and leaf area production (see Fig. lc). Size-dependent comparisons are particularly relevant phenotypic plasticity as a result of plas- for tests of economic models of optimal biomass allocatron. ticity in growth rate versus plasticity in how the phenotypic trait changes through- Developmental Traits that vary during development, particularly if they are related to nodal out plant developmentz9. development or plant developmental stage, may be compared at a common stage (e.g. leaf plastochron number, rosette versus bolting stage) to separate the effects of the rate and ontogenetic pattern of development on Choosing a standard of comparison phenotypic expression. Examples include reproductrve output in relatron to As in any research, the choice of life history stage and aspects of leaf development methodology reflects the goals of a study (see Box 2). Comparisons at a common Two (or more) functionally related phenotypic traits Many analyses of phenotypic responses to environmental effects involve examrning the relative changes age are appropriate for real time pro- in two (or more) phenotypic characters, such as root:shoot and reproductrve:vegetative biomass ratios, cesses such as herbivory, competition, or height/diameter or leaf nitrogen content/ relationships. Interpretations of functronal community structure and natural selec- relationships between two or more traits depend critically on the amount of ontogenetic drift exhibited by tion. For example, the outcome of the each individual trait (see Fig. Id). Explicit analysis of these allometric relationships provides a direct method of interpreting both the changes occurring in each trait independently, and in relation to each other. competitive interactions between two in- dividual plants may be a function of leaf area display at a given time even if the pattern of leaf area display changes truly independent variable (i.e. will not An example: tissue nitrogen concentration drastically during plant developmentao. be altered because of phenotypic ex- and elevated CO, Comparisons at a common size are im- pression) and can be measured without Plants often respond to elevated at- portant in studies of allocation and error. If the relationship is linear, type 1 mospheric CO, levels with reduced tis- functional adjustments in biomass par- linear regression techniques (e.g. analy- sue nitrogen concentrations relative to titioning and resource acquisition. Other sis of covariance, repeated measures ambient CO,-grown plants when compari- developmental indices, such as node analysis of covariance, comparisons of sons are made at a common plant age”“. number, plastochron index28, growth fitted regression parameters) will suffice. It has been assumed that this phenotypic phase” (juvenile, reproductive), height If the relationship is nonlinear, type 1 pattern represents a functional adjust- and diameter of plant@, may be more nonlinear regressions may be compared ment that optimizes plant nutrient content appropriate depending upon the question. using methods described by Mead and when the relative availability of CO, and In all of these cases, these methods allow Curnow and Potvin et al.34 nitrogen is altered”“. However, enriched one to explicitly assess the degree to If plant growth or developmental atmospheric CO, accelerates the growth which patterns of variation in a trait may characters are chosen as the standard of rates of many plant species37. Further- be influenced by plant development. comparison, type 2 regression techniques more, plant nitrogen concentrations are Regardless of which standard of com- are appropriate, as neither variable can often highest in seedlings and sub- parison is deemed appropriate for any be considered truly independent and sequently decrease during growth. Does particular question, explicit examination measured without error. Type 2 linear the reduction in plant nitrogen content of patterns of change in phenotypic traits regression techniques (modifications of in elevated CO, conditions represent a during growth and development, in con- type 1 linear regression techniques listed functional adjustment in this phenotypic trast to comparisons at just one point, above) can be found in many statistical trait to changing resource availability, whether at a particular age or size, en- packages. However, many allometric or is it a consequence of differences in hances the functional interpretation of relationships in plants are nonlinear (i.e. nitrogen concentration between bigger, phenotypic variation. indicative of complex allometry rather more developed plants and smaller, less than simple allomet+‘), and thus pre- developed plants? Statistical tools clude the use of type 2 linear regression This question was addressed in a The choice of a standard of com- techniques. In these instances, type 1 non- study where two species of annual plants, parison will determine what statistical linear regression techniques described Abutilon theophrasti and Amaranthus retrv- tools are most appropriate. For example, above may be employed. In any case, data flexus, were grown with atmospheric CO, if time is the axis against which pheno- should be appropriately transformed to levels of 350 umol.mol-l or 700 ymol~molfl typic traits are compared, type 1 re- achieve normal distributions and hom- and with high or low fertilizer appli- gression techniques are appropriate, ogeneous variances before using stan- cation+. By harvesting individual plants since we can assume that real time is a dard regression techniques. every two days starting three days after PERSPECTIVES

(a) , , (b)

:s ;‘: ._

z 2- \ 2;; e9 I- Tz high N low N 9 cl@ -8 ml1 .01 .I 1 10 .a01 .01 .I 1 10 -10-9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 -9 -8 -7 -6 -5 -4 -3 -2 -I 0 1 2 3 Total plant biomass (g) In {total plant biomass - total plant N} (g)

Fig. 1. The results of a study that examined whether or not the interpretations of phenotypic variation in a physiological trait (%N) would be altered by consideration of the pattern of how that trait changes during growth and developmentz5. The effects of CO, and nitrogen on the growth and tissue nitrogen concentration of two annual plants (Abutilon tbeophrasti and Amaranthus retroflexus) were examined. Only data for A. theophrastiare shown, but identical conclusions were reached from studies of A. retrofiexus. The data show that plants exposed to high CO, (700 ppm; closed circles) grew faster (a) and had lower tissue nitrogen concentrations than ambient COP-grown (350 ppm; open circles) plants when compared at a common time (b). However, no differences were seen when comparisons were made at a common size (c). Furthermore, allometric analyses showed that CO, had absolutely no effect on the pattern of nitrogen use in biomass production (d). These results strongly suggest that the CO,-induced changes in %N that are observed at a common plant age are not the result of functional adjustments in plant nitrogen use, but are probably the result of comparing bigger, more-developed plants grown under elevated CO, with smaller, less-developed plants grown under ambient conditions (see Ref. 25 for a description of the experiments and statistical analyses).

germination, and using nonlinear re- response to resource availability is a written: Ian Baldwin, Fakhri Bazzaz, Eric gression techniques to analyse the datas4, major goal of plant ecology. Because Fajer, Sam McNaughton, Suzanne Morse, it was found that: (1) high CO,-grown plants keep their primary morphological Otto Solbrig, Sean Thomas, David plants had reduced nitrogen concen- surfaces on the outside (i.e. open organ- Tremmel, Peter Wayne and Ian trations and increased biomass relative ization), it has been assumed that they Welsford. Development of these ideas to ambient COzgrown plants when com- can easily branch and add new parts, was supported by NSF grant IBN-9207203 (J.S.C.); Bradley University pared at a common time (Fig. la and b); whereas the closed organization of ani- Research Excellence Award (K.D.M.); (2) tissue nitrogen concentrations did mals is thought to impose tight devel- NSF Dissertation Improvement Grant not vary as a function of CO, level when opmental constraints on morphological (D.D.A.); and United States Department plants were compared at a common size plasticity38. Thus, in contrast to animals14, of Energy Grant to Fakhri Bazzaz. (Fig. lc); (3) the rate of biomass accu- the developmental constraints on pheno- mulation per rate of increase in plant typic variation in plants have not been References nitrogen was unaffected by CO, avail- emphasized. Although strong interactions Campbell,B.D. and Grime,J.P. (1992) Ecology ability (Fig. Id). These results indicate between phenotypic expression, develop 73, 15-29 that, at least for these plants, CO,-induced ment and growth rate in plants have been Tilman, D. (1988) PIanf Strategies and fhe reductions in plant nitrogen concen- recognized for decades7, most functional Dynamics andStructure of Plant Communities, Princeton University Press tration that were observed at a given and evolutionary hypotheses of plant Bloom, AL, Chapin, F.S. IIIand Mooney, H.A. point in time were probably not due to response to resource limitations (e.g. (1985) Annu. Rev Ecol. Syst. 16,363-392 functional adjustments in plant nitrogen optimal partitioning models, plant strat- Sultan, S. (1987) Euol. Biol. 21, 127-178 content, but were probably the result of egy theory) do not fully incorporate these Mooney, H.A. and Winner, W.E. (1991) in accelerated plant growth in high-CO, relationships. A greater appreciation for Responses of Plants to Multiple Stresses conditions coupled with a fixed pattern the degree of phenotypic change during (Mooney, H.A., Winner, W.E. and Pell, E.J., of change in plant nitrogen content as development can only lead to a fuller eds), pp. 129-142, Academic Press plants get bigger. This example clearly understanding of the functional signifi- Crick, J.C. and Grime, J.P. (1987) New Phytol. demonstrates the importance of consider- cance of phenotypic variation in plants. 107,403-414 ing the relationship among phenotypic Evans, G.C. (1972) The Quantitative Analysis of traits, plant growth and plant develop- Plant Growth, University of California Press ment when testing hypotheses about the Acknowledgements Bazzaz, F.A., Garbutt, K., Reekie, E.G. and This is an equal-effort paper, and the Williams, W.E. (1989) Oecologia 79, 223-225 functional role of phenotypic variation. order of authorship was chosen at Bormann, F.H. (1958) in The Physiology of random. We would like to thank several Forest Trees phimann, K.V., ed.), pp. 197-215, Conclusion people, without whose support and Ronald Press Understanding the degree and direc- participation in continuing debates this 10 Knapp, A.K. and Fahnestock, J.T. (1990) Funct tion of phenotypic changes in plants in manuscript would not have been Ecol. 4. 789-798

190 TREE vol. 9, no. 5 May 19.94 _~ PERSPECTIVES

11 Poethig, R.S.(1990) Science 250.923-930 26 Poorter, H. and Remkes, C. (1990) Oecologia coral-zooxanthellae associations in coral 12 Samson, D.A.and Werk,K.S. (1986) Am. Nat 83,553-559 reefs, even though our understanding 127.667-680 27 Poorter, H. and Pothmann, P. (1992) New of the community impact of these associ- 13 Klinkhammer, P.G.L., Meelis, E., de Jong, T.J. Phyrol. 120, 159-166 ations is almost entirely speculative. and Weiner, J. (1992) Funcl. Ecol. 6,308-316 28 Ackerly, D.D., Coleman, .I.%,Morse, S.R. and Moreover, while the evolutionary role of 14 Bernardo, J. (1993) Trends Ecol. &IO/.8, 166-173 Bazzaz, F.A. (1992) Ecology 74, 1260-1269 15 Mooney, H.A.et al (1988) Oecologia 72,502-506 29 Hunt, R. (1990) Basic Growth Analysis, Unwin positive interactions has become clear 16 Evans, G.C. and Hughes, A.P. (1961) Nem Hyman Press over the past decade (e.g. the evolution of Phytol. 60,50-180 30 Tremmel, D.C. and Bazzaz, F.A. (1993) Ecology eukaryotic cells, and flowering plants and 17 Terry. N. (1968) J Exp. Bat. 61, 795-811 74,2114-2124 their pollinators), positive interactions 18 Hughes, A.C. and Evans, G.C. (1962) NeE 31 Mohler, CL., Marks, P.L. and Sprugel, D.G. remain absent from general models of Phyhl. 61, 154-174 (1978)J. Ecol. 66,599-614 community dynamics and organization. 19 Corre. W.J. (1983) Acta Hot. Neerl. 32, 49-62 32 Joliceur, P. (1989) J. Theor. BioL 140,41-49 How can an evolutionary play featuring 20 Rice, %A. and Bazzaz, F.A. (1989) Oecologia 78, 33 Mead, R. and Curnow, R.N. (1983) Statistical strong positive interactions take place on 502-507 Methods in Agriculture and Experimental an ecological stage where positive inter- 21 Pearsall, W.H. (1927) Am 1 Bol. 41,549-556 Biology, Chapman &Hall actions are insignificant? Textbooks, how- 22 Troughton. A. (1956) J Hrit. Grass. Sac 11, 34 Potvin, C., Lechowicz, M.J. and Tardif, S. 56-65 (1990) Ecology 71,1389-1400 ever, strongly support our contention 23 Ledig, F.T. and Perry, T.O. (1965) froc. Sot. 35 Wong, SC. (1979) Oecologia 44,68-74 that positive interactions are currently Am. For, 39-43 36 Hilbert, D.W. (1990) Ann. Bat. 66,91-99 largely overlooked by community ecol- 24 Ledig, F.T., Bormann. F.H. and Wenger. K.F. 37 Bazzaz, F.A. (1990) Annu. Reo. Ecol. Syst 21, ogists. Whereas ecology textbooks earlier (1970) Bat.Gaz. 13. 349-359 167-196 this century devoted as much attention 25 Coleman. .J.S..McConnaughay, K.D.M.and 38 Schmid, B. (1992) Euo/. Trends in Plan& 6. to positive interactions as they did to Bazzaz. F.A. (1993) Oeco/ogia 93, 195-200 45-60 competitive ones, modern textbooks hardly mention positive interactions in a community contextl’l. Recent theoretical (e.g. Refs 11,12) and empirical (e.g. Refs 13,14) work has Positive interactions in communities suggested that positive indirect interac- tions and feedback mechanisms in food- webs may be common important forces species involved; thus, we include fac- in natural communities. In this article, Mark D. Bertness ultative and obligatory facilitations and however, we focus on the direct, non- Ragan Callaway mutualisms. Whereas the ecology and trophic positive interactions that early evolutionary biology of mutualisms has ecologists suggested were critical aspects Current concepts of the role of inter- attracted recent attention7, the role that of community dynamics and organiz- specific interactions in communities they play in the structure and organiz- ation1,2.Direct positive interactions occur have been shaped by a profusion of ation of natural communities has not. when neighbors modify physical and/or experimental studies of interspecific The lack of recent attention paid to biotic conditions and lead to positive competition over the past few decades. the role of positive interactions in com- effects. Although these positive inter- Evidence for the importance of positive munities is at least partly due to their actions have been largely ignored by interactions - facilitations - in uncritical acceptance by early ecologists theoretical ecologists, evidence from a community organization and dynamics and the preoccupation of contemporary wide range of communities has begun to has accrued to the point where it community ecologists with competition emerge during the past five to ten years warrants formal inclusion into community (but see Ref. 8). In addition, much of the indicating that direct positive interac- ecology theory, as it has been in evolu- early development of ecology which high- tions may be common, predictable and tionary biology. lighted positive interactions pre-dated pervasive forces in natural communities the common use of field experiments and in physically harsh environments in Mark Bertness is at the Dept of Ecology and in ecology and thus received little criti- particular. Here, we examine a small sub Evolutionary Biology, Brown University, cal testingg. Moreover, fascination with sample of this evidence and re-evaluate Providence, RI 02912,USA; RaganCallaway is at the Divisionof competition has focused attention on the role of direct positive interactions in BiologicalSciences, University of Montana, communities where competition is con- ecological communities. Missoula, MT 59812, USA. spicuous, potentially distracting ecol- ogists from even recognizing positive Do positive interactions affect interactions. Consequently, while facili- recruitment? arly ecological theory included both tative and/or positive interactions are part Positive interactions during recruit- E positive and negative interactions of most working ecologists’ conventional ment in desert plants were hypothesized among species as important driving forces wisdom, and while anecdotal examples thirty years ago, based on spatial patterns in the structure and organization of natu- can be shown in most communities, the suggesting that neighbors buffer one ral communitiesl.2. More recently, the role general importance of positive interac- another from potentially limiting physical of competition in natural communities tions to community diversity, structure stresses’5.16. But ecologists have tended has received considerable attention (see and productivity is rarely acknowledged. to view these interactions as idiosyn- Refs 3,4), while positive interactions have Modern ecology’s view of positive cratic features of deserts rather than received little attention and are largely interactions is particularly puzzling given examples of general principle and, until ignored in current models of community the prominent role that they play in recently, have not experimentally tested organizatior@. We broadly define posi- many communities and their importance these ideas. Nurse-plant effects and posi- tive interactions as all non-consumer as evolutionary forces. Few ecologists tive density-dependent recruit survivor- interactions among two or more species would deny the potential importance of ship, however, have recently been found that positively affect at least one of the mycorrhizal associations in forests and in other harsh physical environments,