Quantitative analysis of compensatory and catch-up growth in diverse taxa


  • Katie L. Hector,

    Corresponding author
    1. Department of Zoology, University of Otago, 340 Great King St, Dunedin, 9016, New Zealand
      author. E-mail: katiehector1@gmail.com
    Search for more papers by this author
  • Shinichi Nakagawa

    1. Department of Zoology, University of Otago, 340 Great King St, Dunedin, 9016, New Zealand
    2. National Research Centre for Growth and Development (NRCGD), University of Otago, 340 Great King St, Dunedin, 9016, New Zealand
    Search for more papers by this author

author. E-mail: katiehector1@gmail.com


1. ‘Compensatory growth’ and ‘catch-up growth’ are often used interchangeably to describe the faster than optimal growth that occurs following a period of dietary restriction in the development of many animals. Concerns about the statistical analysis of these studies have drawn attention to the risk of false detection in reports of compensatory and catch-up growth.

2. This study aims to quantify the degree to which these compensatory responses occur across the animal kingdom. In addition, this study distinguishes the two terms, ‘compensatory growth’ and ‘catch-up growth’, to clarify the fitness consequences of rapid growth. Compensatory growth refers to a faster than usual growth rate, while catch-up growth implies attainment of control size.

3. Eight meta-analyses and meta-regression analyses were conducted on data extracted from 88 papers, including 11 taxonomic classes. The results confirmed that both growth tactics (i.e. compensatory and catch-up growth) occur across a wide range of taxa and result in decreased direct fitness components.

4. Importantly, the meta-analytic methods used made it possible to identify the specific experimental techniques that most successfully promoted rapid growth after restriction and key differences in the responses of the four major groups (mammals, birds, fish and arthropods) to dietary restriction. Endotherms are more likely to show a compensatory growth response because of their determinate growth; in contrast, the indeterminate and saltatory growth tactics of fish and arthropods reduce the pressure to rapidly achieve a large size.

5. Among the first meta-analyses to be conducted in this field, this study provides valuable support for the premises of compensatory and catch-up growth and also discusses weaknesses in experimental design, and possible solutions, in compensatory growth research. For example, we recommend conducting the experiment within the most linear phase of an animal’s growth to avoid analytical complications arising from size-dependent growth, and our results indicate that dietary dilution more closely resembles quantitative restriction than clutch size and intermittent feeding restriction methods when normal quantitative restriction is not possible.


Under ideal circumstances, animals grow at an optimal rate, limited more by quality control of differentiating tissues than by a lack of resources (Metcalfe & Monaghan 2003). Optimal growth reduces the negative physiological costs of the accumulated cellular damage from rapid growth while maximizing fitness potential. Faster than optimal growth (maximal growth) can decrease cell functioning efficiency, immune function and resistance to physiological stressors (Mangel & Stamps 2001). However, growth rates closer to maximal are commonly observed when an organism has previously experienced a period of limited growth during development. These responses are commonly known as compensatory growth or catch-up growth.

Compensatory growth has long been of interest to scientists (Jackson 1937), primarily because it begs the question: if some animals are willing to grow at a maximal rate, why don’t they all? The answer seems to lie in the altered cost–benefit equation of an animal with a poor start in life. An animal’s adult size is often a major factor in fitness; size is known to affect mate selection, fecundity and offspring survival (Blanckenhorn 2005). Therefore, animals with slow growth during development are at a distinct disadvantage if they reach a small adult size. If the environmental cause of restricted growth passes, the opportunity may present itself for an animal to increase its growth rate to avoid becoming a small adult. In this case, it is more beneficial for the growth-restricted animal to risk the negative consequences of maximal growth than bear the fitness costs of small adult size. In many circumstances, the cost of maximal growth, in the form of accumulated cellular damage, is not paid until after the reproductive phase. For example, in humans, rapid growth in childhood after a ‘small-for-gestational-age’ birth weight is associated with increased risk of late-onset adult diseases such as heart disease, diabetes and obesity (Cottrell & Ozanne 2008). Similar findings have been reported, for example, from the intensive studies of metabolic syndrome in rats (Bol et al. 2009; Porrello et al. 2009). Compensatory growth is also known to decrease the maximum life span of individuals in a number of species (Metcalfe & Monaghan 2003). Thus, compensatory growth allows animals to reach a larger size, which increases reproductive fitness before the consequences of rapid growth negatively affect them.

The theory of compensatory growth has been reviewed by a number of researchers, with a variety of perspectives. Metcalfe & Monaghan (2001) and Dmitriew (2011) provide a broad outline of the forms and consequences of growth compensation across many taxa. Compensatory growth has also been reviewed with specific taxa in mind (domestic fowl: Nir et al. 1996; fish: Ali, Nicieza & Wootton 2003) and with life-history model simulations to support theoretical assumptions (Mangel & Munch 2005). Collectively, these reviews voice strong support for compensatory growth as a measurable, repeatable and taxonomically diverse real-world phenomenon. However, the review by Nicieza & Alvarez (2009) addresses concerns about some of the compensatory growth literature by criticizing the statistical methods used to analyse this type of data. They claim that some analyses may not take into account the size dependence of growth rates, meaning that as an animal gets larger, its growth slows. The relationship between resource allocation to maintenance and to growth changes as an animal grows larger, resulting in a size-dependent growth curve common to most animals when comparing the overall shape of their early growth (West, Brown & Enquist 2001).

The terminology used in this area of research could be considered another impediment to the accurate reporting of compensatory growth (Jobling 2010). In many cases, compensatory growth and catch-up growth are used as interchangeable terms, both meaning that the growth rate of a previously restricted group is significantly higher than a control. This is the explanation that has the most relevance to a trade-off for fitness, owing to the cellular damage incurred with maximal growth. However, in other studies, the term catch-up growth is defined more by the actual ‘catching up’ of adult size rather than the growth. Reaching the same final size as the controls is important in many species where size-dependent fitness traits occur, such as predation, mate choice, social dominance and fecundity (Blanckenhorn 2005). Catching up to the size of normal conspecifics can be achieved by extending the developmental period while continuing at an optimal growth rate (Arendt 1997).While there may be some cost to lifetime reproductive success owing to the extra time spent in a non-reproductive phase (Oli, Hepp & Kennamer 2002), this strategy can often be favourable in variable environments and comes at minimal physiological cost (Wilbur & Rudolf 2006). Thus, it is often unclear what the authors really mean in reporting results as ‘compensatory’ or ‘catch-up’ growth, and this lack of clarity in terminology means that the associated fitness consequences are easily confused. For the sake of clarity in this study, ‘catch-up growth’ will refer to the attainment of a non-significant difference in size between the control and previously restricted animals. This can be achieved at a normal growth rate. In contrast, ‘compensatory growth’ refers to a significantly steeper growth rate of the previously restricted animals, which may or may not result in catching up to the same size of control-fed animals (Fig. S1, Supporting information).

Compensatory growth is hypothesized to have evolved because the fitness consequences are delayed until after reproduction (Metcalfe & Monaghan 2001). Therefore, selection favours those individuals capable of compensatory growth because of their increased size-dependent fitness compared to individuals with limited developmental plasticity in growth (Yearsley, Kyriazakis & Gordon 2004). This selection is dependent on the delay before health consequences occur being sufficiently long to allow unimpaired reproductive effort. The experimental design, in particular the duration of the relevant periods of restriction and realimentation (when animals are returned to a normal diet), is often a seemingly arbitrary decision. While there are a number of reviews on the topic (Wilson & Osbourn 1960; Metcalfe & Monaghan 2001; Ali, Nicieza & Wootton 2003, Monteiro & Victora 2005), there has yet to be a comprehensive meta-analysis of compensatory growth across diverse taxa.

The main aim of this paper was to use meta-analytical techniques to explore a number of hypotheses about compensatory growth. We tested whether: (i) nutritional restriction slows the growth rate and affects the size of a variety of taxa compared to well-fed conspecifics; (ii) restricted animals are able to reach the same final size as controls (‘catch-up’) after a subsequent return to normal feeding amounts; (iii) restricted animals show faster than normal growth (‘compensatory growth’) after nutritional restriction; and (iv) early restriction has an effect on later fitness traits. By analysing each of these hypotheses with a number of moderators included, such as the degree to which an animal was restricted and the method used for restriction, we aimed to gain a better understanding of the variation in the literature and compensatory growth as a whole.

Additionally, a subsequent analysis was performed with the aim of accounting for the different growth tactics and metabolism between the various taxonomic classes. We selected four groups with sufficient data points to explore these possibilities: mammals, birds, fish and arthropods. Birds and mammals both have determinate growth tactics so are most likely to benefit from compensatory growth because growth ceases at maturation (Metcalfe & Monaghan 2003). The higher metabolic rate of birds may affect their ability to grow faster than normal, as a result of the already much higher demands on energy resources of birds compared with mammals (Speakman 2005). Fish have indeterminate growth, so may be under only seasonal pressure to rapidly compensate from restricted growth (Kozlowski 1996). Arthropods show saltatory growth, which proceeds in bursts of growth following ecdysis, so rapid growth may not be necessary as time between instars can be prolonged (Dmitriew & Rowe 2007). The higher starvation resistance and lower fat stores of fish and arthropods as ectotherms may also affect their energy storage responses compared with endotherms (Wang, Hung & Randall 2006). We aimed to identify whether these expectations, based on innate growth tactics and metabolic differences, were relevant and detectable signals within our analyses.

Materials and methods

Data Collection

We collected data from papers published in peer-reviewed journals. Primarily, papers were sourced from searches of ISI Web of Science, the most recent of which was conducted on 5th August, 2010. For inclusion in this analysis, a paper must: (i) be an empirical study (i.e. no computational simulations), (ii) not include animals that have been genetically modified or have a known disorder, (iii) describe an experiment with a treatment and a control group, (iv) report body mass before and after restriction and after realimentation along with estimates of uncertainty (e.g. standard deviation and/or standard error), (v) have quantifiable degrees of diet restriction in a controlled environment (thus excluding correlational data from wild populations in different areas of food abundance) and (vi) report the sample size. Only compensatory growth as a result of dietary restriction, as opposed to temperature or seasonal effects, was considered to be of interest for this study. Most studies used quantitative restriction (i.e. limiting food amounts); however, we included alternative restriction methods, namely diet dilution with indigestible material, intermittent feeding and clutch size manipulations. Diets that limited only one nutrient, e.g. methionine, were not included because they were considered less ecologically relevant.

To investigate the effects of compensatory growth, we collected body size data of restricted and control-fed animals prior to restriction, at the end of restriction and at the end of the experiment (after realimentation). We primarily collected mass data; however, length data were also recorded where available and ranged from total body length to tarsus length. This size estimate would allow us to compare compensatory growth in mass with compensatory growth in length, to account for the possibility of confounding fat accumulation. Growth rates were calculated from the three experimental time points indicated, as discussed in more detail below (Fig. 1).

Figure 1.

 Hypothetical growth curves of control (Con) and diet-restricted (Res) treatment groups representing the comparisons used in the slopes analyses. The shaded area is the restriction period (period A), and the non-shaded area indicates the period post-restriction (realimentation, period B). The comparisons were either control period A (dashed line) against restricted period B (bold line; A vs. B) or control period B (dotted line) against restricted period B (bold line; B vs. B).

Studies that used farmed or model laboratory species were also required to have some fitness-related trait included, such as mortality or fecundity (Table S1, Supporting information). We decided this criterion was the best way to optimize the variety of species included, because some studies of taxonomically diverse wild-caught animals did not include fitness-related traits. This criterion also reduced the representation of agricultural studies (c. 300 agricultural studies with no fitness estimate), which would have overwhelmed non-farmed animals in the analysis. We note that agricultural papers rarely report fitness-related trait data, more commonly focussing on meat quality and characteristics instead.

In summary, we collected data from 88 papers (Table S2, Supporting information), from which we extracted 226 comparisons of the size of restricted animals with their reported control groups. These data included 58 species, spanning eight classes within the phyla Chordata, Arthropoda and Mollusca. Because some papers reported multiple fitness component measures, we collected a total of 207 fitness component comparisons between restricted and control animals, which we categorized into the six broad classes of fitness traits (see Supporting information for analysis of ‘indirect’ fitness traits). For use in the main study, only those fitness traits that were considered ‘direct’ fitness components (survival, number of offspring, lifetime reproductive output, fertility and hatchability) were included. This limited the fitness analysis to 94 comparisons.

Statistical Analysis

Data were analysed using R 2.13.1 (R Development Core Team, 2011) and S-PLUS 8.0.4 (TIBCO; http://www.tibco.com/). Hedges’d was considered the most appropriate measure of effect size because it estimates the magnitude of the difference between two groups. Using this measure, as opposed to Cohen’s d, controls for an upward bias caused by small sample size (Nakagawa & Cuthill 2007).The variance of d was used as the weighting value for statistical analysis of the overall effects, such that estimates with lower variance and therefore more reliability (either from more consistent results or a greater sample size) contributed more to the model (i.e. meta-analytical models). Estimating growth slopes from the two periods observed (A: from the beginning to the end of restriction; and B: from the end of restriction to the end of the experiment, i.e. the realimentation period; Fig. 1) required additional analysis. Details are provided in the Supporting information.

To answer the questions which we had initially posed, three separate sets of meta-analytical models were required: (i) models investigating size differences at each point of interest (before restriction, after restriction and after realimentation at the end of the experiment); (ii) the effect the dietary treatment had on fitness; (iii) the effect of the dietary treatment on growth rates as measured by the growth slopes during the different periods, A or B (Fig. 1). Nicieza & Alvarez (2009) identified size-dependent growth as a prominent confounding variable in analysis of compensatory growth (i.e. control animals are larger and therefore grow at a slower rate, so comparison with post-restriction growth of small restricted animals is biased). As such, we compared the ‘compensatory growth’ (period B) of the previously restricted animals with both period A and period B of the control animals (to give two estimates of growth rates initiated at different sizes; Fig. 1). We also compared differences when the studies were limited to those which had controls showing only linear growth (87 studies, Fig. 1). By only comparing studies where controls showed linear growth over the entire experiment, we could be certain that the size dependence of growth would not be a concern. Identification of linear growth is described in further detail in the supporting information.

Each meta-analysis was analysed as a linear mixed-effects model, using a modified version of the method described by Nakagawa et al. (2007). In all analyses, we controlled for experimental design effects by including both paper identity and taxonomic class as random factors (the former nested within the later). Both null models (model with the intercept; classically considered and referred to as meta-analysis) and scaled best models with moderators (often referred to as meta-regression; Table 1) were used to interpret our results. For scaling, all continuous moderators had the mean subtracted from each value and were divided by two times the standard deviation. This method of scaling and centring allows the outputs of the model to be more fairly interpreted (Gelman 2008; Schielzeth 2010). Binary variables were left unscaled, and sex (which had the values: both, male and female) was analysed with ‘both’ as the reference variable so that the male and female output values could be interpreted as the effect of looking at only one sex as opposed to mixed-sex experiments (Table 1).

Table 1.   Descriptions of the moderators incorporated in the models. ‘Relative’ means calculated as a percentage of the average longevity for each species (see Table S4, Supporting information)
InterceptOverall effect of dContinuousSize measures:
+ large restricted group − small restricted group
+higher fitness for restricted
−lower fitness for restricted
+slower growth for restricted
−faster growth for restricted
TypeMeasurement typeBinary+Mass −length
DegreeDRDegree of restrictionContinuous+More food −less food
AgeRelative age of restriction onsetContinuous+Older −younger (log-scale)
DurResRelative duration of restrictionContinuous+Longer −shorter (log-scale)
PropRealProportion of duration of realimentation:restrictionContinuous+Longer −shorter
FarmedFarmed or laboratory colonyBinary+True −False
FemaleFemale compared to mixed sexBinary+True −False
MaleMale compared to mixed sexBinary+True −False
ClutchSizeRestricted by clutch size manipulationBinary+True −False
IFRestricted by intermittent feedingBinary+True −False
DilutionRestricted by dilution of dietBinary+True −False
AdLibRealRestricted group fed ad libitum for realimentationBinary+True −False
MortalityCompared to reproductionBinary+True −False

Best models were then selected by running the full model using the maximum likelihood method, as opposed to restricted maximum likelihood (REML), because the changes in the AIC (Akaike information criterion, measuring model goodness of fit) values are more relevant under maximum likelihood (Pinheiro & Bates 2000). Least significant moderators were, then, sequentially removed until the AIC value was no longer lowered, and the best model was then reverted to REML so that the effect size estimates could be used in interpreting the data. Full models are reported including all the moderators initially included in the supporting information. Some of the moderators were not directly given from the original papers and were instead calculated by combining data. Namely, the relative age, duration of restriction and duration of realimentation were all calculated as a percentage of the reported maximum longevity of the species and, importantly, were converted to the natural log-scale before being scaled and centred. The proportion of the duration of realimentation to restriction was calculated by dividing the length of the former by the latter.

The four-group analysis consisted of the three classes and one phylum, which constituted the majority of our data set: mammals (= 38), birds (= 39), fish (= 91) and arthropods (consisting of classes Insecta, Arachnida, Branchiopoda and Malacostraca, = 30). We analysed these by performing exactly the same modelling process as above, with study as a random factor, and incorporated the groups as an interaction term with each moderator. Only interaction effects which decreased the AIC value (if any) remained in the final best model. As a multilevel factor, the other groups were regressed against ‘fish’, selected because it had the highest sample size. For estimating overall intercepts by group, each group was fitted as the reference level.

Validation of Meta-Analytical Techniques

To ensure that the analyses were not affected by publication bias, funnel plots of the data sets used in each analysis were visually inspected (Fig. 2, see Fig. S5, Supporting information for fine scale). No obvious asymmetry was detected in the plots, other than that which was expected to reflect true biological heterogeneity in the post-restriction and post-realimentation data sets (Egger et al. 1997; Fig. 2b–c). I2 is a measure of heterogeneity in a meta-analysis and can be interpreted as the percentage of the total variability in a set of effect sizes owing to between-study variability (Huedo-Medina et al. 2006). The high I2 values reported reflect the high degree of inconsistency across studies (Higgins et al. 2003; Table 2). The large contribution of between-study differences (as opposed to between-class differences) to the overall heterogeneity reflects the importance of moderators in controlling for differing experimental techniques (Table 2). In most cases, the best model actually increased the heterogeneity compared to the null model, although the AIC value of all best models was lower than the null models.

Figure 2.

 Funnel plots showing the distribution of the effect sizes (d) extracted from each study plotted against the precision of the study (1/SE). Asymmetry around the null intercept (dashed line) can indicate either publication bias or true biological heterogeneity. For a-h, = 212, 218, 226, 94, 226, 87, 226 and 87, respectively.

Table 2.   Values of the I2 statistic reflecting the heterogeneity of each null and scaled best model. The relative contribution of taxonomic class and source paper to model heterogeneity is also reported as well as the AIC value calculated by maximum likelihood for each model
A vs. BNull99·520·0099·521744
A vs. B linearNull96·1427·0369·12301
B vs. BNull95·938·0787·86665
B vs. B linearNull93·245·2288·01196


Meta-Analysis: Null Models

The null models reflect the overall effect size of d without considering any moderators but still accounting for paper identity nested within taxonomic class. The pre-restriction intercept (i.e. meta-analytic mean) shows that there was no size difference between treatment and control groups prior to restriction (± SE: 0·02 ± 0·03, = 0·642, d.f. = 124, = 0·522; Fig. 3a). By the end of restriction, the restricted group was statistically significantly smaller than controls (± SE: −1·78 ± 0·02, = −9·57, d.f. = 130, < 0·0001; Fig. 3a). By the end of the experiment, the animals in the restricted group had failed to ‘catch-up’ to the size of controls and were still statistically significantly smaller (± SE: −0·54 ± 0·10, = −5·416, d.f. = 138, < 0·0001; Fig. 3a) than those in the control group. There was a statistically significant negative effect on fitness components as a result of the treatment (± SE: −0·25 ± 0·09, = −2·824, d.f. = 57, = 0·0065; Fig. 3a).

Figure 3.

 The intercepts of the null models for each of the eight meta-analyses, grouped as (a) standardized difference between point estimates in size and fitness components and (b) standardized difference between growth slopes. For the former, a negative value of d indicates smaller size or decreased fitness in the restricted group compared with the control group. For the latter, a negative value indicates that the restricted group grew faster than the control group during the period of interest (see Fig. 1). Confidence intervals that span zero indicate no statistically significant effect at α = 0·05.

Meta-analytic models using standardized differences (Hedges’d) between the slopes showed more uncertain estimates (Fig. 3b) than the models using d between point estimates (Fig. 3a). The intercept for period A vs. period B (see Fig. 2a for explanation) shows that there was no detectable difference between the initial growth of the controls and the growth rate of the restricted animals during realimentation (± SE: 0·25 ± 0·27, = 0·894, d.f. = 124, = 0·373; Fig. 3b). However, when only experiments during the linear growth phase of the controls were considered (A vs. B linear, Fig. 1), the restricted animals appeared to show ‘compensatory growth’ at a faster rate than the early growth of controls (± SE: −0·45 ± 0·13, = −2·090, d.f. = 49, = 0·042; Fig. 3b). This effect was also found to be true when comparing the period B growth of all the control animals (B vs. B, Fig. 1), although this result would be expected if size-dependent growth is causing overinflated estimations of compensatory growth, as Nicieza & Alvarez (2009) suggest (± SE: −0·30 ± 0·13, = −2·379, d.f. = 124, = 0·019; Fig. 3b). Their claim of false detection is supported by the finding that the difference for B vs. B is eliminated when only linear control growth is considered (B vs. B linear, Fig. 1; ± SE: −0·15 ± 0·13, = −1·172, d.f. = 49, = 0·247; Fig. 3b).

Meta-Regression: Scaled Best Models

These models take into account the effect of various moderators, as selected by the methods described earlier. To interpret the output of these models, it is necessary to remember that the models are scaled. The intercepts reflect the overall effect size based on the mean values of continuous variables and the default level of binary variables. For example, an intercept that had dietary restriction and farming as moderators would reflect the overall effect based on a moderate degree of restriction and non-farmed animals. The moderator values provided could then be interpreted as the slope of the degree of restriction centred on this intercept and the effect when farmed animals are considered. Inclusion in the best model did not necessarily produce statistical significance in all moderators. For ease of reading, statistical values are only provided for intercepts, but full statistical information for moderators is available in Tables S6–S13 (Supporting information).

There were no statistically significant moderators for the pre-restriction analysis, suggesting experiments were indeed fair and both treatment groups started at a similar size (± SE: −0·04 ± 0·04, = −1·006, d.f. = 124, = 0·316; Fig. 4a). Following restriction, the restricted animals were found to be statistically significantly smaller than controls (± SE: −1·70 ± 0·32, = −5·461, d.f. = 125, < 0·0001; Fig. 4b). The size difference between the control and the treatment groups was heavily dependent on how severe the restriction was, how long the diet lasted and how young the animals were when restriction began. Restriction also showed a greater effect on mass than on length. Intermittent feeding and clutch size manipulation appeared to be less effective than other diet methods.

Figure 4.

 Coefficient plots from meta-regression analyses showing the effects of the treatment on the restricted group in comparison with the control group (a) before the experiment, (b) after restriction, (c) after realimentation and (d) on fitness components. The intercept shows the overall model outcome, and the other points describe the contribution of each moderator to the intercept. Confidence intervals that span zero indicate no significant effect at α = 0·05. For graphs a–d, = 212, 218, 226 and 94, respectively. For binary variables, the coefficient shows the impact of the named effect in contrast to the default value, upon which the intercept depends. The continuous variables indicate the slopes of these variables against the effect of d. The intercept is based on intermediate values of these variables. For example, degree of restriction tends to have a positive slope in relation to d, indicating that more food for the restricted group means a smaller difference in the typically negative relationship between control and restricted size (d). See also Table 1 for information on moderators and Tables S6–S9 (Supporting information) for relevant statistics to these analyses.

Contrary to the results of the null model, animals did achieve ‘catch-up’ growth when taking into account a number of variables, most significantly the duration and severity of the restriction diet (± SE: −0·08 ± 0·31, = −0·244, d.f. = 134, = 0·807; Fig. 4c). Restricted animals were more likely to attain the same length as controls than the same mass, although this could be because length was not as severely affected by restriction.

In support of the null model, fitness components were found to be negatively affected by the diet treatment overall (± SE: −1·17 ± 0·48, = −2·436, d.f. = 52, = 0·018; Fig. 4d). The effect size of the impact on fitness components was even larger when accounting for moderators, especially duration of restriction. Mortality was more negatively affected by treatment than reproduction, and negative fitness consequences were more apparent after longer periods of realimentation. Males and females in single-sex experiments had significantly higher fitness components than mixed-sex experiments. Animals on intermittent feeding regimes were less likely to show deficits in fitness components. These results were extracted from ‘direct’ fitness-related trait measurements only. The analysis of fitness with ‘indirect’ fitness components found no overall effect on fitness-related traits (see Supporting information).

With experimental methods taken into account, both A vs. B and the linear selection of A vs. B (Fig. 1) showed no statistically significant difference in growth rate between period A of the controls and realimentation of the restricted group (A vs. B ± SE: 0·64 ± 0·34, = 1·913, d.f. = 117, =0·058; Fig. 5a; A vs. B linear ± SE: −0·20 ± 0·22, = −0·894, d.f. = 47, = 0·376; Fig. 5b). Both analyses also agreed that compensatory growth was more likely in mass than length and a longer duration of restriction was found to decrease the likelihood of compensatory growth. For all studies, a longer proportion of realimentation decreased the chance of detecting compensatory growth. Single-sex experiments were more likely to show compensatory growth than mixed-sex studies. Intermittent feeding and clutch size manipulation were found to cause slower growth of the restricted group compared to initial growth of controls.

Figure 5.

 Coefficient plots showing the treatment effects on the growth rate of the restricted group after realimentation (‘compensatory growth’) compared to (a) the control group during the restriction period (period A) from all studies, (b) the control group during period A where only control groups in the linear growth phase were included, (c) the control group during the realimentation period (period B) from all studies and (d) the linear control groups during period B (see Fig. 1). The intercept shows the overall model outcome, and the other points describe the contribution of each moderator to the intercept. Confidence intervals that span zero indicate no statistically significant effect at α = 0·05. For graphs a–d, = 226, 87, 226 and 87, respectively. For further explanation on interpreting moderator values, see Fig. 4 and Table 1. Full statistical information for each meta-regression analysis is reported in Tables S10–S13 (Supporting information).

Controlling for experimental methods shows that compensatory growth is achieved by the restricted group when compared with the post-restriction growth of the controls (± SE: −0·57 ± 0·17, = −3·312, d.f. = 121, = 0·001; Fig. 5c). This finding was moderated by the type of measurement, with compensatory growth in mass being slightly weaker. The evidence for compensatory growth could be interpreted as false detection, because faster period B growth is predicted given size-dependent growth of the controls. However, analysis of linear control growth shows that compensatory growth is a robust effect (± SE: −0·80 ± 0·37, = −2·173, d.f. = 45, = 0·035; Fig. 5d). The linear analysis again demonstrates that mass has weaker compensatory growth than length and that, like the full analysis, clutch size and intermittent feeding are unlikely to lead to compensatory growth. The linear analysis also indicates that longer restriction periods are less likely to result in compensatory growth.

Four-Group Analysis

The main effects from each of the eight meta-analyses summarize the relative outcome of dietary restriction for each group (mammals, birds, fish and arthropods; Fig. 6). It is clear that while all groups are much smaller after restriction, arthropods are most likely to catch-up. The difference between controls and previously restricted animals is much less substantial after realimentation in all groups. Mammals were the only group with detectable negative consequences for fitness components. Compensatory growth in birds was statistically significantly faster than control growth in period A, while mammal compensatory growth was statistically significantly faster than control growth in period B. Interaction effects showed that mammals were much more sensitive to the severity of the diet than other groups, but less sensitive to the duration of restriction. The effect of restriction on mass was greater than on length in arthropods, while both measures were affected in the other groups. After a longer period of realimentation, more negative consequences for fitness components are apparent in mammals, while arthropods are more likely to avoid fitness costs. Longer restriction periods resulted in much faster compensatory growth in birds (compared to period A controls) than in other groups. Analysis including groups did eliminate some of the moderators included in the complete data set analysis. For example, the distinction between mortality and reproduction fitness-related traits was no longer important. Full statistical information is provided in the supplementary material (Table S14, Supporting information).

Figure 6.

 Coefficient plots showing the intercepts of each of the four groups (arthropods, = 30; birds, n = 39; fish, = 91; mammals, = 38) when included in the original eight meta-regression analyses as interaction terms (pre-restriction size not shown because there no group effects). Intercepts illustrate the relative strength of size, fitness or growth rate effects between groups. Negative intercepts with confidence intervals not including zero indicate smaller size (post-restriction, post-realimentation), decreased fitness-related traits (fitness) and faster post-restriction growth compared with controls during restriction (A vs. B, A vs. B linear) or realimentation (B vs. B, B vs. B linear).


These meta-analyses and meta-regression analyses support the basic assumptions of altered growth patterns in response to early dietary restrictions: animals may reach the same size as controls after a period of restriction (catch-up growth) and can grow at a faster than optimal rate to achieve this end (compensatory growth; Fig. S1, Supporting information). In reality, the occurrence of both growth patterns is largely dependent on the experimental methods used, as indicated by the large effects of a number of the moderators included and the high heterogeneity among studies (Table 2). In particular, the inclusion of such important moderators as the degree and duration of restriction changed the net result of the post-realimentation analysis, which previously indicated that animals do not ‘catch-up’ to the size of controls after restriction (Fig. 3a). Likewise, the meta-analyses for growth slopes showed equivocal results prior to the inclusion of the preeminent moderators ‘measurement type’ and the duration or proportion of restriction (Fig. 3b). Biologically, these experimental factors have an obvious influence on the growth of animals. In meta-regression analyses, we were able to account for these interstudy differences by scaling and centring data on intermediate values, in order to observe the true effects of ‘catch-up growth’ and ‘compensatory growth’.

Our results also support the claim that compensatory growth comes at a cost to fitness components, at least in ‘direct’ measurements of survival and reproductive output (Mangel & Munch 2005; Fig. 4d, see Supporting information for ‘indirect’ measurements). The magnitude of the effect of duration of restriction on fitness components is particularly notable. Longer than average periods of restriction appear to have a tremendous negative impact on the fitness-related traits of an animal. However, as a cautionary note, all studies included in our meta-analyses were conducted in controlled environments to ensure that feeding amounts could be manipulated and monitored. Therefore, the subjects were spared the real life trials of predator avoidance, mate selection and pathogen exposure, which all clearly have great bearing on the fitness outcomes of wild animals. Studies such as Johnsson & Bohlin (2006), where restricted trout regained body condition but showed increased winter mortality in the wild, are much better indicators of the fitness costs of compensatory growth.

Overall, the analyses show that intermittent feeding and clutch size manipulation are poor ways of conducting compensatory research. Manipulating clutch size seems to cause little restrictive effect on offspring and, therefore, does little to instigate compensatory growth (Alonso-Alvarez et al. 2006). It is possible that intermittent feeding is ineffective because the animals simply gorge on feeding days, a response known as hyperphagia (Ali & Wootton 2000). Recent research on the effect of dietary restriction on longevity found that whether animals were multigeneration laboratory dwellers or not had a significant impact on their response to the diet treatment (Nakagawa et al. in press). It is, therefore, surprising that whether animals were farmed or not had little impact on their potential for catch-up or compensatory growth. The heavy dependence of results on the experimental protocol used suggests that a call be made for unified restriction protocols, at least for similar species. Such protocols would allow the evolutionary context of ‘compensatory growth’ to be much more easily interpreted. Sogard & Olla (2002) give a good example of this strategy in their comparison of compensatory growth between sablefish, Anoplopoma fimbria, and walleye pollock, Theragra chalcogramma. Despite the similar environmental and ecological niches of these two pelagic fish species, the researchers were able to detect a marked difference in the mechanisms of compensatory growth and evolved baseline growth rates by exposing the fish to identical dietary regimes.

It is interesting that, on the whole, the linear analyses (Fig. 5b,d) are actually in agreement with the results derived from the complete slopes analyses (Fig. 5a,c). These results suggest that the impact of researchers not taking size-dependent growth into account when analysing compensatory growth is less influential when analysing a data set with such large statistical power. In smaller, empirical studies like the papers including in this meta-analysis, researchers need to be aware of this confounding factor, as Nicieza & Alvarez (2009) suggest. The conflict between results derived by comparison of the restricted group’s realimentation with part A and part B of the controls relates to size-dependent growth concerns (Fig. 1). Compensatory growth after restriction was faster than the rate of controls at the same time (B vs. B, Fig. 5c), but, importantly, it was no different than the rate observed in the controls during period A (A vs. B, Fig. 5a). Although this difference could be explained by size-dependent growth, the linear comparison makes the growth curve of the controls an unlikely explanation (Fig. 5d). Instead, it is possible that time is the important factor.

The rapid growth after restriction is faster than optimal growth for the age of the animals, but not faster than early on in life. The importance of age-dependent growth has been shown in response to the natural dietary restriction of alpine swift nestlings, Apus melba (Bize, Metcalfe & Roulin 2006). Chicks that were older during poor weather (and therefore poor nutrition) delayed fledging to compensate for wing growth, which had been sacrificed in place of mass during restriction. In contrast, younger chicks lost mass during undernutrition, but were able to reach the same size as controls at fledging by rapid weight gain. Bize, Metcalfe & Roulin (2006) explain the difference in strategies on the varying ‘developmental windows’ of tissues, organs and morphological traits, which in turn alter the stress resistance and tissue preservation hierarchy at different ages. The negative consequences of compensatory growth on fitness components are, therefore, due to rapid growth at an age when physiological restrictions inhibit growth. For example, one proposed mechanism for the later life health consequences of catch-up growth in humans is that the critical period for muscle growth is around 30 weeks in utero (Robinson & Barker 2002). As there is little cell replication after birth, any gain in mass is likely to give a disproportionately high body fat ratio, which leads to similar symptoms as in obesity, although an individual may not be obese. Surprisingly, the present study found that the age (or timing) that restriction was initiated had little effect on the size and growth rate outcomes.

When interpreting the outcome of our four-group interaction models, it is important to remember that the way in which we incorporated groups into the models is conducive to detecting trends within our larger study and reflects contrasts between the groups rather than outright evidence of catching up or not. Most of the heterogeneity in our analysis is derived from interstudy not interclass differences (Table 2), and I2 values were mostly over 90% in this additional analysis (Table S14, Supporting information). As expected, birds and mammals invest more in accelerating growth because of their determinate growth. The severity of the diet for mammals and the duration of restriction in birds were important moderators, reflecting their lower tolerance for reduced fat stores owing to their higher metabolic rate as endotherms (Wang, Hung & Randall 2006).

In contrast, fish seemed relatively unaffected by restriction, with little alteration to growth rate and no detectable consequences for fitness components. Fish have indeterminate growth and are therefore under less pressure than endotherms to rapidly achieve a large final size. Their lower metabolism, as ectotherms, also makes them more resistant to starvation and reduces the cost of depleting fat stores (Wang, Hung & Randall 2006). The saltatory growth of arthropods (‘leaping’ growth after each ecdysis) may account for arthropods being the group most likely catch-up growth in spite of there being no perceptible increase in growth rate as developmental time can be increased (Dmitriew & Rowe 2007). Dietary restriction results in a small larval instar, but if ecdysis is delayed until a threshold size is reached, it stands to reason that restricted animals will reach the same size as controls at the end (Mirth & Riddiford 2007). The cost may come in terms of seasonal timing; taking longer to reach the adult phenotype may result in previously restricted arthropods missing out on seasonal mating, peak food abundance or risking pond desiccation (De Block, McPeek & Stoks 2008). Taken together, these results suggest that what we refer to as ‘compensatory growth’ is best exemplified by mammals, while ‘catch-up growth’ best describes arthropods.

Our analysis has identified a number of influences at work in compensatory growth experiments, and we have noted some areas that could be improved in future. To truly understand whether it was the compensatory growth causing the fitness component deficits and not simply the early life experience of dietary restriction, it is necessary to compare fitness-related traits with permanently restricted animals (a treatment group that is almost always missing in empirical work). Alternatively, the correlation between growth rate and fitness deficits for each individual may disentangle the effects of restriction and compensatory growth (Krause & Naguib 2011). We also recommend that researchers have an established understanding of the study animal’s growth patterns to decide the length of the various treatment periods. The study should occur during the most linear phase of an animal’s growth pattern (if possible), aiming to end with the cessation of linear growth in the controls. This procedure eliminates the influence of size-dependent growth and allows comparisons between species. Further measurements after this period are encouraged as indicators of the impact of compensatory growth on fitness components. Duration and degree of restriction were very influential moderators, and we suggest researchers apply ecologically relevant levels of restriction to best represent likely fitness outcomes. We also suggest that if normal quantitative dietary restriction is not possible, researchers use qualitative restriction, such as diluting food with cornhusks, because this alternative method most closely resembled quantitative restriction (unlike intermittent feeding and clutch size manipulations).

In conclusion, these analyses voice strong support for the theoretical concept of compensatory growth as an evolved type of phenotypic plasticity common to many disparate animal taxa. Dependent on the degree of restriction and the period within which animals were restricted, both catch-up growth and compensatory growth are feasible outcomes following realimentation (Figs 4 and 5). As predicted, there is a cost to fitness components as a trade-off for the larger size, which is manifested as higher mortality and lower reproductive output. Further work is needed to disentangle how much of this cost is generated purely by faster than optimal growth and not by restriction. An additional contribution of the present study may be clarification of the terminology (Fig. S1, Supporting information; cf. Jobling 2010), such that the causes and consequences of ‘catch-up growth’ and ‘compensatory growth’ can be compared and contrasted without confusion. Our results confirm that compensatory growth is a measurable, repeatable and widespread response to early life dietary restriction. With the implementation of more standardized restriction protocols, there can be greater emphasis on uncovering the proximate mechanisms, as well as unravelling the ecological and evolutionary consequences, of compensatory growth across taxa.


The authors are grateful to A. Senior, E. Santos, M. Lagisz, I. Cleasby, J. M. Gaillard and one anonymous reviewer for providing helpful comments on the manuscript. K.H. was financially supported by the University of Otago Postgraduate Award and Publishing Bursary from the Graduate Research Committee. S.N. is supported by NRCGD and the Marsden Fund.