Temperature-size responses during ontogeny are independent of progenitors’ thermal environments

Experimental setup

Both species used in the experiment, Folsomia candida and Proisotoma minuta, were cultured for more than 1 year (encompassing 8–12 generations) in incubators set at either 15 °C or 20 °C under constant dark conditions and fed with dry yeast. Details of the animals that originated the cultures are provided in Marty et al. (2022). The mode of reproduction is mainly asexual (parthenogenetic) in F. candida and sexual in P. minuta. In the case of F. candida—whose life-history is much better known given its common use in ecotoxicological assays (Fountain & Hopkin, 2005)—lifetime reproductive output peaks at ~21 °C (Mallard et al., 2020), which can be regarded as an optimum temperature for the species.

We started the experiment by establishing populations of adult collembolans from several source cultures (i.e., progenitors; F0) in order to obtain offspring of the same age for the following phases of the experiment (henceforth referred as experimental cohorts; F1). For this purpose, we added nine adults of F. candida or 11 adults of P. minuta into each of 80 Petri dishes (90-mm diameter) with a moist substrate of plaster of Paris and activated charcoal (9:1 mixture), in addition to dry yeast as food source provided ad libitum (Supplemental Methods). The plates with F0 adults were placed in their respective temperatures (i.e., 15 °C or 20 °C; progenitors’ temperature) and monitored every day until recently laid eggs were detected. As soon as we detected at least one clutch, all adults were removed from the plates, and their body length was measured (at 10X magnification). We found that the mean adult body length varied between the F0 individuals issued from the 15 °C and 20 °C cultures in F. candida, as individuals from 20 °C were 7.6% larger (mean ± SE; 2,063 ± 28 µm at 15 °C, n = 175; 2,220 ± 27 µm at 20 °C, n = 178; P < 0.001; Table S1), while mean adult body length did not differ in P. minuta (1,354 ± 26 µm at 15 °C, n = 260; 1,302 ± 26 µm at 20 °C, n = 243; P = 0.158; Table S1). The larger body sizes in F. candida at 20 °C could be the result of intraspecific competition (Mallard et al., 2020), given that animal densities were not controlled at this stage and could vary among cultures. Afterwards, we put the plates with fresh eggs belonging to the F1 cohorts in one of the four experimental temperatures (15 °C, 20 °C, 25 °C and 30 °C) (Fig. 1). We draw attention to the fact that all F1 eggs laid in the same plate (coming from the same F0 progenitors) were brought together to the same experimental thermal regime. The thermal range spanned the lowest culturing temperature (15 °C) and temperatures known to be deleterious for collembolan fecundity (30 °C; Mallard et al., 2020; Martínez-De León et al., 2024). We established a total of 80 experimental units: 2 Collembola species × 2 progenitors’ temperatures × 4 treatment temperatures × 5 replicates. A schematic summary of the experimental design is displayed in Fig. 1. We found that the 30 °C experimental treatment induced severe stress in both species, hindering their development: no eggs of Folsomia candida managed to hatch at this temperature, and only few eggs of Proisotoma minuta hatched but did not survive until maturity (Fig. S1). Consequently, we assessed the thermal reaction norms of size (i.e., egg size and size at maturity) and development (i.e., egg and juvenile development) of F1 individuals across three experimental temperature treatments (15 °C, 20 °C or 25 °C), originating from progenitors reared at two temperatures (15 °C or 20 °C).

To measure egg development time, we monitored F1 eggs on a daily basis, and noted the first day in which animals hatched at each Petri dish. When all individuals hatched, we reduced the total number of individuals below 50 (mean ± SD; F. candida: 42.2 ± 10.4 individuals; P. minuta: 25.1 ± 13.7 individuals; Fig. S2). This was done to control for potential effects of varying intraspecific competition on body size, caused when animal densities differ greatly among experimental units (Mallard et al., 2020). We kept monitoring the hatchlings every day until at least one individual reached maturity, that is, when new eggs were detected in the plate for the first time (i.e., progeny from the experimental cohorts; F2). At this point, we collected and counted all collembolans from the plates, stored them in ethanol (70%), and measured their body length (at 10× magnification). No assessment of maturity at the individual level was possible since collembolans are ametabolous invertebrates with indeterminate growth, which means that there are no apparent morphological differences across life stages besides body sizes (Stam, van de Leemkule & Ernsting, 1996). Later, we took pictures of egg clutches (both from the experimental cohorts and their progeny) at 150X as soon as eggs were detected, which were later counted, and their diameter was assessed to estimate egg size. All size measurements were performed with KEYENCE VHX-970F (Keyence, Osaka, Japan). We assessed body size and development using individuals within cohorts instead of individually raised animals, as typically done in life-history experiments (e.g., Stam, van de Leemkule & Ernsting, 1996). Our approach of using cohorts instead of individuals allowed us to understand the underlying mechanisms causing warming-driven body size reductions at the population level in Collembola, as reported in previous studies (e.g., Thakur et al., 2017; Robinson et al., 2018; Mallard et al., 2020). In the present study, using cohorts removes the effect of size structure on mean population body sizes, and thereby allows us to focus on size-at-stage responses (Ohlberger, 2013). In addition, these collembolans thrive when raised in groups (e.g., Sengupta, Ergon & Leinaas, 2017), particularly in the case of the sexually reproducing P. minuta. Nevertheless, individuals within cohorts are expected to display substantial variation in development and body sizes, and thereby yield greater uncertainty on the contribution of each individual on the estimated (population-level) trait values, compared to measurements made on isolated individuals.

Statistical analyses

We modelled sizes (i.e., egg diameter and body length at maturity) and development (i.e., time spent in the transition between life stages) as a function of the progenitors’ temperature (15 °C or 20 °C), the experimental temperature of the cohorts (15 °C, 20 °C, 25 °C or 30 °C) and Collembola species (F. candida or P. minuta). Given that trait plasticity in response to the experimental temperatures could be modulated by the progenitors’ temperature and the species, we initially tested three-way interactions as progenitors’ temperature × experimental temperature × species. Interaction terms and main effects were assessed a priori by means of data visualization (following Zuur, Ieno & Elphick, 2010), and selected or removed from the models based on AIC and their statistical significance. Final models obtained after simplifying the initial three-way interaction models are provided in Table 1. Given that size variables were measured at the individual level within cohorts, we employed linear mixed effects models (R package lme4, v.1.1-26; Bates et al., 2015) using experimental unit (i.e., Petri dish containing a single cohort) as a random intercept to account for the dependency among individuals from the same cohort. Developmental variables were measured at the Petri dish level (i.e., first day in which a life-stage transition was recorded), and were therefore analyzed using linear models. Furthermore, given that temperature-size responses are mainly revealed within the thermal range allowing for positive population growth (Walczyńska, Kiełbasa & Sobczyk, 2016; Blanckenhorn et al., 2021), we obtained a coarse estimate of the thermal performance curve of fecundity by modelling the mean egg production at the first reproductive event. In this case, we employed linear models using per capita egg production (i.e., total number of eggs averaged by the number of adults in a plate) as a response variable. Using the same model selection approach as above, we additionally included mean body size at the Petri dish level as an additive term to account for potential effects of body size on fecundity at the first reproductive event (Liefting et al., 2015; Tully, 2023). However, one important limitation of our estimate of fecundity is that it contains information on both the number of F1 individuals that have reached maturity at the time of observation, as well as the number of eggs laid by each individual (actual fecundity), which warrants caution in the interpretation of these results. Linearity assumptions of all linear models were tested and visually inspected using the package DHARMa (Hartig, 2022).

Body length at maturity of the experimental cohorts was measured on the same age in all individuals of a given cohort, but it might be the case that only the largest individuals of the cohort reached maturity (e.g., due to variation in the pace of development among the individuals of the cohort). In this case, the highest quantiles of the cohort’s body size distribution are a better proxy of size at maturity than the mean cohort body size. We therefore tested changes in body size at the quantiles 0.75, 0.85, and 0.95 of the distribution using linear quantile mixed models (R package lqmm, v.1.5.6; Geraci, 2014), with experimental unit as a random intercept. To simplify the models and enable proper estimation of model parameters, we fitted separate models by species, using centered and scaled body size values. Testing interactions between progenitors’ and experimental temperature precluded the adequate estimation of model parameters, so only main effects were tested in this particular analysis. Final models were selected based on AIC and statistical significance, similarly to the more conventional linear models described above. We performed post-hoc tests to obtain all p-values using the function emmeans (R package emmeans; v.1.7.0; Lenth, 2024). All statistical analyses were carried out in R statistical software (v.4.0.2; R Core Team, 2023).

Transgenerational plasticity did not constrain temperature-size responses

Our findings generally indicate that the progenitors’ thermal environments did not modulate plasticity in body size and development in the following generation. This is consistent with the notion that temperature-size responses are reset at the start of each generation (Forster, Hirst & Atkinson, 2011; Forster & Hirst, 2012), which means that temperature-driven transgenerational plasticity may have minor importance in determining size-at-stage and development in the two Collembola species examined (but see Walczyńska et al., 2015). However, we found that traits concerning the embryonic phase (egg development in F1 and egg size in F2) were, in some cases, affected by the temperature experienced by the previous generation. We found that egg development of the cohorts (F1) was ~16% shorter at 15 °C in P. minuta when eggs were laid by progenitors in warmer environments (20 °C compared to 15 °C). These eggs performed better (i.e., shorter development) despite the mismatch between the thermal environments of progenitors and their offspring, which would show support for transgenerational plasticity based on condition-transfer effects (Bonduriansky & Crean, 2018; Bonduriansky, 2021). Interestingly, the effect of the progenitors’ temperature on egg development faded out when eggs were developed at higher temperatures and were not even detected in F. candida for any of the treatment combinations. Several factors (e.g., egg provisioning or epigenetic factors in the gametes) could be responsible for this variation in transgenerational plasticity and, therefore, provide additional support for the occurrence of condition-transfer effects (Bonduriansky & Crean, 2018). In this regard, we show that egg size in F1, a proxy of egg provisioning (Geister et al., 2009), did not differ between progenitors’ temperatures, which implies that other mechanisms must be involved if condition-transfer effects were to explain our findings. Even though the eggs laid at the higher temperature may have started to develop more quickly before they were transferred to the colder environment (e.g., Geister et al., 2009), such an experimental artifact is unlikely to fully explain our findings. In particular, the magnitude of developmental shortening (−2.20 days) was much greater than the time lag from oviposition to the experimental transfer of eggs (few hours to less than a day). Condition-transfer effects in the context of the temperature-size rule were also suggested by Walzer, Formayer & Tixier (2020), who showed that predatory mites (Amblydromalus limonicus) developed consistently faster when their progenitors were exposed to heat wave conditions. In addition, they reported larger sizes at maturity (following a reverse temperature-size pattern) when offspring matched the thermal environments of their progenitors, indicating the occurrence of anticipatory transgenerational plasticity (Walzer, Formayer & Tixier, 2020).

In the case of egg size in F2, we show that eggs laid at higher temperatures (25 °C) were ~7% larger in P. minuta. We speculate that this is likely the result of greater maternal provisioning of eggs at 25 °C (Pettersen et al., 2019). Developmental costs increase greatly beyond a certain high temperature threshold, so mothers should increase egg sizes to provide enough energy to their offspring to fuel the higher energetic costs (Pettersen et al., 2019). However, why this is the case for P. minuta but not for F. candida, where egg sizes remained similar across temperatures, is unclear. This difference is especially intriguing considering the known plasticity of egg sizes in the latter species (Stam, van de Leemkule & Ernsting, 1996; Tully & Ferrière, 2008; Marty et al., 2022). As opposed to our results, Liefting et al. (2010) reported smaller egg sizes at 20 °C than at 16 °C in the collembolan Orchesella cincta, conforming to the temperature-size rule. Likewise, previous work using the same study system revealed that egg sizes declined from 15 °C to 20 °C in F. candida due to the existence of trade-offs with clutch sizes in the cooler environment (i.e., larger eggs in smaller clutches), while no trade-offs were detected in warmer conditions (Marty et al., 2022). Such inconsistency between these findings and the present results (i.e., no change in egg sizes from 15 °C to 20 °C) might be explained by the trade-offs between egg size and clutch size, which were not accounted for in the present study. Alternatively, we speculate that differences in egg sizes between 15 °C and 20 °C in Marty et al. (2022) might be a transient response due to the shorter acclimation time of collembolan cultures (~6 months), compared to the present study (~1 year). Overall, we showed that egg sizes mostly remained unaltered across different thermal environments, confirming our expectation that eggs typically display weak temperature-size responses.

Responses of body size at maturity affected by growth and development

We found that development generally differed strongly at various thermal regimes in both Collembola species, whereas body size at maturity responded more weakly or not at all. Size at maturity is ultimately determined by both developmental and growth rates, so weak responses of size at maturity suggest a strong coupling between both rates across temperatures in these collembolans. Sengupta, Ergon & Leinaas (2017) reported similar thermal reaction norms of development and growth in Folsomia quadrioculata, resulting in a lack of response of size at maturity across temperatures. In contrast, asymptotic body sizes (i.e., maximum body size in organisms with indeterminate growth) declined at warmer temperatures following the temperature-size rule, as also shown by Mallard et al. (2020) in Folsomia candida. Altogether, these results indicate that size at maturity and asymptotic size might be shaped by different mechanisms, supporting the idea that no singular mechanistic explanation exists for the temperature-size rule across ontogeny (Hoefnagel et al., 2018; Verberk et al., 2021). Given that size at maturity and asymptotic size can display different responses to temperature (Loisel, Isla & Daufresne, 2019), we suspect it to explain the discrepancy between the findings by Thakur et al. (2017) and the results from this study, namely the difference in temperature-induced (adult) body size responses. In the previous study, mean adult body sizes were substantially smaller in warmer environments in F. candida but remained unaltered in P. minuta (Thakur et al., 2017). Given that size measurements were done on adults from populations established during several weeks (i.e., much longer than age at maturity), it is likely that a large proportion of the measured adults were close to their asymptotic body sizes, reaching values more similar to the asymptotic body sizes reported by Mallard et al. (2020) than to the sizes at maturity shown in the present study.

In addition, our results obtained with quantile regressions showed that individuals of F. candida belonging to the highest quantiles (i.e., largest individuals of the cohorts) attained smaller sizes in warmer environments, whereas in Proisotoma minuta a similar response was found for lower quantiles and average body sizes. While it is unclear why temperature affected distinct aspects of the body size distribution of the cohorts between these species, we suspect that differences in the strength of intraspecific competition between species might explain this inconsistency. Strong intraspecific competition has been reported in F. candida, in which few competitively superior individuals monopolize resources through interference (Le Bourlot, Tully & Claessen, 2014), thereby reducing mean body growth rates at the population level (Mallard et al., 2020). Such interference competition could induce great variation in the progress of development within cohorts, causing more competitive individuals to grow quicker and reach size at maturity earlier than the rest of the individuals in the cohort (Le Bourlot, Tully & Claessen, 2014). This would make body sizes in larger competitive individuals to be more regulated by temperature, while growth in smaller individuals from the cohorts would be more affected by resource acquisition (Mallard et al., 2020). Remarkably, Stam, van de Leemkule & Ernsting (1996) reported that individuals of F. candida developed in isolation reached smaller sizes at maturity in warmer environments, matching our results from the highest quantiles of the cohorts. As opposed to F. candida, evidence on the strength and the type of intraspecific competition in P. minuta is still lacking. We speculate that weaker interference competition in P. minuta could induce more synchronous development within cohorts (Le Bourlot, Tully & Claessen, 2014), which could explain why lower quantiles of the size distribution (e.g., quantile 0.75) and average body sizes were more responsive to warming. Furthermore, the range of body sizes at maturity in the cohorts was narrower in P. minuta (mean range across experimental temperatures ± SD: 720.3 ± 106.3 µm) than in F. candida (981.3 ± 162.8 µm; see Fig. 3), which could provide some support for the hypothesis of more synchronous development within our cohorts of P. minuta.

Thermal reaction norms differed among traits and between species

Even though the underlying mechanisms driving trait plasticity can differ across ontogenetic stages, it is plausible that similar patterns might have emerged in the two closely related species. However, this was not the case in the species that we used in this study, confirming previous work showing that these species had distinct body size responses to temperature (Thakur et al., 2017; Marty et al., 2022). Here, we extended on these studies by measuring development and size-related traits in cohorts exposed to various thermal environments. It has been previously showed that Folsomia candida consistently displayed greater plasticity in warmer environments than Proisotoma minuta for adult sizes (Thakur et al., 2017) and egg sizes (Marty et al., 2022). An explanation based on the mode of reproduction, which differs between F. candida (asexual) and P. minuta (sexual), could possibly clarify the distinct degree of plasticity displayed in both species. Species typically reproducing by parthenogenesis (asexual reproduction) may rely heavily on phenotypic plasticity to cope with environmental changes, whereas sexually reproducing species could display less plasticity and instead rely more on genetic variation (Ponge, 2020). In the context of the temperature-size rule, it follows that parthenogenetic species would display stronger size responses than sexually reproducing species. Yet, in the present study we found that the sexually reproducing P. minuta showed greater plasticity than F. candida in some traits (e.g., egg size in F2), supporting the view that phenotypic plasticity needs to be assessed simultaneously for several traits given that plasticity of one trait might be associated with robustness in another trait (Liefting et al., 2015). Therefore, the idea that the degree of plasticity can be generally characterized at the species level, based on the mode of reproduction, seems unlikely according to our results. We also note that considerable genetic variation can evolve in parthenogenetic species in the form of distinct lineages (as in the case of F. candida), and therefore display substantial variation in trait values at the species level (e.g., asymptotic body size, egg size; Tully & Ferrière, 2008; Mallard, Farina & Tully, 2015).

Another aspect of our between-species comparison is that other traits varying between F. candida and P. minuta could additionally affect their body and egg size responses, for instance the differences in their body sizes (smaller terrestrial species -in our case, P. minuta- are more likely to reduce their body size in warmer environments; Horne, Hirst & Atkinson, 2015) or their thermal performance curves of fitness metrics (e.g., population growth rates; Walczyńska, Kiełbasa & Sobczyk, 2016; Blanckenhorn et al., 2021). Nevertheless, our estimate of fecundity (a major component determining population growth) did not change across thermal environments in P. minuta, while it increased strongly from 15 °C to 25 °C in F. candida. This suggests that the lack of size responses in F. candida cannot be explained by experimental temperatures being beyond its “optimal thermal range” (Walczyńska, Kiełbasa & Sobczyk, 2016). These findings are expected to be robust despite the limitations associated with our estimate of fecundity (i.e., combining information on both the number of individuals reaching maturity and their reproductive output), and are further supported by previous studies assessing thermal reaction norms of fecundity in F. candida (Mallard et al., 2020).