Evolutionary adaptation to juvenile malnutrition impacts adult metabolism and impairs adult fitness in Drosophila

Version of Record

Accepted for publication after peer review and revision.

Download
Cite
Share
CommentOpen annotations (there are currently 0 annotations on this page).

405 views

Version of Record published: November 10, 2023 (This version)
Accepted Manuscript published: October 17, 2023 (Go to version)
Accepted: October 12, 2023
Received: September 3, 2023
Preprint posted: January 12, 2022 (Go to version)

1. Of interest
Down the Penrose stairs, or how selection for fewer recombination hotspots maintains their existence

Zachary Baker, Molly Przeworski, Guy Sella

Research Article Dec 6, 2023
Further reading

Abstract
Editor's evaluation
Introduction
Results
Discussion
Materials and methods
Data availability
References
Decision letter
Author response
Article and author information
Metrics

Abstract

Juvenile undernutrition has lasting effects on adult metabolism of the affected individuals, but it is unclear how adult physiology is shaped over evolutionary time by natural selection driven by juvenile undernutrition. We combined RNAseq, targeted metabolomics, and genomics to study the consequences of evolution under juvenile undernutrition for metabolism of reproductively active adult females of Drosophila melanogaster. Compared to Control populations maintained on standard diet, Selected populations maintained for over 230 generations on a nutrient-poor larval diet evolved major changes in adult gene expression and metabolite abundance, in particular affecting amino acid and purine metabolism. The evolved differences in adult gene expression and metabolite abundance between Selected and Control populations were positively correlated with the corresponding differences previously reported for Selected versus Control larvae. This implies that genetic variants affect both stages similarly. Even when well fed, the metabolic profile of Selected flies resembled that of flies subject to starvation. Finally, Selected flies had lower reproductive output than Controls even when both were raised under the conditions under which the Selected populations evolved. These results imply that evolutionary adaptation to juvenile undernutrition has large pleiotropic consequences for adult metabolism, and that they are costly rather than adaptive for adult fitness. Thus, juvenile and adult metabolism do not appear to evolve independently from each other even in a holometabolous species where the two life stages are separated by a complete metamorphosis.

Editor's evaluation

This important study pushes forward the understanding of a major question in evolutionary biology and human health: does juvenile malnutrition affect the performance of the adult? Combining experimental evolution in Drosophila with transcriptomics, metabolomics, and genomics datasets, it provides solid and compelling evidence that adult adaptation is constrained by adaptation at the larval stage. The work will be of interest to a broad audience including evolutionary biologists as well as human health researchers.

https://doi.org/10.7554/eLife.92465.sa0

Introduction

Nutrient shortage is an important and widespread ecological stressor, and a driver of natural selection on animal physiology. Juveniles are often the first to suffer from undernutrition as they are usually competitively inferior to adults. Furthermore, in most species, juveniles have no option to arrest their development and wait for better times; they are forced to continue development despite famine. Prolonged nutrient shortage at a juvenile stage usually results in stunted adult size, but also often has profound effects on adult physiology that do not seem mediated by body size. They include changes in epigenetic marks (Lillycrop and Burdge, 2015), gene expression (May and Zwaan, 2017; Szostaczuk et al., 2020), metabolome (Agnoux et al., 2014), triglyceride accumulation (Lukaszewski et al., 2013; Klepsatel et al., 2020), blood pressure (de Brito Alves et al., 2014), behavior (Akitake et al., 2015), and longevity (Davis et al., 2016). In humans, prenatal undernutrition is correlated with negative consequences for diverse aspects of metabolic function and health at old age (Roseboom et al., 2006; de Rooij et al., 2010).

These physiological responses to developmental nutritional conditions are a form of phenotypic plasticity; that is, a change of phenotype induced by differences in the environment, with no change in genome sequence (Scheiner, 1993; Bateson et al., 2004). We know much less about whether and how adult physiology and metabolism evolve genetically over generations in response to natural selection driven by repeated exposure of the population to juvenile undernutrition. Some authors postulate that phenotypic plasticity in response to novel environmental conditions is usually adaptive, and that evolutionary change will mostly reinforce these initially plastic phenotypes (‘genetic assimilation’) (Baldwin, 1896; Pigliucci and Murren, 2003). Thus, plasticity would predict evolution. However, this prediction does not appear borne out by empirical data. In particular, experimental studies that compared plastic and evolutionary responses of gene expression to novel environments usually found little overlap between genes involved in the two responses; for those that did overlap, evolutionary change tended to counter rather than be in line with plasticity (Yampolsky et al., 2012; Ghalambor et al., 2015; Huang et al., 2016) for an exception, see Josephs et al., 2021. Thus, to understand how adult physiology is molded by exposure to juvenile undernutrition over evolutionary time, inferences based on phenotypically plastic responses should be complemented by the study of genetically based adaptations.

The particular vulnerability of juveniles to famine also raises the question of the degree to which metabolism and physiology of different life stages can evolve independently if they face different nutritional conditions. Anurans and holometabolous insects demonstrate that in the long run evolution can generate spectacular differences in morphology and physiology between juvenile and adult stages. This has led some authors to argue that metamorphosis of holometabolous insects allows the larval and adult phenotypes to be evolutionarily decoupled from each other (Moran, 1994; Rolff et al., 2019). This postulate requires a sufficient supply of genetic variants that affect relevant phenotypes in a stage-specific manner. However, surprisingly little is known to what degree this is the case, whether in holometabolous insects or other animals (Collet and Fellous, 2019). For example, while some enzymes in Drosophila have distinct larval and adult forms encoded by different genes, most metabolic genes are expressed at both stages (https://flybase.org/). Thus, independent evolution of larval and adult metabolism would require independent regulation of expression of the same genes. It may be more parsimonious to expect that most regulatory genetic variants would tend to affect gene expression across lifetime, constraining independent evolution of the juvenile and adult stages. If so, response to selection driven by juvenile undernutrition would lead to similar physiological change in larvae and adults, and these changes might be disadvantageous for adult performance, in particular under conditions of plentiful food.

Experimental data to address the above questions are scarce. Experimental evolution studies on Drosophila demonstrated that adaptation to poor larval nutrition (nutrient-poor diet or larval crowding) may be associated with changes in adult traits such as fecundity (May et al., 2019), lifespan (Shenoi et al., 2016; May et al., 2019), starvation resistance (Kawecki et al., 2021), pathogen resistance (Vijendravarma et al., 2015), or heat tolerance (Kapila et al., 2021). Changes in gene expression and metabolism underlying these correlated responses of adult phenotypes to selection for juvenile or larval undernutrition tolerance have not been explored. Insights into if and how selection driven by juvenile undernutrition shapes adult metabolism would shed light on the importance of evolutionary forces acting on development for understanding adult physiology, with potential implications for human health (Neel, 1962; Prentice et al., 2005).

Here, we use experimental evolution to study the consequences of evolutionary adaptation to larval undernutrition for gene expression, metabolite abundance, and reproductive performance of adult Drosophila melanogaster females. We compare six replicate populations (the ‘Selected’ populations) evolved for over 230 generations on a very nutrient-poor larval diet, to six populations of the same origin maintained on a moderately nutrient-rich standard larval diet (the ‘Control’ populations). Importantly, in the course of experimental evolution, flies of both sets of populations were always transferred to fresh standard diet soon after emergence, maintained on this diet for 4–6 d, and additionally fed ad libitum live yeast before reproduction. Thus, every generation of Selected populations experienced a switch from the poor larval diet to the standard adult diet. Compared to the Controls, larvae of the Selected populations evolved higher viability, faster development and faster growth on the poor diet (Kolss et al., 2009; Vijendravarma and Kawecki, 2013; Cavigliasso et al., 2020). The Selected larvae appear to invest more in protein digestion (Erkosar et al., 2017), are more efficient in extracting amino acids from the poor diet (Cavigliasso et al., 2020), and are able to initiate metamorphosis at a smaller size (Vijendravarma et al., 2012), traits that likely contribute to their improved performance on the poor diet. The Selected populations also evolved reduced dependence on microbiota for development on poor diet (Erkosar et al., 2017), However, they show greater susceptibility to an intestinal pathogen (Vijendravarma et al., 2015) and are less resistant than Controls to starvation at the adult stage (Kawecki et al., 2021). Finally, the Selected and Control larvae also show divergence in metabolite abundance consistent with evolutionary changes in amino acid metabolism, with lower free amino acid concentrations and a higher rate of amino acid catabolism (Cavigliasso et al., 2023). Evolutionary divergence between Selected and Control populations is highly polygenic, involving over 100 genomic regions enriched in hormonal and metabolic genes (Kawecki et al., 2021). Thus, evolutionary adaptation to larval undernutrition in our experimental populations involved a complex suite of genomic and phenotypic changes.

Motivated by the general questions discussed above, in this study we address the following specific questions about the evolutionary divergence between the Selected and Control populations at the adult stage.

First, have the Selected and Control populations evolved major genetically based differences in adult gene expression and metabolite abundance, implying that evolutionary adaptation to larval undernutrition has significant consequences for the physiology of the adult stage? Which pathways have been affected?

Second, in view of the controversy about the relationship between plasticity and evolution summarized above, to what degree do these genetically based evolutionary changes in adult gene expression and metabolome resemble the phenotypically plastic responses to poor larval diet?

Third, are differences in adult gene expression and metabolite abundance between Selected and Control populations similar to those between Selected and Control larvae (reported elsewhere; Erkosar et al., 2017; Cavigliasso et al., 2023)? Such a similarity would be expected either if both life stages were exposed to similar selection pressures or if genetic variants favored by larval undernutrition had pleiotropic effects on gene expression at the two stages. As we argue in ‘Discussion,’ we find the latter interpretation more parsimonious, even it contradicts the notion that larval and adult gene expression patterns can be decoupled from each other in holometabolous insect (Moran, 1994; Rolff et al., 2019).

Fourth, do the metabolic profiles of Selected and Control flies differ along a similar axis as the profiles of fed versus starving flies? Selected flies are less resistant to starvation than Controls and the genomic architecture of divergence between them shares many genes with starvation resistance (Kawecki et al., 2021). This question addresses the link between these two findings, and the answer would aid in functional interpretation of metabolic changes evolved by the Selected populations.

Finally, do adult flies of the Selected populations perform better in terms of fitness than Controls when both are raised on the poor larval diet? This would be expected if most changes in their adult physiology evolved because they improve the performance of adults raised on poor diet. In contrast, if the metabolic changes observed at the adult stage primarily reflect pleiotropic effects of genetic variants that improve larval growth and survival on the poor diet, Selected adults should not have an advantage or even be inferior to Controls even when raised under the conditions under which the former evolved.

Results

Gene expression patterns point to evolutionary changes in adult metabolism

To explore plastic and evolutionary responses of adult gene expression to poor diet, we performed RNAseq on 4-day-old adult females from the Selected and Control populations raised on either larval diet. In this factorial design, the effect of larval diet treatment (poor versus standard) on the mean expression level is a measure of the plastic response. The difference between Selected and Control populations assessed in flies raised on the same diet is a measure of genetically-based evolutionary change (Figure 1A). Potential differences of the plastic responses between Selected and Control populations or the influence of diet on the expression of evolutionary divergence would be reflected in the evolutionary regime × diet interaction.

Figure 1 with 1 supplement see all

Download asset Open asset

Phenotypic plasticity and genetically based evolutionary change in response to larval undernutrition both lead to divergence in gene expression patterns of adult *Drosophila* females.

(A) Design of the experimental evolution and of the gene expression assay. The purpose of ‘relaxation’ was to eliminate the potential effects of (grand)parental environment. The purple arrows indicate the two main factors in the analysis and interpretation, the evolutionary history (evolutionary change), and the effect of larval diet on which the current generation was raised (plasticity). (B) Sample score plot of the fourth and fifth principal components on the expression levels of 8701 genes of flies from Selected (SEL) and Control (CTL) populations raised on poor and standard (std) larval diet. Each point represents a replicate population × diet combination (Figure 1—source data 1). For complete plots of PC1-6, see Figure 1—figure supplement 1. (C) Biological process Gene Ontology (GO) terms enriched (at p < 0.01) in genes differentially expressed between Selected and Control populations, with the number of genes upregulated (red) and downregulated (blue) in Selected populations. (D) GO terms enriched at p < 0.01 among genes showing a plastic response to diet. For details of the GO term enrichment results, see Supplementary file 2.

Figure 1—source data 1 Principal component analysis of adult gene expression: the first six principal component scores of all populations on each diet.: https://cdn.elifesciences.org/articles/92465/elife-92465-fig1-data1-v2.xlsx
Download elife-92465-fig1-data1-v2.xlsx

Prior to collection for RNAseq, the flies were transferred to standard diet upon emergence from the pupa and maintained in mixed-sex group; thus, they were mated and reproductively active. We focused on female abdominal fat body, the key metabolic organ combining the functions of mammalian liver and adipose tissue. It is in the fat body where metabolic reserves of glycogen and triglycerides are stored and mobilized, and where dietary nutrients are converted into the proteins and lipids subsequently transported to the ovaries for egg production (Li et al., 2019). Because a clean dissection of adult fat body is difficult, we performed RNAseq on the ‘carcass,’ consisting of the fat body together with the attached abdominal body wall and any embedded neurons, but excluding the digestive, reproductive, and most of the excretory organs.

Based on the estimation of true nulls (Storey and Tibshirani, 2003), the evolutionary regime (i.e., Selected versus Control) affected adult expression of about 26% of all 8701 genes included in the analysis while the larval diet affected about 45%. Allowing for 10% false discovery rate (FDR), we identified 219 genes as differentially expressed between Selected and Control populations and 827 between flies raised on standard or poor diet (Supplementary file 1). The magnitude of the changes in gene expression was moderate, with the significant genes showing on average about 1.76-fold difference (in either direction) between Selected and Control populations and about 1.30-fold difference between flies raised on poor versus standard diet. While the first three principal component axes appear driven by idiosyncratic variation among replicate populations and individual samples, PC4 and PC5 clearly differentiate the evolutionary regimes or larval diets (Figure 1B, Figure 1—figure supplement 1). The divergence of gene expression patterns between Selected and Control populations, as well as between flies raised on poor and standard diet, was supported by multivariate analysis of variance (MANOVA) on the first five principal components, jointly accounting for 67% of variance (Wilks' λ = 0.052, F_6,5 = 15.1, p = 0.0046 for regime and Wilks' λ = 0.057, F_6,5 = 13.9, p = 0.0055 for diet). We detected no interaction between the evolutionary regime and current diet either in the PCA-MANOVA (Wilks' λ = 0.59, F_6,5 = 0.6, p = 0.41) or in the gene-by-gene analysis (lowest q = 99.9%). Thus, both the evolutionary adaptation and the phenotypically plastic response to larval diet experienced had substantial effects on adult gene expression patterns; however, these effects appeared largely additive (i.e., independent of each other).

The strongest statistical signal from Gene Ontology (GO) enrichment analysis on genes differentially expressed between Selected and Control flies points to genes involved in cuticle development and maturation, including several structural cuticle proteins (Figure 1C, Supplementary file 2, table A). Most of these genes are mainly expressed in epidermis (including in the tracheal system), and nearly all show higher expression in the Selected compared to Control populations. This may reflect differences in the relative amount of the epidermal tissues or in the rate of cuticle maturation. Several other enriched GO terms (phagocytosis, pigment metabolic process, iron ion homeostasis, response to fungus) include multiple genes expressed mainly in hemocytes; these genes are downregulated in Selected flies, suggesting lower abundance of hemocytes.

The other top enriched GO terms point to metabolic processes, notably oxidation/reduction, biosynthesis of purine compounds, and amino acid catabolism (Figure 1C). Differentially expressed genes involved in amino acid catabolism included enzymes involved in catabolism of arginine (Oat), serine (Shmt), branched-chain amino acids (CG17896, CG6638, Dbct), tryptophan (Trh), asparagine (CG7860), and tyrosine (CG1461). Thus, the signal of enrichment in amino acid metabolism was based on multiple differentially expressed genes distributed over multiple branches of amino acid metabolism. A few more genes involved in amino acid catabolism overexpressed in the Selected populations link the aspartate–glutamate metabolism with purines synthesis pathway, which also showed a strong signal of enrichment in differentially expressed genes, and which we examine in more detail below.

Metabolome analysis points to evolutionary changes in amino acid and purine metabolism

Differences in gene expression between Selected and Control flies suggested that evolutionary adaptation to larval diet affected adult metabolism. We therefore used a broad-scale targeted metabolomics approach to measure the abundance of key polar metabolites involved in multiple core pathways in central metabolism. As for gene expression, the analysis was carried out on the fat body with some adjacent body wall and neural tissue (the ‘carcass’) of 4-day-old females from Selected and Control populations, each raised on either poor or standard diet (additionally, we included a starvation treatment that is analyzed in a later section). After filtering for data quality, we retained 113 metabolites that were quantified (normalized to protein content) in all samples.

The first two principal components extracted from the metabolome data clearly differentiated the Selected versus Control populations, as well as flies raised on poor versus standard diet (Figure 2A; MANOVA on PC1 and PC2 scores, Wilks' λ = 0.185, F_2,9 = 19.8, p = 0.0005 and Wilks' λ = 0.055, F_2,9 = 77.3, p < 0.0001, for regime and diet, respectively). Of the 113 metabolites, 57 were found to be differentially abundant between Selected and Control flies at 10% FDR, and also 57 differentially abundant between flies raised on poor versus standard diet (Supplementary file 3). We found no interaction between these two factors either in the univariate analysis (no metabolite passing 10% FDR, lowest q = 0.39) or in the principal component analysis (PCA) (MANOVA, Wilks' λ = 0.918, F_2,9 = 0.4, p = 0.68). Thus, both the evolutionary history of exposure to poor versus standard diet and the within-generation diet treatment had major effects on the metabolome, but, as for gene expression, these effects were largely additive. Therefore, we present the results as a heat map corresponding to the effects of these two factors on abundance of individual metabolites (the first two columns of heat maps in Figure 2B; the third column corresponds to the effect of starvation discussed in a later section).

Figure 2

Download asset Open asset

The effects of phenotypic plasticity and evolutionary change on adult metabolite abundance.

(A) Principal component score plot based on metabolic signatures composed of 113 robustly quantified metabolites in flies from the six Selected and six Control populations raised on both poor and standard diet (Figure 2—source data 1). The flies were in fed condition. Metabolite abundances for two replicate samples were averaged before the principal component analysis (PCA); thus, there is one point per population and diet. The ‘C2’ label indicates control population 2 on either diet; this population deviated from the other Control populations along the PC2 axis; however, we retained it for the univariate analysis. (B) The effects of experimental factors on abundance of individual metabolites. The first two columns of the heat maps indicate, respectively, the effect of evolutionary regime (least square mean difference Selected – Control) and the within-generation (phenotypically plastic) effect of larval diet (least square mean difference poor – standard diet) on the relative concentration of metabolites in fed flies. The third column indicates the main effect of the starvation treatment (least square mean difference between starved and fed flies) on metabolite concentrations in flies from both Selection regimes raised on both diets. Thus, red (blue) indicates that a compound is more (less) abundant in Selected, poor diet-raised and starved flies than in Control, standard diet-raised and fed flies, respectively. Asterisks indicate effects significant at q < 0.1. Annotations to the right of the heat maps indicate interactions significant at q < 0.1: E × S = evolutionary regime × starvation treatment, D × S = larval diet × starvation treatment, E × D × S = three-way interaction; (e) indicates an essential amino acid, C2–C18 for acylcarnitines refers to the length of the acyl chain. For estimates, effects, and statistical tests underlying the figure, see Supplementary file 3; original data are in Supplementary file 8.

Figure 2—source data 1 Principal component analysis of adult metabolite abundance in fed flies: the first four principal component scores of all populations.: https://cdn.elifesciences.org/articles/92465/elife-92465-fig2-data1-v2.xlsx
Download elife-92465-fig2-data1-v2.xlsx

Evolutionary adaptation to larval malnutrition resulted in a striking shift in proteinogenic amino acid concentrations in adult flies: eight essential amino acids were less abundant in the Selected than Control flies, while the reverse was the case for four non-essential ones (Figure 2Bi). Multiple purine compounds were less abundant in the Selected than Control flies, including both purine nucleobases and nucleosides, and the electron carriers NAD, NADP, and FAD; an exception is ADP, which was more abundant in the Selected flies (Figure 2Bii; ATP was not reliably quantified). No such general trend for reduced abundance was observed for pyrimidines (Figure 2Biii), suggesting that the pattern observed for purines is not mediated by changes in synthesis or degradation of nucleic acids.

Selected flies had lower levels of trehalose (the principal sugar circulating in hemolymph) and of lactate than Controls (Figure 2Biv), but higher concentrations of compounds linking glycerol with glycolysis, with no indication of differences in concentration of upstream and downstream glycolysis intermediates (Figure 2Biv). This suggests changes in fatty acid/triglyceride metabolism. While fatty acids were not included in the targeted metabolome analysis, we did quantify the abundance of acyl-carnitines, that is, fatty acid residues attached to carnitine, the molecule that transports them in and out of mitochondria. Compared to the Controls, the Selected flies show accumulation of medium chain (C6–C12) acyl-carnitines (Figure 2Bv). In contrast, acyl-carnitines with long-chain (C16–C18) residues, which are typical for insect triglycerides (Stanley-Samuelson et al., 1988), were less abundant in the Selected than Control flies (Figure 2Bv).

A number of other metabolites showed differential abundance between Selected and Control flies (Figure 2Bvi). More than half of them are derivatives of amino acids, including intermediates in synthesis of NAD+ (kynurenine and kynurenate) and cysteine (cystathionine), neurotransmitters (GABA, N-methylaspartate [NMDA] and N-acetylaspartate [NAA]), and products of amino acid catabolism that do not appear to be functionally linked. Together with increased abundance of isovalerylcarnitine (which carries a residue of deamination of branched-chain amino acids; Figure 2Bv), these results reinforce the notion that adaptation to the poor larval diet has been associated with wide-ranging changes in amino acid metabolism.

In an attempt to integrate metabolite, gene expression, and genome sequence changes associated with adaptation to the poor diet, we performed joint pathway analysis in Metaboanalyst v. 5.0 (Xia et al., 2009). In addition to the metabolites differentially abundant and genes differentially expressed between the Selected and Control populations, we included in this analysis 771 genes previously annotated to single-nucleotide polymorphisms (SNPs) differentiated in frequency between these two sets of populations (after 150 generations of experimental evolution; Kawecki et al., 2021). While most of the nine pathways identified in this analysis were only or mainly enriched in differentially abundant metabolites, two – purine metabolism and alanine, aspartate, and glutamate metabolism – involved a mix of metabolites and expression and SNP candidate genes (Supplementary file 4).

We therefore examined the evolutionary changes in these two pathways (imported from the KEGG database; Kanehisa and Goto, 2000). The alanine–aspartate–glutamate pathway contains multiple links among three of the four amino acids that are overabundant in the Selected flies. However, few of the many enzymes involved in those links showed a signal of differentiation between Selected and Control lines (Figure 3A). In contrast, purine metabolism showed a pattern of upregulation of multiple key enzymes involved in de novo purine synthesis in the Selected population, combined with overabundance of the three amino acids (glutamine, serine, and aspartate) that act as donors of four nitrogen atoms that form the basic purine structure (Kastanos et al., 1997; Salway, 2018) and provide (aspartate) a further amino group in the synthesis of AMP from IMP (Figure 3B). Despite this, compared to the Control flies, the Selected flies appeared depleted of nucleosides and nucleobases derived from IMP. We also found that multiple genes involved in AMP-ATP-cAMP conversion were associated with candidate SNPs differentiated in allele frequency between Selected and Control populations. While difficult to interpret functionally, this adds to the evidence that selection is driven by the dietary regime targeted purine metabolism.

Figure 3

Download asset Open asset

Imprint of evolutionary adaptation to poor larval diet on (A) alanine, glutamate, and aspartate metabolism and (B) purine metabolism pathways of adult female flies.

Highlighted in color are metabolites differentially abundant and enzymes differentially expressed between the Selected and Control populations, as well as enzymes to which at least one single-nucleotide polymorphisms (SNP) differentiated in frequency between the sets of populations has been annotated. Enzymes differentially expressed at the nominal p < 0.05 but not passing the 10% false discovery rate (FDR) are also indicated. Metabolites in gray were either not targeted, not detected, or the data did not pass quality filtering. The pathways have been downloaded from KEGG (https://www.genome.jp/kegg/pathway.html). For the sake of clarity, some substrates and products of depicted reactions that were not differentially abundant or were absent from the data, as well as cofactors, have been omitted. Only genes expressed in female fat body have been included. *5'-nucleosidases: *cN-IIB*, *NT5E-2*, *Nt5B*, *veil*, *CG11883*; none was differentially expressed.

Plastic response predicts in part evolutionary change of metabolome but not of gene expression

As was the case for Selected versus Control populations, the top GO term enriched for genes differentially expressed in flies raised on the poor versus standard larval diet treatment (corresponding to phenotypic plasticity) was ‘chitin-based cuticle development’ (Figure 1D; Supplementary file 2, table B). This is consistent with differences in body size: flies raised on the poor diet are much smaller than those raised on the standard diet (Kolss et al., 2009), and thus likely characterized by a greater surface-to-volume ratio and consequently a greater relative surface of cuticle on the carcass. This is, however, the only similarity between the effects of evolutionary regime and of the diet treatment for most enriched GO terms. No GO terms related to metabolism of amino acids, purines, or any other basic metabolic processes were significantly enriched for the effect of diet treatment. Instead, genes differentially affected by the diet treatment were enriched for many terms related to development, reproduction, nervous system, and behavior (Figure 1D; Supplementary file 2, table B).

This absence of parallelism between the evolutionary and phenotypically plastic responses to the poor larval diet is also visible at the level of individual genes. Only 42 genes were differentially expressed at 10% FDR for both of those factors; at least 15 of them are involved in integumentary (i.e., cuticle, epidermis, or tracheal) development, and most of the remaining ones play a role in the nervous system or transmembrane transport. Of 235 genes that passed the less stringent criterion of being differentially expressed at nominal (not adjusted) p < 0.05 for both factors, 123 showed the same sign of expression difference between poor and standard diet as between Selected and Control populations, and 112 showed the opposite sign, not different from what would be expected at random (p = 0.13, Fisher’s exact test). Across all genes, the correlation between the main effect of diet and the main effect of evolutionary regimes was weakly negative (r = −0.22, p < 0.0001, N = 8701; Figure 4A), as was the corresponding correlation for genes annotated to metabolism (GO metabolic process, r = −0.26, p < 0.0001, N = 4606). Thus, except for genes involved in the cuticle maturation, the plastic (within-generation) response of adult gene expression pattern to larval undernutrition did not in general predict the evolutionary, genetically based response. If anything, evolution tended on average to oppose phenotypically plastic responses of gene expression to larval diet.

Figure 4

Download asset Open asset

Relationship between phenotypic plasticity and genetically based evolutionary change of adult gene expression (A) and metabolite abundance (B) in response to larval diet.

The plastic responses and evolutionary change for each gene and metabolite were estimated, respectively, as the main effect of larval diet and evolutionary regime in the factorial statistical models (see ‘Materials and methods’). Color highlights genes and metabolites with significant effects (q < 0.1) for both diet and regime; in (A), orange symbols indicate significant genes involved in cuticle, integument, or tracheal development (Figure 4—source data 1).

Figure 4—source data 1 The estimated main effects of diet (the plastic response) and of evolutionary regime (the evolutionary change) on adult gene expression and metabolite abundance.: https://cdn.elifesciences.org/articles/92465/elife-92465-fig4-data1-v2.xlsx
Download elife-92465-fig4-data1-v2.xlsx

By contrast, the evolutionary change of metabolome driven by larval diet did to some degree parallel the phenotypically plastic response (correlation across all metabolites r = 0.32, p = 0.0005, N = 113; Figure 4B). Of the 30 metabolites that were significant at 10% FDR for both evolutionary regime and diet treatment, 25 showed the same sign of the difference between Selected and Control flies as the difference between flies raised on the poor versus standard diet (Fisher’s exact test, p = 0.0005). In particular, compared to flies raised on standard diet, flies raised on the poor diet showed a lower abundance of long-chain acyl-carnitines but an excess of medium-chain ones (Figure 2Bv, middle column), and a lower abundance of NAD+ and NADP (Figure 2Bii, middle column) and their precursors nicotinate and kynurenine (Figure 2Bvi, middle column), paralleling the patterns in Selected relative to Control flies (the left column of the respective heat maps). Some of the plastic effects of the poor diet treatment on sugar metabolism were also similar to those of evolutionary response, notably an underabundance of lactate and succinate, and an overabundance of UDP-glucose and of 2-/3-phosphoglycerate and glycerate, the latter accentuated by the underabundance of upstream and downstream glycolysis intermediates (Figure 2Biv).

More than a dozen modified amino acids and products of amino acid degradation were also affected, but these changes were not correlated with corresponding differences between the Selected and Control flies. Although several proteinogenic amino acids showed differential abundance depending on the diet treatment, these differences were also quite different from those due to the evolutionary regime and did not seem to show a consistent relationship with the properties of the amino acids. Finally, with the exception of inosine, the plastic response to diet did not detectably affect the abundance of purine nucleobases and nucleotides that do not contain a nicotinamide group (Figure 2Bii). These results are consistent with the absence of signal of an effect of the diet treatment on amino acid and purine metabolism in gene expression patterns presented above.

Much of evolutionary change in gene expression and metabolome is conserved between life stages

The Selected and Control populations evolved under different larval diets but the adult diet they experienced was the same. Hence, while the divergence between them in larval gene expression and metabolite abundance is presumably driven by immediate nutrient availability, selection at the adult stage would likely result from the need to compensate for carry-over effects of larval diet. It does not seem likely that these two selection pressures would favor broadly similar changes in gene expression and metabolism. However, such a broad similarity would still be expected if the genetic variants favored by selection on the larvae had similar effects across the stages. We therefore explored whether the divergence between Selected and Control populations in gene expression and metabolite abundance was correlated between the two life stages. To do so, we combined the present data from adult carcasses with previously published RNAseq and targeted metabolome data from whole third-instar larvae, collected at generation 190 (Erkosar et al., 2017) and generation 264 (Cavigliasso et al., 2023), respectively.

Of 424 genes that were differentially expressed at nominal p < 0.05 between Selected and Controls at both adults and larvae (Supplementary file 5), 377 (89%) showed the same direction of change on poor diet, more than expected at random (Fisher’s exact test, p < 0.0001, Figure 5A). Eighty-four genes passed the more stringent criterion of 10% FDR at both stages; of those, 78 (93%) showed the same direction of change in the two stages (Fisher’s exact test, p < 0.0001). The estimates of log-fold change were even correlated across all 8437 genes shared between the data sets (r = 0.35, p < 0.0001; Figure 5A). Thus, many of the differences between Selected and Control populations in the adult gene expression patterns paralleled those observed in the larvae.

Figure 5 with 1 supplement see all

Download asset Open asset

The relationship between the evolutionary change (i.e., the difference between the Selected and Control populations) in gene expression and metabolite abundance at the larval and adult stage.

(A) Correlation of evolutionary changes in gene expression (all genes: r = 0.35, N = 8437, p < 0.0001; genes with q < 0.1: r = 0.74, N = 84, p < 0.0001). Five genes (not passing q < 0.1) are outside of the range of the plot. (B) Correlation of evolutionary changes in metabolite abundance (all metabolites: r = 0.37, N = 97, p = 0.0002; metabolites with q < 0.1: r = 0.55, N = 21, p = 0.0091). The estimates plotted are for larvae and adults raised on the poor larval diet (see ‘Materials and methods’); q < 0.1 refers to genes or metabolites that pass the 10% false discovery rate (FDR) threshold for both stages; NS are the remaining genes or metabolites (Figure 5—source data 1, sheet A and B).

Figure 5—source data 1 The relationship between evolutionary change in adults and the corresponding change in the larvae.: https://cdn.elifesciences.org/articles/92465/elife-92465-fig5-data1-v2.xlsx
Download elife-92465-fig5-data1-v2.xlsx

Genes that were differentially expressed in Selected versus Control population in both adults and larvae were enriched in several terms related to transmembrane transport and iron homeostasis (Supplementary file 2, table D). Differences in gene expression of the larvae showed enrichment in lipid and carboxylic acid metabolism, as well as in a number of GO terms linked to cell proliferation and development, including that of the nervous system (Supplementary file 2, table c). The larval differentially expressed genes were not enriched in GO terms linked to amino acid or purine metabolism. However, 18 out of 99 genes in GO term ‘alpha amino acid metabolic process’ and 16 out of 107 in GO ‘purine-containing compound biosynthetic process’ were significantly different between Selected and Control larvae at 10% FDR (Supplementary file 6). Furthermore, the differences between Selected and Control populations across all genes in those GO terms were positively correlated between larvae and adults (amino acid metabolism: r = 0.40, p < 0.0001, N = 94; GO purine synthesis: r = 0.47, p < 0.0001, N = 74). Thus, while the amino acid metabolism and purine synthesis do not show disproportionate changes in gene expression in the larvae compared to other GO terms, the expression of multiple genes in those pathways has clearly been affected in a similar way as observed in adults.

Similarity between the effect of experimental evolution on larvae and adults was also apparent for the metabolome. Across all metabolites, the difference in metabolite abundance between the Selected and Control populations in the adults was positively correlated with the analogous difference in the larvae (r = 0.37, p = 0.0002; Figure 5B). This correlation was even more pronounced across the 21 metabolites that were differentially abundant at both stages (r = 0.55, p = 0.009); of those, 18 showed the same direction of change in larvae and adults, significantly more than expected by chance (Fisher’s exact test, p = 0.011).

In contrast, in the same data set we detected no correlation between the direct (phenotypically plastic) responses of larval and adult metabolome to larval diet, whether across all metabolites (r = 0.01, p = 0.89) or across the 35 metabolites detected as differentially abundant in response to diet at both stages (r = 0.17, p = 0.34; Figure 5—figure supplement 1). Of that last group, 19 showed the same sign of change at both stages while 16 showed opposite signs, not different from random expectation (p = 0.73). This suggests that differences in metabolite abundance accrued at the larval can be erased within a few days of adult life. Hence, differences between Selected and Control adults in metabolite abundance are more likely to be mediated by differences in adult gene expression than by differential accumulation of metabolites during the larval stage.

Metabolic profile of selected flies tends towards starved-like state

In addition to assessing the metabolome of flies directly sampled from the (standard) adult diet, we also assessed the metabolome of Selected and Control flies after they had been subject to 24 hr of total food deprivation (moisture was provided). The effect of the starvation treatment on the metabolic phenotype is also a phenotypically plastic response – to a different form of nutritional stress and one applied to adults rather than larvae. Our motivation was twofold. First, a comparison of fed and starved flies would reveal how the metabolome changes under an ongoing nutritional stress, and thus potentially help interpret metabolome changes driven by genetically based adaptation to recurrent nutrient shortage at the larval stage (i.e., the effect of evolutionary regime). Second, adult flies from Selected populations are less resistant to starvation than Controls (Kawecki et al., 2021). Hence, we expected that the metabolome of the Selected and Control populations might throw some light on the mechanisms underlying these differences in starvation resistance.

As expected, 24 hr of starvation had a strong effect on fly metabolome. PCA cleanly separated starved from fed samples along the first PC axis (Figure 6A), while larval diet mainly defined the second PC axis. Of the 113 metabolites, 74 were detected to be differentially abundant (at 10% FDR) between fed and starved flies (in terms of the main effect of starvation treatment in the full factorial model, Supplementary file 3). Of 39 metabolites affected by both starvation treatment and the evolutionary regime, 29 showed the same direction of change between fed and starved flies as between Control and Selected flies, significantly more than expected by chance (p = 0.006, Fisher’s exact test). In particular, in parallel to Selected flies relative to Controls, starved flies showed a decrease in multiple purine metabolites and two forms of vitamin B3 (nicotinate and nicotinamide), and an increase in medium-chain acyl-carnitines (Figure 2B). For some other metabolites, the effect of starvation differed markedly from that of the evolutionary regime. In particular, several intermediates of the glycolysis pathway were markedly less abundant in starved than fed flies (Figure 2B), whereas the surplus of acyl-carnitines also applied to those with long acyl chains. This possibly reflects a shift to catabolizing lipid stores for ATP generation following the depletion of carbohydrates.

Figure 6 with 1 supplement see all

Download asset Open asset

Effects of starvation on metabolite abundance.

(A) Principal component score plot on metabolite abundance for all conditions (Selected versus Control, poor versus standard larval diet, fed versus starved flies). The two replicate samples from the same population × diet × starvation treatment combination were averaged (Figure 6—source data 1). (B) Metabolites that show significant (q < 0.1) interaction between the evolutionary regime and starvation treatment. Plotted are least-square means obtained from the full factorial general mixed model (Figure 6—source data 2). The bar next to the Y-axis indicates ± standard error of the least-square means (i.e., its width corresponds to 2 SEM); the standard error is identical for all treatments because the least-square means were estimated from the general mixed model and the design is balanced. For compounds showing other interactions, see Figure 6—figure supplement 1. (C) The relationship between the effect of 24 hr starvation (starved minus fed flies) on metabolite abundance in Control flies raised on the standard diet and the evolutionary change (Selected minus Control) quantified in fed flies raised on poor diet (Figure 6—source data 3). Three categories of metabolites of interest are highlighted; ‘acylcarnitines’ only include those with even-numbered acyl chain (resulting from catabolism of triglycerides) but not those with 3C and 5C chains (which are products of amino acid catabolism). Pearson’s correlations: amino acids r = 0.50, N = 19, p = 0.030; purine compounds r = 0.76, N = 16, p = 0.0006; acylcarnitines r = 0.74, N = 9, p = 0.022; other metabolites r = 0.02.

Figure 6—source data 1 Principal component analysis of metabolite abundance in fed and starved flies: the first four principal component scores of all populations under all conditions.: https://cdn.elifesciences.org/articles/92465/elife-92465-fig6-data1-v2.xlsx
Download elife-92465-fig6-data1-v2.xlsx
Figure 6—source data 2 Abundance patterns of metabolites with interactions.: https://cdn.elifesciences.org/articles/92465/elife-92465-fig6-data2-v2.xlsx
Download elife-92465-fig6-data2-v2.xlsx
Figure 6—source data 3 Relationship between response to starvation and the evolutionary change.: https://cdn.elifesciences.org/articles/92465/elife-92465-fig6-data3-v2.xlsx
Download elife-92465-fig6-data3-v2.xlsx

The pattern for proteogenic amino acids also appeared rather different (Figure 2B). However, the focus on the main effects of starvation treatment on amino acid abundance is somewhat misleading because for 6 of the 10 essential amino acids the response to starvation differed between the Selected and Control flies. For all six, the sign of the interaction was the same: while the levels of all these amino acids decreased in Control flies when they starved (q < 0.1), their abundance declined less (lysine), remained similar (threonine, isoleucine), or even increased (phenylalanine, leucine, valine; q < 0.1) in starving Selected flies (Figure 6B). The other three metabolites with significant regime × starvation interaction showed different patterns (Figure 6B). There were also six metabolites with a significant larval diet × starvation or an interaction among all three experimental factors; they did not appear to be functionally connected or to share a common pattern (Figure 6—figure supplement 1).

These results imply that evolutionary adaptation of the Selected populations to larval undernutrition had consequences for the way their essential amino acid metabolism responds to food deprivation at the adult stage. Furthermore, even though in a normal fed state the Selected flies showed lower abundance of most essential amino acids, they could maintain or increase their levels during 24 hr of starvation. This suggests that their low concentrations in the fed state are not due to their absolute shortage but to differential allocation and use. In total, abundance of seven essential amino acids declined significantly upon starvation in Control flies, making the parallel with the difference between Selected and Control flies in fed state more evident. Indeed, the evolved differences in abundance of all proteogenic amino acids between Selected and Control flies were significantly positively correlated with the presumably ancestral response of standard diet-raised Control to starvation; the same was the case for purine compounds and acyl carnitines (Figure 6C).

Differential response of the Selected and Control populations to starvation is supported by regime × starvation interaction on PC1 scores (F_1,10 = 10.7, p = 0.0084) – the metabolic state of Selected flies changed less upon starvation than that of Controls. This does not imply that the metabolome of Selected flies was more robust to starvation. Rather, fed flies from the Selected populations were situated closer to starved flies along the starvation-loaded PC1 axis than fed Control flies (Figure 6A; F_1,17.8 = 14.0, p = 0.0015, GMM on PC1 scores); in the starved condition, they converged to a similar state (F_1,17.8 = 0.0, p = 0.99; see Supplementary file 7 for detailed statistics). Overall, the comparison of the (plastic) response of metabolome to starvation indicates that prominent aspects of the adult metabolic profile of Selected populations evolved in the direction resembling the effects of starvation.

Adult cost of larval adaptation

The above results demonstrate that evolutionary adaptation to nutrient-poor larval diet has led to significant changes in adult metabolism. It is not clear, however, to what degree these changes may have evolved to improve fitness of adults that experienced poor diet as larvae, rather than being carry-over effects or trade-offs of changes driven by selection acting on the larval stage. If the former, adults from the Selected populations should perform better than those from the Control populations when both are raised under the conditions of the poor larval diet regime. If the latter, the fitness of Selected adults raised on poor diet should be similar to or even lower than the fitness of Control adults developed on the same poor diet.

To test these alternative predictions, we compared the fecundity of females from Selected and Control populations raised on the poor larval diet, transferred to standard diet after eclosion, supplemented with ad libitum live yeast for 24 hr before oviposition, and allowed to oviposit overnight. These conditions mimicked those under which Selected populations have been maintained and propagated in the course of their experimental evolution; a female that laid more eggs in this time window would contribute proportionally more offspring to the next generation. The rate at which females can convert nutrients into eggs within this short time window is thus arguably a key measure of their fitness, and is inherently dependent on the metabolism of nutrients, in particular proteins, lipids, and nucleic acids. Despite having evolved under these conditions, Selected females laid only about 2/3 of the number of eggs produced by the Control females, whose ancestors did not face the poor larval diet (Figure 7A). Fly ovaries consist of a variable number of branches called ovarioles; in each ovariole, eggs are produced sequentially one by one. The rate of egg production may thus be limited by the number of ovarioles (Schmidt et al., 2005), which is determined during larval development (Bergland et al., 2008). Indeed, we found that poor diet-raised females of the Selected populations tended to have slightly fewer ovarioles than their Control counterparts (Figure 7B). However, even adjusted for the difference in ovariole number, the Selected flies produced substantially fewer eggs than Controls (Figure 7C).

Figure 7

Download asset Open asset

Selected populations pay a fecundity cost for larval adaptation to poor diet.

(A) The number of eggs laid overnight by females from the Selected and Control (GMM; F_1,8 = 44.5, p = 0.0002, N = 16 per evolutionary regime, 2–3 per replicate population). (B) The number of ovarioles (left + right ovary) per female (GMM; F_1,10 = 4.7, p = 0.056, N = 72 per regime, six per replicate population). (C) The mean number of eggs per ovariole (one-way ANOVA; F_1,10 = 17.1, p = 0.002, N = 6 populations per regime). The females experienced the dietary conditions, under which the Selected populations evolved: the larvae were raised on the poor diet; the adults were transferred to the standard diet and additionally provided live yeast 24 hr before oviposition. Bars are means, error bars correspond to ± SEM, the symbols to means of replicate populations (Figure 7—source data 1).

Figure 7—source data 1 Data on fecundity and ovariole number.: https://cdn.elifesciences.org/articles/92465/elife-92465-fig7-data1-v2.xlsx
Download elife-92465-fig7-data1-v2.xlsx

Drosophila are ‘income breeders’ in that they directly convert acquired nutrients into eggs rather than relying on previously accumulated reserves (O’Brien et al., 2008). The fecundity difference thus implies that the Selected populations are less effective than Controls at converting dietary nutrients into the nutritional and/or generative components of the egg even when raised under the conditions under which the former but not the latter evolved during more than 200 generations.

Discussion

Costly correlated evolution of adult metabolism?

Experimental evolution under chronic larval undernutrition resulted in major shifts in gene expression patterns and metabolite abundance at the adult stage, replicable across six independent experimental (Selected) populations of D. melanogaster. This occurred even though adults of all populations experienced standard diet in the course of the experimental evolution. These evolved differences in adult gene expression and metabolite abundance between Selected and Control populations were broadly positively correlated with the corresponding differences between Selected and Control larvae. This was the case even though expression and metabolites were quantified in whole body in larvae but in dissected carcasses in adults, and despite these assays being performed dozens of fly generations apart. Therefore, the observed correlations between adult and larval evolutionary responses may have underestimated their actual similarity.

Even though adults of both Selected and Control populations experienced the same standard diet, they might still have been subject to differential selection on their adult metabolism. In particular, selection on Selected adults may have favored changes that compensate for lingering consequences nutritional hardship endured during the larval stage. However, it seems difficult to conceive why such compensatory responses of a well-fed adult should be mediated by similar physiological adjustments as those that promote larval growth under severe undernutrition. A more plausible explanation for the similarity in evolutionary changes between larvae and adults is that most genetic variants available for evolution affected gene expression of larvae and adults similarly. As a consequence, many of the adult gene expression changes would represent correlated responses to selection on the larval stage rather than the direct response to selection on adult performance (Lande and Arnold, 1983).

Such correlated responses may be detrimental to the performance of one or both life stages (Collet and Fellous, 2019). Consistent with this prediction, when raised on the poor diet and transferred to the standard diet as adults, the Selected flies were less effective than Controls in converting dietary nutrients into eggs, even though these were the conditions under which the former had evolved and to which the latter were exposed for the first time. (An analogous result on fecundity has been reported in another evolution experiment; May et al., 2019.) We have previously reported a lower fecundity of Selected females compared to Controls when both were raised on standard diet (Kolss et al., 2009); we interpreted that finding as a manifestation of reduced performance of Selected populations when raised on the standard larval diet. The present results imply that Selected adults perform less well than Control adults irrespective of the larval diet on which they have been raised. Other things being equal, a female’s contribution to the next generation under the experimental evolution regime is proportional to the number of eggs she lays within a time window similar to that used to quantify fecundity. Thus, the fecundity reduction we found in Selected adults represents a significant fitness trade-off of the improved ability of the Selected larvae to survive and develop under extreme nutrient shortage.

Amino acid catabolism and starved-like metabolic profile?

While Selected and Control adults differed in abundance of a variety of metabolites, the most striking pattern was the reduced abundance of 8 out of 10 essential amino acids in Selected flies. Fecundity in Drosophila is limited by essential amino acids (Grandison et al., 2009), in particular those in short supply relative to the needs of fly protein synthesis, which for a yeast-based diet are methionine and leucine (Piper et al., 2017). In addition, several essential amino acids (methionine, valine, isoleucine, and in particular leucine, all less abundant in Selected flies) act as signaling molecules that promote protein synthesis by activating TOR complex 1 (Wolfson et al., 2016; Antikainen et al., 2017; Gu et al., 2022). Most of the protein synthesis activity in adult female fat bodies is directed toward synthesizing egg proteins (Piper et al., 2017; Gupta et al., 2022). Thus, while we have not demonstrated a direct causal link, lower abundance of most essential amino acids in the Selected flies is consistent with their lower fecundity.

A reduction in free amino acid abundance could result from a lowered supply from nutrition and/or from a higher rate of their use for protein synthesis. However, these mechanisms should lead to a general depletion of free amino acids, which is not what we observed. Rather, in contrast to essential amino acids, several non-essential amino acids were overabundant in Selected flies; these overabundant amino acids are not enriched in dietary yeast relative to their need for Drosophila protein synthesis (Piper et al., 2017). Thus, rather than being explained by differences in dietary amino acid acquisition or use in protein synthesis, the differences in amino acid abundance between Selected and Control populations appear consistent with differential use of amino acids in metabolism.

As adults, larvae of Selected populations also show lower abundances than Control larvae of multiple amino acids, including six essential, and a higher concentration of uric acid (Cavigliasso et al., 2023), an end product of purine metabolism and the main compound in which nitrogen is excreted in insects (Salway, 2018; Cohen et al., 2020). This suggests that Selected larvae catabolize amino acids and excrete nitrogenous waste products at a higher rate, a hypothesis further supported by their increased accumulation of the heavy isotope of nitrogen ¹⁵N (Cavigliasso et al., 2023). Although we found no difference in the levels of uric acid (urate) in the adult metabolome, it should be kept in mind that the adult metabolome was quantified carcasses, which only include fragments of Malpighian tubules, the excretory organs in which uric acid is synthesized (Cohen et al., 2020). It is notable that Selected flies show overabundance of the three amino acids – glutamine, aspartate, and serine – that contribute nitrogen atoms to purine/uric acid synthesis pathway, as well as overexpression of multiple genes involved in that process. Thus, while we do not have direct evidence for this, it remains possible that Selected flies catabolize amino acids and excrete uric acid at a higher rate than Controls also at the adult stage.

Increased catabolism of amino acids – sourced from autophagy – is one of the hallmarks of the physiological response to starvation (Scott et al., 2004). We found that several differences in metabolome between Selected and Control flies resemble the effects of starvation. In addition to reduction in abundance of several essential amino acids, this includes lower concentration of purine nucleotides and nucleosides, lower levels of trehalose and lactate, and higher abundance of acylcarnitines hinting at increased catabolism of fatty acids. Thus, the metabolic profile of Selected flies, in their normal fed and reproductively active state, resembles that of flies that have been starving. Consistent with the link with starvation, the genomic architecture of differentiation between the Selected and Control populations includes many candidate genes for starvation resistance (Kawecki et al., 2021). It is tempting to speculate that, as a by-product of genetic adaptation to poor larval diet, the Selected populations have become programmed to express a starved-like adult metabolic phenotype, which might explain their low fecundity.

Relationship between phenotypic plasticity and evolutionary change

The evolutionary changes in gene expression did not in general recapitulate the phenotypically plastic responses of adult expression to larval diet. The abundance of a subset of metabolites did evolve in the direction that mimicked the plastic response, but this was not the case for most of the other metabolites. Congruence between the directions of phenotypically plastic response and evolutionary change driven by the same environmental factor has been interpreted as evidence for adaptive nature of the plastic response; conversely, where evolutionary change went in the opposite direction to the plastic response, the latter has been deemed maladaptive (Yampolsky et al., 2012; Ghalambor et al., 2015; Huang et al., 2016; Josephs et al., 2021). We believe this interpretation is not applicable to our results. First, as we argued elsewhere (Cavigliasso et al., 2023), a plastic response to a novel environment may be adaptive in terms of direction but overshoot the optimum phenotype; in such a case, an evolutionary change in the opposite direction would be favored despite the initial plastic response being adaptive. Second, the above interpretation assumes that evolutionary changes are mostly adaptive. However, as we have argued above, the evolutionary responses of adult gene expression and metabolism to larval undernutrition are likely to a large degree maladaptive costs of physiological adaptations favored at the larval stage. Thus, even plastic responses optimal from the viewpoint of adult fitness may have been reversed by the evolutionary change. These considerations imply that assessing the adaptive (or otherwise) nature of phenotypically plastic responses based on the direction of the evolutionary change may be misleading. Furthermore, our experimental results contradict the often-made assertion that phenotypic plasticity ‘drives’ evolutionary change (Baldwin, 1896; Pigliucci and Murren, 2003; Moczek et al., 2011; Laland et al., 2015).

Evolutionary constraints on regulatory flexibility?

Like in all holometabolous insects, most larval tissues in Drosophila disintegrate during metamorphosis and the rest (e.g., brain) undergo extensive remodeling. Most of the adult structures and organs are formed de novo from progenitor cells, resulting in an adult that is very different morphologically from the larvae, and physiologically specialized for a different function (reproduction rather than growth). In particular, cells of the larval fat body dissociate from one another and undergo autophagy and apoptosis during metamorphosis; the adult fat body develops anew from adult progenitor cells although possibly including some remaining larval fat body cells (Li et al., 2019). A major advantage of this complex development is thought to be that it decouples larval and adult gene expression, promoting independent evolution of larval and adult phenotypes (Moran, 1994; Rolff et al., 2019). Contrary to this notion, our results suggest that, at least on the scale of hundreds of generations, evolutionary changes in physiology driven by selection acting on larvae may have maladaptive pleiotropic effects on adult physiology. If this is the case in a holometabolous insect, such evolutionary non-independence of juvenile and adult physiology would likely be more pronounced in species with more developmental continuity between juvenile and adult tissues and organs.

Our study is thus relevant to understanding of constraints on the evolutionary refinement of physiology and life history of metazoans. Complex multicellularity crucially depends on the ability of the genome to express its genes differently in different cell types and life stages. The existence of specialized cells and organs that express greatly different yet highly coordinated and functional metabolic phenotypes from the same genome testifies to the power of regulatory evolution. On the other hand, there has been increased recognition that the ability of evolution to independently shape phenotypes of different life stages and sexes may be significantly constrained. Such constraints are expected to emerge from the complexity of gene regulatory and metabolic networks (Wagner, 2011; Sorrells et al., 2015; Schaerli et al., 2018). One manifestation of such constraints is ‘intralocus sexual conflict’ (sexually antagonistic pleiotropy), whereby simultaneous optimization of female and male phenotypes is hindered by constraints on independent evolution of gene expression in the two sexes (Rice, 1984; Pischedda and Chippindale, 2006; Hollis et al., 2014; Veltsos et al., 2017). Similarly, the developmental theory of aging postulates that gene expression and metabolism are optimized for maximizing performance at a young age and fail to adjust in later age in ways that could improve reproductive lifespan or healthspan (de Magalhães, 2012; Gems and Partridge, 2013), an idea increasingly supported by experimental data (Carlsson et al., 2021). This apparent metabolic inertia of aging individuals might be explained by selection at old age being weak (Medawar, 1952; Hamilton, 1966; Partridge and Barton, 1993). However, our results suggest similar evolutionary constraints linking the metabolism of juveniles and young adults in their reproductive prime, before the age-related decline in the strength of natural selection sets in Hamilton, 1966. Such constraints would hinder evolutionary optimization of juvenile and adult gene expression and metabolism if optima differ between the stages (Collet and Fellous, 2019).

Materials and methods

Diets, experimental evolution, and fly husbandry

Request a detailed protocol

Two diets were used for this study. The ‘standard’ diet consisted of 12.5 g dry brewer’s yeast, 30 g sucrose, 60 g glucose, 50 g cornmeal, 0.5 g CaCl₂, 0.5 g MgSO4, 10 ml 10% Nipagin, 6 ml propionic acid, 20 ml ethanol, and 15 g of agar per liter of water. The ‘poor’ diet contained 1/4 of the concentrations of yeast, sugars, and cornmeal, but the same concentrations of the other ingredients. All experiments were carried out at 25°C and 12:12 hr LD cycle.

Six Selected and six Control populations were all derived from the same base population originally collected in 1999 in Basel (Switzerland) and maintained on the standard food for several years before the evolution experiment started in 2005 (Kolss et al., 2009). The six replicate populations per evolutionary regime constitute the main units of replication in this study; their number was limited by workload considerations. By the time of the experiments reported in this article, the Selected populations had been maintained on the poor larval diet for over 230 generations; the Control populations were maintained in parallel on the standard larval diet. Larval density was controlled at 200–250 eggs per bottle with 40 ml of food medium. Adults of both regimes were transferred to standard diet within 14 d of egg laying (sometimes a day or two later when not enough adults have emerged from the poor diet) and additionally fed live yeast 24 hr before egg collection to stimulate egg production. The target adult population size was 180–200 individuals.

Flies used in the experiments reported here were raised using similar procedures. Prior to all experiments reported in this study, all populations were raised on the standard diet for at least two generations to minimize the effects of maternal environment (Vijendravarma et al., 2010). For RNAseq and metabolome analyses, flies of both Selected and Control populations were raised on both larval diets. Eggs to establish the next generation were collected by allowing flies to oviposit overnight on orange juice agar sprinkled with live yeast. During the egg collection, the eggs were washed with water to remove any traces of diet at the surface – a procedure that also removes much of microbiota. To control the colonization by microbiota, eggs were re-inoculated using parental feces from a mix of adult flies from all 12 populations. These adults were left in a Petri dish and a wedge of food for 48 hr; they and the food were subsequently removed, the feces were washed from the surfaces of the Petri dish with PBS, filtered to remove eggs and debris, and adjusted to OD₆₀₀ = 0.5. Each larval culture was established with 200 eggs transferred to a bottle with 40 ml of poor or standard diet, with 300 µl of the feces suspension pipetted on top; based on plating, this inoculum contains about 10³ CFU. All experiments were performed on females, aged 4–6 days old from eclosion. Selected larvae develop on the poor diet faster than Controls (Kolss et al., 2009; Erkosar et al., 2017). To ensure that the females from Selected and Control populations can be collected synchronously, we initiated the larval cultures in a staggered manner over several days, so that we could collect females from around peak of emergence and at the same time for Selected and Control populations despite the difference in developmental time. For RNAseq and metabolome quantification, adults of both sexes were collected around the peak of emergence, transferred to fresh standard diet, and allowed to mate freely for 3 d before females were collected. For fecundity and ovariole measurements, we collected virgin females.

RNAseq on adult carcasses

Request a detailed protocol

Four-day-old mated female flies were collected in the morning and dissected in PBS. The abdomen was separated, and the gonads, the gut, and the bulk of Malpighian tubes were removed, leaving the ‘carcass,’ consisting of the abdominal fat body attached to the body wall, as well as any hemolymph, oenocytes, neurons, and fragments of Malpighian tubes that remained embedded in the fat body. Precise dissection of the adult fat body without disrupting it is very difficult, which is why carcass is typically used instead (e.g., modEncode project, Brown et al., 2014). From each of the 12 populations raised on either larval diet, we collected one sample of 10 carcasses, that is, 24 samples in total. The samples were snap-frozen in liquid nitrogen and stored at –80°C.

RNA was extracted from the carcass samples using ‘Total RNA Purification Plus’ by Norgen Biotek (#48300, 48400). cDNA libraries (TrueSeq Standard RNA) were generated and sequenced on two lanes of Illumina HiSeq4000 (single read, 150 bp) by the Genomic Technologies Facility of the University of Lausanne following manufacturer’s protocols. Reads were mapped (pseudoaligned) to D. melanogaster reference genome BDGP6.79 using kallisto (Bray et al., 2016), with 22–31 million mapped reads per sample. Read counts per transcript feature output from kallisto were converted to counts per gene using tximport (Soneson et al., 2015). We filtered out genes with very low expression in that we only retained genes with read count per million greater than 2 in at least six samples. We further detected 45 pairs of genes with identical counts across all samples; from each pair we retained only the gene with a lower FBgn number, leaving 8701 unique genes.

Differential expression analysis was performed with limma-voom (Law et al., 2014; Ritchie et al., 2015), using EdgeR normalization (Robinson et al., 2010) implicit in the voom algorithm. Larval diet (poor versus standard), evolutionary regime (Selected versus Control) and their interaction were the fixed factors in the limma model; replicate population was a random factor modeled with duplicateCorrelation function of limma. Adjustment of p-values for multiple comparison was performed using Storey’s FDR q-values (Storey and Tibshirani, 2003) as implemented in procedure MULTTEST option PFDR of SAS/STAT software v. 9. 4 (Copyright 2002–2012 by SAS Institute Inc, Cary, NC). Genes with expression different at q < 0.1 were considered significant for enrichment analyses. GO term enrichment analysis was carried out with bioconductor 3.12 package topGO v. 2.42.0 (Alexa and Rahnenfuhrer, 2021), using the weight method and Fisher’s exact test. Because the tests for enrichment of different GO terms are highly non-independent, no meaningful method for calculating FDR exists (Alexa and Rahnenfuhrer, 2021); therefore, we report uncorrected p-values, focusing our interpretation on the top GO terms with p < 0.01.

Because the FDR threshold focuses on minimizing false positives, it leaves out many genes that have truly differed in expression. Thus, the number of genes that pass the FDR threshold greatly underestimates the number of genes that truly differ in expression, and the relationship between the two depends greatly on statistical power. Therefore, we also estimated the number of genes that differed in expression due to each factor in the analysis as the total number of genes minus the estimated number of ‘true nulls’ (Storey and Tibshirani, 2003).

To study to which degree the gene expression profiles of flies from the two regimes raised on the two larval diets were separable in a multivariate space, we performed PCA on the correlation matrix of the log-normalized expression data for all genes. To test for the separation of the samples in this multivariate space, we performed a MANOVA on the PCA scores, with regime, diet, and their interaction as the factors (using procedure GLM of SAS/STAT software).

Larval gene expression analysis

Request a detailed protocol

The comparison of adult to larval gene expression differences was based on previously published data from an RNAseq study of whole larvae from the Selected and Control populations after about 190 generations of experimental evolution. All larvae were raised on the poor diet in that experiment in either germ-free state or colonized with a single microbiota strain at the high concentration of about 10⁸ CFU (Erkosar et al., 2017). Thus, the bacterial inoculation used for adult RNAseq (feces suspension containing about 10³ CFU) was intermediate between the two larval treatments. To maximize the compatibility of the adult and larval data sets, we remapped the larval reads using the same kallisto algorithm as for adult carcasses and analyzed the expression of the 11,475 genes with the same limma-voom approach, with evolutionary regime, microbiota treatment, and their interaction as the factors. We used the main effect of evolutionary regime from this study for the comparison with the adult results; using just data from the microbiota-colonized treatment led to qualitatively similar conclusions.

Broad-scale targeted metabolomics

Request a detailed protocol

Metabolite abundance was measured using multiple pathway targeted analysis in the carcasses of 4-days-old mated females obtained as described above. From each population raised on each larval diet, we obtained two samples of 10 carcasses (‘fed flies’). Two further samples of 10 carcasses per population and larval diet were obtained from females subject to 24 hr of starvation on nutrient-deprived agarose (‘starved flies’), thus resulting in a three-way design (2 evolutionary regimes each with 6 populations × 2 current larval diets × fed versus starved flies, with 2 biological replicates, for a total of 96 samples, the number limited by cost). The number of samples was limited by the costs of metabolome analysis. The two samples for each population, larval diet and adult starvation treatment (i.e., fed or starved), were dissected by two experimenters, resulting in two experimental batches of 48 samples each. The two batches were also processed for metabolite extraction and analysis separately, on different days.

Extracted samples were analyzed by hydrophilic interaction liquid chromatography coupled to tandem mass spectrometry (HILIC-MS/MS) in both positive and negative ionization modes using a 6495 triple quadrupole system (QqQ) interfaced with 1290 UHPLC system (Agilent Technologies). Data were acquired using two complementary chromatographic separations in dynamic multiple reaction monitoring mode (dMRM) as previously described (van der Velpen et al., 2019; Medina et al., 2020). Data were processed using MassHunter Quantitative Analysis (for QqQ, version B.07.01/Build 7.1.524.0, Agilent Technologies). Signal intensity drift correction was performed on the pooled QC samples and metabolites with CV > 30% were discarded (Dunn et al., 2011; Tsugawa et al., 2014; Broadhurst et al., 2018). In addition, a series of diluted quality controls (dQC) were used to evaluate the linearity of metabolite response; peaks with correlation to dilution factor R² < 0.75 were discarded.

Metabolome analysis

Request a detailed protocol

The peak area data for each compound were log-transformed, zero-centered, and Pareto-scaled separately for the two experimental batches. To test for differential abundance of single compounds, we fitted general mixed models (GMM) to these Pareto-scaled relative metabolite values using procedure MIXED of SAS/STAT software v. 9.4 (Copyright 2002–2012 by SAS Institute Inc). For each compound, we first fitted a full model, with evolutionary regime (Selected versus Control), larval diet (poor versus standard), starvation treatment (starved versus fed), and all their interactions as fixed factors, and population nested within regime, diet × population and starvation × population as random factors. Inspecting the residuals from this model, we detected 13 data points (out of 10,848) with externally Studentized residuals greater in magnitude than ±4.0; these data points were removed as outliers and the model was refitted.

The residuals for 3 of the 113 compounds (creatine, hydroxykynurenine, and trehalose) deviated from normality at 10% FDR (Wilk–Shapiro test); we still report the results for these three compounds but they should be treated with caution. Because the main focus of the article is on the metabolism of flies in their normal fed state, we also analyzed the data from the 48 samples obtained from fed flies separately, fitting GMM with regime, diet, and their interaction, as well as experimental batch as fixed effects, and population nested in regime and diet × population interaction as random effects. The fixed factors in the GMM were tested with type 3 F-tests, using Satterthwaite method to estimate the denominator degrees of freedom. P-values for each factor were adjusted for multiple comparison as for differentially expressed genes. To illustrate the effects of the experimental factors (in particular, the interaction between the effects of evolutionary regime and starvation; Figure 6B), for some metabolites we plotted the estimated marginal means from the GMM. Because the design was balanced and the error variance was estimated from the model, all marginal means had the same standard error; to reduce the clutter in the plots, we plotted this common standard error as a single error bar rather than adding such bars to each symbol representing the mean.

To visualize and test for multivariate differentiation of the metabolome, we performed PCA on log-abundances of the 113 metabolites. The values from the two batches were averaged before this analysis so there was one point per population × diet combination × starvation treatment. As for gene expression patterns, sample scores from the PCA were tested for the effects of experimental factors and their interaction in a MANOVA implemented in procedure GLM of SAS/STAT software.

We also explored the relationship between the evolutionary change in metabolite abundance and the response to starvation for several categories of compounds. We specifically examined if the difference in metabolite abundance between Selected and Control flies raised on poor food, in their normal fed state, was correlated (Pearson’s r) with the difference between starved and fed Control flies raised on the standard diet. These two variables are functions of non-overlapping sets of measurements, thus avoiding spurious correlations due to non-independence of errors.

Fecundity and ovariole number

Request a detailed protocol

Fecundity was only assayed in females raised on poor larval diet, as described above. Male genotype and in particular the seminal fluid proteins males transfer to females during mating affect the short-term fecundity of the female. To ensure that any differences in egg number among populations are driven by female physiology and not by differences between males to which those females were mated, females from all populations were mated to males from a single laboratory population, originally collected from another site in Switzerland (Valais) in 2007. From each population, we collected 25–35 virgin females at the peak day of emergence and transferred them to standard food. Three days later, we split the females into replicates of 10–12 females (2–3 replicates per population depending on the total number of females remaining alive) and placed them together with 10–12 young males on orange juice-agar medium supplemented with live yeast; they were allowed to feed and mate for 24 hr. Subsequently, males were discarded (to prevent potential fecundity reduction due to male harassment) and the groups of females were allowed to oviposit overnight in new bottles with fresh orange juice-agar supplemented with live yeast. Eggs were washed from the medium surface with tap water, collected on a fine nylon mesh, and transferred to a well of a 12-well cell culture plate containing 3 ml of 1% sodium dodecyl sulfate (SDS, Sigma) to facilitate egg dispersion. A photograph of each well was taken under a Leica stereomicroscope with a Canon 60D camera with manual exposure programming, automatic white balance, 3.2 s of exposure time, and ISO 400.

The number of eggs was estimated automatically with Codicount ImageJ plugin, following Perez, 2017. This approach is based on automatically quantifying the area corresponding to the eggs on the image, based on color contrast between the eggs and the background. This total egg area was then converted to an estimate of the number of eggs by dividing it by the area of a single egg, estimated from a separate sample (the same standard egg area was used for all images). Thus, the estimated number of eggs was divided by the number of females in a particular replicate. In one replicate, 1 of 10 females was found dead at the end of oviposition; for the analysis, we set the number of females in this bottle at 9.5.

Flies were raised on the poor larval diet. While the number of ovarioles is fixed by the time of emergence, they are better visible and easier to count if filled with developing eggs (Bergland et al., 2008). We therefore allowed 25 freshly emerged females to mate with 25 males on standard diet for 3 d and subsequently to feed on ad libitum live yeast for 2 d, before collecting and storing them at –80°C until dissections. Six females from each population were haphazardly chosen for ovary dissection. We dissected the ovaries by separating the abdomen and cutting its posterior end, opening the abdomen laterally, and removing the ovaries. We then dipped the ovaries briefly in a solution of crystal violet to improve visual contrast, opened them, and counted the number of ovarioles in each ovary. Dissections and counting were done blindly with respect to the identity of the sample. To normalize the fecundity of each population relative to ovariole number, we divided the mean number of eggs per female in this population by the mean number of ovarioles.

The number of replicates was based on practical considerations, no formal power analysis has been conducted. The estimation of egg number by counting ovarioles was performed blindly with respect to the identity of the sample. Egg number and ovariole number per female were analyzed with a GMM, with regime as a fixed factor and population nested in regime as a random factor. For the number of eggs per ovariole, we only had one data point per population; these were compared between regimes with a one-way linear model.

Data availability

The raw and processed data from the RNAseq on adult carcasses are available from NCBI GEO (accession number GSE193105). Raw data for the previously published larval RNAseq are available from NCBI SRA (accession numbers SAMN07723150-SAMN07723173). Previously published larval metabolome data are available as supplementary material to Cavigliasso et al., 2023. The adult metabolite abundance data are provided in Supplementary file 8, fecundity and ovariole data in Figure 7-source data 1.

The following data sets were generated

(2022) NCBI Gene Expression Omnibus
ID GSE193105. Evolutionary and phenotypically plastic response of adult gene expression to larval undernutrition in Drosophila melanogaster.

https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE193105

The following previously published data sets were used

(2017) NCBI BioProject
ID PRJNA412704. RNAseq on Drosophila larvae genetically adapted to poor diet in microbiota-colonized and germ-free state.

https://www.ncbi.nlm.nih.gov/bioproject/PRJNA412704

References

(2014) Time window-dependent effect of perinatal maternal protein restriction on insulin sensitivity and energy substrate oxidation in adult male offspring
American Journal of Physiology. Regulatory, Integrative and Comparative Physiology 307:R184–R197.
https://doi.org/10.1152/ajpregu.00015.2014
- PubMed
- Google Scholar
1. Akitake Y
2. Katsuragi S
3. Hosokawa M
4. Mishima K
5. Ikeda T
6. Miyazato M
7. Hosoda H
(2015) Moderate maternal food restriction in mice impairs physical growth, behavior, and neurodevelopment of offspring
Nutrition Research 35:76–87.
https://doi.org/10.1016/j.nutres.2014.10.014
- PubMed
- Google Scholar
Software
1. Alexa A
2. Rahnenfuhrer J
(2021)
topGO: enrichment analysis for gene Ontology

R Package Version 2.46.0.
(2017) TOR-mediated regulation of metabolism in aging
Aging Cell 16:1219–1233.
https://doi.org/10.1111/acel.12689
- PubMed
- Google Scholar
1. Baldwin JM
(1896) A new factor in evolution
The American Naturalist 30:441–451.
https://doi.org/10.1086/276408
- Google Scholar
1. Bateson P
2. Barker D
3. Clutton-Brock T
4. Deb D
5. D’Udine B
6. Foley RA
7. Gluckman P
8. Godfrey K
9. Kirkwood T
10. Lahr MM
11. McNamara J
12. Metcalfe NB
13. Monaghan P
14. Spencer HG
15. Sultan SE
(2004) Developmental plasticity and human health
Nature 430:419–421.
https://doi.org/10.1038/nature02725
- PubMed
- Google Scholar
(2008) Quantitative trait loci affecting phenotypic plasticity and the allometric relationship of ovariole number and thorax length in Drosophila melanogaster
Genetics 180:567–582.
https://doi.org/10.1534/genetics.108.088906
- PubMed
- Google Scholar
(2016) Near-optimal probabilistic RNA-seq quantification
Nature Biotechnology 34:525–527.
https://doi.org/10.1038/nbt.3519
- PubMed
- Google Scholar
1. Broadhurst D
2. Goodacre R
3. Reinke SN
4. Kuligowski J
5. Wilson ID
6. Lewis MR
7. Dunn WB
(2018) Guidelines and considerations for the use of system suitability and quality control samples in mass spectrometry assays applied in untargeted clinical metabolomic studies
Metabolomics 14:72.
https://doi.org/10.1007/s11306-018-1367-3
- PubMed
- Google Scholar
1. Brown JB
2. Boley N
3. Eisman R
4. May GE
5. Stoiber MH
6. Duff MO
7. Booth BW
8. Wen J
9. Park S
10. Suzuki AM
11. Wan KH
12. Yu C
13. Zhang D
14. Carlson JW
15. Cherbas L
16. Eads BD
17. Miller D
18. Mockaitis K
19. Roberts J
20. Davis CA
21. Frise E
22. Hammonds AS
23. Olson S
24. Shenker S
25. Sturgill D
26. Samsonova AA
27. Weiszmann R
28. Robinson G
29. Hernandez J
30. Andrews J
31. Bickel PJ
32. Carninci P
33. Cherbas P
34. Gingeras TR
35. Hoskins RA
36. Kaufman TC
37. Lai EC
38. Oliver B
39. Perrimon N
40. Graveley BR
41. Celniker SE
(2014) Diversity and dynamics of the Drosophila transcriptome
Nature 512:393–399.
https://doi.org/10.1038/nature12962
- Google Scholar
(2021) Ageing as “early-life inertia”: Disentangling life-history trade-offs along a lifetime of an individual
Evolution Letters 5:551–564.
https://doi.org/10.1002/evl3.254
- PubMed
- Google Scholar
(2020) Experimental evolution of post-ingestive nutritional compensation in response to a nutrient-poor diet
Proceedings. Biological Sciences 287:20202684.
https://doi.org/10.1098/rspb.2020.2684
- PubMed
- Google Scholar
(2023) Experimental evolution of metabolism under nutrient restriction: enhanced amino acid catabolism and a key role of branched-chain amino acids
Evolution Letters 7:273–284.
https://doi.org/10.1093/evlett/qrad018
- PubMed
- Google Scholar
1. Cohen E
2. Sawyer JK
3. Peterson NG
4. Dow JAT
5. Fox DT
(2020) Physiology, development, and disease modeling in the Drosophila excretory system
Genetics 214:235–264.
https://doi.org/10.1534/genetics.119.302289
- PubMed
- Google Scholar
1. Collet J
2. Fellous S
(2019) Do traits separated by metamorphosis evolve independently? Concepts and methods
Proceedings. Biological Sciences 286:20190445.
https://doi.org/10.1098/rspb.2019.0445
- PubMed
- Google Scholar
(2016) Nutritional limitation in early postnatal life and its effect on aging and longevity in rodents
Experimental Gerontology 86:84–89.
https://doi.org/10.1016/j.exger.2016.05.001
- PubMed
- Google Scholar
(2014) Short- and long-term effects of a maternal low-protein diet on ventilation, O₂/CO₂ chemoreception and arterial blood pressure in male rat offspring
The British Journal of Nutrition 111:606–615.
https://doi.org/10.1017/S0007114513002833
- PubMed
- Google Scholar
1. de Magalhães JP
(2012) Programmatic features of aging originating in development: aging mechanisms beyond molecular damage?
FASEB Journal 26:4821–4826.
https://doi.org/10.1096/fj.12-210872
- PubMed
- Google Scholar
(2010) Prenatal undernutrition and cognitive function in late adulthood
PNAS 107:16881–16886.
https://doi.org/10.1073/pnas.1009459107
- PubMed
- Google Scholar
(2011) Procedures for large-scale metabolic profiling of serum and plasma using gas chromatography and liquid chromatography coupled to mass spectrometry
Nature Protocols 6:1060–1083.
https://doi.org/10.1038/nprot.2011.335
- PubMed
- Google Scholar
(2017) Adaptation to chronic nutritional stress leads to reduced dependence on microbiota in Drosophila melanogaster
mBio 8:e01496-17.
https://doi.org/10.1128/mBio.01496-17
- PubMed
- Google Scholar
1. Gems D
2. Partridge L
(2013) Genetics of longevity in model organisms: debates and paradigm shifts
Annual Review of Physiology 75:621–644.
https://doi.org/10.1146/annurev-physiol-030212-183712
- PubMed
- Google Scholar
1. Ghalambor CK
2. Hoke KL
3. Ruell EW
4. Fischer EK
5. Reznick DN
6. Hughes KA
(2015) Non-adaptive plasticity potentiates rapid adaptive evolution of gene expression in nature
Nature 525:372–375.
https://doi.org/10.1038/nature15256
- PubMed
- Google Scholar
(2009) Amino-acid imbalance explains extension of lifespan by dietary restriction in Drosophila
Nature 462:1061–1064.
https://doi.org/10.1038/nature08619
- PubMed
- Google Scholar
1. Gu X
2. Jouandin P
3. Lalgudi PV
4. Binari R
5. Valenstein ML
6. Reid MA
7. Allen AE
8. Kamitaki N
9. Locasale JW
10. Perrimon N
11. Sabatini DM
(2022) Author Correction: sestrin mediates detection of and adaptation to low-leucine diets in Drosophila
Nature 609:209–216.
https://doi.org/10.1038/s41586-022-05286-9
- PubMed
- Google Scholar
1. Gupta V
2. Frank AM
3. Matolka N
4. Lazzaro BP
(2022) Inherent constraints on a polyfunctional tissue lead to a reproduction-immunity tradeoff
BMC Biology 20:127.
https://doi.org/10.1186/s12915-022-01328-w
- PubMed
- Google Scholar
1. Hamilton WD
(1966) The moulding of senescence by natural selection
Journal of Theoretical Biology 12:12–45.
https://doi.org/10.1016/0022-5193(66)90184-6
- PubMed
- Google Scholar
1. Hollis B
2. Houle D
3. Yan Z
4. Kawecki TJ
5. Keller L
(2014) Evolution under monogamy feminizes gene expression in Drosophila melanogaster
Nature Communications 5:3482.
https://doi.org/10.1038/ncomms4482
- PubMed
- Google Scholar
(2016) Experimental evolution of gene expression and plasticity in alternative selective regimes
PLOS Genetics 12:e1006336.
https://doi.org/10.1371/journal.pgen.1006336
- PubMed
- Google Scholar
(2021) Adaptive and maladaptive expression plasticity underlying herbicide resistance in an agricultural weed
Evolution Letters 5:432–440.
https://doi.org/10.1002/evl3.241
- PubMed
- Google Scholar
1. Kanehisa M
2. Goto S
(2000) KEGG: kyoto encyclopedia of genes and genomes
Nucleic Acids Research 28:27–30.
https://doi.org/10.1093/nar/28.1.27
- PubMed
- Google Scholar
1. Kapila R
2. Kashyap M
3. Gulati A
4. Narasimhan A
5. Poddar S
6. Mukhopadhaya A
7. Prasad NG
(2021) Evolution of sex-specific heat stress tolerance and larval Hsp70 expression in populations of Drosophila melanogaster adapted to larval crowding
Journal of Evolutionary Biology 34:1376–1385.
https://doi.org/10.1111/jeb.13897
- PubMed
- Google Scholar
(1997) Role of mitochondrial and cytoplasmic serine hydroxymethyltransferase isozymes in de novo purine synthesis in Saccharomyces cerevisiae
Biochemistry 36:14956–14964.
https://doi.org/10.1021/bi971610n
- PubMed
- Google Scholar
1. Kawecki TJ
2. Erkosar B
3. Dupuis C
4. Hollis B
5. Stillwell RC
6. Kapun M
(2021) The genomic architecture of adaptation to larval malnutrition points to a trade-off with adult starvation resistance in Drosophila
Molecular Biology and Evolution 38:2732–2749.
https://doi.org/10.1093/molbev/msab061
- PubMed
- Google Scholar
(2020) The influence of developmental diet on reproduction and metabolism in Drosophila
BMC Evolutionary Biology 20:93.
https://doi.org/10.1186/s12862-020-01663-y
- PubMed
- Google Scholar
(2009) Life-history consequences of adaptation to larval nutritional stress in Drosophila
Evolution; International Journal of Organic Evolution 63:2389–2401.
https://doi.org/10.1111/j.1558-5646.2009.00718.x
- PubMed
- Google Scholar
1. Laland KN
2. Uller T
3. Feldman MW
4. Sterelny K
5. Müller GB
6. Moczek A
7. Jablonka E
8. Odling-Smee J
(2015) The extended evolutionary synthesis: its structure, assumptions and predictions
Proceedings. Biological Sciences 282:20151019.
https://doi.org/10.1098/rspb.2015.1019
- PubMed
- Google Scholar
1. Lande R
2. Arnold SJ
(1983) The measurement of selection on correlated characters
Evolution; International Journal of Organic Evolution 37:1210–1226.
https://doi.org/10.1111/j.1558-5646.1983.tb00236.x
- PubMed
- Google Scholar
1. Law CW
2. Chen Y
3. Shi W
4. Smyth GK
(2014) voom: Precision weights unlock linear model analysis tools for RNA-seq read counts
Genome Biology 15:R29.
https://doi.org/10.1186/gb-2014-15-2-r29
- PubMed
- Google Scholar
1. Li S
2. Yu X
3. Feng Q
(2019) Fat body biology in the last decade
Annual Review of Entomology 64:315–333.
https://doi.org/10.1146/annurev-ento-011118-112007
- PubMed
- Google Scholar
1. Lillycrop KA
2. Burdge GC
(2015) Maternal diet as a modifier of offspring epigenetics
Journal of Developmental Origins of Health and Disease 6:88–95.
https://doi.org/10.1017/S2040174415000124
- PubMed
- Google Scholar
(2013) Nutritional manipulations in the perinatal period program adipose tissue in offspring
American Journal of Physiology. Endocrinology and Metabolism 305:E1195–E1207.
https://doi.org/10.1152/ajpendo.00231.2013
- PubMed
- Google Scholar
1. May CM
2. Zwaan BJ
(2017) Relating past and present diet to phenotypic and transcriptomic variation in the fruit fly
BMC Genomics 18:640.
https://doi.org/10.1186/s12864-017-3968-z
- PubMed
- Google Scholar
(2019) Adaptation to developmental diet influences the response to selection on age at reproduction in the fruit fly
Journal of Evolutionary Biology 32:425–437.
https://doi.org/10.1111/jeb.13425
- PubMed
- Google Scholar
Book
1. Medawar PB
(1952)
An Unsolved Problem of Biology

London: H. K. Lewis.
- Google Scholar
(2020) Single-Step extraction coupled with targeted HILIC-MS/MS approach for comprehensive analysis of Human Plasma Lipidome and Polar Metabolome
Metabolites 10:495.
https://doi.org/10.3390/metabo10120495
- PubMed
- Google Scholar
1. Moczek AP
2. Sultan S
3. Foster S
4. Ledón-Rettig C
5. Dworkin I
6. Nijhout HF
7. Abouheif E
8. Pfennig DW
(2011) The role of developmental plasticity in evolutionary innovation
Proceedings. Biological Sciences 278:2705–2713.
https://doi.org/10.1098/rspb.2011.0971
- PubMed
- Google Scholar
1. Moran NA
(1994) Adaptation and constraint in the complex life cycles of animals
Annual Review of Ecology and Systematics 25:573–600.
https://doi.org/10.1146/annurev.es.25.110194.003041
- Google Scholar
1. Neel JV
(1962)
Diabetes mellitus: a “thrifty” genotype rendered detrimental by “progress”?

American Journal of Human Genetics 14:353–362.
- PubMed
- Google Scholar
1. O’Brien DM
2. Min KJ
3. Larsen T
4. Tatar M
(2008) Use of stable isotopes to examine how dietary restriction extends Drosophila lifespan
Current Biology 18:R155–R156.
https://doi.org/10.1016/j.cub.2008.01.021
- PubMed
- Google Scholar
1. Partridge L
2. Barton NH
(1993) Optimality, mutation and the evolution of ageing
Nature 362:305–311.
https://doi.org/10.1038/362305a0
- PubMed
- Google Scholar
1. Perez G
(2017)
Cahiers Techniques de l’INRA

135–142, Une Méthode D'Analyse D'Image Automatique pour Quantifier Rapidement LES Nombres D'Œufs et LES Taux de Parasitisme Chez Trichogramma SP, Cahiers Techniques de l’INRA, Innovations Entomologiques: Du Laboratoire Au Champ: INRA.
- Google Scholar
1. Pigliucci M
2. Murren CJ
(2003) Perspective: Genetic assimilation and a possible evolutionary paradox: can macroevolution sometimes be so fast as to pass us by?
Evolution; International Journal of Organic Evolution 57:1455–1464.
https://doi.org/10.1111/j.0014-3820.2003.tb00354.x
- PubMed
- Google Scholar
1. Piper MDW
2. Soultoukis GA
3. Blanc E
4. Mesaros A
5. Herbert SL
6. Juricic P
7. He X
8. Atanassov I
9. Salmonowicz H
10. Yang M
11. Simpson SJ
12. Ribeiro C
13. Partridge L
(2017) Matching dietary amino acid balance to the in silico-translated exome optimizes growth and reproduction without cost to lifespan
Cell Metabolism 25:1206.
https://doi.org/10.1016/j.cmet.2017.04.020
- PubMed
- Google Scholar
1. Pischedda A
2. Chippindale AK
(2006) Intralocus sexual conflict diminishes the benefits of sexual selection
PLOS Biology 4:e356.
https://doi.org/10.1371/journal.pbio.0040356
- PubMed
- Google Scholar
(2005) Insights from the developing world: thrifty genotypes and thrifty phenotypes
The Proceedings of the Nutrition Society 64:153–161.
https://doi.org/10.1079/pns2005421
- PubMed
- Google Scholar
1. Rice WR
(1984) Sex chromosomes and the evolution of sexual dimorphism
Evolution; International Journal of Organic Evolution 38:735–742.
https://doi.org/10.1111/j.1558-5646.1984.tb00346.x
- PubMed
- Google Scholar
1. Ritchie ME
2. Phipson B
3. Wu D
4. Hu Y
5. Law CW
6. Shi W
7. Smyth GK
(2015) limma powers differential expression analyses for RNA-sequencing and microarray studies
Nucleic Acids Research 43:e47.
https://doi.org/10.1093/nar/gkv007
- PubMed
- Google Scholar
(2010) edgeR: a Bioconductor package for differential expression analysis of digital gene expression data
Bioinformatics 26:139–140.
https://doi.org/10.1093/bioinformatics/btp616
- PubMed
- Google Scholar
(2019) Complete metamorphosis of insects
Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences 374:20190063.
https://doi.org/10.1098/rstb.2019.0063
- PubMed
- Google Scholar
(2006) The Dutch famine and its long-term consequences for adult health
Early Human Development 82:485–491.
https://doi.org/10.1016/j.earlhumdev.2006.07.001
- PubMed
- Google Scholar
1. Salway JG
(2018) The krebs uric acid cycle: a forgotten krebs cycle
Trends in Biochemical Sciences 43:847–849.
https://doi.org/10.1016/j.tibs.2018.04.012
- PubMed
- Google Scholar
1. Schaerli Y
2. Jiménez A
3. Duarte JM
4. Mihajlovic L
5. Renggli J
6. Isalan M
7. Sharpe J
8. Wagner A
(2018) Synthetic circuits reveal how mechanisms of gene regulatory networks constrain evolution
Molecular Systems Biology 14:e8102.
https://doi.org/10.15252/msb.20178102
- PubMed
- Google Scholar
1. Scheiner SM
(1993) Genetics and evolution of phenotypic plasticity
Annual Review of Ecology and Systematics 24:35–68.
https://doi.org/10.1146/annurev.es.24.110193.000343
- Google Scholar
(2005) Geographic variation in diapause incidence, life-history traits, and climatic adaptation in Drosophila melanogaster
Evolution; International Journal of Organic Evolution 59:1721–1732.
https://doi.org/10.1554/05-115.1
- PubMed
- Google Scholar
(2004) Role and regulation of starvation-induced autophagy in the Drosophila fat body
Developmental Cell 7:167–178.
https://doi.org/10.1016/j.devcel.2004.07.009
- PubMed
- Google Scholar
(2016) Evolution of increased adult longevity in Drosophila melanogaster populations selected for adaptation to larval crowding
Journal of Evolutionary Biology 29:407–417.
https://doi.org/10.1111/jeb.12795
- PubMed
- Google Scholar
(2015) Differential analyses for RNA-seq: transcript-level estimates improve gene-level inferences
F1000Research 4:1521.
https://doi.org/10.12688/f1000research.7563.2
- PubMed
- Google Scholar
(2015) Intersecting transcription networks constrain gene regulatory evolution
Nature 523:361–365.
https://doi.org/10.1038/nature14613
- PubMed
- Google Scholar
(1988) Fatty acids in insects: composition, metabolism, and biological significance
Archives of Insect Biochemistry and Physiology 9:1–33.
https://doi.org/10.1002/arch.940090102
- Google Scholar
1. Storey JD
2. Tibshirani R
(2003) Statistical significance for genomewide studies
PNAS 100:9440–9445.
https://doi.org/10.1073/pnas.1530509100
- PubMed
- Google Scholar
(2020) Identification of blood cell transcriptome-based biomarkers in adulthood predictive of increased risk to develop metabolic disorders using early life intervention rat models
FASEB Journal 34:9003–9017.
https://doi.org/10.1096/fj.202000071RR
- PubMed
- Google Scholar
(2014) MRMPROBS suite for metabolomics using large-scale MRM assays
Bioinformatics 30:2379–2380.
https://doi.org/10.1093/bioinformatics/btu203
- PubMed
- Google Scholar
1. van der Velpen V
2. Teav T
3. Gallart-Ayala H
4. Mehl F
5. Konz I
6. Clark C
7. Oikonomidi A
8. Peyratout G
9. Henry H
10. Delorenzi M
11. Ivanisevic J
12. Popp J
(2019) Systemic and central nervous system metabolic alterations in Alzheimer’s disease
Alzheimer’s Research & Therapy 11:93.
https://doi.org/10.1186/s13195-019-0551-7
- PubMed
- Google Scholar
1. Veltsos P
2. Fang YX
3. Cossins AR
4. Snook RR
5. Ritchie MG
(2017) Mating system manipulation and the evolution of sex-biased gene expression in Drosophila
Nature Communications 8:2072.
https://doi.org/10.1038/s41467-017-02232-6
- PubMed
- Google Scholar
(2010) Effects of parental larval diet on egg size and offspring traits in Drosophila
Biology Letters 6:238–241.
https://doi.org/10.1098/rsbl.2009.0754
- PubMed
- Google Scholar
(2012) Chronic malnutrition favours smaller critical size for metamorphosis initiation in Drosophila melanogaster
Journal of Evolutionary Biology 25:288–292.
https://doi.org/10.1111/j.1420-9101.2011.02419.x
- PubMed
- Google Scholar
1. Vijendravarma RK
2. Kawecki TJ
(2013) Epistasis and maternal effects in experimental adaptation to chronic nutritional stress in Drosophila
Journal of Evolutionary Biology 26:2566–2580.
https://doi.org/10.1111/jeb.12248
- PubMed
- Google Scholar
(2015) Gut physiology mediates a trade-off between adaptation to malnutrition and susceptibility to food-borne pathogens
Ecology Letters 18:1078–1086.
https://doi.org/10.1111/ele.12490
- PubMed
- Google Scholar
1. Wagner A
(2011) Genotype networks shed light on evolutionary constraints
Trends in Ecology & Evolution 26:577–584.
https://doi.org/10.1016/j.tree.2011.07.001
- PubMed
- Google Scholar
(2016) Sestrin2 is a leucine sensor for the mTORC1 pathway
Science 351:43–48.
https://doi.org/10.1126/science.aab2674
- PubMed
- Google Scholar
1. Xia J
2. Psychogios N
3. Young N
4. Wishart DS
(2009) MetaboAnalyst: a web server for metabolomic data analysis and interpretation
Nucleic Acids Research 37:W652–W660.
https://doi.org/10.1093/nar/gkp356
- PubMed
- Google Scholar
(2012) Evolution of gene expression and expression plasticity in long-term experimental populations of Drosophila melanogaster maintained under constant and variable ethanol stress
Molecular Ecology 21:4287–4299.
https://doi.org/10.1111/j.1365-294X.2012.05697.x
- PubMed
- Google Scholar

Decision letter

Luisa Pallares

Reviewing Editor; Friedrich Miescher Laboratory, Germany
Detlef Weigel

Senior Editor; Max Planck Institute for Biology Tübingen, Germany
Benjamin R Harrison

Reviewer; University of Washington, United States

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Decision letter after peer review:

[Editors’ note: the authors submitted for reconsideration following the decision after peer review. What follows is the decision letter after the first round of review.]

Thank you for submitting the paper "Evolution under juvenile malnutrition impacts adult metabolism and fitness in Drosophila" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by Detlef Weigel as Senior Editor. The reviewers have opted to remain anonymous.

We are sorry to say that, after consultation with the reviewers, we have decided that in its current state, this work will not be considered for publication by eLife. However, the reviewers agreed that the work represents an extensive collection of datasets that addresses a very timely and important question in biology. We will therefore aim to contact the same editors and reviewers if you decide to make a new submission with a thoroughly revised manuscript.

Specifically, there was consensus in that the manuscript should (a) better contextualize the motivation/results/conclusions using previous data/theory, (b) tone down the overinterpretation of results, and (c) state clearly the biological aspect that is being tested with each experiment and argue whether the experiments/data used for each question are indeed appropriate.

We believe the reviews were very thorough and will help improve the manuscript. We therefore attach them below in their entirety.

Reviewer #1 (Recommendations for the authors):

In this paper, Erkosar et al. explore a fundamental question in evolutionary biology as well as in human health: what are the consequences of juvenile malnutrition for adult performance. They use Drosophila melanogaster, a holometabolous insect, as a study system. By undergoing full metamorphosis, it could be assumed that larval and adult stages have been decoupled in this species. This system therefore allows the authors not only to address a general question regarding juvenile vs adult effects, but also a more specific one regarding the uncoupling of larval and adult stages in this type of insect. A strength of this paper is the use of an experimental evolution set up with 6 replicates each for the selection and control regimes – this allowed the authors not only to study the effects of juvenile malnutrition within a generation but also the long-term effect in adults of 230 generations of larval malnutrition. Previous studies from the authors and others had characterized certain higher-order adult phenotypes (lifespan, fecundity, starvation, etc) in this and other selected populations, and now, Erkosar et al. gather metabolomics and transcriptomics data to connect genetic variation (from a previous study) to expression changes to metabolism to higher-order phenotypes like fecundity and starvation resistance. This type of complete dataset crossing different levels of organization in an organism is not common, and the authors deserve praise for that.

The paper makes two main points that are extremely interesting but should be taken with some caution:

First, selection on larvae malnutrition results in low-fitness adult phenotypes. Using a very complete dataset that combines genomics, transcriptomics, metabolomics, and higher-order phenotyping, it is shown that at all levels, the main physiological change underlying low fecundity and low starvation resistance is purine and amino acid metabolism. The big-picture question here is why adaptation to juvenile malnutrition should constrain adaptation in the adult stages in a holometabolous species. To explain this, the authors argue that larva and adult transcriptional profiles are indeed not decoupled, they are strongly correlated. And therefore, adaptive changes at the larval stage are carried over to the adult stage where they are not adaptive. Although the first part is based on solid data, the second part might not be:

1. The authors show that overall, the expression levels are correlated between larvae and adults (r = 0.42) and that most of the differentially expressed genes have the same sign in both stages. However, the genes that are associated with purine and amino acid metabolism in adults, which the authors argue underlie low starvation resistance and fecundity, are not differentially expressed in larvae. This questions whether the coupling argument can really explain the low-fitness phenotypes in the adult, given that data, it seems that this cannot be concluded. The authors should clarify.

2. In the methods section, the dataset for larvae gene expression used to argue the point above (Erkosar et al. 2017) is said to have been collected for larvae which "were raised on the poor diet". This means that the differential gene expression in that dataset corresponds to the plastic response of larvae to poor diet (i.e., selected larvae on poor diet vs control larvae on a poor diet) and not to the selected response (i.e., selected larvae on poor diet vs control larvae on a standard diet). If this is true, then the correlation estimated in the paper corresponds to plastic response in larvae vs selected response in adults, which will further prevent us from concluding that the expression profile of selected larvae and adults are coupled. The authors should clarify the dataset that was used for this comparison, and how this can change the interpretation of the results.

The second big (in my opinion) conclusion of the study is that plastic response to larvae malnutrition is independent of the selection response to the same stimuli. This sounds counterintuitive but is supported by two independent datasets, gene expression and metabolomics, which suggests it is a solid result. Despite such an interesting pattern, the authors do not really digest this result for the reader, nor contextualize it in any way, despite being other studies exploring the relationship between plastic and adaptive response (e.g. in fish Ghalambor et al. 2015, in flies Huang and Agrawal 2016 Plos Gen, in plants Josephs et al. 2021 Evol letters, etc). The data, nor the discussion of it, help connect the dots: there is a correlation between larval expression profile and adult selected profile, but no correlation with the intermediate step between these two which is the plastic response in the adult?

So, I am left wondering whether conceptually the authors are asking the wrong question here. Given that adults never experience any change in diet, they cannot possibly be addressing plasticity with the experimental design used here? Maybe actual plasticity can only be measured in larvae that experience both diets? Or maybe everything makes sense? In any case I think the authors should discuss better this result for it to be meaningful.

1. It seems like the 'coupling' of larvae and adult expression profiles is a surprising result. However, despite two lines here and there, this is not really contextualized in the text. For someone not working on insects, there is no way of weighing this result. The authors should improve the description of (a) what are the expectations regarding coupling/decoupling and what is previously known about this, (b) the biology of metamorphosis. For example, there are certain things that are "locked" during larval development like size (as accounted for during the ovarioles analysis) and therefore larva and adult phenotypes are coupled. What other things are "locked" during larval development and how can they explain the 'coupled' patterns of gene expression you see? E.g. are some of those pathways things that we should expect to be decoupled/coupled?

2. The result section should be re-organized to bring together data that addresses the same question. For example, the first section 'gene expression patterns point to evolutionary changes in metabolism" uses half of the space describing the comparison selection regime vs plasticity and only at the end addresses the evolutionary changes in metabolism. And then, the second section does discuss the overlap of plasticity and evolution regime but half of this comparison had been described in the first section. This happens again in other result sections and makes the paper harder to read (e.g. pages 7 and 9).

3. Page 4. You say the first PCs don't separate the groups and the PCA is not shown. Could you please add this as a SI figure, and state what is driving those first 3PCs before the signal you are interested in pops up in PC4-5.

4. Page 6, last line. It cites Figure 2 but that PCA is not shown in that figure or anywhere else. This figure will be interesting to put next to the PCA for expression (point 4 above) given that the expression signal is weaker than the metabolomics in separating treatments. Also, you should discuss why this is so in the discussion.

5. First line of page 5. I think it is standard vs poor diet instead of selected and control?

6. Parts of the methods are described in the Results sections, which is OK, but then, that information is omitted from the methods. Please add all relevant info to the methods section.

7. Just a suggestion of course. I think the title does not do justice to the article. It is already known that juvenile malnutrition impacts adult phenotypes. But you went well beyond that! What about something like "evolution under juvenile malnutrition show that larval and adult stages are not decoupled in D. melanogaster resulting in low-fitness adults".

In general, the paper will benefit a lot from better visualization of the results. There is too much text that could go into plots:

8. It will help the reader a lot to have a diagram of your experimental setup. It's hard to imagine plastic response vs selection because the selection regime also experiences plastic response every generation when adults are placed in standard food. You should make a drawing of the life stages of Drosophila and to which diet each one is exposed, and highlight the differences between selection regimes (230 generations, etc) and the plasticity experiment.

9. The overlap between selection and plasticity for gene expression and metabolomics could be made into a plot that shows the number of genes that overlap and the direction of change. Something like x-axis is selection lfc and y-axis is plasticity lfc.

10. Figure 1 could have more information given the first two sections of the results. Panel A is OK, Panel B, instead of showing up and down-regulated for selection regime could instead compare GO terms between selection regime and plasticity – again to highlight that nothing overlaps except for cuticle GO. It will be useful to see the p-values of the enrichment. The plot mentioned in 8 could go as a panel here.

11. Page 6 first section. This could also be made into a plot comparing lfc in larva and adults. And also add the comparison of the GO terms because here there are interesting differences.

12. Page 7 and 9, The results could be shown in plots of PCA, the comparison of lfc between selection and plasticity, and GO terms.

Reviewer #2 (Recommendations for the authors):

Nutritional stress early in life is pervasive and has a wide range of negative effects on adult physiology. Whether the genes underlying the physiological response that allows organisms to overcome stress early in life also enable adult success remains poorly understood. The evolutionary response may favor plastic strategies, where flexibility in response to different stress increases under early-life stressors. Alternatively, genetic variants that enable larval survival may have pleiotropic effects that negatively impact adult fitness. Untangling whether these effects are independent or interactive requires clever experimental design in a tractable system.

Here, the authors use experimental evolution in Drosophila melanogaster to understand evolution to poor nutrition early in life. For >200 generations, larvae were reared in a nutrient-poor diet (SEL diet, 1/4 the concentration of the standard diet). Then, after eclosion, flies are reared under a standard diet. The control (CTL) population was reared on the standard diet for the entire lifespan. For the results presented here, Erkosar et al. performed a reciprocal transplant experiment, where both the control (CTL) and selected (SEL) flies were reared in both CTL and SEL diets. This enabled comparisons between the effects of the evolutionary regime (CTL and SEL flies) and plasticity by varying the within-generation dietary effects (CTL and SEL diet). Combining RNAseq, metabolomics, and genomics with experimental evolution provided a comprehensive measure of how the evolutionary response to dietary stress shaped fly physiology.

A key strength of the study is the experimental design and comprehensive phenotyping. By rearing SEL and CTL lines in both SEL and CTL dietary regimes, the effects of evolution and plasticity can be better understood. In general, the RNAseq and metabolome results suggest that the evolutionary regime and dietary regime act additively in shaping adult physiology, but increased plasticity did not evolve (as no significant interactions were observed). One concern is that gene expression analyses was limited to the adult fat body, where the top hits were primarily associated with cuticular development (SEL flies are smaller). Additional tissues or at different time points in development might provide clearer insights into how evolution under dietary stress shapes gene expression and metabolism. Metabolomics also highlighted the role of purine metabolism. Purine metabolism was upregulated in SEL flies, which were also associated with SNPs that differentiate the genomic response between SEL and CTL lines.

A particularly interesting experiment was to starve the flies and profile the resulting changes to the metabolome. The SEL lines reared on the CTL diet still had a metabolome that more closely resembled the CTL fly after starvation, suggesting that the SEL flies exist in a permanently starved state, regardless of the current diet. The consequences of this permanently starved state also impact fecundity. SEL flies had lower fecundity than CTL flies on the SEL diet, counter to expectations that SEL lines would show greater fecundity in the SEL diet. SEL flies are thus paying a fitness cost for the traits that enabled survival in nutrient-poor diets. However, fecundity was only assayed in the SEL diet, and so it is unknown whether the reduction in fecundity for SEL flies is in response to selection or the diet.

Overall, the authors propose that the results support a scenario where the cross-life stage, pleiotropy drove adaptation to the dietary stress. In other words, SEL flies pay a fitness cost late in life for the strategies used to survive the nutritional stress early in life, with signatures of selection at the genomic, transcriptomic, and metabolomic levels. Particularly for organisms like Drosophila that inhabit ephemeral and fluctuating environments, plasticity is thought to evolve to decouple the physiological needs that differ between adults and juveniles. However, here, through experimental evolution and 'omics rich data, Eroksar et al. provide new insights into the role of pleiotropy in shaping adaptation to early life stress.

I really appreciated the multifaceted approach to understanding adaptation to nutritional stress early in life. The experiments are necessarily complex, and the manuscript is very rich in data-but, I had a difficult time throughout keeping track of how the different experiments linked to different results/conclusions/implications, etc.

One issue is that plasticity wasn't well defined to begin in the introduction. The authors begin at line 46 with the statement "In contrast to such phenotypically plastic within-generation responses", but is not clear how the first paragraph showed phenotypic plasticity. The second paragraph suggests that one can examine plasticity by placing larvae in a different environment than adults. However, this is also the selection regime, where only larvae are reared on nutrient-poor diets and adults are reared on the same standard diet. Plasticity can be assessed in different ways (e.g., change in phenotypic variance, intragenotypic differences, reaction norms, etc.), and so a clearer definition would help ground the complexity of the following experiments. Further, does the starvation experiment count as plasticity in this framework? Clarification would help the reader understand and appreciate the complex experiments that are necessary to address this question.

While the description of the five questions (lines 78-103) helps provide some context in how the reviewers are thinking about plasticity, it was disconnected with the subject material in the introduction and introduced further complications like pleiotropy. It might help by having a simpler statement in the introduction describing how plasticity can be assessed using the selected and control populations.

Another general comment is the error bar in Figure 3B, Figure 6B, etc. It is a little unusual to see just one error bar per panel in the figures. I also see that the error bar is the standard error of the mean-but mean of the main effect (evolutionary regime)? I see the placement is zero centered for the log fold change, which is OK, but what is it for Figure 6B? An additional explanation would be appreciated.

Many of the experiments need additional rationale to describe them. While I got this to some extent in the introduction lines 78-103, the rationale and implications are underdeveloped for several of the key results.

In the experiment associated with Figure 1, why are fat-body/carcasses used? It is described that the fat body is the key metabolic organ, but why wouldn't the whole body be used? For example, given the fecundity results presented later, how does the fat body transcriptome inform these results as opposed to what ovary or whole-body results would show? What is lost by focusing just on this one tissue group? If cuticular development and maturation genes are still most important, wouldn't that still show whole body RNAseq? Was the choice to enable linking the RNAseq with the metabolomics that follows? The rationale should be better clarified.

I'm also not sure I understood also why only the evolutionary regime differential expression results were shown, but not the larval diet results (summarized lines 162-188). I found the lack of parallelism an interesting result from the RNAseq experiment and would find them useful visualized (not as a table) at least in the supplement.

The comparison with Erkosar et al. 2017 was not well justified. In the methods, lines 614-15 it is mentioned that the larval gene expression is derived from an experiment that manipulated the microbiome, with and without microbes. The evolutionary regime was used as the main effect for comparison, but this isn't totally straightforward-why not just subset the 2017 data for only +microbe? why include the germfree at all here? Also, Erkosar 2017 data is from generations 177-200, which isn't stated here. It's very surprising that the same genes are differentially expressed in mostly the same way between larvae and adult carcasses given all these differences in experimental design. The differences between these datasets should be explicit for the reader. However, given the complications listed above, I feel like the comparison between the carcass data and the Erkosar data does not add much to the overall message (and it's not visualized).

I liked the starvation experiment. Like before, it needs to be clear whether plasticity here refers to the std-poor diet and/or fed-starved comparison. It can be both but needed more information here. I think the authors consider the results to show more about the evolutionary regime (lines 373-375), but why isn't this also an aspect of plastic responses to nutritional stress? The interactions between fed-starved and the connection with differential allocation are very interesting results. While most of the discussion of this result is focused on the SEL resembling a starved metabolome, the change for SEL between fed and starved seems less than for the CTL. The CTL-poor seems to change most dramatically between fed and starved, which makes sense given the previous results of this selection experiment. Does this mean that SEL can buffer the nutritional stress better? I'm curious how buffering of this starvation stressor, despite previous evidence that SEL are less starvation resistant (line 331), informs our understanding of the adaptation to nutritionally poor larval diets.

Great to see some overlap between genomic, transcriptomic, and metabolomics in purine synthesis. Please make clear that the Kawecki 2021 genomic data is from ~150 generations.

For the adult costs of larval adaptation, it is interesting that CTL laid more eggs than SEL populations when exposed to a poor larval diet. Would you see the same result if fecundity was assayed for both on the standard diet? i.e., does the evolutionary regime just lead to generally fewer eggs produced by SEL populations? Does it matter for adaptation to nutritionally poor diets if SEL populations are just less fecund? I think the rationale needs some additional justification here. Additionally, in Figure 5, can these be visualized in a way that shows the true range of observed values (e.g., boxplots)?

I think the discussion provides important context, and the authors highlight one potential explanation for the results is pleiotropy across life stages. This is an interesting perspective given that Drosophila are holometabolous insects, and as the authors say, this remains surprisingly understudied. It's not clear whether this cross-life-stage pleiotropy is amplified because of holometabolous insects or also generalizable to other organisms as well.

Reviewer #3 (Recommendations for the authors):

Erkosar et al. use gene expression and metabolomic analyses to investigate the evolutionary change that accompanied over 230 generations of selection by nutrient deprivation from egg to pupation in Drosophila.

The evolved populations were six replicate populations, each selected for larval development and growth under poor nutrient conditions. Each population was kept at ~200 flies at each generation, which may have allowed significant drift when compared to larger populations in similar experimental evolutionary studies however, this regime has resulted in repeated evolution of a few interesting traits. These traits were revealed in earlier work and include faster developmental time, higher egg-to-adult viability, and smaller adult size. Additionally, selected lines showed higher sensitivity to starvation as adults and the allele frequency changes that accompanied selection intersect with those loci associated with selection for starvation resistance.

The picture painted from previous work is of genotypes arising from selection that are more likely to make pupae/adults; to make them faster and more numerously than the control population. The adults that they make however are not as healthy as the adults from the control population. This certainly suggests that adult traits suffer a constraint based on the evolution of juvenile condition. Is that constraint manifest in patterns of gene expression, or in patterns of variation in the metabolome? Hints at metabolic adaptation within these populations came from a prior publication that described a difference in nitrogen or carbon assimilation between these populations, which is at least intriguing.

The current work is important in that it sheds light on the kinds of endophenotypic changes that accompany these evolutionary changes. In addition to the condition-specific patterns of expression and metabolite abundance, the authors also test their ideas about how the transcriptome and metabolome of adults of both populations respond to diet, and if there is a resemblance to flies evolved on poor food and starved flies.

This paper presents several interesting results. One, the evolution that occurred in the larval transcriptome was largely similar to the differences between the populations as adults. This implies a constraint in adult phenotypes on juvenile phenotypes and vice versa that is reflected in the transcriptome.

Within the same experiment, the transcriptome and metabolome of adults from both populations appear to respond similarly to diet. That is, the adults of both populations, despite significant shifts in metabolome and transcriptome, show very similar responses to being placed on the poor diet, when compared to the normal diet. This is an unexpected result given the constraint that may explain the previous result.

That the within-generation response to diet appears independent of the evolutionary response to diet conditions is interesting. The authors could develop their discussion of these results further. I think it is worth considering that the effects of diet with a generation are not necessarily adaptive. An adult on a poor diet might have a metabolome that looks a certain way, reflecting in part the exhaustion of resources. The adult from a larva that has evolved on a poor diet however may more closely approximate an adapted metabolome, there is really no reason to think that these would be the same.

It is also interesting that the effect of diet on the adult metabolome was somewhat similar to the difference between the adult metabolome of the two populations, the evolutionary change. This is in contrast to the seemingly independent effect of diet on the transcriptome, and the transcripts that differ between the populations. I hope that this result is not a 'false positive' due to the small biological sample size of the transcriptome data (n=1) relative to the metabolome data (n=2), or a 'false negative' due to the relatively sparse metabolome coverage (~100 metabolites) that is typical of targeted analysis of fly metabolome compared to the more comprehensive transcriptome. This result is in line with other studies that indicate that endophenotype appears to converge as it goes from: gene > RNA > protein > metabolite, to phenotype. The relative anticipation of the evolutionary response seen in the metabolome differences by diet when compared to the same in a relationship in the transcriptome is neat.

The results described above manifest largely in PCA, in reduced dimensional space, which is appropriate and telling, but may hide variation that occurs within and between populations, and across diets. There is some agreement within the univariate analysis of metabolites and transcripts that are differentially abundant between populations and treatments, suggesting that the 'biological signal' seen in the PCA could be picked apart using the identities of the genes and metabolites. The authors indulge in this pursuit and while creative, take their interpretation too far. The authors make several clever interpretations of their data in order to shed light on potential mechanisms (pathway activity) that may explain their results. Without testing some of these ideas however this amounts to speculation.

The third point of interest is the "Metabolic profile of Selected flies tends towards starved-like state", an argument that the metabolome of selected flies more closely resembles starved flies than does the metabolome of control flies. This is argument rests on MANOVA and some agreement in sign with metabolite levels upon starvation, and those affected by selection. This is an interesting result that could relate to the sensitization to adult starvation in the evolved population. At the univariate level however, the picture is quite complicated (Figure 3B). The interactions between diet, selection regime and the effect of starvation are complex. The authors make a valiant effort at interpretation, however, I don't think their conclusions are well supported from these data. I suggest approaching the univariate analysis more circumspectly.

The authors are over-interpreting the meaning of their data. This is my main objection to what is otherwise an interesting and meaningful study. I think that the authors should include their interpretation of the data, however, it should be presented as speculation, rather than as if the interpretation were itself an observation.

The joint analysis presented in 'Figure 4—figure supplement 1' is problematic for several reasons. I don't think it should be included as it does not add anything to the paper worth discussing.

"[abstract line # 26]: "resulting in deficiency of electron transports and congestion in β-oxidation." Two objections, first, 'deficiency' is different from 'reduced', we have no idea if the reduced level of these RNAs is at all affecting the trait, these are associations and cause and effect is not known. Second, RNA is not protein, and even then, protein abundance is not activity, so either discuss these results in a more speculative light or make a more direct and independent test of electron transport activity/function. Further, congestion in β oxidation is certainly a possible explanation for the patterns of acetylcarnitines and other FA-derivatives that are more/less abundant, however, this too is a speculative explanation and ignores the equally valid evidence that this explanation is wrong.

Similarly, the section that discusses the amino acids [#492-505] describes some amino acids as 'deficient' in the selected flies. These amino acids are reduced in abundance compared to the control flies, but this is not the same as deficient. Deficiency implies that if there were more of these then the outcome would change. That result is unknown. The data are describing an association with abundance, concluding that this is a deficiency is misleading.

In line #385, "This analysis confirmed that the changes in amino acid and purine metabolism contributed to adaptation to poor larval diet." Confirming that something contributes, implies that you have gone beyond association and tested causality. These data do not demonstrate cause or effect but bolster a previously-identified association.

There are two sections of the results (lines 217-290, and 378-404) dedicated to the pathways that might be involved in the transcriptome and metabolome differences seen in these populations. I suggest paring this down, at least as a part of the Results section, as enough of it is speculation. Lines #378-404 are mostly speculative and so this argument should move to the discussion. Similarly, the discussion could be pared down given some of the contradictions that emerge when the data are discussed. For instance, if the paragraphs from lines #492-505 resolves with the axiom: "the differences in amino acid abundance between Selected and Control populations is consistent with differential use of amino acids in metabolism – their interconversion, catabolism and use in the synthesis of other metabolites.", then I'm not sure that it's a good use of space.

Figure 6 could be idealized (made with generic genes) or excluded. I say this because there are at least 10,000 genes from which to pull patterns like these from, so seeing these genes like this does not have meaning (FDR is high). The idea behind this figure can certainly be discussed as it is in the discussion, but this data set is underpowered for this kind of interference.

The authors use the term pleiotropic at several points. I think they are referring to something like antagonistic pleiotropy. I think a short section that introduces this term in this way, with reference to the theoretical or experimental work on the idea in relation to the evolution of aging perhaps would help the reader.

In the bigger picture, there are 'inward' and 'outward' implications for this work. The authors present support for their conclusion that negative trade-offs in adulthood are tied to the endophenotypic changes that occur in juveniles as a result of selection when only applied to juveniles. This support comes from the within-population similarity in adult transcriptome and metabolome to that of larvae. The 'inward' perspective, while tantalizing, is perhaps over-presented in the discussion of these results. Insight into the mechanisms (cellular pathways) by which these trade-offs happen can and should be included, however there are also wider implications of these results for insect evolution, the 'outward' perspective. These results suggest that the evolution of juvenile phenotypes is constrained by the outcomes they associate within the adult, at least on juvenile nutrition inputs/needs. The authors cite some theoretical papers on the topic, and I think the reader would be better served if the authors could reflect on the implications of the constraint they appear to have found. What does this imply for the phenotypic (and environmental) space that the juveniles are constrained to? We know that Drosophila are very choosy when it comes to oviposition, and here we see one of the reasons why that might be the case. Do the authors think that adult choice has evolved in part to deal with this constraint? It's easy to see the negative effects, the burden, of such a constraint. Evolution is powerful though, so why then is such constraint 'allowed'? Could this constraint actually reflect some kind of as-yet unconsidered benefit? For instance, does a constrained gene-trait map allow Drosophila to maintain a smaller, simpler, genome? I am not familiar with the literature on this and other topics that come to mind when the weight of these results are realized. Perhaps the authors could spend some of the discussion to share their interpretation?

There are a few passages that I found hard to understand. When describing part of their hypotheses about the evolution of adult and juvenile phenotypes, they write [#82-85]: "Plastic responses that program metabolism adaptively for a poor nutritional environment in adulthood would be misdirected in the Selected populations because the adults were switched to standard diet; evolution should thus counteract them." Why would plastic responses in adults be affected by selection in juveniles? If they were affected, why would we call them plastic?

Keep in mind that this sentence is followed by [#85-87]: "Conversely, plastic responses that alleviate the consequences of having developed on a poor diet irrespective of adult diet should have become amplified as a result of evolution on the poor diet." It sounds like the authors are predicting that two counteracting effects have both happened and this does not make sense to me. I would like to understand the authors' point here and I would hope that it could be clarified in a revision.

There were several references to the figures that don't seem correct, and other parts of the figure referencing scheme that did not make sense. For example, the reference to 'Figure 2' on line #231, refers to a PC plot, and not the heatmap shown in Figure 2. Line #231 refers to Supplementary Figure file 4, and I'm not sure that's correct. Please make sure that the figure references point the reader to the correct figure and that the figures are referred to and arranged in numerical order.

The experimental design, the sampling scheme and batch design of the analyses were all good. These experiments made the most of the number of samples that were used. The statistical analysis as described is appropriate, however there is enough going on that making the code available would probably help raise the impact of this work.

The authors are clearly aware of the multiple testing problem and I stand by my recommendation to remove figures that make use of data that do not survive FDR correction.

A point I will need clarification on occurs in lines [#601-604]: 'Therefore, we also estimated the number of genes that differed in expression due to each factor in the analysis as the total number of genes minus the estimated number of "true nulls" (Storey and Tibshirani 2003) (https://bioconductor.org/packages/topGO/), using the weight method and Fisher's exact test.' What does this mean? By 'factor in the analysis', I assume you mean regime and diet. Why, and how, would you use topGO to evaluate the number of genes affected by regime or diet?

https://doi.org/10.7554/eLife.92465.sa1

Author response

[Editors’ note: the authors resubmitted a revised version of the paper for consideration. What follows is the authors’ response to the first round of review.]

Reviewer #1 (Recommendations for the authors):

In this paper, Erkosar et al. explore a fundamental question in evolutionary biology as well as in human health: what are the consequences of juvenile malnutrition for adult performance. They use Drosophila melanogaster, a holometabolous insect, as a study system. By undergoing full metamorphosis, it could be assumed that larval and adult stages have been decoupled in this species. This system therefore allows the authors not only to address a general question regarding juvenile vs adult effects, but also a more specific one regarding the uncoupling of larval and adult stages in this type of insect. A strength of this paper is the use of an experimental evolution set up with 6 replicates each for the selection and control regimes – this allowed the authors not only to study the effects of juvenile malnutrition within a generation but also the long-term effect in adults of 230 generations of larval malnutrition. Previous studies from the authors and others had characterized certain higher-order adult phenotypes (lifespan, fecundity, starvation, etc) in this and other selected populations, and now, Erkosar et al. gather metabolomics and transcriptomics data to connect genetic variation (from a previous study) to expression changes to metabolism to higher-order phenotypes like fecundity and starvation resistance. This type of complete dataset crossing different levels of organization in an organism is not common, and the authors deserve praise for that.

We greatly appreciate this praise from the reviewer. To strengthen the evidence for evolutionary non-independence of larval and adult physiology, we now added a comparison with a new data set from larval metabolomics. These data and their analysis from the viewpoint of larval adaptation are published elsewhere (https://doi.org/10.1093/evlett/qrad018), but we used it here to test if we see a similar correlation between the effects of genetically-based evolutionary change on larval and adult metabolite abundance as we see for gene expression. We do (Figure 5B). Furthermore, with this new data set we could also ask if the phenotypically plastic responses of larvae and adults to larval diet are correlated – and they are not (Figure 5 —figure supplement 1). This implies that differences in adult metabolite abundance cannot be explained by some kind of general metabolic "inertia", where difference accrued during larval stage persist into adult stage. I.e., the metabolome differences between Selected and Control populations are more likely to be mediated by the expression of differentiated genetic variants during the adult stage rather than by lingering consequences of expression of these variants during the larval stage.

We believe these new analyses strengthen the case for the interpretation that a large fraction of genetic variants favored by selection on poor larval diet affect larval and adult metabolism similarly, constraining their independent evolution.

The paper makes two main points that are extremely interesting but should be taken with some caution:

First, selection on larvae malnutrition results in low-fitness adult phenotypes. Using a very complete dataset that combines genomics, transcriptomics, metabolomics, and higher-order phenotyping, it is shown that at all levels, the main physiological change underlying low fecundity and low starvation resistance is purine and amino acid metabolism. The big-picture question here is why adaptation to juvenile malnutrition should constrain adaptation in the adult stages in a holometabolous species. To explain this, the authors argue that larva and adult transcriptional profiles are indeed not decoupled, they are strongly correlated. And therefore, adaptive changes at the larval stage are carried over to the adult stage where they are not adaptive. Although the first part is based on solid data, the second part might not be:

1. The authors show that overall, the expression levels are correlated between larvae and adults (r = 0.42) and that most of the differentially expressed genes have the same sign in both stages. However, the genes that are associated with purine and amino acid metabolism in adults, which the authors argue underlie low starvation resistance and fecundity, are not differentially expressed in larvae. This questions whether the coupling argument can really explain the low-fitness phenotypes in the adult, given that data, it seems that this cannot be concluded. The authors should clarify.

We reported in the previous version that the genes that were differentially expressed between SEL and CTL in larvae were not enriched in amino acid and purine metabolism GO terms.

Prompted by the reviewer's comment, we had a closer look at genes involved in amino acid and purine metabolism in the larval RNAseq data. In fact, we found that 18 out of 99 genes in GO "α amino acid metabolic process" and 16 out of 107 in GO "purine-containing compound biosynthetic process" were significantly different between Selected and Control larvae at 10% FDR. These numbers are greater than the corresponding numbers for adults (9 out of 95 for amino acid metabolism genes, 7 out of 78 for purine biosynthesis). These GO terms are not enriched for the larvae because in general many more differentially expressed genes found in the larvae (21%) than in the adults (2.5%). Hence, while the amino acid metabolism and purine synthesis do not show disproportionate changes in gene expression in the larvae compared to other GO terms, the expression of multiple genes in those pathways has clearly been affected. Furthermore, although very few of those genes pass the 10% FDR in both larvae and adults, the differences between SEL and CTL across all genes in those GO terms are positively correlated between larvae and adults (amino acid metabolism: r = 0.40, P < 0.0001, N = 94; purine compound synthesis: r = 0.47, P < 0.0001, N = 74). In contrast, no such correlation is found for the top GO term for adults "chitin-based cuticle synthesis" (r = 0.05, P = 0.65, N = 107).

We had overlooked this in the first submitted version, and we are grateful to the reviewer for their comment that made us re-examine these results. We have now modified the corresponding paragraph of the Results section (l. 252-261) to include the above observations.

2. In the methods section, the dataset for larvae gene expression used to argue the point above (Erkosar et al. 2017) is said to have been collected for larvae which "were raised on the poor diet". This means that the differential gene expression in that dataset corresponds to the plastic response of larvae to poor diet (i.e., selected larvae on poor diet vs control larvae on a poor diet) and not to the selected response (i.e., selected larvae on poor diet vs control larvae on a standard diet). If this is true, then the correlation estimated in the paper corresponds to plastic response in larvae vs selected response in adults, which will further prevent us from concluding that the expression profile of selected larvae and adults are coupled. The authors should clarify the dataset that was used for this comparison, and how this can change the interpretation of the results.

We think there has been some misunderstanding here as to what we mean by the evolutionary response and the plastic response. An evolutionary change in a phenotype is, by definition, mediated by changes in the genetic composition of the population, so it must be assessed in the same environment (a "common garden"). Thus, "Selected larvae on poor diet vs Control larvae on a Poor diet" measures an evolved, genetically-based divergence, not a plastic response. Similarly, "Selected adults raised on poor larval diet vs Control larvae raised on poor larval diet" measures the phenotypic divergence caused by these same genetic changes in adult phenotypes.

The contrast "Selected on poor diet vs Control on standard diet" proposed by the reviewer compares different genotypes in different environments; i.e., it combines effects due to evolution and due to plasticity. Thus, even though these are actually the phenotypes exposed to selection, we do not use them for comparisons e.g. across the life stages (furthermore, we do not have larval gene expression data on standard diet).

We now explain more explicitly at the beginning for Results the meaning of phenotypic plasticity and evolutionary response (l. 149-154).

The second big (in my opinion) conclusion of the study is that plastic response to larvae malnutrition is independent of the selection response to the same stimuli. This sounds counterintuitive but is supported by two independent datasets, gene expression and metabolomics, which suggests it is a solid result. Despite such an interesting pattern, the authors do not really digest this result for the reader, nor contextualize it in any way, despite being other studies exploring the relationship between plastic and adaptive response (e.g. in fish Ghalambor et al. 2015, in flies Huang and Agrawal 2016 Plos Gen, in plants Josephs et al. 2021 Evol letters, etc). The data, nor the discussion of it, help connect the dots: there is a correlation between larval expression profile and adult selected profile, but no correlation with the intermediate step between these two which is the plastic response in the adult?

We have now included these references, as well as another one by Yampolsky et al. (2012). We note that three of those four studies found no or a somewhat negative correlation between the plastic and evolutionary responses; only Josephs et al. found a positive correlation. We do not think this result is counterintuitive, it just suggests that the "evolution follows plasticity" hypothesis proposed by some authors does not seem to be generally supported by this kind of data.

Also, we do not see in what sense the plastic response of the adult is an "intermediate step" between a genetically-based evolved change of larval expression patterns and a genetically-based evolved change in adult expression patterns. The most parsimonious explanation for this pattern is that genetic variants that were the raw material for the evolutionary divergence between Selected and Control populations tend to affect expression at both larval and adult stage, and in the same direction. We now elaborate on the background of this issue in the second and third paragraph of Introduction and further discuss the interpretation of our results in Discussion.

So, I am left wondering whether conceptually the authors are asking the wrong question here. Given that adults never experience any change in diet, they cannot possibly be addressing plasticity with the experimental design used here? Maybe actual plasticity can only be measured in larvae that experience both diets? Or maybe everything makes sense? In any case I think the authors should discuss better this result for it to be meaningful.

Phenotypic plasticity, according to a widely accepted definition, is the expression of different values of phenotypic traits by the same genotype(s) in response to different environmental conditions. This includes cases where the environmental factor of interest is acting at an earlier life stage than the stage when the phenotype is measured. Indeed, as we state in the Introduction, there is a lot of interest in the plastic responses of adult metabolism to juvenile or developmental nutrition, also in the context of human health. Thus, we argue that asking whether this plasticity of adult physiology in response to larval diet anticipates the evolutionary change of the same adult phenotypes driven to the same larval diet is a legitimate and interesting question.

However, the reviewer is right that the fact that the evolutionary change has been driven by larval diet and we are focusing on adult metabolic phenotypes renders the interpretation of the relationship between the plastic and evolutionary change more difficult. This is in part because, as our results suggest, the evolutionary changes of the adult phenotype may be maladaptive. We now elaborate on this in the Discussion (l. 253-272)

Obviously, plasticity of the larvae in response to larval diet and its relationship with the evolutionary change in larval phenotype is an interesting question, which we have addressed in another paper (https://doi.org/10.1093/evlett/qrad018). We found no such relationship.

1. It seems like the 'coupling' of larvae and adult expression profiles is a surprising result. However, despite two lines here and there, this is not really contextualized in the text. For someone not working on insects, there is no way of weighing this result. The authors should improve the description of (a) what are the expectations regarding coupling/decoupling and what is previously known about this, (b) the biology of metamorphosis. For example, there are certain things that are "locked" during larval development like size (as accounted for during the ovarioles analysis) and therefore larva and adult phenotypes are coupled. What other things are "locked" during larval development and how can they explain the 'coupled' patterns of gene expression you see? E.g. are some of those pathways things that we should expect to be decoupled/coupled?

We have now elaborated on the rationale and interpretation of this cross-stage coupling in two places. First, we are bringing up the notion of metamorphosis "decoupling" larval and adult phenotypes in the 3rd paragraph of Introduction. Second, in the discussion we very briefly summarize what happens during the metamorphosis, and specifically describe the fate of larval fat body (cell dissociation, autophagy and apoptosis, with possibly some surviving cells contributing, together with progenitor cells, to adult fat body). While obviously the topic is highly interesting, we did not discuss further on what other organs may carry an "imprint" of larval conditions and how, as we felt this would go too much on a tangent. But a new analysis (Figure 5 —figure supplement 1), which indicates an absence of correlation of the plastic response of metabolome to diet between larvae and adults, implies that such imprint does not play a major role in determining adult metabolism.

2. The result section should be re-organized to bring together data that addresses the same question. For example, the first section 'gene expression patterns point to evolutionary changes in metabolism" uses half of the space describing the comparison selection regime vs plasticity and only at the end addresses the evolutionary changes in metabolism. And then, the second section does discuss the overlap of plasticity and evolution regime but half of this comparison had been described in the first section. This happens again in other result sections and makes the paper harder to read (e.g. pages 7 and 9).

We have restructured the Results section in that we put together the relationship between plasticity and evolution for both gene expression and metabolome together in one subsection (l. 274ff), followed by a subsection that describes the correlation in both gene expression and metabolome between larvae and adult (l. 327ff).

3. Page 4. You say the first PCs don't separate the groups and the PCA is not shown. Could you please add this as a SI figure, and state what is driving those first 3PCs before the signal you are interested in pops up in PC4-5.

We now included a figure supplement with plots of the first six PCs (Figure 1 supplement 1). As far as we can say, the first three PCs seem driven by idiosyncrasies of individual populations and samples. Part of this is measurement variance – including e.g. effects of vial (including that microbiome shared by flies) or drift in the dissection procedure (while samples were dissected in a haphazard order, for practical reasons flies from a given sample were dissected together). However, a part is likely to reflect genetic differentiation among replicate populations, which is expected after 250 generations of independent evolution under rather small populations sizes. This is supported by the fact that the PC scores for each of the first 6 PCs of flies raised on standard and poor food are positively correlated across the 12 populations (r = 0.24 to 0.87). As we cannot really say more about this, we simply added the statement "the first three PC axes appear driven by idiosyncratic variation among replicate populations and individual samples."

4. Page 6, last line. It cites Figure 2 but that PCA is not shown in that figure or anywhere else.

This figure was mis-referenced in the previous version (apologies); now it is figure 2A.

This figure will be interesting to put next to the PCA for expression (point 4 above) given that the expression signal is weaker than the metabolomics in separating treatments. Also, you should discuss why this is so in the discussion.

As for the signal of differentiation between treatments (i.e., evolutionary regime and larval diet) being weaker in gene expression than in metabolome, if we could demonstrate this confidently, this would be an interesting observation worth discussing. However, we do not think the PCA provides sufficient support for this kind of statement. While the PCA represents the data in a standardized way, the underlying data are still of a very different nature and thus differentially subject to random biological variation or measurement error. Furthermore, for RNAseq we only had one sample per population and diet whereas for metabolome we had two (not counting the starved fly data). As indicated in the Methods and the figure legend, the duplicated metabolome samples were averaged for the PCA on metabolome. For these reasons, we think any interpretation of the comparison of PCA results for gene expression versus metabolome would be shaky.

5. First line of page 5. I think it is standard vs poor diet instead of selected and control?

The comparison we are making here is indeed between standard and poor diet, but we wanted to state that the top GO term is the same as for Selected vs. Control. We modified the sentence to say (l.276-8):

"As was the case for Selected versus Control populations, the top GO term enriched for genes differentially expressed in flies raised on the poor versus standard larval diet treatment (corresponding to phenotypic plasticity) was 'chitin-based cuticle development' (Figure 1D; Supplementary file 2 table B)."

6. Parts of the methods are described in the Results sections, which is OK, but then, that information is omitted from the methods. Please add all relevant info to the methods section.

We now added several bits of information to the Methods. However, generally we aimed to minimize redundancies between Results and Methods, with the idea that the design and the essence of our methods should be understood from Results and only readers interested in technical details would refer to Methods.

7. Just a suggestion of course. I think the title does not do justice to the article. It is already known that juvenile malnutrition impacts adult phenotypes. But you went well beyond that! What about something like "evolution under juvenile malnutrition show that larval and adult stages are not decoupled in D. melanogaster resulting in low-fitness adults".

We had a long back-and-forth about this suggestion among the authors. The title proposed by the reviewer exceeded the 120 characters length limit and we did not find a satisfactory way of expressing the same content in 120 characters. Furthermore, we thought this might too strongly focus the reader on a single aspect of our results. Thus, we finally decided for a version resembling the original one, but with an added emphasis on cost to adult fitness.

In general, the paper will benefit a lot from better visualization of the results. There is too much text that could go into plots:

8. It will help the reader a lot to have a diagram of your experimental setup. It's hard to imagine plastic response vs selection because the selection regime also experiences plastic response every generation when adults are placed in standard food. You should make a drawing of the life stages of Drosophila and to which diet each one is exposed, and highlight the differences between selection regimes (230 generations, etc) and the plasticity experiment.

We have now added such a scheme (Figure 1A).

9. The overlap between selection and plasticity for gene expression and metabolomics could be made into a plot that shows the number of genes that overlap and the direction of change. Something like x-axis is selection lfc and y-axis is plasticity lfc.

We have now added such plots for both gene expression and metabolome (Figure 4 A,B).

10. Figure 1 could have more information given the first two sections of the results. Panel A is OK, Panel B, instead of showing up and down-regulated for selection regime could instead compare GO terms between selection regime and plasticity – again to highlight that nothing overlaps except for cuticle GO. It will be useful to see the p-values of the enrichment.

We could not think about a useful graphic way of directly comparing the GO terms between two factors, especially that the overlap consists of just one GO term. One could in principle think of some kind of scatterplot where each point would represent a GO term, with the P-value or the proportion of DEG for plasticity on one axis and evolution on the other axis. But most of the points in our case would be on the axes, and trying to place the long names of GO terms would render the plot unreadable. A Venn diagram is also rather impractical.

Furthermore, as we explain above, we came to realize that the overlap or otherwise of enriched GO terms is of limited usefulness as a tool to assess the similarity of response. Thus, we downplayed the lack of overlap between GO terms, focusing instead on examining the correlations at the level of individual genes.

Still, we did add a plot summarizing the top enriched GO terms (i.e. those with P < 0.01) for the plastic response, as these are or interest in their own right (Figure 1D). We retained the form of the plot, i.e., a horizontal bar plot indicating the number of significantly up- and downregulated genes in each GO category, as this reveals some (mildly) interesting patterns, and is slightly more informative than providing just the number of DEG or a plain list of GO terms.

The plot mentioned in 8 could go as a panel here.

Done, as mentioned above.

11. Page 6 first section. This could also be made into a plot comparing lfc in larva and adults. And also add the comparison of the GO terms because here there are interesting differences.

We now added such a plots for logFC for both gene expression and metabolome (Figure 5A,B), they nicely illustrate the positive correlation.

As for the GO terms, as explained above, we are not convinced that a comparison of significant GO terms is particularly useful in this context. Notably, even though amino acid and purine metabolism GOs do not come up as enriched in the larvae, at the level of expression of genes in these categories there is still a clear positive correlation (l. 356-8). Furthermore, a GO-term analysis of the larval expression data has already been published in Erkosar et al. 2017, albeit using a previous version of mapping and with P-value adjustment which seems overly conservative by current understanding. Thus, rather than expand on the GO term analysis in the context of larva-adult similarity, we decided to play it down. We think that the new Figure 5 conveys the message in a sufficiently convincing way.

12. Page 7 and 9, The results could be shown in plots of PCA, the comparison of lfc between selection and plasticity, and GO terms.

It is not clear how this comments add to the reviewer's comments 3,9 and 10 above. As we already explained above, we now have the PCA plots and the logFC for plasticity and evolution for both gene expression and metabolome, and a comparison of GO terms for gene expression.

For metabolome, a somewhat analogous approach is pathway enrichment analysis. As we reported already in the previous version, we attempted to use a join pathway analysis to look for a joint signal of enrichment in metabolite and gene expression data, with rather disappointing or even misleading results (the supposed second top enriched pathway, "valine, leucine and isoleucine biosynthesis" does not exist in animals). We get similar output if we only use metabolite data. A pathway enrichment-based approach seems more suited to untargeted metabolomic data with a large number of features, of which a rather small fraction is differentially abundant. We have 113 metabolites from predefined groups and half of them (57) are differentially abundant. So, even if a particular pathway is not "enriched", it may show substantial changes. For these reasons, we did not elaborate the pathway analyses and did not include their results in a graphical form, just as a supplementary table (Supplementary file 4).

Reviewer #2 (Recommendations for the authors):

Nutritional stress early in life is pervasive and has a wide range of negative effects on adult physiology. Whether the genes underlying the physiological response that allows organisms to overcome stress early in life also enable adult success remains poorly understood. The evolutionary response may favor plastic strategies, where flexibility in response to different stress increases under early-life stressors. Alternatively, genetic variants that enable larval survival may have pleiotropic effects that negatively impact adult fitness. Untangling whether these effects are independent or interactive requires clever experimental design in a tractable system.

Here, the authors use experimental evolution in Drosophila melanogaster to understand evolution to poor nutrition early in life. For >200 generations, larvae were reared in a nutrient-poor diet (SEL diet, 1/4 the concentration of the standard diet). Then, after eclosion, flies are reared under a standard diet. The control (CTL) population was reared on the standard diet for the entire lifespan. For the results presented here, Erkosar et al. performed a reciprocal transplant experiment, where both the control (CTL) and selected (SEL) flies were reared in both CTL and SEL diets. This enabled comparisons between the effects of the evolutionary regime (CTL and SEL flies) and plasticity by varying the within-generation dietary effects (CTL and SEL diet). Combining RNAseq, metabolomics, and genomics with experimental evolution provided a comprehensive measure of how the evolutionary response to dietary stress shaped fly physiology.

A key strength of the study is the experimental design and comprehensive phenotyping. By rearing SEL and CTL lines in both SEL and CTL dietary regimes, the effects of evolution and plasticity can be better understood. In general, the RNAseq and metabolome results suggest that the evolutionary regime and dietary regime act additively in shaping adult physiology, but increased plasticity did not evolve (as no significant interactions were observed). One concern is that gene expression analyses was limited to the adult fat body, where the top hits were primarily associated with cuticular development (SEL flies are smaller). Additional tissues or at different time points in development might provide clearer insights into how evolution under dietary stress shapes gene expression and metabolism. Metabolomics also highlighted the role of purine metabolism. Purine metabolism was upregulated in SEL flies, which were also associated with SNPs that differentiate the genomic response between SEL and CTL lines.

A particularly interesting experiment was to starve the flies and profile the resulting changes to the metabolome. The SEL lines reared on the CTL diet still had a metabolome that more closely resembled the CTL fly after starvation, suggesting that the SEL flies exist in a permanently starved state, regardless of the current diet. The consequences of this permanently starved state also impact fecundity. SEL flies had lower fecundity than CTL flies on the SEL diet, counter to expectations that SEL lines would show greater fecundity in the SEL diet. SEL flies are thus paying a fitness cost for the traits that enabled survival in nutrient-poor diets. However, fecundity was only assayed in the SEL diet, and so it is unknown whether the reduction in fecundity for SEL flies is in response to selection or the diet.

Overall, the authors propose that the results support a scenario where the cross-life stage, pleiotropy drove adaptation to the dietary stress. In other words, SEL flies pay a fitness cost late in life for the strategies used to survive the nutritional stress early in life, with signatures of selection at the genomic, transcriptomic, and metabolomic levels. Particularly for organisms like Drosophila that inhabit ephemeral and fluctuating environments, plasticity is thought to evolve to decouple the physiological needs that differ between adults and juveniles. However, here, through experimental evolution and 'omics rich data, Eroksar et al. provide new insights into the role of pleiotropy in shaping adaptation to early life stress.

I really appreciated the multifaceted approach to understanding adaptation to nutritional stress early in life. The experiments are necessarily complex, and the manuscript is very rich in data-but, I had a difficult time throughout keeping track of how the different experiments linked to different results/conclusions/implications, etc.

One issue is that plasticity wasn't well defined to begin in the introduction. The authors begin at line 46 with the statement "In contrast to such phenotypically plastic within-generation responses", but is not clear how the first paragraph showed phenotypic plasticity.

We have now modified the beginning of the 2nd paragraph (l. 52ff) to read:

" These physiological responses to developmental nutritional conditions are a form of phenotypic plasticity; i.e., a change of phenotype induced by differences in the environment, with no change in genome sequence (Scheiner 1993; Bateson, et al. 2004). We know much less on whether and how adult physiology and metabolism evolve genetically over generations in response to natural selection… "

We hope this clarifies the distinction between phenotypic plasticity and evolutionary change. What we state in the first sentence is a standard definition of phenotypic plasticity and the second reference cited there (doi:10.1038/nature02725) specifically discusses metabolic changes in adults induced by developmental nutrition as phenotypic plasticity.

The second paragraph suggests that one can examine plasticity by placing larvae in a different environment than adults.

No, this is not what we mean. In general, to study plasticity, one exposes individuals of the same strain or population to different environmental conditions and records the resulting differences in the phenotype. Here, we specifically talk about plasticity of adult phenotype in response to juvenile/developmental nutritional environment. Thus, to study it we need to let individuals develop under different juvenile nutrition treatments and study how this affect adult phenotypes; because we want to be able to attribute the effects specifically to juvenile conditions, the adults must experience the same environment. Of course, one could also study plasticity of juvenile phenotypes in response to juvenile nutrition, or of adult phenotypes in response to adult diets, but these are not a subject of this paper.

The second paragraph has been completely rewritten, so hopefully this should be clear.

However, this is also the selection regime, where only larvae are reared on nutrient-poor diets and adults are reared on the same standard diet.

Yes. The difference between phenotypic plasticity and evolutionary change in response to selection is not in the environmental factor (or experimental conditions) driving it, but in the underlying mechanism. We hope the beginning of the 2nd paragraph makes it clear now. We further explain how we operationally quantify plasticity in our study in the 1st paragraph of the Results.

Plasticity can be assessed in different ways (e.g., change in phenotypic variance, intragenotypic differences, reaction norms, etc.), and so a clearer definition would help ground the complexity of the following experiments.

We are looking at a change of trait means expressed in different environments, as we now state in the first paragraph of Results.

Further, does the starvation experiment count as plasticity in this framework? Clarification would help the reader understand and appreciate the complex experiments that are necessary to address this question.

Absolutely. We now added this sentence to the paragraph where we introduce the starvation treatment (l. 383): " The effect of the starvation treatment on the metabolic phenotype is also a phenotypically plastic response – to a different form of nutritional stress and one applied to adults rather than larvae. "

While the description of the five questions (lines 78-103) helps provide some context in how the reviewers are thinking about plasticity, it was disconnected with the subject material in the introduction and introduced further complications like pleiotropy. It might help by having a simpler statement in the introduction describing how plasticity can be assessed using the selected and control populations.

We have thoroughly rewritten the Introduction, with the new 2nd and 3rd paragraph explicitly introducing the general question of the plasticity-evolution relationship and the evolutionary independence or otherwise of larval and adult phenotypes. These provide the background for the specific questions formulated in the second half of the Introduction.

Another general comment is the error bar in Figure 3B, Figure 6B, etc. It is a little unusual to see just one error bar per panel in the figures. I also see that the error bar is the standard error of the mean-but mean of the main effect (evolutionary regime)?

The points in figure 3B (and the now removed figure 6) are marginal (= least square) means of metabolite abundances from the linear mixed model fitted to metabolite abundance data. The statistical model estimates the overall error variance and uses it as the basis for estimating the standard errors of each marginal mean. If the design is balanced (as is the case for these data), the standard error is the same for all means. We could put these SE bars on each point, but that would render the figure harder to read as there would multiple overlaps, possibly requiring adding some jitter to the points – and this would not add any information because the bars would be identical. We now added this text to Methods to explain this:

" To illustrate the effects of the experimental factors (in particular, the interaction between the effects of evolutionary regime and starvation; Figure 3B), for some metabolites we plotted the estimated marginal means from the GMM. Because the design was balanced and the error variance was estimated from the model, all marginal means had the same standard error; to reduce the clutter in the plots we plotted this common standard error as a single error bar rather than adding such bars to each symbol representing the mean."

I see the placement is zero centered for the log fold change, which is OK, but what is it for Figure 6B? An additional explanation would be appreciated.

Figure 6 has now been removed following a comment by reviewer 3. However, generally, metabolome results are typically zero-centered and scaled for the analysis; this is in part because the abundance estimates from the LC-MS approach we used cannot be meaningfully compared between metabolites. In contrast, the RNAseq data are normalized in a way that preserves the relative differences in the numbers of reads mapping to different genes, and thus allow one to distinguish genes with high and low expression.

Many of the experiments need additional rationale to describe them. While I got this to some extent in the introduction lines 78-103, the rationale and implications are underdeveloped for several of the key results.

In the experiment associated with Figure 1, why are fat-body/carcasses used? It is described that the fat body is the key metabolic organ, but why wouldn't the whole body be used? For example, given the fecundity results presented later, how does the fat body transcriptome inform these results as opposed to what ovary or whole-body results would show? What is lost by focusing just on this one tissue group? If cuticular development and maturation genes are still most important, wouldn't that still show whole body RNAseq? Was the choice to enable linking the RNAseq with the metabolomics that follows? The rationale should be better clarified.

We aimed to focus on changes in the regulation of metabolism during the adult stage. Most of that metabolism takes place in the fat body, and this is in particular where dietary nutrients are converted into egg proteins and lipids, and where metabolic reserves of glycogen and fat are laid down or mobilized. While it is the ovaries that pack the vitellogenin and lipids into eggs, those key materials are imported there from the fat body. For these reasons we focused on the fat body for both gene expression and metabolome. By analogy, studies of the effect of nutrition on rodent metabolism often focus on liver (e.g. Agnoux et al. 2014, 2018; Safi-Stibler et al. 2020).

We now expanded the justification for focusing on the fat body to read:

" We focused on female abdominal fat body, the key metabolic organ combining the functions of mammalian liver and adipose tissue. It is in the fat body where metabolic reserves of glycogen and triglycerides are stored and mobilized, and where dietary nutrients are converted into the proteins and lipids subsequently transported to the ovaries for egg production (Li, et al. 2019). " (l. 156-160)

I'm also not sure I understood also why only the evolutionary regime differential expression results were shown, but not the larval diet results (summarized lines 162-188). I found the lack of parallelism an interesting result from the RNAseq experiment and would find them useful visualized (not as a table) at least in the supplement.

A similar point was made by Reviewer 1. We now added a plot illustrating GO-terms significant for the effect of larval diet (i.e., the plastic response) (Figure 1D), as well as a scatterplot of log-FC due to regime and larval diet (Figure 4A).

The comparison with Erkosar et al. 2017 was not well justified. In the methods, lines 614-15 it is mentioned that the larval gene expression is derived from an experiment that manipulated the microbiome, with and without microbes. The evolutionary regime was used as the main effect for comparison, but this isn't totally straightforward-why not just subset the 2017 data for only +microbe? why include the germfree at all here?

For the larval RNAseq in Erkosar et al. (2017), larvae derived from embryos that were first made germ-free by bleaching, and then mono-inoculated with 300 µl of suspension of a single strain of liquid culture-grown Acetobacter at OD=1 (microbe+) or treated with PBS (microbe-). This microbe+ treatment corresponds to inoculation with approximately 10^8 CFU per larval bottle. By contrast, in the present study we inoculated the embryos without bleaching and using 300 µl of suspension of feces at OD=0.5, which only contains about 10^3 CFU. Thus, in addition to a few other differences, the inoculation treatment used in the present study involved a much smaller microbial dose than the microbe+ treatment of Erkosar et al. (2017).

Furthermore, there is a trade-off in limiting the larval analysis to the microbe+ treatment because it cuts the number of samples by half, reducing statistical power. As a consequence, while the correlation across all genes between larvae and adults is the same whether we take all or only microbe+ data from larvae (r = 0.3522 vs 0.3518), the number genes that pass the FDR threshold is smaller; only 56 genes pass 10% FDR both for "+microbe" larvae and for adults. Still, 53 out of the 56 genes have the same sign of the effect (compared to 78 out of 84 genes when all larval samples are considered). Thus, the general result is robust to using only "microbe+" samples from Erkosar at al.

2017, but the signal is stronger when both "microbe+" and "microbe-" samples are used.

For those reasons we prefer to retain examining the similarities between the effects of regime on larval and adult expression by comparing main effects from the two experiments.

We added the following explanations to the Methods:

" All larvae were raised on the poor diet in that experiment, in either germ-free state or colonized with a single microbiota strain at the high concentration of about 10⁸ CFU (Erkosar, et al. 2017). Thus, the bacterial inoculation used for adult RNAseq (feces suspension containing about 10³ CFU) was intermediate between the two larval treatments. […] We used the main effect of evolutionary regime from this study for the comparison with the adult results; using only data from the microbiotacolonized treatment led to qualitatively similar conclusions."

Also, Erkosar 2017 data is from generations 177-200, which isn't stated here.

We added this information in the methods (it was around generation 190).

It's very surprising that the same genes are differentially expressed in mostly the same way between larvae and adult carcasses given all these differences in experimental design. The differences between these datasets should be explicit for the reader. However, given the complications listed above, I feel like the comparison between the carcass data and the Erkosar data does not add much to the overall message (and it's not visualized).

We would argue that the fact that a pattern is surprising also makes it interesting. The recommendation to remove this comparison goes in the opposite direction to what Reviewer 1 suggested; Reviewer 1 thought that the correlation between the effects of evolution on larval and adult gene expression was a one of the key results of our study, and recommend that we expand on it. This we did as summarized above, including adding an analogous correlation for the metabolome. As to the differences between the ways in which the larval and adult data were obtained (and the time elapsed between the two RNAseq studies): these potentially confounding factors would be expected to weaken the correlation. Thus, the fact that we see this positive correlation despite all the methodological differences implies that the true similarity of larval and adult changes is even greater. We make this point briefly in the Discussion (l. 476-480).

I liked the starvation experiment. Like before, it needs to be clear whether plasticity here refers to the std-poor diet and/or fed-starved comparison. It can be both but needed more information here. I think the authors consider the results to show more about the evolutionary regime (lines 373-375), but why isn't this also an aspect of plastic responses to nutritional stress? The interactions between fed-starved and the connection with differential allocation are very interesting results. While most of the discussion of this result is focused on the SEL resembling a starved metabolome, the change for SEL between fed and starved seems less than for the CTL. The CTL-poor seems to change most dramatically between fed and starved, which makes sense given the previous results of this selection experiment. Does this mean that SEL can buffer the nutritional stress better? I'm curious how buffering of this starvation stressor, despite previous evidence that SEL are less starvation resistant (line 331), informs our understanding of the adaptation to nutritionally poor larval diets.

Indeed, the response of the of the metabolome of Selected flies to starvation in the PCA is quantitatively smaller than that of Controls. However, to say that the metabolic state of Selected is more robust to starvation than that of Control, the starved Selected would have to look less like starved flies than starved Controls. This is not the case; rather, Selected and Controls converge to similar state upon starvation; it is in the fed state where they are different. We elaborated on this interpretation as follows (l. 427ff):

" Differential response of the Selected and Control populations to starvation is supported by regime × starvation interaction on PC1 scores (F_1,10 = 10.7, P = 0.0084) – the metabolic state changed less upon starvation than that of Controls. This does not imply that the metabolome of Selected flies is more robust to starvation. Rather, fed flies from the Selected populations were situated closer to starved flies along the starvation-loaded PC1 axis than fed Control flies (Figure 4A; F_1,17.8 = 14.0, P = 0.0015, GMM on PC1 scores); in the starved condition they converged to a similar state (F_1,17.8 = 0.0, P = 0.99; see Supplementary file 5 for detailed statistics)."

Great to see some overlap between genomic, transcriptomic, and metabolomics in purine synthesis.

Thank you, this is why we think that figure 3 is useful in putting these overlapping results on a common map.

Please make clear that the Kawecki 2021 genomic data is from ~150 generations.

Done (l. 255).

For the adult costs of larval adaptation, it is interesting that CTL laid more eggs than SEL populations when exposed to a poor larval diet. Would you see the same result if fecundity was assayed for both on the standard diet? i.e., does the evolutionary regime just lead to generally fewer eggs produced by SEL populations?

We have already published that, after 64 generations of experimental evolution, the

Selected populations had lower fecundity than Controls when both raised on standard larval diet (Kolss et al. 2009). This, however, can be interpreted as a consequence of the Selected specializing on the poor larval diet at the expense of performance when raised on standard larval diet. We have seen other signs of this, such as somewhat reduced larval viability and substantially smaller adult size of Selected compared to Controls when raised on the standard diet. This kind of trade-off, whereby populations evolving under new conditions lose somewhat in terms of performance in the ancestral conditions is rather expected. So, what we find here indeed suggests that Selected have a generally lower fecundity, suggestive of an larval-adult trade-off (rather than poor larval diet – standard larval diet trade-off). We added a shorter version of this point to the Discussion (l. 496-498):

"We have previously reported a lower fecundity of Selected females compared to Controls when both were raised on standard diet (Kolss, et al. 2009); we interpreted that finding as a manifestation of reduced performance of Selected populations when raised on the standard larval diet."

Does it matter for adaptation to nutritionally poor diets if SEL populations are just less fecund? I think the rationale needs some additional justification here.

Absolutely; eggs to breed the next generation are sampled haphazardly from those laid by the females in a setup identical to that used to quantify fecundity. Thus, a female that lays more eggs will contribute more offspring to the next generation. We now state this explicitly (l. 500-504):

" Other things being equal, a female's contribution to the next generation under the experimental evolution regime is proportional to the number of eggs she lays within a time window similar to that used to quantify fecundity. Thus, the fecundity reduction we found in Selected adults represents a significant fitness trade-off of the improved ability of the Selected larvae to survive and develop under extreme nutrient shortage."

Additionally, in Figure 5, can these be visualized in a way that shows the true range of observed values (e.g., boxplots)?

This is now figure 7. We were not comfortable with using a boxplot because it would conflate two levels of replication (replicate populations and several replicates per population). Furthermore, the nature of the points differs among the three plots: for ovariole each data point is one female, for eggs number if it is the number of eggs laid by a group of females divided by their number, and for egg-to-ovariole ratio it is the average number of eggs divided by the average number of ovarioles for each population. So only the ovariole data represent values for individual females. Therefore, we retained the bar graph but, to illustrate the variation among populations, we added points that represent the means of each population.

I think the discussion provides important context, and the authors highlight one potential explanation for the results is pleiotropy across life stages. This is an interesting perspective given that Drosophila are holometabolous insects, and as the authors say, this remains surprisingly understudied. It's not clear whether this cross-life-stage pleiotropy is amplified because of holometabolous insects or also generalizable to other organisms as well.

While this is not clear empirically, an argument has been made that a key advantage of insect complete metamorphosis is that it eliminates such pleiotropy. Thus, in organisms without metamorphosis such pleiotropy would be more likely. To make this point, we now added the following text to the Discussion (l. 582ff):

" A major advantage of this complex development is thought to be that it decouples larval and adult gene expression, promoting independent evolution of larval and adult phenotypes (Moran 1994; Rolff, et al. 2019). Contrary to this notion, our results suggest that, at least on the scale of hundreds of generations, evolutionary changes in physiology driven by selection acting on larvae may have maladaptive pleiotropic effects on adult physiology. If this is the case in a holometabolous insect, such evolutionary non-independence of juvenile and adult physiology would likely be more pronounced in species with more developmental continuity between juvenile and adult tissues and organs. "

Reviewer #3 (Recommendations for the authors):

Erkosar et al. use gene expression and metabolomic analyses to investigate the evolutionary change that accompanied over 230 generations of selection by nutrient deprivation from egg to pupation in Drosophila.

The evolved populations were six replicate populations, each selected for larval development and growth under poor nutrient conditions. Each population was kept at ~200 flies at each generation, which may have allowed significant drift when compared to larger populations in similar experimental evolutionary studies however, this regime has resulted in repeated evolution of a few interesting traits. These traits were revealed in earlier work and include faster developmental time, higher egg-to-adult viability, and smaller adult size. Additionally, selected lines showed higher sensitivity to starvation as adults and the allele frequency changes that accompanied selection intersect with those loci associated with selection for starvation resistance.

The picture painted from previous work is of genotypes arising from selection that are more likely to make pupae/adults; to make them faster and more numerously than the control population. The adults that they make however are not as healthy as the adults from the control population. This certainly suggests that adult traits suffer a constraint based on the evolution of juvenile condition. Is that constraint manifest in patterns of gene expression, or in patterns of variation in the metabolome? Hints at metabolic adaptation within these populations came from a prior publication that described a difference in nitrogen or carbon assimilation between these populations, which is at least intriguing.

The current work is important in that it sheds light on the kinds of endophenotypic changes that accompany these evolutionary changes. In addition to the condition-specific patterns of expression and metabolite abundance, the authors also test their ideas about how the transcriptome and metabolome of adults of both populations respond to diet, and if there is a resemblance to flies evolved on poor food and starved flies.

This paper presents several interesting results. One, the evolution that occurred in the larval transcriptome was largely similar to the differences between the populations as adults. This implies a constraint in adult phenotypes on juvenile phenotypes and vice versa that is reflected in the transcriptome.

Within the same experiment, the transcriptome and metabolome of adults from both populations appear to respond similarly to diet. That is, the adults of both populations, despite significant shifts in metabolome and transcriptome, show very similar responses to being placed on the poor diet, when compared to the normal diet. This is an unexpected result given the constraint that may explain the previous result.

That the within-generation response to diet appears independent of the evolutionary response to diet conditions is interesting. The authors could develop their discussion of these results further. I think it is worth considering that the effects of diet with a generation are not necessarily adaptive. An adult on a poor diet might have a metabolome that looks a certain way, reflecting in part the exhaustion of resources. The adult from a larva that has evolved on a poor diet however may more closely approximate an adapted metabolome, there is really no reason to think that these would be the same.

It is also interesting that the effect of diet on the adult metabolome was somewhat similar to the difference between the adult metabolome of the two populations, the evolutionary change. This is in contrast to the seemingly independent effect of diet on the transcriptome, and the transcripts that differ between the populations. I hope that this result is not a 'false positive' due to the small biological sample size of the transcriptome data (n=1) relative to the metabolome data (n=2), or a 'false negative' due to the relatively sparse metabolome coverage (~100 metabolites) that is typical of targeted analysis of fly metabolome compared to the more comprehensive transcriptome. This result is in line with other studies that indicate that endophenotype appears to converge as it goes from: gene > RNA > protein > metabolite, to phenotype. The relative anticipation of the evolutionary response seen in the metabolome differences by diet when compared to the same in a relationship in the transcriptome is neat.

The results described above manifest largely in PCA, in reduced dimensional space, which is appropriate and telling, but may hide variation that occurs within and between populations, and across diets. There is some agreement within the univariate analysis of metabolites and transcripts that are differentially abundant between populations and treatments, suggesting that the 'biological signal' seen in the PCA could be picked apart using the identities of the genes and metabolites. The authors indulge in this pursuit and while creative, take their interpretation too far. The authors make several clever interpretations of their data in order to shed light on potential mechanisms (pathway activity) that may explain their results. Without testing some of these ideas however this amounts to speculation.

We have reduced speculation about potential meaning of changes in metabolite abundance and gene expression at specific pathways, but we still think we are justified to identify pathways (in particular amino acid and purine metabolism) given multiple signals from RNAseq and metabolome at both larval and adult stage.

The third point of interest is the "Metabolic profile of Selected flies tends towards starved-like state", an argument that the metabolome of selected flies more closely resembles starved flies than does the metabolome of control flies. This is argument rests on MANOVA and some agreement in sign with metabolite levels upon starvation, and those affected by selection. This is an interesting result that could relate to the sensitization to adult starvation in the evolved population. At the univariate level however, the picture is quite complicated (Figure 3B). The interactions between diet, selection regime and the effect of starvation are complex. The authors make a valiant effort at interpretation, however, I don't think their conclusions are well supported from these data. I suggest approaching the univariate analysis more circumspectly.

The authors are over-interpreting the meaning of their data. This is my main objection to what is otherwise an interesting and meaningful study. I think that the authors should include their interpretation of the data, however, it should be presented as speculation, rather than as if the interpretation were itself an observation.

We modified the text to state clearly that not all metabolite groups show a similarity between the effects of evolution and starvation. We also modified the language to emphasize the speculative nature of these interpretations ("appear consistent with…", "we have not demonstrated a direct link…", "it is tempting to speculate…"). We think it should be clear to the reader that much of interpretations of our data are indeed speculative in the absence of independent experimental tests.

The joint analysis presented in 'Figure 4—figure supplement 1' is problematic for several reasons. I don't think it should be included as it does not add anything to the paper worth discussing.

We have now removed this analysis.

"[abstract line # 26]: "resulting in deficiency of electron transports and congestion in β-oxidation." Two objections, first, 'deficiency' is different from 'reduced', we have no idea if the reduced level of these RNAs is at all affecting the trait, these are associations and cause and effect is not known. Second, RNA is not protein, and even then, protein abundance is not activity, so either discuss these results in a more speculative light or make a more direct and independent test of electron transport activity/function. Further, congestion in β oxidation is certainly a possible explanation for the patterns of acetylcarnitines and other FA-derivatives that are more/less abundant, however, this too is a speculative explanation and ignores the equally valid evidence that this explanation is wrong.

We have now removed any speculation about the significance of differences in NAD, FAD and acylcarnitines.

Similarly, the section that discusses the amino acids [#492-505] describes some amino acids as 'deficient' in the selected flies. These amino acids are reduced in abundance compared to the control flies, but this is not the same as deficient. Deficiency implies that if there were more of these then the outcome would change. That result is unknown. The data are describing an association with abundance, concluding that this is a deficiency is misleading.

We meant to use "deficiency" in a relative sense, analogous to the way "underexpression" is used for genes showing a lower expression in the focal condition than in some other condition. We changed this to "underabundance" or "lower abundance" throughout the ms.

In line #385, "This analysis confirmed that the changes in amino acid and purine metabolism contributed to adaptation to poor larval diet." Confirming that something contributes, implies that you have gone beyond association and tested causality. These data do not demonstrate cause or effect but bolster a previously-identified association.

There are two sections of the results (lines 217-290, and 378-404) dedicated to the pathways that might be involved in the transcriptome and metabolome differences seen in these populations. I suggest paring this down, at least as a part of the Results section, as enough of it is speculation. Lines #378-404 are mostly speculative and so this argument should move to the discussion.

We took care to purge these parts of Results of any speculation as to potential causes or implications of the differences between Selected and Control populations in metabolite abundance. Still, while it could be argued that Figures 1 and 2 could stand on their own, we thought that it would be useful to point the reader to specific patterns in the data, and to briefly explain the role of those compounds in metabolism to less initiated readers. The former lines 378–404 (now l. 250-272) report the joint pathway analysis of Metaboanalyst, as well as describing an overlap between the data for gene expression, metabolome and genomic differentiation for the key pathways. It would be awkward to place it in Discussion. We also note that Reviewer 2 made a specific appreciative comment about this part ("Great to see some overlap between genomic, transcriptomic, and metabolomics in purine synthesis.").

Similarly, the discussion could be pared down given some of the contradictions that emerge when the data are discussed. For instance, if the paragraphs from lines #492-505 resolves with the axiom: "the differences in amino acid abundance between Selected and Control populations is consistent with differential use of amino acids in metabolism – their interconversion, catabolism and use in the synthesis of other metabolites.", then I'm not sure that it's a good use of space.

We are unsure which contradictions the reviewer alludes to. The statement quoted by the reviewer is not an "axiom"; it is one of three potential general reasons for differences in abundance of free amino acids. The other two are (i) a change in their supply from nutrition and (ii) a change in the rate of their use for protein synthesis. The point we are making in this paragraph is that the pattern of differences in amino acid abundance between Selected and Control flies is not consistent with the two latter explanations, leaving the first explanation as the more likely one. We rewrote this paragraph in an attempt to make this point clearer and also removing the somewhat tedious discussion of the abundance of specific amino acids in the diet relative to protein synthesis need (l. 517-524).

Figure 6 could be idealized (made with generic genes) or excluded. I say this because there are at least 10,000 genes from which to pull patterns like these from, so seeing these genes like this does not have meaning (FDR is high). The idea behind this figure can certainly be discussed as it is in the discussion, but this data set is underpowered for this kind of interference.

We removed this figure.

The authors use the term pleiotropic at several points. I think they are referring to something like antagonistic pleiotropy. I think a short section that introduces this term in this way, with reference to the theoretical or experimental work on the idea in relation to the evolution of aging perhaps would help the reader.

This is an excellent point and we now elaborate on the analogy to the early-life inertia theory of aging as well as to "intralocus sexual conflict" over gene expression at the end of Discussion (l. 590ff).

In the bigger picture, there are 'inward' and 'outward' implications for this work. The authors present support for their conclusion that negative trade-offs in adulthood are tied to the endophenotypic changes that occur in juveniles as a result of selection when only applied to juveniles. This support comes from the within-population similarity in adult transcriptome and metabolome to that of larvae. The 'inward' perspective, while tantalizing, is perhaps over-presented in the discussion of these results. Insight into the mechanisms (cellular pathways) by which these trade-offs happen can and should be included, however there are also wider implications of these results for insect evolution, the 'outward' perspective. These results suggest that the evolution of juvenile phenotypes is constrained by the outcomes they associate within the adult, at least on juvenile nutrition inputs/needs. The authors cite some theoretical papers on the topic, and I think the reader would be better served if the authors could reflect on the implications of the constraint they appear to have found. What does this imply for the phenotypic (and environmental) space that the juveniles are constrained to? We know that Drosophila are very choosy when it comes to oviposition, and here we see one of the reasons why that might be the case. Do the authors think that adult choice has evolved in part to deal with this constraint? It's easy to see the negative effects, the burden, of such a constraint. Evolution is powerful though, so why then is such constraint 'allowed'? Could this constraint actually reflect some kind of as-yet unconsidered benefit? For instance, does a constrained gene-trait map allow Drosophila to maintain a smaller, simpler, genome? I am not familiar with the literature on this and other topics that come to mind when the weight of these results are realized. Perhaps the authors could spend some of the discussion to share their interpretation?

We have restructured the discussion in part in response to these comments. We finish it with a section that speculates on potential implications of the constraint, as well as placing it in the broader context of other evidence of constrains on regulatory evolution, such as those thought to be implicated in aging and sexually antagonistic pleiotropy. We did not venture into discussing Drosophila oviposition behavior or genome size; we feel anything we could say about this would be far too speculative and tangential to the topic of the paper.

There are a few passages that I found hard to understand. When describing part of their hypotheses about the evolution of adult and juvenile phenotypes, they write [#82-85]: "Plastic responses that program metabolism adaptively for a poor nutritional environment in adulthood would be misdirected in the Selected populations because the adults were switched to standard diet; evolution should thus counteract them." Why would plastic responses in adults be affected by selection in juveniles? If they were affected, why would we call them plastic?

Keep in mind that this sentence is followed by [#85-87]: "Conversely, plastic responses that alleviate the consequences of having developed on a poor diet irrespective of adult diet should have become amplified as a result of evolution on the poor diet." It sounds like the authors are predicting that two counteracting effects have both happened and this does not make sense to me. I would like to understand the authors' point here and I would hope that it could be clarified in a revision.

This text has now been removed. We now introduce the subject of plasticity versus evolution in more general terms in of the second paragraph of the introduction.

There were several references to the figures that don't seem correct, and other parts of the figure referencing scheme that did not make sense. For example, the reference to 'Figure 2' on line #231, refers to a PC plot, and not the heatmap shown in Figure 2. Line #231 refers to Supplementary Figure file 4, and I'm not sure that's correct. Please make sure that the figure references point the reader to the correct figure and that the figures are referred to and arranged in numerical order.

We apologize for these mistakes in the previous version. We have now double-checked all references to figures and supplementary files.

The experimental design, the sampling scheme and batch design of the analyses were all good. These experiments made the most of the number of samples that were used. The statistical analysis as described is appropriate, however there is enough going on that making the code available would probably help raise the impact of this work.

We appreciate the reviewer's trust in our analyses. We understand the reviewer suggests to make the code available because it might be useful for other researchers. However, making our diverse scripts sufficiently user-friendly and annotated to be useful to others would require a very large amount of work. And even if we did this, the usefulness to other people would be limited. Nothing that we do in terms of statistical analysis is non-standard or novel. In part, we do basic statistical analyses such as PCA, Pearson's correlation, MANOVA or general mixed models, which are implemented in almost any statistical package. We used mostly SAS statistical system, which is now used by a relatively small number of researchers in biology, further reducing the general interest in the scripts. In the other part, we follow standard pipelines in published packages designed for special analyses (RNAseq read filtering and mapping, differential expression analysis, GO-term enrichment analysis). Thus, we are not convinced that including all our scripts would be useful to the readers or would increase the impact of the paper.

The authors are clearly aware of the multiple testing problem and I stand by my recommendation to remove figures that make use of data that do not survive FDR correction.

The philosophy behind the FDR correction used in the omics approaches is that the statistical test on each gene or metabolite tests a different biological hypothesis. We agree that this approach is essential when screening for candidate genes or metabolites, and we do not make any statement about any specific gene or metabolite being differentially expressed/abundant if it does not pass the FDR threshold. However, the number of genes passing an FDR threshold typically greatly underestimates the number of genes that are actually affected (i.e., there are more false negatives than false positives). This is also the case for our data, based on the estimated numbers of true nulls. Therefore, when examining broad patterns of expression changes, such as we do when comparing adult and larval expression or metabolite abundance changes, we went beyond the rather few genes/metabolites that passed the FDR threshold at both stages, and also looked at other genes with nominal P < 0.05, or even at all genes. In this approach, each gene / metabolite is treated as a point estimate and we are not making conclusions about any specific gene / metabolite but about the overall pattern.

A point I will need clarification on occurs in lines [#601-604]: 'Therefore, we also estimated the number of genes that differed in expression due to each factor in the analysis as the total number of genes minus the estimated number of "true nulls" (Storey and Tibshirani 2003) (https://bioconductor.org/packages/topGO/), using the weight method and Fisher's exact test.' What does this mean? By 'factor in the analysis', I assume you mean regime and diet. Why, and how, would you use topGO to evaluate the number of genes affected by regime or diet?

It seems somehow a piece of text got translocated from the end of the preceding paragraph on GO-term enrichment analysis; we apologize for this and thank you for spotting it. We corrected this so the offending sentence (now l. 689-691) now ends after the citation as intended:

https://doi.org/10.7554/eLife.92465.sa2

Article and author information

Author details

Berra Erkosar

Department of Ecology and Evolution, University of Lausanne, Lausanne, Switzerland

Present address
Foundation for Innovative New Diagnostics, Geneva, Switzerland

Contribution
Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing – original draft, Writing – review and editing

Contributed equally with
Cindy Dupuis

Competing interests
No competing interests declared

"This ORCID iD identifies the author of this article:" 0000-0003-1152-6772
Cindy Dupuis

Department of Ecology and Evolution, University of Lausanne, Lausanne, Switzerland

Contribution
Investigation, Writing – review and editing

Contributed equally with
Berra Erkosar

Competing interests
No competing interests declared
Fanny Cavigliasso

Department of Ecology and Evolution, University of Lausanne, Lausanne, Switzerland

Contribution
Formal analysis, Investigation, Writing – review and editing

Competing interests
No competing interests declared

"This ORCID iD identifies the author of this article:" 0000-0002-7764-4934
Loriane Savary

Department of Ecology and Evolution, University of Lausanne, Lausanne, Switzerland

Contribution
Investigation

Competing interests
No competing interests declared
Laurent Kremmer

Department of Ecology and Evolution, University of Lausanne, Lausanne, Switzerland

Present address
Université Côte d’Azur – CNRS – INRAE, Institut Sophia Agrobiotech, Sophia Antipolis, France

Contribution
Investigation, Methodology

Competing interests
No competing interests declared
Hector Gallart-Ayala

Metabolomics Unit, Faculty of Biology and Medicine, University of Lausanne, Lausanne, Switzerland

Contribution
Conceptualization, Formal analysis, Investigation, Methodology, Writing – review and editing

Competing interests
No competing interests declared

"This ORCID iD identifies the author of this article:" 0000-0003-2333-0646
Julijana Ivanisevic

Metabolomics Unit, Faculty of Biology and Medicine, University of Lausanne, Lausanne, Switzerland

Contribution
Conceptualization, Methodology, Writing – review and editing

Competing interests
No competing interests declared

"This ORCID iD identifies the author of this article:" 0000-0001-8267-2705
Tadeusz J Kawecki

Department of Ecology and Evolution, University of Lausanne, Lausanne, Switzerland

Contribution
Conceptualization, Formal analysis, Funding acquisition, Visualization, Writing – original draft, Project administration, Writing – review and editing

For correspondence
tadeusz.kawecki@unil.ch

Competing interests
No competing interests declared

"This ORCID iD identifies the author of this article:" 0000-0002-9244-1991

Funding

Swiss National Science Foundation (31003A_162732)

Tadeusz J Kawecki

Swiss National Science Foundation (310030_184791)

Tadeusz J Kawecki

Research funds of the University of Lausanne

Tadeusz J Kawecki

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank D Promislow for ideas, T Teav at Metabolomics Platform at UNIL for his contribution to sample preparation and data processing, and three reviewers for their constructive comments on a previous version of the manuscript.

Senior Editor

Detlef Weigel, Max Planck Institute for Biology Tübingen, Germany

Reviewing Editor

Luisa Pallares, Friedrich Miescher Laboratory, Germany

Reviewer

Benjamin R Harrison, University of Washington, United States

Version history

Preprint posted: January 12, 2022 (view preprint)
Received: September 3, 2023
Accepted: October 12, 2023
Accepted Manuscript published: October 17, 2023 (version 1)
Version of Record published: November 10, 2023 (version 2)

Copyright

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.