Genetic independence between traits separated by metamorphosis is widespread but varies with biological function

Abstract

Why is metamorphosis so pervasive? Does it facilitate the independent (micro)evolution of quantitative traits in distinct life stages, similarly to how it enables some limbs and organs to develop at specific life stages? We tested this hypothesis by measuring the expression of 6400 genes in 41 Drosophila melanogaster inbred lines at larval and adult stages. Only 30% of the genes showed significant genetic correlations between larval and adult expression. By contrast, 46% of the traits showed some level of genetic independence between stages. Gene ontology terms enrichment revealed that across stages correlated traits were often involved in proteins synthesis, insecticide resistance and innate immunity, while a vast number of genes expression traits associated with energy metabolism were independent between life stages. We compared our results to a similar case: genetic constraints between males and females in gonochoric species (i.e. sexual antagonism). We expected selection for the separation between males and females to be higher than between juvenile and adult functions, as gonochorism is a more common strategy in the animal kingdom than metamorphosis. Surprisingly, we found that inter-stage constraints were lower than inter-sexual genetic constraints. Overall, our results show that metamorphosis enables a large part of the transcriptome to evolve independently at different life stages.

1. Introduction

Despite strong developmental challenges due to complex life cycles, a majority of animals go through metamorphosis [1]. To explain the success of this strategy, metamorphosis must alleviate costs associated with simple life cycles. Antagonistic pleiotropy is created when beneficial and deleterious traits are genetically correlated; for example, when alleles conferring advantage in early life stages produce deleterious correlated effects later in life [2]. In organisms with simple life cycles, genetic correlations between traits expressed in juveniles and adults are expected to be strong and extensive [3], which can lead to maladaptive evolutionary responses that constrain populations from evolving an optimal phenotype [4]. To escape such potent evolutionary constraints, traits may be decoupled when expressed in alternative phenotypes throughout development. The adaptive decoupling hypothesis suggests that metamorphosis facilitates independent evolution of phenotypes expressed at different times of life. In simpler terms, complex life cycles could contribute to alleviation of antagonistic pleiotropy [5].

The morphologies of larvae and adults can be vividly different, showing that, at the macroevolutionary scale, metamorphosis enables the expression of contrasted qualitative characters in distinct life stages. Nonetheless, the comparison of larval and adult morphologies in multiple species reveals that constraints persist between life stages in frogs [6], fish [7] and salamanders [8]. By contrast with discrete traits, the influence of metamorphosis on the microevolution of quantitative traits remains poorly understood. Previous authors have investigated inter-stage genetic constraints in several traits that could be measured across life stages. Single traits, such as responses to cold temperature or antimicrobial peptide expression, appeared genetically independent between life stages of several insects [9–12]; however, the expression of another antimicrobial peptide in the same insect, virulence in a parasite and a locomotory trait in frogs were all genetically correlated [12–14]. Few studies measured inter-stage genetic constraints on more than one trait. Morphological traits of tadpoles and adult frogs were genetically correlated, although to a lesser degree than morphological traits expressed within life stages [15]. Aguirre et al. [16] detected remnant antagonistic pleiotropy when comparing viability traits at four life stages in an ascidian. Yet, these results appear to depend on the nature of the studied traits [17]. To understand the extent of genetic constraints between life stages, it is crucial to quantify genetic correlations beyond a small number of traits chosen because they are easy to measure rather than biologically relevant [17]. A systematic assessment of genetic correlations between life stages can reveal the pervasiveness of genetic constraints and their variability across the phenome and major biological functions.

Here, we used transcriptomics to assess the extent of genetic constraints maintained between larval and adult life stages—using thousands of gene expression quantitative traits measuring gene expression, covering all functions of the organism—in 41 inbred lines of Drosophila melanogaster from the Drosophila Genetic Reference Panel (DGRP) [18]. First, we calculated inter-stage genetic correlations across the transcriptome and tested whether those traits were correlated (correlation different from 0), as well as the direction of that correlation (i.e. positive or negative), or if they had some level of independence (correlation different from 1 and −1). Second, we investigated which biological functions were the most constrained or independent. Finally, we compared the amount of genetic constraint and independence between life stages to two previously published comparable biological cases: genetic constraints persisting between sexes and genetic constraints within sexes reared in environments varying in temperature.

2. Material and methods

(a) . DGRP lines and sample production

We used the DGRP [18], a set of inbred lines derived from an outbred population of Drosophila melanogaster. We chose 50 lines that showed no chromosomal inversions and scored across the range of starvation resistance measured in a previous study [18]. Our final analyses focussed on 41 DGRP lines after removing samples that could not be produced or failed to meet the quality control analyses criteria.

All lines were reared on Bloomington Drosophila Stock Center cornmeal food (https://bdsc.indiana.edu/information/recipes/bloomfood.html) at 25°C and 12 h light and dark cycles. Grand-parental generation consisted of two replicate vials per line of two virgin males and two virgin females placed together for 2 days and further tipped in in another vial to lay for another two days. Egg density was manually regulated to approximate 50 eggs per vial. In the following generation, two replicates of 12 virgin males and 12 virgin females per line, aged two to six days were placed together for 24 h to ensure mating occurred. Up to three males and three females (36 out of the 328 vials contained fewer adults as there were not enough flies available) of these 12 were subsequently placed together in a vial for 48 h to produce a total of eight replicates per line. Egg density was approximately kept to 50 per vial. We independently used the larvae and adults that emerged from those vials for RNA sampling.

Not all of lines were synchronous; in order to avoid creating among-line variance resulting from difference in development times, we did not collect all flies on the same day. During larva collections, once the first ‘wanderer’ (fully grown 3rd instar larva that stop eating and start searching for a pupation site) was observed in the vial, up to 10 larvae per vial were collected from the medium (i.e. not wanderers) and flash frozen with liquid nitrogen. When possible, a further 10 larvae were collected as a backup supply (and used in 12 of the 82 samples). The remainder of the flies were discarded (i.e. adults were never collected from a larva vial). Freezing occurred randomly during the day for the different samples, between 10.00 and 24.00. The collected larvae were not sexed and we expect our samples to be a random mix of males and females. During the collection of adult flies, we scanned vials every day. When the first adults were observed in a vial, we removed them and collected all emerged adults within the following 48 h. The collected adult flies were placed in fresh vials for 4 days to finish maturing and flash frozen in the same manner as were the larvae.

(b) . RNA extractions, RNA-Seq library production and sequencing

(i) . Preparation of RNA samples

The flies that were collected from two vials of each biological replicate were pooled before RNA extraction. As a result of our strict age conditions, we were unable to always gather 20 flies per sample, and the total number of flies per sample ranged from 2 to 31 flies, with a median number of 20 and 17 individuals in larval and adult samples respectively.

For each sample, we extracted RNA using phenol chloroform phase separation. We first added 500 µl of TRIzol (Thermo Fisher Scientific) to the frozen specimens and crushed them with a pestle. We then added 100 µl of chloroform to the crushed specimens, agitated the samples for 15 s and incubated them at room temperature for 3 min. The samples were centrifuged at 12 000 RCF for 15 min at 4°C. We transferred the aqueous phase into fresh tubes, added an additional 250 µl of isopropanol, and inverted the tube four times to ensure mixing. After allowing the tubes to incubate at room temperature for 10 min, the samples were centrifuged at 12 000 RCF for 10 min at 4°C. We removed the supernatant, washed the pellet with 500 µl of 75% ethanol, and again centrifuged the samples at 7500 RCF for 5 min at 4°C. A second wash was then carried out using 500 µl of 75% ethanol. After a final centrifugation at 7500 RCF for 5 min at 4°C, the supernatant was carefully removed. The pellet was left to dry at room temperature for 30 min. RNA was resuspended in 100 µl of nuclease-free water and then the samples were heated to 55°C in a drybath for 15 min. We verified RNA concentration and integrity on the Bioanalyzer (Agilent) with the RNA 6000 Pico Kit according the manufacturer instructions, as well as on the NanoDrop 8000 Spectrophotometer (Thermo Fisher Scientific). All RNA samples were stored at −80°C.

(ii) . Illumina library production

We prepared the cDNA libraries using the Illumina TruSeq Stranded mRNA Sample Preparation Kit, following the manufacturer's protocol and using a Biomek NX^P Automated Workstation (Biomek NXP Span-8 by Beckman). In brief, poly-A-containing mRNA molecules were purified from 1 µg total RNA using poly-T oligo-attached magnetic beads. The purified mRNA was fragmented by adding the fragmentation buffer and then heated to 94°C in a thermocycler for 8 min. We primed the cleaved RNA fragments with random hexamers and used reverse transcriptase to synthesize first-strand cDNA. Second-strand cDNA synthesis, end repair, A-tailing, adapter ligation, and enrich DNA fragments were carried out in accordance with the manufacturer supplied protocols. Each indexed cDNA library was verified and quantified using a dsDNA 915 Reagent Kit on a Fragment Analyzer (Agilent). We equally mixed 12 to 13 indexed cDNA libraries in each final library. The final libraries were quantified by real-time PCR using the KAPA Library Quantification Kit for Illumina Sequencing Platforms (Kapa Biosystems Ltd, SA), adjusted to 10 nM in water, and provided to the Montpellier Genomix platform (http://www.mgx.cnrs.fr/) for sequencing.

(iii) . Illumina library clustering and sequencing conditions

Each final mixed cDNA library was denatured with NaOH and diluted to a concentration of 18 pM using the standard library PhiX (1%) from Illumina. Flow cell was clustered using TruSeq SR Cluster Kit v3-cBot-HS, and then loaded into the Illumina HiSeq 2500 instrument following the manufacturer's instructions. The v3 sequencing chemistry (TruSeq SBS Kit v3-HS) was used, applying a 50 cycle, single-read, indexed protocol. Image analyses and basecalling were carried out using the Hiseq Control software (Illumina HCS) and Real-Time Analysis software (Illumina RTA).

(c) . Bioinformatics

The quality of the obtained sequences was checked using FastQC [19], with particular attention given to per-base-sequence quality. We used Trimmomatic [20] to: (i) remove the first 10 bases from the start of the read; (ii) trim slide windows, using the window size of 4 and an average base quality across each window of Q = 30; and (iii) to keep reads of a minimum of 35 bp at a minimum average quality of 30. To quantify gene expression, we aligned single-end RNAseq reads to the reference genome of D. melanogaster released in Aug.2014 (BDGP Release 6 + ISO 1 IT/dm6) (dm6) using STAR [21]. We applied all default parameters, except for a minimum intron size of 5 and max intron size of 200 000. We used dm6_FlyBase_Genes_assembled_chromosomes-r6.22.gtf for gene annotation. All bioinformatic analyses were done using the Galaxy portal (https://usegalaxy.org.au/).

(d) . Data analysis

We identified reads from 7414 genes that were expressed (more than 10 reads) in both samples of the 41 lines at both stages. To determine which of these 7414 expressed genes were associated with genetic variation—and therefore informative for our analysis of inter-stage covariances—we use linear mixed model analyses to partition the variance in each level of expression within each stage. Gene expression levels, expressed as count per million for each gene in each sample, were log₁₀ transformed. We standardized (mean = 0, s.d. = 1) each expression level within each stage and ran the following univariate mixed models within a restricted maximum-likelihood framework using the function mmer in the package Sommer [22] in R 4.0.4 [23]:

$$Y_{ij}=\mu +\mathrm{Wolbachia}+{\mathrm{Line}}_i+{\epsilon }_{ij}$$ 1

where $Y_{ij}$ is the standardized level of gene expression within a particular stage in line i for replicate extraction j, where $\mu$ is the average expression among lines. Wolbachia is a fixed effect of the presence/ absence of the bacterium Wolbachia in the line, line is a random factor representing the among-line (and thus genetic) variance, and the residual error, ε, is the variance among the two replicate extractions per line, per stage. Of the 7414 gene expression traits analysed, 6384 showed among-line variance at both stages, with statistical support for genetic variance found in both stages in 2422 traits, of which 1875 remained significant at the 5% false discovery rate threshold. Selecting genes according to a significance threshold (particularly in large datasets) is prone to limit the detection of genetic variance [24]. To prevent effect-size and subsequent type II errors from affecting the interpretation of our measure of inter-stage genetic variance, we used the entire subsample of 6384 gene expression traits showing non-zero genetic variance in both stages for further analyses.

We determined the genetic covariance between traits expressed at each stage using bivariate models implemented according to

$$Y=\hspace{0.17em}\mu +Z_l{\delta }_l+\epsilon$$ 2

where $Z_l$ is a design matrix for the line random effect. We modelled the covariance structure among traits expressed at different stages at the line level $({\delta }_l)$ using an unstructured 2 × 2 covariance matrix and at the residual $(\epsilon )$ level using a diagonal covariance matrix. The Wolbachia fixed effect was included as for the univariate model.

We tested whether each expression trait was consistent with an inter-stage genetic correlation of 0, 1, and −1 using three log-likelihood ratio tests (LRT). Tests of null correlations were performed by comparing the unconstrained model (2) to one where the among-trait covariance structure $({\delta }_l)$ was constrained to a diagonal matrix. We tested whether inter-stage correlations were significantly different from 1 and −1 using LRT between the unconstrained model (2) and one where ${\delta }_l$ was constrained to

$$\hspace{0.17em}\left[\begin{array}{cc}{V}_{a_{\mathrm{larva}}}& (-)1\ast \hspace{0.17em}\sqrt{V_{a_{\mathrm{larva}}}}\ast \sqrt{V_{a_{\mathrm{adult}}}}\hspace{0.17em}\\ (-)1\ast \hspace{0.17em}\sqrt{V_{a_{\mathrm{larva}}}}\ast \sqrt{V_{a_{\mathrm{adult}}}}& V_{a_{\mathrm{adult}}}\end{array}\right]$$

Where $V_{a_{\mathrm{larva}}}$ and $V_{a_{\mathrm{adult}}}$ were the among-line variance components obtained from the unconstrained model (2). Values of 1 and −1 were used alternatively. The Sommer function is based on covariance, and thus the correlations values are not limited to −1/+1. Some genetic correlations calculated for low variance genes were thus < −1 or over > 1. In rare cases, LRT suggested a deviation from −1 for correlations under −1, or a deviation from 1 for correlations over 1; we corrected for those as they were not significantly different from the tested limit.

Furthermore, we compared inter-stage correlations with previously published datasets of biologically similar cases, namely, genetic correlations between traits expressed at different temperatures in adult males and adult females (as development time also vary with temperature, differences in maturity existed between those samples) and between males and females. Inter-temperature datasets were collected with microarray technologies, while inter-sexual datasets were collected using RNAseq. Those supplemental datasets [25–27] were kindly provided in the appropriate format by the authors. We used the same 41 lines as those used in our dataset, apart from inter-sexual comparisons, where one of the lines we used was missing, thus only 40 lines were used. We performed the same sequence of analyses on these three datasets: females at 18°C and 25°C, males at 18°C and 25°C, and RNAseq on males and females at 25°C.

Comparing whether gene expression inter-stage correlations were different from inter-temperature and inter-sexual genetic correlations, we applied LRT between the unconstrained model (2) performed on the inter-stage dataset and one where ${\delta }_l$ was constrained to:

$$\hspace{0.17em}\left[\begin{array}{cc}{V}_{a_{\mathrm{larva}}}& \mathrm{corr}\ast \hspace{0.17em}\sqrt{V_{a_{\mathrm{larva}}}}\ast \sqrt{V_{a_{\mathrm{adult}}}}\hspace{0.17em}\\ \mathrm{corr}\ast \hspace{0.17em}\sqrt{V_{a_{\mathrm{larva}}}}\ast \sqrt{V_{a_{\mathrm{adult}}}}& V_{a_{\mathrm{adult}}}\end{array}\right]$$

where $\mathrm{corr}$ corresponds to the genetic correlation of a trait in the tested dataset, and $V_{a_{\mathrm{larva}}}$ and $V_{a_{\mathrm{adult}}}$ correspond to the among-line variance components of model (2). We obtained the number of gene expression traits whose correlations significantly differed between datasets. If compared datasets were similarly constrained, we would expect a 50/50 ratio between significantly more constrained and less constrained correlated gene expression traits. We calculated the deviance from this 50/50 expected ratio using a binomial test.

Finally, we performed Gene Ontology (GO) term enrichment analyses to determine whether inter-stage constrained genes, as well as independent genes, were enriched in biological process and molecular function. We used DAVID 2021 to compare the list of genes that were significantly positively constrained (inter-stage genetic correlations significantly different from −1 and 0, but not from 1), significantly negatively constrained (inter-stage genetic correlations significantly different from 1 and 0, but not from −1) and significantly independent (inter-stage genetic correlation significantly different from 1 and −1, but not from 0) to the background list of the 7414 expressed genes identified in our dataset using the ‘functional annotation chart’ option. All GO terms that were significantly enriched (pvalue) were reported.

3. Results

(a) . Genetic (de)coupling between life stages

When considering the 6384 gene expression traits that showed some level of genetic variance in larvae and in adults, almost a third (29.6%) were significantly correlated between life stages (table 1). Most genetic correlations were positive (1878 at the p-value level, 1146 at 5% fdr; figure 1), with only 12 significant negative correlations (at the p-value level, 1 at 5% fdr). Going beyond detecting only correlations between pairs of genes, we also explored whether metamorphosis enables greater independence between traits by testing whether inter-stage genetic correlations differed significantly from 1. Almost half of the gene expression correlations were significantly lower than 1 (46.1%; figure 1; table 1).

Table 1.

Numbers and percentages of gene expression traits that are genetically correlated or independent.

genetic correlations	larvae–adults	(%)	males–females	(%)
total number of genetically variable gene expression traits	6384		2927
correlation ≠ 0 (p-value)	1890	30	1750	60
correlation ≠ 0 (fdr)	1147	18	1650	41
correlation ≠ 1 (p-value)	2943	46	1395	48
correlation ≠ 1 (fdr)	2487	39	1191	41

Figure 1.

Distribution of inter-stage genetic correlations. The white dashed line corresponds to the median genetic correlation. Colours indicate levels of independence: correlated genes that are significantly different (p-value) from 0, but not from 1 or −1, are in blue; independent genes that are significantly different (p-value) from 1 and −1, but not from 0, are in orange; all other genes are in grey. To simplify representation, correlations less than −1 and greater than 1 (see Material and methods) are not represented.

(b) . Functions of inter-stage genetically (un)constrained traits

We further identified biological processes and molecular functions that were genetically constrained or independent across stages using GO term enrichment analyses. First, we tested which functions were enriched in the positively constrained genes (correlation significantly different from 0 and −1, but not from 1). Inter-stage genetically constrained gene expression traits were significantly enriched in 22 GO terms (electronic supplementary material, table S1). A vast majority of those terms were associated with proteins, including protein synthesis (GO: 0030488, p = 0.006, GO: 0003743, p = 0.027, and GO: 0006413, p = 0.028), transport and localization (GO: 0015031, p = 0.005, GO: 0008104, p = 0.014, and GO: 0034613, p = 0.032), and protein catabolism (GO: 0032435, p = 0.024). GO terms associated with cell division and growth were also enriched in positively constrained gene expression traits, from mitosis (GO: 0000723, p = 0.026, GO: 0006303, p = 0.036, and GO: 0097602, p = 0.040), to tissue organization and morphogenesis (GO: 006024, p = 0.015, GO: 0071711, p = 0.015, and GO: 0007394, p = 0.050). Importantly, four terms associated with detoxification were enriched in genetically constrained traits: GO: 008194, p = 0.012, GO: 0004364, p = 0.020, GO: 0006749, p = 0.042, and GO: 0016757, p = 0.047. Finally, innate immune response (GO: 0045087, p = 0.025) was also enriched in genetically constrained genes.

Genes that showed independent expression across life stages (correlations significantly different from 1 and −1, but not from 0) were significantly enriched in 72 functions and processes (electronic supplementary material, table S2). Functions and processes that varied independent between life stages were predominantly associated with energy metabolism, such as lipid (GO: 0006869, p = 0.002), amino acid (GO: 0006865, p = 0.005, GO: 0003333, p = 0.013, and GO: 0015171, p = 0.014), sugar (GO: 0008643, p = 0.017 and GO: 0015149, p = 0.032), and ion and general transmembrane transport (GO: 0055085, p < 0.001, GO: 0022857, p < 0.001, GO: 0015293, p = 0.007, GO: 0006814, p = 0.024, GO: 0016746, p = 0.029, GO: 0035435, p = 0.030, and GO: 0015116, p = 0.041), metabolism of sugar (GO: 0005975, p < 0.001, GO: 0006098, p = 0.017, GO: 0005977, p = 0.022, and GO: 0009312, p = 0.038), nucleotides (GO: 0006189, p = 0.014, GO: 0006164, p = 0.016, GO: 0009113, p = 0.017, and GO: 0003951, p = 0.041), and lipids (GO: 0006629, p = 0.005, GO: 0008654, p = 0.005, and GO: 0008289, p = 0.012, GO: 0006633, p = 0.018, GO: 0004467, p = 0.024, and GO: 0006631, p = 0.046), oxidation-reduction processes (GO: 0016491, p = 0.002), and protein and amino acid degradation (GO: 0006508, p < 0.001, GO: 0004180, p = 0.002, GO: 0030433, p = 0.007, GO: 0017171, p = 0.009, GO: 0004177, p = 0.030, and GO: 0008237, p = 0.044), in particular endopeptidase activity (GO: 0004252, p < 0.001, GO: 0004222, p = 0.008, GO: 0004866, p = 0.015, and GO: 0004867, p = 0.028). Independent gene expression traits were also enriched in GO terms that related to organs' development: general processes of cell division (GO: 0003777, p = 0.027, GO : 1901673, p = 0.040, GO: 0045317, p = 0.040, and GO: 0007052, p = 0.041), central nervous system development (GO: 0030182, p = 0.019, GO: 0008045, p = 0.027, GO: 0007411, p = 0.030, and GO: 0008088, p = 0.045), chitin and tracheal system (GO: 0035149, p = 0.005 and GO: 0008061, p = 0.008), and, expectedly, gonad development and reproduction function (GO: 0032504, p = 0.003, GO: 0019953, p = 0.003, GO: 0007283, p = 0.006, GO: 0008406, p = 0.016, GO: 0007300, p = 0.038, and GO: 0030513, p = 0.046). Non-stage-specific development functions were also independent across life stages such as cell adhesion (GO: 0007156, p = 0.004, GO: 0098609, p = 0.008, GO: 0007155, p = 0.009, and GO: 0050839, p = 0.014). Finally, several signalling pathways showed significant enrichment of low inter-stage genetic correlations, including G-proteins (GO: 0004930, p = 0.009 and GO: 0007186, p = 0.026) and Wnt-proteins (GO: 00017147, p = 0.047). Other enriched functions can be found in electronic supplementary material, table S2.

Although only 12 genes were significantly negatively correlated between life stages, we detected an enrichment in biological functions associated with symbiosis, muscles development, and oogenesis, although this significant enrichment was led by only a few genes (electronic supplementary material, table S3).

(c) . Comparisons to genetic (de)coupling between sexes

It is hard to interpret (biologically speaking) the percentage of the transcriptome genetically correlated or independent between life stages on its own. We re-analysed two similar previously published datasets describing genetic constraints between males and females, and between adults developed at 18°C or 25°C [25–27]. Results of genetic correlations across temperatures can be found in supplementary materials and we focus here on the comparison between sexes.

A large amount of the transcriptome was genetically constrained across sexes and across temperatures (figure 2; electronic supplementary material). When considering samples collected at 25°C in both males and females, over half (59.8%) of the 2927 gene expression traits that showed some genetic variance in both sexes were significantly correlated among sexes (r ≠ 0; table 1). Among those correlated gene expression traits, almost all showed positive significant intersexual genetic correlations, while only 5 out of 1750 showed negative correlations. As with inter-stage correlations, we assessed whether gene expression was genetically independent between sexes (i.e. r < 1); about half (47.7%) of the studied transcriptome had an inter-sexual genetic correlation significantly below 1 (table 1). A large proportion of the inter-sexual genetic correlations fell between 0 and 1 and were neither significantly independent, nor correlated (figure 2).

Figure 2.

Distribution of inter-sexual genetic correlations. The black line represents the distribution of inter-stage genetic correlations. The white dashed line corresponds to the median genetic correlation. Colours indicate levels of independence: correlated genes that are significantly different (p-value) from 0, but not from 1 or −1, are in blue; independent genes that are significantly different (p-value) from 1 and −1, but not from 0, are in orange; all other genes are in grey. To simplify representation, correlations less than −1 and greater than 1 (see Material and methods) were not represented.

To directly compare inter-stage and inter-sex constraints, as well as which types of constraints were more significant, we considered the 1731 genes that were positively genetically correlated in both datasets. Over one quarter of the genetic correlations (28.5% at p-value, 18.5% at the 5% fdr) differed significantly between the two datasets (figure 3). In most cases (i.e. 88%), genetic correlations were lower between stages than between sexes, which had a probability of 5e⁻⁷¹ of random occurrence.

Figure 3.

Comparison between inter-sexual and inter-stage genetic correlations. Only correlations between 0 and 1 are represented. Grey squares indicate genes whose genetic correlations did not significantly differ between datasets. Green dots denote genes that were significantly more independent across stages than across sexes. Purple triangles represent genes that were significantly more correlated across stages than across sexes.

4. Discussion

Analysing thousands of gene expression traits, we found that around half of the quantitative traits measured were evolutionarily independent between larvae and adults, although inter-stage genetic correlations remained prevalent in one third of the transcriptome. Several biological functions, such as protein synthesis and detoxification, emerged as being under intense constraints. Other functions, such as energy metabolic processes, were found to be mostly free to evolve independently between life stages. Negative genetic correlations appeared rare and concerned less than 1% of measured traits. Adult transcriptomes from other datasets indicated that genetic constraints between larvae and adults were lower than between sexes or across environments (table 1).

The present study provides an overview of evolutionary constraints between life stages at the scale of the phenome. Previous work had, so far, only focused on unique or small a set of traits spread over various taxa [9–16,28], limiting our understanding of how traits separated by metamorphosis are genetically independent or constrained. Our dataset revealed both significant genetic constraints among life stages (r ≠ 0) and widespread evolutionary independence (r ≠ 1) in the same organism (figure 1). Our comprehensive approach, including assessment of which biological functions are mostly (un)constrained, suggests when to expect, or not to expect, genetic constraints between traits expressed before and after metamorphosis.

Gene Ontology Terms enrichment analyses revealed groups of genes that belonged to the same biological function and were more often associated with genetic constraints or genetic independence. Positive genetic covariance was often found in biological functions that are needed throughout the life of an individual, such as protein metabolism and immune response (electronic supplementary material, table S1). Importantly, four GO terms enriched in genetically correlated genes (terms related to glycosyltransferase, GO: 008194 and GO: 0016757, and glutathione metabolism, GO: 0004364 and GO: 0006749) have been particularly studied for their role in insecticide metabolization [29–31]. Comparable studies of genetic constrains spanning a large number of functions are rare, but Blows et al. [32] showed that traits involved in transcriptional regulation were evolving under strong genetic constrains in adult D. serrata males (i.e. within a given life stage and sex). Here, we did not find any transcription term enriched for genetic correlations, suggesting that these constraints do not operate across life stages. Genetic constraints can also vary among populations, species or environments. Some studies report that genes causing sexual antagonism (i.e. genetic correlations across sexes that have opposite effects in each sex fitness) were enriched in functions associated with development [33], but these findings were not confirmed in other studies [34,35].

By contrast, many genes related to energy metabolism were genetically independently expressed. Decoupling energy production and expenditure would present itself as beneficial when larvae and adults have different nutritional needs [36], invest in separate functions (feeding pre-metamorphosis versus reproduction), and have distinct locomotion methods. Importantly, all studied genes were expressed in each larval and adult sample, so that genetically independent functions are not held by discrete, life stage specific, traits. Further, gene expression traits that were genetically independent between life stages were significantly enriched in processes related to organ formation (electronic supplementary material, table S2); this result may be particularly detectable as we studied late-stage larvae that are preparing metamorphosis [37]. We highlighted two signalling pathways in the independently expressed genes: (i) the G-protein signalling pathways, involved in a wide array of functions in Drosophila [38] and in the metamorphosis in several species [39,40], and (ii) the Wnt-protein signalling pathways, whose role has been extensively studied in development in Drosophila [41–43] and metamorphosis in a wide-range of species [44,45].

Identifying the functions under constraints (or not), enables the prediction of evolutionary dynamics, sometimes with community-scale consequences. We found that several immunity gene expressions were genetically correlated between larvae and adults. These correlations could modulate how pathogens influence the evolution of host immunity at both stages. The spread of a virus specifically infecting adults could drive the evolution of elevated adult immunity, which, due to the aforementioned constraints, would lead to elevated immunity in larvae too. Larval pathogens may hence suffer from selection applied by adult pathogens, and vice versa. Similarly, inter-stage genetic correlations of traits involved in insecticide resistance could have deleterious consequences for insect pest control, as selection for resistance at one stage would increase insecticide resistance throughout the life cycle. Such inter-stage dependence due to coupled traits can occur whenever functions are genetically correlated between stages, as was insecticide metabolization and immunity in this study. In order to generalize when and where to expect evolutionary coupling between stages, future studies should test whether the functions revealed by our phenom-wide approach are genetically independent or constrained in other populations and species.

We observed negative genetic correlations between larval and adult expression in only a small number of the genes sampled (37 genes; 0.5%) (electronic supplementary material, table S3). This contrasts with results from previous studies that reported negative genetic correlations between larval speed of development and adult immunity in Aedes aegypti mosquitoes [28] and for viability between life stages of the ascidian Ciona intestinalis [16] Importantly, however, the nature of the traits assessed in the above studies versus those analysed here are different: viability and developmental speed are fitness components that should, in theory, always be maximized, but are constrained by limited resources. With viability traits, negative genetic correlations can correspond to trade-offs [46]. By contrast, intensity of gene expression is of little cost on its own and relationships with fitness vary among genes. Our data therefore show that most of the pleiotropy between traits separated by metamorphosis cause positive genetic correlations, but that negative genetic correlations are nonetheless possible, in particular when driven by resource allocation constraints.

Inter-stage genetic constraints were lower than inter-sexual and inter-temperature genetic correlations (figures 2 and 3). These results are consistent with a previous study on the fly Neodiprion lecontei, which showed that gene expression levels differed more between life stages than between sexes [47]. The greater number of positive genetic correlations between sexes than stages may result from selection being more similar between male and female adults than between larvae and adults [17]. Nevertheless, gonochorism, the separation of male and female function in different individuals, is a more successful strategy (95% of animal species [48]) than metamorphosis (over half of the animal kingdom [1]). Although metamorphosis seems to be more successful at decreasing the genetic constraints between phenotypes than gonochorism, the success of gonochorism may demonstrate a stronger selection on decoupling male and female functions.

Despite revealing a large amount of genetic constraints (or lack thereof), our dataset included several limitations. First, larval samples contained both sexes, which created environmental variance that may have hindered measurable genetic variance. Although sexing larvae is technically feasible, it is an arduous task that was not compatible with the scale of our study. A second limitation, which our study shares with most gene expression analyses, is that transcript numbers may not always reflect protein production as alternative splicing may occur down-stream. As such, alternative splicing may further decrease genetic constraints between stages, and thus reduce genomic conflicts [49]. Thirdly, using inbred lines, such as those of the DGRP panel, has inherent limitations. Phenotypic variation is often higher than in outbred lines [50]. Genetic correlations estimated from inbred lines encompass broad sense heritability, as correlations due to epistasis and dominance are also detected [51]. However, the results on inter-sex (figure 2) and across-environment constraints (electronic supplementary material, figure S1), which we compare our data with, were also produced with inbred lines and thus are expected to contain the same overestimation of variance. Finally, the inter-stage genetic correlations we measured can result from pleiotropy or linkage disequilibrium between larval and adult expression, but can also result from genetic correlations with other traits that themselves are coupled across life stages. Indeed, previous studies have shown that gene expression traits are extensively genetically correlated within a life stage [52,53]. This is, however, a limit shared by other comparable studies [25–27] and remains a topic of investigation on its own right.

Our work is qualitatively different from a number of previous studies that compared sets of genes expressed, or not, at different life stages in animals with complex life cycles, such as in frogs [54,55], a saw fly [47], D. melanogaster [56], and Plasmodium sp. [57]. This approach cannot inform on genetic and evolutionary constraints, even more so if a single genotype is considered. Revealing genetic constraints necessitates selection experiments, mutant analysis or quantitative genetics tools to compare numerous genotypes or populations. By analysing the transcriptome of 41 inbred lines, we found that the strength of genetic constraints between life stage varied widely among genes (figure 1). Comparing the transcriptome of single genotypes at different stages is not sufficient: when genetic constraints among stages are low, other genotypes would be free to express genes differently at each life stages. Similarly, when gene expression is constrained among stages, transcriptional intensity could differ strongly among stages despite being controlled by the same genetic factors. The quantitative genetics tools used in the present study thus provide an unprecedented perspective on genetic constraints between transcriptomes expressed in different life stages.

Genetic constraints become evolutionary constraints (i.e. antagonistic pleiotropy) when selection acts in different directions across genetically correlated traits. Ideally, we would complement our results by measuring the strength and direction of natural selection for each trait at each life stage. This would be possible for a few traits, as shown by Goedert & Calsbeek [15] in Rana sylvatica frogs, but impossible at a phenome-wide scale. An alternative may be to assess whether selection is congruent or antagonist in biological functions that were identified to contain numerous constrained/independent genes in our dataset. For example, measuring selective forces operating on antifungal and antivirus immunity in larvae and adults of the original population, in their original environment, would reveal if the genetic constraints identified in our study result in evolutionary constraints. Ecological context does not only determine selection but also affects phenotypes through genotype-by-environment interactions. The transcriptome of insects [26], like that of plants, vertebrates, and microorganisms [58–60] is under the influence of ubiquitous genotype-by-environment interactions. Genotype-by-environment interactions are likely to also affect the genetic correlations between stages or sexes we report here. An extension of the present work would therefore explore the ecological context of genetic variation for its simultaneous effect on natural selection and heritable variation.

Altogether, our results greatly support the hypothesis that metamorphosis favours the independent microevolution of quantitative traits expressed before and after metamorphosis. This idea aligns with the adaptive decoupling hypothesis initially formulated for the macroevolution of discrete characters. Although a large part of the transcriptome was genetically constrained, our comparison with similar datasets showed that genetic independence between larvae and adult is larger than between sexes or across environments. Our transcriptome-wide results will help in predicting when and where to expect evolutionary constrains, or independence, between functions expressed at distinct life stages.

Ethics

This work did not require ethical approval from a human subject or animal welfare committee.

Data accessibility

New data from this article have been deposited with the Gene Expression Omnibus (GEO) database under accession no GSE226174. This paper also used previously published data, available in GEO under accession nos. GSE67505 and GSE67505, in the ArrayExpress database, accession no. E-MTAB-3216, and in SRA, accession no PRJNA615927.

Supplementary material is available online [61].

Declaration of AI use

We have not used AI-assisted technologies in creating this article.

Authors' contributions

J.M.C.: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, resources, validation, visualization, writing—original draft, writing—review and editing; S.N.: methodology, resources, supervision, writing—review and editing; S.F.: conceptualization, funding acquisition, investigation, methodology, project administration, resources, supervision, writing—original draft, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

This study was funded by a grant from the Santé des Plantes et Environnement department of INRAE to S.F. and an Agreenskills Plus fellowship to J.M.C.