Evolutionary diversification of ancestral genes across vertebrates and insects

Abstract

Background

Vertebrates and insects diverged approximately 700 million years ago, and yet they retain a large core of conserved genes from their last common ancestor. These ancient genes present strong evolutionary constraints, which limit their overall sequence and expression divergence. However, these constraints can greatly vary across ancestral gene families and, in at least some cases, sequence and expression changes can have functional consequences. Importantly, overall patterns of sequence and expression divergence and their potential functional outcomes have never been explored in a genome-wide manner across large animal evolutionary distances.

Results

We focus on approximately 7000 highly conserved genes shared between vertebrates and insects, and we investigate global patterns of molecular diversification driven by changes in sequence and gene expression. We identify molecular features generally linked to higher or lower diversification rates, together with gene groups with similar diversification profiles in both clades. Moreover, we discover that specific sets of genes underwent differential diversification during vertebrate and insect evolution, potentially contributing to the emergence of unique phenotypes in each clade.

Conclusions

We generate a comprehensive dataset of measures of sequence and expression divergence across vertebrates and insects, which reveals a continuous spectrum of evolutionary constraints among highly conserved genes. These constraints are normally consistent between these two clades and are associated with specific molecular features, but in some cases we also identify instances of lineage-specific diversification likely linked to functional evolution.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13059-025-03699-z.

Background

Vertebrates and insects are two clades of bilaterian animals which diverged approximately 700 million years ago (MYA; [1]). On one hand, these clades exhibit relatively comparable body plans and homologous adult tissue types [2]. On the other hand, they also evolved unique biological traits [3–5], and are characterized by rather distinct genomic and molecular evolutionary rates [6, 7].

Notwithstanding their large evolutionary distance, vertebrates and insects share a strong core of conserved genes, representing ~25–65% of their protein-coding complement [8–10]. Since these genes are still recognizable as evolutionarily related after 700 million years, their encoded protein sequence and expression profiles are likely subjected to strong evolutionary constraints. However, even within this set of highly conserved genes, the degree of such constraints can greatly vary. Evolutionary constraints have a direct read-out in the overall levels of molecular diversification (in terms of sequence and/or expression divergence) within each given gene family, and are generally expected to be determined by their different functional requirements. For instance, genes encoding highly structured proteins usually have greater sequence constraints than those with abundant intrinsic disordered regions [11, 12]. Similarly, genes with neural-specific expression in vertebrates show higher expression conservation than those specifically expressed in other tissues [10, 13, 14].

Remarkably, gene duplication can radically alter these intrinsic constraint patterns, both at the sequence and expression levels. Accordingly, highly duplicated genes generally exhibit greater molecular diversification compared to single-copy orthologs [10, 15–17]. This increased molecular diversification is normally due to the reduced constraints upon duplication, as in principle the ancestral function can be carried out by only one of the gene copies while the other one could acquire sequence/expression changes that will not initially be subjected to similar purifying selection pressures. Thus, the resulting molecular diversification is mainly expected to have a largely neutral effect, particularly during the initial stages following gene duplication. However, in at least some cases, such changes eventually lead to the evolution of new functional properties [16]. Therefore, even if this functional interpretation applies only to a limited number of events, understanding patterns of molecular divergence in a gene family may guide sound hypotheses about its functional evolution.

Notably, while changes in sequence or expression alone can drive functional evolution, synergistic modifications at both levels may be more efficient to this aim. Indeed, the combined effect of sequence and expression variation in determining functional evolution has been proved by single-species studies for several groups of paralogs [18–21]. However, sequence and expression changes within gene families have never been investigated together across large animal phylogenies and in a genome-wide manner. On one hand, some landmark studies reconstructed global patterns of gene gains, losses, and duplications throughout animal evolution [8, 9, 22], but they neither investigated sequence divergence within gene families nor complemented their findings with expression data. On the other hand, large comparative transcriptomic studies normally focus on sets of conserved (often 1:1) orthologs [10, 13, 14, 23, 24], but do not usually integrate evolutionary expression changes with information about sequence evolution.

In this work, we use novel measures of sequence and expression divergence to investigate global patterns of molecular diversification of ancestral gene orthogroups within and between vertebrates and insects, with the assumption that they could be informative not only about their molecular constraints but also regarding trends of functional evolution in the two clades. We focused on a set of ~7000 highly conserved gene orthogroups, characterizing general features associated with molecular diversification and identifying groups of genes that exhibit common or unique diversification patterns across vertebrates and insects. Finally, we determined potential connections between molecular diversification and tissue-related attributes, highlighting the important role of highly conserved genes for the evolution of clade-specific phenotypic traits.

Results

Global rates of sequence and expression diversification in vertebrates and insects

We selected a symmetric, time-calibrated phylogeny of eight gnathostome vertebrates and eight insects, with pairs of vertebrate and insect species located at equivalent phylogenetic positions on the two main branches (Fig. 1a). As these two branches presented an equal number of species and equivalent topology, our phylogenetic tree was optimal to directly compare the evolution of molecular traits between the two clades. We specifically focused on protein-coding genes broadly conserved across this phylogeny (6787 orthogroups [10]). For all these genes, we extracted (i) their encoded protein sequence and (ii) their expression profile across seven homologous tissue types (i.e., neural, testis, ovary, muscle, excretory system, epidermis and guts) (see Methods). We then used these features to compute, respectively, the sequence and expression similarities of homologous genes either within vertebrates or within insects (Fig. 1b, c, Additional file 1: Fig. S1a and Additional file 2: Table S1; see Methods), which we used as proxies of molecular diversification in the corresponding gene orthogroup. Importantly, we recalculated these similarities across a wide range of alternative computation approaches, phylogenetic setups, and orthogroups selection strategies (Additional file 1: Fig. S1b, c), consistently obtaining comparable results (see Additional file 3: Supplementary note for a comparison of this and all downstream analyses).

Fig. 1.

Framework overview and definition of sequence and expression similarities. a Left: time-calibrated phylogenetic tree including the scientific acronyms of the 8 vertebrate and 8 insect species considered in this study. Hsa human, Mmu mouse, Bta cow, Mdo opossum, Gga chicken, Xtr tropical clawed frog, Dre zebrafish, Cmi elephant shark, Dme fruit fly, Eba marmalade hoverfly, Aae yellow fever mosquito, Bmo domestic silk moth, Tca red flour beetle, Ame honey bee, Bge cockroach, Cdi mayfly (see Methods for corresponding scientific names). Top right: example of one of the 6787 considered gene orthogroups, where each dot represents a gene. Bottom right: tissues included in our bulk RNA-seq dataset. Evolutionary distances were derived from timetree [1] (MYA: million years ago) and animal silhouettes were generated through Bing Chat by Microsoft (2023) https://www.bing.com/search. b, c Scheme for the computation of the sequence (b) and expression (c) similarity measures. The procedure was performed separately for vertebrates and insects, returning two values per orthogroup. See Additional file 1: Fig. S1a for step-by-step schematics. d, e Distributions of the sequence (d) and expression (e) similarity values for all gene orthogroups (n = 6787) within vertebrates (purple) and insects (orange)

The overall distributions of sequence similarities revealed that protein sequences are in general more conserved in vertebrates compared to insects (Fig. 1d; two-sided Wilcoxon’s test, p-value < 2e-16). This is partially expected because of the shorter generation times [7] and the smaller body sizes of insect species compared to vertebrates. On the contrary, expression profiles across tissues show similar conservation levels between the two clades (Fig. 1e). This is in line with previous studies, which demonstrated how transcriptional networks evolve with largely comparable rates between mammals, birds, and insects [25], or how expression divergence reaches a relatively quick plateau with increasing evolutionary distances, both within mammals [13, 26, 27] and fruit flies [28]. In summary, while sequence changes seem to have been overall more prevalent in insects compared to vertebrates, expression changes contributed similarly to the modification of the ancestral molecular landscape in the two clades.

Comparisons of sequence and expression diversification within and between clades

Next, we separately compared rates of sequence or expression evolution of individual gene orthogroups between the two clades. In general, gene orthogroups show significant and positive correlation both in terms of sequence similarities (Person’s correlation coefficient 0.626, p-value < 2e-16; Fig. 2a) and expression similarities (Pearson’s correlation coefficient 0.665, p-value < 2e-16; Fig. 2b). This suggests that genes that tend to be more or less conserved in vertebrates also evolve with similar relative rates in insects, both from the sequence and the expression perspectives. Thus, the disposition to undergo molecular diversification driven by either type of molecular change seems to be, to a large extent, an intrinsic potential of each gene orthogroup, which is independently but comparably fulfilled in the two clades.

Fig. 2.

Correlation of sequence and expression similarities between and within clades. a, b Correlation of the sequence (a) and expression (b) similarities of all gene orthogroups between vertebrates (x axis) and insects (y axis) (n = 6787). c, d Correlation between the sequence (x axis) and expression (y axis) similarities of all gene orthogroups within vertebrates (c) and insects (d) (n = 6787)

Moreover, we found that conservation levels of these two molecular traits tend to be associated, as we detected positive and significant correlations between sequence and expression similarities both within vertebrates (Pearson’s correlation 0.499, p-value < 2e-16; Fig. 2c) and within insects (Pearson’s correlation 0.450, p-value < 2e-16; Fig. 2d), in line with what was previously shown for more closely related species [26, 29, 30]. However, it should be noted that this association between sequence and expression conservation did not emerge from other studies [31, 32], potentially due to the caveats of using correlations as measures of expression divergence [33] (see Additional file 3: Supplementary note for a comparison between alternative metrics of sequence and expression conservation). Importantly, despite their positive and significant association, sequence and expression similarities show a substantial deviation from perfect alignment (Fig. 2c, d). This indicates that, although genes predisposed to molecular diversification experience generally increased levels of both sequence and expression alterations, the exact interplay between these factors may vary across gene orthogroups and between clades.

Features associated with molecular diversification

We then used sequence and expression similarities to define groups of highly or lowly diversified genes within vertebrates and within insects. In particular, we selected the 500 gene orthogroups with the lowest/highest sequence or expression similarities in each clade, defining eight groups of genes with the most extreme diversification profiles (Fig. 3a and Additional file 2: Table S2). According to our assumptions, highly diversified gene orthogroups should include the ones with weaker overall constraints, but also those that have more often undergone functional evolution; on the other hand, lowly diversified orthogroups are highly constrained and more likely to have strictly preserved their ancestral functions. This assumption is supported by the observation that, even if all the genes in our dataset are ancestral, highly diversified genes are conserved in fewer species compared to all and lowly diversified genes, a pattern particularly evident in insects (Additional file 1: Fig. S1d).

Fig. 3.

Characterization of highly and lowly diversified gene orthogroups. a Distribution of the sequence (top) and expression (bottom) similarity values of all gene orthogroups in vertebrates (left; purple) or insects (right; orange). The highly (n = 500) and lowly (n = 500) diversified gene orthogroups are highlighted in white and black, respectively. b Proportions of highly diversified, all, and lowly diversified orthogroups associated with a lethal, non-lethal, and uncharacterized phenotype. Each plot refers to the corresponding distributions in panel a. See Methods for phenotype definition. c Duplication profiles of highly diversified, all, and lowly diversified orthogroups, where species with duplications are defined as all species with at least two conserved paralogs in a given orthogroup. Each plot refers to the corresponding distributions in a. d Distribution of Tau values for genes included in highly diversified, all, and lowly diversified orthogroups. Each plot refers to the corresponding distributions in a

First, in-depth characterization of these orthogroups revealed that highly and lowly diversified genes were associated with significantly less and more lethal phenotypes, respectively, compared to all genes (one-sided Fisher’s tests: all p-values ≤ 0.001, except for the non-significant p-value of the vertebrate highly diversified orthogroups in terms of sequence; Fig. 3b). Second, and in part consistent with that observation, they also showed significantly higher and lower duplication levels compared to the background (one-sided Fisher’s tests: all p-values ≤ 0.0005; Fig. 3c). This pattern was particularly strong for vertebrates, most likely determined by the two rounds of whole genome duplications that characterize their early evolutionary history [34]. Moreover, these increased duplication levels were observed for highly diversified genes derived from both sequence and expression-based classifications, in line with the idea that gene duplication releases the evolutionary constraints of ancestral genes and potentially triggers their functional evolution through the combination of different molecular mechanisms [35].

Finally, we used a measure of tissue-specific expression known as Tau [36] to test for potential biases in tissue specificity among the selected genes. All highly and lowly diversified gene groups were, respectively, significantly more and less tissue-specific than all genes together (one-sided Wilcoxon’s test, all p-values < 2e-16; Fig. 3d). On one hand, this was consistent with the concept that orthogroups with minimal mutation load will include mainly essential housekeeping genes, characterized by broad expression profiles [37]. On the other hand, it also indicated that gains of tissue specificity might be among the expression changes most employed to generate molecular diversification, and that a substantial proportion of this diversity might contribute to the evolution of tissue-related traits (in line with [10, 13, 14, 26, 27, 38, 39]). Importantly, the fact that also the highly diversified genes from the sequence-based classification show a dramatic increase in tissue specificity is another strong indicator that both sequence and expression modifications might cooperate in determining functional evolution.

Functional categories with common diversification patterns between clades

Then, we focused on those genes that undergo similar extreme levels of molecular diversification (very low or very high) in both vertebrates and insects and through the combined action of sequence and expression changes. First, we selected all the genes belonging to at least three out of four groups of lowly diversified genes (core of the Venn diagram, Fig. 4a and Additional file 2: Table S3), which should include genes that mainly preserve highly ancestral functions. As expected, a GO enrichment analysis on these genes returned categories related to basic and widely conserved cellular organelles (e.g., nucleus, ribosome, nucleolus), molecular functions (e.g., rRNA binding) or biological processes (e.g., translation, mRNA/rRNA processing and splicing) (Fig. 4b and Additional file 2: Table S2).

Fig. 4.

Functional categories with common diversification patterns between clades. a, c Venn diagrams representing the overlap between lowly (a) and highly (c) diversified gene orthogroups defined based on vertebrate/insect sequence (blue) or expression (green) similarity. A common legend for both panels is depicted in a. b, d Significant categories (FDR corrected p-values ≤ 0.05, intersection ≥ 5, precision ≥ 0.05) from a GO enrichment analysis performed on the lowly (b) and highly (d) diversified gene orthogroups common to 3–4 groups depicted in a and c, respectively (center of the Venn diagrams). Results are limited to the top 20 categories, but full reports are available in Additional file 2: Table S3. The GO annotation was built through a human-based GO transfer (see Methods)

Second, we repeated the same analysis for gene orthogroups belonging to at least three out of four groups of highly diversified genes (core of the Venn diagram, Fig. 4c and Additional file 2: Table S3), representing orthogroups with pervasive patterns of molecular diversification. These genes showed clear enrichments in various cilium-related categories (e.g., axoneme), gamete generation, and extracellular regions (Fig. 4d and Additional file 2: Table S3). Many of these enrichments presumably arise from the great variation of male reproductive systems in animals, which are the result of a strong selective sexual and molecular pressure [24, 40]. For example, the axoneme enrichment likely reflects how this ancient microtubule-based structure, a core component of spermatozoa flagella in most animal species [41], was diversified across a wide range of reproductive niches. Importantly, similar results were obtained with both the human-based (Fig. 4b, d and Additional file 2: Table S3) and the fruit fly-based GO annotations (Additional file 1: Fig. S2a and Additional file 2: Table S3).

Functional categories with differential diversification patterns between clades

We next focused on gene orthogroups that are more divergent in one clade compared to the other, indicating that they have been differentially impacted by sequence or expression changes during vertebrate and insect evolution. In order to characterize these orthogroups, we computed the deltas of the respective sequence or expression similarities between the two clades (vertebrates minus insects; Additional file 1: Fig. S3a, b and Additional file 2: Table S1), which showed moderate but significant correlation across all orthogroups (Pearson’s correlation coefficient 0.349, p-value < 2e-16, Additional file 1: Fig. S3c). Given the non-overlapping initial distributions of sequence similarities between vertebrates and insects (Fig. 1d and Additional file 1: Fig. S3a), we applied a z-score transformation to the deltas distribution of sequence and, for coherence, of expression similarities (Additional file 1: Fig. S3d, e). In this way, the sign of the z-scored deltas is reflective of higher diversification rates in insects (positive values) or in vertebrates (negative values), but preserves the same characteristics of the original distributions (Additional file 1: Fig. S3d–f). We then performed a separate gene set enrichment analysis (GSEA) on these z-scored deltas for the sequence and expression similarities (Fig. 5a, Additional file 1: Fig. S2b and Additional file 2: Table S3), highlighting all genes belonging to significant categories (FDR-corrected p-value ≤ 0.01) in Fig. 5b, c.

Fig. 5.

Functional categories with diversification biases between clades. a Top significant categories (FDR-corrected p-values ≤ 0.01) from a GSEA performed on the z-scored deltas of sequence (left) or expression (right) similarities between vertebrates and insects. The GO annotation was built through a human-based GO transfer (see Methods). b, c Same plots as in Fig. 2a, b but where the genes driving all significant GSEA enrichments are highlighted. The genes driving positive enrichment scores (high deltas, more diversified in insects) are highlighted in orange, while those driving negative enrichments (low deltas, more diversified in vertebrates) are highlighted in purple. d Distribution of Tau values for genes in the orthogroups with diversification biases highlighted in b (top row) and c (bottom row) and the relative background composed of all orthogroups. The Tau values are separately plotted for the vertebrate and insect genes included in each orthogroup. The dashed, red lines contain the genes used as input for the Fisher’s exact test depicted in e. e Results of Fisher’s exact tests from the fisher.test function in R (alternative=”greater”) comparing the proportions of tissue-specific genes with diversification biases (Tau ≥ 0.75, dashed lines in d) in each tissue to the same proportions in the background. Tests with p-values ≤ 0.05 are considered significant. Abbreviations: NES, normalized enrichment score

In general, genes with a diversification bias in insects were enriched for several cilium and cytoskeleton-related GO categories, especially when this bias was expression-driven. In contrast, genes with a diversification bias in vertebrates were enriched for functions associated with synaptic structures, ion transport, and actin cytoskeleton. Importantly, while both vertebrate and insect genes with diversification biases tend to be more tissue-specific than their orthologs in the other clade (Fig. 5d), they most likely manifest their potential for functional evolution in different tissue contexts (Fig. 5e). On one hand, the insect group is mainly enriched in testis-specific genes, possibly reflecting the highly variable and specialized ciliary structure of their reproductive traits [42, 43], and consistent with the GSEA results. On the other hand, neural- and muscle-specific genes are strongly over-represented in the vertebrate group, also in line with their observed enrichment for neuronal functional categories [44, 45]. All together, this analysis suggests that at least part of this molecular diversification is likely to result in functional evolution, and might have been differentially used in vertebrates and insects to shape some of their distinct tissue-related traits.

Discussion

This work revolves around a core of ~7000 genes shared between vertebrates and insects, inherited from their last common ancestor approximately 700 MYA. Although vertebrates and insects differ in their overall evolutionary rates due to distinct generation times [7] and body sizes, these differences are consistent across genes and should not introduce biases in inter-gene comparisons. Given their ancient origin, these ancestral genes are likely subjected to strong evolutionary constraints, which limit their divergence in terms of encoded protein sequence and expression profiles. However, even if all ancestral genes are somewhat constrained, they are expected to show some variability in their tolerance to mutations. In fact, some genes keep performing their function through a wide array of sequence or expression alterations, while others are extremely sensitive to modifications of their ancestral molecular state. Moreover, although the majority of fixed molecular alterations in these genes are likely neutral (see Introduction), we know that they can sometimes mediate functional diversification. Thus, characterizing genome-wide landscapes of evolutionary constraints of ancestral animal genes would not only reveal which genes are more or less tolerant to molecular alterations, but also potentially identify those cases where these alterations lead to functional evolution.

Methodologically, we developed novel measures of sequence and expression divergence that align with more standard metrics but offer a few advantages (see Additional file 3: Supplementary note). Moreover, they are directly comparable between clades thanks to the symmetric structure of our phylogeny, where the equal number of species and equivalent topology in the vertebrate and insect branches remove eventual biases due to differential species sampling or evolutionary times (even if the results remain robust to changes in the phylogenetic setup or across different computational approaches; see Additional file 3: Supplementary note). These measures revealed that (i) sequence and expression divergence are highly variable among gene orthogroups but (ii) homologous genes generally share similar evolutionary constraints across clades and molecular layers. Thus, the disposition to undergo molecular diversification driven by either type of molecular change seems to be an intrinsic potential of each gene orthogroup that is independently but comparably fulfilled in vertebrates and insects.

Using this characterization of molecular diversification at the gene orthogroup level, we tried to determine what might drive differences in evolutionary constraints among these highly conserved gene families. While pinpointing causative factors is challenging, we identified some associated biological features. Genes with higher molecular diversification levels tend to belong to orthogroups with more duplicates, be more tissue-specific, and less lethal compared to other genes. This supports the model where duplication lifts at least some of the constraints of the ancestral gene, making it less susceptible to lethal mutations. However, the increased tissue-specificity of more diversified genes suggests that at least part of this diversification might not simply be due to lack of constraints, but might have been positively selected because of its functional effect in specific tissue contexts.

When is molecular diversification more probably linked to a functional outcome? Given the overall high correlation in evolutionary constraints between vertebrates and insects, it could be argued that genes for which diversification levels significantly differ between the two clades have more likely undergone functional changes in one of those clades. Functional evolution of ancestral genes, particularly after gene duplication, can occur through different modalities: for instance, neofunctionalization indicates the emergence of a completely novel function, while specialization implies the evolution of a slightly modified (or specialized) version of the ancestral role. Both novel and specialized functions are expected to be optimized for more restricted biological contexts (e.g., a particular tissue type) compared to more general ancestral functions, in line with the most common regulatory fates observed upon whole genome duplication [16, 46–48]. Consistently, we found that gene orthogroups with a diversification bias in vertebrates were significantly enriched for neural- and muscle-specific genes, whereas genes with higher molecular diversification levels in insects showed over-representation of testis-specificity. This contrast emphasizes that different tissues in vertebrates and insects are presumably most influenced by functional evolution of ancestral genes, and this process likely contributed to the unique complexity of vertebrate neurons and to the highly variable features of insect reproductive systems.

Conclusions

In this study, we present a comprehensive dataset capturing patterns of sequence and expression divergence across vertebrates and insects, revealing a broad and continuous spectrum of evolutionary constraints among deeply conserved genes. While many of these constraints are shared between clades and associated with specific molecular features, we also identify clear cases of lineage-specific divergence, hinting at episodes of functional innovation. This work provides new insights into the interplay between sequence and expression divergence as drivers of evolutionary change, and serves as a reference for future studies on gene evolution and molecular diversification across species.

Methods

Phylogenetic tree

The phylogenetic tree used in this publication is a subset of the phylogeny of bilaterian animals from [10], which we here restricted to the (eight) vertebrates and (eight) insects (sixteen species in total). The vertebrate and insect branches in the resulting time-calibrated phylogeny are monophyletic and symmetric, meaning that each vertebrate is paired with an insect located at an equivalent phylogenetic position and with relatively comparable divergence times (i.e., the two phylogenetic branches share the same topology). The vertebrate species include human (Homo sapiens, Hsa), mouse (Mus musculus, Mmu), cow (Bos taurus, Bta), opossum (Monodelphis domestica, Mdo), chicken (Gallus gallus, Gga), tropical clawed frog (Xenopus tropicalis, Xtr), zebrafish (Danio rerio, Dre) and elephant shark (Callorhinchus milii, Cmi). The insect species include fruit fly (Drosophila melanogaster, Dme), marmalade hoverfly (Episyrphus balteatus, Eba), yellow fever mosquito (Aedes aegypti, Aae), domestic silk moth (Bombyx mori, Bmo), red flour beetle (Tribolium castaneum, Tca), honey bee (Apis mellifera, Ame), cockroach (Blattella germanica, Bge), and mayfly (Cloeon dipterum, Cdi). See [10] for details on the assembly and gene annotation versions for each of the species.

Gene orthogroups

We considered the ancestral gene orthogroups from [10], which were obtained by running Broccoli (v1.2) [49] among all protein-coding genes from the 20 bilaterian species. Importantly, in order to avoid redundant gene homology calls, we selected one representative protein isoform for each gene in each species (i.e., the isoform with the longest coding sequence). These gene orthogroups originally included 7178 orthogroups widely conserved across bilaterian animals, which we filtered to select only the orthogroups that were conserved in at least two vertebrates and two insects (6787 out of 7178, 95% of the original orthogroups). For all the main analyses, we considered all paralogs from all species within each gene orthogroup, which allowed us to capture their overall diversification patterns. However, we also generated four distinct 1:1 orthogroup sets, each composed of a representative paralog per species (when present) selected according to different conservation criteria (Additional file 1: Fig. S1b). The general approach was to identify representative orthologs in each species that either most closely resemble the ancestral state (best-ancestral [BA]) or have diverged the most from it (best-divergent [BD]). This was determined by evaluating the overall sequence and expression similarity of each gene relative to the rest of the orthogroup. For the BA sets, we aimed at selecting the gene in each species with the highest sequence (BA-seq set) or expression (BA-expr set) conservation (step (iii) in Additional file 1: Fig. S1a). For the BD sets, we aimed at selecting the gene in each species with the lowest sequence (BD-seq set) or expression (BD-expr set) conservation (step (iii) in Additional file 1: Fig. S1a). The results were overall highly consistent, in line with the elevated overlap between 1:1 sets (Additional file 1: Fig. S1c), and with specific expected exceptions (see Additional file 3: Supplementary note).

RNA-seq data and gene expression quantification

We used the bulk RNA-seq data from [10], which includes samples for the sixteen species in our phylogeny spanning up to eight homologous adult tissue types. We selected seven of those tissues (neural, testis, ovaries, muscle, excretory, epidermis and gut), excluding adipose because of the missing samples for the elephant shark. As an expression measure for each gene, we used the expression proportions per tissue (i.e., sum across tissues = 1). The expression proportion per tissue is defined as (tissue_expr/all_tissue_expr), where “tissue_expr” is the average quantile normalized log2(TPMs+1) expression of the gene in the target tissue and “all_tissue_expr” the sum of the average quantile normalized log2(TPMs+1) expression values across all tissues. See [10] for further details. All expression files are available in the Supplementary dataset.

Computation of sequence similarities

We derived two representative measures of sequence similarities for each gene orthogroup, one within vertebrates and one within insects. Each of these clade-level measures was computed following a 5-step procedure: (i) computation of pairwise protein sequence similarity between all possible pairs of orthologs, based on protein alignments generated by mafft v7.222 [50] with default parameters (e.g., BLOSUM 62 as default scoring matrix). Sequence similarity was then calculated based on this alignment following the guidelines of the SIAS tool (http://imed.med.ucm.es/Tools/sias.html). In this approach, a score of 1 was assigned to perfectly matched amino acids or pairs with similar chemical-physical properties, while other mismatches or gaps were assigned a score of 0 (see Additional file 2: Table S6 and github repository). Importantly, these pairwise protein sequence similarities were computed considering each gene first as query and then as target, and the final alignment score was divided by the length of the query protein; (ii) for each query gene, computation of average sequence similarity between all its orthologs in each target species (values from step (i)); (iii) for each query gene, computation of average sequence similarity among all target species (values from step (ii); (iv) for each species, computation of the average sequence similarity between all its genes (values from step (iii); (v) for each clade (vertebrates or insects), computation of the average sequence similarity between all its species (values from step (iv). See Additional file 1: Fig. S1a for a detailed schematic of the procedure. Clade-level average sequence similarities for all gene orthogroups are provided in Additional file 2: Table S1. The sequence similarity values for the extra 1:1 orthogroup sets were obtained with a similar procedure but skipping steps (i) and (ii) because of the presence of only one representative gene per species (see Additional file 3: Supplementary note for relative discussion).

Computation of expression similarities

We derived two representative measures of expression similarities for each gene orthogroup, one within vertebrates and one within insects. Each of these orthogroup-level measures was derived following a 5-step procedure: (i) computation of pairwise expression similarity between all possible pairs of orthologs. The pairwise expression similarity was defined as follows:

where n represents the total number of considered tissue types (n = 7), i indicates each of the tissues in turn, x and y correspond to the two species being compared. Briefly, we summed up the absolute differences in expression proportions across corresponding tissues, normalizing this value for the number of comparisons. As this measure would be inversely proportional to the conservation of expression profiles between two genes, we computed the actual expression similarity by subtracting it from one and applying a logit transformation. The resulting low and high expression similarity values correspond to gene pairs with greatly divergent and conserved expression profiles, respectively; (ii) for each query gene, computation of average expression similarity between all its orthologs in each target species (values from step (i)); (iii) for each query gene, computation of average expression similarity among all target species (values from step (ii)); (iv) for each species, computation of the average expression similarity between all its genes (values from step (iii)); (v) for each clade (vertebrates or insects), computation of the average expression similarity between all its species (values from step (iv)). See Additional file 1: Fig. S1a for a detailed schematic of the procedure. Clade-level average expression similarities for all gene orthogroups are provided in Additional file 2: Table S1. The expression similarity values for the extra 1:1 orthogroup sets (represented in) were obtained with a similar procedure but skipping steps (i) and (ii) because of the presence of only one representative gene per species (see Additional file 3: Supplementary note for relative discussion).

Alternative measures for sequence conservation

We first downloaded and formatted human phastCons scores (one file per chromosome) from UCSC: http://hgdownload.cse.ucsc.edu/goldenPath/hg38/phastCons100way/hg38.100way.phastCons/chr*.phastCons100way.wigFix.gz. PhastCons scores are nucleotide-level scores which represent the posterior probability of each nucleotide to be in its most-conserved state based on the comparison with 100 vertebrate genomes [51]. For each human gene belonging to our gene orthogroups, we computed the average phastCons values among all the nucleotides of its coding sequence. Finally, we compared these values with the average similarity values between the same query gene and all its vertebrate orthologs (step (iii) in Additional file 1: Fig. S1a; see Additional file 1: Fig. S4a). Average phastCons value for each human gene is reported in Additional file 2: Table S5.

For the dN/dS analysis, we obtained the dN and dS values for gene pairwise comparisons between humans and other mammals in our dataset (mouse, cow, and opossum) from Ensembl BioMart v99 (released in January 2020 [52];) and calculated the relative ratios. Only comparisons where both orthologs were present within our 6787 bilaterian conserved orthogroups were included in the plots shown in Additional file 1: Fig. S4b.

We also tested alternative computation approaches for sequence similarities. First, we calculated sequence similarity directly using BLOSUM 62 and BLOSUM 45 matrices on the alignments generated by mafft v7.222 [50] (see above). Given that the score directly returned by using BLOSUM 62 and BLOSUM 45 does not range between 0 and 1, we adopted a normalization procedure to make it comparable across alignments. The final alignment score was calculated as:

where aln_score corresponds to the alignment score as computed based on either BLOSUM 62 or BLOSUM 45 across all positions, min_aln_score corresponds to the minimum alignment score (i.e., all protein positions are matched with gaps) and max_aln_score corresponds to the maximum alignment score (i.e., all protein positions are matched with identical amino acids). Importantly, as for the main analyses, the final alignment score is computed by considering each protein in the alignment first as query and then as target. The comparisons between these computation methods and relative results are reported in Additional file 1: Figs. S5–7. Finally, we evaluated the impact of removing leading and trailing alignment gaps on our main method, assuming these gaps represent sequences missing in one of the two species (Additional file 1: Fig. S8). See Additional file 3: Supplementary note for discussion on the comparison between all computation methods.

Testing the influence of the phylogenetic setup on sequence similarities

To evaluate the impact of altering the phylogenetic setup on sequence similarities, we re-computed the average sequence similarities across orthogroups upon removal of all possible combinations of one, two, or three species on each branch or upon removal of species in matched or unmatched phylogenetic positions on the two branches. All relative results are reported in Additional file 1: Fig. S9 and discussed in the Additional file 3: Supplementary note.

Testing the influence of paralog features on sequence similarities

We retrieved the number of paralogs directly from the gene orthogroups used throughout the manuscript. To determine the origin of each paralog, we used the original gene orthogroups from [10], which included four outgroup species. These outgroups allowed us to trace paralog origins up to the last common bilaterian ancestor. For each gene orthogroup, we (i) generated multiple alignments by running mafft v7.222 with the following parameters: --quiet --retree 2 --localpair --maxiterate 500; and (ii) built gene trees by running the fasttree function from FastTree v2.1.10 with the following parameters: -quiet -bionj. We then resolved eventual polytomies with the resolve_polytomy() function from the ete3 python library, and rooted each gene tree with the get_midpoint_outgroup() function from the same library (as described by [49]). We defined duplication nodes on a gene tree as all those nodes with at least one overlapping species in its descending branches, and we dated the duplication back to the last common ancestor of all the descending species, with some exceptions: we assigned NA to duplication nodes with a bootstrap support ≤ 0.5 or to duplication nodes dating back to the root where one of the descending branches only contained one species or the ratio between the number of species on the two descending branches was ≥ 4:1. For each gene, we considered as paralog origin the most recent duplication event from which it derived. All results relative to the association between paralog features and sequence similarity are reported in Additional file 1: Fig. S10. The vertebrate ohnologs shown in Additional file 1: Fig. S10e were obtained from [53]. See Additional file 3: Supplementary note for relative discussion.

Alternative measures for expression conservation

The expression correlations and expression distances (see Additional file 1: Fig. S11a-l) were based, respectively, on Spearman’s correlations and Euclidean distances (more precisely, 1 - Euclidean distance) of expression proportions across tissues. As for the sequence and expression similarities, we derived two representative measures of expression correlation/distance for each gene orthogroup, one within vertebrates and one within insects. These final measures were computed following the same 5-step strategy described for expression similarity (see above). See Additional file 1: Fig. S1a and Additional file 1: Fig. S11a for a detailed schematic of the procedure. All average expression correlation/distance values are provided in Additional file 2: Table S5.

We also evaluated the impact of an alternative approach for expression quantification on the resulting expression similarities. Instead of averaging the contribution of all paralogs (Additional file 1: Fig. S1a, step (ii)), we initially summed the expression of all paralogs from the same species. For this, we used the quantile-normalized log2(TPM+1) expression data at the tissue level from [10]). From these summed values, we calculated the relative expression across tissues and subsequently derived expression similarities (Additional file 1: Fig. S1a, steps (iii) to (v) described above), together with the Taus and the tissue-specificity call as described in [10]. All corresponding results are reported in Additional file 1: Fig. S12 and discussed in Additional file 3: Supplementary note.

Definition of housekeeping orthogroups

The definition of housekeeping genes was based on Tau, a measure of tissue-specificity ranging between 0 and 1 where low and high values correspond to broadly expressed and tissue-specific genes, respectively. Gene orthogroups were defined as housekeeping when all the conserved vertebrate genes presented a Tau value ≤ 0.25 (Tau values from [10]; 158 orthogroups in total, see Additional file 2: Table S5). Thus, in Additional file 1: Fig. S11d–f the expression similarities, correlations, and distances are only shown for vertebrate orthogroups.

Characterization of highly and lowly diversified gene orthogroups

For each clade (vertebrates and insects) and each considered feature (sequence and expression similarity), we defined one group of highly and one group of lowly diversified genes by selecting the 500 gene orthogroups with the lowest/highest similarity values (Additional file 2: Table S2). The phenotypes associated with highly diversified, lowly diversified, and all orthogroups are defined as follows. First, we downloaded gene-phenotype associations from Ensembl v105 [54] for human and mouse and from FlyBase ([55] updated in January 2020) for the fruit fly. We then filtered only for the genes present in our gene orthogroups, and we created one vertebrate- and one insect-specific phenotypic annotation by associating the phenotype of the relative human/mouse or fly genes (respectively) to the whole orthogroup (see Supplementary dataset). “Lethal” phenotypes were defined as everything containing “lethal” or “_die_”, “Not lethal” phenotypes corresponded to all other identified phenotypes, while all the orthogroups with no associated phenotype in the relative clade were labeled as NA. We then associated the phenotypic classification from the human/mouse and fly annotations to the vertebrates and insects gene groups, respectively, and plotted the resulting distributions (Fig. 3b). The number of species with duplications for each of these orthogroups (Fig. 3c) corresponded to the number of species in the relative clade with at least one paralog. The Tau values (Fig. 3d) were taken from [10]. The results depicted in Additional file 1: Figs. S6g–i, S7g–i, S8g–i, S12g–i, S13g–i, S14g–i, S15g–i, S16g–i were generated with the same procedure but starting from alternative measures of sequence and expression similarities or from the additional 1:1 orthogroup sets (see Additional file 3: Supplementary note for relative discussion).

GO annotations

Comparative analyses based on functional GO annotations might be biased by the different GO annotation qualities across species. In order to avoid these biases, we generated a unified annotation for all the species under the assumption that orthologous genes are expected to possess common functional traits (analogous to [10]). First, we built GO annotations both for human and fruit fly. We downloaded the GO annotations (GeneID-GO correspondence) from Ensembl v106 [54] for the two species, and we combined the human annotations with those from clueGO v2.5.5 level 5 [56]. Next, to build a human-based GO annotation file, we assigned the GO annotations of each human gene to all the genes from the other species belonging to the same orthogroup whenever the number of human genes within the orthogroup with that GO annotation was ≥ 1/4 of the total human genes in that orthogroup. A similar strategy was implemented to build a fruit fly-based GO annotation file starting from the fruit fly GO annotation. Finally, we selected only the GO categories with a number of genes included between 3 and 1500 for the human-based annotation and 3 and 500 for the fruit fly one. The resulting annotation files were used for all GO analyses.

GO enrichments

The GO enrichments of common highly and lowly diversified gene groups (Fig. 4b, c and Additional file 1: Figs. S6j, S7j, S8j, S12j) were performed using the human-transferred GO annotation from [10] (see above), and all gene orthogroups as a background. Common sets of highly and lowly diversified genes were defined as gene orthogroups included in at least three out of the four highly/lowly diversified groups (vertebrate/insect from the sequence/expression perspective). Up to the top 20 significant (FDR-corrected p-value ≤ 0.05) categories including at least 5 genes and 5% of the initial set are represented for each group in Fig. 4a, b, while all categories are reported in Additional file 2: Table S3. The GO enrichments for the extra 1:1 orthogroup sets (Additional file 1: Figs. S13j, S14j, S15j, S16j) were performed in the same way but using custom GO annotations that should better represent the gene selection in each particular orthogroup set. We generated these annotations by downloading the human GO annotation from Ensembl v106 [54] and transferring the function of the selected human gene to the relative orthogroup (similar to the approach described in [10] and for the GO annotation above).

Gene Set Enrichment Analysis (GSEA)

We computed the deltas between the sequence and expression similarity measures in vertebrates and insects (Additional file 1: Fig. S3a, b), to which we applied a z-score transformation in order to shift the center of each distribution to zero (Additional file 1: Fig. S3d, e). We then used these z-scored deltas and the GO annotation previously described as input for two GSEAs, which were performed using the fgsea function in R (fgsea package) with the following parameters: minSize=50, maxSize=500. We then only filtered for the categories associated with an FDR-corrected p-value ≤ 0.01 and plotted the 30 categories with the highest/lowest normalized enrichment score (NES) in terms of either sequence or expression (Fig. 5a). All results are reported in Additional file 2: Table S4). The genes highlighted in Fig. 5b, c are the ones listed under “leading edges” in the GSEA results. The GSEA depicted in Additional file 1: Figs. S6k, S7k, S8k, S12k, S13k, S14k, S15k, S16k were generated with the same procedure but starting from alternative measures of sequence and expression similarities or from the different 1:1 orthogroup sets and using for each the custom GO annotation described above (see Additional file 3: Supplementary note for the relative discussion).

Characterization of genes with diversification biases

In order to evaluate the over-representation of tissue-specific genes from particular tissues within the orthogroups with sequence or expression-driven specialization biases, we performed separate Fisher’s exact tests for each clade and each tissue. For each test, we ran the fisher.test function from base R with alternative = “greater” on a matrix containing (i) the number of tissue-specific genes from the tested tissue included in the orthogroups with diversification biases, (ii) the total number of tissue-specific genes in the same orthogroups, (iii) the number of tissue-specific genes from the tested tissue in all other orthogroups, and (iv) the total number of tissue-specific genes in all other orthogroups. The threshold for significant p-values was set at 0.05. The results depicted in Additional file 1: Figs. S6l, S7l, S8l, S12l, S13l, S14l, S15l, S16l were generated with the same procedure but starting from alternative measures of sequence and expression similarities or from the different 1:1 orthogroup sets (see Additional file 3: Supplementary note for the relative discussion).

Authors’ contributions

F.M. and M.I. conceived the study. F.M. generated all results and figures. F.M. and M.I. wrote the manuscript.

Funding

This research has received funding from the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation programme (grant agreement 101002275 to MI) and by the Agencia Estatal de Investigación (PID2020-115040GB-I00/ AEI / 10.13039/501100011033 to MI). FM held a FPI fellowship associated with the grant funded by Agencia Estatal de Investigación and Fondo Europeo de Desarrollo Regional grant reference BFU-2017-89201-P/ AEI/FEDER, UE.

Declarations

Ethics approval and consent to participate

Not applicable.

Competing interests

The authors declare no competing interests.