Pleiotropy increases with gene age in six model multicellular eukaryotes

Abstract

Fundamental traits of genes, including function, length, and Guanine-Cytosine (GC) content, all vary with gene age. Pleiotropy, where a single gene affects multiple traits, arises through selection for novel traits and is expected to be removed from the genome through subfunctionalization following duplication events. It is unclear, however, how these opposing forces shape the prevalence of pleiotropy through time. We hypothesized that the prevalence of pleiotropy would be lowest in young genes, peak in middle-aged genes, and then either decrease to a middling level in ancient genes or stay near the middle-aged peak, depending on the balance between exaptation and subfunctionalization. To address this question, we have calculated gene age and pleiotropic status for several model multicellular eukaryotes, including Homo sapiens, Mus musculus, Danio rerio, Drosophila melanogaster, Caenorhabditis elegans, and Arabidopsis thaliana. Gene age was determined by finding the most distantly related species that shared an ortholog using the Open Tree of Life and the Orthologous Matrix Database. Pleiotropic status was determined using both protein–protein interactions (STRINGdb) and associated biological processes (Gene Ontology). We found that middle-aged and ancient genes tend to be more pleiotropic than young genes, and that this relationship holds across all species evaluated and across both modalities of measuring pleiotropy. We also found absolute differences in the degree of pleiotropy based on gene functional class, but only when looking at biological process count. From these results, we propose that there is a fundamental relationship between pleiotropy and gene age, and further study of this relationship may shed light on the mechanism behind the functional changes genes undergo as they age.

Introduction

Pleiotropy is the phenomenon in which a single gene effects multiple traits, and has wide ranging effects on development (Cheverud, 1996), genetic disease (Ittisoponpisan et al., 2017; Sivakumaran et al., 2011), signaling robustness (Guillaume & Otto, 2012; Papakostas et al., 2014), the evolution of new traits (Armbruster et al., 2009; Lenski et al., 2003), and organismal adaptability (Fraïsse et al., 2019; Hämälä et al., 2020; Kinsler et al., 2020). Pleiotropy is expected to arise in the genome as a result of exaptation, the process through which existing genes and genetic architecture are co-opted for use in novel traits. The role of exaptation in promoting pleiotropy has been demonstrated theoretically using computational models of complex trait evolution (Lenski et al., 2003) and empirically in the flowering vine genus Dalechampia (Armbruster et al., 2009), where pollinator attractant traits are co-opted for plant defense and vice versa. Pleiotropy has also been proposed to arise as a result of co-selection on traits, such as the integration between the growth of the cranial vault and facial masticatory apparatus reported in several primates (Cheverud, 1996). These mechanisms for increasing the number of pleiotropic genes and the number of traits that pleiotropic genes affect suggest that the prevalence of pleiotropy should be positively correlated with gene age. Supporting this idea in humans, older genes have elevated protein–protein interactions (PPI), a commonly used metric to determine the pleiotropic status of a gene (Yin et al., 2016).

The prevalence of pleiotropy does not strictly increase over time, as subfunctionalization following gene duplication events decreases the frequency of pleiotropy in the genome (Guillaume & Otto, 2012; Lynch & Wagner, 2008; Schmid & Sánchez-Villagra, 2010). Subfunctionalization in duplicated genes is predicted to arise through a process of complementary degenerative mutations in the duplicated genes, leading to loss or degradation of multiple functions in the gene duplicates (Force et al., 1999) and promoting the maintenance of duplicated genes in the genome (Lynch & Force, 2000). In the case where the parent gene’s functions are completely segregated between the copies with no overlap in function, the overall prevalence of pleiotropy in the genome should be reduced as two genes are now performing the work of a single gene. Aside from subfunctionalization removing pleiotropic genes from the genome, a gene’s future adaptability is limited by the acquisition of novel traits (Fraïsse et al., 2019), placing a soft limit on the number of processes a gene can be involved in. These forces, acting in opposition to the constant growth in pleiotropy from exaptation and co-selection, could produce a wide range of distributions of pleiotropy across the gene age spectrum.

The relationship between the age of a gene and some fundamental characteristics is well documented. Previous work established the essentiality of older genes (Chen et al., 2012), as well as the tendency to have increased Guanine-Cytosine (GC) content and length compared to young genes (Yin et al., 2016). However, we do not have a clear understanding of how the mean number of traits associated with a gene (pleiotropy) or the variance in traits associated with a group of genes changes as genes age. For example, it is unclear if changes in the prevalence of pleiotropy over time are consistent among different metrics of estimating pleiotropy, such as PPI and biological process (BP) count. It is also unclear if the relationship between gene age and pleiotropy depends on the focal organism or if it is generalizable to large swathes of the tree of life. While previous studies have suggested that pleiotropic genes are more essential than their non-pleiotropic counterparts (Ittisoponpisan et al., 2017), this could be related to either the essentiality of old genes or the tendency of pleiotropic proteins to be hubs in gene signaling networks (De Bruyne et al., 2014; Sadhukhan et al., 2021). Furthermore, the exact dynamics of how the prevalence of pleiotropy changes with a gene’s age are unknown; while previous work demonstrated significant differences in PPI between old and young genes, little is known about changes on a more continuous scale (Yin et al., 2016).

We hypothesized that young genes would have the lowest prevalence of pleiotropy, accounting for the limited time they would have had to become involved in other traits. This expectation also aligns with previous work showing that young genes are less essential than old genes (Chen et al., 2012) and that pleiotropic genes tend to be highly essential. We hypothesized that middle-aged genes would be more pleiotropic than young genes, reflecting the increased time for genes to be recruited into other traits via exaptation. For the oldest genes, we expected either that the prevalence of pleiotropy would continue to climb or that it would reach an equilibrium state, meaning the mean prevalence of pleiotropy in all age bins past a certain age would be equal, reflecting the net outcomes of forces adding and removing pleiotropy from the genome.

To address our hypothesis, we calculated the age and pleiotropic status of protein-coding genes from Homo sapiens, Mus musculus, Danio rerio, Drosophila melanogaster, Caenorhabditis elegans, and Arabidopsis thaliana. We selected these organisms in an attempt to cover a wide range of multicellular eukaryotic life, but were limited to model organisms with well-annotated genomes and protein resources. These organisms, while limited in scope compared to the true diversity of the tree of life, allow us to identify potentially generalizable and species-specific trends in the prevalence of pleiotropy. For a given species, a gene’s age was determined by finding the most distantly related common ancestor that shared an ortholog of that gene. Orthologs were collected from the Orthologous Matrix Database (OMAdb) (Altenhoff et al., 2021) and common ancestors were identified on a phylogeny generated from the Open Tree of Life (synthesis 14.8) (Redelings & Holder, 2017) that included 1,900 species from the OMAdb. To evaluate the prevalence of pleiotropy in genes, we used Gene Ontology (GO) (Gene Ontology Consortium, 2021) BP labels for each gene, as well as PPI from the String DB (Szklarczyk et al., 2023). We elected to use both GO BPs and PPI, as these are complementary measures of pleiotropy (He & Zhang, 2006). BPs assess the kinds of distinct actions a protein may carry out, often with different active sites, while PPI measures total interacting partners and does not capture what a protein is doing with those partners or which active sites are involved in the interaction (He & Zhang, 2006). Our primary goal with this study is to provide a high-level but generalizable understanding of the evolution of pleiotropy, laying the groundwork for future exploration of this fundamental genomic phenomenon.

Methods

Because the datasets for each species were retrieved from databases that used similar terms and formatting, we were able to apply the same pipelines for the analysis of each species individually. The exceptions were A. thaliana, which lacked gene duplication data in the Ensembl database, and C. elegans which had significantly fewer duplicated genes detected than expected from the literature. Therefore both species were excluded from the singleton vs. duplicate analysis. With that in mind, the following sections reference the H. sapiens dataset and its analysis but apply to all species unless otherwise noted.

For each species we retrieved the following number of genes: H. sapiens (n = 19,467), M. musculus (n = 21,128), D. rerio (n = 27,897), D. melanogaster (n = 13,659), C. elegans (n = 16,050), and A. thaliana (n = 25,125).

Gene age determination

Orthologs were downloaded from the OMA database in the form of standard OMA groups (Altenhoff et al., 2021; Kaleb et al., 2019), and the full database was trimmed to only include those ortholog groups with an entry from the focal species (e.g., H. sapiens). We elected to use standard OMA groups rather than Hierarchical Orthologous Groups (HOGs) because of the stringent requirement that all members of the group are mutually orthologous. From these H. sapiens ortholog groups, a list of unique species that had at least one ortholog in at least one group was generated. A phylogenetic tree including only these unique species was trimmed from the Open Tree of Life, and each trimmed tree had at least 1,800 species. Because the Open Tree of Life is a purely relational tree, it does not include branch length data, so all branch lengths were set to 1 during the creation of the OMA-specific phylogeny. The distance from the root of the tree to the most recent common ancestor of each species and H. sapiens was calculated, and the age of each ortholog was set to the age of the oldest common ancestor between H. sapiens and the species in which the ortholog was found. For example, if an ortholog was present in several species, with the oldest one being A. thaliana, then the age of the ortholog would be the distance of the last common ancestor between H. sapiens and A. thaliana to the root of the tree. This means that orthologs with a lower distance from the base of the tree are older than those with a greater distance. If an ortholog could not have an age assigned to it for any reason, it was excluded from the analysis. This process removed 12 A. thaliana genes, 7 D. rerio genes, and no genes for C. elegans, D. melanogaster, or M. musculus. A total of 455 genes were removed in H. sapiens, while this is noticeably more than the other species, it still left 98% of the initial human gene set intact.

Age binning

The number of orthologs assigned to any specific age is widely variable, with older age groups, such as those genes present in the last common ancestor of all eukaryotes, comprising thousands of genes, while more modern age groups, like the genes found only in primates, could have tens of genes or no genes at all. To overcome this uneven sampling across ages, orthologs were grouped into age bins with at least 1,000 orthologs. This binning process started in the oldest ages and progressed by collapsing ages into a single bin until the total ortholog count exceeded 1,000; then a new bin would be created and the process would repeat. Due to differences in total gene counts and the specific distributions of genes, species are not guaranteed to have identical bin counts. Figure S1 plots mean BP vs. time using unbinned data as well as bins where the number of genes was greater than or equal to 500 to demonstrate that our results are robust to alternative binning strategies.

Determining the prevalence of pleiotropy

The prevalence of pleiotropy was determined on a per-gene basis using two methods: GO unique BPs and STRING database PPI. These measures were selected because they provide complementary insight into the pleiotropic characteristics of a gene. A gene’s BP count relates to a purely functional view of pleiotropy, where each unique function the gene carries out is counted as a BP, and having a greater BP count indicates that the gene affects more traits (He & Zhang, 2006; Papakostas et al., 2014; Williams et al., 2023). A gene’s PPI count is agnostic to annotated functional traits and focuses instead on the number of other proteins it interacts with, which provides a different kind of proxy for the degree of pleiotropy (He & Zhang, 2006; Papakostas et al., 2014; Williams et al., 2023). BP counts were calculated by retrieving GO data associated with that gene from the OMA database using the OMAdb Python API (Altenhoff et al., 2021). All GO evidence types were used to determine total BP involvement, but multiple entries of the same process were ignored, and each gene was given the most specific (tip) GO term for a process with which it was associated. For GO analysis, we also used terms from the GO Slim, which is a reduced set of GO terms to avoid multiple counts of related functions. Our GO Slim analysis showed very similar results to using the Generic GO, but as it significantly reduced the size of our dataset, we did not include those analyses in the manuscript. PPI counts were determined for each protein by counting the number of interactions in the STRING database that surpassed the 0.4 confidence score threshold. Confidence scores are provided for each interaction in STRING, generated using the available evidence for a PPI, and the chosen threshold indicated medium to high confidence that the given interaction exists (Szklarczyk et al., 2023). Our results are not dependent on this confidence score threshold and were robust to a much more stringent threshold of 0.95. For both BP counts and PPI, larger numbers indicated more pleiotropy. At no point did, we deploy a hard cutoff to determine if a gene was pleiotropic or non-pleiotropic, instead opting to derive distributions of these pleiotropic measures across ages.

Determination of broad function

Using the GO hierarchy of terms, we determined the broad category that each specific GO term was associated with; for example, pyrimidine nucleobase biosynthetic process is a specific (leaf) term that falls under the metabolic process GO term. The broad categories into which we grouped genes were the child terms of the Biological_Process term (GO:0008150), which is the root of the BP ontology. To each protein, we then assigned a association to a broad functional category if one of its functions fell in that category. Ultimately, 16 functional categories were present in each species, but for simplicity, we elected to analyze only the metabolic, cellular, and developmental processes as these were among the most abundant in each species. If, as was often the case, a gene had GO terms associated with two or more of the three processes we chose to analyze, that gene was included in multiple groups. We plotted immune system processes for the M. musculus and H. sapiens datasets because the immune system has been shown to be highly pleiotropic (Sivakumaran et al., 2011) and provided a basis of comparison for the selected processes. The immune system process gene counts in the other species were too low for effective comparison and were thus excluded.

Duplicated gene analysis

Using the Ensembl biomart (Harrison et al., 2024), we identified the total set of duplicated genes for H. sapiens, M. musculus, D. rerio, D. melanogaster, and C. elegans. Arabidopsis thaliana was excluded from this analysis because it was not available in the biomart dataset. Duplicated genes were identified by selecting “Homologues” under the attributes tab within the dataset of interest (e.g., Human Genes) and then obtaining “Human paralogue gene stable ID” and “Human paralogue associated gene name” data within the “Paralogues” subset. The list of genes with paralogs was exported as a .tsv file, and instances where paralogs were present were treated as evidence of a gene duplication event. This method includes ancestral paralogs, as Ensembl does not differentiate between ancestral and lineage-specific paralogs. A total of 65.6% of H. sapiens genes had paralogs, which is in line with previous estimates of paralogs in the human genome (Ryan et al., 2023). The percentage of genes with paralogs for all species is reported in Table S1. Because the number of duplicated genes detected in this manner for C. elegans was ∼2%, which is significantly lower than the 24%–32% of genes that have been reported in previous studies (Cutter et al., 2009; Ma et al., 2024), we elected to conduct this analysis with only H. sapiens, M. musculus, D. rerio, and D. melanogaster.

PubMed article counts

To determine if there was a relationship between BP count (or PPI) and the depth of study associated with a gene, we conducted a PubMed search for the gene names of each of the proteins in the H. sapiens dataset. We searched titles and abstracts of articles in PubMed using the MetaPub Python package and limited our search to returning the top 1,500 articles for each gene.

Results

The prevalence of pleiotropy increases with gene age across eukaryotes

Genes in middle-aged and older age bins had a greater prevalence of pleiotropy, as measured by BP count, than genes in younger age bins (Figure 1). This finding was also seen in the PPI results (Figure S2). In older age bins the distribution of BPs tended to be unimodally distributed around the median value, but younger genes developed a skewed or bimodal distribution of BP count, with a prominent peak near the mean and a second peak around the single-process baseline. These secondary peaks were not observed in the PPI plots. The statistical relationship between gene age and BP count was tested using a one-way ANOVA, and for each species there was a significant relationship between gene age and BP count (Table 1) as well as between gene age and PPI (Table S2).

Figure 1.

Older genes have an elevated level of pleiotropy as measured by biological process (BP) count. The y-axis shows violin plots built on the log₁₀ (NP count) for all genes in each age bin. The x-axis shows age bins for genes, from oldest on the left to the youngest on the right. The boxplots inside of the violins show the median value (white line) along with the 1st and 3rd quartiles. Whiskers show all values that fall within 1.5 times the interquartile range. Lines were added through the median values of each violin to help visualize overall trends. Orthologs that did not have a BP count were excluded from these plots.

Table 1.

ANOVA tables for the relationship between the number of biological processes a gene is associated with and the age of that gene.

Species	Formula	df	Sum. Sq.	Mean Sq.	F	PR (>F)
Homo sapiens	BP∼Age	1	7,723.26	7,723.26	49.55	<2.01e−12
Mus Musculus	BP∼Age	1	2.46e4	2.46e4	163.89	<2e−16
Danio rerio	BP∼Age	1	452	452	18.33	<2e−5
Drosophila melanogaster	BP∼Age	1	2.3e4	2.3e4	521.29	<2e−16
Caenorhabditis elegans	BP∼Age	1	7,079.67	7,079.67	359.06	<2e−16
Arabidopsis thaliana	BP∼Age	1	1.25e4	1.25e4	148.39	<2e−16

While young ages corresponded to less pleiotropy than middle and older ages, the qualitative dynamics of this pattern were relatively species-specific. In H. sapiens and M. musculus, the mean BP rapidly rose as gene age increased and then leveled out in the older ages. In D. rerio, there was more variability in BP accumulation as age increased, with some age bins having lower mean BP counts than the next youngest age bin. Drosophila melanogaster and A. thaliana exhibited stepwise increases in BP count rather than a smooth accumulation of biological BPs. Finally, C. elegans diverged the most from the other organisms, with its youngest age bins having approximately equal mean BP counts, and only the very oldest bins increasing in mean BP count. These patterns only manifested in the BP plots; when looking at PPI, there was a universally smooth trend of constant accumulation of interactions as gene age increased.

These analyses used all evidence types in the GO to determine the number of BPs associated with a gene. Because some electronic evidence types include functions that are inferred from orthologs, there is the chance that genes that are highly orthologous have an inflated number of functions. To determine if this bias altered our results, we ran our analysis using only BPs that were experimentally validated. We found that the significant relationship between pleiotropy and gene age held in H. sapiens, M. musculus, D. melanogaster, and C. elegans, but the relationship was no longer significant in D. rerio and A. thaliana. Genes from these latter species were very sparsely represented in the experimentally validated dataset, and we believe this severe reduction in the number of genes analyzed accounts for the loss of significance. Relatedly, to determine if highly studied proteins had an inflated number of BPs, we conducted an automated literature search of PubMed for each protein in our human dataset. We then plotted BP and PPI against the number of articles for a protein and found no relationship between these measures (Figures S3 and S4).

The prevalence of pleiotropy differs across age and functional groups

The prevalence of pleiotropy was distinct between functional groups (Figures 2 and S5). Metabolic processes almost always had the lowest prevalence of pleiotropy, and in H. sapiens and M. musculus, immune genes showed the highest prevalence of pleiotropy (when measured by BP). For D. rerio, D. melanogaster, C. elegans, and A. thaliana (for which immune processes were not included), the developmental genes had the highest prevalence of pleiotropy when measured by BP. Interestingly, these differences were not consistent between the PPI and BP plots, likely due to the kind of pleiotropy they represent (Figure S5). Affirming our genome-wide findings, no young functional group had a higher prevalence of pleiotropy than the corresponding old functional group in the same organism. For each species, a two-way ANOVA was conducted to evaluate the relationship between age, trait, and BP count. The relationships between both age and BP as well as trait and BP were significant for all species (Table 2). The relationship between age group and PPI was significant for all species, while the relationship between trait and PPI was significant in all but D. rerio (Table S3).

Figure 2.

The prevalence of pleiotropy is dependent on gene function. Plots show genes in the oldest and youngest time bins (labeled Oldest and Youngest). The y-axis shows violin plots built on the log₁₀ (biological process count) for all genes in a given age bin, separated into four functional groups (metabolic, cellular process, developmental, and immune processes). The boxplots inside of the violins show the median value (white line) along with the 1st and 3rd quartiles. Whiskers show all values that fall within 1.5 times the interquartile range. Orthologs that did not have a BP count were excluded from these plots.

Table 2.

Two-way ANOVA tables for the relationship between biological process count, gene age, and the primary trait associated with a gene.

Species	Factor	df	Sum. Sq.	Mean Sq.	F	PR (>F)
Homo sapiens	Age	1	338.05	338.05	453.56	<2e−16
	Traits	4	166.01	41.5	55.56	<2e−16
Mus musculus	Age	1	359.48	359.48	453.88	<2e-16
	Traits	4	174.26	43.56	55.01	<2e−16
Danio rerio	Age	1	76.46	76.46	140.26	<2e−16
	Traits	3	192.18	64.06	117.52	<2e−16
Drosophila melanogaster	Age	1	503.41	503.41	922.33	<2e−16
	Traits	3	150.33	50.11	91.81	<2e−16
Caenorhabditis elegans	Age	1	87.75	87.75	142.13	<2e−16
	Traits	3	182.00	60.67	98.26	<2e−16
Arabidopsis thaliana	Age	1	283.68	283.68	549.60	<2e−16
	Traits	3	292.62	97.54	188.63	<2e−16

We calculated a bootstrapped mean BP and PPI count (from 50 replicates) for each functional group based on the group with the smallest number of orthologs per species. We then determined 95% confidence intervals around these bootstrapped means and found that these means were distinct based on their disjoint confidence intervals (Figures S6 and S7), suggesting that the observed differences in distributions were likely not due to sample size.

Gene duplication events do not decrease the prevalence of pleiotropy

Genes without paralogs generally had fewer BPs associated with them than genes with paralogs. Several individual age bins across the species showed duplicated genes with a prevalence of pleiotropy that was less than or equal to the nonduplicated genes, but the overall trend supports increased pleiotropy in genes with paralogs (Figures 3 and S8). The exact trends between species varied, with some species showing consistent differences between duplicated and singleton genes throughout the entire age range (H. sapiens, D. rerio, and D. melanogaster) and others showing an inconsistent pattern (M. musculus). For each species, a two-way ANOVA was conducted to evaluate the extent to which gene age and duplication status explained BP count. The relationships between age and BP, duplication status and BP (Table 3), and gene age and PPI (Table S4) were significant for all species, while the relationship between duplication status and PPI was significant in all species except D. melanogaster (Table S4).

Figure 3.

Genes with paralogs are more pleiotropic than genes without paralogs. The y-axis shows violin plots built on the log₁₀ (biological process count) for all genes in a given age bin. The x-axis shows age bins for genes, from oldest on the left to the youngest on the right. Within each age bin, genes without paralogs are plotted first and genes with paralogs are plotted second. The boxplots inside of the violins show the median value (white line) along with the 1st and 3rd quartiles. Lines were added through the median values of each violin to help visualize overall trends. Whiskers show all values that fall within 1.5 times the interquartile range. Orthologs that did not have a BP count were excluded from these plots.

Table 3.

Two-way ANOVA tables for the relationship between biological process count, gene age, and gene duplication.

Species	Factor	df	Sum. Sq.	Mean Sq.	F	PR (>F)
Homo sapiens	Age	1	8,217.78	8,217.78	53.16	<2e−16
	Duplication	1	2.21e4	2.21e4	142.98	<3.21e−13
Mus Musculus	Age	1	1.64e4	1.64e4	110.97	<2e−16
	Duplication	1	4.96e4	4.96e4	334.96	<2e–16
Danio rerio	Age	1	509.25	509.25	21.00	<4.62e−6
	Duplication	1	7,658.53	7,658.53	315.83	<2e−16
Drosophila melanogaster	Age	1	2.23e4	2.23e4	517.50	<2e−16
	Duplication	1	597.49	597.49	13.85	1.99e−4

Discussion

In this study, we have identified a potentially generalizable relationship between gene age and the prevalence of pleiotropy. Our results suggest that young genes accumulate functions as they age, eventually trending toward a plateau that could be indicative of a carrying capacity of function rather than a balance between gain and loss of function rates. We have also shown that genes belonging to metabolic, developmental, cellular, and immune functional groups differ in their prevalence of pleiotropy, and these differences are preserved within age groups. These results lay the groundwork to better understand the evolution and maintenance of pleiotropy in multicellular eukaryotes.

While previous studies suggest that both PPI and BP act as complementary proxies of pleiotropy (Williams et al., 2023), there are differences between the two measures that may explain the qualitative differences observed. PPI measures the number of interacting partners a protein has and is agnostic to the action of the focal protein on those partners. BP describes distinct actions a protein carries out, but any one of those processes could have many PPIs associated with it. In the most extreme cases, as with the human CAPN13 protein, a protein can be associated with a single BP (proteolysis) and have more than 750 annotated PPIs (Sorimachi et al., 2011). We could therefore interpret our results as indicating that gaining more interactions is easier than developing novel functionality, which matches expectations of protein evolution (i.e., micro- and macrotransitions) (Jayaraman et al., 2022). This explanation concurs with the observed results, where increases in PPI are largely consistent between ages (Figure S2), while increases in BP are more abrupt and taper off significantly with age (Figure 1).

Immune system genes tend to have a high prevalence of pleiotropy (Sivakumaran et al., 2011; Williams et al., 2023). Our work expands on these findings by directly comparing the prevalence of pleiotropy in genes across a set of traits, enabling us to determine systemic variations in the prevalence of pleiotropy. We have shown that metabolic genes tend to have low levels of pleiotropy, as measured by PPI and BP, compared to cellular process and developmental genes. These findings are largely stable across the species we investigated (Figures 2 and S5). It is unclear if these relationships are due to differences in the ability of each trait to tolerate pleiotropy or if some traits are simply predisposed to become pleiotropic through evolutionary processes. We do not expect that the differences observed are strictly due to the importance (in a fitness sense) of the traits in question, as development, metabolism, and cellular processes are all fundamental to the survival and reproduction of multicellular organisms. The difference could instead be attributed to the nature of the proteins that are associated with each process. For example, signaling proteins tend to be more pleiotropic than other protein classes (Williams et al., 2023), potentially indicating that the abundance of signaling proteins associated with immunity and development are partially responsible for inflating the prevalence of pleiotropy in genes associated with these traits.

Despite our expectation that gene duplications would reduce the prevalence of pleiotropy, the majority of duplicated genes maintain or increase their prevalence of pleiotropy compared to age-matched singleton genes. A review of the functional changes associated with duplicated genes has found that complete subfunctionalization following gene duplication is likely to be rare (Janiak et al., 2019; Kuzmin et al., 2022). Instead, several processes may act together to maintain pleiotropy in these genes, including partial subfunctionalization, where both genes keep at least some activity for each function; neofunctionalization following subfunctionalization; dosage amplification, where the excess gene products associated with multiple copies are beneficial; and backup compensation, where having a second copy of a critical gene safeguards against loss of function (Kuzmin et al., 2022). When paralogs form complexes, selective pressure can lead to correlated mutations, providing another avenue for paralogs to actively increase pleiotropy, as one member of a complex acquiring a novel function can force others to acquire that same function (Marchant et al., 2019). Critically, our measures of pleiotropy cannot assess how well duplicated genes carry out their shared functions. Thus, a gene that has lost much but not all of its functional capacity following a duplication event (partial subfunctionalization) would still be pleiotropic in our analyses (Janiak et al., 2019).

When looking for examples of this kind of interaction in our data, we find that many duplicates, such as myogenic differentiation 1 (MYOD1) and myogenic factor 5 (MYF5) in mice, share a significant number of BPs despite these genes having distinct roles in specialization and differentiation (Conerly et al., 2016). Critically, even though MYF5 does not induce robust transcription while MYOD1 does, it is still associated with the BP “regulation of DNA-templated transcription.” We are then left to believe that duplication events play a relatively small role in the reduction of pleiotropy in the genome, supporting the hypothesis that the accumulation of pleiotropic functions is instead limited by reduced evolvability as genes acquire novel functions (Fraïsse et al., 2019). Duplicated genes present the same general trend as the singleton genes of increasing pleiotropy over time (Figures 3 and S8), suggesting that our results in Figures 1, 2, S2, and S5 are not qualitatively biased by combining duplicated and singleton genes. Seeking a method to address this potential issue using available GO data, we plotted age vs. experimentally validated GO terms in Figure S9. Despite the severe and uneven reduction in available data across our species, we found that four (H. sapiens, M. musculus, D. melanogaster, and C. elegans) still exhibited increased pleiotropy with gene age, while A. thaliana and D. rerio no longer showed any change in the prevalence of pleiotropy across time.

Our results suggest that gene duplication events play a limited role in maintaining pleiotropic genes in the genome. The accumulation of new functions is thought to be increasingly constrained by the number of current functions (Fraïsse et al., 2019), but other forces might also be at play. For instance, life history trade-offs due to sexual selection have been implicated in the maintenance of antagonistic pleiotropy (Johnston et al., 2013), as has a “selection shadow” when a gene’s function is highly beneficial in early life but detrimental postreproduction (Abdellatif et al., 2023). Furthermore, pleiotropic genes are often essential to organismal fitness (Ittisoponpisan et al., 2017), and this may contribute to their maintenance, as mutations that are necessary to separate the functions of the gene may degrade the essential function(s) of the gene.

The methods used to determine orthology in this study may present some limitations on the interpretation of results. After all, rapidly evolving genes may be misidentified as young genes because rapid sequence divergence “hides” older orthologs. Paired with the observation that pleiotropy slows the rate of evolution (Williams et al., 2023), our methods might bias the identification of highly pleiotropic genes as being very old. Fortunately, our data suggest that not all genes in the oldest age bins are highly pleiotropic, with many having only one or two processes, and not all genes in the youngest age bins have low levels of pleiotropy, as many have dozens of associated BPs. A second point in support of a true effect of gene age is that if the results were an artifact of orthology methods, we would expect rapidly evolving genes to be constrained to the very youngest age bin, but we see that the second and third youngest age bins of many species also show low pleiotropic interactions compared to the middle-aged and older genes.

This work highlights the dynamism of pleiotropy and reveals a previously poorly understood link between a gene’s pleiotropic status and its age. This relationship holds across six distantly related model organisms, suggesting that it could be generalizable among multicellular life. Further work could expand these findings to the single-celled eukaryotes and other domains of life to examine their generality across the tree of life. It is unlikely that the variance in the prevalence of pleiotropy observed between traits is explained by the commonality of genes in the oldest age groups among species, as we observed similar trends amongst the more species-specific young genes. However, it would be interesting to use the pseudo-replication of the same ortholog present in multiple species to study the accumulation of functions. For example, does the same ortholog have the same functions across each species? If not, are there some species where functions are similar while others have diverged? Such analysis could be expanded to conduct maximum likelihood ancestral state reconstructions to determine if the initial functions of a gene bias the other functions it may evolve over time. We also observed that the two self-fertilizing species seem to accumulate pleiotropic functions in a novel way compared to the organisms that sexually reproduce. Genes in self-fertilizing organisms may be less efficiently separated by recombination, reducing co-selection pressure and altering one of the primary mechanisms that gives rise to pleiotropy (Cheverud, 1996).

Our work suggests that a previous observation identifying immune genes as disproportionately pleiotropic in humans (Sivakumaran et al., 2011) may hold more broadly across species and gene age groups. This raises profound questions about the nature of genomic organization and function. For example, is the prevalence of pleiotropy dictated by the importance (in a fitness-related manner) of the trait, or is it intrinsic to the protein type and cellular localization that are necessary for the trait to function? Future investigation into mechanisms at the gene or cellular level could provide fundamental insight into the maintenance of pleiotropy despite the potential for constraining rapid adaptation.

Data and code availability

The data and code used to generate these results are available on Dryad at https://doi.org/10.5061/dryad.m63xsj4fh. Code used in the analysis and generation is also available on GitHub at https://github.com/Reese-Martin/Gene_Age_Project.

Author contributions

R.M. and A.T.T. conceived the project. A.T.T. provided funding. R.M. and A.T.T. designed the analyses, and R.M. conducted them. R.M. and A.T.T. wrote the manuscript.

Funding

This work was supported by the National Institute of General Medical Sciences at the National Institutes of Health (grant number R35GM138007 to A.T.T.).

Conflict of interest

The authors declare no conflicts of interest.