Humanization of yeast genes with multiple human orthologs reveals functional divergence between paralogs

Abstract

Despite over a billion years of evolutionary divergence, several thousand human genes possess clearly identifiable orthologs in yeast, and many have undergone lineage-specific duplications in one or both lineages. These duplicated genes may have been free to diverge in function since their expansion, and it is unclear how or at what rate ancestral functions are retained or partitioned among co-orthologs between species and within gene families. Thus, in order to investigate how ancestral functions are retained or lost post-duplication, we systematically replaced hundreds of essential yeast genes with their human orthologs from gene families that have undergone lineage-specific duplications, including those with single duplications (1 yeast gene to 2 human genes, 1:2) or higher-order expansions (1:>2) in the human lineage. We observe a variable pattern of replaceability across different ortholog classes, with an obvious trend toward differential replaceability inside gene families, and rarely observe replaceability by all members of a family. We quantify the ability of various properties of the orthologs to predict replaceability, showing that in the case of 1:2 orthologs, replaceability is predicted largely by the divergence and tissue-specific expression of the human co-orthologs, i.e., the human proteins that are less diverged from their yeast counterpart and more ubiquitously expressed across human tissues more often replace their single yeast ortholog. These trends were consistent with in silico simulations demonstrating that when only one ortholog can replace its corresponding yeast equivalent, it tends to be the least diverged of the pair. Replaceability of yeast genes having more than 2 human co-orthologs was marked by retention of orthologous interactions in functional or protein networks as well as by more ancestral subcellular localization. Overall, we performed >400 human gene replaceability assays, revealing 50 new human-yeast complementation pairs, thus opening up avenues to further functionally characterize these human genes in a simplified organismal context.

Fig 1. Systematic functional replacement of essential yeast genes with multiple human co-orthologs.

(A) We identified 994 human genes that are orthologs of 709 essential yeast genes. Of these ortholog pairs, we had previously obtained results for 424 pairs with no duplications in either yeast or human lineage (i.e., with 1:1 orthology) [5]. In this study, we tested the remaining set of essential yeast genes that have acquired lineage-specific duplications, classifying them as 1:M (1 yeast to 2 or more human co-orthologs) or M:M (≥2 yeast to ≥2 human co-orthologs) or M:1 (≥2 yeast to 1 human ortholog). There are 140 essential yeast genes with more than one human ortholog, representing 378 ortholog pairs to be tested. In the case of the 1:M category, we obtained 308 informative assays out of 378 testable pairs, whereas in the case of the M:M or M:1 set, we had 29 informative assays out of 90 testable pairs. Replaceability assays were performed in both the hetKO collection and the temperature-sensitive haploid yeast collection. (B) Representative assays performed in yeast hetKO strains for 1:2 (top) and 1:>2 (bottom) are shown. HetKO yeast strains expressing human genes were sporulated, and the sporulation mix was spotted on Magic Marker medium (see Materials and methods) with (yeast gene absent) or without (yeast gene present) G418. Assays were performed with empty vector control (human gene absent) or yeast expression vectors carrying a human cDNA (human gene expression). In the 1:2 class, three different outcomes of human gene replaceability in yeast were obtained. Top panel: both human co-orthologs (Hs-PYCR1 and Hs-PYCR2) can replace their yeast equivalent (Sc-PRO3). Middle panel: one of the two human co-orthologs (Hs-PSMA7 but not Hs-PSMA8) can replace its yeast equivalent (Sc-PRE6). Bottom panel: neither of the two human co-orthologs (Hs-CSTF2 or Hs-CSTF2T) can replace its yeast equivalent (Sc-RNA15). In the 1:>2 class, an example of all human co-orthologs (Hs-KDELR1, two variants of Hs-KDELR2, and Hs-KDELR3) replacing their yeast gene counterpart Sc-ERD2 equally well is shown. The yeast growth assays for these replaceable human genes are shown on the right for each. Haploid yeast gene deletion strains carrying plasmids expressing functionally replacing human genes (colored solid lines) generally exhibit comparable growth rates to the wild-type parental yeast strain BY4741 (black dotted lines). Plotted growth curves display the mean of triplicate growth experiments. 1:M, one-to-many; CSTF2, Cleavage stimulating factor subunit 2; CSTF2T, Cleavage stimulating factor subunit 2 tau subunit; ERD2, ER lumen protein-retaining receptor, endoplasmic reticulum retention defective 2; hetKO, heterozygous diploid knockout; Hs, H. sapiens;; HsA, H. sapiens co-ortholog A; KDELR1, ER lumen protein-retaining receptor 1; KDELR2, ER lumen protein-retaining receptor 2; KDELR3, ER lumen protein-retaining receptor 3; M:1, many-to-one; M:M, many-to-many; OD, optical density; PRE6, Proteasome subunit alpha type-4; PRO3, Delta 1-pyrroline-5-carboxylate reductase; PSMA7, Proteasome subunit alpha type-7; PSMA8, Proteasome subunit alpha type-8; PYCR1, Pyrroline-5-carboxylate reductase-1; PYCR2, Pyrroline-5-carboxylate reductase-2; Sc, S. cerevisiae; ScA, S. cerevisiae ortholog A.

Fig 2. Distribution of replaceability across orthology classes.

Only rarely did all human co-orthologs in one orthogroup replace. Rather, a family of human proteins typically had one or a few replaceable members or none at all. (A) Previously, systematic replacement of essential yeast genes with 1:1 orthologs (1 yeast to 1 human) demonstrated nearly 50% replaceability of essential yeast genes [5]. Here, we tested the replaceability of essential yeast genes with their human counterparts that have acquired lineage-specific duplications in either yeast or human lineage. Of the 308 informative assays obtained in the 1:M class (1 yeast to 2 or more human co-orthologs), 74 human genes replaced their yeast equivalents, whereas 234 did not. Of the 29 informative assays obtained in the M:1 and M:M class (≥2 yeast to 1 or more human co-orthologs), three human genes replaced their yeast equivalents, whereas 26 did not. (B) Combining our previous replaceability assays [5] with the assays done in this study, we have identified 248 essential yeast genes that are functionally replaceable by their human counterparts and 355 that are not. From the perspective of human proteins, 277 replace their yeast versions, whereas 477 do not. Summary of all the humanized yeast assays performed thus far. (C) Nearly half of the essential yeast genes belonging to the 1:1 orthology class were replaceable by their human equivalents (yellow). The distribution of essential yeast genes replaced by at least one human ortholog in the 1:M (both 1:2 and 1:>2 combined) closely matched the 1:1 results. These yeast genes were rarely replaced by all human co-orthologs in an orthogroup (yellow), with the majority of replaceability falling in the differentially replaceable set (green). M:M yeast orthologs were rarely replaceable by any human ortholog, with only three (of 29 tested) human genes replacing three separate yeast orthologs. Nonreplaceable genes are indicated in blue. * indicates data from [5]. 1:M, one-to-many; HsA, H. sapiens co-ortholog A; HsA′, H. sapiens co-ortholog A′; M:1, many-to-one; M:M, many-to-many; ScA, S. cerevisiae ortholog A.

Fig 3. Replaceability of 1:2 orthologs is explained largely by relative divergence of human co-orthologs.

(A) AUCs for the top 30 predictive features for median-collapsed 1:2 orthogroups are shown. The top two predictive features (HsOrthoScore and HsOrthoRank) for this class indicate that replaceability is driven largely by the nearness of the replacing human co-ortholog to the yeast gene relative to a nonreplacing co-ortholog (i.e., most 1:2 orthogroups have only one replacing human co-ortholog, and it is almost always the least-diverged one). (B) The observed trend is even more strongly demonstrated when analysis is restricted to the specific set of 1:2 orthogroups that display differential replaceability (i.e., one human co-ortholog replaces, and the other does not), with AUCs nearing 0.9. The extent of tissue-specific expression also becomes significantly predictive in this set, indicating that human co-orthologs that are more broadly expressed are more likely to replace than their more tissue-specifically expressed co-ortholog. The top 12 features are plotted. (C) More-diverged human co-orthologs in a 1:2 pair do replace in several cases (mostly those in which both co-orthologs replace). When restricting analysis to this set, it is apparent that the most predictive property is sequence similarity to the yeast ortholog, along with proteins that are translated efficiently and are typically in high abundance. Black overlapping bars indicate mean, and error bars indicate standard deviation for 1,000 shuffled AUC calculations for each feature. AUCs were calculated for n = 208 1:2 ortholog pairs, collapsed to 117 and 46 median metaorthologs ([A] and [B], respectively) depending on feature availability. AUCs for MDO assays in (C) were calculated for up to 104 MDO pairs (Materials and methods, S3 Table). The top 12 features are plotted. For full source data, see S1 Data. AA, amino acid; AUC, area under the receiver operating characteristic curve; CAI, codon adaptation index; CBI, codon bias index; Ens74, Ensembl genes version 74; FOP, frequency of optimal codons; Hs, H. sapiens; HsU, H. sapiens UniProt; LLS, log likelihood score; MDO, more-diverged ortholog; RPF, ribosome protected footprints; Sc, S. cerevisiae; ScU, S. cerevisiae UniProt.

Fig 4. Replaceability is explained by relative divergence of 1:2 human co-orthologs from each other and their yeast ortholog.

(A) The average ortholog score of the more-diverged (lighter color) 1:2 co-ortholog is more similar to the less-diverged co-ortholog for co-ortholog pairs that both replace than pairs in which only one or neither of the co-orthologs replace. (B) The average percent amino acid identity for 1:2 co-orthologs is not significantly different between the “all replace” (yellow) class and the “one replaces” (green) class, whereas the “none replace” (blue) class is slightly, but significantly, lower than either. Error bars indicate standard deviation. Raw data for (A) and (B) are available in S2 Data. (C–E) Phylogenetic models depicting gene trees for generic 1:2 orthogroups in the various replaceability classes. In (C), the human co-orthologs belong to the “both replace” class and are thus less diverged from each other but are on average similarly diverged from the yeast ortholog as the “one replaces” class (D), in which the co-ortholog more closely related to the yeast gene is the one that replaces. In (E), the “none-replace” co-orthologs are on average more diverged from the yeast gene. Hs, H. sapiens; HsA, H. sapiens co-ortholog A; HsA′, H. sapiens co-ortholog A′; NS, not significant; ScA, S. cerevisiae ortholog A.

Fig 5. Replaceable 1:>2 human co-orthologs retain orthologous interaction partners and are more central in interaction networks.

(A) AUCs of the top 30 predictive features for median-collapsed informative 1:>2 co-ortholog pairs. The top two most significant features demonstrate the importance of network context in retaining ancestral functions. Specifically, human co-orthologs in highly expanded gene families that have retained a higher fraction of orthologous protein interaction partners with their yeast ortholog (FractionOrthologPartners) are more likely to replace, as well as those that maintain higher centrality in functional interaction networks (betweenness). (B) Similar to the 1:2 case, we further restricted our analysis to a subset of median-collapsed 1:>2 orthogroups that had both replaceable and nonreplaceable human co-orthologs. In this set, the subcellular localization of the human proteins appears to be predictive, in that more broadly localized co-orthologs are more likely to replace than their more organellar-specific co-orthologs. Although this AUC is not significant (indicated by being just less than 2 standard deviations off the mean), it is an obvious trend and does overtake fraction of orthologous partners as the highest performing AUC for this set. (C) Phylogenetic model of a generic 1:M orthogroup showing that the replaceable human co-ortholog has retained more orthologous interactions in a network and is localized in a similar manner to the yeast ortholog. For AUC bar plots, black overlapping bars indicate mean, and error bars indicate standard deviation for 1,000 shuffled AUC calculations for each feature. AUCs were calculated for n = 170 1:>2 ortholog pairs, collapsed to 49 and 26 median metaorthologs ([A] and [B], respectively) depending on feature availability (Materials and methods, S3 Table). The top 12 features are plotted in (B). For full source data, see S1 Data. 1:M, one-to-many; AA, amino acid; AUC, area under the receiver operating characteristic curve; CAI, codon adaptation index; CBI, codon bias index; Ens74, Ensembl genes version 74; FOP, frequency of optimal codons; Hs, H. sapiens; HsA, H. sapiens co-ortholog A; HsA′, H. sapiens co-ortholog A′; HsA′′, H. sapiens co-ortholog A′′; HsU, H. sapiens UniProt; LLS, log likelihood score; RPF, ribosome protected footprints; ScA, S. cerevisiae ortholog A; ScU, S. cerevisiae UniProt.

Fig 6. Simulated protein evolution suggests that diverged duplicates are less likely to bind their ancestral interaction partner.

(A) Inset: Five different types of selection scenarios were considered for a heterodimeric protein complex, considering the effects of amino acid substitutions using the Rosetta molecular modeling platform. (1) AB and AB′ (bind both). (2) AB or AB′, and only the most stable binding interface is considered (bind max). (3) AB, but AB′ was not enforced (bind B). (4) AB, and AB′ is selected against (bind B and not B′). (5) Neither AB nor AB′ is selected for (no bind). Percent of simulations in which an evolved B subunit has the ability to bind the ancestor to A. Divergence is measured by the amount B has diverged from the ancestor of B. (B) Percent of simulations in which an evolved B′ subunit has the ability to bind the ancestor to A. Divergence is measured by the amount B′ has diverged from the ancestor of B. (C) Percent divergence of duplicates when only B or B′ is able to bind the ancestor A. Lighter hues denote a duplicate that is able to bind the ancestor of A, and darker hues denote the nonbinding duplicate. Data and scripts for these figures are available at the following link: https://github.com/a-teufel/Laurent_etal_2020. Figure adapted from [35].