Data sourcesWe have extracted all proteins from the PDB, which are either monomers or homomers; their structure contains at least 80% of their UniProt sequence and has a resolution better than 3 Å (see “Methods” for details). Proteins that are also part of heteromer PDB structures were not included. Next, the sequences were filtered for enzymatic activity, and only the ones with a known Enzyme Commission (EC) number or for which an EC number could be reliably predicted were kept in the dataset (see “Methods”). This resulted in a list of 11968 enzymes (Supplementary Data 1), of which 8248 are homomers, and 3720 are monomers. Analyses of surface hydrophobicity, absence of transmembrane helices, and known heteromer complexes indicate that the quality of the dataset is high (Supplementary notes and Supplementary Figs. 1 and 2). Cellular location was determined with DeepLoc231 for Eukaryotes and PSORTb32 for Prokaryotes, resulting in 7562 cytoplasmic, 894 extracellular, and 1726 other proteins. As the cellular compartments and extracellular environments of Prokaryotes and Eukaryotes are frequently very different, just as their protein secretion/transport mechanisms, wherever it was possible, we analysed Prokaryote and Eukaryote proteins separately (8228 and 3740 proteins, respectively). To remove redundancies in Figs. 2 and 3, we clustered the proteins with MMseqs233 at a 30% sequence similarity cutoff. This resulted in 2435 prokaryotic and 1528 eukaryotic clusters for which the cellular location could be determined, and the cluster centroids were used in the downstream analyses.Fig. 3: Extracellular enzymes are the most frequent among hydrolases.A, B Monomers are most common in hydrolases (EC = 3), while in other enzyme classes, homomers dominate, and their homomer/monomer ratio is between 2–3 in prokaryotes and 1.5–2 in eukaryotes. C, D The frequency of extracellular enzymes is comparably high among the monomers of hydrolases (EC = 3) and lyases (EC = 4), both in prokaryotes and eukaryotes. In absolute terms, however, extracellular hydrolases are much more abundant in the PDB: in total, there are 1521 hydrolase clusters in the dataset, while only 530 lyase clusters and the frequency of monomers is also much lower in lyases. E Graph of Molecular Function GO terms, the terms that are significantly enriched (hypergeometric test) both in prokaryotes and eukaryotes are highlighted with red. The Bonferroni method was used to correct for multiple testing. A diverse array of hydrolase activities, but also carbohydrate binding, toxin activity, or lyase activity on polysaccharides are enriched in both groups. (See Supplementary Data 1 for the full lists of Molecular Function, and Biological Process GO terms.) On panels (A–D), p-values were calculated with tests of proportions and were corrected with the Benjamini-Hochberg method; bars indicate the fraction of quaternary structure in the set, and whiskers represent 95% CI.Extracellular enzymes are predominantly monomersThe analysis of prokaryotic and eukaryotic proteins indicates that despite the huge differences in their biology and composition of their extracellular/intracellular space, enzymes in the two groups show a largely similar pattern: in the cytoplasm, homomers dominate (70% of the clusters, Fig. 2B), while in the extracellular space monomers (70% of the clusters, Fig. 2B), and proteins of other cellular components have intermediate frequencies of homomers and monomers (Fig. 2B). The differences in the frequencies also result in qualitative differences in homomer interfaces: in homodimers, the fraction of the protein surface that is buried in interfaces is much smaller in the case of extracellular proteins than in cytoplasmic ones (Fig. 2C), and in extracellular proteins, the interface is also much less conserved compared to the protein surface (Fig. 2D–F). Conservation was measured as the difference between the ConSurf scores34 of the interface and surface residues. Homomers with small interfaces, particularly below 1000 Å2, are likely to be enriched in quaternary structure (QS) errors6,35 (i.e., can have non-biological, crystallographic interfaces), thus the real frequency of monomers in extracellular proteins is probably even higher, and our estimates are conservative. When only homomers with 1000+ Å2 interfaces are included, in the cytoplasm, 30.2% of the clusters have interfaces that are not significantly more conserved than the solvent-accessible surface (Fig. 2D) while in the extracellular space, 63.0% (Fig. 2F), which is in good agreement with our recent findings showing that in ~35% of homodimers, multimerization is unlikely to contribute to the biochemical function6.Taken together, these findings are in agreement with the hypothesis that in the extracellular space, oligomerisation is either difficult or deleterious (Fig. 2A), and that results in the evolution and dominance of monomer enzymes. Since the pattern is essentially similar in eukaryotes and prokaryotes, it is unlikely to be caused by the limitations of the protein secretion mechanisms of prokaryotes, as eukaryotes can secrete large, assembled complexes. We add that in the final stages of revisions we found out that Monod has already noticed that extracellular proteins are usually monomers36 and hypothesised that it is due to their stabilisation by disulfide bridges. However, this explanation can be ruled out due to the much higher frequencies of disulfide bridges in eukaryotes than in prokaryotes, and the fact that quaternary structure varies in enzymes with the same folds and function.Extracellular enzymes are most common in hydrolases and lyasesNext, we examined whether monomers and extracellular enzymes are enriched in particular enzyme classes. We found that monomers are most abundant among hydrolases (EC = 3), where their frequency is comparable to the frequency of homomers (Fig. 3A, B), while in the case of other enzyme classes, the ratio of homomers to monomers is much higher, ~2–3 in prokaryotes and ~1.5–2 in eukaryotes (Fig. 3A, B). The frequency of extracellular enzymes is highest in the monomers of hydrolases and lyases (EC = 3 and 4, Fig. 3C, D), and the difference between homomers and monomers remains highly significant within these enzyme classes, both in Eukaryotes and Prokaryotes, suggesting that the majority of extracellular enzymes in the PDB are involved in the degradation of macromolecules (Fig. 3C, D).In addition to their distribution in enzyme classes, we also examined the enrichment of Gene Ontology terms in extracellular proteins, compared to the full dataset (Fig. 3E and Supplementary Data 2). Similarly to EC numbers, hydrolase activity is highly enriched among molecular function terms, both in Eukaryotes (p = 2.5e-144) and Prokaryotes (p = 2.9e-47, Fig. 3E and Supplementary Data 2), and all major hydrolase forms are enriched in the dataset (hydrolyses acting on glycosyl bonds, peptidases, lipases, serine hydrolases; Fig. 3E and Supplementary Data 2). The “biological process” terms are less comparable in eukaryotes and prokaryotes due to their different complexity, but in both groups “catabolic process” and its derived terms are the most frequently enriched (see Supplementary Data 2).The above results indicate that hydrolases are the most common enzymes in the extracellular space, however, hydrolases also represent 75% of industrially relevant enzymes37, which may result in research biases in the composition of PDB. Thus, we examined whether excluding them from the data changes the pattern, i.e., whether the enrichment of monomers is a hydrolase-specific phenomenon or a more general trend. We found that their exclusion does not change qualitatively the patterns we observe (Supplementary Fig. 3), although significances are affected, due to the much lower number of clusters (e.g., in eukaryotes, the total number of extracellular clusters drops from 214 to 58, while in prokaryotes to 25, from 105).The evolution of interfaces follows a ratchet-like patternNext, we examined whether the evolution of homomers from monomers (or vice versa) shows any biases, i.e., whether interfaces are more likely to be gained than lost. Using the full set of 12 k eukaryotic and prokaryotic sequences, we identified orthogroups within them using the eggNOG-mapper tool38 (see “Methods”). Altogether, 311 orthogroups had ten or more sequences, and these were used in the downstream analyses. The proteins were aligned, and their rooted phylogenetic trees were used to estimate the evolution of oligomeric status along the phylogeny (see “Methods”), i.e., the ancestral probabilities of being a monomer or homomer for each node (Fig. 4A, see pie charts at each node). Most orthogroups have ancient proteins: 243 have proteins from more than one domain and 92 from all three domains (Bacteria, Archaea, Eukaryota). Ancestral probabilities were estimated in two ways: a binary, and the more accurate multi-state method. In the binary case, proteins were assigned to one of two states: monomers or homomers, irrespective of the number of their subunits. In the multi-state case, the number of subunits (monomer, dimer, tetramer, octamer, etc.) was used as the discrete variable for ancestral character estimation, and we also took into account the homology of interfaces: complexes with similar number of subunits but non-homologous interfaces were treated as separate states in the trees. The estimation of ancestral state was performed with maximum likelihood (ML, Figs. 4 and 5) and also with maximum parsimony (MP, Supplementary Figs. 4 and 5), which show similar results. Trees with only a single quaternary structure type (e.g., all proteins being homomers) were included, but trees where the highest ancestral probability at the root was lower than 51% were excluded from the analyses (31 binary and 57 multi-state trees with ML, and 37 binary and 78 multi-state trees with MP).Fig. 4: The reversion of homomers to monomers is less likely than the evolution of homomers from monomers.A Example phylogenetic trees illustrating ancestral states, with a monomer and homomer root. (Note that the topology of the trees is identical.) The quaternary structure of the proteins at the leaves is indicated with squares, pie charts indicate the probability of being a homomer or monomer at each node. B The probability of changing the quaternary structure (QS) type compared to the root is significantly higher in the trees with monomer roots than in the trees with homomer roots. C Similarly to the binary QS-type, the frequency of subunit losses is significantly lower in trees with homomer roots than the frequency of subunit gains in trees with monomer roots. D The interfaces of homomers that evolved from monomers are much smaller than the interfaces in the trees with a homomer root. E In the trees with homomer roots, the frequency of subunit gains and losses is similar. F The structural similarity of homomers is high within the orthogroups, with TM-scores being above 0.7–0.8 for most orthogroups. However, the average interface overlap is much higher in the orthogroups with a homomer root than in the orthogroups with a monomer root. G Interface overlap also depends on the average size of the interfaces – below 1200 Å2, it drops, suggesting that some of the small interfaces are crystallographic artefacts. The low interface overlap (< 50%) above 1200 Å2 in orthogroups with monomer root indicates that new, independent interfaces evolve much more frequently in this group than in trees with homomer root. On panels (B–D) p-values were calculated with Wilcoxon rank-sum tests. Boxplots display the median, 25–75% interquartile range (IQR), and 1.5 * interquartile range from the hinge (whiskers). Notches are defined as 1.58 * IQR / sqrt(n). On panels (F) and (G), data points beyond the whiskers are shown as outliers.Fig. 5: Extracellular monomers evolved from monomer ancestors.A Example phylogenetic tree, with the quaternary structure and cellular locations of the proteins. Pie charts at the nodes indicate the probability of the number of chains for each node, extracellular proteins at the leaves are highlighted with green. B, C The frequency of extracellular proteins in the orthogroups, depending on the type of analysis (Binary or Multi-state). The fraction of extracellular proteins is higher in orthogroups where the ancestral state was a monomer, however, the difference is only nearly significant in the multi-state comparison due to the small number (7) of trees (Wilcoxon-rank sum tests.). Boxplots display the median, 25–75% interquartile range (IQR), and 1.5 * interquartile range from the hinge (whiskers). Notches are defined as 1.58 * IQR / sqrt(n). Datapoints beyond the whiskers are shown as outliers. P-values are from Wilcoxon rank sum tests. D, E Matrices with the quaternary structure (QS) of the root and extracellular proteins in the orthogroups which do have extracellular proteins. P-values (test of proportions) indicate the difference between extracellular QS in trees with homomer and monomer roots. In the majority of cases, the QS of extracellular proteins is the same as the root, indicating that most extracellular monomers did not lose their quaternary structure but already evolved from monomers, while in the trees with homomer root most of the time, they remained homomers. This pattern is consistent with the predictions of CNE. F The average interface area of homomers scales similarly with the fraction of extracellular proteins in the orthogroups with homomers of monomer ancestral state. G–J Within orthogroups, the average molecular weight (MW) of extracellular homomers is smaller than the MW of cytoplasmic monomers, but there is no difference in their subunits (one-sample Wilcoxon-tests). This suggests that besides CNE, the selection also shapes their quaternary structure. Note that due to the larger size of eukaryotic proteins, prokaryotes and eukaryotes were treated separately, even within orthogroups. K, L The MW of monomers is not affected by their cellular location (one-sample Wilcoxon-tests).To test whether there are trends in the evolution of homomers, we split the trees into two groups depending on whether the root was a homomer or a monomer (Fig. 4A). Next, we examined the frequency of quaternary structure changes in the trees. We found large differences between the phylogenies: in the binary analysis, in the trees where the ancestral state was a monomer, the frequency of multimerization and the fraction of homomers in the tree is much higher than the frequency reversion to a monomer in those trees where the root was a homomer (Fig. 4B and Supplementary Fig. 4A). In the case of homomers with more than two subunits (e.g., tetra- or octamers) reversion to a monomer state is likely to need more than one steps, while a dimer can evolve from a monomer in a single step. To account for this, using the multi-state assignments, we also examined in every tree the frequency of subunit gains and losses, which shows a similar, highly significant difference as in the binary assignment (Fig. 4C and Supplementary Fig. 4B). This is also reflected in the size of the interfaces: homomers in trees with a monomer-root have much smaller (~2x) interfaces than homomers in trees where the root was also a homomer (Fig. 4D and Supplementary Fig. 4C), indicating that once large interfaces were evolved, reverting to a monomer state is much less likely than evolving a new (and relatively small) interface by a monomer. To rule out the possibility that this pattern is caused by major differences in the topologies of the trees (for example, by lower divergence from the root in trees with homomer-root), we calculated the average distance from the root (Supplementary Fig. 6A) and the average distance between the proteins (Supplementary Fig. 6B) for every tree, which show no significant differences between the two tree categories (Supplementary Fig. 6). Finally, in the case of multi-state trees with homomer roots, we also calculated whether the frequency subunit gains and losses differ; surprisingly we found no significant difference between the two (Fig. 4E and Supplementary Fig. 4D). However, this is in agreement with recent experimental findings which show that in the case of homomers that are multimers of dimers, new subunits can be gained and lost relatively easily39,40, at least when the new interfaces are not very large or hydrophobic.Since many of the homomers in the trees with monomer-root have small interfaces, and some of these interfaces might be crystallographic, rather than biological6,35, we also examined whether the pattern changes if homomers with interfaces below 1000 Å2 are assumed to be quaternary structure errors and are treated as monomers in the analyses. The reanalysis of the data with this extended set of monomers with ML shows that the pattern remains similar (Supplementary Fig. 7), thus, our findings are unlikely to be significantly influenced by quaternary structure errors in the PDB.Interface variability depends on the ancestral state of the orthogroupsNext, using the ancestral state of the roots of the multi-state analysis, we examined whether and to what degree the overlap between interface residues depends on the structural similarity of the proteins and the size of the interface. We found that within orthogroups, the overall structural similarity between the subunits of different homomers is high, with the average TM-score being above 0.7–0.8 in the vast majority of the trees (Fig. 4F and Supplementary Fig. 4E). The average overlap of interfaces depends largely on the ancestral state of the tree: in the trees where the root is homomer, it is typically high, above 75%, while in the trees with monomer root, the overlap is much smaller, less than ~50% in most cases (Fig. 4F and Supplementary Fig. 4E), indicating that in these orthogroups different interfaces evolved repeatedly. In addition, interface overlap depends on the size of the interfaces: in orthogroups where the average interface size is below ~1200 Å2, the overlap drops (Fig. 4G and Supplementary Fig. 4F), irrespectively whether the root of the tree is a homomer or monomer, suggesting that, besides rapid evolution, some of these small interfaces do contain crystallographic artefacts. However, in the case of orthogroups with monomer root, even the plateau above 1200 Å2 shows only 25–50% interface overlap (Fig. 4G and Supplementary Fig. 4F), indicating that these orthogroups are characterised by the repeated evolution of homomers with new interfaces.
Extracellular protein evolution indicates the importance of CNE
Next, we examined whether the evolution of quaternary structure in extracellular proteins is more consistent with the predictions of constructive neutral evolution (CNE, “hydrophobic ratchet”), or with the traditional view, which assumes that quaternary structure is adaptive. Extracellular proteins can be seen as a “canary in the coalmine” because the two hypotheses make different predictions: the hydrophobic ratchet predicts that once they evolve, losing interfaces is difficult, thus extracellular monomers will primarily evolve in orthogroups where the ancestral state is also a monomer. In contrast, the adaptive view is less restrictive and does not assume the persistence of quaternary structure without a fitness advantage, thus, it is more permissive for the evolution of monomers from homomers and losing interfaces.To test this, we selected the trees/orthogroups with at least one extracellular protein, where the cellular location could be determined for at least five proteins, and the probability of the most likely ancestral state is higher than 51% (see Fig. 5A for an example). The number of such trees is limited, 35 in the binary and 40 in the multi-state case with the ML ancestral reconstruction method (Fig. 5), and 33 and 37, respectively, with MP (Supplementary Fig. 5). First, we examined the frequency of extracellular proteins, which is higher in orthogroups with a monomer root than in orthogroups with a homomer root, in the binary, and to a lesser degree in the multi-state analyses (Fig. 5B, C and Supplementary Fig. 5B, C). Next, using the most common quaternary structure of extracellular proteins in the orthogroups, we tested whether their quaternary structure changed compared to the root. We found that in the majority of trees, the most common quaternary structure of extracellular proteins is the same as the root (Fig. 5D, E and Supplementary Fig. 5D, E), and it changes compared to the root only in 22% (Fig. 5D, Multi-state analysis) and 20% (Fig. 5E, Binary analysis) of the trees (ML), and even less in the case of the MP method (Supplementary Fig. 5D, E). In the vast majority of the trees where most extracellular proteins are monomers, the root was also a monomer, and the difference between trees with homomer and monomer roots is highly significant for both analyses (Fig. 5D, E and Supplementary Fig. 5D, E). Since in some trees, the majority of proteins are extracellular (see Fig. 5B, C), which can make changes in a quaternary structure very unlikely, thus we repeated the ML analysis using only trees where the fraction of extracellular proteins is less than half of the total number of proteins (Supplementary Fig. 8). This reduced set of trees (29 and 26 for multi-state and binary) still shows a largely similar pattern, as only 27% (Supplementary Fig. 8A, Multi-state analysis) and 26.9% (Supplementary Fig. 8B, Binary analysis) of the trees indicate a change in quaternary structure. The comparison of interface sizes suggests that interface size scales similarly with the fraction of extracellular proteins irrespectively of the root, although the small number of trees with homomer root results in a very high uncertainty for the slope (Fig. 5F and Supplementary Fig. 5F). These findings indicate that reverting to a monomer state from a homomer is difficult even in the case of extracellular proteins, and most extracellular monomers evolve from monomer ancestors, in agreement with the predictions of the CNE model.Extracellular homomers have reduced molecular weightIn the above analysis, we focused on qualitative changes in the quaternary structure (i.e., a change from homomer to monomer). However, selection might also result in more subtle, quantitative changes, like a reduction in the size of complexes or proteins, for example, due to the better solubility of smaller complexes/monomers, or their higher diffusion rates (Fig. 2A). Diffusion rates scale with the molecular weight (MW) of proteins, and complexes, especially with many subunits, are known to have much lower diffusion rates than monomers41. We examined whether extracellular and cytoplasmic proteins within the same orthogroups have similar MW and found that extracellular homomers do show a statistically significant reduction in MW (p = 0.013, Fig. 5G, H), but not their subunits (Fig. 5I, J), or monomers (Fig. 5K, L), indicating that quaternary structure is affected by cellular location, but not protein length. (Note that Prokaryotes and Eukaryotes were treated separately, even within the same orthogroups, because Eukaryotic proteins are generally larger than Prokaryotic ones.) This suggests that selection is also likely to contribute to the loss of subunits/interfaces in extracellular proteins, although the uncertainty associated with the analysis is high due to the small number of orthogroups having both extra- and intracellular homomers.Amino acid synthesis cost is unlikely to be a major determinant of homomer frequencyThe synthesis of amino acids is metabolically costly, and in E. coli, the synthesis cost, measured as the number of high energy ATP P-bonds required for synthesis in the synthesis pathway, can range between 11 P-bonds (Alanine, Glycine, Serine), to 74 (Tryptophan)42, and hydrophobic residues with large side chains are characterised by high synthesis costs (e.g., Tryptophan, Tyrosine, Phenylalanine). Such costs can change when the entire metabolic network is considered or using different substrates43 nevertheless, maintaining a hydrophobic interface is likely to be more costly than having an equivalent solvent-accessible (hydrophilic) surface. Thus, having unnecessary hydrophobic interfaces might be deleterious and is unlikely to be a completely neutral trait. We examined whether the synthesis costs of proteins differ in homomers and monomers of E. coli, and found that, as expected, the surface of subunits (that included interface residues, see “Methods”) has a higher per-residue synthesis cost in homomers (Fig. 6A), particularly in interfaces (Fig. 6B). However, the protein core is more costly in monomers (Fig. 6C), and the total synthesis cost of proteins is also slightly higher in monomers, even when differences in codon adaptation index (which integrate expression levels) are taken into account (Fig. 6D). This raises the possibility that in E. coli (and species with similar metabolic pathways), the high frequency of homomers in the cytoplasm is nevertheless the result of selection against monomers for metabolic reasons, if oligomerisation enables the reduction of total synthesis costs of proteins.Fig. 6: Synthesis costs of homomers and monomers.A Cost of surface residues (including interface) in E. coli. B Difference between interface and solvent accessible surface residues in homomers of E. coli. C Cost of core residues in E. coli. D Cost of full protein (including signal peptides) in E. coli. To remove redundancies, E. coli proteins were clustered at 30% sequence similarity, resulting in 406 homomers and 134 monomers. As expected, the per-residue synthesis cost of the subunit surfaces is somewhat more costly in homomers than in monomers due to the higher synthesis cost of interfaces, however, the synthesis cost of the core and full protein is more costly in monomers, even if codon adaptation index is included as a covariate. E The frequency of homomers and monomers in selected taxa (Gram-negative bacteria, mammals, Plasmodium [malaria parasite]) with different amino acid metabolism (numbers indicate the number of clusters). Plasmodium takes up most amino acids from its host, in mammals most hydrophobic amino acids are essential, while most bacteria can synthesise all amino acids. (See also Supplementary Fig. 11). F Within orthogroups, the per-residue synthesis cost of prokaryotic extracellular and intracellular monomers does not differ significantly. G Between orthogroups, the synthesis cost of extracellular monomers is not lower than the costs of monomers in orthogroups without extracellular proteins. On all panels where present, boxplots display the median, 25–75% interquartile range (IQR), and 1.5 * interquartile range from the hinge (whiskers). Notches are defined as 1.58 * IQR / sqrt(n). Datapoints beyond the whiskers are shown as outliers.This hypothesis can be largely ruled out by the examination of taxonomic groups with amino acid auxotrophies. In mammals, most hydrophobic amino acids (F, I, L, M, V, W) are essential and are taken up from the environment, and two (C, Y) are conditionally essential. Thus, there is no, or just reduced synthesis cost associated with them. Intracellular parasites like Plasmodium sp. or Toxoplasma take up most amino acids from their host44, and parasitic bacteria like Tenericutes (which include Mycoplasma) are also auxotrophic for most amino acids45. In contrast, auxotrophy is rare in plants, and most bacteria (78.4%) can synthesise all amino acids45. Despite these major metabolic differences, the ratio of homomer and monomer enzymes in the cytoplasm does not vary dramatically between these taxa (Fig. 6E and Supplementary Fig. 11), and auxotrophic endoparasites also consistently show the same pattern, even though the number of their currently available PDB entries is very low. In addition, theory suggests46 that in eukaryotes, the cost of protein production is usually not high enough to be perceived by natural selection, while it is in prokaryotes.It has also been suggested that in prokaryotes, extracellular proteins, especially flagellar and fibrous proteins, have reduced synthesis costs because their amino acids are not recycled by the cell47. Using the proteins of the orthogroups, we examined whether this pattern is present also in enzymes (and whether it can contribute to the low frequency of homomers in extracellular proteins). The results indicate that in our dataset, the synthesis costs of extracellular monomers are not lower than the synthesis costs of intracellular monomers, neither within orthogroups (Fig. 6F) nor between orthogroups (Fig. 6G), and also when all prokaryotic extracellular monomers are compared to all prokaryotic intracellular monomers (n = 176 and 1426, respectively; p = 0.487, Wilcoxon rank sum test). Thus, location itself is unlikely to have a major effect on the synthesis cost of these enzymes, and their quaternary structure.However, proteins that are more expressed or abundant were shown to have lower average synthesis costs42,48,49, thus the lower overall metabolic cost of homomers in E. coli can also reflect higher protein abundance or expression, which is in agreement with the hypothesis that macromolecular crowding is a significant factor in the evolution of oligomers (see below).
Homomers are more abundant in the cell than monomers
If the high frequency of oligomerisation in the cytoplasm evolves due to macromolecular crowding, that predicts that oligomers will be formed primarily by proteins that are more abundant than monomers. We tested this hypothesis using the E. coli, yeast and human enzymes, and abundances obtained from the paxDB database50. Besides using our high coverage enzyme dataset from the PDB where the quaternary structure is known, in the case of E. coli and yeast, both the heteromers51 and homomers52 of their entire proteomes are known or predicted, thus their monomers can also be defined, and proteome scale comparisons of enzymes are possible. Our results indicate that in all three species, homomers have higher abundances than monomers (Fig. 7, white panels), the only exception being the comparison in the yeast set of the PDB (Fig. 7D), most likely due to the low numbers of available proteins. In addition, the pattern is present both in cytoplasmic (Supplementary Fig. 12) and non-cytoplasmic (Supplementary Fig. 13) proteins. This is consistent with the hypothesis that homomer interfaces evolve due to macromolecular crowding, i.e., due to the high frequency of interactions in the cell and the minimisation of excluded volume29,30.Fig. 7: Homomers are more abundant than monomers.A Abundance of E. coli enzymes using the full proteome, which includes proteins from the Homomer atlas where oligomerisation was predicted with AlphaFold2, and also the entire PDB, including the low coverage structures. The white panel indicates raw abundances from the paxDB, and the grey panel indicates abundances scaled with the putative number of subunits in each complex, thus it is proportional to the number of particles. B Abundance of E. coli enzymes, but using only the filtered enzyme dataset, with high coverage structures in the PDB. C, D The same for Yeast. E Abundance of Human enzymes, using only the high-coverage PDB dataset. Except for panel (D), in all three species, homomers are more abundant than monomers (panels with white background), but the difference disappears when the scaled abundance (~number of particles) is used. This pattern is consistent with the hypothesis that the high abundance of proteins (macromolecular crowding) facilitates the evolution of homomers. However, it also indicates that not just the number of proteins, but also the number of particles they form is regulated by the cell, suggesting that oligomerisation contributes to the maintenance of cellular homoeostasis. On all panels, boxplots display the median, 25–75% interquartile range (IQR), and 1.5 * interquartile range from the hinge (whiskers). Notches are defined as 1.58 * IQR / sqrt(n). Datapoints beyond the whiskers are shown as outliers. All p-values were calculated with Wilcoxon rank sum tests.In addition, when the abundances are scaled down with the number of subunits of each complex (or, in the case of predicted structures, the putative number of subunits, see “Methods”), thus when the abundance of the complexes/particles they form is compared, in all three species the difference between homomers and monomers largely disappears (Fig. 7, grey panels). This suggests that not only the abundance of enzymes (i.e., the number of catalytic sites), but also the abundance of their complexes (the number of particles they form) is regulated in the cell, the two might be shaped by separate processes, and oligomerisation allows a certain level of uncoupling between the two. Thus, besides biochemical function, the evolution of oligomers and their frequency might also be shaped by the necessity to maintain cellular homoeostasis, most likely by their effect on the properties of the cytoplasm, like viscosity/fluidity, osmotic pressure or diffusion rates.However, several studies have demonstrated that in eukaryotes, oligomerisation increases the half-life of protein complexes and reduces their degradation rate, due to burying ubiquitinoylation sites and degrons in interfaces18,26,53,54. A similar interface effect has not been described in bacteria so far, nevertheless, it may exist, as interface residues are generally enriched at the C-termini of proteins55, and may overlap with C-degrons. Thus (at least in eukaryotes and archaea), the higher abundance of oligomers may be the consequence of oligomerisation, the cause of it, or both. Mallik and Kundu demonstrated that in yeast, the half-life of homomers scales with their buried interface area18, and using the high-coverage enzyme dataset, we examined whether the association between abundance and buried interface area is similarly strong as between abundance and oligomerisation (Fig. 8 and Supplementary Fig. S14). We found that both in E. coli (where the ubiquitin-proteasome pathway doesn’t exist) and human datasets there is a significant positive correlation between the buried interface area of homomers and their abundance, thus the presence of interfaces and complex formation may influence abundances. However, its correlation coefficient (R2) is lower than between quaternary structure and abundance (Fig. 8), especially when only dimers are used, where only E. coli is significant (Supplementary Fig. 14). Thus, the higher abundance of homomers is unlikely to be purely the consequence of lower degradation rates due to having interfaces. The presence of degrons in interfaces is likely to be used as a regulatory mechanism by the cell, however, the fact that interfaces do influence protein abundance through degradation, together with macromolecular crowding, can result in a positive feedback between the two: abundant proteins oligomerise, which in turn reduces their degradation rate, resulting in even higher abundances and interaction frequencies. We hypothesise that besides the hydrophobic ratchet, this “degradation ratchet” might be an additional process causing the accumulation of interfaces over evolutionary timescales, at least in eukaryotes.Fig. 8: The higher abundance of homomers is unlikely to be the by-product of their lower degradation rates due to having interfaces.A E. coli. B Yeast. C Human. The left panels indicate the correlation between quaternary structure and abundance and the corresponding abundance density plots, the right panels indicate the correlation between abundance and the relative surface area of homomers that are buried in the interface. p-values were obtained with the Pearson correlation test. Red vertical lines indicate the average buried surfaced area. While R2 is low in all cases, it is significant in E. coli and Humans, and in both cases the association between buried surface area and abundance is weaker than between quaternary structure and abundance. This indicates that the positive correlation reported previously in eukaryotes between relative interface size and protein half-life is not sufficient to explain the abundance difference between homomers and monomers. (See also Supplementary Fig. 14, showing only dimers).
Cellular location shapes quaternary structure of enzymes
