Protein disorder reduced in Saccharomyces cerevisiae to survive heat shock

Data

We downloaded the yeast (S. cerevisiae) proteome from UniProt (proteome ID: UP000002311)³⁸ as fasta files including only the reviewed proteins (UniProtKB/Swiss-Prot). Removal of duplicates applying the method Uniqueprot2³⁹ (with 100% pairwise sequence identity, keeping the longer sequence) left 5667 proteins (Table S1A). We considered the 16 nuclear chromosomes (matched through http://www.yeastgenome.org⁴⁰, the numbers of proteins per chromosome are given in Figure S1B). The yeastgenome.org resource also provided the annotations of heat-shock response proteins (HSR). Proteins known to interact with HSR proteins augmented this set of HSRs in the following way.

BioGRID (version 3.1.86) provided the data for experimental protein-protein interactions (PPIs) in yeast. After filtering out redundancy (a-b and b-a counted only once) and excluding self-interactions (a-a), we based all subsequent analysis on the single largest connected component of the network. We focused on the most basic network features that allow the comparison and characterization of complex networks. The most elementary characteristic of a node is its degree or connectivity, defined as the number of interactions for a node (here protein), i.e. the number of interactions one protein has with all others. Another important parameter is the betweenness, i.e. the fraction of shortest paths between all other nodes that has to go through a given node. Additionally, we monitored the average degree of neighbors, which depends on the number of nodes and links in the network. These three parameters measured the importance of each node within the network.

Disorder predictions

We applied methods capturing different “flavors” of protein disorder^29,41,42. IUPred (version 1.0) is based on statistical contact potentials and exclusively uses single sequences^28,43. MD (Meta-Disorder)⁴² combines different original prediction methods through machine learning (neural network) with evolutionary profiles and predictions of solvent accessibility and protein flexibility. To some extent disorder is a gradual phenomenon, i.e. proteins may have more or less disorder⁴⁴. On the other hand, prediction methods distinguish between a 30-residue loop resembling “protein disorder” and another resembling a region with “regular structure”²⁹. Thus, protein disorder seems more a binary feature (it is there or not, or present/absent) than a gradual one²⁵. Unfortunately, no argument or data determines one single correct threshold for what constitutes present/absent for protein disorder. Typically, experts use a length threshold of the type: protein disorder is present when at least T consecutive residues in a protein are predicted to be disordered. If so, this protein is considered to contain a long region of disorder. More disorder in this model could imply, e.g. more than one region, or the entire protein. We analyzed many alternatives to choose the threshold for long disorder, and found most to be redundant. We included different views only if they provided relevant information. In particular, we largely focused on one threshold to define “long disorder”: %long30, is the percentage of proteins with at least one region of ≥30 consecutive residues predicted as disordered (alternatives were: %long50 and %long80, i.e. with length thresholds at ≥50 and ≥80, and completely disordered implying no region of 30 consecutive residues without any disorder).

GO term enrichment

We applied BINGO (Biological Networks Gene Ontology⁴⁴, version 2.44) to identify the enrichment of GO (Gene Ontology)⁴⁵ terms in subsets of experimentally annotated proteins. We focused on “biological process” and “molecular function”. For two sets of proteins with annotated biological functions (more precisely: GO numbers) BINGO estimates to which extent their annotations differ in a statistically significant way. We visualized BINGO results with Cytoscape⁴⁶ platform (version 2.8). Our analysis focused on the hypergeometric test in BINGO, which accurately estimates p-values as it tests without replacement. Following the common, procedure for BINGO, we considered p-values >0.05 as significant⁴⁶. Testing multiple hypotheses may give many false positives (Type I error: incorrect rejection of true null hypothesis^47,48). Using BINGO, we corrected for these through the Benjamini and Hochberg correction which provides strong control over the False Discovery Rate (FDR, expected proportion of erroneous null hypothesis rejections among all rejections⁴⁸).

Duplications in response to high temperature reduce protein disorder

In response to high temperature yeast (S. cerevisiae) duplicates the entire chromosome III (for brevity we use chrN to denote ‘yeast chromosome N’ with N as Roman numerals following convention) and fragments from chrIV and chrXII²⁰. The size of the 16 yeast chromosomes varies over six-fold (Table S1). The average protein length is similar between the 16 chromosomes (Figure S1, Table S1). The duplicated chrIII is the 3^rd smallest with 183 genes, of which 153 are mapped and 132 constitute “verified ORFs”. Fewer genes are encoded only by chrI with 90, and chrVI with 125 proteins (Table S1). The relatively small number of genes on chrIII was one reason for choosing it as the first fully synthesized functional yeast chromosome⁴⁹. In contrast to protein length, the percentage of proteins with predicted long regions of disorder differed significantly between the 16 yeast chromosomes (Figure 1).

Figure 1. Protein disorder differs between yeast chromosomes.

The composition of proteins with long regions of disorder (y-axes) differed significantly between the 16 chromosomes of S. cerevisiae (x-axes) and also for the set of HSPs. The three rightmost marks on the x-axes describe: HSPs and the disorder predictions for the HSR-related duplicated fragments on chromosome IV and chromosome XII (frag IV and frag XII). The differences were similar for two different prediction methods (MD in black, IUPred in light gray), and for different thresholds with respect to the minimal length of a disordered region (A: ≥30 consecutive residues predicted in disorder, B: ≥50, C: ≥80). Dashed horizontal lines mark the averages over all chromosomes. Error bars are too small to become visible on the scale chosen. The least disorder content was predicted for chromosome III and chromosome X. Overall, all duplications in response to heat shock treatment reduced the level of protein disorder in the offspring.

The least protein disorder was predicted for chrIII and chrX (Figure 1, Table S2). That means heat response duplicates one of the two chromosomes with the least disorder. In addition, the fragments of chrIV and chrXII that are duplicated along with the entire chrIII also clearly have less disorder than the chromosomes from which they were taken (Figure 1). This enhances the effect of protein disorder reduction in response to high temperatures.

The other low-disorder option is chrX: Why not duplicate chrX in response to high temperatures? ChrX is more than twice as large as chrIII (Figure S1). Thus duplicating chrX would “cost” twice as much. This might be prohibitive. ChrX might also not contain the cell activities important for coping with high temperature. Furthermore, as chrX and chrIII are similar in disorder content while chrX has twice the proteins of chrIII, the duplication of chrX would increase the overall level of proteins with disorder that might become unfolded and thereby “jam” cellular activity more than the duplication of chrIII.

Assume a certain amount of tolerable duplication were tolerable and that number were about 153 proteins (as for chrIII): where in the genome do we find a continuous stretch (within a chromosome) that has 153 proteins with the least disorder? Our results underscored the special role of chrIII (Figure S2): only 3% of all continuous genome fragments with 153 proteins have as little disorder as chrIII (corresponding numbers for chrX: 5%; 29-protein fragment from chrIV: 52%; 64-protein fragment from chrXII: 10%). These figures demonstrate that the duplication of chrIII might be THE optimal choice for a simple way to duplicate 153 proteins with as little disorder as possible.

Heat-shock proteins do not explain the temperature-related duplication

Our results explained why duplicating 150–200 proteins from another chromosome might have been potentially more damaging than the duplication of chrIII in response to high temperatures. In other words, our model might suggest why the duplication of this particular region is better than other duplications. However, what is the selective benefit from the proteins on chrIII? We expected to find the answer to this question in proteins that actively help with coping with heat stress. The immediate suspects are heat-shock proteins (HSP) and the proteins known to interact with these HSPs (HSP-binders). The known HSPs and HSP-binders scatter over all 16 yeast chromosomes (Figure S3). All regions duplicated in response to heat shock contain only one known gene coding for HSPs (HSP30) and one known HSP-binder (TAH1). This implied that 1.3% of all known HSPs and HSP-binders were duplicated in an event that duplicated 0.5% of all genes, i.e. a 2.6-fold over-representation. This statistically insignificant the finding that fewer than 1 in 50 of all HSPs and HSP-binders are duplicated might still be scientifically significant if HSP30 and TAH1/HSP90 were outstandingly important proteins for the given conditions. However, this is not the case, at least not given what is currently known about HSP30. Furthermore, introducing an extra copy of HSP30 into wild-type cells did not increase the ability of the cells to cope with high temperature (Dahan & Pilpel, personal communication).

The set of known HSPs (Figure S3) slightly changed expression levels in response to heat stress during the fixation of the trisomy²⁰ only slightly but almost all HSPs were significantly up-regulated (arrows in Figure S3) when the “refined descendants” replaced the trisomy²⁰. This could imply that the duplicated genes are essential for survival under heat stress. Nevertheless, quite contrary to the naïve expectation, the HSPs and the HSP-binders by no means explained the heat-stress-specific duplications observed experimentally.

Incidentally, HSPs appeared particularly abundant in disorder regions of 30–50 consecutive residues (Figure 1A, in particular for IUPred). It has previously been argued that such disorder is required for proper function of HSPs⁵⁰. In contrast, HSPs are depleted of longer disorder (>50; Figure 1B,C).

Overall, we argue that HSPs could have explained the duplication of many other chromosomes, possibly even better than that of chrIII. Therefore, this explanation is not specific. Thus, we conclude that the duplication of known HSPs and HSP-binding proteins did not explain why chrIII was specifically duplicated. Many HSPs and HSP-binders might remain unknown. However, we have no scientific ground to suspect that the fraction of the unknown HSPs differs between the chromosomes, i.e. that there are particular HSPs on chrIII that remain undiscovered.

GO terms enriched for growth and reproduction in heat stress-duplication

Are any other proteins on chrIII important for growth under high temperature? Simply scanning GO⁴⁵ annotations is insufficient: the question is not whether proteins on chrIII have certain functions, but whether these are overrepresented enough to explain why chrIII and not the other two small chromosomes (chrVI or chrI) are duplicated in response to high temperatures. In order to address this question, we need a GO term enrichment analysis of the duplicated regions⁵¹.

Growth and reproduction might be considered as the most important cell activities in the sense that the organism must grow and proliferate (cells that fail are not observed) even under stress. The GO enrichment analysis seemed to confirm this expectation (Figure 2): the two most abundant GO terms in the heat stress-duplicated regions were those related to (i) sexual reproduction (Figure 2 and Figure S2; “conjugation with cellular fusion”, “reproductive cellular process” and “response to pheromone”) and to (ii) sugar transport (hexose transport process as well as mannose, fructose and glucose transmembrane transporter activity; Figure S4).

Figure 2. GO enrichment of sexual reproduction and nutrient uptake.

The tree gives the complete set of all experimentally annotated GO-terms (Gene Ontology⁴⁵) for any of the proteins on chromosome III that describe biological process (left branch) and molecular function (right branch). The enrichment analysis⁵¹ describes how much chrIII GO-terms are enriched with respect to all other GO-terms from yeast: all terms marked by yellow circles are significantly enriched. Sexual reproduction (7 GO-terms on chrIII) and transport (carboxylic acid and organic anion 4 GO-terms on chrIII) mapped to most overrepresented GO terms on this chromosome.

The major energy source of yeast is sugar, in particular hexose monosaccharides (C6H12O6; e.g. glucose, fructose, mannose). These nutrient sugars are imported into the cell through hexose transporters, which are encoded by HXT genes^52,53. The HXT yeast genes on the duplicated fragment of chromosome IV (HXT3, HXT6 and HXT7) are almost five-fold over-represented with respect to random (yeast has 5667 N_genY genes, 243 N_genD are duplicated, 15 N_genHXT genes are in yeast; in a region with 243 N_genD genes we would find by chance 0.64 HXT genes in the duplicated regions p_chance=N_genHXT *[N_genD/N_genY]). Two HXT genes on the duplicated chrIV fragment (HXT6 and HXT7) appear to encode high-affinity transporters required for growth at very low glucose concentrations (~0.1%⁵⁴), i.e. these two would become particularly important when yeast is cultured under glucose limitation⁵⁴. Interestingly, several works have detected duplication of these two genes (HXT6 and HXT7) in yeast populations evolving under low nutrient availability^8,55. These numbers suggest that heat stress also puts strain upon obtaining the energy needed for growth and reproduction.

Sexual reproduction also appeared crucial for the survival of yeast cultured under heat stress^56,57. Seven of the ten molecular functions to be significantly overrepresented in the heat stress-duplicated chrIII (Table S3) by a standard GO-term enrichment analysis⁵¹ are involved in reproduction. Three of these seven molecular functions are related specifically to sexual reproduction; the others pertain to general reproductive processes (Figure 2). In particular, the reproduction-related processes involve cell fusion (FUS1 and FIG2^58–60), pheromone response (STE50 which is also required for optimal invasive growth and hyperosmotic stress signaling^61,62 and NOT1 that is also involved in several RNA regulation levels⁶³), nuclear fusion, chromosome disjunction, nuclear segregation after mating (BIK1 which is involved in microtubule function during mitosis^64,65), fusion of haploid nuclei during mating; KAR4 or KARyogamy plays a critical role in the choreography of the mating response⁶⁶), cytokenesis (division of cytoplasma and plasma membrane of a cell and its separation into two daughter cells which is also relevant for asexual mitotic growth: CDC10⁶⁷), specification of the site where the daughter cell will form (relevant for budding and asexual growth, also referred to as axial bud selection) and in the developmental process in which the size of a cell is generated and organized (also referred to as morphogenesis: CDC10^67–69). All these genes are also required for the correct localization of other proteins involved in cytokinesis and bud site selection^67,70–73. Other important processes and activities overrepresented on chrIII are related to the avoidance of oxidative stress (e.g. carboxylic acid transport – Figure 2 - which may be important for the survival since during the vegetative asexual reproduction cells were exposed to oxidative stress) and NAD(P)H nitro-reductase activity (Figure 2). The only nitroreductase-related proteins in yeast – HBN1 and FRM2⁷⁴ - are only on chrIII (Table S4). The proteins involved in these two activities (carboxylic acid transport and NAD(P)H nitro-reductase activities) are also implicated in cellular detoxification⁷⁵, which is another task relevant for survival under stress.

All these data supported the view that chrIII is important for sexual reproduction. A seemingly convincing story, until we learned that the laboratory strains of yeast survived through asexual reproduction²⁰, i.e. apparently did not need what is so uniquely enriched in the heat stress duplication. The set of proteins known to be involved in reproduction on chrIII (Table S3) had more disorder than the average for chrIII (Figure S5). Some of these proteins with long disordered regions might not work correctly in heat.

Why duplicate proteins that fail? Not having found a convincing answer, we propose two conjectures: first, sexual reproduction might “frame” another cellular activity of the same protein that is more relevant to the growth conditions applied during the evolution in the laboratory experiment. For instance, CDC10 is also required to maintain cell polarity (GO: 0030011), BUD3 and BUD5 are involved in axial cellular bud-site selection (GO: 0007120), KCC4 a bud neck kinase involved in budding and cell bud growth (GO: 0007117) and BIK1, which is involved in microtubule function during mitosis. All of these activities are related to asexual reproduction. Our second proposition seems more far-fetched, namely that the set of proteins with the strongest GO-enrichment might have been duplicated coincidentally, i.e. the disorder-rich proteins related to sexual reproduction might have been duplicated because they happened to be on chrIII but not due their relevance for the survival in heat. If so, there must be something else we have not found yet on chrIII.

Several other processes were slightly enriched in the duplicated fragments with some relevance for yeast survival in heat but none of those gave a clear explanation (Figure 2): (i) fatty acid elongase, (ii) rRNA (guanine) methyltransferase, and (iii) the importin-alpha export receptor activities. We analyzed these in detail. (i) Fatty acid elongase: currently, only three proteins are known to be involved in lengthening fatty acids; two of those (ELO2 and APA1; Table S3) are on chrIII. Fatty acid elongases are involved in sphingolipid biosynthesis. The sphingolipids are components of the cellular membrane and bioactive signaling molecules that contribute to heat tolerance as they are directly involved in organizational cellular structures (e.g. cell membrane)⁷⁶. (ii) rRNA methyltransferases: three yeast proteins are known to be involved in rRNA (guanine) methyltransferase activity; two of those (BUD23 and SPB1) are on chrIII (Table S4). It is believed that the modification of ribonucleotides optimizes the rRNA structure and represents a way to expand the topological potentials of RNA molecules. It is possible that the loss of modification affects fine-tuning of ribosome function that could give rise to the pronounced cold-sensitivity⁷⁷. (iii) Importin-alpha nuclear export: two yeast proteins contribute toward the importin-alpha export receptor activity; one of those (MSN5) is in the duplicated fragment of chromosome IV. MSN5 knockout mutants show a variety of phenotypes, including carbon-source utilization, defects and sensitivity to high concentrations of ions, severe heat shock, and high pH⁷⁸. Moreover, these mutants are partially sterile⁷⁸. Therefore, this protein appears necessary for cell survival, especially under extreme conditions.

Only one cellular activity related to tRNA synthase appeared overrepresented on the duplicated fragment of chrXII (DUS3 and DUS4 proteins; Table S7). In particular, to the tRNA dihydrouridine synthases, which are responsible for the reduction of the 5,6-double bond of a uridine residue on tRNA (one of the numerous modifications observed on tRNA cytoplasmatic⁷⁹). However, this particular finding appeared less relevant since the corresponding fragment was only duplicated in one of four growth experiments in response to high temperatures²⁰.

One crucial limitation for any functional enrichment study remains the incomplete experimental annotation even for an organism as intensively studied as yeast. It may be that all our speculation above missed the real causation because the functions of the proteins that are really relevant remain uncharacterized. Therefore, we complemented our analysis with one aspect of function for which we have a complete prediction, namely the prediction of sub-cellular localization of all yeast proteins. The experimental localization annotations for yeast are still cover at most 70% of all proteins⁸⁰. However, today’s top prediction methods, such as LocTree3, are very reliable⁸⁰ and can make crucial differences for comparing ‘complete’ data sets⁸¹. We found nuclear proteins to be clearly depleted on chrIII (-4.6 percentage points with respect to the entire proteome; Figure S7A). Other abundant proteins found on chrIII were secreted (extra-cellular) or annotated as endoplasmic reticulum (ER) membrane proteins (each 3.2 percentage points higher than in the full yeast proteome). We also observed significantly more disorder in nuclear proteins (nuclear 77% vs. <40% for non-nuclear; Figure S8). This might explain the depletion of nuclear proteins on chrIII. While these findings were clear, they did not suggest a simple interpretation. The abundance of secreted proteins on chrIII (about 3.2 percentage points more on chrIII than in entire yeast; Figure S7A) implies that in the response to heat shock, more proteins are secreted into the ‘hot’ environment. Given the correlation between habitat and disorder³⁷, we expect that proteins are more likely to sustain high temperatures with less disorder. Unfortunately, a GO enrichment study of the secreted proteins also did not provide the answer we had been hoping for. However, the “secretome” alone could not explain the lower content of disordered proteins on chromosome III (disorder entire yeast-chrIII=50%-43%=7%>3% for secretome; Table 1 and Figure S7A).

Proteins from chrIII less implied in overall PPI network

As proteins cannot be understood without also considering their networks of interaction, we compared the network of experimentally characterized PPIs between the entire yeast and those fragments that are duplicated in heat evolving populations. As for the differential analysis of any experimental annotation, the limitation of such an approach lies in the incompleteness of the experimental data. In all 16 chromosomes, the degree (number of interactions per protein) was lowest for chrIII (average=16±2; Figure 3A). A similar trend was observed for betweenness (number of times that a protein acts as a bridge along the shortest path between two other proteins: average=1800±300; Figure 3B). Furthermore, chrIII is one of the chromosomes with the largest mean value for the average neighbor degree (average=380±40; Figure 3C). Our network analyses confirm chrIII as a good choice for a first line of defense against high temperature because the proteins encoded on this chromosome play less essential roles for the overall PPI network. However, once again, this portrays the duplication as a solution with least possible damage without positively suggesting causation.

Figure 3. PPI network differs between yeast chromosomes.

We began with the entire network of all PPIs with experimental annotations in yeast (Methods), and then differentially analyzed major network features: (A) Degree: The number of PPIs per protein (degree) was minimal for the proteins from chrIII (box in red; lowest mean - black dot and lowest median - black line in the box. (B) Betweenness: betweenness (number of times that a node acts as a bridge along the shortest path between two other nodes) was also lowest for chrIII. (C) Average neighbor degree: plotting the average degree for all network neighbors of all proteins on chrIII (i.e. all those proteins in direct PPI with proteins on chrIII), we observed a much less differentiated view. For this network feature, the proteins from chrIII had one of the highest means (black dot), but one of the lowest medians. Clearly, the proteins from the HSR-duplicated chromosome appeared less involved in the yeast network than expected by chance.