Deep conservation of prion-like composition in the eukaryotic prion-former Pub1/Tia1 family and its relatives

Data

Sets of orthologs for the following protein families were downloaded from OrthoDB database version 10.1 at the Eukaryota level (Kriventseva et al., 2019): Pub1/Tia1 (OrthoDB group 1066369at2579), Nam8/Ngr1 (group 775799at2579) and Pab1 (group 1027234at2759). These data also contain paralogs. In addition, lists of reciprocal-best-hits (RBH) orthologs were made for the same proteins across sets of reference eukaryotic proteomes downloaded from UniProt version 2019_1 (UniProt Cosortium, 2019). These were cross-referenced with the OrthoDB sets to determine the subset of OrthoDB annotations that are such RBH orthologs. Also, only OrthoDB proteins that are from the UniProt reference proteomes were considered for further analysis. FASTA-format files of the Pub1/Tia1/Tiar and Nam8/Ngr1 families of proteins are provided as Supplemental Data (Datas S1 and S2).

Multiple sequence alignment and phylogenetic trees

Multiple sequence alignments were constructed using KMAD (Lange, Wyrwicz & Vriend, 2016), using default parameters. KMAD is designed to handle sequences with intrinsic disorder, a trait expected characteristically of prion-like proteins (Harbi & Harrison, 2014b). Phylogenetic trees were made using PHYML with Bayesian information criterion and aBayes branch support (Guindon et al., 2005), with the exception of a tree including extra paralogs described below. The aBayes branch support is a Bayesian-like transformation of the standard approximate likelihood ratio test implemented in PhyML, that is very fast and has high statistical power (Anisimova et al., 2011).

To construct a phylogenetic tree with extra paralogs, we used a stepwise procedure of tree construction. Examination of BLASTP (Altschul et al., 1997) output indicates that Pab1 and its orthologs were the next most similar RRM-domain-containing proteins in yeast that are widely conserved across eukaryotes (Table S3). Pab1 is a polyadenylate-binding protein that is an important interactor at various stages of messenger mRNA stability, biogenesis and translation. We used similarities to Pab1 to help us to filter for further paralogs to the Pub1/Tia1 and Nam8/Ngr1 proteins as follows. BLASTP was used to compare each ortholog from the Pub1/Tia1, Nam8/Ngr1 and Pab1 protein sets against their own proteomes, and any significant matches (with e-value ≤1E−04) against the majority (>50% length) of a proteome sequence that matched a Pub1/Tia1 or Nam8/Ngr1 sequence in preference to a Pab1 sequence were labelled as potential paralogs of Pub1/Tia1 or Nam8/Ngr1, formed more recently in specific eukaryotic clades. Then, a phylogenetic tree was constructed using the RaxML programme (using the PROTGAMMAJTTF mutation parameters and the ‘–f a’ flag for rapid bootstrap analysis and 100 bootstrap replicates) (Stamatakis, 2015) with the Pub1/Tia1, Nam8/Ngr1 and Pab1 ortholog sets plus the additional potential paralogs proteins as input. The tree was pruned at an appropriate node to remove all of the Pab1 sequences and associated paralogs to leave just those paralogs that were formed more recently than the last common ancestor of the Pub1/Tia1 and Nam8 proteins.

Annotation of RNA-binding (RRM) domains

Positions of RRM RNA-binding domains were taken from annotations in InterPro version 66.0 from the database Pfam (El-Gebali et al., 2019; Mitchell et al., 2019). We checked these annotations against those also for the PROSITE and SMART databases (Letunic & Bork, 2018; Sigrist et al., 2010), and they differ by <1% of extra or fewer domain annotations, which gave us confidence to proceed with the Pfam set. Since there were some OrthoDB sequences that do not have Pfam annotations, and other annotations may be incomplete, we generated additional RRM domain annotations. Firstly, the sequences for the RRM domain annotations were extracted and then compared against the ASTRAL SCOPe database version 2.07 (Fox, Brenner & Chandonia, 2014) using BLASTP (Altschul et al., 1997) with e-value threshold 1E−04, to pull out the identities of protein domain structures that are for RRM domains. Then the ASTRAL SCOPe protein domains plus the sequence fragments for the RRM-domain annotations were then formed into a BLASTP database to compare against the OrthoDB sequences to augment the existing list of RRM-domain annotations (again using e-value threshold 1E−04). These were then reduced for overlap by sorting them in decreasing order of domain length and progressively flagging overlappers further down the list for deletion.

Annotation of prion-like domains and intrinsic disorder

Prion-like domains were annotated using the PLAAC programme with default parameters (Lancaster et al., 2014). Prion-like domains were counted using two thresholds for the PLAAC log likelihood ratio (LLR) score: >0.0 and ≥15.0. Any score >0.0 indicates some predicted propensity to form a prion-like domain; the second threshold of 15.0 was picked since the lowest LLR value for a known fungal prion-forming domain that was picked oput for experimental analysis is ~16.0 in the fission yeast prion-former Ctr4 (Sideri et al., 2017). The programme fLPS was used to detect compositional bias for glutamine and asparagine residues, which is a bias characteristic of the substantial majority of the known prion-forming domains in budding yeast and other organisms (Harrison, 2006; Harrison, 2017). Expected fractional frequencies = 0.05 for both glutamine and asparagine were used in running fLPS. fLPS Binomial P-values indicate the degree of compositional bias, and here we picked the most biased subsequence for either Q or N or Q + N for each sequence. The log10 of these P-values is investigated here for all the data analysis.

To examine for multiple prion-like domains, the sequences were split up into fragments with the ends delimited by RRM domains and other protein domains (plus a 10 amino-acid-residue buffer added onto both ends), which were then input into the PLAAC programme. Prion-forming N/Q-rich domains are highly intrinsically disordered (Harbi & Harrison, 2014b). Intrinsic disorder was annotated using the IUPRED2a and DISOPRED3 programmes with default parameters (Dosztanyi, 2018; Ward et al., 2004). The total percentage of intrinsic disorder was calculated for each whole protein sequence, since there is not an intrinsic disorder score for a specific tract that can be extracted from the output of the disorder annotation programmes. Intrinsic disorder assignments by IUPRED2a and DISOPRED3 are strongly correlated (R = 0.43 for Pub1, R = 0.47 for Nam8, P-values < 1E−20).

Evolution within Saccharomycetes

In Saccharomycetes, the three main proteins from the protein families under study are Pub1, Nam8 and Ngr1. Preliminary analysis of phylogenetic trees indicated that Ngr1 formed by duplication from Nam8 in the last common ancestor of Saccharomycetes. Firstly, we assessed the conservation of prion-like composition across the Saccharomycetes clade for Pub1, and for its paralogs Nam8 and Ngr1. Prion-like composition is largely maintained for all three paralogs across this clade (Fig. 1). The substantial majority of Pub1 orthologs have PLAAC prion-like region scores ≥15.0 (53/65, 82%), and almost all >0.0 (61/65, 94%); less so for Nam8 and Ngr1 (73/114, 64% ≥15.0; 113/114, 99% >0.0).

Evolution across eukaryotes

Furthermore, this prion-like composition is deeply conserved across large diverse clades within the whole eukaryotic domain for the Pub1/Tia1 and Nam8/Ngr1 families (Fig. 2). Overall, 90% of sequences have PLAAC LLR scores >0.0, with 58% ≥15.0. Specifically for Pub1, 84% have scores >0.0 and 60% have scores ≥15.0. Prion-forming N/Q-rich domains are highly intrinsically disordered (Harbi & Harrison, 2014b). There is some degree of correlation of LLR score with overall percentage annotated intrinsic disorder in the sequences, particularly for Pub1 (P < 2E−16; Fig. 3B), but some clades on the phylogenetic tree are observed that have low prion-like character and relatively high intrinsic disorder (Fig. 2).

Prion-like character as judged by the PLAAC programme is maintained in a similar fashion for both the Pub1/Tia1 and Nam8/Ngr1 families (Fig. 3A). PLAAC LLR scores are strongly correlated with the fLPS compositional biases for Q and N residues (R = 0.70 for Nam8, and R = 0.52 for Pub1; P < 1E−20), but substantial clades can be seen on the phylogenetic tree that have low fLPS Q/N residue bias, but high PLAAC LLR scores, such as some of the fungal clades clustering near Saccharomycetes.

The larger tree with additional paralogs whose construction is described in ‘Methods’, indicates that no further major paralogs have formed during eukaryotic speciation, since these would appear as intermittent larger blocks in the tree figure (Fig. S1). Such paralogs have however formed sporadically in evolutionarily more recent clades; some of these appear to have lost the prion-like domains, or intrinsic disorder generally (Fig. S1).

Multiple prion-like domains occur in several diverse clades

Prion-forming proteins are usually characterised as having one distinct prion-forming area (Harbi & Harrison, 2014a; Harbi et al., 2012). Likewise, prion-like proteins are usually annotated by picking out a single region with the highest adjudged propensity for forming prions; the algorithms that have been derived for annotating prion-like proteins are all based on this principle (Harrison, 2017; Lancaster et al., 2014; Ross et al., 2013; Zambrano et al., 2015). However, it is possible that multiple prion-forming regions can occur within the same protein chain. We decided to check for this in the protein families under study here, since from spot-checking they often contain more than one annotated intrinsically disordered region delimited by the structured protein domains (specifically here most often the RRM RNA-binding domain). We parsed each protein sequence into fragments delimited by the positions of protein domains and annotated these fragments for prion-like composition, as described in ‘Methods’. We find that it is not uncommon to have multiple prion-like regions in one protein chain, particulary in the Nam8/Ngr1 family (11% ≥2 prion-like regions for the Nam8/Ngr1 family and 1% for the Pub1/Tia1 family, Figs. 4B and 4C). Indeed, there are specific large clades that maintain two prion-like regions (Fig. 2). Only with a lower PLAAC LLR score threshold of >0.0 does the Pub1/Tia1 family display a sizeable minority of protein sequences with multiple prion-like regions (10% with ≥2 prion-like domains; Figs. 2, 4B and 4C). A large fraction of these extra domains with prion-like composition are internal in the sequence (46% for Pub1/Tia1 and 29% for Nam8/Ngr1 for PLAAC LLR threshold >0.0).

The number of RNA-binding domains is largely fixed at three

The number of RNA-binding RRM domains is largely fixed at three for both the Nam8/Ngr1 and Pub1/Tia1 families of proteins (70% having three RRM domains, with just 17 sequences having >3), but proteins with just one or two RRM domains occur frequently in some specific clades (Figs. 2 and 4A).

Evolution across eukaryotes

Analysis of the evolution across eukaryotes of prion-like composition for the Pub1/Tia1 and Nam8 families, as judged by the PLAAC programme, indicated that such composition is deeply conserved. We can therefore surmise that this composition is important for the protein function. However there are some clades that: (i) have low prion-like character and maintain relatively high annotated intrinsic disorder; (ii) have high prion-like character as judged by PLAAC, but low bias for N and Q residues, as judged by the fLPS algorithm. Sequences from such clades may be good candidates for checking for alternative compositions for prion-like domains in Pub1 that fulfil either its prion-forming ability or its role in stress granule formation. Sequences of type (ii) above would particularly be interesting candidates for further experimental study for prion-like aggregation, since the prion-like composition is made up more from non-Q + N residues that contribute to prion-forming biases in the known S. cerevisiae cases.

Multiple prion-like domains occur in several diverse clades

The occurrence of multiple prion-like domains in proteins raises the possibility of competing prion-like aggregation by the different prion-like domains within the same protein sequence. However, a large fraction of these extra domains with prion-like composition are internal in the sequence, indicating that they would have greater difficulty forming amyloid fibrils. Also, multiple prion-like domains might affect the character of intracellular condensates, that is as either solid-like or gel-like, perhaps making them more solid-like because of the increased prion-like composition relative to other types of intrinsic disorder propensity in the equivalent regions in other orthologs (Kroschwald et al., 2018).

The number of RNA-binding domains is largely fixed at three

We found very strong conservation of a three-domain RRM RNA-binding domain pattern in both the Pub1/Tia1 and Nam8 families. Proteins with just one or two RRM domains occur frequently in a conserved pattern in some smaller clades, which indicates that these are not due to annotation errors and are real evolutionary trends (Figs. 2 and 4A). There are a handful of sequences with four or more RRM domains, but it is unclear whether these may be annotation errors, due to their sporadic occurrence (Fig. 2). One notable cluster that has either as one or two RRM domains (upper left corner of Fig. 2, near clade labelled with blue band), consists mainly of vertebrate Nam8 orthologs that are assigned function in selenocysteine biosynthesis, including TRNAU1AP (transfer RNA selenocysteine 1—associated protein 1) the human RBH ortholog of S. cerevisiae Nam8, a protein that has essential functions in both early and later steps of selenocysteine biosynthesis (Small-Howard et al., 2006). Although no human disorders have been directly linked to mutations in the encoding gene of TRNAU1AP, faults in selenocysteine biosynthesis have been linked to complex disorders of metabolism and the development of muscles, skin, the nervous system and immune system (Schweizer & Fradejas-Villar, 2016). These proteins also tend to have lower annotated intrinsic disorder content and lower prion-like composition (Fig. 2).