The relationship between protein domains and homopeptides in the Plasmodium falciparum proteome

Source data

The UniProt (Boeckmann et al., 2003) reference proteome for Plasmodium falciparum strain 3D7 was downloaded from www.uniprot.org in January 2019. Protein domain annotations for Pf were taken from the Pfam database (El-Gebali et al., 2019). For comparative analysis, three further reference proteomes were downloaded from the same source for the following: another P. falciparum strain (FCH/4), P. yoelii (strain 17XNL), and P. vivax (strain Salvador I).

Annotation

Homopeptides were defined as repetitions of one amino-acid type with a minimum length of three residues (Fig. 1).

Proteins were annotated for intrinsic disorder using the DISOPRED3 and IUPRED2a programs (Dosztanyi et al., 2005; Huntley et al., 2015; Ward et al., 2004). IUPRED2a operates on inputted single sequences, and predicts intrinsic disorder by estimating inter-residue interaction energies (Erdos & Dosztanyi, 2020). It was the best performing single-sequence method for intrinsic disorder annotation, with an area-under-curve (AUC) value of 0.83 for the ROC curve in a recent assessment (Nielsen & Mulder, 2019); DISOPRED3 also had a value of AUC = 0.83 in this assessment, and was one of the best performing methods that use evolutionary information as input. Only regions of predicted disorder ≥30 residues long were considered. A 30-residue length cut-off was used since this is a common threshold or boundary value used in characterizing intrinsically-disordered regions, or in training algorithms for prediction of intrinsic disorder (Atkins et al., 2015). Thus, we used the default “long” parameter choice for the IUPRED2a program. Also, the DISOPRED3 program was run with a 2% expected false positive rate for the algorithm training set, which is in is the author’s recommended parameter range (Jones & Cozzetto, 2015). The results from either intrinsic disorder annotation program were considered separately.

Enrichments & depletions

Because we wish to examine the effects of protein domain structure and intrinsic disorder on the occurrence of homopeptides in Pf, we checked whether there is any deviation from random placement for homopeptides within protein domains and annotated intrinsic disorder. These can be either enrichments relative to background populations or depletions. The background populations were either the whole proteome or the set of protein domain annotations as described below. Enrichments and depletions for homopeptides in protein domains were calculated as depicted (Fig. 1). These were determined for individual amino-acid types in homopeptides using Eq. (1) below for hypergeometric probability, for sampling with replacement: (1) $$P\left(k\right)={\displaystyle \frac{\left(\begin{array}{c}K\\ k\end{array}\right)\left(\begin{array}{c}N-K\\ n-k\end{array}\right)}{\left(\begin{array}{c}N\\ n\end{array}\right)}}$$with the sample counts given by:

k = number of residues in homopeptides in domains of one amino-acid type

n = number of residues in homopeptides in domains

and the background counts given by the quantities:

K = number of residues in homopeptides in the proteome of one amino-acid type

N = number of residues in homopeptides in the proteome

Enrichments/depletions for the amount of homopeptides in specific protein domain types were also calculated with the sample counts given by:

k = number of residues in homopeptides in one domain type

n = number of residues in homopeptides in all domain types

and the background counts given by the quantities:

K = number of residues in one domain type

N = number of residues in all domain-types

Enrichments and depletions for protein domains overlapping or near annotated IDRs were also calculated with the sample counts given by:

k = number of residues in one domain type which near or inside disordered regions

n = number of residues in all domain types which near or inside disordered regions

and the background counts given by the quantities:

K = number of residues in one domain type

N = number of residues in all domain-types

Proximity to IDRs was determined using a 10-residue buffer at either end of the annotated IDRs (Fig. 1).

Gene ontology (GO) category enrichments were also analyzed for specific types of homopeptide enrichment (Huntley et al., 2015). These were calculated by mapping the protein domains onto GO categories, and re-totalling the numbers of residues per GO category rather than per domain.

All enrichments/depletions were calculated using hypergeometric probability with appropriate Bonferroni corrections for multiple hypothesis testing. For example, the Bonferroni correction for enrichments/depletions of homopeptides of individual amino acids in protein domains was P = 0.05/20 = 0.0025, since there are 20 different amino acids being sampled from the same background population.

Propensity

A propensity for homopeptides of a specific amino-acid type to occur in protein domains (P_dom) was calculated as: (2) $${\mathbf{P}}_{\mathbf{d}\mathbf{o}\mathbf{m}}={\text{log}}_{10}[(k/n)/(K/N)]$$

The values of k, n, K and N are as listed above just below Eq. (1). This was calculated for the homopeptide threshold ≥3 residues.

Enrichments and depletions of homopeptides in protein domains

Homopeptides are abundant and pervasive in the Pf proteome, yet it is not clear from a structural perspective which homopeptides are more tolerated in protein domains. We analyzed the preferences of homopeptides of each specific amino-acid type for insertion into protein domains, using three different length thresholds for homopeptides (Table 1). The statistical enrichments/depletions are listed, as well as the fraction of the homopeptide populations for each amino acid, and the propensity (P_dom) of homopeptides of each amino acid for protein domains, calculated as described in “Methods”. For the minimum homopeptide threshold of ≥3 residues length, the most enriched amino acids include the major aliphatic hydrophobic residues valine, isoleucine and leucine, which is to be expected because of the extensive hydrophobic cores of protein domains. Also, the small residues alanine and glycine exhibit highly significant enrichments. For these amino-acid types, the enrichments are only for short homopeptides (of size 3 or 4 residues), since the enrichments disappear for longer homopeptide thresholds (Table 1). Positively-charged homopeptides and other hydrophilic homopeptides are also generally enriched (lysine, arginine, serine, and threonine), while negatively-charged homopeptides are significantly depleted or show no preferences. Lysine homopeptides are the second most abundant in the proteome and are made from AT-rich codons; their relationships with specific protein domains are discussed below. Some amino acids show an enrichment, with a comparable propensity for structural domains (P_dom) as for other amino acids, but these are not significant. Most strikingly though, short asparagine homopeptides are highly significantly depleted within protein domains.

Histograms of homopeptide length also indicate that within protein domains, homopeptides generally lack the longer homopeptide lengths (≥10 residues) that make up the majority of homopeptides outside of protein domains (Fig. 2).

For the longest homopeptide threshold (≥10 residues length), the number of amino-acid types which are comparatively tolerated in protein domains dramatically decreases to one (asparagine; Table 1). Keeping in mind that the enrichment calculations are effectively based on the comparison of different amino-acid types, there should always be at least one enriched amino-acid type unless there are completely no homopeptides at all at a certain threshold. Poly-asparagine homopeptides are depleted in domains until the threshold is extended to 10, which leaves it as the only one existing in domains. The enrichment observed for longer polyasparagine tracts (≥10 residues, Table 1) arises from a small number of specific protein domains that may have a functional linkage for these polyasparagine tracts (Table S1), for example, for specific protein interactions.

The enrichment/depletion results for individual amino-acid types are little affected by the boundary definition of protein domains (i.e., chopping off 3, 5 or 7 residues from the ends of the domains, Table S3), with just some enrichments for glutamine and glutamate becoming significant for these shortened domains. This indicates that these homopeptides significantly occur near the ends of protein domains.

Since lysine and arginine homopeptides are in general significantly enriched in protein domains, and asparagine and aspartate significantly depleted, we examined which individual protein domains are linked to these trends (Table S1). Despite the substantial general depletion of asparagine homopeptides within protein domains, there are 87 individual domains with significant enrichment of asparagine homopeptides, including the low-copy-number Sin-like region and the SacI homology domain; these also stand out when we restrict the analysis to polyasparagine tracts ≥10 residues long (Table S1). The most prominent lysine homopeptide enrichments are for Rifin and PfEMP DBL domains. Specific domains are also linked to hydrophobic or small-residue homopeptides, such as glycine homopeptides arising for ribosomal proteins (Table S1).

Gene ontology enrichments

We examined the enrichments and depletions of GO functional categories associated with homopeptides in protein domains (Table S2). Some top enrichments of GO functional categories for homopeptides in protein domains include: GO:0003899 (DNA-directed 5′-3′ RNA polymerase activity), an enrichment caused by seven different protein domains, and GO:0042578 (phosphoric ester hydrolase activity), which is unique to the SacI homology domain (which is involved in clathrin-mediated endocytosis; three copies in Pf) (Table S2). Inspection of other GO category enrichments indicate that they are also caused by diverse protein domains, for example, GO:0016192 (vesicle-mediated transport) which is linked to homopeptide enrichments in 11 different protein domains, pointing to specific functional significance for homopeptides in the interaction of these proteins. Nonetheless, in general “protein binding” is significantly depleted in the list (Table S2), as are the other high-level “RNA-binding” and “GTP-binding” terms.

The relationship between specific protein domain homopeptides and intrinsic disorder

Intrinsically disordered regions tend to have homopeptides and low-complexity sequences in them (Romero et al., 2001). We surmised that the relationship of different protein domains with homopeptides might be caused by their proximity to or overlap with IDRs of proteins. In general, homopeptides are enriched in the parts of protein domains that are near or in IDRs (Table 2; results for either the DISOPRED3 or IUPred2A program are shown). Also, there is only one individual protein domain type that is significantly depleted in homopeptides near/in IDRs, with the remainder of significant deviations being enrichments (Table 2).

Comparison of trends in other plasmodia

The trends observed for Pf strain 3D7 were validated by analysis of another Pf strain (FCH/4) that was picked from the UniProt reference proteome list (Boeckmann et al., 2003) (Table S4). There is just one small change with enrichments of glutamate homopeptides in protein domains becoming significant (Table S4). Comparisons were also made with proteomes of P. yoelii, a malaria parasite of rodents, and P. vivax, a member of the CVK group of primate-infecting plasmodia (Chaudhry et al., 2018). P. yoelii has an overall approximately even predominance of N and K homopeptides, and P. vivax has predominance of K homopeptides rather than of N homopeptides (Table S4). The depletion of N homopeptides in protein domains is maintained in P. yoelii, but there is no significant depletion/enrichment in P. vivax. K homopeptides also become significantly depleted within protein domains in P. yoelii, despite their similar overall levels to Pf (29% in P. yoelii vs 23% in Pf). The results for homopeptide enrichments in parts of protein domains overlapping IDRs (Table 2) also remain highly significant for these three other Plasmodia proteomes (P-values ~ 0.0).

Homopeptide trends

Homopeptide enrichments within protein domains, such as for hydrophobic (L, I or V) or small (A or G) residues, disappear at longer lengths (≥5 residues). This indicates a limit to their toleration within protein domain cores, for example, because they are not so easily accommodated in regular secondary structures.

We observed a substantial significant relative depletion of short asparagine runs (<10 residues long) in protein domains. Plasmodia have acquired great amounts of N homopeptide tracts during evolution, but statistically these have not been appearing or “landing” within domains. The lack of short intra-domain asparagine runs may be because they interfere with protein folding in some way. For example, they may slow down co-translational protein folding due to a lack of asparaginyl-tRNAs, since levels of asparaginyl-tRNAs in Pf are normal despite the high amounts of asparagine in their coding sequences (Filisetti et al., 2013; Frugier et al., 2010). Thus asparagine homopeptides may be “tRNA sponges” that soak up tRNAs and slow down translation and co-translational folding (Filisetti et al., 2013; Frugier et al., 2010). It is possible that homopeptides, and in particular poly-asparagine homopeptides may make protein domains more prone to misfolding. Although generally slower translation is thought to aid in correct co-translational folding (Waudby, Dobson & Christodoulou, 2019), sometimes faster translation is more desirable through segments that are prone to misfolding (O’Brien, Vendruscolo & Dobson, 2014), or for translational efficiency at buried residue sites or sites that are vulnerable to structurally disruptive mutations (Wang et al., 2015; Zhou, Weems & Wilke, 2009). However, experiments with Pf chaperone Hsp110c, have shown that Pf has cellular mechanisms that are designed to prevent aggregation linked to asparagine tracts (Muralidharan & Goldberg, 2013). A few specific protein domains have enrichment of long polyasparagine tracts. Such tracts may have a specific functional role in these proteins, perhaps for protein or nucleic-acid interaction. Another possibility might be that correct folding of these specific domains is not affected by slow rates of translation, thus N homopeptides can arise in them because of their general abundance in the proteome. Interestingly, the significant depletion of N homopeptides is maintained in the rodent malaria pathogen P. yoelii, but there is an absence of significant depletion/enrichment in the more distantly related P. vivax malaria pathogen from the CVK group. This may indicate that the N homopeptides in the P. vivax protein domains are reduced mainly to those that have functional roles.

Gene ontology enrichments

In the GO enrichments/depletion analysis we see a general trend for depletion of functional categories associated with “binding” (protein binding; RNA binding; GTP binding; ion binding). This suggests that homopeptides may be selected against in structured interaction interfaces, perhaps since they introduce a lack of interaction specificity, or increase the likelihood of off-target binding. Also, the GO results indicate that homopeptide occurrences may be useful information for the discrimination of protein function from the analysis of sequences (Huntley et al., 2014, 2015; Jiang et al., 2016; Le et al., 2019; Le, Yapp & Yeh, 2019; Mutowo-Meullenet et al., 2013).

Intrinsic disorder

We surmise that homopeptides in protein domains are enriched in the parts of the domains near or in IDRs because IDRs generally have more tolerance for insertions/deletions, and are the main determinants of changes in protein length over evolution (Light et al., 2013). Also, our results indicate that the parts of protein domains that can become intrinsically disordered are enriched in homopeptides relative to other domain parts. This may be an important part of encrypting their ability to transition structurally (Narasumani & Harrison, 2015).