Signatures of pleiotropy, economy and convergent evolution in a domain-resolved map of human-virus protein-protein interaction networks

Abstract

A central challenge in host-pathogen systems biology is the elucidation of general, systems-level principles that distinguish host-pathogen interactions from within-host interactions. Current analyses of host-pathogen and within-host protein-protein interaction networks are largely limited by their resolution, treating proteins as nodes and interactions as edges. Here, we construct a domain-resolved map of human-virus and within-human protein-protein interaction networks by annotating protein interactions with high-coverage, high-accuracy, domain-centric interaction mechanisms: (1) domain-domain interactions, in which a domain in one protein binds to a domain in a second protein, and (2) domain-motif interactions, in which a domain in one protein binds to a short, linear peptide motif in a second protein. Analysis of these domain-resolved networks reveals, for the first time, significant mechanistic differences between virus-human and within-human interactions at the resolution of single domains. While human proteins tend to compete with each other for domain binding sites by means of sequence similarity, viral proteins tend to compete with human proteins for domain binding sites in the absence of sequence similarity. Independent of their previously established preference for targeting human protein hubs, viral proteins also preferentially target human proteins containing linear motif-binding domains. Compared to human proteins, viral proteins participate in more domain-motif interactions, target more unique linear motif-binding domains per residue, and contain more unique linear motifs per residue. Together, these results suggest that viruses surmount genome size constraints by convergently evolving multiple short linear motifs in order to effectively mimic, hijack, and manipulate complex host processes for their survival. Our domain-resolved analyses reveal unique signatures of pleiotropy, economy, and convergent evolution in viral-host interactions that are otherwise hidden in the traditional binary network, highlighting the power and necessity of high-resolution approaches in host-pathogen systems biology.

Figure 1. Domain-centric mechanisms of host-virus protein-protein interaction.

(A) A domain-domain interaction (DDI) example : a cyclin domain-containing protein from Saimiriine herpesvirus 2 (red) targets a human CDK6 kinase domain (white). (B) A domain-motif interaction (DMI) example : human retinoblastoma-associated protein (white) contains a linear motif-binding (LMB) domain which recognizes the peptide motif LxCxE (red) in the human papillomavirus E7 protein.

Figure 2. Coverage of human-virus protein-protein interaction network by domain-centric interaction mechanisms.

Fractions of endogenous and exogenous PPIs that can be assigned to different domain-centric interaction mechanisms (DDIs and DMIs). Each mechanism is illustrated using the symbols at the left, with the percentage of interactions described by that mechanism given below. An interaction may be described by more than one interaction mechanism.

Figure 3. Domain-resolved interactions are of high quality.

Endogenous (blue) and exogenous (red) PPIs that are supported by a known DDI or DMI are enriched for confirmed interactions (i.e., interactions reported by at least two publications). Endogenous and exogenous PPIs lacking the support of domain-centric mechanisms are depleted for confirmed interactions. Error bars reflect the standard error based on 1,000 rounds of bootstrap resampling.

Figure 4. Binding site mimicry evolves differently in virus and host proteins.

Two proteins can participate in DDIs with a common target by binding to: (A) different domains in the target; (B) the same domain in the target using different interaction domains; or, (C) the same domain in the target using the same interaction domain. (D) Viral proteins are significantly less likely than human proteins to bind to the same domain of a human protein by means of domain sequence similarity to an endogenous binding partner (Fisher's exact test, two-tailed P<10⁻¹⁰).

Figure 5. Viruses tend to target human proteins containing linear motif-binding (LMB) domains.

We compared domain composition of generic human proteins in the endogenous network (blue) to human proteins targeted by viruses (“viral targets”, red). The vertical axis indicates the fraction of proteins in a group containing LMB domains; the horizontal axis indicates the fraction of proteins containing non-LMB domains. Relative to generic human proteins, human proteins targeted by viruses are significantly more likely to contain an LMB domain (vertical axis; Fisher's exact test, two-tailed P<10⁻¹⁵), and are slightly less likely to contain non-LMB domains (horizontal axis; P = 0.012).

Figure 6. Preferential targeting of LMB domains by viruses is independent of host protein degree.

We partitioned human proteins into LMB domain-containing proteins (filled green points), and LMB domain-free proteins (open black points). We further divided proteins according to endogenous degree. 92% of all human proteins in the network fall into one of the degree 1–20 bins; each bin contains at least 20 proteins. The probability of being a viral target increases with degree for both LMB domain-containing and LMB domain-free proteins. For a given degree, LMB domain-containing proteins are more likely to be viral targets than LMB domain-free proteins.

Figure 7. Viral proteins have a higher fraction of domain-motif interactions (DMIs) than human proteins.

Fraction of DMIs out of the total number of interactions per protein tend to be higher in viral proteins (red) than human proteins (blue) in (A) all interactions in the network (permutation test, two-tailed P = 0.047), and (B) confirmed interactions (P = 0.018). Error bars reflect the standard error.

Figure 8. Viral proteins target LMB domains at greater density than human proteins.

Viral proteins (red) target significantly more unique LMB domains per residue than human proteins (blue) in (A) all interactions in the network (permutation test, two-tailed P = 0.012), and an independently generated human-virus PPI network , (P = 0.049). (B) Viral proteins also have significantly more unique LMB motif types per residue than human proteins (P<0.001). Error bars reflect the standard error.