The research team focused on five large double-stranded DNA viruses: vaccinia virus, the type species of mammal-infecting poxviruses, and four human herpesviruses, namely herpes simplex virus 1, Epstein–Barr virus, human cytomegalovirus and Kaposi’s sarcoma-associated herpesvirus. These viruses are known to encode numerous host-like domains acquired through horizontal gene transfer during their long evolutionary history. To find them systematically, the researchers predicted the three-dimensional structures of 540 viral proteins using AlphaFold2 and then employed Foldseek, a fast structural homology search tool, to identify viral domains that structurally resemble human proteins. After a series of filtering steps designed to eliminate spurious matches between low-complexity folds and partially homologous structures, the pipeline yielded 7,593 human–viral structural homolog pairs derived from 144 distinct viral proteins.

The scale of this mimicry is striking. In herpes simplex virus 1, structurally host-like domains account for roughly 11 percent of the encoded proteome, while in vaccinia virus the figure rises to 31 percent. Notably, despite their structural resemblance, the sequence similarity between viral mimics and their human counterparts is often low, averaging just 28.4 percent within the homologous region and 6.3 percent across full-length proteins. This makes them imperfect mimics: similar enough in fold to bind the same partners, yet different enough in sequence that the interacting surfaces are not identical. The researchers then asked which of these structural homologs actually function in a similar way, defining functional similarity as the sharing of at least one common human interaction partner, which they termed a mutual host target.

By overlaying experimentally characterized human–virus protein interaction networks from databases such as HVIDB with within-human interaction maps from STRING, the team identified 645 triads consisting of a viral mimicking protein, a host-mimicked protein and a mutual host target. These triads involve 229 functionally mimicked human proteins, 40 viral mimicking proteins and 118 mutual host targets. One illustrative example is the KSHV protein ORF72, which structurally resembles the human cyclin CCNO, with both proteins predicted to interact with the human kinase CDK3 through a similar binding region. The set of functionally mimicked proteins, those with direct evidence of shared interactions, became the focus of the subsequent evolutionary analysis.

To measure evolutionary pressure, the researchers computed dN/dS, the ratio of nonsynonymous to synonymous substitutions, for more than 9,000 human genes across ten primate species, including chimpanzee, gorilla, orangutan, macaques and mouse lemur. The results were unambiguous. Human proteins known to bind viral proteins are significantly more conserved than the proteome at large, confirming earlier findings. But the proteins entangled in mimicry are conserved to an even greater degree. Mutual host targets, the human proteins contacted by both the viral mimic and the host-mimicked protein, were the most conserved viral-binding group tested, and functionally mimicked proteins were significantly more conserved than the broader set of structural homologs. When the team searched for signatures of pervasive positive selection using codon-based models in PAML, the picture was equally striking: proteins targeted by mimics showed virtually no positively selected genes at most statistical thresholds, in sharp contrast to the modest but detectable positive selection seen among viral-binding proteins in general.

What explains this exceptional conservation? The researchers found that mimicry-related host proteins sit at crowded hubs of the cellular interaction network. Mutual targets and functionally mimicked proteins have significantly more interaction partners than average human proteins, and targets of mimicry are also more highly expressed across tissues than other viral-binding proteins. Gene ontology analysis showed that these proteins are enriched in antiviral pathways such as cell death, cytokine response and inflammatory signaling, as well as in processes frequently manipulated by viruses, including Golgi regulation, the G1/S cell cycle transition and signal transduction. Within these shared functional categories, mimicked and target proteins were more conserved than their non-mimicked counterparts in the majority of cases, 75.6 percent for mimicked proteins and 63.2 percent for targets. The implication is that essential cellular roles constrain these proteins, leaving the host little room to mutate away from viral mimics without damaging its own interactome.

The analysis then zoomed in to the level of individual amino acids. Using AlphaFold-Multimer to predict complexes between mimicked and target proteins, the team identified interface residues and compared their evolutionary rates with those of buried core residues and exposed surface residues. As expected, interfaces and cores evolved more slowly than surfaces. More revealing was the comparison between two kinds of interface: mimicry interfaces, the residues that physically contact the mutual target, and general interfaces, predicted binding sites for unrelated proteins using the deep learning tool ScanNet and the SVM-based method ISPRED4. In both mimicked and target proteins, mimicry interfaces were significantly more conserved than general interfaces, indicating that the residues exploited by viral mimics are under especially strong purifying selection.

Because viral mimics are imperfect copies, their binding surfaces on the host target are not identical to those of the host-mimicked protein. The researchers therefore subdivided target interfaces into residues contacted by the host-mimicked protein, residues contacted only by the viral mimic, and residues contacted by both. The mutual residues were the most conserved, while residues exclusively touched by the viral protein evolved significantly faster. This pattern hints at a possible escape route: host surfaces used only by the virus, and not required for essential cellular interactions, may tolerate more change, potentially allowing the host to weaken viral interactions selectively. Yet even these faster-evolving regions showed no enrichment for positively selected residues, so the authors caution that this remains a hypothesis rather than demonstrated host escape. The team also compared perfect and imperfect mimicry, examining short linear motifs in disordered viral regions that exactly copy host motifs. Host proteins targeted by these perfect motif mimics, 89 in total, were the most conserved group in the entire study, more constrained even than targets of imperfect domain mimicry, consistent with the idea that perfect mimicry leaves the host no way to discriminate foe from friend.

On the viral side, the picture was more complex. Comparing orthologs across orthopoxviruses, simplexviruses, cytomegaloviruses and rhadinoviruses, the researchers found no consistent difference in evolutionary rate between mimicking and non-mimicking viral proteins, a result they attribute to technical limitations such as small protein sets and limited rate ranges, as well as biological factors including the multifunctional nature of viral proteins. Crucially, however, viral mimics were not enriched for positively selected genes; if anything, the fraction of such genes was lower in mimics than in other viral proteins across all four lineages. In the few mimicking–target complexes that could be modeled with high confidence, interface residues of viral mimics were more conserved than surface residues, mirroring the host-side pattern. The single positively selected residue detected in a viral mimic, asparagine 15 in the vaccinia Bcl-2-like protein OPG035, sits far from the interface with its human target, pointing away from any adaptive remodeling of the binding surface.

The overall message is a reframing of how host–virus conflict plays out at the molecular level. While arms races do occur, particularly in immune and restriction factors, the interactions mediated by mimicry appear governed by constraint rather than escalation. Viruses have exploited the host’s inability to escape: by targeting highly connected, highly expressed, functionally essential proteins through surfaces the host must preserve, mimics anchor themselves to interaction networks that evolution cannot easily redeploy. The conserved nature of these interfaces may also carry practical value, the authors suggest, since stable binding regions could serve as targets for antiviral drugs designed to disrupt mimicry-mediated interactions. And because these principles govern the conservation and evolvability of host–virus interaction networks, they may help predict which viruses can switch hosts and which barriers stand in the way of zoonotic emergence. The study’s data and code are openly available, offering a resource for the growing field of structural virology.

Subject of Research: Evolutionary dynamics of viral molecular mimicry and host protein interactions in large DNA viruses

Article Title: The evolutionary dynamics between viral mimics and host proteins

Article References: Fuchs, R., Schor, O., Naim, B., Tussia-Cohen, D., Mozzi, A., Forni, D., Friedman, S., Haggai, Z., Sironi, M., & Hagai, T. (2026). The evolutionary dynamics between viral mimics and host proteins. Molecular Systems Biology, 22(6), 902-927. https://doi.org/10.1038/s44320-026-00200-1

Image Credits: AI Generated

DOI: 10.1038/s44320-026-00200-1

Keywords: molecular mimicry, virus-host interactions, viral evolution, protein-protein interactions, AlphaFold, dN/dS, positive selection, herpesvirus, poxvirus, structural biology, host-pathogen coevolution, protein interfaces