Mapping the interactome between pathogen proteins and host proteins to reveal new points of therapeutic intervention
Every successful infection depends on a pathogen's ability to manipulate the internal machinery of a host cell. That manipulation happens through direct, physical protein-protein interactions — a pathogen protein binding, blocking, hijacking, or redirecting a host protein. Systematically cataloguing these interactions produces a host-pathogen interaction network (interactome): a map of exactly which pathogen proteins touch which host proteins, and where the vulnerable seams of the infection process lie.
Two complementary experimental strategies dominate host-pathogen interactome mapping:
Yeast two-hybrid (Y2H) screening: • Each pathogen protein (as "bait") is tested pairwise against a library of host proteins ("prey") fused to transcription factor halves • Physical interaction reconstitutes a functional transcription factor, activating a reporter gene • High-throughput: thousands of pairwise tests per week • Detects direct, binary interactions but misses complexes requiring post-translational modification or membrane context
Affinity purification-mass spectrometry (AP-MS): • Each pathogen protein is expressed in human cells with an epitope tag • Tagged protein is pulled down along with anything physically bound to it • Mass spectrometry identifies every co-purifying host protein • Captures multi-protein complexes and interactions that depend on native cellular context (post-translational modification, membrane localization, complex formation) • Cannot distinguish direct binding from indirect co-complex membership without follow-up
Proximity labeling (BioID, APEX): • Pathogen protein fused to a promiscuous biotin ligase or peroxidase • Labels any host protein within a few nanometers, whether stably bound or transiently contacting • Reveals interactions too weak or transient for standard pulldown
Each method has a different false-positive/false-negative profile, so high-confidence interactome maps typically combine two or more approaches and require reproducibility across biological replicates before an interaction is accepted into the network.
The landmark SARS-CoV-2–human interactome (Gordon et al., Nature 2020) used AP-MS on all 29 viral proteins expressed individually in human cells, identifying 332 high-confidence interactions with human proteins — a map that immediately pointed toward existing drugs with activity against several of the implicated host factors.
A raw list of pairwise interactions becomes genuinely useful only once it is represented and analyzed as a network graph:
• Nodes: individual proteins — pathogen proteins on one side, host proteins on the other • Edges: an experimentally validated physical interaction connecting a pathogen node to a host node • Bipartite structure: in the simplest representation, edges run only between the two protein classes, though host-host and pathogen-pathogen interactions can be layered in from existing databases (e.g., BioGRID, STRING, IntAct) • Confidence weighting: each edge can be annotated with the number of independent experiments or methods that support it, giving a quantitative confidence score rather than a binary present/absent call
Once represented this way, the network becomes an object that graph-theoretic tools can analyze — degree distribution, clustering, shortest paths, centrality — turning a wet-lab dataset into a map that computational biologists and structural biologists can mine for mechanistic hypotheses and, ultimately, for candidate points of therapeutic intervention.
Not every node in an interaction network is equally important. Network analysis reveals that connectivity is highly uneven: a small number of proteins accumulate a disproportionate share of interactions. These "hub" proteins — pathogen proteins that reach into many host pathways, or host proteins recruited by many different pathogen factors — are statistically and biologically the nodes most likely to sit at chokepoints of the infection process.
Several complementary network metrics are used to formally identify hub proteins:
Degree centrality: • Simply, the number of edges connected to a node • A pathogen protein with a high degree interacts with many host proteins — a candidate multi-functional virulence factor • A host protein with a high degree is targeted by many different pathogen proteins — a candidate common vulnerability the pathogen has convergently evolved to exploit
Betweenness centrality: • Measures how often a node lies on the shortest path between other pairs of nodes • High-betweenness proteins act as bridges between otherwise separate network modules or host pathways • Removing a high-betweenness node fragments the network far more than removing a random node
Eigenvector / PageRank centrality: • Weights a node's importance by the importance of its neighbors, not just raw connection count • Distinguishes proteins connected to other well-connected proteins from proteins with many low-value connections
Module/community detection: • Clustering algorithms (e.g., Louvain, MCL) group densely interconnected sub-networks into functional modules • Hub proteins frequently sit at the interface between modules, coordinating cross-talk between otherwise distinct host processes (e.g., linking innate immune signaling to vesicular trafficking)
In host-pathogen networks specifically, host hub proteins recruited by unrelated pathogens (viruses, bacteria, parasites) independently are especially interesting — convergent targeting by evolutionarily distant pathogens is strong evidence that the host protein sits at a genuinely rate-limiting step of infection, making it a high-value target across multiple diseases at once.
Hub identification is fundamentally a statistical exercise, and its reliability scales with network size and density:
• Sparse networks (tens of interactions): degree differences between nodes are dominated by sampling noise; a protein may appear highly connected simply because it was over-represented in the screening bait set • Moderate networks (~100–200 interactions): recognizable degree distribution begins to emerge, and the highest-degree nodes usually separate from background • Rich networks (300+ interactions, ideally combining multiple experimental methods and replicate screens): hub identification becomes statistically robust, module boundaries stabilize, and betweenness/eigenvector measures agree with simple degree counts — convergent evidence across metrics is the strongest signal that a candidate hub is real rather than an artifact of one screen
This is why interactome projects invest heavily in scaling up interaction mapping (more bait proteins, more replicate screens, multiple orthogonal methods) before drawing strong conclusions about which proteins are the true hubs — small, noisy networks can be actively misleading about which node matters most.
Connectivity alone does not make an interaction a good drug target — a hub protein can sit at the center of many contacts that are functionally incidental, tolerating disruption without meaningfully affecting the pathogen. What matters for intervention is essentiality: whether disrupting a specific interaction actually degrades pathogen replication, survival, or virulence. Essentiality screening layers functional evidence onto the topological map, separating structurally prominent interactions from functionally critical ones.
Essentiality is established through orthogonal functional experiments layered on top of the physical interaction map:
Genome-wide CRISPR knockout screens: • Host cells are individually depleted of each candidate host protein and then challenged with the pathogen • Cells lacking a truly essential host factor become resistant to infection or show reduced pathogen replication • Cross-referencing screen hits against the interactome pinpoints host proteins that are both physically contacted by the pathogen and functionally required for infection
siRNA/shRNA knockdown panels: • Partial (rather than complete) loss-of-function, useful for essential host genes where full knockout is lethal to the cell itself • Dose-dependent knockdown can reveal a graded relationship between host protein abundance and pathogen fitness
Pathogen-side essentiality: • Conditional or inducible knockdown/knockout of the pathogen protein itself (where genetic tools exist) tests whether the pathogen factor is required for virulence, independent of which host partner it engages • Domain-mapping and point-mutant rescue experiments can isolate exactly which residues mediate the interaction, separating the essential contact from other functions of the same protein
An interaction being "essential" in a single cell-culture assay is only a first pass — real prioritization requires confirming the effect holds across multiple pathogen strains, cell types, and ideally in an animal infection model, since interactions essential in a simplified system sometimes prove redundant once the full complexity of an organism is engaged.
Once essentiality data exists, candidate interactions are typically ranked using a composite score combining several lines of evidence:
• Topological importance (hub status, betweenness centrality) from Stage 2 • Functional essentiality (magnitude of pathogen fitness loss upon disruption) • Structural tractability (is there a defined, druggable binding interface with a resolved or modelable structure?) • Conservation (is the interaction conserved across pathogen strains or even related pathogen species, widening the eventual therapeutic reach?) • Host protein safety profile (does depleting or inhibiting the host protein carry acceptable on-target toxicity risk?)
Interactions scoring well across all five axes graduate from the interactome map into active structure-based drug design or repurposing campaigns — the transition point where computational network biology hands off to medicinal chemistry.
Every validated essential interaction offers two distinct intervention strategies, because an interaction has two ends: the pathogen protein can be targeted directly, or the host protein it exploits can be targeted instead. The two strategies carry very different pharmacological and evolutionary trade-offs, and the interaction network is exactly the tool needed to decide which side of a given interaction is the smarter target.
Targeting the pathogen protein directly is the traditional antimicrobial and antiviral paradigm:
Advantages: • Pathogen proteins are frequently absent from host cells, offering a wide therapeutic window with minimal on-target host toxicity • Decades of precedent and regulatory familiarity (protease inhibitors, polymerase inhibitors, cell-wall synthesis inhibitors) • Direct mechanistic link between drug action and pathogen inhibition, simplifying pharmacodynamic modeling
Limitations: • The pathogen protein is under direct selective pressure from the drug — mutations that reduce drug binding while preserving protein function are actively selected for during treatment • Resistance can emerge within a single course of treatment, especially for pathogens with high replication rates and error-prone polymerases (RNA viruses, HIV) • Narrow spectrum: a drug targeting one pathogen's specific protein sequence typically will not work against related but genetically divergent pathogens
Targeting the host protein that the pathogen exploits is a newer, complementary strategy enabled directly by interactome mapping:
Advantages: • The host protein does not mutate in response to pathogen-specific selective pressure — the pathogen cannot "evolve around" a host-directed drug by mutating its own genome, since the drug never touches the pathogen • A single host-directed therapy can potentially work against multiple pathogens that convergently exploit the same host factor (a genuinely broad-spectrum mechanism) • Existing, already-characterized drugs targeting the host protein for unrelated indications become immediate repurposing candidates
Limitations: • The host protein usually has native cellular functions, so inhibiting it risks on-target toxicity to the patient — the therapeutic window can be narrower and must be carefully established • The pathogen may still adapt indirectly, by evolving to rely on a redundant host pathway instead, though this is a slower and less certain escape route than a simple point mutation • Regulatory and safety evaluation is less standardized than for classical antimicrobials
In practice, the interaction network lets researchers weigh both options for every essential interaction and, where resources allow, pursue host-directed and pathogen-directed candidates in parallel — reducing the chance that resistance to one approach leaves no therapeutic option remaining.
Because pathogen-directed drugs apply direct selection pressure to the very protein they target, resistant variants can be selected for during treatment itself. Host-directed therapy sidesteps this specific escape route entirely, since the drug's target is never under pathogen-driven selection — though it introduces its own distinct safety and tolerability considerations that must be evaluated on a case-by-case basis.
The final payoff of host-pathogen interactome mapping is translational: once a validated, essential, structurally tractable interaction has been identified and a targeting strategy chosen, the network becomes a search tool. It can point toward existing drugs already known to modulate the implicated protein — dramatically shortcutting the path to a testable therapy — or it can define the starting point for an entirely novel drug discovery campaign.
When a host protein newly implicated by the interaction network turns out to already be the known target of an approved or investigational drug, the path to a testable therapy shortens dramatically:
• Cross-reference step: the validated host protein (or its close paralogs) is searched against pharmacological databases (DrugBank, ChEMBL, the Connectivity Map) cataloguing which compounds are known to bind or modulate it • Immediate advantage: an approved drug already has established human pharmacokinetics, safety, and dosing data — years of development are effectively skipped • Rapid experimental validation: candidate repurposed compounds can be tested directly in pathogen infection assays within weeks rather than years • Historical precedent: several existing kinase inhibitors, ion channel modulators, and metabolic drugs developed for unrelated conditions have been redirected toward infectious disease targets once interactome studies revealed a mechanistic link
Network-guided repurposing screens following major viral interactome studies (2020 onward) rapidly narrowed thousands of candidate compounds down to a short list of drugs targeting host factors implicated by the physical interaction map — turning a purely descriptive network into an actionable, testable hypothesis set within months of the interactome being published.
Where no existing drug modulates the implicated protein, the validated interaction still provides an enormously valuable starting point for de novo discovery:
• Structural definition of the target: once essentiality is confirmed, structural biology (X-ray crystallography, cryo-EM, or high-confidence computational structure prediction) resolves the precise binding interface between the pathogen and host protein • Interface-focused screening: high-throughput compound or fragment screens can be focused specifically on disrupting that resolved interface, rather than screening blindly against the whole protein surface • Structure-based design: the resolved interaction geometry becomes the direct input for structure-based small-molecule or biologic design, iteratively optimizing for binding affinity and selectivity at that specific interface • De-risking before heavy investment: because the target has already been shown to be essential and mechanistically linked to infection through the interactome data, novel discovery programs start with substantially higher confidence than target selection based on genomic association alone
In both routes — repurposing and de novo discovery — the interaction network is the connective tissue linking basic infection biology all the way through to a concrete, testable therapeutic candidate.