Fig 1.
B. mallei-human host-pathogen protein interactions.
The set of human-B. mallei protein-protein interactions (PPIs) was created by merging human-B. mallei and orthologous murine-B. mallei interaction data. The set consists of 1,235 unique interactions (gray and purple lines) between 21 B. mallei (green hexagons) and 828 human proteins (pink, blue, and purple circles). Known virulence factors are also indicated in the graph.
Table 1.
Known B. mallei virulence factors that have been shown to attenuate the disease in animal models.
Table 2.
Enrichment of Gene Ontology (GO) terms for human proteins interacting with B. mallei virulence factors.
Table 3.
Topological properties of human proteins interacting with B. mallei.
We evaluated the following properties of the host proteins that interacted with B. mallei proteins based on the human protein-protein interaction (PPI) network [25]: the number of these host proteins in the human PPI network (Np); the average number of interacting partners (in the human PPI network) of each host protein (D); the clustering coefficient, i.e., the number of interactions among the nearest neighbors (C); the average shortest path between any two proteins in the set (SP); the average number of interacting partners in the human PPI network where both partners interact with B. mallei proteins (Di); and the number of host proteins in the largest connected component ( ). The top three rows show the results for the host proteins present in the PPI that interacted with the nine known virulence factors, whereas the three lower rows correspond to host proteins that interacted with all 21 tested B. mallei proteins from the yeast two-hybrid screening (known and putative virulence factors). The results for the randomly selected (498 or 619) human proteins from the entire human PPI network (All PPIs) were generated through 103 random repetitions to create averages and standard deviations. The indicated p-values correspond to the probability of the observed properties being different from the randomly selected set from all PPIs.
Table 4.
Enrichment of Gene Ontology (GO) biological processes in host subnetworks.
LCC represents the number of proteins in the largest connected component annotated with a given term; LIM represents the number of proteins in the largest interaction module for a given term; pGO denotes the probability of the same number of proteins as the LCC being annotated with a given GO term solely through a random selection; pRp denotes the probability that a given number of proteins as the LIM are annotated with a given GO term solely through random selection; pRn represents the probability that a given number of proteins as the LIM are annotated with a given GO term solely through random selection in a random network that has the same degree distribution as our human network. All p-values were assessed using the Benjamini-Hochberg method to meet a maximum false discovery rate threshold of 5% [45]. The table contains only the lowest-level GO terms; the complete data are available in S1 Table.
Fig 2.
Clustering of human proteins targeted by B. mallei known virulence factors.
The graphic shows 116 proteins of the largest connected component of the human PPI network that belong to one or more statistically significant interaction modules. Note that each of these human proteins also interacted with one or more known B. mallei virulence factors. As exemplified by the annotated interaction modules in A-F, the known virulence factors targeted human proteins that were highly interacting among themselves and belonged to the same biological process. For a list of host proteins that compose each interaction module, see S2 Table.
Fig 3.
Focal adhesion pathway as a virulence factor target.
We identified the Kyoto Encyclopedia of Genes and Genomes (KEGG) focal adhesion pathway as enriched with multiple virulence factor targets. The majority (17 of 20) of host proteins interacting with B. mallei virulence factors in this pathway belong to a connected sub-pathway of proteins (yellow boxes and red lines), mainly grouped at the beginning of the pathway. This observation implied that receptors and signaling molecules were likely B. mallei targets, corroborating previous observations that pathogens tend to interfere with host processes related to cell communication and actin cytoskeleton organization. The pathway diagram for the focal adhesion pathway was adapted from the KEGG pathway map [32] with permission from the KEGG database administrators.
Fig 4.
Actin cytoskeleton organization as a virulence factor target.
Human proteins targeted by B. mallei proteins formed an interaction module that was primarily linked to cytoskeleton organization and focal adhesion. Twenty-five of these proteins were involved in cytoskeleton organization processes; 13 of them (red stars) interacted with each other (forming an interaction module), and the remaining 12 proteins (dark red circles) were on average < 2 nodes (< 3 edges) away from the interaction module. The figure also shows the overlap between the cytoskeleton organization interaction module and the focal adhesion pathway (shaded area), where connecting protein interactions from focal adhesion pathway proteins or other proteins appear as smaller circles and dashed lines. Note that all human proteins shown interacted with one or more B. mallei proteins.
Fig 5.
Crosstalk between host pathways targeted by B. mallei virulence factors.
A) The number of shared proteins and shared pathway protein-protein interactions (PPIs) among human proteins interacting with B. mallei that appeared in the focal adhesion pathway and in up to eight other molecular pathways that shared proteins (partially overlapped) with this pathway. The number of shared proteins across pathways was smaller than the number of shared pathway PPIs. B) The location and number of crosstalk interactions affected by B. mallei centered around the focal adhesion pathway and appear as arrows with line thicknesses proportional to the number of shared PPIs. The identity and number of virulence factors that target each pathway are illustrated using a word cloud. By preferentially targeting signaling pathways, the effect of one protein modulated through interaction with a virulence factor could propagate and disproportionally influence a larger number of biological processes.
Table 5.
Putative role of B. mallei proteins inferred from host-pathogen interaction network alignment results.
Fig 6.
Focal adhesion as a central hub for targeting host cells.
The number of shared protein-protein interactions (PPIs) targeted by B. mallei virulence factors is shown as lines proportional to the number of PPIs (only connections with 10 or more interactions are illustrated). Physiological, cancer, and disease pathways were all interconnected via signaling pathways that could be affected through the focal adhesion pathway.
Fig 7.
Host-Pathogen Interactions Alignment (HPIA) algorithm.
The HPIA algorithm is a seed-and extend algorithm that aligns two bipartite graphs, e.g., two different host-pathogen protein interaction networks. A) Given an initial pair of seed nodes (red nodes U1 and u1) from two graphs G1 (left) and G2 (right), the algorithm first aligns seed nodes to each other. Then, it aligns the neighbors of the seed nodes from the first graph (green nodes V1-V6) to the neighbors of the seed nodes in the second graph (green nodes v1-v5) based on the node similarity measure (as defined in Equations 6, 8, and 9). This procedure results in six aligned nodes and five aligned edges. B) The algorithm iteratively selects new seeds and extends around them, e.g., it selects nodes V6 from G1 and v5 from G2 as new seed nodes and, based on the node similarity measure, aligns their unaligned neighbors U2 to u2, creating an additional aligned edge (U2-V6 to u2-v5). C) When the algorithm cannot find any seed nodes of the same type that have unaligned neighbors, it greedily aligns all of the remaining unaligned nodes based on their type and the node similarity measure. Some nodes may remain unaligned if the graphs’ sizes vary, e.g., when there is no match for v6 from G2 in G1. The HPIA algorithm generates a list of aligned nodes and a list of aligned edges inferred from the aligned nodes.