The use of genome wide association methods to investigate pathogenicity, population structure and serovar in Haemophilus parasuis

Abstract

Background: Haemophilus parasuis is the etiologic agent of Glässer's disease in pigs and causes devastating losses to the farming industry. Whilst some hyper-virulent isolates have been described, the relationship between genetics and disease outcome has been only partially established. In particular, there is weak correlation between serovar and disease phenotype. We sequenced the genomes of 212 isolates of H. parasuis and have used this to describe the pan-genome and to correlate this with clinical and carrier status, as well as with serotype.

Results: Recombination and population structure analyses identified five groups with very high rates of recombination, separated into two clades of H. parasuis with no signs of recombination between them. We used genome-wide association methods including discriminant analysis of principal components (DAPC) and generalised linear modelling (glm) to look for genetic determinants of this population partition, serovar and pathogenicity. We were able to identify genes from the accessory genome that were significantly associated with phenotypes such as potential serovar specific genes including capsule genes, and 48 putative virulence factors that were significantly different between the clinical and non-clinical isolates. We also show that the presence of many previously suggested virulence factors is not an appropriate marker of virulence.

Conclusions: These genes will inform the generation of new molecular diagnostics and vaccines, and refinement of existing typing schemes and show the importance of the accessory genome of a diverse species when investigating the relationship between genotypes and phenotypes.

Figure 1

H. parasuis isolate collection displayed by serovar and disease. Each serovar includes those found with cross-reactions. NT denotes 'non-typeable’ by serotyping either due to no reaction or due to more than three reactions to the serotyping antisera. Strains without serotyping data are excluded.

Figure 2

Core genome Neighbor-joining tree (with areas of recombination included) of H. parasuis (500 bootstraps). Trees are overlaid with the populations from the Bayesian analysis of population structure (BAPS), which represent isolates with similar rates of homologous recombination. Further metadata including disease association, serovar and country of origin are also shown. BAPS populations explain the separation of the isolates on the tree into two main clades.

Figure 3

The BAPS populations defined using Bayesian analysis of population structure represent isolates with similar rates of homologous recombination. The BAPS populations have been compared to serovar, disease association and country of origin. Some similarity in the BAPS populations 3 and 5 can be seen based on serovar distribution. Very little difference can be seen between BAPS population when disease association is considered. UK isolates are overrepresented in BAPS populations 1, 3 and 4, while 2 and 5 are of mixed geographic origin.

Figure 4

Heat-map of the shared accessory genes between strains. The plot is ordered by the phylogeny based on the SNPs within the core genome (rows). While the columns are ordered based on the similarity in the presence and absence pattern of accessory genes between isolates, which is represented by a dendrogram along the top of the heat-map. A clear separation can be seen between the clades, and it appears that both the phylogeny and the dendrogram split the population into the two clades, suggesting little recombination occurs between the two clades, but there is recombination within them both.

Figure 5

Discriminant analysis of principal components applied to clinical strains of H. parasuis by disease categories. 80% eigenvalues were retained for the PCA and all eigenvalues were retained for the discriminant analysis. Plots a and b show the first axis of the discriminant function while c, d, e and f show the first two axes. Separation along the axes suggests that genetic differences are present between the phenotypic groups of clinical and non-clinical isolates that are being compared; however the presence of overlap shows that some strains are intermediates. Plots c and d show that geographic origin, do not show much separation of the isolates by geography, those that have separated are only represented by a couple of isolates. On the other hand, the discriminant function based on the BAPS populations shows a lot of separation and so the population structure does have an influence on these isolates genetic content.