Multi-omic latent variable data integration reveals multicellular structure pathways associated with resistance to tuberculin skin test (TST)/interferon gamma release assay (IGRA) conversion in Uganda

Abstract

Understanding the mechanisms of early clearance of Mycobacterium tuberculosis (Mtb) may illuminate new therapeutic strategies for tuberculosis (TB). We previously found genetic, epigenetic, and transcriptomic signatures associated with resistance (resister, RSTR) to tuberculin skin test (TST)/interferon gamma release assay (IGRA) conversion among highly exposed TB contacts. We hypothesized that integration of these datasets with multi-omic latent factor methods would detect pathways differentiating RSTR patients from those with latent infection (LTBI) which were not differentiated by individual dataset analyses. We pre-filtered and scaled features with the largest change between LTBI and RSTR groups for 126 patients with data in at least two of five data modalities: single nucleotide polymorphisms (SNP), monocyte RNAseq (baseline and Mtb-stimulated conditions), and monocyte epigenetics (methylation and ATAC-seq). Using multiomic latent factor analysis (MOFA), we generated ten latent factors on the subset of 33 patients with all five datasets available, four of which were different between LTBI and RSTR (FDR < 0.1). Factor 4, which was the most integrated of the significant factors, showed the greatest difference between RSTR and LTBI groups (FDR < 0.001). Three additional latent factor data integration methods also distinguished the RSTR and LTBI groups and identified overlapping features with MOFA. Using pathway analysis and a cluster-based enrichment method, we identified biologic functions associated with latent factors and found that MOFA Factors 2-4 include functions related to cell-cell adhesion, cell shape, and development of multicellular structures. In summary, latent variable integration methods uncovered signatures associated with resistance to TST/IGRA conversion that were not detected within individual dataset analyses and included pathways associated with cellular interactions and multicellular structures. Author SummaryTuberculosis (TB) is a leading cause of preventable death worldwide. Previous research has identified some genetic and molecular patterns that are linked to resistance to TB in people who are highly exposed to the bacterium that causes the disease but do not develop an infection. We took genetic, epigenetic, and gene expression data from people who were not infected after exposure to TB within in their household and compared them with people who did develop a latent form of TB disease after similar exposure. We used a special method to analyze these datasets in an integrated way, rather than separately. This approach revealed four key factors that clearly distinguished between the resistant individuals and those with latent infection. One of these factors, in particular, showed a strong difference. We found that these factors were related to important biological processes such as how cells stick together, their shape, and how they form tissues. This method allowed us to see new patterns in the biological pathways linked to TB resistance that were not evident when looking at each type of data on its own.