Non-parametric heat map representation of flow cytometry data: identifying cellular changes associated with genetic immunodeficiency disorders

Genetic primary immunodeficiency diseases are increasingly recognised, with pathogenic mutations changing the composition of circulating leukocyte subsets measured by flow cytometry (FCM). Discerning changes in multiple subpopulations is challenging, and subtle trends might be missed if traditional reference ranges derived from a control population are applied. We developed an algorithm where centiles were allocated using non-parametric comparison to controls, generating multiparameter heat maps to simultaneously represent all leukocyte subpopulations for inspection of trends within a cohort or segregation with a putative genetic mutation. To illustrate this method, we analysed patients with Primary Antibody Deficiency (PAD) and kindreds harbouring mutations in TNFRSF13B (encoding TACI), CTLA4 and CARD11. In PAD, loss of switched memory B cells (B-SM) was readily demonstrated, but as a continuous, not dichotomous, variable. Expansion of CXCR5+/CD45RA- CD4+ T cells (X5-Th cells) was a prominent feature in PAD, particularly in TACI mutants, and patients with expansion in CD21-lo B cells or transitional B cells were readily apparent. We observed differences between unaffected and affected TACI mutants (increased B cells and CD8+ T-effector memory cells, loss of B-SM cells and non-classical monocytes), cellular signatures that distinguished CTLA4 haploinsufficiency itself (expansion of plasmablasts, activated CD4+ T cells, regulatory T cells, and X5-Th cells) from its clinical expression (B-cell depletion), and those that were associated with CARD11 gain-of-function mutation (decreased CD8+ T effector memory cells, B cells, CD21-lo B cells, B-SM cells and NK cells). Co-efficients of variation exceeded 30% for 36/54 FCM parameters, but by comparing inter-assay variation with disease-related variation, we ranked each parameter in terms of laboratory precision versus disease variability, identifying X5-Th cells (and their derivatives), naive, activated and central memory CD8+ T cells, transitional B cells, memory and SM-B cells, plasmablasts, activated CD4 cells, and total T cells as the ten most useful cellular parameters. Applying these to cluster analysis of our PAD cohort, we could detect subgroups with the potential to reflect underlying genotypes. Heat mapping of normalised FCM data reveals cellular trends missed by standard reference ranges, identifies changes associating with a phenotype or genotype, and could inform hypotheses regarding pathogenesis of genetic immunodeficiency.