Objective

A major objective in clinical practice, including transplantation, is to develop more focused and “personalized” approaches for both diagnosis and treatment. There have been several recent successes in correlating specific polymorphisms with drug responses; however, the goal of personalized medicine is challenging for multiple reasons. 1) Humans are outbred with numerous genetic polymorphisms, 2) each individual confronts different sets of environmental exposures, and 3) as analysis focuses on smaller subsets or an individual patient the statistical power is increasingly diminished. In addition, the “gold standard” for diagnosis following kidney transplantation is pathological analysis of a graft biopsy; however, studies have shown approximately 20% variation among diagnoses between blinded pathologists.

Design and Methods

To address these challenges we focused on the largest available dataset of microarray data analyzing kidney biopsies [Halloran et al] and combined samples from multiple studies to increase statistical power. After filtering the data, we used multidimensional scaling (MDS) to eliminate 11 samples which were statistical outliers (> 2 SD). The 11 outliers did not correlate with any of the diagnoses and appear to include excessive technical noise. Next, we identified molecular subtypes of each diagnosis based on silhouette width, Dunn index and connectivity criteria using 3 types of clustering algorithms (hierarchical, kmeans and pam). Following identification of the molecular subtypes, we performed GeneOntology and KEGG pathway analysis to identify unique biological functions in each molecular subtypes.

Results

Our analysis indicated that the dataset represented 21 distinct molecular diagnoses (expanded from the 11 pathological diagnoses). The subsets of a pathological diagnosis were often extensively different. For example, we identified 849 genes that were significantly different (fdr <0.05) between 2 subsets of T cell mediated rejection (TCMR) indicating substantial differences in gene expression despite the same pathological diagnosis between the 2 sub-clusters of TCMR. To identify the biological functions of each sub-diagnosis, we identified the significantly enriched Gene Ontology terms and KEGG pathways for each of the 21 molecular diagnoses. These analyses demonstrated biological functions in the molecular subtypes. In future studies, we plan to validate our results in prospective studies.

« Back to 2015 American Transplant Congress