|Program and Abstracts
Back to 2015 Annual Meeting Program
Detection of Novel Cancer Subpopulations Through Integration of Public Microarray Data with Single Cell Gene Expression Profiling
Michael Januszyk, MD, Nadine Jahchan, PhD, Purvesh Khatri, PhD, Robbert C. Rennert, MD, Michael Sorkin, MD, Julien Sage, PhD, Michael T. Longaker, MD, Atul J. Butte, MD, PhD, Geoffrey C. Gurtner, MD.
Stanford University, Stanford, CA, USA.
Single cell gene expression analysis represents an attractive approach to investigate complex cell populations such as tumors for which the granularity afforded by traditional microarray-based analyses is insufficient. However, these high-resolution techniques are typically limited in the number of gene targets that may be simultaneously interrogated and therefore capture only a small fraction of each cell's transcriptome. This limitation may be addressed by screening public microarray data in order to generate an informed list of genes likely to be differentially expressed for each cancer subtype. The resulting transcriptional interrogation can then optimally identify biologically distinct cell subpopulations for prospective isolation, functional evaluation, and selective eradication.
A meta-analysis was performed across six public microarray datasets for human small cell lung cancer (SCLC) comprising 365 samples across eight different platforms. Genes were ranked according to effect size and p-value for tumor versus control samples, and false discovery rates were calculated. The top scoring 48 genes that were significant by both methods, along with the 48 highest rated surface antigen genes, were used to populate a gene list for subsequent single cell evaluation. High throughput gene expression analysis was performed for 400 individual cells from one SCLC line (H446) using the Fluidigm microfluidic platform. Supervised machine learning was applied to identify transcriptionally-defined subgroups among these cells. The non-parametric Kolmogorov-Smirnov test was then used to determine those surface markers best able to distinguish each cluster. Individual cells from each group were then FACS-isolated, and clonogenicity was evaluated after 14 days in culture. Using these surface antigens, prospectively isolated subpopulation cells from four SCLC lines were seeded intranasally into immunodeficient mice in a murine model of SCLC. Tumor weight, volume, and histologic composition were evaluated 1 month following injection of cells.
Three distinct subgroups were identified using K-means clustering, and one cluster exhibited increased expression of genes associated with aggressive malignancy such as PDIA, JAG1, and CD47. DDR1 [p = 2.33e-09] and CD47 [p = 3.38e-08] were identified as the most cluster-distinguishing surface marker genes, and subsequent FACS analysis revealed two distinct populations separable based on DDR1 protein expression. Three sub-populations of SCLC cells (DDR1+, DDR1-, and unsorted) were prospectively isolated using FACS and evaluated in vitro in culture. DDR1+ cells demonstrated significantly greater clonogenic capacity than either DDR1- or unsorted cells, suggesting that this marker may enrich for a more aggressive subset of SCLC cells. Intranasal seeding of cells from each subpopulation produced tumors of similar size and weight; however, histomorphologically, DDR1+ xenografts exhibited a significant increase in the number of mesenchymal NonNE cells, which are required for SCLC metastasis, and these cells were nearly absent in DDR1- tumors. This suggests that selective destruction of DDR1+ cells may significantly decrease the metastatic potential of SCLC tumors.
Identification of novel subtypes among complex cell populations may facilitate the development of new diagnostic and therapeutic techniques centered on the enrichment, depletion, or targeted eradication of specific subpopulations with distinct functional properties.
Back to 2015 Annual Meeting Program