In single cells, noise in gene and protein expression can cause phenotypic heterogeneity. Here, we explored if expression noise could also be observed in complex cellular populations, namely primary human tumors, and the potential relevance of such population-level noise to disease biology. We identified across 341 tissue samples phenotype-specific “stability signatures” - gene sets exhibiting distinct levels of expression noise between tumors and normal tissues. These stability signatures were distinct from the differentially expressed genes typically described in microarray studies, and were validated in nine independent data sets comprising ~1200 tumor and normal tissues from diverse epithelial subtypes. The tumor and normal-specific signatures were associated with distinct pathways regulating either neoplastic development or normal cellular maintenance, and expression of the tumor-specific signature (TSS) in cancers was inversely correlated with metastatic capacity in two different experimental models of metastasis. TSS expression in primary tumors was also a significant predictor of clinical outcome in multiple patient cohorts. A significant fraction of TSS genes possess binding sites for c-Myc, implicating Myc binding as a possible mechanism for TSS regulation. Our results suggest that studying population-level expression noise, which may arise from subtle differences in tissue composition between individuals, may provide insights into disease biology in multicellular organisms.

98th AACR Annual Meeting-- Apr 14-18, 2007; Los Angeles, CA