Data perturbation independent diagnosis and validation of breast cancer subtypes using clustering and patterns

Abstract

Molecular stratification of disease based on expression levels of sets of genes can help guide therapeutic decisions if such classifications can be shown to be stable against variations in sample source and data perturbation. Classifications inferred from one set of samples in one lab should be able to consistently stratify a different set of samples in another lab. We present a method for assessing such stability and apply it to the breast cancer (BCA) datasets of Sorlie et al. 2003 and Ma et al. 2003. We find that within the now commonly accepted BCA categories identified by Sorlie et al. Luminal A and Basal are robust, but Luminal B and ERBB2+ are not. In particular, 36% of the samples identified as Luminal B and 55% identified as ERBB2+ cannot be assigned an accurate category because the classification is sensitive to data perturbation. We identify a "core cluster" of samples for each category, and from these we determine "patterns" of gene expression that distinguish the core clusters from each other. We find that the best markers for Luminal A and Basal are (ESR1, LIV1, GATA-3) and (CCNE1, LAD1, KRT5), respectively. Pathways enriched in the patterns regulate apoptosis, tissue remodeling and the immune response. We use a different dataset (Ma et al. 2003) to test the accuracy with which samples can be allocated to the four disease subtypes. We find, as expected, that the classification of samples identified as Luminal A and Basal is robust but classification into the other two subtypes is not.

Keywords: Breast cancer; Clusters; Diagnosis; Multi-gene Biomarkers; Patterns.

Figure 1a

Average agreement scores relative to cluster A.

Figure 1b

Average cluster agreement scores relative to cluster B.

Figure 1c

Average cluster agreement scores relative to cluster C.

Figure 1d

Average cluster agreement scores relative to cluster D.

Figure 1e

Average cluster agreement scores relative to cluster E.

Figure 1f

Agreement scores for the unclassifi ed samples in Sorlie et al.

Figure 2

Heatmap of 148 uni-genes for the samples in core categories.

Figure 3

An example of a pattern (pattern PA) characteristic of the Luminal A core cluster (Cluster A) and an example of a pattern (pattern NA ) characteristic of the non-Luminal A cases. Notice that P is satisfi ed by all the samples in the Luminal A group, while N is satisfi ed by 88% of the non-Luminal A cases. Both patterns P and N are expressed as bounding constraints on the expressions of genes Liv-1 and Gata-3.

Figure 4

Heatmap of combined Ma et al. and Sorlie et al. data using the 38 genes identified in the latter data. There are four distinct clusters which are separtaed by vertical lines in the plot. The Normals, Luminal A and Basal core samples from Sorlie et al. cluster well enough with samples in the Ma et al. data to make a phenotype identifi cation possible for the latter data. The B core cluster (Luminal B) looks similar to the Luminal A core cluster with some genes over expressed. Core cluster C (ERBB2+) is most similar to Core D (Basal) presumably because the discriminator gene ERBB2 gene is not on the Ma et al. chip set . The sample labels in the Ma et al. data indicate stages of disease (ADH, DCIS or IDC) and the index number of the patient. Notice that samples from the same patient, even if in different stages of BCA, cluster together.