For some years now it has been known that many, if not all cancers can be divided into subtypes according to the genetic changes that occur during their development, and that different subtypes of the same tumour have different prognoses and respond to different drugs. This analysis is more highly developed in common cancers such as breast cancer – the most common form of the disease in women in developed countries – than in rare ones. Testing for HER2 over-expression as a predictor of the sensitivity of breast tumours to trastuzumab is now in routine clinical use. This, however, is only the tip of the iceberg: it should be possible to divide breast tumours into many more subtypes with different perturbations in signalling pathways, and to use this information to predict drug response.

The National Cancer Institute in the US uses a panel of sixty tumour cell lines to screen all potential anti-cancer drugs. This panel still contains relatively few lines representing each tumour type. To explore the relationship between pathway activation and potential drug response in breast cancer in more detail, a large group of researchers led by Joe Gray and Paul Spellman from Lawrence Berkeley National Laboratory, Berkeley, California, USA assembled a panel of about fifty different breast cancer cell lines. They showed using hierarchical consensus clustering of gene expression data that these cell lines modelled the tumour subtypes often referred to as luminal, basal and claudin-low (the basal and claudin-low subtypes having previously been known as basal A and basal B respectively). The researchers tested the sensitivity of each cell line to 77 FDA-approved and investigational anti-tumour drugs by measuring the GI₅₀, or the concentration needed to inhibit cell growth by 50%, of each cell line with each compound.

They used ANOVA statistical analysis to test associations between subtype and drug response in three ways: comparing the three subtypes defined above against each other; comparing the luminal cell lines with the basal and claudin-low lines grouped together; and comparing cell lines with amplification of the gene ERRB2 (ERRRB2^AMP) against those without. A total of 23, or very approximately one-third, of the compounds tested were preferentially active in one or more of the cell line subtypes; each of these had a known molecular target. They then used a network analysis tool, PARADIGM, to identify pathway-based mechanisms that might be involved in these different drug responses. Sub-networks of genes that were up- or down- regulated in each type included an ERK1/2 sub-network activated in basal cell lines; a MYC/MAX subnet activated in claudin-low lines; a FOXA1/FOXA2 subnet upregulated in the luminal sub-type and a CTNNB1 subnet down-regulated in lines over-expressing ERRB2.

Some analyses of the response of the cell lines to different drugs confirmed what is already known about the drugs' mechanisms of action or revealed potential new mechanisms. It was not surprising that a DNA damage response sub-network was up-regulated in basal cell lines that responded to the DNA-damaging agent cisplatin, or that an ERBB2-HSP90 subnet was up-regulated in ERRRB2^AMP lines that responded to geldanamycin, which inhibits HSP90. The clear correlation between gene expression and drug response in cell lines to that seen in whole tumours suggests that large panels of cell lines such taken from sub-types of a single tumour type, such as this one, will be useful in identifying further molecular patterns associated with tumour subtypes and developing a personalised drug response to common cancers.
 

Reference
 

Heiser, L.M., Sadanandam, A., Kuo, W.L. and 40 others, Subtype and pathway specific responses to anticancer compounds in breast cancer, Proc. Natl. Acad. Sci. USA, published online ahead of print 14 October 2011, doi: 10.1073/pnas.1018854108