| Abstract | Antibody-drug conjugates (ADCs) are a promising approach for cancer therapy, combining the specificity of an antibody with the potency of small-molecule toxins. To identify cellular targets for the development of new ADCs, we have set out to identify proteins that (1) are expressed on the cell surface; (2) have high specificity for tumors, with relatively low expression on normal tissues; and (3) can internalize into the tumor cell by a mechanism that enables the delivery and activation of sufficient amounts of toxin to kill cancer cells. Here at the NRC, we have built a pipeline to identify new ADC targets, incorporating public gene expression data mining and glycoproteomic profiling, followed by in vitro screening through a surrogate ADC assay. Public data enable the analysis of large numbers of human tumors and normal tissues, providing a population-based estimate of gene expression. Through curation of the Gene Expression Omnibus, we have built a microarray database that contains >26,000 tumor samples and >8,800 normal samples, all on the Affymetrix HGU133 Plus 2.0 platform. We have also collected RNA-seq data for >5,000 normal samples from the GTEx database, and 6,900 tumor samples from The Cancer Genome Atlas. These samples cover a broad range of tissues: blood, bone marrow, brain, breast, colon, heart, kidney, liver, lung, muscle, ovary, pancreas, prostate, skin, stomach, and uterus. To identify candidates for ADC development, we first classify tumors into subtypes through consensus clustering followed by a Monte Carlo implementation of our iterative ensemble classification methods. Next, we perform differential gene expression analysis between normal tissues and known or novel cancer subtypes. In one example, we have identified 50 breast cancer targets, 7 of which have already been developed as ADCs to the clinical trial stage by others, demonstrating the validity and promise of this approach (Fauteux et al., 2016). Glycoproteomics data are typically derived from small numbers of samples, making a population-based analysis less informative. Therefore, we have integrated glycoproteomic data into our target selection pipeline in two ways. First, glycoproteomics has been used to profile the cell surface of 11 tumor cell lines. Using an approach with high specificity for cell-surface glycoproteins, over 200 cell-surface proteins have been identified for each cell line. This data enables the selection of targets that are amenable to our in vitro functional assay for ADC activity, based on expression in our screen-adapted cell lines. Glycoproteomics has also been used to identify and prioritize targets upregulated during hypoxia or epithelial-mesenchymal transition, two important aspects of tumor biology. For example, cellular glycoproteins from four pancreatic cell lines were profiled under normoxic and hypoxic conditions, identifying >70 proteins upregulated under hypoxic conditions. These glycoproteomic datasets, in conjunction with the public data analysis, are being used to identify promising ADC targets. Based on these target selection methods, we are currently producing and screening thousands of NRC monoclonal and single-domain antibodies generated against a variety of cancer-associated cell surface targets and screening them for ADC activity, in vitro and in vivo. |
|---|
