Pancreatic ductal adenocarcinoma (PDAC) is one of the most lethal adult malignancy due to the propensity of this disease to metastasize. Circulating tumor cells (CTCs) are thought to be enriched for cells with metastatic potential and their characterization offers a means to understand the biological underpinnings of the distal spread of cancer. However, CTCs are rare cells in the blood and their isolation has posed a signifcant technological challenge. Multiple platforms have emerged in capturing these cells with most relying on a positive antibody based selection process (e.g. EpCAM). Our initial molecular characterization of pancreatic CTCs in the genetically engineered LSL-KrasG12D, Trp53flox/flox or +, Pdx1-Cre (KPC) mouse model utilized a microfluidic anti-EpCAM device followed by RNA sequencing to identify aberrant non-canonical WNT signaling in CTC populations (Yu M*, Ting DT*, et al. Nature 2012). Wnt2 was found to increase metastatic potential by enhancing anoikis resistance, a requirement of CTC survival, through activation of Tak1 kinase. However, this study was limited in only providing a partial CTC signature given the variable purity of these purified cell populations and analysis of bulk CTC populations could not provide the resolution to truly assess heterogeneity of these cells. Furthermore, a recent study has shown that cells from the mouse model may disseminate into circulation at an early point in PDAC development through epithelial-to-mesenchymal transition (EMT), which would generate CTCs with very low EpCAM expression and therefore would be missed in an EpCAM based capture device (Rhim AD et al. Cell 2012).

We have overcome these barriers by employing a novel microfluidic isolation device to capture high numbers of CTCs that can be isolated as single cells (Ozkumur E*, Shah AM*, et al. Science Translational Medicine 2013). This device achieves high efficiency negative depletion of normal blood cells providing an enriched population of CTCs in solution that are not biased by a particular extracellular epitope and without any antibody interactions that could affect expression profiles. This has provided a method to truly understand the key transcriptional programs that differentiate CTCs from their primary tumors on an isogenic mouse background.

Five tumor-bearing KPC mice generated a total of 168 single CTCs that were morphologically intact and subjected to a modified single cell RNA sequencing protocol (Tang F et al., Nature Protocols 2010). A total of 75 (45%) of these single CTCs were of sufficient quality for RNA sequencing indicating that the majority (55%) of intact CTCs selected likely had lost viability in the process of hematogenous transit. Single cell RNA-sequencing was also performed on 12 normal leukocytes (WBCs) from a control mouse, 12 mouse embryonic fibroblasts (MEFs), 16 single cells from the mouse NB508 pancreatic cancer cell line, and 34 (min 8 replicates) samples from primary tumors matched to the CTCs.

Unsupervised hierarchical clustering of single cell samples demonstrated clear separation of MEFs, the NB508 pancreatic cancer cell line, and normal WBCs supporting the technical validity of the sequencing approach. Analysis of candidate CTCs identified three major CTC clusters, which were all distinct form matched primary tumors as well as from the NB508 cancer cell line. The must abundant CTC cluster comprised 41 of 75 cells (55%) and was defined by presence of epithelial markers (Krt7, Krt8, Krt18, Krt19) consistent with a “classical” CTC phenotype (CTC-c). The second CTC cluster was defined by enrichment of platelet markers CD41 (Itga2b) and CD61 (Itgb3) (CTC-plt) and a third having enrichment of cellular proliferation genes including Mki67 (CTC-pro). Single cell heterogeneity was assessed by intra-cluster correlation coeffficients, where lower values reflect higher heterogeneity. Not surprisingly, single cell heterogeneity was much higher in CTCs (mean correlation coefficient 0.42, 95% CI 0.36-0.47) compared to cancer cell lines (mean 0.86, 95% CI 0.80-0.91, p-value 1.2 x 10-15), but was notably similar comparing CTCs to single primary tumor cells (mean 0.38, 95% CI 0.28-0.47).

Focusing on the dominant CTC-c cells, we used a non-parametric differential gene expression analysis including a rank product (RP) methodology suitable for large variations in absolute transcript levels found in single cell expression data (Breitling R et al. FEBS Letters 2004). Using a stringent FDR of ≤ 0.01, CTC-c cells had 878 genes with increased expression and 774 genes with reduced expression when compared with matched primary tumors. CTC-c cells were enriched for MAPK, as well as WNT, TGF-β, Neurotrophin, Toll-like receptor, and B-cell receptor signaling pathways. Analysis of a panel of EMT genes with significant differential expression revealed that CTC-c cells were in a biphenotypic state with universal loss of the epithelial markers E-cadherin (Cdh1) and Muc1, while the mesenchymal genes Cdh11 and Vim were found to be expressed much more heterogeneously amongst individual CTCs. Proposed pancreatic cancer stem cell genes were also evaluated and the Aldh1a1 and Aldh1a2 genes were found to be significantly enriched in CTC-c. Expression of both Aldh1a1 and Aldh1a2 in matched primary tumors was done through RNA in situ hybridization (RNA-ISH) revealing a heterogeneous distribution of these stem cell genes in both the stromal and epithelial compartments of the tumor. This highlighted the potential relevance of these stem cell markers in tumor cells dynamically shifting between epithelial and non-epithelial states.

To provide further insight into the potential region from which CTCs emanate from the primary tumor, we selected the most highly enriched CTC transcripts found in ≥ 90% of all classical CTCs. Three genes met these criteria, which were decorin (Dcn), insulin-like growth factor binding protein 5 (Igfbp5), and Kruppel-like factor 4 (Klf4). Each of these genes has been previously implicated in pancreatic cancer development and were evaluated by RNA-ISH in primary tumor specimens to determine if they colocalized in particular tumor cells. Dcn is an extracellular matrix proteoglycan known to be expressed in a wide range of tumor stroma and by RNA-ISH was found primarily in the stromal elements of the tumor. However, both Igfbp5 and Klf4 were found to be focally expressed in cells at the epithelial-stromal interface. Although these genes are co-expressed in a minority of primary tumor cells, they are co-expressed at high levels in 85% of all classical CTCs. Together with the mixed epithelial/mesenchymal marks and enrichment of Aldh1a2 cells in stromal elements, these data point to the majority of viable CTCs emanating from the epithelial/stromal interface.

In summary, we have successfully purified individual pancreatic CTCs using a novel microfluidic device and provided the first comprehensive single cell transcriptome analysis of these rare but exceptional cells. Three major classes of CTCs have been identified that would not have been possible without a single cell approach and we have characterized the major pathways that define these different subsets. The most abundant CTCs were found to have robust expression of keratin genes and are defined by a mixed E/M state with enrichment of Aldh1a1 and Aldh1a2 stem cell genes. These classical CTCs are marked by the co-expression of Igfbp5 and Klf4, which appear to localize to the epithelial-stromal interface in primary tumors. Ultimately, CTC cultures and functional testing will determine the contribution of these genes to CTC metastatic potential. This deep analysis of CTCs at single cell resolution has provided new biological insight into the metastatic cascade that will inform the development of novel therapies to treat this deadly disease.

Citation Format: David T. Ting, Ben S. Wittner, Ajay M. Shah, David T. Miyamoto, Brian W. Brannigan, Kristina Xega, Jordan Ciciliano, Olivia C. MacKenzie, Julie Trautwein, Mohammad Shahid, Haley L. Ellis, Na Qu, Nabeel Bardeesy, Miguel N. Rivera, Ravi Kapur, Sridhar Ramaswamy, Toshi Shioda, Mehmet Toner, Shyamala Maheswaran, Daniel A. Haber. Diversity of circulating tumor cells in a mouse pancreatic cancer model identified by single cell RNA sequencing. [abstract]. In: Proceedings of the 105th Annual Meeting of the American Association for Cancer Research; 2014 Apr 5-9; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2014;74(19 Suppl):Abstract nr NG04. doi:10.1158/1538-7445.AM2014-NG04