Reference Component Analysis of Single-Cell Transcriptomes Reveals Cellular Heterogeneity in Colorectal Cancer
Watch the webinar featuring Shyam Prabhakar, PhD, of the Genome Institute of Singapore, discussing tumor profiling using C1.
Scientists are investigating the role of the microenvironment in tumorigenesis and metastasis more deeply than ever before. Single-cell analysis is likely to become an essential tool for developing targeted cancer therapies. Conventional bulk-sample approaches make it difficult to thoroughly interrogate tumor diversity for specific cell populations and tumor growth catalysts. It is absolutely necessary to go deeper for a census of individual tumor cells to see their true signature. Tumors are highly heterogeneous, with inherent diversity within each cell type. Understanding which tumor cell types are present and their relative abundance, and fully characterizing their functional states can only be achieved through single-cell technology.
“Bulk transcriptomics is certainly useful even when the sample is a heterogeneous mixture of cells, but single-cell transcriptomics is far more informative.”
—Shyam Prabhakar, Genome Institute of Singapore, A*STAR
In 2013, Fluidigm and the Genome Institute of Singapore (GIS) collaborated to open the Single-Cell Omics Centre. Shyam Prabhakar is Associate Director of Integrative Genomics and Group Leader at the GIS, an institute of the Agency for Science, Technology and Research (A*STAR). Prabhakar recently discussed with Fluidigm his intriguing single-cell research into epithelial–mesenchymal transition (EMT), stemness, stromal signatures and colorectal cancer survival. “Until very recently, people had only looked transcriptomically at bulk tumors or known cell types,” he realized. “We didn’t have unbiased analyses of the cell types responsible for tumor development, metastasis or therapeutic resistance nor the mechanisms that regulate these functions.”
Do we see EMT in colorectal cancer?
Recent work by Prabhakar’s team members Huipeng Li and Elise Courtois revealed insights that are challenging to discover without the surgical precision of single-cell sequencing. They went to the deepest resolution possible and saw a distinct transcriptomic signature for EMT in a subset of cells from the tumor mass in a series of colorectal cancer (CRC) patients. Surprisingly, this EMT signature was found not in cancer cells but in cancer-associated fibroblasts, which are mesenchymal to begin with.
This goes against the expectation that colorectal cancer cells undergo EMT and thereby acquire the ability to metastasize, Prabhakar said. “We can't rule out EMT in other tumor subregions that we had no access to, but in the samples we received, EMT is either not happening or happening only rarely.” Recent studies from Enzo Medico and Eduard Batlle’s groups at the University of Torino and IRB Barcelona point in the same direction, leading to the conclusion that the EMT signature observed in bulk tumor studies actually comes from tumor fibroblasts.
This is a critical distinction, and these results highlight the importance of interrogating and finding the signatures of every individual cell because they are different. More importantly, previous analyses based on pooling diverse cell types could lead to false conclusions because it’s difficult to discern which cell type is responsible for the observed changes.
Subpopulations go unseen without single-cell resolution
Single-cell technologies enable researchers to begin exploring cancer mechanisms. Prabhakar explained that in previous studies of tumor phenotypes, metastasis and drug response, there were almost no whole-transcriptome single-cell analyses of solid tumors. He therefore profiled colorectal cancer tumors using C1™ single-cell RNA sequencing to reveal functional states at single-cell resolution.
“We want to do personalized oncology. If we can dissociate the patient tumor into single cells and characterize the tumor phenotype, we could potentially tell the clinician how to treat that particular patient. Currently people are doing this based on tumor DNA sequencing, but I think single-cell RNA sequencing is the next wave. I look forward to upcoming research using the C1 system. There are many questions we can ask with this technology.”
—Shyam Prabhakar, Genome Institute of Singapore, A*STAR
“Colorectal cancer is one of the most common cancers worldwide. Differentially expressed genes from tumor versus normal, that’s really the bedrock of cancer genomics. We use single-cell data to redefine what are the differentially-expressed gene sets in a cell-type-specific manner,” said Prabhakar. For example, genes expressed in stem-like cells appeared to be more strongly expressed in the tumor, but this was often simply because tumor samples contained more stem-like cells. Single-cell analysis facilitates comparisons between cells of the same type instead of comparing heterogeneous cell populations.
The caveats of bulk sample profiling may well be extended to patient stratification and prognosis. In his study, Prabhakar showed a patient stratification method that added value to an existing CRC tumor classification scheme based on single-cell transcriptomic signatures. “Clearly there is additional prognostic information that single-cell transcriptomics can provide at the level of overall survival,” he added.
“Bulk transcriptomics is certainly useful even when the sample is a heterogeneous mixture of cells, but single-cell transcriptomics is far more informative. Single-cell transcriptomics should be relevant for any cancer because tumor heterogeneity affects all aspects of clinical outcome.”
Single-cell advantages with C1
“The C1 platform does a lot of the work automatically, and it also allows imaging,” Prabhakar said. “The system is labor-saving compared to more traditional 96-well-plate methods for single-cell RNA seq since a lot of labor-intensive steps are done on the IFC.” C1 made it easier to get to those subpopulations reliably and consistently.
“C1 was a huge advantage in reaching our EMT findings, allowing us to image and see what’s a single cell, a clump or debris and what’s a doublet. That is absolutely essential. Using a single-cell platform with no imaging, we could mistakenly think we’d discovered a new tumor cell type with this weird transcriptome,” he explained. “You may think you’re looking at a single cell that has undergone EMT when it’s actually an epithelial cell stuck to a mesenchymal cell. You can’t reliably tell the two scenarios apart bioinformatically—you need imaging for this.”
The GIS approach
Facilitated by the SCOC, Elise Courtois from Prabhakar’s lab built a workflow to process primary tumors from 11 CRC patients, which yielded intact single cells of diverse types. Downstream analysis on the C1 single-cell mRNA sequencing system resulted in 1,591 single-cell transcriptomes from tumor and matched-normal samples. In collecting single-cell data, the team realized that current analysis algorithms were insufficient: there were no good bioinformatics tools to process and analyze the data.
“Proving negatives is hard,” Prabhakar said. “We had to develop a new clustering algorithm because existing analysis methods simply could not tell cell types apart.” To create a benchmark dataset, the group performed C1 RNA seq to accumulate approximately 630 single-cell transcriptomes from seven cell lines.
Huipeng Li tested existing RNA seq algorithms using this standard but none reliably clustered cells of the same type together. His solution was to develop the new Reference Component Analysis (RCA) method to cluster single-cell transcriptomes. It worked almost perfectly on the benchmark dataset, and also on a more recently published single-cell dataset. With the RCA algorithm, Li successfully divided colorectal tumor and surrounding normal data into cell-type clusters.
To facilitate his efforts, Li also used two new algorithms developed by his colleague Debarka Sengupta, described in the bioRxiv preprint, “Fast, scalable and accurate differential expression analysis for single cells.” The team generated C1 single-cell data to benchmark the algorithm for normalizing single-cell transcriptomes (pQ) and the algorithm for finding differentially-expressed genes between two cell types (NODES). Emphasizing that “the normalization and differential expression methods we provided are essential for single-cell analysis because they are nonparametric,” he added, “They make almost no assumptions about the nature of the data so they’re quite robust and around 10 times faster than existing approaches.”
Building blocks to future therapies for other cancers
This particular study covered Prabhakar’s work with colorectal tumors, yet he has applied C1 to lung cancer investigations as well. “We’ve already started a subsequent research project doing a parallel study on lung cancer where we’re analyzing a larger dataset.” These findings could potentially improve our understanding of EMT cells and patient cancer treatments. “Our current data are from treatment-naïve primary tumors” Prabhakar said, “but the next step is understanding drug response or even new drug indications.”
Prabhakar sees single-cell transcriptomics becoming an essential tool for cancer biology, biomarker discovery and personalized oncology. “We want to do personalized oncology. If we can dissociate the patient tumor into single cells and characterize the tumor phenotype, we could potentially tell the clinician how to treat that particular patient. Currently people are doing this based on tumor DNA sequencing, but I think single-cell RNA sequencing is the next wave. I look forward to upcoming research using the C1 system,” he concluded. “There are many questions we can ask with this technology.”
Watch Shyam Prabhakar’s webinar, Reference Component Analysis of Single-Cell Transcriptomes Reveals Cellular Heterogeneity in Colorectal Cancer, available on demand at fluidigm.com.
For Research Use Only. Not for use in diagnostic procedures.