August 1, 2016 (Vol. 36, No. 14)
Richard A. A. Stein M.D., Ph.D.
Prospectors Of Old Had It Easy: Pick the Glitter from the Sludge. Today’s Transcriptomic Prospectors Have It Harder
The transcriptome is less a tranquil pool than a turbulent stream that keeps shifting its course and adopting new patterns of gene expression. To survey this stream, one must repeatedly troop up and down its banks. Otherwise, one will never identify its developmental headwaters, homeostatic meanders, or pathogenic cascades.
Along with surveying comes prospecting, that is, dipping into the sediment beneath the flow to assess the cellular and molecular bases of gene expression. Here it is important to distinguish between materials that may seem to be homogeneous or intractably jumbled, but are in fact heterogeneous and, more important, separable. In fact, the growing appreciation of the inter-individual variability that defines cellular populations has become the driving force of efforts to capture and analyze the transcriptome in single cells.
Something of a transcriptomic gold rush is taking shape. The gold, however, may escape all but the best prospectors, those capable of deftly panning unpromising deposits and glimpsing the merest transcriptomic glints in heterogeneous cell populations.
“One of the biggest challenges in oncology is the diversity of cells within a tumor and at different sites within a patient,” says Sohail Tavazoie, M.D., Ph.D., associate professor of systems cancer biology at Rockefeller University. When pondering how to grapple with such diversity, Dr. Tavazoie recalled observations that he had made years ago. He saw that a minority of cells derived from a single clone showed marked phenotypic hetoergeneity.
Expanding on these observations, Dr. Tavazoie reasoned, might improve the understanding of the molecular basis of phenotypic variation. Consequently, he led an effort at his laboratory to generate clones from two different breast cancer cell lines. In this effort, differences between the daughter cells were examined systematically.
Nongenic Variation in Clonal Relief
“Sequencing did not reveal any mutations that were significantly different between the daughter cells,” reports Dr. Tavazoie. “And RNA-seq did not show any differences coming out of the noise.”
However, further studies showed that certain clonal subpopulations, which represented about 1% of the parental population, exhibited a marked morphological variation. Comparing several of these highly variable clones with nonvariable clones unveiled a marked variation in transcript levels. “This suggested that there must be some regulator or regulators that are causing molecular level variation,” explains Dr. Tavazoie.
Analysis of cell-to-cell variation in gene transcript expression identified decreased SNRNP40 levels as one of the factors that increased the metastatic fitness of individual cells. “We found that modulating this gene involved in splicing could recapitulate part of the molecular variation that we saw,” notes Dr. Tavazoie. Several clinical datasets collected from primary breast cancer samples confirmed that decreased SNRNP40 transcript-level expression, which is known to significantly increase the fraction of unspliced pre-mRNAs, is associated with increased metastatic relapse.
These findings illustrated that molecular transcriptomic variation among cancer cell populations drives phenotypic variation, and pointed toward one of the strategies that allows cells to overcome the tumor microenvironment. “We have been very pleased with the pace of the evolution of RNA-seq,” comments Dr. Tavazoie. “Having better analytical tools to improve standardization of the pipelines for spliceoform analysis would be helpful for future studies.”
Sorting and Sequencing
“We apply microfluidics to study cellular heterogeneity,” says Mission Bio CSO and co-founder Dennis Eastburn, Ph.D. “We also perform single-cell analyses at a very high throughput.”
Capturing information from single cells is of increasing interest in biology, and in the case of malignant tumors, circulating tumor cells can provide critical data with therapeutic and prognostic relevance. However, a key limitation of many technologies is the number of cells that can be captured from populations of circulating cells.
“Another consideration,” notes Dr. Eastburn, “is how many of those cells have sufficient RNA sequencing quality to generate meaningful information on the heterogeneity of the disease.”
Work at Mission Bio has attracted the interest of investigators working on circulating tumor cells. While circulating tumor cells can be efficiently purified from very large cell populations obtained during a blood draw, the enrichment methods and platforms often leave background cells that get co-purified. “We want to be able to take partially purified cell populations and pull out tumor cells for a much more detailed RNA-seq analysis,” informs Dr. Eastburn.
One approach that Dr. Eastburn’s team is using to identify rare cells relies on a polymerase chain reaction (PCR) approach to achieve extreme sensitivity and specificity. “Because we are PCR-based and not protein-based, we can look at nucleic acid biomarkers that otherwise one could not assess with antibody-based or fluorescence-activated cell sorting (FACS)-based approaches,” explains Dr. Eastburn. A PCR-based approach allows single cells to be isolated based on a noncoding RNA, mutation, or splice variant prior to characterizing the genomes and transcriptomes of those cells.
In a recent study, Dr. Eastburn and colleagues extended the use of PCR-activated cell sorting (PACS) to perform molecular profiling of TaqMan-targeted cancer cells using a random priming RNA-seq strategy. “We targeted prostate tumor cells expressing a number of nucleic acid biomarkers,” details Dr. Eastburn. “One of the biomarkers we evaluated is an androgen receptor splice variant that correlates with therapeutic resistance.”
This approach helped generate high-fidelity transcriptome measurements of prostate cancer cells. It also allowed transcriptional profiles that otherwise would have remained obscured in the population measurements to be captured.
The future of this strategy critically depends on improving the tools for genomic sequencing. “In addition,” emphasizes Dr. Eastburn, “we need tools able to capture a large percentage of the transcriptome to identify weakly expressed transcripts within individual cells.”
Single Cells, Dual Identities
“We developed a method to simultaneously collect information about the methylome and the transcriptome in single cells,” says Dr. Guoping Fan, Ph.D., professor of human genetics at the University of California, Los Angeles. This method is known as single-cell methylome and transcriptome sequencing (scMT-seq). It involves isolating cytosolic RNA and genomic DNA from a single cells. The RNA is subjected to RNA-seq; the DNA, to DNA methylome profiling.
The new method, Dr. Fan asserts, can simultaneously profile the DNA methylome and the transcriptome in the same cell, both in proximal promoter and in gene body regions. It can also provide information about methylation status and transcriptional activity at individual allelic loci.
In a proof-of-concept study, Dr. Fan and colleagues used scMT-seq to interrogate the transcriptome and methylome heterogeneity among individual sensory neurons isolated from the adult mouse dorsal root ganglion. “Our approach,” recalls Dr. Fan, “allowed us to collect information on the transcriptome, the methylome, and single nucleotide polymorphisms in single cells—information that sheds light on the epigenetic mechanisms of gene expression.”
Experiments using scMT-seq revealed that promoter methylation is inversely correlated with gene expression for genes carrying a low-density CpG promoter (non-CpG island promoter), and that gene body methylation is positively correlated with gene expression for genes containing a high-density CpG promoter or CpG island promoter at the single-cell resolution. Such findings have clinical implications. Nonetheless, as Dr. Fan cautions, more development is needed: “There are still some hurdles we need to clear before these findings can meet the clinical standards and reach the phase when they can be incorporated into clinical decisions.”
No Such Thing as Overexposure
At the single-cell level, correlations between phenotype information and molecular profile information can be hard to see. Although microscopy provides opportunities to dynamically study the phenotypes of individual cells, including behavior, size, movement, and location within a sample, collecting genomic or transcriptomic information from specific cells of interest is currently an experimental challenge.
This challenge, however, may soon become more tractable thanks to efforts such as those undertaken by Santiago Costantino, Ph.D., associate professor of ophthalmology at the University of Montreal. “We developed a new strategy for tagging individual cells,” says Dr. Costantino. “When our approach is used, a cell of interest can be tagged based on morphological criteria, and later it is possible to find that specific cell to perform sequencing.”
The approach developed in Dr. Costantino’s laboratory is called cell labeling via photobleaching (CLaP). Cells are incubated with biotin conjugated to fluorescein to label cell membranes. When the culture media is illuminated with a low-intensity laser beam focused on an individual cell of interest, photobleaching of the dye makes biotin become very reactive. After crosslinking, cells are incubated with streptavidin conjugates. Ulitimately, photobleaching-labeled cells are identified under a fluorescent microscope.
“We can identify cells based on phenotypic characteristics and then analyze the genotype or transcriptome of that specific cell,” explains Dr. Costantino. Experiments in Dr. Costantino’s laboratory confirmed that the reagent used for this study is nontoxic and transferred only to daughter cells during cell division but not to adjacent cells in culture.
Using CLaP, Dr. Costantino and colleagues demonstrated the capture and analysis of single cells of interest. Cells were captured from a heterogeneous population of human retinal pigment epithelial cells and subjected to single-cell transcriptome-wide next-generation sequencing. Coverage distribution across the genome was comparable between tagged and untagged cells, and tagging did not induce any major changes in the gene expression profiles.
While experimental strategies to sequence single cells of interest have been described before, one of their collective shortcomings has been the need for proprietary platforms. “The advantage of our approach,” asserts Dr. Costantino, “is that investigators can use any confocal microscope and then go to a microchip, without the need for any special equipment that people would not already have in their laboratories.”
“Much of our work is focusing on mutations in transcription factors,” says Olivier Elemento, Ph.D., associate professor of physiology and biophysics at Weill Cornell Medical College. “Mutations of this type are very relevant in cancer.”
In cancer research, such mutations have been described even more frequently than other well-studied mutation types including mutations in kinase genes. Yet mutations in transcription factor genes, notes Dr. Elemento, pose a special difficulty: “Kinases can be targeted pharmacologically in many ways using small molecules, but transcription factors are not druggable because it is difficult to find a small molecule that blocks their binding to DNA.”
The pharmacological targeting of oncogenic transcription factors, Dr. Elemento and colleagues decided, could be accomplished by means of a computational approach, one that would search for molecules that capable of modulating transcription factor activity. “We wanted an approach,” explains Dr. Elemento, “that would be agnostic to the mechanism of action.”
The approach developed in Dr. Elemento’s laboratory is based on an in silico screening for chemicals that could specifically disrupt the expression of many genomic targets of a transcription factor, without disrupting the expression of nontarget genes. Using the genetic profiles of cell lines treated with over 1,300 different drugs, Dr. Elemento and colleagues identified genes that are up- or down-regulated in response to the drug. Then the investigators generated connectivity maps reflecting the transcriptional impact of each individual small molecule.
“We used the information about transcription factors and the genes that are modulated to carry out a sort of matchmaking procedure,” details Dr. Elemento. That is, the investigators considered potential matches between drugs and transcription factor genomic targets.
In a proof-of-concept study, Dr. Elemento and colleagues applied this approach to study ERG, an oncogenic pro-invasive transcription factor that is frequently overexpressed in prostate cancer. “We mapped the ERG binding sites in the genome,” recalls Dr. Elemento. “We found about 2,000 target genes. Then, with reference to the Broad Institute’s Connectivity Map, we asked whether there were any drugs that could inhibit most of those 2,000 genes.”
This strategy identified eight candidate drugs that could inhibit many of the targets, and when ChIP-seq combined with drug-induced expression profiling was used to prioritize the list, dexamethasone emerged as the compound with the highest prediction score. ”We showed that dexamethasone is able to disrupt or reverse all the phenotypes induced by ERG,” informs Dr. Elemento.
Building on this finding, Dr. Elemento and colleagues used clinical database electronic medical data to retroactively identify patients who received dexamethasone. Then the investigators determined the likelihood these patients had for developing prostate cancer later in life. “Patients treated with dexamethasone,” reports Dr. Elemento, “had a much lower risk for prostate cancer later in life, reinforcing our experimental data.”