Single-cell transcription site activation predicts chemotherapy response in human colorectal tumors

Abstract

Candidate gene and pathway approaches, and unbiased gene expression profiling, have identified marker signatures predictive of tumor phenotypes, such as drug sensitivity and invasive or metastatic potential. However, application of such information to evaluation of tumors in the clinic is limited by cell heterogeneity in the tumor. We have developed a novel method of fluorescence in situ hybridization (FISH) that can detect transcriptional activation of individual genes at their site in single cells in the interphase nucleus. A major obstacle in the treatment of colorectal cancer is relative insensitivity to the chemotherapeutic agent 5-Fluorouracil (5-FU). Here, we have developed a sensitive approach to predict relative sensitivity of colorectal cancer cells to 5-FU, using FISH with probes targeted to nascent mRNAs to measure the number of individual cells with active transcription sites for a panel of candidate genes. These results reveal that the transcriptional status of four key genes, thymidylate synthase (TYMS), MORF-related gene X (MRGX), Bcl2-antagonist/killer (BAK), and ATPase, Cu(2+) transporting beta polypeptide (ATP7B), can accurately predict response to 5-FU. As proof of principle, we show that this transcriptional profile is predictive of response to 5-FU in a small number of patient colon tumor tissues. This approach provides a novel ability to identify and characterize unique minor cell populations in the tumor that may exhibit relative resistance to chemotherapy.

Figure 1

Defining markers of 5-FU response in human colorectal tumor cell lines using single-cell profiling of transcription site activation. A, flowchart of the strategy used to define a predictive model for response to 5-FU–based chemotherapy. Candidate genes were selected from gene expression profiles of each human colorectal adenocarcinoma cell line. The training set of cell lines selected represents the extremes of sensitivity or resistance to 5-FU. A transcription site activation profile of candidate genes was determined for each cell line. Using leave-one-out analysis, a predictive model that classified the training set of cell lines as resistant or sensitive to 5-FU with the highest accuracy was derived. The predictive marker genes were evaluated for their ability to accurately classify a panel of independent test cell lines as 5-FU–resistant or–sensitive in a blinded study. B, detection of an active transcription site for the gene MRGX in an individual human colorectal adenocarcinoma cell (DLD-1). Nuclei are stained with DAPI and sites of transcription are detected with fluorescent probes labeled in Cy3 (red) and Cy5 (green). Inset, close-up of area of nucleus with both Cy3 and Cy5 probes bound to nascent transcripts. Scale bar, 6 μm. C, transcription site activation profile of 5-FU–resistant and 5-FU–sensitive colorectal tumor cell lines as measured by FISH for nascent mRNAs. Analysis of active transcription sites for each candidate gene in individual cells provides a transcriptional profile for each cell line. Red bars, candidate genes correlated with 5-FU resistance; blue bars, candidate genes correlated with 5-FU sensitivity. Columns, mean for three experiments; bars, SE.

Figure 2

Chemotherapy indicator plot. A, two genes that are poor predictive markers of response to 5-FU treatment. Filled squares, cell lines known to be sensitive. Filled cirlces, cell lines known to be resistant. The decision line is an average of 12 decision boundaries generated from leaving out each of the 12 samples from the training set once. The large error in the decision boundary signifies the dependency of the model on a single sample in the training set. B, the four genes, MRGX, TYMS, BAK, and ATP7B are identified as good predictive markers of response to 5-FU treatment. C, performance of biomarkers in an independent set of blinded test cell lines. Test cell lines A and D, corresponding to RKO and HCT116, respectively, were classified as 5-FU–sensitive (P = 0.05 and P = 0.0005, respectively). Test cell line B, corresponding to SW620 was classified as 5-FU–resistant (P = 0.023). C, the fourth cell line, HCT15, was also classified as 5-FU–resistant (P = 0.099).

Figure 3

Detection of active transcription sites for 5-FU marker genes in paraffin-embedded human colon tumor TMA. A, merge of DAPI, Cy3, and Cy5 channels. Image shows DAPI-stained nuclei containing transcription sites (arrows) for MRGX and TYMS. Scale bar, 5 μm. B, merge of DAPI, Cy3, and Cy5 channels. Image shows DAPI-stained nuclei containing transcription sites (arrows) for ATP7B and BAK. Scale bar, 5 μm. C, active transcription site profile for 5-FU marker genes in colon tumor biopsies from individual patients as measured by RNA FISH. Red bars, genes correlated with 5-FU resistance. Blue bars, genes correlated with 5-FU sensitivity. Columns, mean for three sections from each individual tumor; bars, SE. D, chemotherapy indicator plot for tumors from 15 anonymous patients. Filled diamonds, tumor samples with unknown clinical outcomes. Our predictive model classified 11 samples as sensitive and 2 samples as resistant. The model was unable to classify the remaining two samples with significant confidence.

Figure 4

Prediction of response to 5-FU–based chemotherapy in colon cancer patients. A, active transcription sites for 5-FU marker genes in paraffin-embedded human colon tumor tissues. Image shows DAPI-stained nuclei containing transcription sites for ATP7B and BAK (both shown in green). Scale bar, 5 μm. B, active transcription site profile for 5-FU marker genes in colon tumor samples from seven patients as measured by RNA FISH. The seven patients are designated as follows: patient #1F, female, age 60 y, tumor stage T₃N₂M_x (poorly differentiated adenocarcinoma); patient #4F, male, age 56 y, tumor stage T₃N₁M_x (poorly differentiated mucinous adenocarcinoma); patient #6F, male, age 33 y, unknown tumor stage (metastastic adenocarcinoma); patient #1N, male, age 62 y, tumor stage T₃N₁M_x (well to moderately differentiated adenocarcinoma); patient #4N, female, age 67 y, tumor stage T₃N₂M_x (moderately differentiated adenocarcinoma); patient #5N, female, age 56 y, tumor stage T₃N₂M_x (moderately to poorly differentiated adenocarcinoma); patient #6N, female, age 42 y, tumor stage T₃N₁M_x (moderately differentiated adenocarcinoma). Red bars, genes correlated with 5-FU resistance. Blue bars, genes correlated with 5-FU sensitivity. Columns, mean for six fields from each individual tumor; bars, SE. C, chemotherapy indicator plot for tumors from seven anonymous patients. Filled squares, patients known to be sensitive. Filled circles, patients known to be resistant. Our predictive model classified three samples as resistant and four samples as sensitive.