Phenotypic and transcriptional fidelity of patient-derived colon cancer xenografts in immune-deficient mice

Abstract

Xenografts of human colorectal cancer (CRC) in immune-deficient mice have great potential for accelerating the study of tumor biology and therapy. We evaluated xenografts established in NOD/scid/IL2Rγ-null mice from the primary or metastatic tumors of 27 patients with CRC to estimate their capacity for expanding tumor cells for in vitro studies and to assess how faithfully they recapitulated the transcriptional profile of their parental tumors. RNA-seq analysis of parental human CRC tumors and their derivative xenografts demonstrated that reproducible transcriptional changes characterize the human tumor to murine xenograft transition. In most but not all cases, the human stroma, vasculature, and hematopoietic elements were systematically replaced by murine analogues while the carcinoma component persisted. Once established as xenografts, human CRC cells that could be propagated by serial transplantation remained transcriptionally stable. Three histologically atypical xenografts, established from patients with peritoneal metastases, contained abundant human stromal elements and blood vessels in addition to human tumor cells. The transcriptomes of these mixed tumor/stromal xenografts did not closely resemble those of their parental tumors, and attempts to propagate such xenografts by serial transplantation were unsuccessful. Stable expression of numerous genes previously identified as high priority targets for immunotherapy was observed in most xenograft lineages. Aberrant expression in CRC cells of human genes that are normally only expressed in hematopoietic cells was also observed. Our results suggest that human CRC cells expanded in murine xenografts have great utility for studies of tumor immunobiology and targeted therapies such as immunotherapy but also identify potential limitations.

Figure 1. Histologic features of representative parental human CRC tumors and their derivative xenografts in NSG mice.

Each row in (A)–(C) comprises micrographs of tissue sections from a single parental human tumor or murine xenograft. (A) Left column: tissue sections stained with hematoxylin and eosin (H+E). Subsequent columns, from left to right: tissue sections stained for human leukocyte antigen class I (HLA-ABC), epithelial marker E-cadherin, mesenchymal marker vimentin, endothelial marker PECAM1, and T-cell marker CD3. The rows, from top to bottom, show the histologic features of parental human tumors from D61540, P2726, and P2750, paired with their first-generation xenografts. The histology of the stroma-predominant xenograft developed from P2726 ascites fluid is also shown in the fifth row; the white arrow in the E-cadherin micrograph in this row indicates a small focus of tumor cells. Arrowheads in the PECAM1 micrograph for the P2750.Tu.X1 xenograft indicate areas with PECAM-1-immunoreactive cells. All images were obtained at 200x magnification. White scale bar in the upper left micrograph represents 200 µm.

Figure 2. Pairwise comparison of expression of human genes in CRC tumors and derivative xenografts.

Scatter plots display the normalized transcript counts (in counts per million) for individual human genes in the two samples that are indicated on the x- and y-axes. Red crosses indicate consensus housekeeping genes . Non-housekeeping genes are indicated by violet circles or yellow triangles: violet circles represent genes expressed at some level in both samples, and yellow triangles along the axes represent genes expressed in one sample but not the other. The Pearson correlation coefficient across all genes for each pairwise comparison is indicated in black above the upper left corner of each plot; the correlation for the subset of housekeeping genes is indicated in red. The diagonal line is the locus of points for which x = y. (A) Comparisons of the expression level of human genes in the two halves of the bisected colorectal tumor from D61540 (left panel) and in the ovarian and omental metastases from P2726 (right panel). (B) Comparisons of the expression level of human genes in three parental human tumors (D61540.T2, P2726.Ov, and P2750.Tu, from left to right) and in their first-generation xenografts (D61540.T2.X1, P2726.Ov.X1, and P2750.Tu.X1). (C) Comparison of the human gene expression levels in the atypical stroma-predominant xenograft established from P2750 and in its first-generation xenograft. (D) Mean growth kinetics of CXs (n = 5) derived from the ovarian and omental metastases of P2726 (P2726.Mets.X1) are indicated by black diamonds connected by the bold line, compared with the growth kinetics of the SX derived from the ascites fluid of P2726 (P2726.As.X1) indicated by the gray squares connected by the gray line. All xenografts were established by injecting 2×10⁶ cells into the flank of a NSG mouse. Bars indicate standard error of the mean.

Figure 3. Differential expression analysis of human genes in parental tumors and derivative xenografts.

(A) Conceptual model illustrating the bipartite classification of CRC xenografts as either Carcinoma Xenografts (CXs) or Stromal Xenografts (SXs). (B)Differential expression of all genes between the D61540.T1, D61540.T2, P2726.Om, and P2726.Ov parental human tumors (PHTs) and their derivative CXs D61540.T2.X1, D61540.T2.X2, P2726.Ov.X1, P2726.Ov.X2 (left Venn diagram), or between the P2726.Om, P2726.Ov, and P2750 PHTs and their derivative SXs P2726.As.X1 and P2750.Tu.X1 (right Venn diagram). The number following the “X” in each xenograft name indicates its passage generation. The number of genes with a Benjamini-Hochberg false discovery rate (FDR) of 0.05 or higher, which were interpreted as being “shared,” is indicated in the area of intersection in each Venn diagram. The number of genes with FDR <0.05, which are candidates for differentially expressed genes, are indicated under the labels PHT, CX, or SX. Listed in each Venn diagram are the 20 human genes with the lowest FDR. (C) Relative expression of orthologous human/murine genes in a representative xenograft. The symbols indicate the normalized human (x axis) and murine (y axis) transcript counts for genes with human and murine orthologues in the first generation xenograft from the P2726 ovarian CRC metastasis (P2726.Ov.X1). In the left plot, green ‘X’s indicates the read counts for genes that were preferentially expressed in SXs over PHTs (SX >PHT), and in the right plot yellow ‘+’s indicate the read counts for genes that were preferentially expressed in PHTs over CXs (PHT>CX). The read counts for all other orthologous human/murine gene pairs are indicated by blue circles for genes expressed in both and red triangles for genes expressed in one but not the other. The gray diagonal line represents 1∶1 correlation.

Figure 4. Cluster analysis of mouse and human orthologues in sequenced samples.

Pearson correlation was performed between all samples, and Euclidean distances were calculated. Unsupervised hierarchical cluster analysis of the human and murine transcriptomes, defined across 17,179 orthologous human-mouse gene pairs, in all of the parental tumors and derivative xenografts in which RNA-seq analysis was then performed. The height of the vertical dendrogram arms correlates with the global difference in the transcriptomes between any two samples. The height of the vertical dendrogram arms is proportional to the global differences in gene expression between samples in a cluster group. Along the bottom of the panel, samples are coded as follows: N = normal colon, PHT = parental human tumor, SX = stromal xenograft, CX = carcinoma xenograft, M = mouse stroma. Human and murine transcriptomes are indicated by ‘-H’ and ‘-M’ suffixes, respectively.

Figure 5. Expression of genes encoding high-priority cancer immunotherapy targets in parental tumors and xenografts.

(A) Expression of genes encoding high priority immunotherapy targets in parental tumors or xenografts (indicated by bold text along bottom), or of their murine orthologues in xenografts (indicated by italic text along bottom), is shown in a log-transformed heatmap, with unsupervised hierarchical cluster analysis performed on all samples (x-axis) and genes (y-axis). The total height of the vertical dendrogram arms separating any two samples is proportional to the differences in gene expression across all listed genes between those samples. The scale bar correlates normalized read counts to color. Along the bottom of the panel, samples are coded as follows: N = normal colon, PHT = parental human tumor, SX = stromal xenograft, CX = carcinoma xenograft, M = mouse stroma. (B) Ratio of expression of immunotherapy target genes in EPCAM⁺ cells from the P2726.Ov PHT and its derivative first- and ninth-generation xenografts to their expression in EPCAM⁻ cells from the P2726.Ov PHT. Expression was determined by real-time PCR and LinRegPCR, and normalized to housekeeping gene GAPDH. Down or up arrows indicate genes for which the minimal expression values have been substituted for undetectable transcript in the EPCAM⁺ or EPCAM⁻ fractions, respectively. The asterisk indicates a comparison for which both samples did not have detectable transcript. (C) Ratio of expression of human immunotherapy target genes and their murine orthologues, as inferred from RNA-seq data, in the first- and second-generation xenografts from the P2726.Ov. Down or up arrows indicate genes for which the minimum human or murine read counts, respectively, in each sample were substituted for read counts of 0.

Figure 6. Expression of human B and T cell receptor signaling-associated genes in tumors and xenografts.

(A) Expression of genes associated with B or T cell receptor signaling are shown in the log-transformed heatmap with unsupervised hierarchical cluster analysis performed on all samples (x-axis) and genes (y-axis). The total height of the vertical dendrogram arms separating any two samples is proportional to the differences in gene expression across all listed genes between those samples. The scale bar correlates normalized read counts to color. Along the bottom of the panel, samples are coded as follows: N = normal colon, PHT = parental human tumor, SX = stromal xenograft, CX = carcinoma xenograft. (B) Immunohistochemical staining for PDCD1 in the D61540 parental tumor (top) and its first-generation xenograft (bottom). White scale bar in top micrograph represents 200 µm. (C) Flow cytometric analysis of human PDCD1 (PD-1) and CTLA-4 expression by cells isolated from the D61540 first-generation xenograft.