Abstract
Although the progression of mutated colonic cells is dependent upon interactions between the initiated epithelium and surrounding stroma, the nature of these interactions is poorly understood. Here, the development of an ultrasensitive laser capture microdissection (LCM)/RNA-seq approach for studying the epithelial and stromal compartments of aberrant crypt foci (ACF) is described. ACF are the earliest identifiable preneoplastic lesion found within the human colon and are detected using high-definition endoscopy with contrast dye spray. The current analysis focused on the epithelium of ACF with somatic mutations to either KRAS, BRAF, or APC, and expression patterns compared with normal mucosa from each patient. By comparing gene expression patterns among groups, an increase in a number of proinflammatory NF-κB target genes was identified that was specific to ACF epithelium, including TIMP1, RELA, and RELB. Distinct transcriptional changes associated with each somatic mutation were observed and a subset of ACF display BRAFV600E-mediated senescence-associated transcriptome characterized by increased expression of CDKN2A. Finally, LCM-captured ACF-associated stroma was found to be transcriptionally distinct from normal-appearing stroma, with an upregulation of genes related to immune cell infiltration and fibroblast activation. Immunofluorescence confirmed increased CD3+ T cells within the stromal microenvironment of ACF and an abundance of activated fibroblasts. Collectively, these results provide new insight into the cellular interplay that occurs at the earliest stages of colonic neoplasia, highlighting the important role of NF-κB, activated stromal fibroblasts, and lymphocyte infiltration.
Implications: Fibroblasts and immune cells in the stromal microenvironment play an important role during the earliest stages of colon carcinogenesis. Mol Cancer Res; 14(9); 795–804. ©2016 AACR.
This article is featured in Highlights of This Issue, p. 765
Introduction
Reciprocal interactions between transformed epithelial cells and their associated stroma are a critical determinant of neoplastic growth (1). Stromal cells constitute a large fraction of solid tumors and are important drivers of the carcinogenic process (2). The stroma that is in direct contact with transformed epithelium becomes “activated,” displaying an altered phenotype that produces growth-promoting factors that enhances cancer progression (3, 4). However, little is known regarding the importance of epithelial–stromal interactions at the very early stages of neoplasia when critical steps in cancer progression occur and targeted interventions may elicit the greatest benefit.
Aberrant crypt foci (ACF) are the earliest morphologically identifiable mucosal abnormality found in the human colon (5), a subset of which may progress to colorectal cancer. Although abundant in the distal colon, ACF in the proximal colon are relatively infrequent (<1 per colon) but are more likely to progress to malignancy. Proximal ACF share many molecular features with more advanced neoplasia (6), including somatic mutations to KRAS, BRAF, and APC (7) and may be precursor lesions of “interval” colon cancers that develop in the proximal colon between scheduled colonoscopies. The current study focuses on a detailed characterization of these diminutive (<5 mm) proximal lesions identified by high-definition, magnifying chromoendoscopy. By developing a more comprehensive understanding of epithelial–stromal interactions occurring in these very early lesions, it may be possible to develop a better understanding of how their progression is regulated. This knowledge could ultimately be utilized to identify high-risk patient populations and to develop strategies for preventing cancer development, particularly interval cancers of the proximal colon.
Materials and Methods
Clinical specimens
All colon specimens used in this study were obtained from patients enrolled in an ongoing ACF study at UConn Health-John Dempsey Hospital (Farmington, CT) between 2010 and 2014. Samples were freshly isolated during the colonoscopy procedure and immediately frozen in OCT-embedding media. Each subject provided written informed consent prior to inclusion in the study. High-definition chromoendoscopy was performed in the distal 20 cm of the colorectum and throughout the entire proximal colon using 0.1% indigo carmine dye spray for contrast enhancement. The identification and histologic evaluation of ACF has been described previously (8, 9). In addition, each subject had a histologically confirmed corresponding normal biopsy specimen removed from the same side of the colon, generally within 2 cm of the ACF biopsy. Mutational spectra of ACF were determined using DNA–MS analysis (Sequenom; ref. 8). Only ACF with confirmed, nonoverlapping somatic mutations to either KRAS, BRAF, or APC were selected for further analysis. The study was approved by the UConn Health IRB (#IE-10-068OS-3) in accordance with NIH human research study guidelines.
Laser capture microdissection and RNA extraction
Highly enriched epithelial and stromal RNA samples were obtained by laser capture microdissection (LCM) using an ArcturusXT (Thermo Fisher Scientific). Frozen sections were routinely cut at 9-μm thickness on PEN membrane slides and stored at −80°C until use. After 15 seconds of air drying, sections were rehydrated and dehydrated sequentially in 75% ethanol (EtOH), ddH2O, ddH2O, 75% EtOH, 95% EtOH, and 100% EtOH for 30 seconds each. Each wash solution was treated with ProtectRNA RNase Inhibitor (Sigma-Aldrich). Following the final 100% EtOH wash, slides were washed in xylenes for 5 minutes. Serial sections were laser captured following the manufacturers' protocol. RNA was extracted using the Arcturus PicoPure Frozen RNA Isolation Kit and quantified using a Qubit 3.0 Fluorimeter. The quality of a subset of RNA samples was tested using a 2100 Bioanalyzer (Agilent Technologies).
Ion personal genome sequencing
Sequencing libraries were prepared using two panels: the Ion AmpliSeq RNA Apoptosis Panel, targeting 267 genes involved in the cellular apoptosis pathway, and a customized Ion AmpliSeq Senescence Panel targeting 20 known senescence genes. The following genes were included in the senescence panel: CDKN2a (p16), CDKN2A (p14), MAPK12, MAP2K6, CCND1, BMI1, BRAF, SKP2, E2F1, CDKN1B, MAPKAPK5, MAPK14, MAP2K3, VHL, BACH1, CCNE1, MAPK11, MDM4, NPM1, MAPK13, and MKI67. This panel included specific primers capable of differentiating between the p16 and p14 alternate reading frames of CDKN2A (10). Genes common to both panels were used for normalization (11, 12). Libraries were barcoded using the Ion Xpress Barcode Adapters (ThermoFisher). After barcoding, libraries were quantified using a Bioanalyzer and quantitative PCR, diluted to 25 pmol/L, and pooled in batches of three to four per 318 PGM chip (ThermoFisher). The libraries were incorporated during template preparation using the Ion PGM IC 200 Kit using an IonChef and sequenced, resulting in 800,000 to 1,200,000 reads per panel library with a mean read-depth of 8× for the apoptosis panel and more than 100,000× for the senescence panel. Four samples from one batch belonging to three stromal pairs were excluded from the analysis due to poor library preparation (Supplementary Fig. S1A and S1B). Six gene targets were removed prior to analysis due to possible off-target pseudogene amplification that was identified by GenBank (Supplementary Fig. S1C and S1D). Raw and processed data were uploaded to GEO (GSE77312).
Statistical analysis
Statistical analysis was performed using the Torrent Suite (4.4.0) and the Coverage Analysis and AmpliseqRNA plug-ins. Genes were considered differentially expressed with a fold change >1.5 or <0.66. Gene set enrichment analysis (GSEA) software was downloaded from the Broad Institute GSEA portal (http://www.broadinstitute.org/gsea/index.jsp) and applied to the sample set using each of the genes from both the apoptosis and senescence panels. For the analysis, the GSE77312 dataset was partitioned into the following four subgroups: ACF epithelial cells, ACF stromal cells, normal epithelial cells, and normal stromal cells. GSEA was carried out for all gene lists using signal-to-noise ratios to rank genes. The significance of enrichment was estimated using 1,000 gene permutations, with a statistical cutoff of 0.05 for the nominal P value. All other statistical analyses were performed with GraphPad Prism version 5.0 or in R (3.1.2). All P values are two sided. P <0.05 was considered statistically significant.
Immunofluorescence and quantification
Frozen tissue sections (9-μm thick) were collected on Superfrost (ThermoFisher) slides. All immunofluorescence reactions of the same type were performed in a single batch. The following primary antibodies were used: rabbit antibody to alpha smooth muscle actin (polyclonal; Abcam; Ab5694; 1:200 dilution), CD3+ (monoclonal; eBioscience; UCHT1, 1:200 dilution), and mouse antibody to vimentin [V9] (monoclonal IgG1; Santa Cruz Biotechnology; sc-6260; 1:200 dilution). The following secondary antibodies were used: goat anti-rabbit IgG conjugated to Alexa Fluor 568 (ThermoFisher; A-11036) and goat anti-mouse IgG conjugated to Alexa Fluor 488 (ThermoFisher; A-11001). Briefly, tissue sections were fixed for 30 minutes in acetone at −20°C and blocked in PBS containing 1% BSA (Sigma). Sections were washed with PBS and incubated overnight with primary antibody solution (in PBS with 0.1% BSA). Next, sections were washed and incubated with secondary antibodies (in PBS with 0.1% BSA). The nuclei counterstain was performed with DAPI (100 ng mL−1 in dH2O) for 10 minutes. Slides were washed and mounted in VECTASHIELD HardSet Medium (Vector Laboratories; H-1400). Imaging was done on an Olympus fluorescence microscope with a 40× and 10× objective, using QCapture Pro 7 imaging software (Qimaging). Images were processed and combined using the ImageJ and Photoshop software programs. Activated fibroblasts are commonly identified by their expression of α-SMA (13, 14). However, pericryptal myofibroblasts also express α-SMA and are normally present within the stromal cell population that line the base of colonic crypts (15). Although, pericryptal myofibroblasts are capable of becoming activated, only nonpericryptal α-SMA/vimentin fibroblasts were enumerated in the current study differentiated on the basis of histologic landmarks. To quantify activated fibroblasts within each sample, three independent 40× fields that consisted of more than 50% stroma area were taken from each slide. The number of activated fibroblasts was enumerated by counting the number of dual positive α-SMA and vimentin cells per field. The total area of stroma was calculated by creating an image mask of each field in Photoshop. Activated fibroblast density was calculated by dividing the number of activated fibroblasts by stromal area. Quantification of CD3+ cells followed an identical process but used biopsy samples from different patients due to primary sample depletion.
Results
ACF epithelium and stroma display distinct expression profiles
Our sample set is comprised of 12 ACF biopsies that were collected from the proximal colons of 10 patients. A representative ACF identified in the proximal colon is shown in Fig. 1A. ACF were histologically confirmed and then further selected on the basis of their mutation status as determined by DNA-MS sequencing. Each biopsied ACF had only one detectable somatic mutation in either KRAS (KRASG12D, KRASG12V), BRAF (BRAFV600E), or APC (APCR876*, APCR1450*; Supplementary Table S1). In addition, a corresponding normal biopsy specimen was removed from the same side of the colon, generally within 2 cm of the ACF biopsy and confirmed as histologically normal. As shown in Fig. 1B–F, colonic crypts and stroma were outlined by the LCM software and populations of stromal and epithelial cells were selectively laser captured for analysis. Because the activation of apoptosis and senescence signaling represents critical barriers for early neoplastic progression, RNA extracted from laser-captured cells was subjected to targeted RNA sequencing (RNA-seq) using a panel of 279 genes that cover these pathways. Targeted AmpliSeq panels were used for this study because of their increased sensitivity, providing the ability to quantify even low abundance transcripts. After completing the targeted transcriptomics analysis, principal component analysis (PCA) was used to evaluate broad transcriptional differences across all samples (Fig. 1G). Epithelial and stromal samples were clearly delineated (dotted lines), clustering independently during subanalyses (Fig. 1H and I). These results demonstrate that the epithelium and stromal cells of an ACF are transcriptionally distinct from nearby normal mucosa.
LCM workflow for sample collection and analysis. A, a representative ACF identified in the proximal colon of a patient during high-definition chromoendoscopy (red arrow), using 0.1% indigo carmine dye spray for contrast enhancement. B, H&E slide of a dysplastic ACF harboring an APC r876* mutation. C, unstained PEN serial section of the same sample viewed from the internal camera of the Arcturus XT with slide mark-up indicated. Two capture groups were used: epithelial crypts (red) and the surrounding stroma (green). D, an ultraviolet laser was used to cut the tissue from the slide and epithelial (E) and stromal cells (F) were captured onto separate caps for downstream RNA isolation, library preparation, and sequencing. G, 3D PCA of the RNA-seq data that characterizes the trends exhibited by the expression profiles of ACF epithelium (red), matched normal epithelium (orange), ACF- associated stroma (green), and matched normal stroma (blue). Each sample is represented by a single dot. In 3D plot of epithelial (H) and stromal (I) samples, ACF separate from matched normal.
LCM workflow for sample collection and analysis. A, a representative ACF identified in the proximal colon of a patient during high-definition chromoendoscopy (red arrow), using 0.1% indigo carmine dye spray for contrast enhancement. B, H&E slide of a dysplastic ACF harboring an APC r876* mutation. C, unstained PEN serial section of the same sample viewed from the internal camera of the Arcturus XT with slide mark-up indicated. Two capture groups were used: epithelial crypts (red) and the surrounding stroma (green). D, an ultraviolet laser was used to cut the tissue from the slide and epithelial (E) and stromal cells (F) were captured onto separate caps for downstream RNA isolation, library preparation, and sequencing. G, 3D PCA of the RNA-seq data that characterizes the trends exhibited by the expression profiles of ACF epithelium (red), matched normal epithelium (orange), ACF- associated stroma (green), and matched normal stroma (blue). Each sample is represented by a single dot. In 3D plot of epithelial (H) and stromal (I) samples, ACF separate from matched normal.
Proximal ACF activate antiapoptotic pathways and upregulate NF-κB family genes
To further assess the signaling pathways responsible for the differential clustering observed in the PCA, gene expression patterns were evaluated in laser-captured epithelial cells from ACF and normal mucosa. Differential expression analysis identified 50 significantly altered genes between ACF and normal crypts (Fig. 2A; Supplementary Table S2). To evaluate these changes, GSEA was utilized to identify enriched signatures within the ACF epithelial cells compared with normal epithelia. In general, ACF epithelium demonstrate downregulation of genes involved in apoptosis, referred to as the “apoptosis reactome”, and upregulation of the NF-κB pathway (Fig. 2B). Overall, 14 proapoptotic genes were downregulated in ACF. The most significantly downregulated (6.9-fold, P = 0.00009) gene was hara-kiri (HRK). HRK encodes a protein that is a potent activator of apoptosis that is frequently downregulated in a variety of cancers and is hypermethylated in colorectal cancer. In addition, cytochrome C (CYCS) and caspase-3 (CASP3) expression were downregulated relative to normal (1.5-fold; P = 0.05 and 1.8-fold; P = 0.004, respectively). Among the 10 antiapoptotic genes that were upregulated in ACF epithelial cells, two important NF-κB gene family members, RELA and RELB, were increased by approximately 2-fold (P = 0.00001) and 1.8-fold (P = 0.02), respectively. Several additional NF-κB activators and target genes were also upregulated in ACF. Caspase recruitment domain family member 6 (CARD6) was increased by 3.7-fold in ACF relative to normal (P = 0.02). The mRNA for receptor-interacting serine/threonine protein kinase 2 (RIPK2), known to directly interact with CARD6 (16), was also increased (1.6-fold, P = 0.01). The most significantly upregulated gene present in proximal ACF was tissue inhibitor of metalloproteinase 1 (TIMP1), which was significantly increased (4.7-fold; P = 0.0004). TIMP1 is a key NF-κB target gene that plays a central role in mediating NF-κB activity, as well as contributing to the maintenance of the extracellular matrix.
Comparison of normal and ACF epithelial gene expression patterns. A, heatmap shows 50 significantly altered genes between ACF (red) and normal (blue) epithelium. The TIMP-1 gene shows the highest upregulation (4.7-fold, P = 0.0004), and the HRK gene is the most downregulated (6.8 fold, P < 0.01). B, GSEA identifies multiple differentially expressed gene sets. ACF are negatively enriched for genes within the apoptosis reactome, while positively enriched for genes in the NF-κB pathway. NES, normalized enrichment score.
Comparison of normal and ACF epithelial gene expression patterns. A, heatmap shows 50 significantly altered genes between ACF (red) and normal (blue) epithelium. The TIMP-1 gene shows the highest upregulation (4.7-fold, P = 0.0004), and the HRK gene is the most downregulated (6.8 fold, P < 0.01). B, GSEA identifies multiple differentially expressed gene sets. ACF are negatively enriched for genes within the apoptosis reactome, while positively enriched for genes in the NF-κB pathway. NES, normalized enrichment score.
Mutation-specific transcriptional patterns in ACF
Interindividual variation in gene expression patterns often complicates the results of expression profiling analyses. Thus, to develop a quantitative assessment of mutation-specific effects, potential confounders were controlled for by normalizing each sample to its corresponding matched normal procured from the same patient. After data normalization, we then compared gene expression profiles in ACF between APC versus BRAF or KRAS to determine whether differentially expressed genes may be segregated based upon their mutation status. As shown by the heatmap results depicted in Fig. 3A, a total of 31 genes were differentially expressed between the groups. GSEA (Fig. 3B, left) identified a significant enrichment of MYC target genes (MYC is upregulated 3.5-fold, P = 0.03; Supplementary Table S3), as well as genes within the RB-pathway and the hepatoblastoma gene set in ACF harboring APC mutations. In addition, the M2342 gene set was downregulated in APC-mutant ACF (17). This finding is consistent with earlier studies reporting differential gene expression that is associated with β-catenin inactivation (NES = −2.03, P < 0.001). Analyses of KRAS- and BRAF-mutant ACF epithelia showed up-regulation of gene sets involved in aberrant DNA methylation. In contrast, reduced gene expression was observed for gene sets associated with cell cycling and the telomerase pathway (Fig. 3B, right). Collectively, these findings suggest that somatic mutations in ACF are capable of enforcing transcriptional changes that promote critical cellular processes associated with more advanced neoplasia. Several genes that did not fall into significantly enriched GSEA categories were associated with NF-κB signaling (Supplementary Table S3). For example, the expression of RELB (2.9-fold, P = 0.02) and its downstream effector, BIRC3 (2.9-fold, P = 0.04) were both significantly altered in KRAS and BRAF-mutant ACF in comparison with APC-mutant epithelia. Interestingly, the most upregulated gene (9.5-fold; P = 0.002) in BRAF- and KRAS-mutant ACF epithelia was the secreted tissue inhibitor of metalloproteinase, TIMP2, which plays a key role in antitumor transcriptional activity within the tumor microenvironment.
Oncogenic mutations activate distinct transcriptional profiles. ACF transcriptomes were normalized to patient controls to identify mutation-specific changes. A, heatmap identifies 32 genes differentially expressed genes between RAS-activated ACF and APC-mutant ACF. The TIMP-2 gene shows the highest upregulation (9.5-fold, P = 0.002). B, left, APC-mutant ACF were significantly enriched for gene sets, including MYC target genes, RB pathway, and hepatoblastoma, a pediatric liver cancer that has abnormal Wnt/B-catenin signaling and is associated with individuals with familial adenomatous polyposis; right, KRAS- and BRAF-mutant ACF upregulate gene sets related to aberrant methylation. In addition, gene sets for cell-cycle and telomerase pathway were significantly downregulated, which may be indicative of a senescence phenotype. NES, normalized enrichment score.
Oncogenic mutations activate distinct transcriptional profiles. ACF transcriptomes were normalized to patient controls to identify mutation-specific changes. A, heatmap identifies 32 genes differentially expressed genes between RAS-activated ACF and APC-mutant ACF. The TIMP-2 gene shows the highest upregulation (9.5-fold, P = 0.002). B, left, APC-mutant ACF were significantly enriched for gene sets, including MYC target genes, RB pathway, and hepatoblastoma, a pediatric liver cancer that has abnormal Wnt/B-catenin signaling and is associated with individuals with familial adenomatous polyposis; right, KRAS- and BRAF-mutant ACF upregulate gene sets related to aberrant methylation. In addition, gene sets for cell-cycle and telomerase pathway were significantly downregulated, which may be indicative of a senescence phenotype. NES, normalized enrichment score.
BRAF-mutant ACF are associated with upregulation of senescence-associated genes
Distinct carcinogenic pathways dependent upon BRAF or KRAS-activating mutations have been described in the colon. Thus, the possibility was tested that KRAS and BRAF may activate different transcriptional networks that can be identified even at this early stage of neoplastic transformation. To illustrate these differences, a z-score transformation was applied to gene expression data from laser-captured epithelial cells. As shown in Fig. 4A, a total of 43 differentially expressed genes were identified. Within this panel, 42 genes showed expression differences in the same direction when compared with normal epithelia (Supplementary Table S4), although in each case, the magnitude of change was higher in BRAF- compared with KRAS-mutant ACF. The one exception was p16 (CDKN2A), encoding the tumor suppressor protein (p16INK4A). p16 affords a critical barrier to cancer progression and has been implicated in the “alternative” colorectal cancer pathway (18). Transcripts for p16 were significantly higher (3.5-fold; P = 0.03) in BRAF-mutant ACF compared with both KRAS- and APC-mutant epithelial cells (Fig. 4B). GSEA was also applied to the BRAF- and KRAS-mutant ACF. As shown in Fig. 4C, three distinct gene sets were preferentially activated in BRAF-mutant ACF: histone deacetylase 1 (HDAC1), glioblastoma mesenchymal subtype, and a “cellular transformation” signature, whereas KRAS-mutant ACF were enriched for the PI3K cascade.
KRAS- and BRAF-mutant ACF are transcriptionally distinct. A, left, heatmap compares normalized gene expression profiles between KRAS- and BRAF-mutant ACF. Forty-three genes were differentially expressed between the two groups; right, z-score transformation of gene expression data. B, p16INK4a mRNA is significantly upregulated in BRAF-mutant ACF compared with both KRAS- and APC-mutant ACF. C, GSEA for BRAF-mutant ACF identified enrichment in gene sets, including HDAC1 targets, glioblastoma mesenchymal subtype, a cellular transformation signature, and a negative enrichment for the PI3K cascade. NES, normalized enrichment score.
KRAS- and BRAF-mutant ACF are transcriptionally distinct. A, left, heatmap compares normalized gene expression profiles between KRAS- and BRAF-mutant ACF. Forty-three genes were differentially expressed between the two groups; right, z-score transformation of gene expression data. B, p16INK4a mRNA is significantly upregulated in BRAF-mutant ACF compared with both KRAS- and APC-mutant ACF. C, GSEA for BRAF-mutant ACF identified enrichment in gene sets, including HDAC1 targets, glioblastoma mesenchymal subtype, a cellular transformation signature, and a negative enrichment for the PI3K cascade. NES, normalized enrichment score.
ACF-associated stroma is proinflammatory with the presence of activated stromal fibroblasts
Among the differentially expressed genes identified within laser-captured epithelial cells was a subset that encodes secreted proteins. This observation suggests that early transformed epithelial cells may impart signals that directly contribute to the reorganization of the underlying stromal microenvironment. Thus, LCM was used to capture populations of stromal cells in direct contact with the transformed epithelia. As shown by the heatmap depicted in Fig. 5A, supervised clustering analysis revealed a total of 68 differentially expressed genes (Supplementary Table S5). Six of 9 stromal samples (left side) laser captured from ACF clustered independently from normal stromal samples.
ACF-associated stroma has an altered transcriptome with increased immune infiltration. A, heatmap showing supervised clustering analysis. Sixty-eight genes were differentially expressed between ACF stromal samples and normal controls. Six of 9 samples demonstrate a distinct transcriptional profile from normal controls. B, GSEA determined that cells isolated from the ACF stroma are enriched for gene sets for secreted factors, changes to the extracellular matrisome, and enrichment for the senescent fibroblast phenotype. C, several proinflammatory genes are significantly upregulated in ACF stroma and are leukocyte and lymphocyte specific. D, immunofluorescence of normal and ACF sections confirm increased abundance of CD3+ cells within ACF stroma. Epithelial crypt area is labelled “C” and separated by dashed lines. E, quantification of CD3+ cells from 15 high magnification (40×) fields from five ACF and normal pairs. NES, normalized enrichment score.
ACF-associated stroma has an altered transcriptome with increased immune infiltration. A, heatmap showing supervised clustering analysis. Sixty-eight genes were differentially expressed between ACF stromal samples and normal controls. Six of 9 samples demonstrate a distinct transcriptional profile from normal controls. B, GSEA determined that cells isolated from the ACF stroma are enriched for gene sets for secreted factors, changes to the extracellular matrisome, and enrichment for the senescent fibroblast phenotype. C, several proinflammatory genes are significantly upregulated in ACF stroma and are leukocyte and lymphocyte specific. D, immunofluorescence of normal and ACF sections confirm increased abundance of CD3+ cells within ACF stroma. Epithelial crypt area is labelled “C” and separated by dashed lines. E, quantification of CD3+ cells from 15 high magnification (40×) fields from five ACF and normal pairs. NES, normalized enrichment score.
GSEA identified the enrichment of gene sets for the “extracellular matrisome,” “secreted factors”, and the “senescent fibroblast phenotype” (Fig. 5B). Although the changes were only modest, a number of genes involved in proinflammatory signaling and fibroblast activation were identified. For example, LTA (3.2-fold; P = 0.012), PIK3CD (4.0-fold; P = 0.011), BCL2A1 (3.2-fold; P = 0.02), TBK1 (1.9-fold; P = 0.04), and TRAF4 (1.8-fold; P = 0.04) were each significantly upregulated (Fig. 5C). These results suggest an influx of proinflammatory cells and a release of potent cytokines into the ACF-associated stromal microenvironment. In fact, immunofluorescence analysis of ACF-associated stroma revealed an increase (2.4-fold; P = 0.002) in the number of CD3+ T lymphocytes compared with normal tissue (Fig. 5D and E). The presence of CD3+ T lymphocytes was limited to adjacent stromal tissue with no observable infiltration within the epithelium.
Another gene within the ACF stroma that was significantly differentially expressed was BDNF (4.3-fold; P = 0.013). BDNF has been shown to be secreted by senescence-activated fibroblasts. In addition, several senescence-associated genes were downregulated, including TERT (6.2-fold; P = 0.03) and KI67 (3.8-fold; P = 0.03). ACF-associated stroma also showed increased expression levels of the IAP proteins, X-IAP (1.9-fold; P = 0.02) and Ts-IAP (3.6-fold; P = 0.02; Fig. 6A). These genes are known to contribute to tumor progression via activation of NF-κB and contribute to fibronectin expression, which may increase fibroblast activation (19, 20). Thus, to confirm the presence of activated fibroblasts within the ACF-associated stroma, colon sections were stained for vimentin and α-SMA (Fig. 6B). As shown in Fig. 6C (top), quantification of serial sections identified a significant increase in the density of activated fibroblasts (3.0-fold; P = 0.00001) compared with normal stroma. However, there were no detectable differences in the number of activated fibroblasts in ACF with distinct somatic mutations (KRAS, BRAF, and APC; Fig. 6C, bottom).
Fibroblasts are activated within ACF stroma. A, five genes related to activated senescent fibroblasts were upregulated in ACF stroma. B, immunofluorescence analysis of normal (n = 10) and ACF (n = 11) sections confirm the presence of activated fibroblasts within the ACF-associated stroma. Epithelial crypt area is labeled “C,” and stromal area is labeled with an “S.” Dual-labeled normal appearing myofibroblasts, which line the base on the crypt, are marked with yellow arrows, and activated fibroblasts are labeled with white arrowheads. White scale bar, 60 μm. C, left, quantification of activated fibroblasts from ACF samples performed using three independent 40× fields containing greater than 50% stromal area. Number of nonmyofibroblast dual costained cells were quantified and divided by their calculated stromal area to determine activated fibroblast density; right, normalizing each ACF (n = 11) by its matched normal did not reveal any difference by mutation type.
Fibroblasts are activated within ACF stroma. A, five genes related to activated senescent fibroblasts were upregulated in ACF stroma. B, immunofluorescence analysis of normal (n = 10) and ACF (n = 11) sections confirm the presence of activated fibroblasts within the ACF-associated stroma. Epithelial crypt area is labeled “C,” and stromal area is labeled with an “S.” Dual-labeled normal appearing myofibroblasts, which line the base on the crypt, are marked with yellow arrows, and activated fibroblasts are labeled with white arrowheads. White scale bar, 60 μm. C, left, quantification of activated fibroblasts from ACF samples performed using three independent 40× fields containing greater than 50% stromal area. Number of nonmyofibroblast dual costained cells were quantified and divided by their calculated stromal area to determine activated fibroblast density; right, normalizing each ACF (n = 11) by its matched normal did not reveal any difference by mutation type.
Discussion
Although considerable effort is currently being directed toward the development and application of targeted cancer therapies, preventing cancer offers the greatest promise for reducing overall cancer mortality rates. However, our understanding of the earliest events associated with neoplastic transformation remains limited, particularly in humans. To gain further insight into the earliest stages of colon cancer development, we have focused on ACF. ACF are precancerous epithelial growths commonly found in the human colon. Although ACF are frequently observed in the distal colorectum, they are far less frequent in the proximal colon, averaging less than one lesion per subject. However, the molecular, histologic, and demographic features of proximal ACF indicate that they are more likely to advance to a significant neoplasm. In the current study, we have taken advantage of the accessibility of these early proximal lesions to study molecular features of the epithelial–stromal interactions that regulate their progression. Understanding the nature of these interactions may ultimately uncover specific biomarkers and intercellular signaling pathways that could be monitored and targeted for highly effective cancer prevention. This information could be of central importance for proximal ACF, as these lesions may progress rapidly and lead to the formation of interval cancers in the proximal colon.
The reciprocal interaction between transformed epithelial cells and their associated stroma is an important determinant of early neoplastic growth. The stromal microenvironment within the colon is comprised of extracellular matrix components, secreted factors, and stromal cells, which include fibroblasts as well as resident and infiltrating immune cells. Previous studies of premalignant changes associated with cancer, including head and neck (21), breast (22), and ovarian (23) cancers, have found evidence for a proinflammatory stromal microenvironment. Characteristic changes include increased cytokine signaling and immune cell infiltration. In addition, a mouse model of squamous skin carcinogenesis demonstrated that fibroblasts isolated from the stroma of dysplastic lesions become activated prior to invasion. These activated dermal fibroblasts promote inflammation, mediating immune cell recruitment and altering tumor angiogenesis in an NF-κB–dependent manner (20). Furthermore, stromal fibroblasts associated with advanced colorectal cancer display genomic signatures capable of accurately predicting cancer relapse (24). However, the role of the stromal environment during the earliest stages of neoplastic transformation is less well defined in human. Using a combination of LCM and targeted RNA-seq against a total of 279 gene targets enriched for apoptotic and senescence signaling, we have characterized alterations to the transcriptome of epithelial and stromal cell populations obtained from the human colon. We have identified transcriptionally distinct gene signatures in ACF that cluster independently from normal mucosa and are further dependent upon underlying somatic mutations.
An important observation in the current study is that a single somatic mutation can activate distinct transcriptional changes within epithelial cells. For example, in ACF harboring an APC mutation (e.g., APCR876*, APCR1450*), gene sets for the RB and MYC pathways are significantly enriched (Fig. 3B), which is consistent with activation of the WNT pathway (25). This in turn leads to dysregulation of c-MYC with concomitant activation of the RB pathway (26, 27). As further evidence of mutation-specific effects, KRAS-mutant ACF upregulated a gene set for the PI3K/AKT/MTOR pathway more intensely than BRAF mutants, which is consistent with RAS being upstream of RAF in the growth factor receptor/MAPK signaling pathway. Furthermore, KRAS- and BRAF-mutant ACF displayed upregulation of gene sets related to aberrant methylation, a characteristic of the “alternative” pathway to colorectal cancer (28), where genome-wide hypermethylation may be a key driver to cancer progression. KRAS and BRAF mutations have been identified in both hyperplastic polyps and serrated adenomas with associated CpG island hypermethylation of promoter regions (29). Although several reports have described a limited constellation of DNA methylation changes in ACF, it would be interesting to examine the broader extent of epigenetic modifications in these early lesions. Finally, the presence of either KRAS or BRAF mutations were associated with a downregulation of a set of genes involved in the cell-cycle and telomerase pathway, suggesting that oncogene-induced senescence (OIS) programs may have been activated.
We further observed that senescence is particularly evident in BRAF-mutant ACF that expressed high levels of p16, an important mediator of OIS (Fig. 4B). In fact, OIS may provide an important mechanism for limiting the growth of these early preneoplastic lesions, leading in most cases to their regression and elimination. However, it is likely that not all BRAF/KRAS ACF that undergo OIS are eliminated. For example, a subset of melanocytic nevi, which typically undergo mutant BRAFV600E-induced senescence, is able to bypass senescence and progresses to melanoma (30). One method of senescence bypass is the induction of a secretory phenotype. Senescent cells can undergo changes in cellular metabolism, leading to increased production of secretory factors (31, 32). Within the colonic mucosa, we believe that these secretory factors may exert a protumorigenic influence upon the associated tissue microenvironment that promotes the progression of adjacent premalignant cells (33, 34). In contrast, these senescence changes were not observed in APC-mutant ACF, suggesting a fundamental difference in how these senescence pathways are triggered (35), and possibly underlie the greater malignant potential of APC-mutant lesions.
Activation of the proinflammatory transcription factor NF-κB represents an important mechanism contributing to cancer promotion. NF-κB activation typically occurs during early stages of carcinogenesis (36) and exerts an anti-apoptotic effect while promoting angiogenesis. In colorectal cancer, the NF-κB pathway can become constitutively activated via oncogenic activation of either KRAS or BRAF, while APC can directly regulate NF-κB activation through β-catenin (37). In the current study, we have found upregulation of several proinflammatory NF-κB–related target genes in ACF (Fig. 2B), indicating that the NF-kB/inflammation axis plays a pivotal role in regulating cancer development at the ACF stage.
Several studies have demonstrated that NF-κB signaling plays a fundamental role in OIS activation and the induction of a secretory phenotype. In the current study, we have observed a significant upregulation of secreted TIMP family members in ACF samples examined regardless of their underlying somatic mutation (Figs. 2A and 3A). TIMP-1 secretion by cancer cells induces transdifferentiation of normal fibroblasts into activated fibroblasts in xenograft mouse models and in coculture systems of liver, prostate (38), and colon cancers (39). We also observed TIMP-2 upregulation in BRAF and KRAS-mutant ACF, although the effect of TIMP-2 on tumor progression is less well defined (40). Some studies have correlated high levels of TIMP-2 with an adverse prognosis (41, 42), whereas other studies have observed an opposite effect (43–45). The general consensus, however, is that overexpression of TIMP family proteins elicits effects on ECM turnover and facilitates formation of the tumor microenvironment. These findings suggest that ACF epithelium may interact with stromal cells and induce remodeling of the underlying of the supportive microenvironment (46, 47).
Finally, by directly analyzing adjacent stromal tissue, we have demonstrated for the firsttime that reactive stromal changes accompany the earliest detectable stages of human colonic neoplasia (Figs. 5 and 6). We have observed a modest enrichment of genes reflective of a secretory milieu, with associated alterations to the extracellular matrix. Several genes within the secretory factor gene set were significantly elevated, characteristic of immune cell infiltration, and activation of stromal fibroblasts. Immune cell infiltration into ACF tissue was further demonstrated by CD3+ lymphocyte immunofluorescence. In addition to immune cells, fibroblasts are a major constituent of the extracellular matrix and may become activated by a complex array of stimuli arising during tissue injury. Upon activation, fibroblasts alter their cellular phenotype, increasing expression of α-SMA protein (48). The activated fibroblasts acquire a senescent and highly secretory phenotype, promoting carcinogenesis by the secretion of growth factors and inflammatory cytokines (32, 49). In our study, one key observation was the large upregulation of BDNF (Supplementary Table S5), which is secreted by activated fibroblasts (50). We confirmed the increased abundance of activated fibroblasts within ACF stroma by immunofluorescence of serial sections from the same samples. Taken together, our findings imply that an imflammatory gene expression profile by mutated epithelium, and/or defects in the epithelial barrier related to altered colonic cryptal structure, may trigger this stromal reaction.
In summary, our results indicate that preneoplastic cells within the human colon undergo mutation-specific transcriptional reprogramming. These changes are associated with the induction of epithelial–stromal interactions that may influence the outcome of these early neoplastic changes. Proinflammatory NF-κB pathway signaling in initiated colonic epithelium may mediate the activation of underlying stromal fibroblasts and recruitment of immune cells within the stromal microenvironment. Our results provide new insights into the dynamic, complex interplay that occurs in early colorectal neoplasia, highlighting the role of activated stromal fibroblasts and potential therapeutic targeting of NF-κB for prevention of early neoplasia.
Disclosure of Potential Conflicts of Interest
S. Jackson, K. Varma, and A. Carpino are employees of Thermo Fisher Scientific. No potential conflicts of interest were disclosed by the other authors.
Authors' Contributions
Conception and design: A. Mo, D.W. Rosenberg
Development of methodology: A. Mo, S. Jackson, K. Varma, D.W. Rosenberg
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Mo, S. Jackson, T.J. Devers, D.W. Rosenberg
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Mo, S. Jackson, D.W. Rosenberg
Writing, review, and/or revision of the manuscript: A. Mo, C. Giardina, T.J. Devers, D.W. Rosenberg
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Carpino, T.J. Devers, D.W. Rosenberg
Study supervision: K. Varma, D.W. Rosenberg
Acknowledgments
The authors thank Dr. Jeoffrey Schageman from Thermo Fisher Scientific for his bioinformatics expertise in identifying potential pseudogenes.
Grant Support
This work was supported by the State of Connecticut Department of Public Health, Biomedical Research Applications #2012-0913 and #2015-0901 (to D.W. Rosenberg), and NIH 1RO1CA159976 (to D.W. Rosenberg).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
References
Supplementary data
Molecular characteristics of Samples