Introduction
A growing body of evidence supports the notion that inflammation and colorectal cancer (CRC) are interrelated, including clinical observations and animal models [
1]. The colonic mucosa is in constant contact with a high density of diverse microorganisms [
2]. Antigens from these microbes are recognized by pattern-recognition receptors of the innate immune system. The toll-like receptor (TLR) family represents a critical part of this innate immune recognition, with each TLR recognizing pathogen-associated- or damage-associated-molecular patterns (DAMPs) [
3]. In particular, TLR4 recognizes lipopolysaccharide (LPS) from the outer membrane of Gram-negative bacteria, the most common type of colonic bacteria [
4]. Moreover, TLR4 is a receptor for DAMPs like hyaluronic acid and S100A9 [
5,
6]. Our laboratory has studied the role of TLR4 in intestinal inflammation and colitis-associated neoplasia, supporting the function of TLR4 as a tumor promoter in human tissue and murine models [
7,
8]. Our work in sporadic CRCs also links TLR4 to activation of the neoplastic Wnt/β-catenin pathway [
9].
In this study, we wished to characterize the role of TLR4 in the natural history of sporadic colonic neoplasia. The objective was to identify the prevalence of altered TLR4 RNA expression and tissue localization in sporadic neoplasia, and to determine the relationship between TLR4 expression and survival in CRC. We combined the strengths of transcriptomic array data and immunohistochemical (IHC) staining. Analysis of arrayed data offers a method by which to efficiently query the genomic and protein expression within a given tissue offering insight into the influence of gene expression on patient phenotypes. In an effort to establish the influence of TLR4 on CRC behavior, we drew upon genomic data sets and validated RNA expression profiles with immunofluorescent (IF) staining for theTLR4 protein from tissue microarrays (TMAs) obtained from the National Cancer Institute (NCI).
Our results demonstrate that TLR4 is expressed in adenomas and CRCs. Up to 47% of sporadic colon cancers express TLR4 protein with meaningful impact on survival and other clinical indices. Expression in tumors is localized predominantly in the stromal compartment, with a notable increase in pericryptal fibroblasts in the lamina propria.
Methods
Gene expression profiling
Gene expression arrays were identified by search of the Gene Expression Omnibus (GEO) database [
10]. Data sets were searched using the terms “colon cancer”, “colorectal cancer”, “rectal cancer”, “colon polyp”, and “colorectal neoplasia”. Searches were limited to expression data (messenger RNA). Data sets were included if they contained paired human samples ≥16 subjects, included accompanying clinical data, and had annotation files indicating TLR4. Studies were excluded if they used only animal or cell line models.
Keyword search (November 2011) revealed 170 CRC data sets. 97 pertained to human CRC, and 64 consisted of greater than or equal to 16 samples. 29 contained information on TLR4 expression with clinical characteristics of interest, including demographics, histologic progression of dysplasia, polyp size, histology, initial CRC stage, tumor grade, metastasis, survival (overall [OS], disease specific [DSS], disease free [DFS]), recurrence, and microsatellite instability (MSI).We then reorganized data by pairs of probes to observe the influence of varying transcript length on outcomes. Eleven studies were ultimately selected. A second GEO search was performed to identify series that stratified expression data by tissue compartment (ie, epithelium vs stroma) to further clarify TLR4 localization.
Tissue microarrays
TMA slides were provided by the NCI Cancer Diagnosis Program (CDP). Other investigators may have received slides from these same array blocks. The CDP arranged 279 colon tissue specimens with 182 CRCs of mixed stages and matched normal tissues on two slides [
11]. Neoplastic tissue consisted of adenomas (n = 19) and American Joint Committee on Cancer (AJCC) CRC stage I (n = 24), II (n = 61), III (n = 72), and IV (n = 25). Diverticulitis samples served as inflammatory, non-cancer controls. De-identified clinical data were provided by the CDP. Additional polyps with normal controls were stained on proprietary TMAs (US Biomax).
IF scoring
IF staining was performed on TMAs to detect human TLR4 (Novus Biologicals). Pan-cytokeratin was used as a counterstain to highlight intestinal epithelium (Abcam), and DAPI to counterstain nuclei. TLR4 detection was enhanced using conjugated Tyramide with the fluorochrome Alexa Fluor 488 (Invitrogen). Pan-cytokeratin was detected using an anti-rabbit secondary antibody conjugated with Alexa Fluor 647 (Invitrogen). Stained slides were scanned (Olympus VS120) and viewed using OlyVIA 2.4. A Leica TCS-SP5 Confocal was used for triple IF images. Staining patterns, intensity quantification, and extent TLR4 by surface area were determined by two senior GI pathologists (PAB and MTG) masked to diagnoses. A training subset was independently interpreted and inter-observer variation was determined. Moderate agreement was noted for the stromal score (weighted κ = 0.58 [95%CI 0.28-0.89]); moderate-to-strong agreement was observed for epithelium (weighted κ = 0.68 [95%CI 0.39-0.97]). Disagreement between scoring was settled by consensus. TLR4 signal intensity was scored in the stroma and epithelium. The signal intensity was scored as 0, no TLR4 staining; 1+, low intensity; 2+, moderate intensity; or 3+, high intensity. The extent of surface area with TLR4 was scored on a scale of 0–3 (0: no staining; 1+: present, but <20%; 2+: 20–50%; and 3+: >50%). A TLR4 positivity score was calculated by multiplying staining intensity and surface area data by tissue compartment (range: 0–9) [
7,
12,
13].
To qualify TMA observations, IHC was performed on normal colon, adenomas, and CRCs for TLR4 (Novus Biologicals), smooth muscle actin (α-SMA, Abcam), vimentin (Cell Signaling), and CD68 (Dako) on curls from tissue blocks. Secondary antibody conjugated with horseradish peroxidase was used prior to incubation with the substrate 3,3′-diaminobenzidine. Samples were counterstained with hematoxylin and scored (pathologist MTP).
Approval by the university’s Institutional Review Board was obtained.
Data analysis
Gene expression data
Analysis included quality control assessments of processed data. Differential expression discovery was performed using linear models and empirical Bayes methods (t-tests and ANOVA) via R statistical language [
14]. Survival analyses were conducted using Cox proportional hazards, with results corrected for multiple comparisons using false discovery rate procedures [
15]. Results were assessed for biological relevance. Where processed data were unavailable, raw data were analyzed with Robust Multichip Average using background adjustment, quantile normalization, and summarization; R language was used for Affymetrix chips and compared to processed samples. p-values <0.1 were considered significant. The p-value cut-off of 0.1 was selected as this value represents a favorable compromise between false positive and true positive rates in the setting of background “noise” associated with the identification of differentially expressed candidate RNAs with microarray data [
16].
Tissue microarray data
TLR4 staining intensity, surface area, and intensity score were correlated with clinico-pathologic endpoints. An arbitrary TLR4 intensity score of >3 was selected to denote positive TLR4 staining, with a score of >5 considered strongly positive. R software was used to reveal relationships according to guidance provided by the CDP [
11]. Non-parametric Wilcoxon sum-rank tests were performed for non-normal distributions.
Discussion
We have leveraged available transcriptome databases and well-designed TMAs to address the biologic role of TLR4 in colon dysplasia. The current work both confirms hypotheses engendered from our basic science work and generates new hypotheses about TLR4 signaling and sporadic CRC. In our animal models, we have found that mice constitutively expressing TLR4 have an increased severity of chemically-induced colitis and develop more colonic tumors [
8]. This tumor burden could be attenuated using TLR4-inhibiting antibody. Bringing relevance to humans with colitis-associated cancers (CACs), TLR4 is over-expressed in the majority, with increasing expression with dysplastic progression [
8]. We have further shown that TLR4 leads to activation of the Wnt/β-catenin pathway which is activated in most sporadic CRCs [
9]. Analogous to CACs, we have found an association between TLR4 expression in sporadic CRC and the progression of neoplasia, stage, DFS, and MSS. In particular, an increased expression of TLR4 in the tumor stroma relative to the malignant epithelium was noted. These expression data were validated with IHC showing a similar stroma:epithelium gradient. 35.6% of CRCs demonstrate high levels of TLR4 protein in the tumor stroma, while 3.45% have high levels in the tumor epithelium. Within the stroma, high TLR4 signal localized to the PCMs in particular.
Others using different methodology and smaller numbers demonstrated that TLR4 is associated with tumor stage. Cammarota, et al. have previously reported that stromal TLR4 expression in CRCs is associated with disease progression [
13]. In this series, CRC relapse was predicted by increased stromal TLR4 for stage pT3, lending credence to the predictive capability of this marker [
13]. Our study corroborated these findings using a larger sample of tissues, and answered the subsequent question of whether TLR4 transcripts can be associated with additional CRC endpoints. We confirmed that TLR4 transcript levels were related to colonic dysplasia, CRC stage, and survival.
In a separate series, Wang, et al. demonstrated high TLR4 expression in 20% of CRCs by immunostaining and its association with shorter OS. Both the expression of TLR4 and its co-receptor MyD88 were associated with the presence of liver metastases [
12]. In xenograft models of CRC, TLR4 silencing with RNA interference decreases the metastatic tumor burden in the liver [
32]. Proliferation of TLR4-expressing breast tumors has also been stunted with TLR4-inhibition
in vitro[
33]. In contrast, data from unrelated CRC cell line populations support the loss of expression or down-regulation of TLR4 in metastases compared to earlier stage tumors [
34]. The conflicting observations with respect to TLR4’s role in CRC metastases likely is a reflection of the biologic variation in CRCs, with TLR4 being over-expressed in a subset. Our study did not find a clear association with metastases.
Our study used IF and IHC to understand the location of TLR4 expression in colonic neoplasia. In agreement with Cammarota and Wang, we found that TLR4 protein expression in the stromal compartment was associated with more advanced stages of colon cancer. But we also found that normal stroma has TLR4 positive cells, largely CD68+ macrophages. Our transcriptome data demonstrated high TLR4 expression in adenomas relative to normal tissue and, to a lesser degree, higher expression relative to cancer. We speculate that adenomas may represent a more homogeneous tissue than cancer or that TLR4 plays an important role in tumor promotion from adenoma to cancer.
Our study and Cammarota found that stromal TLR4 expression is associated with cancer outcomes. In addition to the previous documentation of TLR4 expression by the submucosal vascular endothelium or hematopoietic mononuclear cells, our study demonstrated that PCMs also contribute to the TLR4 expression found in the stroma [
13]. These PCMs have previously been recognized as a discrete cell type in colonic adenomas, displaying a unique pattern of surface markers [
35,
36]. Increased density of these fibroblasts has been described in the stroma of digestive tract neoplasia [
37]. They may originate from deeper layers of the intestine, and have been proposed as tumor propagators via the epithelial-to-stromal transition [
38,
39]. We propose that TLR4 on the surface of these PCMs may play a role in the interaction between the epithelial and stromal compartments to promote neoplasia. These myofibroblasts have been shown
in vitro to respond to TLR signals and may therefore contribute to tumor promotion by secreting trophic factors in response to bacterial ligands [
40].
One of the interesting findings among the platforms containing multiple TLR4 probes was a marked divergence of transcripts with clinical outcomes. In particular, the direction and magnitude of specific TLR4 transcript expression on survival was evident, where TLR4 probes fall into two distinct groups, each of which targets a different transcript variant. There exist four recognized mRNA TLR4 products (Figure
1B) [
41]. Four probes from the commercial platform correspond to longer transcripts, while the remaining two probes are associated specifically with shorter mRNAs. The dichotomous relationship between RNA transcripts and clinical outcomes raises the possibility that different TLR4 transcripts or their relative ratios have different biological activities and consequences. The immunology literature supports the notion that alternative splicing of genes involved in innate immunity regulates their function [
42‐
44]. In particular, alternative splicing has been observed in TLR family members expressed in response to LPS [
43]. This splicing phenomenon may explain the opposing survival results observed herein. Epigenetic events, like hypermethylation of gene promoters which occur frequently in CRCs, may also play a role in the expression of varying transcripts [
45]. Other post-transcriptional regulatory events may also contribute; trafficking of transcripts by microRNAs offers another plausible explanation. miR21, a microRNA present in many tumors, also has been shown to down-regulate TLR4 [
46]. We speculate that the type of TLR4 mRNA/protein product regulates biological events, as may non-coding TLR4 transcripts found in genome browsers (Figure
1C). Bench and animal experiments are required to interrogate the mechanism for the functional differences in TLR4 transcripts.
The authors acknowledge the limitations of this study. Most notably, the TMA histologic scoring was based on cores; accordingly, TLR4 positivity may have been underestimated given the heterogeneous nature of CRCs and sampling error inherent in cores. We did not incubate TMA controls with only secondary antibody (TLR4) without the primary antibody; our controls consisted of unmatched, uninvolved colonic tissue. Finally, RNA expression and protein staining conclusions were drawn from unmatched samples in some instances.
Acknowledgements
The authors would like to thank Marc Lippman for critical revision of the manuscript, Sakhi S. Philip and Mansoor M. Ahmed for scanning and photography services, and Cristina Verdejo-Gil for assistance with digital acquisition of images.
Grant support
This study was supported by a Bankhead-Coley Team Science Grant 2BT02 to MTA and DAS, NIH CA137869 and a Crohn’s and Colitis Foundation of America (CCFA) Senior Investigator Award grant to MTA, a CCFA Research Fellowship Award to RS, and National Science Foundation/DTRA (NR66853W) and NIH (MH094759) awards for JC.
Competing interests
The authors do not have any relevant financial interests related to the work described in this manuscript.
Authors’ contributions
DAS participated in the design of the study, acquired the data, interpreted the data, and drafted the manuscript. RS performed the immunofluorescent and immunohistochemical staining. PAB participated in the interpretation and scoring of immunofluorescence. MTG participated in the interpretation and scoring of immunofluorescence. MTP participated in the interpretation and scoring of immunohistochemical stains. MTA participated in the design of the study and interpretation of results. JC participated in the design of the study, performed the statistical analysis, and interpreted results. All authors participated in the preparation of the manuscript as well as reviewed and approved the final manuscript.