Sessile serrated lesions (SSL) are precursors to colon carcinoma, and their distinction from other polyps, in particular hyperplastic polyps (HP), presents significant diagnostic challenges. We evaluated expression patterns in colonic polyps of previously identified colon carcinoma–associated extracellular matrix (ECM) proteins to identify markers distinguishing SSLs from other polyps.
Gene-expression analyses of ECM proteins were performed using publicly available data on preneoplastic colonic polyps. In parallel, we evaluated by IHC the expression of agrin (AGRN) in over 400 colonic polyps, including HP, SSL with and without dysplasia, traditional serrated adenomas (TSA), and tubular adenomas (TA), and compared the consistency of standard histologic diagnosis of SSLs by experienced gastrointestinal pathologists with that of AGRN IHC.
Differential gene expression analysis and IHC identified AGRN, serine peptidase inhibitor (SERPINE2), and TIMP metallopeptidase inhibitor 1 (TIMP1) elevated in SSLs and HPs but decreased in TAs and absent in normal colon. AGRN-positive basal laminae were noted in all TA, TSA, HP, and SSL in distinguishable patterns, whereas other polyps and normal mucosa were negative. SSL with or without dysplasia consistently showed IHC staining for AGRN in the muscularis mucosae, which was absent in HP, TSA, TA, and other polyps. In contrast, histologic evaluation showed only weak interobserver agreement (kappa value = 0.493) in distinguishing SSLs.
Muscularis mucosae–based AGRN immunostaining is a novel biomarker to distinguish SSL from HP, TSA, and TA, with a specificity of 97.1% and sensitivity of 98.9% and can assist in diagnosis of morphologically challenging colonic polyps.
Sessile serrated lesions (SSL) are precursors of colon carcinoma. Their distinction from other polyps with better prognoses, most notably hyperplastic polyps (HP), often presents a significant diagnostic challenge, and histologic evaluation shows only weak interobserver agreement among experts. We identified the extracellular matrix protein agrin in the muscularis mucosae (MM) of SSL biopsies but not in other polyps or normal colonic tissue. IHC staining of the MM for agrin presents a novel biomarker with high specificity and sensitivity and markedly improves the discrimination among polyp subtypes. Immunostaining for agrin in the MM enables pathologists towards more accurate diagnoses of patients with SSLs, can assist with morphologically challenging cases, and may allow evidence-based surveillance of serrated polyps.
Colonic polyps differ in their risk for progression to cancer; consequently, recommendations for removal and further management are dependent on polyp classification (1). Although tubular adenomas (TA) were traditionally considered to be the sole precursor lesions of colorectal carcinoma, the recognition of the serrated polyp route to carcinoma has been an important milestone (2, 3).
The WHO recognizes three categories of serrated polyps: traditional serrated adenomas (TSA), hyperplastic polyps (HP), and sessile serrated lesions (SSL; refs. 4–8). HPs, predominantly arising in rectum and sigmoid colon, lack malignant potential and are by far the most common colonic polyps, accounting for approximately 80% of all serrated polyps (6). In contrast, SSLs, which do have malignant potential, are estimated to represent up to 20% of serrated polyps with a predilection for the right colon (9, 10). TSAs also have malignant potential but represent only 1% of all serrated lesions (11).
Although SSLs show abnormal basal crypt architecture, HPs lack these specific features (9). However, uncertainty exists regarding the minimum criteria for diagnosis of SSL and their distinction from HP. The number of abnormal crypts required has varied among guidelines from one to three SSL-like crypts (5–7). Furthermore, there is significant interobserver variability as to what constitutes an abnormal crypt and poor specimen orientation and other biopsy artifacts accentuate the challenge (12–14). Indeed, substantial interobserver variation among gastrointestinal (GI) pathologists remains and agreement among experts is moderate at best (13, 15–20). These uncertainties often result in the inability to definitively distinguish SSL from HP. The diagnosis of SSL prompts more aggressive surveillance, generally 1 to 3 years, whereas a diagnosis of HP and small TAs (<10 mm) indicates less aggressive surveillance, every 5 to 10 years (5, 18). Collectively, the evidence suggests a need to identify biomarkers for SSLs to guide appropriate management. Suggested biomarkers evaluated, MUC6, ANXA10, and HES1 (21–25) have not been rigorously tested nor are they used in routine practice.
To identify novel biomarkers for SSLs, we evaluated the expression of extracellular matrix (ECM) proteins upregulated in colorectal cancer (26). We investigated human colonic polyps and found differential AGRN positivity in the basal lamina (BL) of TA, TSA, HP, and SSL whereas hamartomatous and mucosal prolapse polyps were negative. Importantly, SSL showed AGRN reactivity in the muscularis mucosae (MM) whereas other colonic polyps and controls were negative. Our study suggests that MM-based AGRN immunostaining represents a novel biomarker to differentiate SSL from HP, TSA, and TA with a high specificity of 97.1% and sensitivity of 98.8%.
Materials and Methods
Patients and samples
Patients were not involved in the study. Formalin-fixed, paraffin-embedded (FFPE) biopsies of normal colon and colonic lesions/polyps were identified from patients (between 2 and 89 years of age) undergoing colonoscopy between 2006 and 2019. Hematoxylin–eosin (H&E) slides were examined by two pathologists (VD, OHY) before inclusion in this study. We analyzed 408 colonic polyps, including HPs (n = 71), SSLs with dysplasia (n = 23) and without dysplasia (n = 166), TSAs (n = 25), TA (n = 64), dysplasia related to inflammatory bowel disease (IBD, n = 29), sporadic adenomas arising in IBD (TA in IBD; n = 9), mucosal prolapse polyps (n = 3). Colonic hamartomatous polyps (n = 18) including juvenile polyps (n = 10) and Peutz–Jeghers polyps (n = 8) served as controls (Supplementary Fig. S9). The splenic flexure was used to differentiate left from right-sided polyps; rectal polyps were categorized separately (see Supplementary Fig. S2; Supplementary Table S2). The SSL samples (n = 189) included 132 (69.9%) polyps from the right colon, 49 (25.9%) from the left colon, and 8 (4.2%) rectal polyps. Among the 71 HP cases, 9 (12.7%) polyps were from the right colon, 32 (45.1%) from the left colon, and 30 (42.2%) from the rectum. HPs were further characterized as microvesicular HPs (MVHP; n = 38) and goblet cell HPs (GCHP; n = 33) by two pathologists (AN, VD; Supplementary Table S2; ref. 27).
To define the serrated polyposis syndrome (SPS), we used criteria proposed by Syngal and colleagues (28). Briefly, diagnosis was based on the following criteria: (i) at least five serrated polyps proximal to the sigmoid colon with ≥2 of these being >10 mm; (ii) any number of serrated polyps proximal to the sigmoid colon in an individual who has a first-degree relative with serrated polyposis; or (iii) >20 serrated polyps of any size, distributed throughout the large intestine. Additional SPS patient samples were provided by the Department of Pathology, University of Utah (Salt Lake City, UT; ref. 29).
The study was approved by the Partners Institutional Review Board protocol (IRB #2017P000061) and the MIT IRB (Protocol #1408006568).
Majority-based pathology classification approach
Approach for majority agreement on 50 diagnostically challenging serrated polyps
We used a majority approach for the diagnosis of SSL. From our cohort of 408 colonic polyps and within the 100 cases validated by the 4 nonpathologists (see Supplementary Materials and Methods), 2 GI pathologists identified 50 diagnostically challenging polyps with a low level of interobserver agreement. Digital whole-slide H&E-stained slides were evaluated by 9 GI pathologists (DP, IB, ARM, LZ, QZ, RC, GL, OHY, VD) using the prior WHO guidelines (6). The pathologists were blinded to endoscopic size, endoscopic appearance, and location of the individual samples and classified the polyps into one of the following four categories: HP, SSL with or without dysplasia, TSA, or unclassified.
Approach for near-universal agreement on 50 SSLs
Given the existing lack of consensus among pathologists, we strove to identify 50 SSLs with near-universal agreement from within the cohort of 408 polyps. Among 65 additional polyps evaluated by 9 GI pathologists (DP, IB, ARM, LZ, QZ, RC, GL, OHY, VD), all blinded to endoscopic size, endoscopic appearance, and location of the samples, we identified 50 SSLs with near-universal agreement.
Interobserver variability among community pathologists
To assess the interobserver variability on selected morphologically challenging polyps (see above), we further assessed 50 community-based pathologists. On the basis of representative images of H&E-stained slides from eight polyps, the pathologists recorded their diagnosis using an audience response system. Each pathologist was required to characterize the polyp into one of five categories: HP, SSL with or without dysplasia, TSA, TA, or Rather not say (Supplementary Fig. S10).
Antibodies, IHC, immunofluorescence, in situ hybridization, and RNA-seq data analysis
Antibodies, standard techniques, RNA-seq data analysis, and detailed information are provided in the Supplementary Materials and Methods and in Rickelt and Hynes (2018, ref. 30).
Identification of ECM proteins in colonic polyps
To identify novel biomarkers to distinguish preinvasive colorectal neoplasms, we investigated the presence of 67 ECM proteins recently identified as increased in tumor samples from patients with colorectal cancer (26). We used two parallel approaches to elucidate their value as potential biomarkers for detection of precancerous colonic lesions. First, we analyzed their expression levels in a publicly available RNA-seq dataset (29) on colonic polyp and control samples (Fig. 1). Second, we performed extensive IHC screening on FFPE samples of diverse polyp types.
Gene expression analyses of 65 ECM molecules identified by proteomics (26) were performed using RNA-seq data (29) from 41 colonic polyps including SSL (n = 21), HP (n = 10), TA (n = 10), and colonic control samples (n = 20). Most individual SSL samples clustered separately from the other polyp and control samples (Supplementary Fig. S1), clearly indicating that this class of polyps is distinct from others. To determine the most abundant ECM candidates upregulated in common colonic polyps, we performed differential-expression analysis comparing individual polyps to normal control tissues using cut-off selection criteria of log2-fold change (FC) >1.0 and adjusted P value (adp) <0.05 (Fig. 2; Supplementary Table S1). This approach identified 5 ECM genes; agrin (AGRN), insulin-like growth factor binding protein, acid-labile subunit (IGFALS), S100 calcium-binding protein A11 (S100A11), serine peptidase inhibitor (SERPINE2), and TIMP metallopeptidase inhibitor 1 (TIMP1), all of which were upregulated in both SSLs and HPs (Figs. 1 and 2; Supplementary Table S1) whereas their expression was lower in TAs.
In parallel, we screened and validated (on FFPE material) commercially available antibodies specific for colorectal cancer–associated ECM proteins (30). The combined IHC and transcriptomics approaches highlighted three ECM-associated proteins (AGRN, SERPINE2, and TIMP1) with reliable antibodies, which were investigated further (Fig. 1).
AGRN and TIMP1 immunostaining in colorectal polyps
To investigate the potential diagnostic utility of AGRN, SERPINE2, and TIMP1, we performed IHC on a cohort of TAs, HPs, SSLs, and TSAs (Supplementary Fig. S2) with antibodies to AGRN, SERPINE2, and TIMP1 (Supplementary Fig. S3). We detected strong staining for AGRN in the BL of blood vessels and crypts of the polyps, and TIMP1 in the cytoplasm of all polyp types investigated (Supplementary Figs. S2A and S3). In the normal colon, AGRN was only present in the BL of blood vessels and negative in the normal colonic crypts. TIMP1 was also negative in the normal colon; however, neuroendocrine cells were positive for TIMP1, as described previously (31). In contrast, variable SERPINE2 cytoplasmic and ECM staining was noted in normal and lesional crypts (Supplementary Fig. S3). Given the lack of a consistent and clear differential reactivity with TIMP1 and SERPINE2 in normal versus polyp tissue, we elected to evaluate AGRN further.
Differential localization of AGRN as a biomarker of SSLs
In total, we evaluated 408 colonic polyps representing the following pathologic diagnoses: HP (n = 71), SSL with (n = 23) and without (n = 166) dysplasia, TSA (n = 25), TA (n = 64), colonic hamartomatous (n = 18), and mucosal prolapse polyps (n = 3; Supplementary Figs. S2B–S2E, S9; Supplementary Table S2). We found positive AGRN staining of the BL in all colonic lesions, whereas hamartomatous, mucosal prolapse polyps, and normal colon mucosa were negative. However, differential patterns of BL localization were noted (Fig. 3); a top-heavy pattern of AGRN positivity was consistently observed in TAs, whereas in TSAs the BL reactivity was uniformly distributed along the length of the colonic crypt. In contrast, in SSL and HP samples, AGRN BL positivity was more prominent in the basal crypts, with weaker staining consistently seen in HPs.
To examine further the biological relevance of these findings, we assessed the BL AGRN positivity in a range of mouse models with conventional adenoma (Supplementary Fig. S4A–S4D) and those with serrated-like (Supplementary Fig. S4E) morphology. In concordance with the human samples, AGRN immunostaining with two anti-AGRN antibodies confirmed the consistent presence of this protein in the BL of murine colonic polyps whereas the BL of adjacent normal colonic mucosa was negative (Supplementary Fig. S4). More importantly, the BL distribution patterns of AGRN in the mouse models paralleled the human data (Supplementary Fig. S4). Collectively, the genetically defined mouse models of colonic polyps recapitulate the distribution patterns of AGRN seen in human TAs and TSAs.
AGRN is expressed in epithelial cells and deposited into the MM in SSL
Strikingly, we also noted the selective presence of MM-based AGRN in human SSLs with and without dysplasia but not in other polyps (Fig. 3; Supplementary Fig. S3). IHC and confocal immunofluorescence microscopy on consecutive sections of SSLs demonstrated AGRN colocalization with SMA and DES (both muscle-specific markers) in the MM (Fig. 4A–D); AGRN was localized to the upper half of the MM (Fig. 4D). AGRN mRNA expression was predominantly localized to crypt-base cells with stronger expression in SSL as compared with HP, TA, and TSA (Supplementary Fig. S5A–S5H), whereas the adjacent MM of SSLs was negative (Supplementary Fig. S5E, S5F, arrowheads). Collectively, these results suggest that AGRN, expressed by colonic epithelial cells, is deposited in the MM in SSLs.
AGRN positivity of the MM as a biomarker for SSLs
To investigate further the utility of MM-based AGRN reactivity in diagnosing SSLs, we investigated 12 anti-AGRN antibodies on human FFPE samples (30), however, we identified only one additional reliable anti-AGRN antibody. Both anti-AGRN antibodies showed similar staining patterns (Supplementary Figs. S5I–S5L); and selectively stained the MM of SSL.
We next identified fifty SSLs with near-universal diagnostic agreement among 9 expert GI pathologists. MM-based AGRN reactivity was noted in all samples (data not shown). In addition, we tested 18 SSL polyps from 10 patients with SPS (a single polyp in 2 patients and 2 polyps in eight patients) and all showed a strong MM-based positivity for AGRN (Supplementary Fig. S6).
To control for bias, because experienced pathologists can recognize a SSL without an AGRN stain, 4 nonpathologists were asked to blindly validate the MM-positivity for AGRN on 100 samples (Supplementary Fig. S7A; Supplementary Table S2), including all cases used in the validation cohort. Notably, there was complete agreement with regard to MM-based reactivity among the 4 observers (Supplementary Fig. S7B), validating the robustness of this criterion.
Finally, we selected 50 diagnostically challenging HPs, SSLs, and TSAs (Supplementary Table S2) and compared the results of AGRN staining with the evaluation performed by 9 GI pathologists on H&E sections (see Materials and Methods). In total, samples #1–46 were scored (Fig. 5A); four samples had to be excluded (samples #47–50; Supplementary Fig. S7C). Complete concordance among the pathologists was achieved in only 13 cases (26%), κ-value = 0.493, indicating only weak overall agreement (Fig. 5A, green). The principal source of disagreement lay in the distinction of HP (Fig. 5A, gray) from SSL (Fig. 5A, orange). On the basis of majority opinion (at least 5 of 9 pathologists), the polyps were classified as follows: 25 SSLs with or without dysplasia, 14 HPs, and 4 TSAs (Fig. 5A, purple). Three samples did not yield a majority and the opinions were widely divergent. The MM staining for AGRN was identified (Fig. 5A, red) in 24 of the 25 cases diagnosed as SSL. In contrast, 19 cases that lacked a majority opinion diagnosis of SSL were negative for AGRN in the MM (Fig. 5A, blue). In the three cases lacking a majority opinion (arrows in Fig. 5A), the results differed from those noted above and are illustrated in Fig. 5B–G. Although SSL represented the majority (5 of 9) opinion in case #25, no AGRN reactivity was noted (Fig. 5B and C). Two cases #26 and #27 both showed MM-based AGRN reactivity, however, for #26 no majority opinion was reached and #27 majority (5/9) opinion was HP, although both cases showed contiguous crypts with basal crypt dilatation and basal serrations (Fig. 5D–G). Finally, we also tested 55 community pathologists by showing them representative images of H&E slides from eight cases and observed a virtually similar level of interobserver agreement (Supplementary Fig. S10).
Collectively, for all samples investigated in this study, the majority reads of the AGRN staining used for final calculations were as follows: MM-based AGRN positive staining in 186 of 188 SSLs with (23/23) and without (163/165) dysplasia, 5/68 HPs, 0/64 TA, 0/25 TSA, and 0/21 hamartomatous and mucosal prolapse polyps (Supplementary Figs. S7D, S9; Supplementary Table S2). AGRN reactivity also assisted in differentiating TA from SSL with dysplasia when the continuity between the SSL and dysplastic component was lost due to fragmentation of the specimen (Supplementary Fig. S8). In addition, loss of BL-based AGRN reactivity was noted in the dysplastic portion of SSL (Supplementary Figs. S8D, S8I, S8J). In summary, these observations suggest that the MM-based AGRN stain could help to assist in identifying dysplasia arising within SSL in cases in which the SSL component is poorly represented. Given that SSL with dysplasia are more likely to progress to cancer it is important to accurately diagnose a dysplastic component of serrated polyps.
Finally, samples of both subtypes of HPs (MVHP, n = 37 and GCHP, n = 32) and nine large HPs (>4.5 mm) were negative for MM-based AGRN (see also Supplementary Table S2).
Power analyses ensured sufficient precision of our estimation. On the basis of sensitivity and specificity, a 95% confidence interval as narrow as 7% was calculated. Collectively, this provides a reliable estimation of MM-based AGRN as a diagnostic marker for SSLs with sensitivity of 98.9% and specificity of 97.1% (Supplementary Fig. S7D).
RNA-seq–profiled colonic polyps corroborate the biomarker capability of AGRN immunoreactivity
We finally sought to validate our AGRN IHC findings on the cohort of polyps from the RNA-seq dataset (29) (Figs. 1, 2; Supplementary Fig. S1). Many SSL (n = 14) and a few HP samples (n = 4) showed higher AGRN expression, most other polyps and controls showed lower AGRN expression (Fig. 6A). We evaluated 10 SSL and 5 HP for MM-based AGRN reactivity (Fig. 6). Blinded evaluation identified 9 of 10 SSL to be AGRN-positive (data not shown). The majority of these SSLs also showed elevated AGRN expression and formed a distinct cluster (Supplementary Fig. S1), three samples (#14, 18, 20) did not conform with our hypothesis (Fig. 6; Supplementary Fig. S1, see figure legend for details).
Finally, in agreement with our prior observation, all 12 samples from patients with SPS in this cohort (29) clustered together and showed high AGRN expression (Fig. 6A; Supplementary Fig. S1). In addition, of these samples all six polyps that we stained were positive for MM-based AGRN reactivity (Fig. 6A).
SSLs have a risk of malignant progression and require accelerated screening for colorectal cancer. However, in routine clinical practice, histopathologic characterization of SSLs has met significant challenges; overlapping histological features with HPs, poor biopsy orientation, resulting in moderate-to-poor interobserver agreement among pathologists (15–17, 19, 20). A biomarker specific for SSLs would clearly help pathologists and allow evidence-based surveillance of serrated polyps, and the results presented here suggest that MM-based AGRN reactivity could serve as such an objective marker for SSL.
There have been several prior attempts to identify SSL-specific biomarkers. Caruso and colleagues (32) performed a gene-array study and compared SSLs to TAs and controls. SSLs showed upregulation of CTSE and TFF1. ANXA10 was identified in SSLs as compared with HPs and validated by IHC as a potential diagnostic marker with a sensitivity of 73% and specificity of 95% in the diagnosis of SSL (24). MUC6 was found to be expressed in SSL but not in HP (21), however, subsequent studies revealed a relatively low sensitivity (22) and lack of specificity for SSL (23). In addition, Delker and colleagues (33) examined gene expression to discriminate between SSLs and HPs and found unique staining patterns for the cell junction protein VSIG1 and MUC17 in SSLs. Finally, the loss of the transcription factor HES1 was observed in the majority of SSLs compared with normal expression in HPs (25); however, TA and also TSA showed variable staining for HES1, diminishing its value as a marker of SSL. Notably, none of these potential biomarkers is currently applied in routine clinical practice.
ECM proteins play an important role in the initiation and progression of colon carcinomas, often related to poor prognosis (34, 35). The diagnostic relevance of ECM proteins in colorectal cancer and their applicability to distinguish individual colonic polyps so far remains untested. Our results demonstrate the differential expression of several ECM genes among colonic polyps and show that AGRN, SERPINE2, and TIMP1 are overexpressed in SSL and HP samples (Figs. 1, 2; Supplementary Figs. S1 and S3). Colonic polyps show distinct patterns of AGRN deposition in BL. Importantly, MM-based AGRN reactivity was restricted to SSLs (Figs. 3, 4; Supplementary Figs. S3 and S5). We confirmed this finding by evaluating a cohort of 50 diagnostically challenging colonic polyps with only weak agreement among GI pathologists. Virtually all cases with a majority read of SSL showed MM-based AGRN reactivity, whereas non-SSL polyps lacked AGRN reactivity in the MM. Another strong validation of the diagnostic value of MM-based AGRN reactivity is the near-universal presence of MM-based AGRN in patients with SPS.
The ECM molecule AGRN, a large multidomain heparan sulfate proteoglycan, is abundantly expressed in developing brain and in virtually all BL of developing organs. Functionally, AGRN aids in the formation of neuromuscular junctions and acetylcholine receptor clustering in the central nervous system (36). The nonneuronal functions of AGRN are only poorly understood. Although little is known about AGRN in carcinogenesis, in recent years, a tumor-promoting role has been reported for several cancer types, including hepatocellular and cholangiocellular carcinoma (37, 38), prostate cancer (39), and oral squamous cell carcinoma (40). Functional studies have shown that AGRN is involved in proliferation, migration, and invasion of liver cancer cells by regulating focal adhesion integrity and to relay mechanosensitive signals into cells to regulate YAP activity to promote tumorigenesis (41, 42). In contrast, the role of AGRN in normal colorectal mucosa, colonic polyps and colorectal cancer has not been investigated extensively. By tissue secretome profiling, AGRN was one of 76 potential colorectal cancer protein biomarkers that may facilitate blood- or stool-based assay development to support clinical management of colorectal cancer (43). Another analysis of the extracellular proteome of colorectal cancer cells identified elevated levels of the C-terminal fragment of AGRN (44), however, its suitability as a biomarker has not yet been assessed.
Although expressed by colonic epithelial cells, AGRN is specifically localized in the MM of SSL. AGRN interacts with several other ECM proteins and a variety of growth factors. In addition, earlier studies have shown AGRN to bind cell-surface adhesion receptors, such as NCAM, integrins, α-dystroglycan and muscle-specific kinase (36). The determination of whether any of these proteins colocalize at the MM of SSLs and interact with AGRN will therefore be of interest.
In conclusion, our study supports the use of MM-based reactivity for AGRN as a novel biomarker to differentiate SSL from HP. Our study uses a majority opinion as the “comparison standard,” and MM-based AGRN expression compared very favorably with this standard, outperforming H&E-based diagnosis by expert pathologists in a set of challenging SSL; whereas in a set of straightforward SSL the results of AGRN staining were concordant with virtually all pathologists. Finally, MM-based AGRN was also universally detected in SPS syndrome. Thus, MM-based immunostaining for AGRN may enable more accurate diagnosis of patients with SSLs and assist with morphologically challenging cases.
Disclosure of Potential Conflicts of Interest
O.H. Yilmaz is an employee/paid consultant for Merck and reports receiving commercial research grants from Exelixis. V. Deshpande is an employee/paid consultant for Incyte and Viela, and reports receiving commercial research grants from Agios and Biotchene. No potential conflicts of interest were disclosed by the other authors.
Conception and design: S. Rickelt, V. Deshpande, R.O. Hynes
Development of methodology: S. Rickelt, V. Deshpande, R.O. Hynes
Acquisition of data (provided animals or samples, acquired and managed patients, provided facilities, etc.): S. Rickelt, C. Condon, M. Mana, C. Pfirschke, J. Roper, L.G.J. Leijssen, K. Boylan, O.H. Yilmaz, V. Deshpande
Analysis and interpretation of data (e.g., sample scoring, statistical analysis, biostatistics, computational analysis): S. Rickelt, C. Whittaker, J. Roper, D.T. Patil, I. Brown, A.R. Mattia, L. Zukerberg, Q. Zhao, R. Chetty, G.Y. Lauwers, A. Neyaz, O.H. Yilmaz, V. Deshpande, R.O. Hynes
Writing, review, and/or revision of the manuscript: S. Rickelt, V. Deshpande, R.O. Hynes
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Rickelt, V. Deshpande
Study supervision: O.H. Yilmaz, V. Deshpande, R.O. Hynes
The authors thank Chenxi Tian for assistance in using the confocal laser-scanning microscope, and all members of the Hynes laboratory for advice and discussions. They thank the Swanson Biotechnology Center at the Koch Institute/MIT, especially Mike S. Brown and Kathleen S. Cormier from the Hope Babette Tang (1983) Histology Facility and Jeff Wykoff in the Microscopy Facility for exceptional technical support, Duanduan Ma from the Barbara K. Ostrom Bioinformatics & Computing Facility for assistance with bioinformatic analyses and power calculations, and Sven Holder for sample sectioning. They also thank Lucia Suarez-Lopez (Koch Institute); Roderick T. Bronson (Harvard Medical School); and Mari Mino-Kenudson, Jeck Williams, and Martin S. Taylor (Massachusetts General Hospital) for discussion, data collection, and advice. The authors also thank Eric R. Fearon (Department of Internal Medicine, University of Michigan, Ann Arbor, MI) for kindly providing slides of colon sections from control and CDX2P-CreERT2 CDX2fl/fl;BrafLSL-V600E/+ mice and Mary Bronner (Department of Pathology, University of Utah, Salt Lake City, UT) for critically reading the manuscript and providing samples previously described by Kanth and colleagues (2016, ref. 29). This work was supported by NIH grants U54-CA163109 (Tumor Microenvironment Network to R.O. Hynes) and R01 CA211184, R01 CA034992 (to O.H. Yilmaz), the MIT Ludwig Center for Molecular Oncology and the Howard Hughes Medical Institute, of which R.O. Hynes is an investigator. O.H. Yilmaz was supported by the Pew Foundation, Sidney Kimmel Foundation, and MIT Center for Stem Cell Research. Facility support was provided by the Koch Institute Swanson Biotechnology Center (Cancer Center Support Grant NIH-P30CA014051). S. Rickelt was supported by postdoctoral fellowships from the Deutsche Forschungsgemeinschaft (DFG) RI2408/1-1 and the MIT Ludwig Center for Molecular Oncology. C. Pfirschke was supported by the MGH ECOR Tosteson and Fund for Medical Discovery Fellowship, and J. Roper by NIH/NCI (1K08CA198002-01).
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.