Dynamic Tumor Growth Patterns in a Novel Murine Model of Colorectal Cancer

Colorectal cancer is the second leading cause of cancer-related deaths in the United States (1). Colorectal cancer most often arises in adenomatous polyps that form throughout the colon. The current model of colorectal tumorigenesis posits a stepwise accumulation of mutations in tumor suppressor genes, oncogenes, and genetic instability genes, giving rise to adenomatous polyps that progress to adenocarcinoma over a period of years (2,3). However, only a small fraction of polyps progress to cancer. Observational studies have shown that over time colorectal polyps can grow, remain static in size, regress (or decrease in size), or resolve altogether (4–6). Because of their malignant potential, polyps that are identified on screening colonoscopy are removed. Increased surveillance is recommended after polyps are found (7), further straining a medical system that already struggles to meet the demand for screening and diagnostic colonoscopies, both in terms of cost (8) and available providers (9). In addition, colonoscopy and polypectomy are not without risk; colonic perforation occurs at a low but predictable rate during colonoscopy (10), and significant post-colonoscopy bleeding is more likely to occur after polypectomy (11).

Current efforts are focused on risk-stratifying patients with colorectal polyps. A potential method for quantifying risk for colorectal cancer would be to identify polyps that are more likely to grow and progress to cancer and those that are likely to be benign. Attempts have been made to predict polyp fate and carcinogenic potential using the size and number of polyps, as well as histologic features of removed polyps (7). It is currently unknown whether analysis of gene expression profiles from polyps can add further specificity to risk predictions. Researchers have attempted to identify differential gene expression profiles among normal colonic mucosa, adenomatous polyps, and colorectal cancer in humans (12), as well as correlate gene expression profiles with colonoscopic findings (13). However, the information gathered from gene expression profiles in human tumors is difficult to interpret functionally because of molecular heterogeneity in neoplastic tissue coupled with the genetic variability inherent in human populations.

Animal models can minimize genetic variation while allowing experimental manipulation of tumorigenesis. Our group has developed a long-lived mouse model with a germline-truncating mutation in the mouse homolog of APC that forms adenomatous colonic polyps, which can be monitored by serial colonoscopies (14,15). We previously showed that 24% to 87% of intestinal tumors in this model progress to adenocarcinoma and 3% develop nodal metastases. These cancers develop without a high level of microsatellite instability or chromosomal gains or losses (15). We have now extended our previous work by meticulously characterizing the natural history of colonic tumors in this model to determine whether colonic tumors have growth patterns similar to those of colonic polyps in humans. We observed five distinct growth patterns as well as benign and malignant pathology. We have also begun to characterize molecular features of tumors with different degrees of invasiveness in this model.

All animals were studied in the Wisconsin Institute for Medical Research (Madison, WI) vivarium under protocols approved by the Institutional Animal Care and Use Committee at the University of Wisconsin (Madison, WI) following American Association for the Assessment and Accreditation of Laboratory Animal Care guidelines. All mice were housed in climate-controlled rooms with 12-hour light/dark cycles and given food and water ad libitum except as discussed below. Mice were generated by crossing SWR females (SWR/J; The Jackson Laboratory; Stock Number 00689) with C57BL/6 Apc^Min/+ males (C57BL/6J-Apc^Min/J; The Jackson Laboratory; Stock Number 002020). F1 Apc^Min/+ hybrids (F1 Min) were identified by genotyping at Apc using a previously described PCR assay (16).

At or within 1 week of weaning (24–35 days of age), F1 Min mice were treated with dextran sodium sulfate (DSS; average molecular weight 500,000; Fisher Scientific) to increase colonic tumor multiplicity. Mice were given a 4% solution of DSS dissolved in water ad libitum for 4 days, followed by 17 days of regular water, and then a second cycle of 4% DSS for 4 days (17). After allowing 30 days for tumor initiation, the colonoscopic surveillance and biopsy protocols were started (Supplementary Fig. S1). A total of 273 mice were enrolled in the surveillance protocol, but only 101 remained on this protocol until they became moribund. Note that 172 mice were randomly removed from the surveillance protocol for various drug studies. Only images of colonic tumors that were collected before treatment were used to assess growth behavior. A total of 12 mice were enrolled in the biopsy protocol.

The distal 4 cm of the colon were visualized in 273 DSS-treated F1 Min mice by colonoscopy using the Karl Storz Coloview system as previously described (18). Briefly, mice were anesthetized using inhaled isoflurane. The distal colon was prepared using enemas of PBS until clear of fecal pellets and debris. Room air delivered by a compressed air pump through the operating sheath of the colonoscope was used to insufflate the colon. The colonoscope was inserted distally and advanced as far proximally as possible, typically until a sharp bend in the colon was encountered, and then slowly withdrawn. The colonic mucosa was inspected during withdrawal, and digital video of the mucosa was recorded. Still images were collected of any tumors as they were encountered. To standardize images for comparison of discrete tumors over time, the colon was maximally insufflated and the colonoscope was positioned so that the entire tumor could be seen with the base of the tumor just at the edge of the screen (demonstrated in Fig. 1). Mice were monitored postprocedurally and allowed to recover from anesthesia. Mice underwent surveillance colonoscopy in this fashion every 3 weeks to monitor the growth pattern of tumors (Supplementary Fig. S1A).

Still images of tumors from 273 animals were annotated for serial comparisons. Each tumor image was compared with the corresponding image from the previous colonoscopy session and rated by a single reviewer as growing, remaining static in size, regressing, resolved, prolapsing from the anus, or unable to assess. Tumor age was defined as the number of days since a tumor was first identified on colonoscopy.

When mice became moribund, they were sacrificed by CO₂ asphyxiation. The entire intestinal tract was removed, opened longitudinally and washed with PBS, fixed for 16 to 24 hours in 10% buffered formalin (Fisher Scientific), and stored in 70% ethanol. Using a stereomicroscope, the total number of tumors in the small intestine and colon was counted. A randomly selected representative sample of 54 small intestinal tumors from 6 mice and 30 colonic tumors from 16 mice that had been followed by surveillance colonoscopy was removed for histologic evaluation. These tumors were paraffin embedded, cut into 5-μm sections, and hematoxylin and eosin (H&E) stained before assessment of tumor invasiveness by a veterinary pathologist.

The operating sheath housing the colonoscope possesses a working channel that allows passage of a 3-French biopsy forceps, which is 5 mm in size when fully opened. A cohort of 12 mice with tumors amenable to biopsy was identified at their first colonoscopy. These tumors underwent biopsy when mice were approximately 90 days of age. Following this initial procedure, tumors were visually assessed by colonoscopy every 14 days, with additional biopsies collected every 28 days until tumors resolved completely or mice became moribund (Supplementary Fig. S1B). During each procedure, two biopsies were collected from each tumor. One was stabilized in RLT buffer with β-mercaptoethanol or RNALater (Qiagen) for RNA purification, and stored at −80°C until processing. The other was fixed in 10% buffered formalin (Fisher Scientific) for 24 hours and then stored in 70% ethanol for paraffin embedding and histologic evaluation.

A set of 12 tumor biopsies from six tumors was chosen for immunohistochemistry (IHC) and gene expression analysis. These paired samples were chosen from the earliest and latest biopsies available for each tumor.

Immunohistochemical analysis of the tumor biopsies was also performed using anti-β-catenin antibody (D10A8; Cell Signaling Technology; ref. 19). Samples were fixed, embedded, and cut as described above. Antigen retrieval was performed using 10 mmol/L citrate buffer heated to boiling for 32 minutes and blocked for endogenous peroxidase using Peroxidazed 1 (Biocare Medical). Slides were blocked in 5% milk in phosphate-buffered saline with Tween-20 (PBST) for 45 minutes, then incubated overnight at 4°C with anti-β-catenin antibody diluted 1:200 in PBST. Slides were treated with horseradish peroxidase (HRP)–conjugated anti-rabbit secondary antibody (MACH 2 Rabbit HRP-Polymer; Biocare Medical), detected with DAB (3,3′-diaminobenzidine) chromogen (Biocare Medical), counterstained with CAT hematoxylin (Biocare Medical), dehydrated, and cover-slipped.

Samples were disrupted in RLT buffer with a Kontes Pellet Grinder (Kimple & Chase). RNA was extracted using the RNEasy Micro Kit (Qiagen) following the manufacturer's instructions. The Nanodrop DU-800 (Thermo-Fisher Scientific) was used to quantify yield and assess purity.

Affymetrix Mouse Gene 1.0 ST GeneChips (Affymetrix) were used for microarray analysis. Hybridization and scanning of the chips were performed by the Gene Expression Center at the University of Wisconsin Biotechnology Center (Madison, WI) according to standard Affymetrix protocol. Chips were normalized by robust multiarray averaging using the XPS system (20). Following preliminary analyses involving per gene ANOVA, differential expression between adenomas and intramucosal carcinomas was measured by a Student t test and fold change. We screened by fold change exceeding 2 and by an unadjusted P value of less than 0.05. The false discovery rate (FDR) in the reported gene list was determined by the q-value method (21), and could not be reduced below 51% given the sample size. Functional categories showing adenoma/intramucosal carcinoma differential expression were assessed by standardized average (over category) log fold changes, as computed in the R package allez (22). We screened for categories with outlying Z scores relative to category size.

RNA quantification was performed on a Nanodrop DU-800 (Thermo-Fisher Scientific). cDNA was generated by reverse transcription of 50 ng of total RNA, which had been stored at −80°C, according to the manufacturer's recommendation (ImProm II Reverse Transcription System; Promega). cDNA was stored at −20°C. Control cDNA was generated from Mouse Total Colon RNA (Clontech).

Commercially available primer and hydrolysis probe sequences for Tnfsf10 (Mm00446973_m1), Lcn2 (Mm01283606_m1), Muc2 (Mm01276696_m1), Tff3 (Mm00495590_m1), RetnlB (Mm00445845_m1), and Tbp (Mm00446973_m1) were identified using the Taqman Assay Search tool (Invitrogen). The hydrolysis probes contained fluorescein amidite (FAM) as a reporter molecule. Reactions were performed in triplicate for each tumor. Assays were performed in a volume of 20 μL, including Taqman Gene Expression Master Mix (Applied Biosystems), primer/hydrolysis probe sets, and 1.5 ng of RNA. A CFX96 Touch Real-Time PCR Detection System (Bio-Rad) was used with the cycling conditions: 2:00 at 50°C, 10:00 at 95°C followed by 45 cycles of 0:15 at 95°C and 1:00 at 60°C. Upon completion of cycles, a melt curve analysis was performed. All values were calculated using the CFX Manager software (Bio-Rad).

Resulting data were analyzed by the quantification cycle (C_t) method as means of relative quantification of gene expression. Values were normalized to an endogenous reference (TBP) and relative expression was compared with a normalized C_t value obtained from Mouse Total Colon control cDNA and subsequently expressed as 2^−ΔΔCt. A Wilcoxon rank sum test was performed for statistical evaluation.

A cohort of DSS-treated F1 Min mice (n = 101) that did not undergo any additional interventions and were followed with colonoscopy every 3 weeks lived an average of 282 days (range, 109–361 days; SD, 56 days). Almost all (99%) of the mice were found to have at least one intestinal tumor. They developed a mean of 12 tumors throughout the intestinal tract (median, 12; range, 0–32), with a mean of 10 in the small intestine (median, 9; range, 0–28) and 3 in the colon (median, 2; range, 0–9). The majority of colonic tumors arose in the distal portion in which they could be easily monitored by colonoscopy. Mean age for developing distal colonic tumors seen on colonoscopy was 121 days. Forty percent of mice had at least one tumor at the time of their first colonoscopy (age, 77–90 days). They continued to develop colonic tumors throughout their lives, with some tumors first identified as late as 306 days of age (Supplementary Table S1).

Colonic tumors in F1 Min mice had five distinct growth patterns—growth, stasis, regression (tumor decreased in size but did not resolve), resolution, and prolapse—which were evident by comparing standardized images that were collected at different times (Fig. 1A–D and Supplementary Fig. S1A). Tumor growth was most frequently seen in newly developed tumors (45% of 21-day-old tumors vs. 11% of 147-day-old tumors). Resolution was also most frequently seen in tumors that had recently formed and tended to be relatively small (7% of 21-day-old tumors compared with 1% of 147-day-old tumors). Conversely, tumors that had been present longer and tended to be larger occluding half of the lumen or more were more likely to remain static in size (36% of 21-day-old tumors vs. 52% of 147-day-old tumors). The percentage of tumors that prolapsed also increased over time (3% of 21-day-old tumors compared with 30% of 147-day-old tumors). These lesions tended to be located near the anus. Tumor regression was seen at a low rate throughout the lifespan of the mice, ranging from 1% to 6% at various time points (Fig. 1D). Thus, the growth patterns of tumors varied with the age of the tumor.

A subset of randomly selected small intestinal and colonic tumors from mice in the surveillance protocol cohort underwent histologic evaluation. Fifty-four small intestinal tumors were collected from 6 DSS-treated F1 Min mice and were analyzed by an experienced veterinary pathologist to assess tumor invasiveness. The 6 mice had a mean age at death of 280 days (range, 185–326 days). Thirty-one tumors were adenomas (57%), 10 were intramucosal carcinomas (19%), and 13 were adenocarcinomas (24%). Many adenomas showed high-grade dysplasia (Fig. 2A–C). Similarly, 30 colon tumors that had been followed by colonoscopy were collected from 16 mice and underwent histologic evaluation by an experienced veterinary pathologist. One was a hyperplastic lesion (3%), 22 were adenomas (73%), 6 were intramucosal carcinomas (20%), and 1 was adenocarcinoma (3%). Again, adenomas often displayed high-grade dysplasia (Fig. 2D–F). Thus, tumors in the small intestine and colon ranged from benign to malignant lesions.

Twelve mice, 6 males and 6 females, with distal colon tumors identified on initial colonoscopy at approximately 80 days of age underwent serial biopsies of tumors every 28 days until moribund as described above (Supplementary Fig. S1B). They lived an average of 272 days (range, 157–409 days). They had a mean of 19 tumors throughout their intestinal tracts (median, 13; range, 8–46), and a mean of two colon tumors monitored by colonoscopy (median, 2; range, 1–5). Because of the variation in their lifespan, different numbers of biopsies were collected from each animal. The mean number of biopsies was seven (range, 2–12). Moribund mice were sacrificed and tumors were collected. Nine tumors were sent for histologic evaluation. There were three adenomas and six intramucosal carcinomas. There were no hyperplastic lesions or invasive adenocarcinomas. Other tumors were not analyzed for a variety of reasons. One mouse died unexpectedly, preventing removal of the colon tumor. Another tumor underwent resolution after three biopsies had been collected. In addition, one tumor prolapsed from the anus, precluding histologic analysis.

Six pairs of biopsies and corresponding whole tumors (three adenomas and three intramucosal carcinomas) from 6 mice were stained for β-catenin (Fig. 3). Interestingly, adenomas (3 of 3) that failed to progress tended to have lower levels of β-catenin within the cytoplasm and nucleus (Fig. 3A and B), whereas those (2 of 3) that progressed to intramucosal carcinomas tended to have higher levels of β-catenin within the cytoplasm and nucleus (Fig. 3C and D).

These same biopsies were analyzed for changes in gene expression that occurred over time. There was no significant differential gene expression between early and late biopsies. However, a modest differential expression was detected between adenomas and intramucosal carcinomas, regardless of whether the biopsies were from an early or late time point (Supplementary Fig. S2). We report a list of 68 genes exhibiting differential expression between adenomas and intramucosal carcinomas (Supplementary Table S2). Of note, Resistin-like β (Retnlb or REML-β), Mucin 2 (Muc2), and intestinal trefoil factor 3 (Tff3) were expressed at a higher level in adenomas relative to intramucosal carcinoma, whereas TNF (ligand) superfamily, member 10 (Tnfsf10) and Lipocalin 2 (Lcn2), were overexpressed in intramucosal carcinoma relative to adenomas (Table 1). These differences were confirmed by quantitative PCR in all cases, except Tnfsf10, for which the difference was trending in the same direction as observed by microarray, but it was not quite statistically significant. In addition to these, a number of genes related to immune function and cell adhesion were differentially expressed, which may have an impact on the invasive and metastatic potential of these tumors (Supplementary Table S2). Although these 68 genes showed the strongest differential expression within this data set, the gene-level P values are not FDR corrected. Although we report genes with at least a 2-fold change in expression level and an unadjusted P value of ≤0.05, because of the FDR, we can predict that about half of these genes will not be validated on quantitative studies. In addition, functional categories in gene ontology showing adenoma/intramucosal carcinoma differential expression were examined (Supplementary Table S3). Differentially expressed gene categories included those related to condensin complex, minichromosome maintenance complex, small nucleolar RNA binding, positive regulation of helicase activity, centromeric heterochromatin, and condensed chromosome, centromeric region (Supplementary Table S4).

We have extended our previous work using a long-lived mouse model of colorectal cancer that carries a truncating mutation in Apc and develops multiple neoplastic lesions throughout the intestinal tract, including tumors progressing to adenocarcinoma. After treating with DSS to increase colonic tumor multiplicity, we meticulously documented the natural history of colonic tumors by serial colonoscopies in a large cohort of experimental animals (424 tumors in 273 mice). Colonic tumors in this model exhibit fates remarkably similar to adenomatous colonic polyps in humans: some grow, some remain static, some regress, and some resolve spontaneously. The fate of newly developed tumors is variable, with alterations in growth patterns over time, but as tumors age their size becomes more stable. Tumors also showed a variety of pathologic fates, ranging from hyperplasic lesions to adenocarcinoma, although the rates of invasive cancer were much lower in the colon than in the small intestine (3% vs. 24%). Interestingly, adenomas that progressed often had a higher level of β-catenin. Moreover, using a small cohort of tumors, we identified differences in gene expression between adenomas and intramucosal carcinomas, including genes linked to colorectal cancer. We did not find differences in gene expression over time when comparing gene expression profiles between early and late biopsies taken from the same tumor.

Numerous studies have documented the variable growth patterns of colorectal polyps in human populations (4–6). Although a certain proportion of colorectal polyps will grow, remain stable in size, regress, or resolve altogether, the underlying mechanisms behind these divergent fates are not understood. Because it can take 10 to 17 years for polyps to progress to cancer in humans (23) and because polyps are removed when found on colonoscopy, longitudinal studies to understand the interactions between the host and the tumor are prohibitively difficult. Our murine model mirrors the growth patterns seen in human colorectal polyps, making it a valuable tool for discovering methods of risk-stratifying colonic tumors, as colonic tumors in these animals can be followed with serial biopsies to assess changes within tumors over time.

Newly developed tumors were more likely to grow or resolve, whereas more established tumors, which remain fairly static in size, seemed to have reached a level of accommodation with the organism. This model could be used to examine factors intrinsic to the host organism that impact polyp growth, such as host immune responses to tumors and the inflammatory response to premalignant lesions. A number of genes related to immune function and increased T-cell activity were relatively overexpressed in intramucosal carcinomas relative to adenomas (Table 1 and Supplementary Table S2), indicating that this model may be useful for investigating the interplay between tumor and host immune system. Being able to predict which tumors will be contained by the host and remain static or regress spontaneously and which will break free of host constraints to continue to grow and progress to cancer would be invaluable both for understanding the natural history of colorectal cancer and for risk-stratifying patients.

Although we have presented an extensive characterization of the growth patterns and histology of colonic tumors in DSS-treated F1 Min mice, we have only preliminarily begun to describe molecular and genetic characteristics of these tumors. In this small sample, we did not find differentially expressed genes between early and late biopsies of tumors followed longitudinally. If these tumors progress along the same adenoma–carcinoma sequence that has been described in human colonic tumors, we might have expected to see changes in gene expression over time. However, this was not observed. This may be related to the dynamics of tumor progression in this animal model, but could also be explained by the small sample size. Alternatively, this may be a manifestation of cancer predetermination; by the time colonic tumors are large enough to be detected, they have already acquired mutations that will cause them to become invasive (24).

We did identify a number of differentially expressed genes that have been linked to intestinal tumors in Apc^Min/+ mice and human colorectal cancer in this small sample of tumors. Resistin-like molecule β (RELM-β, Retnlb), Mucin 2 (Muc2), and intestinal trefoil factor 3 (Tff3) were expressed at a higher level in adenomas relative to intramucosal carcinomas. These genes encode for proteins secreted by goblet cells in the intestine, which play roles in innate intestinal immunity and response to injury (25–27). All have been linked to intestinal tumors in Apc^Min/+ mice and human colorectal cancer (27–33). TNF (ligand) superfamily, member 10 (Tnfsf10) and Lipocalin 2 (Lcn2), were overexpressed in intramucosal carcinoma relative to adenomas and are also linked to Apc^Min/+ intestinal tumors and human colorectal cancer. Tnfsf10, which is involved in the TRAIL apoptotic pathway, is dysregulated in both Apc^Min/+ tumors and human polyps (34–36). Elevated lipocalin 2, a molecule involved in innate immunity and iron metabolism (37), has been found in both Apc^Min/+ tumors and human colorectal cancer (38,39) and may function as either a tumor suppressor or promoter depending on location in the intestine (40). All of these results need to be confirmed in larger samples and by quantitative PCR before definitive statements can be made about their role in tumor formation and progression in this model.

There are important limitations to this model. F1 Min mice carry a germline mutation in Apc, which is analogous to familial adenomatous polyposis (FAP) syndrome in humans. Although hereditary colorectal cancer from syndromes such as FAP and hereditary nonpolyposis colon cancer (HNPCC) accounts for approximately 5% to 10% of all cases of colorectal cancer, the vast majority of cancers arise sporadically. Because it is a germline mutation, additional aberrations are seen in immune function, bone marrow microenvironment, and ammonia metabolism in the Min mouse (41). As with most models of colorectal cancer, the rate of adenocarcinoma is low, and most polyps remain benign adenomas (41). This may be related to the relatively short lifespan of Min mice (typically 3 to 4 months), which does not allow sufficient time for accumulation of additional mutations that would allow tumors to progress to invasive cancer. However, even in our long-lived F1 Min mice with a mean lifespan of about nine months, the rate of invasive lesions was low, particularly in the colon, and there was no evidence grossly of metastatic disease. In this sense, this model does not fully recapitulate the natural history of human colorectal cancer. In addition, we used DSS, an inflammatory agent, to increase colon tumor multiplicity. Although chronic inflammation is certainly a risk factor for colorectal cancer as is seen in the increased incidence of cancer in patients with inflammatory bowel disease, the role of inflammation in sporadic colorectal cancer is incompletely understood (42). The utility of murine models of colorectal cancer lies in their relative simplicity. By introducing this additional element to our model, we have added further complexity and potentially moved the model away from the natural history of sporadic human colorectal cancer in the general population, although we do observe similar outcomes in mice and humans.

This model affords a number of exciting future directions for research. We want to expand our characterization of the molecular and genetic changes identified by our preliminary work with additional IHC and quantitative PCR. This model could be used to investigate chemopreventive interventions. Because the early steps of tumorigenesis for many cases of colorectal cancer are recapitulated here, it provides an opportunity to intervene early in the adenoma–carcinoma sequence. We have carefully characterized the underlying growth patterns of colorectal tumors and can therefore effectively measure relative changes in growth patterns and rates of tumor regression induced by preventive measures. A criticism of mouse models of human colorectal cancer has been that they are not predictive; rather, investigators often report the effect on the multiplicity of intestinal tumors along the entire length of the intestinal tract instead of the colon, which is likely to be the most relevant to human disease. The F1 Min characterized in this study would be highly beneficial in testing newly developed preventive measures as the results are likely to be more predictive of outcomes in human populations. Finally, as we have observed adenocarcinoma in this model, it can also serve as a platform for investigating and characterizing responses to chemotherapeutic agents.

In summary, we have developed a long-lived mouse model of colorectal cancer that develops tumors in the distal colon that can be monitored by serial colonoscopy, shows distinct growth patterns similar to those seen in human colorectal polyps, and develops lesions that progress to adenocarcinoma. Serial biopsies of tumors in this model allow examination of cellular and molecular features of tumors over time. This novel murine model of colorectal cancer can be used to find promising prognostic targets that could lead to improved risk stratification of patients with polyps or cancers identified on colonoscopy, thereby improving this important but imperfect screening tool.

No potential conflicts of interest were disclosed.

Conception and design: T.J. Paul Olson, J.N. Hadac, D.A. Deming, R.B. Halberg

Development of methodology: T.J. Paul Olson, J.N. Hadac, W.R. Schelman, R.B. Halberg

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T.J. Paul Olson, J.N. Hadac, C. Sievers, A.A. Leystra, D.A. Deming, C.D. Zahm, D.M. Albrecht, A. Nomura, L.A. Nettekoven, L.K. Plesh, R.B. Halberg

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T.J. Paul Olson, J.N. Hadac, C. Sievers, D.A. Deming, R. Sullivan, M.A. Newton, W.R. Schelman, R.B. Halberg

Writing, review, and/or revision of the manuscript: T.J. Paul Olson, J.N. Hadac, C. Sievers, D.A. Deming, C.D. Zahm, A. Nomura, L.A. Nettekoven, L. Clipson, W.R. Schelman, R.B. Halberg

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T.J. Paul Olson, C. Sievers, L.A. Nettekoven, L. Clipson

Study supervision: W.R. Schelman, R.B. Halberg

The authors thank the NIH for funding support: T32 CA090217 (T.J.P. Olson), T32 CA009135 (J.N. Hadac, A.A. Leystra, and C.D. Zahm), T32 CA009614 (D.A. Deming), P30 CA014520 (R. Sullivan), R21 HG006568 (M.A. Newton), and R21 CA170876 (W.R. Schelman, G.D. Kennedy, and R.B. Halberg). The authors also thank the Division of Gastroenterology and Hepatology, Department of Medicine, University of Wisconsin School of Medicine and Public Health, and the University of Wisconsin Paul P. Carbone Cancer Center for funding support.

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.