Weight Loss and/or Sulindac Mitigate Obesity-associated Transcriptome, Microbiome, and Protumor Effects in a Murine Model of Colon Cancer

The association between obesity and colon cancer risk is well established (1, 2), with stronger correlations observed for central obesity. Weight loss in adulthood and reduced colon cancer risk are also linked, albeit less well studied. Rapp and colleagues (3) demonstrated that weight loss of >0.1 kg/m²/year is associated with a 50% reduction in colon cancer risk in men but not women. However, other studies found that intentional waist circumference loss of ≥5% or weight loss of ≥20 pounds in postmenopausal women is each significantly correlated with reduced colon cancer risk (4, 5). This discrepancy in findings likely relates to Rapp and colleagues (3) reliance on body mass index to assess obesity and weight loss as well as its inclusion of both premenopausal and postmenopausal women, as women are more prone to central obesity after menopause (6). The effect of weight loss on progression of established colon tumors or adenomas remains unclear, though one study demonstrated that a >5-pound weight loss does not impact adenoma recurrence (7).

Several interrelated molecular mechanisms, including chronic inflammatory signaling, are proposed to explain the obesity-cancer link (8). Obesity in humans and mice stimulates proinflammatory shifts in the immune cell and cytokine profile of the colon. Preclinical studies demonstrated that obesity alters colonic infiltration of cytotoxic CD8⁺ T cells and immunosuppressive regulatory T cells and increases TNFα, IFNγ, IL6, and IL18 expression in the colon (8, 9). A clinical trial using rectosigmoid biopsies showed that weight loss in obese premenopausal women reduces TNFα and IL6 expression levels as well as T-cell and macrophage infiltration in colon epithelium (10). The impact of obesity on the infiltration and activity of immune cells in established colon tumors is less clear. One study found that mismatch repair–proficient colon tumors in obese individuals have significantly lower density of both CD8⁺ and regulatory T cells (11). Others reported that the obesity–colon cancer association is independent of tumor-infiltrating T-cell density (12). However, Wunderlich and colleagues (13) reported that obesity-induced IL6 secretion promotes a shift in macrophage polarization that leads to protumorigenic changes in B- and T-cell recruitment in a mouse model of colitis-associated colon cancer.

Obesity-induced shifts in the colonic immune landscape may relate to changes in gut microbiota, given that these bacteria are closely connected to intestinal immune system development and function. Fecal transplants from high-fat diet (HFD)-fed mice promote the progression of small intestine tumors through a mechanism that involves Toll-like receptor signaling (14). Prebiotic treatment prevents gut microbial dysbiosis and colon tumor progression in mice with HFD-induced obesity (15). However, the interplay between the gut microbiota, local immune activity, and colon tumorigenesis in the context of HFD and obesity remains unclear.

Per epidemiologic and experimental evidence, NSAIDs reduce colon cancer risk (16). Moreover, the function of NSAIDs and the molecular targets for their anticancer effects on colon cancer have been extensively studied (16–18). However, the impact of NSAIDs on obesity-driven colon cancer is poorly characterized, although our previous studies suggest that obese or overweight patients with breast cancer may receive greater benefit from postdiagnostic NSAID use due to elevated local inflammation (19). Sulindac is an NSAID that is metabolized to sulindac sulfide, which suppresses inflammation by inhibiting cyclooxygenase (COX)-1 and -2 (16, 20), and sulindac sulfone, which exerts COX-independent antitumor effects (17, 21). Sulindac's antitumor activity in the colon has been demonstrated in several clinical (22–24) and preclinical (21, 25, 26) studies, but its impact on obesity-associated colon tumorigenesis has not previously been examined.

In the current study, we assessed the impact of three interventions on colon tumor multiplicity and burden in obese mice: (i) sulindac treatment, (ii) weight loss via change to low-fat diet (LFD), and (iii) a combination of both. The interventions started 15 weeks after colon tumor initiation via the chemical carcinogen azoxymethane (AOM) to determine their impact on established tumors and ability to prevent new tumors. Furthermore, because NSAIDs and weight loss interventions dampen obesity-associated inflammatory signals (17, 27), we examined the role of altered inflammatory and immune cell signaling in the tumor, and changes in gut microbial composition, in mediating antitumor effects. Here we report that sulindac and a LFD weight-loss regimen, alone or in combination, reverse obesity-driven changes in gut dysbiosis and tumoral inflammatory and immune signaling pathways via largely nonoverlapping modes of action to effectively reduce colon tumor multiplicity and burden.

Animal studies and procedures were approved and monitored by the University of North Carolina Institutional Animal Care and Use Committee. Male, 12-week-old FVB/N mice were purchased from Charles River Laboratories, Inc., and fed LFD (10% kcal from fat, catalog no. D12450J, Research Diets, Inc.) ad libitum for 1 week of acclimatization. All mice then began 5 weeks of AOM (MilliporeSigma) treatment (10 mg/kg/week, i.p.) to induce colon tumors, continuing on LFD. The mice were then randomized to remain on LFD (n = 45) or change to HFD (n = 85; 60% kcal from fat, catalog no. D12492, Research Diets, Inc.). The LFD fat source was 44% lard and 56% soybean oil; the HFD fat source was 91% lard and 9% soybean oil. Supplementary Table S1 provides additional information regarding nutrient sources and composition of each diet. Fifteen weeks later, 5 mice from each group were euthanized. Colon tumors in these mice were counted and two dimensions of each measured using digital calipers. The remaining LFD-fed mice were randomized to remain on LFD (control, n = 20) or change to LFD containing 140 ppm sulindac (control+Su, n = 20). The remaining HFD-fed mice were randomized to remain on HFD (obese, n = 18) or change to HFD containing 140 ppm sulindac (obese+Su, n = 19), LFD [formerly obese (FOb-LFD), n = 18], or LFD + sulindac (FOb-LFD+Su, n = 20). Preclinical studies have used dietary sulindac supplementation ranging from 160 to 320 ppm (25, 26). Our 140 ppm sulindac dose is approximately 0.02 mg/g body weight per day for control mice, which equates to roughly 250 mg for a 100-kg human when adjusting for allometric scaling (28) and falls within the daily 150–300 mg dose range frequently used in human studies (22, 23). Sulindac was obtained from the NCI's Chemical Repository.

Seven weeks later, the mice were euthanized, blood was collected by cardiac puncture, and serum separated and stored at −80°C. Colon tumors were counted and two dimensions of each measured. Colons from half of the mice in each group were rolled from distal to proximal end, formalin-fixed and paraffin embedded (FFPE) for histology. All tumor tissue was collected from the colons in the other half of the mice, flash-frozen in liquid nitrogen, then stored at −80°C. Body fat was assessed after euthanasia via a Lunar PIXImus Dual Emission X-Ray Absorptiometer (GE Medical Systems) as described previously (29).

Concentrations of several proteins were measured in serum collected at euthanasia. All proteins measured are listed in Supplementary Table S2. All hormones were measured via a BioPlex Pro Mouse Diabetes Panel (Bio-Rad), except insulin-like growth factor 1 and VEGF, which were assessed with Bio-Rad singleplex assays. Cytokines were measured via a BioPlex Pro Mouse Cytokine Panel. All assays were read on a Bio-Rad MAGPIX multiplex analyzer.

A veterinary pathologist, blinded to sample identities, scored FFPE colons (n = 6/group, randomly selected) for dysplasia or neoplasia, using a 1–5 scale (briefly, 1: aberrant crypt foci, 2: polyploid hyperplasia/dysplasia, 3: adenomatous and/or sessile hyperplasia/dysplasia, 4: intramucosal carcinoma, 5: invasive carcinoma) adapted from Meira and colleagues (30). The score reflected the most severe lesion captured in each colon slide, irrespective of percent area and number of tumors.

Colon tumor tissue was excised and combined for each tumor-bearing mouse. Total RNA was isolated from each pooled colon tumor sample via TRIzol reagent (Thermo Fisher Scientific) according to manufacturer's instructions. Total RNA integrity was assessed via RNA ScreenTape analysis (Agilent Technologies). The transcriptome of 5–6 RNA samples per group from the control, obese, FOb-LFD, and obese+Su groups was then profiled via Mouse Clariom S Assay HT (Thermo Fisher Scientific). Total RNA was used to synthesize fragmented and labeled sense-strand cDNA and hybridized onto the Clariom S peg plate. The GeneChip WT PLUS Reagent Kit (Affymetrix) was used to prepare the samples, following manufacturer's instructions. Fragmented and labeled cDNA was used to prepare a hybridization cocktail with the GeneTitan Hybridization Wash and Stain Kit for WT Arrays (Affymetrix). Hybridization, washing, staining, and scanning of the Clariom S peg plate was carried out using the GeneTitan MC Instrument (Affymetrix). Transcriptome Analysis Console Software v 4.0 (TAC, Thermo Fisher Scientific) was used for basic data analysis and quality control, including determination of differential gene expression (P < 0.05, absolute fold change ≥2.0) and identification of enriched canonical pathways (P < 0.05) for pairwise comparisons.

Ingenuity Pathway Analysis (IPA, Qiagen) was used to determine enriched functions and predicted upstream regulators (P < 0.05) for the pairwise comparisons of microarray data, whether there were known regulatory interactions between subsets of genes, and whether these interactions were consistent with the differential expression of those genes.

Gene set enrichment analysis (GSEA version 4.1.0) was performed for the microarray data pairwise comparisons using the Broad Institute MSigDB Hallmark gene sets (version 7.4). A normalized enrichment statistic was generated using default settings (31). A FDR q < 0.05 was considered statistically significant.

Total RNA was isolated as described above from tumor tissues not previously used for the microarray analysis, then reverse transcribed and samples assayed in triplicate for individual genes as described previously (32).

All fecal matter in the colon was collected upon dissection after euthanasia, flash-frozen, and stored at −80° C. Samples were combined with ≤106 μmol/L glass beads (MilliporeSigma), Qiagen ATL buffer, and lysozyme (Thermo Fisher Scientific) and incubated at 37°C for 1 hour with occasional agitation. Proteinase K was added, and the samples incubated at 60°C for 1 hour. Qiagen AL buffer was added, and the samples incubated at 70°C for 10 minutes. Three minutes of bead beating in a Qiagen TissueLyser II at 30 Hz occurred. After centrifugation, supernatants were transferred to a tube containing ethanol. A standard on-column DNA purification method was then used, and DNA eluted in 10 mmol/L Tris (pH 8.0).

DNA was amplified using universal primers targeting the V4 region of the bacterial 16S rRNA gene (33, 34). Primer sequences contained overhang adapters appended to the 5′ end of each primer for compatibility with Illumina sequencing platform. The complete sequences of the primers were:

515F - 5′ TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGGTGCCAGCMGCCGCGGTAA 3′

806R - 5′GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGGGACTACHVGGGTWTCTAAT 3′. Master mixes contained 12.5 ng of total DNA, 0.2 μmol/L of each primer and 2× KAPA HiFi HotStart ReadyMix (KAPA Biosystems). Each sample was amplified as follows: 95°C for 3 minutes, (95°C for 30 seconds, 55°C for 30 seconds, 72°C for 30 seconds) × 25 cycles, hold at 4°C. Each 16S amplicon was purified using AMPure XP reagent (Beckman Coulter), then amplified as follows to add Illumina sequencing adapters and dual‐index barcodes [index 1(i7) and index 2(i5)] to the amplicon target: 95°C for 3 minutes (95°C for 30 seconds, 55°C for 30 seconds, 72°C for 30 seconds) × 8 cycles, 72°C for 5 minutes, hold at 4°C. The final libraries were purified by AMPure XP reagent, quantified, and normalized before pooling. The DNA library pool was denatured with NaOH, diluted with hybridization buffer, and heat denatured before loading on a MiSeq instrument (Illumina). Automated cluster generation and paired-end sequencing with dual reads were performed according to the manufacturer's instructions.

Multiplexed paired-end fastq files were produced from the sequencing results of the Illumina MiSeq using the Illumina software configureBclToFastq. The paired-end fastq files were joined into a single multiplexed, single-end fastq using the software tool fastq-join. Demultiplexing and quality filtering was performed on the joined results. Bioinformatics analysis of the data was conducted using the Quantitative Insights Into Microbial Ecology (QIIME) software (35). Operational taxonomic unit (OTU) picking was performed using pick_de_novo_otus.py. Chimeric sequences were detected and removed using ChimeraSlayer. QIIME software was used to measure alpha and beta indices and produce taxonomic group assignments. Principal coordinates analysis (PCoA) was generated to display intersample distances using the first and second principal coordinates. A cross-correlational analysis was performed to identify linear relationships between bacterial taxa and tumor multiplicity, using Pearson correlation and permutation analysis in which labels were shuffled and correlations calculated, generating a null distribution of correlation coefficients and allowing subsequent calculation of P values.

A power analysis was conducted to determine the number of mice required to ensure power >90% to detect differences in colon adenocarcinoma number, adenocarcinoma incidence, and tumor size. A prior AOM-induced colon tumor study used 23 lean male mice which had 9.2 ± 1.8 tumors and 22 obese male mice which had 13.0 ± 1.3 tumors (36). For our power calculation, we assumed similar variability and effect size and therefore used 20 mice per group.

All data are presented as mean ± SD, with the exception of the microarray, PCoA, and correlation analyses. GraphPad Prism software (GraphPad Software Inc.) was used to assess differences between groups in body weight, body fat, serum proteins, alpha diversity measures, and bacterial taxa via two-way ANOVA, followed by Tukey post hoc test. GraphPad Prism was also used to identify statistical outliers in the serum proteins and bacterial taxa via ROUT method (37). Logistic regression was used to determine differences in tumor incidence. Because of overdispersion in the tumor multiplicity data, a negative binomial generalized linear model (GLM) was used to assess its significance. A χ² goodness-of-fit test was used to determine whether the data fit this model. The tumor burden data were assessed using a gamma GLM with the log link function. Ordinal logistic regression was used to assess the dysplasia or neoplasia scoring. For all statistical analyses of tumor data, the obese mice were compared with the other five groups. P < 0.05 was considered significantly different.

The microarray and 16S sequencing data generated in this study are publicly available in the Gene Expression Omnibus repository (GSE197846) and SRA (PRJNA820039), respectively.

The study design is illustrated in Fig. 1A. At euthanasia, both (FOb-LFD) and FOb-LFD+sulindac (Su) groups had significantly lower body weights than obese and obese+Su mice (P < 0.001 for all), but higher body weights than control and control+Su mice (P < 0.01 for all; Fig. 1B). Sulindac treatment did not significantly impact body weight in any diet group (Fig. 1B). Percent body fat at study endpoint did not significantly differ between non–sulindac-treated control, obese, and FOb-LFD mice, likely due to tumor-associated weight loss in the obese mice during the final 1–2 weeks of study. The obese+Su group had higher body fat percentage than control, control+Su (both P < 0.001), FOb-LFD, and FOb-LFD+Su (both P < 0.01) groups (Fig. 1C).

FOb-LFD mice, versus obese mice, had lower serum levels of IL6 (P < 0.05), GCSF, C-X-C motif chemokine ligand 1 (CXCL1), and VEGF (all P < 0.001; Fig. 1D–G). Obese+Su mice had lower serum levels of IL6, GCSF, CXCL1 (all P < 0.001), VEGF, and monocyte chemoattractant protein 1 (MCP-1; both P < 0.01) than obese mice (Fig. 1D–H). While the mean serum IL1β level was 34% lower in obese+Su mice than obese mice, this between-group difference did not achieve significance (P = 0.06; Fig. 1I). Supplementary Table S2 lists mean concentrations of all analytes measured.

At study week 20, a subset of control and obese mice were euthanized to assess interim colon tumor growth before diet switch. Tumor incidence was 80% in the obese mice versus 40% in control mice (Fig. 2A). Tumor multiplicity (P < 0.01) and burden (P < 0.05), the two primary study endpoints, were significantly greater in obese versus control mice (Fig. 2B and C). At study end, there were no differences between control, obese, and FOb-LFD mice in tumor incidence, but sulindac decreased incidence in all diet groups (all P < 0.001; Fig. 2D). Tumor multiplicity and burden were lower in FOb-LFD mice, versus obese mice (both P < 0.01). Sulindac treatment also decreased tumor multiplicity and burden in all diet groups (all P < 0.001; Fig. 2E and F). Specifically, sulindac (vs. no sulindac) decreased tumor burden by 84% in control mice, 86% in obese mice, and 78% in FOB-LFD mice (Fig. 2F). Furthermore, tumor multiplicity was greater in the obese mice at interim tissue harvest than the endpoint FOb-LFD (P < 0.01), obese+Su, and FOb-LFD+Su mice (both P < 0.001; Supplementary Fig. S1A). Tumor burden was also higher in obese mice at the interim tissue harvest than endpoint obese+Su and FOb-LFD+Su mice (both P < 0.01; Supplementary Fig. S1B). While there was no evidence for a synergistic interaction between LFD intervention and sulindac on colon tumor multiplicity or burden, these outcome measures differed significantly between the obese, FOb-LFD and FOb-LFD+Su groups (P < 0.001 for each), suggesting the two interventions exert additive effects (Fig. 2D and F). Colon pathology scores for dysplasia or neoplasia did not significantly differ between groups, but one-third of the obese mice scored 3 on the five-point scale, indicating adenomatous and/or sessile hyperplasia or dysplasia, while none of the other mice exceeded two (Supplementary Fig. S1C).

To explore possible mechanisms by which weight loss and sulindac treatment reversed the effects of obesity on colon tumor growth, we measured tumoral gene expression in control, obese, FOb-LFD, and obese+Su mice via Affymetrix microarray. GSEA was performed on the full transcriptomic profile of these tumors using the MSigDB Hallmark gene sets to capture changes in biological processes between treatment groups (31, 38). Pairwise comparisons between tumors from obese versus control mice identified enrichment of many gene sets related to inflammatory signaling (TNFA_SIGNALING_VIA_NFKB, ALLOGRAFT_REJECTION, INTERFERON_GAMMA_RESPONSE, INFLAMMATORY_RESPONSE, IL2_STAT5_SIGNALING, IL6_JAK_STAT3_SIGNALING, COMPLEMENT, and INTERFERON_ALPHA_RESPONSE; Fig. 3A). Of the 47 tumoral differentially expressed genes (DEG), IPA indicated that “Cell movement” and “Activation of leukocytes” were among the top five enriched functions (ranked by absolute z-score), with both decreased in control versus obese mice (Fig. 3B). IPA predicted little interaction among proteins encoded by 22 DEG associated with these two functions (Fig. 3C). However, IPA predicted IL6 to be an upstream regulator of six of these DEG (Fig. 3D), and serum IL6 levels were significantly elevated in obese mice versus controls (Fig. 1C).

GSEA using Hallmark gene sets evaluating tumor gene expression from control, FOb-LFD and obese mice revealed several gene sets significantly enriched in obese mice versus the other two groups (Fig. 3A). While ALLOGRAFT_REJECTION, IL2_STAT5_SIGNALING, and COMPLEMENT were the only overlapping immune-related gene sets, EPITHELIAL_MESENCHYMAL_TRANSITION, ANGIOGENESIS, COAGULATION, and KRAS_SIGNALING_UP were all enriched in tumors from obese mice relative to either control or FOb-LFD mice (Figs. 3A and 4A). Moreover, TNFA_SIGNALING_VIA_NFKB and INTERFERON_ALPHA_RESPONSE gene sets were both significantly enriched in tumors from FOb-LFD relative to obese mice, suggesting weight loss may enhance tumoral immune responses. IPA of the obese versus FOb-LFD DEG indicated that “Focal Adhesion” and “Matrix Metalloproteinases” were significantly enriched canonical pathways (Fig. 4B). Moreover, “Chemotaxis,” “Invasion of cells,” “Inflammation of organ,” and “Inflammation of absolute anatomical region” were among the top 10 enriched IPA functions (Fig. 4C). Finally, analysis of the overlap between the two inflammatory IPA functions with the 36 chemotaxis or invasion-associated DEG revealed 17 overlapping DEG (Fig. 4D).

IPA determined all but three of the 17 overlapping genes encode proteins have known interactions with at least one other protein included in this set. Fn1 (fibronectin 1), Mmp9 (matrix metalloproteinase 9), and Tgfbr2 (TGFβ receptor 2) were all focal points within these interactions that were predicted to regulate or be regulated by multiple other genes within the set (Fig. 4E). TGFB1, TGFBR2, and SMAD3 were among the 10 most significant predicted upstream regulators for these DEG (Fig. 4F), and all but four of the 17 genes linked to chemotaxis or invasion and inflammation are TGFB1 targets. Given that “TGFβ signaling” was also included among the top significant canonical pathways for this comparison (Fig. 4B) and Tgfb3 expression was elevated in obese mouse tumors, these data suggest that TGFβ signaling plays a key role in the differential gene expression between obese and FOb-LFD mouse tumors. Unfortunately, there was not sufficient sera remaining to measure serum TGFβ levels.

Relative to tumors from obese+Su mice, tumors from untreated obese mice were enriched for EPITHELIAL_ MESENCHYMAL_TRANSITION and COAGULATION Hallmark gene sets (Fig. 5A). These two gene sets were also significantly enriched in obese mouse tumors relative to tumors from control and FOb-LFD mice (Figs. 3A and 4A). A leading-edge analysis was subsequently performed to determine which genes consistently and significantly contributed to the EPITHELIAL_MESENCHYMAL_TRANSITION or COAGULATION enrichment scores across all three pairwise comparisons. This analysis identified 49 genes of which Lum, Pthlh, Sfrp4, Mmp8, and Serpine1 ranked in the top 10 of the member gene lists for all comparisons (Supplementary Table S3).

The 122 DEG (P < 0.05) for the obese versus obese+Su tumor comparison, like obese versus FOb-LFD, were enriched in the “Matrix metalloproteinases” canonical pathway (Fig. 5B). Sulindac treatment significantly reduced Mmp3 (P < 0.05), Mmp10 (P < 0.01), and Mmp13 (P < 0.05) expression in obese mice, while FOb-LFD reduced Mmp9 and Mmp12 (P < 0.05 for both). In addition, the DEG set was enriched in the “Chemokine signaling pathway” (Fig. 5B), which included elevated Ccl21 (C-C motif chemokine ligand 21) and reduced Ccl11 expression in sulindac-treated obese mice. The top 10 enriched IPA functions for this comparison included decreased “Inflammation of joint” and “Quantity of granulocytes” in the obese+Su mice (Fig. 5C). The 21 DEG linked to the inflammation and granulocyte-related functions also included Ccl11 and Ccl21 (Fig. 5D). Exploration of potential interactions among these DEG demonstrated that Mmp3 was a focal point for a subset of eight interconnected genes within this group that includes Ccl11 (Fig. 5E), suggesting this chemokine may play a role in sulindac-induced suppression of Mmp3 expression. None of these genes were modulated in the obese versus FOb-LFD comparison. However, two functions related to inflammation as well as the “Inflammatory response” canonical pathway were enriched in the obese group relative to FOb-LFD mice, suggesting both interventions affected different inflammatory pathways (Fig. 4B and C).

Additional top IPA functions included “Cell survival,” “Cell viability,” “Neoplasia of cells,” and “Invasion of tissue” (Fig. 5C). The 21 DEG linked to cell survival and viability overlapped almost completely. However, IPA identified very few potential interactions among the proteins encoded by these genes (Fig. 5F). None of these functions appeared in the top 10 list for the FOb-LFD versus obese comparison. Finally, the most significant predicted upstream regulator for this comparison was IL1β, but the 34% reduction in mean serum IL1β level in obese+Su versus obese mice was not statistically significant (P = 0.06; Fig. 1I). IL1β and an inhibitor of the NFκB kinase complex were also significant upstream regulators in our analysis (Fig. 5G). Overall, the substantial overlap between the target genes for these upstream regulators and the genes linked to the pathways and functions described above suggests that suppression of IL1β and/or downstream NFκB signaling mediates sulindac's antitumor effects. A subset of these genes (Ccl11, Ccl21, Ngf, Mmp3, Mmp10, Mmp13) was validated by quantitative RT-PCR (Supplementary Fig. S2), demonstrating that their mRNA levels were significantly different in obese+Su and/or FOb-LFD mice, relative to obese mice, for all genes except Mmp3 (P = 0.09 for the obese+Su vs. obese comparison). Ccl11 (P < 0.01), Ccl21 (P < 0.05), Mmp10 (P < 0.001), and Mmp13 (P < 0.05) were modulated in the obese versus obese+Su comparison. For the obese versus FOb-LFD comparison, Ccl11 (P < 0.01), Ngf (P < 0.05), Mmp10 (P < 0.01), and Mmp13 (P < 0.05) were significantly different.

To assess potential links between treatment-induced changes in the gut microbiota and colon tumor outcomes, 16S rRNA amplicon sequencing of fecal samples collected at study endpoint was performed. PCoA demonstrated fair clustering of samples by treatment group and separation between these clusters, suggesting that diet and/or adiposity and sulindac treatment are responsible for most of the variability in gut microbial composition (Fig. 6A). Obesity did not impact fecal ɑ-diversity, but sulindac reduced the number of OTUs in obese and FOb-LFD mice (P < 0.05 for both) and the Chao1 index values in control (P < 0.05) and obese (P < 0.01) mice (Fig. 6B and C). Phylum level taxonomic assessment demonstrated that percent abundance of Verrucomicrobia was elevated in obese mice, relative to control, FOb-LFD, and obese+Su mice (P < 0.001 for all; Fig. 6D; Supplementary Fig. S3A). A strong diet effect was not observed among any other phyla, though Bacteroidetes abundance was less in the obese+Su group than the other two sulindac-treated groups (P < 0.01 for both), and sulindac increased Cyanobacteria in control (P < 0.05) and obese (P < 0.01) mice (Fig. 6D; Supplementary Fig. S3B and C). At the family, genus, and species levels, the percent abundance in obese versus control and FOb-LFD mice of Bilophila (P < 0.01, P < 0.001), Coprobacillus (P < 0.01, P < 0.001), and Parabacteroides distasonis (P < 0.001, P < 0.01) were higher, and S24-7 (P < 0.001 for both) was lower. Sulindac treatment decreased Ruminococcus flavefaciens in all diet groups (P < 0.05 for control and obese, P < 0.001 for FOb). Furthermore, sulindac treatment decreased Bilophila (P < 0.001), Coprobacillus (P < 0.05), P. distasonis (P < 0.001), Prevotella (P < 0.01), and Akkermansia muciniphila (P < 0.001), and increased S24-7 (P < 0.05), F16, and Roseburia (both P < 0.001) in obese mice (Fig. 6E; Supplementary Fig. S3D–L). Cross-correlational analysis identified 13 taxa significantly correlated with tumor multiplicity (P < 0.05), including positive correlations with Coprobacillus, Prevotella, R. flavefaciens, P. distasonis, and A. muciniphila and negative correlations with F16 and Roseburia (Fig. 6F). These seven taxa were all modulated by sulindac in obese mice, but only Bilophila, Coprobacillus, P. distasonis, and S24-7 were also modulated with diet change and/or weight loss in the FOb-LFD group, relative to obese mice (Supplementary Fig. S3D–L). Neither sulindac nor weight loss significantly changed the abundance of the other six correlated taxa in obese mice (Supplementary Fig. S3M–R).

While obesity positively correlates with colon cancer risk (1, 2), and others have explored the role of inflammatory signaling in obesity-associated colon tumor growth (8, 9), researchers have not previously examined whether weight loss and/or NSAID treatment reverses obesity-induced tumor growth in a mouse model of colon cancer. Our findings suggest that weight loss and/or treatment with the NSAID sulindac—initiated following colon cancer tumor formation via AOM—eliminates the protumor effects of obesity. More specifically, both interventions significantly reduced tumor multiplicity and burden in mice fed a 60 kcal% HFD.

Tumor multiplicity and burden were significantly lower in obese+Su, FOb-LFD, and FOb-LFD+Su mice compared with obese mice, both at study end and week 20. These data indicate that sulindac treatment and weight loss (via a diet switch from 60 kcal% HFD to 10 kcal% LFD), alone or in combination, caused regression of colon tumors in obese mice. In addition, both measures were significantly lower in FOb-LFD+Su mice compared with FOb-LFD mice, suggesting an additive effect with the two interventions. However, there was no evidence of a synergistic effect and no difference in either measure between obese+Su and FOb-LFD+Su mice, likely because sulindac was so effective on its own. Surprisingly, we observed less tumor multiplicity in our current study than in our previous study (36) that used a 45 kcal% fat diet and the same AOM dosing schedule, mouse strain, and animal vendor as here. This disparity may reflect differences in animal husbandry conditions, including microbial communities, or AOM potency as the previous study was completed at NIH using a different AOM lot.

Two prior preclinical studies that examined the impact of diet change and/or weight loss on obesity-associated colon tumorigenesis found no reduction in colon tumor multiplicity (39, 40). This variance from our findings is likely related to study design differences, including the previous studies’ use of C57BL/6 mice, administration of AOM after obesity induction, and the more moderate fat content of their HFDs (40%–45% kcal from fat). Our study used FVB/N mice, which are more sensitive to AOM but less sensitive to HFD-induced obesity, and we initiated tumorigenesis before obesity induction and started weight loss 15 weeks after AOM completion. Given that weight loss regimens began much earlier in the previous studies’ timelines, it's surprising that our weight loss intervention was the only one to successfully reverse obesity's protumor effects. One possible explanation is the C57BL/6 strain's heightened response to HFD feeding relative to most other strains, resulting in a more pronounced obese phenotype that may be less responsive to weight-reversing interventions such as LFD regimens.

A study limitation is our use of a 60 kcal% HFD to induce obesity in mice, which is higher than the average approximately 37 kcal% fat currently consumed in the United States (41). A related limitation is that we cannot distinguish the effects of obesity per se versus a HFD effect independent of obesity. Two well-studied diet-induced obesity regimens are based on either 45 or 60 kcal% fat (mostly lard) regimens (42). We elected to use the higher fat and calorie diet in this study of weight loss and/or sulindac effects to ensure we achieved a robust obesity phenotype in the FVB/N mouse, which relative to the C57BL/6 strain, is sensitive to AOM-induced colon carcinogenesis yet resistant to diet-induced obesity (36). The 60 kcal% HFD regimen recapitulates in numerous mouse strains many comorbidities of human obesity, including hyperglycemia or hyperinsulinemia and chronic systemic inflammation (42). Similar to the 60 kcal% HFD, the macronutrient profile of the 10 kcal% LFD does not reflect a typical human consumption pattern, but it is commonly used in preclinical studies as a control diet to match the purified ingredients of the 45 or 60 kcal% HFD, only modifying their relative proportions. However, this is an additional study limitation, and future studies should consider testing diets with a more balanced macronutrient profile to increase translatability to the human population.

After determining that obesity enhanced, and diet change and/or weight loss suppressed, colon carcinogenesis in our study, we sought to define mediating pathways. Comparison of the tumors in obese and FOb-LFD mice suggested that increased invasion, inflammation, and TGFβ signaling underlie the procancer effects of obesity. TGFβ is a pleiotropic cytokine that regulates immune homeostasis and tumor immune evasion (43). Tumor expression of TGFβ correlates with metastasis and poor prognosis in human colon cancer (44, 45), while TGFβ inhibition can prevent colon cancer metastasis by unleashing a cytotoxic T-cell response against the tumor cells (46). However, mutational inactivation of TGFBR2, a DEG upregulated in our obese mouse tumors, occurs in approximately 30% of all human colon cancer (47), and loss of Tgfbr2 promotes colon tumorigenesis in AOM-treated mice (48). These findings illustrate the dual nature of TGFβ, which typically acts as a tumor suppressor in the early stages of tumor formation but then promotes tumor growth and metastasis in later stages.

Our study demonstrated that sulindac completely suppressed the colon tumor-promoting effects of obesity. Numerous studies have shown that sulindac effectively inhibits human adenoma growth in the colon (22, 23) and tumor growth in preclinical models of colon cancer (21, 25, 26), but its efficacy in reducing obesity-associated colon tumorigenesis has not been tested previously. Aspirin use negates a positive correlation between obesity and colon cancer risk in patients with Lynch syndrome with MLH1 mutations (49). However, a case–control study assessing interactions between regular NSAID use and lifestyle or dietary factors found that an elevated body mass index attenuates the NSAID–colon cancer association, with aspirin primarily driving this interaction. Specifically, an association existed between regular aspirin use and colon cancer risk in normal weight and overweight subjects but not obese subjects. This suggests that obese individuals may need a higher NSAID dose or frequency to achieve the same preventive benefits as normoweight individuals (50). Wu and colleagues (51) reported that a combination of the NSAID salsalate and curcumin inhibits HFD-induced colon tumor growth in AOM-treated mice. The authors speculated that the combination was required to suppress tumor growth because neither treatment alone adequately inhibited tumor-promoting pathways, like NFκB, in the colon epithelium.

Transcriptomic analyses from our study suggested that sulindac is effective at reducing tumor NFκB activity in obese mice, consistent with sulindac's known inhibition of IKKβ (17). This inhibition may be a key factor in sulindac's antitumor effects in the obese setting, as numerous studies have demonstrated that NFκB is an important pathway in colon cancer pathogenesis. Overactivation of this pathway in colon tumors is common and plays a role in regulating tumor cell proliferation, apoptosis, metastasis, and drug resistance as well as angiogenesis and inflammation in the tumor microenvironment (52). Also, obesity increases colon tumor NFκB expression, independent of diet, in the AOM-induced colon cancer model (53). While sulindac is a dual COX-1/2 inhibitor, ample evidence indicates that its antitumor activity is mediated primarily by modulation of other targets, including NFκB (21). This evidence includes large discrepancies between the doses needed for anti-inflammatory versus antitumor activity (17). These differences were demonstrated by measuring the IC₅₀ of several NSAIDs, including sulindac sulfide, for COX-1 and COX-2 inhibition in human mononuclear cells and comparing these values with the IC₅₀ values for growth inhibition in HT-29 human colon cancer cells (54, 55). Further support comes from the fact that sulindac sulfone, a sulindac metabolite with no anti-inflammatory activity, has antitumor effects (21). While we were unable to measure tumor prostaglandin concentrations due to a lack of tissue from the nonobese groups, transcriptomic analyses gave little evidence of between-group tumoral differences in COX activity or prostaglandin synthesis. Although the mechanisms mediating sulindac's antitumor effects remain unclear, our findings suggest that sulindac, or other NSAIDs, could have utility as adjuvant chemotherapy drugs for obese individuals with colon cancer.

GSEA of the tumor transcriptomic profile identified elevation of the EPITHELIAL_MESENCHYMAL_TRANSITION and COAGULATION gene sets in obese relative to control, FOb-LFD, and obese+Su mice. These data suggest that all interventions tested in this study may confer similar repressive effects on these two biological pathways, both of which are important for tumor growth and progression. Moreover, several potential mediators emerged from these gene sets including Lum, Pthlh, Sfrp4, Mmp8, and Serpine1, which have known roles in tumor biology but less well-understood relationships to obesity, weight loss, and NSAID treatment (56–59). Furthermore, there were similar elevations in matrix metalloproteinase (MMP) signaling in obese mice, relative to both FOb-LFD and obese+Su mice, though different genes were affected. Though both interventions modulated inflammatory signaling, potentially enhancing antitumor immune activity, we identified TGFβ as the likely primary pathway mediating the observed effects of FOb-LFD while NFκB appeared to play a prominent role in the effects of sulindac in the obese mice.

We did not assess adipose tissue inflammation, which is a study limitation. Differences in inflammatory signaling from the adipose tissue, particularly mesenteric adipose, may also contribute to the effects of obesity, diet change and/or weight loss, and sulindac treatment on colon tumor growth, and warrant future investigation.

Changes in the gut microbiota composition may have played a role in our interventions’ reversal of obesity-associated colon tumor growth. Several taxa were significantly correlated with tumor multiplicity and modulated by diet change and/or weight loss and/or sulindac, but Prevotella and A. muciniphila were of particular interest, given that both taxa are elevated in patients with colon cancer (60). Flemer and colleagues (60) demonstrated that the Prevotella Cluster was associated with higher MMP13 expression in patients with colon cancer. Sulindac treatment significantly reduced both Prevotella and tumor Mmp13 expression in our obese mice. A. muciniphila is typically decreased with obesity and is linked to several health benefits (61). However, its abundance is increased in colon cancer, particularly in patients with tumors in the distal colon or rectum (60). Sulindac reduced A. muciniphila levels in our obese mice, although the causal relationships between these microbiome changes and colon cancer remain unclear.

Neither Prevotella nor A. muciniphila were affected by the diet change and/or weight loss in FOb-LFD mice, but four taxa were modulated by both interventions in the obese mice and correlated with tumor multiplicity: S24-7 abundance increased while Bilophila, Coprobacillus, and P. distasonis all decreased in abundance. One study found that the abundance of P. distasonis is lower in colon tumor-bearing mice (62), while another report, in contrast with our findings, showed that supplementation with P. distasonis reduces AOM-induced colon tumor formation in obese mice (63). One species of Bilophila, B. wadsworthia, is a sulfidogenic bacterium linked to human colon cancer (64, 65), but its relative abundance was not altered in our study. To our knowledge, we are the first to report linkage of S24-7 or Coprobacillus to obesity and colon cancer.

Given the clear link between obesity and colon cancer, and the continued rise in obesity rates worldwide, there is an urgent need for better colon cancer preventive interventions for the obese population. Here we have shown that a LFD weight loss regimen and/or treatment with the NSAID sulindac effectively reverse the effects of obesity on tumor multiplicity and burden in the AOM model of colon cancer. Transcriptomic profiling suggests that these effects may be mediated, at least in part, through reductions in tumor TGFβ and NFκB signaling and related shifts to more antitumor immune profiles. Changes in obesity-related gut dysbiosis may also contribute to anticancer effects. Our findings indicate that further investigation regarding the effects of NSAID treatment on obesity-associated colon cancer risk and/or progression is warranted, particularly in obese individuals unable to achieve moderate weight loss.

C.M. Ulrich reports grants from NIH during the conduct of the study. No disclosures were reported by the other authors.

L.W. Bowers: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, writing–original draft, writing–review and editing. E.M. Glenny: Data curation, formal analysis, validation, writing–original draft, writing–review and editing. A. Punjala: Formal analysis. N.A. Lanman: Formal analysis. A. Goldbaum: Formal analysis. C. Himbert: Investigation. S.A. Montgomery: Conceptualization, data curation, formal analysis, validation, writing–original draft, writing–review and editing. P. Yang: Investigation. J. Roper: Investigation. C.M. Ulrich: Investigation. A.J. Dannenberg: Conceptualization, resources, writing–review and editing. M.F. Coleman: Conceptualization, data curation, formal analysis, validation, writing–original draft, writing–review and editing. S.D. Hursting: Conceptualization, resources, formal analysis, supervision, funding acquisition, writing–original draft, writing–review and editing.

This study was supported by grants from the NIH R35 CA197627, to S.D. Hursting; R01CA254108, to C.M. Ulrich, J. Roper, and S.D. Hursting; R25 CA057726, to L.W. Bowers; and T32DK07737, to E.M. Glenny.

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.