Association between visceral obesity, metformin use, and recurrence risk in early-stage colorectal cancer

We sought to investigate the association between visceral obesity with disease recurrence and survival in early-stage colorectal cancer (CRC) patients. We also wanted to examine if such an association, if exists, is influenced by metformin use. Stage I/II CRC adenocarcinoma patients treated surgically were identified. L3 level CT VFI (visceral fat index) was used as a metric of visceral obesity and was calculated as the proportion of total fat area composed of visceral fat. N = 492. 53% were males, 90% were Caucasians, 35% had stage I disease, and 14% used metformin. 20.3% patients developed a recurrence over a median follow-up of 56 months. VFI was associated with both RFS and OS in a multivariate model, but not BMI. The final multivariate model for RFS included an interaction term for VFI and metformin (p = 0.04). Confirming this result, subgroup analysis showed an increasing VFI was associated with a poor RFS (p = 0.002), and OS (p < 0.001) in metformin non-users only and metformin use was associated with a better RFS only in the top VFI tertile (p = 0.01). Visceral obesity, but not BMI, is associated with recurrence risk and poorer survival in stage I/II CRC. Interestingly, this association is influenced by metformin use.


Results
Cohort characteristics. After applying the inclusion and exclusion criteria, CT scans for 492 patients were identified and analyzed for visceral obesity quantification. The cohort's median age at the time of tumor diagnosis was 64 years (Inter Quartile Range [IQR] = 54-74), with a majority of the patients being male (53%). Most of the patients were white (90%). A majority of the patients consumed alcohol (62%) at the time of primary tumor diagnosis, whereas 142 (30%) and 32 (6%) patients had never consumed or consumed alcohol only in the past, respectively. There were 88 (18%), 164 (35%), and 223 (35%) current, former, and never smokers, respectively. Colon was the primary tumor site in 361 (73%) and rectum in 131 (27%) patients. 173 (35%) of the patients had stage I disease. 67 (14%) patients were metformin users. The median follow-up of the cohort was 56 months (IQR = 35-92), during which recurrence developed in 100 (20%) patients and 120 (24%) patients died. The liver was the most common site of the first recurrence, with 42 (8.5%) patients developing recurrence in the liver, followed by locoregional (6.5%), lung (3.9%), and other distant sites (1.4%). The median BMI of the cohort was 28.3 kg/m 2 (24.3-32.7), and the median VFI and SMI were 0.44 (0.34-0.55) and 47.5 cm 2 /m 2 (40.1-56.0), respectively. To better understand the relationship between VFI and recurrence, we categorized VFI into 'Top' , 'Middle' , and 'Bottom' tertiles based om if patients fall in the top, middle or bottom 33.33 percentile of the VFI range within the cohort.
Univariate analysis across VFI teritiles. Univariate analysis to compare the sociodemographic and histopathological variables of CRC patients among the different VFI tertile groups revealed that patients in the top tertile were more likely to be older ( (Fig. 2). Neither BMI nor SMI was found to be prognostic for RFS as they were not included in the final model of the multivariate analysis (  Fig. 2).
In the above analysis we analyzed patients with colon (N = 361 [73%]), and rectum (N = 161 [275]) as the primary tumor sites. To potentially account for the differing treatment approaches in patients based on the primary tumor site, we repeated the above analyses as subgroup analysis based on primary tumor site 9,10 . We found that similar to the entire cohort, age, pathological stage, and VFI were retained in the final multivariate model for RFS, whereas, age, race, and VFI were retained for OS as prognostic indicators in patients with colon as the primary site (Supplementary Table S2 www.nature.com/scientificreports/ any significant results potentially due to the low power in the analysis given the small sample size. Additionally, we performed another analysis including the number of lymph nodes examined, if patients received adjuvant/ neoadjuvant chemotherapy or not, and using pathological T stage instead of pathological stage I or II. This additional analysis confirmed our observations with VFI as a significant prognostic indicator for both RFS, and OS (Supplementary Table S3).

Metformin and VFI interaction and subgroup analysis.
Our previous work showed a significant interaction between metformin and obesity when assessing outcomes in lung cancer patients 8 . We sought to assess if a similar interaction between VFI and metformin existed in colorectal cancer by introducing an interaction variable between VFI and metformin use to the multivariate model. This modeling showed that this was indeed the case, whether VFI was used as a continuous or categorical (tertiles) variable. Variables retained in the final model are highlighted in Table 3. A similar analysis for OS did not retain the metformin/VFI interaction variable in the final model. To further explore the interaction between metformin use and visceral obesity, a subgroup analysis examining the association between metformin use and RFS in each of the VFI tertiles was performed. Similarly, the relationship between VFI and outcome in metformin users and non-users was also assessed (Supplementary Table S1 Fig. S1). Metformin use was associated with a significantly better RFS only in the  Fig. S2).

Discussion
A European study suggests that obesity is responsible for around 11% of colorectal cancer (CRC) cases 11 . Our study showed that higher VFI as a metric of visceral obesity is associated with worse RFS and OS. Our findings are similar to that of Fleming et al., who showed that a high visceral to total fat ratio was associated with poor clinical and oncological outcomes in non-metastatic CRC 12 . Interestingly, patients with a high visceral to total fat ratio also had higher levels of IL-6 and tumor necrosis factor α, suggesting that the detrimental effects of obesity may be mediated by inflammation. It has been widely accepted that obesity leads to a chronic inflammatory state which in turn leads to immune dysfunction and the development of cancers 13 . We recently showed that obesity enhances PD-1 mediated T-cell dysfunction at least partly by leptin signaling 14 . This could potentially explain why patients with higher VFI in our study had worse outcomes. An important question in obesity-related research is how obesity is defined. Because of its ease of calculation, BMI has been traditionally used to define obesity. In a study, Sinicrope et al. showed that a BMI ≥ 35.0 kg/ m 2 statistically significantly reduced DFS compared with normal-weight patients in men. However, this adverse effect of obesity was not found in females 15 . However, previous reports have mentioned that visceral obesity may be a more important predictor of outcomes in CRC patients than BMI 16 . We did not see an association between BMI and either RFS or OS in our study. In contrast to Sinicrope et al. gender was not statistically significant in our multivariate model. As explained before, this could be due to BMI's inability to adjust for body composition, contributing to what is known as the "obesity paradox". Of note, unlike the findings by Fleming et al. we Figure 1. Kaplan Meier analysis of recurrence free and overall survival curves for VFI tertiles and BMI categories. Recurrence free and overall survival curves were generated for 492 patients categorized based on VFI into bottom (red), middle (blue), and top (green) teritles and BMI into < 25 kg/m 2 (red), 25-30 kg/m 2 (blue), and > 30 kg/m 2 (green) categories. Kaplan Meier survival curve analysis revealed that higher VFI had poorer RFS (log rank p = 0.01) and OS (p = 0.01), but not BMI. www.nature.com/scientificreports/ did not see any association between skeletal muscle content and recurrence or survival 12 . These could be due to differences in the study population, measurement of SMI, stage distribution, or other confounding factors.
Another interesting finding of this study was that the association between body composition and CRC recurrence is affected by metformin use. Consistent with our previous research in NSCLC, we showed that the detrimental effect of high VFI is only present in non-metformin users. Similarly, the protective effect of metformin was only present in the high VFI group 7 . The possible mechanisms may be AMPK (5′-AMP-activated protein kinase) mediated cell cycle arrest 17 , inhibition of reactive oxygen species generation by inhibiting the Electron Transport Chain 18 , promotion of apoptosis and autophagy of tumor cells 19 , and inhibition of leptin induced T cell exhaustion 20 and increased number of CD3+ and CD8+ tumor infiltrating lymphocytes 21 by metformin. The possible interaction between obesity, metformin, and survival in colorectal cancer could be the reversal of the metabolic and immune dysfunction by metformin within the tumor microenvironment in the high VFI group through the above mechanisms.
However, this interaction between visceral obesity, metformin use, and recurrence was not observed in overall survival. This discrepancy between RFS and OS was also observed by Sauer et al. when comparing preoperative and postoperative chemotherapy in rectal cancer patients 22 . Similarly, Andre et al. found that adjuvant chemotherapy improved only disease free survival and not OS 23 . Possible explanations for this discrepancy could be that the follow up period is small to detect a significant difference, and patient comorbidities and other cofounding variables which might affect OS but not RFS.
To our knowledge, this is the first study that established an association between visceral obesity, metformin use, and clinical and oncologic outcomes in early-stage CRC. However, our study has several limitations, including the retrospective design, relatively small sample size limiting the power of the study, lack of detailed information on treatment, patient co-morbidities, and molecular characteristics of tumors. Also, the dose and duration of metformin were not available. Nevertheless, our findings are hypothesis-generating and consistent with our Univariate and multivariate cox proportional models of VFI tertiles for overall and recurrence free survival. Univariate and multivariate cox proportional model survival curves were generated for 492 patients based on the bottom (red), middle (blue), and top (green) VFI tertiles. Multivariate analysis was performed with patient age, sex, tumor stage, and VFI as covariates for RFS. Patient age, sex, race, and VFI were used as covariates in the multivariate model for OS. Univariate (Wald p = 0.01) and multivariate (p = 0.005) analysis showed that a bottom VFI tertile was associated with a significantly better recurrence free survival. Univariate analysis showed that the bottom VFI tertile (p = 0.01) was associated with a significantly better OS. However, VFI lost its significance and was not included in the final multivariate model for OS.  Clinical data. All patients with Stage I/II colorectal adenocarcinoma undergoing surgery at our institute between 2004 and 2020 were included. Institution databases, cancer registries, and electronic health records were used to extract clinical data. Information on patient age, gender, race (Caucasian, African American, other), smoking status (current, former, and never), alcohol consumption (current, former, and never), height, BMI, metformin use, primary tumor site (caecum, colon, rectosigmoid, and rectum) and pathological T and N stage (as per the 7th or 8th editions of the staging manual of the American Joint Committee on Cancer) at the time of diagnosis, number of lymph nodes examined, received adjuvant therapy or not, tumor recurrence, site (locoregional, liver, lung, and other distant sites), and recurrence-free and overall survival data was extracted. Patients diagnosed with a primary tumor in the appendix and who developed a recurrence > 5 years after primary tumor diagnosis were excluded. Data was collapsed into white (Caucasians) and non-white (African American and others) for patient race, colon (caecum and colon), and rectum (rectosigmoid and rectum) as primary tumor sites for analysis.
CT imaging analysis. This work was performed by authors YRV and SD (Interobserver reliability statistics are explained in Supplementary Text S1). CT scans that were obtained for staging studies, preoperative workup, or surveillance postoperatively were obtained. A single axial cross-sectional image at the L3 vertebral level was identified for analysis. Patients whose entire body cross-sectional axial images within 5 years of primary tumor diagnosis could not be identified were excluded. To quantify visceral adiposity and muscle mass, sliceOmatic software (version 5.0 Rev-16c; Tomovision Software, Magog, Canada) with ABACS+ plugin (Rev-1.0.0; Voronoi Health Analytics, Vancouver, Canada) was used. The images are labeled according to their vertebral level www.nature.com/scientificreports/ (L3) prior to auto-segmentation of image pixels using the Vertebral Label tool. The ABACS+ plugin performs segmentation using preset Hounsfield Unit (HU) ranges for muscle, bone, visceral, subcutaneous, and intramuscular adipose tissues. After segmentation by ABACS+, the Region Growing mode and Tag Lock tool were used to make necessary manual corrections using the preset HU values in the Alberta L3 Manual Protocol available as an additional download. The Tag Surface/Volume tool was used to calculate and export the surface area measurements of muscle, bone, visceral, subcutaneous, and intramuscular adipose tissue. The actual cross-sectional visceral fat area (VFA) was calculated as the sum of visceral and intramuscular adipose tissue areas (True visceral = Visceral + Intramuscular). The total cross-sectional fat area (TFA) was calculated as the sum of true visceral and subcutaneous adipose tissue areas (Total fat area = True visceral + Subcutaneous). Visceral obesity was quantified as an index (visceral fat index, VFI), calculated as the ratio of VFA to TFA (VFA/TFA). The SMI (skeletal muscle index) was calculated after adjusting the L3 cross-sectional skeletal muscle area to the square of the patient's height (muscle area [cm 2 ]/height 2 [m]) as a measure of the total body muscle mass.
Statistical analysis. Since VFI has no established cut-off values, analysis was performed using it as both a continuous and categorical variable as tertiles. Kruskal Wallis rank-sum test and Pearson's chi-square test were used for group comparisons of continuous and categorical variables, respectively. Survival analysis was performed using Kaplan Meier survival curve analysis and Cox proportional modeling. For multivariate Cox proportional modeling, covariates with a p value < 0.2 on univariate cox modeling were used in addition to BMI, SMI, or VFI as covariates. A Backward LR stepwise method with p < 0.1 as tolerance at each step was used to develop a final model. To understand if an interaction between metformin and obesity exists, as shown by us in lung cancer previously, an interaction analysis followed by a subgroup analysis was performed based on VFI tertiles and metformin use 8 . An alpha error of 0.05 was used to assess statistical significance. SPSS (version 26, IBM Software, Armonk, NY) and Prism (version 9.3.1 for Windows OS, GraphPad Software, San Diego, CA) were used for graphing and analysis.
Other. This study is reported as per Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for cohort studies (Supplementary Table S4).

Ethical statement. This retrospective study was approved by the institutional review board of Roswell Park
Comprehensive Cancer Center (RPCCC), Buffalo, New York, USA.

Data availability
Raw data generated in this study is available from the authors upon request.