Energy Balance–Related Factors and Direct Tumor–Adipocyte Contact in Colorectal Cancer: Etiologic Insights from the Population-Based Netherlands Cohort Study

Abstract

Background:

Stroma AReactive Invasion Front Areas (SARIFA) are a histopathologic biomarker characterized by a direct tumor–adipocyte contact at the tumor invasion front. Although the tumor-promoting effect of tumor–adipocyte interactions has often been linked to obesity, it is unclear whether prediagnostic energy balance–related factors are differentially associated with the risk of SARIFA-positive versus SARIFA-negative colorectal cancers.

Methods:

We analyzed data from the Netherlands Cohort Study, including 120,852 participants and 1,826 incident colorectal cancer cases with known SARIFA status. Multivariable-adjusted Cox regression was used to estimate associations between baseline prediagnostic body mass index (BMI), lower-body clothing size as a proxy for waist circumference, and nonoccupational physical activity and colon and rectal cancer risk by SARIFA status.

Results:

In men, BMI was more strongly associated with the risk of SARIFA-positive colon cancer [hazard ratio (HR) per 5 kg/m², 1.57; 95% confidence interval (CI), 1.20–2.07] than SARIFA-negative tumors (HR, 1.22; 95% CI, 1.01–1.48), with similar patterns for lower-body clothing size. In women, weaker associations were observed, but estimates were again higher for SARIFA-positive tumors than SARIFA-negative tumors. Physical activity was inversely associated with SARIFA-positive colon cancer in women (HR per 30 minutes/day, 0.92; 95% CI, 0.85–1.01), but not SARIFA-negative tumors, with no clear SARIFA-specific patterns in men or for rectal cancer.

Conclusions:

Findings suggest a stronger role of body fatness in the development of SARIFA-positive than SARIFA-negative colon cancer and provide novel evidence warranting confirmation in other molecular pathologic epidemiology studies.

Impact:

This study extends previous epidemiologic findings by suggesting that adiposity may influence colon cancer development, resulting in a specific tumor–stroma interaction phenotype.

Introduction

The tumor microenvironment (TME), especially at the invasion front where the tumor cells interact with the host, plays an important role in colorectal cancer progression, and different biomarkers assessing specific components of the TME have been shown to possess prognostic and potential predictive value (1–3). Most studies to date focus on the interaction between tumor and immune cells neglecting other biologically relevant components of the TME, such as endothelial cells, fibroblasts, or adipocytes. The latter are especially interesting as there is increasing evidence that lipid metabolism is a key driver of tumor growth, immune evasion, and therapy resistance (4–7). Metabolic reprogramming is also considered a hallmark of cancer (8).

We established Stroma AReactive Invasion Front Areas (SARIFA) as a novel hematoxylin and eosin (H&E)–based histopathologic biomarker, defined by the direct contact between tumor cells and adipocytes without intervening stromal or inflammatory cells. We and others have shown that SARIFA positivity is strongly associated with poor patient outcomes across multiple different tumor types (9), including colorectal cancer (10–17). SARIFA-positive tumors are characterized by an immune dysregulation, stromal changes, and an upregulation of lipid metabolism (11, 12, 18–20).

Many experimental studies have shown that changes in lipid metabolism play an important role in cancer. This is supported by the well-known link between obesity and cancer growth and progression (21). Our histologic biomarker, SARIFA, may capture or reflect these underlying biological mechanisms. However, in our previous SARIFA studies, we did not find a consistent association between SARIFA positivity and body mass index (BMI)/obesity at the time of cancer diagnosis (9). BMI or weight measured at the time of cancer diagnosis or surgery may not represent BMI or weight in the years before the cancer diagnosis well, as tumors may induce cachexia, i.e., unintentional weight loss and muscle wasting through chronic systemic inflammation. To truly assess whether obesity increases the risk of developing SARIFA-positive cancers, a prospective investigation is needed.

As outlined, we previously assessed SARIFA in relation to BMI measured at the time of cancer diagnosis, which only allowed us to evaluate cross-sectional associations. However, such analyses cannot address whether BMI or other energy balance–related factors likely play a causal role in the development of SARIFA-positive tumors. In the current study, we utilized the prospective design of the Netherlands Cohort Study (NLCS), which provides detailed baseline data on diet and lifestyle, as well as long-term follow-up for incident cancer, including colorectal cancer subtyping. In our earlier NLCS analyses, we identified an association between energy balance–related factors such as BMI and immunohistochemically defined Warburg subtypes, suggesting an interplay between energy balance–related factors and tumor biology (22, 23).

Our current study design enabled us now to investigate whether energy balance–related characteristics at baseline, including BMI, lower-body clothing size as a proxy for waist circumference, and nonoccupational physical activity, differentially influence the risk of developing colorectal cancer according to SARIFA status (SARIFA positive vs. SARIFA negative).

Materials and Methods

Ethical approval

Ethical approval for the NLCS was obtained from the Medical Ethical Committee of Maastricht University Medical Center+ (Maastricht, Netherlands). Written informed consent was obtained from all cohort participants through completion and return of the baseline questionnaire. The NLCS was conducted in accordance with the ethical principles of the Declaration of Helsinki and was approved by the institutional review boards of Maastricht University (Maastricht, Netherlands) and the Netherlands Organization for Applied Scientific Research.

Study design and population

The NLCS was initiated in 1986 and enrolled 120,852 individuals 55 to 69 years of age at baseline (24). Participants completed a mailed, self-administered questionnaire, covering diet, smoking, anthropometry, medical history, physical activity, and other cancer risk factors (24).

The case–cohort design was used for data processing and analysis for reasons of efficiency (25). A random subcohort of 5,000 participants was selected from the full cohort immediately after baseline and was used to estimate the accumulated person-years in the cohort. These subcohort members were actively followed up biennially for vital status information by linking to municipal population registries (24). Only one participant in the subcohort was lost to follow-up after 20.3 years.

Follow-up for cancer incidence was established by annual record linkage with the Netherlands Cancer Registry and Pathologisch Anatomisch Landelijk Geautomatiseerd Archief (PALGA), the nationwide Dutch Pathology Registry (26), covering 20.3 years of follow-up (September 17, 1986, to January 1, 2007). The completeness of cancer follow-up was estimated to be more than 96% (27). After excluding individuals with a cancer history at baseline (except skin cancer), 4,597 incident colorectal cancer cases and 4,774 subcohort members remained for analysis (Fig. 1).

Figure 1.

Flow diagram showing the number of colorectal cancer (CRC) cases and subcohort members. Stepwise inclusion and exclusion process used to derive the final analytic sample of colorectal cancer cases and subcohort members from the NLCS (1986–2006). FFPE, formalin-fixed, paraffin-embedded; NA, not applicable; PALGA, Dutch Pathology Registry; pan-CK, pan-cytokeratin; QC, quality control; TMA, tissue microarray.

Tissue collection and SARIFA status

We previously assessed the SARIFA status (positive vs. negative vs. unknown) for the NLCS colorectal cancer resection specimens as described in a prior publication (13) and in line with previous work on SARIFA in colorectal cancer (9). SARIFA status was established on digitized H&E-stained whole-slide images accessed by using QuPath (https://qupath.github.io/; ref. 28). We and others have previously demonstrated that interobserver variability for assessing SARIFA status in colorectal cancer is low, with κ values ranging from 0.77 to 0.92 (9, 14, 15, 29). SARIFA positivity is defined by a direct tumor–adipocyte contact of at least one tumor gland or at least a group of five or more tumor cells at the invasion front. The presence of one such area is sufficient to categorize the whole case as SARIFA positive. Otherwise, the case is classified as negative. Please refer to Fig. 2 for examples of SARIFA-positive and SARIFA-negative colorectal cancers within the NLCS. For the current study, SARIFA status was available for 1,826 colorectal cancer cases (Fig. 1).

Figure 2.

SARIFA status and prediagnostic energy balance–related factors in colorectal cancer. To assess the relationship between prediagnostic energy balance–related factors, including BMI, clothing size, and physical activity, and the invasion front biomarker SARIFA, we conducted a comprehensive statistical analysis using data from the NLCS, which involved 120,852 participants, including 1,826 colorectal cancer (CRC) cases with available histopathologic SARIFA status. SARIFA positivity is defined by a direct tumor–adipocyte interaction at the invasion front, characterized by the absence of intervening stromal or inflammatory infiltrate, and can be assessed on routinely available H&E-stained slides. In previous studies, we linked SARIFA positivity to upregulated lipid metabolism both at the bulk and spatial transcriptomic level. Moreover, a growing body of research has implicated adipocyte-induced lipid metabolism reprogramming in promoting tumor aggressiveness, particularly in the context of obesity. In this study, we explored whether prediagnostic energy balance–related factors such as BMI, clothing size, and physical activity in particular increase the risk of developing SARIFA-positive colorectal cancers, suggesting that an altered metabolic environment may predispose tumors to adopt this aggressive phenotype. We showed that a higher prediagnostic BMI increased the risk of SARIFA-positive colon cancer. Although the underlying biological mechanism for this association remains unclear, we hypothesize that immune dysregulation induced by adiposity may play a role, potentially influencing tumor growth patterns and the TME. SARIFA positivity has previously been linked to immune dysregulation, which could contribute to stromal remodeling and a more infiltrative growth pattern, though this remains a speculative explanation pending further investigation. Created in BioRender.com Reitsam, N. (2026), https://BioRender.com/19ylm1z).

Energy balance–related factors

At baseline, all NLCS participants completed a mailed, self-administered questionnaire covering anthropometry, diet, physical activity, and other cancer risk factors (24). BMI (in kg/m²) was calculated from self-reported height and weight. Participants also reported their lower-body clothing size (trousers or skirt) based on Dutch sizing labels, which has previously been validated as a reliable proxy for waist circumference in cancer risk prediction within the NLCS (30). To estimate nonoccupational physical activity levels, participants were asked to report the average daily time spent on activities such as walking, cycling, and sports, as described in more detail previously (31).

Cox regression models

After excluding participants with incomplete or inconsistent questionnaires or missing data on exposure variables or confounders, 1,512 colorectal cancer cases and 3,911 subcohort members remained for analyses (Fig. 1). Associations between energy balance–related factors and colorectal cancer risk were examined stratified by sex (men or women), tumor location (colon or rectum), and SARIFA status (positive or negative).

Cox proportional hazards (PH) models were used to estimate hazard ratios (HR) and 95% confidence intervals (CI) for the associations between colon and rectal cancers and BMI (according to sex-specific quartiles and per 5 kg/m² increase), clothing size (according to sex-specific quartiles and per two-size increase), and nonoccupational physical activity (categorized as <30, 30–60, 60–90, or >90 minutes per day and per 30-minute/day increase). Standard errors of the HRs were calculated using the Huber–White sandwich estimator to account for additional variance introduced due to subcohort sampling (32). The PH assumption was assessed using the scaled Schoenfeld residuals (33) and by introducing time–covariate interactions into the models.

All multivariable models were adjusted for age (continuous; years), total energy intake (continuous; kcal/day), family history of colorectal cancer (yes or no), and alcohol consumption (0, 0.1–4, 5–14, or >15 g/day). Age was introduced as a time-varying covariate. Models for BMI and clothing size were additionally adjusted for nonoccupational physical activity (minutes/day), and models for BMI were additionally adjusted for height (continuous; cm). Physical activity models were further adjusted for BMI (sex-specific quartiles). Potential additional confounders were smoking status (never, former, or current), level of education (primary/lower vocational education, secondary/medium vocational education, or higher vocational education/university), processed meat consumption (continuous; g/day), and red meat consumption (continuous; g/day). These potential confounders were included in the final models only if they changed the HRs by ≥ 10% using a backward elimination procedure (34).

We assessed whether the associations between energy balance–related factors and colon and rectal cancer risk differed by SARIFA status (SARIFA-positive vs. SARIFA-negative tumors). This heterogeneity was evaluated using a modified competing risk approach in Stata, adapted for the case–cohort design, as previously described (35, 36). All analyses were conducted in Stata Statistical Software: Release 16 (StataCorp). A P value <0.05 for two-sided testing was considered statistically significant, with no multiple testing correction applied because of the hypothesis-driven focus.

Results

Baseline characteristics

Baseline lifestyle characteristics of subcohort members and colorectal cancer cases, overall and by SARIFA status, are presented in Table 1. Among men, overweight and obesity were more common in colon cancer cases compared with subcohort members, particularly among SARIFA-positive cases (57.2% vs. 46.6%). A similar pattern was seen for rectal cancer in men and colon cancer in women, although less pronounced. By contrast, in women, SARIFA-positive rectal cancer cases were less often overweight or obese than subcohort members (38.9% vs. 43.6%), whereas SARIFA-negative cases were most often overweight or obese (52.9%). The mean lower-body clothing size was similar across subcohort members and cases, regardless of tumor location and SARIFA status.

Table 1.

Baseline characteristics [mean (SD) or %] of subcohort members and colorectal cancer cases according to SARIFA status (positive or negative), by sex and tumor location; NLCS, 1986 to 2006.

Characteristic	Subcohort	Colon cancer			Rectal cancer
		Total	SARIFA negative	SARIFA positive	Total	SARIFA negative	SARIFA positive
Men	​	​	​	​	​	​	​
N	1,971	582	388	194	181	150	31
Overweight/obesitya (%)	46.6	51.9	49.2	57.2	49.7	49.3	51.6
Clothing sizeb	51.7 (2.7)	52.3 (2.6)	52.3 (2.7)	52.3 (2.5)	51.8 (2.5)	51.7 (2.3)	52.1 (3)
Nonoccupational physical activity >60 minutes/day (%)	51.1	51.6	50	54.6	60.8	61.3	58.1
Age (years)	61.3 (4.2)	61.6 (4.2)	61.6 (4)	61.5 (4.5)	61 (3.9)	60.9 (3.9)	61 (4.2)
Total energy intake (kcal/day)	2,164 (500)	2,115 (467)	2,104 (456)	2,137 (488)	2,252 (493)	2,241 (490)	2,306 (510)
Family history of colorectal cancer (%)	5.4	11.2	10.8	11.9	11.1	10.7	12.9
Alcohol consumption (g/day)	15.1 (17.1)	15.4 (16)	15.4 (16)	15.4 (15.8)	16.8 (17.3)	17.6 (17.7)	12.9 (14.4)
Processed meat intake (g/day)	15.9 (16.9)	15.1 (15.1)	15.2 (16)	15.1 (13.1)	17.6 (16.7)	17.9 (17)	16.1 (15.4)
Red meat intake (g/day)	93.8 (41.2)	93.5 (40.5)	91.9 (39.6)	96.7 (42)	92.2 (40.4)	92.6 (39.5)	89.9 (45.3)
Never cigarette smokers (%)	12.7	11.9	10.6	14.4	8.3	6	19.4
University or higher vocational educationb (%)	19.8	24.3	24.8	23.3	17.9	16.2	25.8
Women	​	​	​	​	​	​	​
N	1,940	485	318	167	105	87	18
Overweight/obesitya	43.6	46.8	46.5	47.3	50.5	52.9	38.9
Clothing sizeb	43.4 (2.9)	43.6 (3.5)	43.4 (3)	44.1 (4.2)	43.6 (2.7)	43.8 (2.6)	42.6 (2.9)
Nonoccupational physical activity >60 minutes/day (%)	44.9	38.6	37.4	40.7	44.8	47.1	33.3
Age (years)	61.4 (4.3)	62 (4.1)	62.1 (4)	61.8 (4.3)	61.5 (4.2)	61.2 (4.3)	62.8 (3.9)
Total energy intake (kcal/day)	1,684 (392)	1,679 (388)	1,680 (406)	1,678 (351)	1,691 (345)	1,679 (307)	1,750 (496)
Family history of colorectal cancer (%)	6	9.5	10.4	7.8	10.5	9.2	16.7
Alcohol consumption (g/day)	6 (9.5)	5.9 (9.8)	5.8 (9.9)	6.1 (9.7)	6.1 (9.8)	6.2 (10.3)	5.5 (7.4)
Processed meat intake (g/day)	10.3 (11.6)	10.1 (11.1)	10.3 (11.5)	9.8 (10.2)	11.8 (10.5)	11.8 (9.9)	11.8 (13.5)
Red meat intake (g/day)	81 (38.1)	77 (34.4)	76 (34.3)	78.9 (34.8)	86 (41.8)	86.7 (43.4)	82.4 (33.4)
Never cigarette smokers (%)	57.3	59.4	59.1	59.9	57.1	58.6	50
University or higher vocational educationb (%)	9.5	9.3	10.1	7.8	5.7	6.9	—

Abbreviation: SD, standard deviation.

^aBMI ≥ 25 kg/m².

^bBased on fewer participants due to extra missings.

Among men, high nonoccupational physical activity levels (>60 minutes/day) were reported more often by SARIFA-positive colon cancer cases than by SARIFA-negative colon cancer cases or subcohort members (54.6% vs. 50% and 51.1%). In male rectal cancer cases, high nonoccupational physical activity levels were also more common than in subcohort members, although the difference between SARIFA-positive and SARIFA-negative cases was smaller (58.1% and 61.3%). In women, subcohort members generally reported high nonoccupational physical activity levels more often than cases. This difference was most pronounced for SARIFA-positive rectal cancer cases compared with subcohort members (33.3% vs. 44.9%). By contrast, SARIFA-negative rectal cancer cases reported high nonoccupational physical activity slightly more often than subcohort members (47.1%).

Comparison of other baseline characteristics revealed further differences between subcohort members and cases, as well as by SARIFA status and sex. Compared with subcohort members, cases more often reported a family history of cancer, though percentages varied by SARIFA status. Smoking status and education also differed by SARIFA status and sex. These descriptive findings may further indicate differing etiologies for SARIFA-positive and SARIFA-negative tumors, potentially influenced by sex.

Cox regression analyses

Tables 2, 3, and 4 present HRs and 95% CIs from multivariable-adjusted Cox regression models for colorectal cancer risk according to SARIFA status, examining energy balance–related factors and stratified by sex and tumor location. Age-adjusted results are provided in Supplementary Tables S1–S3 and were comparable with the multivariable-adjusted results. Age was included as a time-varying covariate in all models because of a violation of the PH assumption.

Table 2.

Multivariable-adjusted HRs^a and 95% CIs for associations between BMI and SARIFA status in colorectal cancer, by sex and tumor location; NLCS, 1986 to 2006.

	Medianb	Person-years at risk	Total		SARIFA negative		SARIFA positive		P heterogeneity
			n cases	HR (95% CI)	n cases	HR (95% CI)	n cases	HR (95% CI)
BMI quartiles (kg/m2)
Men: colon
<23.4	22.2	7,993	125	1 (ref)	84	1 (ref)	41	1 (ref)	​
23.4–24.9	24.2	8,343	161	1.21 (0.92–1.60)	120	1.35 (0.98–1.86)	41	0.92 (0.58–1.46)	​
25–26.6	25.7	7,683	154	1.23 (0.92–1.64)	102	1.22 (0.87–1.71)	52	1.25 (0.80–1.96)	​
>26.6	27.8	7,003	166	1.58 (1.19–2.10)	103	1.49 (1.07–2.08)	63	1.75 (1.14–2.69)	0.921
P trend	​	​	​	0.002	​	0.049	​	0.004	​
Continuous per 5 kg/m2	​	31,022	606	1.33 (1.13–1.57)	409	1.22 (1.01–1.48)	197	1.57 (1.20–2.07)	0.410
Men: rectum
<23.4	22.2	7,993	49	1 (ref)	41	1 (ref)	1	1 (ref)	​
23.4–24.9	24.2	8,343	45	0.85 (0.55–1.31)	38	0.84 (0.52–1.34)	7	0.90 (0.32–2.53)	​
25–26.6	25.7	7,683	56	1.19 (0.78–1.82)	48	1.18 (0.75–1.87)	8	1.19 (0.44–3.25)	​
>26.6	27.8	7,003	39	0.93 (0.59–1.48)	30	0.82 (0.49–1.36)	9	1.67 (0.64–4.37)	0.010
P trend	​	​	​	0.795	​	0.846	​	0.275	​
Continuous per 5 kg/m2	​	31,022	189	1.05 (0.82–1.35)	157	1.03 (0.78–1.35)	32	1.19 (0.67–2.12)	0.119
Women: colon
<22.8	21.5	9,014	135	1 (ref)	95	1 (ref)	40	1 (ref)	​
22.8–24.7	23.8	8,914	114	0.85 (0.64–1.13)	72	0.76 (0.54–1.06)	42	1.07 (0.68–1.68)	​
24.8–27	25.7	8,141	130	1.09 (0.82–1.44)	84	1 (0.72–1.39)	46	1.31 (0.84–2.04)	​
>27.1	29.2	8,158	128	1.10 (0.92–1.46)	82	1 (0.71–1.41)	46	1.32 (0.84–2.08)	0.273
P trend	​	​	​	0.288	​	0.685	​	0.155	​
Per 5 kg/m2	​	34,228	507	1.11 (0.97–1.27)	333	1.07 (0.91–1.26)	174	1.19 (0.98–1.45)	0.490
Women: rectum
<22.8	21.5	9,014	31	1 (ref)	23	1 (ref)	8	1 (ref)	​
22.8–24.7	23.8	8,914	20	0.63 (0.35–1.14)	18	0.78 (0.41–1.47)	2	0.23 (0.05–1.10)	​
24.8–27	25.7	8,141	26	0.88 (0.51–1.54)	22	1 (0.54–1.85)	4	0.54 (0.15–1.98)	​
>27.1	29.2	8,158	31	1.07 (0.63–1.83)	27	1.27 (0.71–2.26)	4	0.51 (0.12–2.13)	0.003
P trend	​	​	​	0.613	​	0.334	​	0.457	​
Per 5 kg/m2	​	34,228	108	1.12 (0.88–1.43)	90	1.19 (0.92–1.53)	18	0.83 (0.42–1.65)	0.481

^aHRs were adjusted for age (years; continuous), nonoccupational physical activity (minutes/day), height (cm; continuous), total energy intake (kcal/day; continuous), family history of colorectal cancer (yes or no), alcohol consumption (0, 0.1–4, 5–14, or >15 g/day), processed meat intake (g/day; continuous), and red meat intake (g/day; continuous). Age was included as a time-varying covariate.

^bMedian BMI per quartile based on the subcohort.

Table 3.

Multivariable-adjusted HRs^a and 95% CIs for associations between lower-body clothing size and SARIFA status in colorectal cancer, by sex and tumor location; NLCS, 1986 to 2006.

​	Medianb	Person-years at risk	Total		SARIFA negative		SARIFA positive		P heterogeneity
			n cases	HR (95% CI)	n cases	HR (95% CI)	n cases	HR (95% CI)
Clothing size
Men: colon
≤50	50	10,903	159	1 (ref)	110	1 (ref)	49	1 (ref)	​
52	52	9,750	206	1.44 (1.13–1.85)	135	1.37 (1.03–1.83)	71	1.60 (1.08–2.37)	​
54	54	5,156	105	1.39 (1.03–1.87)	69	1.32 (0.93–1.87)	36	1.55 (0.97–2.46)	​
≥56	56	2,619	80	2.13 (1.52–2.97)	56	2.17 (1.48–3.18)	24	2.03 (1.19–3.45)	0.709
P trend	​	​	​	<0.001	​	<0.001	​	0.006	​
Continuous per two sizes	​	28,428	550	1.42 (1.22–1.65)	370	1.43 (1.20–1.70)	180	1.40 (1.13–1.74)	0.305
Men: rectum
≤50	50	10,903	66	1 (ref)	57	1 (ref)	9	1 (ref)	​
52	52	9,750	59	1 (0.68–1.45)	51	0.99 (0.66–1.49)	8	1.01 (0.39–2.62)	​
54	54	5,156	37	1.24 (0.80–1.92)	30	1.15 (0.71–1.86)	7	1.80 (0.68–4.77)	​
≥56	56	2,619	14	0.93 (0.51–1.71)	9	0.69 (0.33–1.44)	5	2.47 (0.83–7.36)	<0.001
P trend	​	​	​	0.698	​	0.727	​	0.081	​
Continuous per two sizes	​	28,428	176	1 (0.81–1.24)	147	0.94 (0.75–1.18)	29	1.38 (0.82–2.33)	0.151
Women: colon
≤40	40	6,574	90	1 (ref)	62	1 (ref)	28	1 (ref)	​
42	42	8,582	131	1.09 (0.80–1.49)	82	0.99 (0.69–1.43)	49	1.33 (0.82–2.18)	​
44	44	9,270	133	0.99 (0.73–1.34)	91	0.96 (0.67–1.38)	42	1.03 (0.62–1.71)	​
≥46	46	9,454	143	1.05 (0.77–1.43)	90	0.95 (0.66–1.37)	53	1.29 (0.80–2.09)	0.933
P trend	​	​	​	0.966	​	0.739	​	0.555	​
Continuous per two sizes	​	33,880	497	1.12 (0.96–1.30)	325	1.03 (0.88–1.21)	172	1.30 (0.99–1.72)	0.415
Women: rectum
≤40	40	6,574	19	1 (ref)	13	1 (ref)	6	1 (ref)	​
42	42	8,582	25	0.96 (0.51–1.79)	21	1.20 (0.59–2.44	4	0.44 (0.12–1.67)	​
44	44	9,270	29	1.01 (0.55–1.84)	25	1.30 (0.65–2.59)	4	0.40 (0.11–1.48)	​
≥46	46	9,454	34	1.12 (0.62–2.03)	30	1.47 (0.75–2.89)	4	0.38 (0.11–1.39)	0.121
P trend	​	​	​	0.632	​	0.248	​	0.192	​
Continuous per two sizes	​	33,880	107	1.04 (0.82–1.31)	89	1.13 (0.89–1.44)	18	0.66 (0.34–1.29)	0.338

^aHRs were adjusted for age (years; continuous), nonoccupational physical activity (minutes/day; continuous), total energy intake (kcal/day; continuous), family history of colorectal cancer (yes or no), alcohol consumption (0, 0.1–4, 5–14, or >15 g/day), processed meat intake (g/day; continuous), and red meat intake (g/day; continuous). Age was included as a time-varying covariate.

^bMedian clothing size per category based on the subcohort.

Table 4.

Multivariable-adjusted HRs^a and 95% CIs for associations between nonoccupational physical activity and SARIFA status in colorectal cancer, by sex and tumor location; NLCS, 1986 to 2006.

​	Medianb	Person-years at risk	Total		SARIFA negative		SARIFA positive		P heterogeneity
			n cases	HR (95% CI)	n cases	HR (95% CI)	n cases	HR (95% CI)
Nonoccupational physical activity (minutes/day)
Men: colon
≤30	21.4	4,997	105	1.14 (0.86–1.51)	72	1.12 (0.81–1.55)	33	1.18 (0.75–1.87)	​
31–60	42.9	10,100	193	1 (ref)	137	1 (ref)	56	1 (ref)	​
61–90	73.6	6,001	129	1.18 (0.90–1.54)	88	1.13 (0.83–1.55)	41	1.28 (0.83–1.98)	​
>90	130.0	9,925	179	0.96 (0.75–1.23)	112	0.84 (0.63–1.12)	67	1.25 (0.85–1.93)	0.966
P trend	​	​	​	0.455	​	0.137	​	0.446	​
Continuous per 30 minutes/day	​	31,022	606	0.99 (0.94–1.03)	409	0.97 (0.92–1.03)	197	1.02 (0.96–1.08)	0.484
Men: rectum
≤30	21.4	4,997	15	0.53 (0.29–0.97)	13	0.58 (0.30–1.10)	2	0.36 (0.08–1.66)	​
31–60	42.9	10,100	60	1 (ref)	48	1 (ref)	12	1 (ref)	​
61–90	73.6	6,001	49	1.41 (0.94–2.11)	45	1.63 (1.05–2.51)	4	0.56 (0.18–1.78)	​
>90	130.0	9,925	65	1.13 (0.77–1.64)	52	1.11 (0.73–1.68)	14	1.20 (0.55–2.61)	<0.001
P trend	​	​	​	0.014	​	0.033	​	0.177	​
Continuous per 30 minutes/day	​	31,022	189	1.04 (0.98–1.10)	157	1.05 (0.99–1.12)	32	0.98 (0.89–1.07)	0.101
Women: colon
≤30	19.3	7,756	143	1.21 (0.93–1.58)	95	1.19 (0.87–1.62)	48	1.27 (0.84–1.92)	​
31–60	42.9	10,923	165	1 (ref)	112	1 (ref)	53	1 (ref)	​
61–90	75.0	8,000	111	0.92 (0.70–1.21)	66	0.80 (0.58–1.12)	45	1.15 (0.75–1.76)	​
>90	115.7	7,550	88	0.78 (0.59–1.05)	60	0.79 (0.56–1.11)	28	0.77 (0.48–1.25)	0.710
P trend	​	​	​	0.004	​	0.010	​	0.097	​
Continuous per 30 minutes/day	​	34,228	507	0.93 (0.87–1.00)	333	0.92 (0.85–1.01)	174	0.94 (0.84–1.06)	0.905
Women: rectum
≤30	19.3	7,756	26	1.04 (0.61–1.79)	21	1.04 (0.57–1.91)	5	1.03 (0.31–3.37)	​
31–60	42.9	10,923	34	1 (ref)	27	1 (ref)	7	1 (ref)	​
61–90	75.0	8,000	29	1.16 (0.69–1.94)	25	1.24 (0.71–2.18)	4	0.81 (0.23–2.84)	​
>90	115.7	7,550	19	0.79 (0.44–1.42)	17	0.87 (0.46–1.64)	2	0.44 (0.09–2.22)	<0.001
P trend	​	​	​	0.507	​	0.807	​	0.262	​
Continuous per 30 minutes/day	​	34,228	108	1 (0.88–1.15)	90	1 (0.87–1.14)	18	1.04 (0.68–1.57)	0.857

^aHRs were adjusted for age (years; continuous), BMI (kg/m²; quartiles), total energy intake (kcal/day; continuous), family history of colorectal cancer (yes or no), alcohol consumption (0, 0.1–4, 5–14, or >15 g/day), processed meat intake (g/day; continuous), and red meat intake (g/day; continuous). Age was included as a time-varying covariate.

^bMedian daily minutes of physical activity per category based on the subcohort.

Adiposity

Among men, a higher BMI was statistically significantly positively associated with total colon cancer risk (Table 2), with an HR of 1.33 (95% CI, 1.13–1.57) per 5 kg/m² increase and a statistically significant positive trend across sex-specific quartiles (P trend = 0.002). The association with BMI was stronger for SARIFA-positive colon cancers (${\text{HR}}_{5\mathrm{\hspace{0.17em}}\text{kg}\mathrm{/}{\mathrm{m}}^2}$, 1.57; 95% CI, 1.20–2.07; P trend_quartiles = 0.004) than for SARIFA-negative colon cancers (${\text{HR}}_{5\mathrm{\hspace{0.17em}}\text{kg}\mathrm{/}{\mathrm{m}}^2}$, 1.22; 95% CI, 1.01–1.48; P trend_quartiles = 0.049), although the test for heterogeneity was not statistically significant. Lower-body clothing size showed a positive association with total colon cancer risk in men (Table 3), with an HR of 1.42 (95% CI, 1.22–1.65) per two-size increment and a statistically significant positive trend across categories (P trend < 0.001), with results similar after stratification on SARIFA status and the test for heterogeneity not statistically significant. For rectal cancer risk in men, whether overall or stratified by SARIFA status, no statistically significant associations were observed with either BMI or lower-body clothing size, although tests for heterogeneity were statistically significant for categorical modeling of BMI and lower-body clothing size, likely owing to small numbers in categories, resulting in more extreme HRs with wide CIs for some category comparisons (Table 2 and 3).

Among women, higher BMI and lower-body clothing size showed a weak positive association with total colon cancer risk (Table 2 and 3), with an HR of 1.11 (95% CI, 0.97–1.27) per 5 kg/m² increase and an HR of 1.12 (95% CI, 0.96–1.30) per two-size increment, though these associations did not reach statistical significance and neither a significant association nor a significant trend was observed when categorically modeling BMI and clothing size. Stronger associations were observed with SARIFA-positive colon cancers (${\text{HR}}_{\text{per}\mathrm{\hspace{0.17em}}5\mathrm{\hspace{0.17em}}\text{kg}\mathrm{/}{\mathrm{m}}^2}$, 1.19; 95% CI, 0.98–1.45 and HR_{per two sizes}, 1.30; 95% CI, 0.99–1.72) than for SARIFA-negative colon cancers (${\text{HR}}_{\text{per}\mathrm{\hspace{0.17em}}5\mathrm{\hspace{0.17em}}\text{kg}\mathrm{/}{\mathrm{m}}^2}$, 1.07; 95% CI, 0.91–1.26 and HR_{per two sizes}, 1.03; 95% CI, 0.88–1.21), though these associations and tests for heterogeneity were also not statistically significant. For rectal cancer risk in women, no significant associations were observed with either BMI or lower-body clothing size, although the test for heterogeneity reached statistical significance when modeling BMI categorically, again likely due to small case numbers in categories (Table 2 and 3).

Nonoccupational physical activity

Among men, nonoccupational physical activity was not statistically significantly associated with the overall risk of colon cancer (Table 4). Stratification by SARIFA status did not reveal any SARIFA-dependent differences, and the test for heterogeneity was not statistically significant. A nonsignificant, weak positive association was observed with rectal cancer risk (Table 4), with an HR of 1.04 (95% CI, 0.98–1.10) per 30 minutes/day increment and a significant positive trend across categories of nonoccupational physical activity (P trend = 0.014). Similar results were observed for SARIFA-positive tumors, but not SARIFA-negative tumors, with the test for heterogeneity reaching statistical significance in categorical models, but case numbers were small within categories.

Among women, nonoccupational physical activity was associated with a decreased overall colon cancer risk (Table 4), with a borderline statistically significant HR of 0.93 (95% CI, 0.87–1) per 30 minutes/day increment and a significant inverse trend across categories (P trend = 0.004). Stratification by SARIFA status revealed similar associations (HR_{per 30 minutes/day}, 0.92; 95% CI, 0.85–1.01 and HR_{per 30 minutes/day}, 0.94; 95% CI, 0.84–1.06), although a significant trend across nonoccupational physical activity categories was observed only for SARIFA-positive colon cancer risk (P trend = 0.010) and not SARIFA-negative colon cancer risk (P trend = 0.097). The test for heterogeneity was not statistically significant. For rectal cancer risk in women, no statistically significant associations were observed with nonoccupational physical activity (Table 4). Stratification by SARIFA status did not yield additional insights, although the test for heterogeneity reached statistical significance.

Discussion

We present here the first molecular pathologic epidemiology (MPE) study to explore etiologic heterogeneity in colorectal cancer risk by SARIFA status, a novel histopathologic marker reflecting direct tumor–adipocyte contact at the invasion front.

Measures of adiposity (BMI and clothing size) were associated with colon cancer risk in both men and women, with indications of a stronger association for SARIFA-positive than for SARIFA-negative tumors. Tests for heterogeneity were not statistically significant, suggesting no difference in association strength by SARIFA status; however, this test is inherently conservative. The observed inverse association between nonoccupational physical activity and colon cancer risk in women, which also seemed stronger for SARIFA-positive than SARIFA-negative tumors, further supports a potential adiposity- or energy balance–driven mechanism in SARIFA tumor development. The finding of an increased rectal cancer risk associated with higher levels of nonoccupational physical activity in men may be explained by the fact that nonoccupational physical activity likely does not accurately reflect lifetime physical activity in men in the NLCS, as extensively discussed in one of our previous publications (31).

The hypothesis underlying the current study was that SARIFA status may represent a histologic footprint of the effects of adiposity on the tumor–host interface and that for this reason stronger associations between adiposity and the risk of developing a SARIFA-positive colorectal tumor were to be expected than the risk of developing a SARIFA-negative colorectal tumor. Emerging experimental models show that adipocytes can undergo functional reprogramming, contributing to a malignant TME, e.g., via YAP/TAZ-driven dedifferentiation and inflammatory signaling (35) or via cancer cell–induced transformation into cancer-associated adipocytes that secrete pro-inflammatory cytokines such as IL6 and promote tumor invasiveness (36). Epidemiologic studies have also suggested a broader spectrum of obesity-associated cancers than previously recognized, with adiposity influencing risk across multiple tumor sites and subtypes (37). Nevertheless, it is currently not fully understood whether and how adiposity influences the morphology and growth patterns of cancers.

The findings of the current study support a link between adiposity and colorectal tumor morphology. Although these are observational findings, several speculative but biologically plausible mechanisms linked to immune dysregulation may explain the differential associations observed by SARIFA status. Obesity is well known to induce systemic and local immunomodulatory effects on several immune cell subpopulations (38), such as dendritic cells, regulatory T cells (39), and myeloid immunosuppressive cells (40). Notably, obesity has been associated with lower numbers of regulatory T cells in visceral adipose tissue (39). In our previous work, we provided the first evidence of potentially lower numbers of FoxP3-positive regulatory T cells in SARIFA-positive colorectal cancers (10). Lower numbers of regulatory T cells have been linked to a poor prognosis in colorectal cancer in several studies (41–43). Consistent with recent evidence that obesity impairs natural killer (NK) cell–mediated immune surveillance through lipid-induced metabolic paralysis of NK cells within the TME (44), our earlier study also demonstrated a reduction of circulating and tumor-infiltrating NK cells in SARIFA-positive colorectal tumors (18). In prostate cancer, obesity-driven tumor progression has been found to be closely linked to the pro-inflammatory CXCL12–CXCR4/CXCR7 signaling axis (45), underlining the close interconnection between immunologic changes caused by obesity and cancer progression.

Consensus molecular subtype (CMS) classification based on transcriptomic profiling was not available for the NLCS tumors; however, prior work has reported enrichment of CMS1 and CMS4 among SARIFA-positive colorectal cancers (11), supporting the hypothesis that SARIFA reflects a distinct biological tumor phenotype. Future studies should explore associations between energy balance–related factors and these clinically relevant molecular and histopathologic characteristics.

A major strength of our study is the use of the large and well-characterized NLCS with detailed baseline data on diet and lifestyle, as well as long follow-up (over 20 years) and colorectal cancer subtyping. The prospective design enabled us to investigate associations between baseline energy balance–related factors and the risk of developing colorectal cancer, as defined by the presence or absence of a histologically defined invasion front biomarker (SARIFA). The prospective design minimizes the risk of selection and information bias, and the long follow-up yielded large numbers of incident cancer cases, which are needed for MPE studies. However, some limitations merit consideration. MPE studies are inevitably limited by the number of cases with available tumor tissue and successful laboratory analyses, even in large prospective cohorts like the NLCS. On top of this, stratification by sex and tumor location is needed, as associations between energy balance–related factors and colorectal cancer risk are known to differ between anatomic subsites and in men and women. This may have reduced statistical power in some subgroups and our ability to detect associations, requiring replication in even larger, independent cohorts or the pooling of data from multiple cohorts.

In addition, exclusions due to unavailable or nonretrievable tumor material, largely related to block age, may have introduced selection. However, analyses within the NLCS have shown that included and nonretrieved cases are comparable after accounting for age and year of diagnosis, suggesting limited selection related to tissue availability (46). The observed associations between adiposity and colon cancer risk in this study are also consistent with prior NLCS analyses including all incident cases (47), suggesting that case exclusion is unlikely to have materially influenced the results.

Furthermore, the long follow-up for cancer in the NLCS may have allowed changes in body size and physical activity (due to lifestyle changes, cancer progression, or aging) to go undetected, as only a single baseline measurement of energy balance–related factors was available, potentially diluting the associations.

Nevertheless, the NLCS represents an example of the prospective cohort incident-tumor biobank method (PCIBM), a research framework that integrates long-term exposure assessment in cancer-free individuals with systematic collection of tumor tissue from incident cases. Recently, Ogino and colleagues (48) highlighted the NLCS as a flagship PCIBM-applicable cohort as the NLCS enables MPE investigations that uniquely link lifestyle factors, tumor biology, cancer development, and clinical outcomes while minimizing recall and selection bias inherent to case-series designs. Within this framework, our study leverages decades-long prospective data and tumor subtyping to examine etiologic heterogeneity by SARIFA status, illustrating the distinctive capacity of PCIBM-based cohorts to uncover biologically informed exposure–tumor relationships relevant to cancer prevention and precision epidemiology.

Our findings indicated that a higher prediagnostic/baseline BMI may be more strongly associated with a higher risk of developing SARIFA-positive than SARIFA-negative colon cancers, suggesting a potential role of adiposity in the development of this aggressive tumor phenotype. Future studies are essential to elucidate the mechanisms by which adiposity modulates the tumor–stroma interface and immune microenvironment. Such insights could inform the development of targeted prevention strategies and therapeutic interventions and support improved patient stratification in clinical practice.

Supplementary Material

Supplementary Table S1

Supplementary Table S1. Age-adjusted hazard ratios and 95% confidence intervals for the association between body mass index and risk of SARIFA-positive and SARIFA-negative colorectal cancer, stratified by sex and tumor location.

Supplementary Table S2

Supplementary Table S2. Age-adjusted hazard ratios and 95% confidence intervals for the association between lower-body clothing size (as a proxy for adiposity) and risk of SARIFA-positive and SARIFA-negative colorectal cancer, stratified by sex and tumor location.

Supplementary Table S3

Supplementary Table S3. Age-adjusted hazard ratios and 95% confidence intervals for the association between non-occupational physical activity and risk of SARIFA-positive and SARIFA-negative colorectal cancer, stratified by sex and tumor location.

Data Availability

The data are derived from the NLCS and are subject to ethical, legal, and governance restrictions. We further state that participants did not provide consent for unrestricted public sharing of individual-level data and that public deposition could compromise participant confidentiality and privacy. Requests can be addressed to the corresponding authors and will be considered upon reasonable request and approval by instructional review boards.

Authors’ Disclosures

N.G. Reitsam reports other support from Bruker Spatial Biology outside the submitted work. No disclosures were reported by the other authors.

Authors’ Contributions

K. Offermans: Conceptualization, formal analysis, writing–original draft, writing–review and editing. N.G. Reitsam: Conceptualization, formal analysis, writing–original draft, writing–review and editing. B. Grosser: Writing–review and editing. J. Zimmermann: Formal analysis, writing–review and editing. H.I. Grabsch: Conceptualization, writing–review and editing. C.C.J.M. Simons: Conceptualization, supervision, funding acquisition, writing–review and editing. P.A. van den Brandt: Conceptualization, data curation, supervision, funding acquisition, methodology, writing–review and editing. B. Märkl: Conceptualization, supervision, writing–review and editing.