Impact of vitamin D on the colon cancer immune microenvironment: results of a randomized clinical trial of preoperative vitamin D supplementation in patients with stage I-III colon cancer

Abstract

Although vitamin D (VitD) exhibits anti-tumor activity in colorectal cancer (CRC) preclinically, its effects in the human tumor microenvironment (TME) remain unclear. We conducted a randomized, placebo-controlled trial of preoperative high-dose VitD supplementation in stage I–III colon cancer patients to assess its impact on the TME. Forty-two patients received either VitD3 (50,000 IU/day for 7 days, then 10,000 IU/day) or placebo before surgery. Spatial immune-profiling and assessment of VitD receptor (VDR) and CYP27B1 expression were performed on paired tumor samples from 24 patients. VitD significantly increased plasma 25-hydroxyvitamin D levels (P<0.001), increased CD3+CD8+ memory T cells (P=0.03), reduced CD3+CD4+FoxP3+ regulatory T cells (P=0.02) and spatially re-organized the TME, leading to greater T cell and tumor cell proximity. Post-treatment VDR expression was heterogeneous and decreased overall (P=0.02). Spatial transcriptomic profiling of post-treatment resections reflected predominantly repressive VDR activity. These findings support an immunomodulatory role for VitD, warranting further mechanistic investigation.

INTRODUCTION

Colorectal cancer (CRC) is the fourth most common cancer worldwide and the second leading cause of cancer-related death(1). Approximately 20% of colon cancer patients present de novo with metastatic disease, and around 50% will eventually develop metastases(2). Despite significant advances in the treatment of CRC in recent years(3), the five-year survival rate for individuals with metastatic disease is only approximately 14%(4). Consequently, there is a critical need to develop novel and effective therapies to improve survival that are also safe and affordable.

Vitamin D has been broadly reported to exert anti-tumor effects in many cancer types, including CRC, by signaling through the vitamin D receptor (VDR). In addition to inducing cell differentiation and apoptosis, and inhibiting proliferation, angiogenesis, metastatic potential, and cancer stem cell activity, vitamin D also possesses potent anti-inflammatory and immunomodulatory properties, which are particularly relevant as inflammation is an established precursor to and risk factor for CRC(5–10). The rationale for investigating vitamin D in CRC is further supported by several studies demonstrating a significant association between higher circulating 25-hydroxyvitamin D [25(OH)D] levels and improved outcomes in patients with CRC(11–14), including those with stage III colon cancer(15,16). Moreover, the phase II randomized clinical trial SUNSHINE reported that high-dose vitamin D supplementation increased progression-free survival in patients with metastatic CRC(17), although this finding was not reproduced by the SOLARIS phase III randomized clinical trial(18).

A dominant mechanism of action of vitamin D in CRC may potentially be activation of anti-tumor immunity. VDR and 1α-hydroxylase (CYP27B1), which converts circulating 25(OH)D into active 1,25-dihydroxyvitamin D3 [1,25(OH)₂D₃], are present in most cells of the body(19), including CRC tumor epithelial cells(20–22), immune cells(23,24), and fibroblasts(25). While the expression of VDR and CYP27B1 in these cell types suggests that 1,25(OH)₂D₃ can be synthesized locally within the tumor and its microenvironment to drive autocrine and paracrine signaling, the impact of vitamin D in the CRC tumor microenvironment (TME) has thus far remained largely unexplored. We therefore conducted a randomized placebo-controlled trial to explore the effects of preoperative high-dose vitamin D supplementation on the TME in patients with resectable stage I-III colon cancer. We investigated the expression of VDR and CYP27B1 within specific cell types and analyzed changes in immune cells following supplementation using customized, spatially-resolved multiplex immunofluorescence (mIF) assays and spatial transcriptional profiling.

RESULTS

Patient flow and characteristics

A total of 42 patients with newly diagnosed stage I-III colon cancer were enrolled from February 2016 to August 2019 (23 randomized to vitamin D and 19 randomized to placebo) to reach the protocol-specified accrual goal of 24 randomized patients who completed protocol therapy and had sufficient tumor tissue available for analysis (11 vitamin D-treated and 13 placebo-treated patients) (Supplementary Figure S1). Eighteen enrolled patients needed to be replaced due to withdrawal of consent (n=5), not starting or completing protocol-specified vitamin D/placebo treatment (n=6), insufficient tumor tissue available at the time of resection for research (n=3), other reason for inability to collect tissue (n=1), surgery not being scheduled within the protocol-specified window (n=2), and no residual cancer found in the resection specimen (n=1). After exclusions due to absence of residual tumor in the tissue block and inability to retrieve tissue or missing tissue, the remaining pre-treatment biopsies and post-treatment resection specimens underwent tissue-based analysis for assessment of vitamin D receptor status and tumor immune microenvironment characterization (Supplementary Figure S2).

Table 1 shows the baseline demographic and clinicopathologic characteristics of the study population of 24 randomized patients who completed protocol therapy and had tumor tissue collected at resection. Overall, variables were well-balanced between the two groups. Notably, treatment duration did not differ significantly between study arms. The median treatment duration was 8 days for the vitamin D group (interquartile range [IQR] 7–13) and 13 days for the placebo group (IQR 8–13) (P = 0.13), while the mean treatment duration ± standard deviation (SD) was 9.5 ± 3.3 days for the vitamin D group (P=0.36) and 11.5 ± 4.6 days for the placebo group. Supplementary Table S1 summarizes the representativeness of the trial participants.

Table 1.

Baseline demographic and clinicopathologic characteristics of the study population.

Characteristics		Placebo	Vitamin D	Overall	P-valuea	SMD
		(N=13)	(N=11)	(N=24)
Age (median, IQR), years		60.7 (54.2, 69.9)	60.8 (47.7, 77.9)	60.8 (52.0, 70.6)	0.91	0.066
Sex (n, %)	Male	5 (38.5%)	6 (54.5%)	11 (45.8%)	0.68	0.327
	Female	8 (61.5%)	5 (45.5%)	13 (54.2%)
Race (n, %)	White	12 (92.3%)	9 (81.8%)	21 (87.5%)	0.43	0.316
	Others	1 (7.7%)	2 (18.2%)	3 (12.5%)
BMI (median, IQR), kg/m2		29.4 (26.9, 32.4)	28.3 (23.7, 35.9)	29.2 (25.1, 33.4)	0.82	0.051
ECOG PSb (n, %)	0	8 (61.5%)	6 (54.5%)	14 (58.3%)	1	0.142
	1	5 (38.5%)	5 (45.5%)	10 (41.7%)
CEAc (n, %)	Normal	10 (76.9%)	8 (80.0%)	18 (78.3%)	1	0.075
	High	3 (23.1%)	2 (20.0%)	5 (21.7%)
Tumor stage (n, %)	I	2 (15.4%)	2 (18.2%)	4 (16.7%)	1	0.200
	II	6 (46.2%)	4 (36.4%)	10 (41.7%)
	III	5 (38.5%)	5 (45.5%)	10 (41.7%)
Grade of differentiation (n, %)	Well differentiated	2 (15.4%)	1 (9.1%)	3 (12.5%)	0.09	0.946
	Moderately differentiated	10 (76.9%)	5 (45.5%)	15 (62.5%)
	Poorly differentiated	1 (7.7%)	5 (45.5%)	6 (25.0%)
T staged (n, %)	T1	0 (0%)	0 (0%)	0 (0%)	0.86	0.368
	T2	2 (15.4%)	3 (27.3%)	5 (20.8%)
	T3	7 (53.8%)	6 (54.5%)	13 (54.2%)
	T4	4 (30.8%)	2 (18.2%)	6 (25.0%)
N stagee (n, %)	N0	8 (61.5%)	6 (54.5%)	14 (58.3%)	0.62	0.570
	N1	3 (23.1%)	1 (9.1%)	4 (16.7%)
	N2	2 (15.4%)	4 (36.4%)	6 (25.0%)
Primary sitef (n, %)	Right colon	8 (61.5%)	7 (63.6%)	15(62.5%)	0.82	0.483
	Transverse colon	0 (0.0%)	1 (9.1%)	1 (4.2%)
	Left colon	5 (38.5%)	3 (27.3%)	8 (33.3%)
BRAF mutation status (n, %)	Wild type	5 (38.5%)	4 (36.4%)	9 (37.5%)	1	0.423
	Mutated	1 (7.7%)	0 (0.0%)	1 (4.2%)
	Unknown	7 (53.8%)	7 (63.6%)	14 (58.3%)
NRAS mutation status (n, %)	Wild type	6 (46.2%)	4 (36.4%)	10 (41.7%)	0.7	0.200
	Mutated	0 (0%)	0 (0%)	0 (0%)
	Unknown	7 (53.8%)	7 (63.6%)	14 (58.3%)
KRAS mutation status (n, %)	Wild type	5 (38.5%)	1 (9.1%)	6 (25.0%)	0.23	0.839
	Mutated	1 (7.7%)	3 (27.3%)	4 (16.7%)
	Unknown	7 (53.8%)	7 (63.6%)	14 (58.3%)
MSI status (n, %)	MSS	11 (84.6%)	6 (54.5%)	17 (70.8%)	0.18	0.692
	MSI-H	2 (15.4%)	5 (45.5%)	7 (29.2%)
Plasma 25(OH)D (median (IQR)) (ng/ml)		16.8 (14.2, 20.8)	20.1 (16.8, 29.3)	18.7 (15.4, 22.1)	0.12	1.073
Corrected calcium (median (IQR)), mg/dl		9.7 (9.6, 9.9)	9.4 (9.3, 9.5)	9.6 (9.4, 9.7)	0.02	0.975

Abbreviations: SMD: standardized mean differences, IQR: interquartile range, BMI: body mass index, ECOG: Eastern Cooperative Oncology Group, PS: performance status, CEA: carcinoembryonic antigen, MSI: microsatellite instability, MSS: microsatellite stable, MSI-H: microsatellite instability high

^aP-values were calculated with Fisher’s exact test (categorical variables) and Wilcoxon rank-sum test (continuous variables).

^bECOG performance status: 0 - Fully active, able to carry on all pre-disease performance without restriction; 1 - Restricted in physically strenuous activity but ambulatory and able to carry out work of a light or sedentary nature.

^cCEA available in 10 patients in vitamin D arm. After January 2018, CEA < 3.8 was defined as “Normal”; prior to that, CEA < 2.5 was “Normal”.

^dT1 = Tumor invades the submucosa; T2 = Tumor invades the muscularis propria; T3 = Tumor invades through the muscularis propria into pericolonic tissues; T4 = Tumor invades the visceral peritoneum or invades or adheres to adjacent organs

^eN0 = no regional lymph node metastases; N1 = 1–3 regional lymph nodes positive, or any number of tumor deposits are present and all identifiable lymph nodes are negative; N2 = 4 or more regional nodes positive

^fRight colon includes cecum, ascending colon, and hepatic flexure; left colon includes descending, sigmoid, and rectosigmoid colon

Safety

Among the 42 patients who were initially enrolled and randomized, 33 patients took at least one dose of vitamin D/placebo and were evaluable for safety. Five of the 33 patients (15%) (3 assigned to vitamin D, 2 assigned to placebo) reported adverse events, the majority of which were not related to vitamin D/placebo (Supplementary Table S2). One patient randomized to vitamin D reported grade 1 nausea that was possibly related to vitamin D, and another patient randomized to placebo reported grade 2 vomiting and grade 2 diarrhea that were unlikely related to vitamin D/placebo. There were no reports of hypercalcemia or genitourinary stones among the safety evaluable population.

Analysis of plasma 25(OH)D levels

The median baseline plasma 25(OH)D level was 18.7 ng/mL (IQR: 15.4, 22.1) in the overall study population. Most patients (95.8%) had insufficient levels (<30 ng/mL), while only one patient (4.2%) had sufficient levels (≥30 ng/mL) at baseline (Supplementary Figure S3A, Supplementary Table S3). No statistically significant differences in plasma 25(OH)D levels were observed between the two treatment arms at baseline (P=0.12) (Supplementary Figure S3B, Supplementary Table S3).

Patients randomized to the vitamin D group had significantly increased post-treatment plasma levels of 25(OH)D compared to the placebo group (median 59.6 vs. 17.8 ng/mL, respectively, resulting in a median increase of 150%, paired test P=0.001), confirming the efficacy of the administered vitamin D regimen. Conversely, patients randomized to placebo did not demonstrate a significantly increased plasma 25(OH)D level post-treatment (paired test P=0.74) (Supplementary Table S3).

We also investigated the association between baseline plasma 25(OH)D level, change in plasma 25(OH)D level, and clinicopathologic factors (Supplementary Table S4). Baseline plasma 25(OH)D levels did not differ by any demographic, clinical, pathologic, or molecular characteristics tested. Change in plasma 25(OH)D level (difference between post- and pre-treatment levels) was negatively associated with body-mass index (BMI) (Spearman’s correlation = −0.87, P=0.0005), consistent with prior studies(16,26,27). Poorly differentiated tumors were associated with a more substantial change in plasma 25(OH)D levels from baseline in the vitamin D supplementation group (median change of 47.5 ng/mL vs.13.7 ng/mL in well differentiated tumors and 30.6 ng/mL in moderately differentiated tumors, P=0.06).

Colon cancer microenvironment composition

We first assessed the TME composition (Figure 1) in pre-treatment specimens using a custom-built, spatially-resolved mIF panel designed to characterize key cellular components of the TME while simultaneously assessing VDR and CYP27B1 expression (VDR panel). Tissue segmentation and cell characterization were performed using supervised machine learning (Figure 2A). Across 17 tumors, we identified 394,674 cells, including 140,759 cytokeratin (CK)-positive tumor cells, 188,089 immune cells, and 65,826 non-epithelial, non-immune cells (stromal cells). There was no dominant cell population, and the fractional abundance of the main cell population varied greatly across tumors (Supplementary Figure S4A–B). Furthermore, densities of the major cell populations were similar between the placebo and treatment groups (P=0.71, Supplementary Figure S4C).

Figure 1. Study design and analytical workflow.

Patients with stage I–III colon cancer underwent pre-treatment biopsy collection followed by randomization to placebo or vitamin D supplementation prior to surgical resection. Plasma 25(OH)D levels were measured before and after treatment. Pre-treatment biopsies and post-treatment resection specimens were analyzed by multiplex immunofluorescence (mIF), whereas spatial transcriptomic profiling was performed on post-treatment resection specimens only. For mIF analyses, multispectral digital imaging combined with supervised machine learning–based image analysis was used to characterize epithelial, stromal, and immune cell populations. Cell phenotyping enabled identification of T cell subsets, including helper, cytotoxic, regulatory, and memory T cells, as well as macrophages and other immune populations, together with assessment of vitamin D receptor (VDR) and CYP27B1 expression across tissue compartments. Quantitative analyses included cell density, relative abundance, compartment-specific localization, spatial proximity, and cellular neighborhood organization within tumor and stromal regions. Spatial transcriptomic analyses included evaluation of individual gene expression patterns, differential gene expression, and cell-type deconvolution to infer the relative contribution of distinct cellular populations within spatially resolved regions. Portions of figure created in BioRender.

Abbreviations: 25(OH)D: 25-hydroxyvitamin D; mIF: multiplex immunofluorescence

Figure 2. Overview of VDR and CYP27B1 expression in pre-treatment specimens of colon cancer.

(A) Examples of multiplex immunofluorescence (mIF) images and corresponding phenoplots for all cells phenotyped in pre-treatment samples, showing VDR and CYP27B1 coexpression. (B) Relative abundance of VDR expression across 17 pre-treatment biopsy specimens, stratified by main cell type and treatment arm. (C) Examples of multiplex images showing VDR expression in pre-treatment biopsies stratified by tumor histological grade. (D) Treatment effect on VDR expression in all cells (n=17 pre-treatment biopsies; n=18 post-treatment resections). P-values were computed using LMM. Scale bar is 50 um.

Abbreviations: VDR: Vitamin D receptor; CYP27B1: 1-α-hydroxylase; 25(OH)D: 25-hydroxy vitamin D. LMM: linear mixed model.

Assessment of VDR and CYP27B1 protein expression

Although VDR expression varied across individual pre-treatment specimens (Figure 2B), tumor cells had the highest levels, followed by immune and stromal cells (mean VDR percentages 24%, 5%, and 1%, respectively; Supplementary Figure S4D). Given known molecular, histological, and survival differences in CRC by tumor sideness(28–30), we next investigated whether VDR expression differed by tumor location. Interestingly, tumors located on the left side exhibited higher VDR expression in tumor cells compared to right-sided tumors (median 32.1% vs. 21.8%, P=0.08) (Figure 2B; Supplementary Figure S4E; Supplementary Table S5). A similar trend was observed by histologic grade, with higher VDR expression in well/moderately differentiated tumors compared to poorly differentiated tumors (median 28.0% vs. 14.3%, P=0.10) (Figure 2B and 2C; Supplementary Figure S4F; Supplementary Table S5). Finally, VDR expression was not correlated with any other clinical variables, including baseline plasma levels of 25(OH)D (Supplementary Figure S4G), pathological, or molecular variables (Supplementary Table S5). Interestingly, vitamin D supplementation was associated with decreased VDR expression in aggregated cells in post-treatment specimens compared to pre-treatment specimens (P=0.02), while no significant change was observed in post-treatment specimens in the placebo arm (P=0.18) (P-interaction=0.009) (Figure 2D; Supplementary Table S6).

To determine whether these changes reflected shifts in tissue composition, we quantified VDR expression separately within tumor, immune, and stromal compartments by calculating the proportion of VDR-positive cells within each cell type. This analysis revealed a compartment-specific pattern. In tumor epithelial cells, VDR expression decreased following vitamin D treatment (Δ=-12.9%, P=0.001), with a significantly different trajectory compared to placebo (P-interaction=0.04), where no meaningful change was observed. No differences were observed in immune cells (P-interaction=0.16), whereas stromal cells showed an increase in both arms, with a greater increase observed in the placebo arm (P-interaction=0.01) (Supplementary Figure S5-). These findings indicate that distinct cell types in the colon TME respond differently to vitamin D treatment, with tumor cells lowering VDR expression, potentially due to a negative feedback loop, whereas immune cells are unaffected and stromal cells show an increase in VDR expression that is less than that seen in placebo-treated tumors.

CYP27B1 expression in pre-treatment specimens also showed substantial intertumoral variability (Supplementary Figure S6A) and was predominantly localized in tumor cells, followed by immune and stromal cells (mean expression 13%, 3%, and 1%, respectively). A trend toward higher CYP27B1 expression in tumor epithelial cells was observed in left-sided tumors (P=0.11) (Supplementary Figure S6B; Supplementary Table S7), with no significant associations with other clinical or molecular features. Following treatment, CYP27B1 expression increased in tumor epithelial and immune compartments in both treatment arms (P-interaction=0.03 and 0.01, respectively), although this effect was not observed when considering aggregate cell populations (P-interaction=0.53, Supplementary Table S6).

Assessment of VDR-mediated transcriptional activity

To further assess VDR pathway activity, we performed spatial transcriptional profiling using a customized ring-contouring approach to evaluate whole transcriptome profiles of tumor epithelial regions and surrounding stromal regions using the Bruker GeoMx platform. In this approach, cytokeratin expression within tumor cells was used to computationally define concentric rings of stroma 25 μm in width that surrounded each tumor region (Figure 3A). After quality control, 354 whole transcriptome profiles were generated across 15 resection cases. Assessment of expression of CYP24A1 and CAMP, known direct transcriptional targets of VDR, did not show significant differences between treatment arms (Supplementary Figure S7). However, differential expression analysis comparing tumor regions between treatment arms identified 15 differentially expressed genes, the majority of which were upregulated in the placebo arm (Figure 3B); this finding is consistent with the predominantly repressive nature of VDR transcriptional activity. The most differentially expressed gene was CLDN2 (Claudin-2), which encodes a tight junction protein that regulates epithelial cell permeability in the colonic mucosa and is a known direct transcriptional target of VDR signaling(25). This result is consistent with our finding that VDR protein levels were higher in placebo versus vitamin D-treated tumor regions. Additional genes upregulated in tumor epithelium included CDHR2 (cadherin-related family member 2) which encodes an apical membrane-associated protein that regulates intestinal epithelial cell polarity, differentiation, and brush border organization; genes linked to proliferation and the Wnt/β-catenin axis (SP5, MSLN); genes linked to epithelial injury, remodeling, and inflammation (CEMIP, DPEP1, and PRAP1); and genes linked to cellular stress responses, including an antioxidant enzyme (GPX2), and steroid, prostaglandin, and histamine metabolizing enzymes (AKR1C3, AOC1). Overall, these results were consistent with VDR’s known roles in promoting epithelial differentiation and barrier integrity as well as inflammatory responses and mucosal defense.

Figure 3. Spatial transcriptomic analysis of vitamin D-associated changes in the colon cancer microenvironment (n=15).

(A) Representative regions of interest (ROIs) and corresponding areas of illumination (AOIs) used for spatial whole transcriptome profiling. ROIs were defined based on immunofluorescence staining (nuclei, Pan-CK, and CD45), and segmented into epithelial compartment (Center) and concentric stromal regions (Ring 1 and Ring 2) for spatially resolved RNA collection. (B) Differential gene expression analysis comparing vitamin D-treated and placebo groups within tumor-enriched regions (Center). Volcano plot displays log2 fold change versus –log10 adjusted p-value. (C) Vitamin D-associated CAF signature scores in stromal compartments (Ring 1 and Ring 2) stratified by treatment arm.

Abbreviations: VDR: Vitamin D Receptor; CAF: Cancer-associated fibroblasts

Given well-established links between vitamin D signaling and stromal modulation(25,31,32), we next assessed transcriptional changes in stromal regions. Unlike the tumor region analysis, differential expression analysis did not reveal significant expression changes, a result likely due to higher cellular heterogeneity within stromal regions, which contain a mix of fibroblasts, immune cells, and other cell types. As a more sensitive and targeted measure of VDR activity, we therefore evaluated a gene set representing vitamin D-related transcriptional changes in primary cancer-associated fibroblasts derived from primary human CRC samples(25) (see Methods, Supplementary Table S8). Since VDR acts primarily as a transcriptional suppressor, a lower signature score reflects greater VDR activity. Within stromal rings, the score for this signature showed a trend towards lower expression in tumor-proximal stroma (ring 1), but not in tumor-distal stroma (ring 2), from vitamin D-treated tumors as compared to placebo-treated tumors (Figure 3C). While this difference was not statistically significant, the RNA signature decrease in ring 1 was consistent with the spatial pattern of VDR protein expression within stromal cells, where expression was enriched in stromal cells located within 25 μm of tumor cells (Supplementary Figure S8). In summary, while tumor epithelial regions displayed evidence of transcriptional reprogramming related to vitamin D treatment, changes in the stromal compartment were less pronounced, although repressive VDR transcriptional activity was still spatially coordinated with VDR protein expression in this compartment. While these results could be explained by multiple factors, the time interval between receipt of last dose of placebo or vitamin D and surgical resection (7–28 days) may have been too long for identification of more overt VDR-mediated transcriptional changes in all tissue compartments.

Immune microenvironment response to vitamin D supplementation

Given prior reports of changes in immune cell infiltration in tumors treated with vitamin D and its analogs(33–35), we next assessed the effect of vitamin D supplementation on immune cell density using T cell and macrophage markers in our VDR mIF panel (Figure 1). Overall, the fractional abundance of the main cell populations varied substantially across 18 post-treatment colon cancer resections (Supplementary Figure S9A–B). Although no statistically significant changes were observed in tumor (P-interaction=0.43) and aggregate immune cell densities (P-interaction=0.21) overall following vitamin D supplementation (Figure 4A; Supplementary Table S6), the median change in CD3+ T cells was 177% in the treatment arm compared to 44% in the placebo group. To investigate this further, we utilized a dedicated T cell mIF panel to more deeply characterize specific T cell subsets (Figure 1). This confirmed a significant increase in CD3+ T cell density following vitamin D treatment (P=0.03), but no change in the placebo group (P=0.96) (P-interaction=0.054) (Figure 4B; Table 2). Furthermore, vitamin D supplementation decreased CD3+CD4+FoxP3+ regulatory T cells (Figure 4C, P=0.02), and increased CD3+CD8+ cytotoxic T cells (Figure 4D, P=0.02), particularly the CD3+CD8+CD45RO+ memory subset of cytotoxic T cells (P=0.03, Figure 4E; Table 2). Notably, when considering only paired samples, 75% of patients who received vitamin D supplementation had a higher CD3+CD8+ cytotoxic T cell density (Supplementary Figure S9C–D), again largely driven by CD3+CD8+CD45RO+ memory cytotoxic T cells (Supplementary Figure S9C). This result contrasted sharply with the placebo group, where density increased in only 30% of patients. These associations were consistent after adjusting for MSI status, grade of differentiation, baseline BMI, T stage, and primary tumor site individually and in a fully-adjusted multivariable model (Supplementary Table S9).

Figure 4. Impact of pre-operative vitamin D supplementation on changes in the colon cancer microenvironment.

(A) Tile plots showing the percentage change in the specified cell populations after treatment (n=17). (B) Treatment effect on CD3+ T cell density (n=23 pre-treatment biopsies; n=18 post-treatment resections), (C) CD3+CD4+FoxP3+ regulatory T cell density (n=23 pre-treatment biopsies; n=18 post-treatment resections), (D) CD3+CD8+ cytotoxic T cell density (n=23 pre-treatment biopsies; n=18 post-treatment resections), and (E) CD3+CD8+CD45RO+ memory cytotoxic T cell density (n=23 pre-treatment biopsies; n=18 post-treatment resections). P-values were computed using LMM.

Abbreviations: LMM: linear mixed model.

Table 2.

Pre- and post-treatment comparison of immune cell densities according to randomization arm.

T cell Subtype Density (1/mm2)a	Placebo				Vitamin D				P-interactiond
	Pre-treatment	Post-treatment	Estimate (95% CI)c	P-valuec	Pre-treatment	Post-treatment	Estimate (95% CI)c	P-valuec
	(N=12)	(N=10)			(N=11)	(N=8)
	Median (IQR)b	Median (IQR)b			Median (IQR)b	Median (IQR)b
CD3+	1135 (524, 1637)	1053 (796, 2118)	−9.7 (−380.0, 360.6)	0.96	1220 (865, 1909)	2139 (1182, 2425)	636.1 (68.4, 1203.8)	0.03	0.054
CD3+CD4+	868 (357, 1327)	813 (485, 1699)	38.5 (−262.8, 339.8)	0.80	1009 (498, 1616)	1439 (952, 1665)	406.3 (−37.5, 850.0)	0.07	0.17
CD3+CD4+FoxP3-	730 (335, 1090)	750 (435, 1332)	58.2 (−217.6, 334.0)	0.68	892 (441, 1422)	1352 (887, 1579)	477.1 (81.4, 872.9)	0.02	0.08
CD3+CD4+FoxP3+	81 (27, 173)	67 (50, 90)	−17.1 (−53.9, 19.7)	0.36	163 (61, 198)	92 (34, 111)	−76.0 (−140.1, −11.9)	0.02	0.09
CD3+CD8+	184 (98, 381)	246 (114, 419)	−29.0 (−135.0, 76.9)	0.59	305 (190, 553)	648 (218, 781)	209.7 (39.1, 380.2)	0.02	0.01
CD3+CD8+CD45RO-	0 (0, 8)	25 (8, 51)	22.2 (4.9, 39.6)	0.01	2 (1, 8)	24 (8, 37)	20.5 (8.6, 32.4)	0.0009	0.89
CD3+CD8+CD45RO+	165 (98, 370)	203 (108, 342)	−49.1 (−147.5, 49.4)	0.33	305 (185, 551)	614 (200, 765)	188.5 (20.5, 356.5)	0.03	0.009

Abbreviation: MIF, Multiplex Immunofluorescence; VDR, Vitamin D Receptor; ROI, Region of Interest; S.E., Standard Error

^aAggregated cell density, combining intraepithelial and stromal areas, is represented.

^bROIs combined on a per patient basis to aggregate patient-level data.

^cLinear mixed effect model was applied to ROI-level data comparing post-treatment to pre-treatment. Cell density from MIF = Sample timepoint + (1|Patient ID).

^dP-interaction was calculated by including the product term of randomization arm x sample timepoint in the linear mixed effect model with ROI-level data.

To characterize the immune microenvironment beyond T cells, we performed deconvolution(36) on stromal regions from our resected tumor spatial transcriptional profiling dataset in order to assess inferred cellular composition. Using a deconvolution algorithm optimized for spatial transcriptomics, we were able to calculate inferred abundances for 16 different immune cell types, in addition to fibroblasts and endothelial cells. This analysis revealed a trend towards increased T cells in vitamin D-treated tumors, consistent with prior mIF results and an increase in the CD8:CD4 ratio. This analysis also revealed enrichment of myeloid cell populations, particularly macrophages, in vitamin D-treated tumors (Supplementary Figure S9E). In contrast, fibroblasts were less abundant in vitamin D-treated tumors. Consistent with these findings, exploratory analysis of stromal regions showed consistent directional increases in IFN-γ-associated chemokines involved in T cell recruitment (CXCL9, CXCL10, and CXCL11), with no corresponding increase in IFNG expression itself (Supplementary Figure S9F). In parallel, we checked expression of immune checkpoints with a well-established role in CRC, including PDCD1 (PD-1), CTLA4, LAG3, HAVCR2 (TIM-3), and TIGIT. The resultant analysis demonstrated higher expression of HAVCR2 and TIGIT in stromal rings after treatment with vitamin D (Supplementary Figure S10). While these differences did not reach statistical significance by a linear mixed model accounting for repeated measurements per patient, they are suggestive of VDR-mediated remodeling of the immune microenvironment.

Taken together, these findings demonstrate a treatment-specific alteration of T cell subsets associated with vitamin D supplementation and support an immunomodulatory impact of vitamin D on the colon cancer TME that may extend beyond T cells. Notably, this vitamin D-associated shift towards increased cytotoxic T cells and decreased regulatory T cells represents an immune cell profile that has been associated with anti-tumor activity and favorable prognosis(37–39).

Changes in CD8+ T cell spatial configuration after vitamin D supplementation

Considering the known prognostic significance of lymphocytic infiltration at the tumor edge in stage I-III colon cancer patients(40,41), we assessed whether vitamin D supplementation impacted T cells in the tumor center versus the invasive edge in post-treatment resection specimens (Figure 5A). We observed higher densities of CD3+CD8+ cytotoxic T cells (P=0.008; Figure 5B), particularly CD3+CD8+CD45RO+ memory cytotoxic T cells (P=0.0008; Figure 5C), at the invasive edge of the tumor compared to the tumor center in the placebo group (Supplementary Table S10). Conversely, VDR expression was higher in the tumor center as opposed to the invasive edge in all cells (P<0.001, Figure 5D), including tumor cells specifically (P<0.001, Figure 5E), in the placebo group (Supplementary Table S10). However, both cytotoxic T cells and VDR expression were evenly distributed throughout post-treatment tumors from patients who received vitamin D supplementation (Figures 5B–E, Supplementary Table S10). These findings suggest a distinct immunological landscape within the TME, wherein vitamin D supplementation may differentially influence the distribution of cytotoxic T cells and VDR expression between the tumor center and the invasive edge.

Figure 5. Effects of pre-operative vitamin D supplementation on changes in VDR expression and cytotoxic T cell spatial distribution in the colon cancer microenvironment.

(A) Examples of multiplex immunofluorescence (mIF) images and corresponding phenoplots representing CD3+CD8+ cytotoxic T cells, CD3+CD8+CD45RO+ memory cytotoxic T cells, and VDR expression in all cells and tumor cells, highlighting differences in spatial distribution in the tumor center and the invasive edge of colon cancer tissue in post-treatment samples. (B-E) Box plots comparing the aforementioned cell populations in the tumor center and the invasive front, stratified by treatment arm (n=18). (F-G) Box plots showing the mean fraction of CD3+CD8+ cytotoxic T cells (F) and CD3+CD8+CD45RO+ memory cytotoxic T cells (G) within tumor-centered k-nearest neighbor (kNN) regions, in pre- and post-treatment samples stratified by treatment arm. P-values were computed using LMM. Scale bar is 50 um.

Abbreviations: VDR: Vitamin D receptor; LMM: linear mixed model.

Given the uniform distribution of VDR and cytotoxic T cells in vitamin D-treated tumors, we next assessed whether vitamin D supplementation altered co-localization of tumor and immune cells at the individual cell level. Comparison of pre- and post-treatment samples using two different spatial metrics (Gcross and NND), showed a trend towards greater co-localization of tumor cells and CD3+CD8+ cytotoxic T cells, but not CD3+CD4+ helper T cells (Supplementary Figure S11A, Supplementary Table S11). This trend was consistent across a range of tested distances (10–30μm) using the Gcross metric, further supporting the robustness of the results.

To contextualize these findings, which showed CD3+CD8+ cytotoxic T cell spatial reconfiguration at a macroscopic level (tumor center versus invasive edge) and at the level of individual microns, we performed two complementary analyses at an intermediate spatial scale. While Gcross and NND quantify tumor-immune cell proximity, they do not assess whether tumor cells are surrounded by a broader immune microenvironment that could support sustained interactions with CD3+CD8+ cytotoxic T cells. To address this, we performed tumor-centered k-nearest neighbor (kNN) analysis(42), which assesses the composition of cells immediately surrounding each tumor cell. This analysis showed an increased proportion of CD3+CD8+ cytotoxic (Figure 5F, Supplementary Table S11) and CD3⁺CD8⁺ memory cytotoxic T cells (Figure 5G) adjacent to tumor cells following vitamin D supplementation. Finally, to capture the broader organization of the TME, we conducted an agnostic cellular neighborhood (CN) analysis (Supplementary Figure S11B–C), which identified seven distinct neighborhoods representing tumor, stromal, and immune compartments, including tumor-dominant, tumor-intermediate, tumor-stroma interface, stroma-dominant, immune-stromal mixed, CD4 effector-enriched, and CD8 memory-enriched regions. These neighborhoods ranged from tumor-dominant regions to stromal areas, with the tumor-stroma interface in between, and included distinct immune-enriched niches. In both arms, immune-enriched neighborhoods increased in post-treatment resections. However, this shift was more pronounced in the vitamin D arm, with larger increases in CD4 effector-enriched, CD8-memory-enriched, and immune-stromal mixed neighborhoods, along with a greater decrease in stromal and tumor-stroma interface clusters, as compared to placebo (Supplementary Figure S11D). Taken together, these spatial analysis results suggest that vitamin D supplementation is associated with a redistribution of CD3+CD8+ cytotoxic T cells through the tumor microenvironment, with a more even distribution from the invasive edge to the tumor center and greater representation within tumor-adjacent and immune-enriched regions, consistent with a more immunologically active microenvironmental state.

DISCUSSION

This randomized, placebo-controlled trial of preoperative high-dose vitamin D3 versus placebo in patients with stage I-III colon cancer aimed to characterize VDR and CYP27B1 expression and changes in the TME of primary colon tumors in response to vitamin D supplementation (secondary endpoint). While previous studies have reported VDR expression in a variety of cell(43–45) and tumor types(46–48), this is the first study to show VDR protein distribution in multiple cell types simultaneously in different tumor compartments (intraepithelial versus stromal regions) and different locations (tumor center versus invasive edge) in direct response to vitamin D supplementation within a randomized clinical trial. Notably, VDR was predominantly expressed by tumor cells, followed by immune and stromal cells, with wide variance between tumors, indicating the importance of not restricting the examination of VDR expression to only tumor epithelial cells. Although vitamin D supplementation in healthy patients(49,50) and in those with medical conditions, such as disturbed glucose homeostasis(51) and colorectal adenomas(52), has been linked to alterations in VDR target gene expression, few studies have investigated its effect on VDR protein expression(53–55). Furthermore, our results showed that vitamin D supplementation was linked to significant changes in the relative abundance of VDR protein expression.

This randomized trial also found a statistically significant increase in CD8⁺ cytotoxic T cell density, particularly the density of CD8⁺ memory cytotoxic T cells, which have prognostic significance in CRC(56–59), in response to vitamin D supplementation. Similar observations have been reported previously in other tumor types(33,35), suggesting a broader relevance of vitamin D-induced modulation of CD8+ T cell populations in cancer. The preferential increase in CD8+ memory T cells raises the possibility that vitamin D may contribute to the establishment of a more sustained immune response within the tumor microenvironment, rather than a transient activation. Further exploratory analysis showed increased expression of IFN-γ-associated chemokines involved in T cell recruitment, without corresponding increase in IFNG expression, suggesting indirect or stromal-mediated effects of vitamin D. We also noted higher expression of select immune checkpoint genes, such as HAVCR2 (TIM-3) and TIGIT, in stromal regions following vitamin D treatment.

These findings not only improve our understanding of the dynamic relationship between vitamin D and the colon TME but also provide a promising rationale to investigate vitamin D immunomodulation in the treatment of patients with colon cancer. For example, neoadjuvant vitamin D supplementation may potentially enhance tumor responses in patients with resectable colon cancer, both alone and in combination with neoadjuvant immune checkpoint blockade, which has been shown to induce high rates of major pathologic responses in patients with MSI-H and MSS colon cancer in several clinical trials, including the NICHE trials(60–62). Moreover, although a recent randomized phase III clinical trial, SOLARIS, found no statistically significant benefit of high-dose vitamin D on progression-free survival in patients with previously untreated metastatic CRC(18), a potential benefit was seen in patients with left-sided primary tumors. In comparison to right-sided CRC, left-sided CRC is characterized by a “cold” immune microenvironment with fewer T cells and more immunosuppressive myeloid cell subtypes(63). The higher expression of VDR seen in left-sided as opposed to right-sided primary tumors in this preoperative clinical trial may partially explain the benefit of vitamin D in patients with left-sided tumors in SOLARIS. In addition, these observations may also have implications for patient selection. In particular, tumor sidedness and baseline VDR expression may help identify patients more likely to benefit from vitamin D treatment. Future clinical trials could prospectively test the efficacy of vitamin D specifically in patients with left-sided CRC, with comprehensive correlative analyses of baseline, on-treatment, and post-treatment biopsies to examine changes in the TME along with other biological endpoints. Taken together, these results justify further study of immune-related vitamin D activity in CRC.

In addition to its effect on immune cell density, vitamin D supplementation altered T cell organization at multiple spatial scales in the colon TME. We found a distinct and homogeneous distribution of VDR expression and cytotoxic T cells across the tumor center and invasive front in patients randomized to vitamin D, but not placebo, suggesting that localized cell-cell interactions and VDR-driven paracrine signaling pathways may regulate tumor cell and T cell co-localization. A previous study found that vitamin D3 induced T cell attraction to epidermal keratinocytes in a dendritic cell-dependent manner, demonstrating that VDR-driven T cell and epithelial cell co-localization can be regulated by an intermediary cell type(64). Our finding of a trend towards increased myeloid cell abundance in vitamin D-treated tumors suggests that myeloid cells, and macrophages in particular, may represent an important regulator of VDR-induced TME remodeling, potentially modulating the localized immune response through chemokine production. While our findings provide important insights into TME spatial reorganization induced by vitamin D, future research will be required to identify the specific signaling pathways and mechanisms driving this response.

Our study has several notable strengths. While prior cross-sectional studies have reported an association between vitamin D status and TME characteristics, we conducted a randomized, placebo-controlled clinical trial to directly interrogate the impact of vitamin D supplementation on changes in VDR/CYP27B1 expression and the TME. Our dataset was comprehensively annotated with relevant clinicopathologic and molecular data, allowing us to adjust for potential confounding factors. Our use of both protein and RNA-based tissue profiling assays enabled a multimodal assessment of VDR signaling activity and elements of the TME. In particular, our mIF panels were custom designed to evaluate key markers within the vitamin D signaling pathway and enabled the phenotyping of immune cell subsets, overcoming inherent limitations of single-plex immunohistochemistry studies. Our approach using supervised machine learning algorithms at the single-cell level not only improved the accuracy of phenotype determination but also allowed for the reporting of VDR and CYP27B1 expression across individual cells, which is beyond the scope of simple pixel or threshold classification methods. Our measurements, distinguished by their precision and standardization, serve as a valuable benchmark for future CRC tissue profiling studies. Furthermore, our tissue-based approach allowed us to uncover multiple ways in which vitamin D alters the spatial organization of cytotoxic T cells and tumor cells in the colon microenvironment—findings which could not have been identified through dissociative single-cell profiling approaches.

Despite the above strengths, it is critical to recognize some limitations. While this study yielded valuable insights, the modest sample size and the inability to analyze specimens from all patients limited statistical power. Thus, we hope to validate our findings in future larger studies. Additionally, while baseline immune cell densities were not statistically different between treatment arms, some numerical differences were observed. While these are unlikely to account for the observed treatment-associated changes, they warrant consideration when interpreting the results in the context of a modest sample size. In addition, the variations in the duration of vitamin D supplementation and time to surgery may have potentially impacted our results, complicating the interpretation of treatment effects. Finally, while our mIF assays provide high sensitivity and quantitative results, they included a limited number of markers and could only assess pre-selected cell types; conversely, our spatial transcriptional profiling data assessed the entire transcriptome but at regional, and not single-cell resolution, and was limited to post-treatment resection specimens, precluding direct assessment of pharmacodynamic changes. Further studies are needed and are currently underway to investigate additional factors and cell types relevant to vitamin D signaling.

In conclusion, our study demonstrates that vitamin D supplementation in individuals with stage I-III colon cancer induces changes in vitamin D pathway signaling that may reflect an anti-tumor effect within the colon TME. These findings emphasize the therapeutic potential of vitamin D in influencing the dynamic interplay between the immune response and tumor biology in colon cancer. Further investigations are warranted to understand the underlying mechanisms of action and offer new therapeutic strategies that leverage vitamin D supplementation.

METHODS

Trial design and participants

This was a double-blind, placebo-controlled, randomized clinical trial conducted at Dana-Farber/Brigham and Women’s Cancer Center (DF/BWCC) to evaluate the transcriptional and immune targets of vitamin D and the vitamin D receptor (VDR) in patients with stage I-III resectable colon cancer who were scheduled to have their primary tumor surgically resected. This study reports the results of the secondary endpoint of assessing changes in the TME of primary colon tumors in response to vitamin D supplementation, and toxicity of high-dose vitamin D. Patients over the age of 18 were eligible for enrollment if they had 1) a histologically confirmed diagnosis of localized stage I-III adenocarcinoma of the colon with surgical resection planned, 2) no prior radiation therapy or systemic treatment for colon cancer, 3) serum calcium (corrected for albumin level) ≤ 1x institutional upper limit of normal, and 4) an Eastern Cooperative Oncology Group (ECOG) performance status (PS) ≤ 1. Patients were excluded if they 1) regularly used supplemental vitamin D totaling ≥ 2,000 IU/day in the past year, 2) used chronic oral corticosteroid therapy, lithium, phenytoin, quinidine, isoniazid, and/or rifampin (all of which can cause vitamin D depletion), 3) used thiazide diuretics (i.e., hydrochlorothiazide), which can cause hypercalcemia, and were unwilling or unable to discontinue or switch to an alternative anti-hypertensive agent, 4) had pre-existing hypercalcemia, 5) had known active hyperparathyroid disease or other serious disturbances of calcium metabolism in the past 5 years, and 6) had a history of symptomatic genitourinary stones within the past year. Full eligibility criteria are listed in the trial protocol (Supplementary Material).

The institutional review board (IRB) from DF/BWCC approved this study and all participants provided written informed consent. The trial was overseen by the DF/BWCC Data and Safety Monitoring Committee at least every six months during the study. This report followed the Consolidated Standards of Reporting Trials (CONSORT) reporting guideline. (ClinicalTrials.gov Identifier: NCT02172651).

Randomization and Procedures

Patients were randomized 1:1 to receive high dose vitamin D3 or placebo through a computerized block randomization with a block size of 2. Patients in the investigational arm received five 10,000 IU capsules of vitamin D3 (for a total of 50,000 IU) once daily for seven days followed by one 10,000 IU capsule once daily, whereas patients in the placebo arm received five capsules of placebo once daily for seven days followed by one placebo capsule once daily, until their primary tumors were resected. Surgical resection occurred any time after the first seven days of supplementation, but no later than 28 days after initiating study treatment (Figure 1, Supplementary Figure S1).

Correlative Tissue and Blood Biospecimens

Archival tumor tissue from diagnostic pre-treatment biopsies were collected at baseline, while post-treatment tumor specimens were obtained at the time of surgical resection. Tissue samples were formalin-fixed, paraffin embedded and sectioned at 4μm. Blood samples were collected at baseline prior to the start of supplementation, and then immediately prior to surgery (after completion of supplementation) and banked for plasma 25(OH)D measurement (Figure 1). Plasma 25(OH)D levels were measured in a single batch by radioimmunoassay (DiaSorin, Inc.) at Heartland Assays, Inc. (Ames, IA). Blinded duplicates totaling 10% of the total number of samples were included for quality control, with a mean coefficient of variation of 7.3%. All laboratory personnel were blinded to treatment assignment. When duplicate values were available, the average of the values was used for analysis.

Multiplex immunofluorescence protocols

Two spatially-resolved immunofluorescence (mIF) assays were developed to assess nuclear VDR and cytoplasmic CYP27B1 expression in tumor epithelial cells, immune cell populations, and stromal cells at the single cell level. This approach also enabled the characterization of T cell subsets(65). Additionally, the panels included: CD45 (pan-immune cell marker), CD3 (pan-T cell marker), CD4 (helper T cells), CD8 (cytotoxic T cells), and CD68 and CD163 (macrophage markers), FoxP3 (regulatory T cell activity), CD45RO (memory T cells), as well as 2-(4-amidinophenyl)-1H-indole-6-carboxamidine (DAPI) as a nuclear marker and cytokeratin to identify epithelial cells (Figure 1, Supplementary Table S12).

Initially, all antibodies were optimized using chromogenic single-plex staining protocols, which were followed by single-plex immunofluorescence (Supplementary Figure S12). The antibodies were then integrated into a mIF assay that required several steps, including antigen retrieval, antigen detection, and fluorescent labeling with tyramide signal amplification. These stages were conducted on 4μm sections of formalin-fixed paraffin-embedded samples using a Leica BOND RX Research Stainer (Leica Biosystems, Buffalo, IL, RRID:SCR_025548) (Supplementary Figure S13).

Digital image analysis and supervised cell phenotype classification

Digital images were scanned at 200x magnification to examine stained slides using an automated multispectral imaging system (PhenoImager HT, Akoya, Hopkinton, MA, RRID:SCR_023772). A trained pathologist visually reviewed each digital image after image acquisition and spectrum unmixing to confirm the presence of invasive colon cancer and exclude non-neoplastic epithelium and tissue processing artifacts. Up to seven regions of interest were randomly selected from each biopsy, while in resection, three were chosen from the tumor center and three from the periphery to account for heterogeneity (Supplementary Table S13–S16, Supplementary Figure S14). After quality control processes, images were then segmented into tumor epithelial and stromal areas using supervised machine learning (Akoya inForm Tissue Analysis Software2.4.1, Akoya, RRID:SCR_019155) (Supplementary Figure S15A–B).

Following the detection and segmentation of single cells, supervised machine learning was employed to identify main cell types based on a combination of cytomorphology and protein expression at subcellular resolution (Supplementary Figure S15C–E). This analysis allowed the characterization of VDR and CYP27B1 expression, along with the identification of the following phenotypes: CD3⁺CD4⁺ T cells, CD3⁺CD8⁺ T cells, CD3⁺CD4⁻CD8⁻ T cells, macrophages (CD68⁺CD163⁺), other immune cells (CD45⁺CD3⁻CD68⁻CD163⁻), tumor cells (CK⁺), and stromal cells (CK⁻CD45⁻CD3⁻CD4⁻CD8⁻CD68⁻CD163⁻) (Figure 1, Supplementary Figure S15E). R v.4.0 (R Foundation for Statistical Computing, Vienna, Austria, RRID:SCR_001905) was employed to analyze the resulting data.

Quantification of VDR and CYP27B1 expression, immune cell densities, and spatial colocalization

VDR and CYP27B1 expression were separately measured in each cell type. Next, fractional abundance was computed. Immune cell densities were also quantified by tissue compartment and in the combined tissue area.

Spatial relationships between tumor and immune cells were assessed using multiple complementary approaches. Cell-cell interaction was calculated using nearest neighbor distance (NND) and the Gcross function, implemented in the spatstat package v.2.2.0 in R, as previously described(65,66). To characterize local cellular composition, tumor centered k-nearest neighbor (kNN) analysis (k=20) was performed to quantify the fractional abundance of surrounding cell types within each tumor cell neighborhood(42). In parallel, cellular neighborhoods were defined by clustering local cellular composition profiles, enabling the identification of recurrent microenvironmental content across samples(42) (Figure 1, Supplementary Figure S11B–D).

Spatial transcriptional profiling specimen processing

Formalin-fixed, paraffin-embedded slides from post-treatment resections were processed following the fully automated GeoMx-NGS RNA assay (Bruker Technologies) on the Leica BOND RX platform. Slides underwent baking, deparaffinization, heat-induced antigen retrieval (ER2 for 20 min at 100 °C), and proteinase K (0.1 μg/ml) digestion, followed by hybridization with the Human Whole Transcriptome Atlas (WTA) probe mix. Sections were then stained with fluorescent-labeled antibodies targeting tumor cells (pan-cytokeratin [CK]), immune cells (CD45), and nuclei (SYTO 13).

Slides were imaged on the GeoMx Digital Spatial Profiler (Bruker Technologies, RRID:SCR_021660), and regions of interest were selected within tumor areas. Each region was segmented into areas of illumination using a CK-guided contouring approach to define non-overlapping compartments: tumor epithelium (center), adjacent stroma within 1–25 um (ring 1), and stroma within 26–50 um (ring 2) (Figure 3A). Photocleaved oligonucleotide barcodes were collected in 96-well plates and subjected to library preparation. Pooled libraries were analyzed using an Illumina NovaSeq X Plus sequencer (RRID:SCR_024568) with paired-end read mode. Following demultiplexing, sequencing data were processed to generate AOI-level expression counts.

Spatial transcriptional data processing

GeoMx DSP data were processed in R using the GeoMxTools package (v3.6.2, RRID:SCR_023424). Raw count data (DCC files), probe annotation (PKC file), and sample metadata were combined into a GeoMxSet object. AOIs were excluded based on predefined sequencing and quality metrics, including poor alignment performance (<80% read trimmed, stitched, or aligned) and low sequencing saturation (<50%) (Supplementary Figure S16A–E). Each AOI had at least 20 nuclei, and AOIs with relatively low nuclei counts showed no evidence of reduced probe counts or gene detection compared to other AOIs, suggesting no quality issues related to nuclei count. Gene expression values from AOIs that passed quality control thresholds were normalized using upper-quartile scaling and were log₂-transformed prior to analysis.

Derivation of Vitamin D-associated CAF signature

To assess VDR transcriptional activity within stromal AOIs, we evaluated a gene set containing all downregulated genes within a previously published gene set reflecting transcriptional changes that occurred when purified CRC-derived cancer-associated fibroblasts were treated with 1,25(OH)₂D₃ in vitro(25). This set of 23 genes, representing genes repressed by active VDR signaling, is provided in Supplementary Table S8.

Cell type deconvolution of spatial transcriptomic data

Cell type composition was estimated using the GeoMx RNA deconvolution workflow with the SafeTME reference profile from SpatialDecon (36), derived from curated tumor microenvironment cell types and optimized for the GeoMx WTA panel. Cell type deconvolution was performed on normalized gene expression data using the standard linear deconvolution model implemented in SpatialDecon (v1.12.3, RRID:SCR_026836). Gene expression profiles from each AOI were modeled as a linear combination of reference cell type signatures to estimate relative cell type proportions. Only genes overlapping between GeoMx dataset and the SafeTME reference were included in the analysis. Estimated cell abundance scores were computed per AOI, normalized to relative proportions, and used for downstream analysis.

Outcomes

The primary study objective was to identify VDR binding sites in primary colon tumors from patients treated with preoperative high-dose vitamin D3 versus placebo (analysis in process). The secondary study objective, and the topic of this paper, was to characterize changes in the TME of primary colon tumors from patients treated with high-dose vitamin D3 versus placebo. Additional secondary objectives include evaluating the toxicity of preoperative high-dose vitamin D supplementation (reported in this paper), evaluating the transcriptome and correlating expressed genes with identified VDR binding sites (analysis in process), and identifying VDR binding sites in the immediate vicinity of known CRC loci (analysis in process).

Statistical analysis

Patient demographic and clinicopathologic characteristics at baseline according to randomization arm were summarized as descriptive statistics. Continuous variables were presented as median with IQRs, and categorical variables as the number and percentage (n, %). Comparison of two randomized groups were conducted using Fisher's exact test for categorical variables and the Wilcoxon rank-sum test or Kruskal-Wallis test for continuous variables. Changes in plasma 25(OH)D levels were compared between two arms using Fisher’s exact test for categorical vitamin D status and the Wilcoxon rank-sum test for continuous values at each timepoint. Wilcoxon-signed rank test was used for within-group pre- vs. post-treatment comparisons. The correlation of baseline and change in plasma 25(OH)D levels with clinicopathologic factors were assessed using the same statistical methods.

Data from multiple regions of interest (ROIs) for each patient were averaged to assess cellular components of the TME in CRC. Box plots, with medians and IQRs, were used to visualize patient-level data. To model the ROI-level data, we generated a linear mixed-effects model, with cellular components expression level as the dependent variable, patient ID as a random effect, and randomization arm as a fixed effect.

Then, we examined whether baseline nuclear VDR/CYP27B1 protein expression differed by clinicopathologic characteristics, including tumor location, and grade of differentiation. Baseline nuclear VDR and CYP27B1 protein expressions were also assessed by averaging ROIs for each patient. Box plots, with medians and IQRs, were used to visualize patient-level data. Patient-level VDR/CYP27B1 expression were compared by categorized levels of clinicopathologic factors using the Wilcoxon rank-sum test or Kruskal Wallis test.

To determine whether the impact of treatment on VDR/CYP27B1 expression differed by randomization arm, we generated a linear mixed-effects model including three fixed terms: sample timepoint, randomization arm, and their interaction term, and assessed the p-value of the interaction term. Box plots, with medians and IQRs, used to visualize patient-level data.

We then assessed whether cell density or percentage differed between pre-treatment and on-treatment biopsies within each treatment arm. To model the ROI-level data, we generated a linear mixed-effects model, with (1) tumor cell and immune cell densities using the VDR mIF panel, (2) specific T cell subset densities using the T cell mIF panel, and (3) co-localization dynamics between tumor and immune cells (Gcross, NND, kNN) as the dependent variables, patient ID as a random effect, and biopsy timepoint (pre-treatment and post-treatment) as a fixed effect. To determine whether the impact of treatment on these dependent variables differed by treatment arm, we generated a linear mixed-effects model including three fixed terms: biopsy timepoint (pre-treatment and post-treatment), treatment arm, and their interaction term, and assessed the p-value of the interaction term of (1) tumor cell and immune cell densities using the VDR mIF panel, (2) specific T cell subset densities using the T cell mIF panel, and (3) co-localization dynamics between tumor and immune cells (Gcross, NND, kNN).

To explore the impact of potential confounders on the response of T cell subsets densities deploying T cell mIF panel to treatment, we adjusted for MSI status, grade of differentiation, baseline BMI, Tstage, and primary tumor site by adding each variable as the fixed term in the univariate linear mixed effect model above. Due to limited sample size, each covariate was initially included in a separate linear mixed effect model, respectively. To assess the robustness of the findings, we also performed an exploratory analysis by including all covariates in a single multivariable linear mixed effect model, despite the sample size constraints.

To further evaluate spatial configuration of immune cells in post-treatment specimens, we run a linear mixed-effects model including three fixed terms: spatial configuration (invasive front vs. tumor center), randomization arm, and their interaction term, and assessed the p-value of the interaction term. Patient-level data were visualized using box plots, with medians and IQRs reported.

All figures were generated using R software version 4.4.1. All statistical analyses were conducted using SAS 9.4 (SAS Institute, Cary, NC, RRID:SCR_008567), except for the linear mixed-effects model that used lme4 R package(67) (version 1.1–35.5, RRID:SCR_015654), and for standardized mean differences that used tableone R package. Statistical tests were performed using a two-sided significance level of P < 0.05.

STATEMENT OF SIGNIFICANCE.

Colorectal cancer remains a leading cause of cancer-related death with limited long-term survival. We evaluated the effects of short-term, preoperative high-dose vitamin D supplementation on the colon cancer microenvironment in a randomized, placebo-controlled trial. Our findings support vitamin D as an immune modulator and may inform future treatment strategies.

DATA AVAILABILITY

The data generated in this study are available upon request at the time of publication from the corresponding author.