The Cancer Genome Atlas Colonic adenocarcinoma (TCGA-COAD) RNA sequence data from AA (n = 64) and EA (n =284) groups was downloaded as unstranded STAR (raw) counts and processed TPM by using TCGAbiolinks R/Bioconductor package (19). The following clinical metadata variables were also downloaded from TCGA: 1.) age at time of collection (years); 2. sex (male/female); self-identified ancestry (African/European); COAD/READ tumor location (right, left; COAD/READ stage (I,II,III,IV); and MSI status. MSI status was available for 327 out of 348 samples (20). For 17 of 21 samples without MSI-status values, the Microsatellite Analysis for Normal Tumor InStability (MANTIS) scores in the TCGA-COAD-READ database were used to categorize MSI status (21). Samples with MANTIS scores < 0.4 were categorized as MMR-p/MSI-L+MSS and scores ≥ 0.4 were categorized as MMR-d/MSI-H. The raw counts of 15 AA and 18 EA COAD samples from the SUNY Stonybrook/Downstate medical centers were downloaded from Gene Expression Omnibus (GEO) with the accession number GSE146009 (7). The COAD data in GSE146009 was downloaded from 15 AA and 18 EA samples, which were annotated with respect to ancestry but not for MMR/MSI status. The raw counts were used as input for identifying differentially expressed genes (DEGs) using edgeR (22). After using edgeR to normalize the raw counts, the resulting “cpm” counts were used as input into the wilcox.test function in R (v4.0.2) as previously described (23). The threshold for identifying DEGs was the absolute value│log₂ fold change│≥ 1 and adjusted p-value <0.05 The consensus molecular subtypes (CMS) labels reported for the TCGA-COAD-READ dataset based on CMS network and CMS Random Forest (RF) were downloaded from cms_labels_public_all.txt - syn4978511 - Files (24). The abundance of tumor associated T-cells was estimated using CIBERSORT analysis of the TCGA-COAD-READ RNA-sequence data (25).

Assembly of 134 de-identified adult (age > 18) human COAD/READ FFPE tissue samples archived between 2012 and 2024 from three US medical institutions, Stony Brook University Hospital (Stony Brook, NY), New York City Health + Hospitals (NYCH+H)/Kings County Hospital (KCH, Brooklyn, NY), and Henry Ford Health Center (Detroit, MI) was approved by the Stony Brook Institutional Review Board (sIRB2024-0020) with reliance forms reviewed by the Institutional Review Boards for Henry Ford Health Center and Michigan State University. No reliance form was required for NYCH+H/KCH since the research protocol included an honest broker that oversaw HIPAA compliance, was conducted with waiver of consent and was determined to be not human research by its Institutional Review Board (IRB1949860). Only initial surgical resections of treatment-naïve sporadic COAD-READ (excluding inflammatory bowel disease-associated and hereditary COAD-READ syndromes) were selected for analysis. The samples were linked to deidentified clinical metadata curated from electronic medical records by physicians at each of the three medical centers using a common data dictionary as previously described (26). The variables collected for the metadata included: 1.) age at the time of the sample collection (years), 2.) sex (male/female); 3.) ancestry (AA vs. EA) based on self-identification, 4.) ethnicity (all non-Hispanic); 5.) body mass index (BMI, kg/m²); 6.) diabetes mellitus status (type 1 diabetes, type 2 diabetes (T2DM), no diabetes); 7.) smoking (current, former, never); 8.) COAD tumor location (right defined as cecum, ascending colon, hepatic flexure, transverse colon; left defined as including splenic flexure, descending colon, sigmoid, rectum); 9.) COAD stage (1-4); 10.) MMR-d/MSI-H vs. MMR-p/MSI-L+MSS status classified primarily by immunohistochemistry (IHC); 11.) insurance status (Commercial/Medicare; Medicaid Mgd; Medicaid/Self-Pay).

Total RNA was extracted from 5 µm COAD FFPE curls using the RecoverAll™ Total Nucleic Acid Isolation Kit (Thermo Fisher Scientific Inc, Waltham, MA) according to the manufacturer’s recommendation, except paraffin was removed by xylene washes, the protease digestion was extended to 3 hours at 50°C. 200–500 ng of total bulk RNA was reverse transcribed using Superscript IV™ VILO™ cDNA kit (Thermo Fisher Scientific Inc, Waltham, MA) according to the manufacturer’s recommendation qPCR was performed using the Applied Biosystems QuantStudio 3 Real Time PCR System (Thermo Fisher Scientific Inc., Waltham, MA). The 20 μl dual probe (target immune gene/reference gene) PCR reactions included 1-2 μl cDNA (corresponding to 25 ng of RNA), 1 μl 1× TaqMan Universal PCR master mix, 1 μl for each primers/pre mix (for target and reference gene) The reactions, run in triplicate, were incubated in a 96-well optical plate at 95 °C for 10 min, followed by 40 cycles of 95 °C for 15 s and 60°for 10 min. The threshold cycle (Ct) was defined as the fractional cycle number at which the fluorescence passes the fixed threshold. The Ct data were determined using default threshold settings. Taqman^® primer probe set IDs for the target genes were: CXCL10 (HS00171042), CD45 (HS04189704) and CD3D (HS00174158). The Taqman^® primer probe set IDs for the reference genes were: RAB7A (HS01115139) and C1ORF43 (HS00367486). A previous study reported that RAB7a and C1ORF43 were best suited for normalizing RT-qPCR assays of COAD/READ samples (27), particularly because of the low covariance exhibited by these two reference genes. These commercial probe/primer sets have been used extensively in previous publications including the study evaluating the two COAD reference genes (27). The ΔCt were calculated as reference gene Ct – target gene Ct to estimate the log transformation of the ratio of target gene/reference gene templates in the reactions.

TCGA CXCL10 outcomes were expressed as CXCL10 log₂TPM. The RT-qPCR CXCL10 values were expressed as C1ORF43 Ct - CXCL10 Ct and RAB7A Ct - CXCL10 Ct. CD45 values were expressed as C1ORF43 Ct – CD45 Ct and RAB7A Ct – CD45 Ct, and CD3D values were expressed as C1ORF43 Ct – CD3D Ct and RAB7A Ct – CD3D Ct. Spearman’s correlation was used to examine the linear correlation between CXCL10 and continuous variables such as age, CD45 and CD3D values. The Wilcoxon rank sum test (for variables with 2 levels) or Kruskal-Wallis test (for variables with ≥ 3 levels) was utilized to examine the marginal difference in outcomes among categorical variables. For the Kruskal Wallis test, a Dunn’s post-hoc test was used to compare individual groups with each other. Multiple linear regression models were then utilized to examine whether there was a difference in the ancestry level or MMR status after adjusting for COAD/READ location and COAD/READ stage. 134 independent human COAD/READ FFPE tissue samples are expected to have 90% power to detect an increase in R² being 6.5% while the R² of model using covariates alone being 11.5% based on a multiple regression full-versus-reduced-model F-test with a Type I error rate of 0.05 (28). Both R²s in the sample size justification are estimated from TCGA data. Statistical analysis was performed using GraphPad Prism 10 (for some univariate analyses), by using the cor.test () function in R version 4.44 to calculate correlation coefficients and SAS 9.4 (SAS Institute Inc., Cary, NC). Significance level was set at 0.05.

As shown in Figure 1 , CXCL10 was identified as AA vs. EA differentially expressed genes (DEGs), which were expressed at a lower level in AA vs. EA COAD/READ samples in both the TCGA-COAD-READ (64 AA vs. 284 EA) and a smaller SUNY Downstate/Stony Brook (15 AA vs. 18 EA) bulk RNA sequence datasets, using edgeR (22, see Supplementary Table S1 ).

MMR/MSI-status of the TCGA-COAD samples was obtained by downloading the TCGA-COAD-READ metadata and the MSI status reported previously for 327/348 samples (20). Of the 21 samples lacking MSI status values, 17 were categorized by using the MANTIS score (21). No information on the MMR/MSI status was provided for the smaller SUNY Downstate/Stony Brook RNA-seq dataset (7). The distribution of consensus molecular subtype (CMS1, CMS2, CMS3, CMS4, No Label) previously assigned to the TCGA-COAD-READ samples (24) for:1.) AA MMR-p/MSI-L+MSS; 2.) AA MMR-d/MSI-H; 3.) AA MMR undetermined; 4.) EA MMR-p/MSI-L+MSS; 5.) EA MMR-d/MSI-H vs. MMR-p/MSIL+MSS are shown in Figure 2 . CMS1 has been associated with high expression of immune-related genes and MMR-d/MSI-H (22). CMS2 has been associated with a differentiated epithelial cell phenotype. CMS3 has been termed the metabolic subtype because of dysregulated metabolic genes. CMS4 is associated with a high stromal content. Some of the COAD samples could not be readily assigned to a single CMS and have been termed No Label. Thirty-seven (13%) of the 284 EA samples were labeled as CMS1. Thirty-one (84%) of the 37 EA CMS1 samples were also MMR-d/MSI-H. Only two (3%) of 64 AA samples were labeled as CMS1. Only one (50%) of two AA CMS1 samples was also MMR-d/MSI-H.

To reduce the number of false positive DEGs, the Wilcoxon rank sum test was used to determine the overlap between the AA vs. EA DEGs and the MMR-d/MSI-H vs. MMR-p/MSI-L+MSS DEGs (23). The number of AA vs. EA DEGs was reduced to 39 from 420, and the number of MMR-d/MSI-H vs. MMR-p/MSI-L+MSS DEGs was reduced to 738 from 2177 (see Supplementary Table S1 ). The overlap between the 39 AA vs. EA DEGs and the 738 MMR-d/MSI-H vs. MMR-p/MSI-L+MSS DEG lists consists of 7 genes (CXCL10, ALOX15B, IDO1, HCAR2, MARCO, OR2I1P and MTND4P24. Six of seven DEGs were decreased in the AA group and increased in the MMR-d/MSI-H group (CXCL10, ALOX15B, IDO1, HCAR2, MARCO and OR2I1P). The first five DEGs have been linked to macrophage function and in some instances with COAD/READ (29–34). OR2I1P is a pseudogene with unknown function. MTND4P24, which is increased in AA vs. EA and decreased in the MMR-d/MSI-H vs. MMR-p/MSI-L+MSS groups, is a pseudogene with unknown function.

Further analysis of the effects of AA and MMR-d/MSI-H status focused on CXCL10 log₂TPM as the outcome because this gene plays a key role in regulating COAD/READ tumor microenvironment in MMR-d/MSI-H samples (14, 15). Differences in CXCL10 values were significantly associated with ancestry, MMR/MSI status, COAD/READ location, COAD/READ stage, but not sex (see Table 1 ). Age was not significantly correlated with CXCL10 levels. Multiple linear regression models were used to examine associations to CXCL10 expression while adjusting for COAD/READ stage, and with and without COAD/READ location, because of the number of missing location values. As shown in Table 1 , CXCL10 values were lower in AA vs. EA (p -value < 0.0001) and higher in MMR-d/MSI-H vs. MMR-p/MSI-L+MSS (p-value < 0.0001), while controlling for COAD/READ stage and location. Neither COAD/READ stage nor location were significant, while controlling for ancestry and MMR/MSI status. Although MMR-d/MSI-H status has been previously correlated with right colon location (35), no significant multicollinearity was detected between the co-variables (results not shown). Estimated differences in CXCL10 due to ancestry across MMR/MSI status were not significant (see Table 2 ). Consistent with CXCL10’s role as a T-cell attractant was the significant correlation (Spearman’s correlation coefficient r= 0.44, p-value <0.0001) detected between CXCL10 log ₂ TPM values and T-cell abundance estimated by CIBERSORT (see Supplementary Figure S1 ).

Exploratory analysis comparing CXCL10 log₂TPM values to CMS labels within each ancestry group (see Supplementary Figure S2 ), detected no significant difference between CMS labels within the AA group. In contrast, significant differences were detected between CMS labels in the EA group (p-value <0.0001). Dunns post-hoc test detected significantly increased CXCL10 expression in the EA CMS1 group compared with both CMS2 (p-value <0.0001) and CMS3 groups (p-value<0.0001), but not CMS4 or No Label groups. Also, CXCL10 log₂ TPM values were significantly higher in EA CMS4 compared with both CMS2 (p-value<0.0001) and CMS3 (p-value=0.0002) groups, but not CMS1 or No Label groups.

Because only 9 of the AA TCGA COAD/READ samples were MMR-d/MSI-H, an independent set of 134 COAD FFPE samples was assembled from three medical centers that was composed of roughly equal numbers of AA MMR-d/MSI-H, AA MMR-p/MSI-L+MSS, EA MMR-d/MSI-H, EA MMR-p/MSI-L+MSS. Amplifiable RNA by RT-qPCR was recovered from 118 (88%) of the samples. CXCL10 expression was normalized relative to two reference genes as C1ORF43 Ct – CXCL10 Ct and RAB7A Ct – CXCL10 Ct (see Figure 3 ). Loss of 12% of the original 134 samples resulted in the expected power being reduced from 90% to 87%. Univariate analyses of both values confirmed that CXCL10 values were significantly higher in MMR-d/MSI-H vs. MMR-p/MSI-L+MSS (see Tables 2 , 3 ). In contrast to the TCGA-COAD CXCL10 log₂ TPM values, the RT-qPCR CXCL10 values relative to both reference genes trended slightly higher in AA vs. EA but these differences did not reach significance. Differences in C1ORF43 Ct – CXCL10 Ct values were associated with COAD/READ location but not with COAD/READ stage, and differences in RAB7A Ct – CXCL10 Ct values were associated with COAD/READ stage but not with COAD/READ location. Neither RT-qPCR CXCL10 values were significantly associated with age or sex. When the same multivariable model used to analyze the TCGA CXCL10 log₂ TPM results was applied to the RT-qPCR results ( Tables 3 – 6 ), AA C1ORF43 Ct - CXCL10 Ct (p = 0.0438) and RAB7A Ct – CXCL10 Ct (p-value = 0.0497) values were higher than EA values, MMR-d/MSI-H C1ORF43 Ct – CXCL10 Ct (p-value = 0.019) and RAB7A Ct - CXCL10 Ct (p-value = 0.093) values were higher or trended higher than MMR-p/MSI-L+MSS values, while controlling for COAD/READ stage and location.

Parallel RT-qPCR assays were conducted for CD45, a myeloid/general white cell marker, and CD3D, a T-cell marker in the independent set of COAD/READ FFPE samples (see Figure 3 ). In single cell RNA-sequence datasets, CXCL10 has been shown to be highly expressed in myeloid cell types (29). CXCL10 has been shown to be a T-cell attractant (36). Consistent with these reports are the strong positive correlations (see Figure 4 ) observed between C1ORF43 Ct - CXCL10 Ct and C1ORF43 Ct - CD45 Ct (Spearman correlation coefficient = 0.54, p-value < 0.0001), between RAB7A Ct - CXCL10 Ct and RAB7A Ct - CD45 Ct (Spearman correlation coefficient r =0.44, p-value < 0.0001), between C1ORF43 Ct- CXCL10 Ct and C1ORF43 Ct – CD3D Ct (Spearman correlation coefficient r = 0.51, p-value < 0.0001), and between RAB7A Ct - CXCL10 Ct and RAB7A Ct – CD3D Ct (Spearman correlation coefficient r = 0.34, p value < 0.0001).