Serum proteomics links suppression of tumor immunity to ancestry and lethal prostate cancer

There is evidence that tumor immunobiology and immunotherapy response may differ between African American and European American prostate cancer patients. Here, we determine if men of African descent harbor a unique systemic immune-oncological signature and measure 82 circulating proteins in almost 3000 Ghanaian, African American, and European American men. Protein signatures for suppression of tumor immunity and chemotaxis are elevated in men of West African ancestry. Importantly, the suppression of tumor immunity protein signature associates with metastatic and lethal prostate cancer, pointing to clinical importance. Moreover, two markers, pleiotrophin and TNFRSF9, predict poor disease survival specifically among African American men. These findings indicate that immune-oncology marker profiles differ between men of African and European descent. These differences may contribute to the disproportionate burden of lethal prostate cancer in men of African ancestry. The elevated peripheral suppression of tumor immunity may have important implication for guidance of cancer therapy which could particularly benefit African American patients.

Men of African origin bear the highest prostate cancer burden in the U.S. and globally^1,2,3. They are at an increased risk of developing fatal prostate cancer in the U.S and England⁴ and present with more aggressive disease in the Caribbean and sub-Saharan Africa^2,5. The reasons for the observed global prostate cancer health disparities are unclear but are likely related to an array of factors such as access to health care, lifestyle and environment, and ancestral and biological factors^6,7,8.

Previously, we and others described that tumor immunobiology differs between African-American (AA) and European-American (EA) prostate cancer patients^9,10,11,12. A tumor-specific immune-inflammation gene expression signature was more prevalent in prostate tumors of AA than EA patients¹¹. The occurrence of this signature in prostate tumors was associated with decreased recurrence-free survival¹³. Furthermore, regular use of aspirin, an anti-inflammatory drug, may reduce the risk of aggressive prostate cancer, disease recurrence and lethal disease in AA men^14,15. Combined, these findings suggest that inflammation and host immunity may contribute to prostate cancer progression but with notable differences between AA and EA men.

Ancestral factors can influence immune-related pathways¹⁶. Germline genetic variant prevalence and alternative splicing in immune-inflammation-related genes can show large differences amongst population groups^17,18,19. Hence, the immune-inflammation gene expression signature identified in the tumors of AA prostate cancer patients could be due to either tumor biology and the associated microenvironment, ancestral factors, or systemic differences in immune-oncology marker expression.

In this work, we test the hypothesis that a distinct systemic immune-inflammation signature exists in men of African ancestry that associates with prostate cancer using a large cohort of diverse men. Applying large-scale proteomics with Olink technology, we discover the up-regulation of circulating immune-oncological proteins that functionally relate to chemotaxis and suppression of tumor immunity and their association with West African ancestry and lethal prostate cancer. Our findings point to the clinical importance of a serum proteomic signature in prostate cancer patients that may affect men of African ancestry more so than other men.

Large-scale evaluation of immune-oncological proteins in the NCI-Maryland and NCI-Ghana prostate cancer studies

To investigate if men of African descent are differentially affected by a systemic immune inflammation, we utilized two case-control studies with large representations of men of African ancestry: the NCI-Ghana and NCI-Maryland Prostate Cancer Case-Control Studies. Characteristics of the participants in the two studies have been previously described^14,20. For our investigation, limited to men with quality-controlled (QCed) serum proteome measurements, the NCI-Ghana Prostate Cancer Case-Control study included 1143 men of whom 489 were cases and 654 were controls. Cases were older than controls (Supplementary Table 1). Controls had a slightly lower body-mass index (BMI) than cases. In addition, more controls were current smokers than cases (14% vs. 2%). High Gleason score (>7) was reported in 158 out of the 489 patients (32%). Participants in the NCI-Maryland Prostate Cancer Case-Control study included 1647 AA and EA men of whom 819 were cases and 828 were controls. Cases were slightly younger than controls. Cases and controls had similar BMI distributions. In addition, more cases were current smokers than controls (24% vs. 14%). High Gleason score (>7) was reported in 141 out of the 819 patients (17%). In the control population, Ghanaian men tended to be younger, to have a lower BMI, less likely to be diagnosed with diabetes, but with a similar prevalence of current smokers, when compared with AA and EA men.

We assayed 92 circulating immune-oncological proteins in a total of 3094 serum samples containing 1505 controls and 1432 cases along with 157 randomly selected blinded duplicates. To control for any batch effects, the serum samples were assayed in a random order along withthe 5% blind duplicates for intensity normalization (see Methods). Ninety-five percent of the samples passed stringent quality control, leaving 1482 controls (654 Ghanaian, 374 AA, and 454 EA) and 1308 cases (489 Ghanaian, 394 AA, and 425 EA) for our analysis (Supplementary Table 1). The average intra- and inter-plate coefficients of variation calculated based on duplicates were very low at 1.7% and 2.6%, respectively. In addition, the proportion of variance explained by an inter-plate batch effect was rather minimal for each of the serum proteins even before intensity normalization (Supplementary Fig. 1). Out of the 92 serum proteins, 61 had abundance levels above the lower limit of detection (LLOD) in 100% of the samples tested (Supplementary Table 2, Supplementary Fig. 2) and 78 proteins had levels above LLOD in >50% of the samples (Supplementary Fig. 2). Because 10 out of the 92 serum proteins were detected (i.e., had levels above LLOD) in <20% of the samples (Supplementary Fig. 2), only the remaining 82 proteins were included in our analysis (Supplementary Table 3). Next, we assessed how the 82 serum markers correlate with one another in Ghanaian, AA, and EA men with and without prostate cancer using Pearson’s pairwise correlation analysis (Fig. 1 and Supplementary Fig. 3). The top ten observed correlations for each population group are presented in Supplementary Table 4. Most of these relationships have not previously been described. Notably, epidermal growth factor levels strongly correlated with CD40L, a marker of activated T cells, in cases [AA (r = 0.87), and EA (r = 0.77)] and controls [Ghanaian (r = 0.71), AA (r = 0.83), and EA (r = 0.80)]. Interleukin-8 levels correlated consistently with circulating caspase-8 in cases [Ghanaian (r = 0.48), AA (r = 0.74), and EA (r = 0.73)] and controls [Ghanaian (r = 0.69), AA (r = 0.82), and EA (r = 0.80)]. Other robust correlations include interleukin-8 with MCP3, TNFRSF4 with TNFRSF9, and CD83 with TNFRSF9.

Pearson pairwise correlations were estimated for each serum protein pair in African American (a) population controls (n = 374) and (b) prostate cancer cases (n = 394). Source data are provided as a Source Data file.

Full size image

Clinical and socio-demographic characteristics are associated with immune-oncological proteins

Cytokine levels can be influenced by environmental exposures and disease. Therefore, we investigated the association between various socio-demographic and clinical characteristics (age, BMI, education, aspirin use, smoking, diabetes, and PSA) with serum levels of immuno-oncological proteins using a multivariable linear regression model with adjustment for multicomparison analysis (Fig. 2). We restricted this analysis to the control population in the NCI-Ghana and NCI-Maryland studies to exclude the potential confounding effect of prostate cancer in the analysis. Among the exposures, aspirin use, and blood PSA levels showed only few relationships with the profile of the 82 immune-oncology markers. Other exposures and several demographics showed more robust relationships.

The association of the 82 immuno-oncological proteins (as continuous variables) with age, BMI, education, aspirin use, smoking, diabetes, and PSA was assessed in men without prostate cancer using a multivariable linear regression test. P values were adjusted for multiple comparison. An analyte was considered significantly associated with clinical and socio-demographic covariables if the multivariable model yielded a false discovery rate (FDR)-adjusted P < 0.05 on the F-statistic. Analytes that did not have a significant association with any of the clinical/sociodemographic variables in at least one of the population groups are not presented in the heatmap. Blue represents a negative association while red represents a positive association. The significance level (FDR-adjusted two-sided P value-based) for each association is color-coded. Source data are provided as a Source Data file. TI tumor immunity, AA African American, and EA European American.

Full size image

Aging is known to impact the immune system and is a risk factor for many diseases including cancer²¹. In our analysis, aging was most consistently associated with the level of the analytes across the three population groups, showing a significant correlation with almost half of these circulating immune-oncological proteins. For example, PGF, CXCL9, Gal9, Gal1, CX3CL1, TNFRSF12A, CCL23, MMP7, DCN, MMP12, ADGRG1, and PTN positively associated with age in all three population groups. The top-ranked biological functions that associated with these age-related proteins were cell migration and positive regulation of cell communication (Supplementary Fig. 4). Lastly, TNFRSF9, CD27, TNFRSF21, TNFRSF4, and IL12RB1 were positively associated with age while VEGFR2, a tyrosine kinase receptor for VEGF, was negatively associated with age in EA and AA men (Fig. 2 and Supplementary Data 1).

In contrast to the positive association of many of the immune-oncological proteins with age, BMI tended to be negatively associated with these circulating immune-oncological analytes. For instance, 16 of the immune-oncological proteins negatively associated with BMI among Ghanaian men (Fig. 2). This finding may be surprising as obesity is generally thought to be associated with systemic inflammation. On the contrary, serum GAL1, a glycan binding protein that mediates the suppressive function of T_Reg cells²², showed the opposite trend and was positively associated with BMI in all three population groups.

To explore how the social/behavioral environment may affect immune-oncological serum protein levels, we investigated their relationship with educational attainment. For Ghanaian men, 18 of the 82 immuno-oncological markers were negatively associated with their education level (Fig. 2). Among EA men, three of the 82 immune-oncological proteins had significant inverse relationships with the attained level of education (Fig. 2), with two of these markers showing a similar pattern among Ghanaian and EA men.

Previous studies have shown that tobacco smoking increases inflammation²³. Herein, we assessed the association between cigarette use (never, former, vs. current smoker) on the level of immune-oncological proteins in circulation. We found that current smoking was consistently associated with significantly increased level of analytes that regulate angiogenesis (ANGPT2), antigen presentation (CD83), and autophagy (LAMP3), in all three study populations (Fig. 2).

Innate immune system-driven inflammatory processes have been implicated in the pathogenesis of diabetes²⁴. In our analysis, among the proteins that showed an association with self-reported diabetes, a matrix metalloprotease enzyme, MMP7, was positively associated with diabetes in all three population groups (Fig. 2). On the other hand, CX3CL1 was positively associated with diabetes exclusively in men with African ancestry (Fig. 2 and Supplementary Data 1). CX3CL1 is known to regulate insulin secretion²⁵, is elevated in the serum of patients with type 2 diabetes²⁶, and has been implicated in diabetic nephropathy²⁷, validating the findings in our study.

C-reactive protein (CRP) is a commonly measured pro-inflammatory marker in the bodyand has been reported to be associated with worse prostate cancer prognosis^28,29. Because it was not part of our marker panel, we measured blood CRP in 156 plasma samples from population controls of the NCI-Maryland study. Smoking was the only socio-demographic variable that showed association with CRP, however, the observed association was lost when adjusted for multiple testing (Supplementary Table 5). Furthermore, CRP showed positive associations with 24 of the 82 serum proteins (TNFRSF9, IL7, PGF, IL6, Gal9, GZMH, CXCL1, TNFSF14, Gal1, PDL1, HGF, HO1, CD70, TNFRSF12A, CCL3, MMP7, ANGPT2, VEGFA, CCL20, KLRD1, CSF1, CD4, MCP3, and CXCL11).

The systemic immune-oncological cytokine profile in men of African ancestry is distinct from men of European ancestry

To investigate if ancestral population group differences may influence circulating levels of the immune-oncological markers, we performed an unsupervised clustering analysis examining how the levels of the 82 immune-oncological analytes would group men without prostate cancer from Ghana and the US. Notably, these analytes tended to cluster by population group, with levels in Ghanaian men being most distant from EA men while AA samples tended to cluster in between these two groups (Fig. 3), suggesting that the ancestral background may have a significant impact on the global immune-oncological protein profile. We performed an additional statistical analysis of cluster assignments to more formally establish that the immune-oncological protein profile defined by the 82 markers is indeed different between these groups of men. We obtained the cluster assignments by cutting the hierarchical clustering dendrogram to extract K clusters (with K = 2, 3) and tested for differences in their distribution across the population groups (Supplementary Fig. 5). We found significant differences in cluster representation between Ghanaian, AA, and EA men with cluster enrichment by population group at P < 1.e−10, confirming that significant differences likely exist in the global immune-oncological protein profile among them.

Heatmap showing protein profiles for men without prostate cancer. Each row represents a protein (n = 82), and each column corresponds to an individual [n = 1482 (654 Ghanaian, 374 AA, and 454 EA)]. Each individual is color-coded as Ghanaian, AA, or EA in the annotation bar on top of the heatmap. Normalized z-score of proteins abundance is depicted on a low-to-high scale (blue-white-red). Source data are provided as a Source Data file. AA African American and EA European American.

Full size image

To further evaluate the influence of ancestry, we estimated West African ancestry in AA and EA population controls of the NCI-Maryland study and its relationship with the cytokine profile. West African ancestry was determined using 100 validated ancestry informative markers³⁰. The approach showed that, to some extent, the variance in the levels of several immune-oncological analytes can be strongly influenced by the degree of West African ancestry of these individuals (Fig. 4a). The variance in 39 of the analytes were significantly [false discovery rate (FDR)-adjusted P < 0.05] influenced by degree of West African ancestry (Supplementary Table 6, Supplementary Data 2). The levels of 37 analytes were significantly accounted for by West African ancestry even after adjusting for age, BMI, aspirin use, education, income, diabetes, and smoking status (Supplementary Table 7, Supplementary Data 3). CXCL5, CXCL1, MCP2, MCP1, CXCL11, CCL23, PTN, TWEAK, NCR1, IL18, and CCL17 were the top-ranked proteins. West African ancestry contributed to the variance with various effect sizes and explained >10% of the variance among the top 7 proteins (Supplementary Tables 6–7, Supplementary Data 2–3). For instance, 41% and 50% of the variance in the serum levels of CXCL1 and CXCL5, respectively, was accounted for by the degree of West African ancestry (Fig. 4a, Supplementary Tables 6–7, Supplementary Data 2–3).When we compared the levels of these proteins across the three population groups, we observed a significant African ancestry-related trend (Fig. 4b), with 10 of the 82 circulating immune-oncological proteins (CXCL5, CXCL1, CXCL11, MCP2, CCL17, MCP4, CD70, PDL2, MMP7, and CCL19) being significantly elevated in both Ghanaian and AA men compared to EA men (Supplementary Table 8); 13 other markers (MCP1, IL12, CCL23, CD8A, NCR1, TNFRSF4,TNFSF14, TWEAK, IL7, HGF, HO1, TNFRSF21, and ANG1) were inversely related to West African ancestry (Supplementary Table 9).

a Variance analysis for the levels of each of the 82 immune-oncological cytokines assessed as a function of genetic estimation of West African admixture among men without prostate cancer within the NCI-Maryland study (n = 795). The blue plot represents the proportion of variance that can be explained by the degree of West-African admixture while the grey plot represents the residual variance that remains to be explained by factors other than West-African ancestry. b The median levels of the top six West-African ancestry-correlated immune-oncological proteins were compared between Ghanaian (n = 654), AA (n = 374), and EA (n = 454) men. Error bars represent inter quartile range (IQR). Linearized protein abundances (2^NPX) were used to determine median and IQR for each of the proteins. Data are presented as median ± IQR. Source data are provided as a Source Data file.

Full size image

Cytokines associated with suppression of tumor immunity and chemotaxis are upregulated in men of African ancestry

Levels of many of the 82 immune-oncology markers showed a marked association with ancestry. To better define the functional implications of these population group differences, we grouped the 82 proteins into six biological processes according to Olink guidelines (Supplementary Table 10): apoptosis/cell killing, autophagy/metabolism, chemotaxis/trafficking to tumor, suppression of tumor immunity (Th2 response, tolerogenic), promotion of tumor immunity (Th1 responses), or vasculature and tissue remodeling. To gain insight on how activation of these six processes/pathways may differ by population group, we compared process/pathway sum scores between Ghanaian, AA, and EA men without prostate cancer using multicomparison-adjusted significance testing. Of these pathways, chemotaxis, promotion of tumor immunity, and suppression of tumor immunity were different in their predicted activity between AA and EA men (Fig. 5). AA men had higher scores for chemotaxis and suppression of tumor immunity when compared to EA men, indicating higher pathway activity in AA men, but they had a lower score for promotion of tumor immunity. Ghanaian men had even higher scores for chemotaxis and suppression of tumor immunity than both AA and EA men (Fig. 5c, e), indicating a possible association with West African ancestry. The latter was corroborated with our finding that the chemotaxis and suppression of tumor immunity scores positively correlated with the proportion of West African ancestry within the NCI-Maryland cohort, even after holding the other variables constant (i.e., age, BMI, education, aspirin use, diabetes, and smoking history) in the regression analysis (for chemotaxis score: regression coefficient = 5.12 (3.75, 6.49), P < 0.0001; for suppression of immunity score: regression coefficient = 4.02 (2.01, 6.04), P < 0.0001). Even though apoptosis and vasculature-associated cytokines were not significantly different between EA and AA men, we found both processes to be elevated in the Ghanaian men.

Shown are population differences in the abundance of proteins driving (a) autophagy, (b) apoptosis, (c) chemotaxis, (d) promotion of tumor immunity, (e) suppression of tumor immunity, and (f) vasculature. Heatmaps despict levels of process/pathway-associated proteins in relationship to population group (Ghanaian, AA, EA). Shown to the right are the mean score differences for these processes/pathways among the three population groups. Profiles for Ghanaian (n = 654), AA (n = 374), and EA (n = 454) men without prostate cancer are shown. The process/pathway scores are derived from the average z-scores of all the associated proteins. These scores are shown as violin plots and were compared using two-sided Wilcoxon rank sum tests. P values were adjusted for multiple comparison. FDR-adjusted P value significance was coded as <0.0001 (****), <0.001 (***), and ≥0.05 (ns). The violin plots represent median values ± IQR. Source data are provided as a Source Data file. TI tumor immunity. FDR False Discovery Rate, AA African American, and EA European American.

Full size image

Suppression of tumor immunity is associated with reduced survival of prostate cancer patients

Next, we examined the clinical implication of our findings and assessed the association of pathway activity with survival of prostate cancer patients or controls in the NCI-Maryland study using multivariable Cox regression modeling. As of the end of 2018, out of the 819 cases, there have been 202 deaths in our case population, of whom 103 (51%) had a cancer diagnosis as the recorded primary cause of death, and 28% of all deaths (n = 57) were directly attributed to prostate cancer. On the other hand, 99 of the 828 population controls had died by the end of 2018. Median survival follow-up for cases and controls were 8.6 and 6.7 years, respectively. With these data, we built a Cox regression model with the six biological processes/pathways and additional adjustments for other covariables including disease status (see Methods). Among the six pathways, only suppression of tumor immunity showed independent association with survival outcomes among the cases in a disease status-adjusted analysis (i.e., PSA, TNM Stage, Gleason score, and Gleason pattern) defined by the National Comprehensive Cancer Network (NCCN) risk score (Fig.6 with 99% CI, Supplementary Fig. 6 with 95% CI). Prostate cancer patients with an increased activity of this pathway had the highest risk of death from all causes (Fig. 6a, Supplementary Table 11). In contrast, suppression of tumor immunity was not associated with all-cause mortality of population controls (Supplementary Table 12), suggesting that the association with all-cause mortality among cases might be prostate cancer-related. Prostate cancer patients with elevated suppression of tumor immunity at diagnosis had also the highest risk of a prostate cancer-specific mortality, although statistically significant only with a 95% CI (Supplementary Fig. 6), but not with a multicomparison-adjusted 99% CI (Fig. 6b, Supplementary Table 13). Lastly, prostate cancer patients with increased suppression of tumor immunity were also significantly more likely to die from any cancer (prostate cancer or secondary cancer) following the prostate cancer diagnosis (Fig. 6c, Supplementary Table 14), indicating a more general predisposition to cancer in patients with a high suppression of tumor immunity score in this hypothesis generating analysis approach.

We assessed the association of the six pathways defined by the 82 immune-oncology markers with all-cause mortality (n = 202), prostate cancer-specific mortality (n = 57), or mortality due to any cancer after a prostate cancer diagnosis (n = 103) out of the 819 prostate cancer patients followed. The pathway scores were evaluated as continuous predictor variables. Suppression of tumor immunity pathway was distinctively associated with all-cause mortality (a), prostate cancer-specific mortality (b), or a mortality due to any cancer after a prostate cancer diagnosis (c). Multivariable Cox regression analyses were used to assess if the pathways were independently associated with survival of prostate cancer patients in the NCI-Maryland study. For (a–c), the hazard ratios (HRs) were adjusted for age at study entry (years), BMI (kg/m2), self-reported race (AA/EA), education (high school or less, some college, college, professional school), income (<$10k, $10–30 K, $30–60 K, $60–90k, greater than $90k), smoking history (never, former, current), diabetes (no/yes), aspirin use (no/yes), treatment (0 = none, 1 = surgery, 2 = radiotherapy, 3 = hormone, 4 = combination), and NCCN risk score. The HRs indicate the change in risk of dying when the biological process z-score value increases by 1 while holding all the other biological processes’ z-scores and covariates constant. Data are presented as hazard ratios ± 99% confidence intervals. Source data are provided as a Source Data file. TI tumor immunity and CI confidence interval.

Full size image

Elevated suppression of tumor immunity is associated with metastatic prostate cancer

To further corroborate the significance of suppression of tumor immunity in the development of lethal prostate cancer, we assessed its association with prostate cancer aggressiveness per NCCN guidelines (see Methods). Information on TNM stage was only obtainable for the NCI-Maryland prostate cancer patients, hence only these cases were scored according to the NCCN guidelines. Patients with a high suppression of tumor immunity score were at substantially increased odds of being diagnosed with regional or distant metastasis (OR 3.79, 95% CI 1.59–9.04, >median vs. ≤median) (Table 1), consistent with the disease survival data. The data showed a significant trend in the association of elevated suppression of tumor immunity with disease aggressiveness (P trend = 0.004) (Table 1). Although a stratified analysis by self-reported race/ethnicity suggested that high suppression of tumor immunity was associated with metastatic prostate cancer more strongly among AA than EA men, large 95% CIs for the odds ratio precluded the observed difference across racial groups from statistical significance.

Blood levels of TNFRSF9/CD137/4-1BB and pleiotrophin predict lethal prostate cancer among AA men

To identify potential drivers of the relationship between immune-oncology markers and lethal prostate cancer, we applied a cross-validated, regularized Cox regression model using eNetXplorer (see Methods). Included in this model were the 82 immune-oncology markers and 6 patient feature covariates (age, education, BMI, smoking history, aspirin use, and diabetes). Of those patient features, education as a surrogate for socioeconomic status and health care access, BMI, smoking status, and aspirin use have previously been associated with the risk of lethal prostate cancer whereas the direction and strength association for diabetes with prostate cancer and disease outcomes is more uncertain^31,32. Utilizing this method, we could not identify a robust predictive signature of lethal prostate cancer for EA patients. However, for AA patients, lasso regression (alpha = 1) was selected as a predictive model (P = 0.0001) with the best overall performance across alpha values (Supplementary Fig. 7A, Supplementary Data 4). In this model, a signature consisting of TNFRSF9 and pleiotrophin (PTN), both positively associated with the risk of lethal disease, and regular aspirin use (negatively associated with risk) emerged as a top predictor based on two selection criteria: the feature frequency (Fig. 7a) and the weight of the features’ contribution to the prediction (Fig. 7b). These three features combined predicted prostate cancer-specific mortality with an accuracy of 83.7% (SE = 3.8%). Our finding that regular aspirin use was a predictor of improved survival among AA men is consistent with previously published data from this case-control study¹⁴ and the Southern Community Cohort Study¹⁵. The combination of the two proteins alone predicted prostate cancer-specific mortality with 78.2% (SE = 4.2%) accuracy. To gain additional insight of how the prediction of lethal prostate cancer by these two protein markers might be influenced by a patient’s disease status, we added the NCCN risk score to our model. Consistent with the baseline model that did not contain the NCCN risk score, lasso (alpha = 1) remained the most predictive model (Supplementary Fig. 7b, Supplementary Data 5). The NCCN risk score was the top predictor of the lethal disease. The two analytes PTN and TNFRSF9 remained the most predictive features besides the NCCN risk score, and the three features combined predicted prostate cancer-specific mortality with 90% accuracy (Supplementary Fig. 7B, Supplementary Fig. 8). In addition, AA prostate cancer patients with high levels (>median) of both TNFRSF9 and PTN in their blood at diagnosis had the worst prostate cancer-specific survival independent of disease status [(Adjusted HR = 3.09 (1.36, 7.03)] (Fig. 7c). By 10 years, 33% of cases with high levels of both TNFRSF9 and PTN died of prostate cancer compared to only 5% of cases with low levels of one or both of these proteins (Fig. 7c), highlighting the utility of these blood markers for risk stratification of AA prostate cancer patients.

Cross-validated, regularized Cox regression models with different elastic net mixture parameters from ridge (alpha = 0) to lasso (alpha = 1) were implemented to identify a predictive proteomic signature. One-sided P values were obtained empirically by comparing feature frequencies and coefficients against those obtained from random permutations of the response⁷⁴. a Heatmaps of feature frequencies across alpha. Features were ranked by P value for alpha = 1. b Heatmaps of feature coefficients across alpha. Features were ranked by P value for alpha = 1. c Kaplan–Meier plot comparing prostate cancer-specific mortality of AA cases with high levels (>median) of both TNFRSF9/CD137/4-1BB and pleiotrophin (PTN) vs. low levels of either or both proteins. Two-sided log-rank test was used to determine if there were statistically significant survival differences. Adjusted hazard ratio (HR) compares the risk of prostate cancer mortality for those with high levels of both TNFRSF9 and PTN vs. the remaining AA cases. HR estimates were adjusted for potential confounding factors: age, BMI, education, income, smoking history, diabetes status, aspirin use, treatment, and NCCN risk score. In (a and b), P value significance was coded as <0.001 (***), <0.01 (**), <0.05 (*),and <0.1 (.). The exact P values are found in the Source Data file. Source data are provided as a Source Data file. HR hazard ratio.

Full size image

In this study, we describe differences in the expression of immune and chemotaxis-related markers in men from three population groups, with two of them—AA and Ghanaian men—having an ancestral relationship due to the trans-Atlantic slave trade. Most notably, expression of immune-oncology markers related to immune suppression were up-regulated in men of West African ancestry and were associated with prostate cancer mortality. While ancestry can explain some of the observations, other and yet unknown factors contribute to these clinically significant differences in immune function and chemotaxis.

Infections endemic to certain regions have shaped the immune response in affected populations, leaving a lasting genetic and epigenetic footprint³³. As such, population differences in exposures to fatal pathogens have led to population heterogeneity in the immunome. It has been estimated that as many as 360 immune-related genes have been targets of positive selection and have functional variations between populations³⁴. Consistent with these observations, we now report population differences in circulating immune-oncological proteins among Ghanaian, AA, and EA men. We found that the serum proteome-defined immunome of Ghanaian men resembles the immunome of AA men more so than EA men. We identified CXCL5, CXCL1, MCP2, MCP1, and CXCL11 as the top immune-oncological proteins associated with West African ancestry. Four of these chemokines (CXCL5, CXCL1, MCP1, and CXCL11) are known targets of Duffy Antigen Receptor for Chemokines (DARC) binding³⁵. DARC is a non-signaling receptor that binds to both CXC and CC family of chemokines and acts as a depot for chemokines on erythrocytes and as decoy receptor on endothelial cells³⁶. DARC expression modulates the susceptibility to clinical Plasmodium vivax malaria and loss of its expression on erythrocytes, which frequently occurs in sub-Saharan African populations due to germline genetic variants, confers resistance against malarial infection³⁷. Its loss may also influence cancer susceptibility^38,39. Consequently, these individuals lack the ability to sequester the target chemokines, leading to elevated concentration of the chemokines in circulation⁴⁰. Accordingly, we found that CXCL5, CXCL1, and CXCL11 were 2–3-fold higher in sera of Ghanaian or AA men than EA men. Given the angiogenic properties of these chemokines⁴¹, their role in cancer progression has been proposed⁴².

As a key finding, we report that serum proteins regulating chemotaxis and suppression of tumor immunity were elevated in men of African ancestry, suggesting persistent population differences in stimulation of leukocyte recruitment and T cell mediated immune response. Such differences may predispose men of African descent to a distinct tumor microenvironment. Although the direct impact of the peripheral immunome on the prostate tumor microenvironment requires further investigation, we and others have previously reported stark differences in the immune landscape of prostate tumors of AA men as compared to EA men^{9,10,11,12,13}. For instance, programmed cell death ligand-1 (PD-L1), which suppresses T cell–mediated tumor immunity, was found to be overexpressed in AA prostate tumors⁴³. Recent work by Awasthi et al. reported that AA prostate tumors tend to be enriched for immune pathways that are associated with poor clinical outcomes⁴⁴.We show with our current work that elevated, peripheral suppression of tumor immunity associates with lethal prostate cancer and the underlying mechanism may possibly involve an effect on metastasis. Hence, population differences in suppression of tumor immunity may contribute to the disproportionate burden of lethal prostate cancer among men of African ancestry. On the other hand, such differences may offer a therapeutic advantage for immunotherapeutic strategies that are tailored to target immune suppressive pathways. A recent study provided a first indication that differences in the response to cancer vaccines may lead to higher survival rates among AA men⁴⁵.

Differentiating men who have lethal forms of prostate cancer from those with a more slow-growing disease remains a major challenge in clinical oncology. Risk stratification strategies are particularly needed for AA prostate cancer patients who disproportionately bear the prostate cancer burden. This study identified TNFRSF9/CD137/4-1BB and PTN as candidate predictive blood markers for prostate cancer mortality among AA patients. AA patients with high levels of both TNFRSF9 and PTN in their sera had the highest risk of dying from prostate cancer. The membrane form of TNFRSF9 possesses antitumor properties and agonistic anti-TNFRSF9 antibodies are currently in clinical trials^46,47. On the contrary, the soluble isoform of TNFRSF9 that we measured, generated by alternative splicing⁴⁸, has been shown to antagonize antitumor immune response hence promote tumor survival most likely by acting as decoy receptor^49,50. Regulatory T cells described as Tregs are thought to be a major source of secreted TNFRSF9^51,52. Recently, TNFRSF9 mRNA level was shown to be a robust marker of tumor-infiltrating Tregs that suppress antitumor response⁵³. Moreover, high numbers of TNFRSF9–expressing Tregs were associated with poor survival outcomes across multiple human cancers⁵³, consistent with our findings that serum TNFRSF9 associates with lethal prostate cancer. PTN or pleiotrophin, the second protein marker found to be associated with lethal prostate cancer in AA men, may not have the same immune function that soluble TNFRSF9 exhibits. PTN is a secreted cytokine that is developmentally regulated. Normally expressed during embryogenesis as a growth or differentiation factor, it is expressed either at very low levels or not at all in healthy adults^54,55. PTN re-expression in adults is associated with tumor development, metastasis and angiogenesis with elevated expression reported in several cancer sites^56,57,58. This increased expression of PTN has been associated with poor prognosis in colorectal cancer⁵⁶, hepatocellular carcinoma (HCC)⁵⁸ and gliomas⁵⁷. Several hypotheses have been pursued to find out how PTN exerts pro-metastatic effects. For example, it is proposed that PTN promotes cancer progression through increased vascular endothelial growth factor deposition at the vasculature leading to vascular disruption⁵⁷. PTN may upregulate lipid synthesis, contributing to hepatic stenosis and progression of HCC⁵⁸. Serum PTN is a candidate biomarker for occurrence of breast⁵⁹ and small cell lung cancers⁶⁰ and has been shown to associate with metastatic prostate cancer⁶¹, consistent with the findings in this study. In prostate cancer, PTN regulates mesenchymal and epithelial proliferation, with PTN itself being regulated by the androgen receptor during prostate development⁶². To the best of our knowledge, a potential relationship between PTN and TNFRSF9 has not been described. However, PTN has been implicated in the induction of TNF-α expression in peripheral blood mononuclear cells demonstrating a link between PTN and the TNF superfamily in the circulation⁶³.

The immune and inflammatory environment in the circulation has been implicated as a potential influencer of metastasis. Evidence is emerging to indicate that inflammation-activated platelets are pro-metastatic, instigating the formation of a pre-metastatic niche. Lucotti et al. recently demonstrated reduced metastasis in lung cancer models through inhibition of intravascular COX-1-derived thromboxane A2 (TXA2) from platelets via aspirin treatment⁶⁴. Consistent with the animal model data, we recently reported an association between elevated urinary TXB2 (the stable metabolite of TXA2) and metastatic prostate cancer in AA men with prostate cancer⁶⁵, suggesting a distinct inflammatory environment and platelet activity in these men. Platelets can disrupt immunosurveillance of the metastatic cascade through cloaking of natural killer (NK) cells, preventing the NK cells from patrolling and inducing tumor-cell cytolysis⁶⁶. Pre-clinical studies have also implicated other immune cells including macrophages⁶⁷, Tregs⁶⁸, and neutrophils^69,70 as promoters of the metastatic process through protection of tumor cells in the circulation and promotion of tumor cell seeding. Thus, the immune-oncological profile in men of African ancestry may develop in an environment of systemic chronic inflammation and promote metastasis. Future research should test this hypothesis.

Our study has strength and limitations. The major strength is the large sample size, the measurement of 82 immune-oncology markers with a robust technology, and the inclusion of men from Ghana and the U.S. Moreover, we applied multiple testing adjustments in reporting the significance of our observations. However, we collected blood samples in Ghana and the U.S. Although blood sample collection in Ghana followed a protocol that applied standards of practice in the U.S., we cannot exclude that serum preparation and shipping influenced the performance of the immune-oncology marker measurements, yet we are not aware of such an influence. In our survival analysis, we included 57 events for prostate cancer-specific deaths, 103 events for any cancer deaths, and 202 events for all-cause mortality among men with prostate cancer. In these analyses, we found that an elevated suppression of tumor immunity score to beconsistently associated with decreased survival. Yet, additional studies will be needed to further confirm the association of the suppression of tumor immunity signature with lethal prostate cancer.

In conclusion, it is a key finding of our study that suppression of tumor immunity is increased in Ghanaian and AA men, when compared to EA men, and associates with lethal prostate cancer. As such, these findings provide an insight into potential causes of the prostate cancer health disparity. The current study has a large representation of men of African descent who were profiled for their immune-oncological proteome. With the advent of an increasing number of immunotherapies in the drug development pipeline, studies like ours may inform clinical research on population differences in the immune landscape that need to be considered when designing therapies that exploit the immune response.