Multiomic Analysis Links Neighborhood Disadvantage to Inflammatory Proteins and Tumorigenic Markers in ER+ Breast Cancer Plasma and Tumor Samples

Abstract

Residing in disadvantaged neighborhoods has been linked to worsened survival among Black women with estrogen receptor-positive (ER+) breast cancer (BC), yet the underlying multiomic alterations remain underexplored. To investigate associations between neighborhood deprivation and pretreatment steroid hormones, untargeted metabolites, inflammatory proteins, and tumoral gene expression in women with primary ER+ BC and cancer-free controls, pretreatment plasma was collected from ER+ BC patients (n = 91) and controls (n = 141) across three Chicago hospitals. Area deprivation index (ADI) was calculated per participant. Plasma was analyzed via targeted steroid hormone and untargeted metabolomics assays, and Olink’s inflammatory protein panel. Tumor samples (n = 71) were analyzed using the Nanostring Breast Cancer 360 panel. Single-omic analysis and multiomics integration were performed. Elevated inflammatory proteins were observed in cases and controls from disadvantaged neighborhoods (p < 0.05), and tumoral gene expression showed upregulation of inflammatory and proliferation-related genes (p < 0.05). Patients from deprived areas exhibited higher inflammation and antioxidant depletion even within the same tumor grade (p < 0.05). Neighborhood deprivation correlates with pro-inflammatory, proliferative multiomic profiles that may underlie worsened outcomes.

1. Introduction

Breast cancer is the most common cancer in women and the second leading cause of cancer deaths in the US. Black women have a 42% higher mortality rate from all types of breast cancer compared with non-Hispanic white (NHW) women. , The most common breast cancer subtype, estrogen receptor-positive (ER+) breast cancer, is typically associated with better outcomes, but Black women with ER+ breast cancer have more aggressive tumor phenotypes and worse survival compared with NHW women, even after controlling for stage at diagnosis, treatment, and other prognostic factors. , Research by our lab and other groups suggests that this disparity is not due to genetic differences between races, but is associated with socioeconomic factors, particularly neighborhood disadvantage. ,, Due to systemic inequities, Black women are more likely than NHW women to live in disadvantaged neighborhoods. Residing in disadvantaged neighborhoods is associated with worsened outcomes in breast cancer patients, even after controlling for race, suggesting neighborhood status as a key contributor to this disparity. ,,

The transcriptomic and metabolic perturbations driving these worsened outcomes are unclear. Past research on the topic has primarily focused on metabolomic or transcriptomic analysis of Black women with breast cancer, which has revealed dysregulated fatty acid metabolism, altered amino acid usage, cell cycle dysregulation, and dysregulated immune and inflammatory responses. , These studies unfortunately often lack neighborhood-level information, making it unclear how neighborhood status connects to these unfavorable profiles. The area deprivation index (ADI) is an extensively validated measure of neighborhood deprivation using census-based data to quantify variables like income, housing quality, education, and employment. Neighborhood deprivation, as measured by ADI, has been linked to worsened survival among breast cancer patients, , but the connection between ADI and multiomic signatures is underresearched. Applying ADI in a multiomics framework offers a unique opportunity to better understand breast cancer disparities.

There are several mechanisms through which neighborhood status may impact breast cancer outcomes. For example, chronic stress, through altered signaling of the stress hormone cortisol, may increase tumorigenic genes and alter inflammatory profiles, ,, while increased exposure to endocrine disruptors can drive metabolic reprogramming and worsened patient outcomes. However, studies linking neighborhood disadvantage to altered metabolomes and transcriptomes are lacking, making it difficult to study the mechanisms through which neighborhood status impacts ER+ breast cancer outcomes.

Multiomic analysis of ER+ breast cancer patients residing in disadvantaged neighborhoods is a powerful tool to address these knowledge gaps. Transcriptomics, the study of gene expression, can highlight tumoral tumorigenic gene profiles associated with neighborhood status. The investigation of circulating proteins through proteomics, particularly using an inflammatory proteomics panel, can reveal the link between neighborhood disadvantage and inflammation, while metabolomics can further highlight changes in stress hormones and metabolic perturbations related to neighborhood status. This study aims to leverage multiomic approaches to uncover the molecular signatures associated with neighborhood disadvantage, thereby identifying potential biological mediators of outcome disparities.

2. Experimental Procedures

2.1. Experimental Design

This study was a secondary analysis using samples from a previously published study. This study was conducted in accordance with the provisions of the Declaration of Helsinki for the participation of human subjects in research approved by the University of Illinois, Chicago Institutional Review Board (IRB protocol number 2017–1029). Informed consent was obtained from all subjects. Patients aged 24–80 years, with a new diagnosis of American Joint Committee of Cancer (AJCC) stage I–III, ER+ breast cancer were recruited from 3 hospitals in Chicago, IL, between 2018 and 2019. Hospitals were classified as those predominantly serving advantaged or disadvantaged neighborhoods, according to the patient demographics described in their annual Community Needs Assessment reports. Participants without breast symptoms or a personal history of breast cancer who presented for a screening mammogram at corresponding mammography centers were recruited as control subjects. Only control subjects whose screening mammogram was negative (BI-RADS category 1–2) were included in the study population.

Plasma samples were initially collected from 103 breast cancer cases and 150 cancer-free controls. Following analyses conducted for the original study, sufficient plasma remained for additional analysis of steroid hormones in the current study for 91 cases and 141 controls. Tumor tissue for transcriptomic analysis was available for a subset of 71 of the 91 cases.

2.2. Patient Sample Collection

Control subjects donated nonfasting whole blood specimen at the time of study enrollment. The diagnosis of breast cancer in case subjects was confirmed via biopsy. Tumor samples were collected at the time of surgical tumor excision, fixed in formalin, embedded in paraffin (FFPE), and stored and shipped at room temperature. Nonfasting whole blood was collected from individuals with breast cancer case subjects after tumor biopsy but prior to breast surgery or any other breast cancer treatment. Whole blood collected in serum separator tubes was allowed to clot for 40 min at room temperature, and plasma was separated by centrifugation within 1 h of collection (1680g for 10 min at room temperature in a horizontal rotor, tabletop centrifuge). Aliquots of plasma were immediately frozen at −20 °C and transferred to −80 °C within 4 weeks. Proteins and untargeted metabolites were analyzed on the first freeze–thaw cycle, steroid hormones were analyzed on the second freeze–thaw cycle, and glycosylated hemoglobin A1C (HbA1C) was measured on the third freeze–thaw cycle.

2.3. Area Deprivation Index

The residential address at the time of study enrollment was collected from the enrollment survey and geocoded to the census block level using OpenStreetMap (OSM) and Nominatim (OSM’s geocoding service) to derive the latitude and longitude for each residence. Data were visualized using QGIS 3.42.0. We used the area deprivation index (ADI) to rank residential neighborhoods by socioeconomic disadvantage at the census block level.

The ADI uses American Community Survey (ACS) 5-Year Data for its construction and includes 17 factors representing the theoretical domains of income, education, employment, and housing quality. Higher values represent more disadvantage. Publicly available data for census block ADI for 2019 was downloaded from the Neighborhood Atlas (https://www.neighborhoodatlas.medicine.wisc.edu/).

After sample collection, study participants were classified into tertiles of ADI: T1, representing the least disadvantage (ADI score of 0–32.9), T2 (ADI score of 33–65.9), and T3, representing highest disadvantage (ADI score of 66–100). Tertiles were selected to maintain sufficient sample size per group while capturing the distribution of ADI in our cohort.

2.4. Untargeted Metabolomic Analysis

Untargeted metabolomic profiling was performed via gas chromatography/mass spectrometry (GC/MS) by the Carver Metabolomics Core of the Roy J. Carver Biotechnology Center, University of Illinois Urbana–Champaign.

To precipitate protein, 200 μL of plasma was mixed with a 1 mL 3:3:2 mix of isopropanol, acetonitrile, and H₂O. The mixture was vortexed, centrifuged for 5 min at 15,000g, and 600 uL of supernatant was transferred to a separate Eppendorf tube. Tubes were stored at −20 °C prior to delivery to the Carver Metabolomics Core.

50 μL of the internal standard (hentriacontanoic acid, 1 mg/mL) was added to each sample prior to drying down. Samples were derivatized with 50 μL of 40 mg/mL methoxyamine hydrochloride in pyridine (Sigma-Aldrich, MO) for 60 min at 50 °C, then with 50 μL MSTFA + 1%TMCS (Thermo, MA) at 70 °C for 120 min, and following 2 h incubation at room temperature. before derivatization.

Samples were analyzed using an Agilent 7890 gas chromatograph, an Agilent 5975 MSD, and an HP 7683B autosampler. Gas chromatography was performed on a ZB-5MS (60 m × 0.32 mm ID and 0.25 μm film thickness) capillary column (Phenomenex, CA). The inlet and MS interface temperatures were 250 °C, and the ion source temperature was adjusted to 230 °C. An aliquot of 1 μL was injected with the split ratio of 10:1. The helium carrier gas was kept at a 2 mL/min constant flow rate. The temperature program was: 5 min isothermal heating at 70 °C, followed by an oven temperature increase of 5 °C/min to 310 °C, and a final 10 min at 310 °C. Data was collected in positive electron impact mode (EI) at 69.9 eV ionization energy at m/z 30–800 scan range.

Spectra were evaluated using the AMDIS 2.71 (NIST, MD) software using a custom-built database of 460 unique metabolites. All known artificial peaks were identified and removed prior to data mining. All data were normalized to the internal standard hentriacontanoic acid. The instrument variability was within the standard acceptance limit (5%).

Metabolites missing over 40% of data were removed. Remaining missing data values were completed using a trained data imputation algorithm. The imputation performance was tested one metabolite after another by removing one column and doing imputation on them. This process was repeated for each metabolite in the data set, removing and imputing them individually. The imputation performance was evaluated for each specific metabolite considering R square metric. Metabolites with R ² < 0.3, indicating poor imputation performance, were removed.

2.5. Targeted Metabolomic Analysis

Samples were randomized, deidentified, and assigned new IDs for processing and analysis. Samples were analyzed for steroids using ultrahigh-performance liquid chromatography–mass spectrometry (UHPLC-MS/MS) by the Carver Metabolomics Core of the Roy J. Carver Biotechnology Center, University of Illinois Urbana–Champaign.

Steroids chemical standards used to create calibration curves for quantification were acquired from Cayman Chemical (Ann Arbor, MI) and Sigma-Aldrich (St. Louis, MO). At the beginning of the extraction process, 10 μL of 1 μg/mL of labeled surrogate internal standards (testosterone-d₃, estradiol-d₄, progesterone-d₉, corticosterone-d₈, hydrocortisone-d₄, aldosterone-d₄, dehydroepiandrosterone-sulfate-d₆) were spiked into each tube prior to adding 25 μL plasma and 65 μL of methanol. After vortexing, the samples were centrifuged and 50 μL of supernatant was transferred to an LC vial containing 5 μL of 10 ng/mL of instrument standard (CUDA and PUHA). Vials were stored at 20 °C prior to LC-MS analysis.

Ten μL of extract was injected and analyzed with an Agilent 1290 Infinity II UHPLC system (Agilent, Santa Clara, CA), with a Waters Acquity C18 BEH 2.1 × 150 mm, 1.7 um (Waters Milford, MA); Column Temp −60 °C; flow rate 450 μL/min. Mobile phases consisted of 0.1% formic acid in water (A) and 0.1% formic acid in acetonitrile (B). Data was collected on a Sciex 6500+ triple quadrupole mass spectrometer (Sciex, Framingham, MA). Data was acquired in both positive and negative multiple reaction monitoring (MRM) mode.

Metabolites were quantified using Skyline software v24.1 (RRID:SCR_014080) with 4- to 8-point calibration curves normalized to labeled internal surrogate standards. Calibration points with >20% deviation between the known and back-calculated concentrations were excluded to ensure curve accuracy. The concentrations of analytes in subject samples were determined using the resulting calibration curves. Limits of quantitation (LOQ) were established as the lowest concentration at which the analyte could be reliably quantified with acceptable accuracy and precision, defined as a deviation within ±20% of the nominal value. Upper and lower LOQs are described in Supporting Information Table 1. Quantification accuracy was further verified using certified reference values from NIST SRM 1950. Testosterone is quantified at 66% higher than the certified NIST value, which is beyond acceptable limits. As such, testosterone is reported as a semiquantitative peak area ratio between testosterone and its internal standard, testosterone-d3. Estradiol levels are reported as peak area due to inconsistencies observed during quantification, likely attributable to technical issues with the internal standard, estradiol-d4. Missing values were replaced by 1/5th of the minimum positive value of the variable. Metabolites with greater than 50% of missing values were removed from the analysis.

2.6. Olink Proteomic Analysis

Targeted proteomics in plasma was performed using Olink’s proximity extension assay technology. Plasma samples were aliquoted and submitted to Olink Technologies (RRID:SCR_003899) for their inflammatory proteins panel. Sample processing, output data quality check, and normalization of the proteomics data were performed by Olink. To avoid batch effects, samples were randomized across plates. Each plate included interplate controls, which were used to adjust for any plate differences. Proteins with greater than 50% of missing values were removed from the analysis. Final data was reported in log2 scale normalized protein expression (NPX) levels.

2.7. Transcriptomic Analysis

Eighty-two tumor samples were submitted to NanoString Technologies (RRID:SCR_023912) for analysis using the Breast Cancer 360 panel, which comprises 755 curated genes involved in 23 pathways and processes important in breast cancer biology. The panel standard, a DNA oligo blend containing target sequences for all BC 360 probes, was used for normalizing genes. Prior to normalization, zero counts on the raw scale were converted to ones. Expression values were then normalized using the ratio of each gene’s expression to the geometric mean of all housekeeping genes on the panel. This housekeeper normalization step ensured consistency in gene expression measurements across cartridges. Housekeeper genes were used to assess sample integrity by comparing the observed value versus a predetermined threshold for suitability for data analysis. The machine performance was assessed using percentage of fields of view that were attempted versus those successfully analyzed. The binding density of the probes within the imaging area, ERCC linearity, and limit of detection were used as readouts of the efficiency and specificity of the chemistry of the assay. Five samples failed these QC checkpoints and were removed from the analysis, resulting in a total of 77 cancer cases with transcriptomic data. However, 6 of these cases were missing ADI information and were thus excluded from the analysis, resulting in a total of 71 cases included in the final transcriptomic analysis.

2.8. HbA1C Analysis

HbA1C was measured via ELISA (Cloud-Clone Corp, Houston, TX, product number CEA190Hu). The assay was performed according to the manufacturer’s protocol. Briefly, 40 uL of sample was diluted 5-fold in dilution standard. 50 uL of diluted sample and standards were added to precoated wells, followed by incubation at 37 °C with detection reagent A for 1 h. Wells were washed and incubated at 37 °C with detection reagent B for 30 min, followed by another wash step. After incubation with TMB substrate for 15 min at 37 °C, the reaction was stopped and absorption was read using Cytation 5 plate reader at 450 nm. A standard curve was generated and sample concentrations were calculated using a linear regression equation derived from the curve. Of the 232 original samples, 184 samples had enough plasma to be included for this measurement. Five of these samples were removed after measurement due to poor sample quality, resulting in a total of 179 samples used during analyses.

2.9. Statistical Analysis

Covariate-adjusted linear regression modeling was performed for all continuous outcomes: metabolites, proteins, genes, and HbA1C. Two sets of comparisons were performed: cases versus controls within each ADI tertile (reference group = control), and across tertiles among cases only or controls only (reference group = T1). Age and BMI were included as covariates in both models. After covariate adjustment, metabolite were assessed for normality using the Shapiro–Wilk test (α = 0.05), and any non-normally distributed metabolites were log₂-transformed prior to modeling. Proteomic and transcriptomic measurements were all modeled in log₂ scale. Effect estimates are reported as β coefficients with 95% confidence intervals. Multiple comparisons were adjusted for using the Benjamini–Hochberg procedure with a cutoff of 0.05 to control the false discovery rate (FDR). Statistical significance is defined as a raw p-value of 0.05.

Analyte-level differences in transcriptomic, proteomic, and metabolomic data were assessed across ADI tertiles, stratified by tumor grade. For each omics data set, features were covariate-adjusted for age and BMI using linear regression, and log2-transformed residuals were used for subsequent analysis. One-way analysis of variance (ANOVA) was performed within each tumor grade to identify analytes with significant variation across ADI tertiles. Analytes with p-values <0.05 were further evaluated using Tukey’s honest significant difference (HSD) test to identify specific pairwise differences between ADI tertiles. Analyses were performed separately for each -omic data set.

Heatmaps were generated to visualize effect sizes for proteins and transcripts significantly associated with ADI tertile comparisons. Analytes with a raw p-value <0.05 were grouped into three categories: (1) significant in both T2 and T3, (2) significant in T2 only, or (3) significant in T3 only. For each category, effect sizes were used to construct heatmaps using the Seaborn clustermap function in Python. Hierarchical clustering was applied using average linkage and Euclidean distance. To enhance interpretability, color scales were customized per heatmap using the fifth and 95th percentiles of effect sizes within each group.

Protein–protein interaction networks were visualized using the STRING database. An interaction cutoff of 0.700 was used to ensure high interaction confidence. Proteins not meeting this threshold were excluded from the network.

For mediation analysis, models were run using the mediation_analysis function from the pingouin Python package (v0.5.3), with cancer status as the exposure (X), inflammation PC1 as the outcome (Y), and individual race (coded as 0 = White, 1 = Black) or the food access index as the mediator (M). Inflammation was summarized as the first principal component of covariate-adjusted, log2-transformed protein expression values. All models were adjusted for age and BMI. Logistic regression was used to model binary mediators. Indirect (mediated) effects were estimated using ordinary least-squares and were considered statistically significant at p < 0.05. Tertiles with insufficient subgroup counts (<3 samples per Race × cancer_status group) were excluded from analysis to ensure model stability.

For correlation analysis, untransformed values were covariate-adjusted for age and BMI, then non-normally distributed metabolite data and all transcriptomic and proteomic data were log2-transformed. Correlation was performed using Pearson’s or Spearman’s correlation, depending on data distribution. Correlations were performed between HBA1C and uric acid, as well as HBA1C and EN-RAGE.

Multivariate analyses were performed separately for metabolites and proteins using covariate-adjusted data (age, BMI). Multivariate analysis was conducted using principal component analysis (PCA) and partial least-squares discriminant analysis (PLS-DA) in MetaboAnalyst 6.0 (RRID:SCR_015539).

Analyses were performed in Python 3.10 (pandas 1.5, statsmodels 0.14, scipy 1.10) (RRID:SCR_008394) within a Google Collab environment, enrichment plot graphs were generated using R Studio v2024.09.0.+375, and figures were created using Adobe Illustrator v29.1 (RRID:SCR_010279).

2.10. Integrative Multiomic Analysis

The integration of the three-omics data sets using DIABLO applied the workflow proposed by Rohart et al. and Singh et al. using the mixOmics R package v 6.30.0 (RRID:SCR_016889). , Prior to integration, all data sets were log2-transformed and covariate-adjusted for age and BMI. The DIABLO model was built using a weighted design matrix that maximizes the separation between groups while taking into account the correlation between the -omics data sets. Performance of the model was assessed by M-fold cross-validation (10-fold) repeated 10 times using the mixOmics ‘perf’ function and was estimated by classification error rates (CERs). The optimal number of components to include in the model, as well as the method to use for class prediction (“Maximum-”, “Mahalanobis-”, or “Centroids”-based prediction distances), were chosen using the ‘perf’ function to minimize the CERs of the model.

3. Results

3.1. Patient Demographics

A comparison of cases compared to controls by ADI tertiles showed no significant differences in age or BMI (Table ). However, T2 had significantly more Black individuals in the control group compared to cases.

1. Patient Demographics in Cases Compared to Controls.

	T1			T2			T3
	control	cases	p-val	control	cases	p-val	control	cases	p-val
N	39	39		62	28		40	24
% Black	33.30%	15.40%	0.11	77.40%	53.60%	0.03	92.50%	91.70%	1.0
age (mean ± SD)	57.2 ± 11.0	58.9 ± 13.6	0.54	56.6 ± 8.9	59.6 ± 11.0	0.217	56.0 ± 9.9	57.5 ± 12.3	0.609
BMI (mean ± SD)	29.4 ± 8.2	28.5 ± 5.8	0.567	32.2 ± 7.3	33.1 ± 8.7	0.609	32.4 ± 10.4	30.7 ± 4.7	0.363
tumor grade 1		11			6			7
tumor grade 2		21			17			10
tumor grade 3		6			5			7

^aFisher’s exact test.

^bStudent’s t test.

In a comparison of controls or cases across ADI tertile, the number of Black individuals significantly increased across ADI tertiles in both controls and cases (Chi-square p = 1.6 × 10^–8 and 2.1 × 10^–8, respectively), with the highest proportion observed in T3 for each group (Table ). Age and BMI did not differ significantly in cases or controls across tertiles, nor did tumor grade in cases across tertiles.

2. Patient Demographics across ADI Tertiles in Controls Only and Cases Only.

	controls				cases
	T1	T2	T3	P-val	T1	T2	T3	P-val
N	39	62	40		39	28	24
% Black	33.30%	77.40%	92.50%	1.6 × 10–8	15.40%	53.60%	91.70%	2.1 × 10–8
age (mean ± SD)	57.2 ± 11.0	56.6 ± 8.9	56.0 ± 9.9	0.91	58.9 ± 13.6	59.6 ± 11.0	57.5 ± 12.3	0.76
BMI (mean ± SD)	29.4 ± 8.2	32.2 ± 7.3	32.4 ± 10.4	0.12	28.5 ± 5.8	33.1 ± 8.7	30.7 ± 4.7	0.07
tumor grade 1					11	6	7	0.60
tumor grade 2					21	17	10	0.60
tumor grade 3					6	5	7	0.60

^aChi-square test.

^bKruskal–Wallis test.

Among the subset of cancer cases included in the transcriptomic analysis and the integrative multiomic analysis, BMI, as well as the number of Black individuals, significantly increased across ADI tertiles (Supporting Information Tables 2 and 3, respectively).

To ensure our observations were not driven by prediabetic or diabetic status of individuals, we examined HbA1C levels. In the subset of samples used for this HbA1C analysis, demographic trends were consistent with those observed in the full cohort. No significant differences in age and BMI were observed between cases and controls within each tertile (Supporting Information Table 4). In a comparison of controls or cases across ADI tertile, the number of Black individuals significantly increased across ADI tertiles in both controls and cases (Chi-square p = 2.85 × 10^–7 and 8.8 × 10^–7, respectively), with the highest proportion observed in T3 for each group (Supporting Information Table 4). Age and BMI did not differ significantly in controls across tertiles. In cases across tertiles, age and tumor grade did not differ significantly, though BMI did increase in T2 and T3 (p = 0.03) (Supporting Information Table 5).

3.2. Metabolomic and Proteomic Analysis in Cases and Controls by Neighborhood Status

To investigate neighborhood-related differences in circulating metabolites and proteins between cancer cases and cancer-free controls, we performed untargeted metabolomics, targeted steroid hormone metabolomics, and inflammatory Olink proteomics. A total of 82 untargeted metabolites, 13 targeted steroid hormones, and 85 inflammatory proteins were detected. Case-control comparisons were performed within each ADI tertile and analyzed separately for each -omic data set. All metabolite and protein values, effect sizes, and p-values are available in Supporting Information Table 6.

In the targeted steroid hormone metabolomics analysis, several cortisol metabolites were significantly elevated in cases compared to controls in T1 (cortisol: β, 31.39 [95% CI, 10.34–52.45]; P = 0.004; Sum: Corticosterone & 21-Deoxycortisol: β, 1.92 [95% CI, 0.60–3.25]; P = 0.005; 11-deoxycortisol: β, 0.20 [95% CI, 0.07–0.32]; P = 0.002; and the cortisol:cortisone ratio: β, 1.12 [95% CI, 0.26–1.97]; P = 0.011) (Figure A). In T2, DHEA sulfate was elevated in cases (β, 201.64 [95% CI, 20.51–382.76]; P = 0.030) (Figure B). No steroid hormones were significantly different between cases and controls in T3.

1.

Metabolomic and proteomic profiles in cases compared to controls by neighborhood status. (A) Steroid hormones different in cases compared to controls in T1. (B) Steroid hormones different in cases compared to controls in T2. (C) Untargeted metabolites different in cases compared to controls by ADI tertile. (D) Inflammatory proteins different in cases compared to controls by ADI tertile. Red circles indicate FDR < 0.05. Purple squares indicate raw p-value of <0.05. Effect sizes (x axis) were transformed using a signed log₂(x + 1) function, which applies a log₂ transformation to the absolute value after adding 1 to avoid undefined values at zero, while preserving the original sign. This compresses large effect sizes for visualization while maintaining the directionality of associations. Abbreviations: Chol, cholesterol; GA, glyceric acid; Mal, malate; Met, methionine.

Among untargeted metabolites in T1, 3-hydroxybutyric acid and citric acid were elevated in cases versus controls (3-hydroxybutyric acid: β, 86.87 [95% CI, 35.81–137.92]; P = 0.001; Citric acid: β, 4.60 [95% CI, 1.38–7.82]; P = 0.006), while P-cresol and tryptophan were decreased (P-cresol: β, −0.26 [95% CI, −0.47 to −0.05]; P = 0.02; Tryptophan: β, −4.63 [95% CI, −8.34 to −0.92]; P = 0.015) (Figure 1C, top). In T2, 3-hydroxybutyric acid and two fatty acids were elevated in cases compared to controls (3-hydroxybutyric acid: β, 57.18 [95% CI, 20.20–94.16]; P = 0.003; 9-hexadecenoic acid: β, 4.09 [95% CI, 0.15–8.03]; P = 0.042; and 1-monooctadecanoylglycerol: β, 0.44 [95% CI, 0.12–0.76]; P = 0.008). In contrast, glutamic acid, cholesterol, and several amino acids were decreased (Glutamic acid: β, −3.00 [95% CI, −4.73 to −1.27]; P = 0.001; Cholesterol: β, −3.06 [95% CI, −5.12 to −1.00]; P = 0.004; Threonine: β, −11.31 [95% CI, −21.21 to −1.42]; P = 0.026; Tyrosine: β, −10.02 [95% CI, −18.04 to −2.00]; P = 0.015;and Methionine: β, −0.83 [95% CI, −1.59 to −0.07]; P = 0.032) (Figure 1C, middle). In T3, arginine and glyceric acid were decreased in cases compared to controls (Arginine: β, −0.94 [95% CI, −1.69 to −0.19]; P = 0.015; Glyceric acid: β, −2.44 [95% CI, −4.75 to −0.13]; P = 0.039), and no metabolites were observed to be elevated (Figure C, bottom).

Proteomic analysis revealed distinct inflammation patterns across ADI levels. In T1 and T2, many inflammatory proteins were lower in cases compared to controls. The opposite was observed in T3, as three inflammatory proteins, IL-17A TGF-α, and OSM, were elevated (IL-17A: β, 0.35 [95% CI, 0.08–0.61]; P = 0.01; TGF-α: β, 0.27 [95% CI, 0.03–0.51]; P = 0.027; and OSM: β, 0.39 [95% CI, 0.06–0.71]; P = 0.02) while none were decreased (Figure D).

2.

Untargeted metabolites in controls across ADI tertiles and in cases across ADI tertiles. (A) Untargeted metabolites different in T2 or T3 in comparison to T1 in cancer-free controls (p < 0.05). (B) Untargeted metabolites different in T2 or T3 in comparison to T1 in cancer patients (p < 0.05). Black dots represent T2 and orange squares represent T3.

Because chronic inflammation is often associated with elevated blood glucose, we examined HBA1C levels in a subset of samples to explore potential contributions to inflammatory patterns. While T1 cases had significantly elevated HBA1C compared to controls (p < 0.009), the difference was not significant in T2 or T3 (Supporting Information Table 4). Mediation analysis showed that HBA1C did not mediate the observed inflammation differences seen in controls in any tertile (Supporting Information Table 7).

Inflammation is typically higher in cases versus controls due to the pro-inflammatory state of cancer, , so additional analysis was performed to determine other factors that may contribute to the elevation in inflammatory proteins in cases vs controls in T1 and T2 groups. Given the racial composition differences observed in T2 (Table ), we conducted a mediation analysis to assess whether individual race mediated the association between cancer status and inflammation within each ADI tertile.

A significant indirect effect was observed in T2 (β = −1.01, p = 0.044), indicating that race mediated the relationship between cancer status and inflammation. Specifically, controls in T2 were significantly more likely to be Black (p = 0.018), and Black patients exhibited higher inflammation levels (p < 0.001), partially explaining the higher inflammation observed in controls versus cases in this group. In T1, while Black race was associated with higher inflammation (p = 0.046), the indirect effect did not reach statistical significance (β = −1.11, p = 0.164) (Supporting Information Table 8). Because Black people may be more likely to experience food insecurity, resulting in a pro-inflammatory diet, analysis was performed to determine if the participant food access index (FAI), a measure of the percentage of food-insecure neighborhood residents, may mediate inflammation. No significant mediation was observed in any tertile (Supporting Information Table 9).

Multivariate analysis was performed separately for targeted metabolomics, untargeted metabolomics, and proteins, comparing cases and controls within each tertile. Among targeted and untargeted metabolites in cases and controls in T1, T2, or T3, PCA and PLS-DA did not distinguish between cases and controls (Supporting Information Figure 1). Among proteins in cases and controls, PLS-DA distinguished in T1, though it did not distinguish all samples (Supporting Information Figure 2A). PLS-DA cross-validation indicated the fraction of the variance that the model explains in the independent (X) and dependent variables (Y), where R ² = 0.80 and Q ² = 0.12, with a 0.66 accuracy in a two-component model, respectively. The predictive accuracy of the model, as measured by 1000 permutations, was significant (p = 0.021). Proteins with a VIP of >1.0 are visualized in Supporting Information Figure 2B, with many of these proteins being the same as those observed to be significant in our univariate analysis in Figure D. Among proteins in cases versus controls in T2 and T3, PLS-DA and PCA did not distinguish between groups (Supporting Information Figure 3).

3.3. Metabolomic and Proteomic Profiles across ADI Tertiles

To assess the relationship between ADI and metabolomic and proteomic profiles, differences in metabolite and protein levels were analyzed across ADI tertiles, using T1 as the reference group. This analysis was performed separately by -omic group and compared tertiles within controls only and cases only. Omic values, full model estimates, and P-values are reported in Supplementary Table 11.

In controls, several metabolites were significantly altered in both T2 and T3 in comparison to T1. Testosterone was significantly higher in both T2 and T3 in comparison to T1 (Supporting Information Table 10). Among untargeted metabolites, cholesterol, maltose, n-acetylglutamic acid, and nicotinic acid were higher in T2 and T3, while threonic acid was lower. While no metabolite differences were unique to T2, α-ketoglutaric acid, glucuronic acid, and glutamic acid were elevated in T3 compared to T1, while tryptophan was lower (Figure A).

Among cancer steroid hormone metabolites showed a distinct pattern across tertiles. Cortisol, cortisone, and 11-deoxycortisol were all lower in T2 compared to T1 (Supporting Information Table 8), while T3 cases showed an elevation of androstenedione and testosterone (Supporting Information Table 11). In the untargeted analysis, 9-hexadecenoic acid and tetradecanoic acid were decreased in both T2 and T3 compared to T1 (Figure B, top panel). T2 cases also exhibited elevated p-cresol and succinic acid and reduced methyl-alanine (Figure B, middle panel). In T3, 1-mono-octadecanoyl glycerol and 3-hydroxybutanoic acid were decreased, while β-alanine, maltose, and uric acid were increased (Figure B, bottom panel).

Because elevated uric acid is associated with diabetes, correlation analysis between uric acid and HBA1C was performed. No significant association was found (Supporting Information Table 12).

Proteomic analysis revealed a distinct upregulation of inflammatory proteins across ADI tertiles. In controls, several proteins were elevated in both T2 and T3 relative to T1, with only a few proteins being altered in T3 only and many being altered in T2 only (Figure A). The higher number of significant proteins in T2 may be due to sample size imbalance, as there are substantially more T2 controls than in the other tertiles (Table ), potentially enhancing statistical power and group separation. Of the proteins elevated in both T2 and T3, protein–protein interaction (PPI) networks indicated strong interaction between CCL and CXCL chemokines and interleukins (Figure B). Proteins unique to the T2 vs T1 formed a cluster of predominantly CCL and CXCL chemokines and monocyte chemotactic proteins (Figure C). In T3 controls, notable elevations were seen for FGF-21, EN-RAGE, and TSLP (Supporting Information Table 10).

3.

Inflammatory proteins in controls across ADI tertiles and in cases across ADI tertiles. (A) Comparison of overlapping significant proteins in T2 and T3 controls. (B) Inflammatory proteins different in both T2 and T3 in comparison to T1 in cancer-free controls (p < 0.05) (left). Protein–protein interaction (PPI) network of upregulated proteins (right). (C) Inflammatory proteins different in both T2 only in comparison to T1 in cancer-free controls (p < 0.05) (left). PPI network of upregulated proteins (right). (D) Comparison of overlapping significant proteins in T2 and T3 cases. (E) Inflammatory proteins different in both T2 and T3 in comparison to T1 in cancer patients (p < 0.05) (left). PPI network of upregulated proteins (right). (F) Inflammatory proteins different in T3 only in comparison to T1 in cancer patients (p < 0.05) (left). PPI network of upregulated proteins (right).

Because EN-RAGE has been associated with high blood glucose and insulin resistance, we explored whether HBA1C varied across ADI tertiles or correlated with EN-RAGE levels. Results demonstrated that HBA1C did not differ across ADI in controls (Supporting Information Table 5), nor was there a significant correlation between EN-RAGE and HBA1C (Supporting Information Table 12).

In cancer cases across ADI tertiles, there were far more proteins unique to the T3 group than the T2 group (Figure D). Inflammatory proteins elevated in both T2 and T3 compared to T1 formed a cluster of monocyte chemotactic proteins and chemokines (Figure E). In T2 vs T1, only FGF-21 was elevated (Supporting Information Table 10). Proteins elevated in T3 only compared to T1 were highly interconnected, forming a cluster of interleukins, CCL, and CXCL chemokines (Figure F). Full model estimates and P-values are reported in Supplementary Table 11.

Multivariate analysis was performed separately for targeted metabolomics, untargeted metabolomics, and proteins, comparing tertiles in cases only or controls only. Among targeted and untargeted metabolites in cases only and cases only, PCA and PLS-DA did not distinguish between ADI tertiles (Supporting Information Figure 4). Among proteins in cases only and controls only, PLS-DA did not distinguish between tertiles, while PCA provided some separation (Supporting Information Figure 5). In cases, principal component analysis (PCA) explained 14% of the variance (R ² = 0.14, p = 0.005), with significance assessed via 1000 permutations. In controls, PCA explained 10% of the variance (R ² = 0.10, p = 0.001), using the same permutation approach.

3.4. Gene Expression Analysis in Cases across ADI Tertiles

To examine the relationship between neighborhood status and tumoral gene expression, we performed Nanostring analysis using the Breast Cancer 360 panel on a subset of 71 tumor samples. Analysis of 71 tumor samples was conducted to determine gene expression changes by ADI tertile, with T1 set as the reference group.

There were far more genes with significantly altered expression in T3 than in T2 (Figure A). Genes with increased expression in both T2 and T3 compared to T1 include genes associated with DNA repair (e.g., HDAC10, PRKDC) , and the cell cycle (e.g., BRCA2, CDC7, PRC1) , (Figure B). T2-specific upregulated genes (Figure C) were involved in immune response (such as HLA-DRB1, CFD) and estrogen signaling (such as EP300, HDAC5, GRB2). In contrast, T3-specific upregulation (Figure D) was broader, with many upregulated genes being involved in DNA repair (such as RAD51, RAD52, MLH1, POLD1), − cell cycle regulation (such as CDK4, CDC25B, CDKN1B, EIF4E2), , and cancer aggressiveness (MYCN, NOTCH3, and many others) , (Figure D). Downregulated genes were PIP, SOCS3, LAMC2, MAPT, EGLN3, HSPA2, EGF, ITGB3, PAX8, CDCA8, and SKA3. Gene expression, full model estimates, and P-values are reported in Supplementary Table 13.

4.

Tumoral gene expression in cases across ADI tertiles, comparing T2 (n = 22) and T3 (n = 19) to T1 (n = 30). (A) Overlap of significant genes in T2 and T3. (B) Genes significantly altered in both T2 and T3 in comparison to T1 in cases (p < 0.05). (C) Genes significantly altered in T2 only in comparison to T1 in cases (p < 0.05). (D) Genes significantly altered in T3 only in comparison to T1 in cancer patients (p < 0.05).

3.5. Analysis of Analytes Altered across ADI Tertiles by Tumor Grade

To determine the impact of neighborhood status on analytes by tumor grade, we performed ANOVA across ADI tertiles within each tumor grade.

Within tumor grade I, two analytes were elevated in T3 compared to T1:11-ketotestosterone (p = 0.005) and LIF-R (p = 0.002). LIF-R was also elevated in T2 compared to T1 (p = 0.03) (Figure A). Analytes in grade I that were lower in T3 compared to T1 were 9,12-octadecadienoic acid (p = 0.018) and 9-hexadecenoic acid (p = 0.003) (Figure A). 9-Hexadecenoic acid was also lower in T2 compared to T1 (p = 0.049) (Figure A). The only gene altered was HLA-B, which was lower in T2 compared to T1 (p = 0.021) (Figure A).

5.

Changes in analyte levels across ADI tertile, stratified by tumor grade. (A) Analytes altered across ADI tertile within tumor grade I (p < 0.05). (B) Analytes altered across ADI tertile within tumor grade II (p < 0.05). (C) Analytes altered across ADI tertile within tumor grade III (p < 0.05).

Within tumor grade II, two proteins were elevated in T3 compared to T1: CCL11 (p = 0.024) and CXCL5 (p = 0.019), with LAP-TGF-β-1 showing a trend increase (p = 0.051) (Figure B). CXCL5 and LAP-TGF-β-1 were also elevated in T2 compared to T1 (p = 0.031 for both), as were CXCL6 (p = 0.031) and HGF (p = 0.031) (Figure B). Cysteine was lower in T3 compared to T1 (p = 0.005) (Figure B). Analytes lower in T2 compared to T1 were cysteine (p = 0.045) and FGF-23 (p = 0.028) (Figure B).

For tumor grade III, only two analytes showed significant changes. MCP-3 was elevated in T3 compared to T1 (p = 0.034) and T2 (p = 0.012) while lactamide was lower in T3 compared to T1 (p = 0.028) (Figure C).

3.6. Integrative Multiomics Analysis in Cases across ADI Tertiles

To further examine the relationship between neighborhood status and -omics signatures, we performed integrative multiomics analysis using the mixOmics R package. This analysis incorporated metabolomic, proteomic, and transcriptomic data from 65 breast cancer cases (6 samples were excluded due to missing metabolomics data).

A multiblock sPLS-DA was applied using the DIABLO framework. Correlation structure at the component level showed moderate correlation between -omics, as well as decent separation between T1 cases and T2 and T3 cases (Figure A). The clustered image map (CIM) based on the metabolites, proteins, and genes selected on the first component displayed a good classification by ADI tertile, with T2 and T3 cases showing a distinct upregulation of analytes in the upper right quadrant of the CIM (Figure B). Of the eight T2 cases in this cluster, five had an ADI of above 50 and one had an ADI of 49.75, putting them into or just barely below the disadvantaged category. To visualize correlations between analytes, a circos plot was generated using a cutoff of 0.7 (Figure D). Cortisol metabolites were negatively correlated with several inflammatory proteins, while 42 was positively correlated with several inflammatory and cancer-aggressiveness proteins, such as VEGFA. These findings highlight a consistent biological phenotype among patients living in high-ADI neighborhoods, pointing to multisystem biological responses associated with neighborhood disadvantage.

6.

Integrative multiomics analysis comparing T1 (n = 26), T2 (n = 21), and T3 (n = 18) cases using the DIABLO mixOmics framework. (A) Diagnostic plot from multiblock sPLS-DA indicating the correlation between -omics data sets. The discriminative power of each component to separate the samples by neighborhood status is visualized by ellipses. Plot generated using plotDiablo­(). (B) Clustered image map (CIM) for the variable selected by multiblock sPLS-da on component 1. The CIM represents samples in columns (indicated by ADI tertile at the top of the plot) and selected analytes in rows (indicated by their -omics type on the right side of the plot). The color scale reflects normalized values, trimmed to within 3 standard deviations from the mean of an analyte for a given sample. (C) Circos plot from multiblock sPLS-DA representing correlations greater than 0.70 between analytes. The green, blue, and orange lines around the circle represent analyte levels, with lines further away from the circle indicating elevated levels.

4. Discussion

This study examines the multiomics signature of plasma and tumor samples from ER+ breast cancer patients in Chicago, focusing on their relationship with neighborhood status. Our results indicate that patients from disadvantaged neighborhoods exhibit alterations in proteins, metabolites, and genes associated with tumorigenesis and inflammation that may drive worse survival. Our findings expand on past ADI-based breast cancer health disparity studies. While breast cancer patients living in neighborhoods with high ADIs have been reported to experience increased tumor aggressiveness and mortality, , molecular and cellular analyses have been largely limited to measurements of single markers or -omics, such as studies showing an increase in DNA methylation and CRP among high-ADI groups. , Multiomic investigation of the impacts of ADI in breast cancer patients provides greater insight into the potential molecular pathways that may drive worsened patient outcomes.

In a comparison of cases versus controls by ADI tertile, the most affluent tertile had an increase in cortisol metabolites in cases. Cortisol is a chronic stress hormone often elevated during times of stress. Prior research has indicated that women residing in disadvantaged neighborhoods display a blunted cortisol response when experiencing a new stressor, likely due to chronic stress altering the axis responsible for cortisol release. This may explain why only cancer patients from the most affluent areas experienced elevated cortisol metabolites, while those from T2 and T3 saw no significant increases. The blunted cortisol response signature also appeared in the integrative multiomic analysis, where decreased levels of cortisol metabolites in disadvantaged patients were observed. These cortisol metabolites negatively correlated with several elevated inflammatory proteins, indicating a potential link between altered cortisol signaling and inflammation, a known effect of cortisol dysregulation.

The counterintuitive increase in inflammatory proteins in controls versus cases in T1 and T2 appears to be partially mediated by race in T2, though this finding was not significant for T1. Black individuals in our cohort were more likely to reside in high-ADI neighborhoods and displayed higher levels of inflammation, consistent with prior studies linking inflammation to chronic stress due to racial discrimination and pro-inflammatory diets due to food insecurity. While the Food Access Index, a measure of the percentage of neighborhood residents experiencing food insecurity, was not observed to mediate the elevated inflammation in controls, this elevation may be due to individual-level factors not available in our data set.

The impact of neighborhood disadvantage on metabolite and protein profiles becomes more clear when comparing across tertiles within controls only or cases only. High-ADI controls showed elevated EN-RAGE and CXCL5, along with reduced tryptophan and threonic acid, markers linked to chronic inflammation, immune dysregulation, and oxidative stress. − Previous studies have linked neighborhood disadvantage with increased inflammation, with factors such as elevated stress, increased exposure to environmental toxins, lower access to healthy foods, and smoking as a coping mechanism all potentially contributing to this inflammatory profile. , These findings suggest that neighborhood disadvantage may biologically prime individuals toward a pro-tumorigenic state even before disease onset.

In cases only, PPI network visualization showed a highly connected cluster of interleukins, chemokines, PD-L1, and CSF-1 in T3 vs T1, indicating immune dysregulation and inflammation. , The elevation of uric acid seen in T3 vs T1 may further contribute to the elevated inflammatory profile observed in T3 cases. Uric acid is the end product of both fructose metabolism and purine-rich foods (meat, poultry). It is also associated with kidney dysfunction and diabetes, though our analysis did not indicate a link between uric acid and HbA1C. Uric acid is thought to contribute to inflammation through triggering NF-kB signaling. As such, it is possible that the elevation of uric acid in disadvantaged patients, potentially driven by unfavorable diet due to food insecurity, may contribute to a pro-inflammatory profile that worsens patient outcomes.

When comparing analyte changes within specific tumor grades across ADI, chemokines CCL11 and CXCL5 were observed to be elevated in T3 patients with grade II tumors compared to T1 patients with grade II tumors. The same pattern was observed in patients with grade III tumors, who had elevation in chemokine MCP-3 in T3 compared to T1 patients. Furthermore, the decrease in cysteine levels across ADI tertiles among patients with grade II tumors potentially reflects the depletion of cysteine for glutathione synthesis during elevated oxidative stress. Overall, these analyte changes highlights how neighborhood deprivation can increase inflammatory profiles, even among patients with the same tumor grade.

Transcriptomic profiling further supported these findings. T3 tumors showing a unique upregulation of genes associated with worsened patient outcomes, including a significant upregulation of genes associated with cell cycle dysregulation that was not observed in T2 cases. Endocrine disruptors common in deprived neighborhoods, such as PFOA, are known to induce cell progression through the upregulation of CDK4, a gene observed to be upregulated in T3 cases.

Our integrative multiomics analysis confirmed the relationship between inflammatory, immune, and proliferative signaling among patients from disadvantaged neighborhoods. Among patients with an elevated ADI, an upregulation of proteins and genes involved in inflammation and immune dysregulation (MCP proteins, CCL and CXCL chemokines, and CD proteins), , as well as cancer aggressiveness (AXIN1, VEGFA, EIF4E2), , were observed. Metabolites elevated in this group included cholesterol and glutamic acid, metabolites associated with poor outcomes. Elevated cholesterol is associated with worsened cancer outcomes, potentially due to its role in increasing reactive oxygen species, activation of Hedgehog signaling, and induction of T-cell exhaustion. Glutamic acid serves as a nitrogen source for nucleotide synthesis, potentially feeding into the cellular proliferation driven by PI3K-AKT. , This is notable given the observed increase in AXIN1, VEGFA, and HGF, proteins involved in PI3K-AKT signaling, , observed in our disadvantaged cluster.

This research has several limitations. First, we were unable to obtain participant dietary intake information, which limits the interpretation of nutrient-derived metabolites and inflammatory signals. Second, we lack information regarding sample collection time, which prevents us from accounting for variations in cortisol metabolites due to circadian rhythms. Lastly, our mediation analyses were exploratory and constrained by sample size, thereby limiting a full investigation into mediating variables. Further mechanistic studies using in vivo models are essential to uncover the molecular basis of the associations observed in our cohort. Despite these limitations, our findings offer biologically plausible pathways through which neighborhood-level factors may influence cancer biology.

Strengths of the research include census-tract-level ADI information, allowing for an accurate picture of neighborhood status. Additionally, the utilization of cancer-free controls allows us to separate cancer-related molecular changes from those potentially driven by chronic exposure of adverse neighborhood conditions. An additional methodological strength is the use of Olink’s proximity extension assay technology, which allows for exceptional specificity, thus reducing false positives. Furthermore, integrative analysis of -omics both in circulation and within tumor samples allows for a deeper understanding of how neighborhood status may impact patient outcomes through alterations in gene expression, protein, and metabolite profiles, laying the groundwork for future mechanistic studies.

5. Conclusions

Our findings revealed that ER+ breast cancer patients residing in disadvantaged neighborhoods exhibit distinct and unfavorable metabolomic, proteomic, and transcriptomic profiles. Among cancer-free controls residing in neighborhoods with a high ADI, we observed an upregulation of inflammatory proteins. This elevated inflammation was also seen in cancer patients, along with an upregulation of cancer-aggressiveness-associated proteins and genes, upregulation of genes involved in cell cycle regulation, and downregulation of cortisol metabolites. The potential mechanisms through which ADI impacts these analytes are multifactorial and unclear, but may include exposure to environmental toxins, decreased access to healthy foods, and increased levels of chronic stress. Further mechanistic research into ADI-associated factors driving tumorigenesis could lead to the development of novel therapeutics that may ultimately reduce the disproportionately high mortality experienced by Black women and residents of disadvantaged communities with ER+ breast cancer.

All raw targeted metabolomics data and metadata will be available at the NIH Common Fund’s National Metabolomics Data Repository (NMDR) website, the Metabolomics Workbench, https://www.metabolomicsworkbench.org where it has been assigned Study ID ST004145. The data can be accessed directly via its Project DOI: 10.21228/M8125S.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jproteome.5c00447.

• Principal component analysis and partial least-squares-discriminant analysis of metabolites in cases versus controls by tertiles (Supporting Information Figure 1); principal component analysis and partial least-squares-discriminant analysis of proteins in cases versus controls by tertiles (Supporting Information Figure 2); nonsignificant principal component analysis and partial least-squares-discriminant analysis of proteins in cases versus controls by tertiles (Supporting Information Figure 3); nonsignificant principal component analysis and partial least-squares-discriminant analysis of proteins across tertiles in cases only and controls only (Supporting Information Figure 4); and principal component analysis and partial least-squares-discriminant analysis of proteins across tertiles in cases only and controls only (Supporting Information Figure 5) (PDF)

• Lower and upper limits of quantification for quantified metabolites (Supporting Information Table 1); patient demographics for subset of transcriptomic cases (Supporting Information Table 2); patient demographics for subset of integrative analysis cases (Supporting Information Table 3); patient demographics and HbA1C values in cases versus controls for subset of samples included for HbA1C measurement (Supporting Information Table 4); patient demographics and HbA1C values across tertiles for subset of samples included for HbA1C measurement (Supporting Information Table 5); targeted metabolites, untargeted metabolites, and inflammatory protein levels by cases compared to controls (Supporting Information Table 6); mediation analysis testing HbA1C as a mediator of cancer status and inflammation by ADI tertile (Supporting Information Table 7); mediation analysis testing individual race as a mediator of cancer status and inflammation by ADI tertiles (Supporting Information Table 8); mediation analysis testing the food access index (FAI) as a mediator of cancer status and inflammation by ADI tertile (Supporting Information Table 9); metabolites and proteins in controls across ADI tertile (Supporting Information Table 10); metabolites and proteins in cases across ADI tertiles (Supporting Information Table 11); correlation analysis between HbA1C and uric acid and HbA1C and EN-RAGE (Supporting Information Table 12); and genes in cases across ADI tertiles (Supporting Information Table 13) (XLSX)

Z.M.-E., K.F.H., J.F., and H.H. conceived and designed the research; M.W.G., E.A., O.C.D., G.H.R., A.Z., N.P., E.L.H., S.F., S.K., J.J.K., and W.G. contributed to sample procurement. H.H., A.S.C., F.F., and H.M. performed experiments; H.H. and A.S.C. analyzed data; H.H. interpreted results of experiments; H.H. prepared the figures; H.H. and A.S.C. drafted the manuscript; H.H., A.S.C., and Z.M.-E. edited and revised manuscript. All authors reviewed the manuscript.

The authors declare no competing financial interest.