Inherited predisposition to increased systemic inflammation predicts a broad class of disease phenotypes

Summary

Chronic, low-grade systemic inflammation is a polygenic trait captured with the INFLA-score, a composite of C-reactive protein, platelet count, leukocyte count, and granulocyte-to-lymphocyte ratio. We derived a polygenic risk score from the INFLA-score (iPRS) in a multi-ancestry population from the UK Biobank (n = 421,368), then evaluated and used it in a phenome-wide association study among participants in the All of Us (AoU) Research Program. The multi-ancestry iPRS was tested for association with the INFLA-score in AoU (n = 4,833 with biomarker data) via linear regression, adjusting for age, sex, and genetically determined principal components (PCs), and with 2,821 phecodeX-defined phenotypes in AoU (n = 265,068) via logistic regression, adjusting for sex, age, electronic health record (EHR) length, race, ethnicity, and PCs. The iPRS predicted the INFLA-score (R² = 0.026, beta = 0.980, p < 2 × 10⁻¹⁶) and was associated with 47 phenotypes (Bonferroni-corrected p < 0.05). The strongest associations were with blood-related phenotypes: elevated white blood cell count (odds ratio [OR] = 1.19, p = 3.85 × 10⁻⁶⁶), thrombocytopenia (OR = 0.86, p = 5.70 × 10⁻⁴⁴), platelet defects (OR = 0.86, p = 2.47 × 10⁻⁴³), neutropenia (OR = 0.86, p = 5.52 × 10⁻¹⁸), and myeloproliferative disorder (OR = 1.2, p = 2.77 × 10⁻¹⁵). Others included celiac disease (OR = 0.713, p = 2.98 × 10⁻⁴⁶), ankylosing spondylitis (OR = 1.4, p = 1.33 × 10⁻¹⁷), hypertension (OR = 1.04, p = 4.56 × 10⁻¹⁵), rheumatoid arthritis (OR = 1.09, p = 1.02 × 10⁻¹³), and hematuria (OR = 1.05, p = 1.96 × 10⁻¹⁰). Removing major histocompatibility complex SNPs abolished associations with known autoimmune diseases, while all other associations remained. We replicated 17 (42.5%) of 40 significant phenotypes available in the Vanderbilt University Medical Center’s BioVU. Our findings demonstrate that systemic inflammation can be predicted using the iPRS across multiple ancestries, and the iPRS is associated with numerous clinical endpoints. This multi-ancestry iPRS may have future utility in stratifying risk for inflammation-driven conditions across diverse populations.

Chronic, low-grade systemic inflammation influences many common diseases. A polygenic risk score for inflammation (iPRS) was developed, and a PheWAS identified numerous associated phenotypes, including blood-related, autoimmune, and others. The iPRS may have utility in risk stratification for inflammation-driven conditions across diverse populations.

Introduction

Inflammation is an evolutionarily conserved biological process involving the activation of both immune and non-immune cells.¹ This process protects the host by eliminating potentially harmful stimuli, including bacteria, viruses, toxins, and other exposures, and subsequently promotes tissue repair and recovery. While acute inflammation is a protective and rapid response, chronic inflammation is a prolonged, dysregulated response that can lead to sustained tissue damage and contribute to the development of various diseases. In addition, chronic, low-grade systemic inflammation is a heritable, polygenic trait that can be observed in healthy individuals.²

The low-grade systemic inflammation index (INFLA-score) is a composite biomarker, first reported by Pounis et al., that integrates different components of the chronic inflammatory response process: high-sensitivity C-reactive protein (CRP), platelet count, leukocyte count, and the granulocyte-to-lymphocyte ratio.^3,4 Numerous studies have demonstrated the utility of the INFLA-score in predicting health outcomes across various populations. For example, Crotti et al. reported a 1% increase in overall mortality risk per unit increase in the INFLA-score (95% confidence interval [CI]: 1.00–1.03) and a 24% increase in overall mortality among those in the highest quartile (hazard ratio [HR] = 1.24; 95% CI = 1.01–1.52), with particularly elevated mortality risk in older obese adults.⁵ Furthermore, reduction in the INFLA-score following anti-inflammatory statin treatment has been shown to downregulate 8-oxoGuo, a biomarker associated with adverse health outcomes such as diabetes, hypertension, and cardiovascular disease (CVD).⁶

Chronic inflammation is associated with a wide spectrum of diseases, including CVD, cancer, diabetes mellitus, chronic kidney disease, non-alcoholic fatty liver disease, and autoimmune and neurodegenerative disorders.⁷ These diseases collectively account for a significant proportion of global morbidity and mortality. Despite the recognized role of inflammation in complex traits, the genetic underpinnings of chronic inflammation and their broader phenotypic consequences are not fully understood. Polygenic risk scores (PRSs) offer a promising tool for quantifying inherited risk by aggregating the effects of multiple genetic variants. While previous studies have established associations between inflammatory proteins in blood and specific diseases, it is often challenging to rule out reverse causality in such studies, as blood samples are not usually available prior to disease onset, particularly the onset of subclinical disease development processes. Thus, it is possible that disease processes drive increases in inflammatory proteins rather than the inflammatory proteins contributing to disease development. In addition, blood-based biomarkers are not often readily available in large population-based studies. Prior Mendelian randomization studies using genetic variants, which avoid the issue of reverse causality, focus on single inflammation-related proteins (e.g., SNPs associated with CRP) and may not capture the systemic inflammatory state as the INFLA-score was designed to do.⁸ Comprehensive phenome-wide investigations of genetic predisposition to chronic inflammation remain limited, particularly across diverse ancestral populations.⁹

To address this gap, a PRS for the INFLA-score (iPRS), developed previously among European-ancestry participants in the UK Biobank (UKB),¹⁰ was updated in a multi-ancestry cohort from the UKB and used to conduct a phenome-wide association study (PheWAS) across ancestrally diverse populations from the All of Us (AoU) Research Program, which was then validated in the Vanderbilt University Medical Center’s biobank, VUMC BioVU. This approach systematically examined associations across a broad range of phenotypes to identify inflammatory-related traits. In doing so, this work aimed to advance our understanding of the pleiotropic effects of genetic predisposition to chronic inflammation and to inform future screening and prevention strategies.

Subjects and methods

Development of a multi-ancestry iPRS in the UKB

We developed a multi-ancestry iPRS following the approach previously used for European-ancestry participants in the UKB.¹⁰ To group individuals into ancestries, we followed an approach similar to the HARE (harmonizing ancestry and race/ethnicity) method that was implemented in the Million Veterans Program (MVP).¹¹ Specifically, we used the principal components (PCs) that were supplied by the UKB as follows: we defined a training set consisting of individuals within the interquartile range (IQR) of PC1 and PC2 grouped by self-reported ethnic background (Data-Field 21000; White, Black or Black British, Chinese, and Asian or Asian British). The Asian/Asian British group had an oblique distribution on the PC1 vs. PC2 plot, so we fit an ordinary least-squares regression through PC1 and PC2 for this group, rotated PC1 and PC2 about this line, and then calculated the IQR training set accordingly for the Asian/Asian British group. A radial basis support vector machine (SVM) was then trained on PC1–PC15 on the training data (with ancestry labels corresponding to European, African, East Asian, and Central-South Asian) and then applied to the held-out samples. For individuals where the probability of the predicted ancestry from the SVM was >0.95, we assigned the predicted ancestry label. This analysis resulted in 403,584 individuals of European ancestry, 7,103 of African ancestry, 1,550 of East Asian ancestry, and 9,131 of Central-South Asian ancestry.

INFLA-scores were calculated for each participant following the methods in Pounis et al.^3,4 and using ancestry-specific decile limits. The imputed genetic data provided by the UKB underwent rigorous pre- and post-imputation quality control (QC), as detailed previously, and were subsequently filtered within each ancestry group to remove SNPs with a minor allele count <20 and SNPs with a minor allele frequency <0.01.¹² A genome-wide association study (GWAS) of the INFLA-score was performed within each ancestry group using linear regression adjusted for age, sex, and 10 PCs (provided by the UKB) to adequately control for any residual confounding by population structure, using REGENIE v.4.1 with the --step 2, --ignore-pred, and --bsize 200 parameters, which instruct REGENIE to ignore step 1 and to use a block size of 200 to determine the number of variants processed together in memory.¹³ GWAS results across ancestry groups were meta-analyzed using the --meta-analysis command in PLINK v.1.9, which employs fixed-effects inverse-variance weighted meta-analysis. The multi-ancestry meta-analysis GWAS included 421,368 individuals with complete genetic data for 8,367,092 SNPs.

To construct the iPRS, we took an approach similar to other PRS studies of coronary artery disease and major depressive disorder, where they found that multi-ancestry GWAS meta-analysis followed by clumping + thresholding PRS performed similarly to other, more complex PRS methods.^14,15 PRSice-2 was used to perform linkage-disequilibrium-based clumping (250 kb blocks, r² threshold = 0.1) using a multi-ancestry cohort comprising a randomly selected 10% of each ancestry group from the included UKB participants.¹⁶ The lead SNP within each clump reaching genome-wide significance (p < 5 × 10⁻⁸) was selected for inclusion in the multi-ancestry iPRS. After clumping and thresholding using the GWAS meta-analysis output, 1,189 SNPs were selected for inclusion in the iPRS (Table S1).

iPRS evaluation and PheWAS in the AoU Research Program

AoU enrolls adults 18 years old and older who reside in the USA or a USA territory and collects longitudinal electronic health records (EHRs), genomic data, survey data, physical measurements, and data from wearables to create a repository that enables researchers to study the full range of factors that influence health and disease. The present study used Curated Data Repository release v.8 (All of Us Research Program: CDRv8; Controlled Tier C2024Q3R4 and Registered Tier R2024Q3R3) and included participants with longitudinal EHR data, short-read whole-genome sequencing (srWGS) data, and inferred genetic ancestry information and was restricted to participants who identify as female or male at birth.¹⁷ The AoU Research Program is approved by an institutional review board (IRB) that follows the regulations and guidance of the NIH Office for Human Research Protections for all studies, ensuring that the rights and welfare of research participants are overseen and protected uniformly. Analysis was performed in the secure, cloud-based AoU Researcher Workbench (https://workbench.researchallofus.org).¹⁸

iPRS construction in AoU

HAIL v.0.2.134 was used to extract SNPs from the AoU srWGS genetic database. Of the 1,189 SNPs in the multi-ancestry iPRS, 1,182 SNPs passed QC and were available for 280,897 individuals; the 7 unavailable SNPs were omitted from iPRS construction. The iPRS was computed for each AoU participant by summing the number of risk alleles for each SNP, weighted by their effect sizes derived from the UKB GWAS meta-analysis by applying the PLINK v.1.9 --score function. The iPRS was standardized in all analyses by subtracting the mean and dividing by the standard deviation (SD); thus, the effect sizes in regression analyses are interpreted as per one unit increase in SD of the iPRS.

iPRS evaluation

Participants in AoU with available genotype and inflammatory biomarker data for all four measures after the following QC steps (n = 4,833) were included in the analysis to validate the iPRS against the INFLA-score. QC filters were applied to remove implausible values (retaining only 0.011–98 mg/L CRP, 0.1 < x < 100 white blood cells [WBCs], 17–964 platelet count, 0.126–7,909 granulocyte/lymphocyte ratio). Included samples were those with all measures within a year of a CRP measure, and for samples with multiple measures, we selected the set that minimized the average time difference to the CRP measure. INFLA-scores were then calculated following the methods in Pounis et al., as above. Linear regression models were used to test the association between the iPRS with the INFLA-score (the dependent variable), adjusted for age at the time of the CRP measure, sex, and 10 PCs. Additional analyses were performed stratified by ancestry cluster, with derivation of ancestry clusters in AoU previously described.¹⁹

Phenotypic mapping and PheWAS analysis

The PheTK package (v.0.1.47) was used to retrieve phenotypic data and map the International Classification of Diseases (ICD) codes (ICD-9-CM and ICD-10-CM) to phecodeX²⁰ in the AoU dataset. An individual was considered affected for a particular phecode when assigned two or more instances of that phecode. Phecodes having fewer than 20 affected individuals or control subjects were excluded from each analysis, and all logistic regression models successfully converged. Phecodes were classified in 18 categories including blood/immune, CV, congenital, dermatological, endocrine/metabolic, gastrointestinal, genetic, genitourinary, infections, mental, musculoskeletal, neonatal, neoplasms, neurological, pregnancy (excluded in the males-only analysis), respiratory, sense organs, and symptoms.

The PheTK package within the AoU Researcher Workbench was used to run logistic regression to test associations between phecode and the iPRS (independent variable) iteratively across available phecodes after adjustment for sex at birth, the first 5 PCs, EHR length, age at last EHR entry, race, and ethnicity. The first 5 PCs were deemed sufficient to control any residual confounding by ancestry, given that such confounding was already controlled in the derivation of the iPRS, as described above. EHR length was defined for each individual by first calculating the start date of the earliest recorded condition and the end date of the latest recorded condition. If the end date of the latest condition was missing, the start date for the latest condition was used as a proxy for the end date. This calculation captures EHR length across all documented conditions in the dataset and represents the measured time of follow-up for all phenotypes. This provided a time frame indicating how long each patient’s record spanned. Race and ethnicity were self-reported and were included to capture possible variability in environmental and health behavior exposures as well as healthcare access, while PCs were included to control for population stratification. Race categories included American Indian or Alaska Native, Asian, Black or African American, Middle Eastern or North African, Native Hawaiian or other Pacific Islander, White, more than one population, and unknown. Ethnicity was categorized as Hispanic or Latino, not Hispanic or Latino, or unknown. A Bonferroni correction for multiple comparisons based on the number of phenotypes analyzed resulted in a significance level of α = 1.77 × 10⁻⁵ (0.05/2,821 phecodes).

All analyses in AoU were conducted using a Jupyter Notebook environment. The notebook server was running v.6.5.4 and executed using Python 3.10.16 (packaged by conda-forge, built with GCC 13.3.0). In this environment, R v.4.5.0 (released April 11, 2025) was used as the computational kernel for data analysis and visualization.

Post-PheWAS statistical analysis

Shared genetic predisposition to hypertension

Hypertension was selected for secondary analysis as an example of a phenotype for which a strong purpose-built PRS was available. We tested whether the association between the iPRS and the phecodes for hypertension and essential hypertension remained after adjusting for PGS Catalog: PGP000332, trained on hypertension.²¹ This was done by fitting a logistic regression model that included the same covariates used in the PheWAS (sex at birth, the first 5 PCs, EHR length, age at last EHR entry, race, and ethnicity), the iPRS, and PGP000332.

Contribution of SNPs from the major histocompatibility complex region

The iPRS was recalculated after excluding variants located within the major histocompatibility complex (MHC) region (chr6: 28510020–33480577 [hg38]). Among the 165 SNPs located on chromosome 6, 106 were within the MHC region and were therefore excluded, while 59 SNPs on chromosome 6, outside this region, were retained. After excluding the MHC SNPs, the final iPRS comprised 1,083 SNPs, of which 1,077 variants were available for 280,897 individuals in AoU. The PheWAS was repeated to assess the influence of SNPs in the MHC region on observed associations, and a scatterplot was generated to compare the −log10 p values from the PheWAS analysis using the full iPRS SNPs vs. the reduced version excluding the SNPs from the MHC region.

Ancestry-wise evaluation of iPRS and elevated WBC phenotype

An ancestry-stratified analysis assessed whether the relationship between the iPRS and elevated WBC (EWBC) count was consistent across broad ancestries defined by the ancestry_pred column in the srWGS genetic predicted-ancestry TSV file, with the South Asian and East Asian groups combined due to small sample size. To ensure comparability across groups, each ancestry subset was downsampled to the smallest available group size (n = 7,978 for the combined South Asian and East Asian group). Only participants with complete covariate data and either presence or absence of the phecode for EWBC were included. Participants were classified as affected individuals (with the EWBC phecode) if 2 or more instances were recorded or as unaffected subjects (without the phecode) if fewer than 2 instances of the phecode were recorded. Within each ancestry group, a logistic regression model was performed to evaluate the association between standardized iPRS and EWBC affected/unaffected status.

Validation of the iPRS PheWAS in BioVU

The VUMC BioVU is a de-identified biobank linking longitudinal EHR data with clinical biospecimens and DNA.²² Genetic data were ascertained using the Multi-ethnic Genotyping Array (MEGAEx). All data underwent standard QC procedures, phasing, and imputation using the TOPMed r3 reference panel. Results for 88,212 subjects of genetically determined African and European ancestry (n = 15,616 and 72,596, respectively) based on PCs anchored by 1000 Genomes were included in this analysis. Individual risk scores were constructed from 1,185 (99.7%) of the 1,189 iPRS variants and associated weights using the score function in PLINK v.1.9. We then performed a PheWAS using phecodeX codes adjusted for sex at birth, PCs 1–5, length of EHR record, age at last ICD, and EHR-reported race and ethnicity.²⁰ PhecodeX phenotypes were defined using the same affected and unaffected definitions as in the AoU PheWAS, and only phecodes with 20 or more affected individuals or control subjects were considered, as in the AoU PheWAS. A Bonferroni correction for multiple comparisons based on the number of phenotypes to be validated resulted in a significance level of α = 1.25 × 10⁻³ (0.05/40 phecodes available in the VUMC BioVU).

Results

AoU cohort characteristics

The PheWAS cohort consisted of 265,068 participants in the AoU Research Program with complete genetic and EHR data. The mean EHR length was 9 years (3,340 days; median: 7.5 years; SD: 8.1 years; Figure S1A) and was right skewed, with most participants having a relatively short EHR length, as previously reported in the AoU Research Program.¹⁹ The mean age at last encounter in the EHR was 54.8 years (median: 57 years; SD: 17.4; Figure S1B). Table 1 describes participants by gender, ethnic group, race, age at last encounter, and EHR length. The cohort was predominantly female (62%) and evenly distributed across the adult age range. Most participants identified as not Hispanic or Latino (79%), and the largest racial groups were White (57%) and Black or African American (16%), with smaller proportions of participants identifying as Asian (3%), American Indian or Alaska Native (1%), Native Hawaiian or other Pacific Islander (1%), Middle Eastern or North African (0.5%), and more than one racial group (4%).

Table 1.

Characteristics of 265,068 participants in the AoU cohort

Characteristic	Total n = 265,068
Gender, n (%)
Female	164,955 (62)
Male	100,113 (38)
Ethnicity, N (%)
Hispanic/Latino	48,374 (18)
Not Hispanic/Latino	209,680 (79)
Not reported	7,014 (3)
Race, n (%)
American Indian or Alaska Native	3,780 (1)
Asian	6,764 (3)
Black or African American	43,244 (16)
Middle Eastern or North African	1,439 (<1)
More than one population	11,224 (4)
Native Hawaiian or other Pacific Islander	285 (<1)
White	150,678 (57)
Not reported	47,654 (18)
Genetic ancestry, n (%)
Admixed American	47,646 (18)
African	50,812 (19)
Asian	7,978 (3)
European	157,650 (59)
Middle Eastern	982 (<1)
Age at last encounter, n (%)
18–25	12,549 (5)
26–35	33,616 (13)
36–45	37,091 (14)
46–55	41,833 (16)
56–65	55,411 (21)
66–75	53,646 (20)
76 and older	30,126 (11)
EHR length (years), median (range)	7.5 (0–42.8)

Gender, ethnicity, and race are self-reported; age at last encounter and EHR length were defined based on EHR data. Genetic ancestry labels are based on principal component analysis and were provided by the AoU Research Program.

Multi-ancestry iPRS evaluation in AoU

Regression analyses were performed between the iPRS, trained in the UKB, and the INFLA-score calculated among 4,833 participants in AoU with complete biomarker data to confirm that the iPRS accurately captures systemic inflammation in this independent cohort. The iPRS was normally distributed across the population as shown in Figure S2. The association between the PRS and the INFLA-score showed a positive statistically significant relationship among all participants (n = 4,833; beta = 0.980, R² = 0.026, 95% CI = 0.797–1.162, p < 2 × 10⁻¹⁶) and when the cohort was stratified by ancestry group (African, n = 713, beta = 1.325, 95% CI = 0.626–2.024, p = 2.19 × 10⁻⁴; admixed American, n = 920, beta = 1.096, 95% CI = 0.641–1.550, p = 2.74 × 10⁻⁶; and European, n = 3,026, beta = 0.928, 95% CI = 0.721–1.136, p < 2 × 10⁻¹⁶) (detailed results in Table S2).

Phenotypes associated with the iPRS

Of the 3,427 phecodes extracted from the AoU cohort, 2,821 had more than 20 affected individuals in the overall cohort and were included in PheWAS analysis. In the overall cohort, 47 phecodes constituting 13 disease categories (musculoskeletal, gastrointestinal, blood/immune, mental, neoplasms, neurological, CV, dermatological, endocrine, genitourinary, infections, respiratory, and symptoms) were significantly associated with iPRS after Bonferroni correction (p < 1.77 × 10⁻⁵) (Figure 1A; Table 2; complete PheWAS results are provided in Table S3). The QQ plot for association showed upward deviation from the expected line only at the top-right region, indicating minimal inflation and the presence of true associations between the iPRS and multiple phenotypes (Figure 1B).

Figure 1.

INFLA-score PheWAS results

The Manhattan plot is shown in (A) and accompanying QQ plot in (B).

Table 2.

The 47 phenotypes significantly associated with the iPRS

Phecode name	Phecode category	No. of affected individuals	No. of unaffected individuals	Odds ratio (95% CI)	p
Elevated white blood cell count (leukocytosis)	blood/immune	10,973	241,876	1.19 (1.17–1.22)	3.85 × 10−66
Decreased white blood cell count	blood/immune	7,062	253,752	0.82 (0.8–0.84)	2.55 × 10−56
Celiac disease	gastrointestinal	1,526	262,890	0.71 (0.68–0.75)	2.98 × 10−46
Thrombocytopenia	blood/immune	8,859	250,860	0.86 (0.84–0.88)	5.70 × 10−44
Platelet defects	blood/immune	9,056	250,549	0.86 (0.84–0.88)	2.47 × 10−43
Essential thrombocythemia	neoplasms	1,192	262,912	1.4 (1.32–1.49)	3.75 × 10−26
Thrombocytosis	blood/immune	680	263,736	1.49 (1.37–1.61)	3.61 × 10−21
Neutropenia	blood/immune	3,657	259,411	0.86 (0.83–0.89)	5.52 × 10−18
Intestinal malabsorption	gastrointestinal	3,832	258,853	0.87 (0.85–0.9)	7.77 × 10−18
Ankylosing spondylitis	musculoskeletal	700	264,000	1.4 (1.29–1.51)	1.33 × 10−17
Myeloproliferative disorder	neoplasms	2,153	261,430	1.2 (1.14–1.25)	2.77 × 10−15
Hypertension	cardiovascular	115,253	131,903	1.04 (1.03–1.05)	4.56 × 10−15
Essential hypertension	cardiovascular	111,867	135,347	1.04 (1.03–1.05)	7.32 × 10−14
Rheumatoid arthritis and other inflammatory polyarthropathies	musculoskeletal	9,086	252,345	1.09 (1.06–1.11)	1.02 × 10−13
Myeloid	neoplasms	3,047	260,103	1.14 (1.1–1.19)	2.28 × 10−12
Rheumatoid arthritis	musculoskeletal	6,532	256,018	1.09 (1.06–1.12)	5.95 × 10−11
Gastro-esophageal reflux disease	gastrointestinal	69,486	175,035	1.03 (1.02–1.04)	1.14 × 10−10
Hematuria	genitourinary	22,044	228,541	1.05 (1.03–1.07)	1.96 × 10−10
Chronic obstructive pulmonary disease	respiratory	19,183	238,450	1.05 (1.04–1.07)	1.97 × 10−10
Psoriasis	dermatological	6,198	256,450	1.08 (1.06–1.11)	4.28 × 10−10
Papulosquamous disorders	dermatological	7,422	253,809	1.08 (1.05–1.1)	6.25 × 10−10
Lupus	musculoskeletal	3,292	260,799	0.89 (0.99–1.1)	1.05 × 10−09
Mood (affective) disorders	mental	76,802	171,881	1.03 (0.71–1.03)	5.37 × 10−09
Cutaneous lupus erythematosus	musculoskeletal	1,017	263,579	0.82 (1.01–1.06)	1.05 × 10−08
Systemic lupus erythematosus	musculoskeletal	3,067	261,068	0.9 (0.99–1.08)	4.07 × 10−08
Anxiety and anxiety disorders	mental	70,850	173,616	1.03 (1.03–1.24)	4.11 × 10−08
Gastrointestinal inflammation	gastrointestinal	32,265	211,368	1.04 (0.98–1.29)	1.08 × 10−07
Diffuse diseases of connective tissue	musculoskeletal	8,456	253,027	0.94 (0.64–1.04)	1.57 × 10−07
Rheumatoid arthritis with rheumatoid factor	musculoskeletal	2,105	262,558	1.13 (1–1.03)	2.63 × 10−07
Fever	symptoms	16,261	228,186	1.04 (1.02–1.06)	5.62 × 10−07
Demyelinating diseases of the central nervous system	neurological	2,743	261,612	0.91 (1–1.03)	6.59 × 10−07
Major depressive disorder	mental	53,876	195,652	1.02 (1–1.03)	2.70 × 10−06
Nicotine dependence (current and history of)	mental	56,010	191,453	1.02 (0.97–1)	3.36 × 10−06
Fatty liver disease	gastrointestinal	12,213	243,778	1.05 (0.97–1.36)	3.45 × 10−06
Symptoms involving digestive system	gastrointestinal	64,163	171,804	1.02 (0.99–1.12)	3.95 × 10−06
Dizziness and giddiness	neurological	36,712	204,041	1.03 (0.98–1.26)	4.80 × 10−06
Abnormality of urination	genitourinary	61,575	178,128	1.02 (0.98–1.23)	5.42 × 10−06
Hemo onc - by cell of origin	neoplasms	6,847	255,273	1.06 (1–1.04)	6.25 × 10−06
Current tobacco use and nicotine dependence	mental	37,479	214,377	1.03 (0.93–1.01)	7.64 × 10−06
Constipation	gastrointestinal	30,166	214,500	1.03 (1.01–1.06)	8.49 × 10−06
Crohn disease	gastrointestinal	2,606	261,661	1.09 (0.94–1.85)	8.76 × 10−06
Disorders of fluid, electrolyte, and acid-base balance	endocrine/metabolism	44,351	200,810	1.03 (0.85–1.02)	9.64 × 10−06
Sepsis	infections	12,587	247,727	1.04 (0.96–1.6)	9.80 × 10−06
Bacteremia, sepsis, and SIRS	infections	15,099	244,461	1.04 (0.97–1)	1.02 × 10−05
Diffuse goiter	endocrine/metabolism	2,699	261,063	0.91 (0.72–1.03)	1.12 × 10−05
Inflammatory spondylopathies	musculoskeletal	5,420	255,630	1.06 (1–1.02)	1.32 × 10−05
Multiple sclerosis	neurological	2,399	262,128	0.91 (0.92–0.99)	1.35 × 10−05

All p values are <0.05 after Bonferroni correction for 2,821 phenotypes tested.

Among the 47 significantly associated phenotypes, the strongest associations were observed with blood cell phenotypes: EWBC count, thrombocytopenia, platelet defects, neutropenia, and myeloproliferative disorder. Strong negative associations were also found with celiac disease and intestinal malabsorption, and a strong positive association was found with ankylosing spondylitis. Several phenotypes had odds ratios (ORs) indicating a greater than 10% increased odds per SD of the iPRS, including thrombocytosis (OR = 1.49), essential thrombocythemia (OR = 1.40), ankylosing spondylitis (OR = 1.40), myeloproliferative disorder (OR = 1.20), EWBC count (OR = 1.19), myeloid neoplasm (OR = 1.14), rheumatoid arthritis (OR = 1.09) and Crohn disease (OR = 1.09). While most of the phenotypes showed an increased relative odds per SD of the iPRS, 13/47 phecodes showed a negative association (Table 2). Three of these are expected, as they reflect measures of reduced blood cell counts: decreased WBC count (OR = 0.82), thrombocytopenia (OR = 0.86), and platelet defects (OR = 0.86). However, three were less expected, as they are phenotypes with documented relationships to increased inflammation: celiac disease (OR = 0.71), lupus (OR = 0.89), and intestinal malabsorption (OR = 0.87). To assess whether the iPRS might be useful as a risk-stratification biomarker for the association phenotypes, we plotted associations with iPRS decile (Figure S3). For some phenotypes (e.g., MS_706.2, ankylosing spondylitis), we see an approximately linear increase in OR with increasing iPRS decile, whereas for others there appears to be a natural threshold where odds of developing the phenotype are increased only at the highest one or two deciles of iPRS (e.g., CA_120, Hemo onc - by cell of origin) or at all deciles of iPRS compared to the first decile (e.g., GI_529.5, constipation).

Table 3 shows the 47 Bonferroni-significant phecodes per phecode category. The highest number of Bonferroni-significant associations was observed for the musculoskeletal category (9/47), followed by the gastrointestinal (8/47), blood/immune (6/47), and mental health (5/47) categories. Table S4 presents the total count of phecodes per category across the full PheWAS cohort (n = 2,821).

Table 3.

Distribution of Bonferroni-significant phenotypes by phecode category

Phecode category	N (% of 47)
Musculoskeletal	9 (19)
Gastrointestinal	8 (17)
Blood/immune	6 (13)
Mental	5 (11)
Neoplasms	4 (9)
Neurological	3 (6)
Genitourinary	2 (4)
Endocrine/metabolism	2 (4)
Infections	2 (4)
Cardiovascular	2 (4)
Dermatological	2 (4)
Respiratory	1 (2)
Symptoms	1 (2)

Shared genetic predisposition to hypertension

Hypertension was selected as an example phenotype to assess whether the iPRS (and indirectly low-grade systemic inflammation) captures an independent portion of susceptibility not already captured by a PRS purpose-built for the given phenotype. The established hypertension PRS PGP000332 was significantly associated with the hypertension phecode (OR = 1.12, p < 2 × 10⁻¹⁶, Table S3), confirming its utility in this cohort. In a multivariable model adjusted for the same covariates as in the PheWAS, the association between the iPRS and hypertension phecodes remained after adjusting for PGP000332 (Table 4), although the strength of association was attenuated. The variance inflation factor was 1.01, indicating very little collinearity between the iPRS and PGP000332. Thus, the iPRS captures some component of inherited risk for hypertension beyond that represented in an established hypertension PRS.

Table 4.

Odds ratios and p values for hypertension-related phecodes with the iPRS before and after hypertension PRS adjustment

	Before adjusting for hypertension PRS		After adjusting for hypertension PRS
	Odd ratio	p	Odds ratio	p
Hypertension	1.04	4.56 × 10−15	1.03	1.99 × 10−8
Essential hypertension	1.04	7.32 × 10−14	1.03	1.14 × 10−7

Contribution of SNPs from the MHC region

Many of the non-hematologic phecodes associated with the iPRS are established autoimmune diseases. Therefore, we sought to clarify the extent to which an iPRS that did not include SNPs in the MHC region would predict these autoimmune phenotypes. The iPRS was recalculated after removing SNPs located within chr6: 28510020–33480577 and then tested via PheWAS across the same phenotypic set in AoU. After Bonferroni correction, 27 of the 47 significant phenotypes remained significant (Figure 2; phecodes showing differential association based on MHC exclusion are in Table S5, and full results are in Table S3). Of the 20 phenotypes that were no longer significant (after Bonferroni correction) in the MHC-excluded analysis, several had a nominally significant p < 0.05 with an effect size largely consistent with the original analysis; however, attenuation in effect sizes was observed for phenotypes previously linked to immune-related pathways, such as celiac disease (OR before = 0.71, p = 2.98 × 10⁻⁴⁶; OR after = 1.01, p = 0.762), intestinal malabsorption (OR before = 0.87, p = 7.77 × 10⁻¹⁸; OR after = 1.01, p = 0.597), ankylosing spondylitis (OR before = 1.40, p = 1.33 × 10⁻¹⁷; OR after = 1.13, p = 2.41 × 10⁻³), and rheumatoid arthritis (OR before = 1.09, p = 5.95 × 10⁻¹¹; OR after = 1.04, p = 1.61 × 10⁻³) (Table S5). The influence of the MHC region on PheWAS results can be further visualized in the scatterplot shown in Figure 2B with the −log₁₀(p values) for each phecode from the original iPRS (x axis) against the corresponding values from the MHC-excluded iPRS (y axis). The top 10 phecodes with the largest absolute differences in −log10(p value) between the two analyses were identified and labeled on the plot. Most points clustered near the diagonal line on the scatterplot, suggesting overall consistency between the analyses. However, a subset of phenotypes, including those mentioned above, showed notable attenuation in significance after excluding MHC SNPs, indicating that some associations were partially or largely driven by genetic variation in the MHC region.

Figure 2.

PheWAS results after removing MHC-region SNPs from the iPRS

(A) Manhattan plot and (B) QQ plot comparing results to those from the PheWAS of the full iPRS (B). The phenotypes labeled in the QQ plot are the top ten for which the difference in −log10(p value) between the two PheWAS was greatest.

Ancestry-specific association of the iPRS with EWBC count

The iPRS includes the ACKR1 intronic variant chr1:159205095T>C, which is in linkage disequilibrium (LD) with the two variants defining the highly ancestry-differentiated Duffy blood group. Given the influence of the Duffy blood group on neutrophils,²³ the largest component of WBCs, it is possible that the association between the iPRS and EWBC count is due to residual confounding by ancestry in derivation of the iPRS despite application of a multi-ancestry GWAS approach. To address this question, an ancestry-stratified logistic regression analysis was performed to assess whether EWBC count remained significantly associated with the iPRS across different genetically determined ancestries. Analyses were stratified into three broad ancestry groups with sample sizes limited to the size of the smallest group (e.g., 7,978 individuals per group) to ensure balanced comparisons. Within each ancestry group, affected-unaffected distributions were relatively similar, and the standardized iPRS remained associated with EWBC count, although the strength of association was attenuated among individuals of European ancestry (Table 5). This finding indicates the iPRS captures some component of genetic susceptibility to EWBC count across ancestries and is not simply a surrogate for the Duffy blood group.

Table 5.

Association between elevated white blood cell count and iPRS across three ancestries

Ancestry group (n affected individuals/n unaffected individuals)	OR (95% CI)	p
African (275/7,703)	1.01 (1.003–1.014)	0.0039
Asian (211/7,767)	1.01 (1.003–1.014)	0.0013
European (317/7,661)	1.00 (1.000–1.008)	0.0376

Ancestry group shows the three broad groups. South Asians were grouped with East Asians. Each ancestry group was downsampled to a total sample size of 7,978.

Validation of the iPRS PheWAS in BioVU

A validation PheWAS among participants in the BioVU cohort (n = 88,212; 15,616 of African ancestry and 72,596 of European ancestry) replicated 17 (42.5%) of the 40 significant phenotypes identified in the AoU cohort with sufficient affected individuals (out of the 47 phenotypes identified in the AoU PheWAS; Figure 3; Table S6). These replicated phenotypes showed similar strength of association and direction of effect as observed in AoU. For example, positive associations were found for EWBC count (OR = 1.07, 95% CI = 1.06–1.09), ankylosing spondylitis (OR = 1.28, 95% CI = 1.18–1.38), hypertension (OR = 1.02, 95% CI = 1.01–1.03), rheumatoid arthritis (OR = 1.04, 95% CI = 1.01–1.07), and myeloproliferative disorder (OR = 1.04, 95% CI = 1.01–1.08), while negative associations were found for celiac disease (OR = 0.79, 95% CI = 0.75–0.83), multiple sclerosis (OR = 0.92, 95% CI = 0.90–0.95), and systemic lupus erythematosus (OR = 0.93, 95% CI = 0.90–0.95). Seven of the significant phenotypes in AoU were not available in the BioVU cohort. In addition, there were eight phenotypes that were significant in the BioVU cohort (after applying a Bonferroni correction) that were not significant in the AoU PheWAS but showed a consistent association (Table S6). These included known inflammatory conditions, including cholangitis (OR = 0.88, p = 2.23 × 10⁻⁶), sicca syndrome (OR = 0.9, p = 5.34 × 10⁻⁶), chronic hepatitis (OR = 0.09, p = 3.84 × 10⁻⁵), and juvenile rheumatoid arthritis (OR = 1.12, p = 5.84 × 10⁻⁵).

Figure 3.

Replication PheWAS in the VUMC BioVU

Discussion

This study aimed to assess the association of genetic susceptibility to inflammation, captured by the iPRS, and a broad range of clinical phenotypes derived from a diverse cohort within the AoU Research Program. Among the 47 Bonferroni-significant phecodes, the most frequently represented categories were musculoskeletal (9/47), gastrointestinal (8/47), blood/immune (6/47), and mental health (5/47). In total, 13 unique phecode categories were identified, demonstrating that the iPRS is associated with a broad spectrum of phenotypes. The replication analysis in BioVU confirmed and further strengthened our initial findings for more than a third of the phenotypes, including celiac disease, which was the most significant association (p = 6.99 × 10⁻²⁵), several of the blood/immune phenotypes (e.g., EWBC count, p = 2.27 × 10⁻²⁰; decreased WBC count, p = 1.14 × 10⁻⁷; and thrombocytopenia, p = 2.26 × 10⁻⁵), several of the musculoskeletal phenotypes (e.g., ankylosing spondylitis, p = 1.01 × 10⁻⁹; rheumatoid arthritis and other inflammatory polyarthoropathies, 2.23 × 10⁻⁴; and lupus, p = 4.33 × 10⁻⁸), and others discussed further below. Several phenotypes did not pass the threshold for Bonferroni correction in BioVU but showed marginal association signals and were within the same disease class as other replicated phenotypes, for example myeloproliferative disorder (p = 4.77 × 10⁻³), neutropenia (p = 5.35 × 10⁻³), and Crohn disease (p = 5.53 × 10⁻³). These findings further support that inflammation plays a key role in the pathogenesis of multiple complex diseases.

Significant associations with large effect sizes were observed for hematological/blood cell traits, such as EWBC count, thrombocytopenia, neutropenia, and platelet defects. These associations are expected because the INFLA-score is partially defined by these phenotypes and affected-unaffected status in the PheWAS is based on extreme measures of these same phenotypes. Such associations reflect simple overlap with the INFA-score definition. In contrast, associations with other hematologic traits, such as myeloproliferative disorder, do not suffer from this circularity and likely represent biological associations with genetic predisposition to inflammation. In addition, the iPRS was strongly associated with chronic inflammatory conditions, including celiac disease, hypertension, chronic obstructive pulmonary disease (COPD) and gastro-esophageal reflux disease. These findings confirm that low-grade systemic inflammation contributes to a broader range of complex disorders, in line with existing literature. For example, inflammation plays a central role in COPD pathogenesis, with inflammation predominantly affecting the lung parenchyma and peripheral airways, resulting in largely irreversible and progressive airflow limitation.²⁴ This inflammation is characterized by increased numbers of alveolar macrophages, neutrophils, and T lymphocytes. While this phecode was not available in the BioVU replication cohort, the biological plausibility supports the likely true positive association in AoU. Similarly, chronic inflammation is involved in the pathogenesis of hypertension,²⁵ which was significant in the AoU PheWAS and replicated in BioVU. Innate and adaptive immunity contribute to hypertension through vascular inflammation and microvascular remodeling. M1 macrophages, neutrophils, and dendritic cells promote inflammation via cytokine release, while T cell-derived interferon-gamma (IFN-γ) and interleukin-17 (IL-17) exacerbate oxidative stress and endothelial dysfunction. These findings reinforce the relevance of inflammation as an underlying mechanism in complex diseases.

The PheWAS using the MHC-excluded iPRS indicated that SNPs in this region explain the majority of the relationship of the iPRS with known autoimmune traits such as celiac disease, rheumatoid arthritis, intestinal malabsorption, and ankylosing spondylitis. These results are in line with previous research suggesting most autoimmune diseases have a strong human leukocyte antigen (HLA) association. For example, numerous studies suggest a strong association between celiac disease and the MHC, with HLA-B8, HLA-A1, HLA-DR3, HLA-DR7, and HLA-DQ2 implicated.¹⁶ Similarly, ankylosing spondylitis is a highly heritable inflammatory arthritis, with numerous genetic associations that have been identified, with HLA-B27 contributing to approximately 20% of the heritability.²⁶ The HLA associations with ankylosing spondylitis are complex, and multiple non-B27 HLA alleles have been identified as being involved. Rheumatoid arthritis is also HLA driven, with the strongest genetic association located at the HLA-DRB1 gene locus, which is an HLA class II gene.²⁷ However, our HLA-region-excluded iPRS sensitivity analysis suggests that while the MHC region contributes to the inflammatory signal, it does not entirely drive the associations observed between the iPRS and most of our tested phenotypes. The iPRS retains predictive power across a wide range of phenotypes even when the MHC region is excluded. From a screening perspective, the MHC region remains one of the strongest genetic risk factors for autoimmune disorders such as lupus.²⁸ However, the precise contribution of MHC variants to overall disease susceptibility relative to non-MHC polygenic components remains uncertain. Future work integrating iPRS with MHC-specific risk estimates could therefore improve genetic screening and early identification of individuals at high risk of autoimmune vs. inflammatory disease subtypes, potentially guiding preventative or pre-emptive treatment strategies.

Given the multi-ancestry composition of the AoU cohort, the association between the iPRS and EWBC count, the most significant phenotype, was evaluated to verify that the observed association was not disproportionately influenced by any one ancestry group. The results confirmed no single ancestry group exclusively drove the association between the iPRS and EWBC count.

The iPRS was significantly associated with essential hypertension and hypertension in both AoU and BioVU, consistent with the well-established overlap between inflammation and CV risk. Our analyses aimed at disentangling the effects of shared genetic architecture between the iPRS and hypertension PRS suggest that while the iPRS captures some hypertension-related genetic risk, there remains a residual signal likely attributable to distinct inflammatory pathways. In contrast, no significant associations were observed between the iPRS and other CV phenotypes (all p > 0.001 in both AoU and BioVU) such as carditis, atherosclerotic heart disease, or myocardial infarction, despite their well-established inflammation component²⁹ and adequate numbers of affected individuals (e.g., atherosclerotic heart disease, 29,721 affected individuals; myocardial infarction, 11,137 affected individuals; and carditis, 2,625 affected individuals in AoU). However, these observations are not inconsistent with prior findings. While some Mendelian randomization studies have demonstrated evidence for a causal association between IL-6 levels and CVD,^30,31 numerous studies have reported null results for a causal link between CRP and CVD.^31,32 The evidence for WBC traits exerting causal effects on CVD is more mixed.³³

Significant associations were found between iPRS and myeloid/blood neoplasms (e.g., myeloproliferative disorder; p = 2.77 × 10⁻¹⁵ in AoU and p = 4.77 × 10⁻³ in BioVU), while solid cancers showed no significant association (e.g., colorectal cancer, p = 0.03 in AoU and p = 0.11 in BioVU; breast neoplasms/cancers, p = 0.629 in AoU and p = 0.98 in BioVU; and solid tumors, p = 0.73 and not evaluated in BioVU). It is biologically plausible that this iPRS would show stronger associations with blood-related neoplasms, including myeloid malignancies, given the shared biological pathways involving chronic inflammation, immune dysregulation, and hematopoietic cell proliferation. In contrast, the weaker associations observed with solid tumors may reflect the more indirect or tissue-specific roles of inflammation in solid tumorigenesis, which may not be fully captured by this iPRS, which was derived from blood-based markers.

Our study has several notable strengths. The AoU Research Program, which includes individuals from diverse genetic ancestries, racial and ethnic groups, and socioeconomic backgrounds. We opted to adjust for race, ethnicity, and genetically determined PCs in our analyses given that each represents a different factor that might be independently associated with one or more phenotypes. While we acknowledge that this may result in some degree of overadjustment, overadjustment in a logistic regression would result in a loss of statistical power, making our approach conservative. We may have missed some true positive iPRS-associated phenotypes by applying this approach, and those phenotypes falling short of the Bonferroni-corrected threshold for statistical significance warrant further investigation in future studies. The rich EHR data are another strength of this study, and longitudinal EHRs with a median follow-up of 8 years allow inclusion of long-term, clinically meaningful conditions rather than spuriously recorded conditions measured cross-sectionally. The AoU dataset captures real-world, clinically relevant outcomes that may develop over time. In addition, the PheWAS design allowed an unbiased exploration of associations between genetic inflammation risk iPRS and a wide range of clinical phenotypes (2,821 phecodes) using an approach that takes advantage of the fact that the iPRS (1) is a stable metric that captures life-long risk for systemic inflammation rather than transient levels of inflammation measured during a medical visit, (2) is independent of confounders between systemic inflammation and disease states, and (3) was collected on a much larger group of individuals in AoU and BioVU compared to the phenotypes that comprise the INFLA-score, providing our analysis with considerable power gains compared to an analysis of the INFLA-score. Additional strengths of the study design include the use of INFLA-score as a marker of chronic inflammation, as this score is derived from a validated, biologically grounded composite measure which has been shown to have clinical utility.^34,35

Although this study has numerous strengths, some limitations should be acknowledged. For example, there is limited global generalizability, as the findings may be harder to replicate across populations outside the USA due to differences in phenotype prevalences. Such differences may exist even within the USA population itself. For example, variation in medical practice and healthcare access across different states in the USA and different countries globally may affect application of diagnostic codes (e.g., ICD-9 or ICD-10). Although AoU made substantial efforts to include participants from diverse racial, ethnic, and sociodemographic backgrounds, biobanks with differing compositions may yield different results. Moreover, if the SNPs constituting the PRS for INFLA-score are not common in all populations, the applicability of the iPRS may be compromised.

Socioeconomic determinants are well-established modulators of inflammation, with lower socioeconomic status consistently associated with heightened inflammatory activity.³⁶ For example, findings have shown that individuals living in areas of higher deprivation (as measured by the index of multiple deprivation) had elevated levels of CRP and WBC counts. There is also evidence that certain racial and ethnic groups exhibit higher levels of inflammation; for example, racially minoritized populations in the USA have been found to have greater levels of inflammation than White populations.³⁷ However, the specific socioeconomic factors that might influence the relationship between inflammation and each given phenotype will likely differ by phenotype. It will be important to carefully select such factors for evaluation in post-PheWAS analyses focused on individual phenotypes identified here. Such analyses will be important in future genetic studies of inflammation, and we hope that this discovery-based PheWAS will serve as a basis for further, focused investigation of the relationship between inflammation, social determinants of health, and other exposures with respect to the risk of each phenotype.

The results from this study contribute significant information to our understanding of the contribution of low-grade systemic inflammation to a broad range of diseases and phenotypes, which may inform future mechanistic studies and have potential actionable implications for biomarker development. This iPRS could be integrated into risk prediction models to prioritize patients for preventative strategies such as anti-inflammatory lifestyle changes or as a complementary tool to screen for serious and potentially life-threatening conditions (e.g., myeloid malignancies) in an early risk screening or monitoring strategy. Overall, these results indicate that the iPRS has substantial potential clinical utility to identify individuals at increased genetic risk of developing a range of chronic inflammation-related disorders and could help stratify individuals earlier for preventive interventions.

Data and code availability

The scoring file for the iPRS is available through the PGS Catalog (PGS Catalog: PGS012559), and code is available via GitHub (https://github.com/slkerns/iPRS-PheWAS). Access to individual-level data is available through the UKB and AoU resources, respectively, and GWAS summary statistics will be made available upon request.

Acknowledgments

Research reported in this publication was supported by the National Institutes of Health for the project “Polygenic Risk Methods Development (PRIMED) Consortium,” with grant funding for the EndoPhenotype InCorporated PRS (EPIC-PRS) Study Site (National Human Genome Research Institute grant U01HG011720), as well as through the National Cancer Institute under contract no. HHSN261201500003I, task order no. HHSN261000039. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The VUMC BioVU dataset used for the replication analyses is supported by institutional funding, the 1S10RR025141-01 instrumentation award, CTSA grant UL1TR000445 from the National Center for Advancing Translational Sciences/National Institutes of Health, and other project-specific funds. A complete list of funders is available at https://victr.vumc.org/biovu-funding/. Additionally, M.S. was supported by K12AR084232, and H.K.J. was supported by the Wellcome Trust Genetic Epidemiology and Public Health Genomics Doctoral Training Partnership (grant no. 218505/Z/19/Z). The project was also supported in part by the Medical College of Wisconsin Cancer Center Biostatistics Shared Resource.

Declaration of interests

The authors declare no competing interests.