Circulating imidazole propionate and coronary heart disease risk: interplay between histidine intake, fiber, and gut microbiome

Abstract

Background

Imidazole propionate (ImP), a microbial metabolite of histidine, may impair glucose metabolism, but its relevance to coronary heart disease (CHD) risk and potential diet–microbiota regulations remain unclear. We aimed to examine prospective associations of plasma ImP levels and histidine intake with CHD risk, to identify ImP-predicting gut microbes, and to investigate diet–microbiome interactions influencing ImP levels.

Methods

Associations of ImP and histidine with CHD risk were evaluated using Cox models in 7,432 participants from Nurses’ Health Study (NHS), NHSII, and Health Professionals Follow-up Study. Microbiome–diet interactions influencing ImP levels were assessed using fecal metagenome and 7-day diet record data in 296 men from the Men’s Lifestyle Validation Study, with replication in the Mind-Body Study.

Results

Higher plasma ImP was associated with increased CHD risk (HR comparing extreme quintiles = 1.82; 95%CI = 1.17–2.81; p-trend = 0.002), while histidine intake showed a non-significant inverse association. Although histidine intake was not associated with ImP levels, the intake of fiber, especially pectin, emerged as a key negative predictor. We identified 17 distinct ImP-predicting species, including Clostridium and Blautia species. A parametric ImP-microbial score was constructed based on these species to represent the microbial capacity of producing ImP. Further functional characterization uncovered that the microbial urocanate reductase gene urdA was also associated with cardiovascular risk markers. No significant interaction was observed between histidine intake and the microbial score on ImP levels, but ImP levels increased with higher histidine intake and higher microbial score only under low pectin intake (p for 3-way interaction = 0.01). Similar interactions were seen for total fiber (p = 0.09), soluble fiber (p = 0.09), and insoluble fiber (p = 0.11), without statistical significance.

Conclusions

ImP, but not its dietary precursor histidine, was associated with a higher CHD risk. The gut microbial metabolism of ImP appeared context-dependent, with ImP production from histidine associated with a higher ImP-producing microbial capacity and lower fiber intake. These findings highlight the potential role of dietary fiber and gut microbiome in modulating diet-health associations related to ImP metabolism.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12916-026-05012-6.

Background

Coronary heart disease (CHD) has a complex etiology that involves interplay among lifestyle, diet, genetic predisposition, the human gut microbiome, and other factors [1–4]. Accumulating evidence reveals that diet may strongly modulate microbial composition and provide substrates for the microbial production of metabolites, which can subsequently modulate host physiology relevant to CHD etiology [5–7]. Imidazole propionate (ImP), a microbial metabolite derived from histidine, has been linked to insulin resistance and type 2 diabetes (T2D) in human and animal studies [8–10]. Individuals with cardiovascular disease (CVD) [10, 11], heart failure (HF) [12], and elevated diastolic blood pressure [13] were found to have significantly higher plasma ImP concentrations, potentially through pathways involving systemic inflammation [13]. However, to date, there is no prospective study examining the association between plasma ImP concentrations and the risk of developing CHD.

Interestingly, while histidine is the primary precursor of ImP produced by the gut microbiome, the beneficial effects of histidine intake per se on human health were recognized [14, 15], suggesting that probably only a small fraction of histidine intake is shunted to the ImP production pathway in the gut. Yet, our understanding of the microbial compositional features underlying ImP production is still limited. An animal study by Koh et al. identified specific ImP-producing microbial species [8], although the role of the human gut microbiome in the relationship between histidine intake and plasma ImP concentrations remains unclear in free-living individuals. Furthermore, no population-based study has examined the interactions between histidine intake and the gut microbiome on circulating ImP concentrations. Lastly, a recent study found that ImP was not associated with histidine intake, but an unhealthy diet poor in fiber was associated with higher ImP levels [10]. It is intriguing that prior studies consistently demonstrated a positive association between fiber intake and another microbiota-derived metabolite, indole-3-propionate (IPA) [16, 17], as well as a necessary role of dietary fiber in the microbial production of IPA from tryptophan intake [16–18]. Whether fiber plays the same role in the production of ImP remains to be explored.

Therefore, this study aims to address these critical knowledge gaps by investigating: the prospective association between dietary histidine intake, plasma ImP concentrations, and CHD risk; gut microbial features associated with ImP; and the interplay between diet and the gut microbiome in relation to plasma ImP concentrations. These analyses are conducted using data from the Nurses’ Health Study (NHS), NHSII, Health Professionals Follow-up Study (HPFS), and Men’s Lifestyle Validation Study (MLVS). A replication of the microbial findings was conducted in the Mind-Body Study (MBS).

Methods

Study population

The Nurses’ Health Study (NHS) started in 1976 with 121,700 female registered nurses aged 30–55 years enrolled [19]. The NHSII was established in 1989 and consisted of 116,671 younger female registered nurses, aged 25–42 years [19]. The Health Professionals Follow-up Study (HPFS) was initiated in 1986, recruiting 51,529 male professionals aged 40–75 years at baseline [20]. Participants in each cohort were followed every 2–4 years through mailed questionnaires, which gathered data on dietary habits, lifestyle factors, and medical history. Blood samples were collected in the NHS (n = 32,826) between 1989 and 1990 [21], NHSII (n = 29,611) between 1996 and 1999 [22], and HPFS (n = 18,225) between 1993 and 1995 [23]. For the present study, among 13,747 participants from the NHS, NHSII, and HPFS with existing plasma metabolomics data [21–23], we further excluded those without ImP data, with pre-existing CHD, T2D, stroke, or cancer, or who were lost to follow-up after blood collection. This resulted in a final sample of 7,432 participants, including 3,456 women from NHS, 2,889 women from NHSII, and 1,087 men from HPFS (Fig. 1).

Fig. 1.

Overview of study design. The flowchart of the overall study design, sample collections, laboratory assays, and data analysis in the MLVS, NHS, NHSII, and HPFS. Abbreviations: HPFS, Health Professionals Follow-up Study; NHS, Nurses’ Health Study; MLVS, Men’s Lifestyle Validation Study; MBS, Mind-Body Study; ImP, imidazole propionate; CHD, coronary heart disease; T2D, type 2 diabetes; hs-CRP, high-sensitivity C-reactive protein; TC, total cholesterol; HDL-C, high-density lipoprotein-cholesterol; TG, triglycerides; HbA1c, haemoglobin A1c

The Men’s Lifestyle Validation Study (MLVS) is a sub-study within HPFS carried out during 2011–2012, aiming to validate self-reported lifestyle and diet, the participants of which were largely healthy and free of major chronic diseases at enrollment as per eligibility criteria [24]. Among 647 MLVS participants, 307 provided up to two pairs (Set 1 and Set 2) of stool samples, collected 6 months apart. Each pair of stool samples was collected within 24 to 72 h. Two fasting blood samples were obtained via venipuncture, 6 months apart, approximately matching the timing of stool sample collection. Detailed information on the three cohorts and MLVS has been provided in previous publications [24–29]. We excluded individuals who provided no more than one stool sample or had no plasma ImP data (measured in the 2nd blood sample), resulting in 296 MLVS participants and 896 metagenomes included in the analysis (Fig. 1 and Additional file 1: Figure S1).

The Mind-Body Study (MBS) is a sub-study nested within the NHSII (https://nurseshealthstudy.org). Two pairs of fecal samples were collected in this study, following the same collection protocol as the MLVS [30]. Fasting blood samples were collected 3 months after the collection of the second pair of stool samples. After excluding individuals with < 2 stool samples or no existing plasma ImP and dietary data, a total of 205 MBS participants and 788 metagenomes were included in the present study (Fig. 1).

The study protocol was approved by the Institutional Review Boards of Brigham and Women’s Hospital and Harvard T.H. Chan School of Public Health, with informed consent obtained from all participants.

Biomarker measurement in blood samples

In the NHS, NHSII, and HPFS, plasma ImP concentrations and urocanic acid (also known as urocanate) were previously measured across multiple sub-studies at the Broad Institute of Harvard University and M.I.T. (Cambridge, MA), using high-throughput liquid chromatography-tandem mass spectrometry techniques [2]. In the histidine metabolism pathways, ImP can be derived from urocanate by bacterial metabolism [8]. Data from these studies were compiled, and batch effects were corrected [31]. Blinded quality control (QC) samples, comprising 10% of the total, were randomly inserted, and the CVs for ImP were 20.2% in NHS, 18.6% in NHSII, and 21.7% in HPFS. The same metabolomics platform was used to measure plasma ImP concentrations in the second blood samples collected from the MLVS and MBS. Based on 10% QC samples, the average CV of ImP was 19.7% in the MLVS and 15.0% in the MBS, respectively. The average CV of urocanic acid was 81.6% in the MLVS.

In the MLVS, total cholesterol (TC), high-density lipoprotein cholesterol (HDL-C), and triglycerides (TG) were measured in all blood samples using a Hitachi 911 analyzer with reagents from Roche Diagnostics and Genzyme. Hemoglobin A1c concentrations (HbA1c) were measured by turbidimetric immunoinhibition using packed erythrocytes (Roche Diagnostics). The concentrations of high-sensitivity C-reactive protein (hs-CRP) were quantified on a Hitachi 911 system using an immunoturbidimetric high-sensitivity assay with reagents and calibrators from Denka Seiken.

Microbiome sequencing and profiling

We performed shotgun metagenomic sequencing on 896 stool samples from 296 MLVS participants and 788 stool samples from 205 MBS participants. The sample profiling process has been described in detail previously [24, 29]. Briefly, fecal DNA was prepared for sequencing using the Nextera XT DNA Library Preparation Kit, and RNA was prepared by RNAtag-Seq. Metagenomic sequences were generated using Illumina HiSeq paired-end shotgun sequencing. QC was conducted to filter out low-quality sequence reads through KneadData 0.3 (http://huttenhower.sph.harvard.edu/kneaddata). We used MetaPhlAn 4 [32] to perform taxonomic profiling, and HUMAnN 3.6 [33] to perform pathway, enzyme, and gene family profiling of fecal metagenomes. Microbial taxa with a relative abundance below 0.01% or a prevalence under 10% were excluded. After exclusions, 362 species from 10 bacterial phyla were retained for analysis out of 2,201 species in the MLVS. Additionally, 387 pathways and 1,826 enzymes with a prevalence > 10% were included in further analysis. To reduce the impact of random errors, the averages of the two fecal microbiome samples collected in each set were calculated to represent the microbiome profile for that set.

Dietary assessment

A validated semi-quantitative food frequency questionnaire (FFQ) was administered every 4 years, starting in 1984 and 1986 in the NHS, 1991 in NHSII, and 1986 in HPFS, to collect dietary data [34, 35]. Nutrient intake, including histidine, total protein, animal protein, and vegetable protein, and total energy intake, were calculated from FFQ assessments based on the Harvard University Food Composition Database, which is primarily derived from US Department of Agriculture data, supplemented with manufacturers’ data and updated regularly. Dietary pattern scores, including the Alternative Healthy Eating Index-2010 (AHEI-2010) [36], Alternate Mediterranean Diet (AMED) [37], the healthful Plant-based Diet Index (hPDI) [38], Dietary Approaches to Stop Hypertension (DASH) score [39], Diabetes Risk Reduction Diet (DRRD) score [40], World Cancer Research Fund/American Institute for Cancer Research (WCRF/AICR) dietary score [41], Empirical Dietary Index for Hyperinsulinemia (EDIH) score [42], and Empirical Dietary Inflammatory Pattern (EDIP) score [43], were constructed from FFQ data following established methods in previous studies [44]. The average of the two most recent FFQ data prior to blood sample collection was used to represent dietary intake.

In the MLVS, detailed dietary consumptions were assessed using two sets of 7-day diet records (7DDRs) administered 6 months apart during stool sample collections (Fig. 1). With the aid of a food scale, ruler, and instructional support through DVD and telephone, participants measured and recorded the weight of foods in grams both before and after consuming a meal to accurately determine their actual intake. Additionally, they provided recipes for all home-prepared foods, specifying the number of servings per recipe and the portions they consumed. Over 150 nutrients and dietary constituents, including histidine, protein (total, animal, and vegetable protein), and fiber (total fiber, soluble fiber, insoluble fiber, and pectin), were derived from 7DDRs based on the Nutrition Data System for Research at the University of Minnesota Nutrition Coordinating Center. The averages of Set 1 (week 1) and Set 2 (week 2) 7DDRs were calculated to represent dietary intake for each respective set. Log-term diet quality during 1986–2010 was assessed based on the cumulative average of AHEI-2010 derived from FFQ assessments. In the MBS, the same FFQ was administered during each stool sample collection, and the averages of the two FFQ assessments were calculated to represent overall dietary intake.

Covariates assessment

Biennial follow-up questionnaires were distributed to participants across the three cohorts to gather and update information on disease occurrence, demographics, and lifestyle factors such as smoking status, alcohol consumption, body weight, and other relevant variables. Physical activity levels were consistently assessed across the cohorts, and body mass index (BMI) was calculated by dividing weight in kilograms by the square of height in meters as a measure of overall obesity. The AHEI-2010 was used to account for overall diet quality. Covariates collected around the time of blood collection were included in the analysis. In the MLVS and MBS, covariate data were collected similarly around the time of blood collection, with additional assessments on the use of probiotics and antibiotics, surgical procedures, and stool consistency using the Bristol Stool Scale through fecal sample collection questionnaires.

Ascertainment of CHD

In the current study, CHD was defined as incident nonfatal myocardial infarction (MI) or fatal CHD. Incident self-reported nonfatal MI was confirmed through medical record review by study physicians blinded to participants’ exposure status according to the World Health Organization criteria [45]. Deaths were identified from state vital records, the National Death Index, participants’ next of kin, or postal authorities [46]. Fatal CHD was confirmed by hospital or autopsy records, or death certificate with evidence of previous CHD as the most plausible cause. In the current analysis, we included CHD incidences occurring after blood sample collection in the NHS (1989–2020), NHSII (1996–2021), and HPFS (1993–2018) (Fig. 1). Other cardiovascular events, such as coronary revascularization procedures, stroke, etc., were not considered as outcomes in this study.

Statistical analysis

Our analysis consisted of two primary components: analyses of associations of ImP levels, urocanic acid levels, and dietary histidine intake with CHD risk in the three cohorts, and analyses the interplay between diet and microbiome in relation to ImP levels in the MLVS and MBS.

For the first component, plasma ImP and urocanic acid concentrations were batch-corrected and transformed into natural-log z-scores. We imputed missing values for continuous covariates (including BMI and physical activity) using median values and for categorical covariates (including race) using mode, given the low missing rate (< 1%) of these covariates. Cox proportional hazards models, both crude and multivariable-adjusted, were used to assess the association of histidine intake and plasma ImP and urocanic acid concentrations with CHD risk in the NHS, NHSII, and HPFS. Covariates in the adjusted model included age (years), ancestry (White/others), BMI (kg/m²), physical activity (MET-hrs/wk), alcohol intake (categorized as never, < 5 g/d, 5–9.9 g/d, 10–14.9 g/d, 15–29.9 g/d, and ≥ 30 g/d), total energy intake (kcal/d), AHEI-2010, smoking status (never, current, past), family history of myocardial infarction (yes/no), and cohort (NHS, NHSII, HPFS). Renal function was assessed using estimated glomerular filtration rate (eGFR), calculated with the 2021 Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) creatinine equation [47]. Plasma creatinine was obtained from metabolomics profiling. In sensitivity analyses, eGFR was additionally included as a covariate in the multivariable Cox proportional hazards models. A log-rank test was conducted to formally assess differences in survival curves across ImP concentration groups. Proportional hazards assumption was evaluated using Schoenfeld residuals, with no significant violations detected (p > 0.05). Nonparametric Kaplan-Meier curves were used to visually illustrate the association between ImP concentrations and CHD risk. Additionally, multivariable linear regression was applied to assess the associations of dietary factors derived from the FFQ, including nutrients, food groups, and dietary patterns, with plasma ImP concentrations. For analyses of nutrients and food groups, models were adjusted for age, BMI, physical activity, total energy intake, alcohol intake, AHEI-2010, smoking status, and cohorts, while for dietary patterns, models were adjusted for the same covariates except AHEI-2010.

For the second component, in the MLVS, we used multivariable linear regression to estimate the associations of dietary factors derived from the 7DDRs with plasma ImP levels, and the associations of histidine intake and plasma ImP concentrations with cardiometabolic traits (including TC, TG, HDL-C, TC/HDL-C, hs-CRP, and HbA1c) measured at the collection of the second set of fecal samples, adjusting for age, BMI, physical activity, alcohol intake, total energy intake, and AHEI-2010. Principal coordinates analysis (PCoA) based on Bray–Curtis dissimilarity and permutational multivariate analysis of variance (PERMANOVA; marginal = TRUE) were performed to assess species level microbial community structure, adjusting for age, BMI, physical activity, alcohol intake, total energy intake, AHEI-2010 score, probiotic use in the past 2 months, antibiotic use in the past year, and Bristol Stool Scale score. To estimate the contribution of different factors to plasma ImP levels, three Least Absolute Shrinkage and Selection Operator (LASSO) models were constructed to predict ImP concentrations using gut microbiome, dietary intake, or basic characteristics, respectively. The gut microbiome model included all microbial species identified in the MLVS. The dietary intake model incorporated total energy intake and ImP dietary precursors, specifically histidine, total protein, animal protein, vegetable protein, red meat, poultry, fish, dairy, eggs, and whole grains. The basic characteristics model included age, BMI, physical activity, smoking status, alcohol intake, and the cardiometabolic traits, i.e., TC, TG, HDL-C, TC/HDL-C, hs-CRP, and HbA1c. Pearson’s correlation was used to assess the relationship between predicted and actual plasma ImP concentrations across these three LASSO models.

To construct a microbial score to reflect the ImP-producing capacity of the gut microbiome, the relative abundances of microbial species were first normalized using centered log-ratio (CLR) transformation, followed by z-score standardization for further analysis. Multivariable linear regression was then used to estimate the associations between microbial species and ImP concentrations, adjusting for age, BMI, physical activity, alcohol intake, total energy intake, AHEI-2010, probiotic use in the past 2 months, antibiotic use in the past year, and the Bristol Stool Scale score. A linear mixed-effects model was used to evaluate the associations between ImP dietary precursors and ImP-related species, incorporating measurements from the 7DDR and gut microbiome in both sets. The model included the same covariates as the multivariable linear regression. To account for multiple comparisons, p-values were adjusted using the Benjamini-Hochberg method to control the false discovery rate (FDR), resulting in q-values. We then derived an ImP Microbial Score (IMS) to summarize significant ImP-predicting species (FDR q < 0.1). The IMS was calculated by multiplying the abundances of ImP-related species by their respective coefficients and then summing the resulting products. A higher IMS indicates a greater ImP-producing capacity. We further examined the associations of the IMS and ImP-predicting species with CVD biomarkers using multivariable linear regression.

For metagenomic functional profiling in the MLVS, we selected metagenomic pathways and enzymes involved in ImP synthesis (Fig. 4A) from the histidine metabolism pathway (PATHWAY: map00340) based on the Kyoto Encyclopedia of Genes and Genomes (KEGG, https://www.kegg.jp/) dataset. These pathways included HISDEG-PWY: L-histidine degradation Ⅰ, PWY-5028: L-histidine degradation ⅠⅠ, and PWY-5028: L-histidine degradation ⅠⅠⅠ. However, PWY-5028: L-histidine degradation II was present in only two participants, so we excluded it from the analysis. The enzymes included histidine ammonia-lyase (EC 4.3.1.3, hutH) and urocanate reductase (EC 1.3.99.33, urdA). Linear regression models were employed to estimate the associations of these pathways and enzymes abundances with plasma ImP levels, adjusting for the same covariates. Since the prevalence of urocanate reductase (EC 1.3.99.33, urdA) was only 11.4% among participants, we further categorized it as a binary variable (present or not present) for the association analysis. Potential associations of universal metagenomic pathways and enzymes were also analyzed using the same multivariable linear regression models. To identify additional urdA gene homologs, we searched for UniProt IDs with at least 90% sequence similarity to known urdA genes. A total of five urdA gene homologs (UniRef90_A0A143XCN0, UniRef90_A0A2Z6I9Y0, UniRef90_B0NCC6, UniRef90_C8WLE3, and UniRef90_F8DIF2) were identified in the metagenomic gene family profiling. We also calculated the total abundance of these five gene homologs as the sum of urdA gene homologs. Due to the low prevalence of these gene homologs, the same multivariate linear regression was used to estimate the associations between the presence of urdA gene homologs and plasma ImP concentrations.

Fig. 4.

In relation to ImP levels and their associations with CHD biomarkers. (A) Histidine utilization pathway involving ImP synthesis according to KEGG (https://www.genome.jp/kegg/). (B-E) The scatter plots and fitted regression lines with confidence bands illustrate the relationships of microbial pathways and enzymes involving in ImP synthesis with plasma ImP levels. (F) The box plot shows the distributions of plasma ImP in groups with/without urocanate reductases (EC 1.3.99.33). (E) The heatmap shows the association between ImP-related microbial features and CHD biomarkers. Each species or enzyme name in red represents a positive association with ImP, while those in green represent a negative association with ImP. (H) The heatmap shows the association of unstratified and stratified urdA gene homologs with circulating ImP and CHD biomarker concentrations. The numbers on the heatmap represent β coefficients. The bar plot illustrates the prevalence of these gene families. Multivariable linear regression was adjusted for age (years), BMI (kg/m²), physical activity (MET-hrs/wk), alcohol consumption (never, < 5 g/d, 5–9.9 g/d, 10–14.9 g/d, 15–29.9 g/d, ≥ 30 g/d), total energy intake (kcal/d), AHEI, antibiotics use (yes/no), probiotics use (yes/no), colonoscopy (yes/no), and the Bristol Stool Scale score. Abbreviations: ImP, imidazole propionate; EC number, Enzyme Commission number; CI, confidence interval; BMI, body mass index; MET, metabolic equivalent; AHEI, Alternative Healthy Eating Index; MLVS, Man’s Lifestyle Validation Study; CHD, coronary heart disease

To evaluate potential modulation by the gut microbiome on associations between ImP dietary precursors and ImP concentrations, the interactions between dietary intake and IMS on plasma ImP concentrations were examined by including a product term between dietary histidine and IMS in a linear regression model. To further evaluate the role of dietary fiber in modulating the interaction between diet and the gut microbiome, we incorporated a 3-way interaction term among dietary histidine, IMS, and dietary fiber (including total fiber, soluble fiber, insoluble fiber, and pectin) into the multivariable-adjusted model. Additionally, pairwise interaction terms between these three variables were also included in the analysis. In a secondary analysis, we also examined the three-way interactions with other ImP dietary precursors. Furthermore, the participants were stratified by median dietary fiber, and the interaction between histidine intake and IMS on ImP concentrations was analyzed separately within the low and high fiber intake groups.

We validated the IMS, key ImP-related species, and the three-way interactions between diet, IMS, and plasma ImP concentrations in the MBS participants. The IMS was calculated by multiplying the abundances of ImP-related species identified in the MLVS by their respective coefficient directions (1 or -1; and thus semi-parametric) and summing the resulting products. To minimize random errors, microbial features were averaged across all fecal samples. A similar approach was used to calculate the average histidine intake as well as other dietary and lifestyle variables.

All analyses were performed using R v.4.2.0. Statistical significance was defined as a two-sided p-value < 0.05 or q < 0.1. Results with p-values between 0.05 and 0.10 were considered marginally significant associations, which we reported to provide a complete picture of the data patterns while acknowledging the lack of formal statistical significance.

Results

Participant characteristics and correlations between FFQ-assessed diet and plasma ImP concentrations

The study included participants from the NHS, NHSII, and HPFS with valid plasma metabolomics data (Table 1). On average, participants were 51.3 (± 10.2) years old, with those having higher plasma ImP concentrations tending to exhibit a higher BMI, lower total energy intake, and a greater likelihood of smoking. Baseline dietary characteristics according to quintiles of plasma ImP are presented in Additional file 1: Table S12. In the three cohorts, nutrient intakes and dietary pattern scores were generally similar across ImP quintiles, without evidence of systematic gradients. After adjustment, histidine intake and its key dietary sources, including poultry, fish, dairy, and eggs, showed no significant associations with plasma ImP concentrations (Additional file 1: Table S1). Weak negative associations were observed between plasma ImP and red meat intake, whole grain intake, and DASH score (β = −0.036 to −0.026, p < 0.05), while weak positive associations were found with refined grain intake, sugar-sweetened carbonated beverage consumption, and EDIP score (β = 0.026 to 0.036, p < 0.05; Additional file 1: Table S1).

Table 1.

Baseline characteristics of the NHS, NHSII, and HPFS study participants, by quintiles of plasma ImP concentrations assessed in blood sample collection^a

	Total	Plasma ImP quintiles
		Q1	Q2	Q3	Q4	Q5
N. of participants	7432	1487	1486	1486	1487	1486
Age, years	51.3 (10.2)	51.0 (9.9)	51.6 (9.8)	51.6 (10.3)	50.7 (10.2)	51.8 (10.4)
BMI, kg/m2	25.3 (4.7)	24.8 (4.0)	25.0 (4.5)	25.1 (4.6)	25.5 (4.9)	26.1 (5.3)
Cohort
HPFS	1087 (14.6)	235 (15.8)	269 (18.1)	203 (13.7)	168 (11.3)	212 (14.3)
NHS	3456 (46.5)	675 (45.4)	706 (47.5)	721 (48.5)	681 (45.8)	673 (45.3)
NHSⅡ	2889 (38.9)	577 (38.8)	511 (34.4)	562 (37.8)	638 (42.9)	601 (40.4)
Gender
Woman	6345 (85.4)	1252 (84.2)	1217 (81.9)	1283 (86.3)	1319 (88.7)	1274 (85.7)
Man	1087 (14.6)	235 (15.8)	269 (18.1)	203 (13.7)	168 (11.3)	212 (14.3)
Physical activity, MET-hrs/wk	20.2 (26.5)	20.4 (27.5)	19.8 (23.8)	20.0 (23.2)	20.4 (29.8)	20.2 (27.5)
Total energy intake, kcal/d	1832 (496)	1842 (510)	1849 (492)	1836 (485)	1818 (489)	1816 (504)
AHEI	50.5 (10.3)	50.8 (10.2)	50.6 (10.5)	50.8 (10.4)	49.9 (10.1)	50.2 (10.4)
Alcohol intake
Never	2111 (28.4)	444 (29.9)	417 (28.1)	398 (26.8)	408 (27.4)	444 (29.9)
< 5 g/d	2871 (38.6)	566 (38.1)	544 (36.6)	580 (39.0)	577 (38.8)	604 (40.6)
5–9.9 g/d	974 (13.1)	191 (12.8)	204 (13.7)	209 (14.1)	196 (13.2)	174 (11.7)
10–14.9 g/d	606 (8.2)	108 (7.3)	128 (8.6)	131 (8.8)	128 (8.6)	111 (7.5)
15–29.9 g/d	599 (8.1)	129 (8.7)	134 (9.0)	127 (8.5)	113 (7.6)	96 (6.5)
≥ 30 g/d	271 (3.6)	49 (3.3)	59 (4.0)	41 (2.8)	65 (4.4)	57 (3.8)
Smoking
Never	3992 (53.7)	878 (59.0)	805 (54.2)	799 (53.8)	733 (49.3)	777 (52.3)
Past smoker	2689 (36.2)	536 (36.0)	553 (37.2)	528 (35.5)	552 (37.1)	520 (35.0)
Current smoker	751 (10.1)	73 (4.9)	128 (8.6)	159 (10.7)	202 (13.6)	189 (12.7)
Race
White	7288 (98.1)	1459 (98.1)	1452 (97.7)	1466 (98.7)	1463 (98.4)	1448 (97.4)
Others	144 (1.9)	28 (1.9)	34 (2.3)	20 (1.3)	24 (1.6)	38 (2.6)
Family history of MI, %	2231 (30.0)	430 (28.9)	434 (29.2)	447 (30.1)	422 (28.4)	498 (33.5)

^a Data are mean (SD) for continuous variables and n (%) for categorical variables. The data are pooled for NHS, NHSII, and HPFS; the blood collection date: 1993–1995 for the HPFS, 1989–1990 for the NHS, and 1996–1999 for the NHSII

T1–T3, 1st–3rd tertiles; HPFS, Health Professionals Follow-up Study; NHS, Nurses’ Health Study; ImP, imidazole propionate; BMI, body mass index; AHEI, alternative healthy eating index; MET, metabolic equivalent task; MI, myocardial infarction

Association of plasma ImP and histidine intake with CHD risk

In the NHS, NHSII, and HPFS, during 31 years of follow-up in the three cohorts, we documented 225 incident CHD cases among 7,432 participants contributing 175,448 person-years (Fig. 1). In the multivariable-adjusted model, higher plasma ImP concentrations were significantly associated with an increased risk of CHD. The fully adjusted HRs (95% CIs) of CHD comparing the second, third, fourth, and fifth quintiles (Q) to the lowest one of ImP were 1.16 (0.73, 1.86), 1.17 (0.73, 1.87), 1.57 (0.99, 2.47), and 1.82 (1.17, 2.81) (p-trend = 0.002), whereas histidine intake showed a non-significant inverse association with CHD risk (Fig. 2A, C and Additional file 1: Table S2). In the histidine utilization pathways, ImP can be derived from urocanate by bacterial metabolism [8]. Plasma urocanic acid tended to be associated with a high risk of developing CHD (p-trend = 0.02), though the HRs for Q2-Q5 did not reach statistical significance (Additional file 1: Table S2). For example, the HR for Q5 compared to Q1 was 1.22 (95% CI: 0.82, 1.82). The associations remained materially unchanged after additional adjustment for eGFR. In the joint associations between histidine intake and plasma ImP concentrations on CHD risk, participants in the lowest histidine intake tertile and the highest plasma ImP tertile had the highest risk of developing CHD, with an HR (95% CI) of 3.40 (1.71, 6.73), compared with those in the highest histidine tertile and the lowest ImP tertile (Fig. 2B, Additional file 1: Table S3). Risk increased progressively across joint categories (p-trend = 4.9E-04). However, formal testing of multiplicative interaction between histidine and ImP tertiles did not indicate statistically significant interaction (p for interaction = 0.72; Additional file 1: Table S3).

Fig. 2.

Hazard ratio for CHD according to different levels of histidine intake and plasma ImP. The data is pooled for the NHS, NHSII, and HPFS. (A) Hazard ratios were generated by Cox proportional hazards regression model. Q1–Q5,1st–5th quintiles. (B) Joint associations of histidine intake and plasma ImP with the risk of CHD. Hazard ratios were estimated by Cox proportional hazards regression model. T1–T3, 1st–3rd tertiles. The Cox proportional hazards regression model was adjusted for age (years), race (White/others), BMI (kg/m²), physical activity (MET-hrs/wk), alcohol (6 categories: never, < 5 g/d, 5–9.9 g/d, 10–14.9 g/d, 15–29.9 g/d, ≥ 30 g/d), total energy intake (kcal/d), AHEI (quintile), smoking (never, current, past), family history of myocardial infarction (yes/no), and cohort (NHS, NHSII, HPFS). (C) Kaplan-Meier survival curves for plasma ImP levels in relation to CHD. A log-rank test was conducted to formally assess differences in survival curves across ImP concentration groups. (D) Association of plasma ImP and histidine intake with plasma CHD biomarkers were estimated using multivariate linear regression in the MLVS. Plasma ImP and CHD biomarkers were log-transformed and standardized prior to analysis. Effect estimates represent the change in standardized biomarker levels per 1 SD increase in log-transformed plasma ImP or histidine intake. Biomarker units are as follows: TC (mg/dL), TG (mg/dL), and HDL (mg/dL); hs-CRP (mg/L); HbA1c (%). The model was adjusted for age (years), BMI (kg/m²), physical activity (MET-hrs/wk), alcohol consumption (never, < 5 g/d, 5–9.9 g/d, 10–14.9 g/d, 15–29.9 g/d, ≥ 30 g/d), total energy intake (kcal/d), and AHEI. Abbreviations: ImP, imidazole propionate; CHD, coronary heart disease; HPFS, Health Professionals Follow-up Study; NHS, Nurses’ Health Study; MLVS, Men’s Lifestyle Validation Study

7DDR-assessed diet, plasma ImP concentrations, and CVD risk markers in the MLVS

In the MLVS, a total of 296 men were included in the microbiome analysis (Fig. 1), with an average age of 71.6 (± 4.3) years and a BMI of 26.0 (± 3.8) kg/m². Men with higher plasma ImP concentrations tended to be older and have a higher BMI (Table 2). Dietary characteristics in the MLVS are shown in Additional file 1: Table S13. Although minor variations were observed across ImP quintiles, there was no consistent pattern observed for nutrient or food group intake. Histidine intake was not significantly associated with ImP levels. In contrast, significant negative associations (FDR q-value < 0.1) were observed for vegetable protein and pectin, with pectin showing the strongest, inverse association (Set2, β: −0.21; 95%CI: −0.35, −0.07; Additional file 1: Table S4). In addition, after adjusting for potential confounders, plasma ImP was negatively associated with plasma HDL-C and TC, but positively associated with plasma hs-CRP (Fig. 2D, Additional file 1: Table S5).

Table 2.

Baseline characteristics of the MLVS study participants, by quintiles of plasma ImP concentrations assessed in the 2nd blood sample collection^a

	Total	ImP quintiles
		Q1	Q2	Q3	Q4	Q5
N. of participants	296	60	59	59	59	59
Age, years	71.6 (4.3)	70.4 (3.9)	71.7 (4.0)	71.7 (4.1)	72.1 (4.4)	72.4 (4.8)
BMI, kg/m2	26.0 (3.8)	24.9 (3.1)	25.6 (3.5)	26.2 (3.8)	27.1 (4.4)	26.2 (4.0)
Physical activity, MET-hrs/wk	116.9 (50.9)	116.5 (42.8)	117.5 (47.9)	118.6 (55.1)	117.3 (52.9)	114.7 (56.3)
Total energy intake, kcal/d	2152 (553)	2286 (575)	2138 (525)	2100 (545)	2158 (543)	2073 (567)
AHEI	58.1 (9.1)	57.4 (9.0)	60.8 (8.6)	57.2 (9.5)	57.6 (9.4)	57.4 (8.9)
Alcohol intake
Never	47 (15.9)	12 (20.0)	7 (11.9)	12 (20.3)	11 (18.6)	5 (8.5)
< 5 g/d	31 (10.5)	5 (8.3)	2 (3.4)	8 (13.6)	5 (8.5)	11 (18.6)
5–9.9 g/d	43 (14.5)	5 (8.3)	10 (16.9)	7 (11.9)	12 (20.3)	9 (15.3)
10–14.9 g/d	35 (11.8)	15 (25.0)	7 (11.9)	2 (3.4)	4 (6.8)	7 (11.9)
15–29.9 g/d	80 (27.0)	16 (26.7)	19 (32.2)	16 (27.1)	16 (27.1)	13 (22.0)
≥ 30 g/d	60 (20.3)	7 (11.7)	14 (23.7)	14 (23.7)	11 (18.6)	14 (23.7)
Smoking
Never	292 (98.6)	59 (98.3)	59 (100.0)	58 (98.3)	59 (100.0)	57 (96.6)
Current/Past	4 (1.4)	1 (1.7)	0 (0.0)	1 (1.7)	0 (0.0)	2 (3.4)
hs-CRP, mg/dl	1.9 (3.3)	1.4 (1.8)	1.5 (2.8)	1.2 (1.2)	2.6 (4.6)	2.6 (4.3)
TC, mg/dl	181.8 (32.3)	191.4 (31.2)	180.7 (31.9)	183.0 (31.8)	179.7 (32.6)	174.1 (32.5)
HDL-C, mg/dl	55.8 (13.6)	58.6 (12.6)	58.6 (16.0)	56.8 (12.6)	50.9 (11.0)	53.9 (13.9)
TG, mg/dl	99.6 (52.9)	89.6 (36.5)	102.2 (56.7)	94.8 (55.8)	104.2 (42.6)	107.2 (67.5)
TC/HDL-C ratio	3.4 (0.8)	3.4 (0.7)	3.2 (0.8)	3.4 (0.9)	3.7 (0.9)	3.4 (0.8)
HbA1c	5.7 (0.4)	5.7 (0.3)	5.7 (0.3)	5.6 (0.4)	5.8 (0.4)	5.7 (0.3)

^a Data are mean (SD) for continuous variables and n (%) for categorical variables

^b T1–T3, 1st–3rd tertiles; MLVS, Men’s Lifestyle Validation Study; ImP, Imidazole propionate; BMI, body mass index; AHEI, alternative healthy eating index; MET, metabolic equivalent task; hs-CRP, high-sensitivity C-reactive protein; TC, total cholesterol; HDL-C, high-density lipoprotein-cholesterol; TG, triglycerides; HbA1c, haemoglobin A1c

Gut microbial overall composition varied according to ImP levels

We applied three LASSO models with 10-fold cross-validation and evaluated the Spearman correlation between predicted and observed ImP concentrations, which ranged from 0.28 to 0.71 across different factors and time points (Set1 and Set2). Among the three models, the gut microbial model of species and enzyme levels showed the highest predictive performance for plasma ImP concentrations, particularly in Set2. Specifically, the strongest predictive performance was observed with microbial features (Spearman r Set1: 0.57, Set2: 0.71), followed by demographic and clinical factors (Set1: 0.31, Set2: 0.26) and dietary intake (Set1: 0.28, Set2: 0.28) (Fig. 3A, Additional file 1: Table S6).

Fig. 3.

Association of gut microbiome with plasma ImP levels. (A) Performance of different factors in predicting plasma ImP levels using LASSO models. Pearson r is the coefficient of Pearson’s correlation between predicted and actual plasma ImP levels. The bar is colored according to time points of data collection. (B, C) Scatterplots from principal coordinates analysis (PCoA) and permutational multivariate analysis of variance (PERMANOVA), based on Bray–Curtis dissimilarity at species level in Set1 (B) and Set2 (C). The scatter points are colored according to ImP levels and shaped based on the stool sampling time points. (D) For the forest plot, βs (95%CIs) were generated from multivariable linear regression to assess associations of microbial species in Set2 with plasma ImP levels. The model was adjusted for age (years), BMI (kg/m2), physical activity (MET-hrs/wk), alcohol intake (never, < 5 g/d, 5–9.9 g/d, 10–14.9 g/d, 15–29.9 g/d, ≥ 30 g/d), total energy intake (kcal/d), AHEI, antibiotics use (yes/no), probiotics use (yes/no), colonoscopy (yes/no), and the Bristol Stool Scale score. To account for multiple comparisons, p-values were adjusted using the Benjamini-Hochberg method to control the false discovery rate (FDR), resulting in q-values. Results with q-value < 0.1 are shown in the forest plot. For the heatmap, a linear mixed-effects model was applied to evaluate the associations between ImP dietary precursors and ImP-related species in samples from both set 1 and set 2. The bar plot shows the prevalence of these ImP-related species among samples collected in Set2. (E) An approximately maximum-likelihood phylogenetic tree was constructed using FastTree based on the 120 GTDB-defined bacterial marker genes (bac120) from bacterial genomes obtained from NCBI. The orange lines represent species that were positively associated with ImP, while the blue lines represent negative associations. Due to the absence of the marker gene in UBA1394 sp900538575, only the remaining 16 ImP-predicting species were included. (F) The relationship between ImP microbial scores and plasma ImP levels in MLVS and MBS. (G) The heatmap illustrates the Spearman correlation between ImP-related species and plasma ImP levels, while the bar plot shows the prevalence of these species in MBS. Abbreviations: MLVS, Men’s Lifestyle Validation Study; MBS, Mind-Body Study; ImP, Imidazole propionate; BMI, body mass index; AHEI, alternative healthy eating index; MET, metabolic equivalent task

Overall, gut microbial composition differed significantly according to plasma ImP concentrations in both Set1 and Set2 data collections, using PERMANOVA on Bray–Curtis distance of species level (Set1: R² = 0.009, Set2: R² = 0.006, p-value < 0.05 in both Set 1 and Set2) (Fig. 3B, C).

Gut microbe signatures linked to ImP levels and their associations with CVD biomarkers

Based on the superior performance in ImP prediction, Set 2 samples were selected for subsequent analyses, as their metagenomic profiles appeared more relevant to circulating ImP concentrations. For the taxa-wide association analysis, after adjusting for potential confounders, we found 13 species [Ruminococcus gnavus, Clostridium symbiosum, Clostridium scindens, Mediterraneibacter glycyrrhizinilyticus, Enterocloster bolteae (formerly Clostridium bolteae), Clostridium butyricum, Hungatella hathewayi, Sellimonas intestinalis, Eisenbergiella tayi, Dialister invisus, Desulfovibrio fairfieldensis, Clostridium saudiense and Streptococcus thermophilus] showing a positive association with plasma ImP concentrations, whereas four species, including Blautia A sp900066355, Alistipes putredinis, Alistipes communis, and UBA1394 sp900538575, were negatively associated (FDR, q-value < 0.1). Most of them belonged to the Firmicutes phylum, with two negatively associated species from Bacteroidetes and one positively associated species from Proteobacteria. The prevalence of these 17 species ranged from 10.2% (Desulfovibrio fairfieldensis) to 79.9% (Alistipes putredinis) (Fig. 3D, Additional file 1: Table S7). Regarding dietary factors, total protein and vegetable protein intake were negatively associated with the relative abundance of Eisenbergiella tayi, while dairy intake was negatively associated with Streptococcus thermophilus (FDR q-value < 0.1) (Fig. 3D). Non-significant association was found between ImP-predicting species and plasma urocanic acid levels (Additional file 1: Figure S2).

A parametric ImP microbial score (IMS) was calculated by multiplying the abundances of these 17 ImP-predicting species by their respective coefficients and then summing the resulting products. The IMS showed a Spearman correlation of 0.31 and a Pearson correlation of 0.41 with plasma ImP in the MLVS. The IMS was also significantly associated with the ImP levels in the MBS (Spearman r = 0.15; p < 0.05) (Fig. 3F). In the replication analysis of individual key species, 6 out of the 13 positively associated species (Ruminococcus gnavus, Clostridium symbiosum, Clostridium scindens, Enterocloster bolteae, Clostridium butyricum, and Hungatella hathewayi) also showed a positive correlation with plasma ImP concentrations in the MBS, with Ruminococcus gnavus exhibiting the strongest correlation, consistent with the findings in the MLVS. Additionally, the negative association between Alistipes communis and plasma ImP was replicated in the MBS. Meanwhile, the other three negatively associated species, Blautia A sp900066355, Alistipes putredinis, and UBA1394 sp900538575, also exhibited negative correlations with plasma ImP, though these did not reach statistical significance (Fig. 3G). Notably, Alistipes putredinis was also the most prevalent species in this genus in the MBS.

In the MLVS, the positive ImP-predicting species, such as Ruminococcus gnavus (p < 0.05), Clostridium scindens (p < 0.05), and Mediterraneibacter glycyrrhizinilyticus (FDR q < 0.1), were associated with higher TG concentrations. In contrast, negative ImP-predicting species, Blautia A sp900066355 and UBA1394 sp900538575, showed inverse associations (p < 0.05); however, these associations were not significant after multiple testing correction. Meanwhile, Mediterraneibacter glycyrrhizinilyticus was associated with decreased HDL-C, whereas UBA1394 sp900538575 was associated with increased HDL-C (p < 0.05, Fig. 4G and Additional file 2: Table SS1).

Gut microbial functional profiling in relation to ImP levels and their associations with CVD risk markers

First, we focused on metagenomic pathways and enzymes involved in histidine utilization related to ImP synthesis in the MLVS (Fig. 4A). We illustrated the top 20 species contributing to histidine ammonia-lyase (EC 4.3.1.3, hutH) and urocanate reductase (EC 1.3.99.33, urdA). Only three species contributors could be identified for urocanate reductase (EC 1.3.99.33, urdA) in the MLVS participants (Additional file 1: Figure S3). After adjusting for potential covariates, the abundances of L-histidine degradation Ⅰ (β: 0.13; 95%CI: 0.01, 0.25) and L-histidine degradation ⅠⅠⅠ (β: 0.17; 95%CI: 0.05, 0.29) pathways as well as histidine ammonia-lyase (EC 4.3.1.3, hutH) (β: 0.14; 95%CI: 0.01, 0.26) were positively associated with plasma ImP levels (Fig. 4B-D). Moreover, individuals with the presence of urocanate reductase (EC 1.3.99.33, urdA) in the gut had a higher level of ImP (β: 0.51; 95%CI: 0.13, 0.89; Fig. 4E, F), which generally aligned with existing knowledge [8]. For the universal pathway and enzyme association analyses, significant results were only observed in positive associations of plasma ImP with caffeine dehydrogenase (EC 1.17.5.2) and 4-hydroxybutyrate dehydrogenase (EC 1.1.1.61) (FDR q < 0.1). A negative association was found for protein arginine kinase (EC 2.7.14.1) (p = 7.2E-04), though it did not remain significant after multiple testing correction (q > 0.1; Additional file 1: Figure S4 and Additional file 2: Table SS2, 3).

When examining the relationship between enzymes and CVD risk markers, we found that increased urocanate reductase (EC 1.3.99.33, urdA) was significantly associated with lower circulating HDL-C levels (β: −0.12; p < 0.05) in multivariable linear regression, while it showed a marginal association with higher TC/HDL-C ratio (β: 0.12; p = 0.06) and a non-significant positive association with hs-CRP levels (β: 0.10; p = 0.11) (Fig. 4G and Additional file 2: Table SS1). Moreover, we identified five gene homologs with at least 90% sequence similarity to known urdA genes in the MLVS. All urdA gene homologs and their stratified profiling were linked to plasma ImP and CVD biomarkers. After adjusting for potential confounding factors, our results showed that the total presence of urdA gene homologs was significantly associated with higher plasma ImP levels (β: 0.59; 95%CI: 0.25, 0.94; FDR q < 0.1). Similar positive associations, though not statistically significant, were observed for individual gene homologs, such as UniRef90_A0A2Z6I9Y0, UniRef90_B0NCC6, and UniRef90_F8DIF2 (Fig. 4H and Additional file 2: Table SS4). Furthermore, we found that the presence of UniRef90_F8DIF2 gene homolog was associated with increased TG (β: 0.58; 95%CI: 0.03, 1.13; p < 0.05) and TC/HDL-C levels (β: 0.71; 95%CI: 0.15, 1.28; p < 0.05).

Interactions between diet, gut microbiome, and plasma ImP

The IMS did not significantly modify the relationship between dietary histidine intake and plasma ImP concentrations (p for interaction = 0.60). A slight, non-significant negative association was observed between histidine intake and plasma ImP levels (β = −0.11, p = 0.22 in the low IMS group; β: −0.05, p = 0.56 in the high IMS group), regardless of IMS (Fig. 5A). Additionally, apart from whole grain intake (p for interaction = 0.02), no statistically significant interaction was found between IMS and other dietary factors (Additional file 1: Table S8).

Fig. 5.

Interaction between histidine intake, gut microbiome, dietary fiber with plasma ImP levels in the MLVS. (A) The scatter plot and fitted regression lines with confidence band shows the association of histidine intake with plasma ImP levels, stratified by microbial score of ImP. (B) Spearman correlations were used to estimate the association of dietary fibers with plasma ImP and ImP-predicting microbial species and enzymes. Each species or enzyme name in red represents a positive association with ImP, while those in green represent a negative association with ImP. (C) Interaction between histidine intake and ImP microbial score on plasma ImP levels in low or high dietary pectin groups. Multivariable models were adjusted for age (years), BMI (kg/m²), physical activity (MET-hrs/wk), alcohol consumption (never, < 5 g/d, 5–9.9 g/d, 10–14.9 g/d, 15–29.9 g/d, ≥ 30 g/d), total energy intake (kcal/d), AHEI, antibiotics use (yes/no), probiotics use (yes/no), colonoscopy (yes/no), and the Bristol Stool Scale score. Abbreviations: MLVS, Men’s Lifestyle Validation Study; ImP, imidazole propionate; BMI, body mass index; AHEI, alternative healthy eating index; MET, metabolic equivalent task

Given the strong inverse associations of dietary fiber, particularly pectin intake, with plasma ImP (Spearman r = −0.22), IMS (-0.20), and ImP-predicting species (range: −0.20 to −0.15) (Fig. 5B), dietary fiber was additionally included in the interaction analyses. The results revealed a significant three-way interaction among dietary pectin, histidine intake, and IMS in relation to plasma ImP concentrations (p for three-way interaction = 0.01, Additional file 1: Figure S5 and Additional file 1: Table S9), indicating that higher histidine intake combined with a higher IMS (p for two-way interaction = 1.7E-04, Additional file 1: Table S10) predicted increased plasma ImP concentrations only under conditions of low dietary pectin intake (Fig. 5C and Additional file 1: Figure S6). We also found similar three-way interactions for total fiber (p = 0.09), soluble fiber (p = 0.09), and insoluble fiber (p = 0.11), without statistical significance. Similar significant three-way interactions in relation to plasma ImP concentrations were observed among animal protein intake, IMS and dietary pectin (p for three-way interaction = 0.001), total fiber (p = 0.02), and insoluble fiber (p = 0.03), as well as amongn red meat intake, IMS and soluble fiber (p = 0.04) (Additional file 1: Figure S5 and Additional file 1: Table S9). Under low fiber intake, high animal protein intake was associated with higher plasma ImP concentrations among individuals with high IMS (p for two-way interaction < 0.05) (Additional file 1: Figure S6 and Additional file 1: Table S10).

Additionally, significant three-way interactions were found between fish intake, IMS, and dietary fiber (including total fiber, pectin, and insoluble fiber); egg intake, IMS, and dietary fiber (total fiber and insoluble fiber); and nut and seed intake, IMS, and total fiber (p for three-way interaction < 0.05) (Additional file 1: Figure S5 and Additional file 1: Table S9). The two-way interactions between these food groups and IMS were not significant after stratifying participants by low versus high dietary fiber intake, probably due to reduced power in this stratified analysis (Additional file 1: Figure S6 and Additional file 1: Table S10). We attempted to replicate the three-way interactions in the MBS, despite the fact that only total fiber was measured in the MBS using the FFQ. However, the three-way interactions among total fiber, IMS, and histidine intake in relation to ImP levels were absent (Additional file 1: Table S11).

Discussion

Leveraging over 30 years of follow-up data from three large prospective cohorts, the present study revealed a significant association between plasma ImP concentrations and a higher risk of developing CHD, while dietary histidine intake, a precursor of ImP, showed a nonsignificant inverse association with CHD risk. By integrating gut metagenome data, 7DDR-assessed dietary data, and circulating levels of ImP from generally healthy men, we identified 17 ImP-predicting species, including 13 positively associated species (e.g., Ruminococcus gnavus, Clostridium symbiosum, and Clostridium scindens) and 4 negatively associated species (e.g., UBA1394 sp900538575, Alistipes communis, and Alistipes putredinis). Further microbial functional analysis showed that urocanate reductase (EC 1.3.99.33, urdA) and urdA gene homologs were associated with unfavorable CVD risk marker profiles. Notably, the ImP-predicting microbial score did not directly modulate the association between dietary histidine intake and plasma ImP levels, unless the intake of fiber, particularly pectin, was low. The three-way interactions among dietary precursors, high ImP-producing microbial capacity, and low pectin intake remained consistent across major dietary sources of histidine, including animal protein and red meat. To our knowledge, the current study is the first attempt to elucidate the interrelationships along the diet-microbiome–ImP–CHD risk axis among free-living US men and women.

The current study is among the first prospective studies that demonstrated a positive association between ImP concentrations and CHD risk. We also found that ImP concentrations were negatively correlated with HDL-C and positively correlated with hs-CRP. The adverse association of ImP with cardiometabolic health was generally consistent with recent studies reporting higher circulating ImP levels in individuals with CVD [10, 11], HF [12, 48, 49], coronary artery disease among people living with HIV [50], T2D [8, 10], and elevated blood pressure [13]. Interestingly, a cross-sectional study found no association between ImP levels and CVD risk factors, including LDL-cholesterol, HDL-cholesterol, and insulin-resistance, among participants without T2D [13].

The mechanisms underlying the potentially detrimental health effects of ImP remain unclear but may involve the effects of ImP on disrupting glucose metabolism and promoting insulin resistance through the p38γ/p62/mTORC1 pathway [8, 9]. This disruption may, in turn, contribute to atherosclerosis [51]. ImP may promote a pro-inflammatory environment in the gut mucosa by activating MAP-kinase p38γ [8], as Molinaro et al. reported that ImP was associated with elevated pro-inflammatory markers, including hs-CRP, and a reduction in MAIT cell counts originating from the gut mucosa [10]. In addition, ImP inhibits AMPK signaling, a key regulator of cellular energy homeostasis with known anti-atherosclerotic and anti-inflammatory properties [9, 52]. Furthermore, epidemiological and mechanistic evidence suggests that elevated circulating ImP levels are associated with atherosclerosis risk in mice and human cohorts [53]. Nevertheless, a recent in vivo and in vitro study indicated that ImP could lower lipid accumulation through alterations in proliferator-activated receptor (PPAR) signaling pathway gene expression [54]. However, given the complex relationships between lipid metabolism and CHD etiology [55], the relevance of this finding to CHD risk should be further examined in future studies. Taken together, disruption of host metabolic regulatory networks may contribute to CHD pathogenesis, although further mechanistic investigation is warranted.

In contrast to the positive association of ImP with CHD risk, histidine intake was consistently associated with a lower CHD risk in the present (though non-significantly) and previous studies [14]. The inverse associations might be explained by histidine’s protective effect through suppressing pro-inflammatory cytokine expression via the NF-κB pathway in adipocytes [56], and reducing oxidative stress [57]. The divergent associations between ImP and histidine suggest that most health effects of histidine intake are independent of ImP production. This observation, in conjunction with the lack of correlation between histidine intake and ImP concentrations, suggests that only a small fraction of histidine intake is shunted to the microbial pathways. As an analogy, only 3–4% of total tryptophan intake, another essential amino acid, is converted to IPA in the human gut [58], and there is no meaningful correlation between tryptophan intake and IPA levels [16]. It remains to be elucidated regarding the proportion of dietary histidine that is converted to ImP in humans. Moreover, histidine is not the only precursor of ImP since this metabolite can also be the product of urocanate metabolism [8]. Nonetheless, the histidine-ImP pathway is probably the dominant pathway through which ImP is produced by the microbiota, since the abundance of hutH, the gene encoding histidine ammonia-lyase, is approximately two orders of magnitude higher than that of urdA, the gene encoding urocanate reductase [8]. Together, observations regarding ImP or IPA are also good examples that the same amino acid intake may lead to divergent health effects, for which the gut microbiota often plays a significant role.

Indeed, we found evidence of significant associations between the gut microbial composition and ImP levels. Several ImP-related species identified in our study were also suggested to contribute to ImP production in previous experimental or epidemiological studies [8, 10]. For example, Molinaro et al. identified 20 species correlated with plasma ImP in the MetaCardis cohort [10], of which five were Clostridium species and one was a Ruminococcus species. Similarly, we identified five Clostridium species [including Enterocloster bolteae (formerly Clostridium bolteae)] and one Ruminococcus species that were significantly associated with ImP concentrations. Of these, Ruminococcus gnavus, Clostridium symbiosum, Clostridium scindens, and Enterocloster bolteae (referred to as Clostridium bolteae in the referenced manuscript) were identified in both studies as being positively associated with ImP concentrations. In terms of negative microbial predictors, the present analysis found associations of Alistipes putredinis and Alistipes communis with plasma ImP concentrations. Alistipes, a relatively new genus of bacteria, was previously identified as one of the dominant genera in the gut microbiota of longevous individuals [59] and may have protective effects against certain diseases [60–62]. For example, Alistipes putredinis, a potential producer of short-chain fatty acids (SCFAs) [60, 62], may help prevent inflammation and hyperglycemia [61, 63], with this effect being enhanced by higher intake of fiber-rich foods [61]. Moreover, a greater abundance of Alistipes putredinis may strengthen the beneficial association of physical activity with body weight change [64]. The two other negatively associated species, UBA1394 sp900538575 and Blautia A sp900066355, have not been documented before. Overall, the current and previous studies collectively suggest that multiple members of the human gut microbial community across different phyla may be involved in the production of ImP.

To our knowledge, this is the first study among free-living individuals to specifically examine the modulatory role of the gut microbiome in the association between dietary histidine and plasma ImP concentrations. Although no direct interaction was observed between dietary histidine and IMS, their combined effect was associated with increased plasma ImP levels only among men with lower pectin intake. The findings were consistent in the present and previous studies that unhealthy diet poor in fiber may lead to a dysbiotic microbial environment that contributes to higher ImP levels [10, 65]. The observed interaction among fiber, dietary precursors, and human gut microbiome in relation to the production of microbiota-derived metabolites paralles our previous findings regarding tryptophan–fiber–microbiome interactions on IPA production [16]. Intriguingly, Sinha et al.’s study offered mechanistic insights from a series of experiments supporting the three-way interaction: fiber, through the function of fiber-processing bacteria, provides monosaccharides to tryptophan-metabolizing species that subsequently spare the tryptophan to IPA producers and thus promotes the production of IPA [18]. Whether this fiber cross-feeding is also among the key mechanisms underlying the three-way interaction that we observed for ImP deserves further investigation.

The strengths of our study include the use of data from three large cohort studies with long term follow-up to assess the associations between plasma ImP concentrations and CHD risk. For the MLVS, we employed a longitudinal study design with repeated measurements of dietary intake through detailed diet records, gut microbiome, and cardiometabolic risk markers, allowing for a comprehensive analysis of diet–microbiome–metabolite interactions. In addition, the MLVS participants were free of major chronic diseases and maintained relatively stable long-term diets and lifestyles, as required by the study’s eligibility criteria. This design minimized potential reverse causation arising from pre-existing diseases or variability in diet and lifestyle. Furthermore, external replication was conducted for the microbial analyses, enhancing the robustness of our findings.

Despite these strengths, several limitations warrant consideration. First, due to the observational nature of the study, we cannot fully rule out residual confounding or infer causality, even after adjusting for known and potential confounders. Second, measurement error in dietary assessments is inevitable. Although the use of 7DDRs in the MLVS likely improved accuracy compared with FFQ-based assessments in the cohort analyses, some degree of misclassification in dietary and microbiome assessments remains possible. Such errors are likely to be non-differential and would tend to attenuate true associations. Third, although we identified 17 ImP-associated species and constructed an IMS, replication across cohorts was partial and the correlation between the IMS and circulating ImP levels was modest. These findings suggest that microbiome–ImP relationships may vary across populations with different characteristics. Nonetheless, our IMS warrants further validation in larger populations. Additionally, dietary assessments and blood collection were not perfectly synchronized in the studies. For example, in the MLVS, plasma ImP was measured at the timing of second dietary assessment, while dietary intake was also assessed 6 months earlier. In the cohort analysis, dietary intake was derived from the average of the two most recent FFQs prior to blood draw, one of which coincided with blood collection and the other approximately 4 years earlier. Although this cumulative averaging approach reduces random error and better reflects habitual intake, some temporal mismatch may persist. Furthermore, given the modest sample size for analyses involving higher-order interactions, these exploratory findings may be underpowered and should be interpreted with caution. Lastly, as our study population primarily consisted of health professionals, the generalizability of our findings to other populations may be limited.

Conclusions

Plasma concentrations of ImP were associated with a higher risk of developing CHD, whereas an opposite but nonsignificant association was observed for histidine intake across three cohorts of U.S. men and women. In healthy U.S. men, we identified 17 microbial taxa significantly associated with plasma ImP concentrations, which was validated in healthy U.S. women. Notably, the panel of ImP-predicting gut microbial species appeared to modify the association between dietary histidine intake and circulating ImP concentrations under conditions of lower dietary pectin intake in a free-living male population. Taken together, these findings contribute to a deeper understanding of the complex host–microbial crosstalk in histidine metabolism and suggest a potential role for prebiotics in modulating interactions along the diet–gut microbiome–disease axis. The potential implications of these interactions for disease outcomes warrant further investigation in large-scale studies.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ImP

Imidazole propionate

CHD

Coronary heart disease

T2D

Type 2 diabetes

HF

Heart failure

CVD

Cardiovascular disease

MI

Myocardial infarction

IPA

Indole-3-propionate

NHS

The Nurses’ Health Study

HPFS

The Health Professionals Follow-up Study

MLVS

The Men’s Lifestyle Validation Study

MBS

The Mind-Body Study

7DDRs

7-day diet records

FFQ

food frequency questionnaire

DASH

Dietary approaches to stop hypertension

EDIP

Empirical dietary inflammatory pattern

BMI

Body mass index

AHEI

Alternative healthy eating index

MET

metabolic equivalent task

TC

Total cholesterol

HDL-C

High-density lipoprotein cholesterol

LDL-C

Low-density lipoprotein cholesterol

TG

Triglycerides

HbA1c

Hemoglobin A1c concentrations

hs-CRP

high-sensitivity C-reactive protein

HR

Hazard ratio

LASSO

Least absolute shrinkage and selection operator

PCoA

Principal coordinates analysis

IMS

Imidazole propionate microbial score

FDR

False discovery rate

Author contributions

QS designed the study. QS, FBH, QQ, AHE, EBR, obtained funding. EBR, QS, AC, CH and KLI collected biospecimen and collected/generated the data. XL conducted data analysis. XL and YH reviewed data analyses. XL and LZ drafted the manuscript. JL, YL, KHL, AHE, ATC, CH, CZ, FBH, QQ, EBR and QS interpreted the data and provided critically important revisions to the manuscripts. All authors critically revised the manuscript for important intellectual content. QS accepts full responsibility for the finished work and/or the conduct of the study, has access to the data and controlled the decision to publish.

Funding

This study is supported by research grant HL035464 and HL060712 from the National Heart, Lung, and Blood Institute (NHLBI), DK119268, DK126698, DK120870, and DK129670 from the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK), and ES036206 from the National Institute of Environmental Health Sciences (NIEHS). The Men’s Lifestyle Validation Study was supported by U01CA152904 from the National Cancer Institute (NCI). The Health Professionals Follow-Up Study is supported by U01 CA167552 from NCI. The Nurses’ Health Study is supported by UM1 CA186107, and R01 HL034594 from the National Institutes of Health (NIH). The Nurses’ Health Study II is supported by U01 CA176726, R01 CA67262 and U01 HL145386 from NIH. The funders had no role in the study design; in the collection, analysis, and interpretation of data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication. All authors confirm the independence of researchers from funders.

Data availability

The metagenomic sequencing data have been deposited in the NCBI Sequence Read Archive under BioProject accession numbers PRJNA354235 (https://www.ncbi.nlm.nih.gov/bioproject/354235) and PRJNA1164832 (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1164832). Metabolomics data are available upon reasonable request to the corresponding author. Individual-level phenotype and covariate data from the Nurses’ Health Study (NHS) and Health Professionals Follow-up Study (HPFS) are available to qualified researchers through approved applications at https://nurseshealthstudy.org/researchers and https://hsph.harvard.edu/research/health-professionals/resources/for-external-collaborators.

Declarations

Ethics approval and consent to participate

The study protocol was approved by the Harvard T.H. Chan School of Public Health Institutional Review Board, with informed consent obtained from all participants. Patients and the public were not involved in the design, conduct, reporting, or dissemination of this research.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.