Plasma Metabolites and Life’s Simple 7 in REGARDS

Abstract

Background:

The American Heart Association’s Life’s Simple 7 (LS7) is a health metric that captures important factors associated with cardiovascular and cerebrovascular health. Previous studies highlight the potential of plasma metabolites to serve as a marker for lifestyle and health behavior that could be a target for stroke prevention. The objectives of this study were to identify metabolites that were associated with LS7, incident ischemic stroke and mediated the relationship between the two.

Methods:

Targeted metabolomic profiling of 162 metabolites by liquid chromatography-tandem mass spectrometry was used to identify candidate metabolites in a stroke case-cohort nested within the Reasons for Geographic and Racial Differences in Stroke (REGARDS) study. Weighted linear regression and weighted Cox proportional hazard models were used to identify metabolites that were associated with LS7 and incident ischemic stroke, respectively. Effect measures were based on a 1-standard deviation change in metabolite level. Metabolite mediators were examined using inverse odds ratio weighting mediation analysis.

Results:

The study comprised 1,075 ischemic stroke cases and 968 participants in the random cohort sample. Three out of 162 metabolites were associated with the overall LS7 score including guanosine (β=−0.46, 95%CI=−0.65 to −0.27, p=2.87x10⁻⁶), cotinine (β=−0.49, 95%CI=−0.70 to −0.28 p=7.74x10⁻⁶), and acetylneuraminic acid (β=−0.59, 95%CI=−0.77 to −0.42, p=4.29x10⁻¹¹). Guanosine (HR =1.47, 95%CI=1.31-1.65, p=6.97x10⁻¹¹), cotinine (HR =1.30, 95%CI=1.16-1.44, p=2.09x10⁻⁶), and acetylneuraminic acid (HR =1.29, 95%CI=1.15-1.45, p=9.24x10⁻⁶) were associated with incident ischemic stroke. The mediation analysis identified guanosine (27% mediation, indirect effect p=0.002), cotinine (30% mediation, indirect effect p=0.004), and acetylneurminic acid (22% mediation, indirect effect p=0.041) partially mediated the relationship between LS7 and ischemic stroke.

Conclusions:

We identified guanosine, cotinine, and acetylneuraminic acid that were associated with LS7, incident ischemic stroke, and mediated the relationship between LS7 and ischemic stroke.

Graphical Abstract

Introduction

Stroke is one of the leading causes of neurological death and disability in the United States and worldwide.^1,2 In addition to the physical burden, stroke also leads to a financial burden to both patients and their families. It is projected that the total direct stroke-related cost will be more than double to approximately $94 billion in 2035.³ Therefore, stroke prevention is essential to reduce health and economic burdens.

To improve cardiovascular health nationally, the American Heart Association’s Life’s Simple 7 (LS7) describes important factors that influence cardiovascular and cerebrovascular health.⁴ The metrics consist of 4 health behaviors, including body mass index (BMI), diet, physical activity, and smoking, and 3 health factors, including blood pressure, total cholesterol, and fasting plasma glucose, where a higher score is indicative of greater health.⁴ A previous REasons for Geographic and Racial Differences in Stroke (REGARDS) study reported that a higher LS7 score was associated with a lower risk of incident hypertension, which is a potent risk factor for stroke.⁵ Accordingly, the LS7 was also found to have an inverse association with incident stroke, where a higher LS7 score was associated with a lower risk of stroke.⁶ Furthermore, a recent study demonstrated that optimal vascular health determined by LS7 can partially counterbalance the high genetic risk for stroke.⁷ Unfortunately, stroke survivors can experience a decline in the LS7 score after stroke.⁸ These findings emphasize the clinical significance of the modifiable risk factors and highlight the need to understand the underlying biological pathways of these risk factors in order to enhance stroke prevention strategies.

Metabolomic profiling can provide insight into interactions between biochemical processes and environmental factors that influence disease development.^9,10 Our previous studies identified plasma metabolites that were associated with incident ischemic stroke independently of other risk factors.¹¹ We also found several plasma metabolites which link dietary patterns with the risk of stroke,¹² some of which are linked to gut microbial metabolism.¹³ Additionally, we demonstrated that a plasma metabolite, gluconic acid, linked hypertension and incident ischemic stroke, specifically among Black adults, and was also associated with socioeconomic and behavioral measures.¹⁴ These studies suggest the potential of the plasma metabolites, which may serve as biomarkers to reflect the lifestyle or health behaviors that impact the risk of stroke. Since LS7 is a parameter that reflects vascular health, we hypothesize that there are plasma metabolites that are associated with LS7 and link the LS7 with stroke risk.

The aims of this study were to identify plasma metabolites that were associated with a composite health metric defined by LS7, assess whether these metabolites associated with LS7 were also associated with incident ischemic stroke, and determine whether these candidate metabolites were mediators between LS7 and incident ischemic stroke. In this study, we used mediation analysis to unravel the complex relationship between health behaviors determined by LS7 and the risk of stroke that was mediated by the plasma metabolites.

Methods

Data for this study are available from the corresponding author upon reasonable request, and in accordance with REGARDS data sharing policy.

Study Design and Population

The REGARDS study is a population-based, prospective cohort study. The study enrolled 30,239 community-dwelling participants aged 45 years old or older between January 2003 and October 2007. Black individuals and those living in the southeastern United States (i.e., the Stroke Belt), were oversampled.¹³ Demographic, clinical, and lifestyle information were collected by computer-assisted telephone interview, a baseline in-person visit, and self-administered questionnaires.⁶ During the in-person visit, blood samples were collected by venipuncture in EDTA tube followed by 10-minute centrifugation prior to transferring into mailer tubes. Then, specimens were shipped overnight to the central laboratory at the University of Vermont, re-centrifuged at 4°C for 30,000g, and stored at −80°C until metabolomic profiling.¹¹

In this study, we used the stroke case-cohort^15,16 nested within the REGARDS study. This study design consisted of all stroke cases who had an incident stroke from the time of enrollment through April 1, 2019 and random cohort samples (controls) who were sampled from the full REGARDS cohort. The controls were selected by a stratified sampling approach based on age, race, and sex to ensure sufficient representation of high-risk groups. This sampling design has been described and used in prior REGARDS studies.^11,14–18 In this analysis, participants who were lost to follow-up, had a history of stroke at the time of enrollment or had no plasma available for metabolite measurements were excluded. We also excluded participants with hemorrhagic stroke and those without a record of stroke type. All participants provided written informed consent. The study was approved by all participating institutional review boards and conducted in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement (Supplement 1).¹⁷

Covariates and incident ischemic stroke

Race, sex, and smoking status were self-reported by the participants. Participants or their proxies were followed-up by telephone interview at 6-month intervals to ascertain hospitalizations and identify suspected stroke events. Stroke was defined according to the definition from the World Health Organization as a focal neurological deficit lasting >24 hours or non-focal neurological symptoms consistent with stroke in neuroimaging.¹⁸ An incident stroke that occurred during the follow-up period was ascertained by a study nurse who performed an initial review of the medical record that was subsequently confirmed by a team of stroke physicians. Income (below versus above $35,000) and education (high school level versus greater than high school level) were dichotomized, similar to a previous REGARDS study.¹⁹ Region of residence was identified as those who lived in stroke belt or buckle and non-belt at study enrollment. Alcohol use was categorized as never, past, or current user. Additional details of the study design and population are described elsewhere.¹²

Life’s Simple 7

LS7 is a cardiovascular health metric proposed by the American Heart Association.⁴ The seven components of LS7 consisted of modifiable factors, including BMI, physical activity, smoking, diet, blood pressure, total cholesterol, and fasting plasma glucose. Each factor has 3 levels, including ideal, intermediate, and poor, and is given a point score of 2,1, and 0 respectively.^6,20 The overall LS7 score was calculated as the sum of 7 components ranging from 0 to 14. The higher score represents more favorable health. The LS7 data were collected at baseline. In REGARDS, diet classification and physical activity were modified from the original LS7 definitions. The LS7 diet component was derived from the response in the Food Frequency Questionnaire. There were 5 components of diet goals in REGARDS including: fruits and vegetables ≥ 4.5 cups/day; fish 3.5 ounces ≥ 2 servings/week; sodium <1500 mg/day; sweets/sugar-sweetened beverages ≤ 450 kcal/week; whole grains (1.1g of fiber in 10 grams of carbohydrates) 1-oz equivalent servings ≥ 3 servings/day. Participants were classified as having an ideal diet score when 4-5 components were met, an intermediate diet score when 2-3 components were met, and a poor diet score when 0-1 component was met.⁶ For physical activity, participants were classified as having an ideal condition when they had 4 or more times per week of intense physical activity, an intermediate when they had 1-3 times per week, and poor when they had no physical activity.⁶ Other components of LS7 in REGARDS were in accordance with the original LS7 definitions.⁴

Metabolomic Profiling

Plasma samples collected at baseline during the in-person visit were used in the metabolomic profiling. In this study, the targeted metabolomic profiling approach was used to detect a predefined list of metabolites as described in our previous studies.^11,12 The panel consisted of a total of 162 metabolites including amino acids, tricarboxylic acid cycle substrates, nucleotides, and nucleosides known to represent metabolites from a wide range of biological pathways using a validated method performed in our previous studies.^11,12,21,22 Full details of metabolite detection are provided elsewhere.^11,21,23,24 Briefly, polar metabolites were extracted by protein precipitation. EDTA plasma 30 µL was mixed with 70 µL of ice-cold acetonitrile/methanol containing deuterated internal standards.^11,21 The extracted metabolites were separated using dual infinity II 1290 high-performance liquid chromatography pumps (Agilent, Santa Clara, CA) on an Xbridge Amide column (2.1x100 mm 3.5 µm, Waters, Milford, MA). Metabolites were detected by a 6495 triple-quadrupole mass spectrometer (Agilent, Santa Clara, CA). The retention time and mass transition were validated against a purified authentic chemical standard for each compound. All peaks of the detected metabolites were integrated and analyzed using MassHunter QQQ Quantitative Analysis software (Agilent, Santa Clara, CA). Each metabolite was quality controlled and normalized to the nearest human pooled plasma samples injected at regular intervals every 10 injections. Since metabolite levels did not conform to a normal distribution, all values were rank-based inverse normal transformed before statistical analysis.^11,12,14

Statistical Analyses

All statistical analyzes were performed using STATA version 17.0 (StataCorp, LLC., College Station, TX). Baseline characteristics were presented as mean ± standard deviation (SD) and percentage for continuous and categorical variables, respectively. To account for the stratified sampling of the random cohort sample, sample weighting was used in all analyses as previously described.^{11,12,14,25,26} To address the data missingness in this study, we use the following strategies. Similar to prior REGARDS studies,^5,6,8 diet had the highest missingness compared to other components in LS7 in this stroke case-cohort (n=646, 32%). Therefore, we performed additional sensitivity analyses by assigning the missing diet score as 0 (poor), which is the most prevalent category. This approach has been used in previous REGARDS publications.^5,8 The list-wise deletion method was otherwise used for other covariates with missing values.

First, we examined whether the LS7 was a predictor of incident ischemic stroke in this case-cohort using weighted Cox proportional hazard regression models adjusted for age, race, sex, age-race interaction, income, education, alcohol use, and region of residence. This allowed us to assess whether the stroke case-cohort sample is similar to the previous REGARDS study that examined LS7 in the larger cohort (n=22,914).⁶

Next, we identified plasma metabolites that were associated with the overall LS7 score using weighted linear regression adjusted for age, race, and sex. Plasma metabolites associated with LS7 that surpassed the Bonferroni correction threshold of 162 tested metabolites (0.05/162 = 0.00031) were carried forward to the next step. Then, weighted Cox proportional hazard regression adjusted for age, race, sex, and age-race interaction was used to determine the association between these candidate metabolites and incident ischemic stroke similar to prior REGARDS studies.^11,12 Effect measures in all models were based on a 1-standard deviation change in metabolite level. The Bonferroni adjusted p-value was used to account for multiple tests at this stage based on the number of metabolites identified in the first step. Pearson’s correlation was used to demonstrate the correlation of the association between the overall LS7 score and incident ischemic stroke of each metabolite.

To determine whether candidate plasma metabolites were mediators between LS7 and incident ischemic stroke, metabolites that demonstrate an association with LS7 and incident ischemic stroke were evaluated in mediation analysis using the inverse odds ratio weighting approach,²⁷ similar to our previous studies.^12,14 In this mediation analysis, the overall LS7 score was treated as binary exposure (quartiles Q3-Q4 versus quartiles Q1-Q2). Briefly, a logistic regression model between LS7 and the metabolite mediator was fit, adjusting for age, race, and sex. Then the inverse odds weight was calculated by taking the inverse of the predicted odds ratio from the logistic regression model. The direct and total effect of LS7 on incident ischemic stroke was calculated by fitting the Cox proportional hazard models of LS7 on incident ischemic stroke adjusted for age, race, sex, and age-by-race interaction, with and without applying inverse odds weight, respectively. The indirect effect (i.e., the percent mediation of metabolite in the relationship between LS7 and the incident stroke) was calculated from the difference between the total and direct effect. The 95% confidence interval for the indirect effect was calculated from bootstrapping (n=500). Finally, weighted multiple linear regression adjusted for age, race, and sex was used to determine the association between metabolite mediators and 7 components of the LS7.

Results

Study population and baseline characteristics

In this stroke case cohort within the REGARDS study, 1,404 stroke cases were identified through April 1, 2019, and 1,127 participants were in the random cohort sample during the mean follow-up time of 7.1 ± 4.5 years. We excluded participants who had a medical history of stroke at baseline (n = 308), did not have plasma available (n=110), had hemorrhagic stroke (n=122), or who were lost follow-up (n=16). There were 74 participants in the random cohort sample who had ischemic stroke during the follow-up period. These participants were included in the ischemic stroke group, similar to prior REGARDS studies.^11,12,14 The final cohort of 2,043 participants included 1,075 ischemic stroke cases (41% Black individuals) and 968 participants in the random cohort sample (48% Black individuals; Figure S1). Participants who developed incident ischemic stroke during the follow-up period were older and had a higher rate of stroke risk factors than participants in the random cohort sample (Table 1).

Table 1:

Baseline characteristics^*

Characteristics	Controls	Ischemic Stroke
Age, years, mean (SD)	64.6 (9.2)	69.5 (8.7)
Female, %	54.8	49.6
Black race, %	48	41
Stroke risk factors, %
- Hypertension	72.1	84.7
- Diabetes mellitus	20.2	28.7
- Cardiovascular disease	15.6	28.8
- Left ventricular hypertrophy	7.6	14.7
- Atrial Fibrillation	8.8	13.8
- Current Smoker	13.7	16.5
Total Life’s Simple 7 score, mean (SD)	7.58 (2.0)	7.11 (2.1)

^*Number of missing observations prior to sample weighting for each variable (n): hypertension (8), diabetes mellitus (24), coronary heart disease (34), left ventricular hypertrophy (28), atrial fibrillation (48), current smoker (9)

Life’s Simple 7 score associated with incident ischemic stroke

Similar to the full REGARDS study,⁶ the overall LS7 score was associated with a lower risk of incident ischemic stroke in this stroke case-cohort (HR=0.85, 95%CI=0.79-0.92, p<0.001). Since the REGARDS study aims to understand racial differences in stroke, we also evaluated whether the association differed between Black and White participants. There was no racial difference in the association between the overall LS7 score and incident ischemic stroke (p for interaction = 0.95).

Metabolites association with Life’s Simple 7

Since the overall LS7 score was normally distributed (p skewness = 0.20), we next examined the association between plasma metabolites and the overall LS7 score using weighted linear regression adjusted for age, race, and sex. Among 162 metabolites, 21 were associated with the overall LS7 score and exceeded the Bonferroni corrected threshold. These metabolites were dimethlyguanidinovaleric acid, acetylneuraminic acid, C3 malonylcarnitine, uric acid, maleic acid, gluconic acid, cytidine, C6 carnitine, leucine, acetylglutamate, guanosine, indole-3-propanoic acid, adenosylhomocysteine, pyruvic acid, cotinine, acotinic acid, proline, 𝜶-ketoglutaric acid, C7 carnitine, aminoadipic acid, and kynurenic acid (Table 2).

Table 2:

Metabolites associated with Life Simple 7 score.

Metabolites	Life Simple 7
	β (95%CI)	p value
Dimethylguanidino valeric acid	−0.67 (−0.84 to −0.5)	2.26E-14*
Acetylneuraminic acid	−0.59 (−0.77 to −0.42)	4.29E-11*
C3 Malonylcarnitine	−0.56 (−0.72 to −0.39)	7.68E-11*
Uric acid	−0.56 (−0.73 to −0.38)	4.06E-10*
Maleic acid	−0.51 (−0.67 to −0.35)	8.23E-10*
Gluconic acid	−0.50 (−0.67 to −0.33)	1.6E-08*
Cytidine	−0.43 (−0.60 to −0.27)	1.71E-07*
C6 Carnitine	−0.48 (−0.66 to −0.30)	2.19E-07*
Leucine	−0.51 (−0.70 to −0.31)	3.13E-07*
Acetylglutamate	−0.42 (−0.60 to −0.25)	1.92E-06*
Guanosine	−0.46 (−0.65 to −0.27)	2.87E-06*
Indole-3-propanoic acid	0.43 (0.25 to 0.61)	3.14E-06*
Adenosylhomocysteine	−0.40 (−0.57 to −0.23)	5.59E-06*
Pyruvic acid	−0.39 (−0.56 to −0.22)	6.31E-06*
Cotinine	−0.49 (−0.70 to −0.28)	7.74E-06*
Acotinic acid	−0.36 (−0.53 to −0.20)	1.83E-05*
Proline	−0.38 (−0.55 to −0.20)	4.21E-05*
α Ketoglutaric acid	−0.36 (−0.53 to −0.19)	4.84E-05*
C7 Carnitine	−0.37 (−0.55 to −0.18)	1.13E-04*
Aminoadipic acid	−0.33 (−0.50 to −0.16)	1.24E-04*
Kynurenic acid	−0.34 (−0.52 to −0.16)	2.09E-04*

β represents the change of Life Simple 7 score per standard deviation increment of the baseline level of each metabolite.

Model: age + race + sex + metabolite

^*exceeds the Bonferroni correction threshold of 162 tests

Metabolites association with incident ischemic stroke

Among the 21 metabolites that were associated with the overall LS7 score, 9 metabolites were associated with incident ischemic stroke in a weighted Cox proportional hazard model adjusted for age, race, sex, and age-by-race interaction and exceeded the Bonferroni corrected threshold. These metabolites were dimethlyguanidinovaleric acid (HR=1.31, 95%CI=1.18-1.46, p=8.48x10⁻⁷), acetylneuraminic acid (HR=1.29, 95%CI=1.15-1.45, p=9.24x10⁻⁶), C3 malonylcarnitine (HR=1.18, 95%CI=1.07-1.31, p=1.48x10⁻³), uric acid (HR=1.18, 95%CI=1.06-1.31, p=1.80x10⁻³), gluconic acid (HR=1.30, 95%CI=1.16-1.46, p=5.11x10⁻⁶), cytidine (HR=1.24, 95%CI=1.11-1.38, p=9.50x10⁻⁵), acetylglutamate (HR=1.19, 95%CI=1.07-1.33, p=1.04x10⁻³), guanosine (HR=1.47, 95%CI=1.31-1.65, p=6.97x10⁻¹¹), and cotinine (HR=1.30, 95%CI=1.16-1.44, p=2.09x10⁻⁶) (Table 3).

Table 3:

Metabolites associated with incident ischemic stroke

Metabolites	Ischemic Stroke
	HR (95%CI)	p value
Dimethylguanidino valeric acid	1.31 (1.18-1.46)	8.48E-07*
Acetylneuraminic acid	1.29 (1.15-1.45)	9.24E-06*
C3 Malonylcarnitine	1.18 (1.07-1.31)	1.48E-03*
Uric acid	1.18 (1.06-1.31)	1.80E-03*
Maleic acid	1.09 (0.99-1.21)	9.34E-02
Gluconic acid	1.30 (1.16-1.46)	5.11E-06*
Cytidine	1.24 (1.11-1.38)	9.50E-05*
C6 Carnitine	1.05 (0.95-1.16)	3.60E-01
Leucine	1.10 (0.98-1.23)	9.32E-02
Acetylglutamate	1.19 (1.07-1.33)	1.04E-03*
Guanosine	1.47 (1.31-1.65)	6.97E-11*
Indole-3-propanoic acid	0.88 (0.80-0.97)	1.34E-02
Adenosylhomocysteine	1.17 (1.06-1.30)	2.62E-03
Pyruvic acid	1.03 (0.94-1.13)	5.24E-01
Cotinine	1.30 (1.16-1.44)	2.09E-06*
Acotinic acid	1.09 (0.98-1.20)	9.73E-02
Proline	1.09 (0.99-1.21)	8.59E-02
α Ketoglutaric acid	0.99 (0.90-1.10)	9.10E-01
C7 Carnitine	1.12 (1.02-1.23)	2.23E-02
Aminoadipic acid	1.07 (0.97-1.18)	1.71E-01
Kynurenic acid	1.12 (1.01-1.25)	4.02E-02

Hazard ratios (HRs) represent the HR of incident ischemic stroke per standard deviation of the increment of the baseline level of metabolite.

Model: age + race + age*race + sex + metabolite

^*exceeds the Bonferroni correction threshold of 21 tests

Metabolites highly associated with the LS7 score were also highly associated with incident ischemic stroke. There was a negative correlation between the beta coefficient of the LS7 score and the hazard ratio of the incident ischemic stroke (r² = −0.63, p=0.002) (Figure 1).

Figure 1:

Scatter plot between the hazard ratio of incident ischemic stroke and the beta coefficient of the Life’s Simple 7 of the 21 metabolites.

Metabolites as a mediator between Life’s Simple 7 and ischemic stroke

Nine metabolites that demonstrated an association with LS7 and incident ischemic stroke were carried forward in the mediation analysis. We found that guanosine (27% mediation, indirect effect p=0.002), cotinine (30% mediation, indirect effect p=0.004), and acetylneuraminic acid (22% mediation, indirect effect p=0.041), partially mediated the relationship between LS7 and incident ischemic stroke (Table 4).

Table 4:

Metabolite mediators of Life Simple 7 and incident ischemic stroke

Metabolites	Direct Effect		Indirect Effect		% Mediation
	β (95%CI)	p value	β (95%CI)	p value
Dimethylguanidino valeric acid	0.42 (0.12-0.73)	0.007	0.05 (−0.09 to 0.20)	0.462	-
Acetylneuraminic acid	0.37 (0.08-0.66)	0.013	0.11 (0.00 −0.21)	0.041	22
C3 Malonylcarnitine	0.43 (0.15-0.71)	0.003	0.05 (−0.05 to 0.15)	0.352	-
Uric acid	0.38 (0.09-0.66)	0.010	0.1 (−0.01 to 0.21)	0.063	-
Gluconic acid	0.44 (0.15-0.72)	0.003	0.04 (−0.06 to 0.14)	0.437	-
Cytidine	0.44 (0.15-0.72)	0.003	0.04 (−0.05 to 0.13)	0.406	-
Acetylglutamate	0.43 (0.15-0.71)	0.002	0.05 (−0.02 to 0.11)	0.148	-
Guanosine	0.35 (0.06-0.63)	0.016	0.13 (0.05-0.21)	0.002	27
Cotinine	0.33 (0.04-0.63)	0.028	0.14 (0.05-0.24)	0.004	30

Note: Total effect β (95%CI) = 0.48 (0.21-0.74), p value 4.11E-04. The β’s represent Cox regression coefficients which are on the log-hazard ratio scale.

Association between metabolite mediators and components in Life’s Simple 7

To gain further insight into the metabolite mediators and LS7, we next examined the association between 3 candidate metabolite mediators and the components of LS7 in a weighted multiple linear regression adjusted for age, race, and sex. In this analysis, guanosine level was associated with smoking, BMI, diet, and cholesterol. Cotinine (a metabolite of nicotine) was associated with smoking, whereas acetylneuraminic acid level was associated with diet and fasting blood glucose (Table 5).

Table 5:

Metabolites associated with individual components of the Life Simple 7

Factors	Guanosine		Cotinine		Acetylneuraminic acid
	β (95%CI)	p value	β (95%CI)	p value	β (95%CI)	p value
Smoking
- Intermediate	−0.82 (−1.37 to −0.27)	0.004	−0.96 (−1.82 to −0.10)	0.029	−0.12 (−0.77 to 0.52)	0.707
- Ideal	−0.25 (−0.47 to −0.03)	0.025	−1.54 (−1.69 to −1.39)	<0.001	−0.31 (−0.55 to −0.08)	0.008
Body Mass Index
- Intermediate	−0.32 (−0.49 to −0.14)	0.001	0.14 (−0.03 to 0.31)	0.116	−0.08 (−0.28 to 0.13)	0.463
- Ideal	−0.43 (−0.62 to −0.24)	<0.001	−0.02 (−0.21 to 0.18)	0.878	−0.10 (−0.30 to 0.09)	0.302
Physical Activity
- Intermediate	−0.12 (−0.30 to 0.06)	0.189	0.13 (−0.03 to 0.29)	0.105	−0.19 (−0.37 to −0.02)	0.032
- Ideal	−0.14 (−0.32 to 0.03)	0.106	0.15 (−0.04 to 0.33)	0.115	−0.18 (−0.38 to 0.02)	0.074
Diet
- Intermediate	−0.27 (−0.45 to −0.09)	0.004	−0.14 (−0.32 to 0.04).	0.12	−0.28 (−0.48 to −0.09)	0.005
- Ideal*	-	-	-	-	-	-
Cholesterol
- Intermediate	0.29 (0.07 to 0.50)	0.008	−0.07 (−0.28 to 0.13)	0.495	0.19 (−0.05 to 0.44)	0.123
- Ideal	0.37 (0.14 to 0.59)	0.001	0.04 (−0.18 to 0.27)	0.709	0.24 (−0.02 to 0.50)	0.068
Blood pressure
- Intermediate	0.03 (−0.16 to 0.22)	0.764	−0.08 (−0.25 to 0.10)	0.384	−0.09 (−0.29 to 0.12)	0.397
- Ideal	−0.08 (−0.33 to 0.17)	0.542	−0.24 (−0.48 to 0.01)	0.062	−0.15 (−0.40 to 0.10)	0.234
Fasting blood glucose
- Intermediate	−0.03 (−0.29 to 0.22)	0.794	0.1 (−0.11 to 0.32)	0.35	−0.49 (−0.86 to −0.13)	0.008
- Ideal	−0.05 (−0.29 to 0.19)	0.666	0.05 (−0.16 to 0.25)	0.67	−0.80 (−1.16 to −0.43)	<0.001

β represents the change of metabolite level per unit of standard deviation of the baseline level of each analysis.

Model: age + race + sex + smoking + body mass index + physical activity + diet + cholesterol + blood pressure + fasting blood glucose

^*Category had no data

Sensitivity analyses

We performed a sensitivity analysis to determine whether aspirin and statin usage (which may impact the LS7 score), altered the list of metabolites identified in the main analysis. In the sensitivity analysis, we found 24 metabolites associated with the overall LS7 score after adjustment (Table S1). Among these, 18 metabolites overlapped the main analysis. Six additional metabolites were identified including salicylic acid, lactic acid, cyclic AMP, glycocholic acid, glycochenodeoxycholic acid, and UDP N-acetylglucosamine. However, none of these additional metabolites were associated with incident ischemic stroke nor surpassed the Bonferroni corrected threshold.

We also performed sensitivity analyses to account for the data missingness in the diet component of LS7. Imputing missing diet data resulted in 500 additional participants with a complete LS7 score. Similar to the primary analysis, the overall LS7 score was associated with lower incident ischemic stroke (HR=0.85, 95%CI=0.81-0.91, p<0.001). Guanosine, cotinine and acetylneuraminic acid remained associated with LS7. Moreover, 7 additional metabolites were identified in association with LS7 (Table S2); however, none were associated with incident stroke (Table S3).

Discussion

In this study, we demonstrated that a higher overall LS7 score was associated with lower incident ischemic stroke in the REGARDS study, which is consistent with prior findings.⁶ Using targeted metabolomics, we further identified 3 metabolites, including guanosine, cotinine, and acetylneuraminic acid, that were associated with the overall LS7 score, incident ischemic stroke, and mediated the relationship between LS7 and ischemic stroke. These metabolites also demonstrated an association with the individual components of LS7.

Previous work by our group identified guanosine in association with incident ischemic stroke independently of other traditional stroke risk factors.¹¹ In our subsequent studies, we also demonstrated that guanosine was inversely correlated with a plant-based diet, increased stroke risk, and mediated the relationship between a plant-based diet and stroke risk.¹² Furthermore, we also identified guanosine as one of the gut microbiota-associated metabolites which increased the risk of ischemic stroke.²⁸ Consistent with prior studies, we found that guanosine was associated with the diet component of LS7 score and linked the LS7 and incident ischemic stroke. Dietary nucleosides, including guanosine, are absorbed by enterocytes and could influence immune function and intestinal microbiota.²⁹ Accordingly, consistent evidence suggests that guanosine level is associated with diet. Although the mechanism of how guanosine contributes to the increased risk of stroke remains unclear, there is some evidence suggesting that guanosine could lead to the cytotoxicity of endothelial cells.³⁰

Cotinine is the main metabolite of nicotine after tobacco exposure.^31,32 Smoking is a well-established stroke risk factor.³³ There are several mechanisms by which smoking increases the risk of stroke, such as increased platelet aggregation, increased fibrinogen level, direct toxic effect, and accelerated atherosclerosis.^33–35 In this study, we demonstrated that cotinine mediated 30% of the relationship between the overall score of LS7 and the risk of ischemic stroke. We also found that cotinine was associated with smoking but not other components of the LS7. This finding demonstrates that plasma metabolites could be a marker for a specific health behavior that also reflects the risk of stroke. Furthermore, the relationship between cotinine and smoking component of the LS7 identified in our study could be considered as a positive control and demonstrates the validity of our approach regarding both metabolite detection and statistical analysis.

Acetylneuraminic acid is a type of sialic acid, which is a 9-carbon monosaccharide that often serves as the terminal sugar of N-linked or O-linked glycans.^36,37 This monosaccharide plays an important role in the mammalian glycosylation system, immune system signaling and affects the gut microbiome.³⁶ Sialic acid is a component of the mucus layer in the gastrointestinal tract and is also found in meat-based products.³⁷ In our prior study using exploratory factor analysis, we found acetylneuraminic acid as one of the gut microbiota-associated metabolites. This group of metabolites is associated with the risk of stroke and the Southern dietary pattern, which is known to increase stroke risk. ²⁸ In this study, we identified acetylneuraminic acid as one of the metabolite mediators that link LS7 and the risk of stroke. Consistent with our recent study, we also found that acetylneuraminic acid was associated with the diet component of LS7. Collectively, these findings suggest the potential role of acetylneuraminic acid as gut microbiota-associated metabolite, which is related to dietary pattern and impact stroke risk. Increasing evidence suggests that acetylneuraminic acid is associated with inflammatory processes and atherosclerosis, including myocardial infarction.³⁸ In addition, an increase in the level of circulating plasma acetylneuraminic acid is associated with a poor clinical outcome in patients with heart failure.³⁹ It has been proposed that acetylneuraminic acid contributes to atherosclerosis through several mechanisms such as insulin resistance,⁴⁰ regulation of lipid metabolism,⁴¹ and platelet aggregation.⁴² Furthermore, functional metabolomics demonstrated that acetylneuraminic acid causes myocardial injury in vitro and in vivo, which was alleviated by a neuraminidase inhibitor.⁴³

The findings in this study are in line with our prior work, where we identified plasma metabolites that are associated with the risk of stroke independently of other traditional stroke risk factors.¹¹ Some of these stroke-related metabolites link between dietary patterns and stroke risk,¹² whereas others are markers of hypertension,¹⁴ gut microbiome-related metabolites,²⁸ or specifically associated with recurrent stroke.⁴⁴ Collectively, increasing evidence suggests that lifestyle or health behaviors can be reflected quantitatively using plasma metabolites. Using metabolites which integrate endogenous biological pathways and exogenous health behaviors may better determine the stroke risk than the self-reported health behavior. These metabolites may also serve as surrogate markers to provide personalized intervention for stroke prevention. Further studies are warranted to determine the potential of metabolites as biomarkers for stroke prevention at the population level and whether changes in lifestyle or health behaviors can alter the metabolite levels, which ultimately influence stroke risk.

There are several strengths in this study. First, the case-cohort study design included a large number of ischemic stroke cases. Second, this study contained extensive information regarding the characteristics and health behaviors of the participants that are essential for stroke prevention. Third, targeted metabolomic profiling detects biologically relevant metabolites with high sensitivity and specificity.^12,14 However, there are some limitations in this study. This is an observational study design; therefore, the association demonstrated in this study does not necessarily suggest a causal relationship. In addition, the single measurement of the LS7 score in this study limits the interpretation of the changes in LS7 score over time which may influence the risk of stroke. Although the temporality is an essential assumption of the mediation analysis where the mediator should be a consequence of the exposure, the temporality between LS7 and metabolites cannot be definitively established since the association between LS7 and metabolites was cross-sectional by design. Another limitation was the missing diet scores in a relatively large number of participants. However, the sensitivity analyses identified a list of metabolites consistent with the main findings. We also acknowledge the updated Life’s Essential 8 score which includes a sleep health in the metric in addition to the original components of the LS7 score.⁴⁵ However, sleep quality was not collected at baseline in REGARDS so we did not assess Life’s Essential 8 and the metabolites in this study.

Conclusions

In this study, we identified 3 metabolites, including guanosine, cotinine, and acetylneuraminic acid, that were associated with LS7, incident ischemic stroke, and mediated the relationship between LS7 and stroke. The findings in our study provide the potential of metabolites as biomarkers for personalized stroke prevention.

Non-standard Abbreviations and Acronyms:

REGARDS

REasons for Geographic and Racial Differences in Stroke

LS7

Life’s Simple 7