Hepatocyte Growth Factor is Positively Associated with Risk of Stroke: The Multi-Ethnic Study of Atherosclerosis (MESA)

Abstract

Background and purpose

Hepatocyte growth factor (HGF) is positively associated with ischemic and hemorrhagic stroke risk factors. However, understanding of the relation between HGF and stroke is in its infancy. Therefore, we sought to examine the association of circulating HGF with incident stroke using data from the Multi-Ethnic Study of Atherosclerosis. We hypothesized that circulating HGF would be positively associated with an increased risk of stroke.

Methods

Participants aged 45–84 years (n=6,711) had HGF measured between 2000 and 2002, and were followed for incident stroke through 2013 (n=233). Cox proportional hazards regression was used to calculate hazard ratios and 95% confidence intervals for incident stroke. A secondary analysis stratified results by adjudicated stroke type (n=183 ischemic, n=39 hemorrhagic, n=11 other).

Results

After adjustment for potential confounding variables, risk of stroke was 17% higher with each standard deviation increase in HGF (Hazard ratio, 1.17; 95% confidence interval, 1.03–1.34). This association was mainly driven by ischemic stroke, and did not change upon exclusion of cardioembolic strokes, although the number of excluded cases was small. The few hemorrhagic and other types of stroke were not associated with HGF.

Conclusions

Circulating HGF was positively associated with incidence of stroke in a diverse, population-based cohort of men and women from the United States. Our findings support the hypothesis that circulating HGF is a marker of endothelial damage and suggest that HGF may have utility as a prognostic marker of stroke risk.

Introduction

The protein hepatocyte growth factor (HGF) and its receptor c-Met are produced in response to tissue injury and are functional in tissue repair mechanisms. Their favorable effects in the heart and vasculature include anti-inflammatory, anti-fibrotic, and pro-angiogenic actions.¹ Because HGF is released in response to endothelial injury, higher levels of circulating HGF are associated with hypertension, diabetes, smoking, increased age, body mass index (BMI), and presence and severity of atherosclerotic disease in both the heart and lower extremities.^1–12

Despite evidence of a relation of circulating HGF with ischemic and hemorrhagic stroke risk factors, understanding of the connection between HGF and stroke is in its infancy. Although previous studies indicate a positive relation between circulating HGF and both ischemic stroke and prognostic factors for the development of stroke, they are limited in that they were conducted in select populations that may not generalize.^13–16 Additionally, only one of these studies measured HGF before stroke, which is important to establish temporality. Finally, the associations of HGF with hemorrhagic stroke and all stroke (hemorrhagic and ischemic combined) have not been reported.

Therefore, we sought to examine the relation between circulating HGF and incident stroke using data from a large, multi-ethnic, population-based prospective cohort study: the Multi-Ethnic Study of Atherosclerosis (MESA). We hypothesized that circulating HGF would be positively associated with an increased risk of incident stroke. As a secondary analysis, we explored if there were differences in the association with HGF by stroke type.

Methods

Study Population

MESA is a prospective cohort study that was initiated to investigate the prevalence, correlates, and progression of subclinical cardiovascular disease.¹⁷ From 2000–2002, MESA recruited 6814 participants aged 45–84 and free of clinically recognized cardiovascular disease from populations near 6 field centers in Baltimore, Maryland; Chicago, Illinois; Forsyth County, North Carolina; Los Angeles, California; New York, New York; and St Paul, Minnesota. Cardiovascular disease at baseline was defined as heart attack, angina, stroke, transient ischemic attack, heart failure, resuscitated cardiac arrest, or having undergone procedures related to cardiovascular disease. The study participants were non-Hispanic white (38%), African (28%), Hispanic (22%), and Chinese (12%) Americans. 6772 participants had serum HGF measured at baseline as part of an ancillary study (HL98077 – MESA Adhesion Study). Of these 6772 participants, we excluded individuals from all analyses if they had an extreme value of HGF defined as four or more standard deviations beyond the mean (n=31), or had not been followed after baseline (n=30). Our final sample size for statistical analyses was 6711. The Institutional Review Boards at participating centers approved MESA and its ancillary studies, and all participants gave written informed consent.

Baseline Measurements

Measurement of HGF

Venous blood was obtained from fasting participants. Serum separation was performed within 30 minutes of phlebotomy, and aliquots were subsequently stored at −70°C. Circulating levels of HGF protein were measured in serum using a quantitative sandwich enzyme-linked immunosorbent assay (ELISA) with the Quantikine Human HGF Immunoassay kit (R&D Systems, Minneapolis, Minnesota, USA). This method was validated by R&D systems, as specified in the package insert, and verified by the University of Minnesota laboratory that measured HGF for this study. The lower limit of detection was 40 pg/ml. The interassay laboratory coefficients of variation were 12.0, 8.0, and 7.4% at respective mean concentrations of 687, 2039, and 4080 pg/ml for lyophilized manufacturer’s controls, and 10.4% at a mean concentration of 688 pg/ml for an in-house pooled serum control.

Other Baseline Measurements

Sex, age, race/ethnicity, cigarette smoking status, and medication use were self-reported. In addition, participants were asked to bring all medications to the exam. BMI was calculated as weight over height squared (kg)/(m²). Resting blood pressure was measured three times in the seated position using a Dinamap model Pro 100 automated oscillometric sphygmomanometer (Critikon, Tampa, Florida, USA). The average of the last two measurements was used in analyses. Participants were asked to fast for at least 8 hours before their visit. Serum glucose was assayed by a glucose oxidase method on the Vitros analyzer (Johnson & Johnson Clinical Diagnostics, Rochester, New York, USA). Diabetes was defined as use of insulin or other diabetes medication, self-reported physician diagnosis, fasting glucose ≥126 mg/dL, or non-fasting glucose ≥200 mg/dL. Total cholesterol was measured in ethylenediaminetetraacetic (EDTA) plasma using a cholesterol oxidase method (Roche Diagnostics, Indianapolis, Indiana, USA) on a Roche COBAS FARA centrifugal analyzer. After precipitation of non-high-density lipoprotein cholesterol with magnesium/dextran, high-density lipoprotein cholesterol was also measured in EDTA plasma using the cholesterol oxidase method (Roche Diagnostics). Using the Friedewald formula, low-density lipoprotein cholesterol was calculated in plasma samples with a triglyceride value <400 mg/dL. Triglyceride was measured in EDTA plasma using Triglyceride Glycerol Blanked reagent (Roche Diagnostics) on a Roche COBAS FARA centrifugal analyzer.

Stroke Ascertainment

Forms for ascertaining events are available on the MESA website (http://www.mesa-nhlbi.org). A standard adjudication protocol was used to classify events, as previously reported.¹⁸ In brief, a telephone interviewer contacted each participant at intervals of 9–12 months to inquire about all interim hospital admissions, cardiovascular outpatient diagnoses and procedures, and deaths. In order to verify self-reported diagnoses, copies of death certificates and medical records were requested. Hospital medical records were obtained on 99% of hospitalized cardiovascular events, and some information was obtained on 97% of outpatient diagnostic encounters.

Two vascular neurologists independently reviewed and classified stroke events and assigned incidence dates. If a disagreement between reviewers persisted after review and adjudication, the case was discussed and a consensus reached. Reviewers classified stroke as present if a focal neurologic deficit lasted at least 24 hours or until death or, for deficits lasting less than 24 hours, a clinically relevant lesion was evident on brain imaging with computed tomography or magnetic resonance imaging and no non-vascular cause was identified. Patients with focal neurologic deficits from a non-vascular cause were adjudicated as not a stroke. Strokes were classified on the basis of neuroimaging or other tests into three types: ischemic, hemorrhagic, or other. ‘Other’ strokes included those with incomplete evaluations and with rare, specific causes of stroke. Using strict criteria that required a complete stroke evaluation, ischemic strokes were subtyped into large-artery atherosclerosis, cardioembolism, small-vessel occlusion, stroke of other determined etiology, or stroke of undetermined etiology. Hemorrhagic strokes were subtyped into intraparenchymal hemorrhage, subarachnoid hemorrhage, or other hemorrhage.

Incident stroke was defined as the first occurrence of an adjudicated stroke from baseline through the end of follow-up on December 31, 2013. Follow-up for all analyses was calculated as the time elapsed from the baseline examination to whichever came first: first stroke event (regardless of type), loss to follow-up, death, or December 31, 2013.

Statistical Analyses

Unless stated otherwise, analyses were performed using SAS statistical software, version 9.4. A P Value of <0.05 on a two-tailed test was considered statistically significant. Baseline participant characteristics are presented by HGF tertile, and compared using the chi-squared test for categorical variables and the analysis of variance F-test for continuous variables. Stroke incidence rates by HGF tertile were calculated using Poisson regression. A Kaplan-Meier plot was produced using R statistical software. We used Cox regression with restricted cubic splines to visually check the linearity of the relation between HGF and stroke, modeling HGF as a continuous variable, by graphing the natural log of the hazard ratio (HR) on the Y-axis (data not shown).¹⁹ Because we found no evidence of non-linearity, we modeled HGF as a continuous measure. Cox regression was used to calculate HRs and 95% CIs for incident stroke per one standard deviation (259 pg/ml) increase in HGF. Model 1 adjusted for age, race/ethnicity, and sex. Model 2 adjusted for variables in Model 1 plus baseline values of BMI, smoking status, diabetes mellitus, systolic blood pressure, antihypertensive medication use, high-density lipoprotein cholesterol, total cholesterol, low-density lipoprotein cholesterol, triglycerides, lipid-lowering medication use, and field center site. Further inclusion of variables known to be risk factors for stroke and/or associated with circulating HGF—cancer, estimated glomerular filtration rate, diastolic blood pressure, level of education, physical activity, interleukin-6, D-dimer, C-reactive protein and left ventricular hypertrophy—did not appreciably change point estimates, and thus were not included in our final model. By including cross-product terms in models, we tested for interactions with age, sex, race/ethnicity, and hypertension status. We verified the proportional hazards assumption through inspection of ln(−ln) survival curves by HGF tertile and evaluation of the Schoenfeld residuals. As a secondary analysis, we used Cox regression to calculate HRs by stroke type. We censored person-years at the time of the first stroke, regardless of type, for this secondary analysis. As a sensitivity analysis, we restricted follow-up time to three years. As an additional sensitivity analysis, we modeled HGF as a time-varying covariate, using baseline and, when available, visit 2 measurements of HGF. MESA measured HGF in everyone at baseline and in a random sample of 2418 people at visit 2. If a participant did not have HGF measured at visit 2, the measurement from visit 1 was carried forward.

Results

The mean age of MESA participants at baseline was 62 years, and approximately 50% were female. A review of baseline characteristics showed that age, sex, race/ethnicity, systolic blood pressure, use of antihypertensive or lipid-lowering medications, hypertension category, diabetes, smoking status, BMI, total cholesterol, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, and triglycerides differed by HGF tertile. Diastolic blood pressure did not differ by HGF tertile (Table 1). Baseline characteristics of participants according to incident stroke status are also presented (Table I in the online-only Data Supplement).

Table 1.

Characteristics of Participants According to Tertiles of Hepatocyte Growth Factor at Baseline, Multi-Ethnic Study of Atherosclerosis, 2000–2002

	Hepatocyte Growth Factor Tertile			P value for between- group difference
Characteristics (means or prevalences unless otherwise stated)	<809 pg/ml (n=2237)	809–1011 pg/ml (n=2237)	>1011 pg/ml (n=2237)
Range of hepatocyte growth factor, pg/mL	292–808	809–1011	1012–2152
Hepatocyte growth factor ± SD	682 ± 93	906 ± 57	1227 ± 198	N/A
Age, years ± SD	59 ± 9	62 ± 10	65 ± 10	<0.0001*
Male, n (%)	1153 (52)	1018 (46)	996 (45)	<0.0001†
Race/ethnicity				<0.0001†
Non-Hispanic white American	971 (43)	821 (37)	802 (36)
Chinese American	398 (18)	259 (12)	140 (6)
African American	605 (27)	653 (29)	591 (27)
Hispanic American	264 (12)	506 (23)	701 (31)
SBP, mmHg ± SD	122 ± 20	127 ± 21	131 ± 22	<0.0001*
DBP, mmHg ± SD	72 ± 10	72 ± 10	72 ± 10	0.3*
Antihypertensive medication use, n (%)	613 (27)	837 (37)	1037 (46)	<0.0001†
Hypertension categories, n (%)				<0.0001†
Normotensive, SBP <140 and DBP <90 mmHg	1481 (66)	1229 (55)	999 (45)
Controlled Hypertensive, SBP <140 and DBP <90 mmHg	546 (24)	746 (33)	956 (43)
Uncontrolled Hypertensive, SBP ≥140 or DBP ≥90 mmHg	211 (9)	263 (12)	279 (13)
Diabetes, n (%)	139 (6)	258 (12)	436 (20)	<0.0001†
Current smoker, n (%)	204 (9)	268 (12)	396 (18)	<0.0001†
Body mass index, kg/m2 ± SD	27 ± 5	28 ± 5	30 ± 6	<0.0001*
Lipid-lowering medication use, n (%)	308 (14)	360 (16)	415 (19)	<0.0001†
Total cholesterol, mg/dL ± SD	194 ± 35	196 ± 36	192 ± 36	0.0005*
High-density lipoprotein cholesterol, mg/dL ± SD	53 ± 16	51 ± 15	49 ± 14	<0.0001*
Low-density lipoprotein cholesterol, mg/dL ± SD	117 ± 31	119 ± 31	115 ± 32	0.0006*
Triglycerides, mg/dL ± SD	118 ± 76	135 ± 103	142 ± 84	<0.0001*

DBP indicates diastolic blood pressure; N/A, not applicable; SBP, systolic blood pressure; SD, standard deviation.

^*P Value calculated from analysis of variance F-test.

^†P Value calculated from chi-squared test.

During the 74 271 person-years of follow-up (median 12.2, maximum 13.5 years), 233 incident strokes were identified, of which 183 were ischemic (17 were large-artery atherosclerosis, 21 cardioembolisms, 24 small-vessel occlusions, 5 strokes of other determined etiology, and 119 strokes of undetermined etiology [of strokes of undetermined etiology, 4 had two or more causes identified, 28 had negative evaluations, and 87 had incomplete evaluations]); 39, hemorrhagic (32 were intraparenchymal hemorrhages, 6 subarachnoid hemorrhages, and 1 other hemorrhage); and 11, other. The crude incidence rates and 95% CIs of stroke per 1000 person-years across increasing HGF tertiles were 1.64 (1.22–2.21), 2.60 (2.04–3.32), and 5.43 (4.56–6.47) (Table II in the online-only Data Supplement). During a maximum of 13.5 years of follow-up, the proportion that survived free of stroke was significantly different by HGF tertile (log-rank test, P=<0.0001) (Figure 1).

Figure 1.

Percent surviving free of stroke by HGF tertiles, MESA, 2000–2013.

Circulating HGF was a risk marker for incident stroke (Table 2). After adjustment for sex, race/ethnicity, and age, risk of stroke was 30% higher with each standard deviation increase in HGF. Adjustment for other stroke risk factors attenuated the association between HGF and stroke, but it remained statistically significant (HR, 1.17; 95% CI, 1.03–1.34). This association was modestly stronger if we restricted follow-up time to three years (HR, 1.35; 95% CI, 1.05–1.72), but did not appreciably change in an analysis that treated HGF as a time-varying covariate (HR, 1.14; 95% CI, 1.00–1.31). We tested for multiplicative interactions with age, sex, race/ethnicity, and hypertension status, but none was significant (P >0.4).

Table 2.

Adjusted Hazard Ratios for Incident Stroke Per 1 Standard Deviation Increase in Hepatocyte Growth Factor, Multi-Ethnic Study of Atherosclerosis, 2000–2013

Stroke Model*	Number of Strokes/Total Population	Hazard Ratio (95% Confidence Interval)	P Value
All stroke
Model 1	233/6711	1.30 (1.15–1.47)	<0.0001
Model 2	229/6604	1.17 (1.03–1.34)	0.02
Ischemic stroke
Model 1	183/6711	1.34 (1.17–1.53)	<0.0001
Model 2	179/6604	1.21 (1.04–1.40)	0.01
All ischemic strokes except cardioembolic
Model 1	158/6711	1.33 (1.15–1.54)	0.0001
Model 2	156/6604	1.20 (1.02–1.40)	0.03
Hemorrhagic stroke
Model 1	39/6711	1.23 (0.90–1.68)	0.2
Model 2	39/6604	1.09 (0.78–1.53)	0.6
Other stroke
Model 1	11/6711	0.99 (0.53–1.85)	1.0
Model 2	11/6604	0.85 (0.43–1.65)	0.6

^*Cox regression with outcome incident stroke and predictor 1 standard deviation (259 pg/ml) increase of hepatocyte growth factor.

Model 1 - Adjusted for age (continuous), race/ethnicity (Non-Hispanic white American, Chinese American, African American, Hispanic American), and sex (male, female).

Model 2 - Adjusted for Model 1 plus baseline values of body mass index (continuous), smoking status (current, former, never), diabetes mellitus (yes, no), systolic blood pressure (continuous), antihypertensive medication use (yes, no), high-density lipoprotein cholesterol (continuous), total cholesterol (continuous), low-density lipoprotein cholesterol (continuous), triglycerides (continuous), lipid medication use (yes, no), and study cite (Baltimore; Chicago; Forsyth County, North Carolina; Los Angeles; New York, New York; and St Paul, Minnesota).

A secondary analysis stratified results by stroke type (Table 2). In fully adjusted models, risk of ischemic stroke was 21% higher with each standard deviation increase in HGF, and did not change upon exclusion of cardioembolic strokes, although the number of excluded cases was small. Hemorrhagic strokes and other strokes were not associated with HGF, but their numbers were also small.

Discussion

In this multi-ethnic population-based cohort, we found that participants with higher levels of circulating HGF at baseline had a greater risk of incident stroke, independent of other risk factors. This association was mainly driven by ischemic stroke, and did not change upon exclusion of cardioembolic strokes, although the number of excluded cases was small. We were underpowered to detect all but a strong association in hemorrhagic and other types of stroke. Finally, we report a strong positive relation between circulating HGF and many stroke risk factors.

Our study corroborates findings from previous studies, which reported a positive relation between circulating HGF and ischemic stroke,^{15, 16} and between circulating HGF and prognostic factors for the development of stroke.^{13, 14} However, only one of these studies measured HGF before stroke, which is important to establish temporality. It found, in the predominantly white cohort of postmenopausal women that comprises the Women’s Health Initiative (WHI), that risk of incident ischemic stroke was 40% higher for women in the highest quartile of plasma HGF versus the lowest in fully adjusted models.¹ Our study expands on work from previous studies by using a diverse population-based cohort, employing a prospective design, and reporting the relation of HGF with all stroke (ischemic and hemorrhagic combined) and hemorrhagic stroke.

HGF plays a key role in tissue repair by aiding in the protection and regeneration of vascular endothelial cells.^{1, 20–23} Therefore, it should come as no surprise that both the present study and others have reported a strong positive relation between circulating HGF and stroke risk factors that are associated with endothelial damage and dysfunction, including hypertension, diabetes, smoking, age, and BMI.^2–12 Importantly, HGF and stroke were still related after accounting for stroke risk factors. This observation lends credence to the hypothesis that HGF is not simply a surrogate for stroke risk factors but rather a marker of endothelial damage.²³

Biological mechanisms of an HGF stroke relation may differ by stroke type. To explore this possibility, as a secondary analysis, we excluded cardioembolic stroke, which is typically related to atrial fibrillation, from the definition of ischemic stroke, leaving only ischemic strokes predominantly related to atherosclerotic factors. We found that the relation between HGF and ischemic stroke did not change upon exclusion of cardioembolic strokes. However, the number of excluded cases may have been too small to affect the results. In accordance, the WHI investigators reported that the HGF-stroke association did not vary by subtype of ischemic stroke, however, the data were not shown thus leaving unresolved if they had an adequate number of strokes to produce reliable estimates.¹⁶

Study limitations warrant discussion. The limited number of incident stroke events in this study precluded our ability to evaluate the HGF-stroke relation within subgroups of interest. In particular, future research should explore the HGF-stroke relation by race/ethnicity, as race/ethnicity-specific genetic regulation of HGF levels justifies exploration of heterogeneity of phenotypes.²⁴ Also, although we did not detect an association between HGF and hemorrhagic stroke, we were underpowered with only 39 events to date. Finally, HGF was measured in the total cohort only once at baseline, and stroke events could have occurred more than a decade after this measurement. HGF levels are relatively stable over three years, indicating that a single baseline measurement of circulating HGF is reasonable to use as a marker of an individual’s exposure to this protein over this time period.²⁵ Therefore, as a sensitivity analysis, we restricted follow-up to three years so that strokes were closer to the baseline HGF measurement. The association between HGF and risk of stroke was modestly stronger. As an additional sensitivity analysis, we modeled HGF as a time-varying covariate using baseline and, when available, visit 2 measurements of HGF. The advantage of this approach is that it more accurately represents the long-term level of HGF by accommodating changes over time. The association between HGF and risk of stroke did not appreciably change in this analysis. Thus, the timing and number of HGF measurements seem unlikely to drastically change the relation of HGF and stroke risk, but this issue deserves further attention. Future research could examine the association between serial HGF measurements and stroke risk.

Strengths of this study include the prospective design; the large, population-based multi-ethnic sample with a wide geographic distribution in the United States; robust stroke adjudication; and highly standardized assessment of a broad array of stroke risk factors.

In conclusion, circulating HGF was positively associated with incidence of stroke in a diverse, population-based cohort of men and women from the United States. This association was mainly driven by ischemic stroke, as there were few hemorrhagic stroke events during follow-up. Our findings support the hypothesis that circulating HGF is a marker of endothelial damage and suggest that HGF may have utility as a prognostic marker of stroke risk.