Higher plasma transforming growth factor (TGF)-β is associated with kidney disease in older community dwelling adults

The details of the study design, rationale, and study population of CHS have been described previously [25, 26]. Briefly, CHS is a community-based prospective cohort study designed to explore cardiovascular disease risk factors and stroke in individuals 65 years or older. The original cohort of the CHS consisted of 5201 participants who were randomly enrolled from the Medicare-eligibility lists in 4 US communities (Forsyth County, NC; Sacramento County, CA; Washington County, MD; and Allegheny County, PA) in 1989 to 1990. A supplemental cohort of 687 mostly African-American participants was recruited in 1992–1993.

The eligible individuals were aged ≥65 years, not institutionalized, expected to remain in the current community for ≥3 years, not under active cancer treatment, and able to give written informed consent. Follow-up examinations were done at annual visits through 1998/99 and again in 2006/07. Interviews were also conducted through biannual telephone calls. The study was approved by the institutional review boards of the 4 clinical sites including: University of California, Davis (Sacramento County, Sacramento, CA), Johns Hopkins University (Washington County, Hagerstown, MD), Wake Forest University School of Medicine (Forsyth County, Winston-Salem, NC), University of Pittsburgh (Pittsburgh, PA) [27]. All participants gave written informed consent including the use of de-identified samples and data for future analyses.

TGF-β1 was measured in 2011–2012 on EDTA-stored plasma samples from the 1996/97 visit by ELISA (Quantikine Human TGF-β1 Immunoassay; R&D Systems, Minneapolis, MN). Inter- and intra-assay coefficients of variation were between 1.9 and 2.9% and 6.4% to 9.3%, respectively. Platelets are a major source of TGF-β as a result of platelet degranulation, such that if the plasma is not platelet poor, platelet contamination in plasma samples can artificially increase levels of TGF-β [28]. Presumed platelet contamination of the plasma resulting in elevated levels of TGF-β was found in pilot studies at two of the study sites. Therefore, TGF-β was only measured a priori in 1722 participants at the two remaining sites and our final analysis included these individuals. Of the two sites at which TGF-β was not measured, one site (Washington County, Maryland) did not enroll an African American cohort in 1992–1993, leading to modest differences in other participant characteristics across clinic sites.

Blood samples were collected after an overnight fast and stored at −70 °C using standardized protocols at the 1996/97 visit for the baseline measurements and the 2006/07 for the longitudinal analysis of kidney disease progression. The primary outcome of interest was kidney function estimated by serum cystatin C and albuminuria. Cystatin C was measured using a particle-enhanced immunonephelometric assay with a BN II nephelometer on plasma specimens (N Latex Cystatin C; Dade Behring (now Siemens Health-care Diagnostics Inc.), Deerfield, IL, USA) [29]. The assay range was 0.195–7.330 mg/L, with the reference range for young, healthy individuals reported as 0.53–0.95 mg/L. For cystatin C, intra-assay coefficients of variation (CVs) range from 2.0 to 2.8% and inter-assay CVs range from 2.3 to 3.1%. Glomerular filtration rate (GFR) was calculated using the CKD-EPI cystatin C equation [30]. GFR expressed as mL/min per 1.73 m² of body surface area, and serum cystatin C expressed in mg/L. In categorical analyses, CKD was defined as eGFR <60 mL/min/1.73 m². In addition, in the longitudinal analysis we evaluated change in eGFR from the 1996/97 visit to the 2006/07 visit as an outcome since the measurements of kidney function were repeated in a subset of survivors at this time. Albuminuria was assessed using urinary ACR calculated from spot urinary albumin and creatinine levels. Albuminuria was defined as ACR ≥30 mg/g in categorical analyses.

CHS has previously reported on the association between plasma TGF-β levels and cardiovascular disease, including heart failure, myocardial infarction, and stroke through 2010 [31]. It had similarly reported on the association between TGF-β and mortality in another manuscript [32]. Here, we report on the association between plasma TGF-β levels and cardiovascular events and mortality in longitudinal analysis considering the longer duration of follow-up available since our previous publications (2014) [31, 32]. Participants were followed from the baseline visit (1996/97 for this analysis) until death, loss to follow-up, and the end of follow up on December 31^st, 2014. Cardiovascular events (including heart failure, myocardial infarction, and stroke) and death were adjudicated by the CHS outcome-assessment committee on the basis of patient reports, physician diagnoses, medical records, and medication use [33, 34].

We used covariate measurement from the 1996/97 visit. Age, sex, race, and smoking history variables were collected by self-report. Smoking history was determined by questionnaire and categorized as current, former, or never. Blood pressure (BP) was measured twice in seated position after 5 min rest using standard mercury sphygmomanometer. Hypertension was defined by average seated systolic blood pressure ≥140 mmHg, diastolic blood pressure ≥90 mmHg, or by the current use of antihypertensive medications. We defined diabetes mellitus by use of oral hypoglycemic agents or insulin, or as a fasting glucose level ≥126 mg/dL. Height (m) and weight (kg) were measured to calculate body mass index (BMI: kg/m²). Low density lipoprotein (LDL)-cholesterol and triglycerides were measured during CHS year 5 (1992–1993) visit. Serum triglyceride levels were measured with the Olympus Demand System (Olympus, Lake Success, NY, USA); LDL-cholesterol concentrations were calculated by the Friedewald equation [35]. C-reactive protein (CRP) was measured by an enzyme linked immunosorbent (ELISA) assay. Prevalent cardiovascular disease (CVD) was defined by history of myocardial infarction, angina, or stroke.

Clinical characteristics are presented overall and by quartiles of plasma TGF-β. Continuous data are presented as mean ± standard deviation (SD), and categorical variables as counts (%). Variables with skewed distribution are shown as median (inter quartile range, IQR). Covariates were compared across plasma TGF-β quartiles using a linear trend test, and chi-square test for categorical variables. TGF-β was skewed, so in continuous variable analyses it was log base 2-transformed so that the β estimates can be interpreted as per doubling of TGF-β. Correlations were assessed with Pearson correlation coefficient. To evaluate whether plasma levels of TGF-β are associated with kidney function we first conducted cross-sectional linear regression analysis with plasma TGF-β as the predictor variable and kidney function measures (eGFR and log2 ACR) as the outcome variables. To examine whether TGF-β levels correlated with clinically significant CKD, we conducted logistic regression analysis for TGF-β and estimated GFR <60 mL/min/1.73 m², and subsequently for TGF-β and ACR ≥30 mg/g. Considering the potential bias of eGFR equations in the diagnosis of CKD between eGFR 45–59 mL/min/1.73 m², we also evaluated the association between TGF-β levels and CKD defined as eGFR <45 mL/min/1.73 m² [36]. Several models were applied for multivariate adjustment. M1 adjusted for age, gender, and black race. M2 adjusted for the covariates in M1 as well as BMI, hypertension, DM, smoking status, LDL-cholesterol, triglycerides, previous history of CVD, and log2 CRP. Additionally, we adjusted for estimated GFR together with the covariates included in M2 (when evaluating the association with ACR) or ACR (when evaluating the association with eGFR). To evaluate whether baseline TGF-β predicted kidney disease progression, we conducted a similar analysis longitudinally. Only a subset of survivors (n = 478) had cystatin C measurements at the 2006/07 visit. Considering this small number, we were unable to evaluate whether TGF-β associated with incident CKD and the designated outcome included change in CKD-EPI eGFR (from the 1996/97 visit to the 2006/07 visit). Cox proportional hazard models were utilized to evaluate whether baseline TGF-β levels predicted cardiovascular (CV) events, death, or a composite outcome of CV events and death. The same covariates included in the linear regression models were included in the cox proportional hazard models. For all statistical tests, a two-tailed P-value < 0.05 was considered statistically significant. We did not adjust for multiple testing. Analyses were conducted using R (R Development Core Team (2015)) [37].

TGF-β levels were available for 1722 (39%) individuals out of a total cohort of 4413 at CHS year 9 (1996/97) visit. ACR values were available for 1541 (89.5%) of participants with TGF-β levels. Participants with TGF-β measurements differed from those without available measures in several ways. They were younger, more frequently male and black, and had lower blood pressure. In addition, as shown in Additional file 1: Table S1, they less often had prevalent CVD. Among those with available TGF-β levels, the median (IQR) of TGF-β levels was 3482 (2042–6153) ng/L.

The participant characteristics by TGF-β quartiles at baseline are shown in Table 1. Compared to participants in the lower TGF-β quartiles, those with TGF-β levels in the highest quartile were more frequently black, were more likely to have DM, to have higher CRP, and to be current smokers. Age and previous history of CVD were not significantly different according to TGF-β quartiles. eGFR was significantly lower in quartiles 2, 3, and 4 compared to quartile 1. Albuminuria did not differ significantly among the groups. Of note, median TGF-β was 3482 (IQR 2042–6153) for all the participants. Median TGF-β was higher for those with eGFR <60 ml/min/1.73 m² compared to those with eGFR >60 ml/min/1.73 m² (3747 [IQR 2278–6431] vs. 3377 [IQR 1962–6075] ng/L), with p value = 0.05 (based on Mood’s test of medians).

Linear regression analysis revealed that each doubling of TGF-β was significantly associated with lower eGFR (β estimate = −1.54; 95% C.I. -2.44, −0.63; p < 0.001), although this was marginally non-significant with respect to log2 ACR (β estimate = 0.09; 95% C.I. 0, 0.18; p = 0.055) in unadjusted analysis. As shown in Table 2, the association TGF-β and eGFR was slightly attenuated but remained statistically significant after adjusting for demographics in model 1 (β estimate = −1.41, 95% CI −2.27, −0.55) and including cardiovascular risk factors in model 2 (β = −1.18, 95% CI −2.03, −0.32). Additional adjustment for ACR yielded a β estimate of −1.07 (95% CI −1.95, −0.19). Further inclusion of congestive heart failure and claudication in the model had no influence on the association between TGF-β and eGFR. In contrast, the association between TGF-β and ACR was not significant after adjustment of demographics or in any of the other multivariable models. Since the generalized additive model in relation to eGFR showed that the association tended to plateau above the median value of TGF-β (Fig. 1 log2 TGF-β corresponds to approximately 4096 ng/L) we also evaluated the association for quartiles of the biomarker.

When TGF-β levels were modeled as quartiles, we found that TGF-β levels in quartiles 2, 3, and 4 were significantly associated with lower eGFR relative to the first quartile. In the unadjusted analysis eGFR was lower by 4.55 (95% C.I. -7.14, −1.95), 4.92 (95% C.I. -7.52, −2.32), and 4.47 (95% C.I. -7.07, −1.88) ml/min/1.73 m² for quartiles 2, 3, and 4 respectively. These statistically significant associations were slightly attenuated but remained significant after full adjustment (Table 3). We did not find a significant association between higher TGF-β levels and urine ACR in any unadjusted or adjusted analyses.

To determine whether TGF-β levels associated with clinically significant CKD, we evaluated the binary end point of either eGFR <60 ml/min/1.73 m² or with ACR ≥30 mg/g in logistic regression analysis. As depicted in Fig. 2, TGF-β levels in quartiles 2, 3, and 4 were significantly associated with CKD defined as eGFR <60 ml/min/1.73 m². In unadjusted analysis, compared to the participants in quartile 1 of TGF-β levels, participants in quartile 2 had 53% higher odds of eGFR <60 ml/min/1.73 m² (95% C.I. 1.11, 2.1), participants in quartile 3 had 60% higher odds of eGFR <60 ml/min/1.73 m² (95% C.I. 1.16, 2.2), and participants in quartile 4 had 52% higher odds of eGFR <60 ml/min/1.73 m² (95% C.I. 1.11, 2.10). This remained significant for quartiles 2, 3, and 4 after adjusting for demographics and cardiovascular risk factors with OR of 1.59 (95% C.I. 1.11, 2.28), 1.72 (95% C.I. 1.20, 2.45), and 1.50 (95% C.I. 1.04, 2.15), respectively. Of note, when CKD was defined as eGFR < 45 ml/min/1.73 m² [36], the fully-adjusted ORs for CKD were 3.59 (95% C.I. 1.66, 7.77; p < 0.001), 4.58 (95% C.I. 2.16, 9.68; p < 0.001), and 4.24 (95% C.I. 2.0, 9.02; p < 0.001) for quartiles 2, 3, and 4 compared to quartile 1. There was no consistent evidence of an association between TGF-β quartiles and ACR ≥30 mg/g. Quartile 4 of TGF-β levels was associated with 34% higher odds of having ACR ≥30 mg/g compared to quartile 1 in unadjusted analysis but this was no longer statistically significant after adjusting for demographic and cardiovascular risk factors (OR = 1.34, 95% C.I. 0.89, 2.04, p = 0.17).

To evaluate whether TGF-β levels predicted kidney disease progression, we evaluated whether doubling of TGF-β associated with change in eGFR. We found no association between TGF-β levels at baseline and change in CKD-EPI eGFR between the years 1996/97 and 2006/07 in unadjusted or adjusted models. Only 478 individuals had cystatin C measurements at the 2006/07 visit.

Between the 1996/97 visit and the end of follow-up in 2014, 409 out of the studied 1722 participants died with a median follow up of 10.3 years (IQR 5.7–15.8). Of the surviving 1312 participants, 411 were reported to have a CV event and the median follow up was 9.5 years (IQR 5.1-14.5 years). Similar to the previous report [32], doubling of baseline TGF-β was associated with an increased risk of death. In adjusted analysis, the hazard ratio (HR) for death was 1.10 (95% C.I. 1.02, 1.13, p = 0.008). This was somewhat attenuated to a HR of 1.10 (95% C.I. 1.01, 1.10, p = 0.019) and 1.10 (95% C.I. 1.0, 1.12, p = 0.05) for model 1 and model 2, respectively. Also similar to the previous report [31], we found no association between baseline TGF-β levels and CV events in unadjusted analysis (HR was 1.10, 95% C.I. 0.99, 1.20, p = 0.074). After adjusting for the covariates included in model 2, doubling of TGF-β carried a HR of 1.10 for CV events (95% C.I. 1.0–1.22, p = 0.05). Doubling of baseline TGF-β was associated with higher risk of the composite outcome of CV events or death. The unadjusted HR for the composite outcome 1.10 (95% C.I. 1.02, 1.20, p = 0.005). This association remained significant with HR of 1.10 (95% C.I. 1.03, 1.20, p = 0.004) and 1.10 (95% C.I. 1.02, 1.20, p = 0.006) for model 1 and model 2, respectively.