Original Article
Heritability of blood pressure through latent curve trajectories in families from the Gubbio population study Maria Teresa Bonati a, Francesca Graziano a,b, Maria Cristina Monti a,c, Cristina Crocamo a,c, Oscar Terradura-Vagnarelli d, Massimo Cirillo e, Mariapaola Lanti f, Martino Laurenzi d, Mario Mancini g, Alessandro Menotti f, Mario Grassi b, and Alberto Zanchetti a,h
respectively; FIMLEs, ‘Full Information’ Maximum Likelihood Estimates; H2, heritability; HDL, high-density lipoprotein; I, intercept; LCM, latent curve model; MAR, missing at random; MLE, maximum likelihood estimate; Q, curvature; RMSEA, root-mean-square error of approximation; S, slope; SE, standard error; SOLAR, Sequential Oligogenic Linkage Analysis Routine
Background and objectives: Prospective investigations on cardiovascular risk factors in populations provide a unique opportunity to dissect time-dependent quantitative complex traits, such as arterial blood pressure (BP), into their polygenic and environmental components. BP heritability analyses were carried out on 2620 patients belonging to 711 nuclear pedigrees that could be followed up throughout 25 years in the Gubbio Population Study. Methods: Each patient’s BP serial measurements were summarized into individual intercepts (expected values at baseline) and slopes (time-related changes), which were predicted through latent curve models. These models considered either age in years or waves (times from the first survey) as time axis and were linked at a family level in the heritability analyses using additive polygenic– common environment–unique error models adjusted for sex, age and clinical variables. Results: The additive genetic effect explained 32–49% of the variance of SBP values at baseline, the wavedependent analysis with nuclear pedigrees and the sibshousehold matrix accounting for higher heritability values. Heritability of DBP baseline value was lower than that of SBP in analyses by age (5–15%), but fell in the same heritability range as SBP on the analysis by waves (36– 37%). The BP variation over time (slope) explained by an additive genetic effect ranged from 33 to 43% and from 24 to 25% for SBP and DBP, respectively, in the analysis by age. Shared environment also exerted a significant influence, but explained a smaller portion of the variances (4–17%) for both traits. Conclusion: Longitudinal data from the Gubbio population show strong to moderate genetic influences on SBP and DBP baseline values and changes over time with a smaller, though significant, effect of environment. Keywords: age-dependent analysis, blood pressure heritability, longitudinal phenotypes, nuclear and extended pedigrees, sibs-household effects, time-dependent analysis Abbreviations: ACE, additive polygenic–common environment–unique error; BP, blood pressure; C2, household or sibs-household effect according to the analysis, by using extended or nuclear pedigrees,
INTRODUCTION
A
rterial blood pressure (BP) is a genetically complex trait that shows a continuous distribution. BP is polygenic, resulting from the interplay of a large number of allelic variants each with a small and additive effect, and is influenced by gene–gene interactions (epistasis) and by the environment [1]. Each of the few singlenucleotide polymorphisms that have been associated with BP from meta-analyses of genome-wide association studies explains less than 1 mmHg per allele SBP and 0.5 mmHg per allele DBP. This represents only approximately 1% of the phenotypic variance of the traits [2–4]. A genetic influence on BP is evident, as elevated BP values are more common in relatives of hypertensive individuals than in relatives of normotensive individuals. Van den Elzen et al. [5], in a 27-year follow-up study, have shown that parental BP levels were important predictors of BP development in their offspring, across the whole range of parental BP values.
Journal of Hypertension 2014, 32:2179–2187 a
Istituto Auxologico Italiano, IRCCS, Milan, bDipartimento di Scienze del Sistema Nervoso e del Comportamento, Universita` di Pavia, cDipartimento di Sanita` Pubblica, Medicina Sperimentale e Forense, Universita` di Pavia, dCentro di Medicina Preventiva, Gubbio, eDipartimento di Medicina e Chirurgia, Universita` di Salerno, fAssociation for Cardiac Research, Rome, gDipartimento di Medicina Clinica e Sperimentale, Universita` Federico II, Napoli and hUniversita` di Milano, Centro di Fisiologia Clinica e Ipertensione, Italy Correspondence to Professor Alberto Zanchetti, Istituto Auxologico Italiano, Via L. Ariosto 13, 20145, Milan, Italy. Tel: +39 02 619112237; fax: +39 02 619112901; e-mail: [email protected] Received 3 December 2013 Revised 18 June 2014 Accepted 18 June 2014 J Hypertens 32:2179–2187 ß 2014 Wolters Kluwer Health | Lippincott Williams & Wilkins. DOI:10.1097/HJH.0000000000000311
Journal of Hypertension
www.jhypertension.com
2179
Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Bonati et al.
Heritability, that is, the ratio of the variance of the trait caused by genes to the total phenotypic variance, is usually estimated from family or twin studies, which may be cross-sectional or longitudinal. Typically, results from family studies have resulted in lower estimates than those from twin studies [6]. Cross-sectional family studies rely on a single observation of the traits of interest, whereas longitudinal family studies on serial observations over time. BP increases with age at different rates depending on the initial SBP or DBP [7] and with a tracking from childhood to adulthood [8–11]. Heritability of BP traits has been shown to be age-dependent too [6,12]. Therefore, genes may affect not only the levels, but also the longitudinal trends of BP changes [13,14], and the two types of heritability may be linked to independent pools of genes. The largest numbers of studies dealing with BP heritability estimates are based on cross-sectional family studies. Through these studies, adjusted BP heritability estimates have been reported to range from ratios of 0.25 to 0.68 for SBP and from 0.26 to 0.65 for DBP (see [15–22] as a representative selection). Only a few studies deal with BP longitudinal heritability analysis [12,23–34]; the family dataset of these studies is mainly derived from the Framingham Heart Study. The Gubbio Study collected information on BP and other cardiovascular risk factors over a period of about 25 years in a rather stable population, which makes the data particularly suitable to heritability studies.
METHODS Study population and variables The Gubbio Population Study was a prospective epidemiological investigation on BP and cardiovascular risk factors started in 1983 and concluded in 2007 in Gubbio, a town in central Italy. Three surveys (1983–1985, 1988–1992, and 2001–2007) were conducted over about 25 years, as detailed in [35–37]. The surveys targeted the patients aged 5 years and over, living within the mediaeval walls and their close relatives living outside. Among the 6831 participants at the examinations, 51.78% were residents within the mediaeval city [36]. Information on demographic, clinical, anthropometric and environmental variables was collected. Among these variables, DBP and SBP were measured by trained personnel, following the specific WHO Cardiovascular Survey Methods Manual [38], from now on quoted as WHO manual. At each survey, SBP and DBP were measured three times with a manual mercury sphygmomanometer using a cuff of a size adequate to the arm circumference with the patient in a sitting position after 5 min of rest [38]. The average of the second and third measurements was used as the quantitative trait of interest. In individuals on antihypertensive treatment, SBP and DBP were adjusted by adding 10 and 5 mmHg, respectively, as done in previous genetic studies [39,40]. Age was recorded in years; BMI in kg/m2 was calculated from height and weight; serum total cholesterol and high-density lipoprotein (HDL) cholesterol were measured using automated enzymatic methods; information on smoking, drinking habits and physical activity was collected through a questionnaire. 2180
www.jhypertension.com
Drawing and selection of pedigrees At each survey, genealogical information was registered and updated through a structured interview administered to each participant. From these data, nuclear and extended pedigrees were drawn. Nuclear pedigrees are two-generation families with first-degree relationships, that is, parent–offspring and/or sibling. Drawing of extended pedigrees was carried out to include: three generations when applicable, the nuclear family of spouses in the second generation and, consequently, all first cousins – maternal and paternal – in the third one. Pedigrees were included in the analysis only if there were at least two related individuals participating in all three surveys. In order to estimate the sibs-household contribution to phenotypic variance and for sensitivity analysis purposes, the extended families were split into nuclear subpedigrees.
Heritability of longitudinal blood pressure phenotypes Since the traditional variance components model does not accommodate longitudinal phenotypes, a two-step approach has been adopted. In the first step, each patient’s summary statistics were calculated considering two metrics of time, that is, age or waves, as detailed hereafter. 1. Time axis ¼ age: Life span was divided into nine decades, the first decade includes ages 0–9 years old, whereas the ninth decade includes ages 80–89 years old. 2. Time axis ¼ wave: Five time lines, corresponding to waves of measurement, each 5 years long, were considered in the data matrix; the first wave is called 0 and includes BP measurements from 1983 to 1987, whereas the fifth wave includes BP values collected from 2003 to 2007. As shown in Fig. 1, the three surveys of the Gubbio Population Study [35,36] were therefore associated with the waves of measurements. Wave 0 and 1 corresponded to survey 1 and 2. Wave 2, corresponding to the 5-year interval between the end of the second survey and the beginning of the third one, was added as phantom wave because no BP values were measured, whereas BP values of the third survey were split between waves 3 and 4. SBP and DBP measurements for each patient were aggregated over time into the random coefficients (intercept, slope and curvature) by applying a latent curve model (LCM) on equal time points, as the change process is supposedly regular [41], by using the Mplus software v. 7.11 [42]. Linear or quadratic growth trajectories, considering values not available over time as missing at random (MAR) [43], were fitted by case-wise (‘Full Information’) Maximum Likelihood Estimates (FIMLEs) [44]. The linear model implies constant change in BP across equally spaced time assessments. The quadratic model implies differential change in BP across equally spaced time assessments, the amount of BP change depending on where, on the curve, change is considered. Volume 32 Number 11 November 2014
Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Heritability of longitudinal blood pressure traits
SURVEY 1 (1983–1985)
WAVE 0 (1983–1987)
SURVEY 2 (1988–1992)
WAVE 1 (1988–1992)
2215*
2123*
WAVE 2 (1993–1997) WAVE 3 (1998–2002) SURVEY 3 (2001–2007)
WAVE 4 (2003–2007)
1105*
830*
FIGURE 1 The effective number of measured individuals, whose descriptive features are detailed in Table 1, at each wave/survey is reported. The main aim of the drawing is to show how individuals belonging to the informative pedigrees were measured at the two different times of survey 3. Wave 2, inserted as phantom wave, was not imputed and the missing values of the 1105 individuals measured at wave 3 were imputed by latent curve model at wave 4. Observed subjects in the informative pedigrees.
RMSEA below 0.05 and average R2 above 0.50, indicating adequate model fitting. The P values of the LCM fixed parameter estimates (i.e. means, variances and covariances of the individual I, S and Q coefficients) were signed by t test (MLE/SE); the significance level was set at P less than
Heritability of blood pressure through latent curve trajectories in families from the Gubbio population study Maria Teresa Bonati a, Francesca Graziano a,b, Maria Cristina Monti a,c, Cristina Crocamo a,c, Oscar Terradura-Vagnarelli d, Massimo Cirillo e, Mariapaola Lanti f, Martino Laurenzi d, Mario Mancini g, Alessandro Menotti f, Mario Grassi b, and Alberto Zanchetti a,h
respectively; FIMLEs, ‘Full Information’ Maximum Likelihood Estimates; H2, heritability; HDL, high-density lipoprotein; I, intercept; LCM, latent curve model; MAR, missing at random; MLE, maximum likelihood estimate; Q, curvature; RMSEA, root-mean-square error of approximation; S, slope; SE, standard error; SOLAR, Sequential Oligogenic Linkage Analysis Routine
Background and objectives: Prospective investigations on cardiovascular risk factors in populations provide a unique opportunity to dissect time-dependent quantitative complex traits, such as arterial blood pressure (BP), into their polygenic and environmental components. BP heritability analyses were carried out on 2620 patients belonging to 711 nuclear pedigrees that could be followed up throughout 25 years in the Gubbio Population Study. Methods: Each patient’s BP serial measurements were summarized into individual intercepts (expected values at baseline) and slopes (time-related changes), which were predicted through latent curve models. These models considered either age in years or waves (times from the first survey) as time axis and were linked at a family level in the heritability analyses using additive polygenic– common environment–unique error models adjusted for sex, age and clinical variables. Results: The additive genetic effect explained 32–49% of the variance of SBP values at baseline, the wavedependent analysis with nuclear pedigrees and the sibshousehold matrix accounting for higher heritability values. Heritability of DBP baseline value was lower than that of SBP in analyses by age (5–15%), but fell in the same heritability range as SBP on the analysis by waves (36– 37%). The BP variation over time (slope) explained by an additive genetic effect ranged from 33 to 43% and from 24 to 25% for SBP and DBP, respectively, in the analysis by age. Shared environment also exerted a significant influence, but explained a smaller portion of the variances (4–17%) for both traits. Conclusion: Longitudinal data from the Gubbio population show strong to moderate genetic influences on SBP and DBP baseline values and changes over time with a smaller, though significant, effect of environment. Keywords: age-dependent analysis, blood pressure heritability, longitudinal phenotypes, nuclear and extended pedigrees, sibs-household effects, time-dependent analysis Abbreviations: ACE, additive polygenic–common environment–unique error; BP, blood pressure; C2, household or sibs-household effect according to the analysis, by using extended or nuclear pedigrees,
INTRODUCTION
A
rterial blood pressure (BP) is a genetically complex trait that shows a continuous distribution. BP is polygenic, resulting from the interplay of a large number of allelic variants each with a small and additive effect, and is influenced by gene–gene interactions (epistasis) and by the environment [1]. Each of the few singlenucleotide polymorphisms that have been associated with BP from meta-analyses of genome-wide association studies explains less than 1 mmHg per allele SBP and 0.5 mmHg per allele DBP. This represents only approximately 1% of the phenotypic variance of the traits [2–4]. A genetic influence on BP is evident, as elevated BP values are more common in relatives of hypertensive individuals than in relatives of normotensive individuals. Van den Elzen et al. [5], in a 27-year follow-up study, have shown that parental BP levels were important predictors of BP development in their offspring, across the whole range of parental BP values.
Journal of Hypertension 2014, 32:2179–2187 a
Istituto Auxologico Italiano, IRCCS, Milan, bDipartimento di Scienze del Sistema Nervoso e del Comportamento, Universita` di Pavia, cDipartimento di Sanita` Pubblica, Medicina Sperimentale e Forense, Universita` di Pavia, dCentro di Medicina Preventiva, Gubbio, eDipartimento di Medicina e Chirurgia, Universita` di Salerno, fAssociation for Cardiac Research, Rome, gDipartimento di Medicina Clinica e Sperimentale, Universita` Federico II, Napoli and hUniversita` di Milano, Centro di Fisiologia Clinica e Ipertensione, Italy Correspondence to Professor Alberto Zanchetti, Istituto Auxologico Italiano, Via L. Ariosto 13, 20145, Milan, Italy. Tel: +39 02 619112237; fax: +39 02 619112901; e-mail: [email protected] Received 3 December 2013 Revised 18 June 2014 Accepted 18 June 2014 J Hypertens 32:2179–2187 ß 2014 Wolters Kluwer Health | Lippincott Williams & Wilkins. DOI:10.1097/HJH.0000000000000311
Journal of Hypertension
www.jhypertension.com
2179
Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Bonati et al.
Heritability, that is, the ratio of the variance of the trait caused by genes to the total phenotypic variance, is usually estimated from family or twin studies, which may be cross-sectional or longitudinal. Typically, results from family studies have resulted in lower estimates than those from twin studies [6]. Cross-sectional family studies rely on a single observation of the traits of interest, whereas longitudinal family studies on serial observations over time. BP increases with age at different rates depending on the initial SBP or DBP [7] and with a tracking from childhood to adulthood [8–11]. Heritability of BP traits has been shown to be age-dependent too [6,12]. Therefore, genes may affect not only the levels, but also the longitudinal trends of BP changes [13,14], and the two types of heritability may be linked to independent pools of genes. The largest numbers of studies dealing with BP heritability estimates are based on cross-sectional family studies. Through these studies, adjusted BP heritability estimates have been reported to range from ratios of 0.25 to 0.68 for SBP and from 0.26 to 0.65 for DBP (see [15–22] as a representative selection). Only a few studies deal with BP longitudinal heritability analysis [12,23–34]; the family dataset of these studies is mainly derived from the Framingham Heart Study. The Gubbio Study collected information on BP and other cardiovascular risk factors over a period of about 25 years in a rather stable population, which makes the data particularly suitable to heritability studies.
METHODS Study population and variables The Gubbio Population Study was a prospective epidemiological investigation on BP and cardiovascular risk factors started in 1983 and concluded in 2007 in Gubbio, a town in central Italy. Three surveys (1983–1985, 1988–1992, and 2001–2007) were conducted over about 25 years, as detailed in [35–37]. The surveys targeted the patients aged 5 years and over, living within the mediaeval walls and their close relatives living outside. Among the 6831 participants at the examinations, 51.78% were residents within the mediaeval city [36]. Information on demographic, clinical, anthropometric and environmental variables was collected. Among these variables, DBP and SBP were measured by trained personnel, following the specific WHO Cardiovascular Survey Methods Manual [38], from now on quoted as WHO manual. At each survey, SBP and DBP were measured three times with a manual mercury sphygmomanometer using a cuff of a size adequate to the arm circumference with the patient in a sitting position after 5 min of rest [38]. The average of the second and third measurements was used as the quantitative trait of interest. In individuals on antihypertensive treatment, SBP and DBP were adjusted by adding 10 and 5 mmHg, respectively, as done in previous genetic studies [39,40]. Age was recorded in years; BMI in kg/m2 was calculated from height and weight; serum total cholesterol and high-density lipoprotein (HDL) cholesterol were measured using automated enzymatic methods; information on smoking, drinking habits and physical activity was collected through a questionnaire. 2180
www.jhypertension.com
Drawing and selection of pedigrees At each survey, genealogical information was registered and updated through a structured interview administered to each participant. From these data, nuclear and extended pedigrees were drawn. Nuclear pedigrees are two-generation families with first-degree relationships, that is, parent–offspring and/or sibling. Drawing of extended pedigrees was carried out to include: three generations when applicable, the nuclear family of spouses in the second generation and, consequently, all first cousins – maternal and paternal – in the third one. Pedigrees were included in the analysis only if there were at least two related individuals participating in all three surveys. In order to estimate the sibs-household contribution to phenotypic variance and for sensitivity analysis purposes, the extended families were split into nuclear subpedigrees.
Heritability of longitudinal blood pressure phenotypes Since the traditional variance components model does not accommodate longitudinal phenotypes, a two-step approach has been adopted. In the first step, each patient’s summary statistics were calculated considering two metrics of time, that is, age or waves, as detailed hereafter. 1. Time axis ¼ age: Life span was divided into nine decades, the first decade includes ages 0–9 years old, whereas the ninth decade includes ages 80–89 years old. 2. Time axis ¼ wave: Five time lines, corresponding to waves of measurement, each 5 years long, were considered in the data matrix; the first wave is called 0 and includes BP measurements from 1983 to 1987, whereas the fifth wave includes BP values collected from 2003 to 2007. As shown in Fig. 1, the three surveys of the Gubbio Population Study [35,36] were therefore associated with the waves of measurements. Wave 0 and 1 corresponded to survey 1 and 2. Wave 2, corresponding to the 5-year interval between the end of the second survey and the beginning of the third one, was added as phantom wave because no BP values were measured, whereas BP values of the third survey were split between waves 3 and 4. SBP and DBP measurements for each patient were aggregated over time into the random coefficients (intercept, slope and curvature) by applying a latent curve model (LCM) on equal time points, as the change process is supposedly regular [41], by using the Mplus software v. 7.11 [42]. Linear or quadratic growth trajectories, considering values not available over time as missing at random (MAR) [43], were fitted by case-wise (‘Full Information’) Maximum Likelihood Estimates (FIMLEs) [44]. The linear model implies constant change in BP across equally spaced time assessments. The quadratic model implies differential change in BP across equally spaced time assessments, the amount of BP change depending on where, on the curve, change is considered. Volume 32 Number 11 November 2014
Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.
Heritability of longitudinal blood pressure traits
SURVEY 1 (1983–1985)
WAVE 0 (1983–1987)
SURVEY 2 (1988–1992)
WAVE 1 (1988–1992)
2215*
2123*
WAVE 2 (1993–1997) WAVE 3 (1998–2002) SURVEY 3 (2001–2007)
WAVE 4 (2003–2007)
1105*
830*
FIGURE 1 The effective number of measured individuals, whose descriptive features are detailed in Table 1, at each wave/survey is reported. The main aim of the drawing is to show how individuals belonging to the informative pedigrees were measured at the two different times of survey 3. Wave 2, inserted as phantom wave, was not imputed and the missing values of the 1105 individuals measured at wave 3 were imputed by latent curve model at wave 4. Observed subjects in the informative pedigrees.
RMSEA below 0.05 and average R2 above 0.50, indicating adequate model fitting. The P values of the LCM fixed parameter estimates (i.e. means, variances and covariances of the individual I, S and Q coefficients) were signed by t test (MLE/SE); the significance level was set at P less than
