Abstract
Considerable uncertainty exists regarding the genetic architecture underlying common lateonset human diseases. In particular, the contribution of deleterious recessive alleles has been predicted to be greater for lateonset than for earlyonset traits. We have investigated the contribution of recessive alleles to human hypertension by examining the effects of inbreeding on blood pressure (BP) as a quantitative trait in 2760 adult individuals from 25 villages within Croatian island isolates. We found a strong linear relationship between the inbreeding coefficient (F) and both systolic and diastolic BP, indicating that recessive or partially recessive quantitative trait locus (QTL) alleles account for 1015% of the total variation in BP in this population. An increase in F of 0.01 corresponded to an increase of ∼3 mm Hg in systolic and 2 mm Hg in diastolic BP. Regression of F on BP indicated that at least several hundred (300600) recessive QTL contribute to BP variability. A model of the distribution of locus effects suggests that the 816 QTL of largest effect together account for a maximum of 25% of the dominance variation, while the remaining 75% of the variation is mediated by QTL of very small effect, unlikely to be detectable using current technologies and sample sizes. We infer that recent inbreeding accounts for 36% of all hypertension in this population. The global impact of inbreeding on hypertension may be substantial since, although inbreeding is declining in Western societies, an estimated 1 billion people globally show rates of consanguineous marriages >20%.
THE extensive literature on the health effects of inbreeding has largely focused on its impact on reproduction, childhood mortality, and Mendelian disorders (Bittleset al. 1991; Bittles and Neel 1994). Remarkably little has been published on the effects of inbreeding on genetically complex lateonset disorders that account for most of the public health burden of disease. This is despite the observation in other species that the deleterious effects of inbreeding may increase with age, suggesting greater sensitivity of homeostatic mechanisms to inbreeding in later life (Charlesworth and Hughes 1996; Charlesworth and Charlesworth 1999).
We postulated that the quantitative trait, blood pressure (BP), and the related lateonset disorder, essential hypertension, might be mediated by recessive and partially recessive quantitative trait locus (QTL) alleles, which would be influenced by the increased homozygosity found in inbred individuals. In support of this hypothesis, several studies of small inbred communities worldwide have reported an increased prevalence of hypertension (Krieger 1968; Martinet al. 1973; Hurwichet al. 1982; Thomaset al. 1987; Wahid Saeedet al. 1996; Halberstein 1999). In addition, analogous observations have come from experiments in inbred (“spontaneous”) and engineered animal models of hypertension (Stollet al. 2000). To investigate the relationship between inbreeding and BP we studied a large population sample from wellcharacterized genetic isolates from the Dalmatian islands, Croatia (Figure 1).
SUBJECTS AND METHODS
Study population: The village populations of three neighboring islands in the eastern Adriatic, Middle Dalmatia, Croatia (Brac, Hvar, and Korcula—see Figure 1) represent wellcharacterized genetic isolates. Over 100 publications describe the ethnohistory, migration patterns, genealogical reconstruction, biological trait measurements, disease prevalence, and environmental and sociocultural characteristics of this population (Rudan et al. 1987, 1992, 1999; Waddleet al. 1998). Population genetic characteristics of the study population based on a number of serogenetic polymorphisms were reported by Roberts et al. (1992) and Janicijevic et al. (1994). Subsequent analyses of variable number of tandem repeat and short tandem repeat DNA polymorphisms and mtDNA characterized genetic variation in specific islands (Martinovic et al. 1998, 1999; Klaric et al. 2001a,b; Tolket al. 2001). The results indicated that village populations in these islands have preserved separate characteristics over the course of history to the present day. Measures of genetic kinship and genetic distances revealed isolation of individual villages or village groups from each other and from the mainland. Specific village clustering was noted on Brac and Hvar islands, which coincided with known historic processes. An appreciable degree of genetic homogeneity within the studied villages has been noted, which is especially true for the most geographically isolated villages. The 25 villages chosen for this study were founded during one of three periods: the BC era (by admixture of Illyrians, Greeks, and succeeding Romans), the 7th century AD (by Croats who immigrated from Asia), and the 16th to 18th centuries AD (by Croats who left the Balkan peninsula fearing Ottoman expansion). The subsequent tendency toward inbreeding in each village has been influenced by geographic isolation, political (“Pastrovic”) privileges given to residents of certain communities, and by sociocultural factors (Rudanet al. 1992). These island populations present a range of inbreeding patterns at both individual and subpopulation levels, as documented in previous studies reporting endogamy, isonymy, mating choice, genealogical information, and genetic marker distributions. High inbreeding levels have been implicated in at least three Mendelian disorders characterized in neighboring island populations: Mal de Meleda in Mljet (Fischeret al. 2001), hereditary dwarfism in Krk (Kopajticet al. 1995), and hereditary mental retardation in Susak (Bohacek 1964). Measures of diet and lifestyle factors show restricted variation in this population, suggesting its suitability for genetic studies of hypertension (Rudanet al. 1992).
Blood pressure and other measurements: We measured blood pressure, height, and weight between 1979 and 1981 in 2760 adult individuals selected at random from voting lists from 25 isolate villages on three islands (Brac, Hvar, and Korcula) in middle Dalmatia, Croatia (representing a 20% sample of the village populations). In addition, we collected data on body mass index, diet, education level, occupation, and smoking status. This was carried out with the informed consent of participants by the Institute for Anthropological Research in Zagreb, Croatia, in collaboration with the Smithsonian Institute in Washington, DC. None of the examinees had ever received antihypertensive treatment. Blood pressure was measured by a single observer in local health centers and dispensaries between 6 am and 12 noon following standard procedures as described by Weiner (Weiner and Lourie 1969). BP values were adjusted for the major determinants of BP (age, height, weight, and smoking status in the analyses) and were reported separately in males and females. Hypertension was defined as systolic BP ≥160 or diastolic BP ≥95 mm Hg.
Computation of individual inbreeding coefficients: A single researcher (I. Rudan) computed individual inbreeding coefficients independently and blind to BP status for each study participant on the basis of pedigree information on four ancestral generations (five generations where these occurred over a similar time frame) recorded during the initial field work and supplemented by study of parish registries stored in local churches during 19972000. The individual inbreeding coefficients (F) were then computed according to Wright’s path method,
Statistical analysis and modeling: Comparisons of BP among villages were based on systolic and diastolic BP measurements adjusted for age, body mass index (weight/height^{2}), and smoking status. A stepdown multiple regression analysis was performed using MINITAB 12.21 software to investigate the correlation between individual BP measurements and inbreeding coefficients. The model explored the relationship between systolic and diastolic blood pressure (as dependent variables) and a number of explanatory variables: individual inbreeding coefficient (F), island and village of residence, smoking status, and the major known risk factors for hypertension—age, sex, (logtransformed) height, and (logtransformed) weight. Variables that made the least contribution to the explained variation were dropped one at a time until all the remaining variables were statistically significant (defined as P < 0.05 for main effects and P < 0.01 for higherorder effects; Table 2). A model was developed from quantitative genetic theory to derive a lower bound, n_{L}, for the number of genetic loci of equivalent effect contributing to the dominance variance in BP, as
Populationattributable fraction: The populationattributable fraction (PAF) for hypertension (defined as either systolic >160 mm Hg or diastolic >90 mm Hg) was calculated by multiple logistic regression allowing for individual differences in the variables: village, sex, age, height, weight, and smoking. We determined the appropriate regression as a function of all associated variables (including F) and then noted each individual’s probability of being hypertensive if their F was set equal to 0. The sum of all such probabilities, P_{sum}, is an estimate of the number affected in the absence of inbreeding, but with other variables remaining unaltered. Then PAF = 1  P_{sum}/N_{aff}, where N_{aff} is the actual number affected.
Modeling the effects of individual QTL loci: For biallelic loci, the relation between the true number, n say, of recessive QTL loci affecting a trait and n_{L} (see above) is n = n_{L}(1 +γ^{2}), where γ denotes the coefficient of variation of the frequency distribution of locus effects. Following Zeng (1992), we modeled this as gamma with parameter L ≤ 1; i.e., f(x) = x^{L}^{1}e^{}^{x}/Γ(L). This family of distributions has γ^{2} = L^{1}. Since the contribution of a biallelic locus with nonadditive effect, x (or D_{j} in the notation of the appendix), to the dominance variance is just x^{2}, the distribution of such contributions is also gamma, but with parameter L + 2. Hence, for given L, we can compute the minimum proportion of loci contributing any specified proportion of the overall variance. Finally for given n_{L}, we obtain an estimate of the actual minimum number of loci by multiplying this proportion by n_{L} (1 + L^{1}). As shown by Zeng (1992) this number is relatively insensitive to L in the range
RESULTS
Measurements recorded during a survey in 19791981 in an untreated population permitted analysis of BP as a quantitative trait. Body mass index, diet, education level, occupation, smoking status, and inbreeding values among study participants are shown in Table 1 by village of residence. The prevalence of hypertension among individuals with no known inbreeding in their recent ancestry in the study population was ∼20%, and the mean ages of those males and females were, respectively, 45.9 (SD 13.9) and 47.0 (SD 13.9) years. Average inbreeding measures for each of the 25 villages based on Wright’s path method and isonymy gave a consistent pattern of ranking of villages by level of inbreeding. This supports the use of F values as a means of ranking individuals and villages by inbreeding coefficient (Table 1).
We found a highly significant linear correlation between mean inbreeding coefficient of study individuals in each village and the prevalence of hypertension (Figure 2). To explore this further, we performed multiple regression analysis of systolic and diastolic BP on individual inbreeding coefficients (F), controlling for the main recognized determinants of BP (age, sex, height, and weight), village of residence, and smoking status. We found a strong linear correlation between F and adjusted systolic and diastolic BP in both males and females (Figure 3). Both systolic and diastolic BP levels correlated positively with age, weight, and individual inbreeding coefficients and negatively with height and smoking status in both males and females. The regression model explained 3550% of the phenotypic variance in BP. The strongest effect was clearly individual inbreeding coefficients, which alone explained ∼15% of the variance in males and 10% in females in both systolic and diastolic levels (Table 2). An increase in F of 0.01 corresponded to an increase of ∼3 mm Hg in systolic and 2 mm Hg in diastolic BP in both sexes.
The effect of inbreeding (F) on BP depends on the number and dominance properties of QTL alleles, their frequencies, and average effects on the trait (Mukaiet al. 1974; Falconer and Mackay 1996). Using result (1) and taking the total phenotypic variance as an upper limit for the value of V_{D}, we found the genetic component of blood pressure variability in this population to be influenced by not less than several hundred recessive QTL, with 405 and 306 loci for systolic BP and 615 and 375 loci for diastolic BP in males and females, respectively.
The distribution of recessive QTL effects can be approximated as gammatype with mode at zero and parameter L < 1 (Zeng 1992). From our data, if the estimated minimum QTL number is ∼400, and L is between ^{1}/_{16} and 1, then the minimum numbers contributing the upper 25th and 50th percentiles of the distribution are, respectively, 816 and 3055 (Figure 4).
Height was analyzed in a similar fashion since in many populations it shows additive variance but no major dominance component (Krieger 1968; Tambset al. 1992). The results showed that the slope of the regression of F on height did not differ significantly from zero, as predicted (Figure 3), supporting our interpretation of the data.
DISCUSSION
It is widely recognized that essential hypertension is under considerable genetic influence. However, apart from isolated successes in mapping rare monogenic loci, which account for <5% of hypertension, no major progress has been made in defining the genetic basis of essential hypertension (Liftonet al. 2001). A common, often implicit, assumption in mapping studies of such complex traits is that relatively few genetic loci of moderate to large effect account for a large component of the underlying genetic variance despite the paucity of empirical data to support this. We have demonstrated that the effects of recessive QTL on BP are widespread, accounting for 1015% of the total variation in BP in this population. These effects are attributable to a very large number of loci (at least 300600), which will almost certainly show a range of effects on BP.
The model makes several assumptions that may influence these estimates. First, the inbreeding coefficient is based on measures of recent inbreeding (over four to five generations). We therefore calculated isonymy estimates for each village (Table 1) and found that their mean value exceeded the median F value by a factor of 1.35. Since isonymy is widely recognized to overestimate inbreeding (Tay and Yip 1984), this represents an upper bound to the inbreeding estimate. Inflating the F values by 135% in the model reduces the above estimates of minimum QTL numbers by a factor of 1.35^{2} (=1.83). On the other hand, if, as seems likely, V_{D}/V_{P} is nearer 33% than the 100% assumed here, the effect would be to triple the estimates. CavalliSforza and Bodmer (1971), for example, estimated V_{D}/V_{P} to be 0.38 and 0.33 for systolic and diastolic blood pressure, respectively, using the data of Miall and Oldham (1963). Second, as in many genetic models, all loci were assumed to have equal effects, whereas both theory (Brink 1967) and empirical data in animals (Mackay 2001) show that the QTL effects vary widely and may even be of opposite sign. Ignoring this again results in underestimating the true number of QTL. Third, if a substantial proportion of the phenotypic variance is due to epistatic effects, additive and dominance variances may be upwardly biased and lead to an underestimation of the number of QTL. However, many studies suggest that epistatic QTL effects are uncommon (Mackay 2001). Finally, we assume that recombination between adjacent loci is sufficiently frequent that the identitybydescent (IBD) status of any locus can be considered independent of its neighbors. In effect, the method treats tightly linked loci as a single “superlocus” (Flint and Mott 2001), leading to further underestimation of the true number of loci.
The magnitude of the inbreeding effect on BP is large (equivalent to a rise in systolic BP of ∼20 mm Hg and diastolic of ∼12 mm Hg in offspring of firstcousin marriages; F = 0.0625) but very similar to the only other two published estimates we could identify in other isolate populations. Krieger (1968) found a 35 mm Hg increase in diastolic BP associated with a 0.1 increase in F in a study of 3465 children in Brazil and Martin et al. (1973) reported a 728 mm Hg increase in systolic BP in adult Hutterites associated with an increase in F of 0.0625. This may be because inbreeding has a greater influence on lateonset traits than on traits that are subject to early selection (Charlesworth and Hughes 1996). It is also possible that low environmental variation, or underestimation of F due to individuals being related through multiple lines of descent, contributes to the size of inbreeding effect in these isolate populations (Krieger 1968; Martinet al. 1973; Halberstein 1999; Abneyet al. 2001). Thus the observed effect size may be less in more environmentally diverse or outbred populations. The unidirectionality of the effect is also striking and consistent with a linear unidirectional effect seen in an SLeut isolate population (Martinet al. 1973), but the mechanism is unclear. A change in BP with inbreeding is predicted as a consequence of recessive or partially recessive variants with the direction of change toward the value of the more recessive alleles. Physiological homeostasis may also act to support a directional change in BP, for example, through selection against variants tending to reduce BP to maintain circulatory viability. Directional dominance may also occur in lateonset traits when environmental factors are directional (e.g., increase in adult blood pressure due to dietary salt) and when selective constraints are weak compared with blood pressure maintenance in early life.
The estimate of several hundred recessive QTL relevant to human hypertension is realistic and indeed may be conservatively low. It is consistent with a complex and genetically highly variable (Halushkaet al. 1999) system of blood pressure control mediated by cardiac output, blood vessel architecture, renal function, and central nervous system integration and requiring the interaction of homeostatic systems, including baroreceptors, natriuretic peptides, reninangiotensinaldosterone, kininkallikrein, adrenergic receptors, and local vasodilator mechanisms (Liftonet al. 2001). Furthermore, published work from animal models of hypertension supports a polygenic rather than oligogenic basis for hypertension (Liftonet al. 2001) and yet these models probably underestimate the genetic complexity, since they are typically bred to achieve fixation of a small subset of the diversity found in wild populations (Flint and Mott 2001). The greater genetic complexity of a diverse and outbred human population would seem to be selfevident, despite the fact that humans show less haplotype and polymorphic diversity than several other species, including other primates (Reichet al. 2001).
Our minimum estimates of the number of recessive QTL influencing blood pressure control do not in themselves reveal the relative magnitudes of locus effects. There is, however, good evidence for an Lshaped (leptokurtotic) distribution of alleliceffect sizes (Shrimpton and Robertson 1988; Tanksley 1993; Bostet al. 2001; Hayes and Goddard 2001; Mackay 2001; Barton and Keightley 2002). In addition, as shown by Zeng (1992), their distribution can be approximated as gammatype with mode at zero (i.e., with parameter L < 1), implying that most loci contribute little to the overall genetic variation and that the number contributing a large proportion is both small and relatively insensitive to L. The model developed from our data predicts the minimum QTL numbers contributing the upper 25th and 50th percentiles of the distribution are, respectively, 816 and 3055 (Figure 4). Thus, the QTL with the largest effect account individually for a small proportion of the total dominance variation and 5075% of the variation is mediated by many QTL of very small effect, which are probably undetectable using current methods (Terwilliger and Goring 2000).
This study demonstrates an important effect of inbreeding on the genetically complex lateonset disorder, hypertension, which appears to be mediated by a large number of recessive QTL alleles as a result of increased homozygosity. Several factors support the validity of the data and reinforce the conclusions: first, the standard measurement procedures that were adopted and the exclusion of known confounding factors; second, the consistency of findings in diverse populations (Krieger 1968; Martinet al. 1973; Halberstein 1999); third, the linear increase in BP with increasing F (prevalence of hypertension rises by 10% for every increment in F of 0.01 up to F = 0.06); fourth, the overall strength of the effect; fifth, the existence of biologically plausible mechanisms, all of which point to a causal relationship between inbreeding and hypertension. Moreover, the consistency of the observation in a random sample of individuals across 25 villages is not explicable by a kinship effect. In terms of health impact, the results show that 36% of hypertension incidence in this population can be attributed to inbreeding (populationattributable fraction). The population prevalence of hypertension among individuals with no known inbreeding in their recent ancestry is ∼20%, similar to most outbred populations, but it increases steeply among 50yearolds as the inbreeding coefficient rises (Figure 5).
Inbreeding is generally decreasing among nonimmigrant Western societies but it is highly prevalent globally. Consanguineous marriages, defined as a union between individuals related as second cousins or closer (equivalent to F ≥ 0.0156 in their progeny), has been conservatively estimated to occur at 110% prevalence among 2.811 billion and at 2050% prevalence among 911 million people globally (Bittles 1988; Bittleset al. 2001). In addition, the extent of homozygosity by descent in outbred populations may have been underestimated (Broman and Weber 1999). The global impact of inbreeding on hypertension (and stroke) could therefore be significant in health economic terms. In addition, the results provide new insights into the genetic architecture of a common disorder, which should inform and improve the design of QTLmapping studies and explain some of the observed differences in trait distributions among different populations.
APPENDIX: THE EFFECT OF INBREEDING ON A MULTILOCUS PHENOTYPE
Model, notation, and assumptions: The phenotype of the ith individual is modeled by
The effect of inbreeding: Assuming HardyWeinberg equilibrium (HWE),
The components of genetic variance: Table A1 shows the steps needed to compute the additive (V_{A}_{j}) and dominance (V_{D}_{j}) components of total genetic variance (V_{G}_{j}) at locus j, defining
Lower limit for the number of loci, n: We make use of the mathematical result that, for any two sets of real numbers {z_{i}, i = 1,..., n} and {w_{i}, i = 1,..., n}, if
Special cases: The condition
By straightforward though tedious algebra it can be shown that the condition
The implied level of overdominance (ρ) is then 1.73. Extreme overdominance: In the most extreme form of overdominance, all homozygotes have one assigned value (0, say), and all heterozygotes have another (d, say). Then at a single locus, and dropping the suffix j, we have
Conclusions: The models explored here suggest that a sufficient condition for n_{L} (with V_{T} = V_{GT} = Σ_{j}V_{G}_{j}) to be a lower bound for n is likely to be satisfied in most practical circumstances and will fail only in situations of extreme overdominance. In the most extreme such situation, when all homozygotes have the same genetic value and all heterozygotes have a different one, Cauchy’s inequality leads to the result that
Finally, it should be borne in mind that the present method reveals nothing about the relative magnitude of the dominance effects at different loci or of course about the presence of additive effects.
Acknowledgments
The authors thank Professor Bill Hill, Professor Brian Charlesworth, and Dr. Peter Visscher for helpful discussions, comments, and suggestions. This work was supported by the Wellcome Trust (IRDA) grant to H.C. and I.R., the Croatian Ministry of Science and Technology (CMST) grant 01960101 to P.R., 0196005 to P.R., 0196001 to N.S.N., and 0108330 to I.R., and the joint British Council and CMST grant ALIS 054 to H.C. and I.R. I.R. was supported by funds from the UK Medical Research Council, the University of Edinburgh, and the Overseas Research Scheme.
Footnotes

Communicating editor: D. Charlesworth
 Received June 28, 2002.
 Accepted November 19, 2002.
 Copyright © 2003 by the Genetics Society of America