INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Pulmonary function, as measured by FEV₁, is a strong correlate of chronic obstructive pulmonary disease (COPD), and in population studies is predictive of all-cause mortality (1). Environmental exposure, particularly cigarette smoking, is important in the development of COPD; however, only 15% to 20% of smokers develop clinical disease (4). In addition, 15% to 20% of COPD-related mortality occurs in patients with no history of cigarette smoking (5). This suggests the importance of host factors, which may be genetic, in determining the susceptibility to COPD, although few such factors have been positively identified. Homozygous deficiency of the serine protease inhibitor ₁-antitrypsin (₁-AT) causes emphysema in smokers in their fourth decade and nonsmokers in their fifth decade (6), but is present in only 0.02% of the white population of the United States, and accounts for less than 2% of all COPD (7). Recently, mutations in alleles of ₁-antichymotrypsin, another protease inhibitor, have been associated with obstructive lung disease in case-control studies (8, 9), but the relation of this enzyme to COPD has not been evaluated in population studies.

FEV₁ increases from birth to ages 20 to 25 yr, is stable for 5 to 10 yr, and then declines throughout the remainder of adulthood (10, 11). Thus, the FEV₁ in middle-aged and older adults is related to both the maximal attained level of pulmonary function in young adulthood and the rate at which function has been lost. Previous studies suggest that familial factors account for 15% to 60% of variability in pulmonary function (12). In all but one of these studies (15), this familial correlation persisted after adjustment for body habitus. These studies, however, examined the relation of pulmonary function among parents who were adults and their offspring who were still children. The parents' pulmonary function might have already begun to decline with age, whereas their children had not yet attained maximal pulmonary function.

The Framingham Study, a longitudinal cohort study begun in 1948, provides a unique population in which to study the genetics of pulmonary function. Many siblings and spouse pairs were enrolled in the Original Cohort, and the recruitment of the offspring of the original participants for the Offspring Cohort in 1971 produced numerous extended pedigrees for analysis. Spirometric data have been collected during adulthood from each generation of many of these pedigrees. To assess the potential for genetic control of pulmonary function, we assessed familial correlations and performed a segregation analysis using Class D regressive models (19) among participants in the Framingham Study.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Sample

The Original Cohort of the Framingham Study was established between 1948 and 1952 as a random sample of adult residents of the town of Framingham, Massachusetts. Of the approximately 10,000 men and women aged 30 to 59 yr living in the town, 5,209 individuals were enrolled in the study. This group included 596 sibships (two to six siblings/sibship) and 1,644 husband-wife pairs. Subjects have returned every 2 yr for a structured interview, including interim smoking history, and physical examination, including anthropometric measurements. At some visits the subjects underwent spirometry (details are provided subsequently).

Between 1971 and 1975, the Framingham Study was expanded to include a second generation. The Offspring Cohort comprises 5,124 individuals: 2,616 biologic children of the 1,644 Original Cohort husband-wife pairs, 898 biologic children of 378 Original Cohort members with heart disease, 34 stepchildren or adopted children of Original Cohort members, and 1,580 of the offspring's spouses who were not descendants of Original Cohort members. These subjects have returned every 4 yr for an examination similar to that given to the Original Cohort.

Genetic and spousal relationships were reported by the participants in both cohorts of the Framingham Study, and extended pedigrees were constructed. For this analysis, the sample comprises those subjects with acceptable spirometry (defined subsequently) and at least one relative with acceptable spirometry at either Original Cohort Examination 6 (conducted between 1962 and 1964) or at Offspring Cohort Examination 5 (conducted between 1991 and 1994). The sample includes 5,003 individuals from 1,408 extended pedigrees, of whom 2,871 were members of the Original Cohort and 2,132 were members of the Offspring Cohort. The pedigrees comprise first- and second- degree relatives over three generations. This analysis utilizes observations from 3,118 parent-offspring pairs, 2,284 sibling pairs, 101 grandparent-grandchild pairs, 1,280 avuncular pairs, 645 cousin pairs, and 908 spouse pairs.

Smoking Status Determination

For each subject, smoking status was determined as of the time of the spirometry used in the present analysis of pulmonary function. At their initial visit, members of both cohorts were asked if they had ever smoked cigarettes, pipes, or cigars, and at each subsequent visit were asked if they had smoked regularly in the previous 12 mo and how much. Subjects who answered no at each visit were classified as never-smokers. Subjects who reported smoking prior to their first visit or at any subsequent visit, but not at Examination 6 for the Original Cohort or at Examination 5 for the Offspring Cohort, were considered former cigarette or former pipe/cigar smokers. Subjects who reported smoking at the time of their spirometry were classified as current cigarette or current pipe/cigar smokers.

The lifetime amount smoked (pack-years) for current cigarette smokers was estimated from data on the age at which smoking began and on the number of cigarettes smoked per day, collected during the 12 yr of follow-up for the Original Cohort and the 20 yr of follow-up for the Offspring Cohort.

Anthropometric Measurements

Weight and standing height were measured with patients' shoes removed. Waist girth was measured at the level of the umbilicus.

Spirometry

Spirometry for the Original Cohort was performed between 1962 and 1964, using a 6-L Collins water-filled survey spirometer (Warren E. Collins, Inc., Braintree, MA). Each subject performed three FVC maneuvers while standing. FVC and FEV₁ were measured by hand, using back-extrapolation. Only the maneuver with the greatest FVC was recorded in the chart and was available for the present analysis. Nearly all (98.6%) of the 4,259 subjects who attended Original Cohort Examination 6 underwent spirometry, and a review of a sample of the spirograms from this examination indicated that most of the tracings met guidelines relating to smoothness of curve and forceful initial push (20). Raw values were multiplied by 1.10 to correct for body temperature, ambient pressure, saturated with water (BTPS), assuming a clinic temperature of 20° C.

Spirometry for the Offspring Cohort was performed between 1991 and 1994 with a Collins Survey II spirometer connected to an Eagle II microprocessor (Warren E. Collins, Inc.) equipped with pulmonary-function software developed by S & M Instruments (Doylestown, PA) and adapted for use in epidemiologic studies. The software provided automatic correction for BTPS, based on calibrations performed daily by the technicians. Patients performed FVC maneuvers in the standing position while wearing nose clips, and repeated them until at least three acceptable spirograms were obtained, up to a maximum of eight spirograms, using American Thoracic Society (ATS) standards of reproducibility to define acceptable spirograms (21). The majority (86.1%) of the 3,799 subjects who attended Offspring Examination 5 underwent spirometry, and 99.7% of these subjects had at least one acceptable maneuver. The largest FEV₁ from among all acceptable maneuvers was used in the present analysis.

Statistical Analysis

Calculation of standardized residual FEV₁. Adjustment for the effects of gender, age, body size, and smoking history on pulmonary function was performed in an initial stage of the analysis, to reduce the number of parameters fit in the genetic analyses. In exploratory analyses, numerous combinations of age, height, height², weight, waist girth, weight/height ratio, and Ponderal's index (height/weight^1/3) were used as the predictor variables in multiple-linear-regression analyses, run separately for the men and women in the Original and Offspring Cohorts. All models containing any combination of age and height provided approximately equal fit to the data as assessed through R² (data not shown). Residual FEV₁ was defined as the difference between the measured FEV₁ and the FEV₁ predicted by the regression equation, and a standardized residual FEV₁ (srFEV₁) for each subject was calculated by dividing the residual FEV₁ by the standard deviation of residual FEV₁ in the gender- and cohort-specific group. Familial correlations of the standardized residuals were determined with the FCOR program in Statistical Analysis for Genetic Epidemiology (SAGE), release 2.1 (22). The model that yielded the smallest spouse-pair correlation of srFEV₁ presumably provided the most complete adjustment for the effects of body size, which would tend to be correlated among spouses for reasons unrelated to shared genotypes, and therefore was selected for use in the subsequent analyses. The effect of smoking history was added into the final models with categorical variables for current or former smoking status and a continuous variable for pack-years of cigarette smoking for the current cigarette smokers.

The final model used to calculate residual FEV₁ was:

(1)

where pack-years = 0 for never-smokers, former cigarette smokers, and former and current pipe/cigar smokers. The residual FEV₁ obtained from the separate regressions for the men and women in the Original and Offspring Cohorts was standardized as described earlier; thus, for each group, the srFEV₁ had a mean value of 0 and a variance of 1 (data not shown). The use of srFEV₁ as the phenotypic variable in the genetic analyses provided adjustment for gender effects and for both biologic and technical cohort effects on FEV₁.

Familial correlations and segregation analyses. Familial correlations and segregation analyses of srFEV₁ were performed by fitting the Class D regressive model of Bonney (19), using the FCOR and REGC routines in the SAGE software package (22). A commingling analysis was performed to determine whether the data were best described by one, two, or three distributions.

Segregation analysis is used to determine whether the distribution of a trait (in this case, srFEV₁) among related persons can be explained by a specific inheritance pattern consistent with genetic transmission. In regressive models, dependencies among family members are modeled as a Markovian process in which an individual's phenotype is dependent on a set of explanatory variables, including his/her own covariates, a major gene effect, and the covariates and phenotypes of progenitors (23). The general model assumes three population distributions of srFEV₁. The presence of a major gene is assessed by allowing two alleles (denoted A and B) at a single locus, resulting in three "types" of individuals (AA, AB, BB). The mean srFEV₁ for each type is denoted by µ_AA, µ_AB, and µ_BB, with variances ²_AA, ²_AB, and ²_BB, respectively. The frequency of allele A is given by q_A and that of allele B by (1  q_A). Individuals of each type are assumed to transmit allele A to their offspring with transmission probabilities denoted by _AA, _AB, and _BB. For example, in a Mendelian model, the transmission probabilities are specified _AA  1.0, _AB 0.5, and _BB  0.0. Furthermore, the general model allows for correlation between spouse pairs, mothers and offspring, fathers and offspring, and between siblings, denoted as (sp), (mo), (fo), and (sib), respectively. Class D regressive models assume that correlations are equal among all siblings, but the correlations are not assumed to be due solely to common parentage.

Three models of genetic transmission were explored in this analysis: a single Mendelian locus without residual familial correlation, a "mixed model" comprising a Mendelian locus and residual familial correlation, and a model with familial correlation but no major gene. Also tested was a model of "no parent-offspring transmission of a major factor." This model postulates a factor that is equally likely to affect any member of the population, and the likelihood of exposure to the putative factor is determined by the percentage of the population that falls into the distribution with the lowest pulmonary function, represented mathematically as a "transmission probability" set equal to q_A (_{AA  AB  BB}  q_A). The model of "no parent-offspring transmission of a major factor," like the Mendelian and "mixed models," specifies three distributions of srFEV₁, but the probability that an offspring falls into one of the distributions is independent of the distributions of the parents. Each model was compared with a general model in which all parameters are arbitrary (i.e., no particular mode of inheritance is assumed).

The fit of hierarchical models was compared with the likelihood ratio (LR) test, calculated as the difference between the 2 ln likelihood of the models being compared. The LR follows a chi-square distribution, with degrees of freedom (df) equal to the difference between the models in the number of parameters estimated. Among nonhierarchical models, the most parsimonious model is that with the lowest value of Akaike's information criterion (AIC = 2 ln likelihood + 2 [number of estimated parameters]) (24).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The spirometry used in this analysis was done on both cohorts at a mean age of 53 to 54 yr (Table 1). Male and female offspring were heavier than their parents and were less likely to be current cigarette smokers, reflecting changes in body habitus and smoking behavior in the United States between the 1960s and the 1990s. The men and women of the Original Cohort had lower unadjusted pulmonary function (FEV₁ and FVC) values than did the Offspring Cohort subjects. In part, this difference reflects both the shorter height and greater prevalence of smoking in the older generation. Regression of FEV₁ and FVC on age and height among healthy nonsmokers revealed that subjects in the Cohort Offspring had slightly greater pulmonary function than did subjects of a similar age and height in the Original Cohort (data not shown). This difference is attributable either to biologic differences between the cohorts or to changes in the spirometric techniques used in the 1960s and the 1990s.

The final models used to calculate residual FEV₁ for the men and women in each cohort are given in Table 2. Among the men, the relation between pulmonary function and weight was positive in the Original Cohort but inverse in the Offspring Cohort, perhaps reflecting the greater obesity of middle-aged men in the 1990s than in the 1960s. Lifetime cigarette exposure (pack-years) was not available for some former smokers in the Original Cohort, and to maintain comparability of data between the cohorts, adjustment for pack-years was therefore made only for current cigarette smokers. As expected, cigarette smoking was associated with significantly lower cross-sectional pulmonary function (Table 2). Number of pack-years smoked was a strong predictor of pulmonary function. With adjustment for pack-years, current cigarette smoking status was no longer significant for the women in either cohort, although it remained significant for the men. Without a separate adjustment for pack-years, former cigarette smoking predicted significantly lower pulmonary function among the men in either cohort. Among women, former smoking was not associated with a decrement in residual FEV₁, perhaps because of less overall exposure than among the men. Pipe and cigar smoking failed to enter any model as significant predictors of residual FEV₁, but all smoking terms were retained in order to provide the most precise adjustment possible for tobacco exposure.

Familial Correlation

In an arbitrary model that assumes no major gene, the correlation between biologic relatives of srFEV₁ was substantial (Model A in Table 3). A model in which the spouse-pair correlation was fixed at zero (Model B in Table 3) fit the data as well as the arbitrary model (p = 0.2), suggesting that the spouse-pair correlation of 0.045 in the arbitrary model is not significantly different from zero. The correlation of srFEV₁ between mothers and offspring ([mo] = 0.193) exceeded the correlation between fathers and offspring ([fo] = 0.119) in the arbitrary model. When the parent-offspring correlations were set equal to one another ([mo]  [fo] = 0.156; Model C in Table 3), the resulting model provided a less good fit to the observed distribution of srFEV₁ than did the arbitrary or the no-spouse-pair correlation models (p = 0.06). This suggests that the pulmonary function of mothers and their children is more highly correlated than that of fathers and their children. The correlation between siblings of srFEV₁ exceeded the parent-offspring correlations in all models in which these parameters were not fixed (Models A, B, and C in Table 3). A model in which the parent-offspring and sibling correlations were set equal to one another was rejected (p < 0.01), suggesting that the sibling correlation is actually greater than the correlation between parents and children, perhaps reflecting cohort effects or the greater shared environment of offspring within a family. Models assuming no correlation between parents and offspring or no familial correlation at all (Models E and F in Table 3) provided a significantly worse fit to the data than did the arbitrary model (p < 0.001).

Segregation Analysis

A commingling analysis demonstrated that srFEV₁ was better described by a trimodal than by a bimodal or unimodal distribution, and that skewed distributions were superior to normal distributions (data not shown). Because of residual skewness, segregation analyses were performed with both srFEV₁ and its Box-Cox transformation. Results with either variable were comparable, and therefore the nontransformed data are presented in the following sections.

In the arbitrary genetic model (Model 1 in Table 4), individuals homozygous for allele A show much lower pulmonary function than individuals homozygous for allele B, and heterozygotes show intermediate pulmonary function (µ_AA = 1.42, µ_AB = 0.038, and µ_BB = 0.187); however, the transmission frequencies do not correspond to a Mendelian gene (_AA = 0.183, _AB = 0.463, and _BB = 0.161). In addition, familial correlation after accounting for the presence of the major gene remains substantial ([mo] = 0.243, [fo] = 0.152, [sib] = 0.280).

In the model that imposes Mendelian transmission probabilities and permits residual familial correlation (Model 2 in Table 4), allele A is common (q_A = 0.449) and associated with mildly impaired pulmonary function in homozygotes (µ_AA = 0.504), and the residual familial correlation is mildly reduced ([mo] = 0.195, [fo] = 0.104, [sib] = 0.240). This model fits the distribution of srFEV₁ significantly better than a model that allows familial correlation but includes no major gene effect (Model 2 versus Model 3 in Table 4; LR = 133.98, df = 5, p < 0.001). This implies significant transmission of pulmonary function beyond that represented by the residual familial correlation. Comparison of Model 2 with a model that specifies Mendelian transmission probabilities but does not permit residual familial correlation ([mo]  [fo] [sib]  0; Model 4 in table 4) shows that the residual familial correlation significantly improves the fit to the data (LR = 46.21, df = 3, p < 0.001). The comparison of Models 2, 3, and 4 supports the conclusion that both a transmitted factor and residual familial correlation are necessary to explain the distribution of srFEV₁ in this sample.

The arbitrary genetic model (Model 1 in Table 4) fits the distribution of srFEV₁ significantly better than the model that specifies Mendelian transmission probabilities and permits residual familial correlation (Model 2 in Table 4, LR = 14.45, df = 3, p < 0.005). This comparison indicates that the data do not conform to a model that assumes a Mendelian major gene. The deviation from Mendelian transmission probabilities in the arbitrary model is most pronounced for the AA parents, for whom _AA = 0.183 rather than 1.0, suggesting that some parents with severely impaired pulmonary function are not transmitting this phenotype to their offspring.

The model of no parent-offspring transmission of a major factor (Model 5 in Table 4) specifies that there are three distributions of srFEV₁ in the population (in contrast to the familial correlation model, which assumes a single distribution, [Model 3 in Table 4]), and that the probability with which an offspring falls into one of the three distributions is independent of the distributions from which the parents derive. This model provided as good a fit to the distribution of srFEV₁ as the arbitrary model with residual familial correlation, and is more parsimonious.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

We have demonstrated substantial familial aggregation of pulmonary function in a general population sample of middle-aged adults after adjustment for body habitus and smoking history. Genetic models, which allow for a trimodal population distribution of pulmonary function, explained the distribution of srFEV₁ better than models positing a single population distribution; however, the data do not support the existence of a single major gene that determines pulmonary function. The best model suggests that familial aggregation of level of pulmonary function is determined by environmental factors and/ or polygenic influences, as might be expected, given the complexity of lung development and the different biologic systems likely to affect maximal attained FEV₁ and its decline over time.

Our results confirm those of the majority of investigations into the heritability of pulmonary function, which have found significant aggregation of pulmonary function among relatives, and extend this observation to a two-generation population in which all measurements were made in adults. Most studies of familial correlation of pulmonary function have used measurements made during adulthood for parents and during childhood for offspring (12); however, we have demonstrated a strong familial correlation of pulmonary function in a population in which both generations were middle-aged at the time of pulmonary-function measurement, at an age when pulmonary function will reflect not only maximal attained lung size but also the effects of adult decline in pulmonary function.

In our sample, the correlation of srFEV₁ for mothers and offspring exceeded that for fathers and offspring (p = 0.06). This excess correlation may relate to genetic factors, (e.g., maternal transmission of mitochondrial DNA), or to environmental factors that children share more closely with their mothers than with their fathers (e.g., the effect on lung development of prenatal or early postnatal tobacco-smoke exposure) (25). Alternatively, father-offspring correlations may be underestimated in our sample, owing to misclassification from greater underreporting of nonpaternity than of nonmaternity. Using path analysis, consistently larger mother-offspring than father- offspring correlations, and a suggestion of a maternal environmental effect, were demonstrated in families in East Boston (14), but with the same techniques, Coultas and colleagues found no differences in parent-child correlations by sex among Hispanic families in New Mexico (17). Similarly, in a segregation analysis of pulmonary function in the Humboldt Family Study, mother-offspring and father-offspring correlations did not differ (18). These discrepancies may reflect differences in maternal smoking behavior among these different population samples. Alternatively, the substantially greater power of the present study may have enabled demonstration of a significantly greater maternal-offspring correlation.

Sibling correlations in our study exceeded those between parents and offspring (p < 0.01), a finding not reported in previously published analyses of familial aggregation of pulmonary function. This finding may reflect greater measurement error in the FEV₁ data for the parents than for the offspring in the current analysis (given the improvements in spirometric technique in the intervening three decades). Alternatively, the excess sibling correlation may reflect the importance of environmental factors that are shared more closely by siblings than by parents and offspring. For example, siblings are more likely to be concordant for exposure to parental cigarette smoke during infancy and childhood, an exposure that may modify pulmonary function.

Genetic transmission of the level of pulmonary function in families has been examined by segregation analysis in two other populations. The findings of Chen and colleagues of familial aggregation of pulmonary function in the Humboldt Family Study (18) are consistent with our results. In both the Framingham Study and the Humboldt Family Study populations, the most parsimonious model for FEV₁ included a nontransmitted major factor, possibly representing an environmental effect, and residual familial correlation. In both studies, the familial correlations that remained after accounting for the environmental effect were large (in the Humboldt Family Study, [mo]  [fo]  [sib] = 0.184; in our study, [mo] = 0.237, [fo] = 0.150, [sib] = 0.280). These residual familial correlations may represent polygenic effects or unmeasured environmental effects shared by families. In contrast, Rybicki and colleagues reported no familial aggregation of pulmonary function in 56 families selected independently of the presence of pulmonary disease, whereas in 85 families of patients with COPD they found evidence for a major gene without significant residual familial correlation (28). Although among the COPD families, the residual parent-offspring correlation was large in a model with a major gene ([mo]  [fo] = 0.323), inclusion of the familial correlation did not significantly improve the fit of the model, perhaps because of the smaller sample size and lower power of the study. Two factors limit interpretation of Rybicki and colleagues' study. First, smoking status was not included in the regression done to determine residual FEV₁ or as a covariate in some of the segregation-analysis models. In addition, this analysis used the simpler Class A regressive models, in which sibling correlations are assumed to be equal and due entirely to common parentage. In a population in which there is an excess of sibling correlation relative to parent-offspring correlation, the Class A models may be subject to the false inference of a major gene (29). Alternatively, analysis of a more homogeneous group of families, as might occur in restricting the sample to families with a history of COPD, could identify a major gene lost in the overall genetic heterogeneity of the determinants of level of pulmonary function in the general population.

A potential limitation of our analysis relates to possible imprecision in spirometry data collected 30 yr earlier in the Original Cohort than in the Offspring Cohort, prior to the publication of ATS guidelines for standardization of spirometry (21, 30). The use of cohort- and gender-specific regressions, and standardization of the residual FEV₁, minimizes the impact of cohort effects related to spirometric technique. Moreover, the ability to analyze the segregation of pulmonary function when both generations are the same age, and thus have equal potential for exposure to factors that promote decline from maximal pulmonary function (whether these factors are genetic or environmental), is a particular strength of the present study.

In summary, middle-aged biologic relatives in the Framingham Study show substantial familial aggregation of pulmonary function after adjustment for age, body habitus, and smoking status. The pattern of pulmonary-function segregation in this sample is best explained by polygenic effects and/or environmental effects other than smoking. Our failure to demonstrate a single Mendelian gene in this analysis may speak to the limitations of segregation analysis to parse out the probable complex and multifactorial determinants of the level of pulmonary function.