Heritability and individuality of the plasma sodium concentration: a twin study in the United States veteran population

Abstract

Little is known about the population genetics of water balance. A recent meta-genome-wide association study on plasma sodium concentration identified novel loci of high biological plausibility, yet heritability of the phenotype has never been convincingly shown in European ancestry. The present study linked the Vietnam Era Twin Registry with the Department of Veterans Affairs VistA patient care clinical database. Participants (n = 2,370, 59.6% monozygotic twins and 40.4% dizygotic twins) had a median of seven (interquartile range: 3−14) plasma sodium determinations between October 1999 and March 2017. Heritability of the mean plasma sodium concentration among all twins was 0.41 (95% confidence interval: 0.35−0.46) and 0.49 (95% confidence interval: 0.43−0.54) after exclusion of 514 twins with only a single plasma sodium determination. Heritability among Caucasian (n = 1,958) and African-American (n = 268) twins was 0.41 (95% confidence interval: 0.34−0.47) and 0.36 (95% confidence interval: 0.17−0.52), respectively. Exclusion of data from twins who had been prescribed medications known to impact systemic water balance had no effect. The ability of the present study to newly detect substantial heritability across multiple racial groups was potentially a function of the cohort size and relatedness, exclusion of sodium determinations confounded by elevated plasma glucose and/or reduced glomerular filtration rate, transformation of plasma sodium for the independent osmotic effect of plasma glucose, and use of multiple laboratory determinations per individual over a period of years. Individual-level plasma sodium concentration exhibited longitudinal stability (i.e., individuality); the degree to which individual-level means differed from the population mean was substantial, irrespective of the number of determinations. In aggregate, these data establish the heritability of plasma sodium concentration in European ancestry and corroborate its individuality.

INTRODUCTION

Plasma osmolality is the clinically ascertainable index of systemic water balance; a low plasma osmolality is synonymous with a water-excess state, and an elevated plasma osmolality is almost invariably a consequence of water loss. Among the most tightly regulated of physiological parameters, plasma osmolality is determined largely by the plasma sodium concentration in conjunction with its counterbalancing anions.

In mammals, water balance is regulated through central sensing of the plasma osmolality (1) and/or plasma sodium concentration (11), which regulate hypothalamic production of, and release from the posterior pituitary of, the water-retaining hormone arginine vasopressin (AVP; also known as antidiuretic hormone). Once released into the circulation, AVP interacts with its cognate receptor on the basolateral membrane of epithelial cells lining the collecting duct of the nephron and elicits a series of signal transduction events that culminates in the fusion of water channel-containing intracellular vesicles with the apical plasma membrane (14). This insertion event is permissive for water reabsorption from the glomerular filtrate and is one of the final steps in determining composition of the urine.

Regulation of water balance may be impaired, leading to hyper- or hyponatremia. Underproduction of AVP or insensitivity of the collecting duct to the effects of AVP leads to water wasting. Provided that central sensing of the plasma osmolality is intact, and that access to potable water is not limited, the plasma sodium concentration is readily maintained within the normal range via thirst, and hypernatremia (pathologically elevated plasma sodium concentration) is not seen. Excess water retention is potentially more problematic in an ambulatory population. A decreased ability to excrete water can stem from an inability to suppress circulating AVP levels or from a number of clinical conditions associated with decreased effective arterial blood volume. Medications such as thiazide diuretics and antidepressants also commonly impair water excretion, as does chronic kidney disease (5).

Recent evidence supports individuality of the plasma sodium concentration, indicating reproducibility of the phenotype over time in an individual, within the broader normal distribution (31). Because many normally distributed phenotypes are polygenic (23), we investigated the genetic basis for interindividual differences in plasma sodium concentration. An understanding of the genetic basis for such differences can lead to the discovery of new genes regulating systemic water balance under physiological and/or pathological conditions. It can also aid in the interpretation of stably extreme values in the population distribution of plasma sodium concentration by discriminating between pathological states (e.g., tumor-associated AVP excess) and the expected impact of a near-normally distributed phenotype (e.g., cooccurrence of multiple genetic variants). After using a candidate gene-based approach to identify a genetic variant that impacted water balance (26), we sought to explore the heritability of plasma sodium concentration. Previous studies in generally small cohorts detected little to no heritability of plasma sodium concentration (2, 15–17, 20, 22, 28). Using a variety of large family-based cohorts reflecting diverse races and ethnicities, we detected strong heritability in African-American and American Indian cohorts; however, little heritability was detected in cohorts of European ancestry (29). More recently, in a transethnic meta-genome-wide association study on plasma sodium concentration (comprised predominantly of participants of European ancestry), we identified novel genetic loci of high biological plausibility that associated with plasma sodium concentration (3). We infer that heritability is present among Caucasian populations and speculate that limitations in sample size or other factors have precluded its detection. It is notable that every prior investigation of the heritability of this phenotype relied upon single determinations of the plasma sodium concentration.

In the present study, we used the Department of Veterans Affairs Vietnam Era Twin Registry [VET Registry (10)] and corresponding Department of Veterans Affairs VistA electronic health record data (4) to leverage the individuality of the plasma sodium concentration (assessed serially between October 1999 and March 2017 in the routine delivery of patient care) to estimate the cross-ancestry heritability of the plasma sodium concentration in this predominantly Caucasian cohort.

METHODS

Sample Selection

The VET Registry cohort includes 14,738 twins who served in the United States military during the time of the Vietnam conflict (1965–1975) (8). Veterans Affairs outpatient laboratory data were available from the Veterans Affairs Corporate Data Warehouse from October 1, 1999 to March 31, 2017. Outpatient sodium laboratory data were available for 5,974 VET Registry members, of which 5,377 members had laboratory data eligible for study inclusion (see cohort diagram, Fig. 1). Singletons (n = 2,603) were excluded because of lack of data on their twin partner. Twins were excluded (n = 404) in instances of indeterminate zygosity. Data were available from 2,370 twins (i.e., 1,185 twin-pair dyads), of which 1,412 twins were monozygotic (MZ) and 958 twins were dizygotic (DZ).

Fig. 1.

Cohort diagram depicting exclusion criteria. Ineligible laboratory data included plasma sodium concentrations and corresponding (i.e., simultaneous) chemistry panels to which one or more of the following criteria applied: 1) laboratory values missing or with incorrect units, 2) top and bottom 1% value for sodium, glucose, and creatinine determination, 3) glucose >150 mg/dl, 4) visit with estimated glomerular filtration rate (eGFR) > 2 SD units below the population mean or with eGFR within the bottom decile for the entire sample, 5) duplicate data for same visit date (second and subsequent occurrence removed), and 6) laboratory visits occurring within 7 or fewer days of each other. VETR, Vietnam Era Twin Registry.

All blood biochemistry data corresponding to outpatient status were obtained from the electronic health record (VistA; Department of Veterans Affairs) linked to each VET Registry member. Plasma sodium concentration determinations below the 1st percentile or above the 99th percentile were excluded. Plasma sodium concentrations coinciding with simultaneous plasma glucose concentration >150 mg/dl or glucose below the 1st percentile and/or with plasma creatinine concentration below the 1st percentile and above the 99th percentile were excluded. Glucose determinations included both fasting and nonfasting values. Because dysnatremia is common in chronic kidney disease (6), an isotope dilution mass spectrometry-traceable modification of diet in renal disease estimate of glomerular filtration rate (eGFR) was used to exclude plasma sodium concentration values coinciding with eGFR >2 SD below the sample mean. Stratification on sex was not performed at this step because VET Registry participants were all men. Two SD units below the mean corresponded to an eGFR value of 39.8 ml/min. We then excluded readings below the 10th percentile of the remainder, corresponding to an eGFR of 55.9 ml/min. In instances in which plasma sodium concentration was measured within 7 days of another determination, both results were excluded from further analysis. This was done to minimize the impact of acute illness or, potentially, laboratory error necessitating repeated assessment of plasma sodium concentration. Because VistA laboratory data were collected and assayed at many different facilities over many years, we do not have specific information on the method used for the determination of plasma sodium concentrations or the dates when any such method may have been modified by one or more of the clinical laboratories. Exclusions were based on objective biochemical data and not coded diagnoses (e.g., heart or liver disease) because of concerns about the completeness and reliability of the latter.

Plasma sodium concentration was transformed for simultaneous plasma glucose concentration according to the Katz formula (13) to account for the independent osmotic effect of the plasma glucose concentration, as was done previously (3, 29, 31). Because plasma sodium determinations coinciding with plasma glucose > 150 mg/dl were excluded, the effect of the transformation upon the plasma sodium concentration was modest (<1 meq/l). For each participant, the initial plasma sodium concentration was defined as the chronological first plasma sodium determination within the study interval or the only determination in instances in which plasma sodium concentration was measured just once. For each participant, the mean plasma sodium concentration (also referred to as the individual-level mean) was the arithmetic mean of all included determinations for that individual. Note that for Table 4 only, the mean plasma sodium concentration represented the mean of all determinations after excluding the initial plasma sodium determination.

Table 4.

Concordance of categorization of initial plasma sodium concentration and individual-level mean plasma sodium concentration for the same twin in all twins

	Mean Sodium
Initial Sodium	<135 meq/l	135 to <140 meq/l	140 to <142.5 meq/l	≥142.5 meq/l
<135 meq/l	1.3 (28)	3.2 (66)	0.0 (1)	0.0 (0)	4.6% (95)
135 to <140 meq/l	1.7 (35)	38.1 (792)	12.0 (249)	0.3 (7)	52.1% (1,083)
140 to <142.5 meq/l	0.1 (2)	15.6 (324)	14.5 (302)	1.7 (35)	31.9% (663)
≥142.5 meq/l	0.1 (2)	4.0 (83)	6.4 (132)	1.0 (21)	11.5% (238)
Total	3.2 (67)	60.8 (1,265)	32.9 (684)	3.0 (63)	100.0% (2,079)

Concordance values are in percentages, with n values in parentheses. The initial plasma sodium concentration determination was not used in the calculation of the mean plasma sodium concentration for this analysis only. Category concordance (initial vs. mean) was 55.13%. Categories are the same as in Table 3.

Sensitivity analysis tested the impact of concurrent use of medications in classes known to influence systemic water balance. For this drug-excluded analysis, patient pharmacy records were queried for such medications (see medication list in Supplemental Table S1 of the Supplemental Material; the Supplemental Material is available online at https://drive.google.com/drive/folders/1zEJbJfuJ7eAQGLL4zh6FD85Kp3NpwsAi?usp=sharing). A “dispense date” and “days’ supply” for each medication record were obtained. Participant laboratories were excluded if they coincided with an interval of use for one of these medications [i.e., within (dispense date, dispense date + days’ supply)]. Medications prescribed but never picked up by a participant were not grounds for exclusion. A separate sensitivity analysis excluded participants with only a single plasma sodium determination.

Statistical Methods

Descriptive statistics.

The cohort was described by age (at the visit day for an individual’s first eligible laboratory), race, number of eligible laboratory determinations, and initial and mean of plasma sodium, glucose, and creatinine from these eligible laboratory determinations. Participant race was self-reported and was determined from VET Registry records (military records and two registry-wide surveys of participants) and from Veterans Affairs VistA data; therefore, VET Registry race data are more comprehensive than the Veterans Affairs clinical database. Race was unknown for four Veteran participants. These values were summarized for the full cohort and subset by zygosity. The distributions of the number of determinations, transformed plasma sodium determinations, and mean transformed plasma sodium determinations were plotted. Participant mean transformed plasma sodium was plotted against cotwin values, and a fitted line was estimated using a simple linear model to visualize the association for both MZ and DZ twins.

For one subanalysis, distribution of initial and mean transformed plasma sodium concentration was reduced to four categories (<135, 135 to <140, 140 to <142.5, and ≥142.5 meq/l) and subset by zygosity. Additionally, initial transformed plasma sodium and mean transformed plasma sodium (with initial sodium removed) were cross-tabulated using the previous categorization, and concordance was calculated; for only this subanalysis, initial sodium was removed from the mean calculation to prevent an overestimation of concordance. In all other instances, the initial plasma sodium determination was included in the determination of the mean plasma sodium concentration.

Heritability.

Structural equation modeling was used to assess the heritability of the mean plasma sodium concentration per participant, before and after stratification on race; race was not used as a covariate. Modeling assumed that the source of phenotypic variation for an individual was either a genetic (A), shared environmental (C), or individual environmental (E) factor and that the relationships of these factors between MZ and DZ twins were fixed (Supplemental Fig. S1). Structural equation models were fitted via maximum likelihood in R [R version 3.4.3 (25)] using the OpenMX package (18), and results were expressed using likelihood-based confidence intervals (19). The contribution of shared environmental factors to phenotypic variation between individuals within a twin pair was expected to be 0; therefore, the fit of nested models including only AE components versus the full ACE model was assessed by Akaike information criterion and a likelihood ratio statistic. These model comparisons were repeated for sensitivity and secondary analyses in all cases finding similar results between AE and ACE (see Supplemental Tables S7–S11). On the basis of this comparison, results are reported only from the AE models. For the two sensitivity analyses and for the secondary analyses that separately considered the Caucasian and African-American subgroups, heritability was estimated using the same methods.

Individuality.

To assess individuality of plasma sodium concentration, data were simulated under two scenarios for comparison with study data. One scenario (nonindividuality) assumed that all participants shared the same underlying sodium distribution; data points for single sodium determinations were simulated from a normal distribution, with mean and SD estimated from the real-world study data of all plasma sodium determinations. A second scenario (individuality) assumed that each participant had an individual mean; participant means (µ_i for each individual indexed by i) were generated from a normal distribution with mean and SD estimated from the real-world set of mean plasma sodium determinations, and single determinations were simulated from a normal distribution, with individual mean = µ_i and SD estimated as the mean of within-participant SD. In both scenarios, the overall distribution of the number of determinations per participant was fixed and equal to that of the study data. For this analysis, 100 data sets of 22,939 plasma sodium concentration values for 2,370 participants were simulated under each scenario. The mean plasma sodium concentration of simulated participants under the assumption of individuality was more likely to deviate from the study population mean than data simulated under the assumption of nonindividuality. Study data were compared with data simulated under the two scenarios, with deviation defined as the proportion of patient means ≥ 1 meq/l above or below the overall sample mean.

Individuality of plasma sodium concentration was also assessed through modeling. Two models were fit to the study data, allowing for comparison of 1) a simple linear regression model assuming nonindividuality, with only a single intercept (µ, the grand mean in Supplemental Table S20) for all patients, in which the only source of variability is random error (ε_ij: normally distributed with variance of s); and 2) a linear mixed model allowing for individuality, which adds a random intercept clustered at the patient level (α_i: patient i difference from the grand mean, normally distributed with variance of d; here, µ + α_i represents individual underlying sodium for patient i) and captures variation among patients, with additional variation captured by random error (ε_ij: normally distributed with variance of t). Parameters for each model were estimated separately, using maximum likelihood. Comparison of fit was assessed by comparing the likelihood ratio statistic of these models to a 50:50 mixture ${\text{χ}}_{\text{df}=1}^2/{\text{χ}}_{\text{df}=0}^2$ distribution, where df is degrees of freedom (24).

RESULTS

Demographics

A total of 2,370 VET Registry members were included, representing 1,185 twin pairs. All participants were men, 57.8% were MZ, and 85.0% were Caucasian (Table 1). Over the study interval, participants underwent a median of seven (interquartile range: 3−14) plasma sodium determinations. Data corresponding to reduced eGFR or markedly elevated plasma glucose were excluded (see methods). Before analysis, plasma sodium was transformed for the osmotic effect of plasma glucose using the formula of Katz (13) (see methods). The overall sample mean of the initial plasma sodium concentrations (i.e., the first determination for each participant) was 139.5 ± 2.5 meq/l, and the overall sample mean of the individual-level means was 139.3 ± 1.9 meq/l. Demographic data were similar between MZ and DZ twins (Table 1).

Table 1.

Demographic data of the study participants

Characteristic	All Twins (n = 2,370)	Monozygotic Twins (n = 1,412)	Dizygotic Twins (n = 958)
Age, yr [mean (SD)]	58.1 (5.9)	57.8 (6.0)	58.4 (5.6)
Race, %
Caucasian	85.0	84.8	85.3
African-American	11.5	11.3	11.8
Asian	0.2	0.4	0.0
Other	3.1	3.5	2.6
Unknown	0.2	0.1	0.3
Determinations, median (interquartile range)	7 (3, 14)	7 (3, 14)	7 (3, 13)
Sodium, meq/l [mean (SD)]
Initial	139.5 (2.5)	139.5 (2.5)	139.4 (2.6)
Mean	139.3 (1.9)	139.3 (1.9)	139.3 (2.0)
Glucose, mg/dl [mean (SD)]
Initial	103.0 (16.4)	102.5 (16.3)	103.8 (16.4)
Mean	104.9 (12.7)	104.6 (12.5)	105.5 (13.0)
Creatinine, mg/dl [mean (SD)]
Initial	1.0 (0.2)	1.0 (0.2)	1.0 (0.2)
Mean	1.0 (0.1)	1.0 (0.1)	1.0 (0.1)

Data are for monozygotic and dizygotic twins for whom plasma sodium concentration laboratory determinations were available for both members of the twin pair and subject to exclusions outlined in methods. Determinations is the number of individual plasma sodium determinations obtained per participant over the study interval. Initial laboratory values correspond to the first included chemistry laboratory panel wherein sodium, glucose, and creatinine were obtained simultaneously. Mean is the mean of all individual-level means for the laboratory test.

Distribution of Plasma Sodium Concentrations

The number of plasma sodium determinations per participant in the outpatient setting ranged from 1 to 57 (Fig. 2). Participants with a single sodium determination were most common, representing ~12% of the sample, with frequency generally decreasing with increasing number of determinations. The distribution of the number of determinations per participant was similar between MZ and DZ twins (Supplemental Fig. S2).

Fig. 2.

Distribution of participants (n = 2,370) by number of plasma sodium determinations. The final data column represents n ≥ 41 determinations, with a maximum of n = 57 determinations.

The distribution of all plasma sodium determinations considered in aggregate (n = 22,939) resembled a normal distribution, although it was skewed slightly to the left, in the direction of hyponatremia (Fig. 3). The distribution of individual-level means for plasma sodium concentrations (n = 2,370) exhibited a more pronounced central tendency, again with a leftward skew (Fig. 3). When separated by zygosity, the corresponding histograms for MZ and DZ twins were similar (Supplemental Figs. S3 and S4).

Fig. 3.

Distribution of the transformed plasma sodium concentration readings, in increments of 1 meq/l. Distributions of all determinations (n = 22,939 across the entire cohort; gray) and of the individual-level mean plasma sodium concentrations for each participant (n = 2,370 participants; yellow) are shown.

Some plasma sodium determinations and individual-level means were <135 meq/l (Fig. 3), which would generally be considered abnormal. The incidence of abnormal plasma sodium concentration may increase with age (e.g., Ref. 9), and a substantial fraction of participants underwent plasma sodium determinations temporally remote from their twin (Supplemental Figs. S5 and S6). Nonetheless, there was no correlation between measured plasma sodium concentration and the age at which the ascertainment was made (Supplemental Fig. S7 and Supplemental Table S2); therefore, it is unlikely that this analysis was confounded by age. The vast majority of plasma sodium determinations were made in the sixth and seventh decades of life (Supplemental Fig. S7), so inference about age-dependent effects beyond this narrow range cannot be drawn from the present data.

Heritability of Plasma Sodium Concentration

Within-pair correlations in plasma sodium concentration were compared as an index of heritability in MZ and DZ twins using variance components analysis for genetic and environmental influences (Table 2). For this analysis, the individual-level means of all plasma sodium determinations were used. The structural equation model estimates that 41% of the interindividual difference in mean plasma sodium concentration was heritable (i.e., heritability was 0.41, 95% confidence interval: 0.35–0.46); the within-pair correlation was 0.41 in MZ and 0.17 in DZ twins. In a sensitivity analysis that excluded twins who had been prescribed medications (e.g., diuretics and antidepressant and antipsychotic medications; Supplemental Table S1) known to influence water balance (leaving 752 MZ twins and 480 DZ twins), the estimate of heritability was similar at 0.44 (95% confidence interval: 0.36–0.51; Table 2). In a separate sensitivity analysis restricted to participants with two or more plasma sodium determinations (n = 1,116 MZ twins and 740 DZ twind), estimated heritability was 0.49 (95% confidence interval: 0.43–0.54). In a race-stratified series of secondary analyses, heritability in the present study was 0.41 (95% confidence interval: 0.34–0.47) among Caucasian twin pairs and 0.36 (95% confidence interval: 0.17–0.52) among African-American twin pairs (Table 2). Demographic data for participants in the sensitivity and secondary analyses are shown in Supplemental Tables S3–S6. A comparison of nested variance component models for each of the separate analyses in Table 2 is shown in Supplemental Tables S7–S11.

Table 2.

Twin correlations for mean plasma sodium determinations and estimates of genetic and individual environmental variance components

	Correlations		Standardized Variance Components (95% Confidence Interval)
	Monozygotic Twins	Dizygotic Twins	Genetic	Environmental
Primary (full data)*	0.41 (n = 1,412)	0.17 (n = 958)	0.41 (0.35, 0.46)	0.59 (0.54, 0.65)
Sensitivity (drug excluded)	0.41 (n = 750)	0.27 (n = 474)	0.44 (0.36, 0.51)	0.56 (0.49, 0.64)
Sensitivity (>1 determination)	0.48 (n = 1,116)	0.23 (n = 740)	0.49 (0.43, 0.54)	0.51 (0.46, 0.57)
Secondary (Caucasian)	0.41 (n = 1,162)	0.18 (n = 796)	0.41 (0.34, 0.47)	0.59 (0.53, 0.66)
Secondary (African-American)	0.41 (n = 156)	0.10 (n = 112)	0.36 (0.17, 0.52)	0.64 (0.48, 0.83)

Demographic data for the drug-excluded sensitivity analysis, sensitivity analysis after exclusion of participants with only a single plasma sodium determination, and race-stratified secondary analyses are available in the Supplemental Material. Heritability was estimated as the genetic variance component (see also Supplemental Fig. S1).

^*Primary analysis refers to “all twins” in Table 1.

A scatterplot of the mean plasma sodium concentrations for each MZ and DZ twin pair is shown in Fig. 4. The slope estimates for the fitted linear regression lines for MZ and DZ twin pairs were 0.41 (95% confidence interval: 0.34–0.47) and 0.17 (95% confidence interval: 0.08–0.25), respectively.

Fig. 4.

Correlation of mean plasma sodium concentrations within twin pairs. Mean plasma sodium concentration is shown for all monozygotic (MZ; yellow) and dizygotic (DZ; blue) twin pairs. Each twin pair (Twin-1 and Twin-2, order assigned arbitrarily) is represented by a single data point. Slope estimates for the fitted linear regression lines were 0.41 (0.34–0.47) for MZ twin pairs and 0.17 (0.08–0.25) for DZ twin pairs. The identity line is shown in gray.

Individuality of Plasma Sodium Concentration

Individuality of a phenotype is the degree to which any single assessment of that parameter is reflective of its true or intrinsic state. Although unrecognized for many decades, individuality of the plasma sodium concentration was recently shown (31). Individuality is a prerequisite for accurately phenotyping individuals, and accurate phenotyping is a prerequisite for demonstrating heritability. If there is no individual-level reproducibility or “stability” of a variable or phenotype, then its similarity to that of a related individual must be 0 (think height vs. shirt color). Hyponatremia, defined as plasma sodium concentration < 135 meq/l, was present in 4.9% of initial sodium determinations and 2.9% of mean plasma sodium determinations among all twins (n = 2,370 twins; Table 3). Similarly, 11.2% of initial sodium determinations were ≥142.5 meq/l, whereas 3.2% of individual-level means were ≥142.5 meq/l. The range of 135 to <140 meq/l included 52.2% of initial determinations and 60.5% of individual-level means, whereas the range from 140 to <142.5 meq/l included 31.7% of initial determinations and 33.4% of individual-level means. In a separate analysis focusing only on twins with two or more plasma sodium determinations (n = 1,116 MZ twins and 740 DZ twins), 4.5% of initial plasma sodium determinations were <135 meq/l, whereas 2.5% of all mean plasma sodium concentrations were <135 meq/l (Supplemental Table S12). This 2.5% prevalence of hyponatremia is consistent with convention wherein 95% of an unselected population falls within the “reference range” (or within 2 SD units of the mean) for a biochemical laboratory test. (e.g., Ref. 12).

Table 3.

Categorization of plasma sodium among monozygotic and dizygotic twins

		Zygosity
	All Twins (n = 2,370)	Monozygotic (n = 1,412)	Dizygotic (n = 958)
Initial Sodium, meq/l
Mean (SD)	139.5 (2.5)	139.5 (2.5)	139.4 (2.6)
<135, %	4.9	4.8	4.9
135 to <140, %	52.2	52.3	52.1
140 to <142.5, %	31.7	31.9	31.4
≥142.5, %	11.2	11.0	11.6
Mean Sodium, meq/l
Mean (SD)	139.3 (1.9)	139.3 (1.9)	139.3 (2.0)
<135, %	2.9	2.9	2.9
135 to <140, %	60.5	60.7	60.1
140 to <142.5, %	33.4	33.4	33.5
≥142.5, %	3.2	3.1	3.4

This primary analysis refers to “all twins” in Table 1. A threshold of 140 meq/l was used in accordance with convention and because the 139.5–140.5 stratum of all transformed plasma sodium concentration determinations in all participants was the largest (see Fig. 3). A sodium concentration of 135 meq/l is a typical threshold for hyponatremia, and a threshold of 142.5 meq/l was chosen to create an upper quantile similar in size to the hyponatremic category.

In an effort to assess how predictive the initial plasma sodium concentration category was of the mean plasma sodium concentration category for a given individual, we excluded the initial reading from the mean calculation for this subanalysis only. Consequently, participants with a single plasma sodium determination were excluded. When the initial plasma sodium concentration was <135 meq/l (n = 95; Table 4), the mean plasma sodium concentration was <140 meq/l in 99% of participants. When the initial plasma sodium concentration was >142.5 meq/l (n = 238), the mean plasma sodium concentration was ≥140 meq/l in 64% of participants. Conversely, when the mean plasma sodium concentration was <135 meq/l (n = 67), the initial plasma sodium concentration was <140 meq/l in 94% of participants, and when the mean plasma sodium concentration was >142.5 meq/l (n = 63), the initial plasma sodium concentration was ≥140 meq/l in 89% of participants. Although the categories are not equal sized (i.e., not quartiles) and were created based on clinical definitions, the concordance of category assignment for the initial versus mean plasma sodium determination was 55.0%. Results were similar for MZ twins only and for DZ twins only and after excluding twins who had been exposed to medications known to impact systemic water balance (Supplemental Tables S13–S19).

Because the central tendency of the means of individuals’ plasma sodium concentration was greater than that of all determinations from all participants (Fig. 2), individuality of the plasma sodium was assessed in another way. It was possible that a low number of sodium determinations per participant rendered those participants’ means especially susceptible to the impact of one or more relative outlier determinations (e.g., through laboratory error). Therefore, the effect of the number of sodium determinations on the difference between an individual’s mean plasma sodium concentration and the overall sample mean sodium concentration was assessed. If there was no individuality to the plasma sodium concentration, then as the number of determinations increased, the difference between each individual-level mean and population mean should approach 0. In the VET Registry cohort data, the mean distance of the individual-level means from the population mean remained relatively constant irrespective of the number of plasma sodium determinations (Fig. 5A). The paucity of data points among the highest numbers of sodium determinations led to broader confidence intervals over this part of the range. This relationship is also shown in Fig. 5B, which shows the percentage of participants whose mean plasma sodium concentration differed from the population mean by ≥1 meq/l, after stratification on number of sodium determinations. The mean plasma sodium concentration of a participant who had undergone 40–49 or 50–59 sodium determinations was as likely to deviate from the population mean by ≥1 meq/l as that of a participant who had undergone only 0–9 plasma sodium determinations.

Fig. 5.

Individuality of the plasma sodium concentration. A: individual-level mean plasma sodium concentrations (n = 2,370) were calculated and subtracted from the population mean for plasma sodium concentration; the absolute difference of this quantity was then expressed as a function of the number of sodium determinations per participant. Shown is the LOESS regression line with 95% confidence intervals. The difference remains between 1 and 2 meq/l, irrespective of the number of determinations, consistent with individuality of the plasma sodium concentration. B: data from A expressed as the proportion of individual-level data (n = 2,370) for which the absolute difference between the individual-level mean and population mean was ≥1 meq/l. The proportion (y-axis) was approximately 0.5–0.6, irrespective of the number of determinations (x-axis). C: simulation (Simulation-1) of expected outcome assuming that all individual-level participant means were equal to the overall sample mean (see methods). Note that the absolute difference from the overall sample mean dropped off markedly beyond the data point representing 0–9 determinations/participant. D: simulation (Simulation-2) of expected outcome assuming individuality, i.e., that participant individual-level means were not equal to the overall sample mean. Note that this hypothetical distribution closely mirrored that seen with the actual study data in B.

To aid in interpretation of this analysis, data were simulated under two scenarios (see methods). In simulation 1, the individual-level means of the plasma sodium concentrations were held identical to the overall sample mean, that is, there was no individuality. In this hypothetical scenario, the proportion of participants exhibiting an absolute difference between the individual-level mean and population mean of ≥1 decreased rapidly as the number of determinations increased (Fig. 5C). For simulation 2, individual-level means were permitted to vary with respect to the overall sample mean (i.e., individuality was present). Data generated under simulation 2 (Fig. 5D) closely paralleled the actual VET Registry data (Fig. 5B). Moreover, the fit of a linear model to the study data substantially improved when an additional term representing individuality of plasma sodium was added (Supplemental Table S20).

DISCUSSION

The concentration of plasma sodium (along with its counterbalancing anions) is the principal determinant of plasma osmolality and the clinical index of systemic water balance. The present analysis using data obtained in a large study of twin pairs demonstrates, for the first time, that plasma sodium concentration is substantially heritable in individuals of European ancestry [0.41 (95% confidence interval: 0.34–0.47) and 0.49 (95% confidence interval: 0.43–0.54) when participants with only a single sodium determination were excluded]. We previously showed heritability of the plasma sodium concentration using multigenerational family-based cohorts reflecting a range of ethnicities (29); however, heritability was nearly absent in the two cohorts of European ancestry (0.07 in the Framingham Heart Study and 0.11 in the Heredity and Phenotype Intervention Heart Study of the Amish). In our earlier analysis, stratification on sex yielded higher heritability estimates among women; however, the VET Registry cohort comprises men only. We again demonstrated heritability of the plasma sodium concentration in African-Americans (0.36, 95% confidence interval: 0.17–0.52). Heritability was comparable when data from the entire twin cohort were used (i.e., without stratification on race) and when the potential influence of water balance-affecting medications was excluded. This concept of heritability is complemented by a graphical depiction of within-twin pair correlation of mean plasma sodium concentration (Fig. 4).

Visscher et al. (27) noted that “high heritability means that most of the variation that is observed in the present population is caused by variation in genotypes.” However, this does not imply the absence of environmental effects upon the phenotype. An example is the variable of adult height, among the most highly heritable of all traits [heritability: ~0.8 (30)], being influenced over generations by the relatively uniform environmental impact of improved nutrition (27). In general, morphological traits (heritability: 0.25–0.8) are more heritable than so-called “fitness” traits (heritability: ≤0.3) (27), and osmotic set point (i.e., plasma sodium concentration) is an example of the latter. As an environmental influence, water intake is a main determinant of plasma sodium concentration; however, we infer that access to water is neither limiting nor dissimilar among subgroups of the present VET Registry cohort. For perspective, the observed heritability of plasma sodium concentration of 0.41 in VET Registry participants exceeded that of plasma concentrations of creatinine [0.29 (7)], cystatin C [0.35 (21)], and magnesium [0.3 (16)] in other cohorts.

Previous family-based studies have shown insignificant or only very modest heritability of the plasma sodium concentration in Caucasian populations (15, 16, 22). A recent family-based study of 1,128 Swiss participants observed heritability of 0.12 with confidence intervals of approximately 0.02–0.26 (17). Two small twin studies from Australia and Sweden suggested mainly environmental influences on plasma sodium concentration (20, 28). The more recent study from Nilsson et al. (20) estimated heritability of the plasma sodium concentration at 0.22 or 0.28 (depending on the model); however, the confidence intervals in this small study were sufficiently large as to encompass a heritable contribution of 0. A larger Danish study demonstrated a within-pair correlation of plasma sodium concentration in MZ twins that exceeded that of DZ twins; however, participants were elderly (age: 73–95 yr) and 44% of the participants had a plasma sodium concentration below the population reference range (2). This group estimated a genetic contribution in men at 0.29 (0.03–0.52), again with extremely wide confidence intervals. After adjusting for diuretic use, which affected 20% of study participants, and for variables germane to other phenotypes investigated, the heritability estimate was 0.27 (0.00–0.50).

A number of key differences likely led to the unmasking of heritability among European ancestry in the present study. First, this is by far the largest twin study to address phenotype, affording greater statistical power. Second, and unlike our prior large multigenerational cohorts, the degree of genetic relatedness among participants was higher (i.e., all were twins and most were MZ twins). Third, plasma sodium concentration values confounded by severe hyperglycemia or abnormally low GFR were excluded. Fourth, the plasma sodium concentration was transformed for the independent (albeit subtle) osmotic effect of plasma glucose concentration. Fifth, the present study is the first to refine the phenotype through the incorporation of data from multiple plasma sodium determinations per participant. Although plasma sodium concentration is individual, it is imperfectly so, and determining a mean from repeated determinations over time potentially provides a more accurate metric of individual-level water balance. This factor was likely important because when only “initial” plasma sodium determinations were used as the phenotype (in contrast to the mean plasma sodium concentration), heritability estimates were considerably lower [i.e., 0.24 (0.18–0.31); data not shown], and when the analysis was restricted to participants with two or more plasma sodium determinations, the heritability estimate was even higher at 0.49. One caveat in interpreting this distinction is that all data in the present study are “real-world” clinical laboratory data obtained in the context of participants seeking clinical care. Therefore, initial plasma sodium concentrations were potentially obtained in the setting of illness or a clinical change and may have been less reflective of baseline water balance than the mean of all determinations. This caveat is also one of the important strengths of these observations, that the phenotype was (repeatedly) assessed outside the artificial confines of a research environment and hence the result is potentially applicable to other patient care data sets.

In this investigation, we also further explore the individuality of the plasma sodium concentration. Older literature suggested that, within the range of normal values, the plasma sodium concentration was not individual (for a review, see Ref. 31). In contrast, our recent study using clinical plasma sodium concentration data from two large health plans over a 18-yr interval supported individuality of this variable (31). That is, an individual’s plasma sodium concentration may reproducibly differ from the population mean, residing, for example, toward one extreme of the population distribution. VET Registry cohort data were used to corroborate and further refine this phenomenon. The initial plasma sodium concentration obtained in the electronic health record was predictive of the mean of subsequent determinations (Table 4 and Fig. 4). This suggests that a single low plasma sodium concentration, for example, is meaningful in that it is potentially predictive of persistent hyponatremia.

Individuality of the plasma sodium concentration was seen irrespective of the number of sodium determinations obtained (Fig. 5). If plasma sodium concentration determinations were stochastic and each participant’s true underlying plasma sodium concentrations were identical to the population mean, then as the number of determinations for any one participant increased, the difference between the individual-level mean and population mean should asymptotically approach 0. However, this was not seen; the results shown in Fig. 5A clearly show the persistence of a difference, irrespective of the number of plasma sodium determinations obtained over the multiyear study interval. Moreover, this data distribution was recapitulated by a simulation that assumed individual-level means that differed from the population mean (Fig. 5D). These results further support the individuality of the plasma sodium concentration.

Although the VET Registry cohort consists of older men, and a number of exclusions were applied to minimize the impact of variables independent of osmotic setpoint (i.e., for hyperglycemia, abnormal GFR, or use of medications known to impact water balance), this investigation in a large cohort of Vietnam Era twins demonstrates substantial heritability of plasma sodium concentration among those of self-identified European and African ancestry. Moreover, it further establishes the individuality of this important clinical phenotype that is reflective of systemic water balance.

GRANTS

This work was supported by United States Department of Veterans Affairs Grant I01BX003449 (to D. M. Cohen). The United States Department of Veterans Affairs has also provided financial support for the development and maintenance of the Vietnam Era Twin Registry.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

A.K.T., J.G., N.L.S., and D.M.C. conceived and designed research; A.K.T. performed experiments; A.K.T., A.M.K., J.T., K.P.M., C.H.L., C.W.F., J.G., and D.M.C. analyzed data; A.K.T., A.M.K., K.P.M., C.H.L., C.W.F., J.G., N.L.S., and D.M.C. interpreted results of experiments; A.K.T. and D.M.C. prepared figures; D.M.C. drafted manuscript; A.K.T., A.M.K., K.P.M., C.H.L., J.G., N.L.S., and D.M.C. edited and revised manuscript; A.K.T., A.M.K., J.T., K.P.M., C.H.L., C.W.F., J.G., N.L.S., and D.M.C. approved final version of manuscript.