One physiology does not fit all: a path from data variability to “physiogenetics”?

Abstract

Data variability is a costly complication of biomedical experimentation because the same experiment must be repeated a sufficient number of times so that the sample mean becomes a credible representation of the entire population. Since sampling is ideally done randomly in populations normalized for environmental and genetic backgrounds, data variability is viewed as a purely statistical issue reflecting the distribution in the population and captured as the standard deviation of the sampled data. The factors contributing to data variability are not analyzed by statistical methods; for want of a better explanation, data scatter is simply attributed to random noise and/or methodological limitations. In this commentary, evidence is discussed that documents an important role of interindividual biological diversity as a cause for data variability based on studies in which repeated sampling in the same individual permitted statistical comparisons between individuals in the same sample. Significant differences were found for proximal fluid reabsorption and plasma renin concentration between sample means of individuals of the same population. Furthermore, arterial blood pressure varied significantly between individual mice independently of strain and sex. Recognition of the extent of interindividual variability has important implications for data reproducibility, data collection, and data presentation in physiological research. Such nonrandom data variability may have different causes, but DNA modifications by genetic or epigenetic mechanisms could generate phenotype variants without being associated with disease symptoms. Exploration of the heritability of phenotypical diversity in physiology may be defined as “physiogenetics,” and it would thus be the physiological corollary of pharmacogenetics and pharmacogenomics.

the purpose of this commentary is to document that interindividual phenotypical differences or biodiversity can be the cause of data variability independently of identified environmental and genetic background effects, and to discuss some of the practical and scientific implications of biodiversity for physiological research. Biodiversity within a species can be as obvious as differences in height, hair pigmentation, or intelligence, but in most cases it is less apparent and therefore more difficult to distinguish from random noise. In fact, in the hypothesis-driven approach that dominates experimentation in physiology, biological diversity, aside from sex dimorphisms, is usually not discussed as a major contributor to data dispersion. Nevertheless, full exploration of all causes of data variability would seem important since such variability is arguably the main reason why physiological research is often slow, expensive, and fraught with uncertainties.

For a given difference between sample means, the outcome of any statistical analysis depends upon the data dispersion expressed as the standard deviation. While standard statistics can safely predict the probability of a hypothesis to be true or false, they do not address the causes for the data dispersion, including the possibility of interindividual diversity. There is recognition in the statistical literature that such diversity can contribute to dispersion of data, but no easy way is offered to decide whether a given value is different from others in the sample because it comes from a diverse subgroup in the population. It is understood that a number of individual and environmental traits including sex, race or strain, age, time of day, food, housing conditions, and others are potential causes of data variability, but the topic of this discussion is the variability that remains after stratification of these rather obvious sources. In theory, biodiversity as a cause for data variability in physiological research could be statistically established if a variable was determined a sufficient number of times in the same individual. One could then consider these data as a sample that could be used for comparisons with samples from other individuals in the same cohort of subjects. Unfortunately, the majority of physiological studies are not designed to permit such an analysis. In experiments in which only one data point is obtained from an individual (such as one tissue lysate for Western blotting or a single isolated, perfused tubule for transport measurements), there is simply no statistical way to distinguish this data point from others in the sample no matter how different it is from the sample mean. In the following, I will describe three instances in which recent data sets from our laboratory suggest that the assumption of random variability of data from individuals in the same population is almost certainly incorrect.

Proximal fluid reabsorption.

Fluid reabsorption along the proximal tubule can be determined by complete and timed collections of tubular fluid and measurement of the rise in the concentration of a nonabsorbable and freely filterable marker like inulin or iothalamate. The example shown in Fig. 1A shows such measurements in 19 tubules from 4 wild-type mice selected from a recent publication from our laboratory (6). The sample mean for the TF/P_iothalamate ratio was 1.79 ± 0.25 SD with the extremes being 1.43 and 2.16, indicating variability of proximal fractional fluid reabsorption between 30.1 and 53.7%. In Fig. 1B, the same data points are separated into four groups according to the individual of origin. Individual data scatter around means of 1.51, 2.07, 1.61, and 1.92 with consistently lower SD values of 0.09, 0.08, 0.16, and 0.15. Furthermore, as indicated in Fig. 1B, means of individuals 1 and 3 were significantly different from individuals 2 and 4 (ANOVA). Thus, while there is residual, presumably random data scatter with regard to all individuals, some of the overall variability appears to be nonrandom, being the reflection of true differences between individuals of the same population. We have made the same observations in a different data set from an earlier study (Fig. 1, C and D) (2). Furthermore, early micropuncture studies in rats and dogs from an era when data from individual animals were often published in tabulated form also revealed comparable interindividual differences as part of the cause for the data scatter in the entire population (1, 3, 4).

Fig. 1.

Proximal fractional reabsorption (FR) expressed as TF/P_iothalamate (P_io) ratio in 2 data sets selected from previous publications (2, 6). FR is equal to 1 − (1/TF/P_io). A and C: scatter graphs of individual values from 4 mice (n = 19 in A, n = 20 in C). B and D: scatter graphs of the same data shown in A and C with attribution to animals of origin with animal number given on the x-axis. Lines above data points are drawn between data that are significantly different from each other as tested by ANOVA. Horizontal and vertical lines in all graphs indicate mean values and SD.

Plasma renin.

Measurements of plasma renin concentration (PRC) are usually performed by collecting a blood sample and determining the rate of ANG I formation in the presence of saturating levels of renin substrate. In general, a single blood sample is obtained per individual, and values from several individuals are grouped to describe the population mean. PRC values vary considerably between individuals as shown in Fig. 2A, representing measurements in 13 individual mice (5). However, in this particular series, blood collections and PRC measurements were repeated in the same individuals on five subsequent days. All measurements are shown in Fig. 2B, and an overall mean value of 191.6 ng ANG I·ml⁻¹·h⁻¹ ± 84.7 SD was obtained (n = 5 × 13). Averaging results from individual mice revealed that part of the overall variability was due to interindividual differences: SD values were consistently lower for individual data means (31.4, 44.0, 81.6, and 48.1 for the four selected individuals shown in Fig. 2C), and ANOVA comparisons between individual means yielded 13 significant differences within the same population, of which 4 are shown in Fig. 2C.

Fig. 2.

Plasma renin concentration (PRC) in male C57BL/6 wild-type mice. A: scatter graph of individual values (n = 13). Each individual value is the average of 5 measurements/animal. B: scatter graph of all single determinations (n = 65). C: scatter graphs of measurements in 4 selected individuals of 13 in the cohort. Lines above data points are drawn between means that are significantly different from each other as tested by ANOVA. In addition to the 4 significances indicated, there were 9 more when all 13 animals were compared with each other. Horizontal and vertical lines in all graphs indicate mean values and SD.

Arterial blood pressure.

The use of telemetry probes has become the state-of-the-art method of determining arterial blood pressure in conscious small animals like rats and mice. Data collection is handled differently in different laboratories, but in principle an infinite number of data points can be obtained for any individual in the sample. The routine protocol in our laboratory is to collect data every 2 min for 10 s, and these 30 data points are averaged to yield an hourly mean value. When hourly means are averaged for the 12-h daytime and 12-h night time periods, the resulting averages are based on 360 data points in each individual or 1,800 data points when this is done for 5 consecutive days (Fig. 3A). Clearly, there are enough data points to use each individual for a statistical comparison with other individuals in the same sample. Figure 3B shows that the use of hourly values for five consecutive night time periods permits the identification of significant blood pressure differences between individual wild-type mice with an identical C57BL/6 background.

Fig. 3.

Mean arterial blood pressure during the 12-h night time period in male wild-type (WT) mice with C57BL/6 or mixed genetic background. A: mean arterial blood pressure in 5 C57BL/6 mice averaged from hourly means during the 12-h periods of 5 consecutive nights. B: display of individual mean hourly blood pressures in the same 5 mice (n = 60/individual). Numbers on the abscissa indicate animals. Lines above data points in B are drawn between data that are significantly different from each other as tested by ANOVA. Horizontal and vertical lines in all graphs indicate mean values and SD.

Conclusions

The evidence presented in these three examples indicates that biological diversity can contribute importantly to data variability in populations stratified for controllable environmental and genetic background effects. One suspects that these three observations are not exceptional, and that many other phenotypical differences between individuals would be found if experiments were planned or interpreted accordingly. The identification of constitutional differences between individuals in a population thought to be homogeneous has a number of practical and important implications. First, the existence of interindividual differences could be an important contributor to nonreproducibility of data sets, at least in a quantitative sense, since the sample mean could vary substantially dependent on the “physiotype” of each individual selected for sampling. If the number of individuals is small, say less than 10, it is likely that two independent experimental series would yield different mean values. When these mean values were used for statistical comparisons, significances and conclusions could change. In the presence of biodiversity, there is no short-cut around large sample sizes if physiological quantities are to be comparable between laboratories and therefore reliable beyond a reasonable doubt. Second, for a meaningful statistical analysis of micropuncture and similar data sets (Fig. 1), the critical number of observations in a sample must be the number of individuals, not the number of tubules. Determination of independence of sampling should be taken seriously when the data dispersion raises the suspicion of phenotype individuality. To assess biodiversity, the experimental design should consider multiple determinations of a physiological end point in a given individual if possible. Third, data selection and dismissal of so-called “outliers” is a problem for which no easy solution is at hand. Tests for outliers such as Chauvenet's criterion or Grubb's test allow data exclusion when the probability is very low that a value was sampled from a normally distributed population. However, biodiversity may cause a distribution to be non-Gaussian. Thus, in the absence of a rational explanation, such as an identified experimental error, the exclusion of any data seems hazardous since any abnormally high or low value might be a characteristic of an individual and might become less and less of an outlier with increasing sample size. Finally, results from physiological experiments are most commonly presented as bar graphs depicting mean values and their standard errors. One suspects that the use of this type of graph is chosen in part because it optically minimizes data dispersion. Although data dispersion can usually be reconstructed from error bars, data shown as bar graphs can almost always be displayed as scatter graphs, where individual data points provide immediate visualization of data dispersion.

The causes for physiological individuality in the cases discussed above and in similar cases are unknown. If phenotype variants such as high or low plasma renin or high or low proximal fluid reabsorption were shown to be heritable traits, one could argue that genetic polymorphisms or epigenetic DNA modifications contribute to the creation of individual physiotypes. The exploration of the genetics of interindividual physiological diversity may be defined as “physiogenetics,” and it may be seen as complementary to the burgeoning fields of personalized or precision medicine, pharmacogenetics, and pharmacogenomics. In fact, physiogenetics may be the underlying cause of differences in responsiveness to pharmacological and other interventions. For example, constitutional differences in plasma renin concentration like those described above would be predicted to be associated with response diversity to angiotensin-converting enzyme inhibitors and angiotensin receptor blockers. In general, a variant of a target that causes alterations in the efficacy of a drug may well be expected to be associated with an altered physiological phenotype as well.

Although it should be obvious that one physiology does not fit all, a new appreciation of physiological biodiversity and investigations of its underlying causes, especially of physiogenetics, may lead to the identification of important functional modifiers within the spectrum of normality.

GRANTS

This work has been supported by the intramural program of the National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health, Bethesda, MD.

DISCLOSURES

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

AUTHOR CONTRIBUTIONS

Author contributions: J.S. provided conception and design of research; J.S. performed experiments; J.S. analyzed data; J.S. interpreted results of experiments; J.S. prepared figures; J.S. drafted manuscript; J.S. edited and revised manuscript; J.S. approved final version of manuscript.