12Overview Thoughts on Genetics: Walking the Line Between Denial and Dreamland, or Genes Are Involved in Everything, But Not Everything Is “Genetic”

Weiss KM.

Publication Details

“… You know what M. de Talleyrand said, ‘If you don’t know how to play whist you are laying up for yourself a desolate old age.'”

Honoré de Balzac, Lost Illusions

The word “genetic” is used in different ways by different people or by the same people at different times. There are three basic meanings that should be kept clear. The application of “genetics” to social survey problems depends on which meanings are being used. Indeed, the concept of a gene itself is much more complex than protein-coding, a fact that adds underappreciated nuance and complications to proper genetic studies.

“Genetic” can refer to mechanism, that is, what genes do and how they work, a kind of stereotypic biology. When we refer to the fact that p53 is a gene involved in cell cycle regulation in the development of normal tissue, we assume some generality in “normal” people or a standard assay system. “Genetic” is also a population concept that refers to the correlation between phenotype variation and inherited variation in populations. Inherited variation in the p53 gene is associated with abnormal tissue architecture and cancer. Finally, “genetic” refers to somatic change that occurs postinheritance—due to mutation in individual body cells—when the changes are inherited by their descendant cells during development and later life (including aging).

But somatic changes that don't occur in the germ line are not inherited from parent to offspring. Genetics in the form of somatic mutation generates variation in mechanism within individuals, and it is fundamental in the etiology of cancer and a few other diseases, probably including cognitive aging, and may have much greater importance than has yet been documented (Weiss, 2005a, 2005b). Somatic mutation in p53 affects individual cells that can transform to found a clone of cells that constitute a life-threatening cancer. Some somatic changes can in various ways affect gene expression rather than gene sequence itself, and these are known as “epigenetic” effects. A number of epigenetic mechanisms are known and new ones are rapidly being discovered.

The “genetic architecture” of a biological trait refers to the number of genes that contribute to it, the way those genes interact with each other and with the environment, their relative contribution to the trait, and the role of their variation on the trait. It involves all three aspects of genetics. The genetic architecture of traits in multicellular organisms, including human traits of any complexity, is far from completely known. Indeed, “complex” is in the eye of the beholder, and even “simple” traits turn out not to be so simple on close inspection (Scriver and Waters, 1999). Most human genetic epidemiology is largely black-box genetics that searches for statistical correlation between inheritance and trait, initially without knowing (or at the discovery stage even caring) about its mechanism. The situation is made more complex by the expanding definition of “gene” to recognize many functions beyond protein-coding, and these are still being discovered.

The genetic architecture of any trait is the product of its evolution. Evolution is the population-historic process that generates the genomes that construct or affect phenotypes. Evolution involves population size, mating patterns, chance, migration, geographic and social distribution, differential reproduction, and the like. The variation that results depends on mutation rates, the size of the mutation target (number and length of relevant genetic units in the genome), and the effects of reproductive fitness (natural selection). These factors are mediated through demographic history. Evolution helps account for and predict the general trait patterns that we see today, and because humans are a globally distributed species, it is especially useful in explaining the amount and geographic distribution of genetic variation affecting human traits.

The 20th century began with the recognition of Mendel's work on the particulate (discrete-factor) nature of inheritance. Discretely varying (e.g., presence/absence) traits initially seemed inconsistent with the more prevalent quantitative variation (e.g., height, weight) seen in nature and that seemed to be the working material for evolution. But early in the 20th century it was realized that the combined action of many independently segregating particulate genes could, through combinatorial contributions, lead to a quantitative phenotype (to which environmental variation contributes as well). The key fact of that realization (as also noted elsewhere in this volume) is many to many causation: many different genotypes generate essentially the same phenotype. The genes, known generally as polygenes, have been modeled as a homogeneous mix of infinitely many genes, each with two alleles (variant sequence states) of equal frequency, that additively contribute dose-like effects to the trait. The genetic variation in such a model easily generates the typical Gaussian-like trait distributions in a population and the trait-value correlations among relatives, leading to a consistent, complete theory of both variation and inheritance.

This was a fundamental finding, and it can be a lesson difficult to digest for anyone hungering for simple genetic solutions to their favorite traits. It implied that the genetic contributions could not be individually identified. Nonetheless, the basic ideas have been systematically confirmed with ever-increasing depth of understanding during the 20th century. Gene mapping (observational and epidemiological studies to identify genes affecting a trait of interest) and experimental studies have documented these basic points. However, recent genetic and evolutionary research has added important and well-replicated characteristics to the general picture, which dehomogenize polygenic causation to some extent. A considerable amount of data and evolutionary theory show that the distributions of (1) allele frequency, (2) effects of individual alleles on traits they affect, and (3) the effects of alleles on evolutionary “fitness” (natural selection) are skewed. Most alleles in a given gene and population of inference have low frequency, small effect on traits, and small effect on fitness. Since the frequencies of the alleles at a locus must sum to 100 percent, there cannot be too many with high frequency. Most known high-frequency susceptibility alleles have small individual, probabilistic effects on risk and/or risks substantially mediated by the environment, broadly defined and usually only partly known. There is usually a substantial tail of generally rare, strong phenotypic, and/or major fitness effects. The clearest segment of this is the aggregate of lethal mutations, in which the effects are clear and essentially deterministic. The practical upshot is that many complex traits can be called “oligogenic” or “multilocus,” in that embedded in a polygenic background are at least some specific loci with alleles frequent enough and/or effects strong enough that standard methods can detect them with adequate sample design—and new instances are continually being reported. But what fraction of genetic architecture comprises this low-hanging fruit in relation to common disease risk is a contentious question. The definition of “common” is a moving target in the eye of the beholder, and a potential source of bias in interpretation, reporting, or accepting or rejecting study results. Such tractable alleles attract scientific, commercial, and media attention, can satisfy vested interests, and can drive sampling and analytic strategies to find them. There are, of course, clear examples where alleles with substantial effect on disease are at relatively high frequency.

Complicating matters is that genes typically harbor tens to hundreds or more alleles. Most variants are individually rare, so that a given copy of a gene may differ by only one or a few variants relative to the human genome reference sequence. Thus, polygenic contributions involve many genes as well as many alleles within each gene, each with its own population-specific frequency. And because we are diploid, the effects of this variation are genotype-dependent as well.

Because each allele arises by new mutation in some specific individual, most rare alleles are recent and geographically localized, and the more frequent the allele, the older and more geographically dispersed it tends to be. The frequency and/or presence of a given allele varies from place to place. The spatial gradient of frequency is affected by many stochastic demographic factors, but it generally forms a correlation between geographic distance and genetic difference, which applies to individuals and hence to populations. These differences are important for medicine because patients in different populations have different distributions of variation related to a given disease. But the pattern of difference or similarity is probabilistic and subtle, and distant individuals can be more closely related (more similar) at a given gene than individuals from the same population (Witherspoon et al., 2007).

The frequency of an allele and its effect on a given trait are commonly statistically confounded. This is because traits—like disease or behavior— are often defined in terms of deviations from the mean (from normal), and so they can be rare almost by definition: if the frequency of an allele contributing strongly to such a state were common, the mean would be near that allele's effect. An allele could be common in the population and still have a major effect if it results from diversifying selection, such as seen in the immune system, frequency-dependent selection that reduces fitness of any allele that becomes too common, or other demographic or selective situations of that type. Generally with such selection the trait distribution in the population is not unimodal. Sex and sex-related traits are examples.

However, we don't always measure a trait strictly from the mean but sometimes use some desirability standard, such as healthy versus diseased. Hypertension refers to blood pressure elevated enough to pose a presumed universal risk factor for cardiovascular disease per se, not simply relative to the average in the population. In other words, everyone in the population is presumed to be at higher risk with higher blood pressure. Such traits can be common, as for example obesity is in the United States, and certainly genetic risk for obesity could be common. For example, it might be that archetypal human metabolism deposits fat and calcium and so on, such that excessive consumption would elevate blood pressure in anybody regardless of genotype. Such a situation may involve genetic mechanism but not be mappable in observational research: if we all share the genetic mechanism, there may be insufficient or genetic variation contrasts that are too heterogeneous for statistical inference. This usually points to exogenous causal factors, like diet, and does not mean the trait is genetic in the sense that variation in genetic susceptibility is a major cause of variation in risk. Obesity and hypertension are extensively studied traits that are still not well accounted for genetically, but they fit the general many-to-many nature of the polygenic model.

Whether or not a variant affects fitness in the health sense, if it has had little effect on selective fitness (that is, on differential reproduction in the natural selection sense) its frequency in a population depends on the chance aspects of reproducing during its sojourn on the population, and hence it can have any frequency. It can be common and yet manifest major harmful effects if there has been a rapid change in environment, so that what was once benign or even helpful is deleterious in today's environments. This may include many chronic diseases of modernization, like heart disease and diabetes. The new harmful effects can, but need not, also confer a new selective disadvantage to the allele.

These facts are all relevant to the objectives of this volume, and other chapters in this book consider them. They are entirely consistent with what we know about evolution and the mechanistic as well as demographic aspects of genetics. They are consistent with the tens of thousands of papers that have been published on the genetics of human disease, both of single-gene and of complex—multigenic—traits. They are consistent with what we know about behavior as well. These facts are easily documented on the web (e.g., OMIM or PubMed at http://www.ncbi.nlm.nih. gov, and many other places).


Strong, rare effects segregate in families according to the principles Mendel discovered in his pea plant experiments, because like Mendel's carefully selected traits, the effects are closely tied to the action of individual genes. They are replicable among relatives within and between families that carry and transmit them. They are the flags by which single-gene diseases were mapped (that is, the gene's chromosomal location identified so the gene itself could then be found in various ways). And they are the flags by which individual genes contributing substantially to quantitative traits (known as quantitative trait loci, or QTL) have been found. There are hundreds if not thousands of success stories, and mapping results based on a variety of study sizes and designs pour forth new results daily, documenting the elements of causal genetic architecture that I have described (e.g., Wellcome Trust Case Control Consortium, 2007). Usually, however, once the gene is found, many more mutations are discovered whose effects are too small to have flagged the gene in a mapping study, but the aggregate of these effects can be an important, subtle source of trait variation in the population (Weiss and Buchanan, 2003).

However, mapping studies typically account for only a fraction of the familial correlation or aggregation of risk. As one moves down the effect-size scale from strong toward weak, or up the frequency scale from rare toward common, the detectability of the genetic signal often reduces, due to a multiplicity of competing factors of similar strength or frequency, measurement error, and environmental factors. Replicability also declines for most of the lesser-effect signals. The farther along this spectrum one goes, the more tightly designed a study may have to be to characterize or even detect their individual effects, because they may be too small or too embedded in heterogeneous complicating factors like environments. A typical finding of the dehomogenization of polygenic risk concepts, referred to above, is that the more restricted or precisely focused the phenotype definition, the more “mappable” it is to identifiable loci or alleles, and vice versa. But then the results apply to only a small subset of the overall trait. This is the case for such behavioral traits as autism, schizophrenia, even Down syndrome and the like.

A potential complicating factor is a third aspect of genetic causation mentioned earlier, that of somatic mutation. We don't yet know how important this may be, but somatic mutation is probably responsible for cases of behavior disorders related to epilepsy and may be a sleeping giant in behavioral biology that will be silent until attacked explicitly, which is difficult to do with today's technologies (Weiss 2005a, 2005b). The scope of epigenetic effects, including those on behavioral traits, is probably much greater than has been thought (Wong, Gottesman, and Petronis, 2005; Petronis, 2006).

The effects of a given allele (which may be defined as a single nucleotide variant or a pattern of them on a single chromosome, known as a haplotype) usually have large variances. We know from many kinds of data, including from experimental animals, that this variance includes the effects of variation in the genomic background (the rest of the genome) of individuals with the same test variant, as well as the environmental context.

Most of our functional genome involves genes whose action is to regulate the expression of other genes. Consequently, much or most genetic variation involves context, timing, and level of expression. These are quantitative aspects of gene function. Perhaps especially for chronic disease or age-related traits, genotype-specific effects are manifest as dose-response effects on the age-specific hazard function, rather than proteins made dysfunctional by mutations that change their amino acid sequence.

Overall, to varying but not precisely known extents, about 30-60 percent of the variance in most complex traits, including chronic disease, generally called the heritability is due to aggregate effects of genetic variation that can be estimated statistically by familial aggregation of risk, monozygotic versus dizygotic twin contrasts, and other common measures. Family-based estimates are probably somewhat inflated because of undetected confounding between environmental and genetic correlations (various study designs, such as adoptee studies, are often suggested to try to disentangle these), so that heritability is a relative measure applicable to a particular sample or population. For example, the common practice of doing genetic studies on multiply affected families may help mapping significance, but at the expense of overestimates of risk (e.g., Begg, 2002; Terwilliger and Weiss, 2003). Nonetheless, despite the many important technical issues associated with heritability as an indicator, it seems indisputable that the overall genetic contribution to most traits is not trivial and often substantial.

However, the substantial heritability of most traits can lead to a kind of mirage of genetic simplicity. Variation in mapped, specifically identified genes usually accounts for only a fraction, usually a small fraction, of the total amount of familial aggregation (and a consequently smaller amount of overall variation). This is because most of the heritable variation appears to be in the form of polygenic effects, whose aggregate is comprised of effects of numerous genes that are individually too small to be reliably or replicably mapped, or they will be mappable in one but not other populations or samples, because while genetic in the sense of inherited variation, in each individual they will be due to a different set of particular genetic variants.

This substantial fraction of apparent genetic etiology is an epistemological mirage in the sense that its individual components cannot be identified with practicable samples or study designs. To a substantial extent, cases of the trait may be causally so heterogeneous, for reasons described above, that reductionist approaches may be impossible in principle, since they are based on the replicability of observation. These instances of the trait can be characterized as noninferrably genetic, that is, putatively caused by genetic variation but in a way that can't be inferred with practicable or even achievable samples or sample sizes. Causal assertion in such instances may perhaps be true but may border on the mystic.

Note also that heritability refers to the fraction of trait variance that is statistically accounted for by genetic variance. How they do it is an entirely separate question. Thus h2 = 0.40 doesn't imply that 40 percent of cases are caused by a gene (or even by an aggregate of genes) while the remaining 60 percent have nothing to do with genes.

Matters are made more complex—and, importantly, more ephemeral—by the “environmental” contributions, a term I put in quotes because we typically know very little about them except in broad or aggregate terms (e.g., stress, excess dietary intake, etc.). This is a source of major frustration in biomedical genetics today, but it is the expectable product of evolution in which many genes contribute to complex biological phenotypes in contextually dependent ways, and natural selection is tolerant of variation (Weiss and Buchanan, 2003; Buchanan, Weiss, and Fullerton, 2006a, 2006b).

I've tried to illustrate these points heuristically in Figure 12-1. The genetic objective is to predict outcome from inherited genotype. Part A shows a symbolic DNA sequence with several alleles in fonts proportional to their inherent effect on some trait of interest (inherent effect is a rather epistemologically dubious notion itself, but we'll let it pass). Surrounding the sequence are representatives of the factors that interpose between inheritance at the time of fertilization and eventual outcome. Part B shows how these factors overlay each other during life and the consequent difficulty we should expect in reliably inferring the underlying sequence effects from the final phenotype. Each individual differs in these various components (e.g., their degree of transparency or grayness in this heuristic), at least to some extent, even if we somehow know what they are and measure them accurately. The immediate genetic epidemiological problem, the inverse of the causal process itself, is to infer the underlying genotype from a sample of outcomes, and from that to be able to make risk predictions. The figure shows metaphorically why it is problematic to think that genes code “for” complex traits.

FIGURE 12-1. How darkly through a glass?


How darkly through a glass? (Left). At the center, an inherited genotype with nucleotide font size proportional to an assumed inherent effect. Around this are several contributing factors that intervene between the genotype and the effect. (Right). An (more...)

A common strategy to pull the causal rabbit out of the uncertainty hat is meta-analysis, the aggregation of different studies. This approach is taken when existing studies vary widely in their conclusions or in their estimates of a putative risk factor, or it is thought that individual studies may have insufficient statistical power to detect a putative effect. Frequently, as the aggregate study size increases, the net overall estimated effect (e.g., relative risk) becomes smaller. While some substantial effects have been confirmed by meta-analysis, it must be recognized that there has been a continuing flow of meta-analyses finding that what initially seemed to be clear-cut risk factors turn out to have less or no effect. Along with causal and sampling heterogeneity, a likely reason for the latter outcomes is the bias in the design of samples that are responsible for the initial findings, referred to above. It remains difficult to find the rabbit, or to determine if it is the same rabbit in different hats. This makes the assessment of risk very problematic, even for frank disease, for which, unlike many aspects of social behavior, there is no controversy about the desire for risk assessment or intervention.


We do not yet know clearly where on the spectrum from monogenic to polygenic control the genetic components of complex traits lie, and this clearly varies among traits and to some extent among different populations because of their different population history.

There are two possible explanations for the frustration we have faced in attempting to dissect complex traits into their genetic and environmental causes or to identify the factors when they interact. Some think (or fear) that the traits of interest really do lie mainly toward the polygenic range of the etiological spectrum. Alternatively, others think (or hope) that such traits are more likely to be oligogenic (involve only a few major genes) and that the appearance of complexity is due to poor trait definition or identification of environmental factors that affect the phenotypic variance around a modest number of genotypes.

An important point worth repeating is that there are strong reasons to view the genetic contributions as environmentally contextual. The environment can be the physical, psychological, genomic, or ecological setting in which each genotype uniquely finds itself. It may be difficult or epistemologically impossible to separate out interacting factors, but we do have enough evidence from many known secular epidemiological trends that major changes in risk occur rapidly in populations whose gene pool is essentially unchanged, in traits with substantial heritability and even in specific genotypes that are among the strongest and clearest of genetic risk factors, such as the major mutations associated with breast cancer (e.g., Begg, 2002; King, Marks, and Mandell, 2003). When environments are not known, not measured, or change unpredictably, risk estimation on which prospective intervention is based may be more art than science.

It is a sign of scientific victory, not failure, that we basically understand the nature of the uncertainties I have described, even aspects of risk we know that we don't know, or perhaps can't know with current conceptual approaches. Could complex traits really be genetically simple but we're missing something? That seems unlikely to be the general case, after the diversity of approaches that have been taken to the genetics of such traits.

Science should be guided by this victory of understanding rather than decrying it as a source of failure or distress because it doesn't lead to a tidy reduction of causation to a manageable number of clear causal elements. Current knowledge raises the epistemological and methodological question of what to do in the face of a substantial fraction of polygenic control, rather than hoping, against the bulk of the existing evidence, that we can identify a large fraction of individual contributing genes with expensive technological fixes, such as more elaborate statistics, more extensive genotyping technology, and faster computers.


The substantial debate about the nature and extent of genetic control of complex traits becomes more heated when the subject is social or nonpathogenic behavioral traits. Peoples' views are inextricably tied to their social politics, whether that fact is stated or tacit. Many wish to deny that nature controls traits they would rather see as facultative, as manifestations of free will, or as avoidable by lifestyle and societal change. Genetic determinism is a threat to a person's inherent value, something history shows that society can and often does abuse.

Others take the opposite world view. Basically, they see the living world in selfish gene terms, as the genetically determined, materialistic, reducible results of inevitable molecular interactions. They view natural selection as prescriptive, so that our traits today have limited genetic variation and high genetic control, although usually offering politically correct denials that they are not being genetic determinists or essentialists. But since natural selection affects only genes that determine a trait, Darwinian arguments are inescapably deterministic.

Where between these views—one denying genetic involvement in social behavioral traits, and the other dreaming of a simple causal world— does the truth lie? From the mechanistic sense of “genetic,” behavioral traits, like any other traits, clearly involve genes. For behavioral traits, the mechanisms are likely to be very indirect, since genes don't talk, rape, shudder, cheat, or decide what pensions to invest in. But the same is true of diabetes, since genes don't eat, avoid exercise, etc.

In addition to mechanism, we also have every reason to assume that genetic variation affects variation in essentially every trait, social behavior included. This doesn't imply simple genetic architecture, however, and again, as with complex disease traits, behavioral phenotype correlations among relatives do not imply that environments or their nonlinear interaction with genes are unimportant, or the causal genes few. Each case will differ, and the central issues involve the degree to which the latter applies, and to which behavior, no matter what inherent “tendencies” we may wish to ascribe to it, is malleable by culture, just as we ask whether obesity is malleable by diet or exercise. The situation is clearly complex and heavily nuanced (Ryff and Singer, 2005). However, genetics is a particularly strong attractant in social behavior research these days, and it is important to try to understand why this might be.

Some nonscientific factors are at work, including subjective, sociopolitical, status, and desperational views (“nothing else is working”) that are widespread in behavioral and social science. Funding and other entirely human considerations lead genetics to have a strong gravitational attraction for research, not unlike a black hole in space.

From a strictly scientific viewpoint, reductionist causation is convenient, manipulable. That suits our technophilic, interventionist yearning for simple and hence reassuring explanations. A driving force is the familial aggregation of many serious pathological behavioral (psychiatric) traits, even when specific genes have not yet been found. Also, some specific genetic variation with statistical effects on the nature or risk of behavioral traits have been found, as mentioned by various chapters in this book. As with other disease traits, such findings fire enthusiasm for continued searching. But there is a tendency to extend these findings to nonpathological aspects of behavior—to search for genetic determinants of normal social interactions.

I think it is very instructive to recall that Durkheim and the other founders of social science knew that social structure is the emergent result of superorganic interactions, not specific genotypes (basically everyone can speak Chinese, eat with a fork, be a Hindu, etc.) and that social facts exist, change, and are to be explained mainly in terms of their natural level of organization, that is, in terms of other social facts. Those social thinkers were only beginning and certainly didn't get everything right, but they had the central idea.

One way to see the point is perhaps to recognize that just because tallness (and its genetic basis—after all, stature is 90 percent heritable) is correlated with basketball, it does little to explain basketball. But a technophilic excitement over discovering the “basketball genes” that might account for the extremes of height, and so truly could account for a fraction of what we know about contemporary basketball, would distract from the more challenging effort to understand the phenomenon itself. And assuming that some positions of today's style of basketball are optimally played by people with genes for tallness (in a rich nutritional environment) would ignore the facts that basketball originated and flourished when most people were short and tall people were gawky, and the game can be played by people in wheelchairs.

The founders of social science wrote to resist the growing late 19th-century reductionist zeitgeist to explain culture in terms of psychology, and they were right. It remains relevant to ask what the scientific rationale is that leads social scientists today to hunger yet again for rescue by aliens, in this case molecular rescue by genes, another reductionist approach that is too far below the emergent level of the phenomena to fully explain them, and a diversion from trying to fix the problems in social science itself.

Why would social scientists essentially abandon the notion that social facts are to be understood in their own terms and instead seek to reduce these facts to molecular terms? I think the attraction, when it is scientific rather than one of funding politics and the like, rests understandably on two basic, often implicit assumptions.

First, a human being develops from a single cell, and all the information in the cell that affects development is ultimately genetic. Children look like their parents, twins highly resemble each other, and rabbits never give birth to mice. Ergo, genes must determine traits. That's genes as mechanism. Second is the idea that evolution occurs through heritable variation, and since natural selection molds variation adaptively, what we are must be what nature intended us to be, which is controlled by genes.

These premises are substantial but not complete truths. Many factors affect development, most traits of importance to this volume develop gradually with age and are context-dependent, stochasticities of countless sorts demonstrably affect biological processes and organisms. Natural selection is far more tolerant of diversity than one would gather from the melodramatic accounts of selfish genes and ruthless nature that one sees in the media (and often in the literature itself). To a great extent, what behavior is, and evolved “for,” is assessing unpredicted and unclear situations and responding to them. Facultative responses cannot be too highly hard-wired. This should be especially true for humans, who are the social, facultative species par excellence. We should be trying to understand how that works, not to make it into its robotic opposite.


Relevant ethical issues go beyond the routine mechanics of informed consent, and these are especially complex and sensitive when applied to behavioral or social traits rather than outright disease. This is especially true when the causal epistemology is shaky. Genetic findings relate to the subjects involved but also give some information about their relatives. Topics discussed in this volume concern genetic contributions to behavioral traits that may be ultimately related to aspects of health but only indirectly so. Behavioral genetic findings also apply to those that the subject interacts with. Knowledge that you are an inherent gambler or intolerant to stress can be relevant to the lives of your affinal relatives, work cohorts, neighbors, and the like, not to mention those who might wish to exploit your characteristics commercially in one way or another. Let's look before we leap into prenuptial genetic testing to see who is genetically suited for the state of marital bliss, or to tolerate child abuse best (or provoke others to abuse them), or whether you'll be a liberal or conservative.

When social behavior affects health, it can be a legitimate concern of the National Institutes of Health, but genetic variation cuts many ways. It is usually thought that genetic variation responds to the social context, but to a considerable extent it may be that genetic variation contributes to the construction of that context (Odling-Smee, Laland, and Feldman, 2003; Weiss and Buchanan, 2004). As a society, we even hungrily modify our behavior when we're told what our genes make us do, changing the causal landscape on the fly. This is poorly understood territory that deserves more attention before leaping into assumptions of genetic causation.

An important ethical issue is informed consent. Subjects should be informed of the true nature of what can be psychologically or socially intrusive as well as physically invasive research, and also of its risks and benefits, many of which one might only be able to guess at. Informed consent should be frank and clear, and testing should be done to ensure that subjects understand that the investigators are going to be trying to find genetically inherited reasons why they are without adequate pensions, poor, poorly paid, divorced, bad investors, or whatever. Geneticists are aware of the damage that some of their work on disease per se can do if, for example, confidentiality is breached. Employment, insurance, or other forms of discrimination can result. While known abuses may be rare, in the information age the increasing potential for unauthorized data transfer keeps these issues on the table, and society's guardians should always press legislators to ensure that the laws keep up. Probably the greatest risk is in the damage that could be done to people by screeners who don't understand, but make decisions based on, the problematic probabilism of most genotypic risk assessment. What does informed consent actually inform study subjects of?

What do we intend to do with the information we'll collect on genetic variation and behavior? Even at the individual level, discussion quickly slides from abstract investigation to the topic of intervention and prevention. We should acknowledge our historical and clearly resurgent tendency to extend from the individual to the group, which usually includes racial categorization. There is often a disclaimer that a given kind of social genetics is just exploration of the nature of nature, to build models of basic mechanism that can't be done through mouse experiments, and that the work is not intended to identify individuals for policy reasons, nor is understanding mechanism necessarily a step toward genetic determinism in the variation sense. But read carefully: do the same authors—even the same paper—eventually relate their findings to points of intervention? Intervention by whom, and at whose discretion?

Behavioral genetic studies often categorize groups, such as with/ without a test genotype, male/female, or by “race” and report statistically significant differences in some measured outcome. Even when the p-values are convincing and the result replicable, the actual differences are often very modest. Yet focus is naturally drawn to the “significant” mean difference and away from the variance and the much greater overlap between categories. A focus on the mean has potential for unintended social interpretation. Are women different from men in, say, spatial or mathematical ability? Does one “race” have a higher IQ than another? And how sensitive are the distributions to environment? These questions can easily be given less attention than they deserve, because they are less newsworthy and more problematic.

From the late 1800s through the end of World War II, developed societies experienced many aspects of eugenics. The nominal purpose was beneficent, to prevent the human species from deteriorating by accumulating too much defective genetic variation. This was an obvious extension of Darwinian value judgments, since society has always to some extent or other cared for those in need. But the new idea was the notion that modern society protected the unfit in a deeper evolutionary sense when it allowed them to reproduce. Genetic knowledge can identify harmful variants, and they can thus be eliminated.

Eugenics deservedly became a dirty word and is easy to confuse with attempts to help those with life-limiting impairment. Individual eugenics is also routine and voluntary. It is eugenic for parents to decide whether to have children who inherit variation from them that might lead to disease, or to select the sex of their next child, or simply to pick mates with traits they like. Screening for deleterious genetic variation by amniocentesis, and other means, is possible for many different traits, and when the trait is clearly devastating and clearly genetic, as in Tay Sachs disease or thalassemia, this may not be controversial. But the definition of “harmful,” or the threshold of acceptable harm can open a slippery slope of subjective judgment, especially when the causal epistemology is weak.

What about selective abortion based on genetic results for behavioral traits? How undesirable a behavioral trait might be involves subjective judgments or values, which can delve into senses of personal worth, and there are countless ways in which societal discrimination could be imposed on those deemed inherently less worthy. The potential to interpret culturally defined worth as being what was mandated by our history of natural selection (and hence “good”) is a deterministic Darwinian excess to which there is great temptation.

I invite readers who think this isn't dangerous and that we could never plausibly revisit the venomous abuses into which the eugenic movement morphed—that fed the Nazi era in Europe—to read some of the eugenic and human genetic literature of the time (e.g., the main source on human genetics for 20 years is Baur, Fischer, and Lenz, 1931, or see Proctor, 1989; Carlson, 2001). Filter out dated terms and knowledge and you will see that the logic is remarkably similar to what is being widely offered today. Eugenic societal policy, racial hygiene, scientific racism, and social Darwinism were propounded by respected scientists, often the leading scientists, and often in the name of medicine. The traits of interest then have uncanny resemblance to the targets of social and behavioral genetics now.

What starts as “objective” and disinterested study of the facts of life as they are in the raw can lead even well-meaning scientists to accommodate for self-interested reasons to policies that are directly harmful, or even horrific, for many people. What may be benign in normal times turns ugly in societies under stress. To think that those days are over is to be oblivious not just to history but to trends in our own society today. Feelings are heated and opinions sharply divided when it comes to how, when, or whether to study what is “in our genes” regarding heavily culture-dependent aspects of human life and behavior. The risk of societal abuse may be small, but it can't be said to be zero. It's inappropriate to dismiss these concerns as “just political.” This kind of science is by its very nature inherently political: social behavior is politics.


A character in Balzac's Lost Illusions says that “ideas are bipolar.” He means that clever people can find ways to argue opposite points of view from the same data. So far as we understand it today the genetic architecture of diverse traits has the characteristics I've outlined above. I have tried to show why it has been so difficult to identify in specific genetic terms more than a modest fraction of the control and variation of complex traits or even of their familial aggregation.

Work presented in this volume is consistent with that characterization. There are positive findings that may help in the understanding of genetic mechanisms affecting social behavior, but even classic examples of success, like ApoE, have subtleties that may be important to take into account. Extensive data from research on complex disease traits, which one might think would be more tractable than behavioral traits, suggests that a lot of effort, time, distraction, and resources will be committed to chasing down false positive results if the studies are not properly designed to detect genetic factors or if due attention is not paid to tempering results. Given the subtleties of environments in the social-behavioral context, really good design may be hard to come by.

Balzac's epigram at the beginning of this commentary illustrates the bipolar nature of ideas. It's relevant at this stage to keep them in mind. Some people may be born with a propensity to gamble at cards, but that propensity may only be enabled under peer pressure, and peer pressure is a societal structure that can coerce resistant genotypes to gamble as well. Is the cause of the economic disaster of a desolate old age, with its negative health consequences, in the gamblers' genotypes? The gambling system? In the causes of the peer pressure? Or lack of societal guarantees? To build genetic essentialism carelessly into our thinking is to gamble with everyone's future.


  1. Baur E, Fischer E, Lenz F. Human heredity. New York: Macmillan; 1931.
  2. Begg CB. On the use of familial aggregation in population-based case probands for calculating penetrance. Journal of the National Cancer Institute. 2002;94(16):1221–1226. [PubMed: 12189225]
  3. Buchanan AV, Weiss KM, Fullerton SM. Dissecting complex disease: The quest for the philosopher's stone? International Journal of Epidemiology. 2006;35(3):562–571. [PubMed: 16540539]
  4. Buchanan AV, Weiss KM, Fullerton SM. On stones, wands, and promises. International Journal of Epidemiology. 2006;35(3):593–596.
  5. Carlson E. The unfit: History of a bad idea. Cold Spring Harbor, NY: Cold Spring Harbor Press; 2001.
  6. King MC, Marks JH, Mandell JB. Breast and ovarian cancer risks due to inherited mutations in BRCA1 and BRCA2. Science. 2003;302(5645):643–646. [PubMed: 14576434]
  7. Odling-Smee FJ, Laland KN, Feldman MW. Niche construction: The neglected process in evolution. Princeton; NJ: Princeton University Press; 2003.
  8. Petronis A. Epigenetics and twins: Three variations on the theme. Trends in Genetics. 2006;22(7):347–350. [PubMed: 16697072]
  9. Proctor R. Racial hygiene: Medicine under the Nazis. Cambridge, MA: Harvard University Press; 1989.
  10. Ryff CD, Singer BH. Social environments and the genetics of aging: Advancing knowledge of protective health mechanisms. Journals of Gerontology Series B: Psychological Sciences and Social Sciences. 2005;60(Spec No. 1):12–23. [PubMed: 15863706]
  11. Scriver CR, Waters PJ. Monogenic traits are not simple: Lessons from phenylketonuria. Trends in Genetics. 1999;15(7):267–272. [PubMed: 10390625]
  12. Terwilliger JD, Weiss KM. Confounding, ascertainment bias, and the blind quest for a genetic “fountain of youth”, Annals of Medicine. 7. Vol. 35. 2003. pp. 532–544. [PubMed: 14649335]
  13. Weiss KM. Cryptic causation of human disease: Reading between the (germ) lines. Trends in Genetics. 2005;21(2):82–88. [PubMed: 15661353]
  14. Weiss KM. Genetics: One word, several meanings. In: Scriver C, editor. Online metabolic and molecular basis of inherited disease. New York: McGraw-Hill; 2005. Ch. 2.1.
  15. Weiss KM, Buchanan AV. Evolution by phenotype: A biomedical perspective. Perspectives in Biology and Medicine. 2003;46(2):159–182. [PubMed: 12721519]
  16. Weiss KM, Buchanan AV. Genetics and the logic of evolution. Hoboken, NJ: Wiley; 2004.
  17. Wellcome Trust Case Control Consortium. Genome-wide association study of 14,000 cases of seven common diseases and 3,000 shared controls. Nature. 2007;447:661–678. [PMC free article: PMC2719288] [PubMed: 17554300]
  18. Witherspoon DJ, Wooding S, Rogers AR, Marchani EE, Watkins WS, Batzer MA, Jorde LB. Genetic similarities within and between human populations. Genetics. 2007;176(1):351–359. [PMC free article: PMC1893020] [PubMed: 17339205]
  19. Wong AHC, Gottesman II, Petronis A. Phenotypic differences in genetically identical organisms: The epigenetic perspective. Human Molecular Genetics. 2005;14(Spec No. 1):R11–R18. [PubMed: 15809262]