R1. Assortment

Dar-Nimrod points out that the media (as well as researchers themselves) have shaped the public's perception of the relationship between genes and human behavior through uncritical and sensationalist reporting on the results of twin and gene association studies. Indeed, the media are enamored of reports that “two genes predict voter turnout” or whether one is a (twenty-first-century American) liberal or conservative is largely heritable. What the media do not like are disconfirming studies or studies that call reductionist models into doubt. Complexity does not sell.

Lewis presents several helpful critiques, two of which I mention here as additions or corrections to my overall argument: My account of the shaping of offspring behavior by the maternal environment ignores the fact that offspring actively shape that environment by eliciting maternal responses through behavior such as crying, clinging, and proximity seeking; my model of adaptation is a “fit the environment” model that ignores the extent to which organisms are active participants in the construction of their own environmental niches.

Müller, Lenz, & Kornhuber (Müller et al.) make a case for viewing what they term E _act – which includes socially learned behaviors and information – as a non-genetic form of inherited behavior. Their characterization of E _act is reminiscent of what Jablonka and Lamb (Reference Jablonka and Lamb2006) term the (non-genetic) behavioral and symbolic inheritance systems. Human behavior does not occur in a vacuum, and to the extent that E _act inheritance includes such things as culture, language, (cultural) history, and social norms and practices, it is essential to consider E _act when discussing the transmission of human behavior. A noteworthy feature of behavior genetics is the widespread tendency to treat behavioral and symbolic inheritance as if they were genetic inheritance.

For example, according to a study by Alford et al. (Reference Alford, Funk and Hibbing2005), mentioned by Miller, DeYoung, & McGue (Miller et al.), population variance in (twenty-first century American) conservatism and liberalism is attributable more to genetic then environmental variance.Footnote ¹ In fact, Alford et al. claimed that what they were considering was whether one was a “liberal” or “conservative,” but these were measured by asking American citizens a series of questions designed to measure attitudes associated with twenty-first-century American liberalism and conservatism. The problem is that there exists no set of attitudes (“symptoms”) by which we could identify a “liberal” and “conservative” in all times and places. Compare and contrast the “symptoms” of conservatism with the “symptoms” of Type I diabetes (T1D). The symptoms of T1D are basically the same, no matter where they occur, and we can discuss T1D as a disease without referring to any historical or geographical or cultural context (of course, these would be highly relevant if we were considering the etiology of T1D – e.g., the rate of T1D is 45 per 100,000 in Finland and 1 per 100,000 in Venezuela). But we cannot even coherently talk about conservatism, for example, without identifying that we are discussing the “phenotype” characterized by the “symptoms” of a twenty-first-century American conservative as opposed to the “symptoms” of an eighteenth-century German conservative or a twenty-first-century Russian conservative. This is because these divergent “symptoms” amount to very different “phenotypes”; there does not exist (as historians will tell you) a core set of “symptoms” that all conservatives in all times and places exhibit. To assume that there is such a thing is to reify something inherently historical and shifting, to treat it on the model of T1D, and thereby, completely distort it.

This difficulty is rarely considered. The assumption seems to be that just as behavioral genetics need not be concerned with the molecular genetic and biological mechanisms that link genes to behavior, so too it need not be concerned with the conceptual coherence of claiming that a particular set of local and historical behaviors are genetically inherited. Apparently, whatever “phenotype” a twin study claims to be genetically heritable simply is genetically heritable (to think otherwise would be to doubt the methodology). This includes such things as voting behavior (Fowler et al. Reference Fowler, Baker and Dawes2008); credit card debt (De Neve & Fowler Reference De Neve and Fowler2010); mobile phone use (including amount of time spent texting) (Miller et al. Reference Miller, Zhu, Wright, Hansell and Martin2012); and consumer preferences for soups and snacks, hybrid cars, science fiction movies, and jazz (Simonson & Sela Reference Simonson and Sela2011). One wonders whether the results of any twin study – a twin study concerning the heritability of speaking Chinese, or being an Anglican, or shopping at Macy's department stores – would prompt a reconsideration of the methodology (or at the very least, a critical examination as to what it makes sense to propose could be heritable, i.e., is a “phenotype,” in the first place).

Overton & Lerner argue that a paradigm shift (in the Kuhnian sense) requires a competing paradigm, and that despite my comment that the postgenomic view has not yet coalesced into a countervailing paradigm, relational developmental systems (RDS) qualifies as just such a paradigm. I admit my unfamiliarity with this approach, but judging by Overton & Lerner's comments, it seems like a strong candidate for a countervailing paradigm. Halpern notes that much that I argue draws upon Gilbert Gottlieb's developmental systems approach to genetics, and indeed, I have been significantly influenced by his work (as well as by the work of Richard Lewontin). Pléh notes that a debate similar to that I proposed between the genomic and postgenomic worldviews, specifically in relation to the validity of the twin study methodology, existed almost half a century ago between Jenkens and his followers on the one hand and Kagan on the other.

Dickins shows that the flip side, as it were, of the many processes that change DNA structure, which I highlighted, are the processes that preserve DNA integrity. Although DNA mutation may be necessary for evolution, limitless mutation would result in rapid extinction. Hence, an account of the mechanisms that lead to DNA transformation should be matched with an account of those that preserve DNA stability. It is interesting to conjecture what role stability-inducing mechanisms, or their failure, have in human behavior. Schanker notes how brain-derived neurotropic factor (BDNF), which I mentioned as promoting neurogenesis in the hippocampus, is also expressed in the cerebral cortex and other limbic structures and likely plays a critical role in neuronal and behavioral plasticity.

Garzillo & Trautteur present the outlines of a case for viewing DNA, or rather DNA and the cell, as a biological “Turing machine,” against my assertion that “DNA does not contain a determinate code equivalent to the digital code of a computer.” They argue that the “core cellular machinery” is a Turing machine. The core cellular machinery consists of (1) the sequence of bases of the coding gene, which constitute a text specifying the structure of proteins; and (2) the basic machinery of the cell and the expressed enzymes coded by the “minimal gene set.” I see a number of problems with their formulation:

1. 1. The sequence of DNA does not constitute anything like a “text” for specifying the structure of proteins. There are ~25,000 genes in the human genome yet at least 1 million proteins in the human body. In other words, the “expressed enzymes” are not coded in the “minimal gene set.” As noted in the target article, it is the cell, in response to internal and external signals, that determines which isoforms of a given protein will be produced. Furthermore, the exons that the cell combines in different ways to form proteins can be widely dispersed throughout the DNA sequence, challenging the notion of genes as either linear “letters” or “words” that constitute a “text.” Hence, the cell does not function as an “interpreter,” which in the computer science sense of the term either executes the source code directly (which Garzillo & Trautteur equate with the DNA sequence) or translates the source code into some intermediate code that it then executes. The DNA sequence better resembles a database on which the cellular system draws rather than a logical program of instructions (Noble Reference Noble2010).

2. 2. I may well have missed something, but I am not sure how one gets from the “minimal gene set” to the “expressed enzymes.” Without the epigenome, nothing will be expressed (i.e., transcribed). The epigenome itself, however, is not regulated by the minimal gene set (how could it be, if the natural state of the gene set is to be in the “off” position?). Is this problem circumvented by simply postulating that the core cellular machinery includes the expressed enzymes, which then (by what appears to be definitional fiat) obviates the need to account for how these enzymes are able to be synthesized in the first place?

3. 3. Similarly, the minimal gene set can never give rise to differentiated cell types, since what differentiates cell types is not their gene set, but rather gene set activity. Gene set activity is not coded in the gene set, since what differentiates a neuron and a heart cell is not differences in nuclear DNA (nDNA) sequence, but differences in gene transcription. Perhaps Garzillo & Trautteur intended to characterize a “generic cell,” but generic cells do not exist.

Hence, DNA does not constitute a Turing machine, but if anything, an “interaction machine.”

Furthermore, Garzillo & Trautteur argue that the central problem of DNA and the origins of life is how active structures such as ribosomes evolved together with the genes coding for them. However, some of these structures, such as mitochondria, plastids (chloroplasts), and possibly other organelles of eukaryotic cells likely originated as free-living bacteria that were incorporated into another cell in a symbiotic relationship (endosymbiosis). Recent evidence suggests that functionally important regions of ribosomes were recruited and could be relics of an ancient ribonucleoprotein world (Harish & Caetano-Anollés Reference Harish and Caetano-Anollés2012), while retrotransposons are ancient viruses that became incorporated into DNA. The history of both DNA and the cell indicates that they are the products of multiple symbiotic (and non-symbiotic) relationships whereby precursors of cellular organelles were incorporated into precursors of the cell.

Teixeira, Carvalho-Filho, & Silveira (Teixeira et al.) present an interesting overview as to how neogenomic processes might inform investigations of neuronal micro-circuitry. I mention their contribution here because their comments concerning “epigenetic robots” bear a close resemblance to the arguments of Garzillo & Trautteur for viewing the core cellular machinery as a Turing machine. They conjecture that an “epigenetic robot” would be “guided by software (genome) modifiable by both inherent reprogramming tasks (transposable elements) and developmental experiences (epigenome), which in turn should be driven by interactions with robot inner/outer environments, all contributing to the emergence of robot self-programming.” My reservations in regard to this characterization of an epigenetic robot are essentially the same as those in regard to the characterization of the core cellular machinery as a Turing machine: The genome is not equivalent to (or not accurately analogized to) a program and/or a program that engages in (self) reprogramming. Indeed, it has been suggested that DNA be viewed as more akin to hardware and the epigenome to software (Dolinoy et al. Reference Dolinoy, Weidman and Jirtle2007; Jammes et al. Reference Jammes, Junien and Chavatte-Palmer2011), although in the final analysis, analogies to hardware and software may obscure, like a lot of other dichotomies, the complex, dynamic, multifaceted, and fluid nature of the phenomena.

I agree with Glatt's assertion that “statistics alone cannot bridge the gap from molecule to mind,” particularly when the statistical approach depends upon a genetic paradigm that is no longer viable. Glatt argues that attempts to relate polymorphisms to behavior should be driven not by the search for statistical correlations, but by experimentation, for example, in vitro analysis and studies of knockout and transgenic mice. Although such approaches are an important tool in trying to unravel the physiological-behavioral effects, if any, of polymorphic variations, they present a number of potential difficulties.

Consider, for example, the monoamine oxidase gene: monoamine oxidase A (MAO-A) breaks down a class of neurotransmitters known as monoamines, including adrenaline, noradrenaline, dopamine, and serotonin, thereby diminishing their bioavailability. The MAOA gene exhibits polymorphisms in its promoter region characterized by “tandem repeats,” the replication of two or more nucleotide sequences directly adjacent to each other (because the tandem repeats vary in number, they are referred to as variable number tandem repeats [µVNTR]). On the basis of in vitro analysis, the 3.5- and 4-repeat MAOA-µVNTR alleles have been classified as being transcribed 2 to 10 times more efficiently than alleles containing the 3-tandem repeat (Sabol et al. Reference Sabol, Hu and Hamer1998). Hence, the 3-repeat allele is classified as low (l MAOA-µVNTR), for low transcriptional efficiency, and the 3.5- and 4-repeat alleles as high (H MAOA-µVNTR), for high transcriptional efficiency. It is commonly assumed that the differences in in vitro transcription rates translate into different levels of bioavailable MAO-A in the brain, which in turn is presumed to translate into different levels of bioavailable serotonin (5-HT), yielding the following causal schematic: high/low MAOA-µVNTR→high/low levels of brain MAO-A→ high/low levels of brain 5-HT. Finally, the different levels of serotonin in the brain are presumed to translate into differences in behavioral phenotypes.

In fact, it is by no means clear that high and low alleles of MAOA-µVNTR correspond to higher and lower levels of brain serotonin. Studies that have attempted to demonstrate the effects of MAOA-µVNTR genotypes upon in vivo (as opposed to in vitro) brain levels of MAO-A have had mixed, largely negative results (Alia-Klein et al. Reference Alia-Klein, Kriplani, Pradhan, Ma, Logan, Williams and Fowler2008; Cirulli & Goldstein Reference Cirulli and Goldstein2007; Fowler et al. Reference Fowler, Alia-Klein, Kriplani, Logan, Williams, Zhu and Wang2007; Nordquist & Oreland Reference Nordquist and Oreland2010; ). According to a recent review (Nordquist & Oreland Reference Nordquist and Oreland2010, p. 2), “in adult humans, and monkeys with orthologous genetic polymorphisms [polymorphisms having the same function in two different species], there is no observable correlation between these functional genetic variants [of MAOA] and the amount or activity of the corresponding proteins in the brain.” This is not surprising. The brain, like all other organ systems, is characterized by elaborate homeostatic mechanisms: Even if we assumed “greater transcriptional efficiency” of the gene for a given enzyme, we would not expect this to translate into more of that enzyme and more of the physiological effects associated with that enzyme in any straightforward manner.

Furthermore, although these polymorphisms of the MAOA gene have been associated with a bewildering array of phenotypes (see next section), the most well-known association is with aggression (Buckholtz & Meyer-Lindenberg Reference Buckholtz and Meyer-Lindenberg2008): High MAOA is associated with lower aggression and low MAOA with higher aggression (although like all candidate gene association studies, these studies have failed to be consistently replicated). This association was hypothesized on the basis of knockout studies in mice (Shih & Thompson Reference Shih and Thompson1999). MAOA knockout (KO) mice exhibit greater aggression (as well as a number of other behavioral abnormalities). One problem with drawing behavioral inferences from KO mice is that what KO mice manifest are in effect the symptoms of an artificially produced monogenic disorder. It is by no means clear that one can infer, from behavior associated with an artificial monogenic disorder, behavior associated with polymorphisms of that same gene. Finally, if we consider the differences in gene transcription associated with differences in aggression in fruit flies (see next section), it seems very unlikely that any single polymorphism will be a risk factor for behavior.

I mention this cautionary tale not to argue that experimental analysis of polymorphisms is without value, but rather to emphasize the limitations of such analysis.

R2. Gene association studies

I agree with Homberg and Crusio that in addition to environmental stimuli affecting the genome, the genome itself can also influence the impact of environmental stimuli. I did briefly mention something to this effect in the target article, when I commented that, “Of course, offspring may differ in the degree or manner in which specific behavioral phenotypes are shaped by the perinatal environment due to any number of genetic, epigenetic, and micro-environmental differences (such as fetal position), or on the basis of sex” (sect. 9.2). The most well-known example of differences in the impact of environmental stimuli (to use Homberg's formulation) being linked to polymorphic differences concerns polymorphisms of the genes that compose the hepatic cytochrome P450 mixed-function oxidase system and drug metabolism (Wrighton & Stevens Reference Wrighton and Stevens1992).

I am, however, highly skeptical of the examples that Homberg presents: Vulnerability to environmental stressors being influenced by polymorphisms of 5-HTT and DRD4. What I argued in the target article in relation to candidate gene association studies applies as well to gene association studies that posit a specific G × E. Consider that the “stress response” is one of the most diffuse physiological responses in the human organism, involving the hypothalamo–pituitary–adrenal axis and immune system and changes in levels of (to name but a few), corticotropin releasing hormone, glucocorticoid receptor, adrenocorticotropic hormone, epinephrine, norepinephrine, prolactin, growth hormone, gamma-aminobutyric acid, neuropeptide Y, beta receptors, neural killer cell activity, mineralocorticoid receptor, vasopressin, proopiomelanocortin, thyroid stimulating hormone, gonadotropic hormones, luteinizing hormone, follicle stimulating hormone, and oxytocin. That a response that involves proteins coded in thousands of genes, not to mention unknown epigenetic processes, and every major organ system in the body, should be so impacted by polymorphisms on a single gene as to have significant behavioral consequences does not make a lot of sense physiologically or from the standpoint of evolutionary biology.

A good example of just how many proteins we might expect to be differentially transcribed in behavioral variation is provided by an example I considered in the target article: aggression in fruit flies (Drosophila melanogaster). Zwarts et al. (Reference Zwarts, Magwire, Carbone, Versteven, Herteleer, Anholt and Mackay2011) bred a strain of hyperaggressive fruit fly. Using advanced DNA expression analysis they found differences in the transcription levels of 4,038 genes in homozygous hyperaggressive flies versus controls; 1,169 genes were coordinately up or down regulated in all hyperaggressive homozygous flies, with epistatic interactions for over 800 genes. Significant pleiotropy was also observed in that these same genes were involved in a host of basic physiological processes including olfaction, nervous system development, detoxification of xenobiotics, and sex determination, as well as genes of previously unknown origin (for the potential significance of pleiotropy in neuropsychiatric illness, see the commentary of Deutsch & McIlvane). In a situation such as this, no single polymorphism on a single gene (or 2 or 10 genes) could predict, or be a risk factor for, aggression in fruit flies.

Homberg's characterization of polymorphisms of the serotonin transporter linked polymorphic region (5-HTTLPR) as predictors of, or risk factors for, stress-related “maladaptive” behavior draws upon a study by Caspi et al. (Reference Caspi, Sugden, Moffitt, Taylor, Craig and Harrington2003), according to which specific polymorphisms of 5-HTTLPR, combined with stressful life events, increase the risk of depression. Like most candidate gene association studies, however, this study has failed to be consistently replicated (see supplemental table for Charney & English [2012] at http://tinyurl.com/AssociationStudies), and the conclusion of a comprehensive meta-analysis published in the New England Journal of Medicine is as follows:

Furthermore, the specific polymorphisms Homberg mentions are two members of a small group of polymorphisms that have been effectively data-mined for associations (see Charney & English Reference Charney and English2012 and supplemental table at http://tinyurl.com/AssociationStudies). For example, the long and short polymorphic regions of the serotonin transporter gene have been associated with, in addition to many other behavioral and non-behavioral phenotypes, agreeableness, alcoholism, Alzheimer's disease, anger/aggression, anorexia, attachment, attention-deficit/hyperactivity disorder, autism, bipolar disorder, blushing, borderline personality disorder, brain activation by colorectal distention, brain activation in processing errors, breast cancer, bulimia, chronic fatigue syndrome, cleft lip, conscientiousness, contraception use, cooperativeness, creativity, deductive reasoning, depression, epilepsy, extraversion, fearfulness, fibromyalgia, pathological gambling, gastric emptying, harm avoidance, heroin use, attitudes toward individualism and collectivism, insomnia, intelligence, interpretive bias, irritable bowel syndrome, job satisfaction, loneliness, longevity, maternal sensitivity, migraines, neurodermatitis, neuroticism, novelty seeking, number of sexual partners, obesity, obsessive-compulsive disorder, openness, optimism, osteoporosis, panic disorder, parenting, Parkinson's disease, persistence, periodontal disease, postpartum depression, posttraumatic stress disorder, premature ejaculation, premenstrual dysphoria disorder, psoriasis, resiliency to victimization, reward dependence, schizophrenia, seasonal affective disorder, shyness, sleep apnea, smoking, social phobia, sudden infant death syndrome, suicide, utilitarian moral judgments, and well-being. (This list is by no means complete. For complete references for the associations listed here, along with a representative list of associations for three other genes – MAOA, DRD2, and DRD4 – that have been associated with a wide array of phenotypes, see the supplemental table for Charney & English [2012] at http://tinyurl.com/AssociationStudies).

How is it possible that the same polymorphisms of the same gene could simultaneously predict (or be risk factors for) so many different phenotypes (even when, as is sometimes the case, a specific G × E or G × G (gene × gene) interaction is proposed)? A common response to the question (to the extent that it is raised) is to evoke the concept, discussed by Michel , of an “endophenotype.” As Michel characterizes it, an endophenotype describes the various physiological pathways that relate the genotype to behavioral phenotypes. Thus characterized, the concept is certainly important. However, it has been used as a way to explain how the same polymorphism could simultaneously give rise to so many diverse phenotypes (although it remains unclear as to how the same polymorphism could predict, or be a risk factor for, so many different phenotypes). According to Gottesman and Shields (Reference Gottesman and Shields1973), an endophenotype is defined as an “intermediate trait” or an “internal phenotype,” that lies “intermediate” between the genotype and phenotype. The idea is that the endophenotype, which is more “elementary” than the phenotype, can give rise to an array of phenotypes (due to interacting genetic or environmental factors) that all share something with the more primary endophenotype. Furthermore, the genetic basis of endophenotypes is assumed to be less complicated than the phenotypes to which they give rise, to involve fewer genes, and be more “direct” and “deterministic.”

The idea that an endophenotype involves fewer genes and is more direct and deterministic lacks empirical support, and one suspects that this characterization justifies the results of candidate gene association studies more than anything else. If it is proposed that the proliferation of phenotypes associated with the long and short polymorphisms of 5-HTT point to an underlying endophenotype, what more elementary phenotype unites maternal sensitivity, premature ejaculation, irritable bowel syndrome, utilitarian moral judgments, schizophrenia, periodontal disease, and voting? Nor can we divide all of these “phenotypes” into those characteristic of “vulnerability” and those not, or those that that are adaptive and those that are maladaptive (are utilitarian moral judgments adaptive or maladaptive?), particularly given that one and the same trait can be adaptive in one environment and maladaptive in another.

R3. Twin studies

In their commentary, Miller et al. begin with a familiar rejoinder: Those who challenge the twin study methodology from any perspective other than that of statistics – in my case, the perspectives of molecular genetics and developmental and evolutionary biology – fail to appreciate the precise nature of biometric genetics. I anticipated this rejoinder and responded to it at length in the target article (sect. 6.1.1. “Objection 1: Biometric versus biomolecular genetics”). To summarize what I said there: Although biometrical analysis is not concerned with the molecular mechanisms that underlie phenotypic variation, it nonetheless depends foundationally upon certain empirical assumptions. Vilarroya questions the point of my assertion that “biometric genetics must make contact with the natural world at some point.” All that I meant by this was that biometric genetics depends upon certain empirical assumptions, and if these assumptions turn out not to be true, then the validity of the twin study methodology will be called into question. For example, one of the empirical assumptions of the twin study methodology is that MZ (monozygotic) twins share 100% of their segregating genes and their genetic identity remains fixed throughout the life course. To demonstrate that neither of these propositions is true, it was necessary to consider the intermediary mechanisms between genotype and phenotype. The assumption that biometric genetics need not be concerned with advances in molecular genetics (because not concerned with the underlying genetic-molecular mechanisms) has in some ways enabled a methodology developed in the late nineteenth and early twentieth centuries to persist essentially unchanged into the twenty-first (what has changed is the sophistication of the statistical analysis).

According to Miller et al., most of the genetic and epigenetic MZ-twin differences I considered have nothing to do with heritability because they are acquired and not inherited. To the contrary: They can all be inherited, a point I went to great pains to demonstrate throughout the article. Because of the critical importance of this point, I repeat the relevant sections of my article here (references omitted):

If Miller et al. believe that all of the studies in support of these statements are incorrect, then it is incumbent upon them to provide countervailing evidence in support of such a claim.

Similarly, I suspect that Burt's proposal to compare children created via in vitro fertilization (IVF) and gestated and raised by biological mothers with IVF children gestated and raised by non-biological mothers, as a way to disaggregate genetic and neogenetic concordance producing effects, rests upon the assumption that neogenetic effects are solely acquired. (If such studies are undertaken, one hopes that researchers will take into account the following growing body of evidence: Children conceived through assisted reproduction technology (ART) are at an increased risk for negative health outcomes (Allen et al. Reference Allen, Wilson and Cheung2006; McDonald et al. Reference McDonald, Han, Mulla, Murphy, Beyene and Ohlsson2009; Wen et al. Reference Wen, Jiang, Ding, Dai, Liu, Xia and Hu2012); there are significant epigenetic differences between ART and naturally conceived children (Katari et al. Reference Katari, Turan, Bibikova, Erinle, Chalian, Foster and Sapienza2009; Turan et al. Reference Turan, Katari, Gerson, Chalian, Foster, Gaughan and Sapienza2010; van Montfoort et al. Reference van Montfoort, Hanssen, de Sutter, Viville, Geraedts and de Boer2012).

In regard to the distinction Miller et al., as well as Burt, draw between what is inherited and what is acquired, consider the following: MZ twins are genetically identical at one point in their development: before they are MZ twins (i.e., when they are still a single zygote). The moment the zygote divides, they are no longer genetically identical (this is uncontested in regard to mtDNA, but likely true in relation to several neogenetic phenomena). In a situation such as this, what precisely is inherited and what is acquired? Shall we say that what is “inherited” applies only to the predivision zygote and that everything from division on is “acquired”?

Miller et al. state that “several types of evidence suggest that DNA extracted from different cell types is nearly identical. MZ twins are nearly as alike in their SNP genotypes as are two DNA samples from the same person.” Regarding SNPs, as I noted in the target article, “What this [neogenomics] does not call into doubt, however, is the following: MZ twins are significantly more genetically concordant than DZ [dizygotic] twins (and are likely most concordant in relation to SNPs), and this greater genetic concordance plays an important role in a wide range of intertwin phenotypic concordances” (sect. 6). SNPs of nDNA, however, do not occur in mtDNA; and retrotransposon insertions, CNVs, cellular and chromosomal aneuploidy, and the epigenome are not SNPs. We know that MZ twins differ dramatically in their mtDNA, which is inherited solely from the mother and is stochastically partitioned during the formation of the zygote. We know that MZ twins differ in their CNVs (see the findings of the study by Bruder et al. [Reference Bruder, Piotrowski, Gijsbers, Andersson, Erickson, de Stâhl and Dumanski2008]: supplemental data at http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2427204/bin/mmc1.pdf). We know that MZ twins differ in their epigenomes and that these differences change over the life course. And because germline retrotransposition and chromosomal aneuploidy are stochastically distributed, MZ twins are likely discordant for these as well.

Regarding the findings cited by Miller et al. for twin concordance for SNPs, things are likely not quite as they seem. During pregnancy, the mother and fetus exchange a certain number of cells (along with their DNA), most likely through the placenta, with the result that the mother and/or fetus exhibit chimerism, a form of somatic mosaicism in which the mosaic cells contain the DNA of another individual (Jonsson et al. Reference Jonsson, Uzunel, Gotherstrom, Papadogiannakis and Westgren2008). MZ and DZ twins can exhibit, in addition to maternal–fetal chimerism, twin–twin chimerism. Theoretically, chimerism cannot occur in MZ twins, under the assumption that they are genetically identical (so that any intertwin cellular trafficking would not result in somatic mosaicism, but simply the exchange of genetically identical cells). In fact, there are a number of accounts of chimerism in MZ twins (Bourthoumieu et al. Reference Bourthoumieu, Yardin, Terro, Gilbert, Laroche, Saura and Esclaire2005; O'Donnell et al. Reference O'Donnell, Pertile, Sheffield and Sampson2004; Willer et al. Reference Willer, Herrera, Morrison, Sadovnick and Ebers2006), and the incidence is likely significantly higher due to the fact that the possibility of chimerism is only considered (and hence investigated) in cases of dramatic disease related intertwin phenotypic discordance. Cases of inter-MZ-twin chimerism also reveal how genotyping from blood can conceal intertwin genetic discordance.

Consider the following representative (actual) case of blood chimerism between MZ-monochorionic (MZ–MC [monochorionic]) twins (O'Donnell et al. Reference O'Donnell, Pertile, Sheffield and Sampson2004): At delivery, one MZ twin exhibited a normal phenotype, and the other twin showed the dysmorphic features characteristic of trisomy 21. Blood from both twins showed an admixture of normal and trisomic cells (cells with three chromosomes, characteristic of trisomy 21). However, tissue studies of skin and buccal cells of the dysmorphic twin showed only trisomy 21 cells, whereas buccal cells from the normal twin showed only normal cells. In other words, the genetically distinct hematopoietic stem cells of the twins “fused,” leading to a chimeric blood system shared by both twins. If genotyping of the twins were based solely on blood analysis, then their genetic discordances would not have been discovered. A large number of studies have highlighted the potentially confounding effects of blood chimerism when studying genomic and epigenetic variations among discordant MZ twins (Bourthoumieu et al. Reference Bourthoumieu, Esclaire and Yardin2006; Kaplan et al. Reference Kaplan, Foster, Shen, Parry, McMaster, O'Leary and Gusella2010; Machin Reference Machin2009). According to a recent analysis by Erlich (Reference Erlich2011), the effects of chimerism on the detection of variation in intertwin SNPs is substantial:

I am aware, contrary to what Miller et al. say, that heritability indicates the percentage of variance in a phenotype that is due to inheritance, not to variations in DNA. It is precisely for this reason that I speculated in the concluding section of the target article (sect. 10, “Postgenomics”) what might be involved in attempting to estimate variation due to everything inherited via the germline:

According to Miller et al., my “greatest mistake” is that I misunderstand the role of quantitative models, which are concerned with explicating the mechanisms that lead to familial resemblance. Most of the examples I considered in the target article – resemblance between mothers and their offspring in rearing behavior, stress responses, and mating behavior; and resemblances of MZ twins, DZ twins, and singletons with each other and their parents – are examples of familial resemblance. That family members resemble each other is obvious. That DNA, the epigenome, and the pre- and postnatal environment interacting with each other in complex ways contribute to these resemblances is clear.

What is not clear, what is in fact misguided, is the assumption that the twin study methodology can effectively partition and quantify the contribution to phenotypic variation between offspring that is due to differences in what is “inherited” on the one hand and what is “environmental” on the other. “Inherited versus environmental,” or “nature versus nurture,” are artificial and superannuated dichotomies that distort the complexity of the phenomena. Trying to fit environmentally induced epigenetic activation of retrotransposons, or intergenerationally transmitted epigenetic reprogramming, into this dichotomous worldview as represented in standard quantitative genetic models is like trying to locate black holes within Aristotle's dichotomy of the sublunar world of change and the immutable heavens.

This is a fitting place to consider Molenaar's contribution. As someone who has thought long and hard about genetic modeling techniques, Molenaar illustrates the problems that violations of HS1 (the shared genetic identity of MZ twins = 1 and DZ twins = 0.5) and HS2 (this genetic identity remains fixed over the life course) pose to standard methods of analysis in behavior genetics. In this, we are in complete agreement. Indeed, Molenaar's previous work has shown how subject-specific violations of HS1 and HS2 can lead to severely biased estimates in the standard longitudinal genetic factor model. However, where Molenaar and I disagree is in his optimism that this problem can be solved using multivariate, phenotypic, and intraindividual time series data obtained from pairs of genetically related subjects.

The iFACE model that Molenaar has pioneered is a significant improvement over standard quantitative genetic models to the extent that it attempts to incorporate many of the complexities that the standard model ignores. But the limitations of the iFACE model illustrate why this complexity is likely to remain intractable. Space does not permit a full examination of the statistical properties and assumptions of the iFACE model, other than to note that it likewise (and necessarily) rests on dubious empirical assumptions.

In brief, iFACE combines an idiographic filter technique with the standard genetic factor model to look for fixed patterns of factor correlations in a pair of genetically related individuals over time. One of the benefits of iFACE is that it allows genetic correlations between DZ twins to be freely estimated, rather than assumed to be 0.5 from the outset. Moreover, Monte Carlo simulations have demonstrated that iFACE yields fairly accurate estimates of data generated from known parameters (Molenaar Reference Molenaar, Hood, Halpern, Greenberg and Lerner2010). However, the Achilles heel of iFACE lies in the assumptions it must make as to what remains fixed or constrained. The iFACE model must assume stability in the shared environment or in shared genetics in order to estimate the source of phenotypic variation over time (otherwise one could not disambiguate what is driving the variation). In fact, the iFACE model assumes identical, shared environments for DZ twins, at least as the model is presented in existing publications: “The only exception [to parameters that are allowed to be subject specific] is the set of loadings on the common environmental factor: these were constrained to be equal across the two subjects, but could freely vary across the four phenotypic time series” (Molenaar Reference Molenaar, Hood, Halpern, Greenberg and Lerner2010, p. 644).

In other words, the model presumes an answer to the precise question that is at stake in the larger debate: What, exactly, is driving phenotypic variation? Indeed, the assumption of equivalent environmental factors across subjects is every bit as contentious as the a priori assumption of a shared genetic endowment of 0.5 for DZ twins and 1 for MZ twins. Moreover, for any real-world behavioral phenomenon of any interest, environments do differ between twins and change for both twins over time. To even begin to address the methodological problems raised by this reality, one would need reliable and comparable measures of every conceivable environmental influence acting on each of the DZ twins over the time horizon in question (and even then it is not clear how to disambiguate environmental and genetic effects).

Molenaar's work has taken seriously the modeling problems posed by non-linear and non-ergodic genetic processes. However, although iFACE represents an improvement compared to the simplistic factor models used in so much research, it nonetheless assumes an answer to the real question that is at stake. In the end, iFACE shows us that in trying to deal with increasing degrees of complexity and variation in our models, we have no Archimedean point on which to stand. Violations of HS1 and HS2 present deep methodological problems for behavior genetics that cannot be resolved by any existing analytic approach. In fact, I would argue that the real problem lies not in our inability to disambiguate environmental and genetic contributions to phenotypic variation, but with the attempt to do so in the first place. In stating this, I register my agreement with Crusio, who avers that the difficulties that beset human heritability estimates transcend technical solutions. I also agree with Lickliter that we are likely dealing with a degree of complexity such that a full understanding of the route from fertilized egg to mature human exhibiting one or another behavior lies beyond the limits of human understanding.

Burt argues that my assertion that the neogenome behaves like neither an E (environmental) nor G (genetic) parameter is erroneous for the following reason:

First, what Burt ignores is the fact that MZ twins are not genetically identical. Even assuming “loading” onto G and E, heritability estimates in the twin study methodology depend upon the assumption that MZ twins share 100% of their segregating genes. Therefore, it is not the case that neogenetic processes can be partitioned into concordance-producing effects (G) and discordance-producing effects (E) within a model in which the genetic identity of MZ twins is fixed at 1. To the extent that the twin study model treats the genetic identity of MZ twins as fixed in this manner, neogenetic effects will necessarily be excluded.

Second, the use of the expression “absorbed” (“the neogenome will be absorbed into G”) in Burt's commentary is interesting because it implies that something that is not G would be treated as if it were G. So, if MZ twins are more similar because of more similar epigenetic profiles, the epigenome will be “absorbed” into G. Does it matter if the effects of the epigenome are “absorbed” into G in heritability estimates, inasmuch as heritability is concerned with what is inherited, not with DNA per se, as Miller et al. note? Clearly, it does. Consider the intergenerational transmission of environmentally induced changes in sperm count due to exposure to the endocrine disruptor vinclozolin (target article, sect. 5).

If MC–MZ twins were more concordant for low sperm count from vinclozolin-induced epigenetic reprogramming due to their sharing a single blood supply prenatally, and if this epigenetic programming was absorbed into G, then we would falsely conclude that the type of oligospermia we were considering was caused by an inherited defect in the genes (and would likely begin looking for the responsible genes). In other words, the absorption of the epigenome into the genome would mask the true cause of the phenotype. What is more, in this case, the epigenetic reprogramming is environmentally induced and then intergenerationally transmitted: The phenotype itself is manifested in the absence of the original inducing environment. If environmentally induced, inherited epigenetic reprogramming were absorbed into G, we would never consider that the culprit might be an environmental agent. This is also an instance of boundary crossing or the blurring of boundaries, for how are we to classify the inheritance of environmentally induced epigenetic changes in the absence of the original environmental stimulus? Is E inherited? Or is E transformed into G?

Both Battaglia and MacDonald & LaFreniere argue against the claim that MZ twins become more epigenetically and genetically discordant over their lifetimes by citing studies that purport to show that heritability increases with age. I am not sure if their claim is that these studies demonstrate that epigenetic discordances of MZ twins do not, in fact, increase over time, or that, although they may increase over time, they have no effect upon phenotypes. If the former, then clearly the results of a twin study cannot refute the existence of increasing epigenetic discordance, a phenomenon that has been repeatedly demonstrated by advanced molecular techniques (Ballestar Reference Ballestar2009; Fraga et al. Reference Fraga, Ballesta, Paz, Ropero and Setien2005; Kaminsky et al. Reference Kaminsky, Tang, Wang, Ptak, Oh, Wong, Feldcamp, Virtanen, Halfvarson, Tysk, McRae, Visscher, Montgomery, Gottesman, Martin and Petronis2009; Kato et al. Reference Kato, Iwamoto, Kakiuchi, Kuratomi and Okazaki2005; Martin Reference Martin2005; Mill et al. Reference Mill, Dempster, Caspi, Williams, Moffitt and Craig2006; Ollikainen et al. Reference Ollikainen, Smith, Joo, Kiat Ng, Andronikos, Novakovic and Craig2010; Petronis et al. Reference Petronis, Gottesman, Kan, Kennedy, Basile and Paterson2003; Poulsen et al. Reference Poulsen, Esteller, Vaag and Frage2007; Rosa et al. Reference Rosa, Picchioni, Kalidindi, Loat, Knight, Toulopoulou and Craig2008). To deny this would require a refutation of these studies. So, I take the argument to be the latter, namely, that studies that purport to show that heritability increases with age demonstrate that whatever epigenetic (and genetic) changes MZ twins experience over their lifetimes have no effect upon, for example, cognitive development.

Such generalizing from one or two studies concerning one or two phenotypes to all behavioral phenotypes is a common practice in the twin study literature, and it is also an example of the fallacy of “hasty generalization.” Given that the results of a number of other twin studies draw the opposite conclusion–that heritability decreases with age–including the heritability of cognitive ability, such an argument in this context is perhaps more accurately characterized as an instance of the fallacy of neglect of relevant evidence. For example, according to Reynolds et al. (Reference Reynolds, Finkel, McArdle, Gatz, Berg and Pedersen2005):

And as Reynolds et al. (Reference Reynolds, Finkel, McArdle, Gatz, Berg and Pedersen2005, p. 13) note of their own study: “The findings of the present study can be construed as generally supportive of theories proposing the increasing importance of the environment with respect to cognitive aging: Although heritable influences are of greater relative importance for individual differences in cognitive performance, environmental variances increase steadily after age 65.” Other twin studies have reported decreasing heritability for personality (Floderus-Myrhed et al. Reference Floderus-Myrhed, Pedersen and Rasmuson1980; Pedersen et al. Reference Pedersen, Plomin, McClearn and Friberg1988), science scores (Haworth et al. Reference Haworth, Dale and Plomin2009), extraversion and introversion (Viken et al. Reference Viken, Rose, Kaprio and Koskenvuo1994), self-esteem (Jonassaint Reference Jonassaint2010; Raevuori et al. Reference Raevuori, Dick, Keski-Rahkonen, Pulkkinen, Rose, Rissanen and Silventoinen2007), body mass index (Korkeila et al. Reference Korkeila, Kaprio, Rissanen and Koskenvuo1991), and anxiety/depression (Saviouk et al. Reference Saviouk, Hottenga, Slagboom, Distel, de Geus, Willemsen and Boomsma2011).

According to Battaglia, though epigenetic effects are potentially important, the individual and specific impact on brain and behavior is neither well understood nor unambiguously linked to gene expression data. In support of this assertion, he mentions a study by Zhou et al. (Reference Zhou, Yuan, Mash and Goldman2011). Whatever Battaglia's precise intent in mentioning this study, their conclusion unambiguously links epigenetic changes to changes in gene expression and behavior:

According to MacDonald & LaFreniere, “Charney does not present a case that there are important epigenetic effects on adaptive traits like cognitive ability or personality.” To the contrary, all of the examples I presented concern epigenetic effects upon adaptive traits: stress responses, fearfulness, maternal rearing behavior, and mating behavior. Given MacDonald & LaFreniere's focus on cognitive ability, I add the following two excerpts from two recent studies:

MacDonald & LaFreniere assert that many of the processes I highlight (such as epigenetics) are stochastic, and stochastic events are likely to be maladaptive. A direct refutation of this assumption comes from a highly stochastic mechanism that is also highly adaptive, in fact, is the sine qua non for adaptation: the immune system. As mentioned in the target article, stochastic DNA recombination allows for the creation of ~10¹⁵ variable antibody regions to combat rapidly mutating antigens. Furthermore, in noting how epigenetic events have been linked with pathology – which they claim is not surprising given their stochastic nature – they ignore the fact that the epigenome is involved in every aspect of human development starting with cellular differentiation. A neuron differs from a kidney cell not because of differences in its nDNA, but because of differences in its epigenome. Epigenetic differences enable different tissue types and different organs. Hence, if epigenetic processes are maladaptive, then having a brain (as opposed to a kidney) is maladaptive.

Furthermore, stochasticity is playing an increasingly important role in theories of evolutionary adaptation. As against Fisher's standard geometric model of evolution by small steps, that is, the accumulation of many mutations with small benefit, stochastic models of evolution are increasingly being employed. According to Østman et al. (Reference Østman, Hintze and Adami2012, p. 1):

MacDonald & LaFreniere defend the primacy of additive genes in behavioral variation (a basic assumption of the twin study methodology) by citing a statistical analysis by Vissher, according to which for fitness-related traits, “typically around 50% of the phenotypic variation is due to additive genetic variation and…about 80% of genetic variation is additive.” While MacDonald & LaFreniere mention this to challenge the notion of extensive G × E (which would include epigenetic processes to the extent that the epigenome is classified as part of the environment), it is important to consider that such high figures for additive variance entail that epistasis (G × G interaction) is not an extensive feature of complex traits. While this view may be prevalent in the behavioral genetics community, it is certainly not the view of geneticists and evolutionary biologists in general (Phillips et al. Reference Phillips, Otto, Whitlock, Brodie, Wade and Wolf2000, pp. 26–27 [references omitted]):

Aggression is a universal and highly adaptive behavior. As noted above, 800 genes implicated in differences in aggression in Drosophila showed significant epistasis (as well as pleiotropy).

MacDonald & LaFreniere note that contrary to my claim that twin studies are responsible for the principle of minimal maternal effects, recent twin studies of attachment have indicated strong effects of a shared maternal environment. I make two points in this regard: First, the claim of Plomin and Daniels (Reference Plomin and Daniels1987) that in relation to personality the shared rearing environment has an effect statistically indistinguishable from 0 remains highly influential and widely accepted. Second, what MacDonald & LaFreniere identify as maternal effects are effects that the researcher has classified as “shared” and then decided to classify as “maternal.” Although it is true that maternal effects are “shared” (i.e., concordance-producing) environmental effects, to measure maternal effects according to a quantitative genetic model, one must incorporate models developed in animal breeding. These include a measure of maternal performance (P′_m) as part of the offspring's phenotype (the prime indicates that the phenotypic value is a trait possessed by a different individual – the mother – than the individual being considered). For the mathematics involved in estimating maternal effects, see, for example, Bijma (Reference Bijma2006) and Chevrud and Wolf (Reference Chevrud, Wolf, Maestripieri and Mateo2009).Footnote ² As researchers in animal breeding have persistently noted, direct heritability and the response to selection are overestimated when maternal effects are not considered (Barazandeh et al. Reference Barazandeh, Moghbeli, Vatankhah and Mohammadabadi2012; Gregory et al. Reference Gregory, Cundiff and Koch1985; Koivula et al. Reference Koivula, Stranden and Mantysaari2009; Maniatis and Pollott Reference Maniatis and Pollott2002; Russell & Lummaa Reference Russell and Lummaa2009;). They also note that the genetic analysis of maternal effects has proven enormously difficult.

R4. Concluding remarks (with emphasis on twin studies)

Consider the following account of what I have termed minimal (shared) maternal effects in regard to studies of twins (purportedly) reared apart:

Similarly, Burt cites a single study of cognitive ability to suggest that the known differences between the prenatal environments of twins and singletons and between (particularly monochorionic) MZ twins and DZ twins has little effect upon behavioral phenotypes. Both Battaglia and MacDonald & LaFreniere argue that the increase of heritability with age demonstrates an imperviousness to acquired genetic or epigenetic/environmental alterations. These arguments, like almost all defenses of the twin study methodology, share one thing in common: They are denials that one or another aspect of the environment has any effect upon individual behavior. I shall call this “environmental imperviousness.”

Environmental imperviousness actually plays an important role in the twin study methodology: It enables the legitimacy of the equal environment assumption (EEA), historically the most contested aspect of the twin study methodology. In its current incarnation, the EEA does not rest upon the assumption that the environments of MZ twins do not differ in systematic ways from those of DZ twins, but rather that these systematic differences have no effect upon the behavioral phenotype under consideration (Guo Reference Guo2001). Environmental imperviousness ensures that studies of twins “raised apart” in which, for example, none of the twins are separated at birth and some are separated as late as age nine, are free of potential confounding environmental influences. Likewise, it ensures that the results of studies of twins raised together (by far the bulk of twin studies) are not confounded by shared environmental influences. And one can rest assured that if, as Burt suggests, behavior geneticists start using IVF-conceived and differently gestated and reared siblings in heritability studies, studies will appear demonstrating, for example, that the IQ of IVF children is no different from that of naturally conceived children. From this it will be inferred that although IVF children are at greater risk for a range of adverse health outcomes (just as MC–MZ twins are), nonetheless, behavioral geneticists can safely ignore this because none of their behavioral attributes are affected.

It is striking that what is presumed not to matter in human “cross-fostering” studies is deemed to be profoundly important in rodent cross-fostering studies. Consider a typical protocol for a rat cross-fostering study (van der Veen et al. 2008, p. 185):

Yet even with all of this effort to avoid potential confounding environmental influences, laboratory animal studies are plagued by unknown “cage effects” (as well as “handler effects”) that appear to influence complex behavior in profound ways (Valdar et al. Reference Valdar, Solberg, Gauguier, Cookson, Rawlins, Mott and Flint2006). Perhaps the most striking example of this concerns not rodents, but fruit flies (see below). And of course, what are not controlled for in such studies are the effects of the prenatal environment.

It is a reasonable principle that any methodology, particularly one as controversial as the twin study methodology, should be evaluated both on the basis of what the methodology presupposes (e.g., do twins in fact share 100% of their segregating genomes?) and on the basis of its results (i.e., are its results in accord with everything we know thus far about the development, behavior, and evolution of life forms from paramecium to baboons?). Twin studies fail on both of these counts.

Bluntly stated, the principle of minimal maternal effects must be wrong because, as argued in the target article, and as helpfully expanded upon by Swain, Perkins, Dayton, Finegood, & Ho (Swain et al.) and Aitken, maternal effects are omnipresent and far-reaching in human development. Denial of the importance of shared maternal effects leads to an untenable form of human exceptionalism, the hallmark of which is non-adaptive environmental imperviousness (i.e., the absence of phenotypic plasticity). The complement of such environmental imperviousness is the extraordinarily high estimates of heritability that twin studies typically yield. For example, we are told that personality is around 50% heritable (with minimal shared maternal effects) (Bouchard Reference Bouchard2004). Compare this with recent studies on aggression in fruit flies (discussed earlier). For all of the up and down regulation identified in the transcription of over 4,000 genes, heritability estimates for aggression were only 10% (Zwarts et al. Reference Zwarts, Magwire, Carbone, Versteven, Herteleer, Anholt and Mackay2011). This means that by the standard formulation of heritability, 90% of the variation in aggression was due to environment, even though the researchers assumed that they had raised the flies in identical environments.

Given what we now know about heritability of aggression in fruit flies and the extraordinary responsiveness of the developing fly to imperceptible differences in the environment, such high estimates of heritability for human behavior must be wrong. But why? Why not simply assume that the behavior of fruit flies develops in a more environmentally responsive manner, and that what is unique about humans (in addition to their use of language and writing and the extent of culture) is their degree of environmental imperviousness? The problem with such an assumption is that it is diametrically opposed to everything we know about the development of the human brain. Humans are born with brains that are developmentally incomplete. Most neuronal connections are made during infancy and early childhood, and by the time a child is 3 years old, he or she has formed about 1,000 trillion connections (Lagercrantz Reference Lagercrantz2010). The evolutionary “purpose” of these well-known features of brain development is to enable adaptive plasticity within a particular environment. High heritability combined with environmental imperviousness are incompatible with phenotypic plasticity.

Twin studies paint a picture of human behavior as characterized by extraordinarily high heritability (significantly higher than fruit flies), extraordinarily low environmental responsiveness (significantly lower than fruit flies), and prevalent genetic determinism (there is no other way to characterize the assumption that heritability increases with age). The answer as to why twin studies have yielded such a bizarre characterization of human behavior is not hard to find. The underlying assumptions that enable the methodology appear to be confirmed by the results of its application. In other words, the phenomena are interpreted (or distorted) in such a way as to enable (or legitimize) the methodology, while the methodology shapes (or distorts) the phenomenon.

Hence, we are told that the twin study methodology itself has demonstrated that whatever cannot be accommodated by, or whatever might undermine the validity of, the methodology does not matter, and it does not matter because it has no effect upon behavior. The greater similarities of pre- and postnatal environments of twins versus singletons and MZ versus DZ twins have no phenotypic effects; maternal effects, the “bane of heritability estimates,” are minimal or non-existent. Inherited differences in mtDNA – which cannot be accommodated in a model that assumes that MZ twins are genetically identical – do not matter. And as Battaglia and MacDonald & LaFreniere suggest, differences in retrotransposons, CNVs, and epigenomes do not matter either. If we combine such environmental imperviousness with the assumption that all human behavior – being a twenty-first-century American liberal or conservative; amount of time spent texting on a cell phone; consumer preferences for soups and snacks, hybrid cars, science fiction movies, and jazz – is to a large extent heritable, the result is that the subject of twin studies resembles more an automaton than a human organism (or any organism).

The paradigm that underlies twin studies presupposes a biological world characterized by simplicity, symmetry, stability, order, and predictability. It posits a limited number of causal agents – genes, shared environment, nonshared environment – whose contributions to variance in any trait of interest can be separated by simple and intuitive “natural experiments.” The principles of Mendelian inheritance ensure that genetic relatedness is a matter of simple, symmetrical fractions: MZ twins are 1 to 1, MZ twins to DZ twins are 1 to 0.5; siblings are 0.5 to 0.5; germ cells contain 50% of maternal and paternal DNA, and so on. DNA, the sole agent of heritability, is identical in all the cells and tissues of the body, is fixed at the moment of conception, and remains virtually unaffected by the environment throughout the life course, ensuring that the genetic identity of all the relevant subjects never changes. Complex phenotypes are predictable because the path from genotype to phenotype displays such regularity that by relatively simple methods (e.g., candidate gene association studies), we can predict the probability that individuals will possess complex traits on the basis of single nucleotide polymorphisms, often (though not always) without having to take into account any specific attributes of the environment.

All of these assumptions enable statistics to be the driving engine of discovery in behavior genetics. The result is that the ability to make profound discoveries in genetics appears deceptively simple: All that is required is a data set that contains either zygosity and/or genotype information for a handful of polymorphisms and behavioral data (usually in the form of self-reporting). Statistical analysis takes care of the rest. This elevation of statistics has led to the adoption of the methodologies of behavior genetics by researchers in an ever expanding number of disciplines: Economists, sociologists, political scientists, researchers in business, marketing, and management all regularly publish the results of twin and gene association studies that purport to identify the heritability of, and specific polymorphisms for, the behaviors associated with their respective disciplines. Nature, however, does not reveal her secrets so easily.

Why is it assumed that, often with little training beyond statistics, a researcher can in effect become a geneticist but not, for example, a physicist? Why is discovery in genetics deemed so much simpler than discovery in physics, particularly since DNA is a molecule and part of the account of the relationship between genotype and phenotype involves molecular processes? The answer depends upon the simplicity, symmetry, stability, order, and predictability characteristic of the genetic-biological worldview of behavioral genetics.Footnote ³ This worldview, however, is radically at odds with what cutting edge research in molecular genetics and developmental biology (to name two out of a number of scientific disciplines) is revealing. In the emerging post-genomic paradigm, we are confronted with a biological world that is in many ways the opposite of that which has thus far enabled the methodologies of behavioral genetics.