Population geneticists have historically deployed the concept of genes as statistical, rather than material, entities (Griffiths & Tabery, Reference Griffiths and Tabery2008). Although this approach may have sufficed in the era of traditional twin studies of behavior, the advent of genome-wide association studies (GWASs) seems to require some engagement with thorny questions around how variation in DNA sequences might be associated with variations in phenotype. The neoclassical view of the gene as a sequence of DNA encoding a single transcript that uniformly produces a particular protein (Portin & Wilkins, Reference Portin and Wilkins2017) has been adopted by behavior geneticists to provide a biological foundation for their otherwise purely statistical framework. However, decades of empirical findings have long exposed the deficiencies of the neoclassical view. The “postgenomic era” is replete with findings that the same sequence of DNA can be used to derive a variety of transcripts (e.g., Griffiths & Stotz, Reference Griffiths and Stotz2006; McManus & Graveley, Reference McManus and Graveley2011; The ENCODE Project Consortium, 2007, 2012), and that products derived from the same DNA sequence can exhibit divergent structures and functions depending on their cellular context (Piatigorsky, Reference Piatigorsky2007). The cellular signals driving these processes, including epigenetic modifications such as DNA methylation, partly reflect responses to an individual's social and ecological situation (e.g., Meaney, Reference Meaney2010). The activity of the genome develops selectively and responsively to fluctuating physiological conditions, undermining the idea of a prespecified, constant function inherent to a particular sequence of DNA (Neumann-Held, Reference Neumann-Held, Oyama, Griffiths and Gray2001). These realizations necessitate a shift from an “agentive” role for genes to a “reactive” view of the wider genome embedded within the organism as a developmental system (Keller, Reference Keller2014). Related lines of evolutionary thought emphasizing the primacy of development suggest that genes should be seen as followers, not leaders, of adaptive plasticity (Newman, Reference Newman2019; West-Eberhard, Reference West-Eberhard2003).
What is a behavior geneticist to do? Rather than engage with the postgenomic complexities, Madole & Harden (M&H) limit themselves to the neoclassical dogma, with DNA framed as a “specific set of instructions” (target article, sect. 3.3, para. 10) that weathers all the variability that “context” can throw at it (target article, sect. 3.3, para. 8). To an extent, they acknowledge the complexities through a “garden of forking paths” metaphor (target article, sect. 3.3, para. 10), but this path only appears to go forward, from a foundational DNA sequence that is inherited at conception and encodes the same product regardless of spatial and temporal context. However, as noted by Oyama decades ago, the metaphorical “information” in the genome is not static: It develops along with the organism (Oyama, Reference Oyama1985). Further, there is no consideration of the circular causation that runs all the way through organismal functioning (Witherington, Reference Witherington2011) and that is apparent even at the level of DNA transcription. In our view, the neglect of circular causation reflects a neglect of development, long regarded as an afterthought by the field of behavior genetics. In contrast to staid models of behavioral genetics, developmental systems perspectives allow for the multifaceted complexities of ontogeny (Gottlieb, Reference Gottlieb1995; Overton & Lerner, Reference Overton and Lerner2014).
Processes that modify DNA transcription and translation are responsive to temporal and situational changes for the organism. Consider, for example, structural brain anatomy, a phenotype popular for study in the genetics literature. Rather than exhibiting a linear growth trajectory, brain development varies over time across types of growth (e.g., cortical thickness vs. surface area), tissues (i.e., gray vs. white matter), and regions (Fjell et al., Reference Fjell, Chen, Sederevicius, Sneve, Grydeland, Krogsrud and Walhovd2019; Li et al., Reference Li, Nie, Wang, Shi, Lin, Gilmore and Shen2013). Correspondingly, genomic processes relevant to brain development vary across the lifespan as well. For example, while “clusters” of cortical thickness development (i.e., areas of the cortex showing longitudinal intercorrelation over time) were found to overlap substantially with adult “genetic clusters” (i.e., areas previously associated with shared genetic influence), there was only limited overlap between developmental and genetic clusters for cortical surface area, which suggests divergent patterns of developmental organization (Fjell et al., Reference Fjell, Chen, Sederevicius, Sneve, Grydeland, Krogsrud and Walhovd2019). Further, these kinds of developmental processes are sensitive to experience, with epigenetic influences modifying gene expression relevant to neurodevelopment in response to exposures ranging from lead poisoning and child maltreatment to maternal mental health and exercise (Fujisawa et al., Reference Fujisawa, Nishitani, Takiguchi, Shimada, Smith and Tomoda2019; Miguel, Pereira, Silveira, & Meaney, Reference Miguel, Pereira, Silveira and Meaney2019; Robakis et al., Reference Robakis, Roth, King, Humphreys, Ho, Zhang and Gotlib2022; Senut et al., Reference Senut, Cingolani, Sen, Kruger, Shaik, Hirsch and Ruden2012). Behavior geneticists, for whom linear-additive models of gene and environment account for variation in phenotypes, might overinterpret individual statistical associations between single-nucleotide polymorphisms (SNPs) and outcome measures (e.g., brain volume at a particular point in time) and leave unexamined the entwined, dynamic nature of structural brain development.
The interpretation by M&H of polygenic scores as reflecting an individual's genetic “propensity” (target article, sect. 3.1, para. 4) or “risk” (target article, sect. 3.4, para. 3) further highlights a neoclassical view of DNA as an unmoved mover. Their deployment of polygenic scores to compare the size of “genetic effects” on educational attainment across contexts (target article, sect. 3.3, paras. 8–9), for example, also assumes that polygenic scores capture polymorphisms “for” relevant traits in a context-general sense. This account fails to leave explanatory room for multifinality, obscuring plausible biological and/or social intermediaries (e.g., neurodevelopment, sociocultural biases) between SNPs and the target outcome (Kaplan & Turkheimer, Reference Kaplan and Turkheimer2021). In contrast, a developmental systems approach engages with the multilevel complexities of how phenotypic variation is generated (Gawne, McKenna, & Nijhout, Reference Gawne, McKenna and Nijhout2018) and with notions of inheritance that extend beyond DNA (Jablonka & Lamb, Reference Jablonka and Lamb2005). Such an account sees the genome as one resource (among many) used by the developmental system to grow (Overton, Reference Overton, Lerner and Overton2010) and recognizes the importance of developmental change and associated variation in psychobiological processes, such as epigenetic influences on homeostatic self-regulation (Cao-Lei et al., Reference Cao-Lei, Veru, Elgbeili, Szyf, Laplante and King2016). If the neoclassical view of DNA and genes, combined with a neglect of developmental process, remains the foundation of behavior genetics, any amount of methodological and statistical prowess in GWAS approaches will fail to move us forward in terms of understanding the complexities of human behavior.
Population geneticists have historically deployed the concept of genes as statistical, rather than material, entities (Griffiths & Tabery, Reference Griffiths and Tabery2008). Although this approach may have sufficed in the era of traditional twin studies of behavior, the advent of genome-wide association studies (GWASs) seems to require some engagement with thorny questions around how variation in DNA sequences might be associated with variations in phenotype. The neoclassical view of the gene as a sequence of DNA encoding a single transcript that uniformly produces a particular protein (Portin & Wilkins, Reference Portin and Wilkins2017) has been adopted by behavior geneticists to provide a biological foundation for their otherwise purely statistical framework. However, decades of empirical findings have long exposed the deficiencies of the neoclassical view. The “postgenomic era” is replete with findings that the same sequence of DNA can be used to derive a variety of transcripts (e.g., Griffiths & Stotz, Reference Griffiths and Stotz2006; McManus & Graveley, Reference McManus and Graveley2011; The ENCODE Project Consortium, 2007, 2012), and that products derived from the same DNA sequence can exhibit divergent structures and functions depending on their cellular context (Piatigorsky, Reference Piatigorsky2007). The cellular signals driving these processes, including epigenetic modifications such as DNA methylation, partly reflect responses to an individual's social and ecological situation (e.g., Meaney, Reference Meaney2010). The activity of the genome develops selectively and responsively to fluctuating physiological conditions, undermining the idea of a prespecified, constant function inherent to a particular sequence of DNA (Neumann-Held, Reference Neumann-Held, Oyama, Griffiths and Gray2001). These realizations necessitate a shift from an “agentive” role for genes to a “reactive” view of the wider genome embedded within the organism as a developmental system (Keller, Reference Keller2014). Related lines of evolutionary thought emphasizing the primacy of development suggest that genes should be seen as followers, not leaders, of adaptive plasticity (Newman, Reference Newman2019; West-Eberhard, Reference West-Eberhard2003).
What is a behavior geneticist to do? Rather than engage with the postgenomic complexities, Madole & Harden (M&H) limit themselves to the neoclassical dogma, with DNA framed as a “specific set of instructions” (target article, sect. 3.3, para. 10) that weathers all the variability that “context” can throw at it (target article, sect. 3.3, para. 8). To an extent, they acknowledge the complexities through a “garden of forking paths” metaphor (target article, sect. 3.3, para. 10), but this path only appears to go forward, from a foundational DNA sequence that is inherited at conception and encodes the same product regardless of spatial and temporal context. However, as noted by Oyama decades ago, the metaphorical “information” in the genome is not static: It develops along with the organism (Oyama, Reference Oyama1985). Further, there is no consideration of the circular causation that runs all the way through organismal functioning (Witherington, Reference Witherington2011) and that is apparent even at the level of DNA transcription. In our view, the neglect of circular causation reflects a neglect of development, long regarded as an afterthought by the field of behavior genetics. In contrast to staid models of behavioral genetics, developmental systems perspectives allow for the multifaceted complexities of ontogeny (Gottlieb, Reference Gottlieb1995; Overton & Lerner, Reference Overton and Lerner2014).
Processes that modify DNA transcription and translation are responsive to temporal and situational changes for the organism. Consider, for example, structural brain anatomy, a phenotype popular for study in the genetics literature. Rather than exhibiting a linear growth trajectory, brain development varies over time across types of growth (e.g., cortical thickness vs. surface area), tissues (i.e., gray vs. white matter), and regions (Fjell et al., Reference Fjell, Chen, Sederevicius, Sneve, Grydeland, Krogsrud and Walhovd2019; Li et al., Reference Li, Nie, Wang, Shi, Lin, Gilmore and Shen2013). Correspondingly, genomic processes relevant to brain development vary across the lifespan as well. For example, while “clusters” of cortical thickness development (i.e., areas of the cortex showing longitudinal intercorrelation over time) were found to overlap substantially with adult “genetic clusters” (i.e., areas previously associated with shared genetic influence), there was only limited overlap between developmental and genetic clusters for cortical surface area, which suggests divergent patterns of developmental organization (Fjell et al., Reference Fjell, Chen, Sederevicius, Sneve, Grydeland, Krogsrud and Walhovd2019). Further, these kinds of developmental processes are sensitive to experience, with epigenetic influences modifying gene expression relevant to neurodevelopment in response to exposures ranging from lead poisoning and child maltreatment to maternal mental health and exercise (Fujisawa et al., Reference Fujisawa, Nishitani, Takiguchi, Shimada, Smith and Tomoda2019; Miguel, Pereira, Silveira, & Meaney, Reference Miguel, Pereira, Silveira and Meaney2019; Robakis et al., Reference Robakis, Roth, King, Humphreys, Ho, Zhang and Gotlib2022; Senut et al., Reference Senut, Cingolani, Sen, Kruger, Shaik, Hirsch and Ruden2012). Behavior geneticists, for whom linear-additive models of gene and environment account for variation in phenotypes, might overinterpret individual statistical associations between single-nucleotide polymorphisms (SNPs) and outcome measures (e.g., brain volume at a particular point in time) and leave unexamined the entwined, dynamic nature of structural brain development.
The interpretation by M&H of polygenic scores as reflecting an individual's genetic “propensity” (target article, sect. 3.1, para. 4) or “risk” (target article, sect. 3.4, para. 3) further highlights a neoclassical view of DNA as an unmoved mover. Their deployment of polygenic scores to compare the size of “genetic effects” on educational attainment across contexts (target article, sect. 3.3, paras. 8–9), for example, also assumes that polygenic scores capture polymorphisms “for” relevant traits in a context-general sense. This account fails to leave explanatory room for multifinality, obscuring plausible biological and/or social intermediaries (e.g., neurodevelopment, sociocultural biases) between SNPs and the target outcome (Kaplan & Turkheimer, Reference Kaplan and Turkheimer2021). In contrast, a developmental systems approach engages with the multilevel complexities of how phenotypic variation is generated (Gawne, McKenna, & Nijhout, Reference Gawne, McKenna and Nijhout2018) and with notions of inheritance that extend beyond DNA (Jablonka & Lamb, Reference Jablonka and Lamb2005). Such an account sees the genome as one resource (among many) used by the developmental system to grow (Overton, Reference Overton, Lerner and Overton2010) and recognizes the importance of developmental change and associated variation in psychobiological processes, such as epigenetic influences on homeostatic self-regulation (Cao-Lei et al., Reference Cao-Lei, Veru, Elgbeili, Szyf, Laplante and King2016). If the neoclassical view of DNA and genes, combined with a neglect of developmental process, remains the foundation of behavior genetics, any amount of methodological and statistical prowess in GWAS approaches will fail to move us forward in terms of understanding the complexities of human behavior.
Financial support
This research received no specific grant from any funding agency, commercial, or not-for-profit sectors.
Competing interest
None.