From description to explanation: Integrating across multiple levels of analysis to inform neuroscientific accounts of dimensional personality pathology

Abstract

Dimensional approaches to psychiatric nosology are rapidly transforming the way researchers and clinicians conceptualize personality pathology, leading to a growing interest in describing how individuals differ from one another. Yet, in order to successfully prevent and treat personality pathology, it is also necessary to explain the sources of these individual differences. The emerging field of personality neuroscience is well-positioned to guide the transition from description to explanation within personality pathology research. However, establishing comprehensive, mechanistic accounts of personality pathology will require personality neuroscientists to move beyond atheoretical studies that link trait differences to neural correlates without considering the algorithmic processes that are carried out by those correlates. We highlight some of the dangers we see in overpopulating personality neuroscience with brain-trait associational studies and offer a series of recommendations for personality neuroscientists seeking to build explanatory theories of personality pathology.

For nearly a century, personality psychology has focused on describing how people differ from one another. This emphasis has led to a number of important advances in the field, including a growing consensus about the general structure of normal personality (John, Naumann, & Soto, 2008) and the robust role of traits in predicting important life outcomes (Soto, 2019). There is also increasing evidence that the structure of normal personality can be extended to describe abnormal personality and features of psychopathology more broadly (Kotov et al., 2017; Markon, Krueger, & Watson, 2005). These discoveries have contributed to recent systemic changes in psychiatric nosology, including the addition of a dimensionally based Alternative Model for Personality Disorders (AMPD) to the Diagnostic and Statistical Manual of Mental Disorders (DSM-5; APA, 2013), and the approval of the first officially recognized fully dimensional system for diagnosing personality pathology in the International Classification of Diseases (ICD-11; Oltmanns & Widiger, 2019).

Even as research on how to classify and describe personality pathology advances, contemporary personality science is beginning to turn toward explanation, with the goal of elucidating the causal mechanisms from which maladaptive individual differences arise. These efforts have been supported in part by unprecedented technological advances in modern neuroscience. In particular, the rise of magnetic resonance imaging (MRI) has led to a rapid proliferation of research examining the structural and functional correlates of personality traits. Despite these promising developments however, insights into the biological sources of personality and personality pathology have remained elusive, primarily for two reasons. First, early work has often yielded inconsistent or contradictory findings, likely owing to a lack of power caused by the use of small sample sizes (Allen & DeYoung, 2017; Yarkoni, 2015). Fortunately, online data repositories and large public neuroimaging datasets are beginning to remedy this problem, leading to rapid increases in sample sizes in personality neuroscience studies (Eickhoff, Nichols, Van Horn, & Turner, 2016). Here, we focus on a second and more difficult problem facing the field: the proliferation of atheoretical studies that link personality traits to neural correlates without providing a formal account of the computations performed by those correlates. This associational approach (Figure 1) aids in the discovery of circuits related to specific traits, but it does not yield mechanistic accounts of personality, or personality pathology, that span multiple levels of analysis. Instead, to pivot from descriptive to explanatory accounts, we argue that neurobiological investigations must rely on theory-driven studies that formally test, and experimentally manipulate, the hypothesized causal pathways linking a trait to its corresponding behavioral dynamics and neurobiological mechanisms.

Figure 1.

Contrasting the Marr+ approach to personality neuroscience (blue) with existing associational approaches (green). The Marr+ approach integrates descriptive findings from personality psychology with computational, algorithmic, and implementational levels of analysis. Bidirectional influences across levels of analysis leading to mechanistic insights. In contrast, associational approaches often skip multiple levels of analysis, mapping hierarchical traits directly to their neural correlates in the brain.

In the remainder of this commentary, we outline some of the issues and challenges ahead as the field embarks on the path toward explanatory neuroscientific accounts of personality pathology. We begin by highlighting the difficulties posed by the associational approach to personality neuroscience before offering recommendations that can inform the design and analysis of future research. These include harnessing the power of recent dimensional approaches to personality pathology, as well as conducting studies that have a strong theoretical basis, are grounded in findings from comparative psychology (i.e., animal research), and that seek to focally manipulate associations between traits, biology, and behavior. Finally, we conclude by highlighting approaches that hold promise for promoting new explanatory insights into the origins of personality pathology.

Personality Neuroscience: Seeking Explanation, but Stuck in Description

The story of personality neuroscience over the last 50 years has been one of remarkable growth. The field’s origins date back to Eysenck (1967), who was the first to develop a theory of the biological basis of personality traits. From there, early progress was slow, constrained by the technological limitations of the era. The first neuroimaging studies of personality did not emerge until the early 2000’s, and most featured only a handful of subjects (Canli, Amin, Haas, Omura, & Constable, 2004; Canli et al., 2001; J. R. Gray & Braver, 2002). Though few of these initial findings have replicated robustly, they nonetheless signaled the transformative influence of cognitive neuroscience approaches on the field. Today, studies of the neural correlates of personality traits are commonplace, with many reaching sample sizes in the hundreds or even thousands (e.g., Gray, Owens, Hyatt, & Miller, 2018; Owens et al., 2019).

Beyond technological innovation alone, the growth of personality neuroscience has been aided by several major psychometric advances. First, there is now general consensus that five broad factors known as the Big Five—Neuroticism, Extraversion, Agreeableness, Conscientiousness and Openness/Intellect—account for most of the covariation among more specific personality traits (John et al., 2008). Personality pathology adheres to a similar structure, suggesting that maladaptive variants of normal personality traits, such as those in the AMPD, can be used to capture patterns of covariation among psychological symptoms (Kotov et al., 2017). Finally, there is evidence that both personality and personality pathology are hierarchically structured, such that broad higher-order dimensions (e.g., the Big Five, or the domains of the AMPD) account for covariation among more specific lower-order dimensions. Recent discoveries have helped to clarify the nature and structure of levels in the hierarchy that exist both above and below the Big Five (Crowe, Lynam, & Miller, 2018; DeYoung, 2006; DeYoung, Quilty, & Peterson, 2007; Mõttus, Kandler, Bleidorn, Riemann, & McCrae, 2017), and their maladaptive counterparts (Allen, DeYoung, Bagby, Pollock, & Quilty, 2019; Wright & Simms, 2014).

Altogether, personality science and cognitive neuroscience have matured substantially over the last three decades, leading many researchers to begin searching for the neural correlates of personality traits (for reviews, see Allen & DeYoung, 2017; DeYoung, Grazioplene, & Allen, in press). Unfortunately, many of these searches have been atheoretical and exploratory in nature. Indeed, a popular strategy within the field is to obtain a large sample that includes both personality and neuroimaging measures and then attempt to link trait scores to variation in the structure or function of the brain. Often, few explicit or a priori hypotheses are provided, and analyses are run at the whole-brain level with no specific constraints on the circuits being examined. Results are collated and described with a reliance on reverse inference (Poldrack, 2006; 2011), in which the authors draw on other imaging studies to explain why a neural index might be correlated with a particular trait (without having actually tested that explanation). Often, findings are nonspecific, linking many traits to many neural correlates, with little attention paid to how these mappings substantively enhance our understanding of what personality traits are or how they lead to adaptive or maladaptive functioning.

As one example, consider recent findings from the Human Connectome Project (HCP), which examined the surface-based morphometric correlates of the Big Five in two large subsamples and a combined dataset (combined sample N = 1104; Owens et al., 2019; Riccelli, Toschi, Nigro, Terracciano, & Passamonti, 2017). Several structural indices in the dorsolateral and dorsomedial prefrontal cortices (DLPFC/DMPFC) were associated with Neuroticism, Conscientiousness, and Openness/Intellect in both subsamples and the full sample. Agreeableness was also associated with surface area, volume, gyrification, and cortical thickness of the DLPFC in one of the subsamples and the combined sample. Extraversion was negatively related to volume of the right DLPFC, but only in the full sample. In sum, there was compelling evidence that three of the Big Five were associated with the morphometry of the DLPFC and DMPFC, and at least some evidence linking all five specifically to the DLPFC. The authors conclude that the DLPFC and DMPFC may be the two regions of the brain that are most closely associated with personality in general, and they suggest that specific links between traits and DLPFC may reflect the region’s involvement in executive function and cognitive performance (Owens et al., 2019).

Though many of the associations that the authors found in this study replicated across independent subsamples, the broader pattern of findings was strikingly nonspecific, with multiple traits being linked to the same two brain regions. In some ways, this is unsurprising, as personality traits typically do not map onto brain regions in a one-to-one fashion (Allen & DeYoung, 2017; Yarkoni, 2015). Traits are broad, abstract constructs that describe a wide range of cognitive, affective, and behavioral processes, meaning that they are likely to rely on a diverse array of overlapping neural circuits (e.g., one could easily imagine Neuroticism depending on networks related to emotion regulation, attentional vigilance, self-appraisal, and simulation of the future). Links between traits and their neural correlates are therefore likely to be many-to-many, with each trait depending on many circuits, and each circuit being linked to many traits. As a result, knowing that a particular region or network is correlated with individual differences in some trait of interest typically provides little explanatory value, unless we also know what computations that region or network is responsible for, how those computations translate into observable behaviors, and how those behaviors accumulate into dispositional characteristics that remain relatively stable over time.

Explanation requires bridging multiple levels of analysis, and a strong descriptive foundation.

One of the primary reasons that associational studies are limited in their explanatory power is that they often leap across levels of analysis. A useful lens through which to view this problem was provided by Marr (1982). He articulated three levels of analysis that are involved in the study of any information-processing system: 1) the computational level, which describes the purpose of the system; 2) the algorithmic level, which specifies the processes or rules the system uses to achieve its purpose; and 3) the implementational level, which describes how these processes or rules are physically realized in the nervous system. Studies that attempt to link traits directly to neurobiology jump from the computational to the implementational level of Marr’s framework, paying little attention to the algorithmic mechanisms that are vital to understanding how the brain generates stable individual differences in personality (Hallquist, Hall, Schreiber, & Dombrovski, 2018). As a result, these studies tend to promote a naïve reductionism within personality neuroscience, as though simply mapping every trait to its corresponding synapse, cell, region, or network will somehow yield an explanatory account of the consistencies in human experience. To the contrary, because traits reflect stable patterns of thought, emotion, motivation, and behavior—all of which are generated by the brain—they will necessarily be associated with complex neural circuitry (regardless of whether they are derived from the Big Five, AMPD, or any other taxonomic system). There is little reason to believe, however, that simply identifying such circuitry will lead to a compelling neuroscientific account of how personality governs behavior.

Thus, to build explanatory accounts of personality and personality pathology, personality neuroscientists must address all three of Marr’s levels of analysis, describing the problems that traits of interest have evolved to solve (computation), testing alternative models of the lower-level processes that cause variation in those traits as observed in behavior (algorithm), and examining how those processes are represented within the physical hardware of the brain (implementation). By establishing bidirectional links between adjacent levels, one gains greater confidence in an explanatory account. For example, in an experimental task that manipulated social rank through competition, Ligneul and colleagues (2016) found that activity in the rostromedial prefrontal cortex (rmPFC) scaled with competitive prediction errors (i.e., surprise about the win/loss outcome relative to the expected outcome based on previous competitions), as estimated by a mathematical model of social dominance (i.e., an algorithmic → implementational link). Furthermore, by increasing neural excitability of the rmPFC using transcranial direct current stimulation (TDCS), they found that enhancing rmPFC activity led to greater learning from victories relative to defeats, thereby refining and validating the mathematical model of behavior (i.e., an implementational → algorithmic link).

Articulating all three levels of Marr’s framework is necessary for ensuring rigorous explanatory studies of personality pathology, but is it also sufficient? We suspect not. Even the most carefully conceived experimental studies will falter if they are based on a weak descriptive science. Rich descriptions of behavior constrain mechanistic research, inspiring increasingly precise algorithmic and implementational hypotheses, and adjudicating between competing explanations (Krakauer, Ghazanfar, Gomez-Marin, MacIver, & Poeppel, 2017). Without descriptive precision, mechanistic explanations are thus liable to be inaccurate or insufficient.

Neuroscientific investigations into personality pathology have long suffered from a flawed descriptive foundation, owing in large part to limitations of the existing categorical diagnostic system (Abram & DeYoung, 2017; Ofrat & Krueger, 2012). However, there is good reason to think that dimensional models may provide a stronger foundation for neurobiological studies of personality pathology moving forward. First, trait dimensions included in most major personality systems, including the Big Five and AMPD, tend to show continuity across species, suggesting they may adhere more closely to the organization of underlying neural systems that have evolved in response to common adaptive challenges (Gosling, 2008; Latzman, Hecht, Freeman, Schapiro, & Hopkins, 2015). Second, by formalizing patterns of covariation among personality traits, dimensional models can help to constrain and inform studies of personality neuroscience, a point which we elaborate on in the next section.

Given the advantages conferred by a strong descriptive science, we argue that personality neuroscientists should pursue a “Marr+” strategy in building explanatory accounts of personality pathology (Figure 1). This entails addressing description first, by locating a trait within existing structural models of personality pathology (e.g., AMPD), closely examining the range of behaviors with which it is associated, and considering its pattern of associations with other traits. Turning to mechanism, personality neuroscientists should then articulate all three levels of Marr’s framework using a combination of theory and careful experimentation. Adopting such an approach will align personality neuroscience with more established and mature scientific disciplines (e.g., biology) that have historically relied on formal experimentation to build mechanistic accounts of functioning. As a result, researchers will be able to combine the power of a rich descriptive framework with the rigor of theoretically guided experimentation, leading to more detailed, mechanistic accounts of personality pathology. In what follows, we offer several recommendations to aid researchers in implementing the Marr+ approach.

I. Leverage the power of dimensional, structural models of personality.

Personality psychology has long been criticized for prioritizing descriptive models over theory building (Block, 1995), but often the two endeavors are more complementary than adversarial. Indeed, the recent shift to dimensional models of personality pathology is likely to have valuable implications for personality neuroscience. For example, consider the hierarchical nature of dimensional frameworks like the AMPD, in which broader traits (e.g., Detachment) account for patterns of covariation among more specific traits (e.g., Withdrawal, Anhedonia, Intimacy Avoidance) (Markon et al., 2005; Tackett et al., 2012; Wright et al., 2012). At each level of the hierarchy, some mechanisms should account for the covariation among traits that form higher-order factors, whereas other mechanisms should account for the unique variance that distinguishes each trait from its counterparts (Abram & DeYoung, 2015; DeYoung et al., in press). Likewise, because dimensional models remove arbitrary boundaries between normal and abnormal functioning, mechanisms associated with normal range traits should also yield insights into the causal basis of pathological traits (and vice versa).

Personality neuroscience may benefit from considering these features of dimensional models. For instance, Grazioplene and colleagues (2016) recently found that adaptive and maladaptive traits related to Openness were inversely associated with white matter integrity in the frontal lobes after controlling for Intellect, the other major trait that comprises the Openness/Intellect domain (suggesting differential specificity at one level of the hierarchy). Openness is positively correlated with AMPD Psychoticism (Chmielewski, Bagby, Markon, Ring, & Ryder, 2014; DeYoung, Grazioplene, & Peterson, 2012), and severity on the psychosis spectrum is also associated with reductions in frontal white matter integrity (Carletti et al., 2012; Cho et al., 2016; Katagiri et al., 2015; Skudlarski et al., 2013). Thus, this finding capitalizes on a structural distinction in the hierarchy to provide corroborating evidence—at a different level of analysis—for the hypothesis that Openness exists on a common spectrum with Psychoticism (an implementational → computational link; Chmielewski et al., 2014; DeYoung et al., 2012).

Moving forward, these kinds of associational insights, which appear to support convergence between descriptive and implementational levels of analysis, can serve as useful starting points for further experimental work that interrogates the cognitive processes (algorithmic mechanisms) that contribute to personality and personality pathology. For example, there is some evidence that individuals high in Openness have more flexible and interconnected semantic networks (Christensen, Kenett, Cotter, Beaty, & Silvia, 2018), which may allow them to access more remote associations during creative tasks. Indeed, performance on divergent thinking tasks, in which subjects are asked to generate as many distinct answers to an open-ended problem as possible, is also negatively associated with frontal white matter integrity (Jung et al., 2010). Other evidence indicates that creative idea generation reflects a lack of semantic interference by salient conceptual knowledge (Beaty, Christensen, Benedek, Silvia, & Schacter, 2017), and this may represent a promising algorithmic mechanism for Openness. Indeed, performance on a creative task requiring participants to generate verbal responses in the presence of interfering semantic content is linked to increased functional coupling between the executive control and default network, (Beaty et al., 2017), whereas scores on the Openness-Psychoticism spectrum are positively associated with default network coherence, and negatively associated with frontoparietal coherence (Blain, Grazioplene, Ma, & DeYoung, 2019).

Taken together, these findings point to specific algorithmic (excessive idea generation caused by a lack of semantic interference) and implementational (poor prefrontal constraint of default network activity) hypotheses about the nature of Openness. At the extreme, these mechanisms could also underlie the unusual, eccentric, and even delusional beliefs experienced by individuals high in Psychoticism. Such insights are only possible however, when findings are linked to psychometric research that distinguishes Openness from Intellect, and unites Openness and Psychoticism on a common spectrum. Moving forward, these hypotheses can be further validated via the incorporation of more rigorous experimental manipulations (e.g., varying degrees of semantic interference, use of neurostimulation, incorporating task-based fMRI) that place a strong emphasis on observed behavior. Indeed, it is only by assessing the behavioral outputs of traits that we can move closer to identifying the relevant algorithmic mechanisms that will likely serve as targets for future intervention.

The hierarchical structure of personality may also help to generate insights into the level of brain organization that is likely to correspond best to a given trait. This is because traits closer to the top of the hierarchy typically respond to extremely broad classes of stimuli (e.g., punishments, rewards, uncertainty), making them relevant to almost any situation an individual is likely to encounter. In contrast, traits lower in the hierarchy describe responses to more specific classes of stimuli that are present in a limited range of contexts (DeYoung, 2015). Given the range of behaviors with which they are associated, higher-order traits will almost certainly be tied to neural correlates that are distributed diffusely throughout the brain (Abram & DeYoung, 2015; Depue & Lenzenweger, 2015). This is one reason why DeYoung (2006) has hypothesized that personality’s two meta-traits, known as Stability and Plasticity, may be associated with global levels of serotonin and dopamine, respectively (DeYoung, 2013; Wright, Creswell, Flory, Muldoon, & Manuck, 2019). Serotonin and dopamine, which fall into a class of neurotransmitters known as the monoamines, are well-suited to subserve higher-order trait functions because they are phylogenetically old, ubiquitous throughout the brain, and involved in modulating a vast collection of behaviors that are central to personality (Depue, 1995; Depue & Collins, 1999). Thus, the scope of a given trait can aid personality neuroscientists in determining the appropriate level of brain organization from which to study it; broad traits that are conserved across many species are likely to be associated with systems that are distributed diffusely throughout the brain, whereas lower-order traits that show less extensive conservation are more likely to reflect the co-opting of specific circuits within a broader system (and even more likely, the co-opting of circuits from more than one monoaminergic system). This principle could also have important implications for understanding personality pathology, particularly as the field begins to grapple with the causal basis of the p-factor (the higher-order factor that emerges from the positive manifold of psychiatric symptoms). Indeed, the p-factor has a similar psychometric profile as Stability (DeYoung & Krueger, 2018), and is associated with an extremely broad swath of maladaptive behaviors, suggesting it is likely to find its origins in a neural mechanism that affects diffusely distributed brain systems.

Even when a specific trait is not well-described within the hierarchical structure of personality, knowing its relationships to other trait constructs can inform studies of personality neuroscience. The hierarchy is, after all, a heuristic. Personality does not have simple structure, meaning that many lower-order traits will covary with traits outside the primary domain on which they load (Hopwood & Donnellan, 2010). Some individual traits may exist in an interstitial space, cross-loading on multiple higher-order domains (Krueger & Markon, 2014). One such example is the facet of Suspiciousness, which is assigned to the Detachment (low Extraversion) domain in the AMPD, but which consistently exhibits high cross-loadings on two, and sometimes even three other domains: Negative Affect (the maladaptive counterpart to high Neuroticism), Antagonism (low Agreeableness), and less often, Psychoticism (likely low Openness, as discussed above) (Griffin & Samuel, 2014; Krueger, Derringer, Markon, Watson, & Skodol, 2012; Thomas et al., 2013).

The implication of these cross-loadings for neuroscientific investigations is that Suspiciousness is likely to be a blend of neurobehavioral systems that span several domains. For example, individual differences in Suspiciousness may depend on an attentional bias toward threat and uncertainty (reflecting its cross-loading on Negative Affect; Shackman et al., 2016) that becomes pathological when individuals are also prone to detecting false positives (reflecting its cross-loading on Psychoticism; Blain, Longenecker, Grazioplene, & DeYoung, 2020). These tendencies may take on an inherently social quality when paired with difficulties reasoning about the mental state of others (reflecting its cross-loading on Antagonism; Allen, Rueter, Abram, Brown, & DeYoung, 2017), or with deficits in the hedonic enjoyment of interpersonal interactions (reflecting its cross-loading on Detachment; Watson, Stanton, Khoo, Ellickson-Larew, & Stasik-O’Brien, 2019). Knowing how Suspiciousness covaries with traits within the personality hierarchy allows us to leverage the theory surrounding those traits to examine whether Suspiciousness can potentially be conceptualized in terms of multiple systems or processes that converge to produce a complex phenotype.

Explaining complex traits is a difficult problem, and we should not attempt to force a simple, low-dimensional model onto the high-dimensional space of personality (Jolly & Chang, 2019). Nonetheless, the risk of studying such complexity directly is that it may afford the investigator too many interpretive degrees of freedom. For example, if an intruder detection paradigm elicited more basolateral amygdala activity during an fMRI scan, and such activity scaled with individual differences in Suspiciousness, should we conclude that the amygdala somehow encodes this trait, or that this region is computing threat? An alternative account is that amygdala activity in this paradigm reflected sensitivity to novelty (e.g., Camalier, Scarim, Mishkin, & Averbeck, 2019), not threat, and that equivalent correlations with Suspiciousness would have emerged in an auditory oddball task. To avoid such difficulties, it may be useful for personality neuroscientists to focus on developing accounts of traits whose neurobiological interface may be simpler (e.g., dopamine’s role in generating motivated approach behaviors; Guitart-Masip, Duzel, Dolan, & Dayan, 2014) or more easily described in animal models (e.g., threat-related behavioral inhibition). Indeed, understanding these traits may provide crucial building blocks for investigations into interstitial constructs that have complex emergent properties. Moreover, by developing a deeper understanding of the biological systems that contribute to common behavioral repertoires that are conserved across species, personality neuroscientists may even be able to leverage biological evidence to further refine and validate existing trait models.

II. Let theory be your guide: How detailed explanations of behavior can constrain neuroscientific accounts of personality.

With every advance in neuroscience technology, the number of neural indices that we can identify and measure grows exponentially. For personality neuroscientists, linking multifarious neural measures to the wide universe of traits is an exceptionally difficult problem that cannot be overcome without the aid of a strong theory (Abram & DeYoung, 2017; Depue, 1995). Though structural models such as the AMPD and Big Five are atheoretical taxonomies, personality is nonetheless rich with explanatory theories, both new and old, that identify plausible mechanisms responsible for generating individual differences (Denissen & Penke, 2008; Depue & Lenzenweger, 2015; DeYoung, 2015; Fleeson & Jayawickreme, 2015; Gray & McNaughton, 2000; van Egeren, 2009). These theories are indispensable tools, because they help to generate hypotheses about all three levels of Marr’s framework, highlighting the adaptive function of traits (computational level) as well as their possible instantiation in psychological (algorithmic level) and neurobiological (implementational level) processes. In doing so, they connect personality psychology to a variety of other disciplines—cognitive neuroscience, computational psychiatry, developmental psychology, and ethology, to name just a few. Empirical findings from these disciplines can then be leveraged to formulate a priori hypotheses about the origins of individual differences.

One of the main advantages of operating from a strong theoretical foundation is that it pushes one to define Marr’s (1982) highest level of analysis, the computational level, which describes the function or objective of the information-processing system. The computational level of analysis asks, “What is the problem this computation is solving? What purpose does it serve?” In personality psychology, researchers have tried to address these questions by positing that traits reflect variation in systems that respond to classes of stimuli that have been present over the course of evolutionary time, and which hold adaptive significance to humans and other animals (Denissen & Penke, 2008; Depue & Collins, 1999; DeYoung, 2015; Gray, 1973). Examples of these types of stimuli include things like rewards, punishments, and conspecifics— stimuli that humans encounter in a wide variety of situations, and which they must learn to navigate effectively in order to increase their chances of survival and reproduction. As an example, many personality theories posit that Extraversion (the opposite pole of Detachment in the AMPD) reflects between-person variation in mechanisms that enable individuals to respond adaptively to rewards (Depue & Collins, 1999; DeYoung, 2015; Gray & McNaughton, 2000). Indeed, Extraversion is associated with enhancements in cognitive processing following positive primes (Robinson, Moeller, & Ode, 2010), stronger affective reactions following appetitive cues (Smillie, Cooper, Wilt, & Revelle, 2012), and increased motivation and learning under conditions of reward (Skatova, Chan, & Daw, 2013; Smillie, Dalgleish, & Jackson, 2007). Each of these findings is contingent on the presence of an appetitive or rewarding cue being present within the experimental context, a pattern that is consistent with the notion that individual differences in a trait will only manifest when the stimuli to which they respond are present (DeYoung, 2015).

This insight provides one example of how personality theory can offer practical implications for studies of personality neuroscience. If we can only observe the behavioral output of a trait in the presence of certain relevant stimuli, then it seems reasonable to assume those same stimuli will be necessary to observe the trait’s neural basis as well. For example, in several recent studies, Extraversion was positively associated with greater left frontal cortical asymmetry, but only when the electroencephalogram (EEG) recording was made in an approach-relevant motivational context (e.g., when interacting with an attractive experimenter, Wacker, Mueller, Pizzagalli, Hennig, & Stemmler, 2013; or following a positive emotion induction, Wacker, 2018). Others have similarly found that, in both healthy and depressed participants, symptoms of depression associated with low motivation (which typically load on the low pole of Extraversion) were associated with decreased relative left frontal cortical activity, but only during reward anticipation (Nelson, Kessel, Klein, & Shankman, 2018). In contrast, recent meta-analyses have repeatedly found that Extraversion is unrelated to left frontal asymmetry at rest (Kuper, Käckenmester, & Wacker, 2019; Wacker, Chavanon, & Stemmler, 2010).

This kind of interactional effect, in which links between trait and biology depend on the presence or absence of some trait-relevant stimulus, has also been found with other traits and methodologies (e.g., Everaerd, Klumpers, van Wingen, Tendolkar, & Fernández, 2015; Stemmler & Wacker, 2010). The implication is that personality neuroscientists need to think carefully about the types of stimuli that are likely to elicit individual differences in a given trait. In some cases, task-based paradigms may offer a unique advantage to researchers by enabling them to tap into individual differences that would otherwise not be apparent from data collected at rest (Greene, Gao, Scheinost, & Constable, 2018). Understanding the neural basis of rejection sensitivity, for instance, will undoubtedly require the use of paradigms in which other social actors are present, and rejection is at the very least a real possibility for the participant. On the other hand, one must also be careful, when using experimental paradigms, that the strength of a robust task manipulation does not trump the ability to detect individual differences in task behavior (e.g., if an experimentally-induced rejection is too severe, all participants will report near-ceiling levels of rejection, obscuring individual differences; Hedge, Powell, & Sumner, 2018). Finally, one should bear in mind that the flow of knowledge in this case need not always run in the direction of theory to empiricism; studies that empirically examine the types of stimuli that do, or do not, elicit trait differences, both behaviorally and neurally, can be a vital means for refining our theoretical understanding of traits (e.g., knowing whether randomly-generated criticism from a computer elicits the same individual differences in rejection sensitivity as criticism from a peer shapes our theoretical understanding of rejection sensitivity).

Altogether, strong theories encourage researchers to form a detailed representation of a trait, including the problem that the trait has evolved to solve, its observable manifestation in behavior, and the potential cognitive and biological processes that underlie such manifestations. The influential theories of Depue and colleagues provide strong examples of this general approach, synthesizing empirical findings from a variety of disciplines to understand both the algorithmic and implementational levels of a trait (Depue, 1995; Depue & Collins, 1999; Depue & Lenzenweger, 2015). These theories have typically followed a productive four-pronged strategy: 1) build on descriptive and psychometric data from personality psychology to describe a trait’s behavioral, cognitive, affective, and motivational characteristics; 2) consult the animal literature to identify analogous behavior patterns in mammals or other species; 3) use neurobiological research in animals to generate hypotheses about the neural representation of that behavior pattern in humans, and 4) test these hypotheses in humans using careful experimentation and neurobiological assessment. In practice, this approach closely resembles the Marr+ framework we advocate for here.

In one influential application of this approach, Depue and Collins (1999), building off earlier work by Gray (1994), drew a link between individual differences in Extraversion (specifically agentic Extraversion) and an incentive motivation system that is present in many animals. Incentive motivation is mediated by dopaminergic projections that originate in the ventral tegmental area (VTA) of the midbrain and innervate a variety of cortical and subcortical regions, including the ventral and dorsal striatum, amygdala, anterior cingulate, and medial orbitofrontal cortex. In theorizing a connection between Extraversion and the incentive motivation system in animals, Depue and Collins provided personality neuroscientists with a promising proposal for linking Extraversion with neural indices, steering studies of Extraversion toward dopamine and the brain’s reward system. The link between Extraversion and dopamine, demonstrated in a number of studies using pharmacological manipulations, is now one of the most robust findings in all of personality neuroscience (Chavanon, Wacker, & Stemmler, 2013; Depue, Luciana, Arbisi, Collins, & Leon, 1994; Mueller et al., 2014; Rammsayer, 1998; Rammsayer, Netter, & Vogel, 1993; Wacker, Chavanon, & Stemmler, 2006; Wacker et al., 2013; Wacker & Stemmler, 2006). Similarly, neuroimaging studies have provided compelling evidence that Extraversion is associated with structural and functional alterations throughout the mesolimbic reward system (Civai, Hawes, DeYoung, & Rustichini, 2016; Cremers et al., 2011; DeYoung et al., 2010; Grodin & White, 2015; Lewis et al., 2014).

Depue and Collins’ (1999) theory also provided some insight into the algorithmic processes represented by reward-related circuits. For example, the theory noted that activation of VTA dopamine neurons reflects both the presence and magnitude of incentives in the environment, suggesting that midbrain dopaminergic projections may play a role in linking reward cues to subsequent behavioral responses (i.e., approach behavior) (Depue & Collins, 1999). Other animal research has shown that dopaminergic projections from the VTA to the nucleus accumbens and anterior cingulate increase following unpredicted rewards and decrease following unpredicted omissions of reward (Bromberg-Martin, Matsumoto, & Hikosaka, 2010; Schultz, 1998; 2007). This pattern mimics classic reinforcement learning models, in which a prediction error signal encoding the discrepancy between actual and predicted outcomes trains an actor process (i.e., a process that surveys the current environmental state and determines the best action to take) to associate stimuli with rewards or punishments (Sutton & Barto, 1998). Thus, dopaminergic projections from the VTA may constitute a reward prediction error (RPE) signal that helps an organism learn the subjective value of cues or actions in order to shape adaptive behavioral responses (for a useful overview, see Glimcher, 2011).

In humans, this RPE can be captured via an EEG waveform known as the reward-related positivity (RewP; previously referred to as the feedback-related negativity) (Proudfit, 2015). The RewP is a positive-going waveform that appears about 250–300ms after receiving feedback about some outcome and has been linked to dopaminergic signaling in the anterior cingulate (Hauser et al., 2013). Consistent with an RPE account, the waveform spikes following better-than-expected outcomes and declines below baseline following worse-than-expected outcomes (Sambrook & Goslin, 2015). In a notable extension of the animal literature, several studies in personality neuroscience have now shown that Extraversion is associated with the amplitude of the RewP (Bress & Hajcak, 2013; Cooper, Duke, Pickering, & Smillie, 2014; Lange, Leue, & Beauducel, 2012; Smillie, Cooper, & Pickering, 2011), and that the effect is dopaminergically mediated (Mueller et al., 2014). Interestingly, the link between Extraversion and the RewP also shows stability over time, as Kujawa and colleagues (2015) have found that positive emotionality (a temperament analogue to Extraversion) in preschoolers predicts the amplitude of the RewP, both concurrently at age three and six years later at age nine. This finding is particularly interesting in light of fMRI evidence showing temporal stability in the link between Extraversion and activity in the ventral striatum during reward anticipation, pointing to a potential cross-method convergence of findings (Wu, Samanez-Larkin, Katovich, & Knutson, 2014).

Taken together, this line of research represents an impressive demonstration of how a theory that makes predictions spanning traits, behavior, and biology can aid in the generation of stringent, falsifiable hypotheses. These hypotheses can then be tested in careful experiments and used to advance explanatory accounts of personality and personality pathology. Indeed, Depue’s theory of Extraversion, and the subsequent work it has inspired, has led to a series of important insights into personality pathology. To note just one example, there is evidence that the magnitude of the RewP increases from childhood to adolescence (Burani et al., 2019; Speed et al., 2018), and that changes in the volume and activity of the striatum during this same period correspond with changes in agentic Extraversion (Braams, van Duijvenvoorde, Peper, & Crone, 2015; Urosevic et al., 2012). Taken together, this suggests that developmental shifts in the magnitude of dopaminergically-driven striatal RPEs may underlie adolescent-specific reward-seeking behavior, providing a potential target for future intervention efforts.

Ethological perspectives should be used to inform and constrain personality neuroscience research.

One of the many reasons why Depue and colleagues’ theories have proven instructive to modern personality neuroscientists is their emphasis on ethology, or the scientific study of animal behavior under its natural conditions (Mobbs & Kim, 2015). If personality traits reflect the evolution of systems designed to deal with motivationally relevant stimuli, then understanding how those systems have been conserved across species is crucial for understanding the biological basis of traits. Many traits have their roots in neural systems that are evident in other mammals and reptiles (Latzman, Boysen, & Schapiro, 2018; Latzman et al., 2015), suggesting that animal studies, which often yield stronger evidence of causation via direct manipulation of both the environment and biology (e.g., genetic or hormonal manipulations, knockout studies, lesion experiments, planned postmortem tissue analysis; Gosling, 2008), can provide key sources of inspiration to personality neuroscientists. Greater control over experimental conditions enables animal researchers to more precisely define the ecological contexts that give rise to specific trait-relevant behaviors, many of which have analogues in humans. For instance, animal work has played a vital role in shaping our understanding of the ecological contexts that elicit distinct defensive behaviors, with evidence suggesting that more ambiguous threats tend to elicit defensive quiescence, whereas more imminent and clearly dangerous threats tend to be associated with fight, flight, freeze responses (Blanchard & Blanchard, 1989).

Many scientists are already engaged in the type of animal work that has informed modern personality theories (Gosling, 2001; Latzman, Green, & Fernandes, 2017). Their findings can serve to both constrain and inform investigations into personality and personality pathology in humans. For example, understanding the adaptive significance of particular patterns of behaviors in animals can help personality neuroscientists to understand the types of problems that are solved by analogous behaviors in humans (enabling one to effectively articulate the computational level of analysis in the Marr+ framework). Returning to our previous example, animal work has proven to be central to understanding the distinct roles of anxiety and fear in humans. Specifically, trait anxiety has been proposed as a human analogue to defensive quiescence in animals, suggesting it may function to promote risk assessment in the presence of potential danger. In contrast, fear has been linked to fight-flight-freeze behavior, suggesting it facilitates undirected escape in the context of imminent danger (J. A. Gray & McNaughton, 2000).

Clarifying the classes of stimuli, and ecological contexts, that elicit trait-relevant behaviors can also improve the precision of our experimental designs in humans. For instance, animal work on defensive behavior has led researchers to develop human paradigms in which participants make escape decisions based on the proximity of a threat (Qi et al., 2018). Findings from this work have shown that as threats become more imminent, neural activation shifts away from circuitry related to risk assessment, and toward circuitry implicated in motivating escape (Mobbs et al., 2007; Qi et al., 2018). Further, in line with expectations based on the animal work, trait anxiety has been found to predict escape decisions in this paradigm and to scale with the functional coupling of regions related to risk assessment (ventral hippocampus and medial prefrontal cortex), but only in the context of a slow-attacking (i.e., distant) predator (though, caution is warranted given the limited sample size; Fung, Qi, Hassabis, Daw, & Mobbs, 2019). Notably, this result provides a useful example of how articulating Marr’s computational level of analysis, by identifying the distinct problems trait anxiety and trait fear have evolved to solve, can have downstream consequences on our understanding of how those traits affect behavior, and are instantiated in the brain. Moving forward, personality neuroscientists may benefit from understanding the precise sequence of actions an animal implements to navigate different types of threats, as well as other evolutionarily salient challenges, as these actions may help to generate hypotheses about the algorithmic operations that underlie analogous behaviors in humans (Krakauer et al., 2017).

III. To understand a system, push it around and observe what happens

A fundamental challenge in developing explanatory accounts of personality pathology is that nearly all of personality neuroscience is inherently correlational. Due to both practical and ethical constraints, personality neuroscientists usually have very limited control over their primary independent and dependent variables. As a result, it is difficult to know whether any observed association between brain functioning and personality reflects a causal relation, or simply a downstream consequence of a trait’s true cause (DeYoung et al., in press). This applies even to structural studies, as gross anatomical structure is often more malleable than typically assumed (Yarkoni, 2015).

The reality is that true causal explanations are likely to remain elusive in personality neuroscience so long as we are constrained by an inability to manipulate personality. Nonetheless, not all correlational evidence is equal, and some approaches provide greater support for the causal hypotheses embedded within personality theories than others. In particular, studies that “push the system around,” or formally manipulate the links among traits, behavior, and biology are likely to provide some of the strongest evidence in building explanatory accounts of personality pathology, even if they cannot definitively prove causation. Since personality traits cannot easily be manipulated, pushing the system around typically entails altering either the ecological conditions that evoke trait-relevant behaviors or the neurobiological systems hypothesized to underlie those behaviors. In what follows, we briefly review two methods for implementing these types of experimental manipulations and discuss how those methods have contributed to a more precise understanding of the neurobiological systems that underlie individual differences in personality pathology.

Pharmacological manipulations.

Within the cognitive neuroscience toolkit, some experimental approaches allow researchers to test hypotheses about a region, system, or neurotransmitter by manipulating that substrate directly and observing its impact on behaviors relevant to a trait of interest. Recent technological advances, including the growing prevalence of neuromodulatory techniques (e.g., TDCS and transcranial magnetic stimulation), have broadened the menu of experimental options available to manipulate neurobiology. These methods are likely to have increasing relevance for personality neuroscience in the years to come (Crockett & Cools, 2015). At the same time, older techniques, such as pharmacological manipulations, have already proven to be essential tools in explicating the role of specific neurobiological systems in personality and personality pathology.

Pharmacological manipulations typically involve the administration of a specific drug that acts on a neural substrate of interest — usually a neurotransmitter system (note that drugs can have multiple mechanisms of action, so selecting the appropriate agent is vital in these studies). The system targeted by the drug is implicated in personality when a trait moderates its effect on behavior or neurobiological functioning (as assessed via imaging, EEG, or the measurement of a downstream metabolite of the neurotransmitter of interest). For example, as we noted above, one prominent hypothesis in personality neuroscience has been that the metatrait Stability is associated with global functioning of the serotonergic system (DeYoung, 2006). This hypothesis has been indirectly supported by a series of studies showing that traits closely associated to the Big Five domains encompassed by Stability—Agreeableness, Conscientiousness, and Neuroticism—moderate the effect of serotonin manipulations (Brummett, Boyle, Kuhn, Siegler, & Williams, 2008; Kamarck et al., 2009; Manuck, Flory, Ferrell, Mann, & Muldoon, 2000; Manuck et al., 1998). More recently, the first direct evidence supporting the serotonin-Stability hypothesis also emerged, as Wright and colleagues (Wright et al., 2019) found that Stability is associated with individual differences in prolactin response — a hormone that is released from the pituitary in response to increases in serotonin — following the intravenous administration of citalopram, a selective serotonin reuptake inhibitor (SSRI).

Standard pharmacological challenges are beneficial for examining the neurobiological basis of broad traits like Stability, which, as noted previously, is likely to have correlates that are distributed diffusely throughout the brain. However, these types of studies suffer from the same limitations as many associational studies, in that they do not, in isolation, provide much evidence regarding the algorithmic operations that are being carried out by a given neurotransmitter system. One way to circumvent this limitation, and to generate evidence regarding the specific function a neurotransmitter plays in causing individual differences, is to combine pharmacological manipulations with behavioral experiments. For example, in a series of recent studies, Crockett and colleagues found that pharmacologically increasing synaptic serotonin enhanced participants aversiveness to actions that caused harm to themselves or others (Crockett, Clark, Hauser, & Robbins, 2010; Crockett & Cools, 2015). Moreover, the effect of serotonin on harm aversion was moderated by trait empathy as measured by the Interpersonal Reactivity Index, a scale that blends features of Neuroticism and Agreeableness (Crockett et al., 2010). As a next step, it would be interesting to test whether Stability also moderates the link between serotonin and harm aversion. If so, it might suggest that one function of Stability — and serotonin — is to modulate the computed value of actions that are likely to have disruptive or aversive consequences.

Other studies have adopted a similar approach to clarifying individual differences in Extraversion. For example, Depue and Fu (2013) recently examined whether the link between Extraversion and incentive motivation is maintained by an enhanced conditioning of contexts that predict reward. Extending a task from the animal literature to humans, the authors paired a drug reward (methylphenidate) or placebo with two different laboratory contexts in both high and low extraverts, and had participants complete a series of dopaminergically mediated behavioral tasks. They found that high extraverts were more sensitive to the rewarding effects of methylphenidate, such that they exhibited better behavioral performance on the tasks. More importantly, however, high extraverts who received methylphenidate in a particular lab context continued to show performance improvements in that context even in the absence of methylphenidate, suggesting that Extraversion may be characterized by an enhanced ability to associate previously neutral stimuli with the experience of reward. These results help to clarify the ways in which individual differences in Extraversion are maintained over time. By associating contextual stimuli with rewarding cues, Extraverts may acquire a broader network of contexts that elicit incentive motivation, leading to the increased expression of reward-seeking behaviors.

Taken together, these studies represent the potential of pharmacological manipulations for building stronger accounts of the neurobiological components of personality, particularly when combined with a strong theory and careful behavioral experimentation. Nonetheless, it is important to bear in mind that pharmacological manipulations are ultimately correlational in nature, as evidence that a trait moderates the effect of a drug on some neural system does not definitively mean that system causes the trait itself. Yet, even despite these limitations, pharmacological manipulations offer a helpful degree of experimental control that is often unavailable with other methods; using drugs to push the system around is certainly not a perfect technique, but it has proven to be an effective one for learning about the causal sources of personality and personality pathology.

Formal models of behavior and decision-making.

Studies that simultaneously assess behavior and brain function represent a promising avenue for research on the causal basis of personality pathology. However, even with behavioral data in hand, it can be difficult to understand the latent cognitive processes that underlie the behavior, let alone how the brain implements such processes. Indeed, the proliferation of cognitive tasks and neural correlates has increasingly forced cognitive neuroscience to grapple with challenges to specificity (e.g., is the neural architecture supporting social rewards different from other reward modalities? Ruff & Fehr, 2014) and problems with part-versus-whole relationships (e.g., does cognitive control depend on intact working memory circuits?). Behaviorally, the field has more explicitly wrestled problems of task impurity, in which multiple dissociable cognitive processes or systems contribute to a unitary decision (Burgess, 1997). To navigate these challenges, Poldrack and others have argued for the value of pursuing explicit cognitive ontologies that specify the putative cognitive processes/entities and their relationships for any given study or paradigm. Furthermore, they have encouraged cognitive neuroscientists to develop an atlas that maps hypothesized relationships among tasks, processes, and constructs in order to promote convergence among studies and share criteria by which to judge advances in cognitive and biological theories.

Personality neuroscientists already have a good description of how traits map onto each other in a hierarchical structure, but there is much less consensus about mental processes that are specific to each trait and under what controlled conditions one trait would give rise to different behaviors than another. As Poldrack and Yarkoni (2016, p. 604) ask, “Will concepts such as love and hate find their place in a formal ontology of the mind, or are they merely folk psychological abstractions to be abolished as science progresses, in the way that some philosophers once envisioned (Churchland, 1981)?”. Although we need not pursue the eliminative materialism approach of Churchland and others, we must understand better not only how broader traits subdivide into more specific ones, but also the mechanisms that cause them to do so.

Formal mathematical models of behavior and decision-making can help to address this problem, as they encourage researchers to develop, refine, and validate process-based accounts of the latent constructs that contribute to behavior on a given task. Crockett (2016) likens the development of formal models to baking a cake in which one first identifies necessary ingredients, measures their amounts, combines them in a certain order, and finally bakes the mixture. Applying this analogy to personality neuroscience, we can think of ingredients as the cognitive processes that contribute to trait-relevant behavior. Once these are established, formal mathematical models can be used to test competing hypotheses about how much of each of these processes is needed, and in what combination, to produce the behavior of interest (Crockett, 2016). From there, we can begin to think about how relevant personality constructs might relate to variations in our recipe—how does adding a little more of one ingredient, and a little less of another, change the phenotypic expression the trait we are interested in?

The utility of a formal mathematical model lies in its ability to represent observed behavior in terms of hypothesized latent processes that are reflected by the form (equations) and parameters of the model. As one example, sequential sampling models – most prominently the drift diffusion model (DDM; Ratcliff & McKoon, 2008; Ratcliff, Smith, Brown, & McKoon, 2016) – provide a mathematical account of how decision makers choose between two (or more) choices in timed decision tasks. According to the DDM, when a stimulus is presented, participants begin to accumulate noisy evidence in favor of both options. As the relative difference in evidence for one option over another increases with time, participants may cross a decision threshold and respond in favor of the option with stronger evidence. For example, consider the task of determining whether a set of pictures, presented sequentially, are either examples of houses or faces. When a given picture is presented, one scans the picture visually and accumulates relative evidence for it being a house or face, only responding when there is sufficiently strong evidence for one option or the other, depending on time pressure.

This process is parameterized via a formal equation with parameters representing the rate at which information accumulates towards threshold (drift rate) and how much information is needed to reach a decision (boundary). Many variants of the DDM have been developed to test alternative hypotheses about decision processes in a variety of decision tasks (see Ratcliff et al., 2016). Importantly, the DDM makes quantifiable predictions that can be compared to other sequential sampling models of two-choice decision making in the same data, thus investigating the evidence for one model’s superiority over another (Ratcliff & Smith, 2004). While the DDM is just one of many generative models for behavioral tasks, it has an extensive history in cognitive and mathematical psychology, making it a recent target for researchers interested in personality and individual differences (Johnson, Hopwood, Cesario, & Pleskac, 2017).

By comparing competing models on the basis of their ability to reproduce manifest behavior, formal models of decision making can effectively decompose gross behavioral output into the latent processes that contribute to its manifestation (Patzelt, Hartley, & Gershman, 2018). Formal models also force researchers to cast their theories in computational terms, which may aid in overcoming the tendency of humans to construct overly simplistic theories (Jolly & Chang, 2019) that lack the ability to be falsified or make explicit predictions. Once a model is found to provide a reasonable fit to behavior, one can map the cognitive processes represented by the model to its neural implementation in the brain by correlating model-based signals that vary from trial to trial with neural measures (e.g., fMRI measured throughout the task). Such a process has led to pioneering discoveries in neuroscience, including one line of research showing that the subthalamic nucleus provides a “hold your horses” (Frank, 2006) signal that modulates the decision boundary in a DDM framework, a finding that has been further corroborated by neurostimulation studies in humans (Cavanagh et al., 2011). This is just one example of how interrogating the algorithmic operation that gives rise to a particular pattern of behavior can lead to subsequent novel insights into how the brain processes information. Extending these findings to the realm of personality pathology, one could imagine conducting a series of studies testing whether individual differences in decision boundary are associated with Disinhibition. If so, this might suggest that Disinhibited individuals are prone to making decisions before they have accumulated enough evidence to minimize future risk. By capturing this phenomenon across levels of analysis (via behavior and neurobiology), personality pathologists not only gain confidence in their mechanistic account, but they also gain additional targets for prevention and intervention.

At this point, formal models of behavior and decision-making are relatively novel within the literature on personality and personality pathology. Nonetheless, these models provide promising avenues for future research, as they can serve as vital tools in the effort to identify the latent algorithmic processes that contribute to personality function and dysfunction (Hallquist et al., 2018; Patzelt et al., 2018). By integrating computational modeling techniques with the vast array of neuroscientific tools at researchers’ disposal, personality neuroscientists may be able to uncover more mechanistic theories of personality pathology that provide clear predictions for the cognitive processes that bridge between the brain and trait-relevant behavior.

Conclusion

Empirically derived, dimensional approaches to psychiatric nosology are currently reshaping the way researchers and clinicians conceptualize and define personality pathology. This transformation is producing a new era of personality pathology research that overcomes many of the limitations inherent in categorical diagnoses and the traditional case-control design. Having shown the empirical benefits of a dimensional approach, the field is now poised to begin building explanatory accounts of personality pathology. Personality neuroscience, which has been steeped in a dimensional perspective since its conception, is ideally suited to leading this transition.

Yet, developing mechanistic accounts of personality pathology requires more than simply mapping dimensional traits to their neurobiological correlates. Truly explanatory accounts can only be advanced by theory-driven investigations that articulate and test hypotheses about individual differences across multiple levels of analysis. The implications of this perspective are wide-reaching for personality neuroscientists. It will require researchers to hone their knowledge of personality theory, integrate findings from other disciplines (especially cognitive neuroscience and decision science) to develop detailed hypotheses, and implement rigorous and innovative experimental paradigms within sufficiently powered samples. These are difficult challenges, but ones that must be grappled with if the goal is to explain the causal basis of personality pathology. We hope that the recommendations in this commentary will help personality neuroscientists to be well-positioned to meet these challenges and usher in a new era of explanatory personality pathology research.