Trad

Refocusing Neuroscience: Moving Away from Mental Categories and Toward Complex Behaviors Luiz Pessoa1 , Loreta Medina2 , and Ester Desfilis2 1 University of Maryland, College Park, Maryland, USA 2 University of Lleida, Lleida Institute for Biomedical Research Dr. Pifarré Foundation (IRBLleida), Lleida, Spain pessoa@umd.edu; loreta.medina@udl.cat; ester.desfilis@udl.cat August 30, 2021 (original version: May 22, 2021) Mental terms—such as perception, cognition, action, emotion, as well as attention, memory, decision making—are epistemically sterile. We support our thesis based on extensive comparative neuroanatomy knowledge of the organization of the vertebrate brain. Evolutionary pressures have molded the central nervous system to promote survival. Careful characterization of the vertebrate brain shows that its architecture supports an enormous amount of communication and integration of signals, especially in birds and mammals. The general architecture supports a degree of “computational flexibility” that enables animals to cope successfully with complex and ever-changing environments. Here, we suggest that the vertebrate neuroarchitecture does not respect the boundaries of standard mental terms, and propose that neuroscience should aim to unravel the dynamic coupling between large-scale brain circuits and complex, naturalistic behaviors. Open a textbook on the mind and brain, say Cognitive Neuroscience: The Biology of the Mind by Gazzaniga, Ivry, and Mangun (2013, fourth edition). Skimming through it, we see chapters on perception, attention, memory, learning and development, language, motor control, executive functions (“higher” cognitive functions), and consciousness. Aside from consciousness, which comes at the end of the book as a challenging subject, the other topics sound well defined and even intuitive. A central goal of neuroscience is then to uncover how these 2 mental functions are instantiated in the brain (Figure 1A) But where do the above chapter themes come from? Take attention for example (Anderson, 2011). 1 In the West, ideas about attention 1 Quotes and references from this paragraph are from Anderson (2011). have been discussed since the Greeks. For example, Aristotle questions whether “… it is possible or not that one should be able to perceive two objects simultaneously in the same individual time?” This problem is echoed in modern research when Huang and Pashler (2007) state that “[t]his question about [the] possibility of simultaneous selection of two feature values is very fundamental.” Perhaps closer to intuitive notions of attention, consider what is now called the “cocktail party effect,” as formulated a few centuries ago by Stewart (1792/1866): “When two persons are speaking to us at once, we can attend to either of them at pleasure… This power, however, of the mind to attend to either speaker at pleasure, supposes that it is, at one and the same time, conscious of the sensations which both produce.” Figure 1: Neuroscience and mental terms. (A) One of the central problem in neuroscience is to discover mappings between brain and mental functions. (B) Characterizing the coupling between distributed, and dynamic neural circuits, and complex, dynamic behaviors. Photograph reproduced for academic purposes with kind permission from Judy Lehmberg. The terms above—perception, attention, etc.—have long histories, and make for excellent chapter headings in a textbook. But do they provide reasonable conceptual anchors for neuroscience (see also Buszáki, 2020)? Consider perception, which involves processing elements of the environment “through physical sensation”. 2 What is 2 According to Merriam-Webster, perception is the awareness of the elements of environment through physical sensation; or, physical sensation interpreted in the light of experience. the relationship between perception and other mental processes? The classical sequential scheme has long been supplanted by interactive schemes with feedback and other interactions (Figure 1B). Thus, perception is not separate from cognition, and not even action if one adopts “active” frameworks (Aloimonos et al., 1988; Findaly et al., 2003; Merriam and Colby, 2005). More broadly, perception is used to denote a large number of processes only loosely related. For the experimental psychologist or neuroscientist, the concept is just too vague, and lacks enough coherence to provide much conceptual 3 utility. Perhaps perception is so basic that the criticism does not apply to other mental terms commonly used in neuroscience (say, “attention”). Instead, we argue that the problems with what we will call standard mental terms are quite general. Mental categories are routinely used in a dual fashion to denote both the problem—the phenomenon one aims to explain—and the solution—the mechanism proposed to provide the explanation (Hommel et al., 2019). 3 More broadly, mental terms are used in 3 Our treatment follows closely the one by Hommel et al. (2019) that focused on attention. a circular fashion. For example, “emotional processing” is defined in terms of systems that are purported to be part of the emotional brain, such as the hypothalamus or the amygdala; conversely, a structure that plays an important role in “fear” is considered part of the emotional brain. Although the language used in some instances is not necessarily flawed, such linguistic habits are potent enough to lead investigators astray, and limited knowledge is actually gained by sticking to the traditional terminology. More forcefully, we propose that mental terms are epistemically sterile. We support our thesis based on extensive comparative neuroanatomy knowledge of the organization of the vertebrate brain. Evolutionary pressures have molded the central nervous system to promote survival. Careful characterization of the vertebrate brain shows that its architecture supports an enormous amount of communication and integration of signals. The general neuroarchitecture supports a degree of “computational flexibility” that enables animals to cope successfully with complex and ever-changing environments. Here, we suggest that the vertebrate neuroarchitecture does not respect the boundaries of mental terms (see also Cisek, 2019). Thus, we propose that neuroscience should seek to unravel the coupling between large-scale circuits spanning the neuroaxis and complex, naturalistic behaviors (Figure 1B). 4 Large-scale circuits discussed below 4 The photo was taken in Etosha National Park in northern Namibia. The back story of the photo is that the two lions were a mated pair (lions mate on and off for about three days). During that time the female might hunt but the male is usually just following her around “like a puppy” (in the words of the photographer). One wonders if the zebra in the foreground, in assessing potential threat, took into account in any way the status of the lions! include those involving the basal ganglia, amygdala, and superior colliculus/optic tectum systems, which illustrate how the vertebrate neuroarchitecture contains a series of spiraling pathways that communicate and integrate signals across different spatial extents (see also Pessoa, 2017; Pessoa, Medina, Desfilis, 2019). Our critique of a form of “unrestrained mentalism” common in neuroscience does not entail a return to behaviorism. However, we do believe that the careful characterization of behavior is fundamental in neuroscience, and deserves considerable more attention (e.g., Krakauer et al., 2017), in particular ethological approaches focusing on natural behaviors and comparative approaches that consider a range of species. Before proceeding, we stress that the our discussion is selective given space restrictions. In particular, we were unable to review multiple lines of research at the interface between psychology 4 and neuroscience that have emphasized “natural behaviors”; for example, the work by Panksepp and colleagues (see Pankseep, 1998), including his work on play behaviors; as well as the work by Berridge and colleagues on how the liking/wanting distinction informs the understanding of behaviors such as feeding, mating, parental care (Berridge et al., 2009). Emotion and Cognition To illustrate how the semantic separation of mental terms has helped shape the understanding of their neural basis, let us discuss some of the origins of how emotion and cognition are viewed as segregated in the brain. In the conclusion of The Descent of Man, Darwin wrote in 1871 that “the indelible stamp of his lowly origin” could still be discerned in the human mind, with the implied consequence that it was necessary to suppress the “beast within”. This notion was hardly original, of course, and in the West can be traced back to at least ancient Greece. At Darwin’s time, with emotion being considered primitive and reason the more advanced faculty, “true intelligence” was viewed as residing in cortical areas, most notably in the frontal lobe, while emotion was viewed as residing in the basement, the lowly brainstem. The decades following the publication of Darwin’s Origin of Species (in 1859) were a time of much theorizing not only in biology but in the social sciences, too. Herbert Spencer and others applied key concepts of biological evolutionary theory to social issues, including culture and ethics. Hierarchy was at the core of this way of thinking. For the survival of evolved societies, it was necessary to legitimize a hierarchical governing structure, as well as a sense of self-control at the level of the individual (Parvizi, 2009). These ideas, in turn, had a deep impact on neurology. Hughlings Jackson, to this day the most influential English neurologist, embraced a hierarchical view of brain organization rooted in a logic of evolution as a process of the gradual accrual of more complex structures atop more primitive ones. Thus, “higher” centers in the cortex exert control on “lower” centers underneath, and any release from this control could make even the most civilized human act more like their primitive ancestors. This stratified scheme was also enshrined in Sigmund Freud’s framework of the id (the lower level) and the super-ego (the higher level). Against this backdrop, it is not surprising that brain scientists would search for the neural basis of emotion in subcortical territories, while viewing “rational thinking” as the province of the cerebral cortex, especially the frontal lobe. 5 In 1896 the German anatomist Ludwig Edinger (1900) published The Anatomy of the Central Nervous System of Man and other Vertebrates. The book, which established Edinger’s reputation as the founder of comparative neuroanatomy, described the evolution of the forebrain as a sequence of additions, each of which establishing new brain parts that introduced new functions. Edinger viewed the forebrain as containing an “old encephalon” found in all vertebrates. On top of the old encephalon, there was the “new encephalon,” a sector only more prominent in mammals. In one of the most memorable passages of his treatise, Edinger illustrates his concept by asking the reader to imagine precisely inserting a reptilian brain into that of a marsupial (a “simple” mammal). When he superimposed them, the difference between the two was his “new encephalon”. He then ventured that, in the brain of the cat, the old encephalon “persists unchanged underneath the very important” new encephalon. Put differently, the part that was ancestrally present is left unaltered. Based on his coarse analysis of morphological features, his suggestion was reasonable. 5 5 Echoes of Edinger’s ideas are of course evident in the work of Paul MacLean, who proposed the idea of the “triune brain.” But to a substantial degree, his ideas were very much in line with the notion of brain evolution as progress toward the human brain, as in the Aristotelian notion of the scala naturae (Hodos and Campbell, 1969). Given the comprehensive scope of Endinger’s analysis across vertebrates, his views had a lasting impact and shaped the course of research well into the 1960s. More than a century later, knowledge about the brains of vertebrates has expanded enormously. Yet, old thinking dies hard. Antiquated views of brain evolution continue to influence neuroscience, even if implicitly. As an example, consider that most frameworks of brain organization are heavily centered on the cortex. These descriptions view “newer” cortex as controlling subcortical regions, which are assumed to be (relatively) unchanged throughout evolution. Modern research on brain anatomy from a comparative and functional architecture viewpoint indicates, in contrast, that brain evolution is better understood in terms of (1) modification in neuronal populations within the brain’s fundamental units (building blocks) and (2) the reorganization of large-scale connectional systems in which they are engaged, as described below (for a more detailed treatment, see Pessoa et al., 2019). Yet, textbooks often present “cognitive” and “emotional” systems as if they were separate entities (this is especially the case in clinically-oriented materials). In particular, textbooks still discuss the “limbic brain”, a concept that has no stable meaning, and essentially is used as an amorphous synonym for a putative “emotional brain”. 6 What is more, the purported “emo6 Some time ago, one of the authors (L.P.) was invited to write a chapter on the “emotional brain” for a textbook on neuroscience with a clinical angle. For reference, L.P. inspected the chapter of a prior edition published in 2008. The chapter, entitled “The Limbic System”, based its content on the so-called circuit of Papez, a largely speculative proposal from 1937, which somewhat unbelievably is still influential despite being completely outdated. tion/cognition” separation continues to generate ideas of wide public 6 appeal, such as the notion of “System I/System II” popularized by Kahneman (2011). Basic principles of the vertebrate forebrain The brain of all vertebrates is organized according to a common “building plan”, also called Bauplan or morphoplan (Striedter, 2005; Nieuwenhuys and Puelles, 2016), which shares the same basic subdivisions (reviewed in Nieuwenhuys et al., 1998). To unravel the common architecture, it is critical to identify and map the same— technically, homologous—brain regions in different vertebrates. Homology refers to relationships between traits that are shared as a result of common ancestry (Weiss, 1994). For example, the human arm and the bird wing are considered homologous as upper limbs because they arose from a corresponding character in the tetrapod common ancestor through descent with modification. Note, however, that they differ greatly in terms of both detailed structure (distal parts related to adaptive features) and function. Evolutionary divergence can make it difficult to identify homologies. In such cases, the study of embryonic development helps in detecting homologies, as embryos of different species are more similar to one another than adults. This approach has been successfully applied to the comparative study of the organization of the vertebrate nervous system. During development, a series of segments, called “neuromeres”, can be identified that form essential building blocks of the central nervous system (Nieuwenhuys and Puelles, 2016). Sets of neuromeres form key morphological entities that can be identified during development, and give rise to three major territories in adults: forebrain (prosencephalon), midbrain (mesencephalon), and hindbrain (rhombencephalon). In the forebrain, the telencephalon can be subdivided into two major divisions: the pallium and the subpallium, which we will discuss next. (If the reader is unfamiliar with these terms, as first approximation they can mnemonically link “pallium” (from the Latin word for woolen cloak or mantle) with cortex and “subpallium” with the basal ganglia, among other areas in the ventral telencephalon, although this is not strictly correct; see bellow). Pallium In mammals, the pallium includes the cortex in addition to noncortical structures, such as the claustrum and the basolateral complex of the amygdala—we call the latter the pallial amygdala. At present, there are competing views about the overall plan of the vertebrate forebrain (reviewed by Puelles et al., 2013; Striedter, 2005). 7 Controversy is particularly acute regarding the telencephalon, the largest subdivision of the forebrain in amniotes (reptiles, birds, and mammals). This part of the brain shows a high degree of divergence, although it shares the same basic divisions in different groups. One view of the vertebrate morphoplan is the developmental genoarchitecture hypothesis, which is based on shared expression patterns of highly conserved regulatory genes observed at early embryonic stages (Puelles and Medina, 2002; Puelles and Ferran, 2012). When comparative genoarchitecture data are integrated with key morphological landmarks, the embryonic pallium of vertebrates can be subdivided into four (Puelles et al., 2017) or six (Desfilis et al., 2018) compartments that are comparable across species. To unify the proposals of four versus six pallial divisions, we refer to them as medial, dorsal, lateral (more precisely, dorsolateral/lateral), and ventral (more precisely, ventral/ventrocaudal) pallial divisions. 7 7 Recently, new proposals of a ringlike organization of the pallium have emerged (for example, Puelles et al., 2019; García-Cabezas et al., 2019). Although interesting, the proposals are largely cortico- and mammaliancentered, and less suited to explain the pallium organization in non-mammals. As such, ring-like models cannot yet be used to extract general organizing principles that apply to vertebrates more generally. Accordingly, in our discussion we embrace a four/six pallial sector scheme. According to the developmental genoarchitecture proposal, the medial pallium gives rise to the hippocampal formation (Nieuwenhuys et al., 1998; Wullimann and Mueller, 2004; Medina et al., 2017). In mammals the dorsal pallium gives rise to the isocortex (with six layers), the dorsolateral/lateral pallium produces the claustro-insular region, the orbitofrontal cortex, and the perirhinal/lateral entorhinal cortex, while the ventral pallium gives rise to the olfactory cortex and part of the amygdala, the so-called pallial amygdala (Desfilis et al., 2018; Medina et al., 2017; Moreno and González, 2006; Puelles et al., 2017). Based on data on the development of the amygdala (Garcia-Calero et al., 2020), we can also consider that the ventral pallium includes a rostral sector that produces the piriform cortex plus endopirifom nuclei, and a distinct caudal sector that produces the pallial amygdala. Basal ganglia loops Across vertebrates, the subpallium is relatively conserved and contains the striatum, pallidum (not to be confused with “pallium”), parts of the amygdala (subpallial amygdala), and the bed nucleus of the stria terminalis (BST), among others (Gonzalez et al., 2014; Medina et al., 2017; Moreno et al., 2009). Here, we discuss a central element of the neuroarchitecture of tetrapod vertebrates (amphibians, reptiles, birds, and mammals), which involves cortical-subcortical (more generally, pallial-subpallial) forebrain circuits. Classically linked to movement control and disorders, the basal ganglia are now known to be involved in multiple functions, and viewed as essential for sophisticated forms of behavioral control, including learning and regulation of stimulus-driven behaviors, as 8 well as action selection supporting goal-directed behaviors (Yin and Knowlton, 2006; DeLong and Wichmann, 2009; Nelson and Kreitzer, 2014). Work focusing on mammals in the 1970-80s uncovered the cortical-subcortical connectional architecture of the basal ganglia (Alexander et al., 1986; Alheid and Heimer, 1988). Nearly the entire cortical sheet projects to the striatum, and whereas striatal territories receiving cortical input do not directly reciprocate their connections, pathways return to cortex via different parts of the thalamus after an additional step in the pallidum. Together, these studies have led to the important concept of cortico-basal ganglia-thalamo-cortical systems, or basal ganglia loops in short (Alexander et al., 1986; Haber, 2003). An important feature of mammalian basal ganglia loops is that they involve both dorsal (caudate-putamen) and ventral (nucleus accumbens) striatal components. Some cortical areas project to the dorsal striatum (for example, motor and somatosensory areas), while others project to the ventral striatum (in primates, for example, orbitofrontal, prefrontal, and anterior cingulate cortices). A common view is that basal ganglia loops are anatomically and functionally segregated forming multiple parallel circuits, as emphasized originally by Alexander et al. (1986). However, several research groups have described ample anatomical substrates for interactions between circuits, such that multiple opportunities for crossover between streams exist (Joel and Wiener, 1996; Haber, 2010; Hintiryan et al. 2016; Groenewegen et al., 2017; Aoki et al. 2019). We will develop the theme of intercommunication between basal ganglia loops considerably below, as it plays an important role in the intermixing and integration of brain signals that we suggest blur potential mental categories that can be instantiated by the vertebrate neuroarchitecture. Amygdala A central point of the present paper is that standard mental categories are ill-suited to investigating the brain basis of behavior. Here, we describe how the amygdala participates in cortical-subcortical (technically, pallial-subpallial) loops that interlink a very wide spectrum of signals in a way that breaks down the barriers between purported mental domains. In broad terms, the amygdala of mammals consists of (1) a basolateral complex (including lateral, basal, and accessory basal nuclei, and some cortical areas), and (2) an extended amygdala (including the central nucleus, the medial nucleus, and the complex of the bed nucleus of the stria terminalis (Alheid and Heimer, 1988; Amaral et al., 1992). In mammals, we refer to the amygdala as a subcortical 9 structure (with the exception of a few cortical parts; for instance, involved in olfactory processing). Technically, based on the embryonic development of the telencephalon, the basolateral amygdala is mostly of pallial origin and the extended amygdala is mostly of subpallial origin (Martínez-García et al., 2007; Medina et al., 2017). In other words, as the brain develops, the embryonic division that originates the cortex also generates the basolateral amygdala, and the embryonic division that originates the basal ganglia also produces the extended amygdala. 8 8 The embryonic origin and the regulatory genes expressed in each division during development explain the large population of glutamatergic neurons and the typical excitatory projections of the pallial part of the amygdala, as well as the considerable quantity of GABAergic neurons and the inhibitory projections that characterize the subpallial amygdala (reviewed by Medina et al., 2017). More recently, it was shown that part of the medial extended amygdala, previously thought to be subpallial, originates in a new division of the telencephalon, interposed between the subpallium and the hypothalamus, explaining the presence of abundant glutamatergic projection neurons in this region (Morales et al., 2021). The pallial amygdala system First, let us consider the anatomical pathways of the pallial amygdala with other parts of the pallium (McDonald, 1988). It has connections to frontal, parietal, cingulate, prefrontal, insular (both granular and agranular), temporal, olfactory, and hippocampal cortices. Swanson and Petrovich (1998) suggested to naming this sector as the frontotemporal amygdala due to its extensive interconnectivity with the cortex. Although this designation emphasizes the pathways with these parts of the cortex, it does not convey the fact that the pallial amygdala has extensive connectivity with all four pallial sectors discussed previously. Some examples in mammals (based on primates and rodents) are summarized next (Freese and Amaral, 2009). (1) Dorsal pallium: Projections from lateral prefrontal cortical areas (Brodmann areas 8, 45, 46, parts of 9 and 12), with heavier connections originating from more caudal regions; projections from the amygdala to those areas are fairly light. (2) Lateral pallium: Major connections with the orbitofrontal cortex (especially the caudal aspect). (3) Medial pallium: Extensive connections with the hippocampal complex (fields CA3, CA2, CA1, and dentate gyrus; enthorhinal cortex, and subiculum). Projections to the hippocampus are substantially stronger than input from the hippocampus. (4) Ventral pallium: basolateral amygdala nuclei are richly interconnected (e.g., pathways between the lateral and the basal nuclei), as well as interconnected with the piriform cortex. Given the extensive connectivity of the pallial amygdala with all sectors of the pallium, we propose that this area is not devoted to a single function but participates in a broad array of functions across the spectrum of traditional mental domains—emotion, cognition, action. Far from a “danger detector” or a “fear center”, the pallial amygdala is a hub that participates across multiple cerebral networks supporting diverse functions. How is the pallial amygdala organized in other vertebrates? Given the emphasis in the literature on the survival-related functions of the 10 amygdala (here both the basolateral and the extended amygdala), one would expect that the amygdala would be highly conserved and the task of identifying it across vertebrates would be relatively straightforward. Rather, this is far from being the case, and particularly problematic for the pallial amygdala. Despite the challenges, pallial amygdala-like regions have been identified across vertebrates (for discussion of some of the disputes in the literature, see Medina et al. (2017) and Pessoa et al. (2019)). In birds, part of the proposed avian pallial amygdala-like region, the caudal nidopallium, is richly interconnected with all four sectors of the pallium. The reciprocal connectivity with other pallial areas is so extensive that the caudolateral nidopallium is considered functionally analogous (that is, functionally similar but not homologous) to the prefrontal cortex of mammals (Güntürkün and Bugnyar, 2016). Instead, we suggest that the caudolateral nidopallium in birds is functionally similar to the pallial amygdala of mammals, which also shares a similar set of connections with different pallial sectors (Pessoa et al., 2019). In other words, the extensive connectivity of the avian caudal nidopallium that has led some investigators to propose that it is functionally analogous to the mammalian frontal cortex is consistent with the extensive connectivity of the mammalian basolateral amygdala. The pallial amygdala is one of the most associative and integrative brain regions, especially in non-mammals. The connectional systems of the pallial amygdala-like area of birds and reptiles spans multiple levels of the neuroaxis allowing it to be involved in multifaceted signaling (related to external and internal realms). It exhibits exuberant connections with other pallial areas (as in mammals), and additional circuits via the basal ganglia, hypothalamus, subpallial extended amygdala, and thalamus offer the potential for further signal communication, especially in birds where thalamic projections reach a broad spectrum of pallial areas (see next section for elaboration of this point). Thus, the pallial amygdala in all amniotes, notably in mammals, birds, and reptiles, is in a pivotal position for integrating multiple signals and participating in multiple functions that support effective behaviors in complex and dynamic ecological niches. A pallial amygdala-like area has been identified in the ventral pallium of amphibians and teleost fishes (Medina et al., 2017; Moreno and Gonzalez, 2006, and Porter and Mueller, 2020), where several aspects of the connectivity are reminiscent of the amniote organization, including connectivity with other pallial regions (Moreno and González, 2006). For example, in frogs, the pallial amygdala (called lateral amygdala) is reciprocally connected with several pallial areas including the rostral pallium, the lateral pallium, and the olfactory 11 bulb (Roth et al., 2007), and projects to the central amygdala and the BST, and then to the hypothalamus (Moreno and González, 2006; Moreno et al., 2012). The telencephalic system involving the subpallial amygdala Let us turn now to the subpallial amygdala, which in mammals includes the central amygdala. The central amygdala interfaces with the hypothalamus and brainstem in a manner that makes it an important component of endocrine and autonomic reactions to motivationally significant information (Phelps and LeDoux, 2005). The well-established outflow role of the central amygdala is frequently the target of experimental investigation (Figure 2A). Here, we discuss a key component of the connectivity of the subpallial amygdala that places it within the context of cortical-subcortical circuits. In this view, the central amygdala plays a role comparable to that of the striatum in basal ganglia loops (Alheid and Heimer, 1988; Medina et al., 2011; Waclaw et al., 2010; Pessoa et al., 2019). The importance of this perspective is that it extends the functional role of the central amygdala beyond autonomic and endocrine processes, bringing it to bear upon a broad array of processes that are not confined to those traditionally described as “emotion” or “affective” processing. Figure 2: Amygdala circuits. (A) Outflow view of amygdala pathways, where information flows from the basolateral amygdala to the central amygdala, and then to regions important for autonomic and endocrine processing. (B) Basal-ganglia-type circuits involving the extended amygdala (central amygdala [CE] and bed nucleus of the stria terminalus [BST]). A substantial input to the central amygdala originates from the basolateral amygdala, but all pallial sectors are involved to some extent. The arrow in cyan represents the “outflow” arrow of part (A). The loop through thalamus involves the paraventricular nucleus (PVT). Researchers have noted that the central amygdala contains striatum-like GABAergic projection neurons, as well as other properties of striatal neurons (e.g., McDonald, 1982; Alheid and Heimer, 1988; Swanson and Petrovich, 1998). In addition, they share their origin in the striatal embryonic division (Medina et al., 2011, 2017). Accordingly, it has been proposed that the central amygdala should 12 be conceptualized as part of the striatum. As described next, the connectional logic of the central amygdala parallels that of basal ganglia loops in important ways (Figure 2B). The central amygdala receives inputs from the four major pallial compartments. In particular, inputs from the pallial amygdala (part of the ventral pallium) are extensive. The pallial amygdala interfaces with the extended amygdala much like the isocortex (that is, six-layered cortex) interfaces with standard basal ganglia loops (functionally, this also matches the integrative properties of the pallial amygdala which receives massive inputs from across the cortex). The central amygdala projects to the BST (mostly its lateral part), which can be considered a pallidum-like region in terms of its molecular profile and embryonic origin (Bupesh et al., 2011; reviewed by Medina et al., 2017; Nóbrega-Pereira et al., 2010). The BST subsequently projects to the thalamus, which in turn projects to several pallial/cortical targets. The pathways from the BST to the thalamus target the paraventricular thalamic nucleus (PVT) and other midline nuclei (Dong et al., 2001; Dong and Swanson, 2006). 9 In all, the over- 9 The BST also projects to the hypothalamus, thalamus, and brainstem (Dong et al., 2001). all arrangement establishes a pathway through the central extended amygdala and back to the pallium (Kiourac, 2015). Although views of the organization of classical basal ganglia loops are evolving and suggest a more open loop arrangement as opposed to strictly parallel streams (Haber, 2010; Hintiryan et al. 2016; Aoki et al. 2019; Joel and Wiener, 1996; Groenewegen et al., 2016; see also next section), those through the extended amygdala clearly should be conceptualized as fairly open loops. In particular, the inter-pallial connectivity of the basolateral amygdala demonstrates the extensive influence of the extended amygdala loop on pallial function. 10 10 This is even more evident when considering the medial extended amygdala, a nuclear complex originating from multiple embryonic neurons, some glutamatergic and others GABAergic (Medina et al., 2017; Morales et al., 2021). The medial amygdala projects to the hypothalamus directly and by way of the medial BST. It also projects to the thalamus (PVT), to the ventral tegmental area (which has ample modulatory influence on the basal ganglia, amygdala, and prefrontal cortex, among other areas), and to the periaqueductal gray (Canteras et al, 1995). Overall, the medial amygdala is part of social processing networks (Medina et al., 2019) involving the interactions between multiple functional systems. Linking cortical-subcortical connectional systems The organization of cortical-subcortical loops via the striatum constitutes a major large-scale organizational principle of the brain. Whereas basal ganglia loops involving different parts of the cortex were considered originally fairly segregated, recent evidence indicates a considerable amount of cross-talk between them. Understanding their intercommunication is important because it provides potential avenues to investigate interactions between multiple classes of signals—”cognitive”, “affective”, “motor”. An important mode of information exchange between loops occurs via hub regions, namely, regions with fairly extensive connectivity. One such region is the PVT (Figure 3A) (for a review of its connectivity, see Kirouc, 2015). In the extended amygdala system, the 13 pathways from the BST to the thalamus target the PVT. The PVT also projects to the nucleus accumbens, effectively interlinking the extended amygdala and the ventral striatal connectional systems (Figure 3A). The projections of the PVT have a remarkable property. Whereas the majority of neurons in the PVT project to the accumbens, most of them give off collaterals that innervate multiple subcortical targets, including the BST and central amygdala (Moga et al., 1995; Dong et al., 2017; Unzai et al., 2017). In other words, the same PVT neuron impacts responses across multiple target structures. Figure 3: Interlinking of basal gangliatype loops. (A) The paraventricular nucleus of the thalamus (PVT) functions as a hub region given its extensive interconnectivity. (B) The extended amygdala basal ganglia loop engages the paraventricular nucleus of the thalamus (PVT). This region has a considerable projection to the ventral striatum, therefore linking the extended amygdala and ventral striatum basal ganglia loops. The PVT is also richly interconnected with multiple brain regions (see text). (C) Schematic representation of interlinked basal ganglia-type loops emphasizing that their integrative properties complement their organization in terms of separate loops. Abbreviations: BLA, basolateral amygdala; BST: bed nucleus of the stria terminalis; CE, central amygdala; PFC, prefrontal cortex; PVT, paraventricular nucleus of the thalamus; TH, thalamus; VTA, ventral tegmental area. The functional relevance of this organization can be appreciated further by considering additional pathways. The PVT is reciprocally connected with pallial areas, such as the insular cortex, the prefrontal cortex (including orbitofrontal cortex), the hippocampal formation, and the basolateral amygdala. All of these pallial sectors are themselves reciprocally interconnected, and project to the central extended amygdala and nucleus accumbens. In addition, the PVT receives substantial inputs from the hypothalamus and the brainstem. Together, the PVT is a key node for the interchange of affective and reward-relevant information and for modulating behavior in a context-dependent manner. Indeed, recent research is uncovering its critical contributions during both appetitive and aversive processes (Beas et al., 2018; Do-Monte et al., 2015; Penzo et al., 2015; Zhu et al., 2018; Zhou and Zhu, 2019). Figure 3 illustrates another important channel for intercommunication between cortical-subcortical loops involving the ventral tegmental area (VTA). The BST projects to the VTA, a midbrain dopaminergic center that projects to both the ventral striatum (nucleus accumbens) and the extended amygdala (including the central amygdala). The VTA also plays an important role in influencing the pallium, in particular the basolateral amygdala and the prefrontal cortex. Taken together, the VTA occupies a pivotal position for interlinking cortical-subcortical loop-like systems. 14 Additional opportunities for cross-talk exist. Multiple pathways link pallial areas from different compartments, combining signals across basal ganglia loops at the level of the pallium. For example, in mammals, the pallial amygdala is reciprocally connected with the isocortex, the lateral entorhinal cortex, and the hippocampal cortex (Ghashghaei and Barbas, 2002; Pikkarainen and Pitkänen, 2001; Pikkarainen et al., 1999). The interconnectivity at the level of the pallium is not only a property of the mammalian brain, but is present in birds and reptiles, too. For example, as discussed previously, the caudal nidopallium in birds and the equivalent area in reptiles (posterior dorsal ventricular ridge) is an area of the ventral pallium that is linked to all other pallial sectors. In particular, in birds, this area is reciprocally connected with the (1) dorsolateral pallium, which includes an entorhinal-like area as well as an orbitofrontal-like area (Desfilis et al., 2018; Medina et al., 2019; Medina et al., 2017); (2) the Wulst, a dorsal pallial region homologous to the isocortex in mammals; and (3) the hippocampal formation in the medial pallium. In addition, as in mammals, this area in the ventral pallium of birds and reptiles projects to several areas of the subpallium of the forebrain, including the nucleus accumbens in the ventral striatum, the dorsal striatum, and the subpallial amygdala (Medina et al., 2019; Medina et al., 2017) (Figure 2B). Together, the pathways discussed interlink dorsal and ventral basal ganglia loops, which are typically considered to be largely parallel/independent, in ways that are not usually considered (see also Averbeck et al., 2014; Haber, 2003). Importantly, because this organization is found not only in birds and mammals, but also in reptiles, this feature was likely present in the most recent amniote ancestor. The ample crosstalk between loops also suggests a major role of both the basal ganglia (both dorsal and ventral components) in processing non-motor signals, namely, those more closely aligned with contextual, aversive, and appetitive signals. In the end, whereas it is important to understand some of the dominant roles of specific cortical-subcortical basal ganglia loops, it is equally important to understand the many ways in which they are coupled (Figure 2C). Mental Categories and the Vertebrate Neuroarchitecture Let us return to the question of mental categories studied in neuroscience. Are standard terms like “attention”, “memory”, and “decision making” useful for studying and describing the relationship between brain and behavior? More directlty, what should the neuroscientist care about? We argue that a comparative understanding of the 15 general vertebrate neuroarchitecture strongly constrains the classes of mental processes in vertebrates. In particular, the functions supported by the neuroarchitecture do not align themselves well with the standard decomposition. In other words, in part, our argument is that the standard decomposition would require an organization that is relatively modular. We argue, instead, that fundamental principles of the neuroarchitecture indicate that it is not. In particular, the neuroarchitecture is not “additive”—in the sense that new components are added atop an ancestral organization—, as proposed by Edinger. As an example of “non additive” changes, consider the organization of the basal ganglia. Whereas important components of its architecture are conserved across vertebrates, substantial differences are observed, too. In amphibians and reptiles, prominent pathways link the basal ganglia with the optic tectum (Marin et al., 1998; Medina and Reiner, 1995; Reiner et al., 1998), while a less extensive system interlinks the basal ganglia and the pallium. 11 In reptiles and amphibians, basal ganglia loops involving 11 In birds, prominent pathways link the basal ganglia with the optic tectum, too, but extensive connections between the basal ganglia and the pallium are present as well. the thalamus course through the ventral striatum, but an important addition is observed in birds and mammals, both of which include additional loops via the dorsal striatum (caudate-putamen). The considerable development and elaboration of connectional systems involving the cortex/pallium and basal ganglia can be viewed as reflecting, in part, the expansion of the thalamus and pallium in birds and mammals (Medina, 2009; Medina and Abellán, 2009). Taken together, the novel features of the bird and mammalian brain are not only related to the expansion and the increase in complexity of certain territories, but to the reorganization of existing circuits (e.g., a shift away from circuits involving the optic tectum), as new ones emerge (e.g., dorsal basal ganglia pathways through the thalamus). 12 12 Basal ganglia pathways to the optic tectum continue to be important in birds, but new ones through the thalamus were added. We conclude that the architecture of the brain is radically distinct from one that would support circumscribed mental functions. Distributed brain circuits help solve challenging behavioral problems, and our claim is that interactions and integration at different levels are integral to that ability. In so doing, any purported standard mental property (such as “decision making”) is, of necessity, deeply intertwined with others (such as “affective processing”). Attention We now turn to a discussion of a specific mental function, “attention”, and briefly illustrate the difficulties of mapping standard mental terms to the brain. Several investigators have pointed out that “attention” is not a coherent concept, as it is linked to multiple processes (Hommel et al., 2019, and references therein). In fact, it is 16 not even clear if attention is “cause” or “consequence”. For example, Krauzlis et al. (2014) argue that attention arises as a byproduct of circuits centered on the basal ganglia involved in value-based decision making; in their view attention is an effect, not a cause. If one accepts the notion that attention is not a unified concept, how should it be conceptualized? As a first step, we propose to conceptualize it in terms of multiple attention-like selection mechanisms. The advantage of doing so is that selection can be applied across multiple mental domains, including those that are conventionally described as motivational, affective, cognitive, and so on. Doing so allows us to conceptualize the underlying processes as inherently cutting across domains and not, say, in terms of a “cognitive” function as typically done. As an illustration, consider affective attention. Affectively significant visual items, such as those previously paired with shock, are behaviorally prioritized and detected faster (Dukas, 2009; Pessoa, 2013; Le Pelley et al., 2016; Todd and Manaligod, 2018). Thus, they compete with other items more effectively during demanding conditions. Noting that the amygdala is involved in the processing of affectively significant information, and that pathways from the basolateral amygdala reach nearly all levels of the ventral visual system (including primary visual cortex; Amaral et al., 1992), researchers have suggested that such projections provide boosting signals to visual cortex when visual items are negatively valenced. Although this mechanism is frequently highlighted as the key one supporting the enhanced processing of emotion-laden visual items, several other mechanisms are likely involved (Pessoa, 2013). For example, interactions involving the pulvinar nucleus of the thalamus and the basolateral amygdala likely support the behavioral advantage of negatively valenced visual items (Pessoa and Adolphs, 2010). 13 13 Midline thalamic nuclei, projecting to the pallial amygdala and prefrontal cortex, are involved too (Salay et al., 2018). More generally, we propose that both the pallial and the subpallial amygdala should be considered as important structures for selective attention-like processes, too (Pessoa, 2010; Pessoa, 2013). For example, interactions between the basolateral amygdala and frontal and parietal brain regions (possibly involving indirect pathways) likely contribute to selection processes. Other circuits involve the subpallial extended amygdala. For example, projections from the central amygdala to the locus coeruleus can engage the latter area (Cedarbaum and Aghajanian, 1978), which plays important roles in attention-like selection (Ehlers and Todd, 2017). From an evolutionary perspective, it is noteworthy that multiple structures—including the optic tectum, thalamus, and striatum—are also involved in selection processes closely tied to catching prey and avoiding predators (Krauzlis et al., 2018). Clearly, attention-like selection mechanisms are not confined to 17 cortical circuits. To conclude this section, we propose that it is fruitful to conceptualize “attention” even more broadly than in terms of selection processes. Instead, it is useful to consider a broad family of cooperative-competitive mechanisms that emerged philogenetically and that support gradually more sophisticated behaviors (Cisek, 2019 makes a related point in the context of “decision making”). Such concepualization encompasses, for example, circuits involving the extended amygdala and parabrachial nucleus that are relevant for the integration of threat information and feeding behavior (Luskin et al., 2021). Cooperative-competitive mechanisms support a wide range of behaviors, typically combining diverse sources of evidence, including those related to the body and the external environment. This conceptualization helps shift the focus from “understanding attention”, say, to studying how particular brain circuits support particular types of behavior. Threat Assessment If the standard approach of relating functions and brain mechanisms is problematic, how should we proceed? We propose addressing the following question: What neural circuits/systems subserve specific classes of behavior? Animals are confronted with environmental problems that must be solved to insure reproductive success (Fanselow and Lester, 1988). The focus on behavior, especially in terms of the problems it has presumably evolved to solve, is of course the cornerstone of the behavioral ecological approach inspired by ethology (Dewsbury, 1991; Burghardt and Bowers, 2017). As pointedly summarized by Fultot and colleagues (2019): the organism is viewed as a “seeker of stimulation rather than that of a processor of it”, and an animal’s activity is constrained by the environment. Thus, to elucidate families of brain processes requires situating them in the context of complex naturalistic behavior (for a recent multi-author discussion, see Dennis et al., 2021). Consider the threat imminence framework which proposes that, from the standpoint of an animal subject to predation, natural defensive processing should be understood in terms of three key stages (Fanselow and Lester, 1988): pre-encounter, post-encounter, and circastrike. During pre-encounter, the animal’s behavior is constrained by the assessment of the probability of encountering a predator. During post-encounter, behavior generally shifts markedly; animals frequently suppress behavior, taking stock of the situation. Circa-strike behaviors may involve flight (if possible) or fight (usually as a last 18 resort). Several related frameworks have been described, including the ethological approach by the Blanchards and their colleagues (Blanchard and Blanchard, 1988; Blanchard et al., 2011), defensive approach/avoidance systems (McNaughton and Corr, 2004), and the extension of the predator imminence model to humans by Mobbs and collaborators (Mobbs et al., 2015). Ethologically inspired work has a different flavor compared to the standard neuroscience approach. Research themes include foraging, parental care, predator-prey interactions, sexual selection, and social behaviors. Whereas, these topics overlap rather little with those motivated by cognitive psychology, other lists of ecological themes are probably more familiar to a wider group of neuroscientists: signal detection; signal localization; memory acquisition; storage and recall; motivation; coordination; and top-down control (Ewing, 1981). Indeed, some investigators have proposed combining ethological approaches with traditional systems neuroscience, in particular by studying complex behaviors in more natural conditions while recording movement, performing temporally specific perturbations, and recording from large numbers of neurons during freely moving behaviors (Dennis et al., 2021). Let us consider the processes of threat assessment during predatorprey interactions. Far from stereotyped, such processes can be highly complex and flexible in naturalistic conditions (Branco et al., 2019) (Figure 4). The entire process is dynamic such that behaviors are continuously adjusted based on a large set of interacting variables involving both prey and predator while they navigate their mutual environment. Heuristically, we can refer to a level of risk that is continuously monitored and updated, and which is context dependent and based on prior experience (see Kavaliers and Choleris, 2001). For example, certain patches of a habitat may be associated with previous encounters that were more dangerous (Lima and Dill, 1990). At the broadest level, we can refer to a threat as detected or not detected. When threat is not detected, risk assessment might dictate avoiding locations of prior predator encounter, as well as adjusting vigilance levels depending on available cues. As the animal navigates locations of increased risk, they may avoid the territory altogether, but this choice could be overridden by factors such as high levels of thirst or hunger. When threat is detected, ensuing escape responses would be expected. However, animals are not simple stimulus-response devices. If risk is low, prey will continue ongoing activity, but channel it in specific directions; for example, continue grazing but in a manner that at least maintains the distance to the predator. When risk is higher, escape-related behaviors will ensue but with vigor that is 19 Figure 4: Threat assessment as a dynamic process. commensurate with the condition at hand. For example, if risk is moderate, the animal might simply increase its distance from the predator. More generally, the escape process is informed by multiple internal (e.g., hunger, fatigue, bodily health, and sexual arousal) and external (e.g., distance and predator behavior) variables. In particular, escape involves the determination of an adequate route that is tightly coupled with the selection of appropriate refuge. This selection accounts for the safety value of the shelter, the distance and position of the predator relative to the shelter, and potential competition for access (e.g., the burrow is frequently occupied by other animals). For futher discussion, see Branco et al. (2019) and Branco and Redgrave (2020). Ultimately, the term “threat assessment” can be used as a shorthand for a series of interrelated processes that determine an animal’s behavior in the context of potential and actual encounters with predators. As such, the study of the neural basis of threat assessment requires that it be studied in naturalistic settings that approximate the range of behaviors observed in nature, coupled with rich characterization of behaviors of multiple interacting actors, including predators. How is such ecological outlook related to traditional systems neuroscience approaches (Mobbs et al., 2018, 2020; Branco and Redgrave, 2000)? 20 In the preceding discussion of threat assessment, we can identify several instances in which terms like “attention”, “emotion”, or “decision making” could possibly be used descriptively. For example, as an animal navigates its environment and perceived risk increases to moderate levels, it will “pay more attention” to certain aspects of the environment. Cues associated with the presence of predators will gain increased salience. Many of these cues will be laden with affective significance from past encounters, and will engage circuits that are typically described with labels such as “emotion processing” or involving “attention-emotion” interactions. Furthermore, the presence of the predator will invoke a series “attentional” processes related to the acquisition and selection of sensory information, and will be associated with head turning or body movements. 14 14 Perception and attention, therefore, take place in close coordination with— and in the context of—action (Warren, 2021). By and large, the standard neuroscience approach attempts to compartmentalize and isolate behaviors, such as when studying eye/head or “attentional movements” to salient visual items. But, in general, behavioral “decision making” involves a complex interplay of multiple variables that collectively contribute to action choices. As an illustration, consider the following type of decision by a prey: because (a) the hunger level of the animal is high; (b) a relatively unimpeded route to safety is possible; (c) the predator is alone; (d) the predator is not currently approaching; and (e) this type of predator (cheetah, say) has only been more dangerous when attacking the animal from stealth; then the animal may continue grazing. Whereas some of the variables above are temporally stable (past history of encounter with the animal), others may fluctuate temporally (e.g., whether the predator is alone or not). Because behavioral processes build upon a very large set of dynamic variables, whereas the standard account (e.g., “attention mechanisms”, “decision making”) provides a potential heuristic description, it fails to capture the rich interdependence of the multiple mechanisms that support behaviors. In the end, the standard account provides a language that emphasizes independence and separation, where a language of interaction and integration is needed. To conclude, we propose that threat assessment should be viewed as a highly dynamic process. Whereas actions must per force occur sequentially—escape initiation 7→ escape execution 7→ escape termination (Evans et al., 2019)—we suggest that it is necessary to conceptualize threat assessment in a continuous fashion. In this manner, as some mechanisms and processes are engaged, they lead to actions that alter environmental relationships, which in turn are continuously assessed to guide further actions. Large-scale circuits and mental functions 21 What are the neural circuits involved in dynamic threat assessment? We suggest that the overall process cannot be subdivided into separate systems that are engaged during pre-encounter defense vs. post-encounter defense, for example. More generally, there is no single underlying system for threat assessment. But this does not mean that we cannot tackle the problem of the neural basis of behavior. In other words, whereas the standard mental domains do not provide an adequate framework, it is still possible to study the coupling between dynamically engaged distributed neural circuits and complex, dynamic behaviors. By “coupling”, we mean the set of regularities between brain and behavior, in particular how variability in behavior is linked to variability in neural circuits. 15 Importantly, this mapping 15 For the important notion that behavior can act as an enabling constraint on neural activity, see Raja and Anderson (2020). is not one-to-one (one behavior, one circuit), but many-to-many (one behavior can be linked to multiple circuit instantiations, and one circuit can be linked to multiple behaviors) (see Pessoa, 2014). Although relatively little is known in mammals, as a starting point we suggest that it is useful to anchor threat assessment circuits on the superior colliculus and periaqueductal gray (PAG) (Figure 5A). The superior colliculus is often emphasized as a fairly direct sensorimotor interface, but has extensive anatomical connectivity throughout the brain, including extensive visual inputs and outputs to areas regulating head orientation and gaze direction (May, 2006). Several investigators have noted its participation in defensive behaviors (for reviews, see Dean et al., 1989; Brandão et al., 1994, Schenberg et al., 2005, Branco and Redgrave, 2020; for evidence in primates, see Maior et al., 2011, DesJardins et al., 2013, and Forcelli et al., 2016), in addition to well-known involvement in target selection and related functions commonly described as “attentional” (Krauzlis et al., 2014, 2018; Basso and May, 2017). The superior colliculus works in close connection with the PAG (they are bidirectionally connected), and the deep layers of the former may form an integrated anato-functional unit with the latter (Holstege, 1991), a region heavily involved in defensive behaviors (Bandler and Shipley, 1994; see also George et al., 2019). The superior colliculus and the PAG receive inputs from the hypothalamus, too, so bodily context and other state-related signals can be taken into account. The local circuits between the superior colliculus and the PAG do not work in isolation, and have the ability to influence behaviors in several ways, for example, via dopaminergic signaling involving the ventral tegmental area and the substantia nigra (Figure 5A). Importantly, the superior colliculus has the potential to become embedded into large-scale circuits in several ways. The superior colliculus is part of loops with the subcortex via the thalamus and the basal ganglia (for review, see McHaffie et al., 2005) (Figure 5B). 22 Additionally, via the thalamus, the superior colliculus is linked to the pallium (including the pallial amygdala), which also influences the superior colliculus/PAG via the hypothalamus (Figure 5C). Figure 5: Large-scale circuits that participate in threat assessment. (A) Superior colliculus-periaqueductal gray circuit (see dashed outline). Parts in red mark some explicit bridges to/from the circuits discussed in Figures 2 and 3, including the PVT, which illustrates the important role of connector hub regions. (B) The superior colliculus is part of subcortical loops. (C) The superior colliculus is part of loops with the pallium. Abbreviations: Hypothal, hypothalamus; PAG, periaqueductal gray; PVT, paraventricular nucleus of the thalamus; SC, superior colliculus; SN, substantia nigra; Thal, thalamus; VTA, ventral tegmental area. More generally, and critically, the PAG-superior colliculus circuit readily engages with the large-scale connectional systems discussed previously (Figures 2 and 3), for example, via the thalamus, striatum, and pallium (see parts in red of Figure 5A and parts B and C). For example, both the superior colliculus and the PAG receive inputs from the pallium, the hypothalamus, and some midbrain tegmental areas (such as the VTA); these territories, in turn, receive basal ganglia and amygdala inputs, among others. The integrative potential of the PAG-superior colliculus circuit is therefore enormous and, via ascending and descending projections, the circuit is involved in a wealth of behaviors. In the context of threat assessment, the neural circuits engaged (in a species-specific fashion) combine a large number of internal (state of the animal) and external variables (e.g., is there a path to safety?), with prior learning and future-oriented scenario simulations in a situation and context-dependent fashion (e.g., how much time does the animal have?). In this manner, the scale of the circuit engaged is temporally- and condition-specific, ranging from more circumscribed interactions involving fewer brain regions and territories to largescale circuits across the neuroaxis that span a substantial amount of the brain. Thus, one cannot point to the brain and say “here’s where threat assessment happens”. Instead, it is an outcome of the distributed and potentially large-scale mechanisms that support behavior at a specific point in time, and how the brain-behavior 23 coupling evolves temporally. Returning to standard mental functions—perception, attention, cognition, etc.—, can they be used to describe subcomponents of threat assessment? Whereas it is conceivable to use them while taking into account the considerations raised throughout this piece, it would require a major shift from key ways in which they are used by neuroscientists. What is more, in typical usage they define a research agenda that, in many ways, is “reversed”. For example, a considerable amount of energy has been devoted by neuroscientists to uncover “the emotional brain”, an endeavor that we view as futile given that the vertebrate brain does not conform to the boundaries of mental domains. Back to the brain and mental functions What kind of system is the brain? The brain has evolved to provide adaptive responses (“functional responses” in the evolutionary sense) to the problems that living beings face in order to survive and reproduce. Our brief discussion of the vertebrate brain focused on principles of the organization of cortical-subcortical loop-like circuits, as well as forebrain-midbrain interactions. A key goal was to illustrate how the neuroarchitecture supports combinatorial brain connectivity—from region A to region B via multiple routes. Functionally, circuits form dynamically such that specific populations of neurons across areas coalesce into coherent functional units. The overall organization is heterarchical, namely without fixed hierarchies. 16 16 At specific times, some regions can exert a stronger influence on others in a way that could be described in terms of hierarchical control, especially when “higher” regions influence so-called “lower” regions. Unfortunately, such descriptions are almost always unhelpful and carry with them antiquated notions of brain organization. The neuroarchitecture of vertebrates involves long-range circuits that span the midbrain, thalamus, and pallium/cortex, among other regions. We suggest that the anatomy supports a high degree of behavioral flexibility, allowing animals to cope with the multifaceted interactions they engage in involving predators, prey, potential mates, and so on. In species with more malleable behaviors, behavioral success benefits from circuits that can form flexibly, as the number of conditions related to the internal and externals worlds of the animal are exceedingly high. As neural circuits support behavioral elements, we suggest that the level of behavior provides the appropriate language for considering the mapping between brain and mind. The mental domains of the neuroscience vocabulary—attention, cognitive control, etc.—, with their origins detached from the study of animal behavior, provide problematic conceptual anchors. From the present perspective, the conclusion by several authors that categories of mental terms are 24 too heterogeneous to be conceptually useful, is thus unsurprising. Furthermore, based on the present framework, there in no confusion between the phenomenon to be explained and the mechanism used to explain it; for example, in using “attention” to refer to phenomena that engage “attentional mechanisms” (Hommel et al., 2018). The framework also protects against mixing cause and effect; e.g., is attention a causal agent or a functional consequence of circuits with specific roles? (Krauzlis et al., 2014). In conclusion, we suggest that the vertebrate neuroarchitecture does not respect the boundaries of mental terms, and propose that situating research in terms of complex, naturalistic behaviors provides a more promising approach. Ultimately, unraveling the complex dynamic mapping between brain and behavior will require moving past notions of the mind that have dominated neuroscience for a century and a half. Acknowledgements. We thank Emily Dennis for feedback on an earlier version of the manuscript; Evan Thompson, Bryce Huebner, and Kevin Mitchell for discussion; Judy Lehmberg for kind permission to reproduce the photograph in Figure 1 for scientific