36 The Organ of Tomorrow
Overview: Frontal Systems and Control Beyond the Present
The hypothalamus overview began with a strong claim: the brain’s regulation of the body begins there. That was a teaching choice, explicitly qualified, intended to keep the body at the center of the story. We have since opened much of the machinery through which regulation becomes action. Sensory systems estimate the conditions in which the animal must act. Spinal, brainstem, and cortical systems make coordinated movement possible. Cerebellar circuits predict and correct. Basal-ganglia loops regulate access to those controllers, and learning changes which consequences are worth pursuing.
This unit makes a related claim about the frontal lobes.
My claim is that frontal systems extend the regulation of a body by making what is remembered, inferred, or merely possible effective in present control.
The qualification matters here too. Frontal cortex does not introduce prediction into a brain that previously waited for things to happen. The hypothalamus already anticipates demands; the cerebellum already predicts consequences; striatal activity can reflect knowledge of how one state leads to another [@daw2011modelbased]. Evolution did not save all foresight for the front of the human head. What changes is the range of conditions that can organize present behavior, and the flexibility with which the animal can maintain, compare, and revise them.
I call this overview The Organ of Tomorrow because a possible future can govern a body now. A remembered water source can organize a journey before thirst becomes urgent. A commitment can remain effective while an easier opportunity is directly in view. A person can prepare for a consequence that has never occurred, or for another person’s response to an action not yet taken. The title names this reach of control. The tissue that makes it possible belongs to a recurrent system, and much of that system is older than the frontal territories we are about to examine.
36.1 When the present does not specify the action
A hungry animal leaving shelter may find several familiar routes available. Hunger makes food consequential, but it does not specify which route to take. One path may have been blocked yesterday. Another may be safe only when a predator is absent. A third may lead past water that will matter later. The animal’s present needs, remembered experience, and expectations about the journey must influence the same unfolding action.
The difficulty becomes clearer when the scene stays the same but the appropriate response changes. A cue may require one action under a remembered rule and a different action under another. A visible object may invite grasping while a larger task requires leaving it alone. One step in a plan may need to be completed without losing the goal that made the step useful. In each case, something beyond the current sensory display must remain effective.
I will call such information a control state when it changes how the organism uses other information or recruits an action. A remembered rule, an inferred context, an expected outcome, and a pending goal can all serve this role. The term does not imply a little written instruction stored in one neuron. It refers to a state of the nervous system whose content changes what the system does next. The same red light can then mean stop, turn left, or ignore the next cue, depending on the state in which it is encountered.
The future is one important case. The missing information may also concern a past event, a hidden context, an absent object, or another agent’s knowledge. Frontal systems become especially important when these variables must be kept available, transformed, or revised while behavior continues. They participate in those operations through connections with the systems that supply sensory content, memory, bodily significance, and action.
The phrase executive function gathers many of these abilities together. It is useful in the clinic, where a patient may retain language, recognition, movement, and familiar knowledge yet fail to organize them into an effective course of action. The explanatory trouble begins when an executive function becomes an executive: a person behind the forehead who receives reports and issues commands. That merely hides the problem of deciding inside a smaller decider.
Our task is to follow the causal machinery. A rule can change which sensory population gains influence. A prediction can alter the current significance of an outcome. A maintained goal can help stabilize a course of action against distraction. Frontal activity contributes to these changes within recurrent cortical–subcortical systems. It participates in the competition rather than standing outside it.
This is the continuity with Chapter 6, Chapter 34, and Chapter 35. The body already has needs, and its action systems already predict, select, and learn. We now ask how a remembered or possible condition becomes one of the conditions under which those systems operate.
36.2 The evolution and development of frontal cortex
Every vertebrate acts through time. A fish must persist toward shelter despite distraction. A bird must retain information after the relevant object disappears. A foraging animal must leave a declining patch and investigate another. As Chapter 34 showed in its account of ancient action systems, selection, persistence, switching, and learned control long preceded the elaboration of primate prefrontal cortex.
The neural solutions differ across lineages. Mammalian prefrontal cortex (PFC) is part of the pallium, embedded in connections with thalamus, basal ganglia, hippocampal formation, amygdala, hypothalamus, and brainstem. Birds solve overlapping problems with a pallium organized differently from mammalian neocortex. In pigeons, neurons in the nidopallium caudolaterale can sustain delay activity when a stimulus is to be remembered and reduce that activity when the bird is instructed to forget it [@rosecolombo2005]. The operation is recognizable; the tissue is not a miniature human frontal lobe.
36.2.1 What granularity changes
The frontal lobe includes motor and premotor territories as well as prefrontal association cortex. Within that sheet, the organization of cortical layers varies. Granular cortex has a clearly differentiated layer IV, recognizable by its concentration of small neurons. Dysgranular cortex has a less distinct layer IV, while agranular cortex lacks a clearly differentiated one. These architectural types coexist within the primate frontal lobe: many lateral prefrontal association areas are granular, while portions of medial and orbital frontal cortex are agranular or dysgranular [@barbaspandya1989].
Primate evolution greatly elaborated prefrontal association cortex. Monkeys and apes possess an extensive set of granular prefrontal areas, including lateral territories without established direct counterparts in rodents. Rodents have substantial medial and orbital frontal systems for learning and behavioral control, but the primate arrangement includes additional, differentiated cortical territories and their connections [@preussgoldmanrakic1991granular; @balstersetal2020homologs]. This is a substantial difference in the organization of the controller, not simply a change in the size of an otherwise identical frontal lobe.
The distinction matters because cortical layers participate differently in receiving, transforming, and transmitting information. In macaque PFC, inputs from the mediodorsal thalamus (MD) terminate prominently in layers III and IV. Local connections link activity across the cortical depth, while connections within superficial layers can extend laterally across several millimeters. A differentiated layer IV therefore forms part of an organized circuit in which incoming signals interact with ongoing cortical activity [@gigueregoldmanrakic1988md; @kritzergoldmanrakic1995intrinsic].
Architecture also predicts how cortical areas communicate. Barbas and Rempel-Clower showed that differences in laminar differentiation predict the layers from which connections originate and those in which they terminate. Projections from less differentiated to more differentiated prefrontal cortex tend to originate in deep layers and terminate superficially; the reciprocal projections follow a different laminar pattern. Thus, two areas can influence one another through different routes into their local circuitry. The functional implication is that an area’s influence depends partly on where its signals arrive, as well as on what information they carry [@barbasrempelclower1997structure]. This develops the connectional account of cortical organization introduced in Chapter 5.
Granularity is consequently useful as a marker of circuit organization; it does not by itself measure the amount of white matter or the number of connections. The expanded primate prefrontal system combines differentiated input pathways with recurrent local networks and connections to other association areas. One functional example is the maintenance of information after a stimulus disappears. In macaque dorsolateral PFC, experiments implicate NMDA-receptor-dependent recurrent excitation in sustaining activity during a working-memory delay. This mechanism involves pyramidal-cell networks, particularly those associated with layer III. The relevant functional machinery spans the layers; layer IV is one component of it [@wangetal2013nmda].
36.2.2 Evolution works through development
The evo-devo perspective introduced in Chapter 3 provides a way to understand how cortical territories and their circuits change. Developmental signals help establish regional identity; progenitor proliferation and neuronal differentiation populate those regions; growing axons establish their relationships with other systems. Evolution can alter the timing, extent, and interactions of these processes.
Experiments demonstrate that the cortical map is developmentally adjustable. Changing the signal FGF8 in mouse embryos shifts cortical area boundaries. Work on retinoic acid, a signaling molecule derived from vitamin A, connects regional patterning more directly to prefrontal circuitry. Shibata and colleagues identified a prefrontal-enriched retinoic-acid gradient in developing human and macaque cortex. Genetic experiments in mice showed that this pathway contributes to frontal regional identity, connections with the mediodorsal thalamus, and dendritic spine development. These findings identify mechanisms through which changes in development can alter both cortical territory and its wiring [@fukuchishimogorigrove2001fgf8; @shibataetal2021retinoic].
Development also continues to shape the circuit after birth. In human area 9, anatomical measurements show prolonged remodeling of dendritic spines in sampled pyramidal neurons, extending into the third decade. The adult prefrontal system is therefore the product of both early construction and extended maturation [@petanjeketal2011neoteny].
36.2.3 Human expansion: a continuing debate
Human evolution produced an enormous prefrontal association system. In comparisons using cortical boundaries informed by architecture, PFC occupies a larger proportion of cortical gray matter in humans than in chimpanzees or macaques. The continuing debate concerns how much of that difference follows the scaling relationships of larger primate brains, and how much reflects a departure from them [@donahueetal2018quantitative].
Allometry describes how the size of one structure changes in relation to another. If prefrontal cortex normally expands faster than the rest of the cortex, a larger brain will contain a larger prefrontal percentage even while following the same evolutionary scaling relationship. Studies disagree about whether particular human prefrontal measures exceed that expectation. Some support continuity with primate scaling; others identify preferential expansion associated with great apes and humans [@bartonvenditti2013; @gabietal2016neurons; @smaersetal2017expansion; @bartonmontgomery2019proportional].
The substantial elaboration of primate prefrontal association cortex remains the starting point. Human cognition depends on the operation of this enlarged and differentiated system within wider cortical–subcortical networks. Language and social learning allow an instruction acquired from another person to remain effective years later. Nested plans let a person tolerate an unrewarding step because of its place within a larger course. Our question throughout this unit is how those networks allow remembered rules, inferred situations, and possible futures to govern present action. The Deeper Dive examines why measuring the evolutionary expansion of that machinery has proved contentious.
What tissue is being compared? The whole frontal lobe includes motor and premotor cortex as well as PFC. Even “PFC” can mean different territories across studies. Gabi, Herculano-Houzel, and colleagues counted neurons in cortex anterior to the genu, the anterior bend of the corpus callosum. Across their eight sampled primate species, this region contained approximately 8% of cortical neurons, without a conspicuously greater human fraction. This was a neuronal result, not simply a comparison of whole frontal-lobe volumes. The sample included one cortical hemisphere per species and no nonhuman apes; its interpretation also depends on how well the landmark captures comparable prefrontal territories [@gabietal2016neurons].
Donahue and colleagues approached the boundary problem using cortical maps informed by architecture. Their conservative delineation, incorporating granular and dysgranular PFC, occupied approximately 21% of cortical gray-matter volume in humans, 17% in chimpanzees, and 13% in macaques. They also found that the genu landmark omitted more prefrontal territory in humans. This supports taking anatomical boundaries seriously. However, the authors explicitly noted that three species were insufficient to establish a robust nonhuman allometric relationship against which to test human exceptionalism. Their subsequent reply to criticism acknowledged that humans lay on the scaling line fitted to those three species [@donahueetal2018quantitative; @donahueetal2019reply].
What would count as exceptional? A convincing test must consider the expected scaling relationship, uncertainty around it, and the shared ancestry of the species being compared. Barton and Venditti found human frontal measures consistent with primate scaling in their analyses. Smaers and colleagues, using different datasets and evolutionary models, found evidence for departures associated with great apes and humans. These are competing interpretations of comparative evidence. Merely replacing “frontal lobe” with “granular PFC” does not settle their disagreement [@bartonvenditti2013; @smaersetal2017expansion; @bartonmontgomery2019proportional].
Which aspect of the tissue has expanded? Gray-matter volume, neuron number, and white-matter volume measure different properties. Gray-matter volume includes space occupied by neuronal processes and other tissue components, while neuron counts do not describe the arrangement of those processes or their connections. A larger prefrontal volume therefore need not imply a proportionally larger share of cortical neurons. Likewise, granularity describes laminar differentiation; it is not a count of local synapses or long-distance axons [@gabietal2016neurons; @barbasrempelclower1997structure].
How specifically primate is granular PFC? The contrast between rodent frontal organization and the extensive granular lateral PFC of monkeys and apes is substantial. The exact evolutionary origin of particular granular territories is a separate question. Nie and colleagues reported a granular frontal region in Chinese tree shrews with reciprocal mediodorsal-thalamic connections. This constrains claims that such architecture is exclusive to primates, but the resemblance does not by itself establish a one-to-one homology with monkey area 46. Here, homology concerns shared ancestry; functional analogy concerns similar work. The tree-shrew finding broadens the comparative problem without erasing the rodent–primate distinction [@nieetal2024treeshrew].
What does a developmental explanation establish? Demonstrating that a pathway changes cortical boundaries or connections identifies a mechanism available to evolution. Establishing that it caused a particular evolutionary expansion requires additional comparative evidence. Similarly, prolonged spine remodeling in sampled human prefrontal neurons does not establish one age at which all PFC functions finish maturing. These are specific findings about the construction and modification of circuits [@shibataetal2021retinoic; @petanjeketal2011neoteny]. Conserved gross anatomy can also contain molecular differences: a comparison of human, chimpanzee, macaque, and marmoset dorsolateral PFC found broad conservation of cell classes alongside substantial species differences in gene expression. Evolution can modify the workings of a circuit without producing a wholly new cortical area [@maetal2022dlpfc].
36.3 How frontal cortex is organized
36.3.1 Three surfaces and the motor border
The frontal lobe includes much more than PFC. On the lateral surface, its posterior boundary is the central sulcus. Primary motor cortex occupies the precentral gyrus immediately in front of that sulcus. Premotor territories lie farther anterior. On the medial surface, supplementary motor and cingulate motor territories provide additional routes into movement. Frontal eye fields participate in the control of gaze. These action-related regions are neighbors and partners of prefrontal association cortex, not a separate apparatus waiting for its orders.
On a lateral view of a human hemisphere, the superior, middle, and inferior frontal gyri provide a first orientation. Dorsolateral PFC includes association territories across portions of the middle and superior frontal gyri; areas 9 and 46 are familiar reference points. Ventrolateral PFC includes portions of inferior frontal cortex, with areas 44, 45, and 47 often used as landmarks, although their boundaries and classifications vary. The frontal pole, associated with area 10, wraps around the anterior tip onto lateral, medial, and orbital surfaces. A Brodmann number describes an architectonic territory, not a psychological faculty.
The ventral view reveals the orbitofrontal cortex, lying above the eye sockets. The olfactory sulcus separates the gyrus rectus medially from more lateral orbital cortex. OFC contains several distinct areas rather than one reward patch. On the medial wall, the corpus callosum and cingulate sulcus orient the reader to ventromedial, anterior cingulate, and more dorsal medial frontal territories. Ventromedial PFC is a regional label with boundaries that differ across studies; it overlaps medial orbital cortex and portions of the lower medial wall. Likewise, some territory called dorsal anterior cingulate in the decision literature is described as anterior midcingulate in other anatomical schemes. The labels should help us locate an experiment, not substitute for its actual coordinates and anatomy.
36.3.2 Different tissue, different connections
PFC has no single boundary supplied by a sulcus or thalamic projection. The traditional association with the mediodorsal thalamus (MD) remains important, but prefrontal areas receive other thalamic inputs, and frontal–thalamic connections extend beyond simple reciprocal pairs [@mcfarlandhaber2002]. Architecture and connectivity together give a more useful account.
The laminar differences introduced in Section 36.2.1 are patterned across these surfaces. Less differentiated territories occur in posterior orbital and parts of medial frontal cortex, while many lateral and anterior association territories are more fully granular. Barbas and Pandya described basoventral and mediodorsal architectonic trends in rhesus PFC, with systematic intrinsic connections along them [@barbaspandya1989].
This organization helps explain why different frontal regions have access to different information. Tracing in macaques distinguishes orbital, medial, and lateral connectional networks. Orbital territories connect with particular insular, opercular, and temporal areas carrying information about sensory objects and their bodily consequences. Medial territories have stronger relationships with other medial cortical and memory-related systems. Lateral territories connect in differentiated patterns with temporal, parietal, and action-related cortex [@saleem2008complementary; @saleem2014lateral]. The divisions overlap, and the networks connect to one another. There is no anatomical wall separating bodily significance from reasoning.
The table is a working orientation. Its regional labels are broader than individual cortical areas, and its roles describe contributions made through connections.
| Territory | Where to look | Prominent connectional relationships | Contribution emphasized in this unit |
|---|---|---|---|
| Orbital | Ventral frontal surface above the orbits | Insular, opercular, and temporal association cortex; amygdala; striatum; MD; other frontal areas | Relating sensory identity and inferred situation to expected outcomes and their current significance |
| Ventromedial | Lower medial wall and adjoining medial orbital surface | Hippocampal and parahippocampal systems; amygdala; ventral striatum; hypothalamus; medial cortical networks | Relating remembered and possible situations to the concerns and bodily state of the organism |
| Dorsolateral | Middle and superior frontal association cortex | Parietal and other association cortex; premotor territories; associative striatum; MD | Maintaining and transforming task-relevant information, rules, and priorities |
| Ventrolateral | Inferior frontal association cortex | Temporal and parietal association areas; insula and operculum; other frontal and action-related systems | Selecting and retrieving relevant information; contributing to changes in task and response control |
| Dorsal medial and cingulate | Medial frontal wall, with cingulate territories around and within the cingulate sulcus | Lateral and orbital frontal cortex; striatum; thalamus; medial motor and autonomic-related systems | Relating progress, cost, and commitment to persistence, adjustment, and initiation |
| Frontopolar | Anterior tip, extending across several surfaces | Other prefrontal territories and distributed association networks | Keeping alternatives and larger goals effective while a current course is pursued |
The functional evidence gives these relationships meaning. OFC populations can distinguish the value of a particular offered good, the identity of the chosen good, and chosen value across goods; these are different response classes, not evidence that every offer already shares one identity-free code [@padoaschioppaassad2006]. Lateral prefrontal neurons can distinguish remembered locations and abstract rules [@funahashi1989mnemonic; @wallis2001rules]. Frontopolar activity can track evidence favoring an alternative course [@boorman2009alternatives]. These findings orient the unit. The later chapters will examine the tasks and what the measurements establish.
36.4 How the body and world reach frontal circuits
The hypothalamus overview organized incoming information into chemical signals, neural reports from the body, and environmental cues. We retain those distinctions here, while adding the remembered and inferred conditions through which a cue acquires meaning. The question is no longer simply how the body detects a deficit. It is how that deficit changes the processing of a particular food, route, opportunity, or anticipated consequence.
36.4.1 From regulated variables to a changing neural state
The sensors remain where we left them. In Chapter 9, leptin and other metabolic signals reported reserves and nutritional conditions; vagal and brainstem pathways reported events in the digestive tract; arcuate, hypothalamic, and connected circuits organized hunger and feeding. In Chapter 8, the subfornical organ, OVLT, and median preoptic network combined evidence about fluid balance with signals that drinking was already under way. Neither system has to wait for frontal deliberation to protect the body.
There are direct hypothalamic projections to prefrontal cortex. In rhesus monkeys, Rempel-Clower and Barbas found hypothalamic input to all the prefrontal territories they sampled. Much originated posteriorly, while orbital and medial limbic territories also received more anterior and tuberal input. The descending projections were more selective, arising chiefly from orbital and medial cortex [@rempelclowerbarbas1998]. This establishes an anatomical route. It does not establish that each ascending axon carries a labeled message such as hunger, thirst, or sexual motivation. Arousal, physiological state, and particular motivational influences require physiological identification in addition to tracing.
Much of the body’s influence also arrives through intervening structures. Visceral and somatic signals ascend through brainstem and thalamic pathways into insular and other cortical systems. Hypothalamic activity changes the state of amygdalar, thalamic, striatal, and neuromodulatory circuits. The resulting effects can alter sensory processing, learning, autonomic preparation, and action together. There is no requirement that every influence on PFC first pass through the hypothalamus, or that every hypothalamic influence reach PFC directly.
36.4.2 Insula and amygdala: a cue acquires bodily significance
The insula, concealed within the lateral sulcus beneath the opercula, is an important meeting place for information about the body and events that matter to it. Its subdivisions differ. Gustatory, visceral, somatosensory, and more integrative responses should not be collapsed into one undifferentiated feeling signal. Macaque tracing establishes connections from particular insular and opercular territories to orbital and medial prefrontal networks [@saleem2008complementary]. Through these connections, a sensory object can participate in frontal processing together with information about its bodily consequences.
The amygdala supplies another route. Its nuclei receive and exchange sensory and learned information, and its projections reach multiple prefrontal territories with marked regional differences. Rhesus tracing shows that amygdalar input is not confined to a single ventral emotional compartment [@barbasdeolmos1990]. Its relationships with orbital and medial frontal cortex are particularly important for linking an encountered cue to predicted consequences and bodily responses.
A mouse experiment makes the distinction between state and cue concrete. Livneh and colleagues recorded insular neurons responding to learned visual cues for food. Responses were strongly biased toward the food cue during hunger and weakened with satiety. Activating hypothalamic AgRP neurons could restore hunger-like insular responses in sated mice. Circuit mapping and pathway manipulations implicated a route through paraventricular thalamus (PVT) and basolateral amygdala [@livneh2017foodcues]. PVT is a thalamic nucleus; it must not be confused with the hypothalamic PVN.
The demonstrated cortical endpoint was insula. Primate insula–frontal anatomy makes this finding relevant to our larger account, but joining the two studies does not establish one continuous human AgRP-to-PFC circuit. What the mouse experiment does establish is powerful enough: a homeostatic system can change the cortical processing of a learned environmental cue. The body can change what the nervous system makes of the scene before the animal chooses an action.
36.4.3 Hunger and thirst do more than turn a drive up or down
The energy chapter’s sections “Sensors and signals: reserves, meals, and predicted calories” and “The Affective and Behavioral Drive” already showed why a food’s significance changes with the body. Here the anatomical addition is that direct hypothalamic input and routes through insula, amygdala, thalamus, and connected frontal systems provide several opportunities for that change to influence an expected outcome. A food can remain identifiable while losing its present claim on behavior. A Common Currency will examine that separation in detail.
Thirst provides a complementary case. The “Ingestive sensing: prediction before absorption” section of Chapter 8 established that oral and gastrointestinal evidence of drinking can reduce thirst-related activity before the absorbed water repairs the deficit. Frontal choice therefore takes place in a body whose controller is already revising its estimate of what is needed. The relevant state is changing throughout the act of obtaining and drinking water.
Human imaging offers evidence that water’s cortical representation changes with this state. In one study, responses to water in caudal OFC were present during thirst but were no longer detected after drinking to satiety, while primary taste cortical responses were less dependent on thirst [@dearaujo2003water]. This is evidence for state-dependent cortical processing, not a traced line from a particular lamina-terminalis neuron to OFC. The anatomical account remains distributed: oral and visceral evidence, regulatory circuitry, insular and amygdalar relationships, and frontal connections together make water available as an outcome whose current significance can change.
36.4.4 Protection, shelter, reproduction, and other bodies
Food and water make the physiology easy to name. Shelter is more relational. A cavity can protect against a predator, conserve warmth, or provide a place to raise offspring. Whether it is useful depends on the animal’s condition, the properties of the place, what has happened there, and what may happen next. There is no reason to expect a single shelter signal to travel from hypothalamus to PFC.
Instead, several kinds of evidence meet. Visual and other association cortex supplies information about places and agents. Hippocampal and related cortical systems contribute remembered context and routes. Amygdalar connections make learned danger and biological significance available to orbital and medial frontal systems. Hypothalamic, brainstem, and insular influences relate those possibilities to cold, injury, arousal, or other bodily conditions. This is an interpretation of what the established connections permit: the relative contribution of each route must be tested in a particular defensive or shelter-seeking task.
Reproductive and social motives likewise combine physiological state with the circumstances in which action would occur. The hypothalamus overview placed gonadal state, sexual behavior, caregiving, defense, and social relationships within organismal regulation. Hormonal sensitivity in hypothalamic and related circuits changes the animal’s readiness to act. Sensory and temporal association systems distinguish potential partners, offspring, rivals, and their behavior. Connections through amygdala, insula, medial and orbital PFC, and striatum allow that information to interact with remembered relationships and predicted consequences. These relationships provide candidate routes by which reproductive state can influence frontal control; they do not identify a dedicated sexual-drive cable or equate rodent pheromonal circuits with human social life.
The important extension is that an immediate motive can be evaluated within a situation that is partly absent. A safe place may be remembered rather than visible. A potential partner’s response must be anticipated. A dependent offspring may be elsewhere. The older regulatory problem persists while the information governing it becomes more extensive.
36.4.5 Context, memory, and the scene that is not present
Frontal systems also receive the information needed to distinguish these situations. Temporal association connections contribute object and agent information; parietal connections contribute spatial and action-related relationships. Intrinsic frontal connections allow these influences to meet rules and goals represented in other frontal populations [@saleem2014lateral].
The hippocampal contribution is anatomically differentiated. In rhesus monkeys, direct projections from hippocampal formation, including CA1 and subicular regions, preferentially reach particular medial and orbital prefrontal territories [@barbasblatt1995]. This is not a uniform hippocampal projection to all PFC. Wider frontal–memory interactions also involve parahippocampal, retrosplenial, other cortical, and thalamic systems. Nor does a direct hippocampal input imply that every return influence follows the same axons in reverse.
These connections help explain how the same scene can support different actions depending on remembered context. Human OFC activity can distinguish hidden task states that must be inferred from recent events [@schuck2016map]. Hippocampal or ventromedial prefrontal damage can disrupt the construction of detailed remembered and imagined situations [@hassabis2007imagine; @bertossi2016future]. The anatomy places frontal cortex within the construction and use of these situations. It does not place the remembered world inside the frontal lobe.
36.5 Recurrent connections: keeping a condition effective
Incoming information does not accumulate in PFC until a decision is ready to leave. Frontal neurons are active within circuits that continually alter one another. Local recurrent connections, long-range cortical relationships, basal-ganglia pathways, thalamic input, and neuromodulation help determine which population states persist, change, or lose influence. The maintained condition belongs to that activity, not to a separate executive supervising it.
36.5.1 Frontal entry into basal-ganglia circuits
Choosing What to Do, especially “The striatum receives the candidate state of action,” established differentiated but overlapping corticostriatal territories. What It’s Worth extended that account through “Learning is distributed across a striatal gradient” and “The ascending spiral links striatal territories.” We can now give the frontal participants more definite addresses.
Ventromedial and orbital frontal areas project into ventral and rostral striatal territories. Dorsal anterior cingulate projections reach substantial rostral striatal fields, and dorsolateral prefrontal projections are prominent in associative caudate and adjoining striatum. These are biases rather than sealed compartments. Macaque tracing reveals dense focal terminal fields together with more diffuse projections; inputs from orbital, cingulate, and dorsolateral territories occupy both separate and convergent zones [@haber2006striatal]. Anatomical overlap provides an opportunity for interaction, although it does not by itself show that two projections contact the same individual neuron.
A remembered rule can therefore change the cortical input arriving at striatal populations already influenced by context, expected outcome, and dopamine. Activity through the direct, indirect, and hyperdirect routes described in Chapter 34 changes the pattern of inhibitory output from GPi and SNr. In ventral circuits, ventral pallidum is also a major output intermediary. Those influences reach thalamic and subcortical targets, changing both frontal activity and access to ongoing action systems.
The same logic applies whether the behavior is a gaze shift, a reach, or a step in a larger task. The circuit does not need an abstract thought to become a completed motor command before it can matter. A change in cortical state can alter the conditions of competition while the action is being prepared and performed.
36.5.2 Thalamus participates in the state
The thalamus is often drawn as the last relay in a cortex–basal ganglia–thalamus–cortex circle. That drawing hides the continuing interaction. MD and frontal cortex communicate directly and reciprocally, with different MD territories related to different frontal regions. Ventral anterior and ventral lateral nuclei have strong relationships with premotor and motor systems. Macaque tracing also identifies nonreciprocal relationships: cortical input to a thalamic territory can extend beyond the cortex to which that territory projects. These arrangements permit interactions between frontal circuits as well as returns within them [@mcfarlandhaber2002].
The functional importance is visible in mouse experiments. In a delayed spatial task, inhibiting MD input to medial PFC during the delay impaired maintenance and disrupted delay-related cortical activity; the returning cortical projection to MD made a different contribution during subsequent choice [@bolkan2017thalamic]. Another study found that MD supported the cortical representation of an attentional rule without simply copying the rule’s categorical content [@schmitt2017amplification].
These results give substance to the word recurrent. A frontal state that appears to reside in cortex can depend on a thalamic partner remaining active. Holding the Line will examine how persistence can be expressed in sustained firing, changing population activity, and temporarily latent states. Here the essential point is anatomical: keeping a rule effective may require continuing traffic through a circuit, not a signal left behind in one cortical area.
36.5.3 Dopamine changes the operation of frontal circuits
The dopamine account in Chapter 35 distinguished learning, wanting, vigor, and different signals across places and times. Here we add cortical dopamine. Dopaminergic axons reach PFC as well as striatum, and their effects depend on the receptors and local circuits they encounter.
The ventral tegmental area (VTA) is an important source, but the familiar VTA-to-PFC arrow is incomplete for primates. Tracing combined with dopamine-related histochemistry in macaques identified frontal-projecting neurons across a continuum including VTA, dorsal substantia nigra, and retrorubral cell groups. The distribution differed with the frontal target [@williamsgoldmanrakic1998]. This is the dopaminergic substantia nigra associated with pars compacta and adjacent dopamine groups, not the GABAergic SNr output system discussed above. Cortical and striatal dopamine projections have differentiated organizations; they are not two identical copies of one message.
Within PFC, dopamine changes how ongoing synaptic input affects neuronal activity. One especially instructive result concerns D1 receptors in monkeys performing a spatial working-memory task. Modest D1 stimulation sharpened neuronal tuning by reducing responses to nonpreferred directions. Stronger stimulation reduced delay firing more broadly and degraded tuning [@vijayraghavan2007d1]. More dopamine was therefore not equivalent to better maintained control. The effect depended on dose and the activity being regulated.
This supplies a mechanism for changing the effectiveness of a control state without supplying its content. Dopamine need not specify the remembered location to alter how selectively it influences the network. Nor should the striatal D1/direct and D2/indirect diagram be transferred to cortex: cortical receptor effects operate in a different cellular architecture. The example concerns D1 modulation of a particular task and circuit, not a universal formula for every prefrontal operation.
The interaction has return routes. In rats, prefrontal axons form synapses in VTA with target-specific relationships to dopamine and GABA neurons [@carrsesack2000]. Frontal activity can also influence midbrain circuitry through the striatal and pallidal relationships already introduced. Thus neuromodulation is embedded in the same recurrent organization. The controller helps shape the activity of systems that change its own operating conditions.
Dopamine is only one contributor. Noradrenergic, cholinergic, serotonergic, and hypothalamic arousal systems also influence frontal processing. Their presence matters to the argument: a maintained rule is a physiological achievement of a waking, regulated organism. It can become ineffective when the network’s operating state changes even though the rule remains part of the person’s knowledge.
36.6 How frontal activity changes action and the body
The outgoing question is the counterpart of the incoming one. A represented consequence must change something in the organism if it is to guide behavior. Frontal cortex has several routes through which it can alter attention, movement, and bodily preparation, and those routes can be effective together.
36.6.1 Through cortical action systems
Connections with sensory and parietal cortex allow frontal states to change the evidence that remains influential. Connections with premotor and medial motor territories bring rules and goals into networks organizing action. The routes are selective. Different frontal areas contact different action-related partners; there is no single exit from PFC labeled execution.
The distinction between pre-SMA and SMA proper gives a useful anatomical example. In macaques, rostral medial area F6, usually called pre-SMA, receives strong prefrontal input, including from area 46. More caudal F3, SMA proper, has stronger relationships with primary motor and other motor territories. Their different connections help explain how contextual control can influence movement through more than one cortical step [@luppino1993sma]. Selected prefrontal areas also connect with cingulate motor territories [@batesgoldmanrakic1993]. The medial wall is therefore an interface between association-level conditions and motor organization, not one homogeneous motivation area.
A remembered instruction can change a premotor population’s readiness to reach, or a medial motor population’s contribution to a sequence. These cortical changes interact with the basal-ganglia pathways just described. Motor and premotor outputs then engage the descending machinery developed in Unit 5: cortical, brainstem, and spinal systems that coordinate movement, posture, orientation, and sensory correction. The prefrontal goal does not specify every muscle contraction. It changes the conditions under which those established controllers act.
36.6.2 Through basal ganglia and subcortical controllers
Frontal influence also reaches action through striatum and STN, then pallidal and nigral output. As “GPi and SNr provide tonic inhibitory output” in Chapter 34 established, thalamus is one target, and superior colliculus and brainstem systems are others. An action can therefore be influenced through a return to cortical motor networks, through altered access to subcortical orienting systems, or through both.
This arrangement is particularly important for initiation. Representing a possible action, accepting its cost, preparing the body, and recruiting movement can fail in different combinations. The broad selection problem includes vigor, continuation, termination, and switching, as well as beginning. Tomorrow Never Comes will separate these contributions using the effects of prompting and the clinical patterns of apathy, abulia, and akinetic mutism. The anatomy here supplies the routes through which an available goal could become effective.
36.6.3 Through hypothalamic and brainstem regulation
Frontal influence is also physiological. Orbital and medial prefrontal projections reach hypothalamic territories involved in autonomic regulation. Rhesus studies have demonstrated cortical synapses within hypothalamus and mapped hypothalamic projections to brainstem autonomic regions and the spinal intermediolateral column [@barbas2003autonomic]. The anatomical evidence supports a relatively short route from association cortex toward bodily effectors. It does not identify every link as a synapse on the same traced neuron projecting to the spinal cord.
Selected medial and orbital frontal territories also project directly to the periaqueductal gray (PAG) in macaques [@an1998pag]. These projections provide another route into systems coordinating bodily responses and organized behavior. Their effect depends on the cortical origin, PAG target, and active circuit; a line to PAG is not a generic command to freeze, flee, or act. Frontal influences through amygdala and hypothalamus add further routes.
Through this anatomy, a represented situation can help change cardiovascular preparation, arousal, and the organization of defensive or appetitive behavior before the anticipated event occurs. Endocrine effects depend on downstream hypothalamic and pituitary circuitry; the existence of a cortical projection to one hypothalamic region does not imply a direct projection to every neuroendocrine population. The point is more specific: association cortex can participate in regulating the body’s operating state through identified access to regulatory systems.
Feeling the Future develops this prospective preparation and the returning bodily information. It will distinguish measured autonomic changes from stronger claims about how those changes determine choice. The body has been present throughout the unit; that chapter examines one part of the interaction more closely.
36.6.4 Consequences return while the action continues
Imagine the animal moving toward a remembered shelter. Frontal and connected systems help keep the destination and current conditions effective. Cortical and subcortical action systems organize orientation and locomotion. Regulatory systems prepare the body for the effort. As the animal moves, visual, vestibular, proprioceptive, and bodily signals change. An opening may prove blocked; breathing may become labored; a predator may appear. Each change can alter the same networks that are maintaining the course.
The division into incoming and outgoing pathways is an aid to learning anatomy. It is not a claim that the organism first reads the world, then completes a decision, and only then acts. Movement generates evidence. Bodily preparation changes the state against which outcomes are valued. Predictions change before their final consequences are known. Control is the continuing interaction among those processes.
This is the control-theoretic point I want the anatomy to make. A goal is effective when it changes the evolving activity of the controller and, through that activity, the course of the body. A remembered condition can persist as an influence while the particular neurons and movements expressing it change. A changed consequence can reorganize the course before the original action is complete.
36.7 Opening the box without inventing an executive
36.7.1 What the evidence establishes
The anatomy and experiments answer different questions. Tracing establishes origins and terminations, and with appropriate methods can identify synapses. It does not reveal the message carried during a particular behavior. Recording identifies activity related to an outcome, rule, or action; a correlation alone does not establish that the recorded activity is necessary. Perturbation tests causal contribution, but its meaning depends on which cells or projections were changed, when, and what else the manipulation affected. Human imaging gives access to distributed representations and interactions while usually lacking the cellular specificity of animal experiments. Functional coupling is not proof of a direct axonal projection.
Clinical dissociations give these distinctions their force. A person may know a rule yet fail to use it when several demands compete. A visible object may capture behavior despite the larger task. Another person may retain many capacities while rarely initiating action. Similar failures can follow damage outside frontal cortex, including connected thalamic structures [@duncanetal1996goal; @lhermitte1983utilization; @eslingeretal1991thalamic]. Lesions affect pathways and distant activity as well as local tissue. They help identify what a functioning network enables, provided the behavioral analysis is specific.
The species comparisons remain visible for the same reason. Primate tracing supplies much of the differentiated frontal anatomy; rodent circuit manipulations resolve mechanisms inaccessible to routine human experiments; human studies reveal the consequences for language, extended plans, social commitments, and daily life. Their agreement is informative. Their anatomical differences also constrain the account.
36.7.2 Boxes within boxes: the six problems ahead
The hypothalamus overview explained why a recurrent brain cannot be taught by opening every circuit at once. We treated some systems as boxes with identifiable inputs and outputs, then returned to examine their internal machinery. Unit 5 opened the action and learning systems. This overview has now opened the frontal box far enough to locate its major territories, incoming influences, recurrent partners, and routes to action and bodily regulation.
The following chapters take six interacting problems in turn. They are explanatory views of the same organism, not six successive processing stages.
A Common Currency asks how unlike expected outcomes become comparable enough for one choice. Food, water, safety, and information retain different identities and bodily consequences. The chapter examines how shared and outcome-specific representations let them influence the same competition while remaining available for selective revision.
Holding the Line asks how selected information remains effective across delay, distraction, and changing circumstances. Persistence must protect a useful rule without trapping the organism in an obsolete one. The chapter develops the recurrent mechanisms and the distinction between possessing information and using it to govern action.
Feeling the Future examines how anticipated consequences prepare the body and how bodily information participates in ongoing choice. It develops interoception and autonomic evidence, then evaluates the somatic-marker account and its limits within the embodied architecture already established.
The Futures of Others asks how another agent’s knowledge, intentions, and likely reactions influence present behavior. A social world introduces controllers whose actions must themselves be anticipated. Knowing a social rule and using it during an interaction become distinguishable problems.
Wandering When Full builds on Chapter 15. That chapter established the regulatory circumstances and investigatory circuitry through which sampling can improve future controllability. Here the additional questions concern the value of information, directed and random exploration, and how the expected opportunity to use new knowledge changes whether acquiring it is worth the present cost.
Tomorrow Never Comes asks how an available and potentially valuable goal recruits an initiated and sustained course of action. Prompting, effort, vigor, and spontaneous initiation can dissociate. The chapter uses those differences to examine the gap between an action that can be described and a body that begins it.
These problems continually reshape one another. A change in thirst changes the significance of water; that change can reorganize a maintained plan, alter bodily preparation, or end exploration. New information can change the expected consequences of a familiar action. Another person’s likely response can outweigh an immediate opportunity. No chapter hands a finished product to the next cortical region.
The larger course will return to the maps and memories from which remembered situations and possible futures are constructed. Our immediate next step is the comparison problem. The animal can be governed by what is not yet here, but several absent outcomes may compete for the same body. How do consequences as different as food, water, shelter, and information become comparable enough for one course of action to prevail?
That is where A Common Currency begins.