Affect and the Limbic System: Some Hard Problems

SIR: I enjoyed the excellent review of limbic architecture issues by Mega et al.¹ and the many first-rate reviews of limbic and subcortical systems in your summer 1997 special issue. However, none of the articles fully addressed problems concerning the “functional correlates” of the limbic system and the many difficult conceptual issues surrounding the nature of affect. The architecture and connectivity issues alone are formidable, as the borders of the limbic system are vague and have been extended decade by decade like the erosion of a vast neural shoreline. Various “extended” notions about the limbic system include a host of paralimbic, basal ganglia, thalamic and hypothalamic, basal forebrain, and subcortical systems, including even monoaminergic portions of the reticular activating system brainstem core.² As acknowledged by several of the articles, some models of the limbic system suggest that the entire prefrontal systems and heteromodal association cortex in the right hemisphere could be considered part of the limbic system—as its extended association cortex. One wonders what is left—what is not “limbic system” beyond idiotypic or primary cortex and regions such as Broca's or Wernicke's areas. The involvement of so many systems in affect speaks to the “supramodular” nature of affective systems and underlines that affect is not a “channel function.” It is in truth deceptively hard to model, much more so than the modular functions supported in more discrete thalamocortical connectivities.

This difficulty brings up basic conceptual troubles and the lack of any clearly validated “metapsychology” of affect—what Kenneth Heilman understatedly described as affect having several “dimensions.”³ Affect is probably best conceptualized (like attentional and executive functions) as a global state function,⁴ reflecting large-scale interpenetrations of the “state space” of “re-entrant” modules in many subcortical and cortical systems. The distributed and globally re-entrant nature of systems involved in emotion generates the intimidating multidimensionality of affect and its critical binding of many components. These bound modular components probably (at a minimum) include autonomic, endocrine, facial motor and global motor readiness, pain/pleasure, social/signaling, and cognitive (other/self appraisal) encodings. Hardly a “nonspecific” arousal system, affect shows instead this patterned global integration of multiple types of highly distributed information. However, there is typically focus on one or more, but rarely all, of these various features bound together in affect, a version of missing the forest for the trees that has often resulted in the study of emotion resembling the three (or three dozen?) blind men inspecting different portions of the elephant.

Taken as a whole, affect seems best conceptualized as a highly composite product of distributed neural systems that together globally organize the representation of value.⁵ As such, it probably functions as a master system of reference in the brain, integrating encodings done by the more modular systems supported in various relatively discrete thalamocortical connectivities. Given the central organizing nature of affect as a system for the global representation of value, and given evidence that virtually all stimuli elicit some degree of affective “valence tagging,” it would be hard to overestimate the importance of this valence tagging for all kinds of basic operations. The centrality of affective functions is underlined by the intrinsic interpenetration of affect, attentional function, and executive function, and it certainly makes sense that these three global state functions would be highly interdependent. It is logically impossible to separate representation of value from any neural mechanisms that would define attentional foci or that would organize behavioral output.

However, basic questions about the functional correlates of these highly distributed systems are not generally acknowledged. The most basic of these is how these distributed systems support the fundamental biphasic nature of affect (the “plus or minus” nature of all affective valence). This primary feature of affect has been appreciated for as long as there has been human culture: that we have loves and hates, likes and dislikes, attractions and aversions. Yet as fundamental as this is, its neural architecture is still frustratingly unclear. It seems safe to assume that the bipolarity of affect must be grounded in two linked distributed systems that would have to be push-pull and mutually inhibitory, as suggested in the seminal work on emotion and startle probe investigations.^6,7 Either system can activate behavior and inhibit the behavior activated by the opposing system, although, as Heilman points out, it gets complicated in that the positive valence systems responsible for pleasure and “plus” valence tagging cannot be totally equivalent to systems that mediate approach and attachment, nor can the negative affective systems be totally synonymous with those that activate defense, aversion, and detachment. Thus, although “valence tagging” must profoundly color and inform behavioral organization (and probably has large network overlap with executive systems that organize output), I would agree with Dr. Heilman that activation of motor readiness has to be conceptualized as a related but separate module in the system.

There is much work (ably summarized and reviewed by Heilman) that relates these differential affective valences to the two hemispheres (the right biased toward the experience of negative affects, and the left toward benign or positive affects), and also to anterior and posterior cortices. This linkage tempts one to suppose (I think much too neatly and simplistically) that modules for positive affect and negative affect are organized in left and right hemispheres (or in parietal and prefrontal regions). I am natively suspicious of this type of conceptualization, as it seems just too “corticocentric” (consistent with other trends in cognitive neuroscience). I suspect that this obscures the more fine-grained, distributed nature of these positive and negative affective systems, describing only endpoints or global activation tendencies that are derivative of a complex neurodevelopmental course not yet charted that builds highly distributed networks involving counterbalanced complementary structures at multiple levels of the neuraxis: multiple brainstem regions in the reticular core, multiple regions of hypothalamus, multiple basal forebrain regions and amygdala, several paralimbic (especially cingulate versus orbital frontal), and the highest neocortical regions (parietal versus prefrontal, and left versus right hemisphere). The notion that modules for negative and positive affect would be simply centered in higher cortical regions also does not integrate the empirical evidence that stimulation at each level of these distributed systems (brainstem, different portions of hypothalamus, septum versus amygdala, and different cortical regions) can elicit painful or pleasurable effects. Just as Broca's area gets control over bilateral motor neurons, specific right frontal regions probably get control over “dysphoria neurons” that may be highly distributed. At each level of the neuraxis there appear to be differential contributions of complementary and partially opposed structures to negative and positive affect.

But that is another problem—there can't be simple “pain and pleasure” neurons. It is tempting to conceive of regional cortical activations somehow being paired with discrete largely subcortical “reward” and “punishment” centers (such as ventromedial hypothalamus, ventral tegmental area, medial forebrain bundle, and septal regions on the one hand, and perifornical hypothalamus, periaqueductal gray, amygdaloid regions, and so forth, on the other hand). However, the “valence tagging” of experience itself—a felt sense of emotional pain and pleasure attached to various stimuli in consciousness—cannot be reduced to simple activations of pain and pleasure neurons, as this becomes ultimately phrenological and homuncular. And yet there are activations that are inherently pleasurable, such as septal stimulation. That some neural activations “feel good” and others “feel bad” is a most puzzling phenomenon. Why is that the case? Again, these are very basic, even primitive questions, and yet we are still dismally ignorant about these fundamental building blocks. Perhaps even more troublesome is that neuroscience seems more comfortable with ignorance about this than with ignorance about vision or language. Given evidence⁸ that affect neurodevelopmentally underpins the generation of “higher functions,” this may be analogous to understanding calculus before learning how to count on one's fingers.

This bipolarity appears intrinsic to affect as embodying a representation of value: the notion of value, of course, implies either positive or negative. Although this is probably grounded in hypothalamic set point detectors (giving human love and hate their unavoidable biological grounding in homeostasis and survival) early on in life limbic feature detectors (to categorize social events and enable attachment) must be piggybacked on top of the hypothalamus. There is very early recognition (missing in autism) that human objects are a unique class of objects in the world. As MacLean⁹ points out, the limbic system evolved as mammals nurtured their young, suggesting that affect is probably inseparable from functions of attachment and social relations. Given its huge importance, attachment is dismally understood. But here again, thorny conceptual problems go unacknowledged. What enables smiling and the beginnings of symbiotic attachment in the 3-month-old infant? What are those early limbic feature detectors tuned to? What are the structures/networks/modulators crucial to these early affective operations? Obviously these are profound formative processes, marking the real beginnings of a person. Schore's monumental review⁸ addresses some of these issues somewhat later with regard to the practicing subphase, but this is well after the initialization of affect and attachment after the first two or three months. One suspects that both amygdala and various basal forebrain regions, which myelinate early, must be crucial here, probably along with paralimbic cortices and ventral basal ganglia, but early neurodevelopmental foundations for affect and attachment are still terra incognita. Curiously, there is relatively little interest in defining this basic ground, at least relative to other problems.

In any case, your special issue on limbic and subcortical systems is a much-needed counter to the all too “corticocentric” notions dominating some sectors of neuroscience. I would be most interested in the thoughts of Heilman, Mega et al., and other contributors about mapping networks supporting this crucial process of “valence tagging,” the related neurodevelopmental issues of defining early networks initializing affect, and the nature of limbic feature detectors. There is overwhelming evidence that affect serves central organizing functions in the formation of meaning by “valence taggings.” This would also be a point of strong junction between the best in psychoanalysis and a neuroscience that has moved beyond a myopic focus on isolated cognition, toward a more integrative neuroscience of the whole person.