Causal investigations into orbitofrontal control of human decision making

Abstract

Although it is widely accepted that the orbitofrontal cortex (OFC) is important for decision making, its precise contribution to behavior remains a topic of debate. While many loss of function experiments have been conducted in animals, causal studies of human OFC function are relatively scarce. This review discusses recent causal investigations into the human OFC, with an emphasis on advances in network-based brain stimulation approaches to indirectly perturb OFC function. Findings show that disruption of human OFC impairs decisions that require mental simulation of outcomes. Taken together, these results support the idea that human OFC contributes to decision making by representing a cognitive map of the task environment, facilitating inference of outcomes not yet experienced. Future work may utilize similar non-invasive approaches in clinical settings to mitigate decision making deficits in neuropsychiatric disorders.

Introduction

It is well recognized that the human orbitofrontal cortex (OFC) is involved in decision making, and that OFC dysfunction is implicated in neuropsychiatric disorders such as substance use disorder (SUD) [1–3], obsessive compulsive disorder (OCD) [4, 5], and depression [6, 7]. Understanding the precise role of this region in behavior is therefore of critical importance. Over the past 35+ years, work across species has attempted to answer this question, resulting in a range of proposals [8–14]. The diversity and length of this list makes it clear, however, that pinpointing the specific function of OFC is not straightforward. Moreover, the ubiquity of decision-making variables correlated with OFC activity, even in individual studies [15], makes it difficult to identify those that critically contribute to behavior. To get a better understanding of causal contributions, loss of function experiments are required.

Many studies in animals have utilized a variety of behavioral tasks in combination with experimental lesions or inactivation of OFC to understand its function. These studies are summarized in several excellent review papers [8, 16–19]. In contrast, causal studies on the contribution of the human OFC to behavior are sparse. The purpose of this review is to summarize recent studies that use causal methods to test the function of the human OFC, with an emphasis on recent advances in non-invasive brain stimulation techniques designed to indirectly target and modulate activity in OFC networks. We end by discussing what we have learned from these causal studies about the function of the human OFC.

Causal tests of human OFC function I: lesions

How can we gain causal insight into the contribution of human OFC to behavior? A potentially useful approach is to identify tasks and associated behaviors that are sensitive to OFC damage in animals. One such example is the devaluation task [20], in which behavioral responding to cues is measured before and after devaluation of food rewards associated with these cues (Figure 1A). Food rewards can be devalued by selective satiation, or by pairing them with malaise induced by injections of LiCl. Intact animals typically demonstrate a change in response rate after devaluation, such that they avoid cues predicting the devalued outcome. In contrast, lesions or inactivation of the central and lateral OFC in rats [21–23] and non-human primates [24–26] result in continued responding to the cues associated with devalued outcomes (Figure 1B), even though devaluation itself remains unaffected (Figure 1C). The devaluation task is thus a reliable assay of OFC function in animals, and a good starting point for causal investigations in humans.

Figure 1: Adaptive behavior in the devaluation task depends on the OFC.

A. Schematic of the devaluation task. Subjects first learn associations between cues (or actions) and food rewards. In a baseline session, subjects make choices among these cues (or actions). Next, devaluation can be achieved by feeding to satiety on one of the two foods. Choices between cues (or actions) are then reassessed in a probe session, typically under extinction conditions (i.e., cues/actions are offered but no reward is given). B. Subjects with intact OFC reduce choices for cues or actions previously paired with the now devalued reward in the probe choice session, whereas subjects with inactivated OFC continue to choose cues or actions predicting devalued rewards. C. Liking of the devalued food is not affected by OFC inactivation and decreases similarly in both groups.

A common methodological strategy for causal studies in humans is to test participants with “naturally” occurring lesions, such as those caused by strokes, tumors, or metal rods [27]. A recent study tested whether participants with damage to medial OFC and adjacent ventromedial prefrontal cortex (VMPFC) show altered behavior in this task [28]. Similar to what has been observed in animals, relative to control groups, human participants with OFC damage continued to select actions associated with food outcomes that had been devalued through selective feeding [28]. Importantly, participants were not impaired in learning the original action-outcome contingencies, and also appropriately decreased their liking ratings of the consumed food. Thus the impairment was specific to choices in which subjects had to access the new value of the devalued outcome to make a decision.

Participants with OFC damage also show other abnormalities that suggest deficits in decision making [29]. For instance, participants with VMPFC damage are impaired in making choices when the value of objects is related to a configural combination of attributes [30], and differently weight the value of object features [31]. These findings suggest that OFC is critically involved in integrating the value of complex multi-attribute objects.

While studies involving participants with brain damage are indispensable for understanding causal contributions of the human OFC, they have certain limitations that require careful designs and interpretation [32]. Most importantly, naturally occurring lesions are not restricted to specific areas of interest such as the OFC, and thus can include damage to adjacent cortical areas and white-matter tracts. In addition, there is evidence for compensatory adaptations and distributed neurological changes after brain injuries, potentially causing changes in cognition and behavior independent from the loss of the lesioned area itself [33, 34].

Causal tests of human OFC function II: non-invasive brain stimulation

A complementary approach to studying participants with naturally occurring lesions is the use of non-invasive neurostimulation methods. For instance, brain activity can be modulated non-invasively using transcranial magnetic stimulation (TMS) [35], which uses rapidly changing magnetic fields applied to the scalp to induce electric currents in specific underlying brain areas. Depending on the frequency and pattern of stimulation, TMS can have excitatory or inhibitory after-effects that may last several minutes to days, reflecting long-term potentiation or depression [35]. Two commonly used protocols are repetitive TMS (rTMS, pulse trains applied at 1–20 Hz) and continuous theta burst stimulation (cTBS, 3-pulse bursts at 5 Hz). Whereas rTMS is thought to enhance neural activity in the underlying tissue, and cTBS has been shown to have inhibitory effects over motor cortex [36], the effects of any given TMS protocol may vary across brain areas [37]. The main advantage of TMS is that it can be applied experimentally (i.e., in randomized groups) in healthy subjects over specific anatomical locations. In addition, because the effects of TMS are transient, it can be applied at various stages of experimental tasks in order to isolate the contribution of an area to specific cognitive functions, such as perception, valuation, choice, or learning.

Although experiments using TMS to study OFC function are relatively rare, several studies have targeted OFC with TMS in clinical settings such as SUD [38, 39] and OCD [40, 41]. For instance, rTMS over the right OFC (fronto-polar area 2 [Fp2] in the 10–20 EEG system) has been shown to decrease symptom severity in OCD patients and reduce activity in the lateral OFC [41]. Similarly, in patients with alcohol use disorder, cTBS over left OFC (Fp1 in the 10–20 EEG system) has been shown to decrease task-related fMRI activity in the OFC relative to sham [38]. These findings suggest that TMS applied over lateral-anterior OFC can directly alter activity in the underlying tissue.

However, there are limitations regarding to which parts of OFC can be directly targeted with TMS. Unlike the fronto-polar areas targeted in the studies summarized above, areas in the central OFC are too deep relative to the surface of the skull to be directly accessible to TMS. This is not only a problem for OFC but also for subcortical areas such as the striatum and the hippocampus. However, several studies have shown that effects of TMS are not restricted locally to the stimulated tissue, but can be observed in regions that are anatomically and functionally connected to the stimulation site [42–45]. While this potentially complicates the interpretation of TMS results with regards to isolating the function of specific brain areas, it also offers a solution to the problem of perturbing activity in deeper brain structures. Specifically, instead of targeting activity in brain areas directly underneath the coil, TMS studies can be devised to manipulate activity in distributed brain networks [46].

One particularly successful strategy has been to focus on stimulating sites on the cortical surface that, based on resting-state fMRI, are maximally connected to deeper brain areas or networks of interest [47–50]. For instance, rTMS applied over coordinates in parietal cortex that are maximally connected to the hippocampus enhances hippocampus-dependent episodic memory [48]. Importantly, such network-based stimulation also induces a relatively long-lasting modulation of the functional connectivity of the hippocampal network, which scales with the effects on memory [47]. Findings such as these demonstrate the feasibility of indirectly targeting brain areas that are not directly accessible with TMS.

Network-based stimulation of human OFC

The OFC is densely connected to much of the prefrontal cortex (PFC), and more selectively to areas in the parietal and temporal lobes [51]. The intrinsic anatomical connections within the primate OFC have been delineated into a medial and central/lateral network [52], which can also be identified in humans using functional connectivity-based parcellation (Figure 2A) [53, 54]. The medial network comprises medial PFC, posterior cingulate and medial parietal cortex. The central/lateral network comprises dorsomedial PFC, more ventral areas of lateral PFC (LPFC), and posterior parietal cortex [54, 55] (Figure 2B). Of note, the central/lateral network most closely aligns with the lateral OFC areas typically inactivated in animal studies using the devaluation task [21–26], whereas the medial OFC network largely overlaps with the medial OFC and VMPFC areas covered in human lesion studies [28–31].

Figure 2: OFC networks and OFC-targeted brain stimulation.

A. Connectivity-based parcellation of OFC voxels into central/lateral (blue) and medial (red) OFC networks, adapted from [54]. B. Surface plots depict areas on lateral (top) and medial (bottom) surface that are connected with the central/lateral (blue) and medial (red) OFC networks. C. The indicated stimulation site in LPFC has high resting-state connectivity with the target area in the central/lateral OFC. TMS can be applied to individually identified coordinates within LPFC that show maximal connectivity with the OFC target.

We recently tested whether it is possible to selectively target the central/lateral OFC network using a connectivity-based stimulation approach, and whether this would alter behavior in the devaluation task [56]. Hungry subjects learned associations between cues and sweet or savory food odors before performing a choice task in which they were free to choose between cues that were associated with these odors. To individually define stimulation sites, we seeded a resting-state fMRI connectivity analysis in a region of the central/lateral OFC previously shown to represent outcome expectations [57–59]. This identified a stimulation coordinate in LPFC that was maximally connected to the intended OFC target (Figure 2C). After a single session of cTBS or sham with expected aftereffects of 50–60 minutes, participants underwent a second resting-state fMRI scan, and then had unlimited access to food corresponding to either the sweet or the savory food odor. This selective devaluation procedure was followed by a post-TMS choice test.

To examine whether OFC-targeted cTBS modulated OFC network activity, we tested whether the stimulation changed the resting-state fMRI connectivity between OFC and the rest of the brain. We found that TMS reduced the connectedness of the OFC, such that connectivity within the central/lateral OFC network, but not the medial OFC network, decreased in the stimulation relative to the sham group [56].

Disrupting OFC network connectivity using cTBS also altered behavior in the devaluation task. Relative to sham stimulation, OFC-targeted cTBS decreased subjects’ sensitivity to the devaluation procedure, such that they continued to select cues predicting the devalued outcome [56]. Importantly, similar to the effects of inactivation of central and lateral OFC in rats and non-human primates [23, 25], these choices were altered despite the fact that subjects reported liking the devalued odor less, and did not show altered value-based choices more generally. Finally, the effects of cTBS on behavior and OFC network connectivity were correlated across participants in the stimulation group, directly linking cTBS-induced changes in OFC network activity and task behavior [56]. There were no comparable effects on LPFC connectivity, suggesting that the changes in behavior were driven by effects on OFC connectivity rather than the directly stimulated LPFC site.

These findings provide initial evidence that non-invasive network-based stimulation can be used to perturb OFC activity and to disrupt behavior that in animals depends on intact OFC. Notably, although cTBS caused behavioral effects comparable to those of aspiration and excitotoxic OFC lesions [22, 24], as well as pharmacological [25] and optogenetic inactivation of OFC [23], the physiological effects of cTBS are all but certain to differ fundamentally. However, they may involve similar changes in coordinated activity throughout the OFC network.

Open questions about OFC-targeted TMS

The results discussed above demonstrate the feasibility of targeting OFC activity using network-based stimulation. However, several important questions about potential applications and limitations of this approach remain to be explored.

1. What are the physiological effects of OFC-targeted stimulation, and how do they differ between the indirectly targeted and the directly stimulated site? In this regard, is it possible to target OFC by means of alternative entry sites, for instance in the PFC or parietal lobe? This would help mitigating concerns about the involvement of other structures that are difficult to rule out with a single-site stimulation approach.

2. What is the spatial resolution and specificity of OFC-targeted stimulation? Is it possible to selectively target sub-networks of OFC, e.g. anterior vs. posterior or medial vs. lateral OFC? Recent work suggests dissociable contributions of different OFC sub-regions to behavior [16, 18], and the ability to selectively target these networks would allow us to causally test whether a similar heterogeneity exists in humans.

3. What is the time course of the effects of OFC-targeted stimulation on network activity and behavior? Much of our knowledge about the temporal dynamics of TMS comes from studies on motor cortex [36], but evidence suggests that TMS effects differ across brain areas [37]. Our studies suggest that behavioral effects last at least 35 minutes, but physiological short- and long-term dynamics remain to be explored. Studying the temporal profile of OFC-targeted stimulation effects would be important to design better experiments and for developing therapeutic applications.

4. Is it possible to use OFC-targeted stimulation to enhance rather than perturb OFC activity? This would be critical for the development of OFC-targeted therapeutic approaches that aim to enhance brain function. In this regard, it would be important to explore how the effects of different stimulation frequencies interact with oscillatory activity in the targeted areas.

How does the OFC contribute to decision making?

The results from causal studies discussed in this review provide novel insights into how the human OFC contributes to decision making. Most importantly, the human OFC appears critical for adaptive behavior in the devaluation task [28, 56]. Behavior in this task requires two functions: (1) the ability to update the value of a reward, and (2) to perform a mental simulation to access the new value of the outcome when faced with the predictive cue. Because OFC-targeted cTBS [56] and OFC damage [28] do not disrupt the ability to change preferences for the devalued reward itself, OFC may be critical for simulating the value of outcomes that have not yet been experienced. This interpretation is also consistent with findings that OFC damage causes impairments in multi-attribute decision making [29–31]. That is, combining multiple attributes produces a large variety of possible objects (e.g., 3 attributes with 4 levels each = 64 unique objects), such that the value of individual objects may not be readily learned through experience but must be computed or simulated on-the-fly.

Moreover, the idea that OFC supports the simulation of outcomes is in line with studies in rats showing that OFC inactivation does not impair responding based on directly experienced outcomes, but causes a specific failure to respond when likely outcomes have to be inferred [60]. There is correlative evidence for a role of human OFC in such inference-based behavior using similar tasks [61], but causal studies would allow us to draw more definite conclusions. Indeed, unpublished data suggest that OFC-targeted cTBS in humans disrupts behavior that relies on outcome inference but does not impair responding based on direct experience [62].

The notion that OFC is necessary for the simulation of outcomes dovetails nicely with recent proposals that the OFC represents a cognitive map of task space [63, 64]. Within this framework, access to a set of associations or “map” allows individuals to simulate future states or “locations”, given a series of predictive relationships or decisions, even if the particular path between the current and future location has never been traversed. In the absence of this map, such simulations are not possible and individuals must rely solely on previously experienced associations to make a decision.

Conclusions

Initial evidence suggest that network-based TMS approaches can be used to non-invasively perturb OFC network activity in humans. This technique can complement studies in participants with brain damage and help to reveal the specific contribution of the human OFC to decision making. Recent work using this approach provides converging cross-species evidence that the human OFC is required for decisions that are based on inferred or simulated outcomes, as opposed to behavior that can be based on direct experience alone. While it may be a long way toward this goal, these techniques offer a promising avenue for developing novel treatments for neuropsychiatric disorders that are characterized by aberrant decision making and OFC dysfunction.