Reward System Abnormalities in Anorexia Nervosa Navigating a Path Forward
Anorexia nervosa (AN) is associated with the highest mortality rate of any psychiatric illness1 as well as significant health care costs and lost wages. While there have been notable advances in understanding biobehavioral mechanisms of AN, the brain systems that underlie the illness remain poorly understood. Clinically, it is widely accepted that the critical first step in treatment is renourishment—that is, restoring individuals to a healthy body weight. Yet knowing that the primary medical intervention is simply to eat does not, in itself, change behavior. In fact, even after full weight restoration, individuals with AN continue to restrict caloric and fat intake, which is associated with poor longer-term outcomes.2 The severity and persistence of this illness makes understanding the pathophysiology and neural mechanisms of AN a research priority.
In a bottom-up approach to identifying brain mechanisms of illness, neural systems that have been well characterized in healthy individuals are probed to test hypotheses about abnormalities in psychiatric illness. In this issue of JAMA Psychiatry, Frank et al3 present the most recent findings from programmatic research examining reward and related neural systems in AN. Their model hypothesizes that the dopamine system is affected by AN-induced starvation. They propose that this is reflected in heightened prediction error response, a neural activation pattern associated with dopamine and thought to be a critical learning signal in the brain,4 and that stress and high anxiety influence prediction error to perpetuate illness. This kind of integrative model of complex illness with programmatic testing of neurobiological hypotheses is an aspiration in the field of eating disorders. The work draws together ideas about harm avoidance traits, high cortisol levels, and abnormal neural responses to taste, which have been examined to some extent in prior work.5 Furthermore, this study includes data from 56 adolescents and young adults with AN,3 a sample size much larger than most studies in this field.
In the study task, participants were presented with a stimulus and learned which images would be followed by receipt of a sweet taste. The associations become probabilistic, and on some trials, the predictions were violated and sweet liquid was delivered in response to the wrong stimulus or not delivered when it was expected.3 Prediction error occurred when the sweet taste was either omitted or was received unexpectedly. Functional magnetic resonance imaging measures the blood oxygen level–dependent signal response in areas known to reflect the dopamine prediction error signal, such as the striatum. The ingestion of sweet taste as an outcome is presented as having relevance in eating disorders based on an assumption that this relates to a central abnormality in AN. Prior studies from this group have shown differences in neural activation patterns associated with prediction error among adults with AN and recovered from AN, both for taste and monetary outcomes.6,7 Here, they extend this research to study adolescents and young adults, thereby examining this neurocognitive process closer to the onset of illness. The main finding is that there is an elevated prediction error neural response in the striatum among adolescents and young adults with AN, similar to what they have found in adults with AN (although adults also showed increased activation in the orbitofrontal cortex6).
This ambitious model of illness creates an opportunity for us, as a field, to consider all possible interpretations and to be conscious of the tendency for confirmation bias. As such, there are important caveats in interpreting the findings. The bottom-up approach provides information about differences in putative dopaminergic prediction error signaling but does not directly address behavioral disturbances that are central to the illness. Frank et al3 propose a compelling model of illness, suggesting that starvation alters the dopamine system to produce heightened prediction error sensitivity as a physiological adaptation that is intended to drive eating; they further suggest that this goes awry because the ventral striatum signals to the hypothalamus and inhibits appetitive signals. However, the paradigm to date has not tested—in animal models, healthy individuals, or patients with AN—whether passive learning and prediction error response is actually associated with eating behavior. Although the authors suggest that an association with illness-related phenomena is present based on self-reported assessments (primarily harm avoidance, which is ostensibly a trait measure), translational research in eating disorders would be strengthened by getting closer to measurement of actual behavior (eg, direct measurement of restrictive food intake).8
The challenge in understanding the meaning of neural activation patterns is a pervasive conundrum in neuroimaging research; therefore, cognitive neuroscience studies often include manipulations that aim to constrain the interpretation. The prediction error paradigm is essentially a passive learning task, with no instrumental response component. O’Doherty et al9 originally tested learning of the association by measuring when pupillary response shifted from the reward receipt to the cue presentation in an effort to apply rigor and narrow the possible interpretations. Research in decision making has demonstrated that instrumental responses can influence behavior.10 It cannot be assumed that research that includes only passive responding is necessarily relevant to actual decisions about what to eat. In the present study,3 for example, the finding that the groups did not differ in ratings of the pleasantness of the sweet taste may undermine the inference that reward or prediction error was associated with psychopathology.
Understanding the meaning of the neural findings in this study, as in all research, remains a challenge. The patients with AN showed a generally elevated pattern of blood oxygen level–dependent response compared with healthy matched controls. The functional magnetic resonance imaging analysis used absolute prediction error, which does not distinguish positive and negative responses and has been suggested to reflect an attentional or salience signal.9 It may be that the elevated brain response is specific to prediction error, or it may be a brain-wide nonspecific increase associated with attention or engaging in a task. If the reported signal reflects a more general attentional response in AN in the context of receiving sweet liquid, this may still indicate an important mechanism driving behavior but may not be consistent with the proposed model. Using computational methods in the study of eating disorders is critical for a field in dire need of breakthroughs in neuroscience and treatment. Future studies with greater specificity in tasks and analyses will be needed to arbitrate between possible mechanisms underlying aberrant brain responses.
In summary, Frank et al3 put forth an intriguing model of illness in AN, which will allow for new avenues of productive research. The aberrant prediction error–related signals among adolescents and young adults, who are presumably newer to the illness than adults, suggests that starvation affects the dopamine system in patients with AN. This helps to develop the neurobiological understanding of AN, wherein the malnourishment from caloric restriction and dieting behavior lead to neurobiological changes in vulnerable individuals. Across many studies and approaches, abnormalities in dopaminergic frontostriatal systems have been implicated. However, dopamine has multifarious functions with important, distinct roles in learning, motivation, and attention. The role of these systems in illness-relevant behavior merits specific testing. The future of research in AN lies in elucidating which brain systems are critically affected by malnourishment and how these neural effects serve to perpetuate maladaptive eating behavior and persistence of illness.
Corresponding Author: Joanna E. Steinglass, MD, Department of Psychiatry, New York State Psychiatric Institute, 1051 Riverside Dr, Unit 98, New York, NY 10032 (firstname.lastname@example.org).