Schizophrenia in Translation: The Presence of Absence: Habenular Regulation of Dopamine Neurons and the Encoding of Negative Outcomes

Abstract

Many patients with schizophrenia have pronounced deficits in the use of negative feedback to guide problem solving and learning, as seen on tasks like the Wisconsin Card Sorting Test. There is now a compelling body of evidence from nonhuman primates that suggests transient decreases in dopamine cell activity may reflect the occurrence of unexpected negative outcomes, such as the absence of an expected reward, and, generalizing to the human, the occurrence of negative feedback or the absence of expected reward. We present preliminary evidence that habenula projections to the midbrain are capable of producing a transient, but nearly complete, inhibition of dopamine neurons at a population level similar to that observed in behaving primates following an unexpected negative outcome. Human functional imaging studies offer further evidence that the habenula is activated following receipt of unexpected negative feedback or the absence of expected positive feedback. We present initial evidence that patients with schizophrenia lack appropriate modulation of habenula activity in response to feedback. Collectively, these data suggest that the habenula may play a critical role in mediating the feedback-processing deficits of schizophrenia.

Dopamine (DA)-containing neurons in the ventral midbrain are centrally involved in encoding aspects of rewarding stimuli, although some controversy remains regarding the specific nature of the contribution. Early notions suggesting that DA signals the hedonic experience involved in consummatory behaviors have evolved to emphasize more cognitive and motivational aspects of reward processing. There is broad consensus that fluctuations in DA release play a critical role in the learning of stimuli and responses that are associated with reward receipt or removal.¹ Some views emphasize the role of DA in incentive motivation and in mediating the salience of stimuli associated with reward, deemphasizing its role in actual consummatory experience.² Alternately, a number of computational approaches have focused on the role of transient changes in DA activity as signaling the occurrence of an event that was different than expected.^3–6

Contemporary notions regarding the role of DA in reward processing have been strongly influenced by a series of seminal studies conducted by Wolfram Schultz and colleagues in behaving primates.³ In these experiments, extracellular single unit recording techniques were used to monitor the activity of individual DA neurons in the substantia nigra (SN) and ventral tegmental area (VTA) of conscious, active macaque monkeys.⁷ When subjects were presented with an unexpected appetitive stimulus, approximately 75% of the DA cells sampled exhibited a transient increase in firing rate. Repeatedly pairing the reward with a conditioned stimulus resulted in a shift in the phasic response from the presentation of the reward to the presentation of the stimulus with which it was paired. If, however, a fully predicted reward was withheld, a transient (∼ 200 ms) cessation in spontaneous firing occurred precisely at the time the reward would have been delivered. These results suggest that DA neurons encode an error signal that reports the difference between observed and expected events. Increases in activity code for unexpectedly good outcomes, no change occurs when events are fully predicted, and transient decreases occur when an outcome is worse than expected. These transient increases and decreases in DA cell activity may provide a teaching signal that is broadcast to DA target areas where it serves to promote reinforcement learning.

Most of the theorizing and behavioral research in the field has focused on the significance of phasic increases in DA cell activity and the role they play in learning environmental events and responses that predict reward availability. However, the transient interruption in DA cell activity is likely to be just as important. In essence, such transient decreases are an error signal, an indication that an action or event has not resulted in the expected outcome. Such signals are critical for the “unlearning” of associations and predictions that are no longer effective. It is well established that DA neurons are spontaneously active in vivo and that the irregular single spike activity exhibited by these cells contributes to tonic DA levels in terminal areas.^8–10 Other factors that influence the steady state concentration of DA include the density of membrane transporters capable of removing DA from the extracellular space (eg, DA and norepinephrine transporters),¹¹ their location relative to release sites, and the frequency and synchrony of impulse traffic in the mesotelencephalic pathway.¹² While phasic increases in DA release are associated with alterations in neuronal firing pattern, steady state concentrations of DA are strongly influenced by changes in the proportion of spontaneously active DA neurons.¹³ Empirical¹⁴ and computational data¹² estimate basal DA levels in the rat striatum to be in the neighborhood of 30 nM, levels close to those needed to activate DA receptors in a high affinity state (K_d ∼ 40 nM).¹⁵ Given the dependence of tonic DA on impulse-driven release, it seems reasonable to assume that a brief but temporally coherent interruption (100–200 ms) in the spontaneous firing of DA neurons at a population level would result in a transient drop in steady-state DA concentrations, temporarily removing the normal bias on DA receptors in the high affinity state. Regional variations in the density of the DA transporter and/or the proportion of D1 and D2 receptors in high versus low affinity states would be expected to alter the impact of the cessation in activity on tonic DA levels. Thus, DA targets in brain regions containing comparatively low densities of DA transporters may be less affected by this form of information coding than neurons in brain areas where DA uptake is more efficient.

Phasic changes (increases and decreases) in neuronal activity elicited by salient sensory stimuli are driven by afferent inputs to the DA neurons. Projections from the pedunculopontine nucleus contribute to the phasic increase in DA cell firing elicited by some environmental stimuli,^13,16 although glutamatergic projections from the prefrontal cortex and subthalamic nucleus could also be involved.¹⁷ The origin of the pathway(s) responsible for the transient cessation in DA cell activity observed in response to the unanticipated absence of an expected reward has not been established. However, several recent studies have suggested that the lateral habenula (LH) may play an active role in reward-related modulation of DA cell activity.

The LH is a component of the habenular nuclei, a phylogenetically ancient structure positioned below the pineal gland in the dorsal diencephalon.¹⁸ It is made up of a heterogeneous group of neurons that contain a variety of neurotransmitters and neuromodulators.^19,20 Neural activity in this structure has been implicated in a variety of behaviors (reward, maternal and male social behavior), biological functions (stress and nociception) and psychiatric illnesses (depression and schizophrenia).^21–24 The LH is uniquely positioned to facilitate functional interactions between the limbic and basal ganglia circuits in the forebrain and DA and serotonin-containing neurons in the ventral midbrain. Afferents are conveyed within the stria medularis and arise primarily within the internal segment of the globus pallidus (entopeduncular nucleus in the rat), lateral preoptic area, and anterior part of the lateral hypothalamic region, as well as the nucleus accumbens, lateral septal nucleus, diagonal band, and ventral palladum.²⁵ Efferent projections travel within the fasciculus retroflexus and innervate median and dorsal raphe, as well as the pars compacta of the substantia nigra and ventral tegmental area.²⁶ Biochemical and electrophysiological studies indicate that LH projections to the midbrain are capable of altering the activity of DA neurons at a population level. Lesions of the habenula or blockade of impulse traffic in the fasciculus retroflexus increase DA release in the nucleus accumbens, frontal cortex, olfactory tubercle, and striatum, suggesting that the LH exerts a tonic inhibitory influence on midbrain DA neurons.^27,28 Critically, electrical stimulation of the LH inhibits the spontaneous activity of 80–90% of the DA neurons in the SN and VTA.²⁹ The duration of the cessation in firing, while current dependent, maps very closely to that observed in behaving rats and primates experiencing an unanticipated loss of an expected reward.³⁰ The inhibition appears to be mediated indirectly via excitation of GABA-containing neurons as local application of the GABA_A-receptor antagonist, bicuculline, blocks the inhibitory effects of LH stimulation on DA neurons.³⁰ Recent anatomical studies have identified a glutamatergic projection from the LH to the VTA where most of the fibers appear to make asymmetric synaptic contact with GABA-containing neurons.³¹ These results are consistent with electrophysiological data showing that stimulation of the LH excites a subpopulation of nondopaminergic neurons in the VTA and SN.³⁰ These experiments suggest that the LH, via its excitatory inputs to GABA-containing neurons, exerts a powerful inhibitory influence on the activity of midbrain DA neurons.

In light of these electrophysiological and anatomical studies, neuroscientists have begun to focus on the physiological consequences of judgments. Midbrain DA neurons seem to play an important role in mediating a person's response to feedback associated with correct and incorrect judgments. Connections between the habenula and the midbrain were a primary topic in a recent functional magnetic resonance imaging (fMRI) study conducted by Ullsperger and von Cramon.³² This functional neuroimaging investigation used healthy volunteers to assess the neural representation of informative feedback, both positive and negative. By using a difficult perceptual judgment task that elicited a high error rate, the investigators were able to demonstrate a strong interaction between the accuracy of the subject's choice and the type of feedback he or she received. Errors with negative feedback prompted significantly greater hemodynamic signals within the habenula than errors without feedback. Interestingly, correct choices without feedback also showed a tendency to promote greater habenular activity than correct choices with feedback. Taken together, these data suggest that the habenula provides a crucial signal to the midbrain in association with negative feedback if the subject is genuinely unaware of whether a correct choice has been made. This interplay between error likelihood and feedback is likely to play a particularly important role in learning tasks associated with high levels of ambiguity.

Preliminary results from our own fMRI studies³³ have been guided by behavioral and neural changes associated with a high-error (65% accuracy), visual-spatial, match-to-sample task. Studies with healthy volunteers and clinically stable outpatients diagnosed with schizophrenia show that patients do not learn as readily as comparison volunteers. Neither patients nor controls showed improvement in task performance during the first test block when informative feedback was withheld. However, during the second phase of testing, when positive and negative feedback was provided, normal volunteers exhibited a significant improvement in accuracy (average increase = 7%). By contrast, patients did not improve during the feedback phase of the protocol. In association with this improved accuracy, manifest during the second phase of the session, healthy subjects displayed unique neural activity patterns. In error trials where informative (negative) feedback was provided, a robust and concurrent activation occurred in the habenula and the midbrain (Figure 1C). By contrast, in correct trials where positive feedback was provided, a robust activation occurred in the caudate nucleus (Figure 1A). Schizophrenic subjects did not exhibit suprathreshold clusters in the midbrain, habenula, or caudate nucleus (Figure 1B and D).

Fig. 1.

Informative Feedback Differentially Modulates Neural Activity in Normal Volunteers and Schizophrenic Patients. Panels A and C represent fMRI data obtained from 5 healthy volunteers (relevant regions are circled). Panels B and D are derived from fMRI data taken from 5 medicated schizophrenic volunteers (relevant regions have arrows). Fixed effect analyses were applied to echoplanar, event-related fMRI data acquired from a 3 Tesla Philips magnet that employed Sensor coil. Panels A and B illustrate Blood Oxygen Level Dependent (BOLD) signals that covaried with correct trials. Positive feedback was provided. Panels C and D illustrate BOLD signals that covaried with error trials. Negative feedback was provided. Healthy volunteers, Panel A, show that the caudate (right image, circled in white) is active during correct trials associated with positive feedback; this pattern is absent in schizophrenic volunteers (see Panel B). Healthy volunteers (Panel C) exhibit marked midbrain (left image) and habenula (right image) activity during error trials associated with negative feedback. Note that this pattern is also absent in schizophrenic volunteers (see respective images in Panel D).

We interpret this pattern in terms of a corrective feedback circuit. The habenula appears to support the neural traffic necessary for an adaptive response to error. Acting synergistically with the habenula, the midbrain's substantia nigra compacta is inhibited by habenular efferents. This inhibitory signal appears to be mediated by GABA neurons in the SN and VTA. The habenula and nondopaminergic neurons in the midbrain suppress DA neuron firing and generate a representation of “error” through this inhibition.

Correct trials are associated with positive feedback. These positive cues cause an increase in DA cell firing, which, in turn, appears to enhance the cortical signal to the caudate. This modulation may be mediated through DA input to the ventrolateral or dorsolateral frontal cortex. Dopaminergic input to the caudate is not sufficient to generate the large Blood Oxygen Level Dependent (BOLD) response associated with positive feedback during correct trials. But the DA signal to the accumbens and frontal cortex could easily modify the large projections from the cortex to the caudate.

In summary, there is converging evidence that the LH plays a critical role in modulation of DA neuronal activity and is well positioned to support transient inactivations of the DA system in response to unexpected negative outcomes. This system may be a critical player in the larger neuronal networks that mediate reinforcement learning and error processing. Dysfunction in this system could diminish a person's ability to learn from errors. Importantly, this is one of the most characteristic cognitive deficits associated with schizophrenia.