We have shown that LHb-projecting GPi neurons changed their activity differentially depending on the expected reward. Their activity was sometimes differential for the direction of the saccade target, but such changes were smaller than their pronounced reward-related activity. The LHb-projecting GPi neurons were different from other GPi neurons in that their firing rates were lower, their spike durations were longer, and they were located peripherally in the GPi. The recording sites were exactly as expected from a series of anatomical studies on the monkey conducted by Parent and his colleagues (Parent et al., 2001
). These results suggest that the LHb-projecting neurons are a distinct group of GPi neurons, as suggested anatomically (Parent et al., 2001
These properties of the LHb-projecting GPi neurons appear similar to ‘border cells’ described by DeLong and his colleagues (DeLong, 1971
; Richardson and DeLong, 1991
), so called because they were found at the border of the GP. The border cells tended to be activated by the delivery of rewards (as we saw in the LHb-projecting GPi neurons) or aversive airpuffs. However, these authors interpreted the border cells as part of the nucleus basalis of Meynert which projects to the cerebral cortex (Richardson and DeLong, 1991
). The border cells were examined further in subsequent studies especially in relation to dopaminergic denervation (Tremblay et al., 1989
). While the relationship of the border cells with the LHb is unknown, we speculate that at least a subset of these neurons correspond to the LHb-projecting GPi neurons examined here.
Another line of evidence for the reward processing in the GP came from a series of studies showing that some GPi neurons are sensitive to glucose and change their activity in relation to consummatory behavior (Karadi et al., 1995
). Such glucose-sensitive neurons were found in the ventromedial and rostral part of the GPi in the rat and monkey, which roughly matches the location of LHb-projecting neurons.
According to Parent (2001)
, LHb-projecting neurons constitute about 10 % of the total number of neurons in the GPi. Would such a minority of GPi neurons play a significant role in controlling behavior? There are at least two lines of research that may support this view. First, human patients with pallidal lesions often show non-motor symptoms, although the most common symptoms are movement disorders. Even when the patients have no obvious sensorimotor symptoms, they may show a lack of will, motivation, and desire (Miao et al., 2001
; Miller et al., 2006
), or show psychiatric symptoms resembling depression, schizophrenia, and obsessive-compulsive disorder (Laplane et al., 1989
). Second, human brain imaging studies have shown that the GPi is related to reward processing. They report that the ventral striatum (especially the nucleus accumbens) and the dorsal striatum (Knutson and Cooper, 2005
) are commonly activated by expected rewards. In many cases, however, the GPi was also activated (Calder et al., 2007
; Pessiglione et al., 2007
), although in some cases it was not stated so explicitly (Kampe et al., 2001
; Kim et al., 2006
; Tanaka et al., 2007
; Tobler et al., 2007
). These motivation- or reward-related observations on the human GP might reflect changes in the state of the LHb-projecting GPi neurons.
If this is true, how might the LHb-projecting GPi neurons participate in controlling behaviors? We found that there are at least two types of neurons in relation to reward: one negative type (excited by no-reward-indicating targets and inhibited by reward-indicating targets) and the other, positive type (excited by reward-indicating targets and inhibited by no-reward-indicating targets). The response pattern of the negative type GPi neurons was similar to that of the LHb neurons (), and more critically, the responses to the target started earlier in the negative type GPi neurons than in the LHb neurons. The similarity between the two groups of neurons extends to the responses to the reward prediction error: they were excited by the unexpected absence of reward (i.e., on the first trial after the position-reward contingency) but not by the expected absence of reward, and were inhibited by the unexpected presence of reward. These data, showing that the two groups of neurons having similar phase of activation even at the first trial of a block, are consistent with the idea that the negative type GPi neurons have excitatory connections to LHb neurons. In contrast, the positive type GPi neurons started showing reward sensitivity later than that of LHb neurons and did not clearly encode reward prediction error. They are thus unlikely to initiate the reward-related responses in LHb neurons, but could contribute to the later part of the LHb responses.
The suggested excitatory GPi-LHb connection in this study is somewhat puzzling, because a general consensus seems to be that this connection is GABAergic and inhibitory, in the same way as the other GPi efferents are (Vincent et al., 1982
). One possibility is that the excitatory connection is mediated by acetylcholine, as the LHb receives input from the cholinergic nucleus basalis of Meynert and ventro-lateral septum (Herkenham and Nauta, 1977
). This is consistent with the finding that some of the LHb projecting neurons in the rostral part of the entopeduncular nuclues (the GPi counter part of the rat) are cholinergic (Moriizumi and Hattori, 1992
). Another candidate is glutamate, as a relatively high level of AMPA receptor subtypes was found in the LHb (e.g. Petralia and Wenthold, 1992
). It is also possible that the excitatory effect is due to disinhibition such that intra-LHb interneurons, which exert tonic inhibition on LHb neurons, are inhibited by the input from GPi neurons. While the current evidence indicates a possible excitatory connection between the GPi and the LHb, the exact nature of this pathway needs further investigation.
While the negative type GPi neurons had a similar response pattern to that of LHb neurons, a closer examination revealed that many of them were modulated also by the direction of the target () unlike LHb neurons (multiunit activity in , single unit activity in Supplementary Fig. S5
). This suggests that sensorimotor signals, which the GPi neurons have, are removed, and instead reward-related signals are extracted presumably by some LHb local connections. We think that these reward-related signals may originate from the dorsal striatum (caudate and putamen), for the following reasons. First, Tremblay (Tremblay and Filion, 1989
) showed that border cells in the monkey GP, which might correspond to the LHb-projecting GPi neurons (see above), were strongly excited or inhibited by electrical stimulation in the caudate and putamen. Second, Saleem et al. found a strong projection from the monkey striatum to the LHb, probably via the GPi, after injecting an MRI visible anterograde transsynaptic transport agent manganese in the monkey caudate and putamen (Saleem et al., 2002
However, it is unlikely that the dorsal striatum is the only source of inputs to the LHb-projecting GPi neurons. The ventral striatum, including the nucleus accumbens and the ventral putamen, projects to the GPi, specifically to its peripheral regions (Haber et al., 1990
). Also the dopaminergic innervation of the monkey GP is conspicuously high in the peri-GPi region (Lavoie et al., 1989
). These ventral striatal and dopaminergic projections match the anatomical locations of the LHb-projecting neurons. Thus, the LHb-projecting GPi neurons may integrate a number of signals ranging from motivation (via the ventral striatum), reinforcement (via dopamine neurons), and the reward value of a target in a motor context (via the caudate and putamen).
In conclusion, our data suggest that the GPi has two functionally distinct outputs, one involved in motor execution and the other involved in reward evaluation (). The motor execution pathway consists of Striatum→GPi→Thalamus/Brainstem connections. The reward evaluation pathway consists of Striatum→GPi→LHb→Dopamine→Striatum connections. Along this extra-basal ganglia pathway, sensorimotor information is removed and reward information is extracted at the GPi → LHb level. This reward evaluation signal is then used to reinforce/discourage the ongoing action via the dopamine projections to the striatum. The same signal may also be used to control mood and social behaviors via the projections of the LHb to the dorsal and median raphe nuclei which contain serotonin neurons (Kalen et al., 1989
Figure 8 Circuit diagram showing mutual relationship between the lateral habenula (LHb) and the basal ganglia. The GPi has two functionally distinct outputs, one for motor execution (via the motor thalamus or brainstem nuclei) and the other for reward evaluation (more ...)