Predominant Striatal Input to the Lateral Habenula in Macaques Comes from Striosomes

SUMMARY

Striosomes, neurochemically specialized modules in the striatum, are thought to be nodes in circuits extending, via basal ganglia pathways, from mood-related neocortical regions to dopamine-containing neurons of the substantia nigra. Yet striosomes have remained beyond the reach of electrophysiological methods to identify them, especially in non-human primates. Such work is needed for translational as well as for basic science. Here we introduce a method to identify striosomes on-line, in awake, behaving macaques. We combined electrical microstimulation of the striatum with simultaneous electrophysiological recording in the lateral habenula, followed by immunohistochemistry. We demonstrate that striosomes provide the predominant striatal input to the macaque pallido-habenular circuit, known to function in relation to reinforcement signaling. Further, our experiments suggest that striosomes from different striatal regions may convergently influence the lateral habenula. This work now opens the way to defining the functions of striosomes in behaving primates in relation to mood, motivation and action.

eTOC Blurb

Hong et al. introduce a method to identify striosomes in behaving primates by finding striatal sites evoking responses in the lateral habenula, which modulates motivation. This work opens unprecedented opportunities to elucidate striatal functions implicated as vulnerable in neurodegenerative disorders.

INTRODUCTION

Unprecedented progress is being made in the genetic analysis of neuropsychiatric conditions and in identifying affected brain regions, but it remains a formidable challenge to determine how related neural circuits affect these disorders. This gap is in part due to the complexity of interacting networks of circuits modulating mood and affect. It is known, however, that neurons in these circuits modulate the release of dopamine and serotonin, key neurotransmitters implicated in depression, anxiety and related disorders [1–6]. Many of these circuits target the striatum, which is now recognized as a hub in mood regulation as well as in motor control [7–11]. Certain mood-related corticostriatal circuits project preferentially to chemically specialized ‘striosomes’ in the striatum [12, 13], whereas other circuits target the surrounding matrix. These macroscopic zones have distinctive expression patterns of many neuroactive substances, including receptors for dopamine, acetylcholine, opioids and cannabinoids [14–19], and sets of striosomes, distributed in gradients, appear to be related to components of the limbic system. In rodents, such subsets of striosomal neurons have been found to project directly to the dopamine-containing substantia nigra pars compacta (SNc) [20, 21]. Striosomes also project indirectly, via the pallidum, to the lateral habenula (LHb) [22–24], which itself influences the dopaminergic and serotonergic systems of the midbrain, is responsive to motivational value, and can be influenced by the ventral tegmental area [25–29]. Thus striosomes, by at least two descending circuits as examined in rodents, stand in a position to influence major midbrain modulators of mood, motivation and movement. Functional studies of striosomes are difficult and rare, but favor this view, relating striosomal circuits in rodents to reinforcement, motivationally conflicting cost-benefit decision-making, and stress [13, 15, 30–32].

To extend such circuit maps to studies in humans is a critical need in neuroscience and in medicine. Work in non-human primates can help, but achieving delineation of such circuits in primates at resolutions comparable to those in rodents, especially as achieved with genetically based methods, has been extremely difficult for technical reasons. Here, we set out to fill this gap by developing an experimental protocol by which we could identify striosomes in macaques. We did so by microstimulating the striatum of the monkeys while simultaneously recording the spike firing of neurons in the LHb, the target of the striosome-pallido-LHb pathway [33–37].

Our findings demonstrate that a functional connection between the striosomal system and the LHb is present in the non-human primate brain and is the primary source of striatal input to the pallido-habenular circuit; that this striosome-based circuit can both excite and inhibit neurons in the LHb; and that circuits arising from striosomes in widespread regions of the striatum can so function. These findings raise the possibility that the striosomal system in primates can exert a key influence on the balance between positive and negative valence systems.

RESULTS

Microstimulation of the Macaque Striatum Can Evoke LHb Responses

To identify the striatal source of input to the LHb as either striosomes or matrix, or both, we combined striatal electrical microstimulation with LHb multi-unit activity (MUA) recording (Figure 1). In single immunohistochemically stained sections, striosomes have the appearance of rounded or elongated local zones about 500 μm in width that appear separated from each other, but most are interconnected out of the plane of the section in a labyrinthine manner [12, 14, 38, 39]. In three adult rhesus macaque monkeys (monkeys E, F and R), we located the LHb by taking structural magnetic resonance images (MRIs) of each monkey’s brain. Guided by MRI-based coordinates, we sought for units with the neural response characteristics of LHb sites as the monkeys performed a one-direction rewarded (1DR) task [26] (Figure 1A; see STAR Methods). In monkey E, a blunt tipped electrode was located just at the dorsal surface of the posterolateral part of the LHb for MUA recording (Figure S1). This placement resulted in excellent recording stability for the duration of the experiment (Figure 2).

Figure 1. Experimental Protocols.

(A) Schematic diagram illustrating the 1DR behavioral task. After an initial fixation period (dotted circle indicates gaze location), a green or white target appeared to left or right, and the monkey was required to make a saccade to this target. The monkey received juice reward after a saccade to the green target but no reward for a saccade to the white target. Target positions were maintained for 24 trials and then were reversed.

(B and C) Photomicrographs of left LHb of monkey E. Recording sites are marked by deep penetration lesion (arrows) made 10 days before perfusion. Scale bar, 1 mm.

(D and E) LHb MUA of monkey E when a non-rewarding (D) or rewarding (E) target was presented at time 0. Each row in spike raster plots represents one trial. Green curves are spike density functions averaged across trials. See Figure S2 for other monkeys studied.

(F) Microstimulation was applied at 200-μm intervals, 100 times at each depth, along the track in the striatum (left) while MUA was recorded in the LHb (right). Inset: waveform of one stimulation pulse train.

See also Figures S1 and S2.

Figure 2. Neuronal Activity Recorded in Monkey E Performing 1DR Task.

Spike density functions and raster plots (lower and upper portion respectively in each panel) of MUA recorded in the LHb during non-rewarded and rewarded trials of the 1DR task over the course of daily recording to identify the LHb before striatal stimulation. The “n” in each panel denotes the number of trials. Horizontal lines show the mean (dashed black) + 5 SDs (gray) of the values to the left of 0 for each trace. Each hemisphere was recorded separately. Time 0 represents the appearance of the non-rewarding or rewarding target. See Figure S2 for monkeys F and R.

We then began microstimulation within the ipsilateral striatum. We penetrated the striatum with concentric bipolar stimulating electrodes, introduced one per day, while recording MUA in the LHb (Figure 1F). The stimulation was begun at the entrance into the striatum, recognized by the sudden increase in spiking activity as the electrode was lowered. To verify the accuracy of this method of finding the dorsal surface of the striatum, we made a marking lesion at the estimated entrance point in one experiment (Figure S3).

In successive sessions, different anteroposterior levels of the striatum were selected for microstimulation. Microstimulation was applied every 200 μm along the electrode track, 100 times at each depth (see STAR Methods), and the LHb responses were averaged to construct a pseudo-color raster plot (Figure 3D). A response was considered significant if the smoothed MUA changed by at least 1.3 standard deviations (SDs) from the raw baseline MUA (see Figure 6 and STAR methods).

Figure 3. Mapping of Stimulation Response Profiles onto Immunohistochemically Stained Sections.

(A and B) GFAP-stained (A) and Prussian blue-stained (B) sections from right caudate nucleus of monkey F, showing electrode track and marking lesion (bottom arrows). Upper arrows indicate the point at which the electrode entered the striatum. Scale bar, 1 mm. Inset in B shows iron concentration in the region indicated by dotted box (scale bar, 200 μm). See also Figure S3.

(C) KChIP1 immunostained section showing striosomes in bright yellow.

(D) Pseudo-color plot (scale at right) of the baseline-subtracted LHb spike density function recorded at successive depths along the stimulating electrode track, temporally aligned at the stimulation onset (0 ms; vertical black line). The depth (vertical) dimension was scaled and aligned to match the histological image, on which the start and end points of the stimulation track are marked by short white lines. See also Figure S4.

(E) LHb MUA spike density functions as in D, aligned horizontally at the stimulation onset (0 ms) to illustrate the variation of response with stimulation depth. Only the odd numbered stimulation points are shown. The bottom plot shows superimposed LHb responses at all 29 stimulation points.

See also Figures S3 and S4.

Figure 6. Definition and Properties of ESRs.

(A-E) Data analysis pipeline. Left-hand color scale applies to A, right-hand to B-E. (A) Raw MUA was converted to a spike density function, and then to z-scores using the baseline period (218-19 ms before stimulation onset). Each row shows average of 100 stimulation trials for one stimulation depth. (B) Z scores were smoothed in the time dimension. Data from the stimulation artifact period (0–15 ms) were discarded. (C) Smoothed z scores that failed to exceed the response threshold of 1.3 were set to zero, leaving positive and negative candidate responses with non-zero values. The dashed rectangle shows the portion enlarged in panels D and E. (D) Candidate responses that matched in polarity and time at successive stimulation depths were grouped into individual ESRs (white outlines). (E) Individual ESRs that overlapped by at least one stimulation depth were grouped into aggregated ESRs (white outlines).

(F) Distributions of ESR lengths, defined as the number of stimulation depths that belonged to the ESR. Three methods of aggregating individual ESRs are shown: positive ESRs only, negative ESRs only, and aggregation without regard to polarity.

(G) Numbers and proportions of ESRs in each category of spatial relationship to striosomes across all 3 monkeys.

(H) Cumulative distribution of ESR response sizes, defined as the maximum response size at any depth-and-time point in the ESR. Anatomical classification was determined from the stimulation point that elicited the maximum response. See Table S6 for false alarm rates.

(I) Same data as in G, but the ‘miss’, ‘border’, and ‘touch’ classes have been aggregated into a single “not inside striosome” class. The green curve shows the difference between “inside striosome” distribution and the “not inside striosome” distribution.

See also Tables S5 and S6.

We found that the current spread from the stimulating electrode was limited and induced minimal tissue damage, likely due to the small current applied and the concentric bipolar stimulating electrode used. Based on instances such as that shown in Figure 3, we estimated that the effect of the stimulation was detectable for <200 μm. The peak appearing at around 50 ms in Figure 3E corresponds to the red zone at the bottom of the pseudo-color plot in Figure 3D. This response appeared very abruptly at the last stimulation depth, suggesting that the effective zone of current spread was less than 200 μm.

Anatomical Classification of Stimulation Points

We selected eight representative electrode tracks for exhaustive anatomical analysis using the following criteria: histology quality, recording quality, and response values in the group that as a whole were representative of the distribution of values found during preliminary analysis of the entire data set. The distance from every stimulation point to the nearest striosomal boundary was measured (see STAR Methods), without regard to whether or not the stimulation produced an LHb response. For further analysis, each point was then assigned to one of four anatomical classes based on that distance (‘miss’, >100 μm; ‘border’, 50-100 μm; ‘touch’, <50 μm; and ‘inside’, >50 μm inside the boundary). Three of these eight tracks are shown in Figure 4A–4C, together with several examples of the anatomical classes in each track.

Figure 4. Relationship between Anatomically Classified Stimulation Points and Evoked Responses in LHb.

(A-C) Examples of anatomical classification of stimulation points along track 27 in monkey F (A), and tracks 7 (B) and 8 (C) in monkey E. Three points were chosen arbitrarily from each track. White rectangles outline electrode tracks. Each blue line indicates a stimulation point with its anatomical classification. Red curves depict striosome boundaries marked for scoring. Scale bars, 1 mm. See Figures S5 and S6.

(D) Cumulative distributions of maximal smoothed z-scored MUA responses between 15 and 200 ms after stimulation onset for every stimulation point in eight tracks selected for exhaustive analysis (see text). Each curve shows on the y axis the percentage of points within an anatomical class that had maximal responses smaller than the value indicated on the x axis. Vertical lines mark thresholds of, respectively, 1.3 (low) and 2.6 (high). See Tables S1 and S2.

(E) Numbers and proportions of stimulation points classified as having each response level. The red and yellow sectors represent eps.

(F) Most high-level responses were evoked by stimulations inside of striosomes. Histograms show the percentage of all points in each response level group as a function of the distance to the nearest striosome border. The left plot shows points that were outside of striosomes, and the right plot shows points that were inside striosomes, with higher numbers indicating points deeper inside of striosomes. Red numbers show raw counts for the high-level response group (4 inside and 1 at >200 μm outside). See Table S3.

See also Tables S1, S2, S3 and S4, and Figures S4, S5 and S6.

Defining Points in Striatum That Are Functionally Connected with LHb

When we stimulated electrically within the striatum, some stimulation points produced clear responses in the LHb, whereas others did not. Among the points that evoked a clear response, some evoked larger responses than others. To quantify the size of the evoked responses (see STAR Methods), we used the magnitude of the smoothed, z-scored MUA. The LHb response level at each stimulation depth was defined as the maximum response that appeared anywhere within the response period.

Numerical simulations (see STAR Methods and Table S1) showed that with a smoothed z-score response threshold of 1.2, we should expect a false alarm “response” at ~25% of the stimulation points, whereas with a threshold of 1.3, the false alarm rate per depth dropped to ~14%. We therefore set a threshold for “significant responses” at 1.3, and designated each stimulation point at which a response greater than 1.3 was evoked as an effective point (ep). Subsequent analysis suggested that a higher threshold (2.6) would be helpful in identifying striosomes (see below), and the results for this higher threshold are also shown in Table S1.

Most High-Responding eps Are inside Striosomes Identified Immunohistochemically

We used the data from the eight representative electrode tracks to construct cumulative distribution functions of the positive response sizes for each of the anatomical classes. The general distribution of response levels differed substantially between ‘inside’ points and the other three anatomical classes, but the difference appeared only at response levels greater than about 1.2 (Figure 4D). We therefore removed all values less than 1.2, after which a Kruskal-Wallis test returned barely significant p levels for differences between the distributions of the set of ‘inside’ response values compared with ‘miss’ response values (p = 0.047), and ‘inside’ values compared with all other categories aggregated (p = 0.054). The ‘inside’ curve shows another abrupt deviation away from the other curves in the 2.0-2.6 range. We therefore set another threshold at 2.6 to distinguish physiologically “high response” points from “low response” points. The distributions of response values between the ‘miss’, ‘border’, and ‘touch’ categories did not appear different enough to warrant statistical testing.

Non-effective stimulation points were overwhelmingly more common than eps (Figure 4E). Since the proportion of eps that we found (13%) using the threshold of 1.3 was not different from the proportion expected from the null hypothesis simulations, we concluded that the overwhelming majority of eps were false alarms, so an additional criterion was required. Given the size of striosomes, we chose to require more than one ep in succession as the stimulating electrode was advanced in 200 ™m increments. At our threshold for “high response” (2.6), the proportion of eps expected from the null simulations (5 × 10⁻⁶) was very much smaller than the actual proportion found (2%).

When we crossed the four anatomical classes with the three response levels, 80% of the high-response eps were classified as ‘inside’ a striosome, whereas only 11% of all stimulation points tested were ‘inside’ (Table S2). Choosing recording sites based on the size of the LHb response to striatal stimulation could thus provide more than seven times the yield of striosomal recording sites, compared to simply recording blindly.

For certain purposes, such as local field potential or electrical stimulation studies, it might be sufficient to localize a stimulation point to the general vicinity of a striosome (i.e., within 100 μm) rather than requiring it to be strictly inside the striosome. In this connection, it is noteworthy that even low response points provided a larger proportion of striosomal “hits” (58%) than points chosen at random (41%), although this improvement was not significant by Fisher’s exact test (p = 0.13). Furthermore, the Pearson’s linear correlation coefficient between distance to the nearest striosome and response size was −0.2 (p = 0.03), so it seemed likely that useful information could be extracted from lower level responses.

To determine the relationship between response size and anatomy, we constructed histograms of the distance to the nearest striosome for each response level. We found a trend for low-response stimulation points to be closer to striosomes than non-responsive stimulation points (Figure 4F, Table S3). The Kruskal-Wallis test performed on the two distributions confirmed the trend (p = 0.10) but did not detect statistically significant differences.

However, the distance distribution for high-response points was significantly different from the distribution of all points (p = 0.013). This result was confirmed by Fisher’s exact test on the numbers of points strictly ‘inside’ of striosomes versus points that were not, for high-response points compared to all other points (Table S4; p = 0.0005). It should be noted that the five points in the high-response category were not strictly independent, because responses tended to occur at several successive depths. Those five points belonged to three groups of sequential stimulation points from three separate tracks, two of which contained two points each. If we repeat Fisher’s exact test using only one high-response point from each group, the result is still significant (p = 0.03).

An example of a high-response ep that was located inside a striosome is shown in Figure 5. The electrode track can clearly be seen to enter the striosome, with the marker lesion at the final stimulation point still inside the striosome. The final stimulation point on this track evoked a response of 3.0, and the point immediately before it evoked a response of 2.9.

Figure 5. High-Level Responses Coinciding with a Striosome.

(A) Confocal image of hemisection illustrating electrode track (GFAP, green) and striosomes (KChIP1, purple). Scale bar: 5 mm.

(B) Magnified view of electrode track in striatum, ending at a striosome. Scale bar: 1 mm.

(C) Magnified view of the end of track with marker lesion inside the striosome, near the ventral edge. The anatomical classes and response levels of the last five points in the track were, respectively, “miss”/1.41, “miss”/1.16, “touching”/2.18, “inside”/2.87 and “inside”/2.98. Scale bar: 100 μm.

In some sessions, after traversing several millimeters of striatum without finding a significant response, we stopped advancing the electrode and made a marker lesion as soon as a significant response was seen (Figures S4A and S4B). This procedure might have resulted in an overrepresentation of low-response points situated at the border or touching a striosome compared to high-response points inside the striosome. There were some instances in which high-response points were followed by a point that produced no response, in which case we also stopped advancing the electrode and made a marker lesion (Figure S4E).

Definition of Individual and Aggregated ESRs

We frequently saw that stimulation at successive depths evoked similar responses, with a progressive waxing and then waning at each successive depth, suggesting that the stimulating electrode passed near a striosome, with the point of greatest response representing the point of closest approach to the striosome. To test this idea, we grouped eps into effective stimulation regions (ESRs; Figures 6A–6E) when two criteria were met: (1) there were significant responses of the same polarity (positive or negative change in MUA) at two successive stimulation depths at around the same latency, and (2) the series of stimulation depths that meet the first criterion in a pairwise fashion formed a set of stimulation points that were all hypothetically stimulating the same striosome. We defined an “individual ESR” as a set of successive stimulation points that all evoke responses of the same polarity at a similar latency (Figure 6D). When the second criterion is met by a set of individual ESRs, they conjointly form an aggregated ESR (Figure 6E). Individual ESRs can join in either the time or the depth dimension, as long as they overlap by at least one stimulation point.

Individual ESRs could be of either positive or negative polarity. Positive individual ESRs were generally limited to fewer depths than negative individual ESRs, and the overall length of the aggregated ESRs varied correspondingly according to the polarities that we included. Aggregating only positive individual ESRs and ignoring negative ones produced the shortest aggregated ESRs; aggregating only negative ESRs produced aggregates with intermediate lengths; and aggregating both polarities produced the longest aggregates (Figure 6F). Sixty-five percent of positive responses were 2 stimulation points long, whereas only 49% of negative responses and just 47% of mixed responses were of length 2. Also, the 90th percentile of positive response lengths was 4, whereas for negative and mixed responses the 90th percentiles were 5 and 5.8, respectively. There were 72 aggregated ESRs constructed using positive responses only, 73 using negative responses only, and only 50 when aggregation was done including both responses. Consonantly with the ESR length distributions, these frequencies indicate that, if polarities were not required to match, around 20-30 pairs of ESRs of each polarity became merged due to the presence of responses of the opposite polarity at intervening depths.

Our goal was to develop a method for locating striosomes in awake animals, so we chose the aggregation method that produced the shortest aggregated ESRs, and proceeded to analyze positive-only aggregated ESRs. Each aggregated ESR was assigned to one of the four anatomical categories (‘miss’, ‘border’, ‘touch’, and ‘inside’) based on the anatomical class of the stimulation point that elicited the greatest response.

Null hypothesis simulations did not produce a single false alarm ESR at the higher threshold of 2.6 in 500,000 simulated sessions (Table S5). Even at the lower threshold of 1.3, there was only one false alarm ESR out of every ~3 simulated sessions.

Most ESRs Are near Striosomes, and Most High-Responding ESRs Are inside Striosomes

ESRs by definition entailed a response to the electrical stimulation. Since eps were associated with striosomes and constituted only a small fraction of all stimulation points, it was not surprising that a considerably larger proportion of aggregated ESRs were near striosomes (70%; Figure 6G) than of stimulation points overall (41%; Table S2). Likewise, a considerably larger proportion of aggregated ESRs (27%; Figure 6G) than of single stimulation points overall (11%; Table S2) were classified as ‘inside’ striosomes.

We then repeated the cumulative distribution function analysis (Figure 4D) on aggregated ESRs (Figure 6H). The largest difference between the fraction of ESRs that missed a striosome and the fraction that were inside a striosome occurred for response values ranging from 2.58 to 2.68 (Figure 6H). When we combined the data from the ‘miss’, ‘border’, and ‘touch’ classes into a single large “not inside striosome” class (Figure 6I), the largest difference was still in the 2.58 to 2.68 range. This range comfortably encompassed the high threshold that we used in the ep analysis, and we therefore continued to use the high threshold of 2.6 to distinguish high response levels from low levels. We then scored the numbers of ESRs that were above or below that threshold, and inside or not inside striosomes (Table S6). Fifty percent of ESRs that were inside striosomes elicited suprathreshold responses, whereas only 14% of ESRs that were not inside striosomes did. Conversely, 57% of ESRs that elicited suprathreshold responses were inside striosomes, whereas only 17% of subthreshold ESRs were inside a striosome. These differences were statistically significant (p = 0.006, Fisher’s exact test).

Functional Characteristics of Striatal Input to LHb

Preliminary analyses showed no significant correlations between response intensity or latency and anteroposterior (AP) or mediolateral (ML) position within the striatum, (i.e., within the caudate nucleus or the putamen) covering a range of 11 mm in the ML direction and 12 mm in the AP direction (Figures S7A–S7C). This finding was quite surprising, but was not further examined in this study. We focused instead on the dynamics of the responses to the stimulation.

When striatal stimulation evoked a response in the LHb, the response often took the form of alternating patterns of excitation and inhibition, involving as many as five different phases of alternating polarity (Figures 3D, 3E, 7A–7F, and S7D–S7G). Such patterns are a common feature in many parts of the brain, including in the midbrain dopamine-containing substantia nigra-ventral tegmental regions [40], and in the rostromedial tegmental nucleus (RMTg), which modulates them [26]; these patterns may derive in part from the LHb neuron’s intrinsic rebound property [41]. We therefore quantified how widespread such oscillatory patterns were in our data set, and at least initially to characterize their dynamics by computing their autocovariance functions (see STAR Methods). Forty-five of the 60 ESRs examined had valid autocovariance peaks corresponding to transient oscillations, and 15 did not. In 12 of the 45 cases, a covariance peak was chosen at a shorter lag than was selected by the algorithm because the largest peak clearly did not accurately represent the periodicity of the signal (e.g., Figure S7G). When we plotted the distribution of frequencies (Figure 7G), all but one were within the broad range designated as “beta-band” in the literature.

Figure 7. Functional Characteristics of Striatal Input to LHb.

(A) Smoothed z-scored MUA as a function of time (horizontal) and stimulation electrode depth (vertical). Black arrowhead indicates the depth shown in B and C.

(B) Same data from row 29 of A, showing details of MUA temporal pattern.

(C) Autocovariance function showing the greatest maximum after the first minimum, which correctly identified the periodicity of the signal. The green vertical line shows the lag at which the maximum autocovariance occurred.

(D-F) Same as A-C, but for a point on a different track for which the greatest maximum after the first minimum was considered not to represent a true periodicity. See Figure S7 for additional examples. The red vertical line in F shows the lag at which the maximum autocovariance occurred, but this lag was not used in the analysis.

(G) Distribution of periodicities measured at the peak stimulation points in all 60 ESRs except for the 15 that showed no periodicity.

See also Figure S7.

DISCUSSION

Our findings lead to three major conclusions. First, we demonstrate that there is a functional circuit from striosomes to the LHb in the macaque, and that this circuit can serve both to excite and to inhibit spiking of neurons in the LHb. The linkage between striosomes and the LHb, first reported in rodents [22], is thus a characteristic of striatal circuits that is widely conserved. Second, we show that this striosomal circuit arises from far-flung striosomes in both the caudate nucleus and the putamen, suggesting that this circuit could be convergent, with outputs from dispersed parts of the striosomal labyrinth evoking a response in a relatively restricted region of the LHb. Coordinated inputs from functionally distinct parts of the striatum could thus combinatorially signal to the LHb. Finally, our experiments have established for the first time a method by which striosomes in the primate brain can be identified experimentally in vivo, on-line with high probability, opening the way for analysis of the activity patterns of striosomal neurons during multiple behaviors.

Intervention in LHb function by deep brain stimulation has already been introduced for the human brain, and functional MRI findings in human have further registered activity in the habenula in relation to anxiety and depression [29, 42–44]. Much electrophysiological evidence in macaques implicates the LHb in reinforcement learning, motivation, control of action and decision-making through its influence on the dopaminergic and serotonergic systems [29]. It is thus of the greatest interest that such functions have been proposed also for the striosomal system of the striatum [8, 12, 13, 21, 31, 32, 45–50]. Evidence in macaques suggests that at least part of the striosomal system receives direct input from the anteromedial prefrontal and caudal orbitofrontal cortex [12]. Taken together with studies of the pallido-habenular system [29, 35], our findings suggest the possibility that in the primate brain, there could be a master cortico-striosome-pallido-habenular circuit that affects the dopaminergic and serotonergic systems, themselves known to influence mood state, motivation, movement and affective decision-making.

Identification of Striosomes by Simultaneous Microstimulation and Recording

Our key objective was to develop a method to reliably identify striosomes on-line in the behaving monkey. We found that the striatal sites at which electrical stimulation elicited significant modulation in the LHb corresponded to immunohistologically identified striosomes with 70% confidence of accuracy, collectively including a 27% probability of direct hits and a 43% probability of hitting on or just outside the striosome boundary (Figure 6G). Clearly, the method as implemented here was not perfect. There were some matrix sites that elicited an LHb response, and stimulations along electrode tracks occasionally passed through striosomal regions at which the microstimulation did not evoke an LHb response. We could not determine whether these were false or correct negatives. Nevertheless, with careful histological examination to verify recording sites, it should be possible to record from identified striosomes and matrix in behaving non-human primates and thus to establish the response properties of neurons in these compartments. We expect that further refinements of this method (in progress) will yield even higher accuracy.

Striosomal Connection to the Macaque Pallido-Habenular Circuit

Our results build on major preceding work in rodents [35]. In a pioneering study, Rajakumar and colleagues discovered that in the rat, striosomes within the rostral caudoputamen preferentially provide inputs to the LHb via a rostral part of the entopeduncular nucleus (EPN), considered to be the homolog of the internal segment of the globus pallidus (GPi) in primate [22]. Stephenson-Jones et al. [23], in the mouse, reported that the LHb receives disynaptic inputs from the striatum, and that this connection originates from projection neurons in both compartments. Wallace et al. [24] have found, again in mouse, that glutamate/GABA co-releasing somatostatin neurons and glutamatergic parvalbumin neurons in the EPN receive inputs from striosomes and that these, in turn, project to the LHb, in sharp contrast to the Rbp4-Cre-labeled EPN cells that project to ventral anterolateral thalamus and are innervated almost exclusively by matrix.

The neurochemical identity of the LHb-projecting pallidal neurons in the primate is not known. However, it is known that the so-called border neurons encircling the GPi (GPb) are the source of the GPi-LHb projection [33, 51]. The multiphasic responses that we commonly observed, together with the broadly distributed latencies (Figures S7A–S7C), suggest that longer convergent-divergent and/or reverberating pathways probably also play a significant role in the striato-habenular connection. LHb neurons are known to show a sustained post-inhibitory rebound after electrical stimulation of the stria medullaris in vitro [41]. The low frequencies of the oscillatory responses that we saw allow for the possibility that the first cycle of the oscillation is generated by internal dynamics of the LHb, and subsequent cycles by reverberation around basal ganglia loops, possibly including the neocortex.

Striosomes May Be Capable of Exciting or Inhibiting the LHb

Our finding that, when electrically stimulated, some striosomal sites both excite and inhibit the LHb adds to increasing evidence that striosomes have a unique role in basal ganglia function. Employing an antidromic activation technique, Hong and Hikosaka [33] identified the GPb neurons in recordings in macaques and found two groups of task-responsive GPb neurons: one group encoding negative reward prediction errors (RPEs) similar to the responses of LHb neurons, and the other group encoding positive RPEs. In remarkable studies with rats, Shabel et al. [36, 37] demonstrated that transmission from the EPN to the LHb is carried by pallidal fibers that co-express glutamate and GABA, and that the pathway can be excitatory, in contrast to what is believed to be the inhibitory output of the main pallidal segments.

Dual Direct and Indirect Striosomal Inputs to Dopamine Neurons

Strong evidence suggests that striosomes can influence dopamine-containing neurons not only through this pallido-LHb-RMTg-SNc pathway, but also by direct projections to the dopamine-containing neurons of the SNc [20, 21]. It is not known whether the same or different striosomal neurons give rise to these two pathways. Our evidence that the striosome-LHb circuit might be convergent is of particular interest: it has already been hypothesized that this pathway could provide a multi-modal signal to the dopamine system [52]. For the direct striosome-to-SNc circuit, evidence is emerging that striosomes can prominently target discrete clusters of these neurons and their bundled ventrally descending dendrites to form striosome-dendron bouquets and also can target the posterior cell cluster in rodents [20, 21]. There are hints that the striosomal projection has topography within the substantia nigra, but evidence is not complete [21]. The contrast between these two striosomal circuits is particularly intriguing also in that one engages the LHb, in which many neurons are activated by negative RPEs but inhibited by positive RPEs, whereas the other directly engages dopamine-containing neurons, many of which are noted for firing in relation to positive and negative RPEs. This organization raises the possibility that these two striosome-based circuits are, in combination, key to the co-evaluation of positive and negative options.

It is therefore notable that the cortico-striosomal circuit has been shown to affect decision-making under challenging, conflict conditions in which mixed positive and negative options are offered in approach-avoidance paradigms [13, 31, 48]. Decisions involving cost-benefit integration may thus be one key function of these pathways. Learning-related activity patterns have also been observed by in vivo 2-photon imaging [32, 45, 53]. A function for striosomes as the ‘critic’ in actor-critic models [46, 50, 54, 55], or as generating ‘responsibility signals’ acting as value predicting signals in modular reinforcement learning models [47], would be in line with these findings. We were unable here to distinguish by microstimulation the cells of origin of these two striosomal output circuits, but the method that we introduce here should allow direct recording from striosomal neurons as identified by stimulation-recording protocols.

STAR METHODS

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Ann M. Graybiel (graybiel@mit.edu).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Three adult male rhesus monkeys (Macaca mulatta), E, F and R (11, 13 and 14 kg, respectively), obtained from either commercial vendors (monkeys E and F) or the California National Primate Research Center at U.C. Davis (monkey R), were used as subjects in this study. The animal care and experiments were conducted in accordance with the Guide for Care and Use of Laboratory Animals of United States National Research Council. All experimental procedures were approved by the Committee on Animal Care of the Massachusetts Institute of Technology. Monkeys were housed in cages (0.75 m × 1.75 m × 1.75 m), paired with a conspecific whenever possible, or singly. Standard non-human primate biscuits and supplementary foods were provided as required by the guidelines.

Preoperative anesthesia was induced by intramuscular atropine (0.05 mg/kg) and ketamine (10 mg/kg), followed by inhalation of 1-2.5% sevoflurane in 2 liters of O₂ per minute, to implant the headposts and recording chambers described below, secured by ceramic screws and dental acrylic. For all surgeries, the monkeys were maintained on analgesics postoperatively, and prophylactic antibiotics were injected intramuscularly both on the day of surgery and daily thereafter for 1 week.

METHOD DETAILS

Behavioral Task

For each experimental session, we inserted a recording electrode toward the LHb. The final position was determined to be in the LHb based on the neural activity profiles during the performance of the color version of the 1DR task (Figure 1A) reported earlier [26]. A trial started when a small fixation spot appeared at the center of the screen. After the monkey maintained fixation on the spot for 750-1250 ms, the fixation spot disappeared, and a peripheral target appeared on the right or left side, 10° from the fixation spot. The monkey was required to make a quick eye movement (saccade) to the target within 750 ms. Correct saccades were signaled by a tone 200 ms after the saccade. Throughout the experiment, the green target always meant reward (0.3 ml of apple juice), and the white target meant no reward. Within a block of 24 trials, the green target appeared on the same side, and in the next block its side switched to the other side (e.g., from left to right). Even in the unrewarded trials, the monkey had to make a correct saccade; otherwise, the same trial was repeated. In rewarded trials, the juice delivery started and ended simultaneously with a tone stimulus. For monkey R, a classical conditioning task that did not require the eye movement to the target was used. After the eye fixation, the fixation point changed to either blue or red, indicating reward or no reward, respectively. All the task control (e.g., visual display, juice delivery and tone presentation), data acquisition (spike data, eye movement tracking, etc.), spike discrimination, electrical microstimulation, data monitoring and electrode position visualization were handled via software Blip (available at www.robilis.com/blip).

Electrophysiology

One recording chamber was placed over the midline of the occipital cortex, tilted posteriorly by 40°, and was aimed at the LHb. Another chamber for the stimulating electrode was placed over the frontoparietal cortex, tilted laterally by 35°, and was aimed at the GPi. For the stimulation electrode, we used a concentric bipolar electrode (125μm in diameter) with a platinum-iridium wire in its inner core and stainless steel outer tubing (Frederick Haer). The recording electrode for the LHb was a glass-coated tungsten electrode with 60° blunt angle tip (Alpha Omega). Electrodes were advanced by an oil-driven micromanipulator (MO-97A, Narishige). The recording and stimulation sites were determined using a grid system (inserted inside the chamber), which allowed insertion of electrodes at every 1 mm between penetrations. The two lateral grids had 18 (AP) × 18 (ML) holes covering the left and right striatum from the rostral-most caudate (inter-aural 33) to inter-aural 15, and the entire mediolateral span of the striatum [56]. An electrode was introduced into the brain through a stainless steel guide tube, which was inserted into one of the grid holes and then into the brain through the dura. We preferentially recorded MUA in the LHb using low impedance electrodes (about 0.5 MΩ). We chose to monitor MUA rather than single-unit activity because it was more stable across the many recording hours required and avoided damage to the fragile LHb tissue produced by hunting for well-isolated units. MUA offers spatial resolution comparable to that of single-unit recording (hundreds of microns), and can be recorded with a single-channel electrode, unlike singlechannel LFP recordings which suffer from poor spatial resolution due to the possibility of volume conduction. The signal was amplified with a band-pass filter (200-9000 Hz) and sampled at 40 kHz using software Blip.

To record LHb activity, the location of the LHb was identified first by MRI. The electrophysiological features of LHb MUA [57] were also used to locate the LHb using a spike discriminator of Blip. The results of histology for the monkeys (Figures 1B, 1C, S1, S2A and S2D) demonstrated that we were successful in locating the LHb. To determine the influence of the striatum on the LHb, electrical microstimulations were applied to the striatum while recording MUA from the LHb (Figure 1). Hong and Hikosaka [34] reported that electrical stimulations in the striatum of one biphasic 100 μA pulse of 0.2 ms duration did not invoke robust LHb responses, whereas a train of three biphasic pulses at 300 Hz of 133 μA, each lasting 0.2 ms, was sufficient. In the present study, we were able to reduce the current magnitude by half and still observe robust responses. Accordingly, we used a train of three biphasic pulses of 67 μA each lasting 0.2 ms at 300 Hz. The inter-stimulation interval was about 1 s with some variability. The monkeys were kept alert all the time. The stimulation-triggered LHb activity was analyzed online. The electrical stimulations were applied 100 times at every 200 μm along the electrode penetration through the striatum. The MUA recorded at deeper LHb sites in monkeys F and R in response to the appearance of the two targets was similar to that recorded at the surface of the LHb in monkey R (Figures S2B, S2C, S2E, and S2F). Clear identification of typical LHb activity was thus achieved in all three monkeys (Figures 2 and S2).

To find the boundary of the striatum, we lowered the bipolar electrode toward the striatum while monitoring the neural activity recorded on the electrode until we detected typical striatal neuronal activity of busy neuronal spike activity relative to the white matter above the striatum, which is relatively silent. Immediately upon entering the striatum, we switched the bipolar electrode to a stimulating mode and began recording in LHb while delivering electrical microstimulation in the striatum. On one occasion, we made a marking lesion at the entrance of the striatum to verify the accuracy of this method of finding the edge of the striatum (Figure S3); it was accurate. Having this accurate stimulation reference point was very important because the starting point and the end point (with a small marking lesion) of the stimulation track served as the two reference points in histology images to anchor the pseudo-color plot of LHb responses to the penetration track (e.g., Figures 3D and S4; see Data Analysis for further explanation). Experiments were done on both the left and right hemispheres, but only one hemisphere was stimulated and recorded at a time. For monkey R, we stimulated throughout the depth of the striatum, and then made a marking lesion at the appropriate depth if there was an ESR. For monkeys E and F, when a good ESR was found, we made a marking lesion and pulled out the electrode without advancing any further. This way, we could use the end of the physical track of the stimulating electrode as the marking site in addition to the marking lesion made at the end of the track. For a marking lesion, we passed 1 pulse of 10 μA, negative current for 5 s, and waited 15 min before pulling out the electrode.

For monkey E, we skipped the usual survey penetrations for mapping of the LHb. Instead, we found the LHb with the least number of penetrations to minimize possible damage to it. In addition to this, we set the recording electrode (blunt tipped, 60°) at the top portion of t he LHb for recording instead of penetrating the LHb. The raw MUA signal was converted on-line into a spike train using a 4-window discriminator to exclude artifactual signals. However, we did not attempt to isolate single units for recording, which usually requires depth adjustment and therefore may inflict damage to the target tissue. The combined strategy reduced the damage to the LHb of monkey E, and allowed us to record excellent LHb activity until the end of the stimulating experiment (see Figure 2). The excellent recording condition of monkey E also enabled us to record the same LHb site every day while stimulating different parts of the striatum. To access the same site of the LHb for monkey E, we used exactly the same recording parameters for every session: (1) the same grid hole; (2) the same guide tube and its exact position inside of the grid; (3) the same electrode (one glass coated recording electrode was used for the entire experiment for monkey E); (4) the same LHb neural response characteristics; and (5) the same operator (S.H.). To visualize the recording site in the LHb, a marking lesion was made 10 days before we sacrificed monkey E by driving the recording electrode deep inside the left LHb (see Figure S1).

Data Analysis

Putative striosome identification using MUA

At every 200 μm depth along the penetration track, MUA was recorded using a 4-window voltage and timing discriminator, with trigger thresholds adjusted to produce an average trigger rate in a range of roughly 50-100 Hz. The trigger time series was converted to a peri-stimulation spike density function using a 1 ms Gaussian kernel, averaged over a series of 100 stimulations. Spike density functions recorded at various depths were converted to z scores based on a 200 ms baseline period defined as the interval from −220 to −20 ms relative to the stimulation time. For each stimulation depth, the mean and SD of the MUA during the baseline periods for the surrounding 5 depths were used (i.e., including the two depths above and the two depths below). For the first two depths and last two depths, where there were not 5 surrounding depths available, the first and last 5 depths were used, respectively.

Candidate response regions were then grouped into individual (i.e., non-aggregated) ESRs based on their consistency across successive depths (Figures 6A–6E). If two successive depths showed candidate responses of the same polarity (increase or decrease) with maximum magnitude responses within 15 ms of each other, then they were considered to be a “match”, and thus belong to the same ESR. Any candidate responses at a given depth that had no matches at the preceding or following depth were rejected as false alarms. An ESR was thus characterized by the range of depths it occupied as well as by the range of times at each depth. In cases where the time dimensions of two ESRs did not overlap, it was possible for there to be two or more ESRs at different times that overlapped in depth. ESRs of matching polarity were finally aggregated by joining any pairs that overlapped in depth. For comparison, mixed aggregated ESRs were also constructed in which the requirement of matching polarity was dropped. Each final ESR was characterized by its maximum response value and the time and stimulation depth at which the maximum response was evoked. The ESR latencies in Figure S7 represent the time from the onset of stimulation until the earliest point in the ESR, i.e., the time at which the MUA response crossed the threshold of significance in either the positive or negative direction. However, it should be noted that the MUA latencies reported here may not be representative of single-unit latencies.

We repeated the analysis of Figures 4D–4F using negative responses in place of positive responses. The cumulative distribution functions of response sizes again showed that a clear distinction could be made between ‘inside’ points and all other anatomical classes, with a low threshold and high threshold. The histograms of distance to the nearest striosome boundary for the three response levels showed essentially the same result as well. The same four ‘inside’ points that were identified as producing high-level positive responses also produced high-level negative responses, and two “false alarm” high-level negative response points were found outside the striosome boundary, one of which was ‘touching’ and the other was in the 100-150 μm bin. Neither of the “false alarm” points for negative responses was the same point that showed a “false alarm” positive response.

Accuracy of the eps in predicting striosomes

To assess the accuracy of the ep in predicting a striosome (the hit rate), we overlaid the pseudo-color peri-stimulation spike density function and the pseudo-color LHb response plot on top of the images of the KChIP1 histology sections of the striatal penetration sites. This overlaying process was guided by the GFAP to visualize the penetration track and Prussian blue to visualize the marking lesion, together with the onset of spiking activity when the electrode entered the striatum. We were thus able to interpolate all the intervening stimulating sites by anchoring the initial and end points of the pseudo-color peri-stimulation spike histogram and the pseudo-color LHb response plots to the beginning and the end points of the stimulation. To classify the anatomical relationship of each ep to striosomes, we defined a cylindrical current spread zone with a radius of 100 μm along the penetration tracks in the striatal sections (see Figure S5 for details). Then we searched for striosome boundaries within the current spread zone. If a striosome boundary was included, it was classified as a hit; otherwise, a miss. An ESR was defined as a region of two or more consecutive eps that produced responses of similar latency and matching polarity. We then classified the ESRs into 4 categories (Figures 4A–4C): ‘inside’, ‘touch’, ‘border’ or ‘miss’.

To measure the distance between an ep and the nearest striosome, we constructed spherical templates of measurement with radii of 50, 100, 150 and 200 μm, centered along the penetration track (Figure S6). The shortest distance to a striosome from the 5 sections surrounding the ep was selected as the shortest distance for the ep. We measured the distances from all stimulating points without regard to their efficacy along eight representative electrode tracks that were selected on the basis of the following criteria: good histology, good recording, and good fit of response values in the group as a whole to the distribution of values found in the entire data set. We did not examine the histological sections beyond the center section ±2 sections (±100 μm) because of the ambiguity of the alignment of histological sections and their distortion.

The histological situation was complicated for the high-response point that was estimated to be more than 200 μm away from the nearest striosome (Figure 4F). The closest striosome partially encircled the electrode track, and there was also a small blood vessel running through it near the stimulation point. This was thus a rare case in which it was difficult to identify the striosomal boundary at the point of closest approach of the electrode track, and we chose to err on the side of caution by estimating the distance to be more than 200 μm. The correct figure might be as low as 80-100 μm. The blood vessel may also have influenced the flow of stimulation current around and through the striosome. It was thus not extremely surprising that this particular point did not fit the otherwise consistent pattern of the high-response points’ being located inside striosomes.

Autocovariance functions

We used the Matlab “xcov” function with the ‘unbiased’ normalization option. Inconveniently, this option does not normalize to produce unity autocovariance at zero lag, but it offers the great advantage of normalization with respect to the actual number of data points going into the computation at each lag. Unlike the other normalization options, it thus does not impose an artifactual linear taper on the covariance results.

We computed autocovariance functions from the smoothed MUA at the maximal response-eliciting point of each of ESR, found the lag at which the greatest positive maximum occurred, excluding the peak at zero lag, and then converted that lag to an oscillation frequency by taking its inverse. This analysis was done using an in-house automated process, the output of which was inspected visually to eliminate instances in which the automated process was not appropriate (Figures 7C and 7F). In ambiguous cases, the waveforms (Figures 7B and 7E) were also inspected and were considered non-oscillatory if fewer than three clear phases (one and a half cycles) of oscillation could be seen.

Histology

The monkeys were deeply anesthetized with sodium pentobarbital and were perfused with 0.9% saline followed by 4% paraformaldehyde in 0.1 M phosphate buffered saline (PBS). Brains were blocked and stored in 25% glycerol in 0.1% sodium azide (MP Biomedicals, 0210289190) in 0.1 M phosphate buffer (PB) at 4°C until being frozen in dry ice on a sliding microtome and cut into 40-μm coronal sections. Sections were stored in 0.1% sodium azide in 0.1M PB.

GFAP staining typically showed diffuse staining in a roughly 100-200 μm wide region surrounding each electrode track, visible in a sequence of 5-10 sections surrounding the track. The section containing the track itself was usually identifiable clearly on the basis of a dense, tightly confined line of staining in the center of the region of diffuse staining. Prussian blue staining confirmed the precise location of portions of the track, or in some cases the entirety of the track, as well as the marker lesion at the end.

The distance from a given stimulation point to the nearest striosome was first estimated with the aid of a set of templates showing the boundaries at a radius of 100 μm from the electrode track as they would appear in the section containing the track and in the adjacent sections 40 μm and 80 μm away (Figure S5). If the stimulation point was more than about 50 μm interior to the striosome, it was classified as ‘inside’; within about 50 μm of the striosome boundary was ‘touch’, in the range of roughly 50-100 μm ‘border’, and further than that ‘miss’.

For the quantitative point-by-point analysis of the selected eight representative tracks, distances were measured using a template of concentric circles at radii of 50, 100, 150, and 200 μm. Distances of points inside striosomes were measured by centering the template at the actual estimated stimulation location. Distances of points outside striosomes were measured from the closest point on the electrode track that was within 100 μm of the actual estimated stimulation location.

Immunohistochemistry

For immunofluorescence GFAP and KChIP1 double-staining of striatum sections, sections were rinsed 3 times for 2 min in 0.01 M PBS containing 0.2% Triton X-100 (Tx) (Sigma-Aldrich, T8787), and were incubated in tyramide signal amplification (TSA) blocking reagent (PerkinElmer, FP1012) in PBS-Tx (TSA-block) for 1 hr. Then the sections were incubated with primary antibody solutions containing rabbit anti-GFAP [1:500] (DAKO, z0334) and mouse anti-KChIP1 [1:200] (UCDavis /NIH Neuro Mab Facility, #75-003) in TSA-block for 24 hr at 4°C. After primary incubation, the sections were rinsed 3 times for 2 min in PBS-Tx, and then were incubated for 1 hr in the secondary antibody solution containing biotinylated goat anti-mouse [1:400] (Vector laboratories Inc., BA-9200) in TSA-block. After 3 × 2 min rinses in PBS-Tx, they were incubated in TSA-block containing Streptavidin 647 [1:2000] (Life Technologies, A32357) for 1 hr. Then the sections were rinsed 3 times for 2 min in 0.1 M PB, and they were incubated in TSA-block containing goat anti-rabbit Alexa Fluor 488 [1:300] (Life Technologies, A11034) for 1 hr. After 3 × 2 min rinses in PB, the sections were mounted onto glass slides.

After the striatum sections were double-stained with GFAP and KChIP1, they were further stained with Prussian blue (Abcam, iron stain kit ab150674). The slides were hydrated in distilled water for 1 min and were incubated for 3 min in the solution containing equal volumes of potassium ferrocyanide solution and hydrochloric acid solution. Then the slides were rinsed in distilled water for 10 min. After the slides were dried, they were coverslipped with ProLong Gold antifade reagent (Life Technologies, P36930). Sections near the LHb were Nissl-stained (FD NeuroTechnologies, Inc., PS102).

Imaging

The slides were examined microscopically, and then the striatal and LHb regions were imaged with an automated slide scanner (TissueFAXS whole slide scanner, TissueGnostics) fitted with 10X objective. First, GFAP and KChIP1 images were obtained by fluorescent camera with AF488 and AF647 filters, respectively. Then if the section had Prussian blue staining or Nissl staining, the image was obtained by bright-field camera. Imaging parameters such as exposure and thresholds for background were optimized for each slide. The images obtained were viewed and exported with TissueFAXS viewer software (TissueGnostics) and used for further analysis.

Marking Lesions and Track Reconstruction

We found that our initial protocol of marking multiple sites along the electrode tracks by lesions for subsequent analysis of Prussian blue staining was inadequate. For this reason, in later experiments (monkeys E and F), when we found an ESR, we typically made a marking lesion and then retracted the stimulating electrode. This way, we could use the end of the track in combination with the marking lesion as the physical mark of the end point of the stimulation. The finding of an ESR and the decision to make a marking lesion and retract the electrode out were made on-line, based on the on-line recordings from the LHb. The penetration tracks thus tended to stop at an ESR. Although this strategy was successful in finding striosomes and reconstructing the position of the electrode, it also produced a problem in a few instances, in which the electrode track stopped right before entering a striosome. This may have biased the statistics towards more ESRs’ being categorized as being at the border of the striosome. Figures S4A and S4B show examples of such cases. When we occasionally penetrated through a prominent ESR and observed the ESR disappear, we made a marking lesion to determine the whole length of the ESR (Figures S4C–S4E).

QUANTIFICATION AND STATISTICAL ANALYSIS

Initial spot-checking of MUA during the baseline period of several sessions from two different monkeys showed that the data were approximately equally likely to pass or to fail the Lilliefors test for normality. Therefore, to aid in distinguishing physiological responses from fluctuations due to the intrinsic noise of the recording process, we performed numerical simulations of the null hypothesis that LHb MUA is a Poisson process that is not modulated by stimulation. We used two different MUA firing rates for the Poisson process (53 and 94 Hz), corresponding to the 2.5th and 97.5th percentiles of the actual average MUA firing rates calculated on a session-by-session basis. Simulated sessions comprised 100 trials (as in the actual data) at each of 50 stimulation depths (somewhat more than in most sessions, which ranged from 9 to 59 stimulation depths, with a median of 29 and interquartile range of 21 to 41.5).

Statistical analysis results were computed using Matlab and are presented in the main text. The following Matlab functions were used: lillietest (Lilliefors test), kruskalwallis (Kruskal-Wallis test), fishertest (Fisher’s exact test), and corr (Pearson’s linear correlation coefficient).

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit anti-GFAP	DAKO	Z0334
Mouse anti-KChIP1	UCDavis /NIH Neuro Mab Facility	#75-003
biotinylated goat anti-mouse	Vector laboratories Inc.	BA-9200
Streptavidin 647	Life Technologies	A32357
goat anti-rabbit Alexa Fluor 488	Life Technologies	A11034
Bacterial and Virus Strains
Biological Samples
Chemicals, Peptides, and Recombinant Proteins
Critical Commercial Assays
Deposited Data
Experimental Models: Cell Lines
Experimental Models: Organisms/Strains
Rhesus Macaque (Macaca Mulatta)	Primate Center	N/A
Oligonucleotides
Recombinant DNA
Software and Algorithms
MATLAB	MathWorks	https://www.mathworks.com
Blip (behavioral presentation, feedback, data acquisition)	Free software developed by SH	https://www.robilis.com/blip
Eureka (data analysis program)	Developed by SH	Upon request
Other
Stimulation electrodes	FHC	CBAPC113(AG1)
Recording electrodes	Alpha Omega	366-120607-00
Eyelink 1000	SR Research	N/A
MO-97A (Narishige Microdrive)	Narishige	N/A
Lynx-8 (Amplifier)	Neuralynx	N/A
S88X Stimulator	Grass	N/A

HIGHLIGHTS.

• We present a method for on-line identification of striosomes in behaving primates

• Most striatal stimulation sites evoking lateral habenula response are in striosomes

• Stimulation-recording data show that striosomes can both excite and inhibit LHb

• Striosomal outputs from widespread striatal regions provide input to the LHb