Changes in striatal procedural memory coding correlate with learning deficits in a mouse model of Huntington disease

Abstract

In hereditary neurodegenerative Huntington disease (HD), early cognitive impairments before motor deficits have been hypothesized to result from dysfunction in the striatum and cortex before degeneration. To test this hypothesis, we examined the firing properties of single cells and local field activity in the striatum and cortex of pre–motor-symptomatic R6/1 transgenic mice while they were engaged in a procedural learning task, the performance on which typically depends on the integrity of striatum and basal ganglia. Here, we report that a dramatically diminished recruitment of the vulnerable striatal projection cells, but not local interneurons, of R6/1 mice in coding for the task, compared with WT littermates, is associated with severe deficits in procedural learning. In addition, both the striatum and cortex in these mice showed a unique oscillation at high γ-frequency. These data provide crucial information on the in vivo cellular processes in the corticostriatal pathway through which the HD mutation exerts its effects on cognitive abilities in early HD.

Huntington disease (HD) is a progressive and inherited neurodegenerative disorder associated with selective degeneration of medium-sized spiny projection neurons in the striatum (1). Although the disease is well characterized by symptoms such as involuntary choreiform movements, dystonias, and rigidity, these motor symptoms have been found to be preceded by personality, mood, and cognitive disturbances (2). Some postmortem studies have also shown only limited signs of pathologic processes or cell loss in the brain despite substantial clinical evidence of HD (3), suggesting that neuronal—and, more precisely, synaptic—dysfunction, rather than cell death, may predominantly underlie early behavioral manifestations of the HD mutation.

Studies of the effects of the mutant HD gene in several transgenic mouse models, including the R6 lines, have indeed shown changes in the membrane properties, biochemistry, and morphology of fragile striatal cells (4, 5), suggesting compromised functional integrity. They have also shown early and progressive alterations in synaptic plasticity in the corticostriatal pathway (6). These changes have been hypothesized to underlie early cognitive and behavioral deficits in transgenic mice (and patients with HD) by altering striatal information processing and transfer within basal ganglia loops.

To verify this hypothesis, we recorded single-unit activity and local field potentials (LFPs) in the dorsal striatum and cortex, whereas behaviorally naive pre–motor-symptomatic R6/1 mice (7) (14–18 wk old) and age-matched WT littermates acquired an association between a nose-poke (NP) action and reward through operant learning. This procedural learning of an action–reward association is known to critically involve the striatum, a crucial site for the integration of the sensorimotor, limbic, and cognitive information required for the selection of appropriate actions during task learning (8, 9). Because striatal activity patterns have also been shown to change during procedural learning and habit memory formation (10, 11), we recorded activity throughout task learning. As a result of impaired corticostriatal plasticity, activity in the transgenic mice might be expected not to change.

We chose to study the R6/1 line instead of the more widely studied R6/2 line because of the former's delayed onset of neurological symptoms (i.e., clasping) and milder cellular/molecular phenotypes with prolonged lifespan (7–9 mo), without striatal cell loss even at death (6, 12, 13). This progressive and sequential onset of phenotypic events is well adapted for modeling the subtle early changes in neural processing and cognition in HD.

Here, we report remarkable and differential changes in firing properties during procedural learning and task performance of principal projection neurons of the striatum in R6/1 mice compared with WT littermates, associated with dramatic learning deficits. In addition, the striatum and cortex in the transgenic mice showed a unique and excessive oscillation at the high γ-frequency visible at both single-cell and LFP levels. These data provide crucial information not only on the in vivo cellular processes in the striatum and cortex through which the HD mutation exerts its effects on behavioral and cognitive abilities in early HD, but on fundamental aspects of corticostriatal activity that may be central to normal procedural learning and habit memory formation.

Results

Striatal Projection Cells Were Scarcely Encountered During Procedural Learning in R6/1 Mice.

To verify at first the selective vulnerability of striatal projection cells in HD, we classified recorded units into putative medium-spiny projection neuron (MSN) or putative fast-spiking local interneuron (IN) subtypes according to the known waveform parameter [spike width (14); Fig. 1A]. Only 15.4% of the recorded neurons (38 of 246 classified units; Experimental Procedures) were MSN in R6/1, as opposed to 47.0% (108 of 230 classified units) in WT mice. Consequently, 84.2% (208 units) in R6/1 mice, and 53.0% (122 units) of recorded cells in WT mice were narrow-spiking INs [χ²₍₁₎ = 28.12; P < 0.0001; Fig. 1B].

Fig. 1.

Diminished recruitment of putative MSNs recorded in R6/1 mice during behavioral task. (A) Typical waveform of MSN and putative fast spiking INs. Spike peak-to-valley width (d) was used to classify cell types. (B) Frequency histograms of spike widths (d in A) for all cells recorded in WT and R6/1 mice. (C) Mean (±SEM) numbers of INs and MSNs recorded per probe (tetrode) throughout recording sessions in two groups. (****Significant difference between the two genotypes at P < 0.0001.) (D) Schematic representation of tetrode placements in the striatum for all mice of both genotypes (WT, left; R6/1, right). Tetrode positions from different anteroposterior axis were stacked onto a unique coronal plan at 0.7 mm anterior to Bregma. (E and F) Photographs of a mouse carrying electrode implant and head stage during recording and the operant chamber used during recording.

Previous studies have suggested heterogeneous density for IN along the dorsoventral axis of the striatum (15). To rule out the possibility that subtle differences in electrode locations between the two groups (Fig. 1D) might have influenced these proportions, we compared the average numbers of MSN and IN recorded by tetrodes throughout the entire recording sessions (Fig. 1C). Whereas the numbers of INs in the two groups were equivalent [R6/1, 1.80 units; WT, 1.54 units; t₍₁₉₃₎ = 1.27, not significant (NS)], the number of MSNs recorded from probes of R6/1 mice was significantly lower than in WT mice [R6/1, 0.22 units; WT, 1.37 units; t₍₁₉₃₎ = 6.45; P < 0.0001]. Additional comparison between dorsal (< 3 mm depth from skull) and ventral (> 3 mm) tetrodes in only WT mice revealed no difference in the distributions of IN and MSN (Student t test, P value NS for both cell types; Fig. S1A). Finally, comparison between the two genotypes of only dorsal tetrodes (Fig. S1A) confirmed the previous conclusion of a significant decrease in the number of MSNs, but not that of INs, in R6/1 mice [t₍₁₄₄₎ = 1.1 (NS) for IN and t₍₄₇₎ = 1.5 (P < 0.0001) for MSN].

A previous in vitro intracellular recording of R6/2 mice demonstrated the intactness of the waveform parameter (16) used here to classify cell types, excluding the possibility that the HD mutation modified the spike waveform parameter of MSN and led us to erroneously classify them as fast-spiking INs. Because no cell loss has been reported in this line (12, 13), our data suggest that a substantial proportion of MSN in R6/1 mice were inactive during the task.

Active Striatal Narrow Spiking INs in R6/1 Mice Oscillate at High γ-Frequency.

Although global firing rates of the recorded MSN and IN were identical in both genotypes (Table S1), their autocorrelograms revealed differences in firing properties between genotypes (Fig. 2A). We found that only a small proportion (2.42%, n = 6 cells) of striatal cells in WT mice showed oscillatory activity, with a majority (n = 5) at low frequencies (< 10 Hz). In R6/1 mice, in contrast, 53 of the recorded neurons (21.46%) oscillated, the majority (n = 48 cells) at high frequency (60–80 Hz)—hereafter referred to as high γ-frequency—with a mean value of 69.9 Hz ± 0.62 (Fig. 2B). These proportions were significantly different between the two groups [χ²₍₁₎ = 47.35; P < 0.0001]. Among the 48 fast oscillatory cells recorded from seven different R6/1 mice, 46 cells (96% of oscillatory cells) were IN (representing 20.28% of total INs) and only two cells were MSNs (4%, representing 2.63% of total MSNs).

Fig. 2.

High γ-oscillation in the striatum in R6/1. (A) Proportions of cells oscillating at low (<10 Hz, slow) and high (50-80 Hz, fast) frequencies in both genotypes. (B) Relative frequency distribution of γ-oscillations frequencies for all fast oscillatory cells in R6/1 mice. (C) Proportions of pairs of neurons with slow (<10 Hz) and fast (50–80 Hz) oscillations in both genotypes. (D) Relative frequency distribution of oscillatory frequencies among fast oscillatory pairs of cells in R6/1 mice. (E and F) Power spectra of session-wide striatal LFP by FFT averaged for each genotype and training stage (early, E; late, F). (G) Typical example of a coherent IN at a frequency around 70 Hz. Coherence value is display by the black line (left axis) and the phase by the red line (right axis). (H) Percentage of unit-to-LFP coherent INs and MSNs in both genotypes. (I and J) Coherence frequency distribution and phase relationship with high γ-cycle for coherent cells in R6/1 mice.

Entrainment of Striatal INs to LFP High γ-Oscillation in R6/1 Mice.

To explore whether oscillatory single cells contribute to the striatal LFP oscillation (recorded from the same tetrodes) at γ frequency, we first examined power spectra by using fast Fourier transforms (FFTs; Fig. 2 E and F). Averaged FFT for R6/1 mice showed a common enrichment of 40 to 80 Hz during both early and late stages of training, i.e., at the frequencies also seen in unit activity in these mice. The LFP of R6/1 mice also showed increased enrichment for 15 to 25 Hz at a late stage of training, suggesting that learning-related (i.e., behavioral) changes in trained R6/1 mice may contribute to the enrichment of this specific frequency. Contrary to R6/1 LFP, WT LFP showed only a single band of 15 to 25 Hz throughout the two training stages. Second, striatal unit–LFP coherence analysis confirmed the presence of high γ-oscillation in the striatum of R6/1 mice (Fig. 2 G–J). Substantial proportions of INs (52.05% of all INs) and MSNs (29.41% of all MSNs) in R6/1 mice showed strong coherence with phase relationship (mostly between 0 and π) at the γ frequency (68.9 Hz ± 0.6), suggesting the entrainment of striatal cells to high γ synchrony in R6/1 mice.

We then verified whether this oscillatory single-cell activity is generated locally or reflects the activity of a more extended basal ganglia network by examining cross-correlograms (CCs) between pairs of simultaneously recorded cells from both the same probe and the different probes positioned in two hemispheres (Fig. 2 C and D). This global analysis revealed that 14% of CCs showed γ synchrony with a mean frequency of 70 Hz (± 0.5). When the two types of CCs in R6/1 mice were considered separately, 34 (13%) of 259 CCs from the same hemisphere and 18 (17.9%) of 101 CCs from different hemispheres showed synchrony, with respective frequencies of 66.0 Hz ±2.73 and 70.67 Hz ± 1.30 (Fig. S2 A and B). These data suggest a common origin of γ-oscillation beyond the striatum. Among these high oscillatory CCs from the same and different hemispheres, 67% were IN–IN pairs and 29% were IN–MSN pairs. Of 632 CCs calculated in WT mice, only nine CCs (1.4%) revealed rhythmic synchrony, mainly slow. The distributions of oscillatory CCs in both genotypes were significantly different [χ₍₁₎ = 65.14; P < 0.001].

Task-Related Activity of MSNs and INs and Impaired Procedural Learning in R6/1 Mice.

We then analyzed whether the task responsiveness of activated striatal neurons of R6/1 mice was different from that of WT mice (Fig. 3). Throughout different stages of learning, striatal units responded mainly by activation rather than inhibition in both genotypes and cell types (Fig. 3 A–C). The increased activity was found more frequently during an animal's approach to the reward port [i.e., lick (L)] or NP that involved both locomotive activity than during immobile reward consumption [between L and exit (EX)], and for both genotypes [χ²₍₁₎ = 15.7; P < 0.01 for both genotypes combined; Fig. 3C].

Fig. 3.

Task-related firings of striatal neurons and procedural learning ability. Examples of task-responsive activity of an MSN for WT (A) and R6/1 (B) mice around three periods delimited by NP, L, and EX events: period I (between NP and L, green) contains animal's approach to reward following NP action, period II (between L and EX, orange) corresponds to reward consumption, and period III (between EX and NP, blue) is characterized by the initiation of a new trial and NP. Period I is represented to illustrate repeated and circular nature of the events. Thick red bars symbolize periods during which firing rates increased significantly. (C) Proportion of striatal cells (both types combined) presenting inhibition or activation during the three behavioral periods in both groups. (D) Typical acquisition curves for two WT and R6/1 mice. (E) Examples of performance progression within and between sessions over three consecutive days in a representative WT and R6/1 mice. On day n, both mice at midstage of learning performed similarly (70 responses, seventh session for WT, and 77 responses, 32nd session for R6/1 mice), but R6/1 mouse showed slower improvement between sessions than WT mouse did. Graph displays the mean response rate/min averaged every 5 min. (F) The mean cumulative hours required for WT and R6/1 mice to learn the NP–reward association (***P < 0.001). (G) Asymptotic performance levels reached by both genotypes (*P < 0.01). (H) Average running speed throughout trainings. (I) Changes in the proportions of task-responsive MSN and IN across three or four performance levels in both groups of mice. Because R6/1 mice rarely reached more than 200 responses per session, only three levels were analyzed for them. (J) The proportions of MSNs and INs showing the task responsiveness throughout the three (in R6/1 mice) or four (in WT mice) different stages of training shown in I.

At the behavioral level, however, R6/1 mice required much more training time [t₍₁₇₎ = 5.52; P < 0.0001; Fig. 3 D and F] to reach asymptotic performance levels which, in addition, were lower than in WT mice [t₍₁₇₎ = 3.3; P < 0.05; Fig. 3G]. The diminished capacity of R6/1 mice seems to have been caused not by motor deficiencies (Fig. 3H), nor by decreased motivation for milk reward (Fig. S3C), but by an impaired ability to retain learned information over sessions/days (Fig. 3E and Fig. S3). More precisely, we found that, even at late stages of learning, the operant response rates of R6/1 mice tended to progressively increase within each 30-min session, as if they were reacquiring the task. This slow short-term progression was accompanied by poor long-term retention between training sessions.

To explain the learning impairments, we examined whether the proportions of task-sensitive neurons in R6/1 mice remained unchanged with learning as a result of impaired striatal plasticity. In view of the substantial difference in the learning speed, the different learning stages were made equivalent in performance terms (rewards/session) for both groups (Fig. 3I and Fig. S3A) for comparison. In WT mice, the percentage of task-responsive MSNs increased significantly with learning [χ²₍₃₎ = 48.52; P < 0.01], as did that of task-responsive INs [χ²₍₃₎ = 31.32; P < 0.01], consistent with previous reports (10, 17). Surprisingly, in R6/1 mice, these proportions also increased with learning for both MSNs [χ²₍₃₎ = 63.15; P < 0.01] and INs [χ²₍₃₎ = 8.38; P = 0.015]. Therefore, the diminished number of recruited MSNs in R6/1 mice (Fig. 1B), but not the proportion of task-sensitive neurons (Fig. 3J) was tightly correlated with their retarded learning abilities.

Reward Consumption/Immobility Is Associated with High γ-Oscillation in R6/1 Mice.

To further explore the relationship between the γ-synchrony in striatal LFP and behavior, we generated event-triggered spectrograms (± 2.75 s) for reward event and reward zone exit (Fig. 4 A–D). R6/1 LFP showed high γ-enrichment (70–80 Hz) throughout reward consumption (red arrows in Fig. 4 B and D), a spectrum absent in WT mice (Fig. 4 A and B). However, oscillations at 15 to 25 Hz, referred to as β-band (dark arrows in Fig. 4 A–D), were commonly present in both genotypes throughout motor sequences, seemingly in antagonism with high γ-synchrony. Surprisingly, increased β-activity was closely associated in both genotypes with rapid decline of movement speed (white line drawings in Fig. 4 A–D) before reaching the food well (dark arrows in Fig. 4 A, B, and D) and to a lesser degree with the initiation or acceleration of movement (Fig. 4 A, B, and D). Furthermore, we also found that increased γ-power (blue dots) appeared mostly, but not exclusively, when the mouse approached the food well in the operant chamber and consumed the reward (Fig. 4E), in the absence of the β-band (Fig. 4F), supporting its relationship with immobility.

Fig. 4.

PETHs revealing high γ- and β-oscillations in R6/1 mice. (A–D) PETHs around (±2.75 s) the L (A and B) and EX (C and D) for representative WT (A and C) and R6/1 mice (B and D). White lines in A–D represent the average moving speed of the animal. Red arrows represent γ-band, and black arrows and dotted lines represent β-band. (E) Spatial positions of a R6/1 mouse in the operant chamber when his striatal LFP expressed high γ-oscillation (blue dots, green dots show visited pixels). Red circle indicates the location of food well. (F) Spectrogram for 8 s showing the variation of the power densities for low (<30 Hz) and high (70–80 Hz) frequency bands. Dark horizontal bars indicate periods when the mouse consumed reward.

Cortical γ-Oscillation in R6/1 Mice.

Finally, to examine at the presynaptic cortical level the activity changes or the presence of the high γ-oscillation, we recorded from the motor cortex in five WT and eight R6/1 mice submitted to the same behavioral task (Fig. 5). We found increased discharge rates of cortical units in R6/1 mice (Table S2). Therefore, inactive striatal MSNs have not been caused by the diminished cortical excitatory input. Cortical LFPs also confirmed the existence of both high γ- and β-bands as in the striatum of the transgenic mice, whereas only the 25-Hz band was present in WT mice (Fig. 5 A and B). In addition, autocorrelograms revealed that 28.58% of the cortical units recorded from R6/1 mice oscillated, half of them at high γ-frequency (67.59 Hz ± 2.16; Fig. S4). Finally, 42.9% of cells in R6/1 mice were coherent to cortical LFP at 60 to 80 Hz, with phase relationship (Fig. 5 C–E). These data suggest that the cortical γ-oscillation was not merely generated by volume conduction originating from the striatum, and there exists a global corticostriatal high γ-synchrony in R6/1 mice.

Fig. 5.

Cortical high γ-oscillation in R6/1 mice. (A and B) Power spectra of session-wide cortical LFP by FFT averaged for each genotype and training stage. (C) Percentage of cells coherent to cortical LFP in both genotypes. (D and E) Coherence frequency distribution and phase relationship with high γ-cycle for coherent cortical cells in R6/1 mice.

Discussion

Our data in R6/1 mice demonstrate how the HD mutation compromised the capacity of principal striatal output neurons to process learning and task-related information, thereby resulting in an alteration of the information flow in basal ganglia loops. This altered activity pattern is tightly linked to the difficulties of these mice in forming action–reward associations that depend on the integrity of this circuitry. Learning deficits were highly associated with the diminished ability of the transgenic mice to retain the learned information across days, whereas somewhat preserved progression within a session may be supported by other regions or circuits.

Surprisingly, scarce MSNs recorded from R6/1 mice showed task-related firings, similarly to MSNs recorded from WT littermates. In addition, and contrary to our expectation, these cells also were capable of plasticity: the proportion of task-modulated cells substantially increased with training, albeit more slowly, similarly to WT littermates performing at an equivalent level. This suggests that the change in the number of recruited MSNs, rather than the proportion of the recorded MSNs that are task-sensitive, is more closely correlated with the delayed learning in the transgenic mice. Our data thus demonstrate and confirm that the incremental procedural learning requires substantial recruitment and involvement of striatal output neurons.

It is likely that inactive MSNs in R6/1 mice are part of the indirect striatopallidal pathway known to be particularly vulnerable in HD, whereas the small number of MSNs recruited during the task are part of the direct striatonigral pathway, which is preserved in the early stages of HD (18). It is also likely that the full extent of the learning impairments observed here cannot solely be correlated with corticostriatal dysfunction, as hippocampal plasticity or other central and peripheral physiologies have also been shown to be changed in several lines of transgenic mice (19, 20), including R6/1 line. However, as the type of procedural learning studied here has been known to require an intact striatum (21), pathologies of other brain areas including the hippocampus are not likely to be primarily associated with the severe learning deficits observed in R6/1 mice.

A priori, our data do not appear to corroborate previously described elevated striatal activity in the similar R6/2 mice exploring in an open field (22, 23). Along with the differences in transgenic lines and behavioral conditions used for recording, difference in disease advancement could account for the discrepancy. Although we are aware of no study in R6/1 mice, an in vitro intracellular recording study in R6/2 mice once again showed that their striatal MSNs were hyperexcitable at a presymptomatic age (6 wk), consistent with data from Rebec's group (22, 23), and became hypoexcitable later at a symptomatic age (12 wk) (16). Another study of YAC128 transgenic mice also reported biphasic increase and subsequent decrease of synaptic transmission in the corticostriatal pathway with disease progression (24). Therefore, this age-related biphasic temporal evolution of basal ganglia physiology caused by mutant proteins might explain the discrepancy between the excitability observed in previous in vivo recording studies in presymptomatic R6/2 mice (6–9 wk) and our findings in older R6/1 mice (14–20 wk). Our mice might represent a more advanced stage of the disease with corresponding decreased synaptic transmission compared with those younger R6/2 mice.

In contrast to the dramatically reduced number of MSNs recorded in R6/1 mice, putative fast-spiking INs were encountered as often in R6/1 mice as in their WT littermates. More strikingly, the activated INs also responded to the cognitive task, and their proportion increased with training stage as in WT mice. The data from WT mice are consistent with previous studies reporting task-related firing of INs in normal rodents (10, 17, 25). This suggests relatively normal functioning of local striatal INs in R6/1 mice. However, a substantial proportion of the recruited INs of R6/1 mice showed enhanced oscillatory activity at high γ-frequencies (50–80 Hz). Therefore, despite excessive synchrony, striatal INs still responded to afferent (mainly cortical) excitation during task learning, and the two phenomena seem to cooccur as a result of their appearance at different time scales: milliseconds and seconds for γ-oscillation and task-related firing, respectively. Even though task-related IN activity appears normal, it seems highly likely that fine data processing was perturbed by this excessive synchronization of striatal cells. The impaired processing could ultimately be involved in the behavioral impairments of R6/1 mice. Brown suggested a similar explanation for the effect of pathological β-oscillation in Parkinson disease (26).

In addition, highly synchronized pairs of cells observed in our study also led us to suspect cross-structure synchrony within corticostriatal (17, 27), or striatopallidal loops (28). Our data indicate that cortical neurons also oscillate and entrain to cortical LFPs in R6/1 mice, suggesting multiple origins of the γ-synchrony in corticostriatal loops. These data also confirm that the cortical alterations are among the early events in HD (4, 29).

Although γ-oscillation has been associated with high-order information processing (e.g., attention, working memory) in different neural regions such as the cortex, hippocampus, or the olfactory bulb (30), the physiological reality of this fast rhythm in the basal ganglia has only recently been explored (14, 31). This band has been observed in normal awake and anesthetized animals (15, 32, 33) and in humans (34). In normal rats performing a behavioral task, the 70- to 100-Hz oscillation was significantly enriched during rewarded (but also in nonrewarded) trials (14). Furthermore, dopamine (DA) supplementation in Parkinson disease conditions not only abolished pathological 15- to 30-Hz β-oscillation, but also increased γ-oscillation at greater than 70 Hz (34), in association with prokinetic effects. In addition, a pharmacological blockade of DA receptors resulted in a shift from 80 to 100 Hz to 50 Hz synchrony in awake mice (35), whereas injections of psychostimulants or DA agonists in normal rats produced an opposite effect of enhancement of γ-synchrony (14), as in our R6/1 mice. Taken together, these data suggest increased DA transmission in R6/1 mice, as in HD (36), whereas a decreased nigrostratal activity and DA transmission have been reported in the R6/1 mice (37, 38). No explanation is currently available to reconcile this apparent discrepancy.

In sum, the data from R6/1 mice reported here provide crucial information on the in vivo cellular and network processes in the striatum and cortex through which the HD mutation exerts its effects on behaviors and cognitive abilities in early HD. This study also offers insights into fundamental aspects of corticostriatal activity that may be central to normal procedural learning and habit memory formation.

Experimental Procedures

Subjects.

Subjects were R6/1 transgenic mice (n = 19; 11 striatum, eight cortex) and age-matched WT littermates (n = 13; eight striatum, five cortex) obtained from crossbreeding of male R6/1 (7) (C57BL/6 background; Jackson Laboratory) and female C57/BL6 mice (IFFA/Credo). The R6/1 line expresses exon 1 of the human HD gene with an expanded number of CAG trinucleotide repeats (approximately 116–126 repetitions). Genotypes were tested by PCR of tail biopsy specimens. Both male and female mice were used in equivalent numbers for both genotypes in our experiment.

Surgery and Behavioral/Recording Techniques.

The twin tetrodes (39, 40) (Fig. 1E) were positioned bilaterally at the dorsal striatum at 0.7 mm anterior to Bregma, 1.5 mm left and right of the midline suture at 2.0 to 2.5 mm vertical from the skull. For cortical recording, electrode configuration and stereotaxic coordinates were identical except that the electrodes were placed 1.0 to 1.5 mm above the striatal coordinates. Stainless steel tubing containing a tetrode whose position was constantly 1.0 mm above the twin tetrode was used as the animal ground and reference electrode. Following recovery from surgery, cell activity was recorded daily and at the same time the mice were subjected to progressive food deprivation (85–90% of ad libitum weight). Mice were recorded daily while they acquired operant conditioning in an operant chamber (Fig. 1F). The chamber contained 15 NP holes placed over three of its inner walls and a food well that delivered 7 μL of sweetened milk as reward. We maintained the positions of the electrodes throughout the recording sessions to capture dynamic modifications in firing patterns of the same cells during the course of learning. However, if signals were lost, tetrodes were lowered between sessions to search for new cells. Typically, experiments were performed each day in the morning for 30 min, and training continued until mice reached criteria of more than 100 rewards obtained during a standard 30-min session with less than 30% of variation of rewards for three consecutive sessions (i.e., asymptote). Because R6/1 mice exhibited severe difficulty in acquiring the task, when necessary, the training sessions for these mice were extended to 60 or 120 min by using multiple sessions daily.

Cell Type Distinction.

We used a spike width/duration criterion to classify cells into at least two subtypes, although extracellular recording does not allow definite identification of cell type (15, 27). Spike width was defined as the delay between the maximum (i.e., peak) and the minimum (i.e., valley) of the waveform. We classified cells with spike width less than 0.25 ms as putative narrow-spiking local INs, and the remaining cells (spike width > 0.25 ms) were classified as MSNs (Fig. 1). A total of 28 units in WT mice and 13 units in R6/1 mice were discarded from the analysis because they had unclassable waveforms, too few spike counts (i.e., <100), or firing characteristics of tonically active neuron.

Task-Responsive Cells Using Circular Perievent Time Histograms.

To determine whether a cell responds to the behavioral task, we first identified, by using video tracking information, the moment at which a mouse left the food well (area) for EX for each trial ending with reward consumption, in addition to the originally “tagged” behavioral events of NP and first L of the reward consumption (Fig. 3 and Fig. S2). We then constructed three perievent time histograms (PETHs) focusing on periods between two consecutive events within repeated sequences of NP–L–EX. The three periods were between NP and L (period I), between L and EX (period II), and between EX and NP (period III). As each period varies from trial to trial on each session, we adjusted these values to average spike counts for different trials within a session.

To do this, we first calculated mean and SD of durations of each of the three periods for a given session. We then selected only trials whose durations were within ±1.96 SD of the mean, and dropped trials with significantly deviant durations for any of the three periods from the analysis to avoid data contamination. As the number of trials used to calculate PETH would influence the task responsiveness of a given cell, we included only recording sessions with at least 30 trials selected by the previous method in our analysis. By using this criterion, we were able to perform analysis on 157 and 185 neurons in relation to behavior in WT and R6/1 mice, respectively. In addition, for sessions with more than 30 selected trials, only the last 30 trials were used for the analysis to reflect neural activity at an advanced stage of learning with equal statistical power. We then calculated mean durations for the three periods for these 30 selected trials once again. Finally, the mean duration value was attributed to all trials. The firing rate (i.e., spike counts) for each of the three periods was calculated separately and added end-to-end to generate a circular PETH with adjusted number (determined by the mean duration of a period) with constant bin size of 40 ms. The firing rates were then smoothed with a moving average of three bins for statistical analysis.

We used the MOSUM test (41), a structural change test based on regression analysis, to assess statistically whether a striatal cell responded to the task. We chose this test because our data do not allow the identification of baseline firing rates with which changes in firing across behavioral events could be detected. This test is based on the detection of changes, that is, the regression coefficient of subsamples containing changes should be substantially different from the coefficient of the entire dataset. To perform this analysis we used the software R, including the package Strucchange (42). The moving sum of OLS and the boundaries were calculated, respectively, with the functions efp() and boundary() in R. In comparison with human eyes, we found best results with a window size of eight bins (corresponding to 240 ms) and a P value at 0.01.

Statistical Analysis.

Analyses of behavioral and recording data were performed by using a Student t test, and χ², with the significance level set at P < 0.05. Further information on data processing is provided in SI Experimental Procedures.