Transient and selective overexpression of D₂ receptors in the striatum causes persistent deficits in conditional associative learning

Abstract

Cognitive deficits in schizophrenia are thought to derive from a hypofunction of the prefrontal cortex (PFC), but the origin of the hypofunction is unclear. To explore the nature of this deficit, we genetically modified mice to model the increase in striatal dopamine D₂ receptors (D₂Rs) observed in patients with schizophrenia. Previously, we reported deficits in spatial working memory tasks in these mice, congruent with the working memory deficits observed in schizophrenia. However, patients with schizophrenia suffer from deficits in many executive functions, including associative learning, planning, problem solving, and nonspatial working memory. We therefore developed operant tasks to assay two executive functions, conditional associative learning (CAL) and nonspatial working memory. Striatal D₂R-overexpressing mice show a deficit in CAL because of perseverative behavior, caused by interference from the previous trial. D₂R up-regulation during development was sufficient to cause this deficit, because switching off the transgene in adulthood did not rescue the phenotype. We validated prefrontal dependency of CAL by using neurotoxic lesions. Lesions of the medial PFC including the anterior cingulate, infralimbic, and prelimbic cortices impair CAL because of increased interference from previously rewarded trials, exactly as observed in D₂R transgenic mice. In contrast, lesions restricted to the infralimbic and prelimbic cortices have no effect on CAL but impair performance in the nonspatial working memory task. These assays not only give us insight into how excess striatal D₂Rs affect cognition but also provide tools for studying cognitive endophenotypes in mice.

Psychiatric diseases are characterized by their symptoms because, unlike for neurological disorders, reliable biological diagnostic markers are still not available. However, in the last two decades, technological advances in brain imaging, neuropathology, human genetics, and epidemiology have revealed morphological and molecular alterations in the brains of patients with psychiatric disorders. This knowledge can now be used to generate animal models that will help to elucidate the molecular mechanisms underlying the pathogenesis of these diseases.

Brain imaging and neuropathological studies in schizophrenia have found an alteration in dopamine D₂ receptor (D₂R) signaling in the striatum reflected as an increase in the release of dopamine and in the density and occupancy of D₂Rs (1–4). Several studies suggest that, in a subpopulation of patients, the increased dopamine D₂R density may be genetically determined (http://www.schizophreniaforum.org/res/sczgene/meta.asp?geneID=93) (5). Because almost all antipsychotic drugs are directed against D₂Rs, increased D₂R signaling has long been thought to be important for the pathophysiology of schizophrenia.

To determine the behavioral and physiological consequences of increased striatal D₂R density, we generated mice with reversibly increased levels of D₂Rs restricted to the striatum (6, 7). These mice showed selective deficits in cognitive endophenotypes of schizophrenia without a generalized cognitive deficit. Specifically, we found that the acquisition of a prefrontal-dependent spatial working memory task was affected by developmental up-regulation of striatal D₂Rs, and the deficit correlated with changes in the levels of dopamine, dopamine turnover, and D₁ receptor (D₁R) activation in the prefrontal cortex (PFC) (6). This observation may be instructive for the pathophysiology of schizophrenia. Studies in humans, primates, and rodents have found that working memory requires an optimal level of activation of D₁Rs in the PFC, with either an increase or decrease resulting in poorer working memory (8–10). Moreover, patients with schizophrenia consistently show a hypofunction of the PFC (for review, see ref. 11), and D₁R density in the PFC has been correlated with deficits in working memory and executive function in patients (12, 13).

The cognitive deficits in schizophrenia correlate highly with a patient's functional capacity (14), and their severity have the greatest impact on the patient's long-term prognosis. Unfortunately, at present, the cognitive symptoms of schizophrenia are only marginally improved by antipsychotic treatment. Several meta-analyses of published studies suggest that the second-generation, atypical antipsychotics may have a modest and replicable ability to improve semantic memory, attention, and declarative memory but they do not affect working memory or executive functions (15–17). In the CATIE multisite study including >1,000 patients, Keefe et al. (18) concluded that the small improvements in neurocognition observed during typical and atypical antipsychotic treatment were not likely to be greater than placebo or test practice effects. Cognitive symptoms have therefore become a major focus in schizophrenia research, and this has encouraged us to explore in greater detail the nature of prefrontal dependent behavioral deficits in striatal D₂R-overexpressing mice. In particular, we wanted to know whether other types of executive function that are impaired in schizophrenia are also affected by excess striatal D₂Rs.

Therefore, we developed additional tests of prefrontal-dependent behaviors in mice. Although tests for a number of executive functions have been used successfully in rats (19, 20), fewer have been developed for and applied to mouse models, and so far only maze tests have been validated by using prefrontal lesions in mice (6, 21). The use of maze-based working memory tests alone presents several limitations. First, they are limited to the analysis of only one cognitive domain, spatial working memory, whereas many additional components of executive function are disrupted in schizophrenia, which require investigation in animal models. Second, the working memory deficit observed in schizophrenia is apparent across many modalities, not just spatial information processing. Third, maze-based working memory tasks suffer from the specific caveat that the delay length is not precisely controlled, and it is not possible to implement the test completely without delay, thereby making it difficult to dissociate the inability to learn the rules of the task from a working memory deficit. In this study, we have developed and validated an operant-based PFC-dependent task without these limitations.

We chose to explore conditional associative learning (CAL) in mice because it is a form of executive functioning that is deficient in both patients with schizophrenia and individuals with frontal lobe lesions (22–26). CAL tasks comprise arbitrary, fixed associations between members of a set of stimuli and a set of responses that are learned through a process of trial and error. Successful CAL therefore requires strategic regulation of behavior by using internal representations of prior episodic information and previously gained procedural information. Specifically, we adapted for mice an operant conditioning-based CAL task previously described for testing rats (27). This task is directly comparable with CAL tasks used in humans and nonhuman primates, differing only in the number of stimulus–response associations the subject is required to learn. Because we can test the mice on the paradigm with no delay, we also extended this task to measure nonspatial working memory independently of associative learning.

We found that D₂R-overexpressing mice showed decreased performance in the CAL task because of perseveration of responses that were rewarded in the previous trial. As for the deficit in the spatial working memory task we described in ref. 6, developmental up-regulation of D₂Rs was sufficient to induce this deficit. We validated that CAL is sensitive to prefrontal lesions in mice, as has been determined for humans, monkeys, and rats (22, 23, 28, 29). Lesions of the medial PFC in mice impaired performance on the CAL task because of the same type of perseverative behavior observed in striatal D₂R-overexpressing mice. This deficit was evident only when the lesion extended to the anterior cingulate cortex, whereas lesions restricted to the prelimbic and infralimbic cortices did not disrupt CAL but resulted in a deficit after delays were incorporated into the task.

Results

Transient and Selective Overexpression of D₂Rs in the Striatum Causes Persistent Deficits in CAL.

In the CAL task, reinforcement depended on pressing a lever within 5 s in response to one auditory stimulus and withholding from pressing the lever for 5 s in response to a different auditory stimulus. Accuracy on this go/no-go task therefore depends on learning two stimulus–response associations. Learning takes place across sessions, through trial and error.

Mice with striatum-specific up-regulation of D₂Rs tested on this CAL task showed slower acquisition of the task than control littermates. Fig. 1A depicts percentage correct across nine blocks, each comprising five sessions. Percentage correct was calculated by dividing the number of reinforced trials, which was always 55, by the total number of trials (correct and incorrect). Because the mice were generated by using the tetracycline system, which requires the expression of two independent transgenes, we tested whether each individual transgene may have an effect on performance. A comparison of the learning curves in Fig. 1A suggests that the transgenic mice learned the task more slowly than wild-type and single-transgene (tetO-D₂R or CamKII-tTA) littermates. Fig. 1A also shows that the performance of the transgenic mice leveled off well below that of the control groups. The percentage correct across trials was significantly lower for the D₂R-overexpressing mice (F_(3,24) = 9.122, P < 0.0001).

Fig. 1.

Mice with selective overexpression of D₂Rs in the striatum are impaired in conditional associative learning. Developmental overexpression is sufficient to induce the deficit. (A) Percentage correct trials of transgenic and control mice. Each session block consisted of five sessions. Controls included the following: wild-type littermates (n = 5); tetO, mice carrying the tetO-D₂R transgene (n = 5); tTA, mice carrying the CamKII-tTA transgene (n = 7); and D₂R-OE, D₂R-overexpressing mice that carry both transgenes (n = 11). *, P < 0.0001. The solid line represents chance (50%); the dotted line depicts acquisition criterion (78%). (B) Distribution of reinforced lever presses and withholds in the four groups of mice. The number of reinforced responses for each session block is depicted. LP, lever press; WH, withhold. (C) Probability of emitting an error by repeating the previously reinforced response for the four groups (analyzed for last session). *, P < 0.001. (D) Probability of emitting a correct response when repeating a previously reinforced response for the four groups (analyzed for last session). P > 0.05. (E) Percentage correct trials of transgenic and control mice treated with doxycycline: wild-type, tetO, and tTA single transgenic littermates (n = 7); D₂R transgenic, mice carrying both transgenes (n = 5). *, P = 0.05. The solid line represents chance (50%). Error bars represent SEM.

We then analyzed the distribution of reinforced lever press and withhold responses. In each session, there is the opportunity to earn 55 reinforcements, and the mouse determines how they are obtained. It is possible to earn all 55 reinforcements by emitting only one type of response for all trials; however, this is not an efficient strategy because it takes twice as long to earn all reinforcements. If a defect in response inhibition existed, most reinforcements would be earned via the lever press response, and if only one stimulus–response association was acquired, then all reinforcements would be earned via one response type, lever presses or withholds. At the beginning of training, all groups earned the majority of reinforcements through lever presses. As the mice acquired the two stimulus–response associations through training, there was a dramatic decrease in the number of lever presses, concurrent with an increase in the number of reinforced withhold responses. Fig. 1B shows that, at the end of training, all groups earned one-half the reinforcements via each response type [lever pressing (LP), F_(3,24) = 0.572, P = 0.639; withholding (WH), F_(3,24) = 0.549, P = 0.654]. This indicates that the D₂R transgenic mice did not have a defect in response inhibition and were able to learn both stimulus–response associations.

We analyzed the type of errors the mice were making in the last session. Because it is randomly determined which stimulus will be presented for each trial, accurate performance requires flexibility in response selection because the correct response varies from trial to trial. We identified all trials in which a correct response was made and followed by a trial in which the stimulus changed. If on the subsequent trial the mouse repeated the previously reinforced response, which was now incorrect, a value of 1 was assigned. If the mouse responded correctly on the subsequent trial, a zero was assigned. Fig. 1C depicts the probabilities of repeating a previously reinforced response after the stimulus changed and indicates that D₂R transgenic mice were significantly more likely to emit this type of error (overall F_(3,24) = 9.948, P < 0.001).

We then examined the data from the last session to determine whether the D₂ transgenic mice also exhibited an increased probability to emit an error on consecutive trials in which the same stimulus was presented. For such trials, if the mouse repeated a previously reinforced response, a zero was assigned to the trial, whereas if an error was made, a 1 was assigned to the trial. The probability of emitting this type of error was similar across all four groups, as seen in Fig. 1D (overall F_(3,24) = 2.27, P > 0.05). Thus, the defect in D₂R transgenic mice observed in Fig. 1A is specifically due to a failure in switching the response.

We previously found that up-regulation of striatal D₂R expression during development was sufficient to induce a deficit in a prefrontal-dependent working memory task. To see whether the same may be true for the CAL task, we switched off the transgene from 8 to 10 weeks of age by feeding the mice with doxycycline-supplemented food (40 mg/kg in regular chow). We began continuous reinforcement (CRF) training 2 weeks after starting doxycycline treatment, because we previously determined that after 2 weeks on doxycycline the increase in D₂R protein is reversed (6). As seen in Fig. 1E, reversing D₂R expression to normal levels did not reverse the deficit in CAL (P = 0.05).

CAL Is Impaired in Mice with Medial Prefrontal Lesions.

Because this is a study of operant CAL in mice, we examined the dependency of the integrity of the PFC for this task by assessing the performance after lesioning the medial PFC. In Fig. 2A, the “control” group comprises both sham lesion (n = 8) and un-operated-on mice (n = 9). For all analysis, the data were collapsed across un-operated-on and sham lesion groups and referred to as “control,” because ANOVAs did not reveal a significant effect of the sham surgery across any of the behavioral measures. The majority of the control group (15 of 17, 88%) reached the acquisition criterion, 78% correct or better, achieved across five consecutive days (Fig. 2A).

Fig. 2.

Mice with a medial prefrontal lesion including the anterior cingulate are impaired in conditional associative learning. (A) Percentage correct trials of LTA, LNA, and control mice. *, P < 0.006. Each session block consists of five sessions. The solid line represents chance (50%), and the dotted line depicts acquisition criterion (78%). (B) Distribution of reinforced lever presses and withholds for the three groups. The number of reinforced responses for each session block is depicted. LP, lever press; WH, withhold. (C) Probability of emitting an error by repeating the previously reinforced response for the three groups (analyzed for last session). *, P < 0.001. (D) Probability of emitting a correct response when repeating a previously reinforced response for the three groups (last session analyzed). P > 0.05. Error bars represent SEM.

Thirteen mice that received an NMDA-induced lesion of the medial prefrontal cortex (mPFC) were included in this study based on histology. The majority of this group (n = 8, 62%), referred to as “lesioned task acquirers” (LTA), reached the acquisition criteria. In contrast, the remaining mPFC-lesioned mice (n = 5, 38%), did not meet the acquisition criterion and are therefore referred to as “lesioned nonacquirers” (LNA). A comparison of the learning curves in Fig. 2A suggests that the LNA group learned the task more slowly than the LTA or control groups. Fig. 2A also shows that the LNA group performance leveled off well below that of the other groups. Percentage correct across trials was significantly lower for the LNA group (overall F_(2,28) = 6.345, P = 0.006).

The PFC is important for attentional processes. Because latency to emit a response is a measurement of attention, we determined the mean latency to respond across the three groups at the very first session of the CAL task. No difference was found, suggesting that attentional processes may not be grossly affected by the lesion (data not shown) (overall F_(2,28) = 0.011, P = 0.989).

Fig. 2B depicts the distribution of reinforced lever press and withhold responses across nine blocks of five sessions for the three groups. By the end of training, all three groups earn approximately one-half the reinforcements via each response type (lever press, F_(2,28) = 0.66, P = 0.52; withhold, F_(2,28) = 0.66, P = 0.52), indicating that, like D₂R transgenic mice, LNA mice were able to learn both stimulus–response associations and were able to inhibit responses.

We then identified all trials in which a correct response was made followed by a trial in which the stimulus changed and averaged across all trials the probability of repeating the previously reinforced response. Fig. 2C shows that LNA mice were significantly more likely to repeat a reinforced response on the next trial, even if the stimulus changed (overall F_(2,27) = 14.35, P < 0.001). LNA mice did not show an increase in the probability of emitting an error on consecutive trials with the same stimulus (Fig. 2D; F_(2,27) = 14.35, P > 0.05). Thus, the reduction in percentage correct observed in Fig. 2A is related to a bias toward repeating the response that was reinforced on the previous trial even though the stimulus has changed.

Impaired CAL Is Associated with the Amount of Area Lesioned in the Anterior Cingulate Cortex.

We sought to determine whether we could distinguish between LNA and LTA mice based on histology. Fig. 3 depicts the maximum damage for both groups in the regions of the mPFC across rostral to caudal sections. The major difference between the two groups is that the lesion extends more caudally in LNA mice. We next compared the mean percentage of damage in each area of the mPFC across the two groups. The anterior cingulate cortex (Cg) is a dorsal region of the rodent mPFC, and it is subdivided into the Cg1 and Cg2 regions. In the Cg1 region, significantly more damage was observed in the LNA group [supporting information (SI) Fig. S1A; overall F_(1,11) = 17.370, P < 0.005] and in the Cg2 region, damage was only observed in LNA mice (Fig. S1C; overall F_(1,11) = 6.754, P < 0.05). Correlation analysis on data combined from both groups found that, for both Cg1 and Cg2, the mean percentage size of damage collapsed across sections was significantly correlated with percentage correct on the fifth session block (Fig. S1 B and C; Cg1, P < 0.05; Cg2, P < 0.05). Thus, an increase in anterior cingulate lesion size correlated with decreased performance.

Fig. 3.

Schematic representation of the maximal lesions in the medial PFC for the mice that acquired the task (LTA) and that did not acquire the task (LNA).

The prelimbic cortex (PL) and infralimbic cortex (IL) form the ventral region of the rodent mPFC. There was little difference in the amount of damage to the PL and IL regions between the two lesioned groups (Fig. S1 E and G; PL, overall F_(1,11) = 3.255, P > 0.05; IL, overall F_(1,11) = 0.213, P > 0.05). Thus, LTA mice acquired the task, despite a sizeable lesion in the PL and IL cortices. Given that both regions were similarly damaged in the two lesion groups, this suggests that the impairment observed in LNA mice may be unrelated to the damage in the PL/IL areas. We nevertheless determined whether there was a significant relationship between the size of the lesions in PL and IL and accuracy. For this, the data from LTA and LNA mice were combined to assess whether a significant correlation existed between percentage correct on session block 5 and the size of the lesions averaged across all sections. No significant correlation was found (Fig. S1 F and H; PL, P = 0.505; IL, P = 0.654).

Lesion of the PL and IL Is Sufficient to Disrupt Performance on an Operant-Based Nonspatial Working Memory Task.

By adding a delay between the offset of the 10-s stimulus presentation and the lever presentation, we extended the task to test nonspatial working memory. D₂R-overexpressing mice and the group of lesioned mice that did not acquire the CAL task were not tested on the working memory paradigm because the interpretation of their behavior would be made difficult by their poorer level of performance at asymptote. However, to determine whether the delayed version depends on the medial PFC, we compared the performance of LTA and control mice. Fig. 4 displays percentage correct as a function of delay across both groups. In both groups, LTA and controls, a delay-dependent decrease in percentage correct was observed but the lesioned group performed significantly worse across the four delays (overall F_(1,21) = 7.208, P < 0.05).

Fig. 4.

Working memory is significantly impaired in mice with a lesion restricted to the PL and IL cortices. Percentage correct trials for control and LTA mice when tested without delay (0 s) or variable delays (1–10 s) are shown. *, P < 0.05.

Discussion

How altered prefrontal cortical functioning leads to cognitive deficits has become a major focus in the study of schizophrenia. A hypofunction of the PFC may not necessarily explain all of the cognitive deficits, which are heterogeneous and vary between individual patients. However, because the degree of the cognitive symptoms predicts the long-term prognosis of the disease better than the degree of the positive symptoms, understanding the underlying molecular mechanisms of these symptoms is essential for the development of successful treatment of schizophrenia. Animal models addressing the functioning of the PFC are a first step in this direction, and the power of such models critically depends on the availability of specific and rigorous behavioral analyses that can selectively address disease-related endophenotypes in mice.

We developed a model of D₂R up-regulation in the striatum of genetically modified mice (6) that recapitulates the D₂R up-regulation observed in patients with schizophrenia (3, 4). To extend the usefulness of mouse models designed to explore the molecular deficits in cognitive function, we have here applied operant conditioning in mice to measure a well established schizophrenia cognitive endophenotype: CAL. By introducing a delay in the CAL task, we have also developed a working memory task that, as discussed in the introduction, provides several advantages over the commonly used maze-based working memory tasks. We validated both the CAL task and the delayed CAL-working memory tasks by assessing the performance of mice with mPFC lesions. We found that CAL was significantly impaired in both D₂R transgenic mice and mice with mPFC lesions that included the anterior cingulate cortex. Mice with mPFC lesions that leave the anterior cingulate cortex intact were unimpaired on the CAL task but showed significantly disrupted performance on the working memory version. These results are interesting given the observation that patients with schizophrenia and frontal lobe patients are impaired in CAL (22–26) and working memory (11, 30).

The decreased performance observed in the D₂ transgenic and the lesioned mice arose from interference from previously rewarded trials. Although both groups were capable of varying their responses as a function of the current stimulus, they showed a defect when reference memory for the two stimulus–response associations was confused with information from the previous trial. In both groups, reinforcement of a given response created a bias in favor of repeating that response, even when it was no longer relevant. This type of perseverative behavior may be due to increased proactive interference or a deficit in prepotent response inhibition and can therefore be described as a reduction in interference control. Proactive interference occurs when the memory of the previous trial interferes with the information about the current trial, such that the subject responds in a way that was relevant to the previous trial, even though it is no longer appropriate. A defect in prepotent response inhibition occurs when the subject is unable make a response appropriate to the current trial because of the inability to suppress a dominant, automatic, or prepotent response. Our task does not definitively distinguish between these two types of interference, which data from human studies suggest are independent (31).

Our finding is consistent with a prior study showing that humans with PFC lesions more frequently made errors by repeating the previously correct choice selection in a CAL task (27). Our data show that combined lesion of infralimbic, prelimbic, and anterior cingulate cortices results in reduced interference control, whereas lesion of the prelimbic and infralimbic cortices alone does not. This might suggest that the behavior is mediated by the anterior cingulate in isolation, or alternatively, it is the combined destruction of these regions that is necessary to produce the deficit. To conclude that damage to the anterior cingulate is sufficient to disrupt this task, lesions restricted to the cingulate would need to be assessed. Rats with lesions restricted to the anterior cingulate cortex show normal acquisition of a go/no-go conditional discrimination task (32). Similarly, rats with discrete lesions of either the anterior cingulate or prelimbic/infralimbic cortices showed normal acquisition of a biconditional discrimination task (33). Interestingly, in healthy human subjects, an increase in anterior cingulate activity was observed when a prepotent response needed to be inhibited in two different tasks, a variant of the Continuous Performance Test (AX-CPT) and a Stroop task (34, 35). This does not represent a conflict between the anatomical specificity within the mPFC for prepotent response inhibition between humans and rodents because functional MRI activation may reflect sufficiency of an anatomical region to mediate a specific task, rather than necessity.

This cognitive endophenotype is highly relevant to the study of schizophrenia models. The impairment on the CAL task observed in patients with schizophrenia is thought to arise from poor cognitive control rather than impaired reference memory. When patients were informed that they needed to learn multiple stimulus–response associations to perform the task accurately, they still exhibited impaired performance despite demonstrating that they had acquired reference memory for the rules of the task (26). Patients with schizophrenia exhibit a deficit in prepotent response inhibition that correlates with decreased activity in the anterior cingulate cortex (36). Furthermore, there is reduced anterior cingulate activity at rest in patients with schizophrenia (37).

Of particular interest in this study is the finding that up-regulation of D₂Rs in the striatum during development was sufficient to induce the deficit in the CAL task. Switching off excess D₂R expression in the adult by feeding 8- to 10-week-old D₂ transgenic mice with doxycycline did not reverse the deficit in the CAL task. This finding is similar to what we previously observed in the delayed non-match-to-sample T-maze working memory task, in which even switching off the transgene directly after birth did not reverse the behavioral impairment (6). This suggests that both deficits may share common underlying mechanisms. The altered dopamine turnover and D₁R activation that we previously observed in the PFC as a consequence of D₂R up-regulation in the striatum may therefore contribute to the deficits in both the working memory task and in the CAL task. In this context, it would be interesting to study whether in patients with schizophrenia deficits in working memory correlate with deficits in CAL. Our results also draw new attention to the striatum by demonstrating that the striatum plays a central role in the cognitive processes affected in schizophrenia and may achieve this central function in the generation of cognitive symptoms as a result of its interconnections from and back to the PFC (38).

The fact that switching off transgene expression in the adult animal does not reverse the behavioral deficits in both tasks suggests that compensatory mechanisms are induced by early D₂R up-regulation that cannot be reversed by switching off the transgene in the adult animal. As a corollary, our finding further suggests that antipsychotic medication that is mainly directed against D₂R activity may not improve the cognitive symptoms in patients with schizophrenia because they are given too late. Through continued study of this and other mouse models, we should be able to gain insight into the compensatory mechanisms that occur in the developing brain to uncover new molecular targets for developing treatments for the cognitive deficits of schizophrenia.

Materials and Methods

Generation of D₂R Transgenic Mice and Mice for Lesion Studies.

See SI Methods.

Medial PFC Lesions.

Control littermates of D₂R transgenic mice were anesthetized with 1.2% Avertin (2,2,2-tribromoethanol, in 0.22 mM 2-methyl-2-butanol; Sigma–Aldrich) 0.02 ml of solution per gram of body weight. A total of 0.2 μl of 10 mg/ml NMDA (Sigma–Aldrich) was injected bilaterally in a total of four injection sites at a rate of 0.1 μl/min through a 33-gauge internal cannula (Plastics One) angled at 20° toward the midline. The coordinates were 2.0 and 2.4 mm anterior to bregma, 0.9 mm lateral to the midline, and 1.0 mm ventral to the brain surface. Sham animals were anesthetized, their skulls were drilled, and a cannula was lowered to the same coordinates. Postsurgery, all mice were single housed and given 3 weeks for recovery before testing.

Histology.

At the conclusion of the behavioral testing, mice were perfused transcardially with PBS followed by 4% paraformaldehyde. Brains were then fixed in 4% paraformaldehyde overnight then cryoprotected by immersion in 30% sucrose overnight. Brains were sectioned on a freezing Microtome at 40-μm thickness and stained with cresyl violet to assess the extent of lesion-induced neuronal loss and sketched onto drawings from ref. 39.

Behavior and Data Analysis.

See SI Methods.