Overtraining Strengthens the Visual Discrimination Memory Trace Outside the Hippocampus in Male Rats

Lehmann, Hugo; Stykel, Morgan G.; Glenn, Melissa J.

doi:10.3389/fnbeh.2021.768552

ORIGINAL RESEARCH article

Front. Behav. Neurosci., 15 November 2021

Sec. Learning and Memory

Volume 15 - 2021 | https://doi.org/10.3389/fnbeh.2021.768552

This article is part of the Research Topic The Emergent Engram: Multilevel Memory Trace Components and the Broader Interactions View all 8 articles

Overtraining Strengthens the Visual Discrimination Memory Trace Outside the Hippocampus in Male Rats

$\r\nHugo Lehmann*$ Hugo Lehmann^1*

Morgan G. Stykel¹

Melissa J. Glenn²

¹Department of Psychology, Trent University, Peterborough, ON, Canada
²Department of Psychology, Colby College, Waterville, MA, United States

The hippocampus (HPC) may compete with other memory systems when establishing a representation, a process termed overshadowing. However, this overshadowing may be mitigated by repeated learning episodes, making a memory resistant to post-training hippocampal damage. In the current study, we examined this overshadowing process for a hippocampal-dependent visual discrimination memory in rats. In Experiment 1, male rats were trained to criterion (80% accuracy on two consecutive days) on a visual discrimination and then given 50 additional trials distributed over 5 days or 10 weeks. Regardless of this additional learning, extensive damage to the HPC caused retrograde amnesia for the visual discrimination, suggesting that the memory remained hippocampal-dependent. In Experiment 2, rats received hippocampal damage before learning and required approximately twice as many trials to acquire the visual discrimination as control rats, suggesting that, when the overshadowing or competition is removed, the non-hippocampal memory systems only slowly acquires the discrimination. In Experiment 3, increasing the additional learning beyond criterion by 230 trials, the amount needed in Experiment 2 to train the non-hippocampal systems in absence of competition, successfully prevented the retrograde amnesic effects of post-training hippocampal damage. Combined, the findings suggest that a visual discrimination memory trace can be strengthened in non-hippocampal systems with overtraining and become independent of the HPC.

Introduction

Damage to the hippocampus (HPC) can cause retrograde amnesia for memories that are termed episodic or declarative, yet the vulnerability of these memories to HPC damage is not ubiquitous (Scoville and Milner, 1957; Rempel-Clower et al., 1996). For instance, there is evidence suggesting that HPC damage is more likely to spare older or remote memories (Scoville and Milner, 1957; Kim and Fanselow, 1992). This phenomenon has been accounted for by long-term systems consolidation theories, which propose gradual strengthening of the representation in neocortical structures resulting in a decreased or differential contribution of the HPC to the recall process (Squire, 1992; Alvarez and Squire, 1994; Squire and Alvarez, 1995; Nadel and Moscovitch, 1997; Frankland and Bontempi, 2005; Winocur et al., 2010; Sekeres et al., 2018). The distributed reinstatement theory (DRT), however, introduced an alternate account to this time-dependent change; it states that re-experiencing the learning episode can be key in making a memory become established in other networks and more resistant to HPC damage, whereas time is a mere cofactor of the process (Sutherland et al., 2010).

The DRT is founded on three main tenets: (1) that there are multiple memory systems, (2) that for some forms of memory the HPC overshadows other systems, and (3) that the overshadowing can be abated with additional learning (Sutherland et al., 2010). The perspective that there are multiple neural memory systems is not new and is widely accepted (Milner, 1959; Sherry and Schacter, 1987; Packard et al., 1989; Packard and McGaugh, 1996; Squire and Zola, 1996; Nadel and Moscovitch, 1997; White and McDonald, 2002). Less described is the idea that the HPC may overshadow other memory systems, which is supported by the evidence that HPC damage induced after but not before learning causes amnesia for the same type of memory (see Sutherland et al., 2010; Lee et al., 2016). For example, HPC damage after contextual fear conditioning causes profound retrograde amnesia in rats, suggesting that the memory is normally dependent on the HPC (Lehmann et al., 2007, 2013; Sutherland et al., 2008; Broadbent and Clark, 2013; Sparks et al., 2013). The same damage, however, does not impair the rats’ ability to acquire and retain a new contextual fear memory (i.e., no anterograde amnesia), suggesting that a non-HPC memory system is now supporting the memory (Maren et al., 1997; Wiltgen et al., 2006; Lehmann et al., 2009). If other systems can acquire and support a memory in the absence of the HPC, then why do they not do so when the HPC is present at the time of learning? This can be parsimoniously answered by an overshadowing process between the HPC and non-HPC systems, in which the HPC prevents the other systems from acquiring an independent memory (Maren et al., 1997; Fanselow and Poulos, 2004; Driscoll et al., 2005; Lehmann et al., 2006; Sutherland et al., 2006; Lee et al., 2016). According to the DRT and despite the HPC overshadowing, other systems can come to support the memory if there are sufficient additional opportunities to reinstate the memory (Sutherland et al., 2010; Sutherland and Lehmann, 2011). Evidence for the DRT has come from contextual fear conditioning experiments in which distributed fear conditioning episodes prevented the typical retrograde amnesic effects of complete HPC damage (Lehmann et al., 2009). For example, HPC damage after a single conditioning session with 12 context-shock pairings eliminated the memory. Yet, in the same study, complete HPC damage after the same number of context-shock pairings distributed across 11 conditioning sessions prevented the amnesia. Hence, distributing the learning made the memory stronger in other systems as it no longer required the HPC for expression.

The DRT and its postulated process in making a memory become HPC independent has seldom been investigated beyond that of context fear memory. Therefore, we tested the generalizability of the DRT by examining whether additional learning also make a visual discrimination memory become resistant to HPC damage. We selected this type of memory because several memory systems contribute to visual discrimination memory in rats and the HPC overshadows these other systems. Specifically, evidence suggests that damage to the HPC after learning caused severe retrograde amnesia for a visual discrimination memory, but that the same rats were able to relearn the discrimination following the HPC damage (Sutherland et al., 2001; Driscoll et al., 2005; Epp et al., 2008). Given that non-HPC memory systems can acquire and support a visual discrimination memory in the absence of the HPC and in accordance with the DRT, we predicted that overtraining a visual discrimination memory would mitigate the retrograde amnesic effects of HPC damage.

Experiment 1

The present experiment examined whether strengthening a visual discrimination memory, once acquired, using additional learning trials and distributing them over several weeks would increase the likelihood of the memory becoming independent of the HPC. Specifically, rats were trained to criterion on a visual discrimination (80% accuracy on two consecutive days) and then given 50 additional training trials. This number approximates the number of trials that others have suggested is needed to train the non-HPC system in the absence of overshadowing (Epp et al., 2008). Also, in the present experiment, the additional learning was either massed within 5 days (10 trials per day for 5 days; Massed) or distributed across 10 weeks (five trials in one session per week for 10 weeks; Distributed). Because distributed learning is well known to establish stronger memories (Wagner et al., 1973; Bahrick and Phelphs, 1987), the Distributed condition was hypothesized to be the most likely to make the visual discrimination memory become HPC independent; the Massed condition served as a control for the overall number of learning trials and the time interval between learning the discrimination to criterion and the HPC damage.

Methods

Subjects

All procedures were approved by the Trent University Animal Care Committee, which follows the guidelines set by the Canadian Council on Animal Care. The subjects were 38 male Long Evans rats weighing ∼300 g at the beginning of behavioral training. The rats were housed in groups of two in standard laboratory cages and maintained on a 12:12-h light-dark cycle (lights on at 0700 h). Each rat received 25–30 g of rat chow daily and had access to water ad libitum.

Apparatus

The visual discrimination apparatus consisted of a circular pool (140 cm in diameter and 60 cm high) filled with water (21 ± 1°C) to a depth of 32 cm. The pool was located toward the center of a standard laboratory behavioral testing room and was not isolated from the room’s visible cues (e.g., door, shelving, etc.). The water was made opaque using powdered skim milk. A wall (51.5 cm wide and 61 cm high), made of white corrugated plastic, was positioned at the north end of the pool to form two equal-sized compartments, each containing a discrimination stimulus in the center. The discrimination stimuli were a yellow rubber duck (Length × Width × Height; 10 × 10 × 12 cm) and an oval-shaped red buoy (9 × 9 × 12 cm). The stimulus that predicted escape from the water (S+) was attached to a submerged platform (15 cm in diameter and 2 cm below the water surface). The other stimulus (S−) did not predict escape as it was not attached to a platform. An illustration of the apparatus and task is found in Figure 1A.

FIGURE 1

Figure 1. (A) Illustration of the visual discrimination task. Rats were released from the start position and given a maximum of 60 s to swim to the S+ and escape from the water. A trial was considered correct if the rat swam to the S+ and climbed onto the hidden platform without first entering the S– quadrant. A trial was considered incorrect if the rat’s entire body (excluding tail) entered the S– quadrant. The S+ was counterbalanced within groups such that the rubber duck was the S+ for half the rats and the red buoy was the S+ for the other half. The location of S+ was also pseudo-randomly counterbalanced across trials within a swim session so that the rats would learn to swim to the S+ and not a discrete quadrant of the pool. (B) Illustration of the experimental design for the massed training condition (top) and the distributed training condition (bottom). After reaching criterion (defined as >80% accuracy over two consecutive days), all rats were given an additional 50 discrimination trials (each duck represents five trials). In the Massed Training condition, the rats received 50 additional trials over 5 days (10 trials per day). The Recent group received either sham or HPC surgery (Sx) 5–10 days following the last day of training, whereas the Remote group received sham or HPC surgery 10 weeks following the last day of training. In the Distributed training condition, the rats received the additional 50 training trials distributed over 10 weeks (five trials in one session per week). The distributed training group received sham or HPC surgery 5–10 days following their last day of training (approximately 10–11 weeks after reaching criterion). In all cases, the rats were given a minimum of 10 days recovery before testing for retention.

Procedures

Pre-surgical training

Criterion training

The rats, one at a time, were transported to the testing room in a white opaque plastic bucket, which was used as a holding cage between trials. The rats were given 10 discrimination trials per day; for half the rats the S+ was the rubber duck and for the other half it was the red buoy. For all trials, the rats were released from the same starting position facing the southern wall of the pool and given a maximum of 60 s to escape from the water onto the platform hidden below the S+. After escaping from the water, the rats were given an additional 10 s on the platform prior to being removed from the pool and placed in the holding cage. If the rats did not escape the water onto the platform within 60 s, then they were guided to the escape and given 10 s on it before being removed from the pool. The interval between trials was approximately 1 min. The location of the S+ was counterbalanced pseudo-randomly across trials within a session in order for the rats to learn that the S+ predicted escape from the water. This also ensured that spatial cues in the room were not predictive of the platform location and that directional swim strategies (e.g., swim right) would also be unsuccessful to perform the discrimination.

A trial was considered correct if the rat swam directly to the S+ and climbed onto the platform without entering the S− quadrant. A trial was considered incorrect if the rat’s entire torso passed the dividing wall into the S− quadrant, but the trial was permitted to continue until the escape platform was found or the maximum trial duration. Training continued until the rats reached a criterion of 80% (8 out of 10) correct trials on two consecutive days.

Additional learning–massed and distributed sessions

After reaching criterion, the rats received 50 additional trials either massed within five consecutive days (10 trials per day for 5 days; Massed) or distributed across 10 weeks (weekly session of five trials; Distributed). In the Massed condition, the rats received surgery 5–10 days following the last training day (Recent) or 10–11 weeks following the last training day (Remote). Including these two criterion-to-surgery intervals aimed to dissociate the possibility of time-dependent consolidation processes (Squire et al., 2001). Rats in the Distributed condition received surgery 5–10 days after the last day of training. Thus, the criterion-to-surgery in this condition was identical to that of the Massed-Remote condition (10–11 weeks). The rats were assigned to either Sham or HPC surgery. Importantly, the rats were matched across groups according to S+ version (rubber duck and the red buoy), the number of days to reach criterion, and their average performance on the overtraining trials. Figure 1B illustrates the experimental design.

Surgery

The rats were anesthetized with isoflurane (Janssen, Toronto, ON, United States) in 0.8 L/min oxygen (Benson Medical Industries, Markham, ON, United States) at 14.7 PSIA at 21°C and given an analgesic (Metacam, 0.02 ml; 5 mg/mL, s.c.; Boehringer Ingelheim). The rats were then placed in a stereotaxic instrument (Kopf Instruments, Tujunga, CA, United States) and an incision was made along the midline of the scalp, exposing the skull. Eight holes were drilled in the skull above each hemisphere and the lesions were made by infusing N-methyl-D-aspartic acid (NMDA) (7.5 μg/μL in 0.9% saline, Sigma Chemical, St. Louis, MO, United States) into the HPC at 10 injection sites bilaterally. Injections were made using a 30-gauge injection needle attached to a 10 μl Hamilton syringe via polyethylene tubing (PE-50). NMDA was injected into the HPC at a rate of 0.4 μl/min for 45 s at each site with the exception of the last two ventral sites. For these sites the injection lasted 60 s. Following injection, the needle remained in place for an additional 2 min to facilitate the dispersion of NMDA into the HPC. Coordinates of the injection sites and the amount of NMDA injected at each site are summarized in Table 1. Following the NMDA injections, the incision was sutured and the rats were given a prophylaxis (diazepam 0.2–0.6 ml; 10 mg/ml, i.p.; Sabex, Boucherville, QC, Canada) to reduce seizure activity. Sham surgery was identical with the exception that no damage was done to the skull or brain of the rats. All the rats received an analgesic (Metacam, Oral Suspension 0.1 ml; 1.5 mg/mL, p.o.; Boehringer Ingelheim) daily for 7 days following surgery and were given 10–15 days of recovery before retention testing.

TABLE 1

Table 1. Injection coordinates relative to bregma as well as volume of NMDA injected for complete NMDA lesions of the Hippocampus.

Post-surgical retention and retraining

After recovering from surgery, the rats were tested for retention by receiving 10 discrimination trials that followed the same procedures as described for acquisition. The rats that performed below 80% accuracy during retention testing were given additional daily sessions of 10 trials until they reached 80% accuracy.

Histology

Following the completion of behavioral testing, the rats received an overdose of sodium pentobarbital (0.3 ml; 320 mg/ml, i.p. Schering Inc., Pointe-Claire, QC, Canada) and were perfused intracardially with 200 ml of phosphate buffered saline followed by 200 ml of 4% paraformaldehyde. The brains were extracted and stored in 4% paraformaldehyde for 24 h and then transferred to a 30% sucrose/0.1% sodium azide solution for a minimum of 24 h before being sectioned with a freezing microtome (American Optical Corporation, Buffalo, NY, United States) at a thickness of 40 μm. Every 12th section of the HPC was mounted on gelatin-coated glass slides and stained with cresyl violet. The stained sections were examined using a light microscope (Nikon Eclipse 80i; Nikon Instruments Inc., Melville, NY, United States) connected to a 1600 × 1200 megapixel digital camera (MicroFire; Optronics, Fremont, CA, United States) providing a live feed to a Dell Precision Computer. The extent of HPC damage was quantified using unbiased/assumption-free principles and the Cavalieri point-counting method (Mouton, 2002). Specifically, using the Stereologer 2000 program (Stereology Resource Center Inc., Chester, MD, United States), at a magnification of 2×, a systematic sampling grid with an area per point of 0.05 mm² was randomly superimposed over each section. Points that contacted intact HPC cell fields (dentate gyrus and CA fields) were counted. The total number of points counted for each lesion brain was divided by the average number of points for five control rats (Mean = 635.3, SD = 55.49). This gave a proportion of remaining HPC tissue and the remainder of this was used as an estimate of HPC damage for each rat.

Results

Four rats were excluded from the experiment before surgery due to failure to reach criterion on the visual discrimination. After 230 trials (23 training days) these rats were still performing at chance on the discrimination, which is 11 days beyond the average for the other rats. Thus, 34 rats (11 Massed-Recent, 11 Massed-Remote, and 12 Distributed) remained in the experiment.

Histology

The average lesion size, as well as the smallest and largest lesion size, for each group is presented in Table 2 and photomicrographs of an average HPC lesion are found in Figure 2B. A one-way Analysis of Variance (ANOVA) found a statistically significant difference in the lesion sizes across groups, F(2, 13) = 12.14, p = 0.001. Post hoc tests revealed the lesions in the Massed-Remote group were significantly smaller than the lesions in the Massed-Recent or Distributed groups (ps < 0.05), whereas the lesions between the Massed-Recent and Distributed groups did not significantly differ in size.

TABLE 2

Table 2. Descriptive Statistics of the HPC lesion size in each group of Experiment 1.

FIGURE 2

Figure 2. (A) Mean (+SEM) number of days to reach the learning criterion pre-surgically for Sham and HPC rats in the massed/recent, massed/remote, and distributed conditions. No statistical differences were found across groups, suggesting equivalent learning of the visual discrimination prior to the rats being given their overtraining trials. (B) Photomicrographs of coronal brain sections of a lesion rat that sustained 88.1% damage to the HPC. This lesion is representative of the average lesion size for rats across all experiments. Note that the lesions were made after criterion learning and the overtraining trials.

Briefly, extensive neural death was found in all principle cell fields of the HPC (CA1–3, Dentate Gyrus; DG) of all lesion rats. In the Massed-Recent group, minimal sparing was found in the dorsal DG and CA3 of two rats and there was some sparing of the most posterior portion of the HPC in one rat. There was some damage extending into the posterior subiculum in three rats and the ventral subiculum in two rats. Additionally, one rat had minor unilateral damage to the entorhinal cortex. In the Massed-Remote group, very minimal sparing was found in the anterior portion of the HPC and considerably more sparing was found in the posterior portion. Specifically, the most posterior portion of the HPC was considerably spared in four rats. In the Distributed group, minimal sparing of the DG was observed bilaterally in three rats. There was some damage extending into the posterior subiculum bilaterally in two rats and unilaterally in one rat. Additionally, there was some minor damage to the entorhinal cortex unilaterally in one rat. In all groups, minor damage was sustained to the parietal cortex as a result of the cannulas being lowered and withdrawn during surgery.

Behavioral

Pre-operative training: reaching criterion

The average number of days to reach criterion pre-operatively across groups ranged between 11.33 and 15.5 days and is illustrated in Figure 2A. A 2 (surgery) × 3 (training condition) between-subjects ANOVA was performed to ensure that the groups were evenly matched on learning experience prior to surgery. No significant main effects for surgery group, F(1, 28) = 1.19, p = 0.29, or training condition, F(2, 28) = 1.12, p = 0.34, were found and there was no significant interaction between the two conditions, F(2, 28) = 0.12, p = 0.89. This suggests that, on average, the rats in all conditions were able to learn the visual discrimination within the same number of trials.

Post-operative testing: retention

The percentage of correct trials across the retention session (10 trials) is illustrated in Figure 3A. One-tailed one-sample t-tests were conducted to determine whether the rats in each condition swam to the S+ on more trials than would be expected by chance (50%). The sham rats in all groups performed significantly above chance (Massed-Remote: t(5) = 2.13, p < 0.05; Massed-Recent: t(5) = 10.23, p < 0.05; Distributed: t(5) = 8.03, p < 0.05), suggesting they remembered the visual discrimination. The HPC rats in all three conditions, however, did not perform significantly above chance (Massed-Remote: t(4) = 0.93, p = 0.20; Massed-Recent: t(4) = 0.356, p = 0.37; Distributed: t(5) = −1.083, p = 0.16), suggesting that the rats with HPC damage did not remember the visual discrimination.

FIGURE 3

Figure 3. (A) Mean (+SEM) percent of correct trials during the post-surgical visual discrimination retention test for Sham and HPC rats in the massed/recent, massed/remote, and distributed conditions. Each Sham group showed a statistically significant bias (above 50%) in swimming to the S+ (ps < 0.05), suggesting that the control rats remembered the discrimination. In contrast, none of the HPC groups showed a preference for selecting the S+ (ps < 0.05) and showed fewer correct trials than the control rats (p < 0.05). Thus, the HPC rats, regardless of training condition, suffered from retrograde amnesia and the additional 50 training trials beyond criterion learning did not enable the memory to become more resistant to HPC damage. (B) Mean (+SEM) number of days for HPC rats in the massed/recent, massed/remote, and distributed conditions to relearn the visual discrimination compared to their pre-surgical performance (pre-surgical data also shown in Figure 2A). The lesion rats in all three conditions relearned the task in significantly fewer days than it took to learn the task initially (p < 0.05), suggesting that there was some memory trace remaining in non-HPC systems.

A 2 (surgery) × 3 (training condition) between-subjects ANOVA revealed a main effect of surgery, indicating that, overall, the rats with HPC damage performed significantly worse than control rats, F(1, 28) = 19.90, p < 0.05. Neither the main effect for training condition, F(2, 28) = 0.24, p = 0.79, or interaction between the surgery and training conditions, F(2, 28) = 1.74, p = 0.19, were statistically significant, indicating that the rats with HPC damage performed significantly worse than control rats, regardless of training condition.

Post-operative testing: reacquisition

Figure 3B illustrates the average number of days it took the HPC rats in each training condition to reach criterion both pre- and post-operatively. One rat was unable to relearn the visual discrimination post-operatively and was thus excluded from the reacquisition analysis. After 15 days of retraining (11 days beyond the average for the other rats), this rat was still performing no better than chance and was swimming to the right quadrant on every trial. With the exception of this rat, all the HPC rats were able to relearn the visual discrimination post-operatively. A mixed factorial ANOVA with between-subjects variable condition (Massed-Remote, Massed-Recent, and Distributed) and within-subjects variable time-of-learning (Pre-operative, Post-operative) was conducted on number of days to criterion in the HPC rats and revealed only a main effect of time-of-learning, F(1, 30) = 85.22, p = 0.001, indicating that it took HPC rats significantly fewer days to relearn the visual discrimination post-operatively than it took them to initially learn the discrimination. There was no main effect for condition F(2, 30) = 0.52, p = 0.6, and no interaction between condition and time-of-learning F(2, 30) = 1.73, p = 0.19, indicating that this pattern was true regardless of training condition.

Discussion

The main objective of the current experiment was to assess whether additional learning beyond criterion and distributed over several weeks would make a visual discrimination memory more resistant to extensive HPC damage. During the retention test, the Sham rats showed a significant preference for the S+, suggesting they remembered the visual discrimination. In contrast, giving the rats 50 additional discrimination trials after reaching criterion, whether massed in 5 days or distributed over 10 weeks, did not facilitate the visual discrimination becoming independent of the HPC. In all instances, the HPC rats failed to show the same preference as the control rats, suggesting that HPC damage caused retrograde amnesia regardless of learning condition. These findings are consistent with two other studies demonstrating that the HPC is involved in long-term visual discrimination memory (Sutherland et al., 2001; Epp et al., 2008). Although there is evidence that overtraining can make a memory resistant to HPC damage in other tasks (Lehmann et al., 2009; Sutherland et al., 2010; Lehmann and McNamara, 2011), under the parameters of the present experiment, this did not extend to a visual discrimination memory. The second objective of the current experiment was to assess and control for the possibility of long-term memory systems consolidation, whereby the visual discrimination memory could have become independent of the HPC because of processes beyond the additional training. Despite extending the learning-to-lesion interval from 10 days (recent) to 10 weeks (remote), damaging the HPC had equal retrograde amnesic effects. These findings fail to support long-term systems consolidation views and this issue will be discussed in greater depth in the general discussion.

The HPC rats in the current experiment, despite suffering from retrograde amnesia, were able to quickly relearn the visual discrimination. This is consistent with evidence suggesting that HPC damage does not cause anterograde amnesia for this type of memory (Sutherland et al., 2001; Epp et al., 2008). Importantly, this also suggests that non-HPC systems can indeed support a visual discrimination memory. Despite continued distributed training beyond criterion, in the intact brain, the HPC seems to robustly overshadow and prevent the non-HPC systems from establishing a strong enough memory to independently support performance. It is possible that HPC overshadowing cannot be abated in this task, meaning that the non-HPC memory systems are unable to acquire a visual discrimination memory in the presence of the HPC. Alternatively, it is possible that the 50 trials beyond criterion in the current experiment were insufficient to strengthen the memory in the non-HPC memory systems and that even more trials may be required to mitigate the HPC overshadowing.

Experiment 2

The evidence suggesting a lack of anterograde amnesia for visual discriminations comes from studies in which the rats had pre-operative experience in the task and initial learning occurred with the HPC intact (Driscoll et al., 2005; Epp et al., 2008). Thus, it is unknown whether rats with complete HPC damage are able to quickly learn a visual discrimination task if they have had no prior experience. This may provide insight into how many additional trials are needed to mitigate HPC overshadowing. Accordingly, in a second experiment we assessed the number of trials experimentally naïve HPC-damaged rats required to learn a visual discrimination. To address this question, experimentally naive rats received either complete damage to the HPC or sham surgery before being trained on a visual discrimination up to criterion.