Positive Reinforcement Produced by
Electrical Stimulation of Septal Area and Other Regions of Rat Brain1

 

JAMES OLDS2 AND PETER MILNER, McGill University

 

Stimuli have eliciting and reinforcing functions. In studying the former, one concentrates on the responses which come after the stimulus. In studying the latter, one looks mainly at the responses which precede it. In its reinforcing capacity, a stimulus increases, decreases, or leaves unchanged the frequency of preceding responses, and accordingly it is called a reward, a punishment, or a neutral stimulus (cf. 16).

 

Previous studies using chronic implantation of electrodes have tended to focus on the eliciting functions of electrical stimuli delivered to the brain (2, 3, 4, 5, 7, 10, 12, 14). The present study, on the other hand, has been concerned with the reinforcing function of the electrical stimulation.3

 

METHOD

 

General

 

Stimulation was carried out by means of chronically implanted electrodes which did not interfere with the health or free behavior of Ss to any appreciable extent. The Ss were 15 male hooded rats, weighing approximately 250 gm. at the start of the experiment. Each S was tested in a Skinner box which delivered alternating current to the brain so long as a lever was depressed. The current was delivered over a loose lead, suspended from the ceiling, which connected the stimulator to the rat’s electrode. The Ss were given a total of 6 to 12 hr. of acquisition testing, and 1 to 2 hr. of extinction testing. During acquisition, the stimulator was turned on so that a response produced electrical stimulation; during extinction, the stimulator was turned off so that a response produced no electrical stimulation. Each S was given a percentage score denoting the proportion of his total acquisition time given to responding. This score could be compared with the animal’s extinction score to determine whether the stimulation had a positive, negative, or neutral reinforcing effect. After testing, the animal was sacrificed. Its brain was frozen, sectioned, stained, and examined microscopically to determine which structure of the brain had been stimulated. This permitted correlation of acquisition scores with anatomical structures.

 

Electrode Implantation

 

Electrodes are constructed by cementing a pair of enameled silver wires of 0.010-in, diameter into a Lucite block, as shown in Figure 1. The parts of the wires which penetrate the brain are cemented together to form a needle, and this is cut to the correct length to reach the desired structure in the brain. This length is determined from Krieg’s rat brain atlas (11) with slight modifications as found necessary by experience. The exposed cross section of the wire is the only part of the needle not insulated from the brain by enamel; stimulation therefore occurs only at the tip. Contact with the lead from the stimulator is made through two blobs of solder on the upper ends of the electrode wires; these blobs make contact with the jaws of an alligator clip which has been modified to insulate the two jaws from one another. A light, flexible hearing-aid lead connects the clip to the voltage source.

 

The operation of implantation is performed with the rat under Nembutal anesthesia (0.88 cc/Kg) and held in a JohnsonKrieg stereotaxic instrument (11) . A mid-line incision is made in the scalp and the skin held out of the way by muscle retractors. A small hole is drilled in the skull with a dental burr at the point indicated by the stereotaxic instrument for the structure it is desired to stimulate. The electrode, which is clamped into the needle carrier of the instrument, is lowered until the flange of the Lucite block rests firmly on the skull. Four screw holes are then drilled in the skull through four fixing holes in the flange, and the electrode, still clamped firmly in the instrument, is fastened to the skull with jeweler’s screws which exceed the diameter of the screw holes in the skull by 0.006 in. The electrode is then released from the clamp and the scalp wound closed with silk sutures. The skin is pulled tightly around the base of the Lucite block and kept well away from the contact plates. A recovery period of three days is allowed after the operation before testing.

 

 

Testing

 

The testing apparatus consisted of a large-levered Skinner box 11 in. long, 5 in. wide, and 12 in. high. The top was open to allow passage for the stimulating lead. The lever actuated a microswitch in the stimulating circuit so that when it was depressed, the rat received electrical stimulation. The current was obtained from the 60-cycle power line, through a step-down transformer, and was adjustable between 0 and 10 v. r.m.s. by means of a variable potentiometer. In the experiments described here the stimulation continued as long as the lever was pressed, though for some tests a time-delay switch was incorporated which cut the current off after a predetermined interval if the rat continued to hold the lever down. Responses were recorded automatically on paper strip.

 

On the fourth day after the operation rats were given a pretesting session of about an hour in the boxes. Each rat was placed in the box and on the lever by E with the stimulus set at 0.5 v. During the hour, stimulation voltage was varied to determine the threshold of a “just noticeable” effect on the rat’s behavior. If the animal did not respond regularly from the start, it was placed on the lever periodically (at about 5-min. intervals). Data collected on the first day were not used in later calculations. On subsequent days, Ss were placed in the box for about 3½ hr. a day; these were 3 hr. of acquisition and ½ hr. of extinction. During the former, the rats were allowed to stimulate themselves with a voltage which was just high enough to produce some noticeable response in the resting animal. As this threshold voltage fluctuated with the passage of time, E would make a determination of it every half hour, unless S was responding regularly. At the beginning of each acquisition period, and after each voltage test, the animal was placed on the lever once by E. During extinction periods, conditions were precisely the same except that a bar press produced no electrical stimulation. At the beginning of each extinction period, animals which were not responding regularly were placed on the lever once by E. At first, rats were tested in this way for four days, but as there appeared to be little difference between the results on different days, this period was reduced to three and then to two days for subsequent animals. Thus, the first rats had about 12 hr. of acquisition after pretesting whereas later rats had about 6 hr. However, in computing the scores in our table, we have used only the first 6 hr. of acquisition for all animals, so the scores are strictly comparable. In behavioral curves, we have shown the full 12 hr. of acquisition on the earlier animals so as to illustrate the stability of the behavior over time.

 

At no time during the experiment were the rats deprived of food or water, and no reinforcement was used except the electrical stimulus.

 

Animals were scored on the percentage of time which they time responding. Thus the electrical stimulus in the septal area has an effect which is apparently equivalent to that of a conventional primary reward as far as the maintenance of a lever-pressing response is concerned.

 

If we move outside the septal area, either in the direction of the caudate nucleus (across the lateral ventricle) or in the direction of the corpus callosum, we find acquisition scores drop abruptly to levels of from 4 to 6 percent. These are definitely indications of neutral (neither rewarding nor punishing) effects.

 

However, above the corpus callosum in the cingulate cortex we find an acquisition score of 37 percent. As the extinction score in this case was 9 percent, we may say that stimulation was rewarding.

 

 

 

At the thalamic level (section II of Fig. 2) we find a 36 percent acquisition score produced by an electrode placed again in the cingulate cortex, an 11 percent score produced by an electrode placed in the hippocampus, a 71 percent score produced by an electrode placed exactly in the mammillothalamic tract, and a zero percent score produced by an electrode placed in the medial lemniscus. The zero denotes negative reinforcement.

 

At the mid-brain level (section III of Fig. 2) there are two zero scores produced by electrodes which are in the posterior portion of the medial geniculate bodies; here again, the scores indicate a negative effect, as the corresponding extinction scores are 31 and 21 percent. There is an electrode deep in the medial, posterior teigmentum which produces a 2 percent score; this seems quite neutral, as the extinction score in this case is 1 percent. Finally, there is an electrode shown on this section which actually stands 1½ mm. anterior to the point where it is shown; it was between the red nucleus and the posterior commissure. It produced an acquisition score of 77 percent, but an extinction score of 81 percent. This must be a rewarding placement, but the high extinction score makes it difficult to interpret.

 

Behavior

 

We turn our attention briefly to the behavioral data produced by the more rewarding electrode placements.

 

The graph in Figure 3 is a smoothed cumulative response curve illustrating the rate of responding of rat No. 32 (the lowest- scoring septal area rat) during acquisition and extinction. The animal gave a total of slightly over 3000 responses in the 12 hr. of acquisition. When the current was turned on, the animal responded at a rate of 285 responses an hour; when the current was turned off, the rate fell close to zero.

 

The graph in Figure 4 gives similar data on rat No. 34 (the highest-scoring septal rat). The animal stimulated itself over 7500 times in 12 hr. Its average response rate during acquisition was 742 responses an hour; during extinction, practically zero.

 

 

Figure 5 presents an unsmoothed cumulative response curve for one day of responding for rat No. A—5. This is to illustrate in detail the degree of control exercised by the electrical reward stimulus. While this rat was actually bar pressing, it did so at 1920 responses an hour; that is, about one response for every 2 sec. During the first period of the day it responded regularly while on acquisition, extinguished very rapidly when the current was turned off, and reconditioned readily when the current was turned on again. At reconditioning points, E gave S one stimulus to show that the current was turned on again, but E did not place S on the lever. During longer periods of acquisition, S occasionally stopped responding for short periods, but in the long run S spent almost three-quarters of its acquisition time responding. During the long period of extinction at the end of the day, there was very little responding, but S could be brought back to the lever quite quickly if a stimulus was delivered to show that the current had been turned on again.

 

 

DISCUSSION

 

It is clear that electrical stimulation in certain parts of the brain, particularly the septal area, produces acquisition and extinction curves which compare favorably with those produced by a conventional primary reward. With other electrode placements, the stimulation appears to be neutral or punishing.

 

Because the rewarding effect has been produced maximally by electrical stimulation in the septal area, but also in lesser degrees in the mammillothalamic tract and cingulate cortex, we are led to speculate that a system of structures previously attributed to the rhinencephalon may provide the locus for the reward phenomenon. However, as localization studies which will map the whole brain with respect to the reward and punishment dimension are continuing, we will not discuss in detail the problem of locus. We will use the term “reinforcing structures” in further discussion as a general name for the septal area and other structures which produce the reward phenomenon.

 

 

To provide an adequate canvass of the possible explanations for the rewarding effect would require considerably more argument than could possibly fit within the confines of a research paper. We have decided, therefore, to rule out briefly the possibility that the implantation produces pain which is reduced by electrical stimulation of reinforcing structures, and to confine further discussion to suggestions of ways the phenomenon may provide a methodological basis for study of physiological mechanisms of reward.

 

The possibility that the implantation produces some painful “drive stimulus” which is alleviated by electrical stimulation of reinforcing structures does not comport with the facts which we have observed. If there were some chronic, painful drive state, it would be indicated by emotional signs in the animal’s daily behavior. Our Ss, from the first day after the operation, are normally quiet, nonaggressive; they eat regularly, sleep regularly, gain weight. There is no evidence in their behavior to support the postulation of chronic pain. Septal preparations which have lived healthy and normal lives for months after the operation have given excellent response rates.

 

As there is no evidence of a painful condition preceding the electrical stimulation, and as the animals are given free access to food and water at all times except while actually in the Skinner boxes, there is no explicitly manipulated drive to be reduced by electrical stimulation. Barring the possibility that stimulation of a reinforcing structure specifically inhibits the “residual drive” state of the animal, or the alternative possibility that the first electrical stimulus has noxious aftereffects which are reduced by a second one, we have some evidence here for a primary rewarding effect which is not associated with the reduction of a primary drive state. It is perhaps fair in a discussion to report the “clinical impression” of the Es that the phenomenon represents strong pursuit of a positive stimulus rather than escape from some negative condition.

 

Should the latter interpretation prove correct, we have perhaps located a system within the brain whose peculiar function is to produce a rewarding effect on behavior. The location of such a system puts us in a position to collect information that may lead to a decision among conflicting theories of reward. By physiological studies, for example, we may find that the reinforcing structures act selectively on sensory or motor areas of the cortex. This would have relevance to current S-S versus S-R controversies (8, 9, 13, 16).

 

Similarly, extirpation studies may show whether reinforcing structures have primarily a quieting or an activating effect on behavior; this would be relevant to activation versus negative feedback theories of reward (6, 13, 15, 17). A recent study by Brady and Nauta (1) already suggests that the septal areas is a quieting system, for its surgical removal produced an extremely active animal.

 

Such examples, we believe, make it reasonable to hope that the methodology reported here should have important consequences for physiological studies of mechanisms of reward.

 

SUMMARY

 

A preliminary study was made of rewarding effects produced by electrical stimulation of certain areas of the brain. In all cases rats were used and stimulation was by 60-cycle alternating current with voltages ranging from ½ to 5 v. Bipolar needle electrodes were permanently implanted at various points in the brain. Animals were tested in Skinner boxes where they could stimulate themselves by pressing a lever. They received no other reward than the electrical stimulus in the course of the experiments. The primary findings may be listed as follows: (a) There are numerous places in the lower centers of the brain where electrical stimulation is rewarding in the sense that the experimental animal will stimulate itself in these places frequently and regularly for long periods of time if permitted to do so. (b) It is possible to obtain these results from as far back as the tegmentum, and as far forward as the septal area; from as far down as the subthalamus, and as far up as the cingulate gyrus of the cortex. (c) There are also sites in the lower centers where the effect is just the opposite: animals do everything possible to avoid stimulation. And there are neutral sites: animals do nothing to obtain or to avoid stimulation. (d) The reward results are obtained more dependably with electrode placements in some areas than others, the septal area being the most dependable to date. (e) In septal area preparations, the control exercised over the animal’s behavior by means of this reward is extreme, possibly exceeding that exercised by any other reward previously used in animal experimentation.

 

The possibility that the reward results depended on some chronic painful consequences of the implantation operation was ruled out on the evidence that no physiological or behavioral signs of such pain could be found. The phenomenon was discussed as possibly laying a methodological foundation for a physiological study of the mechanisms of reward.

 

NOTES

 

1.	The research reported here was made possible by grants from the Rockefeller Foundation and the National Institute of Mental Health of the U.S. Public Health Service. The authors particularly wish to express their thanks to Professor D. 0. Hebb, who provided germinal ideas for the research and who backed it with enthusiastic encouragement as well as laboratory facilities and funds. The authors are also grateful to Miss Joann Feindel, who performed the histological reconstructions reported here.
2.	National Institute of Mental Health Postdoctorate Fellow of the U.S. Public Health Service.
3.	The present preliminary paper deals mainly with methods and behavioral results. A detailed report of the locus of positive, negative, and neutral reinforcing effects of electrical brain stimulation is being prepared by the first author.

 

The Effects of Amount of Reward and Distribution of Practice on Active and Inactive Memory Traces1

 

EDWARD L. WALKER, University of Michigan; and RYOJI MOTOYOSHI Kyoto University

 

 

Whenever a psychological event occurs such as a reinforced trial in a learning task, it seems highly likely that for a period of time after the trial there is an active trace process. After the disappearance of the active trace phase, there remains an inactive trace variously called permanent trace, permanent memory, or habit strength.

 

Few modern psychologists will deny the existence of an active trace process, but there is little agreement as to its nature and function. Walker (1958) has spelled out in some detail a mechanism or model for the active trace process. In this model the major function of the active trace is the laying down of permanent memory or habit strength. The active trace also produces a negative tendency against the recurrence of the act it represents. The process which produces this tendency has been referred to as action decrement, and the phenomenon, at least in a simple two-choice situation, has been referred to as alternation tendency.

 

Walker (1958) has argued, and it has been found empirically that some factors which operate to produce greater learning from a single trial will also operate to produce greater action decrement and thus a greater tendency to alternate. Walker (1956) has reported more alternation with a water reward for both alternatives than with no reward for either for thirsty animals. Walker and Paradise (1958) have reported a close relationship between the amount of alternation produced by a given stimulus factor and the rate of learning when these particular stimulus factors are the cues for learning. The present report contains one more demonstration of this principle in showing more alternation for a large reward than for a small reward.

 

It was also asserted by Walker (1958) that action decrement has biological utility to the extent to which it operates to produce a negative tendency to perform the same act. It thus offers a degree of protection of the active trace process from disturbance before it completes the process of laying down permanent memory. This characteristic seems to involve a paradox. The mechanism of action decrement protects the organism from further practice and thus from further improvement in performance.

 

This line of reasoning provides two possible explanations for the well-established efficacy of spaced over massed practice. If action decrement produces alternation, then some of the errors which occur in massed practice should not be attributed directly to less habit strength but should be attributed to a temporary tendency to alternate. If the action-decrement mechanism is a protective device, then immediate repetition of a correct choice should produce some deleterious effect upon the trace and result in lesser habit strength than might have accumulated from the same number of well-spaced trials. The present study was designed to try to isolate these two effects in a simple T-maze learning situation with rats.

 

To accomplish this objective, each of the two groups of animals tested for alternation with large and small rewards was further subdivided into Spaced and Massed Groups, although the massing in the latter groups was only partial. The two Spaced Groups had only one trial every 12 hr. in a simple I maze. The two Massed Groups had a pair of trials separated by 30 sec. every 24 hr. Assuming that the active trace process has disappeared within a 12-hr. span, we have the following situation: Performance on all trials for the Spaced Groups are a product of the permanent trace alone. The first trial of each day for the Massed Groups is also a product of the permanent trace. The second trial, following the first as it does by 30 sec., is the product of the active trace process (which is presumed to be affected also by habit strength), and performance of this trial should reflect the character of the trace. Furthermore, to the extent that such second trials produce a degree of interference of the trace of the first trial of each day, learning in the Massed Group should be slowed. The effects of such interference should be apparent in the performance on the first trial of the day, which is presumed to be a product of the permanent trace alone.

 

METHOD

 

Subjects

 

A total of 32 male albino rats was used. They were a part of a group of 41 animals obtained from Rockland Farms. Nine were eliminated for inactivity during preliminary training. They were approximately 120 days old at the beginning of the experiment.

 

Apparatus

 

The apparatus was a simple T maze. The walls were of wood and were 4 in. high. The floor and roof were of hardware cloth. The starting stem and goal arms were 4 in. wide and 18 in. long. Guillotine doors separated the start box from the starting stem and the goal box from the goal stems. The goal boxes were black, and the remainder of the maze was a gray.

 

Procedure

 

Preliminary training. The animals were placed on a fooddeprivation schedule 10 days before the start of the experiment proper. They were fed 6 gm. twice a day on a schedule which gave them food approximately 1 hr. after each experimental session during the experimental phase. During these 10 days they were handled daily and given a 15-min. period of exploration in a linear runway.

 

Amount of reward and action decrement. On Day 11 the animals were introduced to the experimental maze for the first time. For half the animals both sides were baited with one small pellet normally used in the Gerbrands-type Skinner box. For the other half both sides were baited with eight small pellets. Each animal was retained in the start box for 5 sec., permitted a free choice, retained in the goal box for 15 sec., and introduced to the start box for a second trial 30 sec. after removal from the goal box. This operation permitted observation of the frequency of alternation as a function of the amount of reward.

 

Learning phase. Beginning with Day 12 only one side of the maze was baited. For all animals the baited side was the side of first choice—thus the individually preferred side—and the amount of reward was either one or eight pellets as it had been for each animal in the alternation test phase. Half of each of the two groups which differed as to amount of reward was designated at random, with the restriction of equal Ns, as a Spaced Group and the other half as a Massed Group. A spaced animal had a trial every 12 hrs. A massed-group animal had a pair of trials each day separated by 30 sec., and thus each pair of trials was separated by approximately 24 hrs. Training was continued for 15 days. The animals were run between 7:00 and 10:00 A.M. and again between 7:00 and 10:00 P.M. To control for possible effects of time of day, the “experimental day” began in the morning for half the animals and in the evening for the other half.

 

RESULTS

 

Amount of Reward and Action Decrement

 

The hypothesis was that a large reward would produce more action decrement and thus more tendency to alternate than a small reward. Therefore, there should be more alternation after an 8-pellet reward than after a 1-pellet reward. The results may be seen in Table 1. The difference is in the predicted direction and is as large as could be expected from the theory. Yet the chi square value is 1.61, which is significant at approximately the .20 level. While not statistically significant, the finding constitutes another in a growing body of replications of the basic proposition that a factor which operates to produce greater learning on a single trial will operate to produce greater action decrement and, thus, a greater tendency to alternate.

 

 

 

Massed and Spaced Learning for Different Amounts of Reward

 

The results of the learning phase of the study appear in Figures 1 and 2. There is one graph for each of the four groups differing in amount of reward and in conditions of spacing of the trials. In each of the four graphs, the data for the first trial of the day are plotted separately with solid lines. The data for the second trial of the day are plotted in broken lines. Thus, all solid lines and the broken lines in the two graphs of spaced training represent reflections of the permanent trace or habit strength. The broken lines in the massed-training curves reflect the active trace of the first trial of the day, since in the massed-training case, the second trial followed the first by 30 sec., while in spaced training the second trial followed the first by 12 hr.

 

It will be noted that in the groups with the larger reward (8 pellets), both spaced-training curves and the first-trial curve in massed training appear as normal learning curves. The broken curve in the massed-training group, which is presumed to reflect the active trace process, appears quite atypical. The curves in the small-reward groups are much more variable, but in the last few days, especially in the massed-learning group, the plot of second-trial performance is distinctly inferior to first trial performance. Thus, on visual inspection there appears to be a real difference between performance which is a function of the permanent trace or habit strength and performance which is a function of the temporary active-trace process.

 

Table 2 presents a variety of numerical analyses relevant to the question of the efficacy of spaced training and the source of the differences between massed and spaced training. In each instance, the t test was used to test the significance of the differences. In both reward conditions the massed-training groups required a significantly greater number of trials to reach the criterion of errorless performance. In both cases the total number of errors was greater in the Massed Group than in the Spaced Group although in the 8-pellet groups the difference approached, but did not reach, the .05 level. However, there were no significant differences in the number of reinforcements to criterion between Spaced and Massed Groups. If genuine differences in the rates of learning between the massed and spaced conditions, are assumed, the question remains as to the extent to which these differences are attributable to simple alternation and the extent to which they represent significantly slower accumulation of permanent memory or habit strength in the massed-training condition.

 

 

 

 

If action decrement and the negative bias against repetition of an act is a protective device against interruption of the trace, then the massed condition should produce a greater tendency to alternate between the pair of trials in the day. Table 2 shows the mean number of times that a pair of trials within a day represented alternation. As may be seen, the number of alternation pairs is significantly higher in the massed-training groups as compared with the spaced-training groups. It is, therefore, clear that much of the apparent loss of learning efficiency occasioned by massing of trials can be attributed to the fact that massing produced alternation and thus an artifactual loss in efficiency of learning.

 

The question remains whether habit strength is being accumulated at a slower rate in the massed-training groups than in the spaced-training groups. If massing produces alternation, if the effect of alternation is to prevent consecutive occurrences of the same choice, and if it is a wholly effective mechanism, then there will be no instance of repetition of the same response on consecutive massed trials and there will be no difference in learning rate between massed and spaced training reflected in the solid curves in Figure 1, which presumably are relatively pure reflections of habit strength.

 

In Table 2 it can be seen that the number of instances in which the second trial of the pair in massed training represented the same choice as the first trial and, thus, offered an opportunity for trace interruption was in fact quite small, a mean of 2.625 such instances in the 1-pellet group and a mean of 1.375 instances in the 8-pellet group. It would be surprising if such a small number of opportunities for interference with the active trace process produced a significant slowing of the learning process as represented by the first-trial curves. If the first trials of the day are examined independently and the criterion of errorless performance applied to first trials only, it can be seen in Table 2 that the Massed Groups took a greater number of trials to criterion, but the differences are not statistically significant in either case. If the first-trial errors are counted, then the massed 1-pellet group shows a significantly greater number of first-trial errors (p <.02), but there is no difference between massed and spaced practice in the 8-pellet groups.

 

It can therefore be concluded that virtually all the efficacy of spaced training in this simple T-maze learning situation for rats can be attributed to alternation effects produced by massing and that there is little, if any, difference in the rate of accumulation of habit strength under the two conditions. In order to achieve a valid test of the hypothesis that a massed repetition of a particular choice interferes with the trace of the previous trial and thus decreases accumulated habit strength, it would be necessary to achieve a much larger N than the present study provides. This could be achieved simply by running an enormous number of animals, or possibly by massing more than two trials in a two-choice situation, thereby, enforcing repetition of a choice during the active-trace phase.

 

SUMMARY

 

A group of 32 male albino rats were given 10 days of preliminary handling and training in a straight runway under 11 hr. of food deprivation. On Day 11 they were given 2 trials, 30 sec. apart, in a simple, enclosed I maze, with half the animals receiving a 1-pellet reward and the other half receiving an 8-pellet reward as a test of the hypothesis that the larger reward would produce more action decrement and thus a greater tendency to alternate than the smaller reward. This expectation was confirmed.

 

The test of alternation tendency was followed by two learning trials per day for 15 days with the groups at the same reward level, 1 and 8 pellets, as in the alternation test. The side first chosen in the alternation test, thus the preferred side, was the correct side during learning. Each group was divided into a Spaced and a Massed Group during the learning phase of the experiment. The Spaced Group had one trial every 12 hr. The Massed Group had one pair of trials per day separated by an intertrial interval of 30 sec.

 

The expectation that the Massed Group would show a greater tendency to alternate was confirmed, and most of the apparent loss in learning efficiency in the massed condition may be attributed to alternation rather than to slower accumulation of habit strength.

 

NOTE

 

1. This study was carried out while R. Motoyoshi was a visiting scholar at the University of Michigan in the Spring of 1960. His stay in the United States was made possible by a grant from the Rockefeller Foundation to Kyoto University in Kyoto, Japan. The research was supported in part by a grant to the senior author from the Ford Foundation.

 

Sources of Experimental Amnesia1

 

DONALD J. LEWIS, University of Southern California

 

 

It is pointed out that the notion of consolidation is typically restricted to the fixation of learning and that the performance decrement labeled amnesia can be due to several processes subsequent to fixation on the total input-output memory chain. A review of selected experiments shows the likelihood that at least some experimental amnesias are not due to a failure of fixation. For example, when amnesia for an “old” memory is brought about by means of a single electroconvulsive shock (ECS) it is not likely that consolidation “interruption” is the cause. Also when memory returns following amnesic treatment the source must be sought in a performance variable. Other conditions are discussed in which amnesia does not occur even though the learning-ECS interval is extremely short (less than 1 second). Interpretations of these perplexing findings are suggested.

 

Following a learning experience, an experimental trauma to the head which produces unconsciousness will frequently result in an amnesia for that learning experience. The amnesia is indexed by a performance decrement in experimentally treated subjects as compared to controls who did not receive the trauma, and the amnesia may be partial or complete. A partial amnesia is present when the experimental animals show more learning than animals who did not have the learning experience at all but who still show less memory of the learning experience than animals who experience learning but have no amnesic treatment. Amnesia is defined, then to be a performance decrement, following a learning experience that is produced by trauma to the brain. This definition separates amnesia—an empirical event—from its most common explanation—the interference with consolidation.

 

This paper gives a review of some recent experimental studies of amnesia and relates the evidence to current theoretical interpretations; but before turning to the experimental data, the author will focus for a moment on the widely held concept of consolidation in the attempt to present a reasonably general consensus concerning its meaning.

 

CONSOLIDATION

 

Brief History

 

Muller and Pilzecker (1900), following in the tradition of Ebbinghaus, attempted to explain why a second list of nonsense syllables caused the forgetting of a first list. They called this forgetting “retroactive inhibition” and said that it occurred because the neural processes set up by the learning of the second list interfered with the perseverating neural processes established during the learning of the first list. Burnham (1903) is given credit for adding the concept of consolidation to perseveration and relating both to the retrograde amnesia (RA) produced by trauma to the head. Although others adopted these ideas, their acceptance was not universal. Lashley (1918) wrote,

After a review of the experimental literature on perseverative neural processes and consolidation Glickman (1961) concluded, “In the opinion of the writer, the over-all weight of evidence certainly favors the existence of some mechanism of consolidation … [p. 230].” Lewis and Maher (1965), on the other hand, reviewed studies using electroconvulsive shock (ECS) as a source of RA and concluded that the evidence supporting a consolidation process remained unconvincing. This interpretation was contested by McGaugh and Petrinovich (1966), but they were unable to convince Lewis and Maher (1966). McGaugh (1966) has presented a detailed argument favoring a consolidation process and his article, along with that of Glickman (1961), remains the most frequently cited authority for details of the consolidation process.

 

The present article considers once again some aspects of the consolidation process. Such a consideration seems warranted in view of its continuing importance in stimulating research and the influence it has had in the interpretation of data from a number of different fields.

 

PROPERTIES OF CONSOLIDATION

 

A review of the conceptual properties of consolidation can be brief, for they have been stated clearly by those most recently concerned with this notion. Two properties of consolidation seem to be held by all who have treated the subject and a third and fourth probably are held by a majority. The two universal properties have to do with (a) fixation and (b) the time-bound effect. The two properties that are widely but not universally held concern (c) the permanence of the amnesic disruption and (d) the number of memory stages. These are discussed in order.

 

Memory Fixation

 

The first point to be made is that consolidation is conceived to refer to memory fixation or registration; the two terms are frequently used interchangeably. A few quotations will make this point clear.

 

Fixation Is Time Bound

 

The above quotations, among a large number of others that could be chosen, indicate that consolidation refers to the fixation or the input phase of memory. They also substantiate the second point concerning consolidation: that it occurs over a significant period of time and that during this time the trace can be disrupted. Three further quotations should be sufficient on this point.

 

Disruption Is Permanent

 

Another thesis concerning consolidation is that disruption of the trace produces a permanent amnesia. This assumption is stated most forcefully in those studies concerned with recovery of memory following amnesia. Not everyone, however, holds to this assumption (see Deutsch & Deutsch, 1966), but the following recent quotations show that the assumption is not uncommon.

 

Memory Stages

 

A number of theorists, as can be seen from some of the quotations given above, assume several memory stages. The most common assumptions involve simply two stages: an early stage during which memory is susceptible to disruption and destruction, and a later one during which it is not. Others (Barondes & Cohen, 1966) require three stages. The different stages may run consecutively or concurrently. The important point to note about these stages is that they all refer to memory input processes; they all refer to memory fixation.

 

It should be clear that consolidation theory is a rather loose set of assumptions that are more or less widely held by most of those who do research on experimental amnesia. This paper is not an attempt to characterize any specific individual’s notion of consolidation, but rather to present those assumptions that are generally held. There are those who would disagree with any one of the assumptions, but also there are important theorists who would agree with all.

 

Memory Sequences and the Learning-Performance Distinction

 

As has been seen, consolidation theory places its emphasis on learning, on the fixation of memories. Input was similarly emphasized by many learning theorists until Tolman insisted on the distinction between learning and performance. The learning-performance distinction reminds us that it is possible for an organism to learn and to remember even though it does not show this learning in any immediate performance. Failure to show learning, after it has occurred, can be due to many causes. Among the most important of these are the lack of appropriate motivation, lack of appropriate incentive, and the effects of extinction and counterconditioning.

 

If one thinks in terms of the brain processes related to memory, then one can locate the source of amnesia at one or more of several points. First, of course, is registration. It may be that the amnesic agent (AA) prevents the initial registration of whatever brain process underlies that rather permanent change in behavior that is ascribed to learning. If a memory fails to register, there is no possibility of there ever being an expression of that memory. Also, it could be that the memory does register, but that for a period of time fixation is impermanent and susceptible to disruption. This is, of course, one of the assumptions of consolidation theory. But remember that there is more to memory than the fixation of a trace, either through one, or more than one, stage. It can be, for example, that memories are quickly and permanently fixed and that the AAs have their effect on storage2 processes subsequent to fixation. Presumably, memories can be coded and categorized in some fashion and these categories impart meaning to the memories and determine, at least in part, the manner in which they are stored. Storage is perhaps like a complicated filing system with memories categorized along different dimensions and sorted into appropriate bins. These bins are coded at the time of storage, but after the memory is firmly fixed, they serve only to facilitate retrieval. Tentatively assume that the familiarization procedure (Lewis, Miller, & Misanin, 1968a) manipulates a storage process, but this is discussed later. AAs may introduce so much noise into the system at the time of storage that these categorizations and filing processes are obscured. In a sense, the brain is prevented from organizing memories coherently, although there is no disruption of the actual memory fixation.

 

There are other processes still farther along in the memory sequence, past fixation and storage. It may be that memories are fixed, categorized, and stored, but that the AA produces its effects on various mechanisms of memory retrieval. A form of dissociation is one such mechanism that results in an interference with retrieval. It could be that the organism remembers and remembers well, but it no longer associates the memory with its original context and, therefore, the memory does not occur when it should. Or it could be that the organism has lost the motivation to express a certain memory. Or, it could be that various forms of suppression, competition, inhibition, which may be applied long after fixation and storage, actively prevent memories from recurring.

 

It should be clear, from this list of alternatives, which un doubtedly does not exhaust all of the possibilities, that it is possible for amnesia to be produced in a variety of ways and at one or more of the various points along the total memory input-output chain. It has been unfortunate, perhaps, that many investigators of experimental amnesia have concentrated their theoretical attention solely on fixation processes. They have tended to attribute the performance decrement that is due to the administration of an AA exclusively to an interference with consolidation during the fixation of a memory. Without denying the very real possibility that it is at the point of fixation that at least some amnesias are produced, it is essential that investigations be open to other possibilities, for the evidence is increasing that at least some amnesic memory failure is due to other causes than failure to consolidate.

 

Perhaps some theorists maintain that the term consolidation refers to the total memory sequence, but this is a very unanalytic use of the term and it is almost coterminous with the already available “memory.” If consolidation is to be used in this way many difficulties are presented, for a separate term for the fixation process will still be needed, and it is still sensible to inquire whether fixation occurs slowly or very rapidly.

 

Now this paper turns to a selective review of some studies on experimental amnesia which suggest that storage and retrieval processes may be involved. It is not the author’s intention to review exhaustively all of the studies in the area; their number is too vast for a paper of this sort. Nor is there a presentation of a theoretical tour de force proving the consolidation theory a failure. But the author points out some problems for consolidation theory and suggests some alternatives to it. The main emphasis is on those processes which the author feels are illuminated by data from his own laboratory.

 

LEARNING—AA INTERVAL

 

The magnitude of RA is a function of the time intervening between learning and the administration of the AA. If the interval is short the RA is more complete than if the interval is long, and there is some interval which is long enough so that no RA is produced. This statement summarizes the basic data underlying the first two assumptions of consolidation theory which we discussed earlier in this paper. The fact that RA is greatest when the AA is close to the point of learning affords the basic support for the inference that amnesia is due to interference with fixation. And the fact that RA decreases as the learning-AA interval increases is the basis for the assumption that the memory traces stabilize over time until they are no longer disruptible.

 

Using multiple ECSs, a number of experimenters have found behavioral deficits even when the temporal interval between learning and the AA was greater than 80 days. Stone and Bakhtiari (1956), for example, gave ECS 30 days after the last learning trial and 70 days after the first and still produced a behavioral deficit. Braun, Patton, and Barnes (1952) found a deficit even when ECS was given 63 days after the completion of a learning experience that had endured over 56 days. Many researchers believe that such learning-ECS intervals are too long to permit the assumption that interference with consolidation is the cause and that in these situations, at least, the behavioral deficit is due to other causes.

 

Many researchers, using only one ECS do not find amnesia at such long learning-ECS intervals. Heriot and Colman (1962), for example, used an operant chamber and a bar press which was followed by a strong footshock and then ECS at 1, 7, 26, 60, or 180 minutes. The single ECS disrupted behavior up to 60 minutes but not beyond. King (1965) obtained amnesia in a step- through situation with a 5-minute interval separating learning from ECS but not a 15-minute interval. These studies, among a host of others, show the effective amnesic interval to be a matter of minutes, or at most a very few hours. Chorover and Schiller (1965), however, with a step-down apparatus found ECS to be effective at 10 seconds, but not at 30 seconds. A gradient similar to the one reported by Chorover and Schiller (1965) has been found by Quartermain, Paolino, and Miller (1965), in a comparable situation, although their effective amnesic interval extended out to 30 seconds after learning. As a result of their studies, Chorover and Schiller opened the possibility that consolidation occurred quickly within a matter of seconds, and the behavioral deficits reported in many other studies must be due to other processes than failure to consolidate.

 

Replicating an experiment reported by Bures and Buresová(1963) in which RA was found even when a 24-hour interval separated learning from ECS used as an AA, Chorover and Schiller (1966) obtained the same results that the original experimenters did, but they also observed that the animals which had received only a single ECS seemed to show considerable fear and that they were also more active. From these observations Chorover and Schiller (1966) hypothesized that ECS was producing its effect not through destruction of the learning trace but because it broke up response inhibition. They reasoned that the foot- shock, used as the learning stimulus, produced a generalized fear response, indexed by response cessation. ECS served to disinhibit this inhibition of behavior and the animal returned to responding. Thus ECS was assumed to have its effects, in this situation and at long learning-ECS intervals, not on memory, but on response mechanisms.

 

In a clever series of experiments Chorover and Schiller (1966) showed the plausibility of this interpretation. In one experiment, they administered footshock as the animal entered into the smaller of two compartments, but instead of confining him in this compartment after the shock, as is typical, they allowed him to escape. Under these conditions, when escape was permitted, they found no amnesia even when an interval of only 1 minute separated the footshock from ECS. If, however, subjects were confined in the small compartment, given footshock, and not permitted to escape, then ECS did produce amnesia at the relatively long learning-ECS interval reported by Bures and Buresová (1963).

 

In another experiment, using a step-down apparatus, they gave ECS 2 minutes after the footshock and found no amnesia, supporting their previous finding that in this situation amnesia did not occur at intervals greater than 20 to 30 seconds. If, however, immediately following the reception of footshock in the step-down apparatus, subjects were confined in a small compartment and given intermittent footshock for 1 minute with ECS following another minute later, there was marked amnesia. Chorover and Schiller concluded that if their animals were allowed to escape from the place where footshock was received, they developed a discriminated avoidance response—an active response —which was not disruptible by ECS. The ECS did disrupt, however, a passive avoidance response which was a nondiscriminated fear response.

 

This interpretation of the effects of ECS on passive avoidance responding has a great deal of intellectual appeal. Unfortunately, there are data which may not be consistent with this point of view. A recent experiment by Kopp, Bohdanecky, and Jarvik (1966), using a two-compartment, “step-through” apparatus patterned for mice instead of rats, had ECS follow footshock at intervals of 5, 20, 80, 320 seconds, and 1 and 6 hours. Their animals were also allowed to escape from the small compartment where they had received the footshock, since this was considered crucial by Chorover and Schiller (1966). Another group received no ECS at all. The authors found a significant amnesia even at 6 hours. They also gave a tailshock to another group of mice outside the experimental apparatus. These animals showed no effect of tailshock. The conclusion the experimenters drew was that the experimental animals had learned a discriminated avoidance response in this situation as shown by the avoidance behavior of those shocked inside the compartment and not by those shocked outside. Also, even with a learned discriminated avoidance, amnesia appeared at a 6-hour learning-ECS interval.

 

That a discriminated avoidance was learned seems clearcut and at the same time it is not completely conclusive to the issue. The animals could well have discriminated the environment outside of the apparatus as different from the environment inside without discriminating any of the features inside the experimental apparatus. Therefore within the apparatus there would be a generalized fear response rather than a discriminated avoidance response. This would mean, following Chorover and Schiller (1966), that an ECS should have a long-term amnesic effect on the generalized conditioned emotional response inside the apparatus, and thus the data of Kopp et al. (1966) may pose no serious problem.

 

Lewis, Miller, Misanin, and Richter (1967) report data giving some difficulty to Chorover and Schiller’s (1966) interpretation. They used a step-down situation, but gave footshock on the raised platform rather than the floor of the apparatus. Thus the subjects learned to step off the platform rather than to avoid stepping off as is conventional in this situation. When ECS followed this discriminative, active, avoidance response, an effective amnesia was produced. The familiarization time given the animals in the Lewis et al. (A967) experiment, however, was not as great as that given by Chorover and Schiller.

 

Comparison of Amnesic Agents

 

An initial and plausible simplifying assumption seems to have been that the basic neurochemical events underlying learning were fairly similar for different kinds of learning experiences, and that different AAs would have fairly similar effects. This assumption, never strongly held and certainly not crucial to the consolidation point of view, has been found wanting in a number of aspects. It is now clear that the amount of amnesia is a function at least of the species used, the type of AA, its intensity and duration, and the learning situation, in addition to the learning-AA interval. Pearlman, Sharpless, and Jarvik (1961), for example, found amnesia due to ether and pentobarbital limited to a learning-AA interval of 10 minutes, but metrazol would produce amnesia at least 4 days following learning. They tended to think, however, that the response deficit produced by metrazol was due to other causes than the interference with consolidation.

 

Paolino, Quartermain, and Miller (1966) compared different durations of CO2 with different intensities of ECS. They found no RA when CO2 was administered for only 10 seconds. When CO2 was administered for 15 seconds, the RA gradient extended to 1 minute, and it extended to 4 minutes with 25 seconds of CO2. The effective RA gradient for ECS extended to only 30 seconds, and they found that intensity of the ECS current in milliamperes produced no differential effect. They concluded that there are qualitatively different steps in consolidation which are differently affected by different AAs.

 

McGaugh (1966) reported that he could get RA at a learning-ECS interval of 1 hour using a current enduring for 200 milliamperes. This interval could be extended to 3 hours with a current of 400 milliamperes, but no further extension of the interval occurred at 600 milliamperes. This finding was later confirmed by Alpern and McGaugh (1968), Miller (1968), and others. These findings clearly indicate that the duration of the AA can be a determiner of the amount of RA. It now seems, in fact, that some of the conflicts noted above are due either to the degree of original learning (Ray & Bivens, 1968), or to the intensity of the AA (Jarvik Sc Kopp, 1967; Lee-Teng, 1969; Miller, 1968; Ray & Barrett, 1969). The amnesia gradient can probably be described in part by a family of curves in which these variables figure as parameters. Even so, interference with consolidation is not the necessary explanation for these effects, since whatever effect an AA has will depend on its strength of the memory.

 

We may note that most, if not all, of the AAs also have convulsant properties. Many anesthetics produce seizures and are in fact convulsants. Trifluoroethyl, which Alpern and Kimble (1967) found so effective as an AA, is a convulsant at all ordinary temperatures. Add to their convulsant properties the fact that anesthetics can have an inhibitory effect on brain activity and we again have a pattern of neural spiking followed by electrical inhibition. This pattern seems to be part of the RA syndrome, and it has been investigators’ conjecture that inhibition plays a crucial role in producing RA. Such spiking and inhibition may, as most have conjectured, have their effect on the input (consolidation) of the memory trace, but because there are other plausible sources one must be wary of unvaryingly attributing every deficit to interference with consolidation.

 

Extremely Short Learning—AA Intervals

 

With a few notable exceptions, most of the studies considered so far have varied the learning-AA interval within moderate limits. Pearlman et al. (1961) found RA at 4 days with metrazol but attributed it to performance factors rather than to consolidation interference. Alpern and Kimble (1967) did the same with their performance decrement at 24 hours due to potentiated ether. At the short end (less than 10 seconds) of the continuum RA has always been found, and it is just this closeness of learning and the AA that has been the firmest basis for the inference that RA is due to a consolidation failure.

 

Lewis et al. (l968a), however, report a case in which RA either does not occur or is greatly reduced, even when the ECS follows immediately upon the termination of the learning experience. They administered a 5-second footshock as the rat stepped off the small raised platform. Then ECS was given at the instant that the footshock terminated. Under these standard conditions almost total RA was produced. Another group of animals was given considerable familiarization experience in the apparatus before the experimental treatment was initiated. They were placed in the apparatus and allowed to explore for a total of 5 minutes on each of three different occasions. On the fourth trial they received footshock immediately by ECS. These animals showed very little RA. In another experiment (Lewis et al., l968a) RA was demonstrated to be a function of the amount of familiarization. The important point is that very little, if any, RA was shown by familiarized animals for whom zero time intervened between learning and ECS.

 

If, however, the learning-ECS interval is defined as the time intervening between the onset of the footshock and the ECS, then a total interval of only 5 seconds was involved in this experiment, scarcely long enough for “long-term” consolidation to be complete. And yet the data show that it was. To explore this interval more completely, the duration of the footshock was manipulated in another experiment (Lewis, Miller, & Misanin, 1969) with groups of subjects receiving footshock for .5-, 2½-, and 5- seconds duration followed immediately by ECS. Those subjects that received prior familiarization (a total of 15 minutes) again showed very little RA. Thus, even with only .5 second intervening between learning and ECS no RA was produced. At least, if there was RA, it was not statistically reliable and was not retrograde over the intervals that were employed.

 

It is the author’s tentative view that fixation (learning) is almost instantaneous with information input (registration) and that ECS and other AAs have their effects on processes occurring subsequent to fixation. The use of a familiarized situation into which a relatively simple piece of new information is introduced allows one to study a memory trace in simple form. The attempt to reduce the learning situation to a simple one was, of course, the reason for the introduction of the one-trial apparatus. Familiarization further simplifies the situation in that the animal has already explored the situation sufficiently so that he knows where the corners are, the grids, the platform, and so forth.

 

The introduction of footshock represents, then, a relatively “simple” stimulus which is easily integrated into an existing cognitive complex. The subject does not need to spend much time in thought about what has happened and where it has happened and we find that fixation time approaches the speed of neural transmission. This indicates that at least in this simple situation consolidation time is reduced to just about the time it takes for information penetration.

 

It still may be true that there is an interference with consolidation that occurs for more complex information but another interpretation takes a different track. The author proposes that the familiarization operation has its effect on the coding of information for storage subsequent to an extremely rapid consolidation process. Familiarization gives the subject a ready tag, or code, for the input which can then be stored or filed in a manner that makes it quickly accessible. In a more complex situation, information will still penetrate and consolidate rapidly, but the subject may not have the use of this information. An AA will thus work on the storage process—on the “subprogram” that labels information for easy retrieval—and retrieval is made difficult. Of course, to make this interpretation meaningful, operations must be specified that will result in retrieval following amnesia. This problem will be discussed in the section on Memory Return.

 

Another study that manipulates the effective learning-AA interval in an instructive way is that of Miller, Misanin, and Lewis (1969) . The subjects drank in an operant chamber until their lick rate was fairly steady. Then one group was presented with a tone as the conditioned stimulus (CS) which was followed immediately by footshock. Three seconds following the termination of footshock they were given ECS. When the CS was presented again, subjects tended to continue to lick, an indication of RA. For another group, during training, the tone CS and footshock occurred just as with the first group, but the 3-second interval between footshock and ECS was filled by a blinking light. Significantly more RA was produced under this condition. The time interval between CS and ECS was identical in both cases, and there is no reason to suppose consolidation speed should differ. Nevertheless, the two conditions differed in the amount of RA that was produced. It is possible that filling the interval between CS and ECS with another stimulus somehow made the coding of the CS more difficult; successive compound stimuli require more trials to learn than do single stimuli. Much more work needs to be done on this problem, however, and on a complexity variable in general, before its properties are clear.

 

Long Learning—AA Intervals

 

Thus far the effects of ECS given at short intervals following learning have been discussed. Now some recent studies in which longer intervals have been examined are discussed. Schneider and Sherman (1968), using a situation which ordinarily does not produce RA at learning-ECS intervals greater than 2 or 3 minutes at the most, gave footshock followed 6 hours later by a second footshock and ECS 30 seconds thereafter, and they found RA. They also gave a first footshock, followed by another one 30 seconds later with ECS following an interval of 6 hours, but got no RA. RA at this long interval was a feature of the second footshock immediately preceding the ECS. They reasoned that footshock produces (a) learning and (b) arousal; and that ECS disrupts the second property of footshock and not the first. Thus ECS has its effect on the retrieval of a memory and not on its fixation.

 

A similar notion may be contained in the two-part theory of memory consolidation expressed by Ray and Bivens (1968). They found that the amount of RA was a function of the intensity of the footshock used to produce learning. With longer intervals between learning and ECS, RA could be produced only with decreasing footshock intensities. They conjectured that there may be a CS-UCS (unconditioned stimulus) connection based on simple contiguity which is not subject to disruption by ECS. The second component, perhaps solely a performance one, varies with footshock intensity and is susceptible to disruption.

 

In an experiment somewhat similar to that of Schneider and Sherman (1967), Misanin, Miller and Lewis (1968) produced RA with an interval of 24 hours separating learning from ECS. The response they worked with was a cessation of drinking brought about by pairing a CS with footshock while the subject was drinking. They showed that if the sequence was CS, fol lowed immediately by footshock, followed immediately by ECS, a great deal of RA occurred. If, however, 24 hours separated the CS and footshock from the ECS there was no RA. These are expected findings of the kind typically taken to support consolidation theory. However, RA could also be produced 24 hours following learning if the CS immediately preceded the ECS. The rationale that led to this experiment holds that ECS is an inhibitor (Lewis Sc Maher, 1965) for those processes which it follows closely. Thus if an old memory could be rather precisely reactivated and followed immediately by ECS, an amnesia producing inhibition should result. The prediction was confirmed. This study, as well as the one by Schneider and Sherman (1968), indicates that ECS inhibits those memory processes with which it is contiguous, independent of whether these memories are new or old.

 

Two stages of memory can be distinguished (Misanin et al., 1968) although it should be clear that these two stages have little to do with fixation. One stage of memory is that of the “specious present,” the memory that is in momentary awareness, that is being immediately used. This is the “active memory state.” The other kind of memory consists of all other memories that lie outside of awareness. This is the “inactive memory state.” It is investigators’ conjecture, not yet adequately tested, that memories in transition from one state to another are susceptible to inhibition by AAs, although it may be that memories in the active state, whether in transition or not, can be inhibited. This means that the time between learning and the AA is of importance only to the degree that the memory is in the active state, which it usually is when learning is occurring. To the extent that the subject turns to something else so that a memory “recedes” to the inactive state, a time-bound effect will appear until a point is reached at which RA will not occur. A great deal more research will be required to test the implications of this simple conjecture.

 

MEMORY RETURN

 

As has been noted, one common (Luttges Sc McGaugh, 1967) but not universal (Deutsch Sc Deutsch, 1966) assumption by consolidation theorists is that the memory trace is physically destroyed by amnesia. This is why experiments showing spontaneous recovery (Kohlenberg & Trabasso, 1968; Miller, 1968; Zinken & Miller, 1967) have such importance. At the moment, however, the preponderence of the evidence (Chevalier, 1965; Greenough, Schwitzgebel, & Fulcher, 1968; Herz & Peeke, 1967; Luttges & McGaugh, 1967) indicate that there is no spontaneous recovery. The issue is perhaps not yet resolved, and the importance of the topic warrants still further investigation.

 

If it is assumed that the structural change making up an “item” of memory occurs in a relatively short period of time, probably less than 1 second, then most amnesias are not due to an interference with a consolidating engram, but to an interference with memory access. This means that the memory remains but it is unavailable for expression. Lewis, Miller, & Misanin (1968b), in a series of experiments, attempted to return the memory to expression by using a simple reminder.

 

In their first experiment they used four groups of animals in a step-down situation. One group received footshock alone; one group received footshock followed by ECS; a third group received ECS alone; and a fourth group received no treatment. Tests on the following day showed that the footshock alone had produced a great amount of fear and that when this footshock was followed by ECS the fear was largely removed. The ECS alone had no effect. Four hours later in another room, in another apparatus, all of the animals were given a “reminder shock.” The purpose of the second footshock was to remind the animals of the first. Then 24 hours later they were returned to the original footshock situation for a retention test. The data showed a significant amount of memory return in the amnesic groups. This finding was confirmed in three other experiments. A similar finding was reported by Koppenaal, Jagoda, and Cruce (1967), and by Geller and Jarvik (1968), although both of these studies lacked the control groups necessary to make their data convincing. Memory return has also been reported by Flexner and Flexner (1968), who found amnesia due to an intracranial injection of puromycin and a return of memory following a later intracranial injection of saline. Insofar as the memory returned, the Flexners concluded, as the author has, that the amnesia was caused by an interference with memory expression and not its destruction.

 

The general conclusion to be drawn from these studies is that memories are tougher than has been previously claimed. They may be obscured, and their retrieval may be prevented, but they are present and appropriate means can be found to recover them.

 

THEORETICAL CONCLUSIONS

 

The main burden of this paper has been to point out alternatives to consolidation theory. The typical procedure in RA experiments has been to have an animal experience a simple learning task, wait varying time intervals, and administer an AA. Any resulting amnesia has commonly been attributed to the failure of the memory trace to fixate (Glickman, 1961; McGaugh 1966); to a failure in learning. But there is no necessary reason always to attribute a response decrement following learning and an AA to failure at the input end. There is a great deal going on subsequent to fixation as the learning-performance distinction has always made clear. And there is nothing in the design of amnesia experiments that demands that a response (output) failure be always attributable to a failure to fix the input.

 

Certainly the author is not denying now, nor before (Lewis & Maher, 1965), that perhaps some amnesias can be due to the failure of fixation. The author does, however, think that there are other alternatives that need to be explored before consolidation theory becomes firmly fixed as accepted fact, and believes that considerable evidence points to other processes along the total input-output channel as among the sources of amnesia. At least some of these sources lie at the retrieval end of the continuum, actively preventing a fixed and present memory from being expressed. Much of the evidence that leads us to this conclusion has been detailed here. Some of this evidence comes from studies in which no, or greatly reduced, amnesia is found even at a .5-second interval between learning and ECS. Other evidence shows that amnesia will occur at long intervals between learning and ECS if the memory is reactivated. Still other evidence suggests that the memory is still present even though the subject behaves in an amnesic fashion.

 

How do these amnesic retrieval failures occur? It was Lewis and Maher’s (1966) notion that ECS produced a massive inhibition which was conditioned to the stimuli of the situation in which the ECS occurred, or to them through a temporal stimulus generalization gradient. This latter would account for the “time-bound” effect. The studies to which this version of “inhibition theory” was applied involved both multiple learning trials and multiple ECSs, and these multiple ECSs permitted the conditioning of inhibition as experimentally verified by Adams and Lewis (1962). But such conditioning is not as likely in experiments which involve only one ECS. Inhibition does, nevertheless, occur, in the author’s conception, and it is “time bound” in the sense that it applies to those memory processes which it is closest to. Thus ECS can inhibit either new memories or old ones (Misanin et al. 1968). The age of the original learning trace is not important, but the temporal distance between the trace, of whatever age, and the ECS is. That ECS is basically an inhibition of traces rather than a destroyer of them is shown by the several reminder studies mentioned in this paper.

 

It is also a tentative conjecture that other AAs work in a similar inhibitory way to ECS. Most AAs have both a spiking component and a subsequent period of decreasing electrical activity (inhibition). This pattern is shown by anesthetics, ECS, spreading depression, and is reported for many protein synthesis inhibitors. Acetoxycycloheximide may be an exception to this generalization, but its effects are still too contradictory to permit a firm conclusion.

 

NOTES

 

1.	This paper was supported by National Institute of Mental Health Grants MH-07129 and MH-15861. The author wishes to thank Norman Richter for his assistance. Work was begun on this paper while the author was at Rutgers University, and all of the research reported here was performed in the Psychology Laboratory at Rutgers University.
2.	Please note that the term “storage” here refers exclusively to processes subsequent to fixation. In its typical usage it refers either to a combined memory location and fixation process or to fixation alone. The present usage conforms to that proposed by Melton (1963, p. 3).

 

 

 

Electrocortical Correlates of Stimulus Response
and Reinforcement

 

KARL H. PRIBRAM and D. N. SPINELLI, Stanford University School of Medicine;
and MARVIN C. KAMBACK, University of California

 

Three patterns of electrical response were identified in the occipital cortex of rhesus monkeys making a differential discrimination: an input pattern that identifies which stimulus has been displayed; a reinforcement pattern that indicates whether the outcome of the differential response was rewarded or in error; and an intention pattern that occurs prior to the response and predicts which response the monkey is about to make. Neither the reinforcement nor the intention pattern is present while the monkeys perform at chance; at this time, only the differences due to input can be distinguished. These results suggest that more than simple input transmission is occurring in the primary Visual mechanism. The influence of the experience of the organism is apparently encoded in the averaged electrical potentials recorded from the striate cortex.

 

To combine the techniques of electrophysiology with those of behavioral analysis of organisms subjected to cerebral ablations (1), we recorded potential changes that occur in the striate cortex of rhesus monkeys at various instants in a trial during which a visual discrimination is made. We placed a monkey in a restraining chair in front of, and within easy reach of, a 20- by 20-cm translucent panel split vertically down the center. Each half of the panel could be independently depressed; pressure closed a microswitch which sent a pulse to be recorded on magnetic tape (1.3 cm). The pulse also activated a circuit designed to deliver a food pellet into a cup placed under the panel whenever a correct response was made.

 

In front of the monkey, there was, attached to the chair, a small lever which, when pulled, activated a stimulus display. Thus there was reasonable assurance that the monkey would attend (make an observing response) to the display. Initially, during “shaping,” the display covered the entire translucent panel until the animal pressed it; but the duration of exposure was gradually shortened until it lasted for only 0.01 msec. This short duration—in essence a flash—ensured that a transient response occurred in the visual pathways. A transient response was chosen because the techniques of analysis of neuroelectric phenomena are considerably more advanced at present for transients than for changes in steady state. Two stimulus patterns (vertical stripes and a circle) equated for area were generated in a relatively random sequence by slides in a modified Kodak Carousel projector facing the back of the panel. The order of the display of the two patterns was determined in advance, so that the report of the response would be collated by the reinforcing circuit with the pattern displayed. This collation determined whether the response made was correct or incorrect. The occurrence of reinforcement was also recorded on the magnetic tape.

 

Once “shaped,” the monkeys were trained to press the right half of the panel whenever the circle was displayed and to press the left half of the panel whenever the vertical stripes were displayed. One monkey failed to learn the task (a difficult one because of the short duration of the display), and the other two monkeys reached a criterion of 85 percent correct in 200 consecutive trials after 1800 and 2800 trials. Two hundred trials were given daily 6 days a week.

 

The sequence of events that constitutes a trial is therefore as follows: (i) The monkey pulls a lever which initiates a pulse recorded on magnetic tape and (ii) turns on a stimulus display which lasts 0.01 msec. One of two patterns (vertical stripes or circle) is displayed; a pulse to indicate which display is flashed is reported to a reinforcing circuit and recorded on magnetic tape. (iii) After a variable period, the monkey depresses either the right or left half of the display panel. This pressure also initiates a pulse which is recorded on magnetic tape and reported to the reinforcing circuit. This circuit then delivers a food pellet whenever the vertical-stripe display is followed by a press of the left panel and whenever the circle display is followed by a press of the right side of the panel. Reinforcement is also recorded on the tape.

 

Recording of electrical activity from the brain was continuous over sample sessions of 200 trials and, of course, coincided with the recordings of the behavioral events. The sessions chosen were (i) at the beginning of training, after the monkey had been conditioned to press but while he was performing at chance, and (ii) after criterion performance was established. Recordings were made from 12 placements in the striate cortex. All were bipolar (depth of cortex to surface) from an insulated nichrome wire (300 in diameter). The electrical brain signals were adequately amplified before they were recorded on magnetic tape.

 

The tape-recorded results were processed on a small general-purpose digital computer (PDP-8). Brain activity was digitized by an A-to-D converter, and the results of conversion were stored on digital magnetic tape. We devised programs to average the digitized electrical activity forward in time from the onset of the stimulus display (the pulling of the lever) and from the response (the depression of either half of the display panel). Averages were also obtained by running the tape backward from the two time markers; these records indicated what was going on in the monkey’s brain just prior to his turning on the display and making the differential response. Programs were also developed to equate records obtained from unequal numbers of trials, so that correct and incorrect performances could be compared at criterion. Finally, routines to smooth the curves were adapted for photographing the results.

 

For each of the samples recorded, compilations were made of the brain activity (i) after stimulus display, (ii) preceding differential response, and (iii) after differential response. These compilations were then broken down into three categories: circle as opposed to vertical stripes, right as opposed to left panel, and correct as opposed to incorrect outcomes. Reliable differences (2) can be ascertained in the configuration of the brain record evoked by a stimulus display of 0.01 msec (3). In this instance, the circle generated a downward deflection; the two peaks of this deflection are more nearly equal than those generated by the vertical stripes. In the response to stripes, the amplitude of the second peak always exceeded the first. This difference did not change appreciably between the sample taken before learning occurred and the one taken at criterion performance.

 

The records obtained before and after differential response are essentially flat before learning of the problem takes place. No characteristic deflections occur constantly. At criterion, however, a marked difference routinely characterizes correct and incorrect outcomes; nonreinforcement is accompanied by a marked burst of activity in the record (approximately 40 cycles per second). At this time, a difference can also be seen in the brain recording made just prior to the differential response. From this difference, one can predict whether the monkey is going to press the right or the left side of the panel (regardless of whether this will prove to be correct or incorrect). Because this difference in the record prior to response was never observed when the monkey was performing at chance, differences in movement per se probably cannot account for differences in the neuroelectric response.

 

Three types of brain activity were discerned: an input pattern related to the stimulus display and present before as well as after learning, a reinforcement pattern indicating correct or incorrect outcome of the trial, and an intention pattern which occurs prior to the differential response once it has become meaningful.

 

All the brain patterns were not recorded from all 12 electrode placements in the striate cortex. From some, input patterns were obtained best; intention patterns were derived from others, and reinforcement patterns were best obtained from still others. Yet all these brain patterns did occur in the striate cortex—the end station of the anatomically homotopic tracts originating in the retina. These findings suggest that much more than simple input transmission occurs in the primary visual mechanism. At the striate cortex, the neuroelectric signals encode the influence of experience not only with respect to input differences, but also with respect to the organism’s intentions to respond and the outcome of behavior.

 

NOTES

 

1.	Neurobehavioral studies of the functions of the posterior “association” and frontolimbic formations of the forebrain have produced a wealth of evidence [K. H. Pribram, Can. Psychol. 7, 326 (1966); , in Advances in the Study of Behavior (Academic Press, New York, in press); _____, A. Ahumada, J. Hartog, L. Roos, in The Frontal Granular Cortex and Behavior (McGraw-Hill, New York, 1964), p. 28]. Despite this abundance, or perhaps in part as a result of it, a series of dilemmas in regard to interpretation has arisen. Part of the problem has been the limited repertoire of techniques, consisting primarily of ablation of neural tissue and experimental analysis of behavior before and after surgery. Interpretation (as opposed to description) of results is always ambiguous when there is only one mode of presentation of data. Recently, additional methods of neuro-behavioral study have become available, largely as the result of application of computer technology to analysis of electrophysio- logical recordings of the brain. The pioneering observations of D. B. Lindsley [in Electrical Stimulation of the Brain (Univ. of Texas Press, Austin, 1961), p. 3311 and of E. R. John and K. F. Killam [J. Nerv. Ment. Dis. 131, 183 (1960)] have paved the way for experiments in which behavior of the brain, as well as of the organism, during problem solving can be studied. Perhaps, by monitoring some internal as well as external responses of unoperated and lesioned primate solvers of problems, the choice among interpretations may be narrowed.
2.	E R. Spehlmann, Electroencephalogr. Clin. Neurophysiol. 19, 560, (1965);. R. John, R. N. Herrington, S. Sutton, Science 155, 1439 (1967); 5. Sutton, P. Tueting, J. Zubin, E. R. John, ibid., p., 1436.
3.	To determine quantitatively whether reliable differences existed between wave forms in the records, we used the following system of analysis of data. The record was averaged and displayed on the oscilloscope of a PDP-8 computer system. A vertical line was then positioned by a computer program at each inflection point of the displayed wave pattern; these lines then served as break points for analysis of data. The amplitude values of the raw data of the segment of the wave between two break points was then averaged and stored. By this device, the relative amplitude of comparable segments of different waves could be statistically compared by using Student’s t test. This method allows some estimate of the reliability (in the face of variability) of the difference between segments of the waveform, though, of course, it does not determine whether a total waveform is significantly different from another. When reported, a waveform differs from its control at least by P < .05.
4.	Supported by NIMH grant MH 12970 and research career award MH 15,214 to K.H.P.

 

 

Differential-Approach Tendencies Produced by Injection of RNA from Trained Rats

 

ALLAN L. JACOBSON, FRANK R. BABICH, SUZANNE BUBASH,
and ANN JACOBSON, University of California

 

 

Two groups of rats were trained in a Skinner box to approach the food cup when a discriminative stimulus (click or blinking light) was presented. Ribonucleic acid was extracted from the brains of these two groups of rats and injected into two groups of untrained rats. The untrained two groups then manifested a significant tendency (as compared with one another) to react differently to the two stimuli. On the average, the response appeared to be specific to the stimulus employed during training.

 

When RNA was extracted from the brains of rats trained to approach a food cup and was injected intraperitoneally into untrained rats, the rats so injected made Significantly more approaches to the food cup than did controls (rats injected with RNA from brains of untrained rats). (1) We now present evidence that this effect is specific rather than general. Our new experiment is Similar to our earlier one, except that we used two experimental groups, each trained to respond to a different discriminative stimulus.

 

Initially, 16 male Sprague-Dawley rats, aged 50 to 60 days, weighing 220 to 240 g. received magazine training in a Standard Grason-Stadler Skinner box; that is, they were trained to approach the food cup when a discriminative stimulus was presented. For eight rats, the discriminative stimulus was the distinct click (1) . Fotthe other eight rats, the discriminative stimulus was a blinking light. The latter stimulus was produced by blinking the 10-watt house light (within the Skinner box) three times in succession, the three blinks taking a total of approximately 1 second.

 

For the click-trained rats, magazine-training was accomplished in the same fashion as in our earlier study (1).

 

 

The blinking light proved to be a somewhat more difficult stimulus to establish as a signal, and accordingly, minor modifications in the magazine-training procedure were made. After early training, the rats did not respond to the blinking light but they did run to the cup upon hearing the slight noise produced by the pellet’s dropping into the cup. As training progressed, on more and more trials we withheld the pellet until the rat responded to the light alone. Thus, we essentially transferred discriminative control of the approach response from the noise of the descending pellet to the blinking light.

 

Each of these 16 rats was trained to the stimulus as described (1). By the end of training, each rat in both groups approached the food cup promptly and swiftly from any part of the box when the appropriate discriminative stimulus (click or blinking light) was presented, and rarely or never approached the cup in the absence of that stimulus.

 

Upon completion of the training, each rat was killed with ether, and the brain was taken out as quickly as possible. A cut was made on a line joining the superior colliculus to the rostral end of the pons. The tissue posterior to this cut was discarded, as was the tissue of the olfactory bulbs. RNA was extracted from the remaining tissue (1.3 g, average weight) and was dissolved in 2.0 ml of isotonic saline. Approximately 8 hours after extraction, the RNA from each of the rats, light-trained or click- trained, was injected intraperitoneally with a 1.9 cm 22-gauge needle into an adapted untrained rat (1). During adaptation, most animals initially made slight startle responses to the click, but few or no startle responses occurred to the blinking light. By the end of the adaptation series, no animal made any visible response to either the click or the blinking light.

 

Thus eight animals received RNA (RNA-C) from click- trained rats and eight received RNA (RNA-L) from light-trained rats. All were assigned code letters and tested “blind” (1).

 

A session of testing for a given animal consisted of placing that animal in the Skinner box, permitting one minute to elapse, and then delivering a series of ten stimuli (five clicks and five lights in a mixed order, as described below). The stimuli were spaced at least 30 seconds apart. Five such testing sessions were given (1). During the first three test sessions, the order of presentation of stimuli was LCCLLCCLLC; during the last two sessions, the order was CLLCCLLCCL. Each test animal thus received a total of 25 click and 25 light trials. At the beginning of testing, all rats had been deprived of food for approximately 24 hours. After the third test session, all rats were fed 4 to 5 g of Purina Lab Chow. The method of testing and the criterion of response were identical to those used in our first experiment.

 

A comparison of the two judges’ tallies revealed that they agreed on 790 out of 800 trials, that is, on 98.7 percent of the judgments.

 

Each rat received a difference score (C-L) which was obtained by subtracting number of responses to light (L) from number of responses to click (C) for that rat. The Mann-Whitney U test (2) was performed to test the null hypothesis that the C-L scores of the two groups did not differ from each other. The test indicated that the difference between groups injected with RNA-C and RNA-L was significant (P < .001, one-tailed test). The difference in response to click for the two groups was significant by a Mann-Whitney U test (P < .002, one-tailed test).

We may conclude on the basis of the statistical analysis of C-L scores that the two groups differed in their tendencies to react differentially to click and blinking light. The average difference score for the group injected with RNA-C is positive (3.9), whereas the average difference score for the group injected with RNA-L is negative (— 2.8), although the test used compares these scores with each other rather than with zero. A further conclusion is that RNA-C rats responded more to click than did the RNA-L rats. The differences between the groups in response tendencies may be attributed to the RNA preparation injected into the test rats, and hence presumably to the effects which original training produced upon the RNA of the donor animals. Since handling, nutrition, and adaptation to the Skinner box were matched for the two groups, the transfer effect cannot be attributed to these factors. Thus, the new results support our original finding that a response tendency can be transferred by RNA injection, and they further suggest that this effect is to a substantial extent specific rather than general.

 

 

 

Memory in Mammals:
Evidence for a System Involving

Nuclear Ribonucleic Acid*

 

D. J. ALBERT, Mental Health Research Institute, University of Michigan

 

 

The learning of an avoidance response was confined to one hemisphere in rats by starting cortical spreading depression in the other. Removing the medial (but not anterior or posterior) cortex of the trained hemisphere impairs retention of the learning. If the medial tissue is intraperitoneally injected back into the donor animal, there is savings in relearning with the untrained hemisphere. This savings seems to be specific to the previously learned task and the effect occurs only when the tissue from the trained hemisphere is removed after consolidation of the learning has progressed for several hours.

Ribonucleic acid molecules located in the nucleus of some group of cortical cells seem to mediate the savings effect. These results suggest that the effect of injecting the tissue from the trained cortex is to allow ribonucleic acid molecules which have coded the information about the learned response to migrate to the untrained hemisphere and function there as stored learning.

 

1. INTRODUCTION

 

The changes in the central nervous system which constitute learning have generally been thought to consist of changes limited to the synapse, which alter the functional relations between nerve cells and neural circuits [8, 13, 18]. There are, however, several lines of evidence which suggest that this conceptualization is at least an oversimplification. First, Morrell [14] has shown that learning may involve transcortical pathways connecting different parts of the cortex and has concluded that it is unlikely that the changed functional relations between parts of such a complex network could be accounted for by a simple chain of synaptic modifications. More recently, Albert [2, 3] has shown that the processes which occur during the consolidation, or fixation, of learning in the rat do not suggest that new morphological synaptic changes are being strengthened, but rather that more stable temporary changes, such as new macromolecular templates, are involved in forming the permanent retention mechanisms. Finally, experiments with planaria seem to show that learning in this more primitive species is coded in macromolecules [21].

 

The present experiments consider the possibility that macro molecular changes are central to the information storage system in the rat brain. The experiments parallel those done with planaria where evidence was found that learning could be transferred chemically by injecting ribonucleic acid from a trained planarian into an untrained planarian [21],. The design of the present experiments is also to transfer stored information chemically, but the transfer is not between animals, but instead, between one hemisphere and the other in the same animal. Such an intraanimal transfer of learning seemed particularly advantageous because it involves fewer biological assumptions. (Since these experiments were done, however, two reports [4, 7] have appeared describing what appears to be inter-animal chemical transfer of learning in rats, and these studies will be considered in the discussion.)

 

Restriction of the learning to one hemisphere is obtained by using cortical spreading depression to make the other hemisphere nonfunctional during training. A simple avoidance task is used which is known from previous experiments to be easily learned when only one hemisphere is functional [2].

 

The first step in these experiments is to locate the region in the trained hemisphere where the memory of the avoidance response is stored. It is found that a lesion in the medial (but not anterior or posterior) region of the cortex impairs retention, which suggests that the learning may be stored there. To further test the possibility that the learning is stored there and that a chemical storage mechanism is involved, a second experiment is done in which the tissue from the medial cortex is removed and intraperitoneally injected back into the donor animal. When these animals are tested on the avoidance task using the untrained hemisphere, they show savings in relearning while uninjected control animals do not. To demonstrate more clearly that the savings effect is due to the injection of molecules from the trained tissue which have coded the avoidance learning, several additional experiments are presented which show that the savings in relearning is specific to the previously learned task and that the molecules mediating the effect have the characteristics which existing evidence and intuitive reasoning require for a macromolecular information coding system.

 

2. METHOD

 

The experiments all follow a single schedule with minor variations. The subjects were naive, male hooded rats weighing 225—275 g from the Quebec Breeding Farm and Maxfield Animal Supply. Cannulas for starting spreading depression were first implanted in the antero-lateral parietal bone over each hemisphere. On the following day (day 1), the animals were trained on an avoidance task while one hemisphere had cortical spreading depression. It is known that spreading depression disturbs the functioning of the affected hemisphere and that with this task, the learning that occurs while one hemisphere is depressed is recorded mainly in the normal hemisphere [2].

 

On day 2, the animals were again subjected to surgery and this time the tissue from the medial region of the trained cortex (which seems to be the area storing the learning) was removed. The procedure differs at this point depending on the purpose of the experiment. If the tissue was to be injected back into the animal, it was kept ice cold as it was removed. When surgery was complete, any desired manipulations of the tissue were made and it was then injected intraperitoneally back into the donor animal.

 

On day 3, the trained hemisphere was depressed and the animal was tested for savings in relearning with the untrained hemisphere. If the injected tissue has no effect, the animal should require as many trials to relearn as in the original learning, but if there is4ome positive transfer of learning, the animal should relearn in fewer trials. Groups of 6—8 animals were usually sufficient to establish reliable estimates of the effect of the injected tissue.

 

Most of this procedure has been described in detail previously [2] and will be presented only generally here. Unilateral spreading depression was started and maintained for training and testing by 12 per cent potassium chloride (KCl) placed in the polyethylene cannula over one cortex (for details of the cannulas, see Albert [2]). The presence of spreading depression was confirmed by noting a hypesthesia of the body contralateral to the depressed hemisphere. The KCl was left in the cannula for as long as spreading depression was wanted on the cortex and then removed by flushing the cannula with sterile 0.9 per cent NaCl.

 

The avoidance apparatus was a box (10x36x 18 in. deep) with white and black halves separated by a sliding partition. The animal was first placed in the black side for 1 min, and then placed at the end of the white side, facing the white end. If the animal returned to the black side within 5 sec, the trial was counted as an avoidance; if not, the animal was shocked intermittently until it moved to the black side. There was a 1 min. intertrial interval during which the animal remained in the black compartment. This procedure was the same for both the first learning and the test for savings in relearning, except that in the savings test, the animals were purposely shocked as little as possible to avoid making them freeze.

Approximately 15 per cent of the animals were discarded in the course of training and retraining. The major reasons for discarding animals were: too rapid (less than 10 trials) or too slow (more than 30 trials) first learning; development of localized motor seizures which interfered with performance.

 

2.1 Removal and handling of tissue from the trained cortex

 

Tissue from the trained cortex was removed using a suction technique. A 10 ml syringe provided the suction and an attached 15 gauge curved hypodermic needle (whose diameter [1.75 mm] is about the thickness of the cortex) was worked gently into the brain tissue and, in places, under the skull to remove the cortical tissue. When the lesion was finished, gelfoam was slipped into the empty space.

 

Figure 1 shows the location and similarity in size of the anterior, medial, and posterior lesions. The lesions frequently removed the underlying white matter (corpus callosum) but seldom invaded other subcortical structures. The medial lesions gave the most important effects and these began 3—4 mm from the posterior cortical tip and 1—2 mm from the midline and extended about 4 mm anterior and laterally to within 1—2 mm of the rhinal fissure.

 

If the cortical tissue was to be injected without any experimental changes, it was placed in about 5 ml of ice cold 0.9 percent NaC1 as it was removed. Following surgery, it was broken into small pieces (though this was not necessary) by briefly manipulating it by hand in a teflon and glass homogenizer. The tissue was then intraperitoneally injected using a 10 ml syringe and an 18 gauge hypodermic needle, care being taken to avoid getting the fluid into the intestines.

 

 

In some experiments, the tissue was separated into sub- cellular components in order to determine the intracellular location of the molecules causing the savings in relearning. In this case, the tissue was placed in ice cold 0.25 molar (M) sucrose as it was removed and then homogenized using an electric motor drive. To separate the cell nuclei [9, 17] the homogenate was layered over 5 ml of 0.32 M sucrose in a 15 ml centrifuge tube and centrifuged for 10 min at about 800 g in an International Centrifuge bucket rotor (No. 241). To inject the isolated nuclei, the nuclear pellet on the bottom of the centrifuge tube was resuspended in 5 ml of ice cold 0.25 M sucrose.

 

To separate the soluble and solid portions of the nuclear fraction, the isolated nuclei were resuspended in distilled water (8 ml, 0°C) to break them open. The solid material was then sedimented by centrifuging at 100,000 g for 60 min in a Spinco Ultracentrifuge (No. 40 head).

 

A preliminary identification of the kind of molecules mediating the savings in relearning was established using enzymes which selectively break down certain kinds of molecules. Protein was broken down using trypsin. The trypsin used was sterile, lyophilized, 3 times recrystallized (Worthington Biochemicals). Ribonuclease was used to destroy ribonucleic acid and this was 5 times recrystallized (National Biochemical Company).

 

3. RESULTS

 

 

3.1 The Locus of Information Storage

 

The first step in this research is to locate the region in the trained hemisphere where the information about the avoidance response is stored. Previous experiments using the same task seemed to show that if the information is made to transfer from the trained to the untrained hemisphere (by giving the animal one avoidance trial with the trained and untrained hemispheres functional), the information is received in the medial region of the untrained cortex [2, 3]. It seemed reasonable, therefore, that in the trained hemisphere at least part of the learning is localized in the homotopic medial cortex.

 

To test this hypothesis, the animals were given a retention test after the medial cortex of the trained hemisphere was removed. The lesions were made on the day after avoidance training, and on the following day, the animals (N = 10) were tested for savings in relearning the avoidance task while the untrained hemisphere was nonfunctional with cortical spreading depression. Two control groups of 6 animals each were treated in the same way except that the lesions were in the anterior or posterior region of the trained cortex (see Fig. 1 for the locus of the lesions).

 

The animals with lesions in the medial cortex took almost as many trials to relearn (13.5 trials) as in the first learning. In contrast, the animals with anterior or posterior lesions showed significant (P < 0.05; all statistics are two-tailed rank tests) savings, reaching criterion in 4.2 and 5.1 trials, respectively. (There were no differences between groups in original learning in this or subsequent experiments.)

 

These results support the suggestion that some of the learning involved in the avoidance task is stored in the medial cortex of the trained hemisphere, but of course, they are not conclusive since the lesions may have interfered with recall rather than removing the stored information.

 

 

3.2 Injecting the Tiisue from the Trained Cortex

 

To test the possibility that the memory of the avoidance response is chemically coded, the tissue from the medial region of the trained cortex was removed and intraperitoneally injected back into the same animal. The line of reasoning behind this procedure was that when the tissue which is thought to contain the learning is injected, some of the molecules which code the learning may enter the blood vessels and be transported to and into the brain. The molecules are assumed to be structurally labelled with respect to their region of origin in the brain (medial cortex) so tharonce in the brain they are absorbed and alter only the similarly labelled cells in the homotopic region of the untrained cortex.

 

There are, of course, many assumptions in this argument, but those that make it most “biologically unlikely” concern the possibly of getting the molecules into the brain and to the right spot. The main obstacle to getting the molecules into the brain is the blood—brain barrier, and there is evidence that some kinds of large molecules (antibodies) can get through [19]. In addition, the lesion in the trained hemisphere will disturb the blood—brain barrier and should greatly increase the ease of penetration into the brain [11]. As to getting the molecules to the right spot, this seems to require the assumption that the molecules are labelled, but there is evidence that this kind of labelling can occur and that such molecules could migrate to specific regions of the brain. This is shown most clearly in an experiment where antibodies formed to a specific brain tissue (caudate nucleus) and injected intraventricularly, subsequently attacked and disturbed only this specific structure [12].

 

Two groups of 15 animals each were trained on the avoidance task with one hemisphere depressed and on the following day, the medial cortex of the trained hemisphere was removed. In the experimental group, this tissue was intraperitoneally injected back into the donor animal as soon as the surgery was complete; in the control group, there was no injection. Both groups of animals were tested the following day for savings in relearning with the trained hemisphere depressed.

 

The results are clear (Table 2). The animals injected with the tissue from the medial cortex relearning significantly faster (5.0 trials, P<0.01) than in the first learning. They also learned faster than the no-injection control group (18.8 trials, P<0.01) which was slightly slower than in the first learning. To control for the possibility that the faster learning in the experimental group was due to the injection of the cortical tissue but not specifically to the medial tissue where the learning is thought to be stored, another control group was injected with tissue from the posterior region of the trained cortex (Fig. 1). In this group, relearning required 12.1 trials, not significantly better than the first learning or the control group with no injection.

 

To control for experimenter bias as a source of error, the medial injection group and the no-injection control group were repeated blind. The results were the same (Table 2): the animals (N = 8) injected with the tissue from the medial cortex of the trained hemisphere showed significant savings in relearning (7.4 trials) while the no-injection group (N = 8) did not (13.1 trials, P<0.05).

 

The results support the hypothesis that at least part of the memory of the avoidance response is chemically stored in the medial cortex and that when the coded molecules are removed and injected back into the same animal, they can alter the untrained hemisphere so as to make relearning faster.

 

 

3.3 The Specificity of the Injection Effect

 

In order to show more conclusively that the effect of the injected tissue is to actually transfer chemically stored information rather than to simply facilitate learning or performance in general, a slightly different task was given on the relearning day. In this new task, an animal which had previously learned to avoid the white side was put down just over the middle of the box onto the white side (see Fig. 2). The animal now had a choice; it could avoid the shock by running to the black side as previously learned or by running into the far half of the white side. If the effect of the injected tissue from the trained cortex is not specific, about the same number of injected and uninjected animals should learn each response; but if, instead, the injected material actually carries information about the previous learning to the untrained hemisphere, the injected animals should show a greater tendency to relearn the previously acquired task.

 

 

One hemisphere was depressed and the animals trained to avoid the white side; on the following day, the tissue from the medial cortex of the trained hemisphere was removed and injected back into the animals of the experimental group but not those of the control group. The next day, the trained hemisphere was depressed and the animals were trained on the modified avoidance task in which they could avoid shock by moving to the black side or to the end of the white side (Fig. 2). The procedure was the same as in the first training with several exceptions: the animals were put down in the white side, half facing the black and half facing the white side; when the animal chose the white side, the partition separating the black and white sides was not replaced during the intertrial interval, and in order for the movement to be considered a response, the animal had to remain in that end for at least 15 sec; the animal was required to reach a criterion of 7 out of 8 consecutive trials to one side without shock instead of 9 out of 10.

 

The results (Table 3): the animals that were injected with the lesioned tissue all (8) moved to the side of the box (black) that had been safe during the first learning. In the uninjected group, only 3 out of the 8 animals chose this side; the others went to the end of the white side.

 

To control for any side preferences, the experiment was repeated with the white side safe during the first learning. The results were the same. Seven out of eight injected animals went to the white side compared to 2 out of 7 uninjected animals.

It is clear that the animals injected with the tissue from the trained cortex chose to relearn the previously learned task more frequently than the uninjected animals (P<0.01; combined groups). This seems to show that specific chemically coded information about the learned avoidance response is transferred to the untrained hemisphere by molecules in the injected tissue.

 

3.4 The Length of the "Consolidation" Period

 

Since the evidence up to this point strongly supported the conclusion that the memory of the avoidance response is chemically coded, additional experiments were done to relate the molecules causing the savings to existing evidence and intuitive requirements for a molecular information storage system. The first of these experiments considers the length of time following training required for the formation of the molecules mediating the savings. If these molecules are a part of the permanent information storage system in the brain, there should be a consolidation, or fixation, period of several hours after training before the synthesis is complete.

 

The animals were trained with one hemisphere depressed and then at varying times afterward, the medial cortex of the trained hemisphere was removed and injected back into the donor animal. Twenty-four hours later, the trained hemisphere was depressed and the animals were tested for savings in relearning.

 

 

The results are shown in Fig. 3. When the tissue from the trained cortex was removed at 2 or 4 hr after training, the injection did not give rise to savings in relearning (13.5 and 12.6 trials, respectively). By about 7 hr, however, there was significant savings (6.2 trials, P<0.05) and the amount was not significantly less than that of a 24-hr group (5.2 trials). (The 2- and 24-hr groups were run blind. The first learning scores for these groups were 13.8 and 15.0 trials, respectively; the relearning of the 24-hr group was significantly [P<0.05] faster than that of the 2-hr group.)

 

A period of over 4 hr following training is required for the formation of the molecules mediating the savings in relearning. This interval is somewhat longer than the 1—2 hr period that has generally been found for the consolidation of learning, but approaches very closely the 5—10 hr interval that Albert [3], using a new method, has obtained. The finding of an appropriate “consolidation period” for the savings effect clearly suggests that the molecules mediating the savings are a part of the permanent information storage system that is formed following learning.

 

3.5 Intracellular Origin of the Molecules

 

Mediating the Savings

 

One would expect that as with other biological functions, information storage in the nervous system would be associated with a particular place or structure. The grossest physical localization of the critical molecules that could be made with the present preparation would be with respect to the cortical layers and the kind of cell, neuron or glia, but these experiments pose numerous difficulties and so were postponed. An equally important experiment is the subcellular localization of the active molecules within the cortical tissue and this is a comparatively simple problem using the well-known technique of differential centrifugation [9, 17].

 

Following some preliminary experiments which indicated that the active molecules were in the cell nuclei, the following experiment was carried out blind. The animals were first trained; the medial region of the trained cortex was then removed, and the tissue homogenized in ice cold 0.25 M sucrose. The cell nuclei were separated from the homogenate by centrifugation (see Method) and suspended in 0.25 M sucrose. Half the animals (6) were injected with the nuclei and half with the rest of the cellular homogenate. The next day the trained hemisphere was depressed and the animals were tested for savings in relearning.

 

Table 4 shows the results. The animals injected with the isolated nuclei showed savings in relearning (8.4 trials) while the animals injected with the rest of the homogenate did not (12.3 trials, P<0.05).

 

 

A second experiment was done to determine whether the active molecules were in the solid or soluble portion of the cell nuclei. The method was to first isolate the nuclei and break them open by suspending them in distilled water. The solid material was then separated into a pellet by centrifuging at high speed (see Method). The pellet of solid material was suspended in distilled water and injected into half the animals and the soluble material (the supernatant) was injected into the others. The animals were tested for savings in relearning with the untrained hemisphere the following day.

 

Savings in relearning (Table 4) was shown by the group injected with the solid nuclear material (4.3 trials) but not the group injected with the soluble portion (11.4 trials, P<0.05).

 

The results seem clear: the critical molecules for the savings effect are located in the solid material of the nucleus, presumably the nucleolus. Although this conclusion may appear some what uncertain in view of expected cytoplasmic contamination of the isolated nuclei [9], the nuclear material, as isolated, appears to contain the full savings effect and the only subcellular particle which would be almost entirely present there are the nuclei.

---