An automated training paradigm reveals long-term memory in planarians and its persistence through head regeneration

One of the most interesting capabilities of living systems is processing information. Biological information in multicellular organisms comes in at least two types: spatial information, needed to create and maintain specific anatomical structures during embryogenesis and regeneration; and temporal information, abstracted and stored from environmental stimuli over time by the central nervous system (CNS). The intersection of these two fundamental processes has implications for basic neurobiology and engineering of the brain–body interface (Pfeifer and Gomez, 2009; Sampaio et al., 2001), for the synthetic bioengineering of cybernetic systems (Macía et al., 2012; Solé et al., 2007) and for the biomedicine of degenerative brain disease (Murre et al., 2001; Perry and Hodges, 1996). For example, what happens to the personality and mental content of an adult patient with decades of stored memories when their brain is repopulated by the descendants of implanted stem cells (Martino et al., 2011; van Velthoven et al., 2009)? Answering questions about the storage of information in dynamically remodeling biological tissues, and specifically about the dynamics of memory during brain regeneration, requires a tractable model system with both a robust CNS repair mechanism and the ability to learn and remember.

Free-living, planarian flatworms represent the ‘first’ class of organism to have a centralized brain with true synaptic transmission (Sarnat and Netsky, 1985), and share the majority of neurotransmitters that occur in vertebrate brains (Buttarelli et al., 2008). Planarians have primitive eyes and other sensory capabilities, including sensitivity to chemical gradients (Mason, 1975; Miyamoto and Shimozawa, 1985), vibration (Dessì Fulgheri and Messeri, 1973), electric fields (Brown and Ogden, 1968) and magnetic fields (Brown and Chow, 1975; Brown, 1966). Their sensory reception mechanisms are integrated by the worm's nervous system into a rich and complex set of behaviors as they navigate their environment.

Adult stem cell populations (neoblasts) underlie their remarkable regenerative abilities (Reddien and Sánchez Alvarado, 2004; Wagner et al., 2011), and whole worms can regenerate from only a small proportion of the adult worm: a cut off (or damaged) head is rebuilt perfectly within few days (Inoue et al., 2004; Umesono et al., 2011). Recently, planarians have become a popular molecular-genetic system for the investigation of the pathways that allow complex structures such as the head to be regenerated after damage (Aboobaker, 2011; Gentile et al., 2011; Lobo et al., 2012; Newmark and Sánchez Alvarado, 2002; Saló et al., 2009; Sánchez Alvarado, 2006). Thus, planarians are an ideal system in which to probe the dynamics of information stored in the CNS during massive remodeling and repair. While studies have identified several insect organisms in which memories survive the drastic reorganization of metamorphosis (Alloway, 1972; Blackiston et al., 2008; Hepper and Waldman, 1992; Ray, 1999; Sheiman and Tiras, 1996; Tully et al., 1994), planarians are a uniquely tractable system for molecular-biological analyses of large-scale regeneration of adult brains. But can they learn?

Nearly 55 years ago it was demonstrated that planarians could be trained to learn a task, and following amputation of the head, the animals regenerating from the original tail sections remembered the original training (Best, 1963; Corning and John, 1961; McConnell, 1965; McConnell et al., 1959). This stunning finding, suggesting that some memory may be stored outside of the head and imprinted on the new brain during regeneration, led to a myriad of subsequent associative learning studies (Cherkashin et al., 1966; Corning, 1966; Corning, 1967; McConnell, 1965; Morange, 2006; Sheiman and Tiras, 1996). The most common procedure was a classical conditioning protocol based on planarians' well-known photosensitivity (Dasheiff and Dasheiff, 2002; Inoue et al., 2004; Prados et al., 2013; Stephen, 1963). Acquired memories that could survive the process of head regeneration were demonstrated by measuring a direct display of a conditioned response or a faster learning rate (‘savings’) among worm fragments generated from head and tail pieces of previously trained planarians (McConnell et al., 1959).

While learning induced by classical conditioning could be attributed to sensory adaptation rather than consolidation and retrieval of ‘real, encoded’ memory (Halas et al., 1962; Halas et al., 1961), other studies showed that memories formed in more complex discrimination tasks, e.g. eliciting movement in a specific direction in a two-choice maze (Best, 1963; Corning and John, 1961; Corning, 1966; Corning, 1967; Corning et al., 1967; Humphries, 1961; McConnell, 1965; Roe, 1963) or learning to associate odorant cues (Wisenden and Millard, 2001), likewise survived regeneration of the head (Corning, 1966; Ernhart and Sherrick, 1959). The reports of persistent memory in an animal that had to regenerate its entire head (Corning, 1967) suggests approaches for investigating how information can be stored outside of the brain and imprinted on a newly regenerating brain – a truly fascinating possibility.

These remarkable discoveries have not had sufficient impact on the field and were largely abandoned because of practical difficulties inherent in manual experiments. While the basic findings were validated in some cases, they failed to be reproduced in others (Corning and Riccio, 1970; McConnell, 1966), and the whole line of research became abandoned (Rilling, 1996). While modern discoveries such as epigenetic modification (Arshavsky, 2006; Day and Sweatt, 2010; Ginsburg and Jablonka, 2009; Levenson and Sweatt, 2005) and RNAi (Smalheiser et al., 2001) now offer mechanistic explanations of some of the original results, the primary barrier to molecular-level investigations into the dynamics of memory during CNS regeneration has remained: the difficulty of developing a robust learning assay in planarians. Manual behavior experiments involve limited sample sizes, difficulties in precise reproduction of protocols, and lack of quantitative analysis (Corning and Riccio, 1970; Hartry et al., 1964; Morange, 2006; Travis, 1981). As a result of these difficulties, even the capacity for long-term memory planarians has been questioned (Abbott and Wong, 2008; Takeda et al., 2009; Travis, 1981).

As part of our investigations into information processing by dynamically organizing tissues, we have begun to develop automated approaches for eliciting learning and recall in planarians to overcome the problems inherent in manual methods (Nicolas et al., 2008; Oviedo et al., 2008b). We thus developed two platforms that allow automated, parallelized, quantitative and fully objective training and testing of planarians in a wide range of feedback paradigms (Blackiston et al., 2010; Hicks et al., 2006). The aim of this study was to find a learning paradigm that overcomes a number of problems encountered in previous attempts and establishes a modern platform for the use of regenerative planarians for the study of learning and memory.

Best and others (Best, 1963; Best and Rubinstein, 1962a; Koopowitz, 1970) showed that planarians and marine flatworms that had been fed in a familiar environment will start to eat more quickly than naïve worms that never been exposed to the feeding arena before. As in prior studies, their manually performed experiments contained small sample sizes and limited controls (Davenport and Best, 1962; Dufort, 1962), and it appears that there have been no later attempts to use or improve this non-punishing paradigm. Here, we modified this environmental familiarization approach, adapting it to use with a textured substrate (to provide clear haptic cues to the animals) and an automated behavior analysis system (Blackiston et al., 2010). Our protocol minimizes bias caused by manual procedures, allows an unprecedented level of quantitative rigor in behavioral analysis, and applies the procedure to a large sample size in a relatively short time frame. Additionally, in contrast to Best and Rubinstein's protocol, our procedure checks for long-term memory, several days after the training ended. Our results support the findings of Best and Rubinstein, and show a statistically significant shorter feeding delay for the familiarized worms compared with unfamiliarized worms. Most importantly, the memory survives long enough to allow for regeneration after amputation, and indeed we show that memory traces survive entire brain regeneration in a ‘savings’ paradigm. This simple and promising approach opens great opportunities for the use of planarians as model organisms to understand how specific memories survive large-scale regeneration of neural tissues.

For training and testing we used a custom-made fully automated training apparatus (ATA) (Blackiston et al., 2010; Blackiston and Levin, 2012) (Fig. 1A, Fig. 2L,M), which minimized bias caused by manual procedures and facilitated the training and testing of large numbers of control and experimental worms simultaneously under the same conditions, including time of day, temperature and type of arena. However, we settled on a paradigm that requires path tracking of the animals (Fig. 1B) but no complex training algorithm with instantaneous feedback (light or shock) to each animal. Therefore, this protocol could be implemented with any off-the-shelf system capable of multiple video tracking (Marechal et al., 2004; Noldus et al., 2001).

The ATA ‘familiarized’ chamber environment contained a Petri dish with rough-textured floor surrounded by the ATA electrode walls (Fig. 2). Rough-textured Petri dishes were made from commercially available polystyrene 15×60 mm Petri dishes (0875713A, Fisher Scientific, Waltham, MA, USA), and altered by laser etching (versaLASER VL-300, Universal Laser Systems, Scottsdale, AZ, USA). The laser cuts the circles to a depth of 0.2 mm below the level of the dish's floor, but the displaced melted polystyrene also builds up around each circle to a height of ~0.05 mm above the floor of the dish. The pattern (Fig. 2N) is made up of circles drawn at 1.4 mm in diameter and spaced 2.15 mm at their centers. As cut, the outer diameter of each circle ends up being closer to 1.5 and 1.2 mm inner diameter (the trough that the laser cuts for each circle is ~0.3 mm wide).

All planarians used in the present study were Dugesia japonica Ichikawa & Kawakatsu 1964. After examining three planarian species: Dugesia japonica, Dugesia dorotocephala and Schmidtea mediterranea, we found Dugesia japonica to be the most suitable for this project. It has remarkable regenerating capabilities, high tolerance for training and dissection procedures, and is very active. Before experiments, planarian colonies were stored in rectangular plastic containers, filled with Poland Springs natural spring water (Oviedo et al., 2008a). Dugesia japonica has a high tendency to undergo spontaneously fission. In order to prevent spontaneous fission and allow worms to reach a suitable size for the experiment (1–1.5 cm), containers were stored in an incubator at 10°C in continuous darkness (Morita and Best, 1984) and fed once or twice a week with organic beef liver.

In addition to suppressing fission, keeping the worms in darkness has been reported to enhance negative phototaxis (McConnell, 1965) (an important feature for the testing procedure). Worms were kept in continuous darkness during the entire experimental period except for brief periods during water changes and transfers between the experimental environment and their resting Petri dish/well plate. Planarians are more active and display a longer exploration phase when kept at 18°C (as compared with 10°C). The experimental room temperature was also kept at 18°C. Therefore, during the experimental period the worms were held in an incubator at 18°C. The tails' regeneration rate is also higher at 18°C compared with 10°C, allowing testing of the headless fragments of worms after only 10 days after decapitation (Fig. 4). Culturing the worms at high density was also found to be effective in suppressing spontaneous fission (Best et al., 1969). Thus, the worms were held in groups at high density (~12 worms per 2 ml water). This high density required water to be changed every day.

Every morning, during the training phase, the experimental apparatus was cleaned and the water was changed. The worms were taken out of the ATA and placed in Petri dishes with fresh water in the dark for the cleaning period. The familiarized groups were placed in a dish with a rough textured floor and the unfamiliarized groups were placed into standard Petri dishes. Rough-textured and standard Petri dishes were reused during the training, after being thoroughly cleaned with Kimwipes soaked with 70% ethanol, and positionally randomized between trials. The ATA electrodes, used as walls for the ‘familiar’ environment, were also cleaned with Kimwipes soaked with 70% ethanol. At the end of the cleaning procedure the worms were placed back into their experimental environments. In order to suppress fission, the experimental environment was filled with low water levels (~12 worms per 2 ml water) to maintain high density of animals. During the testing sessions, the experimental apparatus (ATA electrodes and dishes) was cleaned between every testing trial. For handling of all worms, we used a plastic transfer pipette with the tip cut off to make a slightly larger opening. During the training, separate pipettes were used for the familiarized and unfamiliarized groups.

Groups of 20–40 experimental worms were placed in an individual ATA chamber (while testing was done on individual animals, familiarization proceeded in groups). The ATA chamber environment contained a Petri dish with rough-textured floor surrounded by the ATA electrode walls (Fig. 2A). The training period lasted 10–11 consecutive days. The chambers were filled with water (~12 worms per 2 ml water) and the lids were closed for darkness. Unfamiliarized (control) worms went through the same procedure, simultaneously with the familiarized (experimental) group, but were placed in the ATA in a standard Petri dish (Fig. 2B). Every morning during the training phase, the worms were taken out of the ATA for water change and cleaning. Before being inserted back into the chambers, the worms were inspected and tail fragments caused by spontaneous fissions were extracted. After a 10 day familiarization period, the worms were taken out of the ATA and divided into smaller groups and were kept in 12-well plates [part number 665102, hydrophobic surface (no treatment); Greiner Bio-One, Frickenhausen, Germany] until testing (12 worms in a well filled with 2 ml water; Fig. 2E). The water in the wells was changed every day. Worms for regeneration experiments were kept in a Petri dish for a 24 h rest phase before dissection and division into smaller groups in small wells (to allow all eaten food to be digested before dissection).

Worms were fed throughout the training period, in order to suppress fissioning and eliminate the possibility of differential starvation levels among worms. The worms were fed in the ATA for 1 h, with one to two small drops of liver (less than what they are capable of consuming; Fig. 2C,D). Feeding took place in the morning after every third day of familiarization training (days 1, 4, 7 and 10). Just before feeding, chambers were filled with an additional ~10 ml of water. On the last morning of familiarization training (day 10), the worms were fed intensively with one to two drops of liver every 20 min, until satiety (revealed by the last drop of liver remaining intact). This procedure ‘synchronizes’ the hunger level of the worms that were tested 4 days later, and suppresses fissioning of the worms during a longer resting phase before testing. In addition, this feeding protocol is designed to create a positive association with the experimental environment. Worms that were tested 11–15 days after the end of training were fed again one to two times before the test.

The ATA contains 12 identical chambers (Fig. 1A). During each testing trial, six familiarized and six unfamiliarized worms were tested simultaneously, each worm in its own individual chamber. All chambers contained a rough textured floor (a separate set of dishes from those used for the training) surrounded by the ATA electrode walls (Fig. 2J,K). A very small amount of liver was spread with a fine paintbrush on a small area of the roughened dishes (Fig. 2J,K,O), and was allowed to dry for ~5 min before being placed in the ATA and filled with 11 ml of water. In the absence of food, worms prefer to stay on the edge of the dish. Therefore, the liver was applied away from the arena wall (Fig. 2J) so that familiarized worms would be more willing to leave the edge and move toward the center of the dish (Fig. 2P). The worms were inserted into the ATA chambers with a plastic transfer pipette, in alternating order, starting with the unfamiliarized. The worms were placed in the chambers, opposite the liver spot. Worm transfer for all chambers averaged <1 min. After all the 12 worms were inside the chamber, the lids were closed and the tracking was initiated.

To identify feeding, we capitalized upon the planarians' strong negative phototaxis (Inoue et al., 2004). Because the worms generally avoid illuminated areas, the quadrant with the spot of liver was illuminated with a strong blue LED light (Azuma et al., 1994; Brown et al., 1968) (Fig. 2L); thus, no worm would stay in this quadrant unless its desire for the liver overcame its natural light aversion (Fig. 2P). As an indication of feeding, we measured how long it took the worms to reach the criterion of three consecutive minutes in the illuminated quadrant containing the liver spot. Any worms that did not reach criterion within 60 min (e.g. never attempted to eat the liver), as well as worms that showed evidence of any health issue such as injuries caused by the transfer pipette or worms that were in the process of fissioning, were not included in the results.

At the end of each testing trial, the worms were inspected individually under a dissection microscope for general health, injuries caused by the transfer pipette, fission, lesions or incomplete head regeneration in the case of the headless fragment worms. In order to avoid possible interference from moving worms for testing in sequential groups, during the evening before testing, worms were divided into two groups of six familiarized and six unfamiliarized worms and each group was placed in a separate well of a 12-well plate, filled with 1 ml of water (Fig. 2I). As in the experimental period, plates were placed in the dark at 18°C until the beginning of the test at the next day.

Worms were decapitated 24 h after the final feeding, which occurred at the end of the familiarization session. So that no brain remained, the worms were decapitated at the point between the auricles and the anterior side of the pharynx (Fig. 2F). Headless fragments were kept in groups of 12 worms in separate wells of the 12-well plates in 2 ml of water (Fig. 2E) in a dark incubator at 18°C. As with the intact worms, water was changed every day. After 7 days of regenerating at 18°C, the headless fragments were capable of eating (Fig. 4). Seven to nine days after decapitation, the regenerated worms were fed to satiety. Three to four days after feeding the worms were tested for recall. The worms were fed a second time in cases when the duration between the first feeding to the recall test was longer than 3–4 days. For example, worms that were tested at day 13 after decapitation were fed at day 7 and then again at day 9 or 10 after decapitation.

In contrast to the headless fragments' regular protocol, where the feeding took place in the worms' home wells, in the savings protocol, the worms were fed in the familiarization arena. Seven to nine days after decapitation, groups of both familiarized and unfamiliarized regenerated worms were inserted into the ATA's chambers with the surrounding electrode surfaces and the rough floor (the familiarization arena; Fig. 2H). After a 30 min exploration phase, drops of liver were placed in the chamber and the worms were allowed to eat until satiety. At the end of the session, the worms were placed back in the multi-well plate (~12 worms per well in 2 ml water; Fig. 2E). In the evening, 3 days after the savings session, the worms were divided into groups of six familiarized and six unfamiliarized worms (Fig. 2I) and placed back in the dark at 18°C until the beginning of the test the next day, 4 days after the savings session.

The ATA's tracking log files were converted to an Excel file for data analysis. Because the delay values were not normally distributed (Kolmogorov–Smirnov test), we used the nonparametric Mann–Whitney U-test to evaluate statistical significance (Bevins et al., 2001). Fisher's exact test was applied to determine statistical significance of the total number of worms that reach criterion in less than 8 min. Tests were one-tailed because the direction was predicted in advance based on the previous work of Best and Rubinstein (Best and Rubinstein, 1962a). To check for any mobility impairment that might be responsible for behavior differences between the familiarized and unfamiliarized worms, the average movement rate (pixels s⁻¹) was calculated for the first minute, when the majority of worms were still engaged in exploration behavior.

Worms were familiarized to the automated behavior analysis platform (ATA) chambers as described in the Materials and methods, and then tracked by the ATA (Fig. 1). The retrieval test for the familiar environment took place 4–15 days after the end of the 10 day familiarization period, during which the familiarized worms were kept and fed in ATA chambers in Petri dishes with a rough-bottomed surface (Fig. 2C). The ‘unfamiliarized’ group was also kept and fed in the ATA but in a standard, smooth-bottomed Petri dish (Fig. 2D). During each test session, six familiarized worms and six ‘unfamiliarized’ control worms were placed individually in the ATA chambers with a rough floor (the familiar environment). A small area of the dish was covered with liver (Fig. 2J,O) and a strong blue light illuminated the quadrant with the liver stain (Fig. 2L). As indication of feeding, we measured how long it took for the worms to reach the criterion of three consecutive minutes spent in the illuminated quadrant near the liver. The testing trials lasted 60 min. To rule out general physical condition differences between the worms, we checked their movement rate during the first minute, a time period while most of the worms were still during their exploration phase before settled down on the liver area. No significant differences were found between the two groups' motility (Table 1).

We tested for recall of a familiar environment 4 days after the familiarization period. Familiarized worms displayed a significantly shorter time to reach criterion compared with the ‘unfamiliarized’ worms (one-tailed U-test, P<0.001; Fig. 3B, Table 2). Similarly, testing for the number of worms to reach criterion in less than 8 min revealed significant differences between the trained and control worms (one-tailed Fisher's exact test, P=0.005; Fig. 3A, Table 2).

Different groups of worms were tested 12–15 days following training. The familiarized worms displayed a significantly shorter time to reach criterion compared with the unfamiliarized control worms (one-tailed U-test, P<0.001; Fig. 3A, Table 2). Testing for the number of worms to reach criterion in less than 8 min revealed a significant difference between the trained and control worms (one-tailed Fisher's exact test, P=0.014; Fig. 3A, Table 2). We conclude that worms can remember a familiar environment for at least 14 days.

The finding that this memory persists for at least 14 days – long enough for the brain to regenerate – allowed us to check the possibility that this memory can survive brain regeneration. Headless fragments regenerated from familiarized worms displayed slightly shorter feeding latency compared with headless fragments from unfamiliarized worms when tested 10–14 days after decapitation (Fig. 3B, Table 2). However, the effect was not statistically significant. We then checked for the phenomenon of savings (see Materials and methods for detailed protocol), as McConnell found in his classical conditioning procedures (McConnell, 1965), where memory was revealed by a significantly faster training in a specific task in groups that had been trained on this task prior to decapitation. Worms that regenerated from headless fragments from original familiarized worms (Fig. 4) displayed significantly shorter feeding latency in the testing assay compared with regenerated worms that had not been familiarized to the environment prior to decapitation (one-tailed U-test, P=0.027; Table 2, Fig. 3B). Testing for the number of worms to reach criterion in less than 8 min revealed a significant difference between the original familiarized worms and control worms (one-tailed Fisher's exact test, P=0.013; Fig. 3A, Table 2). We conclude that some memory of the place familiarization survives decapitation and brain regeneration.

During the last decade, planarians have become an important model organism in the field of developmental and regenerative biology; because of their extensive regenerative capacity (driven by an adult stem cell population) and complex CNS, significant efforts are underway to understand the molecular mechanisms behind neural repair and patterning (Aoki et al., 2009; Gentile et al., 2011; Newmark and Sánchez Alvarado, 2002; Nishimura et al., 2011; Saló et al., 2009; Sánchez Alvarado, 2006; Tanaka and Reddien, 2011; Umesono and Agata, 2009). However, because of their rich behavioral repertoire and ability to learn (Corning, 1967; Oviedo and Levin, 2008), this model system also has the potential to offer unique opportunities for understanding the dynamics of memory during brain regeneration. This question has not only obvious clinical implications for stem cell therapies of adult neurological disorders, but also bears on the fundamental issues of mechanisms of memory encoding and storage in the physical processes of the brain.

While planarians are now being used for studies of drug addiction and withdrawal (Pagán et al., 2012; Raffa et al., 2008; Raffa and Valdez, 2001; Ramoz et al., 2012; Rawls et al., 2011; Rawls et al., 2010; Sacavage et al., 2008), the usage of planarians as a model for learning and memory is still very limited (Nicolas et al., 2008; Nishimura et al., 2010; Oviedo and Levin, 2008). Although extensive work on planarians' learning and memory has long suggested that memories can survive brain regeneration (McConnell, 1966), the limitations of previous manual experiments have led to these important questions being largely neglected by recent workers; these limitations included small sample sizes, difficulties in precise reproduction of protocols, and lack of quantitative analysis (Corning and Riccio, 1970; Travis, 1981). The aim of this work was to find a reliable, state-of-the-art approach that moves beyond past controversies to identify quantitative, objective, high-throughput protocols for eliciting and characterizing planarian long-term memory capabilities. By demonstrating evidence for the acquisition of relatively complex, explicit-like memories, the planarian system becomes even more central in modern research into learning and memory.

Environmental familiarity is a well-accepted paradigm for the study of the explicit memory mechanism in vertebrates (Heyser and Chemero, 2012; Heyser and Ferris, 2013; Teyke, 1989). Although some invertebrates such as bees and ants are capable of spatial memory and environmental recognition (Collett et al., 2003; Horridge, 2005), environmental familiarity has not been frequently used in learning and memory research with invertebrates. Best and Rubinstein (Best and Rubinstein, 1962a) showed that worms display a shorter feeding delay when being fed in familiar environment 90 min after a single 25 min familiarization session. Here we modified their environmental familiarization protocol and adapted it to use with an automated behavior analysis system (Blackiston et al., 2010). This system minimizes bias caused by manual procedures, allows an unprecedented level of quantitative, objective rigor in behavioral analysis and data reporting, and applies the procedure to a large sample size in a relatively short time frame. In addition to more rigorous controls (Davenport and Best, 1962; Dufort, 1962), our protocol allows retrieval after at least 14 days from the end of the training.

Because this protocol measures feeding behavior, the worms' performance in the retrieval test is dependent on their baseline appetite level. We examined different starvation periods between 1 and 30 days (T.S. and M.L., unpublished data) and found differences in the significance and variance of the results as a function of the worms' starvation period, as did Best and Rubinstein (Best, 1963; Best and Rubinstein, 1962a). We observed that the best results, in our procedure, were obtained when the worms were fed 3–4 days before being tested. Future users of this procedure must establish the correct hunger level in the worms to observe the best results in this assay. Because hunger level is a pivotal parameter in this approach and could be affected by many variables, such as manipulation intensity, maintenance temperature, size of the worms, species of worm and type of food, we offer an additional heuristic to other workers reproducing this protocol. As a heuristic, the proper hunger level seems to be achieved when not more than a third of the worms initiate feeding less than 1 min from the start of the testing trial and stay there until criterion is reached. Also, as seen from the results (Fig. 3B), although the general protocol was similar between the different groups, there were still differences in the general latency of feeding between the different categories. Even so, in all of the experiments, both control and experimental groups from each category were from the same colony, trained and tested in the same time and went through identical conditions of feeding and maintenance temperature. As a result, the changes in latency of feeding in each of the categories in both the experimental and control groups indicate the importance of rigor with respect to identical parameters and conditions for the experimental and control worms.

Importantly, in contrast to the most commonly used procedures (classical conditioning protocols), this environmental familiarity protocol cannot be attributed to pseudoconditioning or sensitization effects (Halas et al., 1962; Halas et al., 1961) instead of consolidation and retrieval of ‘real, encoded’ memory and behavior controlled by the brain. Planarian feeding is a truly complex behavior. Although composed of a series of stereotypic actions, it is coordinated and initiated by the CNS (Pearl, 1903; Sheiman et al., 2002). The feeding behavior is dependent on sensory integration (Pearl, 1903), as in our paradigm, of tactile/mechanical stimulation (Best and Rubinstein, 1962b), and chemotactic (Ash et al., 1973; Pearl, 1903) and optical sensations (Inoue et al., 2004).

Previous studies have shown that when food is placed in direct contact with the opening of the folded pharynx, it can activate the reflexes of extending the pharynx and swallowing, even in decapitated worms (Pearl, 1903; Wulzen, 1917). However, activation of these reflexes in decapitated worms is exceptional (Bardeen, 1901; Pearl, 1903) and the worms need to be starved (Bardeen, 1901; Wulzen, 1917) and tested directly after decapitation (Bardeen, 1901; Sheiman et al., 2002; Wulzen, 1917).

We never observed such behavior in our worms (D. japonica, which fasted for less than a week) and, consistent with others' observations (Pearl, 1903; Sheiman et al., 2002), our headless fragments with an intact pharynx did not demonstrate any interest in food until head regeneration (5–7 days after decapitation), even when the tail fragment passed immediately adjacent to the food. Moreover, we observed that extrusion of the pharynx happened just after the head made a first contact with the food, sometimes with a kind of stereotypic, drilling-like movements into the liver. We cannot completely rule out the possibility that the modifications in the peripheral nervous system contribute to the change in feeding latency. However, it is well accepted that the recognition of food and moving directly to it, as in our case, using decision making and a cautious approach, against their natural preference (under the strong light above and away from the edge of the dish; Fig. 2P, supplementary material Movie 1), are behaviors that are controlled by the CNS (Bardeen, 1901; Pearl, 1903; Sheiman et al., 2002). Finally, our results that show that in contrast to intact worms tested 2 weeks after training, regenerated worms with an intact pharynx required ‘retraining’ to demonstrate retrieval (Fig. 3, Table 2), suggesting that the difference found in latency of feeding is due to modification in the CNS and not (or not just) a reflex or peripheral nerve system modification. Thus, our data show the survival of a truly complex, brain-regulated behavior program through the process of head regeneration.

The procedure is ideally suited for an automated apparatus with minimal handling and does not require manual analysis, as was required for example in studies of conditioned response intensity in classical conditioning procedures (Corning, 1967; Prados et al., 2013; Wells, 1967). Our paradigm requires path tracking of the animals but no complex training algorithm with instantaneous feedback (light or shock) to each animal. Therefore, this protocol could also be performed with any of the off-the-shelf systems capable of multiple video tracking (Marechal et al., 2004; Noldus et al., 2001). The protocol avoids operator fatigue and ensures that no scoring biases are introduced into the data by subjective analysis of animal behavior.

While seeking the best complex learning protocol we observed the phenomenon previously called planarians' lethargy (Best, 1963; Best and Rubinstein, 1962b; Corning, 1964; McConnell, 1966; McConnell, 1965). Worms' learning curves during the training phase can suddenly reverse after a steady improvement, while healthy and active worms can begin to refuse to behave at all when inserted into the training apparatuses (Best and Rubinstein, 1962b; McConnell, 1965). Evidence suggests that this phenomenon could be related to familiarization to a dangerous environment, i.e. one in which the animal previously received noxious stimulus (Best, 1963; T.S., unpublished data). The protocol reported here involves natural behavior with minimal handling and without negative reinforcement. This overcomes planarians' lethargy and thus also allows the application to much more sensitive species such as Schmidtea mediterranea (Sánchez Alvarado et al., 2002).

No differences were found in general motility between familiarized and unfamiliarized worms (Table 1). Thus, any behavioral differences are not due to simple changes of overall activity level because of the familiarization protocol. The training occurred in complete darkness and the type and amount of water, food, handling and maintenance were identical between the familiarized (experimental) and the unfamiliarized (control) groups. Therefore, the learned difference between the two environments was mainly tactile. In the majority of their exploration phase, the worms were crawling around the edge on the bottom of the chamber. Hence, the experimental worms could feel the roughness of the floor and the dodecagon shape of the chamber walls, which alternated between delrin plastic and iridium oxide-coated titanium electrode (Fig. 2). Although no shock was delivered and the electrode material does not give off electrolysis products such as metal ions (Blackiston et al., 2010), there is a possibility that additional chemical cues from the electrode metal also facilitated place recognition.

Our results show that planarians can remember previously encountered habitats for at least 14 days (Fig. 3, Table 2). Dugesia japonica regenerates a functional head and CNS after 7 days, and in 14 days the worms are fully regenerated (Agata and Umesono, 2008; Inoue et al., 2004) (Fig. 4). Encouraged by the long-term retrieval, we investigated whether trained worms can display retrieval after decapitation and regeneration of a new head (Corning, 1966; Corning, 1967; McConnell et al., 1959). Worms regenerating from decapitated familiarized worms displayed a slightly shorter average feeding latency compared with regenerated fragments from unfamiliarized worms (Fig. 3, Table 2), but this effect was not statistically significant. Future work will explore longer training phases and further optimize different starvation periods to determine whether the strength of this effect can be increased.

McConnell's original results revealed a pattern of ‘savings’, where the learning curve of retrained animals is better (faster) relative to that of to naïve animals (McConnell, 1965; McConnell et al., 1959). Therefore, we checked for the presence of savings in the regenerated worms. In our savings protocol, regenerated worms were fed in the testing arena (familiarization environment) in a single 3 h session, 4 days before the retrieval test. Therefore, the feeding session was a previously encountered environment for the familiarized worms and a first introduction for the unfamiliarized worms. Worms that had regenerated from headless fragments from original familiarized worms displayed significant shorter feeding latency compared with unfamiliarized worms (Fig. 3, Table 2), suggesting that memory of the original environment was not located exclusively in the brain, and had become imprinted onto the newly built brain during regeneration.

In the past, such results have been received with skepticism (Smalheiser et al., 2001; Travis, 1981). The planarian has a centralized brain that guides behavior (Buttarelli et al., 2008; Sarnat and Netsky, 1985), and it is hard to imagine how memory traces (not just reflex arcs mediated by central pattern generators) can be encoded and stored in tissues remaining after complete head removal. However, such results are now made more plausible by modern discoveries such as epigenetic modification, which occurs in many cell types, not just the CNS (Arshavsky, 2006; Day and Sweatt, 2010; Ginsburg and Jablonka, 2009; Levenson and Sweatt, 2005; Zovkic et al., 2013) and RNAi (Smalheiser et al., 2001). It is likely that brain remodeling (plasticity during learning) and regeneration are both regulated via epigenetic pathways that determine patterns of self-organization of neural (Arendt, 2005; Davies, 2012; Kennedy and Dehay, 2012; Saetzler et al., 2011) and non-neural but electrically communicating cells (Levin, 2012; Mondia et al., 2011; Oviedo et al., 2010; Tseng and Levin, 2013).

It has long been known that regeneration both shapes and is in turn guided by activity of the CNS (Geraudie and Singer, 1978; Mondia et al., 2011; Singer, 1952). Thus, it is possible that experiences occurring in the brain alter properties of the somatic neoblasts and are in turn recapitulated back during the construction of the new brain by these adult stem cells. While exciting previous work in insects (Blackiston et al., 2008; Sheiman and Tiras, 1996) suggested the ability of memories to survive significant rearrangements of the brain and CNS (metamorphosis), planarians provide a unique molecularly tractable model of learned information persisting past complete removal of the brain. Of course, the mechanisms that allow unambiguous mapping (coding and decoding) of environmental sensory facts (e.g. ‘rough floor’, ‘metal walls’, etc.) into physico-chemical aspects of genetic material or neural network topologies are poorly understood not only for this case but for the normal relationship between conscious memory and its physical substratum in the intact brain.

Our data reveal the presence of memory savings in regenerated tail fragments from trained worms. However, no significant results were found in experiments that did not include a retraining component after the brain regenerated, indicating the necessity of CNS modification. These results could be due to insufficient training or a sub-optimal protocol. Alternatively, it is possible that only a rough correlate of the memory is present in the neoblasts, requiring a brief re-exposure to the trigger in order to consolidate into measurable effects on animal behavior (as occurs in the savings paradigm).

We suggest that some trace of memory is stored in locations distributed beyond the brain (because the place conditioning association survives decapitation). A straightforward model implies that information acquired during training must be imprinted on the regenerating (naïve) brain in order to result in the observed subsequent recall behavior. Future work must investigate the properties and mechanisms of such instructive interactions between remaining somatic organs and the regenerating CNS. However, two additional possibilities must be considered.

First is the possibility that the memory is executed entirely by the peripheral nervous system (PNS), not involving the brain in learning or recall. Given the similarities between the planarian brain and that of higher animals [in terms of structure, biochemistry and complex ethology (Nicolas et al., 2008; Oviedo and Levin, 2008; Rawls et al., 2011; Sarnat and Netsky, 1985)], and the fact that worms exhibit no behavior prior to the regrowth of the brain, it is most likely that the planarian brain indeed drives behavior. A pivotal role for the brain is also supported by the need for the savings portion of the paradigm, and the complexity of the behavior that is very unlikely to be implemented by receptor sensitivity and reflex modifications only (e.g. Fig. 2P, supplementary material Movie 1). However, if true, this would suggest a remarkable capacity for integration of complex information in the PNS of an animal that normally has access to an efficient brain, and thus would suggest a research program into the untapped information-processing abilities of the PNS in other advanced organisms.

Second is the possibility that the new brain is regenerated as a tabula rasa and is not imprinted by any traces of the previous memory. Instead, on this model the familiarized worms' PNS (which would have been modified and tuned, e.g. increased/decreased receptor sensitivity to a given stimuli during the training phase) is retraining the new brain, ‘burning’ the association into the new CNS during the short ‘saving’ session (which suffices because it is more efficient than in the unfamiliarized worms, because of the modified PNS sensitivity). We believe this scenario is less likely, because of the behavioral complexity of the learned task (Fig. 2P, supplementary material Movie 1). Experimental and control worms were fed with liver during the entire procedure, and the liver odor would be everywhere in the dish – this means the worms did not have to rely on the rough texture to know that food was somewhere in the vicinity, and both the trained and control groups could have developed positive associations to the smell of the liver. As can be seen in supplementary material Movie 1, the behavior does not resemble a simple reflex modification but rather the whole environment that makes trained worms initiate feeding sooner. We cannot completely rule out the possibility that the modifications in the PNS contribute to change in feeding latency. However, it should be noted that in order for receptor sensitivity to a particular stimulus to change after training, a kind of learning had to take place – the system as a whole (including learning, appropriate modification of the PNS and facilitation of re-training phases) implements an association between the presence of liver and the salient predictor of its presence, the rough surface, out of many other possible sensory modes that could have become more or less sensitized. Thus, this system would provide a novel model in which to examine the interactions between a mature PNS modified by specific experiences and learning in a newly developed brain (Inoue et al., 2004; Koopowitz and Holman, 1988).

Our results, obtained using a highly sensitive, objective, quantitative analysis system, support previous findings of Best and Rubenstein (Best and Rubinstein, 1962a) that planarians are capable of acquiring a relatively complex, explicit-like memories of environmental familiarity. Moreover, this memory survives long enough to allow full regeneration after amputation. Remarkably, headless fragments, regenerated from original environment-familiarized worms, display significant environmental familiarity in a savings paradigm. This simple and promising approach opens great opportunities for the use of planarians as model organisms for modern research of learning and memory. Importantly, planarians are the only molecularly tractable system in which memory and brain regeneration can be studied in the same animal. This is a crucial advantage and allows the investigation of innovative hypotheses as to the role of epigenetic and self-organization mechanisms in memory encoding, brain development and brain regeneration.

We thank Punita Koustubhan for general laboratory assistance; Junji Morokuma and Wendy Beane for advice and help with the planarian model system; Douglas Blackiston and Robert Cook for many helpful discussions about behavioral paradigms; Durwood Marshall, Dany S. Adams and Laura Vandenberg for assistance with statistics; Douglas Blackiston, Michael Romero and Philip Starks for comments on early versions of the manuscript; and Ethan Golden for fabrication of the rough-textured Petri dishes. This work is dedicated to Paul Van Oye and James V. McConnell, two pioneers of learning and memory in planarians.