ABSTRACT: The nudibranch mollusc Melibe leonina has been shown to be useful for neurophysiological studies due to its relatively simple nervous system. As a precursor to understanding nerve pathways for behavior lies the study of the behaviors themselves. Spontaneously occurring behavior of Melibe was observed and analyzed to determine patterns and predictability. Nine individually occurring behaviors and copulat ion made up the behavior catalog that was used. Results showed that there is an exponential transition rate out of the feeding mode of Melibe that there is a correlation between time spent feeding and size of the animal, and that the patterns of behavior can be predicted by a stochastic, one-step transitional probability matrix. Two behavior loops are also present in the Melibe which may demonstrate a hormonal change between them. The predictability of behavior can assist electrophysiological studies with live animal preparations by allowing the researcher to predict and prepare for certain behaviors. INTRODUCTION. Curiosity about how the human brain governs the complexity of human behavior has led many a scientist into the neurobiology field. Unfortunately, the human brain is far too complicated for studies at the current level of underständing. Invertebrate ganglia, with their significantly fewer neurons and simpler structure, have provided a template for initial study with the hope that relationships can be found between these animals and humans. A great deal of time has been spent on trying to find the neural pathways responsible for behavior and learning in invertebrates. In each study, a behavioral paradigm is required in order to relate neurological findings to the whole animal. A good example of the study of neuroethology, the combination of behavior and its neurophysiological basis, is one of Alan Gelperin's experiments on the terrestrial slug Limax maximus. The first part of his study involved behavioral experiments to train the slug to respond positively or negatively to certain food odors through associative learning. He then examined recordings from lip-brain preparations of the slug to try and understand the neurophysiological basis of training (Gelperin, 1983). Marine mollusks have also been used for neuroethological studies. The behavioral hierarchy of Pleurobranchaea californica a carnivorous mollusk, has been described by Davis et.al. (1974), with the intent that an underlying neurophysiological explanation could follow. The technique for deriving this hierarchy involved limiting stimuli to the slug during non-experimental periods, and then testing dominance between two initiated behaviors at a time. Arepresentative model was interpreted from the results. Because of the sensory deprivation involved in such a study, a general idea of natural or "everyday" behavior of the animal cannot be interpreted from a hierarchy. Äfter attempting a dominance study on the nudibranch Melibe leonina found that constant ly limiting the stimulus to the animal would be difficult and actually detract from the animal's natural responses. decided to attempt a less reductionistic approach to behavior by studying behaviors occurring in a simulated natural habitat. In spontaneous behavior, transitions between observable behaviors become more important than dominance. At the neural level, transitions between behaviors and decisions to make those transitions are probably controlled by very different mechanisms than those involved in dominance. while transitions may come in response to changing external and internal states, à dominance hierarchy shows only where one mechanism can overrule another. The model developed to describe patterns of behavior involves stochastic one-step transitions. A stochastic process is one in which a randomly determined sequence of behaviors is considered a sample of one element from a probability distribution (Random House Dictionary, 1987). Essentially, a stochastic model involves transition probabilities between behaviors that depend solely on the present state of the animal, reflecting no influence from past behavioral states. Melibe leonina provides many optimal conditions for this type of study. It is a large, hermaphroditic nudibranch, with a very characteristic set of behaviors. Transitions between behaviors are fairly slow, for easy recognition, and its ganglia is large enough to facilitate neurological studies (Thompson, personal communication). The three distinctive oral hood states of the Melibe provide a physical apparatus for clearly observing behavioral changes. Its environmental conditions are simple to replicate - a ratty blade of Macrocystis kelp, flowing water, and a few snalls. MATERIALS AND METHODS: Melibe leonina were collected off of Del Monte Beach in Monterey during free dives and SCUBA dives. They are found on solitary Macrocystis pyrifera kelp plants, in crowded populations of hundreds of individuals. Tanks under the shed at Hopkins were set up with a continuous sea water flow to hold the animals for further experimentation. Kelp and snails in the tanks provided a substrate and a natural cleaning service, respectively. A total of five holding tanks were used and approximately 100 Melibe were collected over the course of the ten weeks. All animals left at the end of the ten weeks were returned to their original kelp bed. Each of the five tanks for natural behavior observations, also set up under the shed, contained one blade of Macrocystis kelp, one or two snails, and two Melibe. As each animal was placed in a tank, it was measured for volume by water displacement and studied to find unique characteristics for later recognition. One animal from each study tank was injected with methylene blue dye in its right first cerata to aid in differentiation. All study animals were kept in as similar conditions as possible. They were fed every other day with mysids collected off of the kelp in Hopkins Marine Life Refuge. Tanks were cleaned out every two weeks, but never on ar observational day. Melibe were never disturbed unless it became necessary to change something in the tank, if a death occurred, or if the drain clogged. Observations were made in six hour sessions, at fifteen minute intervals. Daytime and nighttime behaviors were recorded for a total of sixty-six hours of observations. Behaviors were recorded on tank sheets (figure 2) and compiled into tables (figure 3). The statistical analysis methods were mainly comprised of chi-square tests for significance of different behavior patterns using the equation Xa-2 (0.-E. P/E. where O is the observed value and E is the expected value. Counting of the various sequences of behaviors was done by an original program written in Pascal on a Macintosh personal computer. Transitional probabilities were assumed to be equal to the observed transitional frequencies. They were used to calculate the probability of occurrence of three behavior sequences using the equation P(ABC)-P(A)P(BIA)P(CB) where A, B, and Care any three behaviors (Oden, 1977). RESULTS: Initial observation of Melibe led to the distinction between nine different behaviors, six of which are common, and three which are less common in spontaneous behavior (figure 4). Copulation was observed to occur during day and night hours, and all other behaviors, other than swimming, were observed to occur spontaneously with copulation, as opposed to observations made by Ageska (1972) which claimed that animals must remain quiescent during copulation. Analysis began with the study of casting/feeding behavior. Total time spent in each behavior was tabulated for every animal. Plots were made of total time spent in a behavior versus the volume of the animal (figure 5). Feeding behavior was well correlated with size, with larger animals spending more time feeding. Time spent resting showed a tendency to decrease as size increased, but the correlation coefficient was low. Transitional frequencies between consecutive behaviors were calculated (figure 7). Using these frequencies, I diagramed all transitions with a frequency greater than ten percent (figure 8). This diagram shows graphically a tendency of the Melibe to remain in one of two behavior loops, F-OH-R/OH or RST-AL-R/AL. The statistical reality of these loops was quantified and it was found that transitions between these loops were significantly lower than between any other combination of 3 or 4 behaviors. Total time spent in each loop for every animal was counted, as well as the average number of intervals within each loop. Volume of animals was graphed against time spent in the F-ÖH-R/OH loop (figure 9). Larger animals tend to spend more time in this feeding loop. Four behaviors, F, OH, RST, and AL had a tendency to last longer than just one interval before transition. I defined a session as a period in which a single behavior continued consecutively. Feeding was the behavior with the most data to support an analysis of the length of its sessions. Sessions were tabulated for each animal and the results graphed (figure 10). An exponential describes the best fit of the data. The equation y-C10 (-kt) calculated for all animals with enough data to support analysis, showed that each antmal had a different decay rate. When volume was graphed versus the mean feeding session length, derived from the k-constant, there was shown to be no significant correlation (figure 6). The k-constant can be converted to mean feeding session length using the equation (mean t) = 0.3017K. To see if a one-step transition frequency matrix is enough to describe longer sequences of behavior, I used the equation presented in the methods to calculate predicted frequencies of sequences of three behaviors (or two-step transitions), starting with the behaviors F, OH, AL, and RST. The one-step probabilities were taken from figure 7. The observed and predicted matrices of occurrences of triplet sequences are in figure 11, with the corresponding chi-square values in figure 12. These matrices were graphed 3-dimensionally in order to visualize the similarities between the predicted and observed frequencies (figure 13) using the computer software "Surfer DISCUSSION: Iwo main conclusions can be drawn from the results of this research. One entails the relationship of physical characteristics to the individual animal's behavior and the other is the relationship among different behaviors. This research also reforms some previously published observations about Melibe behavior in general (Ageska, 1972). Analysis shows that larger animals will spend more time feeding and more time within the "active" feeding loop. Further experimentation could show how a larger animal may have a greater concentration of a "feeding hormone, or that certain neural pathways are stronger that control feeding. Individual animals will tend to remain in the behaviors F, OH, AL, and RST for longer than one fifteen minute interval, yet transitions in and out of roaming behaviors (R/AL, R/OH) are much faster. There may be different bases for the different types of behavior, one "static" and one "transitionary", which can be detected through electrophysiological examination. The importance to the animal of choosing between behavior loops is very interesting. Although transition between loops is infrequent, all animals spend some time in each, presumably because all animals need resting time as well as feeding time. There is probably some hormonal basis that increases the probability of switching from one loop to another, considering the different physiological ramifications involved in the behaviors. The feeding loop allows the animal to be ready to detect food and/or eat it, whereas the resting loop is less energy consuming. Future study involving applied stimuli could show if the defense responses, either crumple or shrug (Scott, 1990), are greater in one of the two loops. The transition frequencies among all of the behaviors led to the formulation of a stochastic model dependent on those frequencies. The use of a stochastic model in descriptions of animal behavior is not new (Bekoff, 1977). Often the process so described is referred to as a Markov process, in reference to its designer. I have chosen to avoid the term "Markov" due to its connotative requirement of stationarity in the data. Stationarity is the property that behavior or behavioral sequences do not change when comparing different time periods. Previous studies that claimed to have found Markovian analyses of animal behavior have been criticized on their lack of stationarity (Bekoff, 1977). In my research, I assumed that stationarity was irrelevant: first, because ! could not determine a proper time interval in which to assume stationarity in ten weeks; second, because 1 pooled the behavior of all animals in each set, effectively neutralizing any substantial motivational changes that would be considered non-stationary The equation for testing a one-step transition model (pg. 5) was taken from an analysis on hermit crab agonistic behavior that was assumed to be non-stationary (Oden, 1977). Using this equation, the frequency of two-step transitions (or 3-behavior sequences) was estimated. Comparing the matrices of predicted and observed occurrences of each sequence yields chi-square values that for thirty-six degrees of freedom show some significant differences. Fortunately, there are several problems with the chi-square as a statistical test in this case. First, the data set in this study is co-variant, leading to an unclear decrease in the degrees of freedom. Also, large six-by-six matrices result in some cells with low expected counts. This will tend to overinflate the chi-square statistic (Steinberg, 1977). Assuming that the chi-square values are overinf lated, the amazing similarities between the predicted and observed sequence frequencies become apparent. This similarity demonstrates that spontaneous behavior sequences in Melibe are stochastic. An animal's transition between its present state and the next future state, within my imposed interval of fifteen minutes, depends only on its present state and not on any previous states. This gives interesting insights into the neurophysiological bases of Me/ibe behavior. It is probable that memory is limited and that circuits that control behavioral transitions are relatively simple. Spontaneous behavior of Melibe leonina is an observable and analyzable subject of study. The information that the study has provided shows that behavior does not have to be elicited in order to be interesting. Possibilities for future research include examining learning in Melibe determining the physiological changes of the animal between behavioral loops and determining the usefulness of the predictive capability of the one-step transition model for whole animal electrophysiological preparations. ACKNOWLEDGEMENTS: First of all I would like to thank my effervescent advisor, Stuart Thompson, for his encouragement, enthusiasm, and patience this quarter. A big thanks to all of the people at Hopkins who are friendly and helpful and always willing to assist a flailing undergraduate, especially Nicole Crane, Bruce Hopkins, Carla (for the super Melibe tanks)), Chris Patton, and Alan Baldridge. To everyone in the 175H class, I never thought 1 could meet such an amazing, talented, friendly group of people in one placel Thanks to Chris Mathes for all of his help, and to his wife and son, Libby and Jonathan for adding a touch of home to the lab. Thanks, Agnieszka and Nat, for always keeping the laughter in the lab and becoming two dear friends. Special thanks to Shari for long mysid collecting mornings and diving trips which became wonderful heart-to-heart talks (Why ask why? Drink a beerl). Thanks to my awesome roomie, Saween, for keeping me on track, making me be social, and teaching me that ! don't have to eat Noodle-Roni every night. And last but absolutely not least, I want to express my gratitude to my mentor, Sam Wang, for hours and hours of generous encouragement and help. You are truly an amazing person and friend. LITERATURE CITED 1. Ageska, R.A. and Nybakken, J. 1972. Contributions to the Biology of Melibe leonina(Gould, 1852), The Veliger, 19: 19-26. 2. Bekoff, M. 1977. Quantitative Studies of Three Areas of Classical Ethology: Social Dominance, Behavioral Taxonomy, and Behavioral Variability. pp 1-46. 7: Hazlett,B.A. (ed.), Quantitative Methods in the Study of Animal Behavior. Academic Press: San Francisco. 3. Davis, W.J., Mpitsos, G.J., and Pinneo, JM. 1974. The Behavioral Hierarchy of the Mollusk Pleurobranchaea, 1. The Dominant Position of the Feeding Behavior. Journal of Comparative Physiology 90: 207-224. 4. Gelperin, A. 1983. Neuroethological Studies of Associative Learning in Feeding Control Systems. pp 189-205 77: Huber,F. and Markl,H. (eds.), Neuroethology and Behavioral Physiology - Roots and Growing Points. Springer-Verlag: New York. 5. Oden, N. 1977. Partitioning Dependence in Nonstationary Behavioral Sequences. pp 203-220. 7n: Hazlett,B.A. Quantitative Methods in the Study of Animal Behavior. Academic Press: San Francisco. 6. The Random House Dictionary of the English Language, second edition, unabridged. Random House: New York, 1987 7. Scott, D.L. 1990. The Shrug Response in Melibe leonina: Behavioral and Neurophysiological Observations. HMS Biology 175H papers. e 8. Senter, R.J. 1969. Analysis of Data: Introductory Statistics for the Behavioral Sciences. Scott, Foresman and Company : Glenview, IIlinois. 509 pp. 9. Steinberg, J.B. 1977. Information Theory as an Ethological Tool. pp 47-74 /n: Hazlett, B.A. (ed.), Quantitative Methods in the Study of Animal Behavior. Academic Press: San Francisco. 10. Taylor, HM. and Karlin,S. 1984. An Introduction to Stochastic Modeling. Academic Press: San Francisco. 398 pp. 1. Thompson, S.H., Personal communications. LIST OF FIGURES 1. Schematic of a Melibe leonina specimen, showing important morphology. 2. Example of a tank sheet, used for recording Melibe behavior. 3. Example of a compiled behavior table from one six-hour observational day, 4. List of behaviors and their physical characteristics, 5. Total time spent feeding/casting and resting versus volume of the animal. 6. Mean feeding session length versus volume of the animal. 7. One-step transitional probabilities for each set of ten Melibe. 8. Diagram of transition probabilities greater than ten percent for both sets of animals. 9. Time spent in the feeding loop versus volume of the Melibe 10. Exponential decay rate of feeding session lengths for two animals. II. Example of observed and predicted matrices for three behavior sequences. 12. Significance of X-square scores calculated from the predicted and observed three behavior sequence matrices. 13. 3-D representation of predicted and observed matrices of sequences of three behaviors. 0 -4 C goe 1 4 . - DAE:S 2 6 9 TIME a et O 7 20 - 29110 Figure 3 Behavior 6 COMON BEHAVIORS. casting/feeding open hood alert posture roaming with open hood roaming with alert posture resting 3 INFREQUENT BEHAVIORS: swimming crumple roamng NATURALLY OCCURRING BEHAVIORS Abbreviation Oneletter Charactert -casting of hood. expulsion of water from hood -ho motion, hood and O tentacles extended, cerata upright -hood up, tentacles in, Al cerata upright -motion of foot on R/OH substrate with open nood -otion of foot with R/AL alert posture -hood,tentacles tucked RST in, cerata down, no motion -side to side body contractions, no 4 contact with subsuat -tight contraction of CR whole body -in of fot substrate with hood down and tucked under Figure 4 .. — SIZE OF MELBE VS. % TIME FEEDING AND RESTING 80 5 60 y -- 15.71+0.4684 R-0.91 . O FEEDM y-3.1465-0.1275x R-0.40 v 20 2 100 200 VOUT &a REIIEE (I) Figure 5 Mean Feeding Session Length vs. Volume of Melibe y -3.4222 - 0.0023x P =0.07 100 50 200 volume (mi) Figure 6 E 2 Figure 0. IL 47.3 21.8 1.5 R/ON 25.3 4.2 2.2 R/AL 30.2 5. RST8 65.5 BEHAVIOR TRANSITIONS (by %) 22.7 20.0 A R/OH 126.8 12 3.! 0. — R/1 44.6 52.6 2.2 97 BEHAVIOR TRANSITIONS (by %) igure 8 VOLUME VS. TIME SPENT IN F-OH-R/OH LOOP 100 6.7537 +0.3101X R-0.87 80 60 40 9 20 o 100 200 300 VOL (mI) FigureS Ezponential Decay of Feeding Session Length in 1A 30 - y - 18.5833 * 10•(-0 0966x) R -098 20 D 1A sessions * intervals (15 min) Exponential Decay of Feeding Session Length in 11A y - 18.5833 * 10°(-0.0966x) R -0.98 intervals Figure 10 8 1 F 0 C3 R 11 Observed matrix P(A,BC 1 I 27 Figure A,-A 8 A 1 1 1 C3 0 88 4 21 26 5310.3 23 73 10 14.2 81.2 6.2 32 23 12 26 41 132 4 03 1.21132 06 Predicted matrix PCA,DPCB, IA,PCC,IB. 15 2 15.5 15 35 26 O.3 12 09 12 33.7 XSOUARE VALUES FOR DIFFERENCE BETWEEN NUMBER OF EDICTED AND ÖBSERVED 3-BEHAVIOR SEQUENCES (ABC) Behavior A Kequare value Significance pe(df-36 SET1 Feeding/casting 62.9 0.005 Open hood 453 O.5 58.2 Alert posture 0025 Resting 55.0 0.025 SET 2 Feeding/casting 0.05 52.3 46.3 Open hood 0.5 O.5 42.5 Alert posture 43.6 Resting 0.5 Figure A Figure 13 BEHAVIOR AEF