{"paper_id":"199d627c-c347-4e0a-87c9-8bac337860ed","body_text":"=== R E V I E W   C O M M O N S   M A N U S C R I P T ===\nIMPORTANT:\nManuscripts subm itted to Review Com m ons are peer reviewed in a journal-agnostic way.\nUpon transfer of the peer reviewed preprint to a journal, the referee reports will be available in full to the handling editor.\nThe identity of the referees will NOT be com m unicated to the authors unless the reviewers choose to sign their report.\nThe identity of the referee will be confidentially disclosed to any affiliate journals to which the m anuscript is transferred.\nGUIDELINES:\nFor reviewers: https://www.reviewcom m ons.org/reviewers\nFor authors: https://www.reviewcom m ons.org/authors\nCONTACT:\nThe Review Com m ons office can be contacted directly at: office@reviewcom m ons.org\n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\n \n \n \n \n \n \n \n \nForgetting in Drosophila \nconsists of an increase in uncertainty rather than \na stochastic loss of memory \n \n \n \n \n \n \n \n \nJunjiro Horiuchi1*, \nNozomi Uemura1, Shiro Horiuchi2, and Minoru Saitoe1 \n \n \n1. Tokyo Metropolitan Institute of Medical Science, Tokyo, \nJapan \n2. Hunter College, City University of New York, USA \n* Corresponding author  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\nAbstract \n \nWhile forgetting has been studied extensively in various organisms, its precise \nnature has often been unclear. Here, we used behavioral experiments in \nDrosophila to determine that a significant aspect of forgetting consists of a \ndecrease in the ability of a memory to induce an appropriate behavior. We tested \nflies for memory retention at various times after training and then separately \nretested both flies that chose correctly and those that chose incorrectly. \nAlthough the ability to choose correctly decreased over time, we could not \nmeasure any differences in memory between flies that initially chose correctly \nand those that chose incorrectly upon retest. This suggests that forgetting is \nunlikely to consist of a spontaneous loss of a memory but instead consists of a \ndecrease in the probability of flies that remember choosing the correct \nbehavioral response. Thus, although flies maintain memory over time, there is an \nincrease in uncertainty associated with this memory. We find that forgetting of \nlong-term memories and accelerated forgetting in old flies occur in a similar \nmanner. \n \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\nIntroduction \n \nThere are various types of forgetting. Anecdotally, in humans, one type may \nconsist of a complete lack of access to a memory where we may completely \nforget to buy milk on the way home from work, and only later recall this memory \nwhen we arrive home and have nothing to drink. In contrast, a different type of \nforgetting consists of an increase in the amount of uncertainty associated with a \nparticular memory. In this situation, we can access the memory, but the contents \nof the memory are unreliable,\n leading to uncertain memory recall. Thus, we may \nremember reading a book, but certain aspects of the plot or storyline may \nbecome hazy over time. \nWhile forgetting has been extensively studied in various \nanimal models, the specific aspects of memory that decrease upon forgetting \nhave not been well characterized. \n \n \nDrosophila have been a highly useful organism for the study of learning and \nmemory, and more recently, forgetting \n(Berry & Davis, 2014; Gao et al., 2019; \nHoriuchi, 2019; Quinn, Harris, & Benzer, 1974; Tully & Quinn, 1985). Similar to \nother animals, Drosophila can learn and form memories, which gradually decay \nover time. This memory decay can be plotted as a memory retention curve or \nforgetting curve \n(see Fig. 1) (Tully & Quinn, 1985), and similar to other \norganisms, as Drosophila age, they suffer an accelerated loss of memory (Mery, \n2007; Tamura et al., 2003). Thus, forgetting can be reliably measured in \nDrosophila. However, specific characteristics of forgetting in flies, whether it \nconsists of a loss of memory of an association, or whether it consists of an \nincrease in uncertainty regarding an association, have not yet been analyzed. \n \nIn Drosophila, learning and memory are often measured using an olfactory \nassociative task in which flies are trained to associate an odor with pain (Quinn \net al., 1974). A population of flies is exposed to an odor and at the same time \nexposed to aversive electrical shocks. Flies are next exposed to a second odor, \nthis time in the absence of electrical shocks. Flies learn to associate the first, but \nnot the second, odor with pain, and subsequently avoid this odor. Memory of this \nassociation can be tested by allowing trained flies to choose between the two \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\nodors in a T-maze (Tully & Quinn, 1985). Immediately after training, a large \nproportion of flies avoid the shock-paired odor and choose the unshocked odor. \nThis proportion decreases as the time between training and testing increases, \nand this decrease is thought to reflect a time-dependent increase in forgetting. \nThus, mutations that improve memory retention without increasing initial \nlearning have been used to identify putative biochemical components regulating \nforgetting (Berry, Cervantes-Sandoval, Nicholas, & Davis, 2012; Berry & Davis, \n2014; Davis & Zhong, 2017; Horiuchi, Yamazaki, Naganos, Aigaki, & Saitoe, 2008; \nShuai et al., 2015; Shuai et al., 2010). In addition, Drosophila mutants with \naltered memory retention curves have been used to identify different memory \nphases, including initial learning \n(LRN), short-term memory (STM), middle-term \nmemory (MTM), anesthesia-resistant memory (ARM), and long-term memory \n(LTM) (Tully & Quinn, 1985). These memory phases occur in a specific temporal \norder and persist for different durations. This temporal sequence suggests that \nforgetting rates may be related to the efficiency of transition between different \nmemory phases. Thus, forgetting may consist of either a progressively smaller \nnumber of flies forming successive memory phases, or a progressive increase in \nuncertainty associated with successive memory phases.  \n \nIn this study, we analyzed forgetting in Drosophila using behavioral analyses to \ndetermine whether it \nconsists of a stochastic loss of memory in an increasing \nsubset of flies, or whether it consists of a gradual reduction of memory strength \nthat occurs throughout the population. Our results do not exclude the possibility \nthat some flies spontaneously forget an association over time. However, our data \nindicate that there is a time-dependent decease in the probability that a fly with \nmemory chooses the non-shocked (appropriate) odor. In other words, a \nsignificant aspect of forgetting in flies consists of a decrease in the ability of a \nmemory to influence a behavior. We refer to this decrease as an increase in \nuncertainty. \n \n \nResults \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\nIn the Drosophila olfactory association training paradigm, odor concentrations \nare chosen such that naïve flies distribute 50:50 when given a choice between \ntwo odors. When flies are given this choice immediately after odor/shock \ntraining, approximately 95% of flies choose the non-shock paired odor \n(hereafter referred to as the correct odor), while approximately 5% choose the \nshock-paired odor (referred to as the incorrect odor). This result suggests two \nextreme possibilities. If a fly that learns the association always chooses the \ncorrect odor when tested immediately after training, 90% of flies must have \nlearned the association while 10% did not. (The 95% of flies choosing the correct \nodor should consist of the 90% that learned and half of the 10% that didn’t.) On \nthe other extreme, we consider the situation where all flies learn the association. \nIn this case, learning must consist of a shift in the probability of choosing the \ncorrect odor from 50% to 95%. \nThese two non-mutually exclusive possibilities \ncan be generalized by the following mathematical model, which expresses \nbehavior as a function of memory and memory strength: \n \nP\n(C) = P(M) * P(C|M) + P(M’) * P(C|M’)     (1) \n \nwhere P(C) is the probability that a fly chooses the correct odor, \nP(M) is the probability that a fly has memory of the odor association, \nP(C|M) is the probability that a fly that has memory chooses the correct odor, \nP(M’) is the probability that a fly has no memory of the association, \nand P(C|M’) is the probability that a fly that doesn’t have memory chooses the \ncorrect odor.  \n \nFlies that don’t form memories \nshould distribute evenly between the odors, \nsimilar to naïve flies; thus, P(C|M’) = 0.5. In addition, flies should either have \nmemory of the association or not; thus, P(M) + P(M’) = 1. Incorporating these \nconstraints, equation (1) can be rewritten as: \n \nP\n(C) = P(M) * P(C|M) + 0.5 * (1 - P(M))     (2) \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\nIn the first situation described above, where a fly with memory always chooses \nthe correct odor, P(C|M) = 1, and P(C) = (P(M) + 1)/2. In the second situation, \nwhere all flies learn the association, P(M) = 1 and P(C) = P(C|M). \n \nWe next considered what happens when forgetting occurs. Again,\n we first \nconsidered two extreme possibilities. If forgetting consists strictly of a stochastic \nloss of an associative memory in a subset of flies, P(M) should decrease over \ntime, while P(C|M) will remain constant. On the other hand, if forgetting consists \nof a gradual reduction in certainty regarding an association that occurs in all flies \nthat learned, P(C|M) will decrease while P(M) remains constant.  \n \nThese two possibilities can be distinguished by testing flies for odor preferences \nat various times after training, and then separately retesting flies that chose the \ncorrect and incorrect odors in a second odor preference test immediately after \nthe first test.  \n \nIf forgetting consists of a stochastic loss of memory, trained flies should consist \nof two separable populations: one population that remembers and a second \npopulation that forgets. \nIn this case, the initial odor preference test should \nseparate flies into two non-equivalent populations. The population of flies that \nchose the correct odor should be enriched for flies that remember, while the \npopulation of flies that chose the \nincorrect odor should be highly enriched for \nflies that forget. When a second odor preference test is given immediately after \nthe first, flies that chose correctly in the first test should again choose correctly \nin the second test at similar or higher probabilities. In contrast, flies that chose \nincorrectly in the first test should distribute between the odors at probabilities \nclose to 50:50 in the second test. Thus, P(C2|C1) > P(C2|C1’), where P(C2|C1) is the \nprobability that a fly that chose correctly in the first test chooses correctly in the \nretest, and P(C2|C1’) is the probability that a fly that chose incorrectly in the first \ntest chooses correctly in the retest. \n \nOn the other hand, if forgetting consists strictly of a gradual reduction in \ncertainty regarding the odor association that occurs in all or most flies in the \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\npopulation, P(M) should remain constant at a value close to 1, while P(C|M) \nshould decrease over time. In this case, trained flies should not separate into two \ndifferent populations. Instead, the probability of a fly choosing the correct odor \nshould result from random or stochastic choices reflecting P(C|M) within a \nrelatively homogeneous population. Thus, retesting flies that initially chose the \ncorrect odor and flies that initially chose the incorrect odor should yield odor \npreferences that are similar to each other, and to the results of the initial odor \npreference test. P(C2|C1) ≃ P(C2|C1’).  \n \nWe first performed retest experiments on flies immediately after aversive \nolfactory training and observed that close to 90% of trained flies chose the non-\nshocked odor when tested immediately after training (Fig. 2). When we \nseparately retested flies that initially \nchose correctly and flies that initially chose \nincorrectly, we found that both populations showed a significant preference for \nthe non-shocked odor. This indicates \nthat a significant percentage of flies that \nlearned chose incorrectly in the original odor preference test. Thus, P(C|M) < 1. \n \nNext, to determine whether \nforgetting at 3 hrs consists of a decrease in the \npercentage of flies that remember, or a decrease in the probability that flies that \nremember choose the correct odor, we repeated the above experiment 3 hrs \nafter training (Fig. 3). Unexpectedly, we found that a significantly higher \npercentage of flies that initially chose the incorrect odor chose the correct odor \nupon retest (compared to the percentage of flies that chose the correct odor in \nthe initial test). Further, we found that a significantly lower percentage of flies \nthat initially chose the correct odor chose the correct odor upon retest \n(compared to the percentage of flies that chose the correct odor in the initial \ntest). It is extremely unlikely that flies that initially chose incorrectly have \nimproved memory during the retest, suggesting that a non-associative effect \nmust be occurring during the initial odor preference test that affects behavior \nduring the retest.  \n \nTo examine this possibility, we measured \nthe behavior of naïve flies subjected to \ntwo consecutive odor preference tests (Fig. 4). As designed in our experimental \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\nparadigm, naïve flies distribute evenly between the two odors when initially \ntested for odor preference. However, upon retest, flies which initially chose one \nof the odors, octanol (Oct), preferred the second odor, methylcyclohexanol \n(MCH) upon retesting, while flies which initially chose MCH preferred Oct. Thus, \nflies tend to alternate their choice of odors, selecting odors that they had \npreviously avoided when given a second chance. While we are not certain why \nflies have this tendency, we include several possible explanations for this \nbehavior in the Discussion. Regardless of why flies behave in this curious \nmanner, we refer to this behavior as an opposite preference tendency \n(T), and \nhave included naïve controls in parallel with trained flies in all subsequent \nexperiments. \n \nWe performed odor retest experiments for trained and naïve flies at various time \npoints after training (Fig. 5). \nWhen we tested flies immediately after training (3 \nmin time point), there were no significant differences in the percentage of flies \nchoosing the correct odor during the first test or during retesting of flies that \ninitially chose the correct or incorrect odors, and all of these scores were \nsignificantly different from the opposite preference tendency in naïve flies (Fig. \n5A). This indicates \nthat a proportion of flies that learned choose the incorrect \nodor during initial testing. Despite choosing incorrectly in the first test, these \nflies retain memory of the odor association and are able to \nchoose the correct \nodor with a high probability on the second test.  \n \nAs the time interval between training and testing increases, the percentage of \nflies choosing the correct odor during the initial test decreases as flies gradually \nforget the association \n(Fig. 5A-D). Flies that chose correctly during this initial test \nshow a decreased tendency to select the correct odor during the retest over time, \nwhile flies that chose the incorrect odor during the initial test show an increase \nin the probability of choosing the correct odor. It is important to keep in mind \nthat for flies that initially chose the correct odor, the opposite preference \ntendency will work against the tendency for flies to choose the correct odor in \nthe retest. \nFor flies that initially chose the incorrect odor, the opposite \npreference tendency will enhance the tendency for flies to choose the correct \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\nodor in the retest. Thus, our data are consistent with a model in which odor \npreferences caused by training are strong immediately after training and \ndominate odor preference scores in both the initial tests and retests. As time \nafter training increases, training-dependent associations decrease in strength, \nand retest scores tend to become the sum of memory-induced preferences and \nopposite odor preferences.  \n \nFlies can form long-term memories (LTMs), which can be measured 24 hrs after \nspaced training, a training protocol consisting of multiple \n(10x) training trials \nwith 15 min rest intervals between each training. LTM has similarities with \nshorter forms of memory since it requires the same gross anatomical structures \nincluding the mushroom bodies and antennal lobes, but it is also\n distinct because \nit has different molecular requirements and uses different neuronal networks \ncompared to short forms of memory. From our odor retest experiments, we find \nthat LTM has qualitative similarities with short-lasting 3 hr and 5 hr memories \n(Fig. 5E). LTM increases the probability that flies will choose the correct odor. \nHowever, flies with LTM will still choose the incorrect odor at some probability, \ndespite retaining memory of the association and maintaining an increased \nprobability of choosing the correct odor.   \n \nIn order to examine \nhow P(M) and P(C|M) are affected by forgetting, we used our \nmeasured values for R(C), R(C2|C1), R(C2|C1’), and R(T) as estimates of P(C), \nP(C2|C1), P(C2|C1’), and P(T) and calculated numerical ranges for Pt(M), Pt(C1|M), \nPt(C2|MC1), and Pt(C2|MC1’) using equation (2) and the following equations.  \n \nPt(C2|C1)  = Pt(M|C1) Pt(C2|MC1) + [1- Pt(M|C1)] [1 – P(T)]   (3) \nPt(C2|C1’) = Pt(M|C1’) Pt(C2|MC1’) + [1- Pt(M|C1’)] P(T)   (4) \nPt(M|C1) = Pt(M) Pt(C1|M) / {Pt(M) Pt(C1|M) + [1- Pt(M)] 0.5}   (5) \nPt(M|C1’) = Pt(M) [1 - Pt(C1|M)] / {Pt(M) [1 - Pt(C1|M)] + [1-Pt(M)] 0.5}   (6) \n \nwhere,  \n \nPt(C2|C1) is the probability at time t that a fly will choose correctly in the second \ntest given that it chose correctly in the initial test, \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\nPt(M|C1) is the probability that a fly that chose correctly in the initial test has \nmemory, \nPt(C2|MC1) is the probability that a fly that has memory and chose correctly in the \nfirst test will choose correctly in the retest,  \nP(T) is the probability that a fly that chose an odor in the initial test will choose \nthe opposite odor in the retest,  \nPt(C2|C1’) is the probability at time t that a fly will choose correctly in the second \ntest given that it chose incorrectly in the first test,  \nPt(M|C1) is the probability that a fly that chose correctly in the initial test has \nmemory,  \nand Pt(M|C1’) is the probability that a fly that chose incorrectly in the initial test \nhas memory,  \n \nBy restricting Pt(M) \nto values between 0 and 1, and restricting Pt(C1|M), \nPt(C2|MC1) and Pt(C2|MC1') to values between 0.5 and 1, we calculated possible \nranges for Pt(M), Pt(C1|M), Pt(C2|MC1) and Pt(C2|MC1'). As seen in Figure 6A, the \nrange of P(M) expands over time, encompassing both situations where P(M) \nremains constant and situations where P(M) decreases over time. In contrast, we \nfind that values for P(C1|M) decrease over time (Fig. 6B). Overall, our data \nindicate that a major component of forgetting is a decrease in certainty, P(C|M), \nthat occurs over time.  \n \nSince aging in many species is associated with an increase in forgetting (Tamura \net al., 2003), we next characterized short-term memory retention in old flies \nusing the retesting procedure (Fig. 7). Overall, we observed reduced memory \nscores at early (3 min and 2 hrs) time points, and lesser, non-significant \nreductions at later (3 hr and 5 hr) timepoints in old flies compared to young flies. \nThese reductions are consistent with age-dependent memory defects described \nin previous reports (Tamura et al., 2003; Yamazaki, Horiuchi, Miyashita, & Saitoe, \n2010). We also observed reductions in retests for old flies that initially chose the \ncorrect odor compared to young flies (3 min and 2 hr time points). We did not \nobserve significant differences in retests for old flies that initially chose the \nincorrect odor compared to young flies. Altogether, our results suggest that \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\nforgetting in old flies is qualitatively similar to forgetting in young flies, but with \nan accelerated rate of forgetting at early (0 hr and 2 hr) timepoints (Fig. 8).  \n \n \nDiscussion \n \nAlthough \nlearning and memory have been studied extensively in Drosophila, it \nhas been unclear precisely how memory decreases over time, and how forgetting \noccurs. For olfactory associative memories in Drosophila, flies form a simple \nassociation between a specific odor and either pain or rewards. Forgetting of this \nassociation could consist of a simple loss of memory or an increase in \nuncertainty regarding this association. Our study demonstrates\n that a large \ncomponent of forgetting consists of a general decrease in the probability that a \nfly that remembers will choose the correct odor. This can be thought of as a time-\ndependent decrease in memory strength or an increase in uncertainty. \n \nIn Drosophila, a general memory retention curve consists of the summation of \nvarious different memory phases, including LRN, STM, MTM, ARM, and LTM \n(Tully et al., 1990; Tully, Preat, Boynton, & Del Vecchio, 1994). Our data suggest \nthat forgetting does not consist strictly of a decrease in the probability that one \nmemory phase is converted into a subsequent memory phase. In other words, it \nis unlikely that the main component of forgetting consists of subsequently \nsmaller subset of flies forming STM, MTM, ARM, and LTM after learning. Instead, \nour data suggest that later memory phases differ from earlier memory phases in \ntheir ability to motivate flies to choose the correct odor.  \n \nPrevious studies have tried to use behavioral tests to separate flies that learned \nfrom flies that did not (Mery, 2007)\n. However, our work indicates that this \nstrategy may have difficulties because we do not find significant differences in \nmemory between flies that choose the correct odor and those that choose the \nincorrect odor. Our results are somewhat surprising since testing is almost \nexclusively used in human education and in mammalian learning studies to \nevaluate learning and memory in individuals. How are memory and forgetting in \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\nour experiments different from memory and forgetting in mammals? Flies used \nin our experiments are from a homogeneous genetic background and were \nraised in identical conditions. More importantly, our olfactory associative \ntraining paradigm is not an operant conditioning paradigm. Flies do not have a \nchoice during training and are forcibly exposed to odors for one minute each and \nshocked 12 times during exposure to the shock-paired odor. Thus, factors such \nas motivation to learn and amount of study are normalized in our study. In this \nsituation, our results suggest that forgetting occurs similarly in all individuals in \na population. When training is increased (compare results from spaced training \n(figure 4E) versus single cycle (figure 4C,D)), flies retain memory for greater \ntime periods, such that memory 24 hrs after 10x spaced training is similar to 3 or \n5 hr memory after single cycle training. This indicates that in flies as well as \nhumans, the amount of training or study strongly affects the rate of forgetting.  \n \nDuring our studies, we observed that naïve flies distribute evenly when given a \nchoice between the two odors used in our study. However, when flies were \nretested with the two odors after an initial odor choice, they preferentially chose \nthe opposite \nodor to the one they had initially chosen. This behavior may have \nsome similarities to a behavior known as Buridan’s paradigm in which a fly will \ncontinuously walk back and forth between two dark lines at opposite ends of an \notherwise featureless arena (Colomb, Reiter, Blaszkiewicz, Wessnitzer, & \nBrembs, 2012; Han, Wei, Tseng, & Lo, 2021). Buridan’s paradigm suggests that \nflies tend to alternate when given two choices instead of focusing on one or the \nother. Alternatively, it is possible that the T-maze used during testing is slightly \naversive to flies. Thus, when flies choose one odor in one arm of the maze, this \nodor becomes associated with an aversive experience, which then affects the \nretest. Consistent with this idea, we find that the opposite preference behavior \nbecomes slightly weaker when the plastic arms of the T-maze are coated with \nfilter paper, and is abolished when the arms of the T-maze are coated with a \nsucrose reward \n(data not shown).  \n \nWhy does \nforgetting consist of a decrease in the probability that a memory \ninduces a behavior? It seems likely that learning and memory originally evolved \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\nto increase the survival of an organism. If this is the case, forgetting should also \nhave evolved as a calculated survival response to a risk. In contrast to the \nsurvival strategy of humans, who have relatively few children per parent and \nraise each child carefully, the survival strategy of flies involves mass production \nof progeny. In this situation, having a certain population of progeny that chooses \nthe incorrect odor may be evolutionarily beneficial, since it can prevent \ncatastrophic loss of the entire population in cases where a situation rapidly \nchanges. Thus, the increase in the proportion of flies choosing the incorrect odor \nmay reflect a calculation that the reliability of a memory as a future predictor \ndecreases as a memory gets older. Preserving the actual memory (M) itself may \nbe important for future events, such as accelerated reinstatement of a behavior if \nthe odor-shock association is re-experienced (sensitization). Thus, decreasing \nP(C|M) over time may be an optimal method of maintaining some memory of an \nassociation while at the same time decreasing its effect on behavior as time \nprogresses. It is fascinating to consider that this type of survival strategy may \nhave evolved into more complex forgetting in\n humans and other mammals.  \n \nMemory retention curves in flies resemble those of humans,\n first described by \nEbbinghaus in 1885 (Ebbinghaus, 1885; Ebbinghaus, 2013). While the time \nscales are vastly different between these curves, their overall shapes are similar, \nwith a rapid reduction in memory at early time points that becomes more \ngradual over time. This suggests that forgetting in flies shares\n aspects of \nforgetting with humans. Consistent with this idea, aging affects memory \nretention in flies, similar to humans and other animals, again suggesting \nsimilarities between forgetting in flies and humans. \n \n \n \nMaterials and Methods \n \nFly lines and maintenance \nw(CS) flies were used in all experiments in this study. Flies were raised at 25 °\nC \nand 60% humidity on at 12 hr: 12 hr light-dark cycle. 3 to 5-day-old flies were \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\nused for young flies, and 20-day-old flies were used for old flies. Flies were aged \nin food vials containing ~50 flies each and were transferred into new vials every \ntwo to three days.  \n \nOlfactory associative training protocols \nThe aversive olfactory associative training procedure has been previously \ndescribed \n(Tamura et al., 2003). Briefly, ~100 flies were placed in a training \nchamber where they were exposed to an odor (the shock-paired conditioned \nstimulus, or CS+) for 1 min and simultaneously shocked 12 times for 1.5 secs at \n60 volts every 5 secs. After a 45 sec air purge, flies were exposed to a second \nodor (the unpaired odor or CS-) for 1 min in the absence of electrical shocks. If \nflies learned, they should associate the CS+, but not the CS-, with pain. Spaced \ntraining consisted \nof ten repeated trainings with 15 min rest intervals between \neach training cycle. All behavioral experiments were repeated 8 to 16 times for \neach time point, alternating 1-octanol and 4 methyl cyclohexanol for the CS+ and \nCS- \nodors. \n \nTesting protocols \nFor testing, flies were \nplaced in an elevator and manually moved to the choice \npoint of a T-maze where they were allowed to choose between the CS+ and CS- \nodors for 90 seconds. Immediately after the initial test, flies that chose each odor \nwere retested with the same odors. \n \nMathematical modelling \nWe investigated t\nime-related changes in memory [Pt(M)] and certainty about \nodor associations [Pt(C1|M)] based on equations (2-6) described in the results. We \nused measured ratios, Rt(C1), Rt(C2|C1’), Rt(C2|C1), and R(T), as estimates for \nPt(C1), Pt(C2|C1’), Pt(C2|C1), and P(T), leaving Pt(M), Pt(C1|M), Pt(C2|MC1), and \nPt(C2|MC1') as unknowns.. Since the number of variables exceeded the number of \nequations, we were unable to obtain precise values, but by restricting Pt(M) to \nvalues between 0 and 1, and restricting Pt(C1|M), Pt(C2|MC1), and Pt(C2|MC1') to \nvalues between 0.5 and 1, we calculated possible ranges for Pt(M), Pt(C1|M), \nPt(C2|MC1) and Pt(C2|MC1') that were used to generate figures 6 and 8. All \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\nmodeling and calculations were performed using Microsoft Excel and GraphPad \nPrism software. \n \n \nAcknowledgments \nThis work was supported by Japan Society for Promotion of Science grants to J.H. \n(21K06403,18K06496) and M.S. (19H01013, 21K18238). \n \n \n \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\nReferences \n \n \nBerry, J. A., Cervantes-Sandoval, I., Nicholas, E. P., & Davis, R. L. 2012. Dopamine \nis required for learning and forgetting in Drosophila. Neuron, 74(3): 530-\n542. \nBerry, J. A., & Davis, R. L. 2014. Active forgetting of olfactory memories in \nDrosophila. Prog Brain Res, 208: 39-62. \nColomb, J., Reiter, L., Blaszkiewicz, J., Wessnitzer, J., & Brembs, B. 2012. Open \nsource tracking and analysis of adult Drosophila locomotion in Buridan's \nparadigm with and without visual targets. PLoS One, 7(8): e42247. \nDavis, R. L., & Zhong, Y. 2017. The Biology of Forgetting-A Perspective. Neuron, \n95(3): 490-503. \nEbbinghaus, H. 1885. Memory: A contribution to experimental psychology. \nEbbinghaus, H. 2013. Memory: a contribution to experimental psychology. Ann \nNeurosci, 20(4): 155-156. \nGao, Y., Shuai, Y., Zhang, X., Peng, Y., Wang, L., He, J., Zhong, Y., & Li, Q. 2019. \nGenetic dissection of active forgetting in labile and consolidated \nmemories in Drosophila. Proc Natl Acad Sci U S A, 116(42): 21191-\n21197. \nHan, R., \nWei, T. M., Tseng, S. C., & Lo, C. C. 2021. Characterizing approach \nbehavior of Drosophila melanogaster in Buridan's paradigm. PLoS One, \n16(1): e0245990. \nHoriuchi, J. 2019. Recurrent loops: Incorporating prediction error and \nsemantic/episodic theories into Drosophila associative memory models. \nGenes Brain Behav, 18(8): e12567. \nHoriuchi, J., Yamazaki, D., Naganos, S., Aigaki, T., & Saitoe, M. 2008. Protein kinase \nA inhibits a consolidated form of memory in Drosophila. Proc Natl Acad \nSci U S A, 105(52): 20976-20981. \nMery, F. 2007. Aging and its differential effects on consolidated memory forms in \nDrosophila. Exp Gerontol, 42(1-2): 99-101. \nQuinn, W. G., Harris, W. A., & Benzer, S. 1974. Conditioned behavior in Drosophila \nmelanogaster. Proc Natl Acad Sci U S A,\n 71(3): 708-712. \nShuai, Y., Hirokawa, A., Ai, Y., Zhang, M., Li, W., & Zhong, Y. 2015. Dissecting \nneural pathways for forgetting in Drosophila olfactory aversive memory. \nProc Natl Acad Sci U S A, 112(48): E6663-6672. \nShuai, Y., Lu, B., Hu, Y., Wang, L., Sun, K., & Zhong, Y. 2010. Forgetting is regulated \nthrough Rac activity in Drosophila. Cell, 140(4): 579-589. \nTamura, T., Chiang, A. S., Ito, N., Liu, H. P., Horiuchi, J., Tully, T., & Saitoe, M. 2003. \nAging specifically impairs amnesiac\n-dependent memory in Drosophila. \nNeuron, 40(5): 1003-1011. \nTully, T., Boynton, S., Brandes, C., Dura, J. M., Mihalek, R., Preat, T., & Villella, A. \n1990. Genetic dissection of memory formation in Drosophila \nmelanogaster. Cold Spring Harb Symp Quant Biol, 55: 203-211. \nTully, T., Preat, T., Boynton, S. C., & Del Vecchio, M. 1994. Genetic dissection of \nconsolidated memory in Drosophila. Cell, 79(1): 35-47. \nTully, T., & Quinn, W. G. 1985. Classical conditioning and retention in normal and \nmutant Drosophila melanogaster. J Comp Physiol A, 157(2): 263-277. \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\nYamazaki, D., Horiuchi, J., Miyashita, T., & Saitoe, M. 2010. Acute inhibition of PKA \nactivity at old ages ameliorates age-related memory impairment in \nDrosophila. J Neurosci, 30(46): 15573-15577. \n \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\nFigures \n \n \n \nFigure 1. \nMemory retention curve. Flies were trained in an odor/shock \nassociation task as described in the text and tested at indicated times after \ntraining. R(C1), the percentage of flies choosing the correct odor, is plotted as a \nfunction of retention time. Data is used with permission from Tamura et al., 2003 \n(Tamura et al., 2003). \n \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \nFigure 2. \nThere is no significant difference in learning between flies that chose \nthe correct or incorrect odors immediately after aversive olfactory conditioning. \nR(C1) refers to the percentage of flies that chose the correct (non-shocked) odor \nin the initial test immediately after training. R(C2|C1) refers to the percentage of \nflies that chose the correct odor in the retest after choosing the correct odor in \nthe initial \ntest, and R(C2|C1’) refers to the percentage of flies that chose the \ncorrect odor in the retest after initially choosing the incorrect odor.  NS indicates \np>0.5. \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \nFigure 3. \nNon-associative effects influence odor retest assays 3 hrs after \nolfactory conditioning. The experiment is similar to that of Figure 1 except that \nthe odor preference test and subsequent retest were performed 3 hrs after \nassociative training. Flies that initially chose correctly, R(C2|C1), show \nsignificantly lower scores upon retest, while flies that initially chose incorrectly \nshow significantly higher scores, R(C2|C1’), suggesting the existence of non-\nassociative effects. **, p<0.01; ***, p<0.001. \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \nFigure 4. \nOpposite odor preference. While naïve flies distribute evenly between \ntwo odors (Oct and MCH) when initially tested for odor preference, those that \ninitially chose Oct prefer MCH when retested, and those that initially chose MCH \nprefer Oct when retested. NS, p>0.05; **, p<0.01; ***, p<0.001. \n \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\n \n \nFigure 5. Initial odor preferences and retest odor preferences at 3 min (A), 2 hrs \n(B), 3 hrs (C), 5 hrs (D), and 24 hrs (E) after associative olfactory conditioning. \nThe opposite odor preferences during retesting of untrained flies, R(T), is shown \nfor comparison in each case. For A, B, C, and D, conditioning consisted of a single \ntraining trial, while for E, spaced conditioning was used. *, p<0.05; **, p<0.01; \n***, p<0.001. \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\n \nFigure 6. (A) Ranges for Pt(M) at indicated times after training. (B) Ranges for \nPt(C|M) at indicated times after training.  \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\n \n \nFigure 7. Initial odor preferences and retest odor preferences after olfactory \nconditioning of old (20-day-old) flies. *, p<0.05; **, p<0.01; ***, p<0.001. \n \n  \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint \n\n \nFigure 8. Ranges for (A) Pt(M) and (B) Pt(C|M) at indicated retention times for \n20-day-old flies. \n \n.CC-BY 4.0 International licenseavailable under a \n(which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made \nThe copyright holder for this preprintthis version posted June 29, 2025. ; https://doi.org/10.1101/2025.06.26.661725doi: bioRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}