Ontogeny of working memory and behavioural flexibility in the free movement pattern (FMP) Y-maze in zebrafish

The acquisition of executive skills such as working memory, decision-making and adaptive responding occur at different stages of central nervous system development. Zebrafish ( Danio rerio ) are increasingly used in behavioural neuroscience for complex behavioural tasks, and there is a critical need to understand the ontogeny of their executive functions. Zebrafish across developmental stages (4, 7, 14, 30 and 90 days post fertilisation (dpf)), were assessed to track development of working memory (WM) and behavioural flexibility (BF) using the free movement pattern Y-maze (FMP Y-maze). Several differences in both WM and BF were identified during the transition from yolk-dependent to independent feeding. Specifically, WM is evident in all age groups, even from 4 dpf. However, BF is not developed until larvae start free feeding, and show significant improvement thereafter, with young adults (90 dpf) demonstrating the most well-defined BF. We demonstrate, for the first time, objective WM processes in 4 dpf zebrafish larvae. This suggests that those wishing to study WM in zebrafish may be able to do so from 4 dpf, thus drastically increasing throughput. In addition, we show that zebrafish follow distinct stages of cognitive development and age-related changes during the early developmental period. Finally, our findings indicate distinct WM and BF mechanisms, which may be useful to study for translational purposes.


Introduction
Maturation of the human central nervous system (CNS) results in altered behaviour and cognitive function (Lupien et al., 2009;Roberts and Lopez-Duran, 2019).Drastic changes in cognitive development take place during the first 10 years of life, enabling the use of more complex decision-making and increases in cognitive control over behaviour, thus resulting in a reduction in reflexive or impulsive responses and increased capacity for planning and cognitive flexibility (Luna et al., 2001).These developments likely reflect morphological changes in the brain to near-adult proportions (Brenhouse and Andersen, 2011;Cromer et al., 2015;Gogtay et al., 2004).However, development continues into adulthood and, particularly in the cognitive domain, the rate of maturation and the age at which adult-levels of ability are achieved varies (Cromer et al., 2015;Gogtay et al., 2004;Lewis et al., 2010;Luciana and Nelson, 2002).For example, frontal lobe functions, such as working memory (WM), reach adult-levels of performance during late-teens to early twenties (Luna et al., 2004;Zald and Iacono, 1998).Indeed, cognitive development has largely been reported as a linear progression of increasing ability from infant to adult (Brenhouse and Andersen, 2011;Cromer et al., 2015).By unravelling how these cognitive processes evolve from childhood to adulthood, it will be possible to identify factors influencing learning, decision-making, and adaptability, leading to improved educational and therapeutic interventions for cognitive challenges across the lifespan.
Zebrafish (Danio rerio) are useful for modelling prenatal, postnatal and early-life (pre-adult) investigation into disrupted neural circuitry and associated behavioural changes that can persist into adulthood (de Abreu et al., 2019;Kozol, 2018;Souza and Tropepe, 2011;Stewart et al., 2014).Zebrafish possess several advantages for assessing the development of cognitive and biological functions.Primarily, the distinction of in utero development of mammals compared to in vitro development of zebrafish larvae permits examination of early developmental stages, while avoiding maternal interference (Ge et al., 2019).Therefore, behavioural examination can begin whilst larvae are still yolk-dependent, compared to free-feeding, allowing comparison of cognitive and behavioural abilities and associated biochemistry at distinct life stages.Although previous studies have investigated behavioural endpoints in zebrafish at several stages of development (eg Valente et al., 2012), to our knowledge, no one has yet presented a comprehensive examination of cognitive development in zebrafish, from larvae to adulthood.
The Free Movement Pattern (FMP) Y-maze has been used in several studies to investigate changes in working memory (WM) and behavioural flexibility (BF) by analysing sequence strings of left and right turn choices made during one hour (1 h) of free exploration by zebrafisfh (Cleal et al., 2020;Cleal and Parker, 2018;Fontana et al., 2019aFontana et al., , 2019b)).The 'chunking' of turn choices into overlapping sequences of four turns, known as tetragrams, allows the identification of repeating patterns within the data.Consecutive alternations (RLRL, LRLR) are significantly over-represented as a search pattern, and drugs that block WM (MK-801, scopolamine, SCH23390) abolish this effect (Cleal et al., 2020).BF can be operationally defined in terms of within-trial adaptations in proportion of consecutive alternations during a trial, which again, is reversed by SCH23390 and MK801 (Cleal et al., 2020) and by amphetamine, which enhances 'habit' like behaviour in the early part of the trial (Cleal et al., 2021).Here, we exploited the FMP Y-maze to investigate the impact of developmental stage on cognitive abilities, testing a cross-section of five age groups in the FMP Y-maze, ranging from yolk-dependent larvae to young adult.We chose the FMP Y-maze to characterise WM and BF for several reasons: 1) It is non-invasive, and requires no intervention to the animal; 2) it is validated in several species (by our group and others) as being a measure of WM and BF; 3) the nature of the task (i.e.simply filming the fish swimming) is such that it can be used at any age without major modification of the protocol, and is not associated with extensive training etc.; 4) there is potential for high throughput with animals tested in parallel.

Animals and housing
Zebrafish were bred in-house at the University of Portsmouth Fish Facility, with eggs collected via the addition of marble-trays to tanks of ~20 adult, AB zebrafish.Offspring were reared in petri-dishes of aquarium-treated water in groups of ~50 embryos from collection until 5 days post fertilisation (dpf) and maintained in a larval incubator at ~28.5 • C on a 14/10-hour light/dark cycle.From 5 dpf larvae were transferred into larger dishes in the incubator.From ~20 dpf larvae were split into groups of 20 fish in 6 L tanks, until 60 dpf when fish were further split into groups of 8-12 fish in 2.8 L tanks on a re-circulating system (Aquaneering Inc., San Diego, CA, USA).From 5 dpf fish were fed on ZM fry food until 30 dpf and then put on a diet of dried fish flakes and live brine shrimp until 90 dpf, 3 times per day during the week and once a day at weekends (am).Aquarium water was maintained at ~25-27 • C, pH 8.4 ( ± 0.4), and fish remained on a 14/10-hour light/ dark cycle.All fish used were experimentally naïve, and any non-drugexposed fish were returned to housing tanks and incorporated into the breeding programme after use.

Ethical statement
Experiments carried out as part of this study were under license from the UK Home Office (Animals (Scientific Procedures) Act, 1986) [PPL: P9D87106F] and with approval from and in accordance with the University of Portsmouth Animal Welfare and Ethical Review Board guidelines.

FMP Y-maze apparatus
For testing larval, juvenile and young adult zebrafish, three different sizes of Y-maze were used (Fig. 1) all with clear acrylic bases and white acrylic walls consisting of 3 arms of equal proportions, at 120 • angle.All mazes were inserted into Zantiks MWP (larval/juvenile) or AD (adult) behavioural unit (Zantiks ltd, Cambridge, UK).The dimensions were as follows: Larval mazes (4-7 dpf): arms were length 1.2 x width 0.4 x depth 0.3 cm (MWP); Juvenile (14-30 dpf): length 2.5 x width 1.0 x depth 0.5 cm (MWP); Adult (90 dpf): length 5 x width 2 x depth 14 cm (AD).For all ages, mazes were filled with aquarium water to allow sufficient swimming depth (3 L for adult, others, maximally filled).For older fish (90 days) we used 50:50 male:female.Sex differences were not analysed, as we have previously seen no sex-related differences in performance (Fontana et al., 2019b).
All units were web-enabled and capable of live streaming video feed of in test activity of zebrafish to a laptop or mobile device.Starting and finishing trials was remote, and logging of arm entries was written into the programme script, recorded automatically and output in an excel spreadsheet.Cameras were located above the maze and enable continuous tracking of individual animals throughout the trial.Experimenter visibility and handling were kept to a minimum to reduce stress and distraction whilst animals were exploring the maze.Ambient light was a maximum of 2 lux in each of the testing units.

Protocol
One-hundred and fifteen larval (4 dpf n = 24, 7 dpf n = 21), juvenile (14 and 30 dpf, n = 52) and young adult (90 dpf, n = 18) AB zebrafish were used for assessing differences in WM and BF in zebrafish developing from larvae to young adults.Sample sizes were calculated based on power analyses from our previous published work, pilot studies, and based on α-levels of 0.05 and power (β) of 0.8 (Cleal and Parker, 2018;Fontana et al., 2019aFontana et al., , 2019b)).Fish were experimentally naïve, and each fish was only used in one trial.There was no training or habituation required prior to testing in the FMP Y-maze.All fish were randomly selected from multiple tanks and clutches prior to testing.For example, for larvae n = 45 (4 dpf n = 24, 7 dpf n = 21) animals were chosen from 5 rearing containers, each containing ~50 embryos.Fish were individually netted or pipetted (dependent on size) directly from home tanks (or dishes) into their respective mazes and placed into behavioural units.Behavioural tests were conducted in Zantiks fully automated behavioural testing units for adults [AD unit] and larvae/juveniles [MWP unit] (Zantiks Ltd., Cambridge, UK).Each fish was free to explore the maze for 1 h, and no changes in environment, novelty or food rewards were introduced during the trial.Logging of data pertaining to arm entries were recorded automatically and analysed (see Section 2.5) on completion of all testing groups as performed in previous studies (Cleal et al., 2020;Cleal and Parker, 2018;Fontana et al., 2019aFontana et al., , 2019b)).When pilot testing the 4 dpf larvae, the majority showed almost no detectable movement in the maze.We therefore induced movement by adding the TRPA1 agonist allyl-isothiocyanate (AIC), which is well established to induce movement in larval zebrafish (Prober et al., 2008) and these treated larvae were used throughout testing.

Data processing and statistical analysis
Y-maze data were collected as a series of binary left and right turns, which in the absence of bias, would be equally likely to be selected (Deacon et al., 2006;Frith and Done, 1983;Gerlai, 1998;Stroe-Kunold et al., 2009).To determine if turn choices in the FMP Y-maze followed a Markov process (ie., chosen randomly) we employed a two-choice guessing task analysis (Frith and Done, 1983) to exploit information theory to detect patterns in large data series by 'chunking' turn choices into small groups of information (i.e.groups of four turns, left, left, right, right-LLRR) (Guze, 1993).We used a third order uncertainty to chunk consecutive turn choices into overlapping groups of four turns, limiting the number of alternative sequences to 2 4 = 16 possible turn configurations, known as tetragrams (LLLL, LLRL, …, RRLR, RRRR).If tetragram selection was completely random, a Markov or 'memoryless' process, the relative frequency of all tetragram sequences would be equal at approximately 6.25 % (100 % of turns/16 tetragram possibilities).Higher rates would be evidence of strategic tetragram selection (Gross et al., 2011).We transformed raw data (arm entry) into tetragram sequences of overlapping choices of turn (i.e. if [L,R,{L,R],R,R} were the first 6 turns, the first 3 tetragram sequences would be; [LRLR], RLRR, {LRRR}).In several previous studies, we have highlighted the prominant role for two particular tetragram configurations, which comprised consecutive alternations (LRLR, RLRL) and repetitions (LLLL, RRRR); therefore particular attention was paid to these patterns (Cleal and Parker, 2018;Fontana et al., 2019aFontana et al., , 2019b;;Gross et al., 2011).
Data were split into two formats for analysis; total percentage use (calculated as a proportion of total turns) of each tetragram sequence for the 1 h of exploration, referred to as 'global' search strategy.This was used as a measure of WM as repetition of previous turn choices must be remembered for patterns of movement to be repeated over 1 h of exploration, requiring memory of arm entries and order of entry.The second type of strategy, known as an 'immediate' search strategy, analysed search pattern configurations over 10 min time bins throughout the trial, equating to six equal, consecutive time bins.This analysis was used to assess BF.Animals that are not able to update information gained during exploration of the FMP Y-maze will likely perform similar strategies over each time bin.However, those that can change behaviour in response to new information would be expected to change movement patterns over time, therefore differences in tetragram usage would be expected with each 10 min time bin.
Data were analysed in GraphPad Prism V9 for Macintosh.Y-maze data were analysed in two ways.For comparing strategies within age groups, we used either a One-way or Two-Way ANOVA.Levine's test assessed homogeneity of variances.Shapiro-Wilk test of normality and boxplot with Tukey confidence interval was used to identify extreme outliers.Tukey's test for multiple comparisons was used as a post hoc analysis.Locomotion was analysed using total turns completed during 1 h of exploration in the FMP Y-maze.Results were considered significant when p ≤ 0.05.Two adult zebrafish (90 dpf, #5, #7) were excluded from analysis due to observed tracking errors during testing.Three extreme outliers (>3 *IQR) were excluded from locomotion analysis (30 dpf, #2, #18, #26), again, likely caused by tracking errors.

Fig. 2 G displays locomotor activity over the full hour of exploration.
There was a significant effect of age on total turns during 1 h of search [One-way ANOVA, F (5, 241) = 150.8,p < 0.0001], demonstrating an increased number of turns with increasing age (Fig. 2G).Tukey's post hoc analysis revealed that all age groups travelled significantly less than 90 dpf young adults.

Discussion
Here, we aimed to understand how executive functions in zebrafish are affected by developmental stage using the FMP Y-maze.Larvae dependent on the yolk-sac showed a repetitive bias towards alternations, at the exclusion of all other strategies.However, a strategic pattern of exploration, inclusive of all 16 tetragram sequences, emerged in freefeeding larvae, and was maintained in juveniles and young adults, all of which showed preferential use of alternation sequences consisting of consecutive left and right turns (LRLR, RLRL), which we and others have (caption on next page) M. Cleal et al. shown to reflect WM (Cleal et al., 2020;Ismail et al., 2022), whilst still using all other strategies for exploration.We further revealed that BF, defined as a changeable strategy which adapted during trial progression, was also only evident from free-feeding (7 dpf), and showed a steady increase in performance as a function of age, up to young adult zebrafish (90 dpf).
The ability to learn from experience and to recall information about the surrounding environment is critical to an organism's survival and ability to exploit their natural habitat (Roy and Bhat, 2016).Numerous studies across different vertebrate species reveal sophisticated behaviour regarding the use of cognitive control to appropriate action selection in the most efficient manner (Howard et al., 2017;Kottler et al., 2019;Monsell and Driver, 2000).However, in order to achieve adult optimised performance, there is continuous development and improvement throughout larval and juvenile stages, not only through synaptic remodelling and plasticity, but also through distinct ontogenic stages (Anderson et al., 2004;Browman and O'Brien, 1992;Grecian et al., 2018;Osborne et al., 2013).
In the FMP Y-maze young 4 dpf larvae were the only age group to search the maze with almost complete exclusivity of search strategy (relying predominantly on alternations and no other pattern of movement), with all tetragram configurations used below 5 %.Ontogenic stage of zebrafish therefore appears to play a critical role in determining the time that sophisticated search strategies emerge.For example, zebrafish have very low motility rates during the early stages of development, whilst dependent on the yolk-sac for survival (Colwill and Creton, 2011).However, free-swimming occurs once larvae are independent feeders and have begun foraging for live prey.This is reported at 5 dpf when resources from the yolk sac have been almost completely consumed and swimming activity and duration have increased in response to the need to search for food (Strähle et al., 2012).Dependence on an available food source, and incomplete development of neural structures could result in under-developed cognitive capacity.The lack of systematic search strategy in 4 dpf larvae may be due to the continued development of anatomical structures and axonal pathways.Analysis of dopaminergic development, critical to motor activity and executive function, showed that despite most dopaminergic neurons being present by 4 dpf, many axonal projections, such as those into the telencephalon (equivalent to prefrontal cortex in mammals, and a crucial area for learning and memory) (Salas et al., 2006) are not completed until 5 dpf (Du et al., 2016).This may provide a potential explanation as to why 4 dpf larvae are unable to develop strategies using all patterns of movement and instead limit their exploration to a single strategy to navigate the maze (Hernandez et al., 2018;Roy and Bhat, 2016).
Older age groups, unlike yolk-dependent larvae, were observed using a specific global search strategy for the full period of exploration in the FMP Y-maze.Although the majority of search configurations were selected randomly, there was evidence of dominant, intermitted use of alternation strategies that relied on equal left and right turns (LRLR, RLRL), performed at a higher frequency than chance.Previous work by our group has demonstrated alternations reflect WM, with drugs that block WM also removing alternations (Cleal et al., 2020).Here we demonstrate that 4 dpf used alternations at least as effectively as juveniles and adults, demonstrating that WM is being utilised during exploration of their environment.
A key aspect of animals with higher cognitive functions, such as learning and memory, is the ability to update their knowledge of a search area based on recent experiences and adapt their behaviour accordingly (Bracis et al., 2015).The FMP Y-maze protocol provides a negative feed-back task, in which animals exploring the maze are continually presented with an absence of novel stimuli, e.g., food or reward.This would be expected to cause a strategy change, based on an animal's ability to alter behaviour in response to environmental cues.We have previously demonstrated that adult zebrafish (6 months old) are able to use information obtained exploring the FMP Y-maze to alter search strategy by increasing the use of alternations over time (Cleal et al., 2020).However, this ability is strongly affected by senescence or inhibiting memory forming or retrieval pathways (Cleal et al., 2020).4 dpf larvae were the only group to show no impact of time on alternation use.The other age groups showed some progression of alternations overtime, demonstrating the initial ability to flexibly update their search strategy over time.However, none showed it to the same extent as young adults (90 dpf), the oldest age group tested, showed the greatest up-regulation of alternation use with increasing time.Young adult zebrafish linearly shifted alternation-use from below chance, to greater than one fifth of movement patterns from the first to last 10 min of exploration, demonstrating substantial overlap with the adult strategy.Changes in alternation use, during extensive exploration, supports a role for experience based, individual learning, influencing search strategy and is reflective of the ability to adapt to their environment.The absence of this behaviour in yolk-dependent larvae is quite possibly reflective of the lack of need for this ability at this developmental stage, due to them being dependent on the yolk sac for food.Whilst other age groups show some degree of flexibility, it is much less than that of the young adults and is thus reflective of an immature cognitive performance that has not yet reached adult-level abilities.
There were several limitations of this study, which should be considered.First, we used a cross section of ages.It is currently unclear whether individual differences could be observed in the developmental process, and indeed, the extent to which this could be observed in FMP Y-maze exploration strategies.Future work may consider FMP-Y maze performance longitudinally in order to consider individual differences.Work in fruit flies (Ayroles et al., 2015) using similar Y maze Fig. 2. Locomotor assessment in the FMP Y-maze reveals age related development of working memory and behavioural flexibility.(A) Schematic of experimental paradigm for tracking free swimming zebrafish and recording of turn data.Fish execute discrete spontaneous turns that comprise a sequence of turn behaviour, which has been broken down in tetragrams to identify patterns in their movement.(B) Image representing the different ages of zebrafish that were examined as part of this study to identify developmental stages at which changes in memory and flexibility can be observed.The ages represented here are from yolk dependent larvae to young adult.(C) Global tetragram usage during FMP Y-maze exploration.Results are given as a percentage of the total number of tetragrams completed in 1 h of exploration and represent the search strategy used to navigate the FMP Y-maze for 4 dpf (+AIC, n = 47), 7 dpf (n = 21), 14 dpf (n = 25), 30 dpf (n = 24) and 90 dpf (n = 16) zebrafish.Patterns show inflated use of strategies consisting of alternating left and right turns (LRLR and RLRL) know as an alternation, which is the only strategy which is represented above random (denoted by the dashed line at 6.25 %).Error bars are + SEM (D) Percentage use of each pair of tetragram sequences (e. g., RRLL+LLRR, LLRL+RRLR).Fish showed a bias towards alternating left and right turn above other sequences, therefore, all sequences were compared to alternations (LRLR+RLRL) in a One-Way ANOVA.Multiple comparisons revealed that alternations were used significantly more than any other strategy, irrespective of age.methodologies has demonstrated clear individual differences in performance, and this would be very interesting to track in a vertebrate to compare.Second, the neural circuits underlying executive function in zebrafish are not presently clear, raising questions about the translational relevance of ontogeny of executive function in this species to, for example, mammals.Third, we included 4 dpf larvae, as we wanted to include both free-feeding and yolk-dependent animals for comparison.However, yolk-feeding larvae were comparatively inactive compared to other age groups; therefore, in order to identify any strategies, it was required for them to be pre-treated with AIC to increase locomotion.It is plausible that the differences observed in BF in the 4dpf larvae were the result of locomotor differences caused by AIC rather than relating to different ages per se.It would not be ethical to incubate the older animals in AIC, of course, as there would be limited value in understanding this beyond the scope of the present study.Therefore, although AIC successfully increases locomotion to the point that a search pattern could be deciphered, no testing has been conducted to investigate the impact of this treatment itself on search strategy.For these reasons, the 4 dpf data should be treated with caution until further testing is carried out.

Conclusion
Here, we present, for the first time, an in-depth analysis of the executive functioning capabilities of developing zebrafish from larval to adult stages.WM was observed in the behaviour of all tested ages of zebrafish.BF was not evident until 7 dpf, but was far from adult levels of ability and showed significant change with increasing age up to young adult.Critically this suggests that groups wishing to study WM in zebrafish may be able to do this from 4 dpf, thus drastically improving the throughput of their work.More broadly, the work supports the theory that an immature executive system in developing fish that does not fully mature until adulthood.It is currently unclear what the underlying mechanisms are for ontogeny of executive function in zebrafish, but our data suggest that they may be an excellent model system for studying this important question in translational neuroscience.Finally, it is critical to consider the intricate role of glutamate in learning and memory, and variations in the glutamatergic system throughout the zebrafish lifespan.The dynamic nature of the glutamatergic system, which transforms from an inhibitory to an excitatory neurotransmitter system from prenatal to postnatal stages in mammals, raises intriguing questions about how these developmental shifts influence cognitive processes.The zebrafish, in which early larval stages of development are equivalent to prenatal stages in mammals, potentially presents a unique model to explore this phenomenon.

Funding
MC was funded by a University of Portsmouth Faculty Ph.D. Studentship.BDF was sponsored by a CAPES foundation studentship (Brazil).CH was funded by Dstl (MOD, UK).MOP receives funding from NC3Rs (NC/W00092X/1).The funders played no role in the design or execution of these experiments, nor in our decision to publish.

Declaration of Competing Interest
The authors declare no conflicts of interest.
Fig. 2.Locomotor assessment in the FMP Y-maze reveals age related development of working memory and behavioural flexibility.(A) Schematic of experimental paradigm for tracking free swimming zebrafish and recording of turn data.Fish execute discrete spontaneous turns that comprise a sequence of turn behaviour, which has been broken down in tetragrams to identify patterns in their movement.(B) Image representing the different ages of zebrafish that were examined as part of this study to identify developmental stages at which changes in memory and flexibility can be observed.The ages represented here are from yolk dependent larvae to young adult.(C) Global tetragram usage during FMP Y-maze exploration.Results are given as a percentage of the total number of tetragrams completed in 1 h of exploration and represent the search strategy used to navigate the FMP Y-maze for 4 dpf (+AIC, n = 47), 7 dpf (n = 21), 14 dpf (n = 25), 30 dpf (n = 24) and 90 dpf (n = 16) zebrafish.Patterns show inflated use of strategies consisting of alternating left and right turns (LRLR and RLRL) know as an alternation, which is the only strategy which is represented above random (denoted by the dashed line at 6.25 %).Error bars are + SEM (D) Percentage use of each pair of tetragram sequences (e. g., RRLL+LLRR, LLRL+RRLR).Fish showed a bias towards alternating left and right turn above other sequences, therefore, all sequences were compared to alternations (LRLR+RLRL) in a One-Way ANOVA.Multiple comparisons revealed that alternations were used significantly more than any other strategy, irrespective of age.Chance selection is represented by the dashed line (~6.25 %).Error bars are ± SEM, * ** *p ≤ 0.0001.(E) Age comparison of total alternations (LRLR+RLRL) used during FMP Y-maze exploration.One-way ANOVA reveals a significant effect of age on alternation use.Chance selection is represented by the dashed line (~6.25 %).Error bars are + SEM, *p ≤ 0.05.(F) Total number of alternations used per 10 min time bin during 1 h exploration in the FMP Y-maze.Two-Way ANOVA reveals a significant effect of age (****p ≤ 0.0001) and time (****p ≤ 0.0001) on alternation use.Individual analysis of each age group revealed a trend of increasing age with increasing use of alternations over time from 7 dpf to 90 dpf.Only 4 dpf, showed no significant effect of time on alternation use.Chance selection is represented by the dashed line (~6.25 %).Error bars are ± SEM, nsnot significant, *p ≤ 0.05, * *p ≤ 0.01, * ** *p ≤ 0.0001.(G) Locomotor activity measured by total number of turns completed in 1 h of free exploration.One-Way ANOVA revealed a significant effect of age on increasing locomotion during exploration.Error bars are ± SEM, * ** *p ≤ 0.0001.M.Cleal et al.