[1] 42w This article has been accepted for publication and undergone full peer review but has not been through the copyediting, typesetting, pagination and proofreading process which may lead to differences between this version and the Version of Record. Please cite this article as

[1] 255w Background and PurposeThe therapeutic effects of fluoxetine (FLX) are believed to be due to its potency for increasing neuronal plasticity and reversing some learning deficits. Nevertheless, a growing amount of evidence shows the adverse effects of the drug on cognition and some forms of neuronal plasticity. Experimental ApproachTo study the effects of chronic FLX treatment we combine an automated assessment of motivation and learning in mice with an investigation of various forms of neuronal plasticity in the central (CeA) and basolateral amygdala (BLA). We use immunohistochemistry to visualize neuronal types and perineuronal nets (PNN), and DI-staining to assess dendritic spine morphology. Gel zymography is used to test FLX's impact on matrix metalloproteinase-9 (MMP-9), an enzyme involved in synaptic plasticity. Key ResultsWe show that chronic FLX treatment in non-stressed mice increases PNN-dependent plasticity in the BLA, while simultaneously impairing MMP-9-dependent plasticity in the CeA. Further, we illustrate how the latter contributes to anhedonia and deficits of reward learning. Behavioral impairments are accompanied by alterations in morphology of dendritic spines in the CeA towards a more immature state, most likely reflecting animals' inability to adapt. We strengthen the link between the adverse effects of FLX and its influence on MMP-9 by showing that behavior of MMP-9 knock-out animals remains unaffected by the drug. Conclusion and ImplicationsIn conclusion, chronic FLX treatment differentially affects various forms of neuronal plasticity, which may explain its contradicting effects on the brain and behavior. Presented findings are of immediate clinical relevance since reported side effects of FLX pose a potential threat to patients.

[1] 81w Deterioration of motivation and related impairments of reward-driven behaviors are core symptoms of mood disorders, including most commonly diagnosed major depression and bipolar disorder (Jorge, 2015;Rakofsky and Rapaport, 2018). Brain circuitry related to reward processing, an umbrella term for motivation, salience, anticipation, pleasure, and satiety, is commonly studied in rodent models. The use of such models enables dissection of reward-driven behaviors and gives insight into the mechanisms of aberrant neuronal plasticity underlying deficits in motivation, reward learning, and reward-based decision making.

[2] 109w Fluoxetine (FLX), a selective serotonin reuptake inhibitor (SSRI) most commonly prescribed for anhedonic symptoms of mood disorders, is also frequently taken by patients without any formal diagnosis. Regrettably, studies describing influence of FLX on cognitive and emotional wellbeing of such individuals are still scarce (Serretti et al., 2010;Olguner Eker et al., 2017). However, there is increasing evidence, in both healthy animals and human subjects, that the drug hinders reward processing (Eisenreich et al., 2017;Flores-Ramirez et al., 2019;Sharp et al., 2019), and can suppress neuronal plasticity (Dringenberg et al., 2014). In particular, it has been reported that chronic FLX treatment decreases amygdala activation in healthy adults (Takahashi et al., 2005).

[3] 189w On the other hand, FLX treatment has been shown to induce neuronal plasticity and to alleviate impairments of certain types of learning, which can, at least partially, explain its therapeutic effects (Maya Vetencourt et al., 2008;Rossi, 2016;Sun et al., 2017;Eavri et al., 2018;Ampuero et al., 2019;Levy et al., 2019;Ohira et al., 2019;Steinzeig et al., 2019). Markedly, some animal studies showed that the drug reverses anhedonia induced by stress as severe as this stemming from repeated social defeat and boosts performance in highly complex cognitive tasks performed with the use of automated behavioral assays (Gottschalk et al., 2018;Marwari and Dawe, 2018). Further, FLX increases neuronal plasticity in fear memory circuitry, i.e., basolateral amygdala (BLA), by shifting neurons surrounded by perineuronal nets (PNNs) -components of the extracellular matrix essential for the synaptic stabilization (Kwok et al., 2011;Flores and Méndez, 2014) -towards a more plastic state, which reverses impairments of fear extinction (Karpova et al., 2011). Even though it is often assumed that FLX should also lessen deficits of reward learning, much less is known about the neural mechanisms through which the drug might influence related brain function (Furukawa et al., 2019).

[4] 128w Here we show that chronic FLX treatment in non-stressed mice not only loosens PNNs surrounding parvalbumin-expressing interneurons in the basolateral amygdala (in line with the previous research, cf. Karpova et al., 2011;Popova et al., 2014), but at the same time reduces matrix metalloproteinase 9 (MMP-9) activity in its central nucleus (CeA). MMP-9 is an extracellularly operating enzyme, intimately involved in synaptic plasticity, learning and memory (Dziembowska et al., 2012;Dziembowska and Wlodarczyk, 2012). Interestingly, reduced activity of MMP-9 has also been observed in patients with depression (Bobińska et al., 2016a(Bobińska et al., , 2016b)). We have previously shown that MMP-9 activity in the CeA is required for learning reward-related but not punishment-related tasks and its blocking affects neuronal plasticity (Knapska et al., 2013;Gorkiewicz et al., 2015;Knapska and Kaczmarek, 2016).

[5] 182w Notably, involvement of MMP-9 in degradation of PNNs has been well documented in both animal models and human patients (Brunswick et al., 2002;Gray et al., 2008;Pollock et al., 2014;Murase et al., 2017;Lorenzo Bozzelli et al., 2018;Wen et al., 2018b;Alaiyed and Conant, 2019;Alaiyed et al., 2019). The PPNs preferentially surround inhibitory neurons (Naegele et al., 1988;Naegele and Katz, 1990;Seeger et al., 1994;Härtig et al., 1995;Morris and Henderson, 2000), such as those expressing parvalbumin and calbindin (Vidal et al., 2006;Ohira et al., 2013;Lensjø et al., 2017;McDonald et al., 2018). The disruption of the interplay between the PNNs surrounding these cells and MMP-9 is thought to contribute to numerous neuropsychiatric and neurodegenerative disorders, including schizophrenia, Alzheimer disease and epilepsy (Wen et al., 2018b). Although the specific PNN substrate of MMP-9 binding has yet to be identified, it is known that regulation of overexpressed enzyme rescues PNN levels and related neuronal plasticity (Wen et al., 2018a;Dwir et al., 2019). Importantly, the native number of neurons surrounded by PNNs is very low in the CeA and thus plasticity related to them may not play a dominant role.

[6] 81w We observed that decrease in MMP-9 activity in the CeA after chronic FLX treatment of nonstressed mice was accompanied by reduced motivation and impaired reward learning. Notably, lack of MMP-9 protected mice from the FLX-induced motivational and cognitive changes, which further points to the direct functional link between the protein and the drug. At the morphological level, animals chronically treated with FLX had a higher proportion of immature spines in the CeA after reward-based training, which suggests alterations in synaptic transmission.

[1] 205w There are studies showing that FLX may affect matrix metalloproteinase 9 (MMP-9) activity and thus influence related neuronal plasticity (Lee et al., 2012(Lee et al., , 2014)). We analyzed the MMP-9 activity level in the CeA and the BLA of non-stressed mice subjected to the 8-week long FLX treatment. CeA and BLA were dissected from FLX-and vehicle-treated wild type (WT) animals and gel zymography was performed on the collected tissue (Fig. 1C, F Kaliszewska et al., 2012;Jasińska et al., 2016). We found that chronic FLX treatment decreased the MMP-9 activity in the CeA, while its level most likely remained unchanged in the BLA, illustrating the circuit specific impact of the drug (Fig 1A, B). However, it is noteworthy that the latter could not be unequivocally verified due to the insufficient sample size. Moreover, the negative influence of FLX was related to the longitudinal drug intake as acute administration of FLX had no effect on MMP-9 activity in either of the examined structures (Fig 1D , E). Notably, as the procedures of chronic and acute administration of the drug were not equivalent, only the relation between the respective FLX-dependent and CTRL MMP-9 activity in a given condition should be compared, rather than the exact protein levels.

[2] 40w We have previously shown that inhibiting MMP-9 activity in the CeA results in deficits of reward learning (Knapska et al., 2013). In the next step we tested the reward processing of non-stressed animals subjected to the 8-week long FLX treatment.

[3] 208w To investigate effects of chronic FLX treatment on reward seeking and reward learning of WT mice, we subjected them to a fully automated cognitive assessment in the IntelliCage system (Fig. 2, Knapska et al., 2013;Puścian et al., 2014). IntelliCage simultaneously records multiple parameters of mouse behavior, which in this context allows for distinguishing between nosepokes, representing seeking behavior and tube lickings, representing consumption. In our experimental protocol, mice first learned that water was accessible by nosepoking in each of the two sides of any conditioning unit (corner) of the IntelliCage system. Then access was limited to one conditioning unit per animal (max. 3 mice per corner). In the subsequent reward learning phase, mice had access to water in one of the holes and to reward (10% sucrose solution) in the other. Performance, defined as a percentage of correct choices (nosepokes to the bottle containing sucrose) made immediately after entering the access corner, and reward consumption, defined as the number of tube lickings, were calculated separately. Notably, the applied measure of performance does not directly reflect preference for a given side of the corner, but rather the ability to perform a correct response instantly when presented with a choice, which is most likely memory based (Knapska et al., 2013).

[4] 49w Consistent with our previous results, vehicle-treated WT mice began to acquire a preference for the side of the corner with reward from the first day of training, reaching performance levels close to 90% by the second day (Fig. 3A). Chronic FLX treatment severely impaired reward learning in WT mice.

[5] 73w Their performance, however improving over time, did not significantly exceed chance level, defined as 50% probability of choosing the correct side of the corner upon entering the conditioning unit, until the fifth day of the training (Fig. 3A). Moreover, to assess impact of FLX on motivation, we measured an increase in the number of nosepokes (reward seeking) within 24h of the reward introduction. FLXtreated WT mice displayed significantly reduced reward seeking (Fig 3B).

[6] 188w It is noteworthy that reward learning and reward seeking deficits in FLX-treated WT mice were neither accompanied by decreased reward consumption, measured as a number of licks nor changed locomotor activity (Supporting Fig. 1A, B), suggesting that observed effects cannot be attributed to reduced thirst, impaired taste discrimination or differences in general activity. To make sure that WT mice treated with FLX did not differ in reward consumption from vehicle-treated controls, we additionally pooled data for all the training days, but still found no significant differences in sweetened water consumption (FLX-treated made 24408.9+/-6785.3 (SEM) licks, vehicle-treated -16460.9+/-5940.4 (SEM) licks, Mann-Whitney U-test, animals: FLX-treated WT n=7, vehicle-treated WT n=13). Since tube licking behavior in the IntelliCage system shows high variability, we also calculated a normalized measure that reflects an increase in the number of licks during the first 12h after reward introduction. Using this measure, we also found no differences between FLX-treated and control animals (FLX-treated -2.67+/-0.62 (SEM), vehicle-treated -1.90+/-0.43 (SEM), t-test for independent samples, animals: FLX-treated WT n=7, vehicle-treated WT n=13). Thus, we conclude that chronic FLX intake impairs reward seeking and learning specifically, rather than reward consumption.

[7] 160w Additionally, we tested the impact of FLX on reward learning and reward seeking in animals lacking MMP-9 protein (MMP-9 KO). We hypothesized, that if FLX impairs reward learning and reward seeking by altering MMP-9 activity, animals lacking this protein should be protected from the negative effects of the drug. MMP-9 KO animals learn the task slower due to the global deficit of MMP-9 in their brains, but they usually reach a level exceeding chance on the third day of training and stably perform approximately 70% of correct responses until the end of training (Fig 3C). In line with our prediction FLX had no influence on the reward learning in MMP-9 KO animals. Neither did it impact their reward seeking (Fig. 3D). Similarly to WT animals, FLX-treated MMP-9 KO mice did not display abnormal reward consumption or activity (Supporting Fig. 1C, D). Taken together, these findings further point to the causal link between the FLX and MMP-9 in reward motivated learning.

[8] 160w All animals subjected to cognitive assessment were sacrificed immediately after the last day of training and their brains collected for the analysis of dendritic spines morphology in the CeA. With this approach we aimed to establish a functional measure of neuronal plasticity reflecting the level of performance in the reward-motivated task. The results revealed that descriptive parameters, such as area, circumference, length, width, and density of dendritic spines in the CeA after training were no different in FLX-treated than vehicle-treated mice (Supporting Fig. 2C-G). In contrast, in FLX-treated animals we observed a shift in morphology of spine population towards a more immature state (Supporting Fig. 2A, B). On the population level we saw, a higher percentage of more elongated, filopodial dendritic spines and a lower percentage of mature, mushroom-like spines. As maturity of dendritic spines is often thought to be a measure of well-established synaptic connectivity, the shift towards immature spines may explain the inability of FLX-treated animals to learn.

[9] 170w Non-stressed animals were subjected to an 8-week long FLX or vehicle treatment and then their brains were collected for testing. To assess whether the FLX treatment leads to (previously reported) increase in plasticity of parvalbumin (PV)-and calbindin (CALB)-positive neurons related to rescue of deficits of aversive learning (Karpova et al., 2011;Lee et al., 2014), we performed a doubleimmunostaining for parvalbumin/calbindin and peri-neuronal nets (PNNs, biotinylated lectin from Wisteria floribunda, Fig 4A-C) in central (CeA) and basolateral amygdala (BLA). Consistently with previous reports, we found that chronic FLX treatment significantly decreased the percentage of PNN surrounding PV-positive neurons in the BLA (Fig 4D). A similar trend was observed for the CALBpositive neurons (Fig 4D), suggesting changes in PNN-dependent plasticity of cells in the BLA of the FLX-treated animals. Since PV-and CALB-positive neurons surrounded by the PNNs in CeA are very scarce (see Table 1), the respective analysis of this brain structure was not possible. Thus, arguably the potential impact of this type of plasticity on function of the CeA is limited.

[1] 166w Currently, affective disorders such as depression, anxiety and fear disorders, including phobias and posttraumatic stress disorder (PTSD), obsessive-compulsive disorder and bulimia nervosa are usually treated pharmacologically, often using serotonin selective reuptake inhibitors (SSRI) such as fluoxetine (FLX). Additionally, millions of healthy subjects are being prescribed FLX and often take the drug for months or even years (Kapczinski et al., 2003;Pomerantz et al., 2004;Reed et al., 2004;Moynihan and Cassels, 2005;Jureidini and Tonkin, 2006;Evans and Sullivan, 2014;Johnson et al., 2017;Mars et al., 2017). Our results suggest that long-lasting, daily FLX treatment, mimicking the one applied in patients, leads to severe impairment of reward learning and anhedonia. Our conclusions are in line with a study, which showed that chronically administered FLX impairs reward-motivated operant learning (Frick et al., 2015). Such side effects might pose a serious threat not only to healthy subjects, but most importantly individuals prone to depression, since FLX may negatively influence motivation and elicit anhedonia-related symptoms (Hoehn-Saric et al., 1990;Bertschy and Baumann, 1995;Raskin et al., 2012).

[2] 252w The traditional conceptualization of anhedonia as 'loss of pleasure' has been recently questioned and a refined definition has been proposed (Treadway and Zald, 2011). The new definition attends more closely to the distinction between deficits in reward consumption, also referred to as consummatory anhedonia and diminished reward seeking, reflecting decreased motivation to pursue rewards. The design of our experiments allowed simultaneous measurement of both motivational components , by independent evaluation of nosepoking behavior ('reward seeking') and tube lickings ('reward consumption'), in each experimental subject. The results show FLX-caused impairment of reward seeking rather than changes in consummatory behaviors. In the first 12 h, when neither FLXtreated nor vehicle-treated mice efficiently performed reward-based task, the nosepoking rate of vehicle-treated mice was higher than that of FLX-treated animals. This result suggests that motivation of non-stressed FLX-treated mice to seek the reward is diminished, even though they find sweetened water rewarding, as reflected by the consumption rates. In fact, even though none of the analyzed variables reflecting licking behavior significantly differed between the FLX-treated and vehicle-treated mice, consumption in the former group showed tendency towards being higher, which makes the result on diminished reward seeking even more striking. The motivation deficit probably impairs learning of the reward location. Animals treated with FLX were unable to learn the position of the reward and made more incorrect responses when entering the corner. For the majority of the training, their choices were at the chance level, thus they had to perform more nosepokes to eventually find the reward.

[3] 52w Moreover, the usage of fully automated behavioral testing enabled prolonged assessment of cognition and motivation in the enriched homecage environment in group-housed animals, thus reducing the experimental and social stress in the subjects. Such approach meets the Refinement criterion of the 3Rs principle, by significantly improving the welfare of the laboratory mice.

[4] 297w Further, ours results are a scarce example of research illustrating effects of long-lasting FLX treatment on female subjects. Both depressed and non-depressed women are at least twice as likely to take antidepressants as men (Pratt et al., 2017;Thunander Sundbom et al., 2017). As the gross majority of studies investigating FLX's influence on reward processing were done in males, it might be feasible that sex differences contributed to the contradicting findings regarding the drug's impact on motivation. However, since Frick et al. described similar behavioral changes in male rats, while Marwari and Dawe contradicting ones in females, we argue this to be unlikely (Frick et al., 2015;Marwari and Dawe, 2018). Nevertheless, potential sex differences in the processing of MMP-9 (Griffiths et al., 2019) together with evidence that females may metabolize FLX faster than males, hence be more susceptive to alterations in dosage and to withdrawal (Hodes et al., 2010), call for more basic research on the subject. Specifically, since interplay of estrogen and its receptors in various tissues with MMP-9 (Hu et al., 2006;Lewandowski et al., 2006;Lobo, 2008;Foresta et al., 2010;Chen and Khalil, 2017;Ahmad et al., 2018;Wolak and Hrabia, 2019;Zhao et al., 2019;Pan et al., 2020), as well as PNNs (Lin et al., 2018;Batista and Hensch, 2019;Uriarte et al., 2020) was previously reported, studying the potential role of the estrous cycle in molecular mechanisms related to longitudinal FLX intake seems especially interesting avenue of research. However, in our findings longitudinal design of the testing protocols, including 5-day-long assessment of reward learning, excludes the possibility that observed effects are attributable to a specific hormonal phase, such studies might put the presented results in a new context. Thus, more studies in females are needed to address the issue of potential sex differences in prevalence of side effects of the drug.

[5] 339w It is noteworthy, that the reported deterioration of reward processing at the behavioral level has been observed for 5 days following the 8-week-long FLX treatment. It is improbable that observed effects were a result of the cessation of chronic FLX use rather than the treatment itself. Firstly, the impaired reward learning occurred from the very first day of training, before the next dose of the drug would have been administered. Further, the study of Klomp et al. showed that FLX treatment as short as 21 days results in elevated serotonin levels persisting for one week after the treatment cessation (Klomp et al., 2014). Also, the changes in MMP-9 activity level and perineuronal nets were observed in tissue collected less than 24h after the last dose of the drug, which excludes interference of FLX treatment cessation. Additionally, presented research is one of the few rodent studies reporting the effects of FLX intake longer than 21 days. As the minimal recommended treatment time ranges between 4 and 6 weeks the more clinically relevant laboratory data is lacking (Pérez et al., 2001;Brunswick et al., 2002). Indeed, it is conceivable that the impairment of reward processing and dysregulation of underlying molecular mechanisms would be less severe or even absent if the treatment lasted for a shorter period of time. This would be in agreement with recent findings illustrating that FLX hinders neurogenesis after 6 and 9, but not after 3 weeks of intake (Ohira et al., 2019). Still, the length of the applied treatment may not be an exclusive explanation of the discrepancies in the literature, as deteriorating effects of FLX have been reported after chronic drug administration significantly shorter than the one used in this research (Takahashi et al., 2005;Dringenberg et al., 2014;Frick et al., 2015;Eisenreich et al., 2017;Flores-Ramirez et al., 2019;Sharp et al., 2019). In addition, as our experiments were performed in an inbred strain from a local colony it may be that the reported effects occur only against this specific genetic background. Thus, further confirmation in outbred mice is needed.

[6] 175w Over the years a number of studies have shown that reward seeking depends on the activity of the central amygdala (CeA, Mahler and Berridge, 2009;DiFeliceantonio and Berridge, 2012;Knapska et al., 2013;Robinson et al., 2014). In line with these results, we observed that chronic FLX treatment decreased activity of matrix metalloproteinase 9 (MMP-9), an enzyme intimately involved in synaptic plasticity (Dziembowska and Wlodarczyk, 2012;Stawarski et al., 2014) in the CeA. Decreased activity of MMP-9 in the CeA was accompanied by severe impairment of reward learning and motivational changes, which aligns with the previous studies, which showed that MMP-9 activity in the CeA is crucial for reward learning (Knapska et al., 2013). Our study confirmed that animals chronically treated with FLX had higher proportion of immature spines (Lai and Ip, 2013) in the CeA after reward-based training. These results suggest that either structural plasticity of dendritic spines is changed in the CeA by chronic FLX treatment or FLX modifies how the reward learning influences spine morphology in the CeA and thus alters formation of stable synaptic transmission.

[7] 249w Previous studies have shown that chronic FLX treatment affects neuroplasticity by recruitment of new neurons (Santarelli et al., 2003;Surget et al., 2008) and enhancement of synaptic plasticity (Karpova et al., 2011;Guirado et al., 2014). In particular, the drug has also been proven to facilitate fear extinction (Karpova et al., 2011) and prevent the return of conditioned fear (Deschaux et al., 2011) in mice. These effects were caused by loosening of the perineuronal nets (PNNs) surrounding parvalbumin-expressing interneurons in basolateral amygdala (Karpova et al., 2011;Popova et al., 2014). We investigated PNNs in the amygdala to test whether similar effects can be observed in FLXtreated mice. Loosening PNNs opens the window of increased neuronal plasticity, which allows for remodeling of memory traces (Gogolla et al., 2009). In the current study we found changes in basolateral amygdala (BLA) PNNs surrounding PV-neurons, similar to what has been reported previously (Karpova et al., 2011). At the same time, a number of studies suggest that MMP-9 increase is related to the degradation of the PNNs, in both animals and humans (Gray et al., 2008;Murase et al., 2017;Wen et al., 2018b;Alaiyed and Conant, 2019;Alaiyed et al., 2019;Fawcett et al., 2019). Also recent studies showed that the serotonin-norepinephrine reuptake inhibitors, such as venlafaxine, increase MMP-9 levels while decreasing the number of PNNs in hippocampus (Tamási et al., 2014;Alaiyed et al., 2019). Thus, we argue that the nature of the PNN-MMP-9 relationship may be specific to a particular neuronal circuit and differ between amygdala and hippocampus or cortex.

[8] 96w However, PV and CALB neurons surrounded by PNNs are very rare in CeA. This suggests that the mechanism underlying impairment of reward learning is different from the one previously observed for improvement of fear extinction after chronic FLX treatment. Our data suggest that observed behavioral effects depend, at least partly, on MMP-9-mediated changes in neuronal plasticity within the CeA. Nevertheless, to explicitly demonstrate that lack of MMP-9 activity in the CeA is sufficient to prevent the negative FLX effects on motivation and learning, additional studies utilizing conditional MMP-9 KO animals combined with viral vectors are needed.

[9] 303w In conclusion, chronic FLX treatment differentially affects various forms of neuronal plasticity and thus elicits contradicting effects on brain and behavior. Presented findings are of an immediate clinical relevance since reported side effects of FLX pose a potential threat to patients. Understanding the neuronal underpinnings of detrimental effects of chronic FLX treatment is crucial for finding strategies to contradict the drug's side effects. Further experiments are needed to determine the cellular mechanisms by which FLX reduces the MMP-9 activity level in the CeA. So far, it has been shown that FLX may directly affect MMP-9 activity (Lee et al., 2012(Lee et al., , 2014)), an effect specifically dependent on the serotonin 5-HT7 receptor (Bijata et al., 2017), sensitivity of which is decreased by chronic FLX treatment (Mnie-Filali et al., 2011). Thus, it is expected that prolonged FLX intake suppresses enzymatic activity of MMP-9 and related neuronal plasticity. The signaling pathway involving the 5-HT7 receptors is a potential link between chronic FLX administration and MMP-9 activity in CeA. Mice were fed with FLX or water (control group) for 8 weeks. Then, their cognitive abilities and motivation were assessed in the IntelliCage system. The protocol consisted of the following phases: (A) the simple adaptation with free access to all water bottles followed by the nosepoke adaptation, when access to water was obtained by performing a nosepoke; (B) preparatory place learning, with access to water limited to one of the four conditioning units; and (C) reward discrimination learning, when one of the bottles contained highly motivating reward (10% sucrose) and the other water. To avoid social influence on learning in two last phases of the experiment the access to bottles in particular conditioning units was randomly assigned to different subjects (max. 3 mice per one conditioning unit). ). For more immunostaining data see Supporting Table 1.

[1] 43w Animals were treated in accordance with the ethical standards of the European Union (directive no. 2010/63/UE), respective Polish regulations and with respect to the principals of the 3Rs. All experimental procedures were preapproved by the Local Ethics Committee no. 1 in Warsaw, Poland.

[2] 225w MMP-9 knockout mice in C57BL/6 background and their respective controls were bred in the Animal House of Nencki Institute of Experimental Biology, Polish Academy of Sciences. Originally MMP-9 homozygous KO mice on a C57BL/6 background were obtained from Dr. Z. Werb (University of California, San Francisco). These mice were bred with C57BL/6NtacF WT mice (purchased from Jax Laboratories) for at least two generations and then maintained and bred continuously with each other as heterozygotes for >10 generations. After that, the colony gene pool was regularly refreshed with newly purchased C57BL/6 mice. Their homozygous progeny (MMP-9 KO and WT mice) used in this study were always littermates derived from several breeding pairs. In the studies concerning the influence of FLX on WT mice we used adult C57BL6/cmdb mice purchased from Medical University in Bialystok (originally bred from pairs bought from Jacksons Laboratories). In this case, animals were housed in the Animal House of Nencki Institute of Experimental Biology for at least 2 weeks before the onset of the experimental procedures. The breeding procedures were conducted as recommended by the International Committee on Standardized Genetic Nomenclature for Mice and in the guidelines provided by the Jackson Laboratory (Benavides et al., 2019, https://www.jax.org, under "strain-background-and-genetic-drift"), both stating that in a new colony strict inbreeding for at least 10 generations is necessary to consider the strain an inbreed one.

[3] 210w All experimental effects of chronic and acute FLX treatment were assessed in non-stressed female mice, without any control or manipulation of their estrous cycle. It is noteworthy, that in the presented studies longitudinal design of the testing protocols, including 5-day-long assessment of reward learning, excludes the possibility that observed effects are attributable to any specific hormonal phase. Performed studied are a scarce example of research illustrating effects of longlasting FLX treatment on female subjects. Both depressed and non-depressed women are at least twice as likely to take antidepressants as men (Pratt et al., 2017;Thunander Sundbom et al., 2017). As further discussed, the gross majority of research investigating FLX's influence on reward processing was done in males and thus the impact of sex in this context remains unclear. Moreover, evidence that females may metabolize FLX faster than males, hence be more susceptive to alterations in dosage and to withdrawal call for more basic research on the subject. Further, to comply with requirements of fully automated behavioral assessment demanding large numbers of animals in tested cohorts, we utilized 3-6 months old MMP-9 KO and WT females to compose desirable group sizes. Each cohort of subjects was balanced for age. The C57BL6/cmdb mice were 3 months old at the beginning of the experiments.

[4] 152w Over a period of 8 weeks, non-stressed mice were subjected to chronic FLX treatment. FLX was administered orally to mimic the route of administration used in human patients. To reduce experimental stress, a voluntary drug consumption protocol was used. Group-housed animals were put in small housing cages for brief (max. 5 minute) separation sessions once a day, when lights were off. For two weeks subjects were then habituated to voluntarily consume 320 µl of liquid caramelized milk. During this period separation sessions were gradually shortened and lasted from 4h on day 1 to 5 minutes on day 14. After the habituation period FLX or a vehicle for delivering the drug was mixed with condensed milk to a total volume of 320µl (on average 30-40µl of FLX dissolved in MiliQ water + 270-280µl liquid caramelized milk, depending on animal's weight). Drug/vehicle dose was 10mg/kg of body weight/day. FLX concentration in water was 7mg/ml.

[5] 81w In experiments with acute FLX treatment, non-stressed animals were taken from the home cage and injected intraperitoneally with the drug or its delivery vehicle, without prior habituation. After the injection, they were returned to their home cages for 30 minutes and then briefly anesthetized with isoflurane and sacrificed by spinal cord dislocation. Drug/vehicle dose was 20mg/kg of body weight/day. FLX concentration in MiliQ water was 2.3-2.4mg/ml with injection volumes of 0.2-0.3 ml per mouse, respectively, depending on an animal's body weight.

[6] 182w Behavioral tests were performed in the IntelliCage (TSE), a fully automated, computer controlled system, which can be used for long-term monitoring of behavior of group-housed mice (Knapska et al., 2013;Puścian et al., 2014). However, this form of behavioral assessment requires obtaining large, well-socialized groups of littermate animals of a given genotype, which poses significant breeding challenges. Thus, to perform the experiments we were compelled to test cohorts of non-identical sizes. Each visit to the operant chamber, as well as each nosepoke and the amount of liquid consumed (number and duration of licks), was recorded for each individual animal. The cage control unit permitted access to particular bottles according to preprogrammed schedules, depending on random assignment of the mice to different test groups within the same cage. Implementation of this procedure together with fully automatized data collection and processing provides high standards of blinding and randomization. The system ran continuously for 11 days. During that time, the behavioral activity of the mice was remotely monitored via internet. Except for the technical breaks and cage cleaning (once a week), the mice were not disturbed.

[7] 464w After 50 days of FLX/vehicle treatment, non-stressed mice were introduced into the IntelliCage environment. FLX and vehicle-treated animals were tested together within the same IntelliCage. Drug administration was continued throughout the adaptation period (6 days). Cohorts of 7-9 mice were subjected to an 11-day IntelliCage protocol. After 6 days of adaptation cognitive function was assessed for the remaining 5 days by introducing a highly motivating reward (10% sucrose solution) in one of the operant chambers and observing the response of the animals. The adaptation phase was comprised of simple adaptation, nosepoke adaptation, and preparatory place learning stages. During all of these stages, 8 bottles containing water were available for consumption. During the simple adaptation phase, doors in all operant chambers (corners) were open and access to water was unrestricted (2 days). In nosepoke adaptation phase all doors were closed by default and opened when an animal put its snout (nosepoke response) into one of the two holes on the operant chamber's walls. When an animal removed its nose from the opening, the door closed automatically. During preparatory place learning access to the drinking bottles was restricted to a single corner (different for different mice). The learning access chamber was assigned randomly, with no more than 3 mice drinking from the same operant chamber. This procedure minimizes social modulation of learning (Kiryk et al., 2011). Reward discrimination learning protocol was based on voluntary choice between two bottles: one containing 10% sucrose solution (reward), the other filled with plain water (neutral stimulus). The percentage of correct choices (nosepokes to the bottle containing sucrose) made immediately after entering the access corner (learning) as well as the increase in nosepoking behavior were measured as indicators of reward-related cognitive functioning. The measures of performance in discrimination task and reward seeking were identical as used in (Knapska et al., 2013). Briefly, task performance was compared to chance level, defined as 50% probability of choosing the correct side of the corner upon entering the conditioning unit. Moreover, to assess impact of FLX on motivation, we measured an increase in the number of nosepokes (reward seeking) within 24h of the reward introduction. We also measured animals' activity and reward consumption (number of licks). The final lick count was presented as a normal number rather than normalized to water consumption, as most animals stop consuming water all together, when given access to 10% sucrose solution. Data were excluded from the analysis when an animal had not learned to obtain water in preparatory place learning and was unable to correctly perform in the subsequent reward discrimination task or in the rare cases when an animal's locomotor activity was extremely low. IntelliCage Controller software alerted us to such cases. We performed 3 replications of the described testing procedure for jointly tested experimental and control groups.

[8] 289w After chronic FLX treatment (8 weeks) non-stressed mice were deeply anesthetized with sodium pentobarbital (100mg/kg) and perfused transcardially with PBS (4°C, 50ml per animal) and, subsequently, with 4% PFA (4°C, 70ml per animal). Brains were dissected, postfixed in 4% PFA (o/n, 4°C) and cryoprotected in 30% sucrose solution (4°C) for at least 3 days. Coronal sections 40μm thick containing pre-and infralimbic cortex, central and basolateral amygdala and hippocampus (coordinates in accordance to the brain atlas, (Paxinos and Franklin, 2001), were cut with the use of a cryostat (-20°C) and kept in PBS (4°C) until further handling. For the immunostaining, slices were washed in PBS (3x10', at room temperature, RT) and blocked with BSA (10% in 0.3% PBST) to avoid nonspecific binding. Next, they were incubated with primary antibodies (mouse anti-parvalbumin, Abcam, 1:10000, RRID:AB_1142183; mouse anti-calbindin, Abcam, 1:5000, RRID:AB_2811302) in 0.3% PBST (24h, 4°C), washed with PBS (3x10', RT) and incubated with a secondary antibody (Alexa 546 goat anti-mouse 1:1000, Invitrogen, 1h, RT) in 0.3% PBST. To simultaneously detect perineuronal nets and parvalbumin or calbindin, we subsequently incubated slices with biotinylated lectin from Wisteria floribunda (1:200; Sigma-Aldrich, 24h, 4°C), washed them with PBS (3x10', RT) and incubated with FITC-conjugated streptavidin (1:1000, Invitrogen, 1h, RT). After the last round of washing with PBS (3x10', RT) slices were mounted (Fluoromount-G, Southern Biotech) on polysine slides. To control for non-specific binding and imaging artifacts 'no primary antibody' and 'no secondary antibody' immunostainings were carried out. Imaging was performed with the use of confocal microscope (Leica TCS SP5). Two independent experimenters (blind to the experimental conditions) used ImageJ software (NIH) to count immune-stained cells. On average, 5 slices per animal were used for bilateral quantification. Average values from two measurements were considered final results.

[9] 218w One day after conclusion of reward-motivated discrimination learning, animals were deeply anesthetized with sodium pentobarbital (100mg/kg) and transcardially perfused with PBS (RT, 50ml per animal) and 1.5% PFA (RT, 70ml per animal). Further procedures were performed as previously described by (Michaluk et al., 2011;Pijet et al., 2019). Briefly, brains were collected and cut into 130 um coronal sections. Slices were then processed for the Dil (1,1-dioctadecyl-3,3,3,3tetramethylindocarbocyanine perchlorate) staining. Dye was delivered using Gene Gun (Bio-Rad) loaded with 1.6 um Dil-coated tungsten particles (Bio-Rad, Hercules, CA, USA). Next, slices were fixed overnight in 0.4% PFA, mounted onto the slides and imaged using the LSM780 confocal microscope (40x/1.4 Oil DIC, Zeiss, at 594 nm wavelength). Dendritic spines were marked and measured semiautomatically using custom software (SpineMagick). Analysis of dendritic spine quantity and morphology. Basic morphological parameters, including spine area, length, width and circumference, as well as spine density were estimated using SpineMagick software (Ruszczycki et al., 2012). Spine length was calculated as the length of the path from spine top to the dendrite along the virtual skeleton of a spine. Analysis of dendritic spine shapes was performed as previously described (Jasińska et al., 2016). Data analysis was performed using custom scripts written by Szymon Łęski in Python, using NumPy and SciPy (Oliphant, 2007;Perez and Granger, 2007) and Matplotlib (Hunter, 2007).

[10] 214w Animals were housed under the conventional facility conditions in the Animal House of Nencki Institute of Experimental Biology for at least 2 weeks before the onset of the experimental procedures. Mice were group housed (7 to 9 same sex conspecifics) in large, rectangular cages (length: 5850 mm, width: 375 mm, height: 200 mm, area: 1802 cm 2 ) with wood shreds bedding. Additionally cages were equipped with 2-3 igloo-shaped shelters and shredding material for nestbuilding. Mice were kept under a 12 h/12 h light/dark cycle with water and food provided ad libitum (filtered water and standard laboratory rodent diet, type 5001). The temperature was maintained at 23-24°C with humidity levels between 35% and 70%. The breeding procedures were conducted as recommended by the International Committee on Standardized Genetic Nomenclature for Mice and in the guidelines provided by the Jackson Laboratory (Benavides et al., 2019, https://www.jax.org, under "strainbackground-and-genetic-drift"), both stating that in a new colony strict inbreeding for at least 10 generations is necessary to consider the strain an inbreed one. All animals were kept under strict veterinary care with daily health inspections. As described in the previous section, at the end of the experiments all mice were subjected to transcardial perfusions performed under deep anesthesia (sodium pentobarbital,100mg/kg) and their brains collected for further processing.

[11] 33w The data and statistical analysis comply with the recommendations of the British Journal of Pharmacology on experimental design and analysis in pharmacology (Curtis et al., 2015(Curtis et al., , 2018;;George et al., 2017).

[12] 328w In procedures assessing cognitive performance, motivation, reward consumption and locomotor activity in animals chronically treated with FLX or vehicle the final group sizes were as follows: WT FLX n=13, WT CTRL n=7, MMP-9 KO FLX n=16 and MMP-9 KO CTRL n=9. To comply with requirements of fully automated behavioral assessment demanding large number of animals in tested cohorts, we utilized 3-6 months old MMP-9 KO/WT females and C57BL6/cmdb mice to compose desirable group sizes. Each cohort of subjects was balanced for age. Notably, we planned for equalization of the group sizes within each experimental condition, within the limits set by the number of offspring of a given genotype/sex/age available in the mouse colony, and thus at the beginning of the procedures the planned group sizes were as follows: MMP-9 KO CTRL n=14, MMP-9 KO FLX n=16 and WT CTRL n=11, WT FLX n=13, within 3 independent replications of the behavioral experiments performed on the cohorts of animals consisting of subjects ascribed to both FLX-and vehicle-treated conditions. However, due to the experimental losses and the necessity of data exclusion in accordance with the predetermined criteria we lost results from 5 WT and 5 MMP-9 KO animals. In case of 2 WT and 4 MMP-9 KO animals randomly assigned to the CTRL group they never obtained a level of 100% consumption of the vehicle in the voluntary drug consumption protocol and thus could not further qualify for behavioral testing. Additionally, data were excluded from the analysis when an animal had not learned to obtain water in preparatory place learning and thus was unable to correctly perform in the subsequent reward discrimination task (1 MMP-9 KO CTRL mouse and 1 WT CTRL mouse) and in a rare case when animal's locomotor activity was extremely low and thus could lead to bias in cognitive assessment (1 WT CTRL mouse). Further, all WT animals from the FLX-and vehicle-treated groups were used for the assessment of changes in dendritic spine morphology in the CeA.

[13] 234w Due to the breadth of applied experimental techniques chronic FLX treatment had to be performed in multiple groups of animals. Notably, to test the influence of the chronic FLX treatment, unconfounded by the behavioral training, on MMP-9 activity and perineuronal nets, additional cohorts were used. In the experiments assessing MMP-9 activity in the central (CeA) and basolateral (BLA) amygdala in animals treated with FLX the sample sizes were as follows: group 1: CeA chronic treatment -CTRL n=8, FLX n=8, group 2: CeA acute treatment -CTRL n=8, FLX n=12, group 3: BLA chronic treatment -CTRL n=6, FLX n=4, group 4: BLA acute treatment CTRL n=8, FLX n=12; please note that due to the insufficient group size of group 3 the reported differences in MMP-9 activity for this condition were not subjected to statistical testing, as described in the Statistical Analysis section. Further, to assess the influence of chronic FLX treatment detangled from behavioral training on changes in the perineuronal nets (PNN), parvalbumin-expressing (PV) and calbindin-expressing (CALB) cells in the BLA and CeA we performed immunostaining procedures on the tissue obtained from FLX treated, but otherwise naïve animals. The sample sizes were as follows: PV + PNN double staining: FLX n=7, CTRL n=7 and CALB + PNN double staining FLX n=4, CTRL n=6. Due to the small sample size <5, the latter group was not subjected to statistical testing, as described in the Statistical Analysis section.

[14] 228w In the experiments on the effects of chronic, long-lasting FLX treatment we implemented a number of randomization procedures, with a goal of minimizing the potential bias in interpretation of the results. Specifically, identically looking, group-housed WT and MMP-9 KO animals were randomly assigned 15-digit electronic IDs and subcutaneously tagged with electronic transponders, encoding the numbers. Subsequently, the ID numbers were randomly assigned to the groups subjected to either FLX or vehicle treatment. In the course of the below described voluntary drug consumption protocol all animals were put in the small housing cages for a brief daily separation sessions (max. 5 minutes), which took place during the dark phase of the light/dark cycle. Every day mice were randomly placed in the small housing cages, without any prior assignment. Doses of drugs for both FLX-and vehicletreated groups were prepared daily in the remote wet laboratory and then brought in ready in the microcups for the voluntary drug consumption session. On each day the experimenter checked the electronic ID number of each separated animal from the distance of 10-15 cm with the use of a remote transponder reader and then administered either FLX or vehicle, in accordance with the document specifying random group assignment. Additionally, animals from both groups were housed together throughout the whole preparatory period and the experiment, to avoid unequal treatment by the animal facility stuff or experimenters.

[15] 170w To minimize bias in the voluntary drug consumption protocol microcups containing liquid caramelized milk, mixed with either FLX or vehicle, and their content looked identically. Further, to assess the effects of chronic FLX treatment on cognitive performance, motivation, reward consumption and locomotor activity we used the IntelliCage, a fully automated behavioral testing system with computerized data collection. Further, as previously described animals were randomly assigned a 15-digit electronic ID number, encoded in the subcutaneously injected transponder device that animal carried throughout the entirety of the experimental procedures. The data from behavioral experiments obtained with the use of the IntelliCage system was further processed based on this enumeration and thus blindly analyzed by the experimenters. The results for any given group were pulled together only at the last step of the analysis for the purposes of visualization and statistical testing. The method of the blind assignment of the 15-digit electronic ID number was also implemented for enumeration of the samples used in the immunostaining, gel zymography and dendritic spine imaging procedures.

[16] 15w The data presented in this article was neither normalized nor transformed in any other manner.

[17] 189w Statistical analyses were performed with Statistica 8.0 (StatSoft) and GraphPad Prism6 software. Normality of distributions was assessed with the Shapiro-Wilk test. When appropriate, group comparisons were analyzed using Student's t-test for independent samples. In case of the data presented in Fig. 3A and Fig. 3C, to account for the repeated measures comparison instead of looking at the P values we took under consideration the t values, calculated separately for each day of training and subsequently compared them with the recommendations from the Dunnett's tables (Dunnett, 1955(Dunnett, , 1964(Dunnett, , 1985)), prescribed for 8 degrees of freedom. The "*" symbols in those figures indicate meeting such defined standard of statistical significance. Datasets that did not meet the criteria for parametric analyses were analyzed using the Mann-Whitney U-Test, with an exception of the data presented in Fig. 4H and in Supp. Table 1, where sample size <5 did not permit for performing the analysis. Chi-squared tests were performed to compare distributions of dendritic spine shapes. The criterion for statistical significance was a probability level of p<0.05, which does not vary throughout the manuscript and is marked by * or # .

[1] 28w Animal studies are reported in compliance with the ARRIVE guidelines (Kilkenny et al., 2010;McGrath and Lilley, 2015) and with the recommendations made by the British Journal of Pharmacology.

[2] 79w In the studies of impact of the chronic FLX treatment on cognitive function and motivation, we used adult wild type C57BL6/cmdb mice and MMP-9 homozygous knock-out mice on a C57BL/6 background. The C57BL6 strain was chosen based on its utility for these and potential further pharmacological studies, as these mice are the most commonly used, best characterized model for such purposes and often experiments with their use are considered a gold standard in biomedical research (Åhlgren and Voikar, 2019).

[3] 121w At least 1 week before the beginning of habituation to drug administration, mice were housed together in groups assigned to the subsequent experimental procedures. To individually identify animals during the drug administration period and enable fully automated behavioral testing in the IntelliCage system, all mice were subcutaneously injected with glass-covered microtransponders (11.5 mm length, 2.2 mm diameter; DATAMARS) under isoflurane anesthesia. Microtransponders emit a unique animal identification code when activated by the magnetic field of the IntelliCage antennas or a portable chip-reader. After injection of a transponder, subjects were moved from the housing facilities to the experimental rooms and adapted to the shifted light/dark cycle (the dark phase shifted from 20:00 -8:00 to 13:00 -01:00 or 12:00 -24:00 accordingly to summer/wintertime).

[4] 103w The behavioral data from the IntelliCage system were analyzed either in the Analyzer (New Behavior) or using custom software written in Python® programming language (Python 2.7) with NumPy, SciPy and PyMICE libraries (RRID:nlx_158570, Kowalski et al., 2016;Dzik et al., 2018). First, raw data files generated by the IntelliCage system were loaded, merged, and tested to identify data segments corrupted due to hardware malfunctions. Corrupted segments (if present) were excluded from further analysis. The relevant parameters (numbers and cumulative durations of visits to specific corners, numbers of nosepokes and tube-lickings) were scored in 12-hour bins and saved in spreadsheets for statistical analysis and plotting.

[5] 224w CeA, BLA and HIPP were dissected from freshly prepared brain sections of naïve, untrained and non-stressed animals subjected to chronic FLX treatment (8 weeks) or vehicle-treated controls. Brains were collected and stored on ice during transportation. Next, they were cut with vibratome (Leica VT1000 S) into 200 μm-thick coronal slices containing structures of interest. Subsequently, slices were put on microscope slides and frozen at -80°C. The appropriate brain structures were dissected with cooled surgical needles. Tissue was pooled from two mice per experimental condition. Further procedures were performed according to the protocols described in (Kaliszewska et al., 2012;Janusz et al., 2013). Briefly, the equal concentration of the protein was mixed with 2× sample buffer Tris-Glycine SDS (Novex). Samples were subjected to electrophoresis under nondenaturing, nonreducing conditions in SDS-PAGE Tris-glycine 8 % acrylamide gels with 0.5 % gelatin (POCH). The gels were than twice washed for 30 min in 2.5 % Triton X-100 and subsequently incubated for 5 days in the zymography buffer (50 mM Tris, pH 7.5, 10 mM CaCl2, 1 μM ZnCl2, 1 % Triton X-100 and 0.02% sodium azide) at 37 °C. The intensity of white bands on the blue background corresponding to the MMP-9 and matrix metalloproteinase-2 (MMP-2) activity was quantified with ImageJ software. The relative activity level of MMP-9 was normalized to MMP-2 (Szklarczyk et al., 2002;Ganguly et al., 2013).

[6] 49w Key protein targets and ligands in this article are hyperlinked to corresponding entries in http://www.guidetopharmacology.org, the common portal for data from the IUPHAR/BPS Guide to PHARMACOLOGY (Harding et al., 2018), and are permanently archived in the Concise Guide to PHARMACOLOGY 2019/20 (Alexander et al., 2015(Alexander et al., , 2019)).