The pharmacological perturbation of brain zinc impairs BDNF-related signaling and the cognitive performances of young mice

Zinc (Zn2+) is a pleiotropic modulator of the neuronal and brain activity. The disruption of intraneuronal Zn2+ levels triggers neurotoxic processes and affects neuronal functioning. In this study, we investigated how the pharmacological modulation of brain Zn2+ affects synaptic plasticity and cognition in wild-type mice. To manipulate brain Zn2+ levels, we employed the Zn2+ (and copper) chelator 5-chloro-7-iodo-8-hydroxyquinoline (clioquinol, CQ). CQ was administered for two weeks to 2.5-month-old (m.o.) mice, and effects studied on BDNF-related signaling, metalloproteinase activity as well as learning and memory performances. CQ treatment was found to negatively affect short- and long-term memory performances. The CQ-driven perturbation of brain Zn2+ was found to reduce levels of BDNF, synaptic plasticity-related proteins and dendritic spine density in vivo. Our study highlights the importance of choosing “when”, “where”, and “how much” in the modulation of brain Zn2+ levels. Our findings confirm the importance of targeting Zn2+ as a therapeutic approach against neurodegenerative conditions but, at the same time, underscore the potential drawbacks of reducing brain Zn2+ availability upon the early stages of development.

receptor TrkB play a crucial role in the plasticity-related modifications of spines by promoting spine pruning and enlargement. pro-BDNF and its receptor p75NTR exert pro-apoptotic activity and negatively affects the dendritic structure and spine density 14,22 . The regulation of spine morphology, a phenomenon defined as structural plasticity, is one of the molecular underpinnings of cognition and learning and memory in particular 23 .
The metalloproteinase, MMP-9, is highly enriched in the hippocampus, the cerebral cortex, and the cerebellum 24 , brain regions where the enzyme is localized in the extracellular matrix near the dendritic spines of excitatory synapses 25 . MMP-9 contributes to learning, memory and neuronal plasticity by promoting the spine remodelling and immobilization of ionotropic receptors that occur upon long-term potentiation (LTP) stimulation 26,27 . In preclinical models of Alzheimer's disease (AD), the functional overexpression of MMP-9 increases the brain levels of mBDNF and prevents the development of AD-related cognitive deficits 28,29 . MMP-9 is structurally characterized by a 3-histidine zinc-binding catalytic motif (His-Glu-Xaa-Xaa-His) and requires zinc (Zn 2+ ) to exert its proteolytic activity 30 .
Zn 2+ is largely present in the CNS, stored within synaptic vesicles at several glutamatergic nerve terminals, and synaptically released upon neuronal activity 31,32 . Zn 2+ affects neuronal processes as well as BDNF signaling 33 . The metal, through MMP activation, promotes the cleavage of pro-BDNF to mBDNF 9,10 . Zn 2+ also modulates BDNF signaling without affecting the MMP activity. This activity is achieved by direct transactivation of TrkB, the preferred and high-affinity BDNF receptor 34 . In addition, in vivo administration of Zn 2+ increases the cortical levels of the BDNF mRNA 35 , thereby indicating a pro-transcriptional activity of the cation.
In a preclinical transgenic AD model, we have previously shown that long-term dietary supplementation of Zn 2+ promotes the activation of MMP-2 and MMP-9, increases the brain levels of mBDNF, and counteracts the development of AD-related pathology and memory deficits 36 .
In this study, we investigated how the pharmacological modulation of brain Zn 2+ affects BDNF signaling, synaptic plasticity, and cognition in young mice. The rationale behind the experimental design and the choice of employing young animals is twofold. First, recent evidence indicate that deficiency in synaptic Zn 2+ affects brain development by impairing BDNF signaling 10 ; second, we aimed at evaluating effects related to the perturbation of brain Zn 2+ homeostasis in young mice, a setting that, on the contrary of aging-or degeneration driven models 37 , is devoid of signs of Zn 2+ deregulation. To manipulate brain Zn 2+ levels, we employed the Zn 2+ (and copper) chelator 5-chloro-7-iodo-8-hydroxyquinoline (clioquinol, CQ) as previous studies have shown that the intraperitoneal administration of CQ promotes a rapid reduction in the brain pool of chelatable Zn 2+ 38-43 . CQ was administered for two weeks to 2.5-month-old (m.o.) mice and effects studied on BDNF-related signaling, MMP activity as well as learning and memory performances.

Results
CQ administration impairs learning, short-and long-term memory performances in young mice. Male C57Bl/6 mice were treated, for 2 weeks, with daily intraperitoneal (i.p.) injections of CQ (30 mg/ kg). The rationale for choosing this concentration and route of administration was threefold: (1) this CQ concentration has been shown to be devoid of significant side effects in rodents including changes in body weight ( Fig. 1B; 25.8 ± 1.6 g in control vs 24.9 ± 0.91 in CQ treated mice, p = 0.64); (2) the dose matches the one employed in clinical trials; (3) the i.p. route is highly effective in reducing the pool of chelatable Zn 2+ in the brain and the hippocampus in particular 38,44 . In agreement with previous studies 38-43 , we obtained a significant decrease of hippocampal Zn 2+ levels (Supplementary Fig. 1; normalized TSQ fluorescence in control: 1.00 ± 0.08 vs 0.49 ± 0.04 in CQ-treated mice, p < 0.001).
The effects of CQ or vehicle on hippocampus-dependent spatial learning were assessed, at the end of the second week of treatment, with the Morris Water Maze (MWM) test, a widely employed task that evaluates memory performances in rodents 45 (Fig. 1A). Compared to vehicle-treated mice, CQ-treated animals displayed learning impairment. CQ-treated mice showed a statistically-significant increase in the time spent to reach the platform upon the first two training sessions ( Fig. 1C; day 1 p < 0.01, day 2 p < 0.001). These learning deficits were transient and disappeared in the final two days of training ( Fig. 1C; day 3 p = 0.069, day 4 p = 0.16). The analysis of memory performances revealed that, compared to vehicle-treated animals, CQ-treated mice performed significantly worse in Short-(1.5 h) and Long-Term (24 h) Memory (STM and LTM, respectively) skills. The CQ-treated group showed a decreased number of platform crosses ( Fig. 1D; for STM: 4.69 ± 0.69 in control vs 2.15 ± 0.54 in CQ-treated mice, p = 0.008; for LTM: 4.3 ± 0.9 in control vs 1.73 ± 0.38 in CQ-treated mice, p = 0.04) and longer times to reach the platform location (Fig. 1E, escape latency for STM: 8.84 ± 1.5 s in control vs 26.38 ± 5.6 s in CQ-treated mice, p = 0.007; escape latency for LTM: 12.9 ± 5 s in control vs 29.58 ± 5.9 s in CQ-treated mice, p = 0.01). No significant differences, between the two study groups, were observed as far as the time spent in the target or opposite quadrants (Fig. 1F,G, T-target, and T-opposite, respectively). Additional analysis showed no significant differences between the two study groups regarding swimming speed or the total distance travelled upon the LTM probe trial (Fig. 1H,I; p = 0.43 for the swimming speed and p = 0.71 for the distance travelled), thereby indicating that the CQ-mediated effects were not driven by decreased motivational cues or impaired swimming abilities.
CQ reduces BDNF signaling in treated mice. Zn 2+ has been shown to promote pro-BDNF maturation in vitro 9 . Western blot (WB) analysis of brain samples revealed a significant decrease of BDNF levels in CQ-treated mice when compared to vehicle-treated animals ( Fig. 2B; see Supplementary Table 1 for the statistics of the WB data). The reduction is region-specific and occurred in the hippocampus, cortex, and striatum but not in the cerebellum. These regional changes were matched by parallel decreased levels of the preferred BDNF receptor, TrkB (Fig. 2C). To assess whether BDNF and TrkB reductions affect BDNF-related signaling, we measured brain levels of the activated -phosphorylated-form of TrkB (pTrkB) and ERK5 (pERK5), a downstream effector of the BDNF-TrkB signaling cascade 46 . WB analysis showed no significant differences between CQ-and vehicle-treated animals in the pTrkB/TrkB ratio (Fig. 2D) while CQ administration induced a significant decrease in pERK5 levels (Fig. 2F).
The neurotrophin BDNF, by activating mechanisms related to functional and structural plasticity, is a potent modulator of LTM 47 . We, therefore, investigated whether the CQ-driven reduction of BDNF and TrkB translates into changes in PSD95 levels, a marker of the post-synaptic integrity. Matching the CQ-related regional changes in BDNF, we found region-specific decreases of PSD95 in the hippocampus, cortex, striatum, but not in the cerebellum (Fig. 2E).
CQ promotes p75NTR signaling in treated mice. Since pro-BDNF-p75NTR activation can modulate neurotoxic activities, we investigated whether the observed CQ-driven impairments of the BDNF-TrkB signaling axis are associated with changes in the pro-BDNF-p75NTR signaling cascade. WB analysis of these pathways showed that, compared to vehicle, CQ treatment promotes a region-specific increase of p75NTR levels along with high levels of phosphorylated ERK 1,2 (pERK 1,2 ; Fig. 2H,I). Surprisingly, the CQ-treatment was also found to decrease pro-BDNF levels in the same brain regions (Fig. 2G). The CQ-driven activation of the p75NTR signaling cascade did not produce overt signs of neurotoxicity in vitro and in vivo ( Supplementary Fig. 2).

CQ treatment does not affect the hippocampal BDNF mRNA levels. Previous in vivo studies have
shown that Zn 2+ supplementation increases the mRNA expression of BDNF 35 . On the contrary, pharmacological manipulations aimed at reducing the brain Zn 2+ levels negatively modulate the BDNF mRNA levels 48 . To address whether CQ-driven Zn 2+ chelation interferes with the BDNF expression, we evaluated mRNA levels in the hippocampi of CQ-and vehicle-treated mice. Quantitative Real-Time PCR (qRT-PCR) analysis showed that CQ administration did not affect the hippocampal BDNF gene expression ( Fig. 2J; 1.00 ± 0.02 in control vs 1.12 ± 0.17 in CQ treated mice, p = 0.54). CQ treatment reduces dendritic spine density in vitro and ex vivo. As BDNF potently modulates synaptic plasticity 34,49 , we first assessed, in an in vitro setting, whether CQ treatment alters the number of dendritic spines and/or their morphology. To that aim, EGFP-transfected primary hippocampal neurons were exposed to CQ (10 µM for 3 days) or vehicle and spines evaluated with confocal microscopy. Confocal imaging analysis showed that, compared to vehicle-treated sister cultures, CQ-treated neurons underwent a significant reduction in dendritic spine density (Fig. 3A,B; spines per µm: 0.51 ± 0.07 in control vs 0.22 ± 0.02 in CQ-treated cultures, p = 0.020). No differences were found, between the two study groups, regarding spine volumes, areas, diameters or lengths ( Fig. 3C-F). To assess whether CQ also impairs spine pruning in vivo, we performed Golgi staining on brain slices of treated animals. Compared to vehicle-treated mice, the hippocampal CA1 pyramidal neurons of CQ-treated mice showed a significant decrease (≈−30%) in their dendritic spine density (  CQ reduces MMP activity in vitro. MMPs, and MMP-9 in particular, promote BDNF maturation through the proteolytic cleavage of pro-BDNF 9 . To address whether CQ interferes with MMP activity we performed a gelatin zymographic essay on MMPs isolated from the medium of cultured hippocampal neurons. After gel electrophoresis, the MMP developing buffer was supplemented with CQ (10 µM) or the high-affinity Zn 2+ chelator TPEN (100 µM, here employed as a positive control) and results compared with those of control gels. CQ almost completely abolished the MMP-9 and MMP-2 activity ( Fig. 4; normalized MMP-9 activity: 1.00 ± 0.09 in control vs 0.08 ± 0.02 for CQ-treated gels, p < 0.001; normalized MMP-2 activity: 1.00 ± 0.17 in control vs 0.02 ± 0.01 for CQ-treated gels, p < 0.001). Gel treatment with TPEN did not induce further MMP reduction, thereby supporting the notion of a maximal inhibitory activity exerted by CQ (Fig. 4A,B).

Discussion
In this study, we investigated the cognitive effects of Zn 2+ chelation in young mice. Previous studies have shown that intraperitoneal CQ administration promotes a robust reduction of the pool of chelatable Zn 2+ in rodents [38][39][40][41][42][43] . The main finding of the study is that a 2-week administration of CQ exerts adverse effects on learning and memory performances of treated mice. These results are in line with evidence supporting the critical role played by Zn 2+ in the modulation of neuronal functioning 32,43,50,51 . The study also indicates that the CQ-dependent effects are mediated by the inhibition of the BDNF neurotrophic axis.
CQ administration affects BDNF signaling, synaptic plasticity, and behaviour. Zn 2+ chelation profoundly affects mBDNF levels and the related signaling pathway. In CQ-treated animals, we observed a significant reduction in the total content of mBDNF. These changes were associated with a decrease of the total amount of the TrkB receptor. The decreased mBDNF levels that we observed were regionally restricted to the hippocampus, cortex, and striatum (Fig. 2). No effects were detected in the cerebellum, a finding that is in line with the hypothesis that the CQ-driven activity on BDNF signaling is selectively mediated by Zn 2+ removal as the BDNF inhibition does not occur in a region, the cerebellum, that is devoid of pools of chelatable Zn 2+ 52 . These results also parallel the behavioural findings as the CQ-dependent cognitive effects are appreciated on a task, the MWM, that evaluates the hippocampus-related memory performances. mBDNF exerts its neurotrophic action by binding to the TrkB receptor, and the study shows that CQ promotes a significant decrease in TrkB receptor levels. Unexpectedly, this result was not associated with a parallel reduction in the phosphorylated, activated, form of TrkB as we find that the pTrkB/TrkB ratio is unaffected by the drug treatment. The design of our experimental setting may help to explain the discrepancy. In our model, the molecular effects of CQ administration were assessed at the end of the 2-week treatment period. Within this time frame, it is conceivable that the CQ-mediated decrease in mBDNF is sufficient to induce a significant spine remodelling and/or elimination. This notion is, indirectly, supported by the observed reduction in PSD95 levels and further confirmed by the in vitro and ex vivo decreases of spine density that we observed (Fig. 3). Thus, the CQ-driven reduction of dendritic spines may have led to a decrease in the spine-associated transmembrane proteins including TrkB, thereby producing a net reduction of the receptor levels without affecting the pTrkB/TrkB ratio. The idea of defective TrkB signaling is further corroborated by the observed CQ-driven decrease in pERK5 levels, a kinase that acts downstream of the BDNF-TrkB neurotrophic axis and is known to promote neurogenesis and neuronal survival 53,54 but is also implicated in learning-and memory-related processes 55,56 .
Our findings support a role for CQ-driven Zn 2+ deprivation in the modulation of synaptic plasticity and dendritic spine pruning. CQ-treated hippocampal neurons show a reduced number of spines when compared with vehicle-treated sister cultures (Fig. 3). Our data are in line with the notion that mBDNF participates in structural plasticity by modulating the spine pruning and morphology [57][58][59] . These plasticity-related phenomena represent the biophysical underpinning of the processes controlling the experience-dependent information storage occurring within the brain circuitry. Thus, the CQ-driven reduction of mBDNF levels may serve as one of the molecular mechanisms underlying the behavioural impairments observed in our model.
In the study, we investigated the Zn 2+ -dependent MMP activity with the aim of providing a mechanistic link between CQ-administration, Zn 2+ chelation, and the reduction of mBDNF levels. Among MMPs, MMP-9 plays a permissive role in favouring structural plasticity and promote the dendritic spine expansion 60 . Experimental evidence indicates that LTP-induced increase in spine size requires MMPs proteolytic activity 26 and, conversely, inhibition of MMPs has been shown to impair spatial learning in adult animals 61,62 . In addition, recent findings have shown that MMP-9 participates, in a Zn 2+ dependent manner, to BDNF maturation 9,10 . In agreement with this notion, our in gel zymography experiments showed that CQ concentrations as low as 10 µM are sufficient to abolish the MMP gelatinolytic activity (Fig. 4). A further mechanistic link relies on the Zn 2+ -dependent transactivation of TrkB 34 , an event that is likely occluded by CQ administration. While the mechanism is possible from a theoretical standpoint, it is unlikely to have occurred in our setting as we did not observe significant changes in TrkB phosphorylation and activation. In addition, a recent report has indicated that the receptor transactivation by Zn 2+ occurs in in vitro settings and only marginally in vivo 49 .
This Zn 2+ -dependent modulation of the MMP activity is of particular relevance when considering the cation role in synaptic plasticity. Zn 2+ is particularly enriched within synaptic vesicles of a subset of hippocampal glutamatergic terminals 31,32,63 . At these synapses, Zn 2+ is released upon neuronal activity and plays a crucial role in the modulation of physiological and pathological processes 32 . Substantiating this critical activity on plasticity and cognition, transgenic mice devoid of the synaptic zinc transporter-3 (ZnT-3 KO mice), and therefore unable to release Zn 2+ , show behavioural and molecular deficits that closely resemble the ones observed in our CQ-treated mice. Further suggesting a commonality of activities between transgenically-or pharmacologically-driven Zn 2+ deprivation, ZnT-3 KO mice, like CQ-treated animals, show alterations in the BDNF-related pathways and plasticity markers 64 . In summary, CQ administration, by chelating Zn 2+ , reduced the MMP activity that, in turn, decreased BDNF maturation and defective structural plasticity, thereby leading to impaired cognitive performances.

CQ administration modulates neurotoxic pathways in a pro-BDNF-independent manner.
mBDNF is strongly related to its precursor form pro-BDNF. The decrease of mBDNF that we found was associated with the overexpression of the p75NTR receptor as well as with increased ERK 1,2 signaling, a pathway that is activated by the pro-BDNF. Our WB data indicate that the treatment also produced decreased of pro-BDNF levels (Fig. 2G). These unexpected results prompted us to investigate whether CQ administration has affected the BDNF gene expression as previous studies have shown that changes in Zn 2+ availability impinge on BDNF expression 35,48 . qRT-PCR experiments, aiming at investigating the issue, indicated that the drug treatment did not alter BDNF mRNA levels. The additional targets of MMPs offer a possible explanation for these results. In addition to BDNF, MMPs are known to promote the cleavage and maturation of other pro-neurotrophins, including pro-NGF, a process that activates a deleterious p75NTR-related signaling 65 . An additional, untested, possibility revolves around CQ effects on other intracellular Zn 2+ -dependent proteases that are involved in the BDNF maturation 10,66 . Moreover, the reduction of mature neurotrophins, including mBDNF, can lead to a feed-forward increase of p75NTR expression and activation 67 . Given the fact that TrkB and p75NTR mediate opposite effects 13,68 , it is therefore conceivable that the mBDNF reduction unleashes a damaging increase in p75NTR levels (Fig. 4H). The hypothesis is unlikely in our setting as the enhanced p75NTR levels did not promote the activation of death pathways. The treatment did not result in increased neuronal toxicity, and the compound was found to be safe and well tolerated (Fig. 1B, and Supplementary Fig. 2). Our p75NTR results, instead fit with previous findings showing that p75NTR-related signaling facilitates the activation of NMDAR-dependent long-term depression 22,69 , negatively modulates dendritic morphology and spine density in hippocampal pyramidal neurons 4,70 , and contributes to synaptic and memory deficits associated with neurodegenerative conditions 71 . Similarly, aberrant activation of ERK 1,2 not only results in the activation of death-related pathways but has been recently implicated in memory impairment in a mouse model of autism 72 .

Conclusions
Our findings support the notion that Zn 2+ , through the activation of the BDNF signaling cascade, is an important modulator of synaptic plasticity, neuronal functioning and cognitive processes. Impairment in the pro-BDNF/ mBDNF signaling cascade may have important clinical implications. Altered neurotrophin biosynthesis and processing has been indicated to be a common feature in several neurological conditions and affect complex human behaviors [73][74][75][76] . Pharmacological tools engineered to restore effective BDNF signaling can, therefore, have a great impact on development-aging-or disease-related physio-pathological settings.
CQ has been employed in preclinical models of neurodegenerative diseases (AD and Huntington's disease) with encouraging results [77][78][79][80] . These robust preclinical findings were matched by positive effects in clinical trials exploring the use of CQ or other metal-protein-attenuating compounds (MPACs) in AD patients [81][82][83][84][85] . Our study underscores the importance of choosing "when", "where", and "how much" modulation has to be applied when dealing with brain Zn 2+ levels. Our study indicates that in young animals, a model conceivably devoid of overt signs of Zn 2+ dyshomeostasis, the CQ-driven deregulation of physiological brain levels of the cation (Supplementary Fig. 1) as well as the related impaired BDNF-signaling have strong functional and structural consequences. Conversely, when aging-and/or disease-related Zn 2+ deregulation has occurred, the use of pharmacological tools, like CQ, to quench the neurotoxic effects of the metal is highly beneficial 80,86,87 , thereby supporting the idea that redistributing Zn 2+ in these contexts may overrun potentially negative side-effects on synaptic plasticity and neurotrophic signaling 37,51 . Thus, our findings do not challenge the importance of targeting Zn 2+ , with CQ or similar MPACs, as a therapeutic approach against neurodegenerative conditions but at the same time highlights the potential drawbacks of reducing brain Zn 2+ availability in the early stages of development.

Materials and Methods
Animal model and treatment paradigm. All the procedures involving the animal handling and their care were approved by the University G. d' Annunzio Institutional Ethics Committee (CEISA protocol no. AD-301). In vivo procedures were performed in compliance with institutional guidelines and in accordance with national and international laws and policies. Mice were bred in the CeSI-MeT animal facility, single-housed, maintained on a 12 h light/dark cycle, and provided with, ad libitum, access to food and water. Behavioral tests were all performed during the early phases of light cycle (8 am-11 am). C57Bl6 male mice were enrolled at 2.5 months of age (m.o.a.) and randomly assigned to a 2-week CQ or vehicle (see below) administration. Animals were treated daily with an intraperitoneal injection of CQ (30 mg/kg body weight) or vehicle. CQ was prepared by first dissolving the drug in DMSO (80 mg/ml) and then diluted in castor oil to a final concentration of 4 mg/ ml (with a final 5% DMSO content).
Mice body weight was measured daily as a gross index of the overall safety of the treatment. A cohort of animals was employed for behavioural assessment, a second group which did not perform a behavioural evaluation, was employed for biochemical analysis.
Neuronal cultures. Primary hippocampal neuronal cultures were prepared as previously described 56,88 .
Briefly, hippocampi were obtained from C57Bl6 fetuses at 15 or 16 embryonic days (E15-E16). Alternatively, some experiments were performed on hippocampi obtained from CD1 mice. Hippocampi were dissected on ice and transferred in a 0.25% trypsin/EDTA solution for 10 min at 37 °C, then pelleted by centrifugation, and dissociated with a fire-polished glass pipette. The neuronal suspension was then diluted in Neurobasal medium supplemented with 0.5 mM l-glutamine, 5% fetal bovine serum, 5% horse serum, 1 × B27 and 0.2% penicillin/ streptomycin and plated onto pre-treated laminin/poly-dl-lysine coated tissue culture glass coverslips. Three days after plating, non-neuronal cell growth was inhibited by adding 5 μM of cytosine arabinofuranoside, to obtain near-pure hippocampal cultures. 25% of the medium was replaced with fresh Neurobasal two times per week. Experiments were performed on cultures between 12 and 18 days in vitro (12)(13)(14)(15)(16)(17)(18).

Morris Water Maze. The Morris
Water Maze (MWM) test was performed as previously described 56,89,90 . In brief, the MWM test was performed with a circular pool (Panlab; 1.2 m diameter) filled with warm (22 ± 1 °C) water. The pool is placed in a room with several intra-and extra-maze visual cues for mice orientation. Mice were habituated to the testing room and placed on the platform 10 s before the first training session to reduce task-related stress. Mice underwent four trials per day for 4 consecutive days. 1.5 and 24 h after the end of the last training trial mice were tested for spatial memory performances. 1.5 and 24 h probe tests consisted of a 60 s free swim in the pool without the submerged platform. Probe sessions were digitally recorded and analyzed. The following parameters were measured: time to reach the location where the platform used to be (escape latency), number of crosses over the platform location (crosses), time spent in the target (T-target) or opposite (T-opposite) quadrants, average speed (cm/s), and total distance travelled during the LTM session.
Tissue collection and reagents employed. Mice were euthanized after the 2-week treatment and tissue samples collected for further analysis. Brains were halved into two hemispheres, snap frozen by liquid nitrogen, and kept at −80 °C until used. For Western blot (WB) analysis, each hemisphere was dissected into sub-regions (hippocampus, cortex, striatum, and cerebellum), snap frozen in liquid nitrogen, and stored at −80 °C until sampling. For quantitative Real-Time PCR (qRT-PCR) hippocampi obtained from CQ-and vehicle-treated mice were stored in RNAlater solution (ThermoFisher Scientific) until RNA extraction. For Golgi staining hemispheres were rinsed in PBS and immediately transferred to Golgi-Cox solution following manufacturer instructions (see below). All standard chemicals and reagents were, unless otherwise specified, purchased from Sigma-Aldrich. Briefly, 3 h after vehicle or CQ administration, mice were sacrificed, brains extracted, and cryosectioned without fixation. 6 µm thick slices were immediately stained with TSQ (100 µM; Santa Cruz Biotechnology) diluted in borate buffer [BB; containing (in mM): 100 boric acid, 25 sodium tetraborate, 75 NaCl, pH 9.5] for 90 s. Slices were then rinsed in PBS and put on coverslips. Images were acquired with a LSM 800 confocal microscope (Zeiss) by exciting the specimen with a 405 nm LED laser source. DRAQ5 (5 µM) was employed to counterstain the neuronal nuclei.
Protein assay. Protein concentration was assayed by employing the Pierce BCA Protein Assay kit (Thermo scientific). Briefly, the assay is a detergent-compatible formulation based on bicinchoninic acid (BCA) for the colorimetric detection and quantitation of total protein. The method combines the reduction of Cu 2+ to Cu + ions by protein in alkaline medium (the biuret reaction) with the high sensitive and selective colorimetric detection of the cuprous cation by using a reagent containing BCA. The purple-colored reaction product of this assay is formed by the chelation of two molecules of BCA with one cuprous ion. This complex exhibits a strong absorbance at 562 nm.
Golgi staining. Hemispheres from vehicle-and CQ-treated mice were stained with an enhanced Golgi-Cox staining system (superGolgi Kit, Bioenno Tech), following manufacturer instructions. Briefly, 150 µm thick brain sections were cut using a vibratome (Leica) and mounted on adhesive microscope slides. Air-dried sections were stained, cleared in xylene, coverslipped in Eukitt mounting medium, and stored in the dark until analyzed. Images from secondary dendrites of CA1 pyramidal neurons were acquired with a DMRD microscope (Leica) equipped with a DFC550 CCD camera (Leica) and a 100× oil-immersion objective (1.3 NA; Leica). A trained experimenter blind to treatment manually counted the spines.
Gelatin zymography. Gelatin zymography was performed as previously described 36 . SDS-PAGE gels (7.5%) were prepared by incorporating 0.1% gelatin during acrylamide/bis-acrylamide polymerization. In all lanes, 10 µl of the supernatant, obtained from cultured hippocampi, was loaded. Gels were subjected to electrophoresis at 125 V constant. After electrophoresis, SDS was removed by washing gels for 30 min with a renaturating solution containing (in mM): 50 Tris-HCl, pH 7.4, and 2.5% Triton X-100. Gels were further washed with large volumes of ddH 2 O to remove residual Triton X-100. To assess MMPs activity gels were incubated overnight at 37 °C in an activation buffer containing (in mM): 50 Tris-HCI, pH 7.4, 120 NaCl, 5 CaCl 2 . Accordingly to the experimental setting, activation buffer was supplemented with CQ (10 µM), TPEN (100 µM), or vehicle (DMSO 1%). Gels were then incubated for 30 min with a developing solution containing (in %): 0.1 Coomassie blue, 40 methanol, 10 acetic acid, 50 ddH 2 O. Staining was stopped when clear gelatinolytic bands appeared. Gels were photographed with a Gel-Doc apparatus (Biorad) and bands quantified by densitometry by using the Fiji distribution of ImageJ (v 1.51) 92 .
RNA extraction and qRT-PCR. The NucleoSpin Tissue RNA isolation kit (Macherey-Nagel) was employed for total RNA extraction. One µg of RNA was retro-transcribed using the High Capacity RNA-to-cDNA Kit (ThermoFisher Scientific). qRT-PCR was performed in a total volume of 25 μl containing 2 × Maxima SYBR Green/ROX qPCR Master Mix (ThermoFisher Scientific), 1 µL of cDNA and 0.3 µM of each primer on an Abi 7900HT Sequencing Detection System (ThermoFisher Scientific). BDNF relative expression was corrected against GAPDH and HPRT1 that were used as endogenous controls. Amplification conditions were: 2 minutes at 50°, 10 minutes at 95 °C, followed by 40 cycles of 15 seconds at 95 °C and 1 minute at 60 °C. A melting curve was run to assess the specificity of primers employed. Samples were run in triplicate. The genes relative fold changes were calculated by the ΔΔCt method.
Neuronal viability assay. To exclude a possible neurotoxic effect of the compound, we assessed cell viability in vitro. Primary hippocampal neurons were exposed for up to 3 days to CQ (10 µM) or vehicle (DMSO). CQ toxicity was evaluated by lactate dehydrogenase (LDH) efflux assay as previously described 93 .
Neuronal viability was also indirectly evaluated in Golgi stained ex vivo brain slices. Briefly, 20× magnification images of hippocampal and cortical areas were acquired and the number of stained neurons per field was counted and averaged.
In vitro analysis of spine morphology. Spine morphology analysis was performed by transfecting primary hippocampal neurons with an EGFP containing construct. Transfection was performed at 7 DIV with Lipofectamine 2000 (ThermoFisher) following manufacturer instructions. Transfected cultures were subsequently treated with CQ or vehicle for 3 days. Primary and secondary proximal dendritic processes of CQ-or vehicle-treated transfected neurons were imaged with a LSM 510 META confocal microscope (Zeiss) by exciting the specimen with a 488 nm Argon laser. z-stack imaging series were acquired and stored for further analysis. Dendrites were considered "proximal" when within 50 μm from the soma. Distances from the neuronal body were measured with the Zeiss proprietary ruler tool. 3-dimensional reconstructions were obtained from stacked images and analyzed with the Fiji 3D Object Counter 94 function. Once images were corrected for artifacts (i.e.: fluorescent particles unrelated to the analyzed neuron/region) and background subtracted, the plugin counted the number of the selected 3D particles present in the stack and automatically provided the volume and the surface area of each object. Spine lengths and diameters were, instead, evaluated with a semi-automated approach with the Reconstruct software 95 .

Statistical analysis.
No methods were employed to predetermine sample size. Statistical analysis was performed by two-tailed Student's t-test. For gelatin zymography ANOVA following Bonferroni post-hoc test was employed. The level of significance was set at p < 0.05. Data were expressed as mean ± SEM. Data availability. The data of the study are available from the corresponding author upon reasonable request.