The intellectual disability PAK3 R67C mutation impacts cognitive functions and adult hippocampal neurogenesis

HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés. The intellectual disability PAK3 R67C mutation impacts cognitive functions and adult hippocampal neurogenesis Charlotte Castillon, Laurine Gonzalez, Florence Domenichini, Sandrine Guyon, Kévin da Silva, Christelle M. Durand, Philippe Lestaevel, Cyrille Vaillend, Serge Laroche, Jean-Vianney Barnier, et al.

Intellectual disability (ID) is characterized by impaired cognitive functions and adaptive behaviors (1).
Although the causes of ID are highly heterogeneous, genetic factors take a large part in the etiology of ID and numerous cases are caused by mutations in genes located on the X chromosome (2). The proteins encoded by ID genes regulate diverse cellular processes including neurogenesis, neuronal F o r P e e r R e v i e w 3 migration, synaptic function, and regulation of transcription and translation (3,4). While abnormalities in synaptic plasticity and dendritic spine morphogenesis have been reported to contribute to cognitive deficiencies in models of ID, alterations in different steps of adult hippocampal neurogenesis have also been linked to cognitive deficits in several models of syndromic ID. For example, selective alterations of proliferation, differentiation, neuronal maturation and survival of adult born neurons or defects in their integration into brain networks have been reported in animal models of ID that display cognitive deficits (5). Few recent data also concur in reinforcing this link between gene mutations leading to ID and dysfunction in adult hippocampal neurogenesis (6,7,8,9). To further investigate this issue, we chose as a case study the X-linked p21-activated kinase-3 (PAK3) gene whose reported mutations result in mild to severe ID (10,11,12).
The Pak3 gene is evolutionary conserved among vertebrates and belongs to group-I PAK family of serine/threonine kinases. PAK3 is highly expressed in brain, in particular in cortex and hippocampus (13). At the cellular level, PAK3 is a downstream effector of the small Rho-GTPases Rac1 and Cdc42 (14). It has been implicated in the LIM kinase pathway that controls actin cytoskeleton dynamics, in the MAP kinase activation and in AMPA-receptor trafficking, mechanisms that play major roles in synaptic plasticity and learning and memory (13,15,16). Functionally, genetic invalidation of the Pak3 gene in mice is associated with selective deficits in hippocampal synaptic plasticity and reduced CREB phosphorylation in cortex and hippocampus (17). PAK3 was also identified as an important player in distinct cellular processes such as in the control of cell cycle exit and differentiation of neuronal precursors in Xenopus embryos (18), in the differentiation of oligodendrocyte precursors (19) and in the time course of interneuron migration during cortical ontogenesis (20).
To date, no study has as yet explored these cellular processes in the adult in vivo. Here, we hypothesized that specific ID mutations leading to PAK3 dysfunction may alter adult hippocampal neurogenesis and neurogenesis-dependent cognitive functions. We therefore generated a novel mouse model of Pak3-dependent ID by knocking-in the R67C missense mutation (Pak3-R67C mouse), strictly identical to the human mutation responsible for moderate to severe non-syndromic ID (21,22). This mutation changes arginine 67 to a cysteine residue in the regulatory domain of PAK3 in F o r P e e r R e v i e w 4 proximity to the p21 GTPase-binding domain. This mutation decreases the binding of PAK3 to Cdc42 while increasing its binding to Rac1 (23). We characterized the behavioral phenotype of this novel ID model and focused, in particular, on adult neurogenesis-dependent tasks, assessing spatial memory and pattern separation function. We then examined whether the R67C mutation affects any of the key steps of basal adult hippocampal neurogenesis by assessing progenitor cell proliferation, differentiation and survival of adult-born neurons. Finally, we explored adult hippocampal neurogenesis in relation to spatial learning by analyzing the effect of the mutation on neuronal survival and the recruitment of newborn neurons by learning.

Generation of the Pak3-R67C mouse model and hippocampus expression of PAK3
To provide insights into the effects of the Pak3-R67C mutation, we generated a knock-in mouse model expressing the mutated Pak3 gene encoding the PAK3-R67C protein. The generation of Pak3-R67C mice was done by mutagenesis of the CGC>TGC mutation in the sequence of the second coding exon in the recombination vector (Fig. 1A). We verified that the HhaI site suppressed by this mutation (22) was absent from the genomic fragment of transgenic males (Fig. 1B). The genomic mutated sequence of exon 2 was confirmed by cloning (data not shown). The mutated allele segregated with Mendelian ratio and mutated hemizygous males and homozygous females were viable and fertile, with normal life span. Mutated hemizygous males and their WT littermate reached adulthood at a normal weight (Pak3-R67C mice: 28.44g ± 0.475, n=9; WT: 28.11g ± 0.633, n=9). PAK3 mRNAs were transcribed at a similar level in brain from transgenic and WT adult males as verified by quantitative RT-PCR (data not shown). As shown by western blotting experiments, the PAK3-R67C protein was detected in adult whole brains extracts from transgenic mice (R67C) as a band of similar molecular weight compared to WT mice (Fig. 1C). The expression of the mutated protein was found to be similar to that of the WT protein, as quantified by western blot (Fig. 1C, mean relative PAK3-R67C protein expression compared to WT expression after normalization to β-Actin: 113. 3 6 We analyzed spatial learning and memory performance in the water maze using two distinct training protocols with two different cohorts of mice. When training was distributed over 11 days (4 successive trials per day), mice of both genotypes showed comparable learning curves ( Fig We then submitted another cohort of WT (n=7) and Pak3-R67C (n=6) mice to a more stringent spatial learning and memory task using a massed training protocol performed within a single day in the water F o r P e e r R e v i e w 7 maze, an experimental design that is more sensitive than distributed training in certain geneticallymodified mice (26) or in mice with altered hippocampal neurogenesis (27,28) and analyzed long-term spatial memory performance 10 days after training. During training, the swim path lengths and time to reach the platform decreased significantly in both genotypes ( Hence, mice from both genotypes learned similarly the task. Spatial long-term memory was probed 10 days later. In contrast to WT mice who spent significantly more time in the target quadrant, Pak3-R67C mice spent as much time in the target quadrant as in the other quadrants ( Fig. 2E Pak3-R67C t(5)= 0.418, p=0.6934). The mean number of crossings of the platform location was significantly smaller in Pak3-R67C than in WT mice ( Fig. 2F; factorial ANOVA, F(1,11)=5.242, p=0.0428) and Pak3-R67C displayed slightly longer distance swum before reaching the target platform position (factorial ANOVA, distance before first crossing: F(1,11)=5.461, p=0.0394). During this probe test, swim speeds were comparable between genotypes (factorial ANOVA, F(1,11)=2.674, p=0.1302). Hence, despite normal spatial learning in both protocols, Pak3-R67C mice displayed longterm spatial memory deficits in the more difficult massed training protocol.

Normal spatial working memory but spatial pattern separation deficit in Pak3-R67C mice
We then examined mice performance in a radial-maze to evaluate their working memory and their ability to finely discriminate spatial configurations. In the working memory paradigm, in which mice had to recover rewards located at the end of the 8 arms of the radial maze without returning to a  Then, to analyze spatial pattern separation function, we submitted Pak3-R67C and WT mice to a spatial delayed non-matching to place (DNMP) paradigm in the radial-maze (29, Fig.3A). During this task (4 trials/day during 10 days), mice (18 WT and 17 Pak3-R67C mice) were tested for their ability to distinguish two spatial configurations that differed by a variable distance between two arms positions of the radial-arm maze. In each trial, mice were submitted to a sample phase in which they can run from a start arm to a single open rewarded arm of the maze and, after a 20s delay, were replaced into the start arm and had to choose between the previously rewarded arm (now non-baited) and a new open and rewarded arm. This newly rewarded arm could be either close or distant to the previously visited arm (Fig. 3A) and the sets of 3 arms varied randomly among trials and during the 10 days of the experiment. Thus, this paradigm tests the ability to spatially discriminate between 2 arms of the maze, either spatially close to each other ("close" configuration) or more separated ("distant" configuration; see methods). In both genotypes, the percent of correct trials in both "close" and "distant" configurations increased regularly over the 10 days of training (Fig. 3B) p=0.0013). These results suggest no specific deficit in Pak3-R67C mice during the training phase of this task, whatever the configuration (close or distant). On day 11, we separated the initial group into two sub-groups of each genotype, which were each submitted to a probe test with 3 successive trials in only one given configuration (either in the close configuration: 9 WT and 9 Pak3-R67C mice or the distant configuration: 9 WT and 8 Pak3-R67C mice). During this probe test, WT mice performed well in both configurations ( Fig. 3C; one-sample analysis to chance: close configuration t (8) hippocampal-dependent spatial pattern separation deficit in Pak3-R67C mice in conditions of higher interference due to the immediate succession of 3 consecutives trials with the same configuration.

Hippocampal pattern separation deficit and high sensitivity to interference in irradiated WT mice and in Pak3-R67C mice
The pattern of deficits observed in Pak3-R67C mice, selectively expressed as a long-term spatial memory deficit when retention was tested several days after massed training and as a spatial pattern separation deficit under high interference, is a common phenotype observed in mice with altered adult hippocampal neurogenesis (7,27,28,29). Moreover, there are hints to suggest that altered hippocampal neurogenesis can affect the proper management of interference (30,31,32). To test this hypothesis more specifically, we used another cohort of mice to examine their performance in a hippocampal-dependent successive spatial delayed non-matching to place task developed by Al Abed et al. (33), which tests memory flexibility and the ability to overcome proactive interference. This task in a radial-maze requires the ability to retain recently visited places, i.e. to remember which arm was the more recently visited arm within each of 3 repeatedly used pairs of arms, over varying intervals ( Fig. 4A-B). This task is therefore characterized by high organizational demand due to the varying delay separating repetitions of similar events (33, and see methods). During behavioral testing, each  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  P  e  e  r  R  e  v  i  e  w 10 mouse was attributed a set of 3 pairs of adjacent arms presented in a pseudo-random order (Fig. 4A).
For each arm pair, the position of the reward was alternated between one trial of that pair and the next one of the same pair (Fig. 4B). During the 15 days of training, mice were submitted to 23 successive trials with a 10 s inter-trial interval (ITI). Performance was measured as the percentage of correct trials over training and, for the last 3 days of training, proactive interference (PI) was calculated as a function of the number of interfering trials (ITn) that corresponds to the number of trials between 2 trials of the same pair (see methods for details). Using c-Fos labeling, Al Abed et al. (33) showed that realization of the task relies heavily on DG activity and pattern separation function. Although this is suggestive of a possible involvement of DG newborn neurons, there is as yet no evidence that adult neurogenesis blockade would cause deficits in this task. To investigate this point, we first tested in this behavioral task performance of WT mice submitted to targeted irradiation of the dorsal DG to focally ablate their adult born neurons (34). For this, a group of 10 C57BL/6 wild-type mice were exposed to X-ray ionizing radiation using the Small Animal Radiation Research Platform (SARRP) of IRSN (Fontenay-aux-Roses, France, see methods), which delivers targeted radiation to the dorsal DG with high accuracy, following the methods and protocol described in (34). Eight non-irradiated sham mice (NIR) served as controls (see methods). To avoid potential advert effects of neuroinflammation caused by irradiation, mice were trained two months after irradiation (29,35 . Performance was then expressed as a function of the number of interfering trials (ITn) between two presentations of the same pair. This revealed that performance of NIR mice became slightly lower (from 80% to 72% correct) as the ITn increased, but remained significantly above chance whatever the ITn ( Fig. 4D; one sample t test ITn 0: NIR: t(7)=9.756, p<0.0001; ITn 1-2: NIR: t(7)=9.888, p<0.0001; ITn 3-4: NIR: t(7)=5.173, p=0.0013). In contrast, performance of IR mice decreased when the ITn increased and was significantly different from NIR mice both for an ITn 1-2 and an ITn 3-4 (  Fig. 4G) was also expressed as a function of the ITn between two presentations of the same pair. This analysis showed that performance of WT mice decreased slightly (from 87% to 65% correct) as the ITn increased, but remained significantly above chance level even for the ITn 3-4 (Fig. 4G). In contrast, performance of Pak3-R67C was significantly lower compared with WT mice, whatever the number of interfering trials, even when dysfunction could be associated with altered adult hippocampal neurogenesis.

Selective alterations of adult DG neurogenesis in Pak3-R67C mice
As previously shown, PAK3 is expressed in the DG (Fig. 1D). To specify the identity of the cells expressing PAK3 in DG, we performed immunofluorescence labeling on lacZ-stained sections of Pak3-tm1b-lacZ reporter mice with neuronal markers. As shown in Fig.5A, most PAK3 expressing cells were localized in the subgranular zone of the DG. They expressed NeuN and some of them coexpressed DCX, suggesting that PAK3 is expressed in both mature and immature neurons of the DG.

Delayed functional and morphological maturation and deficient recruitment of newborn DGCs into spatial memory networks
Guided by the evidence that adult-born DGCs can be recruited upon hippocampal-dependent spatial learning (7,28,36,37,38), we studied the activation of adult-born DGC following spatial memory recall using the activation marker Zif268. For this, we used mice trained in the massed training regimen in the water maze described above, since Pak3-R67C mice can learn this task as well as WT  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  P  e  e  r  R  e  v  i  e  w 15 activated upon recall and not impacted by the dysfunction of PAK3. Finally, the proportion of mature neurons expressing Zif268 in WT mice after memory recall was significantly higher than in the HC condition ( Fig. 6E; Mann-Whitney test: WT: p=0.05). Although this increase after recall did not reach significance in Pak3-R67C mice ( Fig. 6E; Mann-Whitney test: Pak3-R67C: p=0.2207), there was no significant genotype difference in the proportion of Zif268 + /NeuN + cells after recall ( Fig. 6E; Mann-Whitney test: p=0.6015). Thus, memory recall in our training conditions resulted in an activation of a similar proportion of mature DGCs in WT and Pak3-R67C mice, and a selective, learning-related activation of a population of young immature neurons in WT mice, which failed to be activated in Pak3-R67C mice. This suggests that the long-term spatial memory deficit in Pak3-R67C mice is associated with the absence of recruitment of this neuronal population by learning.
This specific phenotype of Pak3-R67C mice has also been observed in another mutant mouse for which deletion of the gene Zif268 led to a similar accelerated death during the critical window of maturation of newborn DGCs and a defect of recruitment during learning, effects that were attributed to delayed functional maturation and altered morphological maturation of young newborn DGCs (28).
Indeed, around 3 weeks of age, newborn DGCs undergo extensive functional synaptic changes that are essential for their functional maturation and their survival (40). At this age, the cells start to express the AMPA-type glutamate receptor GluA1 (41) and also the Cl − ionic co-transporter, KCC2b, that contributes to the formation of mature dendritic spines and functional excitatory synapses (42) and is implicated in the conversion from GABA-induced depolarization to hyperpolarization, a mechanism crucial for synaptic integration of 3-week-old DGCs (43,44). We therefore analyzed GluA1 and KCC2b expression in 18-day-old DGCs of Pak3-R67C and WT mice. We found that the proportion of BrdU-labeled 18-day-old newborn DGCs expressing GluA1 was similar between genotypes (data not shown; Mann-Whitney test: p>0.9999). However, the expression pattern of KCC2b revealed a significant decrease in the expression of this Cl − co-transporter in BrdU-labeled 18-day-old DGCs of Pak3-R67C mice compared to WT mice ( Fig. 7A-B; Mann-Whitney test: KCC2b: p=0.0090). We also analyzed the dendritic morphology of immature neurons (DCX + cells; Fig. 7 C-E). We found that the total length of dendrites of DCX + cells was significantly shorter in Pak3-R67C compared to WT mice  and selection, associated with impairment in their morphological development. These effects might be instrumental in their accelerated death during this time period and lack of recruitment in relation to learning and memory.

Discussion
In the present work, we generated a novel mouse model of ID bearing the missense mutation R67C of the Pak3 gene to characterize the effect of this mutation in cognitive functions and explore the possibility that the cognitive deficits caused by this mutation may be associated with altered adult hippocampal neurogenesis. The mutation generated in our mouse model is strictly similar to the human mutation (22). The knock-in mutation did not alter the expression of the mutated gene suggesting that the mouse phenotype is not due to a defect of Pak3 gene expression or PAK3 protein synthesis. Hemizygous males and heterozygous females displayed normal life and were fertile.
Moreover, the R67C mutation did not alter the volume and general organization of the hippocampal formation, in line with the absence of major brain morphological defect reported in patients bearing this mutation in a three-generation tree (21,22).
Our behavioral characterization showed that Pak3-R67C mice display normal locomotion and exploration, no specific alteration of anxiety-like behaviors and no deficit in spatial working memory in a radial maze. We also report that Pak3-R67C mice do not display spatial learning and memory deficits in a distributed protocol in the water-maze. However, using a more stringent spatial learning task based on a massed protocol, we found that despite normal spatial learning, long-term spatial memory was deficient in Pak3-R67C mice. Moreover, using a delayed non-matching to place task placing varying demands on spatial discrimination ability (29), we showed that Pak3-R67C mice display alterations in spatial pattern separation function and a higher sensitivity to interferences. This  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  P  e  e  r  R  e  v  i  e  w was confirmed in the hippocampal-dependent successive spatial delayed non-matching to place task that models declarative memory of repetitive everyday events (33). This task in a radial-maze is characterized by high organizational demand due to the varying delay separating repetitions of similar events and allows probing memory flexibility and organization, and susceptibility to proactive interference. Our results show that, despite normal spatial working memory, Pak3-R67C mice exhibit a higher sensitivity to proactive interference, suggesting impaired organization of events in memory.
Together, these results demonstrate for the first time a link between PAK3 dysfunction and alterations in spatial memory, in pattern separation function and notably in the management of interference, which are in line with impaired spatial cognitive skills and deficits in attentional and executive functions reported in patients holding mutations in the Pak3 gene (45).
Numerous studies showed that defects in adult hippocampal neurogenesis result in alterations of spatial memory and pattern separation function (29,46,47). Reports also suggest the involvement of dentate gyrus adult-born neurons in behavioral tasks involving proper management of interferences (30,48,49). Here, we first strengthen this point in the successive spatial delayed non-matching to place task by examining performance of normal mice subjected to targeted X-ray irradiation of the dorsal dentate gyrus. Our results showed that mice with near complete absence of adult-born neurons in the dorsal hippocampus have profound deficits in the task characterized by a higher sensitivity to proactive interference. Although our experiment does not address the potential differential contribution of hippocampal adult born neurons in the dorsal and ventral parts of the DG in this task, the similarities between the behavioral phenotype of Pak3-R67C mice and that found after irradiation of the dorsal DG in WT mice reinforce our hypothesis that adult hippocampal neurogenesis might be one neurobiological mechanism altered in our transgenic mice.
PAK3 is expressed in hippocampus (24) and we confirmed here, by a genetic approach, its expression in granule cells of the adult DG, including in immature DCX-expressing neurons. Our analyses showed that, in basal conditions, the R67C mutation of Pak3 does not affect cell proliferation and neuronal differentiation of adult-born DGCs. In contrast, young immature newborn neurons in Pak3-R67C mice undergo accelerated death during the critical period, around 3-4 weeks of their birth, because of their high intrinsic excitability and high capacity to undergo synaptic potentiation (43), are activated by learning, promoting their recruitment and functional integration into memory networks, and hence their subsequent activation upon memory recall (39). We thus investigated the recruitment of DG neurons by measuring Zif268 expression after recall of spatial learning in the massed paradigm for which Pak3-R67C mice display normal learning but deficient long-term memory. We found memory recall resulted in normal activation of mature DG and 28-day-old BrdU + neurons in Pak3-R67C mice, but there was a complete failure of recruitment of a population of young DCX + newborn neurons, indicating that the long-term spatial memory deficit in these mice is associated with the absence of recruitment of this population of young newborn neurons. We therefore propose that the R67C mutation of Pak3 leads to a specific, transient loss of young, not fully mature DGCs during their critical period of maturation. This in turn would impede the recruitment of a population of these young newborn neurons by training, preventing their contribution to long-term spatial memory. Whether this alteration in adult neurogenesis and deficient activation upon learning is also causative for the altered capacity of the mutant mice to cope with interferences and for their deficient ability to finely discriminate events in the spatial domain remains to be investigated. Although there are many unknowns about brain alterations associated with the R67C mutation of the Pak3 gene, the documented contribution of adult hippocampal neurogenesis to spatial memory and spatial pattern separation function (7,27,28,29,46,47,53) reinforces the hypothesis that altered adult neurogenesis in Pak3-R67C mice might be one mechanism instrumental in the selective behavioral deficits displayed by the mice.
Mechanistically, how the R67C mutation of the Pak3 gene could lead to an altered adult hippocampal neurogenesis is as yet unknown. The R67C mutation is located in the regulatory domain of the protein, which changes PAK3 binding to the Rho-GTPase Cdc42 and Rac1, impairing PAK3 binding to Cdc42 while increasing binding to Rac1 (23). Interestingly, it was reported that Cdc42 and Rac1 are differentially regulated during adult hippocampal neurogenesis, suggesting they have stage-specific functions (54). Briefly, Cdc42 plays an important role in the proliferation of neural stem/progenitor cells, is enriched in DCX + immature neurons and its activity is associated with initial dendritic development and dendritic spine maturation. Rac1, in contrast, is mainly operational and crucial in late stages of dendritic outgrowth and spine maturation. Here, we found that the dysfunction of PAK3 does not affect the proliferation of adult born cells, suggesting that Cdc42-PAK3 binding is not required in this step during adult hippocampal neurogenesis. Concerning neuronal maturation, it has previously been shown in cultured hippocampal neurons that the R67C mutation affects spine density, suggesting altered spine growth, from initiation to stabilization of newly formed spines (23). Thus, because dendritic maturation of young adult-born neurons appeared altered by the R67C mutation, the recruitment failure of young DCX newborn neurons when learning occurs is likely due to a defect or a delay in spinogenesis of these young neurons, perhaps because Cdc42-PAK3 binding is reduced when PAK3 holds the R67C mutation. Future work will be necessary to confirm whether the R67C mutation of Pak3 alters the formation of dendritic spines of young adult hippocampal neurons in vivo, and what are the underlying molecular mechanisms. Finally, although we found that young DCX + neurons can express PAK3, we cannot discard a cell non-autonomous function of PAK3. Cell type-specific manipulation of mutated PAK3 using genetic approaches will be one step further for dissociating cellautonomous and non cell-autonomous contribution of PAK3. In summary, this study allowed us to find that the R67C mutation of the Pak3 gene in mice recapitulates some of the human ID phenotypes, especially in the spatial memory domain when the tasks require rapid learning and place a high demand on spatial discriminability and the management of interferences. We also highlighted that the R67C mutation causes dysfunctional maturation and learning-associated recruitment of a population of young adult-born hippocampal neurons during their critical period of integration into pre-existing cortico-hippocampal networks, a defect that represents one possible mechanism contributing to the observed cognitive impairments.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  (1:500, a11075, Molecular Probes). Sections were mounted using Fluoromount medium.

Animal breeding
All the mice were housed in a temperature and light-controlled colony room (12-h light/dark cycle; 21 ± 1°C, 25 ± 5% humidity) in groups of 2-6 per cage with food and water ad libitum. Experiments were conducted during the light phase and performed blind to the genotype. Experimental procedures were conducted in accordance with the guidelines established by the European Communities Council Directive (2010/63/EU Council Directive Decree) and the Animal Experimentation Ethical Committee (CEEA N°59, project N°1408). All efforts were made to reduce the number of mice used and to minimize their suffering.

Irradiation procedure
One group of 8-week old male C57BL/6 wild-type mice were irradiated and tested in the delayed nonmatching to place paradigm in the radial-arm maze (see below). Adult mice (n=10) were exposed to X-ray ionizing radiation using the Small Animal Radiation Research Platform (SARRP, Xstrahl, Ltd., UK) of IRSN (Fontenay-aux-Roses, France). The SARRP can deliver targeted radiation to preclinical animal models with high accuracy (34). Using Cone Beam Computed Tomography (CBCT) from SARRP and magnetic resonance images recorded in parallel and manually superimposed on the SARRP treatment planning system (MuriPlan), the dorsal dentate gyrus was specifically irradiated with the protocol described in (34). Briefly, mice were anaesthetized by i.p. injection of a ketamine (100 mg/kg) / xylazine (10 mg/kg) solution. To irradiate all the dorsal dentate gyrus of adult mice, three isocenters were positioned and, for each, two beams at 180° to the sagittal plane were delivered using a circular 1 mm irradiation field. A fractioned protocol was used to give a total dose of 15 Gy (3 x 5 Gy spaced by 1 day). Throughout the duration of the study, there were no noticeable side effects of irradiation in terms of weight loss or hair loss. Non-irradiated sham control mice (NIR, n=8) were transported to the irradiation facility, anaesthetized, and subjected to the realization of CBCT images, but were not irradiated.

Behavioral procedures
Adult (3-month-old) male C57BL/6 Pak3 hemizygous knock-in mice (Pak3-R67C, n=62) and Wild-Type (WT, n=78) littermate mice, distributed in 5 cohorts, were used in behavioral studies (supplementary table 1). Before the start of behavioral experiments, mice were handled 2 min per day for 3 days by the experimenter. For each behavioral task, a camera connected to a videotracking system (ANY-maze™, Stoelting) was placed above the devices to record behavioral sequences.

Water maze set-up and training procedures
The water maze consisted of a circular tank (150 cm in diameter, 37 cm high) filled with water (21°C ± 0.5°C) to 7 cm below the top of the sidewall, made opaque by adding a white, non-toxic paint (Acusol OP301 Opacifier; Rohm and Haas). The maze was placed in a room containing several different cues on the walls. The pool was divided into four virtual quadrants. A circular escape platform (11.5 cm in diameter placed at 40 cm from the wall), submerged 0.5-cm below the water surface, was located in the center of one quadrant (target quadrant) and remained at a fixed location for each mouse during training. The three other quadrants served as starting points, assigned pseudorandomly and varied on each trial. The maze was placed in a well-lit room (380 lx) containing several extra-maze cues. Data recorded by video-tracking (ANY-maze™) were used to reconstruct swim paths and to calculate averaged swim speed, swim path lengths and time spent in various virtual areas of the maze: the four quadrants, the four platform annuli (48-cm in diameter around the platform position) and a virtual corridor (19 cm in width) set along the wall to quantify thigmotaxis. Memory retention was evaluated during a probe test with the platform removed from pool. We compared the percent time spent in the quadrant that previously contained the platform to chance level (25%) and the number of crossings over the platform site in the quadrant that contained the platform during training.

Distributed protocol. One month after the open-field exploration test, a group of Pak3-R67C (n=15)
and littermate WT mice (n=12) was submitted to a distributed spatial learning task in the water maze.
One day before training, the mice underwent four trials of habituation during which they were trained to find and stay on the escape platform, placed in the center of the tank, for 60 s. Following habituation, the training phase lasted 11 days and consisted of 4 successive trials during which the platform was always located in the same quadrant for a given mouse. The position of the hidden platform was assigned for each mouse in one of the four virtual quadrants of the maze, such that the four positions were equally used in both groups of mice. Mice were introduced into the water maze from three different starting points in the quadrants that did not contained the platform and were allowed to swim freely until reaching the platform. The sequence of starting points was chosen in a pseudorandom order and counterbalanced among individuals. Mice failing to find the platform after 90 s were gently guided to it by hand and a maximum escape latency of 90 s was recorded. Mice remained 60 s on the platform before the start of the next trial. After this training phase, a first probe test was performed 48h after the last training session and consisted of a single 90 s trial without platform. It was followed 1h later by one single trial in presence of the platform, to minimize extinction effects caused by the first probe trial. Another probe trial without platform was performed one week after the last training session.

Massed protocol.
Another cohort of Pak3-R67C mice (n=6) and littermate WT mice (n=7) aged 3.5 months was used. For habituation, one day before training, the mice were submitted to a single habituation block of three trials, with the visible platform (protruding 0.5 cm above the water surface).
A trial started by introducing the mouse into the maze facing the wall at one of the three designated starting points and allowed to swim freely until it reached the platform with a maximum time of 60 s.
The mouse remained 60 s on the platform before being replaced at another starting point for the next trial. The next day, mice were trained to locate the hidden platform. The massed-training procedure (28) consisted in three training sessions separated by 2h, each composed of three blocks, 25-min apart, of three consecutive trials (27 trials in total). At each trial, a mouse was introduced into the maze from any of the three different starting points and allowed to swim freely until it reached the platform. If a mouse failed to find the platform after 60 s, it was gently guided to it and allowed to stay on it for 60 s before the start of the next trial. Ten days after learning, 2 successive 60 s probe tests, 3 min apart, are carried out.
On the first session, a group of littermate mice was allowed to move freely in the maze and collect rewards during 5 min. On the second session, mice were individually placed in the maze for 5 min to retrieve rewards from the baited arm wells. The maze is cleaned with absolute alcohol between each mouse.
Working memory. In this elimination task, mice (WT n=10, Pak3-R67C n=11) had to recover rewards located at the end of the 8 arms of the radial maze without returning to a previously visited arm. Mice were submitted to 1 trial per day for 12 days. At the beginning of a trial, the mouse was placed in the center of the device surrounded by a transparent cylinder. All doors were open. The trial begins when the cylinder is removed. The maximum time for a trial is 5 min. If all rewards are recovered before the end of 5 minutes, the mouse is removed from the device. The number of working memory errors, consisting in entries into an already visited arm during a given trial, was counted.
Delayed non-matching to place paradigm in the radial-arm maze. The delayed non-matching to place (DNMP) paradigm was based on (29) and adapted in (7). During this task, WT (n=18) and Pak3-R67C (n=17) mice, aged 3.5 months, were tested. During training for 10 consecutive days, mice were submitted to 4 trials/day separated by 20 min. Each trial included two phases: a sample phase followed by a choice phase 20 s later (Fig. 3A). In the sample phase, only the start arm and one baited arm were opened. Once the reward was recovered, the mouse was removed from the maze, the maze was cleaned with alcohol and the choice phase started during which one additional arm was opened and was the only arm rewarded. This third arm varied in distance from the sample arm previously rewarded by either one arm (close configuration) or three arms (distant configuration; Fig. 3A). The  Pak3-R67C n=8).
The percentage of corrects choices was analyzed during training and test.

Successive spatial delayed non-matching to place in a radial-arm maze.
To test memory flexibility and organization, and susceptibility to proactive interference, we used the task modeling declarative memory of repetitive everyday events developed by Abed et al. (33). This hippocampal-dependent task in the radial-arm maze requires the ability to retain recently visited places, i.e. which arm was the more recently visited within each of 3 repeatedly used pairs of arms, over varying intervals and intervening arm pairs. The task also requires the ability to organize and update the stored events as they are used, to overcome proactive interference from previous arm visits, and taxes both the retention and organizational components of hippocampus-dependent memory (33). We separately  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  P  e  e  r  R  e  v  i  e  w 29 an alternation rule, a natural mouse behavior. Thus, the reward was always positioned in the arm that was not visited by the mouse during the previous trial with the same pair (Fig. 4B). At the beginning of each daily session, the mouse was placed in the central platform with all arms closed. In the first trial with each pair, at the beginning of a training session, both arms contained a reward. Thus, this trial was just a "sampling" trial, not a "testing" trial, but then, the alternation rule applied and each trial becomes both a "testing" trial and a "sampling" trial. After 10 s, the mouse was given access to one of the 3 pairs of arms (a, b or c) and asked to choose to visit one of the 2 arms. The door of the non-chosen arm was closed as soon as the mouse reached the end of the chosen arm. As the mouse came back in the center, the door of the chosen arm was closed. After a fixed inter-trial interval of 10 s spent in the center, another trial began with 2 opened arms (same pair, or any of the other 2 pairs) and so on. Each daily session consisted of 23 trials with a 10 s inter-trial interval (ITI). Performance was measured as the percentage of correct trials over training and, for the last 3 days of training (days 13,14,15), as a function of the number of interfering trials (ITn) that corresponds to the number of trials with different pairs between 2 trials of the same pair. ITn can therefore vary from 0 (2 consecutive trials with the same pair) to 4 (4 interfering trials with different pairs between 2 trials from the same pair). The percentage of correct trials as a function of proactive interference (PI) was also calculated for the last 3 days of training. PI was dependent on the number of interfering trials between trial n and trial n-1 of the same pair. If the ITn between n-1 and n is small, the interference generated for the trial n+1 is more important and vice versa. Two categories of PI were analyzed, "low PI" for an ITn of 2 to 4 between trial n-1 and trial n, and "high PI" when the ITn is between 0 and 1 (Fig. 4B).

BrdU administration
To study adult newborn cell survival in basal conditions, Pak3-R67C and WT mice were given 3

Double-and triple-labeling analysis and quantification
Sections were analyzed using the Zeiss confocal system (LSM700). Zif268 + /NeuN + cells in each stack was counted manually using Image J software (cells counters plugin) and summed per animal. In total, co-localizations were evaluated on 3800 NeuN + cells per group of mice.

Dendritic morphology analysis of DCX + cells
For morphological analysis of young neurons, we selected DCX + cells from infra-and suprapyramidal blades of the dorsal DG, with vertically orientated dendrites, extending through the molecular layer, and with, at least, tertiary dendritic branches (57, 58). Confocal images were acquired using a Zeiss LSM700 microscope with a X40 objective. For dendritic analysis, 3D reconstructions of the dendritic processes of each DCX + cell were made from sequential Z-stack images that were taken with 1 µm optical Z sections. The dendritic length and number of intersections were measured using the FIJI  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  P  e  e  r  R  e  v  i  e  w   33 image analysis software. Five mice from each genotype were analyzed with eight neurons from each mouse (n=40 from 5 mice per genotype).