Wnt5a is essential for hippocampal dendritic maintenance and spatial learning and memory in adult mice

Significance

The maintenance of neuronal morphology in the adult brain is an understudied area. Here, using tissue-specific deletion in mice, we reveal Wnt5a, a member of the Wnt family of developmental morphogens, as an essential factor for the long-term stability of dendritic architecture in the adult hippocampus. Wnt5a influences synaptic plasticity and related cognitive functions in the mature hippocampus through CaMKII-mediated signaling, Rac1-dependent actin dynamics, and cyclic AMP-responsive element binding-mediated NMDA receptor biosynthesis. In the long-term, Wnt5a-mediated regulation of cytoskeletal signaling and excitatory synaptic transmission is responsible for the maintenance of dendritic arbors and spines in adult CA1 pyramidal neurons. These findings provide insight into the poorly understood structural maintenance mechanisms that exist in the adult brain.

Abstract

Stability of neuronal connectivity is critical for brain functions, and morphological perturbations are associated with neurodegenerative disorders. However, how neuronal morphology is maintained in the adult brain remains poorly understood. Here, we identify Wnt5a, a member of the Wnt family of secreted morphogens, as an essential factor in maintaining dendritic architecture in the adult hippocampus and for related cognitive functions in mice. Wnt5a expression in hippocampal neurons begins postnatally, and its deletion attenuated CaMKII and Rac1 activity, reduced GluN1 glutamate receptor expression, and impaired synaptic plasticity and spatial learning and memory in 3-mo-old mice. With increased age, Wnt5a loss caused progressive attrition of dendrite arbors and spines in Cornu Ammonis (CA)1 pyramidal neurons and exacerbated behavioral defects. Wnt5a functions cell-autonomously to maintain CA1 dendrites, and exogenous Wnt5a expression corrected structural anomalies even at late-adult stages. These findings reveal a maintenance factor in the adult brain, and highlight a trophic pathway that can be targeted to ameliorate dendrite loss in pathological conditions.

Long-term structural maintenance of neuronal networks is essential for sustaining brain functions. The size and pattern of dendrite arbors dictate the ability of neurons to receive and integrate synaptic inputs and are thus critical determinants of information processing in the brain. Established by periods of dynamic growth during development, dendrite arbors and spines are thought to be largely stable in adulthood and are maintained for the lifetime of an organism (1). The significance of this maintenance phase for normal brain functions is underscored by evidence that late-onset retraction of dendritic arbors and spine loss are the most consistent morphological correlates of several neurological and psychiatric disorders, including schizophrenia, major depressive disorder, anxiety, and Alzheimer’s disease (2–5). Thus, specific molecular signals must exist to ensure the maintenance of neuronal morphology and synaptic connectivity in the adult nervous system.

To date, understanding of the molecular cues that maintain adult dendritic patterns has been limited, and largely originates from studies of developmental signals that are deployed later during postnatal life to regulate maturation or stability (1, 6). Brain-derived neurotrophic factor (BDNF) provides an example of an extrinsic signal that is required throughout life for both the establishment and maintenance of neuronal connectivity. Conditional deletion of either BDNF or its receptor, TrkB, leads to a reduction in dendritic complexity of adult cortical neurons (7, 8). Intriguingly, deletion of BDNF or TrkB does not affect dendrite architecture in the adult hippocampus, a brain structure critical for spatial learning and memory and anxiety (7, 9–13). Currently, little is known about dendrite support mechanisms in adult hippocampal neurons. Cornu Ammonis (CA)1 pyramidal neurons in the hippocampus are particularly vulnerable in Alzheimer’s disease and exhibit extensive dendrite arbor loss that correlates with the degree of cognitive decline (14). Identification of adult maintenance mechanisms would be highly relevant to the understanding of the structural basis of hippocampus-dependent behaviors, as well as the etiology of neurodegenerative diseases where extensive dendritic anomalies are manifested late in life.

Here, using neuron-specific deletion in mice, we identify Wnt5a, a member of the Wnt family of developmental morphogens, as an essential factor for the long-term stability of dendritic architecture in the adult hippocampus. Previous studies have implicated Wnt5a in regulating developmental axon and dendrite outgrowth and synapse formation in cultured hippocampal neurons (15–20). Here, we report that Wnt5a deletion does not compromise hippocampal development or maturation in vivo, but results in striking adult-onset defects in dendrite arborization, lengths, and spine densities in CA1 hippocampal pyramidal neurons that manifest after 4 mo of age in mice. Wnt5a is required cell-autonomously in adult neurons to maintain dendritic architecture. Loss of Wnt5a impairs hippocampal synaptic plasticity and spatial learning and memory in adult mice before the onset of dendritic regression, although the behavioral deficits are exacerbated with the appearance of structural abnormalities. Wnt5a acts via calcium- and cytoskeletal-mediated signaling in the adult hippocampus, and unexpectedly, via cyclic AMP-responsive element binding (CREB)-mediated transcription of the obligatory NMDA receptor subunit, GluN1. Finally, we demonstrate that late expression of Wnt5a, even after substantial structural loss, fully restores neuronal morphology, highlighting the growth-promoting capacity of this pathway in the adult brain. Together, these findings reveal that noncanonical autocrine Wnt signaling maintains adult hippocampal connectivity and synaptic plasticity, and provide a trophic pathway that can be targeted to counter structural deficits in pathological situations.

Results

Cell-Autonomous Requirement for Wnt5a in Maintaining Adult CA1 Dendrite Arbors.

We observed Wnt5a expression in the mouse hippocampus at 1 wk after birth, which increased prominently by 2 wk and was sustained at adult stages (Fig. 1 A and B). Wnt5a mRNA was localized throughout the hippocampal formation, and was enriched in the dentate gyrus and the CA1 region. The expression pattern in the hippocampus is consistent with previous findings (21) and that in the Allen Brain Atlas (www.brain-map.org/), where Wnt5a expression has also been noted in the cerebellum and to a lesser degree in the cerebral cortex and olfactory bulb in the adult mouse brain. The onset of Wnt5a expression in hippocampal neurons correlated with the appearance of pre- and postsynaptic proteins (Fig. 1B).

Fig. 1.

Wnt5a is required cell-autonomously for maintenance of adult CA1 dendrite architecture. (A) In situ hybridization shows Wnt5a transcript in the mouse hippocampus at postnatal stages. Wnt5a is absent at the day of birth (P0) but is detected by 1 wk and is abundant in the 3-mo-old hippocampus. (Scale bar, 300 μm.) (B) Onset of Wnt5a protein expression coincides with the appearance of pre- and postsynaptic proteins in hippocampal neurons. Hippocampal homogenates from mice of different ages (ranging from E18 to 6 mo) were immunoblotted with antibodies against Wnt5a, VGLUT1, PSD95, GluA1, GluN1, CaMKIIα, CaMKIIβ, and tubulin. (C) Schematic of the strategy to assess effects of postnatal Wnt5a deletion on neuronal morphologies in adult mice at different ages using CaMKIIα-Cre transgenic mice. (D–G) Representative images of GFP⁺ CA1 pyramidal neurons show that dendritic arbors are normal at 3 mo, but become progressively stunted with age in Thy1-GFP;CaMKII-Wnt5a^fl/fl (KO) mice compared with control Thy1-GFP;Wnt5a^fl/fl mice (WT). (Scale bar, 100 μm.) Shown below are 3D reconstructions of neuronal soma and dendrites from WT and KO mice at 3, 4.5, 6, and 12 mo using Imaris. (H–K) Sholl analyses show that dendritic complexity is unaffected in 3-mo-old KO mice but significantly decreased by 4.5 mo and declines in older KO mice. Results are mean ± SEM from five neurons traced per animal and a total of five mice per genotype. Two-way ANOVA with Bonferroni post hoc test. (L and M) Total dendrite length and number of dendritic branch points are unaltered in 3-mo-old KO mice, but progressively decreases from 4.5 mo to 12 mo in older KO mice. Results are mean ± SEM from five neurons traced per animal and a total of five mice per genotype. **P < 0.01 two-tailed t test. (N) Wnt5a^fl/fl mice were stereotaxically injected with lentiviral vectors expressing GFP-T2A-CRE or GFP at 2 mo and dendritic morphologies analyzed after 3 mo. (O) Dendritic arbors were stunted in Cre-expressing neurons compared with GFP-expressing neurons. (Scale bar, 100 μm.) (P) Sholl analysis shows decreased dendrite arbor complexity in sparsely labeled Cre-expressing neurons compared with controls. Two-way ANOVA with Bonferroni post hoc test. (Q and R) Dendritic lengths and branch points are reduced in Cre-expressing neurons compared with controls. Results are mean ± SEM from 14 GFP-T2A-CRE- and 10 GFP-infected neurons from a total of 10 Wnt5a^fl/fl mice. ***P < 0.001, two-tailed t test.

That prominent hippocampal Wnt5a expression is detected only at postnatal stages was intriguing, given the reported roles of Wnt5a in embryonic processes in the brain (15, 22–25). Wnt5a expression in the postnatal hippocampus, together with evidence supporting the role of Wnts in regulating morphological changes in cultured hippocampal neurons (26, 27), prompted us to address the functions of Wnt5a in hippocampal neurons in vivo. To accomplish this end, we crossed floxed Wnt5a (Wnt5a^fl/fl) mice with the pan-neuronal Nestin-Cre line (28) to delete Wnt5a in all neurons starting at embryonic stages (Fig. S1A). Wnt5a was undetectable in the Nestin-Wnt5a^fl/fl hippocampus throughout postnatal ages, including postnatal day 21 (P21) by in situ hybridization (Fig. S1B). Importantly, quantitative PCR (qPCR) analysis showed that levels of other Wnts were unaltered in the absence of Wnt5a (Fig. S1 C and D), indicating that Wnt5a loss did not elicit compensatory changes in the expression of other Wnt genes in the hippocampus.

Fig. S1.

Wnt5a is dispensable for hippocampal development. (A) Schematic of the strategy for assessing effects of embryonic Wnt5a deletion on neuronal morphology, using Nestin-Cre transgentic mice. (B) Wnt5a transcript is undetectable in the P21 Nestin-Wnt5a^fl/fl hippocampus. (Scale bar, 500 μm.) (C and D) qPCR analyses shows substantial depletion of Wnt5a mRNA in Nestin-Wnt5a^fl/fl hippocampus, whereas other Wnts are unaltered. Results are means ± SEM from n = 7 mice per genotype for 1 and 2 wk, and n = 8 mice per genotype for 4 wk, and from six independent experiments for D. ***P < 0.001, two-tailed t test. (E) Hippocampal cyto-architecture, axonal projections, and CA1 dendritic layers are unaffected in Nestin-Wnt5a^fl/fl mice at 1 mo, revealed by immunostaining with antibodies against NeuN, neurofilament, and MAP2. Lower panels indicate higher magnification images of the boxed regions. [Scale bars, 400 μm (Upper) and 200 μm (Lower).] (F) Golgi staining and dendritic reconstructions show comparable arbor sizes in CA1 pyramidal Nestin-Wnt5a^fl/fl and control neurons at 1 mo. (Scale bar, 100 μm.) (G and H) Quantification of dendritic lengths and dendrite complexity, using Sholl analysis, in 1-mo-old Nestin-Wnt5a^fl/fl mice and Wnt5a^fl/fl littermate controls. Results are mean ± SEM from five neurons traced per animal and a total of four mice per genotype, two-tailed t test. (I and J) Spine densities in CA1 apical dendrites are normal in Nestin-Wnt5a^fl/fl mice at 1 mo. (Scale bar, 5 μm.) Results are mean ± SEM from five mice per genotype.

Despite the early deletion of Wnt5a (Fig. S1 A–C), there were no obvious differences in gross morphology and projections of hippocampal layers between Nestin-Wnt5a^fl/fl mice and control Wnt5a^fl/fl littermates at 1 mo (Fig. S1E), when hippocampal neural circuit establishment should be complete (29–31). Detailed examination of dendritic morphology of CA1 hippocampal pyramidal neurons revealed normal dendrite length, complexity, and spine densities in 1-mo-old mutant mice (Fig. S1 F–J). These results show that Wnt5a is dispensable for establishing dendritic arbors and for spine formation in CA1 pyramidal neurons in vivo, contrary to published reports that Wnt5a promotes neuronal morphogenesis in cultured hippocampal neurons (15, 18, 19, 32).

To next address if Wnt5a might function in the mature hippocampus, we crossed Wnt5a^fl/fl mice with calcium-calmodulin kinase II (CaMKII)α-Cre mice, where Cre expression starts at 2.5 wk after birth and is restricted to forebrain excitatory neurons in the hippocampus and cortex (33). Wnt5a deletion was near complete in the CaMKII-Wnt5a^fl/fl hippocampus by 3 mo of age (Fig. S2 A and B), whereas other Wnts showed normal expression (Fig. S2C). Surprisingly, we observed a decrease in the thickness of the CA1 dendritic layers in 6-mo-old CaMKII-Wnt5a^fl/fl mice using MAP2 immunostaining, although hippocampal cyto-architecture and axonal projections were unaffected (Fig. S2D). To better visualize the morphologies of individual CA1 neurons and their processes and to pinpoint the onset of dendritic defects triggered by the postnatal loss of Wnt5a, we crossed CaMKII-Wnt5a^fl/fl mice with Thy1-GFP-M transgenic mice that have mosaic GFP expression in the hippocampus (34), and analyzed neuronal structure at different ages from 3 to 12 mo (Fig. 1C). Sparsely labeled CA1 pyramidal neurons in Thy1-GFP;CaMKII-Wnt5a^fl/fl mice had normal dendrite arbor complexities and lengths at 3 mo of age (Fig. 1 D, H, L, and M). However, by 4.5 mo, Thy1-GFP;CaMKII-Wnt5a^fl/fl neurons showed striking dendritic deficits that progressively declined in older animals (Fig. 1 E–G, and I–M). Based on Sholl analyses, distal dendrites at distances 50 μm and farther from the soma were more severely affected than proximal dendrites (Fig. 1 I–K). In control mice, the total dendritic length remained remarkably stable between 3 and 12 mo of age (4,866 ± 175 μm at 3 mo vs. 4,978 ± 162.7 μm at 12 mo) (Fig. 1L). In contrast, CaMKII-Wnt5a^fl/fl mice exhibited a pronounced decrease in dendritic length between 3 and 12 mo of age (4,780 ± 132.4 μm at 3 mo vs. 2,810 ± 62.36 μm at 12 mo) (Fig. 1L). Quantification of dendritic spine densities also revealed a significant reduction (31.5% decrease) in 6-mo-old CaMKII-Wnt5a^fl/fl mice compared with controls (Fig. S2 E and F). Despite the profound dendritic shrinkage in adult CaMKII-Wnt5a^fl/fl mice, there was no overt loss of neurons in these animals even at 12 mo (Fig. S2 G and H). Together, these results indicate a requirement for Wnt5a in the maintenance of dendrite arbors and spine densities in adult CA1 neurons.

Fig. S2.

Wnt5a deletion results in spine loss but neuronal viability is intact in adult hippocampus. (A and B) Wnt5a loss is complete by 3 mo in CaMKII-Wnt5a^fl/fl (KO) mice as shown by in situ hybridization and qPCR analyses. (Scale bar, 400 μm.) For qPCR analysis, results are means ± SEM from n = 4 mice per genotype for 1 and 2 mo, and n = 5 mice per genotype for 3 and 6 mo; ***P < 0.001, two-tailed t test. (C) Other hippocampal Wnts transcripts are unaltered in 3-mo-old KO mice. Results are mean ± SEM from six independent experiments. (D) Hippocampal cyto-architecture and axonal projections are normal, but CA1 dendritic layers are reduced in 6-mo KO mice. Lower panels indicate higher magnification images of the boxed regions. [Scale bar, 400 μm (Upper) and 200 μm (Lower).] (E and F) Spine densities are significantly reduced by 6 mo, but normal at 3 mo in KO mice. (Scale bar, 5 μm.) Results are means ± SEM from five mice per genotype at 3 mo, and n = 7 for KO mice and n = 5 for Wnt5a^fl/fl (WT) mice at 6 mo. **P < 0.01, two-tailed t test. (G and H) Nissl staining and cell counts in CA1, CA3, and dentate gyrus layers reveal similar cell densities between KO and WT mice at 6 and 12 mo. (Scale bar, 500 μm.) Results are mean ± SEM from three mice per genotype.

Wnts are known to act either as autocrine (cell-autonomous) or paracrine (noncell-autonomous) secreted factors. Because Wnt5a is deleted from all excitatory hippocampal neurons in CaMKII-Wnt5a^fl/fl mice, we are unable to distinguish between these two modes of signaling using the conditional mutants. To determine whether Wnt5a has a cell-autonomous or noncell-autonomous role in the hippocampus, we performed mosaic analyses by lentiviral delivery of GFP-T2A-Cre or GFP alone into subsets of hippocampal neurons in Wnt5a^fl/fl animals. Viral infections were done at 2 mo and mice harvested 3 mo later for morphological analyses (Fig. 1N). We found stunted dendritic arbors in sparse GFP-labeled CA1 neurons with significant reductions in branch complexity and length in Wnt5a^fl/fl animals infected with GFP-T2A-Cre but not GFP alone (Fig. 1 O–R). Thus, mosaic Wnt5a elimination causes poorly branched and diminished dendritic arbors in isolated Cre-expressing neurons despite the presence of non-Cre–expressing neighboring neurons that are capable of releasing Wnt5a. These results suggest that Wnt5a secreted from adult CA1 neurons maintains neuronal morphology in an autocrine manner, likely because of limited diffusibility attributed to the lipid-modified and hydrophobic nature of Wnts (35).

Wnt5a Is Essential for Hippocampal Synaptic Plasticity.

The dramatic changes in adult neuronal morphology with Wnt5a loss prompted us to ask whether Wnt5a is essential for hippocampal synaptic transmission in vivo. We performed electrophysiological recordings in CaMKII-Wnt5a^fl/fl mice at 3 mo of age, a time when CA1 neuronal morphology is still intact, and at 6 mo when the morphology is impaired. We observed that basal synaptic transmission is normal in CaMKII-Wnt5a^fl/fl mice at 3 mo but not at 6 mo, consistent with impaired dendritic structures at this age (Fig. 2 A–D). We also measured the probability of synaptic release at presynaptic sites by paired pulse facilitation (PPF) analyses and found that PPF was comparable between CaMKII-Wnt5a^fl/fl mice and control littermates at both 3 and 6 mo of age (Fig. S3). The normal presynaptic properties in Wnt5a-deficient mice suggests that Wnt5a acts primarily at postsynaptic sites.

Fig. 2.

Wnt5a loss impairs LTP in the absence of changes in basal synaptic transmission or neuronal morphology. (A–D) Basal synaptic transmission is unaffected in CaMKII-Wnt5a^fl/fl (KO) mice at 3 mo but impaired at 6 mo compared with Wnt5a^fl/fl litter-mates (WT). The input–output relationships and slopes of input–output curves are normal in 3-mo-old KO mice, but significantly different from WT at 6 mo. Results are mean ± SEM from five animals per genotype (n = 18 slices for WT and 17 slices for KO mice) at 3 mo, and from five WT (n = 21 slices) and four KO mice (n = 23 slices) at 6 mo. **P < 0.01, two-tailed t test. [Scale bars, 1 mV (vertical), 2.5 ms (horizontal) for all sample traces.] (E) Impaired LTP at the Schaffer collateral–CA1 synapses in 3-mo-old KO mice. n = 11 slices from 5 WT mice and n = 13 slices from 6 KO mice. Sample traces represent fEPSPs (field excitatory postsynaptic potentials) right before (dashed line) and 1 h after (solid line) θ-burst stimulation. [Scale bars, 1 mV (vertical), 10 ms (horizontal) for both sample traces.] (F and G) Induction and maintenance of LTP are attenuated in KO mice. Results are mean ± SEM from five WT mice (n = 11 slices) and six KO mice (n = 13 slices). *P < 0.05, ***P < 0.001, two-tailed t test. (H) Low-frequency stimulation-induced LTD at Schaffer collateral–CA1 synapses is unaltered in 3-mo CaMKII-Wnt5a^fl/fl mice. Sample traces represent fEPSPs right after (dashed line) and 1hr after (solid line) stimulation. [Scale bars, 1 mV (vertical), 10 ms (horizontal) for both sample traces.] (I) Mean fEPSP slopes are comparable between the genotypes. Results are mean ± SEM from five Wnt5a^fl/fl mice (n = 14 slices) and five CaMKII-Wnt5a^fl/fl mice (n = 16 slices).

Fig. S3.

Wnt5a deletion does not affect presynaptic properties. (A and B) Presynaptic vesicle release is normal in 3- and 6-mo-old CaMKII-Wnt5a^fl/fl (KO) mice as measured by paired-pulse ratios with different interstimulus interval (ms). Results are means ± SEM from four animals per genotype for 3-mo-old animals [n = 17 slices for Wnt5a^fl/fl (WT) and n = 15 slices from KO mice]. For 6-mo-old animals, four WT mice (n = 24 slices) and three KO mice (n = 22 slices) were used. [Scale bars, sample traces in Insets; 1 mV (vertical), 10 ms (horizontal).] Statistical analyses were done using a two-way ANOVA.

Hippocampal neurons exhibit prominent synaptic plasticity in which activity-dependent modulation of the strength of synaptic connections underlies learning and memory (36). Previously, broad-spectrum Wnt antagonists have been shown to affect synaptic structure, plasticity, and cognitive functions in adult organisms (37–41). However, there are 19 vertebrate Wnts, and which Wnt is essential for these functions in the adult brain in vivo remains unknown. To assess the role of Wnt5a in NMDA receptor-dependent long-term potentiation (LTP), an electrophysiological correlate of strengthening of synaptic transmission, we used θ-burst stimulation to induce LTP at Schaffer collateral–CA1 synapses in 3-mo-old mice. Recordings from Wnt5a^fl/fl control slices revealed a robust induction of LTP and a sustained maintenance phase (Fig. 2 E–G). In contrast, CaMKII-Wnt5a^fl/fl slices showed a significant reduction in both induction (274.3 ± 18.2% in control slices vs.. 201 ± 7.9% in CaMKII-Wnt5a^fl/fl slices, P = 0.0002) and maintenance (190.8 ± 13.3% in control slices vs. 151.9 ± 5.4% mutant slices, P = 0.01) phases of LTP (Fig. 2 E–G). The impairment in LTP detected at 3 mo in CaMKII-Wnt5a^fl/fl mice when neuronal structure and basal synaptic transmission are still intact, suggests that synaptic plasticity is more susceptible to the loss of Wnt5a.

To address if Wnt5a contributes to NMDA receptor-dependent long-term depression (LTD), an electrophysiological correlate of weakening of synaptic transmission, we used a standard low-frequency stimulation paradigm to induce LTD in the CA1 hippocampus. In contrast to the LTP defect, we found no differences in LTD between mutant and control mice at 3 mo of age (Fig. 2 H and I). Taken together, these results indicate a specific role for Wnt5a in the potentiation of synaptic efficacy.

Wnt5a Is Essential for Spatial Learning and Memory.

Synaptic plasticity is widely considered to be a cellular mechanism that underlies learning and memory (42, 43). In addition, structural maintenance of synaptic connectivity has been postulated to be critical for life-long memories (44). Given the decreased CA1 LTP in 3-mo-old CaMKII-Wnt5a^fl/fl mice, we subjected Wnt5a mutant mice to behavioral paradigms to evaluate cognitive functions. In the novel-object recognition test, which evaluates the preference of mice to explore a new over a familiar object (45), adult Wnt5a^fl/fl mice showed a significant preference for the novel object, with 3- and 6-mo-old mice spending 70.8 ± 5.5% and 68.1 ± 8.1% of time exploring the novel object, respectively. However, CaMKII-Wnt5a^fl/fl mice showed no preference for the novel object at both 3 and 6 mo of age (Fig. 3 A and B), indicating a deficit in recognition memory.

Fig. 3.

Adult Wnt5a mutant mice show a progressive decline in hippocampus-mediated behaviors. (A and B) WT mice show a significant preference for exploring a new object in the novel-object recognition task, whereas Wnt5a KO mice spent similar amounts of time with familiar and new objects. Deficits in recognition memory were evident in both 3- and 6-mo-old mutants. Dashed line indicates equal amount of time spent exploring new and familiar objects. Results are mean ± SEM from n = 21 WT mice and n = 19 KO mice at 3 mo, and n = 16 WT mice and n = 23 KO mice at 6 mo. *P < 0.05, **P < 0.01 significantly different from 50% time spent with the novel-object, two-tailed t test. (C) Timeline for the Morris water maze (MWM) tasks. (D) Three-month-old KO mice show impaired spatial learning in the Morris water maze test. Results are mean ± SEM from n = 20 WT and n = 19 KO mice. *P < 0.05, **P < 0.01, two-way ANOVA with Fisher’s least-significant difference (LSD) post hoc test. (E) Six-month-old KO mice exhibit more severe deficits in learning and fail to acquire the latency of control animals even on day 12. Results are mean ± SEM from n = 18 WT and n = 20 KO mice. *P < 0.05, **P < 0.01, ***P < 0.001, two-way ANOVA with Fisher’s LSD post hoc test. (F and G) During probe trials, 3- and 6-mo-old KO mice showed less preference for the target quadrant compared with WT mice. Results are mean ± SEM, *P < 0.05, **P < 0.01, two-tailed t test. (H and I) During the memory retention test, both 3- and 6-mo-old KO mice spent less time in the target quadrant compared with control littermates. However, older mutant mice exhibited a marked decay in memory retrieval by the fifth day of the probe trial. White dashed line indicates 25% of time that mice spent in the target quadrant. Results are mean ± SEM; *P < 0.05, **P < 0.01, ***P < 0.001 significantly different from control littermates; ^#P < 0.05 significantly different from 25% line, two-tailed t test.

To test hippocampus-dependent spatial learning in these mice, we used the Morris water maze to test an animal’s ability to use spatial cues to locate a hidden platform in a tank of water (46) (Fig. S4A). Importantly, CaMKII-Wnt5a^fl/fl mice had visual acuity and swimming speeds comparable to littermate controls (Fig. S4 B and C). Three-month-old CaMKII-Wnt5a^fl/fl mice took significantly longer time to locate the hidden platform during the initial training period of 12 consecutive days using four trials per day (see schematic in Fig. 3C), compared with age-matched controls, eventually achieving similar latencies as control mice on the sixth day (Fig. 3D). Six-month-old CaMKII-Wnt5a^fl/fl mice also required significantly more time to find the platform. However, 6-mo-old mutants did not achieve similar latencies as control mice, even when tested on the 12th day (Fig. 3E). These results show that acquisition of spatial learning is impaired in the absence of structural deficits in 3-mo-old Wnt5a mutant mice, but that learning deficits are exacerbated with the appearance of morphological abnormalities in older animals.

Fig. S4.

Behavioral deficits in adult Wnt5a mutant mice. (A) Schematic of the Morris water maze. (B and C) Swimming speeds and visual acuities are similar between CaMKII-Wnt5a^fl/fl mice (KO) and control Wnt5a^fl/fl litter-mates (WT) at both 3 and 6 mo of age. Results are means ± SEM from 20 WT, 19 KO mice at 3 mo of age, and 17 WT, 20 KO mice at 6 mo. (D and E) Three- and 6-mo-old KO mice exhibit fewer crossings over the original platform location during the probe trial compared with control littermates. Results are means ± SEM from n = 20 WT and n = 19 KO mice at 3 mo, and from n = 18 WT and n = 20 KO mice at 6 mo. *P < 0.05, two-tailed t test. (F and G) Both 3- and 6-mo-old KO mice showed deficits in reversal learning compared with WT mice, although the deficits were more pronounced in older mutant animals. *P < 0.05, **P < 0.01, ***P < 0.001, two-way ANOVA with Fisher’s LSD post hoc test.

To evaluate reference memory, we conducted a probe trial on day 13 in which the platform was removed and measured the amount of time that mice spent in the original target quadrant (Fig. 3C). Both 3- and 6-mo-old mutant mice spent significantly less time in the target quadrant and made fewer platform crossings (Fig. 3 F and G and Fig. S4 D and E). We then performed a reversal training by relocating the hidden platform to the opposite quadrant from days 14–25 (Fig. 3C). As in the intial training phase, 3-mo-old Wnt5a mutant mice required more days of training to find the new platform, but eventually reached similar latencies as control mice (Fig. S4F). However, 6-mo-old mutant mice took significantly longer to find the platform compared with control littermates even on day 25 (Fig. S4G). We then conducted probe trials for 7 d (days 26–32) to assess memory retention. Three-month-old mutants maintained a preference for the target quadrant for the 7 d of the probe trial (Fig. 3H), whereas this preference was lost in 6-mo-old mutants by the fifth day (Fig. 3I), suggesting a marked decay in memory retrieval in older mutant animals.

Taken together, the findings from the Morris water maze test support an essential role for Wnt5a in the acquisition of spatial learning and memory storage in adult animals. Notably, the cognitive dysfunction in 3-mo-old CaMKII-Wnt5a^fl/fl mice were consistent with the LTP defects observed at this age but appeared before the onset of anatomical impairments. The more pronounced behavioral defects in 6-mo-old CaMKII-Wnt5a^fl/fl mice suggest a progressive decline in cognitive functions with the manifestation of dendritic abnormalities.

Wnt5a Loss Disrupts Calcium and Cytoskeletal Signaling Pathways and CREB-Mediated Transcription of Glutamate Receptors.

Wnts are known to exert their effects by signaling through three effector pathways: the canonical β-catenin–dependent pathway, a Ca²⁺-dependent pathway, and the planar cell polarity pathway (47). We found comparable levels of nuclear β-catenin and Axin2, c-myc, and NeuroD1, all transcriptional targets of canonical β-catenin signaling (48), between CaMKII-Wnt5a^fl/fl and control Wnt5a^fl/fl hippocampus at 3 mo (Fig. S5 A–C), indicating that canonical Wnt signaling is unaffected by Wnt5a depletion in the mature hippocampus.

Fig. S5.

Wnt5a deletion perturbs calcium and planar cell polarity signaling, but not canonical Wnt signaling in the mature hippocampus. (A and B) Similar levels of nuclear (active) β-catenin in hippocampal nuclear fractions from CaMKII-Wnt5a^fl/fl (KO) mice and control littermates (WT) at 3 mo. The β-catenin immunoblot was reprobed for p84, a nuclear protein, for normalization. Results are means ± SEM from five mice per genotype. (C) Axin2, c-myc, and NeuroD1 transcript levels are unaffected by Wnt5a deletion in 3-mo-old KO hippocampus. Results are means ± SEM from six mice per genotype. (D) Representative immunoblots of synapse-associated signaling and structural proteins. Immunoblots were stripped and reprobed for tubulin for normalization. (E) Densitometric quantification of immunoblots of synaptic proteins in hippocampal homogenates from 3-mo-old KO and WT mice. Densitometric signal intensities for each protein were normalized to tubulin, and values expressed relative to control homogenates.

We next assessed the Wnt–calcium pathway, where Wnt ligands promote an increase of cytoplasmic Ca²⁺ (49, 50). Strikingly, Wnt5a treatment acutely elicited a calcium response in 92% of cultured rat hippocampal neurons transfected with GCaMP3, whereas only 43% of neurons responded to control treatment. Furthermore, the number of calcium transients was fivefold higher in Wnt5a-treated neurons (Fig. 4 A–C and Movies S1 and S2). We then performed biochemical analyses to assess activation of CaMKII, a critical regulator of hippocampal connectivity and functions (51, 52), in vivo, in young adult CaMKII-Wnt5a^fl/fl mice at 3 mo before the appearance of structural anomalies. Using a phospho-specific antibody that detects activated CaMKII (threonine 286 phosphorylation on CaMKIIα and T287 on CaMKIIβ) (53), we found a pronounced attenuation of phosphorylated CaMKIIα (59% decrease) and CaMKIIβ (57% decrease) in CaMKII-Wnt5a^fl/fl mice (Fig. 4 D and E). CaMKII-mediated phosphorylation of the GluA1 subunit of AMPA-type glutamate receptors at a critical serine 831 site (54, 55) has been functionally linked to synaptic plasticity and retention of spatial memory in mice (56, 57). We found a marked decrease in phospho-S831-GluA1 in postsynaptic density fractions from the mutant hippocampus (Fig. 4 D and E). These results suggest decreases in phosphorylation of CaMKII and GluA1 as the molecular underpinnings for the impairments in synaptic plasticity and spatial memory in CaMKII-Wnt5a^fl/fl mice.

Fig. 4.

Wnt5a loss disrupts calcium and cytoskeletal signaling and decreases CREB-mediated GluN1 synthesis. (A) Wnt5a elicits a robust increase in calcium transients in hippocampal neurons. Rat hippocampal neuron cultures were transfected with GCaMP3 and treated with control L- or Wnt5a-conditioned media for 30 min. (Scale bar, 20 μm.) (B) Representative traces of calcium transients. (C) Wnt5a elicits a fivefold increase in calcium transients compared with control media. Results are mean ± SEM from n = 3 independent experiments (total of n = 31 L media-treated cells and n = 26 Wnt5a-treated neurons); ***P < 0.001, two-tailed t test. (D and E) Phosphorylated CaMKIIα/β and phosphorylation of GluA1^Ser831 are significantly reduced in CaMKII-Wnt5a^fl/fl (KO) hippocampus compared with Wnt5a^fl/fl litter-mates (WT) at 3 mo. Immunoblots were stripped and reprobed for total CaMKIIα, -β, and GluA1 for normalization. Results are mean ± SEM from n = 6 mice per genotype; *P < 0.05, ***P < 0.001, two-tailed t test. (F and G) CaMKIV and CREB phosphorylation are attenuated in hippocampal nuclear fractions from 3-mo-old KO mice. Blots were reprobed for total CaMKIV and CREB. Results are mean ± SEM from n = 6 and n = 5 mice per genotype; ***P < 0.001, two-tailed t test. (H and I) GluN1, the obligatory NMDA receptor subunit, is significantly decreased in 3-mo KO hippocampus, but levels of other NMDA receptor subunits, GluN2a/2b, and AMPA-type glutamate receptor subunits, GluA1 and GluA2, are unaltered. Immunoblots were stripped and reprobed for tubulin. Results are mean ± SEM from n = 6 mice per genotype; ***P < 0.001, two-tailed t test. (J) qPCR analysis shows decrease in GluN1 but not GluN2a/b transcripts in the KO hippocampus. Results are mean ± SEM from n = 6 mice per genotype; **P < 0.01, two-tailed t test. (K) A 1-kb region upstream of the mouse GluN1 promoter harbors three putative CRE sites. Hippocampal neurons were cotransfected with luciferase reporter constructs, Cypridina luciferase driven by the GluN1 promoter region, and Gaussia luciferase driven by an SV40 promoter as a control. Neurons were treated with control or Wnt5a media for 6 h and then harvested to measure luciferase activity. (L) Wnt5a treatment significantly increases GluN1 promoter-driven luciferase activity. Deletion of all three CRE sites or just the two proximal binding sites (CRE1 and CRE2) alone abolishes the Wnt5a-mediated response. Results are mean ± SEM for n = 6 independent experiments; *P < 0.05, **P < 0.01, ***P < 0.001, two-tailed t test. (M and N) Rac1-GTP levels are reduced in 3-mo KO mice. Hippocampal homogenates were subjected to GST-PAK-PBD-agarose pull-downs for active Rac-GTP and immunoblotted for Rac1. Rac1-GTP signal intensities were normalized to total Rac1. Results are mean ± SEM from n = 6 mice per genotype; **P < 0.01, two-tailed t test. (O and P) Phospho-JNK levels are significantly decreased in hippocampal homogenates from 3-mo KO mice. Results are means ± SEM from n = 6 mice per genotype; *P < 0.05, **P < 0.01, two-tailed t test.

Calcium signaling within synapses could couple to transcriptional responses via shuttling of a Ca^2+/CaM/CaMKIIγ complex to the nucleus to promote phosphorylation of CaMKIV, which then phosphorylates and activates the transcription factor CREB (58). Phosphorylation of CaMKIV and CREB were significantly reduced in nuclear fractions from 3-mo-old CaMKII-Wnt5a^fl/fl hippocampal tissues (Fig. 4 F and G). Because Wnt5a deletion altered nuclear CREB phosphorylation, we assessed levels of several synaptic proteins (Fig. S5 D and E) and found that only GluN1, the obligatory NMDA receptor subunit, was decreased (Fig. 4 H–J), raising the possibility that GluN1 transcription is CREB-dependent. We did not observe any changes in levels of other NMDA receptor subunits, GluN2a/2b, that are coexpressed and coassembled in the endoplasmic reticulum (ER) with GluN1, in the mature hippocampus (Fig. 4 H–J). We identified three putative CRE sites (CRE1 at −212 bp, CRE2 at −238 bp, and CRE3 at −770 bp) in a 1-kb region upstream of the transcription start site in the mouse GluN1 promoter (Fig. 4K). In a dual luciferase assay, Wnt5a stimulation of hippocampal neurons for 6 h significantly increased luciferase activity compared with control treatment (Fig. 4 K and L). Mutation of just the two proximal CRE elements (−212 to −216 bp and −238 to −242 bp) abolished Wnt5a-induced luciferase activity (Fig. 4L). These results reveal an unexpected role for Wnt5a in enhancing GluN1 transcription through a noncanonical pathway that involves calcium-CaMKII-CREB activation.

We finally examined the planar cell polarity pathway where noncanonical Wnts induce cytoskeletal dynamics by activating small GTPases, such as Rac1 and JNK signaling (59). Rac1 is a critical regulator of the actin cytoskeleton in dendrites and spines (60, 61). Active Rac1-GTP and phospho-JNK levels were significantly reduced in hippocampal homogenates prepared from 3-mo-old CaMKII-Wnt5a^fl/fl mice (Fig. 4 M–P). Rac1 activity can also be influenced by CaMKII activity via CaMKII-mediated phosphorylation of the Rac1-specific GEFs, Tiam1 and Kalirin-7 (62, 63). Taken together, these results suggest that Wnt5a signals via CaMKII and Rac1-mediated signaling, as well as CREB-mediated GluN1 synthesis to maintain synaptic plasticity and structure in the adult hippocampus.

Late Induction of Wnt5a Reverses Dendrite Attrition.

Our results show that adult Wnt5a-deficient mice have profound defects in hippocampal synaptic plasticity, dendrite morphology, and related molecular changes. Could restoring Wnt5a expression prevent or even correct synaptic signaling and dendritic defects in adult CaMKII-Wnt5a^fl/fl mice? To address this question, we expressed Wnt5a using an adeno-associated virus (AAV) virus carrying a Cre-dependent Wnt5a transgene, DIO-Wnt5a, in CaMKII-Wnt5a^fl/fl mice. First, to address if Wnt5a expression rescues signaling defects observed at 3 mo, we delivered Wnt5a at this time point and performed biochemical analyses at 2 wk after viral infection (Fig. 5A). Immunoblotting of hippocampal homogenates revealed that AAV-mediated Wnt5a expression was sufficient to correct the impairments in CaMKII and CREB phosphorylation, GluN1 expression, and Rac1 activity in CaMKII-Wnt5a^fl/fl mice (Fig. 5 B–E). These results suggest a direct and acute role for Wnt5a in regulating calcium and Rac1 activity and enhancing CREB-mediated synthesis of NMDA-type glutamate receptor subunits in the mature hippocampus.

Fig. 5.

Reversal of dendritic attrition by late induction of Wnt5a expression. (A) Schematic of the strategy to assess effects of Wnt5a expression on synaptic signaling in CaMKII-Wnt5a^fl/fl mice. Cre-dependent AAV expressing Wnt5a (AAV-DIO-Wnt5a) was injected into CaMKII-Wnt5a^fl/fl mice at 3 mo and biochemical analyses performed 2 wk later. AAV-DIO-Wnt5a infected mutant mice (Rescue) were compared with CaMKII-Wnt5a^fl/fl mice (KO) or control Wnt5a^fl/fl mice (WT) infected with AAV-GFP. (B–E) Wnt5a expression fully restored CaMKII and CREB phosphorylation, GluN1 expression, and Rac1 activity in CaMKII-Wnt5a^fl/fl hippocampi. Results are mean ± SEM from n = 6 mice per group; *P < 0.05, **P < 0.01, ***P < 0.001, one-way ANOVA followed by Tukey’s post hoc test. (F) Schematic of the strategy to assess effects of Wnt5a expression on neuronal morphology in adult Thy1-GFP;CaMKII-Wnt5a^fl/fl mice. AAV-DIO-Wnt5a was delivered into Thy1-GFP;CaMKII-Wnt5a^fl/fl hippocampus at 6 mo when there is marked dendritic regression, and morphological analyses were performed at 9 mo. Infected neurons were identified by coinfection of AAV-DIO-mCherry. GFP fluorescence was used for imaging. (G) CA1 dendritic arbors in Wnt5a-infected Thy1-GFP;CaMKII-Wnt5a^fl/fl mice (Rescue) were comparable to control Thy1-GFP;Wnt5a^fl/fl mice (WT). However, Thy1-GFP;CaMKII-Wnt5a^fl/fl neurons (KO) infected with AAV-DIO-mCherry had stunted dendritic arbors, as expected. (Scale bar, 100 μm.) (H) Sholl analysis shows that AAV-mediated Wnt5a expression for 3 mo corrects dendrite complexity defects in 9-mo-old Thy1-GFP;CaMKII-Wnt5a^fl/fl mice. (I and J) Dendritic lengths and dendritic branch points are comparable between WT and Rescue neurons, but significantly decreased in KO neurons in 9-mo-old mice. Results are mean ± SEM from n = 5 neurons traced per animal and a total of five mice per group; *P < 0.05, ***P < 0.001, one-way ANOVA followed by Tukey’s post hoc test.

We next addressed whether the dendrite atrophy in adult CaMKII-Wnt5a^fl/fl neurons was permanent or could be reversed by Wnt5a administration well after the onset of structural abnormalities. Thus, we delivered AAV-DIO-Wnt5a into Thy1-GFP;CaMKII-Wnt5a^fl/fl mice at 6 mo when CaMKII-Wnt5a^fl/fl mice exhibit pronounced regression of dendritic arbors, and mice were harvested 3 mo after AAV infections for morphological analyses (Fig. 5F). We simultaneously delivered AAV-DIO-mCherry as a means to label infected neurons that are also GFP⁺ to facilitate tracing of neuronal morphologies in isolated neurons. Strikingly, AAV-mediated expression of Wnt5a in Thy1-GFP;CaMKII-Wnt5a^fl/fl resulted in dendritic arbors that were comparable in complexity and lengths to control Thy1-GFP;Wnt5a^fl/fl neurons (Fig. 5 G–J). As expected, with infection with AAV-DIO-mCherry alone, we observed substantial decreases in dendrite complexity and lengths in Thy1-GFP;CaMKII-Wnt5a^fl/fl CA1 neurons (Fig. 5 G–J). These findings indicate that dendritic attrition in adult mutant neurons can be reversed, and reveals that Wnt5a, remarkably, promotes substantial dendritic growth and branching in the adult brain when neuronal connectivity is thought to be largely immutable.

Discussion

Wnts are evolutionarily conserved signaling molecules that have been classically associated with embryonic patterning and establishment of neural circuits (47, 64, 65). That these classic developmental cues may have critical functions in the adult brain has been implied by recent findings that broad-spectrum blockade or activation of Wnt pathway components affects synaptic structure, plasticity, and cognitive functions in adult animals (37–41). However, surprisingly little is known about which of the 19 vertebrate Wnts is essential for adult nervous system functions in vivo. Furthermore, manipulation of the Wnt pathway through overexpression of antagonists, such as Dickkopf-1 (38, 41), deletion of the Lrp6 coreceptor (40), or the cytoplasmic effector, β-catenin (37), may have consequences on neuronal connectivity and function that are independent of Wnt ligands, via effects on cell–cell adhesion, JNK signaling, and GPCR-mediated cAMP signaling (66–68). Here, we show that deletion of a single Wnt family member, Wnt5a, is sufficient to elicit profound disruptions in synaptic plasticity, structural maintenance, and learning and memory in adult mice, identifying the importance of this particular noncanonical Wnt in later-life functions. Thus, the loss of Wnt5a cannot be compensated for by other Wnts in the adult hippocampus. Together, our results, summarized in the model in Fig. 6, define a causal sequence of events where Wnt5a first influences synaptic plasticity and related cognitive functions in the adult hippocampus through CaMKII-mediated signaling, Rac1-dependent actin dynamics, and CREB-mediated NMDA receptor biosynthesis. In the long-term, Wnt5a-mediated regulation of cytoskeletal signaling and excitatory synaptic transmission is responsible for the maintenance of dendritic arbors and spines. These findings provide insight into the poorly understood structural maintenance mechanisms that exist in the adult brain, and suggest Wnt5a signaling as a molecular target in ameliorating dendrite shrinkage and cognitive decline associated with pathological situations.

Fig. 6.

Model for adult-specific roles for autocrine Wnt5a signaling in maintenance of CA1 dendrite architecture and function. Wnt5a influences synaptic plasticity and related cognitive functions in the adult hippocampus through CaMKII-mediated signaling and NMDA receptor biosynthesis, and also Rac1-dependent actin dynamics in dendritic spines. In the long-term, Wnt5a-mediated regulation of cytoskeletal signaling and excitatory synaptic transmission is responsible for maintenance of dendritic structure.

The finding that embryonic deletion of Wnt5a in neurons did not elicit any structural abnormalities in CA1 pyramidal neurons during development suggests that neuronal Wnt5a is dispensable for the establishment or maturation of hippocampal connectivity in vivo. These results were surprising in the context of reported developmental functions for Wnt5a in cultured hippocampal neurons, and in embryonic processes in other brain regions (15, 18, 19). In hippocampal neurons, several signaling pathways have been shown to influence dendrite morphogenesis, maturation, and stability in vitro and in vivo (1, 6, 69). Thus, in the absence of Wnt5a, other signaling mechanisms, including other Wnt molecules (26, 27), could provide trophic support to hippocampal CA1 dendrite arbors and spines at least for the first several months of life in mice. Alternatively, Wnt5a derived from nonneuronal sources may support hippocampal formation in the absence of neuron-derived Wnt5a. However, the profound defects in adult mice lacking Wnt5a suggest that these mechanisms are unable to compensate for Wnt5a loss at later stages of life. Notably, we demonstrate that Wnt5a, derived from CA1 pyramidal neurons themselves, is critical for sustaining dendritic architecture in the adult hippocampus, implying that specificity for neuronal wiring is intrinsic to active neurons themselves in hippocampal circuits. To date, our limited understanding of the molecular cues that influence neuronal morphology in adult animals has largely come from analyses of cortical neurons in genetically modified mice. Among the examples are adult mice with deletion of BDNF and its receptor TrkB (7, 8), the adhesion molecule δ-catenin (70), and the tumor suppressor phosphatase and tensin homolog deleted on chromosome 10 (PTEN) (71). Our findings identifying Wnt5a as being essential for the maintenance of adult CA1 hippocampal neurons is relevant to understanding the structural bases of hippocampus-dependent behaviors.

We found that synaptic plasticity is most susceptible to the postnatal depletion of Wnt5a. CaMKII-Wnt5a^fl/fl mice had impaired CA1 LTP and related behavioral defects at 3 mo of age, a time when basal synaptic transmission and dendritic morphology are intact. The normal presynaptic properties indicate that Wnt5a acts primarily at postsynaptic sites. Recombinant Wnt5a has previously been shown to acutely modulate NMDAR-mediated synaptic transmission in rat hippocampal slices (72). Our results suggest that Wnt5a likely modifies synaptic strength through CaMKII-mediated signaling events, including the phosphorylation and subsequent trafficking/conductance of AMPA-type glutamate receptors, Rac1-dependent regulation of actin dynamics in dendritic spines, and regulation of NMDA receptor biosynthesis. Attenuation of small GTPase-mediated signaling and excitatory synaptic transmission, both postulated to be critical determinants in stabilizing neuronal connectivity (61, 73, 74), may underlie the gradual attrition of dendritic arbors and spines in later life. Because mice with forebrain-specific deletion of GluN1 have impairments in plasticity at CA1 synapses and spatial memory acquisition (75), these findings suggest that down-regulation of NMDA receptor synthesis contributes, in part, to the functional and behavioral defects that we observed in CaMKII-Wnt5a^fl/fl mice. Although NMDA receptor-mediated LTD is impaired in mice with CaMKII-Cre–mediated deletion of GluN1 (75), that we observed normal LTD responses in CaMKII-Wnt5a^fl/fl mice can be attributed to the fact that residual GluN1 expression in Wnt5a mutant mice may still allow sufficient Ca²⁺ influx to promote LTD, consistent with the view of differential Ca²⁺ requirements for LTD versus LTP (76, 77). Previously, targeted GluN1 deletion has also been reported to result in modest (∼35%) decreases in GluN2a/2b protein expression, but unaltered levels of GluN2a/2b mRNA (78, 79). The decrease in GluN2a/2b subunits was attributed to their aberrant retention in the ER and protein degradation when GluN1 is unavailable (78). That we did not observe any changes in GluN2a/2b protein levels in CaMKII-Wnt5a^fl/fl mice is likely because the GluN1 depletion in our study (34%) is less robust compared with the near-complete depletion previously reported in GluN1 conditional null mice (78, 79). Thus, residual GluN1 expression in CaMKII-Wnt5a^fl/fl mice might be sufficient to avoid ER-associated degradation of the GluN2a/2b subunits. Although the total GluN2a/2b protein content was unaltered in the CaMKII-Wnt5a^fl/fl hippocampus, it is possible that their synaptic localization might be affected by Wnt5a loss, given previous findings that Wnt5a modulates the surface expression of GluN2b in cultured hippocampal neurons (80).

Currently, the Wnt5a receptors that mediate effects on dendritic maintenance and synaptic functions in the adult hippocampus remain to be determined, although likely candidates include the Ror1/2 receptor tyrosine kinases and Frizzled-9. Ror1/2 have been demonstrated to be bona fide Wnt5a receptors in vivo (81). Ror2 is abundantly expressed in mature CA1 dendrites, promotes dendritic activation of noncanonical Wnt signaling and, notably, is necessary for Wnt5a-mediated potentiation of NMDAR currents in acute hippocampal slices (82). Ror2 may function in coordination with Frizzled receptors; in particular, Frizzled-9 is localized to postsynaptic sites in hippocampal neurons, binds Wnt5a via its cysteine-rich domain in biochemical analyses, and the Frizzled-9 cysteine-rich domain is required for Wnt5a-mediated changes in spine densities in cultured hippocampal neurons (32).

The early LTP defects and the cognitive decline followed by retraction of dendrites and spine loss observed in adult Wnt5a mutant mice bear similarities to the progression of events in animal models of Alzheimer’s disease (83). Recent genetic evidence implicates deficiencies in Wnt signaling, largely the canonical arm, in the synaptic dysfunction and cognitive impairments in Alzheimer’s disease (40, 84, 85). Our study emphasizes that noncanonical Wnt signaling is essential for maintaining synaptic function and connectivity in the adult brain. That late induction of Wnt5a expression even after the onset of substantial neuronal atrophy, remarkably restores dendrite morphology in adult neurons, highlights the capacity of the adult nervous system to undergo large-scale structural changes, and suggests a Wnt5a-dependent trophic pathway that could be harnessed for therapeutic purposes in pathological situations.

Materials and Methods

Animals.

All procedures relating to animal care and treatment conformed to The Johns Hopkins University and NIH guidelines. Animals were housed in a standard 12:12 light:dark cycle. The generation of Wnt5a^fl/fl mice has been previously described (86). Hippocampal neuron cultures were established from embryonic day 18 (E18) rat pups, as previously described (87).

Neuronal cell counts were performed as described in Ramanan et al. (88). Lentiviral or AAV vectors were stereotaxically delivered to the hippocampus using coordinates that were previously described (89). Golgi-based analyses of dendrite arbors and spines were performed as described previously (90). Further details of dendrite reconstructions and analyses are included in SI Materials and Methods. Details of in situ hybridization, real-time PCR primers and assays, Rac1 GTPase activity, electrophysiology, calcium imaging, luciferase assays, and the novel-object recognition test can be found in SI Materials and Methods. The Morris water maze test was performed as previously described (91), and the visual acuity was measured as previously described (92).

Statistical Analyses.

All Student’s t tests were performed using two-tailed, unpaired, and a confidence interval of 95%. One-way or two-way ANOVA analyses were performed when more than two groups were compared. Statistical analyses were based on at least three independent experiments and are described in the figure legends. All error bars represent the SEM.

SI Materials and Methods

Animals.

All procedures relating to animal care and treatment conformed to institutional and NIH guidelines. Animals were housed in a standard 12:12 light:dark cycle. The generation of Wnt5a^fl/fl mice has been previously described (86). Wnt5a^fl/fl mice were backcrossed to C57BL/6 background for at least seven generations and maintained on a C57BL/6 background. Nestin-Cre and Thy1-GFP-M transgenic mice were obtained from the Jackson Laboratory, and CaMKII-Cre mice (T29-1 line) were a generous gift from Nicholas Gaiano, NIH, Bethesda. Sprague–Dawley rats were purchased from Charles River. Hippocampal neuron cultures were established from E18 rat pups as previously described (87).

Reagents and Antibodies.

The antibodies used in this study were previously validated for the following applications: Wnt5a (R&D Systems; AF645, Western blotting), VGLUT1 (Millipore; AB5905, Western blotting), PSD95 (NeuroMab; Western blotting), GluA1 [Johns Hopkins Medical Institutions (JHMI) monoclonal antibody core; Western blotting], GluA2 (JHMI monoclonal antibody core; Western blotting), GluN1 (Millipore; MAB363, Western blotting), CaMKII (Cell Signaling; 4436, Western blotting), Tubulin (Sigma Aldrich; T-6199, Western blotting), Rac1 (Millipore; 05-389, Western blotting), P-CaMKII (Thr286) (Cell Signaling; 3316 and 12716, Western blotting), P-CaMKIV (Thr196) (Santa Cruz; sc-28443-R, Western blotting), CaMKIV (Abcam; ab3557, Western blotting), P-CREB (Ser133) (Cell Signaling; 9198, Western blotting), CREB (Cell Signaling; 9104, Western blotting), GluN2a (JHMI monoclonal antibody core; Western blotting), GluN2b (JHMI monoclonal antibody core; Western blotting), Akt (Cell Signaling; 9272, Western blotting), Erk (Cell Signaling; 9102, Western blotting), Nuclear matrix protein p84 (GeneTex; GTX70220, Western blotting), Active β-catenin (Millipore; 05-665, Western blotting), p-JNK (Cell Signaling; 9251, Western blotting), JNK (Cell Signaling; 9252, Western blotting), p-GluA1-Ser831 (Millipore; AB5847, Western blotting), NeuN (Millipore; MAB377, immunohistochemistry), Neurofilament (Millipore; AB5539, immunohistochemistry), MAP2 (Sigma Aldrich; M9942, immunohistochemistry) Synaptophysin (Sigma Aldrich; S5768, Western blotting) and Synapsin 1 (BD Transduction laboratories; 611392, Western blotting). Golgi-Cox staining was done using the FD Rapid GolgiStain kit (FD Neurotechnologies; PK401). Rac1/Cdc42 GTPase activity was measured by a kit (EMD Millipore; 17-441). GluN1 promoter activities were followed with BioLux Cypridina and Gaussia luciferase assay kits (New England Biolabs, E3309S and E3300S). Wnt5a conditioned media and L-cell conditioned media were harvested from Wnt5a producing cells (ATCC CRL-2814) and L cells (ATCC CRL-2648). Both cell lines were first cultured in DMEM supplemented with 10% (vol/vol) FBS for 2 d, and then the medium was changed to Neurobasal medium supplemented with B27 for 24 h.

In Situ Hybridization.

In situ hybridization was performed using a digoxigenin-labeled probe spanning a 572-bp region within Wnt5a exon 2. Mouse brains of various ages were fresh frozen in OCT (Tissue-Tek) and serially sectioned (20 μm) using a cryostat. Sections were postfixed in 4% paraformaldehyde (PFA), washed in PBS, and acetylated with 0.25% acetic anhydride in 0.1 M triethanolamine with 0.9% NaCl. After hybridization with the labeled RNA probe (2 μg/mL) at 68 °C overnight, sections were washed with 0.2× SSC buffer at 65 °C, blocked with TBS containing 1% normal goat serum, and then incubated with alkaline phosphatase-labeled anti-DIG antibody (1:5,000; Roche) overnight at 4 °C. The alkaline phosphatase reaction was visualized with NBT/BCIP, rinsed in PBS, fixed in 4% PFA and mounted in AquaMount (EMD Chemicals).

Real Time-PCR Analyses.

Total RNA was prepared from dissected hippocampal tissues using TRIzol-chloroform extraction (ThermoFisher; 10296-010). RNA was then reverse-transcribed using a RETROscript Reverse Transcription kit (ThermoFisher; AM1710). Real-time qPCR was performed using a Maxima SYBR Green/Rox Q-PCR Master Mix (ThermoFisher; K0221), in 7300 Real time PCR System (Applied Biosystems). The following primers were used for the analyses; Wnt5a-F:5′-CTCGGGTGGCGACTTCCTCTCCG-3′ and Wnt5a-R:5′-CTATAACAACCTGGGCGAAGGAG-3′; Wnt2-F:5′-GTAGATGCCAAG GAGAGGAAAG-3′ and Wnt2-R:5′-CCACTCACACCATGACACTT-3′; Wnt3-F:5′-TGGACCACATGCACCTAAAG-3′ and Wnt3-R:5′-CGTACTTGTCCTTGAGGAAG TC-3′; Wnt3a-F:5′-GCAGCTGTGAAGTGAAGAC-3′ and Wnt3a-R:5′-GGTGTTTCT CTACCACCATCTC-3′; Wnt5b-F:5′-GAGAGCGTGAGAAGAACTTTG-3′ and Wnt5b-R: 5′-GCGACATCAGCCATCTTATAC-3′; Wnt7a-F: 5′-GCCTTCACCTATGCGAT TATC-3′ and Wnt7a-R: 5′- GGTACTGGCCTTGCTTCTC-3′; Wnt7b-F: 5′-GCATGA AGCTGGAATGTAAGTG-3′ and Wnt7b-R: 5′-TGCGTTGTACTTCTCCTTGAG-3′; Wnt8a-F: 5′-TGGGAACGGTGGAATTGTC-3′ and Wnt8a-R: 5′-GCGGATGGCAT GAATGAAG-3′; Wnt11-F: 5′-CCTGGAAACGAAGTGTAAATGC-3′ and Wnt11-R: 5′-TGACAGGTAGCGGGTCTTG-3′; GluN1-F: 5′-CCAGATGTCCACCAGACTAA AG-3′ and GluN1-R: 5′-CATTGACTGTGAACTCCTCTTTG-3′; GluN2a-F: 5′-CTGTGTGGCCAAGGTATAAG-3′ and GluN2a-R: 5′-TCAGTCAGTGGGTCTAT GTC-3′; GluN2b-F: 5′-ATGAGGAACCAGGCTACATC-3′ and GluN2b-R: 5′-GGT CACCAGGTAAAGGTCATAG-3′; Axin2-F: 5′-GAGAGTGAGCGGCAG AGC-3′ and Axin2-R: 5′-CGGCTGACTCGTTCTCCTG-3′; NeuroD1-F:5′-GCTACTCCAAGACC CAGAAAC-3′ and NeuroD1-R:5′-TGTACGAAGGAGACCAGATCA-3′; c-myc-F: 5′-CTCCGTACAGCCCTATTTCATC-3′ and c-myc-R: 5′-TGGGAAGCAGCTCGAATT TC-3′; GAPDH-F: 5′-CCTGCACCACCAACTGCTTA-3′ and GAPDH-R: 5′-CC ACGATGCCAAAGTTGTCA-3′. Each sample was analyzed in triplicate reactions. Data were analyzed using the ΔΔC_t method, normalizing each sample to the internal control, and relative messenger RNA was determined as the percentage of the maximum value observed in the experiment.

Immunohistochemical Analyses.

Mouse brains were fixed in 4% PFA at 4 °C overnight, cryoprotected in 30% sucrose in PBS, frozen in OCT (Tissue-Tek) with dry ice, and serially sectioned (40 μm). For immunohistochemistry with diaminobenzidine (DAB), sections were first incubated with 10% methanol + 3% H₂O₂ for 20 min to quench peroxidase activity and then washed in TBS, permeabilized in TBS containing 0.4% Triton X-100, and blocked using 10% goat serum + 3% BSA for 2 h. Sections were incubated in the following primary antibodies overnight: chicken antineurofilament (EMD Millipore; AB5539, 1:500) or mouse anti-MAP2 (Sigma Aldrich; M9942, 1:1,500). Following TBS washes, sections were incubated with either chicken or mouse anti-HRP secondary antibodies (GE Healthcare; 1:500) for 1 h. Sections were then washed in TBS, incubated with DAB for 2–10 min at room temperature, followed by washes in TBS and then mounted in VectaShield (Vector Laboratories).

For immunofluorescence, sections were washed in TBS, permeabilized in TBS containing 0.4% Triton X-100, and blocked using 10% goat serum + 3% BSA for 2 h. Sections were then incubated in primary antibodies overnight. Following TBS washes, sections were incubated with Alexa-Fluor conjugated secondary antibodies (ThermoFisher; 1:500) for 1 h. Sections were washed in TBS and then mounted in VectaShield (Vector Laboratories).

Neuronal Cell Counts.

Six- and 12-mo mouse brains for neuronal counts were prepared as described in Ramanan et al. (88). Briefly, mouse brains were fixed in PBS containing 4% PFA, and then cryoprotected overnight in 30% sucrose-PBS. Coronal brain sections (20 μm) were stained with a solution containing 0.5% Cresyl violet (Nissl), and cells in CA1, CA3, and dentate granule layers were counted at 40× magnification. For each mouse, 9–12 defined regions were counted and analyzed from three coronal sections using ImageJ.

Viral Injections.

Lentiviruses expressing GFP or GFP-T2A-CRE were generated by subcloning into lentiviral backbones using XhoI and NotI restriction sites. Lentivirus was produced in HEK293T cells using the FUW/Δ8.9/VSVG system. Two collections of media were done 24 and 48 h after transfection, and virus particles pelleted by ultracentrifugation. Virus particles were then resuspended in Neurobasal media and stored at –80 °C until use.

AAV-EF1a-DIO-mCherry and AAV-CMV-GFP were obtained from University of North Carolina Vector Core (2.4 × 10¹³ viral particles per milliliter, stock by Deisseroth laboratory) and University of Pennsylvania Vector Core (5.2 × 10¹³ viral particles per milliliter), respectively. AAV-CMV-DIO-Wnt5a was generated by Vigene Biosciences (2.8 × 10¹³ viral particles per milliliter).

For viral injections, mice were deeply anesthetized with avertin (tribromoethanol, 0.25 mg/g body weight; 2–methyl–2–butanol, 0.16 μL/g body weight) and mannitol (to prevent edema, 10 mg/g body weight). Lentiviral or AAV vectors were stereotaxically delivered to the hippocampus using coordinates that were previously described (89). After injections, mice were allowed to recover and killed for biochemical analyses 2 wk after infections for biochemical analyses or 3 mo later for morphological analyses.

Dendrite Reconstructions and Analyses.

Thy1-GFP mouse brains were harvested and fresh frozen in OCT, and serial coronal sections (200 μm) were prepared using a cryostat. Brain sections were mounted on slides and preserved with PermaFluor mounting medium (Thermo Scientific). GFP-filled CA1 pyramidal neuronal morphologies were imaged using 20× objective to visualize entire dendritic fields (512 × 512 pixels; 1-μm optical scanning) followed with a 3 × 4 tile scan using a Zeiss LSM 710 confocal microscope (Carl Zeiss). Z-stacks were stitched together using Zen imaging software. All of the analyzed neurons were sufficiently bright to allow complete tracing of dendrite arbors, and had intact dendritic arbors. Neuronal somas and dendritic structure were manually traced with Imaris 7.7 software (Bitplane) to create 3D reconstructions of dendritic structures. Dendritic arbors that could not be confidently reconstructed were not used in the analyses. Total dendrite lengths were quantified using Imaris. Dendritic complexities were analyzed by 3D Sholl analysis, and calculating the number of dendrite intersections with concentric spheres that radiated from the soma in 10-μm-radius increments. Five neurons were traced per animal, and the average used as the value for each mouse during statistical analyses.

Golgi staining was performed with FD Rapid Golgistain kit on mouse brains that were fresh-frozen in OCT and serially sectioned (200 μm) using a cryostat. Dendritic arbors were analyzed as described previously (90). Briefly, Golgi-impregnated pyramidal neurons in the CA1 region were traced using Neurolucida software (MicroBrightField) under a Nikon Eclipse E800 microscope, equipped with a motorized stage. Analyses were done by an investigator blinded to the genotype. All of the analyzed neurons were well stained, isolated and had intact dendritic arbors. Dendritic length of each traced neuron was calculated using NeuroExplorer software (MicroBrightField). Five neurons were traced per animal, and the average used as the value for each mouse during statistical analyses.

To measure spine density, we took images of distal apical dendrites (>100 μm away from the soma) of hippocampal CA1 pramidal neurons at 63× magnification using a Zeiss Axiovision microscope with a AxioCam HRC digital camera. The position of each dendritic spine along these dendrite segments was assessed manually and counted using ImageJ software. Spine density for each animal was obtained from at least 15 dendrites per mouse with a total length of 2,000∼3,000 μm traced for statistical analysis.

Rac1 GTPase Activity.

The Rac1/Cdc42 Activation Assay (EMD Millipore; 17–441) was used to detect Rac1 activity in hippocampal homogenates. Briefly, hippocampal lysates were precleared with glutathione-agarose beads (100 μL). Supernatants were incubated with a fusion protein (GST-PAK) coupled to agarose beads. GST-PAK specifically binds to Rac1 or Cdc42 in its GTP-bound form. Rac1-GTP in the pull-downs was detected by immunoblotting using a Rac1-specific antibody.

Slice Preparations and Electrophysiology.

Three- or 6-mo-old male mice were anesthetized with the inhalation anesthetic isoflurane before decapitation. Brains were rapidly dissected out and placed in ice-cold, oxygenated (95% O₂ and 5% CO₂) low-Ca²⁺/high-Mg²⁺ dissection buffer containing 2.6 mM KCl, 1.25 mM NaH₂PO₄, 26 mM NaHCO₃, 211 mM sucrose, 11 mM glucose, 0.5 mM CaCl₂, and 7 mM MgCl₂. The 350-μm transverse hippocampal slices were prepared using a vibratome (Leica; VT1000s) in dissection buffer followed by recovery in a static submersion chamber filled with oxygenated artificial cerebrospinal fluid (ACSF) containing 125 mM NaCl, 5 mM KCl, 1.25 mM NaH₂PO₄, 2 mM CaCl₂, 1 mM MgCl₂, 25 mM NaHCO₃, and 11 mM glucose at 30 °C for at least 1 h before LTP or 2 h before LTD recording.

Slices were transferred to a recording chamber where they were perfused continuously with oxygenated ACSF at a flow rate of ∼3 mL/min at 30 °C. Hippocampal CA1 fEPSP was evoked at 0.033 Hz with a 125-μm platinum/iridium concentric bipolar electrode (FHC) placed in the middle of the stratum radiatum of CA1. Synaptic responses were recorded with ACSF-filled microelectrodes (1–2 MΩ), positioned ∼200 μm away (orthodromic) from the stimulating electrode, and were quantified as the initial slope of fEPSP. Input–output relationships were obtained for each slice with various stimulus intensity and responses were set to ∼45% maximum for LTP experiments and ∼55% maximum for LTD experiments. Input–output curves were plotted as the fEPSP slope against the presynaptic fiber volley amplitude across all stimulation intensities and the slope for the linear fit of the fiber volley–fEPSP slope relationship were obtained from each slice. PPF was recorded with an interstimulus interval of 25−250 ms. PPF data were presented as a ratio of the second response slope relative to the first. LTP was induced by θ-burst stimulation, consists of four trains of 10 bursts at 5 Hz, with each burst consisting of four stimuli given at 100 Hz and 10-s intertrain interval. LTD was induced by long-frequency stimulation consists of 900 single pulses at 1 Hz.

All hippocampal slice electrophysiological recordings were performed by an experimenter blind to the animal genotype. Statistical significance was determined with a two-way ANOVA for PPF and a two-tailed, unpaired t test for input–output relationship, LTP, and LTD. Representative traces are averages of 4 (PPF and input–output relationship) or 10 (LTP and LTD) traces, and stimulus artifacts have been removed for clarity.

Calcium Imaging.

Hippocampal neurons from E18 rat pups were plated onto glass coverslips coated with poly-l-lysine (1 μg/mL; Sigma-Aldrich) and grown in Neurobasal growth medium supplemented with 2% B27, 2 mM GlutaMax, 50 U/mL penicillin, 50 μg/mL streptomycin, and 5% FBS. Neurons were switched to serum-free Neurobasal medium 24 h postseeding and fed twice a week. Neurons were cotransfected with GCaMP3 and TdTomato at 14–15 d in vitro using lipofectamine 2000 (Invitrogen). Briefly, coverslips containing neurons were assembled onto a closed perfusion chamber and continuously perfused with ACSF buffer with 2 mM Mg²⁺. After 10 min of baseline recording (F₀), neurons were perfused with Wnt5a- or L-conditioned media diluted (1:500) in ACSF buffer with 1 mM Mg²⁺ for 30 min. All imaging experiments were performed at 37 °C using a Zeiss spinning disk confocal (Carl Zeiss). The GCaMP3 was imaged at 488-nm excitation and collected through a 505- to 550-nm filter, whereas the TdTomato signal was imaged at 561-nm excitation and 575- to 615-nm emission. Neurons were imaged using a 63× oil objective (N.A. = 1.40) at a rate of six images per minute. Images were analyzed using ImageJ software (NIH) by calculating the normalized change in average GCaMP3 over TdTomato fluorescence intensities from neuronal soma. The fluorescence intensity change is expressed as ΔF/F₀ and the amplitude of fluorescence change (ΔF_max/F₀) represents the extent of GCaMP3 intensity change.

Luciferase Assays.

To monitor GluN1 promoter activity, the GluN1–Cypridina reporter construct was generated by subcloning a 1-kb region upstream of the transcription start site of mouse GluN1 into pCLuc Mini-TK 2 Vector (New England Biolabs). Neurons were cotranfected with GluN1–Cypridina reporter and a control pSV40-GLuc vector that encodes for Gaussia luciferase under the control of the constitutive SV40 promoter. Two days following transfection, neurons were stimulated with Wnt5a or L media diluted 1:500 in culture media. Because the luciferase proteins are secreted, supernatants were harvested and mixed with either Cypridina or Gaussia luciferase substrates and reporter gene activity was monitored with BioLux Cypridina and Gaussia luciferase assay kits (New England Biolabs) using a luminometer.

Behavioral Tests.

Morris water maze.

Mice were trained to find a submerged platform in a water maze in a 12-d training protocol, as previously described (91). Mice were tested in four stages: naïve acquisition, probe trial, reversal learning, probe trials for memory retention (Fig. 3C). During the naïve acquisition phase, mice were trained (four trials per day for 12 d) to find a hidden platform using four visual distal cues equally surrounding the pool. Each mouse was randomly placed in a different area of the pool at the start of each trial with the platform maintained in the same quadrant (target quadrant). The platform was then removed on day 13 during the probe trial. Swimming in each quadrant and specifically the preference for the target quadrant was measured to evaluate spatial memory using a computerized video tracking system (Any-maze) with a camera mounted in the center above the pool. Reversal training began on day 14 when the platform was moved to the quadrant opposite to the original target quadrant. Mice were trained as described for the acquisition phase.

Latency to locate the platform during the acquisition and reversal phases were analyzed by two-way ANOVA followed by a Fisher–LSD post hoc test to examine changes in latency throughout the course of the experiment as well as the effect of the genotype. Probe trials were analyzed by calculating the percentage time spent in the target quadrant and performing a two-tailed t test.

Novel object recognition.

Recognition memory was assessed using the novel-object recognition test. Mice were first removed from their home cages, acclimated to the empty testing arena for 10 min, and subsequently returned to their home cages. The day after acclimation, mice were returned to this arena with two identical objects that they could freely explore for 10 min, after which they were returned to their home cages for 24 h. At the end of the 24-h period, mice were placed back into the arena with one of the objects changed to a novel object, and were allowed to explore both the familiar and novel objects for 5 min. Behavior was monitored from above by a video camera connected to a computerized video-tracking system (Any-maze), and the percentage of time spent with each object was calculated. The objects and arena were thoroughly cleaned between each trial to remove odor cues. Object recognition was analyzed by calculating the percentage time spent with the novel object and performing a one sample t test to determine whether this was significantly above 50%.

Visual acuity.

Visual acuity was assessed by measuring the image-tracking reflex in a virtual cylinder “OptoMotry” (Cerebral Mechanics), as previously described (92). A sine wave grating was projected on the screen rotating in a virtual cylinder. The animal was assessed for a tracking response upon stimulation for ∼5 s. All acuity thresholds were determined using the staircase method with 100% contrast.