Abstract
Individuals with Alzheimer's disease (AD) have an increased incidence of seizures, which worsen cognitive decline. Using a transgenic mouse model of AD neuropathology that exhibits spontaneous seizures, we previously found that seizure activity stimulates and accelerates depletion of the hippocampal neural stem cell (NSC) pool, which was associated with deficits in neurogenesis-dependent spatial discrimination. However, the precise molecular mechanisms that drive seizure-induced activation and depletion of NSCs are unclear. Here, using mice of both sexes, we performed RNA-sequencing on the hippocampal dentate gyrus and identified differentially expressed regulators of neurogenesis in the Wnt signaling pathway that regulates many aspects of cell proliferation. We found that the expression of sFRP3, a Wnt signaling inhibitor, is altered in a seizure-dependent manner and might be regulated by ΔFosB, a seizure-induced transcription factor. Increasing sFRP3 expression prevented NSC depletion and improved spatial discrimination, suggesting that the loss of sFRP3 might mediate seizure-driven impairment in cognition in AD model mice and perhaps also in AD.
- adult hippocampal neurogenesis
- adult neural stem cells
- Alzheimer's disease
- dentate gyrus
- epilepsy
- seizures
Significance Statement
There is increased incidence of seizures in individuals with Alzheimer's disease (AD), but it is unclear how seizures contribute to cognitive decline. Here, we uncover a molecular mechanism by which seizures in AD induce expression of a long-lasting transcription factor in the hippocampal dentate gyrus that suppresses expression of sFRP3, an inhibitor of neural stem cell (NSC) division, accelerating the depletion of a finite pool of NSCs and dysregulating adult hippocampal neurogenesis. We found that restoring sFRP3 expression prevents accelerated use and depletion of NSCs and improves performance in an adult neurogenesis-dependent cognitive task. Our findings have implications for AD, epilepsy, and other neurological disorders that are accompanied by seizures.
Introduction
Alzheimer's disease (AD) is characterized by impairments in hippocampus-dependent memory (Selkoe, 2002). AD is also associated with increased incidence of seizures and epileptic activity that begin early in disease progression and predict earlier and more rapid cognitive decline (Amatniek et al., 2006; Vossel et al., 2013; Vossel et al., 2017). Antiseizure medications improve cognition in both patients and mouse models of disease, which also exhibit spontaneous seizures (Palop and Mucke, 2009; Cumbo and Ligori, 2010; Bakker et al., 2012; Sanchez et al., 2012; Chin and Scharfman, 2013; Vossel et al., 2013; Bakker et al., 2015; Vossel et al., 2021). However, how seizures impair cognitive function in AD is unclear.
Seizures might contribute to cognitive decline through dysregulation of adult hippocampal neurogenesis (AHN). AHN occurs in the dentate gyrus (DG), where neural stem cells (NSCs) produce progenitors that differentiate into dentate granule cells (Ming and Song, 2011). These newborn neurons are required for spatial discrimination, a behavioral ability that reflects pattern separation, the ability to distinguish similar memories, and is an important component of memory processing (Clelland et al., 2009; Sahay et al., 2011; Nakashiba et al., 2012; Danielson et al., 2016). AHN is sensitive to environmental factors, and pathological activity like seizures can dysregulate neurogenesis (Ming and Song, 2011). AHN and pattern separation are altered in both AD and epilepsy patients and related mouse models (Ally et al., 2013; Wesnes et al., 2014; Reyes et al., 2018; Fu et al., 2019; Kim et al., 2021; Madar et al., 2021). Acute seizures stimulate NSC division and neurogenesis, but chronic epilepsy is associated with reduced neurogenesis (Gray and Sundstrom, 1998; Nakagawa et al., 2000; Hattiangady et al., 2004; Sierra et al., 2015). The reduction in neurogenesis with chronic epilepsy might be related to the finding that some populations of NSCs have limited capacity for self-renewal and instead produce a finite number of neural progenitors before losing their neurogenic capabilities (Encinas et al., 2011). This limitation explains how seizures might accelerate the use and depletion of the NSC pool (Sierra et al., 2015). We previously found accelerated depletion in a mouse model of AD neuropathology that expresses mutant human amyloid precursor protein (APP; Fu et al., 2019). APP mice exhibited spontaneous seizure activity that drove increased proliferation and accelerated loss of NSCs as well as spatial discrimination deficits; treatment with an antiseizure medication prevented these alterations (Fu et al., 2019). Similar accelerated depletion of NSCs has been observed in models of epilepsy and other diseases accompanied by seizures, including traumatic brain injury and stroke (Sierra et al., 2015; Koh and Park, 2017; Neuberger et al., 2017).
The Wnt/β-catenin signaling pathway regulates many aspects of cell function and proliferation in both development and adulthood and is influenced by synaptic activity (Lie et al., 2005; Varela-Nallar and Inestrosa, 2013; Tang, 2014; Inestrosa and Varela-Nallar, 2015) making it perfectly poised to link seizures to NSC dynamics. Wnt dysregulation might contribute to several aspects of AD, including APP processing, tau hyperphosphorylation, blood–brain barrier integrity, inflammation, aberrant neurogenesis, synaptic dysfunction and loss, and apoptosis (Inestrosa and Varela-Nallar, 2014; Tapia-Rojas and Inestrosa, 2018; Palomer et al., 2019; Kostes and Brafman, 2023). Wnt signaling has been found to be attenuated in some studies, but augmented in other studies, of the postmortem tissue from AD patients or mice (Boonen et al., 2009). The contrasting results might in part be due to different brain regions/cell types investigated and/or different timepoints in disease progression. Wnt signaling is also activated in kainic acid-induced epilepsy and contributes to aberrant neurogenesis (Huang et al., 2015; Qu et al., 2017; Hodges and Lugo, 2018). Thus, Wnt signaling in the DG is well positioned to mediate seizure-induced dysregulation of NSC proliferation and neurogenesis in APP mice, which was the subject of this work.
Materials and Methods
Experimental model and subject details
This study used heterozygous transgenic mice (line J20) expressing human APP carrying Swedish (K670N, M671L) and Indiana (V717F) familial AD mutations (hAPP770 numbering) driven by the platelet-derived growth factor β-chain promoter (Mucke et al., 2000). This line has been backcrossed for >10 generations onto a C57BL/6J background using nontransgenic (NTG) C57BL/6J mice from The Jackson Laboratory. Age-matched NTG littermates from the same line were used as controls to the heterozygous APP mice. In addition, heterozygous mice expressing a nuclear-localized β-galactosidase (NLS-lacZ) gene under the control of the endogenous Axin2 promoter/enhancer regions of one Axin2 allele were used to monitor Wnt signaling (JAX stock #009120; Lustig et al., 2002). Male and female mice between 2 weeks and 24 months of age were used. Mice were group-housed with ad libitum access to food and water, in cages with pelleted cellulose bedding and EnviroPak nesting material, and maintained on a regular 12/12 light/dark cycle. No specific method of randomization was used, but mice were semirandomly assigned to experimental groups based on the birth order after balancing for age, sex, and genotype. No sex differences were observed. Experiments were performed by investigators who were blinded to the genotype and treatment of the mice. For harvesting of brains, mice were deeply anesthetized with isoflurane or an overdose of a commercial euthanasia solution and flush-perfused transcardially with ice-cold 0.9% saline. Brains were hemisected and either postfixed in 4% phosphate-buffered paraformaldehyde for immunostaining or flash-frozen in dry ice and stored at −80°C for use in biochemical experiments. All experiments were approved by the Institutional Animal Care and Use Committee of Thomas Jefferson University and Baylor College of Medicine.
Immunohistochemistry
Preparation of brains and brain sections, and immunohistochemistry, was performed as previously described (Corbett et al., 2017; You et al., 2017; Fu et al., 2019). Briefly, fixed brains were cryoprotected in 30% sucrose in phosphate-buffered saline (PBS), and coronal sections (30 μm) were cut on a freezing sliding microtome (Microm). Sections were distributed into 10 subseries, each containing every 10th section throughout the rostral–caudal extent of the brain. Each immunostain was performed on one full subseries of sections. Antigen retrieval was performed as described (Hussaini et al., 2013). Primary antibodies used include mouse anti-nestin (Millipore), mouse anti-Prox1 (PhosphoSolutions), rabbit anti-ΔFosB (Cell Signaling Technology), rabbit anti-doublecortin (Cell Signaling Technology), rabbit anti-Ki67 (Thermo Fisher Scientific), chicken anti-β-galactosidase (Aves Labs), chicken anti-GFP (Abcam), and goat anti-sFRP3 (R&D Systems) antibodies. For avidin-biotin/immunoperoxidase immunohistochemistry, secondary antibodies used include biotinylated donkey anti-mouse, goat anti-rabbit, and rabbit anti-goat antibodies (Vector Laboratories). Signal was amplified using the VECTASTAIN Elite ABC-HRP Kit (Vector Laboratories), and diaminobenzidine was used as the chromogen. For immunofluorescence, goat anti-mouse FITC (Jackson ImmunoResearch Laboratories), goat anti-rabbit Alexa Fluor 594 (Thermo Fisher Scientific), goat anti-mouse Alexa Fluor 488 (Thermo Fisher Scientific), goat anti-mouse Alexa Fluor 594 (Thermo Fischer Scientific), goat anti-rabbit Alexa Fluor 350 (Thermo Fisher Scientific), goat anti-chicken Alexa Fluor 488 (Thermo Fisher Scientific), and goat anti-chicken Alexa Fluor 647 (Thermo Fisher Scientific) antibodies were used. Prolong Diamond Antifade Mountant with or without DAPI (Thermo Fisher Scientific) was used. Zeiss AxioImager Z1 with ApoTome attachment was used for bright-field and fluorescent microscopy unless otherwise stated.
Immunoreactive cells in the subgranular zone of the hippocampus were counted in every 10th coronal section throughout the rostral–caudal extent of the hippocampus and summed by an experimenter blinded to genotype and treatment. NSCs and immature neurons were identified by immunophenotyping based on expression of nestin or doublecortin and on morphology following criteria previously published (Encinas and Enikolopov, 2008; Encinas et al., 2011; Fu et al., 2019).
ImageJ was used to measure mean pixel intensities for sFRP3 and ΔFosB quantifications. sFRP3 immunoreactivity was quantified by taking the mean pixel intensity of manually drawn regions around the granule cell layer (GCL), molecular layer, and hilar regions of the DG and normalized to the intensity of the corpus callosum to account for staining variability. ΔFosB immunoreactivity was quantified by taking the mean pixel intensity of manually drawn regions around the GCL of the DG, normalized to the intensity of staining in the stratum radiatum region of CA1 to account for staining variability.
Data are shown normalized to control groups to illustrate genotype- and/or treatment-specific differences.
RNA-sequencing
RNA-sequencing (RNA-seq) was performed as previously described (Stephens et al., 2020). Approximately 300 ng of RNA extracted from the DG of 4-month-old APP mice with high ΔFosB expression and from wild-type NTG mice (four per genotype, including one female, three males in each genotype) were submitted to the University of Pennsylvania Next-Generation Sequencing Core, where library preparation and Illumina hiSeq 2500 paired-read sequencing (100 bp read-depth) were performed. Data were uploaded to Basepair for analysis. In brief, reads were trimmed, aligned to mouse genome mm9, and counted using STAR and FeatureCounts. Differential expression analyses between genotypes were performed via DEseq2 with sex as a secondary factor.
RNA extraction and RT-qPCR
RNA extraction was performed as previously described (Stephens et al., 2020). The Qiagen RNeasy Mini kit (74106) was used. Briefly, hippocampi were submerged in RLT/β-mercaptoethanol buffer, minced with small scissors, and homogenized by passing the lysate through a 21 G needle 15 times. Samples were centrifuged, and the supernatants were transferred to new tubes, and RNA was then purified according to the kit instructions and eluted with nuclease-free water. Final RNA concentration was determined using a NanoDrop One spectrophotometer. Reverse transcription was performed using the TaqMan Reverse Transcription Reagent kit (ABI, N8080234) in accordance with the manufacturer's instructions, also adding 2.5 μM random hexamers and oligo d(T)16 per reaction (ABI, N8080127 and N8080128). The resulting cDNA was diluted in water and used for quantitative PCR, which was performed with an ABI StepOnePlus machine using SYBR Green (ABI, 4309155) as a fluorophore. Each sample was run in triplicate reactions. The primer sets listed below were used to amplify cDNA of Frzb, Wnt2, Wnt9a, sFRP1, sFRP2, sFRP4, and sFRP5. Gapdh was used as the housekeeping gene. Each primer pair was used at concentrations of 0.5 μM per reaction as follows: Gapdh, forward (F), 5ʹ-AATTCAACGGCACAGTCAAGGC-3ʹ and reverse (R), 5ʹ-TACTCAGCACCGGCCTCACC-3ʹ; Frzb, F, 5ʹ-CAA GGG ACA CCG TCA ATC TT-3ʹ and R, 5ʹ-CAT ATC CCA GCG CTT GAC TT-3ʹ; Wnt2, F, 5ʹ-CTA CTG TAT CAG GGA CCG A-3ʹ and R, 5ʹ-GAT GTG TCA TAG CCT CTC C-3ʹ; Wnt9a, F, 5ʹ-TCA AGT ACA GCA GCA AGT TTG-3ʹ and R, 5ʹ-GGT TTC CAC TCC AGC CT-3ʹ; sFRP1, F, 5’-TGT GTC CTC CAT GCG AC-3’ and R, 5’-CAC TTC TTT GAT TTT CAT CCT CAG-3’; sFRP2, F, 5’-GTG TGA AGC CTG CAA AAC CAA-3’ and R, 5’-CTC TGT TGA TGT ACG TTA TCT CC-3’; sFRP4, F, 5’-AGT GTC CAC ATA TCC TGC C-3’ and R, 5’-TAT GGA CCT TCT ACT GAG TTG-3’; and sFRP5, F, 5’-TGA CCA AGA TCT GTG CCC AGT G-3’ and R, 5’-CCA ATC AAC TTT CGG TCC C-3’.
In situ hybridization (ISH)
ISH was performed as previously described (You et al., 2017). Briefly, sections were digested using 1 μg/ml proteinase K for 12 min and then incubated at 65°C overnight with digoxygenin-labeled full-length antisense riboprobe for mouse Frzb (synthesized from Frzb cDNA, IMAGE catalog #4237551). Sense- and no-probe controls were included. Sections were then washed once with 5× SSC/0.5% Tween-20 and seven times with 0.2× SSC/0.5% Tween-20. Sections were then blocked with 10% heat-inactivated sheep serum and then incubated at 4°C overnight in a 1:5,000 dilution of alkaline phosphatase-conjugated sheep antidigoxygenin antibody (Roche, 11333089001). Development of a blue/purple stain for colorimetric detection was achieved via incubation with the chromogen NBT/BCIP (Roche) for 3 h at room temperature. Sections were then washed in PBS-EDTA and fixed with 4% paraformaldehyde for 10 min before mounting onto slides.
Western blot
Western blot was performed as previously described (Corbett et al., 2013). Hippocampi were subdissected from hemibrains and homogenized with a Polytron tissue homogenizer in ice-cold radioimmunoprecipitation assay buffer. Equal amounts of protein were resolved by SDS-PAGE on 4–12% gels, transferred to nitrocellulose, and probed with goat anti-sFRP3 primary antibody (R&D Systems). IR-dye-conjugated secondary antibodies were used for detection and quantification using a LI-COR Odyssey infrared imaging system.
Chromatin immunoprecipitation (ChIP)
ChIP material was obtained in a previous study (Corbett et al., 2017; You et al., 2017). Briefly, hippocampi were subdissected and fixed in 1% formaldehyde. Samples were sonicated to generate genomic fragments 200–1,000 bp in length and precleared with Protein A beads (Millipore) prior to incubation with rabbit anti-ΔFosB (Cell Signaling Technology), rabbit anti-acetylated histone H4 (acetyl K5 + K8 + K12 + K16, Millipore), or rabbit anti-acetylated histone H3 (acetyl K9 + K14 + K18 + K23 + K27) primary antibodies at 4°C overnight. The antibody–chromatin complex was immunoprecipitated with Protein A beads, washed with a series of buffers (Millipore), and then chromatin was eluted and reverse cross-linking performed with Proteinase K. DNA was purified via phenol-chloroform extraction. Final DNA concentration was measured using the NanoDrop One spectrophotometer. qPCR was performed with an ABI 7500 PCR machine using SYBR green as a fluorophore. The following primer pair was used to amplify the Frzb promoter: F, 5ʹ-GGA GAC ACT TTC GTT CCG-3ʹ and R, 5ʹ-CCA AGA GAA CTG TGA TTG TCC-3ʹ.
Golgi–Cox staining
Golgi–Cox staining was performed by the IDDRC Neuropathology Core at Baylor College of Medicine using the FD Rapid GolgiStain Kit (PK 401, FD Neuro Technologies). Hemibrains were impregnated with Golgi solution for 3 weeks and cut into 50-μm-thick coronal sections for imaging and quantification via the Neurolucida software (MBF Bioscience).
Electroencephalography (EEG)
Mice underwent stereotaxic implantation of a headplate comprising six electrodes for recording and analysis of EEG activity, as previously described (Corbett et al., 2017; Fu et al., 2019). Two stainless steel miniature screws (J.I. Morris) were placed over the left and right frontal cortices (AP +1.5 mm, ML ±1.5 mm), and another was placed over the right parietal cortex (AP −2.2 mm; ML ±2.0 mm). A silver depth electrode (A-M Systems) was placed in the left hippocampus (AP −2.2 mm; ML ±2.0 mm; DV 1.8 mm). A ground screw and a reference silver depth electrode were respectively implanted over and inside the cerebellum. Mice were allowed to recover for at least 4 d prior to the onset of recordings, which were performed in their own home cages with ad libitum access to water and feed. Each mouse was recorded for at least 3 d using a tethered Stellate Harmonie acquisition system (v7.0a, Natus Medical) at a sampling rate of 2,000 Hz. Native Stellate, LabChart Pro (AD Instruments) and Spike2 (v7.20, Cambridge Electronic Design) were used for EEG signal processing and analyses.
Pharmacological treatments
For assessment of the effects of reducing seizures on neurogenesis, levetiracetam (Sequoia Research Products) was dissolved in saline and injected intraperitoneally at a dose of 75 mg/kg, three times per day for 2 weeks. Control groups were administered the equivalent volume of saline. Levetiracetam treatment did not successfully reduce seizures in two APP mice, which were therefore excluded from analysis.
For kainic acid-induced seizures, kainic acid (Sigma-Aldrich) was dissolved in saline and injected intraperitoneally at a dose of 15 mg/kg. Control groups were administered with the equivalent volume of saline. Seizures were behaviorally monitored and scored using a modified Racine scale for the first 2 h postinjection.
For the pilocarpine-induced model of temporal lobe epilepsy and recurrent seizure activity, male and female 3–4-month-old wild-type C57BL/6 mice from Charles River Laboratories were used. Mice were initially administered scopolamine methylnitrate and terbutaline hemisulfate (each 2 mg/kg, s.c.; Sigma-Aldrich) to respectively inhibit peripheral effects of pilocarpine and dilate respiratory tracts. Mice were also injected with ethosuximide, a T-type Ca2+ channel inhibitor (150 mg/kg, s.c.; Sigma-Aldrich), which was found to be effective at reducing mortality after seizure induction in C57BL/6 mice (Iyengar et al., 2015). Thirty minutes after pretreatment, mice were injected with either saline (Sal mice) or pilocarpine hydrochloride (Pilo mice, 240–250 mg/kg, s.c.; Sigma-Aldrich) and behaviorally monitored and scored for the first 2 h postinjection. Mice were then administered diazepam (10 mg/kg, s.c.; Henry Schein) to reduce seizure activity and placed in heated cages. While sedated with diazepam, mice were injected with 5% dextrose-lactated Ringer's solution (1 ml, i.p.; Henry Schein) and then transferred to clean cages the following morning. Mice were euthanized 9–12 weeks later. This timepoint was chosen to ensure that Pilo mice were in the chronic stages of epilepsy, which generally start ∼4 weeks post-Pilo, when spontaneous recurrent seizures manifest (Botterill et al., 2019). The brain tissue was extracted and stored as described above for APP mice. All procedures were performed in accordance with Baylor College of Medicine and Nathan Kline Institute IACUC protocols.
Adeno-associated viruses (AAVs)
AAV2 carrying CMV-ΔFosB-IRES2-eGFP (AAV-ΔFosB) or CMV-eGFP (AAV-GFP) were previously developed and characterized (Robison and Nestler, 2011). Previous experiments have demonstrated that AAV2 is neurotropic and achieves stable neuronal gene expression within 18–22 d of infusion into the brain (Corbett et al., 2017; You et al., 2017; Stephens et al., 2024). Plasmid AAV2-sFRP3 was used for AAV mammalian expression of sFRP3 in vivo. The vector backbone, pENN.AAV.CamKII.eGFP.WPRE, was obtained from the University of Pennsylvania Vector Core. The 1,000 bp DNA fragment upstream of the Prox1 gene identified using the program, SnapGene, was used as the promoter for sFRP3 expression. A self-cleaving small peptide, Thosea Asigna virus 2A (T2A; Liu et al., 2017), was used to coexpress sFRP3 gene (Frzb) and the reporter, eGFP. For control, plasmid AAV2-GFP was made using the same vector backbone and contains the same Prox1 promoter region to drive expression of eGFP. AAV serotype 2 viral particles carrying either Prox1-sFRP3-T2A-eGFP (AAV-sFRP3) or Prox1-eGFP (AAV-GFP) were synthesized by the Optogenetics and Viral Vectors Core at the Jan and Dan Duncan Neurological Research Institute. We noted that when virus was delivered without dilution, there were nonspecific effects on cell division in the DG, but that these effects were avoided when we titrated the virus down to lower concentrations.
Retrovirus-GFP (retro-GFP)
The retrovirus containing a GFP tag was a generous gift from Dr. Jenny Hsieh at the University of Texas at San Antonio.
Stereotactic viral infusions
Either 0.5 or 1 µl of virus solution was stereotactically infused unilaterally or bilaterally into the hippocampus at rostral [−1.7 mm anterior/posterior (A/P), 1.2 mm medial/lateral (M/L), and 2 mm dorsal/ventral (D/V) from the bregma] and caudal (−2.7 mm A/P, 2 mm M/L, and 2.1 mm D/V from the bregma) coordinates.
Delivery of recombinant protein
Delivery of recombinant protein was performed using a protocol adapted from another study (Jang et al., 2013b). Micro-osmotic pumps (model 1002, Alzet) were filled with sterile PBS or recombinant sFRP3 (120 ng/day; R&D Systems) per manufacturer’s instructions, attached to cannulae (Brain Infusion Kit 3, Alzet), and primed overnight prior to implantation in 37°C saline. The cannula was targeted to the right ventricle using the following coordinates: −0.3 mm A/P, 1 mm M/L, and 2.5 mm D/V from the bregma (Jang et al., 2013b). Model 1002 micro-osmotic pumps delivered fluid at a rate of 0.25 μl/h for 14 d.
Spatial discrimination task
The task was performed as previously described (Fu et al., 2019). The experimental apparatus consisted of an empty mouse housing cage placed within a three-sided white enclosure, directly touching two sides. To provide visual cues for spatial orientation, the back wall of the enclosure was striped with black tape, the side wall adjacent to the cage had an A4-sized picture taped to it, and a small box was placed to the left of the mouse cage. Two 25 ml Erlenmeyer flasks were placed equidistant to the two corners of the cage facing the striped wall (Fig. 5A). For the training phase, mice were individually placed in the center of the cage and allowed to freely explore for three, 3-min training sessions separated by 3-min rest periods in their home cages. For the test trial, which took place 3 min after the last training trial, one of the two flasks was displaced to two flask lengths from its original location before the mice were placed back in the cage for the single 3-min testing session. During each trial, the amount of time spent exploring each of the two Erlenmeyer flasks was measured by an experimenter blinded to genotype and treatment. The two flask-length distance was chosen as we had previously demonstrated that naive NTG mice, but not APP mice, were able to discriminate this displacement distance (Fu et al., 2019).
Neuron morphology reconstruction and analysis
Spines of Golgi-stained DG granule cells were manually traced and counted with a 100× oil objective and camera lucida system (MicroBrightField) using the live Trace function in the Neurolucida software. Dendritic segments roughly between 80 and 100 μm in length were selected from the middle and outer molecular layers of the DG. Primary dendrites were not included in the analysis.
Dendritic arbors and spines from retro-GFP-infused newborn neurons were imaged from 160-μm-thick coronal sections using a Zeiss LSM 880 confocal microscope. Images of dendritic arbors were obtained using an LD C-Apochromat 40×/1.1 W Korv M27 objective while operating the microscope in a confocal mode. Images were acquired using a 42 μm pinhole and Z-stacks of a 0.55 μm step size. Dendritic spines were imaged from the molecular layer using a 63× oil objective plan-apochromat 63×/1.4 Oil DIC M27 while operating the microscope in a fast Airyscan mode using Z-stacks of a 0.17 μm step size. Image z-stacks were used to create 3D reconstructions of structures using Filament Tracer in Imaris 10.1 (Oxford Instruments). Dendritic arbor values and spine density were extracted from these reconstructions. Spine density was calculated as spines per micrometer dendrite length for both Golgi-stained and retro-GFP-infused neurons. All neuron morphology was imaged, reconstructed, and analyzed in a blinded manner.
Experimental design and statistical analysis
No specific method of randomization of mice was used, but mice were semirandomly assigned to experimental groups based on birth order after balancing for age, sex, and genotype. Experiments were performed and quantified by investigators who were blinded to the genotype and treatment of the mice and were unblinded once summary data were ready to be prepared.
GraphPad Prism 9.0 was used for statistical analyses. For RNA-seq differential gene expression, DEseq2 was used to perform differential gene expression analysis, and p values were adjusted using the Benjamini–Hochberg (BH) procedure. Genes with p-adjusted < 0.05 were considered statistically significant. For comparisons between two experimental groups, unpaired two-tailed Student’s t tests were used. For comparisons between more than two experimental groups, a two-way ANOVA test (when there was normal sample distribution) was used, with repeated measures (RM) when appropriate, followed by multiple-comparison post hoc analyses to compare the differences between individual groups. For correlation between seizure frequency and sFRP3 immunoreactivity, a linear regression was used. A p value of <0.05 was considered statistically significant. The statistical tests, n, and p values for each dataset are provided in the figure legend that accompanies the data.
Results
Regulators of neurogenesis in the Wnt signaling pathway are altered in APP mice
AHN is exquisitely sensitive to environmental cues, and there are a multitude of physiological and pathological stimuli that can affect it. We previously found that the spontaneous seizures exhibited by APP mice aberrantly increased proliferation of adult hippocampal NSCs, thus accelerating the use and subsequent depletion of the NSC pool (Fu et al., 2019). To identify which molecular regulators of AHN might mediate seizure-driven dysregulation of AHN in APP mice, we queried a previously published bulk RNA-seq dataset that compares differential gene expression between the microdissected DG tissue of APP mice with that of NTG mice (Stephens et al., 2020). The APP mice were previously characterized as having high levels of expression of ΔFosB, a seizure-induced transcription factor, in the granule cells of the contralateral DG, which suggest that these mice had robust seizures in the weeks prior to sacrifice (Corbett et al., 2017; You et al., 2017; Stephens et al., 2020). We focused on genes involved in the Wnt signaling pathway that controls cell proliferation to determine which were altered in APP mice with seizures compared with NTG mice (Fig. 1A–C). Using DEseq2 to assess differential gene expression, we found that with a significance threshold of p-adjusted < 0.05 after BH correction, only three genes in the pathway were significantly altered in APP mice: Wnt2 (upregulated; p-adjusted = 0.013), Wnt9a (downregulated; p-adjusted = 2.93 × 10−5), and Frzb (downregulated; p-adjusted = 0.0002; Fig. 1C,D).
Regulators of neurogenesis in the Wnt signaling pathway are altered in APP mice. A, Simplified diagram of the Wnt/β-catenin signaling pathway and inhibitors of the pathway. Wnt ligands are secreted proteins that bind to Frizzled receptors (Fz) and lipoprotein receptor-related protein (LRP) coreceptors to initiate an intracellular signaling cascade, involving Disheveled (Dsh) as a mediator, and leading to accumulation of β-catenin (β-cat) in the cytosol and its translocation into the nucleus, where it interacts with T-cell factor/lymphoid enhancer factor (TCF/LEF) transcription factors to turn on gene transcription. Wnt signaling can be regulated by an array of inhibitors, such as Wise/sclerostin (SOST), Dickkopf proteins (DKKs), Wnt inhibitory factor 1 (Wif-1), insulin-like growth factor binding protein 4 (IGFBP-4), and secreted Frizzled-related proteins (sFRPs). B, Table of genes in the Wnt/β-catenin pathway that were queried in the RNA-seq dataset. Highlighted genes indicate genes that were significantly down- (blue) or upregulated (red) in APP mice relative to NTG mice. C, A volcano plot showing genes in the DG that were differentially expressed between 4-month-old NTG mice and APP mice that had seizures. Differential gene expression was assessed using DEseq2. The horizontal dashed line indicates significance cutoff of p-adjusted < 0.05. The vertical dashed line separates downregulated (left) and upregulated (right) genes. Components of the Wnt/β-catenin signaling pathway are highlighted in green (p-adj < 0.05) and black (p-adj > 0.05). N = 4 mice per genotype. D, Normalized read counts from the RNA-seq dataset of Wnt2, Wnt9a, and Frzb in NTG mice and APP mice with seizures. E, Fold change of Wnt2, Wnt9a, and Frzb mRNA expression in hippocampal samples from NTG mice and APP mice from an independent 4-month-old cohort. N = 8 mice per genotype. F, mRNA expression of sFRP1, sFRP2, Frzb (sFRP3), sFRP4, and sFRP5 in NTG mice and APP mice, normalized to sFRP1 levels in NTG mice. N = 7–9 mice per genotype. *p < 0.05; ***p < 0.001. Values indicate mean ± SEM.
Wnt2 can be regulated by neuronal activity, is involved in stimulating dendritic arborization, and has been implicated in depression and autism (Wayman et al., 2006). However, Wnt2 had relatively low normalized read counts in our samples (Fig. 1D). When we used benchtop RT-qPCR to validate these results in the whole hippocampal tissue of an independent cohort of NTG and APP mice, the difference between genotypes was masked (t(16) = 0.212; p = 0.835; two-tailed unpaired t test; Fig. 1E).
In contrast, Wnt9a and Frzb were more abundantly expressed in the hippocampus and were significantly downregulated in APP mice in both DG (as in Fig. 1C,D) and whole hippocampal samples (Wnt9a, t(16) = 2.612; p = 0.019; Frzb, t(16) = 4.276; p = 0.0006; two-tailed unpaired t test; Fig. 1E). Wnt9a has been found to be involved in hematopoietic stem-cell development and cochlear patterning during development, but there is not much currently known about the role of Wnt9a in the adult hippocampus (Munnamalai et al., 2017; Richter et al., 2018).
Frzb encodes secreted frizzled-related protein 3 (sFRP3), an inhibitor of the Wnt signaling pathway that is normally constitutively expressed and secreted by mature dentate granule cells (Jang et al., 2013b). Notably, sFRP3 was previously found to regulate activity-dependent AHN in wild-type mice (Jang et al., 2013b). Specifically, it was reported that sFRP3 expression was decreased by neuronal activity, and this reduction was found to be necessary for activity-induced neural progenitor proliferation (Jang et al., 2013b). Similarly, APP mice with seizures had decreased expression of Frzb mRNA compared with NTG mice (Fig. 1C–E), and this decrease was specific to Frzb and not to other members of the sFRP family (two-way RM ANOVA revealed effects of gene, F(1.440,20.16) = 708.5; p < 0.0001; genotype, F(1,14) = 8.182; p = 0.013; and gene × genotype interaction, F(4,56) = 13.10; p < 0.0001; Bonferroni’s post hoc tests sFRP1, p > 0.999; sFRP2, p = 0.663; Frzb, p = 0.013; sFRP4, p = 0.065; sFRP5, p > 0.999; Fig. 1F). These findings suggested that regulation of sFRP3 expression might be an important factor in the seizure-induced dysregulation of neurogenesis in APP mice.
sFRP3 is decreased in APP mice in a seizure-dependent manner
To test whether sFRP3 might mediate seizure-induced proliferation of NSCs in APP mice, we first assessed sFRP3 mRNA expression at various ages in NTG and APP mice. In this line of APP mice, spontaneous epileptic activity is evident by 1 month of age, which corresponds with increased NSC division; seizures become abundant by 2 months of age; cognitive deficits become apparent by 3 months of age; and plaque deposition begins ∼6 months of age (Fu et al., 2019). We found that sFRP3 expression in the hippocampus was reduced by 1 month of age in APP mice and remained consistently decreased with age (two-way ANOVA revealed effects of age, F(6,109) = 2.302; p = 0.0394; genotype, F(1,109) = 87.81; p < 0.0001; and age × genotype interaction, F(6,109) = 2.302; p = 0.0394; Holm–Sidak post hoc tests 0.5 month, p = 0.084; 1 month, p = 0.054; 2 months, p = 0.004; 4 months, p = 0.010; 8 months, p = 0.0008; 16 months, p < 0.0001; 24 months, p < 0.0001; Fig. 2A). We used in situ hybridization to confirm that sFRP3 mRNA (Frzb) was expressed in the dentate GCL and was reduced in APP mice (Fig. 2B). Reduced expression of sFRP3 in the hippocampus was also confirmed at the protein level (t(12) = 2.287; p = 0.041; two-tailed unpaired t test; Fig. 2C). In APP mice that received EEG monitoring of seizures, higher seizure frequency corresponded with lower immunoreactivity of sFRP3 in the GCL and the hilus (GCL, F(1,12) = 8.854; p = 0.012; hilus, F(1,12) = 10.20; p = 0.008; linear regression; Fig. 2D). To test whether seizures were necessary for the reduction in sFRP3 expression in APP mice, we assessed Frzb expression in the hippocampus of NTG and APP mice that had been treated with either saline or levetiracetam (LEV, 75 mg/kg), an antiseizure medication that is efficacious at reducing seizure activity in APP mice (Sanchez et al., 2012; Corbett et al., 2017; Fu et al., 2019). APP and NTG mice received intraperitoneal injections of saline or LEV three times a day for 2 weeks, from 1.5 to 2 months of age. We previously found that this treatment regimen was sufficient to prevent the aberrant increase in NSC proliferation in APP mice (Fu et al., 2019). Indeed, LEV treatment restored Frzb expression in APP mice (two-way ANOVA revealed effects of genotype, F(1,16) = 26.93; p < 0.0001; treatment, F(1,16) = 5.637; p = 0.0304; and genotype × treatment interaction, F(1,16) = 4.872; p = 0.0423; Tukey's post hoc tests NTG saline vs NTG LEV, p = 0.999; NTG saline vs APP saline, p = 0.0002; NTG saline vs APP LEV, p = 0.312; NTG LEV vs APP saline, p < 0.0001; NTG LEV vs APP LEV, p = 0.242; APP saline vs APP LEV, p = 0.037; Fig. 2E).
sFRP3 is decreased in APP mice in a seizure-dependent manner. A, Frzb expression in whole hippocampal samples of NTG mice and APP mice at various months of age (m), normalized to NTG Frzb expression at each age. N = 13 NTG, 7 APP (0.5 m); 14 NTG, 7 APP (1 m); 7 NTG, 9 APP (2 m); 8 NTG, 8 APP (4 m); 8 NTG, 8 APP (8 m); 13 NTG, 12 APP (16 m); 4 NTG, 5 APP (24 m). B, In situ Frzb expression in the DG of 2-month-old NTG and APP mice. C, Western blot of sFRP3 in hippocampal lysates of 4-month-old NTG mice and APP mice (left), quantified (right). GAPDH was used as a housekeeping gene. Data are normalized to expression in NTG mice. N = 7 mice per genotype. D, sFRP3 immunoreactivity (IR) in the DG molecular layer (ML), GCL and hilus of APP mice (left), compared with seizure frequency using regression analysis (right). N = 14 mice. E, Frzb expression in the hippocampus of NTG mice and APP mice given three daily intraperitoneal (i.p.) injections of either saline (Sal) or levetiracetam (LEV, 75 mg/kg) for 2 weeks. N = 36 mice per genotype and treatment group. F, Hippocampal Frzb expression in NTG mice 2 h, 4.5 h, 1 d, 3 d, or 7 d after intraperitoneal injection of either Sal or kainic acid (KA, 15 mg/kg, i.p.). N = 6 Sal, 10 KA (2 h); 4 Sal, 8 KA (4.5 h); 13 Sal, 16 KA (1 d); 6 Sal, 7 KA (3 d); 8 Sal, 10 KA (7 d). G, Hippocampal Frzb expression in NTG mice 9–12 weeks post-treatment with either saline or pilocarpine (Pilo, 240–250 mg/kg, subcutaneous). N = 8 mice per treatment group. H, Normalized read counts of Frzb mRNA in the DG of NTG mice (normal ΔFosB) and APP mice with seizures (high ΔFosB) from the RNA-seq dataset in Figure 1B,C. APP mice that had low frequency of seizures (N = 4 mice) and therefore exhibited normal levels of ΔFosB in the DG were also assessed. I, Left, images of ΔFosB IR in two different APP mice with differing ΔFosB and Frzb expression levels. Right, regression analysis comparing ΔFosB IR with Frzb expression in NTG mice and APP mice. N = 7–8 mice per genotype. #p = 0.05; *p < 0.05; **p < 0.01; ***p < 0.001. Values indicate mean ± SEM.
To test whether seizures are sufficient to induce the suppression of sFRP3 expression even in wild-type mice, we examined hippocampal Frzb expression in wild-type mice treated with chemoconvulsants to induce status epilepticus (SE) and chronic epilepsy (Fig. 2F,G). A single injection of an SE-inducing dose of KA (15 mg/kg) decreased Frzb mRNA expression at 4.5 h post-injection, and Frzb mRNA remained decreased for several days before returning to the baseline 7 d after KA (two-way ANOVA revealed effects of time postinjection, F(4,78) = 3.546; p = 0.0104; treatment, F(1,78) = 25.71; p < 0.0001; and time × treatment interaction, F(4,78) = 3.601; p = 0.0095; Holm–Sidak’s post hoc tests 2 h, p = 0.830; 4.5 h, p = 0.008; 1 d, p < 0.0001; 3 d, p = 0.028; 7 d, p = 0.803; Fig. 2F). To determine whether chronic recurrent seizures, similar to those observed in APP mice, can reduce Frzb expression over a longer period of time, we used the pilocarpine model of epilepsy. Pilocarpine induces spontaneous recurrent seizures after an initial latency period that lasts for several weeks (Turski, 2000; Botterill et al., 2019). We assessed hippocampal Frzb mRNA 9–12 weeks after pilocarpine treatment, when chronic epilepsy has typically developed, and found that sFRP3 expression was indeed reduced in pilocarpine-treated mice compared with saline-treated control mice (t(14) = 3.553; p = 0.003; two-tailed unpaired t test; Fig. 2G).
Together, these data suggest that seizures are critical for the suppression of Frzb expression in APP mice. In support of this point, we revisited our RNA-seq dataset, but this time, we also examined the normalized read counts of Frzb mRNA from APP mice that did not exhibit high expression of seizure-induced ΔFosB, indicating low levels of seizure activity (Corbett et al., 2017). Whereas APP mice with high ΔFosB expression had reduced sFRP3 read counts compared with NTG controls, APP mice that did not exhibit high expression of seizure-induced ΔFosB instead had Frzb read counts comparable to that of NTG controls (one-way ANOVA revealed significant difference among means, F(2,9) = 11.55; p = 0.003; Tukey's post hoc tests NTG vs APPhigh ΔFosB, p = 0.003; NTG vs APPnormal ΔFosB, p = 0.357; APPhigh ΔFosB vs APP normal ΔFosB, p = 0.025; Fig. 2H). We also found on a mouse-by-mouse basis that APP mice with higher ΔFosB immunoreactivity in the GCL of the DG had correspondingly lower hippocampal Frzb mRNA expression (F(1,6) = 16.06; p = 0.007; linear regression; Fig. 2I), indicating that Frzb expression in APP mice is closely tied to both seizure activity and ΔFosB expression.
sFRP3 is epigenetically regulated by ΔFosB
sFRP3 expression is regulated by neuronal activity, but the upstream molecular regulators of sFRP3 expression are thus far unknown. Since sFRP3 expression levels corresponded to both seizure frequency and levels of ΔFosB (as shown in Fig. 2D,I), we investigated this relationship further. Notably, ΔFosB is a seizure-induced transcription factor that epigenetically regulates the expression of other activity-dependent genes in the DG of APP mice and pilocarpine-treated mice (Corbett et al., 2017; You et al., 2017; You et al., 2018; Stephens et al., 2020). ΔFosB is also expressed in the mature dentate granule cells that produce sFRP3, and not in the NSCs, as shown by the lack of colabeling between ΔFosB and nestin (Fig. 3A). This expression pattern is consistent with that of sFRP3, which is also expressed by mature dentate granule cells (as shown in Fig. 2B) and was shown in another study to be excluded from nestin+ cells (Jang et al., 2013b). Thus, we hypothesized that ΔFosB might epigenetically regulate expression of sFRP3 in the DG.
sFRP3 is epigenetically regulated by ΔFosB. A, Example images of Nestin (green) and ΔFosB (red) colabeling in the DG of NTG mice and APP mice. Arrows indicate a Nestin + NSC that is negative for ΔFosB labeling. B, Binding of ΔFosB to the Frzb promoter in the hippocampus of NTG mice and APP mice. N = 8 mice per genotype. C, D, Levels of histone H4 (C) and H3 (D) lysine acetylation on the Frzb (sFRP3 gene) promoter in the hippocampus of NTG mice and APP mice. N = 7–8 mice per genotype. E, Binding of ΔFosB to the Frzb promoter in the hippocampus of mice 3 d after treatment with either saline (Sal) or pilocarpine (Pilo, 240–250 mg/kg, subcutaneous). N = 4 mice per treatment group. F, Levels of histone H4 lysine acetylation on the Frzb promoter in the hippocampus of mice 3 d after treatment with either Sal or Pilo. N = 3–4 mice per treatment group. G, NTG mice were given bilateral intrahippocampal infusions of either AAV-GFP or AAV-ΔFosB. Hippocampal Frzb expression was assessed using RT-qPCR one month later. N = 5–8 mice per virus treatment. H, I, Example images of DCX immunostaining (H) and counts of DCX + cells (I) in the DG of NTG mice that were administered unilateral intrahippocampal infusions of AAV-GFP in one hemibrain and AAV-ΔFosB in the other hemibrain 1 month prior to assessment. Quantification is normalized to the cell counts in AAV-GFP-treated hemibrains. *p < 0.05; **p < 0.01. Values indicate mean ± SEM.
To test this hypothesis, we used our previously published ΔFosB ChIP-sequencing datasets to look for putative ΔFosB binding regions in APP mice and pilocarpine-treated mice (Stephens et al., 2020). We identified a region within the Frzb promoter that showed a selective and robust peak in pilocarpine-treated mice and performed ChIP on the hippocampal tissue from NTG and APP mice, as well as from saline- and pilocarpine-treated wild-type mice, to test for ΔFosB enrichment at this region. Consistent with our hypothesis, ΔFosB enrichment at this site on the Frzb gene was greater in APP mice than in NTG mice (t(14) = 2.551; p = 0.023; two-tailed unpaired t test; Fig. 3B).
One mechanism by which ΔFosB downregulates target gene expression is by binding to the target gene and recruiting histone deacetylase 1 (HDAC1) to deacetylate histones around the target gene (Renthal et al., 2008). ΔFosB regulates Fos and Calb1 expression via this mechanism in the hippocampus (Corbett et al., 2017; You et al., 2017), and we hypothesized that it similarly regulates Frzb expression. We therefore assessed histone acetylation around the Frzb gene. Similar to our previous findings with Fos and Calb1, we found hypoacetylated lysine residues on histone H4 (t(12) = 2.372; p = 0.035; two-tailed unpaired t test), but not histone H3 (t(13) = 1.777; p = 0.099; two-tailed unpaired t test), near the Frzb promoter in APP mice compared with NTG controls (Fig. 3C,D) that also corresponded with reduced target gene expression. In pilocarpine-treated mice compared with saline-treated controls, we also found a trend for increase in ΔFosB binding at the Frzb promoter (t(6) = 2.289; p = 0.062; two-tailed unpaired t test; Fig. 3E) and reduced acetylation at histone H4 (t(5) = 3.269; p = 0.022; two-tailed unpaired t test; Fig. 3F). These results support the hypothesis that seizures drive ΔFosB-induced epigenetic regulation of Frzb expression.
To directly test whether ΔFosB is sufficient to reduce Frzb expression, we used AAV to overexpress ΔFosB under the control of the cytomegalovirus (CMV) promoter in the hippocampi of wild-type mice. When compared with mice that received AAV-GFP control virus, mice treated with AAV-ΔFosB had lower levels of Frzb expression in the hippocampus (t(11) = 3.229; p = 0.008; two-tailed unpaired t test; Fig. 3G). In addition, AAV-ΔFosB increased the numbers of newborn immature neurons in the DG (t(6) = 3.301; p = 0.016; two-tailed unpaired t test; Fig. 3H,I). Together, these data support the hypothesis that ΔFosB epigenetically regulates sFRP3 expression in the DG.
Increasing sFRP3 in APP mice prevents accelerated depletion of the NSC pool
sFRP3 is an inhibitor in the Wnt signaling pathway, and knocking it out in wild-type mice increases NSC proliferation and neurogenesis, indicating that sFRP3 functions as a “brake” on neurogenesis (Jang et al., 2013b). Since we found that APP mice have decreased expression of sFRP3 (as shown in Fig. 2) and abnormally increased percentage of Ki67 + dividing NSCs that correspond with accelerated use and depletion of the NSC pool (Fu et al., 2019), we hypothesized that APP mice have an abnormal loss of this “brake” on neurogenesis. To determine whether loss of sFRP3 contributes to the aberrant NSC proliferation in APP mice, we treated NTG and APP mice with recombinant sFRP3 using cannula-driven intracerebroventricular delivery for 2 weeks, starting at 1 month of age, when we first observe epileptic spikes in APP mice (Fig. 4A). APP mice treated with vehicle (PBS) had increased percentage of dividing NSCs (dividing NSCs divided by total NSCs) compared with NTG mice (two-way ANOVA revealed effects of treatment, F(1,21) = 4.797; p = 0.0399; and genotype × treatment interaction, F(1,21) = 6.643; p = 0.0176; Tukey's post hoc test NTG-PBS vs APP-PBS p = 0.030; Fig. 4B), similar to what we previously reported (Fu et al., 2019). In contrast, APP mice treated with recombinant sFRP3 had levels of NSC division comparable to that of NTG controls (Tukey's post hoc test APP-PBS vs APP-sFRP3, p = 0.009).
Increasing sFRP3 in APP mice prevents accelerated depletion of the NSC pool. A, Experiment timeline of intracerebroventricular delivery of PBS or recombinant sFRP3 in NTG and APP mice. B, Nestin and Ki67 colabeling (left) and quantification of percentage of dividing NSCs out of total NSCs (right) in each genotype-treatment group. N = 5–8 mice per genotype and treatment. C, Example images of AAV-GFP virus expression in the DG (top) and of anti-sFRP3 immunostaining in mice infused with either AAV-GFP (middle) or AAV-sFRP3 (bottom). D, E, Unilateral intrahippocampal infusion of AAV-GFP or AAV-sFRP3 for 3 weeks increases hippocampal Frzb mRNA (D) and sFRP3 protein (E). N = 3 hemibrains per virus treatment. F, Experimental timeline of Axin2-lacZ mice administered unilateral intrahippocampal infusions of AAV-GFP and AAV-sFRP3 in opposite hemibrains and then given a single injection of KA (15 mg/kg, i.p.) 19 d after virus infusion to induce SE and cell division. G, Dividing NSCs and β-gal+ dividing NSCs in the DG of mice described in F were identified using Nestin, Ki67, and β-gal immunostaining (left; dotted outlines indicate Ki67+ cell locations) and quantified (right). N = 3 hemibrains per virus treatment. H, Experimental timeline of 1-month-old NTG and APP mice that received bilateral intrahippocampal infusions of either AAV-GFP or AAV-sFRP3 and then tested for spatial discrimination at 3.5 months of age. N = 6–9 mice per genotype and treatment group. I, Example images of Nestin and Ki67 labeling (left) and quantification of percentage of dividing NSCs (right) in the DG of mice described in H. J, Example images of Nestin labeling (left) and quantification of total numbers of NSCs (right) in the DG of mice described in H. *p < 0.05; **p < 0.01; ***p < 0.001. Values indicate mean ± SEM.
We previously found that the accelerated use and depletion of NSCs that lead to altered neurogenesis in APP mice were associated with deficits in spatial discrimination, a neurogenesis-dependent task (Fu et al., 2019). These changes were reversed by treatment with LEV, an antiseizure medication (Fu et al., 2019). To determine whether restoring levels of sFRP3 in the hippocampus in APP mice would similarly normalize the number of dividing NSCs and improve spatial discrimination, we turned to an AAV-driven approach to drive longer-lasting increases in sFRP3 expression. We designed an AAV2 vector containing recombinant sFRP3 with an eGFP reporter after a T2A construct under the Prox1 promoter (“AAV-sFRP3”). Prox1 is expressed in the mature granule cells of the DG, which are the same cells that constitutively express sFRP3. Prox1-targeted AAV should thus more closely mimic endogenous patterns of sFRP3 expression than one with a universal promoter. A construct with only eGFP, and no sFRP3 or T2A, was used as a control virus (“AAV-GFP”). We achieved robust expression in the DG using this virus (Fig. 4C, top). In addition, immunostaining for sFRP3 in AAV-GFP- and AAV-sFRP3–infused DGs revealed overexpression of sFRP3 only in DGs that received AAV-sFRP3 and not AAV-GFP (Fig. 4C, middle and bottom). AAV-sFRP3 increased sFRP3 expression in the hippocampus by ∼20-fold at the mRNA level (t(4) = 4.150; p = 0.014; two-tailed unpaired t test; Fig. 4D) and by ∼4-fold at the protein level (t(4) = 6.432; p = 0.003; two-tailed unpaired t test; Fig. 4E).
To determine whether AAV-sFRP3 produces functional sFRP3 that decreases cell division similar to recombinant sFRP3 and, if so, whether it might do so through altering Wnt signaling, we used the Axin2-lacZ Wnt signaling reporter mice. These mice express lacZ knocked into one endogenous allele of Axin2, which is a target of Wnt signaling (Lustig et al., 2002). In the presence of Wnt signaling, β-galactosidase (β-gal), is produced and can be visualized via immunostaining. In the DG of Axin2-lacZ mice, β-gal is widely expressed in fully mature, calbindin + granule cells but only found in ∼30% of nestin+ cells (Heppt et al., 2020). Because we were interested in the effect of sFRP3 on Wnt signaling in the NSCs, we specifically focused on nestin+ β-gal+ cells in our analysis. To test whether AAV-sFRP3 is functionally active in decreasing cell division, we injected AAV-GFP and AAV-sFRP3 unilaterally into opposite hemibrains of Axin2-lacZ mice and allowed the virus to express for 19 d (Fig. 4F). Because overexpression of sFRP3 does not lead to reductions in cell division under the baseline, unstimulated conditions in which sFRP3 is naturally abundant (Jang et al., 2013b), we injected mice with one dose of KA (15 mg/kg, i.p.) to induce seizures and cell division. We assessed total NSC division 3 d post-KA administration by counting the number of Ki67 + nestin + NSCs in each virus treatment condition and found that AAV-sFRP3 reduced KA-induced NSC division compared with AAV-GFP (t(2) = 7.794; p 0.016; two-tailed unpaired t test; Fig. 4G, graph, left). In addition, of the NSCs that were dividing, the number of β-gal + NSCs was also reduced by AAV-sFRP3 (t(2) = 5.047; p 0.037; two-tailed unpaired t test), suggesting that sFRP3 effectively inhibited Wnt signaling (Fig. 4G, graph, right).
We treated NTG and APP mice with bilateral intrahippocampal infusions of either AAV-GFP or AAV-sFRP3 at 1 month of age (Fig. 4H), which was the earliest timepoint at which sFRP3 was reduced in APP mice. The virus was allowed to express until mice reached 3.5 months of age, the age at which APP mice typically exhibit robust deficits in spatial discrimination (Fu et al., 2019). We found that as expected, when compared with NTG controls, AAV-GFP-treated APP mice had increased NSC proliferation (two-way ANOVA revealed effects of treatment, F(1,27) = 13.41; p = 0.0011; and genotype × treatment interaction, F(1,27) = 15.68; p = 0.0005; Tukey's post hoc tests APP-GFP versus NTG-GFP, p = 0.001, vs NTG-sFRP3, p = 0.003), as well as significant reduction in total number of NSCs (two-way ANOVA revealed effects of genotype, F(1,27) = 10.13; p = 0.0037; treatment, F(1,27) = 8.391; p = 0.0074; and genotype × treatment interaction, F(1,27) = 7.653; p = 0.0101; Tukey's post hoc tests APP-GFP vs NTG-GFP, p = 0.001, vs NTG-sFRP3, p = 0.0008; Fig. 4I,J). However, AAV-sFRP3-treated APP mice showed comparable levels of NSC proliferation (Tukey's post hoc tests APP-sFRP3 vs NTG-GFP, p = 0.592, vs NTG-sFRP3, p = 0.468) and total number of NSCs (Tukey's post hoc tests APP-sFRP3 vs NTG-GFP, p = 0.997; vs NTG-sFRP3, p = 0.992) to NTG controls, suggesting that sFRP3 expression prevented the aberrant increase in NSC division and protected the NSC pool from premature depletion.
The progeny from NSC divisions differentiate into neuroblasts and, eventually, immature neurons. We thus used DCX immunostaining to label and quantify neuroblasts and immature neurons. To our surprise, we found that treatment with AAV-sFRP3 did not affect the number of DCX+ immature neurons and had a mild negative effect on the number of neuroblasts. Compared with APP mice that received AAV-GFP, APP mice that received AAV-sFRP3 had similar numbers of neuroblasts (normalized to average of NTG-GFP, NTG-GFP = 1.00 ± 0.03; APP-GFP = 0.52 ± 0.05; NTG-sFRP3 = 0.86 ± 0.03; APP-sFRP3 = 0.45 ± 0.09; two-way ANOVA revealed effects of genotype, F(1,27) = 81.32; p < 0.0001; and treatment, F(1,27) = 4.537; p = 0.0424; Tukey's post hoc tests NTG-GFP vs NTG-sFRP3, p = 0.125; NTG-GFP vs APP-GFP, p < 0.0001; NTG-GFP vs AAV-sFRP3, p < 0.0001; NTG-sFRP3 vs APP-GFP, p = 0.0002; NTG-sFRP3 vs APP-sFRP3, p < 0.0001; APP-GFP vs APP-sFRP3, p = 0.830) and immature neurons (normalized to average of NTG-GFP, NTG-GFP = 1.00 ± 0.04; APP-GFP = 0.64 ± 0.12; NTG-sFRP3 = 0.88 ± 0.03; APP-GFP = 0.53 ± 0.11; two-way ANOVA revealed effects of genotype, F(1,27) = 22.10; p < 0.0001; Tukey's post hoc tests NTG-GFP vs NTG-sFRP3, p = 0.641; NTG-GFP vs APP-GFP, p = 0.010; NTG-GFP vs AAV-sFRP3, p = 0.001; NTG-sFRP3 vs APP-GFP, p = 0.121; NTG-sFRP3 vs APP-sFRP3, p = 0.016; APP-GFP vs APP-sFRP3, p = 0.766).
Increasing sFRP3 in APP mice normalizes spatial discrimination ability
Adult-born neurons are critical for behavioral pattern separation and spatial discrimination, which is the ability to distinguish between similar contexts (Clelland et al., 2009; Sahay et al., 2011; Nakashiba et al., 2012). To assess whether AAV-sFRP3 also improved neurogenesis-dependent behavior, we tested NTG and APP mice in a spatial discrimination task (Fig. 5A). In this task, mice are trained in an arena with two identical objects (empty Erlenmeyer flasks) for three trials of 3 min each, with a 3 min interval between trials that the mice spend in their home cage. After the last training trial, mice are placed back into their home cage for 3 min, and one object is displaced (displaced object, DO) to two flask lengths away from the original position (Position 2, P2). We previously found that unlike NTG mice, 3.5-month-old naive APP mice were unable to discriminate object displacement to P2 and that LEV treatment rescued this deficit (Fu et al., 2019). Similar to treatment with LEV, increasing sFRP3 through expression by AAV-sFRP3 in the hippocampus enabled APP mice to discriminate object displacement to P2 as well as NTG mice (training vs testing trials for NTG-GFP, t(14) = 4.237; p = 0.0008; NTG-sFRP3, t(14) = 4.700; p = 0.0003; APP-sFRP3, t(10) = 3.172; p = 0.010; two-tailed paired t test; Fig. 5B). In contrast, APP mice that received AAV-GFP infusion were not able to discriminate object displacement to P2 (t(8) = 0.890; p = 0.399; two-tailed paired t test; Fig. 5B). These results suggest that restoring sFRP3 expression in APP mice improves spatial discrimination ability.
Increasing sFRP3 in APP mice normalizes spatial discrimination and spine density in adult-born neurons. A, Spatial discrimination task. Mice are trained with two identical objects placed on one side of the cage for three training trials of 3 min each, with 3 min delays between trials. Three minutes after the last training trial, one object is displaced to P2, and the mouse is placed back in for a 3 min test trial. B, The percentage of time spent with the displaced object (DO) at P2 in virus-treated NTG and APP mice. N = 9–15 mice per genotype and treatment group. C, Experimental timeline of NTG mice and APP mice treated unilaterally with a cocktail of retro-GFP and AAV-CMV-mCherry (AAV-mCh) and retro-GFP and AAV-CMV-sFRP3-mCherry (AAV-sFRP3) in opposite hemibrains. Mice were euthanized after 18 d (data shown in D) or 6 weeks (data shown in E–F) postviral infusion. D, Example images of GFP-labeled dendrite segments (left) and quantification of spine density (right) in NTG and APP mice euthanized 18 d postviral infusion, as described in C. N = 19–55 segments per genotype and treatment group from 3–4 mice per genotype. Representative images were masked to remove dendrites of neighboring neurons. E, Example images of GFP-labeled dendrite segments (left) and quantification of spine density (right) in NTG and APP mice euthanized 6 weeks postviral infusion, as described in C. N = 48–79 segments per genotype and treatment group from 4–6 mice per genotype. F, Example images (left) and Sholl analysis (right) of DG granule cell dendrites from NTG and APP mice euthanized 6 weeks postviral infusion, as described in C. N = 47–90 cells per genotype and treatment group from 4–6 mice. Representative images were masked to remove neighboring neurons. G, Example images of Golgi-impregnated cells in the DG. H, Golgi-impregnated granule cell dendrite segments in the DG molecular layer (left) were analyzed for spine density (right) in NTG and APP mice treated with either AAV-GFP or AAV-sFRP3 for 2.5 months. N = 3–6 mice per genotype and treatment group. I, Prox1 expression was used to identify ectopic granule cells in the hilus in NTG and APP mice treated with either AAV-GFP or AAV-sFRP3, as described in Figure 4H. N = 6–9 mice per genotype and treatment group. *p < 0.05; **p < 0.01; ***p < 0.001. Values indicate mean ± SEM. Scale bar, 5 µm (D, E, H), 20 µm (F).
Increasing sFRP3 in APP mice normalizes spine density in adult-born neurons
Although AAV-sFRP3 normalized NSC division and prevented the aberrant loss of NSCs in APP mice, it did not affect numbers of neuroblasts or immature neurons, suggesting that the improved spatial discrimination performance was not due to increased quantity of newborn neurons. We hypothesized that rather than normalizing quantity, AAV-sFRP3 instead affected the development of the newborn neurons. Development of adult-born granule cells in APP mice is aberrantly accelerated at early stages of cell maturation, with newborn cells in APP mice showing increased spine density compared with those in NTG mice (Sun et al., 2009). However, development is deficient at later cell maturation stages in APP mice, with mature newborn neurons showing decreased spine density as well as reduced dendritic tree size compared with those in NTG mice (Sun et al., 2009). sFRP3 has also been reported to be involved in slowing down the development of newborn cells and that knocking it out results in increased spine density at early stages of cell maturation (Jang et al., 2013b). To assess whether AAV-sFRP3 altered spine density of newborn neurons, we infused into NTG and APP mice a 1:1 cocktail of a retrovirus containing a GFP construct and AAV-mCherry unilaterally in one hippocampus versus retro-GFP and AAV-sFRP3-mCherry in the contralateral hippocampus and examined spine density of GFP-labeled newborn cells after either 18 d or 6 weeks of virus expression (Fig. 5C–E). We found that 18-d-old newborn cells in APP mice that received AAV-mCherry showed increased spine density compared with NTG mice, consistent with prior reports (Sun et al., 2009); however, this aberrant increase was notably prevented in newborn cells from APP mice that received AAV-sFRP3-mCherry (two-way ANOVA revealed significant effects of genotype, F(1,135) = 22.90; p < 0.0001; treatment, F(1,135) = 40.50; p < 0.0001; and genotype × treatment interaction, F(1,135) = 7.233; p = 0.0081; Fisher's LSD post hoc tests NTG-mCherry vs APP-mCherry, p < 0.0001; NTG-sFRP3 vs APP-sFRP3, p = 0.068; Fig. 5D). When we examined GFP-labeled 6-week-old newborn neurons, we found that APP mice that received AAV-mCherry showed significantly decreased spine density compared with NTG mice, but that treatment with AAV-sFRP3-mCherry increased spine density in both NTG and APP mice (two-way ANOVA revealed significant effects of genotype, F(1,247) = 28.06; p < 0.0001; and treatment, F(1,247) = 49.63; p < 0.0001; Fisher's LSD post hoc tests NTG-mCherry vs APP-mCherry, p < 0.0001; NTG-mCherry vs NTG-sFRP3, p = 0.0006; APP-mCherry vs APP-sFRP3, p < 0.0001; Fig. 5E). In addition, we used Sholl analysis, which quantifies the number of dendritic branches that intersect concentric circles of given radii from the soma (distance from soma), to assess the morphological complexity of these 6-week-old newborn neurons (Fig. 5F). Three-way ANOVA revealed significant effects (all p < 0.0001) of distance from soma F(29,7380) = 604.6, treatment F(1,7380) = 15.96, genotype F(1,7380) = 109.1, distance × treatment interaction F(29,7380) = 3.350, distance × genotype interaction F(29,7380) = 26.76, treatment × genotype interaction F(1,7380) = 52.25, and distance × treatment × genotype interaction F(29,7380) = 2.582. Compared with newborn neurons in NTG mice treated with AAV-mCherry (solid black markers), those in APP mice treated with AAV-mCherry (solid blue markers) had reduced numbers of dendritic intersections at most radii, indicating decreased dendritic arborization. Treatment with AAV-sFRP3 increased the complexity of dendritic arbors of newborn neurons in APP mice (open blue markers) compared with treatment with AAV-mCherry (solid blue markers), specifically for dendritic regions 50–120 µm from the soma (Fisher's LSD post hoc tests 50 µm, p = 0.026; 60 µm, p = 0.038; 70 µm, p = 0.004; 80 µm, p = 0.003; 90 µm, p = 0.017; 100 µm, p = 0.014; 110 µm, p = 0.013; 120 µm, p = 0.014). However, AAV-sFRP3 did not fully restore the arborization of newborn neurons in APP mice to NTG levels (black markers), as dendritic arbors 130 µm or farther from the soma were not altered by treatment (Fisher's LSD post hoc tests p > 0.05 for 130–290 µm from the soma). Overall spine density of granule cells in the DG, which was assessed via Golgi staining (Fig. 5G), was unchanged between genotype and treatment groups (two-way ANOVA p > 0.05 for all comparisons; Fig. 5H), suggesting that the effects on spine density are specific to adult-born cells.
Seizure activity has also been shown to increase ectopic migration of adult-born granule cells into the hilus, which disrupts normal network connectivity (Scharfman et al., 2000; Jessberger et al., 2007; Parent, 2007). We previously reported a small increase in the number of Prox1-expressing ectopically migrated granule cells in the hilus of 6-month-old APP mice (Fu et al., 2019). To determine whether AAV-sFRP3 might also reduce ectopic granule cell migration, we assessed the number of Prox1-expressing granule cells in the hilus of AAV-GFP- or AAV-sFRP3-infused NTG and APP mice. However, we found that at this age (3.5 months old), APP mice treated with AAV-GFP do not yet exhibit significantly increased numbers of ectopic hilar granule cells compared with NTG mice treated with AAV-GFP, and that AAV-sFRP3 did not significantly alter the numbers of ectopic hilar granule cells in either NTG or APP mice (two-way ANOVA p > 0.05 for all comparisons; Fig. 5I).
We also examined whether AAV-sFRP3 might have affected seizure incidence. We first assessed ΔFosB expression in NTG and APP mice that had received bilateral infusions of either AAV-GFP or AAV-sFRP3 and found that whereas, as expected, APP mice treated with AAV-GFP showed increased DG ΔFosB immunoreactivity compared with NTG mice, APP mice treated with AAV-sFRP3 did not (two-way ANOVA revealed effect of genotype, F(1,27) = 6.194; p = 0.019; Fisher's LSD post hoc tests APP-GFP vs NTG-GFP, p = 0.004; APP-GFP vs NTG-sFRP3, p = 0.008; APP-sFRP3 vs NTG-GFP, p = 0.483; APP-sFRP3 vs NTG-sFRP3, p = 0.639; APP-GFP vs APP-sFRP3, p = 0.042; Fig. 6). In a separate group of virus-treated APP mice that were implanted with EEG electrodes, we found that treatment with AAV-sFRP3 decreased epileptiform spike frequency over time. At 1 month post-virus infusion, the epileptiform spike rate of APP mice treated with AAV-sFRP3 was 139.5 ± 11.4 epileptiform spikes/h, which decreased to 96.6 ± 10.0 epileptiform spikes/h after 2.5 months of virus expression, whereas APP mice treated with AAV-GFP showed a nonsignificant decrease from 127.0 ± 30.4 epileptiform spikes/h at 1 month postvirus infusion to 87.3 ± 20.8 epileptiform spikes/h at 2.5 months postvirus infusion (two-way ANOVA revealed effects of length of virus expression, F(1,8) = 14.15; p = 0.0055; Sidak's post hoc tests for 1 month vs 2.5 months postvirus infusion in AAV-sFRP3, p = 0.029, and in AAV-GFP, p = 0.094). Similar decreasing trends were observed for seizure frequency in AAV-sFRP3-treated APP mice, which had 0.69 ± 0.15 seizures/day after 1 month of virus expression and 0.33 ± 0.04 seizures/day after 2.5 months of virus expression (two-way ANOVA revealed effects of length of virus expression, F(1,8) = 7.254; p = 0.027; Sidak's post hoc test for 1 month vs 2.5 months postvirus infusion in AAV-sFRP3, p = 0.107). Thus, AAV-sFRP3 might have a mild beneficial effect on the severity of seizure activity, although further investigation is needed to confirm this finding.
AAV-sFRP3 decreased DG ΔFosB expression in APP mice. A, B, Example images of hippocampal ΔFosB immunostaining (A) and quantification of ΔFosB immunoreactivity (IR) in the GCL (B) of NTG and APP mice treated with either AAV-GFP or AAV-sFRP3 for 2.5 months. *p < 0.05; **p < 0.01. Values indicate mean ± SEM.
Discussion
In summary, we found that the suppression of sFRP3 expression is a key driver of seizure-induced alterations in neurogenesis in APP mice. Increasing sFRP3 expression in APP mice normalized NSC division and prevented NSC loss, normalized spine density and increased dendritic arborization of newborn cells, and improved performance in a spatial discrimination task. Epigenetic suppression of sFRP3 expression by the seizure-induced transcription factor ΔFosB is one mechanism that contributes to the downregulation of sFRP3 in conditions with recurrent seizures.
It was somewhat surprising that while AAV-sFRP3 normalized NSC division and total NSCs in APP mice, DCX+ cell numbers remained unchanged. It is possible that there are multiple mechanisms dysregulating neurogenesis in APP mice and that preventing aberrant seizure-induced NSC dynamics revealed the effects of other factors, such as APP/amyloid-β (Aβ). Although we previously found that there are no cell-autonomous differences in cell division in neurospheres derived from NTG and APP mice (Fu et al., 2019), other studies have demonstrated that APP/Aβ can exert inhibitory effects on neurogenesis (Lazarov and Marr, 2010). Therefore, it is likely that APP/Aβ also influences neurogenesis in a non-seizure-dependent manner.
In addition to direct effects on adult neurogenesis, AAV-sFRP3 might have improved overall network function indirectly through preservation of the NSC pool. NSCs have both neurogenic and non-neurogenic functions (Bacigaluppi et al., 2020). Interestingly, when compared with AD patients, subjects who exhibited AD neuropathology but remained cognitively intact showed increased numbers of NSCs, and the number of NSCs also correlated with better cognition (Briley et al., 2016). Indeed, NSCs and progenitor cells can regulate the microenvironment around them through the secretion of various factors, such as different RNAs, lipids, membrane particles, immunomodulatory factors, growth factors, and stem cell factors (Bacigaluppi et al., 2020). Hippocampal NSCs secrete exosomes that can promote synaptic resilience to Aβ oligomers (Micci et al., 2019). NSCs also secrete pleiotrophin, which is critical in newborn neuron development (Tang et al., 2019). NSC-released factors also modulate the functions of non-neuronal cell types. For example, vascular endothelial growth factor, which, in addition to playing a critical role in stem cell maintenance (Kirby et al., 2015), also modulates microglia proliferation, migration, and phagocytic activity (Mosher et al., 2012). Microglia can in turn influence neurogenesis via secretion of factors that can affect proliferation, differentiation, and cell survival of neural progenitors (Ekdahl, 2012; Gemma and Bachstetter, 2013) and via phagocytosis of apoptotic newborn cells (Sierra et al., 2010). Microglia also regulate synaptic plasticity through activity-dependent synaptic pruning (Cornell et al., 2022), which is another important aspect of learning and memory.
Our results suggest that the reduction of sFRP3 expression in the context of seizures in AD might be detrimental to cognitive function in the long run and that restoring sFRP3 can help preserve the NSC pool and thus benefit cognition. In contrast, in a BubR1 hypomorphic (BubR1H/H) mouse model of accelerated aging, reducing sFRP3 decreases astrogliosis and microglial activation and increases cell division and might thus be neuroprotective (Cho et al., 2019). However, BubR1H/H mice show decreased baseline cell division, and knockdown of sFRP3 in this context increases cell division back to wild-type levels. APP mice exhibit increased baseline cell division, and thus increasing sFRP3 in APP mice normalizes NSC division back to levels seen in NTG control mice. Thus, in the context of disease, maintaining optimal levels of cell division through modulation of sFRP3 expression might be critical for improving hippocampal function.
Notably, certain modulators of neuronal activity that are typically regarded as beneficial for cognition, such as electroconvulsive shock therapy and exercise, reduce sFRP3 expression (Jang et al., 2013b). In those conditions, transient sFRP3 reduction is necessary for the context-specific stimulation of neurogenesis, which enhances cognitive function under specific conditions (Sahay et al., 2011; Jang et al., 2013b; Anacker and Hen, 2017). In addition, chronic treatment with antidepressants in the context of depression, in which levels of neurogenesis are reduced, also decreases sFRP3 expression (Jang et al., 2013a; Berger et al., 2020). It is possible that the specific activity patterns that suppress sFRP3 expression, or the manner by which sFRP3 levels are decreased, could affect whether engagement of NSC division leads to depletion of the NSC pool. Indeed, physiological stimulation, such as exploration of a novel environment, more readily activates young granule cells in the DG, while KA-induced pharmacological stimulation activates older granule cells and GABAergic interneurons (You et al., 2020). Interestingly, ΔFosB, which we found to be one regulator of sFRP3 expression, can also be induced by spatial learning and novel environment exposure (Eagle et al., 2015), as well as by chronic treatment with the antidepressant fluoxetine (Vialou et al., 2015), although the pattern of activation is different from that elicited by chronic seizures (Corbett et al., 2017). It is possible that ΔFosB modulates sFRP3 expression even in nonpathological conditions. The beneficial effects of therapeutic inductions of neurogenesis likely depend not just on decreasing sFRP3 expression to “release the brake” on neurogenesis, but also on a combination of other factors that might protect against excessive NSC use and depletion. For example, antidepressants and exercise appear to target and increase proliferation of intermediate neural progenitors rather than the NSCs themselves, which protects against NSC depletion (Kronenberg et al., 2003; Encinas et al., 2006). These findings suggest that the context in which sFRP3 expression is reduced plays a critical role in determining downstream effects and that sFRP3 is an important, but not the sole, signal for controlling NSC activation or neurogenesis. A more complete understanding of such contexts might allow for precise utilization of sFRP3 as a therapeutic agent in different neurological disorders.
Data availability
Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Jeannie Chin (Jeannie.Chin{at}bcm.edu).
Footnotes
This research was supported by National Institutes of Health Grants NS086965 and NS085171 (J.C.). Research reported in this publication was also supported by the Eunice Kennedy Shriver National Institute of Child Health & Human Development of the National Institutes of Health under Award Number P50HD103555 for use of the Neurovisualization Core and the Optogenetics and Viral Vectors Core facilities. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
The authors declare no competing financial interests.
U.T.’s present address: Department of Neurological Surgery, Weill Cornell Medicine, New York City, NY 10065, USA. K.M.’s present address: Department of Anesthesiology, Critical Care and Pain Medicine, Boston Children’s Hospital, Boston, MA 02115, USA. J.C.Y.’s present address: Department of Neurology, Massachusetts General Hospital, Boston, MA 02114, USA. X.Z.’s present address: Department of Neurology, Children’s Hospital of Philadelphia, Philadelphia, PA 19104, USA.
- Correspondence should be addressed to Jeannie Chin at jeannie.chin{at}bcm.edu.
