Abstract
CD2-associated protein (CD2AP) was identified as a genetic risk factor for late-onset Alzheimer's disease (LOAD). However, it is unclear how CD2AP contributes to LOAD synaptic dysfunction underlying AD memory deficits. We have shown that loss of CD2AP function increases β-amyloid (Aβ) endocytic production, but it is unknown whether it contributes to synapse dysfunction. As CD2AP is an actin-binding protein, it may also function in F-actin-rich dendritic spines, which are the excitatory postsynaptic compartments. Here, we demonstrate that CD2AP colocalizes with F-actin in dendritic spines of primary mouse cortical neurons of both sexes. Cell-autonomous depletion of CD2AP specifically reduces spine density and volume, resulting in a functional decrease in synapse formation and neuronal network activity. Postsynaptic reexpression of CD2AP, but not blocking Aβ production, is sufficient to rescue spine density. CD2AP overexpression increases spine density, volume, and synapse formation, while a rare LOAD CD2AP mutation induces aberrant F-actin spine-like protrusions without functional synapses. CD2AP controls postsynaptic actin turnover, with the LOAD mutation in CD2AP decreasing F-actin dynamicity. Our data support that CD2AP risk variants could contribute to LOAD synapse dysfunction by disrupting spine formation and growth by deregulating actin dynamics.
Significance Statement
CD2-associated protein (CD2AP) is a candidate genetic risk factor of late-onset Alzheimer's disease (LOAD) expressed in neurons with an unknown impact on synapse dysfunction, one of the causal LOAD mechanisms. Our research has revealed CD2AP as a new synaptic protein and established a connection between a LOAD genetic variant in CD2AP and synaptic dysfunction independent of β-amyloid accumulation. This study suggests an explanation for the CD2AP-mediated predisposition to AD. Furthermore, we have found that controlling CD2AP's impact on spinal F-actin could be a potential target for therapeutic intervention against LOAD.
Introduction
Late-onset Alzheimer’s disease (LOAD) is the most common neurodegenerative disease and remains without adequate treatment. Unfortunately, the loss of neurons accompanying the pathology that defines AD, β-amyloid (Aβ) plaques, and tau neurofibrillary tangles is likely irreversible and untreatable. However, we may be able to modify early synapse dysfunction that occurs before neurodegeneration. Therefore, dissecting the causal mechanisms of synaptic dysfunction in LOAD is necessary to identify novel promising therapeutic targets.
The gene encoding for CD2-associated protein (CD2AP) was identified as a potential risk factor for LOAD by genome-wide screens of LOAD patients (Hollingworth et al., 2011; Naj et al., 2011) and confirmed by meta-analysis studies (Kamboh et al., 2012; Chen et al., 2015; Kunkle et al., 2019; Gao et al., 2022). Although GWAS identified frequent noncoding variants, whole gene sequencing identified a pathological variant that introduces a mutation in CD2AP (Lys633Arg or K633R), which is highly associated with AD (Vardarajan et al., 2015) with unclear biological significance. The CD2AP expression in the brain is yet unknown but was reduced in the blood of LOAD patients (Tao et al., 2017). Additionally, CD2AP accumulates in advanced AD brain (Camacho et al., 2022). Moreover, the CD2AP susceptibility loci correlate with the burden of neuritic plaques in AD patients (Shulman et al., 2013).
CD2AP is expressed ubiquitously and belongs to the CIN85/CD2AP protein family (Cormont et al., 2003; Bruck et al., 2006; Rouka et al., 2015). CD2AP was initially identified in T cells, but it is most important in kidney glomeruli cells, where its loss of function causes proteinuria and renal dysfunction (Shih et al., 1999; Kim et al., 2003). CD2AP is an actin-binding protein that controls actin polymerization and stability in the glomeruli (Li et al., 2000; Shih et al., 2001) and undifferentiated cells (Welsch et al., 2001; Tang and Brieher, 2013). CD2AP also associates with membranes for endocytosis and endosomal maturation (Monzo et al., 2005; Gauthier et al., 2007; Tolvanen et al., 2015; Ubelmann et al., 2017; Furusawa et al., 2019). In neurons, CD2AP was found to localize more to dendrites than axons (Ubelmann et al., 2017), consistent with a postsynaptic function, and to function in neurite extension (Harrison et al., 2016).
CD2AP was first associated with an AD mechanism by modulating tau toxicity in a Drosophila model of AD (Shulman et al., 2014). It was then implicated in Aβ production but not deposition in an amyloidosis model (PS1APP mice; Liao et al., 2015). We showed that CD2AP loss of function disrupts the endocytic trafficking of Aβ precursor protein (APP), leading to increased Aβ production and intraneuronal accumulation (Ubelmann et al., 2017). However, it is unclear whether this increase in intraneuronal Aβ can cause synapse dysfunction in LOAD, as we observed in the early-onset familial AD (Takahashi et al., 2004; Almeida et al., 2005; Snyder et al., 2005). Alternatively, CD2AP, like other LOAD risk genes, may cause synaptic dysfunction upstream Aβ production (Perdigão et al., 2020). Interestingly, the loss of Cindr, the Drosophila homolog of CD2AP, impaired neuromuscular synapses by interfering with the synapse maturation (Ojelade et al., 2019). Since CD2AP is an actin regulator and the actin cytoskeleton is essential for spines that hold excitatory synapses (Hotulainen and Hoogenraad, 2010), it is hypothesized that CD2AP might function at spines and impact synapses.
Here, we investigate the synaptic function of CD2AP by using knockdown and overexpression approaches in primary cultures of mouse cortical neurons. Our goal is to determine if CD2AP modulates synapses by playing a direct role in spine density and morphology. Interestingly, we discovered that a LOAD-associated mutation in CD2AP results in an aberrant gain of function in spines. Mechanistically, we also established that CD2AP functions in the spines through spinal F-actin, with Aβ production playing a minor role. Therefore, CD2AP risk variants can contribute to LOAD synapse dysfunction by deregulating spinal F-actin independently of Aβ.
Materials and Methods
Animals
All animal procedures were performed under the EU recommendations and approved by the NOVA Medical School (NMS)–Universidade Nova de Lisboa ethical committee (147/2021/CEFCM) and the NMS Animal Welfare Body.
Cell culture, cDNA overexpression, shRNA knockdown, and treatments
Primary neuronal cultures of Mus musculus were prepared as previously (Almeida et al., 2005) from the cortices of Embryonic Day 16 (E16) BALB/c mice of either sex. Briefly, cortices were pooled and dissociated by trypsinization and trituration in Dulbecco's medium Eagle medium (DMEM, Thermo Fisher Scientific) with 10% fetal bovine serum (heat-inactivated FBS, Thermo Fisher Scientific). Dissociated neurons were plated in DMEM with 10% FBS on poly-D-lysine (Sigma-Aldrich)-coated six–well plates (3 × 105 cells/cm2) and glass coverslips (5 × 104 cells/cm2). After 3–16 h, the media were substituted for Neurobasal supplemented with 2% B27 (Invitrogen) and GlutaMAX, or BrainPhys Neuronal Medium (BrainPhys; STEMCELL Technologies) supplemented with 2% NeuroCult SM1 (STEMCELL Technologies) that allows for a more physiological neuronal differentiation and activity (Bardy et al., 2015), and penicillin/streptomycin and kept at 37°C in 5% CO2. Neurons cultured in Neurobasal were used after 21 d in vitro (DIV) and BrainPhys at 15 DIV.
For live cell imaging, primary neurons were grown on 18 mm glass bottom plates (FluoroDish, Thermo Fischer Scientific; 3 × 105 cells/cm2). Before imaging, the medium was exchanged for 37°C prewarmed imaging medium (120 mM NaCl, 3 mM KCl, 2 mM CaCl2, 2 mM MgCl2, 10 mM glucose, 10 mM HEPES) supplemented with B27/SM1.
Neuroblastoma Neuro2a cells (N2a; ATCCCCL-131) were provided by Z. Lenkei (ESPCI-ParisTech). Cells were cultured in DMEM-GlutaMAX (Thermo Fisher Scientific) with 10% FBS (Sigma-Aldrich) at 37°C and 5% CO2. ShRNA infections were performed 24 h after plating for 72 h.
For cDNA overexpression, primary neurons after 7 DIV (BrainPhys) or 12 DIV (Neurobasal) were lipotransfected with Lipofectamine 2000 (Thermo Fisher Scientific) according to the manufacturer's instructions with a few modifications, 500 ng of DNA were used per coverslip, and after the lipid–cDNA complexes were incubated with neurons for 5 min the medium was replaced with conditioned and fresh BrainPhys media (1:1). The cDNA plasmids used were: p-GFP-C1, Cherry-C1, mCherry-actin (Koestler et al., 2008), CD2AP-GFP and CD2AP-myc (Cormont et al., 2003). CD2AP(K633R)-GFP, and CD2AP(K633R)-myc which were generated by site-directed mutagenesis of CD2AP-GFP with the primer 5′GAAATAGCAAAGCTGAAGAAAGCTGTTCTGTTG3′ and 5′CAACAGAACAGCTTTCCTCAGCTTTGCTATTTC3′.
To knockdown CD2AP we used pFLRU lentivirus (LV) encoding yellow fluorescent tag and CD2AP shRNA oligonucleotides sh1 (shCD2AP-1; GTGGAACCCTGAACAATAAG) and sh2 (shCD2AP-2, GGAACCAATGAAGATGAACTTACA) were a gift from Andrey S. Shaw (Zhao et al., 2013), or nontargeting shRNA (shControl; TR30021, OriGene Technologies), prepared by cotransfection with psPAX2 (gift from Didier Trono; Addgene plasmid # 12260) and pMD2.G (gift from Didier Trono; Addgene plasmid # 12259) in STAR-Rdpro cells (ECACC 04072117). LVs were collected by ultraspeed centrifugation and purified with the Lenti-X Concentrator (Takara Bio) according to the manufacturer's protocol. Neurons (7 or 12 DIV) were infected with LVs expressing shControl, shCD2AP-1, and shCD2AP-2 for 5 min at 37°C in 5% CO2. The LV media were replaced with half-conditioned and half-fresh media.
For rescue experiments, cDNA transfection was performed with 10 DIV neurons (BrainPhys) after shRNA infection at 7 DIV.
When indicated, γ-secretase was inhibited with 250 nM γ-secretase inhibitor IX (Calbiochem), and BACE1 was inhibited with 10 µM β-secretase inhibitor compound IV (Merck Millipore, 565788) or 0.1% DMSO (solvent) as a control.
Antibodies and probes
The following primary antibodies were used: anti-CD2AP [Merck Millipore, HPA00326, 1:100 (IF); 1:1,000 (WB)]; anti-postsynaptic density protein-95 [PSD-95; D27E11, Cell Signaling Technology, 3450, 1:200 (IF); 1:2,500 (WB)]; anti-synapsin [Abcam, ab8, 1:200 (IF); 1:2,500 (WB)]; anti-tubulin [T5168, Sigma-Aldrich, 1:5,000 (WB)]; anti-GFP [Sicgen, AB0020, 1:200 (IF)]; anti-vGluT1 [Merck Millipore, MAB5502, 1:100 (IF)]; anti-cortactin [p80/85, Merck Millipore, 05-180-I-25UL, 1:200 (IF)]; and anti-MAP2 [Abcam, ab5392, 1:300 (IF)]. The secondary antibodies were conjugated to Alexa Fluor 488, 555, and 647 (Invitrogen) or HRP (Bio-Rad Laboratories). Phalloidin conjugated to Alexa Fluor 488, 555, and 647 (Invitrogen) was used to detect F-actin (1:500).
Immunofluorescence
Immunofluorescence was performed as previously (Ubelmann et al., 2017). Briefly, primary neurons were fixed with 4% paraformaldehyde/4% sucrose in PBS 1× for 20 min, permeabilized with 0.3% Triton X-100 in PBS 1× for 5 min, and blocked with 2% FBS, 1% BSA in PBS for 1 h at room temperature (RT) before primary antibody incubation for 16 h at 4°C. After washing, secondary antibodies were incubated for 1 h at RT. Coverslips were mounted using Fluoromount-G (Southern Biotech).
Labeling of active synapses with FM4.64
Primary neurons were incubated with the lipophilic dye FM4.64 (10 µM; Thermo Fisher Scientific) for 90 s in a high-potassium saline solution (in mM: 119 NaCl, 70 KCl, 2 CaCl2, 2 MgCl2, 5 HEPES, and 30 glucose), washed with a low-calcium solution (in mM: 150 NaCl, 2.5 KCl, 0.2 CaCl2, 5 MgCl2, 5 HEPES, and 30 glucose) with 1 µM tetrodotoxin (Tocris Bioscience) for 30 s, fixed with 2% paraformaldehyde/4% sucrose in PBS 1× for 10 min and mounted.
Immunoblotting
Cell lysates were prepared using modified RIPA buffer [50 mM Tris–HCl, 1% NP-40, 0.25% sodium deoxycholate, 150 mM NaCl, 1 mM EGTA, and 0.1% SDS, with protease inhibitor cocktail (PIC; Roche)], pH 7.4, as described (Burrinha et al., 2021). Proteins separated by 10% Tris-glycine SDS–PAGE were transferred to nitrocellulose membranes (GE Healthcare) and processed for immunoblotting using the ECL Prime kit (GE Healthcare). Immunoblot images were captured using a ChemiDoc Gel Imaging System (Bio-Rad Laboratories) within the linear range and quantified by densitometry using the Analyze gels function in ImageJ.
Brain synaptosomes preparation
Synaptosomes were prepared from the forebrains (including the cortex and hippocampus) of 6-month-old (adult) Balb6c mice. All steps were performed at 4°C. The samples were homogenized with a pestle in ice-cold buffer 1 [0.32 M sucrose (Sigma-Aldrich); 10 mM HEPES; 2 ml PIC (1×); 1 mM EDTA], pH 7.4, using 10 ml buffer per gram of tissue. The homogenate was centrifuged (10 min, 1,000 × g, 4°C) to obtain a pellet containing nuclear fractions (P1) and the postnuclear supernatant (S1). S1 was centrifuged (15 min, 10,000 × g, 4°C) to generate a pellet that contains crude synaptosomes (P2) and a supernatant fraction (S2). P2 was resuspended in 1 ml of buffer 1, saved, and centrifuged (15 min, 10,000 × g, 4°C), generating the washed synaptosome fraction (P2′). P2′ was lysed by hypoosmotic shock in ice-cold ddH2O + 10 mM HEPES, homogenized with a pestle, and let rotate at 4°C for 30 min to ensure complete lysis. P2′ was further centrifuged (30 min, 21,000 × g, 4°C) to generate a supernatant that contains crude synaptic vesicles (S3) and a pellet that contains synaptosomal membranes (P3). P3 was resuspended in 100 μl modified RIPA buffer.
MEA recordings and analysis
Primary neurons were cultured as described above and plated in single-well MEA at a density of 50,000 cells in BrainPhys media per well coated with poly-L-ornithine (Sigma-Aldrich) for 16 h at 37°C and laminin (Sigma-Aldrich) for 4 h. Primary neurons were transduced with LV expressing shCD2AP-1, shCD2AP-2, or shControl as described. Each MEA well (3Brain Prime HD-MEA) contained 4,096 recording electrodes coupled to a ground electrode. MEA recordings of basal neuronal network activity were performed at 15 DIV for 30 s at 37°C using the BioCamX recording system and BrainWave v.4.5 software (3Brain).
Spikes were detected with the analysis tool “spike detection” of the BrainWave software. The precise timing spike detection algorithm was used to detect negative spikes, with SD of 8.0, 2.0 ms peak lifetime, and 1.0 ms refractory period. The active electrodes were considered in the percentage of the total (4,096). The frequency of electrodes grouped by the mean firing rate was obtained using the histogram function of GraphPad Prism 8.0 with number distribution by bins. The average firing rate refers to the total number of spikes per 30 s of recording. Spiking frequency refers to the number of spikes detected by each electrode per 30 s recording. The cumulative frequency of the firing rate was calculated using Prism 8.0. Bursts were detected by the “burst detector” analysis tool, with a maximum spike interval of 30 ms and a minimum number of five spikes. The burst rate was calculated by dividing the total bursts by 30 s of recording. The number and duration of network bursts, spiking frequency of network bursts, interburst intervals, and spiking frequency were calculated using BrainWave v4.
Fluorescence recovery after photobleaching (FRAP)
To measure F-actin FRAP in spines, primary neurons were transfected with GFP, CD2AP-GFP, CD2AP(K633R)-GFP, and mCherry-actin. GFP and mCherry-actin were imaged using the LSM980 the 63× NA 1.4 oil Plan-Apochromat objective at 37°C, and FRAP was performed using the ZEN (Blue 3.3 edition, Carl Zeiss) bleaching module. Regions of interest (ROIs) of 12 pixels in diameter were placed on 3–6 spines per dendritic segment. Cells were imaged at 1 frame per 1–4 s in a single plane. After five baseline frames, ROIs were photobleached and imaged for 2 min, allowing recovery to reach a stable plateau. FRAP analysis of mCherry-actin fluorescence intensity in bleached spines was performed as described (Koulouras et al., 2018) using EasyFRAP-web (freely accessible platform-independent; https://easyfrap.vmnet.upatras.gr/). Briefly, fluorescence was background subtracted and corrected for photobleaching, and the baseline was set to 100% for each photobleached spine. FRAP recovery and half-time were calculated by fitting an exponential curve.
Image acquisition
Images were acquired on an epifluorescence microscope, the Zeiss Imager Z2 system (Carl Zeiss) equipped with a 20× air EC Plan-Neofluar objective 63× NA 1.4 oil Plan-Apochromat objective and a Zeiss Axiocam 506 mono camera, and a confocal microscope with super-resolution, LSM980–Airyscan 2 (Zeiss) equipped with 63× NA 1.4 oil Plan-Apochromat objective. 3D z-stacks were acquired to enable 3D reconstructions as indicated. The samples were imaged in parallel and using identical acquisition parameters for direct comparison.
Quantitative bioimaging analyses
Image analysis was carried out using Fiji (ImageJ 1.53, https://fiji.sc/), ICY (icy.bioimageanalysis.org; de Chaumont et al., 2012), or Imaris (version 9.5.0; https://imaris.oxinst.com/).
For the quantification of spinal enrichment, spinal and dendritic F-actin, ICY was used to outline a ROI corresponding to a dendritic section manually or automatically based on the GFP signal and outline spines, either by selecting spines with the “ellipse” selection tool or subtracting the dendritic shaft from the dendritic ROI. Each ROI's fluorescence mean intensity and area were obtained with ICY ROI measurements. Spinal enrichment was obtained by dividing the fluorescence intensity in the spine ROI by the fluorescence intensity in the dendritic ROI and presented as a percentage. Spinal F-actin puncta were segmented using the ICY “Spot detector” module in the spine's ROI, and each puncta area was obtained using ICY ROI measurements.
For the quantification of vGluT1, PSD-95, and FM4.64 puncta density or size, dendritic ROIs were outlined with the ICY “Area” selection tool or using ICY “Spot Detector ROI” module “selection based on Channel (GFP),” puncta were segmented with the ICY “Spot detector” module, and ROI's measurements were exported as previously including the dendrite ROI length (Feret's diameter; Burrinha et al., 2021). The “ComDet v.0.5.5” plugin was used for puncta ROI segmentation and colocalization analysis in dendritic ROIs outlined using the “polygon selection” tool using Fiji, as previously (Burrinha et al., 2023).
For the Sholl analysis, single pyramidal neurons expressing GFP were 3D reconstructed using Imaris Filament Tracer. The Filament Sholl analysis was used to quantify neurite number, intersections, and length.
To quantify spine density volume, Imaris “Filament Tracer” was used to 3D reconstruct dendrites based on the GFP (shRNA) or mCherry (OE) signal. The spine module automatically detects spines and provides spine head volume. After spine 3D reconstruction, we classify spines with “Imaris Spines Classifier” plugin with the following customized criteria: stubby (spine length < 1 µm and maximum spine width value > 0.45 µm), mushroom (maximum spine head width > minimum width neck × 2, spine length < head length × 4 and spine length > 1 µm), and long thin (mean spine head values ≥ mean width neck value × 1.1 and spine length > 1 µm). The remaining spines that did not fit the previous criteria were considered filopodia.
When automatic detection was impossible, spine density was manually counted using the point selection tool in Fiji.
To quantify spine motility, dendrites were imaged live 1 frame every 30 s for 30 min; after drift correction and deconvolution with ZenPro 201, dendrites were 4D reconstructed, and the spine volume was measured using Imaris. The spine motility index was measured as described (Dunaevsky et al., 1999) with slight modifications. Briefly, the motility index was obtained for each spine by calculating the ratio of the difference between the largest and smallest spine volume to the average spine volume during the time lapse.
Experimental design and statistical analysis
Each independent experiment was performed with one set of cultures with different conditions processed in parallel to reduce culture-to-culture variability. All cultures were prepared from both male and female mice. Experiments were repeated multiple times, as indicated in the figure legends. Prism 8 (GraphPad) was used for statistical analysis and graphic representation of individual or average replicates with mean ± SD as indicated in the figure legends. The sample size was determined based on pilot studies. Data were tested with the D'Agostino–Pearson's omnibus normality test. For paired data, the Wilcoxon t test was applied. For nonparametric and unpaired data, the Mann–Whitney test was applied. For parametric and multiple comparisons, statistical analysis of data ordinary one-way ANOVA with Holm–Sidak's multiple-comparison test was applied. For nonparametric and multiple-comparison statistical data analysis, one-way ANOVA (Kruskal–Wallis test) with post hoc Dunn's testing was applied. For grouped data, we used two-way ANOVA. Significance was considered as *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001; and ns, not significant.
Results
A pool of CD2AP localizes in the spines
Previously, we localized CD2AP in dendritic endosomes relevant to APP endocytic trafficking in cortical mouse neurons (Ubelmann et al., 2017). To determine if CD2AP localized to synapses, we analyzed CD2AP distribution in synaptic fractions of the mouse adult brain (Fig. 1A). We found CD2AP present in the postsynaptic fraction (P3) positive for PSD-95 and negative for synapsin but not in the presynaptic fraction (S3) (Fig. 1A).
CD2AP is present in dendritic spines. A, Western blot analysis of CD2AP, PSD-95, and synapsin in postnuclear supernatant (S1), crude synaptosomal fraction (P2), crude synaptic vesicle fraction (S3), and synaptosomal membrane fraction (P3) of the 6-month-old brain. Notably, CD2AP is present in the PSD-95-positive fraction P3 and not in the synapsin-positive S3 fraction, supporting a postsynaptic location of CD2AP. B, CD2AP (green) endogenous localization in Cherry (magenta)-expressing dendrites. Scale bar, 10 µm. C, Quantification of the CD2AP and Cherry spinal enrichment, the fraction of dendritic signal in spines (n = 4; NCherry = 42; NCD2AP = 42 dendrites; **p < 0.01; unpaired t test). D, Western blot of CD2AP, PSD-95, and tubulin in primary neurons after 5, 11, and 15 DIV. E, Quantification of CD2AP and PSD95 levels normalized to 11 DIV primary neurons (n = 3; *p < 0.05; **p < 0.01; unpaired t test). F, Colocalization of CD2AP (cyan) and PSD95 (magenta) in GFP-expressing dendritic spines. Scale bar, 5 µm. G, Quantification of CD2AP colocalization with PSD95 as a percentage of CD2AP and PSD-95 (n = 2; NCD2AP = 25; NPSD-95 = 25 dendrites; ns, not significant; Mann–Whitney test). H, CD2AP (cyan) localization lateral to PSD-95 (magenta) in the GFP-expressing spine (green). Scale bar, 1 µm. I, CD2AP (green) colocalization with EEA1 (magenta) and F-actin (cyan) at the spine. Scale bar, 1 µm. J, CD2AP-GFP (CET-R2 rainbow color map; colorcet.com) enrichment at the spine head was recorded by time-lapse Airyscan confocal for 13 s (1 frame per 4.42 s). Scale bar, 0.5 µm. The time lapse is provided in Movie 1-1. Data are presented as mean ± SD.
Next, we investigated whether CD2AP was present in the spines by immunofluorescence analysis in dendrites of mature cortical mouse neurons expressing mCherry to allow the morphological identification of dendritic spines (Fig. 1B). We detected CD2AP puncta in most spines. We quantified the fraction of dendritic CD2AP or mCherry present in the spines. We found that 34% of dendritic CD2AP, more than soluble mCherry (27%), localized to the spines of mature neurons (Fig. 1C).
CD2AP is present in dendritic spines. CD2AP (CET-R2 rainbow color map) dynamic movement into spine head (1 frame per 4.42 s). Scale bar, 0.5 µm.
Furthermore, we measured CD2AP levels during the differentiation and synaptic maturation of primary mouse neurons cultured in BrainPhys media for 5, 11, and 15 DIV (Fig. 1D). CD2AP expression increased threefold, from 5 to 15 DIV, similarly to PSD-95 (Fig. 1E), suggesting a synaptic function for CD2AP. We also analyzed the colocalization of CD2AP with PSD-95 in dendritic spines of GFP-expressing neurons (Fig. 1F). Quantification revealed that 31% of CD2AP puncta colocalized with PSD-95 (Fig. 1G), indicating that CD2AP is present in one-third of the spines. Interestingly, we often observed a CD2AP punctum at the tip of the dendritic spines. This was highlighted in a 3D reconstruction of a GFP-filled spine using Imaris, positioned laterally to the PSD-95 punctum (Fig. 1H). This laterality of CD2AP is consistent with our previous findings of CD2AP endosomal localization and the report of endocytic zones lateral to the PSD (Lu et al., 2007). Indeed, we observed that CD2AP punctum could be laterally associated with the early endosome marker EEA1 to the F-actin-rich spine head (Fig. 1I). Furthermore, we could visualize the movement of CD2AP-GFP puncta into a spine as if associated with an endosome in mature live neurons (15 DIV; Fig. 1J; Movie 1-1).
These data demonstrate that CD2AP localizes to spines in mature primary cortical neurons.
The removal of CD2AP reduces spine density and synapses
The localization of CD2AP in spines led us to investigate whether CD2AP has synaptic function using a knockdown approach. To knockdown CD2AP, we introduced different sequences of small hairpin interfering RNA against CD2AP (shCD2AP-1 and shCD2AP-2) previously reported to knockdown CD2AP (Zhao et al., 2013) and a nontargeting shRNA sequence (shControl) into neurons via LV expressing GFP. We confirmed the CD2AP knockdown through Western blot and immunofluorescence (Extended Data Fig. 2-1A–D).
To determine the impact of CD2AP knockdown on spines, we analyzed spines, morphologically identified using the expression of GFP in shCD2AP- and shControl-treated neurons by epifluorescence microscopy and 3D reconstruction using Imaris. We observed fewer spines after CD2AP knockdown (Fig. 2A). Quantifying spine density, the number of spines per dendritic length (10 mm), revealed that it was reduced by 55 and 60% with shCD2AP-1 and shCD2AP-2 treatment, respectively (Fig. 2B). Furthermore, the volume of the remaining spines decreased by almost 30% after treatment with shCD2AP-1 and shCD2AP-2 (Fig. 2C). We further dissected the changes per spine type, filopodia, long thin, mushroom, and stubby automatically using Imaris and found that both shCD2AP decreased the density of mushroom spines. In contrast, only shCD2AP-1 reduced filopodia; no significant differences were found in long, thin, and stubby spines (Extended Data Fig. 2-1E,F). These results indicate that CD2AP functions in spines, controlling their formation and maturation.
CD2AP knockdown reduces spine density and size and affects synapses. As indicated, primary neurons were treated with shControl, shCD2AP-1, or shCD2AP-2. Data on CD2AP knockdown efficiency are provided in Extended Data Figure 2-1A–D. A, Representative images of dendrites of neurons expressing GFP (green). Below is a 3D reconstruction using Imaris. Scale bar, 10 µm. B, Quantification of dendritic spine density (n = 5; NshControl = 63; NshCD2AP-1 = 61; NshCD2AP-2 = 34 dendrites; ****p < 0.0001; ns, not significant; Kruskal–Wallis test). Data on dendritic spine subtype density are provided in Extended Data Figure 2-1E,F. C, Quantification of dendritic spine volume based on GFP volume marker (n = 5; NshControl = 64; NshCD2AP-1 = 61; NshCD2AP-2 = 34 dendrites; **p < 0.01; ****p < 0.0001; ns, not significant; Kruskal–Wallis test). D, Synapses identified by PSD-95 (magenta, gray) and vGluT1 (cyan, gray) colocalization in neurons expressing GFP (Fire LUT, green). Scale bar, 10 µm. E, Quantification of the synapse density (n = 5; NshControl = 116; NshCD2AP-1 = 130; NshCD2AP-2 = 72 neurons; **p < 0.01; ****p < 0.0001; ns, not significant; Kruskal–Wallis test). F, Quantification of PSD-95 density (n = 5; NshControl = 25; NshCD2AP-1 = 31; NshCD2AP-2 = 15 neurons; *p < 0.05; **p < 0.001; ns, not significant; ordinary one-way ANOVA test). G, Quantification of vGluT1 density (n = 5; NshControl = 36; NshCD2AP-1 = 30; NshCD2AP-2 = 16 neurons; ns, not significant; Kruskal–Wallis test). H, Active synapses labeled with FM4.64 (gray) of neurons upon high-potassium–mediated depolarization. Scale bar, 10 µm. I, Quantification of FM4.64 puncta density (n = 3; NshControl = 110; NshCD2AP-1 = 123; NshCD2AP-2 = 151 neurites; **p < 0.01; ****p < 0.0001; ns, not significant; Kruskal–Wallis test). J, Spines identified with PSD-95 (gray) in dendrites labeled with MAP2 (blue) after treatment with BACEi or DMSO. Scale bar, 10 µm. K, Quantification of PSD-95 density in dendrites of neurons after treatment with BACEi or DMSO (n = 3; NshControl + DMSO = 77; NshCD2AP-2 + DMSO = 90; NshControl + BACEi = 51; NshCD2AP-2 + BACEi = 85 dendrites; *p < 0.05; ****p < 0.0001; Kruskal–Wallis test). L, Spines identified with PSD-95 (gray) in dendrites labeled with MAP2 (blue) after treatment with γ-secretase inhibitor (DAPT) or DMSO. Scale bar, 10 µm. M, Quantification of PSD-95 density after treatment with γ-secretase inhibitor (DAPT) or DMSO (n = 3; NshControl + DMSO = 59; NshCD2AP + DMSO = 81; NshControl + DAPT = 56; NshCD2AP + DAPT = 79 dendrites; **p < 0.01; ****p < 0.0001; ns, not significant; Kruskal–Wallis test). All data are presented as mean ± SD.
Figure 2-1
Knockdown of CD2AP by treatment with lentivirus expressing CD2AP shRNA. (A) CD2AP (grey) and MAP2 (blue) in dendrites of neurons treated with shControl, shCD2AP-1 and shCD2AP-2. Scale bar: 10 µm. (B) Quantification of CD2AP mean intensity in dendrites (n = 3, NshControl = 50, NshCD2AP-1 = 46, NshCD2AP-2 = 42, *P < 0.05, ****P < 0.0001; ns, not significant; Kruskal-Wallis test). (C-D) CD2AP western blot and tubulin as loading control in Neuro2a cells (C) and puromycin-selected NIH 3T3 cells (D) treated with shControl, shCD2AP-1, and shCD2AP-2. (E) Representative 3D reconstruction with spine subtypes using IMARIS of dendrites of neurons expressing GFP treated with shControl, shCD2AP-1, and shCD2AP-2. (F) Quantification of dendritic spine density per subtype in shControl (grey), shCD2AP-1 (magenta) and shCD2AP-2 (green) neurons (n = 3, NshControl = 9, NshCD2AP-1 = 9, NshCD2AP-2 = 9 dendrites, *P < 0.05; ns, not significant, two-way ANOVA test). Data are presented as mean ± SD. Download Figure 2-1, TIF file.
To assess whether CD2AP-dependent reduction in spines affected synapses, we performed immunostaining for vGluT1 and PSD-95 to identify pre- and postsynaptic compartments of excitatory synapses, respectively (Fig. 2D). We automatically quantified their colocalization as a proxy for synapses. The synapse density along dendrites was reduced by 24 and 41% with treatment with shCD2AP-1 and shCD2AP-2, respectively (Fig. 2E). The decrease in synapse density seems mainly due to a reduction in PSD-95 puncta with shCD2AP-1 and shCD2AP-2 treatment (25%; Fig. 2F) as vGluT1 puncta density was unchanged (Fig. 2G). We measured the density of active synapses after shCD2AP treatment by monitoring the density of FM4.64-labeled presynaptic puncta after induction of high-potassium–induced neuronal depolarization (90 s; Fig. 2H). The density of FM4.64 puncta was significantly reduced by 25 and 15% with shCD2AP-1 and shCD2AP-2, respectively (Fig. 2I). The loss of CD2AP likely causes a reduction in synapses due to the lack of functional spines.
We and others have found that reducing CD2AP expression increases intraneuronal Aβ42 (Liao et al., 2015; Ubelmann et al., 2017). In addition, in earlier work, we showed that intraneuronal Aβ42 accumulation induced by a familial AD mutation reduces spines (Almeida et al., 2005). Therefore, we investigated whether blocking Aβ production as previously (Burrinha et al., 2021) could rescue the reduction in spine density induced by CD2AP knockdown. We measured the PSD-95 density in shCD2AP neurons treated with the BACE inhibitor (BACEi) and the γ-secretase inhibitor (DAPT; Fig. 2J–M). Treatment with BACEi and DAPT increased PSD-95 puncta density in neurons treated with shCD2AP-2, but it remained significantly reduced compared with shControl (Fig. 2J–M). Notably, DAPT, but not BACEi, treatment reduced PSD-95 density in shControl neurons, supporting that γ-secretase has synaptic relevant substrates (Bittner et al., 2009; Barthet et al., 2018; Servián-Morilla et al., 2018). These results suggest that Aβ production partially contributes to spine loss induced by CD2AP knockdown. Thus, we discovered that CD2AP depletion reduces synapses by affecting spines, in part independently of Aβ production.
CD2AP knockdown impacts neuronal network activity
To investigate the impact of CD2AP knockdown on basal neuronal activity, we performed multielectrode array (MEA) recordings on primary neurons cultured in BrainPhys for 15 DIV. A representative MEA (4,096 electrodes) activity map shows the spontaneous neuronal activity of the shControl-treated neurons (Fig. 3A). The shCD2AP-treated neurons exhibited less activity. The raster profiles (Fig. 3B) and representative traces (Fig. 3C) indicate lower spiking and burst activity of shCD2AP-1 and shCD2AP-2 neurons compared with shControl.
CD2AP knockdown reduces neuronal activity. Primary neurons treated with shControl, shCD2AP-1, or shCD2AP-2 and basal activity were recorded with an MEA of 4,096 electrodes. A, Representative heatmaps of the spiking activity of active electrodes with >3 spikes/s are outlined in white. B, Representative raster lines showing spiking and burst activity of shControl- (black), shCD2AP-1- (magenta), and shCD2AP-2 (green)-treated neurons. Network bursts are identified below with red squares. C, Representative traces showing burst spiking activity of shControl (black), shCD2AP-1 (magenta), and shCD2AP-2 (green) neurons. D, The percentage of active electrodes (n = 3; ns, not significant; ordinary one-way ANOVA test). E, Frequency histogram of the mean firing rate (Hz) of active electrodes in shControl (gray), shCD2AP-1 (magenta), and shCD2AP-2 (green) neuronal network (n = 3; ns, not significant; two-way ANOVA test). F, The percentage of active electrodes > 3 Hz in shControl (gray), shCD2AP-1 (magenta), and shCD2AP-2 (green) neurons (n = 3; *p < 0.05; ns, not significant; ordinary one-way ANOVA test). G, Average spike frequency (Hz) of shControl (gray), shCD2AP-1 (magenta), and shCD2AP-2 (green) neurons (n = 3; NshControl = 751; NshCD2AP-1 = 452; NshCD2AP-2 = 356 electrodes; *p < 0.05; **p < 0.01; ns, not significant; ordinary one-way ANOVA test). H, Cumulative spike frequency of shControl (gray), shCD2AP-1 (magenta), and shCD2AP-2 (green; n = 3; NshControl = 751; NshCD2AP-1 = 453; NshCD2AP-2 = 356). I, Burst rate per minute (n = 3; NshControl = 753; NshCD2AP-1 = 456; NshCD2AP-2 = 355 bursts; ****p < 0.0001; ns, not significant; Kruskal–Wallis test). J, Interburst spike frequency (n = 3; NshControl = 24; NshCD2AP-1 = 29; NshCD2AP-2 = 22; *p < 0.05; **p < 0.01; ns, not significant; Kruskal–Wallis test). Further analysis is provided in Extended Data Figure 3-1. Data are presented as mean ± SD.
Figure 3-1
CD2AP knockdown reduces neuronal activity. (A) Network burst rate per minute (n = 3; ns, not significant, ordinary one-way ANOVA test). (B) Network burst duration in s (n = 3, NshControl = 24, NshCD2AP-1 = 29, NshCD2AP-2 = 22 bursts, ns, not significant, ordinary one-way ANOVA test). (C) Average spikes per network burst (n = 3, ns, not significant, ordinary one-way ANOVA test). (D) Interburst interval (n = 3, NshControl = 21, NshCD2AP-1 = 26, NshCD2AP-2 = 19 intervals, ns, not significant, ordinary one-way ANOVA test). Data are presented as mean ± SD. Download Figure 3-1, TIF file.
We found that the percentage of electrodes showing neuronal activity was similar in shControl and shCD2AP neurons (Fig. 3D). Most electrodes had low firing activity (∼1 Hz), as revealed by the distribution of active electrodes based on the mean firing rate. Interestingly, the electrodes with higher firing activity seemed reduced upon CD2AP knockdown (Fig. 3E). On average, electrodes firing >3 Hz decreased in shCD2AP-2 neurons and tended to decrease in shCD2AP-1 neurons compared with shControl neurons (Fig. 3F). Moreover, the average spike frequency was reduced in shCD2AP-1 neurons (6.0 Hz) and shCD2AP-2 neurons (5.9 Hz) compared with shControl neurons (6.7 Hz; Fig. 3G). The cumulative frequency of spiking neurons showed a lower frequency of high spiking neurons after shCD2AP-1 and shCD2AP-2 treatment than in shControl neurons (Fig. 3H). Together, these results support that loss of CD2AP mitigates neural excitability.
In the raster profiles (Fig. 2B), we also observed synchronized bursts of activity in the neuronal network. By calculating the burst rate (burst/min) in each active electrode, we found a lower burst frequency in shCD2AP-1 (8 burst/min) and shCD2AP-2 neurons (7 burst/min) compared with shControl neurons (13 burst/min; Fig. 3I). However, no significant changes were observed in the network burst rate (Extended Data Fig. 3-1A), network burst duration (Extended Data Fig. 3-1B), average of spikes per network burst (Extended Data Fig. 3-1C), and interburst interval (Extended Data Fig. 3-1D) with the knockdown of CD2AP. The interburst spike frequency reduced from 717.8 Hz in shControl neurons to 269.7 and 462.8 Hz in shCD2AP-1- and shCD2AP-2-treated neurons, respectively (Fig. 3J). These results indicate that the basic structure of the burst events and the time between them remained stable despite the overall reduction in frequency, consistent with a decrease in the synchronized activity of the neuronal network in the absence of CD2AP. Nevertheless, further investigation is necessary to explain these neuronal activity and excitability alterations.
CD2AP mutant increases spine density more than CD2AP wild-type overexpression
After establishing that CD2AP knockdown reduced spines and synaptic activity, we investigated the impact on synapses of the overexpression of CD2AP and of the K633R coding mutation in CD2AP identified in LOAD patients and associated with a higher risk of LOAD (Vardarajan et al., 2015). We transfected primary neurons with CD2APWT-GFP, CD2APK663R-GFP, and the control GFP plasmid at 7 DIV and evaluated spinal and synaptic alterations at 15 DIV (Extended Data Fig. 4-1A). We mutagenized CD2APWT-GFP to obtain CD2APK633R-GFP (Extended Data Fig. 4-1B). We noted an increase in the number of primary dendrites, their length, and the number of intersections upon overexpression of CD2APWT, which was unchanged by the LOAD mutation (Extended Data Fig. 4-1C–H).
The expression of CD2APWT and CD2APK633R in dendrites was especially evident in dendritic protrusion-like spines (Fig. 4A). The protrusion-like spines contained 50% of dendritic CD2APWT and even more of CD2APK633R (58%; Fig. 4B). This is a notable increase of 47 and 71% compared with the 34% of endogenous CD2AP found in spines (Fig. 1C). We assessed whether the CD2AP-positive protrusions were spines by labeling with PSD-95 (Fig. 4C). Colocalization analysis revealed that 64% of PSD-95 puncta were positive for CD2APWT, while only 47% were positive for CD2APK633R (Fig. 4D). This suggests that while most of the protrusions positive for CD2APWT are spines, the increased number of spine-like protrusions positive for CD2APK633R are likely aberrant and not spines.
The CD2AP LOAD mutation increases spine density and volume but not synapse density induced by CD2AP overexpression. As indicated, primary neurons overexpressing CD2APWT-GFP, CD2APK633R-GFP, or GFP as control. Info on protocols and plasmids used and data on impact on dendrite branching are provided in Extended Data Figure 4-1. A, Representative images of neurons overexpressing CD2APWT-GFP and CD2APK633R-GFP (green). White squares indicate dendritic segments magnified below. Scale bar, 10 µm. B, Quantification of the CD2APWT and CD2APK633R spinal enrichment, the fraction of dendritic signal in spines (n = 5; NCD2APWT = 33; NCD2APK633R = 32; **p < 0.01; unpaired t test). C, CD2APWT or CD2APK633R (gray, green) colocalization with PSD-95 (gray, magenta) alone and merged. Scale bar, 10 µm. D, Quantification of the colocalization of CD2APWT or CD2APK633R with PSD95 (the percentage of PSD-95; n = 3; NCD2APWT = 35; NCD2APK633R = 42 dendrites; ****p < 0.0001; unpaired t test). E, Representative time series of spines expressing GFP, CD2APWT, or CD2APK633R at 0, 3.5, 6.5, and 9.5 min, outlined in yellow, green, blue, and magenta, respectively. The panel on the right shows superimposed spine outlines to reveal spine motility over time. Scale bar, 1 µm. F, Quantification of spine motility index (n = 2; NGFP = 49; NCD2APWT = 112; NCD2APK633R = 92 spines; ****p < 0.0001; ns, not significant; Kruskal–Wallis test). G, Representative images of dendrites of neurons expressing with the volume marker Cherry (gray). Below is a 3D reconstruction using Imaris. Scale bar, 10 µm. H, Quantification of dendritic spine density (n = 3; NGFP = 35; NCD2APWT = 30; NCD2APK633R = 32 dendrites; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001; ns, not significant; ordinary one-way ANOVA). Data on dendritic spine subtype density are provided in Extended Data Figure 4-1A,B. I, Quantification of spine head volume based on Cherry volume marker (n = 3; NGFP = 49; NCD2APWT = 52; NCD2APK633R = 79 spines; **p < 0.01; ****p < 0.0001; Kruskal–Wallis test). J, Synapses identified by PSD-95 (magenta, gray) and vGluT1 (cyan, gray) colocalization in GFP, CD2APWT, or CD2APK633R-expressing neurons (fire LUT, green). Scale bar, 10 µm. K, Quantification of synapse density (n = 3; NGFP = 66; NCD2APWT = 66; NCD2APK633R = 99 dendrites; **p < 0.01; ns, not significant; Kruskal–Wallis test). L, Quantification of PSD-95 density (n = 3; NGFP = 70; NCD2APWT = 89; NCD2APK633R = 97 dendrites; **p < 0.01; ***p < 0.001; ns, not significant; Kruskal–Wallis test). Data on vGlut1 density and PSD-95 puncta area are provided in Extended Data Figure 4-2C,D. M, Active synapses labeled with FM4.64 (magenta) in GFP-, CD2APWT-, or CD2APK633R-expressing neurons upon high-potassium–mediated depolarization. Scale bar, 10 µm. N, Quantification of FM4.64 puncta density (n = 3; NGFP = 54; NCD2APWT = 54; NCD2APK633R = 51 neurites; ****p < 0.0001; ns, not significant; Kruskal–Wallis test). O, Quantification of FM4.64 puncta intensity (n = 3; NGFP = 53; NCD2APWT = 55; NCD2APK633R = 50 neurites; *p < 0.05; ns, not significant; Kruskal–Wallis test). Data on the FM4.64 puncta area are provided in Extended Data Figure 4-2F. Data are presented as mean ± SD.
Figure 4-1
CD2APWT, but not CD2APK633R, overexpression increases neurite number, length, and branching. (A) Overexpression experiment timeline indicating the day in vitro when neurons are transfected and fixed. (B) cDNA plasmids used encoding GFP, CD2APWT-GFP, and CD2APK633R-GFP. (C) Neurons overexpressing GFP, CD2APWT, and CD2APK633R (grey). (D) Quantification of neurites number in neurons expressing GFP, CD2APWT, and CD2APK633R (n = 6, NGFP = 42, NCD2APWT = 37, NCD2APK633R = 46, ***P < 0.001, ****P < 0.0001; ns, not significant, Kruskal-Wallis test). (E) Quantification of the number of intersections per 200 µm radius in neurons expressing GFP, CD2APWT, and CD2APK633R (n = 3, NGFP = 15, NCD2APWT = 17, NCD2APK633R = 17, *P < 0.05, **P < 0.01; ns, not significant, Kruskal-Wallis test). (F) Quantification of neurite length in neurons expressing GFP, CD2APWT, and CD2APK633R (n = 3, NGFP = 15, NCD2APWT = 17, NCD2APK633R = 17, **P < 0.01, ***P < 0.001; ns, not significant, Kruskal-Wallis test). (G) Low magnification (20x) images of neurons expressing GFP, CD2APWT, and CD2APK633R (green) and respective 3D IMARIS reconstructions. Scale bar: 50 µm. (H) Sholl analysis of the number of intersections of neurons expressing GFP (grey), CD2APWT (green), and CD2APK633R (magenta) (n = 3, NGFP = 18, NCD2APWT = 20, NCD2APK633R = 16). Data are presented as mean ± SD. Download Figure 4-1, TIF file.
Figure 4-2
The CD2AP LOAD mutation increases spine density and volume but not synapse density induced by CD2AP overexpression. (A) Representative 3D reconstruction with spine subtypes using IMARIS of dendrites of neurons expressing GFP, CD2APWT, or CD2APK633R. (B) Quantification of dendritic spine density per subtype in GFP (grey), CD2APWT (green), and CD2APK633R (magenta) (n = 3, NGFP = 23, NCD2APWT = 30, NCD2APK633R = 31 dendrites, *P < 0.05, ****P < 0.0001; ns, not significant, two-way ANOVA test). (C) Quantification of vGluT1 density in GFP, CD2APWT, or CD2APK633R neurites (n = 3, NGFP = 71, NCD2APWT = 94, NCD2APK633R = 98 density, **P < 0.01, ***P < 0.001; ns, not significant, Kruskal-Wallis test). (D) Quantification of PSD-95 average area (n = 3, NGFP = 67, NCD2APWT = 93, NCD2APK633R = 91 dendrites, *P < 0.05; ns, not significant, Kruskal-Wallis test). (E) Quantification of FM4.64 average area (n = 3, NGFP = 53, NCD2APWT = 55, NCD2APK633R = 50 neurites, ****P < 0.0001; ns, not significant, Kruskal-Wallis test). Data are presented as mean ± SD. Download Figure 4-2, TIF file.
Dendritic spines are highly motile in vitro and in vivo (Lendvai et al., 2000; Tashiro and Yuste, 2004). This motility is likely related to extensions and retractions as if searching for presynaptic partners. Spine motility also allows for maturation-related morphological changes (Bonhoeffer and Yuste, 2002). To examine rapid spine motility, live neurons were imaged every 30 s for 10 min (Fig. 4E) as previously (Tashiro and Yuste, 2004). We observed spines showing motility with expansions and contractions during imaging. We measured the spine area during the imaging and calculated the spine motility index to quantify the degree of spine motility (Dunaevsky et al., 1999). The motility index was quantified by calculating the difference between the largest and smallest spine area, normalized to the average spine area during the movie. While CD2APWT tended to decrease spine motility, only the CD2APK633R-positive spines significantly reduced motility by >50% (Fig. 4F). This reduced spine motility upon expression of CD2APK633R may be deleterious for spine function.
To analyze spine density and spine head volume analysis, we transfected CD2APWT or CD2APK633R and the volume marker mCherry into primary neurons (Fig. 4G). The spines positive for CD2APWT and CD2APK633R were denser and bulkier than the control (GFP; Fig. 4G). Using mCherry expression, we manually identified and counted the spines. The results revealed a 25% increase in spine density in neurons expressing CD2APWT (8.3 spines/10 µm), which increased by 53% in neurons expressing CD2APK633R (10.1 spines/10 µm) compared with control neurons (GFP; 6.6 spines/10 µm; Fig. 4H). The spine head volume measured upon 3D reconstruction with the Imaris software based on mCherry expression revealed a 26% increase upon overexpression of CD2APWT (0.058 µm3) and by 2.5-fold upon overexpression of CD2APK633R (0.115 µm3) compared with control spines (0.046 µm3; Fig. 4I). Spine subtype analysis using Imaris revealed that neurons expressing CD2APK633 showed a more significant increase in mushroom and stubby spines compared with neurons expressing CD2APWT, which only significantly increased stubby spines (Extended Data Fig. 4-2A,B). These findings, combined with the negative impact of CD2AP knockdown on spines (Fig. 2), suggest that CD2AP is a positive regulator of spine formation and growth. Furthermore, the LOAD K633R mutation augmented the morphologically more mature spines, mushroom, and stubby, suggesting it could enhance CD2AP spinal function.
To assess the impact of CD2APWT and CD2APK633R expression on excitatory synapses, we analyzed vGluT1 and PSD-95 colocalization (Fig. 4J). Synapse density increased with CD2APWT expression (Fig. 4K), as did the density of PSD-95 (Fig. 4L) and vGluT1 (Extended Data Fig. 4-2C). These results indicate that the postsynaptic expression of CD2APWT likely increased PSD-95, consistent with the induction of the formation and growth of mature spines. This postsynaptic maturation may translate into a presynaptic adaptation in the nonexpressing presynaptic neuron since vGluT1 accompanied the increase in PSD-95.
Synapse density only increased with CD2APWT expression (Fig. 4J–L). In CD2APK633R-expressing neurons, synapse density was not different from CD2APWT-expressing neurons, but the PSD-95 area was larger (Extended Data Fig. 4-2D). We also measured the density of active synapses using FM4.64 as before in neurons expressing CD2APWT or CD2APK633R compared with neurons expressing GFP (Fig. 4M). Interestingly, we observed a reduction of active synapses upon CD2APWT expression but not CD2APK633R expression compared with GFP neurons (Fig. 4N). Despite the unchanged density, the presynaptic terminals engaged with CD2APK633R-positive spines were less loaded with FM4.64 (Fig. 4O) and smaller (Extended Data Fig. 4-2E), suggesting a presynaptic impairment induced by postsynaptic CD2APK633R. The increase in synaptic connections due to the CD2APWT-induced spine growth did not reflect on more active synapses, suggesting a slower presynaptic maturation or inhibited synaptic vesicle cycle. Differently, the expression of LOAD CD2APK633R, despite increasing spine density and volume, did not enhance the synaptic connections and weakened synapse activity, which may be by interfering with spine motility and maturation.
These results suggest that CD2AP postsynaptic function is perturbed by the LOAD mutation K633R, negatively impacting spines and synapses.
CD2AP controls spinal F-actin
We have established that a LOAD risk variant affects the CD2AP function in the formation and growth of spines. CD2AP is a known F-actin-binding protein and a regulator of actin dynamics. Hence, we investigated whether CD2AP regulates spines through F-actin.
We colocalized CD2AP with F-actin, labeled with phalloidin, and cortactin, a spinal F-actin regulator (Hering and Sheng, 2003). We observed F-actin foci present in most spines, while CD2AP, like cortactin, was detected in a subset of spines (Fig. 5A). In higher magnification, a partial overlap between CD2AP and cortactin puncta with F-actin foci in spine heads was detected (Fig. 5A). We found a 30% colocalization of CD2AP with F-actin and cortactin (Fig. 5B).
CD2AP modulates spinal F-actin. A, CD2AP (green), F-actin (magenta), and cortactin (cyan) colocalization in dendrites. Scale bar, 5 µm. The white rectangle indicates magnified spines on the right. Scale bar, 0.5 µm. B, Quantification of CD2AP colocalization with F-actin and cortactin (n = 2; N = 13–14 dendrites; ns, not significant; unpaired t test). C, Spinal F-actin (gray; magenta) alone and colocalized with MAP2 (green) in dendrites of shControl-, shCD2AP-1-, or shCD2AP-2-treated neurons. Scale bar, 10 µm. D, Quantification of spinal F-actin in dendrites of neurons treated with shControl, shCD2AP-1, or shCD2AP-2 (n = 4; NshControl = 50; NshCD2AP-1 = 48; NshCD2AP-2 = 49 dendrites; ****p < 0.0001; ns, not significant; ordinary one-way ANOVA). E, Spinal F-actin (gray; magenta) in dendrites of neurons expressing GFP, CD2APWT, and CD2APK633R (green). Scale bar, 10 µm. F, Quantification of spinal F-actin in dendrites of neurons expressing GFP, CD2APWT, and CD2APK633R (n = 3; NGFP = 26; NCD2APWT = 30; NCD2APK633R = 34 dendrites; *p < 0.05; ns, not significant; Kruskal–Wallis test). G, Quantification of F-actin mean intensity in dendrites of neurons expressing GFP, CD2APWT, and CD2APK633R, normalized to GFP (n = 3; NGFP = 27; NCD2APWT = 30; NCD2APK633R = 34; *p < 0.05; **p < 0.01; ns, not significant; Kruskal–Wallis test). H, Quantification of the spinal F-actin average area in dendrites of neurons expressing GFP, CD2APWT, and CD2APK633R (n = 3; NGFP = 26; NCD2APWT = 29; NCD2APK633R = 34 dendrites; *p < 0.05; **p < 0.01; ns, not significant; Kruskal–Wallis test). I, Representative mCherry-actin (magenta) in the spines of dendrites of neurons expressing GFP, CD2APWT, and CD2APK633R (green). The right panels correspond to frames at the indicated times of the corresponding FRAP experiment of mCherry-actin (gray). Scale bar, 1 µm. The corresponding time lapse is provided in Movie 5-1. J, Representative FRAP traces of normalized mCherry-actin fluorescence (A.U.) in the spines expressing GFP (gray), CD2APWT (magenta), and CD2APK633R (green) are shown in I. K, Quantification of mCherry-actin FRAP recovery (n = 4; NGFP = 55; NCD2APWT = 458; NCD2APK633R = 55 spines; *p < 0.05, ns, not significant; Kruskal–Wallis test). Data on FRAP recovery per spine size are provided on Extended Data Figure 5-2A. L, Quantification of mCherry-actin FRAP recovery half-time (n = 4; NGFP = 55; NCD2APWT = 458; NCD2APK633R = 55 spines; *p < 0.05; ns, not significant; Kruskal–Wallis test). Data are presented as mean ± SD.
CD2AP modulates F-actin. mCherry-actin (gray) in the spines of dendrites of neurons expressing GFP, CD2APWT, and CD2APK633R before bleaching at 16.55 s and after during recovery (FRAP). Scale bar, 1 µm. [View online]
Next, we evaluated the impact of CD2AP knockdown on spinal F-actin. We observed less F-actin in the spines of shCD2AP-treated neurons (Fig. 5C). Quantification revealed that 34% of dendritic F-actin is in the spines of neurons treated with shControl. In neurons treated with shCD2AP-1 and shCD2AP-2, the distribution of F-actin between the shaft and spines reduced to 19 and 21% of dendritic F-actin, respectively (Fig. 5D).
In CD2APWT-expressing dendrites, the mean intensity of F-actin mean intensity increased, but the percentage of F-actin in spines (spinal F-actin) and their area was like that of control dendrites (GFP; Fig. 5E–H). In contrast, in dendrites expressing CD2APK633R, the percentage of spinal F-actin and their area increased while the dendritic F-actin was similar to that of CD2APWT-expressing dendrites (Fig. 5E–H). These results indicate that changes in F-actin likely mediate CD2AP-dependent spine expansion. Furthermore, the LOAD mutation in CD2AP may modify its function, inducing a buildup of F-actin in the spines.
To determine the impact of CD2APWT or CD2APK633R overexpression on actin turnover (Koskinen and Hotulainen, 2014), we measured the rate of actin filament assembly by coexpressing mCherry-actin (Koestler et al., 2008) and performing FRAP imaging experiments (Fig. 5I; Movie 5-1). Consistent with previous reports (Koskinen and Hotulainen, 2014), we observed that the fluorescence of mCherry-actin, after photobleaching correction and background subtraction, recovered in GFP-expressing spines at 76% of the prebleached levels (Fig. 5I–K). In spines expressing CD2APWT, mCherry-actin recovered (75%) similarly to control spines, suggesting that CD2APWT does not alter the stable fraction of F-actin (Fig. 5I–K). In contrast, the mCherry-actin recovery in spines expressing CD2APK633R was significantly decreased (65%) compared with spines expressing GFP and CD2APWT, suggesting that the LOAD mutation reduces the dynamic actin turnover by increasing the stable fraction of F-actin (Fig. 5I–K). Differently, the half-time of recovery increased in CD2APWT-expressing spines (9.5 s) and tended to increase in CD2APK633R-expressing spines (10.0 s) compared with GFP-expressing spines (6.5 s; Fig. 5L). This suggests that CD2AP may interfere with the binding of actin monomers to the polymerizing actin filaments or cap the growing F-actin barbed end, resulting in less dynamic actin turnover in spines. To correlate the rate of actin recovery with the head size of the spine, we categorized the spines as small (<0.2 µm2), intermediate, and large (>0.6 µm2; Extended Data Fig. 5-2). The large GFP control spines tended to recover less, likely more stable, while the larger CD2APWT- and CD2APK633R-expressing spines tended to recover more than larger GFP-expressing spines and smaller spines. Interestingly, the intermediate-size spines expressing CD2APK633R were significantly reduced compared with the control or intermediate spines expressing CD2APWT, supporting an increase in the stability of F-actin, especially in intermediate-size spines (Extended Data Fig. 5-2). CD2AP increases FRAP recovery, potentially by polymerizing F-actin, while the LOAD mutation does not, potentially by interfering with F-actin depolymerization.
Figure 5-2
CD2APK633R mutation decreases the mCherry-actin FRAP recovery, especially in the intermediate spine size. (A) Quantification of FRAP time recovery of mCherry-actin in spines expressing GFP (grey), CD2APWT (green) and CD2APK633R (magenta) per size category: small (<0.2 µm2), intermediate (0.2-0.6 µm2) and large (>0.6 µm2) (n = 3, NGFP(<0.2) = 5, NGFP(0.2-0.6) = 21, NGFP(>0.6) = 16, NCD2APWT(<0.2) = 2, NCD2APWT(0.2-0.6) = 34, NCD2APWT(>0.6) = 12, NCD2APK633R(<0.2) = 3, NCD2APK633R(0.2-0.6) = 30, NCD2APK633R(>0.6) = 22 spines, *P < 0.05, **P < 0.01, Mann Whitney test). Data are presented as mean ± SEM. Download Figure 5-2, TIF file.
The CD2AP LOAD mutation does not rescue synapses or spinal F-actin
Finally, we investigated whether the loss of F-actin-positive spines caused by CD2AP depletion could be rescued by the reexpression of CD2APWT or CD2APK633R (Fig. 6). Our findings revealed that CD2APWT rescued the reduction in PSD-95 density induced by shCD2AP-1 and shCD2AP-2 to control levels (Fig. 6A,B). In contrast, the CD2APK633R did not increase PSD-95 density (Fig. 6A,B). We also evaluated the impact on the PSD-95 area and observed a similar trend, with only the expression of CD2APWT significantly increasing the PSD-95 area compared with the neurons treated with shCD2AP-1 (Fig. 6C).
CD2AP wild-type but not mutant rescues spine density. shControl-, shCD2AP-1-, and shCD2AP-2-treated neurons expressing Myc, CD2APWT-MYC, or CD2APK633R-MYC. A, PSD-95 (green), F-actin (magenta) in dendrites of shControl- or shCD2AP-2-treated neurons expressing Myc, CD2APWT, or CD2APK633R (blue). Scale bar, 10 µm. B, Quantification of PSD-95 density (n = 4; NshControl + myc = 67; NshCD2AP-1 + myc = 83; NshCD2AP-2 + myc = 52; NshCD2AP-2 + WT = 18; NshCD2AP-2 + K633R = 23; NshCD2AP-1 + WT = 26; NshCD2AP-2 + K633R = 52 dendrites; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001; ns, not significant; Kruskal–Wallis test). C, Quantification of PSD-95 puncta area (n = 4; NshControl + myc = 66; NshCD2AP-1 + myc = 82; NshCD2AP-2 + myc = 49; NshCD2AP-2 + WT = 18; NshCD2AP-2 + K633R = 23; NshCD2AP-1 + WT = 22; NshCD2AP-1 + K633R = 50 dendrites; *p < 0.05; ns, not significant; Kruskal–Wallis test). D, Quantification of F-actin puncta density (n = 4; NshControl + myc = 67; NshCD2AP-1 + myc = 83; NshCD2AP-2 + myc = 50; NshCD2AP-2 + WT = 17; NshCD2AP-2 + K633R = 24; NshCD2AP-1 + WT = 22; NshCD2AP-1 + K633R = 40 dendrites; *p < 0.05; ***p < 0.001; ****p < 0.0001; ns, not significant; Kruskal–Wallis test). E, Quantification of F-actin puncta area (n = 4; NshControl + myc = 66; NshCD2AP-1 + myc = 82; NshCD2AP-2 + myc = 49; NshCD2AP-2 + WT = 18; NshCD2AP-2 + K633R = 23; NCshCD2AP-1 + WT = 22; NshCD2AP-1 + K633R = 50 dendrites). Data are presented as mean ± SD.
We additionally analyzed the density and area of F-actin puncta in dendrites. We found that CD2APWT but not CD2APK633R rescued F-actin density, similarly to PSD-95 (Fig. 6D). Concerning the F-actin area, the reduction was significant in both shCD2AP-1- and shCD2AP-2-treated neurons, and the rescue by CD2APWT reached significance in shCD2AP-2-treated neurons (Fig. 6E).
According to these results, CD2AP regulates synapses by promoting spine formation and growth. The LOAD mutation K633R impairs CD2AP function and disrupts its ability to control F-actin dynamics within spines. This supports a new specific function for CD2AP in controlling the balance between F-actin polymerization and depolymerization, which are essential for spine formation and growth. Moreover, it suggests that LOAD mutations may lead to faulty F-actin dynamics in spines, which can compromise synapses.
Discussion
Spines depend on the protrusive force of actin polymerization for filopodia formation and spine expansion by recruiting postsynaptic scaffolds, such as PSD-95, which allow synapse formation, plasticity, and strength. The dynamics of F-actin in spines are regulated by actin-binding proteins (Bosch and Hayashi, 2012). Here, we discovered that the Alzheimer's disease risk factor CD2AP, an actin-binding protein, is a postsynaptic protein required for spine formation and expansion, regulating the density of synapses and the electrophysiological activity of neurons via the regulation of F-actin polymerization in dendritic spines. CD2AP's coding variant (K633R) associated with LOAD likely interferes with its regulation of F-actin, making it incapable of rescuing the CD2AP function. Its excess induces spine-like protrusions, which are unable to establish active synapses. All these observations support CD2AP as an essential synaptic regulator whose mutation may contribute to LOAD development by impacting spine morphology and leading to progressive synapse dysfunction (Fig. 7, schematics).
Schematic diagram illustrating the key findings on the role of the Alzheimer's disease risk gene CD2AP in regulating dendritic spine formation and morphology through F-actin. Normally, CD2AP regulates spine formation and morphology via spinal F-actin (control). Loss of function of CD2AP decreases spines by loss of F-actin, losing synapses (CD2AP knockdown). While CD2APWT induces spine growth by polymerizing F-actin, gaining synapses (CD2APWT overexpression), CD2APK633R induces immotile larger spine-like protrusions by deregulated F-actin depolymerization, not gaining active synapses (CD2APK633R overexpression).
How does CD2AP promote spine formation and expansion?
CD2AP is likely a regulator of spine expansion and stabilization into a functional synapse via F-actin. Notably, CD2AP membrane recruitment is required for actin assembly, which also depends on F-actin (Spence et al., 2016). We demonstrated that CD2AP knockdown decreases spinal F-actin. In contrast, overexpression increased dendritic F-actin, supporting that CD2AP promotes F-actin directly (Lehtonen et al., 2002). CD2AP can cap the barbed ends of F-actin in vitro, inhibiting actin polymerization and depolymerization (Tang and Brieher, 2013; Wang and Brieher, 2020). The CD2AP capping of actin filaments can also promote branched F-actin dependent on ARP2/3 (Tang and Brieher, 2013), triggering filopodia maturation into spines through spine expansion (Spence et al., 2016). CD2AP-dependent expansion can sequentially recruit PSD-95, functionalizing the spine (El-Husseini et al., 2000). In agreement, CD2AP knockdown reduced spines, particularly the more mature mushroom subtype, PSD-95, making fewer synapses and compromising neuronal activity. On the contrary, when CD2AP was in excess, more spines formed, notably the stubby spine subtype; PSD-95 increased, with larger spines increasing synapses but not more active.
CD2AP could affect spines through its interaction with cortactin and capping protein (CP), which are known to regulate F-actin in spines (Hering and Sheng, 2003; Fan et al., 2011; Catarino et al., 2013; Cornelius et al., 2021). The CD2AP–cortactin interaction contributes to F-actin stabilization and accumulation in vivo (Wang and Brieher, 2020), and the knockdown of cortactin reduces the spine density (Hering and Sheng, 2003; Catarino et al., 2013; Cornelius et al., 2021), as CD2AP. Cortactin stabilizes the ARP2/3-dependent branched F-actin and delays its depolymerization (Weaver et al., 2001), suggesting a similar role of CD2AP. However, CD2AP may function independently from cortactin since its overexpression leads to longer spines with a minor impact on spine expansion (Hering and Sheng, 2003).
On the other hand, CD2AP–CP interaction inhibits CP (Bruck et al., 2006; Uruno et al., 2006; Takeda et al., 2010), which binds actin filaments, preventing the addition or loss of actin subunits (Billault-Chaumartin and Martin, 2019). However, CP knockdown affects spines differently from CD2AP (Fan et al., 2011), suggesting that CD2AP functions independently in spines. Nevertheless, the complex regulation of F-actin in dendrites (Konietzny et al., 2017) and spines (Honkura et al., 2008) by CD2AP and its interactors will require further investigation.
Impact of a LOAD mutation in CD2AP on spines
The CD2AP coding mutation (K633R; rs 116754410) found in patients with LOAD (Vardarajan et al., 2015) and recently in children with kidney disease (Nandlal et al., 2022) is predicted to be pathogenic. However, it is still unknown how K633R affects CD2AP in an AD-relevant way. Our data suggest that the CD2AP mutation (K633R) causes an aberrant gain of dysfunctional CD2AP in spines, increasing their density, particularly of the mushroom and stubby type and volume. Despite the morphological enlargement, the protrusion-like spines lack PSD-95 assembly and are engaged in less active synapses. This is likely due to the increased stability of F-actin in spines, leading to reduced spine motility and the impaired capacity to establish or sustain normal synapses (Bonhoeffer and Yuste, 2002). Rescue experiments further support that the CD2AP mutant is dysfunctional. It will be necessary to determine the impact of the K633R mutation on endogenous CD2AP and in vivo synaptic activity and plasticity to establish how it contributes to AD.
CD2AP as a risk factor for AD synaptic dysfunction
Individuals carrying CD2AP variants linked to AD are likely less pathogenic than the ones linked to kidney disease. The large CD2AP truncations that lead to kidney disease in children likely cause CD2AP deletion. Differently, a case report of an adult patient with kidney disease and mild cognitive decline had a single missense variation (Tsvetkov et al., 2016). The same mutation has been found in LOAD (Vardarajan et al., 2015).
Our data suggest that CD2AP coding mutations may only slightly alter CD2AP function, which may, with aging, enhance AD development.
Thus, it is crucial to determine their impact on synapses and other disease-relevant mechanisms. The most common variants of LOAD are noncoding and are predicted to alter CD2AP expression; however, it is unknown how CD2AP levels change in the brain. Our data on CD2AP knockdown support the hypothesis that a reduction in CD2AP expression or function may contribute to synapse dysfunction in AD.
Furthermore, the LOAD coding variant in CD2AP (K633R) impact on spine volume and motility suggests it may affect synaptic plasticity, which requires spine expansion to accommodate the increase in synaptic strength that underlies long-term potentiation and memory (Bosch and Hayashi, 2012).
Concerning the role of CD2AP in Aβ-dependent synapse dysfunction in LOAD, we found that the increase in intraneuronal Aβ production induced by CD2AP knockdown in dendrites (Ubelmann et al., 2017) was insufficient to account for the spine loss observed. These results indicate that CD2AP may cause synaptic dysfunction directly and upstream of Aβ production in early AD.
CD2AP may also contribute to the later stages of AD. It has been demonstrated to modulate tau-mediated mechanisms (Shulman et al., 2014) and the integrity of the blood–brain barrier (Cochran et al., 2015) and has been linked to cognitive functioning in familial AD (Manzali et al., 2021). Variations in the CD2AP locus are also associated with the burden of neuritic plaques (Shulman et al., 2013). CD2AP-positive neuronal inclusions, resembling neurofibrillary tau tangles, have recently been detected (Camacho et al., 2022).
Conclusion
CD2AP is a regulator of actin-rich dendritic spines, contributing to spine formation and expansion, mechanisms relevant to synaptic structural plasticity necessary for memory, profoundly affected in AD (Herms and Dorostkar, 2016).
Footnotes
We thank for the gift of plasmids Dr. M. Cormont (Univ. Nice), Dr. A. Shaw (U. Washington), Dr. M. Arpin (Institut Curie), Dr. A. Steffen (Helmholtz Centre for Infection Research), and Dr. D. Trono (EPFL). We thank M. Pinho and F. Mateus for their technical assistance. We thank L. Almeida for the Excel macros. We thank Dr. M. Remondes (iMM) for the valuable insight on MEA analysis. We thank the lab members for their helpful discussions and critical manuscript reading. We thank Dr. S. Marques (NOVA Medical School Animal Facility) and Dr. T. Pereira (NOVA Medical School Microscopy Platform). This project has received funding from iNOVA4Health–UID/Multi/04462/2019, a program financially supported by Fundação para a Ciência e Tecnologia (FCT)/Ministério da Educação e Ciência through national funds and cofunded by FEDER under the PT2020 Partnership Agreement; Maratona da Saúde 2016; and ALZ AARG-19-618007 Alzheimer’s Association, from the research infrastructure PPBI-POCI-01-0145-FEDER-022122 (FCT and Lisboa2020, under the PORTUGAL2020 agreement, European Regional Development Fund). C.G.A. has been supported by CEECIND/00410/2017 and CEEC/iNOVA4Health (FCT); F.S.M. received an FCT doctoral fellowship (PD/BD/128344/2017). J.C. is a recipient of an FCT doctoral fellowship (2020.04851.BD).
The authors declare no competing financial interests.
↵F.S.M.’s present address: Department of Psychiatry, Washington University School of Medicine, St. Louis, Missouri 63110
- Correspondence should be addressed to Cláudia Guimas Almeida at claudia.almeida{at}nms.unl.pt.
