Introduction

The basal ganglia is functionally responsible for sensorimotor coordination, including caudate, putamen, globus pallidus (GP), substantia nigra (SN), and subthalamus nucleus (STN) [1]. Basal ganglia is not only related with movement disorders such as Huntington’s disease, Parkinson’s disease, and dystonia, but also associated with mood disorders such as obsessive-compulsive disorder and depression [2]. More recently, basal ganglia is also suggested to mediate symptoms underlying bipolar disorder (BD) based on the reports of patients who developed BD subsequent to basal ganglia lesions [3]. BD is a mental disorder characterized manic and depressive episodes. Enlargement of putamen was observed in first-episode BD patients [4]. Reduced times of T1 relaxation, an imaging technique sensitive to pH, was observed in the basal ganglia of BD patients [5]. Metabolic abnormalities such as N-acetylaspartate, phosphocholine, glycerophosphocholine, creatine, phosphocreatine levels in the basal ganglia measured by magnetic resonance spectroscopy may be associated with mood states in BD patients [6]. Using resting-state functional connectivity of basal ganglia, the clustering accuracy of discriminating BD patients from healthy controls could achieve 90% [7].

The basal ganglia as a whole is for movement regulation, while each nuclei (brain region) has its distinct role. As the main input nuclei, caudate and putamen receive axons from nearly all parts of cortex to the basal ganglia. In comparison, GP and SN are the main output nuclei [8, 9]. For caudate and putamen, they execute different functions in higher level learning and memory, with the former pivotal in goal-directed action, and the latter serving cognitive functions more limited to stimulus-response [10]. Functional connectivity studies also indicated a clear link between human caudate and putamen to different brain regions, that is, to executive frontal regions and basic sensorimotor areas, respectively [11, 12]. Sixty-four risk loci for BD have been identified by genome-wide association studies (GWAS) and those loci are enriched in genes involved in synaptic signaling [13]. Bulk transcriptomic analyses of subgenual anterior cingulate cortex and amygdala from human postmortem brains with BD have suggested downregulated neuroimmune and synaptic pathways [14]. However, there is still limited understanding of cell type-specific and basal ganglia nuclei-specific alternations associated with BD.

In this study, we performed single-nucleus RNA sequencing of 30,752 nuclei from caudate, putamen, GP, and SN of control human postmortem brain and 24,672 nuclei from BD brain. Integrative analyses of differential expression, gene co-expression (WGCNA), and gene regulatory network (SCENIC) indicated phosphodiesterase (PDE) 5 A, a cyclic guanosine monophosphate (cGMP)-specific phosphodiesterase, to be a significantly upregulated gene and potent target in BD especially in caudate medium spiny neurons (MSNs). Changes of PDE genes in the basal ganglia have been widely reported in psychiatric disorders [15], but the role of PDE5A in BD remains elusive, we then validated the effects of PDE5A upregulation in cultured mouse MSNs in vitro and in adult mouse striatum in vivo. Overexpression of PDE5A in primary cultured mouse lateral ganglion eminence (LGE)-derived striatal neurons led to reduced dendrite complexity, elevated apoptosis, and increased neuronal excitability and membrane resistance. This effect could be rescued by PDE5 specific inhibitor tadalafil. Similarly, overexpression of PDE5A in mouse striatum by stereotaxic injection resulted in a decrease in cGMP levels, an upregulated gene expression profile of neuroinflammation, and BD-like behaviors. Overall, our findings unveiled a causal role of increased PDE5A expression in the basal ganglia of BD, and implies a potential novel application of existing PDE5 inhibitors in the treatment of psychiatric diseases.

Results

To understand cell type-specific gene expression and network changes in BD basal ganglia, we isolated single nuclei from caudate, putamen, SN, and GP of one control and one BD human postmortem brain using modified methods [16,17,18], followed by library preparation for 10× single-nucleus RNA sequencing (donor information in Table S1). Single nuclei from control and BD basal ganglia were filtered based on four criteria: expression of each gene in more than 3 cells, number of expressed genes per cell (1000–10,000), number of transcripts measured per cell (<=50,000), and percentage of mitochondrial genes (<=2%) (Fig. S1). After quality filtering, 30,752 and 24,672 single nuclei from control and BD basal ganglia were kept for subsequent analyses, respectively (Table S2). Note that cells in top 15% pANN (doublet score) were removed as doublets by doubletFinder [19].

Cellular diversity and marker genes in control and BD basal ganglia

All the cell nuclei were clustered using the Seurat software (version 4.1.1) on R (https://github.com/satijalab/seurat). Uniform manifold approximation and projection (UMAP) was used for dimension reduction and visualization of 30,752 and 24,672 single nuclei from control and BD basal ganglia into 20 and 19 clusters, respectively (Fig. 1A, B). For cell type annotation, we applied preferential expression measure (PEM) score to rank the specificity of gene expression pattern to a specific cell type [20]. Briefly, each cluster was PEM-scored to identify the top 100 marker genes, then each gene was matched to a specific cell type based on known cell type marker genes (Table S3), and at last the marker genes of each cell type in each cluster were counted (Fig. 1C, D and Table S4). The 20 clusters in control basal ganglia were annotated as oligodendrocyte subtype 1 and 2 (Oligo-1 and -2, clusters 0 and 3), astrocyte subtype 1 and 2 (Astro-1 and -2, clusters 5 and 8), endothelial cell subtype 1, 2 and 3 (Endo-1, −2, and −3, clusters 7, 9, and 14), oligodendrocyte precursor cell (OPC, cluster 6), microglia (Micro, cluster 4), ependymal cell (Ependyl, cluster 19), and neurons (clusters 1, 2, 10–13, 15–18). The 19 clusters in BD basal ganglia were annotated as Oligo-1 and -2 (clusters 0 and 2), Astro-1 and -2 (clusters 4 and 13), Endo-1, −2 and −3 (clusters 8, 11, and 17), OPC (cluster 7), microglia (cluster 10), Ependyl (cluster 18), and neurons (clusters 1, 3, 5, 6, 9, 12, 14–16). Annotations were mainly based on the expression of known marker genes [21,22,23] such as PLP1 (oligodendrocyte marker), AQP4 (astrocyte marker), APBB1IP (microglia marker), ITIH5, LEF1 (endothelial markers), PCDH15 (OPC marker), and SPAG17 (ependymal cell marker) (Fig. 1E, F). For neuronal clusters, SYT1, MYT1L, GRIA1, CADPS, and GABRB3 are pan-neuronal markers, while PPP1R1B (medium spiny neuron (MSN) marker), DRD1 (MSN-D1 marker), DRD2 (MSN-D2 marker), ELAVL2/CLSTN2 (interneuron (interN) markers), and TH/SLC6A3 (dopaminergic neuron (DaN) markers) are neuronal subtype-specific markers. Thus, we annotated control neuronal clusters as D1-MSN, D2-MSN-1, D2-MSN-2, D1/D2-MSN-1, D1/D2-MSN-2, interN-1, interN-2, interN-3, interN-4, and DaN, respectively (Fig. 1E), and BD neuronal clusters as D1-MSN-C, D1-MSN-P, D2-MSN-C, D2-MSN-P, D1/D2-MSN-1, D1/D2-MSN-2, interN-1, interN-2, and interN-3, respectively (Fig. 1F).

Fig. 1: Cell type taxonomy in control and BD basal ganglia.
figure 1

A, B Unbiased clustering and visualization of snRNA-seq data from control and BD basal ganglia using UMAP, respectively. Cell types were annotated according to expression of known marker genes. C, D Heatmap with scaling by column showed the relative number of predefined cell type markers overlapped with top 100 genes ranked by PEM scores for control and BD basal ganglia, respectively. E, F Markers and expression statistics used for defining cell types in control and BD basal ganglia, respectively. Micro, microglia; D1/D2-MSN-1/2, dopamine receptor D1/D2 containing medium spiny neuron subtype 1/2; DaN, dopaminergic neuron; InterN-1/2/3/4, interneuron subtype 1/2/3/4; Endo-1/2/3, endothelial subtype 1/2/3; Astro-1/2, astrocyte subtype 1/2; Oligo-1/2, oligodendrocyte subtype 1/2; OPC, oligodendrocyte precursor cell; Ependyl, ependymal cell.

Full size image

Note that BD neuronal clusters did not include DaNs. To confirm this, we combined data of control and BD samples and removed batch effects by Seurat. However, the DaN population was still contributed dominantly by cells from control donor (Fig. S2A, B). Furthermore, we examined the expression of DaN marker gene TH in both control and BD samples (Fig. S2C, D). TH was prominently expressed in DaNs from control (Fig. S2C) but almost absent in BD sample (Fig. S2D). These findings suggest the lack of DaNs in the BD sample we tested. Dopaminergic degeneration is a hallmark of Parkinson’s disease [24]. BD has also been linked with dopaminergic dysfunction. Although a large-scale association study has suggested that patients with BD have a significantly increased risk of developing Parkinson’s disease [25], there is no systematic investigations indicating dopaminergic neuron loss in BD basal ganglia. Thus, we would consider the missing DaN cluster in BD basal ganglia here is case-specific but not a general phenomenon or due to the difficulty in capture of DaNs.

For each main cell type, its subtype markers were also shown (Fig. S3). D1/D2-MSNs were clustered into two subtypes with markers FOXP2/GRIN2A/TRHDE and HTR2C/COCH/EYA1 for subtype 1 and 2, respectively (Fig. S3A). Interneurons were divided into three subtypes with markers CDH18/SORCS1/VWC2L, PDE3A/GRIK1, and STXBP6/TMEM163 for subtype 1, 2, and 3, respectively (Fig. S3B). Control basal ganglia had one additional interneuron subtype, namely subtype 4, with NOS1/CRHR2 expression. Oligodendrocytes were clustered into two subtypes with markers OPALIN/PLLP and SEMA5A/ENPP6, respectively (Fig. S3C). Astrocytes were divided into two subtypes with TNC/SLC6A11 and RGS20/GLI3 expression, respectively (Fig. S3D). At last, endothelial cells included three subtypes with markers MECOM/ERG, PDGFRB/RGS5, and APOD/CYP1B1, respectively (Fig. S3E).

Dysregulated PDE genes and ion channels in BD basal ganglia

When comparing cell type ratio between control and BD basal ganglia, we observed that MSNs including D1, D2, and D1/D2 subtypes were dominantly from caudate and putamen but not GP and SN (Fig. 2A and Table S5), consistent with prior knowledge. Interestingly, BD D1-MSNs or D2-MSNs were distinct between caudate and putamen (Fig. 2A). Moreover, BD oligodendrocytes showed a decreased ratio in the caudate but a significantly elevated ratio in putamen, SN and SP (Fig. 2B).

Fig. 2: Differential gene expression and enriched pathways between control and BD basal ganglia.
figure 2

A Cell ratio of each cell type among four nuclei in control and BD basal ganglia, respectively. B Cell ratio of different cell types in each nucleus in control and BD basal ganglia, respectively. C, caudate; GP, globus pallidus; P, putamen; SN, substantia nigra. C number of DEGs for each cell type between control and BD basal ganglia. D Selected upregulated gene ontology biological pathways (GO-BPs) enriched for DEGs between control and BD caudate MSNs. Number of genes involved in each GO term was indicated. Note that PDE genes especially PDE5A was substantially upregulated in BD caudate MSNs including D1, D2, D1/D2 subtypes. Violin plots of PDE5A and PDE1C were shown. E Upregulated GO-BPs enriched for DEGs between control and BD SN interneurons. Those GO terms were further categorized into synapse and ion transport (orange), neural development (green), cell adhesion and migration (blue), signal transduction (red), and others (gray). Upregulated ion channels were evident in BD SN interneurons as shown by violin plots of KCNQ5, KCNJ6, KCNH5, KCND2, KCNQ3, SCN9A, CACNA2D1, and CACNA2D3.

Full size image

Differential expression analysis of each cell type between control and BD basal ganglia nuclei was performed and number of DEGs revealed that the main upregulated differential expression between control and BD basal ganglia was present in caudate MSNs and then SN interneurons, while the main downregulated expression was found in GP microglia and interneurons (Fig. 2C and Table S6). GO Enrichment analysis of those upregulated DEGs indicated that BD caudate MSNs were enriched in positive regulation of synapse assembly, regulation of GTPase activity and response to cocaine (Fig. 2D and Table S6), while BD SN interneurons were enriched in biological processes categorized into synapse and ion transport, neural development, cell adhesion and migration, and signal transduction (Fig. 2E and Table S7).

In BD caudate MSNs, phosphodiesterase 5 A (PDE5A) was one of the highest upregulated DEGs (fold change (FC) = 3.22, p = 0 for D1-MSN; FC = 3.15, p = 0 for D2-MSN; FC = 2.58, p = 2.53E-27 for D1/D2-MSN-1; FC = 2.70, p = 1.35E-16 for D1/D2-MSN-2). Other upregulated PDEs included PDE1A (FC = 2.15, p = 8.67E-08 for D1/D2-MSN-1) and PDE1C (FC = 1.56, p = 3.98E-62 for D2-MSN) (Fig. 2D and Table S6). PDEs are involved in the homeostasis of cAMP and cGMP, which are two key second messengers that modulate a series of neurobehavioral functions including memory and cognition [15]. PDE5A is implicated in Huntington’s disease (HD) [26], Alzheimer disease (AD) [27], and major depression disorder (MDD) [28], while PDE1A and PDE1C are involved in autism spectrum disorder (ASD) [29], schizophrenia (SCZ) [30], MDD [31], Parkinson’s disease (PD) [28], and AD [27].

In BD SN interneurons, potassium channels KCNQ5 (FC = 9.62, p = 7.54E-38), KCNJ6 (FC = 2.34, p = 1.24E-09), KCNH5 (FC = 2.20, p = 1.33E-11), KCND2 (FC = 1.76, p = 1.57E-07), and KCNQ3 (FC = 1.58, p = 2.70E-07), sodium channel SCN9A (FC = 1.45, p = 4.09E-09), and calcium channels CACNA2D1 (FC = 1.83, p = 6.54E-06) and CACNA2D3 (FC = 1.43, p = 1.68E-05) were all upregulated (Fig. 2E and Table S6). This is consistent with the hypothesis that BD may be an ion channelopathy [32]. Other genes are also reported to be involved in BD, such as THSD7B (FC = 4.96, p = 1.57E-21), SNPs of which were associated with BD in a GWAS study [33], and NRG1 (FC = 4.38, p = 9.45E-31), which was associated with schizophrenia and to a less extent, with BD [34] (Table S6).

For downregulated DEGs, BD GP interneurons were enriched in biological processes categorized into synapse and ion transport, neural development, and cell migration (Fig. S4A and Table S7). In addition, BD GP microglia showed decreased biological processes in signal transduction, cell proliferation and migration (Fig. S4B and Table S7).

Interestingly, PDEs and ion channels were downregulated in BD GP interneurons and microglia. PDE1A (FC = −1.76, p = 4.72E-10), PDE4D (FC = −0.61, p = 8.70E-05), potassium channel KCNB2 (FC = −1.09, p = 1.88E-09), KCNT2 (FC = −1.02, p = 1.25E-04), KCNQ5 (FC = -0.91, p = 0.004), and KCNQ3 (FC = −0.70, p = 0.003), calcium channel CACNA1E (FC = −0.91, p = 2.96E-05), sodium channel SCN9A (FC = −0.62, p = 0.006) were downregulated in BD GP interneurons (Fig. S4A and Table S6). Similarly, PDE3B (FC = −0.91, p = 2.71E-27), PDE9A (FC = −0.71, p = 1.37E-11), potassium channel KCNMA1 (FC = −1.27, p = 1.20E-56), and calcium channel CACNA1A (FC = −0.75, p = 8.28E-12) were downregulated in BD GP microglia (Fig. S4B and Table S6).

In short, dysregulated PDEs and ion channels were observed across different BD basal ganglia nuclei and cell types.

MSN-associated gene module and PDE5A-containing hub gene network of BD caudate revealed by WGCNA analysis

To ask whether there might be co-regulated gene modules associated with BD and whether such an association is cell type-specific, weighted gene co-expression network analysis (WGCNA) was performed on 1728 DEGs from all cells between control and BD basal ganglia nuclei. Eleven gene modules were identified (Fig. 3A). To ask whether the 11 gene modules were differentially distributed in the 11 main cell types and 4 basal ganglia nuclei, correlation analysis was performed (Fig. 3B, C). BD caudate MSNs and BD SN endothelial cells were strongly correlated with green and cyan module, respectively, which were consistent with the eigengene expression across cell types in the four basal ganglia nuclei (Fig. 3D, E). A significant portion of the annotated green module genes was enriched in chromatin remodeling (PAK1, BICRAL, GATAD2B) (Table S8). To better explore possible functional consequences of the transcriptional alternations related to green module, Hub gene-based network was generated (Fig. 3F). The network revealed only few annotated genes, such as PDE5A.

Fig. 3: BD caudate MSN-associated gene module and hub gene network by WGCNA analysis.
figure 3

A Dendrogram showing the gene modules identified in DEGs between human control and BD basal ganglia nuclei using WGCNA. Hierarchical clustering of genes with topological similarity was shown. Upper band indicated the module assignments after applying the dynamic tree cut algorithm. Lower band showed the gene modules after merge with highly correlated eigengenes. B Heatmap indicating the correlation between module eigengenes and cell type identities. C Heatmap showing the correlation between module eigengenes and other traits, such as number of transcripts per cell (nCount_RNA), number of expressed genes per cell (nFeature_RNA), and identities of control, BD, and brain regions. D, E Bar plot showing the eigengene expression of the gene module green and cyan among cell types between control and BD, respectively. F Network of hub genes identified in module green. The circle size is positively correlated with module membership of genes, and thickness of line is positively correlated with intergenic correlation. G Representative immunofluorescent staining images of human caudate sections stained with DARPP32 (green) and PDE5A (red). Scale bar, 100 μm. Zoom-in image scale bar, 20 μm. Cell ratio of DARPP32 + PDE5A+ in DARPP32+ cells between healthy control (HC) and bipolar disorder (BD) was compared (Shapiro-Wilk test, n = 7 pairs of donors, 2–6 fields per donor, *p < 0.05).

Full size image

This was consistent with the significantly upregulated RNA expression of PDE5A in BD caudate neurons including D1-, D2-, and D1/D2-MSNs, and interneuron subtypes as revealed by single-nucleus RNA seq (Fig. 2D). To further validate PDE5A upregulation in BD caudate MSNs, we performed immunostaining on caudate sections from 7 pairs of control and BD human postmortem brains (2 pairs from Fudan Brain Bank, and 5 pairs from Douglas-Bell Canada Brain Bank, Table S1), and observed that the ratio of PDE5A+DARPP32+ in DARPP32+ cells was significantly increased in BD caudate (Fig. 3G and Table S9), indicating upregulation of PDE5A in BD caudate MSNs.

PDEs and ion channels are frequent targets of multiple transcriptional factors among different cell types and basal ganglia nuclei of BD

Cellular transcriptional state is mediated by an underlying gene regulatory network (GRN) where a limited number of transcriptional factors (TFs) and cofactors regulate each other and their downstream target genes which are termed as regulons. To determine cell type-specific and nuclei-specific GRNs in BD basal ganglia, we performed a differential regulon activity analysis between control and BD basal ganglia by single-cell regulatory network inference and clustering (SCENIC). The strong differential regulon activity was consistent with the high expression of dominant TFs. For example, XRCC4 was specifically high-expressed in D1- and D2-MSNs of BD caudate (Fig. 4A); ARNTL was predominantly expressed in all MSN subtypes and interneurons of BD putamen (Fig. 4B); CREB5 was specifically high-expressed in astrocytes of BD GP (Fig. 4C); and NR2F2 was uniquely high-expressed in endothelial cells of BD SN (Fig. 4D). Interestingly, PDEs and ion channels were frequently found as downstream targets of those TFs (Fig. 4E). Among different cell types and basal ganglia nuclei, we observed some TFs were shared by 3–4 nuclei, such as SREBF1 (shared by all four nuclei), RUNX2, BCL11A, and SOX6 (shared by three nuclei) (Fig. 4E). Again, PDEs and ion channels often appear as targets of these TFs (Fig. 4E).

Fig. 4: Gene regulatory network of BD basal ganglia by SCENIC analysis.
figure 4

A−D Heatmaps showing the differential regulon activity of cell type-specific transcription factors between BD and control caudate, putamen, GP, and SN, respectively. Lower panels show violin plots of representative TF expression across different cell types between BD and control. E A summary table showing TFs shared by different basal ganglia nuclei with frequent targets of PDEs and ion channels.

Full size image

In short, our results highlight PDEs and ion channels are frequent targets of multiple TFs among different cell types and basal ganglia nuclei of BD.

Overexpression of PDE5A in primary mouse striatal neurons led to reduced dendrite complexity, elevated apoptosis, and increased neuronal excitability and membrane resistance

PDE5A is a cGMP-specific phosphodiesterase and mainly controls cGMP pool via NO/NO-GC activation. The NO/cGMP pathway has been indicated in many processes in the brain including synaptic transmission and neuroprotection [35]. Additionally, cGMP signaling activation can both have a spine formation-promoting effect by VASP [36] and a dendrite-retracting role by RhoA [37]. To understand the potential role of PDE5A upregulation in BD caudate MSNs, we applied primary cultured mouse E16-17 lateral ganglionic eminence (LGE)-derived MSNs as a model to study the effect of PDE5A overexpression on MSN morphology, survival, and excitability. Full length of mouse PDE5A (GeneID: NM_153422.3, 2598 bp) was overexpressed with rAAV2/9-hSyn-PDE5A-EGFP (BrainVTA, Wuhan). Neurons were infected with PDE5A overexpressed (OE) or control virus (NC) at DIV3 and grown to DIV14 for analyses.

PDE5A OE group showed an efficient overexpression as shown by western blot (Fig. 5A) and a defect in dendritic development as shown by MAP2 staining (Fig. 5B). Sholl analysis of neuronal morphology revealed that PDE5A overexpression sharply decreased the complexity of dendritic arbors (number of intersections per 10 μm concentric ring) between 10 and 200 μm radii from the soma (Fig. 5C and Table S10). Furthermore, PDE5A overexpression decreased total dendritic length by ~70%, relative to control MSNs (Fig. 5D and Table S10). This suggests PDE5A upregulation could attenuate MSN dendritic growth and reduce dendritic arbor complexity. To test whether PDE5A inhibition could rescue the defect, we treated OE neurons with PDE5 specific inhibitor tadalafil (10 μM) or vehicle (DMSO) at DIV7 and DIV10 for 12 h. The two groups were named as OE + tadalafil and OE + DMSO, respectively. Expectedly, tadalafil treatment could rescue the decreased dendritic complexity caused by PDE5A overexpression while vehicle did not (Fig. 5B–D and Table S10).

Fig. 5: PDE5A overexpression in primary cultured mouse MSNs led to compromised dendrite arbor development and increased apoptosis.
figure 5

A Western blot of PDE5A expression in control (NC) and PDE5A overexpressed (OE) MSNs. GAPDH was used as internal control. Quantification of PDE5A protein levels was calculated by relative gray value (Shapiro-Wilk test, n = 3 batches, ***p < 0.001). B Representative images of DIV14 MSNs expressing GFP and stained with MAP2 (a dendritic marker) from NC, OE, OE+tadalafil, and OE + DMSO MSNs, respectively. Scale bar, 20 μm. C Sholl analysis shows the number of dendritic intersections plotted against the distance from the center of the soma (2 way ANOVA Sidak’s multiple comparison test, p < 0.05 for 30–130 μm distance, p < 0.05 between NC and OE + tadalafil group in the range of 100–110 μm). D Sholl analysis shows total length of dendrites. 40 neurons each group from three independent batches were statistically analyzed (Shapiro-Wilk test,**p < 0.01 ***p < 0.001). E Representative images of DIV14 MSNs expressing GFP and stained with caspase 3 (an apoptosis marker) from the four groups. Scale bar, 150 μm. White arrow indicated apoptotic cells. F Ratio of caspase 3+ GFP+ cells in GFP+ cells was compared in the four groups (Shapiro-Wilk test, n = 3 batches, 3–4 fields per batch, ***p < 0.001).

Full size image

Subsequently, we examined the effect of PDE5A overexpression on MSN survival by caspase 3 staining. We found a significant increase of apoptotic MSNs by ~50% upon PDE5A overexpression, as shown by the ratio of caspase 3+GFP+ cells in GFP+ cells (Fig. 5E, F and Table S11), indicating PDE5A upregulation could lead to MSN apoptosis. Similarly, to test whether this deficit could be rescued by PDE5A inhibition, we treated OE neurons with tadalafil or DMSO. Tadalafil but not DMSO treatment could prevent OE neurons from apoptosis, recovering to a level similar as that of control group (Fig. 5E, F and Table S11).

We next examined the effect of PDE5A overexpression on neuronal excitability by patch clamp recording on cultured MSNs. The number of action potentials (spikes) produced by depolarizing current steps was recorded. PDE5A overexpression significantly increased spike number in the 0–70 pA range (Fig. 6A, B and Table S12, two-way ANOVA followed by post-hoc Šidák’s multiple comparisons, p < 0.05). Besides, PDE5A overexpression notably increased the excitability of cultured neurons as shown by significantly increased slope per 10 pA (Fig. 6C and Table S12, p < 0.05). Membrane capacitance is often measured as a way of determining cell surface area [38] and determining the propagation velocity of action potentials, which is inversely proportional to membrane capacitance [39]. Although no significant change was observed in membrane capacitance (Fig. 6D and Table S12), PDE5A overexpression caused a notably increase in membrane resistance (Fig. 6E and Table S12, p < 0.05), which may suggest a decrease of open ion channels and thus less leak out, contributing to an increased excitability. Together, these data suggest that PDE5A upregulation could cause an increase in neuronal intrinsic excitability and altered the properties of neuronal membranes.

Fig. 6: PDE5A overexpression in cultured striatal neurons led to increased neuronal excitability and membrane resistance.
figure 6

Whole-cell patch clamp recording was performed on primary cultured mouse E16-17 LGE-derived MSNs infected with control or rAAV2/9-hSyn-PDE5A-EGFP virus (A−E). A Representative traces for neuron spikes elicited by current steps. B The number of action potentials in response to step current injections (from 0 to100 pA: F (10, 320) = 1.337, p = 0.2096; from 0 to 70 pA: F (7, 224) = 2.782, p = 0.0309; 50 pA: p = 0.0185, two-way ANOVA followed by Sidak test). C The slope of the spikes per 10 pA (t (32) = 2.156, p = 0.0248, unpaired two-tailed t-test). D, E Comparison of membrane capacitance and membrane resistance between control and PDE5A OE group, respectively (t (32) = 0.3924, p = 0.6974 for (D); t (32) = 2.556, p = 0.0156 for (E), unpaired two-tailed t-test). All values were presented as mean ± SEM, *p < 0.05. n = 17 neurons from 3 batches of primary culture for each group.

Full size image

Overexpression of PDE5A in mouse striatum resulted in decreased cGMP levels, upregulated expression profile of neuroinflammation, and BD-like behaviors

To further understand the role of PDE5A upregulation in vivo, we utilized a model of PDE5A overexpression by stereotaxic injections of control or PDE5A OE virus into adult mouse striatum to investigate whether the behavioral presentations upon PDE5A upregulation could mimic BD symptoms in patients and delineate the underlying molecular pathways. A battery of behavioral tests were conducted to evaluate the effect of PDE5A upregulation on hyperactivity reflecting positive symptoms (by catwalk, rotarod, and open field tests), social withdrawal (by three chamber test) and anhedonia (by sucrose preference test) reflecting negative symptoms, working memory (by Y maze), sensorimotor gating (by prepulse inhibition (PPI)), depression (by open field and forced swimming tests), and mania-like state (by forced swimming test).

For locomotor activity, PDE5A OE mice showed little change in run and coordination charateristics such as cadence, run duration, speed, and regularity index as shown in catwalk gaiting test (Fig. 7A and Table S13). Moreover, in open field test, PDE5A OE mice displayed neither obvious hyperactivity as shown by distance moved/total moved distance nor depression as shown by performance in the center (Fig. 7C and Table S13). However, PDE5A OE mice did show a remarkable decline in the time spent on the rod in the rotarod test (Fig. 7B and Table S13), indicating a decline in motor balance. This suggests PDE5A upregulation in striatal neurons did not cause hyperactivity, which is associated with positive symptoms.

Fig. 7: PDE5A overexpression led to reduced motor balance, sensorimotor gating, sociability and triggered mania-like state.
figure 7

A series of behavioral tests were performed on control and PDE5A OE mice 4 weeks after stereotaxic injections with one week interval between tests. A Comparisons of run and coordination characteristics by catwalk gaiting analysis (Shapiro-Wilk test). B Comparison of time on rotarod by rotarod test (Shapiro-Wilk test). C Comparisons of the distance moved every 5 min, total distance moved, number of entries into the center, time spent in the center, and duration before first entry into the center by open field test (2 way ANOVA Sidak’s multiple comparison test for left first panel, Shapiro-Wilk test for other panels). D Comparisons of interaction time with stranger mouse 1 (S1) and empty cage (E), interaction time with stranger mouse 2 (S2) and S1, sociability (time spent ratio of (S1 - E) to (S1 + E)), and social novelty (time spent ratio of (S2 - S1) to (S2 + S1)) by three-chamber test (2 way ANOVA Sidak’s multiple comparison test for left first and third panels, Shapiro-Wilk test for other panels). E. Comparison of the percentage of sucrose consumption relative to the total consumption of water and sucrose solution within 36 h by sucrose preference test (Mann–Whitney U test). F Comparison of the percentage of correct alternations in the Y-maze task (Shapiro-Wilk test). G Comparison of pre-pulse inhibition level was assessed at pre-pulse intensities of 74, 78, and 82 dB by PPI test (2 way ANOVA Sidak’s multiple comparison test). H Comparison of immobility time during the last 4 min in water by forced swimming test (Shapiro-Wilk test). n = 10 control and 9 PDE5A OE mice for (A), n = 9 control and 9 PDE5A OE mice for (B−H), due to one death of control group in the process. *p < 0.05; **p < 0.01; ***p < 0.001.

Full size image

The negative symptoms include social withdrawal and anhedonia. We conducted three-chamber test to evaluate social behaviors of the mice. The three-chamber test includes three phases: phase 1 for habituation, phase 2 for the evaluation of sociability to a stranger mouse (S1) verse empty cage (E), and phase 3 for the evaluation of social novelty by introducing another stranger mouse (S2) verse S1. Both control and PDE5A OE mice showed some extent of sociability and social novelty (Fig. 7D and Table S13). However, PDE5A OE mice displayed impaired sociability compared to control mice (Fig. 7D). Next, we performed sucrose preference test to examine anhedonia, the inability to experience pleasure from rewarding activities. No significant difference in sucrose preference and total fluid drank were found between PDE5A OE and control mice (Fig. 7E and Table S13).

In addition to positive and negative symptoms, working memory deficit is a common symptom of psychiatric disorders such as SCZ and BD [40]. Here, we performed Y maze test to evaluate the working memory. PDE5A OE mice showed similar ratio of entering the correct arm and total number of entries as control mice (Fig. 7F and Table S13), suggesting little impact on the working memory. Sensorimotor gating deficits are often detected in individuals with schizophrenia and BD [41, 42]. PPI of the auditory startle reflex is an operational test of sensorimotor gating ability. PPI percentage increases with increasing prepulse intensity from 74 to 82 dB (Fig. 7G and Table S13), indicating a reliable test. However, PDE5A OE mice failed to show increased PPI at a stronger prepulse at 82 dB (Fig. 7G), suggesting impaired sensory gating function. At last, we conducted forced Swimming test to detect the depression or mania-like state. PDE5A OE mice exhibited reduced duration of immobility (Fig. 7H and Table S13), suggesting a decrease in the degree of despair, and possible mimic of mania-like symptoms of BD [43, 44].

Together, the above behavioral results indicate that PDE5A upregulation in striatal neurons in mice leads to partial BD-like phenotypes, namely reduced motor balance, impaired sociability, sensorimotor gating deficits, and mania-like state.

We then confirmed the overexpression efficiency of PDE5A by RT-qPCR (Fig. 8A, left plot, Table S14) and examined whether PDE5A is functionally overexpressed by determining cGMP concentration in GFP-expressed striatal tissue with cGMP ELISA Kit (Fig. 8A, right plot, Table S15). PDE5A is a cGMP-specific phosphodiesterase. The average cGMP concentration of PDE5A OE striatal tissue significantly decreased by 44% (control 1.79 vs. PDE5A OE 1.01 pmol/mL, n = 3 pairs, p < 0.05), indicating a successful and functional PDE5A OE model.

Fig. 8: PDE5A overexpression in mouse striatum reduced cGMP level and induced neuroinflammation.
figure 8

A Left: PDE5A overexpression efficiency by RT-qPCR. Right: cGMP concentration determined by cGMP ELISA kit. n = 3 pairs of GFP-expressed striatal tissue from control and PDE5A OE mice. B Volcano plot displaying DEGs from mRNA-seq analysis based on 3 pairs of GFP-expressed striatal tissues. DEGs were set as log2FC > =1 and <= −1, with p < 0.05. C Bubble plots illustrating gene ontology-biological process (GO-BP) enriched from upregulated and downregulated DEGs, respectively. D Validation of mRNA level of selected upregulated DEGs using RT-qPCR. n = 3 pairs. E Representative immunofluorescent staining images of GFP, GFAP, and CD3 in the striatum of control and PDE5A OE mice. Scale bar, 100 μm. Zoom in image scale bar, 20 μm. Cell number ratio of GFP+/DAPI, GFAP+/DAPI, GFAP+/GFP+, and CD3+/DAPI was compared (n = 3 mice each group, 2–4 sections per mouse). *p < 0.05, **p < 0.01, ***p < 0.001. F Quantification of the mRNA expression levels of neuroinflammatory-related factors in the striatal tissues of control and OE mice (Shapiro-Wilk test, n = 3 mice each group, **p < 0.01).

Full size image

To investigate the underlying molecular mechanisms, we performed mRNA-seq analysis on GFP-expressed striatal tissue from 3 pairs of control and PDE5A OE mice. 385 upregulated genes and 152 downregulated genes were significantly differentially-expressed (Fig. 8B and Table S16). Interestingly, many immune-related genes were upregulated, such as T cell receptor (TCR) components, chemokines, and complements (Fig. 8B). Functional enrichment analysis indicated that upregulated DEGs were mainly enriched in immune-related pathways, including immunity, innate immunity, adaptive immunity, complement pathway, and inflammatory response (Fig. 8C left panel). Downregulated DEGs were enriched in lipid metabolism, cell adhesion, neurogenesis, Wnt signaling, and differentiation (Fig. 8C right panel). Subsequently, we validated the expression of selected genes involved in those immune-related pathways by RT-qPCR (Fig. 8D and Table S14), and confirmed significant upregulation of TCR-related genes (Cd3e, Cd3d, Cd3g, Cd2, Cd8b1, Il7r, and Il2ra), chemokine Cxcl13, complement C3, MHC class II molecule Cd74, and MHC class I component B2m.

These transcriptomic upregulation of lymphocyte activation, co-stimulatory factors, chemokines, and complement system strongly indicated an enhanced pro-inflammatory state in PDE5A OE mice. The neuroinflammation was confirmed on striatal sections, with a significantly increased number of reactive astrocytes and CD3+ T cells in PDE5A OE mice (Fig. 8E and Table S17). Additionally, we examined the expression level of neuroinflammation-related factors, such as interleukin-1β (IL1β), IL6, IL10, tumor necrosis factor α (TNFα), and inducible nitric oxide synthase (iNOS), and found IL1β, TNFα, and iNOS were significantly increased in PDE5A OE mice (Fig. 8F and Table S14).

Discussion

This study conducted single-nucleus RNA sequencing on caudate, putamen, globus pallidus, and substantia nigra from control and BD postmortem brains. The integrative analyses of differential expression, gene co-expression, gene regulatory network revealed significant upregulation of PDE5A, a cGMP-specific phosphodiesterase, particularly in caudate MSNs of BD brains. To elucidate the role of PDE5A in BD etiology, we validated its effects in vitro and in vivo. Overexpression of PDE5A in primary cultured mouse striatal neurons led to reduced dendrite complexity, increased apoptosis and neuronal excitability. This defect could be rescued by PDE5 inhibition. Similarly, in vivo overexpression in the mouse striatum resulted in decreased cGMP level, upregulated neuroinflammatory genes, and BD-like behaviors. These findings strongly support a causal role of increased PDE5A expression in the basal ganglia of BD, suggesting a potential therapeutic avenue using existing PDE5 inhibitors like Tadalafil or Sildenafil for psychiatric diseases.

Traditionally, basal ganglia has been primarily associated with movement disorders like Parkinson’s disease. Recent studies unveiled its significance in psychiatric disorders as well [2, 45]. In BD patients, changes in morphology, neurochemical metabolites and functional connectivity were observed in the basal ganglia [46,47,48]. However, research on the cellular and molecular changes of BD basal ganglia is still limited. Previous studies have indicated that BD is primarily associated with MSNs based on an integrative analysis of single-cell transcriptome from specific mouse brain regions and 42 human genome-wide association studies [49]. Additionally, Gene modules within the striatum exhibited the highest genetic association signals in BD [50]. This aligns well with our finding of a large DEG number in caudate MSNs of BD brain.

The cellular composition in the caudate, putamen, substantia nigra, and globus pallidus varies significantly due to their distinct roles in the basal ganglia and motor control circuitry. The caudate and putamen are composed of 90–95% MSNs with a mix of GABAergic, cholinergic, and other interneurons [51]. The substantia nigra contains substantia nigra pars compacta (SNc) which is rich in dopaminergic neurons (DaNs), and substantia nigra pars reticulate (SNr) which contains mostly GABAergic neurons [52]. The globus pallidus is mainly populated by GABAergic neurons with few interneurons [53]. Besides, glial cell types also varies across those four regions. In our study, we observed that D1-, D2-MSNs are dominantly from caudate and putamen, consistent with prior knowledge. Unexpectedly, BD D1-MSNs from caudate and putamen are two non-overlapping subpopulations, same is true for D2-MSNs. Additionally, the cellular ratio of D1/D2-MSNs increased a bit in BD caudate. Moreover, the ratio of oligodendrocytes decreased in BD caudate but significantly increased in BD putamen, SN, and GP, which is in line with other postmortem studies showing ultrastructural dystrophy and degeneration of oligodendrocytes and dysmyelination in the caudate in SCZ and BD [54].

The cGMP is a ubiquitous second messenger and its signal transduction in cells is activated by ligand-mediated guanylate cyclase (GC) [55]. cGMP exerts its effects through the activation of cGMP-dependent protein kinase G (PKG) and related ion channels. PDE5A is a cGMP-specific phosphodiesterase that hydrolyzes cGMP to 5’-GMP and regulates cGMP signaling in various tissues, including the cardiovascular system, brain, lung, and gastrointestinal tract [56,57,58]. cGMP signaling regulates many physiological processes, including vascular tone, energy metabolism, transcription, cell growth, anti-inflammatory activity, and apoptosis [59,60,61]. Given the critical role of PDE5 in cGMP signal transduction, PDE5 has been targeted pharmacologically for the treatment of various pathological conditions including cardiovascular disease, pulmonary hypertension and erectile dysfunction [57, 62, 63]. Moreover, extensive studies have confirmed the potential therapeutic effects of PDE5 inhibition in various neurological disorders including AD [64], ischemic injury and epilepsy [65, 66]. However, the role of PDE5 in psychiatric disorders is still elusive. Although other PDE family genes have been indicated in psychiatric disorders, such as PDE11A and BD [67], PDE1 and schizophrenia [68]. Our study expands the scope and establishes a causative role of PDE5A in psychiatric diseases especially BD.

We observed that overexpression of PDE5A in mouse LGE-derived striatal neurons led to reduced dendrite complexity, increased apoptosis and neuronal excitability. Additionally, PDE5A overexpression resulted in a decrease in cGMP level. This is consistent with previous studies showing that cGMP signaling could promote spine formation [69], regulate synaptic transmission, and neuroprotection [36]. Under physiological conditions, immune cells can hardly cross the blood-brain barrier. Our finding of increased gene expression of CD3 T cells, chemokines, and complements upon PDE5A overexpression in mouse striatum is reminiscent of neuroinflammation [70]. Intracellular accumulation of cGMP can reduce the production of proinflammatory cytokines [71]. This aligns well with our finding that PDE5A overexpression led to reduced cGMP level and enhanced neuroinflammation.

In addition, upregulation of neuroinflammation-related factors, IL-1β, TNFα and iNOS, is associated with defects in mouse behaviors. Elevation of IL-1β and TNFα, key pro-inflammatory cytokines, is linked with reduced activity, social withdrawal, and diminished exploration [72]. Persistent expression of IL-1β, TNFα, and iNOS could lead to impaired learning and memory [73]. Increased expression of IL-1β and TNFα is also linked to anxiety and depression-like behaviors in mice [74]. At last, overexpression of iNOS leads to increased nitric oxide production, which can result in oxidative stress and neuronal damage. This has also been implicated in anxiety and depression-like behaviors in mice [75]. This is very consistent with the behavioral deficits observed in PDE5A OE mice, namely reduced motor balance, impaired sociability, and sensorimotor gating deficits.

We next searched public datasets of GWAS studies, bulk RNA seq, scRNA seq, and histochemical staining studies towards psychiatric diseases especially BD and schizophrenia. Interestingly, we found PDE family genes are frequently upregulated or downregulated or identified as risk genes from different BD brain regions by different experimental methods (Table 1). The balance of cAMP and cGMP in the nervous system is crucial for learning, memory, neuronal circuits, neurogenesis, neuronal migration, and synaptic plasticity [76,77,78]. In our study, we observed that PDE5A upregulation is particularly evident in caudate MSNs of one BD postmortem brain, which was further validated on 7 BD brains by immunostaining. We then performed a series of in vitro and in vivo validations to prove the causative role of PDE5A upregulation in BD etiology. Interestingly, another group reported PDE5A upregulation to ~5 fold in AD temporal cortex by RT-qPCR (Table 1). Although PDE5A expression may not be elevated in other published BD cohorts (most of them are not of Asian origin), the frequent dysregulation of other PDE genes found in BD or schizophrenia (Table 1) strongly suggests the importance of cAMP/cGMP signaling in psychiatric diseases. Our study unveiled a causative role of PDE5A upregulation-induced cGMP reduction, leading to disrupted dendrite development, increased apoptosis and neuronal excitability, neuroinflammation, and eventually the manifestation of BD-like behaviors. This could serve as a shared mechanisms of the dysregulation of cGMP-related PDE genes in the development of BD or schizophrenia.

Table 1 Dysregulation of PDE expression and psychiatric diseases.
Full size table

We acknowledge a number of limitations in our study. First, we conducted single-nucleus RNA sequencing on caudate, putamen, GP, and SN from only one control and one BD brain, due to limited availability of BD donors. Hence, more cases should be involved when available in future. Second, the decreased cGMP level by upregulated PDE5A was found mainly in BD caudate MSNs, whether this also happens in other brain regions of BD donors is an interesting question for further investigations. Third, the prevalence of the dysregulation of cGMP signaling mediated by PDE5 in BD patients, especially in Asian cohort, is worthy of further survey, considering that PDE5 inhibitors are existing medications and could be easier to translate for new applications.

In conclusion, we established a cell type-specific gene expression profile of BD basal ganglia and particularly unveiled a causative role of PDE5A upregulation in BD etiology via in vitro cultured striatal neurons and in vivo mouse model.

Methods

Human postmortem tissue

The brain tissue samples were obtained from Fudan branch of National Health and Disease Human Brain Tissue Resource Center, the Body Donation Station in Fudan University (Shanghai Red Cross Society). Human postmortem brain tissue from two control and two BD donors of Asian ancestry (Table S1) was obtained under ethic approval 2018C001. Human caudate paraffin sections of five control and five BD donors of Caucasian ancestry were kindly provided by Douglas-Bell Canada Brain Bank (Table S1).

Single nucleus isolation and library preparation

We performed single nucleus RNA sequencing on 8 samples from one control and one BD donors (see Table S1) using 10x single nucleus RNA seq. Single nuclei were isolated from frozen tissues using a modified nuclei isolation protocol [48, 49]. In short, frozen brain tissue (~0.5–1 cm3) was homogenized in 700 μL of chilled homogenization buffer (250 mM sucrose, 25 mM KCl, 5 mM MgCl2, 10 mM Tris buffer at pH 8.0, 1 μM DTT, 0.4 U/μL RNase inhibitor, 0.2 U/μL Superase inhibitor, 0.1% v/v Triton X-100) using a glass dounce. Homogenate was filtered using a 100 μm cell strainer into 2 mL of resuspension buffer (1x PBS with 1% BSA and 0.05 U/μL RNase inhibitor and 0.05 U/μL Superase inhibitor), the solution was then filtered through a 40 μm cell strainer and spinned at 260 x g for 5 min at 4 °C. Nuclei were washed with 3 mL of wash buffer (1x PBS with 1% BSA) and passed through 5 mL of 20–30% Percoll at 700 × g for 10 min at 4 °C to remove cell and myelin debris. Finally, after wash once with wash buffer to remove Percoll residue, nuclei were resuspended into 1 mL of resuspension buffer, stained with trypan blue and numerated using a hemocytometer. The subsequent library preparation was performed according to Single Cell 3’ LT (10x Genomics).

Single-nucleus RNA sequencing data processing

Sequencing reads generated from single nucleus RNA-seq libraries of eight samples were demultiplexed and aligned to the human GRCh38 reference genome using the CellRanger software (version 6.1.1) with default settings, and the raw gene-barcode count matrix per sample was generated using the ‘count’ function with default settings.

Quality control and doublet removal

For each sample, the raw gene-barcode matrix outputted by CellRanger was converted into a Seurat [31] object in R (version 4.1.1) using the Read10X function. Seurat objects corresponding to samples from the same donor were merged into a single Seurat object. Next, for each donor, we used the following four quality control criteria for filtering cells and genes: (1) 1000 = < number of expressed genes per cell <=10,000; (2) number of UMIs per cell <=50,000; (3) fraction of mitochondrial reads per cell <=2%; (4) number of expressed cells per gene >= 3. Afterwards, doublet-like cells were distinguished using the doubletFinder_v3 function from the doubletFinder R package (version 2.0.3) with the parameters ‘PCs = 1:20, pN = 0.25, pK = 0.005, nExp = 9000’. According to the estimated doublet scores, cells with scores in the top 15% were removed.

Clustering using PCA-Louvain based approach and cell type assignment

For each donor, we clustered cells using the standard PCA-Louvain clustering approach, as implemented in the Seurat R package. Specifically, we first normalized the gene-barcode UMI count matrix using the NormalizeData function with the parameter ‘scale.factor = 10,000’, and the ScaleData function is used to scale and center normalized data. Next, top 2000 genes with the highest variance was selected by the FindVariableFeatures function, and these genes were used for PCA dimensionality reduction using the RunPCA function. The FindNeighbors function with the top 20 principal components was used to construct the cell-to-cell nest-neighbor graph. For the control donor, the resolution parameter adopted by the FindClusters function was 0.2, and for the BD donor, the resolution parameter was 0.23.

To preliminarily assign main cell types to clusters, we calculated the top 100 genes with the highest PEM scores for each cluster as the specifically expressed genes of that cluster. We further annotated cell types to these genes according to a known marker gene list (Table S3). Finally, we counted the number of marker genes per cell type for each cluster, and used the Pheatmap R package (version 1.0.12) to draw heatmaps to determine the main cell type for each cluster. To define neuron subtypes, we merged neuron clusters and reran Seurat functions including NormalizeData, ScaleData, FindVariableFeatures, RunPCA, FindNeighbors, and FindClusters, and finally we accurately assigned neuron subtypes according to D1-MSN, D2-MSN, interneuron, and dopaminergic neuron markers.

Differential expression analysis

For each of the four brain regions, differential expression analysis was performed on the same cell type between BD and control groups. Briefly, we used the FindMarkers function from the Seurat R package for differential expression analysis. Differentially expressed genes were screened out using the following four criteria: (1) At least 25% of the cells in the higher expressed group expressed the gene; (2) At least 10% expression percentage difference of the gene between the two groups; (3) The log2-transformed fold-change is larger than 0.5; (4) The adjusted P-value is less than 0.01.

Gene ontology enrichment analysis

Gene ontology biological process (GO-BP) enrichment analysis was conducted via the website http://geneontology.org. P-values were adjusted by FDR (false discovery rate). Only the GO terms with adjusted P-values less than 0.05 were kept.

WGCNA analysis

Weighted gene co-expression network analysis was conducted by WGCNA [79] R package (version 1.71). We took the normalized expressions of 1728 genes identified from the aforementioned differential expression analysis and all cells as the input. Specifically, we used the pickSoftThreshold function to determine the best power (=3). Next, the adjacency function with the parameter ‘type=signed’ and the TOMsimilarity function with the parameter ‘TOMType=signed’ were used to calculate the adjacency matrix and the topological overlap matrix, respectively. Then the cutreeDynamic function with the parameters ‘deepSplit=2, minClusterSize=30’ was used to define co-expression gene modules. Finally, the mergeCloseModules function with the parameter ‘cutHeight = 0.25’ was used to merge similar gene modules.

For the correlation between each gene module and cell type, we calculated the Pearson correlation coefficient between the module eigengene expression vector of all cells, and the binary vector indicating whether each cell belongs to that cell type. Similar methods were used to calculate the correlation between modules and brain regions, as well as the correlation between modules and the BD/control identity.

To determine hub genes of the module green that includes 114 genes, we first calculated the Pearson correlation coefficient between the eigengene expression vector of module green, and the expression vector of each gene, which is defined as the MM (module membership) of the gene. The Pearson correlation coefficient between the expression vector of each gene, and the binary vector indicating whether each cell belongs to BD, was defined as the GS (gene significance) of the gene. Then, the top 60 genes with the largest MM among 114 genes were screened out first, and the top 30 genes with the largest GS were finally regarded as hub genes.

For the plotting of the gene-gene network for hub genes, we retained the edges whose values in the adjacency matrix were in the top 10% among all edges, and drew the network using the Cytoscape software (version 3.7.2). The circle size representing each hub gene is proportional to MM, and the thickness of each edge is proportional to its value in the adjacency matrix.

SCENIC analysis

For each of the four brain regions, we used the pySCENIC [80] Python package (version 0.11.2) to perform single-cell transcription factor (TF) regulatory network analysis. We used the normalized gene expression matrix with all cells of that region and genes filtered out by the geneFiltering function from the SCENIC R package (version 1.3.1) as the input. In particular, for caudate, putamen, SN, and GP, there are 11,840, 11,993, 11,334, and 11,453 genes kept for analysis, respectively.

We first used pySCENIC to infer the regulatory relationship between TFs and their target genes, and we only kept TFs with the number of targets >=20. Then, for each TF module, we used pySCENIC to calculate AUC (the active score) for each cell. For each cell type, if the number of cells >= 20 in both BD and control groups, we calculated the mean AUC for the two groups, respectively. After obtaining the TF-cell type AUC matrices for both BD and control groups, we subtracted the BD matrix from the control matrix to get the differential matrix, and only retained the TFs whose the maximum value of the absolute values of its cell type scores in the differential matrix is greater than 0.01.

Animals

C57BL/6 mice were used in this study and were purchased from Shanghai Slac Laboratory Animal CO., LTD. The day when vaginal plug is detected was considered to be embryonic day 0.5 (E0.5). All experimental animals were kept in an animal facility at Fudan University and all experiments were conducted in accordance with guidelines approved by Fudan University.

Primary MSN culture and adeno-associated virus (AAV) infection

Lateral ganglionic eminence (LGE) is presumed to be the source of the mouse striatal neurons [81]. LGE was isolated from the E16-17 embryonic mouse brains in ice-cold HBSS (Biosharp, BL559A) containing 1% HEPES (Biosharp, BL1061A). The isolated LGE was transferred into a 15 mL centrifuge tube with 3 mL of 0.125% trypsin (Sigma-Aldrich, T4174) and 100 μg/mL DNase I (Sigma, D4527), digested in a 37 °C water bath for 15 min, with gentle shaking every 3 min. The digestion was terminated by adding 3 mL Neurobasal medium containing 10% FBS (Gibco, 10091148), 2% B27 (Gibco, 17504044), and 1% Glutamax (Gibco, 35050061). The mixture was centrifuged at 800 rpm for 5 min, and the supernatant was discarded. The pellet was resuspended in the same medium, with gentle pipetting for 20 times. After centrifugation at 800 rpm for 5 min and discarding the supernatant, the cells were resuspended and plated at a concentration of 50,000 cells/mL into poly-L-lysine-coated 24-well plates, cultured in a Neurobasal medium supplemented with 2% B27 supplement, and 1% Glutamax. rAAV2/9-hSyn-PDE5A-EGFP and control virus were packaged by BrainVTA (Wuhan, China). Full length of mouse PDE5A (GeneID: NM_153422.3, 2598 bp) was overexpressed. Neurons were infected with rAAV2/9-hSyn-PDE5A-EGFP (OE group) or control virus (NC group) at DIV3 and medium was replaced after 12 h. OE neurons were treated with 10 μM tadalafil (MedChemExpress, HY-90009A) or DMSO at DIV7 and DIV10 for 12 h, respectively. The two groups were named as OE+tadalafil and OE + DMSO, respectively. Sholl analysis and immunostaining were performed at DIV14.

Immunofluorescence staining and analysis of dendritic morphology

For brain section staining, sections were rinsed with PBS for 10 min, permeabilized for 20 min with PBS containing 0.2% Triton X-100 and blocked with 5% BSA at room temperature. The sections were incubated overnight with primary antibodies including rabbit anti-GFAP (1:500, Proteintech, 16825-1-AP), mouse anti-CD3 (1:50, Santa Cruz, sc-20047), mouse anti-PDE5A (1:500, Santa Cruz, sc-398747), and rabbit anti-DARPP32 (1:200, Proteintech, 10748-1-AP) at 4 °C. The sections were rinsed with PBS for 3 × 10 min and incubated with corresponding secondary antibodies including CoraLite594-conjugated Donkey Anti-Rabbit IgG(H + L) (1:500, Proteintech, SA00013-8), and CoraLite647-conjugated AffiniPure F(ab’)2 Fragment Goat Anti-Mouse IgG (H + L) (1:500, Proteintech, SA00014-10) for 1 h at room temperature in dark. Nuclei were counterstained with 4′, 6-diamidino-2-phenylindole (DAPI) and the sections were mounted in DAPI Fluoromount-G (SouthernBiotech, 0100-20). In addition, prior to mounting, brain sections were treated with TrueBlack Lipofuscin Autofluorescence Quencher (Biotium, 23007) for 30 s to eliminate spontaneous fluorescence. All images were taken with a confocal microscope (TCS SP8, Leica) and EVOS M7000 (AMF7000, Invitrogen). Analysis was conducted with Image J software (v. 2.0.0-rc-69).

For cell culture staining, neurons on coverslips were fixed with 4% of paraformaldehyde for 20 min at room temperature, rinsed with PBS for 3 × 2 min, permeabilized for 10 min with PBS containing 0.2% Triton X-100 and blocked with 3% BSA at room temperature. The sections were incubated overnight with primary antibodies including rabbit anti-GFP (1:500, Proteintech, 50430-2-AP), mouse anti-MAP2 (1:500, Proteintech, 67015-1-Ig), and mouse anti-caspase3 (1:500, Proteintech, 66470-2-Ig) at 4 °C. The sections were rinsed with PBS for 3 × 10 min and incubated with corresponding secondary antibodies including CoraLite647-conjugated AffiniPure F(ab’)2 Fragment Goat Anti-Mouse IgG (H + L) (1:500, Proteintech, SA00014-10) for 1 h at room temperature in dark. Cell nuclei staining is same as above. All images were taken with a confocal microscope (TCS SP8, Leica), and analyzed with ImageJ Fiji. Sholl analysis and Simple Neurite Tracer are applied to calculate the branching and total length of the tree as described [82].

Western Blot

Cells were lysed in RIPA buffer and incubated for 20 min on ice. Samples were cleared by centrifugation at 12,000 rpm for 15 min, sonicated and protein concentration was measured by BCA kit reagent assay (Beyotime, P0012). Lysates were mixed with 5X SDS-PAGE sample loading buffer and boiled at 95 °C for 5 min. Cell lysates were loaded onto 10% polyacrylamide gels. The proteins were then transferred to 0.45 μm PVDF membranes. Membranes were blocked with 3% BSA for 1 h at room temperature and incubated with primary antibodies including mouse anti-PDE5A (1:1000, Santa Cruz, sc-398747) and rabbit anti-GAPDH (1:10,000, Proteintech, 10494-1-AP) in 3% BSA overnight at 4 °C. Subsequently, the membranes were washed 3 times in TBST and incubated with HRP-conjugated Goat Anti-Mouse IgG(H + L) (1:5000, Proteintech, SA00001-1) and HRP-conjugated Goat Anti-Rabbit IgG(H + L) (1:5000, Proteintech, SA00001-2) for 1 h at room temperature, followed by 3 washes. The antigen–antibody complexes were detected by ECL. Quantification of Western blots was performed using Fiji/ImageJ software.

Patch clamp recording

Neuron culture slides were placed in the recording chamber and superfused with ACSF solution containing the following (in mM): 126 NaCl, 1.6 KCl, 1.1 NaH2PO4, 1.4 MgCl2, 2.4 CaCl2, 26 NaHCO3, and 11 glucose. Whole-cell patch-clamp recordings were made using a MultiClamp 700B amplifier (Molecular Devices). To record neuron excitability, glass electrodes (3−5 MΩ) contained the following (in mM): 125 K-Metanesulfonate, 5 KCl, 10 HEPES, 0.2 EGTA, 2 NaCl, 2.5 Mg-ATP, and 0.25 Na-GTP (pH 7.2−7.3, 270-280 Osmo). Neurons were held at −60 mV, and depolarizing current steps (400 ms, from 0 to 100 pA in 10 pA steps) were applied. The slope of the number of spikes per depolarizing step was used to measure excitability.

Stereotaxic injection

Stereotaxic injection of rAAV2/9-hSyn-PDE5A-EGFP or control virus into mouse bilateral striatum was performed in 7-week-old male C57BL/6 mice (n = 15 pairs). Briefly, mice were anesthetized with tribromoethanol (20 mL.kg-1, i.p.) and placed in a stereotaxic head frame (Model 68001, RWD Life Science). 4 small holes were drilled in the skull for viral injections using a pulled fine-glass capillary (RWD Life Science, GC-3.5) connected to a microliter syringe. Bilateral injections were made for the striatal regions: anteroposterior (AP), +0.74 mm; mediolateral (ML), ±2.0 mm; and dorsoventral (DV), −3.5 mm, and AP, +1.2 mm; ML, ±1.8 mm; DV, -3.0 mm from bregma. 0.125 μL of the virus was injected at each site and a total volume of 0.5 μL virus was injected in each mouse. Behavioral and molecular tests were performed 4 weeks after the surgery.

Behavioral tests

A series of behavioral tests were performed including catwalk gaiting, rotarod, open field, three-chamber, sucrose preference, Y maze, prepulse inhibition, and forced swimming.

In catwalk gaiting test [83], the test mouse was trained 3 times on the CatWalk XT 8.1 (Noldus) to smoothly walk through the corridor 1 day before training. Footprints were automatically detected by the Catwalk XT 10.0 software. Detection settings were as follows, camera gain 20; intensity threshold 0.10; max. speed variation 60%. During the trial, the mouse was allowed to freely pass through the set detection channel, and the running time is 0.5−12 s. The footprints were video-captured and processed to evaluate the movement of the mouse. Each mouse was tested for at least three times.

In rotarod test, motor coordination was evaluated with a rotarod model (Ugo Basile, 47650). The test mouse was acclimated to the testing room and trained at 5 rpm one day before the test. The mouse was placed on a rotating rod at 5 rpm that accelerates up to 40 rpm over the course of 300 s. The time when the mouse falls from the rod was recorded. Average latency to fall of three trials was calculated.

In open field test, the test mouse was adapted for a 30-minute period in a standard open field arena (40 cm × 40 cm) before its locomotor activities were recorded for 60 min. The central area, measuring 20 cm × 20 cm, was also monitored. Ethovision XT videotracking software from Noldus Information Technologies (Leesburg, VA, USA) was utilized to record and analyze parameters, including total distance traveled, time spent in the center, entries y into the center, and the duration of the initial entry into the center.

In three-chamber test [84], the apparatus is an acrylic glass box (102 cm × 47 cm × 45 cm) with a partition dividing the box into three compartments, with a small door that allows free movement of the mice. The test has three phases: (1) habituation (phase 1): pencil cages were placed in the left and right chambers, and the test mouse was placed in the middle chamber to freely explore for 10 min. (2) sociability (phase 2): a weight and gender-matched unfamiliar mouse was placed in a metal cage in either left or right chamber. The test mouse was introduced into the middle compartment and allowed to freely move for 10 min. The time spent in close interaction with an empty cage or the cage with the S1 mouse was recorded. (3) social novelty (phase 3): another weight and gender-matched unfamiliar mouse was introduced into a pencil cage on the other side of the box. The test mouse was placed in the middle chamber and allowed to explore for 10 min. The time spent in close interaction with the S1 and S2 mice was recorded. The behavior of the mice was tracked and recorded using EthoVision XT software. Sociability is calculated as the ratio of interacting time of (S1 - E) to (S1 + E), and social novelty is calculated as the ratio of interaction of (S2 - S1) to (S2 + S1).

In sucrose preference test [85], the test mouse was first accustomed to two identical bottles, one filled with water and the other filled with 1% sucrose, for a duration of 36 h. The test mouse was deprived of food and water for 12 h before presented with a bottle of water and a bottle of 1% sucrose solution. The weight of the two bottles was measured, and the positions of the bottles were exchanged every 12 h, for a total of 3 times. Finally, the weight of the two bottles was measured. The consumption of sucrose solution and water was calculated as sucrose preference (%) = [sugar water consumption / (sugar water consumption + water consumption)] * 100.

In Y-maze test [86], the maze consists of three identical arms at 120° angles to each other, which are named as arm A, B, and C. The test mouse was placed in the center of the Y maze and allowed to move undisturbed for 8 min. The test mouse was monitored when entering into three different arms in a sequence called spontaneous alternating behavior. If the mouse could sequentially pass through the three arms, the trial was considered correct. The spontaneous alternation % = (number of consecutive entries into three arms)/(total number of entries - 2) × 100.

In prepulse inhibition test [87], the test mouse was restricted in a cylindrical device and acclimated to a 70 dB background white noise for 5 min, followed by five 20-ms startling pulses of 120 dB using SR-LAB startle chambers. Then, six sets of tests were performed, each consisting of different stimulus conditions in a pseudorandom distribution: no stimulus, startling tone only, pulse tone only (at 74, 78, or 82 dB), pairing of pulse and startling tone, and a block of five 20 ms startling pulses at 120 dB. Maximum amplitude of fear response for each type of experiment was recorded and the data were analyzed using the average of the six trials. PPI % = [1 - (average response to prepulse before startle stimulus/average response to startling stimulus)] × 100.

In forced swimming test, the test mouse was placed in a transparent cylindrical bucket with a diameter of 15 cm and a height of 30 cm, filled with a certain height of water (the mouse’s hindlegs could not reach the lower end of the bucket), and water temperature was 24 ± 1 °C. The mouse was allowed to swim for 6 min and recorded by camera. Immobility time within the last 4 min was measured.

mRNA-seq and data analysis

Total RNA was extracted from the GFP-expressed striatal tissues from 3 pairs of control and PDE5A OE mice using Trizol reagent (Invitrogen, 15596-026). Libraries were constructed with a VAHTS Universal V6 RNA-seq Library Prep Kit (Vazyme, NR604) and sequenced on a Novaseq 6000 instrument (Illumina). The obtained FASTQ files were aligned to the GRCm38 mouse reference genome with STAR (version 2.5.3a). The raw sequence counts of known genes were quantified using the StringTie software (version 2.2.1). Differential expression analysis was performed using DESeq2 (v1.16.1). Genes were considered to be significantly differentially expressed using a p value of 0.05 and a fold change of 2. GO enrichment analysis was performed using DAVID [88].

Quantitative real-time PCR

For validation, total RNA was extracted from GFP-expressed striatal tissues using RNAprep Pure Tissue kit (Tiangen Biotech., DP431), and reverse-transcribed into cDNA with Evo M-MLV RT kit (Accurate Biology, AG11705). Quantitative real-time PCR was conducted using TB Green Premix Ex Taq (Takara, rr420a) on a StepOnePlus system (Thermofisher, 4376600). The gene expression was normalized using GAPDH as an internal control and calculate the relative expression using the 2-ΔΔCt method [89]. Primer information was summarized in Table S18.

Determination of cGMP concentration

About 0.03 g GFP-expressed striatal tissue was dissected from each mouse, and the tissue was minced. 120 μL of pre-chilled PBS was added for grinding. After centrifugation at 5000 × g for 5–10 min, the supernatant was collected. The cGMP content in the supernatant was determined using the cGMP ELISA Kit (Elabscience, E-EL-0083). Briefly, standard, blank, and sample wells were set up. The standard wells were loaded with 50 μL of serially diluted standard solution, the blank wells were loaded with 50 μL of standard solution & sample dilution solution, and the remaining wells were loaded with 50 μL of the test sample. Immediately, 50 μL of biotinylated antibody working solution was added to each well. The plate was covered and incubated at 37 °C for 45 min. After washing three times, 100 μL of HRP enzyme conjugate working solution was added to each well, and the plate was covered and incubated at 37 °C for 30 min. After washing five times, 90 μL of substrate solution (TMB) was added to each well, and the plate was covered and incubated at 37 °C in the dark for 15 min. The reaction was stopped by adding 50 μL of stop solution, and the OD values of each well were immediately measured at a wavelength of 450 nm using an ELISA reader. A standard curve was fitted on a double logarithmic coordinate axis using a four-parameter logistic function: y = A2 + (A1-A2)/(1 + (x/x0)^p) (A1 = 1.72859, A2 = 0.07314, x0 = 6.19941, p = 1.35357, r2 = 0.98323, where y represents absorbance and x represents concentration).

Statistics

Data are presented as mean ± SEM. For two-group comparisons, if they fit the normal distribution, unpaired Student’s t-tests with Shapiro-Wilk method or two-way ANOVA tests with Sidak’s multiple comparisons method was performed. In case of non-normal data distribution, Mann–Whitney U test was used. All statistical analyses were performed using GraphPad Prism 8 software (version 8.4.3). The value of p < 0.05 was considered statistically significant.