Distinct microglia alternative splicing in Alzheimer's disease

Numerous alternative splicing (AS) events have been documented in Alzheimer's disease (AD). However, cell type-specific AS analysis is still lacking. We described AS events in the hippocampal microglia sorted by CD45 and CD11b from Aβ precursor protein (APP) and non-transgenic (Ntg) mice. GSE171195 dataset was downloaded from GEO database, aligned to GRCm39 genome. Skipped exon (SE), alternative 3’SS (A3SS), retained intron (RI), alternative 5’SS (A5SS), and mutually exclusive exons (MXE) were evaluated using rMATS and maser. Differential expressed genes or transcripts were analyzed via limma. Gene ontology and correlation analyses were performed with clusterProfiler and ggcorrplot R packages. 36,340 raw counts of AS were identified, and 95 significant AS events were eventually selected with strict criteria: (1) average coverage >5; (2) delta percent spliced in >0.1. SE was the most common AS events (68.42%), followed by A3SS and RI. Autophagy genes were mainly spliced in SE events, actin depolymerization genes spliced in A3SS events, while synaptic plasticity related genes were mainly spliced in RI pattern. These significant AS events may be regulated by dysregulated splicing factors in AD. In conclusion, we revealed microglia specific AS events in AD, and our study provides novel pathological mechanisms in the pathogenesis of AD.


INTRODUCTION
Splicing converts precursor mRNA into mature mRNA via spliceosome, a multiprotein-RNA complex with highly specific and stepwise interactions. The spliceosome contains 5 small nuclear ribonucleoproteins (snRNPs; U1 to U6 snRNPs). U1 snRNP binds to 5′ splice site, and U2 snRNP binds to 3′ splice site and polypyrimidine sequences. Next, U5, U4/U6 were recruited to complete intron excision and exon linkage mainly following the "GU-AG" rule [1]. Alternative splicing (AS) enables cells to diversify their proteome by generating distinct characteristics to different protein isoforms through different combination of exons in the mRNA, leading to complex biological functions under conditions of external and internal stimuli. AS plays essential roles in cell differentiation and organ development t [2]. Generally, five basic modes of AS are classified: alternative 5′ splice site (A5SS), skipped exon (SE), mutually exclusive exons (MXE), retained intron (RI), and alternative 3′ splice site (A3SS) [3]. AS dynamically changes during aging, and aberrant AS events have been observed in numerous diseases including cancer through providing malignant protein isoforms [4]. Carcinogenesis involves cellular proliferation, invasion and metastasis, induction of angiogenesis, and immune escape. Recurrent somatic mutations of the splicing machinery components have been observed in human solid tumors and hematological malignancies. AS participates in these processes by regulating the alternative expression of numerous oncogenic or tumor suppressor genes [5].
Aberrant AS events have been characterized in Alzheimer's disease (AD), the most frequent cause of dementia. AD is hallmarked by β-amyloid (Aβ) plaques and hyperphosphorylated tau tangles. Both genetic and environmental risk factors contribute to AD pathogenesis [6]. The triggering receptor expressed in myeloid cells 2 (TREM2) mediates inflammatory responses, microglia activation, and phagocytosis of apoptotic neurons. TREM2 is involved in tau accumulation and Aβ plaque formation and therefore strongly affects risk of AD [7]. TREM2 rare variants caused exon 2 skipping, and activated immune-related functional pathways in AD [8]. Nuclear speckle specific hnRNP D-like prevented AD-related cognitive decline via modulating AS of synaptic gene calmodulin kinase-like vesicle-associated (CAMKV) [9]. Numerous AS events mainly including SE and RI have been documented in both AD human and animal models [10,11].
These studies mainly focused on different models or brain regions. We systematically described AS events in the hippocampal microglia (CD45int CD11b+) sorted with FACSAria II (BD) of Aβ precursor protein (APP) and non-transgenic (Ntg) mice. We revealed hippocampal microglia cell type-specific AS events in AD.

Global summary of alternative splicing events
A total of 36,340 AS events were initially identified in GSE171195 dataset ( Figure 1A). Eventually, 95 significant AS events were collected ( Figure 1B, Table  1). The principal component analysis (PCA) revealed that these AS events clearly classified APP from Ntg group ( Figure 1C). Increased PSI was found in A3SS and SE in APP group, while decreased PSI was found in A3SS and global events ( Figure 1D, 1E). Top two events in each AS event were included in Table 2, and all 95 significant AS events were shown in Supplementary Table 1.

Analysis of SE events
SE events ranked the most frequent AS events, accounting for 68.42% ( Figure 1B). GO analysis of   genes involved in significant SE events were enriched in regulation of autophagy (Figure 2A, 2B), including eukaryotic translation initiation factor 2 alpha kinase 4 (Eif2ak4) and two pore segment channel 2 (Tpcn2). Eif2ak4 tended to harbor longer exons, while Tpcn2 tended to harbor shorter exons in APP ( Figure 2C, 2D).

Analysis of A3SS events
A total of 11 significant A3SS events were identified and ranked top 2 AS events ( Figure 1B). GO enrichment revealed that genes with significant A3SS events were enriched in actin polymerization and depolymerization, including microtubule associated monooxygenase, calponin and LIM domain containing 1 (Mical1), and adducin 1 (Add1) ( Figure 3A, 3B). The sashimi plot showed decreased PSI and a shorter transcript for Mical1 ( Figure 3C), whereas increased PSI for Add1, indicating a longer transcript ( Figure 3D).

Analysis of RI events
RI is third most common events in this study ( Figure  1B). The PSI value referred to level of retained intron.
GO analysis revealed enrichment of RNA splicing as well as synaptic plasticity including axonogenesis, neurotransmitter secretion, neuron projection and axon guidance ( Figure 4A, 4B). Glypican 2 (Gpc2) showed increased RI level and enriched in neuron projection ( Figure 4C). RNA binding motif protein 3 (Rbm3) showed decreased RI level and enriched in RNA splicing ( Figure 4D).

Differentially expressed genes and transcript
Aberrantly spliced mRNAs would be degraded by nonsense-mediated mRNA decay (NMD) pathway. However, some mRNAs may bypass NMD and be translated. Differential expressed genes and transcripts were examined. These samples were well correlated within group ( Figure 5A). 583 up-and 310 downregulated genes were identified ( Figure 5B, Supplementary Table 2). Only two genes with significant AS events showed significant difference between APP and Ntg groups on gene level ( Figure 5C). On transcript level, 1225 up-and 756 down-regulated transcripts were identified (Supplementary Table 3). 17 out of the 26 transcripts, whose host genes with significant AS events were involved in SE event, 6 transcripts were involved in RI event (Table 3). Several genes with significant AS events harbored transcripts with opposite changes, including Rbm3 and Azi2 ( Figure 5D, Table 3). Given Azi2 for instance, SE event resulted in a longer transcript (ENSMUST00000130735) and a shorter transcript (ENSMUST00000133814) ( Figure 5E). In SE event, Azi2 showed increased PSI for the longer transcript in APP group ( Figure 5F). Consistently, the longer ENSMUST00000130735 with increased PSI in APP group was elevated, while the shorter ENSMUST00000133814 with decreased PSI in APP group showed suppressed expression ( Figure 5G). However, the gene level expression of Azi2 showed no significant difference ( Figure 5H).

Analysis of upstream splicing factors
AS events could be regulated by a number of splicing factors (SFs). The expression of 280 SFs obtained from previous study was examined (Supplementary Table 4) [12]. A total of 37 transcripts involved 30 SF genes were significantly dysregulated between APP and Ntg groups ( Figure 6A, Table 4). The PPI networks were constructed using String and Cytoscape ( Figure 6B). The first and second clusters of these SFs were identified using Cytoscape with MCODE app ( Figure  6C, 6D). The correlation of the two clusters and significantly dysregulated transcripts whose host genes were involved in significant AS events were performed ( Figure 6E, 6F).

DISCUSSION
Aβ plaque is a hallmark of AD. Aβ is produced by sequential cleavage of APP, and APP transgenic mice have been widely used ad AD model [13,14]. In this study we explored the AS events in hippocampal microglia sorted by CD45 and CD11b in APP and Ntg mice. All five types of AS events were detected, and genes involved different AS events showed AS type specific GO enrichment. Aberrant AS have been widely recorded in AD patients and animal models [10,11]. We for the first time, investigated cell type specific AS events in AD.
Consistent with previous studies [15,16], SE was the most abundant AS events (68.42%), leading to diverse protein isoforms encoded by a single gene. A3SS (11.58%) and RI (10.53%) showed similar AS event counts. GO analysis of genes involved in significant SE events were enriched in regulation of autophagy ( Figure  2A, 2B), including Eif2ak4 and Tpcn2. Eif2ak4   Atf4 is a key autophagy related gene. Atf4 binds to promoters of numerous autophagy genes including Becn1, Atg7, and Sqstm1, and this subsequently increased transcription of these genes. Therefore, activation of Eif2ak4-Eif2a-Atf4 pathway triggers autophagy [17]. Here, we found that Eif2ak4 tended to generate a longer transcript in AD, the function of this longer isoform may deserve further functional studies. Tpcn2 is a lysosomal non-selective Na + /Ca 2+ channel, Tpcn2 suppresses the fusion between autophagosome and lysosome, leading to the accumulation of autophagosomes in cancer cells [18]. In this study, we found that Tpcn2 tended to produce a shorter transcript in AD.
A3SS occurred due to alternative acceptor sites, and the PSI indicates inclusion of the longer exon in A3SS events. GO enrichment revealed that genes with significant A3SS events were enriched in actin polymerization and depolymerization, including Mical1 and Add1. Tau accumulation is clearly linked to AD pathogenesis. Mical1 was identified as tau interacting protein via proteomic approach. Mical1 changed the interaction properties of tau and potentiated tau toxicity, mediated by Mical1's redox activity on tau Cys322 [19]. Moreover, the protein level of Mical1 was increased in AD patients, and considered to be a potential biomarker of tauopathies [19]. Add1 plays essential role in dendritic morphology. Zheng et al. reported that loss of miR-135a-5p expression increased Rock2 activity and phosphorylation of Add1 at Ser726, leading to dendritic abnormalities and memory impairments in AD [20].
Most mRNA derived from RI event may be degraded via NMD, however, some mRNAs could still be translated [11]. Here, we found that genes spliced by RI were enriched in synaptic plasticity including axonogenesis, neurotransmitter secretion, neuron projection and axon guidance ( Figure 4A, 4B). Gpc2 tended to harbor an intron in APP group ( Figure 4C). Gpc2 has been considered as novel AD risk genes which determine the microglia response to Aβ but not to tau pathology [21].
Moreover, we found that AS events mainly cause dysregulation of genes in transcript level rather than gene expression level ( Figure 5C, 5D). Among the 90 genes involved in significant AS event, only 2 genes (Pld3, Dtwd2) were remarkably dysregulated on gene level. However, on transcript level, 26 transcripts refer to 22 genes involved in significant AS events. Rbm3 and Azi2 harbored transcripts with opposite changes in AD. In addition, AS event level was consistent with corresponding transcripts in Azi2 genes ( Figure 5E-5G), without effect on gene level ( Figure 5H). Moreover, 17 out of the 26 transcripts, whose host genes with significant AS events, were involved in SE event, 6 transcripts were involved in RI event (Table 3). These results suggest that SE and RI events may lead to transcript level dysregulation of different genes in hippocampal microglia in AD. We have also performed GO enrichment analysis of the 17 transcripts with SE event, it seems that these transcripts were not associated with the phenotypes of hippocampal microglia (data were not shown).
The upstream SFs were also explored. 37 significantly dysregulated SF transcripts refer to 30 SFs were identified ( Figure 6A). Two main clusters were selected from SFs PPI networks, and correlation analysis was performed between these SFs and transcripts involved in AS events. Several SFs showed significantly positive or negative correlation with these transcripts (Figure 6E, 6F). These results suggest that dysregulated SFs may govern the AS events of critical AGING AD pathogenesis related genes described here. We have searched several splicing factor databases to explore AS type-specific preferences of these SFs,  In conclusion, we described hippocampal microglia AS in AD. Autophagy related genes were mainly spliced in SE events, actin depolymerization genes were spliced in A3SS events, while synaptic plasticity related genes were mainly spliced in RI pattern. Dysregulation of autophagy, actin depolymerization, and synaptic plasticity have been widely documented in AD. Our results provide a novel therapeutic direction based on AS regulation. These significant AS events may be regulated by dysregulated splicing factors in AD. Our study provides novel pathological mechanisms in the pathogenesis of AD.

Differential expressed gene/transcript analysis
The gene and transcript expression matrices were used for differential expression analysis using limma R package (v3.50.0) [25]. The cutoff values for p value and logFC were 0.5 and 1.5 respectively.

Protein-protein interaction network analysis
The protein-protein interaction (PPI) network analysis was performed using String (https://cn.string-db.org/) and Cytoscape (v3.8.2) [29]. The MCODE app was used to build sub clusters with default settings. The first and second clusters of significantly dysregulated splicing factor transcripts were selected for further transcript correlation analysis.

Transcript correlation analysis
The correlation between significantly dysregulated splicing factor transcripts and significantly dysregulated transcripts whose host genes harbored significant AS events was performed using ggcorrplot R package (v0.1.3). The correlation and p values were calculated. Only correlation results with p value < 0.05 were presented.

Data availability statement
The GSE171195 dataset was deposited in GEO database: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GS E171195. Other data were included in the article or supplementary files.

AUTHOR CONTRIBUTIONS
Data curation, Y.L., J.X. and L.T.; Funding acquisition, X.W.; Supervision, L.C.; Writing-review and editing, X.W. All authors have read and agreed to the published version of the manuscript.