Data pre-processing and snRNA-seq quality control

The dataset is publicly available in the GEO repository (accession number GSE159812), generated using the 3’ tag-based single-cell RNA sequencing protocol (10x Genomics v3) [12]. A total of 15 frontal cortex samples were selected from individuals in the COVID-19 and non-viral control groups. Donor information, including age, sex, group, and number of nuclei, is summarized in Supplementary Table 10.5 in S10 Table, with complete metadata available in the original publication. Raw gene counts were obtained by aligning sequencing reads to the hg38 genome (refdata-gex-GRCh38-2020-A) using CellRanger software (V6.0, 10x Genomics). The cells with high ratio of mitochondrial (more han 10%) and cells with <200 detected features were removed in Seurat 4.0.5 [13]. And technical metrics including median/mean UMI per nucleus and median/mean genes per nucleus are summarized in Supplementary Table 10.7 in S10 Table.

Cells integration and annotations

After quality control, the data were further normalized, scaled, and filtered for highly variable features using Seurat functions NormalizeData, ScaleData and FindVariableFeartures prior to principal component analysis (PCA). Then 15 samples are integrated by Harmony method to reduce the individual difference. The entire dataset is projected onto two-dimensional space by UMAP (Uniform manifold approximation and projection) on the top 20 principal components following the tutorial of Seurat. We annotate cell types referring to the original paper [12] at gene expression levels.

Gene set enrichment analysis

We use irGSEA(v.4.0.0) to analyze and comprehensively evaluate the differential gene sets for cell subsets in the control group and the COVID-19 group. Firstly, we scored individual cells by multiple gene set enrichment methods, and generated multiple gene set enrichment score matrix. Next, we calculated the differentially expressed gene sets for each cell subpopulation in the enrichment score matrix of each gene set by wilcox test. Then the robust rank aggregation (RRA) algorithm in RobustRankAggreg package is used to comprehensively evaluate the results of difference analysis, and the gene sets that are significantly enriched in most of the gene set enrichment analysis methods are screened, visualized and analyzed.

Cells annotations at APA level

According to the tutorial from https://github.com/YangLab/SCAPTURE [14], after the raw data processing following the method described in chapter 3, we get the matrix with PAS transcripts files as the input for single-cell APA analysis, (the process is consistent with single cell gene expression for filtering outliers, normalization, integration, and clustering). The clusters are annotated with the expression of the gene transcripts with specific PAS. We processed 3’ tag-based reads with SCAPTURE to call PAS, remove internal-priming artifacts via its deep-learning classifier, and generate a per-sample PAS-by-cell UMI count matrix. PAS identifiers and cell barcodes were harmonized across samples; lowly expressed PAS were filtered (present in ≥0.5% of cells per sample). The resulting PAS matrices were merged into a Seurat object and underwent the same normalization, integration (Harmony), dimensionality reduction, and clustering procedures as gene expression. For defining cell-type-specific PAS, we first calculated the mean CPM-normalized expression of each PAS within each major cell type in the control group. PAS with a mean expression > 0 in exactly one cell type and 0 in all other cell types were classified as cell-type-specific for that cell type, whereas PAS expressed in more than one cell type were classified as non-cell-type-specific.

APA usage calculation

We quantified per-cell proximal PAS usage using the Proximal Usage Index (PUI). For each gene with ≥2 exonic PAS, PUI_proximal = CPM_proximal/ Σ(CPM of all PAS for that gene), yielding values in [0, 1]; a higher PUI indicates preferential proximal usage. Per-cell APA preference was summarized as the mean PUI across expressed genes. Within each gene (strand-aware), PAS were ordered by genomic position and labeled as Proximal (most 5’), Distal (most 3’), Middle (any intermediates), or Single (only one PAS); we did not use a fixed base-pair distance threshold. For binary contrasts, ‘non-proximal’ denotes the union of Middle + Distal, whereas ‘distal’ refers exclusively to the most 3’ PAS. All transcripts with at least one proximal and one distal PAS site annotated in the database were included in the analysis; thus, transcripts with only a single PAS were excluded.

Cell-type PAS expression was computed from the SCAPTURE PAS-by-cell UMI matrix after per-cell CPM normalization. For each annotated cell type (Idents in Seurat), we calculated the mean CPM per PAS across cells (AverageExpression). Group comparisons (COVID-19 vs. control within the same cell type) were performed on per-cell CPM using FindMarkers (Wilcoxon rank-sum test) to obtain avg log₂ fold-change and adjusted P values; significance was defined as adjusted P < 0.05 and |avg log₂FC| > 0.25.

The GO analysis and KEGG analysis based on the APA

All the subtypes of cell clusters are extracted from the APA usage matrix, and the differential expressions are calculated. Transcripts with adjusted p values less than 0.05 and an average log2 fold change larger than 0.25 are selected after the data is obtained for differential analysis. The multiple transcripts with distinct PAS transcripts could be produced from a single-gene locus, and we use the genes with significant changes of PAS transcripts as the input for Gene Ontology (GO) biological function and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway. GO process and pathway enrichment analysis of genes with significant alterations of PAS were performed by metascape [15]. The similarity score between terms was determined by calculating the Kappa-test scores between each of the two representative terms chosen from the Metascape results (P value ≤ 0.05). Cytoscape [16] was employed to visualize the representative GO terms, and terms were set as nodes, and similarity scores of greater than 0.3 were set as edges.

The validation of PAS usage by IGV

Reads from each cell type are extracted from the bam file, and the bam file is converted to the bigwig (bw) format file using deeptools. The metadata bed file from SCAPTURE is used as the reference and imported to the IGV software, and the interested cell types of bw file are imported to the IGV for visualization.

The 3’-UTR changes in APA

The PAS located in 3’-UTR are extracted, and the corresponding transcripts were analyzed using the movAPA [17] software following its tutorial to calculate 3’-UTR changes. For each gene, movAPA calculated the ratio of PAS transcript counts in the COVID-19 group and the non-viral control group at proximal and distal sites. Differential PAS usage was defined as the usage value in the infected group minus that in the control group, with an absolute difference greater than 0.1 considered a significant change. These APA usage changes were then combined with differential gene expression results to classify genes into four categories.

Global APA levels in neural cells

The PAS were annotated using the latest genome annotation, with 27.9% located in 3’ UTRs, and the remainder distributed across other genomic regions (i.e., 5’ UTR, coding sequence (CDS), and intron) (Fig 1A). Although 3’ UTR PAS are not the majority, they are of particular biological interest because APA events in this region can alter mRNA stability, localization, and translational efficiency by modulating regulatory elements such as microRNA target sites and RNA-binding protein motifs. The poly(A) signals, typically located 15–30 nt upstream of the PAS, are necessary for pre-mRNA cleavage and polyadenylation [18]. The motif enrichment unraveled that the canonical poly(A) signal (AAUAAA) is significantly enriched in the dataset (Fig 1B). Similar to previous studies [3,19], AAUAAA, AUUAAA, and single nucleotide variants of AAUAAA were identified as typical poly(A) signals upstream of the PAS. Fig 1C depicts the genomic distribution of PAS associated with these three poly(A) signal categories, rather than the total PAS distribution shown in Fig 1A. These results support the authenticity of PAS identified by SCAPTURE (S1 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Global APA levels in neural cells.

(A) Distribution of PAS in 3’ UTR. (B) Canonical poly(A) motif (AAUAAA) enrichments for PAS. (C) Genomic distribution of PAS associated with the AAUAAA, AUUAAA, and single nucleotide variants of AAUAAA poly(A) signal categories within the upstream 50 nt of the PAS. (D) Left: UMAP of 16,020 nuclei from the medial frontal cortex of 7 control individuals, identified 7 different clusters. Each point depicting a single cell nucleus, colored according to cluster designation. Right: The mean PAS usage of each single cell for 7 samples after scaling, the expression level is from −3 to 3. (E) Accumulation rate of each cell type based on non-viral control group in solid lines. (F) Proximal rate of each cell type. Statistical comparison between neurons and other cell types was performed using the Wilcoxon rank-sum test (all p-values < 0.001; detailed values are provided in Supplementary Table 10.1 in S10 Table). (G) The numbers of cell-type-specific PAS and their fractions among the total identified PAS in different cell-type populations. PAS diversity values were calculated from normalized mean expression per cell type. (H) Expression levels of MEG3 PAS in neuronal cell types relative to other cell types. Boxplots were generated using standard statistical parameters: the boxes represent the interquartile range (IQR, 25th–75th percentile), the horizontal line within each box indicates the median, and the whiskers extend to 1.5 times the IQR from the lower and upper quartiles.

https://doi.org/10.1371/journal.pone.0324689.g001

We integrated seven samples of frontal cortex from individuals in the control group, generating 16,020 nuclei with 24,813 features. Unsupervised clustering identified seven major cell types (Fig 1D). The cell-type-specific markers uniquely expressed in each subpopulation were obtained from a previously published study [12]. Specifically, glutamatergic neurons are marked by SLC17A7, GABAergic interneurons are marked by GAD1, astrocytes are marked by GFAP, oligodendrocytes are marked by ST18, oligodendrocyte progenitor cell (OPC) is marked by VCAN, microglia are marked by CD74 and endothelial cells by CLDN5 (S1C Fig). S1A Fig shows that glutamatergic neurons represent the largest cell population (51%), followed by oligodendrocytes (18%) and GABAergic interneurons (14%). Astrocytes (9%) and OPC (5%) are present in moderate proportions, while endothelial cells (1%) and microglia (2%) are relatively sparse. Notably, two control samples (FC1 and FC6) exhibit a lower proportion of neurons and a higher proportion of oligodendrocytes compared with other individuals, which may reflect a combination of biological heterogeneity in sampling sites and technical variation. The top 10 most enriched genes in each cluster are shown in the heatmap of S1B Fig, and S1C Fig presents violin plots depicting the differential expression of these cell markers across cell types (S2 Table).

For glutamatergic neurons, the up-regulated gene sets are enriched in Alzheimer’s disease (AD), and amyotrophic lateral sclerosis, belonging to calcium ion signaling and other related pathways, and conversely, down-regulated genes in astrocytes, OPC, oligodendrocytes, microglia, and endothelial cells are also enriched in these pathways. For GABAergic interneurons, genes are mainly enriched in metabolic pathways, the up-regulated genes are enriched in β-alanine, aspartate-alanine-glutamic acid and butanoate metabolism, and citrate cycle, and down-regulated in unsaturated fatty acid biosynthesis and adherens junction. For microglia, genes enriched in antigen processing and presenting are activated. These cells are more likely to participate in inflammation-related responses after infection (S1D Fig).

We calculated the APA usage in frontal cortex cells, the average PUI of each cell is projected into each cell type based on the gene expression in UMAP shown in Fig 1D. The pink color represents the scaled proximal rate mean of each cell, and the glutamatergic neurons and GABAergic interneurons have a relatively deep blue color referring to the mapping of cell subtypes, which indicates they prefer to choose the non-proximal PAS. These results are consistent with the previous reports that the neurons prefer to choose the distal PAS [2,20]. Subsequently, we described the accumulation plot and box plot of each cell type to further systematize the integrated APA usage, and we noticed that the average PUI of neurons is less than that of other cell types (Fig 1E, F). Furthermore, of all the PAS identified in various cell-type populations, neurons account for the greatest number as well as the largest fraction of cell-type-specific PAS (Fig 1G). Therefore, we focused more on changes in PAS levels in neurons compared to other neural cell types.

To address the functional consequences of APA events, we tested for the enrichment of gene ontology and functional pathway annotations among the genes with differential PAS used in neurons (excitatory neurons and inhibitory neurons) compared to non-neuronal nerve cells (S3 Table). Not surprisingly, these genes were highly enriched in biological processes relating to neural development or neural function (S2 Fig). In summary, we observed that a set of genes with neuronal regulatory functions was preferentially subject to distal PAS in neurons.

Neuronal cell types exhibited significantly higher expression of MEG3 PAS. MEG3 has been related to various processes such as necroptosis, apoptosis, inflammation, oxidative stress, endoplasmic reticulum stress, and epithelial-mesenchymal transition, according to recent studies [21–23]. In neurons, the usage of PAS sites was significantly higher in different MEG3 transcripts than in other neural cells (Fig 1H).

Changes in global APA levels in neural cells following infection with SARS-COV-2

The control group (Fig 2A) and positive group (Fig 2C) display different patterns of preference, the shift to proximal in neurons is obvious after being infected with SARS-COV-2, and it indicates the wide changes in APA levels after infections. Each cell type proportion in the non-viral control group, COVID-19 group, and total are calculated. Fig 2B shows that excitatory neurons represent the most abundant cell type. Oligodendrocytes and inhibitory neurons are present in higher numbers compared to other cell types, while endothelial cells and microglia are relatively scarce. The accumulation plot of the infected group described that the neurons’ average PUI is less than other cell types, but the difference becomes smaller after infections (Fig 1E, 2D).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Changes in global APA levels in neuronal cells following infection with SARS-COV-2.

(A) The global preference of APA usage in the infected group. (B) Each cell type proportion in the non-viral control group, COVID-19 group, and the combined dataset (All). (C) The global preference of APA usage in the non-viral control group. (D) Accumulation rate of each cell type based on the infected group with dashed lines. (E) Sankey plot showing relationships among PAS location, PAS type, and cell type. Pink indicates upregulated PAS and blue indicates downregulated PAS after infection. (F) Number of up- and down-regulated PAS per cell type after infection. Red bars indicate up-regulated PAS, and blue bars indicate down-regulated PAS. Statistical comparisons based on normalized proportions are provided in Supplementary Table 10.2 in S10 Table. (G) Heatmaps showing the distribution of upregulated (red) and downregulated (blue) PAS for each cell type in the brain between the non-viral control group and the infected group. PAS not differentially expressed are in white and the numbers of PAS are indicated. The upper part indicates the PAS shared by at least two cell types, and the lower panel indicates the unique PAS of each cell type. The numbers of PAS are annotated on the plots. (H) Dot plot showing the representative GO terms enriched for distal and proximal PAS in different cell types. The upper part indicates the distal PAS, and the lower panel indicates the proximal ones.

https://doi.org/10.1371/journal.pone.0324689.g002

Based on the above results, we further examined APA changes in each cell cluster and assessed statistical significance using the Kolmogorov-Smirnov test. In the cumulative curve plots (S3A, S3B Fig), glutamatergic and GABAergic neurons exhibited pronounced shifts toward proximal PAS usage in the COVID-19 group compared with the non-viral control group. In contrast, astrocytes (S3C Fig), oligodendrocytes (S3D Fig), OPCs (S3E Fig), microglia (S3F Fig), and endothelial cells (S3G Fig) showed only minor changes, with a slight tendency to shift toward non-proximal PAS usage after infection. This statistical analysis confirms that the differences observed in Figs 2A and 2C are supported by quantitative evidence rather than solely by visual inspection. For descriptive purposes only, the reported neurological symptoms of each COVID-19 donor are summarized alongside their sample-level APA metrics in Supplementary Table 10.6 in S10 Table.

The distinct trends observed after infection indicate that APA usage changes among different brain cell types. To further characterize these dynamics, we compared PAS events between the infected and control groups and visualized the results with a Sankey plot (Fig 2E), which illustrates the relationships among PAS genomic location, PAS type, and cell type. The left grey column represents the PAS location in the exon of genes (Fig 2E), and the 3’-UTR accounts for the largest proportion. The second column represents the PAS type. For analytical purposes, all non-proximal PAS were classified as distal PAS to facilitate calculation. The proportion of distal PAS usage is higher than that of proximal PAS, suggesting that brain cells preferentially utilize distal PAS. The third column is the cell type. We then summarized the number of significantly up- and down-regulated PAS in each cell type using bar plots (Fig 2F). Neurons, including glutamatergic neurons and GABAergic interneurons, showed a predominance of downregulated PAS, whereas astrocytes, oligodendrocytes, and OPC exhibited more upregulated PAS following infection (S4 Table). To account for the different number of testable PAS across cell types, normalized proportions and statistical comparisons are provided in Supplementary Table 10.2 in S10 Table, which confirmed the same trends.

The data in Fig 2E and 2F are consistent with the accumulation rate in the proximal rate shift in S3 Fig. In these two types of neurons, the increased proportion of downregulation PAS in proximal sites (Fig 2F) results in the smaller mean PUI after infections, which could explain why neurons have an increasing proximal accumulation rate after infection in S3A, S3B Fig. Besides, the other types of cells have a relatively low proportion of downregulated PAS in proximal sites and more upregulated PAS in distal sites, and the proximal accumulation rates are decreasing in these cell types (S3C-G Fig).

To validate the analysis results, we used the raw bam file from Cellranger, split it by cell types, and converted each file into the WIG format. The proximal and distal PAS usage was then inspected in IGV. Validation examples for selected PAS from each cell type, showing differential expression between the infected and control groups, are presented in S4 Fig. For each group, the first four samples are displayed, with peak heights in the sample tracks representing PAS expression levels. These visual results are compared with the SCAPTURE calculation.

The trends in PAS usage changes across these six cell types are consistent with the results shown in S5 Table. For instance, in oligodendrocytes, at the MT3 gene S4D Fig), the track heights in the control group are higher than those in the infected group, consistent with the results in S5 Table. Compared to the control group, the COVID-19 group exhibits reduced PAS usage in the MT3 gene, with an average log2 fold change of −1.1932.

To further investigate the effects of this significant PAS site usage at the cellular level, we compared the PAS expression of cell types between groups. To this end, we identified more than one thousand differentially expressed PAS (|avg_logFC| > 0.25) in at least one cell type. 370 upregulated and 666 downregulated PAS were identified. Notably, approximately half of the differentially expressed PAS were specifically detectable in the neurons and astrocytes (Fig 2G).

Gene ontology (GO) analysis of differentially expressed PAS further uncovered that downregulated distal PAS in glutamatergic neurons was enriched largely in neuronal development and function (Fig 2H), such as synaptic plasticity, nervous system development, learning or memory, cognition, and so on, highlighting the importance of downregulated expressed distal PAS in neurons after infection, while downregulated expressed proximal PAS in astrocyte and oligodendrocyte were mainly associated with synaptic function after infection.

The APA usage and its correlation with gene expression level

It has been reported that alternative PAS selection within the 3’-UTR can modulate post-transcriptional regulation and influence gene expression levels. We examined this relationship in glutamatergic neurons and GABAergic interneurons by integrating APA usage changes with differential gene expression results. We calculated APA usage by subtracting the usage value in the control group from the infected group using movAPA. We considered changes with an absolute value greater than 0.1 to be meaningful, following thresholds applied in previous studies [24] and taking into account the smaller effect sizes typically observed in single-nucleus datasets. The joint analysis enabled classification of genes into four categories corresponding to the quadrants in Figs 3A and 3B. In the first quadrant, APA usage value of the infected group is relatively larger, and the corresponding gene expression is increased; In the second quadrant, the APA usage value of the infected group is relatively small, and gene expression is increased; In the third quadrant, the APA usage value of the infected group was relatively small, and the gene expression was decreased; while in the fourth quadrant, both APA usage and gene expression are decreased.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. The APA usage and its correlation with gene expression level.

Relationship between APA usage change (y-axis, infected – control, |Δusage| > 0.1 considered significant) and gene expression change (x-axis, avg_logFC) in (A) glutamatergic neurons and (B) GABAergic interneurons. (C) Global gene expression of NEFL in the infected (left) and control (right) group. (D) NEFL gene expression in cell clusters. (E) NEFL-3 APA levels in cell clusters. The y-axis represents scaled PUI values. (F) The PAS usage of NEFL at distal and proximal sites by IGV.

https://doi.org/10.1371/journal.pone.0324689.g003

In glutamatergic neurons (Fig 3A), the genes DDX17, RBM39, FNBP4, and others are located in the first quadrant indicating these genes prefer to use the distal PAS sites and the gene expression level increased. Genes like RPL37, RPL13A, GRINA, ATFB, and CIRBP located in the second quadrant prefer to use the proximal sites with increased gene expression. As for MAP7D2, MAP9 or others in the third quadrant tend to use the proximal sites, and gene expression is relatively downregulated in the infected group. The genes located in the fourth quadrant, such as NEFL, SCD5, and HSP90AA1, predominantly utilize distal PAS sites and exhibit downregulated gene expression level. For GABAergic interneurons in Fig 3B, we find that some genes have similar location patterns. For example, CIRBP is located in the second quadrant in these two types of cells, and NEFL in the fourth quadrant, for both cell types. For other cell types, NEFL is also predominantly located in the fourth quadrant (S5 Fig, S6 Table).

NEFL is a light chain of neurofilaments that functionally maintain the neuronal caliber. Individuals with dementia have elevated levels of NEFL in the cerebrospinal fluid (CSF) [25,26]. In Fig 3C, UMAP visualization is used to illustrate the spatial distribution of NEFL expression across cell types, showing that NEFL is mainly enriched in glutamatergic neurons and GABAergic interneurons, with a visible reduction in the COVID-19 group. The quantitative gene expression levels for each cell cluster are provided in Fig 3D, which confirms a significant decrease in NEFL expression in the COVID-19 group compared to controls. For APA usage (Fig 3E), the PAS site usage in NEFL-3 is decreased after infection. In our APA analysis, NEFL-3 refers to the third annotated PAS within the NEFL transcript. The PAS usage results (Fig 3E) show that NEFL-3 usage is decreased after infection, suggesting that COVID-19 may alter NEFL transcript processing at the 3’-UTR level, potentially contributing to the observed expression changes.

To validate the result of NEFL, we visualized the abundance of PAS sites using wig file of excitatory neurons and inhibitory neurons obtained from raw bam file by IGV shown in Fig 3F.

NEFL is located on the negative strand of the chromosome, with the 5’-UTR positioned on the right and the 3’-UTR on the left. Proximal PAS sites are marked with orange triangles, and distal PAS sites with red triangles in Fig 3F. After infection, the distal PAS sites show higher average peak heights relative to total PAS usage compared with the control group, particularly in sample CV19-FC3. This observation is consistent with the trend shown in Figs 3A and 3B, indicating a shift toward distal PAS usage in NEFL following infection.

Genes with variational APA levels and differential expression widely affected neural function in microRNA-dependent fashion

Previous studies have demonstrated that APA events can influence gene regulation in a microRNA (miRNA)-dependent fashion. Based on this knowledge, we assessed the potential impact of APA events on miRNA binding by identifying miRNA target sites gained or lost in the 3’UTR after APA changes in infection samples [27,28]. In infected samples, APA results in the inclusion or exclusion of miRNA target sites within gene 3’-UTRs.

To determine which miRNAs might be affected, we applied three curated databases-miRcords [29], miRTarBase [30], and TarBase [31]-to the genes showing APA changes and identified 2,404 miRNAs whose target sites are predicted to be impacted (S7 Table) in the four quadrants across all cell types. Figs 4A and 4B summarize these predicted changes, showing the numbers of miRNA target sites gained or lost in each major cell type as a result of APA shifts. Nearly half of these miRNAs being cell-type-specific (Fig 4C, 4D).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Genes with variational APA levels and differential expression widely affected neural function in a microRNA-dependent fashion.

(A) Radial plot showing the number of microRNAs targeting the four quadrants of genes. (B) Heatmap showing the changes in the 3’-UTR of microRNA target genes, either lengthen (red) or shorten (blue). (C) Radar Charts showing the total, (D) specific microRNAs targeting lengthen or shorten 3’-UTR across cell types. (E) Circos plots show the overlap of genes from different cell types in quadrant 2 (left) and quadrant 4 (right). Connecting lines link the same genes that are shared by cell types. (F) GO terms and pathways of quadrant 2 (left) and quadrant 4 (right) genes in different cell types. The network nodes were displayed as pies. Each pie sector is proportional to the number of hits originating from a gene list. The pie charts are colored by cell types, where the size of a slice represents the percentage of genes under the term that originated from the corresponding cell type. Terms with a similarity score > 0.3 are linked by an edge (the thickness of the edge represents the similarity score).

https://doi.org/10.1371/journal.pone.0324689.g004

In order to further investigate the effects of these genes, we examined their distribution across cell types. In quadrant 2, the largest numbers of genes were observed in astrocyte and oligodendrocyte (31 and 51 genes), while glutamatergic and GABAergic interneurons contained 50 and 51 genes, respectively, in quadrant 4 (Fig 4E). Functional annotation enrichment analysis of these predicted target genes revealed terms related to neuronal processes, including “neuron projection development” and “regulation of neurotransmitter levels” in quadrant 4 for glutamatergic neurons, GABAergic interneurons, and oligodendrocyte, “long-term depression” and “GnRH secretion” in both glutamatergic neurons and GABAergic interneurons, and “trans-synaptic signaling” and “regulation of system process” in astrocyte and oligodendrocyte in quadrant 2. “RNA splicing”, “glial cell differentiation”, and “Parkinson disease” were enriched in quadrant 1 (Fig 4F, S6 Fig, S8 Table). These findings are predictive in nature, and further biological assays will be required to confirm the APA–miRNA regulatory mechanisms.

Neurological and psychiatric disease risk genes are significantly altered in APA following infection

To unravel the potential mechanisms underlying the development of neurological and psychiatric effects following SARS-COV-2 infection, fifteen common neurological and psychiatric disorders, including AD, Parkinson’s disease (PD), epilepsy, autism spectrum disorders (ASD), depressive disorder, bipolar disorder (BD), attention deficit hyperactivity disorder (ADHD), multiple sclerosis (MS), ischemic stroke, schizophrenia (SZP), amyotrophic lateral sclerosis (ALS), Huntington disease (HD), migraine disorder, brain aging, and anxiety, were selected for joint analysis using the two databases DisGeNET [32], BioKA [33] and diseases from the GWAS [34] catalogue in the published paper [12]. After compiling all genes in the four quadrants showing variations in APA levels and differential expression across the seven cell types (glutamatergic neuron, GABAergic interneuron, astrocyte, endothelial, microglia, oligodendrocyte, and OPC), we identified a total of 408 genes, of which 267 were classified as risk genes according to the above databases (S9 Table; Fig 5A, 5B). These 267 genes showed significant enrichment for neurological disorders and traits, particularly AD, PD, and schizophrenia. For example, CALM1, a component of the calcium signal transduction pathway [35], was downregulated accompanied with a shorter 3’-UTR after SARS-COV-2 infection (S7A, S7B Fig). A recent study demonstrated that elimination of the Calm1 long 3’-UTR mRNA isoform via CRISPR-Cas9 impairs dorsal root ganglion development and hippocampal neuron activation in mice [36]. The amyloid precursor protein (APP) has received considerable attention due to its proposed role in the pathogenesis of AD [37,38]. We discovered that expression of APP increased after infection, with oligodendrocytes displaying a longer 3’-UTR and astrocytes a shorter 3’-UTR (S7C, S7D Fig). It has been verified that APP 3’UTR length can affect translation efficiencies. A variation in the length of the APP 3’UTR, whether long or short, can cause aberrant expression and eventually increase the risk of AD [39]. Actin constitutes a prominent cytoskeletal protein within eukaryotes organisms, and is recognized for its localization in neuronal synapses. Investigation into beta-actin (ACTB) reveals that two alternative transcripts were induced and terminated at tandem polyA sites, the longer 3’UTR brings higher stability to the transcript, resulting in higher gene expression, in mouse neuronal cells [40]. In our study, the expression of ACTB increased after infection, with a longer 3’-UTR in astrocytes (S7E, S7F Fig). TMOD2 acts as a fibrogenic gene in idiopathic pulmonary fibrosis (IPF) and exhibits a shortened 3’UTR and increased protein expression levels [41]. In the present study, after SARS-COV-2 infection, TMOD2 shows 3’-UTR shortening as well as expression downregulated.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Neurological and psychiatric disease risk genes are significantly altered in APA following infection.

(A) Dot plots showing that genes from four quadrants overlapped with genes from neurological and psychiatric disorders associated gene sets. The pie charts are colored by quadrants, with size indicating the frequency of genes. (B) Network visualizing the overlap between the genes in the four quadrants and genes involved in 15 common neurological and psychiatric disorders. The colors of grey, blue, and orange nodes represent targeting miRNAs, genes and diseases, respectively. The colors of grey and light orange lines represent the connection between genes-miRNAs and genes-diseases.

https://doi.org/10.1371/journal.pone.0324689.g005

The remaining 141 genes that were not assigned to known disease‐associated gene sets are listed in Supplementary Table 10.3 in S10 Table, and their GO functional enrichment analysis is provided in Supplementary Table 10.4 in S10 Table. Although these genes are not linked to known brain disease risk in current databases, they are involved in essential molecular and cellular functions-such as regulation of RNA processing, viral processes, cilium assembly, and translation-that could be relevant to cellular responses and pathophysiological processes during infection.