SARS-CoV-2 Selectively Induces the Expression of Unproductive Splicing Isoforms of Interferon, Class I MHC, and Splicing Machinery Genes

Abstract

RNA processing is a highly conserved mechanism that serves as a pivotal regulator of gene expression. Alternative processing generates transcripts that can still be translated but lead to potentially nonfunctional proteins. A plethora of respiratory viruses, including severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), strategically manipulate the host’s RNA processing machinery to circumvent antiviral responses. We integrated publicly available omics datasets to systematically analyze isoform-level expression and delineate the nascent peptide landscape of SARS-CoV-2-infected human cells. Our findings explore a suggested but uncharacterized mechanism, whereby SARS-CoV-2 infection induces the predominant expression of unproductive splicing isoforms in key IFN signaling, interferon-stimulated (ISGs), class I MHC, and splicing machinery genes, including IRF7, HLA-B, and HNRNPH1. In stark contrast, cytokine and chemokine genes, such as IL6 and TNF, predominantly express productive (protein-coding) splicing isoforms in response to SARS-CoV-2 infection. We postulate that SARS-CoV-2 employs an unreported tactic of exploiting the host splicing machinery to bolster viral replication and subvert the immune response by selectively upregulating unproductive splicing isoforms from antigen presentation and antiviral response genes. Our study sheds new light on the molecular interplay between SARS-CoV-2 and the host immune system, offering a foundation for the development of novel therapeutic strategies to combat COVID-19.

1. Introduction

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) is the pathogen that causes the coronavirus disease 2019 (COVID-19), which to date has killed over 6.7 million people worldwide. SARS-CoV-2 infects the upper respiratory tract and lungs and triggers a systemic inflammatory response that can lead to respiratory failure and death in severe cases [1]. Unlike other coronaviruses and viruses known to interfere with host splicing machinery, such as herpes simplex virus type 1, Epstein–Barr virus, and yellow fever virus, SARS-CoV-2 uniquely blocks interferon (IFN) expression and modulates specific immune cell populations [2,3,4,5,6,7,8]. Given the demonstrated ability of SARS-CoV-2 to interfere with host splicing machinery [9,10], understanding the consequences of perturbations in splicing and unproductive splicing isoform expression for the host’s antiviral response is crucial. Elucidating these mechanisms may help develop strategies to counteract the virus’s immune evasion tactics.

In this study, we particularly focus on annotated unproductive splicing isoforms, which are transcripts that do not code for functional proteins due to alterations in the splicing process. Unproductive splicing isoforms can result from various RNA processing events, such as premature translation-termination codons (PTCs), alternative splice site selection that disrupts the open reading frame (ORF) by retaining partial or full introns, and transcription alternative start and endpoints. Some unproductive isoforms can be targeted for Nonsense-mediated mRNA decay (NMD) or remain untranslated, ultimately reducing protein production.

Considering these observations, our study aims to investigate the transcriptome-wide, isoform-level consequences of SARS-CoV-2-induced disruption in splicing, particularly focusing on the expression of unproductive splicing isoforms of key antiviral genes. We hypothesize that SARS-CoV-2 selectively manipulates the host splicing machinery to promote viral replication and subvert the usual immune response. We utilized bioinformatics approaches and publicly available omics datasets to gain a comprehensive understanding of the consequences of SARS-CoV-2 infection on host splicing. Our integrative analysis combines transcriptomic and nascent-peptide data, enabling a more detailed characterization of the infected host cell landscape.

We found that SARS-CoV-2 infection induces a predominant expression of unproductive splicing isoforms of key IFN signaling genes and ISGs, class I MHC genes, and splicing machinery genes, such as IRF7, OAS3, HLA-B, and HNRNPH1. In contrast, this is not detected for cytokine and chemokine genes, such as IL6, CXCL8, and TNF, which preferentially express productive (protein-coding) splicing isoforms in response to SARS-CoV-2. We further found that several genes overexpressing unproductive splicing isoforms upon SARS-CoV-2 infection also show reduced protein synthesis in infected cells, including RBM5 and SRSF5. Finally, we demonstrated that many proteins and mRNAs affected by the SARS-CoV-2 disruption of splicing described here are also direct targets of viral proteins according to two distinct protein–protein interaction networks (Figure S1).

Our findings have significant implications for the development of therapeutic strategies against COVID-19. By uncovering the molecular interplay between SARS-CoV-2 and host splicing machinery, we provide insights into potential targets for antiviral drugs and immunomodulatory interventions. These discoveries may guide the design of therapies aimed at restoring normal RNA processing during viral infections.

2. Results and Discussion

2.1. SARS-CoV-2 Infection Increases the Abundance of Unproductive Splicing Isoforms

We first analyzed a publicly available RNA-Seq dataset of A549 lung epithelial cells infected with SARS-CoV-2 [2] to investigate the effect of viral infection on gene and RNA isoform expression (Figure 1A). The dataset includes wild-type (WT)- or ACE2-overexpressing (ACE2) A549 cells infected with SARS-CoV-2 at a low (0.2) or high (2.0) multiplicity of infection (MOI) for 24 h (Figure 1A).

Consistent with viral load results (Figure 1A), which was higher in ACE2 cells, we detected the highest number of differentially expressed genes (DEGs) in ACE2-MOI2.0 cells (Figure 1B and Table S1). In these cells, we identified 243 up- and 15 downregulated long non-coding RNAs (lncRNAs), 87 up- and 12 downregulated non-coding RNAs (ncRNAs), and 587 upregulated and 194 downregulated protein-coding DEGs (Figure 1B). The total number of DEGs in ACE2-MOI2.0 cells was markedly higher than those in ACE2-MOI0.2, WT-MOI2.0, and WT-MOI0.2 cells (Figure 1B).

Gene-level expression analysis is standard for RNA-Seq but overlooks alternative RNA processing into productive or unproductive isoforms (Figure 1C), a mechanism cells use for dynamic protein expression control [11]. For example, the abundance of productive isoforms of inflammatory genes in response to tumor necrosis factor (TNF) is mostly regulated through splicing and not by differential transcription [12].

We demonstrated that the proportion of differentially expressed productive transcripts (DETs) of protein-coding genes decreases among upregulated DETs in ACE2-overexpressing cells and is also affected by higher MOI (Figure 1D). Interestingly, the number of upregulated processed transcript and NMD DETs was higher in WT-MOI2.0 cells than that seen in ACE2-MOI0.2 cells (Figure 1E), which was not detected at the gene level (Figure 1B) or for retained intron isoforms (Figure 1E). Transcript-level analysis reveals a more detailed landscape of how genotype (WT or ACE2) and MOI (high or low) might distinctly influence the expression of certain annotated transcript types, in SARS-CoV-2-infected cells.

As ACE2-MOI2.0 cells presented the highest number of DEGs and DETs, we focused our subsequent analyses on those cells. Results for the other cells are available in the Supplementary Material (Figure S2).

2.2. SARS-CoV-2 Selectively Induces Unproductive Isoforms of IFN Signaling Genes While Promoting Productive Expression of Inflammatory Genes

We performed a gene set enrichment analysis (GSEA) on differentially expressed genes (DEGs) and differentially expressed transcripts (DETs) in infected cells to elucidate the biological functions modulated by SARS-CoV-2. Utilizing both productive and unproductive transcripts of protein-coding genes in ACE2-MOI2.0 cells, we examined the impact of each component on selected pathways (Figure 2A). Pathways implicated in pathogen response were chosen based on manual curation of the enrichment analysis and are illustrated in Figure 2. Comprehensive GSEA results can be found in the Supplementary Material (Table S2).

We observed an upregulation in IFN response-related pathways, such as viral sensing, IFN induction, IFN signaling, and interferon-stimulated genes (ISGs) in both productive and unproductive isoforms (Figure 2A) and at the gene level (Figure S2A). However, key genes in the IFN-mediated antiviral response, including MYD88, STAT2, and IRF7, demonstrated a higher fold change in unproductive isoforms (Figure 2B). Notably, the IRF7 gene expressed only unproductive overexpressed isoforms in ACE2-MOI2.0 cells (Figure 2C and Figure S2C), whereas WT-MOI0.2 cells predominantly overexpressed productive IRF7 transcripts (Figure S2D). We propose that SARS-CoV-2 may amplify a constitutive feedback mechanism involving intron 4 retention in the Irf7 gene seen in mice, which suppresses productive IRF7 isoform expression and dampens the IFN response following innate immune stimulation [13].

ISGs exhibited a more balanced overexpression of productive and unproductive isoforms compared to early IFN signaling genes (Figure 2B). Key genes, such as MX1, OAS1, OAS2, and RSAD2, displayed a higher fold change in productive isoforms (Figure 2B), whereas TRIM22 and OAS3 (Figure 2D) showed higher expression of unproductive isoforms. A recent study demonstrated that two common human OAS1 locus variants (rs1131454 and rs10774671) are enriched in European and African COVID-19 patients requiring hospitalization [14]. These variants elevate the expression of an OAS1 transcript with a premature stop codon, resulting in transcript NMD degradation, reduced OAS1 protein expression, and impaired viral clearance [14]. In a similar vein, Frankiw et al. [15] reported that splicing at a frequently used splice site in the mouse Oas1g gene generates an NMD-targeted transcript, while removal of the site leads to increased Oas1g expression and enhanced antiviral response. Hao and Baltimore [13] showed that TNF stimulation that is induced by innate immune gene expression is regulated by splicing, particularly in the regulation of productive isoform expression, rather than transcription.

These findings support our results, emphasizing the significance of the productive versus unproductive balance in regulating key ISGs in response to SARS-CoV-2. Interestingly, gene- and transcript-level ISG expression was increased only in cells with the highest (ACE2-MOI2.0) and lowest (WT-MOI0.2) viral loads (Figure S2A). This suggests that SARS-CoV-2 can silence ISG expression in cells with intermediate viral loads but not in cells infected with either an excess or a paucity of viral particles. Blanco-Melo et al. [2] proposed that this effect, which they also observed for IFNalpha and IFNβ proteins, could stem from a stoichiometric competition between host and viral components, favoring the virus in cells with intermediate viral loads. When few viral particles are present, host components outnumber viral ones, leading to a successful response. Conversely, the presence of numerous viral copies may induce a robust host response, potentially shifting the balance in favor of the host. Our findings lend support to this notion for ISG expression in infected cells as well.

Taking a different perspective, numerous genes involved in promoting and regulating inflammation via NFκB, such as TANK, REL, RELB, and NFKB1, displayed a higher fold change in their productive isoforms (Figure 2B). Conversely, the gene CHUK, which encodes for the NFκB Inhibitor Kinase Alpha (IKK-α), exhibited a greater overexpression of unproductive isoforms (Figure 2B), potentially contributing to increased NFκB-mediated inflammation. Additionally, inflammatory cytokine and chemokine genes, including IL6, CXCL8, CXCL2, and CCL2 (Figure 2B,E), demonstrated a marked positive enrichment of productive isoforms (Figure 2A). No enrichment was detected for the unproductive isoforms of cytokine and chemokine genes in ACE2-MOI2.0 cells (Figure 2A).

These findings align with a clinical hallmark of severe COVID-19: exacerbated lung inflammation driven by hyperresponsive immature neutrophils and classical monocytes [16]. Our results suggest that the critical imbalance observed in COVID-19 patients may result from the biased impact of SARS-CoV-2 on inducing unproductive splicing of IFN-related genes, an effect not seen for key inflammatory genes.

2.3. SARS-CoV-2 Promotes Unproductive Splicing of Class I MHC Genes

Pathways associated with antigen presentation through class I MHC exhibited greater upregulation among unproductive isoforms compared to productive isoforms (Figure 2A). Crucial antigen presentation genes with higher unproductive fold changes included proteasome components PSMB3, PSMB5, PSMB6, PSMB9, and PSMD8. Genes encoding core class I MHC complex elements, such as B2M, HLA-A, and HLA-B, also displayed a higher unproductive fold change. The HLA-B gene revealed five distinct overexpressed retained intron isoforms in ACE2-MOI2.0 cells (Figure 2F and Figure S2C). This upregulation of unproductive isoforms could potentially decrease HLA-B protein production in SARS-CoV-2-infected cells, impairing antigen presentation.

A 10 bp deletion in the 3′ region of intron 1 of the HLA-B*1501 allele was identified in a bone marrow donor, promoting intron retention and resulting in the absence of HLA-B15 antigen from the donor’s serum [17]. Similarly, melanoma cells possess a mutation at the 5′ splice donor site of intron 2 of the HLA-A2 gene, causing the absence of the HLA-A2 antigen and likely protecting these cancerous cells from immune surveillance [18]. Our findings indicate that SARS-CoV-2 infection may also lead to the loss of class I MHC antigens due to intron retention, potentially rendering infected cells incapable of presenting viral antigens to the immune system. Reduced expression of HLA-DR in monocytes represents a molecular hallmark of severe COVID-19 [19]. Other studies have demonstrated that SARS-CoV-2 ORF6 inhibits class I MHC production by blocking the STAT1-IRF1-NLCR5 axis [20] and that ORF3a and ORF7a downregulate class I MHC surface expression in infected cells through distinct mechanisms [21]. Our results expand on this existing evidence by revealing that SARS-CoV-2 also promotes unproductive splicing of key class I MHC genes.

2.4. Unproductive Isoforms of Splicing Factors and Mediators Highly Expressed during SARS-CoV-2 Infection

In ACE2-MOI2.0 cells, significant upregulation of several unproductive transcripts of genes related to various stages of the splicing mechanism was observed (Figure 2A). Key genes encoding proteins involved in recognizing splicing elements for splice site selection, such as HNRNPH1 and HNRNPHL (Figure 2B), and those regulating alternative splicing, including RBM5 (Figure 2G) and SRSF5, displayed a higher fold change of their unproductive isoforms in ACE2-MOI2.0 cells (Figure 2B). Cells often modulate splicing factor abundance by regulating unproductive isoform expression, subsequently affecting the splicing of several other genes in response to stress [22].

SARS-CoV-2 and numerous other viruses exploit this mechanism by hijacking, disrupting, or modulating host splicing machinery. For instance, the herpes simplex virus type 1 protein ICP27 binds to SRPK1 and SR proteins, inhibiting cellular splicing [6], while the Epstein–Barr virus SM protein hijacks the cellular protein SRp20 during alternative splicing [7]. The yellow fever virus NS5 protein interacts with proteins involved in polyadenylation [11], and the influenza A NS1 protein interacts with the U6 snRNA, inhibiting pre-mRNA splicing [16]. Our findings reveal that SARS-CoV-2 selectively manipulates the host’s splicing machinery to induce unproductive transcript expression for a subset of essential antiviral genes and genes associated with the splicing machinery itself. Wang et al. [23] recently demonstrated that COVID-19 patients exhibit global alterations in alternative splicing, and disease severity correlates with the degree of splicing perturbation.

2.5. Nuclear mRNA Export and NMD Pathways Impacted by SARS-CoV-2 Induction of Unproductive Isoform Expression

Genes involved in transporting mature mRNA transcripts to the cytoplasm, such as NUP35, NUP98, NUP205, and POM121C (Figure 2B,H), exhibited higher upregulation of their unproductive isoforms (Figure 2A,B). SARS-CoV-2 ORF6 protein has been consistently shown to interact with and block the nuclear pore, hindering the entry of IFN mediators into the nucleus and the exit of mature ISG mRNAs from the nuclear compartment [4,24,25,26]. Our findings indicate that core nuclear pore genes are also affected at the transcript level through the induced expression of unproductive isoforms by SARS-CoV-2.

Genes involved in the NMD pathway exhibited a shift toward an unproductive expression profile, including genes of the mRNA-EJC complex formation, a key point of NMD (Figure 2A,B). Although the UPF1 gene was downregulated at the gene level, it displayed increased expression of one retained intron isoform at the transcript level (Figure 2I). The protein encoded by UPF1 is essential for recognizing premature termination codons during the NMD process, although the presence of EJC expression does not directly imply NMD activation. NMD-involved proteins can also detect and degrade viral RNA, serving as an additional layer of antiviral response [27]. Ribosomal protein (RP) genes, eventually mistakenly categorized as NMD, but part of the canonical translation process, were also distinctly affected in their productive and unproductive isoforms by SARS-CoV-2 infection (Figure 2A,B). Genes coding for RPs from both large and small subunits, including RPS19, RPL26, and RPL29 (Figure 2J), demonstrated reduced expression of multiple productive isoforms and overexpression of unproductive transcripts (Figure 2B).

Given that SARS-CoV-2 NSP16 protein directly disrupts splicing [9,28], we anticipated detecting differential RNA processing in various genes. Our results confirmed this hypothesis and revealed that the effect is not evenly distributed across different biological processes. In addition to splicing, NMD, and translation-associated genes, upregulated unproductive splicing isoforms are being transcribed from genes of the IFN-mediated immune response and class I MHC complex, while virtually absent in cytokine and chemokine genes.

2.6. Genes with Upregulated Unproductive Transcripts Are Less Translated in SARS-CoV-2-Infected Cells

Given the substantial impact of SARS-CoV-2 infection on the expression of unproductive splicing isoforms of key genes, we investigated whether these alterations could also be detected at the protein level. We analyzed a publicly available translatome dataset of colorectal adenocarcinoma cells (Caco-2) infected with SARS-CoV-2 at an MOI of 1.0 at 2, 6, 10, or 24 h post-infection (Figure 3A) [29]. Nascent peptides were labeled at each time point using heavy isotope-containing lysine (K8) and arginine (R10), enabling their quantification by LC–MS/MS (Figure 3A) [29]. Caco-2 cells are another common model used for epithelial cells and produce the ACE2 protein in higher quantities by default. Here we aimed to detect the molecular similarities present in both cell types upon SARS-CoV-2 infection. The number of differentially translated proteins (DTPs) in response to infection steadily increased from 2 to 24 h (Figure 3B), with the majority being downregulated, except at 2 h post-infection (Figure 3B). Among the 606 DTPs found at any time point (Table S3), 237 also had differentially expressed transcripts (DETs) in ACE2-MOI2.0 cells (Figure 3C). These genes with both DETs and DTPs exhibited a larger fold change of their unproductive isoforms (Figure 3C) and were highly represented among downregulated proteins at 24 h in the translatome (Figure 3C). A potential caveat is that we also observed a higher number of underrepresented proteins compared to overrepresented ones among the productive isoforms of DETs, which suggests that the downregulation is not solely attributed to the regulation of RNA processing (Figure 3C).

We performed an overrepresentation analysis (ORA) with the up- and downregulated DTPs at each time point to verify which pathways were the most impacted by SARS-CoV-2 at the translatome level (Figure 3D). Enrichment results were more significant at 10 h and 24 h. Proteins related to antigen processing via the proteasome and assembly of the class I MHC complex were downregulated at 24 h (Figure 3D), including PSMD8 and SAR1B (Figure 3E). The genes that code for those proteins also presented a positive fold change of their unproductive isoforms in ACE2-MOI2.0 cells in the RNA-Seq (Figure 3E). HLA-A and B2M, core class I MHC components that also had increased expression of unproductive isoforms at the transcriptome level, had a negative fold change at the translatome level, but with a non-significant associated adjusted p-value (Table S3). The same occurred with the proteins STAT1 and RELA (Table S3), which are key transducers of the IFN-mediated antiviral response. Spliceosome E and A complex proteins were also less translated at 10 h and 24 h (Figure 3D). In total, 16 out of 25 splicing proteins that had a reduced translation rate at 10 h or 24 h also presented a higher fold change of unproductive isoforms of their corresponding gene in the RNA-Seq, including RBM5, HNRPNH1, and HNRNPL (Figure 3E). Nuclear pore complex proteins (NUPs) involved in the transport of mature mRNA transcripts to the cytoplasm were also less translated at 10 h and 24 h in Caco-2 cells infected with SARS-CoV-2 (Figure 3D). Among those, NUP35, NUP205, and NUP98 presented a positive fold change of unproductive isoforms in the transcriptome (Figure 3E). General translation factors were upregulated in the translatome (Figure 3D), including ETF1 and PABPC1, which also showed overexpression of their unproductive isoforms in the RNA-seq (Figure 3E). Ribosomal proteins, on the other hand, were mostly downregulated in the translatome (Figure 3D), including RPL29, which also had downregulated productive isoforms and upregulated unproductive transcripts in the RNA-seq (Figure 3E).

The authors who showed that the SARS-CoV-2 NSP16 protein disrupts splicing also demonstrated that the NSP1 viral protein binds to host ribosomes and inhibits translation [9,30]. In our analysis, we observe translation inhibition, as evidenced by the unproductive splicing of multiple ribosomal components (Figure 2), aligning with protein downregulation (Figure 3). The direct interaction of viral NSP1 is also associated with a downregulation of its targets, demonstrated here at both the protein and transcript levels. Our results suggest that the effects of these viral proteins are not generalist but rather lead to splicing disruption and translational silencing of transcripts involved in specific biological functions, namely, the IFN-mediated innate immune response, class I MHC antigen presentation, splicing, and mRNA processing and translation.

2.7. SARS-CoV-2 Proteins Interact with Key Antiviral, Splicing, and Nuclear Transport Proteins

Lastly, we analyzed networks of publicly available experimentally defined protein–protein and protein–RNA interactions between SARS-CoV-2 proteins and host proteins [25,31]. The first protein–protein interaction network (PPI) had 23 viral proteins with significant interactions with 330 host target proteins [25]. The second PPI included 22 SARS-CoV-2 proteins targeting 876 host proteins [30]. It is worth noting that both PPIs are based on overexpression followed by LC-MS/MS, which could lead to non-physiological interactions, which are associated with a high protein concentration.

Several SARS-CoV-2 proteins interfere with the host’s innate immune system (Figure 4A). ORF7b interacts with IL10RB and the MAVS protein (Figure 4B), a key transducer of the signal from viral sensing elements such as RIG-I and MDA5 [31]. NSP13 binds to the TBK1 protein (Figure 4B), another transducer of the viral sensing signaling pathway [32], and ORF3 interacts with IL10RB and IFNGR1 (Figure 4B), the main IFN-gamma receptor. The M protein interacts with both the IRAK1 protein (Figure 4B), a transducer of toll-like receptor-mediated responses [33], and JAK2 (Figure 4B), a mediator of the IFN receptor signaling. None of the ISGs that were altered at the isoform and translation levels appeared as direct targets of SARS-CoV-2 proteins (Figure 4B), suggesting that the virus deploys distinct, complementary mechanisms to interfere with the innate antiviral host response mediated by interferons. RIPK1 was the only host protein related to the NFκB-mediated inflammatory signaling targeted by a viral protein, NSP12 (Figure 4D). Moreover, we did not find any interaction between viral proteins and cytokine or chemokine proteins (Figure 4B–D), which could be related to the transient nature of these factors.

SARS-CoV-2 peptides also heavily interact with host antigen presentation and class I MHC proteins (Figure 4A). The NSP4 viral protein interacts with the HLA-C and HSPA5 proteins, and ORF3 interacts with HLA-A, HLA-C, HLA-E, HLA-G, and the B2M protein (Figure 4B). ORF7b also interacts with the HLA-A and B2M proteins (Figure 4B). HLA-A and B2M are core components of the class I MHC complex [34], which also had an increased expression of several retained intron transcripts in the RNA-Seq (Figure 2B). ORF8 binds to PDIA3 (Figure 4B), a protein involved in antigen loading to the class I MHC complex [34], which also presented strong upregulation of one of its unproductive isoforms, PDIA3-204, in ACE2-MOI2.0 cells (Table S1). The viral protein M interacts with the proteasome protein PSMD8 (Figure 4B). The PSMD8 gene also showed strong overexpression of unproductive splicing isoforms and reduced translation in SARS-CoV-2-infected cells (Figure 2B and Figure 3E). Thus, analysis of the virus-host interactome shows that SARS-CoV-2 proteins interfere with the presentation of viral antigens at distinct levels, through both direct (protein–protein interactions) and indirect (alternative splicing) manipulation of key proteins of the antigen presentation machinery.

The export of mature mRNA transcripts to the cytoplasm was the process with the most host peptides targeted by SARS-CoV-2 proteins (Figure 4A). Several NUPs and mRNA export proteins were targeted by the viral proteins NSP4, NSP9, ORF6, ORF7a, and ORF7b, including RAE1 and NUP98 (Figure 4C). In the early days of the COVID-19 pandemic, Gordon et al. [25] argued that the interaction between ORF6 and the host proteins NUP98 and RAE1 might prevent the export of IFN-stimulated transcripts from the nucleus. Later studies confirmed that hypothesis, showing that ORF6 functions as a potent inhibitor of the IFN-signaling pathway by preventing several key antiviral transducers, such as STAT1 and IRF3 from entering the nucleus and ISG transcripts from leaving the nuclear envelope [4,24,26,35]. Proteins of several other viruses interact with host nuclear pore proteins during the viral replication cycle, including the HIV-1 capsid protein [36,37] and the NS3 protein of the Zika and Dengue viruses [38]. In conclusion, we showed that SARS-CoV-2 affects NUPs and other mRNA nuclear export genes also at the transcriptional and translational levels, increasing the levels of unproductive splicing isoforms of these genes and, therefore, reducing protein translation.

Core splicing machinery proteins are also targeted by SARS-CoV-2 proteins (Figure 4A). The E protein interacts with PRPF40A, CWC27, and RBM5 (Figure 4D), all of which participate in the early steps of RNA splicing. The RBM5 gene also presented increased expression of unproductive transcripts in SARS-CoV-2-infected cells (Figure 2G). Similarly, the hepatitis delta virus (HDV) promotes unproductive RBM5 expression due to the disrupting interaction of the HDV genomic RNA with the splicing factor SF3B155 [39]. The SARS-CoV-2 M protein interacts with PRPF39, a component of the U1snRNP complex, and with the alternative splicing regulators RBFOX2 and RBFOX3 (Figure 4D). RBFOX2 is a hub protein in a network connecting splicing factors to differentially spliced genes, including STAT1 and NUP43, in herpes virus-induced liver hepatocarcinoma patients [6]. In mice, an Rbfox3 isoform promotes the inclusion of an alternative exon containing a premature stop-codon in Rbfox2 transcripts, leading to their degradation via NMD [40]. Three other SARS-CoV-2 proteins, NSP12, ORF7b, and ORF9b physically interact with the splicing factors PPIL3 and SLU7, SYMPK, and PTBP2, respectively. These results suggest that the splicing perturbations promoted by SARS-CoV-2 characterized here might be a result of a combination of multiple viral factors acting together to disrupt host RNA processing, rather than the sole effect of the NSP16 sequestering U2 snRNA described by Banerjee et al. [9].

Two SARS-CoV-2 proteins, N and E, interact with key NMD proteins. The N protein interacts with both UPF1 and PABPC1 (Figure 4D). The rotavirus protein NSP5 also targets UPF1, leading to its proteasomal degradation and facilitating viral replication due to burdening the NMD pathway [28]. The Porcine Epidemic Diarrhea Virus (PEDV) reduces PABPC1 expression, while overexpression of the gene refrains viral replication [41]. Interestingly, similarly to SARS-CoV-2, the PEDV nucleocapsid (N) protein also binds to PABPC1 [41]. The NMD pathway works as a complementary mechanism of antiviral response by promoting viral RNA decay, at the same time as host unproductive RNAs are also degraded [42]. Our results demonstrate that SARS-CoV-2 physically targets core proteins of this mechanism, while also promoting the unproductive expression of the genes that code for those proteins.

The central finding of our study is that SARS-CoV-2 disrupts the host’s splicing of several genes to potentially silence the antiviral responses of infected cells (Figure 5). After entering a human cell, SARS-CoV-2 expresses NSP16, which translocates to the nucleus, where it binds to and sequesters snRNAs U1 and U2 (Figure 5) [9]. This leads to a disruption in splicing, which increases the expression of intron-retaining transcripts, as well as other unproductive transcripts, such as those containing premature stop codons and missing ORFs (Figure 5). These transcripts, contrary to their productive counterparts, do not become translated into canonical proteins in the cytoplasm (Figure 5). Here we showed that this phenomenon heavily affects genes involved in the innate immune response against viruses, including the interferon-mediated response. Surprisingly, cytokine and chemokine genes are not consistently affected, overexpressing mostly productive splicing isoforms, which can be translated into proteins (Figure 5). Viral antigens are mostly presented through the class I MHC pathway, which is also affected by the disruption of splicing promoted by SARS-CoV-2 (Figure 5). Key genes involved in this pathway are less translated in infected cells and are targeted by several viral proteins. This likely helps the virus to avoid detection by the adaptive immune system. Lastly, SARS-CoV-2 also induces the expression of unproductive splicing isoforms of genes from the splicing machinery, as well as genes related to translation and the export of mature mRNAs from the nucleus (Figure 5). Several genes in these pathways also present reduced translation in infected cells and produce proteins or RNAs that are directly targeted by viral proteins. This shows that SARS-CoV-2 disrupts splicing and mRNA processing using multiple resources both at the start of the process as well as downstream.

3. Materials and Methods

3.1. Public Datasets

Raw RNA-Seq reads were obtained from the study of Blanco-Melo et al. [2] (GSE147507). These reads correspond to triplicate samples of WT A549 cells infected with SARS-CoV-2 at an MOI of 0.2 (series 2) or 2.0 (series 5), ACE2-overexpressing cells infected with SARS-CoV-2 at an MOI of 0.2 (series 6) or 2.0 (series 16), and triplicate samples of the corresponding mock-infected cells in each series. Cells were infected for 24 h before RNA extraction. Libraries were prepared with the TruSeq RNA Library Prep Kit v2 (Illumina, San Diego, CA, USA) or TruSeq Stranded mRNA Library Prep Kit (Illumina) and sequenced on the Illumina NextSeq 500 platform [2].

The differential translation rate results table was obtained from the supplementary material of Bojkova et al. [29]. These translatome results were obtained in Caco-2 cells infected with SARS-CoV-2 (MOI = 1.0) for 2, 6, 10, or 24 h and compared to equivalent mock-infected cells. Nascent proteins were labeled at each time point using heavy isotope-containing lysine (K8) and arginine (R10), which allowed for their quantification by LC–MS/MS. Differentially translated proteins in each time point were those with p-value < 0.1.

Protein–protein interaction networks were obtained from the supplementary materials of Gordon et al. [25] and Stukalov et al. [30], and the protein–RNA interaction network was obtained from the supplementary material of Banerjee et al. [9]. The protein–protein interaction networks were obtained from HEK-293T Gordon et al. [25] or A549 Stukalov et al. [30] cells individually transfected with expression vectors each containing one SARS-CoV-2 protein-coding gene. Host proteins that interacted with the expressed viral proteins within cells were identified by affinity purification followed by mass spectrometry [25]. The protein–RNA interaction network [9] was obtained from HEK293T cells individually transfected with SARS-CoV-2 expression constructs for each viral protein. Expressed viral proteins were cross-linked to their partner host RNA molecules, and the protein–RNA complexes were selectively captured with the HaloLink Resin method. RNA-Seq libraries were prepared from the purified protein–RNA crosslink products and sequenced on an Illumina HiSeq 2500. Sequencing reads were then aligned to the human genome, and genes whose RNAs were bound to viral proteins were identified.

3.2. Quantification of Viral Reads

The NC_045512.2 release of the SARS-CoV-2 complete genome was downloaded from NCBI, and the GRCh38.p13 human genome was obtained from the GENCODE website [43]. The raw reads of each RNA-Seq library were aligned with STAR [44] to a reference genome composed of the human and SARS-CoV-2 genomes concatenated. The number of reads mapped to each genome in each library was calculated using an in-house script. The ratio of viral reads was calculated as the number of viral reads in each library divided by the total mapped reads in the library. Results can be seen in Figure 1A.

3.3. Gene and Transcript Expression Quantification

Transcript-level expression quantification was performed from the RNA-Seq raw reads using the alignment-independent software Salmon version 10.1 [45] using the GENCODE GRCh38.p13 v33 human reference transcriptome [43]. The index was built with k-mers of size 31, and the quant function was run using default settings plus the --validateMappings and --nBootstraps 100 flags. We also used the GENCODE v33 reference transcriptome [43] file to create a transcript to gene (tx2gene) dictionary. This dictionary contained the correspondence between each transcript to its parent gene and the information on the biotypes of each gene and transcript. This analysis was based on provided transcript biotypes according to GENCODE annotation and not on splicing event prediction. The biotypes were used as determinants of productive and unproductive isoforms with productive corresponding to protein_coding transcripts and unproductive to processed_transcript, retained_intron, and nonsense_mediated_decay. The presence of unproductive transcripts was associated with alternative splicing and RNA processing. The package tximport was used to import salmon results to R and to summarize transcript-level to gene-level counts. Technical replicate sequencing libraries were collapsed by summing the counts of each transcript or gene from all replicates.

3.4. Differential Gene and Transcript Expression Analysis

Gene- and transcript-level count tables were used to perform differential expression analysis using DESeq2 [46] with default settings. The design was set to be SARS-CoV-2-infected versus Mock samples in each experiment individually. The resulting genes and transcripts log2FoldChange values were shrunk with the lfcShrink function in DESeq2 using the normal mode [46]. Differentially expressed genes or transcripts were those with an adjusted p-value < 0.05 and abs(log2FoldChange) > 2. The Isoformic transcript-level analysis R pipeline was used for plotting and interpretation throughout the analysis [47].

3.5. Sashimi Plots

Sashimi plots were created to visualize splicing events in selected genes using the ggsashimi command line tool [48]. Briefly, raw RNA-Seq reads were aligned to the GRCh38.p13 human genome using STAR [44]. The resulting BAM files and a GTF file of the same version of the genome were used to run ggsashimi for each selected gene. The genomic coordinates of selected genes were obtained using the biomaRt R package [49].

3.6. Custom Gene Set Construction

We created custom sets of genes related to key biological processes and pathways involved in the innate immune response against viruses, antigen presentation via the class I MHC, the splicing machinery and regulation, the transport of RNA transcripts from the nucleus to the cytoplasm, and with NMD (Table S4). We built these custom gene sets using the terms and genes from the curated databases Reactome, Gene Ontology (biological processes, cellular compartment, and molecular function), MSigDB Hallmark pathways, KEGG, and Biocarta (Table S4). The authors selected the pathways and genes included in the custom gene set list based on their scientific experience.

3.7. Functional Enrichment Analysis

Gene set enrichment analysis (GSEA) was performed for RNA-Seq results using the R package fgsea [50] with the custom pathways created. For gene-level enrichment, we used the shrunk log2FoldChange values of all detected genes (no adjusted p-value cutoff) as recommended by the authors. Isoform-specific functional annotation databases similar to those built for genes are practically non-existent. Therefore, we extended the gene-level annotations of the selected genes in each custom pathway to their isoforms (Table S4) and ran separate GSEAs on fgsea with the shrunk log2FoldChange values of productive or unproductive transcripts separately. Significant pathways at the gene and transcript levels were those with a p-value < 0.05.

DTPs were separated into up- and downregulated according to the log2FoldChange values provided by the authors who produced the data [28]. Up- and downregulated DTPs were separately submitted to overrepresentation analysis (ORA) using the fora function in the fgsea package [50]. Significant pathways were those with a p-value < 0.05. For the SARS-CoV-2 versus host protein–protein and protein–RNA interaction networks, we quantified the total number of host proteins or genes in each network that were members of the custom gene sets. Functional enrichment and differential gene, transcript, and protein expression/translation results were represented using heatmaps, bar plots, and line plots created with ggplot2 in R [51].

3.8. Gene-Level Productive versus Unproductive log2FoldChange Plot

We built a gene-level productive versus unproductive log2FoldChange plot for all genes in the custom gene set. For each gene, productive or unproductive log2FoldChange values were calculated as the mean shrunk log2FoldChange value of all productive or unproductive isoforms of the gene. For example, the HLA-B gene has 8 annotated isoforms according to the GENCODE v33 reference transcriptome [43]: 3 protein-coding (productive) and 5 retained intron (unproductive). Thus, the productive fold change of the HLA-B gene was the mean log2FoldChange of its 3 productive isoforms, whereas the unproductive fold change of this gene corresponded to the mean log2FoldChange value of the 5 unproductive isoforms.

3.9. Interaction Networks

We used the igraph [52], tidygraph, and ggraph R packages to build and plot the unified virus versus host protein–protein network.