Omics-guided engineering of CHO cells reveals host targets to support AAV production Authors: Konstantina Tzimou, Guillem López-García, Meryem Irem Sipahi, Lars K. Nielsen, Jesús Lavado-García Published: 2026 ABSTRACT The scalable production of recombinant adeno-associated viruses (rAAVs) remains a critical challenge in gene therapy manufacturing. While HEK293 and insect cell systems dominate current production platforms, each presents significant limitations in cost, scalability, and product quality. Chinese hamster ovary (CHO) cells represent an attractive alternative given their established use in biopharmaceutical manufacturing, yet their inability to efficiently support rAAV assembly has hindered development of CHO-based rAAV bioprocesses. Here, we investigate the molecular basis of this limitation by expressing individual AAV and adenoviral helper proteins in CHO cells and characterizing their effects using integrated transcriptomic and proteomic analyses. Functional enrichment analyses revealed that mitochondrial biogenesis was broadly suppressed across conditions, accompanied by downregulation of antiviral nuclear regulation, signaling networks, including the JAK-STATand interferon-stimulated gene pathways. Functional evaluation of host proteins identified from the omics analyses showed that overexpression of nuclear components like Mx2 and Rad23A supported rAAV production, reaching a yield of 105 vg/mL. Together, these results provide a comprehensive map of CHO cell responses to AAV and adenoviral helper elements, identify candidate host pathways influencing rAAV biogenesis, and highlight both the promise and the remaining barriers toward establishing CHO as a viable platform for large-scale rAAV manufacturing. Introduction As gene therapies advance toward broader clinical application, the manufacturing of recombinant adeno-associated viruses (rAAVs) has become a central challenge in translating laboratory discoveries into scalable treatments. The choice of production platform not only influences yield and cost but also affects critical quality attributes such as vector potency and purity.1 Currently, rAAVs are primarily manufactured via two main strategies: triple transfection in HEK293 cells and baculovirus expression vector systems (BEVS) in Spodoptera frugiperda (Sf9) insect cells. While triple transfection allows relatively straightforward plasmid-based production of rAAVs, it suffers from scalability limitations2 and high costs due to good manufacturing practice (GMP)-grade plasmids and specialized transfection reagents.3,4,5,6,7,8 On the other hand, the BEVS-Sf9 system offers better scalability but results in high levels of empty capsids and non-human post-translational modifications (PTMs), which can impact vector potency and immunogenicity.9,10 Thus, the exploration for alternative platforms for the production of rAAVs is imperative.8,9 Chinese hamster ovary (CHO) cells are an attractive candidate due to their regulatory acceptance, adaptability to large-scale bioreactor production, and world-wide established use in monoclonal antibody and therapeutic protein production. CHO-derived systems benefit from robust expression, well-understood glycosylation patterns, and compatibility with serum-free, chemically defined media.11,12,13 However, unlike HEK293 cells, CHO cells do not endogenously express the adenoviral (AdV) E1A/E1B helper genes, which are considered essential for AAV production. To date, only a limited number of studies have explored the feasibility of rAAV production in CHO cells, and so far this platform has shown poor rAAV productivity, with no commercially viable CHO-based rAAV manufacturing system established to date. In a study by Nagy et al.,14 CHO cells were engineered to express the herpesvirus entry mediator (HVEM) or Nectin-1 receptors to permit infection by recombinant herpes simplex virus type 1 (rHSV-1) vectors providing the necessary AAV rep, cap, and gene of interest (GOI) elements. While this approach enabled the production of multiple rAAV serotypes with a high percentage of full capsids, overall vector yields remained low. Furthermore, the engineered CHO cells displayed reduced viability and viable cell density during production, potentially compromising product quality and scalability of the process. The use of herpes simplex virus (HSV)-based helper systems also raises additional concerns regarding biosafety and regulatory compliance in a manufacturing context. In a different study, Cao et al.15 attempted rAAV production in CHO cells via transient plasmid transfection, supplementing the standard triple-transfection protocol with a fourth plasmid encoding AdV E1A/E1B helper genes that are absent in CHO. This quadruple transfection approach resulted in undetectable viral output and was associated with markedly low expression of cap transcripts, which encode the structural VP proteins of the AAV capsid. To address this, the authors introduced a fifth plasmid expressing the cap gene under the control of a cytomegalovirus (CMV) promoter, which increased cap transcript levels. However, viral titers remained low, and capsid protein expression was still low, highlighting translational inefficiencies and indicating that transcriptional rescue alone is insufficient to support productive rAAV manufacturing in CHO cells. The reasons why CHO cells are unable to efficiently produce rAAVs remain unclear, and thus, a deeper understanding of viral protein-CHO interactions is necessary. The aim of this study is to characterize how CHO cells respond to elements used for rAAV production and to evaluate whether targeted modulation of host factors can enable rAAV production in CHO cells. To address this, we individually expressed AAV and AdV helper proteins in CHO cells to systematically study their effects. By isolating each viral factor, we aim to pinpoint which components elicit adverse cellular responses or are insufficiently supported by the host machinery. To capture the cellular impact of each viral element, we conduct comprehensive transcriptomic and proteomic profiling. This dual-omics approach allows us to observe complementary information on gene expression via transcripts and phenotype modulation, stress pathways, antiviral responses, and apoptotic markers. Through this systematic analysis, we seek to identify host pathways that are activated or suppressed in response to each component and uncover molecular bottlenecks that may hinder rAAV biogenesis in CHO cells. Results Overall protein-specific effects on viability, transcriptome and proteome of CHO-S cells The overall experimental workflow for the omics analyses is depicted in Figure 1A. In brief, a total of 16 plasmids were cloned, 15 encoding a codon-optimized individual AAV or AdV helper open reading frame (ORF) (Rep78, Rep68, Rep52, Rep40, VP1, VP2, VP3, membrane-associated accessory protein [MAAP], AAP, VA RNA I, VA RNA II, E1A, E1B, E2A, E4orf6) under the control of a strong synthetic promoter,16 together with a mock plasmid used as a negative control (NC). These plasmids were individually transfected into CHO-S cultures, followed by monitoring of cell viability and collection of samples at 48 h post transfection (hpt) for transcriptomic and proteomic analyses.Figure 1Experimental workflow and cellular responses to AAV and helper proteins(A) Schematic overview of the experimental workflow. Open reading frames for Rep78, Rep68, Rep52, Rep40, VP1, VP2, VP3, MAAP, AAP, VA RNA I, VA RNA II, E1A, E1B, E2A, E4orf6 were isolated, codon-optimized and assembled into a 15-plasmid expression library together with a mock plasmid (same backbone but no gene of interest) used as a negative control. CHO cells were transfected with single plasmids, one construct per culture. Cells were harvested at 48 hpt for omics profiling, including proteomics and transcriptomic analyses, followed by computational data analysis. NC denotes the negative control (mock-transfected) condition. (B) Time course cell viability for each condition (including NC) measured at 24, 48, and 72 hpt. Data are shown as mean ± SD, (N = 8). Experimental workflow and cellular responses to AAV and helper proteins (A) Schematic overview of the experimental workflow. Open reading frames for Rep78, Rep68, Rep52, Rep40, VP1, VP2, VP3, MAAP, AAP, VA RNA I, VA RNA II, E1A, E1B, E2A, E4orf6 were isolated, codon-optimized and assembled into a 15-plasmid expression library together with a mock plasmid (same backbone but no gene of interest) used as a negative control. CHO cells were transfected with single plasmids, one construct per culture. Cells were harvested at 48 hpt for omics profiling, including proteomics and transcriptomic analyses, followed by computational data analysis. NC denotes the negative control (mock-transfected) condition. (B) Time course cell viability for each condition (including NC) measured at 24, 48, and 72 hpt. Data are shown as mean ± SD, (N = 8). Cell viability was assessed up to 72 hpt (Figure 1B). Over time, differences emerged depending on the construct (Figure 1B). Rep78 and E1A expression caused the strongest cytotoxic effects, reducing viability to around 60% by 72 hpt. In contrast, Rep52 and Rep40 maintained high viability levels, while most other constructs, including VP proteins, AAP, and MAAP, showed intermediate effects, underscoring the divergent impact of individual factors on host survival. For transcriptomic profiling, RNA was harvested at 48 hpt, sequenced, and analyzed (Figure 2A). The analysis compared the transcriptional landscape of cells expressing each viral or helper element to that of mock-transfected controls, revealing substantial variation in transcriptional responses across constructs. The largest numbers of differentially expressed (DE) transcripts were observed for MAAP and VP2, which both triggered strong host transcriptional changes with 2331 and 2789 DE transcripts identified, respectively. In contrast, Rep68, E1B, and E2A induced very few DE transcripts, with identified DE transcripts being as low as 60 for Rep68, 37 for E1B and 251 for E2A, indicating that these proteins have low detectable impact on host gene expression under the tested conditions. Proteomic analysis was performed in parallel on 48 hpt samples using tandem mass tag (TMT) labeling and liquid chromatography-tandem mass spectrometry (LC-MS/MS). In total, 9,569 proteins were identified, of which 6,913 were supported by at least two unique peptides, and the corresponding transcripts for 56% of them were identified in our transcriptomics data (Figure 2B). The expression of AAV and helper proteins was supported by peptide identifications meeting standardized variables (Zq) and false discovery rate (FDR) criteria, with Rep and VP proteins interpreted at the family level due to peptide sequence overlap.Figure 2Summary of omics results and nuclear responses of CHO cells to AAV and helper proteins(A) Numbers of differentially expressed (DE) transcripts per condition, colored by levels of regulation in blue (Log2FC < −0.5), gray (−0.5≤Log2FC ≤ 0.5), and red (Log2FC > 0.5). Transcripts in gray are considered as “no change.” (B) Venn diagram showing the overlap between the identified targets in the transcriptomic and proteomic analysis. The total number of transcripts and proteins is based on all tested conditions. (C) Summary of DAVID enrichment analysis showing up and downregulated biological terms across all conditions. Blocks are enriched terms and color-coded by functional category, with pink blocks denoting antiviral-related terms. (D) Principal-component analysis (PCA) of proteins associated with the GO term nucleus (GO:0005634). Summary of omics results and nuclear responses of CHO cells to AAV and helper proteins (A) Numbers of differentially expressed (DE) transcripts per condition, colored by levels of regulation in blue (Log2FC < −0.5), gray (−0.5≤Log2FC ≤ 0.5), and red (Log2FC > 0.5). Transcripts in gray are considered as “no change.” (B) Venn diagram showing the overlap between the identified targets in the transcriptomic and proteomic analysis. The total number of transcripts and proteins is based on all tested conditions. (C) Summary of DAVID enrichment analysis showing up and downregulated biological terms across all conditions. Blocks are enriched terms and color-coded by functional category, with pink blocks denoting antiviral-related terms. (D) Principal-component analysis (PCA) of proteins associated with the GO term nucleus (GO:0005634). To determine which cellular functions were most affected, we complemented this analysis with proteomic enrichment profiling (Figure 2C). The proteomic data revealed that the main processes altered across conditions were associated with nuclear functions and gene expression regulation, cell cycle control and apoptosis, mitochondrial and metabolic pathways, and antiviral or immune responses. Together, the transcriptomic magnitude and proteomic enrichment patterns provide a coherent overview of the host alterations induced by AAV and helper protein expression. These observations motivated focused analyses of nuclear impacts and apoptotic regulation, mitochondrial modulation, and suppression of antiviral pathways in the following sections. Nuclear impacts and apoptotic regulation Analysis of the proteomics enrichment results across all conditions revealed that many significantly enriched terms were related to nuclear functions, including transcriptional regulation, DNA replication, and chromatin organization (Figure 2C). This observation indicates that the studied proteins have a strong impact on nuclear processes, motivating a focused analysis of this subset. To understand whether these proteins influence nuclear pathways in similar or distinct ways, we examined patterns in the abundance of nucleus-associated proteins. For this purpose, principal-component analysis (PCA) was conducted on the subset of proteins associated with the Gene Ontology (GO) term “nucleus” (GO:0005634) (Figure 2D). A PCA analysis revealed the emergence of three clusters, reflecting similarities and differences in the effects of the studied proteins under the various conditions. The first cluster included VP1, VP2, VP3, E1A, and Rep78, which group closely together, suggesting a shared or highly similar impact on nuclear processes. The second cluster was composed of E4orf6, E2a, VAI, VAII, and E1b, indicating comparable functional roles or effects distinct from the first cluster. Interestingly, E1b was positioned further from the other members of cluster 2 in the PCA analysis, indicating a distinct effect on the nuclear proteome compared to the rest of the helper proteins. Still, it was assigned to cluster 2 due to its function as helper protein and the fact that, together with E4orf6, forms a known functional complex. The third cluster, containing Rep40, MAAP, Rep68, Rep52, and AAP, separated clearly from the other two clusters, highlighting a distinct pattern of influence on the nuclear proteome. Interestingly, the proteins in the first two clusters have been described as essential for the efficient production of rAAVs in HEK293 cells in previous literature. In contrast, the third cluster consisted of proteins that have not been described as essential for rAAV production in HEK 293. This separation may reflect differences in their nuclear interactions or functions that are either supportive but non-essential, or involved in other cellular processes unrelated to rAAV assembly. To further investigate the functional differences between the protein clusters identified in the PCA analysis, we performed a comparative transcriptomic analysis, examining the differential expression of transcripts (−0.5 0.5) associated with the GO term “nucleus” (GO:0005634). DE transcripts were identified for each group, and the enriched terms were visualized to highlight differences in molecular functions, biological processes, and cellular pathways (Figure 3).Figure 3Enrichment analysis of differentially expressed transcripts (−0.5 0.5) linked to nuclear processes in the three clusters previously observedBubble plots display enriched molecular functions (MF), biological processes (BP), and KEGG pathways, where color indicates category and bubble size reflects significance (–log10 p value). Enrichment analysis of differentially expressed transcripts (−0.5 0.5) linked to nuclear processes in the three clusters previously observed Bubble plots display enriched molecular functions (MF), biological processes (BP), and KEGG pathways, where color indicates category and bubble size reflects significance (–log10 p value). Cluster 1, including Rep78, VP1, VP2, VP3, and E1A, showed the highest number of DE transcripts associated with the nucleus, reflecting a broad transcriptional response. Enrichment analysis revealed strong representation of biological processes linked to transcription regulation, chromatin remodeling, and DNA damage response. In addition, unique to this cluster were multiple histone modification-related activities, including histone acetyltransferase (e.g., H3K9ac, H3K27ac, H4K16ac) and histone kinase activities. These terms indicate that proteins from cluster 1 may directly modulate chromatin structure and epigenetic state. Cluster 2, including E1B, E2A, E4orf6, VAI, and VAII, displayed a comparable functional profile to cluster 1, with enrichment in transcriptional regulation, DNA binding, and RNA polymerase II-associated terms, indicating shared engagement in gene expression control. However, it lacked the histone modification-specific terms characteristic of cluster 1, implying that these elements may act as transcriptional regulators or cofactors rather than chromatin modifiers. Last, cluster 3, including Rep40, Rep52, Rep68 AAP, and MAAP, displayed a distinct but partially overlapping enrichment profile compared with clusters 1 and 2. Similar to the other two, it showed enrichment for DNA binding activities and regulation of transcription, indicating the involvement of transcriptional control mechanisms. However, it was uniquely characterized by terms related to the regulation of apoptotic processes and DNA damage response, suggesting that these proteins may participate in stress-induced nuclear signaling. In contrast, cluster 3 lacked enrichment for RNA splicing and RNA processing-related terms, as well as histone modification activities, prominent in cluster 1. These differences indicated that, rather than influencing chromatin remodeling or RNA metabolism, cluster 3 proteins were more closely associated with transcriptional regulation coupled to cellular stress and apoptosis pathways. Nuclear events such as transcriptional activation, chromatin remodeling, and DNA damage response often interface with broader cellular pathways, including those governing cell fate decisions like apoptosis.17,18,19 To explore this dimension, we analyzed proteins involved in pathways central to genome integrity and cell fate determination, identifying those present in our proteomics dataset and mapping their regulation across conditions expressing different viral or helper proteins. The resulting network offers an integrated view of the studied proteins influencing regulation of apoptosis, DNA damage repair, and cell division pathways during expression of viral elements (Table S1). Given evidence that AAV production engages and is influenced by apoptotic pathways in the HEK293 system,20,21,22 we examined how viral and helper protein expression modulates apoptotic signaling in CHO, revealing coordinated changes in the levels of core and regulatory proteins within apoptotic pathways (Figure 4). We were able to identify downregulation in more than one condition of major apoptotic proteins such as p53 and Bax, as well as downregulation of regulators that promote apoptosis (Fas, Notch2, Trim32), but also apoptotic inhibitors (Xiap, Golph3, Mtdh). A number of pro-apoptotic proteins was also found upregulated (highlighted in red) in more than 1 conditions (Melk, Mmp9, Ripk1, Plk1, Ccar1, Rasa1) as well as anti-apoptotic proteins (Erbb2, Myo18α). These observations pointed to a potential activation of early or regulatory stages of apoptosis rather than full execution of the pathway, suggesting that CHO cells may partially engage apoptotic signaling as an adaptive response to viral and helper protein expression. Interestingly, several proteins within this pathway exhibited differential regulation depending on the specific viral element expressed. For example, caspase-3 was upregulated upon expression of VP1, VP2, VP3, and E1A, but downregulated in response to E4orf6, VAI, and VAII. These findings underscore the complexity and context-dependency of the apoptotic responses triggered by different viral components.Figure 4Apoptotic pathways impacted by AAV and helper protein expressionThe schematic illustrates extrinsic and intrinsic apoptotic signaling cascades, including Fas/TNF receptor signaling, mitochondrial apoptosis regulation, and downstream caspase activation. Proteins from these pathways identified in our dataset are color-coded: downregulated proteins are shown in blue (Log2FC < −1), upregulated proteins in red (Log2FC > 1), and proteins with mixed regulation across conditions (up in some, down in others) are highlighted in purple. Proteins shown in white boxes represent pathway components that were either not detected in our dataset or did not exhibit significant up or downregulation under any of the tested conditions. Linked white boxes list the AAV or helper protein conditions under which each protein was found to be down- or up-regulated. Apoptotic pathways impacted by AAV and helper protein expression The schematic illustrates extrinsic and intrinsic apoptotic signaling cascades, including Fas/TNF receptor signaling, mitochondrial apoptosis regulation, and downstream caspase activation. Proteins from these pathways identified in our dataset are color-coded: downregulated proteins are shown in blue (Log2FC < −1), upregulated proteins in red (Log2FC > 1), and proteins with mixed regulation across conditions (up in some, down in others) are highlighted in purple. Proteins shown in white boxes represent pathway components that were either not detected in our dataset or did not exhibit significant up or downregulation under any of the tested conditions. Linked white boxes list the AAV or helper protein conditions under which each protein was found to be down- or up-regulated. Modulation of mitochondrial and metabolic pathways To investigate how viral and helper protein expression influences mitochondrial function, we examined the subset of proteins associated with mitochondrial pathways across all conditions. In total, 434 proteins were identified as being either upregulated (Log2FC > 1) or downregulated (Log2FC < −1). Heatmap analysis revealed a high degree of similarity in the expression profiles of mitochondrial proteins across the samples. Notably, while most of these proteins exhibited consistent expression patterns, Rep40 and MAAP displayed distinct profiles, indicating a divergence from the general trend (Figures 5A and 5B).Figure 5Modulation of mitochondrial and metabolic pathways by viral proteins(A) Heatmap of all identified mitochondrial proteins. (B) Subset of proteins showing opposite regulation in Rep40/MAAP versus the remaining conditions. (C) Enrichment analysis of mitochondrial and metabolism related pathways affected by viral proteins. Bubble color indicates type of regulation (blue for downregulation red for upregulation), bubble size reflects statistical significance (−log10(Benjamini)), and bubble position denotes affected pathways. Modulation of mitochondrial and metabolic pathways by viral proteins (A) Heatmap of all identified mitochondrial proteins. (B) Subset of proteins showing opposite regulation in Rep40/MAAP versus the remaining conditions. (C) Enrichment analysis of mitochondrial and metabolism related pathways affected by viral proteins. Bubble color indicates type of regulation (blue for downregulation red for upregulation), bubble size reflects statistical significance (−log10(Benjamini)), and bubble position denotes affected pathways. We examined genes that were upregulated in MAAP and Rep40 but downregulated in the remaining samples, as well as those showing the opposite pattern. This analysis revealed that the associated proteins were involved in a range of cellular processes (Figure 5B). While a large subset was associated with mitochondrial gene expression, others were implicated in key metabolic pathways, including amino acid metabolism, malate metabolism, glycolysis, the tricarboxylic acid (TCA) cycle, lipid metabolism, and cofactor biosynthesis. Moreover, differences extended to pathways related to stress response, regulation of apoptosis, antiviral activity, and mitochondrial structure-such as cristae formation and organelle biogenesis. Although the effect on mitochondrial function appeared to be a common outcome of viral protein expression, MAAP and Rep40 may mediate this effect through mechanistically distinct pathways, potentially reflecting different roles in host-cell modulation or viral replication strategies. Next to study how viral and helper protein expression affects cellular metabolism, we analyzed pathways related to mitochondrial activity and energy production. The enrichment analysis, summarized in Figure 5C, revealed a consistent pattern of downregulation across mitochondrial and core metabolic pathways in response to the expression of the studied viral proteins. Among the cellular processes analyzed, mitochondrial functions and metabolic pathways essential for energy production appeared particularly susceptible to viral interference, with pathways such as oxidative phosphorylation, the TCA cycle, glycolysis, and lipid metabolism being suppressed, indicating a general reduction in the cell’s energetic capacity. Among the proteins analyzed, E1B produced the most extensive effect, significantly downregulating eight distinct metabolic and mitochondrial terms, followed by E2A, E4orf6, and Rep78, which also affected multiple energy-related pathways. Structural components such as VP1, VP2, and VP3 and AAP showed weaker but still detectable effects on oxidative phosphorylation and carbohydrate metabolism. Together with E2A and E4orf6, E1B appeared to strongly repress terms related to mitochondrial structure and function. Among all biological terms, carbon metabolism and amino sugar and nucleotide sugar metabolism were the most frequently affected, both showing significant downregulation across different viral proteins. These pathways are vital for cellular biosynthesis and energy generation, and their consistent suppression across multiple conditions points to an impairment to the host metabolic activity. Despite the general trend toward suppression, a small subset of terms exhibited upregulation, with the mitochondrial inner membrane being the most notable example. This structure plays a dual role, not only serving as the site for electron transport and ATP synthesis via oxidative phosphorylation, but also participating in cristae architecture, protein import, and metabolite transport.23,24 The upregulation of this compartment, particularly in the context of rAAV production in CHO cells, may reflect viral manipulation to maintain or remodel mitochondrial structure for their benefit, mainly driven by AAV proteins Rep68 and Rep40, and helper E2A, E4orf6, VAI, and VAII. Suppression of antiviral responses To examine how viral and helper protein expression influences the host defense machinery, we analyzed enriched biological processes associated with antiviral and immune responses. As shown in Figure 2C, numerous terms related to antiviral activity were identified across nearly all conditions, yet they were predominantly downregulated. This consistent repression across different viral proteins indicates a broad and systematic suppression of host antiviral mechanisms, suggesting a coordinated viral strategy to dampen innate immune signaling and evade antiviral detection. To better understand how the observed suppression of antiviral pathways manifests at the protein level, we compared the abundance of antiviral-related proteins in each condition to the mock-transfected control (Figure 6). By mapping genes to their associated GO terms, we sought to identify patterns of coordinated suppression that may contribute to the inability of CHO cells to support effective AAV production. Several proteins were found to be downregulated across multiple terms, suggesting their central role in antiviral defense pathways. The enriched downregulated terms include “defense response to virus,” “antiviral innate immune response,” “positive regulation of defense response to virus by host,” “cell surface signaling pathway via JAK-STAT,” “virus receptor activity,” “regulation by host of viral processes,” “IRES-dependent viral translational initiation,” and “positive regulation of viral genome replication.” These terms encompass key facets of host-virus interaction and immune signaling. Notably, core components involved in defense response to virus, such as OAS1, DDX56, Mx2, IFIT1, IFIT2, ZNF1, and ZAP, and positive regulation of antiviral defense by the host, such as STAT proteins, Hsp90a and TOM70 exhibited consistent downregulation across multiple viral elements, indicating a widespread suppression of antiviral responses.Figure 6Regulation of antiviral response pathwaysHeatmap of proteins downregulated in at least one condition and associated with antiviral-related GO terms. Regulation of antiviral response pathways Heatmap of proteins downregulated in at least one condition and associated with antiviral-related GO terms. The JAK-STAT signaling pathway, a critical mediator of antiviral responses, was notably suppressed upon expression of AAV and AdV elements. Key components including JAK1, STAT1, STAT2, and STAT3 were all significantly downregulated upon expression of the majority of the studied viral elements (Figures 6 and 7). This pathway is essential for activating a wide range of interferon-stimulated genes that execute antiviral functions. Consistent with this, we also observed downregulation of several critical antiviral proteins, including Mx1, Mx2, OASL, ZAP (ZC3HAV1), ADAR, BST2 (tetherin), and ISG15 (Figures 6 and 7). These proteins collectively restrict viral replication, degrade viral RNA, and inhibit viral release from infected cells. The combined downregulation of both the signaling pathway and its antiviral targets suggests that the host’s antiviral defense might become disrupted.Figure 7Antiviral response pathways affected by viral and helper protein expressionThe schematic shows two major antiviral pathways: (i) IFNα/β and IFNγ-mediated JAK/STAT signaling leading to induction of antiviral effectors, and (ii) antigen processing and presentation via the MHC class I pathway. Proteins from these pathways that were found downregulated in our dataset are shown in blue (Log2FC < −1). Linked white boxes list the AAV or helper protein conditions under which each protein was found to be downregulated. Antiviral response pathways affected by viral and helper protein expression The schematic shows two major antiviral pathways: (i) IFNα/β and IFNγ-mediated JAK/STAT signaling leading to induction of antiviral effectors, and (ii) antigen processing and presentation via the MHC class I pathway. Proteins from these pathways that were found downregulated in our dataset are shown in blue (Log2FC < −1). Linked white boxes list the AAV or helper protein conditions under which each protein was found to be downregulated. JAK-STAT is not the only pathway found to be downregulated, but we also observed downregulation of antigen processing and presentation pathways. Specifically, key components such as Tapasin, Calreticulin (CRT), MHC class I molecules, and β2-microglobulin (β2M) were consistently downregulated across most of the studied conditions (Figure 7). Their reduced expression suggests a weakened ability of the host cells to present viral antigens that would trigger antiviral pathways. Evaluation of identified target proteins for rAAV production in CHO cells Since, according to the literature, the ability of CHO cells to support rAAV production with transfected codon-optimized genes has not been tested, we first transfected CHO-S wild-type (WT) cells with all the codon-optimized ORFs studied previously and the viral transgene plasmid (pGOI), carrying the egfp gene. To determine whether production of rAAVs can be supported in CHO cells by the same set of helper proteins described as essential for rAAV production in HEK293 cells, we next tested the performance of these minimal gene combinations. For this, we transfected CHO cells with the plasmid with the pGOI and the plasmids encoding the essential elements (Rep78, VP1, VP2, VP3, E1A, E1B, E2A, E4orf6, and VAI, as defined in the literature).1 Moreover, based on the proteomic analysis, several host factors were downregulated upon AAV and helper protein expression. Among these, Rad23A and Mx2 stood out as two of the most strongly downregulated antiviral proteins (Figure 5), while Mmp9 was also upregulated under multiple conditions, upon transfection of VP1, VP2, VP3, E1B, E4orf6, VAI, and VAII (Table S1). To explore their potential roles in rAAV production, Rad23A and Mx2 were transiently overexpressed individually or in combination, and Mmp9 was knocked out through cell engineering (Figure S2) to generate an Mmp9−/− knockout line. Each of these modulations was tested under both setups “all” and “essential,” resulting in the conditions all + Rad23A, all + Mx2, all + Rad23A + Mx2, and the corresponding essential + Rad23A, essential + Mx2, and essential + Rad23A + Mx2, as well as their equivalents in the Mmp9−/− background. All conditions were transfected with the same total amount of DNA, ensuring comparable experimental conditions (Figure 8).Figure 8rAAV titers in CHO-S WT and Mmp9−/− cells under different transfection conditions including identified targetsTiters are expressed as viral genomes per mL (vg/mL). For AAV vector production, “All” genes include Rep78, Rep68, Rep52, Rep40, VP1, VP2, VP3, MAAP, AAP, E1A, E1B, E2A, E4orf6, VAI, and VAII. “Essential” genes include Rep78, VP1, VP2, VP3, E1A, E1B, E2A, E4orf6, and VAI. (A) CHO-S WT transfected with all AAV and helper genes (All) or with all plus Rad23a (+R), Mx2 (+M), or both (+RM), N = 18. (B) CHO-S WT transfected with only essential AAV and helper genes (Essential) or with essential plus Rad23a (+R), Mx2 (+M), or both (+RM), N = 18. (C) CHO-S Mmp9−/− transfected with all AAV and helper genes (All) or with all plus Rad23a (+R), Mx2 (+M), or both (+RM), N = 18. (D) CHO-S Mmp9−/− transfected with only essential AAV and helper genes (Essential) or with essential plus Rad23a (+R), Mx2 (+M), or both (+RM), N = 18. (E) Individual biological replicates (N = 3 technical replicates per biological replicate) for the WT All+RM condition, illustrating variability across biological replicates. Statistical significance in (A) is indicated (∗p < 0.05, ∗∗∗∗p < 0.0001). rAAV titers in CHO-S WT and Mmp9−/− cells under different transfection conditions including identified targets Titers are expressed as viral genomes per mL (vg/mL). For AAV vector production, “All” genes include Rep78, Rep68, Rep52, Rep40, VP1, VP2, VP3, MAAP, AAP, E1A, E1B, E2A, E4orf6, VAI, and VAII. “Essential” genes include Rep78, VP1, VP2, VP3, E1A, E1B, E2A, E4orf6, and VAI. (A) CHO-S WT transfected with all AAV and helper genes (All) or with all plus Rad23a (+R), Mx2 (+M), or both (+RM), N = 18. (B) CHO-S WT transfected with only essential AAV and helper genes (Essential) or with essential plus Rad23a (+R), Mx2 (+M), or both (+RM), N = 18. (C) CHO-S Mmp9−/− transfected with all AAV and helper genes (All) or with all plus Rad23a (+R), Mx2 (+M), or both (+RM), N = 18. (D) CHO-S Mmp9−/− transfected with only essential AAV and helper genes (Essential) or with essential plus Rad23a (+R), Mx2 (+M), or both (+RM), N = 18. (E) Individual biological replicates (N = 3 technical replicates per biological replicate) for the WT All+RM condition, illustrating variability across biological replicates. Statistical significance in (A) is indicated (∗p < 0.05, ∗∗∗∗p < 0.0001). In WT CHO-S cells, transfection with all AAV and helper codon-optimized ORFs supported rAAV production, and this was improved by the transient overexpression of either Rad23A (+R) or Mx2 (+M) (Figure 7A). Combining both Rad23A and Mx2 (+RM) resulted in a 16-fold increase in vector yield compared to transfecting with all plasmids alone, reaching a mean value of around 105 vg/mL of culture (Figure 8A). When only the subset of genes described as essential, according to the literature, in HEK293 cells was provided, rAAV production in CHO was negligible, and no clear benefit of adding Rad23A or Mx2 individually or in combination was observed (Figure 8B). In the case of CHO Mmp9−/−, transfection with all helper genes again supported only low levels of rAAV production, and the addition of Rad23A and/or Mx2 did not enhance yields (Figure 8C). Similarly, transfection with only the essential helper subset yielded very low titers, with no significant effect of Rad23A or Mx2 addition (Figure 8D). To identify the source of variability observed in rAAV yield measurements, we examined the consistency of biological and technical replicates. We found that the variation originated primarily from biological replicates rather than from technical assay noise (Figure 8E). Technical replicates within each biological sample were highly consistent, showing minimal intra-replicate deviation, whereas biological replicates varied substantially. This indicates that the variability reported for the pooled data in the previous analysis reflects inherent biological differences between transfections rather than measurement inconsistencies. Discussion CHO cellular responses upon expression of AAV and helper proteins The number of DE transcripts varied markedly between conditions, reflecting intrinsic differences in the biological role and cellular impact of the individual viral proteins. Regulatory proteins such as Rep78/Rep68, E1B, and E2A primarily modulate specific host pathways involved in transcriptional control, replication, and stress responses, leading to a more targeted transcriptional footprint. In contrast, structural and assembly associated proteins such as MAAP and VP2 are expressed at higher levels and engage broader cellular processes, resulting in more extensive transcriptomic remodeling. All comparisons were performed using identical statistical criteria, and only transcripts meeting the same significance threshold (q value <0.05) were considered, ensuring that differences in DEG counts reflect genuine biological variation in host response. Our findings indicate that the majority of the identified proteins trigger and influence nuclear processes, underscoring the importance of the host nucleus in rAAV production. Despite sharing this general involvement, three different clusters of AAV and helper proteins appear to interact with the nuclear environment through distinct molecular mechanisms. This functional divergence suggests that different subsets of proteins may support rAAV replication and assembly by modulating complementary aspects of nuclear activity, such as transcriptional regulation, chromatin accessibility, or stress response signaling, reflecting a coordinated but multifaceted engagement of host nuclear pathways during viral production. Although the PCA revealed three clearly separated clusters, it is important to consider potential confounding factors that could contribute to this segregation. One obvious candidate is cell viability. However, this is unlikely to be the primary driver of the observed clustering. Conditions expressing Rep78, E1A, and the VP proteins exhibit markedly different viability profiles (Figure 1), yet they consistently group within the same cluster, indicating that viability alone cannot explain the separation. Similarly, differences in the overall number of quantified changing proteins in the proteomic analysis is unlikely to account for the clustering, as the number of significantly up- and down-regulated proteins is comparable across conditions (Tables S1 and S2). While we cannot exclude the contribution of additional, currently unidentified factors, the convergence of constructs with distinct viability and expression burden into the same cluster strongly supports nuclear activity as the dominant axis underlying the observed separation. This interpretation is consistent with the known roles of Rep, E1A, and capsid-associated processes in modulating nuclear function, transcriptional regulation, and DNA replication, which collectively emerge as the most prominent features captured by the PCA. Among these proteins, the capsid VP1, VP2, and VP3 proteins have traditionally been characterized only for their structural roles in virion assembly. However, because AAV assembly occurs in the nucleus, these proteins must be actively imported from the cytoplasm, where they are synthesized, into the nuclear compartment to form complete viral particles.25,26 This nuclear localization aligns with our findings showing that many of the identified proteins are associated with nuclear functions and transcriptional regulation. It also raises the possibility that VP proteins, once inside the nucleus, could interact with host nuclear components and influence processes such as transcriptional activity or chromatin organization. While prior studies have implicated VP1 and VP2 in similar gene expression functions, our data provide evidence that VP3 exhibits similar behavior.27 Such interactions could represent an additional layer of coordination between viral structural elements and the host nuclear machinery, contributing to efficient rAAV replication and genome packaging. The consistent downregulation of mitochondrial and core metabolic pathways in CHO cells upon expression of AAV and AdV helper proteins aligns with a common observation in virology, which is that many viruses actively suppress mitochondrial function and biogenesis to reshape host metabolism, evade immune signaling, or divert resources. One example is measles virus, which has been shown to suppress mitochondrial biogenesis and induce mitochondrial hyperfusion, leading to release of mitochondrial DNA that triggers host innate sensing as part of an antiviral response.28,29 More broadly, viruses including herpesviruses, respiratory viruses, and coronaviruses are known to interfere with mitochondrial dynamics, bioenergetics, and mitochondrial antiviral signaling proteins and pathways to weaken host defenses and reprogram their metabolism.30,31,32,33 Transcriptomic analyses of rAAV-producing HEK293 cells have reported downregulation of genes associated with mitochondrial organization, oxidative phosphorylation, and energy metabolism, suggesting a broad suppression of mitochondrial activity during vector assembly and replication in this platform.34 In addition, AdV helper proteins have been shown to alter host cell metabolism, including the repression of mitochondrial biogenesis and respiratory function.35,36 Therefore, while downregulation of mitochondrial pathways may impose metabolic stress and limit optimal cellular performance, it probably only reflects a general viral response rather than the specific cause of limited rAAV production in CHO cells. In our data, we observed coordinated downregulation of innate immune signaling, JAK-STAT-mediated interferon responses, and antigen presentation pathways across conditions, implying that antiviral responses are not the main bottleneck in rAAV production in CHO cells, as in that case we would expect strong activation and upregulation of these pathways. The suppression of key antiviral mediators, such as OAS1, MX proteins, IFIT family members, and STAT proteins, highlights a viral strategy to evade immune recognition and neutralization. These findings suggest that insufficient rAAV titers in CHO cannot be primarily attributed to strong intrinsic antiviral defenses, but rather point to alternative barriers. To date, there are only two studies reporting the use of CHO cells as a host for rAAV production, and both underscore the biological challenges inherent to this platform. According to Nagy et al., using engineered serum-free, suspension-adapted CHO cells to express HSV entry receptors, they demonstrated the production of multiple rAAV serotypes, albeit with profound impacts on cell viability after co-infection, consistent with host-vector incompatibilities and restriction phenomena in CHO cells.14 In a complementary study, Cao et al. showed that while CHO cells are capable of producing rAAV components following transient transfection, expression of capsid proteins remains a limiting factor, necessitating supplementary expression constructs to elevate rAAV yields.15 Our results build on these foundational observations by providing multi-omics evidence that nuclear regulatory processes, rather than cell viability or simple abundance of DE proteins, are a primary factor distinguishing CHO responses during rAAV component expression. Our results indicate that differences in CHO cell viability or global proteome disruption alone cannot account for the observed limitations in rAAV production. Instead, the ability to restore rAAV production by overexpressing specific nuclear proteins that were downregulated under non-productive conditions points to nuclear regulation as a key determinant of CHO cell permissiveness. In particular, nuclear functions such as DNA replication, genome stability, and nuclear organization emerge as critical bottlenecks, which are consistent with the rAAV life cycle, as both viral genome replication and capsid assembly occur in the nucleus. Codon optimization removes transcriptional constraints but is not the main production bottleneck We observed robust expression of the individual VP proteins under their respective single-element expression conditions, even in the absence of AAP. This efficient expression in CHO cells is likely facilitated by the use of codon-optimized ORF sequences tailored for CHO expression and designed to avoid splicing sites present in the native viral genome. As shown in Table S3, VP proteins were readily detected under their corresponding expression conditions, confirming successful protein production. However, these results should not be interpreted as evidence of efficient capsid assembly in the absence of AAP, as each AAV component was evaluated individually rather than in a combinatorial context. Instead, our findings demonstrate that the VP proteins can be individually expressed in CHO cells, providing a necessary foundation for subsequent studies investigating coordinated expression and assembly of functional capsids. Despite providing codon-optimized versions of both the AAV and helper genes, the resulting rAAV titers remained very low. Each construct was designed to express only the intended protein, with sequences optimized to improve translational efficiency and remove the need for endogenous splicing. For genes that normally produce multiple proteins through alternative splicing, such as Rep78, which is spliced to generate Rep68, the splicing signals were altered so that each transcript encoded a single product. These modifications were designed to bypass transcriptional and post-transcriptional barriers and ensure efficient production of individual proteins, which was confirmed by the detection of all AAV and helper proteins in their respective conditions in the proteomic dataset. However, even with these optimizations, vector yields remained low, suggesting that removal of RNA-level regulatory constraints does not resolve the main bottlenecks, which likely occur at later stages of rAAV assembly or maturation. Helper gene requirements differ between CHO and HEK293 production systems In HEK293 cells, the minimal set of genes required for vector production has been well defined. This essential set includes Rep78, VP1–3, and the AdV helper proteins E1A, E1B, E2A, E4ORF6, and VAI, as established through characterization of AdV and AAV gene functions in multiple studies. In contrast, much less is known about the helper requirements in CHO cells, where only a few studies have explored this system as a potential production platform. In contrast, much less is known about the requirements in CHO cells, where only a few published studies have begun to explore this cell line as potential production platform. Here, we showed that transfection of CHO cells with the essential genes identified in HEK293 fails to support rAAV production, in contrast to transfection with all AAV and helper genes, which produced measurable vector yields, suggesting that the requirements are not fully aligned between the two platforms. It is therefore possible that one or more of the genes considered non-essential in HEK293 may in fact be necessary in CHO, reflecting cell line-specific differences in host-virus interactions. The additional genes included Rep68, Rep52, Rep40, VAII, AAP, and MAAP, which may play supportive or compensatory roles in CHO cells that are redundant in HEK293. In HEK293 cells, the products of these genes assist in critical processes such as genome replication and genome encapsidation (Rep68, Rep52, Rep40), capsid assembly (AAP, MAAP), or RNA stability (VAII). Our findings suggest that, unlike in HEK293, one or more of these functions cannot be fully sustained in CHO without the corresponding AAV or/and helper proteins. This could indicate that HEK293 cells express host proteins capable of complementing or substituting for some of these viral functions, whereas CHO cells lack such endogenous factors. As a result, CHO cells may depend more directly on the complete viral toolkit to achieve efficient genome replication, RNA processing, and capsid assembly, reflecting fundamental differences in host-virus compatibility between the two production systems. Understanding biological replicate variability The substantial variability in rAAV titers across biological replicates, in contrast to the tight consistency among technical replicates, reveals an intrinsic source of heterogeneity in CHO based production rather than measurement error. Such variability within replicates might come from differences in transfection efficiency and cellular responsiveness. Indeed, CHO cells are often considered more challenging to transfect than HEK293 cells, having lower transfection efficiency and often requiring more optimization than HEK293.37,38 Moreover, in this study, cationic lipid-based transfection reagent is used for plasmid delivery. However, comparative studies measuring flow cytometric transfection rates report that polymeric reagents are able to achieve higher transfection efficiency than lipid-based formulations in mammalian systems,39 supporting that lipid-based reagents like FreeStyle MAX might be less robust in driving uniform uptake. These observations are in line with our prior experience, where we again observed variability between biological replicates when transfecting with lipid-based reagents. Already in 1995, it was demonstrated that lipid-DNA complexes are themselves heterogeneous and that cells differ in their ability to internalize, traffic, and express delivered DNA.40 When the transfection mix is added to the cell cultures, small differences in how the complexes interact with each culture can shift the proportion of productive versus non-productive cells, giving rise to variability between biological replicates that cannot be predicted or controlled. Thus, the observed deviations provide insight into the biological noise in transient expression systems, rather than being a technical artifact. In addition to transfection efficiency, the complexity of the multi-plasmid system used in this study likely contributes to the observed heterogeneity. Each production condition involves delivery of a large number of plasmids, up to 18 in total, encoding the various AAV and helper components. Achieving uniform uptake of all plasmids across a cell population is inherently difficult in transient transfection, where stochastic differences in complex formation and delivery can result in cells receiving only a subset of the required constructs. Consequently, only a fraction of cells may express the complete set of factors necessary for efficient rAAV assembly, which introduces additional variability in vector yield between biological replicates. Construction of plasmids encoding multiple genes, similar to the plasmid design used in the established triple transfection method for AAV production, could potentially reduce this variability by minimizing the number of plasmids required for co-delivery. Moreover, since plasmid DNA remains episomal in mammalian cells and is not replicated during cell division, it is diluted as the cells divide. The distribution of plasmids among cells during mitosis is stochastic and cannot be controlled, meaning that cells may gradually lose part of the delivered genetic material over time. In a system that relies on a large number of plasmids, this dilution effect increases the likelihood that cells will no longer carry the complete set of AAV and helper constructs, further reducing the proportion of productive cells and contributing to the observed variability. Functional roles of Mx2 and Rad23A in rAAV production In this study, Mx2 and Rad23A were selected as two of the most strongly downregulated host factors for targeted overexpression. Beyond their established cellular functions, it is worth considering how these proteins might influence rAAV production in CHO cells. Based on our observations, overexpression of Mx2 allowed higher vector yield, suggesting that Mx2, even though it is an antiviral protein, does not always act as a restriction factor. Although Mx2 is known to inhibit replication of several RNA and DNA viruses,41 studies have shown that it has no restrictive effect in replication of influenza virus.42 Moreover, it has been reported that, unlike human Mx2, murine Mx2 does not exhibit antiviral activity.43 Apart the antiviral activity of human Mx2, King et al. showed that it localizes to the cytoplasmic side of the nuclear envelope, modulating the efficiency and kinetics of nuclear import, and that its disruption can delay progression of the cell cycle.44 Even though this function has not been confirmed in nonhuman systems, it is possible that Mx2 plays a similar regulatory role in CHO cells, where its overexpression may enhance nuclear transport efficiency or cell cycle progression, thereby indirectly supporting rAAV production. Rad23A, in contrast, had only a modest effect when expressed alone, consistent with its known function as a ubiquitin-binding shuttle in the proteasomal degradation pathway. Rad23A functions as a ubiquitin-binding shuttle protein that delivers polyubiquitinated substrates to the proteasome and participates in nucleotide excision repair.45 However, it has been reported that its overexpression leads to an increase in production of infectious hepatitis C virus (HCV) in human cell lines. It has also been shown that Rad23A supports HIV-1 production by facilitating the interaction of the viral proteins with the host proteasome machinery, helping the virus regulate host protein degradation and create a more favorable environment for replication.46 Similar mechanisms could operate in CHO, where Rad23A overexpression may alleviate proteotoxic stress caused by accumulation of viral proteins, thereby facilitating efficient folding and assembly. While expression of Rad23A alone lead to only modest improvement in yield, its co-expression with Mx2 resulted in an additive increase, consistent with complementary functions in protein quality control and intracellular trafficking. This suggests that enhancing proteasomal processing capacity can indirectly support productive vector formation in CHO cells. Limitations of the study While this study provides a systematic framework for dissecting CHO cell responses to viral and helper elements, several aspects merit consideration when extending these findings. Because viral proteins can interact and form functional complexes during rAAV biosynthesis, future studies examining combinatorial expression may further clarify how such interactions influence host responses in CHO cells. In the present work, individual protein expression was prioritized to isolate protein-specific effects. During native rAAV replication, both the timing and relative expression levels of Rep and VP genes are tightly modulated through promoter usage, alternative splicing, and helper virus-dependent regulation. In contrast, the strong constitutive and individual expression of all elements used here alters this native expression balance and is therefore expected to bias the observed transcriptomic and proteomic profiles toward responses driven by elevated viral protein load, rather than by temporally coordinated replication and packaging processes. The constructs used in this study encode single, defined ORFs under synthetic promoters, eliminating native promoter usage and splicing. In addition, codon optimization is expected to disrupt potential splice sites, further reducing the likelihood of unintended isoform generation. While proteomic discrimination within Rep and VP families remains inherently limited due to sequence similarity, the design of the expression system supports the presence of a single intended protein species per condition. Moreover, differences in expression levels across transfected cultures and elements may contribute to variability in host responses, despite the use of a common promoter across constructs. In addition, analyses were performed at a single time point, capturing a defined cellular state but not potential dynamic or late-stage effects. Finally, transient transfection inherently introduces cellular heterogeneity, and when multiple plasmids are required, variable plasmid uptake may influence apparent productivity, complicating distinctions between missing viral components and incomplete transfection at the single-cell level. Future prospects: rAAV biomanufacturing potential of CHO cells In this study, we observed measurable production of rAAV in CHO cells, with titers reaching approximately 105 vg/mL. While this demonstrates that CHO cells can support rAAV production, the achieved levels remain far below those required for clinical manufacturing, where intensified HEK293 processes can reach titers around 1014 vg/mL in suspension systems. This difference highlights the gap that must be bridged before CHO cells can be considered a viable alternative platform for large-scale rAAV production. For CHO cells to outperform the HEK293 platform, several key requirements must be addressed. Vector yield must be increased by several orders of magnitude to approach the productivity of optimized HEK processes. This can be pursued through targeted cell engineering to enhance genome replication, capsid assembly, and vector packaging efficiency, as well as through bioprocess engineering strategies to find the optimal culture parameters. In addition, it is important to ensure scalability in controlled bioreactor environments, to translate improvements from small-scale experiments into robust and reproducible manufacturing processes. Product quality should be evaluated and improved across several dimensions, including genome integrity, which can be assessed through sequencing, and the levels of impurities and full to empty viral particle ratios, which depend strongly on downstream purification efficiency and can be optimized together with production yield through design of experiments (DoEs) approaches.47,48 Last, process simplicity can be addressed by reducing the plasmid burden through the design of multi-gene plasmid, as mentioned above, minimizing variability introduced by the co-delivery of numerous DNA species. While there is still significant room for improvement, the number of possible engineering and process innovations provides a strong foundation for future development and improvement of the robustness and scalability of CHO-based rAAV manufacturing. In our study, we generated two large-omics datasets that provide a strong foundation for discovery. Beyond identifying promising cellular targets to modulate rAAV production and quality, deeper mining and analyses can reveal insights into the behavior of viral proteins within CHO cells. These datasets also point to the substantial room for optimization at both process and vector design levels. We anticipate that data-guided strategies, followed by hypothesis-driven validation, offer the potential to translate these insights into measurable improvements in yield and product quality. Looking forward, the continued growth of gene therapy will require diversification of rAAV manufacturing platforms beyond the currently dominant mammalian systems. Insect cell-baculovirus expression systems, particularly those based on Sf9 cells and baculovirus vectors, have already demonstrated industrial success for rAAV production, offering scalability, serum-free suspension growth, and robust process intensification.1 However, additional platforms originally developed for viral particle or vaccine production merit systematic exploration for rAAV manufacturing. These include avian cell lines, such as AGE1.CR.pIX or EB66, which have been optimized for high-density growth, stable virus production, and regulatory compliance, and may provide advantages in scalability and process robustness.49,50,51 Other producer-cell-line-based systems and alternative hosts engineered for efficient capsid assembly and genome packaging could further expand the manufacturing toolbox. Exploring and benchmarking these platforms will be essential to identify systems that balance scalability, product quality, regulatory acceptance, and cost-effectiveness, ultimately enabling broader access to rAAV-based therapies. Conclusions This study provides a comprehensive characterization of CHO cell responses to individual AAV and AdV helper genes at the transcriptome and proteome levels, and evaluates selected host factors for their impact on rAAV production. Our analyses revealed that the effect of AAV and helper proteins can be grouped into distinct clusters, with each one having different effect in nuclear processes. We also observed consistent suppression of mitochondrial and antiviral pathways, while testing of Rad23A, Mx2, and Mmp9 demonstrated that host genes can modulate rAAV productivity. Nevertheless, even under optimized conditions, titers remained orders of magnitude below those achieved in current HEK293 or Sf9-based platforms. These findings underscore not only the potential of CHO as a future host for rAAV production but also highlight the need for further systematic research to establish CHO as a scalable, clinically relevant manufacturing platform with the omics datasets generated here providing a foundational resource to guide that effort. Materials and methods DNA sequences design and plasmid cloning AAV and AdV protein sequences were identified using the UniProt database. The corresponding amino acid sequences were reverse translated into DNA sequences using the Reverse Translation tool available at https://www.bioinformatics.org/sms2/rev_trans.html (Table S1). DNA sequences were subsequently codon-optimized for expression in CHO-S cells using the GenScript codon optimization tool available at https://www.genscript.com/gensmart-free-gene-codon-optimization.html (Table S2). Engineered sequences were analyzed in silico to predict translated products and to exclude unwanted ORFs, using standard bioinformatics tools (SnapGene and ExPASy Translate). Gene synthesis was carried out by Twist Bioscience (San Francisco, CA, USA). Each synthesized gene was cloned into the p100RPU plasmid backbone using NEBuilder HiFi DNA Assembly (New England Biolabs, Ipswich, MA, USA). The cloning reaction mixtures were transformed into Escherichia coli One Shot Mach1 competent cells (Life Technologies, Thermo Scientific, Waltham, MA, USA) following standard protocols. Transformed colonies were selected on LB agar plates containing 100 μg/mL ampicillin. Successful clones were confirmed by whole-plasmid sequencing (Plasmidsaurus, Louisville, KY, USA), and high-purity plasmid DNA was prepared using the NucleoBond Xtra Midi kit (Macherey-Nagel, Düren, Germany) according to the manufacturer’s instructions. Cell culture CHO-S cells from Life Technologies were cultivated and expanded in disposable polycarbonate 125 mL shake flasks equipped with vented caps (Corning Life Sciences, Tewksbury, MA, USA) in CD CHO media supplemented with 8 mM GlutaMAX (Gibco, Life Technologies, Thermo Fisher Scientific, USA) and 0.1% Pluronic F-68 Non-ionic Surfactant (Gibco, Life Technologies). Cultures were maintained at 37°C and 5% CO2 in a Hera Cell 150 incubator (Thermo Fisher Scientific, USA) with agitation at 130 rpm provided by Celltron (Infors HT, Switzerland). Transfection conditions For the individual transfections, 3 mL cell cultures were seeded in 6-well plates and transfected using FreeStyle MAX reagent together with OptiPRO SFM medium (Life Technologies, Thermo Scientific, Waltham, MA, USA) according to the manufacturer’s recommendations at cell density of 1 × 106 cells/mL with the corresponding plasmid for each AAV or AdV element. Transfections with pmaxGFP vector (Lonza, Basel, Switzerland) were applied as positive control for transfection and a mock plasmid with the same backbone and promoter but no coding gene, as a NC. Cell concentration and viability were assessed using a Nucleocounter NC-3000 (Chemometec, Allerod, Denmark) following the manufacturer’s instructions. Transcriptomics For four biological replicates, at 48 hpt, 1.5 mL of cultures were harvested by centrifuging at 300 g for 10 min. Supernatants were discarded and cell pellets were resuspended in 300 μL of DTT/RLT buffer (20 μL of 2M DTT in 1 mL of commercial RLT buffer). Resuspended pellets were vortexed for 20 s and frozen at −70°C until further use. Total RNA was extracted using a Qiagen RNeasy Extraction Kit (QIAGEN N.V., Venlo, the Netherlands) following the manufacturer’s instructions. RNA concentration values were assessed using a NanoDrop 2000 (Thermo Fisher Scientific) spectrophotometer. RNA integrity number (RIN) was measured with an Agilent Fragment Analyzer (Agilent Technologies, Santa Clara, CA, USA). All RNA samples were stored at −20°C until further use for next-generation sequencing (NGS). High-throughput RNA sequencing was performed by Novogene UK (Novogene Co., Beijing, China) utilizing 30 M reads, paired-end, 150 bp strategy based upon Illumina technology. A complete dataset was generated for each condition and treated through the following bioinformatic tools. Quality of FASTQ files was assessed using FastQC.52 After that, pseudoalignment was performed with Kallisto53 using as reference index the latest release of Cricetulus_griseus_chok1gshd.CHOK1GS_HDv1 for CHO-S cells combined with the AdV5 genes described by Donovan-Banfield et al., to account for the added viral genes.54 Following this, the quality of the pseudoalignment was checked using MultiQC55 and pairwise differential expression analyses were performed with Sleuth R package56 based on transcript abundance estimates generated by Kallisto. A likelihood ratio test (LRT) was applied to conduct pairwise comparisons of the condition of interest vs. the mock condition, enabling identification of genes showing significant expression changes across conditions. p values were adjusted for multiple testing using the Benjamini-Hochberg procedure, and genes with a FDR (q value) ≤ 0.05 were considered significantly DE. All DE genes (DEGs) used in the study for functional analysis present a q value < 0.05 in the Sleuth model. These were grouped and annotated using DAVID Functional Annotation Bioinformatics Microarray Analysis.57 Proteomics For three biological replicates, at 48 hpt, 1.5 mL of cultures were harvested by centrifuging at 300 g for 10 min. Supernatants were discarded and cell pellets were frozen at −70°C until further use. Samples were thawed in ice lysed in guanidinium hydrochloride buffer containing 100 mM TCEP, 400 mM CAA, and 500 mM HEPES (pH 8.5), followed by mechanical disruption with a tissue lyser at 30 Hertz for 1′, heating at 95 for 5′ and sonication 5 × 60″ on/30″ off using BioRuptor (Diagenode, Liege, Belgium) to ensure complete lysis. Protein concentrations were determined using a Pierce Rapid Gold BCA assay. For digestion, 20 μL of each sample were diluted with digestion buffer (50 mM HEPES pH = 8.5, 10% ACN) and treated with 200 ng LysC (enzyme to protein ratio 1:50). Samples were vortexted and incubated at 37°C for 4 h, shaking at 700 rpm. About 100 ng of trypsin were added to the sample (enzyme to protein ratio 1:100) and the samples were incubated overnight at 37°C, shaking at 700 rpm. Samples were acidified with 2% trifluoroacetic acid (TFA) (1% final concentration). The pooled samples were subsequently desalted according to the following: SOLAμ columns were activated sequentially with 200 μL methanol, 200 μL Buffer B′ (80% acetonitrile (ACN), 0.1% formic acid (FA)), and two washes of 200 μL Buffer A′ (3% ACN, 1% TFA). Samples were then loaded in 400 μL aliquots and centrifuged at 1,500 rpm for 1 min; this step was repeated as necessary until the entire sample had passed through, and columns could be removed from the plate to ensure complete flow-through. After loading, the columns were washed twice with 200 μL of MS Buffer A (0.1% FA), and bound peptides were eluted with two 30 μL volumes of Buffer B′ (40% ACN, 0.1% FA). Eluates were collected either and transferred into 1.5 mL LoBind Eppendorf tubes (Eppendorf, Hamburg, Germany by centrifugation. The eluates were then dried in a speed vacuum concentrator for approximately 1 h. For TMTpro labeling, dried peptides were reconstituted in 20 μL of 50 mM HEPES buffer (pH 8.5) by vortexing and brief sonication, and peptide concentrations were measured using a Nanodrop spectrophotometer to ensure equal input from each sample. A normalization mixture containing equal parts of all samples was also prepared. TMTpro reagents were equilibrated to room temperature for approximately 10′, dissolved in 20 μL anhydrous ACN, and incubated with gentle mixing to ensure full solubilization. About 20 μg of peptide from each sample was labeled with TMTpro in a final volume of 50 μL with 30 μL sample in HEPES buffer and 20 μL reagent. Labeling reactions were incubated for 1 h at room temperature with shaking at 1,200 rpm. Reactions were quenched by adding 5 μL of 5% hydroxylamine in 200 mM triethylammonium bicarbonate (TEAB), followed by a 15′ incubation at room temperature. Labeled samples were then pooled in equal ratios, and the ACN concentration of the mixture was adjusted to <5% by dilution with 2% TFA. Lysates were subjected to high pH fractionation to increase proteome depth using the Thermo Acclaim PA2 (3 μm, 300 μm × 150 mm, cat #164592), installed in Dionex Ultimate 3000 system (U3000) and combined with a autosamples/fractionator. After fractionation, samples were processed in Orbitrap Eclipse coupled with nanoLC EASY1000 (Thermo Fisher Scientific, Waltham, MA, USA), performed by DTU Proteomics Core facility. For peptide identification, the protein sequence database used was supplemented with the full-length amino acid sequences of all transfected AAV and AdV helper proteins (Table S2), enabling direct detection and relative quantification of viral protein expression in each corresponding condition. Raw data files were processed using Mascot58 and for the comparative analysis of the protein abundance changes, we applied weighted scan-peptide-protein statistical workflow59,60 using iSanXoT package.61,62 It uses as an input a list of quantifications in the form of log2-ratios (e.g., a condition vs. control sample) with their statistical weights and generates the standardized forms of the original variables, computing the quantitative values expressed in units of standard deviation around the averages. The quantitative information is obtained from the spectra and used to quantify the peptides from which the spectra are produced and then the proteins that generate these peptides. In other words, the quantitative information is integrated from the spectrum level to the peptide level and then from the peptide level to the protein level.63 These standardized variables (Zq), express the quantitative values in units of standard deviation.60 All identified proteins with their respective Zq and FDR are included in Table S3. For functional annotations, DAVID57 was used to perform functional enrichment analysis and to extract p values for the enriched processes. Data analysis A small degree of signal leakage or bleeding between adjacent reporter ion channels in the TMT experiment was expected to occur as all channels contain a high expression protein under a strong promoter. This is an inherent property of isobaric labeling chemistry, resulting from the natural isotopic distribution of the reporter tags, which can lead to minor cross-channel signal contribution at the level of peptide-spectrum matches (PSMs). This effect can be reflected in the apparent distribution of identified peptides across conditions. Importantly, this phenomenon did not affect the final quantitative interpretation of the data. The leakage contribution was explicitly corrected during data processing using Proteome Discoverer, and downstream integration from PSMs to peptides and proteins was further handled within the iSanXoT workflow. In addition, we implemented two measures to minimize and control for any residual bleeding effects: (1) careful experimental design of TMT channel assignment to distribute biological replicates across the plex in a way that minimizes systematic bias from adjacent-channel interference, and (2) strict filtering based on protein-level reproducibility, retaining only proteins with a coefficient of variation below 60% across biological triplicates. Together, these steps ensure robust and accurate quantification and prevent distortion of the biological interpretation of the results. Bioinformatics and statistical analyses were performed in Python (v.3.9.6) using commonly applied packages for data handling, statistics, and visualization. Data processing and numerical calculations were carried out with numpy and pandas, while statistical tests were performed with scipy. PCA was implemented with scikit-learn to explore patterns of variance in nuclear proteins, after restricting the dataset to proteins annotated with the nucleus GO term (GO:0005634). Functional enrichment analysis was conducted using GO and KEGG pathway annotations, and results were visualized as lollipop and dot plots, where dot size scaled with –log10(p) values from enrichment statistics. Heatmaps were generated with seaborn and matplotlib, using a blue-white-red gradient to represent downregulation, no change, and upregulation, respectively. Validation of transcriptomic expression One-step quantitative reverse-transcription PCR (RT-qPCR) was performed to validate the expression of selected transcripts identified in the transcriptomics dataset. Total RNA was isolated using the RNeasy Plus Mini Kit (QIAGEN N.V., Venlo, the Netherlands), and reactions were carried out with the Brilliant II SYBR Green qRT-PCR Master Mix (Agilent Technologies, Santa Clara, CA, USA) on a QuantStudio 5 Real-Time PCR System, following the manufacturer’s instructions. Relative transcript levels were quantified using the comparative threshold cycle (ΔΔCt) method, with expression normalized to GAPDH as the endogenous control. Thermal cycling was carried out as follows: reverse transcription at 50°C for 30 min; polymerase activation at 95°C for 10 min; and 40 amplification cycles of 95°C for 30s and 60°C for 1 min, with fluorescence data acquired during the annealing/extension step. A post-amplification melt curve analysis was included to verify specificity of the amplification products. Cell line engineering For the generation of the Mmp9−/− knockout cell line, four different sgRNAs were designed for each gene to knock out. Two in the 5′-UTR region or first exon and two in the 3′-UTR region or last exon, depending on the efficiency and availability of CRISPR-Cas9 target sequences found by CHOPCHOP online tool. CHO-S cells were transfected at 2 × 106 cells/mL with four plasmids encoding the four single guide RNA (sgRNA) sequences and the Cas9:2A:eGFP protein at a final concentration of 1.25 μg/μL of DNA. The plasmid coding Cas9 was used at 0.625 μg/μL and the rest of DNA was equally divided among the sgRNAs used. Briefly, the plasmids were diluted in supplemented culture medium and vortexed for 10 s. The transfection mix corresponded to 5% of the total working volume. Then, FreeStyle Max (Gibco, Life Technologies) was added to the plasmid mix at a ratio of 1:1 (FreeStyle:DNA w/w) and gently inverted 4–6 times. After 10 min incubation at room temperature, the mixture was added to the culture. At 72 hpt, GFP-positive cells were single sorted into 384-well plates (Corning Life Sciences) using fluorescence-assisted cell sorting (FACS) in a Sony SH800 cell sorter (Sony Biotechnology, San Jose, CA, USA). Each well contained 30 μL of the corresponding supplemented culture medium with an additional 25% of 4-days-exhausted medium, 2.5% Instigro CHO (Advanced Instruments, Norwood, MA, USA), 1.5% HEPES (Gibco) and 1% antibiotic-antimycotic (Gibco). After single-cell sorting, the presence of only one cell per well was assessed with a Celigo Imaging Cell Cytometer (Nexcelom Bioscience). Ten days after the sorting, subconfluent clones were transferred to flat-bottom 96-well plates with 180 μL of corresponding culture medium with 1% of antibiotic-antimycotic (Gibco) an epMotion 5070 liquid handling workstation (Eppendorf). Subsequently, cells were expanded to flat-bottom 12-well plates and to 125 mL shake flasks for cell banking. For clone screening, genomic DNA was extracted using QuickExtract DNA Extaction solution (Middleton, WI, USA) and GeneJet Genomic DNA purification kit (Thermo Scientific) following the manufacturer’s instructions. Then, the expected cuts between the sgRNAs were identified by PCR and confirmed by Sanger sequencing (Eurofins Scientific, Luxembourg) (Figure S2). Cloning and expression of selected targets Coding sequences for Mx2 and Rad23A were synthesized (Twist Bioscience, San Francisco, CA, USA) and individually cloned into the p100RPU plasmid backbone using NEBuilder HiFi DNA Assembly (New England Biolabs, Ipswich, MA, USA), following the cloning strategy described previously. Transfections were performed in 3 mL cultures using FreeStyle MAX as transfection reagent and 2 μg of total DNA per mL of culture. Each condition received the standard rAAV production plasmids (helper and AAV plasmids) alone, or in combination with the Mx2, Rad23A, or both expression constructs. rAAV production was subsequently assessed by qPCR. rAAV quantification Cell culture was harvested at 72 hpt and diluted 1:1 with lysis buffer (TrisHCl 400 mM, Triton X-100 1% and MgCl2 20 mM adjusted to pH 7.5). After 1 h incubation at 37°C and agitation at 130 rpm, the lysate was centrifuged for 20 min at 4,000× g and 4°C. The supernatants were stored at −70°C for long-term storage. On the quantification day, samples were thawed in a control manner at room temperature. Then, 5 μL of sample were mixed with 2 μL of 1 U/μL DNase I (Thermo Fisher Scientific, San Jose, CA, USA) and 13 μL of 10X DNase I reaction buffer with MgCl2 (Thermo Fisher Scientific). The mix was incubated for 16 h at 37°C. To inhibit the activity of DNase I, 4 μL of 50 mM EDTA (Thermo Fisher Scientific) were added to the mix followed by a 30 min incubation at 70°C. Lastly, 5 μL of Proteinase K (Thermo Fisher Scientific, San Jose, CA, USA) was added to the sample and incubated for 2 h at 55°C. The Proteinase K was inactivated for 15 min at 95°C. Around 2 μL of the freshly treated sample were mixed with 0.5 μL of 10 μM forward primer (5′-ACGTCAATGGGTGGAGTATTT-3′) and reverse primer (5′- AGGTCATGTACTGGGCATAAT-3′) binding to the GFP sequence with 5 μL of Brilliant III Ultra-Fast SYBR Green QPCR Master Mix (Agilent Technologies, Santa Clara, CA, USA) and 0.15 μL of 2 μM reference dye. Amplification was executed in QuantStudio 5 Real-Time PCR System (Thermo Fisher Scientific, Waltham, MA, USA) with the following conditions: 50°C for 2 min, 95°C for 10 min; 40×: 95°C for 15 s, 60°C for 1 min, 95°C for 15 s, 60°C for 1 min, and 95°C for 1 s. In order to ensure the reliability of the generated data, two controls were added: saturate lysate control (SLC) and transfected lysate control (TLC). The SLC was prepared from a non-transfected cell culture which underwent the lysis protocol in parallel with the transfected samples. Then 5 μL of cell lysate were mixed with 40 ng of pGOI plasmid, 9 μL of 10X DNase I reaction buffer with MgCl2 and 2 μL of 1 U/μL DNase I. SLC control aimed to verify that the DNase step was successful. For the generation of the TLC, a single plasmid transfection was performed with pGOI following the triple transfection, lysis and the quantification protocol. The TLC aimed to guarantee that the effect of DNase I remains independent of whether the plasmid has been transfected or spiked afterward. Plates run in QuantStudio 5 Real-Time PCR System were analyzed via ThermoCloud (Thermo Fisher Scientific). Data and code availability The raw mass spectrometry data is submitted to the ProteomeXchange Consortium via the Mass Spectrometry Interactive Virtual environment (MassIVE) with dataset identifiers PXD069313 and MSV000099445, respectively. NGS data was uploaded to The Sequence Read Archive (SRA) with BioProject ID: PRJNA1343252. Acknowledgments The authors would like to acknowledge generous support from 10.13039/100032285Novo Nordisk Foundation. This work was supported by NNF20CC0035580 and NNF20SA0066621. L.K.N. is supported by NNF14OC0009473. J.L.-G. is supported by NNF22OC0078741 and 10.13039/100010665Marie Skłodowska-Curie Actions (MSCA) postdoctoral fellowship 101105465. K.T. is supported by a DTUQ Alliance PhD grant. Author contributions K.T., experimentation, data analysis, writing the scripts, biological analysis, writing the original draft, reviewing and editing; G.L.-G., Mmp9−/− cell line generation; M.I.S., data analysis; L.K.N., conceptualization; J.L.-G., supervision, conceptualization, data analysis, writing the scripts, writing the original draft, reviewing and editing. Declaration of interests The authors declare no competing interests.