Enhancing gene therapy vectors manufacturing: CEBPA as a master regulator of rAAV biogenesis

Abstract

Background

Recombinant adeno-associated viruses (rAAVs) are the vectors of choice for gene therapy applications due to their favorable safety and efficacy profiles. However, current production platforms—particularly transient transfection-based HEK293 systems—have scalability challenges, primarily due to the high cost of GMP-grade plasmid DNA and low specific productivity when scale up is required. This has become a significant bottleneck in the commercialization of rAAV-based gene therapy products, prompting recent market withdrawals or limited commercialization of gene therapy products based on rAAV vectors. In this context, stable producer cell lines (PCLs), such as HeLaS3-based systems, offer a promising alternative for cost-effective and scalable rAAV manufacturing. Nonetheless, these systems are still in early stages of development and often yield lower titers.

Results

To address these limitations, we employed a CRISPR activation (CRISPRa) screen to identify genetic regulators that enhance rAAV production. This approach revealed several pathways related to protein trafficking and immune response as key contributors to rAAV biogenesis. Notably, we identified CEBPA gene as a master regulator in this context. Overexpression of CEBPA reprogrammed the host transcriptome, activating immune-related pathways and enhancing cellular metabolism. Importantly, when combined with bioprocess intensification strategies, i.e. perfusion, CEBPA-overexpressing cells were able to maintain cell-specific production yield at higher cell density while maintaining high vector quality. This combined approach led to up to 10-fold increase in volumetric productivity compared to non-modified parental cells.

Conclusions

These findings will contribute to the development of a robust and scalable production platform that enhances vector yield without compromising quality. Our approach addresses a critical barrier in gene therapy manufacturing offering a practical path towards broader clinical and commercial viability.

Supplementary information

The online version contains supplementary material available at 10.1186/s13036-026-00622-3.

Background

Recombinant adeno-associated viruses (rAAVs) are the viral vectors of choice for in vivo gene therapy applications due to their favorable safety profile, broad tropism, and ability to support sustained transgene expression [1, 2]. These characteristics have enabled the approval of over 350 clinical trials and the commercialization of eight gene therapy products [3]. AAVs are small non-enveloped viruses carrying a single-stranded DNA genome with 4.7 kb. The AAV genome is flanked by inverted terminal repeats and encodes the information for two genes, the rep gene and the cap gene, and two subgenomic mRNAs important for viral packaging, the assembly activating protein and membrane-associated accessory protein. These genes, essential for viral replication and assembly, are provided in trans when producing rAAV-based gene therapies [4].

The increasing demand for rAAV vectors in gene therapies, particularly those requiring systemic administration, has revealed a critical bottleneck in existing industrial manufacturing platforms. Current Chemistry, Manufacturing and Controls (CMC) practices rely on transient production systems, which ultimately are deemed to be non-scalable and economically unfeasible due to the high costs of GMP grade raw materials [5]. In this context, the development of stable producer cell lines (PCLs) represents a promising alternative, offering a scalable and cost-effective solution capable of supporting the large-scale production demands of rAAV-based therapeutics.

Among the currently available PCL platforms, the HeLaS3-based system is one of the most promising systems for rAAV manufacturing since it is scalable [6]. This system requires the induction of rAAV production, usually through the infection of a helper virus, for instance wtAd5. The wtAd5 provides important functions such as inducing the rAAV promoters, essential for rAAV production, and promotes DNA packaging and virus assembly, which has a direct impact on product quality [7]. Escandell et al. [8] have already reported the implementation of an integrated and streamlined workflow encompassing HeLaS3 clonal selection, bioreactor-scale production, and downstream processing. Despite these improvements, the broader industrial application of the HeLaS3 platform remains limited, primarily due to its inability to consistently achieve the high vector titers required to meet current clinical and commercial demands.

Targeted cell line engineering has emerged as a powerful strategy to enhance productivity in biomanufacturing platforms, as demonstrated in systems such as CHO for recombinant protein production [9], HEK293 for viral vectors [10], and Sf9 insect cells for rAAV production [11]. However, the application of cell engineering strategies to enhance the HeLaS3 platform remains limited, primarily due to an incomplete understanding of the cellular and molecular mechanisms governing rAAV production in this system. Consequently, HeLaS3 cell line engineering often relies on empirical approaches, making the process laborious and less predictable.

CRISPR-Cas9 genome-wide genetic screening has emerged as a transformative technology, enabling unbiased and high-throughput interrogation of gene function across diverse biological contexts. This approach has been widely applied in fields ranging from fundamental cell biology and oncology to pharmacology and drug discovery [12]. The evolution of CRISPR-based tools beyond gene knockout to include modalities such as transcriptional repression [13] and activation (CRISPRa) [14] has further expanded its utility, allowing for complementary strategies to address the same biological question. For rAAV gene therapies, genome wide genetic screens have been applied to the understanding of host-restricting factors in AAV transduction [15] and in identifying genes that improve rAAV titer and quality attributes, in this case full capsids and infectious titers in HEK293T cells [12, 16].

In this work, using a gain-of-function genome-wide CRISPR activation screen, we identified key host regulators that enhance rAAV replication and assembly in HeLaS3 cells. This approach revealed members of the CCAAT/enhancer-binding protein alpha (CEBPA) family as critical transcriptional regulators of rAAV production in this platform. Overexpression of CEBPA resulted in up to ten-fold increase in vector yield compared to non-targeting controls. Furthermore, we present a proof-of-concept for process intensification using CEBPA-overexpressing HeLaS3 cells. These engineered cells demonstrated robust performance at higher cell densities, maintaining cell-specific rAAV productivity comparable to that observed under low-density conditions, thus allowing to achieve higher volumetric yield.

Collectively, these findings establish a novel engineered HeLaS3-based rAAV production system capable of delivering consistent, high-yield vector production. This approach offers a promising solution to meet the stringent demands of CMC for clinical and commercial gene therapy applications.

Methods

Cell culture conditions

HeLaS3 (ATCC CCL-2.2) derived rAAV PCLs were generated as described in [8]. Cells were routinely sub-cultured at 0.3x10⁶ cell/mL in SFM 293 II (11686029, Gibco^TM) supplemented with 4 mM L-Glutamine (25030024, Gibco^TM) every 3–4 days, when cell concentration reached 1-3x10⁶ cell/mL. Cells were maintained at 37 °C in a humidified atmosphere of 5% CO₂ in air under 125 rpm agitation (25 mm orbital diameter) in shake flask. HEK293T cells (ATCC CRL-3216) were maintained in adherent monolayer culture in Dulbecco’s Modified Eagle Medium (DMEM) (10–013-CV, Corning) with 10% FBS (10270106-Thermofisher) in T-flasks at 37 °C in a humidified atmosphere of 5% CO₂ in air. Cell counts were performed by Trypan Blue exclusion method using a Vi-cell XR (Beckman) or Countess (Invitrogen^TM).

Lentivirus production and titration

For lentivirus production, third generation plasmids were used: Gag-Pol plasmid (pMDLg/pRRE: Addgene plasmid # 1225), Rev plasmid (pRSV-Rev: Addgene plasmid # 12,253) and VSV-G envelope plasmid (pMD2.G: Addgene plasmid # 12,259). HEK293T cells were transfected according to jetOPTIMUS® (Polyplus) protocol. Lentiviruses were collected from 48 h post-transfection and stored at 80 °C until used. To titrate the lentivirus encoding the Calabrese library HeLaS3-derived rAAV PCLs were transduced with them and collected after 48 h. DNA was extracted using the DNeasy® Blood & Tissue Kit (50969504, Qiagen), following manufacturer’s protocol. Lentiviral insertions were quantified by ddPCR using the following primers and probes: albumin detection primers (Fw: 5’-GCTGTGAAACTCTGTTGG-3’; Rv: 5’-GACATCCTTTGCCTCAGCAT-3’) and probe (5’-/5 -HEX/AGTGGAAAA/ZEN/TGATGAGATGCCTGCT/3/AbkFQ/-3’); and LTR detection primers (Fw: 5’-AACTAGGGAACCCAC-3’; Rv: 5’- GCTAGAGATTTTCCACACTGA–3’) and probe (5’-/56- FAM/CTTGCCTTG/ZEN/AGTGCTTCAAGTAGTG/3/AbkFQ/-3’). Lentivirus infectious titer was calculated with the following formula:

Western blot

Cell pellets were collected and homogenized in lysis buffer consisting of 4x NuPAGE™ LDS Sample Buffer (NP0007, Invitrogen) and 10x NuPAGE™ Sample Reducing Agent (NP0004, Invitrogen). Samples were run in a 4–12% polyacrylamide gel (NP0321BOX, Invitrogen) at 200 V for 1 h in MOPS Running Buffer (NP000102, Invitrogen). Gel was then transferred to a nitrocellulose membrane with the iBlot 2 Dry Blotting System (Invitrogen™). The nitrocellulose membrane was incubated with the antibodies Anti-CEBPA (ab317442, abcam), diluted 1:15000; Anti-Cas9 (ab271293, abcam), diluted 1:5000; and the Anti-beta-actin (A5441, Sigma), diluted 1:10000. The secondary antibodies used were Anti-rabbit, (NA9341, Cytiva), for CEBPA detection, and Anti-mouse (NA931-1 ML, Cytiva), for dCas9-VP64 and β-actin detection, diluted 1:50000. Membranes were imaged in iBright (Invitrogen) using Enhanced chemiluminescence (ECL; NEL103E001EA, Perkin Elmer).

rAAV production and titer quantification

For rAAV production, cells were centrifuged (300 g, 5 min) and resuspended in production medium consisting of 80% DMEM + 20% SFM 293 II + 7 mM L-Glutamine. To provide helper functions, wtAd5 viruses were added to the cells at a specific Multiplicity of Infection (MOI) of 3 particles/cell. Cells were incubated in 125 rpm agitation (25 mm orbital) at 37 °C in an atmosphere of 5% CO₂ in air atmosphere until harvest. For rAAV quantification, cells were harvested and lysed in 0.5% Na-Deoxycholate, 2 mM Magnesium chloride with 30 U/mL of Benzonse (101695001, Merck) and incubated at 37 °C for 1.5 h. Cell lysates were treated with 60 U/mL DNAse RQ1 (PROMM6101, Promega) at 37 °C for 30 min, followed by a Proteinase K digestion at 37 °C for 30 min. Proteinase K was inactivated by incubation at 95 °C for 20 min. The qPCR was performed in Light Cycler 480® (Roche) using SYBR Green I detection reagent (04707516001, Roche) or Probes Master Reagent (04887301001, Roche). Primers for BGH polyA sequence (Fw: 5′- TCTAGTTGCCAGCCATCTGTTGT-3’; Rv: 5′-TGGGAGTGGCAC CTTCCA-3′) and probe (5’-/56-FAM/TCC CCC GTG/ZEN/CCT TCC TTG ACC/3IABkFQ/-3’) were used for rAAV quantification.

rAAV purification and full-to-empty quantification

For full capsids measurements, infected rAAV producer cells were collected by centrifugation (300 g, 5 min). The cell pellets were resuspended in 1 mL of Lysis Buffer (50 mM Tris-HCL pH 8, 20 mM MgCl₂, 1% Tween) and digested with benzonase at 150 U/mL for 1.5 h at 37 °C. The solution was then incubated with 200 mM NaCl for 15 min at 37 °C. The cells were centrifuged at 4000 g for 5 min, and the supernatant was clarified with a 0.45 μm filter. The supernatant was subjected to several purification steps using the phytips^TM, including 2 cycles of equilibration (50 mM Tris, pH 8, 350 M NaCl, 0.001% Pluronic), 6 cycles of sample capture, 2 washing steps with 2 cycles each (Wash 1 - 50 mM Tris, pH 8, 1 M NaCl, 0.001% Pluronic; Wash 2 - 50 mM Tris, pH 8, 0.001% Pluronic), and 4 elution cycles (50 mM Citric Acid pH 2.5, 0.001% Pluronic). Full-to-empty ratio was determined on the purified samples by mass photometry (SamuxMP, Refeyn).

rAAV infectious units quantification

rAAV infectious titer was determined as previously described with minimal modifications [17]. HeLa RC32 were plated in 96 well plates at a cell concentration of 15,000 cells/well in DMEM + 10% FBS + 4 mM L Glutamine and incubated ON at 37 °C in a 5% CO₂ controlled atmosphere. rAAV dilutions were prepared starting at an MOI of 7x10⁴ vg/cell in 10-fold serial dilutions in DMEM + 1% FBS + 4 mM L Glutamine containing 3.20x10⁸ wtAd5 DRG/mL. Cells were infected by total medium exchange and incubated for 2 h. At this point, the culture medium was diluted 1:2 with DMEM + 10% FBS + 4 mM L Glutamine. At 24 h post-infection, images were acquired with 10x amplification using MICA microscope (Leica) and rAAV infected cells were quantified through the eGFP transgene expression using the following equation:

FACS-Based enrichment and genomic profiling of high rAAV2 producers

To screen for rAAV specific productivity, a total of 5x10⁸ cells were infected with wtAd5 as described previously, ensuring a sgRNA coverage above 1000. For rAAV2 staining of producer cells, the infected PCL was harvested and centrifuged for 5 min at 300 g. Cells were fixed with PBS + 2% PFA (V/V) for 7 min. Afterwards Glycine was added to the solution at a final concentration of 250 mM for quenching. For rAAV2 staining, cells were incubated under agitation at room temperature with the A20 antibody (61055, Progen) followed by a secondary antibody staining of Alexa Fluor 647. Sorting was performed in a BD FACS Aria II. Cells were directly lysed after sorting by addition of lysis buffer (Buffer AL, 50,919,075). DNA was recovered by DNeasy® Blood & Tissue Kit (69504, Qiagen) or QIAamp DNA Micro Kit (56304, Qiagen), following manufacturer’s instructions. The sgRNA sequences from each population were amplified by nested PCR approach targeting the sgRNA area. Next Generation sequencing was performed by Illumina NovaSeq X paired end 150 bp at CD Genomics. Bioinformatic analysis was performed using Galaxy platform (galaxy.eu). The list of genes enriched in high producing population was analyzed in QIAGEN IPA (QIAGEN Inc., https://digitalinsights.qiagen.com/IPA) to determine the main canonical pathways affected.

Cell line engineering

Lenti dCas-VP64_Blast plasmid was a gift from Feng Zhang (Addgene plasmid # 61,425). A clonally derived rAAV PCL was transduced with lentiviruses carrying dCas9-VP64 in serum-containing medium (DMEM + 4 mM L-glutamine + 10% FBS (V/V)). Cells were incubated with 10 µg/mL blasticidin selection (ant-bl-05, InvivoGen) for 11 days. Cells were maintained as described in [8]. The sgRNA sequences to target CEBPA were CEBPA sgRNA_1 5´-CATGGGGCGGCGAACCAGCG-3´; CEBPA sgRNA_2 5´-CCGAGGCGGCCTCTGTCCCC-3´; CEBPA sgRNA_3 5´-GCCTGCTGGGTCCTAGCGCG-3´; and CEBPB sgRNA 5´-CCCTGCCGCCGTCCTCCCGG-3´. To obtain a CEBPA-overexpressing cell line, 0.5x10⁶ cells from the clonally derived rAAV PCL were transduced with lentiviruses carrying the plasmid of interest. Cells were transduced in serum-containing medium (DMEM + 10% FBS (V/V)) and, after 48 h kept in 0.5 µg/mL Puromycin selection (A1113803, Gibco) until viability reached > 90%. Cells were then adapted to suspension cell culture.

Gene expression determination by RT-qPCR

CEBPA and CEBPB genes expression was confirmed by RT-qPCR. RNA was extracted and cDNA synthesized using the SYBR™ Green Fast Advanced Cells-to-CT™ Kit (A35379, Invitrogen™), following manufacturer’s instructions. Gene expression was quantified by qPCR with the Lightcycler480 using the SYBR Green I Master (04887352001, Roche) and normalized with the HPRT1 gene. Primer sequences used for gene expression analysis by RT-qPCR were CEBPA Forward 5´-TCGGTGGACAAGAACAG-3´, CEBPA Reverse 5´-CGCAGGCGGTCATTG-3´; CEBPB Forward 5´-CTTGAGCCCGTACCTGGAG-3´, CEBPB Reverse 5´-GGAGAGGAAGTCGTGGTCC-3´; HPRT1 Forward 5´-CCTGGCGTCGTGATTAGTGAT-3´, HPRT1 Reverse 5´-AGA CGT TCA GTC CTG TCC ATA A-3´.

RNA sequencing and differential gene expression analysis

For RNA extraction and sequencing, 5x10⁶ cells were resuspended in RNAlater stabilization solution (AM7020, Invitrogen) at a final concentration of 1x10⁷ cell/mL. RNA extraction, library preparation and Next Generation sequencing were performed at CD Genomics. RNA transcripts were aligned to the human reference genome (GRCh38 (hg38)) with STAR [18] (version 2.7.11b) under standard configurations. Differential expression analysis was performed with DESeq2 [19]. The data was interpreted to determine canonical pathways, upstream regulators, gene networks and biological functions significantly influenced by CEBPA overexpression with IPA software.

Bioreactor runs and perfusion

rAAV production was performed in a computer-controlled 1 L stirred-tank bioreactor (BIOSTAT® DCU-3, Sartorius) operated in perfusion. HeLaS3 rAAV producer cells overexpressing CEBPA were seeded at a density of 2x10⁶ cells/mL and immediately infected with adenovirus at a MOI of 1 IU/cell. Perfusion was implemented via alternating tangential flow filtration (ATF). The ATF system employed an XCell™ ATF-2 module (Repligen) with a polyethersulfone (PES) hollow fiber membrane (0.5 μm pore size, 1 mm lumen diameter, 0.13 m² surface area) as described in [20]. The specific perfusion rate intended to control lactate accumulation, targeting concentrations below 5 mM at 24 h and below 10 mM at 48 h post infection, corresponding to 4 vessel volumes per day (VVD) and 2 VVD, respectively. Culture volume was maintained via gravimetric feed control. Dissolved oxygen concentration in culture medium was maintained at 45% air saturation by adjusting agitation speed between 70 and 150 rpm. As control, a bioreactor operated in batch was seeded at a cell density of 0.5x10⁶ cells/mL and infected with adenovirus at a MOI of 1 IU/cell (the ideal experimental conditions previously identified in shake-flask). The concentrations of lactate, glucose, glutamine and ammonia were measured using Cedex Bio Analyzer 7100 (Roche, Switzerland). Metabolite consumption/production rates and rAAV2 specific productivity were calculated as previously described [20].

Results

Genome-wide CRISPRa screen identifies pathways modulating rAAV productivity

To have an unbiased approach to uncover mechanisms underlying rAAV biogenesis in the HeLaS3-based production system, we conducted a genome-wide CRISPRa screen targeting over 19,000 human genes using the Calabrese library, which contains guide RNAs designed for transcriptional activation [14] (Fig. 1A). We first established a CRISPRa-compatible HeLaS3 cell line by transducing a previously developed PCL [8] with lentiviral particles encoding dCas9-VP64. This catalytically inactive Cas9 variant [21], fused to the VP64 transcriptional activator, allows programmable gene upregulation via sgRNA-directed binding to promoter regions [22]. Engineered cells maintained stable dCas9-VP64 protein levels (Fig. 1B), normal population doubling time (PDT), and high viability over 31 days in culture (Fig. 1C), confirming phenotypic stability and suitability for gain-of-function screening of the established engineered cell line.

Fig. 1.

Generation of a stable rAAV producer cell line for genome-wide gain of function screening to identify genes involved in rAAV production. A) Schematic overview of the experimental workflow. HeLaS3 rAAV producer cells (PCL) were transduced with lentiviral vectors encoding dCas9-VP64 and a pooled sgRNA library targeting 19,114 genes (56,762 sgRNAs total), generating a heterogeneous cell population with one gene overexpressed per cell. B) Western blot analysis showing sustained expression of dCas9-VP64 over time in suspension cell culture (days 0, 3, 10, 17, and 31). PCL were transduced with lentiviral particles encoding for dCas9-VP64 and grew for several generations. Cas9 antibody was used to immunodetect dCas9-VP64 and β-actin was used as a loading control. C) population doubling time (PDT, left y-axis, circle) and cell viability (right y-axis, square) of dCas9-VP64-expressing PCL over time in suspension culture. D) volumetric rAAV2 productivity (vg/mL ± SD, left y-axis, bars representing the mean) and cell viability (%, right y-axis, full symbol) at various harvest times (48 h, 66 h, 72 h, and 93 h post-infection with wtAd5). Each bar color represents a different condition: parental PCL alone in light grey, PCL + dCas9-VP64 in blue, and PCL + dCas9-VP64 + sgRNA in dark blue. Data represented as mean ± standard deviation (SD) (n = 2)

In this study, rAAV production is triggered by infection with wtAd5, which activates rAAV expression and starts a lytic cycle that causes the release of the rAAV particles. Since the CRISPRa screen was performed using a Fluorescence-activated cell sorting (FACS)-based approach—requiring the discrimination of rAAV productivity at the single-cell level—it was critical to identify a harvesting time point at which rAAV production was maximized while maintaining cell membrane integrity. This was essential for the robust identification of genetic factors enhancing rAAV production in the context of a lytic infection. Therefore, the parental PCL, a cell line stably expressing dCas9-VP64 (PCL+dCas9-VP64), and a cell line co-expressing dCas9-VP64 with the single-guide RNA library (PCL+dCas9-VP64+sgRNA), were infected with wtAd5 at a MOI of 3 infectious units (IU) per cell. Cell viability and rAAV2 titers were assessed at 48-, 66-, 72-, and 93-hours post-infection (hpi) (Fig. 1D). Cell viability remained above ≥80% in all conditions up to 66 hpi, with a marked decline observed at 72 hpi and beyond. Correspondingly, rAAV2 titers peaked at 66 hpi in all cell lines, highlighting this harvesting time point as the best trade-off between high cell viability and high rAAV2 production and used in the CRISPRa screen. Moreover, expression of dCas9-VP64 alone and in combination with sgRNAs led to approximately four-fold reduction in rAAV2 titers compared to the parental PCL. These findings suggest that activation of the CRISPRa system may negatively impact rAAV production, potentially through transcriptional interference or cellular stress responses associated with dCas9-VP64 activity and sgRNA expression.

As depicted in Fig. 1A, PCL+dCas9-VP64 cells were then transduced with CRISPRa library at low lentiviral MOI aiming at one gene overexpression per cell and screened for rAAV productivity after 66 h wtAd5 infection (Fig. 2A). Flow cytometry analysis of AAV capsid staining revealed two distinct populations: cells with high cell specific rAAV production (2.9%) and cells with no or low cell specific rAAV production (97%) (Fig. 2B). Differential enrichment analysis using MAGeCK algorithm [23] of sgRNA sequences recovered from these two sorted populations highlighted several candidate genes with statistically significant enrichment in the high producer group (Fig. 2C). To further explore the biological relevance of the top genes associated with high rAAV production, Ingenuity Pathway Analysis (IPA) was conducted on 576 differentially expressed genes (566 upregulated and 10 downregulated) with log₂ Fold Change (FC) > 0.5 and p-value < 0.05 (Fig. 2D). The analysis revealed a diverse set of enriched signaling pathways, such as S100 Family Signaling (z-score: 4.7; p-value: -log₁₀ = 4.01), RAF/MAP kinase cascade (z-score: 3.46; p-value(-log₁₀) = 2.35), and Protein Sorting Signaling (z-score: 3.32; p-value(-log₁₀) = 2.71), suggesting a role for intracellular trafficking and stress response mechanisms, such as immune response, in enhancing viral vector yield.

Fig. 2.

Genome-wide CRISPRa screen identifies the CEBP gene family as enhancers of rAAV production. A) Schematic of the screening strategy. HeLaS3 rAAV producer cells expressing dCas9-VP64 and a genome-wide sgRNA library were induced for rAAV production via wtAd5 infection. Cells were stained for assembled AAV2 capsids and sorted by FACS based on productivity. Genomic DNA from sorted populations was extracted, and sgRNA sequences were amplified and analyzed by next-generation sequencing. B) FACS plot showing assembled AAV2 capsid signal versus forward scatter (FSC-A). High and low rAAV-producing populations were gated for downstream analysis. C) Volcano plot from MAGeCK analysis comparing sgRNA enrichment between high and low rAAV-producing populations. The x-axis represents log₂ Fold change, and the y-axis shows statistical significance (log₁₀ p-value). D) Ingenuity pathway analysis of genes enriched in the high-producing rAAV population. Bars represent Z-scores; dots indicate -log₁₀(p-values)

CEBPA is a master regulator of rAAV biogenesis

CEBPA was the top hit in the CRISPRa screen, showing the highest level of enrichment among all genes analyzed (log₂FC: 1.7074; adjusted p-value (−log₁₀): 6.10) (Fig. 2C). The CEBPA gene encodes the CCAAT/enhancer-binding protein alpha, a transcription factor known for its role in regulating myeloid lineage commitment and differentiation during hematopoiesis [24]. Despite its well-characterized function in hematopoietic biology, CEBPA has not previously been associated with rAAV production, positioning it as a novel and promising target for cell engineering strategies aimed at improving vector yield. RT-qPCR analysis revealed that CEBPA expression was enhanced when rAAV production is induced through wtAd5 infection, reinforcing its function as a regulator of rAAV biogenesis (Supplementary Fig. 1). Furthermore, to validate the role of CEBPA in rAAV production, three independent sgRNAs were used to induce overexpression. Immunoblot analysis confirmed increased CEBPA expression in all engineered cell lines, when compared to non-targeting (NT) control (Fig. 3A), confirming on-target activity of the selected sgRNAs. To validate the role of CEBPA in enhancing virus production, further testing confirmed that cell lines overexpressing CEBPA consistently showed a 10-fold increase in viral yield compared to the non-targeting control (Fig. 3B).

Fig. 3.

CEBPA overexpression enhances rAAV production through transcriptional reprogramming and immune pathway activation. A) immunoblot analysis of CEBPA expression in producer cell line (PCL) modified with dCas9-VP64 and sgRNAs targeting CEBPA (sgRNA1, sgRNA2 and sgRNA3). β-actin serves as a loading control. B) Volumetric rAAV2 yield (vg/mL ± SD) in PCLs expressing dCas9-VP64 with non-targeting (NT) control, and individual CEBPA sgRNAs (sgRNA1, sgRNA2 and sgRNA 3). Data represent mean ± SD of minimum N = 2 independent experiments. Statistical analysis: unpaired t-test with Welch’s correction (*: p < 0.05, **: p < 0.005****: p < 0.0001). C) Principal component analysis (PCA) plot showing transcriptional variance among NT and CEBPA-overexpressing (expressing sgRNA2) cell lines. RNA was collected from all cell populations and RNA-Seq analysis was carried out. D) Volcano plot of differential gene expression analysis between NT and CEBPA-overexpressing cell lines. The x-axis shows log₂(fold change), and the y-axis shows -log₁₀(false discovery rate (FDR)). CEBPA is significantly upregulated in CEBPA-overexpressing cell lines. E) Pathway enrichment analysis of differentially expressed genes in CEBPA-overexpressing cells. Bars represent Z-scores; dots indicate -log₁₀(p-values)

One of the key advantages of the HeLa-based PCL platform is its ability to produce rAAV particles with a favorable full-to-empty capsid ratio, often outperforming other production systems in terms of the proportion of full particles [8]. To determine whether CEBPA overexpression preserves this characteristic, we assessed the full-to-empty ratio of purified rAAVs by mass photometry. The analysis revealed a ratio of approximately 60% (Supplementary Fig. 2), closely matching the parental PCL values, indicating that CEBPA overexpression does not compromise capsid integrity.

Interestingly, CEBPB gene also emerged as a hit in the CRISPRa screen (log₂FC: 1.53, adjusted p-value(-log₁₀) = 2.42) (Fig. 2C). To explore the hypothesis that the CEBP family plays a role in regulating rAAV production, we also investigated the impact of CEBPB overexpression on rAAV titers. As shown above for CEBPA, RT-qPCR also confirmed that adenoviral infection significantly upregulated CEBPB expression in HeLaS3 cells (Supplementary Fig. 3A). A regulatory relationship between CEBPA and CEBPB has been previously reported [25] suggesting potential cross-regulation within this transcription factor family. To test whether a similar connection exists in our system we engineered PCL cells to overexpress either CEBPA or CEBPB using sgRNA-mediated activation confirming successful induction of both genes by RT-qPCR (Supplementary Fig. 3B). Notably, while CEBPA overexpression led to a robust increase in CEBPB expression (8-fold vs non transduced cells), the reverse was not observed—CEBPB overexpression did not induce CEBPA—suggesting that CEBPA may act upstream of CEBPB in this context. Moreover, functionally, overexpression of either CEBPA or CEBPB led to a significant increase in rAAV production compared to the non-targeting control (10-fold for CEBPA and 8-fold for CEBPB; Supplementary Fig. 3C). Given the stronger regulatory and functional impact of CEBPA on rAAV production, we focused subsequent experiments in this study on further characterizing its role in optimizing rAAV production.

To determine whether the enhancement in rAAV2 titer observed upon CEBPA-overexpression was not clone-specific, we tested its effect in an independent producer cell line not included in the initial screen. Immunoblot analysis confirmed successful induction of CEBPA in this clone (Supplementary Fig. 4A), and viral genome quantification revealed a marked increase in rAAV2 production (Supplementary Fig. 4B). These results demonstrate that the beneficial effect of CEBPA overexpression on rAAV yield is robust and reproducible across distinct producer cell lines.

CEBPA functions as a transcription factor, and its overexpression is expected to induce widespread transcriptional reprogramming. To investigate how CEBPA modulates the cellular transcriptome and correlate induced pathways that potentially influence rAAV production, we performed transcriptomic profiling using RNA sequencing (RNA-seq). Principal component analysis (PCA) revealed that the transcriptomic profile of the CEBPA-overexpressing cell line was markedly distinct from the non-targeting control lines, which clustered closely together (Fig. 3C). This separation indicates broad transcriptional remodeling driven by CEBPA overexpression. Differential gene expression analysis, visualized in a volcano plot (Fig. 3D), further highlighted the changes in gene expression. Among the top upregulated genes were HP (Haptoglobin, log₂FC = 13.11,-log₁₀(False discovery rate (FDR)) = 20.39), an acute-phase protein involved in binding free hemoglobin to prevent oxidative damage and is commonly associated with inflammatory responses and SERPINB2 (Plasminogen Activator Inhibitor Type 2, log₂FC = 12.84,,-log₁₀ (FDR) = 21.01) regulating fibrinolysis and immune signalling and is implicated in cellular stress and inflammation. Supporting this, IPA analysis comparing the CEBPA-overexpressing cell line to the non-targeting control (Fig. 3E) revealed that the most significantly differentially regulated pathways were immune-related signaling cascades. For example, the S100 Family Signaling Pathway showed significant enrichment with a –log(p-value) of 5.80 and a z-score of 2.71. Conversely, the LXR/RXR Activation pathway was among the most downregulated, with a z-score of −2.11, suggesting that suppression of lipid metabolism and nuclear receptor signaling may also contribute to enhanced rAAV production.

Next, we sought to establish an alternative strategy for exogenous CEBPA expressions using cDNA-based constructs. This approach eliminates the need for the sgRNA-Cas9 machinery to overexpress CEBPA gene, which has been shown to hamper rAAV yield (Fig. 1D) which is particularly relevant for manufacturing applications where higher titers are required. CEBPA gene encodes multiple isoforms, including full-length (p42) and an extended (p46) variant with different functions in myeloid differentiation [26], raising the question of which isoform is endogenously induced by the sgRNA approach (Fig. 4A). To address this, we expressed both isoforms individually in PCL cells by lentiviral transduction and compared their expression profiles to those induced by CEBPA sgRNA. Immunoblot analysis revealed that the sgRNA predominantly induces the full-length p42 isoform (Fig. 4B), indicating that this is the physiologically relevant variant under endogenous activation that triggers higher production of rAAV. Importantly, although both CEBPA isoforms led to increased virus production compared to the parental cell line, the p42 isoform consistently resulted in higher titers (Fig. 4C), reaching productivity above 10¹¹ viral genomes per mL (p42 isoform: 1.3 × 10¹¹ ± 0.3 × 10¹¹ vg/mL vs. p46 isoform: 0.8 × 10¹¹ ± 0.09 × 10¹¹ vg/mL, respectively). A GFP-overexpressing control cell line showed titers comparable to the unmodified parental line (0.3 × 10¹¹ ± 0.03 × 10¹¹ vs. 0.4 × 10¹¹ ± 0.2 × 10¹¹ vg/mL respectively) confirming that the enhancement of rAAV production is specifically mediated by CEBPA overexpression.

Fig. 4.

Functional characterization of CEBPA isoforms reveals full-length variant as key driver of rAAV productivity. A) Schematic representation of the three CEBPA protein isoforms generated from a single mRNA via alternative translation initiation codons. These isoforms differ in subcellular localization and interaction profiles, influencing their functional roles. B) Western blot analysis showing that sgRNA-mediated CEBPA overexpression in PCLs primarily induces the canonical full-length isoform (p42-CEBPAoe). β-actin was used as a loading control. cDNA-based induction of CEBPA was validated in two independent experiments performed under distinct culture conditions. C) Volumetric rAAV2 productivity (vg/mL ± SD) in PCLs overexpressing the CEBPA full isoform compared to the original PCL. Cells were harvested 66 hours post-infection and rAAV production quantified by qPCR of DNase-resistant viral genomes. Data represent mean ± SD of a minimum of n = 3 experiments. Statistical analysis was performed using unpaired t-test with Welch’s correction (*: p < 0.05; **: p < 0.01)

To further evaluate the production potential of CEBPA-overexpressing cells for rAAV manufacturing, we measured vector yields across a range of cell concentration at infection (CCI), in shake flask (Fig. 5A). Interestingly, these cells demonstrated a marked sensitivity to the cell density effect—a key limitation in upstream bioprocessing where increased cell concentrations often result in reduced specific productivity and compromised vector quality [27]. At low CCI (0.5 × 10⁶ cells/mL), hereon named low cell density (LCD), CEBPA-overexpression led to cell-specific production yield approximately three-fold higher than control (parental) cells (2.7 × 10⁵ ± 0.7 × 10⁵ vg/cell vs. 1.0 × 10⁵ ± 0.4 × 10⁵ vg/cell, respectively). However, as CCI increased (2 × 10⁶ cells/mL), hereon named high cell density (HCD), the CEBPA-overexpressing PCLs cell-specific production yield was lower than those observed in the parental line (0.05 × 10⁵ ± 0.01 × 10⁵ vg/cell vs. 1.0 × 10⁵ ± 0.3 × 10⁵ vg/cell, respectively), representing a decrease of up to twenty-fold.

Fig. 5.

Impact of CEBPA overexpression on rAAV productivity and metabolic profiles across different cell densities. (A) rAAV2 cell-specific production yield (vg/cell ± SD) in parental PCL and CEBPA-overexpressing PCL (PCL+_p42CEBPAoe) at two different cell concentration at infection (CCI): 0.5 × 10⁶, and 2 × 10⁶ cells/mL. Data represent mean ± SD of a minimum of n = 2 experiments. Statistical analysis was performed using unpaired t-test with Welch’s correction (***: p < 0.001). (B) Time-course of lactate accumulation (mM) in PCL (top) and PCL+_p42CEBPAoe (bottom) cultures at the same CCI. Data represent mean ± SD of a minimum of n = 2 experiments. (C) Comparison of metabolite consumption and production rates (mM/10⁶ cells/hr) at 0.5 × 10⁶ cell/mL for lactate (Lac), glucose (glc), glutamine (Gln), and ammonia (NH₃) between PCL (grey bars) and PCL+CEBPA overexpression (blue bars). Data represent mean ± SD of at least n = 3 independent experiments

Consistent with the findings of Xue et al. (2024), which identified lactate accumulation as a major inhibitor to rAAV production in high-density HeLaS3-based systems, CEBPA-overexpressing cells exhibited rapid lactate buildup under high cell density (HCD) conditions. Within 24 hours post-infection (hpi), lactate concentrations doubled the threshold of 10 mM previously reported to inhibit rAAV production, with CEBPA-overexpressing cells reaching 19.0 ± 0.9 mM compared to 10 ± 1.2 mM in parental cells (Fig. 5B). Furthermore, lactate production rates in CEBPA-overexpressing cultures were approximately 58% higher than those in the parental cell line (−0.65 ± 0.02 mM/10⁶ cells h⁻¹ vs -± 0.0120.43 mM/10⁶ cells h⁻¹) (Fig. 5C), accompanied by an elevated Glc consumption rate, indicating a metabolic shift associated with CEBPA expression. Ammonia and Gln production rates, however, remained similar between both cell lines.

To determine whether this phenotype was specific to cDNA-based overexpression or intrinsic to CEBPA activity, we repeated the experiment using CRISPR-mediated activation with a CEBPA-targeting sgRNA (sgRNA2; Supplementary Fig. 5A). Similar to the cDNA-based approach, CRISPR-activated cells exhibited a marked decline in rAAV titers with increasing cell density. When normalized per cell, the titers drop was even more pronounced (Supplementary Fig. 5B), confirming that the density-dependent phenotype is a robust and reproducible feature of CEBPA activation, independent of the expression method.

Perfusion-based process supports high-titer rAAV production in CEBPA-overexpressing cells

All these findings suggest that CEBPA overexpression induces a hypermetabolic state that, while beneficial for rAAV production under optimal conditions, may exacerbate metabolic stress under high-density culture, thereby limiting scalability. Building on these insights, we hypothesized that a perfusion-based production platform could mitigate the density-dependent phenotype observed in CEBPA-overexpressing cells and enable high-titer rAAV manufacturing. To test this, rAAV2 production using CEBPA overexpression cells was carried out at a CCI of 2 × 10⁶ cells/mL in a 1 L stirred-tank bioreactor operated under perfusion mode during the infection phase, at a variable perfusion rate to maintain lactate concentration similar to that observed in an LCD experiment [20]. In parallel, two control conditions were evaluated: a HCD (2 × 10⁶ cells/mL) shake flask culture, and an LCD (0.5 × 10⁶ cells/mL) bioreactor, both operated in batch mode. While cell viability and viable cell concentration progressively declined throughout infection in all conditions (Fig. 6A upper panel), volumetric yield in the perfusion bioreactor peaked at 48 hpi, reaching 4 × 10¹¹ vg/mL ±0.2 × 10¹¹ (Fig. 6A middle panel). As expected, HCD shake flask culture, operated in batch mode without medium replacement during infection, yielded only 9 × 10⁹ vg/mL—representing a 60-fold decrease compared to the perfusion HCD process. Cell-specific rAAV productivity in the perfusion bioreactor (8 × 10⁵ vg/cell) remained within the same order of magnitude as that observed in the LCD batch bioreactor (4 × 10⁵ vg/cell) (Fig. 6B), indicating that perfusion effectively mitigates the cell density-associated decline in productivity observed in CEBPA-overexpressing cells. In fact, the lactate concentration profile under perfusion closely resembled that of the LCD batch bioreaction (Fig. 6A, bottom panel). Moreover, rAAV particles generated from PCL overexpressing CEBPA under perfusion HCD were evaluated for Critical Quality Attributes and found to be comparable to that obtained with the parental PCL line. This was supported by similar levels of rAAV infectious activity measured in purified particles generated by both cell lines (Fig. 6C). Additionally, the full-to-empty capsid ratio remained around 60% in both cases (Fig. 6D vs Supplementary Fig. S2). These results demonstrate that CEBPA overexpression enhances rAAV titer without negatively impacting vector quality, supporting its applicability in scalable manufacturing settings.

Fig. 6.

Perfusion bioreactor enables high-titer rAAV2 production in CEBPA-overexpressing cells by mitigating density-associated metabolic stress. (A) Upper panel: cell viability (%) (solid lines) and viable cell concentration (dashed lines) over time in batch bioreactor at low cell density (LCD, triangles), perfusion bioreactor at high cell density (HCD, circles), and shake flask at HCD (squares). Middle panel: time-course of rAAV2 titers (vg/mL) for the same three conditions. Data presented as mean ± SD of technical replicates. Lower panel: lactate concentration (mM) over time in the three culture conditions. (B) Cell-specific production yield (vg/cell) of rAAV2 at 48 h post infection in batch LCD, perfusion HCD, and shake flask HCD cultures. C) Infectious rAAV titers (Fold change of rAAV infectivity produced by cells overexpressing CEBPA vs the PCL). rAAV particles from parental PCL and _p42CEBPAoe cells harvested in the perfusion bioreactor were purified and used to infect RC32 cells, and infectious particles were quantified based on GFP expression detected by fluorescence microscopy. Data represents the Fold change ± SD from a minimum of n = 3 independent experiments. (D) Capsid quality analysis by mass distribution in HCD perfusion cultures, analyzed by mass photometry at 66 hours post-harvest using SamuxMP mass photometer equipment. The x-axis represents molecular mass in kilodaltons (kDa). Each histogram displays two distinct populations corresponding to empty and full particles

Discussion

rAAV vectors have become a cornerstone of gene therapy due to their safety profile and ability to mediate long-term gene expression. However, the scalability and cost-effectiveness of rAAV manufacturing remain major challenges, particularly as clinical demand increases. Current platforms often have limited production yields, high production costs, and complex downstream processing. To address these limitations, there is a growing need for innovative strategies that enhance vector productivity and streamline bioprocessing [5], alongside to the development and adaptation as industry standard of robust, scalable manufacturing platforms. In this study, we employed an unbiased genome-wide CRISPR-based genetic screening approach to systematically identify host cell factors that influence rAAV production in HelaS3 based production systems. This production system has been shown to be scalable and with a reduction in manufacturing costs when compared to other platforms [3].

Through this approach, we identified several signaling pathways that play a pivotal role in enhancing rAAV production. IPA analysis highlighted S100 family signaling as the most significantly enriched pathway. This family of calcium-binding proteins is known to regulate key cellular processes such as apoptosis, differentiation, energy metabolism, and inflammation [28]. Notably, MAP kinase signaling and calcium homeostasis, both of which were also enriched in gain of function genetic screen, are functionally linked to S100 protein activity. The role of calcium signaling in viral infection is particularly compelling. Viruses are known to manipulate host calcium dynamics by interacting with membrane-associated proteins, thereby altering intracellular calcium levels [29]. These perturbations can either promote or inhibit calcium influx, ultimately influencing cytosolic calcium concentrations. Such changes can disrupt host cell homeostasis in ways that favor viral replication. Targeted modulation of calcium signaling or its downstream effectors may therefore represent a promising strategy to further enhance rAAV manufacturing efficiency.

In addition to metabolic and stress-related pathways, the genetic screen also revealed a strong enrichment of immune-related signaling, particularly those associated with interferon responses and cytokine production. These findings suggest that the host immune response to AdV infection may play a critical role in shaping the cellular environment for rAAV production. Infection with wild-type Ad5, is known to trigger robust inflammatory responses, including the upregulation of cytokines and chemokines such as IL-1, IL-6, TNF-α, IFN-α/β, IFN-γ, and IL-8 [30], with IL-1 acting as a central regulator [31]. Recognition of AdV by cytosolic DNA sensors, such as cyclic GMP-AMP synthase, activates downstream signaling cascades including type I interferon production and NF-κB-mediated proinflammatory cytokine expression [32, 33]. Another group of pathways identified involves cell death regulation. Given the lytic nature of AdV, it is expected that infection induces apoptosis or other forms of programmed cell death. Among the enriched apoptotic pathways, MYC signaling emerged as a notable hit. MYC has been previously implicated in modulating viral susceptibility, including its role in enhancing Epstein-Barr virus infection in Burkitt lymphoma cells [34]. Altogether, these observations highlight several potential pathways for cell engineering strategies aimed at enhancing rAAV production while modulating host susceptibility to virus-induced cytotoxicity.

CRISPRa screen also identified CEBPA as the most significantly enriched gene associated with enhanced rAAV production. This gene is essential for the normal development of granulocytes, and its disruption impairs the maturation of myeloid progenitors, contributing to leukemogenesis. Notably, germline CEBPA mutations have been identified as a cause of familial acute myeloid leukemia, where an inherited mutation is complemented by a somatic mutation in leukemic cells [35]. Therefore, although CEBPA plays a well-established role in immune responses and hematopoietic regulation, no prior involvement of this gene in rAAV production has been reported to date. Interestingly, CEBPB, another member of the CEBP family, also emerged as a hit, albeit with a lower fold change. Both genes were validated and their overexpression had positive impact in rAAV titer, while maintaining product quality. Importantly, these results also provide functional validation of the genetic screen and support the conclusions from the IPA analysis described above, opening new avenues for future host engineering with improved productivity.

Mechanistically, CEBPA overexpression induced broad transcriptional reprogramming, as evidenced by RNA-seq and principal component analysis. The transcriptomic shift included strong upregulation of immune- and stress-related genes, such as HP and SERPINB2, which are associated with inflammatory responses and cellular stress adaptation. These changes suggest that CEBPA may enhance rAAV production by modulating host pathways that support viral replication and assembly.

The functions of the pathways activated by CEBPA are largely consistent with those identified as enriched in the genetic screen, particularly regarding immune response and cell death pathways. An interesting pathway clearly downregulated in the CEBPA-overexpressing cell lines is LXR/RXR signaling, which is also consistent with the activation of the LPS/IL-1-mediated inhibition of the RXR pathway. Liver X receptors (LXR) form heterodimers with retinoid X receptors (RXR) and regulate gene networks controlling cholesterol and lipid metabolism, as well as inhibiting proinflammatory gene expression in macrophage [36]. Previous studies have shown that activation of the LXR can inhibit infection by HSV-1 or MLV-VSV-GFP in HepG2 cells or macrophages, and that knocking out the genes responsible for LXR production increased cell susceptibility to HSV-1 infection [37]. This could be a potential mechanism for how CEBPA improves rAAV production, specifically by redirecting cellular resources towards viral production. Moreover, the broad range of pathways affected by CEBPA, spanning metabolic, immune, and stress-response networks, suggests that its role is not limited to a single process. These observations support the interpretation of CEBPA as a global regulator influencing multiple cellular functions that may indirectly enhance rAAV biogenesis.

From a bioprocessing perspective, these findings have direct implications for rAAV manufacturing. CEBPA-overexpressing cells achieved a three-fold increase in both volumetric productivity and cell-specific production yield under LCD infection conditions compared to the parental cell line. While increasing cell density is a common strategy to boost volumetric productivity, the cell density effect of CEBPA-expressing cells presented a challenge. The higher lactate production—exceeding inhibitory thresholds for rAAV production as reported by Xue et al. (2024)—would hinder its applicability to rAAV manufacturing. To address this, we implemented a perfusion bioreactor enabling continuous removal of toxic byproducts and nutrient replenishment. This approach successfully increased rAAV titers, reaching up to 4x10¹¹ vg/mL. These results underscore the potential of CEBPA overexpression in host cells, particularly when integrated with process intensification strategies. Notably, although the perfusion bioreactor was operated at a cell density of 2 × 10⁶ cells/mL,—which is lower than the densities typically reported in previous studies using HeLaS3 and HEK293 cells for rAAV production (6 × 10⁶ to 10⁷ cells/mL) [38, 39]—we achieved high volumetric productivity in the CEBPA-expressing background, suggesting that further intensification, aligned with previous literature, could even yield higher volumetric productivities titers exceeding 1 × 10¹² vg/mL.

Conclusion

This study provides a comprehensive roadmap for the development of next generation rAAV production platforms. Through the integration of genome-wide screening, targeted cell engineering, and bioprocess optimization, we demonstrate that it is possible to overcome longstanding limitations in rAAV manufacturing. The identification of CEBPA as a central regulator of rAAV biogenesis not only advances our understanding of host-vector interactions, but together with the discovery of several additional pathways relevant to rAAV production, opens new avenues for the design of high-performance producer cell lines tailored for industrial-scale gene therapy applications.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

F.M., writing – original draft, investigation, methodology, visualization, and formal analysis. M.A., writing – original draft, investigation, methodology, visualization, and formal analysis. R.C., investigation, methodology, and formal analysis. A.R., methodology, P.G.A. supervision, project supervision, project administration, funding acquisition. P.M.A, supervision and funding acquisition. J.M.E., writing – original draft, project conceptualization, supervision, project supervision, investigation, funding acquisition. All authors: writing – review & editing.

Funding

This work was supported by the Research Unit UID/04462: iNOVA4Health – Programme in Translational Medicine, financially supported by Fundação para a Ciência e Tecnologia (FCT)/Ministério da Educação, Ciência e Inovação and the Associate Laboratory LS4FUTURE (LA/P/0087/2020) and Additional support was provided by the FCT research project (PTDC/BTM/ORG/1383/2020). J.M.E is funded by Stimulus of Scientific Employment, Individual Support program (2020.01216.CEECIND) from FCT and F.M by FCT PhD fellowship (2022.11494.BD) and DOI identifier http://doi.org/10.54499/2022.11494.BD.

Data availability

Requests for data, materials and methods shown in the main and supplemental figures are available from the lead contact ( [jose.escandell@ibet.pt](mailto:jose.escandell@ibet.pt)) upon reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors Filipa Moura, Mariana Antunes, Patrícia Gomes-Alves, Paula Marques Alves, and Jose Miguel Escandell are co-inventors on a provisional patent application (No. 119768) related to sequences targeting the gene CEBPA, which was identified as a key gene in this study.