Multi-omics reveal molecular changes during suspension adaptation of HEK293 cells Authors: Benyao Zhang, Shishi Li, Jingjing Liu, Wenhao Su, Xiaohuan Zhang, Xiuxiu Ren, Tingting Zhao, Qiufang Huang, Zihao Ge, Jiangbo Wei Published: 2026 ABSTRACT AbstractAbstractHuman embryonic kidney 293 (HEK293) cells have been successfully adapted from adherent to suspension culture and widely applied in both scientific research and the pharmaceutical industry. Although some studies investigated the variances between established adherent and suspension HEK293 cells of different strains, specific alterations in the cells during this consecutive process of suspension adaptation and possible factors driving this process have not been well described. Here, we adapted adherent HEK293 to suspension with desirable cell growth and high productivity for recombinant adenoviral vectors, and cells at several stages throughout the process were characterized. Slower cell growth, lower glucose uptake, increased lactate production, and weaker cell-surface adhesion were observed in suspension cells compared to their adherent counterparts. We further performed transcriptomics, proteomics, and metabolomics analysis to identify key cellular switches. A total of 2476 differentially expressed genes were found, including 1218 upregulated and 1258 downregulated genes in suspension cells. A similar and correlated pattern was observed in the proteomic study, and 702 differentially expressed metabolites were identified by untargeted metabolomics. In light of enrichment analysis, we summarized that HEK293 adherent cells survived and adapted to suspension culture via structural remodeling, metabolic shift and stress resistance. Our results provide a molecular enlightenment for suspension adaptation and potential directions for rational modification of HEK293 cell lines for future use.Key points• Suspension adaptation reduced adhesion and reshaped the HEK293 cytoskeleton.• Multi-omics revealed metabolic rewiring and enhanced stress resistance.• An optimized suspension line outperformed an internal HEK293 suspension reference.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1007/s00253-026-13912-z.Keywords: HEK293 suspension adaptation, Adenoviral vector, Transcriptomics, Proteomics, Metabolomics Introduction Cell culture refers to a biological process where cells are cultivated in an artificial and controlled in-vitro environment favoring their activities (Weiskirchen et al. 2023). According to the whereabouts of cells, cell culture can be classified as adherent or suspension culture. The most fundamental difference is that adherent cells require a surface or substrate to attach and live, a property known as anchorage dependency (Merten 2015). Alternatively, suspension cells live freely in a floating status without supporting material. Both culture types are widely used in basic research and the pharmaceutical industry, particularly in the bioproduction of proteins, viruses, and cells per se (O’Flaherty et al. 2020; Merten et al. 2011; van der Loo and Wright 2016; Lee and Chang 2024). Use of detaching agents and animal-derived nutrients could introduce extra expense, difficulties in downstream purification, and uncertainties in product safety and batch-to-batch stability (van der Valk et al. 2018). Along with limited space utilization, they become disturbing issues in adherent cell culture, especially in large-scale production. Comparatively, suspension cell culture offers scalability, operational simplicity and avoidance of animal-derived supplements (Biaggio et al. 2015). Some popular production cells are inherently in suspension, such as Spodoptera frugiperda clone 9, whereas others are adherent in nature, including Chinese hamster ovary (CHO), Madin Darby canine kidney (MDCK), Vero cells, and human embryonic kidney 293 (HEK293). Attempts have been made to transform these adherent cells into suspension to take the benefit. Not all of them are amenable to such a change (Wu et al. 2021; Wang et al. 2021; Rourou et al. 2019; Zhang et al. 2023), but HEK293 is a success. HEK293 cell line was established in 1973 and several derivatives have been developed from this lineage for various purposes (Graham et al. 1977). In many cases, they serve as production platforms to package and propagate recombinant adenoviral vectors (rAdV) conveying antigenic transgenes as virus-vectored vaccines targeting at many pathogens, including HIV-1 (Catanzaro et al. 2006), influenza virus (Sayedahmed et al. 2020), Ebola (Gélinas et al. 2019), Zika (Guo et al. 2018), tuberculosis (Smaill et al. 2013), and the most widely distributed adenoviral vaccine hitherto -ChAdOx1-nCoV-19 against severe acute respiratory syndrome coronavirus 2 (Joe et al. 2022; Mendonça et al. 2021; Voysey et al. 2021). Many suspension HEK293 production cells originate from adherent cells via suspension adaptation. Although most research employed a universal strategy by gradually weaning off the serum in the medium, they varied in the gradient of serum reduction and selection of replacement suspension culture medium (SCM) (Caron et al. 2018; Tsao et al. 2001; Jang et al. 2022; Lomba et al. 2021). These differences resulted in variance in cell performance and incomparability of relative results around the globe. Previous systems-level analyses have compared established adherent and suspension HEK293 derivatives and revealed molecular differences associated with cell adhesion, cell motility, cellular component organization, and cholesterol biosynthesis (Malm et al. 2020). However, these studies mainly focused on comparisons among existing HEK293 lineages rather than tracking the progressive intermediate states during serum withdrawal and suspension adaptation. Revealing those alterations will help to get a comprehensive understanding of suspension adaptation and provide a hint for genetic engineering of adherent cells into suspension, e.g., a crucial gene that assists the cell with adaptation. Modification to this particular gene could produce a suspension cell line with excellent homogeneity and stability in a genetic engineering way, rather than heterogeneity by suspension adaptation. Here, we followed a similar approach to adapt the HEK293 adherent cells to suspension by gradual reduction of the serum with a simultaneous medium substitution with four SCMs separately. We compared cell growth and titers of rAdV produced by these cells and screened out a well-adapted candidate with preferable performance for the following studies and future use. We then performed transcriptomic, proteomic and metabolomic analysis to uncover molecular changes in the adapted cells versus their adherent counterpart, and tried to provide some ideas to elucidate the mechanism of suspension adaptation. Materials and methods Adherent HEK293 cell culture HEK293 cells were purchased from the American Type Culture Collection (ATCC, Manassas, VA, USA; CRL-1573). Adherent cultures were maintained in tissue culture–treated T75 flasks in a humidified incubator (BPN-240RHP, Yiheng Scientific Instruments Co, Shanghai, China) at 37 °C in a 5% carbon dioxide (CO₂), 80% relative humidity (RH) atmosphere. Cells were fed in Dulbecco’s Modified Eagle Medium (DMEM, 10566-016, Gibco, Waltham, MA, USA) supplemented with 10% (v.\v.) fetal bovine serum (FBS, 13011-8611, Tianhang Biotechnology, Huzhou, China) and 1% (v.\v.) penicillin–streptomycin (PS, GNM15140-1, Genom Biotech, Hangzhou, China). When cultures reached 80–90% confluence, cells were transferred into new flasks at (4 ± 0.5) × 104 cells/cm2 as seeding density for subculture. An optical microscope (IX51, Olympus Corporation, Tokyo, Japan) was used to view and capture the cells with an accessory software (cellSens Standard, Olympus Corporation, Tokyo, Japan). Sequential adaptation of adherent cells Adherent cells were adapted to serum-free suspension culture by gradually reducing the serum in a 10%, 7%, 3%, 1%, and 0% stepwise gradient. Culture media with different serum concentrations were prepared by mixing the complete DMEM with the SCMs of our selection at defined ratios (v/v): 100% DMEM, 70% DMEM/30% SCM, 30% DMEM/70% SCM, 10% DMEM/90% SCM, 100% SCM, accordingly, and the SCMs to be screened were FreeStyle™293 (Gibco, Thermo Fisher Scientific, Waltham, MA, USA), BalanCD HEK293 (91165, FUJIFILM Irvine Scientific, Santa Ana, CA, USA), CD293 (11913-019, Gibco, Thermo Fisher Scientific, Waltham, MA, USA) and Celer-S001 (FG0104003, BioEngine Life Science, Shanghai, China). In this way, serum elimination and medium substitution were implemented at a synchronized pace. At each serum concentration, cells were maintained for at least three consecutive passages to fully accommodate the current condition. Cells were advanced to the next lower serum level when they reached approximately 80% confluence within 72 h as an indicator that cells had apparently recovered their growth capacity; otherwise, they were maintained at the current serum level until this criterion was satisfied. Suspension adaptation and suspension cell culture When HEK293 cells became stable in the serum-free adherent culture, they were transferred into 125 mL Erlenmeyer flasks at (0.8 ± 0.2) × 10⁶ cells/mL as initial seeding density in 30 mL working volume. Suspension cultures were incubated in an orbital shaker (ZWYC-290A, Zhicheng, Shanghai, China) at 130 rounds per minute (rpm, 25 mm shaking diameter) in a 37 °C, 8% CO₂, 80% RH environment. Cells were passaged every three to four days at the set seeding density. The population doubling time (PDT) was calculated as follows for each passage: where PDT is the population doubling time; t is the culture duration; Nt is the cell number at time point t; N0 is the initial seeding cell number. When PDTs became steady for three consecutive passages, the cells were considered successfully adapted to suspension culture. In some studies, a HEK293-SUS-CD374M (CD374M) cell strain was introduced as a reference. It is a suspension strain previously adapted from HEK293 in our lab by a direct adaptation approach, in which serum was weaned off from 10 to 0% directly and cells were transferred to orbital shaker for gradual adaptation. Cell counting, viability, clumping, and diameter Cell density and viability for both adherent and suspension cultures were determined using an automated cell counter (IC1000, Countstar BioTech, Shanghai, China). Adherent cells were counted after digestion, centrifugation and resuspension, while suspension cells were sampled and counted directly. 10 μL of cell sample was mixed evenly with 10 μL trypan blue stain solution (T10282, Invitrogen, Eugene, OR, USA) and loaded onto a counting chamber (M17, Countstar BioTech, Shanghai, China) for cell counting. Cell number, viability, clumping and diameter were analyzed automatically using the instrument’s proprietary software. Extracellular metabolite analysis The culture supernatants, which were collected from cell cultures after centrifugation at 2,000 g, 10 min, were examined using an automated biochemical analyzer M900 biochemical analyzer (Siemantec, Shenzhen, China) to determine the concentrations of glucose and lactate by oxidase enzyme electrode-membrane method. The glucose consumption rate and lactate accumulation rate were calculated using following equations: where qGlu is glucose consumption rate; qLac is lactate accumulation rate; VCD is viable cell density; dGlu and dLac are concentration changes of glucose and lactate in a given time period; and t is the time period. Cell adhesion assay To quantify the adhesion capacity of cells, adherent cells in 10% serum concentration cultured for 72 h were used to establish a standard curve. Cells were seeded into 96-well plates at an initial density of 4 × 104 cells/well, followed by 1:2 serial dilution to generate five concentrations in triplicates. After incubation at 37 °C for 4 h, the cells were considered tightly attached to the well surface and ready to construct a standard curve. For adhesion analysis, both adherent and suspension HEK293 cells were seeded at 1 × 104 cells/well with another two parallels per group. After incubation at 37 °C for 1 h, the supernatant was carefully removed, and non-adherent cells were therewith cleared away after two repeats of Dulbecco’s phosphate buffered saline (DPBS, 12338-018, Gibco, Grand Island, NY, USA) wash. Each well was then supplemented with 90 µL fresh DMEM and 10 µL water-soluble tetrazolium 8 (WST-8) from a Cell Counting Kit-8 (CCK-8) (40203ES60, Yeasen Biotechnology, Shanghai, China) and incubated for 2 h at 37 °C. The absorbance at 450 nm (OD450) was measured using a microplate reader (SpectraMax M5, Molecular Devices, San Jose, CA, USA). The number of adhered cells was calculated based on the standard curve to quantify adhesion capacity. Adenovirus infection and viral titering The productivity of suspension-adapted HEK293 cells for rAdVs were compared by getting infected with rAdV working stocks, which were established and preserved in a −70 °C cryogenic freezer for long-term storage in Weijiangbo Laboratory, National Vaccine and Serum Institute, at a multiplicity of infection (MOI) of 3 and cell density at 1.5 × 10⁶ cells/mL in 30 mL culture volume. The MOI is the quantitative ratio of infectious viruses against cells to be infected, which can be calculated by: where MOI is the abbreviation of multiplicity of infection; virus titer is the known infectious titer of rAdV working stocks; Vv is the volume of viruses; VCD is the viable cell density; and Vc represents the volume of cell suspension. Cultures were maintained on an orbital shaker at 37 °C, 8% CO₂, 80% RH, and 110 rpm in biological triplicates. The parameters for MOI, seeding density, harvest time, and cultivation conditions were based on design of experiment (DOE)-powered, unpublished data in our laboratory. Two rAdVs were tested here, specifically, a rAdV encoding herpes simplex virus type 2 (HSV-2) glycoprotein D (rAdV-gD2) as an HSV-2 vaccine and another one encoding rabies virus glycoprotein (rAdV-G) as a rabies vaccine. They were constructed on the basis of an Adeno-X™ Adenoviral System 3 (632,269, Takara Bio Inc., Kusatsu, Japan) and antigenic inserts, whose sequence information was gathered from the UniProt database (https://www.uniprot.org/uniprotkb/Q69467/entry for HSV-gD2 sequence and https://www.uniprot.org/uniprotkb/P03524/entry for rabies-G sequence). The detailed construction process is not discussed here. After 48 h post-infection, the mixed cultures of viruses and cells were transferred into 50-mL centrifuge tubes and subjected to three repeated cycles of “freeze-thaw” in liquid nitrogen and a 37 °C water bath to release viral particles. The crude rAdVs were harvested by collecting the supernatant after a 2000 g, 10 min centrifugation of cell lysates. Recombinant viruses were titered using Adeno-X™ Rapid Titer Kit (632,250, Takara Bio Inc., Kusatsu, Japan). Briefly, HEK293 cells in the logarithmic growth phase were digested, centrifuged, resuspended, and then seeded into 24-well plates at 2.5 × 105 cells/mL in 1 mL DMEM per well. While cells were maintained in the 37 °C incubator for 1 h, the virus samples were prepared by ten-fold serial dilution in DPBS, and 50 μL of each dilution was added to each well with three biological replicates in the cell-seeded 24-well plates. The supernatant was discarded after 48 h incubation, while the remaining cells were fixed with 0.5 mL methanol and incubated with mouse anti-hexon antibody and rat anti-mouse antibody one following the other for 1 h each, with two repeats of wash with 100 μL PBS + 1% bovine serum albumin (BSA) between these two steps. An aliquot of 250 μL 3,3' diaminobenzidine (DAB) working solution, formulated by 90% stable peroxidase buffer and 10% DAB substrate (10 ×), was added into each well for 10 min at room temperature (RT) for coloration. Following color development, positive cells with spots were observed and counted under a light microscope, and the titer was calculated. Transcriptomics: RNA extraction, library preparation, sequencing, and data analysis Cell samples containing 6 × 10⁶ cells were collected in the logarithmic growth phase from serum-free suspension-adapted cells, 0% serum adherent cells, 1% serum adherent cells, and 10% serum adherent cells. Independent biological triplicates were prepared for each condition. Total RNA was extracted using Trizol reagent (R0016, Beyotime Biotechnology, Shanghai, China) according to the manufacturer’s instructions. RNA integrity was assessed using an Agilent 5400 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA) and agarose gel electrophoresis. RNA-seq libraries were prepared using the NEBNext® Ultra™ RNA Library Prep Kit for Illumina® (E7770, New England Biolabs, Beverly, MA, USA). Library construction proceeded as follows: mRNA enrichment with oligo(dT) magnetic beads, random fragmentation, first and second-strand cDNA synthesis, end repair and A-tailing, adapter ligation, and polymerase chain reaction (PCR) amplification. Libraries were quantified using a Qubit 2.0 Fluorometer (Thermo Fisher Scientific, Waltham, MA, USA) and diluted to 1.5 ng/µL. Insert size distributions were assessed on an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). Reverse transcription quantitative PCR (RT-qPCR) further quantified libraries with expected insert sizes to determine the effective library concentration (≥ 1.5 nM). Qualified libraries were sequenced by an Illumina NovaSeq 6000 sequencing platform (Illumina Inc., San Diego, CA, USA) using paired-end 150 bp reads, yielding at least 6 G clean data per sample. Raw reads were quality-controlled using fastp (v0.19.7) (Chen et al. 2018; https://github.com/OpenGene/fastp) with parameters (-g: enable polyG tail trimming; -q5: quality cutoff; -u50: maximum allowed percentage of unqualified bases; -n15: maximum allowed number of N bases; -l150: minimum read length after trimming) to obtain clean reads. Q20, Q30, and GC content were calculated for the clean reads, and all downstream analyses were performed based on the high-quality clean data. The filtered reads were aligned to the human reference genome Homo sapiens GRCh38/hg38 from Ensembl release 94 using HISAT2 (v2.2.1) (Kim et al. 2019a, b; https://daehwankimlab.github.io/hisat2/). The corresponding Ensembl GRCh38.94 GTF annotation file was used for gene-level quantification. Gene-level read counts were generated using featureCounts (v2.0.6) (Liao et al. 2014; https://subread.sourceforge.net/featureCounts.html) with paired-end, non-strand-specific counting. The main parameters included -t exon, -g gene_id, -s 0, -Q 10, -C, -B, and -p. Briefly, reads were counted at the exon level and summarized to the gene level according to the gene_id attribute in the GTF annotation file. Multi-mapping reads were excluded from counting by default. Gene expression levels were normalized to fragments per kilobase of transcript per million mapped reads (FPKM) for expression-level visualization. FPKM values were not used as input for differential expression analysis. Differential gene expression analysis was performed using the DESeq2 R package (v1.42.0) (Love et al. 2014; https://bioconductor.org/packages/DESeq2/). The featureCounts-derived gene count matrix was used as input for DESeq2. Genes with a total read count greater than 1 across all samples were retained for differential expression analysis. The DESeqDataSet was constructed using the design formula ~ condition, where condition represented the cell culture state. DESeq2 was run with default settings for size factor normalization, dispersion estimation, independent filtering, and Cook’s distance outlier detection. Differential expression was tested using the Wald test, and log2 fold-change shrinkage was not applied. p-values were adjusted for multiple testing using the Benjamini–Hochberg false discovery rate (FDR) method. Genes with |log2FoldChange|≥ 1 and adjusted p-value < 0.05 were considered significantly differentially expressed. Functional enrichment analyses, including Gene Ontology (GO; https://geneontology.org/) and Kyoto Encyclopedia of Genes and Genomes (KEGG; https://www.genome.jp/kegg/) pathway analyses, were conducted using the clusterProfiler R package (v4.8.1) (Yu et al. 2012; https://bioconductor.org/packages/clusterProfiler/). Pathways with FDR < 0.05 were considered significantly enriched. Gene set enrichment analysis (GSEA) was performed using the GSEA software (v4.3.2) (Subramanian et al. 2005; https://www.gsea-msigdb.org/) with parameters (-permute phenotype; -metric Signal2Noise; minimum gene set size: -set_min 15; maximum gene set size: -set_max 5000; number of gene sets plotted: -plot_top_x50). Proteomics: protein extraction, proteomics analysis, and data processing 2 × 10⁶ cell samples were collected in their logarithmic growth phase from serum-free suspension-adapted cells, 0% serum adherent cells, 1% serum adherent cells, and 10% serum adherent cells, respectively, with independent biological triplicates per condition. Cells were lysed by 2 mL protein lysis buffer (555899, BD Biosciences, San Jose, CA, USA), vortexed gently, and sonicated in an ice–water bath for 5 min. Lysates were centrifuged at 12,000 g for 15 min at 4 °C, and the supernatants were collected. 2 µL of 1 M dithiothreitol was added to the supernatant, and samples were incubated at 56 °C for 1 h for reduction, followed by a 2 min ice bath. 20 µL of 1 M iodoacetamide was then added, and samples were kept at RT in the dark for 1 h for alkylation. Protein concentrations were determined using a Bradford Protein Assay Kit (P0006C, Beyotime Biotechnology, Shanghai, China) according to the manufacturer’s instructions. For sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS–PAGE), 20 µg of each sample was loaded per lane on a 12% gel. Electrophoresis conditions were 120 V for 20 min for the stacking gel and 150 V for 50 min for the resolving gel. Gels were stained with Coomassie Brilliant Blue R250 (1610400, Bio-Rad, Hercules, CA, USA) and destained until bands were clear. For digestion, protein samples were adjusted to a final volume of 100 µL with BD protein lysis buffer. Trypsin and 100 mM tetraethylammonium bromide buffer were added, mixed, and incubated at 37 °C for 4 h for proteolytic digestion. The digestion was quenched by lowering the pH below 3 with formic acid, mixed, and centrifuged at 12,000 g for 5 min at RT. The supernatant was slowly passed through a C18 desalting column. The column was washed three times with wash solution (0.1% formic acid, 3% acetonitrile), and peptides were eluted with elution solution (0.1% formic acid, 70% acetonitrile). Eluates were collected and lyophilized for later analysis. Mobile phase A (99.9% water, 0.1% formic acid) and mobile phase B (80% acetonitrile, 0.1% formic acid) were prepared before the analysis. Lyophilized powder was dissolved in 10 µL of mobile phase A, centrifuged at 4 °C and 14,000 g for 20 min. Sample preparation was taken for sample loading and further liquid chromatography-mass spectrometry (LC–MS) analysis. Vanquish Neo UHPLC system with a C18 precolumn (174,500, 5 mm × 300 µm, 5 µm, Thermo Fisher Scientific, Waltham, MA, USA) was heated at 50 °C in the column oven. The C18 analytical column was ES906 (PepMap™ Neo UHPLC, 150 µm × 15 cm, 2 µm, Thermo Fisher Scientific, Waltham, MA, USA). LC elution conditions were not described here. An Orbitrap Astral mass spectrometer with an Easy-Spray (ESI) ion source (Thermo Fisher Scientific, Waltham, MA, USA) was employed, set at 2.0-kV ion spray voltage and 290 °C ion transfer tube temperature. Data-independent acquisition (DIA) mode was chosen with primary mass spectrometry full-scan m/z acquisition range set to 380–980 and resolution of 240,000 (at 200 m/z). Automatic generation control (AGC), parent ion window size, DIA window count, and normalized collision energy (NCE) were set to 500%, 2 Th, 300, and 25%, respectively. The secondary mass spectrometer m/z range was 150–2000, with a secondary mass spectrometer resolution of 80,000 and a maximum injection time of 3 ms. Raw mass spectrometry data were processed using DIA-NN software (v1.8) (Demichev et al. 2020; https://github.com/vdemichev/DiaNN). Automatic mass calibration was applied for precursor and fragment ions. Carbamidomethylation of cysteine residues was set as a fixed modification, whereas N-terminal methionine loss was considered a variable modification. A maximum of two missed cleavage sites was allowed during peptide identification. To control false discoveries, peptide and protein identifications were filtered using a FDR threshold of 1%. Only peptides with Global.Q.Value < 0.01 and proteins with PG.Q.Value < 0.01 were retained for downstream analysis. Differentially expressed proteins (DEPs) were identified using thresholds of |log2FC|≥ 0.58 and FDR < 0.05. Functional annotation of identified proteins was performed using InterProScan (Jones et al. 2014; https://www.ebi.ac.uk/interpro/), integrating multiple databases including Pfam, PRINTS, ProDom, SMART, ProSite, and PANTHER. GO and KEGG pathway enrichments were conducted using the clusterProfiler R package. Volcano plots were generated using the ggplot2 (Wickham 2016; https://ggplot2.tidyverse.org/) package in R, and hierarchical clustering heatmaps were generated using the pheatmap (Kolde 2025; https://cran.r-project.org/package=pheatmap) package in R. Metabolomics: metabolites extraction, data processing, identification, and analysis Cell samples containing 1.5 × 107 cells from serum-free suspension-adapted cells, 0% serum adherent cells, 1% serum adherent cells, and 10% serum adherent cells each were collected in their logarithmic growth phase and frozen in liquid nitrogen; three biological replicates were analyzed for each condition. Cells were rapidly harvested by centrifugation (1000 g, 3 min, 4 °C). The supernatant was removed, and the cell pellets were immediately quenched in liquid nitrogen to minimize metabolic turnover. The samples were resuspended with pre-cooled 80% methanol by vortex, and then melted on ice and vortexed again for 30 s. After a 6 min sonification and centrifugation at 5000 rpm, 4 °C for 1 min, the supernatant was freeze-dried, redissolved with 10% methanol and was ready to use. Quality control (QC) samples were prepared by pooling equal aliquots from all samples and were injected periodically throughout the analytical run to monitor the stability and reproducibility of the liquid chromatography-mass spectrometry (LC–MS) system. After sample preparation, the solution was transferred into the liquid chromatography–tandem mass spectrometry (LC–MS/MS) system for subsequent analysis. A Vanquish UHPLC system (Thermo Fisher Scientific, Waltham, MA, USA), in conjunction with an Orbitrap Q Exactive™ HF mass spectrometer or Orbitrap Q ExactiveTMHF-X mass spectrometer (Thermo Fisher Scientific, Waltham, MA, USA) for UHPLC-MS/MS analysis. Samples were injected into a Hypersil Gold column (100 × 2.1 mm, 1.9 μm, Thermo Fisher Scientific, Waltham, MA, USA) in a 12 min linear gradient at 0.2 mL/min. The eluents were eluent A (0.1% formic acid in water) for positive polarity mode and eluent B (methanol) for negative polarity mode. The solvent gradient was set as: 2% B, 1.5 min; 2–85% B, 3 min; 85–100% B, 10 min; 100–2% B, 10.1 min; 2% B, 12 min. Q Exactive™ HF mass spectrometer was performed in positive/negative polarity mode with spray voltage on 3.5 kV, capillary temperature at 320 °C, sheath gas flow rate at 35 psi/aux gas flow rate of 10 L/min, S-lens RF (Radio Frequency) level at 60, 350 °C aux gas heater temperature. The raw LC–MS/MS data were converted to mzXML format using ProteoWizard (Kessner et al. 2008; http://proteowizard.sourceforge.io/) and then processed using the XCMS R package (Smith et al. 2006; https://bioconductor.org/packages/xcms/) for peak detection, peak alignment, retention time correction, and metabolite quantification. Peak detection, alignment, and quantification were performed using an in-house optimized XCMS workflow. Metabolite identification was performed by matching precursor mass, adduct ion information, and MS/MS fragmentation spectra against our laboratory-curated in-house MS/MS spectral database. A precursor mass tolerance of 10 ppm was applied, and features that did not meet the matching criteria were considered undetected. The in-house database integrates experimentally acquired MS/MS spectra from more than 15,000 commercial reference standards with curated high-quality MS/MS spectra from public metabolite databases, including HMDB (https://www.hmdb.ca/), RefMetaDB (https://www.metabolomicsworkbench.org/databases/refmet/), and ReSpect for Phytochemicals (http://spectra.psc.riken.jp/). The database contains more than 170,000 metabolites and over 1,000,000 high-quality MS/MS spectra. Following background noise elimination, the preliminary quantitative results were normalized to obtain relative peak areas according to the following formula: Features with missing values in more than 50% of samples within each group were removed. Compounds whose coefficient of variation (CV) of relative peak areas in QC samples was greater than 30% were further excluded. Final data handling was performed on a Linux operating system (CentOS version 6.6) using R and Python. Annotated metabolites were further cross-referenced with the KEGG database, the Human Metabolome Database (HMDB), and the LIPID MAPS database (https://www.lipidmaps.org/) for metabolite classification, pathway mapping, and biological interpretation. Multivariate statistical analyses, including principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA), were conducted to evaluate metabolic differences between groups. Differential metabolites were identified based on variable importance in projection (VIP) > 1 derived from the PLS-DA model, |log2FC|≥ 0.26, corresponding to FC ≥ 1.2 or ≤ 0.833, and adjusted P-value < 0.05. P-values were calculated using a two-tailed Student’s t-test and adjusted for multiple testing using the FDR method. Volcano plots were generated based on log2FC and -log10P-values using the ggplot2 package in R. Hierarchical clustering heatmaps were generated using z-score normalization of metabolite intensities and visualized using the pheatmap package in R (Kolde 2025; https://cran.r-project.org/package=pheatmap). Correlation analyses among differential metabolites were also performed in R to assess relationships between metabolites. Metabolic pathway enrichment analysis of differential metabolites was performed using the KEGG database. Pathways with FDR < 0.05 were considered significantly enriched. Transcriptome–proteome integrated analysis Raw RNA-seq count data were processed with DESeq2 package in R to obtain normalized expression values and gene-level log2FC and adjusted p-values. Protein intensity matrices were log2-transformed and median-normalized prior to analysis. Proteins quantified in fewer than two biological replicates per group were excluded from downstream analysis. For both datasets, PCA and hierarchical clustering were used to assess sample quality and identify outliers. Transcript identifiers were mapped to gene symbols/UniProt accessions using the biomaRt package in R (Durinck et al. 2009; https://bioconductor.org/packages/biomaRt/). Proteomics protein groups were mapped to the corresponding gene symbols when possible. The intersection set was defined and used as the basis for pairwise comparisons and joint modeling. Differential expression for transcripts was determined by DESeq2; differential proteins were identified using the statistical test described in the proteomics analysis. For consistency across omics layers, features were categorized as upregulated, downregulated or not significant using the aforementioned thresholds. For the intersection set, we computed pairwise correlations between transcript and protein. Pearson correlation coefficient and associated p-value were reported to assess linear concordance; Spearman rank correlation was calculated as a nonparametric robustness check. To summarize concordance and discordance at the single-feature level, we implemented a nine-quadrant classification: transcript log2FC (X-axis) and protein log2FC (Y-axis) were partitioned and generated nine sectors that captured concordant up/down, discordant, and nonresponsive categories. Counts per sector were reported and the distribution was tested where appropriate to assess enrichment of concordant vs discordant. To decompose shared and dataset-specific variation and to nominate joint molecular drivers, two-way orthogonal partial least-squares (O2PLS) modeling was applied to the intersection matrices. O2PLS modeling was performed using the OmicsPLS R package (Bouhaddani et al. 2018; https://cran.r-project.org/package=OmicsPLS), based on the O2PLS framework (Trygg and Wold 2003). Model performance metrics (R2X, variance of X matrix; R2Y, variance of Y matrix; Q2, predictive ability) and permutation testing were used to evaluate model robustness and to avoid overfitting. Features were ranked by their absolute joint loadings, and the top contributors were reported as candidate drivers of suspension adaptation. GO and KEGG enrichment analyses were carried out on sets derived from (i) concordant features, (ii) discordant features, and (iii) top O2PLS drivers to probe biological processes and pathways mostly associated with suspension adaptation. Statistical analysis Apart from omics studies whose analytical process has been explained in detail in corresponding sections, all other statistical analyses were performed using GraphPad Prism 10.1.2 (GraphPad Software, Boston, MA, USA). All experiments were performed with three independent biological replicates unless otherwise stated. The data were checked for normality using the Shapiro–Wilk test. For comparisons involving more than two groups, one-way analysis of variance (ANOVA) followed by Dunnett’s multiple comparison test was used. Viral titers were analyzed using two-way ANOVA with cell line and virus type as independent factors. p-values less than 0.05, **p-values less than 0.01, ***p-values less than 0.001, and ****p-values less than 0.0001 were considered statistically significant. Results HEK293 adherent cells were successfully adapted to suspension culture and showed different growth, glucose intake, and lactate generation We employed the sequential adaptation strategy to adapt adherent HEK293 cells to suspension cells by gradually decreasing the serum content from 10%, 7%, 3%, 1%, to 0% with medium replacement at the same time and thereafter transferred the cells to shaking flasks. We screened four commercially available suspension culture media for medium replacement, and cells survived in three of them, which were FreeStyle™293, BalanCD HEK293 and CD293. Survived cells were named in correspondence to their culture medium as HEK293-Sus-Freestyle (FS293), HEK293-Sus-BalanCD (BLCD293), and HEK293-Sus-CD293 (CD293). These cells behaved differently in growth performance, and adherent cells in different serum gradients in the adaptation process also exhibited different growth rates, as shown in Fig. 1A and. All cells, either adherent or suspension, were maintained to keep the same passage number to ensure comparability. Adherent cells manifested a similar growth kinetics. They were initially seeded at 1 × 105 cells/well in 6-well plates and stayed in the latent phase for nearly three days. From day 3, cells entered a logarithmic growth phase until day 7, when they reached a plateau phase due to nutrient depletion or density-dependent contact inhibition, characterized by a comparatively slower, even stationary growth. The density of serum-free adherent cells declined swiftly on day 8 and day 9, indicating a programmed cell death due to nutrient deprivation. HEK293 adherent cells were successfully adapted to suspension culture and showed different growth, glucose intake and lactate generation. A The cell growth curves of adherent cells. The upright legend of percentage referred to the serum concentration and also indicated the percentage of DMEM (multiplied by ten) in the total culture medium. A10: adherent cells in 100% DMEM (10% serum); A7: adherent cells in 70% DMEM/30% SCM (7% serum); A3: adherent cells in 30% DMEM/70% SCM (3% serum); A1: adherent cells in 10% DMEM/90% SCM (1% serum); A0: adherent cells in 100% SCM (0% serum). B The cell growth curve of suspension-adapted cells in three different suspension culture media. C Glucose consumption and lactate production of adherent cells in different serum concentration groups. D Glucose consumption and lactate production of suspension-adapted FS293 cells. E Glucose consumption rate of adherent cells in different groups after normalization In comparison to the adherent cells, adapted suspension cells showed different pattern with an earlier start of logarithmic growth phase from day 1 to day 5. Results also showed that the control CD374M cells achieved maximum cell density at approximately 7 × 10⁶ cells/mL, higher than the density of FS293, BLCD293 and CD293, around 4 × 10⁶ cells/mL. The PDTs within 72 h of FS293, BalanCD293, and CD293 were close to each other, 38.01 h, 39.14 h, and 39.22 h, respectively. Albeit lower than our previously established CD374M cell strain, their PDTs and their maximum density satisfied our production needs. The lower density might result from cell aggregation and incomplete adaptation, which could possibly be solved by addition of anti-clumping agents and protecting agents against shear force, continuing adaptation and optimization of culture parameters. Glucose consumption and lactate accumulation are strictly monitored metabolites during large-scale, cell-based bioproduction because glucose is a necessary carbon source for cells to build the cytoskeleton and proliferate, while on the other side, lactate is an undesirable chemical generated from the glycolysis pathway as a hazard to normal cell function and yield of the target product. Therefore, we were also interested in the glucose and lactate during the suspension adaptation as well as in the adapted suspension cells. We monitored the concentrations of glucose and lactate dynamically in the culture supernatant. As the cells were adapted using a gradual serum replacement method with medium substitution, the initial glucose concentrations in the culture media differed across experimental groups. The glucose concentration data for each group were normalized to eliminate this variability and enable an accurate comparison of glucose consumption. As illustrated in Fig. 1C,, and, the rate of glucose consumption demonstrated a significantly positive correlation with the serum concentration in the culture medium. The qGlu values for adherent cells in 10%, 7%, 3%, 1%, and 0% serum groups and suspension cells (FS293) were 0.27, 0.185, 0.134, 0.135, 0.266, and 0.15 mg/106cells×day-1, respectively. In other words, given the same seeding density, the cells exhibited a gradually slower glucose uptake as adaptation proceeded. Specifically, the 10% serum group (10%-GLU) exhibited the fastest glucose consumption rate, with glucose being nearly used up by day 9. As the serum concentration decreased, the rate of glucose consumption progressively slowed down, with the slowest rate observed in the serum-free group (0%-GLU). Meanwhile, the lactate accumulation rate showed a trend that was highly in a reverse manner to the glucose consumption rate. The gLac calculations were 0.256, 0.160, 0.107, 0.101, 0.198 and 0.072 mg/106cells×day-1, respectively. The 10% serum group had the highest rate of lactate production. As the serum concentration was reduced, the rate of lactate generation decelerated correspondingly, with the serum-free group producing the lowest amount of lactate. The qGlu and qLac data suggest that HEK293 cells cultured in 10% serum concentration were more active in metabolic activities. It was interesting to observe that suspension-adapted cells without serum nourishment had relatively high metabolic rates in glucose consumption and lactate generation. It might be reasoned that cells without serum need to promote their metabolism to accustom themselves to the serum-free environment, but this requires further investigation. Cells in each serum condition and suspension manifested differently in morphology and adhesion ability Cells underwent morphological changes during suspension adaptation. Supplementary Fig. S1 demonstrated that adherent HEK293 cells in DMEM containing 10% serum maintained an epithelium-like morphology, identified as irregular polygons with distinct boundaries. As the adaptation proceeded, some cells could not tolerate the serum-reducing culture, and as a consequence, they detached from the surface and took on a spherical appearance, floating in the medium. We tried to collect these floating cells and subculture them, but the cell viability was less than 80% at the time of collection and even became lower than 10% after passage, indicating that these floating cells could not adapt to the current condition. The epithelial exterior of adherent cells was kept identifiable until complete eradication of the serum component. In the serum-free condition, instead, living cells became more rounded, aggregated as a cluster, and loosened in intracellular and cell-surface connections with blurred boundaries (Supplementary Fig. S2). When transferred and sufficiently adapted to suspension culture from 0% serum adherent culture, most cells existed as spherical individuals, along with a small proportion of cell clumps. There was also an obvious change in cell diameter before and after the suspension adaptation, as shown in the size distribution charts in Supplementary Fig. S2, with a 15.65 μm in diameter of the original adherent cells versus the adapted suspension cells, which had an average diameter of 16.07 μm. Based on our perception of suspension adaptation, a change in the adhesive strength of cells to the surface might occur. Therefore, we performed CCK-8-based assays on adherent cells cultured in different serum concentrations and suspension-adapted cells to evaluate cell adhesion ability. The adhesion ability, as shown in Fig. 2, was defined as the number of cells that attached to the surface over a given time period and was calculated using a high-quality, linear-correlation standard curve whose coefficient of determination, R-squared, was greater than 0.99. Cells routinely cultured with 10% serum showed the highest adhesion (13,655 ± 3000.3), followed by 1% (10,095 ± 1382.8). The 3% serum group showed intermediate adhesion (8976.0 ± 735.4), whereas the 0% serum group exhibited lower adhesion (6988.9 ± 389.9), which was comparable to the value of the suspension cells (7396.5 ± 1892.0). Overall, cells cultured in 10% serum exhibited markedly greater adhesive capacity than cells cultured in low or serum-free conditions, and a loss of adhesive capacity was seen in the suspension cells and in the adherent cells with less serum component. Cells in each serum condition and suspension were different in adhesion ability. Data are presented as mean ± SD (n = 3 biological replicates). Cell adhesion capacity of different cells, as characterized by the number of cells that firmly attached to the plate surface after a set time period. A10: adherent cells in 100% DMEM (10% serum), A7: adherent cells in 70% DMEM/30% SCM (7% serum); A3: adherent cells in 30% DMEM/70% SCM (3% serum), A1: adherent cells in 10% DMEM/90% SCM (1% serum), A0: adherent cells in 100% SCM (0% serum); FS293: well-adapted suspension cells in Freestyle suspension culture medium. One-way ANOVA followed by Dunnett’s multiple comparison test; comparisons were made against A10 Suspension cells were capable of producing recombinant adenoviral vectors encoding different antigens To verify that suspension-adapted cells have practical value for pharmaceutical use as production cells, we compared the adenoviral production capacity of different suspension-adapted cells using two rAdVs. These two rAdVs had a similar adenoviral type 5 backbone but were inserted with different transgenes, rabies G and HSV-2 gD, respectively. The reason why two rAdVs were tested was that the size and properties of the insertion, as well as the final translated product, would influence cell function and cell productivity of the viral vectors, and we would like to screen out a preferable candidate that was capable of producing rAdV effectively and universally, regardless of the insertion sequence. Cell density was adjusted to 1.5 × 106 cells/mL and infected with rAdVs at an MOI of 3. The viruses were harvested 48 h post-infection and titered. The results showed that the FS293 cells achieved the highest yield for both rAdVs, approximately tenfold higher than that of the control CD374M cells, reaching about 9 × 10⁷ infectious unit (ifu)/mL (Fig. 3). Such productivity could meet our current production requirements, evidencing the feasibility of these adapted cells in real industrialized application, and the productivity could be further enhanced by modification of regulatory genes in viral backbone, adjustment of culture parameters, reformation of culture medium and so forth. Productivity of two recombinant adenoviral vectors by three suspension-adapted cells. rAdV-gD2: adenoviral vector encodes herpes simplex virus type 2 glycoprotein D as an HSV-2 vaccine; rAdV-G: adenoviral vector encodes rabies virus glycoprotein as a rabies vaccine. Data are presented as mean ± SD, n = 3 per group Adherent HEK293 cells and suspension-adapted cells had different gene transcriptions PCA analysis revealed a clear separation of samples according to serum concentration, indicating a progressive shift in the transcriptome as serum was gradually reduced. Adherent cells cultured with 10% serum (A10) formed a tight cluster, and fully suspension-adapted cells in FreeStyle™293 medium (SA_F) also clustered distinctly. Notably, A0 (adherent, 0% serum) samples localized closer to the SA_F cluster, suggesting that A0 represents a key transitional state from adherent to suspension phenotypes. Pairwise correlation analysis showed that most within-group correlations exceeded 0.97, whereas correlation of A1_2 with A1_1/A1_3 was 0.94, lower than the correlations in other groups (Supplementary Fig. S3A). We observed heterogeneity within the A1 group: two replicates (A1_1 and A1_3) clustered closely together, while A1_2 was displaced in PCA space (Supplementary Fig. S3B). These findings indicated that cells cultured at 1% serum was possibly in a highly dynamic transitional state with elevated transcriptional heterogeneity. It is also reasonable to explain this as a consequence of random error, and it needs more experiments to confirm. Hierarchical clustering of differentially expressed genes placed A0 in closer adjacency to SA_F, whereas A10 clustered with one A1 sample and the remaining two A1 replicates formed a separate cluster (Supplementary Fig. S3C). Overall, complete serum withdrawal drove adherent cells toward a transcriptomic state similar to suspension cells, while intermediate low-serum conditions produced a bifurcating population in which some cells retain a serum-like profile and others begin to acquire suspension-like features. To elucidate the transcriptional reprogramming underlying adaptation of HEK293 cells from adherent culture to suspension, we performed RNA-seq on three biological replicates per condition and conducted differential expression analysis. Using DESeq2 with thresholds of adjusted p ≤ 0.05 and |log₂FC|≥ 1, we identified 2476 differentially expressed genes (DEGs), of which 1218 were upregulated in suspension (49.1%) and 1,258 downregulated (50.9%), as illustrated in Supplementary Fig. S3D. The balanced distribution of upregulated and downregulated genes indicates bidirectional transcriptional adaptation rather than a unidirectional activation or repression. Differential gene enrichment analysis can relate these DEGs to specific biological functions, processes, and pathways, thereby contributing to our understanding of how these DEGs influence cell function and phenotype. There are several tools, i.e. GO, KEGG and GSEA enrichment analysis, to study and annotate the DEGs, each relying on particular databases and reflecting different aspects of genes. GO enrichment analysis categorizes all genes into three groups based on their functions: cellular component (CC), biological process (BP) and molecular function (MF). As the GO analysis indicated, the upregulated genes in suspension cells versus adherent cells were enriched in “cell–cell adhesion via plasma-membrane adhesion molecules” (GO:0098742), “synapse organization” (GO:0050808), and “regulation of monoatomic ion transmembrane transport” (GO:0034765). There were also upregulations in the pathways that involved in neural functions, such as “regulation of amine transport” (GO:0015837), “modulation of chemical synaptic transmission” (GO:0050804) and “axon development” (GO:0061564) (Fig. 4A). On the opposite side, down-regulated genes were annotated to “antigen binding” (GO:0003823), “receptor ligand activity” (GO:0048018), and “response to nutrient levels” (GO:0031667) (Fig. 4B). Adherent HEK293 cells and suspension-adapted cells had different gene transcription. Data are presented as mean ± SD, n = 3 each group. A and B SA_F vs A10, dot plots of GO enrichment analysis for upregulated (A) and downregulated (B) genes, showing the top 20 significantly enriched biological processes or cellular components. C and D SA_F vs A10, dot plots of KEGG pathway enrichment analysis for upregulated (C) and downregulated (D) genes. The size of the dot represents the number of DEGs enriched in the pathway/term (Count). The color of the dot corresponds to the adjusted p-value, with redder colors indicating higher significance. The X-axis (GeneRatio) represents the ratio of DEGs to the total number of genes in the specific pathway KEGG offered another perspective to identify these DEGs by their biological pathways. The KEGG results showed that upregulated DEGs were significantly enriched in pathways related to signal transduction and neural activities like “Taste transduction” (hsa04742), and down-regulated genes were mainly enriched in “cytokine-cytokine receptor interaction” (hsa04060), “TGF-beta signaling pathway” (hsa04350), and “biosynthesis of amino acids”(hsa01230) (Fig. 4C and). GSEA, on the other hand, provided a more comprehensive gene analysis beyond just examining individual genes, but considering the collective actions of functionally related genes, as a supplement to the GO and KEGG enrichment study. In light of the GSEA method, upregulated genes in suspension-adapted cells were enriched mainly in the “regulation of mitotic nuclear division pathway” (GO:0007088) and “intermediate filament cytoskeleton organization pathway” (GO:0045104). Meanwhile, down-regulated gene networks found in suspension-adapted cells were “cellular response to nutrient pathway” (GO:0031670), and “intrinsic apoptotic signaling pathway” (GO:0097193). Adherent HEK293 cells and suspension-adapted cells had different protein expressions To characterize proteomic changes associated with the transition of HEK293 cells from adherent culture to serum-free suspension culture, we performed DIA-based quantitative proteomics. PCA analysis (Supplementary Fig. S4A) showed that A10 and SA_F samples were clearly separated along PC1 (accounting for 32.72% of the variance) and exhibited good within-group reproducibility. Hierarchical clustering (Supplementary Fig. S4B) demonstrated consistent expression patterns of these differentially expressed proteins (DEPs) within groups. In total, 9991 proteins were identified, of which 806 were considered differentially expressed representing approximately 8.07% of the identified proteome: 414 proteins were upregulated and 392 downregulated in SA_F compared to A10 (|log2Fold Change|≥ 0.58, FDR ≤ 0.05) (Supplementary Fig. S4C). Likewise, DEPs can be interpreted more clearly by enrichment analysis, as demonstrated in Supplementary Fig. S5A and Supplementary Fig. S5B. DEPs were significantly enriched in all three aspects of GO classification: BP, CC, and MF. In terms of BP, up-regulated DEPS were mainly annotated to specific biological processes, including “regulation of cell growth” (GO:0001558), “cell adhesion” (GO:0007155), “oxidation–reduction process” (GO:0055114), and “lipid biosynthetic process” (GO:0008610). Proteins associated with cellular components were differentially expressed, particularly those structural proteins in the extracellular matrix (GO:0031012), extracellular region (GO:0005576), integral component of membrane (GO:0016021) and intermediate filament (GO:0005882). There were also enrichments in glycosaminoglycan binding (GO:0005539), metal ion binding (GO:0046872), and endopeptidase activity (GO:0004175), implying that suspension-adapted cells acclimatized themselves to suspension culture with the assistance of differential expression of some proteins to affect molecular functions. Protein downregulations were clustered into several functions, namely, transport, localization, transmembrane transport (GO:0022857), ion transport (GO:0006813) and immune response-related (GO:0006955) processes in BP; plasma membrane (GO:0098797), integral component of membrane (GO:0016021), extracellular region (GO:0005576) in CC; and transporter activity (GO:0005215), substrate-specific transmembrane transporter activity (GO:0022891), ion binding and metal ion transmembrane transporter activity (GO:0046873) in MF aspect. Tracking down these DEPs via KEGG, proteins participating in cell adhesion molecules, extracellular matrix (ECM)-receptor interaction (map04512), focal adhesion (map04510) and PI3K-Akt signaling pathway (map04151) were differentially expressed in an upregulated manner. Down-expressed proteins were significantly clustered in arrhythmogenic right ventricular cardiomyopathy (map05412), hypertrophic cardiomyopathy (map05410), bile secretion (map04976) and salivary secretion (map04970). These pathways were seemingly irrelevant to cell suspension adaptation; however, it is judicious to notice that ARVC and HCM pathways are involved with cellular skeletons and connexins, and consequently regulate the cell morphology and resistance to mechanical forces; bile secretion and salivary secretion pathways play a regulatory role in membrane transport, secretion and endocytosis of chemicals and nutrients (Fig. 5). KEGG enrichment analysis of differentially expressed proteins between suspension-adapted cells versus adherent cells. A and B SA_F vs A10, up- (A) and down- (B) regulations of differentially expressed proteins in KEGG enrichment. The x-axis represents the ratio of the number of differentially expressed proteins associated with this pathway to the number of background proteins associated with this pathway, the larger the value, the further to the right the point lies, indicating a higher degree of enrichment of differentially expressed proteins in that pathway. The color of the point corresponds to the log10-transformed P-value of the enriched pathway; the redder the color, the smaller the P-value, indicating greater reliability of the test and greater statistical significance of the result. The size of the point reflects the number of differentially expressed proteins in the corresponding pathway; the larger the point, the greater the number of differentially expressed proteins within that pathway Adherent HEK293 cells and suspension-adapted cells were different in metabolic activities To further characterize metabolic shifts of HEK293 cells following adaptation to serum-free suspension culture, we performed untargeted metabolomics analysis on SA-F and A10 cells (n = 3 per group). PCA revealed a distinct separation between the SA-F and A10 groups in both negative (NEG) (Supplementary Fig. S6A) and positive (POS) (Supplementary Fig. S6B) ion modes. A total of 2,742 metabolites were identified using LC–MS/MS. Differential accumulated metabolites (DAMs) were screened based on the criteria (FDR < 0.05, VIP > 1.0, and FC > 1.2 or FC < 0.833). Under these thresholds, 275 DAMs were identified in NEG mode (118 upregulated, 157 downregulated; Fig. 6A) and 427 DAMs in POS mode (123 upregulated, 304 downregulated; Fig. 6B). When enriched via KEGG database, five pathways differed in different degrees between SA-F and A10 cells, as shown in Table 1. Volcano plots provide a visual representation of the overall distribution of differential metabolites. SA_F, suspension-adapted HEK293-SA-FS cells; A10, adherent cells in 10% serum environment. The x-axis represents the change in expression fold change (log2FC) of metabolites across different groups, while the y-axis represents the level of statistical significance (-log10(p-value)). Each point in the figure represents a metabolite, with the size of the point indicating the variable importance in projection (VIP); significantly up-regulated metabolites are represented by red points, and significantly down-regulated metabolites by blue points. A and B Differential accumulated metabolites in NEG mode (A) and POS mode (B) Table 1Altered metabolic pathways between suspension-adapted cells and parental adherent cells based on KEGG enrichmentMapIDMapTitleFDRxynNmap00061Fatty acid biosynthesis0.0069454449166map00591Linoleic acid metabolism0.0241855749166map04216Ferroptosis0.0631474649166map05207Chemical carcinogenesis-receptor activation0.0066944456192Map00982Drug metabolism-cytochrome P4500.0610214656192Note: The first two rows were generated from metabolites detected in negative ion mode (NEG), whereas the last three rows were generated from metabolites detected in positive ion mode (POS). Therefore, the KEGG−annotated differential metabolite set and background metabolite set differed between ion modes, resulting in different n and N values across pathways. MapID, the ID of the enriched KEGG pathway in KEGG database; MapTitle, the name of the enriched KEGG pathway; FDR, adjusted p−value; x, the number of differentially expressed metabolites associated with this pathway; y, the number of background (all) metabolites associated with this pathway; n, the number of KEGG−annotated differentially accumulated metabolites; N, the number of KEGG−annotated background (all) metabolites Altered metabolic pathways between suspension-adapted cells and parental adherent cells based on KEGG enrichment Note: The first two rows were generated from metabolites detected in negative ion mode (NEG), whereas the last three rows were generated from metabolites detected in positive ion mode (POS). Therefore, the KEGG−annotated differential metabolite set and background metabolite set differed between ion modes, resulting in different n and N values across pathways. MapID, the ID of the enriched KEGG pathway in KEGG database; MapTitle, the name of the enriched KEGG pathway; FDR, adjusted p−value; x, the number of differentially expressed metabolites associated with this pathway; y, the number of background (all) metabolites associated with this pathway; n, the number of KEGG−annotated differentially accumulated metabolites; N, the number of KEGG−annotated background (all) metabolites Integration analysis of differentially expressed genes and differentially expressed proteins of adherent HEK293 cells and suspension-adapted cells Linking transcriptomics and proteomics depicts a more comprehensive picture of the complete process and mechanism of gene transcription to its downstream expression of the actual executor in biological processes- the proteins, and pinpoints how the RNAs and proteins influence cell functions. To characterize the regulatory relationship between gene transcription and protein expression during the adaptation process from adherent cells (A10) to suspension cells (SA_F), we performed an integrated analysis of the transcriptomic and proteomic datasets. Supplementary Fig. S7A provides a detailed breakdown of differentially expressed molecules: the transcriptomic analysis identified 1,258 downregulated and 1,218 upregulated genes, whereas the proteomic analysis revealed 392 downregulated and 414 upregulated proteins. In accordance with the Central Dogma of molecular biology, which explains the correlation between genes and proteins, we performed a Pearson correlation analysis on the log2FC of the co-quantified molecules to evaluate the extent to which gene transcription influences protein expression (Crick 1970). We assessed the differential expression statistics across both omic layers. A substantial number of genes and proteins were quantified. Through intersection analysis of the quantified proteins and genes, we identified a set of co-quantified molecules, among which 130 exhibited significant differential expression in both datasets (Supplementary Fig. S7B). As illustrated in Fig. 7A, the transcriptome and proteome profiles for the SA_F vs. A10 comparison exhibited a weak positive correlation trend. This relatively low correlation coefficient suggested that, while transcription served as the basis for protein expression, extensive and significant post-transcriptional or translational regulations occurred during the suspension adaptation of HEK293 cells. Consequently, changes in mRNA abundance did not linearly translate into changes in protein abundance. It could also be technical factors including non-absolute quantification in proteomics, peptide detectability, and ionization-related measurement bias. The Nine-Quadrant Plot (Fig. 7B) offers a more detailed characterization of the relationship between gene transcription and protein expression: Co-regulations were presented as dots located in the red (Quadrant 3) and dark blue (Quadrant 7) regions, which represent genes with matched trends in mRNA and protein levels. Dots located in the green (Quadrants 1, 2, 4) and orange (Quadrants 6, 8, 9) regions represented genes where mRNA and protein changes were inconsistent, as discordant regulation. Notably, the presence of the orange region (mRNA downregulated or unchanged, but protein upregulated) suggests that cells were likely to maintain the homeostasis of key proteins, such as certain stress-response proteins, by enhancing translation efficiency or inhibiting protein degradation. Conversely, the green region (mRNA upregulated, but protein unchanged or downregulated) implied potential translational stalling or accelerated protein degradation, which aligned with the rapid clearance of obsolete structures during cytoskeletal remodeling. Integrated transcriptomic and proteomic analysis reveals the regulatory mechanisms of HEK293 suspension adaptation. Tran, transcriptomic; Prot, proteomic. A Scatter plot shows the correlation of log2FC between the transcriptome (Y-axis) and proteome (X-axis) for co-quantified molecules. The red line represents the linear regression fit, and the Pearson correlation coefficient (R) is 0.249. Green dots indicate molecules significantly differentially expressed in both the transcriptome and proteome datasets, whereas blue dots represent other co-quantified molecules. B Nine-quadrant plot characterizes the correlation of expression changes. Red/dark blue dots (quadrants 3 & 7) indicate co-regulated genes; orange/green dots (quadrants 1, 2, 4, 6, 8, 9) indicate discordant regulation (e.g., orange: protein-level upregulation only; green: transcript-level upregulation only); gray dots (quadrant 5) indicate non-significant molecules. Dashed lines represent the fold-change thresholds. joint loadings plot identifies key drivers contributing to the covariance between the two omics layers. Triangles represent proteins, and circles represent transcripts; markers far from the origin indicate high contribution weight To further identify the key driving factors contributing to the joint variation between the transcriptome and proteome, we employed two-way O2PLS analysis. Figure 7C displayed the top molecules with the highest contribution to the covariance between the two omics’ layers. At the proteome level, we found that proteins such as heat shock protein 10 (Hsp10), a mitochondrial chaperonin indicating energy metabolism and stress response, Na +/K + ATPase alpha 1, which directs ion transport function, and macrophage migration inhibitory factor (MIF) exhibited extremely notable values, indicating that they were core proteins distinguishing the SA_F state from the A10 state. At the transcriptome level, multiple genes identified by ENSG IDs, such as ENSG00000211459 (mitochondrially encoded 12S ribosomal RNA, MT-RNR1), ENSG00000109971 (Hsp8), ENSG00000229807 (X-specific transcript), and ENSG00000169710 (fatty acid synthase), also demonstrated high contribution weights. These high-loading molecules might serve as key regulators, driving the metabolic and structural remodeling during the suspension adaptation process. We then performed a joint GO and KEGG enrichment analysis on the co-differentially expressed molecules to elucidate how transcriptional regulation translated into final protein functional execution. The joint GO analysis (Supplementary Fig. S8A) revealed key features of suspension adaptation across three dimensions: BP, CC, and MF. In terms of BP, terms such as “cellular amino acid biosynthetic process” (GO:0008652) and “glutamine family amino acid biosynthetic process” (GO:0009084) exhibited significant enrichment in the proteome (represented by orange circles) with high enrichment ratios. This confirmed that suspension cells effectively execute metabolic shifts at the protein level to sustain proliferation and functions. Furthermore, the enrichment of ion transport processes, such as “sodium ion transport,” (GO:0006814) implied alterations in the mechanisms regulating membrane potential and osmotic pressure. Regarding cellular component, the analysis highlighted profound changes in cellular structure involving the remodeling of the cytoskeleton and membrane complexes. Actin cytoskeleton and intermediate filaments were significantly enriched in the proteome, which possibly accounted for the morphological transition of suspension cells from an adherent, flattened shape to a spherical form. Of particular note, the sodium–potassium-exchanging ATPase complex displayed an extremely high enrichment ratio and significance in the proteome. This corroborated our previous identification of the key driver protein Na +/K + ATPase alpha 1 in the O2PLS analysis. In terms of molecular function, the analysis revealed an adaptive strategy characterized by the remarkable upregulation of protease inhibitor activity. Peptidase inhibitor activity and endopeptidase inhibitor activity exhibited extremely high statistical significance in the proteome (dark red circles, P-value < 0.00001), whereas the significance at the transcriptional level was weaker. This strongly suggested that SA_F cells accumulated large amounts of protease inhibitors via post-transcriptional regulation to resist proteolytic stress in the serum-free suspension environment or to protect autocrine factors. The joint KEGG analysis (Supplementary Fig. S8B) further validated these findings. Pathways such as cell adhesion molecules (CAMs) and ECM-receptor interaction exhibited the highest significance in the proteomic data (dark red/orange circles), with enrichment ratios significantly higher than those in the transcriptome. This hinted that, although gene transcription was altered, cells adapted to the suspension state through more profound adjustments in protein abundance, e.g., downregulation of specific integrins or switching of adhesion molecules. Regarding the retention of metabolic and secretory functions, glycine, serine and threonine metabolism was enriched in both omics layers, yet the significance at the protein level (orange) was stronger than at the transcriptional level (gray/blue), again underscoring the stable accumulation of metabolic enzymes. Furthermore, pathways closely related to the cytoskeleton, such as HCM and ARVC, which were essentially abundant in the actin/myosin system, also displayed high significance in the proteome, further corroborating that cytoskeletal rearrangement was a core event in suspension adaptation. Taken together, the joint enrichment analysis demonstrated that the suspension adaptation of HEK293 cells was a multi-level regulatory process. While the transcriptome set the blueprint for gene expression, the proteome exhibited more significant and direct functional execution characteristics regarding metabolic enhancement, cytoskeletal remodeling, and protection against proteolysis. Discussion Although there are cases in which adherent HEK293 cells are successfully adapted to suspension culture and are prolific for their target product, only a limited amount of work has focused on the changes in cells during suspension adaptation. Here, we measured and identified some phenotypic changes in several aspects, including cell growth, cell morphology, and cell adhesion strength. First, deduced from the cell growth curve, the cell growth rate decreased during the adaptation to serum-free culture as a result of serum reduction. This suggests that adherent cells may rely on the additional nutrient content and signaling molecules from the serum to maintain growth and division. Serum-rich conditions can provide more growth factors and nutrients, thereby promoting rapid cell expansion (Liu et al. 2023). It is also interesting to see that the cell average diameter increased in suspension-adapted cells compared to adherent parental cells. Loss of contact inhibition and the more spacious volume for cell activities might lead to an increase in diameter (Seirin Lee 2016). Cell adhesion assay showed that cells cultured in 10% serum exhibited the strongest adhesion strength, while both serum-free adapted cells and suspension-adapted cells demonstrated the poorest adhesion without significant difference between these two groups. A conclusion is thereby reached that the existence and concentration of serum have an impact on cell adhesion capacity. Such observations are also reported in a study concerning the adhesion ability of THP-1 cells (Fan and Karino 2008), and a study on human osteosarcoma MG63 cells and human hepatic stellate LX-2 cells (Piazza et al. 2024). When no serum exists in the culture medium, cells adapt to the culture environment by adjusting their dependence on adhesion and remold their own structure. Transcriptomic and proteomic analysis also evidence this conclusion: many of the differentially expressed genes and proteins are associated with cell structure and cell adhesion. Of note, at the metabolic level, we detected a noticeable retard in both glucose consumption and lactate generation rates in suspension-adapted cells. As mentioned earlier, slower lactate generation from cell is considered a useful feature, particularly in large-scale bioproduction since lactate is regarded as a waste from cell metabolic fluxes. This waste product can create a toxic environment for cells by disturbing pH level, changing osmotic pressure and affecting normal metabolism (Pereira et al. 2018), and lactate can also adversely influence the productivity of cells to produce viral vectors (Nadeau and Kamen 2003). In addition to characterizing cell phenotype and providing corresponding interpretations, we also sought a more in-depth exploration into the changes at the molecular level during the adhesion turnover. Hereon, we conclude from the multi-omic data and enrichment analysis, and then propose an omics-based hypothesis, attempting to offer a possible explanation for the mechanism of suspension adaptation: adherent HEK293 cells adapt to suspension culture via cell reprogramming in three aspects—structural remodeling, metabolic shifts and intensification of resistance against external stress. It is important to mention that in this work, we present results from transcriptomic, proteomic, and metabolomic enrichment analyses individually, as well as results from their integration since both perspectives are valuable. Integrated analyses are more reliable because they take into account upstream and downstream relationships, specifically, relationship between transcript abundance and protein expression. However, integrated results may be compromised by several factors, including the detection sensitivity, quantitative characteristics of each omics platform, and specific integration strategy used. Therefore, while placing much emphasis on mechanisms supported by the integrated analysis, we retained each individual omics analyses to provide a richer set of raw data and a more comprehensive reference for subsequent research, analysis or experimental validation—whether by our group or by readers—into proteins, genes, or metabolic pathways of interest. The morphological change from a flattened adherent form to a spherical suspension form is the most striking and determining feature of adaptation, and it is an apparent compass to indicate cellular structural changes. Our multi-omic data provide molecular evidence that this shift is governed by a coordinated restructuring of adhesion complexes and the internal cytoskeleton. On the one hand, cells selectively suppress anchorage-dependent signaling. Joint enrichment analysis showed significant downregulation of pathways related to cell adhesion molecules, ECM-receptor interaction, and collagen-containing extracellular matrix. This suppression effect is consistent with the observed reduction in adhesion capacity and is essential for the cells to acquire anoikis resistance—the ability to survive without physical attachment to the extracellular matrix (Guha et al. 2019). On the other hand, the internal structural integrity is actively maintained. Although adhesion structures are compromised, GO analysis highlights significant enrichment of the intermediate filament components in the proteome (Supplementary Fig. S5A). This reorganization suggests that the cells reinforce their internal mechanical support to sustain a spherical shape and resist the hydrodynamic shear stress inherent to suspension culture. Differential expression of macrophage MIF between adherent and adapted cells further confirms this structural adaptation (Fig. 7C). While known for its role as an immune modulator in inflammation, several diseases and cancers (Aliyarbayova et al. 2025; Guo et al. 2025), MIF has been reported to regulate cell adhesion, migration, spreading, and morphogenesis, partly by modulating Rho GTPase activity (Ridley 2015; Fan et al. 2011), which directly interacts with cytoskeletal components like actin and myosin (Spiering and Hodgson 2011; Nayak et al. 2013; Sumida and Yamada 2015; Arnold et al. 2017). Thus, the elevated expression of MIF likely reflects its role in turning its cytoskeletal dynamics suitable in a continuously motional environment. In addition to structural remodeling, cells adjust their own metabolism to achieve nutritional homeostasis and proliferative capacity. The turnover from a serum-rich, adherent culture system to a serum-free suspension environment imposes dual metabolic challenges: compensating for the loss of exogenous nutrients, and meeting the demands of normal cell function and rapid proliferation under such circumstances. Our results highlight key adaptive strategies of cells in lipid and amino acid metabolism, as well as ion transport regulation. First, the capacity for amino acid biosynthesis is significantly enhanced, as evidenced by joint transcriptomic and proteomic enrichment analysis of cellular amino acid biosynthetic process and glycine, serine, and threonine metabolism (Supplementary Fig. S8A and Supplementary Fig. S8B). The upregulation of these bioactivities reveals that cells make a systemic effort to secure the necessary precursors for protein synthesis and cell function. Second, lipid synthesis is redirected. Metabolomic data showed enrichment of fatty acid biosynthesis and linoleic acid metabolism in Table 1, corroborated by transcriptomic upregulation of lipid biosynthetic process. This enhancement of endogenous lipid production is crucial for three conceivable reasons: compensating for the removal of serum-derived lipids (Hosios et al. 2018; Wu and Näär 2019); potentially alteration of cell membrane composition thereby enhancing resistance in response to the shear stress (Amador et al. 2021); and alteration of cell membrane composition to incur the metabolic shifts by adjusting the chemical transportation gating (Reddy et al. 2019). Third, ion homeostasis is actively managed. The identification of Na +/K + ATPase alpha 1 as a core O2PLS driver (Fig. 7C), combined with the strong evidence from GO enrichment on sodium ion transport, points to a critical adjustment in membrane transport function. Na +/K + ATPase alpha 1 is a key regulator of cell volume and osmotic pressure, and its upregulation suggests the cells are actively reshaping their membrane potential and osmotic balance to cope with the altered physicochemical properties of the suspension culture medium (Old et al. 1995; Elias et al. 2025; Fedosova et al. 2021). Over and above, cells adapted to the suspension culture mode employing another strategy: leveling up stress response to maintain homeostasis. The gradual progression of suspension adaptation means a chronic stress situation where cells must survive oxidative damage, nutrient limitation, and mechanical forces, the last two of which have been discussed as aforementioned. To survive oxidative stress, it is important for cells to alter their metabolic profile and to maintain protein homeostasis, bolstered by protease inhibition (Sapeta-Nowińska et al. 2025; Cafe et al. 2020). The dramatic enrichment and high statistical significance of peptidase inhibitor activity in the proteome is a salient adaptive feature (Supplementary Fig. S5A), suggesting that suspension-adapted cells employ a mechanism, likely involving translational control, to accumulate protease inhibitors such as serpins or tissue inhibitor of metalloproteinases. This strategy is vital in serum-free conditions to protect cell-surface receptors and autocrine growth factors from degradation by endogenous or exogenous proteases, ensuring survival signaling and micro-environmental stability (Ji et al. 2025). Besides, cells need to manage mitochondrial stress. The high loading of the mitochondrial chaperone Hsp10 in the O2PLS analysis indicates activation of the mitochondrial unfolded protein response (Jung et al. 2025; Wardelmann et al. 2021) (Fig. 7C). The elevated expression of Hsp10 helps maintain protein folding integrity and function within the mitochondria and ensures the high ATP output required for the adapted cells to keep their metabolic activities and proliferative state. This is also complemented by the enrichment of antioxidant pathways observed in the metabolome to form a comprehensive defense mechanism against oxidative stress. Conclusively, the successful suspension adaptation of HEK293 cells is achieved through a multifaceted regulatory process. The cells overcome physical constraints via cytoskeletal remodeling, ensure viability through metabolic adaptation, and establish survival stability by activating sophisticated stress defense systems. These three aspects of cell responses interact and interweave as a complex network, assisting the cells to live through the suspension adaptation, and these mechanistic insights provide a molecular roadmap for the rational design and engineering of excellent industrial cell lines. The massive quantity of omics data can also be interpreted in other ways for different purposes, such as exploring key genes and proteins that regulate the adaptation and production of viral vectors, unveiling the relationship between the suspension adaptation and cell tumorigenicity, et cetera (). Venn analysis compares cells at each adjacent serum concentration and the initial 10% serum condition with suspension-adapted cells and identified only three shared genes. Of these, the other two genes, excluding CLDN7, were not considered to be of obvious biological significance, which led us to focus on CLDN7, that was differentially expressed in both mRNA and protein levels (Supplementary Fig. S9). CLDN7 protein belongs to the claudin family, a protein family with 27 members found in mammalian cells. Claudins are tetraspan transmembrane proteins that form tight junction strands between epithelial or endothelial cell sheets and serve as a basic structural component as a physical barrier (Tsukita et al. 2019). They also serve crucial rules in cell permeability, selective ion transportation, maintenance of cell polarity, signal transduction, and notably, cell adhesion and migration (Kim et al. 2019a, b; Otani and Furuse 2020). Back to cell suspension adaptation, it is a process wherein adherent cells are progressively selected and acclimatized to grow and proliferate as suspension cells in a free-floating state, devoid of any solid surface attachment. This process necessitates a profound phenotypic reprogramming, which possibly includes loss of contact inhibition, rearrangement of the cytoskeleton, regulation of cell adhesion molecule expression, acquisition of anti-apoptotic capabilities and metabolic adaptations. CLDN7 is presumed to take an active part in these processes. In adherent cells, high CLDN7 expression is a hallmark of a mature epithelial phenotype, facilitating cell–cell connections, the formation of cohesive monolayers and proliferation (Phattarataratip and Sappayatosok 2020; Fan et al. 2024), and aberrant expression of CLDN7 brings alterations in cell integrity, cell polarity and intercellular contact (Martin et al. 2011). Research has found differential expression of claudins in canine mammary cells from primary tissue culture to establishment of cell line (Hammer et al. 2016), and loss of tight junction usually happens in such a course. Similarly, during adaptation, cells must dissociate from tight junctions to exist as single cells in suspension, and a plausible mechanism to elucidate this is the significant downregulation or functional inhibition of CLDN7. Such reduction weakens intercellular tethering forces thus benefits cell dissociation and promotes cell survival (Arabi et al. 2023). All these theories and researches have set up a rationale and provided an idea that in future work, we can utilize gene-editing tools, e.g., CRISPR-Cas9, to knock down/out CLDN7 in adherent HEK293 cells for suspension adaptation. If feasible, suspension HEK293 cells, originated from adherent status by gene editing, rather than suspension adaptation, will be more homogeneous in cell genome and more stable in cell performance, and ultimately accelerate cell line development for cell-based pharmaceutical applications. In conclusion, we successfully adapted adherent HEK293 cells to suspension culture with preferable cell growth and productivity for recombinant adenoviral vector. We observed alterations in cell growth rate, glucose consumption, lactate generation, and cell-surface adhesion between suspension-adapted cells and the parental adherent cells. Transcriptomics, proteomics, and metabolomics analysis and following GO, KEGG and GSEA enrichment analysis were performed to provide a molecular investigation into the key switches in cells. Based on the differentially expressed genes, proteins, and metabolic pathways, we hypothesize and summarize that suspension adaptation occurs not only in a phenotypic aspect, but also at molecular levels, including gene expression, protein expression and metabolic activities, and all these cellular changes aim to mediate structural components, metabolic activities and stress resistance to survive and adapt in the turnover environment. Supplementary Information Below is the link to the electronic supplementary material. (DOCX 2.11 MB) Author contribution B. Zhang and S. Li contributed equally to this work. J. Wei, S. Li, and B. Zhang designed and initiated the research. S. Li and B. Zhang performed the experiments with the help of J. Liu and W. Su for cell culture, viral production, and cell characterization. S. Li and B. Zhang performed the multi-omics experiments, data analysis and visualization. X. Zhang, X. Ren, Z. Ge, T. Zhao, and Q. Huang determined viral titers. S.Li and B.Zhang drafted the manuscript, which was revised by J. Wei. All authors read and approved the final manuscript. Funding This research did not receive funding. Data availability The RNA sequencing data during the current study are available in the Sequence Read Archive (SRA) under PRJNA1381743. Proteomics data during the current study are available in the iProX, IPX0014727001 Metabolomics data during the current study are available in the MetaboLights, MTBLS13517. Declarations Ethical approval This study does not involve human participants or animals. The HEK293 cell line used in this study was obtained from a commercial source (ATCC), and therefore, specific ethical approval for human tissue use was not required. All experiments were conducted in accordance with the institutional biosafety guidelines. Competing interests The authors declare no competing interests. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Benyao Zhang and Shishi Li contributed equally to this work. References Associated Data Supplementary Materials (DOCX 2.11 MB) Data Availability Statement The RNA sequencing data during the current study are available in the Sequence Read Archive (SRA) under PRJNA1381743. Proteomics data during the current study are available in the iProX, IPX0014727001 Metabolomics data during the current study are available in the MetaboLights, MTBLS13517.