Autophagy and intracellular product degradation genes reduce aggregation of bispecific antibody in CHO cells with a high translational burden

Aggregation of therapeutic bispecific antibodies negatively affects the yield, shelf-life, efficacy and safety of the product. Pairs of stable Chinese hamster ovary cell lines produced two difficult-to-express bispecific antibodies with different levels of aggregated product (10-75% aggregate) in a miniaturized bioreactor system. Here, we analyse the cellular response and link to product aggregation by comparative transcriptome analysis of these CHO cells, to define biological causes and infer strategies to improve yield and quality. Differential expression-and gene set analysis revealed upregulated proteosomal degradation, unfolded protein response and autophagy processes to be correlated with reduction of protein aggregation. Fourteen candidate genes with potential to reduce aggregation were co-expressed in the stable clones for validation. Of these, HSP90B1, DDIT3, AK1S1, and ATG16L1, were found to significantly lower aggregation in the stable producers and two (HSP90B1 and DNAJC3) increased trastuzumab titres by 50% each during transient expression. We suggest our approach to be of general use for defining aggregation bottlenecks in CHO.


Introduction
Mammalian cell lines have become the dominant expression system for therapeutic proteins and are currently used to manufacture more than 80% of all biopharmaceuticals approved on the market [1].Monoclonal antibodies (mAbs) are used for treating a wide variety of diseases including cancer and autoimmune disorders [2].Over the last decade, the biopharmaceutical industry has shifted its focus towards the research and development of bispecific antibodies (BsAbs) because of their ability to bind two different epitopes at ones resulting in improved efficacy [3].There are several different categories and formats of BsAbs, including IgG-like formats such as quadromas, knobsinto-holes, DuetMabs, and dual-variable domains Ig which is created by fusion of two variable domains of mAb resulting in a dual-specific IgG-like molecule [3,4].Despite all of the advantages in biological activity, the production of BsAbs can be challenging as a result of their increased molecular complexity.For instance, one of the challenges is to obtain correct pairing and assembly of the heavy and light chains in order to generate two different sites for epitope binding [3,5].This inherent structural complexity exposes BsAbs to product degradation events with aggregation being the most concerning quality attributes and usually halting the new candidates to preclinical stages [3,6].
Aggregation can occur at multiple stages, from production to storage and administration of the therapeutic protein.The formation of protein aggregates is a major concern during drug development due to its negative impact on the product yield, quality, safety and efficacy [7].Several factors can influence the tendency of a protein to aggregate, including its primary-, secondary-and tertiary structure, as well as glycosylation pattern, and susceptibility to chemical damage.Moreover, cellular stress events associated to the expression of protein levels that surpass the folding-and secretional capacity of the cells have been reported to induce aggregation [7][8][9].Along these lines, the imbalance of expression between antibody heavy and the light chain is known to affect yield and aggregation [10,11].Amongst the numerous strategies applied to reduce BsAbs aggregation are the removal of aggregation-prone regions in the primary structure, improved bioprocessing conditions, media-and storage formulations as well as purification methodologies [7].Furthermore, various genetic engineering strategies have been investigated to prevent intracellularly-formed aggregates and to improve protein secretion [12].Some examples include the overexpression of endoplasmic reticulum (ER) chaperones, vesicle transporters, genes involved in the unfolded protein response (UPR), signal recognition particles, advancement in the signal peptides engineering and the introduction of regulatory elements to fine-tune the ratio between the heavy chain (HC) and the light chain (LC) [10,12].Unfortunately, successful cell engineering findings are not necessarily applicable to all host cell lines and the expressed therapeutics.Many positive results have not been reproducible in settings different from those adopted in the original finding [13,14].This may be the result of the large number of genes investigated as well as their convoluted interactions that make up the protein production machinery of a cell.Additionally, the burden of constantly boosting expression titres is likely to have ramifications for many biological processes throughout the cell.Omics analysis has already helped improve media optimization and aid identification of targets for genetic engineering [15,16].Application of omics technologies also has the potential to strengthen our understanding of the secretion machinery and the biological implications of producing high levels of recombinant proteins, which may in turn guide bioprocess optimization and cell engineering strategies [15,17].In this study, we performed transcriptomic analysis of stable CHO clones and pools producing two difficult-to-express and aggregation-prone BsAbs to define cellular processes linked to both product aggregation and yield.Guided by differential gene expression and gene set analysis, we found four genes (HSP90B1, DDIT3, AKT1S1, and ATG16L1) that significantly reduced the level of aggregated BsAb products when overexpressed and two genes (HSP90B1 and DNAJC3) that increased the titre of trastuzumab in transiently transfected CHO cells.This study also highlights and discusses the importance of cellular homeostasis, finding the right level of transcription of heavy and light chains, and the observed biological differences between the cell lines, such as upregulation of genes involved in endoplasmic reticulum (ER) stress, autophagy, and neurodegenerative diseases.

Model CHO cell cultures
Stable transfectant pools (Ab2-less and Ab2-OPT-agg) and clonal cell lines (Ab1-less and Ab1agg) were generated by nucleofection (Amaxa) of the corresponding bispecific expression plasmid into the AstraZeneca (AZ) proprietary CHO host, a derivative of CHO-K1, and subsequent selection with methionine sulfoximine (MSX).Clonal cell lines were isolated by single cell cloning using fluorescence-assisted cell sorting (FACS).Cells were routinely cultured in CD-CHO (Life Technologies) with 50 µM MSX in Erlenmeyer flasks at 140 rpm, 37 °C, 6% CO2 and 70% humidity.Host cells were maintained in CD-CHO with 6 mM L-glutamine.All cell lines were adapted to AZ proprietary production medium before fed-batch culture.

Fed batch culture and preparation of RNA samples
Fed-batch culture was performed in an ambr15 (Sartorius Stedim) using AZ proprietary production medium and feeds, in a 14-day process.Each cell line was assessed in quadruplicate vessels.Cells were maintained with 50% dissolved O2, 35.5 °C, and controlled pH.Glucose concentration was maintained during the fed-batch culture and measured daily with a glucose/lactate analyser (YSI).Samples for transcriptomic analysis were taken on day 8, triplicate samples were taken from vessels with the highest viability.Aliquots of 5 x 10 6 cells were spun down at 300 g for 5 minutes, the supernatant removed and the pellet snap frozen on dry ice and stored at -80 °C.Cells were then resuspended in 200 µL RNAlater, samples were placed at 4 °C overnight and then stored at -80 °C.Samples were thawed and RNA was extracted according to manufacturer's instructions with the Qiagen RNeasy® Plus Universal Mini Kit (Qiagen, Hilden DE).RNA integrity was checked with an Agilent RNA 6000 Nano kit (Agilent Technologies, Santa Clara, CA, US) on an Agilent Bioanalyzer 2100 and samples with an RNA integrity number (RIN) > 8 were sent for subsequent sequencing.Sequencing was performed on an Illumina platform at paired-end 2x150bp (Illumina HiSeq platform via a commercial service at GATC Biotech, Constance, Germany).

Titre assessment
Sample quantitation was conducted using an Agilent 1260 high performance liquid chromatography (HPLC) system equipped with a binary pump (Agilent Inc.).A POROS™ A 20 affinity column of 2.1 mm × 30 mm, 0.1 mL (Thermo Scientific) was employed using two mobile phases; buffer A: 10 mM sodium phosphate, 150 mM sodium chloride pH 7.2 (used for sample loading and washing of unbound impurities) and buffer B: 12 mM HCl / 150 mM sodium chloride pH 2.0 (used for protein elution step).The calibration curve was prepared using a corresponding reference material with known concentration.The run was carried out at 2 mL/min and the protein elution was monitored by A280 nm.Sample concentration was determined by integration of eluted peak and plotting the peak area against the calibration curve.

Aggregation assessment-2D ProA -SEC
The first dimension ProA method followed the same parameters as the mAb titre method described previously but with an optimized elution (buffer B) at pH 3.0.Utilizing 2D-LC Acquisition Software OpenLAB CDS, a single heart cut was diverted post first dimension detector via 2D-LC valve.This material was stored in a loop capillary before being diverted into the second-dimension flow path.Aggregate analysis was performed via size exclusion chromatography (SEC) in the second dimension on an Agilent 1290 high performance liquid chromatography (HPLC) system equipped with a binary pump (Agilent Inc.).A 6 min isocratic elution at 0.4 mL/min was performed using a mobile phase of 0.1 M sodium phosphate dibasic anhydrous and 0.1 M sodium sulfate adjusted to pH 6.0.Data were collected at 280 nm using a Diode-Array Detector.

Transcriptomics analysis
Gene expression was quantified from raw sequencing data using kallisto [18] cDNA for cricetulus_griseus_picr, Ensemble release 99 [19] as reference with the addition of the antibody sequences.Differential expression analysis was carried out using DESeq2 [20] and the raw counts from kallisto, imported with tximport [21].P-values were calculated with Wald tests and the BH method was used for multiple testing correction.Gene set analysis was performed using Piano [22] with its default settings for its runGSA function, fold changes and padj values from DESeq2, and gene sets were downloaded from MSigDB [23,24].The heatmaps using GO slims in figure 2I were based on the consensus score from gene set statistics calculations with mean, median, sum, Stouffer and tailStrength tests and were calculated with Piano's consensus Heatmap function.Gene sets with the lowest consensus score of one, in either the distinct up-or down direction for either of the two comparisons, were included in the figure.The KEGG [25][26][27] pathway map was colored using Pathview [28], but colors were changed to make it more color-blind friendly.RNA-seq raw data has been deposited at Sequence Read Archive (SRA) and has BioProject ID: PRJNA739051.Reviewers may access the data here before zsubmission: https://dataview.ncbi.nlm.nih.gov/object/PRJNA739051?reviewer=jsj1cjcva4l33ne tnbvfri599.

Cell culture and transfection
ExpiCHO TM cells (Thermo Fisher Scientific) were routinely cultivated in ExpiCHO TM expression medium (Thermo Fisher Scientific), passaged according to the manufacturer's instructions and incubated at 37 °C, 5% CO2, humidified air and 120 rpm.ExpiCHO TM cells were transfected using the standard protocol as suggested by the manufacturer with a total of 20 μg trastuzumab plasmid DNA per transfection.In order to boost the expression level of trastuzumab and to avoid too much translational burden of the cells, a 1:10 ratio between trastuzumab and helper genes were chosen.Moreover, the cells were transfected with 20 μg of the trastuzumab expression construct and 2 μg of the helper genes.An empty vector (pKTH16_empty) was used as a negative control instead of the helper genes.All transfections were made in duplicate and were harvested at day 8 post transfection.

IgG purification
The supernatants, which were harvested at day 8 post transfection, were sterile filtered through a 45 μm filter followed by addition of 200 μL phenylmethylsulfonyl fluoride (protease inhibitor) and stored at 4 °C overnight.The expressed antibodies were purified by Protein A facilitated purification on an ÄktaSTART system (GE Healthcare, USA) using mAbSelectSuRe columns (GE Healthcare).A 20 mM sodium phosphate, 0.15 M sodium chloride (pH 7.3) buffer was used as binding and wash buffer, 0.1 M glycine (pH 2.5) as elution buffer and 1M Tris-HCl (pH 8.5) as neutralization buffer.

Cell culture and transfection
Ab-1-agg and Ab-2-OPT-agg cell lines were transfected by nucleofection with 6 µg DNA and recovered into 2 mL proprietary production medium in a 24 deep well plate.Plates were incubated at 210 rpm, 37 °C, 6% CO2 in a shaking incubator.Four hours post transfection in-house feeds and glucose were added and the temperature was reduced to 34 °C.Cells were fed again day 4 post transfection.Titre was measured on day 6 post transfection using the same methods described previously.VCN and viability was measured by trypan blue exclusion using a CellaVista.

Antibody purification and aggregation assessment
Bispecific mAb was purified from the harvested cell culture supernatant by Protein A affinity chromatography using 200 µL volume PhyTip® columns containing 20 µL of ProPlus (MabSelect SuRe™) affinity resin (PhyNexus Inc., USA) with a Tecan Freedom EVO® 200 robotic liquid handler (TECAN Group Ltd, Switzerland).A DropSense 96 (Trinean, Belgium) was used for measuring sample recovery post-purification.High-throughput aggregation analysis was achieved by size-exclusion chromatography (SEC) and performed on an Agilent 1260 Infinity II UHPLC System equipped with a degasser, quaternary pump, thermostatted multi-sampler, and diode array detector.The method used an Aquity BEH SEC column (1.7µm, 4.6 x 150 mm, 200Å; Waters, Milford, USA) and a mobile phase composed of 50 mM sodium phosphate buffer, 200 mM sodium chloride, pH 6.8 eluted with a sub-4 min isocratic run at 0.15 mL/min.The column oven was set to 25 °C and the injection volume was 10 µL.Data analysis was achieved with Agilent OpenLAB CDS ChemStation Edition, version C.01.07.

Cell lines expressing the same bispecific antibody produced different amounts of aggregated product
This study aimed to identify aggregation bottlenecks of two different BsAb (Ab1 and Ab2) when stably expressed in CHO cells.Two high-producing clones, "Ab1-less" (Ab1 antibody with less aggregation) and "Ab1-agg" (Ab1 producer with more aggregation), engineered with the same Ab1 expression vector but displaying varying degrees of product aggregation, were selected for the Ab1 antibody.For the Ab2, stable transfectant pools "Ab2-less" and "Ab2-OPT-agg" were generated, with the latter involving the use of a vector containing a codon optimized HC (therefore "OPT" in its name).The molecular structures of Ab1 and Ab2 were constructed by inserting a single-chain variable fragment (scFv) in the CH3 region (Fig. 1A) and in the hinge region between the fragment antigen-binding region (Fab) and the fragment crystallizable region (Fc) (Fig. 1B), respectively.The properties of these bispecific formats have previously been described [29].Both pairwise comparisons showed significant differences in titre, cell specific productivity, and aggregation.Ab1-agg expressed more antibody mRNA and resulted in higher titre and cell specific productivity (qP) (2.1 g/L and 12.5 pg/cell.day)compared to Ab1-less (1.4 g/L and 5.1 pg/cell.day)(Fig. 1C-D).Ab1-agg expressed more than twice as much antibody mRNA compared to Ab1-less.The expression of Ab1 heavy-and light chain corresponded to 35000 transcripts per million (TPM) and 85000 TPM, respectively, whilst Ab1-less expressed HC at 14000 TPM and LC at 33000 TPM (Fig. 1E) resulting in a LC:HC ratio of approximately 2.4-fold for both clones.Moreover, Ab1agg showed 5% increase in aggregation compared to Ab1-less (Fig. 1F).Ab2-OPT-agg produced higher titre and qP than Ab2-less (0.85 g/L and 6.6 pg/cell.daycompared to 0.45 g/L and 0.8 pg/cell.day,(Fig. 1C-D), while the expression ratio of the LC and HC was considerably shifted when comparing Ab2-less and Ab2-OPT-agg, with 3.7 for Ab2-less and 2.1 for Ab2-OPT-agg (Fig. 1E).The alteration of LC:HC expression ratio, combined with an expected higher translation efficiency due to the codon optimization lead to higher amount of HC polypeptides in Ab2-OPTagg.Interestingly, the antibody aggregation level for Ab2-OPT-agg was significantly higher compared to clone Ab2-less (75% compared to 10%, respectively) (Fig. 1F).

Early cell death and a rapid accumulation of lactate were distinguishing traits of Ab2-less cells
Viable cell density (VCD), viability, glucose and lactate levels were measured throughout the fedbatch culture (Fig. 2A-H).Ab1-less showed a faster growth rate than Ab1-agg with maximum VCD achieved on day 7 rather than day 10 reporting 27x10 6 cells/mL and 15.5x10 6 cells/mL, respectively (Fig. 2A).No significant difference was observed in viability but Ab1-less reached significantly higher lactate levels between day 7 to day 11 (Fig. 2B-C).Glucose concentration was maintained during the fed-batch culture and was observed to remain steady for both clones throughout the experiment (Fig. 2D).Ab2-less and Ab2-OPT-agg showed a bigger difference in regard to VCD, viability and lactate concentration compared to the Ab1 clones.Approximately at day 5, the VCD and viability began to drop for Ab2-less (Fig. 2E-F).As was also seen for the Ab1 clones, the glucose concentration was similar for both clones up until the final days of cultivation (Fig. 2H).The lactate concentration on day 2 was higher in Ab2-less and continued to increase steadily until the end of the fed-batch culture (Fig. 2G).The differences observed in cell death, growth and division were also supported by the transcriptomic data.In Ab2-less cells, genes enriching the cell death pathway were significantly upregulated, whilst genes enriching chromosome segregation process were significantly downregulated.(Fig. 2I, supplementary GSA results).Moreover, Ab2-less appeared to undergo a cellular transformation (which may be) associated with the cell death since upregulation of cell differentiation and membrane organization were observed.With considerably more GOSlim terms presenting a significant change, the difference was much greater between Ab2-less and Ab2-OPT-agg than between the two Ab1 clones (Fig. 2I, supplementary GSA results).The increased lactate production in Ab2-less may in part be explained by the upregulation of glucose catabolic processes (supplementary GSA results), and the (padj < 0.01) higher expression compared to Ab2-OPT-agg for several glycolysis genes (supplementary DE results).Among these genes, ENO2, HK2, PFKFB3 and PFKFB4 showed the highest upregulation (above 0.5 log2 fold change).

Ab1-agg clone primarily demonstrated an upregulated ER stress and unfolded protein response compared to Ab1-less.
Although clones Ab1-less and Ab1-agg produced the same antibody protein sequence, Ab1-agg produced more antibody mRNA resulting in a higher titre, albeit in a more aggregated form, as was seen in figure 1C-F.When comparing the gene expression of the two clones, the most upregulated gene sets (considering p-values) were associated with ER stress and the UPR in Ab1agg (Fig. 3A, supplementary GSA results).IRE1-mediated UPR, as well as responses to topologically incorrect protein, ER stress, and PERK-mediated UPR were among the most upregulated gene sets in Ab1-agg compared to Ab1-less.Other top upregulated gene sets, involved related processes, such as protein stabilization, retrograde protein transport from ER to cytosol, Golgi organization and vesicle transport.The most upregulated genes (considering fold change and padj < 0.01) related to ER stress and UPR covered a variety of functions.They serve as chaperones protecting polypeptides from misfolding upon translocation (HSPA5, HSP90B1, and DNAJC3), inducing cell cycle arrest and apoptosis (DDIT3) and rearranging disulphide bonds (PDIA4).The remaining genes are part of ERAD and the ubiquitin ligase complex (UBXN8, HERPUD1, and DNAJB9) (Fig. 3B).The selected genes are likely to be crucial for cells to manage an elevated load of aggregation-prone antibody polypeptides inside the ER, and were overexpressed transiently in Ab1-agg aiming to reduce aggregation (Fig. 3C).On average, both HSP90B1 and DDIT3 reduced aggregation from 27 % to 23 %.However, DDIT3 (as well as UBXN8, HSPA5, and HERPUD1) also gave a reduced titre (Fig. 3D).To further assess their impact on antibody production, these genes were co-expressed together with trastuzumab.Here, the luminal chaperones, HSP90B1 and DNAJC3, had a positive impact on protein expression improving titre by 59% and 47%, respectively (Fig. 3E).

Ab2-less had upregulated genes related to autophagy and neurodegenerative diseases compared to Ab2-OPT-agg
The major difference between Ab1-less and Ab1-agg were the upregulation of ER-stress and UPR.However, none of the upregulated gene sets (Fig. 3A) showed a significant (padj < 0.01) up-or downregulation when comparing the two Ab2 producers (Supplementary GSA results).Nevertheless, the most upregulated ER stress genes in Ab1-agg compared to Ab1-less (Fig. 3B) were also upregulated in Ab2-OPT-agg compared to Ab2-less with one exception -DDIT3, a gene responsible for triggering cell cycle arrest and apoptosis (see Supplementary DE results).
Noteworthy, Ab2-less and Ab2-OPT-agg differed in the upregulation of autophagy and related processes, such as autophagosome organization, endosome organization, SNARE interactions in vesicular transport, and the lysosome (Supplementary GSA results).127 of the 421 genes, populating all these sets, were found to be significantly (padj < 0.01) upregulated and 67 downregulated in Ab2-less compared to Ab2-OPT-agg (Fig. 4A).Noticeable among these upregulated genes were, AKT1S1, PIK3C3, ATG101, LAMP2, TOLLIP, ATG16Ll, which are involved in the degradation of cellular material by autophagy [30][31][32].These genes were subsequently transiently expressed in Ab2-OPT-agg to evaluate whether overexpressing these could reduce its aggregation (Fig. 4B).Here, both ATG16L1 and AKT1S1 could significantly reduce the aggregation from approximately 52 % to 45 %.LAMP2 was the only gene that significantly lowered the titre when overexpressed (Fig. 4C).
Other differences in biological processes between Ab2-less and Ab2-OPT-agg included the upregulation of various signalling pathways (JAK-STAT, MAPK, and ERBB), for Ab2-less compared to Ab2-OPT-agg (Supplementary GSA results).Among these pathways were some of the significantly upregulated genes that showed the biggest difference in expression (padj < 0.01, log2 fold change > 3), namely, IL19 (JAK-STAT), AREG (ERBB), PLA2G4E and FLNC (MAPK).Protein synthesis appeared to be downregulated in Ab2-less compared to Ab2-OPT-agg as genes involved in translation initiation and the ribosome were primarily found to be downregulated (Supplementary GSA results).Neurodegeneration is associated with the formation of protein aggregates, and interestingly, more than twice the number of genes related to neurodegenerative disorders Parkinson's and Huntington's disease, were significantly upregulated in Ab2-less compared to Ab2-OPT-agg (supplementary DE results).

Discussion
In this study, by evaluating transcriptomic data from two stable CHO clones, as well as two stable pools, producing two different difficult-to-express BsAbs (Ab1 and Ab2), we have presented the biosynthetic differences between cell lines with different antibody expression profiles, particularly regarding processes closely associated with protein production and aggregation.
Ab1-agg expressed more than twice as much antibody mRNA compared to Ab1-less, possibly leading to increased ER-stress and UPR.It has been reported that the overexpression of an antibody and the enhanced misfolded polypeptide chains in CHO cells correlates with a rise of both BiP (HSPA5) and CHOP (DDIT3) levels [33,34].BiP binds to the Ig HC before its assembly with the LC.The overexpression and incomplete glycosylation of therapeutic proteins have been shown to increase the activity of BiP in the ER.Three ER-membrane bound sensors, PERK, ATF6 and IRE1 are associated with BiP.As a consequence of ER stress, BiP dissociates from these sensors and binds to the misfolded polypeptide, activating UPR to restore the ER homeostasis.Firstly, the UPR tries to restore ER homeostasis by initiating mechanisms to both inhibit further protein synthesis and to increase the protein folding capacity.If this approach fails due to excessive ER stress and high levels of misfolded proteins, UPR activates genes such as DDIT3, which induces cell cycle arrest and promotes cell death via apoptosis [35].BiP, DDIT3, and other ER-stress-related genes involved in IRE1-and PERK-mediated UPR, were upregulated in Ab1-agg.This suggests that the higher abundance of antibody mRNA impaired ER homeostasis, which would explain the reduction in cell growth in Ab1-agg.The use of weak promoters may favour the reduction of aggregate levels in Ab1, however, this may come with a decline of expression titre.Alternatively, the upregulation of helper genes promoting protein folding may mitigate aggregation whilst maintaining desirable expression titres [9].In Ab1-agg, the most upregulated genes considering fold change related to ER stress and UPR were HSPA5, HSP90B1, DNAJC3, DDIT3, PDIA4, UBXN8, HERPUD1, and DNAJB9.Hence, these were suspected to be crucial when the cell folding machinery was overwhelmed during antibody production and were evaluated as helper genes.Overexpression of HSP90B1 and DDIT3 in Ab1-agg reduced antibody aggregation significantly, although DDIT3 also lowered the titre.This reduction in titre is perhaps not surprising as DDIT3 promotes cell death.However, it has previously been shown to increase recombinant production, hence its influence on titre appears to be context-dependent [33].
Co-expression experiments with the first three genes HSPA5, HSP90B1 and DNAJC3, which operate as luminal chaperones that protect polypeptides from misfolding upon translocation, resulted in the highest titre increase compared to the other helper genes for trastuzumab.Here, a significant titre improvement of 59% and 47% was obtained for HSP90B1 and DNAJC3, respectively.HSP90B1 (and HSPA5) are members of a family of glucose-regulated proteins (GRPs), which are found in the ER and protect polypeptides from misfolding upon translocation [36].DNAJC3 is a member of DNAJ family, which is a lumenal ER protein.Upon its interaction with BiP, it selectively binds to misfolded polypeptides in the ER and is believed to be a profolding co-chaperone to BiP [37].HSP90B1 has previously been tested in a co-expression study to improve the productivity of erythropoietin (EPO) in CHO cells, without showing any positive effect [14].Potentially the effect on titre could be further enhanced if their expression was tuned to an optimal level or used in combination with other helper genes.It has also been suggested by different studies that overexpression of several ER-related chaperones or proteins might be necessary to significantly improve the productivity of recombinant proteins [33].
Compared to Ab2-OPT-agg, Ab2-less yielded a lower titre and a markedly less aggregated product while decreasing in viability earlier in the fed-batch process and genes associated with autophagy were upregulated.The higher mRNA expression of the HC, together with an expected increase in translation efficiency due to its codon optimization, was presumed to be the cause of the high aggregation in Ab2-OPT-agg.It has been confirmed in previous studies that if the expression level of HC is too high, it can interfere with optimal expression of recombinant IgG in CHO cells [38].In this scenario, the LC/HC ratio could be optimized by, for example, using different promoters or regulatory elements to either increase the amount of LC or express less HC [10,39].Other alternatives that could be applied to reduce aggregation include changing the hypothermic conditions [40], or switching to a perfusion process [41].
Upon incorrect folding in the ER, polypeptides may be targeted to different degradation paths to avoid the toxic effects of protein aggregates.The primary route of protein degradation is to translocate unfolded polypeptides into the cytosol for degradation by the proteasome.This nonlysosomal degradation machinery requires soluble single peptide species, while protein aggregates require degradation by autophagy [42].In this process, protein aggregates can be bundled and enclosed by an autophagosome that is fused with a lysosome or a late endosome where the aggregates are degraded.Ab2-less showed an upregulation of autophagy, autophagosome and endosome organization, lysosome, endosome to lysosome transport, and SNARE interactions in vesicular transport compared to Ab2-OPT-agg.Six autophagy-related genes that showed significantly higher expression in Ab2-less were overexpressed to see if they could improve the antibody production in Ab2-OPT-agg.Of these genes, both AKT1S1 and ATG16L1 significantly reduced aggregation.Again, the aggregation was only reduced by a small amount, but the impact of these genes could also potentially be further improved by fine-tuning their expression.AKT1S1 regulates mTORC1 activity, which controls autophagy among other biological processes [43] and ATG16L1 is involved in autophagosome formation [44].
The upregulation of autophagy in Ab2-less could be due to the lower viability in this culture, and not caused by protein aggregation.Another plausible explanation for the differences in aggregation levels and biological processes is that Ab2-less managed to re-translocate unfolded polypeptides to the cytosol, but the proteasome was overloaded, resulting in cytosolic aggregates that required autophagic degradation.This type of aggregate degradation is similar to that of neurodegenerative diseases, for example Parkinson's disease, Alzheimer's disease and Huntington's disease.Such diseases can activate programmed cell death, apoptosis and autophagy.For difficult-to-express and aggregation-prone recombinant proteins, valuable insights could be gained from the research on treating such diseases through selective autophagy (aggregaphy) [45].Substances used for treatments of neurodegenerative disorders have been observed to improve titres of recombinant proteins [46] and reduce aggregation [47]; however, it was not acknowledged in these studies that the substances could potentially also be used for treating neurodegenerative diseases.Spectroscopy and imaging have provided valuable information for understanding the mechanisms behind protein misfolding in these diseases and would have been an interesting complement to our transcriptomics data [48].
Ab2-OPT-agg had a considerably higher level of product aggregation compared to Ab2-less.A reason for this could be that aggregates of antibody polypeptides occurred already in the ER as a result of the suggested LC/HC imbalance.A proposed fate for some unassembled HC is to end up in Russell bodies, which are defined as intracellular aggregates of IgG that are not properly folded and have escaped intracellular degradation [49].Potentially a portion of the antibodies was stored in such compartments for Ab2-OPT-agg, while others were secreted in an aggregated form, something a future study would have to confirm.
This study has presented HSP90B1, DNAJC3, AKT1S1, and ATG16L1 as promising helper genes during recombinant antibody production.To our knowledge, it has not previously been described that overexpressing any of these genes can have a positive effect on antibody production by reducing aggregation or improving titres.This study has also highlighted the importance of having a good level of antibody expression and balance between heavy and light chain; here, analogous producers with differences in antibody mRNA levels (Ab1) and the HC:LC ratio (Ab2) displayed significant differences in titre, aggregation and biological responses characterized by transcriptomic analysis.Consequently, finding a favourable balance in expression for difficult-toexpress antibodies is of great importance in order to maintain a cellular balance for higher yield and quality as potential outcome.Viable cell density (VCD), viability, lactate and glucose concentration from 14-day fedbatch cultures and gene set analysis with GOSlims for the two comparisons.Lines show mean values (n=3), error bars represent the standard deviation, and student's t-test, two-sample unequal variance was used to calculate significance (p-value < 0.01 was considered significant).A) Viable cell density (VCD) for Ab1.B) Viability (%) for Ab1 C) Lactate (g/L) for Ab1 D) Glucose (g/L) for Ab1 E) Viable cell density (VCD) for Ab2.F) Viability (%) for Ab2 G) Lactate (g/L) for Ab2 H) Glucose (g/L) for Ab2.I) Gene set analysis with GO slims displaying high-level cellular processes that were significantly (padj < 0.05) up-or downregulated for either of the two comparisons.Ab1-agg Ab1-less Ab2-OPT-agg Ab2-less Figure 3. ER-stress and overexpression of ER-stress genes in Ab1-agg.A) Gene set analysis showing the most significant and distinctly upregulated biological processes in Ab1-agg compared to Ab1-less.A majority of these gene sets were directly associated with ER stress and unfolded protein response.B) Differentially expressed genes involved in protein processing in the ER when comparing clone Ab1-agg to Ab1-less.The fold changes for genes that differed significantly in expression are visualized in red (upregulation) and blue (downregulation).Genes with the biggest upregulation in Ab1-agg were selected for overexpression and are marked with stars.C) Overexpression of selected ER-stress genes in Ab1-agg.HSP90B1 and DDIT3 significantly reduced the antibody aggregation compared to the control overexpressing GFP.Based on three replicates (n=3) and Dunnett's multiple comparisons test with GFP as control was used to calculate significance (** ≤ 0.01).D) Antibody titre for Ab1-agg upon overexpression of selected genes associated with ER-stress.Overexpression of HSPA5, HERPUD1, UBXN8, and DDIT3 significantly reduced the titre compared to the control expressing GFP.Dunnett's multiple comparisons test with GFP used as a control to calculate significance (** ≤ padj 0.01, **** ≤ 0.0001).! E) Antibody titre upon co-expressing the selected genes associated with ER-stress together with trastuzumab (TRAZ).HSP90B1 and DNAJC3 showed a significant titre improvement of 59 and 47 %, respectively.Dunnett's multiple comparisons test with TRAZ as control was used to calculate significance (** ≤ padj 0.01, *** ≤ 0.001).

Figure 2 .
Figure 2.Viable cell density (VCD), viability, lactate and glucose concentration from 14-day fedbatch cultures and gene set analysis with GOSlims for the two comparisons.Lines show mean values (n=3), error bars represent the standard deviation, and student's t-test, two-sample unequal variance was used to calculate significance (p-value < 0.01 was considered significant).A) Viable cell density (VCD) for Ab1.B) Viability (%) for Ab1 C) Lactate (g/L) for Ab1 D) Glucose (g/L) for Ab1 E) Viable cell density (VCD) for Ab2.F) Viability (%) for Ab2 G) Lactate (g/L) for Ab2 H) Glucose (g/L) for Ab2.I) Gene set analysis with GO slims displaying high-level cellular processes that were significantly (padj < 0.05) up-or downregulated for either of the two comparisons.

Figure 4 .
Figure4.Differences between Ab2-OPT-agg and Ab2-less and overexpression of selected autophagy genes in Ab2-OPT-agg.A) Differentially expressed genes involved in autophagy and autophagy-related gene sets (autophagosome organization, endosome organization, SNARE interactions in vesicular transport lysosome, and lysosome) showed 127 genes significantly upregulated and 67 genes significantly downregulated in Ab2-less compared to Ab2-OPT-agg.B) Overexpression of selected autophagy genes in Ab2-OPT-agg.ATG16L1 and AKT1S1 significantly reduced the antibody aggregation compared to the control overexpressing GFP.Based on three replicates (n=3) and Dunnett's multiple comparisons test with GFP as control was used to calculate significance (*** ≤ 0.001).C) Antibody titre for Ab2-OPT-agg upon overexpression of selected genes associated with autophagy.The titre was significantly reduced with LAMP2 overexpression compared to the GFP control.Dunnett's multiple comparisons test with GFP as control was used to calculate significance (* ≤ padj 0.05).