Biotechnology and Bioengineering

In liquid fermentation of filamentous fungi such as Aspergillus oryzae, increased broth viscosity and biomass adhesion to bioreactor surfaces remain major challenges. We previously developed a hyphal dispersion mutant lacking two hyphal adhesion factors, namely cell wall -1,3-glucan (AG) and biofilm galactosaminogalactan (GAG) (AG{Delta}-GAG{Delta} strain). The culture broth of the AG{Delta}-GAG{Delta} strain has low viscosity, which improves mixing and enzyme production. However, mycelia still extensively attach to bioreactor walls and downstream equipment, which impairs mixing and reduces product recovery. The hydrophobin RolA, a surface-active protein of A. oryzae, densely coats conidia and hyphae and contributes to cell surface hydrophobicity. In this study, we disrupted the rolA gene in AG{Delta}-GAG{Delta} (AG{Delta}-GAG{Delta}-{Delta}rolA strain) and evaluated the effects of this disruption on hyphal adhesion to the walls of culture vessels, enzyme production, and bioreactor performance. At the flask scale, the adhesion to glass surfaces was significantly reduced and recombinant enzyme activity was increased by 10%. Improved culture recovery at the end of fermentation further increased total enzyme yield. In a lab-scale stirred-tank bioreactor, both growth and enzyme production were increased. Scaling-up to a 200-L bioreactor showed reduced agitation power consumption while improving hydrodynamic properties. Fermentation of AG{Delta}-GAG{Delta}-{Delta}rolA was successfully scaled up to a 3000-L bioreactor; consistent enzyme activity and improved flow circulation in the bioreactors were confirmed by computational fluid dynamics analysis. Overall, the AG{Delta}-GAG{Delta}-{Delta}rolA strain has increased enzyme production and scalability, supporting its suitability for industrial applications.

Chimeric antigen receptor (CAR) T cell therapy has achieved remarkable therapeutic outcomes in hematological cancers. However, broader clinical use has uncovered substantial challenges arising from intrinsic properties of both T cells and tumor tissues. As the functional phenotype of CAR T cells is affected by the CAR molecular architecture, optimizing CAR constructs continues to be a critical and ongoing task. Here, we present a practical workflow for scalable screening of CAR variants in primary T cells using fitness-guided design and mRNA electroporation. Using a CD19-targeted second-generation CAR, we built a library of point mutants that focused mutagenesis on hinge and costimulatory domains. Amino acid substitutions were prioritized using the sequence-based zero-shot fitness predictor to enrich evolutionarily tolerated variants. From 340 designed variants, we electroporated mRNA encoding 85 constructs into primary human CD8+ T cells and quantified cytotoxicity against CD19-positive Nalm6 cells. Twenty-four variants reproducibly exceeded wild-type cytotoxicity across three runs, and three hits were selected for lentiviral validation. One of the selected variants showed significantly improved cytotoxicity despite lower expression frequency and exhibited higher CD62L within CAR-positive cells, suggesting enhanced intrinsic function with a less differentiated phenotype. This approach enables scalable, rapid discovery of improved CAR domain variants directly in primary T cells.

The trend towards concentrated animal feeding operations (CAFOs) has served to concentrate not only livestock animals but the waste they produce to comparatively smaller areas. The point-source nature of this waste is an opportunity for the recovery and valorization of the nitrogen therein. Such a process would be viable on small to intermediate scales and require minimal inputs at the CAFO. In this study, we demonstrate the potential of the biopolymer cyanophycin to serve as a medium for manure-nitrogen recovery. In the first step, genetically modified strains of Escherichia coli produce intracellular cyanophycin from mock manure hydrolysates. Next, cyanophycin is recovered from microbial biomass via acid solubilization and base precipitation using electrochemically generated acids and bases. Finally, to improve both the yield and recoverable fraction of cyanophycin produced, we leverage the tunability of our genetically engineered system to probe the impacts of cyanophycin synthetase solubility, N-domain activity, and cyanophycin molecular weight on cyanophycin recoverability. Collectively, this work serves as a proof of concept for nitrogen recovery from agricultural waste, aligning with global sustainability initiatives.

Influenza A virus (IAV) defective interfering particles (DIPs) are regarded as promising broad-spectrum antivirals due to their high efficacy demonstrated in animal trials. Previously, an OP7 chimera DIP was generated that showed markedly enhanced antiviral efficacy compared to conventional IAV DIPs. In addition, a high-yield cell culture-based production process for OP7 chimera DIPs was established in perfusion mode. However, due to the technical complexity of this technology, we aimed to combine the advantages of perfusion in terms of yield with that of a simple and less complex batch process. In the present work, we established a fed-batch production process for OP7 chimera DIPs that allows a more straightforward implementation into existing production facilities. Using a feeding regime that was able to overcome the cell density effect, we achieved productive virus infection in stirred tank reactors until a cell density of 10E+6 cells/mL. Furthermore, we obtained a high virus titer of 3.8 log10(HAU/100 uL), corresponding to a 3.4-fold increase compared to a reference batch production. Moreover, the produced material demonstrated a significantly increased interfering bioactivity shown by an in vitro potency assay. Interestingly, this efficacy was similarly high as compared to OP7 chimera DIP material previously produced in perfusion mode, despite twofold lower virus titers. Finally, we demonstrate robustness of OP7 chimera DIP seed virus trains, which is significant step towards later pharmaceutical production. Further, corresponding GMP MDCK-PB2(sus) cell and OP7 chimera virus banks are now available for later biomanufacturing further use. In conclusion, the established fed-batch process offers a valid n additional option for pharmaceutical grade OP7 chimera DIP GMP manufacturing production under GMP conditions required for subsequent clinical testing and for later large-scale manufacturing.

mRNA therapeutics have demonstrated transformative potential in infectious disease, oncology and genetic disorders. Parallel to the rapid clinical advancement of mRNA therapies, research into in vitro transcription (IVT), the pivotal process of mRNA manufacturing, has intensified significantly. However, the majority of these studies omitted cap analog, an essential substrate for co-transcriptional capped IVT (co-IVT), likely to simplify experimental systems. Nevertheless, co-IVT remains the dominant industrial standard for the production of mRNA therapeutics. Here, we report an optimized co-IVT process that achieves a yield of more than 26 g/L while maintaining capping efficiency of over 99.5%, and achieving a 30-fold reduction in double-stranded RNA (dsRNA), the most critical process-related impurity. These performance gains substantially improve mRNA production efficiency and enhance mRNA quality, thereby accelerating the transition to commercial-scale manufacturing of mRNA therapeutics. Moreover, the performance was attained using either of two mechanistically distinct engineered T7 RNA polymerase (RNAP) mutants. Both mutants enabled robust mRNA synthesis under process conditions that have not been previously reported. These observations provide mechanistic insights into IVT and advance our fundamental understanding of this critical bio-manufacturing step.

Three-dimensional (3D) cell culture models, particularly multicellular tumor spheroids, have become essential tools for studying cancer biology, drug screening, and preclinical testing due to their ability to mimic physiological tumor microenvironments. However, traditional invasion assays, such as Boyden-chamber- or Transwell-based systems, often suffer from variability introduced by spheroid handling and transfer, compromising data reproducibility. Here, we present a novel, 3D-printed migration and invasion platform --the MQm-sert-- designed to standardize and streamline spheroid-based invasion assays while maintaining spheroid integrity. Fabricated via fused filament fabrication using biobased polylactic acid, the MQm-sert integrates a hanging-drop spheroid culture system (MQm-sert) with a membrane-based invasion chamber (M-sert), eliminating the need for disruptive spheroid transfer steps. Using synthetic tumor environment mimics (STEMs) composed of breast cancer cells (MDA-MB-231 and MCF7), mesenchymal stromal cells (MSCs), and human pulmonary microvascular endothelial cells (HPMECs), we quantified invasion dynamics and cellular interactions. This innovation significantly reduces experimental variability, as demonstrated by lower variance in invasive cell mass dimensions and cell density compared to conventional workflows. Beyond biological insights, the platform aligns with sustainability goals by leveraging cost-effective, open-source 3D printing, reducing reliance on commercial labware, and addressing key challenges in 3D cell culture standardization.

Multi-stage continuous bioprocessing can increase volumetric productivity, operational consistency, and process throughput, but its design is complicated by coupling among dilution rate, reactor volume, feed allocation, and cellular physiology. Here, we present ContiDesigner, available at https://chemnettools.anc.univie.ac.at/ContiDesigner/, a mechanistic steady-state framework and interactive web tool for the system-level design of continuous fermentation cascades. Comparing one- and two-stage configurations at equal total reactor volume and outlet flow, ContiDesigner reveals how internal flow and reactor volume allocation shape space-time yield and identifies productivity-maximizing operating conditions. Compared with one-stage processes, two-stage cascades favor lower over-all dilution rates, thereby preserving residence time in the production stage. The first-stage dilution rate approaches the corresponding one-stage productivity optimum, but the cascade optimum occurs earlier, reflecting a system-level compromise between biomass generation and production-stage residence time. However, two-stage operation outperforms optimized one-stage operation only when non-growth-associated production in the second stage is sufficiently strong, whereas increasing growth coupling favors one-stage operation. Two case studies demonstrate both the potential and limits of process intensification. An optimized two-stage design is predicted to achieve a more than 1.5 fold increase in space-time yield for poly-R-3-hydroxybutyrate (PHB) production compared with a published experimental five-stage cascade, whereas the lactic acid case study identifies conditions under which staging offers no advantage. ContiDesigner translates these design principles into an accessible workflow to explore feasible operating regions and prioritize cascade designs for experimental evaluation. Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=130 SRC="FIGDIR/small/743657v1_ufig1.gif" ALT="Figure 1"> View larger version (37K): org.highwire.dtl.DTLVardef@ef58faorg.highwire.dtl.DTLVardef@1ba88a4org.highwire.dtl.DTLVardef@160edd3org.highwire.dtl.DTLVardef@9dda34_HPS_FORMAT_FIGEXP M_FIG C_FIG O_LIContiDesigner enables system-level design of continuous fermentation cascades C_LIO_LIHigh stage-one dilution supports biomass generation C_LIO_LILow stage-two dilution preserves productive residence time C_LIO_LIYet two-stage cascades favor lower overall dilution than one-stage systems C_LIO_LITwo-stage advantage requires strong non-growth-associated production in stage two C_LI

Emerging efforts to engineer multicellular kidney tissues for drug discovery, disease modeling, and therapeutic repair require scalable generation of kidney organoids derived from human induced pluripotent stem cells (hiPSCs). Here, we report a method for generating kidney ureteric bud and collecting duct epithelial organoids in stirred tank bioreactors. We demonstrate that extracellular matrix microcarriers facilitate organoid expansion and are compatible with suspension bioreactor methods. The combination of cell-coated matrix microcarriers with stirred bioreactors promotes maturation marker expression compared to static controls comprising matrix-embedded organoids. Our method offers promise for generating epithelial organoids that serve as multicellular models and organ building blocks for scalable biofabrication.

Background: Hydroxyectoine is a high-value compatible solute with applications in cosmetics, healthcare, and biotechnology. Its microbial production has traditionally relied on halophilic organisms requiring high-salinity cultivation, motivating the development of non-halophilic production hosts. Corynebacterium glutamicum is an attractive platform because of its industrial robustness and high metabolic capacity through the aspartate-family amino acid pathway. Results: A synthetic hydroxyectoine pathway was constructed by combinatorially varying the predicted translation initiation rates of four translational units comprising ask, ectAB, ectC, and ectD. A library of C. glutamicum transformants was screened at 40{degrees}C using hydroxyectoine-associated thermoprotection as a functional selection principle. Fast-growing variants generally displayed increased hydroxyectoine formation and higher hydroxyectoine-to-ectoine ratios than a strain carrying the native pathway configuration. The selected combination reached a hydroxyectoine-to-ectoine ratio of 4.7 and was transferred into the lysine-producing strain DM1729SL in order to increase precursor availability and hydroxyectoine production. Carbon-limited fed-batch cultivations showed that both temperature and dissolved oxygen influenced production. The best performance was obtained at 30{degrees}C and 50% relative dissolved oxygen, yielding 12.9 g/L hydroxyectoine with a yield of 0.161 g/g glucose and a volumetric productivity of 0.26 g/L/h after glucose depletion. A subsequent hypoosmotic downshock increased extracellular hydroxyectoine recovery to 14.5 g/L. Conclusions: Thermal stress-based combinatorial pathway balancing enabled efficient de novo hydroxyectoine production in C. glutamicum. Integration of precursor-enhanced host metabolism, controlled oxygen supply, fed-batch cultivation, and osmotic downshock resulted in a substantial improvement over previously reported de novo hydroxyectoine production in this host and establishes C. glutamicum as a promising platform for further hydroxyectoine process development.

Methyl 12-aminododecanoate (ADAME) is a key precursor for Nylon 12 synthesis and an attractive target for sustainable microbial production. However, efficient biosynthesis of ADAME requires the coordinated oxidation and transamination of methyl dodecanoate (DAME), and identifying compatible enzymes for this multistep pathway remains challenging. In this study, we applied a machine learning-based bioinformatic screening approach to identify alternative enzymes for DAME-to-ADAME bioconversion. Among 18 selected distantly related AlkB homologs, AlkBGp01 from Alcanivorax sp. P2S70 showed activity toward DAME, producing 12-hydroxydodecanoic acid methyl ester (HDAME) and 12-oxododecanoic acid methyl ester (ODAME) despite low sequence identity to Pseudomonas putida AlkB. Further screening identified compatible redox partners, AlkG60 from the same species as AlkBGp01 and AlkTp04, whose combination with AlkBGp01 resulted in almost sixfold higher ODAME production than that achieved with P. putida AlkBFGJLT. ODAME production was further increased by more than threefold by conjugating these three newly identified proteins. The optimized single-plasmid system comprising conjugated AlkBGp01, AlkG60, and AlkTp04, together with AlaD and newly identified omega-transaminases EAV41574 produced 0.28 mM gDCW-1 ADAME, representing a 5.6-fold improvement over the previously reported P. putida AlkBFGJLT-based system with AlaD-CV2025. These results demonstrate that machine learning-guided enzyme discovery can identify functional distantly related homologs and compatible enzyme combinations for constructing efficient synthetic pathways. This study also highlights the importance of optimizing pathway architecture, including enzyme conjugation and plasmid configuration, for improving microbial production of bio-based monomer precursors.

Valerolactam is a promising bio-based monomer for polyamide synthesis, but microbial production remains limited by inefficient 5-aminovalerate (5AVA) cyclization, by-product formation, and insufficient process optimization. In this study, Corynebacterium glutamicum was engineered for valerolactam production by expression of davBA from Pseudomonas putida and lysP from Escherichia coli, followed by conversion of 5AVA to valerolactam using the recently identified avaC gene from Collinsella intestinalis. Flask cultivations confirmed efficient valerolactam formation with only minor accumulation of the by-products L-lysine, 5AVA, and glutarate. Batch bioreactor experiments showed that higher glucose concentrations increased titers but also promoted by-product accumulation, whereas increasing the dissolved oxygen setpoint from 30% to 50% improved growth and volumetric productivity. Intracellular cofactor analysis revealed declining energy status and shifts in redox balance during production. Based on these findings, carbon-limited fed-batch cultivation at 50% rDO improved production performance, reaching 3.6 g/L valerolactam with a yield of 0.231 g/g and a volumetric productivity of 0.075 g/L/h, while minimizing by-product formation. These results establish AvaC-based C. glutamicum as a promising platform for sustainable valerolactam biosynthesis.

Microbial fuel cells (MFCs) have been widely investigated as decentralized bioelectrochemical systems capable of converting organic substrates into electricity. However, their long-term autonomous operation is constrained by substrate depletion in the anode compartment, leading to metabolic starvation of electroactive biofilms and a decline in power output. Conventional MFC design treats substrate crossover through the membrane separator as a parasitic loss that reduces coulombic efficiency. In this work, we propose a conceptual inversion of this paradigm by considering controlled cathodic-to-anodic substrate crossover as a passive mechanism to sustain basal microbial metabolism during periods of substrate scarcity. A transport-reaction framework is developed to quantify the balance between membrane-mediated substrate flux and microbial maintenance demand within the anode biofilm. Based on this balance, a dimensionless maintenance crossover Damkohler number (Dam) is introduced to define three operational regimes: starvation-dominated (Dam >> 1), balanced autonomous (Dam {approx} 1), and crossover-dominated (Dam << 1). The framework integrates membrane transport theory with biofilm kinetics to evaluate the effects of separator properties, substrate gradients, and current-dependent electro-osmotic transport on system stability. Order-of-magnitude analysis indicates that achievable crossover fluxes span several orders of magnitude depending on separator characteristics, suggesting that membrane properties critically influence system behavior. This perspective reframes substrate crossover from a loss mechanism to a potential design variable, offering a conceptual tool for enhancing resilience and guiding separator selection in MFCs intended for long-duration, and low-maintenance operation. HighlightsO_LIControlled crossover can sustain microbial metabolism in MFCs C_LIO_LIIntroduces maintenance crossover Damkohler number (Dam) C_LIO_LIIdentifies regimes for autonomous and starvation operation C_LIO_LILinks membrane properties to long-term system stability C_LIO_LIReframes crossover as a design variable, not only a loss C_LI

Pluripotent stem cell-derived pancreatic islet-like cell clusters (SC-islets) have emerged as potential cellular therapy for type 1 diabetes. While SC-islets can theoretically be produced in bioreactors to meet transplantation needs of many recipients per batch, biomanufacturing and translational challenges remain. SC-islet can be cultured in suspension using stirred tank or vertical wheel bioreactors, but these can impart hydrodynamic damage and lead to cellular agglomeration, particularly upon scale-up. We previously described a robust emulsion-based process to encapsulate murine pancreatic beta cells in high-concentration alginate beads which improved graft survival in allogeneic recipients. Here, we present a full pipeline for scalable microencapsulated SC-islet biomanufacturing with extended in vitro bioreactor culture leading into transplantation. We hypothesized that encapsulation would prevent cellular agglomeration, reduce mechanical stress, and preserve differentiation potential during scale-up. Encapsulation of Stage 6 SC-islets prevented cellular agglomeration during extended suspension culture (25 days) and increased cell recovery (91 +/- 3%) compared with non-encapsulated aggregates (60 +/- 10%). No significant differences in glucose-stimulated insulin secretion were observed with vs without microencapsulation. Stage 7 SC-islets matured in the bioreactor were transplanted either as encapsulated islets via the intraperitoneal route or as free SC-islets under the kidney capsule, where they displayed glucose-responsive human C-peptide secretion and remained functional in vivo for up to 98 days. Overall, this work establishes a scalable, robust, transplantation-ready encapsulation platform that supports SC-islet maturation and delivery, providing a generalizable strategy for scaling and transplanting encapsulated organoid systems.

Endothelial cells (ECs) derived from human induced pluripotent stem cells (hiPSCs) are increasingly used to model vascular diseases and test therapeutic strategies. However, the efficiency and reproducibility of differentiation can vary depending on the culture medium and its supplemented factors and stages. Here, we directly compared two defined media, APEL and BPEL, for iPSC-to-ECs differentiation. iPSCs were differentiated over 10 days with sequential growth factor induction, followed by magnetic-activated cell sorting or flow cytometry for CD31+ cells. Both media produced ECs with similar morphology and marker expression, including CD31 and VE-cadherin. Functional assays demonstrated comparable tube formation, indicating equivalent endothelial functionality. Cost analysis indicated that APEL had a higher total reagent cost but generated a higher total cell yield, resulting in a comparable cost per 10 total cells, whereas BPEL was more cost-efficient for producing CD31/VE-cadherin endothelial-specific cells. Our results suggest that APEL and BPEL media are equally effective for generating iPSC-derived ECs, providing flexibility in method selection for vascular disease modeling and drug discovery applications.

Protein engineering and precision fermentation provide an opportunity to increase the value of food proteins by improving their solubility, stability, functionality, or nutritional composition. Here, we use {beta}-lactoglobulin ({beta}LG) as a model protein to investigate how state-of-the-art computational protein design approaches affect these properties. First, the deep learning-based design tool ProteinMPNN was used to alter up to 20% of {beta}LG residues for increased stability. Second, the physics-based modeling platform PyRosetta was used to find positions in {beta}LG accommodating increased branched-chain amino acid (BCAA) content and up to 10 residues were simultaneously exchanged. Experimental characterisation of ProteinMPNN and stabilised BCAA-enriched variants showed similar secondary structure and oligomeric state as native {beta}LG. ProteinMPNN variants gave increased titers and increased thermal stability up to 15 {degrees}C, and this correlated with changes in the rate of surface pressure in droplet tensiometry. Stabilized BCAA-enriched mutants had altered acid solubility. Correlations between computationally derived biophysical metrics and experimental properties are presented and suggest some predictive power for surface hydrophobicity on protein yield.

Machine learning-guided protein sequence redesign is now routinely used to optimize multiple properties relevant for protein engineering, most prominently thermostability and recombinant expression levels. Salt tolerance is a valuable property for "blue-biotechnology"- enabled biomanufacturing, yet no generative computational method exists to redesign proteins for increased salt tolerance. We hypothesized that a training dataset heavily biased toward salt-adapted proteomes would yield a model capable of designing proteins with halophilic properties. To test this, we retrained the sequence redesign model ProteinMPNN on proteins from "salt-in" extreme halophiles such as Haloarcula marismortui, a Dead Sea archaeon that grows optimally near 3-4 M NaCl, roughly six times the salinity of seawater, and accumulates molar concentrations of salts in its cytoplasm. Our model, HaloMPNN, redesigns non-halophilic proteins so that their properties shift towards those of natural halophilic proteins: lower predicted isoelectric point, greater surface acidity, and reduced surface and core hydrophobicity. Redesigning a broad range of non-halophilic proteins with SolubleMPNN, ProteinMPNN, and HyperMPNN shows that this shift is specific to HaloMPNN rather than a generic consequence of sequence redesign. HaloMPNN therefore offers both a route to designing candidate salt-tolerant enzymes and a means of identifying the characteristics that underlie halophilic adaptation.

Lack of robustness and high variability in cardiomyocyte (CM) differentiation from human induced pluripotent stem cells (hiPSCs) remain critical challenges for cardiac regenerative medicine and biological research. Guided by developmental principles of lineage specification, we investigated whether early WNT inhibition could expand the process window of CM differentiation and thereby improve robustness. To test this, we compared early WNT inhibition from days 1--5 with standard inhibition from days 3--5 following a 1-day CHIR99021 (CHIR) treatment for WNT activation. Focusing on CHIR concentration as a critical parameter, we found that early WNT inhibition substantially broadened the concentration window yielding [&ge;]80% differentiation efficiency across distinct hiPSC lines. Extending the analysis to a two-dimensional parameter space incorporating cell density revealed a markedly expanded [&ge;]80% process window. Temporal analysis of lineage-specific transcription factor expression showed that early WNT inhibition suppressed paraxial mesoderm (PM) specification, likely through MSGN1 suppression, in the range of high initial WNT activation that otherwise favored PM fate, thereby extending the process window. It also suppressed definitive endoderm (DE) specification within the expanded window, promoting a more favorable lateral plate mesoderm/cardiac fate. We further demonstrated substantially reduced variability under parameter perturbations and across wells and batches, accompanied by higher differentiation efficiency, without compromising CM phenotypes. These findings demonstrate early WNT inhibition as an effective strategy for expanding the process window of CM differentiation and improving robustness, while exemplifying the process engineering principle that a broader window can support more robust differentiation, with potential relevance to stem cell--based bioprocesses.

Lignocellulosic biomass biotechnological conversion relies on enzymatic hydrolysis of recalcitrant plant cell walls, which limits conversion efficiency and economic viability and renders sugar-release dynamics difficult to predict. Mathematical modeling has therefore emerged as a key approach for elucidating the mechanisms governing enzymatic hydrolysis and predicting its dynamics. Here, a hierarchical adsorption-inhibition framework is developed, beginning with a detailed Dynamic Adsorption-Inhibition Model (DyAIM). This model is then simplified into a Reduced Adsorption-Inhibition Model (ReAIM) by pruning weak adsorption and inhibition interactions. Finally, ReAIM is further simplified into an Effective Activity Model (EAM), in which nonproductive enzyme adsorption onto lignin is represented by a reduction in effective enzymatic activity. Plant cell wall composition is expressed as four structural polymers coupled to six soluble products, with inhibition by mono- and oligosaccharides acting on five functional enzyme pools. Using glucose, xylose and mannose release at two enzyme loadings, the model hierarchy is validated in reproducing saccharification dynamics and in delivering a quantitative product-enzyme inhibition network consistent with literature trends. The model framework was further validated by prediction of saccharification dynamics at an unseen enzyme loading, demonstrating robust performance beyond the calibration conditions. Sensitivity analysis further supported the progressive simplifications adopted in ReAIM and EAM. Maintaining predictive performance across increasing levels of simplification indicates that the hierarchy effectively distinguishes essential mechanisms from dispensable complexity, enabling rational model selection.

Recombinant adeno-associated virus (rAAV) manufacturing remains constrained by low yields, costly purification, and process complexity, limiting cost-effective access to gene therapy. Here, we demonstrate an integrated continuous biomanufacturing platform that transfers proven antibody-manufacturing technology to rAAV production by coupling perfusion to semi-continuous twin-column affinity capture (CaptureSMB). This architecture enables continuous harvest of extracellular rAAV without cell lysis, depth filtration, or endonuclease treatment, sustaining stable operation over five days. Relative to conventional batch processing, perfusion improved capsid or vector genome yield depending on the production process, while affinity capture alone reduced total DNA to batch-comparable levels, with no measurable loss in impurity clearance or potency. By consolidating multiple manual unit operations into one continuous workflow, the platform may reduce manual intervention and process footprint. Its modular design may generalize to other rAAV serotypes, inducible producer cell lines, and viral vector classes, offering a potential route toward more efficient gene therapy manufacturing.

Food production is a significant contributor to global greenhouse gas emissions and deforestation, exacerbated by substantial food waste. Converting food waste into yeast protein offers a sustainable solution to enhance food security and contribute to a circular economy. However, due to the diverse and variable nature of food waste substrates, numerous experimental trials are required to optimise the preprocessing steps, yeast strain selection, nutrient addition, and fermentation conditions. This study presents a hybrid modelling approach where data-driven machine learning is used to predict microbial growth kinetics from process parameters. The hybrid model was trained on a comprehensive dataset consisting of 963 fermentation experiments from 55 publications, enabling transfer learning across 46 yeast strains and 79 food waste substrates. The hybrid modelling method was integrated with Bayesian optimisation, a sequential strategy to optimise expensive-to-evaluate functions, to efficiently maximise yeast biomass growth from different food waste substrates. The utility of the hybrid model was evaluated using five test datasets selected from previous literature and was shown to facilitate an average reduction of 66% in the number of experimental trials required to identify optimal fermentation conditions compared to without using the hybrid model. This proved that the transfer of knowledge between yeast strains and food wastes improved the optimisation efficiency of real, previously published datasets compared to traditional optimisation methods. The novelty and contributions of this study include the collation of the extensive dataset, provided as supplementary material; and the demonstration that transfer learning by training the hybrid model on this heterogeneous dataset can improve the optimisation efficiency for yeast biomass growth on new strains and substrates.