Biomanufacturing has evolved from producing simple recombinant hormones to assembling highly complex biologics in living systems. Yet no matter how sophisticated the molecule becomes, a single truth remains: the earlier manufacturability is addressed, the fewer expensive surprises appear downstream.
Professor Susan Sharfstein, a chemical engineering veteran whose career spans Caltech, UC Berkeley, medical school research, and two decades of academic-industry collaboration, summarizes hard-won lessons for scientists who may never set foot inside a 2,000 L bioreactor facility but still make decisions that determine whether a drug candidate is easy or impossible to manufacture.
This concept is discussed in greater detail in an episode of the Smart Biotech Scientist Podcast, hosted by David Brühlmann, founder of Brühlmann Consulting.
Why “Manufacturability” Starts on Day One
Beyond Efficacy
Early-stage teams often focus on efficacy, safety, and immunogenicity, all critical patient attributes. Sharfstein emphasizes a parallel question: which candidate will a process-development team enjoy producing at scale?
- If five molecules deliver the same therapeutic benefit, the one that folds cleanly, expresses at high titer, and purifies without drama should rise to the top well before IND filing.
- Selecting a complex molecule locks R&D groups into expensive process development rescue projects, extensive cell-line re-engineering, non-platform purification steps, or formulation gymnastics to prevent precipitation.
What “Manufacturability” Covers
In Sharfstein’s experience, manufacturability spans three interconnected zones:
- Upstream: Can the host cell line express and secrete the protein without chronic stress?
- Downstream: Does the protein “behave” on Protein A resin, ion exchange, and polishing columns?
- Formulation: Does it remain soluble and stable at clinical or commercial concentrations? A weakness in any zone can derail timelines, especially when speed-to-clinic trumps exhaustive troubleshooting.
Choosing a Cell Factory: Lessons from CHO, Hybridomas, and Plasma Cells
CHO Cells Still Dominate for Good Reason
Chinese hamster ovary (CHO) cells don’t naturally crank out grams per liter of therapeutic protein, yet industry standards evolved around CHO because the cells:
- Adapt readily to suspension culture.
- Tolerate high osmolarity and mild hypothermia shifts.
- Provide human-compatible glycosylation patterns after careful process control.
When Natural Biology Misleads
Researchers sometimes assume that B-cell-derived hybridomas or plasma cells should outperform CHO because these immune cells natively secrete antibodies. Sharfstein’s laboratory and collaborators found the opposite:
- Hybridomas often underperform in bioreactor conditions; they were not selected for large-scale robustness.
- Plasma cells produce astonishing titers in vivo, yet they stop proliferating and die quickly, unacceptable for a 14-day fed-batch run.
Take-home: evaluate expression hosts by fundamental process criteria, growth, stability, and scalability, not by their biological origin story.
I think that one of the things that people broadly think is that pharmaceutical companies and biopharmaceutical companies, they're only in it for the money. And one of the things when you engage with particularly the scientists and the process development, that they are really passionate about the impact that their drugs have on people's quality of life.
Metabolic Surprises and Stress-Driven Productivity
Glucose, Glutamine, and Unexpected Boosts
While studying hybridomas, Sharfstein’s team decreased glutamine concentrations and saw an increase in secreted antibodies, which was counterintuitive because glutamine is a classic growth substrate. The likely reason is that metabolic stress triggers pathways that enhance secretion.
Temperature-Shift Findings
In collaboration with UCB Celltech, two antibody variants differed by a single framework-region residue: alanine versus glycine.
The alanine version was expressed 10–100-fold higher. Dropping culture temperature rescued expression of the weaker glycine variant, suggesting that folding-rate limitations can be alleviated by mild hypothermia.
Implication: Dialling process parameters, temperature, and osmolarity can partly offset sequence-imposed folding stress, but only if the development team catches the issue early.
Early Red Flags to Catch in the Lab
| Category | Practical Questions to Ask in Shake Flasks or Ambr® Bioreactor Systems |
| Expression | Does the clone reach ≥ 30 pg/cell/day by day 7 without drastic viability loss? |
| Folding / Stress | Do ER-stress markers spike under standard feed? Do temperature shifts improve titers? |
| Glycosylation | Do key glyco-enzymes down-regulate late in culture, leading to heterogeneity? |
| Chromatography | Does the molecule bind Protein A atypically or co-elute with host proteins? |
| Formulation | Does the purified protein aggregate above 100 mg/mL, complicating high-dose delivery? |
- Fail fast: Reject candidates that fail multiple checkpoints before scale-up.
- Iterate smartly: Minor mutations (e.g., substituting alanine for glycine, adjusting CDR charges) can transform an “impossible” molecule into a manufacturable one.
AI and High-Throughput Screens, Promise and Constraints
A New Collaboration with DeepSec AI
Sharfstein’s group is partnering with DeepSec AI to train models that predict expression success:
- Library size: 10⁶ single-domain VHH sequences will be synthesized and transfected into CHO.
- Data loop: measured titers feed back into generative algorithms, refining predictions for future libraries.
- Goal: forecast which novel sequences will express at industrial titers before any wet-lab validation.
Why More Data Still Matters
Despite AlphaFold’s structure revolution, AI models remain only as good as their training data. Sharfstein foresees a positive spiral:
- Cheaper DNA synthesis plus robotics enable larger expression screens.
- Richer experimental datasets strengthen AI predictions.
- Better predictions provide confidence to synthesize bold new libraries, restarting the cycle.
Anticipated Constraints
- Expression does not equal folding alone; a sequence that folds stably in silico may misbehave inside mammalian secretory pathways.
- Without retraining, cell-line specificity models trained in CHO may not extrapolate to HEK293 or novel hosts.
- AI must eventually incorporate chromatography and formulation data for downstream unknowns to provide a full manufacturability score.
Concrete Tips for Process-Minded Scientists
During Clone Selection
- Screen temperature and osmolarity early and discover stress conditions that elevate titers without risking viability.
- Evaluate multiple host lines; a platform CHO line may falter for unusual formats and have backups ready.
- Watch framework mutations, seemingly benign residues that can slash expression, and verify in mini-pools.
I think that I would give my 25 year old self the advice that be really, really open to everything because you have a lot to learn and what you know now is only really a small piece of what you're going as you go out through your career.
During Lead Optimization
- Leverage small mutations, swap charged residues to improve Protein A binding, and adjust variable-region surface patches to reduce aggregation propensity.
- Run parallel downstream screens, simple microscale Protein A tests reveal elution oddities early, and high-throughput solubility assays flag formulation risks.
During Tech Transfer
- Document stress findings; if a 32 °C shift is critical, capture exact timing and ramp rates for CDMO partners.
- Align quality attributes and ensure glycan patterns or charge variants observed at 2 L also meet specifications at 2,000 L.
- Plan for AI integration and share sequence–titer datasets internally; future ML tools thrive on well-curated historical data.
Key Takeaways from Sharfstein’s Perspective
- Manufacturability is a primary design criterion; selecting an easy molecule can save millions and shave months off development.
- Nature provides clues, not blueprints. Plasma-cell biology inspires secretion engineering, but direct transplantation into CHO often fails.
- Subtle sequence choices matter; single amino-acid swaps can dictate ten-fold titer differences.
- Stress can be harnessed; glutamine limitation or mild hypothermia may boost productivity by easing the folding load.
- AI needs experimental fuel; large-library screens in CHO will transform predictive power, but bench data remains irreplaceable.
- Cross-disciplinary mentorship accelerates growth; Sharfstein credits mechanical, chemical, and life-science mentors for her holistic approach.
Final Remarks
Sharfstein envisions biomanufacturing as a virtuous cycle. Robotics and cheap gene synthesis drive vast experimental datasets, AI converts those datasets into predictive manufacturability models, and scientists feed new hypotheses back to the lab, tightening iteration loops and shrinking risk.
Process scientists, data scientists, and molecular engineers will jointly decide which candidate antibodies, VHHs, or multispecifics deserve precious GMP downstream capacity in that ecosystem.
About Susan Sharfstein
Susan Sharfstein is a Professor of Nanoscale Science and Engineering at the University at Albany in Albany, New York. Professor Sharfstein received her B.S. in chemical engineering with honors from Caltech in 1987 and her Ph.D. in chemical engineering from UC Berkeley in 1993. She received an NIH fellowship to pursue postdoctoral studies, initially at UC Berkeley and subsequently at UCLA Medical School.
Professor Sharfstein’s research interests include mammalian and microbial cell bioprocessing, control of protein glycosylation, metabolic engineering, biosensing, development of systems for high-throughput screening, and manufacturing of biohybrid electronic and photonic devices. She was the recipient of an NSF CAREER grant for her studies of monoclonal antibody production. She is the author of over 85 papers and book chapters in the fields of biotechnology and bioprocessing.
She was a 2017-18 recipient of a Fulbright Global Scholar award and spent her sabbatical at Dublin City University, in Dublin, Ireland and University of Queensland in Brisbane, Australia performing proteomic analysis of Chinese hamster ovary cells and studying bispecific antibodies. Dr. Sharfstein was a 2023 recipient of the SUNY Chancellor’s Award for excellence in Scholarship and Creative Activities.
Connect with Susan Sharfstein on LinkedIn.
David Brühlmann is a strategic advisor who helps C-level biotech leaders reduce development and manufacturing costs to make life-saving therapies accessible to more patients worldwide.
He is also a biotech technology innovation coach, technology transfer leader, and host of the Smart Biotech Scientist podcast—the go-to podcast for biotech scientists who want to master biopharma CMC development and biomanufacturing.
Hear It From The Horse’s Mouth
Want to listen to the full interview? Go to Smart Biotech Scientist Podcast.
Want to hear more? Do visit the podcast page and check out other episodes.
Do you wish to simplify your biologics drug development project? Contact Us
