Scalable niomanufacturing of recombinant enzymes for ex vivo conversion of group B red blood cells into ABO O-like cells

What do I want to biomanufacture, and why?

I propose the scalable biomanufacturing of a recombinant enzyme cocktail capable of removing group B antigens from donated red blood cells (RBCs), producing cells with an O-like phenotype within the ABO system. Blood group compatibility depends partly on carbohydrate antigens displayed on the RBC surface. In emergencies and blood shortages, group O RBCs are especially valuable because they lack A and B antigens. However, O-negative and other group O supplies are limited. In 2024, Jensen et al. identified exoglycosidases from the mucin-degrading gut bacterium Akkermansia muciniphila that can remove canonical and extended A and B blood group antigens. The study reported improved compatibility between enzyme-treated RBCs and group O plasma. This suggests that gut microorganisms can provide enzymes for expanding the supply of ABO-compatible blood. My proposed product is not a living bacterium administered to a patient. It is a purified recombinant enzyme cocktail used ex vivo in a controlled blood-processing system. The proposal focuses on group B RBCs because their enzymatic conversion currently appears more efficient and requires less enzyme than the conversion of group A cells. The resulting cells should be described as ABO O-like, rather than universally compatible blood, because other clinically important antigens, including Rh and Kell, would remain present.

What living system would manufacture it

Genes encoding selected A. muciniphila exoglycosidases would be synthesized, cloned and expressed separately in an established recombinant protein-production chassis such as Escherichia coli BL21(DE3). The engineered bacteria would be cultivated in stirred-tank bioreactors. After recombinant expression, the cells would be harvested and lysed. Each enzyme would be purified using chromatography and filtration, followed by endotoxin removal. The purified enzymes would then be combined at an optimized ratio and formulated as a standardized cocktail. The manufacturing workflow would therefore be: Enzyme-gene selection and plasmid construction. Transformation of the production strain. Fermentation and recombinant expression. Cell harvesting and lysis. Enzyme purification and endotoxin removal. Cocktail formulation and activity testing. Ex vivo treatment of group B RBCs. Washing and quality control of the treated cells.

How could automation improve the process?

Automation would be useful during both development and manufacturing. During development, a liquid-handling robot could prepare small expression cultures and test combinations of induction temperature, induction time and media composition. It could also create different enzyme ratios and concentrations in microplates. Automated assays could compare antigen removal at different treatment times and temperatures. Flow cytometry, antibody-based agglutination tests and compatibility tests with group O plasma could be used as readouts. Images and experimental metadata would be collected automatically so that the best conditions could be selected through a design-build-test-learn cycle. At manufacturing scale, sensors and control software would regulate temperature, pH, dissolved oxygen, agitation and feeding in the bioreactor. Automated chromatography, tangential-flow filtration and endotoxin testing would improve consistency between batches. A separate closed blood-processing device could control enzyme addition, incubation, washing and sampling of each RBC unit.

What would be a meaningful scale?

A meaningful initial target would be enough purified enzyme cocktail to treat 1,000 units of group B RBCs in a pilot blood-bank study. Jensen et al. estimated that approximately 8 mg of the limiting enzyme would be needed to treat one 200 mL unit of group B RBCs. Therefore: 1,000 RBC units × 8 mg per unit = 8,000 mg = 8 g of purified enzyme cocktail For an early process estimate, I will assume: Recombinant enzyme yield: 0.5 g/L of fermentation culture. Overall purification and formulation recovery: 50%. Required final product: 8 g. The required fermentation volume would be: 8 g ÷ (0.5 g/L × 0.50 recovery) = 32 L Because the enzymes would initially be produced separately, approximately two 20 L fermentations could provide enough material while allowing for process losses, quality-control samples and failed batches. A tentative manufacturing cycle could take: 1–2 days for fermentation and enzyme expression. 1 day for harvesting and purification. 1 day for formulation and quality control. This gives approximately 3–4 days per enzyme-production batch. Treatment of an individual RBC unit could require approximately 30 minutes of enzyme incubation, followed by washing and quality-control testing. The production yield, recovery and processing times are preliminary assumptions that would need experimental validation.

Important unknowns

Completeness and donor variability: Can the enzyme cocktail consistently remove canonical and extended group B antigens from RBCs obtained from genetically diverse donors? RBC safety and quality: Does enzyme treatment or subsequent washing affect membrane integrity, oxygen transport, deformability, storage life or transfusion survival? Manufacturing and clinical safety: Can the enzymes be produced economically with reproducible activity and acceptably low concentrations of endotoxin, host-cell proteins and residual enzymes? Before clinical use, the treated RBCs would require extensive preclinical validation, compatibility testing, storage studies and controlled clinical trials.

References

Jensen, M. et al. (2024). Akkermansia muciniphila exoglycosidases target extended blood group antigens to generate ABO-universal blood. Nature Microbiology, 9, 1176–1188. https://doi.org/10.1038/s41564-024-01663-4 Rahfeld, P. et al. (2019). An enzymatic pathway in the human gut microbiome that converts A to universal O type blood. Nature Microbiology, 4, 1475–1485. https://doi.org/10.1038/s41564-019-0469-7 Petzold, C. J. & Mukhopadhyay, A. (2025). From bench to biofactory: high-throughput technologies and automated workflows to accelerate biomanufacturing. Current Opinion in Biotechnology, 94, 103320. https://doi.org/10.1016/j.copbio.2025.103320

Use of AI

I used ChatGPT by OpenAI to facilitate the search for relevant scientific papers, help me identify information needed for the scale calculations, and improve the grammar and clarity of my writing because English is not my native language. I reviewed the cited sources and made the final decisions regarding the proposal, assumptions and calculations.

Week 2 HW: DNA Design

HTBAA Homework #2

Project: Recombinant AmGH110A for ex vivo removal of the group B antigen.

Part 0: Set up in Kernel

I created the repository HW2_Gabriela Vera_AmGH110A and the folder vera-amgh110a in the course workspace.

Part 1: Define your expression target

What molecule are you expressing?

I selected AmGH110A, an α-(1,3)-galactosidase from Akkermansia muciniphila associated with removal of the canonical group B antigen from red blood cells (Jensen et al., 2024).

What is the sequence?

The target is gene Amuc_0480 (glaB), corresponding to protein records GenPept ACD04318.1 and UniProtKB B2UNU8; my native coding sequence was derived from NCBI genome accession CP001071.1. The saved CDS is 2,385 bp including its terminal TAA codon and encodes 794 amino acids.

Sequence: ATGGAAGGCAATCTCTCTTTTTCTCTGATGGAAGCTTCTGGACGATCCATATTTTTTCTGATAGAAGGAA TACGGGAACAGAGTATAAAAAATATGTTCAGCCGGATGTTTTCATGGAGTTTTGTAGTTGCAGCGTGTTT GGCGGGATTATTCCCGGCACAGTCCCAGGGCGAGGAAAAAGCTGCCCAATCCGGAACCATAGCCGTAAAA GTTCCTGCCTCCAGCCTGCTGATGACTCGCCAGGAGACGGGGGAAACCCGGTTGGACCGGTCTTTCAGTA ATGCGGGGTTGAGCATAGGCGGCAAAAAATATGCCACGGGAATCGGTACGCACGCTACGTCCATGATTCC CCTGCCCGTTCCGGAAAACCCGAAAGTTTTGAGGCTGGAAGGTGCATGCGGAATTGATGACGGTGCGGAT GGAGACGGAAGCGTGGAATTCCGCGTGATGAGCGGGTCCGAGGTTTTATGGAGTTCCGGCGTGATGAGGA GGGGGATGGCCGCGAAGAAGTTTTCCATTCCGGTGGCGGAAAACGGCATACGGCATCTTTACCTGATGGC CGACCGGGTGGACAATAATTCTTACGACCATGCGGACTGGGTAGATCTGGCCTGGAAGACGACAGGAAGT GGGCAGGGGATGAAGGGAGCTGTGGTGAATGCCTCTGAATTCGGCATGGTTCCCGGCGTCAGAAAAGATC AGGGGCCGGCGCTGCGGGCTGCCGTTTCCGCTCTCCGGAGGCAGGGCGGGGGCGTTTTGAACATCCCCAG AGGCATTTACCATTTTTATCCGGAGGGGGCTTTGAACATGAGTTTCCATATTTCCAACCATGACCAGCCG CTGATTCATCCGGTGTGTGTGCCTCTGGCCGATTTGCGGAATGTCCGCGTGGAGGGCAATGGTTCCCTGT TTCTTTTTCACGGCAAGGTGGTTCCCCTCCTGGTGATGGACAGCGAAAATGTCAGCATCAACCGTTTATC CGTGGATTACGAACGGTCCTGGTGCACGGAGGCGCGCGTCGTGAAGACGGATGACCGGTTCACGGAAGTG GAAATAGACAAAAAGGCCTACCCTTACGAAATCCGGAACAACAGGTTCGTTTTCCAGGGGAAAGGCTGGG AAGAAGGAATGGGAAGCTGCATGGCCTTTGAGAAAGGGACGGGACATATTATCGCCAATACATCGGACAT CGGCTGGAACGGGCATGTGGAGCCGCTGGGCGGGAGCCGTCTCCGTCTGTCCTGGAACCTCAGGCAGAAG GGAATCAAGCCGGGGGATACGCTGGTTCTTCGGAATTATAACCGCCCCCATCCGGGATGCGTGGTGTACC GGGCTCGGAAAACATCACTGAATGACGTATCTCTGCACCAGAGCTCAGGCATGGCCCTGCTGGTCCAGCG ATCGGAGGACTTTCATATGAAAGGCGGCGGAGTCATGGTCAGGAAAGGCACAGGGCGCGTACACACCGCC GGAGCGGACGCAACCCATTTTTCCAATACCCGGGGCGGAATCGTGGTGGAGAAAGCCCTGTTCGAAGGGA TGATGGATGATGCCATCAATGTTCATTCCACCTGCCTGGGCGTGATGGAAGTAGTGGACAGCCATACCCT GAAGTGCAAATACATGCATCGGCAGGCGGTGGGGTTTGAGGTGTTTCTTCCCGGTGAAAAAATCCGTTTC ATCAACGGTCCTACGCTGGAACCCGGCGGAACGGCCACTGTGAAGACGGCAGTGAAAAAGAATTCCGCGG AGATGGTGATTACGGTGGAAGAGCCGCTCCCCTCCTCCGTCAGGGCCGGGGATGCCGTGGAGAATGCGGA TTTTTACCCTTCCGTGGTTTTCCGCAACAATATCGTCCGCAACAACCGTGCCAGAGGATCTCTTTTTACA ACGCCGGAGAGAGTGCTGGTGGAGGGCAATTTGTTTGACCATTCCTCCGGCTCCGCCATTCTATTGGCAG GGGATGCCCAGGGGTGGTATGAAAGCGGCGCCTGCCATGAAGTGGTGATCCGCAAAAATACATTCATTAA CAACCTGACCTCGCGTTACCAGTTTACGAATGCCATCATTTCCATTTACCCGGAAGTCAAGCAGCTGGAC AGGCAGAGGGATTACTATCATCGCAATGTGCTGATAGAAAATAATGTTTTCAAGACGTTCGATGTGCCGC TGCTGTTCGCCATTTCCACGGACAACCTCAAGTTCATCAATAATAAGGTCATTTACAATGACGAGTTTAA GGGATGGGGGCAGAAACCTTTCCAATTCAGAAGATGCGCCAATATTCTGATTAAAGATAACAAGGTGCTG CCTCCCCGCACATGGACCCTTGAGGACTGCAAGCTGGAAAATACCCCATCAGATCAGGTCCGCTTTGGTG GATAA

What host or chassis? The production host is E. coli BL21(DE3), as proposed in Homework 1

Where does the product need to go?

My goal is to produce AmGH110A in E. coli BL21(DE3), purify it, and use the purified enzyme to treat red blood cells ex vivo. The saved CDS retains a predicted signal peptide, so the enzyme’s location in the bacterial cells has not yet been confirmed.

Part 2: Design your genetic parts

No purification tag was added to this construct.

Part 3: Assemble your construct

In Kernel, I assembled pSB1C3 with the T7 promoter → Shine–Dalgarno RBS → native AmGH110A CDS → T7 terminator. The exported construct is 4,537 bp and is annotated as circular. The parts appear in the intended order and orientation on the Kernel map. The CDS has a start and terminal stop, but Kernel’s warning about its ending and the assembly junctions still need review.

Figure 1. Kernel schematic of the 4,537 bp pSB1C3–AmGH110A construct.

Part 4: Optimize and quality-check

Codon-optimize

I used Kernel to optimize the native AmGH110A CDS for expression in E. coli. The optimized sequence still encodes the same 794 amino acids, but 485 codons changed and its GC content decreased from 53.8% to 43.8%. It is 3 bp longer because Kernel added another stop codon. Codon choice can affect how efficiently E. coli produces the protein, although this result does not show whether the enzyme will be active. I saved the optimized sequence separately; the construct shown in this report still contains the native CDS. Figure 2. Kernel codon optimization output; full input and optimized sequences are in the CSV export

Search your sequence

I searched the 2,385 bp native AmGH110A CDS in Kernel. The strongest hits were my saved CDS and the CDS within my cassette, each showing 100% identity and full query coverage with E-value 0. Other library results were short partial matches. This confirms consistency among the objects I saved, but it is not an independent external identification of the enzyme. The original NCBI accession and Jensen et al. support the target assignment. Figure 3. Kernel sequence-search results for the native CDS

Run the safety screen

I submitted the full 4,537 bp construct, including pSB1C3 and the AmGH110A expression cassette, to the SecureDNA biosecurity sequence scanner. It reported “No flagged sequences detected.” This is a sequence-screening result, not a test of whether the enzyme will be expressed or active. Figure 4. SecureDNA biosecurity screen of the full 4,537 bp construct.

Part 5: From design to build

How would you build it?

For a prototype, I would order the 2,467 bp AmGH110A cassette as synthetic DNA and assemble it into a verified pSB1C3 backbone. I chose this approach because the cassette is a few kilobases long and already contains the expression elements. Gibson assembly can join DNA fragments with matching ends, but I would first need to confirm the backbone and design the correct junctions (Gibson et al., 2009). What does synthesis constrain?

DNA synthesis can introduce errors and may be difficult in regions with repeats or extreme GC content. I would review the full 4,537 bp sequence for these features and sequence the assembled plasmid to check for mistakes. If the cassette needed to be synthesized in smaller fragments, error correction could help before assembly (Carr et al., 2004).

Prototype vs. scale.

For a prototype, I would make and verify a small number of plasmids to test enzyme production. At a larger scale, the main need would be to produce and purify much more enzyme through fermentation. I would not need to synthesize a new plasmid for every unit of red blood cells.

Part 6: What don’t you know yet?

Are the final sequence and junctions correct?

I added pSB1C3 in Kernel, and the 4,537 bp GenBank export is annotated as circular. I would inspect both insertion junctions and resolve Kernel’s warning about the end of the CDS, then sequence the physical plasmid after assembly.

Will E. coli produce an active enzyme?

The saved CDS includes a predicted signal peptide, and I have not added a purification tag. I would compare the full-length protein with a version lacking the signal peptide, then measure how much soluble, active enzyme each produces. I would also test whether adding a His₆ tag helps with purification without affecting activity.

Is AmGH110A alone enough for the intended red-cell treatment?

I would test treated red blood cells for remaining B antigens. If antigens remain, I would investigate whether another enzyme is needed.

AI

I used ChatGPT by OpenAI to help me navigate some steps in Kernel, organize the information in this report, and improve the grammar and clarity of my writing. I also used it to explore another possible backbone after finding that adding my complete cassette to pET28a might duplicate expression elements. That search led me to select pSB1C3, which I added to my final construct in Kernel. Examples of questions I asked, paraphrased in English, were: “How should I interpret this Kernel result?”, “Would adding my complete expression cassette to pET28a duplicate any elements?”, “What alternative backbone could accommodate the cassette?”, and “How can I make this explanation clearer and more concise?” I reviewed the suggestions against my saved Kernel results and the cited sources. The final design choices and remaining uncertainties are my responsibility.

References

Jensen, M. et al. (2024). Akkermansia muciniphila exoglycosidases target extended blood group antigens to generate ABO-universal blood. Nature Microbiology, 9, 1176–1188. https://doi.org/10.1038/s41564-024-01663-4

Gibson, D. G. et al. Enzymatic assembly of DNA molecules up to several hundred kilobases. Nature Methods 6, 343–345 (2009). https://doi.org/10.1038/nmeth.1318

Carr, P. A. et al. Protein-mediated error correction for de novo DNA synthesis. Nucleic Acids Research 32, e162 (2004). https://doi.org/10.1093/nar/gnh160

NCBI GenBank. Akkermansia muciniphila ATCC BAA-835, complete genome. Accession CP001071.1. https://www.ncbi.nlm.nih.gov/nuccore/CP001071.1

UniProtKB. Alpha-(1,3)-galactosidase B (glaB, Amuc_0480), accession B2UNU8. https://www.uniprot.org/uniprotkb/B2UNU8/entry

SnapGene. pSB1C3 sequence and map. https://www.snapgene.com/plasmids/basic_cloning_vectors/pSB1C3

Final Project