Eighty-five percent of the human proteome already has at least one monoclonal antibody built against it. Decades of biomedical research produced that library. Almost none of it works where a large share of the disease actually occurs, inside the cell.
A team at the University of Essex reported that the obstacle is more mundane than anyone had pinned down: electrical charge. Their paper, published January 31 in Nature Communications and pushed back into circulation by a fresh press cycle this month, describes a design rule plus AI-driven protein redesign that let them computationally reformat 672 antibodies as intrabodies, fragments built to fold and function in the cytoplasm.
A Library Locked Outside the Building
Antibodies evolved for the bloodstream. That environment is oxidizing, which lets disulfide bonds form and hold the molecule's structure together. The interior of a cell is reducing. Those bonds break, and the fragment misfolds, clumps, or gets degraded.
The practical consequence has shaped neurodegeneration research for years. Tau, TDP-43, alpha-synuclein, the polyglutamine proteins behind Huntington's disease, and SOD1 in motor neuron disease all do their damage inside cells. Conventional antibodies cannot reach them, and small molecules have proven remarkably poor at binding them.
Intrabodies have existed as a concept since the early 1990s. What has never existed is a reliable way to make one. Dropping a known antibody's variable domains into a single-chain format usually produces a nonfunctional protein, which is why functional intrabodies remain scarce despite the enormous antibody catalog available.
The Variable Nobody Had Isolated
The Essex group characterized 45 single-chain variable fragment intrabodies expressed in human cells and analyzed antibody variable domains at scale. Charge emerged as the dominant factor in solubility. Whole-molecule net charge at physiological pH showed a negative linear correlation with solubility across the range of +3 to minus 20, with an R-squared of 0.75. That beat charge measured at a lower pH, beat the isoelectric point, and beat all nine sequence-based solubility calculators they tested it against.
The scale of the mismatch is striking. Seventy percent of antibody variable fragments have an isoelectric point above physiological pH, which makes them badly suited to the cytoplasm. When the team predicted intracellular solubility across 68,551 fragments, only 0.02% of the unmodified fragments met the threshold for high solubility.
They then built new interdomain linkers carrying negative charge, optimized the orientation of the two variable domains, and identified an adjustable charge discrepancy between the heavy-chain framework and the binding loops.
That was the piece that made AI useful. Inverse folding tools, including software from the laboratory of Nobel laureate David Baker, can generate enormous numbers of candidate sequences. The charge rule narrowed the search space and ranked the output, turning a brute-force problem into a targeted one.
Lead author Caitlin O'Shea described the finding plainly in the university's announcement: antibodies usually have the wrong charge to exist inside cells without sticking together, and the redesigned fragments were built to carry the right charge and stay stable. The work was funded by the MND Association.
Six Hundred Sequences, Sixty Targets, Zero Patients
The paper presents over six hundred intrabody sequences targeting sixty cytoplasmic proteins, with binders specific to linear epitopes, conformational epitopes, post-translational modifications, and oligomeric forms. Interactions were experimentally validated for p53, alpha-synuclein, SOD1, polyglutamine, FUS/TLS, UCHL1, and GFP. The remainder are computational designs, not tested molecules, and the sequences are being made openly available to other laboratories.
The near-term value is almost certainly as research tools rather than treatments. Being able to tag a specific misfolded conformer within a living neuron and watch it in real time is something the field has long sought.
A separate group published a parallel AI pipeline in Science Advances that combines AlphaFold2, ProteinMPNN, and live-cell screening, converting 19 of 26 antibody sequences into functional intrabodies, 18 of which had failed conventional conversion. Two teams converging on similar solutions suggests the bottleneck was real and is now breaking.
Where This Actually Lands First
Calling these microscopic medicines what several outlets have done gets ahead of the evidence by a considerable margin.
There is no treatment here. There is no animal efficacy study for a neurodegenerative disease in this paper, no safety data, and no clinical trial. Everything was done in cultured human kidney cells. The single largest unsolved problem is delivery. An intrabody must be produced inside the target cell, which in practice means gene therapy, and delivering a gene into neurons across the blood-brain barrier at therapeutic scale remains unsolved.
Solubility also does not guarantee usefulness. The authors report redesigned fragments that were highly soluble and thermally stable but had lost the ability to recognize their target. When they tested alpha-synuclein binders in cells, only the monomeric form of the protein was present, so oligomer specificity could not be demonstrated there. Earlier charge-based engineering faced similar trade-offs between stability and functionality.
The realistic timeline for any therapy built on this runs a decade or more, and most candidates fail. Anyone managing Alzheimer's, Parkinson's, or motor neuron disease should keep treatment questions with their neurologist, where the answers concern approved options and open trials rather than a molecular platform.
Key Questions Answered
What is an intrabody?
An engineered antibody fragment designed to fold and function inside a cell rather than in the bloodstream, where conventional antibodies operate.
Why do ordinary antibodies fail inside cells?
The cell interior is a reducing environment that breaks the disulfide bonds holding antibody structure together, causing misfolding and aggregation.
What did the researchers find?
That net electrical charge is the strongest predictor of whether a fragment stays soluble inside a cell, with a correlation strong enough to guide AI-driven redesign.
How many did they convert?
They computationally reformatted 672 antibodies and presented over 600 sequences targeting 60 proteins. Binding was experimentally validated for seven targets.
Is this a treatment for Alzheimer's or Parkinson's?
No. There is no efficacy testing in disease models here, no safety data, and no clinical trial. Delivery into neurons remains an unsolved obstacle.
What is the nearest practical use?
Research tools. Fluorescently tagged intrabodies can image specific protein conformations inside living cells, which existing reagents cannot do.
