The research, published in Angewandte Chemie International Edition, focused on phosphofructokinase-1, or PFKP, the platelet isoform of an enzyme that controls a pivotal and rate-limiting step in glycolysis. Glycolysis is the ancient metabolic pathway through which cells break down glucose to generate energy in the form of ATP and to produce molecular building blocks needed for growth and repair. Because PFKP sits at such a critical control point, even modest changes in its behavior can ripple through the entire metabolic network of a cell, altering how efficiently glucose is consumed and how energy production is balanced against biosynthesis.

Enzymes, however, never work in isolation. As Simon Ebbinghaus, one of the lead authors of the study, explains, the molecular environment surrounding an enzyme changes constantly. Cells experience stress, reorganize their metabolism in response to nutrient availability or temperature shifts, and form temporary assemblies in which enzymes cluster together with other biomolecules. These dynamic conditions raise a fundamental question that has remained open in biochemistry: is it mainly the limited physical space around an enzyme that determines how it functions, or is the chemical nature of the molecules sharing that space equally important? The Bochum team designed a series of experiments to disentangle these two effects for the first time in a rigorous, controlled way.

To do so, the researchers studied PFKP in solutions containing so-called crowding agents, molecules added deliberately to mimic the dense interior of a cell. Crucially, they selected crowders that varied widely in size and chemical identity, ranging from small sugar molecules such as glucose to large glucose polymers such as dextran. This design allowed the team to separate the purely physical effect of molecular crowding, in which available volume is reduced, from the chemical effects that arise from specific, weak interactions between the crowder and the enzyme itself. The team then measured how enzyme activity changed with temperature, assessed the structural stability of the protein, and ran molecular dynamics simulations to examine interactions that cannot be observed directly in experiments.

The first result was consistent across all tested environments: at lower temperatures, every crowded solution reduced PFKP activity. This outcome fits the classical picture of macromolecular crowding, in which reduced volume and increased viscosity slow the diffusion of substrates and constrain the conformational movements an enzyme needs to catalyze its reaction. If the story had ended there, it would have confirmed a familiar principle. But when the researchers raised the temperature, the results diverged in a way that no simple volume-exclusion model could explain.

At higher temperatures, a striking difference emerged. Sugar-based crowding molecules, including glucose and the far larger dextran polymer, helped PFKP retain its activity across a broader temperature range than other crowders did. The fact that glucose, a small molecule, and dextran, a polymer many times its size, produced similar protective effects was decisive. If molecular size alone were responsible for the behavior of crowded enzymes, these two agents should have behaved very differently. Instead, their shared chemical identity as sugars pointed to a different mechanism altogether, one rooted in what chemists call soft interactions.

Soft interactions are weak, reversible, non-covalent contacts among the surrounding molecules, the enzyme, and its substrates. Unlike the strong bonds that hold a protein’s structure together, soft interactions flicker on and off constantly, forming a transient chemical cloud around the enzyme. The molecular simulations performed by the team showed that sugar-based crowders can influence regions near the enzyme’s active site, the precisely shaped pocket where the catalytic reaction takes place. By doing so, they affected both the flexibility of the protein and the positioning of its substrates, two factors that together determine how efficiently and how reliably the enzyme can do its job.

This balance between flexibility and positioning lies at the heart of enzyme function. An enzyme must be supple enough to undergo the conformational changes required for catalysis, yet rigid enough to hold its substrates in the correct orientation long enough for the reaction to proceed. The study suggests that sugar-rich environments tune this delicate equilibrium, making PFKP more robust at elevated temperatures precisely when many proteins begin to lose their functional shape. Importantly, the enzyme is not simply protected from unfolding in a passive sense. Its dynamics and catalytic behavior are actively shaped by its local chemical surroundings, meaning the environment participates in defining the enzyme’s working properties rather than merely preserving them.

The implications reach well beyond the test tube. If the chemical composition of the molecules surrounding an enzyme can determine how it responds to thermal stress, then the cell’s interior may be far more chemically sophisticated than the traditional picture of crowding as a purely physical traffic jam suggests. As Ebbinghaus puts it, the chemical composition of the surrounding molecules may actively influence how metabolic enzymes respond to stress. During heat shock, nutrient deprivation, or metabolic reprogramming, the mix of sugars, polyols, and other small solutes inside a cell changes, and this study raises the possibility that such shifts are not incidental byproducts but part of a regulatory strategy that helps keep energy production running under difficult conditions.

Future research will need to determine whether living cells actually exploit these local chemical environments to preserve metabolic activity when challenged, a question the Bochum team identifies as the next step. If confirmed, the findings could reshape how scientists think about metabolic regulation, offering a layer of control that operates through the physical chemistry of the cellular interior rather than through genetic switches alone. The work may also inform biotechnology, where enzymes are increasingly deployed in industrial settings at high concentrations and elevated temperatures, and where engineering the chemical environment around a catalyst could prove as important as engineering the catalyst itself. For now, the study stands as a vivid reminder that in the crowded world of the cell, an enzyme’s performance depends not only on its own structure, but on the company it keeps.

Subject of Research: How soft chemical interactions with surrounding sugar molecules modulate the temperature-dependent activity of the glycolytic enzyme phosphofructokinase-1

Article Title: Robust enzymes

Article References: Robust enzymes. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: enzymes, glycolysis, phosphofructokinase, molecular crowding, soft interactions, protein stability, metabolism, cellular stress, biophysics, sugar molecules, enzyme kinetics, molecular simulations

The team, led by Elwy Abdelkader and Gottfried Otting, used cell-free protein synthesis to produce three versions of GB1, a compact 56-residue immunoglobulin-binding domain that has long served as a workhorse of protein biophysics. In one version, all four valines carried a single fluorine in one stereochemical arrangement, designated (2S,3S)-4-fluorovaline; in a second, the fluorine was installed in the alternative stereochemistry, (2S,3R)-4-fluorovaline; and in a third, both methyl hydrogens of one valine methyl group were replaced, yielding 4,4′-difluorovaline. Mass spectrometry confirmed that the dominant product in each case was the protein in which every valine site had been substituted, although a fraction of molecules retained one canonical valine, particularly in the difluorinated sample where the bacterial valyl-tRNA synthetase struggles to recognise the heavily fluorinated amino acid.

Crucially, the substitutions barely disturbed the protein. Melting temperatures measured by circular dichroism fell only about 10 degrees Celsius, from the wild-type value to a range of 67 to 70 degrees, and the characteristic proton chemical shifts of the native fold were conserved. That structural conservation proved to be more than a reassurance: because the relative ordering of proton chemical shifts was preserved between wild-type and fluorinated samples, the researchers could transfer resonance assignments from conventional two-dimensional NMR spectra to the fluorine dimension, opening a convenient route to assign the fluorine-19 signals without resorting to collections of point mutants. Short-delay proton-fluorine correlation experiments and fluorine-detected TOCSY spectra then linked each fluorine resonance to its parent residue, including stereospecific assignments distinguishing the two fluorines of each difluorovaline side chain.

The central analytical trick relies on two complementary NMR observables that report on the rotation of the CH2F group about its carbon-carbon bond. The first is the three-bond coupling between the fluorine nucleus and the beta proton of the same residue. According to the Karplus relationship, this coupling is large, roughly 30 to 44 hertz, when the fluorine sits trans to the beta proton, and small, around 8 to 9 hertz, in the gauche geometries. A coupling of 38 hertz measured for the fluorine at position 54 therefore unambiguously identified the trans rotamer for that group. The second observable is the so-called gamma effect on carbon-13 chemical shifts: a fluorine atom three bonds away from a carbon-13 nucleus shifts that resonance upfield by several parts per million, and the size of the shift depends on the dihedral angle between the two nuclei. Because the remaining methyl group of each fluorovaline provides a built-in carbon-13 reporter, the gamma effect distinguishes the two gauche rotamers that the proton-fluorine coupling alone cannot separate.

Density functional theory calculations underpinned the interpretation. Computations on the fluorovaline residues of a related protein, for which a 1.3 angstrom crystal structure exists, showed that the gamma effect on the methyl carbon reaches about minus 6.3 parts per million when the fluorine and the methyl carbon are nearly anti-periplanar, but only about minus 5.1 parts per million in the gauche geometries. Calculations on the model compound 1-fluoro-2-methylpropane, sweeping the torsion angle systematically, confirmed the same dihedral dependence whether or not the remaining molecular coordinates were relaxed. Experiment matched theory: the measured gamma effects ranged from about 5.8 to 7.4 parts per million, and they correlated with three-bond fluorine-carbon coupling constants measured by a constant-time HSQC difference experiment, exactly as expected if rotamer populations govern both quantities.

The analysis showed that CH2F groups almost never lock into a single rotamer, even when buried in the hydrophobic core. Only the most deeply buried valine, residue 54, showed a strong preference for one staggered conformation. Solvent-exposed groups sampled multiple rotamers, yet preferential populations could still be identified from the combination of coupling constants and chemical-shift effects. The energy barrier separating the three staggered rotamers of a CH2F group is roughly 4 to 5 kilocalories per mole, high enough to make eclipsed conformations unfavourable but low enough to allow interconversion at room temperature, so the observed spectra reflect a population-weighted average rather than a frozen geometry.

One of the most striking findings concerns where the fluorine atoms end up. In several residues, the preferred rotamer positions the fluorine atom as close as 2.85 angstroms to the carbonyl carbon of the same residue’s backbone, a distance that nominal van der Waals radii would suggest is sterically forbidden. The authors note that standard van der Waals radii may not apply to sp2 carbons bearing electron-withdrawing substituents, and they suggest that a favourable electrostatic interaction between the negatively polarised fluorine and the positively polarised carbonyl carbon may stabilise these geometries. Similar arrangements had been spotted before in a high-resolution crystal structure of a different fluorinated protein, but the new solution-state data show that the effect is systematic rather than a crystallographic artefact. Notably, the interaction is not strong enough to trap the CH2F group in a single rotamer, and it did not produce unusually large changes in the carbonyl carbon-13 chemical shifts.

The study also documented transient contacts between fluorine atoms on neighbouring residues. In GB1, the gamma carbons of valines 39 and 54 lie within about 4 angstroms of each other, and through-space scalar fluorine-fluorine couplings of roughly 1.5 hertz in the monofluorinated sample and 2.7 hertz in the difluorinated sample were detected in fluorine-fluorine TOCSY spectra. These couplings arise from short-range orbital overlap and therefore report direct, if fleeting, fluorine-fluorine contacts. Intriguingly, the difluorinated protein switched which pair of fluorines made contact, from the two gamma-1 fluorines to the gamma-1 fluorine of residue 39 paired with the gamma-2 fluorine of residue 54, suggesting that the extra fluorination nudges the local structure toward the conformation seen in crystal structures of the wild-type protein. The polarity of the carbon-fluorine bonds appears to discourage sustained fluorine-fluorine proximity, which also explains why the two CH2F groups within a single difluorovaline avoid conformations that would bring their fluorines together.

Beyond its structural insights, the work points toward practical applications. Fluorinated amino acids are now commercially available, and proteins labelled with them yield fluorine-19 spectra of exceptional width and resolution while remaining only modestly destabilised. Because the gamma effect translates rotamer populations into measurable carbon-13 shifts, a fluorinated isopropyl group could serve as a sensor of ligand binding that can be read out even on spectrometers not equipped for fluorine detection. The conservation of relative proton chemical shifts between wild-type and fluorinated proteins likewise offers a general shortcut for assigning fluorine spectra in larger systems, where broad lines and chemical-shift anisotropy usually force researchers into laborious mutagenesis. As the authors conclude, the ease with which proteins accommodate fluorine atoms, and the sensitivity of CH2F groups to their chemical environment, make these labels excellent non-invasive probes of protein structure, dynamics and molecular recognition.

Subject of Research: Rotamer preferences of fluorinated valine side chains in the protein GB1 determined by fluorine-19 NMR spectroscopy

Article Title: γ effects identify preferentially populated rotamers of CH2F groups: side-chain conformations of fluorinated valine analogues in a protein

Article References: γ effects identify preferentially populated rotamers of CH2F groups: side-chain conformations of fluorinated valine analogues in a protein. (n.d.). https://doi.org/10.5194/mr-6-257-2025

Image Credits: AI Generated

DOI: 10.5194/mr-6-257-2025

Keywords: fluorine-19 NMR, fluorinated amino acids, GB1 protein, rotamers, CH2F groups, gamma effect, Karplus relationship, DFT calculations, cell-free protein synthesis, through-space coupling, side-chain conformations, protein stability

The scale of the problem the researchers confronted is enormous. The Deafness Variation Database, or DVD, a public resource that catalogs genetic variants associated with hearing loss, contains 381,924 missense variants spread across 224 genes. Missense variants are single-letter changes in DNA that swap one amino acid for another in the resulting protein, and they are notoriously difficult to interpret. Of those 381,924 variants, 303,577—nearly four out of five—are classified as variants of uncertain significance, or VUS. A VUS is essentially a genetic placeholder: laboratory testing has detected a change, but no one can say with confidence whether it disrupts hearing or is harmless. For families, that uncertainty translates into unresolved diagnoses, ambiguous reproductive counseling, and the psychological weight of an answer that is not really an answer.

Professional guidelines from the American College of Medical Genetics and Genomics, known as ACMG/AMP, allow computational evidence to contribute to variant classification, but they leave open exactly how different lines of computational evidence should be weighed and combined. The Iowa team approached this as a formal statistical problem. They built a family of Bayesian models—probabilistic frameworks that start with a prior belief about how likely a variant is to be pathogenic and then update that belief as evidence accumulates. Each type of computational evidence, from an algorithm score to a gene-level property, contributes a quantified strength of evidence expressed in the likelihood ratios that ACMG/AMP recommendations envision. The advantage of this approach is transparency: rather than a black-box verdict, every classification can be traced to explicit, auditable numerical inputs.

The first question the researchers addressed was which of the existing variant-scoring tools deserves the most weight. They compared two widely used predictors, CADD and REVEL, using Bayesian models trained on two different reference datasets: a 2019 snapshot of ClinVar, a public archive of variant interpretations, and a set of variants already labeled within the Deafness Variation Database itself. When accuracy, sensitivity, and specificity were measured head to head, the REVEL model built on the DVD-labeled variants came out on top. That result carries a practical lesson for the field: predictors calibrated on disease-specific, curated data can outperform general-purpose scores, and evidence frameworks for hearing loss should draw on resources tailored to the genes and phenotypes actually at stake.

The team then added a second layer of evidence drawn from evolutionary biology. Genes differ in how tolerant they are to missense change: some proteins can absorb many amino-acid substitutions without losing function, while others are so finely tuned that almost any change is deleterious. Sorting the 224 deafness genes into three bins—tolerant, average, and intolerant—revealed a clear gradient in baseline risk. Variants in intolerant genes carried a prior probability of pathogenicity of 25.7 percent, more than double the 10.7 percent seen in average genes and nearly triple the 8.7 percent in tolerant genes. In other words, the same DNA letter change means something quite different depending on the evolutionary constraint of the gene it lands in, and the Bayesian framework now accounts for that context explicitly.

The most consequential innovation, however, was biophysical. Many missense variants do their damage not by sabotaging a protein’s active site directly but by destabilizing the protein’s three-dimensional fold. A protein that cannot hold its shape is typically degraded by cellular quality-control machinery or accumulates in misfolded, dysfunctional forms, and either outcome can starve the delicate hair cells of the inner ear of the proteins they need to survive and transmit sound. The researchers incorporated predicted changes in folding stability into their Bayesian model, estimating how many kilocalories per mole of destabilization each variant would impose. This protein-folding-informed model outperformed the simpler versions, and crucially it did something the score-based models cannot: it supplied a mechanistic rationale, explaining not just that a variant is likely harmful but how it would break the protein.

The numbers that emerged from this analysis are remarkable. Of the 381,924 missense variants in the database, 54,752—14.3 percent of the total—are predicted to cause modest destabilization of more than 1.0 kilocalorie per mole. A stricter threshold of at least 2.0 kilocalories per mole, indicating more severe structural disruption, captures 22,237 variants, or 6.2 percent of the database. When the analysis is restricted to the prioritized variants of uncertain significance, the fractions climb steeply: 34 percent of prioritized VUSs fall into the modest-destabilization category, and 55.8 percent fall into the severe category. Taken together, the findings suggest that protein misfolding is not an occasional curiosity among deafness variants but a prevalent mechanism, plausibly underlying a large share of hereditary hearing loss that has resisted explanation.

The framework’s power is not merely statistical. Using the protein-folding-informed model, 28,866 variants of uncertain significance reached a posterior probability of pathogenicity above 98 percent and were prioritized as likely pathogenic. From that prioritized set, the team identified twelve probands—individual patients in the database whose genetic diagnoses could be upgraded from uncertain to likely pathogenic or pathogenic. For those twelve families, computational biophysics converted an ambiguous test result into an actionable diagnosis, with implications for prognosis, surveillance, and reproductive counseling. The study also highlights two specific variants that cause clear structural disruption of their proteins, illustrating concretely how biophysical characterization can settle cases that sequence data alone leave open.

Why does the inner ear appear so vulnerable to this particular failure mode? The sensory hair cells that convert mechanical vibration into neural signals are packed with precisely assembled protein complexes, from ion channels to the tip-link proteins of the hair bundle, and many of them are long-lived and rarely replaced. A modest thermodynamic nudge toward unfolding, tolerable in a rapidly turning-over tissue, can be catastrophic in a cell that must last a lifetime. The new findings give that intuition quantitative teeth, showing that destabilization thresholds of one to two kilocalories per mole—small on the scale of protein energetics—correspond to a meaningful probability of disease.

The broader implications reach beyond deafness. Variant interpretation is the bottleneck of modern genomic medicine, and the Iowa study demonstrates a template for breaking it: combine disease-specific curated databases, gene-level evolutionary constraint, and protein-folding physics inside a transparent Bayesian framework aligned with professional guidelines. The same architecture could, in principle, be adapted to cardiomyopathies, epileptic encephalopathies, or any inherited disorder dominated by missense variation. For the hundreds of thousands of families still holding variants of uncertain significance in deafness genes, the message is one of cautious optimism: a large fraction of those ambiguities may now be resolvable, not by waiting for more families to be reported, but by calculating what the variant does to the protein it touches. Twelve diagnoses have already been changed; the framework suggests many more are waiting.

Subject of Research: Computational assessment of protein misfolding as a mechanism of hereditary deafness

Article Title: The prevalence of protein misfolding as a mechanism for hereditary deafness

Article References: Gogal, R. A., Cox, G. M., Kolbe, D. L., Odell, A. M., Ovel, C. E., McCormick, K. I., Hong, B., Azaiez, H., Casavant, T. L., Smith, R. J. H., Braun, T. A., & Schnieders, M. J. (2026). The prevalence of protein misfolding as a mechanism for hereditary deafness. PLOS Genetics, 22(9), e1012085. https://doi.org/10.1371/journal.pgen.1012085

Image Credits: AI Generated

DOI: 10.1371/journal.pgen.1012085

Keywords: hereditary deafness, protein misfolding, variant of uncertain significance, Deafness Variation Database, Bayesian model, missense variants, protein stability, ACMG/AMP guidelines, genetic diagnosis, PLOS Genetics, hair cells, variant interpretation

The central obstacle to studying proteins inside mitochondria has always been delivery combined with signal quality. In-cell NMR, a technique pioneered nearly two decades ago, allows researchers to observe isotopically labeled proteins inside living cells, but applying the same logic to a subcellular organelle is far harder. Mitochondria are small, bounded by a double membrane, and easily damaged during isolation. Earlier work, including a 2014 study by Barbieri, Luchinat and Banci, demonstrated that in-mitochondria NMR was conceptually possible, but the approach has remained technically demanding and limited in resolution. The new study, led by Conggang Li with co-first authors Zeting Zhang and Cai Zhang, builds on that foundation and pushes the method to the point where quantitative, atomic-resolution measurements become routine enough to answer biological questions.

The key innovation lies in sample preparation. The team delivers purified proteins directly into isolated mitochondria using electroporation, a technique in which brief electrical pulses create transient pores in membranes, allowing macromolecules to enter. Crucially, the researchers performed extensive control experiments to verify that the delivered proteins genuinely resided inside the organelles rather than clinging to the outer membrane or leaking into the surrounding buffer. By comparing NMR spectra of the mitochondrial samples with spectra of the supernatant after centrifugation, they confirmed that the signals they recorded arose from proteins within intact mitochondria. They also quantified the intramitochondrial concentration of each delivered protein by lysing the organelles after measurement and comparing integrated peak areas against an internal standard, a step that transforms the method from a qualitative observation into a genuinely quantitative tool.

Two complementary labeling strategies underpin the spectroscopy. For residue-resolved information, the team used uniformly nitrogen-15-enriched proteins and recorded two-dimensional proton-nitrogen correlation spectra, the workhorse experiment of solution NMR. Each cross-peak in such a spectrum corresponds to a specific atom pair in a specific amino acid residue, so changes in peak position, intensity or width report directly on what is happening to that part of the protein. In parallel, the researchers exploited fluorine-19 labeling, incorporating fluorinated amino acid analogues such as 3-fluorotyrosine and 5-fluorotryptophan into their target proteins. Fluorine-19 offers a background-free window into biological samples, because essentially no endogenous mitochondrial molecules contain fluorine, and modern fluorine probes, including the highly sensitive wPSP6F tag developed in the same laboratory, boost sensitivity further. Together, the two approaches allow the same molecular event to be viewed from complementary angles.

The biological centerpiece of the study is alpha-synuclein, the intrinsically disordered protein whose aggregation into Lewy bodies is a hallmark of Parkinson’s disease. Alpha-synuclein has long been suspected of interacting with mitochondrial membranes, particularly cardiolipin-rich regions of the inner membrane, and this interaction is thought to contribute to the mitochondrial dysfunction seen in the disease. Using their in-mitochondria NMR platform, the researchers observed how alpha-synuclein engages the mitochondrial membrane environment, with residue-resolved signal attenuation revealing which parts of the protein chain make contact with membranes. The spectra of nitrogen-15-enriched alpha-synuclein delivered into mitochondria isolated from rat liver and pig brain showed characteristic broadening and attenuation patterns that map the membrane-binding interface at single-residue resolution, something previously achievable only with simplified model membranes in a test tube.

Perhaps the most striking aspect of the work is its treatment of post-translational modifications, the chemical tags that cells attach to proteins to regulate their behavior. The team examined how modifications of alpha-synuclein, including phosphorylation and methionine oxidation, alter its interaction with the mitochondrial membrane. Oxidation is of particular interest because mitochondrial metabolism generates reactive oxygen species, and oxidized alpha-synuclein has been implicated in impaired clearance and altered chaperone binding. The researchers also probed how oxidized alpha-synuclein interacts with the mitochondrial chaperone TRAP1 and with protein disulfide isomerase, using methionine-selective nitrogen-15 enrichment to monitor the oxidation state directly. These experiments demonstrate that the platform can capture regulatory chemistry as it happens inside the organelle, not merely static structure.

To establish the generality of the method, the team turned to a well-characterized benchmark: the B1 domain of protein G, known as GB1, a small, stably folded globular protein that has served as a standard in NMR methodology development for years. By measuring GB1 inside mitochondria, the researchers could assess how the mitochondrial interior, with its crowded macromolecular environment, distinctive pH and high ionic strength, affects protein stability. They also engineered a GB1 variant and monitored its dimerization inside the organelle using fluorine-19 NMR, achieving quantitative insight into how the mitochondrial environment modulates protein-protein interactions. This matters because a growing body of evidence shows that the intracellular milieu, with its excluded-volume effects and specific weak interactions, can shift protein folding equilibria and binding affinities in ways that dilute buffer experiments cannot reproduce.

The technical rigor behind these measurements deserves emphasis. The team measured the internal pH of their mitochondrial preparations using the fluorescent dye BCECF and cross-checked it against NMR chemical shifts of alpha-synuclein, ensuring that spectral changes reflected genuine intramitochondrial conditions rather than artifacts of sample preparation. Deuterated GB1 was used to probe hydrogen exchange and protein dynamics inside the organelle, with time-resolved spectra showing how signals recover as amide protons exchange with solvent. Leakage controls, concentration quantification and comparisons between buffer and mitochondrial samples were performed for every protein studied, establishing a level of validation that will be essential if the method is to be adopted by other laboratories.

The implications extend well beyond method development. Mitochondrial dysfunction sits at the heart of Parkinson’s disease, Alzheimer’s disease, cardiomyopathy, diabetes and aging itself, yet most structural biology on mitochondrial proteins has been performed on purified components or membrane mimetics. A platform that reports on protein structure, stability and interactions inside intact mitochondria bridges the gap between reductionist test-tube biophysics and the messy reality of the organelle. It also complements other emerging approaches, such as dynamic nuclear polarization-supported solid-state NMR, which has recently been used to study nucleic acids and proteins in situ with subcellular specificity, and real-time NMR observations of endogenous cytochrome c translocation within mitochondria reported in 2024. Together these techniques sketch a future in which the atomic-resolution study of biology inside organelles becomes standard practice rather than a tour de force.

For the immediate future, the Wuhan team’s framework offers a template for exploring mitochondria-related biomolecular events that were previously out of reach: how disease-linked mutations change membrane binding, how chaperones protect proteins in the crowded mitochondrial interior, and how the organelle’s unique chemical environment reshapes the energy landscape of folding and aggregation. As the authors note, the approach paves the way toward a more comprehensive understanding of the molecular events governing mitochondrial health and dysfunction. If the method proves as adaptable as in-cell NMR in whole cells has been, the coming years could see atomic-resolution spectroscopy become a routine probe of the organelle whose failure underlies so much of human disease, finally letting researchers watch, atom by atom, what goes wrong inside the powerhouse.

Subject of Research: High-resolution NMR spectroscopy of proteins delivered into intact mitochondria

Article Title: High-resolution NMR spectroscopy of proteins in intact mitochondria

Article References: Zhang, Z., Zhang, C., Xu, G., Du, R., Yu, J., Liu, X., Chai, Z., Wu, Q., Jiang, L., Liu, M., & Li, C. (2026). High-resolution NMR spectroscopy of proteins in intact mitochondria. Nature Chemical Biology. https://doi.org/10.1038/s41589-026-02314-x

Image Credits: AI Generated

DOI: 10.1038/s41589-026-02314-x

Keywords: NMR spectroscopy, mitochondria, in-cell NMR, alpha-synuclein, Parkinson's disease, electroporation, fluorine-19 labeling, protein stability, post-translational modifications, mitochondrial membrane, structural biology, protein dimerization

To understand why the discovery matters, it helps to start with the biology of protein turnover. Inside every cell, proteins that are damaged, misfolded or simply no longer needed are tagged with small molecules called ubiquitin, a molecular label that condemns them to destruction by the proteasome, the cell’s waste-disposal complex. Deubiquitinating enzymes, or DUBs, do the opposite: they snip ubiquitin tags off proteins, rescuing them from degradation and extending their functional lifespan. When this balancing act goes wrong, proteins that should be cleared can accumulate to abnormal levels, and in cancer that often means proteins that drive uncontrolled growth are granted an unnatural stay of execution. USP12 belongs to the ubiquitin-specific protease family, and previous work has linked it to several cancers, but its role in ovarian cancer had remained poorly defined.

The identity of its partner in this study was equally obscure. TBC1D22A is a member of the TBC1 domain family, proteins that act as GTPase-activating proteins for Rab GTPases, the molecular switches that regulate vesicle trafficking, membrane trafficking and intracellular transport. While Rab proteins have increasingly been implicated in tumor progression, TBC1D22A’s specific oncogenic mechanisms in ovarian cancer, and crucially the upstream pathways that regulate it, had never been reported. The new study set out to fill that gap by asking a deceptively simple question: what keeps TBC1D22A stable inside ovarian cancer cells, and what happens when that stability is disrupted?

The answer emerged from an unbiased hunt for TBC1D22A’s molecular companions. Using liquid chromatography coupled with tandem mass spectrometry, or LC-MS/MS, the researchers screened for proteins that physically associate with TBC1D22A, and USP12 surfaced as a candidate interacting deubiquitinase. The team then confirmed the partnership through complementary techniques: co-immunoprecipitation experiments showed that the two proteins can be pulled down together from cell lysates, while immunofluorescence co-localization assays revealed that they occupy overlapping territories within the cell. Together, these approaches established that the association is not a fleeting accident of cellular crowding but a genuine physical interaction, the kind that can support a regulatory relationship.

With the interaction confirmed, the researchers moved on to the central mechanistic question: does USP12 actually control how much TBC1D22A exists inside a cell, and if so, how? The evidence points firmly to post-translational regulation. Cycloheximide chase experiments, which block new protein synthesis and allow researchers to watch existing proteins decay over time, showed that TBC1D22A degraded more slowly when USP12 was present. Proteasome inhibition assays demonstrated that TBC1D22A’s destruction proceeds through the ubiquitin-proteasome pathway. Most tellingly, in vivo ubiquitination assays revealed that USP12 directly reduces the ubiquitin load carried by TBC1D22A, effectively stripping off the degradation tags before the proteasome can act. Importantly, USP12 did not alter TBC1D22A transcription, meaning the enzyme works purely at the protein level, preserving existing TBC1D22A molecules rather than coaxing the gene to make more.

That stabilization has consequences. Through a series of gain-of-function and loss-of-function experiments, the team showed that USP12 promotes ovarian cancer cell proliferation, migration and invasion, and that these effects depend on TBC1D22A. When the researchers removed USP12, TBC1D22A levels fell and the cancer cells became less motile and less invasive; when they re-introduced TBC1D22A in rescue experiments, the aggressive behavior returned. This dependency is the crucial detail: it means USP12 is not simply a general growth promoter acting through many parallel routes, but a specific upstream regulator whose oncogenic influence flows through TBC1D22A. Depleting the downstream partner effectively neutralizes the upstream enzyme, which is exactly the kind of linear dependency that makes a signaling axis attractive as a drug target.

The signaling pathway connecting the pair to malignancy is one of the most famous in all of cancer biology. The PI3K-AKT-mTOR cascade integrates signals from growth factor receptors and relays them to the cell’s growth and survival machinery, promoting metabolism, protein synthesis, proliferation and resistance to cell death. It is hyperactivated in a large fraction of human tumors, including ovarian cancers, and has been the subject of intense pharmaceutical interest for decades. In this study, the USP12-TBC1D22A axis was associated with activation of this pathway, providing a plausible mechanistic bridge between a vesicle-trafficking regulator and the malignant phenotype. The researchers also validated their findings in vivo using subcutaneous xenograft models, in which manipulated ovarian cancer cells were grown as tumors in animals, confirming that the axis influences tumor growth outside the controlled environment of a culture dish.

The clinical data add a layer of real-world relevance. Transcriptomic analysis, tissue microarrays and data from The Cancer Genome Atlas cohorts showed that USP12 is upregulated in ovarian cancer tissues compared with normal tissue, and that its expression positively correlates with TBC1D22A levels, consistent with the stabilizing relationship uncovered in the lab. More ominously, high USP12 expression predicted poorer overall survival and poorer progression-free survival in patients. In other words, the same molecular relationship that drives aggressiveness in laboratory models also tracks with worse outcomes in the clinic, strengthening the case that the axis is not an artifact of experimental conditions but a genuine feature of the disease.

For patients, the implications are cautiously encouraging. Ovarian cancer is often treated with surgery and platinum-based chemotherapy, and while PARP inhibitors have improved outlooks for some patients with homologous recombination deficiency, many tumors eventually become resistant. A deubiquitinase such as USP12 represents a different class of target, and DUB inhibitors have become an active frontier in drug development precisely because these enzymes can be drugged in ways that stabilize-or destabilize-key disease proteins. If the USP12-TBC1D22A axis proves essential in a broader set of ovarian tumors, blocking USP12 could deprive cancer cells of their stabilized TBC1D22A supply and dampen PI3K-AKT-mTOR signaling, potentially complementing existing therapies. The correlation between USP12 and survival also suggests it could serve as a prognostic marker, helping clinicians identify patients whose disease is likely to progress more rapidly.

There are, of course, important caveats. The study relied on cell lines, xenograft models and retrospective analysis of patient cohorts; prospective clinical validation, and demonstration that pharmacologically inhibiting USP12 benefits patients, remain future work. DUBs often have multiple substrates, so any therapeutic strategy would need to account for effects beyond TBC1D22A. Still, the work fills a genuine gap by identifying the first reported upstream regulatory pathway for TBC1D22A in ovarian cancer, and it does so with a methodologically thorough chain of evidence, from unbiased mass spectrometry through mechanistic biochemistry to animal models and human tissue data. As the authors conclude, the USP12-TBC1D22A axis may serve as a promising therapeutic target for ovarian cancer, and for a disease where new molecular handles are scarce, that promise is worth taking seriously.

Subject of Research: USP12-mediated deubiquitination and stabilization of TBC1D22A in ovarian cancer progression

Article Title: USP12 stabilizes TBC1D22A via deubiquitination to activate PI3K-AKT-mTOR signaling and promote ovarian cancer progression

Article References: Lv, X., Gong, R., Cai, X., & Guo, R. (2026). USP12 stabilizes TBC1D22A via deubiquitination to activate PI3K-AKT-mTOR signaling and promote ovarian cancer progression. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-09027-8

Image Credits: AI Generated

DOI: 10.1186/s12967-026-09027-8

Keywords: USP12, TBC1D22A, ovarian cancer, deubiquitinase, ubiquitin-proteasome pathway, PI3K-AKT-mTOR signaling, protein stability, Rab GTPase-activating protein, prognostic biomarker, xenograft model, Journal of Translational Medicine, cancer signaling

The research team, led by Yijie Liu and Jie Teng, worked with fresh leaves of the ‘Gougunao Tea No. 2’ cultivar harvested in Jiangxi Province, China. Their target was a purification strategy called three-phase partitioning, or TPP, a chromatography-free separation technique in which t-butanol and water, normally miscible, are forced apart by adding ammonium sulfate. Under the right conditions, proteins gather into a distinct middle phase between the upper organic layer, which collects oils and lipid-soluble material, and the lower aqueous layer, which retains polysaccharides and other water-soluble compounds. TPP is prized for its simplicity, high recovery rates and easy scalability, and it has already been applied to polyphenol oxidase from several sources. The problem is the solvent: t-butanol is flammable, volatile and ecotoxic, and highly polar organic solvents can disrupt the hydrophobic interactions that hold membrane proteins in their native shape, accelerating the loss of enzyme activity during purification.

To replace t-butanol, the researchers turned to deep eutectic solvents, a class of liquids first described by Abbott and colleagues in 2004. Deep eutectic solvents form when a hydrogen bond acceptor and a hydrogen bond donor self-assemble through an extensive hydrogen bonding network, producing a liquid with a melting point far below that of either component. They are simple to prepare, biodegradable, low in toxicity and highly biocompatible, and their physicochemical properties can be tuned by choosing different component pairs. Previous work has shown that these solvents can stabilize proteins in vitro through hydrogen bonding and hydrophobic forces, which made them attractive candidates for a green version of TPP. However, no one had applied a deep eutectic solvent-based TPP system to a membrane-bound polyphenol oxidase, and the molecular mechanisms behind any such purification remained unexplored.

The team prepared nine different deep eutectic solvents from combinations of hydrogen bond acceptors and donors, including pairs such as tetracaine with thymol, menthol with octanoic acid, and procainamide derivatives with fatty acids. Fourier transform infrared spectroscopy confirmed that each mixture had formed a genuine eutectic: carbonyl stretching peaks shifted to lower frequencies, and the broad hydroxyl absorption bands of the individual components merged into single strong peaks, signatures of newly created intermolecular hydrogen bonds. Proton nuclear magnetic resonance spectroscopy provided further confirmation. The phenolic hydroxyl proton of thymol in the first solvent shifted downfield with clear peak broadening, while the carboxyl protons of several acid-based mixtures vanished entirely from the 8 to 15 ppm region, a consequence of rapid proton exchange within a dense hydrogen bonding network. One solvent, built from menthol and octanoic acid in a one-to-one molar ratio and designated DES-5, showed the largest downfield shift of all, indicating the strongest hydrogen bonding in the entire set.

When the nine solvents were tested against crude tea mPPO extract, DES-5 stood out decisively. It delivered an enzyme activity of 2117 units and a purification fold of 17.43, with a specific activity of 50,652 units per milligram, significantly outperforming t-butanol and every other solvent tested. The authors attribute this to a combination of favorable mass transfer properties and a useful side effect: the solvent appears to denature some contaminating proteins, which raises the purification fold of the target enzyme. Single-factor experiments then mapped how the process responded to changing conditions. Enzyme activity and purification fold both peaked at 45 percent ammonium sulfate, a concentration high enough to salt out proteins through hydrophobic aggregation but low enough to avoid irreversible conformational damage. The optimal solvent-to-extract ratio was 2:1 by volume, beyond which excess solvent began to denature the enzyme itself. Extraction time peaked at 75 minutes, after which prolonged exposure to air and light caused oxidative losses, and the ideal temperature was 30 degrees Celsius, where the lower viscosity of the warm solvent improved mass transfer before thermal inactivation set in.

Response surface methodology refined these parameters into a precise recipe. A Box-Behnken design produced quadratic regression models for both enzyme activity and purification fold, with coefficients of determination of 0.9966 and 0.9992 respectively, indicating excellent fit and predictive reliability. The statistically optimal conditions were 47.2 percent ammonium sulfate, 74.8 minutes of extraction and a solvent ratio of 1.98:1. Rounded to practical values of 47 percent, 75 minutes and 2:1, three replicate validations yielded an average activity of 1967 units and a purification fold of 16.64, with prediction errors below one percent and no statistically significant difference from the theoretical values. The optimized process was thus both accurate and reproducible.

Physical characterization reinforced the case for the green solvent. Scanning electron microscopy revealed that crude mPPO formed irregular porous aggregates with rough edges and abundant protein debris. After t-butanol-based purification, the particles became dense, fragmented shards, a morphology the authors link to organic-solvent-induced aggregation. mPPO purified through DES-5, by contrast, emerged as relatively intact block-shaped particles with denser, smoother surfaces and minimal fragmentation, suggesting the deep eutectic solvent provides a gentler separation environment that preserves the morphological integrity of the protein particles.

To understand the mechanism, the researchers built a homology model of tea mPPO from its UniProt sequence, validated it with Ramachandran and ERRAT analyses, and docked the two DES-5 components into the structure. Both menthol and octanoic acid bound with energies more favorable than the minus 5 kilocalories per mole threshold for spontaneous binding. Menthol’s single hydroxyl group acted as both hydrogen bond donor and acceptor, bridging three backbone residues, Gly301, Lys382 and Gly383, with bond lengths between 2.2 and 2.6 angstroms. Octanoic acid anchored its polar head group through hydrogen bonds to Leu531, Thr331 and Asn200, the shortest at just 1.9 angstroms, while surrounding hydrophobic residues constrained its fatty acyl chain and improved shape complementarity within the binding pocket.

Molecular dynamics simulations in GROMACS then tested whether these interactions stabilize or destabilize the enzyme. In three independent 100-nanosecond simulations per system, the protein backbone of both ligand-bound complexes relaxed and stabilized at root-mean-square deviations of roughly 0.20 to 0.26 nanometers, whereas an additive-free control lacking both ligands never reached a plateau and drifted to 0.30 to 0.35 nanometers, clear evidence that menthol and octanoic acid dampen backbone fluctuations. The dicopper catalytic center, restrained to preserve its native type 3 copper geometry, kept its bridging angle centered at 120 to 130 degrees throughout, confirming that ligand binding does not disturb the active site. The radius of gyration held steady at about 2.22 nanometers in both systems, showing no global unfolding, and free energy landscape analysis revealed funnel-shaped topologies converging on compact, low-energy states. In the menthol system the solvent-accessible surface area rose modestly as surface loops rearranged outward, while in the octanoic acid system it decreased as the ligand shielded surface residues, producing a more compact surface conformation.

Taken together, the experiments and simulations tell a coherent story: a menthol and octanoic acid deep eutectic solvent purifies tea membrane-bound polyphenol oxidase more efficiently than t-butanol precisely because its components bind the enzyme gently, through hydrogen bonds and hydrophobic contacts that stabilize rather than disrupt the protein’s native fold. The optimized protocol, 47 percent ammonium sulfate, 75 minutes, a 2:1 solvent ratio and 30 degrees Celsius, achieves a purification fold above 16 in a single chromatography-free step using a biodegradable solvent. Beyond tea, the authors suggest the approach offers a template for the green purification of other membrane-bound enzymes, a class of catalysts central to cellular processes and industrial biocatalysis but long hindered by purification methods that sacrifice the very activity they seek to recover.

Subject of Research: Green purification of tea membrane-bound polyphenol oxidase using deep eutectic solvent-based three-phase partitioning

Article Title: Molecular dynamics provide insights into the purification of tea membrane-bound polyphenol oxidase through three-phase partitioning utilizing an eco-friendly deep eutectic solvent

Article References: Liu, Y., Hou, S., Xiao, S., Wang, Y., Liu, Y., & Teng, J. (2026). Molecular dynamics provide insights into the purification of tea membrane-bound polyphenol oxidase through three-phase partitioning utilizing an eco-friendly deep eutectic solvent. Food Chemistry: X, 39, Article 104529. https://doi.org/10.1016/j.fochx.2026.104529

Image Credits: AI Generated

DOI: 10.1016/j.fochx.2026.104529

Keywords: polyphenol oxidase, deep eutectic solvent, three-phase partitioning, tea, molecular dynamics, enzyme purification, green chemistry, membrane proteins, molecular docking, food chemistry, biocatalysis, protein stability

Developed by Polina Polunina and Wolfgang Maier of the University of Freiburg together with Alan Rubin of the Walter and Eliza Hall Institute of Medical Research, VEFill is built on LightGBM, a gradient boosting framework well suited to structured, high-dimensional tabular data. Rather than relying on a single source of information, the model integrates a rich set of biologically informed features: evolutionary conservation scores from the EVE generative model, amino acid substitution matrices such as BLOSUM62 and PAM250, physicochemical descriptors of the wild-type and variant residues, and contextual sequence embeddings from ESM-1v, a transformer-based protein language model trained on millions of natural sequences. The authors trained and optimized the model using Bayesian hyperparameter search and stored all underlying data in a structured PostgreSQL database to ensure reproducibility.

The training ground for VEFill was the Human Domainome 1 dataset, a standardized collection of stability measurements generated with an abundance-based protein fragment complementation assay across 522 human protein domains, using site-saturation mutagenesis to introduce every possible amino acid substitution. From this resource, the team assembled a feature-complete subset of 140 domains comprising 136,854 mutations for the full model, and used the broader set of 521 domains, totaling 562,208 mutations, to train reduced-feature versions that depend only on ESM-1v embeddings and positional mean scores.

The performance figures are striking. In a leave-protein-out evaluation, where the model must predict variant effects for protein domains entirely absent from training, the best configuration achieved a coefficient of determination of 0.64 and a Pearson correlation of 0.80. Feature ablation experiments revealed a clear hierarchy of information: models relying solely on evolutionary scores, substitution matrices, or biochemical descriptors performed poorly, with R-squared values below 0.15. Adding ESM-1v embeddings lifted performance substantially, and combining the embeddings with the mean experimental score per position proved even more powerful, indicating that learned sequence representations and data-driven positional priors capture complementary aspects of mutational tolerance.

Perhaps the most practically important finding concerns data scarcity. In controlled subsampling experiments on 28 high-quality datasets, VEFill consistently outperformed a battery of existing imputation methods, including Envision, FactorizeDMS, AALasso, and nearest-neighbor approaches, once at least 20 percent of variants had been experimentally measured. Below that threshold, predictions became markedly harder, and the authors are candid that true zero-shot prediction without any positional context remains challenging, particularly for functionally complex proteins. This matters because a survey of 841 public score sets from the MaveDB repository showed that fully complete datasets are vanishingly rare, with only 44 out of 841 achieving complete coverage of all possible substitutions. The vast majority of real-world experiments therefore fall squarely within the regime where VEFill delivers its greatest benefit.

The team also probed how few measurements are actually needed per position. When per-protein models were trained with a restricted number of substitutions per site, accuracy rose rapidly and began to plateau at roughly four mutations per position. Intriguingly, models trained exclusively on five carefully chosen amino acids, histidine, glutamic acid, asparagine, isoleucine, and glycine, each representing a distinct physicochemical class, matched the performance of models trained on randomly selected substitutions at equivalent coverage. This suggests that sparse, information-efficient mutational libraries could be designed deliberately, prioritizing a small representative set of substitutions per site without sacrificing predictive power.

Noise ceiling analyses added an important dose of realism. Because experimental variability imposes a fundamental upper bound on achievable accuracy, the researchers simulated replicate experiments using reported measurement uncertainties and found that estimated ceilings ranged from 0.68 to 0.99 across domains. VEFill’s per-protein correlations, between 0.52 and 0.91 under an 80/20 split, sat consistently below but generally close to these limits, indicating that much of the remaining discrepancy between predicted and observed scores reflects measurement noise rather than model failure. Error profiles were lowest for near-neutral variants, where data are densest, and predictions at the extreme deleterious and high-activity tails should be read as conservative approximations rather than precise estimates.

Generalization beyond the training data showed both promise and limits. When the cross-protein model was applied to eight full-length proteins from independent MaveDB datasets, including Parkin, PTEN, aspartoacylase, calmodulin, TPK1, and TP53, it performed substantially better on stability-based assays than on activity-based ones, a mismatch the authors attribute to differences between training and test assay modalities and the complexity of cellular phenotypes. A lightweight two-feature version using only ESM-1v embeddings and positional mean scores performed comparably to the full model in most settings, offering a practical alternative when evolutionary scores or other annotations are unavailable. Analysis of protein family composition revealed that a few Pfam families dominated the training data, but retraining on a Pfam-unique subset increased error only modestly, suggesting the findings are robust rather than inflated by family-level redundancy.

Fine-grained evaluations exposed the model’s blind spots in biologically meaningful ways. Proline substitutions consistently produced elevated errors, likely because their unusual conformational constraints disrupt secondary structure in ways that sequence-based features do not fully capture. In the TRIM44 zinc finger domain, histidine-to-cysteine mutations were poorly predicted when entire positions were held out but accurately recovered when single variants were withheld, hinting at context-dependent chemistry, such as zinc coordination, that only fine-grained positional information can resolve. The authors suggest that future versions could incorporate explicit structural features, cross-species transfer learning, or systematic assay harmonization to close these gaps.

Overall, VEFill arrives at a moment when the variant effect field is scaling rapidly, with MaveDB now listing more than 2,600 public datasets and over 1,100 for human proteins. By providing an interpretable, scalable tool that turns partially complete mutational maps into denser, more usable resources, the study lowers the experimental barrier for variant prioritization, protein engineering, and the clinical interpretation of missense mutations. The code and trained models have been released openly, inviting the community to apply sparse-library design strategies and, ultimately, to build the comprehensive Atlas of Variant Effects that the field has long envisioned.

Subject of Research: Machine learning-based imputation of missing scores in deep mutational scanning datasets across human protein domains

Article Title: VEFill: accurate and generalizable deep mutational scanning score imputation across protein domains

Article References: Polunina, P. V., Maier, W., & Rubin, A. F. (2026). VEFill: accurate and generalizable deep mutational scanning score imputation across protein domains. Molecular Systems Biology, 22(6), 979-1002. https://doi.org/10.1038/s44320-026-00203-y

Image Credits: AI Generated

DOI: 10.1038/s44320-026-00203-y

Keywords: deep mutational scanning, variant effect prediction, machine learning, gradient boosting, protein stability, ESM-1v, protein language models, MaveDB, Human Domainome, imputation, missense variants, bioinformatics

The Indian hedgehog protein belongs to the Hedgehog family of signaling molecules, a group of secreted morphogens that is conserved across vertebrates and includes Sonic hedgehog, Indian hedgehog, and Desert hedgehog. While Sonic hedgehog is the most prevalent member and governs limb growth, neural patterning, and organ development, Indian hedgehog is the star of the skeletal show. It regulates chondrocyte differentiation, ossification, and skeletal morphogenesis, and it does so partly through its interaction with parathyroid hormone-related peptide, a molecule that controls the proliferation and maturation of cartilage cells in developing fetal bones. Mice engineered to lack Indian hedgehog display profound abnormalities in bone formation, and in humans, mutations in the IHH gene have already been linked to brachydactyly type A-1, a condition marked by shortened fingers and toes, and to acrocapitofemoral dysplasia, characterized by short limbs, a relatively large head, and a narrow thorax. One well-known example is the Asn100Asp substitution, which is associated with brachydactyly type A1.

The starting point for the new analysis was the dbSNP database, the National Center for Biotechnology Information’s comprehensive catalog of human genetic variation. The researchers identified a total of 436 single nucleotide polymorphisms within the IHH gene, of which 280 were classified as non-synonymous, meaning that the single-letter DNA change actually alters the amino acid sequence of the resulting protein. Non-synonymous variants are the most consequential class of common genetic variation because they can directly reshape a protein’s structure and function, although many turn out to be harmless. The challenge, and the central aim of the study, was to separate the potentially disease-causing variants from the benign majority using computational methods that are far faster and cheaper than laboratory experiments.

To accomplish this triage, the team deployed a battery of seven independent prediction algorithms: PANTHER, FATHMM, SIFT, PolyPhen, Mutation Assessor, SNPs&GO, and Provean. Each tool interrogates a mutation from a different angle. FATHMM relies on a hidden Markov model and flags variants with scores above 0.4 as potentially pathogenic. PolyPhen evaluates sequence homology and assigns probability scores, treating values above 0.5 as deleterious. SIFT weighs sequence conservation and amino acid properties, classifying mutations with scores below 0.05 as harmful, while Provean applies a cutoff of minus 2.5. The researchers adopted a deliberately stringent consensus rule: only variants predicted to be damaging by at least five of the seven tools advanced to the next stage. This approach, previously applied to genes such as HBA1, BRCA1, and BRCA2, narrowed the field from 280 non-synonymous variants to 116 candidates with a high likelihood of functional impact.

Structure and stability came next. A protein’s three-dimensional fold is essential to its function, and mutations that destabilize that fold can trigger misfolding and loss of activity. The team used two complementary tools, I-Mutant 3.0 and DynaMUT, to estimate how each candidate mutation would change the protein’s free energy of stability. I-Mutant 3.0, which is built on a support vector machine with roughly 80 percent accuracy, classifies mutations as neutral, destabilizing, or stabilizing based on predicted changes in free energy, while DynaMUT combines graph-based signatures with normal mode dynamics to reach a consensus verdict. The verdict was striking: 103 to 104 of the candidate mutations were predicted to destabilize the protein. Even more telling, 53 of the 116 deleterious variants fell within functionally significant regions of the protein, particularly the stretch between amino acids 28 and 202, which is involved in metal binding, cholesterol transfer, lipidation, glycosylation, and autocleavage, with calcium and zinc ions contributing to structural integrity.

To visualize what these mutations actually do to the protein’s architecture, the researchers retrieved the native structure of the N-terminal signaling domain of Indian hedgehog from the Protein Data Bank, using the structure with the identifier 3K7G. This domain is the business end of the protein, the part that binds the PTCH1 receptor on target cells, so mutations here are especially likely to disrupt signaling. Mutant models were generated by homology modeling with SWISS-MODEL, achieving more than 90 percent sequence identity with the native protein, and were validated with PROCHECK, which assesses stereochemical quality through Ramachandran plots. Most mutant models placed more than 90 percent of residues in favored regions with G-factor values above 0.5, confirming that the models were reliable enough for further comparison. Superimposing the mutant structures onto the native one with PyMOL revealed root-mean-square deviation values exceeding 1 angstrom for many variants, indicating significant structural deviation, while energy minimization analyses showed altered energy profiles that suggested the mutations could substantially reshape the protein’s interaction landscape.

The most revealing experiments were computational docking studies that asked how the mutations affect the handshake between Indian hedgehog and its receptor, PTCH1. Under normal conditions, PTCH1 suppresses a protein called Smoothened; when a Hedgehog ligand binds PTCH1, the brake is released and the signaling cascade proceeds. Using the PatchDock server, which exploits shape complementarity principles, and refining selected solutions with FIREDOCK, the team compared the binding behavior of native and mutant proteins. The results were nuanced. Several mutations, including G69S, L166M, G137D/E142K, and A184V/A184D, produced markedly higher geometric shape complementarity scores than the wild-type protein, with A184V and A184D showing the greatest potential to enhance Ihh-PTCH1 binding and possibly hyperactivate the downstream pathway, a scenario consistent with diseases of inappropriate bone growth. Meanwhile, mutations C28Y, Y49C, and E68G displayed strongly negative atomic contact energies, indicating favorable binding energetics, whereas R66H and R32P showed reduced values that could stabilize the complex in unusual ways. Intriguingly, some variants such as G29R and G202C improved structural complementarity while simultaneously worsening desolvation energy, a reminder that shape and energetics do not always move in lockstep. The overall pattern echoes the E95K mutation already associated with brachydactyly type A1, lending plausibility to the idea that altered receptor binding is a genuine disease mechanism.

The study also zoomed out to place Indian hedgehog in its cellular context. Using the STRING database, the researchers constructed a protein-protein interaction network containing Ihh and ten partner proteins, connected by 41 edges with a high average local clustering coefficient of 0.89 and a PPI enrichment p-value of 3.62e-13, indicating statistically significant connectivity. The network includes PTCH1 and PTCH2, the receptors that negatively regulate the pathway; Smoothened, the signal transmitter; HHIP, a negative regulator that restricts Hedgehog diffusion; DISP, which helps release Hedgehog from producing cells; the co-receptors Cdo and Boc; Hedgehog acyltransferase, which performs the palmitoylation essential for signaling activity; GAS1, a positive regulator; and PTHLH, the parathyroid hormone-related peptide that governs chondrocyte behavior at the growth plate. Mutations in many of these partners are independently associated with skeletal malformations, including Gorlin syndrome, and the authors argue that structural changes in Ihh could ripple through this entire network, disrupting the crosstalk between Hedgehog, Wnt, BMP, and Notch signaling that coordinates skeletal formation and repair.

The practical implications extend toward the clinic. Because the deleterious mutations appear to alter receptor binding and pathway activity, the authors suggest that the Ihh protein could serve as a potential drug target, and that restoring signaling balance with small molecules or biologics might one day help patients with IHH-related disorders. Alternative strategies could target downstream effectors of the Wnt or BMP pathways, activate SMAD proteins, or inhibit negative regulators such as HHIP. Biomarkers derived from gene expression signatures or protein interaction profiles altered by IHH mutations, including changes in PTHLH expression or GLI transcription factor activity, could support earlier diagnosis and personalized therapy. The authors are careful to note that these are computational predictions, and that structural and biochemical experiments will be needed to validate the functional consequences of the flagged variants. Even so, by distilling 436 raw variants down to a prioritized set of 53 mutations concentrated in the receptor-binding domain, the study provides exactly the kind of focused roadmap that experimental geneticists and drug developers need to translate a database of single-letter changes into a deeper understanding of skeletal disease.

Subject of Research: Bioinformatics prediction of pathogenic non-synonymous variants in the IHH gene linked to bone development abnormalities

Article Title: Predicting pathogenic variants in the IHH gene associated with bone development abnormalities: a bioinformatics approach

Article References: Sunkar, S., Pavankumar, G., Hariharan, V., Namrata, K., Nachiyar, C. V., & Athista, M. (2025). Predicting pathogenic variants in the IHH gene associated with bone development abnormalities: a bioinformatics approach. Discover Biotechnology, 2(1), Article 8. https://doi.org/10.1007/s44340-025-00016-z

Image Credits: AI Generated

DOI: 10.1007/s44340-025-00016-z

Keywords: IHH gene, Indian hedgehog, nsSNPs, bone development, brachydactyly, acrocapitofemoral dysplasia, Hedgehog signaling, PTCH1, protein stability, molecular docking, bioinformatics, skeletal dysplasia

To appreciate why this matters, it helps to understand how the glutamine synthetase system has worked for decades. Chinese hamster ovary cells, or CHO cells, are the workhorses of the biotechnology industry, producing the majority of recombinant therapeutic proteins on the market, from monoclonal antibodies to enzyme replacement therapies. When researchers want a CHO cell line to make a therapeutic protein, they typically link the gene for that protein to the gene for glutamine synthetase, an enzyme that allows cells to synthesize glutamine, an amino acid essential for growth. When the cells are grown in medium lacking glutamine and containing methionine sulfoximine, an inhibitor of glutamine synthetase, only cells that produce enough of the enzyme survive. The logic is straightforward: cells that have integrated more copies of the construct, or that express it more strongly, make more glutamine synthetase and therefore grow better under selection. Over time, this enriches the population for high-producing clones.

But the system has a well-known weakness, and it lies in the intrinsic biochemistry of glutamine synthetase itself. The protein is remarkably stable inside the cell, persisting long after it is made. That stability means that even cells with modest expression of the construct accumulate enough enzyme to survive selection. In practical terms, the selection stringency is low: the survival bar is easy to clear, and cells with suboptimal productivity slip through alongside the rare high performers. Traditional workarounds involve escalating concentrations of methionine sulfoximine or laborious rounds of gene amplification and clone screening, both of which add months to cell line development and introduce regulatory and manufacturing complexities. Methionine sulfoximine, after all, is a chemical inhibitor that must be rigorously controlled and eventually removed from the production process, and amplified gene copies can be unstable over long culture periods.

The Inha team approached the problem from an angle that has become increasingly fashionable in cell engineering: proteostasis, the tightly regulated balance between protein synthesis and protein degradation. Rather than trying to control how much glutamine synthetase mRNA is transcribed or translated, they attacked the protein itself, engineering variants that carry modular degrons, short peptide sequences that tag a protein for destruction by the cell’s own degradation machinery. By fusing these degrons to glutamine synthetase, the researchers created what they call destabilized GS, or dGS, variants that are actively and continuously broken down inside the cell. A cell can only maintain sufficient glutamine synthetase activity if it produces the protein at a very high rate, which in turn requires high expression of the linked construct. The degradation tag effectively raises the survival bar, making selection far more stringent without changing the selection conditions at all.

The researchers tested this concept in CHO cells engineered to express either a reporter protein or Etanercept, a fusion protein that combines an antibody fragment with a tumor necrosis factor receptor and is used to treat autoimmune conditions such as rheumatoid arthritis. When the standard, stable glutamine synthetase was used as the selection marker, the resulting cell pools showed the familiar pattern: a broad distribution of productivity, with many mediocre producers surviving selection. When the destabilized GS variants were used instead, the picture changed markedly. The dGS-selected pools were selectively enriched for cells with markedly enhanced productivity, demonstrating that accelerating protein turnover translates directly into stronger selection pressure. The effect worked across both the reporter system and the commercially relevant Etanercept construct, suggesting the strategy is general rather than product-specific.

Molecular analysis of the enriched pools revealed something particularly interesting about where the productivity gains came from. A common assumption in cell line development is that higher productivity comes from gene amplification, meaning the cells have acquired more copies of the therapeutic gene, often through selection regimes designed to force amplification. But when the team examined the dGS pools, they found that the improvements were driven primarily by increased per-copy mRNA abundance rather than by an increase in gene copy number. In other words, the cells that survived the stringent selection were not those with the most genes, but those that made the most messenger RNA from each copy they had. This distinction is significant for manufacturing, because per-copy expression efficiency can be more stable across long production runs than amplified gene arrays, which are prone to silencing or loss over time.

The performance numbers underscore the practical potential of the approach. The top-performing dGS pool achieved an Etanercept titer of 870 milligrams per liter in shake-flask batch culture, a respectable figure for a non-optimized, drug-free system that did not rely on extensive gene amplification. Equally important, the pool maintained stable production over 80 days of continuous passaging, indicating that the high-producing phenotype was not a transient artifact of selection but a durable property of the enriched cell population. Stability over weeks of culture is a critical requirement for commercial manufacturing, where a production campaign can run for many weeks and any decline in titer translates directly into lost product and higher costs. The fact that these results were obtained without chemical selection agents or amplification regimes suggests the method could simplify both the development timeline and the regulatory dossier for new biologic drugs.

The broader significance of the study lies in its demonstration that proteostasis engineering, the deliberate manipulation of protein degradation rates, can serve as a design tool for selection systems, not just as a way to study protein turnover. By choosing or tuning a degron, researchers can in principle dial in exactly how much of a survival protein a cell must produce to stay alive, converting a binary drug-based selection into a graded, high-stringency filter that favors the best producers. Because the strategy operates at the level of protein stability rather than transcription or translation, it is conceptually orthogonal to existing optimization approaches and could be combined with them. The authors suggest that this capability has the potential to accelerate cell line development and reduce manufacturing complexity in biopharmaceutical production, two goals that the industry pursues relentlessly given the enormous cost and time pressures of bringing a biologic to market.

There are, of course, questions that future work will need to address before degradable selection markers become standard practice in industrial cell line development. The study was conducted at shake-flask scale, and performance in the stirred-tank bioreactors used for commercial manufacturing will need to be confirmed. The long-term behavior of degron-tagged proteins under the metabolic stresses of large-scale culture, and the interaction of accelerated protein degradation with the cellular burden of producing a therapeutic protein at high levels, remain areas for further investigation. Nevertheless, the core result stands as a striking proof of concept: by making an essential enzyme deliberately fragile, the researchers turned the cell’s own protein quality control system into a powerful engine for finding its best producers. In an industry where every percentage point of titer improvement carries real economic weight, teaching selection systems to exploit proteostasis may prove to be one of the more consequential ideas to emerge from cell line engineering in recent years.

Subject of Research: Proteostasis engineering of glutamine synthetase for stringent, drug-free selection of high-producing Chinese hamster ovary cell lines

Article Title: Proteostasis engineering of glutamine synthetase for stringent, drug-free selection in Chinese hamster ovary cells

Article References: Kim, H., Park, J. U., Ha, J., & Baik, J. Y. (2026). Proteostasis engineering of glutamine synthetase for stringent, drug-free selection in Chinese hamster ovary cells. Applied Microbiology and Biotechnology. https://doi.org/10.1007/s00253-026-14015-5

Image Credits: AI Generated

DOI: 10.1007/s00253-026-14015-5

Keywords: CHO cells, glutamine synthetase, proteostasis, degrons, protein degradation, selection stringency, cell line development, biopharmaceutical manufacturing, Etanercept, per-copy mRNA abundance, drug-free selection, protein stability

The quantity at the heart of the study is ΔΔG, the change in a protein’s folding free energy caused by a mutation. When ΔΔG is favorable, the mutated protein holds together at least as well as the original; when it is unfavorable, the mutation destabilizes the structure. Getting this number right matters far beyond academic curiosity. Enzyme engineers hunting for catalysts that survive industrial temperatures rely on stability predictions to prioritize which of thousands of possible mutations to synthesize. Clinicians interpreting genomic variants need to know whether a single amino acid substitution in a human protein is likely to sabotage its folding, a frequent mechanism underlying inherited disease. Experimental measurement of ΔΔG is slow and expensive, so accurate computational surrogates have enormous leverage.

Why has prediction remained hard? Because the information that determines stability lives at multiple distances simultaneously. A mutation’s immediate neighborhood — the few residues packed around it in the folded structure — sets the local physicochemical environment: hydrogen bonds broken or formed, hydrophobic pockets filled or emptied, electrostatic contacts gained or lost. Yet the global context matters too. Distant segments of the chain can pack against the mutation site in three-dimensional space, allosteric communication can propagate perturbations far from their origin, and the overall sequence encodes the evolutionary constraints that keep the fold viable. A model that sees only the local scene misses long-range consequences; one that sees only the global picture blurs the fine chemical detail where many decisive interactions live.

The new framework, developed by Yuming Zhao, Ziqi Liu, Benzhi Dong, and Dali Xu, confronts this two-scale problem with an explicitly parallel design built on top of ESM-2, a large protein language model trained on billions of evolutionary sequences. ESM-2 converts each amino acid sequence into a rich numerical embedding in which positions carry contextual information learned from the statistics of natural proteins. From these embeddings, the researchers compute a difference sequence — a position-by-position numerical portrait of what the mutation changes. That difference representation then feeds two independent processing branches running side by side.

The first branch is a convolutional neural network, a family of architectures descended from image-recognition research. Convolutional filters slide across the sequence and respond only to a finite window of positions, making them naturally suited to capturing short-range motifs: the local pattern of chemical-property changes surrounding the mutation. The second branch is a Transformer, the architecture behind modern large language models, whose self-attention mechanism lets every position weigh information from every other position in the sequence. This gives the model unrestricted reach for sequence-wide dependencies, the long-range couplings that convolutions cannot span. Crucially, the two branches do not process the data one after the other; they operate simultaneously, and a gated fusion mechanism — a learned weighting layer — decides adaptively how much of each representation to blend into the final prediction for any given input.

The authors frame the architectural question precisely: is there complementary value in modeling short-range patterns and long-range dependencies independently and in parallel, rather than successively within a single pathway? Their ablation experiments address this directly. Introducing the serial version of the architecture provides the largest gain over raw ESM-2 embeddings alone, confirming that structured local-global processing is the core of the method’s power. Switching from serial to parallel processing then yields additional improvements across most evaluated metrics. In other words, allowing the local and global views to be formed without contaminating each other — and then fused on learned terms — extracts information that a strictly sequential pipeline leaves on the table.

Two further ingredients push the model toward physical realism. First, explicit physicochemical difference features encode how the wild-type and mutant residues differ in fundamental properties such as size, charge, and hydrophobicity, giving the network direct descriptors of mutation-type heterogeneity rather than forcing it to rediscover basic chemistry from embeddings. Second, the researchers confronted a subtle logical property of ΔΔG itself: antisymmetry. The stability change of mutating residue A to B is, in principle, the mirror image of mutating B back to A. Deep networks do not respect such symmetries unless they are built in or encouraged. The team used reverse-mutation data augmentation, training on both directions of each mutation, and embedded a transitivity constraint in the joint loss function — a thermodynamic regularization that ties predictions together so the model’s outputs obey the consistency rules that real free-energy cycles demand.

The performance evaluation rested on two widely used benchmarks. S669 is the standard single-mutation dataset for ΔΔG prediction, while PTmul-NR tests generalization to multiple mutations and other, harder regimes. On both, the parallel framework achieved results comparable to state-of-the-art sequence-based methods. On S669 specifically, the complete model delivered numerical improvements over the serial baseline in the three canonical metrics: the Pearson correlation coefficient, which measures how well predictions track experimental values; the root-mean-square error; and the mean absolute error. Notably, the reduction in mean absolute error remained statistically significant even after correction for multiple comparisons — a rigorous standard that many benchmark studies skip — and the parallel design incurred only limited additional computational cost, an important practical consideration for labs scanning large mutational libraries.

The significance of the work extends beyond its leaderboard position. It contributes an architectural lesson to the fast-growing field of protein language model applications: as pretrained embeddings become commodities, the differentiating factor is increasingly the downstream network that interprets them. Demonstrating that parallel local-global extraction with gated fusion beats serial processing — and that thermodynamically motivated constraints measurably improve consistency — gives other developers concrete design principles. It also aligns with a broader trend of injecting physical prior knowledge into machine learning models, a strategy that tends to improve generalization on exactly the sparse, noisy datasets that characterize experimental biophysics.

There remain, of course, honest limits. Sequence-based models infer stability without explicit three-dimensional structures, so their mechanistic interpretability is indirect, and benchmark datasets like S669 carry well-documented biases, including errors and overrepresentation of certain protein families, that cap achievable accuracy for everyone in the field. The authors are careful to describe their method as comparable to, rather than decisively surpassing, the strongest competitors — with the statistically robust gain in mean absolute error as the clearest architectural dividend. Still, for protein engineers hoping to design thermostable enzymes and for geneticists triaging which variants in a patient’s genome are most likely to destabilize a protein, the message is encouraging: reading a mutation’s local neighborhood and its global context simultaneously, disciplined by the laws of thermodynamics, squeezes genuinely better predictions out of the same underlying biological information. As protein language models continue to improve, frameworks like this one — architected to respect both the chemistry of the immediate environment and the physics of the whole fold — are positioned to ride that rising tide directly into the lab.

Subject of Research: Prediction of mutation-induced protein stability changes using parallel local and global feature modeling on protein language model embeddings

Article Title: Parallel local–global feature modeling with thermodynamic regularization for predicting protein stability changes

Article References: Zhao, Y., Liu, Z., Dong, B., & Xu, D. (2026). Parallel local–global feature modeling with thermodynamic regularization for predicting protein stability changes. BMC Bioinformatics. https://doi.org/10.1186/s12859-026-06669-9

Image Credits: AI Generated

DOI: 10.1186/s12859-026-06669-9

Keywords: protein stability, ΔΔG prediction, ESM-2, deep learning, Transformer, convolutional neural network, gated fusion, thermodynamic regularization, protein engineering, bioinformatics, S669 benchmark, machine learning