The phenomenon at the heart of the study is called mistranslation. Ribosomes, the molecular machines that read messenger RNA and assemble proteins, normally interpret each three-letter codon with high fidelity. But when tryptophan runs short, two distinct errors emerge. In one, ribosomes stall at tryptophan codons and slip out of their reading frame, generating entirely different protein sequences from unaltered mRNA, a behaviour the field has dubbed translational sloppiness. In the other, a subtler and arguably stranger process, the tryptophan-transfer RNA is misloaded with phenylalanine, so the ribosome inserts the wrong amino acid while continuing to read the message in frame. The resulting proteins, termed W>F substitutants, carry phenylalanine wherever tryptophan should have been. Previous work had shown that these substitutants accumulate in tumours and correlate with T cell activity, but whether they were merely inevitable accidents of a stressed translation apparatus or actively regulated by the cancer cell’s genome remained an open question.

To answer it, the team built an elegant genetic tool. Their dual-reporter vector places a single tryptophan codon inside the sequence of SIINFEKL, a well-studied peptide that is displayed on the cell surface by the mouse H-2Kb major histocompatibility complex when correctly produced. Downstream, in a different reading frame, sits a turbo-green fluorescent protein that can only be expressed if the ribosome frameshifts at that same tryptophan codon. The system therefore reports both mistranslation modes simultaneously: surface presentation of SIINFEKL signals tryptophan-to-phenylalanine substitution, while green fluorescence signals frameshifting. When the researchers introduced this construct into A549 lung cancer cells and treated them with interferon-gamma, tryptophan-free medium, or both, they observed strong SIINFEKL induction under conditions that depleted tryptophan, and mass spectrometry confirmed that endogenous cellular proteins carried the same W>F substitutions. By comparing peptide intensities, they estimated the median extent of mistranslation at roughly one percent of the relevant proteome, a strikingly high error rate for a process once assumed to be vanishingly rare.

With the reporter validated, the team launched a genome-wide CRISPR-Cas9 knockout screen using the Brunello library, treating the cells with interferon-gamma plus tryptophan depletion and then sorting, by fluorescence-activated cell sorting, the rare cells that retained frameshifting but had lost SIINFEKL presentation. This strategy filtered out genes affecting tryptophan levels or reporter expression and homed in on factors specific to codon reassignment. Alongside expected hits such as WARS1, the tryptophanyl-tRNA synthetase, and antigen-presentation genes including B2M and TAP1, the screen surfaced two unexpected candidates: ADAR1, an adenosine-to-inosine RNA-editing deaminase, and FTSJ1, a 2′-O-methyltransferase that decorates the anticodon loops of several transfer RNAs. Neither had previously been implicated in mistranslation, and their emergence suggested that cancer cells actively invest in the machinery of translational error.

The two enzymes turned out to operate through fundamentally different mechanisms. ADAR1, the team found, is a global enabler of mistranslation. This enzyme resolves double-stranded RNA structures by converting adenosine to inosine, thereby preventing the activation of antiviral sensors such as protein kinase R that would otherwise shut down protein synthesis. When the researchers knocked out ADAR1, global translation dropped, and W>F substitutant production collapsed even though tryptophan remained depleted and IDO1 remained induced. Crucially, re-expressing a wild-type ADAR1 restored substitutant production, whereas a catalytically dead mutant did not, proving that the editing activity itself is required. Proteomics revealed that ADAR1 loss also suppressed histidine-to-glutamine substitutions under histidine starvation, indicating that its influence extends across multiple mistranslation programmes rather than being confined to tryptophan.

Digging deeper, the researchers connected ADAR1 to the ribosome-associated quality control pathway, or RQC, a conserved system that detects stalled and collided ribosomes and resolves them. Knocking out factors acting at or upstream of ribosome splitting, including ZNF598, RACK1 and ABCE1, suppressed W>F substitutants, while disruption of more downstream effectors had little effect. ADAR1-deficient cells showed reduced protein levels of ZNF598, RACK1 and ABCE1, and the authors propose that ADAR1 sustains the expression of this quality-control machinery, which in turn is essential for the aberrant translation events that flourish under nutrient deprivation. Intriguingly, the interferon-inducible p150 isoform of ADAR1, long known to respond to type I interferons, was also induced by interferon-gamma, a type II interferon central to anti-tumour immunity, hinting at a previously underappreciated role for the enzyme in the tumour-immune battleground.

FTSJ1, by contrast, proved to be a specialist. Loss of the enzyme reduced W>F substitutants without touching global translation, frameshifting, or other known mistranslation events such as histidine-to-glutamine and phenylalanine-to-tyrosine substitutions. Transfer RNA sequencing showed that FTSJ1 loss specifically erased methylation marks at positions 32 and 34 of the tryptophan tRNA anticodon loop, modifications that depend on the cofactors THADA and WDR6 respectively. Catalytically inactive FTSJ1 mutants failed to rescue substitutant production, and the effect was reproduced across lung, glioblastoma, breast, melanoma and prostate cancer cell lines, underscoring the generality of the mechanism. The specificity is remarkable: even though FTSJ1 also methylates the phenylalanine tRNA, phenylalanine-to-tyrosine mistranslation was unaffected, pointing to a uniquely tailored role for the tryptophan tRNA modification.

The biochemical explanation emerged from in vitro experiments with recombinant human WARS1, the synthetase that normally charges tryptophan tRNAs with tryptophan. Under tryptophan scarcity, WARS1 can be tricked into loading phenylalanine instead, but this mischarging is inefficient. Using microscale thermophoresis and aminoacylation assays, the team showed that methylated tryptophan tRNA binds far more readily to phenylalanine-bound WARS1 than does unmethylated tRNA, while binding to tryptophan-bound WARS1 is unaffected by the methylation state. In other words, FTSJ1’s chemical marks act as a molecular handshake that only matters when the enzyme is carrying the wrong amino acid. Normal translation proceeds untouched when tryptophan is abundant, but when the immune system strips tryptophan away, the methylated tRNA is perfectly positioned to accept phenylalanine and keep the protein assembly line running.

The consequences ripple outward to cancer immunology. W>F substitutants generate novel peptides, or neoepitopes, presented on HLA molecules, and some of these, such as the broadly shared TMBIM6 W>F neoepitope, can be recognised by T cells and targeted by adoptive cell therapies. The new study shows that FTSJ1 knockout reduces the presentation of W>F neoepitopes in vitro and in xenograft tumours infiltrated by T cells, and correspondingly dampens T cell activation in co-culture assays. This cuts both ways: on one hand, tumours with high FTSJ1 may diversify their immunopeptidome in ways that could invite immune recognition; on the other, the adaptive benefit of sustaining protein synthesis during tryptophan starvation may outweigh that risk, which may explain why both ADAR1 activity and FTSJ1 expression are elevated in tumour tissues compared with adjacent healthy tissue in clinical proteomics datasets. The authors suggest that phenylalanine’s close resemblance to tryptophan makes these substitutions relatively harmless to the cancer cell, allowing translation to continue while limiting the ribotoxic stress that uncharged tRNAs would otherwise provoke.

The study reframes mistranslation not as noise but as a regulated, multi-layered survival programme that tumours deploy under immune pressure. ADAR1 provides the global permissive environment by maintaining ribosome quality control and suppressing antiviral translational shutdown, while FTSJ1 fine-tunes the tryptophan tRNA to exploit phenylalanine as a surrogate substrate. That two RNA-modifying enzymes converge on the same error pathway suggests the process is controlled at multiple stages, and the authors suspect further regulators remain to be found. Given that substitutant-derived neoepitopes are already being explored as immunotherapy targets, understanding the genetics of their production opens a potential new lever: manipulating ADAR1, FTSJ1 or their cofactors could either amplify the neoepitope supply to energise anti-tumour T cells or throttle the translational escape hatch that tumours rely on when the immune system tightens the metabolic screws.

Subject of Research: Genetic regulation of tryptophan-to-phenylalanine mistranslation in cancer cells under immune-mediated tryptophan depletion

Article Title: Global and specific mechanisms stimulate mistranslation in cancer

Article References: Global and specific mechanisms stimulate mistranslation in cancer. (n.d.). https://doi.org/10.1038/s41556-026-02088-3

Image Credits: AI Generated

DOI: 10.1038/s41556-026-02088-3

Keywords: mistranslation, cancer, ADAR1, FTSJ1, tryptophan depletion, W>F substitutants, ribosome quality control, tRNA methylation, WARS1, interferon-gamma, neoepitopes, translational sloppiness

The scale of the finding is what makes it remarkable. Proteome diversity has long been acknowledged in principle: heterozygosity in the germline introduces single nucleotide polymorphisms that alter protein sequences, somatic mutations accumulate in tissues over a lifetime, transcriptional errors occasionally corrupt RNA messages, and the ribosome itself is an imperfect machine that misincorporates amino acids at measurable rates. What was missing was a way to detect, quantify, and evaluate the functional consequences of all these variant classes on a proteome-wide scale. The Weizmann team developed exactly that, combining deep mass spectrometry-based proteomics with genomic and transcriptomic data to distinguish variants that are genetically encoded from those that arise through errors during translation, a category the authors call non-genetic substitutions.

The technical challenge is considerable. When a mass spectrometer identifies a peptide whose sequence differs from the reference proteome at a single position, that difference could stem from a germline SNP, a somatic mutation, an RNA editing event, or a mistranslation event in which the ribosome inserted the wrong amino acid. Each explanation carries very different biological implications, and each requires different supporting evidence. By cross-referencing protein-level observations with matched DNA and RNA sequencing from the same tissues, the researchers could partition the 7,215 unique substitutions into their mechanistic classes, confidently localizing each variant to a specific position in a specific protein and assigning it to a genetic or non-genetic origin.

One of the most striking results concerns the abundance of mistranslated proteins. Intuitively, one might expect translational errors to be random noise, with the frequency of any given substitution determined only by the chemical propensities of the misincorporation event. Instead, the team found that the abundance of both genetic variants, such as SNPs and somatic mutations, and mistranslated protein variants mirrors their allele frequencies in the human population. A substitution that is common in the human gene pool tends to appear more abundantly at the protein level, even when it arises through non-genetic means. This unexpected correlation suggests that the translation machinery is tuned, whether by tRNA abundance, codon usage, or amino acid availability, in ways that systematically bias which alternative amino acids appear at which positions, effectively echoing evolutionary statistics in the day-to-day errors of protein synthesis.

This observation leads to the study’s most provocative conceptual claim: non-genetic substitutions may provide a distinct route for exploring protein sequence space, circumventing the mutational constraints imposed by the genetic code. The genetic code restricts which single-nucleotide changes are possible at each codon, and many amino acid pairs are unreachable by any single point mutation. Translational errors are not bound by those rules. A ribosome that misincorporates an amino acid can, in principle, place chemically distant residues at positions that mutation alone could never reach in one step. The proteome, in this view, is not merely a snapshot of the genome but a continuously sampled cloud of sequence variants, some of which may occasionally exhibit useful or biologically meaningful properties without ever being fixed in DNA.

To ensure that the non-genetic variants were not artifacts of the mass spectrometry pipeline, the team went beyond computational analysis and provided experimental validation of non-genetic substitutions on selected purified proteins. Confirming that the substituted proteoforms genuinely exist as intact molecules, rather than as spectral misassignments, is a critical step for a field where false discovery rates in variant peptide identification have historically been a serious concern. The validation anchors the computational survey in biochemical reality and strengthens the case that mistranslation produces stable, detectable protein species in healthy tissue.

The researchers also asked whether non-genetic variation is merely background noise or whether it responds to physiological conditions. In proteome-wide analyses of cancer-derived cell lines subjected to amino acid starvation, they observed specific and recurring non-genetic variation patterns. When cells run short of particular amino acids, the ribosome’s error profile shifts in characteristic ways, producing reproducible substitution patterns across the proteome rather than a random scatter. This condition dependence implies that the landscape of protein variants within a cell is dynamic and environmentally responsive, adding a layer of proteome plasticity that operates entirely below the level of gene expression regulation.

Perhaps most intriguingly, the survey identified hundreds of substituted non-genetic proteoforms that recur consistently in multiple healthy individuals or map to annotated protein functional sites. Recurrence across independent people is difficult to explain by chance, and localization to known functional sites, such as catalytic residues or binding interfaces, raises the possibility that some of these variants carry functional consequences. The authors propose that such substitutions constitute a novel class of functional protein phenotypic variants: sequence changes that arise without any change to the genome, persist as real molecular species, and potentially alter protein behavior. If even a fraction of these variants modulate enzyme activity, binding affinity, or protein stability, the effective functional repertoire of human proteins is substantially larger than annotations based on reference sequences alone would suggest.

The implications ripple outward into several areas of biology and medicine. For personalized medicine, the finding complicates the standard model in which an individual’s disease risk and drug response are read off their genome. If two people with identical genotypes can harbor different repertoires of non-genetic protein variants, shaped by diet, tissue type, and cellular stress, then proteome-level profiling may capture medically relevant information that genomics cannot. For evolutionary biology, the study revives interest in the idea that translation errors could serve as a transient, reversible testing ground for amino acid changes, allowing proteins to probe sequence space before any genetic commitment is made. And for basic biochemistry, the population-level mirroring of mistranslation frequencies hints at deep co-evolution between the genetic code, the translation apparatus, and the amino acid composition of proteins.

Collectively, the findings indicate that non-genetic amino acid substitutions in human proteins provide an abundant and previously underappreciated source of functional proteome expansion. The central dogma remains intact as a description of information flow, but the study shows that the final stage of that flow is far noisier, more structured, and potentially more consequential than the textbook version implies. Every healthy tissue, according to this work, carries thousands of protein variants coexisting with their canonical forms, some genetically encoded and some born of the ribosome’s imperfections, and at least some of them recur predictably across individuals and map onto the very sites where protein function is decided. The human proteome, it turns out, is not a single set of molecules but a population of them, and that population is far larger than the genome alone could ever predict.

Subject of Research: Proteome-wide detection of genetically encoded and non-genetic amino acid substitutions in healthy human tissues

Article Title: Encoded and non-genetic protein variants expand human functional proteome

Article References: Tretyachenko, V., Leiman, T., Morgenstern, D., Levin, Y., Asraf, O., Dahan, O., Dahary, D., & Pilpel, Y. (2026). Encoded and non-genetic protein variants expand human functional proteome. Nature. https://doi.org/10.1038/s41586-026-11124-z

Image Credits: AI Generated

DOI: 10.1038/s41586-026-11124-z

Keywords: proteomics, proteogenomics, mistranslation, protein variants, single amino acid substitutions, central dogma, mass spectrometry, translation errors, functional proteome, human tissues, sequence space, Weizmann Institute