MainTryptophan (W, Trp) is an essential amino acid that can be depleted in cancer cells in response to interferon (IFN)-γ secretion by activated T cells in the tumour microenvironment1. IFNγ induces the transcriptional activation of indoleamine 2,3-dioxygenase (IDO1), an enzyme that degrades tryptophan into kynurenine2,3,4,5. In some cancer cells, limiting tryptophan levels cause ribosome stalling and frameshifting, resulting in mistranslated proteins from unaltered messenger RNA6,7,8,9,10. The ability of cancer cells to use frameshifting to sustain aberrant protein production under amino acid-depleted conditions, termed ‘sloppiness’9, was causally linked to the activation of the mitogen-activated protein kinase pathway8,9.In addition to ribosomal frameshifting, in-frame protein synthesis continued under Trp-depleted conditions via a specific codon reassignment mechanism, reassigning phenylalanine (F, Phe) to tryptophan codons without altering the transcript sequence11. This translational error, resulting in amino acid substitutions, was termed ‘W > F substitutants’11. W > F substitutants are enriched in cancer cells, and their level correlates with IDO1 expression and T cell activity11. Both substituting and frameshifting expand the proteome and immunopeptidome with aberrant proteins and neopeptides, respectively8,9,10,11,12,13,14,15,16. In the absence of tryptophan, the tryptophanyl-tRNA synthetase (WARS1) mis-aminoacylates the tryptophan tRNAs (tRNATrp) with phenylalanine, producing W > F substitutants11. However, whether W > F substitutants result from inevitable errors in the mRNA translation machinery or are regulated in cancer remains unknown (Fig. 1a).Fig. 1: A genetic screen uncovers genetic regulators of W > F substitutants.Full size imagea, Model depicting the consequences of T cell-secreted IFNγ on cancer cell W > F mistranslation. b, Scheme of the drW-vector, containing a single Trp codon, able to simultaneously measure frameshifting and W > F substitutant events. c, Median fluorescence intensity (MFI) ± s.d. of H-2Kb-bound SIINFEKL in A549-drW cells treated as indicated for 48 h. d, MS of A549 cells subjected to the indicated treatment procedures for 48 h. The heatmap depicts the number of tryptic W > F peptides found in 2/2 replicates (top row in black), and the log2 protein intensities for control proteins (marked blue). Each box represents the average of two replicates. e, Quantification of tryptic W > F peptide numbers (d) ±s.d. identified by MS in A549 cells treated for 48 h as indicated (n = 2). The average number of tryptic peptides found in mock-treated cells was subtracted from every condition. f, Intracellular Trp (dots) and kynurenine (Kyn, squares) levels (μM), median ± s.d. in A549-drW cells, upon the indicated treatments maintained for 48 h. g, MFI ± s.d. of out-of-frame (OOF)tGFP in A549-drW cells treated as indicated for 48 h. h, log2 ± s.d. of the ratio between H-2Kb-bound SIINFEKL increased MFI signal over mock-treated cells versus the increased OOFtGFP MFI signal over mock-treated cells (from c and g) in A549-drW cells treated for 48 h with the indicated treatment. i, log2 ± s.d. of the ratio between H-2Kb-bound SIINFEKL increased MFI signal over mock-treated cells versus the increased OOFtGFP MFI signal over mock in A549-drW expressing sgRNA against WARS1, B2M or a non-targeting (NT1) control. The ratio was normalized to sgNT1. The indicated treatment was maintained for 48 h. j, Model depicting the Brunello library screen in A549-drW.1 cells treated with -Trp/IFNγ for 48 h and selection criteria of the screen (cells able to frameshift (OOFtGFP signal) but not substitute (APC-H-2Kb-bound SIINFEKL signal)). k, Volcano plot depicting enriched sgRNAs in the A549-drW.1 cell population sorted as in j. Data points represent the average of two replicates and three or four sgRNA pooled per gene. P values were calculated by MaGeCK. FDR, false discovery rate. Throughout the figure, **P F events and mechanistically explain this by improved binding of methylated tRNATrp to phenylalanine-mischarged WARS1.ResultsA genetic screen identifies genetic regulators of W > F substitutantsTo uncover genes that control mistranslation following IFNγ-induced tryptophan depletion, we designed a genetic screen based on a dual-reporter plasmid (drW-vector) that allows the simultaneous detection of both W > F substitutants and +1 frameshifting events from a single tryptophan-residue placed in the context of the SIINwEKL peptide (Fig. 1b). On one hand, stimulation of W > F substitutants produces the SIINFEKL immunopeptide that is presented by the H-2Kb receptor9,11,17. On the other hand, a downstream out-of-frame (+1 frame) tryptophan-less turbo-GFP (OOFtGFP) coding sequence allows monitoring frameshifting events emerging from the same tryptophan codon10.We stably transduced the drW-vector with H-2Kb into A549 cells, isolated from lung carcinoma, a tissue known for its ability to produce W > F11. We measured SIINFEKL generation and presentation following IFNγ, IFNγ + IDO1 inhibitor (IDOi), tryptophan depletion (−Trp), the combined treatments (−Trp/IFNγ) and depletion of two control amino acids (histidine (−His) and arginine (−Arg)). Both IFNγ and −Trp/IFNγ treatments resulted in high SIINFEKL signal, while tryptophan and the control depletions did not (Fig. 1c). The effect of IFNγ was negated by IDOi in accordance with the required tryptophan depletion for W > F production. During −Trp/IFNγ treatment, no mutations in SIINwEKL mRNA were detected in A549-drW cells (Extended Data Fig. 1a,b). As a negative control for W > F events, we designed a SIINaEKL reporter (drA-vector; Extended Data Fig. 1c), which produced only background signals, indicating the high reliance of the SIINFEKL signal generated from the drW-vector (Extended Data Fig. 1c). We substantiated these results using mass spectrometry (MS), showing induction of endogenous W > F substitutants upon either IFNγ or −Trp/IFNγ treatments and, to a lesser extent, following tryptophan starvation (Fig. 1d,e). Using peptide intensities, calculating the W > F:wild-type (WT) ratio, we quantified the median extent of mistranslation at 1% (Extended Data Fig. 1d). We further examined the expression of control proteins, namely the induction of IDO1 (Fig. 1d), as well as the similar abundance of actin B (ACTB), histones 1–4 (H1–H4), WARS1 and the WT counterparts of the mistranslated peptides (Fig. 1d and Extended Data Fig. 1e). Finally, we confirmed the depletion of tryptophan and the induction of kynurenine using metabolomics (Fig. 1f).In contrast with W > F substitutants, we observed low, though specific, levels of OOFtGFP signal when tryptophan was depleted either alone or in combination with IFNγ (Fig. 1g). IFNγ treatment alone was not sufficient to induce OOFtGFP above background levels, as the controls IFNγ + IDOi, −His or −Arg conditions (Fig. 1g). We confirmed the specificity of the OOFtGFP signal using the drA-vector (Extended Data Fig. 1f) and excluded transcriptional mistakes as the origin of the detected OOFtGFP (Extended Data Fig. 1a,b). Additionally, the drF-vector, expressing SIINFEKL followed by a STOP codon and in-frame tGFP, also showed background tGFP levels, corroborating frameshifting as the cause of tGFP expression (Extended Data Fig. 1c,f). The presence and functionality of the construct were verified by the drastic increase of SIINFEKL presentation upon IFNγ18 or −Trp/IFNγ, with only a marginal increase upon the other treatments (Extended Data Fig. 1c). Thus, frameshifting and codon reassignment are two distinct and independently regulated mechanisms, and drW-vector activity reflects them.To identify genetic drivers of W > F events, we normalized the W > F signal to OOFtGFP expression by dividing the SIINFEKL fold change by the corresponding OOFtGFP fold change. This will filter out events affecting tryptophan levels or reporter expression, and accurately recapitulated SIINFEKL induction by either IFNγ or −Trp/IFNγ, and reduction therof upon IFNγ + IDOi treatment (Fig. 1h). To minimize heterogeneity and enhance signal-to-noise ratio, we selected the A549-drW.1 clone, which displayed induction of OOFtGFP and the strongest induction of SIINFEKL upon −Trp/IFNγ treatment (Extended Data Fig. 1g,h). As expected, knockout (KO) of either WARS1, which drives W > F substitutions11, or B2M, an essential antigen-presentation component19, markedly reduced the SIINFEKL/OOFtGFP ratio after −Trp/IFNγ treatment by lowering SIINFEKL levels, while maintaining OOFtGFP (Fig. 1i and Extended Data Fig. 2a,b).Next, we transduced cells with the Brunello CRISPR-Cas9 library20 and treated them with −Trp/IFNγ for 48 h to robustly induce both W > F SIINFEKL and OOFtGFP (Fig. 1c,g,h), while restricting the screen to factors acting downstream of IFNγ-mediated IDO1 activation. Cells with reduced SIINFEKL presentation but intact OOFtGFP expression were isolated by FACS (Fig. 1j) and compared with unsorted bulk cells by sequencing. MAGeCK analysis21 identified sgRNAs enriched in the SIINFEKL-low population (Fig. 1k), including expected hits such as WARS1 and antigen-presentation genes (B2M, IRF7, TAP1 and STAT22). These were excluded along with sgRNAs targeting general transcription factors and protein degradation genes. We then examined the top eight candidates (ADAR, AMOTL, CSNK2A, DPH6, FTSJ1, MBNL, UXT and XPOT; Extended Data Fig. 2a,b) and focused on the strongest hits. Exportin-T (XPOT), a tRNA export protein23 whose KO reduced W > F production without affecting peptide presentation (Extended Data Fig. 2c,d), and two unexpected RNA processing factors. The adenosine deaminase RNA specific enzyme 1 (ADAR1) and the tRNA (cytidine(32)/guanosine(34)-2’-O)-methyltransferase (FTSJ1).ADAR1 sustains general mistranslation during nutrient deprivation conditionsWe first investigated ADAR1, a deaminase that resolves endogenous double-stranded (ds)RNA through A-to-I editing, thereby preventing Protein Kinase R (PKR)-mediated stress responses and translational repression24,25,26 (Fig. 2a). To validate ADAR1 as a W > F regulator, we tested two Brunello-derived sgRNAs27 targeting ADAR in A549-drW.1 cells (sgADAR#1 validated with TIDE analysis28 (Extended Data Fig. 3a) and sgADAR#3), alongside a positive control targeting WARS1 (sgWARS1). Both sgADARs significantly impaired W > F signal, comparable with sgWARS1, without a notable effect on OOFtGFP expression, resulting in a decreased SIINFEKL:OOFtGFP ratio (Fig. 2b and Extended Data Fig. 3b,c). ADAR1 exists as an IFNα-inducible ADAR1p150 and the constitutive ADAR1p110 isoform26. Western blotting confirmed the specific induction of ADAR1p150 by IFNα, and to a lesser extent, IFNγ (Fig. 2c and Extended Data Fig. 3d,f). While sgADAR#3 targets both isoforms, sgADAR#1 selectively represses ADAR1p150 under IFNγ and IFNα treatments (Fig. 2c and Extended Data Fig. 3d–f). Both IFN types, but not tryptophan depletion, induced a W > F SIINFEKL signal in A549-drW.1 cells, although IFNα had much lower efficiency than IFNγ (Fig. 2d and Extended Data Fig. 3g,h). This likely reflects the stronger induction of IDO1 by IFNγ (Extended Data Fig. 3d,i). Because the robust IFNγ response obscured the more modest increase induced by IFNα, we separately compared IFNα-treated and mock-treated cells, confirming a significant induction of IDO1 (Extended Data Fig. 3d,i). Thus, both IFNα and IFNγ induce the activation of ADAR1 to promote W > F substitutants following IDO1 induction.Fig. 2: ADAR1 sustains mistranslation.Full size imagea, Model depicting the ADAR1 deaminase function that resolves double-stranded (ds)RNA structures, by converting adenosine (A) to inosine (I, A-to-I). b, log2 ± s.d. of the ratio between H-2Kb-bound SIINFEKL increased MFI signal over mock-treated cells versus the increased OOFtGFP MFI signal over mock in A549-drW.1 treated for 48 h with −Trp/IFNγ and expressing the indicated sgRNA. Ratios are normalized to sgNT1. c, Scheme of the ADAR1 p150 and p110 isoforms with their functional domains and regions targeted by sgADAR (sgADAR#1 and sgADAR#3) used in this study, adapted from elsewhere26 (top). Immunoblot analysis for ADAR1 and tubulin (for equal loading) in A549-drW.1 cells expressing sgNT1 or sgADAR#1 and treated as indicated for 48 h (bottom). d, log2 ± s.d. of the MFI ratio as in b, in A549-drW.1 cells treated as indicated for 48 h. e, Number of sites (±s.d.) showing at least a 10% increase in A-to-G editing compared with mock in A549-drW.1 expressing either sgNT1 or sgADAR#1 (n = 2) and treated as indicated for 48 h. f, log2 ± s.d. of the MFI ratio as in b in A549-drW.1 cells treated for 48 h with mock or -Trp/IFNγ. Cells expressed either sgNT1 or sgADAR#1. sgADAR#1 expressing cells were additionally transduced with an empty vector (EV) plasmid or plasmid encoding sgRNA-resistant WT or H910Y-E912A catalytic mutant ADAR1. A549-drW.1 sgNT1 were transduced with EV plasmid for uniformity. Ratios are normalized to sgNT1. g, Intracellular Trp (dots) and kynurenine (Kyn, squares) levels (μM), median ± s.d. in A549-drW.1 cells either expressing sgNT1 or sgADAR#1, upon the indicated treatments maintained for 48 h. h, Scheme of vector in which tGFP and the SIINFEKL peptide sequence are cloned in the +1 frame downstream of the only tryptophan codon (top). log2 ± s.d. of the MFI ratio as in b in A549 cells transduced with the plasmid (as in top) and treated for 48 h with −Trp/IFNγ (bottom). Cells expressed the indicated sgRNA. The two segments of the x axes (separated by //) contain two different experiments (normalized to each respective sgNT1 mock treated) here represented in the same graph for comparison. i, Quantification of tryptic W > F peptide numbers ± s.d. identified by MS in A549-drW.1 expressing either sgNT1 or sgADAR#1 (n = 2) and treated for 48 h as indicated. The average number of peptides found in the mock-treated sgNT1 cells was subtracted from every condition. j, MFI ± s.d. of OPP incorporation in A549-drW.1 expressing sgNT1, sgADAR#1 and an sgRNA-resistant form of WT-ADAR1 (ADAR1), treated with the indicated condition for 48 h. CHX, cycloheximide, used at 0.1 mg ml−1. k, Quantification of tryptic H > Q peptides, mean ± s.d. in A549-drW.1 as in i. The number of H > Q peptides was merged for the two histidine codons: CAC and CAT. Throughout the figure, *P F/OOFtGFP in mock or −Trp-treated A549-drW.1 cells (Extended Data Fig. 5a–c).To exclude impaired antigen presentation as the cause of the reduced SIINFEKL signal upon ADAR1 loss, we used the OOFtGFP-SIINFEKL reporter9 (Fig. 2h), where SIINFEKL presentation depends on frameshifting, not on W > F. As expected, −Trp/IFNγ induced OOFSIINFEKL and OOFtGFP. Their ratio was marginally affected by sgWARS1, whereas sgADAR#1 slightly but significantly increased it, consistent with enhanced IFN signalling (Extended Data Fig. 3j), resulting in more presentation in ADAR1-deficient cells. Thus, ADAR1-loss does not impair antigen presentation, which cannot explain the reduced W > F signal (Fig. 2h and Extended Data Fig. 3m,n). As a positive control, B2M-KO selectively reduced OOFSIINFEKL, decreasing the OOFSIINFEKL:OOFtGFP ratio (Fig. 2h and Extended Data Fig. 3o,p).MS analysis of endogenous peptides confirmed the reporter results: ADAR-KO abolished W > F substitutant generation following IFNγ and −Trp/IFNγ treatment (Fig. 2i). Control protein abundance excluded sample-quality effects (Extended Data Fig. 3q,s), while sustained IDO1 induction and persistent tryptophan depletion excluded restoration of Trp as the cause (Fig. 2g and Extended Data Fig. 3q). Notably, reduced ADAR1 expression was not visible in proteomics analysis, in line with the relatively low expression of the ADAR1p150 isoform when compared with ADAR1p110 (Fig. 2c and Extended Data Fig. 3d). Together, these results confirm at the endogenous level that ADAR1 activity is required for W > F substitutant generation.ADAR1 maintains the ribosome-associated quality control machinery to sustain aberrant protein productionWe next asked whether ADAR1 broadly regulates substitutants or acts specifically on W > F.O-propargyl-puromycin (OPP) incorporation assays33 revealed reduced global translation following ADAR loss under all conditions tested (Fig. 2j), an effect rescued by ADAR1 re-expression. Proteomics analysis further showed that sgADAR#1 suppressed histidine-to-glutamine (H > Q) substitutions34 (Fig. 2k and Extended Data Fig. 3r,t). Taken together, these results suggest that ADAR1 enhances substitutants in tumours and generally affects translation.Because ADAR1 sustains global translation, we assessed mistranslation during 24 h translation inhibition. Cycloheximide and harringtonine abolished translation, W > F substitutants and frameshifting (Extended Data Fig. 5d–f), whereas ADAR1 loss selectively inhibited W > F production (Extended Data Fig. 3b,c), indicating a mechanism beyond global translational repression. The RQC pathway plays a pivotal role in regulating translation under stress conditions, including amino acid depletion (Fig. 3a)35. Loss of factors acting before, during or immediately downstream of ribosome splitting suppressed W > F substitutants, whereas disruption of more-downstream effectors had only minor effects (Fig. 3b). Frameshifting was inhibited by at least one sgRNA for all RQC factors except ABCE1. (Extended Data Fig. 5g). Altogether, these results suggest that RQC plays a key role in the production of aberrant proteins in A549 cells during IFNγ-induced Trp depletion.Fig. 3: ADAR1 sustains mistranslation by maintaining expression of the RQC.Full size imagea, Simplified model depicting the principal players involved in the RQC pathway. b, MFI ± s.d. of H-2Kb-bound SIINFEKL in A549-drW.1 cells transduced and treated as indicated for 48 h. Statistics indicate comparison with sgNT1. c,e,g,i,l,n, Immunoblot analysis for ZNF598 (c), RACK1 (e), ABCE1 (g), ADAR1 (i), IDO1 (l), WARS1 (n) and tubulin (for equal loading) in A549-drW.1 cells expressing sgNT1 or sgADAR#1 and treated as indicated for 48 h. Immunoblots c and n are obtained from the same membrane, therefore the corresponding tubulin is represented twice. d,f,h,j,k,m,o, Quantification of the relative ZNF598 (d as in c, n = 5), RACK1 (f as in e, n = 5), ABCE1 (h as in g, n = 4), ADAR1p150 (j as in i), ADAR1p110 (k as in i, n = 4), IDO1 (m as in l, n = 4), WARS1 (o as in n, n = 3) expression (vs. tubulin) ± s.d., normalized to the sgNT1 control. Throughout the figure, *P F, but not OOFtGFP reporter signal, resulting in a marked decrease in the SIINFEKL/OOFtGFP ratio under −Trp/IFNγ conditions (Fig. 4b and Extended Data Fig. 6a,b). FTSJ1-KOs were confirmed by TIDE analysis28 (Extended Data Fig. 6c,d).Fig. 4: FTSJ1 is required for endogenous W > F substitutants.Full size imagea, Model depicting tRNATrp and tRNAPhe methylation (M) by FTSJ1 in positions 32 and 34. b, log2 ± s.d. of the ratio between H-2Kb-bound SIINFEKL increased MFI signal over mock-treated cells versus the increased OOFtGFP MFI signal over mock of A549-drW.1 treated for 48 h with −Trp/IFNγ and expressing the indicated sgRNA. The ratio was normalized to sgNT1. c, log2 ± s.d. of the MFI ratio as in b in A549 cells transduced with the OOFtGFP-SIINFEKL plasmid (Fig. 2h) and treated for 48 h with −Trp/IFNγ. Outlined dots (sgNT1 and sgWARS1) indicate previously shown data (Fig. 2h) and are shown here for comparison. The ratio was normalized to sgNT1. d,e, Quantification of tryptic W > F peptide numbers ± s.d. identified by MS in A549-drW.1 expressing either sgNT1 or sgFTSJ1#2 (d), or FTSJ1-KO and rescue clones #1 and #3 (n = 2, e). Cells were treated for 48 h as indicated. The average number of peptides found in the mock-treated sgNT1 cells was subtracted from every condition. f, Intracellular Trp (dots) and Kyn (squares) levels (μM), median ± s.d. in A549-drW.1 cells either expressing sgNT1 or two clones of FTSJ1-KO (#1 and #3), upon the indicated treatments maintained for 48 h. Outlined symbols (sgNT1) indicate previously shown data (Fig. 2g) and are shown here for comparison. Throughout the figure, *P F endogenous substitutants but stable expression of control proteins (Fig. 4e, Extended Data Fig. 6j–l and Supplementary Fig. 1b). FTSJ1 phenotypes cannot be explained by increased tryptophan levels (Fig. 4f). Re-expression of sgFTSJ1-resistant FTSJ1 restored W > F production, confirming an on-target effect (Fig. 4e and Extended Data Fig. 6j–l). Together, these results establish FTSJ1 as a key regulator of W > F substitutant generation.FTSJ1-mediated tRNA methylation specifically promotes W > F substitutantsTo determine whether FTSJ1-dependent tRNA methylation underlies W > F production, we performed tRNA-seq37,38. FTSJ1-loss had minimal effects on tRNA levels, altering only tRNAArg(TCT) (Extended Data Fig. 7a). In contrast, it induces changes in base-calling signals at positions 32–34 of mainly tRNAPhe and tRNATrp, consistent with known FTSJ1-dependent modifications39,40 (Fig. 5a and Extended Data Fig. 7b and Supplementary Figs. 2 and 3). These changes are negated by FTSJ1 re-expression (Extended Data Fig. 7c,d and Supplementary Figs. 4 and 5). No effect on base-calling signal of tRNASec was observed despite FTSJ1 being associated with stop codon read-through and metastasis in melanoma41 (Supplementary Figs. 2 and 3). Thus, as expected, FTSJ1 is required for tRNATrp modifications.Fig. 5: FTSJ1-induced tRNA 32–34 methylation is needed for substitutants across several cell lines.Full size imagea, tRNA modification levels determined by alteration in base calling from tRNA sequencing in A549-drW.1 sgNT1 versus FTSJ1-KO #1 (n = 1). The Sprinzl number indicates the location within the tRNA. The z-score indicates the difference in base calling levels between FTSJ1-KO and sgNT1. b, log2 ± s.d. of the ratio between H-2Kb-bound SIINFEKL MFI levels of −Trp/IFNγ versus mock treatment, in A549-drW.1 cells expressing sgNT1 or sgFTSJ1#2. Additionally, cells express an EV or sgRNA-resistant WT, A26P, or K28A FTSJ1 plasmids. Ratios are normalized to sgNT1-EV. c, Quantification of W > F tryptic peptide numbers ± s.d. identified by MS from A549-drW.1 expressing sgFTSJ1#2 and EV or sgRNA-resistant WT, A26P or K28A FTSJ1 (n = 2). Cells were treated for 48 h as indicated. The average number of tryptic peptides found in the mock-treated sgNT1 was subtracted from every condition. d, Model depicting tRNA-Trp and -Phe methylation (M) by FTSJ1 and its cofactors THADA (position 32) and WDR6 (position 34). e, log2 ± s.d. of the ratio between H-2Kb-bound SIINFEKL increased MFI signal over mock-treated cells versus the increased OOFtGFP MFI signal over mock in A549-drW.1 treated for 48 h with −Trp/IFNγ and expressing the indicated sgRNA. The ratio was normalized to sgNT1. f, MFI ± s.d. of OPP incorporation in A549-drW.1 expressing sgNT1 and FTSJ1-KO clones #1 and #3, treated with the indicated condition for 48 h. CHX was used at 0.1 mg ml−1. g, Quantification of tryptic H > Q peptide numbers ± s.d. in A549-drW.1 as in c. The number of H > Q peptides was merged for the two histidine codons: CAC and CAT, and the average number of tryptic peptides found in the mock-treated sgNT1 cells was subtracted from every condition (n = 2). h, Quantification of F > Y peptides ± s.d. in A549-drW.1 as in c. The average number of tryptic peptides found in the mock-treated sgNT1 cells was subtracted from every condition (n = 2). i, log2 ± s.d. of the MFI ratio as in e. Cells expressed the drW-vector and were treated for 48 h with −Trp/IFNγ. For every cell line, the ratio obtained for cells expressing sgFTSJ1#2 is normalized to their respective sgNT1. Outlined dots (A549) indicate previously shown data (Fig. 5e) here represented for comparison. Throughout the figure, *P F production under tryptophan-shortage conditions.FTSJ1-mediated methylation of tRNATrp at positions 32 and 34 requires THADA and WDR6, respectively40,43,44 (Fig. 5d). To determine which modification contributes to W > F production, we generated THADA-KO and WDR6-KO cells and measured SIINFEKL induction. Both KOs reduced SIINFEKL induction without affecting OOFtGFP expression (Fig. 5e and Extended Data Fig. 8c,d), albeit to a lower extent than sgFTSJ1#2. Neither FTSJ1 nor THADA expression changed upon IFNγ treatment, and despite a slight reduction in WDR6 levels (Extended Data Fig. 8e), no changes in tRNATrp modifications at positions 32–34 were detected (Extended Data Fig. 7e,f and Supplementary Data Figs. 6 and 7). Together, these findings indicate that FTSJ1-dependent methylation of tRNATrp at position 32 and 34 promotes efficient W > F production.Because FTSJ1 modifies multiple tRNAs40,45, we asked whether its effect on W > F production reflects a broader role in translation. Unlike ADAR1 (Fig. 2j), FTSJ1-loss did not affect global translation under any condition tested (Fig. 5f). Moreover, it did not affect H > Q34 mistranslation during histidine deprivation (Fig. 5g) or F > Y11 (Fig. 5h) mistranslation during phenylalanine deprivation, despite efficient FTSJ1-KO and rescue (Extended Data Fig. 8f,g). This lack of effect was observed even though control proteins and WT counterpart peptides were unaffected (Extended Data Fig. 8f,g and Supplementary Figs. 1d, e) and tRNAPhe being a known major target of FTSJ1. Therefore, FTSJ1 promotes tryptophan substitutant events specifically, without interfering with global translation or being involved in other known mistranslation events.We further expanded our findings to cell lines other than A549 (RA glioblastoma, MDA-MB-231 breast cancer, A375 melanoma and PC3 prostate cancer). sgFTSJ1#2 supresses W > F production in all examined cell lines (Fig. 5i and Extended Data Fig. 8h–o). Thus, FTSJ1 expression is required for W > F production in different cancer types.FTSJ1-mediated tRNATrp modifications promote mis-aminoacylation without affecting aminoacylation of cognate tRNAAminoacylation starts with highly specific binding of the amino acid to the corresponding aminoacyl-tRNA synthetase, followed by activation, tRNA binding and amino acid transfer to the tRNA (Fig. 6a). WARS1 normally binds tryptophan with high specificity but can mis-aminoacylate tRNATrp with phenylalanine during tryptophan depletion, leading to W > F substitutions11. Given the central role of WARS1 in this process and the specific effect of FTSJ1 on W > F production, we hypothesized that FTSJ1 promotes phenylalanine mischarging of tRNATrp. To test this hypothesis, we performed in vitro aminoacylation experiments using recombinant human WARS1 (rhWARS1), amino acids and tRNA extracted from A549-drW.1-sgNT1. Tryptophan bound WARS1 with much higher affinity than phenylalanine, although both interacted at amino acid concentrations ≥10 mM (Fig. 6b). Consistent with previous findings11, rhWARS1 charged tRNATrp with tryptophan or phenylalanine, but not arginine (Fig. 6c), supporting phenylalanine as a preferred non-cognate substrate.Fig. 6: FTSJ1 methylation increases tRNATrp binding affinity to Phe-bound WARS1.Full size imagea, Model depicting WARS1-mediated aminoacylation. The amino acid (AA; Trp or Phe) first binds to WARS1 and is activated to an aminoacyl-adenylate intermediate. tRNATrp then binds to AA-bound WARS1 and the amino acid is transferred to the tRNA, producing aminoacyl-tRNATrp. The tRNATrp will be released, after which the enzyme returns to its original state. b, Normalized response ± s.d. of red labelled recombinant human WARS1 (rhWARS1) incubated with incremental concentration of either Trp (dots) or Phe (squares). Dots represent the average of three independent experiments. Dissociation equilibrium constant (KD) ± s.e.m. is indicated. c, Relaxed amino acid specificity of WARS1 toward phenylalanine. Aminoacylation activity of rhWARS1 was measured as pmol (mean ± s.d.) of amino acid charged onto tRNA after incubation with 10 OD ml−1 of tRNA from A549-drW.1 sgNT1 in the presence of the indicated amino acid (10 μM Trp, Phe or Arg). Dots represent independent experiments (n = 2). d,e, Role of FTSJ1 in mis-aminoacylation. Aminoacylation activity of rhWARS1 was measured as pmol (mean ± s.d.) of Phe (d) or Trp (e) charged onto tRNA isolated from A549-drW.1 sgNT1, FTSJ1-KO#1 or Rescue#1. Dots represent independent experiments (n = 3). f,g, Normalized response ± s.d. of red labelled rhWARS1 incubated with incremental concentration of tRNA isolated from A549-drW.1 sgNT1 (dots), FTSJ1-KO#1 (squares) or FTSJ1-Rescue#1 (triangles) and 10 mM Trp (f) or Phe (g). Resc, Rescue. Dots represent the average of three independent experiments. The RNA concentration at which 50% of rhWARS1 is bound (BC50) ± s.e.m. is indicated. Throughout the figure, *P F production in tryptophan-deprived conditions, where phenylalanine can bind WARS1.FTSJ1 stimulates W > F neopeptidesW > F substitutants can generate HLA-I-presented neoepitopes that diversify the cancer immunopeptidome and may be targeted by adoptive T cell therapies10,11,12,14,46 (Fig. 7a). To assess the impact of FTSJ1 on T cell activation, we first confirmed that FTSJ1-KO strongly reduced endogenous W > F production in IFNγ-only-treated cells (Fig. 7b), simulating active T cell conditions, while leaving control proteins and WT peptides unchanged (Extended Data Fig. 8e and Supplementary Fig. 1f). OT-1 T cells, recognizing SIINFEKL-H-2Kb (ref. 47), show reduced activity against cells expressing sgFTSJ1#2 (Fig. 7c and Extended Data Fig. 9a; KO validation in Extended Data Fig. 9b). This effect is in line with a lower level of W > F substitutants (Fig. 7b), indicating the potential of FTSJ1 to boost W > F neoepitopes in IFNγ-induced tryptophan-depleted environments.Fig. 7: FTSJ1 is required for W > F substitutants and neopeptide presentation and recognition.Full size imagea, Model depicting the T cell co-culture experiment. Cancer and T cells are co-cultured, where T cell-mediated killing is expected upon IFNγ treatment when FTSJ1 is present. b, Quantification of tryptic W > F peptide numbers ± s.d. identified by MS in A549-drW.1 expressing either sgNT1 or FTSJ1-KO clones #1 and #3 (n = 2). Cells were treated for 48 h as indicated. The average number of peptides found in the mock-treated sgNT1 cells was subtracted from every condition. c, MFI ± s.d. of intracellular TNF used to assess T cell activation in A549-H-2Kb (control) or A549-drW.1 cells transduced with sgNT1 or sgFTSJ1#2, pre-treated with IFNγ or mock for 72 h, and used in co-cultures with OT-1-derived T cells for 4 h at an effector:target (E:T) ratio of 1:1. d, Ratio of the number of W > F peptides versus number of WT peptides ± s.d. determined via immunopeptidomics in A549-H-2Kb or A549-drW.1 transduced with the indicated sgRNA and treated with mock or IFNγ for 72 h. e, Model depicting in vivo experiment. A549MART1FTSJ1-KO and Rescue clones are engrafted into mice, followed by TCRMART1 T cell (or mock PBS) injection to induce IFN-mediated Trp shortage in the tumour and the consequent production of W > F peptides. Tumours were processed with immunopeptidomics. f, Ratio of the number of W > F peptides versus number of WT peptides ± s.d. determined via immunopeptidomics in A549MART1FTSJ1-KO (n = 5) or Rescue (n = 6) xenografts. Mice were treated with PBS or TCRMART1 T cell for 96 h. Each dot represent a tumour obtained from a single animal. g,h, MFI ± s.d. of CD137 used to assess T cell activation in A549HLA-A*24:02 transduced with sgNT1 or sgFTSJ1#2, pre-treated with IFNγ or mock for 72 h (g) or loaded with 1 nM TMBIM6W>F for 24 h (h) and used in co-cultures with TCRTMBIM6W>F.1 T cells overnight at an E:T ratio of 1:2. i, Model depicting the effect of ADAR1 on RQC maintenance and FTSJ1 on tRNATrp charging with phenylalanine, necessary for W > F substitutant production. Throughout the figure *P F/WT peptide ratio was markedly lower in sgFTSJ1#2 (Fig. 7d) cells despite a modest reduction in WT peptides (Extended Data Fig. 9e). We also examined the broadly shared TMBIM6W>F neoepitope13. TMBIM6W>F, but not its WT counterpart, was significantly reduced in sgFTSJ1#2 transformed cells (Extended Data Fig. 9f,g). These findings demonstrate that FTSJ1-dependent W > F substitution extends to the immunopeptidome.To assess FTSJ1 function during T cell-induced tryptophan depletion in vivo, we generated FTSJ1-KO (TIDE in Extended Data Fig. 10a) and rescue A549MART1/HLA-A*02:01 clones, which are recognized by TCRMART1 T cells (Fig. 7e). FTSJ1-KO and rescue were confirmed by proteomics, with no changes in control proteins, WT peptides, or IFNγ-induced IDO1 and WARS1 expression (Extended Data Fig. 10b,c). As expected, W > F production was markedly reduced in FTSJ1-KO cells (Extended Data Fig. 10b,d).A549MART1FTSJ1-KO and rescue cells were engrafted into immunodeficient mice and treated with TCRMART1 T cells or PBS13. Immunopeptidomics confirmed T cell activity (by an increased WT peptide presentation) and revealed an increased number of treatment-specific W > F peptides in T cell-infiltrated tumours (Extended Data Fig. 10e,f). Consistent with our in vitro findings, A549MART1FTSJ1-KO tumours presented fewer W > F neoepitopes than rescue tumours (Extended Data Fig. 10f). Although WT peptide presentation was modestly reduced, the W > F/WT peptide ratio remained significantly lower in FTSJ1-KO tumours, and induction of WT peptides upon T cell treatment was comparable between KO and Rescue (Fig. 7f and Extended Data Fig. 10e,g). These results demonstrate that FTSJ1 is required for efficient induction of W > F neoepitopes by activated T cells in vivo.We further assessed the functional impact of FTSJ1 on W > F neoepitope recognition in vitro using TCRTMBIM6W>F.1 T cells and A549HLA-A*24:02 cells. As expected, IFNγ enhanced T cell reactivity, whereas this response was significantly reduced in sgFTSJ1#2 cells, consistent with their lower TMBIM6W>F neoepitope production (Fig. 7g and Extended Data Fig. 9f). Notably, both cell types elicited similar T cell responses when loaded with synthetic TMBIM6W>F peptide (Fig. 7h), excluding defective peptide presentation. These results indicate that FTSJ1 promotes W > F neoepitope generation and subsequent T cell recognition following IFNγ exposure.ADAR and FTSJ1 activity and expression are favoured in cancerConsistent with its known pro-oncogenic role48,49, ADAR1 activity was elevated in lung tumours (Clinical Proteomic Tumour Analysis Consortium; CPTAC), as reflected by increased A-to-G editing events (Extended Data Fig. 10h). Likewise, FTSJ1 expression was higher in tumours than in adjacent tissue (Extended Data Fig. 10i–m). While FTSJ1 has cancer-specific functions50,51, its role in promoting W > F mistranslation may contribute to cancer adaptation during tryptophan stress following anti-tumour activity.DiscussionWe identified two RNA-modifying enzymes, ADAR1 and FTSJ1, as key regulators of substitutant events during nutrient deprivation in cancer cells (Fig. 7i). Both are required for efficient W > F substitutant production but act through distinct mechanisms. ADAR1 broadly promotes substitutants by maintaining RQC, whereas FTSJ1 specifically enhances W > F reassignments by improving tRNATrp binding to phenylalanine-bound WARS1. FTSJ1-loss did not affect tryptophan-tRNATrp aminoacylation or other amino acid depletion-induced substitutants, suggesting a specialized role in adapting to tryptophan shortage. A-to-I editing and FTSJ1 expression are elevated in tumours, and W > F substitutants are enriched in cancer proteomes, correlating with T cell activity markers11. These findings suggest W > F substitutants may represent an adaptive mechanism that supports translation during tryptophan depletion while limiting ribotoxic stress induced by uncharged tRNAs52. Because phenylalanine closely resembles tryptophan, these substitutions may not be deleterious to cancer cell survival.ADAR1 has two isoforms, p110 and the type I IFN-inducible p150 (ref. 53). We found that type II IFNγ also induces ADAR1p150, suggesting previously unrecognized roles in anti-tumour immunity54. Through A-to-I editing of endogenous dsRNA, ADAR1 prevents activation of immune sensors such as MDA5 and PKR, thereby limiting IFN signalling49 and preserving translation55. Indeed, ADAR-KO increased interferon responses, inhibited mRNA translation and suppressed RQC expression in our model. In line with this positive impact of ADAR1 on cell survival under stress conditions, it is not surprising that its expression and activity are enhanced in cancer proteomes and transcriptomes48,49,56.Our findings link ADAR1 to RQC expression maintenance and show that this is required to sustain mistranslation. While the RQC machinery detects and resolves ribosome collisions57, only proteins acting upstream of ribosome splitting were essential for maintaining W > F mistranslation during IFNγ-induced tryptophan depletion. In contrast, downstream RQC factors had only minor effects on W > F substitutant production. Most RQC factors suppressed frameshifting, except ABCE1, which promotes ribosome splitting58. ABCE1 resembled ADAR1 in selectively affecting W > F substitutions without altering frameshifting, suggesting a possible direct connection. Further studies are required to define how ADAR1 interacts with RQC during amino acid stress.Unlike ADAR1, FTSJ1 does not influence global translation or substitutant events beyond W > F, as far as we tested. This specificity first reflects that FTSJ1-mediated tRNATrp modifications are dispensable for binding to tryptophan-bound WARS1 under normal conditions. Second, W > F specificity is shown by tRNAPhe, which is hypomethylated in FTSJ1-KO cells without affecting F > Y substitutants. In contrast, when WARS1 is bound to phenylalanine, FTSJ1-mediated tRNATrp methylation enhances tRNA binding. These findings suggest that W > F mistranslation is regulated differently from both canonical tryptophan codon translation and other mistranslation events. As aminoacylation requires many WARS1 conformational changes59,60, we propose that structural differences between tryptophan and phenylalanine alter WARS1 conformation, making phenylalanine-bound WARS1 dependent on methylated tRNATrp, thereby explaining the specific role of FTSJ1 in promoting W > F mistranslation.Cancer cells reprogramme translation to support growth, and both ADAR1 and FTSJ1 are frequently upregulated in tumours. Adar promotes tumour growth in mice by suppressing PKR-mediated translational inhibition48,49, and we observed increased A-to-G editing in tumour tissues relative to adjacent healthy tissue. FTSJ1 is likewise elevated in cancer51 and has been associated with tumour progression51 and reduced T cell infiltration61. Together, high ADAR1 and FTSJ1 expression may help sustain protein synthesis in nutrient-poor tumour environments. Although enhanced W > F substitutant production could generate immunogenic neoepitopes, its adaptive benefit in maintaining translation during tryptophan depletion may outweigh the risk of anti-tumour immune recognition.Identifying FTSJ1 as a specific driver of W > F substitutants both in vitro and in vivo shows that this mistranslation process is actively regulated. The additional involvement of ADAR1 and potentially XPOT indicates that W > F substitutants are controlled at multiple stages of translation. This added complexity suggests that further regulators remain to be discovered. Given the potential clinical relevance of substitutants13, uncovering new regulators may provide alternative strategies to induce them. Overall, our findings show that FTSJ1-mediated tRNA methylation is essential for W > F substitutants and that multiple translational processes, including the RQC, regulate W > F peptide production.MethodsCell cultureCell lines were obtained from ATCC. As previously described9,10,11,13,14, A549, A375, MDA-MB-231 and HEK 293T cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM; Gibco). RA and PC3 cells were maintained in Roswell Park Memorial Institute (RPMI) 1640 Medium (Gibco). All media were supplemented with 10% fetal bovine serum (Serana) and 100 U ml−1 penicillin−streptomycin (Gibco). Cell lines were cultured in a humidified incubator at 37 °C with 5% CO2 and were routinely tested negative for Mycoplasma contamination by PCR.Tryptophan-free DMEM/F12 medium was obtained from US Biologicals, and histidine-, phenylalanine- and arginine-free DMEM were custom made (Cell Culture Technology). In contrast to amino acid sufficient media, depleted media were supplemented with 10% heat-inactivated, dialysed fetal bovine serum (Gibco). MG-132 (Selleckchem) was used at a final concentration of 10 µM for 4 h. IFNγ (PeproTech), 1-methyl-L-tryptophan (Sigma) cycloheximide (CHX) (MedChem Express) and harringtonine (harr; Sigma-Aldrich) were used as indicated.T cell cultureTCRTMBIM6W>F.1 and TCRMART-1 transduced T cells were maintained as described in ref. 13 in RPMI 1640 medium supplemented with 10% human serum (One Lambda), 100 U ml−1 penicillin–streptomycin and with 50 U ml−1 IL-2 (Proleukin, Novartis). OT-1 T cells were isolated as described in ref. 11 using a Dynabeads Untouched Mouse CD8 Cells kit (Invitrogen) according to the manufacturer’s protocol and maintained in RPMI 1640 medium containing 10% fetal bovine serum, 50 µM 2-mercaptoethanol (Sigma), 100 U ml−1 penicillin−streptomycin, 100 µg ml−1 IL-2 (ImmunoTools), 5 µg ml−1 IL-7 (ImmunoTools) and 10 µg ml−1 IL-15 (ImmunoTools). When co-cultured with cancer cells, T cell were maintained overnight in RPMI 1640 with 10% fetal bovine serum, supplemented with or without IFNγ.Generation of plasmidsThe drW and drF-vector, were generated using primers listed in Supplementary Tables 1 and 3. To generate the drA-vector, the GeneArt site-directed mutagenesis system (Invitrogen) was used according to the manufacturer’s protocol using the primers listed in Supplementary Table 2. The sgRNAs were cloned in pLentiV2-puro using the Zhang laboratory protocol62,63 with primers listed in Supplementary Table 4 (ref. 20). pCDH-Hygro H-2Kb was used from ref. 9. pCDH BFP was cloned from pCDH-CAG-NCreIntN-EF1a-mTagBFP2 (gift from S. Je (Addgene, #160507; RRID:Addgene_160507)) using the primers listed in Supplementary Table 5. Gene fragments were obtained from IDT for ADAR1, ADAR1 H910Y-E912A, FTSJ1, FTSJ1 A26P and FTSJ1 K28A. All produced vectors were Sanger-sequenced (Macrogen).sgRNA library transformationThe human pLentiCRISPR-v2 pooled KO library (Brunello) was a gift from D. Root and J. Doench (Addgene; plasmid #73178)20. Endura electrocompetent cells (Lucigen) were transformed with Brunello library plasmids as described in ref. 64. Plasmid DNA was prepared from bacterial cultures using PureLink HiPure Plasmid Maxiprep kit (QIAGEN).Lentiviral production and transductionHEK 293T cells were used as described in ref. 11. Vectors of interest, pMDL RRE, pVSV-G and pRSV-REV plasmids were transfected using polyethyleneimine (Polysciences). Cells were transduced by adding, previously frozen, lentiviral supernatant supplemented with Polybrene (Sigma). Cells were selected using 5 µg ml−1 blasticidin (Invivogen), 2 μg ml−1 puromycin (Bio-connect) or 50–1,000 µg ml−1 hygromycin B (Gibco).Reverse transcriptionTotal RNA was extracted from A549-drW.1 cells using TRIzol reagent (Invitrogen), according to manufacturer’s instructions. Samples were treated with Turbo DNase (Thermo Fisher Scientific) and RNA was purified using the Zymo RNA Clean and Concentrator-5 kit (Zymo Research). RNA was converted to complementary DNA with the High-Capacity cDNA Reverse Transcription kit (Applied Biosystems), according to the manufacturer’s instructions, using random hexamers for priming. Samples were then sent for Sanger sequencing (Macrogen) with primers in Supplementary Table 6.Genome-wide CRISPR/Cas9 screenTwo replicates of A549-drW.1 cells transduced with the pLentiCRISPR-v2 Brunello library were seeded, maintaining a 300× coverage at MOI 0.3. Cells were selected with puromycin and cultured for 8–9 days before treatment. MG-132 was added for 4 h. Cells were stained with H-2Kb-bound to SIINFEKL antibody (BioLegend, clone 25-D1.16, 1:200 antibody) for 30 min at 4 °C (ref. 9). After two washes, cells were resuspended in BSA–PBS with 4,6-diamidino-2-phenylindole (DAPI) (Invitrogen). The top 1% GFP+APC− cells of the −Trp/IFNγ group were sorted. DNA was extracted using a Puregene Cell kit (QIAGEN) according to the manufacturer’s protocol. Sequencing libraries were prepared using the primers listed in Supplementary Tables 7 and 8. PCR replicates were pooled and cleaned using the ISOLATE II PCR with Gel cleanup kit (Bioline) followed by CleanNGS beads (CleanNA) before analyses on a 2100 Bioanalyzer using a 7500 chip (Agilent). Then they were quantified with a KAPA qPCR Library Quantification kit (Roche, KK4824) and sequenced with 75-bp single reads on a NextSeq550 using a High Output kit v.2.5 (75 Cycles) (Illumina) with the addition of 20% PhiX Sequencing Control V3 (Illumina, FC-110-3001) in the sequence pool.CRISPR/Cas9 screen analysisRead counts per sgRNA were obtained from raw read count files using the caRpools library in R65. To assess relative changes between starting and final cell population, MAGeCK software was employed using bulk sgRNAs for normalization21. Only the genes identified by at least three sgRNAs were plotted. log2 fold change and the raw P value (using permutation) of this gene in negative selection were plotted in the volcano plot.Western blottingCells were lysed in 1× Laemmli buffer and BCA-quantified. Proteins were separated by SDS–PAGE gels and transferred to 22 μm pore size nitrocellulose membranes (Santa Cruz). Staining was performed using ADAR1 (Santa Cruz, 1:200 dilution), Tubulin (Santa Cruz, 1:10,000 dilution), IDO1 (Cell Signaling Technology, 1:1,000 dilution), ABCE1 (Abcam, 1:1,000 dilution), RACK1 (Cell Signaling, 1:1,000 dilution), ZNF598 (Invitrogen, 1:1,000), WARS1 (Invitrogen, 1:1,000 dilution), antibodies. IRDye 680RD donkey anti-mouse (LI-COR, 1:10,000 dilution), IRDye 800CW goat anti-rat (LI-COR, 1:10,000 dilution) and IRDye 680RD donkey anti-rabbit (Li-COR, 1:10,000 dilution) were used as secondary antibodies. Visualization was performed by using an Odyssey infra-red scanning device (LI-COR).Proteomics sample preparationCell pellets were lysed in 1× S-Trap Lysis buffer (5% SDS and 50 mM TEAB pH 8.5), boiled and sonicated. Then, 30 µg of protein was digested using S-Trap micro-columns (ProtiFi) according to the manufacturer’s protocol. In brief, samples were reduced and alkylated using TCEP (5 mM, 15 min, 55 °C) and CAA (20 mM, 10 min). The samples were acidified and a methanol TEAB buffer was added, before loading on the S-Trap column. Trapped proteins were washed four times with the methanol TEAB buffer and then digested for 2 h at 47 °C using trypsin (Sigma-Aldrich, 3 μg per sample).For 2D LC–MS/MS, cell pellets were processed the same, after which 200 µg aliquots were digested with 20 µg trypsin using S-Trap Mini spin columns. The 100-µg digest was subjected to basic reversed-phase (HpH-RP) high-performance liquid chromatography for offline peptide fractionation as described previously10, with the flow rate set to 0.3 ml min−1 at the start and end of the method. Peptides were eluted at a constant flow of 100 μl min−1 in a 35-min gradient containing a nonlinear increase from 5.5–28% solvent B. Collected fractions were concatenated to a total of 12 fractions per sample.Proteomics mass spectrometryFor single-shot proteomes, data-independent acquisition (DIA) analyses were performed on the Orbitrap Astral mass spectrometer (Thermo Scientific) connected to a Vanquish Neo nano-LC system (Thermo Scientific) in a 30 samples-per day (30SPD) LC–MS method. The Vanquish Neo was operated in the trap-and-elute mode and peptides (~1 µg of digest) were loaded onto a Pepmap 100 C18 5 µm trap column (300 µm × 5 mm, Thermo Scientific), before separation on the analytical column (AUR3-25075C18-TS, 1.7 µm/75 µm × 25 cm, IonOpticks AU) mounted into an Easyspray ion source (Thermo Scientific). The column was heated at 50 °C with flow rate 0.5 µl min−1 at the start of the method. Solvent A was 0.1% formic acid/water and solvent B was 0.1% formic acid/80% acetonitrile. Peptides were eluted at 0.4 µl min−1 in a 35.7 min effective gradient, containing a non-linear increase from 8% to 45% solvent B, followed by a 0.4-min ramp to 99% solvent B and 3.4 min wash at 0.75 µl min−1 flow rate at the end. The column was equilibrated using the ‘fast equilibration’ script in combined control mode at a 1,450-bar limit. The Orbitrap Astral was run in DIA mode, with full MS scans being collected in the Orbitrap analyzer at 240,000 resolution at m/z 200 over a 380–980 m/z range. The default charge state was 2+, the normalized AGC target was set to 500% (equivalent to 5 × 106 charges) and the maximum injection time was 5 ms. For DIA MS2, a normalized HCD collision energy of 25% was applied to a 380–980 m/z precursor range using non-overlapping isolation windows of 2 Th, with window placement optimization. Scans were acquired in the Astral analyzer over a 100–1,000 m/z range, with the normalized AGC target set to 500% (equivalent to 5 × 104 charges) and a maximum injection time of 3 ms. DIA analyses of HpH-RP fractions (~1 µg peptides per fraction) were performed using the same instrumental setup and settings, but with a shorter (60SPD|70SPD) LC gradient.Generation of proteomics mutant databaseThe human proteome was obtained from UNIPROT66. Each substituted database was generated as described11. In brief, all occurrences of either tryptophan, phenylalanine or histidine were replaced by other amino acids in a separate database (FASTA file). As per the database, tryptophan is substituted with phenylalanine, phenylalanine with tyrosine, or histidine with glutamine.Proteomics database search and filteringRaw MS data were converted to the mzML format using ThermoRawFileParser (v.1.4.2) with centroiding enabled for MS2 spectra to ensure compatibility with downstream analysis. The converted mzML files were analysed using DIA-NN (v.1.8.1) for DIA peptide and protein identification and quantification. Search parameters were configured to enable trypsin digestion and allowing up to one missed cleavage. Searches were performed using precursor m/z values of 300–1,800 (charge states +1 to +4), fragment m/z values of 200–1,800 and peptide lengths of 7–50 amino acids. Methionine excision was enabled. Variable modifications included protein N-terminal acetylation and methionine oxidation, with a maximum of five variable modifications per peptide. Protein inference was performed using the relaxed setting, and identifications were filtered at a q-value threshold of 0.01 (1% FDR). Quantification was enabled.Extend of mistranslationOnly the W > F peptides identified in two out of two replicates (with Ln intensity > 9) and their respective WT matched peptides were selected for the analysis. The average intensity of the peptides that were found in both replicates was divided by the average intensity of the WT counterparts and plotted in log2. For the percentage of W > F substitutants, the median of the W > F/WT ratio was used.Immunoprecipitation of HLA-peptidesImmunopeptidomics was conducted as described in ref. 13. In brief, cell pellets or tumour samples were lysed11,13,67 and the W6/32 antibody bound to protein-A sepharose 4B beads was used for immunoaffinity purification.Immunopeptidomics mass spectrometryHLA-eluted peptides were analysed on an Orbitrap Exploris 480 Mass spectrometer connected to an Evosep One LC system (Evosep Biotechnology). Before LC separation with the Evosep One, peptides were reconstituted in 0.1% formic acid and loaded on Evotip Pure (Evosep) tips. Peptides were then eluted/separated using the ‘Extended Method’ (88-min gradient) on an EV1137 (Evosep) column with nanospray, Easyspray ion source (Thermo Scientific) and EV1086 (Evosep) emitter. On the Exploris 480, data-dependent acquisition was performed at resolution 60,000 with MS1 mass range 350–1,700 m/z, normalized AGC target was set to 100% and maximum injection time of 50 ms. Dynamic exclusion was set to 10 s. and MS2 spectra were acquired at 15,000 resolution. The top ten precursors per cycle were HCD fragmented with charge states 2–4, and the top five precursors per cycle were HCD-fragmentated if singly charged. The MS2 isolation window was 1.1 m/z, normalized collision energy was 30, normalized AGC target was 50% and the maximum injection time was 100 ms.Vulcano plot for differentially expressed peptides from immunopeptidomicsThe Protti package of R was used68 with the standard parameters. MaxLFQ Intensity for each peptide was used for the quantification. The standard parameters of completeness MAR = 0.7 and completeness MNAR = 0.1 were used for assigning missing values. Data imputation was performed using the Ludovic method.Generation of mutant databaseThe database used for peptide search was created similar to ref. 13. In brief, an amino acid window (±10 AAs) around all tryptophans (W) of proteins in the UniProt database UP000005640 was acquired. The W of interest in these peptides were converted to phenylalanine (F), and in the case of multiple Ws in the amino acid window, all the Ws were converted to F. These peptides were added to the UniProt database to generate the final database for peptide search.Database search and filteringThe native non-specific-HLA workflow (with preset parameters; peptide mass tolerance: ±20 ppm, calibrate mass: True, digestion: non-specified, peptide length: 7–35, mass range: 200–5,000 Da, missed cleavage: 2, fragment max charge: 2, fixed modifications: [57.02146 cysteine alkylation], variable modifications: [15.9949 methionine oxidation, 42.0106 N-terminal acetylation, -17.026500 N-terminal glutamine (Q) or cysteine (C) cyclization, 0.984016 N-terminal phenylalanine amidation]) in Fragpipe39 v.20.0 was used to identify peptides from the acquired mass spectra. Peptides identified by Fragpipe were mapped back to the proteome to confirm truly substituted peptides from canonical peptides. Peptides between 8 and 12 amino acids were retained for downstream quality checks such as peptide length distribution.Animal studiesThe in vivo experiments were approved by the Netherlands Cancer Institute Animal Experimental Committee. Experiments were performed under the approval AVD30100202519090 and DEC NKI (OZP ID 12051). Mice were bred and maintained in accordance with institutional, national and European guidelines for Animal Care and Use. The WP 39.1.11952. A549MART1FTSJ1-KO clone and corresponding rescue, were mixed with Cultrex (R&D systems, lot 1698229) in a ratio of 1:1 and injected into the right mammary gland #4 at 1 × 106 cells in 40 µl per NOD-SCID IL2R-null (The Jackson Laboratory) (NSG) mice, aged 6–8 weeks. The maximum permitted disease end points were not exceeded in any experiment. Animals were housed in a certified animal facility, providing 12-h light–dark cycles, regulated at 21 °C and 55% relative humidity. Mice were kept in individually ventilated cages, and food and water were provided ad libitum. Tumour size was monitored as described in ref. 13, by caliper measurement. When tumours on average reached a volume of 150–200 mm3, mice were randomly divided over groups receiving PBS or 10 × 106 DMF5 T cell injection via the tail vain, in a volume of 100 µl PBS. Starting from the day of T cell injection, mice daily received 100,000 units of IL-2 in 100 µl PBS per mouse for 3 consecutive days. Tumour material was collected 96 h after injection, snap frozen and stored at −80 °C. Processing of tumour material is conducted as described here in Methods for proteomics sample preparation.Measurement of tGFP and H-2Kb-bound SIINFEKLCells were detached and incubated with APC anti-mouse H-2Kb-bound to SIINFEKL antibody (BioLegend, clone 25-D1.16; 1:200 in PBS/0.1% BSA), plus when needed Live/Dead Fixable near-IR dead cell stain kit (1:1,000 dilution, Invitrogen) for 30 min on ice. Cells were rinsed two times with PBS–BSA 0.1% and resuspended in 0.1% BSA–PBS, with DAPI (Invitrogen) when required. Analyses was performed on the Attune NxT (Thermo Fisher Scientific) using Attune NxT software v.4.2 and the data were analysed using FlowJo v.10 software (FlowJo).OPP analysisOPP analysis was conducted as in ref. 10. In brief, after treatment, cells were treated with CHX (0.1 mg ml−1) for 5 min where necessary, and then supplemented with 10 μM OPP (Life Technologies) for 60 min at 37 °C. Cells were detached and fixed overnight in 70% ethanol at 4 °C, permeabilized with 0.1% Triton X-100 (Sigma), and blocked with 3% BSA (Sigma) in PBS. Subsequently, the click-it reaction was performed using click-it reagents and picolyl azide AF488 (all from Thermo Fisher Scientific). The cells were analysed on the Attune NxT using software described above.OT-1 T cell SIINFEKL recognition assaysOT-1 T cell SIINFEKL recognition assays were conducted as described in ref. 11 except that the co-cultured samples were then incubated for 4 h at 37 °C. Next, the cells were pelleted, blocked with PBS–BSA 0.1% and stained with Live/Dead Fixable near-IR dead cell stain kit (Invitrogen) and anti-mouse CD8-VioBlue antibodies (Miltenyi, 1:100 dilution). Afterwards, the cells were fixed and permeabilized (eBioscience Foxp3 Transcription Factor Staining Buffer Set (Invitrogen)). Cells were then stained with PE-conjugated anti-mouse TNF (Miltenyi, 1:100 dilution) and APC-conjugated anti-mouse IFNγ (Miltenyi 1:100 dilution). Cells were washed and analysed on a BD LSR Fortessa (BD Biosciences). Data were analysed using FlowJo v.10 software (FlowJo).T cell activationT cells were thawed and cultivated at a concentration of 1 × 106 cells per ml in RPMI 1640 medium supplemented with 10% human serum (One Lambda). T cell activation was performed as in ref. 13. In brief, for the cells loaded with the TMBIM6W>F peptide (GenScript Biotech), the peptide was loaded 24 h before the ending of the treatment at a concentration of 1 μM. Then, 100,000 cancer cells were co-cultured in RPMI medium for 16 h in a 96-well U-shaped plate with 50,000 TCRTMBIM6W>F.1 T cells. Cells were incubated with TCRmβ-PE (Miltenyi, 1:100 dilution), CD8-VioBlue (Miltenyi, 1:200 dilution), CD137-APC (Miltenyi, 1:100 dilution) and with 1:1,000 Live/Dead Fixable near-IR dead cell stain kit (Invitrogen) in PBS–BSA 0.1% on ice in the dark for 30 min. Cells were washed twice with PBS–BSA 0.1%. Samples were analysed on a BD LSR Fortessa (BD Biosciences). Data were analysed using FlowJo v.10 software (FlowJo).DNA extraction and TIDE analysisDNA extraction and TIDE analysis were performed as in ref. 13. In brief, cells were lysed at 55 °C in a 500-μl solution of 100 mM Tris, pH 8.5 mM EDTA, 0.2% SDS and 200 mM NaCl containing 2 µl of proteinase K (20 mg ml−1 stock solution from Sigma) overnight. After centrifugation, DNA was pelleted adding isopropanol. Then, PCR was performed using Phusion Polymerase (Thermo Fisher) following the manufacturer’s instruction. Primers used for amplifying the target genes are listed in Supplementary Table 9. Cells expressing a control sgRNA (sgNT1) were always used as reference. The PCR product was purified from agarose gel (QIAGEN) and sent for Sanger sequencing (Macrogen). The analysis was conducted via tide.nki.nl ref. 28.RNA-sequencingCells were collected in RLT buffer and snap frozen in liquid nitrogen. Total RNA was isolated using the RNeasy Mini kit (QIAGEN), including an on-column DNase digestion (QIAGEN), according to the manufacturer’s instructions. The quality and quantity of the total RNA were assessed by the 2100 Bioanalyzer using ‘Agilent RNA 6000 Nano’ (G2938-90034, Agilent Technologies). Total RNA samples having an RNA integrity number > 8 were subjected to library generation using the TruSeq stranded mRNA library preparation kit, according to the manufacturer’s instructions (document #1,000000040498 v.00, Illumina) with incorporation of xGen UDI-UMI adaptors (Integrated DNA Technologies). The stranded mRNA libraries were analysed on a 2100 Bioanalyzer instrument following the manufacturer’s protocol ‘Agilent DNA 7500 kit’ (Agilent Technologies), diluted to 10 nM, and pooled equimolar into multiplex sequencing pools for paired-end sequencing on the NovaSeq 6000 Illumina sequencing instrument. Paired-end sequencing was performed using 54 cycles for read 1, 19 cycles for read i7, 10 cycles for read i5, and 54 cycles for read 2, using the NovaSeq6000 Reagent kit v.1.5 (100 cycles) (Illumina). The fastq files of mRNA-seq data were aligned to the reference genome (GRCh37/hg19) using the STAR alignment tool69. Transcript abundances were quantified with Salmon70, and count data were variance-stabilized using the variance stabilizing transformation implemented in DESeq2.A-to-I editing enrichment calculationThe tool JACUSA2 (ref. 30) was used to detect site-specific editing events in RNA-seq data (CPTAC). The fastq files of mRNA-seq data were aligned to the reference genome (GRCh37/hg19) using the STAR69 alignment tool and then sorted and indexed using SAMtools. The call-2 feature of JACUSA2 was used to compares two different conditions. Each condition was first compared with the reference, and then with each other. To increase confidence in the output, several filtering steps were implemented, where each site must have a minimum of 15 reads covering that base in both the samples and the JACUSA2 score must be at least 2. Additionally, only the sites where there was an enrichment of ≥10% RNA editing in one condition versus the other were considered. The output was the sites at which an enrichment in RNA editing had been observed in one condition when compared with another.tRNA-seqTotal RNA was isolated using TRIzol reagent (Invitrogen), according to manufacturer’s instructions. Nano-tRNA-seq sample preparation, tRNA quality control, library preparation and Nanopore sequencing were conducted by IMMAGINA Biotechnology in Italy. Bioinformatics analysis was performed according to ref. 37. Alignment of sequences causes the experimental tRNA numbering to be in a + 1 position compared with the canonical numbering.rhWARS1 productionThe rhWARS1 was produced as described in ref. 11. In brief, cells were grown at 37 °C until OD600 of 0.7. Next, protein expression was induced by addition of 0.4 mM IPTG and the cells were grown overnight at 18 °C. After lysis, the recombinant rhWARS1 protein was purified using nickel beads, after which the protein was reconstituted in 25 mM Tris, pH 8.0, 200 mM NaCl and 1 mM TCEP.WARS1 aminoacylationTotal RNA was isolated using TRIzol reagent (Invitrogen), according to manufacturer’s instructions. Next the small RNA fraction was isolated using the RNA clean & concentrator kit (Zymo), according to their protocol. The resulting RNA fraction was deacylated by adding one volume of 1 M Tris-HCl at pH 9 and incubated for 3 h at room temperature. RNA was precipitated by adding one volume of 20% potassium acetate at pH 4.5 and 2.5 volumes of cold ethanol. RNA was pelleted and dissolved in H2O. tRNA was aminoacylated by incubating 10 OD ml−1 RNA with 2 μM WARS1, 3 mM ATP and 10 μM [3H]Trp (American Radiochemcals) or [14C]Phe (Revvity) in buffer: 50 mM HEPES, pH 7.5, 30 mM NH4Cl, 70 mM KCl, 20 mM MgCl2 and 2 mM dithiothreitol for 30 min at 37 °C. AA-tRNA was precipitated with trifluoroacetic acid on ice for 45 min, followed by filtration through nitrocellulose filters (Sartorius). A control with amino acids but no RNA was included. Filters were dissolved in 1 ml Soluene-350 (Revvity) by shaking for 10 min. For radioactivity counting, 10 ml Ultima Gold XR (Revvity) was added and disintegrations per minute were measured in a Liquid Scintillation counter. tRNA charging was calculated as pmol of amino acids using the specific activity of 3H and 14C. Values obtained with the control without tRNA were subtracted.WARS1 binding affinityThe binding affinity of WARS1 protein was determined using Microscale Thermophoresis (MST). rhWARS1 was first labelled with Dy547P1 dye (Dyomics). A fixed concentration of labelled WARS1 (20 nM) was mixed with a series of increasing concentrations (twofold dilution steps) of either amino acids or RNA (in the presence of 10 mM amino acid) in buffer containing 100 mM Tris pH 8, 10 mM MgCL2, 40 mM KCl, 1 mM TCEP, 3 mM ATP and 0.05% Tween20 in MiliQ. Deacetylated small RNA fraction samples were prepared as described for the WARS1 aminoacylation experiment.Measurements were performed on a Monolith NT.115 instrument (NanoTemper Technologies). Data were exported to GraphPad Prism (GraphPad Software) and fitted using a one-site binding model. From amino acid titrations, the dissociation constant (KD) was calculated, whereas from mRNA titrations the BC50 (the RNA concentration at which 50% of rhWARS1 is bound) was calculated. Differences in KD or BC50 values were assessed for statistical significance during fitting using Akaike’s Information Criterion. For representation, data were normalized using the fitted background and maximum binding (Bmax) values.MetabolomicsCell pellets were reconstituted in 50 μl of internal standard mix (10 μM l-Trp-d5 and 1 μM l-Kyn-d4 in water) by sonication for 10 min. Next, the samples were mixed with 30 μl of trifluoroacetic acid. After centrifugation, the samples were diluted tenfold in water and 20 μl was subjected to LC–MS/MS consisting of an UltiMate 3000 Autosampler and HPLC pump (Thermo Scientific) and API4000 MS/MS (Sciex). Calibration was carried out using a set of aqueous standards. Separation was performed on a Symmetry C18 column (2.1 × 150 mm, particle size 3.5 µM, Waters). Mobile phase A (0.1% formic acid in water) and B (methanol) were used in a 5 min gradient from 20 to 95% B, maintained for 3 min followed by re-equilibration at 20% B. Multiple reaction monitoring for acquisition were l-Trp (205.1/187.9 and 205.1/146.3), l-Kyn (209.3/192.1 and 209.3/146.1), l-Trp-d5 (210.3/192.0 and 210.0/192.0) and l-Kyn-d4 (213.2/196.1 and 213.2/150.2).FTSJ1 enrichment calculationThe gene expression datasets were obtained from the CPTAC. For each tumour type, the expression levels of FTSJ1 were extracted and categorized into two groups: tumour-adjacent tissues and tumour tissues, based on patient indices. The normalized intensity values were derived from the tandem mass tag proteomics dataset.Statistics and reproducibilityThe data in all figures represent independent cultures, as described. Flow cytometry gating strategies are depicted in Supplementary Figs. 8 and 9. Statistical analysis was performed as indicated in the figure legends. Exact P values are listed in Source Data. Data with sample size F and WT counterpart as in Fig. 1d and Extended Data Fig. 1e (n = 2). e, log2 intensity of the tryptic WT peptides corresponding to the tryptic W > F peptides represented in Fig. 1d. Each box represents the average of two replicates. A549 cells were treated as indicated for 48 hrs. f, Median fluorescence intensity (MFI) ± s.d. of out-of-frame (OOF)tGFP signal in A549 cells expressing the drW, drA or drF-vectors. Cells were treated as indicated for 48 hrs. Outlined dots (drW-vector) indicate previously shown data (Fig. 1g). g,h, MFI ± s.d. of H-2Kb-SIINFEKL (g) or OOFtGFP (h) signal in A549-drW cells. Ctr indicates the polyclonal population, #1-5 indicate clones. Treated as indicated for 48 hrs. Throughout the figure, *P