INTRODUCTION
Mutations in genes of the RAS-mitogen-activated protein kinase (MAPK) pathway occur in over 40% of all human cancers, leading to persistent pathway activation and fueling some of the most aggressive malignancies
1. Hyperactive RAS-MAPK signaling rewires core biosynthetic processes, including protein synthesis, RNA processing, and DNA replication
2. Despite decades of effort, the pursuit of anti-KRAS therapy has been hindered by the complex regulation of RAS signaling and the historically undruggable nature of RAS proteins
3. The development of covalent KRAS mutant-specific and pan-RAS inhibitors represents a major advance
1. However, adaptive resistance rapidly emerges through multiple routes, including feedback reactivation of upstream components such as receptor tyrosine kinases (RTKs)
4 and wild-type RAS, acquisition of secondary RAS mutations
5, and activation of compensatory survival signaling
6, 7. These multifaceted mechanisms explain why both single-agent strategies targeting individual RAS nodes and current clinical combinations of kinase inhibitors invariably encounter resistance, underscoring an urgent need to elucidate the fundamental mechanisms through which RAS sustains tumor growth and drug resistance.
Cancer proliferation requires elevated ribosome biogenesis and increased translational output
8. To meet this demand, tumors boost global protein synthesis through dysregulation of tRNAs
9, rRNAs
10, ribosomal proteins
11, 12, and translation factors
13. Beyond such global upregulation, recent studies are beginning to reveal a codon-specific layer of translational control that shapes cellular proteome composition in cancer. For example, increased expression of specific tRNAs can selectively enhance the translation of transcripts enriched for their cognate codons, thereby promoting breast cancer metastasis
9, 14. Similarly, dysregulated tRNA modifications preferentially enhance decoding of subsets of codons required for stress or oncogenic responses
15, 16. These findings highlight that codon content itself can selectively shape onco-translational programs. However, whether and if so, how oncogenic signaling pathways harness such codon-selective mechanisms remains largely unexplored, particularly in a systematic manner.
Rare codons, typically characterized by poor codon-anticodon pairing and decoded by low-abundance tRNAs, exhibit slow elongation kinetics
17, impose translation bottlenecks
18, and often serve as checkpoints for protein quality control
15, 19. Under stress, these bottlenecks are exacerbated by impaired tRNA charging and reduced elongation factor activity
20. Given that oncogenic signaling can profoundly reprogram protein synthesis even under nutrient-limited conditions, we hypothesize that rare codons may constitute a previously unrecognized regulatory interface between oncogenic signaling and translation.
Here, we uncover a mechanism by which RAS-MAPK signaling leverages the translation of the rare arginine codon, CGA, to promote tumor growth and therapeutic resistance. We demonstrate that a set of CGA-enriched genes, which encode ribosomal proteins, RNA splicing factors, and cell cycle regulators, is selectively upregulated by RAS signaling, thereby promoting tumorigenesis and drug resistance. Mechanistically, CGA decoding is accelerated by METTL13-mediated eEF1A K55 dimethylation. At the regulatory level, the stability and activity of METTL13 are enhanced via RSK-dependent phosphorylation, linking RAS signaling to codon-selective translation. Importantly, METTL13 inhibition disrupts this axis, suppresses RAS-driven tumor growth, and overcomes both intrinsic and acquired resistance to RAS pathway inhibitors.
RESULTS
RAS signaling specifically hyperactivates the translation of rare arginine codon CGA
Elevated ribosome biogenesis and protein synthesis are required for excessive proliferation of cancer cells and are theoretically coupled to the sustained growth of drug-tolerant persisters. These growth-promoting functions are largely regulated by RAS-MAPK and PI3K-mTOR (mammalian target of rapamycin) signaling pathways
21. To systematically investigate whether distinct pro-proliferation signals are associated with specific translational features, we individually introduced eight well-established oncogenic mutations, including
ALK-NPM1,
KRASG12D,
BRAFV600E,
MEK1K57N,
ERK2E322K,
EGFRT790M,
PTEN knockout (KO) and
cMYC overexpression (OE), into the normal epithelial cell line MCF10A to obtain genetically stable transformants (Supplementary Fig. S1a). After four weeks of culture, these engineered cells all converted into transformed cells exhibiting typical tumor-like characteristics such as altered morphology, changes in expression of molecular markers — including reduced E-cadherin (loss of epithelial adhesion), increased pERK (MAPK activation), and elevated p-P70 (mTOR activation) — dependent on individual oncogenic signaling (Supplementary Fig. S1b–d). Among these, we selected three representative lines (
KRASG12D,
MYC OE,
PTEN KO, three prevalent onco-drivers) along with an empty vector (EV) control for further studies. Polysome profiling revealed that all three transformed lines exhibited increased abundances of both total ribosomes and polysomes (actively translating ribosomes) compared to the EV control (Supplementary Fig. S1e), consistent with elevated translational activity as a hallmark of cancer.
We next performed ribosome profiling (Ribo-seq) to assess overall features of ribosome-bound mRNA fragments associated with individual onco-drivers (Fig. 1a; Supplementary Fig. S1f–j). Differential ribosome codon reading (Diricore) analysis focusing on the ribosomal A site which accepts the incoming aminoacyl-tRNA and decodes its codon, revealed a pronounced decrease in occupancy of the arginine CGA codon in MCF10A
KRAS G12D cells and of the proline CCC codon in MCF10
MYC OE and MCF10A
PTEN KO cells (Fig. 1b). This CGA-specific decoding enhancement was further recapitulated in cells harboring the
KRASG12V mutation (Supplementary Fig. S1k), suggesting a general translational feature of oncogenic KRAS signaling. Specifically, in EV cells, ribosome protected fragment (RPF) distribution over the CGA codon exhibited a strong peak at the ribosomal A site (–15), indicating pronounced ribosome stalling and inefficient decoding of this rare codon
22, 23. In contrast, specifically in MCF10A
KRAS G12D cells, ribosome stalling at CGA was markedly reduced, suggesting improved translation efficiency (TE), whereas no such significant reduction occurred in
PTEN KO or
MYC OE cells (Fig. 1c; Supplementary Fig. S1l,m).
Interestingly, although ~20% of human genes completely avoid CGA usage, around 512 genes are significantly enriched (> 3-fold compared to CGA average frequency at 0.62%; Fig. 1d) with this codon. Notably, more than half of these genes (261 genes) were enriched in four key biological processes: translation, RNA splicing, cell cycle regulation, and mitochondrial function (noting that mitochondrial protein translation was poorly detected by Ribo-seq) (Fig. 1e). From a compositional perspective, the core machinery constituting these processes is predominantly composed of CGA-enriched proteins, making it possible to coordinately regulate their collective functions through CGA translation (Fig. 1f). Proteins encoded by these genes, such as ribosomal proteins (RPs)
24, SRSF splicing factors
25 and CDK4
26, have been reported to be crucial for tumor growth and drug resistance (Fig. 1f; Supplementary Fig. S1n and Table). Analyses of TCGA datasets across colorectal, lung, and pancreatic cancers, which are frequently associated with RAS-MAPK activating mutations, revealed that these mutations were consistently associated with elevated CGA usage (with the common arginine codon CGG analyzed as a control; Supplementary Fig. S1o). Moreover, across all three cancer types, higher CGA usage correlated with advanced tumor grade and poorer patient survival (Supplementary Fig. S1p, q).
Consistently, CGA-enriched genes showed higher and smoother translational signals in MCF10AKRAS G12D cells (Supplementary Fig. S1r), further supporting improved translation. Integrated RNA-seq and Ribo-seq analyses revealed that those CGA-enriched gene groups showed significant elevation of TE in MCF10AKRAS G12D cells, but not in the PTEN KO or MYC OE cells (Fig. 1g; Supplementary Fig. S1s). Consistent with TE analysis, SILAC-based proteomics revealed a significant upregulation in the abundance of high-CGA proteins in MCF10AKRAS G12D cells compared to the EV controls (Fig. 1h; Supplementary Table). Together, these findings suggest that KRASG12D signaling alleviates the rate-limiting translational constraint imposed by the rare arginine codon CGA.
To further validate our findings, we developed a codon-specific TE reporter system using short-lived luciferase to monitor CGA-dependent translation in real time (Fig. 1i; Supplementary Fig. S1t, u). Application of this reporter across multiple cancer cell lines and under inhibitor treatments, we consistently observed that activation of the RAS-MAPK pathway enhanced CGA TE. Importantly, the abundance of CGA-decoding tRNAUCG remained constant, indicating the enhanced CGA TE was not due to changes in tRNA levels. Notably, a comparison of CGA TE among MCF10A-transformed lines with the eight onco-drivers revealed that only those able to activate RAS-MAPK signaling showed a significant elevation in CGA TE (Fig. 1j; Supplementary Fig. S1v). Furthermore, treatment with various RAS-MAPK pathway inhibitors markedly reduced CGA TE in KRASG12C-driven H358 lung cancer cells (Fig. 1k; Supplementary Fig. S1w). These results reveal a selective regulatory mechanism whereby RAS signaling promotes translation of the CGA-enriched genes, which encode core components of translation, splicing and cell cycle progression, thereby supporting sustained tumor growth.
Enhanced CGA decoding promotes RAS-driven tumorigenesis and drug resistance
We reasoned that elevated CGA TE supports RAS-driven growth by boosting the production of CGA-enriched proteins. To test this, we developed a CGA index, defined as the ratio of protein abundance of three representative high-CGA proteins (RPL15, SRSF1, CDK4; CGA usage ratio at 4.0-, 5.7-, 4.8-fold of average, respectively) to three low-CGA proteins (ACTN, TUBB, VCL; CGA usage ratio at 0.35-, 1.0-, 1.5-fold of average, respectively) (Fig. 2a; see the Materials and Methods section for details). While the three high-CGA proteins are well documented to promote tumorigenesis across various malignancies, the three low-CGA proteins are widely used as housekeeping controls for data normalization across groups
24-26. To validate whether the CGA index accurately reports CGA translation capacity, we established patient-derived organoids (PDOs) from 13 colorectal cancer (CRC) cases. We found that the CGA index, as measured by immunoblot analysis (Supplementary Fig. S2a, b) correlated with both RAS mutant status (Fig. 2b; Supplementary Table) and tumor recurrence status (Fig. 2c). To directly compare CGA decoding capacity across PDO models, we introduced a dual-fluorescence-based codon translation reporter into PDOs with high versus low CGA indices (
e.g., high: CRCO-03, CRCO-22; low: CRCO-02, CRCO-06). CGA TE was quantified by normalizing the EBFP signal (CGA reporter) to the EGFP signal (internal control) (Fig. 2d; Supplementary Fig. S2b). This analysis confirmed significantly elevated CGA TE in high-CGA-index PDOs, validating the CGA index as a faithful surrogate for actual CGA codon translation activity (Fig. 2e, f; Supplementary Fig. S2c, d).
Further analysis of clinical proteomics data from CRC patients revealed that individuals carrying RAS/RAF mutations exhibited a higher CGA index score (Fig. 2g). More importantly, a high CGA index score was strongly associated with non-response to first-line therapies and worse prognosis (Fig. 2h, i, datasets from Li
et al.
27 and Martinez-Val
et al.28). Consistently, expanding the index to include the top 50 high-CGA proteins independently recapitulated these trends (Supplementary Fig. S2e–g).
Furthermore, in
KRASG12D CRCO-03, we observed notable heterogeneity in CGA TE, with higher efficiency detected in highly proliferative cells (Ki67
+) (Fig. 2e, k; Supplementary Fig. S2h, i). This finding aligns with a recent report revealing upregulated CGA translation during M phase
29. Treatment with a KRAS
G12D inhibitor suppressed overall CGA TE in PDOs but did not affect CGG TE as a control (Fig. 2j, k; Supplementary Fig. S2h, i). Notably, a small subset of cells escaped this suppression, maintaining both high CGA translation capacity and proliferative signals (Fig. 2j, k). These cells are considered a potential source of drug resistance, suggesting that CGA TE may play a critical role in mediating therapeutic resilience.
We then generated drug-resistant cells from the inhibitor-sensitive lung cancer cell line H358 through long-term KRAS
G12C inhibitor AMG510 treatment
6 (Fig. 2l). Using SILAC-based proteomics, we observed that the expression of high-CGA proteins increased as resistance developed (Fig. 2m; Supplementary Fig. S2j and Table). To further validate this finding, we overexpressed high-CGA protein sets (
e.g., CDK2/4, RPL15/26/38 with CGA codon optimized to CGG to ensure efficient expression) or tRNA
UCG (to enhance global CGA TE) in naive H358 cells. Both manipulations effectively conferred inhibitor resistance, supporting the functional role of CGA-enriched proteins in promoting drug tolerance (Fig. 2n). Together, these results position CGA-specific translation as a critical node in RAS-driven tumorigenesis and as a key contributor to drug resistance.
RAS signaling enhances CGA translation via the METTL13-eEF1A K55me2 axis
We next investigated how RAS-MAPK signaling upregulates the TE of the CGA codon. Although global translation enhancement was also observed in the PTEN KO and MYC OE models, the increase in CGA TE was specific to RAS activation (Fig. 1b; Supplementary Fig. S1e). Moreover, the abundance of the CGA-decoding tRNAUCG remained unchanged in KRASG12D cells, ruling out tRNA availability as an explanation for this selective effect (Supplementary Fig. S1v). We therefore hypothesized that RAS-MAPK signaling might enhance ribosome recognition or loading of the tRNAUCG. This hypothesis was supported by polysome-associated tRNA profiling, which revealed a significant increase in translating tRNAUCG (polysome-bound) in RAS-activated cells (Supplementary Fig. S3a).
Given that tRNA delivery to the ribosomal A-site is mediated primarily by eEF1A, and that eEF1A is subject to extensive methylation
13, 30-33 (N-ter me3, K36me3, K55me2, K79me3, K165me2; Fig. 3a), we asked whether eEF1A expression or methylation status contributes to this process. First, although knockdown of total eEF1A decreased global translation, it had no selective effect on CGA TE in MCF10A
KRAS G12D cells (Fig. 3b; Supplementary Fig. S3b, c), indicating that its overall abundance is not the regulatory factor. Using a candidate siRNA-based approach, we next perturbed the major methylations of eEF1A to monitor CGA TE alteration with the codon-specific reporter. This screen identified METTL13 as specifically essential for the
KRASG12D-induced enhancement of CGA TE; METTL13 catalyzes both N-terminus (N-ter me3) and K55 (K55me2) methylation of eEF1A (Fig. 3b; Supplementary Fig. S3c–e).
In parallel with the CGA reporter-based TE candidate screen, we performed a cell growth-based CRISPR-Cas9 screen using a customized epigenetics-focused gRNA library, targeting 36 ribosomal modifiers along with approximately 600 chromatin modifiers and regulators (Supplementary Fig. S3f). This screen allowed us to identify key growth dependencies on translational or transcriptional regulators, specifically for MCF10AKRAS G12D cells without affecting MCF10AEV, MCF10AMYC OE and MCF10APTEN KO cells. Strikingly, METTL13 again emerged as the top candidate essential for KRASG12D-driven tumor growth (Fig. 3c). Depletion of METTL13 had no measurable effect on the growth of EV, PTEN KO and MYC OE MCF10A cells, highlighting a unique dependency of METTL13 in KRAS-mutant cells (Supplementary Fig. S3g, h). No chromatin modifiers or regulators were identified as significant hits, suggesting that targeting translational regulation may represent an effective therapeutic strategy for RAS-MAPK-driven tumors.
Next, we performed METTL13 knockdown followed by Ribo-seq in MCF10AKRAS G12D cells to assess how METTL13 affects genome-wide CGA decoding. Consistent with the reporter assays, METTL13 loss led to increased ribosome occupancy at the CGA codons, indicating attenuated TE of CGA-enriched genes, exemplified by ribosomal proteins, SRSFs, CDK4 and CSNK2A, and resulted in reduced abundance of their corresponding proteins (Fig. 3d–f; Supplementary Fig. S3i, j and Table). We further evaluated the effect of METTL13 on the abundance of tRNAs bound to 80S monosomes and polysomes in SW837 cells, a colorectal cancer line harboring KRASG12C mutation. METTL13 depletion selectively decreased the abundance of tRNAUCG associated with polysomes, aligning with the observed reduction in CGA codon TE (Fig. 3g; Supplementary Fig. S3k–m). Accordingly, OE of tRNAUCG in METTL13 KO cells restored the expression of high-CGA proteins such as RPL15, SRSF1, and CDK4 (Fig. 3h).
Since METTL13 contains a dual catalytic methyltransferase domain, we sought to determine which eEF1A methylation event, N-ter me3, K55me2, or both, is required for enhanced CGA translation under KRAS activation. We found that both the eEF1A K55R substitution and the corresponding G58R catalytic dead mutation
13 in the K55-specific methyltransferase domain of METTL13 abolished its ability to promote CGA TE in SW837 cells, while the eEF1A N-terminal His-tagging (blocking N-ter me3) and the corresponding METTL13-E545A catalytic mutation showed no effect (Fig. 3i; Supplementary Fig. S3n). Published structural analyses of the ribosome translation process indicate that a hydrogen bond between eEF1A K55 and E43 residues may be compressed during codon recognition, suggesting that methylation at K55 could modulate eEF1A conformation during the loading of tRNA
UCG onto the ribosome. In contrast, N-ter me3 resides within a flexible region that is remote from the core functional domains of eEF1A, including the ribosome-binding, tRNA-binding, and GTPase regions, suggesting that it may be involved in other processes (Supplementary Fig. S3o; EMD 29758–29760
34).
Consistent with previous reports that eEF1A K55 methylation regulates global translation output
13, we observed that METTL13 disruption consistently reduced global translation, as evidenced by diminished translational RPFs in MCF10A
KRAS G12D cells and lower polysome abundance in SW837 cells (Supplementary Fig. S3i, k). Importantly, our findings extend beyond this global effect by revealing an additional codon-based preference, suggesting a more sophisticated regulatory mechanism that fine-tunes the translation of a specific subset of mRNAs critical for tumorigenesis.
Together, these findings provide compelling evidence that the RAS-MAPK pathway enhances the translation of CGA codon through a specific METTL13-eEF1A K55me2 regulatory axis.
RAS-driven tumorigenesis requires RSK-mediated phosphorylation of METTL13
The direct molecular link connecting RAS-MAPK signaling to METTL13 remains elusive. By immunoblotting, we noticed that the protein levels of METTL13 and eEF1A K55me2 were significantly elevated in MCF10AKRAS G12D cells and other RAS-MAPK-activated MCF10A cells (Fig. 4a; Supplementary Fig. S4a), indicating that METTL13 may function as a downstream effector of the pathway. In contrast, METTL13’s mRNA levels remained comparable across all four MCF10A-derived cell models (Supplementary Fig. S4b). Further analysis of TCGA datasets from lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LSCC), and CRC again demonstrated a positive correlation between RAS pathway activation and METTL13 at the protein level, but not at the mRNA level (Supplementary Fig. S4c).
Employing endogenous immunoprecipitation from the respective MCF10A cell lines followed by Phos-Tag SDS-PAGE analysis, we detected stronger phosphorylation signals for METTL13 protein in MCF10A
KRAS G12D cells compared to the other models (Fig. 4a). Mass spectrometry analysis identified a phosphorylation site at serine 267 (S267) within the linker region between the two methyltransferase domains of METTL13 (Supplementary Fig. S4d). Further analysis of published phosphoproteomic datasets confirmed that S267 phosphorylation is readily detectable with its levels positively correlated with RAS-MAPK activation
35, 36 (Supplementary Fig. S4e). Structural modeling with AlphaFold3 indicated that S267 is located at the periphery of a zipper-like structure within the METTL13 linker, where it forms a hydrogen bond with D318 (Fig. 4b). Of note, the phosphorylation at S267 is expected to disrupt this hydrogen bond, triggering conformational changes in METTL13 (Supplementary Fig. S4f).
To identify the kinase responsible for S267 phosphorylation, we performed an unbiased screen of 72 kinase inhibitors using an in-cell ELISA, which nominated the RAS-MAPK-RSK signaling axis (Fig. 4c). The consistent suppression of pS267 by inhibitors targeting multiple nodes of this pathway — ranging from EGFR and RAF to MEK and ERK — suggested that the terminal effector kinase likely belongs to the MAPKAPK family. Supporting this, sequence analysis confirmed a conserved RSK consensus motif (R-X-R-X-X-S) flanking S267 (Fig. 4d). Based on these convergent findings, we narrowed our focus to all eight candidate MAPKAPKs for functional validation. Through immunoprecipitation analysis, we found that RSK1 and RSK2 exhibited robust cellular interaction and catalytic activity toward METTL13S267. RSK3 and RSK4 also stimulated pS267, despite showing weaker interaction with METTL13 under our assay conditions. In contrast, other candidates, such as MAPKAPK2/3/5 and MNK1, failed to either bind or phosphorylate METTL13 (Supplementary Fig. S4g, h). Finally, in vitro kinase assays confirmed that RSK1 directly phosphorylated METTL13 at S267 (Supplementary Fig. S4i, j). Collectively, these results identify RSK1/2 as the primary kinases mediating METTL13 phosphorylation downstream of oncogenic RAS signaling.
Importantly, this phosphorylation event significantly enhanced METTL13 protein stability. We found that RSK OE upregulated METTL13 protein levels; conversely, in cycloheximide (CHX) chase assays, both mutation of the S267 phosphorylation site and inhibition of RSK reduced the METTL13 protein half-life by approximately three-fold (Fig. 4e; Supplementary Fig. S4k). Furthermore, this modification stimulated METTL13’s catalytic activity toward eEF1A K55 by at least five-fold in vitro (Fig. 4f; Supplementary Fig. S4l). Consistently, METTL13 purified from MCF10AKRAS G12D cells exhibited a stronger in vitro methyltransferase activity than that from MYC OE or PTEN KO MCF10A cells (Supplementary Fig. S4m). Indeed, S267 phosphorylation of METTL13 is required for its effective methylation towards eEF1A in A549 (KRASG13S) and A375 (BRAFV600E) cells (Supplementary Fig. S4n). Consistent with the CRISPR-Cas9 screen (Fig. 3c), METTL13 loss significantly inhibited the proliferation of A549 and A375 cells, and the S267A mutant could not rescue this effect (Supplementary Fig. S4o).
The hydrodynamic tail vein injection (HTVi) based spontaneous liver tumor model was chosen because it allows rapid tumor initiation from a defined onco-driver, and the liver has negligible endogenous METTL13 expression
37 (Fig. 4g; Supplementary Fig. S4p). These features make it an ideal system for assessing the tumorigenic role of METTL13
in vivo across different oncogenic signaling backgrounds. We again found that METTL13 specifically promoted tumorigenesis driven by activation of the RAS pathway (
e.g.,
KRASG12D,
BRAFV600E, and
EGFRT790M), but had no obvious effect on tumors driven by
MYC OE or
YAPS5A (Fig. 4h; Supplementary Fig. S4q, r). Moreover, the S267 phosphorylation of METTL13 is essential for its pro-tumorigenic activity, to a degree comparable to the requirement for its eEF1A K55 methyltransferase activity (Fig. 4i; Supplementary Fig. S4s, t).
To further evaluate the biological and clinical relevance of this phospho-methyl regulatory axis, we generated a S267 phosphorylation-specific antibody (Supplementary Fig. S4u) and measured METTL13 pS267 levels in transformed MCF10A cell lines and human tumor specimens, including samples from LUAD and CRC patients. Consistently, transformed cells and tumor tissues with activated RAS signaling exhibited upregulation of METTL13 protein, S267 phosphorylation, and eEF1A K55me2 levels (Fig. 4j; Supplementary Fig. S4v). Notably, S267 phosphorylation signal showed a significant association with RAS-MAPK pathway activity and metastasis status, supporting its potential value as a candidate marker for RAS-MAPK activation and prognostic stratification (Fig. 4k; Supplementary Fig. S4w, x).
Together, these results define a novel ribosome-associated phospho-methyl regulatory axis downstream of RAS signaling. Our findings show that RAS-MAPK-RSK-mediated phosphorylation of METTL13 at S267 stimulates its activity toward eEF1A K55me2, subsequently enhancing the translation of CGA-enriched genes and promoting RAS-MAPK driven tumorigenesis (Fig. 4l).
METTL13-CGA axis sustains proliferative persisters in RAS-MAPK-driven PDOs
Having established a role of the RAS-METTL13-eEF1A K55me2 axis in controlling CGA TE, we sought to define its functional role in tumorigenesis and its therapeutic relevance. IF analysis of PDOs revealed a heterogeneous distribution of METTL13 pS267 and eEF1A K55me2, with the strongest signals coinciding with high CGA TE and localizing to highly proliferative (Ki67+) cell populations (Supplementary Fig. S5a, b). This spatial correlation linked the METTL13-eEF1A K55me2 axis activation directly to the proliferative and drug-resistant cells identified in our earlier findings (Fig. 2j, k).
To functionally test this link, we knocked out METTL13 using lentiviral CRISPR-Cas9 in PDOs from three recurrent CRC patients (Fig. 2c). METTL13 ablation significantly reduced the proportion of proliferating cells (marked by pRB and Ki67) and suppressed organoid growth (Fig. 5a, b; Supplementary Fig. S5c). This effect was specific to RAS-driven tumors, as METTL13 KO had no impact on PDOs lacking RAS pathway activation (Supplementary Fig. S5d–f). Importantly, although single-agent RAS-MAPK pathway inhibition (MRTX1133, KRASG12D inhibitor; Encorafenib, BRAFV600E inhibitor) left behind a population of persister cells, co-targeting METTL13 eradicated these residual cell populations (Fig. 5c, d).
We hypothesized that this effect stems from impaired translation of CGA-enriched cell-cycle regulators. To test this, we co-cultured control and METTL13 KO PDOs, followed by IF to examine CDK expression. As expected, METTL13 KO specifically reduced CDK2 and CDK4 (high-CGA) protein levels, but not CDK1 (low-CGA) (Supplementary Fig. S5g–i). While CDK6 protein was also reduced, this was likely due predominantly to a decrease in its mRNA (Supplementary Fig. S5h, i). Consequently, METTL13 loss decreased proliferation (pRB, Ki67), and increased apoptosis (cleaved caspase-3) (Supplementary Fig. S5h). Furthermore, METTL13 KO also reduced levels of other high-CGA proteins, such as resistance-associated proteins RPL15 and SRSF1 (Fig. 5e).
Targeting METTL13 overcomes therapy resistance in RAS-driven tumors
The high mortality of advanced CRC is largely attributed to acquired drug resistance and tumor recurrence. All 13 CRC patients included in this study had received first-line treatment (chemotherapy, with or without VEGFR inhibitors) and subsequently underwent surgical resection (Fig. 2). Three patients (CRCO-03, -05, -22) developed recurrence within one year. We therefore evaluated the therapeutic potential of targeting the METTL13-CGA axis in the corresponding PDOs, particularly in combination with RAS-MAPK pathway inhibitors.
We first compared the efficacy of METTL13 loss to that of existing RAS-MAPK inhibitors. In monotherapy settings, METTL13 loss exerted stronger tumor-suppressive effects than RAS-MAPK inhibitors alone (Fig. 5f). Clinically, to counteract drug resistance to RAS-MAPK inhibitors, combination regimens with CDK, EGFR, or MEK inhibitors are commonly recommended
38. We compared METTL13 targeting-based combinations with these standard treatment strategies. Strikingly, combining RAS-MAPK inhibitors with
METTL13 KO achieved therapeutic responses comparable to or exceeding those of standard combination regimens in all three recurrent PDOs (Fig. 5g). Similar results were observed in three additional PDOs (CRCO-01, -07, and -10) that exhibited high CGA index but no documented recurrence (Supplementary Fig. S5j).
Notably, the combination of METTL13 loss with RAS inhibitors produced a uniformly robust response across all six PDOs, overcoming the patient-specific variability observed with current clinical combination therapies (e.g., with CDK, EGFR, or MEK inhibitors) (Fig. 5g; Supplementary Fig. S5j). For instance, while CDK inhibitor-based combinations were effective in CRCO-3, -5, and -10, they were markedly less effective in the other three models compared to the METTL13-based strategy.
To examine whether targeting METTL13 can enhance the clinical efficacy of KRAS
G12D inhibitors, we collected 10
KRASG12D-mutant PDOs (including 3 CRC and 7 PDAC) to test whether
METTL13 KO could sensitize them to MRTX1133. Viability analyses showed that CRC PDOs were largely resistant (IC
50 > 1 µM), whereas PDAC PDOs were generally more sensitive (IC
50 < 0.2 µM), in line with previous reports that CRC frequently displays intrinsic resistance to KRAS inhibition
4,39,40. Importantly, METTL13 loss markedly increased sensitivity to MRTX1133 across all 10 CRC and PDAC PDOs, generally by 3 to 6-fold (Fig. 5h; Supplementary Fig. S5k). These results underscore the robustness of METTL13 inhibition in suppressing drug resistance compared with current standard therapeutic approaches, likely by limiting the translation of multiple CGA-enriched oncogenic proteins.
Finally, we tested whether METTL13 suppression could block both intrinsic and acquired drug resistance in vivo. In xenografts derived from intrinsically resistant SW837 cells, intratumoral injection of METTL13-targeting siRNA significantly inhibited tumor growth and enhanced the efficacy of AMG510 (Fig. 5i, j). To model acquired resistance, we established AMG510-resistant H358 tumors and Trametinib (MEKi)-resistant A375 tumors (Fig. 5k). In both models, tumors resumed growth after ~3 weeks of treatment, confirming resistance acquisition. Subsequent METTL13 knockdown in vivo effectively reduced the expression of high-CGA proteins (e.g., RPL15 and CDK4) and markedly suppressed tumor progression (Fig. 5j; Supplementary Fig. S5l), demonstrating that METTL13 plays a pivotal role in both intrinsic and acquired resistance states of RAS-MAPK tumors.
In summary, targeting the METTL13-CGA axis eliminates a core translational dependency of RAS-driven persisting cells and effectively circumvents intrinsic and acquired resistance.
DISCUSSION
Our study identified enhanced translation of the rare arginine codon CGA as a distinctive signature of RAS signaling-driven tumors. Mechanistically, the RAS-MAPK effector kinase RSK phosphorylates and stabilizes the methyltransferase METTL13 at S267, which in turn stimulates METTL13 activity on the dimethylation of eEF1A at K55 to promote efficient CGA decoding. Targeting METTL13 disrupts this regulatory axis, inhibits CGA translation, and suppresses tumor growth and progression driven by RAS signaling. Notably, this CGA translational activation is not seen in MYC- or PTEN loss-dependent tumor cells, underscoring its specificity to the RAS-MAPK pathway (Supplementary Fig. S6).
Previous studies established that METTL13-mediated eEF1A K55 dimethylation promotes global translational output and supports tumorigenesis, with this modification reported to enhance eEF1A GTPase activity
13. Here, our findings reveal a distinct model in which oncogenic RAS-MAPK signaling activates METTL13-eEF1A K55me2 to drive codon-biased translation of CGA-enriched transcripts, thereby linking signal-responsive ribosome regulation to selective translational output in cancer.
A codon-specific translation program sustains oncogenic KRAS
CGA is among the rarest codons in humans and represents a translation bottleneck with slow decoding kinetics, which is a strong signal for the recruitment of degradation factors to the ribosome and for triggering mRNA decay
41, 42. We propose that these inherent translational characteristics may render CGA particularly sensitive to modulation by the METTL13-eEF1A axis, especially in the context of oncogenic KRAS signaling. Our results show that CGA codons are enriched in a particular set of genes essential for RAS signaling-driven proliferation. This genomic codon composition enables cancer cells to upregulate multiple crucial genes by enhancing the translation of a single codon, a highly efficient strategy for sustaining biomass accumulation in the nutrient-limited tumor environment. Notably, many of the CGA-enriched genes identified in our study, such as ribosomal proteins (
e.g., RPL15
24), splicing factors (
e.g., SRSF1
25), and cell cycle regulators (
e.g., CDK4
26), have been individually implicated in tumor growth, metastasis, and therapeutic resistance. Despite their functional importance, these proteins have traditionally been considered “undruggable” due to their essential roles in normal cellular physiology.
CGA translation as a potential mechanism of acquired resistance to RAS-MAPK inhibitors
Current clinical combination strategies pairing a direct KRAS inhibitor with immune checkpoint blockade or conventional cytotoxic drugs show limited durability
43-45, underscoring that our understanding of KRAS signaling regulation remains incomplete and that the crucial mechanisms underlying resistance remain to be discovered. In our study, we observed that when treated with KRAS inhibitors, most cells in patient-derived organoids ceased proliferating and showed decreased CGA translation, while a small fraction of cells maintaining high CGA translational activity continued to proliferate. These findings suggest that sustained CGA translation supports residual KRAS signaling and may constitute a mechanism of acquired resistance to KRAS inhibition.
Of note, clinical resistance to RAS-MAPK pathway inhibitors, particularly in advanced-stage malignancies, is multifactorial and can be driven by clonal diversity, increased mutational burden, and activation of bypass signaling pathways. Our findings suggest that the METTL13-CGA axis represents a previously unrecognized layer of translational adaptation that may coexist with these genetic and signaling-based resistance mechanisms to support tumor growth.
Targeting translation of multiple effectors through a ribosomal handle
Having identified METTL13 as an upstream regulator that controls multiple effectors
via CGA translation, we evaluated its therapeutic potential in RAS-MAPK-driven cancers. Dual inhibition of METTL13 and RAS pathway targets exhibited greater efficacy than current clinical combinations targeting the RAS pathway together with CDK, EGFR or MEK inhibitors
46. This supports a strategy in which targeting an upstream regulatory “handle” that coordinates multiple downstream effectors may be more effective than blocking individual signaling components. Such a “one-handle, multiple-effector” strategy holds the potential to offer new therapeutic avenues, particularly for cancers driven by undruggable targets.
METTL13, initially named FEAT (Faint Expression in normal tissues, Aberrant OE in Tumors), is minimally expressed outside of the testis
37. Its function as an eEF1A K55 methyltransferase has been confirmed in previous independent studies
13, 31, although the mechanistic interpretations regarding how eEF1A methylation affects translation output appear to differ. Despite these uncertainties, encouragingly,
METTL13 KO mice showed no overt abnormalities over a two-year observation period, except for a mild compromise in fertility (unpublished data), suggesting a favorable safety profile. Nevertheless, future studies should carefully evaluate whether METTL13 inhibition affects normal tissues or organs with high translational demand, particularly under stress or regenerative conditions.
Codon-biased translation as a signaling-linked phenomenon
Our findings extend the concept of codon-biased translation implicated in stress responses
47-49, ageing
50, 51 and cancer
16, 52, 53. Two principal mechanisms have been proposed: altered tRNA abundance, which enhances translation of mRNAs enriched for cognate codons
9, 54, 55, and tRNA modifications that modulate tRNA stability or decoding efficiency at the wobble position
16, 56, 57. Unlike these earlier studies, the CGA codon we identified is selectively driven by the activation of RAS signaling, showing that codon-biased translation can operate at the level of a single codon linked to a given pathological setting. Together, these findings highlight the potential of targeting codon-specific translation to selectively inhibit aberrant protein synthesis while minimizing the toxic effects observed in global translation inhibition trials
58.
Despite these advances, many questions remain unanswered. For example, are codon-specific programs activated by specific upstream signals, and how are these programs dynamically regulated? What are the molecular determinants defining each codon program? Are a few key effectors sufficient to mediate cellular adaptation, or do they act within larger regulatory networks? Beyond the CGA codon “awakened” by RAS-MAPK signaling, we also identified additional codon-selective activation by other oncogenic pathways, such as the proline codon CCC in
MYC OE and
PTEN KO cells. Interestingly, a recent study found that MYC-amplified neuroblastomas exhibit increased proline uptake compared to their low-MYC counterparts
57. By exploiting this metabolic dependency, a proline-restricted diet was shown to trigger codon-specific proteome rewiring and promote neuroblastoma differentiation. These observations suggest that codon-specific translation programs may represent a general principle across cancers driven by distinct oncogenic pathways and warrant further exploration.
Ribosome-based epigenetic regulation
The prevailing view holds that rRNAs mediate nearly all ribosomal functions, whereas ribosomal proteins are largely regarded as static structural contributors to protein synthesis
59, 60. However, accumulating evidence, including our findings, supports the notion that post-translational modifications of the protein components of the translation machinery can modulate ribosomal output with functional consequences. Methylation of ribosomal proteins and translation factors, such as RPL40 by SMYD5 and eEF1A by METTL13, has been shown to enhance translational capacity and promote tumorigenesis
11-13. These observations highlight the importance of protein modifications within the translation apparatus in fine-tuning how mRNA information is interpreted, extending beyond the well-established regulatory roles of tRNAs and rRNAs.
The interplay between mRNA features and the ribosomal regulatory state must be precisely coordinated to allow cells to adapt to environmental fluctuations and long-term evolutionary pressures. Based on our findings in this study, we propose a previously overlooked ribosome-centric epigenetic layer as a key determinant of cellular function at the codon level. This layer acts analogously (codon vs transcription factor binding motifs) to chromatin modifications in transcription, providing a dynamic yet heritable mechanism that integrates signaling inputs to modulate translational outcomes. Deciphering this ribosome-based regulatory logic will not only reveal how distinct codon programs are decoded across physiological and pathological contexts but also offer a conceptual and practical framework for precise translation reprogramming in disease treatment.
Limitations of the study
Several limitations should be acknowledged. First, we currently lack direct structural and biochemical evidence linking the methylation status of eEF1A to CGA translation efficacy. Future studies using structural approaches and reconstituted in vitro translation systems, incorporating tRNAUCG, unmethylated or K55-dimethylated eEF1A, and ribosomes, will be required to directly define the molecular mechanism by which eEF1A K55 dimethylation may influence CGA codon decoding. Second, our clinical analysis is retrospective in nature, utilizing established patient tissue cohorts. Prospective validation in large patient cohorts will be required to determine whether the correlation between METTL13 pS267 and poor prognosis is robust across multiple cancer types, and whether METTL13 pS267 has clinical utility as a biomarker. Third, the in vivo validation relied primarily on RNAi-mediated METTL13 ablation due to the absence of selective METTL13 inhibitors. The development of potent and specific inhibitors will be critical for defining the enzymatic requirement of METTL13 and assessing its therapeutic potential in preclinical models. Finally, although our data support a codon-level mechanism of translation reprogramming, the contribution of additional factors or processes influenced by the METTL13-eEF1A K55me2 axis cannot be excluded.
MATERIALS AND METHODS
Cell culture
HEK293T, SW837 and A375 cells were cultured in high-glucose Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% Fetal Bovine Serum (FBS) and 1% Penicillin-Streptomycin (P/S) at 37 °C in 5% CO2. H358 and A549 cells were cultured in RPMI 1640 medium supplemented with 10 % FBS and 1% P/S at 37 °C in 5% CO2. MCF10A cells were cultured in DMEM/F12 medium supplemented with 20% horse serum, 1% P/S, 10 μg/mL insulin, 2 μg/mL hydrocortisone and 100 ng/mL cholera toxin at 37 °C in 5% CO2. Transformed MCF10A cells were cultured in DMEM like HEK293T. Cell lines used in this study were regularly tested for mycoplasma.
Generation of stable cell lines
Stable cell lines were generated by lentiviral transduction. HEK293T cells were co-transfected with the lentiviral expression vector and packaging plasmids psPAX2 and pMD2.G using Lipofectamine 2000 (Invitrogen) per the manufacturer’s instructions. Viral supernatants were collected 48 and 72 h post-transfection, filtered through a 0.45-μm syringe filter, and used to infect target cells in the presence of 8 μg/mL polybrene (Sigma). After 24 h, medium was replaced, and cells were selected with puromycin (2 μg/mL) or blasticidin (5 μg/mL) for 2–4 days until a stable population emerged. Knockdown or OE efficiency was validated by qRT-PCR or immunoblotting. For transformed MCF10A cell lines, cells were maintained in epithelial medium for 3–4 weeks post-selection until most cells underwent oncogene-induced apoptosis; remaining cells exhibiting morphological transformation and altered growth rates were considered transformed. Then changed the medium to regular DMEM.
For the generation of AMG510-Resistant H358 Cells. The protocol for generating AMG510-resistant H358 cell lines was adapted from Xue
et al.6. In brief, parental H358 cells were seeded at 50% confluence in 10-cm dishes and initially treated with 2 nM AMG510. The drug-containing medium was refreshed every three days. The AMG510 concentration was increased by 2 nM every six days, unless significant cell death was observed, in which case the concentration was maintained until viability recovered. After 120 days of selection, the concentration increment was raised to 5 nM every six days. The selection process continued until day 180 to establish stably resistant populations.
PDO culture
Human colorectal and pancreatic tissues were obtained and processed in accordance with the guidelines of Zhongshan Hospital and relevant Chinese regulations (B2024-577). Fresh tumor tissues were minced and digested to single cells, then resuspended in extracellular matrix (ECM). Tumor cells were cultured in organoid basal medium supplemented with kit additives (bioGenous) at 37 °C and 5% CO2, forming organoids within one week. Medium was refreshed every 3 days, and organoids were passaged every 10–14 days via mechanical and enzymatic dissociation. Organoid identity was confirmed by histology, immunostaining, and gDNA sequencing.
Genetic modification of PDOs
Early-passage (P0–P1) organoids were dissociated into single cells and incubated with lentiviral supernatant containing 8 μg/mL polybrene and 10% FBS for 3 h at 37 °C with shaking at 650 rpm. Cells were washed twice with basal medium and re-embedded in ECM. Puromycin selection (2 μg/mL, 2 days) was applied after organoid re-formation. Knockdown or OE efficiency was assessed by immunostaining.
Plasmid and cloning
Gene inserts or shRNA/sgRNA sequences were cloned into lentiviral vectors using restriction enzyme-based cloning or Gibson assembly. Plasmids were amplified in E. coli (Top10), purified with EndoFree MaxiPrep kits, and verified by Sanger sequencing. All plasmid sequences have been deposited in the SnapGene repository.
Immunoblotting
Cells, organoids, or tissues were lysed in RIPA buffer containing protease and phosphatase inhibitors. Protein concentration was determined by BCA assay and Ponceau S staining. Equal amounts of protein (20–30 μg) were separated by SDS-PAGE and transferred to nitrocellulose membranes. Membranes were blocked with 5% non-fat milk in TBST, incubated with primary antibodies overnight at 4 °C, followed by HRP-conjugated secondary antibodies. Signals were detected using enhanced chemiluminescence (ECL) and imaged on a ChemiDoc system (Bio-Rad).
Codon translation reporter assay
Dual-luciferase reporter constructs contained 10× repeats of the test codon sequence between Gaussian-luciferase (secreted, control) and Nano-luciferase (with a PEST degradation signal peptide, half-life < 30 min). For the assay, cells were transfected for 24–48 h, and medium was replaced 30 min before measurement. Accumulated secreted Gaussian-luciferase activity in the medium was measured (Beyotime kit) as an internal control. Cells were lysed, and Nano-luciferase activity was quantified (Promega kit). Codon TE was calculated as the Nano-to-Gaussian luciferase signal ratio.
Dual-fluorescent protein reporter plasmids were constructed by inserting 10× repeats of the codon of interest between coding sequences for EGFP (control) and EBFP fluorescent proteins. The reporter was transfected into PDO by lentiviral system, and selected by puromycin. The PDOs were fixed with cold 4% PFA and 0.25% glutaraldehyde for 30 min at 4 °C. Then they were washed with cold PBS twice to remove the ECM. After diluting with PBS to 5–10 organoids per 100 μL, they were aliquote into 96-well plates, centrifuged, carefully removed the PBS, and then a microplate reader was used to measure fluorescence.
siRNA library screen
Cells were reverse-transfected in 48-well plates with an siRNA library (three siRNAs per gene) using RNAiMAX (Lipofectamine). To ensure sustained knockdown, transfections were repeated every 3 days at 30% confluence for a total of 3 rounds. After 6 days, quadruplicate wells were transferred to 96-well plates and co-transfected medium refreshed.
CRISPR-cas9 screen
A custom lentiviral sgRNA library targeting 900 epigenetic regulators (6 sgRNAs/gene, 5,400 total) was constructed in the lentiCRISPR-v2 backbone (PicoHelix). Lentivirus was produced from 293T cells. MCF10A cells were infected at a MOI of 0.2–0.3 and selected with puromycin (2 μg/mL, 3 days). A population of 5 × 10
6 cells was maintained to preserve library representation. Genomic DNA was extracted on day 3 (baseline) and day 20. sgRNA sequences were amplified by two-step PCR and sequenced on an Illumina platform. Enriched or depleted sgRNAs were identified by comparing read counts between time points. The screen was performed as adapted from Shalem
et al.61.
SILAC assay
Cells were cultured in SILAC media lacking either arginine or lysine, supplemented with 10% dialyzed fetal bovine serum and 1% P/S For heavy and light labeling medium, the following stable isotope-labeled amino acids were added: 13C615N2 L-lysine (K8), 13C615N4 L-arginine (R10) for heavy labeling; 4,4,5,5-D4 L-lysine (K4), 13C6 L-arginine (R6) for medium labeling; and unlabeled L-lysine and L-arginine for light labeling, each at a final concentration of 0.1 mg/mL. All amino acids were used at 0.1 mg/mL. Cells were cultured for 7–8 passages to achieve complete isotope incorporation, which was confirmed by MS.
For protein extraction, cells were washed three times with PBS and lysed in 4 M urea buffer (100 μL per 1 × 106 cells) followed by sonication for 1 min. Protein concentration was determined using a BCA assay, and equal amounts from each SILAC condition were combined at a 1:1:1 ratio.
For LC-MS/MS analysis, 30 μg of mixed protein was reduced with 4 mM DTT (30 min), alkylated with 8 mM iodoacetamide, and precipitated with methanol. Proteins were digested overnight at 37 °C with MS-grade trypsin (Promega). Peptides were desalted and analyzed by LC-MS/MS for identification and quantification.
Polysome profiling and tRNA qPCR
Approximately 2 × 107 cells were incubated with 400 μM CHX (MCE) for 8 min and then pelleted. Pellets were washed twice in PBS with 400 μM CHX and immediately lysed in 200 µL cold lysis buffer (100 mM KCl, 10 mM MgCl2, 50 mM Tris-Cl, pH 7.4, 0.5% NP-40) for 10 min on ice. The lysates were clarified by centrifugation at 700× g for 5 min at 4 °C to discard the cell nucleus and 12,000× g for 10 min at 4 °C to discard mitochondria and debris. Lysates were then loaded onto 10–50% sucrose gradients and ultra-centrifuged in an SW41 Ti swinging-bucket rotor (331362, Beckman) at 36,000 rpm for 2 h at 4 °C. All gradients were fractionated using a Biocomp gradient fractionator while monitoring A260.
300 µL gradient fractions were subjected to RNA extraction by 1.2 mL Trizol (ThermoFisher Scientific), and recovered by small RNA kit (Zymo). tRNA were denatured at 75 °C for 10 min, and were subjected to reverse transcription reactions (TOYOBO). qPCR primers and reaction were referenced by Tsuyoshi Udagawa
et al.62.
Ribosome profiling and data analysis
The construction of RPF libraries was performed as previously described
11. About 1 × 10
7 cells were digested with RNase I (EN0601, ThermoFisher Scientific) and ribosomes were collected using MicroSpin S-400 HR columns (27514001, Cytiva). Then RPFs were extracted using a small RNA kit (R1070, Zymo Research) and purified by urea-PAGE. RNA fragments of 25–40 nt were excised and recovered. After end repair, libraries were generated from the RPFs using the Small RNA Library Prep Kit (NR811, Vazyme) and sequenced on a Illumina NovaSeq 6000 platform (conducted by Nanjing Gaoxin Precision Medicine Technology Co., Ltd.)
After sequencing, raw reads were filtered by Trim Galore (v0.6.10) with parameters “---fastqc --paired -a AGATCGGAAGAGCACACGTCTGAACTCCAGTCAC -a2 GATCGTCGGACTGTAGAACTCTGAACGTGTAGATCTCGGTGGTCGCCGTAATT” and the reads mapping to human rRNA and tRNA were removed using Bowtie2 (v2.2.5). Then the cleaned reads were aligned to the human reference genome (hg38) using STAR (v2.7.11b) to report the best alignments with parameters “--outSAMmultNmax 1”. The indexes for BAM files were generated by samtools (v1.9) index and visualization was achieved using IGV. The cleaned reads were also aligned to the human reference transcriptome (hg38) using Bowtie2 (v2.2.5) to report the best alignments with the local model. PCR duplicates were removed using Picard MarkDuplicates (v2.26.2). The metagene analysis was performed using RiboMiner (v0.2). Ribo-seq quality control analyses followed the standards established by Liu
et al.61 and Mao
et al.63, 64.
To analyze codon-specific ribosome stalling, we performed Diricore analysis on the Ribo-seq data. To ensure analytical robustness and sufficient read coverage, only genes with a Transcripts Per Million (TPM) value greater than 10 were included in the analysis. For these selected genes, the A-site codon occupancy was calculated based on the distribution of uniquely mapped RPFs. The A-site offset was set to –15 nucleotides, a parameter determined by the metagene periodicity analysis of the P-site relative to the start codons. The differential occupancy of each of the 64 codons at the A-site was then compared between different experimental conditions to identify specific codon-level translational changes.
RNA sequencing data analysis：Raw reads were filtered by Trim Galore (v0.6.10) with parameters “--illumina --fastqc --paired” and aligned to the human (hg38) reference genome using hisat2 (v2.2.1). The counts matrix was generated using subread (v2.0.2) featureCounts with default parameters. The used gene annotation gtf file was downloaded from the Ensemble website (
ftp.ensembl.org/pub/release-112/gtf/homo_sapiens/). The differentially expressed genes were identified using the R package DEseq2 (v1.48.1) with a Padj-value cutoff < 0.05 and |fold change| cutoff >= 2. The gene expression counts heatmap was generated by the R package pHeatmap (v1.0.13) (
github.com/raivokolde/pheatmap).
HTVi
Plasmid DNA for HTVi was prepared to ensure endotoxin-free quality and concentrations above 1 mg/mL. A total of 10 µg of pT3 (onco-driver plasmids) and pPB (PB transposase) plasmids were mixed at a mass ratio of 25:1 and diluted with saline to a final volume of 1 mL. Three-week-old C57BL/6J mice were used for the injection. The plasmid mixture (1 mL) was injected via the tail vein within 5 s to create the required high-pressure condition for efficient delivery into the liver. After one week, liver tissues were harvested from the mice for qPCR analysis to assess the expression of onco-drivers and the efficiency of METTL13 knockdown. For tumor progression assessment, mice were monitored for 4–6 weeks for signs of abdominal swelling and weight loss, indicative of tumor development. At the study endpoint, mice were euthanized, and their liver tumors were collected for imaging, weight measurement, and immunohistochemical analysis.
Protein purification
HEK293E cells were cultured to a density of 1–1.5 × 106 cells/mL and transfected using PEI-pro method with DNA-PEI complexes (50–100 μg plasmid DNA at PEI:DNA ratio of 1.5–2:1). Cells were harvested 48 h post-transfection, washed with PBS, and lysed in lysis buffer supplemented with protease and phosphatase inhibitors. After sonication and centrifugation, FLAG-tagged proteins were captured using anti-FLAG M2 affinity gel-beads with 4 h incubation at 4 °C. Beads were washed three times with binding buffer, and proteins were eluted with 4 mg/mL FLAG-peptide in elution buffer. Protein concentration was determined by Bradford assay.
In vitro methyltransferase and kinase assay
Methyltransferase assay: recombinant FLAG-METTL13 (wild-type and S267A mutant) was purified from HEK293E cells. Highly phosphorylated METTL13 was obtained by co-transfecting with HA-RSK1 (1:5 ratio) and treating with 20 ng/mL PMA to activate ERK-RSK signaling. eEF1A substrate was purified from E. coli with an N-terminal His-tag to block native N-terminal trimethylation. Reactions contained 50 nM METTL13 and 50–8,000 nM eEF1A in 20 μL reaction buffer (100 μM SAM, 0.5 mM TCEP, 50 mM Tris, pH 7.5, 0.05% Triton X-100) incubated at room temperature for 120 min. Reactions were terminated with 1% TFA, and methylation was quantified using Mtase-Glo Kit (Promega) by measuring SAH-derived luminescence. Kinetic parameters were determined using GraphPad Prism.
Kinase Assay: Recombinant FLAG-RSK was purified from PMA-stimulated HEK293E cells. METTL13 (wild-type and S267A) and RPS6 substrates were purified from E. coli to avoid pre-phosphorylation. Reactions contained 4,000 nM METTL13 and 50 nM RSK with 50–1000 μM ATP in 20 μL kinase buffer (0.5 mM TCEP, 50 mM Tris pH 7.5, 0.05% Triton X-100, 2 mM MgCl2) incubated at room temperature for 10 min. Reactions were terminated with 1% TFA, and ADP production was measured using ADP-Glo Kit (Promega). Kinetic constants were calculated using GraphPad Prism.
Long-term proliferation assay
Cells were seeded at 5,000 cells/well in 24-well plates (H358). Drug treatments were initiated the following day. siRNA transfections were performed every 3 days with medium replacement (containing inhibitors) 24 h post-transfection. After 20 days, cells were washed with ice-cold PBS, fixed with 4% paraformaldehyde, and stained with 0.1% crystal violet.
Xenograft assay
Cell line-derived xenografts: four-week-old female athymic nu/nu mice (BALB/c background) were housed under specific pathogen-free conditions. H358 (2 × 106), SW837 (1 × 106), and A375 (1 × 106) cells were suspended in PBS and injected into mammary fat pads. Treatment began when tumors reached ~100 mm3, with randomization to treatment groups (n ≥ 5 mice/group). In vivo siRNA (2 nmol in PBS) was administered intratumorally every 5 days. Experiments were conducted in a blinded manner.
All animal procedures were approved by the Fudan University Institutional Committee and conducted in accordance with Shanghai Medical Experimental Animal Care Commission guidelines.
Organoid assays
Proliferation Assay: Organoids were plated in 96-well black/clear bottom plates (20–40 ducts/well in 15 μL ECM, n = 3). Viability was measured using Living Cell-Fluo Kit (biogenous) via fluorescent dye oxidation-reduction. Baseline measurements (Day 0) were taken 24 h post-seeding, with subsequent measurements every 3 days. Between measurements, organoids were washed twice with basal medium (30 min each) before medium replacement.
IC50 Determination: Organoids were plated in 384-well white plates (~20 ducts/well in 10 μL ECM, n = 4). Inhibitor treatments were applied 48 h post-seeding, with viability assessed after 72 h using ATP-Glo Kit (biogenous). Data were analyzed using Microsoft Excel and GraphPad Prism.
IF and Imaging: Organoids were fixed in 4% PFA/0.25% glutaraldehyde (1 h, 4 °C), washed with cold PBS, and permeabilized/blocked in PBS containing 0.5% Triton X-100 and 10% BSA (2 h, RT). Primary antibodies were fluorescently labeled using FlexAble Kit (Proteintech) and incubated overnight at 4 °C. After washing, about 100 organoids were resuspended in 50 μL PBS and dropped on a slide. Aspirated the PBS and pipetted 20 μL antifade buffer to cover the organoids and then placed the coverslip. Confocal microscopy was performed by SP8 LSCM (Leica), and image analyzed by LAS X software.
Calculation of the CGA index
The CGA Index was calculated to quantify the relative abundance of CGA-enriched proteins in biological samples, based on normalized proteomic data. The calculation was performed as follows:
First, the raw protein abundance values were normalized across all samples using Z-score normalization to ensure comparability. The CGA Index was then computed as the ratio of the average normalized abundance of high-CGA proteins to the average normalized abundance of low-CGA proteins, according to the formula:
n = number of high-CGA proteins
High CGAnorm,i = Z-score normalized abundance of the i-th high-CGA protein
m = number of low-CGA proteins (CGA enrichment ratio < 1.5, common control protein)
Low CGAnorm,j = Z-score normalized abundance of the j-th low-CGA protein
Proteins were classified as high-CGA or low-CGA based on their CGA codon enrichment ratio, which was calculated as the frequency of CGA codons relative to the genomic average.
Ethics approval and consent to participate
This study was conducted in strict accordance with the Declaration of Helsinki and was approved by the Ethics Committee of Zhongshan Hospital, Fudan University (Approval No. B2024-577). Written informed consent was obtained from all individual participants (or their legal guardians) prior to their inclusion in the study.
DATA AVAILABILITY
High-throughput sequencing data were deposited to the GEO with accession numbers GSE310997 and GSE319633.
The Author(s) 2026. Published by Higher Education Press. This is an Open Access article distributed under the terms of the CC BY license (https://creativecommons.org/licenses/by/4.0/).
