Alternate RNA decoding results in stable and abundant proteins in mammals

Data availability

The analyses described in this work use sequencing and proteomic data from previously published datasets deposited in public data repositories. CPTAC raw MS proteomic data were downloaded from the CPTAC Data Portal with the following PDC study IDs: PDC000120 (BRCA), PDC000127 (CCRCC), PDC000234 (LSCC), PDC000153 (LUAD), PDC000270 (PDAC) and PDC000125 (UCEC). CPTAC genomic and transcriptomic sequencing reads were accessed through the GDC Data Portal with the following dbGaP study accessions: phs000892 (CPTAC-2) and phs001287 (CPTAC-3). Whole exome sequencing data from the following PDC studies were analysed: PDC000127, PDC000446, PDC000204, PDC000221, PDC000234, PDC000153, PDC000489, PDC000270, PDC000393, PDC000125, PDC000439, and PDC000464. Access to controlled data was granted after application to NCBI (project no. 24007: Investigation of Mistranslation Rates in Cancer). RNA-seq data for healthy human tissues17 were downloaded from ArrayExpress with the identifier E-MTAB-2836, and corresponding raw MS data were downloaded from PRIDE with project accession PXD010154. Mouse transcriptome sequence data were downloaded from ArrayExpress under the identifier E-MTAB-10276. The mouse MS proteomic data were downloaded from PRIDE with the dataset identifier PXD030983. Primary cell MS data were downloaded from PRIDE with the following accession numbers: PXD008511 (B cells), PXD008512 (hepatocytes), PXD008513 (monocytes) and PXD008515 (natural killer cells). Immunoprecipitation–MS proteomic data were downloaded from MassIVE with accession MSV000088555. Supporting information, data and documentation are available at decode.slavovlab.net.

Code availability

Software, data-analysis pipelines and other supporting documentation are available at decode.slavovlab.net. The code for reproducing all the analyses and figures presented is freely available at GitHub (github.com/SlavovLab/decode).

References

Cantwell-Dorris, E. R., O’Leary, J. J. & Sheils, O. M. BRAFV600E: implications for carcinogenesis and molecular therapy. Mol. Cancer Ther. 10, 385–394 (2011).

Article

CAS

PubMed

Google Scholar

Hart, J. R. et al. The butterfly effect in cancer: a single base mutation can remodel the cell. Proc. Natl Acad. Sci. USA 112, 1131–1136 (2015).

Article

ADS

CAS

PubMed

PubMed Central

Google Scholar

Wright, A. & Vissel, B. The essential role of AMPA receptor GluR2 subunit RNA editing in the normal and diseased brain. Front. Mol. Neurosci. 5, 34 (2012).

Article

CAS

PubMed

PubMed Central

Google Scholar

Parker, J. & Friesen, J. D. “Two out of three” codon reading leading to mistranslation in vivo. Mol. Gen. Genet. 177, 439–445 (1980).

Article

CAS

PubMed

Google Scholar

Savitski, M. M., Nielsen, M. L. & Zubarev, R. A. ModifiComb, a new proteomic tool for mapping substoichiometric post-translational modifications, finding novel types of modifications, and fingerprinting complex protein mixtures. Mol. Cell. Proteomics 5, 935–948 (2006).

Article

CAS

PubMed

Google Scholar

Cox, J. & Mann, M. MaxQuant enables high peptide identification rates, individualized p.p.b.-range mass accuracies and proteome-wide protein quantification. Nat. Biotechnol. 26, 1367–1372 (2008).

Article

CAS

PubMed

Google Scholar

Wilhelm, M. et al. Deep learning boosts sensitivity of mass spectrometry-based immunopeptidomics. Nat. Commun. 12, 3346 (2021).

Article

ADS

CAS

PubMed

PubMed Central

Google Scholar

Picciani, M. et al. Oktoberfest: open-source spectral library generation and rescoring pipeline based on Prosit. Proteomics 24, e2300112 (2024).

Article

PubMed

Google Scholar

Yang, K. L. et al. MSBooster: improving peptide identification rates using deep learning-based features. Nat. Commun. 14, 4539 (2023).

Article

ADS

CAS

PubMed

PubMed Central

Google Scholar

Leduc, A. & Slavov, N. Impact of protein degradation and cell growth on mammalian proteomes. Preprint at bioRxiv https://doi.org/10.1101/2025.02.10.637566 (2025).

Clark, D. J. et al. Integrated proteogenomic characterization of clear cell renal cell carcinoma. Cell 179, 964–983 (2019).

Article

CAS

PubMed

PubMed Central

Google Scholar

Krug, K. et al. Proteogenomic landscape of breast cancer tumorigenesis and targeted therapy. Cell 183, 1436–1456 (2020).

Article

CAS

PubMed

PubMed Central

Google Scholar

Gillette, M. A. et al. Proteogenomic characterization reveals therapeutic vulnerabilities in lung adenocarcinoma. Cell 182, 200–225 (2020).

Article

CAS

PubMed

PubMed Central

Google Scholar

Dou, Y. et al. Proteogenomic characterization of endometrial carcinoma. Cell 180, 729–748 (2020).

Article

CAS

PubMed

PubMed Central

Google Scholar

Cao, L. et al. Proteogenomic characterization of pancreatic ductal adenocarcinoma. Cell 184, 5031–5052 (2021).

Article

CAS

PubMed

PubMed Central

Google Scholar

Satpathy, S. et al. A proteogenomic portrait of lung squamous cell carcinoma. Cell 184, 4348–4371 (2021).

Article

CAS

PubMed

PubMed Central

Google Scholar

Wang, D. et al. A deep proteome and transcriptome abundance atlas of 29 healthy human tissues. Mol. Syst. Biol. 15, e8503 (2019).

Article

PubMed

PubMed Central

Google Scholar

Batut, B. et al. Community-driven data analysis training for biology. Cell Syst. 6, 752–758 (2018).

Article

CAS

PubMed

PubMed Central

Google Scholar

Mordret, E. et al. Systematic detection of amino acid substitutions in proteomes reveals mechanistic basis of ribosome errors and selection for translation fidelity. Mol. Cell 75, 427–441 (2019).

Article

CAS

PubMed

Google Scholar

Ma, C. et al. Improved peptide retention time prediction in liquid chromatography through deep learning. Anal. Chem. 90, 10881–10888 (2018).

Article

ADS

CAS

PubMed

Google Scholar

Wen, B., Wang, X. & Zhang, B. PepQuery enables fast, accurate, and convenient proteomic validation of novel genomic alterations. Genome Res. 29, 485–493 (2019).

Article

CAS

PubMed

PubMed Central

Google Scholar

Mohler, K. & Ibba, M. Translational fidelity and mistranslation in the cellular response to stress. Nat. Microbiol. 2, 17117 (2017).

Article

CAS

PubMed

PubMed Central

Google Scholar

Liigand, P., Kaupmees, K. & Kruve, A. Influence of the amino acid composition on the ionization efficiencies of small peptides. J. Mass Spectrom. 54, 481–487 (2019).

Article

ADS

CAS

PubMed

Google Scholar

Serrano, G., Guruceaga, E. & Segura, V. DeepMSPeptide: peptide detectability prediction using deep learning. Bioinformatics 36, 1279–1280 (2020).

Article

CAS

PubMed

Google Scholar

Gessulat, S. et al. Prosit: proteome-wide prediction of peptide tandem mass spectra by deep learning. Nat. Methods 16, 509–518 (2019).

Article

CAS

PubMed

Google Scholar

Gabriel, W. et al. Prosit-TMT: deep learning boosts identification of TMT-labeled peptides. Anal. Chem. 94, 7181–7190 (2022).

Article

ADS

CAS

PubMed

Google Scholar

Wiśniewski, J. R., Hein, M. Y., Cox, J. & Mann, M. A “proteomic ruler” for protein copy number and concentration estimation without spike-in standards. Mol. Cell. Proteomics 13, 1535–9484 (2014).

Article

Google Scholar

Wu, Q. et al. Translation affects mRNA stability in a codon-dependent manner in human cells. eLife 8, e45396 (2019).

Article

PubMed

PubMed Central

Google Scholar

Drummond, D. A. & Wilke, C. O. Mistranslation-induced protein misfolding as a dominant constraint on coding-sequence evolution. Cell 134, 341–352 (2008).

Article

CAS

PubMed

PubMed Central

Google Scholar

Quax, T. E., Claassens, N. J., Söll, D. & van der Oost, J. Codon bias as a means to fine-tune gene expression. Mol. Cell 59, 149–161 (2015).

Article

CAS

PubMed

PubMed Central

Google Scholar

McCormick, C. A. et al. mRNA psi profiling using nanopore DRS reveals cell type-specific pseudouridylation. Preprint at bioRxiv https://doi.org/10.1101/2024.05.08.593203 (2024).

Mathieson, T. et al. Systematic analysis of protein turnover in primary cells. Nat. Commun. 9, 689 (2018).

Article

ADS

PubMed

PubMed Central

Google Scholar

Kong, A. T., Leprevost, F. V., Avtonomov, D. M., Mellacheruvu, D. & Nesvizhskii, A. I. MSFragger: ultrafast and comprehensive peptide identification in mass spectrometry-based proteomics. Nat. Methods 14, 513–520 (2017).

Article

CAS

PubMed

PubMed Central

Google Scholar

Chen, S. et al. A genomic mutational constraint map using variation in 76,156 human genomes. Nature 625, 92–100 (2024).

Article

ADS

CAS

PubMed

Google Scholar

Giansanti, P. et al. Mass spectrometry-based draft of the mouse proteome. Nat. Methods 19, 803–811 (2022).

Article

CAS

PubMed

PubMed Central

Google Scholar

Lawrence, M. S. et al. Mutational heterogeneity in cancer and the search for new cancer-associated genes. Nature 499, 214–218 (2013).

Article

ADS

CAS

PubMed

PubMed Central

Google Scholar

Specht, H. et al. PSMtags improve peptide sequencing and throughput in sensitive proteomics. Preprint at bioRxiv https://doi.org/10.1101/2025.05.22.655509 (2025).

Slavov, N. Single-cell proteomic technologies: tools in the quest for principles. Annu. Rev. Biophys. 55, 253–275 (2026).

Leduc, A., Khoury, L., Cantlon, J., Khan, S. & Slavov, N. Massively parallel sample preparation for multiplexed single-cell proteomics using nPOP. Nat. Protoc. 19, 3750–3776 (2024).

Article

CAS

PubMed

PubMed Central

Google Scholar

Huffman, R. G. et al. Prioritized mass spectrometry increases the depth, sensitivity and data completeness of single-cell proteomics. Nat. Methods 20, 714–722 (2023).

Article

CAS

PubMed

PubMed Central

Google Scholar

Sun, L. et al. Evolutionary gain of alanine mischarging to noncognate tRNAs with a G4: U69 base pair. J. Am. Chem. Soc. 138, 12948–12955 (2016).

Article

ADS

CAS

PubMed

PubMed Central

Google Scholar

Netzer, N. et al. Innate immune and chemically triggered oxidative stress modifies translational fidelity. Nature 462, 522–526 (2009).

Article

ADS

CAS

PubMed

PubMed Central

Google Scholar

Danecek, P. et al. Twelve years of SAMtools and BCFtools. GigaScience https://doi.org/10.1093/gigascience/giab008 (2021).

Chen, S., Zhou, Y., Chen, Y. & Gu, J. fastp: an ultra-fast all-in-one FASTQ preprocessor. Bioinformatics 34, i884–i890 (2018).

Article

PubMed

PubMed Central

Google Scholar

Kim, D., Paggi, J. M., Park, C., Bennett, C. & Salzberg, S. L. Graph-based genome alignment and genotyping with HISAT2 and HISAT-genotype. Nat. Biotechnol. 37, 907–915 (2019).

Article

CAS

PubMed

PubMed Central

Google Scholar

Pertea, M. et al. StringTie enables improved reconstruction of a transcriptome from RNA-seq reads. Nat. Biotechnol. 33, 290–295 (2015).

Article

CAS

PubMed

PubMed Central

Google Scholar

Pertea, G. & Pertea, M. GFF Utilities: GffRead and GffCompare. F10000Research 9, 304 (2020).

Article

Google Scholar

Garrison, E. & Marth, G. Haplotype-based variant detection from short-read sequencing. Preprint at https://doi.org/10.48550/arXiv.1207.3907 (2012).

Wang, X. & Zhang, B. customProDB: an R package to generate customized protein databases from RNA-seq data for proteomics search. Bioinformatics 29, 3235–3237 (2013).

Article

CAS

PubMed

PubMed Central

Google Scholar

Lautenbacher, L. et al. Koina: Democratizing machine learning for proteomics research. Nat. Commun. 16, 9933 (2025).

Huber, F. et al. matchms—processing and similarity evaluation of mass spectrometry data. J. Open Source Softw. 5, 2411 (2020).

Article

ADS

Google Scholar

Wan, K. X., Vidavsky, I. & Gross, M. L. Comparing similar spectra: From similarity index to spectral contrast angle. J. Am. Soc. Mass Spectrom. 13, 85–88 (2002).

Halloran, J. T. & Rocke, D. M. Matter of time: faster percolator analysis via efficient SVM learning for large-scale proteomics. J. Proteome Res. 17, 1978–1982 (2018).

Article

CAS

PubMed

PubMed Central

Google Scholar

Huttlin, E. L. et al. The BioPlex network: a systematic exploration of the human interactome. Cell 162, 425–440 (2015).

Article

CAS

PubMed

PubMed Central

Google Scholar

Marino, A. et al. Aging and diet alter the protein ubiquitylation landscape in the mouse brain. Nat. Commun. 16, 5266 (2025).

Article

ADS

CAS

PubMed

PubMed Central

Google Scholar

Li, J. et al. Proteome-wide mapping of short-lived proteins in human cells. Mol. Cell 81, 4722–4735 (2021).

Article

CAS

PubMed

PubMed Central

Google Scholar

Nettling, M. et al. DiffLogo: a comparative visualization of sequence motifs. BMC Bioinformatics 16, 387 (2015).

Article

PubMed

PubMed Central

Google Scholar

Behle, A. et al. Manipulation of topoisomerase expression inhibits cell division but not growth and reveals a distinctive promoter structure in Synechocystis. Nucleic Acids Res. 50, 12790–12808 (2022).

Article

CAS

PubMed

PubMed Central

Google Scholar

Erdős, G., Pajkos, M. & Dosztányi, Z. IUPred3: prediction of protein disorder enhanced with unambiguous experimental annotation and visualization of evolutionary conservation. Nucleic Acids Res. 49, W297–W303 (2021).

Article

PubMed

PubMed Central

Google Scholar

Eddy, S. Accelerated profile HMM searches. PLoS Comput. Biol. 7, e1002195 (2011).

Article

ADS

MathSciNet

CAS

PubMed

PubMed Central

Google Scholar

Moffat, L. & Jones, D. Increasing the accuracy of single sequence prediction methods using a deep semi-supervised learning framework. Bioinformatics 37, 3744–3751 (2021).

Article

CAS

PubMed

PubMed Central

Google Scholar

Hu, G. et al. flDPnn: accurate intrinsic disorder prediction with putative propensities of disorder functions. Nat. Commun. 12, 4438 (2021).

Article

ADS

CAS

PubMed

PubMed Central

Google Scholar

Peng, Z. & Kurgan, L. High-throughput prediction of RNA, DNA and protein binding regions mediated by intrinsic disorder. Nucleic Acids Res. 43, e121 (2015).

Article

PubMed

PubMed Central

Google Scholar

Steinegger, M. & Soding, J. MMseqs2 enables sensitive protein sequence searching for the analysis of massive data sets. Nat. Biotechnol. 35, 1026–1028 (2017).

Article

CAS

PubMed

Google Scholar

Zhao, B. et al. DescribePROT: database of amino acid-level protein structure and function predictions. Nucleic Acids Res. 49, D298–D308 (2021).

Article

ADS

CAS

PubMed

PubMed Central

Google Scholar

Meng, E. et al. UCSF ChimeraX: Tools for structure building and analysis. Protein Sci. 32, e4792 (2023).

Article

CAS

PubMed

PubMed Central

Google Scholar

Chen, T. & Guestrin, C. XGBoost: A scalable tree boosting system. In Proc. 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD '16)785–794 (ACM, 2016).

Seplyarskiy, V. et al. A mutation rate model at the basepair resolution identifies the mutagenic effect of polymerase III transcription. Nat. Genet. 55, 2235–2242 (2023).

Article

CAS

PubMed

PubMed Central

Google Scholar

Cheng, J. et al. Accurate proteome-wide missense variant effect prediction with AlphaMissense. Science 381, eadg7492 (2023).

Article

CAS

PubMed

Google Scholar

Download references

Acknowledgements

We thank M. Collins for help with analysis; N. Bandeira, A. Raj, Z. Ignatova and B. Karger for detailed feedback; O. Kok and S. Zheng for help with revisions; and members of the Slavov laboratory for discussions and suggestions.

Funding

The work was funded by an Allen Distinguished Investigator award through The Paul G. Allen Frontiers Group to N.S., an NIGMS award (R01GM144967) to N.S., NCI awards UG3CA268117 and UH3CA268117 to N.S., an NIGMS award (R35GM148218) to N.S., an NIA award (R01AG092460) to N.S., and a Bits to Bytes award from MLSC to N.S.

Author information

Authors and Affiliations

Departments of Bioengineering, Biology, Chemistry and Chemical Biology, Single Cell Proteomics Center, Northeastern University, Boston, MA, USA

Shira Tsour, Rainer Machné, Andrew Leduc, Simon Widmer, Eunice Koo & Nikolai Slavov

Program in Medical and Population Genetics, Broad Institute of MIT and Harvard, Cambridge, MA, USA

Jeremy Guez & Konrad J. Karczewski

Analytic and Translational Genetics Unit, Massachusetts General Hospital, Boston, MA, USA

Jeremy Guez & Konrad J. Karczewski

Parallel Squared Technology Institute, Watertown, MA, USA

Nikolai Slavov

Authors

Shira Tsour

Rainer Machné

Andrew Leduc

Simon Widmer

Eunice Koo

Jeremy Guez

Konrad J. Karczewski

Nikolai Slavov

Contributions

Study design, supervision and raising funding: N.S. Data analyses: S.T., R.M., A.L., S.W., E.K. and N.S. gnomAD analyses: J.G., S.T. and K.J.K. Initial draft: S.T. and N.S. Writing: all authors approved the final manuscript.

Corresponding author

Correspondence to

Nikolai Slavov.

Ethics declarations

Competing interests

N.S. is a founding director and CEO of Parallel Squared Technology Institute, which is a non-profit research institute. S.T. is an employee of Alnylam Pharmaceuticals. All other authors declare no competing interests.

Peer review

Peer review information

Nature thanks Yitzhak Pilpel, Mikhail Savitski and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data figures and tables

Extended Data Fig. 1 Systematic identification and validation of amino acid substitutions.

(a) Number of tumor and normal samples analyzed from each CPTAC dataset. (b) Number of samples analyzed for each healthy tissue from the label-free dataset. (c) Distribution of the percentage of each transcript with a read that is included in the patient-specific databases. (d) Distribution of the number of transcripts with 100% sequence coverage included in each patient-specific protein database. (e) Non-substitution modifications identified in the dependent peptide search are majorly comprised of post-translational modifications, and include artifacts and chemical derivatives from MS analysis. (f)–(k) (Continued on the next page) (f) The number of modified peptides identified as having an amino acid substitution or other type of post-translational or chemical modification. (g) Mass error distributions for SAAP and all peptides identified in the database search show no significant differences. The lower, middle, and upper lines of the boxplots correspond to the first quartile, median, and third quartiles, respectively. The upper whisker extends from the third quartile to the largest value and the lower whisker extends from the first quartile to the smallest value, each at most 1.5XIQR of the hinge. Data beyond the whiskers are outliers that are plotted as individual data points. (h) Butterfly plots showing a systematic mass shift in MS2 spectra between SAAP and BP for a representative SAAP with median RAAS = 1.2 in ribosome-binding protein 1 isoform 1 (RRBP1). The fragmentation spectra were predicted by the Prosit TMT model26. (i) Cumulative density distributions of p-values (MaxQuant) and FDR-controlled q-values computed using only SAAP. Red dashed line indicates confidence threshold for SAAP inclusion in further analysis. (j) Over 80% of substitutions identified from lysine (K) or arginine (R) are at sites of missed cleavage or are substitutions between K and R. (k) Observed and predicted (DeepRT+,20) retention times show strong agreement for all main peptides identified in standard database search and for SAAP. (l) TMT and label-free spectra for the same SAAP provide complementary evidence fragments and are in strong agreement with Prosit predictions25,26. (m) Observed spectra for SAAP quantified in both TMT and label-free datasets are in stronger agreement with the correctly matched prosit model, i.e. TMT spectra with Prosit TMT26 and label-free spectra with Prosit HCD25 than with the mismatched Prosit model.

Extended Data Fig. 2 Establishing confidence in AAS abundance.

(a) SAAP with high RAAS ≥ 1 are identified with the same FDR-controlled confidence as SAAP with low RAAS < 1. (b) SAAP with high RAAS ≥ 1 are identified with as many fragment ions providing evidence at the site of alternate translation as SAAP with low RAAS < 1. (c) SAAP with high RAAS ≥ 1 have similar mass error distributions as SAAP with low RAAS < 1. The lower, middle, and upper lines of the boxplots correspond to the first quartile, median, and third quartiles, respectively. The upper whisker extends from the third quartile to the largest value and the lower whisker extends from the first quartile to the smallest value, each at most 1.5 X IQR of the hinge. Data beyond the whiskers are outliers that are plotted as individual data points. (d-f) SAAP abundance estimates are unlikely to be affected by differences in ionization efficiency between base peptides and alternatively translated peptides. The ionization efficiency distributions for SAAP and BP are indistinguishable (d), correlate strongly (e), and there is negligible fold change between them (f). (g-i) SAAP abundance estimates are unlikely to be affected by differences in peptide detectability between base peptides and alternatively translated peptides. The peptide detectability distributions for SAAP and BP are indistinguishable (g), correlate strongly (h), and there is negligible fold change between them (i). (j) Fourth-root spectral angle for SAAP-predicted vs. BP-predicted spectra. N = 12,124 spectra from 1,891 unique SAAPs. The x-axis shows the spectral angle between the observed spectrum and the Prosit-predicted spectrum for the SAAP; the y-axis shows the same for the BP. Pearson r = 0.72, p = 0.0002, computed by permutation. (k) Scatter plot of median spectral angle vs. median RAAS shows that the fourth-root spectral angles are independent of the substitution ratios. Pearson r = 0.016, two-sided correlation p = 0.45, computed against a non-correlating beta distribution. (l) Substitution ratios (RAAS) computed for the same site of alternate translation from peptides in different enzymatic digests of tonsil are consistent for peptides across the range of peptide abundances. Pearson r = 0.77 and two-sided correlation p-value computed against a non-correlating beta distribution = 2.5e-12. (m) Base peptides with missed cleavages are generally an order of magnitude more lowly abundant than their fully cleaved counterparts. (n) Correlation of BP (top panel) or SAAP (middle panel) abundance with shared peptide abundance across samples. Shared peptides are peptides found in both the encoded and alternatively translated proteoforms. BP correlation to shared peptides decreases with increasing RAAS, while SAAP correlation to shared peptides tends to increase, especially at RAAS > 1, as indicated by the difference in these correlations (bottom panel), and in support of the hypothesis presented in Fig. 2a. Abundances are computed with MS2-level intensities.

Extended Data Fig. 3 Quantification of substituted amino acid peptides (SAAP).

(a) Distributions of substitution ratios for all SAAP identified in each dataset computed for each experiment (TMT set, CPTAC data) or sample (label-free data), using MS1 precursor ion intensities. N indicates the number of RAAS computed at the MS1 level in each dataset. (b) Median RAAS were computed for each unique SAAP-BP pair using reporter ion intensities (MS2, CPTAC data) or precursor ion intensities (MS1, label-free data). Distributions of median RAAS across all SAAP in a dataset are shown. N indicates number of unique SAAP-BP pairs identified in each dataset. (c) Median RAAS across all SAAP identified in each sample were computed using reporter ion intensities (MS2, CPTAC data) or precursor ion intensities (MS1, label-free data). Distributions shown are of these medians across all samples in a dataset. N indicates the number of samples in each dataset. (d) Substitution ratio distributions shown in (a), (b), (c) have consistent medians, highlighting variability in RAAS across datasets. (e) Upset plot showing overlap in unique SAAP identified across all datasets. Dataset combinations require at least 10 shared SAAP to be included in visualization. (f) Heatmap displaying the percentage of samples in each dataset in which SAAP identified in 6+ datasets are found. Hierarchical clustering shows a cluster of shared SAAP that are commonly identified across majority of samples in addition to 6+ datasets. (g) To confirm variability in RAAS across datasets, we looked at the subset of SAAP that were identified in at least 1 sample in at least 6 datasets. LUAD and LSCC substitutions consistently have the lowest RAAS, while PDAC substitutions have the highest RAAS. N indicated the number of RAAS computed for shared SAAP in each dataset. (h) Boxplots highlighting the difference between RAAS in CPTAC datasets relative to RAAS computed in LUAD. Only SAAP shared between LUAD and the compared dataset are used. Each data point is a log10(RAAS) difference computed for a unique SAAP-BP pair. (i) RAAS as a function of the minimum number of codon-anticodon mismatches needed for incorporating the detected amino acid across all datasets. (j) An example of a substitution that can be partially explained by synthesis errors arising from significantly (t-test) higher abundance of the amino acyl-tRNA ligase supplying the alternatively translated amino acid relative to the abundance of the amino acyl-tRNA ligase supplying the encoded amino acid. *: q-value < 10−3, **: q-value < 10−5, ***: q-value < 10−20. (k) RAAS negatively correlates to the codon stability coefficient, an empirical measure of codon usage. n denotes number of codons, r is Pearson correlation, p is two-sided correlation p-value computed against a non-correlating beta distribution, and the red line is the ordinary least squares fit. (l) RAAS distributions for SAAP identified and validated in human hepatocytes. (m) The stability of SAAP relative to BP in primary human B cells is inversely proportional to their RAAS. r is Pearson correlation, p is two-sided correlation p-value computed against a non-correlating beta distribution. (n) Same as (m) but in NK cells. The lower, middle, and upper lines of the boxplots in panels h-j correspond to the first quartile, median, and third quartiles, respectively. The upper whisker extends from the third quartile to the largest value and the lower whisker extends from the first quartile to the smallest value, each at most 1.5XIQR of the hinge. Data beyond the whiskers are outliers that are plotted as individual data points.

Extended Data Fig. 4 Associations between codons, incorporated amino acids and RAAS.

(a) Relative codon frequencies in the full transcripts of the set of all proteins with identified substitution sites. The total count for each codon was divided by the total count of all codons for the same amino acid. Codon groups (per amino acid, separated by vertical lines) were sorted by amino acid property groups. Within each codon group, codons were sorted by their relative frequencies. All other panels are aligned with this sorting, see (e) for x-axis labels. These relative frequencies (bar heights) were also used in Fig. 3b (x axis). (b) RAAS dotplot for codons (columns) and incorporated amino acids (rows), sorted and color-coded by amino acid property groups. (c) RAAS dotplot for codons by datasets, i.e., cancer types and healthy tissues. (d) RAAS dotplot for all codons without further subsetting. Note, that the median RAAS values (colors) correspond to the y-axis values in Fig. 3b. (e) RAAS distributions for each codon. The lower, middle, and upper lines of the boxplots in e correspond to the first quartile, median, and third quartiles, respectively. The upper whisker extends from the third quartile to the largest value and the lower whisker extends from the first quartile to the smallest value, each at most 1.5XIQR of the hinge. Data beyond the whiskers are outliers that are plotted as individual data points.

Extended Data Fig. 5 Substitution ratios depend on substitution and tissue types.

(a) RAAS dotplot for all encoded and incorporated amino acids. p-values were computed by two-sided Student’s t-test. See Methods: RAAS dotplots for more details. (b) Violin plots of RAAS medians for each substitution type in every dataset. N indicates number of substitution identified in each dataset. (c) RAAS dotplot as in (a) but by chemical properties of the encoded and incorporated amino acids. (d) Heatmap of median RAAS by substitution type for substitution types with variance <10% across datasets. (e) Heatmap of median RAAS by substitution type for substitution types with variance >50% across datasets. (f) RAAS dotplots for encoded (left panel) and incorporated (right panel) amino acids. (g) Median RAAS values for SAAP grouped based on the encoded amino acid (left panel) or incorporated amino acid (right panel) correlate strongly and significantly across all datasets (Pearson correlation). (h) Substitution types with significantly higher RAAS in a given tissue type (cancer and healthy samples) relative to all other tissues analyzed (t-test, Benjamini-Hochberg FDR-corrected). (i) RAAS distributions (boxplots) and number of SAAP identified (barplot) for substitution types that are significantly higher in a given tissue (colored) relative to all other tissues (gray) analyzed. Colors indicate the same tissue types as in (h). The lower, middle, and upper lines of the boxplots correspond to the first quartile, median, and third quartiles, respectively. The upper whisker extends from the third quartile to the largest value and the lower whisker extends from the first quartile to the smallest value, each at most 1.5 X IQR of the hinge. Data beyond the whiskers are outliers that are plotted as individual data points.

Extended Data Fig. 6 Associations of substitutions with cancer.

(a) RAAS fold change between tumor and patient-matched normal adjacent tissue samples with median of distribution shown in red. N indicates the number of patient-specific RAAS values compared. (b) Patient-level RAAS distributions stratified by clinical tumor stage. No significant associations were measured between RAAS and tumor stage. The lower, middle, and upper lines of the boxplot correspond to the first quartile, median, and third quartiles, respectively. The upper whisker extends from the third quartile to the largest value and the lower whisker extends from the first quartile to the smallest value, each at most 1.5 X IQR of the hinge. Data beyond the whiskers are outliers that are plotted as individual data points. (c) RAAS for the N → G substitution in serine/threonine-protein phosphatase PP1-beta catalytic subunit (PP1CB) is significantly higher in tumor samples than in patient-matched normal adjacent tissue, for the majority of patients in LUAD and LSCC. q-values are derived from two-sided Student’s t-test p-values that were adjusted for multiple testing with the Benjamini-Hochberg method.

Extended Data Fig. 7 Proteins with high RAAS organized by functional groups.

RAAS dotplots for all proteins with significantly high median RAAS (p < 10−5) from each functional category shown in Fig. 5c.

Extended Data Fig. 8 Amino acid sequence context around substitution sites.

(a) Counts (text) and enrichments (gray scale: p-values of cumulative hypergeometric distribution tests) of amino acids surrounding the amino acid substitution sites. Lysine (K) and arginine (R) are enriched directly upstream of the substitution sites. Tryptophan (W), methionine (M), glycine (G) and cysteine (C) are enriched directly adjacent to substitution sites. (b) As (a) but for substitution types (encoded:incorporated) vs. their position in identified peptides. Substitutions by glycine (→ G) or alanine (→ A) are enriched within the 3 N-terminal amino acids of base peptides (N1 to N3), i.e., directly after the trypsin cleavage sites (K or R). Various substitutions involving arginine (N), methionine (M) or glutamate (E) as either the encoded or the incorporated amino acid are enriched distant from the N- and C-termini (>9). Only substitution types with at least one significant enrichment (p < 10−10) are shown in (b). (c) Sequence difference logos were calculated for all unique sequences surrounding substitution sites, subset for all observed substitution types (encoded→incorporated amino acids), and plots were only generated if any of the positions −3 to +3 around a substitution site showed a significant enrichment with p < 10−10 (*: p < 10−3, **: p < 10−5, ***: p < 10−10), (computed with R package DiffLogo57, see Methods: Sequence Difference Logos for details) and all resulting logos are shown. The logos were grouped by common patterns (rows from top to bottom): (i) Substitutions by glycine or alanine (Q → A, Q → G, M → G, L → G) are enriched directly upstream with lysine (K) or arginine (R), i.e. they are preferentially observed at the N-terminus of base peptides, next to the trypsin cleavage sites (K or R). (ii) Substitutions of glutamine (Q → A) or arginine (N → G) are flanked by cysteine (C) enrichments. (iii) substitutions E → N, T → V and N → M are flanked by methionine (M) enrichments. (iv) Substitutions E → C, I → Q, L → Q and P → T are flanked by tryptophan (W) enrichments. (d) Sequence difference logos of selected subsets of substitution sites. AAS denotes the site of the substitution and numbers refer to adjacent positions in the protein sequence. The y-axis shows the Jensen-Shannon divergence of the selected set of sequences (number n of sequences is indicated) compared to all other sequences in our data; *** indicates enrichment significance p < 10−10, computed with R package DiffLogo57, see Methods: Sequence Difference Logos for details.

Extended Data Fig. 9 Substitution proximity in domains, 1D and 3D protein structures.

(a) Pfam domains significantly enriched in substituted peptides (FDR-adjusted p-values < 0.05, see Methods for details). (b) The number of base peptides decreases exponentially with the number of distinct SAAP detected per base peptide. This reflects the abundance bias of AAS detectability, such that many distinct SAAPs can be detected for highly abundant peptides. (c) Clusters of substitutions in the 1D sequence of a thioredoxin domain. Pfam annotations are shown as black lines, and predicted secondary structures (S4pred), α-helices and β-sheets, are shown with gray bars. (d) Clusters of substitutions in the 1D sequence of an actin ATPase domain. Pfam annotations are shown as black lines, and predicted secondary structures (S4pred), α-helices and β-sheets, are shown with gray bars. (e) High density region of substitutions in marginal zone B- and B1-cell specific protein (MZB1). A high RAAS substitutions cluster in β-sheet and unstructured regions is immediately followed by a lower RAAS cluster in an α-helix region. (f) Three 1D clusters of substitutions are found in the glycolytic region of aldolase B (ALDOB). (g) A set of three substitutions distant in the 1D sequences but clustered together in the 3D structure of the Ras-related protein, RAP1A. The substitutions have high RAAS, reflected in their color-coding (as in the legend in (c)). (h) Many substitutions are identified in the ribosome complex, some of which cluster across complex subunits in the 3D protein structure. (i) High and low RAAS substitutions cluster in the 3D protein structure of the proteasome complex.

Extended Data Fig. 10 Associations of substitutions with protein properties.

(a) Protein RAAS is negatively correlated to protein abundance with Spearman r = −0.57 and two-sided correlation p-value = 5.9e-153. The intensity was calculated as the median over all “leading razor protein” intensities of all data sets (CPTAC and tissues) where a given base peptide was identified. (b) RAAS is positively correlated to the disordered score (IUPred3 prediction) with Spearman r = 0.17 and two-sided correlation p-value = 5.2e-15. RAAS was computed as the median RAAS for each unique BP/SAAP pair across all datasets (CPTAC and tissues) and the disordered score is the score of the protein at the AAS site. (c) as (b) but for the mean conservation score (MMseqs2 score via the DescribePROT database). Spearman r = −0.15 and two-sided correlation p-value = 2.5e-39. (d) Allele frequency in the gnomAD database for all possible missense variants in alternatively translated codons. (e) Allele frequency in the gnomAD database for missense variants coding for identified substitution in alternatively translated codons. Sites with allele variation frequency ≥10−3 correspond to 131 SAAPs. (f) Observed / expected ratios for missense variants coding for identified substitution in alternatively translated codons, determined from analysis of gnomAD database (see Methods). Missense variants are less constrained with increasing RAAS (Holm-adjusted two-sample exact Poisson rate-ratio test between hightest and lowest quartiles=3.3e-7). Data point colors correspond to RAAS quartiles as in (d) or low or high AlphaMissense (AM) controls. Error bars represent 95% Poisson confidence intervals on the obs/exp ratio. N = 7,069 substitutions with 1/4 in each RAAS quartile. (g) Upset plot showing overlap in unique SAAP identified between human and mouse tissues.

Rights and permissions

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Reprints and permissions

About this article

Cite this article

Tsour, S., Machné, R., Leduc, A. et al. Alternate RNA decoding results in stable and abundant proteins in mammals.

Nature (2026). https://doi.org/10.1038/s41586-026-10678-2

Download citation

Received: 03 October 2024

Accepted: 18 May 2026

Published: 24 June 2026

Version of record: 24 June 2026

DOI: https://doi.org/10.1038/s41586-026-10678-2