Threonine fuels glioblastoma through YRDC-mediated codon-biased translational reprogramming

Xujia Wu; Huairui Yuan; Qiulian Wu; Yixin Gao; Tingting Duan; Kailin Yang; Tengfei Huang; Shuai Wang; Fanen Yuan; Derrick Lee; Suchet Taori; Tritan Plute; Søren Heissel; Hanan Alwaseem; Michael Isay-Del Viscio; Henrik Molina; Sameer Agnihotri; Dennis J Hsu; Nu Zhang; Jeremy N Rich

doi:10.1038/s43018-024-00748-7

. Author manuscript; available in PMC: 2025 Jul 1.

Published in final edited form as: Nat Cancer. 2024 Mar 22;5(7):1024–1044. doi: 10.1038/s43018-024-00748-7

Threonine fuels glioblastoma through YRDC-mediated codon-biased translational reprogramming

Xujia Wu ^1,², Huairui Yuan ¹, Qiulian Wu ¹, Yixin Gao ², Tingting Duan ¹, Kailin Yang ³, Tengfei Huang ¹, Shuai Wang ¹, Fanen Yuan ¹, Derrick Lee ¹, Suchet Taori ¹, Tritan Plute ^4,⁵, Søren Heissel ⁶, Hanan Alwaseem ⁶, Michael Isay-Del Viscio ⁶, Henrik Molina ⁶, Sameer Agnihotri ^4,⁵, Dennis J Hsu ^1,⁷, Nu Zhang ^2,^✉, Jeremy N Rich ^1,^8,^✉

PMCID: PMC11552442 NIHMSID: NIHMS2032325 PMID: 38519786

Abstract

Cancers commonly reprogram translation and metabolism, but little is known about how these two features coordinate in cancer stem cells. Here we show that glioblastoma stem cells (GSCs) display elevated protein translation. To dissect underlying mechanisms, we performed a CRISPR screen and identified YRDC as the top essential transfer RNA (tRNA) modification enzyme in GSCs. YRDC catalyzes the formation of N⁶-threonylcarbamoyladenosine (t⁶A) on ANN-decoding tRNA species (A denotes adenosine, and N denotes any nucleotide). Targeting YRDC reduced t⁶A formation, suppressed global translation and inhibited tumor growth both in vitro and in vivo. Threonine is an essential substrate of YRDC. Threonine accumulated in GSCs, which facilitated t⁶A formation through YRDC and shifted the proteome to support mitosis-related genes with ANN codon bias. Dietary threonine restriction (TR) reduced tumor t⁶A formation, slowed xenograft growth and augmented anti-tumor efficacy of chemotherapy and anti-mitotic therapy, providing a molecular basis for a dietary intervention in cancer treatment.

Glioblastoma (GBM) represents the most common primary malignant brain tumor with a median survival of less than 2 years¹. GBM was originally designated as GBM multiforme due to marked intratumoral morphologic variation. Modern molecular analyses have shown that GBMs display striking spatial and temporal variation in genomic and epigenomic states and the tumor microenvironment, including the vasculature and immune responses². Most GBMs recur within 2 to 3 cm of the original resection cavity often with a nodular pattern of growth, suggesting clonal recurrence. Failure to achieve cure for patients with GBM is multifactorial, including the presence of GSCs, a stem-like population that exhibits resistance to radiotherapy and chemotherapy and generates a hierarchy of cell types within tumors³. Although the impact of GSCs remains controversial, defining their molecular regulation may offer therapeutic paradigms to improve the clinical care of patients afflicted with GBM.

Many cancers, including GBM, display accelerated protein synthesis⁴. However, a deep understanding of translational adaptation in response to oncogenic signaling is lacking, especially in cancer stem cells. Although normal embryonic and tissue-resident stem cells display low global translation rates⁵, mounting evidence suggests that translational regulation in cancer stem cells depends upon tumor type and cell-specific oncogenic signals^6,7. Protein translation requires fine tuning of multiple components and factors, including messenger RNA (mRNA), ribosomes, tRNA and other regulators^4,5. Post-transcriptional modifications of tRNA permits modulation of not only global efficiency in translation of transcripts but also selective efficiency of codons. The human genome encodes ~500 tRNA species that recognize 61 codons for 20 amino acids, for which over 40 types of tRNA modifications have been identified⁸. Dysregulation of tRNA modifications is prevalent in cancer⁹, but how cancer stem cells co-opt tRNA modifications and metabolism to fulfill their translational needs remains largely unexplored.

Among tRNA modifications, t⁶A is highly conserved across evolution¹⁰. This modification localizes to position 37 in the anticodon stem loop and is exclusively found in tRNA species that decode ANN codons. Modified tRNA species are locked in a three-dimensional structure that defines translational efficiency of codons. Impaired t⁶A biosynthesis causes an autosomal recessive disease, Galloway–Mowat syndrome, characterized by defective nervous system development¹¹. Despite the requirement of t⁶A in the nervous system, the role of t⁶A in brain cancer remains poorly characterized.

Based on this background, we hypothesized that cancer stem cells in brain tumors differentially regulate translation through post-transcriptional tRNA modifications to reveal potential therapeutic paradigms. Indeed, we found that GSCs display preferential regulation of protein synthesis with selective dependency on YRDC, a rate-limiting enzyme in t⁶A generation. Concordant with the effects of t⁶A locking tRNA tertiary structures, GSCs display a YRDC-dependent codon bias in translation to sustain mitosis. Leveraging the role of threonine as an enzymatic substrate for YRDC to generate t⁶A, we developed a strategy to translate these observations into a targeting strategy by restricting dietary threonine, resulting in inhibition of tumor proliferation and in vivo growth.

Results

GSCs exhibit high translation rates

To examine the rate of protein synthesis of GSCs, we first injected O-propargyl-puromycin (OPP), an alkyne analog of puromycin that is used for quantification of protein synthesis¹², into the contralateral hemispheres of mice bearing patient-derived xenografts. Brains were harvested after 30 min, and different cell populations were separated by flow cytometry. Human CD147 was employed to enrich for patient-derived tumor cells¹³, and CD133 and SRY-box transcription factor (TF) 2 (SOX2) were used to distinguish stem cell populations¹⁴. OPP signal intensity represented global protein synthesis rates (Fig. 1a). Tumor cells displayed higher percentages of rapidly translating cells (OPP^hi) than non-neoplastic cells (Fig. 1b–d and Extended Data Fig. 1a–c). CD133⁺ GBM cells exhibited higher translation rates than CD133⁻ tumor cells, as measured by the higher frequency of OPP^hi cells (Fig. 1e,f and Extended Data Fig. 1d,e). To address the limitations of using CD133 alone as a GSC marker¹⁴, we included SOX2, a fate-determining GSC marker¹⁵, for validation. Most CD133⁺ GSCs displayed high SOX2 levels (Fig. 1g). CD133⁺SOX2^hi tumor cells showed higher OPP incorporation than CD133⁻SOX2^lo/− tumor cells (Fig. 1h,i), supporting the idea that GSCs are relatively translationally active.

Fig. 1 | — a, Graphic illustration of in vivo protein translation measurement in different cell populations. Human (h) CD147 is used to mark patient-derived tumor cells; CD133 and SOX2 are used to distinguish GSCs. b–i, Gating strategy (b,g), representative histogram plot (c,e,h) and statistical quantification (d,f,i) (n = 6 mice per group) of OPP flow cytometric analysis of the indicated cell populations in GSC23-derived intracranial tumors. The cutoff used to define high (OPP^hi) and low (OPP^lo) OPP signal is 10³ on the logarithmic scale. j–l, Quantification of translational activities in scRNA-seq data of 28 early-passage GSC cultures derived from 24 patients and 14,207 malignant cells from seven patients with GBM. In j, n = 65,655 for all GSCs and n = 14,207 for all tumor cells. In k, n = 64,417 for GSCs and n = 1,238 for tumor-like GSCs. In l, n = 1,971 for GSC-like tumor cells and n = 12,236 for differentiated tumor cells. m, GSEA analysis of GOBP: positive regulation of translation in RNA-seq data of matched GSCs and DGCs (GSE54791). NES, normalized enrichment score. n,o, Immunoblots showing puromycin incorporation in the indicated cells. NSC11, neural stem cell 11; HNP1, human neural progenitor 1. p, Representative images of in vitro OPP incorporation in the indicated cells. Scale bar, 20 μm. DAPI, 4,6-diamidino-2-phenylindole. In d,f,i, boxes represent data within the 25th to 75th percentiles, whiskers depict the range of all data points, and horizontal lines within boxes represent median values. In j–l, violin plots represent the overall distribution of data points. Box plots show median, upper and lower quartiles; whiskers depict 1.5 times the interquartile range. In n–p, immunoblots and images are representative of three independent experiments with similar results. Two-tailed paired t-test for d,f,i. Two-tailed unpaired t-test for j–l. Weighted Kolmogorov–Smirnov statistic test for m.

Next, we mined a large single-cell RNA sequencing (scRNA-seq) dataset that included GSC cultures and tumor cells from patients with GBM¹⁶ (Extended Data Fig. 1f). Stemness marker PPP1R14B and differentiation marker GFAP distinguished GSCs from other tumor cells in two principal components (Extended Data Fig. 1g). A translational signature from the Gene Ontology (GO) database (GO biological process (GOBP): positive regulation of translation) was used to infer translational activity, which was higher in GSCs than in bulk tumor cells (Fig. 1j). To rule out the potential bias of comparison between in vitro GSC cultures and in vivo tumor cells, we segregated each cell population into two parts with distinct differentiation states (Methods) and reperformed the comparison (Extended Data Fig. 1h,i). Consistently, most GSCs showed higher translational activities than tumor-like GSCs (Fig. 1k), and GSC-like tumor cells exhibited higher translational activities than differentiated tumor cells (Fig. 1l).

To explore whether GSCs display increased translation in vitro, we analyzed transcriptomes of a panel of cultured GSCs and matched differentiated tumor cells (differentiated glioma cells; DGCs)¹⁵. Gene set enrichment analysis (GSEA) revealed enriched translational signature in GSCs (Fig. 1m). To measure protein synthesis in vitro, we pulse-treated cells with puromycin for 30 min and detected puromycin incorporation by immunoblotting. GSCs exhibited greater puromycin incorporation than DGCs or neural stem cells (NSCs) (Fig. 1n,o and Extended Data Fig. 1j,k). Increased protein translation was not restricted to a specific subcellular compartment, as indicated by upregulation of OPP signals throughout cells (Fig. 1p). Together, GSCs exhibit accelerated translation in vivo, which can be recapitulated in vitro.

Cancer stem cells, including GSCs, are not obligatorily quiescent and actively divide in dedicated niches¹⁷. We confirmed that GSCs proliferate faster than paired DGCs in vitro (Extended Data Fig. 1l,m). To rule out the possibility that the increased translation in GSCs is simply a byproduct of increased proliferation, we used a pan-cyclin-dependent kinase (CDK) inhibitor, alvocidib¹⁸, to reduce the proliferation of GSCs to a level comparable to that of paired DGCs (Extended Data Fig. 1n,o). Under similar proliferation rates, GSCs treated with alvocidib still exhibited higher global translation rates than paired DGCs (Extended Data Fig. 1p), indicating the existence of other translational regulators. Also, the increased translation in GSCs is not simply attributable to increased metabolic activity, as we did not observe a consistent alteration of metabolic markers, including mammalian target of rapamycin (mTOR), AMPKα and eIF2α, in GSCs compared to DGCs (Extended Data Fig. 1q). Together, neither increased proliferation nor metabolic activity appear to account for increased translation in GSCs.

CRISPR screening of tRNA modifiers in GSCs

Dysregulation of tRNA modifications serves essential roles in translational regulation and cancer progression^8,9. To investigate potential tRNA modifiers that contribute to GSC growth and translational activation, we performed CRISPR knockout screening in two patient-derived GSCs targeting 111 genes reported or predicted to regulate tRNA modification from the GO database, the Reactome database and prior reports^8,19 (Fig. 2a and Supplementary Table 1).

Fig. 2 | — a, Graphic illustration of CRISPR knockout screening targeting tRNA modification genes. sgRNA, single-guide RNA. RRA, robust ranking aggregation. b,c, Gene rank of negative selection results for GSC456 (b) and GSC23 (c) cells in CRISPR screens. Values lower on the y axis indicate greater gene essentiality. The top three ranked genes are labeled. d, Significant hits (P < 0.05, two sided, calculated with the MAGeCK algorithm) in negative selection results from b,c. Red color labels the enzymes involved in t⁶A biosynthesis. The Venn diagram shows hits overlapping in both screens. e–i, Gene rank of tRNA modification genes in genome-wide CRISPR knockout screens of five GSCs. More positive BF scores indicate higher confidence of essentiality. Top hits and *YRDC* are highlighted. j, Gene rank plot showing differences of average quantile normalized BF (qBF) scores for each tRNA modification enzyme between GSCs and NSCs. Data are from genome-wide CRISPR knockout screens of 24 GSCs and four NSCs. Data are z transformed. Top hits and *YRDC* are highlighted. Diff, difference. k, Chronos analysis of *YRDC* dependency in different cell lines in DepMap CRISPR knockout screens (n = 1,077 independent screens in total). Score < −0.6 is used as the cutoff of essentiality. Boxes represent data within the 25th to 75th percentiles, whiskers depict the range of all data points, and horizontal lines within boxes represent median values. l, Heatmap showing z-transformed BF scores of *YRDC* in genome-wide CRISPR knockout screens of five carcinoma cell lines and one non-transformed epithelial cell line.

Depletion of guide RNA (gRNA) content of GSCs on day 14 compared to day 1 (baseline) identified YRDC as the top essential tRNA modifier in both GSC screens (Fig. 2b,c and Supplementary Tables 2 and 3). By contrast, most gRNA species enriched on day 14 were specific for each GSC (Extended Data Fig. 2a), which may reflect the heterogeneity of GSCs. We identified nine significant hits in the screen of GSC456 cells and 12 significant hits in the screen of GSC23 cells, of which YRDC and RPUSD2 were the only two genes overlapping (Fig. 2d). YRDC encodes an enzyme that functions with the downstream KEOPS complex (LAGE3, OSGEP, TP53RK, TPRKB and GON7) and O-sialoglycoprotein endopeptidase-like 1 (OSGEPL1) to catalyze the biosynthesis of t⁶A on cytoplasmic and mitochondrial tRNA species, respectively¹⁰. In the GSC23 screen, three of the five subunits of the KEOPS complex (encoded by TPRKB, GON7 and TP53RK) were depleted on day 14 (Fig. 2d), highlighting the crucial role of cytoplasmic t⁶A in GSC propagation.

To confirm these observations, we mined published genome-wide GSC screens^16,20,21 with the BAGEL algorithm²², using higher Bayer factor (BF) scores to indicate greater confidence of gene essentiality. Among the tRNA modification enzymes, YRDC ranked as one of the top essential genes in multiple GSC screens (Fig. 2e–i). To explore GSC-specific dependencies, we compared BF scores between GSCs and NSCs for each gene. YRDC ranked as the second most GSC-specific tRNA modifying gene by fitness relative to NSCs (Fig. 2j). To investigate the pan-cancer essentiality of YRDC, we next explored screen results from the Cancer Dependency Map (DepMap) project²³. YRDC was essential in the majority of cancer cell lines, while the dropout effects of YRDC in non-neoplastic cells were only modest (Fig. 2k). We found similar results in another genome-wide screen dataset²⁴, as YRDC was essential across different carcinoma cell lines but non-essential in non-transformed epithelial cells (Fig. 2l). Collectively, these data suggest a specific requirement for YRDC in GSCs and other cancer cells, providing the rationale for targeting YRDC in GBM and other malignancies.

YRDC is essential for GSC maintenance and translation

In the transcriptomic data of GSCs versus matched DGCs¹⁵ and GBM versus normal brain cortex, tRNA modification enzymes were differentially expressed, as we identified 19 enzymes enriched in GSCs and 17 enzymes enriched in GBM (Fig. 3a,b and Supplementary Table 4). Overlap between these two sets highlighted YRDC as one of the two enzymes upregulated in both GSCs and GBM (Fig. 3c). Preferential expression of YRDC in GSCs compared to both DGCs and NSCs was further confirmed by immunoblotting (Fig. 3d,e). Analysis of chromatin immunoprecipitation (ChIP)–sequencing (ChIP–seq) data^15,25 showed extensive enrichment of acetylation of histone H3 on lysine 27 (H3K27) (H3K27ac) on the YRDC promoter in GSCs (Extended Data Fig. 2b), indicating that YRDC is actively transcribed in GSCs. Proneural (PN) DGCs are reprogrammed to GSCs by expressing a set of core neurodevelopmental TFs:POU3F2, SOX2, SALL2 and OLIG2 (ref. 15). Upon reprogramming, the H3K27ac signal increased on the YRDC promoter (Extended Data Fig. 2c).

To understand the upstream regulator of YRDC, we scanned TF-binding motifs on the YRDC promoter and identified 568 TFs potentially associated with YRDC (Supplementary Table 5), of which 36 TFs were enriched in GSCs compared to both DGCs and NSCs (Extended Data Fig. 2d). OLIG1 and OLIG2 were the top two TFs based on their expression patterns (Extended Data Fig. 2e). Loss of OLIG2, but not OLIG1, reduced YRDC expression (Extended Data Fig. 2f–i), suggesting that OLIG2 plays a more important role in driving YRDC transcription. We further identified peaks enriched within the YRDC promoter in OLIG2 ChIP–seq data¹⁵ (Extended Data Fig. 2j). We predicted two potential OLIG2-binding sites, which we verified by ChIP followed by quantitative PCR (qPCR) (ChIP–qPCR) analysis in two GSCs (Extended Data Fig. 2k,l), indicating that OLIG2 directly binds to the YRDC locus. OLIG2 expression positively correlated with YRDC expression in the Chinese Glioma Genome Atlas (CGGA) and the Mack_GBM dataset²⁵ (Extended Data Fig. 2m,n).

Next, we interrogated YRDC contributions to GSC maintenance and protein translation. Targeting YRDC (Extended Data Fig. 3a) decreased GSC viability (Fig. 3f–h) and 5-ethynyl-2′-deoxyuridine (EdU) incorporation (Fig. 3i and Extended Data Fig. 3b). Depleting YRDC diminished GSC self-renewal, as assayed by impaired sphere formation frequency using extreme limiting dilution assays (Fig. 3j–l) and reduced sphere size (Fig. 3m and Extended Data Fig. 3c), demonstrating the crucial role of YRDC in GSC maintenance. By contrast, loss of YRDC did not substantially diminish NSC proliferation (Extended Data Fig. 3d,e), demonstrating the specific requirement for YRDC in GSCs.

As YRDC catalyzes the formation of t⁶A on tRNA, we hypothesized that YRDC functions through t⁶A to regulate protein translation in GSCs. Mass spectrometry (MS) showed that targeting YRDC reduced t⁶A on tRNA (Fig. 3n,o), consistent with its enzymatic activity. YRDC-depleted GSCs exhibited slower translation rates (Fig. 3p). mTOR signaling²⁶ and the eukaryotic initiation factor (eIF)2α-mediated integrated stress response (ISR)²⁷ are two essential pathways in translational regulation. YRDC deprivation did not substantially diminish the phosphorylation of mTOR or induce activation of eIF2α (Extended Data Fig. 3f), suggesting that reduced translation was not mediated through these two pathways.

Threonine dynamically regulates t⁶A and translation

Threonine is an essential substrate of YRDC¹⁰ (Fig. 4a). We traced threonine using stable isotope-labeled threonine ([¹³C₄,¹⁵N]l-threonine) for 6 h, resulting in about 30% labeled t⁶A in GSCs, with lower labeled levels in DGCs (Fig. 4b), which suggests the rapid influx of threonine to t⁶A formation . To test whether threonine availability dynamically regulates t⁶A levels, we modulated threonine concentrations in culture medium and analyzed t⁶A by MS. Threonine supplementation increased t⁶A without affecting total adenosine levels at 72 h. By contrast, TR decreased t⁶A without reducing total adenosine levels (Fig. 4c). Consistent with the changes in t⁶A, threonine supplementation facilitated protein synthesis, while TR reduced protein synthesis (Fig. 4d,e). These results demonstrate that there is a concentration-dependent relationship between threonine levels and t⁶A biosynthesis and rates of protein translation.

As threonine is an essential amino acid, we next asked whether the inhibitory effect of TR on protein translation was nonspecific due to cellular sensing of an amino acid deficiency. Mammalian cells sense amino acid levels mainly through mTOR complex 1 (mTORC1)-mediated signaling²⁸ and the general control nonderepressible 2 (GCN2)–eIF2α-mediated ISR pathway²⁷. Upon threonine supplementation, mTOR and eIF2α phosphorylation levels did not change (Fig. 4f). As arginine has been reported to activate mTOR activity^29,30, we included arginine as a control. Arginine supplementation increased puromycin incorporation as well as mTOR phosphorylation, while threonine supplementation accelerated translation without affecting mTOR phosphorylation levels (Extended Data Fig. 4a), supporting the notion that threonine abundance promotes increased translation by a different mechanism than arginine. Increased puromycin incorporation under threonine supplementation was reversed upon YRDC loss (Fig. 4g,h), indicating that threonine supplementation is unable to promote increased translation when t⁶A biosynthesis is impaired. Unlike arginine, TR in general did not alter mTOR activity (Fig. 4i and Extended Data Fig. 4b), supporting the idea that the effect of threonine was independent of mTOR signaling. Although phosphorylation of eIF2α was increased at threonine concentrations lower than 2 μM, TR from 800 μM to 4 μM for 72 h in culture medium did not affect eIF2α phosphorylation (Fig. 4i and Extended Data Fig. 4b), yet translation was reduced in this range of TR (Fig. 4e). Furthermore, we observed stable charging levels in all three detectable tRNA^Thr species (tRNA^Thr_AGT, tRNA^Thr_CGT, tRNA^Thr_TGT)³¹ over this range of threonine manipulation (Fig. 4j), which reinforces the notion that translational effects were not mediated by alteration of tRNA^Thr charging and decoding. Together, these findings suggest that threonine availability fuels translation through YRDC and t⁶A biosynthesis with limited contributions from nonspecific effects of threonine deficiency.

Given the impact of threonine on GSC growth, we hypothesized that GSCs reprogram their metabolism to augment threonine availability. Indeed, GSCs displayed higher levels of intracellular threonine than DGCs and NSCs (Fig. 4k and Extended Data Fig. 4c), which could be derived either from greater uptake or lower degradation. In mammalian cells, ASCT1 (SLC1A4) and ASCT2 (SLC1A5) are the major transporters contributing to threonine uptake³². However, neither SLC1A4 nor SLC1A5 was consistently preferentially expressed by GSCs (Extended Data Fig. 4d,e). Threonine is mainly catabolized through three independent pathways involving threonine dehydrogenase (TDH), threonine aldolase (TA) and serine dehydratase–threonine deaminase (SDS)³³. TDH is a pseudogene in humans³⁴, and TA has minimal enzymatic activity in mammals³⁵, suggesting that SDS is the major downstream catabolic enzyme in human cells (Extended Data Fig. 4f). GSCs express low SDS levels (Extended Data Fig. 4g), suggesting that reduced catabolism in GSCs likely contributes to threonine accumulation. To test this hypothesis, we overexpressed the degradative enzyme SDS in GSCs, resulting in reduced intracellular threonine levels and impaired t⁶A biosynthesis (Fig. 4l–n). Together, these results support the idea that high threonine levels in GSCs are due to reduced degradation, which facilitates t⁶A formation.

Threonine functions mainly through YRDC in GSCs

To better understand the underlying mechanism, we performed matched RNA-seq and tandem mass tag (TMT)-labeled quantitative proteomic analysis in patient-derived GSCs upon either YRDC deprivation or TR. GSCs cultured in control medium and threonine-restricted medium displayed strong differences as mapped by principal-component analysis (PCA), while targeting YRDC decreased the distances induced by TR (Fig. 5a,b), suggesting that the effect of TR was partially diminished by YRDC loss. To explore YRDC-dependent threonine effects, we defined differentially expressed genes (DEGs) induced by TR in cells with intact YRDC expression and then compared them to those in cells with YRDC depletion. Most TR DEGs were no longer differentially expressed compared to baseline conditions upon targeting YRDC expression (Fig. 5c–f), indicating that these alterations were mostly dependent on YRDC. GO enrichment analysis based on the YRDC-dependent downregulated genes identified pathways mainly related to cell cycle regulation, including ‘mitotic cell cycle’, ‘DNA-templated DNA replication’, ‘nuclear chromosome segregation’, ‘regulation of chromosome separation’ and ‘attachment of spindle microtubules to kinetochore’. YRDC-dependent upregulated genes were associated with cell differentiation and metabolic processes of multiple biomolecules, including ‘cellular nitrogen compound’, ‘metal ion’, ‘fatty acid’ and ‘carbohydrate’ (Fig. 5g–j). Together, these results demonstrate that the metabolic function of threonine is mainly mediated through YRDC in GSCs, which facilitates cell cycle progression and involves in multiple cellular metabolic processes.

Fig. 5 | — a,b, PCA analysis of transcriptomic (a) and proteomic (b) data of GSC456 cells with or without YRDC knockdown cultured in control or threonine-restricted (TR, 4 μM, 72 h) medium. Solid circles indicate cells with normal *YRDC* expression. Dashed circles indicate cells with YRDC knockdown. Dim., dimension. c–f, Volcano plot showing DEGs and differentially expressed proteins between GSC456 cells with intact *YRDC* expression cultured in control and threonine-restricted media (4 μM, 72 h) in transcriptomic data (c) and proteomic data (e). All genes were projected to the same comparison in GSC456 cells with YRDC knockdown in transcriptomic data (d) and proteomic data (f), and coloring indicates the status of these genes in c,e. Those DEGs and differentially expressed proteins from c,e that turned stable in d,f are defined as YRDC-dependent alterations. Cutoff, log₂ |fold change| > 1 and adjusted (adj.) P < 0.05. g–j, GO enrichment analysis of biological process (GOBP) of YRDC-dependent downregulated and upregulated genes upon TR (4 μM, 72 h) in transcriptomic data (g,h) and proteomic data (i,j) of GSC456 cells. IMP, inosine monophosphate; miRNA, microRNA; ncRNA, noncoding RNA; rRNA, ribosomal RNA; snRNP, small nuclear ribonucleoprotein. In a,c,d,g,h, transcriptomic data are from three biologically independent samples per group. In b,e,f,i,j, proteomic data are from four biologically independent samples per group.

Threonine and YRDC fuel mitosis with ANN codon bias

Dysregulation of tRNA modifications often leads to alterations in tRNA abundance or function⁹. In human cytoplasm, the t⁶A moiety exclusively decorates ANN-decoding tRNA species³⁶. To understand the role of t⁶A for its carriers, we performed transcriptome-wide tRNA sequencing (tRNA-seq) under YRDC knockdown or TR. Most tRNA isodecoders remained stable (Extended Data Fig. 5a,b) without differences in the expression levels of either ANN-decoding or non-ANN-decoding tRNA isodecoders (Fig. 6a and Extended Data Fig. 5c), suggesting that t⁶A does not contribute to the stability or abundance of tRNA. We next asked whether t⁶A contributed to the function ofANN-decoding tRNA in GSCs, as revealed by structural analysis³⁷. Based on the codon specificity of the t⁶A carriers, we hypothesized that t⁶A regulates translation not only at a global level but also in an ANN codon-biased manner. First, we calculated ANN codon usage frequencies across the principal coding sequences (CDS) annotated by APPRIS³⁸ for each coding gene (Supplementary Table 6). In proteomic analysis, loss of YRDC tended to decrease the quantity of proteins enriched with ANN codons (Fig. 6b), suggesting that YRDC facilitates the translation of ANN codon-enriched transcripts. GO enrichment analysis of those downregulated proteins upon YRDC targeting identified top pathways specifically related to cell cycle progression and cell division (Extended Data Fig. 5d). Consistent with YRDC-knockdown effects, we observed that the downregulated proteins in TR tended to possess higher ANN codon frequencies, although the difference was not statistically significant (P = 0.0797) (Extended Data Fig. 5e). Overlap of YRDC-knockdown and TR effects identified 18 co-downregulated proteins enriched with ANN codons (Fig. 6c,d), which were related to mitotic cell cycle regulation in GO enrichment analysis (Fig. 6e). GO enrichment analysis of the downregulated proteins exclusively associated with YRDC knockdown identified the ‘cerebral cortex cell migration’ pathway. By contrast, the downregulated proteins exclusively associated with TR were enriched with proteins encoded by mitosis-related genes, indicating the complexity of TR in cell cycle regulation (Extended Data Fig. 5f). Together, threonine and YRDC fuel the translation of cell cycle-related transcripts in an ANN codon-dependent manner, which was consistent with increased G0/G1 phase and decreased S/M phases under YRDC deprivation and TR (Extended Data Fig. 5g,h).

Fig. 6 | — a, Expression of ANN-decoding and non-ANN-decoding tRNA isodecoders in tRNA-seq of GSC456 cells upon YRDC knockdown. Data are from three biologically independent samples per group. CPM, counts per million. b–d, ANN codon frequencies of differentially expressed proteins (b), Venn diagram of downregulated proteins (c) and ANN codon frequencies of overlapping differentially expressed proteins (d) in proteomics of GSC456 cells with the indicated treatments. e, GO enrichment of co-downregulated proteins from c. KD, knockdown; GOCC, GO cellular component. f, Comparison of transcriptomics and proteomics upon YRDC knockdown in GSC456 cells. Colored dots indicate translational dysregulation. Orange labels nine of 18 co-downregulated proteins that are downregulated at the translation level. g, Distribution of ANN codon frequencies in CDS. Orange dots indicate the nine proteins from f. h,i, Representative immunoblots from three independent experiments showing t⁶A targets in GSCs with the indicated treatments. j,k, Correlations between ANN codon occupancy alteration upon YRDC knockdown and TR at the ribosome A site (j) and the A + 1 site (k). Red dots show overlapping stalled codons. The black line shows linear regression. Ribosome profiling data are from two biologically independent samples per group. l,m, Codon frequencies of each ANN codon in humans. Coloring indicates the stalling status of ANN codons in j. n–p, Overlapping ANN codon frequencies of differentially expressed proteins in proteomics of GSC456 cells with the indicated treatments. q, Distribution of overlapping ANN codon frequencies in CDS. Orange dots indicate six t⁶A targets. r, Graphic illustration of two synonymous reporters. MSCV, murine embryonic stem cell virus promoter. Luc, luciferase; Rluc, *Renilla* luciferase. s,t, Luciferase activities of the indicated reporters and qPCR analysis of *YRDC* in 293T cells with the indicated treatments (n = 3 independent experiments). Data are presented as mean ± s.d. In a,b,d,n–p, boxes represent data within the 25th to 75th percentiles, whiskers depict the range of all data points, and horizontal lines within boxes represent median values. In b–d,f,n–p, proteomic data are from four biologically independent samples per group. In f, transcriptomic data are from three biologically independent samples per group. Two-tailed unpaired t-test for a,s,t. Two-tailed Mann–Whitney test for b,d,n–p. Over-representation test corrected by FDR for e. Two-tailed Pearson correlation for j,k.

By combined analysis of matched proteomic and transcriptomic data, we observed that nine of the 18 co-downregulated proteins were unchanged at the RNA level upon YRDC targeting (Fig. 6f), indicating that their protein levels were likely regulated by translational buffering. The majority of these genes (six of nine) harbor ANN codon frequencies greater than the 75th percentile across the whole human CDS (Fig. 6g). These six targets decreased at the protein level upon YRDC loss or TR (Fig. 6h,i), while alterations at the RNA level were minimal (Extended Data Fig. 5i,j). Collectively, we identified six proteins as the most prominent translational targets of t⁶A in GSCs: SPC25 component of NDC80 kinetochore complex (SPC25), microtubule-associated serine–threonine kinase like (MASTL), Rac GTPase-activating protein 1 (RACGAP1), cellular inhibitor of PP2A (CIP2A), centrosomal protein 55 (CEP55) and non-SMC condensin I complex subunit G (NCAPG). In published screen datasets^16,20,21,39, SPC25, MASTL, RACGAP1 and NCAPG were essential in both GSCs and GBM generally, with CIP2A essential in a subset of GSCs and variable dependencies of CEP55 in different screens (Extended Data Fig. 6a). In clinical datasets, the protein abundance of these targets positively correlated with YRDC protein abundance and with each other in the Proteomic Data Commons (PDC) GBM cohort⁴⁰ (Extended Data Fig. 6b,c). High expression levels of these targets were associated with higher glioma grade in the Cancer Genome Atlas (TCGA) and CGGA datasets (Extended Data Fig. 6d,e). Together, these results indicate the requirement of the t⁶A downstream targets in GSCs and GBM.

To gain deeper insight into the regulation of t⁶A at a single-codon resolution, we performed ribosome profiling⁴¹ of GSCs with YRDC targeting or TR. The isolated ribosome footprints were about 30 nucleotides in length (Extended Data Fig. 7a) and showed a strong three-nucleotide periodicity on the CDS frame (Extended Data Fig. 7b), which is consistent with the characteristics of ribosome profiling data⁴². The transcripts with downregulated translation efficiency upon YRDC loss were also enriched with cell cycle regulation pathways (Extended Data Fig. 7c,d), further supporting the role of t⁶A in cell cycle regulation. During translation, tRNA pairs with three binding sites within the ribosome, specifically, the A site (aminoacyl), the P site (peptidyl) and the E site (exit), with enrichment of ribosome footprints at the A site indicating longer ribosome dwell time and ribosome stalling. To investigate ribosome stalling events, we leveraged the CONCUR pipeline⁴³ to infer the ribosome A site codon occupancy for each codon. YRDC loss and TR showed a similar distribution of ANN codon occupancy at the A site (based on the correlation coefficient and significant P value in Fig. 6j). We further observed that five ANN codons were increased under both YRDC loss and TR (AAA, AGT, ATG, AGG and AGC) (Fig. 6j), suggesting that inhibition of t⁶A biosynthesis and threonine limitation both result in similar patterns of ribosome stalling. By contrast, this pattern of codon stalling was completely lost at the A + 1 site (Fig. 6k). YRDC loss and TR were more likely to affect frequently used ANN codons in human CDS (Fig. 6l,m). Amino acid deprivation leads to ribosome stalling at specific codons encoding related amino acids^44,45. We observed no ribosome stalling at threonine codons under TR (4 μM at 72 h) (Fig. 6m), which is consistent with the minimal alteration of tRNA^Thr charging levels (Fig. 4j) and further supports the idea that ribosome stalling under threonine limitation was mediated by t⁶A instead of impaired tRNA^Thr charging and decoding. At the protein level, downregulated proteins under YRDC targeting and TR were enriched with the overlapping stalled ANN codons (Fig. 6n,p), and the six t⁶A downstream targets also exhibited high levels of overlapping ANN codon frequencies (Fig. 6q). By contrast, although we identified 21 overlapping non-ANN codons stalled under both conditions (Extended Data Fig. 7e), they were not enriched in the differentially expressed proteins of proteomic data (Extended Data Fig. 7f–h). Additionally, the previously identified t⁶A downstream targets possessed low frequencies of these overlapping non-ANN codons (Extended Data Fig. 7i), suggesting that the effects on non-ANN codons are nonspecific to YRDC- and threonine-mediated t⁶A function. Collectively, these findings reveal the function of t⁶A at a single-codon resolution, providing a better understanding of the interplay between threonine, t⁶A and their requirement for decoding specific ANN codons.

We next developed two synonymous reporters with different ANN codon frequencies in one of our target genes, SPC25 (that is, ANN hi-SPC25 and ANN lo-SPC25). These sequences were placed upstream of that for luciferase, followed by the sequence for internal ribosome entry site (IRES)-driven Renilla luciferase as an internal control (Fig. 6r). Despite achieving similar levels of YRDC knockdown, the translation of ANN hi-SPC25 was more sensitive to YRDC loss that that of ANN lo-SPC25, as measured by lower luciferase activity (Fig. 6s) with comparable transcript expression (Extended Data Fig. 7j). Similarly, the translation of ANN hi-SPC25 was more sensitive to TR than that of ANN lo-SPC25 (Fig. 6t and Extended Data Fig. 7k). These results further support the notion that threonine and YRDC regulate ANN codon-biased translation and that one possible advantage of upregulating YRDC in GSCs is to facilitate the synthesis of ANN-rich genes that promote cell cycle progression.

Dietary TR inhibits tumor growth

We next asked whether YRDC expression and threonine abundance promote tumor growth in vivo. Targeting YRDC in patient-derived xenografts resulted in reduced tumor growth (Fig. 7a,b and Extended Data Fig. 8a–e), which translated into prolonged survival of tumor-bearing mice (Fig. 7c). Analysis of tumor-bearing brains revealed that loss of YRDC diminished tumor cell mitosis (Extended Data Fig. 8f–i). These results support the development of agents targeting t⁶A biosynthesis in GBM treatment. However, inhibitors directly targeting YRDC or the downstream enzymes involved in t⁶A biosynthesis are not currently available. We therefore asked whether TR could be an alternative approach to target t⁶A and tumor growth. Strict TR in vitro with 2 μM threonine in the medium inhibited proliferation of GSCs and NSCs, but TR at slightly less restriction (4 μM) substantially reduced cell viability of GSCs with minimal effects on matched DGCs and other non-malignant cells (Fig. 7d), demonstrating that threonine and t⁶A availability are especially limiting in GSCs.

Fig. 7 | — a, Representative in vivo bioluminescence imaging (left) and quantification (right, n = 6 mice per group) of mice bearing the indicated xenografts. Images were acquired when the first neurological sign occurred in any cohort. Scale bar, 1 cm. b, Tumor growth curve from in vivo bioluminescence analysis of mice bearing the indicated xenografts (n = 5 mice per group). Data are presented as mean ± s.e.m. c, Kaplan–Meier survival curves of mice bearing the indicated xenografts (n = 6 mice per group for GSC456 cells and n = 5 mice per group for GSC468 cells). d, Cell viability of the indicated cells cultured in control or TR medium for 72 h. Data are from four independent experiments. e–h, Representative in vivo bioluminescence imaging and quantification (e,f, n = 5 mice per group; scale bars, 1 cm) and representative images of hematoxylin and eosin (H&E)-stained brain sections (g,h; scale bars, 1 mm) of tumor-bearing mice fed the indicated diets. Data were acquired on day 21. i, Kaplan–Meier survival curves of tumor-bearing mice fed the indicated diet (n = 5 mice per group). j, MS analysis of tissue t⁶A levels of xenografts with the indicated dietary treatment (GSC456, n = 9 mice on the control diet and n = 8 mice on the TR diet; GSC468, n = 8 mice per group). k–m, Gating strategy (k), representative histogram plot (l) and statistical quantification (m) (GSC456, n = 5 mice on the control diet and n = 4 mice on the TR diet; GSC468, n = 4 mice per group) of in vivo OPP flow cytometric analysis of tumor cells from tumor-bearing mice fed the indicated diet. MFI, median fluorescence intensity. In a,d–f,j,m, data are presented as mean ± s.d. One-way ANOVA followed by multiple comparisons for a. Two-way ANOVA followed by multiple comparisons for b,d. Log-rank test for c,i. Two-tailed unpaired t-test for e,f,j,m.

Next, we investigated the safety and efficacy of dietary TR in vivo. According to the minimal requirement of dietary threonine (about 0.18%, wt/wt) for rat maintenance⁴⁶, we fed mice either a control diet (0.82% threonine, wt/wt) or a TR diet (0.2% threonine, wt/wt) and monitored body weight. We did not observe significant weight loss in mice fed the TR diet (Extended Data Fig. 9a), despite 50% reduction of serum threonine concentrations within 3 d, which was maintained (Extended Data Fig. 9b). The TR diet did not cause significant weight loss or pathological abnormalities in any organ examined, including the brain, liver, heart, spleen, lung and kidney (Extended Data Fig. 9c,d). As multiple similarities are shared between GSCs and NSCs³, we asked whether the TR diet was toxic to NSCs. In adult mammalian brains, ventricular–subventricular zone GFAP⁺ astrocytes have been characterized as the largest NSC population⁴⁷, which give rise to intermediate progenitors and DCX⁺ neuroblasts in neurogenesis⁴⁸. Dietary TR did not alter the numbers of ventricular–subventricular zone GFAP⁺ cells (NSCs) (Extended Data Fig. 9e–g). The numbers of DCX⁺ cells (neuroblasts) remained unchanged (Extended Data Fig. 9h,i), indicating that the neurogenic function of NSCs was intact. Together, these results demonstrate the negligible toxicity of dietary TR on NSCs and different tissues.

To determine the value of dietary TR as a therapeutic approach, we initiated either a TR diet or maintained a control diet for 7 d before GSC implantation intracranially and then maintained diets throughout the study. Tumor-bearing mice fed the TR diet showed reduced tumor volume and extension of survival (Fig. 7e–i). Consistent with in vitro effects, TR in vivo suppressed t⁶A formation (Fig. 7j), inhibited tumor protein synthesis (Fig. 7k–m) and reduced tumor cell cycle progression (Extended Data Fig. 9j–m). Similarly, suppression of translation was not due to inhibition of mTOR signaling or activation of the ISR pathway, as we did not observe differential mTOR and eIF2α phosphorylation in tumor tissues from mice exposed to different diets (Extended Data Fig. 9n). To further understand the interplay between threonine and YRDC in vivo, we performed dietary TR with or without targeting YRDC expression in xenografts, revealing suppressed protein synthesis in tumors with loss of YRDC expression (Extended Data Fig. 9o–q). Upon YRDC loss, dietary TR only slightly decreased the fraction of OPP^hi cells and had no further effect on the median fluorescence intensity of the OPP signal (Extended Data Fig. 9p,q). Collectively, these data suggest that dietary TR functions mostly through YRDC and limitation of threonine impairs protein translation and proliferation of GSCs in vivo.

Dietary intervention potentiates standard therapeutics

We next asked whether dietary TR could improve outcomes in combination with standard treatment modalities. To better mimic the clinical use of the TR diet, we initiated dietary intervention and drug treatment 7 d after GSC implantation (Fig. 8a). GSCs display therapeutic resistance, including to chemotherapy⁴⁹. Temozolomide, the standard first-line chemotherapeutic drug, only modestly improved the survival of GSC-bearing mice as a monotherapy, but combining temozolomide with dietary TR provided combinatorial benefit with longer survival in vivo (Fig. 8b,c and Extended Data Fig. 10a). As no YRDC inhibitors have been developed, we interrogated drug sensitivity data from the Cancer Therapeutics Response Portal (version 2)⁵⁰ and gene expression data from the Cancer Cell Line Encyclopedia⁵¹ to identify compounds for which efficacy tracked with YRDC expression. High YRDC expression in cancer cell lines correlated with high sensitivity to an H3K27 trimethylation (H3K27me3) demethylase inhibitor (GSK-J4) and CDK inhibitors (BRD-K30748066, dinaciclib and alvocidib) (Fig. 8d). Although the CDK inhibitor alvocidib is no longer under active clinical development, we previously demonstrated that it is an active agent against GSCs⁵². We thus sought to investigate the potential combinatorial benefit with TR and observed that dietary TR potentiated the efficacy of alvocidib and prolonged the survival of tumor-bearing mice (Fig. 8e). These studies suggest that dietary manipulation of threonine can augment the efficacy of targeted therapeutics.

Fig. 8 | — a–c, Graphic illustration (a), tumor growth curve from in vivo bioluminescence analysis (b, n = 5 mice per group) and Kaplan–Meier survival curves (c, n = 5 mice per group) of GSC468-bearing mice with the indicated treatment. TMZ, temozolomide. d, Therapeutic efficacy prediction of drugs for YRDC (Methods). The blue dot shows the top resistance drug, and red dots show the top sensitive drugs for high *YRDC* expression. e, Kaplan–Meier survival curves (n = 5 mice per group) of GSC468-bearing mice with the indicated treatment. f,g, Immunoblots showing YRDC expression in GBM (T), matched peripheral tissues (P), non-neoplastic epilepsy tissues (N), benign meningioma (BM) and glioma with different grades. h, Representative immunohistochemistry staining showing YRDC expression in LGG and GBM. Scale bar, 100 μm. i,j, Heatmap showing the activities of translational regulation pathways in RNA-seq data of TCGA_GBM and the GTEx brain cortex (i). The pathway used for inferring translational activity is colored red in i and compared in j. k,l, Violin plot of translational activity in RNA-seq data of TCGA_LGG, GBM (grade II, n = 216; grade III, n = 241; grade IV, n = 152) (k) and the CGGA (grade II, n = 188; grade III, n = 255; grade IV, n = 249) (l). m, Kaplan–Meier survival curves of CGGA_GBM (*IDH* wild type) based on *YRDC* mRNA expression. The top 25% and the bottom 25% are defined as high and low groups, respectively. MST, median survival time; m, months. n, Pearson correlation of *YRDC* expression and translational activity in RNA-seq data of CGGA_GBM (*IDH* wild type) (n = 183). The red line shows linear regression. o, Graphic abstract of this study. In b, data are presented as mean ± s.e.m. In j–l, violin plots represent the overall distribution of data points. In i–n, the n number indicates patients. In f–h, data are representative of three independent experiments with similar results. Two-way ANOVA followed by multiple comparisons for b. Log-rank test for c,e,m. Two-tailed Pearson correlation for d,n. Two-tailed unpaired t-test for j. One-way ANOVA followed by multiple comparisons for k,l.

In multiple clinical datasets, including TCGA, the CGGA, Rembrandt and Gravendeel, YRDC expression positively correlated with glioma grade (Extended Data Fig. 10b–e). YRDC was preferentially expressed in GBM tissues at the protein level compared to in paired peripheral brain tissues (Fig. 8f), non-neoplastic brain tissue, meningiomas and low-grade gliomas (LGGs) (Fig. 8g,h). GBM can be classified into three molecular subtypes based on its intrinsic transcriptomic and genomic dimensions: PN, mesenchymal and classical⁵³. YRDC expression was slightly higher in the PN subtype, without differences between mesenchymal and classical subtypes (Extended Data Fig. 10f). The PN subtype expresses high levels of OLIG2 and often harbors PDGFRA alterations and isocitrate dehydrogenase 1 (IDH1) mutations⁵⁴. However, YRDC expression did not correlate with PDGFRA expression, PDGFRA copy number alterations, IDH mutations or MGMT methylation in GBM (Extended Data Fig. 10g–j). Additionally, we assigned a panel of patient-derived GSCs into the three subtypes using a well-recognized classifier⁵³, but no correlation between YRDC expression and GSC subtype was observed (Extended Data Fig. 10k). Overall, these data suggest that YRDC is generally upregulated in GSCs and GBM instead of being restricted to any molecular subtype.

To better understand translational activation in GBM, we performed single-sample GSEA (ssGSEA) to assign tissues with a score based on a panel of translational regulation signatures in bulk RNA sequencing data. Positive translational regulation pathways were upregulated, and most of the negative translational regulation pathways were downregulated in GBM (Fig. 8i). Both positive and negative regulation of translational initiation signatures were enriched in GBM. Translational activity increased in GBM compared to in normal tissue (Fig. 8j), accelerated as glioma progressed from LGG to GBM (Fig. 8k,l) and predicted poor survival in IDH-wild-type GBM (Fig. 8m). YRDC expression positively correlated with increased translational activity (Fig. 8n).

Discussion

Crosstalk among microenvironmental cues, amino acid metabolism, tRNA post-transcriptional modification and translational reprogramming remains poorly understood in tumor biology. Here, we report that cancer stem cells in GBM (that is, GSCs) are characterized by high translation rates, which are driven by YRDC- and threonine-mediated t⁶A modification on ANN-decoding tRNA. Rewired metabolism in GSCs leads to threonine accumulation, which facilitates t⁶A biosynthesis through YRDC and causes ANN codon-dependent translational reprogramming to fuel cell cycle progression. Depletion of YRDC or dietary restriction of threonine dampens t⁶A levels, leading to suppressed translation and compromised GSC survival and GBM growth (Fig. 8o). In clinical data, YRDC is enriched in GBM and correlates with translational activity.

Embryonic stem cells maintain low levels of translation⁵. Mechanisms underlying the disparity in translation levels between normal and cancer stem cells have been unclear. One possibility is that oncogenic mutations in cancer stem cells drive active translation, as oncogenic pathways required for cancer stem cell maintenance, including RTK–RAS, PI3K–AKT, MYC and β-catenin–WNT, promote translation in many malignancies^3,4. Another possibility is that, in addition to a dormant state shared with normal stem cells, GBM harbors a second active proliferating population of GSCs, as revealed by scRNA-seq analysis from IDH-mutant oligodendrogliomas and astrocytomas^55,56. Collectively, our observations support the complexity of translational regulation in cancer stem cell biology.

t⁶A localizes at position 37, which is next to the anticodon (positions 34, 35 and 36). Structural analysis reveals that t⁶A regulates tRNA function through stabilizing anticodon loop conformation, facilitating codon–anticodon pairing and enhancing domain closure of the ribosome around the codon–tRNA complex³⁷. However, increasing evidence suggests that functions of t⁶A are more complex with cell type- and context-specific effects^11,36,57. Here, we found that t⁶A does not contribute to the stability and abundance of tRNA. Instead, t⁶A modifications promote the decoding ability of ANN-decoding tRNA in GSCs, resulting in an ANN codon-biased proteomic shift. The effects of t⁶A modification vary among different tRNA species. Loss of t⁶A tends to cause ribosome stalling on ANN codons that are frequently used in human CDS. Codon usage and tRNA abundance have coevolved such that preferentially used codons correlate with the abundance of cognate tRNA species within cells⁵⁸. Thus, ANN-decoding tRNA species that are frequently used during translation may serve as the main executors of t⁶A signaling, so that cells need not control every ANN-decoding tRNA, enabling an economical and robust strategy for t⁶A signaling in translational regulation.

t⁶A formation may be regulated by environmental cues such as CO₂ and bicarbonate⁵⁹. Here, we found that t⁶A levels can be governed by threonine availability. Aside from serving as a building block, threonine is also required in mouse embryonic stem cells for S-adenosylmethionine production and histone methylation through a TDH-mediated catabolic pathway⁶⁰. However, because TDH is a pseudogene with no corresponding enzymatic function in humans³⁴, the importance of threonine to humans is likely unrelated to S-adenosylmethionine production and histone methylation. Here, our studies identified a metabolic role of threonine in t⁶A biosynthesis and translational reprogramming, which translates into a well-tolerated dietary therapy. We recently found that dietary restriction of another amino acid, lysine, inhibits tumor growth through epigenetic remodeling of endogenous immune responses in preclinical studies⁶¹. These findings collectively contribute to additional layers of metabolic regulation during tumorigenesis, offering therapeutic paradigms to improve the clinical care of patients afflicted with GBM.

Methods

This study complies with all relevant ethical regulations and was approved by the Clinical Research Ethics Committee of the First Affiliated Hospital of Sun Yat-sen University, the Institutional Review Board of the Case Western Reserve University and the Institutional Animal Care and Use Committee of the University of Pittsburgh.

Human glioma and non-neoplastic brain tissues

All pathologically diagnosed glioma samples, their adjacent brain tissues, benign meningioma and non-neoplastic brain tissue (epilepsy) used in this study were collected from excess surgical resection samples from the Department of Neurosurgery at the First Affiliated Hospital of Sun Yat-sen University with written informed consent. The study was approved by the Clinical Research Ethics Committee of the First Affiliated Hospital of Sun Yat-sen University ([2020]322). All patient studies comply with the Declaration of Helsinki.

Glioblastoma stem cell derivation and cell culture

GBM tissues were obtained from excess surgical resection samples from patients at the Case Western Reserve University with written informed consent from patients and in accordance with an institutional review board-approved protocol (090401). All samples were examined by neuropathologists. All patient studies were carried out in accordance with the Declaration of Helsinki. Patient-derived GSCs were obtained and maintained as previously described⁵². The GSC23 sample was derived from a recurrent GBM biopsy specimen from a 63-year-old male patient and was provided as a generous gift from E. Sulman (NYU Langone Health)⁵². The GSC456 sample was derived from a GBM biopsy from an 8-year-old female patient and was provided as a generous gift from D. Bigner (Duke University)⁶². The GSC468 sample was derived from a GBM in our laboratory and transferred via a material transfer agreement from Case Western Reserve University⁶³. The GSCRKI line was derived from a GBM and transferred via a material transfer agreement from the MD Anderson Cancer Centre⁶³. The GSC3028 line was derived from a recurrent GBM from a 65-year-old female patient⁵². The GSC387 line was derived from a GBM from a 76-year-old female patient⁵². To minimize in vitro cell culture-based artifacts, patient-derived xenografts were propagated as a renewable source of GSCs. The NSC11 line (hNSC11, Alstem) was derived from human IPS cells. HNP1 human neural progenitors (HN60001, ArunA Biomedical) are fully differentiated and were derived as adherent cells from the hESC WA09 line. All GSC and NSC lines were cultured in Neurobasal medium (21103049, Gibco) supplemented with B27 without vitamin A (12587010, Gibco), 20 ng ml⁻¹ recombinant human EGF (236-EG-01M, R&D Systems), 20 ng ml⁻¹ recombinant human bFGF (4114-TC-01M, R&D Systems), sodium pyruvate (11360070, Gibco), GlutaMAX (35050 061, Gibco) and streptomycin–penicillin (15140122, Gibco) at 37 °C with 20% oxygen and 5% carbon dioxide. Media contain 800 μM l-threonine unless otherwise noted. Matched serum DGCs were maintained in DMEM (11995065, Gibco) supplemented with 10% FBS (26140079, Gibco) to maintain differentiation status. The non-malignant brain cultures (NM176 and NM177) were derived in our laboratory from resection specimens from patients undergoing surgery to treat epilepsy⁶³ and cultured in mixed medium composed of half DMEM medium supplemented with 10% FBS and half Neurobasal medium. Human normal astrocytes (1800, ScienCell) were derived from a 23-week-old human fetus of unknown sex (not reported by the manufacturer) and cultured in Astrocyte Medium (1801, ScienCell) supplemented with 10% FBS. The 293T line was purchased from ATCC (CRL-3216) and cultured in DMEM medium supplemented with 10% FBS. The customized Neurobasal medium without l-threonine was purchased from Boca Scientific. Aside from removing threonine from the customized medium, all other components are the same as for the Neurobasal medium (21103049, Gibco). Short tandem repeat analyses were performed to authenticate the identity of each tumor model used in this study on a yearly basis. Mycoplasma testing was performed at least once a year to ensure no contamination. All experiments conform to relevant regulatory standards.

Xenograft models

All mouse experiments were performed in accordance with the Institutional Animal Care and Use Committee of the University of Pittsburgh (protocol 21049014). All mice were housed in specific-pathogen-free conditions at an ambient temperature of 20–26 °C and with 30–70% humidity and a 12-h–12-h light–dark cycle before use. Both male and female mice were used in studies. For intracranial xenografts, healthy NSG mice (NOD. Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ, strain 005557, Jackson Laboratory), 4–6 weeks old, were randomly selected for intracranial injection by implanting 10,000 GSCs into the right cerebral cortex at a depth of 3.5 mm following standard procedures. Housing conditions and animal status were supervised by a veterinarian. Animals were monitored until neurologic signs were observed, at which point they were killed. Neurologic signs included hunched posture, gait changes, lethargy and weight loss. Brains were harvested, fixed in 4% paraformaldehyde and then subjected to paraffin-embedded sectioning. H&E staining was performed on sections for histological analysis. Mouse survival was analyzed in GraphPad Prism software, and statistical significance was tested with the log-rank test. Mouse brains implanted with firefly luciferase-labeled GSCs were monitored by bioluminescence imaging. Animals were treated with d-luciferin (50 mg per kg, P1042, Promega) intraperitoneally and anesthetized with isoflurane for imaging analysis. Bioluminescence images were captured with an IVIS imaging system (PerkinElmer).

Threonine-restricted diet and drug treatment in vivo

Special diets with defined threonine levels were purchased from Envigo. The control diet contained 0.82% threonine (wt/wt; TD.01084, Envigo), and the threonine-restricted diet contained 0.20% threonine (wt/wt; TD.220352, Envigo). Non-essential amino acids were proportionally adjusted to keep nitrogen constant. Vitamins were increased to compensate for losses during irradiation sterilization. For examining the safety and efficacy of the diets, tumor-free NSG mice were fed with either the control or the threonine-restricted diet. At the indicated time points, mouse body weight was monitored, blood was sampled through tail bleeding in the morning, organ weight was measured and organs were collected for paraffin-embedded sectioning followed by H&E staining and immunofluorescence analysis. For examining the effect of dietary threonine intervention on tumor growth, NSG mice, 4–6 weeks old, were fed either the control or the threonine-restricted diet from 1 week before GSC inoculation, and this was maintained until the end point. For examining the combinatory effects of dietary TR with chemotherapy and anti-mitotic therapy, dietary intervention and drug treatment started from 1 week after GSC inoculation and were kept throughout the whole study. Temozolomide (20 mg per kg, HY-17364, MCE), alvocidib (10 mg per kg, HY-10006, MCE) or vehicle was administrated intraperitoneally three times per week until the end point. All mice were randomized between each cohort.

Protein synthesis measurement in vivo and in vitro

To measure protein synthesis in vivo, mice were inoculated intracranially with 10,000 GSCs into the right cerebral cortex following standard procedures and monitored until the onset of neurological symptoms. At the end point, 20 μl of 1.5 mM OPP (1407-5, Click Chemistry Tools) was injected into the contralateral hemisphere using the following coordinates: −0.5, 1.2, −2 mm. Mice were killed 30 min later, and brains were harvested and processed for cell digestion, antibody labeling and the click reaction using the Click-iT Plus Alexa Fluor 647 Picolyl Azide Toolkit (C10643, Thermo Fisher Scientific) and then subjected to flow analysis. Anti-human CD147 antibody was used to enrich for patient-derived tumor cells. Anti-human CD133 and anti-SOX2 antibodies were used to distinguish GSCs. All flow samples were run on Attune Flow Cytometers (Thermo Fisher Scientific), and data were analyzed with FlowJo version 10. To evaluate protein synthesis in vitro, GSCs were treated with 1 μM puromycin for 30 min followed by immunoblotting with anti-puromycin antibody or 15 μM OPP for 30 min following by the click reaction. Quantification of puromycin signal from immunoblots was conducted by grayscale analysis using ImageJ and normalized to the tubulin signal. Information about antibodies is available in the Nature Research Reporting Summary linked to this article.

CRISPR knockout library design, screen and data analysis

A total of 1,310 sgRNA oligonucleotides targeting 111 genes involved in tRNA modification (ten gRNA species for each gene and 200 negative controls) (Supplementary Table 1) were synthesized and cloned into the lentiCRISPR v2-Blast vector by Synbio Technologies. The lentiviral pooled library was transduced into two patient-derived GSC lines (GSC456 and GSC23) at an MOI of 0.3 and a coverage of more than 1,000-fold with two replicates for each cell line. Two days after transduction, blasticidin was added (5 mg ml⁻¹) for selection. Five days after transduction, a portion of cells with 1,000-fold coverage were harvested as the day 1 time point. The rest of the cells (with a doubling time of ~3 d) were then cultured for an additional 14 d and were harvested with 1,000-fold coverage at the day 14 time point. Genomic DNA was extracted using the QIAamp DNA Mini Kit (51304, Qiagen), and a two-step PCR procedure was performed using Q5 High-Fidelity 2X Master Mix (M0492, NEB) to amplify sgRNA sequences. Primers and reaction systems for two-step PCR are provided in Supplementary Table 1. PCR products were purified, pooled and then sequenced using the Illumina PE150 platform in CD Genomics. Raw FASTQ data were processed to remove low-quality reads. The MAGeCK algorithm⁶⁴ was used for sgRNA counting and statistical tests. P < 0.05 (two tailed) was used as the cutoff to define significant hits. sgRNA counts and MAGeCK test results are listed in Supplementary Tables 2 and 3.

Immunoblotting

Immunoblotting was performed using standard methods. Information about antibodies is available in the Nature Research Reporting Summary linked to this article.

Immunofluorescence and immunohistochemistry staining

Immunofluorescence and immunohistochemistry staining were performed as previously described using standard procedures⁶¹. Information about antibodies is available in the Nature Research Reporting Summary linked to this article.

RNA extraction and quantitative real-time PCR

Total RNA was extracted using the TRIzol Reagent (15596018, Life Technologies) and the Direct-zol RNA Miniprep kit (R2052, Zymo Research), and then we proceeded to reverse transcription using the High-Capacity cDNA Reverse Transcription Kit (4368814, Thermo Fisher Scientific). Quantitative real-time PCR was performed using the SYBR Green master mix (4309155, Thermo Fisher Scientific) on a CFX96 Touch Real-Time PCR Detection System (Bio-Rad), and data were normalized to ACTB values. The primers used in this study are listed in Supplementary Table 7.

Northern blot for detecting tRNA^Thr charging levels

The detection of tRNA^Thr charging levels was performed as described⁶⁵. Briefly, total RNA was extracted and resuspended in acidic buffer (10 mM acetate buffer, pH 4.5, 1 mM EDTA), normalized by biomass and run in a 6.5% acetate–urea PAGE gel (pH 5.0) at 4 °C. For deacylation control, samples were pretreated in 1 M Tris buffer (pH 9.0) at 37 °C for 50 min. After electrophoresis, the gel was stained with SYBR Safe (S33102, Invitrogen) to visualize total tRNA and then transferred onto Hybond-N⁺ hybridization membranes (GERPN119B, Sigma) at 4 °C. Membranes were cross-linked with UV irradiation, blocked and hybridized with 100 mM of biotin-labeled oligonucleotide probes targeting tRNA^Thr (Supplementary Table 7). HRP-Conjugated Streptavidin (1:5,000, N100, Thermo Fisher Scientific) was used for signal detection.

tRNA extraction and mass spectrometry analysis of t⁶A

The PureLink miRNA Isolation Kit (K157001, Thermo Fisher Scientific) was used for tRNA enrichment. In total, 85% ± 5% of the purified small RNA species are tRNA. tRNA-containing small RNA species (hereafter referred to as tRNA) have been proven to be highly precise for the detection of t⁶A and other tRNA modifications⁶⁶. A total of 1–3 μg purified tRNA was digested for 1 h at 37 °C using the Nucleoside Digestion Mix (M0649S, New England Biolabs), spiked with 20 ng per μg tRNA heavy standard ([¹³C₄,¹⁵N]t⁶A; T405562, Toronto Research Chemicals), extracted with methanol and then dried out using a SpeedVac. Samples were resuspended in 25 μl of 50% acetonitrile and injected onto a HILIC column (normal-phase chromatography). To detect t⁶A, we modified the MS scan to include a SIM and PRM scan targeting t⁶A (heavy and endogenous). To detect standard nucleosides within a linear range, we had to dilute the samples tenfold and inject a smaller volume (20-fold final dilution). The raw data were then searched for t⁶A and heavy t⁶A as well as a few other deoxyribonucleosides and ribonucleosides, and the corresponding peaks were integrated manually using Skyline software version 23.1.1.268. The signal of the 412 → 413 transition is defined as intact t⁶A (MS¹ level), and the signal of the 413 → 281 transition is defined as one of the fragments of t⁶A after application of collision energy (MS² level). The MS² level is typically more specific and was used when available; otherwise the MS¹ level was used. We used the heavy internal standard signal to convert the MS peak area into a relative concentration, and data were displayed in units of ng per μg tRNA. For tumor tissues, data were further normalized to adenosine peak area due to the greater deviation of total adenosine among samples. Processed MS results are shown in Supplementary Table 8.

Threonine tracing and mass spectrometry analysis of t⁶A

The customized Neurobasal medium without l-threonine (Boca Scientific) was used for tracing. The amino acid [¹³C₄,¹⁵N]l-threonine (607770, Sigma) was added into the medium at a final concentration of 800 μM, which keeps all components the same as in the Neurobasal medium (21103049, Gibco) except for isotope labeling. Cells were traced for 6 h, collected and then subjected to MS analysis of t⁶A without spiking in the heavy standard. The raw data were searched for t⁶A (unlabeled) and [¹³C₄,¹⁵N]t⁶A (labeled), and the corresponding peaks were integrated manually using Skyline software version 23.1.1.268. The labeling efficiency was calculated with the following formula: ([¹³C₄,¹⁵N]t⁶A (labeled))(labeled t⁶A + unlabeled t⁶A)⁻¹. Processed MS results are shown in Supplementary Table 8.

Proliferation and neurosphere formation assay

Cell proliferation experiments were performed by plating cells in a 96-well plate at a density of 2,000 cells per well. CellTiter-Glo (Promega) was used for measurement, and data were normalized to day 0 values. In vitro extreme limiting dilution assays were used to assess neurosphere formation capacity. Cells with decreasing numbers per well (100, 50, 20, 10, 5, 2) were plated into each well of a 96-well plate. The presence and number of neurospheres in each well were recorded 7–14 d after plating. Data were analyzed using software available at http://bioinf.wehi.edu.au/software/elda.

Click-iT EdU incorporation assay

The EdU incorporation assay was performed using the Click-iT EdU Cell Proliferation Kit (C10339, Thermo Fisher Scientific). Cells were incubated with 10 μM EdU for 2 h. Five random fields per condition were acquired with a ×40 magnification objective. The percentage of EdU-positive cells was determined according to the number of EdU-positive cells and DAPI-positive cells using ImageJ.

Cell cycle analysis

Cells were fixed with ice-cold 70% ethanol overnight at 4 °C and then washed with PBS twice. Cells were then stained with PI/RNase Staining Buffer (550825, BD Biosciences) for 30 min at room temperature and then subjected to flow analysis using the LSRFortessa (BD Biosciences). Data were analyzed with ModFit LT version 3.3.

Chromatin immunoprecipitation followed by PCR

ChIP assays was performed using the Magna ChIP A/G Chromatin Immunoprecipitation Kit (17-10085, Millipore). Briefly, cells were cross-linked with 1% formaldehyde for 10 min and quenched with 0.125 M glycine. Nuclei were extracted, lysed and sonicated to shear chromatin to 200–500 bp. The sheared chromatin was diluted and then incubated with anti-OLIG2 antibody (2 μg, AF2418, R&D Systems) or IgG overnight followed by incubation with magnetic Protein A/G beads. After washing, elution and reverse cross-linking, ChIP DNA was purified and used for PCR analysis. Primers targeting two OLIG2-binding sites on the YRDC promoter are listed in Supplementary Table 7.

Threonine quantification

Mouse serum or cell line samples were deproteinized using the 10kD Spin Column (ab93349, Abcam) and analyzed with the Threonine Assay Kit (ab239726, Abcam). Threonine levels in each cell were normalized to total protein amount. For detection of threonine levels by MS, a detailed method and processed results are provided in our prior reports^61,67.

Codon usage analysis

Homo sapiens GRCh38 CDS sequences were downloaded from the Ensembl database (Ensembl Genes 105). To avoid multiple splice variants from the same gene affecting downstream analysis, principal splice isoforms were filtered using annotations from the APPRIS database (Genecode 39, Ensembl 105)³⁸. For genes with multiple annotated isoforms, the transcript with the highest score was chosen as the representative. FASTA sequences were processed with the SeqinR package, and codon usage was calculated with CoRdon (version 1.16.0) in R. ANN codons were defined as codons that start with adenine, with N denoting any nucleotide. ANN and non-ANN codon frequencies of each gene were determined by codon counts normalized to CDS length. Codon frequencies of each ANN codon across human principal CDS were calculated by the total number of each ANN codon divided by the total number of all codons. The results of codon usage analysis are listed in Supplementary Table 6.

Tandem mass tag-labeled quantitative proteomics

Cell pellets were lysed with sodium dodecyl sulfate-free lysis buffer (20 mM Tris-HCl, pH 7.4, 100 mM KCl, 0.5% NP-40, 1× proteinase inhibitor, 1× phosphatase inhibitor). Fifty micrograms of total protein was subjected to proteomic analysis. Briefly, cysteines were reduced and alkylated with DTT and IAA. Proteins were extracted by Wessel–Flügge extraction and digested with trypsin. Peptides were labeled with TMTpro with >99% label efficiency and satisfactory stoichiometry. Samples were pooled and fractionated with a two-step fractionation scheme consisting of strong cation exchange and high-pH reversed-phase chromatography. Peptides were analyzed by an Orbitrap Fusion Lumos (Thermo Scientific) mass spectrometer operating in positive-ion MS² mode. Spectra were queried against the human proteome at 1% FDR using the Sequest HT search engine through Proteome Discoverer version 2.5. Protein intensities were derived from peptide reporter ion intensities, and a maximum spectral cofragmentation of 25% was allowed. Further data analysis was performed within the Perseus statistical environment. Protein intensities were log₂ transformed. Potential contaminants were removed. It was required that a protein be quantified in at least four replicates in at least one group. Proteins that did not meet this requirement were removed, resulting in quantitation of 7,861 proteins with 12 missing values. Intensities were normalized to the median in each experiment (therefore some values are negative). The missing values were replaced by random low-abundant signals from a normal distribution. The groups were compared using two-tailed Student’s t-test truncated with permutation-based FDR correction (P = 0.05). Differentially expressed protein lists were established by setting the fold change threshold at 2 and P < 0.05. clusterProfiler was used for GO enrichment, and FactoMineR was used for PCA analysis. The MS proteomics data have been deposited at the ProteomeXchange Consortium via the PRIDE⁶⁸ partner repository with the dataset identifier PXD049966. Processed results are shown in Supplementary Table 9.

RNA sequencing and data analysis

Total RNA was extracted and used for library construction using the Illumina TruSeq Stranded Total RNA Library Prep Kit at Novogene. The library was sequenced as a paired-end 150-bp read. Raw FASTQ reads were trimmed using Trim Galore, followed by transcript mapping with HISAT2 to the human reference genome (hg38). SAMtools was used for sorting, indexing and format conversion from SAM files. Quantification and differential analysis were performed using featureCounts and DESeq2. The DEG lists were established by setting the fold change threshold at 2 and adjusted P < 0.05 using DEseq2. clusterProfiler was used for GO enrichment, and FactoMineR was used for PCA analysis.

tRNA sequencing and data analysis

Library preparation and sequencing were performed by Arraystar. Briefly, tRNA was isolated from total RNA by electrophoresis on TBE–urea gels and then subjected to m¹A and m³C demethylation using the rtStar tRF&tiRNA Pretreatment Kit (AS-FS-005, Arraystar). Demethylated tRNA was purified and then partially hydrolyzed according to the Hydro-tRNAseq method⁶⁹. Libraries were constructed using the NEBNext Multiplex Small RNA Library Prep Set for Illumina kit (E7300L and E7850L, New England Biolabs). Size selection of ~140–155-bp PCR-amplified fragments (corresponding to a size range of ~19–35-nucleotide tRNA fragments) was performed. Libraries were sequenced on an Illumina sequencer. For data analysis, sequencing quality was examined with FastQC software, and trimmed reads (trimmed 3′-adaptor bases by Cutadapt) were aligned to the cytoplasmic mature tRNA sequences obtained from GtRNAdb³¹ using BWA software with a maximum of two mismatches. Uniquely mapped reads for each isodecoder were kept, and differentially expressed tRNA isodecoders were screened by setting the fold change threshold at 2 and adjusted P < 0.05 using the edgeR package.

Ribosome profiling and data analysis

GSC456 cells with or without YRDC knockdown or TR (4 μM, 72 h) were collected for ribosome profiling. Cells were washed twice with ice-cold PBS containing 100 μg ml⁻¹ cycloheximide, and cell pellets were snap frozen with liquid nitrogen. Ribosome footprint extraction and library construction were conducted at TB-Seq. Briefly, cells were resuspended in ice-cold lysis buffer (20 mM Tris-HCl, pH 7.4, 100 mM NaCl, 5 mM MgCl₂, 1% Triton X-100, 1 mM DTT, 20 U ml⁻¹ Turbo DNase I, 0.1% NP-40, 100 μg ml⁻¹ cycloheximide), and the soluble cytoplasmic fraction was isolated by centrifugation as described⁷⁰. Supernatants were collected and then digested with RNase I for 45 min at room temperature. Monosomes were purified by size exclusion chromatography on MicroSpin S-400 HR columns (GE Healthcare) as described⁷¹. Size selection of footprints with a length of 25–33 nucleotides was performed by electrophoresis with 15% TBE–urea gels. Libraries were prepared using a SMARTer smRNA-Seq Kit (Takara Bio) and sequenced on an Illumina NovaSeq 6000 sequencer with single-read, 1 × 50 cycles.

For data analysis, clean reads after quality control and adaptor removal were aligned to human rRNA and tRNA sequences using Bowtie (version 1.3.1) to remove contamination. Unaligned reads were further aligned to the H. sapiens transcriptome using Bowtie (version 1.3.1) with default parameters. To construct the transcriptome, the mRNA with the longest CDS was selected for each coding gene. In the case of equal CDS length, the longest transcript was selected. Data quality was evaluated, and P site offsets were determined using riboWaltz⁷². To examine the codon occupancy on the ribosome A site, clean reads were aligned to the H. sapiens genome (GRCh38) using Bowtie (version 1.3.1) with default parameters, following CONCUR pipeline⁴³ analysis and codon enrichment calculation and normalization as described⁷³. The differences of the codon occupancy on the ribosome A site between different conditions were calculated as the differences of normalized codon counts and presented as the log₂ transformed means of two biological replicates. Codons with a value greater than 0 were considered stalled under treatment. For translation efficiency analysis, ribosome footprints were counted as the number of reads aligned to the CDS region for each gene using featureCounts. Raw ribosome footprint counts and matched RNA-seq counts were subjected to Xtail analysis⁷⁴. Genes with an adjusted P value less than 0.05 and a fold change greater than 2 were considered significantly differentially translated. clusterProfiler was used for GO enrichment.

ANN codon reporters

ANN codon reporters were generated by gene synthesis with the pLV-Puro-MSCV plasmid backbone from VectorBuilder. A series of synonymous mutations were introduced into the SPC25 sequence to mutate the ANN codons without altering the corresponding amino acid sequence. SPC25 synonymous mutants with different ANN codon frequencies were infused upstream of the sequence for luciferase, followed by that for IRES-driven Renilla luciferase as an internal control. Next, 293T cells transfected to express the reporters were subjected to either YRDC knockdown or TR for 48 h. Cells were analyzed with the Dual-Luciferase Reporter Assay System (E1910, Promega). Luciferase activities normalized to Renilla luciferase activities represent the protein amounts of SPC25 mutants, and data were normalized to the control values in each reporter group. Transcript expression was analyzed by qPCR, and data were normalized in the same manner. Primers targeting Luc and Rluc are provided in Supplementary Table 7.

Single-cell RNA sequencing reanalysis

The scRNA-seq data of 65,655 cells from 28 early-passage GSC cultures derived from 24 patients and 14,207 malignant tumor cells from seven GBM tumors¹⁶ were used for reanalysis. The predetermined annotation and PCA coordinates were according to the original data and code. Translation activities were inferred using AUCell (version 1.20.2)⁷⁵ based on GOBP: positive regulation of translation. Scores were normalized between 0 and 1 by subtracting the minimum and dividing by the range, and two-tailed unpaired t-test was used for comparison. To rule out the potential bias of comparison between in vitro GSC cultures and in vivo tumor cells, we used a logistic regression classifier to identify tumor-like GSCs within all GSCs and GSC-like tumor cells within tumor cells. The first two principal components were used as inputs. On the training set, a logistic regression classifier was trained using the best hyperparameter determined from fivefold cross-validation using the caret package in R. We repeated the 80–20 train–test split randomly (stratified) 30 times and tested the accuracy and stability of these models. We found that the model was robust with high accuracy, and we picked the best-performing model (highest test accuracy) to predict the entire dataset. The correctly classified GSCs were reassigned as ‘GSCs’ (n = 64,417), and the misclassified GSCs were reassigned as ‘tumor-like GSCs’ (n = 1,238). The correctly classified tumor cells were reassigned as ‘differentiated tumor cells’ (n = 12,236), and the misclassified tumor cells were reassigned as ‘GSC-like tumor cells’ (n = 1,971).

Data mining of genome-wide CRISPR screens

Genome-wide screens^16,20,21 analyzed by the BAGEL algorithm²² were used to investigate the dependencies of YRDC in other GSCs and NSCs. For comparison between 24 GSC screens and four NSC screens, quantile normalization was performed on BF scores. Differences in average quantile normalized BF scores between GSC and NSC screens for each tRNA modification enzyme were transformed to Z scores. Screens from the DepMap project analyzed by Chronos²³ were used to investigate the dependencies of YRDC in other cell lines. Any screen with a median score less than −0.6 was considered as YRDC essential. Screens of different carcinoma cell lines and a non-transformed epithelial cell line were from a prior report²⁴. To investigate the dependencies of t⁶A downstream targets, processed data analyzed by the BAGEL algorithm were downloaded^16,20,21,39, followed by a query of each target.

GSC dataset interrogation

RNA-seq data of three pairs of GSCs and DGCs are from GSE54791 (ref. 15) and were analyzed using the limma package. To infer translation activities in GSCs, GSEA was used based on GOBP: positive regulation of translation. To identify dysregulated tRNA modifiers in GSCs, 111 tRNA modification enzymes were analyzed by setting the threshold of log₂ |fold change| to 0.8 and adjusted P < 0.05 (two tailed; Supplementary Table 4). RNA-seq data of 44 GSC samples and nine NSC samples are from GSE119834 (ref. 25) and were analyzed using the limma package. Molecular subtypes (PN, classical and mesenchymal) were determined with the ‘ssgsea.GBM.classification’ package⁵³. Three empirical classification P values for three subtypes were generated for each cell, and the molecular subtype was defined by the smallest P value. ChIP–seq data of H3K27ac from GSCs, DGCs and NSCs are from GSE54047 (ref. 15) and GSE119755 (ref. 25). ChIP–seq data of OL1G2 are from GSE54047. Integrative Genomics Viewer was used for peak visualization.

Transcription regulator analysis

TF-binding motifs on the YRDC promoter (within 2 kb upstream from the transcription start site) were scanned throughout the JASPAR 2020 CORE database (H. sapiens) using the TFBSTools package in R. A total of 568 potential TFs associated with YRDC were identified with a relative score larger than 0.8 (Supplementary Table 5). The co-upregulated TFs were established by setting the fold change threshold at 2 and the P value less than 0.05 in GSCs versus DGCs (GSE54791 (ref. 15)) and GSCs versus NSCs (GSE119834 (ref. 25)).

Drug therapeutic efficacy prediction of YRDC

Drug therapeutic efficacy prediction was performed by interrogating drug screening data from the Cancer Therapeutics Response Portal version 2 dataset⁵⁰ and mRNA expression of cell lines from the Cancer Cell Line Encyclopedia⁵¹. For each compound, the Pearson correlation coefficient and P value (two tailed) between YRDC expression and the AUC value across all cancer cell lines was calculated. The correlation coefficients for all compounds were z transformed and used for determination of gene rank. Greater coefficients predict drug resistance, and smaller coefficients predict high drug sensitivity in cells with high YRDC expression. The correlation results are listed in Supplementary Table 10.

Public glioma patient datasets

TCGA_GBM and all normal brain cortex samples from GTEx (brain, cortex; brain, anterior cingulate cortex (Ba24); and brain, frontal cortex (Ba9)) were downloaded from UCSC Xena, followed by differential expression analysis using DESeq2. The dysregulated tRNA modification enzymes were identified by setting the threshold of log₂ |fold change| > 1 and adjusted P < 0.05 (two tailed; Supplementary Table 4). ssGSEA analysis of a panel of translational regulation pathways from GOBP datasets was conducted using the GSVA package with the ssGSEA method in R. The PDC_GBM proteome⁴⁰ was used for YRDC, SPC25, RACGAP1, CIP2A, CEP55 and NCAPG protein abundance evaluation and correlation analysis. Public glioma databases from GlioVis (http://gliovis.bioinfo.cnio.es) were used for YRDC, SPC25, MASTL, RACGAP1, CIP2A, CEP55, NCAPG and OLIG2 mRNA expression and correlation analysis. To analyze the correlation between YRDC and OLIG2 expression, the Mack_GBM dataset²⁵ was also used. For survival analysis, only IDH-wild-type GBM samples were used, and data were analyzed using the Kaplan–Meier curve and the log-rank test.

Statistics and reproducibility

No statistical methods were used to predetermine sample sizes, but our sample sizes are similar to those reported in previous publications^52,61. All mice were randomized for in vivo treatment. For experiments not involving mice, cells were randomized into experimental groups. Data collection and analysis were not performed blind to the conditions of the experiments. No data were excluded from the analyses. All statistical analyses were performed using R, Python or Prism 9 software and are described in the figure legends. Data distribution was assumed to be normal, but this was not formally tested. Unless otherwise indicated, data in the figures are presented as mean ± s.d. For all statistical tests, P < 0.05 was taken to indicate statistical significance unless otherwise stated.

Extended Data

Extended Data Fig. 3 | — a. Immunoblots showing YRDC expression in GSCs with or without YRDC knockdown. b, c. Representative images of EdU incorporation assay (b) (scale bar, 50 μm) and sphere formation assay (c) (scale bar, 100 μm) in GSCs with or without YRDC knockdown. d. Immunoblots showing YRDC expression in NSCs with or without YRDC knockdown. e. Cell viability of NSCs with or without YRDC knockdown (n = 5 independent experiments). f. Immunoblots showing p-mTOR^S2448, mTOR, p-eIF2α^S51, eIF2α expression in GSCs with or without YRDC knockdown. in e, data are presented as mean ± s.d. in **a-d** and f, immunoblots and images are representative from three independent experiments with similar results. Two-way ANOVA followed by multiple comparison for e.

Extended Data Fig. 4 | — a, b. Immunoblots showing puromycin incorporation, p-mTOR^S2448, mTOR, p-eIF2α^S51, eIF2α expression in GSC456 cultured in indicated media for 72 h. Ctrl, 800 μM threonine and 400 μM arginine. 5xThr, 4000 μM threonine. 5xArg, 2000 μM arginine. TR, 8 μM threonine. AR, 4 μM arginine. c. MS analysis of intracellular threonine levels in indicated cells. Left, n = 5 biologically independent samples per group; right, n = 3 biologically independent samples per group. d, e. QPCR analysis of *SLC1A4* (d, n = 3 independent experiments) and *SLC1A5* (e, n = 3 independent experiments) in matched GSCs and DGCs. f. Graphic illustration of threonine metabolism in human. Red indicates enriched and green indicates depleted in GSCs. Crosses indicate nonfunctional catabolic pathways in humans. g. QPCR analysis of *SDS* in matched GSCs and DGCs (n = 3 independent experiments). In c-e, and g, data are presented as mean ± s.d. In a and b, immunoblots are representative from three independent experiments with similar results. One-way ANOVA followed by multiple comparison for c. Two-way ANOVA followed by multiple comparison for d, e and g.

Extended Data Fig. 5 | — a, b. Volcano plot showing the differentially expressed gene analysis of ANN and non-ANN decoding tRNA isodecoders in tRNA-seq data of GSC456 with indicated treatments. Cut-off: |log₂FC|>1, adjusted P < 0.05. c. Expression of ANN decoding and non-ANN decoding tRNA isodecoders in tRNA-seq of GSC456 upon TR. d. GO enrichment analysis of downregulated proteins under YRDC knockdown in proteomics of GSC456. e. ANN codon frequencies of differentially expressed proteins upon TR in proteomics of GSC456. f. GO enrichment analysis of proteins downregulated only in YRDC knockdown (left), and proteins downregulated only in TR (right) in proteomics of GSC456. g, h. Flowcytometric analysis of cell cycle in GSC456 with indicated treatments (n = 3 independent experiments). i,j. QPCR analysis of indicated t⁶A targets in GSCs with indicated treatments (n = 3 independent experiments). In c and e, boxes represent data within the 25th to 75th percentiles, whiskers depict the range of all data points and horizontal lines within boxes represent median values. In **g-j**, data are presented as mean ± s.d. In a-c, tRNA-seq data are from 3 biologically independent samples per group. In e, proteomic data are from 4 biologically independent samples per group. Two-tailed unpaired t test for c. Overrepresentation test corrected by FDR for d and f. Two-tailed Mann-Whitney test for e. Two-way ANOVA followed by multiple comparison for **g-j**.

Extended Data Fig. 6 | — a. BF scores of *SPC25, MASTL, RACGAP1, CIP2A, CEP55* and *NCAPG* in genome wide CRISPR knockout screens of GSCs and GBM cell lines (n = 24 independent screens per group). The greater positive BF score indicates higher confidence of essentiality. Data are presented as mean ± s.d. b. Correlations between YRDC and SPC25, RACGAP1, CIP2A, CEP55, NCAPG protein abundance in proteomics of PDC_GBM (n = 99). Red lines show linear regression of the data. MASTL was not detected in this dataset. c. Correlation heatmap of targets from b. The size of circle indicates the value of correlation coefficient. Red colour represents positive correlation and blue colour represents negative correlation. d, e. Violin plot of *SPC25, MASTL, RACGAP1, CIP2A, CEP55, NCAPG* expression (log₂(count+0.5)) in RNA-seq data of TCGA (Grade II, n = 226; Grade III, n = 244; Grade IV, n = 150) (d) and CGGA (Grade II, n = 232; Grade III, n = 194; Grade IV, n = 225) (e) datasets. Violin plots represent the overall distribution of data points, horizontal lines show median, upper, and lower quartiles. In b, d and e, the number of n indicates patients. Two-tailed Pearson correlation for b and c. One-way ANOVA followed by multiple comparison for d and e.

Extended Data Fig. 7 | — a. Distribution of read length of ribosome footprints in ribosome profiling of GSC456. Data are presented as mean ± s.e.m. b. Heatmap showing the 3-nt periodicity of ribosome footprints with different read lengths (25-35 nt) on CDS. Data are plotted as the signal of the first nucleotide of ribosome P site on three-nucleotide (0, 1, 2) codon frames. The small table on the right provides detailed experimental setup for each group. c. Volcano plot showing the differential translation efficiency (TE) of GSC456 upon YRDC knockdown (cut-off: |log₂FC|>1, adjust P < 0.05). Red indicates genes with upregulated TE and blue indicates genes with downregulated TE. d. GO enrichment analysis of genes with downregulated TE upon YRDC knockdown. e. Non-ANN codon occupancy alteration upon YRDC knockdown and TR at ribosome A-site. Black dots and labels show overlapped stalled codons under both conditions. f-h. Overlapped-non-ANN codon frequencies of differentially expressed proteins in proteomics of GSC456 with indicated treatments. i. Distribution of overlapped-non-ANN codon frequencies in CDS. Yellow dots indicate 6 t⁶A targets.j, k. QPCR analysis of transcript expression of indicated reporters in 293 T with indicated treatments (n = 3 independent experiments). In f-h, boxes represent data within the 25th to 75th percentiles, whiskers depict the range of all data points and horizontal lines within boxes represent median values. In j and k, data are presented as mean ± s.d. In a-e, ribosome profiling data are from 2 biologically independent samples per group. In f-h, proteomics data are from 4 biologically independent samples per group. Overrepresentation test corrected by FDR for d. Two-tailed Mann-Whitney test for f-h. Two-tailed unpaired t test forj and k.

Extended Data Fig. 8 | — a, b. QPCR analysis (a, n = 3 independent experiments) and immunoblots (b) showing YRDC expression in two GSCs with or without YRDC knockdown for in vivo tumorigenesis. c. Representative in vivo bioluminescence images of tumour-bearing mice derived from GSC468 with or without YRDC knockdown at indicated timepoint. Scale bar, 1 cm. d, e. Representative images of H&E-stained brain sections of tumour-bearing mice derived from GSC456 (d) and GSC468 (e) at indicated timepoint. Scale bar, 1 mm. f-i. Representative images (f, h) and quantification (g, i, n = 15 randomly selected fields examined over 3 mice per group) of phosphorylated histone H3 (p-H3, red) in brain sections derived from indicated xenografts. α-tubulin (green) shows the cytoplasm and DAPI (blue) shows the nucleus. Scale bar, 20 μm. In a, g and i, data are presented as mean ± s.d. In b, immunoblots are representative from three independent experiments with similar results. In c-e, images are representative from 5 mice. One-way ANOVA followed by multiple comparison for a, g and i.

Extended Data Fig. 9 | — a, b. Body weight (a, n = 5 mice per group) and serum threonine levels (b, n = 3 mice in control group at Day 14, n = 4 mice per group in others) of tumour-free mice fed with indicated diets at indicated time points. c, d. Weight (c, n = 5 mice per group) and representative H&E-stained images (d) of indicated organs from tumour-free mice fed with indicated diets for 30 days. Scale bar, 100 μm. e. Representative H&E-stained brain section of tumour-free mice fed with indicated diets. Rectangles show the region of interest for NSC analysis. Scale bar, 2 mm. f-i. Representative images (f, h) and quantification (g, i, n = 10 randomly selected fields examined over 3 mice per group) of V-SVZ GFAP⁺ cells (NSCs) and DCX⁺ cells (neuroblasts) in tumour-free mice fed with indicated diets for 30 days. Rectangles show the enlarged region. V, ventricle. V-SVZ, ventricular–subventricular zone. Scale bar, 100 μm. j-m. Representative images (j, l) and quantification (k, m, n = 20 randomly selected fields examined over 4 mice per group) of phosphorylated histone H3 (p-H3, red) in brain sections of indicated xenografts fed with indicated diets. α-tubulin (green) shows the cytoplasm and DAPI (blue) shows the nucleus. Scale bar, 20 μm. n. Immunoblots showing p-mTOR^S2448, mTOR, p-eIF2α^S51, eIF2α expression in tumour tissues from indicated xenografts fed with indicated diets. Each lane represents a sample from one mouse. o-q. Representative histogram plot (o), quantification of the percentage of OPP-hi tumour cells (p, n = 4 mice per group), and quantification of the median fluorescence intensity (MFI) (q, n = 4 mice per group) of OPP flow cytometric analysis in GSC456 derived xenograft with indicated treatments. In **a-c**, g, i, k, m, p and q, data are presented as mean ± s.d. In n, immunoblots are from three independent experiments with similar results. Two-way ANOVA followed by multiple comparison for a and b. Two-tailed unpaired t test for c, g, i, k and m. One-way ANOVA followed by multiple comparison for p and q.

Extended Data Fig. 10 | — a. Representative in vivo bioluminescence images of GSC468-bearing mice with indicated treatments at indicated timepoints. Scale bar, 1 cm. b-e. Violin plot of *YRDC* expression in RNA-seq data of TCGA_LGG, GBM (Grade II, n = 226; Grade III, n = 244; Grade IV, n = 150) (b), CGGA (Grade II, n = 232; Grade III, n = 194; Grade IV, n = 225) (c), Rembrandt (Grade I-II, n = 100; Grade III, n = 85; Grade IV, n = 130) (d) and Gravendeel (Grade I-II, n = 32; Grade III, n = 85; Grade IV, n = 159) (e) datasets. f. Violin plot of *YRDC* expression in different subtypes in RNA-seq data of TCGA_GBM. PN, proneural, n = 46; CL, classical, n = 59; MES, mesenchymal, n = 51. g. Correlations between *YRDC* and *PDGFRA* expressions in RNA-seq data of TCGA_GBM (n = 160). Red line shows linear regression. **h-j**. Violin plot of *YRDC* expression in different *PDGFRA* status (diploid, n = 106; gain or amplification, n = 30) (h), *IDH* status (WT, wildtype, n = 142; mutant, n = 8) (i) and *MGMT* methylation (methylated, n = 56; unmethylated, n = 66) (j) in RNA-seq data of TCGA_GBM. k. Violin plot of *YRDC* expression (log₂(FPKM)) in different subtypes in RNA-seq data of GSCs (GSE119834). PN, proneural, n = 10; CL, classical, n = 16; MES, mesenchymal, n = 18. The number of n indicates biologically independent cell lines. In b-f and h-k, violin plots represent the overall distribution of data points. In a, images are representative of 5 mice. In **b-j**, the number of n indicates patients, and the expressions are presented as log₂(count+0.5). One-way ANOVA followed by multiple comparison for b-f and k. Two-tailed Pearson correlation for g. Two-tailed unpaired t test for **h-j**.

Supplementary Material

NIHMS2032325-supplement-s1.xlsx^{(10MB, xlsx)}

NIHMS2032325-supplement-s2.pdf^{(5MB, pdf)}

Acknowledgements

We thank the Metabolomics and Lipidomics Core for MS analysis, the Biospecimen Core for histology analysis and the Flow Cytometry Core Facility at the University of Pittsburgh. We thank the Proteomics Resource Centre at the Rockefeller University for MS analysis of t⁶A and TMT-labeled quantitative proteomics. This work was supported by the National Natural Science Distinguished Youth Foundation of China (82125024 to N.Z.), Special Funds of the National Natural Science Foundation of China (82341007 to N.Z.), the Guangdong Province Regional Joint Fund—Key Project (2022B1515120023 to N.Z.), the Major Program of the National Natural Science Foundation of China (82192894 to N.Z.), the Science and Technology Planning Project of Guangzhou (202103000019 to N.Z.), start-up funds from the University of Pittsburgh (to J.N.R.), NIH grants (R35CA197718, R01CA238662, R01CA268634 and R01NS103434 to J.N.R.), the Defense Health Agency (HT9425-23-1-0689 to J.N.R.), and the American Cancer Society Lisa Dean Mosley Cancer Stem Cell Consortium (CSCC-LEAD-22-186-01-CSCC to J.N.R.).

Footnotes

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Code availability

No custom code was generated for this study.

Competing interests

The authors declare no competing interests.

Extended data is available for this paper at https://doi.org/10.1038/s43018-024-00748-7.

Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s43018-024-00748-7.

Data availability

RNA-seq, tRNA-seq and ribosome profiling data that support the findings of this study have been deposited in the Gene Expression Omnibus under accession code GSE229673. The MS proteomics data have been deposited at the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD049966. sgRNA counts for screening and processed MS data are provided in the Supplementary Tables. Human glioma transcriptomic data were derived from the TCGA Research Network (http://cancergenome.nih.gov/) and the CGGA database (http://www.cgga.org.cn/). All other data supporting the findings of this study are available from the corresponding author on reasonable request. Source data are provided with this paper.

References

1.Tan AC et al. Management of glioblastoma: state of the art and future directions. CA Cancer J. Clin. 70, 299–312 (2020). [DOI] [PubMed] [Google Scholar]
2.Mathur R et al. Glioblastoma evolution and heterogeneity from a 3D whole-tumor perspective. Cell 187, 446–463 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
3.Gimple RC, Yang K, Halbert ME, Agnihotri S & Rich JN Brain cancer stem cells: resilience through adaptive plasticity and hierarchical heterogeneity. Nat. Rev. Cancer 22, 497–514 (2022). [DOI] [PubMed] [Google Scholar]
4.Truitt ML & Ruggero D New frontiers in translational control of the cancer genome. Nat. Rev. Cancer 16, 288–304 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
5.Saba JA, Liakath-Ali K, Green R & Watt FM Translational control of stem cell function. Nat. Rev. Mol. Cell Biol. 22, 671–690 (2021). [DOI] [PubMed] [Google Scholar]
6.Blanco S et al. Stem cell function and stress response are controlled by protein synthesis. Nature 534, 335–340 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
7.Signer RA, Magee JA, Salic A & Morrison SJ Haematopoietic stem cells require a highly regulated protein synthesis rate. Nature 509, 49–54 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Suzuki T The expanding world of tRNA modifications and their disease relevance. Nat. Rev. Mol. Cell Biol. 22, 375–392 (2021). [DOI] [PubMed] [Google Scholar]
9.Dedon PC & Begley TJ Dysfunctional tRNA reprogramming and codon-biased translation in cancer. Trends Mol. Med 28, 964–978 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Thiaville PC, Iwata-Reuyl D & de Crecy-Lagard V Diversity of the biosynthesis pathway for threonylcarbamoyladenosine (t⁶A), a universal modification of tRNA. RNA Biol. 11, 1529–1539 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
11.Arrondel C et al. Defects in t⁶A tRNA modification due to GON7 and YRDC mutations lead to Galloway–Mowat syndrome. Nat. Commun 10, 3967 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Hidalgo San Jose L & Signer RAJ Cell-type-specific quantification of protein synthesis in vivo. Nat. Protoc 14, 441–460 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Zhai K et al. Pharmacological inhibition of BACE1 suppresses glioblastoma growth by stimulating macrophage phagocytosis of tumor cells. Nat. Cancer 2, 1136–1151 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
14.Lathia JD, Mack SC, Mulkearns-Hubert EE, Valentim CL & Rich JN Cancer stem cells in glioblastoma. Genes Dev. 29, 1203–1217 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Suva ML et al. Reconstructing and reprogramming the tumor-propagating potential of glioblastoma stem-like cells. Cell 157, 580–594 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Richards LM et al. Gradient of developmental and injury response transcriptional states defines functional vulnerabilities underpinning glioblastoma heterogeneity. Nat. Cancer 2, 157–173 (2021). [DOI] [PubMed] [Google Scholar]
17.Batlle E & Clevers H Cancer stem cells revisited. Nat. Med 23, 1124–1134 (2017). [DOI] [PubMed] [Google Scholar]
18.Asghar U, Witkiewicz AK, Turner NC & Knudsen ES The history and future of targeting cyclin-dependent kinases in cancer therapy. Nat. Rev. Drug Discov 14, 130–146 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
19.de Crecy-Lagard V et al. Matching tRNA modifications in humans to their known and predicted enzymes. Nucleic Acids Res. 47, 2143–2159 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
20.MacLeod G et al. Genome-wide CRISPR–Cas9 screens expose genetic vulnerabilities and mechanisms of temozolomide sensitivity in glioblastoma stem cells. Cell Rep. 27, 971–986 (2019). [DOI] [PubMed] [Google Scholar]
21.Toledo CM et al. Genome-wide CRISPR–Cas9 screens reveal loss of redundancy between PKMYT1 and WEE1 in glioblastoma stem-like cells. Cell Rep. 13, 2425–2439 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
22.Hart T, Brown KR, Sircoulomb F, Rottapel R & Moffat J Measuring error rates in genomic perturbation screens: gold standards for human functional genomics. Mol. Syst. Biol 10, 733 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Dempster JM et al. Chronos: a cell population dynamics model of CRISPR experiments that improves inference of gene fitness effects. Genome Biol 22, 343 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
24.Hart T et al. Evaluation and design of genome-wide CRISPR/SpCas9 knockout screens. G3 7, 2719–2727 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
25.Mack SC et al. Chromatin landscapes reveal developmentally encoded transcriptional states that define human glioblastoma. J. Exp. Med 216, 1071–1090 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
26.Ma XM & Blenis J Molecular mechanisms of mTOR-mediated translational control. Nat. Rev. Mol. Cell Biol 10, 307–318 (2009). [DOI] [PubMed] [Google Scholar]
27.Pakos-Zebrucka K et al. The integrated stress response. EMBO Rep. 17, 1374–1395 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Vellai T How the amino acid leucine activates the key cell-growth regulator mTOR. Nature 596, 192–194 (2021). [DOI] [PubMed] [Google Scholar]
29.Chantranupong L et al. The CASTOR proteins are arginine sensors for the mTORC1 pathway. Cell 165, 153–164 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
30.Saxton RA, Chantranupong L, Knockenhauer KE, Schwartz TU & Sabatini DM Mechanism of arginine sensing by CASTOR1 upstream of mTORC1. Nature 536, 229–233 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
31.Chan PP & Lowe TM GtRNAdb 2.0: an expanded database of transfer RNA genes identified in complete and draft genomes. Nucleic Acids Res. 44, D184–D189 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
32.Kandasamy P, Gyimesi G, Kanai Y & Hediger MA Amino acid transporters revisited: new views in health and disease. Trends Biochem. Sci 43, 752–789 (2018). [DOI] [PubMed] [Google Scholar]
33.Chen G & Wang J Threonine metabolism and embryonic stem cell self-renewal. Curr. Opin. Clin. Nutr. Metab. Care 17, 80–85 (2014). [DOI] [PubMed] [Google Scholar]
34.Edgar AJ The human l-threonine 3-dehydrogenase gene is an expressed pseudogene. BMC Genet. 3, 18 (2002). [DOI] [PMC free article] [PubMed] [Google Scholar]
35.Ogawa H, Gomi T & Fujioka M Serine hydroxymethyltransferase and threonine aldolase: are they identical? Int. J. Biochem. Cell Biol 32, 289–301 (2000). [DOI] [PubMed] [Google Scholar]
36.Wang JT et al. Commonality and diversity in tRNA substrate recognition in t⁶A biogenesis by eukaryotic KEOPSs. Nucleic Acids Res. 50, 2223–2239 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
37.Murphy FVT, Ramakrishnan V, Malkiewicz A & Agris PF The role of modifications in codon discrimination by tRNA^Lys_UUU. Nat. Struct. Mol. Biol 11, 1186–1191 (2004). [DOI] [PubMed] [Google Scholar]
38.Rodriguez JM et al. APPRIS: selecting functionally important isoforms. Nucleic Acids Res 50, D54–D59 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
39.Kim E Table_DepMap2018Q4_BAGEL1.txt. figshare 10.6084/m9.figshare.12317654.v1 (2020). [DOI] [Google Scholar]
40.Wang LB et al. Proteogenomic and metabolomic characterization of human glioblastoma. Cancer Cell 39, 509–528 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
41.Brar GA & Weissman JS Ribosome profiling reveals the what, when, where and how of protein synthesis. Nat. Rev. Mol. Cell Biol 16, 651–664 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
42.Ingolia NT, Ghaemmaghami S, Newman JR & Weissman JS Genome-wide analysis in vivo of translation with nucleotide resolution using ribosome profiling. Science 324, 218–223 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]
43.Frye M & Bornelov S CONCUR: quick and robust calculation of codon usage from ribosome profiling data. Bioinformatics 37, 717–719 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
44.VanInsberghe M, van den Berg J, Andersson-Rolf A, Clevers H & van Oudenaarden A Single-cell Ribo-seq reveals cell cycle-dependent translational pausing. Nature 597, 561–565 (2021). [DOI] [PubMed] [Google Scholar]
45.Hsu DJ et al. Arginine limitation drives a directed codon-dependent DNA sequence evolution response in colorectal cancer cells. Sci. Adv 9, eade9120 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
46.National Research Council (US) Subcommittee on Laboratory Animal Nutrition. Nutrient Requirements of Laboratory Animals: Fourth Revised Edition, 1995 (National Academies, 1995). [PubMed] [Google Scholar]
47.Doetsch F, Caille I, Lim DA, Garcia-Verdugo JM & Alvarez-Buylla A Subventricular zone astrocytes are neural stem cells in the adult mammalian brain. Cell 97, 703–716 (1999). [DOI] [PubMed] [Google Scholar]
48.Doetsch F & Alvarez-Buylla A Network of tangential pathways for neuronal migration in adult mammalian brain. Proc. Natl Acad. Sci. USA 93, 14895–14900 (1996). [DOI] [PMC free article] [PubMed] [Google Scholar]
49.Chen J et al. A restricted cell population propagates glioblastoma growth after chemotherapy. Nature 488, 522–526 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
50.Seashore-Ludlow B et al. Harnessing connectivity in a large-scale small-molecule sensitivity dataset. Cancer Discov. 5, 1210–1223 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
51.Barretina J et al. The Cancer Cell Line Encyclopedia enables predictive modelling of anticancer drug sensitivity. Nature 483, 603–607 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
52.Qiu Z et al. Transcription elongation machinery is a druggable dependency and potentiates immunotherapy in glioblastoma stem cells. Cancer Discov. 12, 502–521 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
53.Wang Q et al. Tumor evolution of glioma-intrinsic gene expression subtypes associates with immunological changes in the microenvironment. Cancer Cell 32, 42–56 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
54.Verhaak RG et al. Integrated genomic analysis identifies clinically relevant subtypes of glioblastoma characterized by abnormalities in PDGFRA, IDH1, EGFR, and NF1. Cancer Cell 17, 98–110 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
55.Tirosh I et al. Single-cell RNA-seq supports a developmental hierarchy in human oligodendroglioma. Nature 539, 309–313 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
56.Venteicher AS et al. Decoupling genetics, lineages, and microenvironment in IDH-mutant gliomas by single-cell RNA-seq. Science 355, eaai8478 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
57.Joshi K, Bhatt MJ & Farabaugh PJ Codon-specific effects of tRNA anticodon loop modifications on translational misreading errors in the yeast Saccharomyces cerevisiae. Nucleic Acids Res. 46, 10331–10339 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
58.Bulmer M Coevolution of codon usage and transfer RNA abundance. Nature 325, 728–730 (1987). [DOI] [PubMed] [Google Scholar]
59.Lin H et al. CO₂-sensitive tRNA modification associated with human mitochondrial disease. Nat. Commun 9, 1875 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
60.Shyh-Chang N et al. Influence of threonine metabolism on S-adenosylmethionine and histone methylation. Science 339, 222–226 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
61.Yuan H et al. Lysine catabolism reprograms tumour immunity through histone crotonylation. Nature 617, 818–826 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
62.Bao S et al. Glioma stem cells promote radioresistance by preferential activation of the DNA damage response. Nature 444, 756–760 (2006). [DOI] [PubMed] [Google Scholar]
63.Gimple RC et al. Sorting nexin 10 sustains PDGF receptor signaling in glioblastoma stem cells via endosomal protein sorting. JCI Insight 8, e158077 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
64.Li W et al. MAGeCK enables robust identification of essential genes from genome-scale CRISPR/Cas9 knockout screens. Genome Biol. 15, 554 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
65.Pavlova NN et al. Translation in amino-acid-poor environments is limited by tRNA^Gln charging. eLife 9, e62307 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
66.Chan CT et al. A quantitative systems approach reveals dynamic control of tRNA modifications during cellular stress. PLoS Genet. 6, e1001247 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
67.Wang X et al. Purine synthesis promotes maintenance of brain tumor initiating cells in glioma. Nat. Neurosci 20, 661–673 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
68.Perez-Riverol Y et al. The PRIDE database resources in 2022: a hub for mass spectrometry-based proteomics evidences. Nucleic Acids Res. 50, D543–D552 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
69.Gogakos T et al. Characterizing expression and processing of precursor and mature human tRNAs by hydro-tRNAseq and PAR-CLIP. Cell Rep. 20, 1463–1475 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
70.Ingolia NT, Brar GA, Rouskin S, McGeachy AM & Weissman JS The ribosome profiling strategy for monitoring translation in vivo by deep sequencing of ribosome-protected mRNA fragments. Nat. Protoc 7, 1534–1550 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
71.Aeschimann F, Xiong J, Arnold A, Dieterich C & Grosshans H Transcriptome-wide measurement of ribosomal occupancy by ribosome profiling. Methods 85, 75–89 (2015). [DOI] [PubMed] [Google Scholar]
72.Lauria F et al. riboWaltz: optimization of ribosome P-site positioning in ribosome profiling data. PLoS Comput. Biol 14, e1006169 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
73.Bornelov S, Selmi T, Flad S, Dietmann S & Frye M Codon usage optimization in pluripotent embryonic stem cells. Genome Biol. 20, 119 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
74.Xiao Z, Zou Q, Liu Y & Yang X Genome-wide assessment of differential translations with ribosome profiling data. Nat. Commun 7, 11194 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
75.Aibar S et al. SCENIC: single-cell regulatory network inference and clustering. Nat. Methods 14, 1083–1086 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

NIHMS2032325-supplement-s1.xlsx^{(10MB, xlsx)}

NIHMS2032325-supplement-s2.pdf^{(5MB, pdf)}

Data Availability Statement

[R1] 1.Tan AC et al. Management of glioblastoma: state of the art and future directions. CA Cancer J. Clin. 70, 299–312 (2020). [DOI] [PubMed] [Google Scholar]

[R2] 2.Mathur R et al. Glioblastoma evolution and heterogeneity from a 3D whole-tumor perspective. Cell 187, 446–463 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R3] 3.Gimple RC, Yang K, Halbert ME, Agnihotri S & Rich JN Brain cancer stem cells: resilience through adaptive plasticity and hierarchical heterogeneity. Nat. Rev. Cancer 22, 497–514 (2022). [DOI] [PubMed] [Google Scholar]

[R4] 4.Truitt ML & Ruggero D New frontiers in translational control of the cancer genome. Nat. Rev. Cancer 16, 288–304 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R5] 5.Saba JA, Liakath-Ali K, Green R & Watt FM Translational control of stem cell function. Nat. Rev. Mol. Cell Biol. 22, 671–690 (2021). [DOI] [PubMed] [Google Scholar]

[R6] 6.Blanco S et al. Stem cell function and stress response are controlled by protein synthesis. Nature 534, 335–340 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R7] 7.Signer RA, Magee JA, Salic A & Morrison SJ Haematopoietic stem cells require a highly regulated protein synthesis rate. Nature 509, 49–54 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R8] 8.Suzuki T The expanding world of tRNA modifications and their disease relevance. Nat. Rev. Mol. Cell Biol. 22, 375–392 (2021). [DOI] [PubMed] [Google Scholar]

[R9] 9.Dedon PC & Begley TJ Dysfunctional tRNA reprogramming and codon-biased translation in cancer. Trends Mol. Med 28, 964–978 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R10] 10.Thiaville PC, Iwata-Reuyl D & de Crecy-Lagard V Diversity of the biosynthesis pathway for threonylcarbamoyladenosine (t⁶A), a universal modification of tRNA. RNA Biol. 11, 1529–1539 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R11] 11.Arrondel C et al. Defects in t⁶A tRNA modification due to GON7 and YRDC mutations lead to Galloway–Mowat syndrome. Nat. Commun 10, 3967 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R12] 12.Hidalgo San Jose L & Signer RAJ Cell-type-specific quantification of protein synthesis in vivo. Nat. Protoc 14, 441–460 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R13] 13.Zhai K et al. Pharmacological inhibition of BACE1 suppresses glioblastoma growth by stimulating macrophage phagocytosis of tumor cells. Nat. Cancer 2, 1136–1151 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R14] 14.Lathia JD, Mack SC, Mulkearns-Hubert EE, Valentim CL & Rich JN Cancer stem cells in glioblastoma. Genes Dev. 29, 1203–1217 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R15] 15.Suva ML et al. Reconstructing and reprogramming the tumor-propagating potential of glioblastoma stem-like cells. Cell 157, 580–594 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R16] 16.Richards LM et al. Gradient of developmental and injury response transcriptional states defines functional vulnerabilities underpinning glioblastoma heterogeneity. Nat. Cancer 2, 157–173 (2021). [DOI] [PubMed] [Google Scholar]

[R17] 17.Batlle E & Clevers H Cancer stem cells revisited. Nat. Med 23, 1124–1134 (2017). [DOI] [PubMed] [Google Scholar]

[R18] 18.Asghar U, Witkiewicz AK, Turner NC & Knudsen ES The history and future of targeting cyclin-dependent kinases in cancer therapy. Nat. Rev. Drug Discov 14, 130–146 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R19] 19.de Crecy-Lagard V et al. Matching tRNA modifications in humans to their known and predicted enzymes. Nucleic Acids Res. 47, 2143–2159 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R20] 20.MacLeod G et al. Genome-wide CRISPR–Cas9 screens expose genetic vulnerabilities and mechanisms of temozolomide sensitivity in glioblastoma stem cells. Cell Rep. 27, 971–986 (2019). [DOI] [PubMed] [Google Scholar]

[R21] 21.Toledo CM et al. Genome-wide CRISPR–Cas9 screens reveal loss of redundancy between PKMYT1 and WEE1 in glioblastoma stem-like cells. Cell Rep. 13, 2425–2439 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R22] 22.Hart T, Brown KR, Sircoulomb F, Rottapel R & Moffat J Measuring error rates in genomic perturbation screens: gold standards for human functional genomics. Mol. Syst. Biol 10, 733 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R23] 23.Dempster JM et al. Chronos: a cell population dynamics model of CRISPR experiments that improves inference of gene fitness effects. Genome Biol 22, 343 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R24] 24.Hart T et al. Evaluation and design of genome-wide CRISPR/SpCas9 knockout screens. G3 7, 2719–2727 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R25] 25.Mack SC et al. Chromatin landscapes reveal developmentally encoded transcriptional states that define human glioblastoma. J. Exp. Med 216, 1071–1090 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R26] 26.Ma XM & Blenis J Molecular mechanisms of mTOR-mediated translational control. Nat. Rev. Mol. Cell Biol 10, 307–318 (2009). [DOI] [PubMed] [Google Scholar]

[R27] 27.Pakos-Zebrucka K et al. The integrated stress response. EMBO Rep. 17, 1374–1395 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R28] 28.Vellai T How the amino acid leucine activates the key cell-growth regulator mTOR. Nature 596, 192–194 (2021). [DOI] [PubMed] [Google Scholar]

[R29] 29.Chantranupong L et al. The CASTOR proteins are arginine sensors for the mTORC1 pathway. Cell 165, 153–164 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R30] 30.Saxton RA, Chantranupong L, Knockenhauer KE, Schwartz TU & Sabatini DM Mechanism of arginine sensing by CASTOR1 upstream of mTORC1. Nature 536, 229–233 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R31] 31.Chan PP & Lowe TM GtRNAdb 2.0: an expanded database of transfer RNA genes identified in complete and draft genomes. Nucleic Acids Res. 44, D184–D189 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R32] 32.Kandasamy P, Gyimesi G, Kanai Y & Hediger MA Amino acid transporters revisited: new views in health and disease. Trends Biochem. Sci 43, 752–789 (2018). [DOI] [PubMed] [Google Scholar]

[R33] 33.Chen G & Wang J Threonine metabolism and embryonic stem cell self-renewal. Curr. Opin. Clin. Nutr. Metab. Care 17, 80–85 (2014). [DOI] [PubMed] [Google Scholar]

[R34] 34.Edgar AJ The human l-threonine 3-dehydrogenase gene is an expressed pseudogene. BMC Genet. 3, 18 (2002). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R35] 35.Ogawa H, Gomi T & Fujioka M Serine hydroxymethyltransferase and threonine aldolase: are they identical? Int. J. Biochem. Cell Biol 32, 289–301 (2000). [DOI] [PubMed] [Google Scholar]

[R36] 36.Wang JT et al. Commonality and diversity in tRNA substrate recognition in t⁶A biogenesis by eukaryotic KEOPSs. Nucleic Acids Res. 50, 2223–2239 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R37] 37.Murphy FVT, Ramakrishnan V, Malkiewicz A & Agris PF The role of modifications in codon discrimination by tRNA^Lys_UUU. Nat. Struct. Mol. Biol 11, 1186–1191 (2004). [DOI] [PubMed] [Google Scholar]

[R38] 38.Rodriguez JM et al. APPRIS: selecting functionally important isoforms. Nucleic Acids Res 50, D54–D59 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R39] 39.Kim E Table_DepMap2018Q4_BAGEL1.txt. figshare 10.6084/m9.figshare.12317654.v1 (2020). [DOI] [Google Scholar]

[R40] 40.Wang LB et al. Proteogenomic and metabolomic characterization of human glioblastoma. Cancer Cell 39, 509–528 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R41] 41.Brar GA & Weissman JS Ribosome profiling reveals the what, when, where and how of protein synthesis. Nat. Rev. Mol. Cell Biol 16, 651–664 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R42] 42.Ingolia NT, Ghaemmaghami S, Newman JR & Weissman JS Genome-wide analysis in vivo of translation with nucleotide resolution using ribosome profiling. Science 324, 218–223 (2009). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R43] 43.Frye M & Bornelov S CONCUR: quick and robust calculation of codon usage from ribosome profiling data. Bioinformatics 37, 717–719 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R44] 44.VanInsberghe M, van den Berg J, Andersson-Rolf A, Clevers H & van Oudenaarden A Single-cell Ribo-seq reveals cell cycle-dependent translational pausing. Nature 597, 561–565 (2021). [DOI] [PubMed] [Google Scholar]

[R45] 45.Hsu DJ et al. Arginine limitation drives a directed codon-dependent DNA sequence evolution response in colorectal cancer cells. Sci. Adv 9, eade9120 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R46] 46.National Research Council (US) Subcommittee on Laboratory Animal Nutrition. Nutrient Requirements of Laboratory Animals: Fourth Revised Edition, 1995 (National Academies, 1995). [PubMed] [Google Scholar]

[R47] 47.Doetsch F, Caille I, Lim DA, Garcia-Verdugo JM & Alvarez-Buylla A Subventricular zone astrocytes are neural stem cells in the adult mammalian brain. Cell 97, 703–716 (1999). [DOI] [PubMed] [Google Scholar]

[R48] 48.Doetsch F & Alvarez-Buylla A Network of tangential pathways for neuronal migration in adult mammalian brain. Proc. Natl Acad. Sci. USA 93, 14895–14900 (1996). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R49] 49.Chen J et al. A restricted cell population propagates glioblastoma growth after chemotherapy. Nature 488, 522–526 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R50] 50.Seashore-Ludlow B et al. Harnessing connectivity in a large-scale small-molecule sensitivity dataset. Cancer Discov. 5, 1210–1223 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R51] 51.Barretina J et al. The Cancer Cell Line Encyclopedia enables predictive modelling of anticancer drug sensitivity. Nature 483, 603–607 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R52] 52.Qiu Z et al. Transcription elongation machinery is a druggable dependency and potentiates immunotherapy in glioblastoma stem cells. Cancer Discov. 12, 502–521 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R53] 53.Wang Q et al. Tumor evolution of glioma-intrinsic gene expression subtypes associates with immunological changes in the microenvironment. Cancer Cell 32, 42–56 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R54] 54.Verhaak RG et al. Integrated genomic analysis identifies clinically relevant subtypes of glioblastoma characterized by abnormalities in PDGFRA, IDH1, EGFR, and NF1. Cancer Cell 17, 98–110 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R55] 55.Tirosh I et al. Single-cell RNA-seq supports a developmental hierarchy in human oligodendroglioma. Nature 539, 309–313 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R56] 56.Venteicher AS et al. Decoupling genetics, lineages, and microenvironment in IDH-mutant gliomas by single-cell RNA-seq. Science 355, eaai8478 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R57] 57.Joshi K, Bhatt MJ & Farabaugh PJ Codon-specific effects of tRNA anticodon loop modifications on translational misreading errors in the yeast Saccharomyces cerevisiae. Nucleic Acids Res. 46, 10331–10339 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R58] 58.Bulmer M Coevolution of codon usage and transfer RNA abundance. Nature 325, 728–730 (1987). [DOI] [PubMed] [Google Scholar]

[R59] 59.Lin H et al. CO₂-sensitive tRNA modification associated with human mitochondrial disease. Nat. Commun 9, 1875 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R60] 60.Shyh-Chang N et al. Influence of threonine metabolism on S-adenosylmethionine and histone methylation. Science 339, 222–226 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R61] 61.Yuan H et al. Lysine catabolism reprograms tumour immunity through histone crotonylation. Nature 617, 818–826 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R62] 62.Bao S et al. Glioma stem cells promote radioresistance by preferential activation of the DNA damage response. Nature 444, 756–760 (2006). [DOI] [PubMed] [Google Scholar]

[R63] 63.Gimple RC et al. Sorting nexin 10 sustains PDGF receptor signaling in glioblastoma stem cells via endosomal protein sorting. JCI Insight 8, e158077 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R64] 64.Li W et al. MAGeCK enables robust identification of essential genes from genome-scale CRISPR/Cas9 knockout screens. Genome Biol. 15, 554 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R65] 65.Pavlova NN et al. Translation in amino-acid-poor environments is limited by tRNA^Gln charging. eLife 9, e62307 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R66] 66.Chan CT et al. A quantitative systems approach reveals dynamic control of tRNA modifications during cellular stress. PLoS Genet. 6, e1001247 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R67] 67.Wang X et al. Purine synthesis promotes maintenance of brain tumor initiating cells in glioma. Nat. Neurosci 20, 661–673 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R68] 68.Perez-Riverol Y et al. The PRIDE database resources in 2022: a hub for mass spectrometry-based proteomics evidences. Nucleic Acids Res. 50, D543–D552 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R69] 69.Gogakos T et al. Characterizing expression and processing of precursor and mature human tRNAs by hydro-tRNAseq and PAR-CLIP. Cell Rep. 20, 1463–1475 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R70] 70.Ingolia NT, Brar GA, Rouskin S, McGeachy AM & Weissman JS The ribosome profiling strategy for monitoring translation in vivo by deep sequencing of ribosome-protected mRNA fragments. Nat. Protoc 7, 1534–1550 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R71] 71.Aeschimann F, Xiong J, Arnold A, Dieterich C & Grosshans H Transcriptome-wide measurement of ribosomal occupancy by ribosome profiling. Methods 85, 75–89 (2015). [DOI] [PubMed] [Google Scholar]

[R72] 72.Lauria F et al. riboWaltz: optimization of ribosome P-site positioning in ribosome profiling data. PLoS Comput. Biol 14, e1006169 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R73] 73.Bornelov S, Selmi T, Flad S, Dietmann S & Frye M Codon usage optimization in pluripotent embryonic stem cells. Genome Biol. 20, 119 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R74] 74.Xiao Z, Zou Q, Liu Y & Yang X Genome-wide assessment of differential translations with ribosome profiling data. Nat. Commun 7, 11194 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[R75] 75.Aibar S et al. SCENIC: single-cell regulatory network inference and clustering. Nat. Methods 14, 1083–1086 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

Threonine fuels glioblastoma through YRDC-mediated codon-biased translational reprogramming

Xujia Wu

Huairui Yuan

Qiulian Wu

Yixin Gao

Tingting Duan

Kailin Yang

Tengfei Huang

Shuai Wang

Fanen Yuan

Derrick Lee

Suchet Taori

Tritan Plute

Søren Heissel

Hanan Alwaseem

Michael Isay-Del Viscio

Henrik Molina

Sameer Agnihotri

Dennis J Hsu

Nu Zhang

Jeremy N Rich

Abstract

Results

GSCs exhibit high translation rates

Fig. 1 |. GSCs exhibit high translation rates.

CRISPR screening of tRNA modifiers in GSCs

Fig. 2 |. CRISPR screening of tRNA modifiers in GSCs.

YRDC is essential for GSC maintenance and translation

Fig. 3 |. YRDC is essential for GSC maintenance and translation.

Threonine dynamically regulates t6A and translation

Fig. 4 |. Threonine dynamically regulates t6A and translation.

Threonine functions mainly through YRDC in GSCs

Fig. 5 |. Threonine functions mainly through YRDC in GSCs.

Threonine and YRDC fuel mitosis with ANN codon bias

Fig. 6 |. Threonine and YRDC fuel mitosis with ANN codon bias.

Dietary TR inhibits tumor growth

Fig. 7 |. Dietary TR inhibits tumor growth.

Dietary intervention potentiates standard therapeutics

Fig. 8 |. Dietary intervention potentiates standard therapeutics.

Discussion

Methods

Human glioma and non-neoplastic brain tissues

Glioblastoma stem cell derivation and cell culture

Xenograft models

Threonine-restricted diet and drug treatment in vivo

Protein synthesis measurement in vivo and in vitro

CRISPR knockout library design, screen and data analysis

Immunoblotting

Immunofluorescence and immunohistochemistry staining

RNA extraction and quantitative real-time PCR

Northern blot for detecting tRNAThr charging levels

tRNA extraction and mass spectrometry analysis of t6A

Threonine tracing and mass spectrometry analysis of t6A

Proliferation and neurosphere formation assay

Click-iT EdU incorporation assay

Cell cycle analysis

Chromatin immunoprecipitation followed by PCR

Threonine quantification

Codon usage analysis

Tandem mass tag-labeled quantitative proteomics

RNA sequencing and data analysis

tRNA sequencing and data analysis

Ribosome profiling and data analysis

ANN codon reporters

Single-cell RNA sequencing reanalysis

Data mining of genome-wide CRISPR screens

GSC dataset interrogation

Transcription regulator analysis

Drug therapeutic efficacy prediction of YRDC

Public glioma patient datasets

Statistics and reproducibility

Extended Data

Extended Data Fig. 1 |. GSCs display high translation rates.

Extended Data Fig. 2 |. YRDC expression is driven by OLIG2.

Extended Data Fig. 3 |. Specific requirement of YRDC in GSCs than NSCs.

Extended Data Fig. 4 |. Threonine metabolism in GSC supports t6A formation.

Extended Data Fig. 5 |. YRDC and threonine shift the proteome supporting mitosis with ANN codon-bias.

Extended Data Fig. 6 |. Data mining of t6A downstream targets in clinical datasets.

Extended Data Fig. 7 |. YRDC loss reduces translation of cell cycle-related transcripts.

Threonine dynamically regulates t⁶A and translation

Fig. 4 |. Threonine dynamically regulates t⁶A and translation.

Northern blot for detecting tRNA^Thr charging levels

tRNA extraction and mass spectrometry analysis of t⁶A

Threonine tracing and mass spectrometry analysis of t⁶A

Extended Data Fig. 4 |. Threonine metabolism in GSC supports t⁶A formation.

Extended Data Fig. 6 |. Data mining of t⁶A downstream targets in clinical datasets.