B-progenitor acute lymphoblastic leukemia (BCP-ALL) is a hematologic malignancy arising from B-lymphoid progenitor cells. 1, 2 Through whole-genome, exome, and transcriptome sequencing, numerous subtypes have been identified, including those harboring DUX4 and NUTM1 rearrangements and the CDX2/UBTF subtype. 3-10 We previously reported MEF2D and ZNF384 fusions and DUX4-rearranged cases in 203 patients and defined 14 expression-based subtypes across 1223 cases. 11, 12 Using bioinformatics approaches, we analyzed transcriptomic data from 3371 BCP-ALL samples across 23 datasets. Hierarchical clustering revealed relationships among samples (Figure 1A; Supporting Information S2: Table 1), using the gene set listed in Supporting Information S2: Table 2. Of the 3371 cases, 3223 were assigned definitive molecular subtypes (proportions and case numbers in Supporting Information S2: Table 3; fusion and mutation profiles in Supporting Information S1: Figure 1A). Uniform Manifold Approximation and Projection (UMAP) visualization closely mirrored the hierarchical clustering results (Figure 1B). This large cohort enabled high-resolution delineation of the BCP-ALL genomic landscape. We identified 23 BCP-ALL subtypes via UMAP and hierarchical clustering. Notably, a candidate novel cluster (n = 17) was found adjacent to the CDX2/UBTF and TCF3: : HLF subtypes. (Figure 1A, B; Supporting Information S1: Figure 2; Supporting Information S2: Table 4; for discovery details, see Supporting Information S1: Methods: Molecular Features Searching Based on UMAP Embedding). The most differentially overexpressed gene in this cluster compared with all established subtypes was TWIST2 (Figure 2A, B; Supporting Information S1: Figure 3A, B; Supporting Information S2: Table 5). We therefore designated this entity the “TWIST2-high” subtype. ALLcatchR13 validated our classification, showing 93% consistency (2998 out of 3223 samples) between predicted and annotated labels, with high credibility across most subtypes (Supporting Information S1: Figure 4A, B). However, predictions for samples with high TWIST2 expression had low confidence (0. 75) and did not align with known subtypes, suggesting unique characteristics for the TWIST2-high subtype (Supporting Information S1: Figure 4B, C). Analysis of an independent BCP-ALL dataset (309 patients) from the Guangzhou Medical University Affiliated Women and Children's Medical Center identified two samples (A22082800018 and A23020800098) with abnormally high TWIST2 expression (FPKM 12. 26 and 12. 97). When these samples were incorporated into the original dataset for clustering analysis, they clustered with the TWIST2-high subtype, supporting the stability of this subtype (Supporting Information S2: Table 1; Supporting Information S1: Figure 1B). All BCP-ALL samples with available survival data (n = 667) were selected for survival analysis. 12 The TWIST2-high subtype had an inferior prognosis, with an overall survival of less than 50% at 3 years (Supporting Information S1: Figure 5). Differential gene expression analysis between the TWIST2-high subtype and the other groups identified 1856 significant genes, including FOXP2 and DPT (Figure 2A). ssGSEA revealed enrichment of the epithelial–mesenchymal transition (EMT) hallmark, consistent with the known oncogenic role of TWIST2 in epithelial tumors (Supporting Information S1: Figure 3C). Transcription-factor enrichment analysis implicated TWIST2 itself as a potential regulator of this signature (Supporting Information S1: Figure 6A). TWIST2-high cases lacked recurrent subtype-defining genetic alterations (fusions, sequence mutations, or copy-number profiles), indicating that elevated TWIST2 expression defines a genetically independent entity. Although TWIST2-high samples showed highly similar global expression profiles, they separated into two subclusters distinguished by discrete driver events (Figure 1A; Supporting Information S1: Figure 6B; Supporting Information S2: Table 6). RNA-based variant calling revealed hotspot isocitrate dehydrogenase 1/2 (IDH1/2) mutations (IDH1 R132 or IDH2 R140) in 4/17 cases (24%) and a recurrent fumarate hydratase (FH) missense variant (NC₀00 001. 11: g. 241506090 C>T; NP₀00 134. 2: p. Ala273Thr, hereafter FH A273T) in 7/17 cases (41%) (Figure 2C). These alterations were mutually exclusive. The FH A273T variant was uniquely confined to the TWIST2-high subtype across the entire 3371-case BCP-ALL cohort and was absent from 871 T-ALL and 3552 AML RNA-seq datasets (Supporting Information S2: Tables 1, 4, and 7–9). Whole-genome sequencing of four samples (representing three patients, including diagnostic and remission pairs) confirmed FH A273T at the genomic level, with variant allele frequencies consistent between RNA-seq and WGS. In the remission sample BL125C, FH A273T persisted at VAF 0. 508, confirming germline heterozygosity, whereas leukemic cells showed loss of heterozygosity (LOH) of the wild-type allele (Supporting Information S1: Figures 7–9; Supporting Information S2: Tables 4 and 9–18). AlphaMissense classified FH A273T as pathogenic (Supporting Information S2: Table 19). From the perspective of protein structure, A273T is located in the turn-structure of the protein and may affect enzyme activity by influencing the conformation of FH (Figure 2D; Supporting Information S2: Table 20). CRISPR-Cas9-mediated FH knockout or reconstitution with FH A273T equivalently impaired fumarase activity compared with wild-type rescue (Figure 2E–G), but neither manipulation altered TWIST2 expression (Supporting Information S1: Figure 10A). Across the full cohort, IDH1/2 hotspot mutations occurred in only 0. 5% of cases but were significantly enriched in TWIST2-high samples (P = 5. 16 × 10−7, binomial test; Figure 2C; Supporting Information S1: Figure 10B, C; Supporting Information S2: Tables 4 and 8). Using the gene signature from Yasuda et al. for ssGSEA scoring, we found that scores were relatively high in TWIST2-high samples (Supporting Information S1: Figure 11). 10 Unsupervised clustering within TWIST2-high cases cleanly separated FH A273T-dominated (TWIST2-high/FH-like) and IDH1/2-mutated (TWIST2-high/IDH-like) subclusters (Supporting Information S1: Figure 10C, D). FH and IDH1/2 are crucial enzymes in the tricarboxylic acid (TCA) cycle. FH mutations lead to fumarate accumulation, while IDH1 (R132) and IDH2 (R140) mutations produce (R) -2-hydroxyglutarate (R-2-HG). Both metabolites inhibit TET-dependent DNA demethylation, indicating a convergent oncogenic mechanism between these mutations. 14-16 In the only case where germline DNA was available (BL125), we could confirm that FH A273T was present as a heterozygous germline variant and that an LOH event led to monoallelic expression from the mutated allele in the ALL sample, in line with the traditional Knudson two-hit model for cancer predisposition. 17 In 3371 samples, RNA sequencing identified one sample carrying the IDH2 p. Y179D mutation and another sample with the FH p. G288V mutation. Additionally, two RNA-seq samples harbored the FH p. S347T and FH p. R261I mutations, respectively, but did not carry any IDH1/2 mutations. Notably, these mutations were annotated by AlphaMissense as being strongly pathogenic (Supporting Information S2: Table 19). To investigate the potential synergy between TWIST2 overexpression and the FH A273T mutation, we recapitulated this co-occurrence in FH-reconstituted BCP-ALL cells. Unexpectedly, TWIST2 overexpression in the presence of the FH A273T mutation resulted in reduced proliferative capacity (Supporting Information S1: TWIST2 Overexpression; Supporting Information S2: Tables 21 and 22; Supporting Information S1: Figures 12–16). However, the suppressive function of TWIST2 in established BCP-ALL cells is consistent with the findings reported by Thathia et al. and suggesting a context-dependent oncogenic role for TWIST2 distinct from other BCP-ALL subtypes. 18 Although IDH1/2 and FH mutations are significantly enriched and mutually exclusive in the TWIST2-high subtype, we could not identify a direct mechanism underlying TWIST2 upregulation. Due to the small proportion of patients in the TWIST2-high subtype, the statistical power of our prognosis-related studies may be insufficient. In summary, these larger datasets enabled a more detailed characterization of the genomic landscape of BCP-ALL and led to the identification of a novel subtype characterized by high TWIST2 expression. This TWIST2-high subtype displays a distinct gene expression profile, is frequently associated with alterations in IDH1/2 or FH genes, and appears to be linked to adverse clinical prognosis. The results published here are in whole or in part based upon data generated by the Therapeutically Applicable Research to Generate Effective Treatments (https: //www. cancer. gov/ccg/research/genome-sequencing/target) initiative, phs000218, and datasets from databases dbGap, EGA, DDBJ, NGDC, and GEO. Tao Zeng: Writing—original draft; writing—review and editing; visualization; formal analysis. Ling Zhang: Validation; investigation; writing—original draft. Wenxin Yin: Validation; writing—original draft; formal analysis. Cuiping You: Writing—original draft; formal analysis. Henrik Lilljebjörn: Writing—original draft. Qian Wang: Resources. Weina Zhang: Resources. Xiaotian Ji: Validation. Yuliang Wang: Writing—review and editing; writing—original draft; formal analysis. Yongjing Liu: Formal analysis; writing—original draft; writing—review and editing. Yali Xie: Writing—original draft; writing—review and editing; formal analysis; data curation. Xiaoxi Feng: Writing—original draft; writing—review and editing; formal analysis. Xiang Zhang: Resources. Hua Jiang: Resources. Thoas Fioretos: Writing—original draft. Gang Xiao: Validation; investigation; resources; funding acquisition. Jie Jin: Resources. Suning Chen: Resources. Jinyan Huang: Resources; supervision; formal analysis; data curation; writing—original draft; writing—review and editing; conceptualization; project administration; funding acquisition. The authors declare no conflicts of interest. This study was supported by the National Key Research and Development Program of China (2019YFE0108100), Key Research and Development Program of Zhejiang Province (2024SSYS0024), the National Natural Science Foundation of China (NSFC) General Program (No. 82270159, 32170663, 82170142 and 82470222), National Key R&D Program of China (2022YFE0133200), the Department of Science and Technology of Zhejiang Province through program (2023R01012). The code used in this analysis can be found at https: //github. com/JhuangLab/twist2/ and https: //jhuanglab. github. io/twist2/. Data generated by this study is available upon request to the corresponding author. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.
Zeng et al. (Wed,) studied this question.