Articles Pathophysiologic and clinical implications of molecular profiles resultant from deletion 5q Vera Adema,a Laura Palomo,b Wencke Walter,c Mar Mallo,b Stephan Hutter,c Thomas La Framboise,d Leonor Arenillas,e Manja Meggendorfer,c Tomas Radivoyevitch,a Blanca Xicoy,f Andrea Pellagatti,g Claudia Haferlach,c Jacqueline Boultwood,g Wolfgang Kern,c Valeria Visconte,a Mikkael Sekeres,h John Barnard,i Torsten Haferlach,c Francesc Sole,b,1 and Jaroslaw P. Maciejewski a,1* a Department of Translational Hematology and Oncology Research, Lerner Research Institute Cleveland Clinic, Taussig Cancer Institute, 9500 Euclid Avenue, Cleveland, OH 44195, USA b Myelodysplastic Syndrome Research Group, Josep Carreras Leukaemia Research Institute, Institut Catala d'Oncologia-Hospi- tal Germans Trias i Pujol, Universitat Autonoma de Barcelona, Badalona, Spain c MLL Munich Leukemia Laboratory, Munich, Germany d Department of Genetics and Genome Sciences, Case Western Reserve University, Cleveland, OH, USA e Laboratori de Citologia Hematolo gica, Servei de Patologia, Hospital del Mar and GRETNHE, Cancer Research Program, IMIM (Hospital del Mar Medical Research Institute), Barcelona, Spain f Hematology Service, Institut Catala d'Oncologia (ICO)-Hospital Germans Trias i Pujol, Institut de Recerca Contra la Leuc emia Josep Carreras, Universitat Autonoma de Barcelona, Badalona, Spain g Blood Cancer UK Molecular Haematology Unit, Nuffield Division of Clinical Laboratory Sciences, Radcliffe Department of Medicine, University of Oxford and Oxford BRC Haematology Theme, Oxford, United Kingdom h Leukemia Program, Department of Hematology and Medical Oncology, Cleveland Clinic, Cleveland Clinic Taussig Cancer Institute, Cleveland, OH, USA i Department of Quantitative Health Sciences, Cleveland Clinic, Lerner Research Institute, Cleveland, OH, USA Summary Background Haploinsufficiency (HI) resulting from deletion of the long arm of chromosome 5 [del(5q)] and the accompanied loss of heterozygosity are likely key pathogenic factors in del(5q) myeloid neoplasia (MN) although the eBioMedicine 2022;80: 104059 consequences of del(5q) have not been yet clarified. Published online 23 May 2022 Methods Here, we explored mutations, gene expression and clinical phenotypes of 388 del(5q) vs. 841 diploid cases https://doi.org/10.1016/j. with MN [82% myelodysplastic syndromes (MDS)]. ebiom.2022.104059 Findings Del(5q) resulted as founder (better prognosis) or secondary hit (preceded by TP53 mutations). Using Bayes- ian prediction analyses on 57 HI marker genes we established the minimal del(5q) gene signature that distinguishes del(5q) from diploid cases. Clusters of diploid cases mimicking the del(5q) signature support the overall importance of del(5q) genes in the pathogenesis of MDS in general. Sub-clusters within del(5q) patients pointed towards the inherent intrapatient heterogeneity of HI genes. Interpretation The underlying clonal expansion drive results from a balance between the “HI-driver” genes (e.g., CSNK1A1, CTNNA1, TCERG1) and the proapoptotic “HI-anti-drivers” (e.g., RPS14, PURA, SIL1). The residual essen- tial clonal expansion drive allows for selection of accelerator mutations such as TP53 (denominating poor) and CSNK1A1 mutations (with a better prognosis) which overcome pro-apoptotic genes (e.g., p21, BAD, BAX), resulting in a clonal expansion. In summary, we describe the complete picture of del(5q) MN identifying the crucial genes, gene clusters and clonal hierarchy dictating the clinical course of del(5q) patients. Funding Torsten Haferlach Leukemia Diagnostics Foundation. US National Institute of Health (NIH) grants R35 HL135795, R01HL123904, R01 HL118281, R01 HL128425, R01 HL132071, and a grant from Edward P. Evans Foun- dation. Copyright Ó 2022 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/) Keywords: Myelodysplastic syndromes; 5q deletion; Haploinsufficiency; TP53; CSNK1A1 *Corresponding author. E-mail address: maciejj@ccf.org (J.P. Maciejewski). 1 These authors contributed equally to this work. www.thelancet.com Vol 80 Month June, 2022 1 Articles (CDR-1)4 is associated with a higher-risk MDS or acute Research in context myeloid leukaemia (AML). The distal CDR spanning 1.5Mb on 5q32-q33.2 (CDR-2)5 is present in the classical Evidence before this study 5q- Syndrome and conveys better prognosis. The prevailing theory in del(5q) is that haploinsuffi- The clinical del(5q) phenotype may also be affected ciency (HI) stemming from deletion and not simply loss by the actual boundaries of the deletion,6 along with of heterozygosity (LOH) is the culprit in clonal evolution. other chromosomal defects and somatic mutations To date no haploinsufficient gene has been found to be co-existing in a hierarchical relationship with del the pivotal leukaemogenic factor conveying growth (5q).7 Significant progress in the understanding of the advantage, but various other genes have been found to disease mechanisms resulting from del(5q) has been be important for phenotypic features or for propensity made. However, it has not been entirely clarified to acquire subsequent specific lesions. This study repre- which haploinsufficient (HI) genes on 5q are associ- sents the most comprehensive compendium of geno- mic and RNA sequencing data of patients with MDS ated with the disease development and malignant pro- carrying del(5q). Unlike previous studies, we based our gression. analysis on an innovative copy number adjustment that Here, we examine a large cohort of del(5q) patients to mitigates the clonality effect on the haploinsufficient clarify the molecular relationships of these lesions with genes on 5q. other somatic defects and determine their impact on the resultant expression patterns and the morphologic Added value of this study and clinical phenotypes. Our study based on a large cohort, has defined distinct expression signatures of molecular subtypes of del(5q) according to the position in the clonal hierarchy and Methods correcting previous assumptions that del(5q) was the Patients and samples: A total of 400 samples (388 ultimate ancestral lesion at the level of a stem cell. We patients and 12 follow-up samples) with MNs (myeloid have also provided a minimal haploinsufficient (HI) neoplasms) with del(5q) (Table 1) determined by con- expression signature of del(5q) including most indica- ventional G-banding cytogenetics and fluorescence in tive genes and their pathophysiologic role. Further- situ hybridization (FISH) were included in the study. A more, we defined HI-driver (clonal expansion) and HI- anti-drivers (brakes, proapoptotic) genes and their com- cohort of patients with MN diploid status for 5q plex interactions in del(5q) neoplasia affecting the clini- [n = 844 samples (840 patients and four follow-up sam- cal course. Clonal drive overcomes proapoptotic ples)] were also collected for comparison purposes. All pressure (HI-anti-drivers) through escape mechanisms samples included were profiled for gene sequencing promoted by accelerator events (e.g., CSNK1A1, TP53 and cytogenetics was available for all. mutations). Ethics Information: All samples were collected with previous written informed consent in accordance with Implications of all the available evidence institutional ethics committee approvals following the Our analysis of a compendium of del(5q) genomics con- revised Declaration of Helsinki. A total of 51 samples firms some of the previous findings, revises previous were collected at the Josep Carreras Leukaemia assumptions, discovers new haploinsufficient candidate Research Institute (2010/4102/I) in between 2009 and genes and precisely circumscribes molecular relation- 2014 on behalf of the MDS Spanish group, 185 samples ships (e.g., del(5q) with TP53 mutations) and generates were collected at the Cleveland Clinic (IRB-5024) in a minimalistic expression signature of del(5q). between 2002 and 2017, and 1,008 samples were col- lected at the MLL Munich Leukemia laboratory in between 2005 and 2018. We think our cohort is indeed representative because: (i) cohorts from 3 centres of excellence were collected in the same time period and Introduction all the diagnostic criteria and data collected were syn- Deletions in the long arm of chromosome 5 [del(5q)] can chronized; (ii) our sample size is large enough to accom- be found in different myeloid neoplasia (MN) and are modate for the biologic diversity, including rare the most common karyotypic alterations in myelodys- subtypes and; (iii) is not hampered by a single institu- plastic syndromes (MDS). Del(5q) can be found alone or tional bias. with other cytogenetic abnormalities and is associated Next-Generation Sequencing (NGS): Samples under- with specific morphologic features.1 However, del(5q) is went targeted deep sequencing (tumour sample; the only cytogenetic abnormality used to define a pheno- n = 141), whole exome sequencing (WES; tumour and typically distinct WHO MDS subtype,2 originally germline sample; n = 95) or whole genome sequencing described by Van den Berghe and colleagues in 1974.3 (WGS; tumour sample; n = 1,008). To combine the data Del(5q) encompasses two commonly deleted regions from the three platforms (Table S9) we selected 181 (CDRs). The centromeric CDR on 5q31.2-5q31.3 genes (on the basis of their implication in the 2 www.thelancet.com Vol 80 Month June, 2022 Articles del(5q) cohort (n = 388) isolated del(5q) del(5q) + 1 del(5q) in CK n (%) 188 (49) 36 (9) 164 (42) Female, n (%) 132 (70) 23 (64) 71 (43) Age, median years (min-max) 73 (35 94) 73 (42 88) 73 (24 89) Classification, n (%) Low Risk MDS low risk 154 (82) 25 (69) 37 (22) CMML-1 1 (0.5) 1 (3) 0 High risk MDS high risk 20 (11) 5 (14) 78 (48) CMML-2 0 0 0 sAML 10 (5) 4 (11) 30 (18) pAML 3 (1.5) 1 (3) 11 (7) MDS/MPN-U 0 0 8 (5) Blood counts (%) Anaemia (<10g/dL) 117 (69) 19 (58) 106 (68) Thrombocytopenia (<100 £ 109/L) 17 (10) 12 (36) 113 (72) Thrombocytosis (>450 £ 109/L) 24 (14) 2 (6) 2 (1) Neutropenia (<1.8 £ 109/L) 51 (46) 17 (55) 88 (72) % BM blasts, median (min-max) 3 (0 76) 3 (0 45) 8 (0 87) Table 1: Clinical features of del(5q) cohort. CK: Complex Karyotype:  3 cytogenetic alterations; Low risk: MDS low-risk: MDS-SLD (MDS with single lineage dysplasia), MDS-RS (MDS with ring sidero- blasts), MDS-MLD (MDS with multilineage dysplasia), MDS del(5q) (MDS with isolated del(5q)), MDS-U (unclassifiable MDS); CMML-1 (chronic myelomono- cytic leukemia-1); High risk: MDS high-risk: MDS-EB-1 (MDS with excess blast-1), MDS-EB-2 (MDS with excess blast-2); CMML-2 (chronic myelomonocytic leukemia-2); sAML (secondary AML); pAML (primary AML); MDS/MPN-U (myelodysplastic/myeloproliferative neoplasm unclassifiable). BM: bone marrow. (CBC data available for: iso-del(5q): anemia: n = 169; thrombocytopenia: n = 169; thrombocytosis n = 169; neutropenia: n = 111; del(5q) +1: anemia: n = 33; thrombocytopenia: n = 33; thrombocytosis n = 33; neutropenia: n = 31; del(5q) in CK: anemia: n = 156; thrombocytopenia: n = 156; thrombocytosis n = 156; neu- tropenia: n = 122). pathogenesis of MN; Table S2). After variant calling, all Haploinsufficiency expression analysis: RNA-Seq samples were analysed together based on a strict in- results from 995 patients (170 del(5q) and 825 diploid house variant filtering process in line with current sci- patients) with MN were analysed. Patients were further entific standards. There was no institutional biased in classified according to the karyotype, BM blast % and terms of distribution of somatic mutations. RNA- TP53 mutational status. After data processing (supple- sequencing (RNA-seq) was performed only on samples mental methods), a total of 148 del(5q) and 752 diploid collected at the MLL Munich Leukemia laboratory. patients were included in the analysis. Gene expression Clonal hierarchy: To determine dominant/ancestral was measured in counts per million (CPM), assigned to and sub-clonal/secondary hits and del(5q) for each each gene (see supplemental methods). Adjustments for patient: Copy Number Variation from WES data was cal- del(5q) clonality: Linear models were fit per gene using culated based on the allelic imbalance of informative the R package limma8 with log2(CPM+0¢1) as the heterozygous SNPs within the 5q deleted region in WES response variable and del(5q) CNV clonality (centred at germ line samples. Copy Number Variation analysis from 50% clonality), del(5q) status (isolated, +1, CK), diploid WGS data: Copy number variant (CNV) calls were cre- karyotype (normal, 1+, 2+, CK), TP53 mutational status ated using GATK4 (version 4.0.2.1), following the “best (wild type, mutant) and blast risk [BM blasts %: low practices” somatic CNV workflow (https://software. (<5%), high (>5%)] as covariates. broadinstitute.org/gatk/best-practices/). Del(5q) haploinsufficient signature: Sparse Prediction RNA-Seq: Total RNA from whole bone marrow (BM) Modeling: models for predicting del(5q) vs. diploid status samples was extracted. RNA-Seq libraries were con- were built using a Bayesian sparse logistic regression structed from ribosomal RNA depleted RNA using Tru- approach as implemented in the R package HTLR.9 Seq Total Stranded RNA kit (Illumina) and sequenced The htlr function from HTLR was used to train the pre- on a NovaSeq 6000 platform. diction models with the default parameters (t prior), www.thelancet.com Vol 80 Month June, 2022 3 Articles Figure 1. Cytogenetic and molecular characterization of 5q: (a) CNV analysis based on the cytogenetic break points was used to determine the most commonly deleted region in del(5q). (b) Mutation distribution in isolated del(5q), compound del(5q), and diploid patients. (c) Percentage of mutant patients in the most common mutated genes (mutations on chromosome 5 are not included). 4 www.thelancet.com Vol 80 Month June, 2022 Articles Figure 2. Clonal architecture and subclonal hierarchy: reconstruction of the clonal hierarchy using an allelic imbalance method for WES samples (with available paired sample) and by CNV analyses for WGS, were compared to the VAF of most common mutations adjusted by zygosity and copy number. (a) Distribution of ancestral (navy blue), co-dominant (yellow) or subclonal (light purple) del(5q) in isolated del(5q) and com- pound del(5q). (b) Kaplan-Meier curves for ancestral, secondary and co-dominant del(5q) event with significant differences (Log rank test). (c) Mutational distribution of the most representative mutations in dominant, co-dominant or subclonal del(5q) in patients with isolated del(5q) or compound del(5q). (d) Exemplary cases of clonal architecture of isolated del(5q) patients (left) and in compound del(5q) patients (right). (e) Percentage of clonal burden of del(5q) (red), increased subclonal mutations (yellow), and decreased subclonal mutations (blue). Lines connect paired samples. Doted lines indicate undetectable mutation after treatment. www.thelancet.com Vol 80 Month June, 2022 5 Articles except for initial states, which were chosen using the patients without 5q (P = 0¢0001; [Fisher's exact test]) bias-corrected Bayesian classification approach option. (Figures 1b; S1c). Final gene predictors were those having non-zero model For the mutational profile, we selected the most coefficients (calculated using the HTLR nzero_idx func- commonly mutated genes in MNs (n = 181). Comp- tion). Models were trained using MHI(50%) and MHI del5q showed a higher frequency of TP53 mutations vs. (25%) data with 900 samples (148 del(5q) samples) and iso-del5q and mutant diploid cases (P < 0¢0001; [Chi2- expression from 57 del(5q) genes. Class predictions test]), and a significantly lower frequency of hits in were made using the posterior median of the regression SF3B1, TET2, and ASXL1, vs. iso-del5q (Figure 1c). coefficients. Prediction error was assessed using ten- When we focused on mutations on 5q, CSNK1A1 (5q32) fold cross validation and feature selection stability was was the most commonly mutated gene in iso-del5q assessed by counting the number of times the feature cases. Canonical CSNK1A1E98 (n = 13) and was selected across the ten fits. CSNK1A1D140V (n = two) were identified.10,11 Neither Statistical analyses: Statistical analyses were per- canonical DDX41 (5q35.3) frame-shift germ line, nor formed on GraphPad Prism (version 8.0.2) and R statis- somatic missense ATP-binding domain mutations were tical software packages (version 3.6.0). Chi-square or found within deleted areas and the only three DDX41 Fisher’s exact test were performed for categorical varia- mutations in del(5q) were all heterozygous.12,13 RAD50 bles, while Mann-Whitney and Wilcoxon tests were (5q31.1; n = four) was found in a hemizygous configura- used for pairwise continuous variables. The overall sur- tion, whereas NPM1 (5q35.1; n = two) mutations were vival (OS) for diagnosis samples was defined from diag- always heterozygous and associated with comp-del5q in nosis to death or last follow-up. All deaths, whether or secondary-AML (s-AML). not related to MDS, were considered as the endpoint of Subclonal hierarchy and dynamics. Zygosity- and copy the follow-up interval. Kaplan-Meier method was used number-adjusted VAF of somatic mutations were used. to estimate the OS curves, and log rank test was used To determine the clonal architecture, the CNV analyses for comparisons between groups. All p-values reported were performed following allelic imbalance calculations are two-sided and considered statistically significant for heterozygous SNPs located in the CDRs of del(5q)14 when P <005. (Figure S2a-b). Del(5q) occurs in diverse configurations Role of the funders: Funding sources had no role in (Figure S2c-e) with coinciding somatic mutations: (i) as study design, data collection, data analyses or results a dominant event, more frequent in iso-del5q (61%) interpretation, in writing of the manuscript, or in any than in comp-del5q (46%; P = 0¢03; [Fisher's exact test]); aspect related to the study. (ii) as a sub-clonal event with similar frequencies in comp-del5q (27%) and iso-del5q (24%); or (iii) as a co- dominant event with unresolved rank of somatic altera- Results tions/del(5q), more commonly in comp-del5q (27%) Clinical and genetic features of del(5q) patients. We than in iso-del5q (15%; P =0¢04; Figure 2a). Patients included 388 patients with del(5q)-associated (Table 1) with a dominant del(5q) had better overall survival and 844 MN without del(5q) (Table S1; Figure 1a-b); (OS; P =0¢0296; [Kaplan-Meier, log rank test]) than RNA-seq results from BM (obtained retrograde flushing those with del(5q) in a co-dominant or subclonal config- of discarded de-identified BM harvest kits) of 64 healthy uration (Figure 2b). Del(5q) occurring in a dominant individuals [median age 59 (range, 26 85)] were configuration coincided with secondary CSNK1A1 muta- included as reference. Patients were grouped according tions (ten%) in iso-del5q, and TP53 (26%) in comp- to del(5q) status: isolated del(5q) (iso-del5q) and com- del5q. However, when del(5q) was a sub-clonal event, it pound del(5q) (comp-del5q): including del(5q) plus any was often preceded by dominant ASXL1 mutations other alteration (+1-del5q) or in the context of complex (26%) in iso-del5q and dominant TP53 (56%) in comp- karyotype (CK-del5q) (Figure S1a-b). To delineate the del5q (Figure 2c). Figure 2d shows exemplary cases boundaries of the del(5q), we conducted copy number with ancestral iso-del5q and sub-clonal CSNK1A1 and variant (CNV) analyses using conventional cytogenetics. SF3B1. When a clear clonal hierarchy could not be math- The CDRs in our del(5q) cohort (CDR-A; Figure 1a) ematically resolved, del(5q) co-ranked with TP53, ASXL1, included CDR-1 and CDR-2.4,5 The commonly retained and SF3B1 mutations in iso-del5q and TP53 and regions6 (CRR) centromeric (p12-q13) and telomeric DNMT3A in comp-del5q. (q35.1-q35.3) were mostly co-deleted in CK-del5q (36% Sequential BM and granulocyte fraction samples were and 52% of patients, respectively; Figure 1a). NGS analy- available from 7 and 3 patients, respectively. Nine patients sis for 181 selected genes (Table S2) yielded 2,251 received at least 6 cycles of LEN (LEN cycles: pt1:8, pt2: 9; somatic mutations in 1,034 (83%) cases (Table S3). pt3: 6, pt4:8; pt5: 9, pt6: 13; pt7:10; pt8:7; pt9:32), and pt10 While a total of 71% of patients with iso-del5q carried was during disease progression. Most consistently, del(5q) one mutation (12% on 5q), 85% of patients with comp- decreased in response to LEN (pt1, pt2, pt3, pt4, pt6, pt7). del5q harboured one mutation (8% on 5q; P =0¢001; Concurrent mutations showed diverse dynamics; (i) [Fisher's exact test]) but were less mutated than decrease/ disappearance of mutations (e.g., DNMT3A, 6 www.thelancet.com Vol 80 Month June, 2022 Articles Figure 3. Haploinsufficiency (HI) analysis of selected genes on 5q. (a) Box plot showing the definition of HI expression defined when the expression in del(5q) was lower than the percentile 25th of the diploid expression level. (b) Gene expression (Log2 CPM) in del(5q) cases (pink) and diploid cases (blue). Dashed lines show median values. (c) Exemplary cases of different correlation types between expression and clonality in isolated del(5q) and compound del(5q). Green error bar shows the expression of the given gene in diploid cases. Black error bar shows the expected expression at 50% and 100% clonality of del(5q). (d) Heatmap using the spares model-based clustering of del(5q) and diploid patients (see methods), including the 57 HI genes. Karyotype group, TP53 mutational status, BM Blast % group and % of clonality are also depicted. www.thelancet.com Vol 80 Month June, 2022 7 Articles CSNK1A1, ASXL1) paralleling del(5q) contraction (pt1, pt2, (n = 66) and -seven (n = 233) contained the majority of pt3, pt4); (ii) expansion of pre-existing mutations (e.g., non-del(5q) LR-MDS. Cluster-six had the highest per- PRPF8, RUNX1, ASXL1) during LEN treatment, while del centage of abnormal karyotypes. Cluster-seven, the larg- (5q) clone decreased (pt5 was responder to LEN while pt8 est cluster, included the majority of normal or non-CK and pt9 nine progressed despite therapy); (iii) progressing karyotypes and showed a high frequency of SF3B1 and patients were characterized by acquisition of new indepen- DNMT3A mutations (n = 29). dent subclones (e.g., pt10; ASXL1; Figure 2e). In terms of marker HI genes, Cluster-three exhibited Haploinsufficiency of genes located on 5q. Of 405 genes higher CXXC5 mRNA levels vs. del(5q) defined Cluster- on 5q, 188 were located within CDR-A (CDR-1: 41, CDR- one. Cluster-four was characterized by low expression of 2: 55 genes) and 128 were deleted across all patients TGFBI. Cluster-five showed overall differences on the (according to CNV analysis). For dichotomized thresh- expression pattern in gene Cluster-two (RPS14, old analysis, we defined mRNA expression to be HI in NDUFA2, SMAD5) and gene Cluster-four (SIL1, del(5q) when it was 25th percentile of the diploid CTNNA1, among others). The expression pattern of expression (Figure 3a-b). We noticed a variable land- Cluster-7 was most distinct from that of Cluster-one. scape of HI expression in some of the classically HI Minimal haploinsufficient signature of del(5q) and genes, probably due to normal tissue contamination. haploinsufficient patterns within del(5q). Using the 57 We then queried the relationship between mRNA and HI marker genes, we performed a sparse Bayesian pre- corresponding 5q ploidy. Genes with a higher-than- diction analyses to identify the minimal del(5q) gene expected expression, due to the del(5q), may show a par- signature that could distinguish del(5q) from diploid tial compensation or deletion of a methylated allele cases. Using the 50% clonality adjusted data, the mini- (PPP2AC, RPS14), while those with a steeper slope are mal gene signature contained UBE2D2, C5orf24, PURA, affected by opposing effects (DDX46, PURA), while SIL1, and HSPA4 genes and misclassified only one del some did not correlate with the del(5q) clonal burden (5q) sample (0¢1% observed error rate for clustering (TAF7, EGR1) (Figure 3c). To identify causative genes, within 5q deleted samples; Figures 4a; S4a). Even when subsequent analyses were restricted to the 57 genes this five-gene signature was applied to the 25% clonality (q23.3-q33.1) showing evidence of HI and negative adjusted data, the error rate was only 6¢7%. When we ploidy slope in isolated del(5q) (Figure S3, Table S4). conducted pairwise associations between the 57 genes, Some previously implicated genes in the del(5q) region we found a striking correlation with distinct groups of did not follow these rules. Remarkably, decline of highly correlated 5q genes. Even though diploid patients expression of CDC25C, DIAPH1, SPARC, RAD51 and also exhibited some significant correlations, those corre- EGR1 genes either did not correlate with the del(5q) lations were stronger and more significant in del(5q). clonal burden or was too variable to be pathognomonic. Repeating the signature search using the 25% clonality Other genes were excluded because they were not adjusted data resulted in a seven HI gene signature deleted consistently enough [APC: 144/148, DDX41 (PURA, SIL1, HSPA4, CSNK1A1, TCOF1, CTNNA1, and one/148, NPM1 one/148 (deleted/total)] to assume the FAM13B) with only 11 del(5q) and eight diploid patients ubiquitous role in the pathogenesis of del(5q). misclassified (one% error rate; Figure S4b-c). When we Ploidy-adjusted expression value of the 57 selected applied model-based sparse sub-clustering within del genes were used for unsupervised clustering to deter- (5q) patients a nine gene signature was found (TGFBI, mine differences in the expression levels between del CTNNA1, PURA, CXXC5, DCTN4, SMAD5, HARS, (5q) and diploid patients for genes on 5q (Figure 3d). As TMEM173, and RPS14) to obtain six different expression expected, del(5q) cases clustered together and showed clusters demonstrating heterogeneity of HI within del consistent HI of 5q marker gene expression (Tables S5 (5q) patients (Figure 4b; d, Tables S7 and 8). Cluster and 6). Cluster-one (n = 146) included almost all del(5q) comparison showed that cluster-A and cluster-E had the cases, except for eight patients (“mis-categorized” to majority of CK but cluster-A showed low mutational other clusters). It was characterized by low risk MDS burden and lowest expression of TGFBI, CTNNA1, and (LR-MDS), presence of anaemia /neutropenia and low PURA. Cluster-E showed more 5q as secondary hit, mutational burden, with TP53 being the most com- high number of TP53 mutations (n = 20), and HI for monly mutated gene. Diploid cluster-two (n=133) fea- SMAD5, RPS14, and TMEM173 but less profound HI of tured a normal karyotype, frequent ASXL1 and TET2 PURA and CXXC5. Cluster-B and cluster-F included the mutations, and profound down-modulation of RPS14 majority of isolated del(5q). Ancestral 5q hits and deeply and NDUFA2 mRNA. Clusters-three, -four, -five depressed expression of TGFBI, CTNNA1, and PURA (n = 138, 90, 94, respectively) included most of the high (but not CXXC5 or RPS14) were grouped in cluster-B. risk MDS (HR-MDS). Cluster-three was enriched for Cluster-C pattern was the most common (n = 62), with thrombocytopenia and SRSF2 mutations; cluster-four 50% of ancestral del(5q), frequent TP53 mutations for anaemia, thrombocytopenia, and ASXL1 and SRSF2 (n = 23) and low expression of CTNNA1, PURA, and mutations. Cluster-five was characterized by pancytope- HARS. Interestingly, RPS14 mRNA was not uniformly nia and frequent ASXL1 mutations and CK. Cluster-six HI in del(5q) patients. By comparison diploid cluster- 8 www.thelancet.com Vol 80 Month June, 2022 Articles Figure 4. Minimal gene signature of del(5q). (a) Heat map including the del(5q) 5 gene-minimal signature according to a sparse Bayesian prediction analysis that differentiates between del(5q) and diploid patients. Expression of del(5q) patients was adjusted to a 50% del(5q) clonality representing the expression of del(5q) and diploid patients. Blue colours represent a low expression while red colours represent up-regulation of the expression (white means neutral). (b) Heat map including the del(5q) 9 gene-minimal signature within del(5q) patients. Blue colours represent a low expression while red colours represent up-regulation of the expression (white means neutral). www.thelancet.com Vol 80 Month June, 2022 9 Articles two contained patients with a down-modulation of associations leading to apoptosis/ proliferation/ differ- RPS14 but no relationship with p21 levels or TP53 muta- entiation and disassociation in MDS. tions was seen in diploid cases vs. those with del(5q) and Using an integrative-molecular approach, we dem- low RPS14 expression. onstrated: (i) del(5q) patients have similar but less com- Pathogenic interactions. We identified several sub- plex mutational landscapes as other MDS subtypes with classes according to their function with regard to clonal mutational signatures dominated by exclusively hemizy- advantage: (i) directly or indirectly acting proapoptotic gous CSNK1A1 mutations, along with an increased fre- genes (BAD, BAK1, BAX, among others) (Figure 5a) quency of TP53 mutations; (ii) a subset of HI genes whose HI facilitates subclonal escape via e.g., hypomor- (n = 57) was selected on significant expression reduction phic TP53 mutations or via CSNK1A1 mutations. HI lev- with a negative correlation with 5q clonality. Minimal els of CSNK1A1 may promote apoptosis (of MDM2 expression signature points towards the presence of dephosphorylation leading to p21 increase). UBE2D2 marker genes for del(5q) irrespective of their functional HI in del(5q) may result in accumulation of nuclear importance. For analytic purposes, minimal signature p5315 contributing to mutational pressure; and (ii) genes may help to design RNA based tests for del(5q) tumour suppressor genes whose HI directly provides using minimal number of target genes and to identify selective pressure for del(5q) CSNK1A1, CTNNA1, del(5q) cells in single cell RNA-seq. While reproducibil- C5orf24 (TECRG1) genes. ity of minimal signature may be an issue, our study has We then focused on selected interactions between identified 57 consistently HI genes, which can be added well-established mutations and the newly defined HI 5q to improve the precision of identification of del(5q); (iii) genes. CSNK1A1 mutations were only found in del(5q) del(5q) neoplasms result from complex interaction patients (12% vs. none in diploid cases); CSNK1A1 and between HI-driver (clonal expansion) and HI-anti-driv- TP53 mutations did not coincide; only 15% were co- ers (brakes, proapoptotic) genes; (iv) clonal drive eventu- mutated with TP53 (two% vice versa, Figure 5b). All del ally overcomes proapoptotic pressure (HI-anti-drivers) (5q) showed CSNK1A1 HI expression [15% of diploid through escape mechanisms promoted by accelerator patients were CSNK1A1 low expressors (le; 75th per- events; (v) using gene clustering according to expres- centile; Figure 5c)]. Diploid CSNK1A1-le also had low sion reduction, we identified subsets of diploid patients expression of RPS14 (P < 0¢001, [Fisher's exact test]; with 5q gene repression patterns similar to del(5q) Figure 5d). Moreover, CSNK1A1 expression did not cor- patients; vi) we determined a minimal HI gene signa- relate with TP53 mutations (Figure 5e). Irrespective of ture of del(5q) involving essential genes. Integration of CSNK1A1 mutations or wild type expression level, these findings led to a new and more complete concept CDKN1A (p21) as a downstream effector of TP53 was of pathogenesis of del(5q). generally up regulated in del(5q) except for TP53 In contrast to some previous studies postulating that mutants showing p21 mRNA levels comparable to those del(5q) is an obligatory primary hit,16 we demonstrated of diploid patients (P = 0¢125; [Mann-Whitney test]; that del(5q) is not always an initiating event. It may be Figure 5f). Similarly, del(5q) cases with HI RPS14 preceded by other mutations (e.g., mutations in TP53), a expression also showed decreased p21 expression succession characterized by an early advanced pheno- (P < 0¢0001; [Mann-Whitney test]; Figure 5g). type. Subclonal TP53 mutations likely represent a later While expression of CDC25C, DIAPH1, and RAD50 escape mechanism resulting from apoptotic pressure genes did not correlate with ploidy (FDR for ploidy slope produced by HI of various genes. When del(5q) was (SL): 0¢988; 0¢072; 0¢307; respectively; Figure 5h), EGR1 dominant, the most common secondary hits in isolated expression showed a positive ploidy slope (FDR SL: del(5q) were mutations in CSNK1A110,17,18 and for com- 0¢211), but those with a relatively lower expression were pound del(5q) TP53 mutations. The individual fate of 75% of TP53 mutants (40/53; P <0¢0001; [Chi2-test]) the clinical course of del(5q) is thus influenced by the (Figure 5j). APC, when involved in the deleted region, hierarchy and the succession of molecular events, add- was indeed HI (P <0¢0001) and correlated with the pres- ing further complexity to del(5q) sub-entity. ence of CSNK1A1 mutants (n = eight/nine) (Figure 5k). For the purpose of this discussion, we focused on the When deleted (21/167) DDX41 were indeed HI; all these ancestral del(5q) characterizing the classical del(5q) patients had a CK and 18/21 of these patients also har- entity (iso-del5q). Among HI genes involved in ancestral boured TP53 mutations (Figure 5l). del(5q) there must be some HI-drivers which provide the impetus for clonal expansion, thus enabling selec- tion needed to eventually overcome HI-anti-driver genes Discussion producing the phenotypic dysplasia and apoptosis of del Del(5q) results in various biologic features due to the HI (5q). Such a pressure leads to acquisition of accelerator of individual genes contained in the affected region. We events such as CSNK1A1 mutations, TP53 mutations or investigated the molecular mechanisms of del(5q) deletion, and/or monosomy 7. including HI genes and mutations affecting cellular Putative HI-drivers include the previously function to elucidate the phenotype-genotype “established” del(5q) genes: CSNK1A1,17 CTNNA1,19 and 10 www.thelancet.com Vol 80 Month June, 2022 Articles Figure 5. Pathogenic mutational interactions in del(5q): (a) Expression of selected pro-apoptotic genes showing a significant higher expression in del(5q) compared to diploid patients. (b) Frequency of mutant CSNK1A1 that are co-mutated with TP53 (left) and frequency of TP53 mutants that include CSNK1A1 mutations (right). (c) Expression (Log2 CPM) of CSNK1A1 in del(5q) patients and diploid patients. (d) RPS14 expression (Log2 CPM) comparison between diploid patients with low expression (le) of CSNK1A1 and diploid patients with normal expression of CSNK1A1. (e) CSNK1A1 expression comparison between del(5q) patients WT (solid colour) or MT (grey stripes) for CSNK1A1, TP53. Diploid WT patient expression is also depicted. (f) CDKN1A expression comparison between del(5q) patients WT (solid colour) or MT (grey stripes) for the CSNK1A1 and TP53 genes. Diploid WT patient expression is also depicted. (g) CDKN1A expression in del(5q) vs. del(5q) low RPS14 expressors and diploid patients vs. diploid low RPS14 expressors (cluster-2). (h) Correlation between the percentage of del(5q) (according to CNV analysis) and expression (Log2 CPM) of excluded genes located on 5q. Green error bar represents the expression of the given gene in diploid patients, TP53 wild type, and <5% bone mar- row blasts. Black error bar represents the estimated expression of the given gene at 50% and 100% of del(5q) clonality. [Isolated del (5q) patients (blue line), compound del(5q) (pink line)]. (i) EGR1 expression in del(5q) and diploid cases TP53 mutant (solid colour) or wild type (grey stripes). (j) Correlation between % of del(5q) (according to CNV analysis) and expression (log2 CPM). Dots represents expression and clon- ality of patients without deletion involving APC or DDX41. (k) APC expression in del(5q) and diploid cases. On the right frequency of CSNK1A1 mutants and wild type in del(5q) APC low expressors. (l) DDX41 expression in patients with del(5q) and DDX41 deleted, del(5q), and diploid cases. www.thelancet.com Vol 80 Month June, 2022 11 Articles also some new genes such as C5orf25 (TCERG1).20 Thus, In summary, we provide an integrated analysis of the del(5q) results in a complex interaction between the HI- genome of del(5q) patients. We identified key HI genes anti-drivers which resist the expansion promoted by the (drivers vs. anti-drivers) and gene clusters, that together HI-driver genes. The acquisition of accelerator hits will with acquired mutational events and their hierarchy shift the balance between the proapoptotic and pro-pro- might determine the clinical course and complexity of liferative functions in favour to the last. Of note is that del(5q). We also provide a substrate for future genera- we included as accelerators CSNK1A1 gain of function tion of mouse models for confirmatory and genetics mutations which shift the balance towards pro-prolifer- studies to further characterize the HI-drivers and HI- ative function Wnt/b-catenin.18 TP53 alterations may anti-drivers in del(5q). cooperate with the pro-tumorigenic effects or cyto- genetic alterations like monosomy 7 or del(17p) may ultimately lead to a more aggressive phenotype. Finally, Contributors EGR1 upregulation in response to apoptotic stress will V.A. carried out the experimental/analytic procedures, be attenuated via TP53 inactivation,21,22 consistent with interpreted the data, and wrote the manuscript. L.P., M. our results. M., and V.V. collected data and performed DNA molecu- HI-anti-driver genes include e.g., RPS14, HSPA4, lar studies. W.W., M.M., S.H., T.L., T.R., and Jo.B. con- SIL-1, and UBE2D2 and all promoted the increased apo- ducted bioinformatic analysis. Jo.B. established statistical ptosis in del(5q).23 Whereas we show that RPS14 is not analysis and conducted haploinsufficiency analysis. V.V., uniformly HI, our results are otherwise consistent with L.A., B.X, A.P., J.B., M.S., provided clinical specimens earlier reports,24,25 as del(5q) cases with RPS14 deletion and important insights on the manuscript. T.H., C.H., did harbour most of the TP53 mutations. Together with and W.K. provided molecular data RNA-Seq and WGS or in the absence of RPS14, HI of HSPA4 and SIL1 and helped to interpret results and edit the manuscript. genes may also contribute to cell cycle arrest and apo- J.P.M, T.H., and FS supervised the study and provided ptotic pressure in del(5q).26 PURA is another del(5q) patient specimens and data. V.V. and J.P.M. have verified gene which represses TP53 and BAX and thus its HI the underlying data. J.P.M. formulated and edited the promotes apoptosis.27 Interestingly, PURA can either manuscript. J.P.M. and F.S. equally contributed. All function as a homodimer (activating the transcrip- authors read and approved the final manuscript. tion) or as a repressive heterodimer with PURB located on chromosome 7p. Simultaneous deletion of PURA and PURB correlates with an enhanced pro- Declaration of interests gression to sAML in MDS.28 Another HI-anti driver Dr. Sekeres received consultation fees from BMS/Cel- gene, UBE2D2 may also be consequential to del(5q) gene and Kurome. All the other authors declare no con- pathogenesis. Knock-down of UBE2D2 decreases flicts interests. TP53 ubiquitination and degradation.15 Finally, increased expression of CDKN1A (p21), caspase-3, -8, and -9 and BAX genes supports notions of high apo- Data sharing statement ptotic pressure in del(5q) cells. All information are provided in the appendix. Requests The recently published single cell barcoding strategy for additional information can be sent to the corre- by knockout of HI loci of Csnk1a1, Apc, and Egr1 showed sponding author. RNA-sequencing results have been that only HI of Csnk1a1 resulted in a clonal expansion as reported in Hershberger CE et al., Leukemia 2021. shown in serial transplantations, while HI of Apc or Egr1 did not.29 Our study is compatible with these experiments also assigning a prominent role of Acknowledgements CSNK1A1 in the clonal expansion, despite its duality in This work was supported in part by Torsten Haferlach function (apoptosis vs. enhancing the Wnt/b-catenin Leukemia Diagnostics Foundation. US National Insti- pathway). The latter pathway can be augmented by the tute of Health (NIH) grants R35 HL135795, CSNK1A1 gain of function mutations. A similar effect R01HL123904, R01 HL118281, R01 HL128425, R01 may be a result of APC down-modulation in deleted HL132071 (J.P. Maciejewski), and a grant from cases, as previously described.30 Edward P. Evans Foundation (J.P. Maciejewski). US Despite considerable effort, our study may suffer National Institute of Health (NIH) grants R01 from the obvious shortcomings inherent to retrospec- CA217992, R01 LM013067, R21 CA248138 (T. tive study with samples collected at different times. LaFramboise). Blood Cancer UK, grants 13042 and Another limitation is that genomic and expression data 19004 (A. Pellagatti and J. Boultwood). Instituto de were based on bulk DNA and RNA. Further studies Salud Carlos III, Ministerio de Economia y Competi- with single cell technologies are needed to determine vidad PI/14/00013; PI/17/0575 (F Sole); 2017 the differences during haematopoiesis of del(5q) and SGR288 (GRC) Generalitat de Catalunya (F. Sole); non-del(5q) cells. economical support from CERCA Program/ 12 www.thelancet.com Vol 80 Month June, 2022 Articles Generalitat de Catalunya, and Fundacio Internacional 15 Saville MK, Sparks A, Xirodimas DP, et al. Regulation of p53 by the Josep Carreras and “la Caixa” Foundation (F. Sole). ubiquitin-conjugating enzymes UbcH5B/C in vivo. J Biol Chem. 2004;279(40):42169–42181. 16 Mossner M, Jann JC, Wittig J, et al. Mutational hierarchies in mye- lodysplastic syndromes dynamically adapt and evolve upon therapy Supplementary materials response and failure. Blood. 2016;128(9):1246–1259. 17 Ribezzo F, Snoeren IAM, Ziegler S, et al. Rps14, Csnk1a1 and Supplementary material associated with this article can miRNA145/miRNA146a deficiency cooperate in the clinical pheno- be found in the online version at doi:10.1016/j. type and activation of the innate immune system in the 5q- syn- ebiom.2022.104059. drome. Leukemia. 2019;33(7):1759–1772. 18 Smith AE, Kulasekararaj AG, Jiang J, et al. CSNK1A1 mutations and isolated del(5q) abnormality in myelodysplastic syndrome: a retro- References spective mutational analysis. Lancet Haematol. 2015;2(5):e212–e221. 1 Giagounidis AAN, Germing U, Haase S, et al. Clinical, morpholog- 19 Liu TX, Becker MW, Jelinek J, et al. Chromosome 5q deletion ical, cytogenetic, and prognostic features of patients with myelodys- and epigenetic suppression of the gene encoding alpha-catenin plastic syndromes and del(5q) including band q31. Leukemia. (CTNNA1) in myeloid cell transformation. Nat Med. 2007;13 2004;18(1):113–119. (1):78–83. 2 Arber DA, Orazi A, Hasserjian R, et al. The 2016 revision to the 20 Montes M, Coiras M, Becerra S, et al. Functional consequences for World Health Organization classification of myeloid neoplasms apoptosis by transcription elongation regulator 1 (TCERG1)-medi- and acute leukemia. Blood. 2016;127(20):2391–2405. ated Bcl-x and Fas/CD95 alternative splicing. PloS One. 2015;10 3 Van den Berghe H, Cassiman JJ, David G, Fryns JP, Michaux JL, (10):e0139812. Sokal G. Distinct haematological disorder with deletion of long 21 Stoddart A, Fernald AA, Wang J, et al. Haploinsufficiency of del(5q) arm of no. 5 chromosome. Nature. 1974;251(5474):437–438. genes, Egr1 and Apc, cooperate with Tp53 loss to induce acute mye- 4 Le Beau MM, Espinosa R, Neuman WL, et al. Cytogenetic and loid leukemia in mice. Blood. 2014;123(7):1069–1078. molecular delineation of the smallest commonly deleted region of 22 Yu J, Baron V, Mercola D, Mustelin T, Adamson ED. A network of chromosome 5 in malignant myeloid diseases. Proc Natl Acad Sci p73, p53 and Egr1 is required for efficient apoptosis in tumor cells. U S A. 1993;90(12):5484–5488. Cell Death Differ. 2007;14(3):436–446. 5 Boultwood J, Fidler C, Soularue P, et al. Novel genes mapping to the 23 Parker JE, Mufti GJ, Rasool F, Mijovic A, Devereux S, Pagliuca A. critical region of the 5q- syndrome. Genomics. 1997;45(1):88–96. The role of apoptosis, proliferation, and the Bcl-2-related proteins 6 Jerez A, Gondek LP, Jankowska AM, et al. Topography, clinical, in the myelodysplastic syndromes and acute myeloid leukemia sec- and genomic correlates of 5q myeloid malignancies revisited. J Clin ondary to MDS. Blood. 2000;96(12):3932–3938. Oncol Off J Am Soc Clin Oncol. 2012;30(12):1343–1349. 24 Schneider RK, Schenone M, Ferreira MV, et al. Rps14 haploinsuffi- 7 Makishima H, Yoshizato T, Yoshida K, et al. Dynamics of clonal evo- ciency causes a block in erythroid differentiation mediated by lution in myelodysplastic syndromes. Nat Genet. 2017;49(2):204–212. S100A8 and S100A9. Nat Med. 2016;22(3):288–297. 8 Ritchie ME, Phipson B, Wu D, et al. Limma powers differential 25 Boultwood J. The role of haploinsufficiency of RPS14 and p53 acti- expression analyses for RNA-sequencing and microarray studies. vation in the molecular pathogenesis of the 5q- syndrome. Pediatr Nucleic Acids Res. 2015;43(7):e47 Rep. 2011;3(Suppl 2):e10 9 Li L, Yao W. Fully Bayesian logistic regression with hyper-LASSO 26 Rosam M, Krader D, Nickels C, et al. Bap (Sil1) regulates the molec- priors for high-dimensional feature selection. J Stat Comput Simul. ular chaperone BiP by coupling release of nucleotide and substrate. 2018;88(14):2827–2851. Nat Struct Mol Biol. 2018;25(1):90–100. 10 Schneider RK, Adema V, Heckl D, et al. Role of casein kinase 1A1 27 Kim K, Choi J, Heo K, et al. Isolation and characterization of a in the biology and targeted therapy of del(5q) MDS. Cancer Cell. novel H1.2 complex that acts as a repressor of p53-mediated tran- 2014;26(4):509–520. scription. J Biol Chem. 2008;283(14):9113–9126. 11 Heuser M, Meggendorfer M, Cruz MMA, et al. Frequency and 28 Lezon-Geyda K, Najfeld V, Johnson EM. Deletions of PURA, at prognostic impact of casein kinase 1A1 mutations in MDS patients 5q31, and PURB, at 7p13, in myelodysplastic syndrome and pro- with deletion of chromosome 5q. Leukemia. 2015;29(9):1942–1945. gression to acute myelogenous leukemia. Leukemia. 2001;15 12 Polprasert C, Schulze I, Sekeres MA, et al. Inherited and somatic (6):954–962. defects in DDX41 in myeloid neoplasms. Cancer Cell. 2015;27 29 Stoddart A, Fernald AA, Wang J, et al. Haploinsufficiency of del(5q) (5):658–670. genes, Egr1 and Apc, cooperate with Tp53 loss to induce acute mye- 13 Maciejewski JP, Padgett RA, Brown AL. M€ uller-Tidow C. DDX41- loid leukemia in mice. Blood. 2014;123(7):1069–1078. related myeloid neoplasia. Semin Hematol. 2017;54(2):94–97. 30 Li L, Sheng Y, Li W, et al. b-Catenin is a candidate therapeutic tar- 14 McKenna A, Hanna M, Banks E, et al. The genome analysis toolkit: get for myeloid neoplasms with del(5q). Cancer Res. 2017;77 A MapReduce framework for analyzing next-generation DNA (15):4116–4126. sequencing data. Genome Res. 2010;20(9):1297–1303. www.thelancet.com Vol 80 Month June, 2022 13