An artificial intelligence-assisted clinical framework to facilitate diagnostics and translational discovery in hematologic neoplasia.

BACKGROUND: The increasing volume and intricacy of sequencing data, along with other clinical and diagnostic data, like drug responses and measurable residual disease, creates challenges for efficient clinical comprehension and interpretation. Using paediatric B-cell precursor acute lymphoblastic leukaemia (BCP-ALL) as a use case, we present an artificial intelligence (AI)-assisted clinical framework clinALL that integrates genomic and clinical data into a user-friendly interface to support routine diagnostics and reveal translational insights for hematologic neoplasia. METHODS: We performed targeted RNA sequencing in 1365 cases with haematological neoplasms, primarily paediatric B-cell precursor acute lymphoblastic leukaemia (BCP-ALL) from the AIEOP-BFM ALL study. We carried out fluorescence in situ hybridization (FISH), karyotyping and arrayCGH as part of the routine diagnostics. The analysis results of these assays as well as additional clinical information were integrated into an interactive web interface using Bokeh, where the main graph is based on Uniform Manifold Approximation and Projection (UMAP) analysis of the gene expression data. At the backend of the clinALL, we built both shallow machine learning models and a deep neural network using Scikit-learn and PyTorch respectively. FINDINGS: By applying clinALL, 78% of undetermined patients under the current diagnostic protocol were stratified, and ambiguous cases were investigated. Translational insights were discovered, including IKZF1plus status dependent subpopulations of BCR::ABL1 positive patients, and a subpopulation within ETV6::RUNX1 positive patients that has a high relapse frequency. Our best machine learning models, LDA and PASNET-like neural network models, achieve F1 scores above 97% in predicting patients' subgroups. INTERPRETATION: An AI-assisted clinical framework that integrates both genomic and clinical data can take full advantage of the available data, improve point-of-care decision-making and reveal clinically relevant insights promptly. Such a lightweight and easily transferable framework works for both whole transcriptome data as well as the cost-effective targeted RNA-seq, enabling efficient and equitable delivery of personalized medicine in small clinics in developing countries. FUNDING: German Ministry of Education and Research (BMBF), German Research Foundation (DFG) and Foundation for Polish Science.

Cryptic TCF3 fusions in childhood leukemia: Detection by RNA sequencing.

Acute lymphoblastic leukemia (ALL) is the most frequent malignancy in childhood and adolescence. In more than 60% of cases of this heterogeneous disease, a genetic marker is identified via cytogenetic or molecular analyses. TCF3 gene fusions occur in 5%-11% of ALL patients. In < 1%, the TCF3 alteration in ALL leads to a TCF3-HLF fusion gene. Even though this is a very rare event, the detection of a TCF3-HLF fusion gene is associated with a very poor prognosis with incurable relapses in almost all patients. The frequent TCF3-PBX1 fusion gene, which is detectable in 5%-10% of childhood B-cell precursor ALLs and ~3.8% of adult B-cell precursor ALLs, is associated with a rather good prognosis, that is, an observed event-free 5-year survival of approximately 85%. Thus, the distinction of the different partner genes fused to TCF3 is essential for risk assessment. To verify RNA sequencing as a tool for detection of known and unknown fusion genes, we screened 200 cases of pediatric B-cell precursor ALL with "targeted" RNA sequencing in a pilot project in comparison to classical cytogenetic analyses (chromosome R-banding analysis), fluorescence in situ hybridization, and PCR. We observed a TCF3 fusion gene in 6.5% (13/200) of the patients. Ten (5%) patients displayed a TCF3-PBX1 fusion gene, two (1%) patients a TCF3-FLI1 fusion gene, and one (0.5%) patient a TCF3-HLF fusion gene. For the TCF3 fusions, we obtained discrepant results with the different methods, which are described in the article. Taken together, translocations leading to TCF3 fusion genes might appear cryptic and may remain undetected by a single method.

Frequency and prognostic impact of PAX5 p.P80R in pediatric acute lymphoblastic leukemia patients treated on an AIEOP-BFM acute lymphoblastic leukemia protocol.

PAX5 is a member of the paired box (PAX) family of transcription factors involved in B-cell development. PAX5P80R has recently been described as a distinct genetic B-cell precursor (BCP) acute lymphoblastic leukemia (ALL) subtype with a favorable prognosis in adults. In contrast, an unfavorable outcome has been observed in children. Our aim was to determine the frequency of PAX5P80R in childhood BCP-ALL treated according to the Associazione Italiana Ematologia ed Oncologia Pediatrica-Berlin-Frankfurt-Muenster (AIEOP-BFM) ALL 2000 protocol and to evaluate its clinical significance within this study cohort. The analyses included 1237 patients with ALL treated in the AIEOP-BFM ALL 2000 trial with complete information for copy number variations (CNVs) of IKZF1, PAX5, ETV6, RB1, BTG1, EBF1, CDKN2A, CDKN2B, and ERG. A customized TaqMan genotyping assay was used to screen for PAX5P80R . Sanger sequencing was used to confirm PAX5P80R -positive results as well as to screen for second variants in PAX5. Agilent CGH + SNP arrays (e-Array design 85 320; Agilent Technologies) were performed in PAX5P80R -positive patients to verify additional CNVs. Almost 2% (20/1028) of our BCP-ALL cohort were PAX5P80R -positive. White blood cell counts higher than 50 000/µl as well as male sex were significantly (P < .05) associated with PAX5P80R . Most of the PAX5P80R -positive cases were 10 years of age or older. PAX5P80R -positive samples were enriched for deletions affecting PAX5, IKZF1, CDKN2A, and CDKN2B. Compared to PAX5P80R -wildtype BCP-ALL, PAX5P80R -positive patients showed a significantly reduced 5-year overall survival (P = .042). Further studies should evaluate the interaction of PAX5P80R with other genetic aberrations to further stratify intermediate risk pediatric BCP-ALL.

Implementation of RNA sequencing and array CGH in the diagnostic workflow of the AIEOP-BFM ALL 2017 trial on acute lymphoblastic leukemia.

Risk-adapted therapy has significantly contributed to improved survival rates in pediatric acute lymphoblastic leukemia (ALL) and reliable detection of chromosomal aberrations is mandatory for risk group stratification. This study evaluated the applicability of panel-based RNA sequencing and array CGH within the diagnostic workflow of the German study group of the international AIEOP-BFM ALL 2017 trial. In a consecutive cohort of 117 children with B cell precursor (BCP) ALL, array analysis identified twelve cases with an IKZF1plus profile of gene deletions and one case of masked hypodiploidy. Genetic markers BCR-ABL1 (n = 1), ETV6-RUNX1 (n = 25), and rearrangements involving KMT2A (n = 3) or TCF3 (n = 3) were assessed by established conventional techniques such as karyotyping, FISH, and RT-PCR. Comparison of these results with RNA sequencing analysis revealed overall consistency in n=115/117 cases, albeit with one undetected AFF1-KMT2A fusion in RNA sequencing and one undetected ETV6-RUNX1 fusion in conventional analyses. The combined application of RNA sequencing, FISH, and CGH+SNP array reliably detected all genetic markers necessary for risk stratification and will be used as the diagnostic standard workflow for BCP-ALL patients enrolled in the AIEOP-BFM ALL 2017 study. Prospectively, consistent collection of genome-wide CGH+SNP array as well as RNA sequencing data will be a valuable source to elucidate new prognostic lesions beyond established markers of pediatric ALL. In this respect, RNA sequencing identified various gene fusions in up to half of the IKZF1plus (n = 6/12) and B-other (n = 19/36) cases but not in cases with hyperdiploid karyotypes (n = 35). Among these fusions, this study reports several previously undescribed in frame PAX5 fusions, including PAX5-MYO1G and PAX5-NCOA6.

From a variant of unknown significance to pathogenic: Reclassification of a large novel duplication in BRCA2 by high-throughput sequencing.

BACKGROUND: Germline mutations in BRCA1/2 significantly contribute to hereditary breast and/or ovarian cancer. Here, we report a novel BRCA2 duplication of exons 22-24 in a female patient with bilateral breast cancer at age 35 and 44. The duplicated region was initially detected by gene panel sequencing and multiplex ligation-dependent probe amplification. However, the location and orientation of the duplicated region was unknown. Therefore, it was initially classified as a variant of unknown significance. METHODS: The spatial directional characterization of the BRCA2 duplication was achieved by targeted enrichment of the whole-genomic BRCA2 locus including exons and introns, and subsequent high-throughput sequencing. Subsequently, bioinformatics tools and a breakpoint-spanning PCR were used for identification of location and orientation of the duplication. RESULTS: The duplicated region was arranged in tandem and direct orientation (Chr13(GRCh37):g.32951579_32960394dup; NM_000059.3 c.8754 + 651_9256+6112dup p.(Ala3088Phefs*3)). It is predicted to result in a frameshift and a premature stop codon likely triggering nonsense-mediated mRNA decay. Consequently, it is regarded as pathogenic. CONCLUSION: This case study demonstrates that a comprehensive characterization of a structural variant by breakpoint assessment is crucial for its correct classification. Therefore, sequencing strategies including non-coding regions might be necessary to identify cancer predispositions in affected families.

The identification of pathogenic variants in BRCA1/2 negative, high risk, hereditary breast and/or ovarian cancer patients: High frequency of FANCM pathogenic variants.

NGS-based multiple gene panel resequencing in combination with a high resolution CGH-array was used to identify genetic risk factors for hereditary breast and/or ovarian cancer in 237 high risk patients who were previously tested negative for pathogenic BRCA1/2 variants. All patients were screened for pathogenic variants in 94 different cancer predisposing genes. We identified 32 pathogenic variants in 14 different genes (ATM, BLM, BRCA1, CDH1, CHEK2, FANCG, FANCM, FH, HRAS, PALB2, PMS2, PTEN, RAD51C and NBN) in 30 patients (12.7%). Two pathogenic BRCA1 variants that were previously undetected due to less comprehensive and sensitive methods were found. Five pathogenic variants are novel, three of which occur in genes yet unrelated to hereditary breast and/or ovarian cancer (FANCG, FH and HRAS). In our cohort we discovered a remarkably high frequency of truncating variants in FANCM (2.1%), which has recently been suggested as a susceptibility gene for hereditary breast cancer. Two patients of our cohort carried two different pathogenic variants each and 10 other patients in whom a pathogenic variant was confirmed also harbored a variant of unknown significance in a breast and ovarian cancer susceptibility gene. We were able to identify pathogenic variants predisposing for tumor formation in 12.3% of BRCA1/2 negative breast and/or ovarian cancer patients.

Routes of Clonal Evolution into Complex Karyotypes in Myelodysplastic Syndrome Patients with 5q Deletion.

Myelodysplastic syndrome (MDS) can easily transform into acute myeloid leukemia (AML), a process which is often associated with clonal evolution and development of complex karyotypes. Deletion of 5q (del(5q)) is the most frequent aberration in complex karyotypes. This prompted us to analyze clonal evolution in MDS patients with del(5q). There were 1684 patients with low and intermediate-risk MDS and del(5q) with or without one additional cytogenetic abnormality, who were investigated cytogenetically in our department, involving standard karyotyping, fluorescence in situ hybridization (FISH) and multicolor FISH. We identified 134 patients (8%) with aspects of clonal evolution. There are two main routes of cytogenetic clonal evolution: a stepwise accumulation of cytogenetic events over time and a catastrophic event, which we defined as the occurrence of two or more aberrations present at the same time, leading to a sudden development of highly complex clones. Of the 134 patients, 61% underwent a stepwise accumulation of events whereas 39% displayed a catastrophic event. Patients with isolated del(5q) showed significantly more often a stepwise accumulation of events rather than a catastrophic event. The most frequent aberrations in the group of stepwise accumulation were trisomy 8 and trisomy 21 which were significantly more frequent in this group compared to the catastrophic event group. In the group with catastrophic events, del(7q)/-7 and del(17p)/-17 were the most common aberrations. A loss of 17p, containing the tumor suppressor gene TP53, was found significantly more frequent in this group compared to the group of stepwise accumulation. This leads to the assumption that the loss of TP53 is the driving force in patients with del(5q) who undergo a sudden catastrophic event and evolve into complex karyotypes.

Breast cancer patients suggestive of Li-Fraumeni syndrome: mutational spectrum, candidate genes, and unexplained heredity.

BACKGROUND: Breast cancer is the most prevalent tumor entity in Li-Fraumeni syndrome. Up to 80% of individuals with a Li-Fraumeni-like phenotype do not harbor detectable causative germline TP53 variants. Yet, no systematic panel analyses for a wide range of cancer predisposition genes have been conducted on cohorts of women with breast cancer fulfilling Li-Fraumeni(-like) clinical diagnostic criteria. METHODS: To specifically help explain the diagnostic gap of TP53 wild-type Li-Fraumeni(-like) breast cancer cases, we performed array-based CGH (comparative genomic hybridization) and panel-based sequencing of 94 cancer predisposition genes on 83 breast cancer patients suggestive of Li-Fraumeni syndrome who had previously had negative test results for causative BRCA1, BRCA2, and TP53 germline variants. RESULTS: We identified 13 pathogenic or likely pathogenic germline variants in ten patients and in nine genes, including four copy number aberrations and nine single-nucleotide variants or small indels. Three patients presented as double-mutation carriers involving two different genes each. In five patients (5 of 83; 6% of cohort), we detected causative pathogenic variants in established hereditary breast cancer susceptibility genes (i.e., PALB2, CHEK2, ATM). Five further patients (5 of 83; 6% of cohort) were found to harbor pathogenic variants in genes lacking a firm association with breast cancer susceptibility to date (i.e., Fanconi pathway genes, RECQ family genes, CDKN2A/p14ARF, and RUNX1). CONCLUSIONS: Our study details the mutational spectrum in breast cancer patients suggestive of Li-Fraumeni syndrome and indicates the need for intensified research on monoallelic variants in Fanconi pathway and RECQ family genes. Notably, this study further reveals a large portion of still unexplained Li-Fraumeni(-like) cases, warranting comprehensive investigation of recently described candidate genes as well as noncoding regions of the TP53 gene in patients with Li-Fraumeni(-like) syndrome lacking TP53 variants in coding regions.

Telomere shortening, TP53 mutations and deletions in chronic lymphocytic leukemia result in increased chromosomal instability and breakpoint clustering in heterochromatic regions.

Complex karyotypes are associated with a poor prognosis in chronic lymphocytic leukemia (CLL). Using mFISH, iFISH, and T/C-FISH, we thoroughly characterized 59 CLL patients regarding parameters known to be involved in chromosomal instability: status of the genes ATM and TP53 and telomere length. Interestingly, a deletion of the ATM locus in 11q, independent of the cytogenetic context, was associated with significantly diminished risk (p<0.05) of carrying a mutation in TP53. In patients with loss or mutation of TP53, chromosomal breakage occurred more frequently (p<0.01) in (near-) heterochromatic regions. Median telomere length in patients with complex karyotypes was significantly shorter than that of healthy controls and shorter than in all other cytogenetic cohorts. Furthermore, the median telomere length of patients carrying a TP53 mutation was significantly shorter than without mutation. We conclude that telomere shortening in combination with loss of TP53 induces increased chromosomal instability with preferential involvement of (near-) heterochromatic regions.

GT198 (PSMC3IP) germline variants in early-onset breast cancer patients from hereditary breast and ovarian cancer families.

GT198, located 470 kb downstream of BRCA1, encodes for the nuclear PSMC3-interacting protein, which functions as co-activator of steroid hormone-mediated gene expression, and is involved in RAD51 and DMC1-mediated homologous recombination during DNA repair of double-strand breaks. Recently, germline variants in GT198 have been identified in hereditary breast and ovarian cancer (HBOC) patients, mainly in cases with early-onset. We screened a cohort of 166 BRCA1/2 mutation-negative HBOC patients, of which 56 developed early-onset breast cancer before the age of 36 years, for GT198 variants. We identified 7 novel or rare GT198 variants in 8 out of 166 index patients: c.-115G>A (rs191843707); c.-70T>A (rs752276800); c.-37A>T (rs199620968); c.-24C>G (rs200359709); c.519G>A p.(Trp173*); c.537+51G>C (rs375509656); c.*24G>A. Three out of 7 identified variants (c.-115G>A, c.519G>A and c.*24G>A) with putative pathogenic impact were found in HBOC patients with breast cancer onset at ≤ 36 years. The nonsense mutation c.519G>A p.(Trp173*) was located within the DNA binding domain of GT198 and is predicted to induce nonsense-mediated mRNA decay. Functional analyses of c.-115G>A, and c.*24A>G indicated an influence of these variants on gene expression. This is the second study that gives evidence for an association between pathogenic GT198 germline variants and early-onset breast cancer in HBOC.

Haploinsufficiency of ETV6 and CDKN1B in patients with acute myeloid leukemia and complex karyotype.

BACKGROUND: Acute myeloid leukemia with complex karyotype (CK-AML) is a distinct biological entity associated with a very poor outcome. Since complex karyotypes frequently contain deletions of the chromosomal region 12p13 encompassing the tumor suppressor genes ETV6 and CDKN1B, we aimed to unravel their modes of inactivation in CK-AML. RESULTS: To decipher deletions, mutations and methylation of ETV6 and CDKN1B, arrayCGH, SNP arrays, direct sequencing of all coding exons and pyrosequencing of the 5'UTR CpG islands of ETV6 and CDKN1B were performed. In total, 39 of 79 patients (49%) showed monoallelic deletions of 12p13 according to karyotypic data and 20 of 43 patients (47%) according to genomic profiling. Genomic profiling led to the minimal deleted region covering the 3'-UTR of ETV6 and CDKN1B. Direct sequencing revealed one novel monoallelic frameshift mutation in ETV6 while no mutations in CDKN1B were identified. Furthermore, methylation levels of ETV6 and CDKN1B did not indicate transcriptional silencing of any of these genes. ETV6 and CDKN1B had reduced expression levels in CK-AML patients with deletion in 12p13 as compared to CK-AML without deletion in 12p13, while the other genes (BCL2L14, LRP6, DUSP16 and GPRC5D) located within the minimal deleted region in 12p13 had very low or missing expression in CK-AML irrespective of their copy number status. CONCLUSIONS: ETV6 and CDKN1B are mainly affected by small monoallelic deletions, whereas mutations and hypermethylation play a minor role in CK-AML. Reduced gene dosage led to reduced gene expression levels, pointing to haploinsufficiency as the relevant mechanism of inactivation of ETV6 and CDKN1B in CK-AML.

Histone methyltransferase Suv39h1 deficiency prevents Myc-induced chromosomal instability in murine myeloid leukemias.

Suv39h1 mediates heterochromatin formation in pericentric and telomeric regions by trimethylation of lysine 9 of histone 3 (H3K9me3). Yet, its role in the induction of chromosomal instability is poorly understood. We established a leukemia model by retrovirally expressing Myc in wild-type and histone methyltransferase Suv39h1-deficient hematopoietic cells and characterized the resulting leukemias for chromosomal instability. All mice that received cells overexpressing Myc developed myeloid leukemia with a median survival of 44 days posttransplantation. Myc-overexpressing wild-type leukemias demonstrated clones with numerical chromosomal aberrations (5/16). In secondary transplantations of these leukemic cells, structural changes, mostly end-to-end fusions of chromosomes, appeared (10/12). In contrast, leukemic cells overexpressing Myc with reduced or no Suv39h1 expression had a normal karyotype in primary, secondary, and tertiary transplantations (16/16). Myc-transduced Suv39h1-deficient cells showed less critically short telomeres (P < 0.05) compared with Myc-transduced wild-type bone marrow cells. Gene expression analysis showed upregulation of genes involved in the alternative lengthening of telomeres (ALT) mechanism. Thus, we hypothesize that loss of Suv39h1 implies activation of the ALT mechanism, in turn ensuring telomere length and stability. Our data show for the first time that Suv39h1 deficiency may prevent chromosomal instability by more efficient telomere stabilization in hematopoietic bone marrow cells overexpressing Myc.

Cytogenetic follow-up by karyotyping and fluorescence in situ hybridization: implications for monitoring patients with myelodysplastic syndrome and deletion 5q treated with lenalidomide.

In patients with low and intermediate risk myelodysplastic syndrome and deletion 5q (del(5q)) treated with lenalidomide, monitoring of cytogenetic response is mandatory, since patients without cytogenetic response have a significantly increased risk of progression. Therefore, we have reviewed cytogenetic data of 302 patients. Patients were analyzed by karyotyping and fluorescence in situ hybridization. In 85 patients, del(5q) was only detected by karyotyping. In 8 patients undergoing karyotypic evolution, the del(5q) and additional chromosomal aberrations were only detected by karyotyping. In 3 patients, del(5q) was only detected by fluorescence in situ hybridization, but not by karyotyping due to a low number of metaphases. Karyotyping was significantly more sensitive than fluorescence in situ hybridization in detecting the del(5q) clone. In conclusion, to optimize therapy control of myelodysplastic syndrome patients with del(5q) treated with lenalidomide and to identify cytogenetic non-response or progression as early as possible, fluorescence in situ hybridization alone is inadequate for evaluation. Karyotyping must be performed to optimally evaluate response.

Clonal heterogeneity in childhood myelodysplastic syndromes--challenge for the detection of chromosomal imbalances by array-CGH.

To evaluate whether copy number alterations (CNAs) are present that may contribute to disease development and/or progression of childhood myelodysplastic syndromes (MDS), 36 pediatric MDS patients were analyzed using array-based comparative genome hybridization (aCGH). In addition to monosomy 7, the most frequent chromosome aberration in childhood MDS, novel recurrent CNAs were detected. They included a loss of 3p14.3-p12.3, which contains the putative tumor suppressor gene FHIT, a loss of 7p21.3-p15.3, a loss of 9q33.3-q34.3 (D184) and microdeletions in 17p11.2, 6q23 containing MYB, and 17p13 containing TP53. In this small patient cohort, patients without CNA, patients with monosomy 7 only and patients with one CNA in addition to monosomy 7 did not differ in their survival. As expected, all patients with complex karyotypes, including two patients with deletions of TP53, died. A challenge inherent to aCGH analysis of MDS is the low percentage of tumor cells. We evaluated several approaches to overcome this limitation. Genomic profiles from isolated granulocytes were of higher quality than those from bone marrow mononuclear cells. Decreased breakpoint calling stringency increased recognition of CNAs present in small clonal populations. However, further analysis using a custom-designed array showed that these CNAs often did not confirm the findings from 244k arrays. In contrast, constitutional CNVs were reliably detected on both arrays. Moreover, aCGH on amplified DNA from distinct myeloid clusters is a new approach to determine CNAs in small subpopulations. Our results clearly emphasize the need to verify array-CGH results by independent methods like FISH or quantitative PCR.

Telomere shortening and chromosomal instability in myelodysplastic syndromes.

Telomere shortening and chromosomal instability are believed to play an important role in the development of myeloid neoplasia. So far, published data are only available on the average telomere length in myelodysplastic syndromes (MDS) and acute myeloid leukemia (AML), but not on the telomere length of individual chromosomes. We used a new technique, telomere/centromere-fluorescence in situ hybridization (T/C-FISH), which combines fluorescence R-banding and FISH using a probe against the telomere repeats to measure the telomere length of each chromosome arm in 78 patients with MDS. In line with the previous results, patients with MDS showed significantly shorter telomeres than those of healthy controls. Telomere lengths did not differ significantly between distinct morphological subtypes of MDS. However, there was a significant difference in telomere length between patients with an isolated monosomy 7 and patients with a normal karyotype (P < 0.05). Notably, patients with an isolated monosomy 7 showed significantly longer telomeres than patients with a normal karyotype in many chromosome arms, among them 7p and 7q. Neo-telomeres were found in two patients with a complex karyotype, in one case at the fusion site of a dic(14;20). Normal and aberrant metaphases of the same patient did not differ in telomere length, thus indicating to telomere shortening as a basic mechanism affecting all hematopoietic cells in patients with MDS. In some MDS subtypes, like MDS with isolated monosomy 7, telomeres may be stabilized and even increase in length because of the activation of telomerase or alternative mechanisms.

Analysis of array-CGH data using the R and Bioconductor software suite.

BACKGROUND: Array-based comparative genomic hybridization (array-CGH) is an emerging high-resolution and high-throughput molecular genetic technique that allows genome-wide screening for chromosome alterations. DNA copy number alterations (CNAs) are a hallmark of somatic mutations in tumor genomes and congenital abnormalities that lead to diseases such as mental retardation. However, accurate identification of amplified or deleted regions requires a sequence of different computational analysis steps of the microarray data. RESULTS: We have developed a user-friendly and versatile tool for the normalization, visualization, breakpoint detection, and comparative analysis of array-CGH data which allows the accurate and sensitive detection of CNAs. CONCLUSION: The implemented option for the determination of minimal altered regions (MARs) from a series of tumor samples is a step forward in the identification of new tumor suppressor genes or oncogenes.