This is a retrospective single-center case series of 56 pediatric patients with high-risk hematologic malignancies and blood disorders treated at Columbia University Medical Center (CUMC) between January 2014 and June 2017. Twenty-eight cases have been previously reported (11). High risk was defined as having a prognosis of <50% overall survival at 5 years, having relapsed disease, being refractory to previous therapies, or having a rare hematologic disorder or unusual presentation for age. When an underlying hereditary cancer predisposition was suspected, patients and their parents were also offered constitutional WES.

Participants signed consent for constitutional or tumor-normal WES either as part of an Institutional Review Board (IRB)-approved protocol (AAAB7109) or they signed the clinical consent.¹ Written consent for clinical WES was obtained after the risks and benefits were explained to the patient and/or caregiver, which included the potential disclosure of medically actionable secondary findings, defined as germline disease-causing mutations unrelated to the condition for which sequencing was being performed. Patients could opt-in/opt-out of learning secondary findings and/or having these results appearing in the electronic medical record (EMR), having their samples and/or data stored for future research, either with or without identifiers, and future contact. Results not reported included carrier status, variants of uncertain significance (VOUS) in secondary findings except as related to cancer, and mutations related to adult-onset conditions for which the genetic link is unclear or for which no known intervention is of proven benefit. Samples were obtained in the course of clinical care and either prospectively banked for future molecular testing (IRB-AAAB7109) or freshly collected at the time of relapse. IRB approval was obtained for this retrospective analysis of de-identified patient and clinical genomics data (IRB AAAQ8170 and AAAP1200).

Nucleic acid preparation and high-throughput sequencing were performed following previously described standard protocols in the Laboratory of Personalized Genomic Medicine at CUMC (11). Sequencing was performed on fresh or frozen bone marrow or peripheral blood containing at least 50% lesional cells. Flow-sorting was used to either purify or enrich tumoral cells in the event of mixed chimerism, in patients who had received an allogeneic stem cell transplant, or when the amount of tumoral cells was below the threshold. Fresh or formalin fixed paraffin embedded tissue was used in cases of organ infiltration or chloroma. Three CLIA-certified, College of American Pathologists and New York State Department of Health approved assays were used. WES of tumor and normal tissue (buccal swab or peripheral blood) and RNA sequencing of tumor tissue was performed when possible. DNA sequencing of 467 cancer-associated genes (Columbia Comprehensive Cancer Panel, CCCP) was used when available tumoral or normal tissue was inadequate. Constitutional WES was performed when a constitutional disease was suspected on samples from both patients and parents (trio).

Whole exome and transcriptome libraries were prepared as previously described (11) and sequenced on a HiSeq2500 using paired-end 125-cycle x2sequencing. Variant calls were independently made on tumor and germline material. Variants were subject to filtering. In normal DNA, variants were passed through a “reference range filter” of cancer predisposition genes, genes relevant to pharmacogenomics, and variants relevant to patient care; a “reportable range filter” which includes COSMIC (cosmic70 provided by Annovar) variants in the patient’s mutation report file and variants in genes recommended by American College of Medical Genetics (ACMG) for the reporting of secondary findings (14); as well as a frequency filter, which includes variants whose minor allele frequency in the 1,000 genomes (phase 1, version 3, release date 11/23/2010) is less than 1%. Somatic mutations in the tumor were identified by subtracting all variants called in normal tissue (output at minor allelic fraction of ≥5%) from variants called in the tumor (output at a minor allelic fraction of ≥10%). The approach maximized the number of variants output to minimize the likelihood of filtering out actionable mutations prior to molecular tumor board discussion. Copy number variation (CNV) was determined from the WES data and fusion transcripts and relative gene expression were identified from the RNA-seq data. Tumor and normal DNA and tumor RNA sequence analysis was previously described (11). Quality statistics for tumor-normal WES are presented in Table S1 in Supplementary Material.

After initial review by a molecular pathologist, interpretation of each tumor-normal WES and RNA-seq results was conducted in a weekly multidisciplinary molecular tumor board. The tumor board included members from pediatric oncology, molecular pathology, cytogenetics, medical genetics, cancer biology, and bioinformatics. Clinical relevance of variant calls, CNV, and fusions was based on published laboratory data documenting gain or loss of function mutations in known oncogenes and tumor suppressor genes, respectively or predicted by in silico algorithms or determined based on previously reported gene fusions or gene fusions expected to have the same effect as previously reported fusions involving one of the partner genes. Additionally, factors such as recurrence of the identified variant and allelic fraction were considered when addressing clinical significance (11). The final clinical report provided a tiered description of somatic mutations based on disease-association and evidence for clinical actionability. Tier 1 included known, tumor type-specific actionable somatic mutations; Tier 2 included somatic mutations in targetable pathways, actionable somatic mutations in other tumor types and somatic mutations in well-established cancer genes; Tier 3 included other somatic mutations in cancer genes; and Tier 4 included somatic VOUS. Reporting of germline variants was based on whether they were known to be pathogenic, recommended by the ACMG for reporting of secondary findings (14) or VOUS in cancer genes or genes potentially important for patient care. Accession number for all genes and fusions reported are provided in Table S2 in Supplementary Material. Reports were sent to the primary oncologists and posted to the EMR in accordance with patient opt-in/opt-out preferences selected at the time of informed consent. An overview of the PIPseq workflow is presented in Figure 1. Datasets are available through cBioPortal for Cancer Genomics² (15, 16).

For the purposes of this study, we defined clinically impactful findings as any molecular test result that, when integrated with a patient’s history, symptoms, and other clinical findings, could lead to a change in the assessment or management of the patient. This definition is in agreement with the recommendations provided by the Association for Molecular Pathology to evaluate the clinical utility of a molecular test (17). To evaluate the potential clinical utility of tumor and germline alterations, these clinically meaningful results were subcategorized into the following five categories: (1) diagnostic; (2) prognostic; (3) identification of a therapeutic target; (4) other clinically impactful information, including pharmacogenomics or findings that resulted in a significant refinement of a therapeutic plan; and (5) recommendations for health maintenance interventions or genetic counseling for the patient and other at-risk family members. Genetic alterations were considered “targetable” if: (1) an FDA approved drug or experimental drug was available that inhibited the target directly or inhibited its downstream signaling pathway; (2) there was pre-clinical evidence to support efficient targeting of the aberrant function of the mutated gene and/or potential clinical benefit; and (3) there was information on age-appropriate dosing. “Targetable” mutations were further categorized using a five-tier system, previously described (18, 19). This tiering system evaluates the potential clinical benefit of targeting an altered gene based on published preclinical and clinical data.

Molecular profiling was performed on 27 bone marrow or peripheral blood samples obtained from 20 patients with ALL (mean age, 9.5 years; median age, 8.0 years; range, 1.5–23.9 years) (Table 1). Three patients had T-ALL and 17 had B-ALL (Figure 2). Sequencing was performed at time of diagnosis in 10 patients and at time of relapse in the remaining 10 patients. Paired diagnostic and relapsed samples were available for sequencing from three patients while serial relapse and refractory samples were sequenced from two other patients. One patient with T-ALL had constitutional WES only, because of strong family history of cancer. The mean mutational load across patients was 705.2 variants (SD = 1,735.6; median = 93.0; range = 30–5,950). After filtering, 81 somatic variants were identified in 34 genes, most of them classified as Tier 2 (53/81, 65.4%) (Figure 3; Table S2 in Supplementary Material). The mean number of somatic variants per sample, after filtering, was 3.0 (median = 2.0; range = 0–13) and was not significantly different between relapsed and diagnostic samples (3.71 vs. 2.23, p = 0.19). Signaling pathways most frequently, aberrantly activated were the RAS pathway (NRAS, n = 8; KRAS, n = 4; PTPN11, n = 1) and the JAK-STAT pathway (IL7R, n = 2; SH2B3, n = 1; JAK3, n = 1). As expected, several transcription factors critical for the normal development and differentiation of B- and T-cells were also recurrently mutated, such as PAX5 (n = 4) and NOTCH1 (n = 5). Additionally, three B-ALL samples carried mutations in genes responsible for resistance to chemotherapy (CREBBP, n = 2; NT5C2, n = 1). Analysis of CNVs revealed loss of CDKN2A in two samples and IKZF1 in three samples.

Nine unique fusions were identified using RNA-seq in nine patients. Fusions were conserved in paired diagnostic and relapse samples, as well as, in serial relapsed and refractory samples and were the only driver alteration identified in 4/9 cases. Gene fusions involved a kinase in four cases and a transcription factor in the remaining five cases, likely causing constitutive activation of a signaling pathway and aberrant cell differentiation, respectively (Figure 3). Analysis of RNA-seq data also allowed the recognition of a BCR-ABL1-like ALL expression pattern in a patient with relapse ALL and no identifiable driver mutation. Further testing led to the identification of a NUP214-ABL1 fusion by real-time polymerase chain reaction (RT-PCR) (20) and subsequent addition of dasatinib to conventional chemotherapy. Interestingly, the analytic pipeline failed to identify the NUP214-ABL1 fusion despite adequate coverage and good quality of the specimen. Further analysis using three different pipelines, as well as, repeat RNA sequencing on a subsequent refractory sample, also failed to recover the fusion. We speculate that the fusion might have been present in a smaller subclone below the level of detection of current sequencing technologies.

The analysis of serial samples from five patients was particularly interesting (Figure 4). Three patients had B-ALL and two patients had early thymocyte precursor (ETP) ALL. In two patients with B-ALL, the gene fusion present at diagnosis was the only genetic alteration maintained at relapse, suggesting that the fusion represented the main driver present in the founding clone. Conversely, none of the somatic mutations found in the diagnostic samples were retained in the relapsed samples, possibly indicating the eradication of several subclones by the initial therapy. The third patient with B-ALL had relapsed MLL-rearranged infant ALL that switched to MLL-rearranged AML at the completion of one cycle of blinatumomab. In this case, the KMT2A-AFF1 fusion and all somatic gene mutations present in the relapse ALL sample were shared in the AML sample, but the AML sample had also acquired several new genetic alterations, some in genes more commonly mutated in myeloid diseases (e.g., TET1 and TET2). The two patients with ETP also experienced a relapse with phenotypic switch to a myeloid, monocytic leukemia. The first patient presented with relapsed ETP and, while receiving the first cycle of re-induction chemotherapy with standard chemotherapy (vincristine, prednisone, PEG-aspargase, and doxorubicin) plus everolimus, the leukemia evolved into myelomonocytic leukemia with loss of T-cell markers and expression of myeloid markers. The two samples (ETP-ALL and AML) carried identical genetic alterations. The patient experienced a second ETP-ALL relapse after stem cell transplant. In this sample, the activating mutation found in NOTCH1 at diagnosis was lost, but two new NOTCH1 mutations were acquired together with a loss of function mutation in FBXW7, suggesting that the leukemia was “addicted” to activated NOTCH1 signaling. Interestingly, while receiving palliative chemotherapy with a gamma-secretase inhibitor, the ETP-ALL again switched immunophenotype and evolved into AML. The second patient with ETP relapsed during standard chemotherapy as an isolated thymic granulocytic sarcoma. The diagnostic and the two relapse myeloid samples in this case also shared identical genetic alterations with both the same gene fusion, as well as, the somatic gene mutations identified in all samples.

Approximately 37% of gene variants and copy number losses identified by tumor-normal WES were considered targetable (Table 2). However, only 2 of the 15 patients (13%) carrying targetable mutations in their tumor received targeted therapy. In contrast, 60% of the transcriptome findings were targetable and 4/6 patients whose leukemia carried targetable gene fusions received effective targeted therapy (Table 2). For example, one patient with a second relapse of ALL occurring post-transplant achieved complete remission with no minimal residual disease 2 months after receiving standard maintenance therapy with dasatinib for a SMARCC2-PDGFRB fusion. Overall, 19/20 patients with ALL carried at least one targetable alteration.

The clinical impact of the genomic alterations identified in ALL through NGS goes well beyond the findings of targetable lesions, frequently in ways that are not easily quantifiable. In four patients, identification of an IKZF1 deletion, FOX1-ABL1 fusion, and a specific RNA-seq profile led to a refined diagnosis of BCR-ABL like ALL and related poor prognosis (20). In one patient with relapsed B-ALL, the finding of a NT5C2 mutation, responsible for resistance to nucleoside analogs (48, 49), had clear implications for subsequent therapy (Table 3). Last, for the two patients with ETP that evolved to AML, the identification of shared DNA and RNA alterations between the ALL and AML samples led to the conclusion that the AML was not a second cancer or a secondary AML, but rather a immunophenotypic switch of the same leukemic clone; this assessment clearly impacted the subsequent clinical management of both patients.

Twenty-eight bone marrow, peripheral blood, or chloroma samples, obtained from 23 patients with myeloid neoplasms, underwent molecular profiling (mean age, 9.9 years; median age, 10.0 years; range, 0.9–23.9 years) (Table 1). Sequencing was performed at diagnosis in 8 patients and at time of relapse in 15 patients. Paired diagnostic and relapsed samples were available for sequencing in two patients, while serial relapse and refractory samples were sequenced from two other patients. About 1 patient had relapsed chronic myeloid leukemia (CML), 1 patient had a myeloproliferative neoplasm (MPN), and the remaining 21 patients had AML (Figure 2). Among the 21 patients with AML, 10 were classified as AML with recurrent genetic abnormalities (50) [biallelic mutation of CEBPA, n = 1; NPM1, n = 2; KMT2A rearrangement, n = 2; inv16 (p13.1q22), n = 2; t(8;21)(q22;q22.1), n = 1; RUNX1 mutation, n = 2], one patient had AML associated with Down syndrome, one patient had AML with myelodysplasia related changes, and the remaining nine patients had AML not otherwise specified. In two patients suspected to have an underlying genetic predisposition to cancer, only constitutional WES was performed. Constitutional WES was also performed in one patient after WES findings on the tumor-normal samples raised the possibility of germline genetic alterations.

The mean mutational load across patients was 101.7 variants (SD = 89.74; median = 79.0; range = 14–457). After filtering, 77 somatic variants were identified in 48 genes, most of them classified as Tier 2 (69%) (Figure 3; Table S2 in Supplementary Material). All patients had at least one genomic alteration. The mean number of somatic variants per sample after filtering was 2.85 (median = 2.0; range = 0–13) with similar mean number in diagnostic and relapsed samples (3.6 vs. 2.4, p = 0.30). WT1 and FLT3 were the most commonly mutated genes (n = 8 and n = 7, respectively) as well as members of the RAS pathway (PTPN11, n = 5; NRAS, n = 4; and KRAS, n = 3). As expected, gene mutations commonly seen in adult AML, such as in TET2, DNMT3A, ASXL1, EZH2, and in members of the spliceosome machinery (50, 51), were only rarely encountered in our cohort. Internal tandem duplications of the FLT3 gene are not easily identifiable through NGS, however, the presence of FLT3ITD was confirmed in two AML samples through PCR. RNA sequencing identified seven unique gene fusions in 12 samples from 10 patients, of which four fusions were the product of cryptic chromosomal inversions and translocations not identified by traditional chromosomal banding. Unlike what was observed in the ALL samples, gene fusions identified in the AML samples never involved kinases, but rather involved transcription factors or co-activators and co-repressors of transcriptional complexes likely critical for the proper development and differentiation of hematopoietic precursors. These fusions are considered class II mutations in the classical “two-hit” model proposed more than a decade ago by Gilliland (52).

The analysis of the two diagnostic-relapse pairs and the serial relapsed samples showed a simple model of clonal evolution, different from the pattern of clonal evolution described in adult AML (Figure 4). One diagnostic AML sample, characterized by a KMT2A rearrangement, lost the FLT3 TKD mutation and gained a NRAS mutation at relapse; one AML relapsed sample with a NUP98-NSD1 fusion, gained only a new mutation in Runx1 and RIT1 in a subsequent refractory sample. In another patient with AML, which likely evolved from a myelodysplastic syndrome (MDS), the MDS and secondary AML samples shared a mutation in IDH1 and the AML gained additional mutations only in NRAS and WT1 genes. These patterns of clonal evolution are consistent with what was previously described in the TARGET cohort of paired diagnostic and relapsed samples in pediatric AML (8).

Twenty-seven of the 77 variants identified were considered targetable (35%) (Table 2) with 19/23 patients (82%) carrying at least one targetable genetic variant. However, only one patient received targeted therapy. The patient, with a previously known t(8;21)(q22;q22.1) AML and found at relapse to carry an activating mutation of c-KIT, received targeted therapy with imatinib; the total peripheral white blood cell count normalized and the blast count significantly decreased, allowing a sustained control of disease for ~9 months until imatinib was discontinued due to toxicity. Three of the seven fusions were also considered targetable, however, none of the patients received targeted therapy. Importantly, 2/77 somatic variants and 5/7 gene fusions carried prognostic significance, in one case (CBFA2T3-GLIS2 fusion) leading to the decision to proceed to stem cell transplant as consolidation therapy. Finally, findings from WES/RNA-seq led to the correct diagnosis in two cases. In one child, the finding of mutated PTPN11 and SETBP1, together with cytogenetics and clinical findings, led to the correct diagnosis of AML evolved from JMML rather than de novo AML. In a second patient, the initial diagnosis of AML with minimal differentiation was corrected to megakaryoblastic leukemia following the finding on CBFA2T3-GLIS2 fusion by RNA-seq (Table 3). Overall, the findings obtained by NGS of myeloid leukemia samples provided clinically relevant information in 20 patients (86%).

Two samples from two patients with hepatosplenic gamma-delta T-cell lymphoma (HSTCL) and two samples from two patients with anaplastic large cell lymphoma (ALCL) were sequenced (mean age, 13.8 years; median age, 16.0 years; range, 5–18 years) (Table 1; Figure 2). One patient also had constitutional WES because of family history of cancer. Tumor tissue was obtained at diagnosis in two cases and at relapse in the remaining two. The mean mutational load was 248.5 variants (SD = 374.1; median = 75.0; range = 36–808). After filtering, 12 somatic variants were identified in 9 genes (Figure 3; Table S2 in Supplementary Material). The mean number of somatic variants per sample was 3.0 (median = 2.5; range = 0–5). The two HSTCL samples shared activating mutations in KRAS and STAT5B, as well as, in genes involved in the epigenetic modulation of transcription. Importantly, mutations in KRAS and STAT5B were considered targetable, although neither patient received a targeted therapy (Table 2). The finding of a STAT5B mutation in one of the two patients, together with the cytogenetic finding of isochromosome 7q, was particularly important in confirming the diagnosis of HSTCL as the patient was previously erroneously diagnosed and treated for T-ALL (Table 3).

Two patients with Langerhans cell histiocytosis and one patient with extensive cutaneous and mucosal involvement by indeterminate cell histiocytosis had tumor samples sequenced (Table 1; Figure 2). Tumor-normal WES was done in one sample and targeted panel sequencing was performed in the remaining two samples. Neither variants commonly identified in histiocytic disorders (53), nor variants of significance in cancer related genes were identified in these samples. Three patients presented with hemophagocytic lymphohistiocytosis (HLH) and underwent constitutional WES to investigate an underlying inherited immunodeficiency disorder or familial HLH (54) (Table 1; Figure 2). One patient with several congenital abnormalities was referred to our institution for management of HLH and stem cell transplantation. However, he was later found to have a compound germline mutation in MLL2 leading to the diagnosis of Kabuki syndrome (55) and stem cell transplant was averted. The remaining two children with HLH were found to carry a germline mutation in C1QA and XIAP, respectively and stem cell transplantation was recommended for both children (Table 4).

Sequencing of germline tissue was performed in 50/56 patients. In six patients, only constitutional WES was performed, because of a strong suspicion of a cancer predisposing condition (n = 3) or an underlying condition predisposing to the development of hemophagocytic lymphohistiocytosis as described above (n = 3) (Table 1). Constitutional WES was requested in 2/6 patients based on clinical observations and in the absence of any family history suggestive of possible predisposition to cancer. One patient developed persistent thrombocytopenia following completion of planned chemotherapy for AML and was referred for stem cell transplant; mild thrombocytopenia was noted during the screening work-up of her donor sister leading to the suspicion of an underlying familial platelet disorder; constitutional WES identified a germline mutation in RUNX1 leading to a change in donor selection and referral of the affected family members to the clinical genetic service. The second patient had persistent anemia and thrombocytopenia, but no definitive diagnosis was obtained from the work-up, which included analysis of the bone marrow aspirate and biopsy. Constitutional WES identified a de novo germline frameshift mutation in GATA2 (56), thus explaining the clinical findings of peripheral blood cytopenias and leading to a recommendation for stem cell transplant. In the remaining 44 patients, germline tissue was sequenced through WES/RNA-seq. Germline findings were identified in 7/44 patients (16%) and included variants in three genes recommended for return by ACMG (Figure 3; Table 4). The three patients and their families were referred to a clinical geneticist for the return of results. Additionally, for one of these three patients, a 1-year-old girl with Down syndrome related AML refractory to chemotherapy, the finding of germline TP53 mutation contributed to the early decision to proceed to palliative care rather than intensive therapy with curative intent. Importantly, the family history for these three patients was not suggestive of an inherited cancer syndrome and a referral to clinical genetics would have not been requested. Overall, clinically impactful germline variants were found in 12/50 patients (24%) (Table 4).