Targetable kinase gene fusions in high-risk B-ALL: a study from the Children’s Oncology Group

B-lineage acute lymphoblastic leukemia (B-ALL) is characterized by recurring sentinel chromosome abnormalities that are important markers for prognosis and treatment stratification and frequently deregulate oncogenes or encode proteins that play a critical role in leukemogenesis, some of which are viable therapeutic targets.^1-3  Although long-term survival in childhood ALL now exceeds 85%, several genetically defined subgroups continue to have poor survival. One high-risk subtype is Philadelphia chromosome-like (Ph-like) or BCR-ABL1–like ALL, defined by the presence of a gene expression profile (GEP) similar to that of Ph⁺ ALL but lacking the canonical BCR-ABL1 fusion.^4-8  Ph-like ALL accounts for approximately 15% of childhood B-ALLs, with the frequency increasing to 20% to 25% in adolescents and young adults, and is associated with a higher risk of relapse and death.^6,9-13  A majority of patients with Ph-like ALL harbor somatic genetic alterations that activate kinase and cytokine receptor signaling.^6,9,11,14-16 

Previous work by our group described the genomic landscape of Ph-like ALL in children, adolescents, and young adults.^6,14,15,17  Approximately 50% of patients with Ph-like disease overexpress cytokine receptor–like factor 2 (CRLF2), resulting from a rearrangement that fuses the immunoglobulin heavy chain locus (IGH) with CRLF2 or, alternatively, an interstitial deletion within the pseudoautosomal region of chromosome X/Y that fuses CRLF2 to the P2RY8 gene promoter.^18-22  Approximately half of patients with CRLF2 rearrangements harbor activating point mutations of JAK1 or JAK2.^{9,18,20,23,24}  This subset may be amenable to targeted therapy with JAK2 inhibitors, such as ruxolitinib, although response has varied in preclinical studies of CRLF2-rearranged ALL.^25,26  The remaining patients with Ph-like ALL harbor a variety of kinase alterations, including fusions involving ABL-class genes (ABL1, ABL2, CSF1R, and PDGFRB) sensitive to ABL1 tyrosine kinase inhibitors (TKIs) or rearrangements that create JAK2 fusion proteins or truncating rearrangements of the erythropoietin receptor (EPOR) that are sensitive to ruxolitinib in vitro.^{6,11,14,16,27,28}  JAK2 inhibitors have demonstrated efficacy in patients with myelofibrosis harboring JAK2 mutations²⁹ ; however, clinical data regarding their efficacy in patients with ALL harboring JAK-activating alterations are limited.

Anecdotal reports describe substantial clinical responses to TKIs in patients with Ph-like ALL and specific genomic rearrangements, providing a compelling rationale for developing strategies to identify such patients in real time and test TKI therapy in genetically defined subsets.^6,16,28,30  However, a wide variety of genomic alterations are present in patients with Ph-like ALL, making diagnostic molecular characterization complex. Our analysis of 1725 patients with B-ALL identified multiple fusion partners for ABL1 (n = 6), ABL2 (n = 3), PDGFRB (n = 4), and JAK2 (n = 10), some with different genomic breakpoints producing distinct fusion transcripts,⁶  as well as 4 different genomic rearrangements that produce an activated, truncated EPOR.¹⁴  Other studies have demonstrated a wide variety of ABL-class and JAK2 fusions.^31-33  Herein we describe the development of an algorithm for genomic testing to identify Ph-like ALL and targetable kinase and cytokine receptor alterations in B-ALL suitable for treatment stratification. This algorithm was applied to an unselected cohort of 1389 consecutive patients with high-risk B-ALL enrolled in Children’s Oncology Group (COG) clinical trials.

We retrospectively studied 1389 consecutively enrolled, newly diagnosed patients with B-ALL between 2010 and 2014. Samples were selected based on availability of cryopreserved leukemia blasts from patients meeting eligibility criteria for COG clinical trial AALL1131 (NCT01406756). The criteria for enrollment in AALL1131 included National Cancer Institute (NCI) high-risk status at diagnosis (age ≥10 years and/or initial white blood cell [WBC] count ≥50 000/µL; 884 [64%] of 1339 patients). The remaining 36% of patients (505 of 1389) analyzed had NCI standard-risk ALL (age <10 years and WBC <50 000/µL) and were originally enrolled in COG AALL1131 at the time of diagnosis (n = 37) because of central nervous system or testicular leukemia or prior corticosteroid treatment. Alternatively, patients were enrolled in the COG standard-risk ALL trial AALL0932 (NCT01190930) at the time of diagnosis (n = 468) and became eligible for COG AALL1131 at the end of induction because of minimal residual disease (MRD) levels ≥0.01% (Figure 1).^34,35  Because we were interested in identifying patients with new targetable kinase fusion genes amenable to precision-medicine therapies, NCI standard-risk patients with known BCR-ABL1 (targetable fusion already known) or ETV6-RUNX1 fusions (known not to have targetable kinase fusion genes) were not included in the samples selected for analysis. Patients with Down syndrome were also excluded from these analyses. Patients and/or their parent(s)/guardian(s) provided informed consent for clinical trial participation, banking, and future research. All analyses were approved by the institutional review boards of Nationwide Children’s Hospital, the University of New Mexico, and St. Jude Children’s Research Hospital.

All patient cases were analyzed using a TaqMan LDA polymerase chain reaction (PCR) assay on a 384-well microfluidic card (supplemental Figure 1, available on the Blood Web site). The GEP used to define Ph-like ALL by prediction analysis for microarray³⁶  was used to develop an 8-gene quantitative assay optimized to identify patients with Ph-like disease with targetable kinase gene alterations.^11,37  An integrated score between 0 and 1 was generated from the 8-gene assay, with a predictive score ≥0.5 considered Ph-like. Assays for expression of the 8 genes involved in the LDA predictor (JCHAIN, SPATS2L, CA6, NRXN3, MUC4, CRLF2, ADGRF1, BMPR1B) were performed in duplicate along with individual reverse transcriptase PCR (RT-PCR) assays for P2RY8-CRLF2 (2 splice forms), BCR-ABL1 (b2a2/b3a2 and e1/a2 isoforms), and ETV6-RUNX1 (4 isoforms). Both BCR-ABL1 and ETV6-RUNX1 fusions were included on the LDA card as validation for use in future prospective screening algorithms. Expression of predictor and fusion genes was internally normalized to an endogenous control gene (EEF2), and levels were expressed in units of ΔC_t (predictor/fusion C_t – EEF2 C_t). Patient cases of Ph-like disease that were BCR-ABL1⁺ or ETV6-RUNX1⁺ were excluded from further analysis because either a targetable kinase lesion had already been identified (BCR-ABL1) or a targetable kinase fusion had not been identified in comprehensive genome/RNA-sequencing analyses of ETV6-RUNX1 ALL.^38-40 

Any patient case determined to have elevated CRLF2 expression (CRLF2_high; defined as ΔC_t ≤6 by LDA) was assessed for P2RY8-CRLF2 by Taqman PCR on the LDA card or for IGH-CRLF2 by fluorescence in situ hybridization (FISH).⁹  Those with CRLF2_high were also assessed for the presence of a concomitant JAK1 or JAK2 mutation by Sanger sequencing.⁹  All identified coding variants were confirmed to be somatic by comparison with matched remission DNA.

The remaining Ph-like CRLF2_low patient cases were subject to multiplex and singleplex RT-PCR assays for previously identified kinase fusions; 2 µg of total RNA isolated using TRIzol (Life Technologies, Carlsbad, CA) was used to prepare complementary DNA (cDNA) with a High-Capacity cDNA Reverse Transcription Kit (Life Technologies), and 2.5 µg of cDNA (equivalent to 50 ng of starting RNA) was used to test samples for previously identified kinase fusions involving ABL1, ABL2, CSF1R, JAK2, NTRK3, and PDGFRB by RT-PCR (supplemental Table 1).^6,15 

IL7R mutations were analyzed in 173 Ph-like samples as previously described.⁴¹  These included patient cases that were LDA⁺ but were either negative by other assays (eg, JAK mutational analysis, multiplex PCR fusion testing) or not analyzed using RNA sequencing methods.

CRLF2_low patient cases with no kinase gene fusion identified by multiplex or singleplex PCR and CRLF2_high patient cases without CRLF2 rearrangement underwent RNA kinome capture and total-stranded whole-transcriptome sequencing (RNAseq), performed in parallel. All identified fusions were confirmed via bidirectional Sanger sequencing. RNA kinome capture and sequencing were performed using an Agilent SureSelect RNA Kinome Capture Kit (Agilent Technologies, Inc., Santa Clara, CA) including 612 genes (571 kinases). Library preparation was performed using capture baits and RNA-sequencing library preparation kits (Agilent Technologies or Illumina; New England Biolabs, Inc., Ipswitch, MA) and sequenced using the Illumina HiSeq 2000. Fusion screening was accomplished by Geospiza GeneSifter (Geospiza, Inc. [PerkinElmer], Seattle, WA) with ChimeraScan.^42,43  RNAseq was performed using the TruSeq library preparation on the Illumina HiSeq 2000 platform. To detect kinase fusions, samples were analyzed using CICERO⁶  and FusionCatcher.⁴⁴  Sequence mutations were analyzed using the GATK pipeline.⁴⁵  Genomic data have been deposited at the European Genome Phenome archive accession EGAS00001001952. Sequencing metrics (supplemental Table 2) and processing details are provided in the supplemental data.

Of 1389 patient cases analyzed, 341 (24.6%) had a Ph-like GEP by LDA. Those with BCR-ABL1 (n = 46; all NCI high risk) or ETV6-RUNX1 (n = 11; all NCI high risk) were excluded from further analysis. The remaining 284 patient cases (20.4%) of Ph-like ALL, including 198 (22.4%) of 884 with NCI high-risk ALL and 86 (17.0%) of 505 with NCI standard-risk ALL, that met eligibility criteria for AALL1131 underwent testing according to the algorithm (Figure 1; supplemental Table 3). The sex distribution of patients with Ph-like ALL (57.4% male) was similar to that of patients with non–Ph-like disease (55.0% male; P = .473). Patients with Ph-like disease comprised a higher proportion of those ≥10 years of age (25.8%) as compared with those age <10 years (16.7%; P < .0001) and had a significantly higher diagnostic WBC count using a cutoff of 50 × 10⁹/L (40.5% vs 30.4%; P < .001), 100 × 10⁹/L (24.3% vs 13.8%; P < .001), or 200 × 10⁹/L (8.5% vs 5.4%; P = .049).

Among these 284 patient cases of Ph-like ALL, 155 (54.6%) were CRLF2_high, with P2RY8-CRLF2 in 63 (22.2% of those with Ph-like disease) and IGH-CRLF2 rearrangement in 61 (21.5%). Among the 61 patient cases with confirmed IGH-CRLF2 fusions, 60 (98.4%) had a CRLF2 ΔC_t ≤4, whereas 47 (74.6%) of 63 patient cases with P2RY8-CRLF2 had CRLF2 ΔC_t ≤4 (P = .0001). Overall, rearrangement of CRLF2 was identified in 124 (80%) of 155 LDA⁺CRLF2_high patient cases, with an overall frequency of 43.7% (124 of 284) in patient cases of Ph-like ALL. Of the 124 patient cases of CRLF2-rearranged disease, 63 (50.8%) harbored concomitant activating mutations in either JAK1 (n = 10), JAK2 (n = 46), or IL7R (n = 8). Notably, 1 patient harbored activating alterations in both JAK1 and IL7R (Figure 2).

The 160 patients with Ph-like disease without CRLF2 rearrangement, including 129 with CRLF2_low disease and 31 with CRLF2_high disease lacking CRLF2 rearrangement, were assessed for additional kinase alterations, initially by RT-PCR for known fusions and, if negative, by kinome capture and RNA sequencing (supplemental Table 1).⁶  Of these, 35 had kinase fusions involving ABL-class genes, JAK2, or EPOR identified by RT-PCR. Identification of other kinase fusions in RT-PCR-negative patients was achieved primarily with RNA sequencing. Of the 30 patients with a fusion involving ABL-class genes, JAK2, or EPOR identified by RNA sequencing, kinome capture was performed in parallel in 27, with the kinase fusion identified in 16 of the 27, whereas 11 were negative by kinome capture. Those not identified by kinome capture included 2 new fusions (known kinases with new 5′ partners) and 9 patients with IGH-EPOR rearrangement. Two patients with IGH-EPOR identified by RNA sequencing were not tested using kinome capture. Similarly, JAK-STAT–activating mutations were only identified by RNA sequencing. Two patient cases with a kinase fusion identified by kinome capture did not undergo further RNA sequencing (NUP153-ABL1 and TNIP1-PDGFRB). Together, 113 of 156 patient cases with Ph-like ALL without CRLF2 rearrangement (72.4%) with evaluable material harbored a genomic alteration predicted to activate kinase or cytokine receptor signaling (Figure 3), including 40 with ABL-class fusions (ABL1, n = 17; ABL2, n = 4; CSF1R, n = 4; and PDGFRB, n = 15), 14 with JAK2 fusions, 11 with EPOR rearrangement, 1 each with an NTRK3 and LYN fusion, and 11 with an activating FLT3 mutation (10 internal tandem duplications similar to those identified in acute myeloid leukemia and 1 missense V579G mutation). We also observed sequence mutations and copy-number alterations in genes activating JAK-STAT signaling in 20 patients, including IL7R (n = 14), SH2B3 (n = 7), and JAK1 (n = 2). Notably, 3 patients harbored a concomitant IL7R mutation and SH2B3 deletion, and 2 patients harbored an FLT3 internal tandem duplication mutation with SH2B3 deletion. Nineteen patients harbored missense mutations in Ras pathway genes, including KRAS (n = 11), NRAS (n = 6), NF1 (n = 1), and PTPN11 (n = 1). One of the NRAS mutations was identified in a patient with ZMIZ1-ABL1 and another in a patient with an SH2B3 missense mutation. We also identified 1 P2RY8-CRLF2 fusion by LDA in a patient with CRLF2_low gene expression (ΔCt, 7.4). Forty-two patients cases with Ph-like disease (14.8%; including 12 [7.7%] of 155 CRLF2_high and 30 [23.3%] of 129 CRLF2_low; P < .001) lacked a kinase alteration by RNAseq analysis, 39 of which had an LDA value ≤0.65. Four patient cases lacked material for complete evaluation (supplemental Figure 2).

All fusion-positive patient cases were confirmed to have in-frame fusion of the 5′ partner to the 3′ kinase gene and retention of the tyrosine kinase domain of the 3′ fusion partner. The 17 ABL1 fusions identified included 5 known 5′ partners (ETV6, NUP214, RANBP2, RCSD1, and ZMIZ1, with alternate breakpoints identified in ETV6 and NUP214) and 3 novel 5′ partners: CENPC1, NUP153, and LSM14A; the 4 ABL2 fusions involved previously known 5′ partners (RCSD1 and ZC3HAV1). The 4 CSF1R fusions involved 1 known 5′ partner (SSBP2) and 1 novel partner (TBL1XR1); the 15 PDGFRB fusions included 1 previously known 5′ partner (EBF1) and 3 novel partners (ETV6, TNIP1, and ZMYND8). The 14 JAK2 fusions involved 8 different 5′ partners, 5 of which were known (ETV6, SSBP2, BCR, PAX5, and PCM1) and 3 of which were novel (RFX3, USP25, and ZNF274), with alternate breakpoints identified in SSBP2, BCR, and PAX5. We detected the previously reported ETV6-NTRK3 fusion in 1 patient and a GATAD2A-LYN fusion in another (Table 1; Figures 3 and 4). Although a majority of the fusions involved interchromosomal translocations, several fusions could have arisen from interstitial deletions within 1q24.2-q25.2 (RCSD1-ABL2), 5q14.1-q23.2 (SSBP2-CSF1R), 5q32 (TNIP1-PDGFRB), 5q32-q33.3 (EBF1-PDGFRB), 9p24.1-p24.2 (RFX3-JAK2), 9p13.2-p24.1 (PAX5-JAK2), or 9q34.12-q34.13 (NUP214-ABL1).

In total, 238 (83.8%) of 284 patients with Ph-like ALL harbored a genetic alteration activating kinase or cytokine receptor signaling. Those with higher LDA values were more likely to have targetable kinase alterations and particularly ABL-class fusions; 98% of patients (50 of 51) with LDA values ≥0.8 harbored a CRLF2 rearrangement (n = 28) or other kinase fusion (n = 22), whereas 30.2% (26 of 86) with LDA values between 0.5 and 0.59 harbored a targetable fusion involving CRLF2 (n = 21) or another kinase (n = 5; P < .0001; supplemental Figure 3). We observed differences in the spectrum and distribution of kinase alterations in the standard- vs high-risk cohorts (supplemental Figure 4). Compared with the standard-risk cohort eligible for AALL1131, patients with Ph-like ALL meeting NCI high-risk criteria had a higher prevalence of CRLF2 rearrangement (48.5% vs 33.7%; P < .02) and ABL-class fusions (17.7% vs 5.8%; P = .0087), with fewer Ras pathway alterations (3.0% vs 12.8%; P = .0044) and instances of no kinase alteration being identified (9.1% vs 27.9%; P < .0001; Table 2).

Among the remaining 41 patient cases with available material for RNAseq, 12 harbored fusions in additional genes not involved in kinase or cytokine receptor signaling, including MEF2D (n = 4) and ETV6 (n = 2; supplemental Table 3).

To determine whether the screening algorithm in this study was sensitive for detection of potentially targetable kinase fusions, we analyzed patient cases with LDA values below the threshold of 0.5. Of the 1048 LDA-negative patient cases, 6 had P2RY8-CRLF2 fusion detected by LDA and 3 harbored IGH-CRLF2 detected by FISH. The LDA values for these patient cases with CRLF2 rearrangement ranged from 0.419 to 0.488, and all were CRLF2_high with CRLF2 ΔC_t values ranging from 2.2 to 5.8. A total of 194 additional patient cases underwent the multiplex/singleplex PCR panel (3 with LDA values between 0.25 and 0.3 [all standard risk]; 132 with LDA values between 0.40 and 0.449 [standard risk, n = 68; high risk, n = 64]; and 59 with LDA values between 0.45 and 0.499 [standard risk, n = 31; high risk, n = 28]), and 63 patient cases underwent RNA sequencing (31 with LDA values between 0.40 and 0.449 and 32 with LDA values between 0.45 and 0.499). The 194 patients analyzed had characteristics similar to the 854 patients not tested, with respect to sex, Hispanic ethnicity, age, and initial WBC count (P = .299-1.0) but were more likely to be NCI standard risk (P < .001). No potentially targetable kinase fusions involving ABL1, ABL2, CSF1R, PDGFRB, JAK2, NTRK3, or EPOR rearrangement were identified. We did identify 1 potentially targetable fusion involving the fibroblast growth factor receptor 1 gene, HOOK3-FGFR1, with an LDA value of 0.47. Other fusions identified in this cohort included TCF3-PBX1 (n = 4) and MEF2D (n = 3) and 2 rearrangements each involving ETV6, PAX5, IGH, and NUTM1 (supplemental Table 4).

Patients with Ph-like ALL have significantly inferior event-free and overall survival, emphasizing the clinical importance of identifying those who could potentially benefit from targeted therapy.^4-6,10-12,15  Here we determined the frequency of kinase-activating lesions in 1389 patients with B-ALL enrolled in or eligible for COG protocol AALL1131 for high-risk ALL identified as Ph-like by LDA (n = 284; 20.4%). Of note, we were not able to assess outcome in these patients, because these data were not yet mature and were not available for analysis. It will be interesting to determine whether MRD-based risk-directed therapy can salvage the poor outcome for these patients, as has previously been reported for patients in St. Jude clinical trials, although a number of patients with Ph-like ALL in those trials underwent stem-cell transplantation because of elevated MRD.⁴⁶ 

We observed rearrangement of CRLF2 in 43.7% of patient cases of Ph-like ALL, with similar prevalence of P2RY8-CRLF2 and IGH-CRLF2. This differs from CRLF2 fusion partner ratios of 2.1:1 and 5:1^18,19  for P2RY8:IGH reported in other studies^18,19,47  and was likely a result of the different ethnic composition and NCI risk stratification of our study cohort and the comprehensive FISH analyses performed. Of the 61 IGH-CRLF2 fusions identified, 56 (91.8%) occurred in NCI high-risk patients and 5 in standard-risk patients. The frequency of self-declared Hispanic/Latino ethnicity was 27.6% among the 1389 patients analyzed, 37.0% (106 of 284) among patients with Ph-like ALL (P = .0002 for Ph-like vs overall), 30.2% (19 of 63) among patients with P2RY8-CRLF2, and 59% (36 of 61) among patients with IGH-CRLF2 (P = .0012 for P2RY8- vs IGH-CRLF2). Similar to prior reports, point mutations of JAK2 or JAK1 were present in approximately half (45.2%) of those with CRLF2-rearranged disease.^6,9,15,20,23  Coexpression of genomic lesions causing CRLF2 overexpression and JAK1/2 mutations is transforming in vitro and promotes lymphoid transformation.^18,48  Regardless of JAK status, cells with CRLF2 deregulation are frequently sensitive to JAK and mTOR inhibitors, although variations in response have been observed.^25,49 

An additional 22.8% of patients with Ph-like ALL harbored targetable kinase fusions involving ABL-class genes, JAK2, or EPOR (45.0% of LDA⁺CRLF2_low patient cases and 5.8% of LDA⁺CRLF2_high patient cases). Functionally, overexpression of fusion proteins encoded by ABL1, ABL2, CSF1R, or PDGFRB rearrangement phenocopies BCR-ABL1 in vitro, with similar transforming capabilities and analogous sensitivity to the ABL inhibitors imatinib and dasatinib.^6,15  These data suggest that patients harboring ABL-class fusions are candidates for testing ABL inhibitors in combination with chemotherapy, an approach that has revolutionized treatment of pediatric Ph⁺ ALL.^50-52  In addition, the novel GATAD2A-LYN fusion is predicted to activate SRC signaling and also respond to dasatinib^31,53 ; functional studies are ongoing.

Other targetable genomic lesions in patients with Ph-like ALL lacking CRLF2 rearrangement include JAK2 fusions and EPOR rearrangement, both of which are sensitive to ruxolitinib in vitro and in vivo, suggesting that small-molecule JAK inhibitors merit testing in this Ph-like ALL subset.^14,25,49  Other subgroups of Ph-like ALL include patients who harbor activating sequence mutations or copy-number alterations activating JAK-STAT signaling, including IL7R, SH2B3, and JAK1 (collectively 6.3%). Preclinical studies suggest the efficacy of JAK inhibitors should also be assessed in patients harboring these JAK-STAT pathway alterations.²⁵  Of note, 3.9% of patients with Ph-like disease in this study harbored FLT3-activating alterations, with strong preclinical data suggesting US Food and Drug Administration–approved FLT3 inhibitors are logical agents to test in this subgroup.⁵⁴  The remaining ∼15% of patients with Ph-like disease had other genomic lesions that do not seem to target cytokine or receptor signaling, with no obvious therapeutic implications.

The population of patients in this study was based on eligibility criteria for the COG AALL1131 clinical trial, which included those with NCI high-risk ALL and those with NCI standard-risk disease with central nervous system or testicular leukemia, steroid pretreatment, or MRD ≥0.01% at end of induction. This study was not designed to assess the frequency of Ph-like ALL and kinase alterations in an unselected cohort of patients with standard-risk ALL. Despite this, we still observed significant differences in the genetic characteristics of patients with Ph-like ALL based on initial NCI risk stratification, even though 79.5% (66 of 83) of those with Ph-like ALL with NCI standard-risk disease were MRD⁺ vs only 56.5% (108 of 191) of patients with high-risk Ph-like ALL. The incidence of Ph-like ALL was modestly lower in the tested NCI standard-risk patients compared with the NCI high-risk patients (17.0% vs 22.4%; P = .019). This is likely because patients with Ph-like ALL are older and tend to have higher WBC count at diagnosis and are therefore more likely to present with NCI high-risk features. In total, 10.9% of NCI high-risk patients had Ph-like disease with CRLF2 rearrangement compared with 5.7% of tested standard-risk patients (P = .0012; Table 2). There were similar overall percentages of patients with P2RY8-CRLF2 (4.5% of high-risk and 4.8% of tested standard-risk patients); however, poor-prognosis IGH-CRLF2 status was much more common in high-risk than tested standard-risk patients (6.3% vs 1.0%; P < .001). We also observed an increased prevalence of other targetable kinase fusions in the high-risk vs tested standard-risk cohort (6.3% vs 1.0%). The frequency of JAK2 fusions was similar between the NCI high-risk and tested standard-risk patients, with increased frequency of ABL-class fusions (4.0% vs 1.0%; P = .0012) and EPOR rearrangement (1.2% vs 0%; P = .01) observed in NCI high-risk compared with tested NCI standard-risk patients with ALL. Compared with our previous report on unselected patients with NCI standard-risk ALL, where Ph-like status was defined by prediction analysis for microarray analysis of microarray data, the frequency of Ph-like ALL was increased in our current selected standard-risk cohort (10% vs 17%); however, the frequency of targetable kinase fusions was similar in both standard-risk cohorts (0.6% vs 1.0%).⁶  These data suggest that targetable kinase fusions are less common in patients classified as having standard-risk ALL based on initial NCI criteria of age and WBC count, even among those with a poor initial response to therapy, and likely help to explain the lower frequencies of these lesions reported by some groups that have screened cohorts of unselected patients with ALL.⁴⁶ 

Given the ongoing discovery of new fusions and new splice variants of known fusions, screening with RT-PCR assays for known fusions will not identify fusions with novel breakpoints or 5′ fusion partners. The most comprehensive approach would be to perform unbiased RNA sequencing on all patient cases of ALL or on the subset of patient cases defined as having a Ph-like GEP signature. This approach is currently challenging, particularly for clinical trials involving large numbers of patients from multiple sites (>1800 patients with B-ALL are enrolled in COG ALL trials by >200 centers per year), but will likely become feasible as sequencing costs continue to decline and the speed of bioinformatic analysis improves. We explored a kinome capture RNA-sequencing approach in the current study but found that this assay lacks significant cost and efficiency advantages over full RNA sequencing. Other technologies that might be used include the assay based on Nanostring (Seattle, WA) and a commercially available anchored multiplex PCR assay (ArcherDx, Inc., Boulder, CO) based on targeted next-generation sequencing with the potential ability to detect any fusion associated with a 3′ kinase gene partner.

These data show that 20.4% of children, adolescents, and young adults with B-ALL and high-risk clinical features as defined in this study have Ph-like ALL defined by the LDA ≥0.5 threshold. Among these patients, 83.8% had cytokine receptor– or kinase-activating alterations, confirming and extending our previous observation that, although Ph-like ALL has a complex range of genomic lesions, the underlying alterations seem to be associated with a limited number of pathways, including ABL-class and JAK-STAT signaling, that are sensitive in vitro to US Food and Drug Administration–approved TKIs and JAK small-molecule inhibitors.^6,28  Although somewhat complex, the LDA screening and downstream testing approach we have developed and described in this study is feasible for screening patients in the context of treatment stratification. Our data demonstrate that the LDA threshold of 0.5 is robust, with only 1 of 194 patient cases with an LDA value <0.5 tested having a kinase fusion (HOOK3-FGFR1). Similar to our prior reports,^6,9,18,47  a small subset of patients with CRLF2 rearrangement (9 [6.8%] of 133) did not have a Ph-like GEP but did exhibit high CRLF2 expression. On the basis of these results, we implemented the LDA assay and downstream testing algorithm in COG AALL1131 in 2016. Patients with NCI high-risk Ph-like ALL and ABL-class fusions are eligible to have dasatinib added to their backbone chemotherapy regimen. Those patients with Ph-like ALL and JAK-STAT pathway lesions, including CRLF2 rearrangement with or without JAK mutations, JAK2 fusions, EPOR rearrangement, or IL7R rearrangement, are eligible to enroll in COG AALL1521 (NCT02723994), in which ruxolitinib is added to combination chemotherapy.

The authors wish to thank Erik Zmuda at Nationwide Children’s Hospital for technical assistance and Joshua Stokes from Biomedical Communications at St. Jude Children’s Research Hospital for the pie charts.

This work was supported by grants from St. Baldrick’s Foundation (S.P.H., C.G.M., and M.L.L.) the Leukemia Lymphoma Society (W.L.C., S.P.H., and C.G.M.), the National Institutes of Health, National Cancer Institute (U10 CA 180886, U10 CA 180899, and P30 CA021765), and the National Institutes of Health, National Institute of General Medical Sciences (P50 GM115279). M.L.L. is the Benioff Distinguished Chair of Children’s Health. S.P.H. is the Jeffrey E. Perelman Distinguished Chair in Pediatrics at the Children’s Hospital of Philadelphia. K.G.R. is an American Society of Hematology Scholar.

The opinions and assertions contained herein are the private views of the authors and are not to be construed as the official policy or position of the US Government, the Department of Defense, the Department of the Air Force, or the US Food and Drug Administration.

Contribution: S.C.R., R.C.H., K.G.R., and S.P.H. prepared the manuscript; E.S., A.S., H.J., I.-M.C., M.V., K.G.R., Y.S., D.P.-T., and Z.G. performed experiments; S.C.R., R.C.H., Y. Liu, Y. Li, J.Z., K.G.R., J.E., M.D., N.A.H., A.J.C., M. J. Borowitz, B.L.W., C.G.M., C.L.W., and S.P.H. analyzed data; E.A.R., A.L.A., M. J. Burke, W.L.S., P.A.Z.-M., K.R.R., W.L.C., M.L.L. and S.P.H. provided patient samples; and all authors edited and approved the manuscript.

Conflict-of-interest disclosure: The Ph-like gene expression classifier used in this work is covered in part by issued US Patent #8568974 (inventors: C.L.W., R.C.H., and I.-M.C.; assignees: STC.UNM and Sandia Corporation) and pending US application #20140322166 (inventors: C.L.W., S.P.H., C.G.M., I.-M.C., K.G.R., and R.C.H.; applicants: STC.UNM, St. Jude Children’s Research Hospital, and the Children’s Hospital of Philadelphia on behalf of the Children’s Oncology Group).

Correspondence: Shalini C. Reshmi, Institute for Genomic Medicine, Department of Pathology, Room C0839, 700 Children’s Dr, Nationwide Children’s Hospital, Columbus, OH 43205; e-mail: shalini.reshmi@nationwidechildrens.org; and Richard C. Harvey, 2325 Camino de Salud NE, CRF-107 (MSC 08-4640), Albuquerque, NM 87131; e-mail: rharvey@salud.unm.edu.