Current developments in molecular monitoring in chronic myeloid leukemia

Abstract

Molecular monitoring plays an essential role in the clinical management of chronic myeloid leukemia (CML) patients, and now guides clinical decision making. Quantitative reverse-transcriptase-polymerase-chain-reaction (qRT-PCR) assessment of BCR-ABL1 transcript levels has become the standard of care protocol in CML. However, further developments are required to assess leukemic burden more efficiently, monitor minimal residual disease (MRD), detect mutations that drive resistance to tyrosine kinase inhibitor (TKI) therapy and identify predictors of response to TKI therapy. Cartridge-based BCR-ABL1 quantitation, digital PCR and next generation sequencing are examples of technologies which are currently being explored, evaluated and translated into the clinic. Here we review the emerging molecular methods/technologies currently being developed to advance molecular monitoring in CML.

Introduction

CML is caused by the formation of the Philadelphia (Ph) chromosome, a reciprocal translocation between chromosomes 9 and 22, resulting in the formation of the BCR-ABL1 fusion gene. This fusion gene encodes an activated tyrosine kinase, which dysregulates cell proliferation, differentiation and apoptosis. Historically, CML was treated with chemotherapeutic agents such as hydroxyurea, busulfan and interferon. In 2001, TKIs were approved by the Food and Drug Administration as therapy for CML, and dramatically improved the 8-year survival rates of CML patients to ~85% [Kantarjian et al. 2012; Kalmanti et al. 2015]. TKIs are such an effective therapy for CML due to their specific inhibitory activity against BCR-ABL1. Ph positive (Ph+) cells and BCR-ABL1 transcripts can be monitored using cytogenetic and molecular assays. Certainly, demonstration of milestone molecular responses during TKI therapy is associated with clinical response, including overall survival (OS), progression-free survival (PFS) and event-free survival (EFS) [Hanfstein et al. 2012; Marin et al. 2012a, 2012b; Etienne et al. 2014; Hehlmann et al. 2014]. Due to the strong association between molecular-response (MR) milestones and long-term treatment outcomes, molecular results now guide clinical decision making, and play an essential role in the clinical management of CML patients [Baccarani et al. 2013; NCCN, 2016].

Quantitation of BCR-ABL1 transcript levels using qRT-PCR has become the standard-of-care protocol in CML molecular monitoring [Baccarani et al. 2013; NCCN, 2016]. A ratio of BCR-ABL1 transcripts is calculated by measuring unknown samples against a standard curve derived from serial dilutions of accurately quantified standards and normalized to the expression of one of three recommended reference genes; ABL1, BCR and GUSB [Hughes et al. 2006]. Variations in reagents, controls genes, probe/primer sets, and PCR machines can lead to significant differences in BCR-ABL1 levels reported by various laboratories [Branford et al. 2008]. Moreover, there is a limited availability of reference standards, which has hindered comparability between BCR-ABL1 levels quantified across laboratories. Therefore, to aid comparisons of BCR-ABL1 results internationally, standardization of BCR-ABL1 transcript levels was undertaken. Exchange of patient samples with an international reporting scale (IS) calibrated laboratories enabled calculation of laboratory-specific conversion factors for in-house qRT-PCR protocols and alignment of BCR-ABL1 values to the IS. This initiative has led to the implementation of reporting of BCR-ABL1 transcripts on the IS [Hughes et al. 2006; Branford et al. 2008]. More recently, the first World Health Organization International Genetic Reference Panel for quantitation of BCR-ABL1 by qRT-PCR was developed [White et al. 2010]. These standards are now used for the production and dissemination of secondary standards for improved interlaboratory calibration.

Current recommendations for CML molecular monitoring are based upon quantification of BCR-ABL1, as assessed by qRT-PCR with or without cytogenetics. Molecular assessments are made at diagnosis and at 3, 6 and 12 months after beginning TKI therapy. BCR-ABL1 transcript levels at these milestone timepoints define MR. The European LeukemiaNet 2013 recommendations define an optimal response as BCR-ABL1 transcript levels measured on the IS as <10% at 3 months, <1% at 6 months, and <0.1% from 12 months onward [Baccarani et al. 2013]. Patients with BCR-ABL1 transcript levels >10% at 6 months and >1% from 12 months onward are classed as failed therapy. BCR-ABL1 levels that fall between the optimal and failed levels indicate a warning zone where more frequent molecular monitoring is recommended [Baccarani et al. 2013]. The NCCN Clinical Practice Guidelines define therapy failure as >10% BCR-ABL1 at 3 months or 6 months [NCCN, 2016]. Both guidelines recommend a change in treatment to an alternate TKI when a patient fails therapy. Achievement of a deep molecular response [(DMR) MR⁴ or MR^4.5] is an emerging goal in CML, as it is a prerequisite for entry into treatment discontinuation studies [Mahon et al. 2010; Ross et al. 2014]. Undetectable BCR-ABL1 from qRT-PCR assays with inadequate sensitivity may lead to inappropriate or premature cessation attempts. Well defined guidelines have been developed to ensure adequate sensitivity levels are achieved to detect residual disease down to MR⁴ or MR^4.5 [Cross et al. 2015].

Current advances in molecular monitoring are focused towards addressing ongoing clinical challenges in CML. These challenges include (i) the development of efficient methods which facilitate fast, inexpensive and sensitive BCR-ABL1 quantitation in emerging economic regions; (ii) quantitation of low levels of BCR-ABL1 to assist detection and monitoring of MRD; (iii) sensitive and accurate detection of mutations that drive resistance to TKI therapy; and (iv) the discovery of molecular markers that predict response to therapy or disease progression. The emerging molecular methods and technologies currently being explored to advance molecular monitoring in CML will be reviewed.

Advances in BCR-ABL1 quantification

GeneXpert cartridge-based BCR-ABL1 quantification

Advances in BCR-ABL1 quantification, whereby inexpensive and sensitive results on the IS can be efficiently obtained from various laboratory settings will be advantageous to CML molecular monitoring in emerging economic regions. A cartridge-based automated qRT-PCR system, the Xpert BCR-ABL1 Monitor Assay, which measures BCR-ABL1 p210 transcripts, has been developed by Cepheid (California, USA). The GeneXpert system utilizes microfluidics to directly process peripheral blood samples in a cartridge for RNA extraction, RT-PCR and fluorescence detection in an ‘all in one’ self-contained instrument [Dufresne et al. 2007]. A major advantage of this type of system is a simplified workflow. Therefore, this system has potential to provide faster turnaround times, reduce hands-on time, and requires less technical expertise compared with the current diagnostic BCR-ABL1 qRT-PCR assay. In contrast to most current BCR-ABL1 qRT-PCR assays that determine absolute BCR-ABL1 copy number, the GeneXpert system calculates the BCR-ABL1/ABL1 ratio using the delta-crossing threshold (Ct) comparative-quantitation method. Early comparisons of the GeneXpert assay with a standard qRT-PCR assay indicated the GeneXpert Ct values and assay efficiencies correlated highly with standard qRT-PCR assays, indicating that the system had the potential to translate into a clinical setting, despite differences in quantitative methods [Jobbagy et al. 2007; Winn-Deen et al. 2007]. However, the system did not report BCR-ABL1 values on the IS, which hindered interpretation and interlaboratory comparison of results. Consequently, the system adopted reporting BCR-ABL1 transcript levels on the IS, whereby a conversion factor specific to the GeneXpert assay was developed and used to adjust the BCR-ABL/ABL1 ratio to the IS [Enjeti et al. 2015]. Additionally, a batch-specific standard curve and an efficiency value were also introduced to ensure cartridge-lot-specific calibration [Enjeti et al. 2015]. This quality-control improvement to the assay facilitates convenient inter-run and cross-laboratory comparison. Certainly, accurate measurement down to 0.01% BCR-ABL1 IS, a level sufficient to detect MR⁴, has been consistently achieved [Cayuela et al. 2011; O’Dwyer et al. 2014].

Improvements to the GeneXpert BCR-ABL1 assay has led our laboratory to introduce the Cepheid system as a screening method to detect new cases of p210 BCR-ABL1 CML. Approximately 20% of the samples received in our laboratory are analyzed to exclude CML rather than to confirm a CML diagnosis. The GeneXpert assay is a rapid and reliable procedure, for these cases. At this stage, our standard qRT-PCR method is used for all other samples for a number of reasons: (i) we are an international reference laboratory and must maintain our current method and (ii) we have a strong interest in assessing the kinetics of response, for which our current qRT-PCR protocol has been optimized [Branford et al. 2008, 2012, 2014]. Atypical transcripts, which are less frequently detected in CML, are not targeted in the current GeneXpert BCR-ABL1 assay and will not be detected. Therefore, all negative cases with a strong indication of CML are screened for the presence of atypical transcripts. Interestingly, a novel use of the cartridge system was demonstrated recently where comparable BCR-ABL1 values were achieved using RNA extracted from fresh blood and corresponding dried-blood spots that had been mailed across the globe. High concordance was achieved between the fresh blood and the blood spots (R² = 0.94) [Sala Torra et al. 2016]. The study showed promise for the inexpensive shipping of patient samples from low-resource regions and the generation of reliable results to improve patient monitoring.

The GeneXpert system has not been validated to detect BCR-ABL1 levels below 0.01% BCR-ABL1 (MR⁴); this is primarily due to divergence of correlation from qRT-PCR results below this level, demonstrating the inaccuracy of GeneXpert at very low transcript levels [O’Dwyer et al. 2014; Enjeti et al. 2015]. However, no comparative study has been conducted with a large sample size to accurately determine the concordance between the technologies. Cepheid will soon introduce an enhanced sensitivity assay which can reliably test at a level of MR^4.5 sensitivity on the IS [Day et al. 2015]. Once, this sensitivity level is robustly achieved, the assay could be used to routinely monitor residual disease and aid selection of patients eligible for treatment discontinuation.

Minimal residual disease monitoring

The aim of current developments in MRD monitoring is to enhance the detection of low levels of residual leukemic cells. Molecular methods with increased sensitivity allow for the examination of associations between very DMRs (e.g. down to MR⁶) and treatment-free remission (TFR) success. The Sokal clinical-risk score is the only variable associated with a significant probability of failing TFR, whereby, low Sokal-risk patients were less likely to experience molecular recurrence [Mahon et al. 2010]. Other prognostic factors for successful TKI cessation remain elusive [Mahon et al. 2014; Rousselot et al. 2014]. Modification of conventional qRT-PCR methods and application of digital PCR and deoxyribonucleic acid (DNA) PCR are approaches being employed to improve assay sensitivity and detection of residual leukemic cells.

Modifications to conventional quantitative reverse-transcriptase-polymerase-chain-reaction assays

Modification of conventional qRT-PCR assays to increase sensitivity is an attractive option, as specialized equipment (e.g. a digital PCR platform) or additional expertise would not be required. Alteration of the priming strategy of the reverse-transcription (RT) reaction, whereby random pentadecamer primers are used instead of random hexamers, has been reported [Ross et al. 2008]. This alteration resulted in increased assay sensitivity. However, a mean of 2.3-fold higher BCR-ABL1 copy number was observed with the use of pentadecamer primers compared with random hexamers, indicating that the modification results in overestimation of BCR-ABL1 transcripts [Ross et al. 2008]. Consequently, this modification has not been adopted for routine clinical monitoring in CML. Maximizing the amount of template RNA added to the RT reaction is another strategy proposed to increase sensitivity of the conventional qRT-PCR assay. Concentration of total mRNA has been achieved using oligo-dT-bound magnetic beads to pull down poly-A-tail mRNA transcripts [Yeung et al. 2015]. This modification resulted in approximately sixfold increased sensitivity, and maintained high correlation with conventional qRT-PCR (R² = 0.930). Importantly, the modified protocol also resulted in a higher proportion of samples achieving MR^4.5 sensitivity (91% versus 61%) and increased the sensitivity of the assay to MR⁵ (78% versus 9% ⩾ 5 log) [Yeung et al. 2015]. Examination of the clinical significance of this assay will reveal its clinical utility.

Digital polymerase chain reaction

The conventional BCR-ABL1 qRT-PCR assay utilizes the standard curve absolute quantification method to determine mRNA levels, where samples with unknown BCR-ABL1 levels are quantitated by extrapolating a value from a standard curve of known quantities [Branford et al. 1999]. This quantification method relies on the comparison between amplification cycles and is therefore sensitive to inefficient or variable amplification, which can limit the sensitivity of the method [Freeman et al. 1999]. Digital PCR offers an advantage over conventional PCR because reactions are partitioned into thousands to millions of nanoliter reactions and amplification is undertaken in many spatially separated reactions [Vogelstein and Kinzler, 1999]. Partitions can be generated by emulsion [droplet digital PCR (ddPCR)], as used by the Bio-Rad (California, USA) and RainDance PCR systems (Massachusetts, USA), or by carrying out PCR in nanoscale-reaction chambers, such as Fluidigm’s integrated fluidic circuit chips (California, USA) or Life Technologies’ QuantStudio3D (California, USA). The aim of digital PCR (dPCR) is to have either a single or no template molecules in each partition; as a result, after PCR amplification, each partition can be scored as positive or negative (digital readout). The absolute quantity of molecules can be determined without the need for a standard curve. A major advantage of dPCR is the precise quantification of nucleic acids, facilitating measurement of small percentage differences. dPCR therefore has the potential to detect BCR-ABL1 with greater sensitivity and precision than qRT-PCR [Whale et al. 2012]. dPCR was used to determine the copy-number concentration of a certified plasmid reference material for BCR-ABL1 quantification [White et al. 2015]. Moreover, the requirement for standardization by conversion of the BCR-ABL/ABL1 ratio to IS is reduced.

The application of dPCR for MRD monitoring in CML is a growing area of research. Initial studies using both nanofluidic and ddPCR to quantitate BCR-ABL1 mRNA levels could reliably quantitate down to 0.01% BCR-ABL1 IS, equivalent to MR⁴ [Goh et al. 2011; Jennings et al. 2014]. Although, below this level, false negative results were issues for both platforms. Another study has reported robust quantitation of BCR-ABL1 levels from 0.0032% to 10% IS across three dPCR platforms: Bio-Rad QX200, RainDance RainDrop System, and Applied Biosystems QuantStudio3D [Huang et al. 2015]. This study successfully detected BCR-ABL1 down to MR⁵ and MR^5.5, demonstrating dPCR approaches enable increased sensitivity detection levels. However, another study reported that BCR-ABL1 values (MR³–MR⁵ range) generated by the Bio-Rad QX200 ddPCR system were consistently higher than values generated by conventional qRT-PCR [Franke et al. 2015]. This resulted in significant differences in the distribution in MR classification between the quantitative methods, which could potentially impact the clinical management of CML patients. Consequently, any newly developed method should not be used without careful evaluation and comparison with conventional qRT-PCR. These studies have demonstrated that dPCR offers increased sensitivity relative to conventional qRT-PCR and may aid future comparisons of DMR rates across different CML therapies; however, the clinical value of the improved sensitivity dPCR offers is yet to be demonstrated.

Deoxyribonucleic acid polymerase chain reaction

BCR-ABL1 mRNA expression levels are not directly related to the number of Ph+ leukemic cells, and hence the detection of persistent leukemic cells using standard qRT-PCR is reliant on the expression of BCR-ABL1. The detection of the BCR-ABL1 genomic DNA (gDNA) breakpoint is an alternative strategy for identifying Ph positivity. There are a number of advantages with DNA-based, as opposed to RNA-based, molecular monitoring. DNA is stable and therefore less subject to degradation during sample handling; is technically less challenging to isolate at a high quality, and because the number of molecules detected directly relates to the number of Ph+ cells, it is theoretically easier to standardize interlaboratory measurements. However, the BCR-ABL1 breakpoints are dispersed throughout ~5 kb of BCR within introns 13 and 14, and ~140 kb region of ABL1 intron 1 [Grossman et al. 1989; Mills et al. 1991; Zhang et al. 1995]. Consequently, a limitation of DNA-based techniques is that patient-specific BCR-ABL1 breakpoints must be characterized to enable specific assay design and subsequent quantitation. Specific PCR oligonucleotide design is further complicated by the high frequency of repeat elements at breakpoint sites [Chen et al. 1989].

Bubble PCR, inverse PCR, long-range PCR followed by nested PCR and next-generation-sequencing (NGS) molecular techniques have been utilized to successfully detect BCR-ABL1 genomic breakpoints [Zhang et al. 1995; Waller et al. 1999; Bartley et al. 2010; Ross et al. 2010; Sobrinho-Simoes et al. 2010; Pagani et al. 2014; Linhartova et al. 2015; Alikian et al. 2016]. To circumvent patient-specific assay design after characterization of the BCR-ABL1 genomic breakpoint, a library of primer pairs and probes has been developed that allows selection of pretested patient-specific DNA quantitative PCR assays [Bartley et al. 2015]. These assays were combined with a highly multiplexed and sequential hybrid primer PCR to quantitate BCR-ABL1 down to MR⁵ sensitivity level, resulting in detection of MRD in 50 samples where MRD was not detected by qRT-PCR assays [Bartley et al. 2010, 2015]. These advances in DNA PCR in an MRD setting have resulted in increased sensitivity of the estimate of BCR-ABL1 positivity compared with the qRT-PCR mRNA assay. Comparison of four BCR-ABL1 quantitative methods, qRT-PCR, DNA qPCR, mRNA dPCR and DNA dPCR, demonstrated that these methods were 100% concordant in detecting the presence of BCR-ABL1 at levels greater than MR⁴ [Alikian et al. 2016]. However, the results were highly discordant when measuring BCR-ABL1 levels below MR⁴, indicating each method holds unique prognostic information at these low levels. To assess the clinical relevance of detectable BCR-ABL1 DNA, a study by Ross and colleagues measured BCR-ABL1 DNA levels in patients prior to stopping imatinib in a clinical trial [Ross et al. 2010]. All patients had undetectable BCR-ABL1 transcripts using strict sensitivity criteria. The aim was to determine whether more sensitive detection of BCR-ABL1 could identify patients destined for molecular recurrence after stopping. Although a higher level of sensitivity could be achieved using a DNA-based assay, BCR-ABL1 DNA was detected in seven of eight patients who sustained undetectable BCR-ABL1 mRNA levels after imatinib cessation [Ross et al. 2010]. This study indicated that BCR-ABL1 DNA presence before imatinib cessation is not predictive of subsequent relapse. Consequently, until further clinical association studies are performed, the clinical utility of DNA-based methods is limited. Accordingly, BCR-ABL1 genomic-breakpoint detection and quantitation has not been implemented into routine molecular monitoring in CML.

Detection of resistance mutations

BCR-ABL1 kinase domain (KD) mutations are a major mechanism of CML resistance to TKI therapy, and accounts for ~50% of acquired resistance in CML. Many different single KD mutants are associated with resistance to imatinib [Gorre et al. 2001; Von Bubnoff et al. 2002; Branford et al. 2003]. The second-generation inhibitors, nilotinib, dasatinib and bosutinib, are active against the majority of imatinib-resistant mutations, but some confer clinical resistance to nilotinib (Y253H, E255K, E255V, F359V and F359C), dasatinib (V299L, T315A, F317L, F317I, F317V and F317C), bosutinib (V299L), or all second-generation inhibitors (T315I) [Shah et al. 2007; Hughes et al. 2009; Muller et al. 2009; Jabbour et al. 2012; Khoury et al. 2012]. BCR-ABL1 KD mutation analysis is recommended to identify mutations that are drivers of resistance [Soverini et al. 2011]. Each mutation has differences in sensitivity to second-generation TKIs, therefore BCR-ABL1 mutation is used to guide selection of salvage therapy after TKI failure [Baccarani et al. 2009, 2013]. BCR-ABL1 KD mutations are commonly detected by Sanger sequencing of BCR-ABL1 cDNA. Sanger sequencing, however, can only detect mutations present above a 10–20% allele load [Hughes and Branford, 2006]. Molecular techniques with greater sensitivity, such as mass-spectrometry genotyping, identify mutations down to 0.5% mutation load [Parker et al. 2011]. When mutations conferring resistance are detected at low levels, they may lead to subsequent clinical resistance [Parker et al. 2011]. Furthermore, the detection of multiple KD mutations is associated with substantially inferior responses compared with patients with no or a single KD mutation [Parker et al. 2012, 2016].

Compound mutations (more than one mutation in a single molecule) have also emerged as a potential driver of acquired resistance to first- and second-generation TKIs, and show different resistance profiles compared with individual mutants [Shah et al. 2007; Khorashad et al. 2013; Zabriskie, 2014]. Ponatinib is a third-generation TKI that overcomes resistance mediated by single mutations in the BCR-ABL1 KD that cause resistance to first- and second-generation TKIs, and is the only TKI that inhibits the T315I ‘gatekeeper’ mutation. There are conflicting reports regarding whether compound mutations confer resistance to ponatinib [Zabriskie, 2014; Deininger et al. 2016]. For patients treated in the chronic phase, ponatinib has been shown to induce durable responses regardless of BCR-ABL1 mutation status, with no single or compound mutations consistently conferring primary or secondary resistance [Deininger et al. 2016]. This may not be the case for patients treated with ponatinib with advanced-phase disease. Advances in NGS-based mutation detection approaches could aid identification of true compound BCR-ABL1 KD mutations to guide clinical decision making regarding the selection of salvage therapy after TKI failure. dPCR or denaturing high-performance liquid chromatography combined with direct sequencing are methods that have been reported to detect BCR-ABL1 mutations [Ernst et al. 2009; Oehler et al. 2009]. However, NGS-based sequencing technologies are emerging as preferred methods for quantifying low-level-resistant mutations.

Molecular barcoding in next-generation sequencing

Routine Sanger sequencing and genotyping techniques do not distinguish between true compound mutations or mutations in trans on separate alleles or polyclonal mutations [Khorashad et al. 2013; Parker et al. 2014a]. Nested PCR amplification of the BCR-ABL1 KD, followed by cloning and Sanger sequencing or NGS sequencing techniques were originally used to detect compound mutations [Khorashad et al. 2013; Soverini et al. 2013; Kastner et al. 2014]. However, subsequent studies demonstrated that sequencing artifacts and PCR recombination can lead to false-positive detection of compound mutations, and hence the rate of compound mutations may have been overestimated [Parker et al. 2014a; Deininger et al. 2016]. Novel advances and refinement of the NGS-based sequencing approaches by the incorporation of unique molecular ID tags into the NGS protocol demonstrated that sequencing errors can be identified and removed from analysis [Parker et al. 2014b]. This resulted in the ability to distinguish compound mutations from PCR-artifacts and polyclonal mutations and specific detection of true compound mutations. Alternatively, Deininger and colleagues devised an algorithm to predict false-positive compound-mutation frequencies [Deininger et al. 2016]. The algorithm was based on the recombination rate of their NGS protocol, which was determined by measuring the frequency at which false-compound mutations were generated when RNA samples from two patients, each with distinct single BCR-ABL1 KD mutations, were mixed. Only compound mutations with an observed frequency greater than the false-compound mutation rate were considered true-positive compound mutations [Deininger et al. 2016].

Long-read next-generation sequencing

Detection of compound mutations by NGS technology can be limited by short-read lengths. Longer-read lengths enable sequencing of single molecules, facilitating detection of compound mutations and splice isoforms [Rhoads and Au, 2015]. Cavelier and colleagues utilized Pacific Biosciences’ single-molecule long-read sequencing to detect mutations in the BCR-ABL1 transcript [Cavelier et al. 2015]. Using this technology, the entire 1578 bp BCR-ABL1 major-fusion transcript was sequenced [Cavelier et al. 2015]. This assay successfully detected mutations down to 1% sensitivity and identified all mutations detected by routine Sanger sequencing. This assay detected five low-level mutations not detected by Sanger sequencing. In addition, multiple BCR-ABL1 transcript isoforms were identified in two CML patients, including a previously reported 35 bp insertion, which does not confer resistance to TKI treatment [O’Hare et al. 2011]. This method readily detected compound mutations; however, a low rate of PCR recombination artifacts was reported. Further validation of this method will be required to adequately assess the false-positive compound-mutation detection rate. Molecular barcoding or development of an algorithm to identify false-positive compound mutations could be used in conjunction with long-read sequencing to more robustly identify sequencing errors [Parker et al. 2014b; Deininger et al. 2016].

Duplex deoxyribonucleic acid sequencing

Duplex sequencing is an alternative approach to conventional NGS sequencing, whereby both strands of a single DNA molecule are sequenced. In this protocol, gDNA is captured with individual biotinylated DNA oligonucleotides, and each of the two DNA strands are tagged with unique molecular barcodes and sequenced [Schmitt et al. 2012]. This method distinguishes between true mutations and amplification, or sequencing-error artifacts, as only true mutations are present at complementary positions in both strands, whereas artifacts are seen in only one strand. The duplex-consensus sequence method has an advantage over single-strand consensus sequencing because sequencing errors that occurred during the first round of PCR can be identified and eliminated [Schmitt et al. 2012, 2015a]. This approach has been reported to improve DNA sequencing accuracy by >100,000 fold [Schmitt et al. 2016]. This methodology was used for the detection of ABL1 KD mutations in heavily pretreated patients, with the majority of patients having received three or more TKIs prior to ponatinib therapy [Schmitt et al. 2015b]. Preliminary data from nine of 29 patient samples demonstrated an average detection limit of one in 11,412 cells (<0.01%), and notably, rare subclones were detected in four patients at baseline, which were undetectable by NGS or Sanger sequencing assays [Schmitt et al. 2015b]. These mutations later emerged during ponatinib treatment, indicating that the mutant subclones pre-existed to ponatinib therapy [Schmitt et al. 2015b]. A limitation of this sequencing strategy is that amplification is not specific to BCR-ABL1 fusion-gene gDNA. Consequently, it cannot be determined whether detected mutations are present in the BCR-ABL1 fusion gene or the wild-type ABL1 gene. Therefore, prospective studies are required to demonstrate the association between low-level-resistance mutations detected by duplex sequencing and drug resistance, or disease progression.

Prediction of therapy outcomes and disease progression

BCR-ABL1 transcript type as a predictor of response to tyrosine kinase inhibitor therapy

Identification of molecular markers that predict response to TKI therapy at time of diagnosis is an active area of research in CML. Whether BCR-ABL1 transcript type is predictive of response to therapy has re-emerged as an area of interest in CML research. Several BCR-ABL1 transcript types have been detected in patients with CML, including the common e13a2 (b2a2), e14a2 (b3a2) and the rare e1a2 and e13a3 transcripts, among others [Shtivelman et al. 1986; Mills et al. 1991; Faderl et al. 1999]. The clinical significance of the uncommon BCR-ABL1 transcripts is difficult to assess due to their rarity, however, the largest study of e1a2 transcript types indicated this transcript is associated with inferior outcomes to TKI therapy [Verma et al. 2009]. In contrast, two case reports have indicated that patients with the e13a3 transcript tend to display good sensitivity to TKI therapy [McCarron et al. 2015; Ha et al. 2016]. The association between transcript type and outcome has long been debated. In the pre-TKI era, patients harboring e14a2 breakpoints had shorter duration of chronic phase compared with those with e13a2 breakpoints (average 36.6 versus 56.1 months) [Grossman et al. 1989]. Moreover, Mills and colleagues reported that a higher proportion of e14a2 patients progressed to blast crisis than those with e13a2 (8/10 versus 3/12) [Mills et al. 1991]. These studies suggested the e14a2 transcript was associated with inferior outcomes. In contrast, several studies reported that there was no association between transcript type and OS in patients treated with interferon (IFN) or chemotherapy (Table 1) [Zaccaria et al. 1992; Shepherd et al. 1995; Colleoni et al. 1996]. Due to these conflicting studies, transcript type was not incorporated into the clinical management of CML patients.

Table 1.

Studies that have assessed the prognostic significance of BCR-ABL1 transcript type with therapy outcomes in CML patients.

Study	Treatment	Patients	Transcript types	Outcome
				CCyR	MMR	DMR	EFS/FFS	PFS/TFS	OS
Pre-TKI
Mills et al. [1991]	IFN, HU	80	e14a2 or e13a2*	–	–	–	–	Y	–
Zaccaria et al. [1992]	IFN, HU	72	e14a2 or e13a2	–	–	–	–	–	N
Inoue et al. [1992]	IFN	26	e14a2 or e13a2*	Y~	–	–	–	Y	–
Shepherd et al. [1995]	IFN, HU	119	e14a2 or e13a2	–	–	–	–	–	N
Colleoni et al. [1996]	IFN, HU	32	e14a2 or e13a2	–	–	–	–	–	N
Post-TKI
Lucas et al. [2009]	IM 400	71	e14a2* or e13a2	Y	–	–	N	–	N
Sharma et al. [2010]	IM 400, IFN or HU+IM	87	e14a2, e13a2* or both	Y	–	–	–	–	–
Castagnetti et al. [2012]	NIL 300 or 400	201	e14a2 or e13a2	N	N	–	N	N	N
Hanfstein et al. [2014]	IM 400 or 800 or IFN+IM	1105	e14a2*, e13a2 or both	N	Y	Y	–	N	N
Dmytrenko et al. [2015]	IM	508	e14a2* or e13a2	–	Y	Y	Y	N	N
Lin et al. [2016]	IM 400	166	e14a2*, e13a2 or both*	–	N	–	–	–	–
Jain et al. [2016]	IM 400 or 800, DAS or NIL	481	e14a2*, e13a2 or both	N	Y	Y	N	Y	N
Adelaide cohort	IM 400, 600 or 800	523	e14a2*, e13a2 or both	–	N	Y	N	N	N

IM, imatinib; NIL nilotinib; DAS, dasatinib; IFN, interferon; HU, hydroxyurea; CCyR, complete cytogenetic response; MMR, major molecular response; DMR, deep molecular response (achievement of MR⁴ or MR^4.5); EFS, event-free survival; FFS, failure free survival; PFS, progression-free survival; TFS, transformation-free survival; OS, overall survival; ^*Transcript type associated with superior response; Y, transcript type was significantly associated with indicated outcome; N, transcript type was not associated with outcome; –, not investigated/reported; ~, this study reported complete hematologic response rather than CCyR; Adelaide cohort, as described in Branford et al. [2014].

In the TKI era, several studies have assessed the prognostic value of the common BCR-ABL1 transcript type with response to TKI treatment (Table 1). A small study of 71 patients expressing the e14a2 or e13a2 transcripts treated with imatinib (400 mg) indicated that patients harboring e14a2 achieved significantly higher cumulative incidence of complete cytogenetic response (CCyR) compared with e13a2 patients, however, no significant difference in OS or EFS between transcript types was reported [Lucas et al. 2009]. In this study, patients with e13a2 transcripts had higher pretreatment pCrKL/CrKL (phospho-CrKL/CrKL) ratio, suggesting e13a2 transcripts have higher BCR-ABL1 kinase activity. In direct contrast, another small study indicated that the e13a2 transcript was associated with a higher rate of CCyR achievement in imatinib-treated patients [Sharma et al. 2010]. No association between transcript type and overall CCyR or major molecular response (MMR) rates, nor FFS, PFS or OS in nilotinib-treated patients has been detected [Castagnetti et al. 2012]. A larger study of 1105 CML patients treated with frontline imatinib demonstrated that the e14a2 transcript was associated with superior achievement of MMR and MR⁴; however, no association with CCyR, PFS or OS was detected [Hanfstein et al. 2014]. Therefore, the authors concluded no risk prediction according to BCR-ABL1 transcript type could be made at diagnosis. Another study of 508 patients treated with imatinib also reported superior achievement of MMR and DMR in patients with the e14a2 compared with e13a2 transcripts [Dmytrenko et al. 2015]. A lower rate of EFS in e13a2 compared with e14a2 patients was also reported, but no difference in PFS or OS was detected. A more recent study of 481 CML patients treated with frontline imatinib (400 or 800 mg) (n = 268), dasatinib (n = 105) or nilotinib (n = 108) also demonstrated that the e14a2 transcript was associated with superior achievement of MMR and MR^4.5 [Jain et al. 2016]. No significant differences in the 5-year probabilities for EFS or OS were detected between transcript types. However, an inferior probability of 5-year transformation-free survival (TFS) was observed between patients with the e13a2 transcript compared with the e14a2 transcript, and the authors concluded transcript type may impact the probability of response to TKI treatment [Jain et al. 2016]. Overall, in the pre-TKI and post-TKI eras, alternate transcript types were associated with outcome; the e13a2 or the e14a2 transcripts tended to be associated with superior outcomes, respectively (Table 1), and inconsistent prognostic significance has been reported across studies.

We have previously investigated treatment outcomes in 528 patients consecutively treated with frontline imatinib [Branford et al. 2014]. That study’s patient cohort provided an opportunity to assess treatment outcomes according to transcript type. The e14a2, e13a2 or both transcripts were detected in 229, 198 and 96 patients, respectively, and five patients had atypical BCR-ABL1 transcripts. MRs (where imatinib cessation for any reason was treated as a competing risk), FFS, PFS and OS were assessed, as stratified by transcript type. In this cohort, BCR-ABL1 transcript type was not predictive of 3-month early molecular response (EMR) (p = 0.212), 12-month MMR (p = 0.701), 48-month MR⁴ (p = 0.068), FFS (p = 0.074), PFS (p = 0.414) or OS (p = 0.393). However, transcript type was predictive of 48-month MR^4.5, whereby patients harboring the e14a2 transcript were more likely to achieve MR^4.5 compared with patients with the e13a2 or both transcripts (38%, 25%, 25% CI of MR^4.5, respectively; p = 0.007). Our data therefore only support an association with MR^4.5 and not with any survival outcomes. Differences in prognostic significance between studies may be related to differences in sample sizes, BCR-ABL1 assays, treatment modalities and outcomes assessed across studies. Given the inconsistent prognostic significance across studies, most notably for survival outcomes (Table 1), the evidence at this stage does not support the incorporation of the transcript type into clinical decision making.

Additional chromosomal abnormalities

In addition to the Ph chromosome, other chromosomal abnormalities can be observed at diagnosis in ~7% of CML patients [Fabarius et al. 2011]. Additional chromosomal abnormalities can be detected by G- or R-banded bone-marrow metaphases or FISH cytogenetics. High-risk ‘major route’ cytogenetic abnormalities, including Ph duplication, isochromosome 17q and trisomy 8 are associated with significantly inferior MRs and shorter OS compared with patients with the standard BCR-ABL1 translocation [Fabarius et al. 2011]. Whilst the results of these cytogenetic tests offer prognostic information, the tests are laborious and only have a limited detection sensitivity of ~1–5% Ph+ cells [Kantarjian et al. 2008]. With the majority of CML patients receiving TKI therapy and most patients rapidly achieving CCyR, this test lacks the sensitivity required to monitor low-level residual disease. Single nucleotide polymorphism (SNP) microarray studies in CML have revealed heterogeneous submicroscopic copy-number variation, which to date have not been yet been linked to clinical outcomes in CML [Soverini et al. 2015]. Whole-genome sequencing (WGS) and RNA-sequencing (RNAseq) are emerging technologies being used for the detection of structural variations and fusion genes [Annala et al. 2013; Zhou et al. 2013; Favero et al. 2015; Korfi et al. 2015]. WGS structural-variation data were used to identify amplification of BCL2, in addition to T315I and F359V BCR-ABL1 KD mutations, in a ponatinib-resistant CML patient [Korfi et al. 2015]. Importantly, in-vitro functional studies demonstrated the patient’s primary CD34+ CML cells showed increased sensitivity to the BCL2 inhibitor, ABT-263, demonstrating the potential of WGS to identify mechanisms of resistance and alternative therapeutic targets in CML.

Detection of fusion genes is one of the most clinically relevant applications of RNAseq [Yoshihara et al. 2015]. One study has reported detection of a novel fusion gene (RNF213-SLC26A11) in a CML patient using RNAseq, however, the clinical significance of this finding is uncertain [Zhou et al. 2013]. Fusion transcripts represent a small proportion of the sequencing reads generated by RNAseq, therefore, is an inefficient method for fusion-gene detection due to the high number of sequencing reads required. Targeted sequencing panels that facilitate deeper sequencing of regions of interest offer an alternative approach for detecting fusion genes. Single-primer enrichment technology has been utilized by the Ovation Fusion Panel Target Enrichment System (NuGen, California, USA), allowing targeted RNAseq of known fusion genes or fusion genes with a known partner, and is reported to identify more fusions while using fewer reads than standard RNAseq [Scolnick et al. 2015]. This method successfully detected BCR-ABL1 fusions in Universal Human Reference RNA, however, the detection of other clinically significant fusions was not reported, and therefore the clinical utility of this technique is yet to be demonstrated.

Conclusion

Advances in molecular techniques have great potential to progress molecular monitoring in CML. Cartridge-based BCR-ABL1 detection is an example of a technology currently being translated into routine molecular monitoring of CML. Similarly, NGS technologies have enabled substantial advances for sensitive and accurate detection of therapy-resistant BCR-ABL1 KD mutations, facilitating personalized treatment approaches. In contrast, increases in the sensitivity of BCR-ABL1 detection and the prognostic significance of BCR-ABL1 transcript types are areas of research which currently have uncertain clinical significance, inhibiting translation into clinical management of CML. Importantly, advances in genetic technologies are paving the way for identification of novel mechanisms of therapy resistance and disease progression, which may have the potential to be incorporated into CML molecular monitoring in the future.