Chronic myeloid leukemia: the paradigm of targeting oncogenic tyrosine kinase signaling and counteracting resistance for successful cancer therapy

CML accounts for 15–20% of all cases of leukemia in adults [1]. Clinical hallmarks of CML are leukocytosis, a left shift in the differential count, and splenomegaly. The natural history of the disease follows a triphasic course with an initial chronic phase (CP), an intermediate accelerated phase (AP) and a final, fatal blastic phase (BP)(Fig. 1). CP may last several years and is characterized by the expansion of the myeloid cell compartment, although cells still retain the capacity to differentiate and function normally. Symptoms in this phase are generally mild and many patients are asymptomatic, being often diagnosed incidentally after a routine blood test. AP, that may have a variable duration from weeks to years and cannot always be recognized, is characterized by the appearance of more immature cells in the blood, frequent constitutional symptoms, and a less favorable response to therapy. The final stage is BP, where immature cells predominate and survival is measured in months. Progression from CP to BP is characterized by an increase in genetic instability leading to the accumulation of genetic/cytogenetic defects additional to the Ph chromosome and increased likelihood of drug resistance (Fig. 1). Although TKIs have greatly improved patient outcomes, up to 5% of patients may still progress from CP to BP and the prognosis of such patients remains quite poor [2]. Comprehensive catalogs of the additional genetic and functional defects observed in BP patients have been compiled [3, 4], but the mechanisms underlying disease progression have not been clarified yet.

Before the advent of targeted therapy, the gold standard for pharmacologic treatment was α-interferon (α-IFN), which was associated with a not negligible toxicity and a median survival time of approximately five years [5]; upfront allogeneic stem cell transplant was the only curative option. TKIs have revolutionized the life expectancy and quality of CML patients and have led to the introduction of the concept of ‘functional’ or ‘operational cure’ [6]. This is defined as avoidance of progression and resistance and durable freedom from any disease sign and symptom despite the possible presence of residual leukemic cells. At first, it was envisioned that functional cure could be achieved only with lifelong TKI treatment. More recently, however, several clinical trials have shown that 40 to 60% of the patients who achieve a deep and durable reduction or clearance of residual BCR-ABL1 transcripts (‘Deep Molecular Response’) after several years of TKI treatment may safely interrupt their therapy without relapsing (‘Treatment-Free Remission’ (TFR); see [7‐9] for detailed reviews on this issue, that is out of the scope of the present manuscript). Current clinical research is therefore focusing on avoiding resistance and increasing the rate of patients successfully achieving TFR.

It was 1960 when a simple light microscope enabled Peter Nowell and David Hungerford to observe that a minute acrocentric chromosome was consistently detectable in the bone marrow cells of CML patients [10]. This chromosome was named ‘Philadelphia’ (Ph) after the city where its discovery took place. In 1973, once again just a microscope was enough for Janet Rowley to uncover that the Ph chromosome was the result of a reciprocal translocation between chromosomes 9 and 22: the t(9;22)(q34;q11) [11]. The subsequent leap forward came when the first molecular biology techniques became available. By the mid-1980s, it could be established that the t(9;22) translocation resulted in the juxtaposition, on the Ph chromosome, of Abelson 1 (ABL1), the human homologue of the v-abl oncogene carried by the Abelson murine leukemia virus (A-MuLV) located on the long arm of chromosome 9, to a gene of unknown function on the long arm of chromosome 22, which was called BCR for Breakpoint Cluster Region, since DNA breaks occurred in a relatively small genomic region [12, 13]. Association of the Ph chromosome with B-cell acute lymphoblastic leukemia (B-ALL) was also discovered [14]. A smaller 7.0 kb mRNA, as opposed to a CML Ph chromosome 8.5 kb mRNA product, was observed in B-ALL patients [15, 16]. Furthermore, the BCR-ABL1 protein product in B-ALL samples was 185/190 kDa (p190^BCR-ABL1) as opposed to the 210 kDa BCR-ABL1 protein product (p210^BCR-ABL1) detectable in CML samples [15, 17]. The differences in the Ph chromosome gene product in B-ALL versus CML were found to be the result of a different localization of BCR breakpoints: in B-ALL, they were mapped within the minor breakpoint cluster region (m-BCR) whereas in CML, they fell within the major breakpoint cluster region (M-BCR) (Fig. 2a). A third region where breakpoints may more rarely cluster is the so called μ-BCR (Fig. 2a). Depending on the breakpoint, and after alternative splicing, different BCR-ABL1 transcripts may result (Fig. 2b). Further studies showed a high but not absolute correlation between the p210^BCR-ABL1 form and CML, and between p190^BCR-ABL1 and B-ALL, questioning whether specific forms of BCR-ABL1 may play a role in the aetiology of each leukemia. A p230^BCR-ABL1 isoform (typical of a subset of CML once called chronic neutrophilic leukemia) resulting from the μ-BCR was later uncovered [18] (Fig. 2a-b). Over the years, additional, more rare fusion schemes have also been reported (Additional file 1: Figure S1).

Seminal was the discovery that the protein derived from the chimeric BCR-ABL1 gene had tyrosine kinase activity, that derived from normal ABL1 but was deregulated as a consequence of the translocation, and correlated with the ability to induce malignant transformation [19].

The BCR-ABL1 protein acquires some domains from BCR and others from ABL1 [20]. Domains from BCR include, depending on the genomic breakpoint position (Fig. 2c):

Domains from ABL1 include (Fig. 2c):

The reason why native ABL1 has a tightly regulated kinase activity whereas BCR-ABL1 shows constitutive activation lies essentially in the fact that BCR-ABL1 loses the the N-terminal “cap” (N-cap), a region with a signal sequence for myristoylation playing a critical regulatory role. The N-terminal myristic acid group binds a deep hydrophobic pocket in the C-terminal lobe of the kinase domain. Interaction of the myristoylated N-cap with the C-terminal lobe is critical to maintain an autoinhibited state. Loss of this region, together with fusion of BCR sequences encompassing the oligomerization domain and Y177, abrogate the physiologic control of the kinase.

The understanding of native ABL1 functions (recently reviewed in [21]) was the key to unravel how BCR-ABL1 may promote cellular transformation. The ABL1 protein is implicated in a wide range of cellular processes, including regulation of cell growth and survival, oxidative stress and DNA-damage responses, actin dynamics and cell migration, transmission of information about the cellular environment through integrin signaling. To this purpose, ABL1 interacts with several cellular proteins – including signalling adaptors, other kinases, phosphatases, cell-cycle regulators, transcription factors and cytoskeletal proteins. Overall, it seems that the ABL1 protein serves as a key hub that integrates signals from various extracellular and intracellular sources to control cell cycle and apoptosis. Two major mechanisms have been implicated in the malignant transformation by BCR-ABL1: a) altered adhesion to bone marrow stroma cells and extracellular matrix, and b) constitutively active mitogenic signaling and reduced apoptosis [22]. Several cellular cascades are hijacked by BCR-ABL1 to promote CML. They include the RAS/RAF/MEK/ERK pathway, the JAK2/STAT pathway, the PI3K/AKT/mTOR pathway (reviewed in [23]).

How slightly different BCR-ABL1 isoforms (p190^BCR-ABL1 vs p210^BCR-ABL1) can trigger such diverse diseases (CML has an indolent course and TKI therapy results in stable remissions in the great majority of cases; Ph + ALL is much more aggressive, responses to TKIs are not durable and prognosis is relatively poor) has long been under investigation. Besides the clearly different cell of origin, several studies over the years have addressed the issue of which pathways may be differentially activated by the two isoforms, up to two very recent quantitative comparative proteomic studies comparing their respective ‘interactomes’ and ‘phosphoproteomes’. [24, 25] Both studies showed, surprisingly, no differences in the extent of autophosphorylation and kinase activation. However, they identified differential interactions, differential signaling networks and also differential intracytoplasmatic localization [24, 25].

CML is considered a paradigm for precision medicine in that it is caused by a single deregulated protein that exhibits a ‘druggable’ gain of function and is expressed in leukemic cells but not in normal cells. The success of targeted therapy in CML has not yet been replicated in other malignancies since cancer is most frequently the result of stepwise accumulation of multiple genetic defects [26]. How can BCR-ABL1 be necessary and sufficient for disease initiation and maintenance? And is it really sufficient?

In vitro culture systems demonstrated that BCR-ABL1 can transform immature hematopoietic cells, some fibroblast cell lines, and hematopoietic cell lines rendering them growth factor-independent. In addition, several groups reported that a CML-like disease could be induced in mice transplanted with bone marrow infected with a BCR-ABL1 retrovirus. In contrast, mutant isoforms of BCR-ABL1 carrying inactivating mutations in the SH1 domain, or mutants lacking the BCR coiled coil domain, did not induce leukemia. All these studies [27‐30], conducted around the 90s, converged to demontrate that BCR-ABL1 is indeed the causative agent of CML and fostered the search for small molecule inhibitors. On the other hand, evidences have also been brought that challenge this view. There are marked strain differences in disease induction after BCR-ABL1 retroviral expression, suggesting that the genetic background may influence the ability of the oncogene to initiate CML [29]. Even more interestingly, a conditional knock-in mouse in whom the human BCR-ABL1 cDNA was knocked into the endogenous mouse Bcr locus so that it could be conditionally expressed with different tissue-specific Cre transgenes under the added control of the native Bcr regulatory elements, was found not to develop leukemia during its lifetime, despite expression of a constitutively active BCR-ABL1 tyrosine kinase was observed in the hematopoietic progenitors [31]. The authors thus postulated that i) physiologic BCR-ABL1 expression may be insufficient for development of a CML-like disease; ii) in the retroviral or transgenic models, non-physiologic, very high levels of BCR-ABL1 expression due to multiple copies of the oncogene and expression from a very active retroviral promoter, non-specificity of expression timing and locale and maybe also random insertion-site mutations could artificially select for disease development [31]. This study was published in 2013, but the idea that additional cooperating events might be required for the induction of CML was, indeed, not new. Between the 80s and the 90s, initial evidences were brought in support of the existence of a putative event preceding the acquisition of BCR-ABL1 at least in a proportion of patients. Studies of X chromosome inactivation and glucose-6-phosphate dehydrogenase genotype had raised the hypothesis that clonal hematopoiesis might precede the acquisition of the Ph chromosome [32, 33]. In addition, starting from the 90s, five reports had been published about the detection of BCR-ABL1 transcripts in circulating leukocytes of up to 65% of healthy individuals when using sensitive polymerase chain reaction (PCR)-based assays [34‐38]. Overall, 380 samples have been analyzed in these studies. BCR-ABL1 was detected in cord blood and newborns (up to 40%), children and adolescents (up to 56%), adults (20–59 yrs.; up to 65%) and elderly (> 60 yrs.; up to 65%). For unknown reasons, the e1a2 rearrangement (leading to p190^BCR-ABL1) was much more frequently detected than the e13a2 or e14a2 rearrangements (leading to p210^BCR-ABL1). It might be argued that in all the studies a nested reverse transcription (RT)-PCR strategy was used to enhance sensitivity, although such an approach has the known drawback of being more prone to contamination. Unfortunately, there is no follow-up information available for BCR-ABL1-positive cases. The latency period between acquisition of the Ph chromosome and overt clinical development of CML is unknown and it is likely to be highly variable. Atomic bomb survivors could develop CML up to 40 years later. On the other hand, there are reports of children > 1 year of age who were diagnosed with CML [39]. In spite of the technical issues, these data, together with case reports of patients with detectable Ph chromosome in their bone marrow cells but otherwise asymptomatic (with a follow-up of few years only, however) [40, 41] raise, among others, the hypothesis that other events are needed before a true malignant expansion can occur and overt CML may develop. Mathematical models predict that 2 or more genetic hits in the hematopoietic stem cells may be needed for CML to develop [42, 43]. Although CP CML has long been considered a genetically homogeneous entity, the power of next generation sequencing (NGS) is now changing this view. A few years ago, targeted NGS-based resequencing of the 25 most commonly mutated genes in myeloid leukemias/myelodysplasias revealed ASXL1, TET2, RUNX1, DNMT3A, EZH2 and TP53 mutations in 5 out 15 chronic phase CML patients at diagnosis [44]. In the same study, analysis of individual hematopoietic colonies showed that the great majority of mutations were part of the Ph + clone. However, targeted resequencing of subsequent samples during TKI treatment revealed that the DNMT3A mutation found in the Ph + cells of a patient at diagnosis was also present in the Ph- clone, implying that it preceded BCR-ABL1 acquisition. [44] Now we know that DNMT3A, TET2 and ASXL1 mutations, among others, may indeed be found in healthy elderly individuals, where they correlate with the risk of hematologic cancer and all-cause mortality (‘CHIP’, clonal hematopoiesis of indeterminate potential) [45‐47]. Such mutations are thought to represent the first hit, leading to a clonally expanded pool of pre-leukemic hematopoietic stem cells from which overt leukemia may subsequently evolve through the acquisition of additional, disease-shaping genetic lesions [48]. Most recently, a NGS-based screen of 92 myeloid-associated genes in 300 serial samples from 100 CP CML patients at diagnosis and after TKI therapy showed evidence of DNMT3A, TET2, ASXL1, BCOR and CREBBP mutations in both diagnosis and follow-up samples, despite response to TKI therapy and BCR-ABL1 transcript clearance [49]. This further indicates that up to 10% of CML patients may have CHIP-related mutations and reinvigorates earlier hypotheses of a multistep pathogenesis of CML – arising, at least in some cases, from pluripotent stem cells of a pre-existing Ph- clone that enjoys a growth advantage.

Prospective serial screening of healthy individuals to determine whether the presence of the BCR-ABL1 oncogene in their blood predicts for future CML development would be of great interest. To this purpose, the use of digital PCR would enable to conjugate high sensitivity with a more precise and accurate count of BCR-ABL1 transcripts. However, because CML occurs at a frequency of 1–2 cases per 100,000 per year, a very large cohort would be needed, together with analysis of an equal number of individuals without detectable BCR-ABL1 transcripts.

Whether or not the only genetic (or epigenetic) hit, BCR-ABL1 is the main disease driver in CP CML, as testified by the remarkable clinical efficacy of TKIs. Based on the structural and functional features of BCR-ABL1, two inhibitory strategies have been devised. ATP-competitive inhibitors bind the kinase domain in the cleft between the N-terminal lobe and the C-terminal lobe. In contrast, allosteric inhibitors do not compete with ATP binding and rather bind to sites that are important regulators of kinase activity (Fig. 3).

It was 2001 and imatinib was still undergoing phase I-II trials when C. Sawyers’ group reported that BCR-ABL1 could escape from inhibition [73]. The analysis of a handful of patients with BP CML who had relapsed after an initial response had shown reactivation of BCR-ABL1 kinase activity despite continued imatinib treatment. A mechanism interfering with imatinib binding was hypothesized, and the entire kinase domain was sequenced in search of point mutations at some BCR-ABL1-imatinib contact residue. Strikingly, an identical substitution of Threonine to Isoleucine at residue 315 (T315I) was identified in six out of nine patients [73]. Initially, this finding casted a shadow over the long-term stability of responses to targeted therapy, since at that time it was difficult to predict how frequently such mutations would arise, thus neutralizing imatinib efficacy. Later on, however, it was realized that the earlier in the disease course TKI therapy is commenced, the lower is the relapse rate and the degree of genetic instability responsible for mutation acquisition. So, if TKI-resistant mutations remain, even nowadays, a challenge in patients with AP and BP, they arise much less frequently in CP patients who receive front-line TKI therapy [74]. In this setting, less than 30% of patients who fail therapy are found to harbor mutations (Soverini et al., unpublished).

Threonine 315 was later named ‘the gatekeeper’ residue, because it is strategically positioned to control the accessibility of the ATP-binding pocket. On binding, the hydroxyl group of Threonine 315 forms a hydrogen bond with imatinib, and the side chain present at position 315 also sterically controls the binding of the inhibitor to hydrophobic regions adjacent to the ATP-binding site [51, 75]. The substitution of Threonine with the bulkier and more hydrophobic Isoleucine was shown to eliminate this hydrogen bond, required for high-affinity inhibitor binding, and to create a steric hindrance interfering with imatinib placement [73, 75]. Notably, Threonine 315 is essential for imatinib binding but not for ATP binding. This means that the catalytic activity, hence the tumour-promoting function, is preserved in the imatinib-resistant T315I mutant. A strikingly identical amino acid substitution was later observed at homologous positions in the kinase domain of c-KIT (T670I) and PDGFRα (T674I) in imatinib-resistant gastrointestinal stromal tumors and hypereosinophilic syndromes, respectively [76, 77], further highlighting the central role of this highly-conserved ‘gatekeeper’ threonine in controlling the accessibility of the ATP-binding pocket. Accordingly, the T315I confers resistance to all the currently approved second-generation TKIs (dasatinib, nilotinib and bosutinib) and only the third-generation TKI ponatinib has demonstrated in vitro and in vivo activity against this mutant.

As the number of imatinib-resistant patients increased, sequencing of the kinase domain revealed a plethora of additional mutations. At present, more than 50 different mutation hotspots are known (Table 2). However, marked differences in IC₅₀ values (the intracellular concentration of the drug required to inhibit by 50% proliferation or viability of a BaF3 cell line engineered to express a given BCR-ABL1 mutant) have been observed across these mutants, suggesting that the degree of insensitivity to imatinib may be variable [78]. Imatinib-resistant mutations have been detected at contact residues (F317L, Y253H), in the phosphate-binding loop (P-loop)(G250E, E255K), in the A-loop (H396R), and in other regions of the kinase domain where amino acid substitutions may possibly force the equilibrium towards the active conformation of the kinase, whom imatinib is unable to bind. In vitro sensitivity profiling, corroborated by clinical experience, have identified much smaller spectra of resistant mutations for second-generation TKIs (Table 2) and these spectra are essentially non-overlapping (with the exception of the T315I mutation, as anticipated above). Hence, BCR-ABL1 kinase domain mutation screening is recommended in patients who fail TKI therapy, since detection of specific mutations influences the choice of the second- or subsequent-line TKI [79]. Ponatinib was rationally designed to bind mutant BCR-ABL1 as effectively as it binds native BCR-ABL1. Indeed, it is the only currently available option for T315I-positive patients [80]. Anecdotal reports, however, suggest that under the selective pressure of ponatinib, the T315I may further change into T315M or T315L [81, 82].

Sequencing of TKIs in patients who fail multiple lines of therapy has more recently brought up the issue of compound mutations. A compound mutant arises when two mutations are acquired by the same BCR-ABL1 molecule, thus by the same clone, as opposed to polyclonality where two clones acquire a single mutation each (Additional file 3: Figure S3). The term ‘compound mutant’ was coined at the very dawn of the second-generation TKI era – when dasatinib treatment of some imatinib-resistant patients was found to result in the acquisition of dasatinib-resistant mutations by BCR-ABL1 molecules already harbouring imatinib-resistant mutations [83]. Double compound mutants are by far the most frequent; compound mutants with three or even four mutations may also, occasionally, be detected – but too many mutations seem to be poorly tolerated [84, 85]. Detection of compound mutants might have important clinical implications. According to two recent studies, the IC₅₀ values of second-generation TKIs and of ponatinib experimentally derived for many compound mutants are much higher than those each single mutant would exhibit [86, 87]. Such in vitro data suggest that i) the great majority of compound mutants are likely to be highly resistant to all second-generation TKIs; ii) some compound mutants might be challenging even for ponatinib. Very recently, a study in mice has predicted mutations interfering with asciminib binding. Such mutations (A337V, P465S, V468F, I502L) hit different residues as compared to those detected in case of resistance to ATP-competitive inhibitors, hence the hypothesis that combining both inhibitory modes might prevent mutation-driven resistance [70].

Kinase domain mutations are the most extensively studied mechanism of TKI resistance (mainly because of its actionability), but they are neither the only nor even the most frequent one (Fig. 5) [88]. Little, however, is known about other mechanisms, that have been investigated only in cell line models or in very small subsets of patients. In the pivotal study by Sawyer’s group, 3 patients who were negative for the T315I mutations were found to carry multiple copies of the BCR-ABL1 gene by fluorescence in situ hybridization analysis and a 4–20-fold increase in BCR-ABL1 transcript levels [73]. This mechanism, most frequent in advanced phase patients, can be overcome by the more potent second-generation TKIs. BCR-ABL1-independent mechanisms have also been reported or hypothesized to occur in imatinib-resistant patients. Activation of compensatory pro-survival/anti-apoptotic pathways may play a role. In this regard, overexpression or hyperactivation of some members of the SRC family of kinases (LYN, HCK), key effectors downstream of BCR-ABL1, have been described in cell lines and in some imatinib- and nilotinib-resistant patients [89‐92]. This was one of the rationales that prompted the clinical development of dasatinib and bosutinib, dual SRC/ABL1 inhibitors. More recently, other molecules have been implicated in BCR-ABL1-independent TKI resistance and evaluated as therapeutic targets in in vitro studies: FOXO1 [93], β-catenin [94], STAT3 [95], the nucleocytoplasmatic transport molecules RAN and XPO1 [96], Cobll1 and NF-κB signalling [97], the AXL tyrosine kinase [98]. However, it is premature to tell whether these recent findings will translate into more effective therapeutic strategies for resistant patients.

Primary resistance (i.e., upfront failure to achieve a satisfactory response to therapy, as opposed to relapse after an initial response) has been linked to altered expression levels and/or function of the transporter molecules responsible for imatinib influx/efflux. Efflux proteins like the P-glycoprotein (Pgp or MDR1) encoded by the ABCB1 gene, have been shown to play a role in some in vitro studies [99, 100]. Certain ABCB1 polymorphisms have also been reported to predict for response to imatinib [101‐103], although there is not complete concordance among different studies, most probably because of the heterogeneity in patient populations and of the relatively small sample sizes. The expression and function of the human organic cation transporter 1 (hOCT1), mediating imatinib uptake, have also been linked to differences in response rates in imatinib-treated patients [104, 105]. For some second-generation TKIs like dasatinib and nilotinib, transport in and out of cells is known not to rely on these molecules, which explains why the limited efficacy of imatinib may be overcome by switching to another drug [106, 107].

It is also well established that CML stem cells are intrinsically insensitive to TKIs, mainly because they do not require BCR-ABL1 kinase activity for their survival. CML stem cells thus survive TKI therapy and represent a dangerous reservoir from which resistance/relapse may originate. In addition, stem cell persistence is thought to be (one of) the reason(s) why treatment-free remission may not be pursued in approximately half of the cases. Several molecules and pathways have been identified in an attempt to eradicate CML stem cells (extensively reviewed in [108]), but very few combinations of TKI plus drugs targeting such molecules/pathways have so far progressed from preclinical to clinical testing.

Last but not least, it is important to remember that in many cases, a sudden increase in disease burden as assessed by BCR-ABL1 transcript level measurement, or even a relapse, must be ascribed not to a biological rissue but to patient nonadherence to therapy [109‐112]. Compliance represents a major problem for all chronic, self-administered treatments. Although CML is a life threatening disease if not properly treated, and although TKIs are generally well tolerated, patients’ perception regarding the importance of regular TKI assumption and regarding the burden of adverse reactions may be very different from physicians’ perception. This results in non-intentional or even in intentional lack of compliance, which may have serious consequences if not timely identified and addressed.