Molecular pathogenesis of the myeloproliferative neoplasms

In MPN, constitutive activation of the JAK/STAT signalling pathway is a critical mediator of the pathogenesis. A point mutation in exon 14 of the JAK2 gene results in a single amino acid (valine to phenylalanine) substitution and conformational change in the JH2 pseudo-kinase domain of JAK2. This results in constitutive tyrosine phosphorylation activity by disrupting the normal inhibitory action of the JH2 domain. This JAK2 V617F transcript therefore drives constitutive activation of the JAK/STAT pathway in the absence of EPOR, MPL or G-CSFR ligand binding. This mutation is detected in 95% of PV patients and approximately 50% of ET and PMF patients [5, 37, 38]. The resulting disease phenotype is subject to several additional variables including homo or heterozygosity of the JAK2 V617F, variant allele frequency, additional co-operating mutations and/or external influences including iron deficiency. The remaining 5% of PV patients are almost entirely accounted for by mutations in exon 12 of the JAK2 gene through predominant activation of EPOR signalling pathways driving an erythrocytosis [4]. These JAK2 exon 12 mutations have not been seen in PMF or ET.

The majority of JAK2 V617F negative ET and PMF patients have detectable mutations in MPL or CALR [7, 8, 38]. These mutations drive disease through activation of MPL receptor and subsequent downstream JAK/STAT activation. Generally, the driver mutations occur in a mutually exclusive manner. A number of activating MPL mutations have been identified in the transmembrane domain encoded by exon 10 in both familial and sporadic forms of MPN [8, 39]. These gains of function mutations including W515L and S505N constitutively activate downstream JAK/STAT signalling by removing an inhibitory element and inducing dimerization, respectively [39, 40]. Additional activating or augmenting mutations identified in MPL transmembrane domain by deep mutational scanning screens have also been previously identified in MPN patients demonstrating an inherent susceptibility in the MPL gene [41].

Calreticulin (CALR) is an endoplasmic reticulum chaperone protein which in mutant form will interact directly with the TPO receptor MPL driving TPO independent activation. Numerous mutations in the CALR gene have been described with the majority classed as type 1 resulting from a 52-bp deletion in exon 9 or type 2 with a 5-bp insertion in exon 9. The subsequent activation of MPL and downstream JAK/STAT signalling is dependent on a positively charged C terminus resulting from a frameshift in exon 9 and enabling the CALR lectin binding domain to maintain a stable interaction with MPL [42]. There is a recognised distinct clinical phenotype between patients with detectable type 1 and type 2 CALR mutations in both patients and murine models. Type 1 deletions are significantly over-represented in myelofibrosis and produce a more pronounced MPN phenotype in mice [43].

A small minority of ET and PMF patients fall into the “triple negative” category with no detectable mutation in JAK2, MPL or CALR. The incorporation of additional genetic tests into the work-up of these patients has challenged the notion of true triple negativity in MPN. Retrospective evaluations demonstrate that some of these patients may have other genetic markers of clonality detectable or subsequently test positive for a driver mutation. A few patients remain with characteristic phenotypic and morphological features and no detectable genetic abnormalities [44].

Frequently, these JAK/STAT activating driver mutations are the only detected genetic abnormality in MPN patients with one large study reporting this in 45% of patients using a targeted myeloid next-generation sequencing (NGS) panel [15]. The presence of JAK2, CALR or MPL mutations alone are sufficient to generate an MPN phenotype, albeit polyclonal in nature, in murine models [45]. One study of gene expression profiling by microarray analysis demonstrated features of activated JAK/STAT signalling in MPN patients regardless of clinical phenotype or mutational status [2]. Ruxolitinib, a JAK1/2 inhibitor, is effective across all mutant driver backgrounds [46]. It is therefore clear that constitutively activated JAK/STAT signalling is a key feature of disease pathogenesis.

The presence of the JAK2 V617F mutation is also evidently more than a simple switch for excess proliferation. There is significant heterogeneity in terms of the variant allele frequency (VAF), and therefore, clonal size measured in peripheral blood granulocytes is detectable across the MPN patient population [47]. Patients with homozygosity or high VAF tend towards a PV phenotype rather than ET [48, 49]. And yet, there are many PV patients with a low VAF and similarly ET patients with high VAF. A rare subgroup of patients presenting with splanchnic vein thrombosis (SVT) frequently exhibit normal or near normal blood counts with a small JAK2 V617F clone detectable [50]. The JAK2 V617F mutation is also detectable in individuals with CHIP who exhibit no MPN phenotype. Despite the normal blood counts, these individuals have been observed to have a significantly increased risk of cardiovascular disease [20]. Paradoxically, JAK2 V617F positive ET cases are significantly more likely to have thrombotic complications despite lower platelet counts than their CALR mutated comparators [51]. One suggestion in these cases is that the mutant JAK2 results in qualitative changes enhancing the pro-thrombotic phenotype. There is evidence to suggest enhancing endothelial–erythrocyte interactions via activation of Lu/BCAM or enhanced neutrophil extracellular trap formation [52, 53]. A recent study has demonstrated that in patients with a low VAF (< 20%) and therefore small mutant clones within the peripheral blood granulocytes, there is significant heterogeneity of clone size within the reticulocytes and platelets measured using a quantitative polymerase chain reaction established to measure JAK2 V617F RNA. In many cases, the clonal sizes in the reticulocytes and platelet populations were much higher than the granulocytes perhaps from late expansion of erythroid and platelet precursors [54]. We may not therefore have been accurately assessing clonal size in many of our “low allele” patients, and granulocyte VAF may underestimate the qualitative effect of the mutant JAK2 presence.

The complexity of STAT signalling has identified roles for STAT proteins in oncogenesis and tumour suppression, occasionally with conflicting roles in the same tumour type [55]. Investigations of STAT protein recruitment, phosphorylation and ultimately dominance of transcriptional control in MPN have focused on the role of STAT5, STAT1 and STAT3. STAT5 activation was identified early as a key mediator of MPN pathogenesis with experimental work able to demonstrate a dependence on STAT5 to generate a MPN phenotype [56, 57]. STAT5 transcriptional activity is upregulated by the expression of JAK2 V617F in cell lines [37]. In analysis of ex vivo colony forming assays from ET and PV patients, transcriptional analysis demonstrated an enrichment of STAT5A targets with nuclear phosphorylation of STAT5A identified in JAK2 V617F position colonies from both ET and PV patients but not wild type colonies whilst a recent phospho-proteomics approach identified STAT5 and STAT3 as differentially phosphorylated in JAK2 V617F mouse haematopoietic cells [58, 59]. Conditional expression of a null STAT5a/b gene resulted in a failure of a JAK2 V617F mouse model to develop polycythaemia but did not abrogate the risk of myelofibrosis [60]. In an alternative JAK2 V617F mouse model STAT5 deletion resulted in loss of the PV phenotype which could be rescued by STAT5 re-expression [61]. Taken together, STAT5 signalling appears to play a key mediator role in generating the PV phenotype.

Identification of enhanced enrichment of Interferon-gamma target genes in ET in comparison to PV highlighted STAT1 signalling as a potential mediator of differential molecular response between the two disorders. In keeping with this, phosphorylated STAT1 was detectable in ET patients and not PV patients ex vivo [58]. Murine models subsequently demonstrated the loss of STAT1 producing a phenotype favouring erythropoiesis at the expense of megakaryopoiesis and with a reduction in fibrosis [62]. STAT1 phosphorylation at serine 727 may drive proliferation and restrain megakaryocyte differentiation in blast phase MPN. Blocking this serine phosphorylation resulted in different functional outcomes in comparison to blocking tyrosine phosphorylation with ruxolitinib [63]. These results have suggested that an altered balance between STAT1 and STAT5 signalling may be one possible cell intrinsic mechanism of phenotype determination. However, erythroblasts harbouring the JAK2 exon 12 mutations which drive erythrocytosis and are only associated with PV have a transcriptional profile which cannot be distinguished from JAK2 V617F positive ET erythroblasts with no differential in STAT1 activation [64].

Constitutive activation of STAT3 was identified in a number of MPN patients from granulocytes in advance of the discovery of JAK2 V617F, whilst higher levels of STAT3 tyrosine phosphorylation have been identified in JAK2 V617F positive individuals and as a result of JAK2 V617F expression in murine models [65‐67]. A murine model of STAT3 hyperactivity induced by deletion of suppressor of cytokine signalling (SOCS) 3 spontaneously develops myeloproliferative and lymphoproliferative pathology with serine phosphorylation of STAT3 critical [68]. STAT3 deletion results in an altered MPN phenotype in JAK2 V617F mice with reduced neutrophilia and enhanced thrombocytosis present [69].

The canonical tyrosine phosphorylation, nuclear translocation and transcription factor activity are only one role of the STAT proteins. Unphosphorylated STAT proteins appear to have important roles in the normal maintenance of the epigenome in HSC and progenitor cells [70]. As mentioned, serine phosphorylation of STATs may affect transcriptional control and the methylation of STAT3 by Enhancer of Zeste 2 (EZH2) identified as a mediator of transcriptional control in other solid tumours requires investigation in MPN to fully understand the dynamics at play between STAT proteins and final phenotypes [63, 68, 71]. Therefore, as with all aspects of molecular biology in MPN, differential STAT1/STAT3/STAT5 mobilisation is likely to tell only part of the story.

In normal health, intracellular signalling cascades are closely regulated positively by ligand binding to cell surface receptors and negatively by a number of intracellular components acting predominantly as phosphatases or targeting proteins for ubiquitination and subsequent proteasome-mediated degradation. The SOCS proteins are critical negative regulators for JAK/STAT signalling. SOCS3 is a key negative regulator of EPO signalling and therefore erythropoiesis through its interaction with JAK2 and EPOR. There is conflicting evidence on the role of SOCS proteins in the regulation of mutant JAK2. One study has demonstrated that in the presence of the JAK2 V617F mutant, SOCS3 undergoes tyrosine phosphorylation, losing the ability to negatively regulate JAK2 and may in fact potentiate the effect by stabilising the mutant JAK2 [78]. A more recent investigation has demonstrated that mutant JAK2 exhibits comparable inhibition by SOCS3 in in vitro kinase assays [79]. Knockdown of SOCS3 in JAK2 V617F expressing Ba/F3 cells further enhances phosphorylation of STAT3, STAT5 and extracellular signal-regulated kinase 1/2 (ERK1/2) suggesting a tumour suppressor role for SOCS3 in the mutant state [80]. One small study has suggested aberrant regulation of SOCS3 expression in PMF with observation that SOCS3 promoter regions were methylated in 32% of PMF patients but not ET or PV patients [81]. The JAK inhibitor ruxolitinib can clearly reduce the expression of SOCS3 in a number of cellular contexts outside of MPN [82, 83] and we and others have shown that RNA sequencing data confirm reduced expression in MPN cell line models [84, 85]. Histone deacetylase inhibition may upregulate SOCS3 expression in MPN [86‐88].

CBL encodes the E3 ubiquitin ligase Cbl proto-oncogene (CBL) and is an important regulator of many tyrosine kinases. CBL mutations have been identified in many malignancies and at a low but significant level in MPN [15]. These mutations drive myeloproliferation and result in activated JAK/STAT and PI3K/AKT signalling in murine models [89]. In addition, loss of specific phosphatase activity of phosphatase and tension homolog deleted on chromosome 10 (PTEN), a regulator of PI3K/AKT signalling can drive an MPN phenotype [90]. Conversely, higher expression of dual-specificity phosphatase 1 (DUSP1) may be required in the JAK2 V617F context to protect the cells by moderating JNK/P38 MAPK signalling and protecting against accumulating DNA damage [91].

Figure 2 highlights active intracellular signalling cascades identified in MPN and examples of intracellular negative feedback mechanisms. Understanding the complexities of this intrinsic regulation of intracellular signalling in vivo in the heterogenous cellular contexts of MPN patients may help to improve our understanding of pathogenesis and treatment options.

As larger cohorts of individuals continue to be analysed it is clear that the presence of particular mutations is associated both with the disease phenotype and ultimate prognosis. Chronic phase PV and ET patients are significantly more likely to have no additional mutations when compared to PMF patients. The occurrence of particular genes has been significantly associated with disease phenotype. For example, NFE2 mutations correlate with a PV phenotype, spliceosome component mutants and epigenetic regulators EZH2 and ASXL1 are more frequently observed in PMF, whilst other genes including IKZF1 are almost exclusively observed in blast phase disease [15, 96, 97]. Early studies demonstrated clear prognostic implications of particular co-operating mutations. In a cohort of 483 European PMF patients with validation in an American group, ASXL1, EZH2 and SRSF2 mutations independently predicted shortened survival. IDH1/2 or SFSF2 mutations were associated with leukaemic progression in these cohorts with TP53 strongly associated in another cohort [98, 99]. The negative prognostic impact of ASXL1 appears to be evident when present with another high risk mutation but not solely on its own [100]. In a large study of 2035 patients, eight genomic subgroups were identified within the MPN spectrum, each reflecting a different proportion of PV, ET and PMF patients with variable risks of leukaemic or fibrotic transformation and overall survival. Genetic factors including TET2, SRSF2 and ASXL1 mutations contributed to over 50% of the risk factors for fibrotic transformation from PV or ET using a predictive modelling approach. Similarly, over a third of the risk of leukaemic transformation was attributable to genetic factors including TP53 mutation [15]. Figure 3 demonstrates mutated genes identified as contributors to leukaemic and fibrotic transformation from this predictive modelling approach integrated with results from multivariate analysis approaches in other large cohorts [15, 98, 101, 102]. Individualised prognostic evaluations can now be achieved at diagnosis based on the incorporation of genetic and clinical information, and molecular data are now routinely incorporated into prognostic scoring systems including MIPPS70 + and GIPSS [101, 103].

The variant allele frequency reported by next-generation sequencing (NGS) myeloid panels provides a rough estimate of clonal size and dominance, reflecting both the number of affected cells and whether these cells are hetero or homozygous for the mutation of interest. Variant allele frequencies of driver and co-occurring mutations are seldom identical, reflecting a clonal hierarchy comprising dominant and sub-clones with differing genetic abnormalities and resulting competitive advantages. Application of single-cell technologies in myeloproliferative patients has demonstrated complex clonal hierarchies with dominant and sub-clones demonstrating distinctive transcriptomic signatures within individual patients [16]. The clonal complexity is greater in MPN in comparison to CHIP patients but much less than that observed in AML [104].

This complexity reflects clonal evolution over time, occurring as cells acquire additional genetic or epigenetic changes driving further divergence from the cell of origin. This can significantly alter the clonal structure of the disease with smaller sub-clones developing the competitive advantage to establish dominance. Patients progressing to myelofibrosis and AML often demonstrate greater genetic complexity within clones with increasing numbers of mutations detectable. Ultimately, this results in transcriptional and functional changes which are observed in patients cells underlying the progressive disease phenotype [105]. In chronic phase MPN, the rate of acquisition of new mutations is generally considered to be a slow process. Only two additional mutations were observed over the equivalent of greater than 130 patient years in one study [99]. However, this generalisation, like many, does not reflect the significant heterogeneity across the MPN patient spectrum with both acquisition of new clones resulting from new mutations or minor sub-clones establishing dominance observed in patients transforming from chronic phase to blast phase disease [104, 106]. Meanwhile, approximately one-third of PMF individuals receiving the JAK inhibitor ruxolitinib demonstrated clonal evolution on therapy acquiring new mutations [107]. Perhaps as our repositories of sequential pre- and post-MPN progression sequencing data increases, we may understand the genetic, epigenetic and extrinsic factors explaining the propensity for clonal evolution in a number of individuals.

Recent comprehensive genomic profiling with a whole genome sequencing approach and phylogenetic reconstruction based on the numerous somatic mutations occurring during life have provided evidence that acquisition of both driver and common co-occurring mutations may occur many decades prior to disease presentation. In a number of cases, the acquisition of JAK2 V617F or DNMT3A mutations were predicted to have occurred in utero. This approach reveals a significant variability in the rates of clonal expansion, dependent on the mutational landscape of the clone [95]. A concept of mutational compatibility can be theorised from much of the genetic data available in myeloid malignancy to date. A number of individual mutations are frequently observed together, whilst others appear strictly mutually exclusive of each other [99]. In some cases, this mutational compatibility clearly drives a competitive advantage for the cell such as that demonstrated by FLT3-ITD and NPM1 mutations in AML, rapidly establishing a clonal dominance and driving a clear phenotype [104]. In MPN, this mutational compatibility is more subtle but evident from the high frequencies of many mutations including TET2, ASXL1 and DNMT3A occurring alongside drivers observed across the MPN patient spectrum. It is interesting that many of these co-occurring mutations have been observed to be the initiating mutation prior to the acquisition of one of the classical driver mutations or may occur subsequently to it [108, 109]. This bi-directional co-occurrence suggests more than a simple random coincidence in the acquisition of these mutations.

Acquisition order also appears to be a critical determinant of the resulting disease phenotype. Order of acquisition can be inferred from colony assays or single-cell studies allowing genotyping of the clonal structure. In the case of TET2, it is evident that HSCs with this mutation are transcriptionally altered, driving expansion of the clone within the HSC compartment but with limited excess production of terminally differentiated megakaryocytes or erythrocytes until a second hit with the JAK2 V617F mutation occurs. In contrast, JAK2 V617F first cells behave in precisely the opposite manner [108]. Consistent with these findings, loss of TET2 has been observed to promote clonal expansion and self-renewal of HSCs in murine models [110, 111]. TET2-first individuals tend towards an ET phenotype with higher numbers of single mutant cells in the HSC compartment. In contrast, TET2-second individuals show a predominance of double mutant cells in the HSC compartment with a tendency to PV [108]. Figure 4 demonstrates this pictorially. Similarly, loss of function DNMT3A mutations appears to enhance the self-renewal of HSCs and DNMT3A null HSCs are serially transplantable to a significantly enhanced degree in comparison to normal HSCs [112, 113]. Loss of function of another epigenetic regulator EZH2 has also been observed to enhance self-renewal of the JAK2 V617F positive HSC in a murine model [114]. JAK2 V617F only clones in contrast do not promote an MPN disease phenotype on serial transplantation [111]. Early observations that JAK2 V617F positive individuals could transform to JAK2 V671F negative blast phase disease appear to highlight the ability of pre-JAK2 clone with self-renewing capacity in the stem compartment to later establish dominance over the JAK2 clone [115]. Recent advances have moved away from a traditional model of haematopoiesis in which precursors progress through a series of discrete intermediate stages with decreasing differential potential at each stage to a continuum where the boundaries between stem and progenitor cells are increasingly blurred [116]. The epigenetic regulation of transcriptional control affected by loss of normal TET2 and DNMT3A may allow HSC to access alternative transcriptional programmes, shift along this continuum and promote self-renewal. It is interesting that in CHIP individuals, TET2 and DNMT3A clonal fractions are approximately 25% and 14% smaller than ASXL1 clonal fractions suggesting a differential ability to promote clonal expansion [117]. In any instance, it is clear that the field in which the JAK2 mutation is sown is very different in the context of an earlier mutation. Whether these TET2 or DNMT3A mutant first cells are primed for JAK2 mutagenesis or the subsequent frequency of JAK2 V617F mutations simply reflects random chance and high turnover is not clear at this stage.

Further cellular context is provided by the cell of origin and ultimate differentiation bias of this cell. Detection of driver mutations in multiple different mature cell compartments including erythroid, megakaryocyte, granulocyte, monocyte and in some cases lymphoid cells provide evidence of an initiating cell with multipotency implicating the HSC population [118]. In keeping with this, JAK2 and CALR mutants are detectable within CD34+, CD38- HSC and CD34+, CD38+ progenitor populations [119]. A murine model incorporating a humanised ossicle niche has demonstrated engraftment of myelofibrosis from the CD34+, CD38− HSC population only, with no engraftment in the CD34+, CD38+ progenitor population, suggesting that the initiating cell resides within this HSC population [120]. A knock-in JAK2 V617F model has also demonstrated the critical role of the HSC population in initiation and maintenance of the polycythaemia phenotype [121]. Therefore, involvement of the multi-potent HSC population appears to be important.

Yet, multipotency of these cells in experimental conditions may not reflect the true balances and equilibriums at play in a complex multi-factorial environment such as the bone marrow. In fact, some of these cells appear to have an in-built differentiation bias from a very early position in the haematopoietic hierarchy. In mouse models, platelet biased HSC populations have been identified at the apex of the hierarchy, with enhanced capability of short- and stable long-term platelet production without loss of self-renewal ability and retain the ability in a proportionally limited manner give rise to lymphoid biased HSCs [122]. It is not clear if HSC exist with other lineage bias [123]. Application of single-cell technologies has demonstrated this megakaryocyte differentiation bias in HSCs in myelofibrosis patients with the majority of megakaryocyte progenitors transcriptionally distinct from normal, with proliferative and fibrosis enhancing gene signatures evident [124]. Similarly, single-cell studies in JAK2 V617F positive ET patients have demonstrated an expanded population of megakaryocyte primed HSCs with increased sensitivity to interferon alpha signalling [125]. These findings point towards an inherent bias in phenotype, at least towards ET or PMF, depending on the bias on the initiating stem cell acquiring the first mutation.

The interaction between the malignant clone and stromal cells by means of pro-inflammatory and cytokine signalling is a key determinant in the formation of a fibrotic phenotype. Myofibroblasts responsible for the deposition of collagen are derived from multipotent mesenchymal progenitor stromal cells [126‐128]. These myofibroblasts evolve over time, undergoing a maldifferentiation process to lose ability to support haematopoietic tissue and contribute to marrow fibrosis. The presence of the malignant haematopoietic clone and resulting inflammatory milieu provide a continuous drive for myofibroblast differentiation and vicious cycle of fibrosis [128]. Removing Gli1⁺ mesenchymal stromal cells or interfering with PDGFRA signalling in these cells can ameliorate the fibrotic phenotype [126, 127].