Pankit J. Vachhani, MD, Assistant Professor of Medicine at the O’Neal Comprehensive Cancer Center at the University of Alabama, and Srdan Verstovsek, MD, United Energy Resources, Inc., Professor of Medicine and Hematologist-Oncologist at MD Anderson Cancer Center, debate the usefulness of variant allele frequency (VAF) as a prognostic biomarker for myeloproliferative neoplasms (MPNs).
VAF Can Predict Outcomes in MPNs
By Dr. Vachhani
We should start by discussing two points to set the stage. One, VAF, otherwise referred to as allele burden, informs us on the fraction of sequencing reads harboring the variant/mutation. A single VAF number does not capture the heterogeneity of mutations that may exist in a homozygous versus heterozygous situation and the different frequencies within primary clone versus subclones. Further, the exact cell population (malignant cells vs bystander normal tissue) that is being measured may affect the results. This has traditionally been performed using polymerase chain reaction (PCR; quantitative PCR or droplet digital PCR) and, what is most often used in a clinical setting, next-generation sequencing (NGS). Testing for VAF from blood or marrow should yield comparable results. However, data show that VAF from blood or marrow is a composite measure of alleles from a heterogenous and variable mixture of mature leukocytes and therefore does not report information from the critical MPN hematopoietic stem cell and progenitor cell population.
Two, prognostic biomarkers inform a likely outcome when present independent of the treatment received. Prognostic biomarkers can be positive or negative. VAF—as a marker of disease burden—is indeed a biomarker, but we must ask the follow-up question, “What exactly is the biomarker prognostic for?” In the context of MPNs, the prognostic risk could be for thrombosis, leukemic transformation, fibrotic transformation, abnormal blood counts or spleen size, and risk of death, among others.
As we know, MPNs are marked by the presence of a driver gene mutation—JAK2, MPL, or CALR, in the vast majority of cases. In polycythemia vera (PV), nearly 98% of patients have JAK2 mutations (most of which are the classic JAK2 V617F mutation). On the other hand, in essential thrombocythemia (ET) and myelofibrosis (MF), JAK2 V617F is found in 50% to 60% of cases, while MPL and CALR mutations are found in 5% to 10% and 15% to 30% of cases, respectively. About 10% to 15% of ET and MF cases are what we call “triple-negative,” referring to the absence of one of the three classic driver gene mutations.
In the field of MPNs, VAF as a biomarker has been explored mostly in the context of JAK2 mutation. Few reasons have contributed to this situation: (1) CALR and MPL mutations are found in a minority, albeit substantial (when referring to ET and MF) amount of patients; (2) unlike JAK2 mutations (barring the rare exon 12 mutations in PV), CALR has more than 50 mutations, although a 52 base pair deletion (type 1) or a five base pair insertion (type 2) account for more than 80% of mutations; (3) the different baseline VAFs of the least present MPL W515L and W515K suggests a functional difference between the two variants; and (4) the CALR clone appears to have a clonal advantage compared to the JAK2 clone, which precludes the use of the same VAF thresholds.
Some of the earliest suggestions of the impact of JAK2 VAF came from our attempts to understand the nature of MPNs. Two major things that play a role in the phenotypic manifestation of MPNs (especially PV and ET) are the order of gene mutation acquisition and the JAK2 gene dosage. For example, we now know that JAK2 V617F homozygosity is found in about 25% to 30% of PV patients but only 2% to 4% of ET patients. Unlike PV, in ET, the homozygous state (roughly denoted by VAF >50%) has a higher thrombotic risk compared to the heterozygous state (hazard ratio [HR]=3.97) than wild-type (HR=1.0) or heterozygous patients (HR=1.49). Additionally, the JAK2 V617F VAF levels are higher in PV than ET. Studies have shown that JAK2 V617F VAF >50% makes ET unlikely, and if it is PV, there’s a high chance of progression to post-PV MF. Mayo Clinic investigators showed in a study of 176 ET patients that JAK2 V617F VAF correlated directly and significantly with platelet and leukocyte counts, palpable splenomegaly at diagnosis, and venous thrombosis occurring after diagnosis.
In another study performed in Italy, the investigators sought to correlate JAK2 V617F VAF with major clinical outcomes in patients with PV. From a set of 173 patients, the investigators determined that VAF correlated with measurements of erythropoiesis, myelopoiesis, and neutrophil activation. Patients with VAF >75%, compared to those with <25%, had a greater risk of enlarged spleen and pruritus. In these patients, the risk of requiring chemotherapy or developing major cardiovascular events was significantly higher (relative risk=7.1).
In a separate prospective Italian study, investigators enrolled 338 patients with PV, of whom 320 (94.7%) carried the JAK2 V617F mutation. All eight patients who progressed to post-PV MF had VAF >50%, and this association retained significant in multivariable analysis. By contrast, the risk of developing acute myeloid leukemia (AML) or thrombosis was not significantly related to mutant allele burden. The increased risk of fibrotic transformation was previously noted by other studies, too: In PV, the fibrotic transformation was significantly higher among homozygous than heterozygous groups (23% vs 2% in the Mayo Clinic study and 11.5% vs 1.4% in an Italian study). In ET, the fibrotic transformation was significantly higher among homozygous than heterozygous ET patients (14.3% vs 4.7% in an Italian study).
A third Italian study showed that five years post-diagnosis, JAK2 allele burden became the only independent factor for predicting subsequent thrombotic events in patients with PV or ET. In a more recent study, JAK2 V617F VAF >50% (HR=2.4) and previous venous thrombosis (VT; HR=2.8) were confirmed as independent predictors of future VT in patients with PV. The impact of JAK2 V617F VAF >50% on VT was particularly significant in conventional low-risk patients. All together, we have data to show clear associations of VAF in PV-related outcomes, although the very specific risk of thrombosis seems to be unclear, given some contrasting results across different studies.
Also worth mentioning is a study from a Danish general population that showed that in individuals with JAK2 VAF ≥1%, compared to <1%, the odds ratio (OR) for venous thromboembolism was 2.8 for JAK2 V617F-positive non-MPNs and 6.0 (95% CI, 1.3-28; P=.04) for mutation-positive MPNs compared with non-mutated individuals. Effectively, this shows that JAK2 VAF, even at really low levels, can have an impact on thrombotic risk and that there is an effect of mutant JAK2 dosage on the thrombotic risk.
Even in MF, we have evidence supporting the prognostic role of VAF. The investigators from Mayo Clinic assessed 129 patients with primary MF with allele burden information (irrespective of JAK2 mutation status) and divided them into quartiles based on JAK2 V617F allele burden. There was shortened overall and leukemia-free survival for the lower quartile but not for the upper quartile allele burden group, which was validated with multivariable analysis. In a similarly conducted study from Italian investigators, 127 JAK2-mutated patients were divided into quartiles. Patients in the lower quartile had shorter time to anemia and leukopenia and did not progress to large splenomegaly. Furthermore, survival was significantly reduced in the lower quartile compared to upper quartiles and JAK2 wild-type patients. In multivariate analysis, factors associated with reduced survival were age, blast count >1%, and JAK2 V617F burden within first quartile. It is no surprise then that when MD Anderson Cancer Center MPN experts devised a simple prognostic model in MF, a 50% cutoff dichotomized JAK2-mutated patients into those with high VAF and favorable survival and those with low VAF and unfavorable survival. Ultimately, their risk model had only two main factors: age and driver mutation information along with VAF for JAK2.
Beyond prognostic value, JAK2 VAF also has predictive value. We now have data from a study of 381 patients treated with interferon for MPNs (340 of whom had PV or ET) where the results suggest that interferon could be discontinued in those who achieved complete hematologic response, particularly in patients with a driver VAF <10% at time of discontinuation. In MF, the notable impact of ruxolitinib dosing is well known. In addition, a study showed that the probability of spleen response to ruxolitinib in subjects with a JAK2 V617F allele burden ≥50% was 5.5-fold higher than in subjects with JAK2 V617F allele burden <50% or any other mutation (OR=5.51). In contrast, data from the PERSIST trials showed that pacritinib leads to spleen volume reduction in MF patients, irrespective of JAK2 V617F VAF.
The next wave of research will hopefully help clarify the various VAF thresholds from both prognostic and predictive standpoints. It is likely that these will be different based on the gene of interest, specific mutation, MPN, therapeutic context, and the outcome of interest. Additional data on the VAF measurements of CALR/MPL genes and non-driver genes will be important. For example, we have some data suggesting that rising TP53 mutation VAF may predict AML transformation. Finally, we may even have prognostic models that account for the interaction of VAF thresholds with other mutations and well-known risk factors in MPNs.
Currently, This Is Not a Useful Biomarker
By Dr. Verstovsek
Molecular markers are known to contribute to the disease existence, persistence, and progression in MPNs. The presence of JAK2, CALR, or MPL mutations is seen in about 90% of patients, and testing for them is part of diagnostic process—everybody tests for them. The presence of additional mutations, or non-driver mutations that do not activate the JAK/STAT pathway, which can be tested through NGS in larger academic centers, may contribute to the complexity and aggressiveness of the disease. ASXL1, EZH2, and IDH mutations, and others, are part of the prognostic scoring systems. We certainly have a good understanding of the need for identification of certain mutations, either for diagnostic or prognostic purposes, and that is becoming a standard practice.
However, when we talk about VAF, or allele burden, which is the percentage of cells in the patient sample that are affected by whatever mutation you are testing for, whether we have good evidence that allele burden of a JAK2 mutation, for example, makes any difference for the patients, that is hard to say, so I would say no. In addition, whether a modification (e.g., by therapy) of the percentage of cells affected by a given mutation makes any difference for the patient’s outcome is not well understood, so I would say no.
Let’s say that we have, at the time of diagnosis, a genetic test done that can identify the number of cells affected by the JAK2, ASXL1, or EZH2 mutations. Would that give me useful information for overall outcomes for the patients? Would I be able to say that a patient who has 25% of cells affected by ASXL1 mutation, versus the other one that may have 75% of cells affected by ASXL1 mutation, would have a different outcome? No, we don’t have that information at all.
How about during therapy, should we say that decreased allele burden is a clinically valuable result? I argue no. I am not aware of any convincing evidence that the modification of even a JAK2 mutation allele burden makes an overall difference for the outcome of the patient. Perhaps there are some signs that this is a possibility, particularly when we talk about interferon for patients with PV. It has been published by several groups that long-term interferon therapy can, in about 14% to 15% of patients after five years of therapy, completely eliminate the JAK2 mutation—meaning it’s not detectable by the standard PCR test, which would be in the realm of 0.1 sensitivity.
That is something that is completely different than just saying certain levels of the allele burden at the start or during therapy make an important difference for overall outcomes. Perhaps there is a possibility that elimination of detectable mutation makes a difference for patients. On the other hand, we already know that, particularly in patients with MF, there might be different mutations present to different degrees affecting different types of the cells. There might be presence of different clones of the cells in the bone marrow, which is in fact known to exist in MF. Affecting the one cell type—let’s say dropping the JAK2 mutation allele burden from 50 to five—may allow the other clone to be proliferating higher and perhaps affect the overall outcome of a patient in a negative way, to be pessimistic.
This is not evidence-based, I’m just bringing this to the discussion. Do we understand the complexity of the biology in the MPNs? Do we understand the competition between the clones that have a different mutation, particularly in MF? We don’t. We are not ready at this moment to allow allele burden to become a prognostic factor or to signify during therapy that, if modified, there would be any different outcomes for the patient.
I hesitate to say that a certain degree of a decrease of any of these mutations with any of the new therapies that are being developed would be particularly meaningful. I am not aware of a good connection between the allele burden modification of any of the mutations and the different clinical outcome. This is all exploratory, and it’s not ready for prime time.
This topic is debate-worthy because in the closely related MPNs, such as chronic myeloid leukemia (CML), if you have a therapy that can eliminate the clone and have a molecular response that is at the level of 0.001 detectability (extremely low), then that might mean functional cure. There are evidence-based data or experiences in the closely related conditions that molecular response may mean a lot, but we are not at that level with the Philadelphia chromosome (Ph)-negative MPNs. We are hopeful that we can aim for molecular response to become clinically relevant in Ph-negative MPNs, but the connection between the molecular response and clinical relevance does not currently exist.
There is a lot of evidence that with new medications in development for MF, you can modify the levels of fibers in a bone marrow. You can modify the levels of cytokines. You can modify in samples from the patients that are on these investigational agents, allele burden of different mutations, and many call that disease modification, despite the fact that there is no real connection between these biological alterations in those measurements and the clinical outcome of the patients. There are many different studies that are underway in MF where there is evidence of biological modification, but none showed a connection between modifications in biological parameters, like allele burden, and clinical outcomes. That’s why I say we are not ready yet to assign a clinical relevance to allele burden modifications. That’s why my position is no.
However, the field is evolving rapidly, and I hope that some of these new medications prove me wrong and that in a year or maybe three to five years, I will be on the other side, and I’ll argue, “yes.”