Introduction

Multiple prospective clinical trials cumulatively involving more than 2000 patients with chronic myeloid leukemia (CML) have examined the feasibility and safety of treatment-free remission (TFR) after withdrawal of tyrosine kinase inhibitor (TKI) treatment following achievement of a sustained deep molecular response (DMR), variably defined as a molecular response (MR) of BCR-ABL1 ≤ 0.01% (MR4.0) or better on the International Scale (IS) [1,2,3,4,5]. Despite slightly differing inclusion criteria and definitions of molecular relapse being used these trials have shown remarkably consistent results, with TFR rates of around 40–60% at 2 years, and the majority of instances of molecular relapse occurring in the first 6 months after TKI withdrawal. There is considerable interest among patients and clinicians in the potential to stop TKI treatment [6]. The recent CML treatment recommendations of the US National Comprehensive Cancer Network (available at www.nccn.org) and the European Society of Medical Oncology [7] for the first time included criteria for a TFR attempt outside the clinical trial setting. With increasing numbers of patients attempting TFR it is important to have longer-term data on the safety and durability of this approach.

The Australasian Leukaemia & Lymphoma Group (ALLG) CML8 study (TWISTER; ACTRN 12606000118505) commenced accrual in July 2006 and enrolled 40 imatinib-treated patients with undetectable measurable residual disease (UMRD; approximately equivalent to MR4.5, or BCR-ABL1 ≤ 0.0032%) by real-time quantitative reverse transcriptase-polymerase chain reaction (RQ-PCR). We previously reported a TFR rate of 42.7% with a median follow-up of 3.6 years [1]. The French STIM study commenced at around the same time and had very similar inclusion criteria [2]. The long-term outcome of the 100 patients in the STIM study was recently updated with a median follow-up of 6.4 years and reported 38% of patients alive and in TFR [8]. These two studies offer us the longest available follow-up of patients in TFR.

In the TWISTER study we previously demonstrated the presence of persistent BCR-ABL1-positive cells, detectable by highly sensitive individualized genomic DNA PCR (sensitivity MR6.2), even in patients with sustained TFR for up to 5 years [1, 9]. The intronic BCR-ABL1 fusion sequence is essentially unique to each individual patient [10], effectively eliminating the risk of false positive results, and this enables us to achieve a level of sensitivity and specificity which is significantly greater than that of conventional RQ-PCR [9]. In the present study we extend the period of follow-up to demonstrate the long-term durability of TFR and the safety of TKI discontinuation in selected patients. In addition we show evidence for a gradual reduction in the level of circulating BCR-ABL1-positive cells during TFR.

Materials and methods

Treatment and monitoring by RQ-PCR

The inclusion criteria, monitoring, and re-treatment schedules for the TWISTER study were described previously [1]. Eligible patients had sustained UMRD for at least 2 years on imatinib treatment. In subsequent years there has been a move away from UMRD to define absolute levels of BCR-ABL1, such as MR4.5 (BCR-ABL1 ≤ 0.0032%). For comparison, the equivalent median sensitivity of RQ-PCR on the day of stopping imatinib was MR4.8 (range 4.0–5.3). RQ-PCR monitoring was performed in a central laboratory (SA Pathology, Adelaide) [11, 12] monthly for the first 12 months after discontinuation, every 2 months in the second year, then every 3 months thereafter. The protocol-defined trigger to restart imatinib was two consecutive positive RQ-PCR results at any level, without requiring either an increase in BCR-ABL1 or loss of a major molecular response (MMR; BCR-ABL1IS ≤ 0.1%). In patients who restarted imatinib there was 1 year of central monitoring during re-treatment [1]. The protocol was subsequently amended to add a further 4 years of follow-up following completion of the monitoring or re-treatment phase during which the patients were monitored in their respective local laboratories, all of which report BCR-ABL1IS [13]. At last follow-up the only patients being monitored in the central laboratory were those patients treated in Adelaide. We circulated a data collection form to investigators to update survival and disease status (including molecular response and CML treatment). The study was approved by the ethics committees of the participating hospitals. All samples were collected with informed consent in accordance with the Institutional Ethics-approved protocols and with reference to the Declaration of Helsinki.

DNA Q-PCR

The BCR-ABL1 genomic breakpoint sequence was identified in 26 patients (13 who had relapsed and 13 in stable TFR) for whom we had diagnostic material available. Genomic DNA was extracted from peripheral blood leukocytes and BCR-ABL1 DNA PCR was performed as described previously [9, 14]. Serial testing was performed during the protocol-defined follow-up period, and at later time points in 20 Adelaide patients (9 patients in long-term TFR and 11 patients who restarted imatinib treatment). Briefly, the amount of amplifiable DNA in a sample was determined by interpolation from a standard curve for the GUSB control gene. Nested PCR was performed with 500 ng amplifiable genomic DNA in each first round PCR and real-time Q-PCR in the second round to give a semiquantitative value. Typically, 20 replicates were performed to give a limit of detection of MR6.2 (one in ~2 million cells). At very low levels of measurable residual disease (MRD) the assay is digital: each PCR replicate contains either one target molecule or none. When all replicates were positive, a standard curve (comprising serial dilution of the patient’s diagnostic DNA) was used to estimate copy number [9, 14], with the modification that Poisson statistics were used to exclude replicates that gave outlying values [15]. This modification was necessary due to the potential for minor differences in PCR efficiency in a nested assay to produce widely varying threshold cycle numbers in the second round real-time Q-PCR.

Statistical analysis

Statistical tests were performed using GraphPad Prism 7 statistical software (GraphPad Prism Inc., La Jolla, CA, USA). R version 3.4.1 (R Foundation for Statistical Computing, Vienna, Austria) was used to estimate the cumulative incidence of molecular response by the Kaplan−Meier method and to prepare loess curves (locally weighted scatterplot smoothing, an adaptation of the least squares method for averaging serial data) to fit the BCR-ABL1 DNA data over time. A P value of <0.05 was considered statistically significant.

Results

Patient characteristics

Forty patients were enrolled and their characteristics at study entry were previously described [1]. Key features are summarized in Table 1. At the time of last follow-up the median age of the patients was 69 years (range 38–91 years).

Table 1 Patient characteristics at study entry
Full size table

Overall and molecular relapse-free survival

Of the 40 patients enrolled, 37 remained alive at last contact, 18 of whom were in continuous TFR with a median follow-up of 8.6 years (range 5.7–11.2 years). No patient progressed to accelerated phase or blast crisis, and all three deaths were considered unrelated to CML (one each from plasma cell myeloma, cardiac failure, unknown cause with UMRD on the last available test). The three deaths were all among patients who had recommenced imatinib treatment and therefore had no effect on molecular relapse-free survival, which was 45% at 8 years (95% confidence interval 31.9−63.4%) (Fig. 1). No patient recommenced TKI treatment without prior molecular relapse. The latest molecular relapse was at 27 months, as previously reported [1].

Fig. 1
figure 1

Molecular relapse-free survival (MRFS). Kaplan−Meier estimate of the proportion of patients remaining alive with stable UMRD (95% confidence interval indicated by dashed lines)

Full size image

BCR-ABL1 DNA PCR during stable TFR

DNA Q-PCR was performed in nine Adelaide patients who remained in stable TFR for a median of 8.6 years (range 5.7–10.8 years). In two of the patients (#4, 5) BCR-ABL1 DNA became undetectable in the last 4 years whereas, remarkably, two of the patients (#8, 9) had detectable BCR-ABL1 DNA at every measurement over 6.5 and 10.5 years, respectively (Fig. 2 and Supplementary Figure 1). In order to test whether there was a reduction in BCR-ABL1 DNA over time we plotted a loess curve using the quantitative values from all nine patients. This showed a continuous rate of decline during the first few years of TFR with an inflection at 3 years, beyond which the data were too sparse to determine whether there was any further decline (Fig. 3a). In the six patients who had quantifiable BCR-ABL1 DNA in more than 50% of samples during the first year of TFR, the median level of BCR-ABL1 reduced significantly from MR5.0 during the first year after imatinib discontinuation to MR6.1 in the sixth year of follow-up (P = 0.03; Fig. 3b).

Fig. 2
figure 2

BCR-ABL1 DNA monitoring in nine patients in long-term TFR. BCR-ABL1 DNA values for every patient are indicated by dots. The black dots  represent positive values, and the white dots represent undetectable values

Full size image
Fig. 3
figure 3

BCR-ABL1 DNA values over time. a, b Values of patients in long-term TFR and c, d for patients re-treated with imatinib. a, c The graphs represent the loess analysis. The BCR-ABL1 DNA values for every patient are indicated by dot-dashed lines. The blue solid line represents a fitted loess line (locally weighted scatterplot smoothing). The gray zone represents the 95% confidence interval of the loess line. For illustrative purposes the undetectable values are assigned a value of MR6.2 corresponding to the limit of detection of the method. The total number of patients (red) and samples (black) are shown for each year. b, d Decline of BCR-ABL1 DNA. The blue dots represent the median of DNA BCR-ABL1 values per year per patient. Median with 95% CI are represented. Wilcoxon matched-paired signed rank test performed. IM imatinib

Full size image

Response to imatinib re-treatment

Twenty-two patients met the study criteria for re-treatment with imatinib, and all 22 regained UMRD after a median of 3.3 months of re-treatment (range 0.0–17.4 months). One patient lost UMRD, but remained in MR4.5 (BCR-ABL1 0.002%); this patient had spontaneously returned to UMRD at the time of restarting imatinib. No patient has developed imatinib resistance. Two patients switched to second-generation TKIs after regaining UMRD on imatinib: one to dasatinib because of chronic imatinib toxicities (periorbital edema requiring blepharoplasty, and nausea), and one to nilotinib with the aim of achieving a deeper molecular response to enable a subsequent TFR attempt.

BCR-ABL1 DNA Q-PCR during imatinib re-treatment

DNA Q-PCR was performed in 11 Adelaide patients who relapsed and recommenced imatinib treatment. We previously performed DNA PCR in these patients after regaining UMRD (median 3.2 months; range 1.1–9.9 months) and showed that the level of BCR-ABL1 DNA returns to a level similar to that prior to imatinib discontinuation [1]. In this follow-up analysis we analysed the BCR-ABL1 DNA level over time with a median of 6.2 years (range 1.3–9 years) of follow-up during re-treatment (Fig. 3c) Again we performed a loess analysis, which appeared to show three distinct phases: a rapid decline in BCR-ABL1 DNA during the first 2 years of re-treatment, followed by the same biphasic pattern that we described above for the patients in continuous TFR (Fig. 3c). In nine patients there were sufficient numbers of samples after the first year to compare the values of BCR-ABL1 DNA. The median level of BCR-ABL1 DNA reduced significantly from MR4.7 during the first year of re-treatment to MR5.9 in the fifth year of re-treatment (P = 0.02; Fig. 3d).

Second and subsequent TFR attempts

Twelve of the 22 patients who regained UMRD later attempted TKI discontinuation for a second time (TFR2). In contrast to the primary study definition of molecular relapse [1], the trigger for re-treatment after TFR2 was at the discretion of the treating physician. For these 12 patients the median time to regain UMRD after restarting imatinib after the first TFR attempt was similar to the overall cohort (median 3.2 months; range 0.9–17.4 months), and the median duration of TKI re-treatment prior to TFR2 was 5.9 years (range 1.7–9.7 years). Three of these patients attempted TFR2 in a clinical trial (ACTRN12615001169538, a Phase Ib study of lenalidomide in combination with imatinib for adult patients with CML in second molecular remission). The clinical outcome of the 12 patients is summarized in Table 2. Six patients restarted TKI treatment: four after loss of MMR (restarted imatinib at 1, 3, 8, and 72 months, respectively); and two after loss of MR4.5, remaining in MMR (restarted TKI at 2 months (imatinib) and 7 months (nilotinib)). The latter patient (#30) subsequently stopped nilotinib 32 months later, and remained in TFR3 with UMRD at last follow-up after 16 months off treatment. One patient (#25) died from an unknown cause after 13 months in TFR2 with no detectable BCR-ABL1 transcripts. The remaining five patients remained alive and in MMR (n = 1) or stable MR4.5 (n = 4) after a median of 17 months (range 12–44 months) following TKI discontinuation.

Table 2 Duration of re-treatment and outcome of second TKI discontinuation attempt
Full size table

Discussion

In this updated report of the TWISTER study we confirm the stability of TFR with a median follow-up in excess of 8 years. No molecular relapse has been observed beyond 27 months after imatinib discontinuation, consistent with the STIM study in which the latest relapse was reported at 22 months [8]. Notably, one patient stopping imatinib for a second time lost MMR after 6 years. Extremely late relapses have been reported even after allogeneic stem cell transplantation [16, 17]. At present, there are very few patients with follow-up in TFR for more than a decade. Therefore, until we have data from much larger numbers of patients followed in very long-term TFR we continue to recommend RQ-PCR monitoring indefinitely at least twice per year. We also provide long-term follow-up on the stability of re-treatment with imatinib: no patient developed resistance and the commonest cause for discontinuation was a second TFR attempt.

We used highly sensitive BCR-ABL1 DNA PCR to monitor the depth of response in serial samples taken during stable TFR. The number of patients in whom this analysis could be performed was limited by sample availability. The precision of any MRD assay operating near the limit of detection will be determined primarily by sampling error, which can be modeled using the Poisson distribution [18, 19]. Nevertheless, by pooling the data from the evaluable patients we were able to demonstrate a clear reduction in the proportion of BCR-ABL1-positive cells circulating in the peripheral blood. The reduction of MRD during the time in TFR could be explained by two hypotheses: gradual extinction of long-lived lineage-committed cells that lack self-renewal capacity or depletion of slowly proliferating CML precursor cells. Multiple studies have shown an association between immunological cell subsets and successful TFR [20,21,22,23,24], and immunological depletion of residual leukemic cells could occur even in the absence of TKI treatment. However, stochastic effects at play in the switch between self-renewal and proliferation may also lead to the random extinction of leukemic stem cells [25], so that no active process need be postulated to explain this observation.

Around half of TFR attempts result in re-treatment with a TKI and almost all of these patients rapidly regain a DMR [4, 5]. After UMRD is restored conventional RNA-based RQ-PCR cannot be used to determine whether there is further response in the CML clone. Using highly sensitive patient-specific BCR-ABL1 DNA PCR we were able to show that there is a gradual reduction in the CML clone with prolonged re-treatment, even though UMRD is regained typically within a few months. The triphasic response that we observed in patients receiving re-treatment is consistent with mathematical models describing a steep first phase of response reflecting the depletion of differentiated leukemic cells [26,27,28]. This burst of proliferation at molecular relapse must arise from a residual population of CML progenitor cells, and the slower second phase of reduction in BCR-ABL1 could correspond to depletion of this progenitor population by continuing TKI treatment [26,27,28].

The number of patients in a second DMR after an unsuccessful attempt at TFR is increasing over time, and the question of how to manage patients who may wish to have a second TFR attempt is gaining importance. The French CML collaborative group recently reported on 70 patients who recommenced imatinib after an unsuccessful TFR attempt, and who later stopped TKI treatment for a second time after regaining a DMR (RE-STIM) [29]. The estimated proportion of patients remaining in stable MMR without treatment was 42% at 2 years. In our own cohort of 12 patients attempting TFR2 the proportion of patients remaining in TFR2 was 60% (if two patients restarting TKI without loss of MMR are excluded). Considering the small number of patients (three of whom were in a Phase 1 study that might have influenced the outcome of the TFR attempt) these proportions are broadly similar. Several factors might contribute to the change in status of these patients from relapse at TFR1 to remission at TFR2. Firstly, the A-STIM study demonstrated that when the trigger for re-treatment is MMR the proportion of patients remaining in TFR rises from around 40% to around 60% [3]. This suggests that perhaps 20% of patients who were re-treated in TWISTER (and STIM) might have sustained MMR at the first TFR attempt, had the re-treatment trigger been less strict (noting that one of our patients originally “relapsed” without loss of MR4.5). Differing criteria for TKI re-treatment will influence the reported rate of TFR. Secondly, the total duration of TKI treatment in patients who achieved TFR at the second cessation attempt was very long at a median of 12.7 years (range 11.3−14.8 years). An interim analysis of the EuroSKI study reported that each additional year of DMR on imatinib prior to discontinuation was associated with approximately 2–3% absolute increase in the probability of TFR at 6 months after stopping [30]. Our results showing reducing levels of BCR-ABL1 DNA during prolonged re-treatment could potentially help explain why a longer duration of TKI treatment results in a higher proportion of patients in stable TFR. The optimal duration of TKI treatment for an individual patient cannot be determined at present, and there is a need for reliable biomarkers to predict TFR outcome. Prospective clinical trials with consistent inclusion criteria for patients attempting TFR2 are needed to help identify factors that may predict the outcome of a second discontinuation attempt.

Including the three patients who entered a phase 1 study of TFR2, a total of 24 out of 40 patients in the TWISTER study (60%) have eventually achieved TFR. No patient developed hematological relapse, TKI resistance, or progression of CML. Unnecessarily prolonged TKI treatment may expose the patient to rare late-emerging TKI toxicities, impairment of quality of life, and increased financial burden. Our results confirm the long-term safety and durability of TFR after imatinib and show, for the first time, evidence that MRD continues to fall in these patients despite the absence of ongoing therapy.