Lack of survival advantage with autologous stem-cell transplantation in high-risk neuroblastoma consolidated by anti-GD2 immunotherapy and isotretinoin

Since 2003, high-risk neuroblastoma (HR-NB) patients at our center received anti-GD2 antibody 3F8/GM-CSF + isotretinoin – but not myeloablative therapy with autologous stem-cell transplantation (ASCT). Post-ASCT patients referred from elsewhere also received 3F8/GM-CSF + isotretinoin. We therefore accrued a study population of two groups treated during the same period and whose consolidative therapy, aside from ASCT, was identical. We analyzed patients enrolled in 1st complete/very good partial remission (CR/VGPR). Their event-free survival (EFS) and overall survival (OS) were calculated from study entry. Large study size allowed robust statistical analyses of key prognosticators including MYCN amplification, minimal residual disease (MRD), FCGR2A polymorphisms, and killer immunoglobulin-like receptor genotypes of natural killer cells. The 170 study patients included 60 enrolled following ASCT and 110 following conventional chemotherapy. The two cohorts had similar clinical and biological features. Five-year rates for ASCT and non-ASCT patients were, respectively: EFS 65% vs. 51% (p = .128), and OS 76% vs. 75% (p = .975). In multivariate analysis, ASCT was not prognostic and only MRD-negativity after two cycles of 3F8/GM-CSF correlated with significantly improved EFS and OS. Although a trend towards better EFS is seen with ASCT, OS is near identical. Cure rates may be similar, as close surveillance detects localized relapse and effective salvage treatments are applied. ASCT may not be needed to improve outcome when anti-GD2 immunotherapy is used for consolidation after dose-intensive conventional chemotherapy.

Our group adopted ASCT consolidation of HR-NB in the 1980s based on the concept that high-dose alkylators can overcome chemoresistance [10,11]. Subsequently, after promising results with high-dose cyclophosphamide in induction [12], as well as with 3F8 in phase I and II trials [13,14], we undertook a non-ASCT study using dose-intensive induction followed by consolidation with
For EFS, the only two events other than PD were in ASCT patients: 1) acute leukemia 12 months from NB diagnosis and 4.5 months from study entry (NB subsequently relapsed); and 2) death from pulmonary fibrosis 78 months from NB diagnosis and 70 months from study entry.

Univariate and multivariate analyses of prognostic factors
In univariate analyses (Table 2), ASCT was not prognostic for EFS (hazard radio [HR] = 0.68, p = 0.13) or OS (p = 0.975). Post-MRD negativity was significantly associated with better EFS and OS ( Figure 2). Longer time from 1st chemotherapy to 3F8 and longer time from ASCT or last chemotherapy to 3F8 were significant for better EFS (p = 0.012 and p = 0.022, respectively). HAMA as a time-dependent variable was not significant for EFS (p = 0.564) but marginally significant for OS (p = 0.058).
Since ASCT patients had longer time from 1st chemotherapy to 3F8 (Table 1), we undertook subset analyses. Among ASCT patients, time from 1st chemotherapy to 3F8 and time from ASCT to 3F8 were not significant for EFS (HR = 0.97 per month, p = 0.78 and HR = 0.97 per month, p = 0.86 respectively). In contrast, among non-ASCT patients, these variables were significant (HR = 0.91, p = 0.027, and HR = 0.49, p = 0.0146, respectively). When patients were split by the median time (7.75 months) from 1st chemotherapy to 3F8, ASCT was not significant for EFS (p = 0.44 for < 7.75 months group, p = 0.98 for ≥ 7.75 group). When patients were stratified by the median time (1.55 months) from ASCT or last chemotherapy to 3F8, ASCT was not significant (p = 0.134 for < 1.55 months group, and p = 0.61 for ≥ 1.55 months group). Based on these subset analyses, we concluded that univariate effect of ASCT on EFS (Table 2) was most likely a result of association between EFS and longer time from 1st chemotherapy or longer time from ASCT.
In the final multivariate analysis (Table 3), there was no significant impact of ASCT on EFS (p = 0.098) or OS (p = 0.655). As in the univariate analyses, post-MRD negativity correlated with significantly better EFS and OS (p < 0.001). Since time from 1st chemotherapy and time from ASCT or last chemotherapy were associated with ASCT, we fitted the multivariate model stratified by these variables dichotomized at median values. The effect of ASCT on EFS was not significant in a multivariate model adjusted for the post-MRD variable (p = 0.16 when stratified by time from 1st chemotherapy, and p = 0.21 when stratified by time from ASCT or last chemotherapy). The same held true for OS (p = 0.85 and p = 0.96, respectively).

Toxicity
As noted in previous reports [8,35], 3F8/GM-CSF had manageable toxicities-hence, treatment was outpatient for the ASCT and non-ASCT groups. 3F8 caused grade 1-2 generalized pain and urticaria. Anaphylactoid reactions to 3F8 occurred on day 1 of cycle 2 in four patients, on day 3 of cycle 2 in one patient, and on day 1 of cycle 3 in one patient. Posterior reversible encephalopathy syndrome without sequelae developed in one patient in cycle 4. Common side-effects of isotretinoin were grade 1-2 dry skin and cheilitis. There were no long-term toxicities linked to protocol therapy.

DISCUSSION
HR-NB patients whose consolidative therapy of 1st CR/VGPR included 3F8/GM-CSF + isotretinoin had similar EFS and near-identical OS whether these biological agents were administered following ASCT or conventional chemotherapy (Figure 1). In the multivariate analysis, ASCT was not significantly prognostic for either EFS or OS (Table 3). With its large cohort of non-ASCT patients, the current study's patient population may well be unique because ASCT has been part of all major studies since 2000 [9]. The experience, therefore, affords an opportunity not otherwise available to reassess whether ASCT (which is not standard for other extracranial solid tumors [36,37]) should be routine for HR-NB. Revisiting this issue is especially timely given that the recent update of the landmark CCG study [17] showed no OS advantage with ASCT [38], and a recent meta-analysis found that ASCT for HR-NB did not improve OS [39].
The ASCT and non-ASCT patients were enrolled on study and treated during the same period by the same Because of an incomplete response to induction, 2nd-line therapy was needed to achieve 1st CR/VGPR. team and underwent the same surveillance (tests, schedule, duration) thereby avoiding biases seen with open-label trials [40]. Large study size (n = 170) allowed robust statistical analyses to assess relevant prognostic factors. An additional noteworthy point is that the two cohorts had similar clinical and biological characteristics, including MYCN, contemporary induction, and pre-MRD-positivity (Table 1). UHR status was significantly less common in the ASCT cohort -a clinical difference that might suggest a better outlook for the ASCT patients, but UHR was not prognostic ( Table 2).
Because of the need to recover from ASCT toxicities, the time from initiation of induction chemotherapy and the time from ASCT or last chemotherapy to 3F8 were significantly longer for the ASCT cohort (Table 1). These differences could be considered selection bias that favored a superior outcome for the ASCT cohort because these patients would have more durable remissions whereas patients with early PD post-ASCT would not be eligible for 3F8/GM-CSF. The timing, however, was not prognostic in the multivariate analysis where the only variable significantly prognostic for both EFS and OS was MRD measured after two cycles of immunotherapy (similar to other studies [8,35]) ( Table 3).
Once viewed as a promising treatment for a range of poor-prognosis pediatric and adult extra-cranial solid tumors, ASCT became standard only for HR-NB [9,36,37]. The randomized CCG study [17], and the pilot tandem program that yielded excellent results [31], used TBI. Because of toxicity concerns, however, TBI was not used in recent national trials [2,3,32]. Assessments of ASCT ought to take into account differences in cytoreduction regimens. In the sole randomized trial of ASCT regimens for HR-NB reported to date [33], busulfan-melphalan was associated with significantly better EFS and OS compared to the widely-used carboplatin-etoposide-melphalan [3,26]. The patients in our study were referred after these and tandem ASCT regimens [3,[31][32][33]. Irrespective of cytoreductive regimen, however, extensive experience over decades shows an uncertain benefit of ASCT for refractory NB [41,42] -and the EFS and OS data in the current report would appear to undermine the rationale for the routine use of ASCT in HR-NB patients in 1st CR/VGPR in the contemporary era.
Another recent finding that raises uncertainty about ASCT for HR-NB is the lack of benefit from ex vivo purging [3]. In the COG randomized study of purging, occult NB contamination of peripheral blood stem cells was rare, suggesting that post-ASCT relapse could be attributed to the failure of the myeloablative regimen (carboplatin-etoposide-melphalan) to ablate the MRD that survived induction.
Advances in therapy since 2000 could account for the lack of survival advantage with ASCT in our experience. Thus, local control of soft tissue NB is excellent with dose-intensive chemotherapy, resection, and RT [16,24,43]; eradication of chemoresistant histologically-evident NB in BM is reliably achieved with anti-G D2 immunotherapy [35]; and novel salvage therapies have emerged. Regarding this last point, three ASCT and 14 non-ASCT patients are in continuous 2nd CR/ VGPR and off all therapy with lengthy follow-up since relapse (42+ -109+ months). HR-NB relapse has long been viewed as a systemic and ultimately lethal event; EFS has, therefore, been considered the most meaningful measure of efficacy. For curability of HR-NB, however, long-term OS may now supersede EFS endpoints given recent developments offering hope that the equivalence between relapse and lethality may no longer hold true. Thus, close monitoring [44] may now be detecting    [8], which might be controlled by surgery and/or focal RT, supplemented by systemic therapies that are non-cross-resistant with prior treatments. Examples include chemotherapy regimens [45,46] and novel agents [47,48]. Of note, a multi-modality salvage program centered on intrathecal 131 I-labeled mAbs has yielded prolonged 2nd CR/VGPRs in patients with isolated relapse in the brain [49]. A bivalent vaccine combined with oral β-glucan has shown promise in consolidating 2nd CR/VGPR [50].
With cure of relapsed HR-NB being a realistic possibility, long-term OS gains increased importance. OS stands out as 1) the gold standard for evaluation of a treatment's efficacy; 2) the acid test for drug approval by the Food and Drug Administration; and 3) the driving force for advances in cancer therapeutics [40,51,52]. OS is 100% accurate for event and time, and it takes into account safety (toxic complications), which is a major concern with ASCT [53]. EFS is a surrogate endpoint for early assessment of efficacy, but its validity requires confirmation, either through correlation with OS or by meta-analysis [40,51,52]. This point is well illustrated by the disappearance of a long-term OS advantage for ASCT in the randomized CCG study [38] -an update reported after a meta-analysis had already identified no OS advantage with ASCT for HR-NB [39].
It would appear that the corrected results of the landmark CCG study [38], and the critical importance of long-term OS, support a reevaluation of ASCT for HR-NB. An additional consideration is that the randomized ASCT studies -conducted in 1982-1985 [25], 1991-1996 [17], and 1997-2002 [26] -have uncertain contemporary relevance, given the lower dose-intensity of their induction regimens and the absence or irregular use of local RT and anti-G D2 mAbs. The three randomized studies also preceded modern advances in salvage therapy and in the early detection of recurrent NB [44].
In conclusion, our experience, combined with a critical review of ASCT for HR-NB reaching back 30 years [39] and the loss of survival advantage with ASCT in a major study [38], suggests that ASCT may not improve outcome when local RT, anti-G D2 mAbs, and isotretinoin are used for consolidation after dose-intensive induction chemotherapy. A definitive confirmation of this welcome possibility would require a prospective randomized trial. Discontinuing ASCT for HR-NB would be consistent with the general consensus among pediatric oncologists that this highly toxic treatment in all other extracranial pediatric solid tumors is no longer recommended [36,37].

MATERIALS AND METHODS
Beginning in 2003, patients with HR-NB (stage 4 at age > 18 months or MYCN-amplified stage 2/3/4/4 s at any age) received 3F8/GM-CSF + isotretinoin to consolidate 1st CR/VGPR [54] documented following ASCT (patients referred to MSKCC) or conventional chemotherapy. UHR disease was defined as requiring 2nd-line therapy (because of an incomplete response to induction) to achieve this 1st CR/VGPR. This report concerns the HR-NB patients in 1st CR/VGPR enrolled on protocol 03-077 (NCT00072358) and the successor protocols 09-158 (NCT01183416) and 09-159 (NCT01183429). Major organ toxicity had to be grade < 2 by Common Terminology Criteria for Adverse Events Version 2.0, except absolute neutrophil count (ANC) ≥ 500/µl and platelet count ≥ 10,000/µl were acceptable. Informed written consents for treatments and tests were obtained according to MSKCC institutional review board rules.

Protocol treatment
Immunotherapy cycles comprised priming doses of GM-CSF (Leukine, Immunex) × 5 days, followed by 3F8/GM-CSF × 5 days. 3F8 (prepared as described [55]) was intravenously infused over 30-60 minutes, with dosing at 20 mg/m 2 /day or, in a pilot study within the 09-158/ 09-159 protocols, 80 mg/m 2 /day (high-dose 3F8) for the 1st two cycles. GM-CSF was injected subcutaneously at 250 µg/m 2 /day for the five days of priming and the 1st two days of 3F8, and then increased to 500 µg/m 2 /day. GM-CSF was not given if the ANC was > 20,000/µl. These cycles were separated by 2-to-4-week intervals through cycle 4 and then by 6-to-8-week intervals through 24 months from study entry. Treatments were deferred if patients formed elevated human anti-mouse antibody (HAMA) titers (measured as described [56]). Isotretinoin was administered orally (× 6 courses, as described [17]) between cycles of 3F8, beginning post-cycle 2.

Extent-of-disease evaluations and correlative studies
Disease status was assessed every 3 months for > 36 months by histology of BM aspirates and biopsies obtained from bilateral posterior and bilateral anterior iliac crests, 123 I-metaiodobenzylguanidine (MIBG) scan, and computed tomography or magnetic resonance imaging of chest/abdomen/pelvis [44]. Disease status was defined by INRC [54], modified to incorporate 123 I-MIBG findings. CR: no evidence of NB, including normal 123 I-MIBG scan. VGPR: volume of primary mass reduced > 90%, normal 123 I-MIBG scan, BM(−) by histology. PD: new lesion or > 25% increase in an existing lesion.

Statistical analysis
The difference between clinical and biological features of the ASCT and non-ASCT cohorts was tested using Fisher's exact test for categorical variables and Wilcoxon rank-sum test for continuous variables. Event-free survival (EFS) was defined as time from start of 3F8 to PD, secondary malignancy, or toxic death, and was censored at last follow-up in the absence of these events. Overall survival (OS) was defined as time from start of 3F8 to death or last follow-up. The Kaplan-Meier method was used to calculate the probability of EFS and OS. The log-rank test was used to compare survival curves. Median follow-up was calculated using the Kaplan-Meier method and reversing the censoring indicator. Prognostic impact of clinical and biological features on EFS and OS was tested by univariate Cox proportional hazards regression. HAMA and post-MRD were evaluated as time-dependent variables. ASCT and variables significant in the univariate analysis (p < 0.05) were tested in the multivariate model. Time from the 1st dose of induction chemotherapy to 3F8 and time from ASCT or last dose of chemotherapy to 3F8 were correlated, so only one of them was chosen for multivariate model. Since these variables were correlated with ASCT, we also fitted multivariate models stratified by these variables dichotomized at their medians. All analyses were done using R version 3.0.2 (http://cran.r-project.org/).