Intermittent BRAF inhibition in advanced BRAF mutated melanoma results of a phase II randomized trial

Combination treatment with BRAF (BRAFi) plus MEK inhibitors (MEKi) has demonstrated survival benefit in patients with advanced melanoma harboring activating BRAF mutations. Previous preclinical studies suggested that an intermittent dosing of these drugs could delay the emergence of resistance. Contrary to expectations, the first published phase 2 randomized study comparing continuous versus intermittent schedule of dabrafenib (BRAFi) plus trametinib (MEKi) demonstrated a detrimental effect of the “on−off” schedule. Here we report confirmatory data from the Phase II randomized open-label clinical trial comparing the antitumoral activity of the standard schedule versus an intermittent combination of vemurafenib (BRAFi) plus cobimetinib (MEKi) in advanced BRAF mutant melanoma patients (NCT02583516). The trial did not meet its primary endpoint of progression free survival (PFS) improvement. Our results show that the antitumor activity of the experimental intermittent schedule of vemurafenib plus cobimetinib is not superior to the standard continuous schedule. Detection of BRAF mutation in cell free tumor DNA has prognostic value for survival and its dynamics has an excellent correlation with clinical response, but not with progression. NGS analysis demonstrated de novo mutations in resistant cases.

1. There was any crossover planned? The intermittent was significantly worse than the continuous. How do you managed this question? 2. In the Extended data eTable 1. -Patient Characteristics there is no p value to analise the homogeneous distribution of the two groups, please provide.
3. The stratification ratio 1:1 was not applied according with the M1a or M1b and the have different prognosis? Did it affect your results? Can you show some evidence? 4. For the e GeneReadQIAact Custom DNA Panel, did you determinate a OLB for each mutation? Can you explain?
Reviewer #4: Remarks to the Author: This study reports the results of a randomised phase II study that aimed to assess the efficacy and safety of continuous versus intermittent schedules of administration of Vemurafenib in combination with Cobimetinib, in previously untreated BRAFV600-mutation positive patients with unresectable locally advanced of metastatic melanoma. Beyond the initial trial outcomes in the study design, authors have conducted an exploratory translational sub-study in cell-free DNA using a subset of the patients. Authors should be commended for this well designed and well conducted study and the concise and well written report. However I have few comments that I hope, will help to better interpret the results. In their study protocol (p 58), it is stated "With this sample size we would have a power of 80% to detect a 23% difference in the percentage of patients free of progression at 1 year…". However, when looking at Figure 1A, we can see that the 12 month survival rates difference is much larger than 23% (about 60%-25% =35%), in addition the observed median between the two arms (16.2 versus 6.9) resulted in a much higher median difference compared to the expected median difference of 4 months (10-6 months) in the sample size calculation. Despite these findings that I found rather interesting, the main conclusion of the study is based on the single Log-rank test p-value. A known weakness of Log-rank test is it fails (no power) if the 2 hazard cross. Even though the hazard curves are not ploted I suspect the proportional hazard assumption does not hold for the PFS survival curves in Figure 1A (for the extended eFigure 3, the OS curves are crossing consequently the hazard crossed and the proportional hazard is violated). I would suggest to formaly assess the proportional hazard assumption and consequently use more powerful test when the proportional hazard does not hold. Minor comment: please include 1-year and 2-year survival rates in Figure 1a extended eFigure 3 as planned in the SAP.

1
Response to Reviewers specific comments: Reviewer #1 (Remarks to the Author): In their report Gonzalez-Cao report on the second trial testing intermittent versus continuous BRAF+MEK inhibition. The trial confirms the previous data, and once more shows that all animal models arguing for a drug holiday from MAPK pathway targeting were wrong. These data are very important, as the animal model data drew so much attention that (also in my experience) many patients wanted to follow drug holidays and it was very hard to convince them not to do. The manuscript is clear and well written.

ANSWER: No modifications are suggested. Thank you.
Reviewer #2 (Remarks to the Author): This manuscript describes a 70-patient study of intermittent versus continuous dosing of vemurafenib with cobimetinib in patients BRAF mutated metastatic melanoma conducted by the Spanish Melanoma Group. A study published in Nature Medicine last year examined intermittent dosing dabrafenib and trametinib in a similar population also showing that continuous dosing was superior. These initial findings were surprising to many and a number of possible explanations were discussed including the possibility that the long half-life of trametinib may have blunted the drug withdrawal effect such that any benefit of intermittent dosing was obscured. The half-life elimination of vemurafenib is approximately 2-1/2 days and the half-life elimination of cobimetinib is about 2 days so that full elimination of both drugs is anticipated within 2 weeks, an advantage over the previous study. Furthermore, the current findings suggest that the superiority of continuous dosing maybe a class effect of these medications. Based on this, the paper could be an important addition to the medical literature.
However, the manuscript requires major revisions prior to publication. The brief report format is challenging given the volume of data presented and some improvement in summarizing these data could be helpful. The supplementary data section is 70 pages long and much of these data could be summarized more effectively or culled. For example, it might be better if the cell-free DNA findings over the course of treatment could be summarized graphically rather than including a large number of figures with individual patient data. The comments at the end of the article summarizing the findings are very brief, perhaps by necessity given the brief report format, but there is no substantive discussion of the primary clinical findings, the superiority of continuous dosing. The cell-free DNA findings are interesting, both replicating and expanding on existing published data. A further discussion of the relatively weak predictive value of serial cell-free DNA levels in predicting disease progression would be desirable, particularly in patients on the continuous dosing arm.

ANSWER: We have incorporated a new paragraph at the end of the manuscript discussing the primary findings and the weak predictive value of BRAF identification in cfDNA at progression:
"Moreover, the current findings suggest that the superiority of continuous dosing may be a class effect of BRAF/MEK inhibitors and reject the previous hypothesis from Algazi et al posing that the long half-life of trametinib may have blunted the drug withdrawal effect. 6 In our study, as half-life elimination of vemurafenib and cobimetinib are approximately 72 and 48 hours, respectively, the full elimination of both drugs is anticipated within two weeks, so the intermittent schedule is causing a subsequent intermittent drug withdrawal in plasma. These results come to confirm the detrimental effect of the intermittent dosing, contrary to previous results in animal models arguing for the benefit of a drug holiday over the MAPK pathway inhibition." "Finally, the study confirms that identification of BRAFV600 mutation in pretreatment cfDNA is associated with a dismal prognosis, with striking differences between treatment arms mainly in preBRAF+ patients. Clinical utility of BRAF testing in pretreatment cfDNA could be particularly helpful for patients with pretreatment normal LDH levels because positive results of BRAF testing identifies patients with a dismal prognosis into this subgroup. Interestingly, while serial monitoring of BRAF mutation in cfDNA throughout treatment has a good correlation with clinical response, BRAF testing did not capture disease progression in a significant number of patients, mainly in those treated with the standard schedule. Understanding the role of BRAF mutant clones in melanoma resistance to BRAF inhibition is key in order to conduct rational drug development in this field. The identification of different mechanisms of resistance in plasma samples at progression could be helpful for guiding the research on novel targeted agents as salvage therapy for every individual case." Although the authors' intent is generally understandable, there are some passages that would benefit from editing for flow and clarity. For example on line 133, "At this respect, targeted NGS analysis … showed mutations in…." ANSWER: The language has been reviewed. We have modified the sentence in line 133 to: "In this regard, targeted NGS analysis of eight cases at radiological progression revealed mutations in NRAS, KRAS, PIK3CA and TP53, as well as amplifications in BRAF, PDGFRA and KIT".
3 Labeling of some of the figures also makes it more difficult to track the main findings. For example, Figure 1B would be easier to read if the groups were labeled more descriptively (rather than G1, G2, G3, G4). The same is true with extended data figure 5. There are also some formatting problems including some boxes that are present presumably because the authors did not designate the text font as "symbol." ANSWER: We have included a box in each figure clarifying the label of Figure 1B and eFigure 5 Reviewer #3 (Remarks to the Author): The paper is well justified and conducted clinical and translational trial performed to answer an important hypothesis on metastatic melanoma treatment.
Although is a high value research and worthwhile for publication, some points need to be addressed:

In our opinion, it is unlikely that the results of the primary end point of the study have been affected by a heterogeneous distribution of patients between the treatment arms.
We have not included this new information in the revised manuscript in order to follow Reviewer #2's recommendation related to the extension of the supplementary appendix "The supplementary data section is 70 pages long and much of these data could be summarized more effectively or culled". Please, let us know if these data should be included in the manuscript. This study reports the results of a randomised phase II study that aimed to assess the efficacy and safety of continuous versus intermittent schedules of administration of Vemurafenib in combination with Cobimetinib, in previously untreated BRAFV600mutation positive patients with unresectable locally advanced of metastatic melanoma. Beyond the initial trial outcomes in the study design, authors have conducted an exploratory translational sub-study in cell-free DNA using a subset of the patients. Authors should be commended for this well designed and well conducted study and the concise and well written report. However I have few comments that I hope, will help to better interpret the results. In their study protocol (p 58), it is stated "With this sample size we would have a power of 80% to detect a 23% difference in the percentage of patients free of progression at 1 year…". However, when looking at Figure 1A, we can see that the 12 month survival rates difference is much larger than 23% (about 60%-25% =35%), in addition the observed median between the two arms (16.2 versus 6.9) resulted in a much higher median difference compared to the expected median difference of 4 months (10-6 months) in the sample size calculation. Despite these findings that I found rather interesting, the main conclusion of the study is based on the single Log-rank test p-value. A known weakness of Logrank test is it fails (no power) if the 2 hazard cross. Even though the hazard curves are not ploted I suspect the proportional hazard assumption does not hold for the PFS survival curves in Figure 1A (for the extended eFigure 3, the OS curves are crossing consequently the hazard crossed and the proportional hazard is violated). I would suggest to formaly assess the proportional hazard assumption and consequently use more powerful test when the proportional hazard does not hold.
ANSWER: The p value obtained in the PFS analysis (p=0.07) was significant according to the design of the study that used a high α error level for sample size calculation (error α = 0.1). Although with this design we had a 10% probability of incorrectly rejecting the true null hypothesis, the advantage was that the number of patients included and randomized was reduced compared with other conventional study designs. In our opinion, we should not focus on the P value alone to decide whether the experimental treatment arm is clinically different, because it is essential to consider the magnitude of treatment differences and the power of the study. In this study the magnitude of the difference between both schedules is clinically relevant and it is concordant with the previous data that showed detrimental results with the experimental schedule.
We have included information about the sample size calculation in the Supplementary Online Content in the Method Section: "Using the method described by Brookmeyer R., for an error α = 0.1 and an error β = 0.20, it will be necessary to include 34 evaluable patients per treatment group, using Log-rank (Mantel-Cox) 2-sided test. With this sample size we would have a power of 80% to detect a difference of 23% in the percentage of patients free of progression to 1 year (with an error α = 0.1)" We have tried to modify the strength of our main conclusions, including a final sentence: "These results come to confirm the detrimental effect of intermittent dosing, contrary to previous results in animal models arguing for the benefit of a drug holiday over the MAPK pathway inhibition." Minor comment: please include 1-year and 2-year survival rates in Figure 1a extended eFigure 3 as planned in the SAP.