3rd Immunotherapy of Cancer Conference (ITOC3)
Is PD-L1 really a predictive biomarker?
Dr Paolo Ascierto - National Tumor Institute, Naples, Italy
My talk will be about PD-L1, if the PD-L1 is a real predictive biomarker that can help us in selecting patients.
PD-L1 is a target for lots of immunotherapies – what has your biomarker research shown so far?
First of all I think that I should be clear with the definition of a predictive biomarker. A predictive biomarker is a marker that is able to select the patients that can have a benefit for a treatment. So it should be a sort of black and white. If present, patients can respond or not to a treatment, if not the same, but it should be clear that this is the ideal biomarker. So the question now is is the PD-L1 expression the right biomarker, the right predictive biomarker, because we need a predictive biomarker because the new checkpoint inhibitors are good but the cost is very high, it sometimes may be not sustainable.
Having said that, coming back to the PD-L1, my personal feeling is that PD-L1 is not the right marker. Why is this? If you look at the data on melanoma, consider that several trials, mainly the BMS trial, found that also the PD-L1 negative patients got an important outcome after this kind of treatment. But before discussing about this data, I think that also it’s useful to make some consideration about the PD-L1.
PD-L1 first is an immunological marker, it’s not a molecular marker like the BRAF or the KRAS. What does it mean? This means that it’s a dynamic marker, it can be an inducible marker, interferon gamma radiotherapy, and in other situations can be upregulated or downregulated. And you can find at different times different expressions of PD-L1.
Third, the topographic localisation of PD-L1; PD-L1 is present not in all the tumour microenvironments. Normally you can find PD-L1 at the limits of the tumour with the infiltration. This may be important because if you take a biopsy not in the part of the tumour where PD-L1 is expressed, you can have a negative effect, negative. This is another important question we should evaluate on biopsy or in the old tumour samples, another issue.
Then we have a different system, different monoclonal antibody, different scoring system, different cut-off. Consider that the four most important companies, BMS, MSD, AstraZeneca and Roche use different antibodies with different scoring systems. Three of them use the evaluation on tumour size, one, the Roche, uses the evaluation of tumour [?? 3:29], it’s a little bit different. So, having said that, the question is again if the PD-L1 is the right marker.
In melanoma, as I said, the original sin was in the phase I trial from Suzanne Topalian and colleagues. At that time they found that 9 patients in the 25 patients PD-L1 positive responded, while in the 17 patients PD-L1 negative nobody responded. For this reason at the time all of us said we’ve got the marker. But later, with other studies, not only melanoma but also in lung cancer, it was clear that in the PD-L1 negative there is a group of 15-20% of patients who can respond, despite being PD-L1 negative. This is the problem. We cannot avoid treatment in the PD-L1 negative, in this group of patients. Later, in the phase III trial from BMS, it was performed a prospective evaluation. Patients were stratified according to PD-L1 status; as you know that the pre-planned analysis is more important than a retrospective analysis. This pre-planned analysis in the first trial, 037, it was clear that in the PD-L1 positive the response rates were 44%, higher of course, but all of us know that in PD-L1 positive the performance is better. But in the negative it was confirmed that 20% of patients negative got an important response. But most important were the details in 066: first line nivolumab in a BRAF wild-type population. If you look at the response rate 53% in the PD-L1 positive, 33% in PD-L1 negative.
So consider that you are using the PD-L1 marker in order to select patients. We can decide that we can treat just the patients PD-L1 positive. We have the 53% who can respond but in the same group there is the 47% who don’t respond while in the negative you have 33% who can respond. Is this the right marker for selecting patients with melanoma? No.
But there is another important question in melanoma. You know that now we have the combination of ipilimumab and nivolumab and if you look at the data in terms of progression free survival in the PD-L1 positive the PFS is similar in the combination and in the nivolumab arm monotherapy. You know the combination is more toxic so the question now is if the PD-L1 might be the right marker for selecting treatment for patients, it’s a little bit different. But also in such a case first we need to wait for overall survival data because the overall survival rate was demonstrated in 066 that also the PD-L1 negative have a very good performance and outcome. Second, if you look at the response rate it’s higher in PD-L1 positive with respect to PD-L1 negative. Then the percentage of patients PD-L1 positive in this study, in this group, PD-L1 positive was the lowest found in a clinical trial, 21.5%. So we should be cautious when we discuss this data but mainly if you look at the kinetics of action of the combination you can see that the combination is able to get earlier, deeper, longer response. So this is what makes the difference. In the future we can select the treatment based on patients’ characteristics more than PD-L1.
In lung cancer the situation is a little bit different. We have also a lot of data and sometimes contrasting data. From the BMS trial in the squamous it’s clear in the CHECKMATE-017 that PD-L1 negative and positive can reach good outcomes. So PD-L1 in this group of patients probably is not able to select patients. There are other interesting data from the KEYNOTE-010. KEYNOTE-010 is a randomised trial performed in only PD-L1 positive patients; one of the criteria for selecting patients was that all patients should have more than 1%, cut-off was 1% PD-L1 expression. They found that in the strongest positive patients, more than 50%, this forecast a better outcome but this is clear. But if you look at the Forest plot in these patients, also in the group from 1 to 49%, the performance of ipilimumab was better with respect to docetaxel. So I don’t know which is the reason because we should select patients on the basis of strongest or more. Probably I think that also these data are not useful in order to decide if PD-L1 in lung cancer may be the right marker or not.
More interesting are the data from Roche because they use another scoring system, in only immune system cells and not all in tumour cells. They found four different subgroups. The most interesting data that in the patients without expression on tumour cells, immune system cells, of PD-L1 the survival curves were similar to the control arm of docetaxel. Of course, these can be discussed but there are also in such cases some considerations. First, in this study were all the patients squamous or non-squamous, we don’t have the data according to the different cancers. Second, if you consider the PD-L1 tumour standalone it’s not significant, so what is significant is the special immune system cell. But if you look, the curves are similar. So considering that the PD-L1 treatment has less side effects than docetaxel from an ethical point of view there is also an impact on quality of life of this kind of treatment. Similar data were found in the CHECKMATE-057, non-squamous with nivolumab PD-L1 positive reached a good outcome and of course there was a significant effect on nivolumab with respect to docetaxel. But in the negative, a different cut-off, there was no difference in terms of overall survival. But those in such case is the same question of the Roche trial.
So, in conclusion, what I can say is that before to consider the PD-L1 as the right marker I think that we should test other markers where the so-called breaking away action is probably more evident.