Two post-hoc analyses of KEYNOTE studies presented at ESMO 2019
Conflicting results from two analyses presented yesterday at ESMO Congress 2019 give more cloud than clarity regarding the role of tTMB as a predictive biomarker for pembrolizumab in non-small-cell lung cancer (NSCLC).
The first of the two post-hoc analyses of KEYNOTE studies reported an association between higher tTMB levels and improved clinical outcomes in previously treated and untreated patients receiving pembrolizumab monotherapy for PD-L1-positive NSCLC (Abstract LBA79). Comparing pembrolizumab with chemotherapy in patients with advanced NSCLC of mixed histology and a tumour proportion score (TPS) ≥1%, the analysis included 253 (24%) patients from KEYNOTE-010 and 793 (62%) patients from KEYNOTE-042. High and low tTMB was defined using a prespecified cut-point of 175 mutations/exome, derived from meta-analysis of tTMB and gene expression data from pembrolizumab clinical trials across multiple tumour types. tTMB was significantly associated with overall survival (OS), progression-free survival (PFS) and objective response rate (ORR) in the pembrolizumab arms but not the chemotherapy arms. In both studies, improvements with pembrolizumab were confined to patients with TMB ≥175 mutations/exome (high): OS hazard ratios for KEYNOTE-010 and KEYNOTE-042, respectively, were 0.56 and 0.62 for high tTMB. tTMB was not associated with TPS in either treatment arm.
In the second analysis, there was no association between tTMB and outcomes in patients receiving combination pembrolizumab plus platinum-based chemotherapy for advanced, previously untreated NSCLC of mixed histology (Abstract LBA80). The analysis included around half of the patients treated in each of the KEYNOTE-021, KEYNOTE-189 and KEYNOTE-407 trials. tTMB was not significantly associated with benefit of pembrolizumab plus chemotherapy versus chemotherapy alone: using the same tTMB definitions as the first analysis, OS hazard ratios for KEYNOTE-189 and KEYNOTE-407, respectively, were 0.64 and 0.74 for high tTMB.
"The novel aspect of these data is not the relationship between pembrolizumab and TMB, which is consistent with previous reports, but the lack of a relationship between pembrolizumab plus chemotherapy and TMB,” says Dr Kevin Litchfield from The Francis Crick Institute, London, UK.Before jumping to possibly premature conclusions, differences between the monotherapy and combination therapy trials that may have influenced the lack of association should be studied more closely. The first is PD-L1 status. "PD-L1 expression was a requirement only for monotherapy trials and, as we know that pembrolizumab benefit is associated with the degree of expression, this may be an important factor,” says Litchfield. Sample size and/or the statistical power of the studies may also have an influence. Over 400 patients received pembrolizumab in KEYNOTE-042, compared with only around 200 in the biggest combination trial, KEYNOTE-189. "However, it may also be that, as datasets mature and become bigger, an association between tTMB and outcomes emerges for combination therapy,” continues Litchfield, citing the KEYNOTE-189 hazard ratios for PFS for high and low tTMB—0.32 and 0.51, respectively—which show a trend towards an association.
Beyond those caveats, biological mechanisms, such as differences in the immune response to chemotherapy stimulation, may be involved, although more data would be needed to support such speculation. Prof. Tony Mok from the Chinese University of Hong Kong, Investigator on the KEYNOTE-042 study and a co-author on Abstract LBA79, believes that the methodology for both the single-agent and combination therapy analyses requires further scrutiny. He is particularly interested in the definition of the cut-point for high/low tTMB, which he regards as flawed. "The 175 mutations/exome cut-point for the analyses was derived from across all the different tumour types treated in pembrolizumab trials. However, we now know that while the 10–20% of patients with the highest PD-L1 levels appear to benefit the most from pembrolizumab, the absolute cut-points differ for each tumour type.1 So one standardised cut-point will not be suitable for clinical decision making across the board.” He is also concerned that the limited numbers of patients with tissue available for TMB analysis may distort the balance of the populations studied. "If we want to have the final answer, we need to conduct a prospective study,” he says.
In summary, the analysis suggests that tTMB is at best a weak predictive marker with probable limited clinical utility. "Even in the high tTMB setting in KEYNOTE-042, the ORR of 34.4% shows that two-thirds of patients did not respond to pembrolizumab,” says Litchfield. "To be of use, it is likely that TMB would have to be combined with other predictors of checkpoint inhibitor response, such as immune filtration scores and biomarkers of immune evasion, but the danger is that the process would then become too complex for realistic clinical implementation.”