Although study findings have identified tumor mutational burden (TMB) as an important factor in responses to cancer therapy, establishing a definition for expression of this biomarker and harmonizing assays to measure it have proved challenging, according to Naiyer A. Rizvi, MD.
Plasma assays for determining TMB are becoming more diagnostically relevant, said Rizvi, director of thoracic oncology and immunotherapeutics at the Herbert Irving Comprehensive Cancer Center in New York, New York.
What is important now is figuring out the respective values for determining high levels of TMB in each tumor type, since it has been determined that each disease may have its own cutoff for high TMB. Investigators in 1 analysis found that high TMB expression was regularly associated with better survival outcomes; however, the cutoff point for determining this level is not consistent across tumor types, assays, and clinical trials.
Rizvi reviewed recent TMB findings during a presentation at the European Society of Medical Oncology 2019 International Congress on Targeted Anticancer Therapies.
“TMB is a clinically important biomarker in concert with PD-L1 expression for singleagent PD-L1 therapy,” he said. “On its own, it may be an effective predictive biomarker [for response to] combination immunotherapy irrespective of the patient’s PD-L1 expression.”
An analysis of the association between nonsynonymous somatic TMB and overall survival (OS) after treatment with immune checkpoint inhibitor therapy, which included combination or monotherapy with CTLA-4 or PD-1/L1 inhibitors, was assessed using the MSK-IMPACT assay in 1662 patients across multiple tumor types.1
Those patients within the top 20% of TMB expression for their tumor type were observed to have better OS as a result of treatment, and investigators said that the data are most likely indicative of a benefit across all cancer types. Interestingly, the cutoff value for high TMB varied widely, with the highest value observed in colorectal cancer at 52.2 mutations per megabase versus estrogen receptor–negative breast cancer at 4.4 mutations per megabase (Figure).1
Rizvi also pointed out that it is important to consider that different commercial panels may yield varying results and that most platforms require a certain number of mutations to produce an accurate assessment for TMB. “It is important to know [where you are looking]. Which part of the exome you are sampling for TMB can also influence what your TMB estimate will be,” he said.
He went on to compare whole-exome sequencing (WES), the gold standard of genomic profiling, with the FoundationOne CDx and the MSK-IMPACT next-generation sequencing–based assays. He said that although these panels are fairly accurate, the comparison is not 1:1 with WES and there is a lack of standardization for TMB calculations between these tests.
Rizvi highlighted the current efforts of Friends of Cancer Research to collaborate with different pharmaceutical companies, the FDA, the National Cancer Institute, and academic centers in an effort to harmonize the data. The working group will create a set of universal best practices for defining and aligning TMB across platforms. Although he says it is unlikely that these companies will change their methods to mimic other tests, at least there may be a greater understanding of how these data are related.2
Utility of TMB as a Biomarker
Rizvi pointed out that TMB expression does not seem to correlate with the magnitude of PD-L1 expression or immune infiltration in all patients. However, when both high TMB and PD-L1 expression are observed, patients are considered likely to benefit from immunotherapy.
In the phase III CheckMate 227 trial of nivolumab (Opdivo) in combination with ipilimumab (Yervoy) in patients with stage IV or recurrent non–small cell lung cancer (NSCLC) not previously treated with chemotherapy, progression-free survival (PFS) was assessed in patients with TMB ≥10 mutations per megabase.3 The findings showed that although TMB was predictive of response in patients treated with the combination immunotherapy regimen, PD-L1 expression and TMB were mutually exclusive and there was no apparent correlation between the 2 biomarkers.
Figure. High TMB Cutoffa in Select Tumor Types1
Figure. High TMB Cutoffa in Select Tumor Types1 In the CheckMate 026 trial that tested nivolumab monotherapy against chemotherapy in treatment-naïve patients with stage IV or recurrent NSCLC with PD-L1 expression ≥5%, the response rate to therapy was 26% versus 33%, respectively.4 But in an exploratory analysis of patients with high TMB, better response rates were observed with nivolumab than with chemotherapy (47% vs 28%), and although the factors do not co-associate, patients who had high TMB and PD-L1 expression ≥50% together had a significantly improved response rate of 75%.
In another negative phase III trial, IMvigor211, which compared atezolizumab (Tecentriq) with chemotherapy in patients with metastatic urothelial carcinoma after disease progression on platinum-based chemotherapy, patients with high TMB saw an OS benefit with atezolizumab over chemotherapy (11.3 vs 8.3 months; HR, 0.68).5 When investigators looked at patients with both high TMB and PD-L1 expression ≥5%, the OS improved further (17.8 vs 10.6 months) and the hazard ratio was even more favorable, at 0.50. These results indicate that, for single-agent PD-1/L1 inhibitors, TMB is probably not sufficient as a stand-alone biomarker, Rizvi said.
However, that might not be the case for all immunotherapy regimens. The singlearm, phase II CheckMate 568 trial in NSCLC was important for 2 reasons, he said.6 For one thing, it helped estimate the threshold of TMB for a positive response to therapy to be ≥10 mutations per megabase in this tumor type. In addition, the combination improved response and PFS in patients with high TMB regardless of PD-L1 expression, indicating that it may serve as a stand-alone biomarker for some therapies, as with this immunotherapy combination.
Although promising results in patients with high TMB have been observed in some trials, Rizvi cautions that it does not tell the whole story. A new exploratory analysis of the CheckMate 227 trial shows TMB is predictive of improved PFS with combination immunotherapy, but these results did not translate to OS, where the hazard ratio for high and low TMB was 0.77 and 0.78, respectively (23.03 vs 16.2 months).7
Similarly, when investigators superimposed TMB with the presence of neoantigens, patients with low subclonal fractions responded well to pembrolizumab (Keytruda), regardless of TMB expression.8 Conversely, all patients with high subclonal fractions did not do well with pembrolizumab.
TMB in Plasma Offers Advantages in Heterogeneity
The ability to test TMB in plasma offers an advantage over obtaining invasive tumor samples that present difficulties for many patients, Rizvi said. An analysis of the POPLAR and OAK trials of atezolizumab in patients undergoing second-line or higher treatment for NSCLC showed that blood TMB (bTMB) was evaluable in about 75% of patients and was an actionable biomarker for predicting PFS in this subset of patients.9
“What is not clear is that when you look at the OS data, there does not seem to be that much of a difference [in patients] with high or low TMB,” Rizvi said.
Another interesting finding from the analysis was that for patients who had both tumor and plasma evaluable for TMB, the shared variance was observed in 59%. For the remaining patients, TMB was present in either the tumor or blood only. Rizvi said this concordance is not optimal.
“One of the hypotheses is that the tumor and the plasma were not collected at the same time…that could have influenced what the mutation landscape was within the patients over time,” he said.
Investigators also looked at circulating tumor DNA samples that they ran on both plasma and tumor panels and found that concordance was good. He said this indicated that what is observed in TMB collected in tumor samples (tTMB) is just different from what is seen in the plasma.
“The other possible conclusion is that the plasma is just more representative of tumor heterogeneity, and perhaps, in some ways, this could be a better assay than from tumor [samples].”
Some other challenges with plasma assays are getting a reliable estimate of TMB when there is a patient with low shedding of tumor DNA. With the GuardantOMNI assay, Rizvi said TMB has been measured within 85% accuracy, with as low as 0.5% maximum somatic mutation allele fraction in a blood-based assay.
In the phase III MYSTIC study, which looked into the frontline combination of durvalumab (Imfinzi) and tremelimumab versus durvalumab alone or platinum-based chemotherapy in patients with stage IV NSCLC, TMB was evaluated using the GuardantOMNI assay in plasma and FoundationOne CDx in tumor samples.10 The overall correlation for tTMB versus bTMB was 0.6.
“Again, we are seeing a similar phenomenon as far as the concordance not being perfect,” Rizvi said. “In my mind, I think this does reflect the heterogeneity in what is shed [by the tumor] versus [a] single biopsy site.”
He did point out that ≥10 mutations per megabase as defined in tTMB was correlative with bTMB ≥16 mutations per megabase in the MYSTIC analysis. Patients determined to have high TMB and who received the immunotherapy combination derived a PFS benefit in an otherwise negative trial.
Going forward, clinical trials will best testing bTMB as a druggable target in patients who may benefit from immunotherapy. The B-FAST clinical trial (NCT03178552), a NSCLC umbrella trial with separate arms for patients with driver oncogenes as well as patients who are bTMB positive, is currently enrolling and will be testing the efficacy of atezolizumab in patients with high TMB.