Identifying correlates of response to treatment will enhance immune checkpoint blockade, given that the scope of benefit is unevenly distributed to the minority of patients who achieve a response.
Tumor Mutation Burden Challenges
TMB was one of the first recognized correlates of response to immune checkpoint inhibition, but challenges remain in applying it to clinical practice, because of the substantial overlap in TMB values between responders and nonresponders, “raising questions about how best to use TMB to discriminate between these 2 groups,” Dr Vokes said.
Another challenge is the heterogeneity among TMB values generated by the different sequencing platforms used to measure TMB. The distribution of TMB values across different tests differs when the values are plotted linearly, but standardizing them into TMB z-scores substantially decreases the heterogeneity.
In 2 sets of independent patient cohorts—one from Dana-Farber Cancer Institute and one from Memorial Sloan Kettering Cancer Center—patients who achieved complete or partial responses and those with durable clinical benefit from immune checkpoint inhibition were shown to have higher TMB values than nonresponders. Standardizing TMB values from the 2 cohorts into z-scores preserved the relationship between TMB and response to therapy.
When the 2 cohorts were joined for an analysis of response, the highest rates of durable clinical benefit were observed in the patients with the highest TMB values and the lowest rates of durable clinical benefit correlated with the lowest TMB values. Treatment responders, however, were found across the spectrum of TMB values. A receiver-operator analysis showed that the optimal TMB threshold was associated with a sensitivity and specificity of approximately 60%.
“Application of this threshold would have led to 30% of patients being treated without response and failing to treat 12% of patients who would have responded,” Dr Vokes said. Using other TMB thresholds again demonstrated a trade-off in overtreatment versus undertreatment.
It is likely that TMB is not associated with response in a vacuum, Dr Vokes said, but rather it interacts with biologic features. Correlates of response may also be found within the tumor microenvironment.
Cancers with high tumor infiltration and a high expression of cytokines and enzymes involved in the tumor immunity cycle may represent “hot” (T-cell–inflamed) tumors that are more amenable to checkpoint inhibitor therapy, Dr Vokes said.
“Conversely, cancer cells that evade immune detection or manage to prevent the infiltration of T-cells into the tumor microenvironment are classified as ‘cold’ or T-cell–noninflamed tumors, and these may be less amenable to therapy,” she said. Activation of certain oncogenic signaling pathways, such as MAP kinase, WNT/beta-catenin, and PTEN, may lead to T-cell exclusion that contributes to resistance to immune checkpoint inhibition.
Genomic features that appear to correlate with treatment response include mutations in DNA repair enzymes and expression signatures of interferon-gamma signaling or PD-L1, all of which likely promote a T-cell–inflamed environment. Many of the studies identifying these pathways, however, come from small cohorts and have not been confirmed in subsequent cohorts.
A better understanding of the relevant biology that contributes to tumor response to treatment, and how best to integrate multiple genomic correlates of response into better prognostication are needed, Dr Vokes said.
By aggregating 249 whole exome sequencing tumors across different cancer types, Dr Vokes and colleagues were able to recapitulate the association between P10 loss and resistance to immunotherapy and to identify an association between cell-cycle signaling and response. But even with this large cohort, many of the findings were statistically underpowered, she said.
Other markers of the inflamed tumor microenvironment, such as the T-cell–inflamed gene-expression profile, may also interact with TMB to help select the patients who are most likely to respond to immune checkpoint inhibitor therapy.
Multivariate models that incorporate clinical variables (ie, smoking status, PD-L1, and histology) and integrate multiple transcriptomic and genomic features are being assessed to improve response prediction, Dr Vokes said.