Which biomarkers and how many should we use to improve outcomes with immunotherapy?

Tumour heterogeneity is one of the main challenges surrounding the identification and use of biomarkers for immunotherapy. It is unlikely that any particular biomarker will be associated with every cell within a tumour. For example, with tumour mutational burden (TMB), it has been shown that relying on a single tumour sample can lead to a high rate of misclassification, with 20% of lung and 52% of bladder tumours having at least one biopsy with high TMB but low clonal TMB overall (Cell Rep. 2020;31:107550). The misclassification rates are reduced dramatically by the use of a more representative sampling methodology. Tumour heterogeneity partially explains why only a few biomarkers of those identified in preclinical studies are validated in the clinical setting.

Problems with capturing and representing heterogeneity can be overcome by separating tumour sampling temporally as well as spatially. Dynamic changes in biomarkers are particularly pertinent for the development of new medicines. In phase I and II trials, it is important to establish if investigational agents are interacting with the target as determined in preclinical studies. On-treatment changes in biomarkers, such as alterations in gene expression or T-cell receptor (TCR) clonality, can also provide insights into possible mechanisms of resistance, information that can be used to prioritise therapeutic combinations.

I believe that the future lies in the use of an array of different biomarkers for an individual patient. An example of this can be seen from the prospective phase II ADAPTeR study carried out by my group, which looked at the use of nivolumab in 15 treatment-naïve patients with metastatic clear cell renal cell carcinoma (Cancer Cell. 2021;39:1497–1518.e11). The dynamic monitoring of biomarkers was combined with the multifaceted profiling of 115 multiregion tumour tissue samples to identify the best predictor for immunotherapy. For the patients involved, this was revealed to be a combination of single-cell resolution, TCR sequencing and RNA sequencing, with application of the combination to bulk sequencing. Mining existing datasets – both academically generated datasets and those from pharma-led clinical trials – and combining the data in a harmonised way will help to provide a better understanding of different biomarkers, their co-occurrence and their interaction in different patient settings.

Will the use of multiple biomarkers be feasible in the clinical setting? Patients are already increasingly being given the opportunity to have quite comprehensive profiling of their tumours carried out early in the disease process, for example through The Cancer Genome Atlas Program in the US and the 100,000 Genomes Project in the UK. Such profiling provides a wide range of information – including genotype, the status of genetic mutations, copy number, TMB and mutational signatures – all within a single test, and compares favourably with current standard practice, which often involves conducting multiple tests at different stages of the patient journey to identify suitability for different therapies or clinical trial eligibility. From a cost-effectiveness point of view, having the information documented upfront would avoid the costs associated with repeat testing. Clinical application will also be facilitated by technology, which is moving at an incredibly fast pace in this field. There is currently a great deal of research activity in multi-omics and methods that combine DNA/RNA sequencing and methylation profiling, and I foresee it will not be long before we see these techniques entering the clinic. We only need to look at the DNA and RNA sequencing methods we use in everyday practice; these clinically validated instruments were just research tools 10 years ago. At the end of the day, it is not really a question of whether the use of multiple biomarkers in the clinic is feasible; the complexity of the tumour microenvironment and the need to evaluate patients for immunotherapy eligibility mean that it has to be feasible.