Evaluation of the tumor microenvironment using image analysis for clinical trials.
The number of clinical trials examining immunotherapy and combination therapies with immunotherapy as a backbone has exceeded 1000 in 2016. While PD-L1 by immunohistochemistry has been successfully employed as a companion diagnostic for several therapies across a handful of indications, PD-L1 is only one potential biomarker for showing predictive response. PD-L1 does not provide a complete depiction of the tumor and it is widely documented that some PD-L1 negative patients have responded positively to immunotherapy. As more combination therapies are examined, more informative markers will need to guide when and how a treatment or combinations of treatment should be considered.
Recently, efforts have focused on the tumor microenvironment as an indicator of response to immunotherapy. Specifically, “hot” tumors with involvement of the immune system have been suggested to have statistically higher response to treatment protocols. While some have focused on the inflammatory gene panels and mutational burden, adequately assessment of various types of T cell responders in the area of the tumor could provide insight into a patient’s prognosis. The environmental makeup remains extremely complex, with some cells enhancing immunosuppression and others inducing potent anti-tumor responses. Furthermore, evaluation of proximity and morphology of the differing types of T cells will be critical in assessing the status of the existing immune involvement.
In the case of standard immunohistochemistry, the challenge in assessment lies within the inherent semi-quantitative analysis output. Manual pathology assessment of a single IHC marker is constrained by the intensity and percent of cells stained. Additionally, standardization and reproducibility remain a challenge in the clinical trial setting. Conversely, immunohistochemistry allows for the evaluation of differential expression among the heterogeneity of the sample by which no other testing modality is suitable. Employing image analysis and machine learning from analytics software such as Definiens, allows for standardization of the data captured for a given marker; thereby making the analysis of the marker under study more robust and powerful. In the current example, a standard IHC marker is assessed and Definiens is to develop an enumeration algorithm to assess the overall intensity of the marker as well as the density of the marker in the tumor region (reported in cells/mm2). Subsequent validation show the correlation of the analysis using the software compared with the manual assessment meet the preset criteria. Furthermore, the repeatability of the assay when using image analysis consistently meets criteria significantly better than manual assessment. Overall, the employment of image analysis for any given marker helps to improve pathology assessments and standardizes interpretation as well as systematically identifies the optimal cut-point threshold.