Preclinical Immunotherapy Safety and Toxicity Assessment
Importance of Immunotherapy Safety Assessment
One unfortunate example of under-prediction of toxicity was observed with the CD28 superagonist mAb TGN1412. In this case, there was a lack of CD28 expression on the CD4+ effector memory T cells of the species used for preclinical safety testing. Preclinical testing therefore failed to predict the cytokine release syndrome (CRS) later seen in humans.
Also, despite CAR-T cell therapy breakthroughs and encouraging results, potentially life-threatening toxicities are intimately associated with this cell therapy, with CRS being a particularly severe complication. However, in comparison with adverse events seen with immune checkpoint inhibitors (mostly delayed-onset autoimmune adverse events), CAR-T cell toxicities tend to be acute, less diverse, and therefore more predictable.
In Vitro Assays: Providing Predictive, Safety Relevant Data
Since TGN1412, nonclinical studies supporting clinical testing of biological agents acting on immunomodulatory targets have become more rigorous. Now, in vitro testing for cytokine release inducing activity on PBMC or T cells is regularly incorporated into these tests.
Cytokine release is often also evaluated for biological agents that have not been characterized as an immunomodulatory agent but targeting the immune cell compartment.
Interpretation of in vitro cytokine release data is challenging, due to variations between PBMC donors and their level of response. It’s also hard to determine the in vitro cytokine threshold which leads to clinical adverse events.
Although this translatability is not absolute, it can be predictive. For example, in vitro cytokine release was investigated for the bispecific molecule MEDI-565, which targets both the CEA antigen and CD3. The in vitro study was performed on PBMC, demonstrating T cell proliferation and cytokine release upon both CD3 and CEA target engagement. These results were used along with other parameters to determine the minimal anticipated biological effect level (MABEL) and select the maximal safe starting dose.
In addition to cytokine release, off-target binding needs to be documented, including specificity and toxicity for human tissues distinct from the intended target. This is needed to identify potential toxicity towards certain organs.
Using immunochemistry procedures, cross reactivity with a range of human tissues should be determined to identify unknown sites of the target antigen, and unexpected targets due to cross-reactive epitopes.
Select the Right In Vivo Model for Toxicology Studies
Generally, due to the human specificity of immuno-oncology agents and similarities in target expression, cynomolgus NHPs are the most frequently used species in toxicology studies.
Syngeneic mice are the most frequently-used rodent model in immuno-oncology, at the moment. The main advantages of syngeneics are that these mouse models have a fully-competent immune system, allowing the evaluation of mechanisms of action and the study of drug activity.
Using such models requires that a mouse surrogate exists, or that the test molecule cross-reacts with the mouse target.
During the process of generating lead therapeutic candidates, the development of a mouse surrogate antibody is possible. This is a potential solution to the limited safety testing possible with humanized monoclonal antibodies with restricted species cross-reactivity. These surrogates can instead facilitate proof-of-concept studies in rodent efficacy models.
Alternatively, if a surrogate doesn’t exist, genetically engineered mouse models (GEMM) can be used. In these models, a transgene or KI gene has been introduced which replaces the murine proteins with their human counterpart, allowing the evaluation of human biological therapies. The KI model has already shown its utility in predicting potential side effects like autoimmune or pro-inflammatory potencies with candidate human antibodies. For instance, CTLA-4 KI mice treated with anti-hCTLA4 antibody developed autoimmune effects observed in patients treated with anti-hCTLA4.
Syngeneic or GEMM are probably the models which best represent the interaction between a fully immunocompetent system and the host cells. Nevertheless, although still little explored, the utility of humanized mice in non-clinical safety studies can offer some advantages, recapitulating many of the toxic, pro-inflammatory phenotypes observed in human patients experiencing CRS.
For instance in PBMC humanized mice, lymphopenia and CRS are observed after a-CD3 or TGN1412 administration. In a couple of hours after administration, loss of human CD45+ cells is observed with TGN1412, whereas only loss of human T cells is observed with a-CD3, alongside release of human cytokines with both treatments.
Beyond PBMC mice, humanized mice generated through inoculating human hematopoietic cells (from human fetal liver or cord blood stem cells) into immunocompromised mice (e.g. NOG® or NSG™) can be used in safety testing. These models can be used to monitor CRS, as well as additional adverse events associated with immunotherapy treatments.
For example, immunotoxicity and side effects of a-CTLA4 (ipilimumab) has been shown in CD34+ mice engrafted with hepatocellular carcinoma (HCC) patient derived xenograft (PDX) models, which was consistent with clinical data. In this study, mice treated with ipilimumab showed significant body weight loss while the group with a-PD-1 (pembrolizumab) was healthy.
Pathological analysis confirmed that the group treated with ipilimumab developed massive cell infiltration and damage in the liver, lung, and kidney while the control and pembrolizumab-treated groups remained normal.
As long as there are open questions remaining around immunotherapy, the development of preclinical models will need to carry on improving and refining to predict clinical activity and toxicity. As a current starting point, further developments of in vitro platforms as well as transgenic and/or humanized mice will provide useful models for solving a range of problems in preclinical immunotherapy assessment.