Mechanosensitivity of PD1 Localization on the CD8 T-cell Membrane as Interrogated Using STORM Microscopy and Monte Carlo Simulations

Introduction: Solid tumors represent up to ninety percent of all cancers and pose significant challenges to designing effective therapies due to the highly complex microenvironments they establish. These microenvironments suppress anti-tumor immune responses largely mediated by cytotoxic CD8 T-cells carrying receptors (TCRs) that recognize cognate antigens for killing. Cancer cells inhibit TCRs by upregulating expression of checkpoint inhibitor molecules such as PD-L1, enabling them to escape killing. Despite the demonstrated importance of force in PD-L1 inhibition however, little is known about the mechanosensitivity of PD-L1 T-cell inhibition. Additionally, recent studies implicate cancerous extracellular vesicles (EVs) as potent immunosuppressors due to their upregulated levels of PD-L1. However, given the size of EVs ( < 150nm), it is unclear how they are capable of such potent immunosuppressive behavior. Our collaborators demonstrated that the adhesion molecule ICAM-1 is necessary for EVs to successfully suppress T-cells. Furthermore, our lab previously demonstrated that the potency of nanoparticle/cell interactions is dependent on mechanosensitive multivalent binding and clustering. Thus, we hypothesize that PD-L1 inhibition is mechanosensitive and, in the case of EVs, potentiated by the formation of PD1 nanoclusters on the T-cell membrane. To test this, we utilize STORM superresolution microscopy in conjunction with biophysical Monte Carlo simulations of EV/T-cell interactions to provide novel insight into the localization of PD1 molecules under varying mechanical conditions. We use these results to inform a stochastic model of intracellular TCR signaling and find that EV levels of PD-L1 are capable of significant T-cell inhibition when clustering is accounted for.

Materials and

Methods: For PD-L1 inhibition studies, we use primary mouse CD8 T-cells whose cortical tension has been modulated using the cytoskeletal inhibitor ROCKi. T-cells are purified from splenocytes using negative selection and then stimulated with ɑCD3 and ɑCD28 antibodies (2μg/ml each) for 24 hours prior to treatment. Treatment is performed using either soluble, slide-bound, or TAM EV PD-L1 and in the presence of ɑCD3/CD28 antibodies. After treatment, localization of PD-1 and LFA-1 molecules is observed using STORM microscopy. In silico, we use Monte Carlo simulations of PD-L1/T-cell interactions and simulate changes in the T-cell cortical tension by modulating excess membrane area (Aex), defined as the ratio of the curvilinear area of the membrane and area of the EV projected onto the membrane, where increasing Aex corresponds to decreasing cortical tension. In the simulation, the fluid cell membrane is discretized as a triangulated mesh and evolved in accordance with the Metropolis criteria, which considers the thermodynamic cost associated with attempted moves (eg. displacing a vertex, diffusing a protein) and accepts it with a certain probability dependent on that cost. Independent simulations are allowed to reach thermal equilibrium and then sampled for microstates to obtain representative multivalency states and local molecular densities that can be mapped to STORM results. The calculated molecular densities are input into a stochastic ordinary differential equation model of TCR signaling that predicts T-cell activation markers given reaction rates known from prior studies.

Results, Conclusions, and Discussions: Using Monte Carlo simulations, we use Djikstra’s algorithm to find the shortest distance along the membrane between bound molecules of the same species and find that TCR and PD-1 molecules have distinct distributions of distances from that of LFA-1 molecules (Fig.1). Specifically, a higher fraction of these populations were found within a 150 nm distance from each other relative to LFA-1, suggesting that TCR and PD-1 molecules favor smaller clusters. Using the number of bound molecules within this 150nm neighborhood, we calculate a local molecular density to use as initial conditions for our stochastic model of TCR signaling and find that CD8 T-cell activation as measured by phosphorylated AKT levels (AKT*) is sensitive to clustering behavior, enabling significant reduction of AKT* levels with EV levels of PD-L1 when the effects of clustering are accounted for. Specifically, we first demonstrate that T-cell activation without PD-L1 inhibition is mechanosensitive by performing parameter scans for the clustering parameters TCR density and Aex (cortical actin tension) and finding that there exist critical thresholds for each parameter when the other is held constant in order to induce significant AKT* expression (Fig.2A). We then demonstrate that using TCR/PD-1 densities and Aex values found from Monte Carlo simulations demonstrating clustering behavior results in significant reduction in AKT* levels when even a single molecule of PD-L1 is present (Fig.2B). To validate these findings, we use STORM microscopy (Fig.3) to image the localization of PD-1 and LFA-1 molecules on the membranes of primary mouse CD8 T-cells after treatment with TAM EVs to provide novel evidence of the formation of EV-induced PD-1 nanoclusters. Taken together, these results bolster the growing evidence supporting the general importance of mechanosensitivity in CD8 T-cell activation and provide novel insight into how molecular clustering may be a mechanism by which TAM EV PD-L1 is capable of inhibiting T-cells so potently.

Acknowledgements (Optional): We acknowledge funding from the NCI under the Physical Sciences in Oncology Network and the NHGRI under the T32 fellowship program.