Delayed dynamics of migratory response to CTLA-4 blockade reveals a mechanistic view on potential T cells reinvigoration following immune checkpoint blockade

Cytotoxic T-lymphocyte-associated antigen 4 (CTLA-4) and programmed cell death protein 1 (PD-1) receptors, two clinically relevant targets for immunotherapy of cancer, are negative regulators of in immune cell activation and migration. However, optimizing therapeutic outcomes still requires fundamental research to reach a comprehensive insight into the coherent function of immune regulators. Here, we investigated the statistical dynamics of T cells migration as a measure of the functional response to these pathways in an experimental setup of immune checkpoint blockade. For this purpose, we used a previously developed 3-dimensional organotypic culture of patient-derived tumor spheroids. Experiment-based dynamical modeling remarked distinct characteristics of the receptors regulation followed through with the modification of their proportions in the immune modulation. We demonstrated that time-delayed kinetics of PD-1 activation just overrides its relatively more efficient cell-level function which potentially makes an operative contribution to the functional dominance of CTLA-4 in the tumor microenvironment. Simulation results showed good agreement with data for tumor cells reduction and active immune cells count observed in each experiment. These analyses propose a new mechanistic view on relative immunogenicity of PD-1 and CTLA-4 inhibitors manifested in literature and point the possible inherent obstacles in checkpoint inhibition-based immunotherapy of cancer to address in the future. Significance Ex vivo monitoring of temporal response to PD-1 and CTLA-4 in the closure of T cell movement dynamics and elucidating their feasible commitment to the kinetic constraints at cell-level resolution. Delayed dynamics of migratory response to CTLA-4 inhibition revealed a mechanistic view on potential T cell reinvigoration following immune checkpoint blockade.


Introduction
Immune regulatory mechanisms modulate immune response, primarily to allow immune recovery and quench autoimmune reactions. Several mechanisms are involved in immune tolerance and malignant tumors use these mechanisms to escape immune rejection. Among them, immunosuppressive signaling pathways have a critical role in the regulation of chronic immune responses (1). Cytotoxic T-lymphocyte-associated antigen 4 (CTLA-4) and programmed death 1 (PD-1), two negative co-stimulatory receptors, attenuate T cells activation mainly through intrinsic cellular mechanisms (2).
Blockade of PD-1 and CTLA-4 receptors eliminates their downstream signals, which enables T cells reinvigoration and boosts antitumor response. The clinical success of checkpoint inhibitionbased cancer treatment, is beholden to the fundamental research that provides a mechanistic understanding of the immune regulation and tolerance. However, improving therapeutic outcomes in Pd-1 and CTLA-4 blockade requires an even more detailed mechanistic insights particularly on comparative aspects of checkpoints function, which emerge from the coordinated nature of immunoregulatory mechanisms and the compensatory relationship between them (2,(19)(20)(21).
Despite the shared objective for PD-1 and CTLA-4 pathways, many studies indicate some peculiar specificities of each pathway; for example, in the timing of these receptors function, their corresponding immune cells population, the predominant operating environment, and the downstream transducing molecules (22)(23)(24)(25)(26). While TCR activation upregulates both receptors expression, the existence of CTLA-4 cytoplasmic reserves donate to this receptor the exclusive possibility of rapid intracellular trafficking to adjust its membrane recruitment according to TCR signal strength (27)(28)(29). These molecular and mechanistic characteristics can eventually appear in the dynamics of tumor-T cell interaction., However, in spite of the potential clinical relevance, comparative studies characterizing the kinetics of checkpoint induced immune inactivation, remained rare.
Several previous studies considered the dynamics of tumor-immune interaction in the presence of immunosuppressive pathways inhibitors for model-based investigation of their therapeutic outcomes (30)(31)(32)(33)(34). For example, a mathematical framework provided justifications for observations such as sustained tumor rejection in combined radiotherapy and checkpoint blockade despite tumor recurrence in exclusive regimens. The model highlighted their synergistic effect based on the immunogenicity of irradiation and intensification of abscopal effects in concurrent treatment (31). Also, an experiment-based mathematical model was used for effective dose prediction and optimal sequential strategy for radiotherapy and anti-PD(L)-1 treatment (32). While most of the models consider cell-cell interaction networks, the integration of related molecular processes can elucidate the relative effects of intracellular peculiarities and intercellular communication in response to immune checkpoint inhibition.
Herein, we aim to resolve the kinetic pattern govern PD-1 and CTLA-4 functions to address the possible therapeutic limitation originated from differential cell level response to these pathways.
For this purpose, we used a previously developed ex vivo system that recapitulate the tumor microenvironment and makes possible the precise monitoring of the tumor response to immune checkpoint blockade (35,36). The system made up of a 3D microfluidic culture of organotypic tumor spheroids derived from patient samples that retain autologous immune cells (Fig.1a). The applied microfluidic device allowed for the controlled medium treatment with anti-PD-1 and anti-CTLA-4 antibodies as well as short-term evaluation of response to these inhibitors. Previously published data, based upon cytokine measurement and immune profiling of tumor spheroids, confirmed the above system as a novel platform for biomarker identification and systematic evaluation of checkpoint blockade outcome (37).
Here, using time-lapse imaging and single-cell tracking, time series of immune cell movements extracted in three conditions of individual and combined PD-1 and CTLA-4 blockade, to derive the state of immune cells activity and follow temporal response to immunosuppressive factors.
The retrogressive movement capability of immune cells, as well as distinct dynamics of their response to PD-1 and CTLA-4 pathways, delicately captured by temporal parameter changes of a heterogeneous random walks model. Finally, a model of tumor-immune interaction based on a system of ordinary differential equations revealed how characteristic dynamics of PD-1 and CTLA-4 activation potentially imposes limitations on tumor response to PD-1 blockade.

Cell migration tracking
Patient derived organotypic tumor spheroids were studied for migration activity of immune cells in response to checkpoint blockade. Three and 10 spheroid cultures (biopsies form melanoma patients responsive to PD-1/PD-L-1 blockade) were investigated respectively for anti-CTLA-4 + anti-PD-1 treatments and exclusive inhibition of PD-1. Response to CTLA-4 was analyzed in five thyroid tumor cultures derived from a patient nonresponsive to PD-1/PD-L-1 blockade, the same sample was also accounted for six control experiment with no drug treatment. The detectable moving cells were tracked in each experiment (Supplementary Note 1) to extract cell trajectories, and related quantities such as cell velocity and persistence using the Tracking package of IMARIS software.

Calculation of mean squared displacement of the cells
Mean squared displacement was evaluated as a substantial criterion to investigate the diffusive behavior of the cells. The MSD of a walker is defined as follows:

Bayesian inference method for parameter estimation of the migration model
We used heterogeneous random walks (38) for modeling the migration of tumor-infiltrating

Simulation of cell trajectories
The time-varying autoregressive model of the first order was evaluated for analysis of cell movements. In this way, we used the estimated activity and persistence of the cells to simulate trajectories based on the same autoregressive process. The model parameters were further estimated for the simulated trajectories to be compared with the input parameters and confirm the accuracy of the applied algorithm. The later statistical verification of the simulated trajectories determines the accuracy of this modeling paradigm.

Mathematical Modeling
A set of biological assumptions have been considered to infer a mathematical model of checkpoint induced immune modulation. These assumptions outline the theoretical underpinning of the work and reduce the parameters and equations to those sufficient for rational description of the experimental data.

Accessibility of tumor cells:
The experimental results indicate that even in the case of two pathways blockade a fraction of tumor cells would survive. We supposed that this is because of the limited access of tumor infiltrating lymphocyte to the population of tumor cells, a well-known phenomenon play a role in tumor-associated immune resistance.(39) Assuming a spherical geometry for tumor spheroids, equation (7) relates the accessible fraction of tumor cells to their total population (40).

Direct and indirect tumor -T-cells interaction:
The subpopulation of accessible tumor cells may be directly killed by T-lymphocytes and in turn, stimulate PD-1 and CTLA-4 pathways via direct contact. The effects of T cell exhausting factors other than PD-1 and CTLA-4 (i.e. free radicals, secreted cytokines, oxygen limitation, …) are assumed to be dependent on the total population of tumor cells (regardless of the direct accessibility to T cells) (41,42).

Up-regulation of PD-1 expression:
We assume that sustained presence of cancer cell antigens in the cultured tumor microenvironment leads to the up-regulation of PD-1 surface expression. The rate of this induction is considered to be independent of cancer cell populations.

No proliferation and recruitment in 3D channel microenvironment:
No immune cell recruitment occurs in this experimental setup, and tumor cell proliferation can be ignored because of limited nutrient availability in this condition.

Lymphocytes activity modeling: It is assumed that tumor cell death occurs in proportion to
lymphocytes activity which appears in their migratory behavior. Therefore, the dynamics of activity in tumor interacting lymphocytes is modeled, instead of their population changes.

PD-1 signaling pathways
As stated above genetic, transcriptional, and translational regulation of PD-1 expression cause a It is assumed that all of the three subpopulations of active lymphocytes may be deactivated by CTLA-4 receptors with rate along with inhibitory factors other than PD-1 and CTLA-4 (i.e., radical formation, hypoxia …) with rate of performance. PD-1 signaling pathway can exclusively inhibit the cells in full expression state of PD-1 ( ). Cytotoxic activity of lymphocytes total population causes the death of tumor cells at a rate of (Eq. 6).
Finally, the presumption of cell to cell contact in CTLA-4 and PD-1 pathways activation as well as tumor cell death process leads to a system of ordinary differential equations as follows:

Mean squared displacement of the cells:
It was observed that the slope of the MSD log−log plots of the cells takes values smaller than one at many time intervals, which is a characteristic of subdiffusive motions. Furthermore, in all of the three experimental conditions, the slope of the curves has a decreasing trend and almost tends to zero (Fig. 3d-f, blue curves). Temporal pattern of movement activity and persistence of the cells are illustrated in Fig. 4a-c. In each experiment, while the activity parameter shows considerable changing behavior, population-averaged persistence parameter displays moderate variations over the time of observation (Fig. 4a-c, insets). In the case of PD-1 blockade, ensemble activity of the cells displays a relatively constant decrease over time from the very beginning, while blocking the CTLA-4 receptor, drives the cells to a more rapid decrease of the motion but after a short initial delay (Fig. 4b, c). In the case of drug combination, the migratory activity of the cells showed more intensive fluctuations but only a mild overall decrease during the time course of the experiment (Fig. 4a). Model simulation for PD-1 blockade: In the absence of CTLA-4 inhibitor, activation of the relevant signaling pathway affects the dynamics of lymphocytes activity (Fig. 2).

Temporal variation of cell migration parameters in checkpoint blockade experiments:
In this scenario, the parameters ξ and set as the values inferred from the previous simulation and similarly the initial values experimentally determined for this condition. In addition, tumor cell death data are used for model validation.
Training the experimental diagram, the value of 9.6 * 10 −6 ℎ −1 −1 was estimated by the solver for lymphocytes inactivation rate triggered by CTLA-4 signaling cascade ( ) (Fig. 4b, blue curve Model simulation for CTLA-4 blockade: In this case, up-regulation and consequent activation of PD-1 pathway modifies the dynamics of lymphocytes activity (Fig. 2). To set the initial values of the model we supposed that the whole population of activated lymphocytes initially belongs to the subpopulation lacking PD-1 receptor. In this standing, the parameters and were learned so that the model could explain the experimental data (Fig. 4c, blue curve).
Because of the possibly unrealistic assumption above, we attempted to correct the initial condition treatments, Fig. 4c yellow curve). The inferred values of the parameters represented in Table 1.

Comparing in vivo population of reinvigorated cells in PD-1 and CTLA-4 blockade: In addition
to the investigated spheroids (average size of 70 µm (37)) we ran the model for a more realistic tumor size about 100 3 (37) better mimicking in vivo condition and the consequent therapeutic implications of the model.

Results demonstrate that the population of immune cells deactivated by CTLA-4 receptor , by
nearly ten-fold, out-numbers the cells whose inactivation associated with PD-1 receptor.
Presuming that this model outcome is a good representation of the evolved tumor situation, this suggests that the inhibition of the CTLA-4 pathway would reactivate a larger population of immune cells than PD-1 pathway blockade.  following checkpoints inhibition, the sub diffusive behavior of the cells even at the beginning of the assessment, demonstrates T cells tendency for physical engagement rather than pointless receding movements in tumor environment, (Fig. 3,4).
In the current study, the motion pattern of tumor-associated immune cells was investigated in 3D ex vivo culture, and their motion behavior leveraged for the dynamical modeling of tumor-immune cell interaction dominated by immunosuppressive pathways. To this end, time series of cell migration events were applied to infer the extent of immune modulation related to each pathway; using this functional response, immune cells inactivation was quantified and the relative kinetics Experiment-based estimation of the parameters revealed the greater potency of immune modulation for PD-1 rather than CTLA-4 receptor (see Table 1), in agreement with the previous findings demonstrated that PI3K/Akt signaling transversed by PD-1 and CTLA-4 pathways employing different molecular networks which is more effective in PD-1 signaling cascade (25). Importantly, the model could explain how, despite the functionally more efficient signaling cascade initiated by PD-1, the presence of cytoplasmic vesicles store CTLA-4 receptor, eventually leads to a more dominant immunosuppressive effect of this receptor at the level of lymphocyte populations (43).
The model assertion that the specific regulation of CTLA-4 expression, mainly relying on vesicular transport, may precipitate its suppressive function, has been acknowledged by a study aimed at the modeling of CTLA-4 intracellular trafficking, showing the significance of CTLA-4 recycling in the kinetics of its molecular function. They show that even in the presence of high-affinity ligands, the intracellular CTLA-4 resources, lasts for several hours thanks to its cytoplasmic reserves and only thereafter its surface expression turns to be dependent on de novo synthesis (44). It should be noted that this relatively broad immunogenicity may lead to the activation of other immunosuppressive pathways and a more intense reinvigoration of regulatory cells, possibly resulting in the limited therapeutic response to anti-CTLA-4 compared with anti-PD-1 treatment (43,(51)(52)(53)(54)(55)(56). Indeed, our analyses further remark the importance of pretreatment immune profiling and highlights the significance of identifying predictive biomarkers as a strict irrevocable goal for improving response to CTLA-4 blockade (43,55).
On the other hands, considering the importance of the initial response and priming of anti-tumor immunity for immunotherapy outcome, this profound immune excitation predicted here provide some rationale for combined and alternative therapeutic strategies; patient treatment with immunostimulatory drugs with lower toxicity following initial activation of the immune system by CTLA-4 blockade may improve clinical outcomes (57). Moreover, in patients without baseline population-related biomarkers for anti-PD-1 treatment, this strong immune restoration may establish the condition for improved objective response (51,54,58).
The presented quantitative insight into the distinct PD-1 and CTLA-4 kinetics and functions can be useful for the development of more practical models for planning effective treatment schemes.
More generally this study highlights the need for more accurate understanding of the kinetics of response to checkpoints inhibitors in order to develop more effective therapeutic strategies based on sequential stimulation of the immune system.