Quantitative systems pharmacology model of GITR‐mediated T cell dynamics in tumor microenvironment

Abstract

T cell interaction in the tumor microenvironment is a key component of immuno‐oncology therapy. Glucocorticoid‐induced tumor necrosis factor receptor (TNFR)‐related protein (GITR) is expressed on immune cells including regulatory T cells (Tregs) and effector T cells (Teffs). Preclinical data suggest that agonism of GITR in combination with Fc‐γ receptor‐mediated depletion of Tregs results in increased intratumoral Teff:Treg ratio and tumor shrinkage. A novel quantitative systems pharmacology (QSP) model was developed for the murine anti‐GITR agonist antibody, DTA‐1.mIgG2a, to describe the kinetics of intratumoral Tregs and Teffs in Colon26 and A20 syngeneic mouse tumor models. It adequately captured the time profiles of intratumoral Treg and Teff and serum DTA‐1.mIgG2a and soluble GITR concentrations in both mouse models, and described the response differences between the two models. The QSP model provides a quantitative understanding of the trade‐off between maximizing Treg depletion versus Teff agonism, and offers insights to optimize drug design and dose regimen.

Study Highlights.

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

T cell interaction in the tumor microenvironment (TME) is a key component of immuno‐oncology therapy. Anti‐GITR antibodies have modulated T‐cell profiles in the TME and demonstrated compelling efficacy in syngeneic mouse tumor models.

• WHAT QUESTION DID THIS STUDY ADDRESS?

This report developed a novel quantitative systems pharmacology (QSP) model of an anti‐GITR agonist antibody to describe the kinetics of intratumoral regulatory T cells (Tregs) and effector T cells (Teffs) in the TME of two syngeneic mouse tumor models.

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

The QSP model provides a quantitative understanding of the trade‐off between the complexity of maximizing Treg depletion versus Teff agonism, and offers insights to optimize drug design and dose regimen.

• HOW MIGHT THIS CHANGE DRUG DISCOVERY, DEVELOPMENT, AND/OR THERAPEUTICS?

The novel QSP model described in this report provides insights into an important class of molecules that have the dual pharmacology of immune agonism and effector function mediated cellular depletion. This approach can be used to achieve a dynamic balance of complexed immune responses that need to be understood to optimize drug design and dosing regimen.

INTRODUCTION

Glucocorticoid‐induced tumor necrosis factor receptor (TNFR)‐related protein (GITR) is a cell surface receptor constitutively expressed at high levels on regulatory T cells (Tregs) and at low levels on naïve and memory T cells and other immune cell types, such as natural killer cells (NK), and macrophages. GITR has been shown to inhibit Tregs and to extend the survival of T‐effector cells (Teffs) by serving as a co‐stimulatory immune checkpoint molecule. ¹ , ² The effect of crosslinking or targeting GITR on various cell types differs with the cell type and immunological context.

GITR ligand (GITR‐L) is expressed on many activated cells of the immune system as well as non‐immune cell types, such as endothelial cells. ³ Engagement of GITR by GITR‐L or an agonistic anti‐GITR antibody (an antibody able to bind GITR and mimic the agonistic activity of the ligand) has been shown to increase T cell proliferation and effector function, cytokine production, and cytolytic function, and protect T cells from activation induced cell death. ⁴ , ⁵ Studies in GITR transgenic mice suggest that GITR engagement may increase levels of memory CD4⁺ cells (CD44⁺/CD62L⁻). ⁶ Therefore, targeting GITR focuses the immune activation and memory T‐cell generation within the tumor microenvironment (TME) rather than systemically, thereby maintaining efficacy while reducing systemic toxicity and providing a novel option to enhance antitumor immunity in cancer treatment.

Studies utilizing a GITR antibody, DTA‐1, have shown compelling antitumor activity in syngeneic tumor models, and demonstrated that DTA‐1 agonism contributes to antitumor activity by promoting the activation and expansion of cytolytic CD8⁺ T cells and NK cells. ⁷ , ⁸ In tumor‐bearing animals, DTA‐1 treatment can regress established tumors depending upon CD8⁺ T cells and NK cells, and to a lesser extent CD4⁺ T cells. This immunity is long‐lived as mice are resistant to tumor rechallenge. ⁷ , ⁹ Preclinical data also suggest that Fc‐γ receptor (Fc‐γR)‐mediated depletion of Tregs in combination with agonism of GITR enhances Teff activation, resulting in an increased intratumoral Teff:Treg ratio and tumor shrinkage. ¹⁰ In addition, elimination of DTA‐1's FcγR interaction through the N297A mutation abrogates its antitumor activity in a mouse tumor model while not affecting GITR saturation properties in the tumor and draining lymph. ¹⁰ These results support the hypothesis that agonist antibodies targeting GITR require co‐engagement with activating FcγRs for their tumoricidal activities. Therefore, a detailed understanding of FcγR binding versus GITR binding properties and the resulting effects on both Tregs and Teffs will be essential for the design of anti‐GITR antibodies with an optimal efficacy and safety profile.

METHODS

Antibody production and treatment

DTA‐1.mIgG2a was generated by cloning the DTA‐1 hybridoma variable regions into a mouse IgG2a backbone and was produced from HEK293 cells (Invitrogen). Tumor‐bearing mice were typically treated 8–9 days after tumor inoculation (~70 mm³) by intraperitoneal (i.p.) injection of antibodies.

Mouse syngeneic tumor models

BALB/c female mice were obtained from Charles River Laboratory or Taconic with wild type and FcγRIIB−/− and Fc common γ chain−/− genotype. The BALB/c derived colorectal carcinoma cell line Colon26 was obtained from the DCTD Tumor Repository, National Cancer Institute, Frederick, MA. The mice were inoculated with subcutaneous injections of 5 × 10⁵ Colon26 cells onto the right flank. The A20 murine B lymphoma cell line was obtained from American Type Culture Collection. The mice were inoculated with subcutaneous injections of 1 × 10⁶ A20 cells onto the right flank.

All cells were tested and found to be free of mycoplasma and viral contamination in the IMPACT VIII PCR assay panel (RADIL, MU Research Animal Diagnostic Laboratory). Tumor diameter was measured by electronic caliper every 2–3 days and tumor volume was determined. All procedures were performed in accordance with the standards of the US Department of Health and Human Services and were approved by the Novartis Animal Welfare Committee.

In vitro and in vivo studies

Tumor dissociation and analysis by flow cytometry

Antibodies to detect specific cell types and receptors were purchased from Becton, Dickinson (BD) Biosciences, eBioscience, BioLegend, and Sino Biologicals. Single cell suspensions from excised tumors were obtained by mincing the tumors, treatment with collagenase, and treatment with red blood cell lysis buffer. Cells were stained by saturating doses of antibodies before incubation with fluorochrome conjugated antibodies and were counted using an LSR‐II flow cytometer (BD). The cell density in the tumor was calculated by dividing the number of cells of interest to the total number of cells extracted.

In vitro T cell stimulation assays

Splenocytes were stimulated with CD3 and CD28 specific antibodies (R&D systems) together with indicated concentrations of GITR specific antibodies. Cell proliferation was determined after 72 h of incubation with CellTiterGlo (Promega).

Pharmacokinetic/pharmacodynamic study

In Colon26 tumor bearing mice, serum DTA‐1.mIgG2a concentration, soluble GITR (sGITR; a target engagement biomarker) level in serum, and tumor sizes were measured following a single intravenous (i.v.) dose of 0.3, 1, 3, 5, 10 or 15 mg/kg of DTA‐1.mIgG2a as previously reported, ¹¹ and intratumoral Treg and Teff profiles were determined at the i.v. dose of 5 mg/kg.

In A20 tumor bearing mice, serum DTA‐1.mIgG2a concentration and sGITR level were measured after a single i.p. dose of 15 mg/kg of DTA‐1.mIgG2a, Treg and Teff profiles were determined at the doses of 1 and 5 mg/kg, and tumor sizes were measured after a single i.p. dose of 0.1, 1, 2.5, and 5 mg/kg.

Serum concentration of DTA‐1.mIgG2a was quantified by a liquid chromatography‐tandem accurate mass spectrometry assay using several peptide variants and stable isotype labeled internal standard. Anti‐drug antibodies (ADAs) were measured using biotinylated DTA‐1.mIgG2a as a capture antibody and an Alexa fluor 647 donkey anti‐mouse (H + L) as the detection antibody.

The level of sGITR in the serum was determined using an enzyme‐linked immunosorbent assay. Tumor sizes were measured three times a week and determined by using caliper measurements. Tumor volumes were calculated as (length × width × width)/2. For the analysis of tumor infiltrating immune cells, mice were euthanized and tumors were collected and dissociated on a GentleMACS Octo dissociator. The cell suspension was stained with a cocktail of antibodies and run on a LSR Fortessa cytometer to identify different immune populations.

Mathematical modeling

Quantitative systems pharmacology model development

A quantitative systems pharmacology (QSP) ¹² model was constructed to include pharmacokinetics (PK), sGITR, and cell activity as a result of treatment of tumor models with the drug, anti‐GITR antibody DTA‐1.mIgG2a (Figure 1). The model includes (1) absorption, distribution, metabolism, and elimination of DTA‐1.mIgG2a, (2) DTA‐1.mIgG2a binding, and (3) impact on Treg and CD8⁺T cells in the tumor. All drug molecules in the model are assumed to have a single valency. A simplified version of the intracellular GITR signaling cascade was implemented where the trimer complex consisting of the drug bound to GITR on the tumor cell and the FcγRIV receptors on the macrophages either lead to proliferation of the cell type or lead to initiation of antibody‐dependent cellular phagocytosis (ADCP) via the macrophages. To simplify the model, IL‐10 and IL‐2 are assumed to represent the larger sets of anti‐inflammatory and pro‐inflammatory cytokines, respectively.

FIGURE 1.

Model development for GITR mediated T cell dynamics. (a) The schematic of QSP model for GITR mediated T cell dynamics in mouse TME. (b) The chronology of events in the mouse TME following drug treatment. The weight of the arrows in (b) indicates the relative magnitude of the different conflicting processes in each panel for the best possible treatment outcome, the thicker arrows indicating a faster process while a thinner arrow indicating a slower process. The pharmacokinetics were modeled as an empirical two‐compartment model with central and peripheral compartments, and a tumor compartment which comprised of most of the cellular interactions. The tumor compartment consisted of interactions of Tregs, CD8+ (Teff) cells, and macrophages (FcγRIV) cells. Without drug, the tumor compartment Teff cells are arrested in an inactive state under IL‐10 mediated downregulation as indicated in the left panel in (b). In the center panel in b, upon drug administration, the Treg cells are depleted through ADCP, lifting the IL‐10 mediated downregulation and allowing Teff cells to be activated. In the right panel in b, the activated Teff cells under the influence of IL‐2, can now proceed to differentiate to CTL cells and clear the tumor, which is the intended outcome of the drug. ADCP, antibody‐dependent cellular phagocytosis; CTL, cytotoxic T lymphocyte; QSP, quantitative systems pharmacology; Teff, effector T cell; TME, tumor microenvironment; TMDD, target‐mediated drug disposition.

Model development workflow

The model was developed in stages with the PK model developed first, followed by the cellular model which involved the cellular interactions under disease conditions in the absence of the drug. ¹³ Figure S1 shows the sequence of model development and Figure 1a shows the model schematic with the twin mechanism of the action of the drug: proliferation of GITR+ cells and ADCP via macrophages. ¹⁴ , ¹⁵ The two modules were then ported together based on known rates of drug distribution to the tumor ³ and fit to data from Colon26 and A20 mice. Because ADA mediated clearance was observed in the mouse studies, a semi‐mechanistic ADA model was added to match the observed PK. ¹⁶ More details of the model development are described in Supplementary Information (Model development workflow and Additional modeling details).

The key parameters for the Colon26 and A20 model are shown in Table 1. A full set of parameters are provided in Tables S1 and S2. In brief, literature values were used to parameterize the CD8 expansion leading tumor killing. ¹⁷ , ¹⁸ , ¹⁹ These models differed in populations of tumor infiltrating lymphocytes (TILs), and Treg subtype (high or low GITR expressing). All other parameters were kept constant between the two models. Ordinary differential equations were used to describe all interactions. Molar quantities were converted to measurement units in the analysis to allow for improved model verification, such as confirming mass balance. The model was built in MATLAB SimBiology version 5.6 (Mathworks). The model calibration was done using an iterative approach. ¹³ A full list of model reactions and a Symbiology Project File is provided as Supplementary Information.

TABLE 1.

Calculation of GITR in the tumor for Colon26 and A20 mouse models.

GITR+ cell types in tumor	Average GITR expression	Cell numbers (cells/mm3) (Internal data)	Volume of cellular fraction of tumor (mm3)	Total cell numbers	Tumor receptor expression (pmol)
Colon26
Treg cell (FOXP3+)	189,000	116	63.46	7362	0.0023
B cell	16,500
CD4+ cell	60,000
CD8+ cell	41,000	105	63.46	6648	0.0004
NK cell	39,000
Macrophage cells	19,000	885	63.46	56,166	0.0018
Amount of GITR in tumor					0.0045 pmol
Measured volume of tumor					79.33 mm3
Cellular fraction of tumor (80% tumor volume)					63.46 mm3
Acellular fraction of tumor (20% tumor volume)					15.87 mm3
Effective concentration of GITR in tumor (amount/acellular volume)					0.285 nM
Effective concentration of GITR on Treg cells					0.15 nM
Effective concentration of GITR on CD8+ T cells					0.025 nM
Effective concentration of GITR on macrophage cells					0.11 nM
A20
Treg cell (FOXP3+)	189,000	922	101	92,904	0.029
B cell	16,500
CD4+ cell	60,000
CD8+ cell	41,000	935	101	94,248	0.0064
NK cell	39,000
Macrophage cells	19,000	770	101	77,582	0.0024
Amount of GITR in tumor					0.038 pmol
Measured volume of tumor					126.25 mm3
Cellular fraction of tumor (80% tumor volume)					101 mm3
Acellular fraction of tumor (20% tumor volume)					25.25 mm3
Effective concentration of GITR in tumor (amount/acellular volume)					1.5 nM
Effective concentration of GITR on Treg cells					1.15 nM
Effective concentration of GITR on CD8+ T cells					0.25 nM
Effective concentration of GITR on macrophage cells					0.095 nM

Abbreviation: Treg, regulatory T cell.

Cell types modeled

The major immune cells considered in the tumor were CD8 cells, Tregs, and FcγRIV expressing macrophages as effector cells. ¹⁴ , ¹⁵ The model was set to match the percentage of each cell type as measured in a representative tumor from the corresponding tumor model. Figure 1a indicates how the depicted cells in the model schematic correlate to model species. CD8 cells are counted as the sum of inactivated and activated Teff cells, and the cytotoxic T lymphocyte (CTL) cells. Treg cells are counted as the sum of high and low GITR expressing Treg cells.

Dataset preparation

The data used for modeling included serum concentrations versus time of DTA‐1.mIgG2a, serum sGITR in the Colon26, and A20 models. Published kinetic data on the modulation of intratumoral Treg and CD8 cell population from disaggregated tumors following administration of DTA‐1.mIgG2a, the same antibody as used in this study, was used to describe the immune cell kinetics. ¹⁰ The pool of GITR receptors on CD8 and Treg cells was computed from the number of receptors per cell, the number of TIL counts (in the tumor) and typical numbers of circulating lymphocytes (for systemic levels). ²⁰ , ²¹

Sensitivity analysis

A local sensitivity analysis was conducted to explore the parameter space centered on the Colon26 and A20 nominal parameters to understand how individual parameters impacted the numbers of active and inactive CD8⁺ cells, high and low GITR expressing Treg cells and the CTL cells (see Supplementary Information for details).

RESULTS

Model calibration to pharmacokinetic/pharmacodynamic study

The model was calibrated to the single dose PK study in Colon26 mice across all dose levels. The model is able to describe the nonlinear PK, the higher observed clearance at lower doses (0.3 and 1 mg/kg) due to target‐mediated drug disposition (TMDD; Figure 2a,b). The main parameter that describes this accelerated clearance is the GITR concentration expressed by peripheral blood cells (Table S3). It was assumed that GITR expression in the peripheral blood was the same for the Colon26 and the A20 models.

FIGURE 2.

Model fitting for single‐dose PK data from the Colon26 mice (intravenous doses of 0.3–15 mg/kg) (a) and A20 mice (i.p. dose of 15 mg/kg) (b). The Colon26 model informed TMDD parameterization in the model while the A20 model informed first order absorption of the drug. For the 0.3 and 1 mg/kg doses, there was a significant discrepancy between C _max observed compared to the other doses. For these two doses, solid lines are simulations at the nominal dose. Dashed lines are simulations where an estimated dose was used to better match the observed C _max. As it was not clear the source of this discrepancy, the nominal and the estimated dose were carried forward to the other simulations. C _max, maximum concentration; TMDD, target‐mediated drug disposition.

As shedding of GITR is part of the process of activating CTLs, sGITR is a potential biomarker of activation. Our model was able to describe the net increase observed in total sGITR, which is due to the formation of sGITR‐antibody complex (Figure S2). However, a steep decline of sGITR by day 7 was observed. It is likely due to the impact of ADAs on the clearance of the complex, therefore, an empirical ADA model was added. The model was in general agreement with the overall sGITR profiles including the decline phase. However, due to the highly variable timing of ADA development, the model was not able to exactly match with the timing of the individual dose groups (Figure S2).

Treatment of the tumors with DTA‐1.mIgG2a resulted in a strong reduction in Tregs for the Colon26 model and an increase for the A20 model. These changes were observed only in the tumor with no significant change observed in draining lymph nodes. The change in Treg populations were monitored for 7 days for the Colon26 and A20 models. For the Colon26 tumor model, treatment with 5 mg/kg DTA‐1 showed a decrease in Treg cells to 5% by day 5 followed by a slight recovery by day 7 (Figure 3a,b). For the A20 model, treatment with 5 mg/kg DTA‐1.mIgG2a showed continual Treg increase until day 7, whereas, at 1 mg/kg DTA‐1, Treg increase stabilized at day 3 (Figure 3c,d).

FIGURE 3.

Model fitting for CD8 and Treg cells in tumors following single‐dose DTA‐1.mIgG2a treatment in Colon26 (intravenous dose) mice (a, b) and A20 (i.p. dose) mice (c, d). Dashed lines are simulations where an estimated dose was used to better match the observed pharmacokinetic data (Figure 2). The dots and whiskers represent the observed concentration and the 95% confidence intervals. Treg, regulatory T cell.

The model describes two phases of the CD8⁺ T cell profile, an initial decline due to ADCP of the CD8⁺ T cells, followed by a rapid expansion of the CD8⁺ T cells into CTLs. It describes the CD8⁺ T cell profile in the Colon26 model well, but not in the A20 model (Figure 3b,d). In the Colon26 model with 5 mg/kg DTA‐1.mIgG2a, a slight reduction in CD8⁺T cells was observed until day 5 followed by a rapid increase at day 7. In comparison, in the A20 tumor for a dose of 1 mg/kg, the CD8⁺ T cells decline briefly until day 3 and rebound at day 7. Whereas at 5 mg/kg, the CD8⁺ T cells slightly increase until day 3 before increasing rapidly at day 7, which is not predicted by the model. The reason for this disagreement is not clear, but one hypothesis is that both CD8⁺ cells and Treg cells in the A20 model may have different distribution of GITR expression compared to the Colon26 model.

Trade‐off between maximizing Treg depletion and CD8 ⁺ T cell expansion

The magnitude of CD8⁺ proliferation and CTLs produced in response to treatment with a GITR agonist antibody depends on the balance of three factors that are influenced by drug treatment: (1) removal of high GITR expressing Tregs leading to lowering of IL‐10 and release of CD8⁺ from the state of inactivity, (2) proliferation of CD8⁺ cells by GITR stimulation and IL‐2 activity, and (3) removal of CD8⁺ cells through ADCP. The interplay of these three factors is illustrated in Figure 1a,b. If the dose is too low, enough high GITR expressing Tregs are not removed, leading to the enough anti‐inflammatory activity that CD8⁺ cells remain in an inactive state. If the dose is too high, too many CD8⁺ cells are depleted through ADCP resulting in a smaller pool of CD8⁺ cells differentiating to form CTL cells. Figure 3 shows model simulations overlayed on intertumoral Treg and CTL cells. In both models, there appears to be a threshold dose that leads to CTL induction. This suggests there exists an optimum dose at which the anti‐inflammatory activity is suppressed to the extent that a suitable number of CD8⁺ cells differentiate into CTLs.

Prediction of CTL production at tumor clearing doses

Based on the assumption of the maximum number of tumor cells one CTL can kill, a minimum number of CTL cells needed to be produced to clear the tumor can be determined, allowing the comparison of model prediction with observed tumor growth inhibition across drug doses, even though the model does not directly predict tumor clearance. In particular, the doses predicted to produce sufficient (or insufficient) number of CTLs to result in tumor clearance should agree with the observed tumor clearance (or lack of tumor clearance). Hence, the agreement (or lack of it) was used as a measure of model validation.

Using the size of the tumor, density of cells in the tumor, and typical CTL:tumor cell kill ratio, we estimated the minimal number of CTLs required to clear the tumor for the Colon26 and the A20 models (Table 2). Doses that are predicted to produce more than 0.79 × 10⁶ and 13 × 10⁶ CTLs per tumor are predicted to fully clear the tumor in the Colon 26 model and A20 model, respectively.

TABLE 2.

Calculation of the minimum number of CTLs needed for tumor clearances in Colon26 and A20 models

Tumor model	Parameter description	Value	Unit	References
Colon26	Volume of Colon26 tumor	7.9E−2	cm3	Internal data
	Total number of all cells in tumor	7.9E6	Cells	Del Monte 21
	Peak number of CTL cells needed to completely clear tumor cells	0.79E6	Cells	Assume 10 cells killed/CTL
A20	Volume of A20 tumor	0.13	cm3	Internal data
	Total number of all cells in tumor	130E6	Cells	Del Monte 21
	Peak number of CTL cells needed to completely clear tumor cells	13E6	Cells	Assume 10 tumor cells killed per CTL

Abbreviation: CTL, cytotoxic T lymphocyte.

Next, a range of doses was simulated in each tumor model from 0.1 to 15 mg/kg and the peak number of CTLs was predicted (Table 3). The optimal dose was determined as the minimum dose which would generate a sufficient number of CTLs to kill all of the tumor cells in the Colon26 and the A20 models. To validate these predictions, we compared the doses that were predicted to be tumor clearing to efficacious doses in the Colon26 and A20 models. At doses over greater than 1 mg/kg in the Colon26 and greater than 5 mg/kg in the A20 model, a sufficient number of CTLs to fully clear the tumor is predicted by the model. In the Colon26 model, these predictions are consistent with the doses that were observed to clear the tumor (Figure 4), suggesting that the model can quantitatively capture the degree of CTL expansion. In the A20 mouse model, the computational model predicts that 1 mg/kg should be sufficient, whereas only limited tumor killing is seen at that dose, however, at the next dose evaluated experimentally (5 mg/kg), the computational model is in good agreement with the observed data of tumor inhibition.

TABLE 3.

Prediction of CTLs produced following DTA‐1.mIgG2a treatment in Colon26 and A20 models.

Tumor model	Predicted minimal number of CTLs needed to clear tumor (106 cells)	DTA‐1.mIgG2a dose (mg/kg) a	Complete tumor growth inhibition observed b	Predicted peak CTLs (106 cells)
Colon26	0.79	0.3	No	7.8E−4
		1	Yes	1.39
		3	Yes	0.86
		5	Yes	0.84
		10	Yes	0.83
		15	Yes	0.81
A20	13	0.1	No	6E−3
		0.3	No data	8.6E−3
		1	No	22
		3 c	No data	11
		5	Yes	12
		10	No data	12
		15	No data	12

Abbreviation: CTL, cytotoxic T lymphocyte.

^a Single‐dose as intravenous for Colon26 mice and i.p. for A20 mice.

^b Observed data are presented in Figure 4.

^c 2.5 mg/kg was studied and complete response was not observed.

FIGURE 4.

Observed tumor volume versus time profiles following single‐dose DTA‐1.mIgG2a treatment in Colon26 mice (intravenous dose) (a) and A20 mice (i.p. dose) (b).

In addition, the model provides an explanation for the sharp dose response observed in the studies. Low doses are insufficient to trigger the positive feedback loop (the production of IL‐2 by activated T‐cells leading to the further activation). However, the first dose level that crosses this activation threshold leads to a full response. Similarly, in the A20 model, with different Treg dynamics, a sharp dose response is also observed (Table 3; Figure 4).

Sensitivity analysis

Sensitivity analysis was conducted to understand which parameters have the largest influence on the outputs and the results are shown as heat maps (Figure S2). For both models, the formation of the drug:GITR:FcγRIV trimeric complex was the most sensitive process, followed by tumor drug disposition and drug elimination. The formation of the trimeric complex is the key step both in receptor clustering leading to GITR activation, but also in FcγRIV clustering leading to ADCP‐driven cell depletion. The other sensitive parameters included initial number of macrophages in the tumor, the GITR and FcγRIV expression on the macrophages, and the endocytosis of the GITR receptor. These suggest that GITR expression as well as the presence of FcγRIV bearing immune cells could be biomarkers of response to anti‐GITR antibody, and or inform optimal dosing regimens for individuals or populations.

DISCUSSION

Despite the very promising results in the clinic, many patients still do not respond to anti‐programmed death‐1 (PD‐1) or anti programmed death‐ligand 1 (PD‐L1) blockade, and thus it is important to investigate alternative ways to improve the objective response rates in patients with advanced cancer or to patients progressing after anti‐PD1 or anti‐PD‐L1 therapy. Antibodies against GITR, with the potential for Treg depletion as well as Teff costimulation, may be focused within areas in or near the tumor and thus could potentially achieve this objective. Understanding these dual pharmacology drugs with a direct immunomodulatory effect combined with an indirect effect through cell depletion requires a thorough comprehension of the balance of activating T cells through receptor crosslinking while depleting of Treg cells. The goal is to promote cytotoxic effector T cell generation and to dampen the immunosuppressive effects by depleting FoxP3⁺ CD4⁺ T regulatory cells. ⁴ , ²² , ²³ Thus, the success of anti‐GITR antibodies depends on achieving a coordination of activating T cells through receptor crosslinking and depleting T cells through ADCP. However, these two different mechanisms may have differential effects on individual cell types based on the relative expression of the target as well as the availability of effector cells, and take place on distinct time scales. Furthermore, once CD8⁺ T cells have been sufficiently activated, they produce proinflammatory cytokines which create a positive feedback loop. This makes it challenging to design optimal drug and dose regimens.

We have shown here that a QSP model can be a useful tool in achieving this objective and leveraging a mixture of in vitro and in vivo data to understand the pharmacology of DTA‐1 in two different murine tumor models. Although not representing the full complexity of the known biology, our model was able to provide insight into the key mechanisms at play in the antitumor response. The model was able to describe the PK of the drug and the dynamic profiles of sGITR, Treg, and CD8⁺ T cells over the observed time, whereas also predicting the required number of CTLs to kill the tumor. Ultimately, the goal of this type of modeling is to bridge from preclinical studies to clinical studies.

DTA‐1 has been used extensively in the literature to modulate GITR in mouse through GITR agonism. It has been showed to promote regression of mouse tumors and induce long term antitumor immunity. ¹⁰ It is also known to induce Treg depletion through Fc/FcγR interaction, boosting CD8⁺/Treg ratio, and aiding clearance of the tumor. However, being a GITR agonist, DTA‐1 also stimulates CD8⁺ T cell proliferation. To our knowledge, it has not been conclusively proven whether Treg depletion or CD8⁺ T cell proliferation is the determining factor controlling change in CD8⁺/Treg ratio. Our model suggests that Treg cell depletion is necessary initially to reduce the immunosuppressant effects of IL‐10 cytokines. Once the tumor immune system recovers from the effects of IL‐10 via Treg cell depletion, the CD8⁺ T cells recover to an active state before having multiple proliferation and differentiation steps to form CTL cells. Thus, the CD8⁺/Treg ratio changes are based on Treg depletion occurring first. One of the goals of this work was to understand the differences in the anti‐tumor response in different mouse tumor models. To account for the observed behavior, we hypothesized that there are two populations of Treg cells, one with high GITR expression and the other with low GITR expression. The Tregs with high GITR expression were assumed to be high enough to trigger ADCP on drug binding, whereas the Tregs with low GITR expression were assumed to be too low to activate ADCP. Furthermore, the Colon26 tumor was considered to mostly have high GITR expressing Tregs, whereas the A20 tumor was considered to have mostly low GITR expressing Tregs. These assumptions aided simulation of different Treg profiles in Colon26 and A20 models. However, it remains as a hypothesis and would need experimental verification.

The GITR agonists that have been studied in phase I clinical trials for treatment of advanced solid tumors showed minimal efficacy as monotherapy but more promising as combination therapies, such as combining with PD‐1 inhibitors. ¹¹ , ²⁴ , ²⁵ , ²⁶ , ²⁷ , ²⁸ Therefore, further understanding and quantifying the mechanisms in humans is an important next step to inform clinical study design and selection of combination therapy. The QSP model is mechanistic and many of the parameters are have been directly measured experimentally. Using the parameters measured in humans with the consideration of multiple tumor types in humans, such as the number of Teff and Treg cells, the model could be extended to simulate clinical intervention. Because our model quantitatively characterized the complexed relationship of GITR agonist and antagonist mechanisms, which are on many of the same cell types impacted by PD‐1 inhibitors, it could be used as the basis for a human QSP model to explore the viability of GITR treatment in conjunction with checkpoint inhibitors such at PD‐1 antibodies. Additional cancer immune pathways can be included in the model to further characterize the pathway inter‐plays and thus to inform the selection of novel combinations therapy in clinical trials, or to select patients who are more likely to respond. Because of the complexity and species differences of the positive and negative feedback loops and GITR expression profiles, ²⁹ a straightforward extrapolation of the mouse to human pharmacodynamic relationship would not be expected to yield success without the mechanistic model.

In conclusion, our model is the first QSP model that provides a mechanistic characterization of anti GITR therapy. The model helps to understand the dynamic balance of complexed immune responses that are needed to produce an effective anti‐tumor response. More broadly, this model provides insights into an important class of molecules that have the dual pharmacology of immune agonism and effector function mediated cellular depletion. Therefore, it can be used in combination with models of other immunotherapy agents to understand and optimize combination therapies.

AUTHOR CONTRIBUTIONS

Y.J., K.M., J.F.A., B.G., and D.A.K. wrote the manuscript. Y.J. and B.G. designed the research. D.A.K., K.M., Y.J., J.F.A., L.G., and B.G. performed research. K.M. and J.F.A. analyzed data. J.M.B. contributed new reagents/analytical tools.

FUNDING INFORMATION

This study was funded by Novartis Pharmaceuticals Corporation. Applied BioMath performed the modeling work under contract with Novartis.

CONFLICT OF INTEREST STATEMENT

Yan Ji and Deborah Knee are employees and stockholders of Novartis. Joshua F. Apgar and John M. Burke are Cofounders of Applied BioMath. Bruce Gomes was an employee of Novartis at the time of this work. Kumpal Madrasi and Lore Gruenbaum were Applied BioMath employees at the time of this work.

Supporting information

Appendix S1

Ji Y, Madrasi K, Knee DA, et al. Quantitative systems pharmacology model of GITR‐mediated T cell dynamics in tumor microenvironment. CPT Pharmacometrics Syst Pharmacol. 2023;12:413‐424. doi: 10.1002/psp4.12925