Dose optimization of ClpP agonists using an in vitro microfluidic perfusion platform and in silico pharmacokinetic-pharmacodynamic modeling

Abstract

Small molecule activators of the mitochondrial caseinolytic protease P (ClpP agonists) can disrupt tumor metabolism and deprive tumors of their energy needs. The imipridone, ONC201, is a ClpP agonist currently undergoing clinical evaluation across multiple cancer types, while additional analogs with improved potency and selectivity are in preclinical development. Preclinical studies in mice have demonstrated a unique pharmacokinetic-pharmacodynamic (PK-PD) relationship for ONC201 characterized by prolonged pharmacology following a single dose. This motivated the selection of an initial human dosing regimen of every three weeks, and subsequent dose exploration studies in mice led to dose intensification in human patients. However, a systematic analysis of ClpP agonist PK-PD relationships has not been performed, and the optimal exposure profile for ClpP agonists remains undefined. To address this gap, we combined PK-PD modeling with a microfluidic perfusion platform as an animal-alternative approach for translational PK-PD of ClpP agonists. We demonstrate that the anti-proliferative effect on triple negative breast cancer cells correlates with the magnitude and duration of ClpP agonist exposure above a threshold concentration required for ClpP activation. Moreover, we demonstrate that PK-PD model simulations using parameters derived from microfluidic perfusion datasets can successfully predict the anti-tumor efficacy of a ClpP agonist in a mouse tumor xenograft study. These studies support the translational relevance of the animal-alternative in vitro PK-PD platform and its utility to help guide dose optimization of ClpP agonists as cancer therapeutics.

INTRODUCTION

The mitochondrial caseinolytic protease P (ClpP) is an ATP-dependent protease located within the mitochondrial matrix that has emerged as a promising anticancer drug target [1–4]. The ClpP protease is part of the ClpXP protein complex formed with the caseinolytic mitochondrial matrix peptidase chaperone subunit X (ClpX) which recognizes protein degrons and serves as a chaperone to target mitochondrial proteins for degradation by the ClpP protease. By degrading misfolded or damaged proteins, ClpP is believed to play an important role in maintaining mitochondrial function. Additionally, it plays a key role in the mitochondrial unfolded protein response which helps with recovery from damage or other metabolic stress. ClpP overexpression has been identified in multiple cancer types and is associated with poor prognosis in breast cancers [5].

The first small molecule ClpP agonists, acyldepsipeptide antibiotics (ADEPs), were designed as antibacterial agents owing to their ability to increase bacterial ClpP activity and trigger bacterial cell death [6]. Subsequent studies demonstrating that ADEPs also bind to human ClpP began to uncover the potential role for ClpP agonists in treating human cancers [7]. The most advanced ClpP agonist compound for the treatment of cancer is the imipridone-based compound, ONC201 (previously known as TIC10), which is currently undergoing clinical evaluation in multiple cancer types [8, 9]. It was discovered from a screen for induction of tumor necrosis factor-alpha-related apoptosis-inducing ligand (TRAIL) in certain cancer cell lines, and it has been proposed to serve as an antagonist of the D2 dopamine receptor [10]. Emerging evidence suggests that its primary mechanism as an anti-cancer therapeutic may be related to its function as a ClpP agonist [11, 12]. This triggers the integrated stress response through induction of activating transcription factor 4 (ATF4), leading to disruption of mitochondrial function and inhibition of oxidative phosphorylation.

ONC201 has demonstrated growth inhibitory properties across various cancer cell lines and tumor growth inhibition in multiple mouse tumor models [9, 13–15]. Its broad preclinical efficacy and favorable safety profile led to its advancement into clinical evaluation across a variety of tumor indications as a single agent and in combination regimens [16, 17]. Based on the observation that anti-tumor activity was observed in mice after a single dose and the pharmacodynamic (PD) effects appeared to extend well beyond the pharmacokinetic (PK) exposure profile, the recommended Ph II dose (RP2D) was 625 mg orally every three weeks [8, 9]. Subsequent preclinical investigations demonstrated that ONC201 exhibited dose- and schedule-dependent effects on tumor progression in mouse tumor models, and near-maximal anti-tumor efficacy was observed with weekly dosing of ONC201 [18]. This led to modification of the clinical protocol to also evaluate weekly dosing with one or two consecutive daily doses in both adult and pediatric patients [19–21]. Additional small molecule ClpP agonists have since been developed with improved potency relative to ONC201, although none of the compounds has been used for extensive dose optimization. One of the most potent ONC201 analogs is TR-107, which has improved ClpP agonist potency and low nM anti-proliferative effects in multiple cancer cell lines [3]. To facilitate the application of ClpP agonists as cancer therapeutics, it is critical to understand the PK-PD relationship connecting the magnitude and duration of ClpP agonism to downstream pharmacological effects.

Successful drug discovery and development requires a clear understanding of these PK-PD relationships to inform compound design and to optimize dosing regimens [22]. A recent report described the use of virtual PK-PD modeling and machine learning to define PK drivers of efficacy for drug discovery programs [23]. This work importantly cautioned against generically defining PK drivers; instead, these must be evaluated for each pharmacological target. Moreover, a single simple PK metric (e.g., average concentration) is often not adequate to describe the pharmacological response which can be driven by a more nuanced combination of exposure magnitude and duration. While such model-based predictions are invaluable for drug discovery and development, it is equally important to test model predictions experimentally. Unfortunately, the current preclinical toolbox of experimental in vitro and in vivo assays is limited in its ability to systematically explore these PK-PD relationships, and preclinical drug testing in animals is pursued despite known deficiencies because of a lack of suitable alternatives [24]. To address the limitations of preclinical animal models for establishing PK-PD relationships for experimental therapeutics, we and others have developed microfluidic perfusion systems that enable evaluation of cell responses to precisely controlled drug exposure profiles [25–30]. The input drug exposure profiles can comprise steps of different durations or dynamic profiles that mimic the actual in vivo PK in species such as mice or humans. Unlike traditional preclinical animal experiments, the in vitro PK-PD studies avoid PK variability, unanticipated toxicity limitations, and the need to find compounds suitable for in vivo administration. Here, we apply custom microfluidic perfusion systems in combination with PK-PD modeling to delineate the exposure-response relationships for ClpP agonists and to guide dose optimization for the treatment of triple negative breast cancer (Fig. 1).

Figure 1.

Overview of the translational pharmacokinetic-pharmacodynamic (PK-PD) workflow combining an in vitro microfluidic perfusion platform with in silico PK-PD modeling. The desired drug PK profile is applied via perfusion of cells grown in a microfluidic channel slide, and PD response is captured through live-cell imaging of cell growth or biomarker dynamics. The resulting data are used to parameterize a PK-PD model, which then provides additional predictions for testing in the microfluidic perfusion platform in an iterative manner. Portions of the figure were created with BioRender.com.

MATERIALS AND METHODS

Chemicals

TR-107 was provided by Madera Therapeutics (Cary, NC) or purchased from Selleck Chemicals LLC (Houston, TX, USA). ONC201 was purchased from Cayman Chemical (Ann Arbor, MI, USA). Tetramethylrhodamine methyl ester (TMRM) perchlorate and Hoechst 33342 were purchased from MedChemExpress (Monmouth Junction, NJ, USA).

Cell Culture

MDA-MB-231 cells constitutively expressing red fluorescent protein (MDA-MB-231-RFP) under a CMV promoter were purchased from GenTarget (San Diego, CA, USA). Parental MDA-MB-231 cells were purchased from the Tissue Culture Facility within the Lineberger Comprehensive Cancer Center at the University of North Carolina at Chapel Hill. Cells were cultured in DMEM (Fisher Scientific, Waltham, MA, USA) supplemented with 10% fetal bovine serum (Sigma, St. Louis, MO, USA) and 1% penicillin-streptomycin (Fisher Scientific, Waltham, MA, USA), and they were maintained in a humidified incubator at 37°C and 5% CO₂.

Static Proliferation Assays

MDA-MB-231-RFP cells were added to clear-bottom, tissue culture-treated, black 96 well plates (Sigma, St. Louis, MO, USA) at a density of 2,000–3,000 cells per well in 100 μL media (DMEM + 10% fetal bovine serum + 1% penicillin-streptomycin). Cells were allowed to adhere overnight at 37°C and 5% CO₂ within a humidified incubator. Drug dilutions in 100 μL media were added to each well to achieve the final desired drug concentration in 200 μL total media. The plate was positioned on an AutoLCI automated live cell imaging system (World Precision Instruments, Sarasota, FL, USA) within the incubator and serial brightfield and red fluorescence images were acquired every 12 h for 1 week. The percent confluency of cells in each image was determined using the AutoLCI Analysis App (World Precision Instruments, Sarasota, FL, USA) and imported into GraphPad Prism (Dotmatics, Boston, MA, USA) for analysis.

Live-Cell Imaging of Changes in Mitochondrial Membrane Potential

MDA-MB-231 cells were added to clear-bottom, tissue culture-treated, black 96 well plates (Sigma, St. Louis, MO, USA) at a density of 4,000 cells per well in 200 μL media (DMEM + 10% fetal bovine serum + 1% penicillin-streptomycin). After overnight incubation at 37°C and 5% CO₂ within a humidified incubator, the media was removed and replaced with 100 μL of phenol red-free DMEM (Fisher Scientific, Waltham, MA, USA) supplemented with 10% fetal bovine serum and 1% penicillin-streptomycin along with 20 nM TMRM and 10 ng/mL Hoechst 33342. Following an additional incubation for 1 h at 37°C and 5% CO₂ within a humidified incubator, an additional 100 μL of media was added to each well containing 20 nM TMRM, 10 ng/mL Hoechst 33342, and 2X the desired final concentration of the ClpP agonist (TR-107 or ONC201). The plate was imaged longitudinally using an Olympus IX83 inverted fluorescence microscope with a motorized stage (Evident Scientific, Waltham, MA, USA) containing a stage top incubator (PeCon, Erbach, Germany) to maintain a humidified 5% CO₂ atmosphere. The microscope was maintained within a custom environmental enclosure heated to 37°C. Brightfield and fluorescence images were acquired at each timepoint with a UPLXAPO20X objective using a DAPI filter set for Hoechst 33342 fluorescence and a Texas Red filter set for TMRM fluorescence. Images were analyzed using ImageJ. Background subtraction in each channel was performed with a rolling ball radius of 50. For nuclear counts, the Hoechst 33342 channel was converted to a binary image using the Triangle method, and the Analyze Particles function was used to quantify the number of cells per image. For the mitochondrial membrane potential (MMP), the average signal in the TMRM channel was calculated for each image. The average MMP was divided by the total nuclear count in each image to determine the average normalized MMP per cell as a function of time for each treatment condition.

Microfluidic Perfusion Setup: Type I

Type I microfluidic perfusion systems were designed to enable automated switching of drug exposure to cells growing in a μ-Slide VI 0.4 microfluidic channel slide (Ibidi USA, Fitchburg, WI, USA). Fluid flow was controlled by pressurization of input reservoirs (Duran PressurePlus glass bottles or polypropylene conical tubes) containing microfluidic reservoir caps (PreciGenome, San Jose, CA, USA) using FlowEZ pressure controllers (Fluigent, Le Kremlin-Bicetre, France) connected to an input pressure source of a 5% CO₂ gas mixture. Drug and media inputs were connected to a SwitchEZ valve (Fluigent, Le Kremlin-Bicetre, France) that was controlled using the OxyGEN software (Fluigent, Le Kremlin-Bicetre, France) or the Fluigent SDK for MATLAB (MathWorks, Natick, MA, USA). The fluidic path connecting the input reservoirs and the microfluidic channel slide included 1/16” OD × 0.02” ID FEP tubing (IDEX Health & Science, Oak Harbor, WA, USA) and a segment of 1/16” OD × 0.0025” ID PEEK resistor tubing (IDEX Health & Science, Oak Harbor, WA, USA) to provide the desired pressure drop as a function of flow rate. Target operating parameters were 500 μL/h total perfusion flow rate with an applied pressure of ~3–4 psi. The outlet of the microfluidic channel slide was connected by 1/16” OD × 0.02” ID FEP tubing to 250 mL glass bottles (Fisher Scientific, Waltham, MA, USA). Functional qualification of the input step exposure profile programs at the selected flow rates was confirmed through quantification of the relative concentration vs. time profile in the perfusate of a blue food coloring dye to serve as a drug surrogate. Perfusate samples were routed through an Agilent G1315B diode array detector with a UV-Vis flow cell (Agilent, Santa Clara, CA, USA). Absorbance values at 630 nm were recorded every 0.8 seconds.

Microfluidic Perfusion Setup: Type II

Type II microfluidic perfusion systems were designed to enable automated mixing of input drug and media mixtures as a function of time to mimic physiological drug PK profiles. The setup was identical to the Type I microfluidic perfusion systems except no SwitchEZ valves were included in the fluidic path. Instead, modulation of drug and media input flow rates was achieved through modulation of the input reservoir pressures using the FlowEZ pressure controllers. Desired input drug PK profiles were converted to flow rate ratios of the drug and media inputs using a MATLAB script, and the desired flow rate setpoints were achieved through control of the corresponding drug and media reservoir pressure inputs. To reproduce drug PK profiles with high fidelity, typical time steps between pressure changes were 5–15 min. The Fluigent SDK for MATLAB allowed direct control of the FlowEZ pressure controllers through the MATLAB script. Functional qualification of the input PK profile program to achieve the desired output PK profile was confirmed through quantification of the relative concentration vs. time profile in the perfusate of a blue food coloring dye to serve as a drug surrogate. Perfusate samples were collected in 96-well plates at 15-min intervals using an automated fraction collector (BioRad, Hercules, CA, USA). Absorbance values at 630 nm for each fraction were determined using a SpectraMax M2 microplate reader (Molecular Devices, San Jose, CA, USA).

Microfluidic Perfusion In Vitro PK-PD Studies

MDA-MB-231-RFP cells were loaded into the microfluidic slides at a seeding density of 5×10⁴-1×10⁵ cells/mL according to the manufacturer’s instructions. Cells were allowed to adhere overnight at 37°C and 5% CO₂ within a humidified incubator. Input reservoirs containing media alone or media plus drug at the desired concentration were prepared one day prior to treatment initiation and allowed to equilibrate overnight at 37°C and 5% CO₂. The Type I or Type II microfluidic perfusion system was first flushed with media and then connected to the microfluidic slide containing the cells. The entire fluidic path of the perfusion system was placed within a humidified incubator and maintained at 37°C and 5% CO₂ for the duration of the study. The microfluidic slide was positioned on an AutoLCI automated live cell imaging system which was also contained within the incubator. Serial brightfield and red fluorescence images were acquired every 8 h for 1 week. The percent confluency of cells in each image was determined using the AutoLCI Analysis App and imported into GraphPad Prism for analysis.

PK-PD Modeling

A PK-PD model was developed in MATLAB SimBiology (MathWorks, Natick, MA, USA). Input drug PK profiles for the Type I perfusion systems were modeled as step functions:

$$\mathit{\text{Conc}}\left(t\right)=\left\{\begin{array}{c}{\mathit{\text{Conc}}}_{\mathit{\text{interval}}}\mathit{\text{if}}0\le t\le t_{\mathit{\text{interval}}}\\ 0\mathit{\text{if}}t>t_{\mathit{\text{interval}}}\end{array}\right.$$ (1)

where t is the time (h), t_interval is the duration of the desired exposure interval (h), Conc is the concentration of drug at time t (nM), and Conc_interval is the desired concentration of drug during the step exposure interval. Input drug PK profiles for the Type II perfusion systems were modeled as a one-compartment PK profile with first-order absorption:

$$\frac{d{\mathit{\text{Drug}}}_{\mathit{\text{oral}}}}{dt}=-k_a\cdot {\mathit{\text{Drug}}}_{\mathit{\text{oral}}}$$ (2)

$$\frac{d\mathit{\text{Drug}}}{dt}=k_a\cdot {\mathit{\text{Drug}}}_{\mathit{\text{oral}}}-\frac{CL}{V}\cdot \mathit{\text{Drug}}$$ (3)

where Drug_oral is the amount of drug administered orally (micromole), Drug is the amount of drug in plasma (micromole), k_a is the first-order absorption rate constant after oral administration, CL is the drug clearance (L/h), and V is the drug volume of distribution (L). The CL and V parameters represent apparent values (CL/F and V/F) which are normalized by the unknown bioavailability (F) of the drug compound. Weight-based drug doses for in vivo mouse studies were converted to dose amounts using a body weight of 0.025 kg. Drug concentrations in plasma are defined as:

$$\mathit{\text{Conc}}=\frac{\mathit{\text{Drug}}}{V}$$ (4)

Unbound drug concentrations in plasma or cell culture media were calculated using the following equations:

$$f_{u,\mathit{\text{med}}}=\frac{\mathit{\text{Dilution}}}{\left(\frac{1}{f_{u,pl}}-1\right)+\mathit{\text{Dilution}}}$$ (5)

$${\mathit{\text{Conc}}}_u=\mathit{\text{Conc}}\cdot f_u$$ (6)

where f_u,med is the fraction unbound in cell culture media (dimensionless), f_u,pl is the fraction unbound in plasma (dimensionless), Dilution is the dilution factor for % serum (i.e., 10 for a dilution to 10% serum-containing media; dimensionless), and Conc_u is the unbound concentration of drug in plasma or media (nM). These PK exposure profiles were linked to the PD effects on cell proliferation through an indirect response model:

$$\frac{d\mathit{\text{Cells}}}{dt}=k\mathit{\text{growth}}\cdot \mathit{\text{Cells}}\cdot \left(1-{\left(\frac{\mathit{\text{Cells}}}{\text{max}\_\mathit{\text{cells}}}\right)}^{n\_\mathit{\text{growth}}}\right)\cdot (1-I\mathit{\text{max}}\cdot \mathit{\text{Effect}}4)$$ (7)

$$\frac{dEf\mathit{\text{fect}}1}{dt}=k\mathit{\text{transit}}\cdot \left(\mathit{\text{Effect}}4-\mathit{\text{Effect}}1\cdot \left(\frac{{\mathit{\text{Conc}}}_u^{n\_PD}}{EC{50}^{n\_PD}+{\mathit{\text{Conc}}}_u^{n\_PD}}\right)\right)$$ (8)

$$\frac{\mathit{\text{dEffect}}2}{dt}=k\mathit{\text{transit}}\cdot \left(\mathit{\text{Effect}}1\cdot \left(\frac{{\mathit{\text{Conc}}}_u^{n\_PD}}{EC{50}^{n\_PD}+{\mathit{\text{Conc}}}_u^{n\_PD}}\right)-\mathit{\text{Effect}}2\right)$$ (9)

$$\frac{\mathit{\text{dEffect}}3}{dt}=k\mathit{\text{transit}}\cdot (\mathit{\text{Effect}}2-\mathit{\text{Effect}}3)$$ (10)

$$\frac{\mathit{\text{dEffect}}4}{dt}=k\mathit{\text{transit}}\cdot (\mathit{\text{Effect}}3-\mathit{\text{Effect}}4)$$ (11)

where Cells is the number of cells, kgrowth is the first order growth rate of the cells (1/h), max_cells is the maximum number of cells for the logistic growth equation, n_growth is an exponent to modify the shape of the growth curve (dimensionless), Imax is the maximum inhibition of cell growth rate by the drug (dimensionless), ktransit is the first order rate constant for the propagation of the drug effect on cell growth rate (1/h), Effect(1–4) are the transit compartments for the propagation of the drug effect on cell growth rate (dimensionless), Conc_u is the unbound concentration of drug in plasma or media (nM), EC50 is the unbound drug concentration resulting in half-maximal propagation of the drug effect on cell growth rate (nM), and n_PD is the Hill coefficient for the PD effect of drug concentration on cell growth rate (dimensionless). The initial value of the Effect1 compartment was set to 1 while the initial values of Effect2, Effect3, and Effect4 were set to 0. The initial number of cells was defined from the measured percent confluency value at t = 0 h for each imaging position for in vitro studies or from the initial tumor volume for each treatment group for the in vivo mouse tumor xenograft study. Conversion between cell number and the measured outputs of percent confluency or tumor volume was achieved using the following equations:

$$\%\mathit{\text{Confluency}}=\frac{\mathit{\text{Cells}}}{\text{max}\_\mathit{\text{cells}}}\cdot 100$$ (12)

$$V_{\mathit{\text{tumor}}}=\mathit{\text{Cells}}\cdot \frac{V_{\mathit{\text{cell}}}}{\mathit{\text{frac\_cell}}}\cdot 1\times {10}^6$$ (13)

where %Confluency is the percent confluency of the cells (dimensionless), V_tumor is the approximated tumor volume based on random close packing of spherical cells (mm³), V_cell is the volume of an individual cell (L), and frac_cell is the cell volume fraction of the tumor (dimensionless).

Statistics

Cell proliferation studies were performed with three (96-well plate) or four (microfluidic perfusion) technical replicates per treatment condition. Cell percent confluency is reported as the mean +/− standard deviation. One-way ANOVA with Tukey multiple comparisons tests were performed using Prism. Statistical significance was defined as a p value less than 0.05.

RESULTS

Static Exposure-Response Studies

Initial exposure-response studies using constant drug concentrations were conducted in 96-well plates to establish the relationship between ClpP agonist drug exposure and anti-proliferation response in MDA-MB-231-RFP triple negative breast cancer cells. Cell growth kinetics under the different treatment conditions were monitored every 12 hours through live-cell imaging, and the cell percent confluency was calculated at each time point over the 1-week treatment period (Fig. 2A–B). The estimated IC50 values based on percent confluency for TR-107 were 18 nM and 27 nM after 4 d and 7 d, respectively. For ONC201, the estimated IC50 values based on percent confluency were 903 nM and 934 nM after 4 d and 7 d, respectively. Both compounds exhibited steep exposure-response relationships around these IC50 values (Fig. 2C). However, TR-107 showed growth inhibitory IC50 values that were ~30–70X lower than ONC201 at timepoints >3 d, consistent with previous reports [3]. It is important to note that reported IC50 values are time-dependent and will be affected by the underlying cell growth kinetics, and PK-PD models accounting for drug effects on cell growth rate modulation can offer improved translational PK-PD predictions [31].

Figure 2.

Effect of mitochondrial caseinolytic protease P (ClpP) agonists on the growth of MDA-MB-231-RFP cells in a 96-well plate format. (A) Continuous treatment with TR-107 at a range of fixed concentrations for 1 week. (B) Continuous treatment with ONC201 at a range of fixed concentrations for 1 week. (C) Summary of the exposure-response relationship for TR-107 and ONC201 after 4 days and 7 days of treatment. Data are shown as mean ± standard deviation.

Dynamic Exposure-Response Studies

While the prior static studies define the relationship between a constant drug exposure profile and the corresponding effects of ClpP agonists on cell proliferation, they do not reveal how the transient drug exposure profiles under physiological conditions may affect the resulting efficacy. Therefore, we employed a Type I microfluidic perfusion system to expose MDA-MB-231-RFP cells to precisely controlled transient drug exposure profiles. This system employs automated valves to turn on or off the application of drug at defined timepoints over the 1-week treatment period. The ability of the programmable perfusion system to apply the desired transient drug exposure intervals was confirmed through quantification of a dye surrogate in the perfusate using a flow cell-based diode array detector (Fig. S1A). Because the cells are grown in a microscopy-compatible channel slide, cell growth kinetics could be assessed by the same live-cell imaging approach as that used for the static assays in 96-well plates.

Initial perfusion studies were conducted to evaluate the response of the MDA-MB-231-RFP cells to constant drug exposures while under perfusion in the channel slide. At the applied flow rate of 500 μL/h, the calculated shear stress in the channel is expected to be ~0.01 dyn/cm² and is at least an order of magnitude below even the low expected shear stress experienced by cells from interstitial fluid flow [32]. Consistent with these expectations, the observed response under perfusion was similar to what had been observed in the standard static 96-well plate assays. In the perfusion system, the estimated IC50 values based on percent confluency for TR-107 were 19 nM and 15 nM after 4 d and 7 d, respectively. Again, TR-107 showed a steep exposure-response relationship for concentrations around these IC50 values (Fig. 3).

Figure 3.

Effect of ClpP agonists on the growth of MDA-MB-231-RFP cells in a Type I microfluidic perfusion system. (A) Input TR-107 media exposure profiles applied over the 1-week study duration. (B) Growth of MDA-MB-231-RFP cells following treatment with the TR-107 exposure profiles shown in (A). (C) Summary of the exposure-response relationship for TR-107 after 4 days and 7 days of treatment. Data are shown as mean ± standard deviation.

The next set of perfusion studies aimed to address whether the pharmacological response following ClpP agonist treatment correlated with average drug exposure, or the integrated area under the curve (AUC). A standard experimental approach to test this PK metric is the dose fractionation study in which each treatment group receives the same total drug dose but with a different dosing frequency and amount per dose. Such dose fractionation studies are typically conducted in preclinical animal models, although the Type I microfluidic perfusion system enables these studies to be conducted in an automated manner without the need for animal studies. The previous dose response experiments showed that a constant exposure of 20 nM TR-107 exerted a strong anti-proliferative effect in the MDA-MB-231-RFP cells. Therefore, this 20 nM daily exposure (AUC = 480 nM*h) was used as the basis for an in vitro dose fractionation study using the perfusion system (Fig. 4A). Despite having identical total daily exposure levels, the cell growth curves revealed notable differences among the treatment groups (Fig. 4B). Longer exposure durations with lower exposure levels (i.e., 60 nM for 8 h or 20 nM for 24 h per day) led to significantly stronger anti-proliferative effects after 1 week of treatment than shorter exposure durations with higher exposure levels (i.e., 480 nM for 1 h, 240 nM for 2 h, or 120 nM for 4 h per day). Transient exposure of 120 nM for 4 h per day also led to significantly greater cell growth inhibition after 1 week of treatment compared to exposures of 480 nM for 1 h or 240 nM for 2 h. Similar trends were seen with ONC201, where sustained exposure at 2000 nM led to more pronounced cell growth inhibition than the same total daily dose compressed into shorter time intervals (Fig. S2).

Figure 4.

Effect of TR-107 dose fractionation on the growth of MDA-MB-231-RFP cells in a Type I microfluidic perfusion system. (A) Input TR-107 media exposure profiles applied over the 1-week study duration. Note that each treatment arm includes the same total daily applied dose of TR-107. (B) Growth of MDA-MB-231-RFP cells following treatment with the TR-107 exposure profiles shown in (A). Data are shown as mean ± standard deviation.

Although the dose fractionation studies demonstrated that sustained ClpP agonism can achieve greater anti-proliferative efficacy, these studies did not directly compare different exposure durations at the same exposure levels. Therefore, additional Type I perfusion studies were used to investigate the impact of ClpP agonist exposure duration. The first study examined daily dosing of 40 nM TR-107 for different durations each day (Fig. 5A). Cell growth inhibition under constant exposure of 40 nM or transient exposure of 40 nM for 8 h per day led to significantly greater cell growth inhibition after 1 week of treatment than daily exposure durations of 1 h, 2 h, or 4 h (Fig. 5B). Additionally, exposure to 40 nM TR-107 for only 1 h or 2 h per day did not lead to significantly different cell proliferation after 1 week of treatment compared to untreated cells. Examination of different daily exposure durations is relevant to the optimization of PK half-life, although this study design does not identify if there is a minimum sustained duration of ClpP agonism required to maximize efficacy. To answer this question, a second Type I perfusion study was conducted in which 40 nM TR-107 was applied continuously or for a single duration of 4 h, 8 h, 24 h, or 48 h out of the entire 1-week study (Fig. 6A). An exposure duration of 48 h led to a response that was not significantly different from continuous exposure, whereas the shorter durations of 4 h, 8 h, and 24 h resulted in significantly lower anti-proliferative effects than continuous exposure (Fig. 6B). Additionally, exposure to 40 nM TR-107 for 4 h or 8 h did not lead to significantly different cell proliferation after 1 week of treatment compared to untreated cells. Similar trends were also observed with ONC201 (Fig. S3). For both ClpP agonists, exposure durations of 48 h yielded nearly the same maximal growth inhibition as the sustained exposure for 1 week, consistent with prior studies showing that exposure of BT474 cells to ONC201 for >48 h was sufficient to maximize growth inhibition [33]. Timelapse imaging of MDA-MB-231 cells following constant treatment with TR-107 or ONC201 showed that MMP declined steadily over the first several days of treatment across the cell population, although there was variability in the time to loss of MMP across individual cells within each treated population (Fig. S4).

Figure 5.

Effect of TR-107 daily repeat dose exposure duration on the growth of MDA-MB-231-RFP cells in a Type I microfluidic perfusion system. (A) Input TR-107 media exposure profiles applied over the 1-week study duration. Note that each treatment arm includes the same applied concentration of 40 nM TR-107 but with a different duration of exposure each day. (B) Growth of MDA-MB-231-RFP cells following treatment with the TR-107 exposure profiles shown in (A). Data are shown as mean ± standard deviation.

Figure 6.

Effect of TR-107 single dose exposure duration on the growth of MDA-MB-231-RFP cells in a Type I microfluidic perfusion system. (A) Input TR-107 media exposure profiles applied over the 1-week study duration. Note that each treatment arm includes the same applied concentration of 40 nM TR-107 but with a different duration of exposure for the single applied dosing interval. (B) Growth of MDA-MB-231-RFP cells following treatment with the TR-107 exposure profiles shown in (A). Data are shown as mean ± standard deviation.

PK-PD model development and translational PK-PD predictions

A PK-PD model framework was developed to translate the experimental datasets into a quantitative exposure-response relationship for ClpP agonists (see PK-PD Modeling in Materials and Methods). The results shown in Fig. 4 demonstrate that the pharmacological response to ClpP agonists in the MDA-MB-231-RFP cells is not driven by either average drug exposure or peak drug exposure. Based on this finding, the pharmacological response in the PK-PD model was captured using a sigmoidal response relationship (i.e., Hill equation). This results in a response that is a function of both the magnitude and duration of ClpP agonist exposure, and it is consistent with the expected saturable and cooperative engagement of the ClpP heptameric ring by the small molecule agonists to trigger ClpP activation [34]. Model parameters for the ClpP agonist, TR-107, were estimated through simultaneous fitting to multiple independent datasets generated using the Type I microfluidic perfusion system (Fig. S5), and the final estimated model parameter values are provided in Table S1.

An important application of the PK-PD model is to enable simulation of the anticipated response to different ClpP agonist dosing regimens. To demonstrate the translatability of the PK-PD model, simulations were conducted to mimic previously reported mouse tumor xenograft studies with TR-107 [3]. In this study, mice bearing MDA-MB-231 subcutaneous tumors were treated with TR-107 at two different intermittent dosing regimens. The reported PK parameters for TR-107 in mice were used to define the drug exposure profiles (Fig. 7A), and the PD response was defined as the change in tumor volume for each treatment regimen. Model parameters governing the PD response were derived from the in vitro studies conducted in the Type I perfusion platform (Table S1). To facilitate in vitro to in vivo translation, PD parameters were based on unbound drug exposures for TR-107 using previously reported plasma protein binding data [3] and extrapolation to cell culture media using Equations 5–6. The resulting model simulations of the anticipated tumor response show very good concordance with the previously reported tumor growth data (Fig. 7B). The PK-PD model was then used to explore the potential efficacy of dosing regimens that had not been tested in the mouse tumor xenograft models (Fig. 7C). These simulations confirmed the expectation that additional dose intensification for TR-107 might yield improved efficacy, consistent with prior studies conducted with ONC201 in mice [18].

Figure 7.

PK-PD model-based predictions of MDA-MB-231 tumor growth inhibition in mice using parameters derived from the Type I microfluidic perfusion system. (A) TR-107 unbound plasma PK profiles in mice following administration in four weekly cycles of either 4 milligrams/kilogram (mpk) twice-daily on days 2–4 of each week or at 8 mpk twice-daily on days 2 and 5 of each week. (B) Model-predicted (solid lines) vs. observed in vivo tumor volume (solid circles) as a function of time. (C) Model predictions for the anti-tumor efficacy of hypothetical twice-daily treatment with TR-107 given across a range of dose levels on each day throughout the study duration.

Dose exploration using Type II perfusion systems

While PK-PD models are powerful translational tools for drug discovery and development, they carry inherent risks based on the underlying assumptions. The Type II perfusion systems can serve as a complementary experimental approach that provides an additional layer of confidence in efficacious dose predictions. Unlike the Type I perfusion systems that utilize automated valves to apply defined intervals of drug exposure, the Type II perfusion systems can apply drug exposures that mimic physiologically relevant PK profiles. To illustrate this application, the Type II perfusion system was employed to mimic the previously published study of TR-107 treatment of mice bearing MDA-MB-231 tumor xenografts [3]. The reported PK parameters for TR-107 in this study were used to define the input PK profiles corresponding to a 1-week cycle from the two different dosing regimens tested in mice (Fig. 8A). Input media concentrations were adjusted to provide similar unbound drug exposures using Equation 5. The ability of the programmable perfusion system to apply the desired dynamic drug exposure profile was confirmed through quantification of a dye surrogate in perfusate samples collected every 15 minutes over the course of a 24-h PK profile (Fig. S1B).

Figure 8.

In vitro to in vivo translatability of the microfluidic perfusion platform. (A) Input TR-107 media exposure profiles applied over the 1-week study duration in a Type II microfluidic perfusion system. The TR-107 profiles are derived from the unbound plasma PK profiles for TR-107 administered at 4 milligrams/kilogram (mpk) twice-daily on days 2–4 or 8 mpk twice-daily on days 2 and 5. (B) Growth of MDA-MB-231-RFP cells following treatment with the TR-107 exposure profiles shown in (A). Note that the dosing regimens were applied in two separate experiments for the different dosing intervals. (C) Comparison of microfluidic perfusion platform-observed relative cell growth (open squares and circles) vs. in vivo-observed relative tumor growth (solid lines with solid circles). The nondimensional time scale, τ, corresponds to the time to reach a relative growth of ~20 for the untreated or vehicle groups (i.e., 168 hours for the in vitro datasets and 25 days for the in vivo datasets). Data in (B) are shown as mean ± standard deviation, and data in (C) are shown as median to facilitate visual comparison.

The PD response following a 1-week cycle of each treatment regimen was determined through quantification of relative cell growth through longitudinal live-cell imaging (Fig. 8B). Comparison of the relative growth curves from the Type II perfusion system and the previous mouse tumor xenograft study can be facilitated by use of a nondimensional time scale corresponding to an equivalent relative growth window for the untreated cells or tumors. The nondimensional time scale is based on the relative growth rate of the cells or tumors, which can be calculated through modeling of the growth curves or approximated as the time it takes for the untreated or vehicle groups to reach a given fold increase relative to their initial value at the beginning of the treatment period. This approach facilitates comparison of the treatment-induced effects within the Type II perfusion system and the mouse tumor xenografts. The results from the microfluidic perfusion studies align very closely with those observed in the mouse tumor xenograft study when the growth curves for each study are scaled by the time it takes for the untreated or vehicle groups to increase ~20-fold relative to their initial value (i.e., 168 hours for the in vitro datasets and 25 days for the in vivo datasets) (Fig. 8C). These results establish the feasibility for using the microfluidic perfusion system as an alternative approach to mouse tumor xenograft studies for dosing regimen exploration.

Consequently, the Type II perfusion system was used to experimentally test the PK-PD model predictions shown in Fig. 7C for continuous twice-daily dosing of TR-107. Input PK profiles in media were adjusted to achieve equivalent unbound drug exposures for TR-107 as previously measured in mice and simulated using the PK model (Fig. 9A). The PD response of the MDA-MB-231-RFP cells following a 1-week treatment with twice-daily dosing of TR-107 across a range of dose levels was determined through longitudinal live-cell imaging (Fig. 9B). All of the treatment groups showed significant reductions in cell growth compared to untreated cells. Moreover, the predicted efficacy of the dosing regimens within the microfluidic perfusion system is consistent with the PK-PD model predictions for in vivo tumor growth inhibition using the same dosing regimens (Fig. 9C). The microfluidic perfusion data therefore support the PK-PD model predictions that twice-daily dosing of TR-107 over the entire treatment period can drive much more profound cell growth inhibition than the intermittent dosing regimens.

Figure 9.

Model-informed dose exploration within the microfluidic perfusion platform. (A) Input TR-107 media exposure profiles applied over the 1-week study duration in a Type II microfluidic perfusion system. The TR-107 profiles are derived from the simulated unbound plasma PK profiles for TR-107 administered twice-daily for the duration of the study. (B) Growth of MDA-MB-231-RFP cells following treatment with the TR-107 exposure profiles shown in (A). (C) Comparison of model-predicted relative tumor growth (solid lines) vs. microfluidic perfusion platform-observed relative cell growth (solid circles). The nondimensional time scale, τ, corresponds to the time to reach a relative growth of ~20 for the untreated or vehicle groups (i.e., 168 hours for the in vitro datasets and 25 days for the model-predicted relative tumor growth). Data in (B) and (C) are shown as mean ± standard deviation.

DISCUSSION

Agonists of the mitochondrial protease, ClpP, have emerged as a promising new class of cancer therapeutics that target mitochondrial function to disrupt tumor metabolism. ONC201 is a first-in-class imipridone compound that has shown clinical activity across various cancer types. While multiple mechanisms of action have been proposed for ONC201, ClpP serves as an important biological target for this class of molecules. Multiple analogues of ONC201 have been developed to identify more potent and selective activators of ClpP, and TR-107 is an example of one of the most advanced analogues currently in preclinical development. Successful application of these compounds as cancer therapeutics requires a detailed understanding of the extent and duration of ClpP agonism needed for efficacy. However, the availability of tools to conduct such a systematic exposure-response analysis represents a major gap in the current translational toolbox.

Early target validation studies may involve genetic manipulations or other pharmacological perturbations that result in sustained target modulation but fail to account for the transient nature of target modulation expected with drug administration in physiological organisms. Preclinical animal studies often therefore provide the only experimental insights into temporal PK-PD relationships. Unfortunately, these preclinical animal models are constrained by species-specific differences in PK and PD, and detailed PK-PD investigations for dose optimization in animal models must be weighed against cost, time, and ethical considerations. Therefore, there is an urgent need to address this gap in the translational toolbox with novel alternative methods or new approach methods (NAMs) that overcome the limitations of current in vitro and in vivo preclinical approaches [25, 35–37]. Moreover, the FDA’s Oncology Center of Excellence has spearheaded Project Optimus to transform how dose optimization and dose selection are performed in oncology [38]. Specifically, this initiative encourages a shift away from the standard maximum tolerated dose paradigm for dose selection. Instead, drug developers are encouraged to pursue early incorporation of dose exploration studies and a rational approach to dose selection based on a more nuanced understanding of a drug’s exposure-response relationship. In this study, we have described an approach that combines PK-PD modeling with in vitro microfluidic perfusion systems for the systematic investigation of drug exposure-response relationships in an animal-alternative format.

PK-PD models provide a quantitative framework for understanding and exploring drug exposure-response relationships, and the use of such models can increase the probability of success in drug discovery and development [39]. A critical aspect of PK-PD model development is selecting the underlying model structure comprising the system of equations that translate the biological mechanisms into a mathematical framework [40]. Equally important is the selection of the drug exposure metric incorporated into these equations that links PK to the downstream PD response [23]. Parameterization of these models requires datasets that relate the response of the biological systems to the given drug exposure metrics. These datasets often are derived from measurements taken at the endpoint of an assay with static drug exposures, but such snapshots can fail to capture nuances in the temporal response dynamics within biological systems. For this reason, the studies presented here leveraged live-cell imaging to capture dynamic exposure-response datasets. With the added ability to precisely control the applied drug exposure profiles, the microfluidic perfusion systems generated datasets that could inform the PK-PD models while also providing experimental validation of model predictions.

The microfluidic perfusion systems described here are designed to be fit-for-purpose to address specific PK-PD questions that arise in drug discovery and development. As with any approach, there are important limitations that must be considered. The current platform does not incorporate mechanisms for generating drug metabolites that may contribute to pharmacological activity. Therefore, this approach is not suitable for drugs where the pharmacologically active form requires in vivo metabolic conversion, although any metabolites can be directly evaluated if they are synthetically available. The Type I perfusion systems are ideally deployed early in drug discovery programs once a tool compound is available to characterize the impact of pharmacological target perturbation. These initial studies do not require knowledge of drug PK, and the studies can be conducted with compounds that may not even be suitable for in vivo use. The Type I studies provide general insights to guide drug design based on the required modulation of the pharmacological target to achieve efficacy, and these findings should be broadly applicable across a class of compounds with a shared target and mechanism of action. However, the Type II perfusion systems are most effectively deployed with well-characterized lead compounds or more advanced drug candidates. Typically, the PK of these compounds will have already been measured in the relevant species (e.g., mouse, human) or predicted PK profiles will have been generated using approaches such as in vitro-in vivo extrapolation, allometric scaling, or physiologically based pharmacokinetic modeling. Unbound concentrations of drug in plasma are often used as a suitable surrogate for drug exposure at the site of action (e.g., tumor), although the Type II perfusion systems can be used to directly mimic measured or predicted drug exposure profiles in tissues or compartments other than plasma. Importantly, this platform also provides the unique ability to experimentally evaluate the pharmacological response of human cells to human drug PK profiles in a nonclinical setting. This can serve to increase confidence in human efficacious dose predictions to guide the design of first-in-human clinical studies of investigational drug candidates. Moreover, the platform can be leveraged to conduct human-relevant dose exploration in a nonclinical setting, providing complementary insights that support initiatives such as Project Optimus that aim to transform the processes for clinical dose selection and optimization.

Using this combination of PK-PD modeling and microfluidic perfusion systems, we evaluated how the magnitude and duration of ClpP agonist exposure affected the anti-proliferative response in a human triple negative breast cancer cell line. Under constant exposure, the ClpP agonists demonstrated a threshold effect with a very steep response curve transitioning from near maximal response to negligible response around the threshold value. This was observed for both TR-107 and ONC201, and it is likely a consequence of the mechanism of action in which the compounds must engage the binding sites within the ClpP heptameric ring to trigger activation of the protease. Subsequent studies in the microfluidic perfusion systems demonstrated that simple PK metrics such as peak or average drug exposures over the dosing interval were insufficient to explain the observed PD response. Instead, the magnitude and duration of ClpP agonist exposure above a threshold concentration required for ClpP activation determined the pharmacological effect. With daily dosing of a ClpP agonist, constant exposure above this threshold led to greater cell growth inhibition than transient exposure durations of 1 h, 2 h, 4 h, or 8 h per day. This observation can help to guide half-life optimization for novel ClpP agonists. Furthermore, we confirmed prior observations that sustained exposure above the efficacy threshold for >48 h yields a pharmacological response that is similar to that achieved with a constant drug exposure profile. However, imaging studies revealed variability across individual cells in the time to loss of MMP following ClpP agonist treatment. While a dosing regimen of two consecutive daily doses per week of ONC201 is currently being evaluated in clinical studies, sustained ClpP agonism beyond two days may help drive deeper or more sustained anti-tumor efficacy in light of anticipated exposure and response variability by affecting a greater proportion of the cell populations within a tumor lesion. The PK-PD simulations and microfluidic perfusion studies evaluating intermittent vs. daily dosing of TR-107 provided further evidence in support of sustained ClpP agonist exposure as a strategy for enhancing efficacy. Prolongation of plasma PK half-life through compound optimization or alternative formulations may represent promising strategies for future design iterations of ClpP agonists.

An important caveat to the conclusions presented here is that these studies have focused exclusively on the anti-cancer efficacy of drugs acting through an assumed primary mechanism of ClpP activation. In general, the conclusions should be generalizable for all ClpP agonists, including TR-107 and ONC201, with a shared mechanism of action. However, it is possible for differences in polypharmacology profiles, even among drugs with the same primary mechanism of action, to skew these exposure-response relationships. Future iterations of the microfluidic perfusion platform will also aim to incorporate toxicity-relevant cell populations to simultaneously evaluate safety and efficacy for dose exploration and optimization in an animal-alternative manner. Although the evidence presented here points toward continued dose intensification of ClpP agonists as a strategy to improve efficacy, it is unknown whether tolerability will likewise be decreased with more sustained ClpP agonist exposure due to on-target or off-target mechanisms. While tolerability in the clinical setting with ONC201 has been excellent to date, most of the studies have examined relatively infrequent dosing regimens that do not provide sustained ClpP agonism beyond 1–2 days per week. It will be important to monitor the tolerability of dosing regimens that aim to maintain sustained ClpP agonism for consecutive days as is currently being studied in the clinical setting for ONC201.

CONCLUSION

In this study, we leveraged an animal-alternative platform combining PK-PD modeling with microfluidic perfusion systems to elucidate dynamic exposure-response relationships for ClpP agonists as cancer therapeutics. By using the microfluidic perfusion systems, these exposure-response relationships can be investigated in a systematic manner without the significant time, expense, and ethical barriers imposed by conducting such studies in traditional preclinical animal models and even without the requirement for compound optimization and scale-up typically required to enable such in vivo studies. Since dose optimization must ultimately be guided by the balance between efficacy and toxicity, future efforts will aim to integrate cells or tissues that enable simultaneous evaluation of both metrics. Moreover, outside of testing in animals or humans, the current experimental toolbox does not allow direct testing of efficacious dose predictions derived from PK-PD modeling. This study demonstrated how microfluidic perfusion systems can be deployed both to generate datasets that inform PK-PD models and to directly test PK-PD model predictions. Although these studies focused on applications related to the discovery and development of cancer therapeutics, the described approach can be applied broadly across other therapeutic areas that have in vitro assays which are suitable for transformation into dynamic in vitro PK-PD assays using the perfusion platform. The long-term objective of this approach is to significantly reduce the number of animal studies required to inform efficacious dose predictions while improving confidence in human dose selection for ClpP agonists and other investigational therapeutics.