Fig 1.
Overview of ADC mechanisms of action.
Intact ADCs extravasate from the blood into tumor tissue to enter the tumor microenvironment (TME). The intact ADC typically binds to the cancer cell, internalizes, and releases the payload inside the cell. The payload can directly kill the cell or diffuse into nearby cells for bystander effects. These are generally considered Targeted Delivery. Four proposed target-independent mechanisms of efficacy are also shown: (1) Free payload from systemic deconjugation or metabolism in healthy tissue and intravasation can also enter the tumor as free payload (Systemic Release). (2) Fcγ receptor mediated uptake of ADCs, often driven by FcγR1, releases payload inside myeloid cells like macrophages (Mac Release), and the payload can kill nearby cancer cells (bystander killing). (3) Extracellular proteases can cleave the linker to release payload outside of cancer cells, relying on the bystander effect to enable the payload to enter cells (TME Release). (4) An indirect mechanism of efficacy is immune stimulation, which could be driven by mechanisms such as immunogenic cell death, in which cancer cells killed by targeted delivery release damage-associated molecular patterns (DAMPs) to activate T cells for further cancer cell killing. Activated T cells can kill cancer cells by releasing perforin (PFN) and granzyme B (GzmB), Fas/FasL binding, and interferon gamma (IFNγ) and tumor necrosis factor alpha (TNFα) cytokine release to induce apoptosis (Immune Effects). The activation of T cells can enhance/drive tumor efficacy depending on the immune microenvironment. Created in BioRender. Calopiz, M. (2026) https://BioRender.com/9s9zws3.
Fig 2.
Hybrid Agent-Based Model (SimADC) Combines Drug Gradients with Cellular Agents.
(A) A grid of thousands of cells incorporates blood vessels (source and sink of ADCs and payloads from the plasma), cancer cells that divide or die with probabilities based on local conditions, T cells for immune cell killing, and macrophages for Fc-mediated payload release. (B) The grid is overlayed with drug gradients (including bound and unbound ADC and free payload in the intracellular and extracellular space) that are used to capture bystander killing and determine the probability of cell death. (C) The updated model incorporates targeted delivery (inclusive of bystander effects) as in our previous models but also adds T cell killing (baseline and activated), macrophage-mediated payload release (mac release), extracellular ADC linker cleavage (TME release), and uptake of free payload from the plasma (systemic release). Created in BioRender. Calopiz, M. (2026) https://BioRender.com/doavg9f and https://BioRender.com/a2tocth.
Fig 3.
Independent calibration of payload potency and baseline T cell activity.
Experimental data from Rios Doria et al. [8] (top row) were used to train the simulations (bottom) on payload efficacy and T cell effects. (A) The cell doubling time was fit to match the untreated animal model tumor growth, and pharmacodynamic parameters were fit to efficacy data using an anti-EphA2 antibody with either a Tubulysin or PBD payload in a nude mouse model that lacks T cell effects. SimADC was able to capture the magnitude and dose response of these effects. (B) For T cell killing, the growth rate in a nude mouse host versus immunocompetent syngeneic host was compared. The baseline (non-activated) T cell killing rate was fit to account for the slower growth rate when intratumoral T cells are present. Error bars represent standard deviation; each simulation point is the result of 300 simulation runs (100 simulations run in triplicate).
Fig 4.
Experimental and simulated efficacy of ADCs in immunocompetent mice.
To capture the activation of immune cells following ADC treatment, the percentage of active T cells in the model was increased from 10% (untreated) to 18% (EphA2-Tubulysin) or 22% (EphA2-PBD) based on values reported in Rios-Doria et al. [8]. While only measured for one dose of each ADC (5 mg/kg for EphA2-Tubulysin and 0.3 mg/kg for EphA2-PBD), these values were kept constant across doses. Error bars represent standard deviation; each simulation point is the result of 300 simulation runs (100 simulations run in triplicate).
Fig 5.
Treating nude vs. syngeneic mice with T-DXd.
(A) Using the CT26 nude mice growth rates previously calibrated from Rios-Doria et al. [8], we simulated T-DXd with a 20-fold decrease in potency (to account for the differences in sensitivity between human NCI-N87 cancer cells and mouse CT26 cancer cells) and find agreement with the experimental data showing little efficacy in mouse cancer cells grown in the nude mouse model. (B) Using the growth rates from the syngeneic mouse model, we simulated T-DXd treatment and find a similar efficacy to the experimental data. We also simulated the impact of activated T cell killing alone (no payload efficacy) and see similar tumor volumes, indicating that efficacy is mainly driven by T cell activation in this model. Error bars represent standard deviation; each simulation point is the result of 300 simulation runs (100 simulations run in triplicate).
Fig 6.
Fc receptor-mediated internalization of ADCs can drive high efficacy in macrophage-dense tumors.
Our macrophage model pharmacokinetic and pharmacodynamic parameters were calibrated to literature experimental data from Li et al. [10], who measured (A) intratumoral free payload concentrations and (B) efficacy in tumors with high macrophage infiltration (~50% macrophages in the tumor) after dosing a targeted (CD30 receptor binding) and a non-targeted (IgG, mainly binding to Fc receptors on macrophages) ADC with an MMAE payload. Our simulations show similar concentrations of intratumoral payload, also resulting in similar efficacy with both types of ADCs after model calibration. Error bars represent standard deviation; each simulation point is the result of 300 simulation runs (100 simulations run in triplicate).
Fig 7.
TME release of payload can drive efficacy in tumor xenografts.
Experimental data from Tsao et al. [11] (top) shows the growth of untreated HER2-negative MDA-MB-468 xenografts injected in the mammary fat pad of SCID mice (control) or treated with three injections of 10 mg/kg T-DXd on days 0, 7, and 14. Experimental data (Experiment A) were collected from a mouse with HER2-positive and HER2-negative dual-implanted tumors and data (Experiment B) were collected from a mouse with only one HER2-negative tumor. In simulations, a cleavage rate constant (kCleave) of 4x10-5/s was able to capture the efficacy in the HER2-negative tumor. Error bars represent standard deviation; each simulation point is the result of 300 simulation runs (100 simulations run in triplicate).
Fig 8.
SimADC predicts responses to systemic DXd in mouse xenografts.
(A) The plasma clearance parameters for free DXd were fit to experimental data from Okamoto et al. [26] following intravenous delivery. We simulated a dose of 0.04 mg/kg (vs. 1 mg/kg used in the experiment) to match the initial concentration. (B) Our previously-calibrated DXd parameter set was validated by comparison to experimental data from Kumazawa et al. [27] (left) after calibrating our cell doubling time parameters to the control curve measured from human SC-6 xenografts in nude mice. We doubled our maximum probability of cell killing to adjust for the change in potency (0.045 for DXd vs. 0.09 for exatecan) and dosed a total of 2 mg/kg (vs. a total of 50 mg/kg used in the experiment) based on our equivalent model dose from (A). We simulated the following dosing regimens as outlined in Kumazawa et al. Each color corresponds to the dosing strategy for each curve, and the arrows represent the dosing schedule for each experiment and simulation run: (i, green) q4d x 3, (ii, red) q4d x 4, and (iii, blue) q7d x 3. Error bars represent standard deviation; each simulation point is the result of 300 simulation runs (100 simulations run in triplicate).
Fig 9.
Minimum threshold of HER2 expression in mice and humans for efficacy of targeted delivery alone.
The growth rates of tumors with varying low expression levels were simulated in SimADC following treatment with 5.4 mg/kg of T-DXd. Using tumor regression (below starting volume, as defined by the horizontal dashed line) at 21 days (defined by the vertical dashed line) as the threshold for efficacy, (A) 50,000 HER2/cell were needed in mice while (B) 30,000 HER2/cell were sufficient in the human simulations. The difference occurs from scaling the pharmacokinetic parameters from mice to humans. Error bars represent standard deviation; each simulation point is the result of 300 simulation runs (100 simulations run in triplicate).
Fig 10.
Predicted efficacy in HER2 low (10,000 HER2/cell) tumors in humans and cell killing mechanisms.
Simulations tested control (no treatment), targeted delivery (5.4 mg/kg T-DXd only), baseline T cell killing only, and additive (baseline T cell killing plus targeted delivery) and synergistic (activated T cell killing plus targeted delivery) mechanisms. (A) With 10K HER2/cell, tumor growth is slowed but does not regress with targeted delivery of T-DXd alone. Infiltrating T cells without treatment also slow growth but don’t regress the tumor. However, a combination of 5.4 mg/kg of T-DXd with infiltrating T cells (baseline T cell killing) can regress the tumor, and activation of the T cells (synergy) can drive a strong response. (B) The first 2000 cells killed by each treatment regimen are plotted over time. Cells killed due to direct action of DXd (targeted delivery) vs T cell killing are plotted to separately to show their relative contributions. Compared to T-DXd treatment alone, the presence of T cells reduces the number of cancer cells killed by the drug with a further reduction following activation due to competing mechanisms. The T cell killing rate is not impacted as strongly by the T-DXd treatment, but it is substantially enhanced through activation. Error bars represent standard deviation; each simulation point is the result of 300 simulation runs (100 simulations run in triplicate).
Fig 11.
Target-independent mechanisms alongside targeted delivery show tumor growth control in humans.
Simulated tumor volumes on day 21 (+/- standard deviation) were plotted for three target-independent mechanisms of action (mac release, TME release, and systemic release) acting in concert with targeted delivery of cancer cells. The dotted line shows initial tumor size. Lower HER2 targeting (10K HER2/cell) reduced tumor sizes to ~50% with targeted delivery alone. Each of the three target-independent mechanisms added to this efficacy, with the greatest improvement from mac release, which was the only mechanism that resulted in tumor regression. Note that mac release and TME release are dependent on the number of macrophages and extracellular protease concentrations in the TME, respectively. Error bars represent standard deviation; each bar is the result of 300 simulation runs (100 simulations run in triplicate). The dotted line is the cutoff for 1000 mm3, in which tumor volumes below 1000 mm3 achieve regression.
Fig 12.
Efficacy of combined mechanisms alongside targeted delivery to moderate and high HER2-expressing tumors.
(TD = targeted delivery, ICadd = immune cell additive activity, ICsyn = immune cell synergistic activity, SYSrel = systemic release, MACrel = mac release, TMErel = TME release) The addition of systemic release to TME release was enough to regress tumors. The addition of either systemic release or TME release to mac release was enough to drive further tumor regression. For comparison with the combined target-independent mechanisms, tumors with moderate HER2 expression (100,000 HER2/cell) and high HER2 expression (1,000,000 HER2/cell) underwent simulated treatment, and both resulted in very strong efficacy, consistent with the high efficiency of targeted payload delivery. Error bars represent standard deviation; each bar is the result of 300 simulation runs (100 simulations run in triplicate). The dotted line is the cutoff for 1000 mm3, in which tumor volumes below 1000 mm3 achieve regression.
Table 1.
Summary of efficacy across varying HER2 expression and target-independent mechanisms.
Table 2.
Pharmacokinetic parameters for plasma concentration of free payload in mice and humans.
Table 3.
Pharmacokinetic and simulation parameters of ADCs in mice and humans.