Mechanistic modeling of tumor immune microenvironment reveals strategies to enhance antibody-drug conjugates’ efficacy Authors: Constantinos Harkos, Lance L Munn, Triantafyllos Stylianopoulos, Rakesh K Jain Published: 2026 ABSTRACT AbstractBackgroundAntibody-drug conjugates (ADCs) and bispecific antibodies represent a rapidly advancing frontier in oncology, yet the abnormal tumor microenvironment (TME) hinders their delivery and reduces efficacy. Emerging immunomodulatory ADCs (IM-ADCs) demand mechanistic mathematical models that couple drug transport with immune dynamics.MethodsHere, we present a mechanistic framework for the delivery of HE-S2 ADC, an anti-programmed cell death ligand 1 (PD-L1) antibody bearing the bifunctional immunomodulator D18. Our model integrates cancer-immune cells interactions, TME properties, such as dysfunctional vessels, elevated interstitial fluid pressure, tissue hydraulic conductivity, and vascular permeability, spatiotemporal distributions across growing tumor and adjacent host tissue, convective-diffusive transport, ADCs binding and internalization kinetics and tumor-draining lymph node biology governing antigen presentation and the generation of effector CD8+ T cells. Parameters were calibrated simultaneously with the murine MC38 and B16 tumor growth data and effector CD8+T cell data following treatment with D18, anti-PD-L1, and ADC.ResultsOur mechanistic spatiotemporal model captures the superior antitumor efficacy of the HE-S2 ADC relative to its individual components and provides mechanistic predictions for unmeasured variables, such as spatiotemporal dynamics of drug/immune-cell distributions. It explains reduced intratumoral D18 exposure via rapid clearance, while antibody/ADC achieves higher tumor retention through leaky tumor vasculature. The model suggests a reinforcing loop in which improved ADC exposure enhances CD8+T cell infiltration, driving tumor shrinkage that lowers fluid pressure and improves drug delivery. Parametric analyses findings support TME normalization strategies that increase functional vessel density prior to ADC administration; however, such approaches should preserve sufficient vascular permeability by maintaining vessel pore radius >~40 nm, ensuring pores remain large enough for ADC extravasation and effective intratumoral delivery.ConclusionThe proposed mechanistic model successfully captures how TME properties regulate the delivery and efficacy of IM-ADCs while suggesting TME normalization as a potential strategy to improve treatment outcomes.Keywords: Immunotherapy, Antibody-drug conjugates - ADC, Tumor microenvironment - TME WHAT IS ALREADY KNOWN ON THIS TOPIC Antibody-drug conjugates (ADCs) and bispecific antibodies are among the fastest-growing therapeutic modalities in oncology, but abnormalities of the tumor microenvironment (TME) limit their delivery and antitumor response. Antibody-drug conjugates (ADCs) and bispecific antibodies are among the fastest-growing therapeutic modalities in oncology, but abnormalities of the tumor microenvironment (TME) limit their delivery and antitumor response. WHAT THIS STUDY ADDS The model predicts spatiotemporal biomarkers (i.e., interstitial fluid pressure (IFP), intratumoral drug distribution, CD8+T cell infiltration) that help explain observed treatment outcomes. The model predicts spatiotemporal biomarkers (i.e., interstitial fluid pressure (IFP), intratumoral drug distribution, CD8+T cell infiltration) that help explain observed treatment outcomes. Mechanistic modeling indicates that during ADC treatment, CD8+T cell infiltration reduces tumor volume, lowering IFP and further improving intratumoral delivery, reinforcing a positive loop. The model suggests the use of TME normalization approaches that increase functional vessel density before immunomodulatory ADC (IM-ADC) dosing, while maintaining a vessel wall pore radius >~40 nm. Mechanistic modeling indicates that during ADC treatment, CD8+T cell infiltration reduces tumor volume, lowering IFP and further improving intratumoral delivery, reinforcing a positive loop. The model suggests the use of TME normalization approaches that increase functional vessel density before immunomodulatory ADC (IM-ADC) dosing, while maintaining a vessel wall pore radius >~40 nm. HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY The study highlights TME factors that enable better IM-ADC delivery and efficacy. The study highlights TME factors that enable better IM-ADC delivery and efficacy. It suggests TME normalization as a strategy to boost ADC delivery and efficacy. It suggests TME normalization as a strategy to boost ADC delivery and efficacy. Introduction Antibody-drug conjugates (ADCs) and bispecific antibodies (BsAbs) are among the fastest-growing classes of drugs in oncology.1–3 ADCs consist of monoclonal antibodies (mAbs) conjugated to a payload—most commonly a cytotoxic drug with the aim of improving the safety and efficacy of the conjugated payload by restricting its delivery to cells expressing the target antigen, such as a tumor antigen, recognized by the chosen mAb.1–3 On the other hand, bispecific antibodies can bind to two different antigens or epitopes simultaneously. One arm of the antibody can bind to cancer cells and the other arm to immune cells, bringing the immune cells into close proximity, promoting immune cell-induced cytotoxicity against tumor cells. Toll‐like receptor 7 and 8 (TLR7/8) immunomodulatory ADCs (IM-ADCs), also referred to as immune-stimulating antibody conjugates (ISACs) have been tested in the clinic, such as the formulations NJH-395 for TLR7,4 SBT-6050 for TLR8,5 and BDC-1001 for both TLR7/8.6 Overall, early clinical experience suggests a key challenge for this class: strong immune stimulation can lead to systemic immune-related toxicities that limit dose escalation and achievable exposure. For example, NJH-395 was associated with dose-limiting and other immune-related toxicities in the Phase I study, which constrained dosing.4 In contrast, BDC-1001 has shown improved tolerability and evidence of clinical activity, but the most meaningful benefit was observed at higher dose levels (e.g., the Recommended Phase 2 Dose), consistent with the need for higher exposure for efficacy.7 While recent reviews and preclinical studies highlight the potential of IM-ADCs and BsAbs,3 8–11 the mechanisms of their interactions with the immune system and the tumor microenvironment (TME) remain poorly understood. The abnormal and immunosuppressive TME not only hinders the delivery of therapeutic agents, but also renders them ineffective once they accrue in tumors.12–15 Better elucidation of these interactions will be vital to the development of the next generation of antibody therapeutics. Mathematical modeling has been demonstrated to be an important tool in studying antibody-based therapeutics, especially helping us to understand the pharmacokinetics and pharmacodynamics (PK–PD), payload release and stability, and interactions with target cells.16–18 For ADCs, models span cellular processing and intracellular exposure,19 antigen heterogeneity and bystander killing,20 21 systemic deconjugation and drug-antibody ratio (DAR) dynamics,22 tumor-growth PK-PD and regimen effects,23 tumor disposition and translational PK–PD,24 spatial tumor transport and penetration,25–27 and biomarker-linked PK–PD, including DNA-damage markers.28 These studies are summarized in table 1. Similar modeling approaches have also been used for bispecific antibodies, where physiological-based pharmacokinetic modeling and quantitative systems pharmacology modeling capture whole-body PKs, T-cell trafficking/engagement, synapse formation, and size-dependent biodistribution/extravasation, and reveal optimal dosing while avoiding cytokine release syndrome.29–33 Table 1Summary of recent representative mathematical models for ADCs, highlighting the model type, key biological processes considered, TME integration and major outcomesReferenceModel typeKey biological processesTME propertiesOutcomeJumbe et al23PK–PDTumor growth kinetics and delayed tumor killing (delay between plasma concentrations and tumor response).Tumor compartment, tumor proliferation rate.Same total T-DM1 dose over 21 days gave similar tumor response whether given as one, two, or three doses. This supports schedule-independent activity across the tested regimens and suggests efficacy is driven by overall exposure.Shah et al24PK–PDCD30 binding and internalization, intracellular processing and payload release, tumor payload drives tumor growth inhibition.Tumor compartment, plasma-tumor transport, tumor size effects, vascularized tumors and avascular micrometastases.Tumor disposition model predicts xenograft tumor and plasma payload concentrations and translated model predicts progression-free survival and complete response rates comparable to clinical results.Bender et al22PKAntibody PK, distribution and clearance, DM1 deconjugation in the central compartment across DAR0–DAR7, plasma stability informed by in vitro dataNoneDeconjugation drives T-DM1 PK. Higher DAR deconjugates faster. T-DM1 clears ~2× faster than total trastuzumab.Cilliers et al25PBPK+Krogh cylinder tissue distribution model.Whole-body ADC/antibody PK, tumor penetration, HER2 binding, internalization, DAR-dependent distribution, deconjugation, co-administration of unconjugated antibody.Tumor compartment represented by a Krogh cylinder with radial diffusion from vessels, vascular permeability, antigen binding, and binding-site-barrier effects.Captures systemic organ concentrations and heterogeneous T-DM1 tumor distribution; predicts that DAR, linker stability/deconjugation, receptor expression, trafficking, and unconjugated antibody co-administration can strongly affect ADC penetration and efficacy.Singh and Shah19PK–PDInternalization, degradation, payload efflux.NoneEstimates intracellular exposure drivers.Burton et al26Systems pharmacology/distributed-parameter mechanistic tumor transport model.ADC extravasation, diffusion, antigen binding, internalization, intracellular payload release, payload efflux, spatial and heterogeneous bystander effects.Realistic 3D tumor microvascular network, spatial drug gradients, vessel geometry, heterogeneous antigen expression, payload washout.Predicts that payload exposure far from vessels depends strongly on payload diffusivity. In heterogeneous tumors, payload accumulation in antigen-negative cells increases with dose but depends weakly on the percentage of antigen-positive cells.Singh and Shah20PK–PDHER2 binding and internalization, intracellular processing with MMAE release, MMAE efflux and exchange between cell types, tubulin binding and tubulin occupancy drives cytotoxicity, transit compartments capture delay in cell killing.Antigen heterogeneity.Explains bystander killing in coculture where payload generated in HER2-high NCI-N87 affects HER2-low GFP-MCF7. Cytotoxicity in both cell types requires similarly high tubulin occupancy. A ~59–60% increase in tubulin occupancy corresponds to half-max bystander killing.Byun et al21PK–PDReceptor-mediated endocytosis, payload release, bystander killing.Antigen heterogeneity.Predicts effect of Ag+/Ag⁻ ratios on payload release.Vasalou et al28PK–PDTumor penetration and HER2 binding, internalization and trafficking, payload deconjugation and tumor released DXd, non-specific uptake, DNA-damage biomarkers with a time delay vs tumor DXd.Tumor compartment, HER2 expression level, internalization rate and non-specific uptake.Tumor released DXd was higher in tumors with higher HER2. DNA-damage markers increased and stayed elevated, but their changes happened later than the tumor DXd peak. The JIMT-1 data were best explained by a slower internalization rate.Chen et al27PK/layered tumor-distribution model.Antibody binding kinetics, HER2 binding, intratumoral penetration, receptor occupancy, ADC/immunotoxin cytotoxicity and efficacy.Tumor represented as concentric layers extending from the vasculature to distant tissue regions, incorporating binding-site-barrier effects and intratumoral distribution gradients.Shows that binding microconstants influence tumor distribution, with lower kon improving penetration and antibodies with lower kon showing more homogeneous intratumoral distribution and trends toward improved efficacy.This workCombination of PK–PD and distributed parameter mechanistic modeling.ADC biodistribution, immune response of innate and adaptive immune cells.Tumor compartment with tumor growth at the expense of the host tissue, spatial variations of vasculature, vessel permeability, hydraulic conductivity and IFP.Predicts tumor growth and shrinkage after therapy, immune infiltration, spatial drug gradients and the role of spatial variations in the TME in ADC efficacy.ADC, antibody-drug conjugates; 3D, three-dimensional; DAR, drug-antibody ratio; GFP-MCF7, Green Fluorescent Protein-expressing Michigan Cancer Foundation-7 cells; HER2, Human epidermal growth factor receptor 2; IFP, interstitial fluid pressure; MMAE, monomethyl auristatin E; NCI-N87, National Cancer Institute-N87 cells; PBPK, physiological-based pharmacokinetic modeling; PD, pharmacodynamics; PK, pharmacokinetics; T-DM1, trastuzumab emtansine ; TME, tumor microenvironment. Summary of recent representative mathematical models for ADCs, highlighting the model type, key biological processes considered, TME integration and major outcomes ADC, antibody-drug conjugates; 3D, three-dimensional; DAR, drug-antibody ratio; GFP-MCF7, Green Fluorescent Protein-expressing Michigan Cancer Foundation-7 cells; HER2, Human epidermal growth factor receptor 2; IFP, interstitial fluid pressure; MMAE, monomethyl auristatin E; NCI-N87, National Cancer Institute-N87 cells; PBPK, physiological-based pharmacokinetic modeling; PD, pharmacodynamics; PK, pharmacokinetics; T-DM1, trastuzumab emtansine ; TME, tumor microenvironment. Nonetheless, few of the existing models treat the tumor as a distinct compartment with its own properties, and few consider the tumor vasculature, while none of them feature properties of the TME, such as spatial variations of the abnormal vasculature, uniform elevation of interstitial fluid pressure (IFP), and the dense extracellular matrix, which can strongly hinder drug delivery. Furthermore, the existing ADC models focus on standard cytotoxic ADCs. We have previously developed mathematical models to investigate the roles of vascular permeability, hydraulic conductivity, interstitial diffusion, IFP and lymphatics in the delivery of nano-sized drugs, including antibodies.34–38 Here, we extend our modeling framework to incorporate an IM-ADC, which has a payload that goes beyond standard cytotoxic agents, reflecting emerging designs that use immunotherapeutic agents as payloads, overcoming the limitations of chemotherapeutic agents.10 Specifically, a recent study constructed an IM-ADC by conjugating anti-programmed cell death ligand 1 (PD-L1) with a bifunctional immunomodulator D18.11 D18 is a bifunctional immunomodulator that enhances antitumor immunity through two primary mechanisms: (1) increasing KDM5A abundance, which suppresses PTEN expression and increases PD-L1 abundance via the PI3K-AKT-S6K1 pathway, and (2) activating TLR7/8 signaling pathways. This dual action drives a potent immune response by increasing T-cell activation and expansion, boosting populations of CD103+tumor-infiltrating dendritic cells, and ultimately enhancing the overall recruitment of activated CD8+T cells to tumors.39 The resulting IM-ADC, termed HE-S2, combines the anti-PD-L1 antibody, which blocks the programmed cell death protein 1 (PD-1)/PD-L1 interaction, with TLR7/8 signaling pathway activation. It also upregulates its PD-L1 target, thereby eliciting a stronger response to PD-1/PD-L1 blockade. Our model considers: (1) the delivery of HE-S2 ADC, (2) the interactions of cancer cells with immune cells, (3) pathophysiological properties of the TME, such as the dysfunctional tumor vessels, the uniformly elevated IFP, the hydraulic conductivity, and the permeability of the vasculature, (4) temporal and spatial distribution of ADCs, immune cells and tumor cells inside the tumor and the surrounding host tissue, (5) transport of ADCs through the vasculature and between the tumor-host tissue via convection and diffusion, (6) ADCs binding and internalization kinetics and (7) the tumor draining lymph nodes where the antigen presentation and the development of effector CD8+T cells (cytotoxic immune cells that specifically target the tumor cells) takes place. Our math model aims to reveal the effects of the TME on the efficacy of ADCs in eradicating solid tumors. The framework can also be adapted in the future to model bispecific antibodies, offering a versatile tool for studying next-generation antibody therapeutics. Methods Model compartments and implementation The mathematical model consists of three compartments (figure 1 and online supplemental figure S1): the vascular compartment, the tumor-draining lymph nodes and the tumor-host tissue compartments. The blood and tumor draining lymph nodes are assumed to have homogeneous concentrations of the modeled variables. For these compartments, ordinary differential equations are employed to describe temporal dynamics. However, the tumor-host tissue compartment is modeled as a sphere, assuming spherical symmetry (figure 1, online supplemental figure S1). For this compartment, partial differential equations are used to describe both temporal and spatial dynamics of the modeled variables. A detailed description of the model equations can be found in online supplemental information. Schematic of the model. The figure shows the three modeled compartments: tumor-host tissue, TDLN and blood. The blood and TDLN compartments consider temporal variations while the tumor-host tissue compartment is modeled as a sphere, considering both spatial and temporal variations of the model variables and parameters. The model integrates properties of the tumor and host tissue microenvironment by considering differences among those regions in the pore sizes of the vessels, the fraction of functional blood and lymphatic vessels and the hydraulic conductivity of the interstitial space. The tumor region changes size depending on the available nutrients, the proliferating tumor cells and the immune system response against the tumor. Key immune cells and processes of the immune system are modeled, such as cancer cell killing, antigen production, antigen uptake and presentation in the TDLN and production of effector cells. The transport of antibody-drug conjugates, as well as the transport of the individual components of the conjugate, are modeled. The ADCs can be transported through the vasculature along the tumor-host tissue interstitial space to the TDLN and back to the blood supply. Transport is governed by both convection and diffusion. The modeled IM-ADC, HE-S2, is an anti-PD-L1 antibody conjugated to the bifunctional immunomodulator D18. This IM-ADC acts through (1) ICB of PD-L1, (2) activation of the toll-like receptor 7/8 signaling pathway, and (3) enhancement of ICB sensitivity via PD-L1 upregulation. These mechanisms synergize to strengthen the immune response against the tumor. Created in BioRender. Stylianopoulos, T. (2026) https://BioRender.com/5xqjlhy. ADC, antibody-drug conjugate; APC, antigen-presenting cell; ICB, immune checkpoint blockade; IM-ADC, immune-stimulating ADC; NK, natural killer; PD-L1, programmed cell death ligand 1; TDLN, tumor-draining lymph node. At the symmetry side of the 1D interval (ie, the tumor center) the initial cancer cell density was assumed to have its peak value, and then it decreases in the radial direction. Regions of the domain where the cancer cell density is above a predefined threshold are assumed to be the tumor region and the rest of the region is classified as host tissue.34 35 The model parameters (online supplemental table S1) change from their abnormal-tumor value (at the tumor center) to their normal-host tissue value (at the tumor margin) as a function of cancer cell density, using a step function with a transition zone. Depending on the density of the cancer cells, the availability of nutrients and the interaction of the immune cells with the tumor cells, the tumor domain can either expand or shrink. The model equations for all compartments were solved using COMSOL Multiphysics (COMSOL, Burlington, Massachusetts, USA). Modeling the immune response The model simulates the key immune response processes, such as tumor cell killing, antigen uptake and antigen presentation. The immune cells are divided into innate and adaptive immune cells. The innate immune cells are divided into two categories. Immune cells, such as natural killer (NK) cells induce cytolysis of tumor cells and thus, produce antigen and immature antigen-presenting cells (IAPCs), such as dendritic cells that can uptake and present antigen. As shown in figure 1, the killing of cancer cells by the immune cells produces antigen in the tumor-host tissue compartment. There, the IAPCs can uptake the antigen and become antigen-presenting cells (APCs). All immune cells, as well as the antigen they produce, can be transported through the lymphatic vasculature to the tumor-draining lymph nodes (TDLNs). There, more antigen uptake and production of APCs can take place. In the TDLNs, APCs present the antigen to immature CD8+T cells to become effector CD8+T cells, which specifically target tumor cells. The resulting effector cells, as well as the rest of the immune cells, can be transported from the lymph nodes to the blood and through the vasculature back to the tumor-host tissue compartment to enhance the immune response against the growing tumor. It is worth noting that the source of innate immune cells, as well as the activation of adaptive immune cells, is affected by the amount of proinflammatory cytokines produced by the immune cells. Modeling individual components and the conjugate We model the transport and the effects of the ADC and the individual components that form the conjugate. ADC transport is modeled through convection and diffusion across the vessel walls, among the tumor-host tissue, to the lymphatics and back to the blood supply considering crucial parameters of the TME such as: the amount of the functional vascular density (functional vessels’ surface per unit volume), the pore size of the vessels wall (ie, vascular permeability), the hydraulic conductivity of the tumor interstitial space and the pressure gradients that drive the convective transport. The model also integrates properties of the drug (ADC or the individual components), such as the drug size, which affects the diffusion coefficient and the clearance of the drug as well as binding and internalization kinetics. The diffusion coefficients and the clearance rates of each drug were calculated based on their size.40 41 A detailed description of the model parameters and their values is provided in online supplemental table S1 and online supplemental table S2. The modeled ADC is HE-S2, which is an anti-PD-L1 conjugated to the bifunctional immunomodulator D18.11 We performed fitting to experimental data11 to calibrate model parameters related to tumor cells and their interactions, as well as those quantifying the effects of each component of the ADC. We simultaneously fit murine tumor growth curves from two cell lines—melanoma B16 and colon adenocarcinoma MC38—for the following groups, depending on the administered drug: control, the payload D18, anti-PD-L1 antibody and the ADC. Each administration was modeled using an exposure-averaged continuous source term beginning at the start of the treatment period, rather than modeled as repeated discrete bolus injections. We also fit the available murine tumor necrosis factor (TNF)-α+ CD8+ T cells data for the MC38 cell line for each administered drug that were captured at the end of the experiment (day 17 of the simulation).11 To do this, we used LiveLink for MATLAB (The MathWorks, Natick, Massachusetts, USA) to run multiple COMSOL simulations for each drug. For the simulations, we modeled the effect of each drug (ADC and ADC components) on the immune response and antitumor efficacy. The intensity of each effect is associated with a modeled parameter, which can be estimated by fitting tumor curves from the administration of the individual components of the ADC. Specifically, for D18 administration and TLR7/8 signaling pathway activation, we modeled the passive diffusion of D18 across the IAPCs and APCs membrane. Cells that internalized D18 were assumed to exhibit increased cytokine production, scaled by the parameter wc. (equations: 1, 12, 23, 58, 65, 73 at online supplemental information). For the case of the anti-PD-L1 antibody administration, we modeled the binding and internalization kinetics into the tumor cells. Cells that were bound by the antibody were assumed to experience an increased immune killing rate, scaled by parameter wa (equations 41, 43, 44 at online supplemental information). For the ADC case, we modeled the binding and internalization kinetics of the conjugate to immune and cancer cells. We also incorporated increased cytokine production induced by the payload D18 (parameter wc), an increased immune killing rate due to the anti-PD-L1 antibody (parameter wa), and increased sensitivity to PD-L1 blockade induced by D18 via PD-L1 upregulation (parameter ws; equations 42, 43 in the online supplemental information).11 More information on how these effects were applied to the model equations can be found in the online supplemental information. Results Model calibration with published data The model calibration to the tumor growth data is shown in figure 2. Importantly, we did not fit a separate parameter set to each individual growth curve. Instead, we performed a simultaneous calibration using all tumor-volume observations across the four treatment groups within both cell lines (MC38 and B16), comprising 36 tumor-volume data points in total. Although the full model contains multiple parameters, only 12 were adjusted during calibration (online supplemental table S2). Most parameters were fixed based on literature or previous estimates (online supplemental table S1) or were shared across curves to enforce a consistent mechanistic structure. In particular, the drug-mechanism effect parameters wc, wa, and ws were treated as global parameters shared across both cell lines, whereas only a limited set of tumor-immune interaction parameters was allowed to be cell-line-specific (eg, intrinsic tumor growth and immune-mediated tumor cell killing). Simultaneously, the model was also calibrated to the murine TNF-α+ CD8+ T cells data for the MC38 cell line. Here, we calibrated the modeled increase of the average concentration of CD8+T cells in the tumor from the control, as shown in figure 2. CD8+T cell data were not available for B16 tumors. The figure shows the fitting of model parameters to murine MC38 and B16 tumors.11 The fitting is performed for the following modes of administration: control, the payload D18, anti-PD-L1 antibody and ADC. On the left, each tumor growth curve presents the model predictions, and the bullet points represent the experimental data. The vertically dashed black lines show the start of the treatment. On the right of each growth figure, the simulated versus experimental volume is presented. The closer the dots are to the diagonal, the more accurate the fitting is. Similarly, the upper-right figure shows the model’s fit to CD8+T cells. For each calibration, the MAE of the fitting is displayed. ADC, antibody-drug conjugate; MAE, mean average error; PD-L1, programmed cell death ligand 1. As shown in figure 2, the model fits the tumor growth curves. The mean average error (MAE) is reported for each fit; the closer the simulated volume is to the experimental volume, the closer the MAE is to zero, indicating a better fit. The figure shows that the model reproduces the experimental data with good accuracy. D18 and anti-PD-L1 individual administrations reduce tumor volume compared with the control group. This occurs due to increased proinflammatory cytokine production induced by D18 and enhanced immune cell-mediated tumor cell killing with anti-PD-L1, respectively, and we observe that the model captures this effect. On the other hand, with ADC administration, these effects synergistically inhibit tumor progression. In addition to individual effects, when D18 and anti-PD-L1 are combined, D18 increases sensitivity to immune checkpoint blockade, further enhancing the immune response against the tumor. Altogether, this synergy explains the stronger reduction in tumor growth observed with ADC treatment. Simultaneously with the calibration of the growth curves, we ensured that the increase of the CD8+T cells in the model reflects the increase in the available experimental data.11 Specifically, on day 17 of the simulation, experimental data show a negligible increase during administration of the payload and the antibody compared with control, whereas there is more than a threefold increase with ADC treatment. Mechanistic insights of model predictions Following the calibration of model parameters, we investigated the behavior of model variables that cannot be measured experimentally to provide mechanistic insights into the underlying processes. We generated plots to further investigate the changes in the model variables that led to reduced tumor growth after ADC administration. Figure 3 presents some of the model variables from the MC38 data fitting. The distributions of these variables are shown in the 1D tumor-host tissue domain with spherical symmetry, where 0 is the tumor center and the colored solid lines indicate the tumor margin at different time points. For most injection modes, we observe that the IFP within the tumor, a common biomarker of treatment response, approaches the vascular pressure and then returns to normal values (ie, 0 mm Hg) as we move into the host tissue (figure 3). This is not the case, however, with the ADC treatment on day 17, where the reduced tumor volume led to a decrease in IFP towards normal values. Distribution of model variables in the tumor host-tissue domain at various time points (day 0, 8.5, 17 of the simulation). The results present the distribution of: (1) interstitial fluid pressure, (2) drug concentration and (3) effector CD8+T cells from the fitting of MC38 tumors. For each variable, graphs for the following modes of injection are presented: control, the D18 payload, anti-PD-L1 antibody and the antibody-drug conjugate. The x-axis of the graphs depicts the radial direction from the tumor center. The vertical solid lines indicate the tumor-host tissue boundary, which is different at various time points due to tumor progression. PD-L1, programmed cell death ligand 1. Figure 3 also depicts lower concentrations of D18 drug in the tumor tissue compared with the anti-PD-L1 antibody and the ADC. This can be explained by the significant size difference between the D18 drug and the antibody or ADC. D18 is a small molecule that can be cleared rapidly. Also, nanosized particles, such as antibodies and ADCs, are preferentially delivered through the tumor vasculature, where enlarged vessel-wall pores facilitate their extravasation. That is not the case with the normal vasculature in the host tissue, where nanosized particles are larger than the available vessel wall pores and are physically excluded by the normal tissue. On the other hand, small molecules like D18 can pass through the normal vessels wall and have larger concentrations in the host tissue. The reduced concentration of intratumor D18 is due to elevated IFP, which reduces convective transport inside the tumor. The model can also predict the distribution of CD8+T cells (figure 3). The results show that CD8+T cells infiltrate the tumor region only in the ADC case, whereas in all other cases we observe reduced numbers of CD8+T cells within the tumor compared with the host tissue. This indicates a more robust immune response against the tumor in the case of ADC administration Role of tumor microenvironment in ADC efficacy Next, we studied how physiological properties of the TME can improve ADC delivery by performing parametric analyses in which two TME parameters were varied simultaneously for each heatmap. For these simulations, the two parameters shown on the heatmap axes were varied, while all other model parameters were fixed to the values listed in online supplemental tables S1 and S3. The resulting heatmaps are presented in figures 4–6. Specifically, we varied the tumor vessel wall pore radius, the hydraulic conductivity of the tumor interstitial space, and the tumor functional vascular density. The vessel wall pore size determines the permeability/leakiness of the vessels and thus controls the fluid that enters the tumor interstitial space from the vessels and, as a result, the IFP. It also determines steric and hydrodynamic interactions for the transvascular transport of the drug from the vessels into the tumor. The larger the pores relative to the drug size, the lower these interactions are, and thus transport increases. The hydraulic conductivity of the tumor determines the clearance of the fluid from the tumor. The higher the conductivity, the more efficient the transport of the fluid is towards the tumor periphery, which also affects IFP. For each parametric analysis, the tumor volume, ADCs inside the tumor, the effector CD8+T cells inside the tumor and the effector CD8+T cells inside the TDLNs are presented. The effect of TME properties on model predictions. The figure shows heatmaps of ADC treatment for a range of tumor vessel wall pore radius and tumor interstitial space hydraulic conductivity values. The heatmaps present four types of results: the tumor volume, the mass of the ADCs inside the tumor, the mass of effector CD8+T cells inside the tumor and the mass of effector CD8+T cells inside the tumor draining lymph nodes. The intratumoral ADC and effector CD8+T cell quantities represent total masses integrated over the tumor volume. ADC, antibody-drug conjugate; TME, tumor microenvironment. By varying the hydraulic conductivity of the tumor tissue along with the vessel wall pore radius (figure 4), we observe that increasing the pore size allows for a larger amount of ADC to pass through the larger pores. For smaller vessel wall pore sizes, the vessels become less leaky, reducing fluid leakage and IFP so that the impact of the hydraulic conductivity to clear excessive fluid from the interstitial space diminishes. For larger pore sizes, the IFP increases and thus, increasing the hydraulic conductivity can cause the drug to wash out of the tumor to the host tissue more rapidly. Therefore, better therapeutic outcomes are observed for vessel wall pore radii greater than 30 nm and hydraulic conductivities less than 10−6 cm2/mm Hg-s. Interestingly, the total CD8+T cell mass in the tumor increases with tumor volume, largely because larger tumors provide more tissue space to accommodate more cells, whereas smaller tumors contain fewer cells overall. The opposite effect is seen for the CD8+T cell mass in the lymph, which, as expected, decreases when the mass in the tumor increases. The parametric analysis in figure 5 depicts the effect of the functional vessels and tumor hydraulic conductivity on the efficacy of ADC treatment. More functional vessels in the tumor reduce tumor volume by transporting more ADCs to the tumor site. This is mainly observed when the vascular density surpasses 70 (1/cm). Larger amounts of functional vessels cause an increase in trafficking of the immune cells, resulting in lower amounts in the TDLNs and larger amounts in the tumor. More CD8+T cells in the tumor induce a better treatment response. The effect of TME properties on model predictions. The figure shows heatmaps from ADC treatment for a range of values of the hydraulic conductivity of the tumor interstitial space and the tumor functional vascular density (the amount of functional vessel surface in a unit volume). The heatmaps present four types of results: the tumor volume, the mass of the ADCs inside the tumor, the mass of effector CD8+T cells inside the tumor and the mass of effector CD8+T cells inside the tumor draining lymph nodes. The intratumoral ADC and effector CD8+T cell quantities represent total masses integrated over the tumor volume. ADC, antibody-drug conjugate; TME, tumor microenvironment. The last analysis (figure 6) shows the effect of the functional vasculature and vessel permeability on the efficacy of ADC treatment. When the tumor vessel wall pores are small (<40 nm in radius), antitumor effects are not strong because not enough ADCs are able to pass through the vasculature. When the pores are large enough (>40 nm in radius), the number of functional vessels plays a critical role in the treatment outcome, with more functional vessels enhancing immune response. The results suggest that pores >40 nm in radius must be combined with a vascular density larger than 70 (1/cm) for optimal therapeutic outcomes. Also, as in figure 5, more functional vessels increase CD8+T cell trafficking from the TDLNs to the tumor, thereby improving treatment response. The effect of TME properties on model predictions. The figure shows heatmaps of ADC treatment for a range of tumor vessel wall pore radii and tumor functional vascular density (the amount of functional vessel surface per unit volume). The heatmaps present four types of results: the tumor volume, the mass of the ADCs inside the tumor, the mass of effector CD8+T cells inside the tumor and the mass of effector CD8+T cells inside the tumor draining lymph nodes. The intratumoral ADC and effector CD8+T cell quantities represent total masses integrated over the tumor volume. ADC, antibody-drug conjugate; TME, tumor microenvironment. Figure 6 shows an interesting effect at low functional vascular density, where increasing vessel wall pore radius can increase tumor volume despite increasing total intratumoral ADC mass. This is because the effector CD8+T cell values represent total intratumoral mass, not local concentration or density. Thus, when tumor volume is larger, a similar total CD8+T cell mass is distributed over a larger volume, reducing the effective CD8+T cell concentration and immune-mediated tumor-cell killing. This effect emerges from the balance among ADC delivery, tumor volume, and immune-mediated killing. Discussion Our mechanistic mathematical model provides a framework for exploring the spatiotemporal dynamics and efficacy of ADCs, specifically HE-S2, within the complex TME. It aligns with experimental data indicating that ADCs enhance tumor response compared with the administration of the individual components that make up the conjugate.11 Additionally, it simulates unobserved aspects of the experiment, such as IFP, drug, and immune cell spatiotemporal distribution. The model predicts lower D18 levels in the tumor than in antibody/ADC because D18 is a small molecule that clears rapidly and is less retained, whereas antibody/ADC shows greater tumor delivery and retention through leaky tumor vasculature. It can also be excluded from normal tissue. In addition, the model suggests a positive feedback loop under ADC treatment: improved intratumoral ADC exposure promotes a stronger CD8+T cell response and infiltration, thereby increasing tumor cell killing and reducing tumor volume. Tumor shrinkage then lowers intratumoral IFP, which alleviates transport barriers and can further improve intratumoral delivery/retention of the ADC, reinforcing the therapeutic effect. Additionally, the model results depict how TME properties affect ADC efficacy. Specifically, parameters, such as the tumor vascular pore size, hydraulic conductivity of the tumor interstitial space and the functional vascular density emerge as critical determinants of the efficacy of ADCs. Specifically, our analysis shows that ADC efficacy is maximized when tumor vessels are permeable enough (pore radius >40 nm) and sufficiently abundant (vascular density >70 (1/cm)) to enable ADC extravasation and immune trafficking. At the same time, hydraulic conductivity should not be too high, because it can increase convective washout to host tissue, reducing intratumoral ADC retention and weakening the CD8+T cell response. These findings support the administration of TME normalization agents that increase functional vascular density prior to ADC administration.14 42 43 However, these approaches should preserve sufficient vascular permeability by maintaining vessel pore radius >40 nm, ensuring the pores remain large enough for ADC extravasation and effective intratumoral delivery. It is important to note that reduced IFP alone is not sufficient to ensure improved ADC delivery. IFP, vascular perfusion, and vascular permeability are related but distinct determinants of macromolecular transport. In our model, elevated IFP directly reduces the transvascular pressure gradient and therefore limits convective extravasation of ADCs into the tumor. However, even if pressure gradients are restored, heterogeneous or insufficient perfusion can still limit the amount of drug reaching tumor vessels, particularly in poorly perfused tumor regions, such as the tumor center. Thus, successful TME normalization for ADC delivery requires a balance among improved perfusion, adequate vascular density, sufficient vascular permeability, and reduced IFP. This distinction is important because some normalization strategies, including anti-vascular endothelial growth factor (VEGF)-based approaches, may improve perfusion and reduce IFP but can also reduce vascular permeability or vascular volume, thereby decreasing uptake of large therapeutics such as antibodies and ADCs. Conversely, strategies that alleviate vessel compression or improve vessel functionality while preserving sufficient permeability may enhance extravasation and intratumoral delivery. Consistent with this interpretation, the model’s predictions suggest that these delivery-favorable TME changes may be achievable with existing TME-modulating strategies. For example, vascular normalization approaches (eg, anti-VEGF therapy) may be most relevant in hyper-permeable tumors, whereas stress-alleviation approaches (eg, losartan or ketotifen) may be more relevant in tumors dominated by vessel compression.44–47 The dominant physical barrier to delivery depends on the vascular phenotype. In tumors with hyper-permeable but poorly functional vessels (eg, subsets of glioblastoma), anti-VEGF “vascular normalization” can reduce vessel hyperpermeability/pore size while improving perfusion and functional vessel density, and clinical studies in patients with glioblastoma treated with anti-VEGF agents (eg, cediranib) show that increases in perfusion consistent with normalization are associated with improved outcomes.44 In tumors with abundant compressed/collapsed vessels, improving delivery requires vessel decompression by alleviating solid stress; this can be achieved by repurposing agents such as losartan, which reduces fibrosis/solid stress, re-opens vessels, and improves perfusion and drug delivery in desmoplastic tumor models, and has been evaluated clinically in combination regimens.45 More recently, ketotifen has been shown to restore vessel functionality/perfusion and to enhance T-cell infiltration and response to anti-PD-L1-based therapy in preclinical models, consistent with a stress-alleviation/immune-supportive remodeling mechanism.46 Finally, for tumors that display both hyper-permeability and vessel compression, a combined strategy, normalizing abnormal vessels while also alleviating solid stress, may be required to optimize perfusion and intratumoral delivery.47 An additional translational consideration for IM-ADCs/ISACs is the potency of the immune-stimulating payload. Similar to ultra-potent cytotoxic ADC payloads, highly potent payloads can provide strong pharmacologic activity but might also narrow the therapeutic window by increasing the risk of systemic toxicity. This concern is particularly relevant for TLR7/8 agonists, for which systemic administration has been limited by narrow therapeutic indexes, variable exposure, and dose-limiting toxicities.48 Although the clinical tolerability of HE-S2 has not been established, D18 exhibits low-nanomolar TLR7/8 agonist activity,11 raising potential safety concerns for single-agent immune-stimulating ADCs, especially if payload exposure approaches levels associated with clinically administered TLR7 agonists such as 852A.49 These considerations further motivate the use of mechanistic modeling to support the rational design of next-generation IM-ADCs. For example, models such as the one developed here could be extended to evaluate whether lower-potency immune-stimulating payloads, higher but safer dosing, alternative dosing schedules, or TME-normalization strategies can preserve intratumoral immune activation while reducing systemic toxicity risk. Despite the strength of our analysis, our model has certain limitations as several assumptions were made to keep the number of model parameters minimal and maintain a level of complexity that allows a reliable numerical solution. Specifically, the model did not account for the ADC surface charge and configuration which, along with the ADC size, can affect the transport properties.12 15 Additionally, we assumed a uniform vessel wall pore radius while there is a distribution of pore radiuses. Also, no changes in the diffusion coefficient of the ADCs were considered due to variations among tumor types. The tumor host tissue domain assumed spherical symmetry but, in reality, variations exist in both the polar and azimuthal directions. Furthermore, only key processes of the immune response were modeled as modeling all aspects of the immune system is impossible due to the vast complexity. Finally, individual repeated administrations from the He et al11 dosing schedule were not simulated as separate bolus injections. Instead, treatment was approximated using an exposure-averaged continuous source term beginning at the start of treatment, because time-resolved plasma and intratumoral PK data were unavailable. Explicit multidose PK modeling may affect quantitative fitted parameters but this approximation is not expected to alter the main mechanistic findings of our study. Nevertheless, the model is able to establish generalizable principles in ADC delivery despite these limitations. It tries to bridge the gap of ADC modeling by incorporating immune system interactions and the effect of TME. This study aims to establish a predictive framework that can guide future experimental studies and inform clinical translation. Supplementary material Footnotes Funding: This project was supported in part by funding from the European Research Council (ERC) under the European Union’s Horizon Europe research and innovation program (grant agreement no 101141357) to TS. This project was also supported by grants R01-CA247441, R21-EB031982, and U01-CA261842 to LLM; and grants R01-CA259253, R01-CA208205, R01-NS118929, U01-CA261842, and U01-CA224348, Outstanding Investigator Award R35-CA197743, and grants from the National Foundation for Cancer Research, Jane’s Trust Foundation, Niles Albright Research Foundation, and Harvard Ludwig Cancer Center to RKJ. The funders did not influence the results/outcomes of the study. Views and opinions expressed are those of the authors only and do not necessarily reflect those of the European Union or the ERC Executive Agency. Neither the European Union nor the granting authority can be held responsible for them. Provenance and peer review: Not commissioned; externally peer reviewed. Patient consent for publication: Not applicable. Ethics approval: Not applicable. Data availability free text: All data relevant to the study are included in the article. All relevant data supporting the findings of this study are presented within the manuscript and figures. Data availability statement All data relevant to the study are included in the article or uploaded as supplementary information. References Associated Data Supplementary Materials Data Availability Statement All data relevant to the study are included in the article or uploaded as supplementary information.