Maximizing response to intratumoral immunotherapy in mice by tuning local retention

Direct injection of therapies into tumors has emerged as an administration route capable of achieving high local drug exposure and strong anti-tumor response. A diverse array of immune agonists ranging in size and target are under development as local immunotherapies. However, due to the relatively recent adoption of intratumoral administration, the pharmacokinetics of locally-injected biologics remains poorly defined, limiting rational design of tumor-localized immunotherapies. Here we define a pharmacokinetic framework for biologics injected intratumorally that can predict tumor exposure and effectiveness. We find empirically and computationally that extending the tumor exposure of locally-injected interleukin-2 by increasing molecular size and/or improving matrix-targeting affinity improves therapeutic efficacy in mice. By tracking the distribution of intratumorally-injected proteins using positron emission tomography, we observe size-dependent enhancement in tumor exposure occurs by slowing the rate of diffusive escape from the tumor and by increasing partitioning to an apparent viscous region of the tumor. In elucidating how molecular weight and matrix binding interplay to determine tumor exposure, our model can aid in the design of intratumoral therapies to exert maximal therapeutic effect.

Intratumoral administration enables the use of potent immune modulators to treat cancer. As a result, the number of trials investigating local administration of cancer therapies has boomed1. Modalities undergoing intratumoral testing2,3 include small molecules4, nucleic acids5, proteins6, viral-vector-based agents7, and cell therapies8,9.

Response to most local therapies hinges on tumor exposure10, which depends on a drug’s pharmacokinetic properties. Amongst the features that influence pharmacokinetics, molecular size governs several transport phenomena relevant to tumor retention (e.g. extravasation, diffusion, accessible interstitial volume)11. Moreover, molecular size varies tremendously across therapeutic modalities. Previous characterizations of the biodistribution of large locally injected particles has provided a pharmacokinetic perspective on nanoparticle (~10–100 nm), liposomal (~20 nm−1 µm), viral-vector (~100 nm), and cell-based therapies (~10–20 µm)12,13. However, for smaller-sized protein therapies (e.g. peptides, nanobodies, cytokines, antibodies) ranging from 1 to 10 nm in molecular radius, it remains unclear quantitatively how their differences in size impacts tumor exposure. For proteins, tumor retention is also affected by target binding. Drug pharmacokinetics seems intuitive, yet analyses outlining exactly how target binding and molecular size impacts the tumor exposure of intravenously injected biologics have proven pertinent to their design, optimization and clinical translation14,15,16,17,18,19,20,21. Intratumorally administered biologics would benefit from similar foundational pharmacokinetic analysis, which presently does not exist.

Pharmacokinetic insight can inform improvements to immune modulators, such as the cytokine interleukin-2 (IL-2). IL-2 is capable of eliciting durable tumor regression via natural killer (NK) and T cell activation but is plagued by severe systemic toxicity. Thus IL-2 has been clinically evaluated as a local therapy in different formats (e.g. wild-type22,23, PEGylated24, fused to fibronectin targeting diabody25,26, fused to GD2 binding antibody27, and fused to epithelial cell adhesion molecule binding antibody28). Using IL-2 as an example, here we report the role of its pharmacokinetics in controlling therapeutic efficacy after intratumoral injection and define strategies to maximize anti-tumor effect by tuning local retention. To tune local retention, we fuse IL-2 to proteins of different size and matrix-binding affinity. By varying these two attributes, we systematically alter the local pharmacokinetic features of IL-2 fusion proteins and define how each interact to impact local retention and anti-tumor efficacy in mice. To extend beyond the empirically testable parameter space, we employ computational tools to predict IL-2’s intratumoral exposure across a spectrum of theoretical sizes (1 kD to 1000 kD) and matrix-binding affinities (micromolar to subnanomolar). Although matrix-binding increases local persistence of intratumorally injected agents as reported previously29, we find that tumor exposure can also be improved significantly by increasing molecular size, and maximal exposure is obtained by combining both approaches. Tracking intratumorally injected proteins using positron emission tomography (PET) reveals that increased molecular size not only slows overall diffusive escape from the tumor, as our computational model predicts, but increases the proportion of injected protein halted in potentially viscous regions of the tumor30. Collectively, this framework delineates tunable pharmacokinetic features to aid in the engineering of local immunotherapies for maximal anti-tumor effect.

Engineering IL-2 fusions of varying size and matrix affinity

To determine whether an intratumoral immunotherapy’s efficacy depends on its pharmacokinetic features, we sought to create a panel of immunotherapeutic agents varying in molecular weight and collagen-binding affinity. Recently, we and others validated that collagen, an abundant matrix protein, is an effective and tumor-agnostic target for retaining locally administered immunotherapies29,31,32,33,34. To target immunotherapies to collagen, we used the ectodomain of murine leukocyte-associated immunoglobulin-like receptor-1 (LAIR), an endogenous collagen-binding protein35. In order to eliminate LAIR’s collagen affinity while preserving its size, we engineered an inert version of LAIR using yeast surface display36. Wild-type LAIR displayed on the surface of yeast specifically bound to collagen-related peptide (CRP), a soluble mimetic of collagen (Fig. 1a and Supplementary Fig. 1a, b)37. We generated a library of yeast displaying LAIR mutants (Fig. 1a, b) and sorted clones with diminished binding to CRP (Fig. 1b, c and Supplementary Fig. 2a, b). When expressed recombinantly, these LAIR mutants still retained residual binding to native collagen (Supplementary Fig. 3a–j). Based on the sequences of enriched yeast clones (Supplementary Fig. 2a, b), we then rationally combined mutations to residues deemed critical for CRP binding38. Two mutations together, at position 41 (R41A) and position 43 (E43A), eliminated binding to collagen and constitute LAIRx, an inert size-matched protein (Fig. 1d).

a Schematic of yeast library displaying mutagenized-LAIR (white) as Aga2 (gray) fusion proteins. LAIR expression was detected using a fluorescent antibody against c-myc epitope (green) and binding to biotinylated collagen-related peptide (CRP) was detected using fluorescent streptavidin (SA-AF647, red). b Successive flow assisted cells sorts isolating LAIR mutants with diminished binding to biotinylated CRP. c Equilibrium binding titration of biotinylated CRP on enriched yeast from panel b (mean + s.d.; n = 8) compared to wild-type LAIR-displaying yeast (n = 1). Binding was measured by median fluorescence intensity of SA-AF647 by flow cytometry. d LAIR and LAIRR41A E43A, henceforth LAIRx, binding to collagen measured by enzyme-linked immunosorbent assay (ELISA) (mean + s.d., n = 6 for LAIR and n = 3 for LAIRx). e Collagen binding affinity (Kd, equilibrium dissociation constant) of IL-2 fusion proteins in small and large molecular weight formats (kD, kilodaltons) measured by ELISA (mean + s.d; n = 3 for all groups except n = 6 for LAIR-IL2). f Dose-dependent proliferation of CTLL-2 cells exposed to IL-2 fusion proteins (mean + s.d.; n = 3) gSchematic of B16F10 tumor study timeline (top) and treatment components (bottom). Mice were inoculated with 1 × 106 B16F10 cells subcutaneously in the right flank on day 0. Treatments were administered on day 6 and 12. TAA, tumor-associated antigen; i.tu., intratumoral; i.p., intraperitoneal. h, i Tumor growth (left, mean + s.d.) and survival (right) over time (n = 10 mice for the PBS (i.tu.) and PBS (i.tu.) + TA99 (i.p.) groups, and n = 7 mice for all other groups). Tumor area is shown until a mouse in the group is euthanized. Statistical significance of each treated group’s tumor area and survival (top right corner) was calculated by using a one-tailed Student’s t-test and log-rank Mantel-Cox test, respectively, versus the PBS (i.tu.) + TA99 (i.p.). Other survival comparisons (legend adjacent) were generated by a log-rank Mantel-Cox test. *P < 0.03; **P < 0.002; ***P < 0.0002; ****P < 0.0001; n.s. not significant. Source data for panel c–f and h, i are provided in the Source Data file. Panels a and g created with BioRender.com.

Full size image

Like many immunotherapies, IL-2 elicits exposure-dependent anti-tumor effects, making it a model immunomodulator to test the impact of pharmacokinetic properties on therapeutic efficacy. We fused IL-2 to LAIR or LAIRx, generating two small-format (32 kD) immunotherapies LAIR-IL2 and LAIRx-IL2 (Supplementary Fig. 4a–c) with collagen affinities of 220 nM and >1 μM, respectively (Fig. 1e). Because CRP is an imperfect collagen mimetic, LAIR mutants demonstrating higher-affinity CRP-binding did not exhibit increased affinity to native collagen (Supplementary Fig. 5a–d). Instead, we exploited avidity to increase binding strength by using dimers of LAIR and LAIRx. To passively increase molecular weight, we also included a large inert protein, mouse serum albumin (MSA) containing a histidine to glutamine mutation (H464Q) that abrogates neonatal Fc receptor-mediated recycling39. The large-format (137 kD) immunotherapies (Supplementary Fig. 4c), LAIR-LAIR-MSAH464Q-IL2, LAIR-LAIRx-MSAH464Q-IL2, and LAIRx-LAIRx-MSAH464Q-IL2 demonstrated collagen affinities of 13 nM, 130 nM, and >1 µM, respectively (Fig. 1e). All IL-2 fusion proteins were equivalently bioactive in inducing T cell proliferation (Fig. 1f).

Size and matrix affinity of local IL-2 determine response

As a single agent, IL-2 therapy cannot overcome immune resistance barriers posed by the tumor microenvironment. However, we and others have demonstrated previously that IL-2, when combined with a tumor-targeting antibody, can achieve exposure-dependent tumor control mediated by local stimulation of cytotoxic CD8+ T and NK cells29,40,41,42,43. Therefore, we treated mice bearing established, immune-infiltrated (Supplementary Fig. 6), subcutaneous flank B16F10 melanoma tumors with an antibody directed against a tumor-associated antigen tyrosinase-related protein-1 (anti-TYRP-1, or TA99) intraperitoneally and with intratumoral injections of our IL-2 fusion proteins (Fig. 1g). All IL-2 fusion proteins imparted tumor growth delay (Fig. 1h, i); however, treatment with the large-format fusions (Fig. 1h) further slowed tumor growth (p = 0.01 for LAIRx-LAIRx-MSAH464Q-IL2 vs. LAIRx-IL2) and enhanced survival (p = 0.01 for LAIRx-LAIRx-MSAH464Q-IL2 vs. LAIRx-IL2; p = 0.04 for LAIR-LAIRx-MSAH464Q-IL2 vs. LAIR-IL2) compared to their small-format counterparts (Fig. 1i). Within each size tier, therapies with collagen binding improved overall survival (Fig. 1h, i). However, the difference in survival after treatment with the two large-format collagen binding variants LAIR-LAIR-MSAH464Q-IL2 and LAIR-LAIRx-MSAH464Q-IL2 of different collagen affinity was apparently not statistically significant (Fig. 1h). Strikingly, the tumor growth delay (p = 0.25) and survival benefit (p = 0.21) of targeted small-format LAIR-IL2 was equivalent to untargeted large-format LAIRx-LAIRx-MSAH464Q-IL2. Previous work has shown that systemically delivered large-format IL-2 (>70 kD) controls tumors better than wild-type IL-2 (15 kD) due to increased tumor exposure arising from size-based enhancements in circulatory half-life40. To compare solely the local effects of LAIR-IL2 and LAIRx-LAIRx-MSAH464Q-IL2, we matched the systemic IL-2 exposure by supplementing local treatments with LAIRx-LAIRx-MSAH464Q-IL2 delivered intraperitoneally (Supplementary Fig. 7a). We found again that the tumor growth delay and survival benefit (Supplementary Fig. 7b, c) of small targeted LAIR-IL2 was indistinguishable from large untargeted LAIRx-LAIRx-MSAH464Q-IL2. Their therapeutic equivalency indicates that, even for therapies injected directly into the tumor, molecular size is very relevant to local exposure.

Model predicts duration of local treatment activity

To understand and predict how a treatment’s pharmacokinetic properties skew therapeutic efficacy, we recapitulated the kinetic and transport dynamics occurring during and after intratumoral injection into a computational model. During an intratumoral injection, the fluid in the tumor interstitium (i.e. extracellular space between cells in the tumor) is instantaneously displaced by the volume injected. Theoretically, for an injection into a well-mixed tumor interstitium, the volume held-up in the tumor (Vholdup) is related to the tumor’s size (Vtumor) and interstitial void volume fraction (εtumor):

The model assumes that any injected volume not held-up is released immediately into circulation, which is in agreement with observations of systemic dissemination during intratumoral injection44,45. (Fig. 2a) After injection, held-up IL-2 fusion protein toggles between four states within the tumor: unbound, bound to collagen, bound to IL-2 receptor (IL-2R), or concurrently bound to both targets (Fig. 2b). Protein-binding interactions are dictated by association (kon) and dissociation (koff) rate constants, and the concentration of a complex and its constituents. The model includes two drug removal processes: (1) consumption by IL-2R+ cells in the tumor and (2) clearance from blood. Consumption was captured using the internalization rate (kinter) of the IL-2:IL-2R complex. Since IL-2 induces NK and T cell proliferation, we incorporated a logistic growth function to simulate the increase in IL-2R in response to IL-2 consumption. Since uniformly high interstitial pressures within tumors are thought to eliminate fluid pressure gradients across vessel walls and interstitium, our model of intratumoral drug transport describes diffusive, not convective, processes18,46. Protein intravasation (kintrav) and extravasation (kextrav) are captured by diffusive transport across a two-pore model of the intratumoral capillaries with a capillary permeability that depends on the protein’s molecular weight, a relationship validated by Schmidt et al.16. Within the blood, the rate of plasma clearance (kclear) uses another validated empirical relationship also dependent on the protein’s molecular weight16. Due to the lack of functional lymphatic vessels observed in tumors, transport into the lymphatic system is not modeled here47,48. We translated the relationships that define kinetic, transport, and cellular dynamics into a system of ordinary differential equations (ODEs) (Supplementary Tables 1–3, the code generated is also publicly available49.) Solving the ODEs yields the total extracellular concentration of the injected IL-2 fusion protein, target-bound and freely diffusing, in the tumor over time. (Supplementary Fig. 8a) To correlate the model output with biological activity, we evaluated the tumor concentration of locally-injected IL-2 fusion protein ([Intratumoral IL-2 Fusion Protein]) over time using this relationship:

where EC50,IL2, the IL-2:IL-2R half maximal effective concentration, is 2.4 × 10−7 M (Eq. 2)50. When the intratumoral injected IL-2 concentration far exceeds EC50,IL2, the fractional activity approaches 1. However, when the concentration equals EC50,IL2, thereby eliciting a half-maximal response, the fractional activity is 0.5. (Supplementary Fig. 8b) The area-under-curve (AUC) of the fractional activity over time estimates how many days after local administration the injected protein is active within the tumor. (Supplementary Fig. 8b)

Model simulating the disposition of intratumorally injected IL-2 collagen-binding fusion protein over time (t). a During an intratumoral injection, IL-2 fusion protein is instantaneously transferred from the syringe to the tumor compartment (circle) of volume capacity Vholdup and blood compartment (square) of volume Vblood b After injection, diffusion of the injected protein between the tumor and blood compartment occurs based on the rate of intravasation (kintrav) and extravasation (kextrav). Removal of the injected protein from blood occurs at a rate of clearance (kclear). Ø indicates removed protein. The italicized rates depend on the molecular weight of the injected protein. Within the tumor interstitium (inset), the injected protein toggles between target-bound states. The injected protein (free or collagen-bound) binds IL-2 receptor (IL-2R) according to rate of association (kon,IL2R) and dissociation (koff,IL2R) with IL-2R. When bound to IL-2R, the injected-protein-IL-2R complex is internalized and degraded at the rate of IL-2R internalization (kint,IL2R). IL-2 signaling increases the concentration of IL-2R over time. The injected protein (free or IL-2R bound) binds collagen according to the rate of association (kon,col) with collagen and rate of dissociation with collagen (koff,col). Glossary of the model’s variables, initial conditions, and equations are provided in Supplementary Tables 1–3. c Model’s prediction of the intratumoral activity duration of an injected IL-2 fusion protein varying in molecular weight (kD) and collagen affinity (Kd equilibrium dissociation constant in M units) The circles on the heatmap indicate the model’s prediction for the proteins LAIR-LAIR-MSAH464Q-IL2 (red), LAIR-LAIRx-MSAH464Q-IL2 (orange), LAIRx-LAIRx-MSAH464Q-IL2 (yellow), LAIR-IL2 (purple), and LAIRx-IL2 (pink). Each protein’s predicted days of activity is adjacent to color bar. d Percentage of mice from Fig. 1h, i surviving 100 days after treatment (outlined box). On day 100, surviving mice were inoculated with 0.1 M B16F10 on the contralateral left flank. The percentage of mice surviving for another 100 days after tumor rechallenge (filled box). e Correlation between the model’s predicted intratumoral activity in panel c and hazard ratio for treatments from Fig. 1h, i compared to PBS (i.tu.) + TA99 (i.p.) treated control group. Hazard ratio computed by log-rank method. Source data for panel d, e are provided in the Source Data file. Panel a and b created with BioRender.com.

Full size image

By inputting a spectrum of collagen affinities and molecular weights into the model, we generated a heat map relating the size and collagen affinity of injected immunotherapies to their duration of activity in the tumor. (Fig. 2c). The activity duration predicted by the model is 1.33 days (~32 h) for LAIR-LAIR-MSAH464Q-IL2, 0.99 days (~24 h) for LAIR-LAIRx-MSAH464Q-IL2, 0.44 days (~11 h) for LAIRx-LAIRx-MSAH464Q-IL2, 0.24 days (~6 h) for LAIR-IL2, and 0.13 days (~3 h) for LAIRx-IL2 (Supplementary Fig. 8b and Fig. 2c) The model’s rank-order of treatment activity duration aligns exactly with their elicited in vivo survival and immunological memory (Fig. 2d) as well as the treatment’s hazard ratio (Fig. 2e).

The primary in silico sink for locally injected IL-2 fusion proteins was clearance from circulation and not uptake by T cells, in part due to the low initial abundance of IL-2R+ cells in our poorly inflamed B16F10 tumor model51. (Supplementary Fig. 9a) Despite significant drug dissemination to blood both during and after the simulated injection (Supplementary Fig. 9b), the predicted total systemic exposure remained orders-of-magnitude lower than tumor exposure (Supplementary Fig. 9c). Based on a sensitivity analysis of the model’s rates and initial conditions, the parameters-in addition to collagen affinity and molecular weight- that perturb the output tumor exposure are: tumor collagen concentration, the initial tumor concentration of injected protein, and collagen’s half-life (Supplementary Fig. 10). Concentration estimates were derived from tumor hydroxyproline content measurements52(Supplementary Fig. 11a and Supplementary Table 2) and from calculation of injection hold-up with reported values for tumor void volume fraction16,53,54. (Supplementary Table 1) Collagen turnover, a slow process facilitated by tumor macrophages, led to negligible measured drug degradation (Supplementary Fig. 11b–e) and thus is not simulated in the model55. Nonetheless, even large variation in our choices for concentrations and half-life still corroborate the model’s finding: tumor exposure is indeed impacted heavily by molecular weight and matrix targeting of proteins that are injected intratumorally.

Large proteins are predisposed to tumor matrix entrapment

To validate the model’s drug disposition, we tracked 89Zr (78-hour half-life) labeled proteins after injection into B16F10 tumors using PET/CT-imaging. To minimize the contribution of cell-residualized radionuclide to total intratumoral activity, we removed IL-2 from our fusion proteins. (Supplementary Fig. 12a–c) For consistency, we also adjusted the protein concentration after radioisotope labeling to ensure each tumor was injected with exactly 0.1 nmol protein in 20 μL (Supplementary Table 1). Mice were imaged at multiple time points after injection (Fig. 3a, b and Supplementary Fig. 13a–e). In contrast to systemically-administered agents, intratumorally injected proteins were concentrated more at the tumor than any other organ. Partial volume correction (PVC) and isotope-decay correction of the PET images enabled accurate quantification of protein in the tumor after injection (Fig. 3cand Supplementary Fig. 13f). Notably, injected tumors initially held-up only 30% of the injected dose (I.D.) due to the injected volume (20 μL) exceeding the B16F10 hold-up volume (~6.6 μL) (Supplementary Fig. 14a). Interestingly, the volume held-up was independent of tumor size for the range (20–80 mm3) evaluated in our PET study. (Supplementary Fig. 14a). Very large B16F10 tumors (200–300 mm3) that better approximate the size of human tumors, however, hold up 66% of our injected-dose which is still only ~13 μL, suggesting that the tumor interstitium is not as rapidly well-mixed as previously thought (Supplementary Fig. 14a). As a result of injecting volume in excess of the tumor’s capacity, we detected immediate protein accumulation in peripheral organs, albeit at concentrations significantly lower than at the tumor and difficult to resolve by PET imaging. (Supplementary Fig. 14b–f)

a Representative isotope decay- and PVC-corrected whole body PET/CT 3D intensity projections of B16F10 tumor-bearing mice at 0, 6, 24, and 48 h after intratumoral injection of 89Zr-labeled fusion proteins LAIR-LAIR-MSAH464Q, LAIR-LAIRx-MSAH464Q, LAIRx-LAIRx-MSAH464Q, LAIR, and LAIRx. CT image (color bar in grayscale, Hounsfield units) is used for anatomical reference. PET emission (red-yellow color scale from 0–100 %ID) is normalized for all images to the maximum activity intensity at time 0 to allow inter- and intra-group comparisons. b Injected B16F10 tumors in whole body PET/CT 3D projections of panel a are enlarged to visualize injected proteins local distribution. cQuantification of intratumoral activity after injection of 89Zr-labeled LAIR-LAIR-MSAH464Q (mean + s.d., n = 5), LAIR-LAIRx-MSAH464Q (mean + s.d., n = 5), LAIRx-LAIRx-MSAH464Q (mean + s.d., n = 4), LAIR (mean + s.d., n = 4), or LAIRx (mean + s.d., n = 4) in corrected PET images. Fusion protein molecular weight (kD, kilodaltons) and affinity (Kd, equilibrium dissociation constant) to collagen type I, measured in Supplementary Fig. 12, displayed adjacent to legend. d Comparison of tumor exposure of fusion proteins predicted by the model and measured in vivo (mean ± s.d., n = 4–5). Tumor exposure is calculated as the area under the tumor concentration of the injected protein-time curve for a day after injection. Panel c, d are provided in the Source Data file. ID injected-dose, AUC area-under-the-curve.

Full size image

The amount of intratumorally retained protein decreased more rapidly for the small 15 kD proteins LAIR (t1/2 = 48 min) and LAIRx (t1/2 = 40 min) compared to their larger 95 kD equivalent-affinity counterparts LAIR-LAIRx-MSAH464Q (t1/2 = 2.8 h) and LAIRx-LAIRx-MSAH464Q (t1/2 = 2.5 h), respectively (Fig. 3c). LAIR-LAIR-MSAH464Q demonstrated the slowest rate of escape from the tumor (t1/2 = 7.9 h). (Fig. 3c). The model’s predicted rate of tumor escape for these proteins agrees with their observed diffusive escape. (Supplementary Fig. 15a–c). However, the model consistently underestimated their absolute tumor exposure, due to a small fraction of all injected-proteins persisting in the tumor, even at the experimental endpoint. (Fig. 3d and Supplementary Fig. 16) Maximum intensity projections of each individual PET image showed that the persisting proteins were localized to the tumor margin (Supplementary Fig. 17). Quantitative analysis of injected tumors revealed <0.001% of the injected dose residualized inside cells, persuading us that the contribution of non-specific protein catabolism is negligent. (Supplementary Fig. 11b–e). The margins of a B16F10 tumor is matrix-rich, particularly in collagen29; however, even inert LAIRx and LAIRx-LAIRx-MSAH464Q accumulated in this region (Supplementary Fig. 17). The matrix has been shown to considerably slow the diffusion30 and convection56 of proteins, rendering them entrapped. (Fig. 3b). We observed that the entrapped fraction differed as a function of both molecular weight and collagen binding. The large collagen-binding proteins LAIR-LAIR-MSAH464Q and LAIR-LAIRx-MSAH464Q plateaued similarly to 37.3% and 33.0% of the held-up dose, respectively (Fig. 3c). The large inert protein LAIRx-LAIRx-MSAH464Qplateaued to 17.8% of the held-up dose (Fig. 3c). Amongst the small proteins, 11.1% of held-up LAIR endured compared to only 7.5% of LAIRx (Fig. 3c). Our PET image observations, validated by gamma-counter measurements on excised organs (Supplementary Fig. 16), are consistent with a previous report that proteins within a tumor can become entrapped in viscous matrix regions and that large proteins are more susceptible to entrapment than small proteins30. However, this underappreciated dynamic is not recapitulated by the current model.

Contribution of size to exposure outweighs affinity to FN-EIIIB

We sought to evaluate whether the role of size and binding affinity extends to IL-2 fusion proteins targeting less abundant (Supplementary Fig. 18a) yet tumor-specific matrix proteins. We fused IL-2 to the 15 kD nanobody NJB2, previously shown to bind alternatively spliced extra-domain B (EIIIB) domain of fibronectin (FN-EIIIB), a tumor-specific glycoprotein, with nanomolar affinity. (Supplementary Fig. 18b)57,58 To generate an inert size-matched comparator, we fused IL-2 to NJT6, a nanobody that lacks a murine target. To increase the molecular weight while maintaining affinity, MSAH464Q was incorporated between the nanobody and IL-2 (Supplementary Fig. 18b). The 32 kD small format, NJB2-IL2 and NJT6-IL2, and 98 kD large format, NJB2-MSAH464Q-IL2 and NJT6-MSAH464Q-IL2, equivalently activated a T cell line in culture (Supplementary Fig. 18c).

We treated mice bearing established subcutaneous flank B16F10 tumors intraperitoneally with TA99 and intratumorally with our IL-2 nanobody panel (Fig. 4a). Consistent with earlier observations, large-format IL-2 delayed tumor outgrowth more than small-format IL-2. (Figs. 1g and 4b, c) When comparing equivalent affinity constructs, increasing molecular weight improves the overall survival efficacy: NJB2-MSAH464Q-IL2 v. NJB2-IL2 (p = 0.003) and NJT6-MSAH464Q-IL2 v. NJT6-IL2 (p = 0.03). However, the EIIIB-targeting only improved the treatment efficacy for the small format NJB2-IL2 over NJT6-IL2 (p = 0.05) and not the large-format NJB2-MSAH464Q-IL2 vs. NJT6-MSAH464Q-IL2 (p = 0.92). In agreement with earlier observation that the benefit of size can outweigh the benefit of targeting, the survival with large untargeted NJT6-MSAH464Q-IL2 was identical (p = 0.13) to small targeted NJB2-IL2. The pharmacokinetics of locally injected EIIIB-targeting IL-2 fusion proteins, particularly their molecular weight, dictates their overall effectiveness.

One purpose of intratumoral administration is to increase the local bioavailability of a therapy (the other purpose being to minimize toxic systemic exposure). Here we explore how the pharmacokinetic properties of locally injected biologics, specifically molecular size and matrix-binding affinity, can be tuned to maximize tumor retention and thereby boost anti-tumor responses in mice.

Local administration is a seriously considered option for potent immuno-modulators. Several dose-limited protein therapies are currently in trials as intratumoral therapies: OX40 agonist antibody (NCT03831295), CD137 agonist antibody (NCT03792724), bispecific antibody MDX-447 (NCT00005813), anti-PD-1 and anti-CTLA-4 antibodies (NCT03058289), and tumor-targeted IL-2 and TNFα (NCT04362722). The absence of a fundamental pharmacokinetic framework for intratumorally injected agent biodistribution restrains their design optimization. Direct intratumoral injection of an immunotherapy ensures access to immune cells residing within the tumor microenvironment. But since local treatment can act as an in situ vaccine, a robust influx of target immune cells may not occur for several hours to days after treatment. Prolonging tumor residence of injected agents to guarantee productive engagement with desired immune cells is critical to safely maximizing efficacy. Furthermore, to sustain biological effect, the local drug concentration must remain within its therapeutic window. The product of tumor drug concentration and duration of bioavailability equates to local drug exposure. Intratumoral drug exposure has predicted overall survival of humans treated with checkpoint blockade59,60, and of mice treated with IL-2, as demonstrated in the present work.

Using computational and experimental methodologies, evaluating different IL-2 fusion proteins highlighted the decisive role of size and matrix binding on exposure-driven efficacy (Figs. 1 and 4). To our surprise, modest differences in molecular weight of IL-2 fusion proteins significantly impacts their efficacy. Biodistribution studies on intratumorally injected dextrans12 and liposomes13 have found that a critical diameter between 30–50 nm and 100–200 nm, respectively, marks a major inflection in tumor retention. LAIR (~3 nm) and LAIR-LAIR-MSA (~7 nm) are comparatively smaller and poorly-retained yet we find their size-based differences in retention are nevertheless therapeutically meaningful. It is well established that size controls a molecule’s diffusivity within the tumor interstitium and its vascular permeability, thus larger proteins escape the tumor more slowly (Fig. 3b). However, we observed that intratumorally injected proteins are subject to an additional size-based phenomenon: entrapment (Fig. 3c). Alexandrakis et al. first identified entrapment when reporting that an intratumorally injected protein exhibits two interstitial diffusivities and that protein partitioning to “fast” (~10−7 cm s−1) and “slow” (~10−9 cm s−1) diffusing states depends on its molecular weight. They found that 20% of bovine serum albumin (69 kD) injected into a melanoma tumor model partitions into a slow diffusing state, appearing immobilized. Consistent with this measurement, we observed 18% of LAIRx-LAIRx-MSAH464Q(97 kD) and 8% of LAIRx (15 kD) appear entrapped in B16F10 melanoma tumors (Fig. 3c). PET imaging revealed that our injected proteins become entrapped in the tumor periphery, an area rich in collagen matrix and responsible for slowing protein diffusivity29 (Fig. 3b and Supplementary Fig. 17). Size-based entrapment could partially explain why larger IL-2 fusions imparted better tumor growth delay and survival benefit than smaller IL-2 fusions. However, we find that the small and large protein versions persist at intratumoral concentrations of 50 nM and 200 nM, respectively, both exceeding the EC50 for IL-2Rɑβɣ activation (Supplementary Fig. 16). Further studies into how entrapment in viscous tumor regions impacts cytokine bioavailability are warranted.

Tuning tumor exposure of locally injected proteins can also be accomplished by functionalization with a tumor-localization domain. To date, researchers have appended local immuno-modulators with tumor-localization domains that bind acellular matrix proteins (collagen29,31,32,33,34, fibronectin splice variants26, tenascin C61, etc.) and cellular targets27,28. We find that successful localization depends on target abundance. FN containing EIIIB and collagen are both long-lived matrix proteins; however, FN containing EIIIB is far less abundant than collagen in B16F10 tumors (Supplementary Fig. 18a)58,62. As a result, we observe that the survival elicited by NJB2-MSAH464Q-IL2 (Kd,EIIIB 1 nM) treatment is inferior to that of LAIR-LAIR-MSAH464Q-IL2 (Kd,collagen 13 nM). Our model predicts that tumor cell targets of similar abundance to collagen (with an expression of ~105–106 receptors per cell) may also impart localization. Conceivably, achieving productive tumor localization with a cellular target will require balancing binding affinity and target turnover in order to minimize target-mediated drug degradation. The modeling approach described enables follow up analyses on the extent drug exposure is affected by alternative targets (varying in abundance and turnover) and even other pharmacokinetic properties (e.g. charge, stability, and hydrophobicity). With minor modification, the existing model can predict the retention of other immune modulators. In a vast parameter space, employing computational tools in drug design as done here can reveal the principles and major criteria that best drive improvements in efficacy.

Depending on the immune agonist and its mechanism of action, eliciting a desired biological effect may depend on drug exposure in the tumor and/or other lymphoid tissues such as the tumor-draining lymph node (tdLN). For IL-2 therapy of a single lesion, tumor exposure alone sufficiently predicts response. However, for co-stimulatory or co-inhibitory molecules, tdLN exposure may be a better indicator of effectiveness63. Fortunately, for biologics, prolonged tissue residence enhances lymphatic drainage64. Therefore, extending tumor interstitium persistence after local injection may be a simple way to ensure and augment drainage of agents to the functional lymphatic vessels found at the tumor periphery65. Our model may still provide a general ranking for proteins of different sizes whose actions depend on tdLN exposure, and not tumor exposure directly. Alternatively, transport to the tdLN can be empirically defined, using relationships established for subcutaneously injected drugs66, and appended to the model framework.

Importantly, prolonged tumor exposure will not necessarily translate to enhanced efficacy for all immunotherapies67. Some cytokines that coordinate anti-tumor immunity, such as type I68 and type II69 interferons, elicit effects (such as upregulation of PD-L1/L270, T cell activation induced cell death71, etc.) detrimental to anti-tumor immunity at prolonged high tumor concentrations. The exposure-response relationship for other cytokines may also hook at high levels of exposure. The LAIR and LAIRx fusion proteins and computational model described here afford the opportunity to probe for the optimal signal duration of various cytokine therapies.

By simplifying the tumor into a single compartment, our model does not address how spatial heterogeneity in antigen distribution, hydraulic conductivity, and interstitial fluid pressure (IFP) within different tumors impacts the exposure and efficacy of injected drugs. It is widely accepted that uniformly elevated IFP inside tumors inhibits transvascular and interstitial convection rendering diffusion the dominant mode of transport within tumors18,46 - just as we have modeled. However, the pressure drop at the interface between tumor and normal tissue can create an outward seeping flow at the tumor margin. Concurrently, the low hydraulic conductivity of the matrix-rich interface also can impede convective fluid flow56. The interplay of these dynamics results in intratumorally injected proteins stalled in the margin of B16F10 tumors. (Supplementary Fig. 17) For tumor models displaying different matrix composition, solid stress, size, etc. their combined transport dynamics may result in a different localization pattern for intratumorally injected proteins72. Understanding and integrating all spatial dynamics that describe intratumoral micro-pharmacokinetics, although experimentally and computationally intensive, can refine the proposed model. The impact of intratumoral drug distribution is a critical issue that contributes to varied local pharmacodynamic responses to small molecule inhibitors73and antibody-drug conjugates74,75.

Once local drug candidates are identified, the procedures to locally administer them are critical to their successful translation. The optimal injection technique that maximizes intratumoral drug distribution and initial retention is unknown. We find that B16F10 tumors accommodate only ~6.6–13.3 μL of injected volume, implying that our injection only accessed a tumor subregion. Studies in this tumor model typically test volumes from 10–30 µL, and thus a bulk of a locally injected dose is actually systemically injected. Injecting volumes in excess of tumor capacity, a common practice in preclinical mouse models, occurs in the clinical setting as well. Investigational local therapies often follow guidance established by the first intratumoral oncolytic viral therapy talimogene laherparepvec (T-VEC, trademarked Imlygic) which advises injecting between a tenth to as much as 30-fold greater fluid volumes than a tumor can theoretically hold. As a result, systemic dissemination during intratumoral injection is not uncommon and has been observed for agents as small as tracers and as large as viral vectors45,76. Whether some drug leakage from an injected lesion is necessary to engender a robust systemic immunological response has not been explored. Clinically, the consequences of imperfect local retention are diminished efficacy and increased toxicity due to unintended systemic exposure of potent drugs. In an assessment of intratumoral injections, severe (grade ≥ 3) immune-related systemic toxic effects occurred in 4% of T-VEC injections and 2% of investigational local therapies due to drug leakage77. Standardized instructions for intratumoral injection, aimed at reducing immediate systemic exposure while maximizing tumor coverage, are much needed. Currently, clinical protocols suggest scaling intratumoral injection volume by lesion size;1,78but as we observed in mice, larger lesions may accommodate the same initial hold-up volume from a single injection as small lesions – a poorly understood phenomenon. Defining the optimal injection number, volume, rate, and pressure as a function of tumor and drug properties is needed to improve delivery. While generating guidelines based on observations in syngeneic mouse models – as we have done here – is helpful, their validation in tumors of clinically-relevant size and tissue heterogeneity is still required.

Clinical outcomes of local immunotherapies depend on both their administration and local pharmacokinetics. To the latter end, we have outlined here a foundational modeling framework establishing the pharmacokinetic underpinnings of local IL-2 therapy efficacy in mice. Our analysis presents a step towards the critical evaluation of locally injected biologics, which should help the growing field of intratumoral pharmacology rationally design the properties and dosing regimens of future local immunotherapies.