Increased Tumor Penetration of Single-Domain Antibody–Drug Conjugates Improves In Vivo Efficacy in Prostate Cancer Models

Antibody–drug conjugates (ADC) have opened a new field of targeted therapeutics based on “hybrid” drugs combining desirable targeting properties of biologics with the potency of small-molecule cytotoxic payloads. For ADCs, the protein carrier is typically a mAb that specifically binds to a target antigen expressed on cancer cells, increasing the delivery of the small-molecule payload to the site of action in vivo. To date, seven ADCs have been FDA-approved (1, 2), most recently enfortumab vedotin and trastuzumab deruxtecan, with a large pipeline of >70 in clinical trials. However, one drawback of antibodies is slow tumor penetration. The tumor uptake of antibodies is limited by their extravasation rate, and they tend to penetrate only a few cell layers outside of blood vessels due to their rapid antigen-binding rate relative to intratumoral diffusion (3, 4). In the clinic, unconjugated antibodies are often well tolerated, such that they can be delivered at very high doses that saturate receptors on cell layers closer to the blood vessel, enabling the antibody to diffuse farther through the tumor. However, the payload toxicity of ADCs limits the dose and frequency of administration, restricting tumor penetration depths and allowing regrowth between doses (typically given every 3 weeks for current therapies; ref. 5). Therefore, devising design and treatment strategies that increase tumor penetration may yield greater efficacy and improve clinical success rates for ADCs and other protein–drug conjugates.

Many current ADCs in late-stage clinical trials are targeting receptors with high expression (e.g., TROP-2, HER2, folate receptor, and Nectin-4) where tissue penetration can play a significant role in efficacy. Previous work based on coadministration of trastuzumab with the ADC T-DM1 (ado-trastuzumab emtansine) demonstrated that higher tissue penetration could yield better efficacy in a mouse model of highly expressed HER2-positive cancer (6). On the basis of a literature review of studies involving ADCs with a range of drug-to-antibody ratios (DAR; ref. 7), ADCs that delivered the same payload dose at a lower DAR (i.e., higher protein doses) generally yielded an increase in efficacy in animal models with moderate to high expression, likely as a result of better tissue penetration (8). Although bystander effects can help mitigate transport challenges by allowing the payload to diffuse deeper into the tissue following release from the protein, analysis of literature data and predictive simulations indicated that higher tissue penetration is more efficient at improving efficacy even when using bystander payloads (8). When the ADC is more uniformly distributed, the same total amount of cytotoxic payload delivered to the tumor is spread more homogeneously, so cells adjacent to vessels that receive an overdose of payload with heterogeneous ADC distribution now receive fewer payloads. Therefore, implicit in this approach is that the amount of payload delivered per cell exceeds the intrinsic potency of the payload, thereby maintaining a lethal cellular dose while increasing penetration to reach more cells. Similarly, lower target receptor expression can improve efficacy for extremely potent ADCs by enabling deeper tissue penetration. Lower receptor expression limits the amount of bound and internalized ADC for perivascular cells but increases tumor tissue penetration because penetration depth is inversely proportional to expression (7, 9). On the basis of these and other results (10), tumor tissue penetration is critical for the overall efficacy of ADC treatment and should be analyzed when optimizing protein–drug conjugates for maximum effectiveness.

Besides increasing the dose or coadministering the ADC with unconjugated antibody, there are protein engineering approaches to develop scaffolds for improved tissue penetration (11–14). This may be particularly advantageous for highly expressed targets that would require very large antibody doses to saturate the tumor or for targeting metastases with varying expression levels. Reducing affinity is one method of increasing tissue penetration (15). However, for large, bivalent IgGs, the intrinsic affinity of each Fab arm must be very low to penetrate tissue with highly expressed antigens (e.g., 270 nmol/L K_d; ref. 16). Instead of modulating affinity, lower molecular weight proteins can penetrate tissue farther before being immobilized through target binding (17) because of their faster diffusion coefficient (which is the dominant mode of transport in tumors due to elevated interstitial pressure; ref. 18). Smaller scaffolds are already being tested in the clinic and some, such as caplacizumab—a bivalent single-domain antibody (∼30 kDa) used to block platelet aggregation (19), have already received FDA approval, providing precedent for these formats (https://www.fda.gov/news-events/press-announcements/fda-approves-first-therapy-treatment-adult-patients-rare-blood-clotting-disorder). A major limitation for these smaller scaffolds is their rapid renal clearance (11, 20, 21). However, fusion to albumin-binding domains can extend plasma half-life and avoid glomerular filtration by increasing the effective molecular weight while still enabling better tissue penetration (22–26).

In this study, we investigated the tumor tissue pharmacokinetics and efficacy of three different prostate-specific membrane antigen (PSMA)-binding single-domain antibody (Humabody)–drug conjugates and an IgG–drug conjugate (referred to collectively as ADCs or simply antibodies/antibody constructs if no drug is present). Humabodies are fully human, single heavy chain variable (V_H) domains generated using a proprietary transgenic mouse platform (Crescendo Biologics; refs. 27, 28). The mice lack murine heavy chain, κ light chain, and λ light chain expression but contain human heavy chain genes. After in vivo maturation and development, multiple V_H domain Humabodies can be constructed with a range of functionality to optimize target engagement and improve therapeutic benefit. We used this flexible platform to test the impact of two main design parameters with ADCs: the cellular internalization rate and the plasma clearance rate. The first construct (V_H2-V_H1-HLE, MW = 44.7 kDa) is a fusion of three V_H domains: two PSMA-binding domains to different epitopes (V_H1 and V_H2) conjugated to an albumin-binding “half-life extender” (HLE) V_H domain to yield V_H2-V_H1-HLE. Targeting multiple epitopes results in antigen cross-linking and rapid internalization, while binding albumin can slow plasma clearance. Therefore, this construct was designed for rapid internalization and slow clearance. The second antibody construct, designed to isolate the impact of internalization (V_H1-HLE, MW = 31.1 kDa), consists of a monovalent PSMA-binding domain (V_H1) connected via a linker to the albumin-binding domain, resulting in the V_H1-HLE construct. The lack of ability to cross-link receptors slows internalization. The third construct (V_H2-V_H1, MW = 29.7 kDa) is identical to V_H2-V_H1-HLE except without the HLE domain to compare the impact of different clearance rates. All Humabody constructs were site-specifically conjugated at the C-terminus with either a potent DNA-alkylating agent (DGN549, an indolinobenzodiazepine DNA-alkylating monoimine; ref. 29) for potency/efficacy studies or a fluorescent dye for in vitro and in vivo kinetics and distribution experiments. The fluorescent antibody constructs (also referred to as fluorescent antibodies in this study) were used to determine the overall tumor uptake (%ID/g), tumor tissue distribution, in vitro spheroid penetration, and in vivo single-cell pharmacokinetics for comparison with the in vitro and in vivo efficacy. Combined with a computational model, these results demonstrate that smaller monovalent protein scaffolds can increase tissue penetration enhancing in vivo efficacy in mouse models of prostate cancer.

The Krogh cylinder model has been validated for tissue distribution of antibodies as well as smaller molecules by our group and others (3, 9, 30–32). A 1-D cylinder model consisting of only radial gradients was used to represent the distribution of antibody constructs because they are permeability-limited. The model tracked the concentration of free target as well as free, bound, and internalized antibody. Equations and parameters used for simulations are located in the Supplementary Table S1. The equations describe the rate of antibody extravasation from a blood vessel, the diffusion and subsequent binding to a PSMA receptor in the tissue, at which point the therapeutic is internalized and degraded. Modifications were made to parameters based on the size of the antibody constructs tested. All parameters used in the Krogh cylinder were gathered from the literature or measured independently (i.e., not fit to tissue distribution data).

Both fluorophore and drug (Supplementary Methods) conjugation reactions occurred at the C-terminus of each single-domain antibody. Alexa Fluor 680 (AF680), Alexa Fluor 647 (AF647), or Alexa Fluor 488 (AF488; Thermo Fisher Scientific) were conjugated to each antibody through sortase labeling of the LPTGX motif present at the C-terminus. Briefly, NHS ester dyes were reacted with propargyl amine in aqueous solution buffered with 7.5% sodium bicarbonate for 30 minutes at room temperature, followed by purification on reverse-phase HPLC. The alkyne dye was then reacted with a GGGX peptide (synthesized on a CEM Liberty Blue peptide synthesizer), where X is the nonnatural amino acid azidohomoalanine, via copper-catalyzed click chemistry similar to previously published protocols (33, 34) to generate GGGX-dye. The peptide-dye product was purified on reverse-phase HPLC and reacted with the antibody constructs (5:1 ratio) for 10 minutes with 2 μmol/L sortase and 5 mmol/L calcium chloride in aqueous HEPES buffer. The reaction mix was filtered with a Costar Spin-x Centrifuge Tube Filter (Thermo Fisher Scientific) containing 300 μL of HisPur Ni-NTA Resin (Thermo Fisher Scientific) to remove unreacted protein (containing a His-tag) and the His-tagged sortase and then purified via size-exclusion chromatography/FPLC. The fluorescent constructs were concentrated with AMICON 10 kDa molecular weight spin filters to 10–40 μmol/L with a degree-of-labeling (DoL) approximately 0.7 dyes/protein as confirmed with absorption readings from a NanoDrop 1000 Spectrophotometer (Thermo Fisher Scientific; Supplementary Fig. S1). After purification, Native-PAGE gels were scanned on an Odyssey CLx to ensure all free dye was removed (Supplementary Fig. S2). Drug conjugates were purified and characterized via Q-ToF Mass spectrometry (Supplementary Fig. S3).

The DU145 cell line (RRID: CVCL_0105, PSMA-negative, certified by ATCC via short-tandem repeat analysis) and PSMA transfected DU145 cell line generated by Kampmeier and colleagues (DU145-PSMA; ref. 35) were received from Crescendo Biologics. The transfected cell line was confirmed for PSMA expression but the transfected cell line and CWR22Rv1 lines were not reauthenticated. Cells were cultured two to three times per week up to passage number of 50 and grown in RPMI1640 supplemented with 10% (v/v) FBS, 50 U/mL penicillin, and 50 μg/mL streptomycin at 37°C with 5% CO₂. Mycoplasma testing was performed annually using the Mycoalert Testing Kit (Thermo Fisher Scientific, NC9719283). All animal studies were approved and conducted in compliance with the Institutional Animal Care and Use Committee of the University of Michigan (Ann Arbor, MI) and Association for Assessment and Accreditation of Laboratory Animal Care International. Pharmacokinetic and in vivo tumor distribution animal studies were conducted in 4–8-week-old homozygous female nude (RRID: 2175030, Foxn1nu/nu, Jackson Laboratories) mice. For in vivo tumor distribution and growth studies, the nude mice were inoculated in both flanks with 5 × 10⁶ cells in Matrigel (Thermo Fisher Scientific, CB40234A), one flank with DU145-PSMA and the other with DU145 cells. For tumor growth studies, tumor volume was measured with calipers 3× per week using the formula: volume = 0.5*length*width². Corrected sample size for each treatment group was calculated as 10 based on an a priori power test (details in Supplementary Materials).

Plasma clearance was measured after tail-vein injection of 0.7 nmol of fluorescent antibody (or 0.067 nmol of J591 anti-PSMA IgG). Plasma samples were obtained through retro-orbital sampling 10 μL of whole blood, then mixed with 15 μL of PBS-EDTA (10 mmol/L), and centrifuged at 3,000 × g for 1 minute. A total of 18 μL of the resulting plasma was frozen at −80°C until further analysis. The antibody concentration was determined by scanning 15 μL of plasma in a 384-well black-walled Plate (Corning) on the NIR Odyssey CLx Scanner (LI-COR) and comparing the signal intensity to a calibration curve of known antibody concentration to signal intensity at the same DoL and scan settings. The plasma concentration at each time point was normalized to the initial concentration, and then the clearance was fit to a biexponential decay using PRISM (GraphPad). Absolute plasma concentrations at 1 minute were compared with theoretical initial concentrations calculated on the basis of the dose and estimated plasma volume of the mouse.

The biodistribution of fluorescent antibody was conducted as described previously (7, 36, 37) with a single bolus dose delivered for each antibody. Briefly, 24 hours after tail-vein injection of 0.7 nmol of fluorescent antibody (or 0.067 nmol of J591 anti-PSMA IgG), animals were euthanized, and organs were resected. Organs were then homogenized by mechanical disruption, incubated with 1:1 RIPA buffer (Thermo Fisher Scientific)/PBS solution supplemented with 5 mg/mL Collagenase IV (Thermo Fisher Scientific) for 1.5 hours, disrupted using a FB-120 Sonic Dismembrator, and incubated in 1:1 RIPA buffer/0.025% trypsin-EDTA solution for 1.5 hours. After homogenization, organs were serially diluted and scanned on the Odyssey CLx scanner to ensure fluorescence detection was in the linear range. The signal intensity was compared with a calibration curve and normalized to organ weight and homogenate volume to calculate the percent injected–dose per gram (%ID/g).

As described previously (7, 32, 36), the tumor distribution of fluorescent antibody was analyzed using fluorescence microscopy 24 hours postinjection. Briefly, 0.7 nmol fluorescent antibody (or 0.067 nmol of J591 anti-PSMA IgG) was administered via tail-vein injection once the tumor volume was approximately 250 mm³. The animal was imaged 24 hours postinjection. Hoechst 33342 (Thermo Fisher Scientific) was administered 15 minutes before euthanasia via the tail vein at 15 mg/kg to label functional vasculature in the tumor. Tumors were then resected, flash-frozen in optimum cutting temperature (OCT) compound using isopentane chilled on dry ice, and sectioned for histology on a cryostat (10-μm slices). Before imaging, tumor slices were stained for 30 minutes with anti-mouse CD31 (BioLegend, 102402) conjugated to Alexa Fluor 555. Microscopy was performed using an upright Olympus FV1200 confocal microscope equipped with a 20× objective and 405 nm, 543 nm, and 635 nm lasers. High-resolution tumor images were obtained by stitching smaller images with the Olympus software. Images were exported and analyzed using ImageJ image analysis software as described previously (7, 32, 36).

Resected tumors were digested into a single-cell suspension using a Tumor Dissociation Kit (Miltenyi Biotec, 130–096–730) and passed through a 40-μm filter to remove clumped cells. The suspension was washed 2 times with PBS and then analyzed using an Attune Acoustic Focusing Cytometer (Life Technologies). Quantitative fluorescent beads (Quantum Simply Cellular anti-Mouse IgG) were used to convert fluorescent intensity units from flow experiments to number of proteins and later to number of payloads delivered on the basis of DoL or DAR.

Antibody constructs with a C-terminal cysteine residue were conjugated to a potent DNA-alkylating agent (DGN549) via maleimide chemistry similar to previous reports (29), in a reaction buffer containing 25% N,N-dimethylacetamide and 75% 50 mmol/L potassium phosphate-5 mmol/L EDTA, pH 6.0 for 3 hours at 25°C and then purified by Illustra NAP-25 columns (GE Healthcare) to achieve an average DAR of approximately 1.0. The free DGN549 payload has an IC₅₀ = 20 pmol/L ± 15 pmol/L in DU145 cells. The potency of the ADC constructs was then assessed by seeding DU145-PSMA cells at a density of 8,000 cells per well in 96-well plates for cell viability assays. Titrations of DGN549-conjugated antibodies were replaced daily for 6 days. At the endpoint, cells were washed twice with media and then incubated with PrestoBlue Cell Viability Reagent (Thermo Fisher Scientific, A13261) for 25 minutes at 37°C in a 1:10 dilution in media. The fluorescence (560/590, Ex/Em) of each well was then measured using a Biotek Synergy plate reader to measure final cell viability. Background signal from wells without cells was subtracted from all samples, and then, viability was normalized to untreated cells.

For the high expression DU145-PSMA xenografts, female homozygous Foxn1 nude mice (Jackson Laboratories, 002019) were injected in the left flank with 5 × 10⁶ DU145-PSMA cells. Mice were assigned into six treatment groups: PBS vehicle control (n = 9), nonbinding control V_H-HLE-DGN549 (n = 6), V_H1-HLE-DGN549 (n = 9), V_H2-V_H1-HLE-DGN549 (n = 10), V_H2-V_H1-DGN549 (n = 9), and J591-DGN549 (n = 9). Treatments were administered in a single dose as tumors reached an approximate volume of 250 mm³. Doses were matched to the amount of payload delivered for the maximum tolerated dose (MTD; 90 μg/kg for Humabody–drug conjugates, 30 μg/kg for J591). However, these doses all caused complete responses in CWRr22Rv1 tumors (Supplementary Fig. S4), so all the doses were decreased 3-fold to 30 μg/kg for Humabody–drug conjugates, 10 μg/kg for J591. The J591 ADC was given at a 3-fold lower payload dose to match the toxicity between the agents as estimated from weight loss (Supplementary Fig. S5). In general, the Humabody–drug conjugates were better tolerated, enabling a higher payload dose. Tumor sizes were monitored three times per week until the study endpoint.

Similarly, for the moderate expression CWR22Rv1 xenografts, male CB.17 SCID mice were implanted subcutaneously on one flank with 1 × 10⁷ CWR22Rv1 human prostate carcinoma cells. After 20 days, animals with individual tumor xenograft volumes of 75 to 126 mm³ were sorted into groups (n = 10) with a group mean tumor volume of 106 mm³ and dosing was initiated. HLE Humabody–drug conjugates were administered in a single dose on day 1 of the study. Non-HLE Humabody–drug conjugate was administered in three, 10 μg/kg doses, administered on day 1, 3, and 5 of the study (30 μg/kg cumulative dose). Tumor size in mm³ was monitored individually and a group median tumor volume calculated. Animals reaching the endpoint volume of 1,000 mm³ or the end of the study were euthanized.

Tumor spheroids were cultured using custom 384-well plates described previously (38). Briefly, 3,000 cells suspended in 25 μL of seeding media comprising 19.5 μL complete RPMI culture media, 5 μL of 1.2% (w/v) methocellulose (Dow Corning), and 0.5 μL Matrigel were added to alternate wells of the 384-well hanging drop plate. The edges of the 384-well plate were lined with sterile gauze soaked in sterile distilled water (Gibco) containing 0.1% penicillin/streptomycin. The 384-well plate was sandwiched in a 96-well clear well plate containing sterile water to minimize media evaporation from the hanging drops. Media changes were made every 2 days by removing 9 μL media from the drop and adding in 10 μL of fresh media (to adjust for evaporation), repeated twice for each drop. Spheroids were cultured for 7 days until they attained a diameter of 400–500 μm.

To study distribution, the hanging drops were incubated in the 384-well plate with V_H1-HLE-AF680, V_H2-V_H1-HLE-AF680, V_H2-V_H1-AF680, or J591-AF680 at a final concentration of 30 nmol/L, by replacing 10 μL of the 25 μL of the spheroid media with 10 μL of 75 nmol/L (2.5× concentration) fluorescent antibody. The drug concentration in the media was assumed to be constant over the course of the incubation (i.e., no depletion effects). After 24 hours of incubation, the spheroids were individually extracted from the wells and washed two times with PBS before being fixed with 4% formaldehyde, frozen in OCT, and stored at −80°C until further processing. Frozen OCT blocks were sectioned for histology on a cryostat (16-μm slices), and sections were stained ex vivo with Hoechst 33342 (Thermo Fisher Scientific) for 5 minutes and J591-AF555 for 30 minutes before imaging on the Olympus FV1200 confocal microscope. Multichannel imaging of spheroid sections was performed with a 20× objective with 405 nm (Hoechst-33342), 543 nm (AF555), and 635 nm (AF680) lasers. Image analysis was performed to generate a Euclidean distance map (Supplementary Fig. S6).

Plot values are shown as mean ± SD. Data were analyzed for statistical significance (P < 0.05) with an unpaired, two-tailed t test with Welch correction using GraphPad Prism 8 for macOS.

The in vitro toxicity of DU145-PSMA cells was measured after 6 days using a PrestoBlue viability assay (Supplementary Fig. S7). V_H1-HLE-DGN549 had the lowest potency, followed by both V_H2-V_H1-HLE-DGN549 and V_H2-V_H1-DGN549. The J591-DGN549 ADC had the highest potency based on protein concentration, likely due to the multiple payloads per antibody. Potency based on payload concentration was similar for V_H2-V_H1-DGN549, V_H2-V_H1-HLE-DGN549, and J591-DGN549 (Supplementary Fig. S7B). These results are consistent with the internalization rates for these ADC constructs, with the biparatopic Humabodies (V_H2-V_H1-HLE-AF680 and V_H2-V_H1-AF680) capable of inducing surface-crosslinking and driving internalization (Supplementary Fig. S8; ref. 39), the J591 ADC inducing internalization (40), and V_H1-HLE having the slowest cellular uptake.

Tumor growth studies done in athymic nude or SCID mice with PSMA-expressing xenografts showed significant growth inhibition for all PSMA-targeted ADCs relative to untargeted controls. In the xenograft model with higher PSMA expression (DU145-PSMA), V_H1-HLE-DGN549 shows the greatest efficacy among all experimental groups (P < 0.05, Supplementary Fig. S9). V_H2-V_H1-HLE-DGN549 showed better efficacy than V_H2-V_H1-DGN549 given in a single bolus dose as a result of V_H2-V_H1-HLE's prolonged circulation and tumor exposure. The albumin-binding V_H domain (HLE) presumably reduced renal filtration to slow clearance (K_D = 51.9 nmol/L for mouse albumin with k_off = 3.41 × 10⁻²/s and k_on = 6.57 × 10⁵/M*s as measured by surface plasmon resonance). Individual DU145-PSMA growth curves for all mice in each treatment group are displayed in Supplementary Fig. S10. Overall, tumor growth inhibition was generally lower in the DU145-PSMA tumors compared with CWR22Rv1 xenografts, likely due to a combination of the larger starting tumor size (250 vs. 100 mm³), which can reduce vascular density (41) and higher PSMA expression, resulting in greater heterogeneity in intratumoral distribution. In CWR22Rv1 tumors, V_H1-HLE-DGN549 and J591-DGN549 were equally effective. Fractionated dosing of the rapidly cleared V_H2-V_H1-DGN549 in the CWR22Rv1 model to mitigate rapid renal clearance resulted in similar efficacy as the more slowly cleared V_H2-V_H1-HLE-DGN549, further highlighting the benefit of prolonged exposure. In contrast to the observed trend for in vitro potency of the Humabody–drug conjugates, V_H1-HLE-DGN549 had surprisingly higher efficacy than either V_H2-V_H1-HLE-DGN549 or V_H2-V_H1-DGN549 in both tumor models (Fig. 1A and B). To better understand this effect, a detailed study of the distribution at the systemic, organ, tissue, and cellular level was initiated using near-infrared fluorescence to delineate the contribution of the multiple ADC delivery and processing steps in vivo toward efficacy.

To study distribution, fluorescent antibodies were injected into nude mice bearing antigen-positive DU145-PSMA cells in the left flank and DU145 (PSMA-negative) cells in the right flank. (The clearance of the fluorescent antibodies was similar to nonfluorescent antibody (Supplementary Fig. S11) and ADCs fluorescently labeled via lysine side-chains (Supplementary Fig. S12). The slowly cleared V_H1-HLE-AF680, V_H2-V_H1-HLE-AF680, and J591-AF680 antibody had similarly high tumor uptake in the antigen-positive tumor as expected (∼30%ID/g) with no statistically significant differences, while the more rapidly cleared V_H2-V_H1-AF680, had much lower uptake (∼7% ID/g), Fig. 2A. Rapid plasma clearance of V_H2-V_H1-AF680 also resulted in lower uptake in the antigen-negative tumor and all other organs compared with V_H1-HLE-AF680, V_H2-V_H1-HLE-AF680, and J591-AF680 (Fig. 2B). Because the tumor uptake, plasma clearance, and in vitro toxicity did not sufficiently explain the trends in tumor growth inhibition, further studies on the intratumoral distribution of these agents were performed.

Previous work in the literature indicates that smaller protein scaffolds result in higher tissue penetration (e.g., refs. 15, 17, 19, 24). We adapted the Krogh cylinder model for a spherical geometry and, after gathering parameters on binding kinetics, plasma clearance, and receptor expression, conducted simulations of tissue and spheroid distribution after 24 hours as depicted in Fig. 3A. The computational model predicted the smaller V_H1-HLE would exhibit the greatest tumor penetration, and that the rapidly internalizing V_H2-V_H1 and V_H2-V_H1-HLE would only target a few cell layers into the spheroid, similar to the J591 antibody. Notably, this model assumes that the specific, reversible albumin-binding of HLE does not reduce the diffusion into tissue in contrast to nonspecific albumin-sticking from lipophilic agents (42).

These predictions were validated in vitro using spheroids made up of DU145-PSMA cells. Histology images of the spheroids are shown in Fig. 3B. Ex vivo staining of the tissue slides indicated uniform PSMA expression throughout the spheroids (red), and incubation with antigen-negative DU145 spheroids showed rapid and uniform distribution (which was only detectable at high concentrations due to a lack of binding, Supplementary Fig. S13). The absolute penetration distances were dependent on the cell packing, with spheroids seeded at lower cell density (2,000 vs. 3,000) and grown for longer periods of time (14 days vs. 7 days) resulting in lower penetration (Supplementary Fig. S14), likely due to higher cell density/lower void fraction, but the relative pattern of penetration remained consistent. These results were maintained across approximately 10 independent experiments (each done on separate days) containing approximately 25 spheroids per fluorescent antibody (Supplementary Fig. S15). Quantitative image analysis confirmed the higher penetration of V_H1-HLE-AF680 relative to the other fluorescent antibodies (Fig. 3C). The experimental data complements the computational predictions as V_H1-HLE-AF680 was demonstrated to distribute farther than any of the other antibodies tested.

The spheroid experiments provided a high-throughput, well-controlled, and robust environment to quantify tissue penetration. To account for additional considerations in vivo, such as plasma clearance and extravasation, tumoral distribution was further analyzed in DU145-PSMA xenografts.

Tumor histology images and whole-animal imaging (Supplementary Fig. S16) captured the overall fluorescent antibody distribution following intravenous delivery at the same doses as the efficacy study. Similar trends were seen for in vivo tumor distribution as those predicted by the computational model and demonstrated in the spheroid experiments. V_H2-V_H1-AF680 and V_H2-V_H1-HLE-AF680 exhibited limited penetration similar to that of the perivascular distributed J591-AF680 antibody, while the V_H1-HLE-AF680 had greater penetration targeting more cells (Fig. 4). The J591 antibody did not show any increase in penetration at later times (Supplementary Fig. S17 at 72 hours), consistent with a pseudo-steady state between vascular delivery and cellular degradation preventing deeper penetration over time. ADCs with fluorescently labeled lysines showed a similar pattern as the antibodies lacking a payload, indicating the payload was not impacting the tumor distribution (Supplementary Fig. S18).

A Euclidean distance map of the histology images was compiled to provide a semiquantitative look at the differences in tumor penetration between fluorescent antibodies (Supplementary Fig. S19) and was consistent with the in vivo computational predictions using the Krogh cylinder model (Supplementary Fig. S20). However, the images do not provide absolute quantification of payload delivery with single-cell resolution that would allow for precise correlations between efficacy and penetration.

Flow cytometry can provide absolute measurements of antibody uptake at the single-cell level (e.g., Supplementary Fig. S21, in vitro). The single-cell measurements enabled quantification of the fraction of cells targeted in the tumor and the number of payloads per cell for the targeted fraction (representative histograms Supplementary Fig. S22). Gating on the in vivo histograms, V_H1-HLE-AF680 targeted more cells than the other antibodies (P < 0.05; Fig. 5A), consistent with the computational model and histology experiments. This finding is also consistent with in vivo efficacy in the same xenograft, where V_H1-HLE-DGN549 showed the most tumor growth inhibition of the ADC constructs tested. While V_H1-HLE-AF680 targeted the largest fraction of cells, the number of payloads per cell in these targeted cells was lower than V_H2-V_H1-HLE-AF680 (but not statistically significant due to high variability, Fig. 5B), which explains the similar tumor-averaged total uptake (%ID/g). Using an AF647 tag (and its corresponding calibration curve) for higher sensitivity due to lower total uptake, we found that V_H2-V_H1-AF647 reached a slightly lower fraction of targeted cells, and the number of payloads per cell was approximately 1/3 compared with V_H2-V_H1-HLE-AF680), resulting in the much lower overall %ID/g in Fig. 2. Note that the J591-AF680 antibody was administered at a lower payload dose, resulting in fewer payloads per cell even though it has a similar uptake efficiency (∼30 %ID/g).

Following the first wave of ADC approvals, various protein–drug conjugates are being developed on the basis of affinity, toxicity, and stability to achieve an optimal therapeutic effect. Although the heterogeneous, perivascular distribution of these agents is well documented (9, 43–46), the impact on efficacy is not routinely isolated and investigated. Here, we demonstrated the improved penetration of a single variable heavy chain unit conjugated to an albumin-binding domain (V_H1-HLE-AF680) over other ADC constructs in a three-dimensional spheroid cell culture system and in vivo. As a result, V_H1-HLE-DGN549 exhibited higher in vivo efficacy than the other ADC constructs despite having lower potency in vitro.

In vitro cellular toxicity assays indicated that the rapidly internalized biparatopic antibody conjugates (V_H2-V_H1-HLE, V_H2-V_H1) and J591 had the highest potency, followed by V_H1-HLE (Supplementary Fig. S7). The greater in vitro potency of the biparatopic ADC constructs is likely due to receptor surface clustering that causes rapid internalization of the drug (39, 47, 48). In monolayer cell culture, there are no mass transport/drug delivery limitations, so cellular internalization/payload release is the rate-limiting step in payload uptake and the key determinant for cell killing. However, in vivo, the extravasation of ADC from the blood is the rate-limiting step and determines total tumor uptake (at subsaturating doses, which is typical for highly expressed antigens). Therefore, the faster internalization rate (relative to binding kinetics and diffusion) did not deliver more total payload to the tumor, as seen by the similar tumor biodistribution for V_H1-HLE-AF680 and V_H2-V_H1-HLE-AF680 (Fig. 2). The rapid internalization instead limited the penetration distance of V_H2-V_H1-AF680 and V_H2-V_H1-HLE-AF680 in tumor spheroids (Fig. 3) and in vivo (Fig. 4), which therefore reached fewer cells (Fig. 5). Although the amount of V_H1-HLE-AF680 and V_H2-V_H1-HLE-AF680 delivered to the tumor was the same, the slower internalizing V_H1-HLE-AF680 was able to penetrate farther in the tumor to target and kill more cells than V_H2-V_H1-HLE-AF680 as illustrated by the diagram in Fig. 6.

One of the main disadvantages of using single-domain antibodies or other small protein scaffolds in place of antibodies is rapid renal clearance (23). V_H2-V_H1-AF680 demonstrated poor tumor uptake caused by this rapid kidney filtration. Conjugating an albumin-binding domain to the monovalent and biparatopic constructs slowed clearance nearly 100-fold, leading to increased tumor uptake, comparable with a mAb (Fig. 2).

Interestingly, some of the in vivo efficacy trends differed between the moderate and high expressing xenografts (Fig. 1). In both models, the V_H1-HLE-DGN549 demonstrated the greatest efficacy, but the relative efficacy of the other agents varied between the two tumor models. In the CWR22Rv1 tumor model with lower expression, V_H1-HLE-DGN549 and the J591-DGN549 ADC showed almost identical tumor growth inhibition, as did V_H2-V_H1-HLE-DGN549 and V_H2-V_H1-DGN549. But in the DU145-PSMA tumor model, each protein–drug conjugate exhibited distinct efficacy. The most likely explanation for differences in efficacy between the DU145-PSMA and CWR22Rv1 tumor models is the level of PSMA expression: DU145 has relatively high expression (∼10⁶ receptors/cell), while CWR22Rv1 has moderate expression (< 10⁵ receptors/cell). With more available targets for binding in a higher expression cell line, more ADC is required to saturate the first few cell layers, thus hindering diffusion to regions distal from vasculature. Cell death often correlates with intracellular concentration (49), and V_H1-HLE-DGN549 maintained efficacy (even exceeding V_H2-V_H1-HLE-DGN549 in vivo, Fig. 1) with a lower (but still lethal) number of payloads per cell than V_H2-V_H1-HLE-DGN549 based on in vitro measurements (Fig. 5; Supplementary Fig. S21). This is consistent with V_H2-V_H1-HLE-DGN549 delivering more of the potent DGN549 payloads than needed to kill targeted cells (i.e., overkill) in this model. Another contributing factor for the differences in degree of response between tumor models could be the starting tumor size. Treatments were generally less efficacious in the DU145-PSMA tumor model, where doses were administered at an average tumor size of 250 mm³, versus 100 mm³ in the CWR22Rv1 tumor model.

The slower clearance due to the presence of a HLE domain (i.e., higher plasma AUC) was expected to increase efficacy in the mouse model of prostate cancer based on greater drug exposure. Previous studies have demonstrated that exposure (i.e., plasma AUC) correlates with tumor uptake and response to ADCs (50, 51). This is supported by the improved in vivo efficacy of V_H2-V_H1-HLE-DGN549 compared with V_H2-V_H1-DGN549 (Fig. 1) in the DU145-PSMA tumor model (single dose each). The plasma clearance and biodistribution data confirmed that V_H2-V_H1-HLE-AF680 is cleared more slowly and has significantly higher tumor uptake than V_H2-V_H1-AF680 (Fig. 2). According to single-cell measurements, V_H2-V_H1-HLE-AF680 targeted a higher number of cells than V_H2-V_H1-AF647 with more payloads per cell in each targeted cell (Fig. 5). Alternatively, V_H2-V_H1-HLE-DGN549 and V_H2-V_H1-DGN549 showed almost identical efficacy in the CWR22Rv1 tumor model with lower PSMA expression where V_H2-V_H1-DGN549 was administered in three fractionated doses of 10 μg/kg. Fractionated dosing of the faster cleared V_H2-V_H1-DGN549 increased tumor exposure time compared with a single bolus dose. The similar efficacy of V_H2-V_H1-HLE-DGN549 and V_H2-V_H1-DGN549 in this system further highlights the significance of extended tumor exposure to in vivo efficacy.

Taken together, the differences seen with the in vivo efficacy studies cannot be explained solely by total tumor uptake or in vitro potency, given that V_H1-HLE outperformed V_H2-V_H1-HLE and J591 in vivo with similar tumor uptake (Fig. 2) and lower in vitro potency (Fig. 1). The best correlate for efficacy was the delivery of a lethal cellular dose of payload to the maximum fraction of cells within the tumor. The in vitro spheroid and in vivo histology images agreed with a predictive computational model based on the fundamental binding kinetics, tumor physiology, target-specific expression/internalization, and biophysical properties (e.g., diffusion) of the ADCs.

These results can be used to help design more effective ADCs for efficient payload delivery and cell killing. Importantly, while this work demonstrates that the V_H1-HLE-DGN549 is the most effective in this system, it does not indicate that monovalent binding agents are always the most efficacious agents in vivo. Rather, the work highlights the need to customize the construct based on the target properties (expression, internalization), payload (with or without bystander effects, MTD), and properly scale these results to the clinic (i.e., clearance rates). Likewise, the PSMA expression in these animal models was moderate to high, and the payload used is extremely potent. Under these conditions, the constitutive internalization rate of PSMA was sufficient for tumor cell killing in these animal models without the need to drive more rapid internalization through biparatopic clustering. Many current ADCs demonstrating therapeutic response in late-stage clinical trials have high expression targets (sacituzumab govitecan, trastuzumab deruxtecan, mirvetuximab soravtansine, and enfortumab vedotin; ref. 52) and may benefit from methods that improve penetration to reach more cells at a lower, yet lethal payload dose. However, driving faster internalization could be very useful under circumstances with lower expression or slower constitutive turnover. In fact, it raises the intriguing possibility of driving internalization of targets that are internalized too slowly for efficient payload delivery, which are generally considered “non-internalizing.” Although rapid internalization was not beneficial for this PSMA model, biparatopic clustering in theory could also be used to overcome resistance mechanisms due to poor cellular internalization/trafficking (48). These data could be compared with truly “non-internalizing” ADCs that release their payload outside the cell (53, 54).

In summary, single-domain antibodies provide a promising platform for controlling the internalization kinetics, binding affinity, size, and plasma clearance of these ADC constructs. While the use of a variable heavy-chain binding domain provides improved transport characteristics (e.g., blood vessel permeability, diffusion), rapid clearance from the blood through the kidneys results in short plasma half-lives that limit therapeutic potential. The addition of a small albumin-binding domain improved plasma half-lives without significantly impacting the extravasation and tissue penetration of the agents, allowing “antibody-like” targeting efficiencies. The investigation of four ADC constructs (V_H1-HLE, V_H2-V_H1, V_H2-V_H1-HLE, and J591) showed markedly different in vitro potencies, tissue penetration, plasma clearance, and efficacy. Interestingly, neither plasma clearance/AUC nor in vitro potency correlated with tumor response. The systemic, organ, tissue, and single-cell pharmacokinetic data indicated that the ability to reach a maximum number of cells with a lethal payload dose was the driving factor in efficacy. These results were corroborated by a computational Krogh cylinder model. This combination of computational modeling and experimental quantitative pharmacology can be used to scale these results to other targeting systems and the clinic for more efficient design of highly efficacious agents.

S. Vance is an employee/paid consultant for Crescendo Biologics and has ownership interest (including patents) in Crescendo Biologics PSMA antibody patents. Q. Qiu is an employee/paid consultant for Dyne Therapeutics. A.O. Abu-Yousif is an employee/paid consultant for Takeda. T. Sandal is an employee/paid consultant for Crescendo Biologics and has ownership interest (including patents) in Crescendo Biologics PSMA antibody patents. J. Legg is an employee/paid consultant for and has ownership interest (including patents) in Crescendo Biologics. L. Thompson is an employee/paid consultant for and has ownership interest (including patents) in Crescendo Biologics PSMA. N. Goodwin is an employee/paid consultant for Roche Diagnostics GmbH and has ownership interest (including patents) in Crescendo Biologics PSMA antibody patents. G.M. Thurber is an employee/paid consultant for Nodus Therapeutics, reports receiving a commercial research grant from Crescendo Biologics, Takeda Pharmaceuticals, and Eli Lilly, has ownership interest (including patents) in Advanced Proteome Therapeutics, and is an unpaid consultant with Abbvie, AstraZeneca, ImmunoGen, Immunomedics, Novartis, Roche, Seattle Genetics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: I. Nessler, E. Khera, S. Vance, A.O. Abu-Yousif, T. Sandal, J. Legg, L. Thompson, G.M. Thurber

Development of methodology: I. Nessler, E. Khera, A. Kopp, Q. Qiu, L. Thompson, N. Goodwin, G.M. Thurber

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): I. Nessler, E. Khera, A. Kopp

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): I. Nessler, E. Khera, A. Kopp, A.O. Abu-Yousif, G.M. Thurber

Writing, review, and/or revision of the manuscript: I. Nessler, E. Khera, S. Vance, A. Kopp, T.A. Keating, A.O. Abu-Yousif, J. Legg, L. Thompson, G.M. Thurber

Study supervision: T.A. Keating, G.M. Thurber

We would like to thank Dr. Shuichi Takayama and lab members for assistance with setting up the tumor spheroids experiments and Dr. Geeta Mehta and lab members for discussion around tumor spheroids processing and quantification. DGN549 was provided by ImmunoGen, Waltham, MA. This study was funded, in part, by Takeda Pharmaceutics, Crescendo Biologics, NSF GRFP (to I. Nessler), and NIH Grant R35 GM128819 (to G.M. Thurber).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.