Investigating PSMA-Targeted Radioligand Therapy Efficacy as a Function of Cellular PSMA Levels and Intratumoral PSMA Heterogeneity

The prostate-specific membrane antigen (PSMA) is a plasma membrane glycoprotein highly overexpressed in metastatic castration-resistant (mCR) prostate cancer. It has been associated with cell migration, invasiveness, folate uptake, bone metastases, and poor prognosis (1). PSMA-targeting radioligands bind with high affinity to the extracellular domain of PSMA and deliver ionizing radiation predominantly to PSMA⁺ cells. Radioligand therapy (RLT) with ¹⁷⁷Lu-PSMA (and ²²⁵Ac-PSMA) has been introduced as a late-stage therapeutic alternative in mCRPC. Data from clinical studies suggest that PSMA-RLT is effective in patients with mCRPC as it delays disease progression, reduces serum prostate-specific antigen levels, and improves bone pain (2–4).

Eligibility for PSMA-RLT depends, among others, on PSMA expression on ⁶⁸Ga-PSMA PET/CT images. More than 90% of patients with mCRPC present with PSMA⁺ lesions (5). However, ∼50% of patients with mCRPC do not respond to PSMA-RLT despite documented PSMA expression (2, 4). Potential resistance mechanisms include insufficient tumor radiation dose, upregulated DNA damage response pathways, germline or somatic mutations, and low or heterogeneous tumor PSMA expression. The degree of PSMA expression likely affects tumor cell ¹⁷⁷Lu-/²²⁵Ac-PSMA uptake. ¹⁷⁷Lu/²²⁵Ac cytotoxicity is a consequence of DNA single- and double-strand breaks induced by beta particles (6, 7). Therefore, the degree of PSMA-radioligand uptake is likely associated with the degree of DNA damage and tumor cell death.

Here we investigated PSMA-RLT efficacy as a function of tumor PSMA expression levels. We also modeled tumor heterogeneity by creating subcutaneous xenografts with various fractions of PSMA⁺ tumor cells to assess RLT responses. We aimed to determine whether a threshold value for minimum PSMA levels required for therapeutic response exists.

Study 1: ¹⁷⁷Lu-PSMA617 RLT efficacy as function of PSMA expression/cell (Fig. 1A). RM1-YFP (PSMA-, YFP-only), RM1-PSMA low (PSMA⁺; ∼0.5 × 10⁴ PSMA/cell), RM1-PSMA medium (PSMA⁺⁺; ∼1.3 × 10⁴ PSMA/cell), and RM1-PSMA high (PSMA⁺⁺⁺; ∼4.9 × 10⁴ PSMA/cell; Supplementary Fig. S1A), respectively, were injected into Nod Scid gamma (NSG) mice (12 mice/group; day 0) to generate xenografts.

Study 2: The impact of the fraction of PSMA⁺ versus PSMA⁻ RM1 tumor cells on ¹⁷⁷Lu-PSMA617 RLT efficacy (Fig. 2A). PSMA⁻ and PSMA⁺⁺⁺ cells were mixed at different ratios (PSMA⁻:PSMA⁺⁺⁺ = 100:0, 75:25, 50:50, 25:75, 0:100) and injected into NSG mice (10 mice/group; day 0).

Study 3: To confirm the data obtained in the RM1 model, we tested the relevance of PSMA expression for RLT outcome in a human-derived PC model (study 3; Fig. 3A). NSG mice were injected with PC-3-PIP (PSMA⁺⁺⁺) and PC-3 (PSMA⁻) at different ratios (PSMA⁻:PSMA⁺⁺⁺ = 100:0, 66:33, 66:33, 0:100; 12 mice/group). The ratios of PSMA⁺ to PSMA⁻ cells was changed to 0% to 33% to 66% to 100% based on the results of study 2 (Fig. 2B; Supplementary Fig. S4A).

Study 4: Finally, to further elucidate the consequences of PSMA heterogeneity, PSMA⁻ and PSMA⁺⁺⁺ RM1 cells were mixed at different ratios (PSMA⁻:PSMA⁺⁺⁺ = 100:0, 66:33, 66:33, 0:100) and injected into NSG mice (10 mice/group; day 0) for treatment with ²²⁵Ac-PSMA617 (Fig. 4A).

Baseline tumor volume was determined on day 4 (RM1 model) or day 19 (PC-3 model). One to 2 days later, mice were treated with ¹⁷⁷Lu-PSMA617 (studies 1, 2, 3) or ²²⁵Ac-PSMA617 (study 4). Tumor volume was measured 3×/week (RM1 model) or 1 to 2×/week (PC-3 model) by CT volumetry. In the RM1 model, tumor growth studies were terminated already at day 14 post-RLT because of tumor size and mouse condition (beginning ulceration) as per veterinarian instruction. Absolute tumor volumes for all studies are shown in Supplementary Figs. S2A to S2D.

In a subset of mice (three mice/group; randomized group allocation), tumors were resected 2 days post-RLT. We chose this time point based on our preliminary biodistribution data (data not shown), which suggested a higher ¹⁷⁷Lu-PSMA617 uptake at 48 hours versus 4 hours following injection of 30 MBq ¹⁷⁷Lu-PSMA617 into mice with C4-2 tumors. The percent injected ¹⁷⁷Lu-/²²⁵Ac-PSMA617 activity incorporated into the tumors was determined by ex vivo gamma-counting. DNA damage and PSMA expression in the tumor samples were analyzed by anti-53BP1 and anti-PSMA IHC, respectively.

To ensure comparable tumor growth, characteristics across groups and for comparison to tumor volumes in RLT-treated mice tumor volumes of untreated mice were measured by CT (5 mice/group).

Murine RM1 (ATCC CRL-3310) sublines expressing no, low, medium, or high levels of human PSMA were described previously (8, 9). PC-3 were provided by K. Pillarsetty (Memorial Sloan Kettering Cancer Center). C4-2 were a kind gift from G. Thalmann (University of Bern, Bern, Switzerland). 22Rv1 cells were purchased from ATCC (CRL-2505). Cells were thawed 2 to 3 weeks prior to injection into mice. Cells were maintained in RPMI1640 medium supplemented with 10% FBS (Omega Scientific) at 37°C, 5% CO₂ and 20% O₂. Contamination with mycoplasma was excluded before preparing the cells for injection into mice using the Venor GeM Mycoplasma Detection Kit (Sigma-Aldrich). Human cell lines were authenticated in August or September 2019 using the Promega powerplex16 System and the small tandem repeat alleles were searched on the DSMZ database (Laragen Inc.).

All studies were approved by the UCLA Animal Research Committee (#2005-090). Intact male, 6 to weeks old NSG mice were obtained from the UCLA Radiation Oncology Animal Core. Mice housed under pathogen-free conditions with food and water ad libitum, and a 12- to 12-hour light–dark cycle. Veterinarian staff and investigators observed the mice daily to ensure animal welfare and determine if humane endpoints (e.g., hunched and ruffled appearance, apathy, ulceration, severe weight loss, tumor burden) were reached. All interventions were performed under anesthesia (2% isoflurane). RM1 (a total of 0.1 × 10⁶ cells in PBS: Matrigel = 1:1) or PC-3 sublines (a total of 5 × 10⁶ cells in 100 μL Matrigel) were inoculated subcutaneously into the left shoulder region. Mice weighed (mean ± SD) 30.9 ± 2.8 g (study 1), 29.5 ± 2.0 g (study 2), 28.7 ± 1.3 g (study 3), and 25.5 ± 0.6 g (study 4) at study start, and 29.2 ± 3.2 g (study 1), 31.4 ± 2.7 g (study 2), 25.6 ± 3.4 g (study 3), and 29.6 ± 0.75 g (study 4) at the endpoint.

Tumor volume was anlayzed by CT as described previously (8–10). Briefly, using OsiriX v.10.0.2 (Pixmeo SARL), tumors were delineated on ≥7 CT slices and the compute volume function was used to derive the tumor volume.

In studies 1 and 2 (RM1 model), mice received 60 MBq of ¹⁷⁷Lu-PSMA617 (84 GBq/μmol; UCLA Biomedical Cyclotron Facility) intravenously into the tail vein (9). In study 3 (PC-3 model), mice received 40 MBq ¹⁷⁷Lu-PSMA617 (84 GBq/μmol; ref. 11).

In study 4, mice with RM1 tumors received 40 kBq ²²⁵Ac-PSMA617 (130 MBq/μmol; UCLA Biomedical Cyclotron Facility) intravenously into the tail vein. For synthesis, [²²⁵Ac]Ac(NO₃)₃ was acquired through the Isotope Program, Office of Nuclear Physics, Department of Energy's Office of Science, and dissolved in 0.1 M HCl. Labeling of the precursor PSMA617 (in 1 M NaOAc/10 mg/mL gentisic acid) with the ²²⁵Ac-solution commenced at 90°C, 30 minutes, pH ∼5.5, and resulted in ²²⁵Ac-PSMA617 (>92% purity by radio thin-layer chromatography).

For ¹⁷⁷Lu-treated tumors, tumors were resected, weighed, and incorporated activity was measured in a gamma-counter (Cobra II Auto-Gamma; Packard Instrument Co.) with ¹⁷⁷Lu detection energy window of 189 to 231 keV. Gamma counting was delayed 24 hours for the ²²⁵Ac-treated tumors to allow ²²⁵Ac to reach the secular equilibrium state with its daughter nuclides. The activity of ²²⁵Ac was inferred from ²²¹Fr counts detected in the energy window with limits 170 to 260 keV. Data were decay corrected to the time of tumor resection and expressed as percent injected activity per gram tissue (%IA/g).

Tumor tissues were stained for the DNA damage marker 53BP1 and PSMA as described previously (9). Briefly, tissues were fixed in 10% formalin overnight and transferred to 70% ethanol for storage until radioactivity had decayed. Paraffin-embedded sections (4 μm) were de-paraffinized and rehydrated. Endogenous peroxidase was blocked (3% hydrogen peroxide/methanol, 10 minutes). Antigens were retrieved in heated 0.01 M citrate buffer, pH 6.0 (95 °C, 25 minutes). Specimens were incubated overnight at 4°C with a polyclonal rabbit anti 53BP1 antibody (1:2,000; Novus, NB100-304) or a mouse anti-PSMA antibody (clone 3E6; 1:50; DAKO, M362029-2) in BSA. For detection, the Dakocytomation Envision System labeled polymer horseradish peroxidase (DakoCytomation, Carpinteria) and the diaminobenzidine reaction (#BDB2004 L; Biocare Medical) were used according to the manufacturers' instructions. The sections were counterstained with hematoxylin. All slides were mounted with Cytoseal (Thermo Fisher Scientific) and scanned digitally at 20× magnification using ScanScope AT (Leica Biosystems, Vista).

For xenografts, semiautomated analysis of samples was performed using the Definiens Developer XD and Tissue Studio (Definiens AG). The immunoreactive score (IRS) was calculated by multiplying the staining intensity with the percent positive cells in the sample. For comparison of the relative PSMA expression between experimental groups in studies 2 to 4, a relative IRS was calculated by correcting for background staining (IRS of interest—IRS of PSMA- tumors) and normalization to the 100% PSMA⁺⁺⁺ group (i.e., 100% group = 100%).

PSMA expression on human samples was determined using a tissue microarray. This array (kind gift of Dr. J. Said, UCLA) contained human prostatectomy samples from 76 patients. For each patient, matched normal and tumor tissue specimen (three each) were available. A total of 199 tumor samples were available for analysis (29 samples without core or diagnostic material in the core). Analysis was performed by an experienced pathologist (DD) using a semiquantitative scoring system (0 = negative, 1 = weak, 2 = strong staining).

Cells (0.2–0.5 × 10⁶) were stained with an anti-hPSMA-APC antibody (dilution 1:10, incubation for 30 minutes at 4°C in the dark; clone REA408, Miltenyi Biotec). Absolute PSMA antigenic sites on RM1 and C4-2 lines were determined using the Quantum Simply Cellular human IgG beads (Bang Laboratories) according to the manufacturer's instructions. Samples were measured on a 5-laser LSR II cytometer (BD Biosciences) and analyzed using FlowJo (ThreeStar) software.

RM1 sublines were seeded on six-well plates (5,000 cells/well). On the next day, cells were irradiated (x-ray) with 2 or 5 Gy (dose rate 5.4 Gy/min; x-ray energy 300 kV at 10 mA, filtration 1.5 mm Cu and 3 mm AI, resulting in a half value layer of 3.0 mm Cu) or left untreated. Days to reach confluence were recorded as a measure of radiosensitivity. Alternatively, on day 4 (RM1) or day 6 (PC-3) after seeding, cells were prefixed by adding 4% formaldehyde/PBS to the culture medium for 2 to 5 minutes before medium removal and fixation with 4% formaldehyde/PBS for 15 minutes. Cells were stained with crystal violet solution [0.05% (w/v) crystal violet/MilliQ-water; 15 minutes], washed (three to four times, water) and dried.

RM1 cells (10⁵ cells/well) were seeded in triplicates for each condition and incubated overnight. Cells were washed once with RPMI1640/5% BSA (binding medium) and incubated for 30 minutes in binding medium before addition of ¹⁷⁷Lu-PSMA617 (0.2 nmol/L = 4.2 kBq per well). To determine specific binding, the PSMA inhibitor 2-(phosphonomethyl)pentane-1,5-dioic acid (PMPA; 10 μmol/L) was added to one triplicate per cell line. Following 1 hour incubation at 37°C, supernatant was transferred to gamma-counting tubes. Cells were washed with glycine.HCl pH 2.8 to remove membrane-bound radioligand and the wash solution was collected into a second set of tubes. Using 0.3N NaOH, cells were lyzed and collected into a third set of tubes. All samples were measured in a gamma-counter (see above). Data were decay corrected to the time of radiosynthesis completion. For analysis, the percent activity in the supernatant, membrane-bound, internalized, and overall taken up were determined. Specific binding was calculated by subtracting the values from samples incubated with PMPA from those of samples without inhibitor. Data were normalized to the cell number, which was determined in triplicate on the day of the experiment for each cell line.

Data are presented as mean ± SD. One-way ANOVA with Holm–Sidak correction for multiple testing was used to compare multiple groups. Significance was set at P < 0.05. Nonlinear centered second order polynomic regression was used for regression analyses unless indicated otherwise. GraphPad Prism (version 7, GraphPad Software, Inc.) was used for all statistical analyses. Data were analyzed blinded. A power analysis was performed (G*Power3.1) in relation to the primary goal of comparing tumor growth between experimental groups using the following parameters: t test for difference in mean of independent groups, alpha error 0.05, effect sizes of 1.3 SD (9, 10), 80% power.

To corroborate our hypothesis that heterogenous PSMA expression could attenuate PSMA-RLT efficacy in patients with prostate cancer, we analyzed PSMA positivity in prostate cancer specimens. Per semiquantitative analysis of PSMA expression, 48 of 199 (24%) samples scored 0, 26 of 199 (13%) scored 1+, 67 of 199 (34%) scored 2+, and 58/199 (29%) scored 3+. In line with previous studies (12–15), we thus confirmed PSMA heterogeneity on a per cell (varying PSMA staining intensity) and per lesion (spatial differences) basis (Supplementary Fig. S3).

To determine ¹⁷⁷Lu-PSMA617 efficacy as function of cellular PSMA expression levels (Fig. 1A, study 1), we used RM1 sublines with varying levels of PSMA (PSMA⁻, PSMA⁺, PSMA⁺⁺, PSMA⁺⁺⁺; Supplementary Fig. S1A). Differences in PSMA expression were maintained in vivo and did not affect general radiosensitivity (x-ray) of these cells in vitro or tumor growth in vivo (Supplementary Figs. S1B–S1E).

PSMA-RLT induced the most pronounced growth retardation in PSMA⁺⁺⁺, followed by PSMA⁺⁺ (PSMA⁺⁺ vs. PSMA⁺⁺⁺, P = 0.0015) and PSMA⁺ tumors (PSMA⁺ vs. PSMA⁺⁺, P = 0.0016). The low PSMA expression on PSMA⁺ was sufficient for an antitumor effect of RLT (PSMA⁺ vs. PSMA⁻, P < 0.0001). Tumors without PSMA expression did not respond to RLT (P < 0.0001 PSMA⁻ vs. PSMA⁺; PSMA⁻ RLT vs. untreated, P = 0.3517; Fig. 1B; Supplementary Table S1 and Supplementary Fig. S1C).

To investigate if the differences in PSMA-RLT efficacy across groups could be explained by differential ¹⁷⁷Lu-PSMA617 uptake, we measured ¹⁷⁷Lu-PSMA617 activity in vitro and in resected tumors. In line with published data (16), ¹⁷⁷Lu-PSMA617 effectively bound to cells in a PSMA-dependent manner and a large fraction of the activity was internalized. We confirmed a near perfect correlation between radioligand binding and PSMA expression (flow cytometry vs. membrane-bound activity, R² = 0.9490; vs. internalized activity, R² = 0.9992; vs. overall uptake, R² = 0.9992; Supplementary Fig. S1F).

Similarly, and in agreement with the tumor growth data, ¹⁷⁷Lu-PSMA617 tumor uptake positively correlated with in vitro ¹⁷⁷Lu-PSMA617 binding (R² = 0.9924; Supplementary Fig. S1F) and cell surface PSMA expression (R² = 0.9797; PSMA⁺⁺⁺ vs. PSMA⁺⁺, P < 0.0001; PSMA⁺⁺ vs. PSMA⁺, P = 0.0959; PSMA⁺ vs. PSMA-, P = 0.7789; Fig. 1C and D, Supplementary Table S1 and Supplementary Fig. S1D).

To further explore the mechanistic basis for the PSMA dependent differences in therapeutic efficacy, we quantified DNA damage in the resected tumors. The number of 53BP1 foci tended to increase with ¹⁷⁷Lu-PSMA617 uptake (R² = 0.9803; Fig. 1G) and with higher tumor cell PSMA levels (R² = 0.7154; PSMA⁺⁺⁺ vs. PSMA⁺⁺, P = 0.1473; PSMA⁺⁺ vs. PSMA⁺, P = 0.3563; PSMA⁺ vs. PSMA⁻, P = 0.6483; Fig. 1E, F, and H; Supplementary Table S1).

Next, we tested the efficacy of PSMA-RLT as function of PSMA⁺ to PSMA⁻ cell ratios in a given tumor (Fig. 2A, study 2). Following injection of 0%, 25%, 50%, 75%, and 100% PSMA⁺ RM1 tumor cells, respectively, we cross-checked PSMA expression in the resulting xenografts (Supplementary Fig. S4). The presence of stroma/connective tissue resulted in <100% cancer cells in all cases. However, comparison of PSMA IRS in between experimental groups was in line with the percent PSMA⁺ cells injected, except for the group 25% PSMA⁺ tumor cells. Relative PSMA IRS were (mean ± SD): group “100%,” 100 ± 13.9%; group “75%,” 75.5 ± 18.2%; group “50%,” 54.2 ± 6.2%; group “25%,” 50.2 ± 5.7%; group “0%,” 0 ± 0%.

Tumors containing only PSMA⁺⁺⁺ or PSMA⁻ tumor cells, respectively, exhibited similar responses to ¹⁷⁷Lu-PSMA617 as in study 1. Effectiveness of ¹⁷⁷Lu-PSMA617 increased with increasing fractions of PSMA⁺ tumor cells (100% vs. 75%, P = 0.0177; 75% vs. 50%, P = 0.0012; 50% vs. 25%, P = 0.5492; 25% vs. 0%, P < 0.0001). Congruent with the presence of ∼50% PSMA⁺ tumor cells in the 25% groups, the difference in tumor growth between the groups 25% and 50% PSMA⁺⁺⁺ cells was not significant (Fig. 2B; Supplementary Table S1).

¹⁷⁷Lu-PSMA617 uptake in resected tumors 2 days post-RLT tended to increase proportionally with the fraction of PSMA⁺ cells (R² = 0.9398; 100% vs. 75%, P = 0.7318; 75% vs. 50%, P = 0.4701; 50% vs. 25%, P = 0.7318; 25% vs. 0%, P = 0.0042; Fig. 2C; Supplementary Table S1 and Supplementary Fig. S4B).

Similarly, the number of 53BP1 foci tended to be higher in experimental groups with a higher fraction of PSMA⁺ cells (R² = 0.9372; 100% vs. 75%, P = 0.3330; 75% vs. 50%, P = 0.9036; 50% vs. 25%, P = 0.9787; 25% vs. 0%, P = 0.9036) and higher ¹⁷⁷Lu-PSMA617 uptake (R² = 0.9861; Fig. 2D–F; Supplementary Table S1).

To confirm our findings obtained in RM1 tumors in a human prostate cancer-derived model, we investigated the impact of PSMA heterogeneity in PC-3–based xenografts (Fig. 3A). PC-3 cells grow slower in vivo than RM1 cells (Supplementary Fig. S2), and express ∼11× more PSMA than PSMA⁺⁺⁺ RM1 cells (Supplementary Fig. S1A). PC-3 (PSMA-) and PC-3-PIP (PSMA⁺⁺⁺) displayed similar radiosensitivity (x-ray) in vitro and tumor growth in vivo (Supplementary Fig. S5).

Following injection of PC-3:PC-3-PIP cells in ratios of 0:100, 33:66, 66:33, or 100:0, relative PSMA IRS in tumors were (mean ± SD): group “100%,” 100.0 ± 36.3%; group “66%,” 41.3 ± 11.7%; group “33%,” 19.2 ± 7.9%; and group “0%,” 0.0 ± 0.0%. Treatment with ¹⁷⁷Lu-PSMA617 resulted in tumor shrinkage and effective disease control in mice with PC-3-PIP but not with PC-3 tumors. As in the RM1 model, the fraction of (PSMA⁺) PC-3-PIP cells was correlated with ¹⁷⁷Lu-PSMA617 RLT efficacy (100% vs. 66%, P < 0.0001; 66% vs. 33%, P = 0.0001; 33% vs. 0%, P < 0.0001; 0% vs. untreated PSMA⁻ tumors, P = 0. 9787; Fig. 3B; Supplementary Table S1; Supplementary Fig. S5D); at the end of the observation period (day 88), 9 of 9 mice were alive in the group “100%” PSMA⁺ cells, 5 of 9 in “66%,” and 0 of 9 mice in the groups “0%” and “33%” PSMA⁺ cells, respectively.

Consistently, ¹⁷⁷Lu-PSMA617 uptake (R² = 0.9827) and DNA damage (R² = 0.7910) were correlated with PSMA expression in the PC-3 model (uptake: 100% vs. 66%, P < 0.0001; 66% vs. 33%, P = 0.0046; 33% vs. 0%, P = 0.0009; 53BP1: 100% vs. 66%, P = 0.0159; 66% vs. 33%, P = 0.0.5838; 33% vs. 0%, P = 0.4323; Fig. 3C–F; Supplementary Table S1; Supplementary Figs. S5E and S5F).

Taken together, studies 1 to 3 suggest that degree of PSMA expression is correlated with ¹⁷⁷Lu-PSMA617 uptake and degree of DNA damage, all resulting in a better therapeutic outcome.

PSMA heterogeneity may increase the reliance on the crossfire effect for effective tumor targeting (study 4; Fig. 4A). To test this hypothesis, we treated mice with tumors containing various fractions of PSMA⁺ tumor cells with ²²⁵Ac-PSMA617, which has a shorter range in tissue than ¹⁷⁷Lu.

In agreement with injection of 0%, 33%, 66%, or 100% PSMA⁺⁺⁺ RM1 cells, relative PSMA IRS in resected tumors were (mean ± SD): group “100%,” 100 ± 34.4%; group “66%,” 62 ± 3.6%; group “33%,” 34 ± 7.7%; group “0%,” 0 ± 0%.

Similar to ¹⁷⁷Lu-PSMA617, effectiveness of ²²⁵Ac-PSMA617 increased with increasing fractions of PSMA⁺ tumor cells (100% vs. 66%, P = 0.0020; 66% vs. 33%, P = 0.0028; 33% vs. 0%, P = 0.0120; 0% vs. untreated PSMA- tumors, P = 0.5401). Differences between groups tended to be more accentuated (studies 2 vs. 4). Overall, treatment with ²²⁵Ac-PSMA617 resulted in improved tumor control, higher tracer uptake and DNA damage levels compared with ¹⁷⁷Lu-PSMA617 RLT (Fig. 4B; Supplementary Table S1; Supplementary Fig. S6).

In agreement with the differences in therapeutic efficacy between groups, ²²⁵Ac-PSMA617 uptake correlated with the fraction of PSMA⁺ cells in resected tumors (R² = 0.9967; 100% vs. 66%, P = 0.0291; 66% vs. 33%, P = 0.0532; 33% vs. 0%, P = 0.0416). Correspondingly, higher uptake resulted in enhanced DNA damage (R² = 0.8329; 100% vs. 66%, P = 0.0497; 66% vs. 33%, P = 0.3142; 33% vs. 0%, P = 0.3142; Fig. 4C–F; Supplementary Table S1; Supplementary Fig. S6).

Thus, although radioisotopes with long ranges in tissue may, at least partly, be able to compensate for PSMA heterogeneity because of their crossfire, the higher linear energy transfer of isotopes like ²²⁵Ac may still result in enhanced radiation-induced cytotoxicity.

Here we investigated whether PSMA expression/cell and the fraction of PSMA⁺ cells, as markers of tumor heterogeneity, are determinants of RLT efficacy in murine models of prostate cancer. We demonstrate that both are critically important for effective tumor targeting with PSMA-RLT: The PSMA expression level is proportional to the radioligand tumor uptake which, in turn, is associated with the degree of DNA damage.

Several studies have demonstrated intrapatient variations in both PSMA expression and the fraction of PSMA⁺ cells in given tumors. These variations can arise (i) from dynamic processes (12–14, 17–19), including drug-induced alterations (10, 20–23); (ii) from the presence of highly PSMA expressing neovasculature in cancer types without PSMA expression in tumor cells (5, 14, 17, 24–27); and (iii) from areas within the tumor containing PSMA⁻ cells (12, 13). PSMA heterogeneity might still result in positive image contrast on ⁶⁸Ga-PSMA PET scans (24, 28–30) and thus, in the decision to administer RLT. Therefore, PSMA heterogeneity might be an underappreciated contributor to RLT failure.

Currently, differences in PSMA expression can most accurately be captured by analyzing tumor biopsies (e.g., IHC). However, biopsying every lesion is not feasible in patients with mCRPC. A potential future alternative to biopsies may be the radiomic analysis of ⁶⁸Ga-PSMA tumor uptake in which phenotypic information is extracted from PET/CT images (31–33). Because of the limitations of radiomics (34, 35) and biopsies, clinicians rely on “expression levels” by means of PET standardized uptake values (SUV) and ⁶⁸Ga-PSMA uptake above liver activity in the majority of metastases (but not necessarily in all) is considered a condition for RLT initiation (4, 37). SUVs inform about interlesion PSMA heterogeneity but can be confounded by small lesion size (partial-volume effect). PET imaging cannot, or only to a limited extent (36), resolve intralesion heterogeneity although intralesion heterogeneity/homogeneity in target expression might be one factor underlying interlesion differences in ⁶⁸Ga-PSMA SUVs.

This study suggests that low PSMA expression is associated with markedly attenuated, yet still measurable RLT effects. This finding is in line with the observation that the overall lower PSMA expression in PSMA⁻ tumors with PSMA⁺ neovasculature was sufficient for an antitumor effect of ¹¹¹In-J591, a radio-labeled PSMA-binding antibody (26). However, undertreatment not only reduces RLT efficacy, but might also select treatment-resistant tumor clones.

One solution to account for potential intra- and interlesion PSMA heterogeneity and to optimize safely delivered maximal tumor doses is individual patient dosimetry (38–41). This approach provides overall tumor radiation dose measurements independent of intra- and interlesional PSMA heterogeneity. Presently, patients usually receive a standard activity (mostly ∼7 GBq) without dosimetric considerations of the maximal doses that could be achieved with minimal healthy organ damage. However, using dosimetry and pharmacodynamic modeling, it has been suggested that lesions with high PSMA expression could be treated with activities higher than the standard activity without exceeding the tolerable biological effective dose for organs at risk such as kidney and salivary glands (42–44). Therefore, it seems feasible to optimize RLT efficacy in patients with high and homogenous PSMA expression by increasing the administered activity (and peptide amount). In patients with low PSMA levels and/or high interlesion heterogeneity, dosimetry might help to determine whether the tumor absorbed doses that could be achieved are sufficient to induce a significant therapeutic effect. It would also indicate the fraction of metastases that could be targeted successfully versus the number of metastases that would likely receive an insufficient dose. Therefore, dosimetry would provide valuable information for the decision whether to initiate RLT or to choose an alternative treatment regimen. By treating only those patients who are likely to benefit, more patients would be spared an unsuccessful (and costly) treatment regimen, and success rates of RLT would be increased.

It is possible that PSMA-PET/CT uptake intensity as measured by SUV parameters may be predictive of response/nonresponse to RLT (e.g., definition of a SUV cut-off). Uptake intensity by PET may correlate with absorbed doses which in turn may contribute to optimizing RLT activities (45, 46). Indeed, first clinical data suggest that ⁶⁸Ga-PSMA uptake (SUV) on pretherapy PET/CT correlates with the absorbed ¹⁷⁷Lu-PSMA dose and prostate-specific antigen response (37, 47, 48). Consistently, a recent ¹⁷⁷Lu-PSMA RLT clinical trial in which patients with low ⁶⁸Ga-PSMA relative to ¹⁸F-fluorodeoxyglucose uptake were excluded (4) reported higher response rates than reported for less stringently selected patients (2). A potential limitation of both, dosimetry- and PSMA-PET/CT-based approaches is the presence of PSMA⁻ lesions that might remain undetected and can give rise to tumor relapse. Thus, patients may benefit from additional PET/CT scans with an alternative tracer (e.g., fluorodeoxyglucose; ref. 4). A further cause for relapse not captured by dosimetry or PET might be micrometastatic disease below the imaging detection limit.

RLT efficacy might be improved by pharmacologic upregulation of PSMA (21). In this study, increases in target availability positively correlated with ¹⁷⁷Lu-/²²⁵Ac-PSMA617 tumor uptake and antitumor effect. This finding is in line with a study showing higher uptake of a PSMA-binding PET probe in LNCaP versus 22Rv1 (lower PSMA expression than LNCaP) xenografts (49). The RM1 models used here express similar PSMA levels as 22Rv1, but at least 5x and 11x less PSMA than C4-2 and PC-3-PIP, respectively (Supplementary Fig. S1A). We previously showed that drug-induced PSMA upregulation in C4-2 tumors did not enhance ¹⁷⁷Lu-PSMA617 efficacy (10). Therefore, although patients with low PSMA uptake on PET/CT might benefit from PSMA induction, increases in PSMA levels beyond a certain, yet to be defined threshold, might not improve RLT outcome.

Finally, low-dose irradiation (x-ray) of tumors was shown to enhance cancer immunotherapy by re-programing macrophages to a state that promotes tumor infiltration and cell kill by cytotoxic T cells (50). It is thus possible that—in immunocompetent animal models and in patients—RLT against low PSMA expression tumors may synergize with immunotherapy.

PSMA RLT is effective even in tumors with low PSMA levels or with a small number of PSMA⁺ cells. However, optimal antitumor efficacy of RLT hinges, among other factors, on homogenously high target expression. Clinical studies designed to determine intra- and interlesion PSMA heterogeneity and to optimize PSMA-RLT for each patient are highly warranted.

C.G. Radu holds ownership interest (including patents) in Sofie Biosciences and Trethera Corporation. No potential conflicts of interest were disclosed by the other authors.

Conception and design: C.G. Radu, K. Lueckerath

Development of methodology: C.E. Mona, K. Lueckerath

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K. Current, C. Meyer, C.E. Magyar, J. Almajano, R. Slavik, A.D. Stuparu, C. Cheng, D.W. Dawson, J. Czernin, K. Lueckerath

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K. Current, C. Meyer, C.E. Magyar, D.W. Dawson, K. Lueckerath

Writing, review, and/or revision of the manuscript: K. Current, C. Meyer, C.E. Magyar, J. Czernin, K. Lueckerath

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K. Current, C.E. Mona, J. Almajano, R. Slavik, D.W. Dawson

Study supervision: J. Czernin, K. Lueckerath

We thank L. Wei for assistance with radioligand injections and acknowledge the UCLA Translational Pathology Core Laboratory and Rana Riahi for IHC stainings. This study was partially funded by the US Department of Energy, Office of Science Award (DE-SC0012353), the Prostate Cancer Foundation (17CHAL02), and the UCLA SPORE in Prostate Cancer (P50 CA092131). K. Lueckerath received a scholarship from the German Research Foundation (Deutsche Forschungsgemeinschaft, DFG, grant no. 807454).

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.