RO4, a high-affinity humanized antibody against the juxtamembrane region of mesothelin for targeted cancer therapy

Abstract

Background

Mesothelin (MSLN) is a surface antigen highly expressed in several solid tumors, including mesothelioma, ovarian, and pancreatic cancers. However, therapeutic efficacy of MSLN-targeted agents is often compromised by shed MSLN (SM), which acts as a soluble decoy and accumulates in tumor microenvironments, reducing antibody engagement at the tumor surface.

Methods

To overcome this barrier, we generated antibodies targeting the membrane-proximal, non-shed region of MSLN using a peptide encompassing major cleavage sites for rabbit immunization. From 200 B-cell clones, 14 antibodies specific to the juxtamembrane region of MSLN were identified. The lead candidate, RO4, underwent detailed characterization and humanization to improve clinical applicability.

Results

Humanized RO4 (hRO4) exhibited enhanced binding affinity to MSLN and specifically recognized tumor-associated, non-shed epitopes. Structural analysis confirmed precise epitope engagement near the cleavage site. hRO4 effectively inhibited mesothelin shedding in vitro and enabled potent tumor eradication when expressed in chimeric antigen receptor (CAR) T cells in NOD scid gamma mouse models.

Conclusions

Targeting a non-shed epitope of MSLN with hRO4 overcomes a critical limitation of conventional MSLN-directed therapies. By avoiding decoy interference and enhancing tumor-specific targeting, hRO4-based therapeutics offer promising clinical potential for improving outcomes in MSLN-positive cancers.

Statement of Significance We developed RO4, a high-affinity antibody that binds a non-shed juxtamembrane epitope, thereby avoiding interference from SM. Humanized RO4 retains strong binding and enables potent CAR T-cell activity in preclinical models. These findings support RO4 as a promising candidate for next-generation mesothelin-targeted immunotherapies.

Introduction

Mesothelin (MSLN) is expressed on many types of cancer cells, including mesothelioma, ovarian, and pancreatic cancers, establishing it as a promising therapeutic target [1]. Normal expression of MSLN is limited to a few organs, such as the pleura, pericardium, and peritoneum. MSLN is synthesized as a precursor protein of 622 amino acids [2]. This precursor is processed inside the cell and cleaved by furin, resulting in the mature form of MSLN, which spans amino acids 296 to 598 and is attached to the cell membrane through a glycosylphosphatidylinositol anchor [3]. MSLN is shed from cells by the action of several proteases [4, 5]. Shed MSLN (SM) acts as a barrier to antibody-based therapies targeting MSLN. Our previous research indicates that several proteases can release SM by cutting at several sites near the cell membrane. The levels of SM are significantly elevated in tumors and ascites [6, 7]. The primary shedding sites are located after amino acids 584, 586, and 591 [4, 5, 8]. We have previously immunized mice with a MSLN peptide (582–598), which encompasses these shedding sites, and isolated mAb 15B6. This antibody binds to full-length MSLN but does not bind to SM. The Fv of mAb 15B6 produces very active chimeric antigen receptor (CAR) T cells and bispecific antibodies [8–10].

To develop an antibody with distinct characteristics from mAb 15B6, capable of recognizing potentially a different peptide conformation, rabbits were used as immunization hosts due to their diverse immune system, which often reacts to a broader range of antigens compared to mice [11, 12]. Furthermore, rabbits possess a larger genetic diversity and a more mature immune system, which can lead to stronger and more specific antigen responses [13–15]. From the immunized rabbits’ peripheral blood, we performed B-cell cloning and isolated a highly potent novel mAb, RO4. We also humanized mAb RO4 via complementarity determining region (CDR) grafting. In this paper, we describe the binding characteristics of mAb RO4, the structure of the antibody peptide complex, its effectiveness in inhibiting the shedding of MSLN, and its efficacy in hRO4–CAR T cells.

Materials and methods

Cell lines

Cell lines OVCAR8, KLM1, T3M4, RH29, SW48, and A431 have been previously described [4, 5, 16–18]. The luciferase-expressing derivatives of these cell lines, namely, OVCAR8-luc, KLM1-luc, A431-luc, RH29-luc, T3M4-luc, and SW48-luc, were obtained from R. Hassan, M. Ho, and C. Alewine at the National Cancer Institute (Bethesda, MD). All cell lines were cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum and 1% penicillin–streptomycin. The cells were incubated at 37°C in a 5% CO₂ atmosphere.

Isolation of rRO4 clone targeting the C-terminal region of mesothelin

The C-terminal peptide of MSLN, spanning amino acids 582–598 (IPNGYLVLDLSMQEALS), was conjugated with Keyhole limpet hemocyanin (KLH) and used as an immunogen in rabbits. Four rabbits were immunized multiple times with 200 μg of the peptide, and high antibody titers were confirmed. Peripheral blood mononuclear cells (PBMCs) were isolated from fresh blood collected from one of the immunized rabbits. B Cells enriched for the free peptide were plated into fifty 96-well plates for a single round of B-cell cloning. The conditioned medium from these cultures was screened by Enzyme-Linked ImmunoSorbent Assay (ELISA) for reactivity with the free peptide. Subsequently, the binding affinity to both full-length and truncated MSLN was assessed by ELISA, resulting in the identification of 14 mAb clones. Among the 14 clones, the RO4 clone demonstrated optimal binding characteristics. The variable region (V-region) of the RO4 clone was sequenced, and the antibody sequences were cloned into mammalian expression plasmids. Transfected mammalian cells produced antibodies, and the culture supernatants were collected. Antibody validation was performed using ELISA, confirming the specificity and reactivity of the expressed antibodies. The sequence data for rRO4 is provided in Fig. S3.

Immunohistochemistry staining with rRO4

Tissue sections were cut at 5-μm thickness. Automated immunohistochemistry (IHC) staining was conducted using a Leica Biosystems’ BondRX system under the following parameters: Epitope Retrieval 2 (EDTA) for 20 min, followed by application of RO4 antibody at a concentration of 0.02 μg/ml for 30 min. The Bond Polymer Refine Detection Kit (Leica Biosystems #DS9800) was utilized, then 3,3′-diaminobenzidine chromogen was applied to visualize the antigens, and sections were counterstained with hematoxylin. As a negative control, Rabbit monoclonal IgG XP Isotype Control (Cell Signaling Technology) was used in place of the primary antibody. Images of the stained sections were captured using the Aperio Scanscope FL (Leica Biosystems, Buffalo Grove, IL) whole slide scanner. Tissue evaluation was performed by an independent research pathologist who assessed MSLN expression using the RO4 antibody. The tissues were organized into a microarray (FDA999w3; TissueArray.com) consisting of 99 cores, which included normal tissue samples from 32 human organs, collected from three different individuals. The majority of the tissues were sourced from autopsies. The layout of the cores is shown in Fig. S2B.

ELISA for rRO4 binding to MSLN protein

Microtiter plates were coated with variants of MSLN protein: MSLN (296–599 M) or MSLN (296–599 V), each at a concentration of 1 μg/ml, using 50 μl per well. After coating, the plates were washed to remove unbound protein. The rRO4 antibody was then added under various concentrations to assess binding. Detection of bound rRO4 was achieved using a horseradish peroxidase–conjugated goat anti-rabbit IgG (Catalogue No. 111-005-046; Jackson ImmunoResearch Laboratories, Inc.) followed by the addition of 3,3′,5,5′-tetramethylbenzidine substrate (Thermo Fisher Scientific). The reaction was terminated with 2 M sulfuric acid. Absorbance at 450 nm was measured using an automated plate reader (Molecular Devices Corp., Sunnyvale, CA).

Affinity measurement of antigen to antibody by surface plasmon resonance

Affinity measurements between antigen and antibody were performed using a Biacore 8000 system (Cytiva) following the protocol provided by Genscript Probio (NJ). The experiments were conducted at 25°C using HBS-EP+ as the running buffer. The antibody was immobilized on a Series S Sensor Chip with Protein A to facilitate capture. Antigen samples were prepared at various concentrations for experiments. These samples were injected sequentially over the surfaces of flow cells 1 and 2 to observe the association phase of the binding interactions. This was followed by the injection of running buffer to monitor the dissociation phase. The binding kinetics are presented as sensor grams in Figs. 2B–G. Quantitative binding data, including kinetic constants and affinity measurements, are provided in Table S2.

Figure 2.

Crystal structure and binding affinities of mAbs 15B6 and RO4. (A) Upper panel: Crystal structure of the mAb 15B6 Fab in complex with the C-terminal peptide of MSLN. Left: The heavy (VH) and light (VL) chains of the Fv are shown as cartoon models, overlaid with a semi-transparent surface. The bound peptide seen in crystal (from N584 to S598) is shown as cartoon in black interacting with both VH and VL. Middle: The MSLN C-terminal peptide sequence from N584 to S598 with interacting residues colored in red and proteolytic sites indicated with arrows. Right: The 90° rotated view of the peptide bound to the mAb 15B6 is given. The peptide is rendered as cartoon in black with side chain residues shown as stick models and labeled with the same color codes. The residue M593, important for the M/V variants, is labeled in red. Lower panel: Crystal structure of the mAb RO4 Fab in complex with the C-terminal peptide of MSLN. Left: VH and VL are rendered as cartoon models, which is overlaid with a semi-transparent surface in green and cyan, respectively. The bound peptide in black is shown as cartoon model. Middle: The sequence of the MSLN peptide from I582 to S598 is given showing the critical residues for the epitope in red and the proteolytic sites indicated by arrows. Right: The 90° rotated view is given. (B-G) Binding sensorgrams of mAbs to MSLN deletion protein by Biacore. MSLN (296-599, M) (B,C,D) or MSLN (296-599, V) (E,F,G) was used for measurement. Sample concentrations tested were 70, 35, 17.5, 8.75, 4.38, 2.19 and 1.09 nM (B,C), 200, 100, 50, 25, 12.5, 6.25, and 3.125 nM (D,G), or 390, 195, 97.5, 48.75, 24.38, 12.19, and 6.09 nM (E,F). The binding data is shown in supplementary Table S2.

Crystallization and structure determination of the RO4 Fab fragment with the C-terminal peptide of MSLN

The Fab fragment of the RO4 antibody was prepared using the Pierce Fab preparation kit (Thermo Fisher Scientific) following the manufacturer’s instruction as described previously [19]. Purified Fab was concentrated to 10 mg/ml using an Amicon Ultra concentrator (Millipore) with a molecular weight cutoff of 30 kDa. The synthetic MSLN C-terminal peptide (residues 582–598) powder (GenScript) was dissolved in buffer containing 20 mM Tris, pH 7.5, and 100 mM NaCl to a final concentration of 4.4 mg/ml. The peptide solution was mixed with concentrated Fab at a molar ratio of 3:1 and left on a horizontal orbital shaker at 4°C overnight before setting up the crystallization trays.

The crystallization was performed robotically by the hanging-drop vapor diffusion method. An aliquot of 0.24 μl protein complex was mixed with 0.24 μl reservoir solution (100 μl) containing 0.1 M Bis–Tris, pH 6.5, and 16% PEG 10000. Plate-shaped crystals appeared 10 days after incubating at 4°C. Crystals were treated with a cryoprotectant solution made of 80% glycerol and reservoir solution in a 1:3 ratio prior to flash cooling in liquid nitrogen.

An X-ray diffraction dataset was collected at 100 K using the NYX (19-ID) beamline at the National Synchrotron Light Source II, Brookhaven National Laboratory. Frames of diffraction data images were processed using HKL2000 [20]. The structure was solved by the molecular replacement method using the BALBES program in the CCP4 program suite [21, 22]. The atomic model was built using the modeling program COOT and refined in Phenix [23, 24].

Detection of surface MSLN expression by flow cytometry

Cells were harvested and stained with either h15B6, hRO4, or no antibody for 30 min. After washing, cells were stained with anti-human IgG(H + L) conjugated to AlexaFluor-647. Data were obtained using a Sony ID7000 spectral analyzer (Sony Biotechnology). A total of 10 000 counts were collected and analyzed using the FlowJo v10.4.2 software.

Flow cytometry–based internalization assay

Anti-MSLN antibodies were labeled with Alexa 647 antibody labeling kit (A20186; Thermo Fisher Scientific). Then, 1 × 10⁵ cells were plated in a 24-well plate. After 4 h of culturing at 37°C, media was changed to the media containing antibody (2 µg/ml in 300 ml). After incubating for indicated times, cells were washed with glycine stripping buffer and PBS before trypsinization. Flow cytometry analysis was performed after resuspended in the FACS buffer (PBS containing 0.02% FBS).

Generation and expansion of CAR T cells

Plasmid DNA encoding a second-generation CAR was packaged into lentivirus. Human PBMCs were stimulated with anti-CD3/CD28 Dynabeads (Thermo Fisher Scientific) and transduced using protamine sulfate at a concentration of 10 μg/ml. Additionally, the culture medium was supplemented with 300 U/ml of interleukin-2 (IL-2) (Miltenyi Biotec) in RPMI-1640 medium to support cell growth and CAR expression. The transduced PBMCs were maintained by replenishing the culture with fresh medium containing IL-2 every other day. On Day 8 post-transduction, the cells were harvested for transduction efficacy analysis. This involved staining the cells with anti-EGFR-AF488 (R&D Systems, FAB9577G) to assess CAR expression and anti-CD3-APC (BioLegend, 300312) to confirm T-cell identity [25, 26]. The stained cells were obtained using a Sony ID7000 spectral analyzer (Sony Biotechnology), and the resulting data were processed using FlowJo software. The cells were collected and cryopreserved prior to Day 11 to ensure viability and functionality for further experiments [17].

Collection and preparation of ascites samples

Ascites samples were obtained from patients diagnosed with mesothelioma at the National Institutes of Health (NIH) under an Institutional Review Board–approved protocol (ClinicalTrials.gov identifier NCT01950572). Written informed consent was obtained from all participants prior to enrollment. The study adhered to the International Conference on Harmonization–Good Clinical Practice guidelines.

Luminescent-based cytolytic assay

Effector CAR T cells were cocultured with luciferase-expressing target cells at various effector-to-target (E:T) ratios for 20 h. This setup was designed to assess the cytolytic activity of the CAR T cells against the target tumor cells.

Animal model and cell implantation

A total of 5 million OVCAR8 or KLM1 luciferase-expressing cells were implanted either intraperitoneally or subcutaneously into NOD scid gamma (NSG) mice 6 to 10 weeks of age. This approach allowed for the establishment of a model to study the effectiveness of CAR T-cell therapies in an in vivo setting [8].

CAR T-cell administration

On Day 7 post-implantation, the mice received a single intravenous injection of 5 million cells. The cells injected included mock-transfected T cells, h15B6 CAR T cells, or hRO4 CAR T cells, depending on the experimental group. Tumor progression and CAR T-cell efficacy were monitored using the Xenogen IVIS Lumina imaging system (PerkinElmer) following intraperitoneal injection of 1.5 mg of D-luciferin (PerkinElmer). The bioluminescence signal flux from each mouse was quantified as radiance (photons per second) using Living Image software, which facilitated the noninvasive tracking of tumor growth and response to treatment. Mice were monitored daily for signs of distress or illness, and humane endpoints were strictly observed [8]. Mice showing any signs of sickness were euthanized in accordance with the ethical guidelines.

Statistics

Data are presented as mean ± standard deviation (SD) from three independent experiments, unless otherwise specified. Statistical analyses were performed using GraphPad Prism software. The Mann–Whitney U test was employed to compare differences between two independent groups. The cumulative probability of survival, defined as the time until tumor volume reached 400 mm³, was estimated using Kaplan–Meier survival curve analysis. Differences in survival distributions among the groups were assessed using the log-rank (Mantel–Cox) test and Wilcoxon test.

Results

Isolation of mAb RO4

To produce new antibodies to MSLN, we immunized four rabbits with the peptide from the C terminus of MSLN (IPNGYLVLDLSMQEALS). The peptide was attached to KLH. We isolated 14 B-cell clones reacting with full-length MSLN but not SM (Table 1, Fig. S1). We used recombinant MSLN (296–591) as a surrogate for SM [4, 5, 8]. Nine of these mAbs (RO2, RO3, RO4, RO5, RO6, RO10, RO12, RO13, and RO14) bind to full-length MSLN but do not bind to MSLN (296–591). The affinity of the nine mAbs to full-length MSLN was quantified via ELISA, leading to the exclusion of three mAbs (RO10, RO13, and RO14) with low affinity. The binding of the remaining mAbs to OVCAR8 cells was then quantified by flow cytometry. Four mAbs (RO3, RO4, RO5, and RO12) demonstrated a geometric mean fluorescence intensity >5.5, indicating strong binding to live cells (Table 1). Their specificity was assessed by immunohistochemistry. mAbs RO3, RO5, and RO12 non-specifically bind to many types of normal cells and were discarded. Because mAb RO4 only reacted with normal cells known to express MSLN, it was selected for further analysis. Examples of strong membranous staining with mAb RO4 in MSLN-positive mesothelioma cells are shown in Figs. 1A and 1B, with no nonspecific staining of normal tissues (Fig. S2). Mesothelial cells lining the mesentery and peritoneum were positive (Fig. S2). Rare epithelial cells in the tonsils (samples E1, E2, and E3) also tested positive for MSLN. MSLN expression was absent in other tissues. Occasional false-positive cytoplasmic staining was noted in the stomach, kidneys, and uterus, but no membranous staining was observed.

Table 1.

Anti-MSLN antibodies

Name	Reactivity in FACSa log GeoMean	Reactivity to MSLN-Hisb affinity (KD, nM)	Reactivity to MSLN (296–585)c	Reactivity to MSLN (296–591)c	Cross-reactivityd to normal tissue by IHC
RO1	4.40	18.6	−	+	ND
RO2	5.46	40.3	−	−	ND
RO3	5.57	18.9	−	−	+
RO4	5.48	26.5	−	−	−
RO5	5.57	43.9	−	−	+
RO6	4.56	26.5	−	−	ND
RO7	5.15	26.4	−	+	ND
RO8	5.20	35.4	−	+	ND
RO9	5.15	12.0	−	+	ND
RO10	4.03	91.6	−	−	ND
RO11	4.06	35.4	−	+	ND
RO12	5.61	28.8	−	−	+
RO13	3.89	>100	−	−	ND
RO14	4.06	99.8	−	−	ND

^aGeometric mean fluorescence intensity in a FACS analysis using OVCAR8 cells. OVCAR8 cells express human MSLN.

^bDefined by an ELISA measurement of mAb concentration (20 h, 25°C).

^cReactivity to MSLN fragment in ELISA. Parts of the results are shown in Supplementary Fig. S1. +, positive; −, negative.

^dCross-reactivity to normal tissue was evaluated by immunohistochemistry (IHC). +, mAb reacts some normal tissues; −, mAb does not react with normal tissue.

Figure 1.

Binding of rRO4 antibody to fixed tumor samples. (A) Microscopic image of the IHC staining of human malignant mesothelioma samples with rRO4 antibody. Scale bar represents 3 mm. (B) An enlarged image of the area enclosed by the square in the center of the Fig. 1A. Scale bar indicates 200 µm.

Peptide interacts with 15B6 and RO4 differently as revealed by the structure of RO4 with bound peptide

To elucidate at atomic resolution the interaction between the C terminus of MSLN and mAb RO4 and to understand the structural basis for the observed difference in affinity between RO4 and 15B6, we crystallized the Fab fragment of mAb RO4 in complex with the 17-residue peptide from the MSLN C terminus and determined the structure of this complex to 1.52-Å resolution. The structure revealed a peptide conformation distinct from that seen in the complex with 15B6 (Fig. 2A). Although all 17 residues of the peptide are clearly defined in the electron density map, the first four (I582–G585) are apparently stabilized by crystal contacts, whereas the rest (Y586–S598) except one (L587) form the binding epitope, participating in critical interactions mostly in the forms of hydrogen bonding and van der Waals interactions as listed in Table S1.

The RO4 binds the peptide along the groove between variable light (VL) and variable heavy (VH) subunits, which is different from binding of 15B6 to the same peptide. In the latter case, the peptide lies across the surface perpendicular to the inter-subunit groove (Fig. 2A). Unlike the extended conformation of the peptide seen when binding to 15B6 (PDB: 7U8C), the epitope of the peptide interacting with RO4 has only the first five residues (Y586–D590) adopting an extended conformation that is recognized by the CDR1 and CDR3 of light chain; the other seven residues form a two-turn helix (L591–M597) that lays snuggly in the groove created between the heavy and light chain CDRs. All six CDRs of the Fab heavy and light chains are involved in antigen recognition, although the LC appears to be engaged more with the peptide as indicated by the buried surface area calculation (377 Å² for LC and 221 Å² for HC). As binding affinity is often determined by the number of H-bonds, we found that the peptide forms 11 H-bonds with RO4, significantly more than it does with 15B6.

Humanization of mAb RO4

To humanize mAb RO4, the CDRs of the VH and VL regions of the parental antibody were identified as antigen-contacting residues using the combined Kabat, IMGT, and Paratome CDRs, as described previously [27]. As a result, the CDRs of mAb RO4 were grafted onto the framework of human germline sequences IGHV3–66^*01, IGHJ4^*01, IGKV1–27^*01, and IGKJ4^*01. Figure S3 details the amino acid sequences of the humanized Fv (hRO4), its alignment with the original rabbit Fv (RO4), and the human VH and VL germline sequences used in the humanization process. Changes from rabbit to human sequences are marked in pink boxes.

Biacore analysis of antibody affinities

Position 593 of MSLN is often methionine and less frequently valine. Since both antibodies bind to this region, we determined the binding affinities of hRO4 and h15B6 antibodies to the M and V variants of MSLN (Figs. 2B–G and Table S2). Our analyses, using Biacore, revealed differences in binding affinities between the two antibodies. With the M593 variant of MSLN (MSLN (296–599, M)), hRO4 demonstrated a higher affinity (KD = 4.64 nM) compared to h15B6 (KD = 77.8 nM). With the V593 variant (MSLN (296–599, V)), hRO4 had a superior affinity (KD = 83.5 nM) compared to h15B6 (KD = 482 nM).

Expression of MSLN on human cell lines

Using flow cytometry, we also investigated the binding affinities of the humanized antibodies hRO4 and h15B6 to various MSLN-expressing cancer cell lines, employing 1 μg/ml of each antibody, and detected the bound antibody with anti-human IgG conjugated to Alexa Fluor 647 (Figs. 3A–F). The cell binding assays were conducted on ovarian cancer (OVCAR8), pancreatic ductal adenocarcinoma (KLM1 and T3M4), malignant mesothelioma (RH29), and colorectal cancer (SW48). hRO4 demonstrated higher binding affinities than h15B6. In OVCAR8 cells, hRO4 exhibited a geometric mean fluorescence intensity (GM) of 12 810, which is 1.2 times higher than that of h15B6, which had GM of 10 684 (Fig. 3A, Fig. S4). In KLM1 cells, hRO4 showed GM of 2587, reflecting a 1.6-fold increase over h15B6’s GM of 1599. In RH29 cells, hRO4’s binding (GM 7445) was 2.2 times stronger than that of h15B6 (GM 3351) (Fig. 3C, Fig. S4). The most pronounced differences were observed in SW48 and T3M4 cells, where hRO4’s binding was 6.6 and 65 times stronger, respectively, compared to h15B6, with GM of 384 versus 58.3 for SW48 and 1976 versus 30.5 for T3M4. Neither antibody exhibited binding to epidermoid carcinoma cells (A431), which are negative for MSLN expression.

Figure 3.

Characterization of hRO4 and h15B6. Binding of hRO4 and h15B6 to OVCAR8 (A), KLM1 (B), RH29 (C), SW48 (D), T3M4 (E), and A431 (F). Individual cell lines were analyzed by flow cytometry after staining with 1 µg/ml of either h15B6 (red), hRO4 (blue), or no antibody (orange), with anti-human IgG (H + L) conjugated to AF647. The stained cells were then analyzed using a Sony ID7000 spectral analyzer (Sony Biotechnology). Data acquisition and subsequent analysis were performed using FlowJo software. h15B6 and hRO4 inhibit MSLN shedding (G). A total of 50 μg/ml of h15B6, hRO4, or hIgG1 antibody was incubated with each of the 15 000 cells for 24 h. The media were collected and shed MSLN was detected by human MSLN R-plex kit (mesoscale diagnosis). ^*P < .001.

Differential internalization of hRO4 and h15B6 antibodies across tumor cell lines

To evaluate the internalization efficiency of hRO4 and h15B6 antibodies, we performed quantitative fluorescence-based uptake assays in OVCAR8, KLM1, and T3M4 cell lines using labeled antibodies and measured signal intensities at 4 and 22 h post-incubation (Fig. S5).

In OVCAR8 cells, the hRO4 antibody demonstrated robust internalization, reaching a fluorescence signal of 5356 at 4 h, whereas h15B6 exhibited significantly lower uptake with a signal of 1191, resulting in a 4.5-fold difference between the two antibodies. Internalization of hRO4 plateaued at 4 h, while h15B6 internalization continued to increase up to 22 h.

In KLM1 cells, hRO4 showed a fluorescence signal of 1404 at 4 h, in contrast to 153 for h15B6, representing a 9.2-fold difference in uptake. Similar to OVCAR8 cells, hRO4 internalization was saturated at 4 h, while h15B6 internalization persisted up to 22 h. When comparing cell lines, hRO4 uptake in OVCAR8 cells was 3.6-fold higher than in KLM1 cells, and h15B6 uptake was 7.8-fold higher in OVCAR8 than in KLM1.

In T3M4 cells, the disparity in uptake was even more pronounced. At 4 h, hRO4 showed a signal of 300, while h15B6 was minimally internalized, with a signal of only 12.6, resulting in a 24-fold difference. Both antibodies continued to be internalized up to 22 h; however, the h15B6 signal increased only slightly to 20 at 22 h, which was comparable to the negative control antibody MCP11 (18.6), indicating negligible internalization of h15B6 in T3M4 cells.

These results highlight significant differences in internalization kinetics and magnitude between hRO4 and h15B6 antibodies, as well as variability among tumor cell lines, which may have important implications for antibody-based therapeutic strategies targeting mesothelin.

Inhibition of MSLN shedding

Using five different cell lines known to shed varying amounts of MSLN, we evaluated the ability of mAb RO4 to inhibit MSLN shedding, as this antibody targets the region where shedding occurs. As illustrated in Fig. 3G, MSLN shedding was significantly inhibited in all cell lines by both mAbs hRO4 and h15B6. The extent of shedding inhibition varied among the cell lines when treated with mAb hRO4 versus mAb h15B6. Using RH16 cells, mAb hRO4 resulted in a 54% reduction in MSLN shedding, whereas mAb h15B6 resulted in a 25% reduction (P = .0001). In RH29 cells, mAb hRO4 led to an 80% reduction compared to a 48% reduction with mAb h15B6 (P = .0001). In OVCAR8 cells, mAb hRO4 reduced MSLN shedding by 92%, while mAb h15B6 achieved an 86% reduction (P = .0001). In KLM1 cells, the reductions were 74% for mAb hRO4 and 61% for mAb h15B6 (P = .001). By contrast, in H226 cells, mAb hRO4 and mAb h15B6 resulted in reductions of 23% and 14%, respectively, with no statistically significant difference observed (P = .154).

Specific cytotoxicity of hRO4 CAR T cells against MSLN+ tumor cells

To ascertain the therapeutic potential of mAb hRO4, we engineered CAR T cells incorporating the single-chain antibody variable fragment (scFv) of hRO4, h15B6, or SS1 and included the hinge and transmembrane domain from a CD8 molecule along with a 4-1BB endo-domain [17, 25, 26, 28] (Fig. S6). The epitopes of RO4 and 15B6 mAbs are located within the C-terminal region of MSLN (residues 584–598). In contrast, SS1 mAb binds shed MSLN, indicating that its epitope is not located within the C-terminal region [4]. Additionally, a truncated human epidermal growth factor receptor was included in the vector for tracking the CAR T cells (Fig. S6). We evaluated the cytotoxic activities of these CAR T cells against several human cancer cell lines known to express MSLN. The PBMC derived from four healthy donors was used in our study. The transduction efficiency of activated hRO4, h15B6, and SS1 CAR T cells was 36.5%, 53.5%, and 50.4%, respectively (Fig. S7). Although variability in transduction efficiency was observed among hRO4, h15B6, and SS1 CAR T cells, these differences do not account for the functional disparities observed in cytotoxicity assays. All functional comparisons were conducted after normalizing CAR expression levels (based on EGFR+ cells). Thus, the observed differences in cytotoxic activity reflect intrinsic functional properties of each CAR construct rather than variations in transduction efficiency.

The cytotoxicity assays involved co-culturing hRO4 CAR T cells with cancer cells at various E:T ratios for 24 h in fresh culture medium, reducing potential interference from SM which builds up in the medium over time. The cell lines used were OVCAR8 (ovarian cancer), KLM1 and T3M4 (pancreatic cancer), RH29 (malignant mesothelioma), and SW48 (colon cancer). Ovarian Cancer (OVCAR8): At E:T ratios of 1:1 and 3:1, hRO4 CAR T cells demonstrated cytotoxicities of 56.6% and 89.2%, respectively, compared to 50.9% and 84.5% for h15B6, and 30.7% and 72.2% for SS1. While no significant difference was observed between hRO4 and h15B6 (P = .06 and P = .19), both showed a statistically significant advantage over SS1 CAR T cells (P < .05; Fig. 4A, Fig. S8A). Pancreatic Cancer (KLM1): At an E:T ratio of 1:1, hRO4 CAR T cells showed higher cytotoxicity (35.2%) than h15B6 (19.2%) and SS1 (27.5%), though differences were not statistically significant. At a 3:1 ratio, hRO4 efficacy increased to 84.9%, significantly exceeding SS1 (70.2%, P = .03), but remained comparable to h15B6 (78.0%, P = .17) (Fig. 4B, Fig. S8B). Malignant Mesothelioma (RH29): At an E:T ratio of 1:1, hRO4 CAR T cells showed higher cytotoxicity (86%) than h15B6 (67%) and SS1 (78%), though the differences were not statistically significant. At a 3:1 ratio, all three CAR T cells—hRO4, h15B6 (91.8%), and SS1 (81.6%)—achieved complete cell killing (100%), with no significant differences observed (Fig. 4C, Fig. S8C). Colon Cancer (SW48): At an E:T ratio of 1:1, hRO4 CAR T cells exhibited greater cytotoxicity (25%) compared to h15B6 (0.66%) and SS1 (43%), with a statistically significant difference between hRO4 and h15B6 (P = .001). At a 3:1 ratio, hRO4 CAR T (69%) also significantly outperformed h15B6 (13%) and SS1 (77%), with the difference between hRO4 and h15B6 remaining significant (P = .001) (Fig. 4D, Fig. S8D). Pancreatic Cancer (T3M4): At an E:T ratio of 1:1, hRO4 CAR T cells showed significantly higher cytotoxicity (60.7%) compared to h15B6 (25%) and SS1 (74.7%), with a 2.4-fold increase over h15B6 (P = .0006). At a 3:1 ratio, hRO4 (86.7%) also significantly outperformed h15B6 (52.3%), showing a 1.7-fold difference (P = .0012), while SS1 achieved 97.7% (Fig. 4E, Fig. S8E). MSLN-Negative Epidermoid Carcinoma Cells (A431): Minimal cell killing was observed, validating the specificity of the CAR T cells (Fig. 4F, Fig. S8F). The enhanced cytotoxicity of hRO4 CAR T cells across multiple cancer cell lines, compared to h15B6 CAR T cells, underscores their potential as a more universally applicable therapeutic option for the treatment of a broad spectrum of MSLN-positive cancers.

Figure 4.

The activities of CAR T cells. Cells were cocultured with hRO4 (blue), h15B6 (red), SS1 (black), and MOC (x) CAR-transduced T cells at the indicated E:T ratio for 20 h, remaining tumor cells were lysed using cell culture lysis 5× reagent (E1531; Promega) for 15 min. Luciferase activity, indicative of the number of surviving target cells, was then measured in the lysates. This measurement was conducted using luciferase assay reagent (E1500; Promega) on a VICTOR X5. Results were analyzed as percent killing based on luciferase activity in wells with tumor cells alone: % killing = 100 − (relative light units [RLU] from wells with effector and target cells)/(average RLU from wells with target cells) × 100. (A) OVCAR8, (B) KLM1, (C) RH29, (D) SW48, (E) T3M4, (F) A431. Error bars are included in the graph; however, they are not visible as they are smaller than the dimensions of the symbols used.

In vivo evaluation of antitumor efficacy of hRO4 CAR T cells

hRO4 CAR T cells exhibited cytolytic activity against several MSLN-expressing cells, prompting further investigation using a human MSLN-expressing cancer cell mouse model. In these mouse experiments, OVCAR8 cells were injected intraperitoneally into NSG mice, as depicted in Fig. 5A. Following confirmation of tumor engraftment, 5 million of SS1-CAR T, h15B6 CAR T, and hRO4 CAR T cells were administered intravenously on Day 8. In the group receiving SS1-CAR T cells, which targets an epitope on SM, no tumor regression was observed; however, in the groups treated with h15B6 and hRO4 CAR T cells, the tumor signals rapidly diminished. By Day 22, all treated tumors were eliminated, with no recurrence observed at Day 28 of the experiment. Next, human pancreatic cancer cell line KLM1 was implanted subcutaneously in NSG mice (Fig. 5B). Once the tumor volume exceeded 100 mm³, both h15B6 and hRO4 CAR T cells (5 million cells each) were administered intravenously. In the hRO4 CAR T group, rapid tumor regression was noted, with complete tumor regression by Day 30, persisting until Day 43. In the h15B6 CAR T group, all but one mouse exhibited complete tumor regression. There is no statistical difference in response.

Figure 5.

The antitumor activities in vivo. (A) 5 Million of OVCAR8-luc cells were injected into mice peritoneally. On Day 8, 5 million mock, SS1, h15B6, or hRO4 CAR T cells were injected intraperitoneally. Ventral images of mice were measured every week by bioluminescent imaging (n = 5 per group). Right color bar indicates quantitation of bioluminescence. Maximum scale is 3.0e⁸ (p/sec/cm²/sr). (B) 5 Million of KLM1-luc cells were injected into mice subcutaneously. On Day 8, 5 million of mock (open circles), h15B6 (closed circles), or hRO4 CAR T cells (closed squares) were injected intraperitoneally. Tumor size was measured twice a week by calipers. Experiments were conducted three times, and a representative result is shown. Arrow indicates CAR T injections. Values represent mean ± SEM.

Shed MSLN levels in patient’s ascites and culture supernatants of human mesothelioma cells

Clinical ascites samples from six mesothelioma patients obtained from patients at the NIH Clinical Center showed a range of SM concentrations from 87 to 2851 ng/ml (average 977 ng/ml). The culture media from human mesothelioma cells displayed SM concentrations ranging from 1088 to 15104 ng/ml (average 6118 ng/ml) (Table S3).

Shed MSLN inhibit CAR T-cell function

To estimate whether SM can inhibit CAR T-cell function, we added recombinant MSLN with a C-terminal deletion at residue 585 or residue 591 to RH29 and KLM1 cells, which mimics SM found in cancer patients [8]. We then determined whether the cytotoxic activity of CAR T cells containing hRO4 Fv, h15B6 Fv, or SS1 Fv was blocked. Figure 6A shows that the 585 MSLN deletion blocks the activity of SS1 CAR T cells on RH29 cells but does not block the activity of h15B6 and/or hRO4 CAR T cells. As shown in Fig. 6A, deletion of MSLN at position 585 significantly reduced the cytotoxic activity of SS1 CAR T cells against RH29 cells, with cell killing decreasing from 95.8% to 37.5% (P = .0286), whereas the cytotoxic activities of h15B6 and hRO4 CAR T cells were not affected (P = .486 and P = .629, respectively). A similar trend was observed with the deletion at position 591, where SS1 CAR T-cell-mediated cytotoxicity against RH29 cells was again significantly impaired (P = .029), while the activities of h15B6 and hRO4 CAR T cells remained unaffected (P = .200 and P = .114, respectively), as shown in Fig. 6B. These findings were further validated using KLM1 cells, as shown in Fig. 6C, where the 585 deletion impaired the activity of SS1 CAR T cells (P = .029), but had no significant effect on h15B6 or hRO4 CAR T cells (P = .771 and P = .999, respectively). Consistently, Fig. 6D demonstrates that deletion at position 591 also inhibited SS1 CAR T-cell-mediated killing of KLM1 cells (P = .029), while the cytotoxic activities of h15B6 and hRO4 CAR T cells were not affected (P = .400 and P = .999, respectively).

Figure 6.

Truncated MSLN (296–585) or MSLN (296–591) block the activity of SS1 CAR T (black lines), but not h15B6 (red lines) and/or hRO4 CAR T (blue lines). RH29 cells (A, B), KLM1 cells (C, D) were incubated with SS1, h15B6, or hRO4 CAR T cells (E:T ratio was 5:1). MSLN (296–585) (A, C) and MSLN (296–591) (B, D) were added as a competitor. After 20 h, cell killing was measured by a luminescent-based cytolytic assay. The concentration on the horizontal axis shows the final concentration of truncated MSLN present in each well. ^*P < .05, compared to conditions without truncated MSLN.

Discussion

In this study, we developed and characterized a novel high-affinity anti-MSLN antibody, RO4, using rabbit immunization against a C-terminal peptide that is absent from SM. Structural analysis (Fig. 2A; Supplementary Table S1) revealed that in the RO4/peptide complex, the peptide adopts a conformation that is very different from the one observed in the 15B6/peptide complex. A part of the peptide epitope adopts a two-turn helix when interacting with the RO4, using residues that are not contacted by the previously reported 15B6 antibody, such as L589 and S598. The broader epitope contact and formation of significantly greater number of H-bonds likely underlie the higher binding affinity observed for RO4 in comparison to 15B6 (Figs. 2B–G) and distinguishes it mechanistically from antibodies that may be affected by SM interference.

Following humanization, hRO4 maintained high binding affinity and specificity, as demonstrated by flow cytometry analyses showing stronger binding to MSLN-positive cancer cell lines such as T3M4 and SW48 (Figs. 3D–E). This was especially evident in tumor models expressing lower or heterogeneous levels of MSLN, where hRO4 CAR T cells outperformed h15B6 CAR T cells in cytotoxicity assays (Figs. 4D–E; Supplementary Figs. S8D–E). Notably, hRO4 CAR T cells showed a statistically significant advantage over h15B6 in SW48 and T3M4 models, with up to 2.4-fold higher cell killing at lower E:T ratios.

The resilience of hRO4 CAR T cells to inhibition by SM was further confirmed using recombinant proteins ending at amino acids 585 and 591, mimicking naturally shed variants. These truncated proteins did not inhibit the function of hRO4 or h15B6 CAR T cells, while significantly impairing SS1 CAR T cells that target the N-terminal domain (Figs. 6A–D), reaffirming the advantage of targeting non-shed epitopes.

In vivo, both h15B6 CAR-T and hRO4 CAR-T cells achieved complete tumor regression in the OVCAR8 intraperitoneal model (Fig. 5A) and the KLM1 subcutaneous pancreatic tumor model (Fig. 5B). In OVCAR8 and KLM1 cells, the binding capacity of h15B6 and hRO4 antibodies to the cell surface was comparable (Fig. S4), whereas the uptake of h15B6 was lower than that of hRO4 (Fig. S5). Nevertheless, no significant differences in cytotoxicity were observed between h15B6 CAR T cells and hRO4 CAR T cells in these models (Fig. 4). These findings suggest that in OVCAR8 and KLM1 cells, CAR T-cell cytotoxicity may be influenced by antigen-binding capacity, but not by internalization efficiency. In contrast, in SW48 and T3M4 cell models, h15B6 exhibited both lower binding affinity (Fig. S4) and reduced uptake (Fig. S5) compared to hRO4. Consistently, hRO4 CAR T cells demonstrated higher cytotoxic activity than h15B6 CAR T cells. These results indicate that, in these cell lines as well, CAR T-cell cytotoxicity may depend on antigen-binding ability, whereas antibody internalization does not appear to play a significant role.

RO4 is a MSLN-specific monoclonal antibody, and its specificity was evaluated by immunohistochemistry using a human tissue microarray (Fig. 1 and Fig. S2). In normal tissues such as the stomach, kidney, and uterus, immunoreactivity was predominantly observed in the cytoplasm, suggesting intracellular localization rather than functional surface expression. Given that CAR T-cell activation requires sufficient levels of surface antigen, the actual risk of off-tumor toxicity appears to be limited. Moreover, the integration of affinity-tuned receptors and dual-recognition circuits offers additional safeguards to mitigate the potential for normal tissue reactivity [31–33].

Taken together, our data demonstrate that the high-affinity antibody hRO4 confers both mechanistic and functional advantages in MSLN-targeted immunotherapy. hRO4 binds with high affinity to a non-shed epitope, is resistant to interference from soluble antigen, and outperforms lower-affinity antibodies in specific tumor contexts, suggesting broad therapeutic utility [29, 30]. Various antibody-based therapeutic formats—such as bispecific antibodies, antibody–drug conjugates (ADCs), and immunotoxins—have been developed for cancer treatment. However, a common limitation among these modalities is the reduced therapeutic efficacy caused by the decoy effect of SM, which can sequester circulating antibodies. Therefore, antibodies like RO4, which selectively bind to membrane-bound mesothelin without recognizing SM, are expected to provide improved therapeutic benefit by circumventing this decoy mechanism. Moreover, hRO4 demonstrates broader binding across diverse MSLN-positive tumor cell types compared to h15B6 (Fig. 3), suggesting its potential utility in treating a wider patient population.

Future investigations should focus on validating the clinical potential of hRO4 in tumor types characterized by high levels of MSLN shedding, as well as optimizing therapeutic designs for tumors with low or heterogeneous antigen expression.

Author contributions

Masanori Onda (Conceptualization [equal], Data curation [equal], Formal analysis [equal], Investigation [equal], Methodology [equal], Software [equal], Writing—original draft [equal], Writing—review & editing [equal]), Xiufen Liu (Resources [equal]), Wenlong Liu (Data curation [equal]), Jingyu Zhan (Data curation [equal], Formal analysis [equal], Methodology [equal]), Carolyn A. Maslanka (Data curation [equal], Formal analysis [equal], Methodology [equal]), Di Xia (Data curation [equal], Formal analysis [equal], Methodology [equal]), Mitchell Ho (Methodology [equal]), and Ira Pastan (Conceptualization [equal], Supervision [equal], Writing—original draft [equal])

Conflict of interest

M.O., X.L., M.H., and I.P. are inventors on patents on RO4 antibodies and have assigned all rights to the NIH. M.H. holds the position of Editor-in-Chief for Antibody Therapeutics and is blinded from reviewing or making decisions for the manuscript.

Funding

This research was supported by the Intramural Research Program of the National Institutes of Health (NIH). The contributions of the NIH authors were made as part of their official duties as NIH federal employees, are in compliance with agency policy requirements, and are considered Works of the United States Government. However, the findings and conclusions presented in this paper are those of the authors and do not necessarily reflect the views of the NIH or the U.S. Department of Health and Human Services.

Data availability

The data generated in this study are available from the corresponding author upon request. The atomic coordinates for the structure of RO4 in complex with C-terminal peptide of MSLN have been deposited in the Protein Data Bank (www.pdb.org) under accession code 9P4C (https://doi.org/10.2210/pdb9P4C/pdb).

Ethics and consent statement

Not required.

Animal research statement

All procedures involving animals were conducted under a protocol approved by the Institutional Animal Care and Use Committee (IACUC) at the NIH, ensuring adherence to federal and institutional guidelines for animal welfare.