Identification of CADM1 as an Immunotherapeutic Target and Evaluation of a Novel CADM1-Targeting Antibody-Drug Conjugate in Preclinical Osteosarcoma Models

Abstract

Due to the paucity of validated cell surface osteosarcoma-specific targets, patients with this condition have long been excluded from the benefits of antibody-drug conjugate (ADC) therapy observed in patients with several solid and hematologic malignancies. Our comprehensive surfaceome profiling approach previously identified osteosarcoma-specific cell-surface antigens that are highly expressed in osteosarcomas but minimally expressed in normal tissues. As a result, one such antigen, CADM1, was selected for the generation of an ADC. We tested a CADM1-targeting ADC with a tesirine payload (SG3249) in vitro in osteosarcoma, rhabdomyosarcoma, and neuroblastoma patient-derived xenograft cell lines. In vivo, we tested six CADM1-expressing osteosarcoma patient-derived xenograft models. The CADM1 ADC demonstrated significant antitumor activity in vitro across the osteosarcoma, rhabdomyosarcoma, and neuroblastoma cell lines. Additionally, it effectively reduced tumor volume and extended event-free survival in all six osteosarcoma PDX models tested. Notably, the CADM1 ADC achieved a major complete response in one model (OS2), complete responses in two models (OS1 and OS33), and partial responses in three models (OS9, OS17, and OS31). Based on these results, clinical development of CADM1-targeted therapies for osteosarcoma and other CADM1-expressing pediatric solid tumors may be warranted.

Graphical Abstract

Introduction

Osteosarcoma is the most common primary bone cancer in children and young adults. The long-term survival rate for patients with localized disease has been 60–70% since the implementation of adjuvant chemotherapy in the 1980s (1). However, improvement in survival has been limited over the past 4 decades. Although increased understanding of the biology of osteosarcoma has revealed its heterogeneity and molecular aberrations, few recurrent targetable alterations have been identified (2). Therefore, novel therapeutic strategies are urgently needed.

Immunotherapies targeting cell-surface antigens with antibody-drug conjugates (ADCs) have exhibited promising activity in hematologic malignancies and solid tumors (3–5). Several cell-surface proteins are known to be expressed in osteosarcomas, such as HER2, GPNMB, and IGF1R (6–8). However, the results of osteosarcoma treatment with monoclonal antibodies and ADCs in clinical trials have been disappointing (9–12), possibly due to low expression of surface antigens or a lack of antitumor activity of the payload drug conjugates. The paucity of known osteosarcoma-specific cell-surface antigens has been an obstacle to the development of additional ADCs.

Previously, we developed a high-throughput approach that integrates proteomic and transcriptomic profiling data from osteosarcoma cell lines, patient-derived xenografts (PDXs), and patient tumor samples. This approach identified high-confidence osteosarcoma cell-surface antigens, such as LRRC15, MMP14, MRC2, and CD276. ADCs targeting these proteins exhibited encouraging antitumor activity in preclinical testing (13–16).

CADM1, which we identified as being highly expressed in osteosarcomas in our surfaceome profiling, is a cell surface glycoprotein that plays a crucial role in various biological processes, including cell adhesion, cell recognition, immune response, and tumor suppression. However, CADM1-targeted therapy for osteosarcoma is lacking. We hypothesized that a CADM1-targeted ADC could be a new effective treatment option for osteosarcoma. Hence, we developed a CADM1-ADC with a tesirine payload (SG3249). Tesirine is a DNA-damaging agent that induces both intrastrand and interstrand cross-linking. We tested the CADM1-ADC against osteosarcoma cell lines in vitro and in osteosarcoma PDX models.

Materials and Methods

Cell culture

The standard osteosarcoma cell lines HOS (cat. #CRL-1543; RRID: CVCL_0312) and U2OS (cat. #HTB-96; RRID: CVCL_0042) and the neuroblastoma cell lines IMR-32 (cat. #CCL-127; RRID: CVCL_0346) and BE (2)-C (cat. #CRL-2268; RRID: CVCL_V007) were purchased from the American Type Culture Collection. The rhabdomyosarcoma cell line RH30 (cat. #CSC-C0500; RRID: CVCL_0041) was purchased from Creative Bioarray. Additionally, we tested seven PDX-derived osteosarcoma cell lines (OS17, OS31, OS33, OS36, OS39R, OS42, and OS43), two patient-derived osteosarcoma cell lines (OS252 and MDA-SA174), one Ewing sarcoma cell line (MDA-SA140), and one leukemia cell line (K562 RRID: CVCL_K562, negative control; kindly provided by Dr. Joya Chandra, The University of Texas MD Anderson Cancer Center). All cell lines were employed in flow cytometry, cytotoxicity, and internalization assays. For RNA sequencing (RNA-seq) analysis, we used 13 additional patient-derived osteosarcoma cell lines (OS242, OS300, OS308, OS337, OS340, OS342, OS354, OS365, OS366, OS368, OS377, OS379, and OS396). All PDX and patient-derived osteosarcoma cell lines were established in our laboratory from 2006 to 2021 as described previously (17,18), cultured in Dulbecco’s modified Eagle’s medium (cat. #30–2002; American Type Culture Collection) with 10% fetal bovine serum (cat. #Mt35011cv; Thermo Fisher Scientific) and maintained in a humidified incubator at 37°C with 5% CO₂. All cell lines were authenticated by short tandem repeat (STR) profiling and routinely tested for Mycoplasma contamination using MycoAlert (Lonza, cat. #LT07–118) according to the manufacturer’s instructions and found to be negative. Mycoplasma testing is performed monthly, and the most recent test was completed in January 2024. All experiments were performed on cells passaged no more than 15 times from thawing.

Flow cytometric analysis

Seven osteosarcoma (HOS, U2OS, OS252, OS17, OS31, OS33, and OS39R), one rhabdomyosarcoma (RH30), two neuroblastoma (IMR-32 and BE [2]-C), and one Ewing sarcoma (MDA-SA140) cell line were analyzed for surface CADM1 expression. Samples of 1 × 10⁶ cells each were washed twice with phosphate-buffered saline (PBS) and resuspended in PBS with 1% bovine serum albumin. To assess CADM1 expression, cells were incubated with a primary antibody against CADM1 (cat. #TAB-1448 CL; Creative Biolabs) for 30 minutes on ice. Following incubation, cells were washed twice with PBS containing 1% bovine serum albumin. A secondary antibody, APC-conjugated goat anti-mouse IgG (cat. #115–136-146; Jackson ImmunoResearch Laboratories; RRID: AB_2338651), was incubated with the cells for an additional 30 minutes on ice while protected from light. Flow cytometry was performed using a FACS Fortessa instrument (BD Biosciences) and analyzed using FlowJo software (RRID: SCR_008520). Unstained cells from among the respective cell types were used as controls to set voltages for forward scatter, side scatter, and fluorescence.

Cytotoxicity assay

Eight osteosarcoma cell lines (OS17, OS31, OS36, OS39R, OS42, OS43, OS252, and MDA-SA174) were evaluated for payload sensitivity using nine commercially available cytotoxic agents: SG3199 (cat. #HY-101161), PNU-159682 (cat. #HY-16700), nemorubicin (cat. #HY-15794), calicheamicin (cat. #HY-19609), DM4 (cat. #HY-12454), dolastatin 10 (cat. #HY-15580), monomethyl auristatin E (cat. #HY-15162), taltobulin (cat. #HY-15584), and mertansine (cat. #HY-19792). Cells were seeded in 96-well plates at 5,000 cells/well and treated for 5 days across the following concentration ranges (respectively): 5.0–1 × 10⁻⁶ μmol/L, 78.1–4 × 10⁻⁵ μmol/L, 390–2× 10⁻⁴ μmol/L, 120–6.14 × 10⁻⁵ nmol/L, 1.28–0.66 mmol/L, 1.6–1.6 × 10⁻⁹ mmol/L, 0.39–2 × 10⁻⁷ mmol/L, 0.39–2 × 10⁻⁷ mmol/L, and 1.35–7 × 10⁻⁷ mmol/L. Cell viability was determined using the alamarBlue assay and IncuCyte live-cell imaging system (Essen BioScience; RRID: SCR_019874) according to the manufacturer’s protocol. For CADM1-ADC antitumor activity, HOS, SaOS2, OS17, OS31, and K562 cells were seeded in 96-well plates at 5,000 cells/well and treated with CADM1-ADC in a 5- or 10-fold serial dilution series (5000 nmol/L to 0.0001 nmol/L). Plates were placed in an IncuCyte live-cell imaging system with images acquired every 3 hours using phase contrast and fluorescence channels. Imaging data were analyzed using Prism 10 software (GraphPad Software; RRID: SCR_002798), and viability was assessed based on confluence and fluorescence intensity.

Internalization and bystander assay

Osteosarcoma cells were seeded in 96-well plates at a density of 5,000 cells/well. Cells were allowed to grow until they reached about 70–80% confluence before initiation of the internalization assay. The CADM1-ADC and a human anti-CADM1 antibody (clone PTA021_A1; Creative Biolabs), both at 5 μg/mL, were mixed with IncuCyte Fabfluor-pH Dye (cat. #4722; Sartorius) in separate tubes at a molar ratio of 1:3 in cell culture media, resulting in a 2x final assay concentration. The mixture was incubated at 37°C for 15 minutes to facilitate conjugation. Following incubation, 50 μL of the Fabfluor-antibody mix was added to each well of the cell plates. The plates were then placed in the IncuCyte live-cell analysis system. Images were captured every 30 minutes using a 10x objective, and the appropriate fluorescent module was configured for 5 days. Imaging data were collected using the integrated software with the IncuCyte system. Internalization of the ADC-dye complex was assessed based on fluorescence intensity and localization within the cells. A human IgG1-ADC (WBPX1222–240321003; WuXi Biologics) and an anti-IgG1 monoclonal antibody (cat. #SAB4700520; Sigma; RRID: AB_10897971) were used as isotype controls.

The CADM1-positive cell line HOS and the CADM1-negative cell line K562 were selected for the bystander assay. HOS and K562 cells were incubated separately with the CADM1-ADC; the same was done with SG3199, the free payload of the ADC (cat. #HY-128952; MedChemExpress), which served as a positive control. CADM1-ADC and free payload concentrations of 0.07 nmol/L and 0.01 nmol/L were used for the incubation. After 5 days of culturing at 37°C, the conditioned culture medium was transferred to CADM1-negative K562 cells in the plates, and the plates were placed in the IncuCyte live-cell analysis system for 60 hours. Images were captured every 30 minutes using a 10x objective, the imaging data were collected, and the proliferation of K562 was assessed using the integrated software with the IncuCyte system.

RNA-seq analysis of osteosarcoma cell lines

Gene expression profiling was performed with 17 osteosarcoma cell lines using 150-bp paired-end RNA-seq, achieving an average coverage of 26.5 million reads per sample (range, 23.3–29.4 million reads). Read quality was assessed using FastQC (version 0.11.8; RRID: SCR_014583). Subsequently, reads were aligned to the human genome (GRCh38.p12, NCBI Assembly Accession ID GCF_000001405.38) using TopHat2 (version 2.1.1, RRID: SCR_013035). Aligned reads were quantified for each gene using HTSeq software (version 0.11.0; RRID: SCR_005514) with gene annotations from GENCODE version 29 (GRCh38.p12; RRID: SCR_014966). RNA-seq expression data for 101 osteosarcoma patient samples were obtained from the Therapeutically Applicable Research to Generate Effective Treatments (TARGET; RRID: SCR_014514) data matrix (via the Center for Cancer Genomics website [https://ocg.cancer.gov/programs/target/data-matrix; accessed May 29, 2019]). Comprehensive clinical information was available for 98 of the samples. Additional RNA-seq expression data for 2,007 samples across 16 pediatric cancer types were retrieved from the St. Jude PeCan data portal (https://pecan.stjude.cloud; accessed April 7, 2020). RNA-seq expression data for normal tissues were obtained from the Genotype-Tissue Expression Portal (V8, https://www.gtexportal.org; accessed August 26, 2019; RRID: SCR_013042). This dataset includes RNA-seq profiles for 29 different tissue types, totaling 16,695 samples.

Immunohistochemical staining and H-scores

Tissue microarrays were assembled, consisting of 13 normal human tissue arrays (Table 1), 70 human osteosarcoma arrays, and 19 PDX arrays with samples representing various cancer histologies. The immunohistochemical staining was performed using a polyclonal anti-CADM1 antibody (cat. #s4945; Sigma-Aldrich; RRID: AB_532287) at a dilution of 1:10,000. The tissue sectioning process was constructed as described previously (13). Quantitative evaluation of the immunohistochemical staining was carried out by a pathologist (R.L.) through direct microscopic examination. Staining intensity and distribution were assessed in tumor cells, PDX samples, and normal tissues. Staining results were quantified as H-scores, which were calculated by multiplying the staining intensity by the percentage of positively stained cells, with H-scores ranging from 0 to 300. Additionally, cytoplasmic staining patterns were examined. Representative images of the immunohistochemical staining on scanned slides were obtained using Halo software (version 3.1.1; Indica Labs; RRID: SCR_008442).

Table 1.

Assessment of CADM1 expression in normal tissues using immunohistochemistry: comparative analysis of staining intensity, proportion of membrane-positive cells, and derived H-scores.

Normal tissue	Membranous intensity	Membranous %	Membranous H-score
CARDIAC MUSCLE	0	0	0
CEREBELLUM	0	0	0
COLON	2	50	100
KIDNEY	2	10	20
LIVER	1	10	10
LUNG	3	20	60
OVARY	0	0	0
PANCREAS	3	5	15
SKELETAL MUSCLE	0	0	0
STOMACH	2	60	120
TESTIS	2	90	180
TONSIL	2	5	10
SKIN	0	0	0

CADM1-ADC generation

An anti-CADM1 (clone PTA021_A1) monoclonal antibody was manufactured as described for U.S. Patent US20120093826A1(19). The antibody was diluted to a final concentration of 5 mg/mL. Tesirine (SG3249) (CAS. #1595275–62-9; cat. #ADC-S-031; Creative Biolabs) is a pyrrolobenzodiazepine dimer payload with a protease-cleavable valine-alanine linker, and SG3199 is the released warhead of SG3249. A 10-mmol/L solution of tris (2-carboxyethyl) phosphine was added to the antibody solution to facilitate reduction. The reduction mixture was incubated at 33°C for 2 hours. After the reduction step, the mixture was allowed to cool to room temperature. SG3249 dissolved in dimethyl sulfoxide was then added to the antibody solution for conjugation. The conjugation reaction was allowed to proceed for 1 hour at room temperature. The reaction was quenched by the addition of N-acetyl cysteine (1 μmol/L, 100 μmol/L of a 10-mmol/L solution). To remove unreacted SG3249 and exchange the buffer, tangential flow filtration was performed, and the buffer was changed to PBS. The CADM1-ADC concentration was determined by correcting A280 using the SG3249 A280/A330 ratio. Analytical characterization included size-exclusion chromatography (SEC; Tosoh TSKgel SuperSW mAb column), revealing 20.6% high molecular weight species, and hydrophobic interaction chromatography (HIC; TSKgel Butyl-NPR column) using a salt gradient. The weighted average of drug-to-antibody ratio (DAR) was 3.76, with 7.35% unconjugated antibody. ADCs are modular therapeutics comprising an antigen-specific monoclonal antibody covalently linked to a potent payload. In this study, our ADC specifically used an IgG1 backbone and a cleavable linker. This architecture enables tumor-targeted binding, internalization, and lysosomal processing to release the payload and kill cells. Membrane-permeable payloads can mediate a bystander effect, and some ADCs can trigger immunogenic cell death, both of which may augment antitumor activity. Depending on the isotype and Fc engineering, Fc-mediated effector functions may contribute to cell death, although they were not examined in this study (20).

PDX testing

Osteosarcoma PDX models were obtained through the Pediatric Preclinical In Vivo Testing consortium as described previously (17,18). This is a National Cancer Institute–funded initiative aimed at evaluating novel agents directed against pediatric solid tumors and leukemia preclinical models. Detailed characteristics of these models can be accessed via multiple sources, including the PedcBioPortal (https://pedcbioportal.kidsfirstdrc.org/study/summary?id=pptc). For this study, six osteosarcoma PDX models (OS1, OS2, OS9, OS17, OS31, and OS33) were used for in vivo testing. Female CB17SC scid^−/− mice (4–6 weeks old; Taconic Farms; RRID: IMSR_TAC: CB17SC) were used to carry the subcutaneous flank xenografts. All mice were housed under barrier conditions, and experimental procedures were performed according to protocols approved by the Institutional Animal Care and Use Committee at MD Anderson. Once tumor volumes reached about 150 mm³, mice were randomly assigned to three groups (treatment, vehicle, and negative control) with 10 mice per group. Treatments were administered intraperitoneally as single doses on day 0. The treatment group received 3 mg/kg CADM1-ADC in PBS (21), whereas the control group received PBS only. As a negative control, 3 mg/kg CD19-ADC in PBS was administered to the control cohort. Also, a CADM1-negative Ewing sarcoma PDX model (MDA-SA140) was treated with 3 mg/kg CADM1-targeting ADC. Mice were monitored weekly for changes in body weight and tumor volume, which were recorded to assess treatment efficacy and overall health.

Statistical methods

A tumor event was defined as a quadrupling of tumor volume from the day treatment began as measured in two diameters in millimeters using calipers with the formula $([\text{diameter}1/2+\text{diameter}2/2]^3*0.5236)/1000$ (17). The exact time of each event was estimated by interpolating between measurements before and after the event, assuming log-linear growth. Differences in event-free survival (EFS) among the experimental groups (n > 2) were assessed using the Gehan-Breslow-Wilcoxon test. Additionally, differences in minimum relative tumor volume (the ratio of current tumor volume to volume at treatment initiation) were evaluated using the Wilcoxon rank-sum test. The P-value threshold for significance for the statistical methods was set at <0.05. The response categories consisted of progressive disease (PD), which was further categorized as PD without growth delay (PD1) or PD with growth delay (PD2) in treated mice. The remaining categories were stable disease (SD); partial response (PR); complete response (CR), defined as the absence of measurable tumor mass for at least one recording; and maintained CR (MCR), characterized by the absence of a measurable tumor mass for at least three consecutive weekly recordings as described previously (Appendix: Statistical Methods). Each mouse was assigned a numerical score from 0 to 10 based on its objective response measure: PD1 = 0, PD2 = 2, SD = 4, PR = 6, CR = 8, and MCR = 10. The overall group response was determined by the median score of the evaluable mice. In cases where the median score fell halfway between two objective response measure categories, the lower response category was assigned as the objective response.

Data availability

The data generated in this study will be available from the GEO (RRID: SCR_005012) under accession number GSE309827 and will be made available upon request from the corresponding author. Noncommercially available materials and experimental protocols will be made available to not-for-profit or academic requesters upon completion of a material transfer agreement by contacting RGorlick@mdanderson.org.

Results

Proteomic analysis confirms CADM1 expression in osteosarcoma

We prepared surface protein extracts and completed proteomic analysis with patient-derived osteosarcoma cell lines and osteosarcoma PDX models as described previously (13). We used quantitative mass spectrometry to profile the surface proteins of these cells. We confirmed that CADM1 was frequently expressed in osteosarcoma cell lines and PDX models, supporting its suitability as a therapeutic target for osteosarcoma (Fig. 1A).

Figure 1.

A, Expression of CADM1 in osteosarcoma (OS) models. Left panel: heat map of mass spectrometry–based expression (Z-scores) of 881 cell surface proteins across osteosarcoma PDX models and osteosarcoma cell lines. CADM1 and additional previously reported surface targets (MRC2, MMP14, LRRC15, and CD276) are indicated. Right panel: magnified view highlighting CADM1 expression along with that of the other previously reported candidate genes. B, CADM1 was expressed at higher levels in osteosarcoma than in a range of normal tissues. The boxes represent quartiles 1 and 3 of the data. The bars represent the median values. TPM, transcripts per million. C, Expression of CADM1 and candidate surface targets in osteosarcoma. Left panel: heat map analysis of gene expression across the TARGET osteosarcoma cohort (n = 98) and an independent osteosarcoma cell line set (OSC; n = 17). Expression of 214 genes encoding surface proteins with clinical annotations, including sex, survival status, metastatic disease, and histologic response to therapy, is shown. Candidate targets, including CADM1, LRRC15, MT1-MMP, MRC2, and CD276, are highlighted. Right panel: magnified representation of these five candidates across the same cohorts. D, CADM1 expression in osteosarcomas and other pediatric cancers. ACT, adrenocortical carcinoma; AML, acute myeloid leukemia; BALL, B-cell acute lymphoblastic leukemia; CPC, choroid plexus carcinoma; EPD, ependymoma; FPKM, fragments per kilobase million; HGG, high-grade glioma; LGG, low-grade glioma; MB, medulloblastoma; MEL, melanoma; MLL, mixed-lineage leukemia; NBL, neuroblastoma; WLM, Wilms tumor; RB, retinoblastoma; RHB, rhabdomyosarcoma; TALL, T-cell acute lymphoblastic leukemia. The boxes represent quartiles 1 and 3 of the data. The bars represent the median values.

Transcriptomic analysis confidently identifies CADM1 overexpression

To validate that CADM1 can serve as a therapeutic target, we employed a transcriptomic analysis as reported previously (13). Specifically, we compared RNA-seq data for human osteosarcoma samples obtained from the TARGET project with RNA-seq data for 29 types of normal tissue from 16,695 samples in the Genotype-Tissue Expression project. Gene expression of CADM1, measured in transcripts per million (TPM), is markedly higher in osteosarcoma tissue (mean: 136.4; median: 106.3 TPM) than in normal tissues, where expression remains within single-digit TPM levels, except for thyroid, testis, pituitary, and brain tissue, in which it does not exceed 30 TPM. This differential gene expression supports the rationale for selecting CADM1 as a target for ADC development (Fig. 1B).

Furthermore, we used hierarchical clustering analysis to illustrate the expression patterns for CADM1 and the previously reported candidate cell surface protein as therapeutic targets in the osteosarcomas from the TARGET data set and in 17 osteosarcoma cell lines. This analysis profiled 214 genes (rows) across TARGET-OS patient tumors (columns, left) and an osteosarcoma cell-line panel (columns, right). Candidate and previously reported cell-surface targets were annotated for druggability; notably, CADM1 was consistently identified across datasets (Fig. 1C). Also, we found that the CADM1 expression levels in osteosarcomas were higher than in most other types of pediatric cancer (Fig. 1D) based on RNA-seq data for 2,007 patient samples in the St. Jude PeCan data portal.

Validation of CADM1 as a cell-surface antigen in osteosarcoma

We performed flow cytometry to validate the cell-surface localization and protein expression levels for CADM1 in seven osteosarcoma cell lines (HOS, U2OS, OS252, OS17, OS31, OS33, and OS39R). All cell lines exhibited target protein expression, although the levels varied (Fig. 2A).

Figure 2.

Validation of CADM1 expression in osteosarcoma. A, Flow cytometric analysis of seven osteosarcoma cell lines. The gray histograms denote unstained controls, and the colored histograms denote anti-CADM1 staining. B, Representative immunohistochemical membranous staining for CADM1 in a human osteosarcoma sample, an osteosarcoma PDX sample, and a normal tonsil sample. C, Tissue microarray H-scores (0–300) for CADM1 across 70 human osteosarcoma, 19 osteosarcoma PDX, and 13 normal human tissue samples. Each dot represents one case, and the horizontal lines indicate the median values. Data presentation and statistical analyses are described in the Materials and Methods section.

We then performed immunohistochemical staining using an osteosarcoma tissue microarray with samples from 70 patients and 19 PDX models, the results of which are shown in Fig. 2B. CADM1 was expressed in 79% of the patient samples and all the PDX models. We also conducted immunohistochemical staining with a microarray containing representative samples of 13 normal human tissues with relative staining intensity, proportion of membrane-positive cells, and derived H-scores (Table 1). We observed the following H-scores: 120 and 180 in the gastric glands and testis, respectively (moderate intensity), and ≤60 in the kidney, lung, pancreas, tonsil, and liver (low intensity). CADM1 expression was absent from the skin, ovary, cerebellum, skeletal muscle, and cardiac muscle. The H-scores for the immunohistochemically stained patient samples, PDX tissues, and normal healthy human tissue samples are summarized in Fig. 2C.

In vitro testing of ADC payloads in osteosarcoma cell lines

We tested a panel of nine commercially available payloads (SG3199, PNU-159682, nemorubicin, calicheamicin, DM4, dolastatin 10, monomethyl auristatin E, taltobulin, and mertansine) in vitro in eight osteosarcoma cell lines (OS17, OS31, OS36, OS39R, OS42, OS43, OS252, and MDA-SA174) to select a payload for ADC development in osteosarcoma treatment. The half-maximal inhibitory concentrations (IC₅₀s) indicated that SG3199, the warhead release from tesirine (SG3249), was the most potent payload against these osteosarcoma cell lines (Fig. 3). The in vitro potency observed in our osteosarcoma cell lines is consistent with established clinical treatment patterns, where DNA-damaging agents such as doxorubicin, cisplatin, and high-dose methotrexate remain the backbone of therapy. In contrast, microtubule inhibitors (including taxanes and vincristine) have demonstrated limited clinical efficacy against osteosarcoma and are not routinely used, which aligns with their weaker activity in our preclinical models (National Comprehensive Cancer Network, Osteosarcoma, Version 2.2025). Moreover, a key feature of pyrrolobenzodiazepine dimers that enables their “bystander effect” is their hydrophobic, or lipophilic, nature. This property allows a pyrrolobenzodiazepine payload to diffuse from targeted cancer cells to kill nearby untargeted cells, which may be advantageous in heterogeneous tumors such as osteosarcoma (22). In conclusion, our test results, together with the above-mentioned existing evidence, led us to select tesirine as our payload of choice.

Figure 3.

Results of in vitro testing of cytotoxic payloads in osteosarcoma cell lines. The IC₅₀ values demonstrate that SG3199, a pyrrolobenzodiazepine dimer payload, was the most potent payload against seven osteosarcoma cell lines.

Properties of the CADM1-ADC and CD19-ADC

We developed an ADC with a monoclonal anti-CADM1 antibody (clone PTA021_A1) and SG3249 and characterized it using SEC and HIC. We determined that the CADM1-ADC–weighted average of the drug-to-antibody ratio was 3.76 (Supplementary Fig. S1).

We developed a CD19-ADC using the same payload and linker as with the CADM1-ADC. The CD19-ADC–weighted average of the drug-to-antibody ratio was 4.17 (Supplementary Fig. S2). Supplementary Figure S3 illustrates the schematic structure of the CADM1 ADC. A CD19-ADC with the same payload and linker as those for the CADM1-ADC (Creative Biolabs) was used as a negative control.

Selective in vitro antitumor activity of the CADM1-ADC in osteosarcoma models

We assessed the in vitro cytotoxicity of the CADM1-ADC in four osteosarcoma cell lines (HOS, SaOS2, OS17, and OS31) and the CADM1-negative leukemia cell line K562, which was confirmed to be CADM1-negative by flow cytometry. The IC₅₀ values demonstrated that the CADM1-ADC significantly inhibited proliferation in CADM1-positive osteosarcoma cells but had less of an effect on K562 cell proliferation. SG3199, the free payload of the CADM1-ADC, exhibited no selectivity for CADM1-positive or CADM1-negative cell lines, reflecting its nonselective nature. The CD19-ADC, as an isotype control for CADM1-ADC, exhibited no significant cytotoxic effects at the concentrations tested. These in vitro results validated the antigen specificity of the CADM1-ADC (Fig. 4A).

Figure 4.

In vitro testing demonstrated selective cytotoxicity of the CADM1-ADC. A, IC₅₀ values for the CADM1-ADC, CD19-ADC, and payload SG3199 against osteosarcoma cell lines and CADM1-negative K562 cells. B, Results of flow cytometric analysis of the rhabdomyosarcoma cell line RH30, the neuroblastoma cell lines IMR-32 and BE (2)-C, and the Ewing sarcoma cell line MDA-SA140. The IC₅₀ values in the cell lines are shown. CADM1 was highly expressed in the rhabdomyosarcoma and neuroblastoma cell lines.

Selective in vitro antitumor activity of the CADM1-ADC in other pediatric solid tumors

We evaluated the antitumor activity of the CADM1-ADC in vitro using the rhabdomyosarcoma cell line RH30, two neuroblastoma cell lines (IMR-32 and BE[2]-C), and a CADM1-negative Ewing sarcoma cell line (MDA-SA140) as a negative control. Flow cytometry confirmed high CADM1 expression on the rhabdomyosarcoma and neuroblastoma cell lines. In vitro cytotoxicity assays demonstrated that the CADM1-ADC effectively reduced the viability of these CADM1-positive cell lines. In contrast, flow cytometric analysis of MDA-SA140 cells confirmed the absence of CADM1 expression, which correlated with a higher IC₅₀ value in the cytotoxicity assay (Fig. 4B).

Internalization and bystander killing effect of the CADM1-ADC

To evaluate the ADC’s internalization properties, we treated the CADM1-positive osteosarcoma cell lines HOS, OS17, and OS31 with the CADM1-ADC for 5 days at 37°C. We observed internalization of our ADC in all three cell lines, reaching a steady-state plateau starting from day 4 (Fig. 5A). As described previously (23), we validated the bystander killing effect of the CADM1-ADC. We treated HOS cells with the ADC for 5 days, after which we transferred the conditioned media onto untreated CADM1-negative K562 cells, as represented by the assay setup (Fig. 5B-D). The CADM1-ADC elicited a bystander effect as evidenced by a decrease in the proliferation of CADM1-negative K562 cells in HOS-conditioned medium. Conversely, CADM1-negative K562-conditioned medium did not induce any bystander effect or cytotoxicity when transferred onto untreated CADM1-negative K562 cells. Exposure to SG3199 inhibited the proliferation of CADM1-negative K562 cells (Fig. 5E). These findings support the known mechanism of action of ADCs, whereby the payload is released upon internalization with antigen-positive cells. Presumably, following internalization, the protease-cleavable valine–alanine linker is expected to be enzymatically cleaved in lysosomes to release the active payload, allowing the free payload to be released extracellularly and potentially kill CADM1-negative cells in the microenvironment by passive diffusion.

Figure 5.

Internalization of the CADM1-ADC and bystander effect assay. A, HOS, OS17, and OS31 osteosarcoma cells were treated with Incucyte Fabfluor-pH–labeled 5 μg/mL CADM1-ADC, an anti-CADM1 antibody, human IgG1 ADC, and a human anti-IgG1 antibody (isotype control). Significant internalization was observed in CADM1-ADC–treated HOS, OS17, and OS31 cells. B-D, Bystander effect with CADM1-ADC conditioned media. B, Schematic of the bystander assay. CADM1-negative K562 recipient cells were exposed for 5 days to conditioned media transferred from CADM1-ADC–treated HOS donor cells. C, conditioned media were derived from CADM1-ADC 5-day-treated K562 donor cells. D, Free payload–treated media were included as a positive control. E, Proliferation (60 hours) of K562 cells exposed to CADM1-ADC (0.01 or 0.07 nmol/L; HOS-derived or K562-derived), SG3199 (0.01 or 0.07 nmol/L), or drug-free medium.

In vivo antitumor activity of the CADM1-ADC in osteosarcoma PDX models

Based on a prior study using the same pyrrolobenzodiazepine payload (tesirine) in an ADC, we selected a 3-mg/kg dose of the CADM1-ADC for in vivo testing (21,23). The CADM1-ADC was generally well tolerated at this dose. After administration of a single dose, the CADM1-ADC induced significantly prolonged EFS (T-C average: 98.34 days [i.e., difference in median time-to-event for the treatment and control groups]) in all tested mice across six CADM1-expressing osteosarcoma models (OS1, OS2, OS9, OS17, OS31, and OS33; P < 0.05) (Fig. 6A). The CADM1-ADC induced MCR in one model (OS2), CR in two models (OS1 and OS33), and PR in three models (OS9, OS17, and OS31). We treated the same six osteosarcoma models with the CD19-ADC. Two models (OS9 and OS17) exhibited PD1, three models (OS1, OS2, and OS31) exhibited PD2, and one model (OS33) had a PR (Supplementary Table S1) (Fig. 6B and C). The variations in relative tumor volume indicated that the target was responsible for the differing effectiveness of the ADCs. The average maximum weight loss was 10.1% across all models tested (Fig. 6D). RNA-seq data and immunohistochemical H-scores demonstrated that OS1 had higher CADM1 expression than did OS33, OS17, OS2, OS9, and OS31 (Fig. 6E and F).

Figure 6.

Tumor volume and survival after CADM1-ADC treatment in preclinical osteosarcoma models. A, The CADM1-ADC induced significant improvement in EFS when compared with the control in all six osteosarcoma models tested (P < 0.05). B, The CADM1-ADC induced objective responses, consisting of MCR in one osteosarcoma model (OS2), CR in two models (OS1 and OS33), and PR in three models (OS9, OS17, and OS31). C, Of the CD19-ADC–treated osteosarcoma models, two had PD1 (OS9 and OS17), three had PD2 (OS1, OS2, and OS31), and one had a PR (OS33). The pale-colored lines represent individual mice, whereas the darker colored lines represent cohort median values. D, Graphs correlating mouse body weight change with EFS. Median weight loss is shown in blue-gray for treatment with the CADM1-ADC and light blue for that with the CD19-ADC. E, RNA-seq data for PDX samples. F, immunohistochemical (IHC) H-scores for PDX samples. FPKM, fragments per kilobase million. G, Graph of relative tumor volumes showing PD in the Ewing sarcoma model after CADM1-ADC exposure.

We treated the Ewing sarcoma PDX model MDA-SA140 with the same CADM1-ADC to perform a comparative analysis with our CADM1-positive models. Flow cytometry revealed a lack of CADM1 expression (Fig. 4B), and we observed PD (Fig. 6G). These findings further validated the specificity of the CADM1-mediated cytotoxicity observed in in vivo osteosarcoma models.

Discussion

Over the past few decades, genomic profiling and the development of preclinical models have led to progress in understanding the biology of osteosarcoma. Molecular characterization of osteosarcoma PDX and patient samples has provided opportunities to identify therapeutic targets and design biomarker-stratified clinical trials. However, clinical trials of targeted therapies such as tyrosine kinase inhibitors have yet to demonstrate them to be curative (24–26). This suggests that targeting pathway alterations may not suffice in disease control because alternative pathways may negate target inhibition through compensatory regulations or redundancy. Single-agent immune checkpoint inhibitors have also produced disappointing results in recent trials (27). Thus, novel therapeutic strategies for osteosarcoma are urgently needed (28).

ADCs targeting surface antigens have exhibited promising clinical activity in several cancer types and have attained regulatory approval for the treatment of multiple solid and hematologic malignancies (3–5). To identify high-confidence cell-surface proteins that are enriched specifically in osteosarcomas, we developed an integrated transcriptomic and proteomic surfaceome profiling approach (13). Using it, we identified several overexpressed cell-surface antigens, such as LRRC15, MMP14, MRC2, and CD276. Investigators previously tested ADCs targeting these proteins in osteosarcoma models and demonstrated encouraging antitumor activity (13–16).

In the present study, we validated that CADM1, another protein identified in our surfaceome profiling, is highly expressed in osteosarcoma cell lines, PDX models, and patient samples. CADM1 is a transmembrane glycoprotein belonging to the immunoglobulin superfamily that is crucial for cell-cell adhesion, recognition, and interaction and maintaining tissue homeostasis. CADM1 also plays a role in the immune system by being essential for mast cell development (29), regulating T-cell retention in the intestinal tract, and enhancing natural killer cell–mediated cytotoxicity (30). Additionally, CADM1 behaves heterogeneously in carcinogenesis. Authors reported that it acts as a tumor suppressor for non-small cell lung cancer (31) but promotes tumor growth for adult T-cell leukemia/lymphoma (32). Furthermore, chimeric antigen receptor T-cell therapy targeting CADM1 is being explored for small-cell lung cancer. CADM1 was found to be expressed in 83% of small cell lung cancer samples, and membrane co-localization of CADM1 and 4.1R protein was associated with advanced tumor grade for this cancer (33). Moreover, a CADM1-targeting ADC is being explored in the treatment of adult T-cell leukemia/lymphoma (34).

CADM1 was reported to be highly expressed on the surface of osteoblasts but not osteocytes (35). Clear membranous staining for CADM1 was detected in most osteosarcoma samples (46/57), whereas less than 20% (5/28) of chondrosarcoma samples were CADM1-positive. The authors concluded that CADM1 is a novel osteoblastic adhesion molecule that is expressed transiently during osteoblastic maturation and a useful diagnostic marker for osteosarcoma (35).

In the present study, we tested a novel CADM1-ADC with six osteosarcoma PDX models, finding that it significantly inhibited tumor growth in all six. The results demonstrated MCR in one model, CR in two, and PR in three, with improved EFS in all models following single-dose treatment. The CADM1-ADC exhibited selective cytotoxicity and inhibited the proliferation of CADM1-expressing neuroblastoma and rhabdomyosarcoma cell lines, highlighting its potential for treating other CADM1-expressing pediatric solid tumors. These findings are particularly significant, as OS PDX models have rarely shown responses in previous Pediatric Preclinical Testing Program and Pediatric Preclinical Testing Consortium testing. This ADC, along with the recently approved B7-H3 ADC, has the potential to transform this landscape (36).

The efficacy of ADCs is dependent in part on their cytotoxic payload. In our study, the pyrrolobenzodiazepine dimer payload, SG3199, had the greatest potency against osteosarcoma cells. This aligns with recent studies in which LRRC15-targeting ADCs with a nemorubicin payload were more effective at inhibiting osteosarcoma growth than were those with a monomethyl auristatin E payload (37). However, identifying the best antibody-linker-payload combination still requires further investigation. We are developing three Good Manufacturing Practices–compliant CADM1-ADCs, each with distinct linker-payload combinations, to determine the most effective and safe candidate for a clinical trial. In vitro and in vivo studies of these novel ADCs with preclinical osteosarcoma models are underway.

We acknowledge our study’s limitations, particularly the need to evaluate CADM1 toxicity in normal CADM1-expressing tissue. On-target, off-tumor toxicity remains a key limitation of ADC development, as many target antigens for solid tumors are expressed in normal tissues. Nevertheless, several FDA-approved ADCs achieve acceptable safety profiles because their cytotoxic payloads are primarily released following internalization, whereas normal tissues typically express lower antigen levels and lack the proliferative activity that makes tumor cells especially vulnerable to chemotherapeutic agents. Linker stability, selective antibody-antigen internalization, and optimized dosing further contribute to this therapeutic balance. Trastuzumab deruxtecan exemplifies this principle, demonstrating efficacy against HER2-positive cancers despite HER2 expression in normal tissues such as the heart and lung (38). To further minimize off-tumor effects, ongoing strategies include antibody affinity optimization and innovative linker-payload systems that activate preferentially in the tumor microenvironment, thereby enhancing the therapeutic index and expanding the clinical utility of ADCs. Also, in this study, we assessed toxicity according to mouse body weight loss and clinical observations of distress and mortality as detailed in our previous preclinical testing (13). Given the lack of robust data to illustrate the interaction between human anti-CADM1 antibody and mouse CADM1 protein, the present study design is unable to fully evaluate the on-target, off-tumor effect. Further studies are being performed to evaluate the potential toxicological profile of the CADM1-ADC in humans, including dose-limiting toxicities; on-target, off-tumor effects; off-target effects; and potential species-specific responses that may not be evident in preclinical models. Moreover, we found that CADM1 was highly expressed in several other solid tumors. Whereas preclinical testing of the CADM1-ADC in treatment of those cancers is necessary to validate its clinical potential, that is beyond the scope of the present study and will be investigated in the future.

In conclusion, our surfaceome profiling identified CADM1 as a target on osteosarcoma cells, which we validated in PDX models and human tumor samples. Preclinical testing of the CADM1-ADC demonstrated promising responses, supporting its potential in therapy for osteosarcoma. These findings suggest a pathway for developing CADM1-targeted cytotoxic agents for treatment of osteosarcomas and other pediatric solid tumors.

Appendix: Statistical Methods

For solid tumor experiments, an event is defined as a quadrupling of tumor volume from day 0, whereas for acute lymphoblastic leukemia (ALL) experiments, an event is defined as a hCD45 cell percentage exceeding 25%. In both cases, the exact time-to-event is estimated by interpolating between the measurements directly preceding and following the event assuming log-linear growth. For brain tumor experiments, an event is defined as an animal becoming moribund or experiencing a severe neurologic deficit, and the time-to-event is based on the day on which the event is noted. Differences in EFS between experimental groups (e.g., treated vs. controls) are tested using the G^ρ test described by Harrington and Fleming (Biometrika 69:553–566, 1983; α = 0.05, two-sided alternative) with ρ = 1, which is equivalent to the Peto and Peto modification of the Gehan-Wilcoxon test. This test is more efficient than the Mantel-Cox log-rank test under shift (i.e., translation) alternatives (Tarone and Ware, Biometrika 64:166–80, 1977), which is a pattern we have observed in previous testing data with similar xenograft mouse models.

The objective response categories are PD without and with growth delay (PD1 and PD2, respectively; defined only for treated mice), SD, PR, CR, and MCR.

For solid tumor experiments, objective response categories are defined as follows:

• PD with <50% tumor regression throughout the study and >25% tumor growth at the end of the study.

• PD1 with PD and when the mouse’s time-to-event is ≤200% of the Kaplan-Meier (KM) median time-to-event in the control group.

• PD2 with PD and when the time-to-event is >200% of the KM median time-to-event in the control group.

• SD with <50% tumor regression throughout study and ≤25% tumor growth at the end of the study.

• PR with ≥50% tumor regression at any point during the study but a measurable tumor throughout the study period.

• CR with disappearance of a measurable tumor mass during the study period.

• MCR with no measurable tumor mass for at least three consecutive weekly readings at any time after treatment has been completed.

• For ALL experiments, objective response is defined as follows:

• PD when hCD45 never <1% during the study period and the mouse reaches event (hCD45 >25%) at some point during the study period.

• PD1 with PD and when the mouse’s time-to-event is ≤200% of the KM median time-to-event in the control group.

• PD2 with PD and when the mouse’s time-to-event is >200% of the KM median time-to-event in the control group.

• SD when hCD45 never <1% and the mouse never reaches event during the study period.

• PR when hCD45 <1% at least once during the study period but not CR.

• CR when hCD45 <1% for at least two consecutive weekly readings during the study period regardless of whether event is reached at a later time point.

• MCR when hCD45 <1% for at least three consecutive weekly readings at any time after treatment has been completed.

Overall group response is determined by the median response among evaluable mice as follows: each mouse is assigned a score from 0 to 10 based on its response—PD1 = 0, PD2 = 2, SD = 4, PR = 6, CR = 8, and MCR = 10—and the median for the group determines the overall response. If the median score is halfway between an objective response number category, the objective response is assigned to the lower response category (e.g., an objective response score of 9 is scored as a CR). Studies in which toxicity is greater than 25% or in which the control group is not SD or worse are considered unevaluable and are excluded from analysis. Treatment groups with PR, CR, or MCR are considered to have had an objective response. Agents inducing objective responses are considered to be highly active against the tested line, whereas agents inducing SD or PD2 are considered to have intermediate activity, and agents producing PD1 are considered to have a low level of activity against the tested line.

For brain tumor experiments, magnetic resonance imaging is time-consuming and involves a relatively high rate of animal death after scanning, probably due to anesthesia overdose. Therefore, statistical analysis for these experiments will most often be limited to comparison of EFS.

For combination testing projects, the primary objective is generally to demonstrate that the combination is significantly more effective than either agent used at its optimal single-agent dose/schedule. This condition is termed therapeutic enhancement, which represents a therapeutic effect for which a tolerated regimen of a combination treatment exceeds the optimal effect achieved at any tolerated dose of monotherapy associated with the same drugs used in the combination.¹ This definition is operationalized as follows: therapeutic enhancement is considered present when the tumor growth delay (T-C) for a combination is greater than the tumor growth delay for each of the single agents tested at its maximum tolerated dose and when the EFS distribution for the combination treatment is significantly better than the EFS distributions for both of the single agents tested at their maximum tolerated doses. To control the experiment-wise type I error at 5%, statistical tests are evaluated at the Bonferroni-corrected significance level α = 0.01 due to the five comparisons being made (combination vs. agent 1 alone, combination vs. agent 2 alone, agent 1 vs. control, agent 2 vs. control, and combination vs. control). Testing is considered unevaluable for therapeutic enhancement if either single agent used alone produces a median EFS beyond the observation period. If a treatment group exhibits excessive toxicity (>25% toxic deaths), therapeutic enhancement is not evaluated.

Below is an example of summary graphs produced for two solid tumor experiments. These plots include 1) KM EFS curves, 2) tumor volume growth curves, and 3) relative tumor volume (RTV) growth curves. For ALL experiments, only two plots per experiment will be shown: KM EFS and hCD45 cell percentage versus days (comparable to the tumor volume growth curves for solid tumors). For the brain tumor experiments, only KM EFS curves will be shown.

Summary tables for solid tumor experiments will summarize results for each experimental group (e.g., treated vs. controls) and include the following columns:

• Study, the alphanumeric code for the specific experiment.

• Tumor, the alphanumeric code for the tumor model.

• Grp, experimental group, usually C(ontrol) or T(reatment).

• N, the total number of mice entering the experiment.

• N_d, the number of mice experiencing toxic death.

• N_x, the number of additional mice excluded from the analysis.

• N_a, the number of mice in the analysis.

• N_ev, the number of events.

• KMmed, the KM estimate of the median time-to-event (days).

• EFS T-C, the difference in median time-to-event (days) between the T and C groups.

• EFS T/C, the ratio of median time-to-event for the T and C groups.

• EFS P-value, computed using an exact log-rank test.

• V₀ mean, the mean of initial tumor volume (cm³).

• V₀ P-value, computed using the Peto and Peto modification of the Gehan-Wilcoxon test.

• Mean (±SD) minRTV, the mean (± SD) per-mouse minimum relative tumor volume.

• Tumor volume P-value, computed using the Wilcoxon rank-sum test.

• One column for each category of objective response (i.e., PD, PD2) showing the number of mice in each category.

• Resp rate, the response rate, defined as the percentage of mice having PR or better.

• Med resp, the median response evaluation.

Summary tables for ALL experiments will include the same columns as those for the solid tumor experiments, except hCD34 cell percentage will be reported instead of tumor volume or relative tumor volume, as appropriate. Summary tables for brain tumor experiments will not include any of the columns concerning tumor volume or objective response.