The B7-H3–Targeting Antibody–Drug Conjugate m276-SL-PBD Is Potently Effective Against Pediatric Cancer Preclinical Solid Tumor Models

Pediatric cancer mortality has declined by 65% since 1970 (1). A large proportion of this progress can be attributed to marked reduction in deaths from leukemia, which has been achieved through optimization of chemotherapy regimens (2). In contrast, there has been only modest improvement in survival probabilities for children, adolescents, and young adults with many high-risk solid malignancies despite dramatic intensification of cytotoxic therapy (3), such that survivors face lifelong complications due to intensive, multimodal therapy, underscoring a critical need for development of new therapies for pediatric cancer (4–6). Novel therapeutic approaches are thus required to improve outcomes for high-risk childhood solid malignancies.

While immune checkpoint inhibitors have shown remarkable efficacy in many adult solid malignancies, there has not been a significant activity signal in clinical trials for patients with pediatric solid tumors (7, 8), at least in part due to the fact that pediatric tumors generally have low mutation burdens and thus few neoantigens, and generally low expression of PD-L1 (9–11). CD19-directed chimeric antigen receptor (CAR) engineered T-cell therapy has shown remarkable efficacy in B-cell leukemias (12–14), but as yet few convincing durable responses in solid malignancies have been observed for CAR T cells targeting surface antigens relevant for pediatric solid tumors. Antibody–drug conjugates (ADC) couple the specificity of antibodies with highly potent cytotoxic agents, conferring increased potency and potential for bystander killing not seen with mAbs alone (15–18). There are currently 10 FDA-approved ADCs for a variety of human cancers, dozens in clinical development, but in children only ADCs against leukemias and lymphomas have been actively developed.

The B7-H3 protein (encoded from the CD276 gene) is a checkpoint molecule that is highly expressed in pediatric solid tumors, with reportedly limited expression in normal tissues (19–25). Overexpression of B7-H3 is correlated with tumor progression, metastasis, and poor clinical outcome across a variety of malignancies (26, 27). mAbs targeting B7-H3 cause limited toxicity and have demonstrated some antitumor activity in clinical trials (28, 29). Furthermore, intrathecal delivery of ¹³¹I-8H9 (omburtamab) is in late-phase clinical development for neuroblastomas with brain and/or leptomeningeal metastasis, to date showing a manageable toxicity profile with tumoricidal doses (30). Recent work has concerned development of B7-H3 CAR-T cells (23, 24), with one study demonstrating efficacy in Ewing sarcoma, osteosarcoma, and medulloblastoma xenografts (22). Finally, a pyrrolobenzodiazepine (PBD)-armed ADC against CD276 (m276-PBD, a precursor to m276-SL-PBD) showed antitumor activity in a variety of adult cancer preclinical models, including superiority over a monomethyl auristatin E (MMAE)-armed anti-CD276 ADC (31). While both PBD and MMAE are DNA-binding agents, the superiority of PBD is likely due to PBD functioning irrespective of the stage of the cell cycle, thereby eradicating both the tumor and the tumor vasculature. Together, these data suggest CD276 is a rational target for further ADC development. Here, we present results from the Pediatric Preclinical Testing Consortium (PPTC) investigation of m276-SL-PBD across a large and diverse panel of pediatric cancer patient-derived (PDX) and cell line–derived xenograft (CDX) models.

PDX RNA sequencing (RNA-seq) data were gathered from the PPTC dataset available on cBioPortal (32). Normal tissue RNA-seq data were gathered from the Genome-Tissue Expression (GTEx) resource (https://gtexportal.org/home/). Patient tumor RNA-seq data were collected from the Treehouse Childhood Cancer Initiative (https://treehousegenomics.soe.ucsc.edu/), which included data from the Therapeutically Applicable Research to Generate Effective Treatments initiative (https://ocg.cancer.gov/programs/target). The associated data and code used to generate the figures are available in an open-access GitHub Repository at https://github.com/nathankendsersky/pptc_cd276.

Tissue microarrays (TMA) including human normal childhood tissues, human neuroblastoma primary tumors, and PDX arrays for the histologies studied here were constructed as described previously, with normal tissues obtained from surgical or autopsy specimens at the Children's Hospital of Philadelphia (Philadelphia, PA), as described previously (33–35). Staining was performed with a polyclonal B7-H3 antibody (R&D, AF1027) at a 1:1,000 dilution. Aperio GENIE Classifier (Leica Biosystems) was used to annotate tumor regions and exclude stroma and necrosis. Tumor classification was reviewed and any discordant classifications were manually annotated by a pathologist (J. Pogoriler). For each tumor core, Aperio Membrane V9 algorithm (Leica Biosystems) was used to quantify staining intensity and percentage of tumor cells stained. H-scores were calculated by the following formula [(1 × %cells intensity = 1) + (2 × %cells intensity = 2) + (3 × %cells intensity = 3)]. Cores in which fewer than 100 cells were available for evaluation were excluded.

The antibody used to create the m276-SL-PBD ADC, called m276-SL, is a modified version of the fully human m276 IgG1 antibody (31) engineered to contain four mutations (S239C, L234A/L235A/P329G) in the Fc domain (36). The S239C mutation was used for site-specific maleimide-based PBD conjugation via a dipeptide (valine-alanine) linker (36) while the LALA-PG mutations were used to prevent undesirable interactions of the ADC with Fc receptors that could potentially result in off-target killing (37). Further details about the synthesis of the modified ADC will be published separately.

CB17SC scid^−/− female (5–8 weeks) mice (Taconic Farms or ENVIGO) were used to propagate subcutaneous xenografts. All mice were maintained under barrier conditions and experiments were conducted using protocols and conditions in accordance with the Institutional Animal Care and Use Committee at The University of Texas Health San Antonio (UTHSA; San Antonio, TX), M.D. Anderson Cancer Center (Houston, TX), and the Children's Hospital of Philadelphia (Philadelphia, PA). The number of mice per group (control and treatment) was varied as indicated, with 10, 2, or 1 animal(s). For randomized controlled trials with either 10 or 2 mice per arm, 0.5 mg/kg of m276-SL-PBD or (0.5 mg/kg) of 100 μL/20 g PBS was administered intraperitoneally once a week for 3 weeks (days 1, 8, and 15). A dose of 0.5 mg/kg was chosen because it was the lowest effective dose discovered in Seaman and colleagues (31). For the single mouse trial (SMT; N = 1) study (38), a single dose of 0.5 mg/kg m276-SL-PBD was administered intraperitoneally. Treatment was initiated when tumor volumes reached 200 mm³, and tumor volume and event-free survival (EFS) were monitored as reported previously (29).

A tumor event was defined as a quadrupling of tumor volume from the day of treatment initiation, where tumor volume was estimated from caliper measurements as (4/3) × π × ((length + width)/4)³. The exact time to event was estimated by interpolating between the measurements directly preceding and following the event, assuming log-linear growth. Differences in EFS between experimental groups of N > 2 were evaluated using the Gehan–Wilcoxon test, and differences in minimum relative tumor volume (RTV; ratio of current tumor volume to volume at treatment initiation) were evaluated using the Wilcoxon rank-sum test.

Area over the curve (AOC) is the area above the RTV growth curve but below the event threshold of RTV = 4. It was calculated for each mouse using the trapezoid rule and was then divided by 4; a day of complete tumor remission, therefore, adds exactly 1 to the AOC. The AOC is correlated with time to event, but also captures the maximum extent and durability of tumor remission. It is a continuous, composite measure of tumor response over the course of the experiment and is therefore well suited for genomic analyses of association between tumor response and specific mutations.

The objective response categories are progressive disease (PD, which is subdivided among treated mice into PD without and with growth delay, PD1 and PD2, respectively), stable disease (SD), partial response (PR), complete response (CR; no measurable tumor mass for one recording), and maintained CR (MCR; no measurable tumor mass for at least three consecutive weekly recordings) as described previously (39). Response rate is defined as the percentage of mice with PR or better. Mice experiencing a possibly treatment-related death (i.e., drug toxicity), mice with failed engraftment, and mice that unexpectedly die for reasons unrelated to treatment are excluded from statistical analyses of time to event, minimum tumor volume, and objective response.

For experiments of groups of N = 2 mice per group, we did not perform any statistical tests of group differences but have reported all other summary statistics. For the SMT, we report time to event, AOC, minimum RTV, and the objective response measure for each PDX or CDX model tested.

CD276 mRNA was generally highly expressed in pediatric solid tumors, compared to generally low expression in a panel of 1,641 normal tissues samples from 25 organs (Fig. 1A). Likewise, CD276 mRNA was highly expressed in each PDX model studied here (32). Ewing sarcoma models had the lowest median mRNA expression of CD276 (median = 37.42 transcripts per million, TPM, normalized by trimmed mean of M-values) whereas osteosarcomas (median = 165.92 TPM) and Wilms tumors (median = 142.29 TPM) showed the highest levels of mRNA expression (Fig. 1B). IHC of three tumor TMAs showed strong staining across the tested PDXs (Fig. 2A–C). Osteosarcoma models showed the highest expression of CD276 (median H-score = 243.0). Overall, we observed a moderate correlation (R = 0.527, P = 7.13e-05) between mRNA (TPM) and protein expression (median H-Score; Supplementary Fig. S1). For the normal tissue array, the membrane algorithm was applied to assist in reproducible evaluation of the intensity of membranous staining, with pathologist review to determine which cell types were staining in any given tissue (Supplementary Fig. S2). Strong staining (3+) was seen in the basal layer of the epidermis of the skin. Weak (2+) staining was present in the adrenal cortex, bladder urothelium, basal layer of the esophageal epithelium, prostatic epithelium, pancreatic islet tissue, ovarian stroma, splenic stroma, and a subset of pituitary cells. Staining of the bone marrow hematopoietic cells as well as the epithelium in the ileocecum, colon, and stomach varied from negative (0) to weak (2+). Equivocal (1+) staining was present in a subset of cardiac myocytes, a subset of aortic stromal cells, a subset of gastrointestinal tract stromal cells (appendix, colon, small intestine, stomach), gallbladder epithelium, basal urothelium of the ureter, fallopian tube epithelium and subsets of cortical neurons, renal podocytes and subset of renal tubular cells, pancreatic exocrine tissue, hepatocytes, pulmonary epithelium, a subset of germinal center cells (lymph node, tonsils, spleen), salivary gland acinar cells, and a subset of developing spermatozoa. Staining was negative in other tissues including cerebellum, mature bone, skeletal muscle, and thyroid.

The efficacy and toxicity of m276-SL-PBD was initially evaluated in 26 pediatric PDX and CDX models in randomized controlled trial with either 10 or 2 mice per arm (Table 1; Supplementary Table S1). This was followed by a SMT using a single dose of m276-SL-PBD in 47 distinct models (33 PDX, 14 CDX; Table 2; ref. 40). m276-SL-PBD was generally well tolerated with a treatment-related mortality rate of 5.5% when administered in three weekly doses and 0% when administered as a single dose. In the randomized trials with three doses given weekly, late weight loss was observed in the Ewing sarcoma, rhabdomyosarcoma, and Wilms tumor models, but not in the osteosarcoma and neuroblastoma models (Supplementary Figs. S3–S6). No significant weight loss was observed in the SMT with a single administered dose of m276-SL-PBD (data not shown).

In the randomized trials, m276-SL-PBD induced significantly prolonged EFS in all tested mice across all models. The minimum RTV in treated mice was significantly less than the starting mean tumor volume for all models studied with the exception of one of five osteosarcoma models (OS-9) and one of 12 neuroblastoma models (NB-1691; Fig. 3A; Supplementary Figs. S3–S7). m276-SL-PBD showed objective responses in all models except the NB-1691 neuroblastoma model (PD1 response) and the OS-9 osteosarcoma model (SD response; Table 1). All Ewing, rhabdoid tumors, and Wilms models, and four of five osteosarcomas and 10 or 12 neuroblastomas showed a MCR (N = 16) or CR (N = 5). The overall response rate (PR or better) was 92.3% in the randomized trial (Supplementary Table S2). In the SMT, a single dose of m276-SL-PBD significantly prolonged EFS and decreased minimum RTV in models across all histologies tested (Figs. 3B and C; Supplementary Figs. S8 and S9). The overall response rate was 91.5% with 31 of 47 (66%) models showing a MCR (Table 2; Supplementary Table S2). The 10 models studied in both trial designs showed concordant responses (KT-10, KT-11, Rh-18, Rh-30, Rh-36, SK-NEP-1, ES-1, NB-1643, OS-1). Across both the randomized trial and SMT, we observed no correlation between CD276 expression (mRNA or protein) and objective response category (Supplementary Fig. S10).

B7-H3 is an immunomodulatory cell surface protein in the B7 superfamily, which includes other well-characterized members such as PD-L1 (41). Here we confirm and extend the previous reports of high expression on the majority of high-risk pediatric solid tumors, with significant normal childhood tissue expression by IHC limited mainly to the basal layer of skin (not previously reported), and weak expression noted in a variety of different childhood tissues (Supplementary Fig. S2; refs. 20–25). Whether or not this normal tissue expression will be a liability for an ADC that requires internalization remains unknown. In addition, B7-H3 has been shown to play a role in metastasis, and high B7-H3 expression is associated with poor outcomes across a variety of malignancies (23). However, the precise mechanisms that lead to B7-H3 overexpression and how the protein exerts its protumor function are not defined. Early studies suggested B7-H3 played a role in T-cell costimulation (42, 43). Other data point to a potential immunosuppressive, checkpoint mechanism (44, 45). Future studies are required to more precisely define the role of B7-H3 in normal human development as well as pediatric cancer tumor cell-intrinsic and extrinsic mechanisms of immune evasion.

B7-H3 has emerged as a promising immunotherapeutic target for a variety of cancers, with several immunotherapeutic platforms targeting this protein in development, including antibodies, bispecific antibodies, ADCs, and CAR T cells (22–24, 28–31). Because solid pediatric malignancies tend to have low mutational burden and immunologically cold microenvironments, which hinder adaptive immunotherapeutic approaches (9–11), ADCs may be ideal platforms for differentially expressed proteins in the plasma membrane.

We used the m276 antibody platform and covalently linked PBD with a cleavable valine-alanine linker and a drug–antibody ratio (DAR) of 2. The m276-SL-PBD ADC showed potent antitumor activity against pediatric solid xenografts at 0.5 mg/kg regardless of whether this was administered once or weekly for 3 weeks. Overall response rates above 90% in both the randomized and SMTs, with limited observed toxicity, is an encouraging signal for clinical development. Importantly, m276 binds human and mouse B7-H3 with equal affinity (31), which further supports a potential therapeutic index in pediatric patients. The weight loss in the multidose experiments in some of the sarcoma and Wilms models may be due to a cumulative effect of PBD, as has been previously observed in clinical trials of PBD-family agents (46), as well as the sensitivity of CB17SC scid^−/−/−animals to DNA-damaging agents. No skin toxicity was observed in the mice treated in this study.

Our results with m276-SL-PBD confirm the utility of the SMT design as an efficient and accurate design for screening large panels of xenograft models. A retrospective analysis of 67 agents tested by the Pediatric Preclinical Testing Program using N = 8 or 10 mice per group evaluated whether a single mouse chosen at random predicted the group median objective response. Out of 2,134 total comparisons, the single mouse matched the group median response 78% of the time and fell within one response category away 95% of the time (40). More recent work analyzed 30 agents tested by the PPTC from 2015 to 2018 and observed exact and near match rates of 82% and 94%, respectively (47).

It is very likely that ADC payloads may need to be tailored to potentially unique cytotoxic vulnerabilities and mechanisms or drug resistance in childhood cancer. Especially in the relapse setting after dose intensive chemotherapy, many of the favored payloads in licensed ADCs like DM1 may not be ideal as they are substrates for P-glycoprotein and other drug efflux pumps. In addition, it was recently demonstrated that m276-SL-PBD confers advantages over m276-MMAE due to PBD-mediated eradication of tumor-associated vasculature in adult tumor xenograft models (31). However, ADCs with PBD-family cytotoxic agents have shown toxicity in the clinic (48), and the antitumor activity seen at 0.5 mg/kg must be viewed in the context of a dose and schedule that can be achieved in the clinic. It is reassuring in our study that deescalation from 0.5 mg/kg once a week for 3 weeks to a single 0.5 mg/kg dose had no impact on antitumor activity, suggesting that with an ideal target, stable linker, and brisk internalization kinetics, PBD containing ADCs directed at B7-H3 can be developed clinically. Our data support further development of B7-H3 ADCs for childhood cancers.

S. Seaman reports a patent for 62/947.135 pending and a patent 10604582 issued. Y. Feng reports a patent for 62/947135 pending. B. St. Croix reports a patent for 62/947,135 pending and a patent 10604582 issued. No disclosures were reported by the other authors.

N.M. Kendsersky: Data curation, visualization, writing–original draft, writing–review and editing. J.Lindsay: Data curation, visualization, writing–original draft, writing–review and editing. E.A. Kolb: Conceptualization, funding acquisition, writing–review and editing. M.A. Smith: Conceptualization, resources, formal analysis, supervision, writing–original draft, project administration, writing–review and editing. B.A. Teicher: Conceptualization, writing–review and editing. S.W. Erickson: Data curation, formal analysis, visualization. E.J. Earley: Data curation, formal analysis, visualization. Y.P. Mosse: Conceptualization, supervision, investigation, writing–review and editing. D. Martinez: Data curation, formal analysis, investigation, visualization. J. Pogoriler: Data curation, formal analysis, investigation, visualization. K. Krytska: Data curation, formal analysis, investigation. K. Patel: Data curation, formal analysis, visualization, methodology. D. Groff: Data curation, investigation. M. Tsang: Data curation, investigation. S. Ghilu: Data curation, investigation. Y. Wang: Data curation, investigation, methodology. S. Seaman: Conceptualization, resources. Y. Feng: Resources. B. St. Croix: Conceptualization, resources, writing–original draft, writing–review and editing. R. Gorlick: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, writing–review and editing. R. Kurmasheva: Conceptualization, supervision, writing–original draft, writing–review and editing. P.J. Houghton: Conceptualization, resources, supervision, funding acquisition, investigation, writing–original draft, project administration, writing–review and editing. J.M. Maris: Conceptualization, resources, supervision, funding acquisition, investigation, writing–original draft, project administration, writing–review and editing.

This work was supported by NCI U01 CA199287 (to J.M. Maris), NCI R35 CA220500 (to J.M. Maris), NCI U01 CA199221 (to Richard G. Gorlick), NCI U01 CA199297 (to P.J. Houghton), CPRIT RP160716 (to P.J. Houghton), NCI F31 CA239424 (to N.M. Kendsersky), NCI F31 CA254244 (to J. Lindsay), NCI U01 CA199222 (to Gregory J. Gatto), and the Giulio D'Angio Endowed Chair (to J.M. Maris).

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