SGN-CD228A Is an Investigational CD228-Directed Antibody–Drug Conjugate with Potent Antitumor Activity across a Wide Spectrum of Preclinical Solid Tumor Models

This article is featured in Highlights of This Issue, p. 419

Melanotransferrin (CD228, MELTF, MFI2, p97) is a glycophosphatidylinositol (GPI)-anchored membrane protein first described as a melanoma-specific antigen (1). CD228 shares high homology with other members of the transferrin family of iron-binding proteins and has 37% to 39% amino acid identity with human serum transferrin, lactoferrin, and ovotransferrin (2, 3). Two forms of CD228 have been reported: a GPI-anchored variant and a shorter secreted variant that lacks the C-terminal GPI anchor and can be transported across the blood-brain barrier (4). CD228 is an oncofetal antigen with demonstrated expression in fetal colon and umbilical cord tissue and high expression in multiple solid tumors, such as melanoma, non–small lung cancer (NSCLC), triple-negative breast cancer (TNBC), colorectal, and pancreatic cancer (refs. 3, 5; Fig. 1). Importantly, CD228 has shown highly restricted low-level expression in normal tissues limited to the skin epidermis, brain endothelium, kidney tubules, and sweat and salivary gland ducts (6, 7), indicating that this antigen is a promising target for cancer therapies. While there is no clear consensus on the function of CD228, studies have shown that it plays a role in endothelial cell migration and angiogenesis, plasminogen activation, and differentiation (8). In a preclinical mouse model, L235, a CD228-directed antibody, delayed tumor growth and decreased metastasis to the brain, suggesting that CD228 plays a role in tumor proliferation and migration (9).

Antibody–drug conjugates (ADCs) employ linker chemistry to combine the specificity and favorable pharmacokinetics of a monoclonal antibody with the cytotoxic potential of a chemotherapeutic drug. The primary mechanism of action (MOA) of ADCs involves binding to the target antigen on tumor cells, internalization of the antigen/ADC complex, and release of the cytotoxic payload after cleavage of the linker [reviewed in (10)]. ADCs have demonstrated promising activity against various treatment-refractory hematologic malignancies and solid tumors, and 12 ADCs are currently approved by the FDA for clinical use (11). While these ADCs have improved patient outcomes and broadened treatment options, there is opportunity to build upon the observed activity and expand the number of patients that may benefit from ADC therapeutics targeting unique antigens.

CD228 has been explored preclinically as an ADC target. L49 is a murine IgG1 antibody directed to human CD228 that was isolated after immunizing BALB/c mice with melanoma and lung cancer cell lines and was demonstrated to bind to CD228 with a dissociation constant (K_D) of 1 nmol/L (12, 13). While L49 holds promise as an ADC antibody backbone (13), the internalization of the CD228 antigen and the potential of CD228-directed antibodies to deliver drugs to cancer cells have previously been unclear. Hellstrom and colleagues reported that melanoma cells continue to express CD228 after prolonged exposure to L49 (12). More recently, another CD228-directed antibody, hup97, was used together with a secondary antibody conjugated to the potent microtubule-disrupting agent monomethyl auristatin E (MMAE) as a reporter of antibody internalization (14), which showed weak cytotoxicity to tumor cells compared with antibodies against antigens known to internalize efficiently (such as L antigen and CD71). However, compared with these antigens, CD228 was expressed at low levels in the tested cell lines, making the results uncertain. Another study explored L49 as the antibody backbone of a CD228-directed ADC, in which the antibody was conjugated to the antimitotic drug monomethyl auristatin F (MMAF) with a proteolytically cleavable valine-citrulline (Val-Cit) dipeptide linker (15). In the study, the ADC was readily internalized by the melanoma cell lines tested; however, only cells with high CD228 receptor expression (80,000–240,000 sites/cell) showed sensitivity.

Here, we investigated a novel ADC, SGN-CD228A, which uses the L49 antibody to direct cytotoxic MMAE to tumor cells. SGN-CD228A is composed of humanized L49 (hL49) conjugated with an average of 8 molecules of MMAE via a novel PEGylated glucuronide linker (Supplementary Fig. S1; ref 16). We show that the antibody backbone of SGN-CD228A has high specificity and affinity for CD228. In vitro, SGN-CD228A was efficiently internalized and active in cancer cell lines across multiple tumor types, and its cytotoxicity correlated with CD228 expression and internalization and sensitivity to the MMAE payload. In contrast to previously tested ADCs directed to CD228 (15), this combination of antibody, linker, and payload was significantly more active, with the glucuronide linker achieving greater potency than the Val-Cit linker in melanoma models in vitro and in vivo. Moreover, SGN-CD228A was effective at inhibiting tumor growth in multiple xenograft models of melanoma, TNBC, and NSCLC. Overall, these findings support the development of SGN-CD228A for the treatment of solid tumors.

Hybridoma cell lines producing the two murine anti-CD228 IgG1 mAbs, L235 (ATCC) and L49 (13), were grown as recommended. hL49 was produced from stably transfected CHO-DG44 cells, and cL49, by transiently transfected 293F cells. Antibodies were purified from culture media with MabSelect Protein A column (Amersham). hIgG1k (#I5154, Sigma-Aldrich) was used as a flow cytometry control.

All cancer cell lines were obtained from ATCC, German collection of microorganisms and cell cultures (DSMZ), or the European Collection of Authenticated Cell Cultures and grown according to the recommendations provided. Cell lines were expanded upon receipt, frozen in aliquots between passage 3 and 5, and only kept in culture for <6 weeks at a time during experiments. Cell line stocks were authenticated and Mycoplasma tested at IDEXX BioAnalytics using STR-based DNA profiling and multiplex PCR. Mycoplasma testing was performed every 3 weeks while cells were in culture using Lonza MycoAlert Mycoplasma Detection Kit (#LT07–318). A stable RPMI-7951 cell line expressing human CD228 was generated by lentiviral transduction with a custom plasmid containing full-length human CD228 cDNA (RefSeq# NM_005929.5) driven by the CMV promoter. CD228-expressing cells were selected with 1 μg/mL puromycin.

K_D values for hL49 and cL235 were determined using a saturation binding assay. Briefly, parental or CD228-expressing RPMI-7951 cells were incubated with CD228-directed antibodies labeled with AF647 (Life Technologies). Fluorescence was analyzed by flow cytometry (Attune NxT cytometer). Quantitative flow cytometry (qFACS) was used to determine the cell-surface abundance of human CD228 receptors on all cell lines using unlabeled mouse L49 and L235 and an anti-mouse FITC secondary antibody and calibration beads (QIFIKIT; Dako)

Formalin-fixed, paraffin-embedded tumor tissues were purchased from US Biomax Inc. All samples stained with the rabbit polyclonal anti-CD228 antibody (#HPA004880; Sigma-Aldrich) were processed on Bond-Max autostainer (Leica). Antigen retrieval was performed using EDTA-based pH 9 Bond Epitope Retrieval Solution 2 (Leica) for 20 minutes at 98°C to 100°C, and nonspecific background was blocked with Protein Block (Dako). Isotype-matched rabbit IgG1 (#11–000–003; Jackson Immuno Research) was used as negative control. Automated IHC staining was performed with Bond Polymer Refine Detection (DAB) and AP Red Detection kits (Leica). Slides were incubated with primary anti-CD228 antibodies (1 μg/mL) for 45 minutes, followed by DAB chromogen staining and hematoxylin counterstaining. Slides were rinsed and dehydrated to allow for cover slipping. Images were captured using a slide scanner (Leica, Aperio AT2). Slides were evaluated and scored by a pathologist and images were taken using a Zeiss Axiovert 200M microscope (Carl Zeiss). IHC images of representative staining in Fig. 1 were generated as part of an IHC analytical validation at Mosaic Laboratories (CellCarta).

A detailed experimental procedure for the preparation of the drug-linkers has been previously described (17) and is also included in the Supplementary Materials and Methods (Supplementary Fig. S2). Conjugation of the quenched-fluor linkers is described in the Supplementary Materials and Methods (Supplementary Figs. S3 and S4).

Reagents were titrated (2,000–0.1 ng/mL for ADCs or 500–0.03 nmol/L for unconjugated MMAE) and incubated with cancer cell lines for 96 hours. Cell viability was measured using the Cell-Titer Glo viability assay (Promega) following the manufacturer's instructions, and cytotoxic activity was measured with an EnVision Multimode Plate Reader (Perkin Elmer). EC₅₀ values were calculated using GraphPad Prism 7.0 (RRID:SCR_002798).

The binding kinetics of hL49 to its target, CD228, were assessed by biolayer interferometry (BLI) on an Octet RED384 instrument (ForteBio). In the monovalent binding assay, hL49 was loaded onto anti-human IgG-Fc (AHC) biosensors (ForteBio) in BLI assay buffer (0.1% BSA, 0.02% Tween 20, 1x PBS, pH 7.4), and CD228 was prepared as 2.5-fold serial dilutions (100–0.41 nmol/L). In the bivalent assay, high precision streptavidin (SAX, ForteBio) biosensors were saturated with 5 μg/mL biotinylated hL49 in BLI assay buffer, and CD228 was prepared as 2.5-fold serial dilutions (50–0.2 nmol/L). Association time was 300 s and dissociation time was 1,200 s. Sensorgrams were generated at 21°C and globally fitted with the 1:1 Langmuir isotherm model (Rmax unlinked) after a reference subtraction of the antigen-loaded 0 nmol/L analyte. Rate constants and K_D values were calculated using ForteBio Data Analysis Software.

A2058 cells were dosed with 2 μg/mL SGN-CD228A and pre-incubated on ice for 30 minutes. SGN-CD228A–treated cells were washed, then immediately fixed and permeabilized (0-hour time point) or further incubated at 37°C for the 4-hour time point. Cells were stained with an anti-human antibody, anti-lysosomal associated membrane protein-1 (LAMP-1) antibody (#562622; BD), and Hoechst. Images were collected using IN Cell Analyzer 2200 (GE Healthcare). In another instance, cells were incubated with hL49 conjugated to 8 copies of AF647 and 2 copies of the TQ5WS quencher on ice for 30 minutes. Cells were washed and incubated at 37°C in media. Images were collected every hour. The amount of AF647 liberated from the antibody and quencher was quantified (total pixel intensity) and normalized to the number of cells (Hoechst positive) for each time point.

Cells were resuspended at 2.5×10⁵/mL and treated with 20 ng/mL ADCs (SGN-CD228A or hL49-vc-MMAE) for 24 hours. The culture supernatant was saved for mass spectrometry (MS) analysis. Adherent cells were trypsinized and combined with pelleted nonadherent cells. An aliquot was removed for enumeration and determination of average cell diameters (Vi-Cell XR2.03, Beckman Coulter). Untreated cell pellets were used for standard curves, spiking with varying MMAE concentrations. Both spiked and treated cell pellets were mixed with internal standard and then fully suspended by vortexing in ice-cold MeOH followed by placement at −20°C (15 minutes) and centrifugation (16,000 × g, 5 minutes, 4°C). The methanolic supernatants were mixed with 0.2% formic acid in water. Untreated culture medium samples were spiked with MMAE standards, and all culture media samples were mixed with internal standard and acidified to 0.2% formic acid. All samples were subjected to solid-phase extraction (MCX 96-well plate, Waters), and the recovered eluates were dried and reconstituted in 95% acetonitrile with 0.1% formic acid for quantification by LC/MS-MS. To derive an equation for the quantitation of released drug in the experimental unknown samples, the peak area for each MMAE standard was divided by the peak area obtained for the internal standard. The resultant peak area ratios were plotted as a function of the standard concentration and were fitted to a curve using linear regression. The peak area ratios obtained for MMAE to internal standard in the experimental samples were converted to drug amounts using the derived equation.

Animal handling and experimentation were performed with Institutional Animal Care and Use Committee approval at Seagen, Champions Oncology, and Crown Biosciences, which are fully accredited by the Association and Accreditation of Laboratory Animal Care. In all xenograft studies, no weight loss or treatment-related toxicities were observed for mice treated with any of the test articles.

Briefly, 2.5×10⁵ −1×10⁶ cells were injected subcutaneously into female athymic nude mice (RRID:RGD_5508395). When tumors reached approximately 100 mm³, mice were randomized into study groups and dosed with test articles via intraperitoneal injection. Tumors were measured twice weekly, and tumor volumes were calculated with the formula V = ½ × L × W². Animals were euthanized when tumor volumes reached 500 to 1,000 mm³. Mice showing durable regressions were terminated around day 40 to 66 after implantation.

Patient-derived xenograft (PDX) studies were performed at Champions Oncology and Crown Biosciences. Patient informed consent for specimen collection was obtained through an Institutional Review Board protocol or as part of Champions’ CLIA-certified clinical program for drug testing on a personalized mouse model. To Crown's best knowledge, the Materials were collected by Crown's suppliers and partner collaborators in compliance with applicable local laws, regulations, ethical standards, and informed consent. PDX models were generated by implanting athymic nude mice with tumors from 61 human patients (22 TNBC, 4 melanoma, and 35 NSCLC). When tumors reached 100 to 300 mm³, 2 mice/model were treated with single intraperitoneal dose of SGN-CD228A (3 mg/kg), and percent change in tumor volume for each mouse was calculated at time of best response or 7 days post dose. A similar study with 5 additional melanoma and 2 NSCLC models was performed in NOD/SCID mice (RRID:IMSR_JAX:001303). The LU0697 squamous NSCLC model was grown in BALB/c nude mice (Beijing Anikeeper Biotech Co., Ltd), and the LU5200 adenocarcinoma NSCLC model was grown in NOD/SCID mice (3 animals/group). Mice were randomized and dosed (intraperitoneally) when tumors reached approximately 100 mm³, and studies were terminated 28 days post dosing.

We performed RNA sequencing (RNA-seq) on 22 cell lines. Cell pellets from each cell line were submitted to vendors for RNA extraction, library preparation using poly-A selection, and Illumina sequencing. FASTQ files containing raw sequence reads were analyzed in-house using STAR v2.5.2b (RRID:SCR_004463; ref. 18) to align reads to a combined hg38/mm10 human/mouse genome reference. Gene-level transcripts per million reads (TPM) quantification was performed using RSEM v1.3.0 (19). TPM values were normalized to a sum of 1 million by species for each sample and averaged across one or two biological replicates per cell line to determine a single transcript profile per cell line. Data analyzed in this study were obtained from DepMap 21Q3 Public figshare https://doi.org/10.6084/m9.figshare.15160110.v2, Dataset doi:10.6084/m9.figshare.9201770.v2.

RNA-seq data (log-transformed TPM) from the Cancer Cell Line Encyclopedia (CCLE) were downloaded from the DepMap Portal. Forty-two cell lines from CCLE overlapped those from our analysis. In total, 49 of 50 cell lines had CD288 RNA-seq quantification from either CCLE or internal profiling, with 15 represented in both. To arrive at a final quantification for each cell line that accounts for batch differences between CCLE and our internal workflow, we normalized across both datasets using a rank-based percentile approach. The CCLE and Seagen CD228 TPM values were replaced by their percentile relative to the other samples in the same dataset, and then the percentiles were averaged for the cell lines that overlapped between datasets. The linear model predicting SGN-CD228A sensitivity with dependent variables MMAE sensitivity, and normalized CD228 mRNA expression (described above) was fit using function lm from R v3.6.0. In this model, the coefficients for both dependent variables are significantly different from zero, with P < 10⁻⁶ for both.

For correlation analysis in PDX models, % tumor growth inhibition (TGI) was calculated as 100×(1-(Tf-Ti)_treated/(Tf-Ti)_untreated), where Tf is tumor volume on the day the first animal was sacrificed, and Ti is tumor volume at dosing. Mean %TGI (typically mean of 2 animals) was used in the statistical model. Percent tumor volume change = (100×((Tn-Ti)/Ti), where Tn is the smallest tumor volume measured). Low and high CD228 expression groups were based on RNA level < 5.20 log₂(TPM+0.1), representing the highest 1/3 of CD228-expressing NSCLC models and top half of CD228-expressing TNBC models. Linear correlation was analyzed using Pearson's product-moment correlation option with x = log₂(TPM) and y = %TGI via cor.test in R v3.6.0.

Data generated in this study are available within the article and its Supplementary Data files or from the corresponding author upon reasonable request.

CD228 is highly expressed in melanoma patient samples (15) and has recently been described as colorectal cancer biomarker (5, 20). Analysis of CD228 RNA levels in The Cancer Genome Atlas (TCGA) demonstrated broad CD228 expression in many tumors (Fig. 1A; Supplementary Table S1), with highest median CD228 mRNA expression in skin cutaneous melanoma. We examined breast tumors based on hormone receptor (HR)/HER2 status (Fig. 1B). While a large range of expression was observed across all three subtypes, CD228 was expressed at a higher level in TNBC than HER2+ or HR+/HER2- cancers.

Analysis of matched RNA and proteomic data (21) from TCGA breast cancer samples indicated high correlation between CD228 RNA and protein levels (Fig. 1C). We tested a large collection of tumor microarrays by IHC using a commercial polyclonal antibody, which further confirmed the broad CD228 expression (Fig. 1D). Given the extensive characterization of CD228 expression in melanoma, we used melanoma to identify an RNA cutoff that was likely to predict IHC positivity. Using this value, we found good agreement between positivity on IHC microarrays and RNA expression by tumor type (Fig. 1D). As further validation, we stained tumor sections across different tumor types with a novel, IHC-optimized CD228 antibody. The results showed high CD228 positivity in patients with pancreatic adenocarcinoma, mesothelioma, and melanoma and higher positivity in squamous lung cancer than the adeno subtype (Fig. 1E; representative images in Fig. 1F). Consistent with the literature, IHC assessment of normal CD228 expression showed detectable protein levels in a limited number of tissues, namely, outer plexiform layer of the retina, brain endothelium, and juxtaglomerular apparatus and medullary tubules of the kidney (Supplementary Fig. S5A and S5B). However, RNA in situ hybridization did not identify any signal in the juxtaglomerular apparatus or brain endothelial cells (Supplementary Fig. S5C). Similar studies in cynomolgus monkey further confirmed the restricted normal tissue expression profile of CD228, and preclinical toxicology studies resulted in a safety profile consistent with that of other MMAE-conjugated ADCs (22–24) and a non-targeted version with the same linker and drug–antibody ratio (DAR; ref. 25). Overall, the limited presence of CD228 in normal tissues and its broad expression in solid tumors indicate that this antigen is a suitable ADC target.

ADCs comprise monoclonal antibody, linker, and cytotoxic drug or payload. In this study, MMAE, a synthetic analog of the naturally occurring microtubule-disrupting compound dolastatin-10 (26), was chemically conjugated to the antibody backbone via a PEGylated glucuronide linker incorporating a self-stabilizing maleimide (mDPR; Fig. 2A; ref. 16). Conjugation takes place at the cysteine residues that make up the interchain disulfide bonds of the antibody, resulting in DAR of 8 (Supplementary Fig. S1; ref. 16). Upon ADC internalization, lysosomal β-glucuronidase cleavage of the glucuronide portion of the linker releases free MMAE for tubulin binding and induction of G₂–M cell-cycle arrest and apoptosis (16, 27). L49 was investigated as the antibody backbone of a CD228-targeting ADC. L49 bound strongly to control SK-MEL-5 and SK-MEL-28 cells endogenously expressing CD228, while binding was greatly reduced in CRISPR/CAS9 knockout cells (Fig. 2B). Conversely, L49 did not bind RPMI-7951 cells, which lack endogenous CD228 expression, but recognized CD228-overexpressing RPMI-7951 cells. Consistent results were obtained when the mouse antibody (mL49) and a human-mouse chimeric variant (cL49) were conjugated to MMAE via the glucuronide linker and tested for cytotoxic activity against the above engineered cell lines (Fig. 2C), confirming specificity of L49 to human CD228. These L49 ADCs also showed strong cytotoxicity against melanoma A2058 cells (∼50,000 CD228 receptors/cell, Supplementary Table S2) compared with other commercially available anti-CD228 antibodies conjugated to the same drug-linker (Fig. 2D). cL49 was also used as an ADC to compare the PEGylated glucuronide MMAE drug-linker to vc-MMAE and mc-MMAF in a melanoma cell panel with a range of CD228 expression. Consistent with published data, vc-MMAE and mc-MMAF ADCs were only active in cells with high CD228 copy numbers, whereas cL49 paired with the DAR8 PEGylated glucuronide MMAE linker was active in cells with lower expression (Supplementary Table S2). L49 was humanized (hL49) by grafting the specificity-determining residues onto a human immunoglobulin backbone. The framework residues that differed between the mouse and human variable and heavy chain sequences were optimized for antigen binding in competition and saturation experiments (Supplementary Fig. S6). The hL49 HALC variant was selected for further evaluation, and its affinity to CD228 was confirmed using BLI, a label-free technique that enables the direct measurement of molecular interactions. hL49 HALC yielded K_D of 6.9 nmol/L (Fig. 2E). Together, these results demonstrate that hL49 is highly specific for CD228, has low nM affinity, and is a superior ADC compared with other CD228-directed antibodies. Therefore, hL49 was conjugated to the PEGylated glucuronide MMAE drug-linker to generate the ADC SGN-CD228A.

The intracellular delivery of cytotoxic payload by ADCs involves antigen binding on target cells, internalization, and trafficking to lysosomes. Consistent with this MOA, SGN-CD228A was detected on the plasma membrane of A2058 cells following incubation on ice to allow ADC binding (Fig. 3A, left). After a 4-hour incubation at 37°C, the majority of SGN-CD228A internalized and colocalized with LAMP-1 staining vesicles (Fig. 3A, right), indicating trafficking to lysosomes (white arrows). ADC cell-surface binding and intracellular trafficking kinetics were further assessed by an orthogonal assay in which hL49 was conjugated to a dye and quencher pair using the glucuronide portion of the linker as used in SGN-CD228A. Cells with > 15,000 CD228 copies/cell generally had moderate to high total fluorescence intensity that correlated with CD228 expression levels (Fig. 3B; Supplementary Table S3). In contrast, CD228-negative cells, such as WM115, had minimal fluorescence signal. Next, CD228 cell-surface expression was determined by qFACS in a panel of carcinoma cell lines [>15,000 copies/cell in 42 of 53 cell lines tested (Fig. 3C; Supplementary Table S3)]. Similar to patient RNA-seq and IHC data, the qFACS panel showed a range of expression and was used to evaluate parameters associated with ADC activity, namely, CD228 RNA levels, CD228 surface levels and internalization, and cytotoxic sensitivity to MMAE and SGN-CD228A (Supplementary Table S3). Because traditional metrics of cytotoxicity (such as IC₅₀) depend on cell division rates, in this analysis of diverse cell lines we used growth rate (GR) metrics (28). GR metrics are based on GR inhibition over the course of the assay and are insensitive to cell division number. Sensitivity to MMAE and SGN-CD228A was measured as area over the dose response curve (GR_AOC). Correlation matrix of all factors showed that CD228 expression (both RNA and cell surface) and internalization are highly correlated to each other (Fig. 3D). Of these 3 variables, CD228 RNA expression was the best predictor of SGN-CD228A sensitivity (r = 0.65; Fig. 3D; Supplementary Table S4). Interestingly, compared with CD228 expression metrics, MMAE sensitivity (GR_AOC MMAE) was similarly correlated to SGN-CD228A sensitivity (r = 0.52; Fig. 3D) but not to CD228 expression (r = 0.04). When MMAE sensitivity was combined with CD228 RNA expression in a linear model, the ability to predict SGN-CD228A efficacy was considerably improved compared with either variable alone (R² = 0.671 vs. 0.482 or 0.271; Fig. 3E). Taken together, these results suggest that sensitivity to MMAE is an equally important indicator of ADC activity as CD228 expression and internalization.

The in vivo antitumor activity of SGN-CD228A was evaluated in cell line–derived xenograft (CDX) models implanted with melanoma tumor cell lines Colo-853 (Fig. 4A) or A2058 (Fig. 4B). In the Colo-853 model, a single intraperitoneal dose of 0.33, 1, or 3 mg/kg SGN-CD228A produced 1/8, 5/8, and 8/8 durable remissions, respectively. In the A2058 model, 1 mg/kg SGN-CD228A produced tumor growth delay and 3 mg/kg produced durable remissions in 8/8 mice. In both models, the non-binding control antibody conjugated to the same drug-linker (hIgG1-ADC) had minimal antitumor activity. To gain a comprehensive understanding of the antitumor activity of SGN-CD228A across multiple tumor types, %TGI by a single 1 mg/kg intraperitoneal dose was determined in 6 additional CDX models (Fig. 4C). Over 70% TGI was observed in 5/6 models with >32,000 CD228 copies/cell, while the antitumor activity was lower in models with <24,000 copies/cell (Fig. 4C). Finally, the efficacy of a single dose of 1 or 3 mg/kg SGN-CD228A was categorized on the basis of response rate (Fig. 4D), which showed complete response (CR) rate of 64% with 3 mg/kg SGN-CD228A and 29% with 1 mg/kg SGN-CD228A. These findings demonstrate that the in vivo antitumor activity of SGN-CD228A is dose-dependent and results in CR in several cancer models, especially in those with higher expression.

Considering the promising results observed with SGN-CD228A in CDX models, we next investigated the effect of the linker on ADC efficacy. Athymic nude mice implanted with A2058 cells were administered SGN-CD228A (DAR8 PEGylated glucuronide MMAE) or hL49-vc-MMAE, which consists of hL49 conjugated to an average of 4 molecules of MMAE via the Val-Cit dipeptide linker (Supplementary Fig. S7A). A single dose of 1 or 3 mg/kg SGN-CD228A produced 4/5 and 5/5 CRs, respectively, whereas the same concentrations of hL49-vc-MMAE dosed weekly for 3 total doses produced no durable remissions in mice (Fig. 5A). Similarly, in the Colo-853 model, 3/5 or 5/5 CRs were observed after treatment with 1 or 3 mg/kg SGN-CD228A (Fig. 5B). To further understand the differences between the linkers, in vitro cytotoxicity studies were performed with SGN-CD228A variants in which hL49 was conjugated to 4 or 8 MMAE molecules via the glucuronide linker in the absence of mDPR and PEG₁₂, enabling a direct comparison to the Val-Cit linker (Fig. 5C; Supplementary Fig. S7B). Comparing the DAR4 ADCs, the glucuronide linker showed improved cytotoxic activity over the Val-Cit linker, especially in cells with lower cell-surface CD228 expression. Additional improvement in cytotoxicity was observed with DAR8 compared with DAR4 glucuronide linker, although the extent of improvement was limited in the higher antigen-expressing cell lines. Furthermore, similar levels of internalization were observed between the dipeptide and glucuronide linkers (Supplementary Fig. S8), but differences emerged when released intracellular and extracellular MMAE was measured. Despite the glucuronide linker having twice as much drug per antibody, the two linkers released approximately the same amount of total MMAE over 24 hours (Fig. 5D, left). However, intracellular MMAE levels more than doubled in cells treated with the glucuronide linker (∼3-fold in SK-MEL-5 cells and ∼8-fold in A2058 cells), suggesting the overall intracellular retention of MMAE was greater with the glucuronide linker than the dipeptide linker (Fig. 5D, middle and right panels). These findings indicate that CD228 is a promising ADC target for the PEGylated self-stabilizing glucuronide linker technology, likely because it allows more MMAE to accumulate within cells on a per-receptor basis than the dipeptide linker technology.

Antitumor activity of SGN-CD228A was further evaluated in PDX models spanning multiple tumor types. In the squamous NSCLC model, a single dose of 1 or 3 mg/kg SGN-CD228A resulted in rapid tumor regression, producing durable remissions in 1/3 and 3/3 mice, respectively (Fig. 6A, left). In the adenocarcinoma model, while none of the tumors were eliminated, mean tumor volume 21 days post dosing was 20 mm³ in the 3 mg/kg SGN-CD228A group compared with 652 mm³ in the untreated group (Fig. 6A, right). To obtain a broad sense of activity across multiple tumor types, a single dose of SGN-CD228A (3 mg/kg) was tested in a mouse PDX trial using 70 tumor models (9 melanoma, 39 NSCLC, and 22 TNBC). CD228 expression in the PDX samples was variable (similar to that in primary patients; Fig. 1) and highly correlated between RNA and protein (Supplementary Fig. S9). The PDX models were divided into low and high CD228 RNA expression groups (Fig. 6B), and overall response rate (ORR, >30% tumor shrinkage based on RECIST criteria) was calculated. Consistent with the CDX study findings, SGN-CD228A had strong antitumor activity in melanoma PDX models (56% ORR), all of which were considered high CD228 expressors. Similarly, SGN-CD228A produced a high response rate in TNBC (59% ORR, all models), which improved to 78% ORR in high CD228-expressing models (n = 9) compared with 46% ORR (n = 13) in low-expressing models. In contrast, SGN-CD228A was less active in NSCLC (26% ORR). However, in high CD228-expressing models, the ORR improved to 53% (n = 15) compared with 8% (n = 24) in low-expressing models. Comparison of CD228 RNA expression and response to SGN-CD228A, as measured by mean %TGI, were significantly correlated in NSCLC only (Fig. 6C).

Here, we describe the preclinical profile of SGN-CD228A, an investigational, CD228-directed, MMAE-linked ADC investigated in a Phase I clinical trial (NCT04042480) for safety, tolerability, and efficacy in patients with advanced solid tumors. CD228 is a promising ADC target in part due to its high expression on tumor tissues and low expression on normal tissues. RNA-seq and IHC measurements consistently showed elevated CD228 RNA and protein levels in multiple tumor types, suggesting that CD228 is an excellent target for delivery of cytotoxic payloads. CD228 overexpression has been reported to increase cellular proliferation (29, 30), whereas CD228 knockdown suppresses growth and metastasis (31). Migration of human microvascular endothelial cells and SK-MEL-28 cancer cells was increased with CD228 upregulation but inhibited by the CD228-directed antibody cL235 (32). These results indicate not only that CD228 could play a role in angiogenesis and metastasis but also that a therapeutic directed against CD228 could inhibit its activity. Together, these observations suggest that targeting CD228 with an ADC may be a potential strategy to optimize efficacy by delivering a cytotoxic payload as well as inhibiting key aspects of tumor biology.

The relationship between ADC target expression and tumor response to treatment has been a major focus in the ADC field. Overall, there has not been a strong correlation between the two factors. Generally, target expression itself has not proven to predict response, and patients with little to no detectable target expression by IHC can have clinical responses. These observations suggest that ADC efficacy is dependent on several factors, such as antibody binding affinity and internalization, ADC toxicity, and properties of the payload including permeability and mechanism of cytotoxicity. Here, in a large panel of cell lines spanning multiple tumor types, a linear model incorporating both CD228 RNA levels and MMAE sensitivity was a better predictor of SGN-CD228A in vitro activity than CD228 levels alone. Similar results have been reported for other ADCs (33–35), suggesting that intrinsic sensitivity to the cytotoxic drug is an important parameter of ADC responses. While a comparable analysis was not applied to the in vivo experiments in this study, given the lack of strong correlation with CD228 expression, it is likely that MMAE sensitivity, as well as other factors present in the tumor microenvironment (36), contributes to ADC activity in vivo. Therefore, consistent with results with other cancer therapeutics (37), it is likely that the ADC payload is as important for efficacy as the ADC target, and more studies are needed to better understand the mechanisms of payload sensitivity.

L49 was selected as the antibody backbone of SGN-CD228A due to its high specificity to CD228 and its ability to outperform other anti-CD228 antibodies as an ADC. The Val-Cit dipeptide is an effective ADC linker that is utilized in 4 clinically approved ADCs (brentuximab vedotin, polatuzumab vedotin, enfortumab vedotin, and tisotumab vedotin; ref. 38) and has shown preclinical activity with CD228-directed vc-MMAE conjugates; however, in comparison, the PEGylated glucuronide linker showed greater preclinical activity (Fig. 5A and B; ref. 15). The clinical relationship between target expression and ADC response has not typically been as clear as the preclinical observations. However, vedotin ADCs are currently approved without the need for target selection in indications where there is enrichment of patients with relatively high tumor target expression. CD228 has a wide distribution of expression, and therefore, we hypothesize that it may require an optimal linker/payload combination to be effective in patients with lower expression. In SGN-CD228A, the incorporation of a self-stabilizing maleimide in the glucuronide linker results in enhanced intratumoral drug delivery, and the long PEG₁₂ chain helps minimize plasma clearance, providing a wider therapeutic window in preclinical models compared with faster clearing conjugates containing no PEG or shorter PEGs (16, 17). This novel drug-linker also allows for stable and homogeneous drug loading and the generation of highly potent DAR8 conjugates. In our study, a single 3 mg/kg dose of SGN-CD228A (DAR8) produced durable remissions in melanoma CDX models, while three doses of the same concentration of a DAR4 CD228-directed ADC containing a Val-Cit linker did not (Fig. 5A and B). In vitro, the antitumor effect of the glucuronide drug-linker was particularly pronounced in cell lines with lower CD228 expression and is likely related to increased intracellular retention of MMAE (Fig. 5C and D), although the mechanism behind this is still unknown. Similar observations have been reported for the HER2-directed ADCs ado-trastuzumab emtansine (T-DM1; ref. 39) and trastuzumab deruxtecan (T-DXd). Compared with T-DM1, T-DXd has a higher DAR (3–4 vs. ∼8), allowing efficient delivery to HER2-expressing tumor cells, increased antitumor activity in HER2-low expressing PDX models (40), and improved efficacy in patients with HER2-low breast cancer (41). Future efforts directed at improved linkers that increase intracellular drug concentration and retention are needed. Even though MMAE is very potent, ADC delivery presents an opportunity to modify intracellular accumulation, and the modular nature of ADCs enables the development of optimal therapeutics through selection of the right linker and payload for the right target. Overall, the findings in this study show that the glucuronide linker is a promising alternative to the Val-Cit dipeptide and should be included in the repertoire of drug-linkers for the generation of potent ADC therapeutics. Data from the clinical study of SGN-CD228A (NCT04042480) will enable a better understanding of how the linker and DAR impact the therapeutic window for MMAE ADCs.

L. Westendorf reports a patent for US2020/0246479A1 pending. T.S. Lewis reports other support from Seagen Inc. outside the submitted work; in addition, T.S. Lewis has a patent for US2020/0246479A1 pending. S. Sandall reports a patent for US 2020/0246479 pending to Seagen, Inc. No disclosures were reported by the other authors.

R. Mazahreh: Conceptualization, resources, formal analysis, validation, investigation, writing–original draft, writing–review and editing. M.L. Mason: Conceptualization, data curation, formal analysis, validation, investigation, methodology, project administration, writing–review and editing. J.J. Gosink: Data curation, formal analysis, writing–review and editing. D.J. Olson: Resources, data curation, investigation, writing–review and editing. R. Thurman: Resources, data curation, formal analysis, investigation, methodology, writing–review and editing. C. Hale: Data curation, formal analysis, visualization. L. Westendorf: Resources, writing–review and editing. T.A. Pires: Investigation, writing–original draft, writing–review and editing. C.I. Leiske: Resources, writing–review and editing. M. Carlson: Formal analysis, investigation, methodology. L.T. Nguyen: Conceptualization, resources, writing–review and editing. J.H. Cochran: Investigation. N.M. Okeley: Investigation, writing–review and editing. R. Yumul: Investigation, writing–review and editing. S. Jin: Investigation, writing–review and editing. I.J. Stone: Investigation, writing–review and editing. D. Sahetya: Investigation, writing–original draft. A. Nesterova: Investigation, writing–review and editing. S. Allred: Investigation, writing–review and editing. K.M. Hensley: Formal analysis, supervision, writing–review and editing. R. Hu: Investigation. R. Lawrence: Resources, supervision, investigation. T.S. Lewis: Conceptualization, supervision. S. Sandall: Conceptualization, formal analysis, supervision, investigation, writing–original draft, writing–review and editing.

We thank Esther Trueblood for assistance with IHC. We thank Iliyana Mikell for providing help with data presentation, writing support, and editing support. We thank Christine O'Day for thoughtful discussions about experimental results. This work was funded by Seagen Inc.