Preclinical Antitumor Efficacy of BAY 1129980—a Novel Auristatin-Based Anti-C4.4A (LYPD3) Antibody–Drug Conjugate for the Treatment of Non–Small Cell Lung Cancer

C4.4A (LYPD3) is a glycosylphosphatidylinositol (GPI)-anchored, highly glycosylated cell surface protein that has been shown to be upregulated in migrating keratinocytes during wound healing (1). It was first described as a metastasis-associated cell surface protein in rat pancreatic tumor cells (2) and since then, has been associated with carcinogenesis in several different cancers (3–5). In cancer, C4.4A has been suggested to be involved specifically in tumor cell invasion via interaction with the extracellular matrix (2, 6).

C4.4A is strongly overexpressed in non–small cell lung cancer (NSCLC) with preferential expression in squamous cell carcinoma (SCC) subtype compared with the other two most common NSCLC subtypes; adenocarcinoma (AC) and large-cell carcinoma (LCC) (4, 5, 7, 8). Lung cancer is the most frequently diagnosed cancer with an estimated 1.8 million new cases in 2012 (9). NSCLC accounts for 85% of all lung cancers and SCC, being the second most frequent histologic subtype occurring in 30% of NSCLC cases, explains approximately 400,000 deaths per year worldwide (10).

Overexpression of C4.4A has also been detected in SCC of the head and neck (HNSCC) including esophageal SCC (ESCC) subtype (3, 11). On the transcriptional level, approximately 50% of primary lung cancers and 75% of lung cancer metastases express C4.4A mRNA, whereas no expression has been detected in normal lung tissue (12, 13). In addition, C4.4A is expressed in colorectal (6, 14, 15) and breast cancer (16).

Overexpression of C4.4A has been shown to correlate with a malignant phenotype and poor prognosis in NSCLC (5, 8), colorectal cancer (15) and ESCC (11), while in breast cancer, it is associated with a good prognosis (16). Under normal physiologic conditions, C4.4A expression is limited to skin keratinocytes as well as esophageal endothelial cells and placental cells (1, 12). The recently published C4.4A knock-out mice are born viable, no gross abnormalities are observed, and surprisingly, also the epidermal development is normal (17). Therefore, targeting C4.4A with a specific antibody–drug conjugate (ADC) represents a unique opportunity to selectively treat C4.4A-positive tumors with high unmet medical need.

Here, we report the development and preclinical evaluation of BAY 1129980 (C4.4A-ADC), a C4.4A-targeting human immunoglobulin G1 (hIgG1) antibody (C4.4A-Ab) conjugated via cysteine side chains and a noncleavable alkyl hydrazide linker to a novel, highly potent microtubule-disrupting auristatin W derivative (18–20). The C4.4A-ADC is efficacious both in vitro and in vivo in a variety of cell line and patient-derived xenograft (PDX) models representing various C4.4A-expressing cancers with high efficacy in NSCLC models.

Cell lines were acquired from American Tissue Culture Collection (ATCC) unless otherwise noted and cultured according to the provider's instructions. The cell lines used were: A549 (CCL-185), lung adenocarcinoma; NCI-H292 (CRL-1848), lung mucoepidermoid carcinoma; NCI-H322 (#95111734, Sigma-Aldrich), human bronchioalveolar carcinoma; SCC-4 (CRL-1624) and SCC-9 (CRL-1629), both head and neck (H&N) cancer cell lines; FaDu (HTB-43), H&N nasopharyngeal SCC; SCaBER (HTB-3), urinary bladder squamous carcinoma; HCT-116 (CCL-247), colon carcinoma; BxPC3 (CRL-1687), pancreatic adenocarcinoma; A431 (CRL-2592), skin epidermoid carcinoma; MCF-7 (HTB-22), mammary adenocarcinoma; CHO (#85050302, ECACC), Chinese hamster ovary cells; HEK293 6E (21), human embryonic kidney cells, and LLC-PK1 and L-MDR1 cells (both obtained from Prof. A. H. Schinkel at Netherlands Cancer Institute, Amsterdam, the Netherlands). All cell lines were obtained between 2002 and 2013 and authenticated using short tandem repeat DNA fingerprinting at Leibniz Institute DSMZ-German Collection of Microorganisms and Cell Cultures (DSMZ) before the experiments.

A549 cells (no endogenous C4.4A expression) were stably transfected with pCMV UCOE8 Dest vector (EMD Millipore) containing the full-length human C4.4A encoding cDNA sequence or a mock vector pCMV UCOE8 Dest using the FuGENE HD transfection reagent (#04709691001, Roche) to produce hC4.4A:A549 and mock:A549 lung cancer cell lines, respectively. C4.4A-overexpressing cells were selected using puromycin (1–10 μg/mL) and identified by flow cytometry. CHO cells expressing full-length human or murine C4.4A were generated correspondingly.

Recombinant soluble human, murine, and cynomolgus monkey C4.4A orthologues were generated by cloning the cDNA of the respective species-specific C4.4A extracellular domains into mammalian expression vectors based on pTT5 (NRC Canada) constructs with a C-terminal hexahistidine tag. Vectors were transfected into HEK293 6E cells (21) and expression was performed for 5 days at 37° C and 5% CO₂ in F17 Medium (ThermoFisher). Recombinant proteins were purified from supernatants by His-tag affinity chromatography (IMAC) on NiNTA superflow resin (Qiagen).

High-quality total RNA extracted from lung tissue collection consisting of 22 NSCLC samples and 18 normal lung tissues adjacent to tumor samples (Charité) (22) were analyzed for gene expression using an Affymetrix HG-U133Plus2.0 DNA-oligonucleotide array as described in the Supplementary Methods. The gene expression data have been deposited in the ArrayExpress repository with E-MTAB-5231.

Immunohistochemistry (IHC) for C4.4A was performed in NSCLC patient samples and in both cell line-derived xenograft (CDX) and PDX samples. Tissue microarray (TMA) blocks containing paraffin-embedded tumor biopsies from patients representing different NSCLC subtypes were obtained from Asterand UK Ltd. Fresh-frozen samples (hC4.4A:A549, mock:A549, NCI-H292, NCI-H322 and FaDu xenograft tumors) and formalin-fixed paraffin-embedded (FFPE) samples from CDX tumors (NCI-H292, NCI-H322, SCC-4, FaDu, SCaBER) were generated in-house as described in Supplementary Methods. PDX samples were obtained as FFPE sections on slides from EPO GmbH.

IHC for C4.4A was performed either manually (for fresh-frozen samples) or using an autostainer (for FFPE samples) as described in the Supplementary Methods. Stained IHC samples were evaluated by a trained pathologist (Provitro GmbH or Ventana). The percentage of tumor cells staining positive for C4.4A at each intensity level (0, 1, 2, or 3) was calculated and an H-score was calculated as described previously (23). An intensity score was also calculated for each sample representing the highest intensity (on a scale of 0 to 3+) with staining at that level or higher in at least 30% of tumor cells (membrane staining only).

The C4.4A-Ab (BAY 1112623) was generated by phage display panning using the n-CoDeR library of BioInvent International AB (24) as described in the Supplementary Methods and previously (24–27). All positions in the complementarity-determining regions (CDR) of the C4.4A-Ab not required for antigen binding were reverted to human germline sequences as previously described (27). Binding of C4.4A-Ab to recombinant C4.4A protein, as well as its affinity constants were determined using immunoblotting, surface plasmon resonance, and flow cytometry as described in the Supplementary Methods. The antibody-binding sites on cell lines were determined with the Quantum Simply Cellular Kit (#816, Bangs Laboratories) according to the manufacturer's instructions. Internalization studies were performed with CypHer5E fluorescent-labeled C4.4A antibodies as described previously (28) using hC4.4A:A549, NCI-H292, BxPC3 and FaDu cells with different C4.4A receptor levels. Cells were treated with various concentrations (7–34 nmol/L) of labeled C4.4A-Ab and an isotype control antibody (BAY 1059506, Bayer Pharma AG) and internalization was kinetically measured using the INCell Analyzer 1000 (GE Healthcare). Antibody internalization was analyzed by determining granule count and granule intensity per cell via microscopy. Mock:A549 and wild-type A549 cells were used as negative controls. Intracellular trafficking of C4.4A-Ab was examined in the same cell lines as described in the Supplementary Methods.

In addition, to confirm the internalization of the C4.4A-Ab, a cytotoxicity experiment using Hum-ZAP System (ATS Bio) was used as described in Supplementary Methods.

An overview of all the antibodies used is provided as Supplementary Table S1.

The synthesis of C4.4A-ADC (BAY 1129980) from an auristatin W derivative and C4.4A-Ab is described in Supplementary Methods and in Supplementary Fig. S1A. The drug-to-antibody ratio (DAR) and purity of the ADC were determined by reversed-phase HPLC analysis and size exclusion chromatography coupled with multiangle light scattering (SEC-MALS), respectively, as described in Supplementary Methods.

The identification and quantification of active metabolites formed after intracellular degradation of the C4.4A-ADC was based on synthetic reference compounds. The metabolites were determined in previously mentioned cell lines in vitro and as described in Supplementary Methods. The biological profile of the observed metabolites, including activity in a tubulin-inhibition assay and cellular transport properties were determined as described in Supplementary Methods.

To determine IC₅₀ of cell viability cells were treated with 0.01–100 nmol/L of C4.4A-ADC or C4.4A-Ab, or 0.03–300 nmol/L of vinorelbine (2570983, Hexal), or paclitaxel (T-1912, Sigma) and cell viability was determined at 72 hours using the CellTiter-Glo Luminescent Cell Viability Assay (#G7573 and #G7571, Promega). The IC₅₀ was determined using GraphPad Prism(GraphPad Software, Inc) or Microsoft Excel (Microsoft).

All animal experiments were conducted in accordance with the German animal welfare law and approved by local authorities.

For the cell line-based subcutaneous tumor models, 1–5 × 10⁶ NCI-H292 or NCI-H322 lung cancer cells, SCC-4 or FaDu H&N SCC cells, or SCaBER urinary bladder SCC cells were suspended in 50%–100% Matrigel (Basement Membrane Matrix, BD Biosciences) and injected subcutaneously to the left flank of female NMRI nu/nu mice (24–26 g, 8–10 weeks, Taconic). Studies with the six NSCLC PDX models (Lu7064, Lu7126, Lu7343, Lu7433, Lu7466 and Lu7700) were performed at EPO GmbH (29). Tumor fragments obtained from in vivo passage on mice were implanted subcutaneously in the inguinal region of male NMRI nu/nu mice (Taconic). Tumor volume (0.5 × length × width²) and body weight was determined at least twice weekly. Mice were allocated to different treatment and control groups (n = 8–10/group) by stratified randomization based on their primary tumor size (median tumor volume of approximately 50–150 mm³). Unless otherwise indicated, C4.4A-ADC and the control ADC (BAY 1132711, Bayer Pharma AG; Supplementary Table S1), which contains the same linker payload and conjugation via cysteine residues to the antibody, were administered intravenously (i.v.) every fourth day and repeated for a total of three times (Q4D×3).

Final tumor weight was determined at the end of the study. Treatment response was defined using the RECIST criteria (30). Progressive disease (PD) was defined as greater than 20% increase in tumor size. Partial response (PR) was defined as greater than 30% reduction in tumor size and complete response (CR) as an absence of any palpable and visible tumor mass at study end. Stable disease (SD) was defined as a reduction of tumor volume <30% or an increase of tumor volume ≤20%. Minimum effective dose (MED) was defined as lowest dose resulting in reduction of tumor size by more than 50%. Treatment to control ratios (T/C) were calculated based on mean tumor volumes.

In the three experiments using NCI-H292 model, that is, the experiments with one treatment cycle, two treatment cycles, or combination with cisplatin, treatments were started on day 6 (mean tumor size of 67 mm³), day 8 (148 mm³), or day 6 (62 mm³) after tumor cell inoculation, respectively. For SCC-4, NCI-H322, FaDu, and SCaBER models, treatments were started on day 4 (79 mm³), day 4 (124 mm³), day 5 (65 mm³), or day 6 (93 mm³) after tumor cell inoculation, respectively. For the NSCLC PDX models, treatments were started on day 7 (Lu7466, Lu7700), day 8 (Lu7064), day 11 (Lu7433), day 12 (Lu7343), or day 18 (Lu7126) after tumor cell inoculation with average tumor sizes of 75–110 mm³. For the NCI-H322 model, C4.4A-ADC and control ADC were administered on days 4, 8, 12, 25, 29, and 33. C4.4A-ADC metabolite concentrations were measured ex vivo in NCI-H292 tumors and plasma as described in Supplementary Methods.

Cisplatin (Sigma, 3 mg/kg, intraperitoneally (i.p.)), paclitaxel (Bristol-Myers Squibb or Fresenius Kabi, 24 mg/kg, i.v.), vinorelbine (Hexal, 2 or 5 mg/kg, i.p.), carboplatin (Hexal, 75 mg/kg, i.v.), and docetaxel (TEVA/Ratiopharm, 12.5 mg/kg, i.v.) were used as standard-of-care (SOC) compounds. Unless otherwise stated, cisplatin was administered every third day for a total of five times (Q3D×5) and paclitaxel once weekly for a total of 5 times (Q7D×5). In the NCI-H292 combination study, cisplatin was administered on days 8, 11, 14, and 20. In the NCI-H322 study, cisplatin was administered on days 4, 7, 11, 25, 28, 32, 35, 39, 42 and 46. In the FaDu study, cisplatin was administered every third day for a total of six times (Q3D×6) and paclitaxel once weekly for a total of 3 times (Q7D×3). Vinorelbine was administered intraperitoneally on day 6 and 20 (NCI-H292) or on days 4, 11, 25, 32, 39 and 46 (NCI-H322). Carboplatin and docetaxel were administered once weekly for a total of three (Q7D×3) and four (Q7D×4) times, respectively. PBS was used as the vehicle control. In the control experiment with the NCI-H292 model, naked C4.4A-Ab was administered at 4 or 10 mg/kg (QD7 × 3) starting on day 7.

The mRNA transcript levels in tissues were analyzed with one-way ANOVA followed by unpaired t tests. For the comparison of tumor volume, the log-transformed data were analyzed using one-way ANOVA and Tukey's HSD test or Kruskal–Wallis test, followed by Dunn's test with Holm–Bonferroni correction. For PDX models, the log- or square root transformed data were analyzed using a linear mixed-effects model and the comparisons were performed using model contrasts. Unless otherwise indicated, statistics were calculated at the last time point at which the vehicle group remained in the experiment. The analyses were performed using R software (version 3.2.2). P values were adjusted for multiple comparisons and values lower than 0.05 were considered statistically significant.

To assess the suitability of C4.4A for antibody-directed drug targeting, the expression of C4.4A was analyzed in a set of tumor samples with DNA microarrays and IHC. A statistically significant overexpression of C4.4A mRNA transcript was observed in NSCLC compared with normal lung tissue adjacent to the tumor, with higher expression in NSCLC-SCC than in NSCLC-AC tissues (Fig. 1A).

IHC analysis of a set of NSCLC tumor samples (n = 87) demonstrated strong C4.4A expression on the surface of tumor cells (Fig. 1B). In summary, 91.2% (31/34) of NSCLC-SCC samples, 35.3% (12/34) of NSCLC-AC samples, and 42.1% (8/19) of samples of other subtypes of NSCLC (mixed histology, large cell, and unspecified) were found to be C4.4A-positive at any staining intensity.

The monoclonal antibody BAY 1112623 (C4.4A-Ab) was selected to constitute the antibody portion of a C4.4A-specific ADC due to its high specificity toward the C4.4A antigen. More specifically, C4.4A-Ab was found to react strongly with full-length human and mouse recombinant C4.4A as well as with the recombinantly expressed C4.4A domain 1 (S1) in an immunoblot analysis following SDS-PAGE (Supplementary Fig. S2). In contrast, C4.4A-Ab failed to detect the C4.4A domain 2 (S2), indicating that the antibody recognizes an epitope within the S1 domain. Notably, binding of C4.4A-Ab to recombinant C4.4A occurred under nonreducing conditions only, stabilized by disulfide bridges. Furthermore, the C4.4A-Ab showed high monovalent binding affinity to human recombinant C4.4A (K_d = 60 nmol/L) and cross-reactivity to recombinant murine (K_d = 120 nmol/L) and cynomolgus monkey (K_d = 34 nmol/L) C4.4A protein.

Binding of C4.4A-Ab to C4.4A was further analyzed by flow cytometry on hC4.4A:A549 cells and cell lines endogenously expressing C4.4A. C4.4A-Ab bound specifically to hC4.4A:A549 cells (EC₅₀ = 2.3 nmol/L) and endogenous C4.4A in NCI-H292 cells (EC₅₀ = 0.04 nmol/L), whereas no binding to mock:A549 cells was observed. Cross-reactive binding to the cellular mouse antigen was shown in CHO cells transfected with murine recombinant C4.4A (EC₅₀ = 0.8 nmol/L; Supplementary Table S2).

Efficacious ADC-mediated cytotoxicity requires that the delivery of the ADC into cells is driven by antigen receptor binding and subsequent internalization. A highly specific, target-dependent, and significant internalization was demonstrated for the fluorescence-labeled C4.4A-Ab, as it was internalized into hC4.4A:A549 cells with an estimated half maximum signal intensity of 55 min, but not into mock:549 cells (Supplementary Fig. S3A–I). Moreover, an isotype control antibody showed only minor internalization after a long exposure (>24 hours; Supplementary Fig. S3A–II). The target-specific internalization of C4.4A-Ab was also confirmed by a Hum-ZAP assay using mock:A549, hC4.4A:A549 and endogenously C4.4A-expressing MCF7 cells (Supplementary Fig. S4). The microscopic analysis of costaining studies demonstrated a strong colocalization of the C4.4A-Ab with markers of the lysosomal (LAMP1 and Rab7 antibodies) and endosomal compartments (Rab5 antibody) in hC4.4A:A549 cells (Supplementary Fig. S3C). In NCI-H292 cells, a strong-to-moderate colocalization of C4.4A-Ab with Rab7 and minor colocalization with Rab5 was observed. In BxPC3 cells a moderate and in FaDu cells a moderate but heterogeneous costaining were detected.

The ability of C4.4A-Ab to selectively bind to C4.4A-positive tissues was investigated by IHC in xenograft tissues derived from hC4.4A:A549 or mock:A549 lung cancer cells grown as tumors on NMRI nu/nu mice. Positive staining was observed only in the hC4.4A:A549 tumor tissue but not in the mock:A549-derived tissue (Supplementary Fig. S5A). Furthermore, C4.4A-Ab staining was evident in cancer cell line–derived tumor models with endogenous C4.4A expression including the NCI-H292 and NCI-H322 cell lines (Supplementary Fig. S5B and S5C).

As the C4.4A-Ab demonstrated the essential features required for the targeting component of an ADC, that is, target antigen specificity and internalization into cells, it was subsequently modified with highly potent N-methyl auristatin W derivatives. BAY 1110086, identified previously among N-carboxyalkyl-N-methyl auristatins as a compound with high cytotoxic potency and reduced efflux properties, was selected and coupled to cysteine residues of the C4.4A-Ab via a noncleavable hydrazide linker (Supplementary Fig. S1A) (31). The C4.4A-ADC BAY 1129980 with an average DAR of n = ∼4 was selected as a candidate for further characterization and preclinical validation (Fig. 2A) (18). Upon internalization, the ADC is colocalized to lysosomes where it is degraded to the active metabolites BAY 1112179 and BAY 1136309 (Supplementary Fig. S3B–II). These metabolites show strong tubulin-inhibition comparable with monomethyl auristatin F (MMAF; Supplementary Fig. S1B and S1C). In the cellular efflux assay the metabolites BAY 1112179 and BAY 1136309 showed a basal-apical transport of 3 nmol/L/s and 3.6 nmol/L/s, respectively, indicating very low efflux out of the cells.

After intravenous administration of 7.5 mg/kg C4.4A-ADC mice bearing NCI-H292 tumors, BAY 1136309 was the main metabolite detected in the tumor with high and long-lasting exposure: maximum concentrations of about 530 μg/L were measured 24 hours after administration (Supplementary Fig. S6). AUC was calculated to be about 59 mg·h/L and terminal half-life was about 36 hours. Concentrations of BAY 1136309 in plasma were below the limit of detection (5 μg/L), indicating that the metabolite is formed in the tumor tissue after proteolytic cleavage of the C4.4A-ADC.

In vitro cytotoxicity of C4.4A-ADC was tested in C4.4A-positive and -negative cancer cell lines. High potency at subnanomolar range (IC₅₀ = 0.05 nmol/L) was observed in hC4.4A:A549 lung cancer cells. Moreover, a remarkable selectivity (over 1,000-fold) compared with mock:A549 cells was observed (Fig. 2B), demonstrating the target dependency of the cytotoxic effect. In cell lines with endogenous C4.4A expression, C4.4A-ADC showed high potency with IC₅₀s at single- to double-digit nanomolar range and even at subnanomolar range (IC₅₀ of 0.6 nmol/L) in NCI-H292 lung cancer cell line (Fig. 2C). As expected, no linear correlation was found between the number of C4.4A-Ab–binding sites on the cell lines and in vitro potency (Supplementary Table S2).

Antitumor efficacy of C4.4A-ADC was tested in the C4.4A-positive NCI-H292 human NSCLC xenograft model in three settings: as monotherapy using one or two treatment cycles, and in combination with cisplatin.

Monotherapy treatment with C4.4A-ADC (Q4D×3) halted tumor growth dose dependently with a minimum effective dose (MED) of 1.9 mg/kg (Fig. 3A and B; Supplementary Table S3). In contrast, both SOCs, cisplatin and paclitaxel, as well as the control ADC failed to inhibit tumor growth at the maximum tolerated dose (MTD). In addition, vinorelbine treatment was markedly less efficacious compared to C4.4A-ADC (P = 0.0001). C4.4A-ADC was well tolerated; no fatalities or body weight loss of over 10% were observed with any of the C4.4A-ADC doses used. Treatment with vinorelbine at 5 mg/kg led to a dramatic drop in body weight and therefore, the second treatment with vinorelbine was postponed until mice regained normal weight.

Efficacy of C4.4A-ADC was further evaluated in NCI-H292 tumors that were allowed to regrow after an initial treatment cycle with C4.4A-ADC. The first treatment cycle with 15 mg/kg C4.4A-ADC (Q4D×3) resulted in a marked delay of tumor growth with a significantly reduced tumor volume, as compared to vehicle, cisplatin, or control ADC (Fig. 3C and D). Notably, the regrown tumors were still sensitive to an additional treatment cycle with C4.4A-ADC, and a reduction in tumor growth compared with vehicle (P = 0.00001), control ADC (P = 0.00711), and cisplatin (P = 0.00063) was observed.

The study to determine the potential of C4.4A-ADC and cisplatin as combination therapy demonstrated that both monotherapy and combination therapy reduced NCI-H292 tumor volumes significantly (both P < 0.001), whereas cisplatin alone, dosed at MTD, had no significant (P = 0.11783) effect on tumor growth on day 19, the last measurement, including all experimental groups (Fig. 3E and F). However, termination of treatment resulted in tumor re-growth in C4.4A-ADC–treated animals and progressive disease in all mice on day 40. Combination of C4.4A-ADC with cisplatin resulted in stable disease in 4/8 and partial response in 4/8 mice on day 19, demonstrating additive antitumor efficacy. In this study, no body weight loss over 10% was observed in the C4.4A-ADC monotherapy group. In cisplatin monotherapy and combination treatment groups, 2/8 and 4/8 of mice, respectively, showed body weight losses of 10% to 20%. However, indicated treatment pause allowed the mice to regain normal body weight (Supplementary Fig. S7). The combination treatment increased body weight loss only marginally compared to cisplatin monotherapy.

The naked C4.4A-Ab was shown to have no in vivo efficacy without the ADC moiety (Supplementary Fig. S8).

In the C4.4A-positive NCI-H322 human lung cancer xenograft model (Fig. 4A and B; Supplementary Table S3; and Supplementary Fig. S9A), C4.4A-ADC administered at 3.75 and 7.5 mg/kg suppressed tumor growth until day 39, whereas control ADC or vinorelbine had no effect. Cisplatin dosed at MTD (3 mg/kg) showed minor but significant antitumor efficacy. The animals in all other groups except the two C4.4A-ADC groups had to be sacrificed between days 39 and 50 due to large tumor volumes (Fig. 4A and B). Due to slow tumor growth in all groups in the beginning of the study, a treatment pause was scheduled for days 12 to 24. C4.4A-ADC was well tolerated with no signs of body weight loss in any of the treatment groups.

In order for the C4.4A-ADC to function, the target cells need to exhibit sufficient C4.4A expression. The hypothesis whether this is already sufficient for ADC activity, was tested with the FaDu human nasopharyngeal SCC model and the human SCaBER urinary bladder SCC model exhibiting medium and high C4.4A expression, respectively. C4.4A-ADC showed no antitumor activity in either model (Supplementary Figs. S9C and S9D and S10A and S10B; Supplementary Table S3).

Finally, the antitumor efficacy of C4.4A-ADC was evaluated in six NSCLC PDX models representing different C4.4A expression levels, as categorized by H-scoring. Two models (Lu7466 and Lu7064) express C4.4A at high levels (H-scores 280 and 240, respectively). Two models (Lu7126 and Lu7433) exhibit medium (H-scores 190 and 100, respectively) and one model (Lu7343) low (H-score 80) C4.4A expression. One model (Lu7700) was C4.4A-negative (H-score 0) (Supplementary Table S3).

Strong antitumor efficacy was observed for C4.4A-ADC at all doses used in the Lu7466 NSCLC-AC model (Fig. 5A and B; Supplementary Fig. S9I). Furthermore, dose-dependent efficacy was observed in the Lu7064 pleomorphic cell carcinoma model (Supplementary Figs. S6A, S6B, S9E), the Lu7126 NSCLC-SCC model (Fig. 5C and D; Supplementary Figs. S9F and S11C) and the Lu7433 NSCLC-SCC model (Supplementary Figs. S9H and S11D–S11F). In these models with high or medium C4.4A expression, tumor growth was inhibited using C4.4A-ADC at the two highest doses tested (7.5 and 15 mg/kg, both Q4D×3). In addition, distinct efficacy was observed on day 29 with the lowest dose (3.75 mg/kg), the last time point when the vehicle group remained within the experiment in the Lu7433 model (Supplementary Fig. S11E). The SOC compounds paclitaxel and carboplatin were active in these four PDX models. In the low C4.4A-expressing Lu7343 NSCLC-SCC model, treatment with C4.4A-ADC led to initial disease stabilization. However, tumors demonstrated progressive growth shortly after the end of the treatment and no significant difference to the vehicle control was observed (Fig. 5E and F). In contrast, both paclitaxel and carboplatin showed activity resulting in disease stabilization. Expectedly, no antitumor efficacy of C4.4A-ADC was evident in the C4.4A-negative Lu7700 NSCLC-AC model (Fig. 5G and H; Supplementary Fig. S9J). Both paclitaxel and carboplatin showed comparable activity in this model with a transient disease stabilization; however, these were followed by tumor progression shortly after the end of the treatment cycles. Docetaxel showed efficacy in the two PDX models (Lu7126 and Lu7466) where it was tested. Both C4.4A-ADC and docetaxel were highly efficacious as monotherapies in the Lu7466 model, but due to the high sensitivity of the models to the single compounds no additive benefits could be observed with the combination treatment. A positive correlation was found between the level of C4.4A expression (presented either as H-score or percentage of cells positive at any intensity) and the in vivo efficacy in the PDX models (Fig. 6), suggesting that higher levels of antigen expression would lead to improved sensitivity to the C4.4A-ADC.

In all PDX models, treatment with C4.4A-ADC was well tolerated without any notable body weight loss in any of doses applied. The treatment with the selected SOCs resulted in no major side effects and no marked body weight loss, with the exception of cisplatin, which led to a reversible body weight loss of approximately 10% in the Lu7433 model.

C4.4A overexpression has been shown to predict increased mortality in NSCLC (5). Our studies confirmed the previous finding (5) of high prevalence of C4.4A expression in NSCLC on both mRNA and protein level. Interestingly, we detected particularly strong expression and high prevalence of C4.4A in the SCC subtype of NSCLC. Previously, C4.4A expression in normal human organs and tissues has been found to be restricted to skin, placenta, and esophagus (1, 3, 12). The tumor-selective expression profile and efficient internalization capability of C4.4A present an intriguing opportunity for ADC-based targeted treatment of C4.4A-positive cancers, such as NSCLC and particularly its SCC subtype.

The C4.4A-specific antibody (C4.4A-Ab) selected for these studies was demonstrated to be selective for and rapidly internalized by the cancer cells and revealed the preferred intracellular trafficking into lysosomes upon binding to C4.4A. We used a fully human C4.4A-Ab coupled via cysteine residues and a noncleavable linker to a novel antimitotic auristatin W derivative to generate a C4.4A-ADC (BAY 1129980) and evaluated its therapeutic potential in vitro and in vivo (18). The microtubule-disrupting auristatin derivative was chosen due to the proven effectiveness of this effector class in the clinic (32). In addition, we confirm our previous findings (31) that N-carboxylalkyl-N-methyl auristatin W derivatives are highly efficacious microtubule-disrupting agents that can be optimized for reduced efflux properties and allow for stable linker attachment.

The linker in C4.4A-ADC was optimized for high stability to prevent nonspecific release of the toxophore and unwanted uptake by target-negative tissues, thereby aiming to minimize the potential side effects of the ADC. Noncleavable linker ADCs have previously been shown to release their toxophores (linker metabolites) only after proteolytic antibody degradation in the lysosomal compartment (33). Prominent examples, such as trastuzumab emtansine (T-DM1) (34), are capable of efficaciously mediating antitumor activity with minimal systemic toxicity. As the employed antibody C4.4A-Ab is a hIgG1, a potential contribution of the antibody Fc part to the mode of action of the ADC still needs to be determined. However, it is unlikely that the targeting antibody would induce antibody-dependent cell-mediated cytotoxicity (ADCC) as the naked antibody did not demonstrate efficacy in our in vivo control experiments.

Under normal physiological conditions, C4.4A is expressed in skin keratinocytes (1). The strong reaction of the anti-CD44v6 bivatuzumab mertansine ADC against CD44v6-positive human skin (35) is one of the few cases reported where microtubule-disrupting agents have shown severe target-dependent toxicity in the clinic. Importantly, in contrast to CD44v6, C4.4A is not expressed in the basal layers of the skin responsible for skin regeneration (1). Nevertheless, to detect any potential target-mediated side effects already at the preclinical stage, a human/mouse cross-reactive antibody was selected as the targeting moiety in C4.4A-ADC. This approach was chosen due to the comparable C4.4A expression patterns between mouse and humans (36). C4.4A-ADC was well tolerated at efficacious doses and with repeated dosing schedules in all experiments. Only reversible skin reddening was observed and exclusively at doses markedly higher than the MED.

In vitro, C4.4A-ADC demonstrated potent anti-proliferative activity with subnanomolar to double-digit IC₅₀s in cell lines endogenously expressing C4.4A. In hC4.4A:A549 cells, a subnanomolar IC₅₀ of 0.05 nmol/L was determined while no effect was observed in mock:A549 cells, indicating high target selectivity and ADC stability in vitro (i.e., no unwanted toxophore release was observed). Our data show that C4.4A receptor-mediated uptake of C4.4A-ADC results in its degradation, toxophore release, and eventually inhibition of tumor cell viability. The anti-proliferative activity of C4.4A-ADC was dependent on C4.4A expression, but there was no linear correlation observed between the number of C4.4A antibody binding sites and ADC potency in vitro. This missing correlation is not due to intrinsic cell line differences in sensitivity to microtubule interfering agents as all responded in the low nanomolar range to treatment with the microtubule directed agents vinorelbine and paclitaxel. Our data suggest that ADC efficacy in cell lines is more likely to depend on internalization efficiency, intracellular trafficking and toxophore release. Altogether, the in vitro studies indicate that the cytotoxic efficacy of the ADC is mainly driven by C4.4A expression and effective lysosomal trafficking.

The marked in vivo efficacy of C4.4A-ADC was demonstrated in CDX and PDX models with homo- and heterogeneous C4.4A expression patterns to optimally mimic the situation in the clinic and using doses and schedules earlier shown to be efficient even in less sensitive models (37, 38). C4.4A-ADC was particularly potent in two medium to high C4.4A-expressing NSCLC xenograft models, NCI-H292 and NCI-H322, resulting in equal or superior efficacy compared to SOC compounds. Interestingly, in NCI-H292 model, C4.4A-ADC was shown to have high potential for additive antitumor effects if combined with cisplatin. Moreover, tumors that regrew after cessation of treatment remained responsive to a second round of C4.4A-ADC, indicating that repeated and effective treatment with C4.4A-ADC could be possible in the clinic. The lack of efficacy with control ADC observed in the NCI-H292 and NCI-H322 models supports the C4.4A-specific antitumor activity of the C4.4A-ADC. However, the observed lack of efficacy in two models with medium (FaDu) and high (SCaBER) C4.4A expression showed that antigen expression in some cases is not sufficient for in vivo efficacy, and that additional parameters such as inefficient trafficking and processing may limit the activity, as observed in vitro e.g., in the FaDu model.

In the PDX models, one treatment cycle with C4.4A-ADC resulted in dose-dependent antitumor effects in the medium-to-high C4.4A-expressing NSCLC PDX models Lu7466, Lu7064, Lu7126, and Lu7433 that represented different histologies and subtypes, including NSCLC-SCC. In Lu7466, complete responses were observed in over 90% of the mice regardless of the dose used. Interestingly, the efficacy of C4.4A-ADC observed in the model derived of a patient resistant to treatment with cisplatin and vinorelbine, Lu7126, suggests the possibility of antitumor response even in a clinical setting following the development of resistance to SOC chemotherapy. In addition to the efficient lysosomal trafficking shown in vitro, the in vivo studies demonstrate the importance of homogenous C4.4A expression. A good correlation between efficacy and high H-score and percentage of C4.4A-expressing cells was observed in the NSCLC PDX models. This is conceivable, as the C4.4A-ADC has a noncleavable linker and delivers non-cell–permeable toxophore metabolites. The weak efficacy observed in the Lu7343 model with lower H-score and lower percentage of cells supports this. C4.4A expression may also serve as a predictive marker for patient selection, but the threshold for response has to be determined during clinical studies.

Multiple ADCs targeting various antigens in several cancer types are currently under evaluation in preclinical and clinical studies (39) and auristatin-based conjugates are utilized in approximately half of the ongoing clinical studies with ADCs (32). However, to our knowledge, there are only a few other ADCs in development for the treatment of NSCLC and none of them target C4.4A, making our approach unique.

In conclusion, C4.4A-ADC is a promising therapeutic candidate for the treatment of C4.4A-expressing cancers, such as NSCLC and particularly its SCC subtype. C4.4A-ADC, in monotherapy or in combination with the current cancer therapies including immune checkpoint inhibitors, could provide new options for treating C4.4A-positive human malignancies with high unmet medical need. A phase I study (NCT02134197) with C4.4A-ADC (BAY 1129980) is currently ongoing.

J. Willuda, H.-G. Lerchen, C. Lange, F. Dittmer, and G. Leder have ownership interest (including patents) in Bayer AG. C. Pena, K. Mclean, H. Apeler, R. Jautelat, and K. Ziegelbauer have ownership interest in Bayer AG. L. Linden, C. Kopitz, and J. Tebbe are co-inventors on a patent. All authors except S. El Sheikh are employees of Bayer AG or Bayer LLC.

The linker payload technology has been licensed from Seattle Genetics. Studies with NSCLC PDX models were performed at EPO GmbH.

Conception and design: J. Willuda, L. Linden, H.-G. Lerchen, B. Stelte-Ludwig, C. Pena, F. Dittmer, R. Beier, S. El Sheikh, H. Apeler, K. Ziegelbauer, B. Kreft

Development of methodology: J. Willuda, L. Linden, H.-G. Lerchen, C. Kopitz, B. Stelte-Ludwig, C. Pena, C. Lange, O. von Ahsen, J. Müller, S. El Sheikh, G. Leder

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Willuda, L. Linden, C. Kopitz, B. Stelte-Ludwig, C. Pena, C. Lange, S. Golfier, C. Kneip, P.E. Carrigan, O. von Ahsen, J. Müller, R. Beier, S. El Sheikh, G. Leder

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Willuda, L. Linden, C. Kopitz, B. Stelte-Ludwig, C. Pena, C. Lange, P.E. Carrigan, J. Schuhmacher, O. von Ahsen, J. Müller, R. Beier, S. El Sheikh, G. Leder, H. Apeler, K. Ziegelbauer

Writing, review, and/or revision of the manuscript: J. Willuda, L. Linden, H.-G. Lerchen, C. Kopitz, B. Stelte-Ludwig, C. Pena, C. Lange, P.E. Carrigan, J. Schuhmacher, O. von Ahsen, J. Müller, F. Dittmer, J. Tebbe, G. Leder, H. Apeler, B. Kreft

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): B. Stelte-Ludwig, C. Pena, J. Müller, G. Leder, R. Jautelat

Study supervision: J. Willuda, L. Linden, K. Ziegelbauer, B. Kreft

Other (design and chemical synthesis of the ADCs): H.-G. Lerchen

The authors gratefully acknowledge Norman Dittmar, Karola Henschel, Katrin Jänsch, Nicole Kahmann, Anja Klinner, Monika Klotz, Jessica Möllmann, Beatrice Oelmez, Kirstin Seifert, Juliane Szengel, Bianka Timpner, Jana Wätzold, Elke Fischer, Juergen Wedlich, Georg Zeidler, Thorsten Boldt, Sebastian Deitz, Anna DiBetta, Beate König, Dirk Wolter, Susanne Bendix, and Bettina Muchow for excellent assistance. The authors thank Dr. Lisa Dietz for measuring the metabolites and Dr. Anette Sommer for a critical review of the manuscript. Aurexel Life Sciences Ltd. (www.aurexel.com) is acknowledged for editorial support funded by Bayer AG.

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.