A bispecific antibody-drug conjugate targeting pCAD and CDH17 has antitumor activity and improved tumor-specificity.

Introduction
Antibody–drug conjugates (ADCs) are an innovative class of targeted therapeutics that have the potential for improved cancer treatment. ADCs combine the selectivity of monoclonal antibodies (mAbs) with the potent cytotoxic activity of small molecule drugs (payloads), enabling targeted payload delivery through antigen-specific binding. To induce cell cytotoxicity, ADCs bind to the surface antigen target, internalize within the cell, and subsequently traffic to the lysosome where the antibody-payload linker or antibody is degraded, releasing the cytotoxic payload to act on its intended target.1,2 Despite the promise of ADCs, there remain challenges in their development and clinical implementation, such as achieving sufficient tumor specificity that minimizes damage to normal tissues induced by the highly potent payload.3,4
P-cadherin (pCAD, placental Cadherin, Cadherin-3; encoded by CDH3) is a cell-surface glycoprotein and a member of the classical cadherin superfamily. It is involved in calcium-dependent cell–cell adhesions of the epithelium and is associated with various tumor-promoting processes, including cell invasiveness, metastasis, and tumorigenesis.5,6 pCAD is overexpressed in several cancer types while being restricted in normal tissues,6 presenting an attractive antigen for delivering potent cytotoxic payloads. Indeed, PCA062, a first-in-class pCAD-targeting ADC, demonstrated potent in vitro and in vivo antitumor activity in several pCAD-expressing tumors and achieved a favorable safety profile in non-human primate toxicology studies prior to progressing into clinical trials.7,8 Additionally, another pCAD-targeting ADC, BC3195, has recently entered Phase I clinical trials for advanced solid tumors.9
CDH17 (LI-Cadherin, liver-intestinal Cadherin, Cadherin-17; encoded by CDH17) is similarly part of the cadherin superfamily and plays an essential role in cell adhesion and organ development, particularly in the formation and maintenance of the intestinal epithelium.10,11 In colorectal cancer (CRC), CDH17 has been implicated in regulating integrin signaling for cell adhesion, promoting cell proliferation, inhibiting apoptosis, and facilitating metastasis.12,13 Its expression has been associated with prognostic importance and has therefore emerged as a key CRC marker.14 While various CDH17-based therapeutics are under investigation for treating CRC, including an antibody,15 an antibody-conjugated to a photosensitizer,16 and a T-cell engager,17 no CDH17-targeted ADCs have reached clinical trials. Notably, pCAD and CDH17 are both highly expressed in CRC, presenting an opportunity to leverage their combination to improve ADC specificity to CRC cells expressing both antigens.
Bispecific antibodies (bsAbs) are designed to simultaneously target two distinct antigens. Previous studies combining bsAbs with the ADC approach have demonstrated advantages, including enhanced ADC efficacy through improved internalization and lysosome-directed degradation. For example, coupling the rapidly internalizing protein PRLR with the clinically approved ADC target HER2 was shown to enhance internalization and degradation of HER2, resulting in increased cell death compared to a monospecific HER2 ADC.18 Similarly, coupling the lysosomal membrane protein CD63 with HER2 induced ADC lysosomal accumulation in HER2+ cells, leading to increased cell cytotoxicity.19 Other bispecific ADCs have been developed that couple the tyrosine kinase receptor EGFR with targets such as cMET, aiming to overcome secondary pathway mutations that cause treatment resistance,20 or with MUC121or TROP2,22 which are highly co-expressed on cancer tissues. Notably, the EGFR x TROP2 ADC and the EGFR x cMET ADC are currently in clinical trials.
While dual targeting to improve tumor specificity is not a new concept, our work uniquely contributes by identifying pCAD and CDH17 as an optimal antigen pairing for targeting CRC. Further, toward identifying a pCAD x CDH17 bispecific, we designed a novel screening approach where we screened CDH17 arms and affinity-tuned pCAD arms by focusing on avidity-driven antibodies. We used in vitro cancer cell lines and assays that assess bsAb binding and internalization of the cytotoxic molecule monomethyl auristatin E (MMAE) leading to cell proliferation inhibition of dual-antigen-expressing cells compared to single antigen-expressing cells. The whole screening approach was conducted in two phases. First, we screened CDH17 arms and selected a high avidity candidate as the lead, based on its ability to produce the greatest difference between a monovalent (one-armed) antibody and its corresponding bivalent antibody for cell binding and inhibiting cell proliferation. Next, we generated a panel of affinity-tuned pCAD arms and similarly screened for in vitro cell cytotoxicity and binding differentials. Through this rationally designed two-phase workflow, we identified a lead pCAD x CDH17 bsAb, which we subsequently conjugated with MMAE payload. In preclinical in vivo studies using mice bearing dual flank tumors of CRC, we demonstrated that the pCAD x CDH17 bispecific ADC inhibited the growth of dual-antigen-expressing tumors while sparing single antigen-expressing tumors. Overall, these results support the development of bispecific ADCs and validate pCAD x CDH17 as paired targets, with the potential to offer clinical benefit to cancer patients.

Results

pCAD and CDH17 dual target expression is high in solid tumors and limited in normal tissues
To validate the suitability of the targets for a bispecific ADC approach, we first set out to evaluate the bulk mRNA levels of CDH3 (the gene encoding pCAD) and CDH17 in both human normal tissue and cancer samples using publicly available datasets from The Cancer Genome Atlas (TCGA)23 and Gene-Tissue-Expression (GTEX).24 Our analysis revealed that CDH3 mRNA is overexpressed in various solid tumors, including CRC, and found at moderate levels in normal esophagus, skin, prostate, and reproductive tissues (Figure 1ai). CDH17 mRNA is similarly overexpressed in several gastrointestinal solid tumors, including pancreas and stomach cancer, in addition to CRC. In normal tissues, CDH17 mRNA is moderately expressed in the large and small intestine (Figure 1aii). These findings are also reflected in publicly available scRNAseq datasets for healthy normal donors from Tabula Sapiens,25 where CDH3 single-cell RNA expression is widespread across tissues, and CDH17 expression is mostly expressed in the small and large intestine (Figure S1A). We also integrated multiple publicly available scRNAseq datasets26,27 collected from CRC patients to confirm that cancer cells are the cell type responsible for high CDH17 and CDH3 expression, instead of other cell types present in the CRC tumor microenvironment (Figure S1B).

While detectable mRNA levels of CDH3 and CDH17 are present in specific normal tissues independently, our primary interest for the dual-targeting approach was on evaluating the co-expression of CDH3 and CDH17. Encouragingly, our co-expression bioinformatics analysis demonstrated that the CDH3 and CDH17 dual-expressing population is predominantly isolated to cancer cells, with minimal dual-positive expression observed in normal tissues (Figure 1b). These RNA-level findings were corroborated by cell surface protein expression, demonstrated by positive immunohistochemistry (IHC) staining in a set of CRC samples (Figure S2A). To compare to the co-expression of pCAD and CDH17 in normal tissues, we conducted IHC on a set of normal tissue blocks. Significantly, our analysis confirmed single antigen staining on expected normal tissue liabilities such as the colon and prostate, for CDH17 and pCAD, respectively (Figure S2Bi), with a notable absence of dual-antigen expression across normal tissues (Figure S2Bii).

Generation of bsAb phase I: selection of lead CDH17 arm
Based on the suitability of pCAD and CDH17 as antigens for a dual-antigen targeting approach to increase tumor-targeting selectivity, we next set out to generate a cross-arm avidity binder following the schematically outlined in Figure 2a.

The first phase of the selection process involved identifying an avidity-driven CDH17 binding arm. We generated 24 unique anti-CDH17 IgG antibodies to be functionally profiled. To enable screening for high avidity binders, we also generated 24 monovalent antibodies to be compared to the corresponding 24 bivalent CDH17 antibodies. First, we evaluated the 48 total monovalent and bivalent candidates using an in vitro inhibition of cell proliferation assay in two CDH17+ cell lines, SNU16 (stomach cancer cell line), and an engineered CDH17-overexpressing CRC cell line, HT29CDH17+ (Figure 2b, Figure S3). Our analysis demonstrated that IC50 values ranged from 1.8E–6 nM to 10.9 nM in SNU16 cells, and 8.4E–6 nM to 11 nM in HT29CDH17+ cells. To enable selection based on highly avid arms, we calculated the ratio between IC50 values of corresponding monovalent and bivalent antibodies and filtered the list of candidates based on those with a high ratio observed across both cell lines tested (Figure 2c). In the majority of clones (22/24), we observed a larger fold change in proliferation inhibition in the SNU16 cell line compared to the overexpressing HT29 line, likely due to the lower CDH17 expression in SNU16 cells. The top 7 CDH17 arms selected were: CDH17 1, CDH17 2, CDH17 5, CDH17 9, CDH17 11, CDH17 15, and CDH17 22, as highlighted in Figure 2c. The 48 total antibodies were also assessed in a fluorescence-activated cell sorting (FACS) binding assay in HT29CDH17+ cells, and the ratio between binding IC50s of corresponding monovalent and bivalent antibodies was similarly quantified (Figure S4A). Across the 7 CDH17 arms selected from the in vitro inhibition of proliferation assay, all monovalent antibodies had lower or similar binding to HT29CDH17+ cells compared to the respective bivalent antibodies (Figure S4B).
Upon filtering the 24 candidates, we next combined all 7 CDH17 arms with a pCAD arm to generate 7 pCAD x CDH17 bsAbs, named bsAb 1 through 7 (Figure 2di; QC in Figure S5). To evaluate whether the avid property of the CDH17 arm was maintained in the bispecific format, we compared the proliferation inhibition capacity between corresponding monovalent and bivalent antibodies, and bsAbs, in an engineered dual-expressing CRC cell line, HT29pCAD+CDH17+ (Figure 2dii). Our results demonstrated that all 7 bsAbs retained higher efficacy at internalizing and inhibiting proliferation compared to their corresponding monovalent antibodies, with similar potency to their bivalent equivalents. The potency of our bsAbs was further translated to dual-positive cell selectivity, upon assessing the cell binding of the 7 bsAbs in single positive expressing cell lines (HT29pCAD+ and HT29CDH17+) compared to a double positive expressing cell line (HT29pCAD+CDH17+) (Figure 2ei). All 7 bsAbs showed increased dual -antigen-specific binding compared to the CDH17-only expressing cell line, quantified by the fold change between their measured Amax values (Figure 2Eii). While we achieved a lack of selectivity for CDH17-only expressing cells using our avidity screening approach, the difference between the dual-expressing cell line and the pCAD-only expressing cell line was minimal, justifying the search for an avidity-driven pCAD arm. Before the subsequent phase of our approach, however, we further narrowed our 7 CDH17 candidates to a lead CDH17 arm that had the most avid arm, CDH17 11, based on the combined proliferation inhibition and FACS binding assay results. Overall, in the first pCAD x CDH17 bsAb search phase, we demonstrated improved selectivity of our lead CDH17 arm to dual-expressing cells compared to CDH17-only expressing cells.

Generation of bsAb phase II: selection of lead affinity detuned pCAD arm
The second phase of the selection process sets out to address the remaining activity on pCAD-only expressing cells, in an attempt to mitigate any potential toxicity to pCAD normal tissue liabilities as previously discussed. To generate additional pCAD candidates for testing, we detuned the affinity of the pCAD arm by alanine scanning the light chain complementary-determining region 3 (CDR3), where each amino acid of the CDR was systematically replaced by an alanine, generating a panel of 8 light chain single alanine variants named pCAD var 1 through 8 (Figure 3a).

Using the pCAD alanine variant monovalent antibodies, we ran an in vitro inhibition of cell proliferation assay in an engineered pCAD+ cell line (HT29pCAD+) (Figure 3b). The calculated IC50 values ranged between 0.02 and 11 nM and the Amax values ranged between −9.5% and 62.7% inhibition, where all alanine variants achieved a less potent IC50 compared to the original pCAD monovalent mAb. We also ran the pCAD alanine variant monovalent antibodies, along with the pCAD monovalent mAb, in a FACS binding assay using the same pCAD+ cell line (Figure 3ci). The IC50 values ranged between 0.23 and 100 nM, with most candidates achieving the desired decreased binding compared to pCAD mAb. Because the FACS binding assay provided a larger dynamic range for comparing pCAD alanine variants than the proliferation inhibition assay, we opted to prioritize the FACS binding results for pCAD arm selection. Taking the fold change of the IC50 values between the alanine variant and the parental pCAD arm, we selected those that had an intermediary of 1.5 to 5-fold less activity (Figure 3cii). From this list, we selected three variants that had low, intermediate, and high binding, pCAD var 1, var 6, and var 2, respectively, to combine with our lead CDH17 avidity-driven arm CDH17 11, to correspondingly generate bsAb 8, bsAb 9, and bsAb 10 (QC in Figure S5).
To evaluate the specificity of our three affinity-detuned pCAD x CDH17 bsAbs, we ran all candidates in an in vitro inhibition of proliferation assay, across a panel of cell lines (HT29 parental with low expression of both antigens, HT29pCAD+, HT29CDH17+, and HT29pCAD+CDH17+) (Figure 3d). As expected, across all bsAbs, no activity was observed in the parental double negative cell line, reduced activity was observed in the CDH17+ line, and high activity was observed in the dual-expressing cell line. Moreover, compared to the bsAb with the parental pCAD arm (Figure 3di), bsAb 10 was able to induce the largest reduction in activity in the pCAD-only cell line compared to bsAb 8 and bsAb 9 (Figure 3dii), as quantified by the increased IC50 and decreased Amax values (Figure 3diii), and fold change of IC50 values between single expressing cell lines and the dual-expressing cell line (Figure 3div). Overall, in phase II, we generated a pCAD arm using an alanine scanning method that achieved specificity to dual-expressing cells.

pCAD x CDH17 bispecific ADC is selective for dual-expressing tumors in dual flank in vivo mouse models
To evaluate the in vivo antitumor activity and specificity of a pCAD x CDH17 bispecific ADC, we conjugated our lead pCAD x CDH17 bsAb (bsAb 10) with the well-validated cytotoxic payload MMAE for our preclinical proof-of-concept, allowing us to focus on the antibody discovery and dual targeting aspects of the molecule. MMAE, a potent antineoplastic agent that binds to tubulin and microtubules to induce cell death, is one of the most commonly employed payloads for ADCs in the clinic.28 We also conjugated MMAE to the parental pCAD x CDH17 bsAb 5 and affinity detuned pCAD x CDH17 bsAb 8, which represents the selected alanine variant pCAD arm with lower cell binding affinity, to be used for comparative purposes. All three pCAD x CDH17 bsAbs with the mutations for conjugation were profiled by SDS-PAGE and analytical size exclusion chromatography (SEC) (Figure S6), and the resulting pCAD x CDH17 mmAE ADCs were tested in vitro using the HT29 panel previously used to confirm expected profiles (Figure S7A).
For our in vivo experiments, we utilized dual flank cell line-derived xenograft models to comprehensively evaluate the efficacy and specificity of the pCAD x CDH17 bispecific ADC against dual-antigen expressing tumors compared to single-antigen expressing tumors. One model was implanted with HT29 cells expressing both pCAD and CDH17 on the right flank, and with single positive HT29CDH17+ cells on the opposite flank, referred to as the CDH17 dual flank model (Figure 4ai). The second model was similarly implanted with HT29pCAD+CDH17+ cells on the right flank and single positive HT29pCAD+ cells on the opposite flank, referred to as the pCAD dual flank model (Figure 4aii). Importantly, both models maintained similar growth rates between flanks. Additionally, CDH17 and pCAD expression were evaluated and characterized using IHC in implanted engineered HT29 cell lines. At both 15- and 32-days post-implantation, CDH17 and pCAD were robustly expressed in the respective engineered xenografted tumors, as depicted in the representative images in Figure S7B at day 32. This confirms that the upregulation of CDH17 and pCAD in HT29 engineered cells is maintained throughout in vivo growth.

Upon administering a single intravenous dose of 2.5 mg/kg of the parental pCAD x CDH17 (bsAb 5) MMAE ADC, we observed expected in vivo responses that aligned with our in vitro observations. In the CDH17 dual flank model, the dual-antigen expressing tumor exhibited durable remission in vivo. Conversely, the CDH17-expressing tumor on the opposite flank continued to increase in tumor volume, demonstrating growth similar to that observed in the vehicle-treated CDH17-expressing tumor (Figure 4bi). Expectedly in the pCAD dual flank model, the single dose of the parental pCAD x CDH17 mmAE ADC led to remissions in both the dual-antigen expressing tumor and the pCAD-expressing tumor, indicating nonspecific targeting of the ADC (Figure 4bii).
Next, we treated both dual flank models with our lead alanine variant pCAD x CDH17 (bsAb 10) MMAE ADC. In the CDH17 dual flank model, we observed a trend consistent with the results of the parental pCAD x CDH17 mmAE ADC, as the CDH17 bsAb arm remained identical (Figure 4ci). In contrast to the parental pCAD ADC, the pCAD dual flank model exhibited a divergent response between the dual-antigen expressing tumor and the pCAD-expressing tumor upon treatment with bsAb 10 mmAE ADC, similar to our in vitro assay results (Figure 4Cii). We further performed pharmacokinetics studies for measuring total antibody and ADC levels in the plasma and demonstrated similar exposure between the parental and affinity detuned ADCs up to 72 hours (Figure S8A), with comparable total antibody and ADC half-lives (62 and 38 hours for bsAb 5 ADC, respectively; 77 and 40 hours for bsAb 10 ADC, respectively). Comparing the in vivo results between bsAb 5 mmAE ADC and the lead alanine variant bsAb 10 mmAE ADC confirms the successful affinity detuning of the pCAD arm, reducing tumor inhibition in pCAD+ single-antigen expressing tumors. Interestingly, when we treated both dual flank models with the alternative alanine variant pCAD x CDH17 (bsAb 8) MMAE ADC, we observed a difference in tumor responses exclusively in the CDH17 dual flank model (Figure S8Bi). The pCAD+ tumor in the treated pCAD dual flank model reached a tumor volume similar to that of the dual-expressing tumor indicating nonspecific dual targeting (Figure S8Bii), as expected based on the in vitro proliferation inhibition assay results of bsAb 8 on HT29pCAD+ cells, underscoring the importance of the pCAD arm in driving selectivity. Our in vivo findings validate the increased specificity of our pCAD x CDH17 bispecific MMAE ADC toward dual-antigen expressing tumors, leading to selective and significant in vivo antitumor activity and achieving complete responses.

Discussion
In the past several years, there has been a marked increase in ADCs entering clinical trials for solid tumors. ADCs hold immense potential in cancer therapy, but their clinical application faces challenges because of the need for highly potent payloads to effectively target and kill tumor cells. This necessity for payload potency underscores the importance of selecting target antigens that exhibit high and robust expression on tumor cells while being limited or absent on normal tissues, establishing a large therapeutic window. As a result, early clinical development and US Food and Drug Administration-approved ADCs for solid cancers have predominantly focused on antigens like HER2, characterized by a substantial differential expression between tumor and normal cells.1,29,30 Despite the emergence of a handful of clinically approved ADCs, the majority are HER2-targeting,1 demonstrating the challenge of identifying suitable antigens. To diversify our approach, we focused on dual targeting CDH17 and pCAD, both of which have been separately used as targeting antigens due to their prevalence in solid cancer tissues, such as CRC. Here, our hypothesis centered on addressing tumor specificity by requiring the co-expression of two surface cell antigens on tumor cells, bypassing potential normal tissue liabilities associated with each antigen when targeted individually, and for more specific delivery of the cytotoxic payload (MMAE). Other potential strategies to enhance tumor specificity include the selection of dual payloads that can elicit a synergistic effect, and more advanced payloads that target cancer cell-specific vulnerabilities.
In this study, we rationally selected for an avidity-driven CDH17 arm and pCAD arm by using antibodies in both a monovalent and bivalent format, selecting those that necessitated two arms for high cell binding and payload delivery. This avidity-driven criteria resulted in a substantial increase in binding affinity and inhibition of proliferation efficiency, specifically, in cells expressing both antigens compared to cells expressing either pCAD or CDH17 alone. Further, we demonstrated the ability to affinity detune pCAD arms, generating a broader diversity of antibodies. This expansion allowed us to identify arms with weaker binding affinities to pCAD+ expressing cells, while preserving their ability to internalize. These arms were subsequently coupled to CDH17 and tested to determine the optimal pairing that achieved a significant difference in response between dual-expressing and pCAD-only expressing cell lines. Ultimately, an intermediate cell binding candidate that maintained sufficient but less binding compared to the parental pCAD arm emerged as the ideal option. The suitability of the lead BsAb was confirmed when conjugated to MMAE and tested in vivo with two dual flank models, where it exhibited a high degree of pCAD and CDH17 dual-positive tumor specificity, highlighting the effectiveness of our rational bsAb design and selection process. This unique avidity-driven approach can be implemented with other co-expressing tumor antigens that are being considered for dual-specific targeting. Notably, whereas the in vitro data demonstrated strong specificity of the lead bsAb ADC, the in vivo dual flank study revealed some nonspecific binding of the CDH17 arm, though minimal and likely less pronounced if the timeline was extended beyond 32 days. Further investigation using monospecific ADC equivalents (i.e., pairing the CDH17 or pCAD Fab of the bispecific with the Fab of an isotype control) would be valuable to further evaluate whether dual targeting achieves increased specificity and overall improves efficacy.
While the data presented here identified a functional pCAD x CDH17 bispecific ADC, there are limitations regarding the therapeutic potential and translational relevance of this molecule. Our in vitro and in vivo studies used an engineered HT29 cell line to model pCAD and CDH17 expression, allowing us to select candidates and demonstrate selective activity of the bispecific ADC. However, a single engineered cell line does not capture the full spectrum of heterogeneity of pCAD and CDH17 expression levels seen in patient samples. Certain tumors exhibit a higher proportion of cells that co-express both antigens, while others predominantly consist of either pCAD or CDH17 single-expressing tumor cells resulting in a small proportion of dual positive cells. Future studies using a broader panel of CRC cell lines with varying expression levels of pCAD and CDH17 would provide greater insight into the contribution of each arm. Assessing pCAD and CDH17 protein expression in CRC patient samples using IHC would also be critical for understanding co-expression patterns and for selecting more representative CRC cell lines. Additionally, further in vivo validation would be required to assess the translational aspect of a pCAD x CDH17 ADC. This includes using patient-derived xenograft models to demonstrate a more clinically relevant assessment of the bispecific ADC’s efficacy in heterogeneous settings, as well as evaluating the impact of the bystander effect, where the released payload crosses the cell membrane following lysosomal degradation of the ADC and enters neighboring cells to exert its cytotoxic effects beyond the initially targeted cell.
In summary, our preclinical investigation establishes the framework of a unique avidity-driven screening approach utilizing pCAD and CDH17 as a dual-antigen pairing for CRC. Through a systematic screening process involving CDH17 arms and affinity-detuned pCAD arms, we identified and selected monovalent arms that exhibited optimal properties when paired as a bispecific ADC. The preclinical data here of a pCAD x CDH17 bispecific MMAE ADC warrants further clinical validation and investigation in the pursuit of an efficacious treatment option for CRC patients.

Materials and methods

Immunohistochemistry
Tissue microarrays (TMAs) were acquired through a collaboration with the Institute of Pathology of the University Hospital Basel. 3-µm sections were prepared and processed for IHC, using a Ventana Discovery Ultra fully automated immunostainer (Ventana Medical System, Tucson, USA): sections were deparaffinized, retrieved with the Cell Conditioning #1 antigen retrieval reagent (Ventana Medical System), then incubated for 60 min at room temperature with the rabbit monoclonal anti-pCAD, clone EPR22426 (Abcam Cat# ab242060; 1:2000 dilution) or with the rabbit monoclonal anti-CDH17, clone EPR3996 (Abcam Cat# ab109190; 1:500 dilution) primary antibody, followed by incubation with ready-to-use OmniMap anti-rabbit antibody, ChromoMap DAB Kit, counterstained with Hematoxylin/Bluing reagent, dehydrated and cover slipped. Images were acquired using the Aperio AT2 slide scanner (Leica Microsystems, Wetzlar, Germany). Normal tissue TMAs, containing representative sections of all major body organs, were stained with identical processes, and scored 0 to 4 for negative, low, medium low, medium high, and high.

Tabula sapiens scRNAseq dataset processing
Raw gene expression count of 454,069 cells from 15 normal human subsets across 24 organs were downloaded from the Tabula Sapiens Consortium.25 The original cell type labeling was inherited. Pseudobulk counts were created by adding gene expression counts in cells of the same cell type and from the same patient. Trimmed mean of M values (TMM) normalization from the edgeR package followed by counts per million (CPM) normalization was applied to remove donor and cell type size effect.

Public CRC scRnaseq dataset processing
Raw count from two public CRC scRNAseq datasets were downloaded from Jabbari et al.26 and Qian et al.27 The SCTransform function from the Seurat package31 was applied for normalization before the Symphony package32 was leveraged for automatic cell type annotation referencing our internal immune-oncology scRNAseq-profiled cell types. Pseudobulk counts were created by adding gene expression counts in cells of the same predicted cell type and from the same patient. TMM normalization from the edgeR package followed by CPM normalization were applied.

Cell culture
SNU16 (ATCC) was cultured in RPMI 1640 (Gibco) supplemented with 10% fetal bovine serum (FBS) (Gibco). HT29 CDH3 knockout cell line was made with a viral vector that expressed Cas9 and a pCAD guide RNA. For engineering overexpressing cell lines, HT29 cell lines were transduced with a lentivirus construct expressing hpCAD, hCDH17, or hpCAD and hCDH17 under an EF1A promoter and were cultured in McCoy’s 5A media (Gibco) supplemented with 10% FBS (Gibco). All cells were cultured at 37°C under 5% CO2 and passaged before reaching confluency up to 25 passages. Cells were validated for pCAD and/or CDH17 expression in FACS or were tested for sensitivity to CDH17+pCAD+ ADCs in a cytotoxicity assay prior to use.

Inhibition of cell proliferation assay
Cells (SNU16, HT29 pCAD KO, HT29 pCAD+, HT29 CDH17+ or HT29 pCAD+CDH17+) were seeded in a culture-treated 384-well clear plate (2000 cells per well in 40 μL culture medium) and incubated at 37°C under 5% CO2 for 24 hours. Serially diluted samples (5 µL) were added to each well and a final concentration of 34 nM of anti-human Fc-VC-MMAE ADC (DAR 2.0) (5 uL) was added to each well. The plate was incubated at 37°C for 120 hours, then 25 uL of CellTiter-Glo (Promega) was added to each well, incubated at 25°C for 10 min, followed by measuring luminescence using the PHERAstar FSX Microplate reader. IC50 values were calculated using Graph Pad Prism 8 software. All assays were performed in quadruplicate technical replicates.

Flow cytometry binding assay
Binding to tool cell lines (HT29 pCAD KO, HT29 pCAD+, HT29 CDH17+, and HT29 pCAD+CDH17+ cells) was evaluated using FACS. All antibodies were diluted in FACS buffer and added to cells followed by a 1 hour incubation. Unbound antibodies were washed off, and cells were then incubated with Alexa Fluor® 647 AffiniPure Goat Anti-Human IgG (H+L) (1:250) (Jackson ImmunoResearch Laboratories, Cat# 109-605-003). After a 30 min incubation, unbound secondary antibody was washed off. Cells were then resuspended in FACS buffer with DAPI (1:5000) (ThermoFisher Scientific, Cat# 62248) for FACS detection using the Miltenyi MACSQuant. The mean fluorescence intensity of the cells in the live gate was plotted against the log antibody concentration, and the IC50 was determined by nonlinear regression fitting.

Antibodies
All antibodies were cloned into CMV-promoter-driven expression plasmids for mammalian cell expression and subsequently produced by transient transfection in the HEK-293T cells. For the identification of avidity-driven antibodies, full mAbs were produced as wild-type huIgG1. The monovalent one-armed antibodies were produced using the knobs-into-holes technology.33 The full-length antibody heavy chain was modified to encode the T366W: S354C mutations, and the Fc was modified to contain the Y349C:T366S:L368A:Y407V mutations. The light chain, mutated heavy chain, and Fc were co-transfected into HEK-293 cells. Purification was done by affinity chromatography on a mAb SelectSure resin (Cytiva) followed by a polishing step by SEC. Antibodies were characterized for integrity and homogeneity by SDS-PAGE analysis and mass spectrometry.
The bispecific antibodies were also produced using the knobs-into-holes technology. The anti-pCAD antibodies harbor a Kappa light chain and the anti-CDH17 antibodies harbor a lambda light chain. The anti-pCAD heavy chain was modified to encode the T366W: S354C mutations, and the anti-CDH17 heavy chain was modified to contain the Y349C:T366S:L368A:Y407V mutations. The conjugated bispecific antibodies evaluated in vivo were designed with mutations E152C:S375C for facilitating conjugation and mutations D265A:P329A for silencing Fc effector functions. The purification of bispecific antibodies was done by affinity chromatography on a mAb SelectSure resin (Cytiva) followed by a first polishing step by affinity chromatography on a CaptureSelect LC Lambda resin (ThermoFisher Scientific) and a second polishing step by SEC. Antibodies were characterized for integrity and homogeneity by SDS-PAGE analysis and mass spectrometry.
Other antibodies used in this study were purchased from commercial vendors: Alexa Fluor® 647 AffiniPure Goat Anti-Human IgG (H+L) (Jackson ImmunoResearch Laboratories, Cat# 109-605-003), anti-CDH17 (Abcam #109190), anti-pCAD (Sigma-Aldrich, Cat# HPA001767).

bsAb conjugation
Ten mg of each antibody was incubated with RMP Protein A resin (GE) at a ratio of 10 mg Ab to 1 mL resin in phosphate-buffered saline (PBS) for 15 min with mixing in an appropriately sized disposable column. Cysteine HCl was added to a final concentration of 20 mm and incubated with agitation for 30 min at room temperature to allow the reactive cysteines to be deblocked. The resin was quickly washed with 50 column volumes PBS on a vacuum manifold. The resin was then re-suspended in an equal volume PBS containing 250 nM CuCl2. Reformation of antibody interchain disulfides was monitored by taking time points. At each time point, 25 mL of resin slurry was removed, 1 mL of 20 mm MC-VC-MMAE was added, and the tube flicked several times. The resin was spun down, supernatant removed, and then eluted with 50 mL antibody elution buffer (ThermoFisher Scientific). The resin was pelleted and the supernatant analyzed by reverse-phase chromatography using an Agilent PLRP-S 4000A.
Once it was determined that the antibody reformed its interchain disulfide bonds, the resin was washed with 10 column volumes of PBS and the resin was resuspended in an equal volume PBS and 8 equivalents of MC-VC-MMAE in dimethyl sulfoxide (DMSO), with a final concentration of 10% DMSO in the reaction and then incubated at room temperature for 3 hours. The resin was then washed with 50 column volumes PBS. The ADC was eluted from the protein A resin with antibody elution buffer. The ADC was then buffer exchanged into PBS. Samples were sterile filtered with a 0.22 mm syringe filter. The following analyses were performed: 1) analytical SEC to determine percent monomer, 2) mass spectroscopy to determine DAR, 3) limulus amebocyte lysate test to determine endotoxin load, and 4) protein concentration was determined by A280 utilizing extinction coefficient and molecular weight of antibody or BCA protein assay (Thermo).

In vivo dual flank model efficacy studies
Studies were conducted in accordance with ethical guidelines and regulations set by the Novartis BioMedical Research Institutional Animal Care and Use Committee, as well as the Guide for the Care and Use of Laboratory Animals. Cell lines were confirmed to be free of mycoplasma and mouse viruses before use and cultured in medium according to ATCC guidelines. 5 × 10E6 cells were resuspended in 200 uL of ice-cold Hank’s Balanced Salt Solution (HBSS; Gibco, Cat# 25200–056) containing 50% Matrigel (Corning, Cat# 354234). Female NSG mice (6–12 weeks of age) were inoculated subcutaneously in the right flank with HT29pCAD+CDH17+ cell line and concurrently inoculated on the left flank with either the HT29pCAD+ or the HT29CDH17+ cell line. Tumor-bearing mice were randomized into treatment groups 15 or 16 days after inoculation, once the average tumor volumes on both flanks reached 200–225 mm3. Tumors were measured twice weekly by calipering in two dimensions. Tumor volume was calculated using a modified ellipsoid formula: tumor volume – L × W2 × π/6, where L is the longest axis of the tumor, and W is perpendicular to L. Tumor ulcerations were observed in some instances, at which point mice had to be euthanized early. Serum samples were collected 1 hour, 24 hours, and 72 hours post-dose for quantification of total mAb and total ADC-MMAE concentrations.

Abbreviations

ADCAntibody–drug conjugate
bsAbBispecific antibody
CDH17Cadherin 17
CDR3Complementary determining region 3
CRCColorectal cancer
DARDrug antibody ratio
FACSFluorescence-activated cell sorting
GTEXGene-Tissue-Expression
IHCImmunohistochemistry
MMAEMonomethyl auristatin E
pCADP-cadherin
SECSize exclusion chromatography
TCGAThe Cancer Genome Atlas

Supplementary Material

KMAB_S_2024_0104_R2_final supp materials.docx