CEACAM5-positive colorectal, lung, or gastric cancer cell lines were purchased from ATCC (SK-CO-1, SNU-C1, H508, LS174T, SNU-16, H727, H2122) or DSMZ (MKN-45), and cultured according to the respective datasheets. CEACAM5-negative A549 cells (ATCC), and two primary epithelial cell lines, i.e., HBEpiC (ScienCell Research Laboratories) and CCD841CoN (ATCC), were used as controls. The number of CEACAM5 and CD47 antigens/cell was determined by QIFIKIT (Agilent DAKO) according to the manufacturer’s instruction as previously described (41).

MC38 (Kerafast) were silenced for mouse CD47 cell surface expression and were afterwards stably transfected to express human CD47 and human CEACAM5. A stable clone named 3C12 was isolated by cell sorting (Beckman #Cytoflex SRT) and cultured like mentioned in the original MC38 datasheet. The number of human CEACAM5 (203’000 molecules/cell) and human CD47 (40’000 molecules/cell) expressed at the surface of the clone was determined by QIFIKIT (Agilent DAKO) according to the manufacturer’s instruction as previously described (41).

Peripheral blood mononuclear cells (PBMCs) from healthy donors were isolated from buffy coats (Centre de Transfusion Sanguine Genevois, Switzerland), and monocytes differentiated into macrophages by culturing PBMCs in presence of 20 ng/mL human M-CSF (Peprotech). At least three donors were tested per cell line. To assess the impact of soluble CEACAM5 (sCEACAM5), 0.02, 0.05 and 0.1 μg/mL sCEACAM5 (Biorad) was added to selected experiments.

After six days, differentiated macrophages were detached from the culture flask and 30.000 cells/well seeded in clear flat bottom 96-well plates (Costar). Plates were incubated for 48 hours at 37°C and 5% CO₂. Subsequently, supernatants form the wells were removed and replaced by 0.5 µM of calcein red-orange (ThermoFisher Scientific) diluted in PBMC medium. After 30 min at 37°C, plates were washed twice with 150 µL of phosphate-buffered saline per well. After the last washing step, 100 µL of PBMC medium containing 2 mg/mL of irrelevant human IgG1 to saturate the FcγR1 expressed on macrophages was added.

In the meantime, target tumor cells were stained with 1 µM of calcein-AM (ThermoFisher Scientific) for 30 min at 37°C. After washing, cells were resuspended at 1.8 x 10⁶ cells/mL and stained target cells were opsonized with antibodies of interest for 20 min at room temperature (RT). Subsequently, 100 µL of opsonized stained target cells were transferred to each well and plates were incubated for 2h 30 min at 37°C. After incubation, medium was removed and replaced with 100 µL of PBMC medium.

Plates were acquired using a CellInsight CX5 imager (ThermoFisher Scientific) and values corresponding to the average number of target cells number engulfed per macrophage were extracted using the imager’s HCS Studio software. Values were used to calculate the phagocytosis index, defined as the number of target cells engulfed by one hundred macrophages.

PBMCs from healthy donors were used as a source of effector T-cells. At least three donors were used for each cell line tested. To assess the impact of sCEACAM, 0.02, 0.05 and 0.1 μg/mL sCEACAM5 was added to selected experiments.

PBMCs were centrifuged at 300 g for 5 min and the cell pellet resuspended in Roswell Park Memorial Institute (RPMI) 1640 medium supplemented with 2% fetal calf serum (FCS), 2 mM L-glutamine, 25 µg/mL gentamicin (all from Sigma-Aldrich), at a density of 1x10⁷ cells/mL. 5x10⁵ PBMCs were plated in Ultra-Low Attachment Sterile 96-well round bottom plates (Corning) and 1x10⁴ target cells added. Antibodies were diluted in RPMI-1640 containing 2% (v/v) FCS at 4x final concentrations and 25 µL of the diluted antibody was added to the plate. The highest final concentration was 6.66 nM, followed by seven 1/5 serial dilutions. Wells containing only effector cells mixed with target cells, or target cells only, served as controls. Plates were incubated at 37°C for six hours.

For readout, 5 µL of lysis buffer (Cytotoxicity detection kit PLUS (LDH), Roche) was added to wells containing only the targets cells and incubated for at least 5 min. The lysis step was verified by visual inspection under the microscope. These wells served as positive control of 100% target cell lysis.

In the meantime, the detection mixture was prepared according to the manufacturer’s instructions. 100 μL of the freshly prepared detection mixture was added to the plate containing the diluted ADCC samples and incubated at RT, protected from light, for 15 min. After incubation, 50 μL of stop reagent was added. Optical density was measured using a microplate reader at 490 nm (Spectra i3Max). The percentage of specific lysis was calculated by using the following equation:

Effector + target=baseline without Ab; max. target=target cells with lysis buffer; target=target cells only (spontaneous LDH release).

Specific lysis results were then reported in GraphPad Prism to establish potency curves.

Animal studies were performed in accordance with the Swiss Veterinary Office guidelines and approved by the Cantonal Veterinary Office (Geneva, Switzerland; #GE134 and #GE299). Mice were randomly assigned into different groups. Study endpoints included tumor volume >1500 mm³, body weight loss >15%, and other graft-versus-host-disease symptoms. Anesthesia was performed by inhalation of isoflurane 3%, for 10 min. maximum, euthanasia with an intraperitoneal (IP) injection of pentobarbital 150 mg/kg (lethal dose).

In the xenograft model, 7-8-week-old NOD SCID female mice (NOD.Cg-Prkdc^scid/J; Charles River) were injected subcutaneously (SC) with 3x10⁶ SNU-C1 target cells. Treatment with NILK-2401 intravenously (IV) at 20 mg/kg or IgG1 isotype control at 20 mg/kg (n=8 each) twice/week was started on day 1. The last injection on day 43 was given IP for technical reasons.

NILK-2401 was also tested in a syngeneic model with 6- to 20-week-old female human CD47/human SIRPα transgenic mice (C57BL/6-Sirpa^tm1(SIRPA)Cd47^tm1(CD47)/Bcgen; Biocytogen) engrafted SC with 0.5x10⁶ MC38-hCEA-hCD47 cells expressing human CEACAM5 and human CD47 ( Supplementary Table S1 ). Treatment was started at day 1 post engraftment; NILK-2401 was dosed IP at 30 mg/kg twice/week (n=6), an IgG1 isotype antibody at 30 mg/kg served as control (n=8).

Mice were controlled daily for clinical symptoms and potential adverse events. Animals were weighed twice/week using a digital balance. For tumor growth follow-up, tumors were measured thrice/week in two dimensions (length, width) using a caliper and the volume was expressed in mm³ using the formula:

The tumor volume measurement was used to calculate the area under the curve (AUC) for each individual mouse. To estimate the AUC, the same time frame was used for all mice. The statistical analysis of AUC was performed with data starting from the day of the engraftment (d0), until the first day of euthanasia due to tumor volume >1500 mm³. AUC reflects the tumor growth curve between two defined time points and allows comparison between groups.

The percentage of tumor growth inhibition (TGI) at indicated time points in comparison to the control group was determined based on tumor volumes, using the formula:

Tt = mean tumor volume of treated at time t. T0 = mean tumor volume of treated at time 0. Vt = mean tumor volume of control at time t. V0 = mean tumor volume of control at time 0.

A Kaplan-Meier survival analysis was done using GraphPad Prism software. Mice euthanized due to ulceration were excluded from the analysis.

Please see the Supplementary Data for additional information.

The κλ body platform (40) was used to generate a human CEACAM5xCD47 BsAb selectively blocking CD47 on CEACAM5-positive tumor cells ( Figure 1 ).

The anti-CD47 arm was isolated from the naïve dummy kappa libraries using LCB’s phage display platform. Due to ubiquitous CD47 expression, the anti-CD47 arm was designed to minimize CD47-mediated monovalent binding of the CEACAM5xCD47 BsAb to CEACAM5-negative cells. The selection of the final anti-CD47 arm (called K2) was based on in vitro and in vivo pharmacology, pharmacokinetic and toxicology/tolerability studies to identify the CD47-arm with the most appropriate affinity to maximize efficacy on one hand and tolerability on the other hand. A detailed description of the characterization of the anti-CD47 arm is summarized in (42).

The isolation of anti-CEACAM5 sequences was performed using proprietary phage display libraries displaying scFv from the dummy naïve lambda libraries. Fifty-eight different selection strategies were performed using soluble recombinant human CEACAM5 and/or cynomolgus CEACAM5, or cells expressing human CEACAM5 or cynomolgus CEACAM5, as well as combinations of both approaches. Generated κλ bodies were tested in several in vitro assays, to assess their activity and safety, and the most promising candidates were subjected to an affinity maturation process (lead optimization). Following the in-depth characterization of lead-optimized candidates, a candidate matching the target product profile was identified (i.e., K2AC100), selected as clinical lead candidate, and termed NILK-2401.

As shown above, no binding to or killing of CEACAM5-negative cells or primary epithelial cells was observed in vitro with NILK-2401. Specificity of binding to CEACAM5 and CD47, and thus likelihood of absence of off-target effects, was likewise addressed using PEAK-cells expressing different CEACAMs ( Supplementary Figure S1 ) and analysis of whole blood samples ( Figures 5A–D , Supplementary Table S3 ).

The in vivo anti-tumor activity of NILK-2401 was investigated in a mouse xenograft model, using SNU-C1 as target cells, as well as in a syngeneic model using human SIRPα/human CD47 transgenic mice with MC38 transfected with human CEACAM5 and human CD47 ( Figure 6 ).

NOD SCID mice were exposed to 20 mg/kg NILK-2401 IV twice weekly vs. control IgG1. NILK-2401 significantly delayed tumor growth, with a TGI of 53% at d35, i.e., 1^st day of euthanasia due to tumor volume >1500 mm³ ( Figures 6A–C ).

In hSIRPα/CD47 transgenic mice, NILK-2401 treatment at 30 mg/kg IP twice/week significantly suppressed tumor growth of MC38-hCEACAM5/hCD47, with a TGI of 84.3% at d22 (i.e., 1^st day of euthanasia due to tumor volume >1500 mm³; Figures 6D–F ). Prolonged treatment with NILK-2401 resulted in almost complete tumor regression followed by relapse of the tumors several days after stopping the treatment. At the end of the experiment (52 days post engraftment), all isotype control-treated mice were euthanized due to high tumor volume, whereas NILK-2401-treated mice showed a survival rate of 83% ( Figure 6G ).

The PK profile of NILK-2401 was evaluated in cynomolgus monkeys as part of a non-GLP single-dose study and in Tg32 FcRn humanized mice.

The anti-CEACAM5 arm is not cross-reactive to cynomolgus CEACAM5, while the anti-CD47 arm of NILK-2401 does recognize the cynomolgus target with similar affinity (see above).

Following IV administration at 0.5 mg/kg ( Figure 7A , Supplementary Table S4 ), NILK-2401 concentrations decreased steadily until the last time point (i.e., 1,008 hours), although the decrease seemed slightly higher in one monkey between 0.25 and 168 hours. After 168 hours, elimination was similar across all animals.

After IV administration at 20 mg/kg ( Figure 7B , Supplementary Table S4 ), concentrations decreased steadily till the last time point in two monkeys. In the two other monkeys, concentrations declined steeply after 168 hours and were below the limit of quantification after 672 hours. Concentrations during the first 24 hours in three monkeys were higher compared to the other monkey (range: 508-1618 μg/mL vs. 173-473 μg/mL).

NILK-2401 PK seemed to be more than dose-proportional for C_max and AUC_0-inf. No gender differences in PK were observed.

PK data were confirmed in Tg32 human FcRn mice ( Figure 7C , Supplementary Table S5 ), although several values were flagged because they did not fulfill at least one of the conditions for reliable derivation. NCA of the mean PK data showed a dose-related increase in C_max and AUC_last. Clearance (CL) seemed slightly higher after injection at 0.5 mg/kg compared with the 20 mg/kg dose level (0.283 vs. 0.183 and 0.203 mL/h/kg, respectively, depending on the inclusion or exclusion of outliers).

Doses of 0.5 and 20 mg/kg, given as single IV administration, were well tolerated over the 6-week observation period (data not shown). No abnormalities in clinical signs, including observation of the injection site, and food intake were observed, and no toxicologically meaningful changes were present in body weights, safety pharmacology (i.e., neurological behavior, ECG, blood pressure and respiratory function), clinical pathology (i.e., hematology, clinical chemistry, and coagulation) and at post-mortem examinations.

Mortality was observed in one low-dose female monkey immediately after administration. No clinical signs were observed before death and no grossly abnormal observations were recorded at necropsy. Following histological examination of the full list of collected tissues, a clear cause of death was not defined. However, based upon the single incidence of mortality and since no signs of toxicity were noted in other animals, the death of this animal was considered unlikely to be related to NILK-2401 treatment.

Likewise, NILK-2401 was well tolerated in human FcRn Tg32 mice (data not shown); no clinical signs, no weight loss, or mortality were observed following injections of 0.5 and 20 mg/kg NILK-2401, respectively.

The immunogenicity potential of NILK-2401 was evaluated by quantification of the pre-existing NILK-2401-reactive CD4 T-cell repertoire in the blood of 14 normal donors with diverse HLA molecules in parallel to controls including Adalimumab (anti-TNFα), Trastuzumab (anti-HER2 mAb), and KLH. The therapeutic antibodies Adalimumab and Trastuzumab are described to be potentially non-immunogenic and immunogenic proteins, respectively (43, 44), while KLH is a highly immunogenic T-cell dependent antigen (45).

Across the 14 KLH responding donors, nine donors responded to Adalimumab, and exhibited a mean response rate of 0.2 specific CD4 T-cells per million. Six donors responded to Trastuzumab and exhibited a mean response of 0.1 specific CD4 T-cells per million, which is at the cut-off between low and moderate T-cell response. Five donors responded to NILK-2401 and exhibited a mean response of 0.07 specific CD4 T-cells per million. Overall, these results show that NILK-2401 has a T-cell response rate in the category of low risk of immunogenicity ( Supplementary Figure S6 ).

When combining two different mechanisms of action, i.e., inhibition of CD47 and re-targeting of T-cells to the tumor, equivalent activity of NILK-2401 plus NILK-2301 therapy at >600-fold (LS174T) to >3,000-fold (MKN-45) lower doses compared to NILK-2301 monotherapy was observed, i.e., ≈0.02 vs. 13.3 and, ≈0.03 vs. 100 nM, respectively. Combination with NILK-2401 can thus boost NILK-2301 activity (i.e., Emax) by 15 to 250% ( Figure 8 ).