Discovery of ginisortamab, a potent and novel anti-gremlin-1 antibody in clinical development for the treatment of cancer

ABSTRACT

Gremlin-1, a high-affinity antagonist of bone morphogenetic proteins (BMP)-2, −4, and −7, is implicated in tumor initiation and progression. Increased gremlin-1 expression, and therefore suppressed BMP signaling, correlates with poor prognosis in a range of cancer types. A lack of published work using therapeutic modalities has precluded the testing of the hypothesis that blocking the gremlin-1/BMP interaction will provide benefits to patients. To address this shortfall, we developed ginisortamab (UCB6114), a first-in-class clinical anti-human gremlin-1 antibody, currently in clinical development for the treatment of cancer, along with its murine analog antibody Ab7326 mouse immunoglobulin G1 (mIgG1). Surface plasmon resonance assays revealed that ginisortamab and Ab7326 mIgG1 had similar affinities for human and mouse gremlin-1, with mean equilibrium dissociation constants of 87 pM and 61 pM, respectively. The gremlin-1/Ab7326 antigen-binding fragment (Fab) crystal structure revealed a gremlin-1 dimer with a Fab molecule bound to each monomer that blocked BMP binding. In cell culture experiments, ginisortamab fully blocked the activity of recombinant human gremlin-1, and restored BMP signaling pathways in human colorectal cancer (CRC) cell lines. Furthermore, in a human CRC – fibroblast co-culture system where gremlin-1 is produced by the fibroblasts, ginisortamab restored BMP signaling in both the CRC cells and fibroblasts, demonstrating its activity in a relevant human tumor microenvironment model. The safety and efficacy of ginisortamab are currently being evaluated in a Phase 1/2 clinical trial in patients with advanced solid tumors (NCT04393298).

KEYWORDS:

• Ab7326
• BMP signaling
• ginisortamab
• gremlin-1
• tumor microenvironment
• UCB6114

Introduction

Gremlin-1 and bone morphogenic proteins (BMPs) regulate the proliferation and differentiation of stem cells.¹ BMPs are secreted growth factors belonging to the transforming growth factor β (TGFβ) family.² Enhanced BMP signaling via SMAD 1/5/8 upregulates a large number of genes, including the inhibitors of DNA-binding (ID) proteins ID1, ID2, and ID3, and favors terminal stem cell differentiation.² Suppressed BMP signaling engendered by BMP antagonists, such as gremlin-1, favors dedifferentiation and a stem cell phenotype.¹ The extracellular interaction with gremlin-1 prevents BMPs-2, −4, and −7 from binding to their cell surface receptors, thereby blocking signal transduction from the BMP receptor and subsequent SMAD phosphorylation.^3,4

Initially described in 1998,⁵ gremlin-1 is highly conserved between species.⁶ Gremlin-1 is important in embryonic limb, kidney, and lung development,⁷ as well as in adult brain, ovary, bone, and intestinal tissues.^8–10 Gremlin-2/protein related to Dan and Cerberus (PRDC) has also been identified that, while having a relatively low protein sequence similarity, has a shared domain structure with gremlin-1 and also acts as an antagonist of BMPs-2,-4, and −7.^11,12

Aberrant expression of gremlin-1 has been associated with a wide range of diseases, with preclinical studies showing that gremlin-1 has a role in the carcinogenesis, progression, and invasion of multiple cancers.^13–15 Gremlin-1, secreted by tumor stroma (or, in some cases, the tumor cells themselves), antagonizes BMP signaling, thereby promoting tumor cell expansion, renewal, and a more invasive mesenchymal phenotype.^16,17 Gremlin-1 is upregulated in multiple solid tumor types, with high mRNA levels being observed in >60% of colorectal, pancreatic, and esophageal cancers and in approximately 50% of bladder, breast, and lung cancers.¹⁸ Loss of the tight balance between gremlin-1 and BMP signaling can lead to hereditary mixed polyposis syndrome, resulting in early colorectal cancer (CRC) development if left untreated.^19,20 In contrast, there is little evidence in the literature for involvement of gremlin-2 in cancer, and our analysis of gene expression profiling interactive analysis (GEPIA)²¹ data suggests it is typically downregulated in many cancer types relative to normal tissue (data not shown).

Investigation of the impact of gremlin-1 inhibition in cancer has been limited by a lack of inhibitory therapeutic agents for this target. One in vivo study using an antigremlin-1 antibody displayed efficacy in a chronic hypoxia/SU5416-induced pulmonary arterial hypertension model in mice.²² However, to our knowledge, this has not been developed clinically and does not appear to have been evaluated in oncology models. Another recent report described an anti-gremlin-1 antibody that is efficacious in mouse models of castration resistant prostate cancer.²³

Here, we describe the discovery and in vitro characterization of ginisortamab (UCB6114) and antibody (Ab) 7326 mouse immunoglobulin G1 (mIgG1). Ginisortamab is a first-in-class, fully human IgG4P monoclonal antibody developed to block gremlin-1 binding to BMPs-2, −4, and −7. To carry out mouse in vivo model studies, the chimeric Ab7326 mIgG1 was created, which maintains the same ginisortamab human variable regions cloned with mouse IgG1 constant regions. Ab7326 mIgG1 has been previously shown to limit multiple myeloma disease progression,²⁴ and to reduce CRC tumoroid growth in mouse models.²⁵ Ginisortamab is currently being evaluated in a Phase 1/2 clinical trial for the treatment of advanced solid tumors (ClinicalTrials.gov, NCT04393298; EudraCT, 2019–002598–78).

Results

Discovery of ginisortamab from naïve human phage display library and generation of Ab7326 mIgG1

Attempts to generate antibodies that block BMP binding to gremlin-1 through immunization of rats and rabbits (data not shown) successfully resulted in the production of antibodies that bound to gremlin-1, but they did not restore BMP signaling in the ID1-luciferase reporter assay described below. In order to identify rare functional gremlin-1/BMP blocking antibodies, a large naïve single-chain variable fragment (scFv) library was panned. Ginisortamab was derived from the human antibody variable regions discovered from the library bio-panned with successive rounds against purified human and mouse gremlin-1 protein. Binding specificity was determined using ELISA against human and mouse gremlin-1 to select a cross-reactive clone. The variable regions were sub-cloned into mouse or human antibody expression vectors to produce full-length IgG or fragment antigen-binding (Fab) constructs for characterization in different assays.

For human trials, ginisortamab has been developed as a fully human IgG4P antibody. The nomenclature IgG4P signifies that the IgG4 heavy chain sequence has been modified in its hinge region by mutating serine at position 241 to proline to reduce the occurrence of chain exchange (separation of the two heavy chain-light chain pairs of the IgG followed by reassociation with equivalent heavy-light chain pairs of IgG4 of different specificity).

As described in subsequent assays, the variable regions of ginisortamab cross-react with mouse and rat gremlin-1. However, to carry out mouse in vivo model studies, the chimeric Ab7326 mIgG1 was generated, which maintained the same ginisortamab human variable regions against gremlin-1 cloned with a mouse IgG1 constant region. This modification confers a longer half-life in rodents and reduces the chance of anti-drug antibodies developing in long-term pre-clinical studies. The mouse IgG1 isotype also has similarly low antibody-dependent cellular cytotoxic activity to the human IgG4 isotype.²⁶

Ginisortamab and Ab7326 mIgG1 are equally potent at restoring BMP signaling in a reporter assay

To determine the activity of anti-gremlin-1 antibodies, a BMP signaling reporter assay was used (human embryonic kidney [HEK]293 cells with an ID1-luciferase reporter). This cell line allowed quantitation of BMP signaling by measuring the levels of luciferase produced, which is proportional to the level of BMP signaling.

The half maximal effective concentration (EC50) of BMP-4/7 heterodimer was used to determine the concentration of exogenous gremlin-1 used in the assay (gremlin-1 concentration used was the lowest concentration of protein that reproducibly achieved 100% inhibition of BMP signaling at its EC50). Ginisortamab or Ab7326 mIgG1 were titrated into the assay, and the percentage restoration of the BMP-4/7 signal was determined to establish potency of the antibody against gremlin-1. Figure 1 shows representative concentration-response curves of ginisortamab in the reporter assay against (a) human gremlin-1, (b) mouse gremlin-1, and (c) a representative concentration-response curve of Ab7326 mIgG1 in the reporter gene assay against mouse gremlin-1. The calculated mean half maximal inhibitory concentration (IC50) values are summarized in Table 1. The data showed that ginisortamab inhibited both human and mouse gremlin-1 in a concentration-dependent manner, with mean IC50 values of 8.2 nM and 9 nM, respectively. The chimeric equivalent anti-gremlin-1 antibody Ab7326 mIgG1 also inhibited mouse gremlin-1 in a dose-dependent manner with a mean IC50 value of 18.5 nM, which was comparable to ginisortamab against human and mouse gremlin-1 (Figure 1c and Table 1). To confirm that this antibody could similarly restore BMP homodimer signaling that was previously antagonized by gremlin-1, Ab7326 mIgG1 was tested in the reporter assay against human gremlin-1 with BMP-4 or BMP-7 homodimers. Gremlin-1 inhibited signaling induced by both BMP-4 and BMP-7, and Ab7326 mIgG1 was able to fully restore this signal (Supplementary Figure S1).

Figure 1. Representative concentration-response curves in the BMP signaling reporter gene assay of ginisortamab against human gremlin-1, mouse gremlin-1, and Ab7326 mIgG1 against mouse gremlin-1. Ginisortamab can inhibit both (a) human and (b) mouse gremlin-1 in a concentration-dependent manner with mean IC50 values (n = 5) of 8.2 nM and 9 nM, respectively. Ab7326 mIgG1 also inhibits mouse gremlin-1 (c) in a dose-dependent manner with mean IC50 values (n = 5) of 18.5 nM, which is comparable to ginisortamab against human and mouse gremlin-1 in this assay system with the concentrations of ligands used within the detection limits of the assay.

Ab: antibody; BMP: bone morphogenetic protein; mIgG1: mouse immunoglobulin G1; IC50: half maximal inhibitory concentration.

Table 1. Ginisortamab IC50 for the inhibition of gremlin-1 antagonism of BMP-4/7 heterodimer activity, as measured in the BMP signaling reporter gene assay.

Antibody	Protein	Geometric Mean IC50 (95% CI), nM (N = 5)
Ginisortamab	Human gremlin-1	8.2 (5.7–11.7)
Ginisortamab	Mouse gremlin-1	9.0 (5.4–15.1)
Ab7326 mIgG1	Mouse gremlin-1	18.5 (10.6–32.4)

Ab: antibody; BMP, bone morphogenetic protein; CI: confidence interval; IC50: half maximal inhibitory concentration; mIgG1: mouse immunoglobulin G1.

Ginisortamab and Ab7326 mIgG1 have comparable affinity for mouse and human gremlin-1 by surface plasmon resonance (SPR)

The kinetic values for human and mouse gremlin-1 binding to immobilized ginisortamab are summarized in Table 2 (n = 3; sensorgrams shown in Supplementary Figure S2). The geometric mean affinities (equilibrium dissociation constant [KD]) of ginisortamab for human and mouse gremlin-1 were 87 pM and 61 pM, respectively.

Table 2. Affinity of ginisortamab binding human and mouse gremlin-1, as demonstrated by SPR analysis.

Sample	Mean Ka x 10^7 (M^−1s^−1)	Mean Kd x 10^−3 (s^−1)	Geometric mean KD (pM)
Human gremlin-1	1.90	1.58	87.0
Mouse gremlin-1	3.56	2.19	61.0

Ka: association rate constant; Kd: dissociation rate constant; KD: equilibrium dissociation constant; SPR: surface plasmon resonance.

The affinity of Ab7326 mIgG1 binding mouse gremlin-1 was also determined. The kinetic values for mouse gremlin-1 binding to immobilized Ab7326 mIgG1 are summarized in Table 3 (n = 3). The KD value for mouse gremlin-1 binding to Ab7326 mIgG1 was 52.8 pM.

Table 3. Affinity of Ab7326 mIgG1 binding mouse gremlin-1, as demonstrated by SPR analysis.

Sample	Mean Ka x 10^Citation7 (M^−1s^−1)	Mean Kd x 10^−3 (s^−1)	Geometric mean KD (pM)
Mouse gremlin-1	2.33	1.17	52.8

Ab: antibody; Ka: association rate constant; Kd: dissociation rate constant; KD: equilibrium dissociation constant; mIgG1: mouse immunoglobulin G1; SPR: surface plasmon resonance.

Ab7326 Fab/human gremlin-1 crystal structure reveals BMP blocking mechanism

The gremlin-1/Ab7326 Fab structure (PDB code 8B7H) revealed a gremlin-1 dimer with a Fab molecule binding to each gremlin-1 monomer. The gremlin-1 structure was as described previously (5AEJ),²⁷ forming an antiparallel, non-covalent dimer arranged in an arch. Each monomer adopts the characteristic finger-wrist-finger arrangement with a cysteine-knot motif toward the wrist end opposite the fingers (Figure 2). The Ab7326 Fab contacts gremlin-1 mainly through Finger 2, perpendicular to the axis of the gremlin-1 dimer (Figure 2a). The Ab7326 Fab uses the heavy chain and light chain to bind to gremlin-1 and binds predominantly to one monomer, burying a surface area of ~1700 Ų. The binding interface is mainly located at the base of the fingers close to the cysteine-knot motif. The finger and wrist loops (residues I82–S89, Finger 1; P138–K145, Finger 2; and H109–E113, wrist) were unresolved in the final model due to their structural flexibility. These loops were resolved in the structure of gremlin-1 alone as they were stabilized by crystal contacts.²⁷

Figure 2. Interaction of Ab7326 Fab with gremlin-1. (a) cartoon representation of the gremlin-1/Ab7326 Fab complex (PDB code 8B7H). Gremlin-1 dimer (cyan and dark blue) with finger and wrist regions labeled (F1, F2, W). One copy of Ab7326 Fab (heavy chain: orange; light chain: yellow) binds each monomer of the gremlin-1 dimer. (b) close-up of the paratope-epitope interactions with the CDR loops of the heavy and light chains color coded and labeled (HCDR1: magenta; HCDR2: light purple; HCDR3: green; LCDR1: red; LCDR2: brown; LCDR3: wheat). (c) key residues involved in the interactions of HCDR3 loop with gremlin-1 are shown in sticks and labeled. The dotted lines indicate hydrogen bonds.

Ab: antibody; CDR: complementarity determining regions; Fab: fragment antigen-binding; F: finger region; HC: heavy chain; HCDR: heavy chain CDR; LC: light chain; LCDR: light chain CDR; PDB: Protein Data Bank; W:wrist region.

In the gremlin-1/Ab7326 Fab crystal, the crystal contacts are through the Fab molecules, and the gremlin-1 dimers sit unrestricted by crystal contacts in solvent channels (the only gremlin-1 contacts being those with the complementarity determining region [CDR] loops of Ab7326 Fab), revealing the flexibility of the finger loops (Supplementary Figure S3a). In addition, the structural flexibility of the gremlin-1 molecule is highlighted by the B-factor analysis of the complex (Supplementary Figure S3b). The majority of Fab contacts are through the heavy chain CDR (HCDR) loops, which form a series of interactions with gremlin-1 residues (Figure 2b). Several interactions are with the HCDR3 loop (S105, Y106, Y107), which makes hydrogen bond interactions with the backbones of K126, F128, and T130 of gremlin-1, as well as hydrophobic interactions with the side chain of F128 (Figure 2c). Other interactions are with the HCDR1 loop (T30, D31, Y32, Y33), which mainly forms hydrogen bonds with the backbone of F128 and the side chains of K126 and K127 (Supplementary Figure S4). The HCDR2 loop (D52, E54, D55) also contacts gremlin-1 through interactions mainly with the side chains K126 and K153. The light chain CDR (LCDR) loops make contact farther away from the cysteine-knot motif, toward the end of the fingers and the other monomer. The LCDR1 loop (Y31, S33, Y38) forms hydrogen bond interactions with both gremlin-1 monomers, with the backbone of R148 of one monomer and the backbone of I110 of the other, and hydrophobic interactions with the side chain of R148. The LCDR3 loop (Y97, Y98) also contacts the side chain of R148, mostly through a hydrogen bond network with the backbone of Y97 and Y98. There is no contact with the LCDR2 loop.

The gremlin-1/Ab7326 Fab structure was used in combination with the published gremlin-2/growth differentiation factor 5 (GDF5) structure (5HK5)^28,29 to highlight how Ab7326 blocks BMP binding. GDF5 is a member of the BMP family with a low affinity for gremlin-2 and is currently the only BMP to have had a crystal structure solved showing its interaction with a gremlin protein. Its physiological relevance to gremlin-1 or gremlin-2 biology is unknown, however the GDF5 binding site on gremlin-2 has been demonstrated to use the same key residues as its high affinity ligand BMP-2, which furthermore are conserved in gremlin-1.²⁹ Thus gremlin-2/GDF5 is a relevant model structure to use to elucidate the mechanism by which Ab7326 may block gremlin-1 interactions with BMPs. The epitope of Ab7326 consists of residues K126, K127, F128, T129, T130, R148, K153, and Q154 on one gremlin-1 monomer and I110 on the second gremlin-1 monomer (Figure 3). To understand the positioning of this epitope in relation to the GDF5 epitope on gremlin-1, we modeled the GDF5 epitope onto gremlin-1 using the gremlin-2/GDF5 complex structure. The overall sequence identity of gremlin-1 and −2 was 53% rising to 71% when ignoring the less well conserved N-terminal region (Supplementary Figure S5). The gremlin-2/GDF5 structure and mutagenesis studies highlighted regions on gremlin-2 that were critical for BMP binding, including key residues (F104, I106, and F117) that are conserved in gremlin-1²⁹ (Figures 3a and S5). After mapping the GDF5 epitope onto gremlin-1, it was clear that there was no direct overlap of the GDF5 and Ab7326 Fab epitopes (Figure 3a). The two epitopes were, however, adjacent and when GDF5 was modeled onto the gremlin-1/Ab7326 Fab structure, there was a clear clash between the Ab7326 and GDF5 (Figure 3b). This clarifies that Ab7326, although binding a different epitope to GDF5, would block its interaction with gremlin-1.

Figure 3. Ab7326 sterically blocks gremlin-1 ligand interaction. (a) surface rendered image of the gremlin-1 dimer (cyan and dark blue) from the gremlin-1/Ab7326 Fab structure (Fabs removed for clarity) with the Ab7326 (red) and GDF5 (green) epitopes highlighted. Close-up shows epitope residues for Ab7326 (red) and key epitope residues for GDF5 (green²⁹ in sticks and labeled. (b) gremlin-1 dimer (cyan and dark blue, surface rendered) with GDF5 dimer (green cartoon, modeled using the gremlin-2–GDF5 structure 5HK5²⁹) Ab7326 Fab (orange: heavy chain and yellow: light chain, surface rendered) bound. Close-up shows clash of modeled GDF5 with Ab7326 Fab.

Ab: antibody; Fab: fragment antigen-binding; GDF: growth differentiation factor.

Given the structural conservation of the TGFβ superfamily³⁰ and data showing that mutation of key residues on gremlin-2 (F104, I106, and F117) disrupts both GDF5 and BMP2 binding,²⁹ it is reasonable to assume that other known gremlin-1 ligands (BMP-2, −4, and -7) will have a similar binding mode to that seen in the gremlin-2/GDF5 structure and that Ab7326 would also block their interaction with gremlin-1.

Ginisortamab inhibits gremlin-1 antagonism of BMP signaling pathways in human CRC cell lines

To demonstrate that ginisortamab has activity against endogenous human gremlin-1 and can modulate BMP pathway signaling in human CRC cells, we sought to develop an in vitro human CRC cell line assay system. We first characterized the expression of gremlin-1 and relevant BMP pathway components in three human CRC cell lines, HCT 116, DLD-1, and LS174T (Supplementary Figure S6). Gene expression analysis of HCT 116 by qPCR showed that this line had negligible expression of gremlin-1, but higher expression of BMP-2, −4, −6, and −7 (BMP-6 also signals through SMAD1/5/8).³¹ DLD-1 showed a similarly low expression of gremlin-1 and generally higher BMP gene expression. LS174T expressed gremlin-1 as well as BMP genes, raising the possibility that this cancer cell line may directly modulate BMP signaling by autologous gremlin-1 expression (Supplementary Figure S6a). BMP ligands cause receptor activation by assembling a heteromeric receptor complex comprising both Type 1 and Type 2 receptors. BMP-2 and −4 utilize Type 1 receptors BMPR1A or 1B and Type 2 receptors BMPR2, ACVR2A, or ACVR2B. In addition to these, BMP-7 can use ACVR1 as its Type 1 receptor.³² Analysis of BMP receptor genes in the three CRC lines indicated that there was broad gene expression of both Type 1 and 2 receptors involved in BMP signaling (Supplementary Figure S6a). When surface BMPR protein expression was measured by flow cytometry, all three CRC lines clearly expressed BMPR1A, ACVR1, BMPR2, and ACVR2A, suggesting that they had the necessary receptor combinations available to allow BMP-2, −4, and −7 to bind and transmit signal (Supplementary Figure S6b).

To investigate if inhibition of gremlin-1 by ginisortamab can modulate BMP signaling in human CRC cell lines, we first measured the phosphorylation of SMAD1/8 (pSMAD1/8) as the proximal downstream signaling component of the BMPR pathway. Addition of recombinant human gremlin-1 for 16 h led to a clear reduction in pSMAD1/8 in all three CRC cell lines, thus demonstrating active BMPR signaling that can be antagonized by gremlin-1 (Figure 4a). Pre-incubation of ginisortamab with gremlin-1 before addition to cells maintained pSMAD1/8 signaling, demonstrating that ginisortamab blocked gremlin-1 antagonism of the BMP pathway in human CRC cells. Addition of ginisortamab alone to HCT 116, DLD-1, or LS147T CRC cell lines did not alter the pSMAD1/8 signal (Figure 4a). This suggests that despite endogenous gremlin-1 gene expression in LS174T cells, it is either not of sufficient concentration or activity to affect ongoing BMP signaling. We then tested whether ginisortamab can reverse preexisting gremlin-1 antagonism of BMP signaling. HCT 116 cells were incubated with gremlin-1 for 16 h, followed by addition of ginisortamab for the last 30 to 120 min before pSMAD1/8 analysis. Addition of ginisortamab for 30 min was sufficient to fully restore the pSMAD1/8 signal (Figure 4b). The concentration dependency of this effect was seen when we titrated ginisortamab against a fixed concentration of 15 nM gremlin-1 in HCT 116 cell cultures (Figure 4c). Ginisortamab was able to reverse the pSMAD1/8 inhibition induced by 16 h of incubation with gremlin-1 in a concentration-dependent manner, with an IC50 of 1.578 nM.

Figure 4. Ginisortamab inhibits gremlin-1 antagonism of BMP signaling pathways in human CRC cell lines. (a) CRC cell lines HCT 116, DLD-1, and LS174T show decreased pSMAD 1/8 in the presence of recombinant human gremlin-1, which can be inhibited by ginisortamab. Gremlin-1 (15 nM) and ginisortamab or hIgG4 control (150 nM) were pre-incubated together for 30 min before adding to CRC cells for 16 h before pSMAD 1/8 analysis by phospho-flow cytometry. Graph shows mean and individual data points of normalized % pSMAD 1/8 signal from n = 2/3 independent experiments. (b) Ginisortamab can reverse recombinant human gremlin-1 suppression of pSMAD 1/8 in HCT 116 CRC cells. HCT 116 were incubated with 15 nM gremlin-1 for 16 h before addition of 150 nM ginisortamab for 30 to 120 min. Graph shows mean and individual data points of normalized % pSMAD 1/8 signal from n = 3 independent experiments. (c) Ginisortamab exhibits dose-dependent reversal of pSMAD 1/8 inhibition. HCT 116 cells were incubated with 15 nM gremlin-1 for 16 h before addition of a titration of ginisortamab for the final 120 min before pSMAD 1/8 analysis. Graph shows individual data points of normalized % pSMAD 1/8 signal from n = 3 independent experiments and non-linear fit curve; IC50 for ginisortamab = 1.578 nM. (d) quantitative RT-PCR analysis of ID1 mRNA level in CRC cell lines stimulated with gremlin-1 in the presence of hIgG4 or ginisortamab. Fold changes are shown relative to hIgG4. Graph shows mean and individual data points from n = 3 independent experiments.

BMP: bone morphogenetic protein; CRC: colorectal cancer; hIgG4: human immunoglobulin G4; IC50: half maximal inhibitory concentration; ID: inhibitor of DNA-binding protein; mRNA: messenger ribonucleic acid; pSMAD: phosphorylated SMAD; RT-PCR: real-time polymerase chain reaction.

To understand if the recovery of BMPR signaling, as observed by an increase in SMAD1/8 phosphorylation, translates to downstream gene transcription, we measured ID1, ID2, and ID3 expression by qPCR in CRC cell lines. In each cell line, we observed that 24-h incubation with recombinant human gremlin-1 led to downregulation of ID1 (Figure 4d), ID2, and ID3 (Supplementary Figure S7) mRNA expression. When ginisortamab was co-incubated with gremlin-1, the downregulation of ID1, ID2, and ID3 was inhibited (Figures 4d and S7).

Ginisortamab restores BMP signaling in CRC cells and fibroblasts through inhibition of fibroblast-derived gremlin-1

In vitro, a high expression level of gremlin-1 was found in human primary fetal pancreatic fibroblasts (Supplementary Figure S8). To mimic the interaction between gremlin-1 expressing fibroblasts and CRC cells in the tumor microenvironment (TME), we seeded CellTracker Green 5-chloromethylfluorescein diacetate (CMFDA)-labeled CRC cell lines (HCT 116, DLD-1, or LS174T) and CellTrace Violet-labeled fibroblasts either on their own or in direct contact co-cultures for 72 h in the presence of ginisortamab or human IgG4 isotype control.

Co-cultures were sorted by fluorescence-activated cell sorting (FACS), and expression of ID genes was assessed in both sorted CRC cell lines and fibroblasts (Figure 5). ID1, ID2, and ID3 gene expression was downregulated in all three CRC cell lines in the presence of fibroblasts. Moreover, this effect was specifically inhibited by ginisortamab, thus confirming the gremlin-1-mediated antagonism of BMP target gene expression in human CRC cells. No effect of ginisortamab on ID gene expression was observed in the three CRC cell lines when cultured alone, given their lack of (HCT 116 and DLD-1) or low (LS174T) gremlin-1 expression (Figures 5 and S9).

Figure 5. Ginisortamab inhibits endogenous gremlin-1-mediated antagonism of BMP target gene expression in human CRC cells and fibroblasts. (a) Representative immunofluorescence images of CellTracker Deep red-labeled and CellTracker green CMFDA-labeled HCT 116 cells cultured alone or in direct co-culture. Cultures were imaged 72 h after seeding. Representative flow cytometry dot plots showing the gating strategy to isolate CellTracker green CMFDA-labeled CRC cells and CellTrace violet-labeled fibroblasts from co-cultures. TO-PRO3 was added to discriminate between viable and dead cells. (b) quantitative RT-PCR analysis of ID1 mRNA level in CRC cell lines and fibroblasts cultured on their own or in direct co-culture in presence of 150 nM hIgG4 or ginisortamab. CRC cell lines and fibroblast co-cultures were incubated for 72 h and sorted prior to mRNA analysis to isolate pure populations of CRC cell lines (sorted CRC cells) and fibroblasts (sorted ‘Fbs). Fold changes in CRC cell lines and fibroblasts are shown relative to either CRC cell lines or fibroblasts alone treated with hIgG4, respectively. The left y-axes depict gene expression change in CRC cells, and the right y-axes depict gene expression change in fibroblasts. Graphs show mean and individual data points from n = 3 independent experiments.

BMP: bone morphogenetic protein; CRC: colorectal cancer; CMFDA, 5-chloromethylfluorescein diacetate; DAPI, 4′,6-diamidino-2-phenylindole; FACS: fluorescence-activated cell sorting; Fb: fibroblast; hIgG4: human immunoglobulin G4; ID: inhibitor of DNA-binding protein; mRNA: messenger ribonucleic acid; RNA: ribonucleic acid; RT-PCR: real-time polymerase chain reaction.

Interestingly, we observed an upregulation of ID1 and ID3 expression in fibroblasts when seeded in direct cell–cell contact with CRC cells compared with those seeded on their own, probably due to the exposure to higher levels of CRC-derived BMPs. This effect was exacerbated by ginisortamab treatment in both fibroblasts alone and in direct co-cultures with CRC cells. These results show that ginisortamab treatment, through inhibition of gremlin-1, induces BMP signaling pathways in both CRC cell lines and fibroblasts.

Discussion

Gremlin-1 has been a prospective target of interest in oncology and other diseases for several years. The lack of a human therapeutic antibody to gremlin-1 has precluded testing the effects of blocking the gremlin-1–BMP interaction and restoring BMP signaling in the oncology setting. Here, we present ginisortamab, the first anti-human gremlin-1 antibody to be developed for the potential treatment of cancer. Discovered using a human-derived phage display library panned against human gremlin-1 protein, ginisortamab binds and inhibits the antagonistic function of human and mouse gremlin-1 protein.

A truncated form of gremlin-1 (A43–D160) was used for crystallography, as attempts to crystallize the gremlin-1/Ab7326 complex with full length protein were unsuccessful. The truncated gremlin-1 protein was successfully crystallized alone (data not shown) and aligned closely with the published structure, 5AEJ (RSMD of 0.721); all amino acids (aa) were clearly resolved except for the N-terminal 6aa (V49-D160 were visible). In contrast, significant flexibility of gremlin-1 was observed in the Ab7326 Fab complex structure where gremlin-1 sits unconstrained in solvent channels (Supplementary Figure S3a). Several regions were unresolved including the finger loops and the N-terminal 25aa, highlighting the dynamic nature of gremlin-1. Flexibility in the N-terminus is necessary as it undergoes a significant conformational change when BMP ligands bind.²⁸ In addition, the overall B-factors for the gremlin-1 dimer in the Ab7326 Fab complex structure were high (Supplementary Figure S3b) highlighting the flexibility in this region of the crystal lattice. However, the structure clearly revealed the contact epitope of Ab7326 on gremlin-1, enabling its mechanism of action to be determined.

The specificity and pharmacological activity of ginisortamab and the mouse chimeric Ab7326 mIgG1 were characterized in several in vitro studies. In vitro binding assays and cell experiments demonstrated that ginisortamab and Ab7326 mIgG1 bound with high affinity to human and mouse gremlin-1. Both antibodies inhibited the interaction of gremlin-1 with BMPs, resulting in restoration of the BMP signal in a reporter assay.

Cancer cells reside in a complex milieu of cells including stromal cells, parenchymal cells, endothelial cells, immune/inflammatory cells, and extracellular matrix, collectively known as the TME.³³ Homotypic and heterotypic interactions between the different cells within the TME can profoundly affect the development and the progression of cancer. Previous studies have demonstrated high levels of gremlin-1 in CRC patients, specifically within the stromal compartment of the TME, including cancer-associated fibroblasts.³⁴ To investigate the effects of ginisortamab on human CRC cell lines, a co-culture system mimicking human CRC – fibroblast interactions in the TME was developed, enabling endogenous BMP signaling and antagonism by gremlin-1.

All three of the human CRC cell lines analyzed in our studies constitutively expressed BMPs, Type 1 and 2 BMP receptors, and had active BMP signaling that could be inhibited by gremlin-1 as measured by SMAD1/8 phosphorylation and ID gene expression assays. Furthermore, the human primary fibroblast line used here endogenously expressed gremlin-1, allowing us to measure the effects of ginisortamab in co-cultures without addition of recombinant BMPs or gremlin-1, thereby providing a much closer mimic of physiological conditions. Exposure of CRC cell lines to gremlin-1-secreting fibroblasts in this direct cell–cell co-culture system impaired activation of BMP signaling, as measured by ID gene expression. Treatment of co-cultures with ginisortamab blocked fibroblast-secreted gremlin-1 and increased BMP availability, restoring BMP signaling pathways in both CRC cells and fibroblasts. These results demonstrate the potent ability of ginisortamab to modulate the BMP signaling pathway in human cancer cell lines.

In CRC cells, BMP signaling has been shown to promote tumor-suppressive functions, including reduction of cancer cell proliferation, invasion, and motility, as well as inhibition of epithelial-mesenchymal transition.¹⁶ This tumor-suppressive activity may be attributed to BMP signaling driving the differentiation of cancer stem cells (CSC), limiting their proliferation and self-renewal properties. In healthy intestines, high BMP concentrations restrict stemness of Lgr5+ stem cells and maintain epithelial homeostasis by preventing premalignant hyperproliferation in response to damage.³⁵ In agreement with this mechanism, a number of studies have identified gremlin-1 as a stemness factor that is expressed in tumors to promote CSC self-renewal and maintain a pool of CSCs, despite the presence of morphogen BMPs.^36,37 CSC sub-populations are resistant to DNA-damaging chemotherapy and radiotherapy treatments, and are therefore considered responsible for chemoresistance, recurrence, and metastasis.^38–40 This suggests that targeting stemness factors such as gremlin-1 in the TME could be highly beneficial in enhancing the efficacy of current cancer treatments.⁴⁰

Experimental evidence of the functional anti-tumor effect of Ab7326 was initially documented in a mouse model of CRC.²⁵ Blocking gremlin-1-mediated BMP antagonism using Ab7326 mIgG1 promotes colorectal tumoroid differentiation and restrained tumoroid growth by downregulating stemness genes such as Lgr5 and upregulating pro-differentiation genes such as Krt20.²⁵ Restoration of BMP signaling in CRC cells and fibroblasts by ginisortamab may therefore inhibit cancer progression and metastasis by switching tumor cells from a stem cell to a differentiated phenotype. Recently, the efficacy of Ab7326 has been examined in vivo in the 5TGM1/KaLwRij mouse model of multiple myeloma. These studies showed that mice treated with Ab7326 had a decrease in multiple myeloma whole body tumor burden of up to 81.2% when treated prophylactically.²⁴ Overall, these data suggest that inhibition of gremlin-1 activity by ginisortamab may potentially lead to a novel treatment for patients with cancer.

In conclusion, ginisortamab is a potential first-in-class, high-affinity, anti-human, gremlin-1 antibody that demonstrates the ability to restore gremlin-1-suppressed BMP signaling. These data, together with the growing body of mouse model data,^24,25 provided the rationale to evaluate the safety, pharmacokinetics, and anti-tumor activity of ginisortamab in the ongoing Phase 1/2 clinical trial for the treatment of participants with advanced solid tumors (NCT04393298).

Materials and methods

Discovery of ginisortamab using a phage bio-panning campaign

Ginisortamab was discovered using a phage bio-panning campaign designed to identify cross-species-reactive clones using recombinant gremlin-1 protein expressed and purified from supernatants of murine myeloma NS0 cells (R&D Systems, 5190-GR and 956-GR, for human and mouse proteins, respectively). A large naïve library of approximately 1000 E. coli TG1 transformants (UCB Pharma) was used for antibody discovery. This library was constructed using natural human variable region sequence templates of IgG- and IgM-derived variable heavy chains, IgK- and IgL-derived light chains, in a scFv format and was constructed within a pUC119-based phagemid vector. Library aliquots were rescued using M13KO7 helper phage (Thermo 18311019) to provide predominantly monovalent display upon phage virions. Briefly, anti-gremlin-1 antibodies were enriched from the library over three rounds of bio-panning using Nunc-Immunotubes (Thermo 444202), wherein gremlin-1 was passively adsorbed to the surface of the tubes at a concentration of 10 µg/mL in phosphate-buffered saline (PBS). The presence of gremlin-1 binding scFv-displaying phage particles was revealed using the anti-c-myc tag mouse antibody 9E10 (Thermo, 13–2500) and the appropriate anti-mouse-IgG-horseradish peroxidase (HRP) conjugate (Jackson ImmunoResearch, 115-036-071).

The DNA encoding the scFv fragment was sub-cloned into mammalian expression vectors for transient expression from HEK293F cells (as derivatives of the pIMMS-2 vector), as N-terminal fusions with the mouse IgG1-crystallizable fragment (Fc [mFc]) region (UniProtKB/Swiss-Prot: P01868 IGHG1_MOUSE) Gly103 – Lys324 contains the core hinge disulfides for further binding and functional assays. VR4405 was selected as having the characteristics desired for progression as a therapeutic candidate, and further assessments were made following sub-cloning of the variable region sequences separately into full length human and mouse IgG formats.

Cloning of Ab7326 mIgG1 from ginisortamab variable regions

To generate human gamma-4 and mouse gamma-1 full-length immunoglobulins for further characterization, the nucleotide sequence encoding the variable heavy and light chain from VR4405 was sub-cloned into the corresponding vectors for HEK293F or CHO-SXE cell expression via transient transfection, using chemical or electroporation methods, respectively.

BMP dependent reporter assay to test for restoration of BMP signaling

To measure effects of activity of anti-gremlin antibodies on BMP signaling, HEK-ID1 luciferase reporter cells⁴¹ were plated in poly-D-lysine coated 96-well plates in assay media (Dulbecco’s Modified Eagle Medium [DMEM] with 0.5% fetal bovine serum [FBS], 1X L-glutamine, and 1X non-essential amino acids) and incubated at 37°C, 5% CO₂ until adhered. Antibody was prepared in assay media and titrated threefold across the plate. Human gremlin-1 (R&D Systems, 5190-GR) or mouse gremlin-1 (R&D Systems, 956-GR) protein was added to the antibody titration and pre-incubated for 45 min at 37°C, 5% CO₂, followed by a second pre-incubation with BMP-4/-7 heterodimer (R&D Systems, 3727-BP), BMP-4 (R&D Systems, 314-BP), or BMP-7 (R&D Systems, 354-BP) for 45 min at 37°C, 5% CO₂, prior to addition to the cells. Cultures were incubated for 24 h at 37°C with 5% CO₂. Luciferase signal per well was detected using Steady Glo (Promega, E2510). Restoration of the maximal BMP signal determined in the absence of gremlin-1 was measured and plotted as mean ± standard deviation from quadruplicate measurements.

SPR assay to test affinity of ginisortamab and Ab7326 mIgG1 for gremlin-1

Ginisortamab or Ab7326 mIgG1 were immobilized, and human or mouse gremlin-1 were titrated, respectively. The affinity of ginisortamab – gremlin-1 binding was determined by SPR using the Biacore T200 (GE Healthcare Biosciences AB) at 25°C. Ginisortamab (in 10 mM NaAc, pH 4.5) was immobilized on a CM4 Series S Sensor Chip (GE Healthcare Biosciences AB) via amine coupling chemistry using the option “Aim for Immobilized Level” to achieve approximately 200–300 response units. HBS-EP+ buffer (10 mM HEPES pH 7.4, 0.15 M NaCl, 3 mM ethylenediaminetetraacetic acid [EDTA], 0.05% Surfactant P20; GE Healthcare Biosciences AB) was used as the running buffer with a flow rate of 10 µL/min. A reference surface was prepared by activating and deactivating the appropriate flow cell. HBS-EP+ buffer with an extra 150 mM NaCl (final concentration, 300 mM NaCl) was used as the running buffer for the affinity assay. Human and mouse gremlin-1 were titrated from 10 nM to 0.625 nM over the immobilized ginisortamab and reference flow cell for 120 s at a flow rate of 30 µL/min followed by 300-s dissociation. A buffer control (0 nM) was run at the start and end of each titration set. The surface was regenerated by a 60-s injection of 10 mM HCl at a flow rate of 10 µL/min. Binding assays involving mouse gremlin-1 and Ab7326 mIgG1 were performed as above, with the exception that Ab7326 mIgG1 was immobilized on the sensor chip surface. Background subtraction binding curves were analyzed using the Biacore T200 Evaluation Software (Version 3.0; GE Healthcare Biosciences AB) using 1:1 binding fitted with local Rmax.

Human gremlin-1 protein production for crystallography

Truncated human gremlin-1 (Ala43-Asp160), optimized for expression in E. coli, was cloned into the BamHI and XhoI sites of a modified pET32a vector (Merck Millipore), generating a vector encoding the gremlin-1 sequence with an N-terminal 6His-TEV tag.

Human gremlin-1 was expressed in E. coli strain BL21 (DE3). Bacteria were grown in 2×TY/Amp media at 37°C until exponential phase (OD₆₀₀ of 3), after which the culture was supplemented with a MOPS and glycerol feed mix. Protein expression was induced by addition of 300 μM IPTG and cells incubated at 17°C, 180 rpm for 16 h. Cells were harvested and resuspended in lysis buffer at 4°C. The insoluble fraction was harvested by centrifugation at 3500 g for 30 min at 4°C. Pelleted inclusion bodies were washed three times by resuspending in wash buffer containing 0.5% Triton X-100, followed by centrifugation at 21,000 g for 15 min. An additional two washes were performed using wash buffer without Triton X-100. Inclusion bodies were resuspended in 8 M urea denaturing buffer, stirred for 16 h at room temperature, and clarified by centrifugation at 21,000 g for 15 min. The solubilized inclusion bodies were purified, under denaturing conditions, using a two-step purification protocol and then refolded using rapid-dilution method. After 5 days of refolding, the gremlin-1 protein was dialyzed against 20 mM HEPES, 100 mM NaCl, pH 7.5 and further purified using heparin affinity and size-exclusion chromatography. Purified/refolded human gremlin-1 protein was characterized by sodium dodecyl-sulfate polyacrylamide gel electrophoresis (SDS PAGE) and demonstrated to have the expected molecular weight and correct arrangement of disulfide bonds using liquid chromatography mass spectrometry and to be active in a cell assay (ID1 reporter assay).

Expression and purification of Ab7326 fab for crystallography

The Ab7326 Fab was expressed in CHO-SXE cells and purified by MabSelect SuRe (Cytiva 17543803) protein A affinity, and size exclusion chromatography to isolate 100% monomer.

Gremlin-1/Ab7326 fab complex formation and structure determination

For complex formation, a 1:1 gremlin-1/Ab7326 Fab complex was made and purified by size exclusion. The isolated complex was concentrated to 6 mg/mL prior to crystallization trials. Gremlin-1/Ab7326 Fab complex crystals were grown using the hanging-drop method by mixing a solution of gremlin-1/Ab7326 Fab complex at 6 mg/mL and 0.2 M potassium fluoride, 20% polyethylene glycol 3350, and 20% glycerol in a 1:1 ratio. Diffraction data were collected at the Diamond Light Source (Harwell Science and Innovation Campus) and were processed using X-Ray Detector Software (XDS).⁴²

Gremlin-1/Ab7326 Fab structure was solved by molecular replacement using Phaser,⁴³ published mouse gremlin-2 structure coordinates,²⁸ and a Fab model available from proprietary Fab coordinates. The resultant model of gremlin-1/Ab7326 Fab complex contained a copy of gremlin-1 dimer with 2 Fabs (Figure 2). Model corrections were made with Coot⁴⁴ and coordinates were refined using Refmac⁴⁵ and Phenix.⁴⁶ Final coordinates were validated with Molprobity.⁴⁷ A summary of model refinement statistics is shown in Table 4.

Table 4. Data collection and refinement statistics of X-ray data used for gremlin-1/Ab7326 Fab complex (PDB code: 8B7H).

Diffraction Statistics
Wavelength (Å)	0.97949
Space group	C2221
Cell dimensions	a = 65.74 Å, b = 80.65 Å, c = 252.26 Å; α = 90°, β = 90°, γ = 90°
Resolution range^a (Å)	50.96–1.95 (2.00–1.95)
Completeness (%)	99.3 (98.7)
Multiplicity	5.6 (5.6)
I/sigma	18.3 (2.7)
Rmerge	0.051 (0.604)
Refinement Statistics
Resolution range (Å)	50.96–1.95
Rcryst	0.2651
Rfree	0.2750
No. non-H atoms	4039
Protein	3977
Ligand	0
Water	62
Average B-factor (Å^2)	60.13
Protein	60.50
Solvent	36.66
RMSD bonds (Å)	0.011
RMSD angles (°)	1.560
Ramachandran plot
Favored (%)	90.8
Allowed (%)	8.4
Outliers (%)	0.8
Rotamer outliers (%)	0.88

^aValues in parenthesis correspond to the highest resolution shell.

Ab: antibody; Fab: fragment antigen-binding; H: hydrogen; No, number; PDB: Protein Data Bank; Rcryst: crystallographic R-factor; Rfree: free R-factor; Rmerge: measure of agreement among multiple measurements of the same reflection; RMSD: root-mean-square deviation.

Human CRC and fibroblast cell line culture

Human CRC cell lines HCT 116 (colorectal carcinoma) and DLD-1 (Dukes’ type C colorectal adenocarcinoma) were purchased from AddexBio (distributed via Caltag, C0009005 and C0009007, respectively), and LS174T (Dukes’ type B colorectal adenocarcinoma) was purchased from Sigma (87060401). All CRC lines were maintained in culture in DMEM media (Gibco 21969–035) supplemented with 10% FBS (Gibco 16140–071) and Glutamax (Gibco 35050–038), and passaged using TrypLE Express dissociation solution (Gibco 12604–013). Human primary pancreatic fibroblasts were purchased from Neuromics (SC00A5, lot 005C) and human primary rectal fibroblasts were purchased from Sciencell (2960, lot 16865). Fibroblast lines were maintained in culture in MSC-GRO Vitro Plus III low serum complete medium (Neuromics, PC00B1), passaged using TrypLE Express, and used for experiments within 10 passages. All cell lines were routinely tested and found negative for mycoplasma.

pSMAD1/8 phospho-flow assay on CRC cell lines

Cells were plated at 1.25 × 10⁵ per well in a 24-well plate (Corning 353047) and incubated for 24 h in complete DMEM (10% FBS/Glutamax) before addition of 15 nM (300 ng/mL) recombinant human gremlin-1 (R&D Systems, 5190-GR) or media-only control for 16 h. Next, 150 nM (22.5 µg/mL, or a titration, as indicated) ginisortamab or a matched human IgG4 isotype control (Biolegend 403702) were either pre-incubated together with gremlin-1 for 30 min before addition to cells or added for the last 30 to 120 min of the 16-h gremlin-1 incubation (as indicated in Figure 4). At the end of the assay, cells were fixed, permeabilized, and stained for pSMAD1/8 according to the BD Phosflow protocol. Briefly, cells were detached with TrypLE Express and transferred to a 96-well plate (Corning, 3799). Cells were fixed with Lyse/Fix (BD, Phosflow 558049) for 15 min at 37°C, washed with PBS (Gibco) then permeabilized with cold Perm III (BD, Phosflow 558050) for 30 min on ice. Cells were washed with stain buffer (PBS, 2% FBS, 0.09% NaN₃) and incubated for 1 h at room temperature with a 1:20 dilution of PE-labeled anti-SMAD1 (pS436/pS465)/SMAD8 (pS465/pS467) (BD 562509) before washing again with stain buffer and analyzed on a BD FACS CANTO II flow cytometer. Data analysis was performed using FlowJo v10.6.0 and GraphPad Prism v8.1.1 software. Normalized percentage recovery of pSMAD1/8 signal was calculated using % recovery = ([Value – Background]/[Max signal – Background]) × 100%, where Background = pSMAD1/8 signal in presence of gremlin-1 + hIgG4 and Max signal = media-only control.

BMP receptor expression profiling on CRC cell lines

The surface BMP receptor expression profile of CRC cell lines was measured by flow cytometry using biotin-labeled anti-BMP receptor antibodies obtained from R&D Systems: BMPR1A (BAF820), BMPR1B (BAF505), BMPR2 (BAF811), ACVR1 (BAF637), ACVR2A (BAF340), ACVR2B (BAF339), and goat IgG isotype control (BAF108). Cells were first incubated for 20 min at room temperature with 1:1000 dilution of Live Dead Aqua (Invitrogen, L34957A) to identify dead cells. Cells were washed in stain buffer (PBS, 2% FBS, 0.09% NaN₃) and incubated with 50 µg/mL of the indicated biotinylated antibody (diluted in stain buffer) for 30 min at room temperature. Cells were washed again with stain buffer before incubating with 1:125 dilution of streptavidin-PE (Invitrogen, 12-4317-87) for 30 min at room temperature followed by further washing before analysis on a BD FACS CANTO II flow cytometer. Data analysis was performed using FlowJo v10.6.0.

Real-time qPCR

Total RNA was obtained using the RNeasy Plus Minikit (Qiagen 74104) and 1 μg of RNA was then reverse transcribed with SuperScript IV VILO Master Mix (ThermoFisher 11756050) according to manufacturer instructions. qPCR was performed using TaqMan Fast Advanced Master Mix (ThermoFisher 4444963) and Taqman expression probes. The relative expression values for each gene of interest were normalized to the geometric mean of the expression levels of GAPDH, B2M, RPLP0, and HPRT1, which were used as housekeeping control genes. The Taqman probe ID list is presented in Table 5.

Table 5. Taqman probe ID list.

Gene	Taqman Probe ID	Gene	Taqman Probe ID
GREM1	Hs00171951_m1	ACVRL1	Hs00953798_m1
GREM2	Hs00254699_m1	ACVR1	Hs00153836_m1
BM-2	Hs00154192_m1	ACVR2A	Hs00155658_m1
BMP-4	Hs00370078_m1	ID1	Hs00357821_g1
BMP-6	Hs01099594_m1	ID2	Hs04187239_m1
BMP-7	Hs00233476_m1	ID3	Hs00171409_m1
BMPR1A	Hs01034913_g1	GAPDH	Hs02758991_g1
BMPR1B	Hs01010965_m1	B2M	Hs00187842_m1
BMPR2	Hs00176148_m1	HPRT1	Hs02800695_m1
ACVR2B	Hs00609603_m1	RPLP0	Hs99999902_m1

ACVR: activin receptor; ACVRL: activin receptor-like; B2M: beta-2-microglobulin; BMP: bone morphogenetic protein; BMPR: BMP receptor; GAPDH: glyceraldehyde-3-phosphate dehydrogenase; HPRT: hypoxanthine-guanine-phosphoribosyltransferase; ID: inhibitor of DNA-binding protein; RPL: ribosomal protein L.

Co-culture of fluorescently labeled CRC cell lines and pancreatic fibroblasts

For all co-culture experiments, CRC cell lines and pancreatic fibroblasts were detached and labeled with CellTracker Green CMFDA (ThermoFisher, C2925) and CellTracker Deep Red (ThermoFisher, C34565), respectively, according to manufacturer instructions. Briefly, CellTracker Green CMFDA and CellTracker Deep Red (5 µM) were incubated with CRC cell lines and pancreatic fibroblasts for 20 min at 37°C. Upon labeling, both CRC cell lines and pancreatic fibroblasts were resuspended in MSC-GRO Vitro Plus III low serum complete medium (Vitro Biopharma, SC00B1) and each cell type was seeded at a density of 2.6 × 10⁴cells/cm² and incubated at 37°C for 72 h in the presence of 150 nM ginisortamab or human IgG4 isotype control.

For gene expression analyses in CRC cell lines and pancreatic fibroblasts, co-cultures were detached using TrypLE Express dissociation solution (ThermoFisher 12604021) and resuspended in PBS containing 0.5% bovine serum albumin (BSA) and 2 mM EDTA and separated by FACS. TO-PRO3 (ThermoFisher, T3605), a nucleic acid stain used for distinguishing dead cells from live cells, was added to each condition prior to sorting. Sorted CRC cell lines and pancreatic fibroblasts were separately lysed for RNA extraction.

Abbreviations

aa	=	amino acid
Ab	=	Antibody
ACVR	=	activin receptor
BMP	=	bone morphogenetic protein
BMPR	=	BMP receptor
BSA	=	bovine serum albumin
CDR	=	complementarity determining region
CMFDA	=	5-chloromethylfluorescein diacetate
CRC	=	colorectal cancer
CSC	=	cancer stem cell
DMEM	=	Dulbecco’s Modified Eagle Medium
EC50	=	half maximal effective concentration
EDTA	=	ethylenediaminetetraacetic acid
Eudra CT	=	European Union Drug Regulating Authorities Clinical Trials Database
ELISA	=	enzyme-linked immunosorbent assay
Fab	=	fragment antigen-binding
FACS	=	fluorescence-activated cell sorting
FBS	=	fetal bovine serum
Fc	=	crystallizable fragment
Fb	=	fibroblast
GDF	=	growth differentiation factor
GEPIA	=	Gene Expression Profiling Interactive Analysis
HCDR	=	heavy chain CDR
HEK	=	human embryonic kidney
HEPES	=	4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid
HRP	=	horseradish peroxidase
IC50	=	half maximal inhibitory concentration
ID	=	inhibitor of DNA-binding protein
IgG	=	immunoglobulin G
IPTG	=	isopropyl β-D-1-thiogalactopyranoside
Ka	=	association rate constant
Kd	=	dissociation rate constant
KD	=	equilibrium dissociation constant
LCDR	=	light chain CDR
mFc	=	multimeric fragment crystallizable regions
mIgG1	=	mouse immunoglobulin G1
MOPS	=	3-(N-morpholino)-propanesulfonic acid
NCT	=	National Clinical Trial
PBS	=	phosphate-buffered saline
PDB	=	Protein Data Bank
PDRC	=	protein related to Dan and Cerberus
PE	=	Phycoerythrin
pSMAD	=	phosphorylated SMAD
qPCR	=	quantitative polymerase chain reaction
R&D	=	research and development
Rmax	=	maximal binding response
RSMD	=	root-mean-square deviation
RT-PCR	=	Real-time polymerase chain reaction
scFv	=	single-chain variable fragment
SPR	=	surface plasmon resonance
TGFβ	=	transforming growth factor beta
TME	=	tumor microenvironment

Data sharing statement

Data from non-interventional studies are outside of UCB’s data sharing policy and are unavailable for sharing.

Acknowledgments

The authors acknowledge Carlos Martinez-Fleites, PhD, for solving the gremlin/Ab structure, Gill Holdsworth, PhD, for providing expertise in BMP biology, Laura McLaughlin, PhD, for characterizing the CRC cell lines. The authors also thank Radhika Bhatia, PhD, of UCB Pharma, for publication and editorial support, and Kheng Bekdache, PhD, of Ashfield MedComms, an Inizio company, for medical writing support that was funded by UCB Pharma in accordance with Good Publication Practice (GPP) guidelines (http://www.ismpp.org/gpp-2022).

Disclosure statement

All authors were employees of UCB Pharma, UK at the time the work was conducted. ND, PJ, LK, DM, MS, AS, and TR: ownership of UCB Pharma stocks and shares.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/19420862.2023.2289681

Additional information

Funding

This work was funded by UCB Pharma, UK.