Functional patient-derived organoid screenings identify MCLA-158 as a therapeutic EGFR × LGR5 bispecific antibody with efficacy in epithelial tumors

Patient-derived organoids (PDOs) recapitulate tumor architecture, contain cancer stem cells and have predictive value supporting personalized medicine. Here we describe a large-scale functional screen of dual-targeting bispecific antibodies (bAbs) on a heterogeneous colorectal cancer PDO biobank and paired healthy colonic mucosa samples. More than 500 therapeutic bAbs generated against Wingless-related integration site (WNT) and receptor tyrosine kinase (RTK) targets were functionally evaluated by high-content imaging to capture the complexity of PDO responses. Our drug discovery strategy resulted in the generation of MCLA-158, a bAb that specifically triggers epidermal growth factor receptor degradation in leucine-rich repeat-containing G-protein-coupled receptor 5-positive (LGR5+) cancer stem cells but shows minimal toxicity toward healthy LGR5+ colon stem cells. MCLA-158 exhibits therapeutic properties such as growth inhibition of KRAS-mutant colorectal cancers, blockade of metastasis initiation and suppression of tumor outgrowth in preclinical models for several epithelial cancer types. Batlle and colleagues develop an organoid platform for functional antibody screening and identify a therapeutic bispecific antibody that binds EGFR and LGR5 and that shows efficacy across epithelial tumor patient-derived xenograft models in vivo.

I t is widely established that only a subset of colorectal cancer (CRC) cells, the so-called cancer stem cells (CSCs), exhibit long-term tumorigenic potential, whereas the tumor bulk is relatively short-lived and poorly tumorigenic. The distinctive properties of CSCs are linked to expression of a genetic program reminiscent of healthy colonic stem cells 1,2 . CSCs are characterized by elevated levels of WNT pathway components, including LGR5, zinc and ring finger 3 (ZNFR3) and ring finger protein 43 (RNF43), all of which are part of the WNT-receptor complex 1,[3][4][5][6][7][8] . While the WNT receptor transduces downstream signals and sustains self-renewal in healthy colon stem cells, it is dispensable in most CRCs due to downstream mutations in the WNT pathway resulting in constitutive activation and signaling [9][10][11] .
Growth and survival of CRC cells depend on mitogenic signals triggered by receptor tyrosine kinases (RTKs) of the epidermal growth factor receptor (EGFR) family 12 . Therapeutic agents targeting RTK signaling such as cetuximab, a monoclonal antibody against EGFR, are used to treat patients with metastatic CRC 13,14 . However, current anti-EGFR therapies have several limitations. They are not effective against CRC with activating mutations in the RAS oncogene family (principally KRAS); and in patients with RAS wild-type, only a subset derive meaningful therapeutic benefit 13,15,16 often with treatment-limiting adverse effects 17,18 .
Here, we developed a therapeutic strategy based on blocking proliferation of CSCs in CRC by leveraging the dual-targeting capabilities of bAbs. We generated a large biobank of PDOs that recreate the cellular heterogeneity and organization of CRCs [19][20][21] , including dependency on self-renewal and proliferative properties of CSCs 3,5,6 . Therapeutic bAb candidates were screened against CRC PDOs of multiple genotypes and phenotypes to assess their functional activity and compared to matched healthy colon mucosa organoids.

NATuRE CANCER
Through this approach we identified MCLA-158, an LGR5 × EGFR bAb with potent and selective growth inhibitory activity in vitro and in vivo against both wild-type and oncogenic KRAS-mutant CRCs.

Results
Generation of bispecific antibodies targeting CSCs. Both healthy colon stem cells and CRC stem cells are characterized by high transcriptional expression of leucine-rich repeat-containing G-protein-coupled receptor (LGR)4, LGR5, RNF43 and ZNRF3 (refs. 1,3,[5][6][7][8] ) yet few antibodies against the ectodomains of these proteins have been reported. Phage libraries generated from immunized humanized transgenic mice (MeMo) and synthetic libraries were screened to generate large panels of bAbs against these CSC surface markers and EGFR using the strategy described in Fig. 1a. The generation of the panel for Erb-B2 receptor tyrosine kinase 3 (ERBB3, commonly known as HER3) is described elsewhere 22 . Close to 400 target clones were selected from immune and synthetic libraries (Fig. 1a). Each WNT clone was expressed in a bispecific IgG format paired with a non-binding arm to measure monovalent target interaction, whereas EGFR clones were produced in a bivalent antibody format. Characterization of these IgG crude productions demonstrated that clones with good affinity, stability and ligand-blocking activity could be isolated from both library formats against all targets (Supplementary Table 1). At the end of the selection and production process, the WNT × RTK panel contained more than 500 bAbs, derived by combining 54 WNT Fab arms (plus a control Fab against the tetanus toxoid (TT)) with 4 EGFR and 4 HER3 Fab arms (Fig. 1a).

Establishment of a CRC patient-derived organoid biobank.
We established a living biobank of CRC PDOs from fresh surgical specimens as previously described 20 . The key clinical-pathological features of CRCs from which we expanded PDOs are detailed in Supplementary Tables 2-4. A total of 99 tumor samples were collected from 68 patients treated at two different hospitals (University Medical Center Utrecht and Meander Medisch Centrum) from which 61 primary CRC and 11 CRC liver metastases PDOs were derived and banked. The success rate of tumor PDO generation was 72.7%, similar to that reported for previous PDO biobanks 20,21,23,24 . For 31 patients, we also established PDOs from tumor-adjacent healthy mucosa (Supplementary Table 5). Figure 1b and Extended Data Fig. 1a summarize the mutations in components of the four main CRC driver pathways (WNT, RTK/ RAS, TP53 and transforming growth factor-β) in each PDO line according to exome sequencing results. The complete catalog of genetic alterations present in PDOs is included in Supplementary Tables 6-8. Chromosome amplifications frequently present in CRC (such as 20q, 13, 8q and 7) and chromosome losses (such as 18, 15, 17p, 14, 8p, 4 and 5) were also highly represented in the PDO biobank (Extended Data Fig. 1b). The most abundant mutational signature in the PDOs was signature 1, which corresponds to deamination of cytosine at CpG dinucleotides leading to T > G mutations 25 (Extended Data Fig. 1c-e). In addition, six PDOs were hypermutated, as shown by a very high frequency of single-base substitutions and small indels (Extended Data Fig. 1c-e), which coincided with an elevated frequency of mutational signature 6 owing to DNA mismatch-repair deficiency 25 (Extended Data Fig. 1c-e).
Overall, these analyses illustrate that the spectrum of genetic alterations captured in the PDO biobank are largely consistent with that reported for the CRC dataset of The Cancer Genome Atlas (TCGA) 9 (Extended Data Fig. 1a).
PDO dependency on RTK mitogenic signals. Epidermal growth factor (EGF) is a non-redundant mitogen for both healthy and tumor stem cells in the colon and it is included in the standard PDO culture medium 20,21 . Removal of EGF slowed down the growth of the majority of PDOs, although the extent of this effect varied greatly among them (Fig. 1c). Heregulin (HRG) is an EGF-like growth factor that binds HER3. Although HRG is not an essential stem cell factor 20,21 , by substituting EGF for HRG we identified a subset of PDOs that can utilize HER3 mitogenic signals for expansion (Fig. 1c).
The above experiments revealed that a substantial fraction of PDOs bearing activating mutations in EGFR downstream pathway components, including many with canonical KRAS-activating mutations, exhibited lower growth rates in the absence of EGF and HRG (Fig. 1c). We corroborated this observation by culturing a subset of KRAS wild-type and mutant PDOs over a wide range of EGF concentrations (Fig. 1d,e and Extended Data Fig. 2a,b). As an example, C55T carried a KRAS-G12V mutation, yet, similar to some KRAS wild-type PDOs such as C20T ( Fig. 1e and Extended Data Fig. 2b), it became growth-arrested at low EGF concentrations ( Fig. 1d and Extended Data Fig. 2a). C37T carried a KRAS-G12D mutation and exhibited a milder response to low EGF concentrations than C55T ( Fig. 1d and Extended Data Fig. 2a) but comparable to that of some KRAS wild-type PDOs such as C39T ( Fig. 1e and Extended Data Fig. 2b). These findings confirm that mutations in EGFR pathway components do not confer complete EGFR signaling independence to CRCs, as previously observed 21 . Our results also underscore the power of organoids to predict EGFR responses.

High-content image-based screening of RTK responses in PDOs.
CRC PDOs are formed by heterogenous cell types that encompass CSCs and their progeny and adopt complex three-dimensional (3D) organizations. To fully capture the outcome of drug-target interactions in this complex environment, we utilized a high-content image-based screening method. Figure 2a illustrates the profound and differential modifications in PDO morphology caused by EGF and HRG using two representative and previously characterized PDOs 20 . In response to EGF, P18T grew into large extending structures with a collapsed central branching lumen but developed spherical and swollen lumens in response to HRG (Fig. 2a). In contrast, P14T PDOs displayed a single round lumen when cultured with HRG but formed multiple lumens in response to EGF (Fig. 2a). The number of lumens, lumen area and organoid area discriminated well between P14T and P18T PDOs cultured without RTK-stimulating factors or cultured in either EGF or HRG-supplemented medium (Fig. 2a). Similarly, EGFR-and HER3-blocking antibodies caused specific alterations in PDO morphology in a dose-dependent manner that could be distinguished using morphological parameters (Fig. 2b). Based on these observations, we calculated a multiparametric score from multiple morphological measurements of PDOs that was more robust than any single feature in discriminating RTK-induced responses. The organoid models were ordered by their APC and TP53 mutation status. c, Organoid size was compared between PDOs cultured with either EGF (5 ng ml −1 ) or HRG (5 ng ml −1 ) and the absence of growth factors over a culture period of 7 days. '%' indicates change relative to absence of EGF; '-' indicates that no data were collected. Each data point represents the mean of n = 4 independent cultures. d, Dose-response curves to EGF in KRAS mutant PDOs. Organoid size was measured at day 7 and referred to as % of maximum. Each data point represents mean ± s.d. of n = 4 independent cultures. e, Dose-response curves to EGF in KRAS wild-type (wt) PDOs. Each data point represents mean ± s.d. of n = 4 independent cultures.   C110T  C55T  C82T  C36T  C8T  C22T  C39T  C31T  C43T  C87T  C15T  C65M  C28T  C54T  C26T  C20T  C14T  C44T  C115T.III  C113T  C78M  C37T  C92T  C7T  C9T  C45T  C40T  C57T  C25T  C31M  C27T  C47T  C66T  C75T  C18T  C23M  C29T  C96T  C2T  C6T  C42T  C60T HYPERMUTATED HRG RESPONSIVE -1 -1 1 1 -1 1 --1 1 1 1 1 ----1 -1 1 --

Articles
NATuRE CANCER Bispecific antibody screen. We chose P14T and P18T as models to perform the primary bAb screen. These two organoids are KRAS wild-type, depend on RTK signaling (Fig. 2a,b) and are models for CRCs that carry mutations that constitutively activate the WNT pathway 20 . It is however well established that 10-15% of all CRCs depend on WNT receptor-mediated signaling 20,[26][27][28] .
To screen for bAbs in this CRC subset, we also included P19Tb, which only expands in medium supplemented with WNT3a 20 . The panel of >500 WNT × RTK bAbs was screened in duplicate at high (10 µg ml −1 ) and low (2 µg ml −1 ) antibody concentrations in the presence of EGF, HRG or WNT3a. We calculated a multiparametric score for each PDO and antibody treatment as a surrogate of potential therapeutic responses.
Results from the primary screen are shown in Fig. 2c. The Zʹ factor was 0.33 (s.d. = 0.22, n = 8). Bispecific antibodies combining the EGFR-targeting arm Fab232 with a subset of LGR4-, RNF43-, ZNFR3-and particularly LGR5-targeting arms modified P14T and P18T growth patterns cultured in EGF-dependent growth conditions as shown by changes in the multiparametric score (left red box in Fig. 2c). Likewise, the HER3-targeting arm Fab264 combined with LGR4-, RNF43-, ZNFR3-or LGR5-targeting arms triggered responses in these two PDOs in HRG-supplemented medium (right red box in Fig. 2c). However, none of the antibodies modified substantially the growth features of P19Tb in WNT3a-supplemented medium ( Fig. 2c) implying lack of WNT signaling inhibitory activity in the bAb panel.
We selected the 28 bAbs that most robustly modified CRC organoid growth patterns in the primary screen; 14 contained the EGFR Fab232 antibody arm coupled to distinct arms against WNT pathway components and the other 14 were based on the Fab264 HER3-targeting arm combined with WNT-targeting arms. The activity of these antibodies was subsequently characterized on an extended panel of 22 CRC PDOs and one healthy mucosa organoid model (C51N) (Fig. 2d,e). This secondary screen not only confirmed the activity of several EGFR-or HER3-based bAbs in multiple PDO lines, but also revealed two unexpected activities in a subset of bAbs. First, the Fab232 EGFR arm coupled to a control TT arm exerted almost no effect on PDO growth, but its activity was greatly potentiated in combination with several LGR5 arms (Fig. 2d). This synergism was, however, not observed in the HER3-based antibodies, which exhibited equivalent growth inhibitory activity in combination with the TT-, LGR5-or other WNT-targeting arms (Fig. 2e). Second, several bispecific antibodies inhibited the growth of PDOs bearing activating KRAS mutations (Fig. 2d,e).
The bAb consisting of the Fab232 EGFR arm and the Fab072 LGR5 arm showed growth inhibitory activity in the largest fraction of PDOs (Fig. 2d, arrow). This bAb reduced the growth rate of 52% of the CRC organoid models, including many KRAS-mutant PDOs. We named this bAb MCLA-158.
Molecular characterization of MCLA-158. MCLA-158 is a native bispecific IgG binding both EGFR and LGR5 (Fig. 3a). It contains two CH3-engineered heavy chains that enforce heterodimerization through the inclusion of charged residues in the CH3 interface (dubbed DEKK 29 ) and a κ light chain in germline configuration (Fig.  3a). MCLA-158 produced by transient transfection and purified by protein A capture and gel filtration resulted in an essentially pure bAb, as demonstrated using mass spectrometry and analytic cation exchange. The main product-related contaminants were trace amounts of half bodies resulting from un-dimerized heavy chains (Fig. 3b), homodimers of heavy chains containing the DE mutation pair (Fig. 3b) and charged variants of the bAb production in line with that observed in parental monoclonal antibodies (Fig. 3c).
In Scatchard assays, the affinity of MCLA-158 for its targets was in the subnanomolar range: 0.22 nM for the anti-EGFR arm and 0.86 nM for the anti-LGR5 arm (Supplementary Table 9). To characterize the binding epitopes recognized by the bAb, shotgun mutagenesis analysis was applied 30 . Mutant libraries of recombinant EGFR and LGR5 expressed on cells were constructed and MCLA-158 binding was analyzed by flow cytometry analysis. Alanine substitutions in residues I462, G465, K489, I491, N493 and C499 were shown to reduce binding activity of MCLA-158 to EGFR (Extended Data Fig. 2c). These residues map to domain III of EGFR (Fig. 3d, blue surface) and overlap with the surface bound by EGF (Fig. 3d, in yellow) implying that binding of MCLA-158 in this region inhibits EGF interaction with its receptor 31 . Indeed, this was confirmed in an EGF-driven cell death assay 32 using the EGFR binding Fab (Fab232) of MCLA-158 reformatted as a bivalent IgG (Fig. 3e). The degree of inhibition in this assay was similar to that of cetuximab (Fig. 3e).
Mutagenesis analysis of MCLA-158 binding to LGR5 identified D43, G44, M46, F67, R90 and F91 as important contact residues (Extended Data Fig. 2d). These residues (blue surface in Fig. 3f) map to the N-CAP (orange) and first leucine-rich repeat domain (dark teal), in a region proximate to but not overlapping the R-spondin (RSPO) binding region (yellow surface) 8 , which is consistent with the anti-LGR5 Fab domain of MCLA-158 having no effect on RSPO binding. Indeed, binding of the LGR5 arm of MCLA-158 reformatted as a bivalent IgG (Fab072) to LGR5-overexpressing CHO cells was only weakly inhibited by high RSPO concentrations (Fig. 3g). This is in contrast to the activity of two RSPO-blocking antibodies OMP88R20 and bivalent Fab049 IgG (Fig. 3g); the latter was also generated in this study (Supplementary Table 1).

The LGR5 arm of MCLA-158 recognizes stem-like tumor cells.
We next studied the capacity of the LGR5 arm included in MCLA-158 (Fab072) to recognize endogenous LGR5 levels at the cell surface. In dissociated P18T organoids, the monospecific LGR5 arm included in MCLA-158 (Fab072 IgG) identified a subset of cells that expressed elevated LGR5 messenger RNA levels, shown by a, Characterization of the morphological change of P18T and P14T in response to 5 ng ml −1 EGF or 5 ng ml −1 HRG. Pictures illustrate morphology adopted in different conditions and the graph depicts measurements of lumen counts, lumen area and organoid area. Each data point represents a well mean of ~100 organoids. Scale bars, 200 μm. GF, growth factor. b, Effects of either EGFR-or HER3-blocking antibodies on P14T cultured with EGF, HRG or no growth factor. Graph depicts measurement of lumen complexity versus width on different treatments. The size of each data point indicates antibody doses (from smallest to largest, 1, 2.5, 10 and 25 μg ml −1 ). Each data point represents the mean measurement of all PDOs growing in one culture well. Note that antibody-mediated growth inhibition in EGF-stimulated growth conditions has different morphological effects than in HRG-stimulated growth conditions, indicating the requirement of a multiparametric score. Scale bars, 200 μm. Ab, antibody; ctrl, control; neg, negative. c, Primary bAb panel screen. Changes in multiparametric scores triggered by different bAbs on P14T, P18T and P19Tb PDOs cultured with EGF, HRG or WNT3a. '%' indicates change relative to absence of growth factor. Red boxes indicate the antibodies considered for secondary screening. L4, LGR4 Fab arm; L5, LGR5; RN, RNF43; ZN, ZNFR3; TT, tetanus toxoid (control Fab). Each data point represents a mean of n = 2 independent culture wells. d, Secondary screen of bAbs containing EGFR Fab arms on PDOs supplemented with EGF. Red intensity indicates % of growth inhibition in each PDO calculated as multiparametric score. Percentage of PDOs that showed responses to each antibody (bottom). KRAS mutation status (right). Arrow indicates MCLA-158. Each data point represents a mean of n = 2 independent wells. e, Secondary screen of bAbs containing HER3 Fab arms on PDOs supplemented with HRG. Each data point represents a mean of n = 2 independent wells. LGR5       P = 1 × 10 -9 P = 2 × 10 -10 P = 3 × 10 -9 P = 5 × 10 -10 P = 1 × 10 -8 P = 3 × 10 -9 P = 2 × 10 -9 P = 1 × 10 -8 P = 7 × 10 -5 P = 0.0003   Fig. 3b). MCLA-158 also elevated the apoptosis index as measured by the number of cells with a condensed nucleus (Extended Data Fig. 3c). Of note, these effects were not observed when the EGFR arm of MCLA-158 (Fab232) was coupled to the TT control arm (Fig. 4b,c and Extended Data Fig. 3a-c). Consistent with these findings, high-concentration MCLA-158-treated CRC organoids displayed delayed kinetics of recovery after the withdrawal of antibody treatment compared to cetuximab (Extended Data Fig. 3d).
Organoid initiating capacity has been extensively used as functional readout of healthy and cancer stem cell activity 1,34-36 . At high doses (10 μg ml −1 ), both antibodies decreased organoid size (Fig.  4d), yet MCLA-158 was particularly efficient at preventing organoid initiation compared to cetuximab (Fig. 4e). To further assess the specificity of MCLA-158 toward LGR5 + CSCs, we knocked down LGR5 levels in P18T PDOs using a short hairpin RNA (shRNA) vector ( Fig. 4f-i and Extended Data Fig. 3e). Downregulation of LGR5 did not cause major changes in expression of stem cell/WNT target genes ( Fig. 4f) nor did it modify PDO growth kinetics (Fig. 4g), implying that, as in the case of healthy crypt stem cells 8 , LGR5 function is dispensable for CSC expansion. Notably, LGR5 knockdown conferred a large degree of insensitivity to MCLA-158 ( Fig. 4h,i), further confirming specific targeting of the LGR5 + cell population.
We next compared in vivo antitumor activity of MCLA-158 versus cetuximab on subcutaneous xenografts generated by inoculation of P18T organoids ( Fig. 4j and Extended Data Fig. 4a,b). MCLA-158 caused a significant reduction in the mean tumor volume compared to both PBS (vehicle) and cetuximab treatment from day 2 onward (Fig. 4j and Extended Data Fig. 4a,b). Of note, cetuximab-and PBS-treated mice showed comparable growth kinetics over the observation period as measured by tumor volume and survival (Extended Data Fig. 4a,b). Histological inspection of MCLA-158-treated P18T xenografts revealed reduced cellularity and a prominent decrease in the number of Ki67 + cells (Fig. 4k). MCLA-158 also showed superior growth inhibitory capacity relative to cetuximab in subcutaneous xenografts generated from inoculation of C31M, a PDO bearing a KRAS-G12D mutation (Extended Data Fig. 4c).
Activity of MCLA-158 against LGR5 + PDOs and patient-derived xenografts. We next compared the activity of the two antibodies across a range of PDO models. MCLA-158 reduced growth rates by at least 50% in 11 out of 21 organoids tested, including two metastasis-derived PDOs (C0M and C31M) and outperformed the growth inhibitory capacity of cetuximab in the majority of these CRC models (Fig. 5a). The different therapeutic potency of the two antibodies was particularly evident in a subset of PDOs bearing KRAS-activating mutations (C0M, C55T, C27T, C31M and C25T; Fig. 5a). Further supporting these observations, we measured large differences (8 to 125-fold) in IC 50 between cetuximab and MCLA-158 in both KRAS wild-type and mutant PDOs (Fig. 5b) and this differential effect was even more pronounced in culture conditions with elevated EGF concentrations (Supplementary Table 10). PDOs that responded to MCLA-158 contained higher percentages of LGR5 + cells than non-responder PDOs (Fig. 5c and  Supplementary Table 11). Moreover, stratification of PDOs according to high versus low % of LGR5 + cells (defined as above or below average) predicted growth inhibition by MCLA-158 ( Fig. 5d and Supplementary Table 11).
To extend our observations beyond the CRC PDO biobank, we selected 24 patient-derived xenograft (PDX) models that coexpress elevated LGR5 and EGFR mRNA levels according to RNA-sequencing data. MCLA-158 demonstrated significant antitumor activity in the majority of PDX models; CRC (three out of five; individual growth kinetics in Extended Data Fig. 5), esophageal squamous cell carcinoma (four out of six), gastric adenocarcinoma (six out of eight) and squamous head and neck cancers (two out of five), both in a KRAS-wild-type and KRAS-mutant setting (Fig. 5e).

Limited responses of healthy colonic stem cells to MCLA-158.
EGF is a mitogen for healthy colonic stem cells 37 and treatment with EGFR pathway inhibitors causes gastrointestinal toxicity in a subset of patients 38 . The detrimental effect of EGFR inhibitory antibodies was evident in two out of five healthy mucosa PDOs included in our biobank (C71N and C57N; Fig. 5b). However, the response triggered by MCLA-158 in these two healthy mucosa-derived PDOs was substantially weaker than that of cetuximab (Fig. 5b). An advantage of the PDO-based screening system described herein is the possibility of testing antibody responses on pairs of healthy and tumor organoids derived from the same patient. Figure 5f shows a dose-response curve of C55T PDO, which was generated from a KRAS-G12V-mutant CRC, versus C55N, a healthy PDO expanded from adjacent healthy mucosa. At 1 µg ml −1 , MCLA-158 exerted a robust (maximal) tumor growth inhibitory effect without affecting the expansion of healthy mucosa-derived PDO culture. In contrast, cetuximab reduced tumor growth only at concentrations approximately 100-fold higher than those required for MCLA-158. At this concentration, growth inhibition of C55N was equivalent to that of C55T (Fig. 5f). Therefore, MCLA-158 beneficially distinguished between tumor and healthy tissue and exerted a potent antitumor response, whereas cetuximab did neither in these models. Notably, C55N PDO exhibited substantially reduced LGR5 cell surface expression and contained fewer LGR5 + cells compared to its tumoral counterpart C55T (Fig. 5g-i). This finding, together with our observations that downregulation of LGR5 levels inhibits MCLA-158 activity (Fig. 4h,i), supports the notion that specific  Table 11). d, Percentage of growth inhibition in PDOs with % LGR5 + cells above (high) or below (low) average (Supplementary Table 11). e, Volume of individual CRC, esophageal, gastric and head and neck PDX xenografts treated either with vehicle (control) or MCLA-158 (25 mg kg −1 per week for 6 weeks). Mutations in EGFR pathway components are indicated. f, Dose-response curves to cetuximab and MCLA-158 of PDOs C55T and C55N in 2.5 ng ml −1 EGF. Each data point represents a mean ± s.e.m. of n = 6 independent cultures. g, LGR5 surface levels in the indicated PDOs measured using a LGR5 (Fab072) × TT antibody. Staining of TT × TT antibody on C55T was used as a negative control. Frequency plot was normalized to the mode and shows a representative experiment. Bar indicates the gate used to quantify LGR5 + cells.   89  81  80  77  72  64  61  56  54  50  49  47  47  38  33  31  10  <1  <1 11  <1   57  65  41  39  <1  11  15  47  32  36  52  4  28  70  43  47  18  14  <1  <1  25   WT G12D WT G12V G12S G12D  G12V  WT  WT  WT  WT  WT  WT  WT  WT  WT  WT G12R WT G12D   C5T  C2T  C20T  C22T  C17T  C27T  C6T  C7T  C14T  C26T  C42T  C28T C25T  C37T  C0M  C31M  C55T  P18T P8T P19Tb P14T Articles NATuRE CANCER targeting of CRC by MCLA-158 is due to higher LGR5 expression in tumor compared to healthy colonic stem cells.

MCLA-158 inhibits metastasis formation.
In experimental models of advanced CRC, LGR5 + tumor cells are required for metastasis formation 3 . This finding prompted us to test the effects of MCLA-158 on orthotopic xenografts (PDOXs) (Fig. 5j,k and Extended Data Fig. 4d-f). The three different PDOX models used for these experiments were selected because they carried KRAS-activating mutations (Supplementary Table 12), generated metastases and expressed detectable LGR5 mRNA levels. Mice bearing model LM-CRCX3 were treated with either cetuximab or MCLA-158, whereas models M001 and M005 were treated only with MCLA-158. As shown in Fig. 5j, MCLA-158 exhibited superior therapeutic capacity to cetuximab in terms of reducing the size of the primary CRC generated by LM-CRCX3. MCLA-158 also completely prevented development of local and distant metastases, whereas cetuximab exerted no effect on the disseminated disease ( Fig. 5k and Extended Data Fig. 4d).
In the M001 and M005 models, not only did MCLA-158 treatment cause a significant reduction of primary CRC growth (Fig. 5j), it robustly blocked the formation of metastasis ( Fig. 5k and Extended Data Fig. 4e,f). In model M005, two mice developed peritoneal metastases despite MCLA-158 treatment ( Fig. 5k and Extended Data Fig. 4e). However, there were no major differences in LGR5 mRNA levels between MCLA-158-resistant metastases and those arising in vehicle-treated mice (Extended Data Fig. 4g). Therefore, in a small fraction of cases, LGR5 + metastatic cells can apparently bypass MCLA-158 inhibitory effects.
Transcriptional response to MCLA-158 treatment. MCLA-158 binds LGR5 but does not block RSPO binding (Fig. 3g). To exclude the possibility of non-conventional modulation of WNT signaling by MCLA-158, we performed RNA sequencing of P18T and C55T exposed to MCLA-158 or cetuximab (Fig. 6a-d). This experiment revealed that the expression of the WNT target gene, LGR5 + intestinal stem cell (ISC) and Paneth cell signatures was upregulated rather than inhibited by MCLA-158 (Fig. 6a,b). Cetuximab also induced upregulation of WNT, ISC and Paneth cell signatures implying that these transcriptional effects are direct consequence of EGFR inhibition (Fig. 6a,b). These findings are consistent with a recent study demonstrating that cetuximab treatment enforces an ISC and Paneth cell-like phenotype in CRC 39 . MYC proto-oncogene, BHLH transcription factor (MYC) is a direct β-catenin/TCF target gene 40 and plays a causal role in the expansion of CRC stem cells due to WNT activating mutations 41,42 . Despite increased WNT/ISC gene levels upon MCLA-158 and cetuximab treatment, the expression of MYC was downregulated (Fig. 6c,d). MCLA-158 was particularly efficient at suppressing MYC expression compared to cetuximab (Fig. 6c) and this differential effect was even more evident in the C55T KRAS-mutant PDO (Fig. 6d). We further investigated global differences in response through gene set enrichment analysis (GSEA; Fig. 6e-h). The biological responses of P18T to high-dose MCLA-158 and cetuximab treatment (10 μg ml −1 ) were largely overlapping (Fig. 6e). In contrast, changes in gene expression provoked by low-dose MCLA-158 (1 μg ml −1 ) were of much higher magnitude and significance than those of cetuximab, involving downregulation of MYC target genes, suppression of the mTOR biosynthesis pathway and the mitogenic program (Fig. 6f). This differential response was amplified in the C55T KRAS-mutant PDO line (Fig. 6g,h), which remained largely unresponsive to low-dose cetuximab (Fig. 6h). From these results, we conclude that both antibodies elicit qualitatively similar transcriptional changes, yet MCLA-158 exerts a much more potent response than cetuximab.

MCLA-158 induces EGFR internalization and degradation.
To address the mechanistic basis for the enhanced therapeutic properties of MCLA-158 compared to cetuximab, we investigated whether its activity was correlated with binding to EGFR, binding to LGR5 or the combination of both over a range of antibody concentrations (Fig. 7a). For these experiments we used as models PDOs derived from primary CRCs (P18T and C55T), metastasis (C1M) and healthy mucosa (C51N and C55N) (Fig. 7a). In all cases, the LGR5 arm (Fab072) combined with the EGFR arm (Fab232) in the same antibody exerted profound synergetic growth inhibitory effects on CRC PDOs compared to the individual arms combined with the TT control arm (Fig. 7a). Therefore, physical linkage of the EGFR and the LGR5 arms in full-length bispecific IgG format is strongly associated with synergistic inhibition of CRC organoid growth.
We next studied the cellular distribution of EGFR in PDOs upon antibody treatment (Fig. 7b). Addition of cetuximab did not modify EGFR distribution, which remained basolateral for the duration of the treatment and colocalized with cetuximab in the basolateral membrane (Fig. 7b, white arrows). In contrast, 60 min after the addition of MCLA-158, EGFR displayed a punctuated cytoplasmic distribution, which overlapped with the localization of MCLA-158 (Fig. 7b, white arrowheads), implying that both receptor and antibody were rapidly internalized. Treatment of PDOs with LGR5 × TT and EGFR × TT bAbs demonstrated that the cytoplasmic localization of MCLA-158 results from binding to LGR5, a constitutively internalizing cell surface protein 43 and is independent of EGFR binding (Extended Data Fig. 6a). Of note, MCLA-158 was not internalized by healthy colon organoids and remained basolateral for the duration of treatment (Extended Data Fig. 6b). Similarly, there was no EGFR internalization in C47T (Extended Data Fig. 6b), a PDO model that contains no LGR5 + cells (Supplementary Table 11). We also noticed that EGFR staining intensity was strongly reduced after 24 h, whereas MCLA-158 remained in the cytoplasm (Fig. 7b, open arrowheads). Western blot analysis of PDO P18T lysates confirmed a time-dependent decrease in total EGFR levels under MCLA-158 treatment, beginning at 6 h and continuing to decrease at 72 h (Fig. 7c). After washing out MCLA-158, EGFR levels continued to be downregulated during at least another 72 h (Fig. 7d). In contrast, EGFR levels upon cetuximab treatment remained unchanged throughout the treatment (Fig. 7c).

Discussion
CRC-derived PDOs predict drug and radiation responses in patients with CRC [44][45][46][47] . Here we reveal the potential of PDO biobanks for drug discovery. Our approach has several advantages over traditional pharmaceutical strategies. First, PDO biobanks permit unbiased functional testing of large panels of drug candidates across a relevant cross-section of patient genotypes and phenotypes at the first stage of the discovery pipeline. This approach avoids making mechanistic assumptions upfront and bypasses the common reliance on simplistic model systems such as cell lines, which may at best only partially reflect the pathophysiology of a disease state. We generated PDOs from 72% of patient samples, a success rate that compares favorably to other patient-derived tumor model systems such as PDXs. Of note, recently reported optimized PDO growth conditions enable expansion of near 100% of samples 21 . A second advantage of our approach is that PDOs are better surrogates of tumor biology than cell lines as they recapitulate the heterogeneity, organization and vulnerabilities of the tumor of origin 20,21,48 . These advantages are illustrated by the discovery that a large fraction of CRCs carrying KRAS-activating mutations still display EGF dependency 21 . Cell lines selected to expand in standard culture conditions had failed to reveal this feature. Third, the use of healthy-tumor PDO pairs from the same patients facilitates the early selection of drug candidates with a large therapeutic index. However, we recognize that all model systems have limitations reproducing complex pathophysiology and can give rise to conflicting observations. For instance, subcutaneous PDX models with KRAS mutations were less sensitive to MCLA-158 than KRAS-mutant PDO models. This could reflect murine adaptations, distortions created by PDO culture conditions or tumor heterogeneity, highlighting the challenges of therapeutic drug discovery. Nonetheless, the unique properties of MCLA-158 would have not been elucidated via reliance on traditional pharmaceutical strategies. Cetuximab and panitumumab are therapeutic monoclonal antibodies that bind EGFR and prevent ligand-dependent downstream signaling; they are approved for treatment of RAS wild-type metastatic CRC 13,14 . However only ~30% of patients respond to these drugs, whereas a far greater proportion of RAS wild-type metastatic tumors are sensitive to EGFR inhibition in preclinical settings [49][50][51][52] . Consistent with our results, this dependency can even be observed in tumors with acquired RAS mutations 51 . Investigation of EGFR monoclonal antibody cocktails with preclinical potency greater than cetuximab have shown clinical responses in patients who progressed on EGFR monoclonal antibody treatment 53 and this activity is mechanistically associated with EGFR receptor internalization and degradation [53][54][55] . However, their use substantially increases on-target off-tumor toxicities compared to licensed EGFR monoclonal antibodies, limiting their therapeutic application 53,56 .
By applying our PDO discovery approach we report the identification of MCLA-158, a bAb that combines the potency of monoclonal antibody cocktails via EGFR degradation with selectivity for tumors driven by WNT dysregulation. Our data suggest that the differential effect of MCLA-158 on tumor compared to healthy PDOs is facilitated by the elevated expression of LGR5 in CRC, a consequence of constitutive WNT pathway activation. Upregulation of the WNT target gene LGR5 is detected in the majority of CRCs (its expression is particularly elevated in CSCs 1,3-8,57 ) and it correlates with lymph node metastases 58,59 . Indeed, LGR5 + cells are required for metastatic outgrowth in preclinical CRC models 3 . Our data indicate that MCLA-158 effectively targets the highly mitotic LGR5 + CSC population that supports organoid growth and that initiates primary and metastatic tumors. Mechanistically, we provide evidence that the increased potency of MCLA-158 compared to cetuximab is due to LGR5-dependent internalization of EGFR, which ultimately leads to its degradation. MCLA-158 shows in vivo antitumor activity in preclinical models of other cancer types characterized by LGR5 expression such as esophageal squamous cell carcinoma, gastric carcinoma and head and neck squamous cell carcinoma (HNSCC). Based on its differentiated therapeutic profile an ADCC-enhanced development candidate of MCLA-158 (petosemtamab) is currently undergoing clinical evaluation in different populations of patients with solid tumors (NCT03526835).

Methods
This study complies with all relevant ethical regulations. The Biobank Research Ethics Committee of the University Medical Center (UMC) Utrecht (TCBio) approved the biobanking of PDOs. All donors participating in this study signed informed consent. Experiments with mice were approved by animal experimentation committees of each of participating institution as detailed for each case below. Testing EGF-blocking capacity of EGFR arms. A biotin-EGF-based competition assay was performed using serum. Goat anti-human IgG-Fc (Bethyl Laboratories, cat. no. A80-104A; 5 µg ml −1 ) was coated overnight. Wells of the enzyme-linked immunosorbent assay (ELISA) plates were washed 3× (PBS and 0.02% v/v Tween 20) and blocked (PBS (pH 7.2) and 5% BSA) for 1 h at room temperature. rhEGFR-Fc at 5 µg ml −1 (RND Systems, cat. no. 344-ER) was captured for 1 h at room temperature and plates washed again 3×. Then, 1.5 nM biotinylated EGF (Invitrogen, cat. no. 3477) in blocking buffer with or without different dilutions of serum was added to the plates and incubated for 1 h at room temperature, after which plates were washed again 3×. HRP-streptavidin (BD, cat. no. 554066) was added to the plates for 1 h at room temperature and plates were washed again and visualized by TMB/H 2 SO 4 staining. Quantification was performed by means of OD450 nm measurement.
Antibody library construction and panning. From successfully immunized mice, the lymph nodes were used for the construction of 'immune' phage antibody repertoires. RNA was extracted from the lymphoid tissue using Trizol Reagent LS (Invitrogen) and 1 µg of total RNA was used in a room temperature reaction using an IgG-CH1 specific primer. The resulting complementary DNA was then used to amplify the polyclonal pool of VH-encoding cDNA using in-house developed VH-specific primers. PCR products were purified and ligated into Fab phagemid MV1043 that already contained the IGKV1-39 common light chain (cLC). Libraries were rescued using helper phage VCS-M13. All phage libraries contained >10 6 transformants and had an insert frequency of >80%. Materials from individual mice were kept separate throughout this procedure.
Phage libraries were rescued according to standard procedures. Briefly, Fc-or HIS-tagged fusion protein and cells expressing the target antigens were used for panning in a single selection round. Approximately 1,000 individual colonies resulting from the selections on each antigen were picked, rescued and screened for binding to antigen-positive and antigen-negative cells in FACS. Clones that bound specifically to the target antigen of both species were identified and sequenced to establish a VH gene sequence. The VH sequence was aligned to the VH of all other clones that met these criteria. Clones were grouped into 'superclusters' (defined as a group of clones sharing the same VH V-gene usage and having at least 70% sequence identity in HCDR3 and the same HCDR3 length) and into 'clusters' that use the same VH gene and 100% identical HCDR3.
Generation of Fabs and bispecific antibodies. VH regions of cLC Fabs resulting from these selections were recloned into a wild-type IgG1 vector or into vectors for expression of bAbs: the KK vector (T366K and L351K) and DE vector (L351D and L368E). Wild-type IgG1 constructs were expressed individually; bispecific IgGs were prepared by mixing equal amounts of DNA of KK and DE vectors. Expression was conducted in Freestyle 293-F cells (Invitrogen, cat. no. P/N 51-0029) after transfection using polyethylenimine (PEI; Polysciences) according to the manufacturer's recommendations. IgGs were purified over protein A, buffer exchanged to PBS and quantified using Octet (Fortébio). Larger batches of protein were purified over protein A and gel filtration.
IgGs (both mono-and bispecific) were analyzed for binding in flow cytometry and ELISA. Briefly, for flow cytometry, cells were incubated with 5 µg ml −1 of IgG for 1 h on ice. Cells were washed twice with ice-cold PBS and 1% BSA and IgG Articles NATuRE CANCER detected by incubation with a fluorescent-labeled anti-human IgG antibody for 30 min on ice. Cells were washed again, resuspended in PBS and 1% BSA and measured in a FACSCanto cytometer (Becton Dickinson).
For ELISA, antigens were coated overnight to MAXISORP ELISA plates. Wells of the ELISA plates were blocked (PBS (pH 7.2) and 5% BSA) for 1 h and incubated with selected antibodies (5 µg ml −1 diluted in PBS and 5% BSA) for 1 h at 25 °C. The ELISA plates were washed three times (PBS and 0.05% v/v Tween 20) and bound IgG was detected with 1:2,000 diluted HRP-conjugate (goat anti-human IgG, Becton Dickinson). The plates were washed again and visualized by TMB/H 2 SO 4 staining. Quantification was performed by means of OD450 nm measurement.
EGFR domain mutations and antibody competition assays. Phage minipreps of Fab clones belonging to the different superclusters were tested for binding to CHO cell clones stably expressing either wild-type EGFR or EGFR-variant III in FACS, as described above.
Anti-EGFR Fabs (as Fabs expressed on phage) were tested for binding to EGFR in the presence of an excess of control, literature-derived IgG, namely ICR10 (Abcam, cat. no. ab231; domain I), EGFR.1 (Thermo Fisher Scientific cat. no. MS-311-P; domain II), matuzumab (made in house; domain III) or cetuximab (Merck clinical batch; domain III) as described by Cochran et al. 60 .
Serum stability binding assays. To get an indication of the stability of the bispecific IgGs, all IgGs were incubated at 40 °C for a week in serum-containing medium (DMEM high glucose (Gibco, cat. no. 41966-029) supplemented with 9% fetal bovine serum (Thermo Fisher Scientific, cat. no. SV30180)) and were then tested for binding in a binding ELISA (essentially as described above). This ELISA consisted of coating 2 µg ml −1 antigen (rhLGR4-Fc, rhLGR5-Fc, rhZNRF3-Fc or rhRNF43-Fc), using 100 µl of 5 µg ml −1 IgG and blocking in 5% BSA in PBS. In this assay, the binding to antigen of the same IgG (in serum-containing medium) kept at 2-8 °C (refrigerator) was compared to that of IgG kept at 40 °C. The percentage loss of binding after 1 week incubation at 40 °C was calculated.
Native mass spectrometry. Native mass spectrometry was performed according to the procedure of de Nardis et al. 29 .
Cation exchange chromatography. Cation exchange high-performance liquid chromatography (HPLC) was performed to assess the charge heterogeneity and retention time of the IgGs. The experiments were performed at ambient temperature on a Dionex HPLC system equipped with an SP STAT 7 µm column and a UV-vis detector, with 10 µg sample injected in each run. A gradient of 25 mM phosphate buffer (pH 6.0) with NaCl concentrations increasing from 0 to 1 M was applied to separate the antibodies. The data were analyzed using Chromeleon software.
Isoelectric focusing. FocusGels with pI range 6-11 (Web Scientific, cat. no. 1006-03) were run on a GE Healthcare Multiphor II electrophoresis unit at 10 °C. Then, 10 µg of untreated sample was loaded to the sample next to a high pI range marker (GE Healthcare, cat. no. 17047301 V). The electrophoresis program consisted of three phases: initial focusing for 10 min at 500 V followed by 90 min at 1,500 V and finally a focusing phase at 2,000 V for 10 min. Subsequently, the gel was fixed and stained using colloidal Coomassie dye (Pierce, cat. no. 24590).

Shotgun mutagenesis. The epitopes on EGFR and
LGR5 recognized by MCLA-158 were determined by shotgun mutagenesis analysis as previously described 30 . Two mutation libraries were made from the two antigens: one library encompassed amino acids 300-520 (ligand-binding domain L2 or domain III) of human EGFR (GenBank reference sequence NP_005219.2) and the other encompassed amino acids 22-560 (N-terminal domain until the first transmembrane helix) of human LGR5 (GenBank reference sequence AAH96324.1). The LGR5 expression construct was truncated at amino acid 834 to increase cell surface expression of the receptor. In-house-developed antibodies targeting a different epitope were used as control antibodies for the expression of the mutants. An amino acid residue was considered as a critical residue if the binding activity or reactivity of MCLA-158 was reduced by more than 75%, as compared to the unmodified amino acid sequence.
Affinity determinations. MCLA-158 was radiolabeled with 125 I using IODO-GEN according to the protocol described by van Uhm et al. 61 . The immunoreactivity of the antibody after radiolabeling was investigated with the method described by Lindmo et al. 62 . Steady-state cell affinity measurements of 125 I-MCLA-158 were performed with CHO cells (DSMZ, cat. no. ACC 110) expressing either EGFR or LRG5 to investigate the affinity toward EGFR and LGR5, respectively. In addition, measurements were performed with DLD1 cells (DSMZ cat. no. ACC 278) expressing EGFR and LGR5 to determine the apparent affinity for cells expressing both targets. The assay used a constant concentration of target (cells) and the amount of radio ligand was titrated without violating the assumptions behind affinity measurements at steady-state conditions. Nonspecific binding was assessed by the presence of 100-fold molar excess of unlabeled MCLA-158 in a parallel series. The assay was repeated twice and the estimated K D value was reported as the mean of three independent experiments. Estimation of the K D values was performed using GraphPad Prism v.6.0 and the nonlinear regression equation 'One-Site-Total and Nonspecific binding' with the constraint that K D values must be greater than 0.

A431 apoptosis inhibition assay.
To test the capacity of the Fab232 monoclonal antibody to block EGF-mediated signaling, a cell-based assay was used to measure its ability to prevent EGF-induced apoptotic cell death of A431 cells 32 . A431 cells (DSMZ, cat. no. ACC 91) were plated at 1,500 cells per well in 96-well tissue culture plates and grown overnight. The next day, antibody was added at the indicated concentrations together with 62.5 ng ml −1 (10 nM) of recombinant, human EGF (R&D Systems, cat. no. 236-EG) and cells were grown for 3 days. After 3 days, the number of metabolically active cells was determined by addition of alamarBlue (Invitrogen, cat. no. DAL1100) and measurement of the fluorescence at 590 nm emission with 560 nm excitation. Fab232 monoclonal antibody was tested for its EGF-blocking effects in this assay compared to cetuximab and irrelevant control IgG.

RSPO-blocking assay.
For assessing the capacity to block RSPO, a FACS assay was used. Binding of antibody Fab072 and Fab049 at 100 ng ml −1 ; OMP88R20 (Oncomed, described in patent no. US-8628774-B2 at 50 ng ml −1 ) to LGR5 + CHO-K1 cells was assessed in FACS analysis in the presence of increasing concentrations of the ligand RSPO-1, RSPO-3 or RSPO-4 (ranging from 0.05 μg ml −1 to 19 μg ml −1 ; R&D Systems, cat. no. 4645-RF/CF). The MFI signal obtained at a certain ligand concentration was normalized to that obtained in the absence of ligand (set at 100%).
Colorectal cancer organoid biobank. The collection of patient data and tissues for the generation and distribution of organoids was performed according to the guidelines of the European Network of Research Ethics Committees, following European, national and local laws. The Biobank Research Ethics Committee of the UMC Utrecht (TCBio) approved the biobanking protocol under which this research was performed (12-093 HUB-Cancer). All donors participating in this study signed informed consent. Patients were identified by their treating physician/ specialist in the hospital and selected from those undergoing resection of the tumor. Patients received a letter with the information at least 2 days before their visit to the hospital. Furthermore, the biobank study coordinator in the hospital visited the patient and explained the research project and answered any questions related to the project. In addition, the patients were informed that they could retract permission at all times, without having to provide a reason for doing so. Available organoids can be requested at techtransfer@huborganoids.nl. P18T, P19Tb and P14T PDOs have been previously described 20 .
CRC organoids were established as described 20 . We did not preselect samples based on any criteria. We derived and stocked PDOs from 61 primary CRCs and 11 CRC liver metastases from a total of 99 tumor samples corresponding to 68 patients treated at two different hospitals (UMC Utrecht and Meander Medisch Centrum). For 31 patients, we also established PDOs from adjacent healthy mucosal tissue. We failed to establish 27 samples: 9 samples were contaminated; 14 tumor samples organoids were not obtained after seeding cells owing either to reduced amount of tumor cells in the surgical sample, inefficient dissociation or because suboptimal growth conditions; and PDOs from 4 samples could not be propagated in the long term.
CRC and healthy PDO expansion medium has been previously described 20 . EGF and HRG concentration determines the sensitivity of the screening assay. EGF was added at 2.5 ng ml −1 , 5 ng ml −1 or 10 ng ml −1 during screening or at 10 ng ml −1 or 50 ng ml −1 during expansion. HRG was used at 5 ng ml −1 . Wild-type organoid expansion medium is equivalent to the tumoroid expansion medium, with one replacement, 70 ml enriched advanced DMEM is replaced by a combination of 20 ml enriched advanced DMEM and 50 ml Wnt3A-conditioned medium 20 . Healthy tissue-derived colon organoids and colon tumoroids with wild-type APC are WNT-dependent and therefore require supplementation with WNT for expansion.
Sequencing of PDO biobank. DNA was extracted from the organoids and matched blood samples and the pairs were confirmed to originate from the same individuals using genotyping (Fluidigm). Exome sequences were selected for using by bait capture (Agilent) and sequenced on a HISEQ 2000 sequencer, with 75-bp paired-end reads. Average exome coverage was 147×. In addition, samples were genotyped using the Affymetrix SNP6 array, for genome-wide SNP coverage to be used by the copy-number analysis.
We identified driver mutations based on a list of 726 genes causally implicated in cancer (http://cancer.sanger.ac.uk/census/). For recessive cancer genes, we reported nonsense, frameshift, essential splice site and stop-lost mutations, as well as missense and in-frame mutations at recurrent positions according to the Cosmic database (http://cancer.sanger.ac.uk/cosmic). For dominant cancer genes, we reported missense or in-frame indels previously reported in other studies.

NATuRE CANCER
Among copy-number variants, we identified a set of regions in individual samples as amplicons (total copy number ≥5 if ploidy was lower than 2.7, otherwise total copy number ≥9) or homozygous deletions (total copy number of 0) and reported variants in genes or chromosomal regions previously reported 9 as implicated in CRC.
Antibody screen. For screening, the antibodies were randomized in v-bottom 96-well master plates (Greiner Bio-One 736-0118). The antibody concentrations applied in the primary screen were 10 µg ml −1 and 1 µg ml −1 (Phage panel) or 40 µg ml −1 and 4 µg ml −1 (Fab panel); and in the validation screen the antibody doses were 10 µg ml −1 and 2 µg ml −1 . The antibodies were added to the organoids 30 min after seeding. For most models the end point (day of fixation) was day 9, but in some PDOs with high growth rates, day 7 or 8 was chosen to prevent overgrowth of the control wells.
High-content imaging. Upon fixation and staining the nuclei and the actin cytoskeleton 64 , imaging of the organoids in the hydrogel was performed using an ImageXpress Micro XLS system (Molecular Devices MetaXpress v.5.3) 65 . Briefly, 20 to 24 z-stacks of each well in a 384-well plate were captured using a ×4 objective, with a z-step size of 50 µm. Confocal images were captured using an ImageXpress Micro Confocal (Molecular Devices MetaXpress v.6.2.3 for Figs. 1b-e, 2a-e, 5a,b and 7a and Extended Data Fig. 2a,b; and v.6.2.3 for Fig. 7b) or a Nikon Ti A1 confocal laser microscope (NIS-Elements software for Extended Data Fig. 6a,b) using a ×20 0.75 NA objective and processed in ImageJ. For antibody localization studies, the organoids were incubated for the indicated times with 1 µg ml −1 antibody before fixation in 4% PFA. Upon washing in TBP (0.1% Triton X100 (Sigma-Aldrich) + 0.5% BSA (Sigma-Aldrich) in PBS) the antibodies were detected with a goat-anti-human-Cy5 secondary antibody (1:2,000 dilution; Thermo Fisher Scientific) before washing in TBP and fixation and staining for the nuclei and actin cytoskeleton. Endogenous EGFR levels were detected using the mouse monoclonal antibody anti-EGFR (1:40 dilution; Thermo Fisher Scientific, cat. no. MA5-13319;) followed by a goat-anti-mouse-Alexa488 secondary antibody (1:2,000 dilution; Thermo Fisher Scientific).
Image analysis. Captured images were stored on a central data server, accessible by the Crown Bioscience Ominer 3D image analysis platform integrated into the KNIME Analytics Platform (v.2.11.3) 64,65 . The software analyzes the number and structure of the objects (nuclei and cytoskeleton) detected in each well, as >500 different features. After quality control checks, the Zʹ factor 66 is calculated to determine the separation window between positive (no growth factor) and negative controls (EGF, HRG or WNT3A). The Zʹ factor is defined as Z ′ = 1 − 3×(s.d. of positive control+s.d. of negative control) |mean of negative control−mean of positive control| . The z score-normalized data were analyzed to train an optimal subset of 3-20 features that distinguish the reference treatment effect from the negative control morphology. This subset of features is then used to calculate the Euclidian distance and scaled between zero and one to calculate the normalized multiparametric response. This unified score of morphology change was used to discriminate hits in the compound screens across all organoid models.

Cell cycle analysis in PDOs.
To assess the effect of the antibodies on cell cycle progression, organoids were treated for 48 h with a low dose of antibody (2 µg ml −1 ). Ethynyl-deoxyuridine was then added to the medium at a final concentration of 10 µM and incubated for 2 h. The organoids were then collected and processed to a single cell suspension. The Click-iT EdU Alexa Fluor 647 Flow Cytometry Assay kit was used according to the manufacturer's protocol. The organoids were counterstained with DAPI (1 µg ml −1 ) for 30 min on ice and then analyzed using the Gallios flow cytometry machine (v.1.2).
Subcutaneous PDX models. P18T and C31M were grown for 7 days before disaggregating into a single cell suspension for injection. Female NOD.CB17/ AlhnRj-Prkdcscid/Rj mice (Janvier Labs) aged between 6-8 weeks were used. Xenografts were initiated by subcutaneously injecting 100 µl of BME2:PBS (50:50) solution containing 200,000 single P18T or C31M cells. At the experimental end point, mice were killed and tumors were collected. One tumor was manually cut into small pieces approximately 0.5 mm × 0.5 mm × 0.5 mm (width × length × height). The pieces were then placed into four flanks of recipient NOD-SCID mice, using one piece/flank. A trocker was used to implant the pieces into the mice. Once the tumor volume on a mouse reached an average of 50 mm 3 , mice were injected once a week with PBS (vehicle), cetuximab (0.5 mg per mouse) or MCLA-158 (0.5 mg per mouse). Experiments were approved by the Animal Care and Use Committee of Barcelona Science Park (CEEA-PCB-14-000053). Maximum tumor burden allowed by the institutional review board (1,200 mm 3 per mouse) was not exceeded.
For CRC, gastric, esophageal and HNSCC, we selected PDXs belonging to different tumor types that showed elevated EGFR and LGR5 mRNA expression levels according to available RNA-sequencing data. The main mutations present in these PDX models are listed in Supplementary Table 14 and the strain of mice used for each model is in Supplementary Table 15. These tumor grafts were either tumor fragments or a cell suspension that had been collected from donor mice bearing established primary human tumors. When mean tumor size reached 100-150 mm 3 , mice were randomized over the treatment arms to receive an intraperitoneal (i.p.) injection of MCLA-158 in PBS. For HNSCC, mice were injected with 25 mg kg −1 per week MCLA-158 depending on their weight (n = 4 mice per group). For gastric/esophageal/CRC, mice were injected with 0.5 mg per week MCLA-158 (about 25 mg kg −1 ) regardless of the weight of the animal (gastric/esophageal, n = 4 mice per group; CRC, n = 8 mice per group). Mice were treated once a week for 6 weeks, followed by a 3-week observation period. Tumors were measured twice a week using calipers and tumor volume was calculated by assimilating them to an ellipsoid using the formula, length × width 2 × ½. The percentage change in tumor growth from baseline was calculated using the tumor volume data on day 0 and the tumor volume data on the last day at which all mice in both groups were still in the experiment. These experiments were approved by committee of the Ethics of Animal Experiments of Crown Bioscience (study nos. P1030 and E0591-U1913, E0591-U1906, E0591-U1907). In 10 out of 203 mice, the maximum tumor size allowed by the institutional review board (3,000 mm 3 ) was exceeded due to very fast tumor growth between the last measurement and the day of being killed.

Orthotopic CRC models.
To establish the LM-CRCX3 orthotopic CRC model, a small fragment of a unique lung metastasis from a patient with CRC previously treated with fluoropyrimidine-based chemotherapy was obtained with the informed consent of the patient. The tumor had a KRAS mutation (G12D) and was microsatellite stable. Briefly, a small tumor fragment of 4 × 4 mm 3 maintaining tridimensional structure was anchored to the serosa of the cecum of two 5-week-old male athymic nude mice (strain Hsd:Athymic Nude-Foxn1nu) purchased from Envigo. Twenty mice were implanted with LM-CRCX3 at passage no. 2 and 19 days later, when homogeneous small masses were detected at palpation, mice were randomized and allocated into three treatment groups (n

NATuRE CANCER
Research Involving Animals, developed by the Council for International Organizations of Medical Sciences. The maximum size allowed by the ethics committee for orthotopic colorectal tumors is 1,200 mm 3 and this limit was not exceeded.
For M001 and M005 orthotopic CRC models, carcinoma tissues were obtained from patients with CRC upon patient surgery. Then, 3 × 10 5 patient-derived tumor cells suspended in 50 ml of PBS were injected into the cecum wall of 8-10-week-old NOD-SCID female mice (NOD.CB17-Prkdcscid/NcrCrl) purchased from Charles River Laboratories 68 . PDX genotypes were characterized by exome sequencing. Main driver mutations are detailed in Supplementary Table 12. Both PDX models M001 and M005 were derived from liver metastases of patients with stage IV advanced CRC. Experiments with mice were conducted following the European Union's Animal Care Directive (86/609/EEC) and were approved by the Ethical Committee of Animal Experimentation of the Vall d'Hebron Research Institute (ID 17/15 CEEA and 18/15 CEEA). Mice were treated MCLA-158 (0.5 mg per animal in PBS i.p. once per week) starting on the day that orthotopic tumor growth was confirmed by microcomputed tomography. Control mice were injected with the corresponding amount of vehicle (PBS). Tumor growth was monitored by microcomputed tomography imaging during treatment. The limit established by the ethics committee for orthotopic experiments is that the tumor mass growing in the cecum must not substantially interfere with the normal physiological functions of the animal (ability to eat, drink, walk and groom) or cause pain or distress. To reach the end point of metastasis formation, all mice were killed when more than one-third of the cohort showed generals signs of illness. These limits were not exceeded in the current experiments. At this end point, mice were killed and complete necropsies were performed. Organs were macroscopically inspected for the presence of metastases. Primary carcinomas in the cecum, liver, lungs and any other visible tissue affected were collected for histological analysis. Standard H&E staining was performed on repeated sections of each tissue to confirm the presence of metastatic lesions.

Data availability
Organoid exome sequencing data have been deposited at European Genome-Phenome Archive (study no. EGAS00001004584). RNA-sequencing data of P18T and C55T organoids treated with antibodies are deposited at Gene Expression Omnibus (GSE186531). Microarray data of Fab072 + versus Fab072 − tumor cells are deposited at Gene Expression Omnibus (GSE190543). Further information and requests for resources and reagents should be directed to the corresponding authors. All requests for raw and analyzed data and materials will be reviewed promptly by the corresponding author to verify whether the request is subject to any intellectual property or confidentiality obligations. Any data and materials that can be shared will be released via a material transfer agreement. Source data are provided with this paper.

Reporting Summary
Nature Research wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in reporting. For further information on Nature Research policies, see our Editorial Policies and the Editorial Policy Checklist.

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend,

Software and code
Policy information about availability of computer code

Data
Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: -Accession codes, unique identifiers, or web links for publicly available datasets -A list of figures that have associated raw data -A description of any restrictions on data availability Organoid exome sequencing has been deposited at European Genome-Phenome Archive (study number EGAS00001004584). RNA sequencing data of P18T and CSST organoids treated with antibodies is deposited at Gene expression Omnibus (GSE186531). Microarray data of Fab072+ versus Fab072-tumor cells are deposited at Gene expression Omnibus (GSE190543). Further information and requests for resources and reagents should be directed to the corresponding authors. All requests for raw and analyzed data and materials will be reviewed promptly by the corresponding author to verify whether the request is subject to any intellectual property or confidentiality obligations. Any data and materials that can be shared will be released via a material transfer agreement. Source data are provided with this paper.

Field-specific reporting
Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection.

Life sciences study design
All studies must disclose on these points even when the disclosure is negative.

Sample size
No statistical method was used to predetermine sample size. Except for some experiments with PDXs in fig. Se, we used at least 5 mice per group, which is sufficient to detect meaningful differences based on historical record with similar experiments. We reproduced therapeutic results of antibody treatments using multiple different in vivo models. For the majority of in vitro experiments, we used n = >3 for according to historical experience with similar experiments. The sample size typically results in standard error <25% of the mean.

I
Data exclusions No data from in vitro experiments were excluded. For in vivo experiments, animals were excluded only if they died or had to be killed according to protocols approved by the animal experimental committees.

Replication
As reported in figure legends and methods, the findings were reliably reproduced with similar results. Therapeutic outcomes upon antibody treatments were reproduced using multiple different in vitro and in vivo models. Representative experiments included in the manuscript were repeated the indicated number of times with similar results as follows; Fig.3c; 3 times, Fig. 7b; 4 times, Fig 7c; 3 times, Fig. 7d; 2 times, Fig. 7e; 3 times, Extended data Fig. 6a-b; 4 times; Extended data Fig. 6c; 3 times for Pl8 and 1 time for C55T and C82N. For the rest of experiments, only data that we were able to replicate at least two times with equivalent results were included in the manuscript.
Randomization For in vivo experiments, animals were randomized, with each group receiving mice with similar tumor sizes. For in vitro experiments, all samples were analyzed equally with no subsampling; therefore, there was no requirement for randomization.

Blinding
Data collection and analyses were not performed blinded to the conditions of the experiments. For in vivo experiments, blinding was not performed because identification of mice was required for treatment administration and husbandry. For in vitro studies, experiments were not performed in a blinded manner because investigator need it to identify samples to perform treatments.

,I
We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material, system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response.
Goat anti-human lgG-Fc (Bethyl Laboratories cat# #A80-104A: 5 µg/ml for coating in anit-EGFR competition assays. OMP88R20 (generated as described for monoclonal antibodies) -50ng/ml in RSPO blocking assays Therapeutic antibodies: Monoclonal antibodies against WNT, RTK and control targets (generated as part of the article as described in the materials and methods and used as indicated in different experiments) Bispecific antibodies against WNT, RTK and control targets (generated as part of the article as described in the materials and methods and used as

Validation
All antibodies generated in this study were validated for specificity and binding by FACS and binding antigen positive/negative cell lines. Antibodies used for IF, IHC, flow cytometry and western blot have been previously validated on supplier's website and used in multiple studies previously.

Animals and other organisms
Policy information about studies involving animals: ARRIVE guidelines recommended for reporting animal research Note that full information on the approval of the study protocol must also be provided in the manuscript.

Human research participants
Policy information about studies involving human research participants All relevant patient population co-variates from which PDOs were derived are reported in Supplementary Table 3 We derived and stocked PDOs from 61 primary CRCs and 11 CRC liver metastases obtained from 68 patients treated at two different hospitals (University Medical Center Utrecht (UMCU) and Meander). Patients were identified by their treating physician/specialist in the hospitals. Patients were selected from those who are undergoing resection. The patients who will be included were diagnosed with colorectal cancer. Patients received a letter with the information at least 2 days prior to their visit to the hospital. Furthermore, biobank study coordinator in the hospital visited the patient and explained the research project and answered any questions related to the project. In addition, the patients were informed that they can retract the permission at all times, without having to provide a reason for doing so.
The collection of patient data and tissue for the generation and distribution of organoids has been performed according to the guidelines of the European Network of Research Ethics Committees (EU REC) following European, national, and local laws.  2c-d): Gates were set according to antibody control (TT) staining. We collected cells that stained positive or negative above and below this threshold as indicated in the figure.
-Fab072 antibody staining of dissociated xenografts (Fig 3h): We sorted alive (DAPI negative) epithelial tumor cells (Epcam+). Gates were set according to antibody control staining. We collected the brightest 10% of the cells. Negative cell fraction corresponded to center of the major pick as indicated in the figure.