Dimerization of kringle 1 domain from hepatocyte growth factor/scatter factor provides a potent minimal MET receptor agonist

Hepatocyte Growth Factor/Scatter Factor (HGF/SF) and its cognate receptor MET play several essential roles in embryogenesis and regeneration in post-natal life of epithelial organs such as liver, kidney, lung, and pancreas, prompting a strong interest in harnessing HGF/SF-MET signalling for regeneration of epithelial organs after acute or chronic damage. The limited stability and tissue diffusion of native HGF/SF, however, which reflect the tightly controlled, local mechanism of action of the morphogen, have led to a major search for HGF/SF mimics for therapy. In this work, we describe the rational design, production and characterisation of K1K1, a novel minimal MET agonist consisting of two copies of the kringle 1 domain of HGF/SF placed in tandem. K1K1 is highly stable and displays biological activities equivalent or superior to native HGF/SF in a variety of in vitro assay systems and in a mouse model of liver disease. These data suggest that this engineered ligand may find wide applications in acute and chronic diseases of the liver and other epithelial organs dependent on MET activation.


Introduction
Hepatocyte Growth Factor/Scatter Factor (HGF/SF) was simultaneously and independently discovered as a hepatotropic factor in the serum of hepatectomized rats (Nakamura et al, 1984) and as a scatter factor (SF) inducing morphological changes and epithelial cell migration in medium conditioned by human embryo fibroblasts (Stoker & Perryman, 1985;Stoker et al, 1987). After the discovery of its cognate receptor, the tyrosine kinase receptor MET (Bottaro et al, 1991), important studies using targeted gene disruption or blocking antibodies demonstrated embryonic developmental defects in placenta (Uehara et al, 1995), liver (Schmidt et al, 1995), skeletal muscles of limb and diaphragm , and kidney (Santos et al, 1994;Woolf et al, 1995). Collectively, these studies demonstrate an early and wide involvement of HGF/SF-MET signalling in organ development.
In post-natal life, the plasma level of HGF/SF rises immediately after injury of several organs such as liver, kidney, and heart (Lindroos et al, 1991;Nakamura et al, 2000;Zhu et al, 2000;Matsumoto & Nakamura, 1997). HGF/SF is essential for the regeneration of kidney (Zhou et al, 2013), liver (Borowiak et al, 2004), and skin (Chmielowiec et al, 2007) and, importantly, treatments with recombinant HGF/SF prevents fibrosis after experimental injury in kidney (Liu, 2004) and lung  providing a strong rationale for harnessing HGF/SF-MET signalling for regenerative medicine.
The mature, biologically active species of HGF/SF is a two-chain (/), disulphide-linked protein structurally related to the blood proteinase plasminogen. The -chain (69 kDa) comprises an N-terminal (N) domain homologous to the plasminogen activation peptide and four kringle (K) domains and contains the high affinity-binding site for MET. The -chain (34 kDa) is homologous to the catalytic domain of serine proteinases (SPH) (Donate et al, 1994); it is devoid of enzymatic activity but contains a secondary binding site for MET. During tissue remodelling and at sites of tissue damage, the activation of the clotting/fibrinolytic pathways promotes the cleavage of the single chain pro-HGF/SF into mature two-chain HGF/SF leading to MET activation. In this way, HGF/SF-MET signalling promotes cell survival, division, migration and a complex morphogenetic program underlying tissue regeneration (Miyazawa et al, 1994(Miyazawa et al, , 1996. The discovery of two alternative HGF/SF splice variants, NK1 and NK2 (Miyazawa et al, 1991;Chan et al, 1991;Cioce et al, 1996) enabled further progress toward the identification of regions important for receptor binding and activation (Hartmann et al, 1992;Lokker et al, 1992). NK1, a partial receptor agonist containing the N-domain and first kringle domain (K1) is monomeric in solution but assembles as a head-to-tail non-covalent dimer in crystal structures offering an attractive bivalent receptor-activating configuration (Chirgadze et al, 1999;Ultsch et al, 1998). Heparin and heparan sulphate proteoglycans cause dimerization of NK1 in solution (Schwall et al, 1996) and are essential for the partial agonistic activity of NK1 but not of HGF/SF (Zioncheck et al, 1995;Sakata et al, 1997;Schwall et al, 1996) and, on the strength of the available biochemical data and partial crystal structures, alternative models of HGF/SF-MET activation were proposed involving with 1:2 or 2:2 stoichiometries (Fig 1A) ( Chirgadze et al, 1998;Donate et al, 1994;Miller & Leonard, 1998).
A recent cryo-electron-microscopy (EM) structure of HGF/SF in complex with the extracellular domain of MET (MET927) fused to a dimerization motif has offered new insights into the binding mechanisms (Uchikawa et al, 2021). This structure unambiguously shows a single HGF/SF molecule bridging two MET receptors by engaging both the NK1 and SPH binding sites with a second HGF/SF molecule and heparin stabilising the interaction. In contrast, a second structure of the NK1-MET927 complex confirmed the presence of the complex with 2:2 stoichiometry postulated on the basis of the early crystal structures of NK1.
This structure validates independent studies on the N and K1 domain of HGF/SF produced in yeast (Holmes et al, 2007) or by total chemical synthesis (Raibaut et al, 2013;Simonneau et al, 2015) suggesting K1 domain as the main or sole domain of HGF/SF responsible for the primary, high-affinity interaction between NK1 or HGF/SF and the SEMA domain of MET.
On the strength of the structural, biochemical and mutagenesis data available, we rationalised that a covalent dimer containing a tandem repeat of the K1 domain of HGF/SF could constitute a minimal and stable MET agonist able to diffuse effectively across tissue boundaries owing to the absence of the N-terminal domain of HGF/SF responsible for high-affinity binding to heparan sulfate proteoglycans (HSPG). Here we report the development of this new MET ligand which we name K1K1. We describe its structure and demonstrate that K1K1 is stable and displays biological activity, equivalent or superior compared to HGF/SF, in both in vitro and in vivo models. This new MET ligand may constitute a major step towards the prospect of harnessing MET signalling for the therapy of acute and chronic degenerative diseases of major epithelial organs.

Results
Rational for the design of K1K1. The molecular architecture of full MET receptor, its extracellular fragment and the various agonists used in this study are presented figure 1B. The concept of connecting two HGF/SF kringle 1 domains in tandem for the creation of the potent MET agonists described in this work comes from the initial observation that NK1 monomers form dimers upon crystallisation providing two MET receptor binding sites at opposite ends (Ultsch et al, 1998;Chirgadze et al, 1999). Within the structure of the NK1 dimer, the N-and C-termini of both kringle domains are all located in close proximity at the centre of the dimer structure, suggesting that a crosslinker could be made to link two kringle domains for mimicking the arrangement of K1 domains in NK1 dimeric complex. The recreation of such a configuration was first achieved using the streptavidin molecular platform combined with Cterminally biotinylatined synthetic kringle 1 (K1B) (Simonneau et al, 2015). Oligomerisation would place the K1B C-termini close together (~ 2 nm) around the streptavidin core and project the K1-MET binding site located around residue Glu159 away from it, mimicking the arrangement of K1 domains in NK1 crystal structures (Fig. 1C). The remarkable potency of streptavidin-K1B complex confirmed our assumption that multimerisation would lead to strong receptor activation and thus completed earlier observations for the location of the high-affinity receptor binding site in the first kringle domain (Lokker et al, 1994;Chirgadze et al, 1999;Gherardi et al, 2006). As potential heterogeneity of the streptavidin-K1B complex as well as the streptavidin moiety itself precluded its development as a therapeutic and would hamper structural studies, we hoped that strong agonistic activity could be achieved as well by associating two K1 domains, in tandem and covalently (Fig. 1C). Importantly, the activity of a such a covalent K1 dimer would no longer rely on a protein dimerization or multimerization assembly as required for wild-type NK1 (Sakata et al, 1997;Schwall et al, 1996) and streptavidin-K1B.
In wild-type NK1, the distance between the last cysteine in one kringle and the first cysteine in the other kringle is roughly 1 nm, giving possibility to associate the two kringle 1 domains in tandem. An asset of such a design is to enable the production of the molecule by recombinant methods. Therefore, we designed a linker to bridge the distance between the C-terminus of the first K1 domain and the N-terminus of the second one. We chose for that purpose the four amino acid long linker, SEVE, which is naturally present between kringle 1 and kringle 2 (K2) in HGF/SF. This native linker showed no contacts with neighbouring kringle domains in wildtype NK2 reported by Tolbert and colleagues (PDB code 3HN4) (Tolbert et al, 2010) and in NK2 in complex with heparin (PDB code 3SP8) and was therefore believed to not constrain the relative orientation of the K1 domains. Furthermore, a poly-histidine tagged variant, designated K1K1H6, was produced to facilitate certain assays as well as a single kringle 1 domain poly-histidine tagged variant (K1H6) as a monovalent control. As earlier studies had revealed a secondary low-affinity heparin binding site in the kringle 1 domain affecting biological activity (Lietha et al, 2001), we designed two additional variants, K1K1S2 and K1K1S4. In these variants, we have introduced 2 or 4 reverse-charge mutations in each kringle (schematically shown in Fig. 1B) in order to test the implication of the basic residues naturally present at those positions and to evaluate their importance for the biological activity.
Upon expression in Escherichia coli, K1K1 was abundantly present in inclusion bodies.
Expression at 18 °C and induction with a low IPTG concentration (0.1 mM) allowed the production and extraction of fully folded protein from "non-classical" inclusion bodies using a mild, non-denaturing arginine-based extraction method derived from the work of Jevševar and colleagues (Jevševar et al, 2005). Expression in these inclusion bodies offers an inexpensive and abundant source of recombinant protein with relatively few contaminants as opposed to soluble bacterial expression. The low-affinity heparin binding site in the kringle domain still allowed an effective single-step purification by heparin Sepharose® affinity chromatography following arginine extraction (Fig. S1A). An additional size exclusion chromatography step was used to remove possible traces of aggregation to yield crystallography-grade material with UPLC-MS analysis confirming the purity and identity of the proteins as shown in Fig. S1B and S1C for K1K1 and K1K1H6 respectively. Figure S1D shows all purified recombinant proteins used in this study on Coomassie-stained gel under reducing conditions. Structure determination by x-ray diffraction and SAXS. Purified proteins were used for crystallization experiments using commercial sparse matrix screens and several conditions resulted in the growth of protein crystals at 17 °C within days. The collected datasets enabled the determination of the molecular structures at 1.8 Å resolution and 1.7 Å for K1K1 and K1K1H6 respectively. Both proteins crystallized in space group P 1 21 1 with K1K1 having two molecules per asymmetric unit and K1K1H6 having one.
Structurally, K1K1 and K1K1H6 are nearly identical with a backbone root mean square deviation (RMSD) of 0.8 or 1.8 Å (Fig. 2B), depending on which of the two molecules in the asymmetric unit of K1K1 is used for alignment with K1K1H6. Both proteins show an elongated "stretched-out" conformation with the MET binding sites exposed on opposite ends and the Nand C-terminus located at the centre. In K1K1H6, good electron density was also observed for the poly-histidine tag possibly due to stabilizing interactions between the histidine residues and residues of the N-terminal kringle domain.
Surprisingly, like the revelation of seeing the first structures of NK1 that presented a likely biologically relevant dimer, both crystal structures show a pseudo C2 symmetry around a central axis and present both receptor binding sites in nearly identical orientation on either side of the molecule (located around residue Glu37 in the first kringle and residues Glu120 in the second kringle, equivalent to NK1 residue Glu159). One can envision a binding mode in which two MET receptor monomers are brought in proximity by binding on opposite sides of K1K1, forming a 2:1 receptor activation complex. K1K1 therefore presents the most minimalistic peptide-based receptor agonist mimicking the binding and receptor activation mechanism proposed for wild-type NK1 but with the important difference that ligand dimerization is not required. This covalent "mimicry" is even more evident when comparing the structures of the NK1 dimer and K1K1. Alignment of the first kringle domain of K1K1 with one of the two kringle domains within the NK1 dimer places the second K1K1 kringle close to the position of the kringle domain of NK1 protomer 2 (Fig. S2A). However, the straightened conformation observed in the crystal structure of K1K1, "misaligns" the second K1K1 kringle domain through a rotation by 109.7 degrees and a translation of roughly 14 Å (Fig. S2B). No contacts between the two K1 domains or the kringle domains with the SEVE linker are observed. The linker is more stretched out compared to the two NK2 structures supporting our idea that it is highly flexible and therefore allows significant conformational freedom (Fig. S2C). This flexibility was confirmed with SAXS measurements of K1K1 in solution which generated an envelope that perfectly accommodates K1K1 volume-wise but only in a bent conformation ( Fig. 3A). This is also evident from comparing the experimental scatter curve with the one generated based on the K1K1 crystal structure using CRYSOL indicating that the overall volume is the same but the distribution and therefore conformation differs significantly.
Notably, the K1K1 solution SAXS envelope is compatible with the two kringle domains in the orientation found within the NK1 dimer (Fig. S2D). Excitingly, SAXS measurements of K1K1 in complex with MET567 gave a molecular mass estimation of 86.8 kDa and a calculated envelope that corresponds to a 1:1 complex with a "pan handle" extrusion which could accommodate K1K1 bound to the SEMA domain. Using a low-resolution crystallographic dataset of NK1 in complex with MET567 (unpublished, briefly described in Blaszczyk et al. (Blaszczyk et al, 2015)), we could generate a model based on the alignment of K1K1 with the crystal structure of the NK1-MET567 complex. The theoretical scatter curve matched the experimental scatter curve remarkably well (Chi 2 =3.6) with also in this model a slightly bent K1K1 conformation fitting best with the SAXS envelope. Moreover, the recently published cryo-EM structure of the NK1 dimer in complex with two SEMA domains (Uchikawa et al, 2021) perfectly accommodates our K1K1 SAXS envelope overlapping both NK1 kringle domains and supports the proposed model for receptor activation by K1K1 (Fig. 3C, Fig. S2E).
Overall, we have generated compelling structural support for a minimal MET receptor agonist that comprises two covalently linked MET binding sites (see Supplementary Video 1). The structure of K1K1 and K1K1H6 are available from the PDB data bank with id code 7OCL and 7OCM respectively. The SAXS data is available from SASBDB database with id codes SASDLM9, SASDLN9, and SASDLP9. Having established MET receptor binding and remarkably potent activation of the MET signalling pathway by both K1K1 and K1K1H6, we subsequently focused on demonstrating its potency in various phenotypic biological assays. We first performed a cell viability assay based on metabolic activity on MDCK cells incubated overnight (16 hours) in the presence of anisomycin, an apoptosis inducer, which prevents protein synthesis and leads to cell death. We looked at the cell survival after the addition of HGF/SF, K1K1, K1K1H6 and NK1 at different concentrations during anisomycin treatment. While HGF/SF was most effective, K1K1 and K1K1H6 were equals in preventing cell death and much more potent than native NK1 in this assay (Fig. 4E). We also used this assay to validate the specific activities of different preparations of K1K1 and K1K1H6 showing very little batch-to-batch variation and similar relative activity when compared to HGF/SF and NK1 ( Fig. S4A) confirming the reliability of the protein production and purification protocol.
An MDCK scatter assay (Stoker & Perryman, 1985) was used as a sensitive phenotypical assay to determine the minimal concentration needed to activate MET and induce the epithelialmesenchymal transition and cell scattering. K1K1 was able to exert an effect at a concentration down to 1 pM, ten times lower than native HGF/SF and one thousand times lower than NK1 ( Fig. S4B).
To examine the role of the low-affinity heparin binding sites in the kringle domain, we used the analogues K1K1S2 and K1K1S4 with specific reverse-charge mutations altering heparin binding. These variants were compared with HGF/SF and K1K1 in an MDCK scattering assay ( Fig. 5A). Both variants had reduced activity with observable scattering for K1K1S2 down to 300 pM and no scattering observed at 100 pM while K1K1 still showed activity at a concentration ten times lower (10 pM). K1K1S4 only showed activity in the high nanomolar range and was inactive at 1 nM. We further performed a 3D morphogenesis assay using MDCK cells ( Fig. 5B and Fig. S5A and B) and confirmed the results obtained with the scatter assay, indicating that the reverse-charge mutations reduced the biological activity of K1K1S2 and K1K1S4. The loss of biological activity induced by the mutations was also observed in SKOV3 cells using a quantitative Boyden chamber migration assay (Fig. 5C). Finally, we examined the impact of the mutations on biological activity at the level of the MET receptor signalling pathway by investigating the phosphorylation status of MET, Akt, and ERK (Fig. S5C).
Altogether these data indicate that K1K1S2 and K1K1S4 are weaker agonists than K1K1 by one and two orders of magnitude respectively and suggest that the low-affinity heparin binding site within the K1 domain plays a significant role in MET activation by K1K1.

In vivo activation of MET signalling.
Having established the superior potency of K1K1 over NK1 in different in vitro assays that sometimes matched or even surpassed native HGF/SF, we were keen to start several in vivo mouse studies to look at the effects of K1K1. We first validated that K1K1 is diffusible and biologically active in vivo, whatever the injection route (intraperitoneal (I.P.) versus intravenous (I.V.)), by measuring phosphorylation of the MET receptor, Akt, and ERK in liver homogenates 10 minutes after injection (Fig. S6). Since no significant difference was observed between I.P. and I.V. administration routes, all subsequent injections were done by I.P for experimental convenience. To study the effect of different In vivo efficacy of K1K1 in the treatment of alcoholic steatohepatitis. To evaluate the in vivo efficacy of K1K1 applied to a liver disease model, we used a validated mouse model of subchronic alcohol exposure, the adapted Lieber DeCarli (LDC) model (Lieber & Decarli, 1989) in order to study steatosis, a common feature of several liver diseases.
As expected, alcohol consumption induced steatosis in liver of mice fed with alcohol (LDC + Vehicle group) which was not observed in mice fed with a control diet (Control + Vehicle) or a control diet with K1K1 treatment (Control + 10 µg K1K1) ( Fig. 7A and Fig. S7A). The treatment with different doses of K1K1 (0.4, 2, and 10 µg) significantly decreased the steatosis in mice fed with alcohol (LDC + K1K1) (Fig. 7B). Regarding the factors involved in steatosis improvement, K1K1 treatment was able to significantly increase mRNA expression of ApoB, PPARα and LDLR (Fig. 7C). The highest doses of K1K1 (2 and 10 µg) also significantly decreased the mRNA expression of TNFα and IL-6 in LDC mice, while a moderate increase of both markers was observed in untreated mice compared to control health mice. Interestingly, while mRNA expression of MET decreased by ethanol exposure alone, it was induced by K1K1 treatment (Fig. S7B).
Taken together these results show that K1K1 significantly improved steatosis in a mouse model of liver injury. Our data suggest that this therapeutic effect is mediated in part by the increase of protective factors such as ApoB and PPARα and the lowering of proinflammatory cytokines, induced by K1K1 treatment.

Discussion
HGF/SF-MET signalling is essential for liver development (Schmidt et al, 1995;Uehara et al, 1995) and regeneration (Borowiak et al, 2004;Huh et al, 2004;Liu, 2004) and pre-clinical studies employing either recombinant HGF/SF (Horiguchi et al, 2009), HGF/SF gene delivery (Ueki et al, 1999) or HGF/SF-transfected mesenchymal stem cells (Moon et al., 2019) all demonstrated a strong therapeutic potential for HGF/SF in pre-clinical models of liver diseases. However, all failed to offer a robust platform toward the use of HGF/SF in the clinic.
Indeed, both gene therapy and cell-based therapies meet considerable regulatory challenges and therapeutic applications of recombinant HGF/SF face strong and well-known limitations: the protein has a complex multidomain protein susceptible to proteolytic degradation, is unstable in physiological buffers, and exhibits limited tissue diffusibility and penetration.
Hence, harnessing the HGF/SF-MET pathway in regenerative medicine will rely on engineered and improved HGF/SF derivatives in terms of stability and tissue availability.
Several molecules have been developed in recent years to overcome the limitations of HGF/SF in therapy. These included engineered variants of the natural growth factor (Cassano et al, 2008;Sinha Roy et al, 2010), agonistic monoclonal antibodies (Gallo et al, 2014;Kim et al, 2019), synthetic cyclic peptides (Sato et al, 2020), and aptamers (Ueki et al, 2020). All the engineered variants of HGF/SF developed thus far, however contain the high-affinity HSPG binding site present in the N-domain and thus share with HGF/SF the feature of limited diffusion and tissue penetration. The design of a new MET ligand with reduced affinity for HSPG, therefore was the prime objective of the present study. Building on available structures, mutagenesis data, and the biological activity of multimeric assemblies of the K1 protein (Simonneau et al, 2015), we rationalised that a K1K1 covalent dimer may have yielded a MET agonist with attractive physico-chemical and biological properties.
The structural data reported in this work show that K1K1 displays a perfect symmetrical positioning of the K1 units in its crystalized form and thus readily opens the possibility of a direct MET dimerization in a 1:2 K1K1:MET configuration. SAXS data indicate a bending of K1K1 when in solution, nicely fitting the natural position of the K1 domains within the NK1 dimer bound to MET as observed in recent cryo-EM structural analysis (Uchikawa et al, 2021).
Thus, the structural data suggest that flexibility is important for enabling the proper positioning of K1 domains upon binding to MET.
K1K1 displayed a vastly superior stability in physiological buffers compared to native HGF/SF and a potency approaching activity at picomolar concentrations in vitro ( Fig. 4A and   Fig. 5A). Importantly, the improved diffusibility into tissues as a result of reduced binding to heparin (Fig S8) clearly is a further and major component of the remarkable activity displayed by K1K1 in the studies in vivo (Figs 6 and 7). SPR analysis confirmed a ten-fold reduction of heparin binding in K1K1 (KD ~5.6 M) compared to NK1 (KD ~0.43 M) while the affinity of the K1K1 mutants S2 and S4 was below the limits of sensitivity of SPR (Fig. S8). Interestingly, these mutants displayed reduced biological activity compared to K1K1 indicating that the presence of the low affinity site for heparan sulfate in the K1 domain plays a role in the potent agonistic activity in vitro as well as in vivo. Seen from a different angle, K1K1 can be regarded as a tool that can be easily manipulated and varied to investigate more in depth the role of heparan sulfate molecules in ligand-receptor binding and receptor activation, independently of the strong HS-binding N-domain.
Altogether, the structural, biochemical, and biological data reported here suggest that K1K1 might serve as a lead candidate for protein therapy of several human pathologies where activation of the HGF/SF-MET signalling axis can be beneficial. Activation of the MET pathway has been known to ameliorate alcoholic steatohepatitis as well as liver fibrosis or cirrhosis in pre-clinical models of liver disease (Matsuda, 1997;Matsuda et al, 1995;Tahara et al, 1999;Taı̈eb et al, 2002) and, conversely, deletion of the MET gene in postnatal mice causes hyperlipidemia, severe steatosis, and progressive liver disease (Kroy et al, 2014). No specific drug has been approved as yet for the treatment of alcoholic and non-alcoholic liver disease (Sharma et al, 2020) but we submit that the K1K1 protein data presented here may emerge as a strong candidate for clinical development owing to its favourable physical chemistry and its strong activity in a well-established clinical mouse model of alcoholic steatohepatitis ( Fig. 7 and S7A).
Finally, we wish to emphasize that HGF/SF is a broad-acting morphogen that controls development and regeneration of other major epithelial organs beyond liver: HGF/SF -MET signalling is essential for lung alveologenesis (Calvi et al, 2013) and for kidney development (Ishibe et al, 2009). Thus, we surmise that potent MET agonists may find many applications beyond liver disease notably in the large number of patients with acute and chronic kidney damage (Zhou et al, 2013) and chronic lung emphysema (Shigemura et al, 2005) and may well find an urgent application in reducing lung damage and fibrosis in patients suffering from COPD (chronic obstructive pulmonary disease) or severe COVID-19 (George et al, 2020;Zuo et al, 2010).

Cloning of K1K1 and K1K1 variants
The cDNA encoding K1K1 and K1K1H6 was generated using a fusion PCR reaction after separate PCR amplification of two K1 domains from a wild-type human NK1 template. The

Protein expression
Recombinant human NK1 protein (residues 28-209) was expressed in Pichia pastoris and purified as described in Chirgadze et al., 1999(Chirgadze et al., 1999. Recombinant human MET567 was expressed in CHO Lec 3.2.8.1 cells and purified as described in Gherardi et al., 2003(E Gherardi et al., 2003. Human HGF/SF was stably expressed in NS0 myeloma cells and purified as described in Gherardi et al., 2006(Ermanno Gherardi et al., 2006. For the expression of K1K1 and K1K1 variants, a single colony of freshly transformed E. coli BL21(DE3) was used to inoculate an overnight 5 mL Luria broth (LB) culture containing 100 g/mL of ampicillin. This culture was used to inoculate 500 mL of LB with ampicillin shaking at 250 rpm and grown to an optical density (600 nm) of 0.6-0.8 at 37 °C. The culture was then cooled down to 18 °C while shaking after which IPTG was added to a final concentration of 0.1 mM. The culture was grown at 18 °C with shaking for 24 h after which the bacterial cells were harvested by centrifugation for 30 min at 10,000 rcf at 4 °C and the cell pellet was stored at -80 °C.

K1K1 extraction from inclusion bodies
The frozen bacterial cell pellet was thawed on ice and resuspended in 25 mL 50 mM Tris pH 8.5, 500 mM NaCl, with the addition of one tablet of cOmplete EDTA-free Protease Inhibitor Cocktail (Roche), 1 unit of Pierce Universal Nuclease and 10 g of lysozyme. The suspension was incubated rotating at 4 °C for 30 min before being placed on ice and sonicated using ten pulses of 30 s with 60 s pause in between using a Sonic Ruptor 400 (Omni International) with a OR-T-750 tip at 100% power output. The suspension was centrifuged at 10,000 rcf for 10 min, the supernatant discarded, and the pellet resuspended in 25 mL 50 mM Tris pH 8.5, 500 mM NaCl with 0.4% Triton X-100 using a glass Potter-Elvehjem tube and PTFE pestle. The suspension was incubated at 4 °C rotating for 30 min after which it was centrifuged at 10,000 rcf for 10 min, the supernatant discarded, and the pellet resuspended in 25 mL 50 mM Tris pH 8.5, 500 mM NaCl with 0.025% NP40 using the Potter-Elvehjem tube and pestle. The suspension was again incubated at 4 °C rotating for 30 min, centrifuged for 10 min at 10,000 rcf, the supernatant discarded, and the pellet resuspended in 25 mL 50 mM Tris pH 8.5, 500 mM NaCl using the Potter-Elvehjem tube and pestle. The suspension was once more incubated at 4 °C rotating for 30 min, centrifuged at 10,000 rcf for 10 min, and the supernatant discarded.
The final pellet was resuspended in 20 mL 50 mM Tris pH 8.5, 500 mM NaCl, 2 M arginine, 0.5 mM GSSG, 5 mM GSH. This suspension was incubated for three days at 4 °C on a rotary wheel.

K1K1 purification
After three days of incubation in Tris buffer with arginine, the inclusion body suspension was centrifuged at 20,000 rcf for 30 min at 4 °C. The supernatant was transferred to a new centrifuge tube and centrifuged again at 20,000 rcf for an additional 30 min at 4 °C. Unless a pellet was visible, in which case a third centrifugation was performed in a new tube, the cleared supernatant was diluted 100 times in 2 L of 50 mM Tris pH 7.4, 150 mM NaCl and filtered.
The diluted supernatant was loaded onto a 5 mL Heparin HiTrap column (GE Healthcare) and eluted with a gradient up to 1 M NaCl, in 50 mM Tris pH 7.4. Peak fractions were pooled, concentrated, and loaded on a HiLoad 16/600 Superdex 200 pg column (GE Healthcare) equilibrated in 50 mM Tris pH 7.4, 500 mM NaCl. Peak fractions were pooled and concentrated to 5 mg/mL before being used or stored after flash freezing. For K1K1S2 and K1K1S4, instead of Heparin HiTrap™ column, the first step purification was done on a 5 mL HisTrap™ HP column (GE Healthcare).

Protein crystallization and x-ray diffraction
For protein crystallization, K1K1 and K1K1H6 were dialyzed in 10 mM Tris pH 8.5, 100 mM NaCl and concentrated to 12 mg/mL and 11.8 mg/mL respectively. A pre-crystallization screen performed with different protein to crystallization solution ratios (1:2, 1:1, 2:1) in a 24-well plate using hanging drop. After 24 h, good crystals were obtained in a 2:1 protein-to-solution ratio condition. In addition to the cryo-protection intrinsic to the Morpheus screen, additional glycerol was added up to 25% of the final volume before crystals were collected using a CryoLoop (Hampton Research) and flash frozen and stored in liquid nitrogen. Crystallizations of K1K1 went through a similar optimization procedure and yielded good crystals in a condition of 100 mM Tris/Bicine pH 8.5, 30 mM Sodium nitrate, 30 mM Sodium phosphate dibasic, 30 mM Ammonium sulfate, 12.5% w/v PEG 1000, 12.5% w/v PEG 3350, 12.5% v/v MPD. Data was collected at the ESRF in Grenoble, France, at beamline at BM14 for K1K1H6 and ID23-1 for K1K1 and was solved by molecular replacement using the NK1 kringle domain taken from 1BHT.pdb (Ultsch et al., 1998) (Liebschner et al, 2019) using the kringle domain from 1BHT.pdb (Ultsch et al., 1998) followed by several rounds of refinement using Phenix refine (Afonine et al., 2012) and WinCoot (Emsley et al, 2010) (v0.9.4.1). The dataset of K1K1 and K1K1H6 had an upper resolution limit of 1.61 and 1.39 respectively, a resolution cut-off was set to 1.8 for K1K1 and 1.7 for K1K1H6. The data collection and refinement parameters and statistics are given in table 1. The data collection and refinement statistics are given in Supplementary table S1. The molecular graphics presented in this manuscript were generated with PyMOL (v2.4.1) and Chimera (v1.15) (Pettersen et al, 2004).

Small angle X-ray scattering (SAXS)
Solution scattering data were collected at ESRF BM29 using a sec-1 frame rate on Pilatus 1 M detector located at a fixed distance of 2.87 m from the sample, allowing a global q range of 0.03-4.5 nm with a wavelength of 0.01 nm. SEC-SAXS experiments were carried out using Nexera High Pressure Liquid/Chromatography (HPLC; Shimadzu) system connected online to SAXS sample capillary. For these experiments, 35 L of each sample at the concentrations indicated in the Table were injected into a Superdex 200 PC 3.2/300 Increase column (GE Healthcare), pre-equilibrated with 25 mM Tris pH 7.4, 150 mM NaCl. Frames corresponding to protein peaks were identified, blank subtracted and averaged using CHROMIXS2. Radii of gyration (Rg), molar mass estimates and distance distribution functions P(r) were computed using PRIMUS in the ATSAS package (v3) (Petoukhov et al, 2012). Modelling of flexible loops and glycosylation were carried out using COOT, CORAL and SASREF. Comparison of experimental SAXS data and 3D models from structural models was performed using Supplementary   Table S2.

AlphaScreen™ phosphorylation assay
Measurement of ERK1/2 (T202/Y204) and Akt1/2/3 (S473) activation were performed on HeLa cells using AlphaScreen™ SureFire Kits (#TGRES500 and #TGRA4S500) according to manufacturer protocol with minor modifications. Into a 96 well plate (#353072, Flacon, Corning), 10 000 HeLa were seeded in 200 µL of DMEM supplemented with 10% FBS and let attached and grown for 24 h. Cells were next starved in DMEM supplemented with 0.1% FBS for 2 h. Cells were treated for 8 min with MET agonist from 0.3 to 500 pM with HGF/SF, from 3 pM to 50 nM with K1K1 or K1K1H6 and from 3 to 1000 nM with K1H6. Cells were quickly washed in cold PBS and AlphaScreen™ lysis buffer was added for 15 min under middle shaking (600 rpm on orbital rocker, MixMate, Eppendorf). Immediately, 5 µL of lysate in analyzed by addition of acceptor and donor bead mixes for 2 h at 23 °C. The emitted signal intensity was measured using standard Alpha settings on an EnSpire® Multimode Plate Reader.
Measured are expressed has experimental duplicates (mean+/-SD, n=2). The data were subjected to a non-linear regression analysis using four-parameter sigmoidal equation using SigmaPlot® software (v13).

Surface plasmon resonance (SPR) analysis
Affinity constants were measured using a Biacore T200 (GE Healthcare) at 25 °C using PBST (PBS + 0.05% Tween 20) as running buffer and a flow rate of 30 L/min. For the measuring of affinity for MET, the receptor fragment MET567 comprising the N-terminal SEMA domain and the cysteine-rich/PSI domain was immobilized at low-density (515 RU) using aminecoupling on a CM5 sensor chip (GE Healthcare). A blank inactivated reference channel was used to correct for non-specific binding. 300 seconds injections were done using serial-diluted samples of K1K1, NK1, K1K1S2, and K1K1S4 testing a high range of 2 M-0.125 M and low range of 0.8 M-12.5 nM injecting for 300 seconds to reach equilibrium. For the measurement of affinity for heparin, a heparin-coated SC HEP0320.a chip (Xantec) was used.
With no reference channel available, all four channels were used for analysis. K1K1, NK1, K1K1S2, and K1K1S4 were injected in a wide range of concentrations (10 M to 156 nM) for 300 seconds. Biacore Evaluation software version 3.2 was used for calculating affinity constants. Curves were plotted using Graphpad Prism (v5.04).

SKOV-3 cell migration assays
Agonist-induced cell migration was measured by seeding 50,000 SKOV-3 cells resuspended in serum-free RPMI supplemented with 0.1% BSA (BioFroxx) in each of the upper wells of a 96-well Boyden chamber (AC96, Neuro Probe). HGF/SF, K1K1, K1K1S2, and K1K1S4 was loaded in the lower compartment and cells were left to migrate in a humidified 37 °C incubator with 5% CO2 for 6 h. Afterwards, non-migrated cells were removed, the cells migrated over the collagen coated (100 µg/mL Purecol, Nutacon) 8 m pore-size membrane (Neuro Probe) were fixed in 4% paraformaldehyde in PBS and stained using HCS CellMask™ Green stain (ThermoFisher Scientific). Fluorescent intensity was measured using an Azure C600 (Azure Biosystems) and migration was quantified after control-adjustment and charted in Graphpad Prism.

Madin-Darby canine kidney (MDCK) scatter assay
The assay was performed according to Stoker et al. (Stoker et al, 1987). MDCK cells were seeded at low density (2,000 cells/well on a 12-well plate, #353043 Falcon, Corning) in DMEM supplemented with 10% FCS to form compact colonies. After treatment, when colony dispersion was observed and the cells were fixed and colored by Hemacolor® stain (Merck, Darmstadt, Germany) according to the manufacturer's instructions. Representative images were captured using a phase contrast microscope with 40× and 100× magnification (Nikon Eclipse TS100, Tokyo, Japan).

Morphogenesis Assay
Wells of a black thin clear bottom 24-well plate (#4ti-0241, Brooks Lifesciences) were coated with 200 µL of 2 mg/mL type 1 rat tail collagen (354249, Corning/BD Biosciences) diluted in DMEM equilibrated in 1/10 v/v of 7.5% w/v sodium bicarbonate solution (#25080-060, Gibco, ThermoFisher). Cells (500-700) were seeded into a 500 µL layer of a 1:1 collagen/Growth Factor Reduced MatrigelTM (#354230, Corning/BD Biosciences) mixed in DMEM balance with bicarbonate and were treated twice a week for one month. The cells were then fixed for 24 h with 4% paraformaldehyde in PBS, coloured for 3 h with 0.1 mg/L Evans Blue in PBS (#E2129, Sigma Aldrich), extensively washed in PBS and finally nuclei stained with 3 nM DAPI is PBS. Cell were observed using inverted microscope observed and confocal microscope (LSM880 Zeiss) with long distance 10× objective using 488 nm excitation Argon laser and LP 500-750 nm emission filter. 3D reconstitutions (14 µm slices) and maximum intensity projections of z-stack were realized with Zen software (v8.1).

Survival Assay
In The data were normalized to percentage of control cells signal and subjected to a non-linear regression analysis using four-parameter sigmoidal equation with minimum=0 and maximum=100 using SigmaPlot software (v13 and v14.5). week-old C57BL/6 NRJ mice weighing 19-21 g (Janvier Labs) were intraperitoneally or intravenously injected with K1K1 (5 µg/mouse) and sacrificed after 10 min of treatment. Livers were immediately perfused with PBS supplemented with protease and phosphatase inhibitors, collected and then rapidly frozen in liquid nitrogen for subsequent protein extraction and Western blot analysis.

Liver steatosis model
Eighty C57BL/6J female mice of 8 to 10 weeks ( ̴ 19 g) provided by Janvier Labs were used in this study. They were housed in temperature-and humidity-controlled rooms, kept on a 12 h light-dark cycle. Animal procedures were conducted in accordance with French government policies. Mice were fed with a control liquid diet of adapted Lieber DeCarli ad libitum during a 7-day habituation process. Body weight and liquid diet consumption were monitored every two days. Mice were then separated in 6 groups: unlimited access to adapted LDC containing ethanol and control mice were pair-fed with isocaloric control diet over this feeding period. On the 11th day, mice were sacrificed.

Hematoxylin-erythrosine B staining
The livers embedded in paraffin and sectioned at 4 µm thick were deparaffinized and rehydrated in successive baths of xylene, absolute ethanol, ethanol 95% and tap water. The slides were stained with hematoxylin (Harris), rinsed, and stained with erythrosine B 0.8% (Sigma). The slides were finally dehydrated and mounted with Eukitt (GmbH). Slides were analyzed on a Zeiss Axio Imager M2 microscope. For histological evaluation, a score between 0 and 3 (3 = 75% steatosis) for steatosis was assigned to each section of the liver (3 areas per section).

Real-time quantitative RT-PCR
Livers in 350 µL of RA1 and 3.5 µL of β-mercaptoethanol were mixed and frozen at -80 °C until RNA extraction. RNA extraction was carried out using the nucleospin RNA kit (Macherey-Nagel). The RNAs were quantified using a Nanodrop and the retro-transcription made using the high-capacity cDNA reverse transcription kit (Thermo Fisher