The crystal structure of a multidomain protease inhibitor (HAI-1) reveals the mechanism of its auto-inhibition

Abstract

Hepatocyte growth factor activator inhibitor 1 (HAI-1) is a membrane-bound multidomain protein essential to the integrity of the basement membrane during placental development and is also important in maintaining postnatal homeostasis in many tissues. HAI-1 is a Kunitz-type serine protease inhibitor, and soluble fragments of HAI-1 with variable lengths have been identified in vivo. The full-length extracellular portion of HAI-1 (sHAI-1) shows weaker inhibitory activity toward target proteases than the smaller fragments, suggesting auto-inhibition of HAI-1. However, this possible regulatory mechanism has not yet been evaluated. Here, we solved the crystal structure of sHAI-1 and determined the solution structure by small-angle X-ray scattering. These structural analyses revealed that, despite the presence of long linkers, sHAI-1 exists in a compact conformation in which sHAI-1 active sites in Kunitz domain 1 are sterically blocked by neighboring structural elements. We also found that in the presence of target proteases, sHAI-1 adopts an extended conformation that disables the auto-inhibition effect. Our results also reveal the roles of non-inhibitory domains of this multidomain protein and explain the low activity of the full-length protein. The structural insights gained here improve our understanding of the regulation of HAI-1 inhibitory activities and point to new approaches for better controlling these activities.

Introduction

Multidomain proteins usually consist of several well-folded domains connected by linker regions (1–3). Such proteins can adopt extended tertiary structures with high interdomain flexibility and dynamics. They can also form a compact structure supported by interdomain interactions, and one example for the latter case is the structure of plasminogen, the zymogen precursor of the primary fibrinolytic protease plasmin (4), in which the seven domains wrapped up into compact tertiary structure. It is possible that these two types of tertiary structures can be interconvertible, and both interdomain interaction and linkers are likely critical determinants of the structure type and the relative orientation of individual domains (4). Interdomain linkers tend to have high variability in both length (typically from 2 to 18 amino acid residues) and sequences (5), as compared with the domains that are often more constricted in these parameters because of requirements of conservation of fold and function. Short linkers are more likely lead to a rigid conformation of multidomain protein. However, the relationship between linkers and the conformational dynamics of multidomain proteins is highly empirical and not well-understood.

Hepatocyte growth factor activator inhibitor 1 (HAI-1)³ is a membrane-bound multidomain protein containing an extracellular region that consist of a MANEC (motif at N terminus with eight cysteines) PAN/apple-like domain (MANEC or N) (6), an internal PKD-like domain (I), a Kunitz domain 1 (K1), a low-density lipoprotein receptor class A (LDLRA) domain (L), and a Kunitz domain 2 (K2) followed by a C-terminal single-span transmembrane region (7). HAI-1 has been shown to be essential to the integrity of the basement membrane during placental development (8), because HAI-1-deficient mice died in utero due to placental defects. HAI-1 is also important in maintaining postnatal tissue homeostasis including keratinization of the epidermis (9), hair development (10), colonic epithelium integrity (11, 12), proliferation and cell fate of neural progenitor cells (13), and tissue injury and repair (14, 15).

HAI-1 has been demonstrated to inhibit a number of serine proteases, including matriptase, HGFA (16, 17), hepsin (18), trypsin (19), and prostasin (20). The interaction between HAI-1 and matriptase is critical for tissue morphogenesis and cellular biology. An imbalance in the HAI-1:matriptase ratio results in tumorigenesis (21, 22). For instance, a mild overexpression of matriptase relative to HAI-1 results in spontaneous squamous cell carcinoma, a phenotype that can be effectively reversed back to wild type by additional expression of HAI-1 (22). It was reported that embryonic death in mice caused by gene deletion of HAI-1 can be rescued by simultaneous deletion of the matriptase gene (23). Hence, HAI-1 is thought to have tight functional relationship with matriptase to keep a balanced level of enzyme activity to maintain homeostasis.

At present, no structural information is available for full-length form of HAI-1 or its similar multidomain target proteases. Previous studies on HAI-1 focused on the HAI-1-MANEC domain (6), the HAI-1-IK1 fragment (15), and the HAI-1-K1 domain in complex with either matriptase (serine protease domain (SPD)) (24) or HGFA (serine protease domain) (25), revealing some of the structural condition of the primary inhibitory interactions of HAI-1 with protease domains. Many questions remain on this multidomain protein: how HAI-1 arranges its domains in respect to tertiary structure; what is the relationship between the overall conformation/tertiary structure and function of HAI-1? Also, how does HAI-1 facilitate matriptase activation and inhibition of enzymatic activity?

In this study, we report the crystal structure of the full-length extracellular domain of HAI-1 (sHAI-1). The structure showed that sHAI-1 has a compact conformation mediated by the MANEC domain and a long but well-ordered linker (residues 354–374, named 365-linker). The mechanism of sHAI-1 auto-inhibition was due to sterically blockage on the active site in Kunitz domain 1. The compact conformation was further confirmed to exist in solution by small-angle X-ray scattering (SAXS). The SAXS also showed that the presence of the target protease caused a dramatic conformational change of the tertiary structure of sHAI-1 to extended conformation. Thus, this study illustrated that the compact conformation of sHAI-1 is not suitable for interacting with target proteases, and conformational transition to the open form is required to inhibit the target proteases. In addition, the long linker does not necessarily lead to an extended structure of a multidomain protein (sHAI-1) but also acts as an adhesive to help HAI-1 adopting an auto-inhibitory conformation.

Results

Auto-inhibition of HAI-1 activity through domain interplay

To determine the contribution of individual domains of HAI-1 for the inhibition of matriptase, we produced the recombinant proteins representing the extracellular part of human HAI-1 with different domain truncations, including the NIK1LK2 (sHAI-1), NIK1L, IK1L, and K1 fragments (Fig. 1A). These recombinant soluble proteins were expressed either in Drosophila S2 cells (NIK1LK2, NIK1L, and IK1L) or in Pichia pastoris X-33 cells (for K1). Their inhibitory activities toward matriptase (serine protease domain) were analyzed using a chromogenic substrate-assisted enzymatic assay. The NIK1LK2 construct, including all extracellular domains of HAI-1, is referred to as sHAI-1.

Figure 1.

Auto-inhibition of HAI-1, shown by the lower inhibition of sHAI-1 compared with other HAI-1 variants against matriptase. A, schematic domain construction of HAI-1 variants. The primary translation product of HAI-1 is described as HAI-1, and variants containing different domain constructs are named sHAI-1, NIK1L, IK1L, and K1. Domains are shown as rectangles and labeled as follows: N, MANEC domain; I, internal domain; K1, Kunitz domain 1; L, LDLRA domain; and K2, Kunitz domain 2. TM is short for the transmembrane region. The residue numbers of the N- and C-terminal regions of these variants and the glycosylation sites are also labeled accordingly. B, results of inhibition assays, demonstrating that the MANEC domain and Kunitz domain 2 blocked matriptase binding, whereas the internal domain and the LDLRA domain stimulated the binding affinity.

The sHAI-1 was 2-fold less potent than the HAI-1 variant lacking the Kunitz domain 2 (NIK1L) based on results of the half-maximal inhibitory concentration (IC₅₀) measurements (Fig. 1B), suggesting that the Kunitz domain 2 somehow restrains the inhibitory activity of sHAI-1. On the other hand, the NIK1L construct was again 690-fold less potent than a construct without the MANEC domain (IK1L), demonstrating that the MANEC domain decreases the inhibitory activity of sHAI-1 even more. The inhibitory capability against matriptase of sHAI-1 was 1060-fold weaker than that of the IK1L fragment, together supporting the possibility that the surrounding domains (both MANEC domain and Kunitz domain 2) have a synergistic effect in regulation of sHAI-1 inhibitory activity.

The Kunitz domain 1, the major inhibitory domain of HAI-1, showed 6- to 7-fold less potent inhibitory capability than IK1L (Fig. 1B), consistent with the previous report that the internal and LDLRA domains stimulate Kunitz domain 1 inhibition of matriptase (15). The results obtained here are consistent with a previous study of Kojima et al. using rat HAI-1 and the full-length rat matriptase (26). They showed that the deletion of the MANEC domain and the Kunitz domain 2 of rat HAI-1 enhanced the inhibitory activity and that the deletion of internal domain and LDLRA domain weakened the inhibitory activity.

Crystallization and structure determination of sHAI-1

To access the roles of sHAI-1 domains in the auto-inhibition of HAI-1 activities, the crystal structure of sHAI-1 was determined. As a multidomain protein, the long linkers between domains likely increase the flexibility of sHAI-1, posing a challenge on its crystallization ability (24). We screened more than 1000 crystallization conditions and finally obtained preliminary crystals of sHAI-1. Unfortunately, these crystals diffracted only to 7–8 Å from synchrotron X-ray sources. The crystal was confirmed to contain only sHAI-1 with no sign of degradation by SDS-PAGE analysis of washed crystals (Fig. 2A). Following intensive efforts to improve the crystal diffraction including screening of cryoprotectant and pH conditions and soaking with different additives (such as calcium), a data set of a moderate resolution (3.8 Å) was obtained. The structure was then determined by molecular replacement, and 80% of the primary sequence (324 of 406 residues) was built into the final structure. The final model was refined to an R_work/R_free of 0.25/0.33 (Table 1) with reasonable geometry. The MANEC domain, internal domain, Kunitz domain 1, and Kunitz domain 2 were well-defined, where the LDLRA domain was untraceable in the electron density maps and not included in the final model.

Figure 2.

Reliability of sHAI-1 crystal structure. A, SDS-PAGE analysis confirmed the presence of the intact sHAI-1 in crystals. B, 2F_o − F_c electron density of linkers regions. The model of sHAI-1 (sticks) is well-fitted into the electron density map (gray) contoured at a level of 1.0 σ.

Table 1.

X-ray data collection and refinement statistics of sHAI-1

Data collection and scaling
Beamline	BL18U1
Space group	P41212
Cell dimensions
a, b, c (Å)	95.426, 95.426, 124.495
α, β, γ (°)	90, 90, 90
Rmerge (%)	15.37 (92.2)
Rp.i.m. (%)	3.1 (18.4)
Rmeas (%)	15.7 (95.3)
I/σ(I)	17.47 (4.45)
Completeness (%)	97.20 (85.37)
Redundancy	25.2 (27.6)
Refinement
Resolution (Å)	38.09–3.82 (3.98–3.82)
Rwork	0.25 (0.31)
Rfree	0.33 (0.26)
Protein residues total/build	406/320
Average B factor (Å2)	170.90
Validation
Ramachandran favored (%)	73
Ramachandran allowed (%)	20
Ramachandran outliers (%)	7
Root mean square deviations from ideal
Bond lengths (Å)	0.005
Bond angles (°)	1.25

Despite the limited resolution, the current structure was highly reliable. The collected X-ray data had good quality with high redundancy (25.2) and strong signal (I/σ of 17.47; Table 1). The molecular replacement solutions were robust with a translation function score of 10.4 (excess 8.0 usually indicates a definitely correct solution) (27) and a log-likelihood gain of 379.4. In addition, most modeled sHAI-1 residues were represented well in the final electron density map, including the long linker connecting to the Kunitz domain 2 (Fig. 2B).

Crystal structure of sHAI-1 reveals a compact tertiary structure

In the structure of sHAI-1, the four domains (the MANEC domain, the internal domain, the Kunitz domain 1, and the Kunitz domain 2) cluster into a globule rather than an extended conformation (Fig. 3A). These four observed domains are arranged in a tetrahedral manner with the internal domain, the Kunitz domain 1 and the Kunitz domain 2 forming a pocket into which the MANEC domain inserts (Fig. 3B). The MANEC domain plays a critical role in maintaining the compact conformation of sHAI-1. Electrostatic potential analysis of the internal domain and the Kunitz domain 1 reveals that the positively charged concave surface it creates matches favorably with the negatively charged surface of the MANEC domain (Fig. 3C). Similarly, the negatively charged surface of the Kunitz domain 2 is preferred by the positively charged surface of the MANEC domain opposite to the Kunitz domain 2 (Fig. 3C). Three consecutive residues (Tyr¹²³-Glu¹²⁴-Gln¹²⁵) of the MANEC domain insert into the gap between the internal domain and the Kunitz domain 1 and form a hydrogen bonding network with the internal domain (Asp¹⁷⁰, Lys¹⁷², and Thr²³⁷) and the Kunitz domain 1 (Leu²⁹¹ and Glu²⁹⁵; Fig. 3, D and E). Additionally, helix 3 and β strand 5 of the MANEC domain tether the Kunitz domain 2 to the core of sHAI-1 through polar interactions (Fig. 3F).

Figure 3.

Crystal structure of sHAI-1 and interdomain interaction. A, overall compact arrangement of sHAI-1. The N (salmon), I (blue), K1 (yellow), 365-linker (forest green), and K2 domains (purple) were observed and packed closely together. Glycans at Asn⁶⁶ (green) and the P1 site (Arg²⁶⁰, yellow) are shown as sticks and spheres. The missing LDLRA domain and linkers are drawn with a gray dashed line. B, MANEC domain embedded into the concave (right) formed by the rest part of sHAI-1 (left, surface in gray). C, electrostatic potential analysis of sHAI-1 revealed opposite charge existing in the contact area between MANEC domain and IK1-K2. B and C are rotated clockwise along y axis for 60° compared with A. D, the hydrogen bonding network in E was supported by the well-defined electron density map (2F_o − F_c, gray) at contoured level of 1.0 σ. E and F, hydrophilic interaction between MANEC domain and the rest of sHAI-1. G, hydrophobic interactions between 365-linker and MANEC, and Kunitz domain 1. H, negative charged 365-linker was attracted by the positively charged P1 site loop.

The internal domain and Kunitz domain 1 are arranged in a V-shape (Fig. 3A), quite similar to the previously published crystal structure of the IK1 fragment (PDB code 5ezd) (15), despite the difference in length of protein constructs and crystal packing environments. This comparison demonstrates that the IK1 has a relatively rigid interdomain interface and thus is also critical for maintaining the compact conformation of sHAI-1. Kunitz domain 2 appears to be loosely attached to the globule core of sHAI-1 with 4.2% of its surface contacting with the neighboring domains. The LDLRA domain is even more labile to an extent that it is completely disordered and could not be modeled in the current structure.

An interesting feature of the new structure is that a long linker (365-linker, residues 354–374) between the LDLRA domain and the Kunitz domain 2 is well defined by its strong electron density (Fig. 2B). The 365-linker wedges between the MANEC domain and Kunitz domain 1 and tethers Kunitz domain 2 onto the protein core. This long linker appears to be crucial for maintaining the compact conformation of sHAI-1. It “glues” the MANEC domain and Kunitz domain 1 together via hydrophobic interactions. Leu³⁶³ and Ile³⁶⁶ of the 365-linker together with Leu²⁸⁴ of Kunitz domain 1 cover the hydrophobic patch of MANEC domain (formed by Phe⁷⁵, Leu¹⁰¹, Phe¹¹⁷, Ile¹¹⁹, and Phe¹³⁷) (Fig. 3G). Furthermore, most residues of the 365-linker close to the positively charged P1 site of the Kunitz domain 1 (residue Arg²⁶⁰) are negatively charged (Fig. 3H).

Structural basis of HAI-1 auto-inhibition

Using the crystallographic approach, we found sHAI-1 adopting a compact conformation. In this conformation, the key inhibitory residue (Arg²⁶⁰, P1) of the Kunitz domain 1 was not fully solvent-exposed. Superposition of the HAI-1-K1·matriptase-SPD complex (24) structure to sHAI-1 by the Kunitz domain 1 alignment illustrates that the access of the target proteases is impeded because of steric hindrance from the neighboring HAI-1 domains. Such steric hindrance stems from three structural elements.

First, the MANEC domain is positioned where it would overlap extensively with matriptase protease domain (Fig. 4A), thus limiting the inhibitory activity of sHAI-1 in its compact conformation. Three glycosylation sites of HAI-1 (Asn⁶⁶, Asn²³⁵, and Asn⁵²³) were predicted and located in MANEC domain, internal domain, and the cytoplasmic tail of HAI-1, respectively (Fig. 1A). An N-linked glycan moiety in the MANEC domain (Asn⁶⁶) was clearly visible in the current structure. This N-linked glycan is close to the P1 site with a distance of 8.1 Å (Fig. 4B). In humans, N-linked glycans can be quite extensive and large molecules (28), suggesting that the glycan can block an additionally larger surrounding area from protease accessibility. We observed that the deglycosylation of Asn⁶⁶ indeed led to stronger inhibitory activities (Fig. 4, E–G, and Table 2).

Figure 4.

Auto-inhibitory mechanism of sHAI-1. A, superposition of crystal structure of sHAI-1 with the HAI-1-K1·matriptase-SPD (PDB code 4isl) complex shows serious clashes between the MANEC domain, the 365-linker, and Kunitz domain 2 of HAI-1 and matriptase-SPD. B, the 365-linker is quite close to the active site (P1 site, Arg²⁶⁰) with a shortest distance of 8.4 Å. The glycan at Asn⁶⁶ is 8.1 Å away from P1 site and can block the access of the P1 site because of the high flexibility of glycan (the light blue sector indicates the inaccessible volume because of the glycan). C, structural clashes between Kunitz domain 2 and matriptase-SPD. D, superimposition of the crystal structure of sHAI-1 with the trypsin·Kunitz domain complex (PDB code 4U30) shows the clash between trypsin and the internal domain and the Kunitz domain 1 (red circle). The P1 residue (Lys³⁸⁵) is showed as spheres and colored in purple. E, deglycosylated IK1L shows no difference in the IC₅₀ values compared with the non-deglycosylated IK1L, which suggests that the potential glycosylation site on Asn²³⁵ does not affect the inhibition of matriptase. F and G, the inhibitory capability of deglycosylated sHAI-1 and NIK1L are stronger than that of the non-glycosylated counterpart, indicating that glycan at Asn⁶⁶ decreases the inhibitory activity against matriptase.

Table 2.

IC₅₀ for reaction of the serine protease domain of matriptase inhibited by HAI-1 variants

Variants	Wild type	Deglycoslysed
	nm	nm
sHAI-1	106.10	18.73
NIK1L	69.13	26.17
IK1L	0.10	0.08
K1	0.67	0.69

Second, the 365-linker is close to the P1 residue (Arg²⁶⁰) and blocks the accessibility of Arg²⁶⁰ (Fig. 4, A and B). Some residues of this linker possess large side chains, including Phe³⁶⁰, Asp³⁶¹, Tyr³⁶³, Arg³⁶⁵, His³⁶⁷, Phe³⁶⁸, and Tyr³⁷¹, and have strong interactions with the P1 loop of the Kunitz domain 1 (Figs. 2B and 3G), which imposes an additional barrier to the P1 loop insertion into the active site pocket of matriptase.

Third, the Kunitz domain 2 was also found collide slightly with matriptase (Fig. 4C) and needs to be moved away before matriptase can access to the inhibitory loop region of the Kunitz domain 1. The slightly collision observed here explains the relatively weaker effect from Kunitz domain 2 in down-regulation of the inhibitory activity of sHAI-1 against matriptase.

HAI-1 has two Kunitz-type domains: Kunitz domain 1 and Kunitz domain 2. Kunitz domain 2 was demonstrated to have potent inhibitory activity against trypsin (29), although it cannot inhibit HGFA and matriptase (24). In this compact conformation, the active site (Lys³⁸⁵, P1) of Kunitz domain 2 is also not fully exposed to the solvent and is impeded by internal domain and Kunitz domain 1 (Fig. 4D). It should be emphasized that the active P1 residues in the two Kunitz domains are not totally blocked but remain half-solvent-exposed.

X-ray structure of sHAI-1 was supported by SAXS solution structure

To identify sHAI-1 conformation in solution, we used the SAXS method to determine the overall shape of the sHAI-1. sHAI-1 showed no sign of aggregation at the concentration of 3 mg ml⁻¹ at pH 7.4. The molecular mass was estimated to be 60 kDa from the experimental SAXS data and was very close to the theoretical value of 54 kDa (Table 3). The radius of gyration (R_g) estimated from the Guinier plot was 30.4 Å, which was in excellent agreement with R_g of 31.0 Å obtained from a different plot (pair-wise distance distribution P(r) function), demonstrating the high quality of the sHAI-1 SAXS data. The pair-wise distance P(r) distribution also gave a maximal dimension (D_max) of 106 Å (Fig. 5, A and B).

Table 3.

SAXS parameters for sHAI-1 and sHAI-1·matriptase complex

The radii of gyration (R_g) in real space are estimated from the SAXS data using Guinier analysis, and the R_g of Reci is obtained from pair-wise distance distribution P(r) function. D_max, mass, and mass real are the maximum particle dimensions, molecular mass estimated from I(0), and molecular mass calculated from the sequence, respectively. The χ² value of the fit of D_max is given.

Samples	Real space		Reci		Dmax	Mass kDa		χ2
	Rg	I(0)	Rg	I(0)		SAXS	Seq
	Å		Å			kDa
HAI-1	30.4	2.96E02	31.0	3.09E02	106	60	54	1.07
sHAI-1·matriptase	45.9	1.86E02	43.0	1.93E02	158	103	82.5	1.01

Figure 5.

SAXS analysis reveals that sHAI-1 is in compact conformation in solution. A, experimental SAXS curve (open circles) of sHAI-1. B, the pairwise distance distribution P(r) derived from the scattering profile of sHAI-1. C, the model of sHAI-1 tertiary structure based on the envelope obtained from the experimental SAXS data by placing the LDLRA domain at the tail region. Meanwhile, the missing linkers are modeled with gray dashed lines. D, fit and residual plots for sHAI-1. Shown are the experimental SAXS data (black) and the theoretical scattering curve (orange) generated from the sHAI-1 model. The computed sHAI-1 model is fitted to the experimental SAXS profile with χ² value of 3.26. The fit and residual plots are for q (Å⁻¹) (x axis) versus log-intensity (y axis). E, the structure model of sHAI-1 (cartoon) and the experimental envelope were superimposed together.

The ab initio molecular model was constructed from the SAXS data. The envelope of sHAI-1 is shaped like a tadpole with a short tail, demonstrating the compact conformation of sHAI-1. The sHAI-1 crystal structure model can be fitted into the head of this molecular envelope. The LDLRA domain that was missed in the crystal structure was placed to the tail region (Fig. 5E). The theoretical scattering curve generated from such sHAI-1 model fitted quite well to the experimental curve (χ² of 3.26 using FoXs) (30) (Fig. 5D). These results indicated that solution sHAI-1 adopts a compact conformation similar to what we observed in crystals. This is quite interesting considering the loose interaction between the Kunitz domain 2 and the remaining part of sHAI-1 in the presence of the long linker (365-linker). It is difficult to explain the presence of the tail in the SAXS envelop. One possibility might be that this tail reflects the flexibility of the LDLRA domain of the sHAI-1, which can adopt a range of different conformations, leading to its disorder in the crystal structure.

sHAI-1 adopts a non-compact conformation upon matriptase binding

To study the conformation of sHAI-1 bound to target serine proteases, we prepared the molecular complex of sHAI-1 with just the SPD of matriptase (molecular mass = ∼28 kDa). The molecular mass calculated from the SAXS data was 103 kDa, which was higher than the theoretical value (82.5 kDa). The hydrodynamic radius (R_g) estimated from Guinier plot was 45.9 Å, close to the R_g obtained from pair-wise distance distribution function P(r) (43.0 Å) (Table 3). The maximal dimension of the complex was much bigger (D_max of 158 Å; Fig. 6, A and B) than the one of sHAI-1 alone (106 Å), showing that the complex adopts a more extended conformation. The ab initio SAXS model was then constructed, and the resulting molecular envelope of the sHAI-1·matriptase complex looked like an elongate rod that is wide in the middle part and narrow at its two ends. Again, this shape demonstrated that sHAI-1 adopts a much more extended conformation in the inhibitory complex.

Figure 6.

SAXS analysis of sHAI-1 in complex with matriptase-SPD (sHAI-1·SPD) reveals sHAI-1 adopts an extended conformation that was much different from sHAI-1 alone. A, experimental SAXS curve (open circles) of sHAI-1·SPD complex. B, the pair distance distribution P(r) of the complex. C, experimental envelope of the sHAI-1·SPD complex at three orthogonal views. A molecular model of the sHAI-1·SPD complex was built to fit the SAXS envelop. D, a schematic highlights the conformational transition of sHAI-1 from the compact to extended state upon matriptase-SPD binding.

No crystal structure of the sHAI-1·matriptase complex is known yet. To explain the SAXS molecular envelope of the sHAI-1·matriptase complex, we manually positioned the largest fragment of sHAI-1·matriptase complex, the HAI-1-K1·matriptase-SPD complex (24), to the middle of the envelope and placed the MANEC domain in the larger end, whereas the Kunitz domain 2 was positioned in the smaller end. The internal domain and LDLRA domain can then be positioned accordingly (Fig. 6C). Taken together, we present a model explaining the conformational transition of sHAI-1 upon binding of matriptase (Fig. 6D). In this model, the overall conformation of sHAI-1 was in an open state where the structural hindrance imposed by the MANEC domain, 365-linker, and Kunitz domain 2 has been removed, whereas the internal domain and Kunitz domain 1 were still organized in a V-shape and clamped the 60 loop of matriptase (15), and the Kunitz domain 1 can make tight interactions with matriptase using classical Kunitz-type standard mechanism inhibition (15).

Discussion

The working model of how HAI-1 forms complex with matriptase

Our structural studies reveal that sHAI-1 adopts a compact and auto-inhibitory conformation mediated by the MANEC domain and the 365-linker. For membrane-bound form of HAI-1, the intact MANEC domain and the 365-linker likely also maintains a compact and auto-inhibitory conformation and exhibits similar inhibitory activity toward matriptase in vivo (31).

We came up with a working model for HAI-1. HAI-1 is expressed as a membrane-bound form and exists as a compact auto-inhibitory conformation (Fig. 7A). Membrane-bound HAI-1 can shed into the extracellular milieu primary in the 58- and 40/39-kDa forms. The 58-kDa form is most likely the same as our recombinant sHAI-1 and adopts a compact conformation (31) (Fig. 7A). The 40/39-kDa form likely consists of the MANEC domain, the internal domain, and Kunitz domain 1, without the 365-linker (32). The MANEC domain of the 40/39-kDa form likely moves from active site of Kunitz domain 1 (Fig. 7A), explaining the much higher potency of the 40/39-kDa form in inhibition of its target proteases. In the presence of activated two chain matriptase, sHAI-1 adopts an extended conformation (Fig. 7B) so that its Kunitz domain 1 can bind to the SPD of target proteases. It is likely that the non-catalytic domains of matriptase can stimulate the conformational transition of HAI-1 from the compact to open state. This is supported by the fact that HAI-1-NIK1LK2 inhibited the entire extracellular domain more strongly than the SPD alone of matriptase (31). In addition to the interaction between matriptase-SPD and HAI-1 Kunitz domain 1, the CUB2 domain in matriptase also interacts with the Kunitz domain 2 of HAI-1 (34). Another previous study has shown that the matriptase LDLR4 domain was located near the SPD domain but opposite of the active site cleft of SPD (35). However, the validity of this model awaits further validation by biophysical approaches or antibody mapping.

Figure 7.

Working model of HAI-1. A, models of membrane-bound HAI-1 and its variants. Left, the models of sHAI-1 of the 58- and 40/39-kDa forms. Right, the membrane-bound model of HAI-1. B, membrane-bound form HAI-1 in complex with activated matriptase in two chain form. Matriptase is a cell surface glycoprotein type II transmembrane serine protease, consisting of an N-terminal cytoplasmic domain, a signal-anchor transmembrane domain, a non-catalytic stem domain (containing a SEA domain, two CUB domains, and four LDLRA domains), and a C-terminal SPD. Matriptase is synthesized as a zymogen and is completely activated by cleavage at Gly¹⁴⁹–Ser¹⁵⁰ and Arg⁶¹⁴–Val⁶¹⁵. HAI-1 inhibited the activated matriptase with its Kunitz domain 1 forming a tight non-covalent complex with SPD. Additional interactions are found between the CUB2 domain and Kunitz domain 2. SEA, sea urchin sperm protein, enteropeptidase, agrin; CUB, complement C1r/C1s-urchi embryonic growth factor, bone morphogenetic protein 1.

The role of linkers in multidomain proteins

Linkers can serve as a covalent connection between the domains of a multidomain protein. In addition, linkers can also affect the function of the multidomain protein in various ways (36). In HAI-1, there are four linkers between the five domains, including the 160-linker (residues 155–165), the 245-linker (residues 241–249), the 310-linker (residues 301–318), and the 365-linker. These linkers are classified into three classes according to their length as short linkers (2–5 residues), medium-sized linkers (6–15 residues), and long linkers (more than 16 residues) (5). The 245-linker and the 365-linker are well-defined in the electron density maps in the sHAI-1 structure, whereas the electron density of the 160-linker and the 310-linker were invisible, and they cannot be modeled.

The 365-linker connecting the LDLRA and the Kunitz domain 2 is quite long (20 residues), and this long linker appears to be a major contributor to the compact conformation of sHAI-1. The long 365-linker tethers spatially separated LDLRA domain and Kunitz domain 2 onto the core of sHAI-1. Furthermore, we observe that the 365-linker appears to sterically obstruct the target proteases binding to the active P1 site. This observation is consistent with a previous report where deletion of the 365-linker enhanced the inhibitory activity of HAI-1 markedly (26). The critical role of this linker is further supported by its high sequence conservation among known mammalian species (Fig. 8). Sequence conservation is not a typical phenomenon of linkers in multidomain proteins. In general, the 365-linker acts as a structural scaffold and aids in LDLRA and Kunitz domain 2 looping around Kunitz domain 1 to maintain the compact conformation and regulate the inhibitory activity of sHAI-1.

Figure 8.

Sequence alignment of the 365-linker among different mammal lines. Conserved residues of the 365-linker among mammal lines are highlighted by color coding (white).

The 245-linker connects the internal domain to Kunitz domain 1. Unlike the typical random coil conformation of an interdomain linker, this linker is observed in α-helix conformation in accordance with a well-defined electron density. This linker conformation helps to maintain a more fixed orientation between the internal domain and Kunitz domain 1 as observed in the current study and a previous study (15).

Unlike the 245- and 365-linkers, both the 160- and 310-linkers were disordered, reflecting their flexible propensities. The flexibility of the 160-linker was also shown by a ∼15° rotation of helix 3 (residues from 143 to 153), which is next to the 160 linker in the sequence, when compared with the crystal structure of sHAI-1 and the NMR structure of the HAI-1-MANEC domain. In summary, the linkers are not only important to maintain the compact conformation of sHAI-1 but also appear to participate in the overall inhibitory activity regulation. This is also an example showing that longer linkers (e.g. the 365-linker) do not necessarily lead to higher conformational freedom of the tertiary structure.

A new mode of Kunitz domain association in a multi-Kunitz domain protein

A unique feature of the current HAI-1 conformation is that the two Kunitz-type domains are separated by a domain in sequence but are spatially closed to each other. The reactive site of Kunitz domain 1 and Kunitz domain 2 are both blocked by neighboring structural elements, which explains why the 58-kDa fragment of HAI-1 (NIK1LK2) is not the most potent inhibitor of serine proteases in vivo. This is different from other inhibitory proteins that contain multiple Kunitz-type domains, where the Kunitz-type domains are linked sequentially and at least one of its Kunitz domains is not interrupted by other domains. Examples of these multi-Kunitz-type domain proteins include 1) tissue factor pathway inhibitor with three Kunitz domains, and the Kunitz domains in the inhibitors are not interrupted by another domain (7, 38); 2) bikunin, composed of two Kunitz-type domains packed close together with the protease binding site in Kunitz domain 1 exposing to solution, but the protease binding site in the Kunitz domain 2 impeded by Kunitz domain 1 (39); and 3) HAI-2, an inhibitor with high homology to HAI-1 but contained only Kunitz domain 1 and Kunitz domain 2, was reported to inhibit HGFA with Kunitz domain 1 as the functional domain in human, but in mouse the truncated HAI-2 contained only Kunitz domain 2 was also an efficient HGFA inhibitor. Whether the discrepancy results from different domain arrangement of these two Kunitz domains in human and mouse is still unknown (40–43). It appears that a special arrangement of Kunitz domains represents a new level of inhibitory activity regulation for multi-Kunitz-type proteins, in addition to the Kunitz domain sequences.

A new way to intervene with matriptase-mediated cancer formation

A lower level of HAI-1 in tumors has been shown to be associated with an overall poor prognosis in several cancers (44). Additionally, matriptase is consistently overexpressed in a wide variety of human tumors of epithelial origin, and high matriptase levels are correlated with poor clinical outcome. The structural insight gained from the current study and the structure model of sHAI-1 in complex with matriptase-SPD provides a new way to intervene with HAI-1 inhibitory activity. The compact conformation of sHAI-1 disfavors its inhibition of target proteases. Reducing the auto-inhibition may be one way to enhance HAI-1 inhibitory activity. This can be achieved by perturbing the interaction between the 365-linker and Kunitz domain 1 and the MANEC domain.

Experimental procedures

Recombinant protein expression and purification

sHAI-1 (Gly³⁶–Pro⁴⁴¹) and Kunitz domain 1 was expressed in Drosophila S2 cells and X-33 cells, respectively. Protein purification of sHAI-1 and Kunitz domain 1 was done as described previously (45). Cloning, expression, and purification of HAI-1-NIK1L (Gly³⁶–Ser³⁷⁰) was according to our previous method (24). To express HAI-1-IK1L (Thr¹⁵⁴–Ser³⁷⁰), the cDNA was generated by PCR using the full-length cDNA of HAI-1 isoform 2 and cloned into the expression vector pMT/Bip/V5-His-A (Invitrogen). Drosophila expression system (S2 cells) was utilized to generate recombinant HAI-1-IK1L as secreted protein. Protein expression and purification was did the same as sHAI-1. Expression, purification, and refolding of the SPD of matriptase basically followed our previous method (15). The complex of sHAI-1·matriptase was formed by mixing sHAI-1 and matriptase-SPD at a molar ratio of 1:1.5 in 150 mm NaCl and 50 mm Tris·HCl, pH 7.4, for half an hour. The mixture was then concentrated and purified by size-exclusion chromatography (Superdex 200). The complex fraction was collected and concentrated to ∼10 mg ml⁻¹ for further usage.

Determination of the half-maximal inhibitory concentrations (IC₅₀)

IC₅₀ values of sHAI-1 were measured in the reaction buffer (150 mm NaCl, 20 mm Tris·HCl, pH 7.4, and 0.5/1000 Tween 20). The initial concentration of sHAI-1 was 6 mg ml⁻¹ and was serially diluted at a ratio of 1:3 using the reaction buffer, and then 10 μl of each diluted sample was incubated with matriptase at the room temperature for 10 min in a 96-well polysterene plate. The proteolytic reactions were initiated by adding 40 μm Pefachrome® tPA (tissue plasminogen activator) to the wells to final volumes of 100 μl. The kinetics of p-nitroanilide release was monitored spectrophotometrically at 405 nm for 10 min. Control treatments without sHAI-1 were performed to ensure the activity of matriptase, and treatments in the absence of both sHAI-1 and matriptase were performed to exclude any effects of the reaction buffer. The IC₅₀ values of NIK1L, IK1L, and Kunitz domain 1 were determined in the same way as for sHAI-1. Deglycosylation of HAI-1 (sHAI-1, NIK1L, and IK1L) was performed by adding a trace amount of deglycosylation enzyme (peptide-N-glycosidase F) into protein samples and incubating at 37 °C for ∼2 h. IC₅₀ measurements of deglycosylated HAI-1 were completed in the same way as sHAI-1.

SAXS data collection and structural modeling

To study the conformation of sHAI-1 in solution, SAXS data were collected on the BL19U2 Beamline at Shanghai Synchrotron Radiation Facility. The energy range of this beamline is 7–15 keV, and the focus spot size (μm²@12keV) is 320 × 43 (H × V) with a beam divergence angle of 46 × 27 (H × V). The samples were placed in reusable quartz capillaries at 283 K. All data sets were measured with an exposure time of 1 s. Three concentrations (3, 6, and 9 mg ml⁻¹) protein in 150 mm NaCl and 20 mm Tris·HCl, pH 7.4, were prepared and analyzed to ensure a possible monomer state. The SAXS data of sHAI-1 were collected between every two buffers. Averaging of the scattering data and background buffer subtraction were performed using bioSAXS-RAW software (46). The qualities of the scattering curves were analyzed using the program PRIMUS (47) to ensure there was no obvious aggregation or radiation of the sample. The forward scattering I(0) and radius of gyration (R_g) were determined by Guinier analysis, assuming that they were at very small angles (s ≤ 1.3/R_g). The indirect Fourier transform method implemented in the program ScÅtter was used to estimate the pair-wise distance distribution function P(r) and the maximum particle dimensions, D_max (48). Molecular mass was calculated from the SAXSMow2 program (49). Ab initio modeling was performed using DAMMIF program (50). At least 10 calculated DAMMIF solutions were performed, aligned, and compared. Subsequently, meaningful shapes were picked up and averaged by DAMAVER program (51). The SAXS data of the complex of sHAI-1·matriptase was also collected and analyzed in the same way as sHAI-1.

Crystallization and data collection of sHAI-1

Using sitting drop vapor diffusion, reproducible crystals were finally obtained in 1.6 m ammonium sulfate, 20% glycerol, and 0.08 m sodium acetate, pH 4.6. These crystals were dipped into a precipitant solution of 1.8 m ammonium sulfate, 0.08 m sodium acetate, pH 4.6, with 25% glycerol and then frozen into liquid nitrogen quickly for X-ray data collection. The diffraction data were collected on the BL18U1 Beamline at Shanghai Synchrotron Radiation Facility.

Structure determination and refinement

The diffraction data were indexed, integrated, scaled, and merged using the xia2 automated data processing pipeline (52), which utilizes XDS (53) and the CCP4 suite (54). The crystal structure was solved by Phaser-MR of Phenix (27), using the published crystal structures of the HAI-1-IK1 fragment (PDB code 5ezd) (15) and the NMR structure of the HAI-1-MANEC domain (PDB code 2msx) (6) as search models. Homology modeling of the LDLRA domain was performed by SWISS-MODEL using the LDLRA domain of low-density lipoprotein receptor as template (PDB code 1f8z) (55). Kunitz domain 2 was modeled by the program Sculptor using the structure of the Kunitz domain 1 (PDB code 4isl) as template (24). Determination of the LDLRA domain and Kunitz domain 2 was also proceeded by MR. Subsequent model building was performed in COOT (56) and refined using Phenix.refine (37, 57). The atomic coordinates and structure factors have been deposited in the PDB under the accession code 5H7V. All graphic representations were prepared using PyMOL.

Author contributions

M. L. crystallized the sHAI-1 and carried out other studies including SAXS. C. Y. generated the recombinant proteins and solved the X-ray structure. B. Z., L. J., and Y. J. participated in various experiments including X-ray data collection. M. L., C. Y., and M. H. analyzed the data. C. Y. and M. H. conceived and designed the study. M. H., M. L., C. Y., and J. K. J. finalized the manuscript.