Directed structural modification of Clostridium perfringens enterotoxin to enhance binding to claudin-5

Abstract

Clostridium perfringens enterotoxin (CPE) binds to distinct claudins (Clds), which regulate paracellular barrier functions in endo- and epithelia. The C-terminal domain (cCPE) has the potential for selective claudin modulation, since it only binds to a subset of claudins, e.g., Cld3 and Cld4 (cCPE receptors). Cld5 (non-CPE receptor) is a main constituent in tight junctions (TJ) of the blood-brain barrier. We aimed to reveal claudin recognition mechanisms of cCPE and to create a basis for a Cld5-binder. By utilizing structure-based interaction models, mutagenesis and assays of cCPE-binding to the TJ-free cell line HEK293, transfected with human Cld1 and murine Cld5, we showed how cCPE-binding to Cld1 and Cld5 is prevented by two residues in extracellular loop 2 of Cld1 (Asn¹⁵⁰ and Thr¹⁵³) and Cld5 (Asp¹⁴⁹ and Thr¹⁵¹). Binding to Cld5 is especially attenuated by the lack of a bulky hydrophobic residue like leucine at position 151. By downsizing the binding pocket and compensating for the lack of this leucine residue, we created a novel cCPE-variant; cCPE_Y306W/S313H binds Cld5 with nanomolar affinity (K _d 33 ± 10 nM). Finally, the effective binding to endogenously Cld5-expressing blood-brain barrier model cells (murine microvascular endothelial cEND cell line) suggests cCPE_Y306W/S313H as basis for Cld5-specific modulation to improve paracellular drug delivery, or to target claudin overexpressing tumors.

Electronic supplementary material

The online version of this article (doi:10.1007/s00018-014-1761-6) contains supplementary material, which is available to authorized users.

Introduction

Claudins which form the backbone of tight junctions (TJ) are expressed in a tissue-specific manner and regulate paracellular permeability in endo- and epithelia [1, 2]. The mammalian claudin family comprises at least 26 different subtypes, which are tetraspan transmembrane domain proteins. Most, if not all, claudin subtypes form polymers by both cis-interactions (with claudins within the same cell membrane) and trans-interactions (with claudins of the neighboring cell membrane). Just recently a crystal structure of claudin-15 (Cld15) comprising the four transmembrane helices and the extra- and intracellular loops (PDB ID:4P79, [3]) was released, nevertheless the molecular mechanisms of claudin–claudin interactions are still unclear. However, nature has provided us with a protein that can be used as a molecular tool. Clostridium perfringens enterotoxin (CPE) binds to claudin-3 (Cld3) and claudin-4 (Cld4) [4], which were first reported as CPE receptors [5]. CPE can be used to elucidate mechanistic details of claudin’s extracellular interactions and to target claudins for modulation of the barrier function. CPE also binds to several similar members of the claudin family—the “CPE receptors” [6], but not to other claudins such as Cld2 and Cld5 (“non-CPE receptors”). Cld1 was initially reported to be a non-CPE receptor [7], but it was later shown that CPE does interact with Cld1, albeit weakly [8, 9]. The crystal structure of CPE comprises three domains [10, 11]. The C-terminal domain of CPE contains the claudin-binding site [12], which consists of a deep and strongly hydrophobic pit formed by three tyrosines (Tyr³⁰⁶, Tyr³¹⁰ and Tyr³¹²; triple-Tyr pit) and a shallower and less hydrophobic/non-polar pit framed by three leucines (Leu²²⁵; Leu²⁵⁴; Leu³¹⁵) [9]. The two N-terminal domains of CPE cause cytotoxicity by mediating pore formation in the plasma membrane of the host mucosa cells, leading to cell death [13, 14]; whereas the C-terminal domain of CPE (cCPE) on its own is non-cytotoxic [15, 16]. However, since cCPE binds to extracellular loop 2 (ECL2) of distinct claudins [6], it increases paracellular permeability for ions and solutes by affecting the TJ reversibly [7] and was thus suggested to be a promising selective modulator of claudins [15, 16]. cCPE enhances drug absorption in the rat jejunum [17] and was shown to be effective in mucosal vaccination [18, 19]. Furthermore, since tumor proliferation and growth are associated with deregulation of claudin subtype expression and function [20], CPE constructs can be used to target tumors that overexpress claudin [21–24] and labeled cCPE was successfully used to detect Cld4-overexpressing cancer cells in vivo [25]. Unlike other drug absorption enhancers like chitosan and sodium caprate [26–28], cCPE acts specifically on its receptors; however, the determinants of this specificity are still unclear. Moreover, claudin subtype-specific binders are not yet available. We and others demonstrated that the claudin-binding properties of cCPE can be selectively changed by multiple mutations in cCPE [9, 29]. After we showed that Cld3 and Cld4 interact with cCPE by different mode of actions, we were able to create cCPE-variants binding predominantly either to Cld3 or Cld4. In addition, we showed that the contact between a bulky hydrophobic residue (Leu¹⁵⁰ in Cld3 and Leu/Met¹⁵¹ in Cld4) and the triple-Tyr pit of CPE is one of the key interactions in cCPE–claudin binding [9].

In this study, we analyzed the cCPE–claudin interaction in more detail (I) to understand the binding mechanism and (II) to lay a foundation for the development of cCPE as a claudin subtype-specific biological for tissue-specific TJ- and claudin modulation. We aimed to identify the molecular features that underlie the strong interaction between cCPE and Cld3 and Cld4, the weak interaction with Cld1 and the lack of an interaction with Cld5. Cld5 is a non-CPE receptor. However, since it is responsible for sealing the blood–brain barrier, our main aim was to develop a cCPE-variant with enhanced specificity for Cld5, as this might allow the modulation of even a non-CPE receptor claudin. Using a rational molecular and structure-based approach, we studied the mechanisms of interaction between cCPE and the ECL2 of Cld1, Cld3 and Cld5 at the level of individual residues. With this approach, it was possible to deduce information about the structural properties of the ECL2 of these claudins. On the basis of these molecular insights, we were able to change individual amino acid residues in cCPE in such a way that a cCPE-variant was produced with nanomolar affinity to Cld5. We proved the concept that directed modifications of cCPE allow a targeted interaction with distinct claudin subtypes, even with non-CPE receptor claudins. These findings open the way to specific pharmacological interventions to improve tissue-specific paracellular drug delivery or tumor treatment.

Materials and methods

Plasmids

For the construction of plasmid encoding GST-CPE_194-319 (GST-cCPE) fusion protein, cDNA of CPE (kindly provided by Dr. Y. Horiguchi, Osaka, Japan) was amplified by PCR and cloned into pGEX-4T1 (GE Healthcare, Munich, Germany), using EcoRI and SalI [30]. Plasmids encoding GST-CPE_194-319 with single or multiple mutations (L238F, I258 K, V259Y, A302 M, A302Y, S305P, Y306A Y306 W, S307R, Y310H, Y310K, S313H, L315A) were generated by site-directed mutagenesis of pGEX-4T1-CPE_194-319, as described earlier [30]. Plasmids based on pEYFP-N1 encoding murine Cld3_WT-YFP [30], murine Cld5_WT-YFP [31], Cld5_R145Q, Cld5_E146A-YFP [32], human Cld1_WT-YFP [33], Cld5_D149N, Cld5_T151L, Cld5_D149N/T151L-YFP, ChF, ChF_D148N/T150L-YFP and ChG-YFP [34] have been described previously. Cld1_D150N, Cld1_T153V, Cld1_D150N/T153V-YFP and ChG_D148N/T150L-YFP were generated by site-directed mutagenesis of Cld1_WT-YFP or ChG-YFP, respectively. pEGFP-Cld4 encoding human Cld4 with N-terminal GFP was kindly provided by Dr. W. Hunziker (Singapore). Murine Cld6, −7, −8 and −9 were obtained from OriGene Technologies, Inc. (Rockville, MD, USA) and plasmids originating from pEYFP-N1 encoding Cld6, −7, −8 and −9-YFP were generated by restriction-free cloning [35].

Antibodies

Phycolink^® anti-GST R-phycoerythrin-conjugated antibodies (Europa Bioproducts Ltd, Cambridge, UK), goat HRP-anti-mouse (Invitrogen, Carlsbad, USA), mouse anti-GST (Sigma-Aldrich, Steinheim, Germany) and mouse anti-GFP/YFP (Clonetech Mountain View, USA).

Expression and purification of cCPE-constructs

CPE_194-319 with N-terminal GST fusions as well as GST (control) was expressed in E. coli BL21. Bacteria were grown to A600 = 0.6–0.8, expression induced by addition of 1 mM isopropyl-β-D-thiogalactopyranoside. 3 h after induction, bacteria were harvested (10 min, 20.000×g, 4 °C) and lysed in PBS (phosphate buffered saline) with 1 % (v/v) Triton X-100, 0.1 mM PMSF, 1 mM EDTA, protease inhibitor cocktail (Sigma-Aldrich, Steinheim, Germany) and sonicated with Vibra Cell™ Model 72434 (BioBlock Scientific, Strasbourg, France) by 10 × 1 s pulses. Insoluble cell debris was removed by centrifugation (20,000×g for 1 h at 4 °C). Proteins were purified from supernatants using glutathione-agarose (Sigma-Aldrich, Steinheim, Germany) and dialyzed against PBS. Protein concentration was determined with BCA Protein Kit (Thermo-Scientific, Rockford, USA).

Labeling of GST-cCPE

Labeling of GST-cCPE with AlexaFluor^®647 was performed utilizing the AlexaKit (Invitrogen) and GST-cCPE_K257A. This mutant interacts similarly to cCPE_WT with all claudins tested so far and was used to prevent inhibition of binding to claudins by coupling the amino-reactive fluorophore to Lys²⁵⁷—which is in close vicinity to the claudin-binding pocket of cCPE. The resulting Alexa647–cCPE showed a labeling of 9 mol AlexaFluor^®647 per mol GST-cCPE_K257A.

Cell culture and transfection

HEK293 cell cultivation, transient or stable transfection, as well as selection of stable HEK293 cell lines expressing murine Cld6-, Cld7-, Cld8- or Cld9-YFP were performed as described [32].

cEND cells were maintained as described [36]. For experimental setups, cEND cells were allowed to reach 80 % confluence and incubated in serum reduced DMEM with 2 % (v/v) of dextran-coated charcoal-treated FCS (Fetal calf serum) for 24 h.

Live cell imaging

For live cell imaging, the transfected cells were transferred to 1 ml DMEM, 10 mM N-(2-hydroxyethyl)piperazine-N’(2-ethane- sulfonic acid) pH 7.5 without phenol red. For cCPE-binding, cells were incubated with 2 µg/ml Alexa647-cCPE. To study subcellular localization of claudin-variants, the plasma membrane was visualized by addition of 20 µl trypan blue, 0.05 % in PBS. Cells were examined with a LSM 510 META system, using an Axiovert 135 microscope equipped with a Plan-Neofluar 100×/1.3 objective (Zeiss, Oberkochen, Germany). YFP was investigated at λ_exc. 488/λ_em. 505–550 nm, AlexaFluor^®647 λ_exc. 633/λ_em. 650–710 nm and trypan blue λ_exc. 543/λ_em. >590 nm. The thickness of optical sections was <0.9 µm.

Cell surface biotinylation

Cell surface biotinylation was performed as described [32]. Briefly, cells were incubated for 60 min at 4 °C with 0.4 mg/ml EZ-link-NHS-SS-biotin (Pierce, Rockford, IL, USA) in PBS with Ca²⁺/Mg²⁺. 50 mM glycine in PBS with Ca²⁺/Mg²⁺was used for quenching. Lysis was performed in RIPA buffer (50 mM Tris/HCl, pH 7.5; 150 mM NaCl; 1 mM EDTA; 1 % Nonidet-P40 (Calbiochem); 0.5 % Na-deoxycholate, 0.1 % Na-dodecylsulfate) with Complete Protease Inhibitor Cocktail (Roche, Indianapolis, IN, USA). The 10,000×g supernatant was incubated with UltraLink immobilized NeutrAvidin™ Protein Plus (Pierce, Rockford, IL, USA) for 2 h on a shaker at 4 °C. Beads were washed four times with RIPA buffer containing Complete Protease Inhibitor Cocktail (Roche, Indianapolis, IN, USA). Bound proteins were eluted with Laemmli buffer and analyzed by SDS-PAGE and Western blot with anti-GFP antibodies. For each claudin-YFP construct, the intensities of the immunoreactive bands were normalized to the respective intensity for Cld_WT-YFP detected in parallel. Surface biotinylation rate was calculated by dividing the normalized intensity in the cell surface fraction by the normalized intensity in the lysate.

Cellular cCPE-binding assay

Two days after transient transfection or 1 day after plating of stable lines, HEK293 cells expressing claudin constructs were incubated with 0.5 µg/ml GST-cCPE constructs (30 min. 37 °C) in 24 well plates (Techno Plastic Products AG, Trasadingen, Switzerland). cEND cells were prepared as described above and incubated with 5 µg/ml GST-cCPE (30 min. 37 °C). Cells were washed twice with ice-cold PBS (with Ca²⁺ and Mg²⁺), fixed (10 min with 2.4 % (w/v) paraformaldehyde in PBS) followed by quenching (20 min with 100 mM glycine in PBS) and blocking (10 min with 1 % (w/v) BSA, 0.05 % (v/v) Tween 20 in PBS). Cells were incubated with PhycoLink^® anti-GST-R-phycoerythrin conjugate in 1 % (w/v) BSA; 0.05 % (v/v) Tween 20; 2 µM Hoechst 33258 in PBS for 1 h and washed 3x with PBS. Bound GST-cCPE was detected via fluorescence intensity of PhycoLink^® anti-GST at λ_exc. = 545 ± 12 nm/λ_em. 578 ± 12 nm; claudins were detected at λ_exc. 506 + 5 nm/λ_em. 525 ± 5 nm for YFP tagged-, or λ_exc. 488 ± 5 nm/λ_em. 510 ± 5 nm for GFP-tagged claudins. Cell number was monitored with Hoechst dye at λ_exc. 365 ± 12 nm/λ_em. 480 ± 12 nm. Non-transfected HEK293 cells were used as a negative control (fluorescence intensity of GST-cCPE bound to Cld3_WT was 15.4 ± 0.8 times higher than the negative control). Fluorescence intensity of bound cCPE was normalized to amount of claudins (YFP or GFP intensity) or cell number (Hoechst intensity).

Equilibrium dissociation constants (K _d) were determined in cellular cCPE-binding assays with stably Cld5-expressing HEK293 cells and varying concentrations of the GST-cCPE constructs (1–4600 nM). The cells were fixed directly after cCPE incubation, without a previous washing step. The normalized fluorescence intensity of bound cCPE was plotted against the cCPE concentration. K _d was calculated using the non-linear regression analysis for a single-site, specific binding in GraphPad Prism version 5.0 (San Diego, CA). Non-specific binding was accounted for by subtracting the fluorescence signal yielded for non-transfected HEK293 cells after GST-cCPE incubation at the respective concentrations.

Structural bioinformatics and molecular modeling

Based on the crystal structure of murine Cld15 (PDB ID: 4P79, [3]), homology models of Cld1, Cld3, Cld4 and Cld5 were created. The alignment between murine Cld15, murine Cld1, murine Cld3, human Cld4 and murine Cld5 (Fig. S1) reveals a similarity score of 55 % (Cld1), 59 % (Cld3), 61 % (Cld4) and 48 % (Cld5). Gaps and extensions in the loop regions of the structural template were adjusted manually. For interaction models of the ECL2 with cCPE, on the one hand, the cCPE crystal structure was used as deposited (PDB ID: 2QUO, [12]); on the other hand, the interaction models were built consulting our previously reported ECL2 models and cCPE interactions models [1, 9, 32]. All manual reciprocal dockings, manipulations, optimizations of ECL2 models or cCPE models (cCPE_Y306W, cCPE_S313H and cCPE_Y306W/S313H, based on PDB ID: 2QUO), as well as calculations of hydrophobic and electrostatic potentials on the molecular surfaces, were performed with Sybyl X2.0 (Certara USA Inc., St. Louis, MO, USA). Models were energetically minimized using the AMBER7 FF99 force field.

Statistics

Unless stated otherwise, results are shown as mean ± SEM. Statistical analyses were performed using GraphPad Prism version 5.0 (San Diego, CA). Firstly, normality tests were performed (D’Agostino and Pearson omnibus, Shapiro–Wilk and Kolmogorov–Smirnov test). Data sets showing normal distribution were analyzed using an unpaired, one-tailed Student’s t test. Data sets not showing normal distribution were analyzed using Wilcoxon Signed Rank Test. p < 0.05 was taken as significant.

Results

The sequence of the claudin ECL2 turn—comprising the residues N/D, P, L/M/V, V/T, P/A/D—is essential for CPE interaction [30]. Relative to the highly conserved proline in this turn [1, 37, 38], residue positions prior and following that proline are designated here as P−1, P, P+1, P+2, P+3; (Fig. 1a). This designation helps in the following position-dependent analysis of cCPE interaction with different claudins and their mutants.

Fig. 1.

Residues at Positions P−1, P+1 and P+2 inhibit binding of cCPE to Cld1 and Cld5. a Sequence alignment of ECL2 (extracellular loop 2) of Cld1, −3, −4 to −9 (ClustalW2) displayed using Geneious Pro 5.4.4. Red frame: striking difference between Cld3 and Cld4 and Cld1 and Cld5. green: bulky hydrophobic; dark green: small hydrophobic; dark blue: polar uncharged; blue: basic; magenta: acidic; cyan: Tyr; black: Pro; orange: Cys. Helix, Turn and β-strand: secondary structure elements of the ECL2; CPE-binding: core residues of the CPE-binding region; TMH: transmembrane helix. b HEK293 cells transiently expressing Cld3_WT (orange columns), Cld1_WT (red columns) or Cld3-mimicking Cld1 mutants (gray columns) and c HEK293 cells transiently expressing Cld3_WT (orange columns), Cld5_WT (beige columns), Cld3-mimicking Cld5 mutants (gray columns) or Cld3/Cld5 chimeras (ChF, ChG, checkered and striped columns, scheme on the right depicts Cld3 (orange) and Cld5 (beige) portions of the chimeras) were incubated with 0.5 µg/ml GST-cCPE_WT. Bound cCPE was detected using anti-GST antibodies in a plate reader. b cCPE_WT binds to the Cld3-mimicking Cld1 mutants (Cld1_D150N; Cld1_T153V; Cld1_D150N/T153V) more strongly than to Cld1_WT. c cCPE_WT binds to the Cld3-mimicking Cld5 mutant Cld5_D149N/T151L and the Cld3/Cld5 chimeric mutants (ChF_D148N/T150L and ChG_D148N/T150L) much more strongly than to Cld5_WT or to ChF and ChG. Results are mean ± SEM (n ≥ 5); *p < 0.05

Claudins: substitutions in Cld1 and Cld5 that mimic Cld3, enhance cCPE-binding

Cld3 and Cld4 share a common sequence motif in the turn region of ECL2 (N^(P−1) P^(P) L/M^(P+1) V^(P+2) A/P^(P+3)) which mediates cCPE-binding [9, 30]. We analyzed specific residues (P−1, P+1 and P+2; Fig. 1a) of the turn region of Cld1, Cld3 and Cld5, and their contributions to the binding of cCPE_WT. Cld1 exhibits the sequence ₁₅₀D^(P−1) P^(P) L^(P+1) T^(P+2) P^(P+3) and Cld5, the sequence ₁₄₉D^(P−1) P^(P) T^(P+1) V^(P+2) P^(P+3) (Fig. 1a). Cld1_WT shows only weak binding of cCPE_WT while Cld5_WT does not bind cCPE_WT when compared to Cld3_WT (Fig. 1b, c). Cld3-mimicking constructs of Cld1, Cld1^(P−1)_D150N and Cld1^(P+2)_T153V, showed increased binding of cCPE_WT. Interestingly, Cld1^(P+2)_T153V displayed a much greater increase in cCPE-binding than did Cld1^(P−1)_D150N. Surprisingly, double substitution of positions P−1 and P+2 (Cld1^{(P−1)/(P+2)}_D150N/T153V) gave the lowest increase in binding (Fig. 1b). We next analyzed the effect of Cld3-mimicking substitutions of Cld5 at positions P−1 (Cld5^(P−1)_D149N) and P+1 (Cld5^(P+1)_T151L) and their double mutant (Cld5^{(P−1)/(P+1)}_D149N/T151L). In contrast to Cld1, the single mutants Cld5^(P−1)_D149N and Cld5^(P+1)_T151L revealed no or only very slightly increased binding of cCPE_WT. Only the double mutant Cld5^(P−1/P+1)_D149N/T151L showed greatly increased binding of cCPE_WT (Fig. 1c). Similarly, the corresponding double substitution D148N/T150L in Cld3/Cld5 chimeras containing the ECL2 (ChF) and ECL1 (ChG) of Cld5 and the rest of Cld3 strongly increased the binding of cCPE_WT (Fig. 1c).

This indicates that these residues (P−1 and P+1 in Cld5 or P+2 in Cld1) of the ECL2 turn region interfere with cCPE-binding to Cld1 and Cld5. Nevertheless, the changes in cCPE-binding to HEK293 cells expressing the Cld3-mimicking constructs could also be due to differences in the availability of the mutants for cCPE on the cell surface. Therefore, the subcellular localization of Cld3-mimicking mutants of Cld1 and Cld5 was analyzed by confocal microscopy, in comparison to the respective Cld_WT. We detected Cld1_WT and Cld5_WT enrichment at contacts between two claudin-expressing cells (Fig. S2A, B), as demonstrated previously [33]. This indicates trans-interaction between claudins. A similar enrichment was obtained for Cld1^(P−1)_D150N, Cld1^(P+2)_T153V, Cld1^{(P−1)/(P+2)}_D150N/T153V, Cld5^(P−1)_D149N and Cld5^(P+1)_T151L. Cld5^{(P−1)/(P+1)}_D149N/T151L exclusively displayed marked loss of contact enrichment, indicating that only for Cld5^{(P−1)/(P+1)}_D149N/T151L trans-interaction was directly or indirectly affected (Fig. S2A, B). To analyze whether the mutations affect the amount of claudins in the plasma membrane, cell surface biotinylation was performed (Fig. S2C). Only the mutations T151L^(P+1) and D149 N/T151L^{(P−1)/(P+1)} in Cld5 clearly reduced the levels of claudin in the plasma membrane; Nevertheless, confocal microscopy analyses showed that Alexa647-cCPE binds to HEK293 cells expressing Cld5_D149N/T151L-YFP (Fig. S2B, panel III). In addition, neither microscopy nor surface biotinylation indicated increased levels of Cld1^(P−1)_D150N, Cld1^(P+2)_T153V or Cld1^{(P−1)/(P+2)}_D150N/T153V in the plasma membrane. Furthermore, the Cld3/Cld5 chimeras ChF and ChG occurred at high levels in the plasma membrane and their subcellular distribution was essentially unchanged by D148N/T150L substitution (ChF, [34] and ChG, unpublished data). In summary, the results exclude the possibility that the differences in cCPE-binding to these constructs are due to differences in their subcellular localization. All analyzed claudin constructs reached the plasma membrane. This indicates that the substitutions did not induce misfolding of the protein which would interfere with the binding assays.

Taken together, our data indicate that residues of the ECL2 turn region at positions P−1, P+1 in Cld5 and at positions P−1, P+2 in Cld1 prevent strong interaction with cCPE. Consequently, to create cCPE-constructs with extended binding to Cld5_WT, cCPE has to be modified in such a way that it is able to complement the differences in the turn region of the ECL2 of Cld1 and Cld5 relative to Cld3 and Cld4.

cCPE: in silico analysis of interaction with Cld1 and Cld5 identified cCPE positions that can be modified to enhance binding

Utilizing previously established interaction models of cCPE with the ECL2 of Cld3 and Cld4 [9], we generated working models for the interaction of cCPE with Cld1 and Cld5 (Fig. 2). The interaction models indicated how the ECL2 turn fits into the hydrophobic binding pocket of cCPE. The previous homologous interaction model of cCPE and Cld3 indicates that ECL2 turn positions P−1 (Asn¹⁴⁹ in Cld3, Asp¹⁵⁰ in Cld1 and Asp¹⁴⁹ in Cld5) and P+2 (Val¹⁵² in Cld3, Thr¹⁵³ in Cld1 and Val¹⁵² in Cld5) lie on the peripheral rim of the binding pocket, which consists of two pits. The large hydrophobic pit (triple-Tyr pit) of cCPE is formed by a triple-Tyr motif (Tyr³⁰⁶, Tyr³¹⁰ and Tyr³¹²). Cld3 and Cld4 contain a large and bulky hydrophobic side chain at P+1 (Leu¹⁵⁰ and Leu¹⁵¹, respectively), completely filling the space of the triple-Tyr pit (Fig. 2b). Unlike all reported cCPE-binding claudins (e.g., Cld3 and Cld4), Cld5 lacks this bulky hydrophobic residue at P+1 of the ECL2—which was previously determined to be a key interaction partner for cCPE [9]. The smaller residues, Thr^(P+1) of murine or Ser^(P+1) of human Cld5, are unable to fill the triple-Tyr pit of cCPE (Fig. 2d). These findings lead to the conclusion that the triple-Tyr pit of cCPE requires modification (downsizing) to fit to Thr/Ser^151(P+1) of Cld5. Downsizing the triple-Tyr pit (magenta arrows in Fig. 2d) should compensate for the void space, as left by the lack of a large hydrophobic residue at position P+1 in Cld5 (green square in Fig. 2d). This should enable binding of cCPE to Cld5 and lead to a cCPE-variant with shifted specificity towards Cld5. In this study, we will refer to this cCPE-modification as “pit-filler”. Based on the working models, we selected the following residues Leu²³⁸, Ile²⁵⁸, Val²⁵⁹, Ala³⁰² and Tyr³⁰⁶ of cCPE as potential “pit-filler” positions.

Fig. 2.

Scheme of the claudin–cCPE Interaction. a Top view onto the surface of cCPE, (white) with the claudin-binding pocket and the triple-Tyr pit (green) and the rim of the triple-Tyr pit (blue). Dashed lines represent the cross section for the schemes. b Interaction of the ECL2 of Cld3 (gray) with the claudin-binding pocket of cCPE (black) in comparison with c Cld1 and d Cld5. Introduction of a positive-charged residue on the rim of the triple-Tyr pit which could interact with D^(P−1) (blue cloud) could lead to enhanced binding of cCPE to Cld1 and Cld5. To enable binding to Cld5, the huge gap between Thr/Ser^(P+1) (T/S) and the triple-Tyr pit (green square) has to be filled by substitutions around the triple-Tyr pit (magenta arrows, “pit-filler”)

Moreover, to enhance cCPE-binding to Cld1 or Cld5, a positively charged residue should be introduced at the peripheral rim of the triple-Tyr pit, as this can counteract the negatively charged Asp^(P−1) of Cld1 and Cld5 (Asp¹⁵⁰ and Asp¹⁴⁹, respectively, Fig. 2c, d), or interact with Gln^146(P−5) of Cld1 (“electrostatic angler”). In the latter case, the introduction of a positive-charged residue should only enhance binding to Cld1, since Cld5 exhibits an arginine at the respective position (Arg^145(P−5), Fig. 1a). Ser³⁰⁷ and Tyr³¹⁰ are located at the peripheral rim of the triple-Tyr pit and thus the interaction model suggested that these positions in cCPE could be used to introduce positive or polar residues (“electrostatic angler”), to permit hydrophilic interactions with either Asp^(P−1) in Cld1 and Cld5 or with Gln ^(P−5) in Cld1. Interestingly, substitution S307R was incorporated in multiple mutants that also apparently affect cCPE-binding to claudins [29].

cCPE: S313H of the triple serine mutant (305,307,313) contributes most to the binding of both Cld1 and Cld5

cCPE_{S304A/S305P/S307R/N309H/S313H} was shown to enhance cCPE-binding to Cld1 and Cld5 and cCPE_{S305P/S307R/S313H} was shown to enhance cCPE-binding to Cld1 [29]. To characterize the influence of each substitution of cCPE_{S305P/S307R/S313H} in detail, the corresponding cCPE single (cCPE_S305P and cCPE_S313H), double (cCPE_S305P/S307R) and triple mutants (cCPE_{S305P/S307R/S313H}) were generated and their binding to Cld1- and Cld5-constructs was investigated in cellular binding assays with HEK293 cells expressing Cld1_WT, Cld1^(P−1)_D150N, Cld1^(P+2)_T153V and Cld1^{(P−1)/(P+2)}_D150N/T153V (Fig. 3a), or Cld5_WT, Cld5^(P−1)_D149N, Cld5^(P+1)_T151Land Cld5^{(P−1)/(P+1)}_D149N/T151L (Fig. 3b).

Fig. 3.

Substitution S313H in cCPE makes the biggest contribution of cCPE_{S305P/S307R/S313H} to binding to Cld1 and Cld5. HEK293 cells transiently expressing a Cld1_WT (red columns), or Cld3-mimicking mutations of Cld1 (gray columns) or b Cld5_WT (beige columns), or Cld3-mimicking mutations of Cld5 (gray columns) were incubated with 0.5 µg/ml GST-cCPE_WT or GST-cCPE mutants. Bound cCPE was detected using anti-GST antibodies in a plate reader. Quantification normalized to cCPE_WT (relative binding) for each claudin construct. cCPE_S313H and cCPE_{S305P/S307R/S313H} showed a strong increase in binding to Cld1_WT and Cld1_D150N/T153V. a cCPE_S307R only showed a strong increase in binding to Cld1_WT. b Only cCPE_S313H and cCPE_{S305P/S307R/S313H} showed a strong increase in binding to Cld5_WT, Cld5_D149N and Cld5_T151L, but not to Cld5_D149N/T151L. Results are mean ± SEM (n ≥ 5); *p < 0.05

Claudin-1: Compared to cCPE_WT, the mutants cCPE_{S305P/S307R/S313H} and cCPE_S313H bound more strongly to Cld1_WT and Cld1^{(P−1)/(P+2)}_D150N/T153V, but no more strongly to Cld1^(P−1)_D150N and Cld1^(P+2)_T153V. In contrast, cCPE_S307R and cCPE_S305P/S307R exhibited stronger binding only to Cld1_WT, but not to the other Cld1 constructs (Fig. 3a).

Claudin-5: Compared to cCPE_WT, cCPE_{S305P/S307R/S313H} and cCPE_S313H showed much stronger binding to Cld5^(P+1)_T151L, somewhat stronger binding to Cld5_WT and Cld5^(P−1)_D149N and slightly but significantly stronger binding to Cld5^{(P−1)/(P+1)}_D149N/T151L. cCPE_S305P/S307R, cCPE_S305P and cCPE_S307R showed no clear increase in binding to any Cld5 construct (Fig. 3b).

In summary, S307R slightly strengthens binding to Cld1 but not to Cld5. S313H increases binding to Cld1 and Cld5, and has a stronger effect on binding than does S307R. Apart from cCPE_S307R, the introduction of a positive-charged or polar residue to the rim of the triple-Tyr pit (“electrostatic angler”) did not enhance binding of cCPE to Cld1 or Cld5 (cCPE_S307R, cCPE_Y310H and cCPE_Y310K). Binding data and our model (Fig. S3) suggest that the introduced Arg³⁰⁷ of cCPE_S307R most probably interacts with Gln^146(P−5) of Cld1. Cld5 exhibits an Arg^(P−5) at the corresponding position, which could explain why cCPE_S307R shows no increased binding to Cld5 but to Cld5_R145Q, especially when combined with substitutions S305P and S313H (cCPE_{S305P/S307R/S313H}, Fig. S3C).

cCPE: combination of Y306W with S313H enhances specificity to Cld5

After analyzing the contributions of single substitutions S305P, S307R and S313H to the binding behavior of the broadly specific claudin binder cCPE_{S305P/S307R/S313H}, we studied the binding ability of the potential “pit-filler” mutants to Cld1 and Cld5. None of the single mutants (cCPE_L238F, cCPE_V259Y, cCPE_A302M, cCPE_A302Y, cCPE_Y306W) exhibited increased binding to Cld1 or Cld5 compared to cCPE_WT in cellular binding assays (Fig. 4a). Since cCPE_S313H showed enhanced binding to Cld5_WT, without decreasing the binding to Cld3 or Cld4, we decided to create constructs combining S313H with the potential “pit-filler” mutants. Thus, the following constructs were designed: cCPE_L238F/S313H, cCPE_V259Y/S313H, cCPE_A302Y/S313H and cCPE_Y306W/S313H. Binding of these constructs was then compared in cellular binding assays to the binding of cCPE_WT and cCPE_S313H to Cld1_WT and Cld5_WT (Fig. 4b).

Fig. 4.

cCPE_Y306W/S313H shows increased binding to Cld5. HEK293 cells transiently expressing Cld1_WT (red columns), Cld5_WT (beige columns) were incubated with 0.5 µg/ml GST-cCPE_WT, a GST-cCPE single mutants, b GST-cCPE double mutants containing substitution S313H and GST-cCPE_Y306A/Y315A as negative control. Bound cCPE was detected using anti-GST antibodies in a plate reader. Quantification normalized to cCPE_WT (relative binding) for the respective Cld_WT. a Only cCPE_V259Y and cCPE_Y306W show binding to Cld1 or Cld5 which closely resembles that of cCPE_WT. b cCPE_Y306W/S313H shows increased binding to Cld5 compared to cCPE_S313H and cCPE_WT while also showing weaker binding to Cld1 compared to cCPE_S313H. Results are mean ± SEM (n ≥ 5); *p < 0.05

Relatively to cCPE_WT, cCPE_V259Y/S313H demonstrated decreased binding to Cld1 and Cld5. cCPE_L238F/S313H and cCPE_A302Y/S313H showed stronger binding to Cld1 and Cld5 but bound more weakly than cCPE_S313H. In contrast, cCPE_Y306W/S313H bound much more strongly to Cld5 with a significant increase compared to cCPE_S313H. cCPE_Y306W/S313H exhibited decreased binding to Cld1_WT compared to cCPE_S313H.

To analyze the mechanism by which cCPE_Y306W/S313H enhanced cCPE-binding to Cld5 in more detail, we conducted cellular binding assays with HEK293 cells expressing Cld5_WT, and the Cld3-mimicking mutants Cld5^(P−1)_D149N, Cld5^(P+1)_T151L or Cld5^{(P−1)/(P+1)}_D149N/T151L (Fig. 5a). In contrast to the binding of cCPE_WT to the respective Cld5 construct, cCPE_Y306W/S313H showed marked increases in binding to both Cld5_WT and Cld5^(P−1)_D149N; in both cases, the binding was stronger than with cCPE_S313H. In addition, cCPE_S313H showed the greatest increase in binding to Cld5^(P+1)_T151L and also slightly enhanced binding to Cld5^{(P−1)/(P+1)}_D149N/T151L. As predicted for a “pit-filler” mutant, our designed construct cCPE_Y306W/S313H showed weaker binding than cCPE_S313H to Cld5^(P+1)_T151L and to Cld5^{(P−1)/(P+1)}_D149N/T151L. In addition, we analyzed the binding of these cCPE-constructs (cCPE_Y306W/S313H, cCPE_S313H, cCPE_WT and, as negative control, cCPE_Y306A/L315A) to Cld1, −3, −4, –5, −6, −7, −8 and −9 (Fig. 5b). Compared to cCPE_WT, the combined double mutant cCPE_Y306W/S313H showed weaker binding to Cld3, −4, −6, −7, −8 and −9, the strongest binding to Cld5 and slightly stronger binding to Cld1.

Fig. 5.

cCPE_Y306W/S313H shows shifted specificity towards Cld5 and enhanced binding to cEND cells. HEK293 cells a transiently expressing Cld5_WT, or Cld3-mimicking mutations of Cld5 or b transiently expressing (I) Cld1_WT, (IV) Cld5_WT or stably expressing (II) Cld3_WT, (III) Cld4_WT and (V) Cld6 _WT—(VIII) Cld9_WT were incubated with 0.5 µg/ml GST-cCPE_WT (gray columns), GST-cCPE_Y306W/S313H (red columns), GST-cCPE_S313H (blue columns) or GST-cCPE_Y306A/L315A (black columns) as negative control. Bound cCPE was detected using anti-GST antibodies in a plate reader. a cCPE_Y306W/S313H binds better to Cld5_WT and Cld5_D149N than cCPE_S313H and cCPE_WT, but binds more weakly to Cld5_T151L and Cld5_D149N/T151L than cCPE_S313H. b cCPE_Y306W/S313H binds better to Cld5_WT than cCPE_S313H and cCPE_WT and better to Cld1_WT than cCPE_WT but not better than cCPE_S313H. cCPE_Y306W/S313H binds more weakly than does cCPE_WT to all other claudins. Results are mean ± SEM (n ≥ 5); *p < 0.05. c Representative concentration-dependent binding of GST-cCPE_WT (gray line) and GST-cCPE_Y306W/S313H (red line) to stably Cld5-expressing HEK293 cells used for determination of K _d. d cEND cells, which endogenously express Cld5, were incubated with 5 µg/ml cCPE-Variants. Bound cCPE was detected using anti-GST antibodies in a plate reader. Quantification normalized to cCPE_WT (relative binding). Results are mean ± SEM (n ≥ 5); *p < 0.05

cCPE: Y306W/S313H binds to Cld5 with a nanomolar K_d

To quantify the affinity of cCPE_Y306W/S313H for Cld5_WT, concentration-dependent binding of the cCPE-variant to Cld5-expressing HEK293 cells was analyzed. cCPE_Y306W/S313H bound to Cld5_WT with a K _d of 33 ± 10 nM (n = 5). In contrast, the K _d value for cCPE_WT could not be determined precisely, since even with high cCPE_WT concentrations (up to 4600 nM), the binding was not clearly saturated (Fig. 5c). However, this result indicates that the K _d of cCPE_WT for Cld5_WT is at least ≫1.0 µM and therefore the affinity of cCPE to Cld5_WT was increased at least 30-fold by the Y306W/S313H substitution.

cCPE: Y306W/S313H shows enhanced binding to cells with endogenous Cld5

To investigate whether the positive effect of cCPE_Y306W/S313H on binding to Cld5 can also be observed in cells expressing Cld5 endogenously, we used the brain microvascular endothelial cEND cell line. These cells have been shown to express Cld5 at the protein and mRNA levels and are used as an in vitro model for the blood–brain barrier [39, 40]. We performed cellular binding assays of cCPE_WT, cCPE_S313H, cCPE_Y306W/S313H and cCPE_Y306A/L315A with this cell line (Fig. 5d). cCPE_Y306W/S313H showed a fivefold increased binding to cEND cells compared to the binding of cCPE_WT and a 2.5-fold increased binding compared to cCPE_S313H. This validates the enhanced binding of cCPE_Y306W/S313H to Cld5 observed in HEK293 cells in a cellular system expressing Cld5 endogenously. Interestingly, the difference in binding between cCPE_Y306W/S313H and cCPE_S313H even exceeds the difference observed in the HEK293 cell system.

Discussion

We aimed to understand the mechanism by which Clostrisium perfringens enterotoxin binds selectively and with high affinity to distinct claudins. In this study, we used a structure-based rational molecular approach to reveal mechanistic details about the cCPE–claudin interactions and indirectly to deduce information about the extracellular structural properties of interacting claudins, particularly Cld1 and Cld5. Insights into the interaction were gained by combining structural information, molecular modeling, cellular binding assays and mutagenesis. Firstly, we asked the question as to why cCPE only interacts weakly or not at all with Cld1 and Cld5. By sequence comparison, we identified two key differences between Cld1 and Cld5 and other claudins interacting with cCPE, namely Asp^150(P−1) and Thr^153(P+2) in Cld1 and Asp^149(P−1) and Thr^151(P+1) in Cld5 (Fig. 1a). Subsequently we created Cld3-mimicking mutants of Cld1 and Cld5 and analyzed their cCPE-binding behavior. Furthermore, we can provide a molecular explanation as to why the core substitutions of the cCPE-variants (cCPE_{S305P/S307R/S313H}) reported by Takahashi et al. [29] enhance Cld1 binding and allow Cld5 binding.

Our main goal, however, was to combine the knowledge gained from these analyses to create a cCPE-variant with a high affinity and a shifted specificity towards Cld5, which is not naturally bound by cCPE_WT. This could be used for the development of a blood–brain barrier modulator.

Claudins: ECL2 conformations differ as a result of Asp^(P−1) (Cld1 and Cld5) or Asn^(P−1) (Cld3 and Cld4)

The conformation of the ECL2 differs depending on the amino acid residue at position P−1. The results of the cellular binding assays of the Cld3-mimicking mutants of Cld1, Cld5 and the Cld3/Cld5 chimeras show that presence of Asp^(P−1) drastically disturbs the interaction of cCPE with Cld1 and Cld5 (Fig. 1b, c); this is in line with the results of the substitution of Asn^(P−1) to Asp in Cld3 and Cld4 [8, 9] and the substitution mapping of ECL2 peptides of Cld3 and Cld5 [30]. Firstly, it was assumed that Asp^(P−1) in claudins interrupts the binding of cCPE due to electrostatic interference with residues on the peripheral rim of the binding pocket. However, our results for cCPE_S307R, cCPE_S305P/S307R and cCPE_{S305P/S307R/S313H} provided no evidence for a direct interaction of Asp^(P−1) in Cld1 and Cld5 with Arg³⁰⁷ or His³¹³ in cCPE (Fig. 3). According to our data, position Arg³⁰⁷ in cCPE seems rather to interact with Gln^(P−5) of Cld1 (Fig. S3), which further confirms not only our previously reported cCPE–ECL2 interaction models of Cld3 and Cld4 [9], but also is in agreement with our current interaction model between cCPE and ECL2 of Cld5 (Figs. 2, 6), where the N-terminal helix of the ECL2 of Cld1 and Cld5 lies across the triple-Tyr pit.

Fig. 6.

Suggested conformational differences in the turn region of ECL2 of Cld1 and Cld5 compared to Cld3 and Cld4. cCPE is illustrated (cartoon, white) with the calculated surface (white); substituted residues are depicted as ball and stick in magenta. a Cross section of the detailed interaction model of cCPE_WT with the Cld3-ECL2 (cartoon, orange, with the turn defining residues as ball and stick; red: oxygen, blue: nitrogen). b Sketch of the interaction of the ECL2 of Cld3 and Cld4 (orange, dashed line: possible H-bond with backbone amid; red: oxygen, blue: nitrogen) and the claudin-binding pocket of cCPE_WT (black line) and cCPE_S313H (dashed gray line). c Sketch of the interaction of the ECL2 of Cld1 and Cld5 (beige, dashed lines: possible H-bonds with backbone amides; red: oxygen, blue: nitrogen) and the claudin-binding pocket of cCPE_WT (dashed black line) and cCPE_S313H (gray line). d Cross section of the detailed interaction model of cCPE_S313H with the Cld5-ECL2 (cartoon, beige, with the turn defining residues as ball and stick; red: oxygen, blue: nitrogen). The residue at position P-1 has a crucial influence on the structure of the ECL2. a, b While Asn^(P−1) (Cld3 and Cld4) can only form one hydrogen bond with the backbone amides of the ECL2-turn region and thus causes a more flexible and wider ECL2-structure which fits optimally for cCPE_WT, c, d Asp^(P−1) of Cld1 and Cld5 can form a hydrogen bond network with two backbone amides of the ECL2-turn region and the polar side chain of Thr^(P+1) (Cld5) or Thr^(P+2) (Cld1), which makes the turn structure narrower and more rigid. In summary: Cld1 and Cld5 possess a narrower and more rigid turn region of the ECL2 than Cld3 and Cld4

All of these results indicate that the different cCPE-binding behavior of Asp^(P−1) and Asn^(P−1) in ECL2 of claudins is more probably caused by an altered intramolecular conformation of the turn region of the ECL2 of Cld1 and Cld5 compared to the ECL2 of Cld3 and Cld4. According to the ECL2 models, Asp^(P−1) of Cld1 and Cld5 is able to form a stronger H-bond network with the backbone amide hydrogens of the residues P+1 and P+2 and/or one hydroxyl group (Thr/Ser^(P+1) Cld5 and Thr^(P+2) Cld1) compared to just one possible interaction for Asn^(P−1) of Cld3 and Cld4 (Fig. 6). This leads to a more intra-turn orientation of the side chain of residue P+2—Thr^153(P+2) (Cld1) and Val^152(P+2) (Cld5), making the ECL2 of Cld1 and Cld5 slimmer and more rigid compared to the protruding conformation of residue P+2 in the bulkier ECL2 of Cld3 or Cld4 (Fig. 6). This slimmer conformation of the ECL2 would explain the positive effect of the substitution S313H in cCPE_S313H, cCPE_{S305P/S307R/S313H} and cCPE_Y306W/S313H on binding to Cld1 and Cld5.

Cld5^(P−1)_D149N does not bind cCPE due to the lack of a bulky hydrophobic residue (Leu) at position P+1 that fills the triple-Tyr pit of cCPE. However, the Cld5 mutant containing Leu at position P+1, Cld5^(P+1)_T151L does not bind cCPE well, since it still possesses the slimmer and more rigid turn conformation of Cld5_WT caused by Asp^149(P−1). This is supported by the finding that cCPE_S313H, which fits better to the slimmer ECL2 conformation, shows the strongest binding to Cld5^(P+1)_T151L (Fig. 3b). Taken together, the turn conformation of ECL2 for Cld1, Cld3, Cld4 and Cld5 looks very similar, but it exhibits distinct slight conformational differences (Fig. S4), which are reflected in the divergent binding behavior and selectivity of cCPE versus these claudins.

cCPE: narrowing the claudin-binding pocket by S313H enhances Cld1- and enables Cld5-binding

Besides confirming that the triple mutant (cCPE_{S305P/S307R/S313H}) increases binding to Cld1 and that the single substitution S313H contributes most to this increased binding [29], we here demonstrate that the single substitution S313H alone enables binding to Cld5. This can be explained by our molecular interaction models. The small serine at position 313 of cCPE allows the interaction with the bulky ECL2 of Cld3 and Cld4 (Fig. 6a, b). The replacement of the serine side chain by the lager imidazole ring of histidine changes the shape and narrows the binding pocket of cCPE (Fig. 6c, d). Since Cld1 and Cld5 possess a slimmer shaped ECL2 than Cld3 and Cld4, the residue at P+2 (Val^152(P+2) in Cld5 and Thr^153(P+2) in Cld1) cannot interact properly with Ser³¹³ at the peripheral rim of CPE`s binding pocket (Fig. 6c, d), unlike the residue at P+2 in the ECL2 of Cld3 and Cld4 (Val^(P+2) in both cases). Thus, a gap remains between the rim (dotted black line in cartoon, Fig. 6c) and the ECL2 of Cld1 and Cld5. S313H fills this particular space in the binding pocket. Consequently cCPE-constructs containing S313H fit much better to the slimmer conformation of the ECL2 of Cld1 and Cld5 compared to the bulkier turn of ECL2 of Cld3 and Cld4.

This suggested interaction with the particularly slim ECL2 turn is also supported by the lack of additional positive effects of S313H on binding to Cld1 and Cld5 constructs in which the conformation of the ECL2 is altered (e.g. Cld1^(P−1)_D150N, Cld1^(P+2)_T153V and Cld5^{(P−1)/(P+1)}_D149N/T151L). However, additional contribution of an electrostatic effect by S313H on some claudins seems feasible, in particular for Cld1 and Cld7 both exhibiting Thr at position P+2.

The following modeling results can be seen as explanation for our findings that cCPE_S313H does not show strong inhibition of binding to Cld3, −4, −6 and −9 and even exhibits slightly increased binding to Cld7 (Thr^(P+2) like Cld1, see Fig. 1a). All the latter claudins are predicted to possess a wider ECL2 than Cld1 and Cld5, due to Asn^(P−1), but since Asn^(P−1) allows a more flexible turn compared to Asp^(P−1) of Cld1 and Cld5; it is reasonable to assume that the side chain at P+2 might also be allowed in an intra-turn orientation in the ECL2 of Cld3, Cld4 and Cld6–Cld9 (Fig. 6). In addition, apart from Asn^(P−1), all these claudins also possess a bulky hydrophobic residue at position P+1 which fills the triple-Tyr pit of the binding pocket (Figs. 1a, 2a, 6a) and thus contributes very strongly to CPE-binding.

cCPE: “pocket filler” mutant Y306 W downsizes the triple-Tyr pit and shifts specificity towards Cld5

The interaction between Cld3 or Cld4 with cCPE is processed by a knob and hole interaction of the bulky hydrophobic Leu^(P+1) and triple-Tyr pit, respectively [9]. In Cld5, this bulky hydrophobic knob is missing; it contains instead very small residues at position P+1 (Thr^151(P+1) in murine Cld5 or Ser^151(P+1) in human Cld5) and it is thus a non-receptor claudin. To enable or enhance cCPE-binding of Cld5, the triple-Tyr pit had to be altered in such a way that it can compensate for the lack of bulky hydrophobic residues like Leu or Met at position P+1 of the ECL2 of Cld5. To achieve this, several putative “pit-filler” mutants were identified in silico, and then created and tested in vitro. We postulated that a true “pit-filler” mutant has to have increased binding to Cld5_WT, while showing decreased or no binding to claudins with a bulky hydrophobic residue at position P+1. Indeed, the introduction of the larger tryptophan side chain to replace tyrosine at position 306 of cCPE shows this effect when combined with S313H (Figs. 5a, b, 7). Compared to the binding pocket of cCPE_WT, the modeled binding pocket of cCPE_Y306W/S313H shows clear downsizing of the triple-Tyr pit (Fig. 7a, b). In the interaction models of cCPE_WT and cCPE_Y306W/S313H with Cld5, substitution Y306W fills the void space (green in Fig. 7c) between Thr^(P+1) of Cld5 and the triple-Tyr pit of cCPE (Fig. 7c, d). This explains why cCPE_Y306W/S313H shows enhanced binding to Cld5_WT compared to cCPE_WT and cCPE_S313H, but decreased binding to all claudins which have a bulky hydrophobic residue at position P+1, such as in Cld1, Cld3, Cld4, Cld5^(P+1)_T151L, Cld5^(P−1/P+1)_D149N/T151L and Cld6–Cld9 (Fig. 5a, b). The knob (P+1) of these claudins fails to interact with the closed hole (triple-Tyr pit of CPE). Thus, the effect of Y306W is due to an altered receptor site for the Thr/Leu^(P+1) of Clds and of course always supported by other side chains such as the mutation S313H. Furthermore, cCPE_Y306W/S313H shows enhanced binding to Cld5_WT, with a nanomolar K _d (33 ± 10 nM), which is of the same order of magnitude as the K _d of cCPE_WT for the classic CPE receptor claudins Cld3 and Cld4 (K _d≈10 nM) [7, 41]. cCPE_Y306W/S313H also shows a strongly enhanced binding to the microvascular endothelial cEND cell line (Fig. 5d); these cells express Cld5 endogenously and are an in vitro model for the blood-brain barrier.

Fig. 7.

Substitution Y306W downsizes the triple-Tyr pit, making it a better fit for Cld5. Cross section of the claudin-binding pocket of a cCPE_WT and b cCPE_Y306W/S313H and the interaction model of Cld5 with c cCPE_WT and d cCPE_Y306W/S313H. cCPE is illustrated (cartoon, white) with the surface displayed (white) and the triple-Tyr pit defining residues in ball and stick (blue); substituted residues are depicted as ball and stick in magenta. ECL2 of Cld5 is shown in beige, with the residues of the turn region as ball and stick (white, C and H; blue, N; red, O). c Thr^(P+1) of Cld5 is not able to fill the triple–Tyr pit of cCPE_WT (highlighted in green), d by substituting Y³⁰⁶ with Trp the triple-Tyr pit was shrunken and closes the space to Thr^(P+1) of Cld5

Claudin–15 structure confirms intramolecular stabilization of the ECL2 by Asn/Asp^(P−1) and Lys^(P+7)

Recently two crystal structures of modified ECL2 peptides in complex with CPE were released (PDB ID: 3ZJ3 [42] and 4P5H [43]), which show opposing orientations of these peptides. However, the low resolutions of 3.37 Å and 3.38 Å do not allow explicit interpretation of H-bond interaction. Additionally and more importantly, a discrepancy concerning cCPE-binding to peptides versus binding to full-length claudins has already been observed since a Cld4 ECL2 peptide does not bind, whereas full-length Cld4 binds strongly [9, 30]. Thus, for the generation of homology models, the Cld15 structure (PDB ID: 4P79 [3]) with a resolution of 2.4 Å was used as a structural template for Cld1, –3, –4 and –5. This new crystal structure provides us with ideal means of assessing the performance of our claudin interaction models.

Our previous studies were based on initial ECL2 models with a helix-turn-helix motif [1, 6, 9, 30, 32]. We observed a remarkable structural agreement between these initial ECL2 models and the Cld15 structure in the N-terminal and especially in the central part, the turn region of ECL2, which is the interaction interface with cCPE (Fig. 8). This is demonstrated by the low RMSD values between Cld15 ECL2 (Ile¹³⁵-Tyr¹⁵¹) to the initial Cld5 model (Leu¹³⁶-Val¹⁵³) = 0.69 Å (pairwise fitting of backbone atoms). The only difference between the initial ECL2 model and the Cld15 structure concerns the C-terminal part (Lys¹⁵⁷-Gly¹⁶¹ in Cld5 and the corresponding Lys¹⁵⁵-Gly¹⁵⁹ in Cld15) for which we previously predicted a helical conformation, whereas the Cld15 structure shows a short beta strand that is interacting with the ECL1. However, this part is according to our results not directly participating in the essential CPE-binding region (Fig. 1a). Thus, the interacting interface of the initial ECL2 model of Cld5 to cCPE is nearly identical to the now solved crystal structure of Cld15.

Fig. 8.

Superposition of the interacting CPE interface for the initial Cld5-ECL2 model (cartoon, beige) and the crystal structure Cld15 (PDB ID: 4P79, cartoon, cyan) shows a high similarity. Residues of the turn region depicted as sticks (white, H; blue, N; red, O). Pairwise fit of the back bone atoms of the N-terminal helices and the turn region (Ile¹³⁵-Tyr¹⁵¹ in Cld15 and Leu¹³⁶-Val¹⁵³ in Cld5) for the Cld5-ECL2 model (beige) and the Cld15 structure (cyan), demonstrates a high similarity and reveals an RMSD of 0.69 Å. The crystal structure of Cld15 confirms our initially postulated turn stabilizing interaction of Lys^157(P+7) in Cld5 [1, 32] (Lys¹⁵⁵ in Cld15) with the back bone carbonyl group of Asp^149(P−1) (Asn¹⁴⁸ in Cld15) as well as the involvement of the residue at position P−1 in the turn conformation

More importantly, the crystal structure of Cld15 also indicates an involvement of Asn^(P−1) in stabilization of the ECL2’s turn structure, which is in line with our analysis of the impact of the Asn/Asp^(P−1) difference in the ECL2 of claudins on binding of cCPE (Fig. 6). Furthermore, this crystal structure, also confirms our already previously [1, 32] postulated stabilization of the ECL2 turn by interaction of Lys^157(P+7) with the back bone carbonyl group of Asp^149(P−1) (Lys¹⁵⁵ and Asn¹⁴⁸ in Cld15) and the N-terminal helix of ECL2 (Fig. 8). Finally, the crystal structure confirms the main interacting turn conformation of our ECL2 models for claudins with cCPE.

In summary, we elucidated the molecular mechanism by which cCPE binds strongly to its receptor (Cld3), but only weakly to Cld1 and not at all to the non-receptor Cld5. By utilizing these insights, we generated a cCPE-variant with a specificity shifted away from the receptor claudins (e.g., Cld3, Cld4 and Cld6–Cld9) towards a non-receptor claudin (Cld5), by downsizing the binding pocket of cCPE, to complement the lack of a bulky hydrophobic residue in the ECL2 of Cld5. This newly developed cCPE-variant, with nanomolar affinity for Cld5, shows decreased binding to other claudins. Moreover, it demonstrated efficient binding to endogenous Cld5-expressing blood–brain barrier model cells, thus laying the foundation for high affinity tissue-specific claudin targeting. Therewith, we are able to provide the proof of principle that cCPE can be modified by multiple mutations in such a way that the specificity for claudins can be shifted even towards non-CPE receptor claudins. Apart from this new cCPE–claudin interaction pattern, differences in the structural properties of the ECL2 of claudins have been deduced and can be used to develop claudin subtype-specific binders as TJ modulators for tissue-specific drug delivery or for targeting cancer cells that overexpress claudin.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

TJ

Tight junctions

CPE

Clostridium perfringens Enterotoxin

cCPE

C-terminal domain of Clostridium perfringens enterotoxin

Cld

Claudin

ECL

Extracellular loop

PDB

Protein data bank

RMSD

Root mean square deviation