Caspase-4 disaggregates lipopolysaccharide micelles via LPS-CARD interaction

Lipopolysaccharides (LPS) are a major component of the outer membrane of Gram-negative bacteria and are pathogen-associated molecular patterns recognized by the TLR4/MD2 complex that induces an inflammatory response. Recently, the cytosolic receptors caspase-4/-5/-11 that bind LPS inside the cell and trigger inflammasome activation or pyroptosis, have been identified. Despite the important roles of caspase-4 in human immune responses, few studies have investigated its biochemical characteristics and interactions with LPS. Since caspase-4 (C258A) purified from an Escherichia coli host forms aggregates, monomeric proteins including full-length caspase-4, caspase-4 (C258A), and the CARD domain of caspase-4 have been purified from the insect cell system. Here, we report the overexpression and purification of monomeric caspase-4 (C258A) and CARD domain from E. coli and demonstrate that purified caspase-4 (C258A) and CARD domain bind large LPS micelles and disaggregate them to small complexes. As the molar ratio of caspase-4 to LPS increases, the size of the caspase-4/LPS complex decreases. Our results present a new function of caspase-4 and set the stage for structural and biochemical studies, and drug discovery targeting LPS/caspase-4 interactions by establishing a facile purification method to obtain large quantities of purified caspase-4 (C258A) and the CARD domain.

Lipopolysaccharides (LPS) are a major component of the outer membrane of Gram-negative bacteria. Their structure comprises three major parts, lipid A, a core oligosaccharide, and the O-antigen. Lipid A contains two glucosamine units with hexa-acyl chains and two phosphate groups, and is referred to as an active unit that causes the inflammatory response in mammalian hosts 1 . LPS is also referred to as an endotoxin, and is a representative pathogen-associated molecular pattern recognized by the host innate immune system. LPS micelles in humans are recognized by the LPS binding protein (LBP), and LPS is transferred to CD14 and forms a heterodimeric complex with TLR4/MD2 on the surface of immune cells [2][3][4] . While the complex formation initiates signal transductions resulting in the expression of genes involved in host defense 3,5 , excessive amounts of LPS can result in a lethal septic shock and even death. Recently, caspase-4/-5 from human and caspase-11 from mouse were reported as cytosolic LPS receptors. Caspase-4/-5/-11 are enzymes belonging to a family of cysteine proteases and are classified as inflammatory caspases. These enzymes are synthesized as a latent zymogen containing the caspase activation and recruitment domain (CARD), and large and small subunits (Fig. 1). The physiological functions of caspase-4/-5/-11 have been recently elucidated. In 2013, Kayagaki et al. and Hagar et al. independently found that caspase-11 is activated by cytosolic LPS and leads to non-canonical inflammasome activation 6,7 . The next year, Shi and colleagues showed that caspase-4 and caspase-11 directly bind to cytosolic LPS through their CARD domain and suggested that LPS binding induces oligomerization of caspase-4 or caspase-11, and that oligomerization is required for caspase-4 or caspase-11 activation and formation of non-canonical inflammasome 8 . Activated caspase-4 induces inflammasome formation as well as pyroptosis, a form of inflammatory cell death through cleavage of Gasdermin D 9 . Several studies have also investigated how LPS enters cells and demonstrated that outer-membrane vesicles (OMVs) secreted by living Gram-negative bacteria enter the host cell through endocytosis 10 . Escherichia coli O111:B4 LPS could bind to the cholera toxin B subunit followed by endocytosis via the GM1 ganglioside 6,7 , or release of bacteria from the vacuoles via small interferon-induced guanylate binding proteins 11,12 . However, the mechanism of LPS recognition by caspase-4 through the CARD domain and the events occurring after binding to LPS at the molecular level are still unclear. Contrary to previous suggestions that caspase-4 is oligomerized by LPS binding 8 , another study suggested that caspase-4 just binds to large LPS micelles that provide a surface for caspase-4 activation resulting in high molecular weight complexes 13 . Since E. coli has endogenous LPS, full-length caspase-4/-5/-11 purified from E. coli are isolated as high molecular weight aggregates; therefore, all biochemical studies have been performed with caspase-4 purified from the Bac-to-Bac Baculovirus Expression System 8,13 . Here we established the expression and purification procedures for the monomeric caspase-4 (C258A) variant and its CARD domain from an E. coli system at the mg scale to understand the roles of caspase-4 during its interactions with LPS. With the purified CARD domain and caspase-4 (C258A), we demonstrate for the first time that these proteins not only bind to large LPS micelles, but also break them into lower molecular weight complexes in vitro. We also found that as the molar ratio of caspase-4 to LPS increases, the size of the caspase-4/LPS complex decreases as observed by gel filtration chromatography and transmission electron microscopy (TEM). Our results provide a fundamental basis to understand the role of caspase-4 during non-canonical inflammasome formation by cytosolic LPS, and the availability of E. coli purified monomeric caspase-4 (C258A) and CARD domain will facilitate further biochemical and structural studies as well as drug discovery for sepsis and inflammation related diseases.

Results
Overexpression and purification of recombinant caspase-4 (C258A) and CARD domain using E. coli cells. Despite the important roles of caspase-4, biochemical and structural studies on caspase-4 have been limited due to difficulties in purification. Although wild type caspase-4 would provide information regarding caspase-4 activation by LPS, we could not detect intact full-length caspase-4 on an SDS-PAGE gel when we tried to purify wild type caspase-4 in the E. coli host. Therefore, instead of using wild type caspase-4, we decided to study the LPS/caspase-4 complex using the caspase-4 (C258A) variant and the CARD domain that plays a critical role in LPS recognition. According to a previous study 8 , caspase-4 (C258A), a catalytically inactive variant, purified from E. coli cells was eluted as about 600 kDa aggregates based on the size exclusion chromatography (SEC), whereas caspase-4 expressed and purified from Sf21 insect cells was a monomer as determined by analytical ultracentrifugation and static light scattering. Since protein purification from insect or mammalian cells is normally more expensive, more cost and time effective methods are required for performing structural and biochemical studies as well as for drug screening targeting the interaction between LPS and the CARD domain of caspase-4. Therefore, we asked how we could establish a purification method for monomeric caspase-4 (C258A) and the CARD domain from E. coli, which has not been reported before. We then noticed that unlike LPS, lipid IV A , a precursor of the LPS biosynthetic pathway, did not activate caspase-4 as much as LPS and induced a lower molecular oligomer than LPS did in a native gel 8 . Therefore, we hypothesized that if E. coli produces lipid IV A as its only LPS, it may allow expression and purification of caspase-4 and the CARD domain in the monomeric form. To test this, we transformed pET28b-caspase-4 (C258A) and pET28b-CARD into ClearColi BL21 (DE3) cells that produce lipid IV A instead of full-length LPS 14 and C41 (DE3) 15 that produces LPS. With two different E. coli hosts, we compared the expression level of caspase-4 (C258A) and the CARD domain containing 1-80 amino acids of caspase-4, as suggested by UniProt 16 (Fig. 1). The expression levels of caspase-4 (C258A) from ClearColi BL21(DE3) were much higher than those in C41 (DE3) under the same conditions (Fig. 2a). In addition, the CARD domain was rarely expressed in C41 (DE3), but was well expressed in ClearColi BL21 (DE3) based on an SDS-PAGE gel (Figs 2a and S1). When we purified Caspase-4 (C258A) and CARD domain from ClearColi BL21 (DE3), the proteins were eluted at ~600 kDa from a gel filtration column ( Supplementary Fig. S2), implying that caspase-4 still binds to lipid IV A micelles maybe with lower affinity compared to LPS. Therefore, to obtain monomeric caspase-4 (C258A) or CARD domain, we had to design a method to dissociate the caspase-4/ lipid IV A interaction. Shi et al. 8 reported that when Tween 20 was present, caspase-4 did not form a high molecular weight oligomer with LPS in a native gel. It is highly possible that caspase-4 expressed in the cytoplasm of E. coli may bind to LPS during lysis, so if we lysed cells in the presence of 1% Tween 20, caspase-4 could not form complexes with lipid IV A , thereby resulting in monomeric proteins. Based on this hypothesis, we added 1% Tween 20 to the lysis buffer and removed it during Ni-NTA affinity chromatography, and the resulting caspase-4 (C258A) was eluted at ~49.7 kDa from a gel filtration column ( Supplementary Fig. S2) and monomeric caspase-4 (C258A) was also determined via analytical ultracentrifugation (Fig. 2b,d). The purity of caspase-4 (C258A) was determined by SDS-PAGE ( Supplementary Fig. S3). We noticed that after Ni-NTA affinity chromatography, two major protein fractions were purified for caspase-4 (C258A). The lower molecular weight fraction in lane 4 of Supplementary  Complex sizes of caspase-4 (C258A/LPS) or CARD/LPS depend on the molar ratio between caspase-4 (C258A) and LPS or CARD and LPS. After purification of caspase-4 (C258A) and the CARD domain from E. coli, we wanted to test whether these proteins have the ability to bind LPS as previously reported for caspase-4 purified from sf1 cells 8,13 . To identify complex formation and the relative size of the complexes, we incubated fixed concentrations of caspase-4 (C258A) (12.5 μM) or CARD domain (23 μM) with an increased amount of Ra-LPS composed of lipid A and inner-and outer-core polysaccharides but no O-antigen region. After incubation, the reaction mixtures were injected into a gel filtration column, and the retention times for the samples were compared. As shown in Fig. 3a, purified caspase-4 (C258A) formed higher molecular weight complexes as the LPS amounts were increased. These complexes were abolished if the CARD domain was deleted (Δ80 caspase-4(C248A)) (Fig. 3b). When we looked at the data in detail, we noticed several interesting observations. First, LPS/Caspase-4 (C258A) formed a ~600 kDa complex, and we did not observe a lower molecular weight complex even though enough caspase-4 remained in a monomeric form. Second, to ensure that all monomeric caspase-4 participated in the complex formation, 2-5 molar equivalent of Ra-LPS was required. This implied that the LPS and caspase-4 complex does not form in a 1:1 stoichiometry. Third, considering that the monomeric molecular weights of caspase-4 (C258A) and Ra-LPS are 45.4 kDa and ~3.9 kDa, respectively and based on the rough stoichiometry we observed (1:2-5 = caspase-4 (C258A): Ra-LPS), about 9-10 molecules of caspase-4 (C258A) might be participating in the complex formation. These phenomena were also observed when we monitored the CARD domain and the Ra-LPS complex formation, as shown in Fig. 3c. For the CARD/Ra-LPS complex, the first CARD/LPS complex in Fig. 3c was eluted at ~136 kDa. Since the Superdex 200 10/300 GL column has an optimum separation range from 10 kDa to 600 kDa ( Supplementary Fig. S6), we noticed a gradual increase in the complex as the concentration of Ra-LPS was increased. Intriguingly, we observed a discrete and symmetric UV trace from each complex regardless of the molar ratio between the proteins/Ra-LPS. (Fig. 3a,c). This implies that caspase-4/Ra-LPS form a complex of a specific size depending on the molar ratio rather than on non-specific micelles. Our results demonstrate that caspase-4 (C258A) and the CARD domain purified from E. coli bind to LPS micelles and that the 1-80 amino acid region of caspase-4 is enough to recognize LPS. Notably, the minimum concentration of Ra-LPS used in this study was 6.25 μM. Even considering the dilution effect during gel filtration chromatography (24 mL column volume), the minimum Ra-LPS concentration we used was about ~130 nM. According to a previous study, the hydrodynamic radius of Kdo 2 -lipid A that is the core structure of Ra-LPS is nearly 80 nm at a concentration above CMC (41.2 ± 1.6 nM) at 25 °C 17 , and the hydrodynamic diameter of the 2000 kDa dextran was calculated as 54.4 nm 18 . Therefore, if CARD simply binds to the surface of Ra-LPS micelles, the complexes would be eluted in a void volume of the column rather than being eluted at 136-600 kDa, as observed in Fig. 3. The results imply that caspase-4 does not simply bind to a large LPS micelle; rather, caspase-4 reduces the size of the LPS micelle.

Caspase-4 (C258A) disaggregates large LPS micelles to smaller complexes via its CARD domain.
The above implication raises the following questions. Does caspase-4 bind large LPS micelles and disaggregate them to smaller LPS/caspase-4 complexes? If the molar ratio of Ra-LPS and caspase-4 determines the size of the complex, are large LPS micelles broken into smaller complexes when we add an increased amount of caspase-4 into a fixed amount of Ra-LPS micelles? To answer these questions, 23 μM of Ra-LPS was titrated with increased amounts of caspase-4 (C258A) or the CARD domain and analyzed using a transmission electron microscope (TEM). Based on the TEM images, we observed that large LPS micelles were dramatically disaggregated into smaller complexes by adding 0.92 μM of caspase-4 (C258A) or the CARD domain corresponding to a 1:0.04 molar ratio of the Ra-LPS:protein (Fig. 4a,c). As the amount of proteins increased, the size of the Ra-LPS/protein complex was decreased. This disaggregation ability of caspase-4 (C258A) is reminiscent of LBP protein 2 that we used as a positive control in our TEM experiment (Fig. 4d). We also observed that disaggregation of LPS micelles by caspase-4 (C258A) completely relied on the presence of the CARD domain (Fig. 4b,c).

Discussion
Caspase-4 is known as a receptor for intracellular LPS, and induces non-canonical inflammasome formation and pyroptosis. Numerous biological studies related to caspase-4 have found that it plays roles in acute and chronic diseases such as sepsis and cancer 6,7,19 . Accordingly, targeting caspase-4/LPS interactions is important for drug development. However, few studies 8,13 have sought to understand the biochemical characteristics of caspase-4 or the CARD domain as an LPS receptor in vitro. Biochemical, structural, and pharmaceutical studies require a facile method to obtain purified caspase-4 or the CARD domain in large quantities. In this study, we have established an efficient expression and purification system for full-length caspase-4 (C258A) and its CARD domain using E. coli cells. With purified caspase-4 (C258A) and CARD, we showed that these proteins bind to LPS micelles as well as break large LPS micelles into smaller LPS/protein complexes in an LPS/protein molar ratio-dependent manner (Figs 3, 4 and 6). We also showed that caspase-4 disaggregates LPS equally well regardless the presence of the O-antigen region (wild type E. coli O55:B5 LPS containing O-antigen vs. Ra-LPS without O-antigen, Supplementary Fig. S8). Altogether, we found that caspase-4 has a similar function to LBP that binds and disaggregates LPS micelles depending on the molar ratio between LBP and LPS during the LPS transfer cascade from LBP/CD14 to TLR4/MD2 2,4 . Until the discovery of caspase-4 as a cytosolic LPS receptor, the LBP/CD14/TLR4-MD2 system was representative of the LPS recognition system in mammalian hosts. LBP binds LPS and then CD14 binds to LPS-bound LBP and induces changes in LBP structure. The structural change transfers LPS from LBP to CD14, and LPS is eventually transferred from CD14 to MD2 and forms the LPS/ TLR4/MD2 complex 4 . The LPS/TLR4/MD2 complex induces intracellular signaling pathways. During the LPS transfer process, large LPS micelles are effectively disaggregated when LBP and CD14 are presented together 4 or when the molar ratio of LBP to LPS is increased 2 . Previously, two different explanations were suggested for high molecular weight entities observed by pore-limited native gel analysis and gel filtration chromatography for LPS and caspase-4/-5/-11 recognition. One study suggested that LPS binding induces caspase-4 oligomerization to large molecular entities 8 and activates caspase-4. On the other hand, another study suggested that caspase-4  binds to LPS aggregates or LPS-rich membrane interfaces on which caspase-4 can assemble in the appropriate proximity or orientation to active caspase-4 13 . In contrast to previous suggestions, we have found that caspase-4 binds large LPS micelles, and disaggregates them to smaller LPS/caspase-4 complexes (Fig. 4), which may provide the appropriate proximity and orientation to active caspase-4. According to a recent study, the expression level of Guanylate Binding Protein (GBP) is increased by OMV stimulation. This study also confirmed that GBP binds to LPS molecules on the surface of the OMV, which facilitates LPS recognition by caspase-11 20 . While it is not currently clear whether caspase-4 itself has the full ability to extract a single LPS molecule from the OMV or bacteria, large LPS micelles in the cytosol formed from the OMV or by the lysis of whole bacterial cells by GBP would be recognized by caspase-4 and are disaggregated to smaller LPS/caspase-4 complexes. Thus, caspase-4 and GBP seem to have similar functions carried out by the LBP/CD14/TLR4/MD4 system for LPS-dependent signaling, including binding to LPS, disaggregating LPS micelles, and activating downstream signaling cascades. So far several proteins including LBP 2 , albumin, hemoglobin, and high-density lipoprotein have been reported to possess disaggregating ability for LPS micelles 21 . Still many questions remain to be answered to understand LPS/caspase-4 recognition, such as whether different sizes of LPS/caspase-4 complexes induce different levels of downstream effects including protease activity of caspase-4, inflammation activation, and pyroptosis; how caspase-4 binds to LPS at the molecular level; and whether LPS binding induces structural changes in caspase-4 or not. We have now set the stage to study some of these questions by establishing large quantities of purified caspase-4 (C258A) and the CARD domain. While caspase-4 has been known to recognize and bind LPS using the CARD domain, we found a fundamental role of caspase-4 that disaggregates LPS micelles to form stable LPS/caspase-4 complexes in LPS dependent signaling implying that each cell may form different sizes of LPS/caspase-4 complexes depending on the caspase-4 expression levels and the amount of LPS exposed to the cytosol of the cell during infection.

Experimental Procedures
Materials. ClearColi BL21(DE3) were purchased from Lucigen (USA). All PCR amplification executed using KOD Hot Start DNA Polymerase (Novagen, USA) or Pfu polymerase (ELPiS, ROK) and a T3000 thermocycler (Biometra). Sequencing and primer synthesis were done at Macrogen (Seoul, ROK). His-tag human LBP protein purified from HEK293 was purchased from Sino Biological. Ra-LPS mutant WT-LPS-from E. coli O55:B5 and all other chemicals were purchased from Sigma-Aldrich (Korea) unless stated otherwise.
Purification of human caspase-4 CARD domain. Cell pellets were suspended with 30 mL of buffer B.
Cells were lysed by sonication and soluble fractions were separated by centrifugation (20,000 × g for 40 min at 4 °C). The supernatant was heated at 70 °C for 30 min and spun down at 20,000 × g for 30 min to remove denatured proteins. The resulting supernatant was loaded on a pre-equilibrated 1 mL HisTrap column (GE Healthcare, USA) with buffer B and washed with 10 CV of buffer B and 10 CV of buffer B containing 0.1% Tween20 and 50