Discovery of a caspase cleavage motif antibody reveals insights into noncanonical inflammasome function

Significance

Inflammasomes, cytosolic multicomponent complexes, sense patterns of pathogenesis or metabolic changes and initiate a cellular response via the activation of the inflammatory caspases (iCasps). A more comprehensive substrate analysis for iCasps will greatly improve our understanding of this pathway and provide potential blood-based biomarkers for inflammasome activation in disease. Antibodies typically exhibit high affinity and specificity for a single modification site, in direct contrast to a protease that exhibits a more degenerate recognition profile. Here, we generate antibodies that defy this convention and exhibit a degenerate recognition motif similar to the iCasps and apply these new tools to elucidate hundreds of putative iCasp substrates. Our findings reveal insights into inflammasome function.

Abstract

Inflammasomes sense a number of pathogen and host damage signals to initiate a signaling cascade that triggers inflammatory cell death, termed pyroptosis. The inflammatory caspases (1/4/5/11) are the key effectors of this process through cleavage and activation of the pore-forming protein gasdermin D. Caspase-1 also activates proinflammatory interleukins, IL-1β and IL-18, via proteolysis. However, compared to the well-studied apoptotic caspases, the identity of substrates and therefore biological functions of the inflammatory caspases remain limited. Here, we construct, validate, and apply an antibody toolset for direct detection of neo-C termini generated by inflammatory caspase proteolysis. By combining rabbit immune phage display with a set of degenerate and defined target peptides, we discovered two monoclonal antibodies that bind peptides with a similar degenerate recognition motif as the inflammatory caspases without recognizing the canonical apoptotic caspase recognition motif. Crystal structure analyses revealed the molecular basis of this strong yet paradoxical degenerate mode of peptide recognition. One antibody selectively immunoprecipitated cleaved forms of known and unknown inflammatory caspase substrates, allowing the identification of over 300 putative substrates of the caspase-4 noncanonical inflammasome, including caspase-7. This dataset will provide a path toward developing blood-based biomarkers of inflammasome activation. Overall, our study establishes tools to discover and detect inflammatory caspase substrates and functions, provides a workflow for designing antibody reagents to study cell signaling, and extends the growing evidence of biological cross talk between the apoptotic and inflammatory caspases.

Inflammasomes, cytosolic multicomponent complexes, sense patterns of pathogenesis or metabolic changes and initiate a cellular response via the activation of the inflammatory caspases (iCasps) (1–3). Consequently, the inflammasome plays a key role in protecting the host from pathogens, but aberrant activation can also lead to conditions such as gout, diabetes, and liver disease (4–6). Humans express three iCasps (1/4/5), whereas mice express two iCasps (1/11). Multiple different types of inflammasome complexes exist, which have been broadly classified as canonical or noncanonical. Upon detection of various microbial or endogenous stimuli by sensor proteins, canonical inflammasomes assemble and activate caspase-1, leading to cell death and production of proinflammatory cytokines (e.g., IL-1β and IL-18). More recently, the noncanonical inflammasome with distinct features from the canonical inflammasome was discovered. Here, cytosolic lipopolysaccharides (LPS) or oxidized phospholipids directly bind caspase-4/5/11, leading to oligomerization, caspase activation, and inflammatory cell death called pyroptosis (7–10).

A landmark advance in the inflammasome field was the discovery that pyroptosis is driven by inflammatory caspase-mediated cleavage of gasdermin D (GSDMD) (11–13). Cleavage of GSDMD relieves an autoinhibited state, leading to assembly of a multimeric GSDMD pore in the plasma membrane and release of cytoplasmic molecules (14, 15). While the initial discovery of the noncanonical inflammasome occurred in macrophages, many nonmyeloid cells, such as endothelial and epithelial cells, express both caspase-11 and GSDMD, indicating that this essential pathway may have additional functions. For example, caspase-11 has been shown to control cytoplasmic bacterial growth independent of pyroptosis (16). In another instance, caspase-11 confers protection in a model of inflammatory bowel disease because of activity in both the myeloid and nonmyeloid compartments (17, 18). Notably, several cell types such as mast cells and neurons contain very low or no GSDMD, which would prevent pyroptosis in response to pathway activation (19). However, lack of a comprehensive substrate analysis for the noncanonical inflammasome, beyond GSDMD and a few other proteins, limits our understanding of this pathway (20, 21). Additionally, a large dataset of substrates could be leveraged to explore these cleaved substrates as blood-based biomarkers of inflammasome activation in various pathological settings. Thus, we set out to address this technical challenge by developing a workflow with an antibody toolset to specifically detect cleavage products generated by the inflammatory caspases.

Caspases selectively cleave substrates at primary sequence motifs (P4-P3-P2-P1) that contain an Asp at P1. While caspases tolerate significant sequence diversity at P2 + P3, apoptotic caspases (3/6/7) prefer substrates with a P4 aspartic acid (e.g., DxxD), whereas iCasps prefer substrates with a hydrophobic P4 residue (e.g., W/IxxD) (22, 23). Recent work has shown that iCasps possess the ability to recognize and cleave GSDMD independent of the exact cleavage site sequence through an exosite interaction (24, 25). However, since the iCasps exhibit robust activity against the canonical WEHD tetrapeptide, other substrates likely exist. In the last decade, a series of mass spectrometry-based methods have been developed to identify substrates of a given protease, which have enabled the discovery of a small set of caspase-1 substrates (26–29). Strategies that employ direct positive enrichment for newly generated N or C termini within a substrate offer the potential to identify proteolysis events with low abundance and/or stoichiometry while simultaneously mapping both the protein identity and cleavage site (29, 30). Monoclonal antibodies (mAbs), which are essential tools to studying signaling pathways, typically exhibit high affinity and specificity for a single modification site, in direct contrast to a protease that exhibits a more degenerate recognition profile. Several groups have reported strategies to generate anti-posttranslational modification mAbs or mixtures of mAbs with degenerate specificities similar to those of the enzymes that catalyze the modification (30–32). However, these mAbs often require additional sequence features beyond the modified residue, and the structural basis for their unique recognition precludes further optimization.

Here, we set out to generate mAbs that exhibit a degenerate recognition motif similar to the iCasps and therefore enable the study of these enzymes and their substrates in cells and tissues. Critically, such a reagent must avoid recognition of peptides with a P4 Asp for several reasons; activation of apoptotic caspases often occurs at some basal level in cell culture or can occur downstream of canonical inflammasome activation. This background pool of proteolysis products with P4 Asp can greatly exceed the small pool of iCasp products and hinder their identification. Furthermore, a previously reported caspase motif mAb mixture fails to recognize most iCasp products (30). To overcome these technical challenges, we first developed a rabbit immune phage strategy that combined degenerate peptides with a small set of peptides from known substrates to yield two mAbs with the desired recognition profile. Using a panel of biochemical and cell-based assays, we demonstrate that these mAbs, called CJ2 and CJ11, selectively detect the C-terminal cleavage products generated by iCasps (i.e., P4 is hydrophobic) but not the apoptotic caspases (i.e., P4 is Asp). We structurally characterized one such mAb, CJ11, in complex with the liberated C-terminal peptides from IL1β and IL-18 to reveal the basis for an unusual highly selective but degenerate substrate recognition mechanism. Using an immunoprecipitation followed by mass spectrometry (IP–MS) strategy, we leverage this tool mAb to identify hundreds of putative substrates of the noncanonical inflammasome and reveal key functions of this complex in activating caspase-7 as well as targeting components of the spliceosome. Our work sheds light on the potential functions of the noncanonical inflammasome and provides a tool for identifying key substrates of the inflammatory caspases in healthy and disease contexts.

Results

Generation of pAb with Pan-iCasp Substrate Specificity.

Our goal was to generate an antibody with degenerate recognition similar to that of the iCasps (1/4/5/11). We first precisely designed target and decoy peptides using the experimentally determined recognition motifs for the iCasps and apoptotic caspases (3/6/7) (22, 23, 33). We synthesized two peptide libraries as target antigens: W/YxxD and I/LxxD, where P3 was an equimolar mixture of E, V, and Q and P2 was an equimolar mixture of H, S, and T (Fig. 1A).

Fig. 1.

Overview of rabbit immunization strategy to generate antibody with inflammatory caspase-like specificity. (A) Design of two different degenerate peptide libraries. (B) Sequences of C-terminal peptide products in known inflammatory caspase substrates generated upon proteolysis. (C) ELISA results with purified pAbs shows specific recognition of WxxD and/or IxxD targets with minimal cross-reactivity to DxxD control (n = 3 with error bars representing SD). (D) ELISA results with purified pAbs demonstrate strong recognition of five peptides corresponding to proteolysis products generated in known inflammatory caspase substrates.

We also synthesized two control peptide libraries in which P2 and P3 had the same degeneracy, but either the C-terminal carboxylate was capped with an amide (WxxD-NH₂ or IxxD-NH₂) or the P4 position was changed to D (DxxD). These peptide libraries were synthesized using a split and mix method, and Edman sequencing confirmed the desired degeneracy of each library. Given the robust ability of rabbits to generate anti-peptide antibodies, we then immunized rabbits with the target peptide libraries (34).

Purified polyclonal sera (pAbs) from each immunized rabbit were characterized by enzyme-linked immunosorbent assay (ELISA) against the peptide libraries and peptides from known caspase-1/11 substrates (GSDMD, IL-1β, and IL-18) to identify the best rabbit(s) (Fig. 1 A and B). Each pAb showed a preferential response to the target WxxD or IxxD peptide libraries but weak to no binding to the DxxD control (Fig. 1C). Several of the IxxD-immunized rabbits also bound the WxxD library. Several of the rabbit pAbs (rabbit 37 and 39) showed some binding to DxxD, indicating that stringent selections during the subsequent mAb generation would be required to remove such antibodies. Consistent with the results against the degenerate peptide pools, all pAbs showed strong binding to all five human and mouse substrates (Fig. 1D). As a final and more stringent screen, we performed Western blot analysis with these pAbs on lysates from mouse bone marrow derived macrophages (BMDMs) that were stimulated with two different stimuli (LPS/cholera toxin B or adenosine triphosphate) designed to activate the noncanonical and canonical inflammasomes (8). Encouragingly, each pAb was able to detect at least one and often multiple unique bands that were unique to lysates of stimulated BMDMs (SI Appendix, Fig. S1). Combined, our results indicate that each immunized rabbit contains a mixture of individual mAbs with a range of specificities.

Discovery and Characterization of Pan-iCasp Substrate mAb.

Given the diversity of antigens used for immunization, we hypothesized that our desired mAbs with iCasp specificity would be quite rare within the antibody pool. We therefore designed a rabbit immune phage library strategy to select for mAbs with this specificity using both W/IxxD peptide pools and individual product peptides from known iCasp substrates (Fig. 2A).

Fig. 2.

Discovery of monoclonal antibodies with inflammatory caspase-like specificity. (A) Schematic of rabbit immune phage workflow. (B) Both CJ2 and CJ11 exhibit selective binding to WxxD and IxxD targets with no binding to DxxD as detected by ELISA. (C) ELISA results show that CJ11 exhibits broader recognition of substrate proteolysis products than CJ2. (D) Addition of a single residue after the C-terminal Asp to product peptides eliminates binding to CJ2 and CJ11 by ELISA. (E) Peptide specificity profile of CJ11 shows strong preference for hydrophobic residues at P4 and strict recognition of Asp at P1 with some degeneracy at P3 and P2. All ELISAs were performed in triplicate, with errors bars representing SD.

Multiple phage panning tracks were performed along with stringent counterselections against the DxxD control. After phage ELISA screens to confirm specificity, we generated IgGs for 14 unique clones for further characterization. Interestingly, only two out of the 14 IgGs (CJ2 and CJ11) showed the desired P1 and P4 specificity and exhibited similar binding to both WxxD and IxxD pools by ELISA (Fig. 2B). Importantly, the mAbs showed no binding to the control peptide pools (DxxD and W/IxxD-NH₂). We next characterized binding to the cleavage products from human and murine GSDMD, IL-1β, IL-18, and caspase-11 (Fig. 2 C and D). For IL-1β, we evaluated both the canonical (B) and noncanonical (A) cleavage products as defined in Fig. 1B. Impressively, CJ11 showed strong binding to all of these peptides, whereas CJ2 only exhibited binding to a subset of the targets (Fig. 2C). The addition of a single amino acid after the P1 Asp to either the GSDMD or caspase-11 peptides abolished or greatly reduced binding of both CJ2 and CJ11, indicating that direct recognition of the C-terminal carboxylate was required for high affinity binding (Fig. 2D).

We next performed a peptide profiling experiment to reveal the preferred recognition motifs of CJ2 and CJ11. We used phage display selections against both IgGs using two degenerate peptide libraries (X₁₂-COOH and X₉D-COOH, where X is any of the 20 amino acids) (35). In agreement with our ELISA data, both CJ11 and CJ2 only tolerated Asp at P1 and did not show recovery of any peptide with a P4 Asp (Fig. 2E and SI Appendix, Fig. S2). When profiled against the more diverse X₁₂ library, CJ11 enriched for peptides with predominantly hydrophobic residues at P4 with more restrictive recognition at P3 (Glu/Gln) and P2 (Ser/Thr). Interestingly, only peptides with a P1 Asp were recovered, suggesting that CJ2 and CJ11 do not recognize peptides with a highly similar P1 Glu. The X₉D library was used to more deeply profile mAb recognition within caspase cleavage products. Here, CJ11 exhibited similar strong hydrophobic recognition at P4 with expanded diversity at P3. Thus, we successfully isolated two mAbs (CJ2 and CJ11) with broad recognition for known iCasp substrates that require both a free C terminus Asp at P1 and hydrophobic residue at P4 for efficient binding.

Structural Basis of Degenerate Recognition by CJ11.

To illuminate the basis of the unusual specificity of CJ11, we determined cocrystal structures of its Fab with peptides representing the liberated C termini of mouse IL-1β (LFFEVD) and IL-18 (GDLESD) to 2.0 and 3.0 Å resolution, respectively (SI Appendix, Table S1). In both CJ11-IL-1β and IL-18 complexes, CJ11 targets the free C-terminal carboxylic and aspartic acid groups while simultaneously forming an array of main-chain interactions along the peptides (Fig. 3 and SI Appendix, Fig. S3).

Fig. 3.

Structural basis of degenerate peptide recognition by CJ11. (A) View of the IL-1β + CJ11 complex. The mouse IL-1β peptide (₂₁LFFEVD₂₆) is shown in stick representation (orange) and the light chain (LC) and heavy chain (HC) of CJ11 are shown in gray and cyan cartoon representation, respectively. (B) Electrostatic surface representation of CJ11 with IL-1β peptide shown in stick representation. (C) Close-in view of IL-1β Asp26 interaction with CJ11. Hydrogen bond and ionic interactions are shown by dotted lines (black) and water molecules are shown as red spheres. Residues indicated by an asterisk (*) indicate an interaction with the protein backbone. (D) View of interactions across the IL-1β + CJ11 complex. (E) Superposition of the IL-18 + CJ11 complex with the IL-1β + CJ11 complex, with the IL18 peptide (₃₀GDLESD₃₅) residues labeled and shown in purple stick representation.

These observations begin to rationalize how CJ11 can paradoxically achieve high selectivity in the context of a degenerate consensus motif. Below, we focus on the CJ11-IL-1β complex structure in detail, since it is of highest resolution.

The IL-1β ₂₁LFFEVD₂₆ motif adopts an “L-shape” that docks onto a strong electropositive surface patch at the intersection between the heavy chain (HC) and light chain (LC) of CJ11 (Fig. 3 A and B and SI Appendix, S3 A and B). The free acidic terminus of IL-1β, Asp26, directly engages the backbone amides of HC-Tyr98 and HC-Thr99 from complementarity-determining region 3 (CDR3), while the Thr99 hydroxyl also forms a hydrogen-bonding interaction (Fig. 3C). These close-fitting interactions explain why an extension of, or modification to, the carboxyl terminus of the peptide is not compatible with CJ11 binding. Orthogonally, the acid side chain of Asp26 forms a tight ionic interaction with the guanidino group of HC-Arg95 and also binds to the HC-Thr100 side chain and HC-Ala32 backbone amide through water-mediated interactions (Fig. 3C). Three structural observations suggest why an aspartic acid, and not a glutamic acid, is so strongly preferred by CJ11. First, Asp26 forms a direct hydrogen bond to its own amide backbone to stabilize and orient the side chain for productive CJ11 binding (Fig. 3C), whereas a similar coordination scheme would be geometrically disfavored by a longer glutamic acid. Second, HC-Arg95 of CJ11 is itself intimately coordinated to the Fab scaffold through a surrounding water network, suggesting that the guanidine group is optimally prepositioned to bind an aspartic acidic (Fig. 3C). Third, the overall snug fit of both the carboxylic acid and side chain of Asp26 implies a precise lock-and-key recognition event that would sterically preclude the larger glutamic acid residue from binding (Fig. 3 A–D). Thus, CJ11 organizes a multipoint network to form six interactions with Asp26 of IL-1β through a distinctive arrangement that rationalizes its high selectivity to bind terminal aspartic acid residues (Fig. 3C).

To optimally position the terminal Asp26 of IL-1β for binding, CJ11 coordinates the five preceding residues primarily by exploiting the peptide backbone (Fig. 3D). The amide and carbonyl of Leu21 are both bound directly by LC-Asn31 (CDR1), the Phe22 carbonyl and Phe23 amide are coordinated to LC-Asn97 (CDR3) through water molecules, the Glu24 carbonyl interacts with the backbone amide of HC-Ser52 (CDR2), and the Val25 carbonyl hydrogen bonds to the hydroxyl of LC-Try34 (CDR1) (Fig. 3D). This elaborate backbone coordination scheme most likely underlies the degenerate nature of the CJ11 consensus binding motif. In fact, Leu21 and Phe22 side chains are completely solvent exposed, indicating degeneracy at these positions and rationalizing why Gly30 and Asp31 from IL-18 (₃₀GDLESD₃₅) are permissive for CJ11 binding (Fig. 3 D and E and SI Appendix, Fig. S3 C and D). The IL-1β Phe23 and Val25 side chains bind into a small or large polar cleft on CJ11, respectively, but are not strict binding determinants because Leu32 and Ser34 within the IL-18-CJ11 complex highlight the permissive nature of these side-chain docking clefts on CJ11 (Fig. 3E and SI Appendix, Fig. S3 B and D). Notably, relative to Phe23 on IL-1β, Leu32 of IL-18 packs against CJ11 in a manner that enforces a distinct local backbone geometry, which serves to direct Asp31 and Gly30 toward solvent (Fig. 3E). Beyond Asp26, Glu24 is the only side chain from IL-1β to interact specifically with CJ11 and that makes hydrogen bonds to the backbone amides of HC-Gly54 and HC-Gly55 (CDR2) (Fig. 3D). However, the Glu24 side chain is also largely solvent exposed, which rationalizes the lack of strict consensus for CJ11 binding but the preference for a glutamic acid or glutamine at this position. While we used the noncanonical IL-1β ₂₁LFFEVD₂₆ cleavage site for our structural analysis, structural modeling using the canonical ₁₁₁EAYVHD₁₁₆ cleavage site confirms that this peptide will also be capable of binding CJ11 as shown in Fig. 2. The terminal Asp is completely conserved between both peptides. The change from Val to His at P2 would maintain the backbone recognition, and the side chain is readily accommodated by the surrounding polar environment. The change from Glu to Val at P3 is also tolerated since it is partially solvent exposed, and the Val would also make favorable Van der Waals interactions. The other changes at P4 to P6 occur at highly exposed positions, which enable degenerate recognition. Overall, our structural analysis has permitted a molecular level dissection of the unique coordination strategy that CJ11 Fab employs to selectively recognize degenerate peptide motifs that terminate with a free aspartic acid.

CJ11 Enables Selective Detection of iCasp Substrates in Cells.

To validate the ability of our antibodies to detect cleaved iCasp substrates within cell lysates, we designed several experiments. We focused only on CJ11 because of the broader specificity profile. We first evaluated the ability of CJ11 to selectively immunoprecipitate the cleaved form compared to the full-length form of four known substrates: IL-1β, IL-18, caspase-11, and GSDMD. FLAG-tagged constructs encoding for either the full-length or N-terminal fragment of the human proteins were expressed in HEK293 cells. Cells were then lysed, and immunoprecipitations were performed with either CJ11 or anti-FLAG mAbs followed by anti-FLAG Westerns for detection. Strikingly, CJ11 selectively immunoprecipitated only the cleaved forms of IL-18, caspase-11, and GSDMD (Fig. 4A).

Fig. 4.

Selective immunoprecipitation of cleaved forms of inflammatory caspase substrates with CJ11. (A) Anti-FLAG immunoblots of CJ11 or anti-FLAG immunoprecipitation from HEK293 lysates that overexpress full-length (F) or cleaved (Cl) forms of four caspase substrates. CJ11 selectively pulls down only the cleaved form of three substrates. (B) CJ11 immunoblots of CJ11 immunoprecipitation from immortalized mouse macrophages. CJ11 pulls down multiple unique bands upon stimulation with intracellular LPS, and these unique bands are not detected in the absence of Casp1.

Since caspase cleavage of IL-18 removes only a small fragment from IL-18, both the cleaved and full-length forms migrate with similar sizes. We failed to detect any cleaved IL-1β, potentially because of low levels of expression.

We next moved to an experiment with immortalized mouse macrophages to gain a more complete and unbiased view of antibody performance (8). Briefly, wild-type or Casp1 KO (knockout) macrophages were stimulated with cytosolic LPS to solely focus on caspase-11 substrates. Following cell lysis, we used CJ11 to immunoprecipitate proteins and detected the proteins via Western blot with CJ11. We detected very few bands in the absence of LPS or Casp1, establishing that CJ11 exhibits very low background enrichment. Quite strikingly, CJ11 enriched for many unique bands in wild-type macrophages treated with LPS (Fig. 4B). These results highlight several features of CJ11. First, even in a complex cell lysate with endogenous C termini and presumably some cleavage products from apoptotic caspases because of a low level of apoptosis, CJ11 exhibits a very high selectively for iCasp substrates. Second, activation of the inflammasome results in the cleavage of a multitude of substrates, which can be efficiently enriched by CJ11 immunoprecipitations.

Discovery of Putative Noncanonical Inflammasome Substrates.

Based on our validation of CJ11, we sought to leverage this reagent to identify substrates of the noncanonical inflammasome and reveal potential functions beyond GSDMD cleavage-induced pyroptosis. Caspase-11 in endothelial cells mediates lethal septic shock. Thus, we elected human EA.hy926 endothelial cells as a physiologically relevant model for studying caspase-4, the human equivalent of murine caspase-11. EA.hy926 cells express very little RNA for caspase-5 and inflammasome sensors, NLRP3 and NLRC4, that are essential for caspase-1 activation (11). We stimulated human EA.hy926 cells with intracellular LPS to activate the caspase-4 noncanonical inflammasome (SI Appendix, Fig. S4). Additionally, this setup enabled us to solely focus on substrates of the noncanonical inflammasome since EA.hy926 cells are unable to trigger activation of caspase-1 downstream of caspase-4 activation due to absence of NLRP3 sensor protein. We collected cell pellets and matched supernatants 1 h poststimulation. We then performed immunoprecipitations with CJ11 followed by mass spectrometry analysis.

From both the cell lysates and matched supernatants, we identified 4,220 unique peptides (Dataset S1). We filtered the resulting list of peptides to identify those peptides and corresponding proteins that were more than twofold enriched upon LPS simulation (Fig. 5A), identifying 406 peptides derived from 328 proteins (Dataset S2).

Fig. 5.

Discovery of noncanonical inflammasome substrates with CJ11-based IP–MS. (A) Volcano plot showing quantitative caspase-4 substrates in EA.hy926 cells. Peptides identified in the lysate and supernatant are colored blue and yellow, respectively. Fold enrichment in LPS-treated samples is relative to untreated samples that were also subject to CJ11-base IP–MS. (B) Sequence logo from peptides enriched at least twofold upon LPS treatment. Amino acids after the Asp (labeled 1 to 6) correspond to the sequence from the native full-length protein. (C) Portion of gene oncology analysis on caspase-4 substrates. (D) Substrate interaction network for spliceosome components that were cleaved. (E) Western blot analysis confirms that both GSDMD and caspase-7 are cleaved upon activation of caspase-4 but not upon CRISPR/Cas9 removal of CASP4. Full-length GSDMD and caspase-7 are indicated with black arrows, and cleaved forms are indicated with open arrows.

The sequence motif for these peptides contained only D at P1 and a high prevalence of hydrophobic residues (L, V, I, F, and M) at P4, confirming that a majority of these proteolysis events are due to iCasp activity (Fig. 5B). Interestingly, proline was the third most abundant amino acid at P4, suggesting that human caspase-4, like mouse caspase-11, possesses a distinct substrate specificity from caspase-1 (33). Furthermore, CJ11 enriched for peptides with significant diversity at P3 and H/S/T at P2, which confirms previous in vitro work on caspase-4 specificity (22).

Overall, only a few of these putative substrates were previously identified caspase-1 substrates (HNRPK, TIF1B, MCM3, and GSDMD), with the remainder representing new iCasp substrates (Dataset S3) (26, 27). Beyond differences in caspase, this result suggests that differences in cell type or inflammasome complex (e.g., canonical versus noncanonical) lead to cleavage of unique substrate pools. We performed gene ontology enrichment analyses using Protein ANalysis THrough Evolutionary Relationships to better understand the cellular processes affected by caspase-4. Processes such as RNA splicing, regulation of gene silencing, and Golgi to endosome transport were enriched alongside cell compartments such as the spliceosomal complex (Fig. 5C and Dataset S4). A majority of the proteins involved in RNA splicing physically interact and are components of the spliceosome, suggesting that the caspase-4 noncanonical inflammasome alters or inhibits RNA splicing (Fig. 5D).

Notably, we found that caspase-7 was cleaved under our experimental conditions, suggesting potential cross talk between the noncanonical inflammasome and an apoptotic caspase. To confirm that this proteolysis event was due to caspase-4, we used CRISPR/Cas9 to knockout CASP4 in EA.hy926 cells (SI Appendix, Fig. S5). Upon stimulation with LPS, we observed that cleavage of both GSDMD and caspase-7 occurred in the wild-type EA.hy926, but no cleavage could be detected in the absence of caspase-4 (Fig. 5E). Furthermore, cleavage at this Asp198 is known to activate caspase-7. Thus, we have discovered that activation of the noncanonical inflammasome can result in downstream activation of caspase-7 and likely eventual apoptosis in the absence of GSDMD-mediated pyroptosis. This finding agrees well with previous work on the canonical inflammasome and a caspase-11 genetic study (19, 36). Since activation of caspase-7 occurs in this context, it is possible that some of the putative noncanonical inflammasome substrates are in fact cleaved by caspase-7 at a site in which P4 is not Asp. To evaluate whether CJ11 is capable of detecting such products, we performed immunoprecipitation-Westerns using CJ11 and lysates from EA.hy926 cells stimulated with TNF-related apoptosis-inducing ligand, a potent activator of the extrinsic apoptotic pathway for 2 or 6 h. We observed several cleavage products detected by CJ11, indicating that some substrates of the apoptotic caspases can be detected (SI Appendix, Fig. S6). Given the short time scale of our LPS stimulation experiments (1 h) and that activation of caspase-7 would occur after caspase-4 activation, we anticipate such substrates would be of lower abundance compared to those substrates of the noncanonical inflammasome. Ultimately, future studies will be required to fully confirm each of the putative substrates of the noncanonical inflammasome. Overall, these results validate the utility of our iCasp motif mAb as an important tool to study the function of the noncanonical inflammasome and the iCasps in general.

Discussion

The predominant function of the noncanonical inflammasome is proteolytic activation of GSDMD to initiate pyroptotic cell death in response to intracellular pathogens or other damage signals. However, other functions likely exist as suggested by the protective phenotype observed in inflammatory bowel disease models and highlighted by the fact that several cell types such as neurons or mast cells express caspase-4/5/11 and only low levels of GSDMD (19). Given that cellular localization and substrate concentration can dictate substrates of a particular protease, it is essential to employ strategies that identify protease substrates in a cellular context. Here, we developed an antibody toolset capable of selectively detecting cleavage products generated by the iCasps by both Western blot and IP–MS. These tools will be essential to inflammasome and caspase fields to further elucidate their biological functions.

Interestingly, the similar recognition motif of a mAb and the inflammatory caspases is achieved via distinct modes of molecule recognition. For caspase-1/4/11, the P1 Asp lies buried in a pocket comprised of electrostatic contacts with Arg179 and Arg341 and a hydrogen bond with Gln283 (SI Appendix, Fig. S7). For CJ11, the P1 Asp lies in a pocket composed entirely of residues from CDRH3 that recognize both the C-terminal carboxylate through main-chain amines in Tyr97 and Thr98 and the Asp side chain via ionic interaction with Arg95. By analogy to the caspase mode of recognition, we anticipate mutation of HC.Thr99 or HC.Tyr32 to Arg could provide enhanced P1 Asp recognition. Although the P2 side chain is surface exposed upon binding to the caspase, the mAb partially buries the P2 side chain in a pocket containing HC.Ile50 and LC.Tyr93, where these two bulky residues sterically block recognition of peptides that contain larger side chains at the P2 position. Finally, while the exact molecular basis for lack of P4 Asp/Glu recognition by CJ11 remains unclear, we expect that a P4 Asp would be partially solvent exposed and not have steric clash or charge repulsion with the mAb. However, a side-chain carboxylate at P4 might offend the adjacent carbonyl of P5, which is tightly engaged by LC.Asn31 (Fig. 3D).

Given the limited number of reported substrates of the noncanonical inflammasome, we focused our efforts on elucidating putative caspase-4 substrates in human endothelial cells. In total, IP–MS experiments with our mAb revealed over 300 proteins that were cleaved upon noncanonical inflammasome activation with minimal overlap with existing and smaller iCasp datasets (26, 27). We hypothesize that the substrates could be unique to caspase-4 and/or human endothelial cells. Furthermore, our method directly enriches for the iCasp cleavage products and therefore could detect proteolytic events of low abundance and/or stoichiometry. At a global level, the overall sequence motif agrees with the canonical caspase-4 cleavage motif, but we note the presence of a set of substrates with a hydrophobic residue (e.g., Val/Leu) at P3 (Fig. 5B). A recent study found that the iCasps bind and cleave GSDMD in a manner independent of the exact cleavage site sequence (24). However, while caspase-4 was unable to cleave a NFLTD-AFC probe that corresponds to the GSDMD cleavage site, caspase-4 was readily able to cleave a WEHD-AFC probe, indicating that not all substrates require this exosite. Additionally, upon binding of GSDMD into the exosite, caspase-4 can now readily cleave the NFLTD-AFC probe, indicating that GSDMD binding increases the catalytic activity of caspase-4. This more active form of caspase-4 could also be capable of cleaving additional substrates with less preferred cleavage site sequences. In analyzing our list of substrates, we observed multiple proteins (e.g., SRRT, SEPT2, TMED8, CNOT1, and CCT3) that have tetrapeptide cleavage sites identical or closely related to that of GSDMD, suggesting that these proteins either also bind caspase-4 via an exosite or only undergo caspase-4–mediated proteolysis after the formation of the more active GSDMD:caspase-4 complex. Future studies are needed to identify the full range of substrates that fall into these categories.

Cross talk between forms of cell death, such as apoptosis, necroptosis, and pyroptosis, is an emerging theme. We found that caspase-7 is cleaved and activated upon activation of the caspase-4 noncanonical inflammasome. While pyroptosis is predicted to occur rapidly after GSDMD activation by iCasps, multiple conditions exist in which GSDMD processing is limited or fails to occur altogether. First, cells such as neurons and mast cells express little to no GSDMD and upon activation of canonical inflammasome undergo apoptosis because of caspase-7 activation (19). As distinct stimuli would activate the noncanonical inflammasome in these cells, cleavage of caspase-7 by caspase-4 provides a fail-safe mechanism to ensure removal of these cells via apoptosis. Second, GSDMD function can be inhibited due to genetic polymorphisms (37). Third, GSDMD can be inhibited by pathogen effector proteins, such as the enteroviral protease 3C (38). Finally, the recent report of a small molecule inhibitor of GSDMD pore formation will make understanding the consequences of inflammasome activation in the absence of GSDMD-mediated pyroptosis even more critical (39).

Upon inflammasome-mediated cleavage and activation of GSDMD, GSDMD forms pores in the plasma membrane, which enables release of cytoplasmic molecules. Indeed, we detected 96 peptides released into the supernatant only 1 h post-LPS stimulation (Dataset S2). These peptides could serve novel blood-based biomarkers of inflammasome activation and, by extension, pyroptosis. Similarly, previous work on the apoptotic caspases has shown that caspase cleavage products can be detected in patient sera postchemotherapy (40). Collectively, our work describes an antibody toolset for detecting iCasp substrates and reveals over 300 noncanonical inflammasome substrates, including caspase-7, enabling insights into the biological function of this complex.

Materials and Methods

Peptide immunizations were performed in rabbits. Immune phage libraries were generated from the immunized rabbits, and mAbs were generated using in vitro phage display selections. Antibodies were characterized using a panel of biochemical and cell-based experiments. Proteomics experiments to identify caspase-4 substrates were performed in human EA.hy926 cells using CJ11 for affinity enrichment of peptides from cleaved caspase substrates. Additional details for the material and methods described in this paper are provided in SI Appendix.

Data Availability

Crystal structure data have been deposited in the Protein Data Bank (7JWP, 7JWQ).