Mechanism of filament formation in UPA-promoted CARD8 and NLRP1 inflammasomes

NLRP1 and CARD8 are related cytosolic sensors that upon activation form supramolecular signalling complexes known as canonical inflammasomes, resulting in caspase−1 activation, cytokine maturation and/or pyroptotic cell death. NLRP1 and CARD8 use their C-terminal (CT) fragments containing a caspase recruitment domain (CARD) and the UPA (conserved in UNC5, PIDD, and ankyrins) subdomain for self-oligomerization, which in turn form the platform to recruit the inflammasome adaptor ASC (apoptosis-associated speck-like protein containing a CARD) or caspase-1, respectively. Here, we report cryo-EM structures of NLRP1-CT and CARD8-CT assemblies, in which the respective CARDs form central helical filaments that are promoted by oligomerized, but flexibly linked, UPAs surrounding the filaments. Through biochemical and cellular approaches, we demonstrate that the UPA itself reduces the threshold needed for NLRP1-CT and CARD8-CT filament formation and signalling. Structural analyses provide insights on the mode of ASC recruitment by NLRP1-CT and the contrasting direct recruitment of caspase-1 by CARD8-CT. We also discover that subunits in the central NLRP1CARD filament dimerize with additional exterior CARDs, which roughly doubles its thickness and is unique among all known CARD filaments. Finally, we engineer and determine the structure of an ASCCARD–caspase-1CARD octamer, which suggests that ASC uses opposing surfaces for NLRP1, versus caspase-1, recruitment. Together these structures capture the architecture and specificity of the active NLRP1 and CARD8 inflammasomes in addition to key heteromeric CARD-CARD interactions governing inflammasome signalling.

and Caspase-1 without giving any indication of the experimental procedures. The level of detail they ascribe to their analyses -namely amino acid level specificity -represents a drastic overreach of the evidence they present. This is particularly the case when they decide between model orientations "by visual inspection of the charge complementarity". Again, their analysis is plausible, but wholey unsupported by structural evidence or documented procedures. I strongly recommend aiming for a higher level of rigour in their analysis and presentation.
On the whole, I enjoyed the concept of the paper. In particular, I thought the octamer structure to be revealing in determining orientation of binding. The EM data seem adequate, and, at first glance, appear to be superior in resolution to the related structures cited in a recent bioRxiv deposition. The topic is of interest, and is pertinent to the field of inflammasome biology. The figures are beautifully done, and clear in what they attempt to convey. However, I find the interpretation of the data lacking in rigour, and overreaching in its scope. I do not recommend publication of this manuscript in its present form, but I encourage the authors either to introduce experiments (with experimental details) that support their plausible claims, or formulate an interpretation of the existing data consistent with the conclusions that can be drawn from the evidence at hand.
Concerns and comments: Fig1 -The FSC plots in e) and h) are not at 1 at 10Å, even though you claim 3.5 and 3.7 Å resolution. This is not good. In general, the falloff of the curve in e) is strange, as the map-model correlation is better than the half-map correlation. How do you explain this? -You seem to be using a 0.5 criterion for the map-model. A more conservative crition seems to be warranted here given your FSC curves -0.7 is used in some literature and may be more appropriate. Page 5 -"One possible explanation is that CARD8UPA does not follow the CARD helical symmetry but is orderly associated with the central CARD8CARD filament." -Card8CARD and UPA are covalently linked. How can UPA be "orderly associated", but not visible in the average? Wouldn't that be disordered? Have you tried focused classification instead of just subtraction and asymmetric classification? Also, your nomenclature is confusing. It sounds like you have two different filaments: "no UPA density was observed in the CARD8-CT filament structure, in which only the CARD8CARD filament was visible" You have one filament here: CARD8-CT. It's made of a CARD and UPA -one you can see, one you can't. Fig 3 -It seems to me that the octamer in 3a) and the detailed interactions in 3c) are modeling exercises and are unsupported by structural data presented here. This is not very clear in the text or the figure caption. What is the evidence that ASC or Caspase-1 filaments will stack on top of Card8? Nothing of the sort is presented. Page 9 -"The functional role of UPA in inflammasome signalling is suggestive of its ability to dimerize or oligomerize such that its presence promotes CARD filament formation." -The data presented are not suggestive to me of any dimerisation of UPA (which is a feature also present in the schematic model in Fig4g). I see no structural or biochemical evidence presented here that corroborates the assertion of UPA dimerisation, though CARD dimerisation does seem to involve UPA presence. Page 9 -"Based on these data and analyses, we propose a model of UPA-induced NLRP1CARD filament formation in which flexibly linked UPA dimers or oligomers promote the intrinsic tendency of NLRP1CARD dimerization and its helical polymerization to mediate inflammasome formation (Fig. 4g)." -And yet you don't see UPA in either structure... can you hypothesize why? Also, the orientation of CARD dimerisation would necessitate an asymmetry in the UPA domain on the N-Terminal end of your constructs. How would these differences be bridged in your model? Page 10 -"To further elucidate the structural basis for the NLRP1−ASC interaction, we analysed the modelled interfaces type by type." -You do not describe how you arrive at this model. It does not seem to be based on structural data presented here, and, although plausible, remains speculative. You go on to give amino-acid level interactions for this model in the absence of structural data, or methods for how you generated it. This is not acceptable. Page 11 -"and that UPA oligomerization decreases the threshold for filament assembly by locally Page 4 -ASC and UPA still not defined. Page 5 -"By systematically optimizing cleavage conditions, we purified short (~100-200 nm) filaments that still contained some uncleaved MBP-tagged proteins." -I don't get it. You optimised cleavage conditions but still have MBP-tags in your filament? What were you optimising? Fig1 -1b) is great, and very helpful. Well done.
-Your 2D classes in 1d) are surrounded by a poorly-defined haze. Is this MBP? Fig2d -the detail images are labeled Type I etc, but this label is not in a consistent location making it hard to find. the colour of this text is also not coordinated with the interaction type colour in the overview. Alternatively, the border of the detail could be coloured.  5a -why "ASC or caspase-1" ? Which did you use? If the fit is the same, just say which you used and tell us the other fit is identical. If the point is only to show the orientation difference between a) and b), then be more explicit about that. Page 11 -"hieratical" => hierarchical ? Page 11 -"We posit that these CARD filaments are decorated by the flexibly linked UPA subdomains ..." Why are you positing this? Is there concern that UPA is not bound? Could it be cleaved? Do you have an antibody against the UPA region?
subdomain of FIIND. The molecular mechanism by which the active UPA-CARD fragments of 11 NLRP1 and CARD8 assemble the inflammasome complexes and selectively recruit ASC or 12 caspase-1 to mediate downstream signaling remains unresolved. In this manuscript,

13
Hollingsworth, David, Li, and colleagues employ a combined structural and biochemical approach 14 to investigate the structural mechanism governing NLRP1-and CARD8-mediated inflammasome 15 assembly and signaling.

16
They solved the cryo-EM structures of oligomeric UPA-CARD assembly of NLRP1 and 17 CARD8, in which both CARDs form central helical filaments, resembling other known CARD 18 filamentous structures, whereas the UPA is outside and flexibly linked to CARD filaments without 19 an ordered organization. The CARDs of NLRP1 further forms a thicker filament by a unique 20 dimerization interface. They also identify the structural basis in NLRP1 and CARD8 CARD 21 filamentous assembly which enable NLRP1 and CARD8 to discriminate between ASC and procaspase-1 for diverse signalings. Furthers structural characterization of ASC-CARD and caspase-23 1-CARD interaction by an engineered CARD-CARD fusion protein provide more insights for 24 understanding NLRP1-ASC-caspase-1 and CARD8-caspase-1 signaling. Overall, the 25 experiments are well performed, and structural data are solid, which provides valuable insights 26 into the molecular mechanism of NLRP1 and CARD8 inflammasome signaling, and has potential 27 to be of broad interest in the innate immune field.

28
Responses: We thank the reviewer for the positive comments.

30
However, there are still some gaps that need to be addressed in order to firmly establish the

46
NLRP1 or CARD8 UPA-CARD and CARD induced similar levels of LDH release ( Fig. 3f and g).
Finally, purified MBP-fused UPAs from NLRP1 and CARD8 were already oligomers even with the the UPA subdomains lower the threshold needed for inflammasome signaling due to their ability 2 to oligomerize, which is particularly important in physiological contexts when the endogenous 51 protein concentrations are low.

106
We agree that all potential interfaces should be modeled extensively, and while not 107 presented, we had inspected every permutation of the potential CARD8/ASC/caspase-1 and "impossible" interfaces have predicted steric and/or electrostatic clashes to justify why these 110 interactions do not occur. However, the CARD domains of ASC, caspase-1, CARD8 and NLRP1 111 are quite different in sequence despite having the same fold (Table below)      on modeled structural compatibility (Fig. 6a, c).

223
Of note, although not directly shown for the CARD8 and the NLRP1 system, unidirectional    calculation of a map at 3.9 Å with many discernable side chains (Fig. 5, below), particularly some 272 at protein-protein interfaces. The map-model FSC is now at a 4.1 Å resolution, much more similar 273 to the resolution indicated by the map-map FSC. Below we also share the fitting of ASC CARD and 274 caspase-1 CARD into the density (Supplementary Fig. 4, below).

284
There are three type I interactions, four type II interactions, and one type III interaction between

342
UPA interface abolished NLRP1 inflammasome signaling in cells while maintaining these results provide a strong premise for a bona fide functional role of UPA oligomerization in  the concentration of these dimer mutants in a cellular system and also investigated the ability of 377 the UPA-CARD mutants to form filaments (Supplementary Fig. 3a-b, below). While the Y1445A 378 mutant abolished filament formation, the other mutants resembled the wild-type UPA-CARD.

379
Given this new data, the reviewer's suggestion and that the dimerization surface of NLRP1 is only 380 conserved in mammalian NLRP1, but not in more remote NLRP1 or any CARD8, we toned down 381 discussion of the functional significance of NLRP1-CARD dimerization.

517
Detailed modelled CARD-CARD type I-III interactions between NLRP1 (green) and ASC (gold).

531
The data presented are not suggestive to me of any dimerisation of UPA (which is a feature also 532 present in the schematic model in Fig4g). I see no structural or biochemical evidence presented 533 here that corroborates the assertion of UPA dimerisation, though CARD dimerisation does seem 534 to involve UPA presence.

535
Responses: The reviewer has identified a key lack of experimental evidence in our original 536 submission for which we planned to amend in the revision. To investigate whether the UPA 537 domain facilitates assembly of the UPA-CARD inflammasome, we titrated NLRP1 and CARD8

538
UPA-CARD versus their CARD-alone counterparts for filament formation in vitro and LDH release 539 in cells, and examined UPA alone for its aggregation property (new Fig. 3 below). In vitro, 0.25 540 µM CARD8 or NLRP1 UPA-CARD ( Fig. 3b and d) induced filament formation; by contrast, only 541 at 15 µM CARD8 CARD formed filaments (Fig. 3a) and even at 30 µM, NLRP1 CARD failed to form 542 filaments (Fig. 3c)

545
NLRP1 or CARD8 UPA-CARD and CARD induced similar levels of LDH release ( Fig. 3f and g).

546
Finally, purified MBP-fused UPAs from NLRP1 and CARD8 were already oligomers even with the 547 large MBP tag that often prevents oligomerization (Fig. 3e). Thus, while not absolutely required,

574
Additionally, we recently released a pre-print of our human DPP9-NLRP1 complex structure

596
CARDs or those along the z-axis of the filament given the linker size. If the position of this flexibly 597 linked UPA dimer/oligomer was stochastic relative to the central CARD helix, it would completely 598 average out during cryo-EM image processing (Fig. 3h, above).

600
6. Comment: How would these differences be bridged in your model? Page 10 -"To further 601 elucidate the structural basis for the NLRP1−ASC interaction, we analysed the modelled 602 interfaces type by type." -You do not describe how you arrive at this model. It does not seem to go on to give amino-acid level interactions for this model in the absence of structural data, or 22 Responses: We apologize for any oversight in communicating our modelling. We now discuss 607 how we arrive at these models both in the Methods section and in the Results section.

Comment:
Page 11 -"and that UPA oligomerization decreases the threshold for filament the full names of these proteins, as they are not very helpful, and they are perhaps even you mean 4 Å resolution? Or 4 of each molecule? Rephrase.