Protein Aggregates and EC Activation in CABMR Biopsies. We initially surveyed CABMR biopsies, excluding patients with confounding conditions characterized by protein aggregates including primary amyloidosis and post-transplant lymphoproliferative disorder (PTLD, Supplementary Table 1).30 In immune-EM, we observed that certain adluminal cells contained filamentous inclusions enriched for C9 which appeared to be contained within intracellular vesicles (Fig. 1a). We did not detect filamentous inclusions in glomerular ECs or in control tissues from transplant patients undergoing routine surveillance biopsies.
Congo Red and thioflavin clinically identify protein aggregates in tissues, but these dyes also reportedly fluoresce in the presence of collagen fibrils.31,32 We indeed observed dye fluorescence in laminations in arterioles with allograft vasculopathy (AV, Supplementary Fig. 1a), sites well known to contain collagen,9 as well as morphologically fibrosed interstitial regions (Supplementary Fig. 1b). The concurrent presence of fibrosis precluded evaluation of putative protein aggregates at these sites. However, in 3 of 8 patients we surprisingly observed congophilic staining in adluminal cells lacking AV (Fig. 1b). These adluminal cells costained for C9 and thioflavin (arrows, Fig. 1c) but not collagen (arrows, Fig. 1d). C9 + Thioflavin + vessels moreover expressed VCAM-1 (Fig. 1e) at ~ 3- to 10-fold higher levels than TUNEL (Fig. 1f).
The C9 Component of MACs Forms Aggregates in ECs. We used PRA-treated HUVECs to mechanistically analyze patient-level findings. As previously observed,22 PRA-treated HUVECs did not show increased cell death (Fig. 2a). In kinetic immune-EM studies, PRA caused C9 to colocalize within large, perinuclear structures (Fig. 2b) developing filamentous inclusions (Fig. 2b). Contemporaneously, thioflavin fluorescence increased in both a dose- and time-dependent manner (Fig. 2c,d). Under I.F., thioflavin showed punctate staining colocalizing with C9 in large, perinuclear vesicles (Fig. 2e). Thioflavin but not type I collagen fluorescence increased with PRA (Fig. 2e,f), excluding confounding collagen staining as a cause for increased thioflavin signals. These results recapitulated ultrastructural and histologic findings in CABMR.
Following treatment, PRA generates alloAb, anaphylatoxins, and MACs. To identify culprit mediator(s) causing thioflavin fluorescence, we performed sera fractionation and recombination studies. We separated PRA sera into its IgG + and IgG- fractions and found that while these fractions showed only minimal effects individually, combining the IgG- and IgG + fractions significantly potentiated thioflavin fluorescence (Fig. 2g). This ruled out IgG or contaminant(s) as a principal cause for increased thioflavin staining and suggested a role for C’ activity.
To test C9 aggregation, we treated HUVECs with the IgG + fraction of PRA combined with C9-deficient reference sera. This permitted IgG binding, anaphylatoxin formation, and oligomerization of MAC proteins to form C5b-8 which has pore-forming capabilities33 but lacks C9. IgG-induced C’ activation with C9-deficient sera minimally increased thioflavin fluorescence compared to the IgG + fraction alone (Fig. 2h, lane 3 vs lane 4). Addition of C9, which alone showed no effects (lane 1) significantly rescued thioflavin staining (lane 4 vs lane 5). We separated PRA-treated HUVECs into soluble and insoluble fractions, and detected increased C9 within the SDS-insoluble pellet (Fig. 2i) that, unlike pools of C9 within SDS-soluble supernatants, became resistant to mild proteinase K (PK) digestion, together indicating generation of insoluble C9 (Fig. 2j).
To further verify C9 aggregates, we tandemly expressed elements within the TSP, LDLRA, and MACPF domains of C9 that form b-sheets (www.uniprot.org, prosite.expasy.org/scanprosite/), a conformation frequently adopted in protein aggregates. Each element, ranging from ~ 6–9 kD, was tandemly expressed intracellularly and separated by a flexible glycine linker to allow antiparallel b-sheet stacking which occurs upon native assembly of MACs (Fig. 2k). We linked these elements to FLAG as this tag is less prone to aggregation relative to GFP, RFP, and luciferase reporters.
Of elements tested, AA197-270, an element within the MACPF domain containing the majority of predicted IDRs, strongly induced high molecular weight aggregates in HEK293 cells (Fig. 2l). AA197-270 colocalized within large perinuclear punctae, phenocopying distributions of native C9 aggregates in PRA-treated HUVECs (Fig. 2m). In fractionation studies, AA197-270 aggregates exclusively appeared within the insoluble pellet (Fig. 2n). Original and uncropped Western blot films for Fig. 2 are shown in Supplementary Fig. 3. C9 may form aggregates.
Intracellular C9 Form Aggregates Within the Endolysosomal Pathway. We asked whether C9 aggregates formed at the cell surface and/or intracellularly. We previously found that MAC internalization required CME.23 Exploiting this, we pre-treated HUVECs with CME inhibitors, Dynasore and PitStop2, and tested effects on C9 aggregates. Both CME inhibitors increased C9 levels at the cell surface (Fig. 3a, arrows), and this significantly reduced thioflavin fluorescence in PRA-treated HUVECs. Pitstop2 significantly decreased insoluble C9:soluble C9 ratios (lane 2 vs lane 4, Fig. 3b), an effect phenocopied by siRNA vs dynamin-2 (DNM2), the EC-specific dynamin isoform whose membrane scission activity generates endosomes (Fig. 3c). Following MAC assembly, the majority of C9 aggregates form intracellularly.
MACs vigorously assemble under cell-free conditions at pH < 6.34,35 While C9 is unlikely to encounter such conditions within intravascular or interstitial space, intracellular endolysosomes routinely reach intraluminal pH 4.5–5.5 to regulate vesicular maturation and trafficking.36 Based on this, we tested whether C9 aggregates could form within the endolysosomal system. We resuspended C9 in buffers of varying pH and found that at pH ≤ 6.5, C9 began to form ring-like structures whose aggregation increased in direct proportion to buffer acidity (Fig. 3d) and protein concentration (Fig. 3e). Among MAC components, C9 showed the highest thioflavin fluorescence (Fig. 3f), altogether suggesting that endolysosomes favor C9 aggregation.
Rab5 GTPase activity regulates endolysosomal trafficking, and we asked if Rab5 activity was required for intracellular C9 aggregates. We stably transduced ECs with Rab5 DN (S43N) constructs where Rab5 is locked in an inactive GDP-bound conformation. Compared to Rab5 WT EC controls, Rab5 DN ECs lacked thioflavin staining (Fig. 3g). Rab5 activity mediates intraluminal acidification of Rab5 + vesicles, a key step required for endosome maturation,36 and blocking this process via buffer alkalinization with NH4Cl or bafilomycin, a V-ATPase inhibitor, significantly reduced thioflavin staining (Fig. 3h). In contrast, AA197-270 aggregates which bypass internalization pathways mediated by Rab5 were unaffected by these treatments (Supplementary Fig. 1c). Original and uncropped Western blot films for Fig. 3 are shown in Supplementary Fig. 4. C9 aggregates form within the endolysosomal system in a manner requiring Rab5 activity and endosome acidification.
C9 Becomes a Substrate for Aggrephagy. Aggrephagy is a form of selective macroautophagy enabling degradation of protein aggregates which might otherwise cause cytotoxic effects within cells.37 Proteins containing IDRs frequently become aggrephagic substrates,38 prompting us to test whether C9 became a substrate for aggrephagy. PRA upregulated markers of autophagic flux including LC3-II and P62 (Fig. 4a) and generated increased GFP + RFP + LC3B punctae in ‘traffic light’ HUVECs. (Fig. 4b). In co-immunoprecipitations (co-IPs), C9 became ubiquitinylated (Fig. 4c), the first step towards targeting proteins to autophagosomes. At ~ 2 hrs, C9 became contained within LC3B + membranes (Fig. 4d) displaying markers specific for aggresomes (Fig. 4e-g). Blocking autophagosome-lysosome fusion with chloroquine (CQ) caused gross enlargement of C9 + vesicles (Fig. 4h), indicating C9 trafficking through a macroautophagic pathway. We performed pulse-chase studies in PRA-treated HUVECs and found that siRNA depletion of aggrephagy mediators, ATG5 or ATG16L, increased C9 (Fig. 4i,j). To strengthen these data, we tested AA197-270 in HUVECs. AA197-270 (FLAG) colocalized with Thioflavin + Hsp70 + aggresomes (Fig. 4k), and AA197-270 cells showed high molecular weight aggregates and increased autophagic flux which became decreased with ATG5 siRNA (Fig. 4l). Original and uncropped Western blot films for Fig. 4 are shown in Supplementary Fig. 5. These data indicated that intracellular C9 becomes an aggrephagic substrate.
C9 Aggrephagy Activates NF-kB and Causes EC Activation. Aggresomes sequester and activate NF-κB proteins.39,40 We transduced HUVECs with an NF-kB luciferase reporter and performed a limited siRNA screen and found that siRNA against 33 of 52 genes annotated to ‘selective macroautophagy’ (KEGG) significantly inhibited NF-κB (Fig. 5a). Concurrent with this, a re-analysis of Western blots in Fig. 4i-j showed that siRNA against autophagy genes in PRA-treated HUVECs reduced phosphoP65 (pP65), and this concurrently reduced transcripts previously found to be NF-kB-dependent (Fig. 5b,c).21 Human ECs express sufficient costimulatory molecules to directly prime CD4 + CD45RO + memory T cells (Tmem) in EC:T cell cocultures,22 and ECs transfected with siRNA vs autophagy genes showed significantly reduced ability to elicit activation (HLA-DR), effector (IFN-g) and proliferative (CFSE dilution) responses in alloimmune Tmem (Fig. 5d,e).
Orthogonally, AA197-270 aggregates upregulated NF-kB marked by NIK (Fig. 5f) in both a dose- and time-dependent manner (Fig. 5g,h). To identify salient genes and/or pathways, we performed RNA seq which showed 2,160 and 2,528 genes that were significantly upregulated and downregulated, respectively, by AA197-270 aggregates (Supplementary Fig. 1d). Notably, GSEA uncovered a gene signature for autophagy (Supplementary Fig. 1e) as well as signatures related to allograft rejection, IFN-g, and NF-kB responses (Supplementary Fig. 1f). NF-kB and interferon signaling genes were strongly upregulated in AAV197-270 cells (Supplementary Fig. 1g). Based on this, we examined functional intersection between these pathways, specifically testing IFN-g priming of NF-kB genes, a phenomenon observed in PRA-treated HUVECs22 as well as macrophages where IFN-g may prime inflammasomes. We pre-treated AA197-270 cells with IFN-g and noted that inflammatory genes previously found to be NF-kB-dependent in PRA-treated HUVECs22 including CCL5 and IL-6 became synergistically or additively potentiated, respectively, by IFN-g (Fig. 5i). Similar to PRA-treated HUVECs, gene-specific knockdowns of ATG5 significantly reduced NF-kB (Fig. 4l) and inflammatory genes in AA197-270 cells (Fig. 5j). Original and uncropped Western blot films for Fig. 5 are shown in Supplementary Fig. 5. Aggrephagy of C9 promotes NF-κB and EC activation.
ZFYVE21 is Required for C9 Aggrephagy. While Rab5 activity was required to form C9 aggregates (Fig. 3c), prior studies have linked Rab5 activity to aggregate degradation, i.e., aggrephagy.37 ZFYVE21 is a conserved Rab5-associated protein implicated in cell motility,41 and we previously found that ZFYVE21 mediated NF-kB activity.24,26 The functions of ZFYVE21 have not been connected to macroautophagy, and in this context we tested roles for ZFYVE21 in mediating C9 aggrephagy.
Rab5 + endosomes sequester ZFYVE21 to shield this protein from proteasome degradation, allowing its rapid upregulation following PRA.24,26 became upregulated in PRA-treated HUVECs contemporaneously with LC3-II (Fig. 6a,b). To colocalize internalized MACs with ZFYVE21, we treated HUVECs with PRA spiked with AF647-labeled C9 protein, an approach forming intracellular C9 + vesicles.24 ZFYVE21 colocalized with C9 + vesicles containing LC3B (Fig. 6c,d, Supplementary Fig. 2a), and ZFYVE21 + C9 + punctae became grossly enlarged with CQ (Supplementary Fig. 2b). These data indicated that ZFYVE21 trafficked through an autophagic pathway. In pulse-chase studies, ZFYVE21 siRNA ablated aggrephagic flux while potentiating C9 (Fig. 6e,f), and this ablated GFP + RFP + autophagosomes (Fig. 6g). We found that in contrast to PRA, serum starvation induced autophagic flux in ECs without upregulating ZFYVE21 (Supplementary Fig. 2c,d), and ZFYVE21 siRNA in serum-starved ECs showed no effects on LC3-II (Supplementary Fig. 2c,e), indicating a principal role for ZFYVE21 in selective macroautophagy, i.e., C9 aggrephagy, but not non-selective macroautophagy.
We tested if ZFYVE21 could interact with one or more ATG8 family proteins which critically anchor signaling proteins to aggresomes.37 We performed pulldowns in ZFYVE21-FLAG ECs treated with CQ to block degradation of ATG8 family proteins which might otherwise become substrates for aggrephagy.42,43 In co-IPs, ZFYVE21 principally interacted with LC3A and LC3B (Fig. 6h,i). ATG8 family proteins including LC3B bind adapter proteins via LC3B Interacting Regions (LIRs, W/F/Y-x-x-I/L/V where x = any amino acid) and recently identified GABARAP interacting motifs (GIMs, W/F-V/I-x-V).43 While ZFYVE21 did not contain GIMs, we identified 2 LIRs between AA57-60 and AA146-149 (Supplementary Fig. 2f). Alanine substitutions showed that L60, Y146, and L149 mediated ZFYVE21 binding to LC3B (Fig. 6j,k). Original and uncropped Western blot films for Fig. 6 are shown in Supplementary Fig. 6. ZFYVE21 is a mediator of C9 aggrephagy, a process required for NF-kB and EC activation.
RNF34-P62 Complexes Mediate C9 Aggrephagy. We sought to identify E3 ubiquitin ligases utilized by ZFYVE21 to mediate C9 aggrephagy. We perform proteomic analyses of the ZFYVE21 interactome in ZFYVE21-GFP HUVECs treated with PRA for 2 hours. Three separate preparations yielded spectra for 3,347 proteins, 586 of which were uniquely upregulated by PRA (Supplementary Fig. 2g). PRA-induced proteins included Rab5A/B/C, MAC proteins (C5/6/7/8a/8g/9), and inflammasome proteins (NLRP3, casp1)25 which previously appeared in prior MAC + Rab5 + proteomic datasets.24 We additionally noted V-ATPase subunits (ATP6V1A/B2/C1/D/E1/H) and autophagy-related proteins (ATG3, ATG8G, P62), further supporting a role for vesicular acidification and ZFYVE21 within aggresomes, respectively.
We identified upregulated proteins in our proteomic datasets containing RING and HECT domains, domains known to show E3 ubiquitin ligase activity. We cross-referenced this list against proteins showing z-scores ≥ 2 in a prior genome-wide siRNA screen for NF-kB.23 Among proteins identified, we focused on RNF34. RNF34 is an E3 ubiquitin ligase containing a FYVE domain allowing colocalization to Rab5 + endosomes and a RING domain enabling ubiquitinylation of immune-related substrates.44,45 RNF34 has not been connected to macroautophagy but was previously found to directly bind ZFYVE21 in a process that enhanced its stability in HUVECs.26
We thus considered whether RNF34 interacted with ZFYVE21 within aggresomes. RNF34 pulled down with ZFYVE21 following PRA treatment, and this interaction persisted to 2–4 hr, timepoints when aggresome formation occurred (Fig. 7a). At these times, RNF34 was found to heavily colocalize with ZFYVE21 + Thioflavin + aggresomes (Fig. 7b). RNF34 siRNA reduced C9 ubiquitinylation while potentiating C9 (Fig. 7c,d), and this resulted in attenuated NF-kB (Fig. 7c,d) and decreased EC-mediated T cell activation in EC:T cell cocultures (Fig. 7e). Based on this, we tested RNF34 ubiquitinylation of C9 in cell-free ubiquitinylation studies (Fig. 7f). Among E2 conjugating enzymes tested, we selected UBCH5a due to its strong activity. We found that UBCH5a allowed RNF34 to directly K48 ubiquitinylate C9 in a manner requiring its RING domain which is known to contain E3 ubiquitin ligase activity (Fig. 7g). We subsequently co-transfected CQ-treated HUVECs with RNF34 siRNA in the presence of either Ub-WT or Ub-DN constructs and found that RNF34 siRNA, like Ub-DN, blocked C9 ubiquitinylation (Fig. 7h), indicating that RNF34 ubiquitinylates C9.
Ubiquitinylated cargo require binding to proteins containing ubiquitin binding domains (UBDs).46 C9 and ZFYVE21 lack annotated UBDs and did not bind K48 Ub chains in the absence of E2 conjugating enzymes (Supplementary Fig. 2h). In studies above, we tested P62, a UBD-containing protein, which, in contrast to C9 and ZYVE21 showed robust binding to K48 Ub chains (Supplementary Fig. 2h). In ZFYVE21-FLAG ECs, P62 pulled down with ZFYVE21 (Fig. 7i) and heavily colocalized with ZFYVE21 + punctae concurrently showing thioflavin staining (Fig. 7j). In pulse-chase studies, P62 siRNA potentiated C9 while inhibiting LC3-II (Fig. 7k). P62 (SQSTM1) siRNA strongly inhibited PRA-induced NF-kB luciferase activity (Fig. 5a), and strongly decreased phosphoP65 in PRA-treated HUVECs (Fig. 7k). Original and uncropped Western blot films for Fig. 7 are shown in Supplementary Fig. 7. Collectively, our data support a function for ZFYVE21 as an adaptor, bridging LC3 + aggresomes to RNF34-P62 complexes to mediate C9 aggrephagy.
Signaling Responses Related to C9 Aggrephagy Occur In Vivo. We examined the relevance of salient signaling processes in vivo. We initially tested whether C9 aggregates could occur in AV. To do this we employed a humanized model of ischemia reperfusion injury (IRI).27 In this model, human artery segments are subjected to hypoxia in organ culture prior to being implanted as interposition xenografts in descending aortae of SCID/beige immunodeficient mice pre-engrafted with human lymphoid cells. This protocol activates complement (C’) and forms MACs in adluminal ECs but also the media, a region typically spared from collagen deposition in AV.9 This allowed us to assess thioflavin staining without confounding collagen fluorescence which complicated our prior anayses in CABMR biopsies (Fig. 1). Compared to controls, C’-treated grafts developing AV showed increased thioflavin staining in medial regions (Fig. 8a), and C’-treated tissues showed increased insoluble C9 (Fig. 8b). Our data showed that C9 aggregates may form during AV.
To further test C9 signaling responses, we employed a second approach where we injected C57/Bl6 mice (H-2b) with anti-H2b Ab, a treatment forming non-cytolytic MACs on ECs.26 Following treatment with anti-H2b Ab but not but not control MOPC Ab, C9 + ECs co-staining for thioflavin were visualized (Fig. 8c). Concurrent with this, anti-H2b Ab-treated hosts showed increased NIK whose intensity became significantly attenuated in C3−/− mice (H-2b) that bind IgG but lack the ability to form MACs (Fig. 8d). Concomitant with loss of NF-kB activity, C3-/- mice showed decreased thioflavin staining in ECs (Fig. 8e), indicating a role for terminal C’ activation for aggregate-induced EC activation. Insoluble C9 aggregates were significantly reduced by Dynasore (Fig. 8f), indicating that, similar to PRA-treated HUVECs, C9 aggregates formed intracellularly in vivo. Autophagy reporter (AR) mice are on the C57/Bl6 background (H-2b) and express mRFP-GFP-LC3B, allowing autophagic flux to be assessed via GFP + RFP + MFIs. We injected AR mice with anti-H2b Ab and observed increased GFP + RFP + regions (Fig. 8g) colocalizing with C9 (Fig. 8h), indicating C9 aggrephagy. C9 aggrephagy was blocked by Dynasore (Fig. 8i). Internalization of terminal C’ proteins induces C9 aggregates, C9 aggrephagy, and NF-kB in vivo.
ZFYVE21 Mediates RNF34-Mediated Aggrephagy and NF-kB In Vivo. To examine role(s) for ZFYVE21 in vivo, we generated ZFYVE21fl/fl mice lacking exon 3 which mediates ZFYVE21 colocalization to early endosomes.41 Per Fig. 9a, we generated global ZFYVE21−/− mice. After confirming loss of ZFYVE21 in multiple tissues (Fig. 9b), we injected ZFYVE21−/− mice with anti-H2b Ab and relative to age- and gender-matched littermate controls, ZFYVE21−/− mice showed increased thioflavin staining ((Fig. 9c) and increased insoluble C9 (Fig. 9d). ZFYVE21−/− kidneys treated with anti-H2b Ab showed increased LC3B which, like P62, is known to become an autophagic substrate in vivo,42,43 indicating decreased autophagic flux (Fig. 9e). Notably, loss of ZFYVE21 virtually abolished expression of RNF34, and tissue loss of RNF34 was associated with significantly attenuated NF-kB marked by NIK (Fig. 9e,f) and reduced EC activation marked by VCAM-1 in glomerular ECs (Fig. 9g).
To further delineate roles for RNF34, we utilized a third approach incorporating collagen-fibronectin gels.24,26 HUVECs were embedded in collagen-fibronectin gels and implanted subcutaneously into SCID/beige mice, allowing HUVECs to self-assemble into perfused microvessels in vivo.47 Three weeks post-implantation, hosts bearing collagen-fibronectin gels and injected with PRA showed perfused Ulex + microvessels co-staining for C9 and VCAM-1 (Supplementary Fig. 2i). PRA-treated gels additionally showed increased punctate thioflavin and C9 staining under I.F. (Supplementary Fig. 2i, Fig. 9h). In follow-up studies, HUVECs were transduced with control or RNF34 shRNA prior to gel embedding and subcutaneous implantation. Here, we found that RNF34 shRNA significantly potentiated both thioflavin staining and C9 MFIs (Fig. 9i), indicating a role of RNF34-mediated C9 aggrephagy in vivo. Original and uncropped Western blot films for Fig. 8 are shown in Supplementary Fig. 8. ZFYVE21 stabilizes RNF34 to mediate C9 aggrephagy, NF-kB, and EC activation in vivo.
ZFYVE21 in ECs Dictates Alloimmune Tissue Injury. We analyzed effects of ZFYVE21 in a skin allograft model of CABMR.26 In CABMR, ECs become principal targets for MACs.6,15 To define roles for ZFYVE21 in ECs, we crossed Cdh5-Cre x ZFYVE21fl/fl mice to generate ZFYVE21 EC−/− mice. Subsequently, male ZFYVE21 EC−/− mice, ZFYVE21−/− mice, and littermate controls (WT) were injected with anti-H-2b Ab to form MACs, and twenty-four hours later MAC-treated male skin grafts were placed onto SCID/bg recipients passively receiving female splenocytes. Twenty-one days later, compared to MOPC Ab-treated controls, WT skin grafts treated with anti-H2b Ab showed increased epidermal thickening (Fig. 10a) and CD45 + immune cell infiltrates (Fig. 10b), both of which became significantly reduced in similarly-treated ZFYVE21 EC−/− and ZFYVE21−/− skin grafts. We noted that the tissue readouts above were reduced in ZFYVE21 EC−/− mice to a degree approximating that of ZFYVE21−/− mice globally lacking ZFYVE21, indicating a principal role for ZFYVE21 in ECs in mediating alloimmune tissue injury in our disease model. On further phenotyping, we found that ZFYVE21 EC−/− grafts showed decreased cytokines (Fig. 10c) and chemokines (Fig. 10d) vs comparably-treated WT hosts. Finally, to generalize findings we analyzed public RNA seq data from complement-mediated conditions including CABMR (n = 110), rheumatoid arthritis (n = 23), and lupus nephritis (n = 21). These analyses showed significant correlations between genes marking EC activation and aggrephagy (Supplementary Fig. 2j) as well as moderate-high correlations between aggrephagy genes with ZFYVE21 (left, Supplementary Fig. 2k) and RNF34 (right). ZFYVE21 in ECs modulates MAC-induced tissue injury in vivo.