To determine whether antibodies deposit in SAH livers, we collected explanted liver tissues from SAH patients during liver transplantation at Johns Hopkins. Liver tissue sections with H&E staining from SAH patients showed histologic features of SAH including macrovesicular steatosis, neutrophilic lobular inflammation, ballooning hepatocyte degeneration, Mallory- Denk bodies, and portal and pericellular fibrosis (Figure 1A). Immunohistochemistry (IHC) staining by using anti-human immunoglobulin antibodies demonstrated massive IgA and IgG deposition in ballooned hepatocytes in SAH livers, while none of the hepatocytes were stained with anti-human immunoglobulin antibodies in liver tissue sections from healthy donors except for positive staining in some hepatic sinusoid cells (Figure 1B and C). To further confirm the deposition of immunoglobulins in SAH livers, the presence of immunoglobulins in liver tissue homogenates form SAH (n=7) or healthy donors (n=7) was assessed by Western blot analysis and ELISA assays. Western blot analysis demonstrated that the levels of IgA and IgG were dramatically increased in all SAH livers as compared with the donor livers (Figure 1D and E). The IgM but not the IgE level was also significantly increased in SAH livers. The increase of IgA, IgG and IgM levels in SAH liver tissue homogenates was further confirmed by ELISA. IgA and IgG isotypes were major immunoglobulins in SAH livers (Figure 1F). Further analysis of IgG subclasses demonstrated that the IgG subclass levels - predominantly IgG1 – were significantly higher in SAH livers than that in healthy donors (Figure 1G—figure supplement 1). On the basis of these findings, we performed IHC staining for human IgG in SAH livers from 45 patients with liver transplantation and 10 donor livers in a clinical pathology lab at Johns Hopkins in a double- blind manner. The IgG+ hepatocytes in scanned slides of stained tissues sections were analyzed by using HALO^TM Image Analysis Software. Positive cells were reported as percentage stained surface area of total annotated area by digital analysis (figure supplement 1). Few IgG+ hepatocytes were identified in donor livers. In contrast, on average ∼40% of the hepatocytes (ranging from 4% to 80%) were IgG positive in the SAH livers (Figure 1H). These findings demonstrated the deposition of immunoglobulin antibodies in ballooned hepatocytes in SAH livers.

IgG, especially IgG1, plays a critical role in the classical completement activation pathway. To determine if immunoglobulin in ballooning hepatocytes induces activation of complement, C3d and C4d were analyzed in SAH livers. IHC staining showed the presence of both C3d and C4d in ballooning hepatocytes in SAH livers but not in the donor livers (Figure 2A and 2B). Double staining for IgG and complement fragments C3d or C4d showed IgG co-stained with C3d or C4d in ballooning hepatocytes (Figure 2C and 2D). These results indicated that IgG deposition in hepatocytes was associated with activation of complement. Furthermore, complement activation including the presence of C3d and C4d in SAH liver was confirmed by Western blot analysis (Figure 2E and 2F). Finally, we asked whether antibodies (Ig) extracted from SAH livers exhibit hepatocyte killing efficacy in an ADCC assay. Compared to isotype control human Ig from healthy donors, no increased hepatocyte killing was observed when PBMCs (effector cells) from healthy donors were added into cultured human hepatocytes (target cells) in the presence of serum Ig from SAH patients. However, the hepatocyte killing efficacy was significantly increased when the same levels of Ig extracted from SAH livers were added into the hepatocytes/PBMCs co-culture system (Figure 2G). These results demonstrated that immunoglobulin antibodies deposited in hepatocytes of SAH livers could induce activation of complement but more importantly, exhibited antibody-dependent cellular cytotoxicity of hepatocytes. Therefore, deposition of immunoglobulin antibodies may contribute to the hepatocyte ballooning degeneration and necrotic damage in SAH.

To further define the Ig and C4d deposition on the membrane of ballooned hepatocytes, we performed multiplex analyses by co-staining liver tissue sections with multiple cell markers for hepatocytes (Heppar1), Kupffer cells (CD68), hepatic sinusoid endothelial cells (CD32) and hepatocyte membrane (β-catenin). Images of confocal microscopy showed the presence of both IgG and IgA in hepatic sinusoid endothelial cells but not hepatocytes in the donor livers, while the majority of ballooned hepatocytes co-stained with IgG and IgA in SAH livers (new Figure 3A). Further, co-staining with β-catenin detected heavy IgG and IgA deposition on hepatocyte membrane (Figure 3B). Interestingly, the IgG and IgA on hepatocyte membrane were co-stained with C4d as well (Figure 3C).

We used the human proteome microarray (HuProt), comprised of 21,240 individual purified human proteins, to perform antibody profiling assays¹⁵. Each liver specimen was treated to release tissue-deposited immunoglobulins (Figure 4A). After neutralization, the extracted antibodies from each liver sample were separately probed to the HuProt arrays, using isotype- specific secondary antibodies to obtain the IgG, IgA, IgM, and IgE autoimmune signatures of the same liver sample (Figure 4B). Many positive human proteins were recognized by each of the four Ig isotypes in all five SAH samples. Each antibody profiling assay was performed in duplicate and only those reproducible signals were scored. A substantial fraction of autoantigens was shared by the antibodies in all five SAH livers, regardless of the Ig isotype (Figure 4C—figure supplement 2). The total numbers of the shared autoantigens recognized by the IgG, IgA, IgM, and IgE isotypes were 346, 319, 194, and 10 (Suppl. Figure S2), respectively, suggesting that the shared autoantibodies of the IgG and IgA isotypes were the most prevalent, while the counterparts of IgE isotype were the scarcest. With tissues of five donor livers, the numbers of positive human proteins were much lower (Figure 4C—figure supplement 3).

Although 95 shared IgG autoantigens were identified by the donor liver samples, 79 (83.2 %) of them were also shared by the SAH livers (Figure 4C). More importantly, a large fraction (i.e., 267 proteins) of the SAH-shared autoantigens were not found in the donor livers (Figure 4C)

We applied the above approach to a group of five livers explanted from patients with AC, AIH, PBC, NASH, HCV, and HBV-infection. SAH still exhibited the highest number of shared IgG autoantigens, while HBV showed highest numbers of shared IgA and IgE autoantigens, and PBC showed the highest number of shared IgM autoantigens (figure supplement 3). The numbers of shared IgG, IgA and IgM autoantigens were much higher in SAH livers (n=859) as compared with AC (n=349), HBV (n=735), other liver diseases (n<428) or HD livers (n=448). Although the shared IgE autoantigens were the lowest in all seven liver diseases, each disease showed a distinct autoantibody signature (figure supplement 3).

We compared the shared autoantigens recognized by SAH immunoglobulins to their counterparts from the other six liver diseases. By Venn diagram analysis 45, 41, 68 and 8 autoantigens were commonly recognized by IgG, IgA, IgM and IgE isotype autoantibodies in liver tissue extracted from different liver diseases (figure supplement 4). 188 unique IgG autoantigens, 45 unique IgA autoantigens and 7 unique IgM autoantigens were recognized by tissue homogenates from the SAH livers, whereas the second highest in this category was HBV livers in which 1 unique IgG autoantigen, 88 unique IgA autoantigens and 25 unique IgM autoantigens were identified (Figure 4D— figure supplement 4). The third in this category was PBC in which 7 unique IgG autoantigens, 2 unique IgA autoantigens and 81 unique IgM autoantigens were identified. The tissue homogenates from HCV livers recognized 32 unique IgG autoantigens, 4 unique IgA autoantigens and 3 unique IgE autoantigens, while only 4 unique IgA autoantigens were recognized by tissue homogenates from AC livers (figure supplement 4), showing disease-distinct autoantibodies in diseased livers regardless of their etiology. A large number of unique autoantigens were recognized by Ig recovered from SAH livers (table supplement 1) indicating immunoglobulins (especially IgG) deposited to the SAH livers (Figure 1) might play an important role in pathogenesis.

Antibodies secreted into the gut mostly target bacteria and bacterial products²¹. Immunoglobulins deposited in SAH livers might be from the gut, and these immunoglobulins may be antibodies targeting intestinal bacterial antigens. To test this, we employed a bacterial proteome array, comprised of 4,256 purified E. coli proteins encoded by a commensal strain K12, to do antibody profiling assays¹⁶ (Figure 4E). Using the same liver tissues and approach described above, we obtained the immune signatures of the 40 livers from 7 diseases and 5 HD in duplicate.

Many more bacterial than human antigens were recognized by immunoglobulins from alcoholic livers (AC and SAH) compared with healthy donor livers (figure supplement 5). The differences in shared bacterial antigens between alcoholic livers and healthy donor livers were more pronounced across all four Ig isotypes. For instance, only one bacterial antigen was commonly recognized by IgG antibodies from HD livers, but 435 shared bacterial antigens (about 10.2% of 4,256 purified E. coli proteins) were recognized by IgG antibodies in the five SAH livers (Figure 4F), and 466 shared bacterial antigens were recognized by IgG antibodies in the five AC livers (figure supplement 5). The numbers of the commonly shared bacterial antigens identified by the IgG antibodies extracted from the other five liver diseases were much lower, ranging from 7 to 63 (figure supplement 5). In addition, the numbers of IgA, IgM or IgE recognized bacterial antigens by liver tissue homogenates from SAH or AC were much higher than that in other liver diseases, except PBC showing a higher number of IgM antigens than AC (figure supplement 5). This observation suggests that chronic or acute alcoholic liver disease is associated with specific antibacterial antibody signatures.

A large number of shared bacterial antigens were recognized by liver tissue homogenates from each liver disease. We found that 1, 6, 24 and 5 E. coli antigens were commonly recognized by IgG, IgA, IgM and IgE isotype antibodies in liver tissue homogenates extracted from different liver diseases (Figure 4F—figure supplement 6). Unique IgG or IgA bacterial antigens were only identified by using liver tissue from SAH or AC (Figure 4G—figure supplement 6). More unique bacterial antigens were recognized by the SAH than AC IgA (110 vs. 54), IgM (54 vs. 1) and IgE (45 vs. 1), but less so by the SAH IgG (53 vs. 83) (table supplement 2). Notably, 69 unique IgM bacterial antigens were recognized by PBC liver tissue. Common anti-E. coli protein antibodies are present in diseased livers regardless of etiology, but unique anti-E. coli IgG and IgA antibodies exist only in alcoholic livers predominantly in SAH.

The prevalent anti-bacterial immunoactivity of the immunoglobulins in SAH livers suggested that immunoglobulins deposited in SAH livers might be derived from leaky gut and these anti- bacterial antibodies cross-react with human liver proteins. We used proteins from E. coli (strain K12) immobilized on magnetic beads to capture immunoglobulins pooled from the SAH livers (Figure 5A) (E. coli captured-Ig) which were then released and incubated on the HuProt arrays to determine whether these bacterium-recognizing antibodies could also cross-react with human proteins (examples shown in Figure 5B). At a stringent cutoff value (e.g., S.D.=10) 694, 796, 451 and 42 human proteins were reproducibly identified by E. coli protein captured IgG, IgA, IgM and IgE antibodies from the five SAH livers, respectively (Figure 5C and antibody recognized protein sets in table supplement 3). Strikingly, many of these E. coli binding Ig recognized proteins (74.06%, 76.46%, 48.39% and 49.41% respectively) were also found to be the autoantigens recognized by immunoglobulins recovered directly from the five SAH livers.

These results demonstrated that there exist a large number of cross-reacting antibodies that recognize both bacterial and human proteins in the SAH livers.

To determine if autoantigens recognized by Ig from diseased livers were specifically presented in cellular components, we performed GO enrichment analysis on antibody recognized protein sets. Proteins recognized by IgG antibodies from SAH livers were significantly over-represented in 8 cellular components including cytosol, cytoplasm, nucleus, mitochondrion, focal adhesion extracellular exosome, mitochondrial intermembrane space and ruffle membrane, while autoantigens recognized by IgA antibodies from SAH livers were over-represented in cytosol and cytoplasm (Figure 6A—table supplement 5). These proteins are involved in a number of biological processes including nucleobase-containing small molecule interconversion, protein phosphorylation, signal transduction, regulation of cell proliferation, mitochondrial ribosome reassembly and mitochondrion organization (Figure 6B and C). With the exception that cellular components were identified by IgM antibodies from PBC livers (table supplement 5), no cellular component was recognized by Ig from other diseased livers. Interestingly, E. coli-captured IgG and IgA from SAH livers recognized 5 out of 8 cellular components which were recognized by Ig directly extracted from SAH livers (Figure 6D). This was observed only in SAH livers (table supplement 4). These results further demonstrated the presence of unique cross-reacting anti-E. coli autoantibodies in the SAH livers.

Alcohol-derived leaky gut may promote translocation of gut bacterial products and Peyer’s patches IgA-secreting plasma cells to the liver¹³. To determine if the migration of bacteria and/or bacterial products from the bowel to liver occurred in SAH, we performed immunohistochemistry staining for the gram-negative bacterial (E. coli) product livers. Both LPS and LTA were in liver tissue from SAH patients cf. controls (Figure 7A), especially in the inflammatory areas. The increase of LPS levels in SAH liver tissues was confirmed by using Pierce™ Chromogenic Endotoxin Quant Kit (Figure 7B).

Immunohistochemistry staining for CD20+ and CD138+ cells revealed that most of these cells were localized in the inflammatory areas of SAH livers, with none in HD livers (Figure 7C). These results suggest that increased gut bacterial antigens in the SAH liver are associated with increased B cells and plasma cells.

Based on our findings that immunoglobulins extracted from SAH livers but not present in the serum exhibited hepatocyte killing efficacy (Figure 2G), we suspected that immunoglobulin antibodies deposited in hepatocytes might not be produced systemically but by plasma cells infiltrating the liver. To determine if infiltration of B and plasma cells in the SAH liver is associated with expression of genes coding for certain immunoglobulins or fragments of immunoglobulins, we analyzed RNA seq data generated from SAH and healthy donor livers (GEO GSE143318). Specifically, we examined whether genes coding for immunoglobulins or fragments of specific immunoglobulins are upregulated in SAH liver samples relative to donor liver samples. We identified 175 genes coding for immunoglobulins or fragments of immunoglobulins and compared their expression levels between SAH and control samples.

Interestingly, 87 identified genes representing immunoglobulin fragments showed a robust and stable expression in SAH livers but not in healthy donor livers (Figure 7D). These results suggest that infiltrating B and plasma cells secrete immunoglobulins that not only have anti-intestinal antigen activity but also cross-react with hepatocyte antigens.

Explanted liver tissues and blood were collected from patients with SAH or AC who were referred for liver transplantation at Johns Hopkins hospital¹⁴. Explanted liver tissues from patients with other liver diseases were obtained from the Liver Tissue Procurement and Distribution System at the University of Minnesota, which was funded by NIH Contract# HHSN276201200017C. All studies were approved by the Johns Hopkins Medicine Institutional Review Boards (IRB00107893).

Cut sections were prepared from formalin-fixed paraffin embedded liver tissues for staining with IgG (ab200699), IgA (ab200699 or GTX20770), C4d (ab167093), C3d (ab136916), CD20 (Dako, Santa Clara, CA), CD138 (Abcam, Cambridge, MA), E. coli LPS (Abcam, Cambridge, MA), LTA (Thermo Fisher, Waltham, MA). Vectastain Elite ABC Staining Kit and DAB Peroxidase Substrate Kit (Vector Laboratories, Burlingame, CA) were used to visualize the staining according to the manufacturer’s instructions. Diaminobenzidine tetrahydrocholoride and blue alkaline phosphatase (Vector Laboratories) were used as brown and blue chromogen and hematoxylin as nuclear counterstaining. Echo Revolve microscope (Echo Laboratories Inc) was used for taking image pictures.

Liver protein lysates used for this assay contained similar concentrations of protein. Each SAH and donor liver sample were added on a precoated ELISA plate (Thermofisher Scientific, Waltham, MA) to determine the total IgG (BMS2091), IgA (BMS2096), IgM (BMS2098) and IgG subclasses (991000).

PBMCs were isolated from heparinized peripheral blood samples from healthy volunteers using Ficoll-Paque plus density gradient medium.

ADCC was determined by a calcein-acetyoxymethyl release assay (calcein-AM, C3100MP, Thermofisher Scientific). Calcein-AM labeled primary human hepatocytes were cultured with Ig from SAH livers, patient serum or human IgG isotype control in a 96 well plate at a density of 1x10⁴ cells per well in triplicate and PBMCs were added as effector cells at an effector: target cell ratio of 5:1 respectively. Antibody-independent cell mediated cytotoxicity (AICC) was measured in wells containing target and effector cells without the addition of AH or control IgG antibodies. The following formula was used to calculate ADCC: % ADCC= 100 x (mean experimental release –mean AICC) ÷ (mean maximum release – mean spontaneous release).

Fifty-milligram liver tissues from SAH or the controls were homogenized in the lysis buffer (Cell Signaling Technology, MA). The total protein concentrations of each liver samples were determine using a standard curve generated with BSA at different known concentration using the Quick Start™ Bradford Protein Assay (Bio-Rad, U.S.A.). On the basis of the measure protein cencetrions, 25 micrograms of total proteins of each liver sample were boiled in NuPAGE LDS Sample Buffer (Thermofisher, MA) and subjected to electrophoresis in a 4-12 % Bis-Tris gradient PEG gel (Thermofisher, MA). All samples were tested in several parallel gels. One gel was stained with SimplyBlue™ SafeStain according to the product manual (Thermofisher, MA), and the other one was subjected to the western blot assay using the Trans-Blot Turbo RTA Midi 0.45 µm LF PVDF Transfer Kit (Bio-Rad Laboratories, CA). After transferring the total proteins to the PVDF membrane, the membranes were incubated with the IRDye labeled human immunoglobulin antibodies specific recognizing IgG, IgM, IgA, and IgE, or mouse monoclonal antibodies against human C3d (Bio-Rad Laboratories, CA) and C4d (Santa Cruz, CA) respectively, followed by probing with Alexa 647 labeled Goat anti-Mouse IgG (H+L) (Thermofisher, MA). After scanning with Odyssey CLx Imaging System, the signals were calculated by the corresponding software and then analyzed by Excel.

Sequential multiplex immunofluorescence staining on formalin-fixed, paraffin-embedded liver sections was performed, as previously described³². Images were acquired on a Zeiss LSM 900 confocal microscope. Acquired images were processed and analyzed using FIJI³³. The following antibodies were used: HepPar1 (Catalog NBP2-45272, Novus), CD68 (Catalog M0876, Dako), CD32 (Catalog 53151, Cell Signaling Technology), IgG (109-005-088, Jackson ImmunoResearch), IgA (Catalog ab124716, Abcam), β-catenin (Catalog 610154, BD Biosceinces), C4d (Catalog BI- RC4D, BIOMEDICA).

The liver tissues were subjected to a Dounce homogenizer in a lysis buffer (0.1 M Glycine pH 2.0, 150 mM NaCl) to elute the binding antibodies (Ig). After a five-minute spin down with 20,000 g at 4°C, the supernatants were transferred to new 15ml tubes and neutralized with 1M Tris-base buffer to pH 7.0 immediately. Then, the antibody enrichment was performed using Protein L coupled magnetic beads (Thermofisher, MA) according to the manual. The antibodies extracted from 4.8-gram liver tissue pieces were subjected to the protein microarray assays using human proteome microarray HuProt^® array and E. coli strain K-12 bacterial proteome microarray respectively to screen their corresponding antigens^{15, 16}. Data analysis including the criteria for positive hits was performed as before^{15, 16}.

The total lysates of E. coli were immobilized on the magnetic beads using the kit of Pierce NHS- Activated Magnetic Beads according to the product manual (ThermoFisher, MA). Then, these E. coli protein magnetic beads were incubated with the extracted total autoantibodies from liver samples to capture the active antibodies. After eluting with Glycine pH 2.0, the eluted antibodies were neutralized with 1 M tris-base buffer to pH 7.0 immediately. Finally, these antibodies were subjected to the human proteome microarray.

The total RNA was isolated using Qiagen RNeasy© kit. After the RNA quality was assessed by capillary electrophoresis (bioanalyzer), cDNA libraries were prepared using TruSeq© RNA Library Prep Kit and sequenced with an Illumina NextSeq500. Base-calling and fastq conversion was performed using RTA (2.4.11) and Bcl2fastq (2.18.0.12), respectively. Raw sequencing files were uploaded to the NIH GEO database.

Adaptor sequences were trimmed from the raw reads using Cutadapt¹⁷. Trimmed reads were then mapped to reference genome GRCh38 using STAR aligner with default parameters¹⁸. The number of counts per gene was estimated using the “quantMode” command in STAR. Batch effect was corrected using Combat seq. Differentially expressed genes (DEGs) were then identified using DESeq2 ¹⁹. Genes with adjusted P < 0.01 and log2 fold change > 1 were chosen as DEGs.

DAVID²⁰ was used to conducted GO analysis to find out enriched GO terms (cellular components and biological process). All enriched terms were chosen with a threshold p-value of 0.05.

External data requests can be directed to the corresponding authors, who will respond within 1 week and help facilitate the request. Access to clinical datasets will be available based on approval through the IRBs of the Johns Hopkins University and might be subject to patient privacy.