Suppression of a broad spectrum of liver autoimmune pathologies by single peptide-MHC-based nanomedicines

Peptide-major histocompatibility complex class II (pMHCII)-based nanomedicines displaying tissue-specific autoantigenic epitopes can blunt specific autoimmune conditions by re-programming cognate antigen-experienced CD4+ T-cells into disease-suppressing T-regulatory type 1 (TR1) cells. Here, we show that single pMHCII-based nanomedicines displaying epitopes from mitochondrial, endoplasmic reticulum or cytoplasmic antigens associated with primary biliary cholangitis (PBC) or autoimmune hepatitis (AIH) can broadly blunt PBC, AIH and Primary Sclerosing Cholangitis in various murine models in an organ- rather than disease-specific manner, without suppressing general or local immunity against infection or metastatic tumors. Therapeutic activity is associated with cognate TR1 cell formation and expansion, TR1 cell recruitment to the liver and draining lymph nodes, local B-regulatory cell formation and profound suppression of the pro-inflammatory capacity of liver and liver-proximal myeloid dendritic cells and Kupffer cells. Thus, autoreactivity against liver-enriched autoantigens in liver autoimmunity is not disease-specific and can be harnessed to treat various liver autoimmune diseases broadly.

N anoparticles (NPs) coated with mono-specific type 1 diabetes (T1D), experimental autoimmune encephalomyelitis (EAE) or collagen arthritis (CIA)-relevant peptide-major histocompatibility complex (MHC) class II (pMHCII) molecules can restore normoglycemia in diabetic animals or motor function in paralyzed mice, and resolve joint swelling and destruction in arthritic mice 1 . pMHCII-NPs directly trigger sustained ligation of cognate antigen receptors on autoantigen-experienced FoxP3 -CD25 -T-cells, promoting their differentiation into T-regulatory-type-1 (TR1)-like cell progeny in a phagocyte-independent manner, followed by systemic expansion 1,2 . Consequently, these compounds cannot trigger TR1-like cell formation or expansion in mice that are either disease-free or do not express the cognate autoantigen 1 . These in vivo-expanded TR1-like cells then broadly suppress the polyclonal T-cell responses underlying T1D, EAE, and CIA development in a disease-specific manner, by suppressing local autoantigen presentation and antigen-presenting cell (APC) activation in a cognate antigen-dependent but non-antigenspecific manner (i.e. by recognizing cognate pMHC molecules on costimulation-competent, autoantigen-loaded APCs) 1 .
Although AIH, PBC, and PSC are considered as distinct diseases, there is a group of patients presenting with features of both cholestatic liver disease and AIH. Furthermore, PBC is frequently associated with extra-hepatic autoimmune conditions 8 . The existence of these overlap syndromes suggests that activation of T-cells targeting such liver-enriched autoantigens may contribute to various liver autoimmune conditions. In that case, pMHCIIbased nanomedicines displaying epitopes from antigens relevant to one disease (e.g. from PDC-E2 in PBC) might be able to trigger the formation and expansion of epitope-specific TR1 cells capable of blunting both the corresponding liver autoimmune disease (e.g. PBC) and other liver autoimmune diseases.
We sought to test this hypothesis by asking if pMHCII-based nanomedicines displaying epitopes from various PBC-relevant or AIH-relevant antigens could blunt liver autoimmunity broadly. We find that pMHCII-based nanomedicines displaying epitopes from various liver-autoimmune disease-relevant antigens can blunt not only the relevant liver autoimmune disease (i.e. PDCbased nanomedicines blunt PBC) but also their irrelevant counterparts (i.e. PSC and AIH in addition to PBC). Remarkably, they do so without impairing the ability of the host to mount antibody responses against exogenous antigens, to clear viral or bacterial infections or to kill metastatic allogeneic tumors. Thus, hepatocyte and cholangiocyte autoimmune insults can readily trigger the stimulation of peripheral T-cells recognizing liver-prevalent selfantigens, and such T-cell responses can be harnessed by pMHCIIbased nanomedicines to treat liver autoimmunity broadly.

Results
TR1 cell formation and expansion by PBC-relevant pMHCII-NPs. NOD.c3c4 mice, which carry anti-diabetogenic regions from C57BL/6 chromosomes 3 and 4, spontaneously develop a form of autoimmune biliary disease that resembles human PBC 9 . Like >90% of PBC patients, these mice develop autoreactive T-cell and B-cell responses against the dihydrolipoyl acetyltransferase (E2) and dihydrolipoyl dehydrogenase-binding protein (E3BP) components of the PDC complex [10][11][12] , leading to biliary epithelial cell destruction, cholestasis, small bile duct proliferation, and liver failure.
Suppression of local and proximal APCs. We have previously shown that T1D-relevant antigen-specific TR1-like CD4+ T-cells selectively suppress the proinflammatory and antigen-presenting capacity of pancreatic lymph node-associated APCs by recognizing cognate pMHC class II complexes on autoantigen-loaded APCs draining the pancreas (the source of autoantigenic    PCLNs of control-NP-treated animals. This indicated that, in NOD.c3c4 mice, treatment with PDC-E2 166-181 /IA g7 -NPs selectively downregulates the pro-inflammatory capacity of CD11b+ cells draining the liver ( Fig. 5a and Supplementary Fig. 2b), presumably because they are loaded with PDC-E2 antigenic material shed from the injured liver and thus display the TR1 cells' cognate pMHC class II complexes on their surface. pMHC-NP treatment also decreased the secretion of CCL4, IFNγ, and IL-10 by PCLN-associated CD11b+ cells, but this reduction was also seen in MLN-associated CD11b+ cells ( Supplementary  Fig. 2b). Interestingly, Kupffer cells isolated from PDC-E2 166-181 / IA g7 -NP-treated mice ( Supplementary Fig. 2a) also secreted significantly lower levels of 8 of these 30 mediators ( Fig. 5b and Supplementary Fig. 2c). Thus, systemic expansion of PDC-E2specific TR1 CD4+ cells in NOD.c3c4 mice is associated with dramatic inhibition of the pro-inflammatory properties of local Liver B-regulatory (Breg) cell recruitment and formation. The liver and the PCLNs, but not non-liver-draining inguinal LNs (ILNs) of PDC-E2 166-181 /IA g7 -NP-treated mice harbored significantly higher numbers of PDC-E2 166-181 /IA g7 -tetramer+ cells and B-cells than those from control-NP-treated animals (Fig. 5c).
In addition, the B-cell and tetramer+ T-cell numbers in liver, albeit not PCLNs, were correlated (Fig. 5d). Furthermore, the liver and PCLN, but not the MLN B-cells of PDC-E2 166-181 / IA g7 -NP-treated mice produced IL-10 in response to LPS, whereas neither the liver nor the PCLN B-cells of control NPtreated animals produced IL-10 ( Fig. 5e). Collectively, these observations suggested that, in the liver-draining lymph nodes, PDC-E2 166-181 /IA g7 -NP-induced TR1-like cells promoted Breg cell formation. These B-cells had disease-specific immunoregulatory activity in vivo, because transfer of purified PCLN Bcells from PDC-E2 166-181 /IA g7 -NP-treated NOD.c3c4 mice suppressed the transfer of disease into NOD.scid.c3c4 hosts reconstituted with splenocytes from diseased NOD.c3c4 donors. In contrast, the pancreatic lymph node-associated B-cells from BDC2.5mi/IA g7 -NP-treated NOD mice, which can protect NOD. scid mice from diabetes transfer 1 , had no effect on the ability of splenocytes from sick NOD.c3c4 mice to transfer PBC into NOD. scid.c3c4 hosts, suggesting that these B cell-mediated immunoregulatory effects are antigen-specific (Fig. 4h).
To investigate this further, we ascertained the ability of PDC-E2 166-181 /IA g7 -specific TR1-like cells to promote the differentiation of PDC-E2 166-181 peptide-pulsed conventional B-cells isolated from NOD.Il10-eGFP reporter mice (IL-10/eGFP-Bcells) into CD1d high /eGFP+ and CD5+/eGFP+ progeny in vivo (upon adoptive transfer into PDC-E2 166-181 /IA g7 -NP-treated hosts). As shown in Fig. 5f-g, there was a clear formation of Breg cells in the hosts' spleen, liver, and PCLNs (containing cognate TR1 cells) but not in the MLNs (lacking cognate TR1 cells). Thus, PDC-E2-specific TR1 cells promote the recruitment and differentiation of conventional B-cells into Breg-like cells. This outcome was driven by cognate pMHC class II interactions between the host's TR1 cells and the transfused eGFP-B-cells, because it only occurred when the donor B-cells were pulsed with cognate (PDC-E2 166-181 ) but not an irrelevant (BDC2.5mi) peptide (Fig. 5h).
Continued versus intermittent treatment. We next examined if the size of the cognate TR1 cell pool arising in blood in response to therapy could be used to gauge the need for re-treatment. As the liver is a large organ, we suspected that the bloodresidence time of the PDC-E2-specific TR1 cells in diseased NOD. c3c4 mice would be significantly shorter than in diabetic NOD mice, where cells can persist in the circulation for months after treatment withdrawal 1 . The blood tetramer+ T-cell content from most mice declined to~50% of the original values within 4-6 weeks after treatment withdrawal. Re-treatment rapidly restored these values (Fig. 6a, b). Intermittent therapy given up to 50 weeks of age did not compromise the pharmacodynamic or therapeutic effects of pMHCII-NPs ( Fig. 6c-g), as compared to mice treated continuously, supporting the safety of these compounds, even when administered for prolonged periods of time.
pMHCII-NPs versus the standard of care. Ursodeoxycholic acid (UDCA, a hydrophilic bile acid) is the standard of care for PBC. Although effective in~50% of patients when given early, it is ineffective at advanced stages of PBC 13 .
Intake of UDCA-supplemented chow by 6-week-old NOD.c3c4 mice for 14 weeks had a small therapeutic effect on the progression of PBC, as manifested by reductions in liver scores and liver weight, bile duct proliferation and mononuclear cell infiltration, albeit not serum ALT or TBA levels, CBD scores, CBD diameter, or bile duct involvement (Fig. 7a-d). However, when UDCA was given at 24 weeks, it had none of these effects, except for a very significant reduction in CBD diameter, possibly because of its anti-cholestatic effects (Fig. 7e-g). In contrast, PDC-E2 166-181 /IA g7 -NPs had substantial therapeutic effects in both 6-week-old and 24-week-old animals (Fig. 7a-h).
Therapeutic effects in another PBC model. The NOD.c3c4 model does not fully recapitulate the immunopathology of human PBC, characterized by female prevalence, progression to liver fibrosis, and absence of liver cyst formation. B6 mice carrying a deletion of the IFNγ 3′-untranslated region adenylate uridylaterich element (ARE) (ARE-Del +/-) have a dysregulated Ifng locus, and develop a form of PBC that, like the human disease, primarily affects females and is associated with up-regulation of TBA, production of anti-PDC-E2 autoantibodies, portal duct and lobular liver inflammation, bile duct damage and fibrosis 14 .
Treatment of female (NODxB6.IFNγ ARE-Del -/-) F1 mice with PDC-E2 166-181 /IA g7 -NPs triggered TR1 cell formation/expansion and suppressed the upregulation of TBA and ALT levels, liver inflammation and fibrosis, as compared to mice treated with control NPs (Supplementary Fig. 3a-d). Similar results were obtained in B6.IFNγ ARE-Del -/mice treated with NPs displaying an IA b -binding PDC-E2-derived epitope (PDC-E2 94-108 / IA b -NP) ( Supplementary Fig. 3e, f), indicating that the therapeutic activity of these compounds is not a peculiarity of the NOD genetic background or its unique MHC class II molecule.
Disease versus organ specificity. Given that PDC-E2 is an autoantigen expressed in virtually all cell types, our results begged the question of whether PBC-relevant nanomedicines (i.e. PDC-E2 166-181 /IA g7 -NP) are disease-specific or not.
PSC is a chronic cholestatic disease characterized by inflammation of intra-hepatic and extra-hepatic bile ducts leading to a fibroobliterative cholangitis with periductal fibrosis around medium and large bile ducts and degenerative changes of the biliary epithelium, in the absence of anti-mitochondrial autoantibodies 17 . Abcb4 knockout mice (lacking the multidrug resistance protein 3) develop a form of cholangitis similar to human PSC that is caused by impaired biliary phospholipid secretion 17 .
AIH is characterized by a portal mononuclear cell infiltration of the liver parenchyma that is associated with presence of ANAs and/or smooth muscle (AIH type 1) or anti-liver kidney microsomal or anti-liver cytosol type 1 autoantibodies, which target the microsomal cytochrome CYP2D6 or FTCD, respectively (AIH Type 2) 18 . Recently, it has been shown that infection of NOD mice with a replication-defective adenovirus encoding human FTCD (Ad-hFTCD) triggers a form of chronic AIH that resembles AIH type 2 19 .
Collectively, these observations suggest that hepatocyte (AIH) and bile duct epithelial (PBC and PSC) damage in liver    Fig. 6 Effects of intermittent vs. continuous therapy. a and b Changes in the circulating frequency of tetramer+CD4+ T-cells in response to intermittent retreatment (as a function of circulating tetramer+ T-cell levels). a shows profiles of one mouse, where the green arrows indicate the timing of individual doses and double arrowheads indicate that two doses were given in that particular week(s). Treatment was withdrawn at 24 weeks of age. Mice were monitored for persistence of tetramer+ cells in blood once every two weeks. When the percentages of tetramer+ cells fell below 50% of the original values, treatment was re-initiated and when the percentages of tetramer+ cells in blood recovered, treatment was withdrawn again. b shows the average values ± SEM corresponding to cohorts of NOD.c3c4 mice intermittently treated with PDC 166-181 /IA g7 -NPs (n = 19) or left untreated (n = 16) after withdrawal of continued therapy (twice a week from 15 to 24 weeks of age). Data are from four experiments. c and d Percentage of tetramer+CD4+T-cells and mean fluorescence intensity staining for TR1 markers in tetramer+CD4+ T-cells from the mice studied in a and b. Data in c correspond to n = 12 pMHCII-NP-treated and 13 untreated mice/organ, respectively, from four experiments. Data in d correspond to n = 6 mice/group/organ. e-g Serum TBA and ALT levels (e; n = 10, 11, 8 and 7, from left to right, respectively), macroscopic (f; n = 14, 16, 19, and 8, from left to right, respectively) and microscopic (g; n = 15, 12, 7, and 8 from left to right, respectively) scores from mice treated intermittently (from 15 to 50 weeks of age), or continuously (twice a week from 24 to 38-44 weeks of age). Averaged data correspond the mean ± SEM. P values were compared via Mann-Whitney U (c-g) or two-way ANOVA (b) NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-019-09893-5 ARTICLE NATURE COMMUNICATIONS | (2019) 10:2150 | https://doi.org/10.1038/s41467-019-09893-5 | www.nature.com/naturecommunications autoimmunity results in the delivery of significant amounts of liver-prevalent autoantigens, including PDC-E2, CYPD2D6, and FTCD to local and proximal APCs. In turn, this enables autoreactive CD4+ T-cell priming (a sine qua non requirement for pMHC-NP-induced TR1 cell formation 1 ), cognate TR1 cell generation by pMHC-NPs, Breg cell formation, and suppression of the pro-inflammatory capacity of local and proximal autoantigen-loaded APCs.
Normal immunity is spared. We next investigated if persistent expansion of PDC-E2 166-181 /IA g7 -specific TR1 cells results in suppression of normal immunity against infection and cancer.
Cohorts of NOD.c3c4 mice received doses of PDC-E2 166-181 / IA g7 -NPs or control NPs twice a week for 9 weeks. At the end of therapy, the mice were given an i.v. injection of recombinant vaccinia virus. The viral titers in the ovaries of females 14 days after infection were similar in both cohorts of mice and substantially lower than those found at the peak of infection, indicating that pMHCII-NP therapy did not impair cellular immunity against the virus-infected cells (Fig. 10a).
To probe this further, we infected PDC-E2 166-181 /IA g7 -NPtreated or untreated NOD.c3c4 mice with a laboratory strain of influenza (HKx31 -H3N2-) i.p. to induce heterologous (shared) immunity against a subsequent lethal infection with an    Fig. 7a). Similar results were obtained in mice infected with the intracellular pathogen Listeria Monocytogenes (LM). LMinfected PDC-E2 166-181 /IA g7 -NP-treated and untreated NOD. c3c4 mice were equally efficient at clearing the bacteria from both the spleen and liver, consistent with unimpaired immunity against this intracellular pathogen ( Fig. 10d and Supplementary  Fig. 7b). This outcome was also true when the mice were infected with LM shortly before initiation of pMHC-NP treatment. As shown in Supplementary Fig. 7c-f, diseased NOD.c3c4 mice infected with LM immediately before initiation of treatment and treated for 5 consecutive weeks had, at the end of follow up, significantly reduced disease scores, as well as numbers of LM colony forming units (cfu) in both the liver and spleen. Importantly, this reduction in splenic and liver LM cfu was similar in pMHC-NPs-treated vs. untreated mice. Thus, treatment suppressed liver inflammation without impairing the ability of the host to clear the pathogen, presumably because the mechanisms involved in clearance of this intracellular pathogen are not impaired by TR1 cell-driven immunoregulation.
Thus, despite targeting a systemically expressed antigen, PDC-E2 166-181 /IA g7 -specific TR1 cells do not impair cellular or humoral immunity against local or systemic foreign antigens. This is potentially so because bacterial/viral antigenic load in local APCs transiently overwhelms the APCs' ability to present PDC-E2 epitopes to cognate TR1-like CD4+ T-cells, hence the manifestation of their immunoregulatory properties.

Discussion
We have used mice undergoing PBC, PSC, or AIH, as well as NSG mice humanized with PBMCs from PBC patients to investigate whether hepatocyte and/or cholangiocyte destruction in autoimmunity results in the stimulation of autoreactive T-cells capable of responding to disease-relevant and irrelevant pMHCIIbased nanomedicines. We found that nanomedicines displaying various liverprevalent antigenic peptides triggered TR1-like cell formation and expansion in mice undergoing various liver autoimmune diseases, as well as in NSG mice humanized with PBMCs from PBC patients. As a result, these nanomedicines effectively blunted PBC, PSC, and AIH in various genetic backgrounds by suppressing liver inflammation, even when initiated at the peak of disease severity. In the liver, disease suppression involved TR1 cell-driven local Breg cell formation, required both IL10 and TGFβ, could be transferred by both tetramer+CD4+ T-cells and PCLN-associated B-cells from treated donors, and was associated with profound suppression of the pro-inflammatory capacity of both liver and liver-proximal myeloid DCs as well as Kupffer cells. In contrast, a nanomedicine displaying a pancreatic beta cell-specific epitope was unable to trigger cognate TR1 cell responses in NOD.c3c4 mice undergoing liver autoimmunity in the absence of pancreatic autoimmunity, consistent with the sine qua non requirement for autoantigen-experience in T-cell responsiveness to pMHCII-NPs 1 .
Importantly, suppression of liver inflammation by these nanomedicines did not compromise immunity against viruses (vaccinia, influenza), intracellular bacteria (Listeria), or metastatic (liver) allogeneic tumors. The TR1-like CD4+ T-cells that are triggered by pMHC-based nanomedicines can only effect regulation when they engage cognate pMHC class II of professional APCs that are loaded with endogenous autoantigen 1 . Such APCs must capture autoantigen shed from the damaged liver cells and therefore are only present in significant numbers in the target organ or in draining lymphoid tissue. As a result, it is not surprising that pMHC-based nanomedicines do not impair immunity against systemic infections or against vaccines, as the liverdistal APCs that orchestrate these immune responses are not loaded with liver-derived autoantigens. For intra-hepatic or liverproximal immunity, such as against a LM infection, the infected liver APCs may be overwhelmed with LM-derived antigens, decreasing their ability to elicit TR1 cell function and suppression (by dilution of cognate pMHCs at the expense of pathogenderived pMHC complexes below the threshold required for TR1 cell activation). These cells may therefore be spared from suppression. Given the short half-lives of myeloid-derived APCs (days), replacement of these APCs by uninfected ones might be sufficient to support continued TR1-mediated immunoregulation. Alternatively, the mechanisms involved in clearance of this intracellular pathogen and allogeneic liver tumor metastases may not be impaired by TR1 cell-driven immunoregulation.
Collectively, these results have several important implications for our understanding of both normal immunity and treatment of autoimmunity. First, they demonstrate that tissue destruction in specific autoimmune diseases has the potential to trigger the stimulation or possibly outright activation of autoreactive T-cells recognizing many, perhaps all, of their antigenic components, suggesting that the antigenic repertoires in autoimmune diseases may be much more extensive than currently thought. We note that the pMHCs tested herein were designed in silico, using online MHC-binding algorithms, raising the possibility that any peptide capable of binding to self-MHC molecules, from a whole host of proteins expressed by hepatocytes and/or cholangiocytes, might be recognized by, and be able to activate peripheral T-cells. Second, our observations imply that penetrance of central and peripheral T-cell tolerance to highly expressed antigens is remarkably incomplete, even in disease-resistant genetic backgrounds. From an evolutionary standpoint, such pervasive autoreactivity may have been sustained because it functions as a source of regulatory cells capable of extinguishing pathology. Third, from a translational standpoint, this study has identified disease-modifying compounds for several complex liver autoimmune diseases that share common immunopathological pathways and represent unmet clinical needs 20,21 . Finally, these observations suggest that a few pMHCII-based nanomedicines displaying ubiquitous epitopes and HLA class II molecules encoded in oligomorphic HLA loci might be sufficient to treat various liver autoimmune diseases without impairing normal immunity. e Serum anti-DNP antibody titers upon KLH-DNP immunization (n = 3/group). f-k Percentages of tetramer+ cells (f), macroscopic PBC scores and liver weight (g), microscopic PBC scores (h), liver images (i), microscopic tumor scores (j), and survival rates (k) of untreated vs. PDC 166-181 /IA g7 -NP-treated NOD.c3c4 (n = 5/group) or untreated Balb/c mice (n = 7) challenged with CT26 cells (overlapping 100% survival in untreated vs. PDC 166-181 /IA g7 -NPtreated mice). Scale bar in (i): 100 μm. l-p Percentages of tetramer+ cells (l), macroscopic (m), and microscopic PBC scores (n), and liver weight and metastasis number (o), and liver images (p) of untreated vs. PDC 166-181 /IA g7 -NP-treated NOD.c3c4 (n = 5 and 4) or untreated B6 mice (n = 6) challenged with B16/F10 cells. q Survival rates of the mice in l-p (overlapping 100% survival in untreated vs. PDC 166-181 /IA g7 -NP-treated mice). Data correspond to mean ± SEM. P values were compared via Mann-Whitney U except for k, q (log-rank), or c (two-way ANOVA) intervals from NOD.c3c4 mice. NOD.Abcb4 -/mice were obtained by backcrossing the mutant Abcb4 allele from FVB/N-Abcb4 -/mice onto the NOD/Ltj background for six generations, followed by intercrossing. (NODxB6.IFNγ ARE-Del -/-) F1 mice were generated by intercrossing IFNγ ARE-Del −/− and NOD/LtJ mice. NOD. Il10 tm1Flv (Tiger) mice were obtained by backcrossing the Il10 tm1Flv allele from C57BL/6.Il10 tm1Flv mice (Jackson Lab) onto the NOD/Ltj background for 10 generations. These studies were approved by the institutional animal care committee of the Cumming School of Medicine at the University of Calgary.

Methods
Human subjects. HLA-DRB4*0101+ PBC patients were recruited under informed consent approved by the Institutional Ethics Review Board at Hospital Clinic (see Supplementary Table 1 for demographic and other patient details). All the work with human participants complied with all the relevant ethical regulations and was approved by the Hospital Clinic human ethics review board. . PE-conjugated pMHC class II tetramers were produced using biotinylated pMHC monomers. pMHC class II tetramer staining and phenotypic marker analysis were done as follows. After avidin incubation (15 min at RT), blood leukocytes, and single cell suspensions from spleen, lymph node, liver mononuclear cells, and bone marrow cells were stained first with pMHC tetramer (5 μg ml −1 ) in FACS buffer (0.05% sodium azide and 1% FBS in PBS) for 60 min at 37°C, and later with FITC-conjugated antimouse CD4 (5 μg ml −1 ) and PerCP-conjugated anti-mouse B220 (2 μg ml −1 ; as a 'dump' channel) for 30 min at 4°C. After washing, cells were fixed (1% paraformaldehyde in PBS) and analyzed with FACScan, FACSaria, or BD LSRII flow cytometers. For phenotypic analyses, the cells were incubated with anti-FcR Abs, and then stained with cell surface marker antibodies diluted 1:100 in FACS buffer (at 4°C for anti-CD49b and anti-LAP Abs, and at 37°C for anti-LAG-3 Abs) followed by pMHC tetramer, FITC-conjugated anti-mouse CD4 (5 μg ml −1 ) and PerCP-conjugated anti-mouse B220. Upon staining, cells were washed, fixed, and analyzed by flow cytometry. FlowJo software was used for all analyses.
pMHC monomers and peptides. Recombinant pMHC class II monomers were purified from supernatants of CHO-S cells transduced with lentiviruses encoding a monocistronic message in which the peptide-MHCβ and MHCα chains of the complex were separated by the ribosome skipping P2A sequence. The peptide was tethered to the amino terminal end of the MHCβ chain via a flexible GS linker and the MHCα chains were engineered encode a BirA site, a 6xHis tag, a twin strep-tag, and a free Cys at their carboxyterminal end. The secreted, self-assembled pMHC class II complexes were purified by sequential nickel and Strep-Tactin ® chromatography and used for coating onto NPs or processed for biotinylation and tetramer formation as described above. The epitopes encoded in the murine monomeric constructs were selected based on predicted MHCII-binding capacity using RANKPEP (http://imed.med.ucm.es/cgi-bin/rankpep_mif.cgi) using 7.54 as the threshold score. PDC-E2 166-181 had a score that fell below the threshold but was selected for experimentation because it is contained within one of the lipoylbinding domains of PDC-E2, an antigenic target for AMAs. For CYPD and FTCD epitope prediction, we used a second online algorithm (GPS-MBA) (http://mba. biocuckoo.org/) and peptides predicted by both RANKPEP and GPS-MBA were selected for experimentation. hPDC-E2 122-135 , hPDC-E2 249-262 (both contained within the lipoyl-binding domain of PDC-E2), and hPDC-E2 629-643 have been pMHCII-NP therapy for PBC in various genetic backgrounds. Cohorts of 15week-old male and/or female NOD.c3c4 mice with established PBC were left untreated or treated with 20 μg of pMHCII-NPs or Cys-NPs (i.v.) twice weekly for 9 weeks unless indicated otherwise. Liver disease scoring involved macroscopic evaluation of cyst content (0-5 for all experiments, except Fig. 6f, where cyst content was scored from 0-8), liver weight, and CBD diameter (0-4), as well as microscopic evaluation of bile duct involvement (0-4), bile duct proliferation (0-4), and mononuclear cell infiltration (0-4) 23 . In other experiments, treatment was initiated at the peak of disease (24 weeks of age) and given twice a week for 14-20 weeks. Intermittent treatment involved treating mice twice a week from 15 to 24 weeks of age, then withdrawing treatment until the percentages of tetramer + cells dropped to~50% of the levels seen at treatment withdrawal (measurements in peripheral blood were done once every 2 weeks), re-treating mice twice a week until the percentages of tetramer+ cells reached original values, and repeating this cycle until 50 weeks of age.
In experiments involving (NOD x B6.IFNγ-ARE-Del -/-) F1 and B6.IFNγ-ARE-Del -/mice, 10-week-old male and female mice were treated for 5-6 weeks. Histopathologic severity in the liver was assessed by scoring the extent of portal inflammation, lobular inflammation, and granuloma formation from 0 to 4, and bile duct damage from 0 to 2. The extent of portal inflammation and bile duct damage were scored from 0 to 4 based on the ratio between affected vs. unaffected area. The extent of lobular inflammation and granuloma formation were scored from 0 to 4 based on number of lesions per specimen 14 . Inflammatory scores were obtained by adding the scores for both severity and lesion number. The severity of fibrosis was scored on a 0-6 scale as follows 24 : 0, no fibrosis; 1, fibrous expansion in few portal areas with or without small fibrous septa; 2, fibrous expansion in most portal areas with or without small fibrous septa; 3, fibrous expansion in most portal areas with very few portal-to-portal bridging; 4, fibrous expansion in all portal areas with marked bridging (portal-to-portal and portal-to-central); 5, marked bridging with very few nodules (incomplete cirrhosis); and 6, complete cirrhosis.
Studies using NOD mice involved treating cohorts of 10-week-old pre-diabetic female NOD/Ltj mice with 20 μg of pMHCII-NPs or Cys-NPs i.v. twice weekly for 5 weeks.
pMHCII-NP therapy for AIH in NOD mice. We induced AIH by infecting 5-6-week-old female NOD/Ltj mice with an adenovirus encoding human FTCD (Ad-hFTCD, 10 10 plaque forming units (PFU) i.v.), as previously described 19 . Four weeks later, cohorts of mice with established AIH were treated with 20 μg of pMHCII-NP s or Cys-NPs (i.v.) twice weekly for 5-6 weeks. Histopathological scoring was done using the Ishak scale as above 24,26 .
pMHCII-NP therapy in human PBMC-reconstituted NSG hosts. PBMCs from HLA-DRB4*0101+ PBC patients (recruited under informed consent approved by the Institutional Review Board at Hospital Clinic) were depleted of CD8+ T-cells (to reduce the magnitude of GvHD in the hosts) using anti-CD8 mAb-coated magnetic beads (Miltenyi Biotech, Auburn, CA) and injected i.v. (2 × 10 7 ) into 8-10-week-old NSG hosts. Mice were treated with 30-40 μg pMHC-NPs starting on day 5 after PBMC transfusion, twice a week for 5 consecutive weeks, or left untreated. Therapy-induced expansion of cognate CD4+ T-cells was measured in liver, peripheral LNs, and spleen (Supplementary Table 1). The gender, age, antimitochondrial autoantibody status, and type of pMHC-NP tested for each patient are summarized in Supplementary Table 1. A mouse was considered a responder if the percentage of tetramer+ T-cells in at least two different organs were higher than the mean ± 10 standard deviation values seen in untreated hosts.
Evaluation of general adaptive immunity. Evaluation of cellular responses to Vaccinia infection was performed as previously described 1 . Briefly, pMHCII-NPtreated and untreated female mice were injected i.v. with 2 × 10 6 PFU of recombinant Vaccinia Virus (rVV) and sacrificed on days 4 and 14 after infection. Samples were processed for pMHCII tetramer staining and rVV titer measurements. Briefly, both ovaries were collected in DMEM containing 2% FBS, homogenized, freeze-thawed three times followed by sonication (three rounds, 20 s each). Serial dilutions of the lysates were added to confluent BSC-1 cell cultures at 37°C for 1 h, washed twice with serum-free DMEM and then overlaid with DMEM containing 2% FCS and 0.4% carboxymethyl cellulose (CMC; Sigma, Saint Louis, MO). On day 3, the overlay was discarded, and the cell layers were stained with crystal violet to count the number of plaques.
To evaluate cellular responses to Influenza infection, pMHCII-NP-treated and untreated mice were first primed i.p. with the HKx31 (H3N2) strain at 10 6 EID 50 per mouse. One cohort of mice was sacrificed 7 days after priming and processed for tetramer staining to confirm presence of pMHC-NP-specific TR1-like cells during priming. Other cohorts of primed mice were re-infected 30 days later with an intranasal dose of PR8 virus, a lethal H1N1 strain of Influenza (8 × 10 4 EID 50 per mouse), under anesthesia. PR8-challenged mice were weighed daily and scored clinically from 0 to 4 based on the extent of ruffled fur, reduced motility, huddled appearance, and rapid and/or labored breathing as previously described 27 . Mice were sacrificed 7 days later and processed for tetramer staining and influenza titer measurement. Briefly, lungs were collected in serum-free DMEM, homogenized and freeze-thawed three times. Serial dilutions of the lysates were added to confluent MDCK cell cultures at RT for 1 h and washed. Cultures were then overlaid with DMEM containing 0.4% CMC and TPCK-trypsin for 2-3 days, washed, fixed, and stained with crystal violet to count plaques. Cellular immunity to intracellular bacteria was determined by infecting pMHCII-NP-treated and untreated mice i.v. with 10 3 cfu of LM. In some experiments, mice were sacrificed 7 days or 14 days after infection and samples processed for tetramer staining and bacterial load measurements. Briefly, spleen IA g7 or BDC2.5mi/IA g7 tetramer + (2 × 10 5 ) T-cells FACS-sorted from the spleen and liver or pancreas-draining lymph nodes of PDC-E2 166-181 /IA g7 -NP-or BDC2.5mi/IA g7 -NP-treated donors, respectively.
Histology and immunohistochemistry. Livers were fixed in 10% formalin for 2 days, embedded in paraffin, cut into 5 μm sections and stained with H&E or Picrosirius Red. We scored (~0.5 cm 2 ) sections from the four liver lobes from each mouse (right and left, median and caudal) and a minimum of four portal triads per lobe section (16 portal triads/mouse). For immunohistochemistry, liver tissues were embedded in Tissue-Tek OCT, sectioned into 30 µm cryosections and stored on slides at −80°C. Slides were fixed in chilled acetone, washed with PBS, treated with a 1:10 dilution of 30% H 2 O 2 in PBS, washed with PBS, blocked with 10% normal goat serum in PBS, washed again, and stained with anti-mouse CD4 (GK1.5) or CD8 (Lyt-2) antibodies (1.5 h, 4°C). After washing, the slides were stained with a biotinylated goat anti-rat secondary antibody (1:200 dilution), incubated with horseradish peroxidase (HRP)-conjugated streptavidin, followed by 3,3-diaminobenzidine (DAB) substrate. Slides were counterstained with hematoxylin before mounting.
ALT and TBA assays. ALT levels in serum were determined using a kit from Thermo Fisher Scientific following the manufacturer's protocol. Briefly, serum samples were mixed with pre-warmed (37°C) InfinityTM ALT (GPT) Liquid Stable Reagent at 1:10 ratio and OD readings were taken for 3 min at 1 min intervals in a nanodrop at a 340 nm wavelength, 37°C. The slope was calculated by plotting absorbance vs. time using linear regression and multiplied with a factor to obtain ALT levels in serum (U/l) as described in the kit. Serum TBA levels were analyzed using a TBA Enzymatic Cycling Assay Kit (Diazyme, Poway, CA) following the manufacturer's protocol but using 96-well plates instead of cuvettes 14 .
Serum levels of anti-mitochondrial PDC-E2 antibodies were determined via ELISA. Briefly, ELISA plates were coated with PDC-E2 protein (5 μg ml −1 , 100 μl) (SurModics Inc, Eden Prairie, MN) overnight at RT. Plates were washed, blocked using 3% dry skim milk in PBS (pH 7.4, 150 μl), and incubated with serially diluted serum samples (100 μl, at 1:250 dilutions prepared using reagent diluent) for 2 h at RT. Wells were washed and incubated with 100 μl of HRP-conjugated anti-mouse IgG (1:2000 in reagent diluent) for 2 h at RT, and washed. Finally, wells were incubated in the dark with 100 μl of DAB substrate for 20 min at RT. Upon stopping the enzymatic reaction with 50 μl 2 N H 2 SO 4 , the absorption was measured at a 450 nm wavelength using an ELISA plate reader. The positive antibody activity (PAA) levels were calculated by calculating the mean OD ± 2 SD of the control NOD serum samples (positive index) and by dividing the OD values corresponding to NOD.c3c4 serum samples by the positive index, whereby values > 1.0 correspond to PAA.
Statistical analyses. Unless specified, sample size values mentioned in the figure legends correspond to the total number of mice examined, pooled from different experiments. Data were compared in GraphPad Prism 6 using Mann-Whitney Utest, Kruskal-Wallis test, Chi-Square, Log-Rank (Mantel-Cox), Pearson correlation, two-way ANOVA or multiple t-test analyses using the Holm-Sidak correction. P-values < 0.05 were considered statistically significant. Only statistically significant P values are displayed in figures.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Raw data used to generate the figures of the manuscript are available from the authors upon request.