The lupus susceptibility allele DRB1*03:01 encodes a disease-driving epitope

The HLA-DRB1*03:01 allele is a major genetic risk factor in systemic lupus erythematosus (SLE), but the mechanistic basis of the association is unclear. Here we show that in the presence of interferon gamma (IFN-γ), a short DRB1*03:01-encoded allelic epitope activates a characteristic lupus transcriptome in mouse and human macrophages. It also triggers a cascade of SLE-associated cellular aberrations, including endoplasmic reticulum stress, unfolded protein response, mitochondrial dysfunction, necroptotic cell death, and production of pro-inflammatory cytokines. Parenteral administration of IFN-γ to naïve DRB1*03:01 transgenic mice causes increased serum levels of anti-double stranded DNA antibodies, glomerular immune complex deposition and histopathological renal changes that resemble human lupus nephritis. This study provides evidence for a noncanonical, antigen presentation-independent mechanism of HLA-disease association in SLE and could lay new foundations for our understanding of key molecular mechanisms that trigger and propagate this devastating autoimmune disease.

S ystemic lupus erythematosus (SLE) is an autoimmune disease that afflicts millions of individuals worldwide 1 . The main impediment to finding a cure for SLE is incomplete understanding of its genetic and molecular mechanisms. Salient cellular aberrations observed in SLE include enhanced endoplasmic reticulum (ER) stress, an activated unfolded protein response (UPR), mitochondrial dysfunction and aberrant cell death, associated with a pro-inflammatory state and generation of autoantibodies against nuclear antigens, which are implicated in target tissue damage, such as lupus nephritis (LN) [1][2][3][4] . The molecular mechanism that triggers this cascade of events is largely unknown.
The contributions of genetic factors to SLE risk or disease severity are, likewise, incompletely understood. Susceptibility to the disease has been associated with more than 100 loci 5,6 , with a particularly strong involvement of human leukocyte antigen (HLA) genes 7,8 . Meta-analysis of genomic studies has identified DRB1*03:01 as the single most significant SLE-associated HLA allele in Europeans 9 . The underlying mechanism, however, is unknown. The prevailing hypotheses concerning HLA-disease association postulate presentation of self or foreign antigens by HLA molecules 4,8 ; however, the antigen presentation hypotheses have not yet been empirically validated in SLE.
Seeking better understanding of the functional role of HLA molecules in SLE, we examined an antigen presentationindependent mechanism, modeled after the 'MHC Cusp' theory 10,11 , which postulates that in addition to presenting antigenic peptides to T cells, major histocompatibility complex (MHC) molecules express signal transducing ligands that, upon binding to non-MHC receptors trigger allele-specific cell activation events. Under certain environmental circumstances and permissive background genes, this activation can provoke aberrant cellular events that facilitate autoimmune diseases. At the focus of the MHC Cusp theory is an α-helical cusp-like conformational motif that is shared by all products of the MHC gene family, irrespective of their primary sequences, and independent of antigen presentation. In HLA-DR molecules, the cusp involves the third allelic hypervariable region (TAHR; residues 65-79) of the DRβ chain. In rheumatoid arthritis (RA) the cusp region that is coded by disease-associated HLA-DRB1 alleles encompasses a 'shared epitope' (SE), which has been shown to act as a ligand that activates in vitro and in vivo pro-arthritogenic events in mice [12][13][14][15][16][17] .
Here, we examined whether the TAHR of the DRβ chain coded by the SLE-susceptibility allele DRB1*03:01 may be directly contributing to SLE pathogenesis. Our findings reveal that reminiscent of SE-coding DRB1 alleles in RA, the SLE-risk allele DRB1*03:01 encodes what we designated here as a 'lupus epitope' (LE) in the TAHR of the DRβ chain that, in the presence of interferon gamma (IFN-γ), activates signature lupus transcriptomes in mouse and human macrophages, and triggers SLE-characteristic cellular aberrations, including ER stress, UPR, mitochondrial dysfunction, necroptosis and production of pro-inflammatory cytokines. In the presence of IFN-γ, primary mouse bone marrow-derived macrophages (BMDMs) from non-immunized transgenic mice that carry the LE-coding DRB1*03:01 allele exhibit similar cellular aberrations ex vivo, and in vivo administration of IFN-γ to those transgenic mice produces an SLE-like disease, including formation of antidouble-stranded DNA (dsDNA) antibodies and renal immuno-histopathology changes akin to LN.
Quantitative RT-PCR (qRT-PCR) analysis showed no significant effect on epitope-activated gene expression in the absence of IFN-γ. In its presence, however, substantial increased expression was found for representative macrophage activation gene markers ( Supplementary Fig. 1c). To characterize TAHR epitopes-driven transcriptional landscapes, and to differentiate between the respective contributions of the different epitopes versus IFN-γ, RNA-seq analyses were performed using two models: 1. Transcriptional modulation by allelic epitopes in the presence of IFN-γ, versus IFN-γ alone (Model A), or 2. Transcriptional modulation by allelic epitopes in the presence of IFN-γ, versus epitope alone (Model B).
In contrast to the upregulation of many SLE-relevant genes by 65-79*LE in RAW 264.7 macrophages, 65-79*PE downregulated many pro-inflammatory and pro-SLE genes. Representative genes illustrating the dichotomous modulation of SLE pathogenesis-related gene expression by 65-79*LE versus 65-79*PE are shown in Fig. 2c and Supplementary Table 2. In this context, it is notable that DRB1 alleles that encode a PE motif (70-DERAA-74) have been previously found to significantly reduce the risk of various autoimmune diseases, including SLE [17][18][19] .
To identify transcriptional patterns that are shared between Model A and Model B, we performed bioinformatic metanalyses encompassing compiled data sets of RAW 264.7 cells in the two models ( Supplementary Fig. 3). A substantial overlap was found between Model A and Model B, with 773 upregulated and 378 downregulated DEGs that were common between the two models in 65-79*LE-stimulated macrophages, and 573 common upregulated and 286 downregulated in 65-79*SE-stimulated macrophages (Supplementary Fig. 3a and 3b). 65-79*PE upregulated only 16 and downregulated no DEGs that were shared by the two models.
GO-BP analysis of upregulated DEGs in 65-79*LE-and 65-79*SE-stimulated cultures when Model A and Model B were meta-analyzed revealed distinct enrichment patterns between the two epitopes. As shown in Supplementary Fig. 3c, d, consistent with the disease relevance of their respective DRB1-coding alleles, epitopes 65-79*LE-activated transcriptomes were enriched in terms involving cell division, DNA damage and repair, and nuclear reorganization, whereas transcriptomes in cultures activated by 65-79*SE showed preponderance of GO terms related to proinflammatory and innate immune system signaling, as well as cellular stress response. Additionally, KEGG analysis revealed higher significance among pathways involving cell cycle and SLE in 65-79*LE-stimulated cells, while the higher statistically significant KEGG pathways terms in 65-79*SEstimulated cells included TNF signaling pathways and several TNF-dependent disease processes, including RA ( Supplementary  Fig. 3e), consistent with the established disease associations of the two respective coding DRB1 alleles.
Thus, taken together, RNA-seq analyses identified allelespecific, functionally distinct, and clinically relevant activation of transcriptional landscapes. 65-79*LE activated an SLE-relevant transcriptome, 65-79*SE activated a pro-RA transcriptome, and 65-79*PE activated an autoimmune-protective transcriptome, consistent with the respective disease associations that the three coding DRB1 alleles demonstrate in humans.
The LE activates a cascade of lupus-associated cellular aberrations. To determine the functional impact of the three TAHR epitopes we studied their effects in IFN-γ-stimulated mouse RAW 264.7 and human THP-1 macrophages. ER stress and UPR aberrations have been long implicated in SLE pathogenesis 3,28,29 . As shown in Fig. 3a, RNA-seq analyses identified many ER stress and proteasome degradation genes that were differentially regulated by 65-79*LE. We therefore determined the functional impact of 65-79*LE on these cellular processes. In the presence of IFN-γ, THP-1 cells activated by 65-79*LE, but not by 65-79*SE or 65-79*PE, showed increased expression of the ER stress marker phosphorylated inositol-requiring enzyme 1 α (pIRE1α), C/EBP homologous protein (CHOP) and glucose-regulated protein 78 (GRP78) (Fig. 3b), as well as the proteasomal degradation marker p62 and overabundant poly-ubiquitinated proteins ( Fig. 3c-e). Evidence of allele-specific 65-79*LE-triggered ER stress and UPR was found in RAW 264.7 macrophages as well ( Supplementary Fig. 4).
The ER and mitochondria are functionally entangled 30,31 . We therefore sought to characterize indicators of mitochondrial function, including ATP levels, mitochondrial membrane potential (ΔΨm) and mitochondrial reactive oxygen species (ROS) levels. In the presence of IFN-γ, macrophages stimulated with 65-79*LE, but not with 65-79*SE, 65-79*PE, or with 2 other 15mer peptides corresponding to the TAHR coded by other control alleles, DRB1*04:03 and DRB1*15:01, demonstrated significantly, and dose-dependently, decreased cellular ATP levels ( Fig. 3f and Supplementary Fig. 1). Corroborating evidence for 65-79*LEactivated mitochondrial dysfunction was the finding of an allelespecific effect of this epitope on mitochondrial membrane potential. In the presence of IFN-γ, 65-79*LE, but not 65-79*SE or 65-79*PE, significantly suppressed mitochondrial membrane potential and increased mitochondrial ROS both in human THP-1 and mouse RAW 264.7 macrophages (Fig. 3f, g and Supplementary Fig. 4). Reduction of intracellular ATP levels by 65-79*LE could be completely prevented in THP-1 macrophages by Baricitinib (Fig. 3h), a Janus Kinase inhibitor known to block IFN-I-activated pathways 32 . Additionally, a clear trend toward ATP normalization was found in cultures treated with a monoclonal antibody against IFN-I receptor, IFNAR2, as well (Fig. 3i). Thus, IFN-I appears to play a role in 65-79*LE-activated cell aberrations, consistent with the IFN-I-associated GO-BP terms enrichment effects by this epitope (Fig. 2i, j), and the known role of IFN-I in SLE.
Crosstalk between mitochondria and ER has been previously documented, with ample evidence indicating that mitochondrial dysfunction is often secondary to ER stress [39][40][41] . To determine whether the mitochondrial dysfunction observed in 65-79*LEtreated macrophages is secondary to ER stress, we used sodium 4-phenylbutyrate (4-PBA), a chemical chaperone that inhibits protein misfolding, thereby inhibiting ER stress 42 . Relevant to the present study, 4-PBA has been shown to prevent SLE manifestations in experimental lupus models 43,44 . As expected, 4-PBA inhibited 65-79*LE-activated ER stress and proteasomal degradation (Fig. 6a). Further, prevented the reduction of intracellular ATP levels (Fig. 6b), reversed mitochondrial membrane potential loss (Fig. 6c), and normalized mitochondrial ROS levels (Fig. 6d), in an allele-specific fashion. Moreover, 4-PBA inhibited 65-79*LE-activated necroptosis signaling as evidenced by decreased expression of the necroptotic protein markers RIP1 and pMLKL, as well as cell death in mouse and human macrophages (Fig. 6e-g).
To better map the specific pathway involved, we tested three selective inhibitors that target the transcription factor 6α (ATF6α)-, protein kinase R-like endoplasmic reticulum kinase (PERK)-or inositol-requiring enzyme-1 (IRE1)-mediated pathways ( Supplementary Fig. 6h). The findings suggest that inhibition of the ATF6α-, but not PERK-or IRE1-mediated pathways significantly increased cellular ATP levels, suggesting that LE-activated UPR pathway is mediated by ATF6α (Supplementary Fig. 6i-6k).
Disease-promoting effects of physiologically folded LEexpressing HLA-DR molecules in transgenic mice. To determine the functional effects of the LE in its physiologically folded conformation, we performed ex vivo and in vivo experiments in transgenic mice that express on their cell surfaces heterodimeric HLA-DR molecules consisting of a monomorphic DRα chain and a polymorphic DRβ chains coded by one of the three alleles DRB1*03:01 (LE); DRB1*04:01 (SE), or DRB1*04:02 (PE). These transgenic mouse lines are referred to as LE-Tg, SE-Tg and PE-Tg, respectively. Their cell surface-expressed HLA-DR molecules differ from each other by a few amino acid residues in the 65-79 region of the DRβ chain ( Supplementary Fig. 1b).
Genomic DNA analyses of the three Tg mouse lines using the MiniMUGA SNP array showed that the three Tg mouse lines include identical background strain compositions, with minor difference in the calculated percentage of genomic contribution by C57BL/6J (B6J), C57BL/10J (B10J) and SWR mouse strains (detailed in the Methods section). We therefore used these 3 WT mouse strains as genetic background controls in key experiments. Fig. 4 Allele-specific activation of cell death, pro-inflammatory cytokine production and DNA damage by the LE. a-c Cell death a, viability b, and ROS levels c, in THP-1 cells exposed to different allelic epitopes (n = 3). d Heatmap of DNA damage-related genes (Model B) in THP-1 cells exposed to different allelic epitopes (n = 4). e DNA damage, quantified by the comet assay, in THP-1 cells exposed to different allelic epitopes (n = 3). f-h Supernatant levels of the pro-inflammatory cytokines TNF-α f, IL-6 g, and IL-1β h, in THP-1 macrophages exposed to different allelic epitopes (n = 3). Data represent mean ± SEM. One-way (b, c, e, h), or two-way (a, f, g) ANOVA. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. See also Supplementary Fig. 5.
Consistent with the findings in mouse and human macrophage cell lines discussed above, when exposed to IFN-γ, primary bone marrow-derived macrophages (BMDMs) from LE-Tg showed evidence of ER stress and UPR activation (Fig. 7a,  b), and reduction of intracellular ATP levels (Fig. 7c). Such effects could not be seen in LE-Tg BMDM in the absence of IFN-γ, nor could they be seen in the presence or absence of IFNγ, in the control, SE-Tg or PE-Tg BMDMs, or in the 3 genetic background control WT mouse strains, B6J, B10J and SWR ( Supplementary Fig. 7). Like the effect of 65-79*LE on the mouse and human macrophage cell lines ( Fig. 4 and Supplementary Fig. 5), in the presence of IFN-γ, BMDMs from LE-Tg showed increased proinflammatory milieu, as evidenced by increased production of TNF-α and nitrite (Fig. 7d, e). Such proinflammatory effects could not be seen in LE-Tg BMDM in the absence of IFN-γ.
To ascertain that BMDMs from the three genetic background control strains are capable of responding to an exogenously added LE epitope, we measured their proinflammatory response to the allelic peptides in the presence or absence of IFN-γ. Interestingly, like mouse RAW 264.7 and human THP1 macrophage cell lines, in the presence of IFN-γ, BMDM of all the three control mouse strains (B6J, B10J and SWR) showed increased TNF-α and nitrite production in response to exogenously-added LE peptide 65-79*LE, but not to 65-79*SE or 65-79*PE (Supplementary Fig. 5f, g).
These findings demonstrate that LE-triggered proinflammatory effects are strain-and species-independent.
Given our findings of an LE-activated IFN-I transcriptome ( Fig. 2 and Supplementary Fig. 2), and the known pathogenic role of IFN-I in both human SLE and mouse models of the disease 26 , we quantified by qPCR the expression levels of salient IFN-I genes in HLA-DR humanized mice expressing on their surface LE-, or SE-positive HLA-DR molecules (LE-Tg and SE-Tg mice, respectively). As supplementary Fig. 8 demonstrates, in the presence of IFN-γ, BMDMs from LE-Tg mice, but not from control, SE-Tg mice, showed increased gene expression of several known SLE-associated IFN-I genes, such as Ifi44, Ifi44l, and Irf7, along with markedly curtailed expression of Dnase1l3, a gene known to protect against SLE in both humans and mice [45][46][47] .
IFN-γ has been long found to contribute to SLE disease development in both humans and mice 15,16 . Given the observed obligatory co-factor role of IFN-γ in vitro at the transcriptome and cell function levels as noted above, we proceeded to determining its in vivo effect in Tg mice. Intraperitoneal administration of IFN-γ resulted in increased serum levels of anti-dsDNA antibodies in LE-Tg, but not in SE-Tg mice (Fig. 7f, g), without a significant difference between male and female mice (Fig. 7h). IFN-γ-treated LE-Tg mice developed glomerulonephritis, as evidenced by proteinuria (Fig. 7i) and histopathology (Fig. 7j, k). Specifically, examination of kidney sections of LE-Tg mice treated for 24 weeks with IFN-γ showed increased glomerulitis with mesangial and/or endocapillary hypercellularity, crescent formation and immune deposits, as compared to LE-Tg mice treated with PBS, or SE-Tg mice treated with PBS or IFN-γ (Fig. 7k). In addition, kidney sections of LE-Tg mice treated with IFN-γ showed increased glomerular C3 deposits compared to LE-Tg mice treated with PBS, or SE-Tg mice treated with either PBS or IFN-γ (Fig. 7l, m).
Thus, taken together, the findings of this study demonstrate that in the presence of IFN-γ, the LE activates lupus signature transcriptomes and a cascade of lupus-associated cellular events in vitro in an allele-specific manner. When exposed to IFN-γ in vivo, naïve transgenic mice carrying the LE-coding SLE risk allele DRB1*0301 develop serological and histological findings analogous to human SLE. A proposed model is illustrated graphically in Fig. 7n.

Discussion
It has been long proposed that the mechanistic basis of HLAdisease association in lupus involves the presentation of putative self or foreign antigens that lead to an adaptive immune system-mediated target organ damage. The new insight offered by this study is that an entire cascade of SLE-associated events, including transcriptional, cellular, serological, and clinical manifestations, can be triggered, independent of antigen presentation, by a short epitope coded by allele DRB1*03:01, known as a major risk factor in human SLE.
Studying mouse and human macrophage cell lines stimulated with short synthetic peptides corresponding to the TAHR coded by three DRB1 alleles, as well as BMDMs derived from transgenic mouse lines that express physiologically folded distinct allelic HLA-DR molecules, we demonstrated allele-specific activation of a cascade of pathogenic events that includes transcriptional, cellular and disease phenomes that closely resemble human SLE and its experimental mouse models. Given that synthetic peptides are incapable of presenting antigens, and the fact that macrophage cell cultures are devoid of T cells, these findings support a diseaserelevant, DRB1 allele-specific, antigen presentation-independent mechanism.
Consistent with its known effects in human SLE and experimental models of the disease 15,16 , IFN-γ was identified here as an   Supplementary Fig. S9. Data represent mean ± SEM. One-way ANOVA (c-e, h, m), Twoway ANOVA (g), or Wilcoxon sign rank test (f, i, j). *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. obligatory co-factor for LE-activated transcriptional and cell biology effects. In this context, it is of interest that while both Model A (epitope + IFN-γ versus IFN-γ alone) and Model B (epitope + IFN-γ versus epitope alone) of RNA-seq data analysis uncovered bioinformatics evidence of a lupus signature transcriptome activation by the LE, Model B was found to be more robust in both mouse and human macrophages. One possible explanation might be that Model B more closely mimics the physiologic conditions in vivo, where the expression levels of DRB1-coded HLA-DR molecules is relatively constant, while IFN-γ levels could vary periodically, secondary to fluctuating environmental conditions. It is plausible that such changing circumstances could allow the LE to realize its SLE-triggering potential. Further research into this question, and more detailed mapping of the LE pathway and its crosstalk with IFN-γ-activated pathways, will help to dissect the respective roles of LE and IFN-γ and the mechanism of their interaction in SLE.
The LE overcame strain and species barriers, as the effects of this human gene product could be found in mouse (RAW 264.7 and BMDMs of diverse mouse strains) as well as human (THP-1) cells. Additionally, LE effects were found both in cell lines (RAW 264.7 and THP-1) and primary cells (BMDMs). This versatility attests to the fundamental nature of the mechanisms involved in LE-activated events. It is noteworthy that the MHC Cusp theory discussed below is partly based on the hypothesis that class II β chain TAHRs may have preserved fundamental, non-antigen presenting functions through evolution, consistent with the principles of trans-species polymorphism 11,48,49 .
Different from the strong female preponderance in human SLE, here, LE effects on anti-dsDNA antibody levels showed no difference between male and female mice. The reason for this apparent discrepancy is unknown and requires further study; however, it is worth mentioning that although a higher incidence and severity of SLE disease is well documented in female mice in some experimental models such as NZBW 50 , sex bias is less prominent in the MRL/lpr model 51 . Notably, a large genomewide association study involving genotype data from over 27,000 individuals identified no gender-based differential SLE risk for any HLA allele 52 , suggesting that gender bias in humans involves mechanisms influenced by non-HLA risk factors.
The penetrance of nephritis in IFN-γ-treated LE-Tg mice was low, but statistically significant. It is worth noting that anti-dsDNA antibody formation, proteinuria histopathology and immunofluorescence findings, specifically in LE-Tg mice, were all consistent with LN development. We submit that LE-Tg + IFN-γ may not be a robust model of murine SLE, and there are several popular models with higher penetrance and disease severity. However, our LN data are not presented to propose a new robust lupus model, but rather to substantiate the disease relevance of the in vitro data. Susceptibility and disease severity in SLE are multifactorial. It might be helpful to explore in future studies whether the LE can act as a disease facilitator in established models of SLE.
This study has focused on DRB1*03:01, the best characterized and the single most significant (OR = 1.87; p = 1.17 × 10 −58 ) 9 SLE-associated HLA allele. However, it should be mentioned that SLE has been reported to associate with other HLA genes as well. While some of these associations are attributable to the strong linkage disequilibrium in the MHC region, DRB1*15:01 has been found to independently associate with the disease as well (OR = 1.33; p = 1.92 × 10 −12 ) 9 . The mechanistic basis of that association is unknown and is beyond the scope of this first study. It is worth noting, however, that a recent association study of SLE using Immunochip genotyping data 52 identified differential risk associations at a single amino acid residue level between the DRB1*15 allele, in which risk-associated residues were found primarily inside the peptide-binding pocket, and allele DRB1*03:01, in which risk-associated residues were found primarily at the region 70-77 of the DRβ chain, overlapping with the cusp region (amino acid residues 65-79).
The distinct topologies of SLE-critical amino acid residues coded by the two alleles, along with our findings that the 65-79*LE, but not an equivalent peptide corresponding to DRB1*15:01, suppressed ATP cellular levels ( Supplementary  Fig. 1d), together suggest that the functional effects on disease risk by DRB1*03:01 and DRB1*15:01 are dissimilar. While allele DRB1*03:01 encodes a disease-driving LE, as demonstrated here, allele DRB1*15:01 may contribute to disease risk through antigen presentation. The apparent disparities between the respective functional effects of the two alleles on disease risk might also provide a mechanistic basis for an imputation-based conclusion that DRB1*03:01/DRB1*15 heterozygosity confers a greater risk for SLE than either DRB1*03:01/DRB1*03:01 or DRB1*15/ DRB1*15 homozygosity 52 .
Different from the empirical data-based model presented here, the above-mentioned gene association study suggested that SLE disease risk conferred by allele DRB1*03:01 is attributable to a single amino acid residue located at the peptide binding groove: Tyr26 in the risk allele DRB1*03:01 versus Phe26 in the non-risk allele, DRB1*03:02, indirectly implicating antigen presentation. That conclusion was offered because both alleles share an identical amino acid sequence in the TAHR but differ by a single amino acid residue in position 26, located at the DR3 peptide groove 52 .
That hypothesis is seemingly in conflict with our findings. However, despite continuous refining efforts, the accuracy of single nucleotide polymorphism-based imputation of HLA genotyping, especially in respect to the DRB1 locus 53,54 , remains uncertain. Nonetheless, irrespective of the reliability of computational HLA fine genotyping, the theory examined here and the antigen presentation hypothesis as mechanistic bases for HLAdisease association are mutually non-exclusive, as previously discussed 10,11,20,55 . Indeed, targeted sequencing efforts to examine the genetic association within the HLA class II region in lupus identified a regulatory genetic variant located between HLA-DRB1 and HLA-DQA1 that could explain the majority of the genetic risk in this locus 56 . This effect is tagged by the SNP rs9271593 within an intergenic regulatory locus known as XL9 separating the HLA-DRB1 and HLA-DQA1 genes. The lupusassociated variant in this SNP tags the lupus-associated classical alleles DRB1*03:01 and DRB1*15:01, and is associated with increased expression of HLA-DRB1, HLA-DQA1, and HLA-DQB1 in monocyte-derived dendritic cells 56 . These findings support an antigen presentation independent effect of the HLA class II genetic effect in lupus, consistent with our findings.
It should be noted that DRB1*03:01 is associated with many other autoimmune diseases besides lupus. We submit that as demonstrated here, the LE activates several fundamental cell biology aberrations (e.g., UPR, mitochondrial perturbations), which are implicated in many diseases besides SLE. Additionally, SLE risk, like many other autoimmune conditions, depends on multiple genes, as well as environmental factors and disease phenomes are likely determined by intricate gene-gene and geneenvironment interactions. The fact that KEGG pathway transcriptome analyses of LE-activated macrophages identified several other diseases that are known to associate with DRB1*03:01 in both human and mouse macrophages (Fig. 2f) lend further support to the mechanism proposed here.
This study demonstrates an allele-specific, antigen presentation-independent cascade of events that culminate in the development of serological and tissue damage akin to human SLE. The trigger of this cascade of events involves a short epitope coded by HLA-DRB1*03:01, the single most significant genetic risk factor for SLE in diverse populations. Our model (Fig. 7n) proposes that a combination of necroptotic cell death with increased abundance of free DNA fragments plus a proinflammatory milieu, together stimulate anti-dsDNA antibody formation. The model further proposes that those antibodies can form immune complexes with DNA fragments and precipitate in glomeruli, and cause LN. Importantly, the curtailed IFN-γactivated transcription of Dnase1l3, -a gene that codes for an extracellular DNA degradation enzyme, whose deficiency is strongly implicated in the pathogenesis of human SLE 47 and murine disease models 46 -in BMDM of LE-Tg ( Supplementary  Fig. 8) suggests a possible facilitating mechanism of anti-dsDNA antibody formation by increasing antigenic (DNA) exposure. This scenario deserves further research.
The findings of this study might open the door to better understanding of the precise molecular and genetic mechanisms involved in HLA-disease association. Exploration of such mechanisms in the context of human diseases could potentially identify new preventive, diagnostic, and therapeutic solutions.

Methods
Mice. Transgenic (Tg) mice were generated as previously described 24,57,58  Transgenic lines LE-Tg, SE-Tg and PE-Tg were developed using classical pronuclear microinjection. For line LE-Tg, the transgene was injected into (C57BL/ 6 x DBA/2)F1 embryos 59 . Lines SE-Tg and PE-Tg were created injecting (C57BL/ 10 x SWR)F1 embryos 57,58 . Genomic DNA from four representative mice from each line was analyzed using the MiniMUGA SNP array, containing~9,700 SNPs 60 . The three lines showed to be highly inbred (98.5% to 99.5% homozygous genotypes) and mice from each line were almost genetically identical.
In agreement with the history of breeding the genomes of the three lines are very similar and resemble recombinant inbred strains with C57BL/6J (B6J) and C57BL/10J (B10J) as main backgrounds. After the analysis of the 366 polymorphic SNPs present in the MiniMUGA array it was estimated that the percentage backgrounds were 55.6% C57BL/10J and 44.4% C57BL/6J for line LE-Tg; 59.6% C57BL/10J and 40.4% C57BL/6J for line SE-Tg; and 69.1% C57BL/10J and 31.5% C57BL/6J for line PE-Tg. Using a special function from a proprietary spreadsheet developed by Transnetyx, we could rule out the presence of "diagnostic SNPs" from C57BL/6N (NIH substrain) in the three Tg lines. With the same tool we could also rule out traces of DBA/2 alleles in line 0301. However, we estimated the presence of a very small (~1%) proportion of SWR/J genome in lines SE-Tg and PE-Tg.
All mice were housed under specific pathogen-free and temperature-controlled (25°C) conditions with a 12 h light-dark cycle. Ex vivo experiments were carried out using 10-12-week-old male mice. For in vivo experiments, female or male mice were used at 8 weeks of age. Mice were administered intraperitoneal (IP) injections containing 100 μl of either PBS, or recombinant mouse IFN-γ (10 µg per injection) twice a week. All animal experiments were approved by the Institutional Animal Care & Use Committee (IACUC) at the University of Michigan.
Reagents. All materials used, along with vendor names and catalog numbers are listed in Supplementary Table 5.
RNA-Sequencing. Samples for RNA-seq were prepared and processed as previously described 20 . Briefly, RAW 264.7 cells were treated with 15mer ligands 65-79*LE, 65-79*SE or 65-79*PE (100 μg/mL) with or without IFN-γ (5 ng/mL). Media were refreshed after 48 h, and at 72 h, cells were lysed with Trizol (Invitrogen). Total RNA was isolated using the Direct-Zol TM RNA miniprep kit (Zymo Research) according to manufacturer's instructions. RNA from THP-1 cells was isolated 72 h after treatment with 15mer ligands 65-79*LE, 65-79*SE or 65-79*PE (100 μg/mL) with or without IFN-γ (10 ng/mL) using the RNeasy mini kit (Qiagen) according to the vendor instructions. Genomic DNA was removed using the Turbo DNA-Free kit (Ambion). All samples used for library generation had an RNA integrity number of 7.5 or higher.
The TruSeq Stranded mRNA Sample Preparation Kit (Illumina) was used for library generation, according to manufacturer's protocol, using 200 ng of total RNA. For each sample, 100 bp single-end reads were generated using Illumina's HiSeq2500v4. Raw read counts were obtained using featureCounts from the Rsubread1.5.0p3 package and Gencode-M12 gene annotations (RAW 264.7) or featureCounts from the Rsubread-1.6.1 package and Gencode28-hg38 gene annotations (THP-1) using only uniquely aligned reads. Data normalization and differential expression analysis were performed using DESeq2-1.16.1 within R-3.4.1 using an adjusted p-value threshold of 0.05. Mouse gene annotations were verified using the MGI database (http://www.Informatics.jax.org/index.shtml) 20 , only protein-encoding genes were considered. Differentially expressed genes with a fold change (FC) of > 1.5 and adjusted p-value (Padj) of < 0.05 were used for Gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis using the DAVID bioinformatics database (version 6.8, https://david. ncifcrf.gov/) 62 . GO and KEGG terms with a false discovery rate (FDR) of < 5% (< 0.05) were considered significant. Unsupervised clustering of DEGs was visualized using a web server heatmapper (http://www.heatmapper.ca) 63 using average linkage and Pearson correlation distance.
Upstream regulator analysis. Upstream regulator analysis was performed using Advaita Bio's iPathwayGuide (iPG, Advaita Bioinformatics, MI, https://advaitabio. com/ipathwayguide/) 64 . All differentially expressed genes were used to predict upstream regulators that are likely to be activated. Upstream regulators with an adjusted p-value of <0.05 were included in the analysis.
qRT-PCR. cDNA was synthesized from total RNA using High-Capacity cDNA Reverse Transcription Kit (Thermo Fisher) according to the vendor's protocol. qRT-PCR was performed on a StepOnePlus or a CFX384 Touch (Bio-Rad) Real-Time PCR system, with a total reaction volume of 10 or 20 µL using Fast SYBR TM Green Master Mix (all reagents from Applied Biosystems). The primers used are listed in the KEY RESOURCES table. Analysis was done with StepOne or CFX Maestro Software using the ΔΔCT method.
Cell death and viability assays. Lactate dehydrogenase (LDH) release from cells was measured using the Cytotoxicity Detection Kit PLUS (Roche), according to the manufacturer's protocol. Absorbance was measured at 490 nm using a microplate reader (Bio-Tek, Synergy™ H1). Cytotoxicity was calculated as described 65 .
Cell viability was evaluated using the methylthiazolyldiphenyl-tetrazolium bromide (MTT) method 66,67 . Briefly, cells were rinsed with PBS and a 0.5% (w/v) MTT solution was added for 3 h. Next, formazan crystals formed were solubilized with isopropanol (Sigma) and optical density was measured at 540 nm using a microplate reader (Bio-Tek, Synergy™ H1).
Mitochondrial function assays. Intracellular ATP levels were measured with the ATPlite Luminescence Assay System (PerkinElmer) following the vendor protocol. Mitochondrial membrane potential was determined using the tetramethylrhodamine (TMRE) dye (Invitrogen), following the manufacturer's protocol. Mitochondrial ROS was measured using MitoSOX Red superoxide indicator (Molecular Probes) following the supplier's protocol.
Immunoblots. Cells were collected and lysed with RIPA buffer (Sigma) containing phosphatase inhibitor phosSTOP and protease inhibitors cOmplete mini EDTAfree (Roche). Protein concentration was quantified using RC DC™ Protein Assay Kit (Bio-Rad). Proteins were separated using SDS-PAGE Gels (Thermo Fisher) and transferred to PVDF Western Blotting membranes (Roche). The membranes were blocked with 5% bovine serum albumin (BSA, Sigma). Blots were incubated overnight with primary antibodies listed in the KEY RESOURCES table, followed by 1 h incubation with appropriate secondary antibody. Membranes were developed using SuperSignal TM West Pico Plus ECL substrate (Thermo Scientific) or Clarity TM Western ECL Substrate (Bio-Rad). Images were acquired using the Omega Lum™ C Imaging System (Gel Company) or ChemiDoc™ Imaging System (Bio-Rad).
Immunocytochemistry for ubiquitinylated proteins. Cells were cultured on cover glass and fixed with a 4% paraformaldehyde solution. Fixed cells were subsequently permeabilized with 0.1% triton-X100, blocked in 2% BSA, followed by overnight incubation with mono and polyubiquitinylated conjugates monoclonal antibody (1:200) and 1 h incubation with secondary antibody, labelled with Alexa Fluor 647. Coverslips were mounted onto slides using Prolong Diamond with DAPI (Invitrogen). Cells were imaged using a Nikon E800 Epifluorescence microscope.
Quantification of DNA damage. DNA damage was quantified using a Comet Assay (Trevigen), based on previously described protocols 68,69 . The slides were observed by fluorescence microscopy (Cytation 5, GFP filter). A total of 100 cells in each slide were scored according to the length and intensity of the tail in relation to the head as described 68,69 . Each cell was scored on a scale of 0 to 4 by two independent observers blinded with regards to the samples and the average from both counts was used for calculation 69 .
Quantification of cytokines, nitrites, and Serum anti-dsDNA-IgG levels. Cell supernatants were used for quantification of cytokines (IL1-β, IL-6 and TNF-α) by commercial ELISA kits following the protocols provided by the manufacturer (KEY  RESOURCES table). To measure nitrite concentrations, cell culture supernatants were collected and immediately subjected to a Griess Reagent assay (Promega), according to the suppliers' protocol.
Serum anti-dsDNA-IgG levels were measured at week 5, 10, 16, 20 and 24 using an ELISA Kit (Alpha Diagnostic International) according to the manufacturer's instructions.
Renal histopathology and immunofluorescence. Renal histopathology was interpreted as described before 70 . Briefly, mouse kidneys were harvested after 24 weeks of treatment, fixed in formalin, and 3 µm sections were stained with periodic acid-Schiff (PAS) or hematoxylin and eosin (H&E). The kidneys were scored for glomerulitis, extracapillary proliferation and mesangial expansion by a pathologist (E.F.), who was blinded with regards to the experimental groups. Images were taken at a 400 × magnification, with a BX41 microscope (Olympus) and an Olympus DP73 camera using cellSens imaging software.
Complement deposition was quantified by C3 immunofluorescence. Five µm frozen sections were stained with FITC-conjugated anti-C3 (Immunology Consultants Laboratory) for 1 h at 4˚C, followed by Hoechst (Invitrogen) counterstaining. Images were captured by a Nikon E800 Epifluorescence and Brightfield microscope (Nikon, Melville, NY) using a 10x objective. Glomerular C3 staining was scored on a 0-3 basis in a blinded manner in at least eight glomeruli per section. Positive controls were identically processed using frozen kidney sections from 28 weeks old female New Zealand Mixed (NZM) 2328 mice.
Statistics and Reproducibility. Raw data related to all graphed figures are specified in Supplementary Data File 4. Unless otherwise specified, data are presented as mean value ± standard error of the mean (SEM). All statistical analyses were performed using the Graph Pad Prism 8.0 statistical software (Graph Pad Software: https://www.graphpad.com/). One-way, or two-way ANOVA were used for analysis of normally distributed in vitro data. The statistical significance of the differences of anti-dsDNA IgG (Figs. 7f, h) levels and albumin-to-creatinine ratios (ACR) (Fig. 7i) among the different in vivo mouse treatment groups was analyzed by a Wilcoxon sign rank test, using the measured levels in PBS-treated LE-Tg, and the measured levels in IFN-γ-treated SE-Tg mice as the theoretical medians for statistical analysis compared to the measured levels in IFN-γ-treated LE-Tg mice, under the assumption that treatment of LE-Tg with IFN-γ induces higher levels. The significance of the differences in renal histopathology scores among the different mouse treatment groups (Fig. 7j) was analyzed by a Wilcoxon sign rank test as well, using a theoretical median of zero under the assumption that the various treatments do not induce any glomerular pathology. For all experiments, the statistical test used and the number of replicates are indicated in the figure legends. pvalues < 0.05 were considered statistically significant.