NF-κB RelB suppresses the inflammatory gene expression programs of dendritic cells by competing with RelA for binding to target gene promoters

Introduction

Current studies estimate that there are ~150 life-long autoimmune diseases that are characterized by dysregulation of the adaptive immune system with no known cures¹, affecting ~5%–8% of the world population². There is also a growing list of over 40 genetically encoded autoinflammatory diseases characterized by the dysregulation of innate immune responses³. One of the most common immune response pathways associated with autoimmune and autoinflammatory diseases is NF-κB^4,5,6,7,8. NF-κB plays a central role in inducing the expression of genes involved in cell survival, differentiation, and inflammation^{9,10,11,12,13,14}. The NF-κB family is comprised of multiple hetero- and homo-dimeric transcription factors that are made up of the subunits RelA, cRel, RelB, p52, or p50^15,16. Genetic loss of the RelB subunit in mice results in multi-organ inflammatory pathology characterized by several phenotypic lesions, including (1) mixed inflammatory cell infiltration in organs such as the lung and liver, (2) splenomegaly due to extramedullary hematopoiesis, and (3) thymic atrophy¹⁷. We previously identified a pronounced dysregulation of interferon (IFN)-stimulated gene (ISG) expression in fibroblasts obtained from a patient with combined immunodeficiency (CID) and autoimmunity due to a rare homozygous RELB mutation that results in complete loss of RelB protein^18,19,20. This dysregulation in the IFN-stimulated gene program was also seen in dendritic cells (DCs) derived from RelB⁻^/⁻ mice. Several studies have implicated the loss of RelB in DCs as a key driver of autoimmunity in RelB⁻^/⁻ mice, demonstrating the sufficiency of either the absence or presence of RelB in DCs alone in promoting or suppressing the inflammatory pathology, respectively, in the lung, liver, spleen and thymus^21,22,23. Transcriptomic profiling in DCs revealed large-scale upregulation of ISGs, raising the possibility that pathology arising from loss of RelB may be an interferonopathy, a group of inherited autoinflammatory diseases characterized by dysregulation of the IFN pathway^24,25,26,27. However, genetic ablation of IFN signaling in IFNAR^−/−RelB⁻^/⁻ mice did not improve critical aspects of the RelB^−/− pathology²⁰. Instead, further analysis of DCs derived from IFNAR⁻^/⁻RelB⁻^/⁻ mice revealed a subset of dysregulated NF-κB motif-containing pro-inflammatory genes that are likely drivers of RelB⁻^/⁻ pathology, and suggested that it is in fact a relopathy (an NF-κB/Rel-associated autoinflammatory disease)^28,29,30. The observations that NF-κB proinflammatory genes were hyper-expressed in the setting of RelB loss in patient-derived fibroblasts, and that NF-κB signaling was dysregulated in a separate study of patients harboring a different RELB mutation leading to reduced RelB protein³¹, together call for further investigation of how the loss of RelB activity leads to the dysregulation of NF-κB motif-containing pro-inflammatory genes.

RelB is thought to both activate and repress transcription of metabolic and inflammatory genes, while RelA is a potent transcriptional activator that induces inflammatory gene expression^32,33,34. In the absence of stimulation, RelB is localized to both the nucleus and cytoplasm. RelB can form dimers with RelA that are non-functional in binding DNA or IκBs³⁵, presumably because they have an unusually intertwined quaternary structure³⁶. In the cytoplasm, RelB can also stabilize p100/p105 interactions that sequester RelA in a non-productive complex^{37,38,39,40,41,42}. In the nucleus, RelB can repress gene expression activity by competing with RelA for binding κB sites⁴³ and potentially facilitate repressive chromatin modifications via histone deacetylase and histone methyltransferase activities^44,45. RelB is also implicated in regulating DNA methylation via the Daxx protein⁴⁶ and in increasing IκBα stability⁴⁷. However, it remains unclear which of these mechanisms is responsible for mediating autoimmunity in RelB-deficient patients and mice^18,19. Given the direct role of RelB loss in DCs in promoting autoimmune pathology, the role of DCs in other autoimmune diseases^{21,22,23,48,49,50,51}, and the high expression of RelB in healthy mouse and human DCs^52,53,54, we sought here to identify the mechanism by which RelB represses inflammatory gene expression in DCs. We found that RelB is in fact associated with pro-inflammatory genes in DCs prior to maturation stimulation. Upon stimulation, RelA is recruited to NF-κB response genes, and this recruitment is potentiated in RelB⁻^/⁻ DCs at locations that were bound by RelB prior to stimulation. To test the role of RelB DNA binding in functional DC responses, we then generated a RelB DNA binding mutant mouse strain (RelB^DB/DB), in which RelB lacks three amino acid side chains that make specific contacts with κB site nucleotides. Phenotypic analysis of these mice and genome-wide RNA-seq and ChIP-seq analyses performed in DCs derived from them, showed that the RelB DNA binding mutant is defective in suppressing the inflammatory pathology. Our results support a model in which RelB competes with RelA for κB sites, and thus RelB functions to dampen the expression of immune response genes in DCs and reduces the risk for auto-inflammatory and autoimmune disease.

Results

Characterization of IFN-independent multi-organ inflammation in RelB
⁻
^/
⁻ mice

Loss of RelB in mice results in multi-organ inflammatory pathology, including mixed lymphocytic and granulocytic infiltrates in the lung and liver, splenomegaly due to increased extramedullary hematopoiesis, and thymic atrophy¹⁷. Transcriptomic studies performed in both patient fibroblasts harboring a loss-of-function RELB mutation and DCs from RelB⁻^/⁻ mice revealed a pronounced dysregulation of ISG expression²⁰, raising the possibility that RelB deficiency may be characterized as an interferonopathy. However, we recently showed that genetic ablation of type I IFN signaling in an IFNAR⁻^/⁻RelB⁻^/⁻ compound knockout mouse strain did not improve critical aspects of RelB-deficient pathology, including unchanged splenomegaly, runted growth, and elevated serum cytokine levels²⁰. These surprising findings have prompted us to examine other aspects of RelB deficiency including thymic atrophy and visceral inflammation that have been previously reported^17,22,23. As previously reported, IFNAR⁻^/⁻RelB⁻^/⁻ mice showed runted growth compared to IFNAR⁻^/⁻ controls (Fig. 1a), and displayed splenomegaly²⁰. Thymi from IFNAR⁻^/⁻RelB⁻^/⁻ mice were smaller in size (Fig. 1b) and weight (Fig. 1c) than IFNAR⁻^/⁻ controls, indicating thymic atrophy similar to that previously described in RelB⁻^/⁻ mice¹⁷. Histologic sections of the lung and liver appeared unremarkable in wild-type (WT) and IFNAR⁻^/⁻ mice; in contrast, there was moderate to marked mixed lymphocytic and granulocytic inflammation arranged in diffuse and perivascular distributions in the lungs and livers of IFNAR⁻^/⁻RelB⁻^/⁻ mice (Fig. 1d, e), again similar to the previously described findings for RelB⁻^/⁻ mice¹⁷. Previous studies have shown that RelB deficiency limited to DCs is sufficient to drive inflammatory pathology in the liver, spleen, lung, and thymus of RelB⁻^/⁻ mice^21,22,23. While IFN signaling was shown to be dysregulated in DCs derived from RelB⁻^/⁻ mice²⁰, the phenotypic and histopathologic features of IFNAR⁻^/⁻RelB⁻^/⁻ mice provide further evidence that the pathology arising upon RelB deficiency is largely driven independently of IFN signaling.

Loss of RelB leads to elevated RelA binding at genomic RelB-bound κB sites

Considering these findings, we focused our attention on the previously identified NF-κB-driven pro-inflammatory gene expression that remained dysregulated in bone marrow-derived dendritic cells (BMDCs) derived from IFNAR⁻^/⁻RelB⁻^/⁻ mice²⁰ and was also dysregulated in patient-derived fibroblasts from RelB-null patients²⁰. We asked whether RelB and RelA may compete for binding to the same κB sites, as RelB-containing heterodimers have shown DNA binding specificities similar to RelA heterodimers⁵⁵^,56. We approached this question by first examining whether the loss of RelB results in altered RelA binding at NF-κB motifs that would normally be bound by RelB in WT conditions. To answer this, we generated BMDCs from both RelB⁻^/⁻ and WT mice and performed ChIP-seq for the NF-κB subunits RelA and RelB. We first analyzed the distribution of the relative binding signal of all RelA binding sites between RelB^−/− and WT BMDCs and found that they were normally distributed and centered at a 1:1 ratio. Binding sites at the 90th percentile displayed a log₂fold change (FC) of 0.62 in binding between RelB^−/− and WT BMDCs, whereas binding events in the bottom 10% displayed a log₂FC of –0.74 (Fig. 2a; Supplementary Table S1). To assess whether sequences bound by RelB in WT BMDCs display altered RelA binding in RelB^−/− BMDCs, we selected RelB binding events induced > 2-fold by LPS stimulation (10 ng/mL) in WT BMDCs and analyzed the subsequent changes in RelA binding at these locations in RelB^−/− BMDCs. K-means clustering recognized three predominant RelA binding patterns at these sites (Fig. 2b). Cluster A (n = 1580) contained binding events that have elevated binding by RelA in RelB^−/− BMDCs with an average of log₂FC = 0.53 between RelB^−/− and WT BMDCs (Fig. 2b, c; Supplementary Table S1). Cluster B (n = 2135) contained RelA binding events that were unchanged in RelB^−/− BMDCs with an average of log₂FC = −0.08 between RelB^−/− and WT BMDCs (Fig. 2b, c; Supplementary Table S1), and binding events in cluster C (n = 719) displayed a reduction in RelA binding with an average of log₂FC = −1.09 between RelB^−/− and WT BMDCs (Fig. 2b, c; Supplementary Table S1). Furthermore, motif analysis of sequences within these clusters confirmed the specificity of these binding events to RelA and RelB with ~60%–80% of peaks within each cluster containing NF-κB binding elements (Fig. 2d). These data reveal that while the majority of RelA DNA binding events are unchanged in the absence of RelB, the loss of RelB affects binding of RelA to a portion of NF-κB binding sites that are also bound by RelB in a WT setting. Gene ontology (GO) analysis of genes annotated to genomic sites in each of the clusters revealed “NF-kappaB signal transduction” (Cluster A, P-value = 7.75E–03; Cluster B, P-value = 2.25E–02), “innate immune response” (Cluster A, P-value = 5.61E–03; Cluster B, P-value = 1.84E–02) and “inflammatory response” (Cluster A, P-value = 3.76E–02; Cluster B, P-value = 4.88E–04) terms enriched in genes from clusters A and B; these terms were absent from genes in cluster C. Instead, the top terms in cluster C were “heart morphogenesis” (P-value = 4.95E–02), “cell junction disassembly” (P-value = 3.34E–02), and “NK cell-mediated cytotoxicity immune response” (P-value = 1.18E–02). These data suggest that RelA and RelB may compete throughout the genome for binding to NF-κB binding elements at or near genes with known inflammatory functions.

**Fig. 2: Loss of RelB leads to elevated RelA binding to a subset of genomic κB sites.**

Elevated RelA binding to promoter regions is correlated with elevated gene expression

We next sought to understand whether sites with elevated RelA binding in the absence of RelB are enriched at or near genes that are hyper-expressed in RelB^−/− BMDCs. We first annotated all RelA binding events to the nearest gene, resulting in some genes being annotated to several RelA binding events. Since we were specifically interested in RelA binding events that showed elevated binding as a consequence of the loss of RelB and associated with response to stimulation, we filtered our data to assign each gene to only the one RelA binding event that displayed the highest FC between RelB^−/− and WT BMDCs. After filtering, we found that the ratio between RelA binding events in RelB^−/− vs WT BMDCs was normally distributed and centered around 1.26. RelA binding events in the top 90th percentile displayed log₂FC = 0.66 and the bottom 10th percentile had a log₂FC = –0.22 (Fig. 3a; Supplementary Table S2). Using RNA-seq data sets from WT and RelB^−/− BMDCs, we then asked whether the pro-inflammatory genes previously reported to be hyper-expressed in RelB^−/− BMDCs contain NF-κB regulatory elements that resulted in elevated binding by RelA. To answer this, we selected genes induced > 2-fold upon TLR stimulation in either WT or RelB^−/− BMDCs, and then plotted the RelB^−/− hyper-expression phenotype (x-axis) against the elevated RelA binding phenotype (y-axis) and highlighted the previously reported IFN-independent NF-κB pro-inflammatory genes that were hyper-expressed in RelB^−/− BMDCs (red points) (Fig. 3b). We found that 98 of 117 (84%) of these genes fell within the elevated binding and hyper-expressed right upper quadrant (Fig. 3b). Further, loss of RelB resulted in a statistically significant increase (P < 0.01) in RelA binding at regulatory regions of genes that are hyper-expressed in an IFN-independent manner over genes whose expression is unchanged between RelB^−/− and WT BMDCs (Fig. 3c). We then compared the genome browser tracks of RelA binding events at or near the promoter regions of previously reported IFN-independent NF-κB pro-inflammatory genes that were hyper-expressed in RelB^−/− BMDCs²⁰. While many RelA binding events were located at the promoter regions of these genes, the average distance between RelA binding events and their respective transcription start sites (TSSs) was −1352 bp (Supplementary Table S3). We observed that the loss of RelB resulted in elevated binding by RelA at or near the TSS of previously reported NF-κB pro-inflammatory genes including Cd40, Ikbke, Map3k14, Ccl5, Ccl22, and Cd80 (Fig. 3d). These data provide evidence that RelA and RelB may compete for NF-κB binding sites, and that in a RelB-deficient context, hyper-expressed pro-inflammatory genes have elevated binding by RelA at or near their respective TSSs.

A targeted mutant of the RelB DNA binding domain also shows autoimmune pathology

These data suggest so far that RelB may suppress the NF-κB pro-inflammatory gene expression in BMDCs, and potentially the inflammatory pathology, via competition with RelA for binding to sites near the regulatory promoter regions of these genes. To test this hypothesis further, we generated a novel RelB mutant mouse strain (RelB^DB/DB) with the targeted ablation of its high-affinity DNA binding. Using the previously reported crystal structure of RelB that reveals the protein–DNA binding interface⁵⁷, we designed and characterized RelB variants with a single amino acid substitution (RelB^Y120A) converting Tyr120 to alanine (Y120A), or a triple substitution within the amino-terminal DNA binding domain of RelB, converting Arg117, Tyr120, and Glu123 to alanine (R117A, Y120A, and E123A) (Fig. 4a). After we confirmed that the ectopic expression of these RelB protein variants was similar to that of WT RelB protein in RelB-deficient 3T3 cells (Supplementary Fig. S1), we measured their ability to bind DNA via electrophoretic mobility shift assay (EMSA). We found that WT RelB protein was able to bind κB sites, while both single RelB^Y120A and triple RelB^{R117A/Y120A/E123A} mutant proteins were unable to bind to κB sites, confirming the directed loss of DNA binding function (Fig. 4b). To ensure maximal reduction in DNA binding affinity, we decided to use the triple mutant design for generating the RelB DNA binding mutant mouse (RelB^DB/DB).

**Fig. 4: RelB requires its DNA binding function to suppress autoimmune pathology.**

Upon knockin embryonic stem cell generation, blastocyst complementation, germline transmission, and breeding to homozygosity, RelB^DB/DB mutant mice appeared runted and smaller than WT littermates. At 8 weeks of age they displayed a hunched posture, scaly skin, and enlarged abdomens, similar to RelB^−/− mice¹⁷ (Fig. 4c). RelB^DB/DB mice showed marked splenomegaly as reported in RelB^−/− mice, with spleens from RelB^DB/DB mice weighing on average 1.7× more than the spleens from WT controls (Fig. 4d). To determine whether this phenotype was type I IFN-dependent or -independent, we bred the new RelB^DB/DB mutant strain with the IFNAR^−/− mouse strain to produce a compound mutant. Similar to our previous findings²⁰, splenomegaly was found to be independent of IFN signaling with spleens from IFNAR^−/−RelB^DB/DB mice weighing more than spleens from WT and single IFNAR^−/− mice on average by 2.1- and 1.9-fold, respectively (Fig. 4d). Blinded histologic analysis of spleens from RelB^−/− and IFNAR^−/−RelB^−/− mice revealed a severe contraction in white pulp, expansion of red pulp, and increased extramedullary hematopoiesis when compared with WT and IFNAR^−/− controls (Supplementary Figs. S2, S3), consistent with prior reports^17,20. Analysis of spleens from RelB^DB/DB mice similarly displayed an intermediate contraction in the white pulp, expansion of red pulp, and increased extramedullary hematopoiesis compared to WT (Supplementary Figs. S2, S3). Spleens from IFNAR^−/−RelB^DB/DB mice also revealed an intermediate contraction in white pulp; however, little to no expansion of the red pulp nor increase in extramedullary hematopoiesis was noted when compared with IFNAR^−/− controls (Supplementary Figs. S2, S3). In the lungs, blinded histologic analysis revealed moderate to marked perivascular mixed lymphocytic and neutrophilic infiltrates in RelB^−/− and IFNAR^−/−RelB^−/− as previously reported^17,20, while no significant inflammation was noted in WT and IFNAR^−/− controls (Fig. 4e; Supplementary Fig. S4). Lung histology in RelB^DB/DB and IFNAR^−/−RelB^DB/DB mice similarly showed moderate to marked perivascular inflammation, yet contained predominantly lymphocytic infiltrates (Fig. 4e; Supplementary Fig. S4). Consistent with lung findings and prior studies^17,20, blinded histologic analysis of livers from RelB^−/− and IFNAR^−/−RelB^−/− mice revealed moderate to marked mixed lymphocytic/neutrophilic inflammation in periportal and centrilobular patterns (Fig. 4f; Supplementary Fig. S5). Again, RelB^DB/DB and IFNAR^−/−RelB^DB/DB livers showed a similar degree of inflammation to RelB^−/− and IFNAR^−/−RelB^−/− livers, but infiltrates were predominantly composed of lymphocytes (Fig. 4f; Supplementary Fig. S5). Thymi from RelB^DB/DB mice weighed less than WT controls, indicating thymic atrophy as seen in RelB^−/− mice (Supplementary Fig. S6). Thymi from IFNAR^−/−RelB^DB/DB mice showed size variation intermediate between RelB^DB/DB and WT controls (Supplementary Fig. S6). Finally, in the serum, RelB^DB/DB and IFNAR^−/−RelB^DB/DB mice showed elevated levels of both pro-inflammatory cytokines IP-10 and IL-6, consistent with elevated levels previously reported in RelB^−/− mice^20,22,58 (Supplementary Fig. S7). Together, this phenotypic and histological analysis of the RelB^DB/DB mutant strain provides compelling evidence that the DNA-binding mutant of RelB fails to suppress critical aspects of the autoimmune pathology associated with the loss of RelB.

RelB DNA binding mutant DCs hyper-activate pro-inflammatory gene expression

We then addressed whether the RelB DNA binding mutant is able to suppress the induction of IFN-dependent and/or IFN-independent pro-inflammatory genes that are hyper-expressed in RelB^−/− mice. We compared BMDCs derived from WT, RelB^−/−, and RelB^DB/DB, as well as IFNAR^−/−, IFNAR^−/−RelB^−/−, and IFNAR^−/−RelB^DB/DB compound mutant mice in which type I IFN signaling is ablated yet IFN-independent gene expression remains intact. We stimulated BMDCs from all genotypes, collected cells at several time points (0 h, 1 h, 3 h, and 8 h), and then performed RNA-seq and differential gene expression analysis.

To identify genes that were hyper-expressed by the complete loss of RelB, we selected transcripts that were both induced by CpG (0.1 µM) (log₂FC > 1) in either WT or RelB^−/− BMDCs and hyper-expressed (FC > 1.5) in RelB^−/− BMDCs relative to WT BMDCs. We then plotted the expression of these genes in BMDCs derived from WT, RelB^−/−, RelB^DB/DB, IFNAR^−/−, IFNAR^−/−RelB^−/−, and IFNAR^−/−RelB^DB/DB mice (Fig. 5a). Using k-means clustering, we identified an IFN-dependent hyper-expressed cluster (B; 93 genes) containing genes hyper-expressed in RelB^−/− BMDCs (average fold differences of 1.25× to 1.5× compared to WT BMDCs at all observed time points), with absent induction in all IFNAR^−/− containing BMDCs (average fold differences of 0.5× to 0.6× in all IFNAR^−/− containing BMDCs compared to WT BMDCs at the late 8 h stimulation time point) (Fig. 5a; Supplementary Table S4). Notably, cluster B genes were also hyper-expressed in BMDCs derived from RelB^DB/DB mice, with fold differences ranging from 1.35× to 1.5× between RelB^DB/DB and WT BMDCs at all observed time points (Fig. 5a). As expected, motif analysis of cluster B produced the IFN sensitive response element (ISRE) as the top statistically enriched motif, and GO analysis revealed IFN signaling pathways “interferon beta” and “interferon alpha” among the top terms (Fig. 5b). Individual genes in this IFN-dependent hyper-expressed cluster B included established ISGs such as Mx1, Isg20, Oasl2, Ifi35, Ifit1, Isg15, Oaslb, all of which were all also hyper-expressed in BMDCs derived from RelB^DB/DB mice but not in BMDCs derived from compound IFNAR^−/−RelB^DB/DB mice (Fig. 5c). Notably, IFN-dependent genes previously described to be hyper-expressed in RelB^−/− BMDCs in response to polyI:C did not show a consistent hyper-expression phenotype in BMDCs derived from RelB^DB/DB mice²⁰ (Supplementary Fig. S8), indicating slight differences in the patterns of IFN-dependent gene expression in RelB^−/− and RelB^DB/DB DCs upon CpG vs polyI:C stimulation.

**Fig. 5: RelB suppresses type I IFN signaling and pro-inflammatory genes via its DNA binding function.**

Analysis of cluster A (178 genes) identified genes that were hyper-expressed in all RelB^−/− containing genotypes, but were not hyper-expressed in BMDCs derived from single IFNAR^−/− mutant mice, consistent with previously established IFN-independent pro-inflammatory genes²⁰. BMDCs derived from the RelB^DB/DB and IFNAR^−/−RelB^DB/DB compound mutant mice showed similar induction and hyper-expression of cluster A genes to BMDCs derived from RelB^−/− and IFNAR^−/−RelB^−/− mice (Fig. 5a). Specifically, cluster A genes showed average fold differences ranging from 1.3× to 1.5× between RelB^DB/DB BMDCs and WT BMDCs, 1.6× to 2.0× between IFNAR^−/−RelB^DB/DB BMDCs and WT BMDCs, and 1.2× to 1.3× between IFNAR^−/−RelB^DB/DB BMDCs and single IFNAR^−/− BMDCs at all observed time points (Fig. 5a; Supplementary Table S4). Motif analysis of cluster A genes detected NF-κB as the top statistically enriched motif, and GO analysis revealed NF-κB activating pathways “CD40”, “LPS” and “TNF-alpha signaling” among the top GO terms (Fig. 5b). Notably, genes within this IFN-independent cluster A included Map3k14, Ikbke, Ccl22, Ccl5, Cd40, Cd80, and Cd86, which are NF-κB-inducible pro-inflammatory genes and previously shown to be hyper-expressed in RelB^−/− and IFNAR^−/−RelB^−/− BMDCs²⁰. Here, we found that these genes were also hyper-expressed in RelB^DB/DB and IFNAR^−/−RelB^DB/DB BMDCs relative to WT and single IFNAR^−/− BMDCs (Fig. 5c, d). Stimulation with polyI:C showed similar effects (Supplementary Fig. S9). Together, these data provide strong evidence that RelB suppresses IFN-independent pro-inflammatory gene expression via direct binding to DNA and competition with RelA for κB sites.

Discussion

Here we aimed to elucidate the mechanism by which RelB suppresses the hyper-expression of pro-inflammatory genes and inflammatory histopathology seen in RelB^−/− mice^17,20. We found that the RelB-deficient pathology is IFN-independent, as assessed in the lung, liver, and thymus. As previous studies have pinpointed the loss of RelB in DCs alone to be sufficient in driving inflammatory pathology in the liver, spleen, lung, and thymi of RelB^−/− mice^21,22,23, we investigated dysregulated gene regulatory mechanisms in RelB^−/− DCs. Our RNA-seq and ChIP-seq analyses of BMDCs from RelB^−/− mice revealed elevated binding by RelA to a subset of genomic κB sites that were bound by RelB in resting WT BMDCs. Our analyses revealed that elevated binding by RelA to the promoter regions of known IFN-independent pro-inflammatory genes was correlated with their hyper-expression. These data support the hypothesis that RelB inhibits the expression of pro-inflammatory genes by competing with RelA for binding to their promoter regions.

To further test this hypothesis, we generated a novel RelB^DB/DB mouse containing mutations of three specific amino acids that abrogate κB site-specific DNA binding. Indeed, we found that the functional loss of RelB DNA binding resulted in the hyper-expression of IFN-independent pro-inflammatory genes in BMDCs similar to the hyper-inflammatory phenotype seen by the complete loss of RelB²⁰. Further, RelB^DB/DB mutant mice appeared runted and smaller than WT littermates at 8 weeks of age, similar to the RelB^−/− mice¹⁷. RelB^DB/DB mice displayed hunched posture, scaly skin, enlarged abdomens, thymic atrophy, and splenomegaly like RelB^−/− mice^17,20. Further, RelB^DB/DB mice revealed moderate to marked perivascular inflammation in the lungs, and moderate to marked periportal and centrilobular inflammatory infiltrates in the liver. Like RelB^−/− mice, these hallmarks of the inflammatory pathology were not rescued by the generation of IFNAR^−/−RelB^DB/DB compound mutant mice. Notably, while RelB^−/− and IFNAR^−/−RelB^−/− mice showed mixed lymphocytic and neutrophilic infiltrates in both the lung and liver as previously reported, RelB^DB/DB and IFNAR^−/−RelB^DB/DB mice instead displayed predominantly lymphocytic infiltrates in both the lung and liver, suggesting that while RelB^−/− and RelB^DB/DB mice share a common primary auto-inflammatory mechanism, there are likely compounding mechanisms caused by the complete loss of RelB leading to additional neutrophilic infiltrates and slight variations in spleen pathology.

The literature provides a rich set of potential molecular mechanisms by which RelB controls gene expression, ranging from altering IKK activity, to stabilization of the repressive p100/p105 complex, to DNA methylation and establishing repressive histone modifications, to simple competition with RelA for κB sites³⁷^{,38,41,45,46,47}^,59,60. Our RelA/RelB ChIP-seq and RNA-seq studies suggest that competition between the two NF-κB family members is the key mechanism, and the fact that the RelB DNA binding mutant near-phenocopies the RelB knockout provides further support for this simple model of balanced NF-κB target gene expression control.

Our findings do not determine the exact function of RelB upon binding to DNA and do not rule out the involvement of cytoplasmic functions. Epigenomic profiling of the repressive H3k9me2 modification did not reveal RelB-dependence even though SMAC mimetics applied to DCs, which induce RelB among other signaling pathways, led to statistically significant changes (data not shown). Further biochemical characterization of DCs derived from the RelB DNA binding mutant may also reveal additional gene regulatory mechanisms. It is worth noting that the Nfkb2 p100 protein, which has the potential to limit how much RelA-containing dimer may be activated, has also been reported as an inhibitor of DC activation, with Nfkb2^−/− DCs showing elevated MHC class II and costimulatory molecule expression and an enhanced ability to induce CD4⁺ T cell responses⁶¹. However, the autoimmune pathology is sporadic in only some mice and is mild, with the great majority having unaffected lifespans^62,63,64,65. These data support our conclusion that the stabilization of p100 function by RelB is not the key mechanism leading to lethal multi-organ pathology in RelB^−/− mice.

Further, while our mechanistic studies were performed in DCs, our phenotypic characterization of the autoimmune pathology was done utilizing mice with a global RelB DNA binding mutant. While this mutant may also have defective functions in other immune cells such as medullary thymic epithelial cells (mTECs), which have also been implicated in the RelB knockout pathology, previous work has demonstrated that correcting RelB deficiency in DCs is sufficient to restore appropriate mTEC and T-regulatory (Treg) cell numbers in the thymus²³ and improve autoinflammatory pathology²¹, suggesting that DC pathology is causal of changes in the thymic niche. However, whether this niche-modulatory function of RelB in DCs is dependent on high-affinity DNA binding remains unknown. Therefore, further studies of how DCs derived from the RelB DNA binding mutant influence other immunomodulatory cell types in autoimmune pathology may be warranted.

The concept of NF-κB dimers competing for κB sites also applies to the p50 subunit, which on the one hand, is a dimerization partner for RelA, but on the other hand, may homodimerize and have antagonistic effects on RelA-mediated gene activation. Given that p50 homodimers lack a transactivation domain⁶⁶, but are able to bind DNA and compete with RelA:p50 for binding to κB sites at regulatory regions, p50 homodimers are thought to act as repressors of NF-κB gene expression. Indeed, overexpression of p50 was shown to suppress RelA-induced expression of several immune response genes such as H2-K1, ICAM1, and IL2^67,68,69. The p50 homodimer was also reported to interfere with the binding of IFN regulatory factors on ISREs that contain a triple G motif (e.g., GGGRA), thus resembling a half-site of the palindromic κB element⁷⁰.

How could RelB function both as a transcriptional activator for organogenic chemokine and metabolic genes³³^,54^,71^,72 and as a transcriptional repressor for pro-inflammatory genes³²? One possibility is that RelB’s role is determined by its dimerization partner, whether it is p50 or p52. As a RelB:p50 dimer, RelB may have similar DNA binding specificity to RelA:p50⁵⁵ and thus compete with the RelA:p50, which is a strong pro-inflammatory transcriptional activator. However, p52 has been shown to confer a different DNA sequence specificity to NF-κB dimers^15,73. Thus, the RelB:p52 heterodimer, a weaker transcriptional activator, may be binding to promoter elements that are not accessed by RelA:p50, such as on organogenic chemokine genes⁷⁴. Differences in mRNA half-life in the two sets of genes may amplify the specificity due to RelA:p50’s transient and RelB:p52’s long-lasting activity dynamics. Further studies are needed to substantiate this conceptual framework.

Our present findings highlight that NF-κB RelA transcriptional activity, once released from IκB proteins, is further finetuned by RelB in accessing the κB sites of target genes. As such, we provide mechanistic insights into the autoinflammatory mechanisms that drive RelB-deficient pathology. While substantial inflammatory consequences of RelB deficiency are seen phenotypically, our findings suggest that they are caused by small changes in RelA binding to target gene promoters and target gene expression in DCs. As such, RelB-deficient pathology and other relopathies may likely require therapeutic approaches focused on finetuning these slight changes in dysregulated gene regulatory mechanisms. Given the extensive list of pathologies linking NF-κB to autoimmunity and dysregulation of IFN-dependent and IFN-independent pro-inflammatory gene expression^6,7,8, the present analysis may inform ongoing efforts to elucidate the molecular mechanisms that mediate them.

Materials and Methods

Design of RelB DNA binding mutants

The pBABE-puro vector containing RelB variants was constructed by first amplifying the coding region of murine RelB cDNA corresponding to amino acids 1–558 by PCR, and inserting it into the EcoRI and BamHI sites of the polylinker. Mutagenesis using Agilent QuikChange II Site-Directed Mutagenesis Kit (Agilent Technologies) was performed for RelB point mutations. DNA binding mutants within the amino-terminal of RelB were generated by making single or triple substitutions at residues Arg117, Tyr120, and Glu123 to Ala (R117A, Y120A, and E123A). RelB-Y120A was obtained by PCR using the oligonucleotides (5′-tggcatgcgcttccgcGCcgagtgcgagggccgc-3′ and 5′-gcggccctcgcactcgGCgcggaagcgcatgcca-3′). RelB-R117A/Y120A/E123A mutant was obtained using the oligonucleotides (5′-cagcgtggcatgGCcttccgcGCcgagtgcGCgggccgctcggcc-3′ and 5′-ggccgagcggcccGCgcactcgGCgcggaagGCcatgccacgctg-3′).

Generation of cell lines harboring RelB DNA binding mutants

Immortalized Relb^−/−nfkb2^−/− fibroblast cells were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM) supplemented with 10% bovine calf serum (BCS), 1% penicillin-streptomycin and 1% l-glutamine. Platinum-E (Plat-E) cells were cultured in 10% fetal calf serum, 1% penicillin-streptomycin, 1% l-glutamine, and supplemented with 1 μg/mL puromycin and 10 μg/mL of blasticidin. For transfection, Plat-E retroviral packaging cell line⁷⁵, was plated on 10-cm plates 16 h prior at 50% confluency in DMEM supplemented with 10% fetal calf serum, 1% penicillin-streptomycin, and 1% l-glutamine. Cells were transfected using 300 μL of Opti-MEM medium (Thermo Fisher Scientific) and polyethylenimine (PEI; 1 μg/μL in 1× PBS, pH 4.5, Polysciences #23966-2) with 7 μg of retroviral construct DNA, pBABE-puro empty vector (EV control) or RelB-expressing constructs (4:1 ratio of PEI (μL):plasmid DNA (μg)). The transfection complex (Opti-MEM medium, PEI reagent, plasmid DNA) was incubated for 15 min at room temperature, then added drop-wise to Plat-E cells, and incubated for 6 h. Transfection media was then replaced with fresh DMEM containing 10% fetal calf serum, 1% penicillin-streptomycin, and 1% l-glutamine. Cells were further incubated for a total of 48 h prior to collecting viral supernatants. Virus-containing supernatant was filtered through a 0.45-μm filter and used to infect RelB^−/−nfkb2^−/− fibroblast cells with the addition of 4 μg/mL Polybrene. 48 h post-infection of cells with viruses, the stably transduced cells were then selected with 2.5 μg/mL puromycin for a total of 72 h. After selection with puromycin, the puromycin-containing medium was removed and cells were passaged twice for recovery prior to being expanded in culture for experiments. EV, pBABE-puro vector without the RelB gene fragment, is used as a negative control, to maintain a stable retrovirally transduced RelB knockout cell line, meanwhile, pBABE-puro containing full-length RelB (RelB-WT) is used as a positive control for RelB-expressing cell line.

Biochemical characterization of RelB DNA binding mutants

Western blotting and EMSA were performed with standard methods as described previously^54,76,77. Western blots were probed with antibodies specific for RelB (Santa Cruz Biotechnology, Cat# sc-226), and Actin (Santa Cruz Biotechnology, Cat# sc-1615). For EMSA, fibroblast cells were stimulated with 1 ng/mL TNF, nuclear extracts were harvested at the indicated times (h), and previously published protocols were followed^78,79.

Generation of the RelB DNA binding mutant mouse strain

The RelB^DB/DB mutant mouse line was generated by Ingenious Targeting Laboratory (iTL, www.genetargeting.com) using the targeted Neomycin selection cassette for the conventional knockin model. A donor sequence encoding the mutant RelB protein was used to generate, via homologous recombination, a tagged embryonic stem cell line carrying a Neo resistance marker, that was implanted to yield heterozygous mice. These mice were then bred with a mouse line constitutively expressing the Flp recombinase to remove the Neo resistance marker included in the homologous donor sequence. We then back-crossed the resultant mice with WT C57BL/6J mice to remove the Flp background and generate the completed, germline heterozygous mutant knockin line. Further, a subsequent round of heterozygous mating was performed to obtain the homozygous knockin of RelB^DB/DB mice.

Mouse husbandry

WT, knockout, and knockin mice were housed in pathogen-free conditions at the University of California, Los Angeles. RelB^−/− mice were generated by breeding RelB^+/– mice, IFNAR^−/−RelB^−/− mice were generated by breeding IFNAR^−/−RelB^+/– or IFNAR^+/–RelB^+/– mice, and RelB^DB/DB mice were generated by breeding RelB^+/DB mice. All mice used for experiments were between 4 and 12 weeks on the day of the experiment, both male and female mice were used for experiments.

BMDCs

Bone marrow cells were isolated from mouse femurs and cultured with 20 ng/mL GM-CSF and 10 ng/mL IL-4 to produce BMDCs with half the media being replaced on days 3 and 6 as previously reported³⁸. Cells were stimulated with CpG (0.1 µM) (Invivogen ODN 1668, Cat# tlrl-1668) or Poly(I:C) HMW (10 μg/mL) (Invivogen; Cat# tlrl-pic) and collected at specified time points in Invitrogen™ TRIzol™ Reagent (Cat# 15-596-018) for RNA and was extracted using Qiagen RNeasy Mini Kit (Cat# 74106) as described⁵⁴.

Transcriptome profiling

RNA was used for RNA-seq as described⁸⁰. Briefly, libraries were prepped using the KAPA Stranded mRNA-Seq Kit Illumina® platform (KR0960-v3.15) using 1 µg of RNA per sample measured using a Qubit 2.0 fluorometer. Final libraries were checked via agarose gel and multiplexed with a maximum of 24 samples per sequencing run. Libraries were sequenced using Illumina HiSeq 3000 with single-end 50-bp reads at the UCLA Technology Center for Genomics & Bioinformatics.

Bioinformatic analysis

Reads were trimmed using cutadapt (1) (cutoff q = 20) and mapped to the mm10 genome. Processed reads showed high-quality reads and alignment scores. The October 2014 version of the Ensembl database was used to extract gene annotation information. CPM values were generated using edgeR (5) to normalize the raw count data based on sequencing depth. To permit FC calculations, a pseudocount of 1 CPM was added. Induced genes were selected using a log₂FC > 1 cutoff for any stimulated time point relative to 0 h unstimulated control, and transcripts with empty gene names were removed. Data were z-scored and plotted using the pheatmap R package. Fold differences of genes within heatmaps were calculated by first calculating the fold differences for all individual genes between genotypes of interest and WT or IFNAR^−/− control CPM (Genotype X)/CPM (Genotype Y) for each individual time point. Average fold differences were then calculated by averaging the fold differences of all genes within each cluster for each individual time point. GO and motif analysis was done via Homer suite considering regulatory regions within −1 kb to 1 kb from the TSS. Line graphs of individual genes were generated using GraphPad Prism.

ChIP sample generation and data preprocessing

ChIP-seq was performed as previously described^81,82 with anti-RelA antibody (Cell Signaling Technology, D14E12 or Cat# 8242) or RelB antibody (Cell Signaling Technology, D7D7W or Cat# 10544). Approximately 1 × 10⁷ BMDCs were used per sample. Two chemical crosslinkers, DSG and PFA, were used during sample preparation at concentrations of 1.0 mM and 1.0%, respectively. Sonication was performed on a Covaris M220-focused ultrasonicator. ChIP-seq libraries were prepared using KAPA HyperPrep Kits (Roche) and sequenced using Illumina HiSeq 3000 with single-end 50-bp reads at the UCLA Technology Center for Genomics & Bioinformatics. Reads were aligned using Hisat2 to the mouse genome (NCBI37/mm9). Peaks were called if they were enriched compared to input samples and had a false discovery rate of < 0.01 using HOMER software⁸³. To compare peaks across samples, a master probe which contained all peaks from every sample was generated using BEDTools⁸⁴. Then SeqMonk was used to find raw read counts for each peak in the master probe. RPKMs were generated using the length of each peak and the depth of sequencing for each sample.

Tissue isolation, fixation, and scoring

Spleens, lungs, and livers were isolated from age-matched mice immediately after being euthanized and subsequently rinsed with PBS. After the removal of excess PBS, spleens were weighed. Fixation was done in 10% formaldehyde for 46–48 h. Tissue was processed, sectioned, and H&E-stained by the UCLA Translational Pathology Core Laboratory (TPCL). Histology slides were blinded and scored. Red pulp expansion was quantified and scored as adequate = 0, expanded = 1, and markedly expanded = 2. White pulp contraction was quantified and scored as adequate = 0, reduced = 1, and markedly reduced = 2. Extramedullary hematopoiesis was scored as present = 1, increased = 2, marked = 3. Lung and liver inflammation was quantified and scored as absent = 0, scant = 1, mild = 2, moderate = 3, and marked = 4. Lymphocyte and neutrophilic infiltrates were scored as absent = 0, scant = 1, mild = 2, moderate = 3, marked = 4.