RelA ubiquitination is directed to the amino-terminal RHD
To identify the ubiquitin acceptor residues responsible for RelA ubiquitination, we first investigated the ubiquitination of broad truncation mutants (the amino-terminal RHD or the carboxyl-terminal transactivation domains or TAD). The RHD was strongly ubiquitinated (, upper panels using an N-terminal specific antibody), while the TAD was not ubiquitinated at all when expressed in isolation of the RHD (, lower panels). These findings are in keeping with the fact that COMMD1 and SOCS1 both bind to the RHD (Maine et al 2007
, Ryo et al 2003
) and suggest that the TAD fails to be ubiquitinated either as a result of a lost interaction with the ubiquitin ligase or because it lacks the lysine(s) that serve as acceptor site(s).
RelA ubiquitination is promiscuous and involves non-degradative polyubiquitin chains
RelA can be targeted by non-degradative ubiquitination
While RelA ubiquitination has been shown to be stimulus-dependent and to affect RelA stability, only a small amount of the cellular pool of this protein is degraded in response to pro-inflammatory stimuli. Unlike IκB-α, whose stimulus-induced ubiquitination results in substantial degradation, RelA ubiquitination does not affect overall cellular levels of this protein and is only noticeable when specific cellular compartments are examined or when new protein synthesis is blocked (Geng et al 2009
, Maine et al 2007
, Yang et al 2009
). All this suggested that not all RelA ubiquitination may be for degradative purposes, and thus we set out to examine whether alternative forms of ubiquitination, such as lysine-63 (K63) linked chains, also take place. RelA was expressed together with ubiquitin and the presence of ubiquitinated RelA was detected after enrichment of ubiquitinated proteins on Ni-NTA columns (). Using a mutant form of ubiquitin that only has the K63 site available for polyubiquitin chain formation, we found evidence that RelA can be ubiquitinated in a K63-dependent manner (, left panels). Conversely, other lysine linkages were also evident, since a ubiquitin mutant lacking K63 was also able to recover polyubiquitinated RelA (, right panels).
To confirm these findings and exclude the recovery of mixed polyubiquitin chains containing endogenous and mutant ubiquitin, we utilized a ubiquitin replacement strategy. This was accomplished through recently developed cell lines where endogenous ubiquitin expression is repressed by Tet-inducible shRNA against all 4 human genes encoding ubiquitin, while mutant ubiquitin is concurrently expressed (Xu et al 2009
). After Tet-induction, we immunoprecipitated RelA () and found that ubiquitinated RelA could be recovered from cells expressing the K63R and K48R ubiquitin mutants, confirming the conjugation of RelA with non-K48 and non-K63 linked chains. Next, we utilized an antibody that is specific for K63-linked polyubiquitin chains and found that cell stimulation with TNF resulted in the accumulation of polyubiquitinated RelA, and this material was immunoreactive with the K63-specific antibody ().
Redundant lysines can be targeted for ubiquitination
Having identified the RHD as a potential site for ubiquitin conjugation we undertook a mutagenesis approach to try to identify the ubiquitin accepting residue(s). Progressive mutagenesis of all lysines in the RHD (N) failed to abrogate RelA ubiquitination (). The apparent discrepancy between this result and the findings in , where the RHD deletion mutant (TAD) was not ubiquitinated, suggested that the prior result was more likely related to the structural alterations inherent to the deletion rather than to the actual loss of absolutely required lysines. Moreover, we found that K/R mutations in the TAD (C) also had persistent ubiquitination. Only a mutant completely devoid of all lysines (All) was ultimately able to abrogate all RelA ubiquitination, indicating the highly redundant nature of lysines as acceptor sites for ubiquitination of this protein.
Mass spectrometry identifies ubiquitin acceptor sites in the amino-terminus of RelA
The ubiquitin acceptor sites proved to be highly redundant precluding the identification of physiological target sites using a mutagenesis approach. Therefore, we decided to utilize mass spectrometry to identify physiologic acceptor sites and polyubiquitin branching in a non-biased manner. To this end, ubiquitinated RelA was purified through a sequential affinity procedure termed bimolecular affinity purification (BAP) and the recovered material was analyzed by mass spectrometry. The procedure was performed 3 times, and in aggregate allowed the identification of 7 ubiquitin conjugated lysines from among the 18 lysines present in RelA (). Three of the ubiquitin conjugated residues corresponded to lysines previously described as acetylation sites. In addition, these data confirmed the presence of K48-linked polyubiquitin chains, as well as other linkages including K29, K33 and K63 (data not shown).
RelA polyubiquitination occurs at sites of acetylation
RelA polyubiquitination overlaps with acetylation target lysines
In order to understand the physiologic relevance of RelA ubiquitination at the sites identified, we introduced point mutations in the indicated lysines. We focused further analysis on 4 mutants of RelA which are schematically displayed in . In the first mutant all 4 lysines not known to be acetylation targets were changed to arginines (K4R-NonAc). The other 3 lysines (K123, K310 and K315) have been reported to be acetylation sites with two of these sites (K123 and K315) being immediately adjacent to a preceding lysine (marked with an asterisk) that can also be acetylated (Buerki et al 2008
, Chen et al 2002
, Kiernan et al 2003
, Rothgiesser et al 2010
). Based on this, K122 and K314 were also included in the analysis. Given that K310 has received particular attention as an acetylation site, we generated two mutants affecting the lysines targeted for acetylation excluding or including K310 (K4R and K5R, respectively). Finally, a mutant containing most of the lysines identified was also generated (K8R).
Next, we examined whether these mutations affected RelA acetylation. RelA was co-expressed with p300, the most potent acetyltransferase that targets this protein. After RelA precipitation, the recovered material was immunoblotted with an acetyl-lysine reactive antibody. Wild-type (WT) RelA or the K4R-NonAc mutant targeting sites not known to be acetylation sites were clearly acetylated by p300 (). On the other hand, the RelA K310R mutant was poorly acetylated, while the K4R mutation also showed strongly impaired RelA acetylation. The combined K5R mutation had the most dramatic reduction with nearly no detectable acetylation (), indicating that these 5 lysines are indeed the main acetylation sites targeted by p300. Moreover, since the K310R mutation was sufficient to abrogate most acetylation, we concluded that this site is responsible for most of the signal or that the process of acetylation requires cooperativity between the K310 site and the other lysines.
Acetylation acceptor lysines are primarily responsible for RelA polyubiquitination
The ubiquitination status of these RelA mutants was examined next. RelA was co-expressed with His-tagged ubiquitin, followed by enrichment of ubiquitinated proteins on Ni-NTA beads. Western blot analysis revealed intact ubiquitination of the K4R-NonAc mutant, while the K4R mutant was hypo-ubiquitinated compared to the WT protein (). Additional experiments confirmed that the mutations of K122 and K314 included in the K4R mutant were required for this effect (data not shown), suggesting that they indeed function as redundant sites for both ubiquitination and acetylation.
RelA polyubiquitination is primarily directed to acetylation sites
Interestingly, the K8R mutant, which includes the acetylation sites and the 4 non-acetylation sites identified by mass spectrometry was ubiquitinated to a similar extent as the WT protein. Moreover, additional lysine mutations (K8R+K310R, and others) often resulted in a paradoxical increase in RelA ubiquitination (data not shown), suggesting that extensive perturbations of RelA after multiple lysine mutations may result in polyubiquitination due to disturbed secondary and tertiary structures.
The K310R and K5R mutants were similarly examined. Expression of RelA along with His-tagged ubiquitin resulted in the upward shift of the high-molecular ladder above RelA, indicating that this laddered material corresponded to ubiquitinated RelA (, first and second lanes). While K310R showed no differences, the K4R and K5R mutants displayed reduced ubiquitination. Upon precipitation of ubiquitinated proteins, the pulldown was immunoblotted for RelA confirming confirmed that K4R and K5R are hypo-ubiquitinated (). The K5R mutation not only abrogated the basal ubiquitination of RelA, but also led to dramatic reductions in polyubiquitination after TNF stimulation, which were evident at the level of total RelA ubiquitination and K63-linked polyubiquitination ().
Polyubiquitinated RelA is not concurrently acetylated
These findings indicated that lysines that are acetylation target sites are also responsible for most polyubiquitin conjugation to RelA, implicating that polyubiquitinated RelA cannot be concurrently acetylated. To test this notion, ubiquitinated RelA was purified using the BAP procedure () and the acetylation of ubiquitinated RelA was examined. The acetyl-lysine specific antibody was reactive with the non-ubiquitinated form of RelA, and never produced any signal against ubiquitinated RelA, supporting the hypothesis that acetylated RelA is refractory to ubiquitination.
RelA acetylation and ubiquitination compete with each other
Ubiquitination inhibits RelA acetylation
Next we extended these findings in vivo, where we set out to promote RelA ubiquitination to examine its effects on the levels of acetylated RelA. p300-dependent acetylation of GST-tagged RelA was monitored by precipitation and immunoblotting with an acetyl-lysine reactive antibody. First, we used increasing expression of His-tagged ubiquitin to drive RelA ubiquitination (). This led to greater amounts of ubiquitinated RelA (, bottom panel) and a parallel decrease in the levels of acetylated RelA (, top panel). Interestingly, this effect was observed even at the level of the non-ubiquitinated form of RelA and corresponded to a progressive decrease in RelA-p300 co-precipitation (, third panel), suggesting that occupation of lysines is not the only mechanism by which these two modifications compete with each other.
To substantiate these results in a complementary setting, RelA ubiquitination was promoted through 2 alternative approaches (Maine et al 2007
, Mao et al 2009
): the concurrent expression of Cul2 and COMMD1 (components of the ligase complex that targets RelA), and expression of GCN5, a cofactor that binds to this ligase and promotes RelA ubiquitination (to avoid any potential cross-talk with acetylation, the E575Q acetyltransferase deficient mutant of GCN5 was utilized). Both Cul2/COMMD1 and GCN5 promoted strong RelA ubiquitination (, bottom panel and were associated with a dose-dependent reduction in RelA acetylation (, top panel), and reduced co-precipitation of p300 (not shown).
Increased acetylation inhibits RelA ubiquitination
These experiments indicated that ubiquitination of RelA inhibits its acetylation, and we set to examine whether the converse could be true (). RelA acetylation was promoted by expression of CBP, a homolog of p300. Increased RelA acetylation was paralleled by decreased RelA ubiquitination noted by precipitating RelA and immunoblotting for ubiquitin (, top panel; also evident as high molecular weight material in RelA western blots). Altogether, the data indicated that these two PTMs can target the same sites and antagonize each other in vivo.
Generation of rela−/− reconstituted fibroblasts
In order to examine the functional contribution of RelA modification at these five lysines, mutations of these sites were introduced in the corresponding murine homolog of RelA (designated here as mRelA). It is important to note that all the lysines identified by mass spectrometry are tightly conserved among mammals and for simplicity we will continue to refer to these sites according to their position in human RelA.
We generated K to R mutations, which abrogate all modifications of these residues while preserving the overall charge at these positions, as well as K to Q mutations, which mimic lysine acetylation. These mRelA mutants were stably expressed in rela−/−
mouse embryo fibroblasts utilizing lentiviral infection. Comparable expression of the desired mutants as well as relevant NF-κB regulatory proteins was confirmed by immunoblotting (). Expression of classical IκB proteins and other NF-κB subunits became greater upon reintroduction of mRelA, and this effect was similar across all cell lines developed. Since these proteins are themselves expressed from NF-κB responsive genes (Hayden and Ghosh 2008
), these data suggested that the transcriptional activity of the mutants was preserved. Indeed, mRNA expression of Nfkbia
, the gene encoding IκB-α, was largely similar across all cell lines ()
Modifications of RelA acetylation sites have diverse function in endogenous gene expression
RelA K/R and K/Q mutants are able to dimerize and bind DNA
The RHD is required for dimerization and DNA binding, and this latter function involves indirectly K122 and K123 (Chen et al 1998
). Thus, we examined first whether the RelA mutants could dimerize, bind to IκB-α and to their cognate DNA. Immunoprecipitated mRelA from the stable cell lines demonstrated co-precipitation of endogenous p105, p50 and IκB-α () indicating that they can form heterodimers and bind to IκB-α. The K4Q and K5Q mutants displayed slight predilection for p50 compared to IκB-α binding, consistent with prior reports that acetylation has an inhibitory effect on RelA-IκB-α interactions (Chen et al 2001
). Similarly, nuclear levels of RelA K5Q were slightly higher in unstimulated cells, while post-stimulation levels were comparable to WT and K5R stable cells (Fig. S1
). Next, we assessed the ability of these mutants to bind to DNA using a DNA/protein co-precipitation assay using nuclear extract and biotinylated κB-containing oligonucleotides (). All mutants were coprecipitated to a similar extent by DNA in unstimulated cells () and in extracts from TNF stimulated cells (data not shown). Altogether, the K/R or K/Q mutations did not impair dimerization or DNA binding, a pre-requisite to examine the contribution of these residues to gene expression.
Lysine modifications play gene-specific roles in transcription
Next, we examined the effects of modifications at these residues on global TNF-inducible gene expression in K5R and K5Q reconstituted fibroblasts. After filtering the data for genes regulated by at least 2-fold in response to TNF in WT cells, we identified 185 genes that were differentially regulated by a factor of 2-fold in either the K5R or K5Q cells compared to the WT cells (). Contrary to our initial expectation, the acetyl-mimicking K5Q mutation did not lead to a global activation of gene expression. Rather, many TNF-inducible and RelA-dependent genes in these cells were specifically underexpressed in K5Q cells. Nevertheless, an activating effect of the K5Q mutation was also seen (lower part of the gene cluster in ).
Key findings from this analysis were confirmed by qRT-PCR, such as the dramatic repression of Vcam1
expression in K5Q reconstituted fibroblasts ( and S2
); the converse increase of Mmp13
in K5Q expressing cells was also observed ( and S2
). Moreover, similar alterations were noted for genes whose expression levels were too low to pass filter criteria of the microarray analysis such as Icam1
, and Cxcl10
, which demonstrated repressed expression in K5Q cells ( and S2
) and Mmp3
, which demonstrated exaggerated late induction in K5Q cells (Fig. S2
The function of RelA modifications on transcription and chromatin association is gene-specific
Inducible RelA ubiquitination can be detected on chromatin
Vcam1 and Icam1 expression was evaluated in further detail by qRT-PCR, which confirmed the repressive effect of RelA K5Q observed in the microarray analysis (). In addition, the K4Q mutation was similarly repressive, while isolated K310 mutations had minimal effects on these and other genes examined (Vcam1, Icam1, Mmp13, Mmp3, Csf2, Saa3, Il6, Ptgs2 and Cxcl10 - and data not shown).
Cells reconstituted with K5R, a mutant unable to be modified on any of these residues, displayed increased expression of Vcam1
. We initially speculated that this was due to loss of degradative ubiquitination, given the inhibitory effects of ubiquitination on Icam1
expression (Burstein et al 2005
, Maine et al 2007
, Mao et al 2009
). However, the unexpected inhibitory effects of the K5Q mutant, suggested that loss of acetylation could also account for the increased expression in the K5R expressing fibroblasts. To clarify this question, the presence of ubiquitinated RelA at this promoter was examined by chromatin IP (ChIP) and re-ChIP using ubiquitin and RelA sequential immunoprecipitations. This confirmed the physiologic induction of RelA ubiquitination on the human ICAM1
Histone acetylation correlates with the transcriptional effects of K5 mutations
We next evaluated the potential mechanism(s) for the transcriptional effects of K5 mutations. ChIP analysis of the Vcam1
or the Icam1
promoters indicated that RelA recruitment was not impaired even for the repressive RelA K5Q mutant (). In fact, the recruitment of RelA K5R and K5Q to the Icam1
promoter persisted at 5 h compared to RelA WT, consistent with their decreased ubiquitination. Similarly, recruitment of CBP to this region could not account for differences in expression (Fig. S3
). However, histone acetylation of the Vcam1
promoters had a general correlation with gene expression, being decreased in RelA K5Q reconstituted fibroblasts and somewhat elevated -particularly after TNF- in RelA K5R cells (). The converse findings were made for Mmp13
, a gene that was selectively induced in RelA K5Q cells (). Again, RelA or CBP recruitment did not correlate with differences in Mmp13
expression ( and S3
). However, histone acetylation in this promoter region was dramatically increased in RelA K5Q expressing fibroblasts, consistent with their increased expression of Mmp13