O-GlcNAcylation of High Mobility Group Box 1 (HMGB1) Alters Its DNA Binding and DNA Damage Processing Activities

Abstract

Protein O-GlcNAcylation is an essential and dynamic regulator of myriad cellular processes, including DNA replication and repair. Proteomic studies have identified the multifunctional nuclear protein HMGB1 as O-GlcNAcylated, providing a potential link between this modification and DNA damage responses. Here, we verify the protein’s endogenous modification at S100 and S107 and found that the major modification site is S100, a residue that can potentially influence HMGB1-DNA interactions. Using synthetic protein chemistry, we generated site-specifically O-GlcNAc-modified HMGB1 at S100 and characterized biochemically the effect of the sugar modification on its DNA binding activity. We found that O-GlcNAc alters HMGB1 binding to linear, nucleosomal, supercoiled, cruciform, and interstrand cross-linked damaged DNA, generally resulting in enhanced oligomerization on these DNA structures. Using cell-free extracts, we also found that O-GlcNAc reduces the ability of HMGB1 to facilitate DNA repair, resulting in error-prone processing of damaged DNA. Our results expand our understanding of the molecular consequences of O-GlcNAc and how it affects protein–DNA interfaces. Importantly, our work may also support a link between upregulated O-GlcNAc levels and increased rates of mutations in certain cancer states.

Graphical Abstract

INTRODUCTION

O-GlcNAcylation is a posttranslational modification in higher-order eukaryotic organisms involving the addition of a single N-acetylglucosamine (GlcNAc) sugar moiety onto serine and threonine residues of protein substrates. Unlike other types of glycosylation whose targets are routed toward the cell surface, O-GlcNAc substrates remain intracellular where the glycosyl-transferase that adds the modification (O-GlcNAc transferase or OGT) and the glycosidase that removes it (O-GlcNAcase or OGA) also reside. This enzymatic cycling means that akin to protein phosphorylation, O-GlcNAcylation can be transient, reversible, and sensitive to the cellular conditions.¹ O-GlcNAc levels are nutrient- and stress-responsive, and their misregulation at a global level is implicated in cellular abnormalities and disease states. At the protein level, O-GlcNAc acts beyond simply adding bulk to the modified side chains, as direct interrogations on specific protein targets demonstrate that GlcNAc can act as molecular “grease or glue” that can modulate protein interfaces.² Thus, O-GlcNAc can influence protein folding, inhibit protein aggregation, participate in posttranslational modification (PTM) crosstalk, and regulate protein–protein interactions.

O-GlcNAcylation is highly enriched in the nucleus, where various chromatin modifiers, transcription factors, scaffold proteins, and nucleocytoplasmic enzymes are known to be modified. Posttranslational modifications on these targets affect diverse cellular processes such as DNA replication, cell cycle control, transcriptional regulation, epigenetics, and DNA repair processes.^1,3 Recently, the chromosomal protein high mobility group box 1 (HMGB1, previously HMG1 in older publications) was identified as a substrate for O-GlcNAcylation, specifically at serines 100 and 107⁴ (Figure 1a). HMGB1 is a non-histone chromosomal protein that participates in various cellular functions. Its protein sequence is composed of two DNA-binding HMG-box domains (boxes A and B) joined by a short linker region and is terminated by a flexible tail of an uninterrupted stretch of aspartic and glutamic acid residues.^5,6 Each HMG box is characterized by a trihelical L-shaped fold that contours to DNA substrates.⁷ Both HMG boxes also have high proportions of basic residues that interact with the DNA backbone, as well as specific helix residues that can intercalate between DNA bases.⁸ These structural features impart HMGB1 with a preference for distorted, non-B-form DNA structures. Specifically, HMGB1 has shown higher affinity in vitro for supercoiled DNA,⁹ DNA minicircles,¹⁰ looped DNA,^11,12 cisplatin-DNA adducts,¹³ and four-way junction DNA¹⁴ over linear forms. HMGB1 is also able to induce torsional and topological alterations upon binding linear DNA by causing bending, kinking, looping, and supercoiling processes.¹⁵ This preference for noncanonical DNA structures allows HMGB1 to act as a “DNA chaperone” by working with histone proteins to facilitate chromatin organization¹⁶ or with other nuclear proteins for transcriptional co-regulation^17,18 and DNA damage recognition and repair.¹⁵ For instance, HMGB1 has been shown to function as a cofactor in nucleotide excision repair (NER), facilitating error-free repair on both psoralen interstrand and UV-induced intrastrand cross-links.¹⁹

Figure 4.

S100A mutation exhibits biochemical properties distinct from unmodified or HMGB1(gS100) variants. (a) HMGB1(S100A) mutant shows comparable ability to preserve negative supercoils as unmodified HMGB1 control. (b) HMGB1(S100A) exhibits enhanced oligomerization on four-way junction DNA over unmodified HMGB1. The effect of the mutation is similar to the effect of S100 O-GlcNAcylation as in Figure 3b. (c) The amount of HMGB1(S100A) required to form the maximum number of minicircles is indicated by *. The saturation binding curves and half-maximal ratio for minicircle formation is similar to HMGB1(gS100). For quantification of this and a replicate experiment, please see Figure S5. All results shown in this figure are representative of at least 2 experiments.

Figure 1.

HMGB1 is O-GlcNAc-modified at S100 and S107 in cells. (a) HMGB1 has two high-mobility group domains (boxes A and B) and a highly acidic tail. Previous proteomic studies localize O-GlcNAc modification of HMGB1 on the B box domain. (b) Chemoenzymatic labeling and biotin pulldown confirm that HMGB1 is O-GlcNAc modified in H1299 cells. O-GlcNAcylated proteins were enriched using chemoenzymatic labeling and bioorthogonal reaction with a cleavable biotin-tag. Nup62 and β-actin pulldowns are shown as positive and negative controls, respectively. (c) The major sites of HMGB1 O-GlcNAc modification are S100 and S107. H1299 cells were transfected with HA-tagged wild-type HMGB1 and FLAG-tagged S-to-A mutant HMGB1, and lysates were subjected to chemoenzymatic labeling and pulldown. HA and FLAG immunoblot signals were imaged simultaneously to allow quantitative comparison using densitometry. Bar graphs represent means of two separate biological experiments; error bars represent ± SEM. Statistical significance was calculated using two-tailed Student’s t test. (d) Top, structure of HMGB1 B box domain (yellow) bound to bent DNA (white) in a V(D)J recombination complex (PDB: 6CIJ). Bottom, S100 points into the DNA interface and is only a few residues away from a known, conserved intercalating residue (F103).

Here, we examined the consequences of HMGB1 O-GlcNAcylation to its DNA binding properties and its role in DNA damage processing. We first confirmed that the protein is O-GlcNAc modified in human non-small-cell lung carcinoma, H1299, cells and identified the major modification site to be serine 100. We then utilized selenocysteine-based expressed protein ligation (EPL) to prepare homogeneously O-GlcNAcylated HMGB1 at S100 (gS100). We demonstrated through in vitro biochemistry that the modification alters HMGB1’s ability to interact with known DNA substrates such as four-way junction DNA, negatively supercoiled DNA, and nucleosomal DNA. Several of these substrates are model scaffolds for damaged DNA or intermediates of DNA repair pathways, suggesting that O-GlcNAc may play a role in DNA damage processing. We tested this possibility in a cell-free assay and demonstrated that O-GlcNAc resulted in a loss of HMGB1’s known ability to facilitate error-free repair of interstrand cross-linked DNA.

RESULTS AND DISCUSSION

HMGB1 Is Endogenously O-GlcNAc-Modified, and the Major Site of Modification Is S100.

We found the O-GlcNAc status of HMGB1 from multiple proteomics studies that employed a variety of detection methods.^20-22 In a glycosite mapping study of primary human T cell proteins, a tryptic glycopeptide was detected corresponding to residues 97–112, in which two serine residues at positions 100 and 107 (S100 and S107) were found to be O-GlcNAc-modified.⁴ This study utilized the metabolic chemical reporter Ac₄GalNAz to enrich and identify O-GlcNAc sites. However, because reporters of this type can result in increased metabolic flux,²³ we first wanted to confirm that HMGB1 is endogenously O-GlcNAcylated. We employed chemoenzymatic detection of endogenously O-GlcNAc-modified proteins from H1299 cells.²⁴ Protein lysates were first subjected to enzymatic modification by an engineered galactosyltransferase, GalT-(Y289L), which transfers an azide-bearing N-azidoacetyl-galactosamine residue, GalNAz, to O-GlcNAc moieties. The azide tag of the resulting disaccharide allowed for further functionalization with a biotin handle through a Cu(I)-catalyzed click reaction²⁵ using an alkyne-azo-biotin reagent.²⁶ Following streptavidin agarose enrichment and washing of nonbinders, proteins were eluted through the reduction of the azo linker. Subsequent SDS-PAGE separation and Western blotting enabled detection of bona fide O-GlcNAc-modified proteins, including HMGB1, and the highly O-GlcNAcylated Nup62 protein as positive control (Figure 1b). Notably, the non-O-GlcNAcylated β-actin was not enriched in our pulldown lanes.

In order to confirm that modification occurs at S100 and S107 and to compare the relative amounts of modifications at these two sites, we analyzed variants of HMGB1 bearing serine to alanine mutations at these positions. H1299 cells were cotransfected with a FLAG-tagged mutant, either HMGB1(S100A) or HMGB1(S107A), and an HA-tagged wild-type HMGB1(WT) as an internal positive control. Following cell lysis, we again performed chemoenzymatic labeling and biotin pulldown to enrich the O-GlcNAcome. We imaged our inputs and pulldowns simultaneously during Western blotting and normalized the signal of the FLAG mutant to the HA control (Figure 1c). Importantly, an HA-HMGB1(WT) versus FLAG-HMGB1(WT) control experiment showed no effect on pulldown efficiency, indicating that any loss in signal is from reduced O-GlcNAcylation of our mutants. Consistent with the published proteomics data, both the S100A and S107A mutations resulted in reduction of the enrichment, confirming that O-GlcNAcylation occurs at both of these sites. The S100A mutant resulted in a greater loss of O-GlcNAcylation than the S107A mutant, indicating higher modification stoichiometry at the S100 site. Simultaneous mutation of both putative sites in the FLAG-HMGB1(AA) mutant reduced the pulldown efficiency to insignificant levels, indicating that these two residues are the main sites for O-GlcNAcylation.

The S100 glycosite is located at the first helix of the box B domain, where it points toward the bound DNA, based on a solution NMR structure of the human HMGB1 B box domain in complex with duplex DNA²⁷ and crystal structures of HMGB1 bending the DNA in a V(D)J recombination complex (PDB: 6CIJ, Figure 1d)²⁸ and of the Drosophila ortholog HMG-D structure bound to linear DNA.²⁹ Notably, it is only a few residues away from conserved intercalating residue F103 (Figure 1d, bottom). Variant analysis at position 100 of the HMG-domain-containing proteins also identified this serine as a key determining residue for binding certain DNA structures in a sequence-independent fashion.²⁹ Taken altogether, this information led us to hypothesize that O-GlcNAcylation at this S100 could alter HMGB1 interactions with DNA and potentially affect downstream processes where these interactions are necessary.

Semisynthesis of O-GlcNAc HMGB1(gS100).

Currently, the only way to study the site-specific consequences of O-GlcNAc on protein function is to generate homogeneously modified protein and subject it to direct biochemical testing.³⁰ For this purpose, our lab utilizes EPL^31,32 to prepare and study a variety of proteins.³³ This ligation technique traditionally involves the reaction between one peptide fragment bearing an N-terminal cysteine residue as the nucleophile and another bearing a C-terminal thioester as the leaving group. HMGB1 incidentally has three cysteine residues that can theoretically be useful for EPL reactions: C23, C45, and C106. However, none of these sites are conveniently located for the incorporation of O-GlcNAcylated serine at position 100. Hence, we used a cysteine surrogate, selenocysteine (Sec), that would allow us to use a nearby alanine (A94) as a ligation site. Selenocysteine at this position would initially facilitate the ligation reaction, after which it could be converted back to alanine chemoselectively through a radical deselenization reaction.^34,35

As schematized in (Figure 2a), we prepared HMGB1(gS100) using a three-fragment, two-ligation strategy. We first used E. coli to recombinantly express N-terminal residues 1–93 fused to an intein domain from Anabaena variabilis (AvaE).³⁶ Thiolysis of the intein by the addition of sodium mercaptoethanesulfonate (MesNa) afforded the N-terminal thioester fragment 1. Subsequent analyses by MALDI-MS demonstrated that the initiator methionine had been completely removed. We then used solid phase peptide synthesis to prepare the middle fragment 2 corresponding to residues 94–105. This peptide has three important features: an alanine-to-selenocysteine substitution at position 94 (Sec94), a tri-O-acetyl-protected O-GlcNAcylated serine residue at position 100, and a C-terminal hydrazide group. Finally, we also prepared C-terminal fragment 3 through the hydrolysis of a recombinantly expressed HMGB1(106–215)-AvaE fusion protein. Again, cleavage of initiator methionine from this intein fusion protein was efficient. All fragments were purified by semipreparative reversed phase HPLC (RP-HPLC).

Figure 2.

Semisynthesis of HMGB1(gS100). (a) Synthetic scheme outlining the preparation of O-GlcNAcylated HMGB1. Synthetic peptide 2 was incubated with thioester 1. The ligation product was deselenized in the same pot before purification. The resulting product 4 was activated and converted to a thioester prior to the addition of recombinant fragment 3. The ligation product was purified prior to deprotection of the O-acetate protecting groups of the sugar moiety to obtain HMGB1(gS100). (b) Characterization of synthetic HMGB1(gS100) using analytical RP-HPLC (C4 analytical, 0–70% solvent B over 60 min) and MALDI-MS.

Fragments 1 and 2 were subjected to EPL conditions, resulting in facile formation of ligation product. Following buffer exchange to remove the ligation catalyst MPAA, addition of TCEP converted Sec94 to A94, corresponding to intermediate product 4. This product was purified by RP-HPLC, and the C-terminal hydrazide was then converted to a thioester using azide activation and thiol displacement.³⁷ Addition of fragment 3 to the same pot resulted in the formation of ligation product. After purification and deprotection of the sugar’s O-acetyl protecting groups, we obtained HMGB1(gS100) at >95% purity by analytical RP-HPLC (Figure 2b). At each step of the synthesis, fragments, intermediates, and products were characterized by MALDI mass spectrometry to confirm their identities (Figures S1 and S2). Unmodified HMGB1 was also prepared from hydrolysis of its corresponding intein fusion followed by RP-HPLC (Figure S3a). Notably, circular dichroism of recombinant unmodified HMGB1 and semisynthetic HMGB1(gS100) showed closely overlapping spectra, suggesting that O-GlcNAc does not cause dramatic changes to secondary structure (Figure S3b).

O-GlcNAcylation at S100 Reduces Interaction with Negatively Supercoiled DNA.

HMGB1 engages different binding modes and structural features when interacting with different DNA structures.³⁸ In the case of supercoiled DNA, the major mechanism involves ionic interactions with the DNA backbone for torsion conservation. These interactions are mediated by basic residues within the individual HMG box domains. The isolated box A, with its higher proportion of basic residues, exhibits slightly higher affinity for supercoiled DNA³⁹ compared to the isolated B box. However, in the isolated B box, mutation of K96 or R97 to alanine causes a loss in affinity for supercoiled DNA, confirming the importance of net charges for this interaction.⁸ An additional mechanism for the isolated B box DNA binding is the intercalation of F103, which can insert into distorted portions of the DNA. Again, mutation of this position to a less bulky serine residue reduces the ability of the isolated B box domain to introduce supercoils onto DNA.⁸

In order to determine the consequence of O-GlcNAc on supercoiled DNA binding, we used a well-established assay to compare the abilities of our protein variants to protect supercoiled DNA from topoisomerase I-mediated relaxation. Given that HMGB1 is known to interact preferentially to negative over positive supercoils,⁹ we isolated negatively supercoiled pUC19 plasmid from bacteria and preincubated it with varying amounts of HMGB1 or HMGB1(gS100). Prokaryotic topoisomerase I, which can only relax negative supercoils,⁴⁰ was then added to these complexes to initiate relaxation of the DNA. We then digested the proteins with Proteinase K, and the resulting DNA topoisomers were resolved in a fluorophore-free agarose gel followed by staining with the high-sensitivity GelRed dye. The pUC19 plasmid alone runs as expected, showing a major band corresponding to the supercoiled form and smaller amounts of nicked and relaxed isomers. Addition of topoisomerase I in the absence of HMGB1 efficiently converted most of the supercoiled plasmid into relaxed forms as well as bands that we attribute to positively supercoiled forms⁹ (Figure 3a). As expected from previous studies, increasing amounts of HMGB1 prevented this relaxation, with HMGB1 preserving virtually all of the negatively supercoiled DNA at a 200-fold molar excess. Interestingly, HMGB1(gS100) had a reduced ability to prevent relaxation, and this was most evident when comparing the topoisomer distributions at a molar ratio of <100:1. We repeated the experiment and confirmed that O-GlcNAcylation increased the amount of HMGB1 required to maintain negative supercoils. S100’s proximity to both the basic patch and F103 indicates that O-GlcNAc may be interfering with these interactions, resulting in the reduced efficiency of supercoiled DNA binding.

Figure 3.

S100 O-GlcNAcylation alters HMGB1 interactions on DNA. (a) Negatively supercoiled DNA (10 nM was preincubated with HMGB1 or HMGB1(gS100) before treatment with topoisomerase I). HMGB1(gS100) binds less to negatively supercoiled DNA than unmodified HMGB1 based on the higher molar ratio needed to preserve supercoiled structures. (b) Four-way junction DNA (100 nM, “free probe”) is bound by HMGB1 to form a slower-migrating complex “I”. At higher molar ratios, oligomerization results in even slower migration and streaking on the gel. HMGB1(gS100) exhibits higher oligomerization propensity than unmodified HMGB1. (c) Histones and 601 DNA (either 147 bp or 218 bp) were formed into mononucleosomes (“MN”, 50 nM) and subjected to complexation with HMGB1. HMGB1(gS100) binds 147 bp nucleosomes better than HMGB1 (top), and this effect is more pronounced in the 218 bp nucleosomes (bottom). (d) Schematic of the DNA circularization assay. DNA (123 bp, 125 nM) can be ligated by T4 ligase to form linear (L1, L2, …) and cyclic (C1, C2, …) products. In the presence of HMGB1, DNA bending allows formation of exonuclease III-resistant DNA minicircles. (e) The amount of HMGB1 required to form the maximum amount of minicircles (indicated by *) is lower for HMGB1(gS100) than unmodified protein. For quantification of this and a replicate experiment, please see Figure S5. All results shown in this figure are representative of at least 2 experiments.

O-GlcNAcylation at S100 Enhances HMGB1 Oligomerization on Four-Way Junction and Nucleosomal DNA.

HMGB1 exhibits high affinity for damaged DNA and intermediates of DNA repair. One representative structure specifically recognized by HMGB1 is the four-way junction (4WJ) or cruciform DNA.¹⁴ 4WJs recapitulate the biological structure of Holliday crossover junctions, key intermediates during homologous recombination and homology-directed double-strand break repair.^41,42 4WJ DNA consists of four halfcomplementary strands that form four duplexes or “arms”, leaving a junction or “hole” at the center. Mechanistic studies of HMGB1–4WJ binding demonstrated that the isolated box A and B domains can individually bind 4WJ DNA with similar affinities.⁴³ In the full-length protein where both domains are present, the more basic A box occupies the hole where the DNA is most distorted, while the B box interacts with one of the arm regions of the cruciform.⁷ Notably, this higher affinity exhibited by the A box over the B box when the didomain is present is a common feature of HMGB1 binding to other highly distorted DNA structures.^13,44,45 After the formation of the stable 1:1 complex, additional HMGB1 will oligomerize, producing higher molecular weight HMGB1–4WJ complexes.⁴⁶ HMGB1 oligomerization occurs as well on other structures including supercoiled, linear, and circular DNA, and the B box and its flanking linker regions appear to be the major contributor to this process.³⁹

HMGB1 has been shown to bind with nanomolar affinity to 4WJ DNA using gel retardation or electrophoretic mobility shift assays (EMSAs).¹⁴ We similarly utilized EMSAs to establish the effect of HMGB1 O-GlcNAcylation on 4WJ binding. We first prepared 4WJ DNA using published singlestrand sequences¹⁴ and added these to varying amounts of HMGB1 or HMGB1(gS100) to allow complex formation. Separation by native polyacrylamide gel electrophoresis and subsequent staining with GelRed enabled the visualization of HMGB1–4WJ complexes. At lower molar equivalents, we observed a shift of the free probe (Figure 3b) to a slower-migrating band (marked “I”), representing a stable 1:1 complex of HMGB1 with 4WJ DNA. Further increase in HMGB1 equivalents resulted in HMGB1 oligomerization visualized as unresolved staining that appeared higher than complex I (Figure 3b). We did not observe a dramatic difference in the ability of HMGB1 and HMGB1(gS100) to form complex I, with both variants showing comparable amounts of bound DNA at lower molar ratios. In separate EMSA experiments where we utilized a more gradual HMGB1 titration gradient, we confirmed that 1:1 binding affinity is not affected by O-GlcNAc (Figure S4). On the other hand, the oligomerization process is altered by the O-GlcNAc modification, as the HMGB1(gS100) variant formed larger oligomer complexes more readily (Figure 3b). We estimate that this enhancement in oligomerization is about 4-fold, based on the similarity in staining pattern between the 200:1 HMGB1:4WJ and the 50:1 HMGB1(gS100):4WJ lanes and in the higher ratio lanes that follow. Importantly, we observed the same difference in oligomerization in a replicate experiment.

HMGB1 also binds to nucleosomal DNA to displace other proteins already present on chromatin or to reconfigure the region for subsequent binding of other DNA-associated proteins.^38,47 Binding predominantly occurs near the entry/exit sites of the nucleosomes rather than the less accessible histone-bound DNA.⁴⁸ The presence of linker DNA further enhances binding by facilitating protein oligomerization.⁴⁹ In order to test whether O-GlcNAc affects the binding of HMGB1, we again used EMSAs to visualize HMGB1-mononucleosome complexes. Briefly, we purified and refolded histones and assembled mononucleosomes using either 147 bp or 218 bp DNA bearing the 601 nucleosome positioning sequence. Mononucleosomes were then incubated with varying amounts of HMGB1, and complexes were resolved with native polyacrylamide gel electrophoresis followed by GelRed staining.

Using mononucleosomes assembled with the minimal DNA length (147 bp, Figure 3c, top), HMGB1 exhibited weak binding, with diffuse bands forming only at molar ratios of 50 or more. HMGB1(gS100) showed slightly better binding/oligomerization evident from the relatively faster disappearance of the unbound nucleosomes. As expected, binding was generally enhanced when using mononucleosomes assembled with the longer 218 bp DNA (Figure 3c, bottom), with HMGB1 requiring lower molar ratios for complexation. With these nucleosomes, we observed a more pronounced effect of O-GlcNAcylation; the unbound nucleosome band completely shifted at a molar ratio of 20:1 HMGB1(gS100) compared to 50:1 for the unmodified variant. Given that the effect is more defined when extended linear DNA is present and when higher HMGB1 amounts are used, our results point to enhanced oligomerization as the major effect of O-GlcNAcylation, similar to what we have seen in the 4WJ binding studies. Taken together, these data suggest that O-GlcNAc modification improves HMGB1 binding/oligomerization on both 4WJ and nucleosomal DNA.

The enhancement in oligomerization we observed for HMGB1(gS100) is distinct from the effects of B box mutations that affect direct protein–DNA interactions. Specifically, K96A and R97A mutants of the isolated B box showed decreased affinity for 4WJs, while the intercalation mutant F103S demonstrated neither loss in affinity nor gain in oligomerization propensity.⁸ Hence, while we cannot rule out the possibility that O-GlcNAc affects these ionic and intercalation interactions, it is possible that the enhancement toward oligomerization occurs through additional mechanisms. For instance, cross-linking studies have shown that HMGB1 oligomerization is dependent on the presence of DNA and can occur via intermolecular protein–protein interactions likely through the flanking linker regions of the B box domain.⁵⁰ Given that O-GlcNAc is known to relieve protein interfaces on certain protein targets, the release of these surfaces may afford additional protein–protein interactions that can result in the enhanced oligomerization we observed in our experiments.

O-GlcNAcylation at S100 Improves HMGB1 Ability to Circularize Linear DNA.

Although HMGB1 binding is greater for distorted DNA structures, HMGB1 is also able to bend and kink linear DNA. Structural studies have shown that HMG-box domain-containing proteins^27,29,51 can introduce a wide range of bending angles (around 30–130°), with HMGB1 capable of bending the backbone up to 90°. One biochemical approach widely used to examine this property is through ring closure/circularization assays, wherein the bending activity of multiple HMGB1 molecules on short, linear DNA drives the formation of DNA minicircles.⁵² We used a circularization assay (Figure 3d) to compare the relative abilities of HMGB1 and HMGB1(gS100) to circularize a short (123-bp) fragment that was predigested with KpnI to generate sticky ends. Addition of T4 ligase would result in a mixture of linear (L2, L3, …) and cyclic (C2, C3, …) DNA products containing multiple copies of the 123-bp sequence. To isolate the cyclic products, exonuclease III (Exo) can be added, resulting in the digestion of linear DNA. When the 123-bp DNA is initially incubated with HMGB1 prior to T4 ligation, DNA bending and oligomerization by HMGB1 will facilitate intramolecular cyclization, leading to the formation of Exo-resistant 123-bp cyclic products termed minicircles.^50,53

We quantified the amount of minicircles formed by measuring the band intensities at different HMGB1:DNA ratios. As expected, the amount of minicircles increased with increasing molar ratios of HMGB1 but decreased after reaching a maximum (Figure 3e). This falloff is consistent with results from other studies and has been rationalized as the consequence of excess HMGB1 preventing proper alignment of the ends and/or preventing access to the T4 DNA ligase.⁵² We performed this assay using both HMGB1 and HMGB1(gS100) and quantified replicate experiments to determine the effect of O-GlcNAcylation. Regression fitting of specific binding using the premaximum region of the curves allowed us to calculate and compare the amounts of HMGB1 or HMGB1(gS100) required to produce half-maximal amounts of minicircles (Figure S5). Interestingly, HMGB1(gS100) demonstrated a significant reduction in this value (4.903 versus 33.504, p = 0.0002), suggesting enhanced DNA bending.

The circularization assay has been used as a direct readout of DNA bending. However, there are multiple molecular events that lead to the formation of DNA minicircles. These events include (1) the initial low-affinity binding of HMGB1 to linear DNA, (2) introduction of the DNA bend, (3) higher-affinity binding of HMGB1 to distorted DNA, and (4) protein oligomerization. Given that full-length HMGB1 has a binding footprint of only about 20-bp DNA,⁹ it would take multiple HMGB1 molecules to bring together the DNA ends for efficient circularization. Thus, protein oligomerization⁵³ appears to be the major contributing process for the formation of minicircles in this assay. This is supported by the observation that the same structural determinants that modulate oligomerization of the B box (i.e., the flanking linker regions) are also crucial for its ability to form DNA minicircles and for its binding to preformed DNA minicircles.^52,54 Given our results from the 4WJ and nucleosome binding studies, it is likely that the enhancement in circularization we observed for HMGB1(gS100) is a consequence of its improved oligomerization ability. Indeed, regression fitting of the data in Figure S5 also determined a significant (P = 0.0022) 2-fold increase in the Hill cooperativity coefficient for HMGB1(gS100) over the unmodified variant, which we interpret as enhanced binding of subsequent monomers akin to oligomerization.

Loss-of-Function O-GlcNAc Mutant of HMGB1 at Position 100 Exhibits Altered Biochemical Behavior.

A convenient way to study the functional consequences of O-GlcNAcylation in living systems is by introducing serine/threonine-to-alanine mutations to the O-GlcNAcylated site of the protein of interest. Any phenotype associated with the use of such mutants can then be reasonably ascribed to the chronic loss of the protein’s O-GlcNAc status. We envisioned that this approach in living systems would allow us to investigate the functional effects of HMGB1 O-GlcNAcylation in chromatin-related processes. In order to confirm that the introduction of an alanine mutation at position 100 does not exhibit any perturbations to the biochemical behavior of HMGB1, we expressed and purified HMGB1(S100A) and subjected this protein to the same experiments described in previous sections. In DNA supercoiling assays, HMGB1(S100A) minimally affected the ability of HMGB1 to bind and protect negatively supercoiled DNA from topoisomerase I relaxation (Figure 4a). However, in the EMSA with the 4WJ DNA substrate, HMGB1(S100A) behaved differently from wild-type, unmodified HMGB1 (Figure 4b) with its complexes closely resembling those formed by HMGB1(gS100) at higher molar ratios. At lower stoichiometries, HMGB1(S100A) exhibited a significant reduction in binding affinity for the formation of 1:1 HMGB1–4WJ complexes (Figure S4) compared to either unmodified HMGB1 or HMGB1(gS100). Additionally, in DNA circularization assays, HMGB1(S100A) exhibited a half-maximal ratio closer to the HMGB1(gS100) rather than the unmodified variant (Figure 4c, Figure S5). Together, these data demonstrate that the S100A substitution has a direct effect on the biochemical behavior of HMGB1, strongly indicating that the expression of this loss-of-function mutant cannot be used in living systems to accurately study the functional role of HMGB1 O-GlcNAcylation.

O-GlcNAc Modification of HMGB1 Results in Error-Prone Processing of DNA Lesions.

Given the well-documented role of HMGB1 in DNA damage repair pathways and our observations demonstrating how O-GlcNAc alters HMGB1 interactions with DNA structures that represent targets or intermediates of repair mechanisms, we next tested whether HMGB1(gS100) participates differentially in the repair or processing of DNA lesions. We have shown previously that HMGB1 binds a 57-bp ICL-damaged DNA substrate with high affinity and specificity.⁵⁵ Using the same EMSA conditions, we found that O-GlcNAc results in increased oligomerization (Figure S7) similar to our findings with other DNA structures. In contrast, HMGB1(S100A) behaves similarly to unmodified protein. In order to investigate the biological consequence of this altered affinity in the context of DNA damage processing, we then utilized a cell-free assay⁵⁶ wherein purified HMGB1 can be added into cell extracts from which endogenous HMGB1 protein has been depleted. To these extracts, the interstrand cross-linked (ICL) reporter plasmid pSupFG1 is introduced. The cell extracts are then activated and DNA processing is allowed to proceed. Mutation frequencies on the reporter plasmid can then be determined through blue-white screening.⁵⁶ Importantly, the ICL is introduced onto the reporter plasmid through a triplex-forming oligonucleotide (TFO), which allows specific targeting of a defined region in the plasmid.⁵⁶ This allowed us to assess the types of mutations that persist at the end of the assay.

To deplete HMGB1, we used siRNA treatment of human osteosarcoma U2OS cells resulting in ~90% depletion of HMGB1 (Figure S6). Although no detectable increase in mutation frequencies was observed in the undamaged plasmid as a result of the HMGB1-knockdown, ICL-damaged plasmids showed ~30-fold induction in mutagenesis in HMGB1-depleted extracts compared to ~4-fold with wild-type extracts (Figure 5b), confirming the role of HMGB1 in error-free processing of these lesions in U2OS extracts. We then supplemented HMGB1-depleted extracts with unmodified HMGB1, HMGB1(gS100), and HMGB1(S100A) in the DNA repair assays (Figure 5a). Complementing the HMGB1 depleted extracts with recombinant HMGB1 resulted in the expected lower level of mutagenesis in ICL-damaged plasmids (~7-fold), comparable to the mutagenesis yielded from wild-type extracts. This demonstrates that add-back of purified proteins can act as a surrogate for depleted HMGB1. Interestingly, complementing the extract with HMGB1(gS100) reduced the error-free processing of ICL-damaged plasmids, resulting in high mutation frequencies comparable to HMGB1-depleted extracts. Further, we found that the background mutation frequency in undamaged plasmid was also significantly higher when the extracts were supplemented with HMGB1(gS100) compared to all other experiments (Figure 5b). These results suggest an increase in spontaneous replication errors, a reduction in the processing of DNA damage, or a combination of these when HMGB1 is O-GlcNAcylated. Meanwhile, complementing the HMGB1 depleted extracts with HMGB1(S100A) protein also resulted in a reduction of mutation frequencies in ICL-damaged plasmid, although the magnitude of reduction was less than when wild-type HMGB1 was used, suggesting somewhat less efficient ICL processing in the mutant. In contrast to HMGB1(gS100), the background mutation frequency of undamaged plasmid from HMGB1(S100A) add-back was unaffected.

Figure 5.

O-GlcNAc modification of HMGB1 results in error-prone processing of the TFO-directed ICLs in human U2OS whole cell extract. (a) Immunoblot demonstrating supplementation of HMGB1-depleted extract with purified HMGB1, HMGB1(gS100), and HMGB1(S100A) proteins. Wild-type, non-HMGB1 depleted extract was used as control. (b) Mutagenesis assays showing spontaneous mutation frequencies (undamaged plasmid) and TFO-directed ICL induced mutation frequencies (ICL-damaged plasmid) using HMGB1-depleted cell extracts without or with HMGB1/HMGB1(gS100), or HMGB1(S100A) protein add-back. The error bars indicate ± SD. P-values were calculated from one-way ANOVA followed by post hoc Tukey. (c) Mutation spectra generated from sequencing N = 10 mutant colonies from each of the experimental groups in B. Representative sequencing data are shown in Figure S7.

We also sequenced the mutations in the mutation-reporter plasmids to characterize the mutation spectra as a function of the add-back experiments. The sequencing results across the different experiments revealed four distinct types of mutations (Figure 5c, Figure S8). Base substitutions, insertions, and small deletions were mostly localized to and around the targeted thymine base in the TFO-binding region of sequences isolated from wild-type extract. In addition to these types of mutations, sequences recovered from HMGB1-depleted extracts also showed multiple consecutive A → C transversions and T → C transitions in the TFO-binding region, resulting in what we termed “C-tracts.” Sequences from both the HMGB1 and HMGB1(S100A) supplemented groups showed the presence of all four types of mutations at different frequencies. Interestingly, no base substitutions were detected in mutants from the HMGB1(gS100)-supplemented group. Although our limited sequencing data preclude assessment of statistical significance of differences in the observed mutation spectra, taking together the results from these experiments suggests that O-GlcNAcylation of HMGB1 interferes with the repair of ICL-damaged DNA, resulting in error-prone processing and potentially altered mutation signatures.

CONCLUSIONS

Through protein semisynthesis and biochemistry, we discovered that the PTM O-GlcNAc changes the way HMGB1 interacts with different types of DNA structures. These alterations likely occur by influencing known features that determine HMGB1–DNA interactions, specifically (1) the ability to form ionic interactions with the basic linker region located N-terminally of the S100 O-GlcNAc site and (2) HMGB1’s oligomerization propensity. The positive charges from the K96 and R97 residues are important for HMGB1 binding to negatively supercoiled DNA, and given that O-GlcNAc at S100 reduces the affinity of HMGB1 for negatively supercoiled DNA, the sugar moiety may be interfering with the ionic interactions required from this basic region. Interestingly, the S100A mutation does not cause the same loss in affinity for negatively supercoiled DNA, suggesting that O-GlcNAc’s effect on the ionic interactions may be based on its bulk. On the other hand, oligomerization on 4WJ, nucleosomal, and short linear DNA (in circularization assays) is highly enhanced by O-GlcNAc, and this effect is comparable to the enhancement in oligomerization by the S100A mutation in all three assays. Importantly, the distinct behaviors of unmodified HMGB1, HMGB1(gS100), and HMGB1(S100A) in our biochemical experiments indicate that loss-of-function mutation experiments in living systems may not faithfully interrogate the consequences of O-GlcNAcylation.

We also discovered that the perturbations in HMGB1–DNA interactions translate to alterations in activity during DNA damage processing. We observed that while unmodified HMGB1 can efficiently participate in NER-dependent processing of ICL-damaged plasmids,^19,56 O-GlcNAcylation at S100 inhibits this activity. We previously proposed that HMGB1’s participation in NER occurs at least in part via damage-specific architectural modification, i.e., through induction of negative supercoiling in ICL-damaged plasmids.⁵⁶ The observation that HMGB1(gS100) suffers from a loss in affinity for negative supercoils in our biochemical studies may partially explain the loss in error-free processing. On the other hand, HMGB1(S100A), which essentially performed similarly to unmodified HMGB1 in in vitro negative supercoiling assays, also demonstrated a milder loss in DNA processing efficiency. This indicates that other mechanisms may also be at play. We hypothesize that the increased oligomerization propensity of HMGB1(gS100) and HMGB1(S100A) may also contribute to inefficient DNA damage processing, for instance by sequestering the damage site from repair proteins or by interfering with the formation of productive protein complexes such as that with the NER cofactor XPA.⁵⁶ Importantly, we have found that the oligomerization of HMGB1(gS100) on an ICL-damaged DNA substrate is markedly enhanced (Figure S7), and this effect is not recapitulated by the HMGB1(S100A) mutant albeit having demonstrated increased oligomerization on other DNA structures. This further highlights the insufficiency of this mutant as a tool for studying the effects of O-GlcNAc.

While speculative, our results may have implications for O-GlcNAc in cancer. The overall levels of O-GlcNAcylation are increased in tumors and cancer cells compared to healthy tissue,⁵⁷ and O-GlcNAc promotes tumor survival and tumori-genesis in xenografts.^57,58 However, evidence for O-GlcNAc contributing to the initiation of cancer is much more limited. Given our results, we postulate that increased O-GlcNAcylation of HMGB1 may contribute to the accumulation of DNA mutations in cells, thus promoting the transition from healthy to diseased states.

Notably, the S100 residue is also known to be phosphorylated.^59,60 Given that the interplay between O-GlcNAc and phosphorylation is widely documented for different proteins,⁶¹ reciprocal crosstalk between these two PTMs can occur wherein the presence of O-GlcNAc may block the ability of HMGB1 to be phosphorylated or vice versa. This may have important implications especially if phosphorylation affects HMGB1 in a different or opposite manner than what we observed for O-GlcNAc. Importantly, not all PTMs can induce dramatic alterations to protein structure and function; for instance phosphorylation studies of the HMGB1-related protein HMGN1 have recently shown mild to neutral effects on DNA binding.⁶² Thus, understanding the direct effects of phosphorylation of HMGB1 at S100 will be a critical point of interest for future studies, as well as understanding the dynamics of these two PTMs in the context of living systems.

HMGB1 is a jack-of-all-trades protein within the nucleus; hence our approach could enable future investigations on the consequences of O-GlcNAc to HMGB1’s participation in other nuclear processes. Additionally, since the discovery of a proinflammatory function when HMGB1 is released extracellularly by necrotic cells,⁶³ it has also been annotated with numerous roles and functions in immune activation and signaling and is more recently widely studied as a molecule of interest in diverse clinical contexts.^64,65 Our synthetic strategy may be extended to study these other functions,⁶⁶ as well as other PTMs (e.g., phosphorylation at S100) that occur within this region.