Acetylation of C/EBPα inhibits its granulopoietic function

CCAAT/enhancer-binding protein alpha (C/EBPα) is an essential transcription factor for myeloid lineage commitment. Here we demonstrate that acetylation of C/EBPα at lysine residues K298 and K302, mediated at least in part by general control non-derepressible 5 (GCN5), impairs C/EBPα DNA-binding ability and modulates C/EBPα transcriptional activity. Acetylated C/EBPα is enriched in human myeloid leukaemia cell lines and acute myeloid leukaemia (AML) samples, and downregulated upon granulocyte-colony stimulating factor (G-CSF)- mediated granulocytic differentiation of 32Dcl3 cells. C/EBPα mutants that mimic acetylation failed to induce granulocytic differentiation in C/EBPα-dependent assays, in both cell lines and in primary hematopoietic cells. Our data uncover GCN5 as a negative regulator of C/EBPα and demonstrate the importance of C/EBPα acetylation in myeloid differentiation.

H ematopoiesis is a precisely controlled process involving differentiation of multipotential hematopoietic stem cells (HSCs) into specific lineages, regulated by transcription factors. CCAAT/enhancer-binding protein alpha (C/EBPa) is one of the transcription factors that is crucial for both myeloid differentiation and maintenance of quiescence in adult HSCs. The role of C/EBPa in granulopoiesis is demonstrated through Mx1-Cre-driven conditional disruption of C/EBPa in adult mice, resulting in a differentiation block during the transition from common myeloid progenitors to granulocyte monocyte progenitors and increased HSC self-renewal 1,2 . Moreover, expression of C/EBPa in both leukaemic cell lines and in human CD34 þ -enriched HSCs lead to granulocytic differentiation [3][4][5] . Changes in C/EBPa expression and function are associated with impaired myeloid lineage fate decisions while dysregulation of C/EBPa function is frequently associated with leukaemogenesis [6][7][8][9][10] . Acetylation catalysed by lysine acetyltransferases (KATs) is an important post-translational modification (PTM), and KATs play pivotal roles in numerous cellular processes. More importantly, histones are not the only physiological acetylation targets 11 . C/EBPa involvement with KATs has been previously reported [12][13][14][15] , but the role of acetylation in modulating C/EBPa function has not been addressed.
In this study, we identify the ability of KAT GCN5 to interact and acetylate C/EBPa on at least lysine K298 and K302 in the basic DNA-binding domain (DBD), which results in impaired DNA-binding activity. Moreover, wild-type (WT) GCN5 but not the mutant, catalytically inactive GCN5 represses C/EBPa transactivation and short hairpin RNA (shRNA) knockdown of GCN5 results in upregulation of C/EBPa activity. C/EBPa acetylation is detectable in myeloid cell lines, primary leukaemic samples and drops dramatically during granulocytic differentiation of 32Dcl3 cells on granulocyte-colony stimulating factor (G-CSF) treatment. An acetylation mimetic mutant of C/EBPa shows loss of DNA binding resulting in the loss of transcriptional activity as predicted by molecular dynamics (MD) simulations. Our study is the first to provide an understanding of how C/EBPa activity is regulated by KAT at the post-translational level.

Results
C/EBPa is acetylated in its C-terminal region by GCN5. Given C/EBPa expression normally triggers differentiation of immature myeloblasts into mature granulocytes, we investigated why some leukaemic cell lines remain undifferentiated despite high levels of expression of non-mutated C/EBPa. One possibility is the inactivation of WT C/EBPa protein by PTM. Therefore, we investigated whether acetylation is a negative regulator of C/EBPa activity. Using a pan-acetyl-lysine antibody, acetylated C/EBPa was detected in leukaemic cell lines HL-60 and Molm-14 (Fig. 1a). Next, we sought to investigate which acetyltransferase/s represses C/EBPa activity. A C/EBPa-dependent promoter-luciferase reporter construct was used to measure the transcriptional activity of C/EBPa. Co-transfection of various acetyltransferases with a C/EBPa expression plasmid was performed in 293T cells lacking endogenous C/EBPa. Among the acetyltransferases tested, GCN5 was found to repress C/EBPa transcriptional activity (Fig. 1b). Cotransfection of p300 and C/EBPa resulted in an increase in the C/EBPa transactivation activity as previously reported (Supplementary Fig. 1a) 14 . Co-transfection of GCN5 antagonized the ability of C/EBPa to activate the reporter plasmid in a dosedependent manner, without affecting C/EBPa protein levels (Fig. 1b, Supplementary Fig. 1a-c). GCN5 alone had no effect on luciferase activity, indicating that GCN5 specifically mediates downregulation of the reporter through interaction with C/EBPa (Fig. 1b). Expression of these acetyltransferases and C/EBPa were confirmed by using western blot ( Supplementary  Fig. 1b). A co-immunoprecipitation experiment showed both GCN5 and PCAF (p300/CBP associated factor, a GCN5 homologue) interact with C/EBPa ( Supplementary Fig. 1d). To address the question whether acetyltransferase activity of GCN5 is required to repress C/EBPa transcriptional activity, an acetyltransferase-defective mutation of GCN5 (-HAT, Y621A/ P622A) was used in place of WT GCN5 (ref. 16). Indeed, the mutated HAT-deficient form failed to repress C/EBPa transcriptional activity, indicating that acetyltransferase activity is essential for repression (Fig. 1c, Supplementary Fig. 1c). In a reciprocal experiment, C/EBPa transcriptional activity doubled when endogenous GCN5 was efficiently knocked down (shGCN5 # 3; Fig. 1d,e). Taken together, our data indicate that GCN5 inhibits C/EBPa transactivation in an acetyltransferase-dependent manner, although the role of other HATs in this process cannot be excluded.
Subsequently, we performed an in vitro acetyltransferase assay using GCN5 and C/EBPa peptides. K298, K302 and K326 were identified as the sites of acetylation by GCN5 (Fig. 1f, Supplementary Fig. 1e-g and Supplementary Table 1). These lysine residues have high degree of evolutionary conservation across different species, suggesting crucial role for C/EBPa function ( Supplementary Fig. 1h). K298 and K302 are exposed on the basic DBD, whereas K326 resides in the leucine zipper dimerization domain (Fig. 1g) 17,18 .
To further investigate the protein domains involved in the C/EBPa-GCN5 interaction, we performed co-immunoprecipitation assays in 293T cells ( Supplementary Fig. 1i). While immunoblot analysis using FLAG and V5 antibodies revealed that GCN5 interacts with C/EBPa WT, C/EBPa 1-207, and C/EBPa p30/120-358, it failed to interact with C/EBPa 204-358 ( Supplementary Fig. 1j). By performing pull-down assays with FLAG antibody for C/EBPa-TAD1 (Transactivation domain 1), and C/EBPa-DBD separately, we were unable to detect any interaction between GCN5 and TAD1 or DBD domain of C/EBPa ( Supplementary Fig. 1k). Collectively, these observations suggest that the GCN5 interaction domain in C/ EBPa lies in the N-terminal region of C/EBPa ( Supplementary  Fig. 1l).
The relevant lysine residues (K298, K302 and K326) were substituted with arginine to generate non-acetylated mimetic forms of C/EBPa (referred to as K3R). We further tested whether a pan-acetyl antibody is able to detect acetylation differences between C/EBPa WT and K3R or C/EBPa-DBD and C/EBPa-DBD K3R ( Supplementary Fig. 1m). Immunoprecipitated C/EBPa WT or K3R mutant showed no difference in acetylation using a pan-acetyl antibody, both with (lanes 4 and 5) and without (lanes 2 and 3) GCN5 co-transfection. In addition, co-transfection with DBD or DBD K3R did not show any acetylation signal using a pan-acetyl antibody (lanes 6 and 7). Immunoprecipitated WT, K3R, DBD, and DBD K3R were detected by using V5 antibody. These results are in accordance with our domain-mapping data, suggesting that the C/EBPa DBD domain does not interact with GCN5, and therefore no acetylation signal is observed from either DBD or DBD K3R when co-transfected with GCN5 ( Supplementary Fig. 1l).
To detect acetylation of C/EBPa in cells at K298, K302 and K326, site-specific anti-acetyl-C/EBPa antibodies were generated using synthetically acetylated peptides. The acetylated and non-acetylated forms of these peptides were first confirmed by mass spectrometry. Our antibodies were able to readily recognize acetylated C/EBPa at K298, K302 and K326. When a non-acetylated mimetic form of C/EBPa, that is, K3R was used, no signal was detected, confirming that the antibodies we generated are capable of specifically detecting acetylated C/EBPa ( Supplementary Fig. 1n). Consistently, western blotting with these site-specific acetylation antibodies showed an increase in acetylated C/EBPa when GCN5 and C/EBPa were co-expressed in 293T cells ( Supplementary Fig. 1o). We also examined whether K298, K302 and K326 were acetylated in HL-60 and Molm-14, and the results are consistent when probed with site-specific antibodies (Fig. 1h). These data indicate that our acetylationspecific antibodies were able to detect C/EBPa acetylation in the DBD of C/EBPa.
Loss of C/EBPa acetylation on myeloid differentiation. We looked at whether endogenous C/EBPa is acetylated at K298, 302 and 326 in vivo and if the acetylation status of C/EBPa changes with respect to myeloid differentiation. Within the hematopoietic system, expression of C/EBPa is detectable in early myeloid precursors and its expression is necessary and sufficient for neutrophilic differentiation 5,19,20 . We used non-leukaemic 32Dcl3 cells to assess C/EBPa acetylation status on differentiation. Murine 32Dcl3 cells are dependent on interleukin-3 (IL-3) for survival and proliferation, and readily differentiate into mature granulocytes on removal of IL-3 and addition of G-CSF 21 4 TK promoter in a dose-dependent manner and is dependent on the GCN5 histone acetyltransferase (HAT) domain. 293T cells were transiently transfected with p(CEBP) 4 TK, pRL-null, and pcDNA6 expression plasmids for C/EBPa and with GCN5 or GCN5 (-HAT), respectively. Protein expression of corresponding constructs were shown in Supplementary Fig. 1c. Luciferase activity was measured in duplicate for each experiment and data are shown as mean ± s.d. (N ¼ 3). (d) Knockdown of endogenous GCN5 results in an increase in C/EBPa transactivation potential. GCN5 knockdown in 293T cells enhanced C/EBPa transactivation capacity on a minimal p(CEBP) 4 TK promoter with pRL-null and pcDNA6 C/EBPa plasmids. Firefly luciferase readings were normalized against internal control renilla luciferase. Luciferase activity was measured in duplicate for each experiment and data are shown as mean±s.d. (N ¼ 3). (e) Western blot demonstrating knockdown efficiencies of endogenous GCN5 by 3 different shRNAs in 293T cells used in d. (f) GCN5 acetylates C/EBPa at K298, K302 and K326. C/EBPa peptides were incubated with the GST-HAT domain of GCN5 in the presence of 3 H-labeled acetyl-CoA. Tritium incorporation by C/EBPa peptides was measured by scintillation counting. Error bars represent mean ± s.e.m. Experiments were performed twice in duplicate. (g) Structure of the C/EBPa basic region-leucine zipper domain bound to DNA. Arrows indicate the location of the acetylated lysine residues. (h) Endogenous C/EBPa is acetylated in the basic leucine zipper region in HL-60 and Molm-14 cells. C/EBPa protein was immunoprecipitated using a rabbit antibody recognizing total C/EBPa protein, followed by immunoblotting against acetylated C/EBPa antibodies (K298, K302, and K326). Rabbit anti-GFP antibody was used as a control for IP. **Po0.01 and ***Po0.001; Student's unpaired t-test (b,d and f).
FACS and Giemsa-stained cytospins analysis revealed an efficiency of Z80% in granulocytic differentiation of 32Dcl3 cells on day 4 on G-CSF induction ( Supplementary Fig. 2a). Western blot results showed enrichment of acetylated C/EBPa for K298 and K302 in 32Dcl3 cells before differentiation by G-CSF is initiated (Fig. 2a). On G-CSF induction, there was a reduction in acetylation levels for K298 and K302, which coincides with the loss of GCN5 protein expression. We then used acetylation-specific antibodies to determine the endogenous levels of C/EBPa acetylation in primary acute myeloid leukaemia (AML) samples in which C/EBPa levels could be detected by western blot. Undifferentiated human CD34 þ cells expressed such low levels of C/EBPa protein that we were unable to compare their C/EBPa acetylation signal with AML. Therefore, we used G-CSF treated CD34 þ cells on day 7, in which C/EBPa protein expression was detected by western blot. C/EBPa was acetylated at K298 and K302, and GCN5 was higher in AML samples than in partially differentiated human CD34 þ cells (day 7; Fig. 2b, Supplementary Fig. 2c). We did not observe a stronger acetylated signal at K302 in AML samples #3, #4 and #5 as compared with other AML samples (Fig. 2b). We were also unable to detect any signals using acetylated K326-specific antibody in AML samples and in G-CSF-induced differentiated 32Dcl3 cells (Supplementary Fig. 2b). Our AML sample size was insufficient to distinguish the difference between K302 acetylation, and lack of K326 acetylation (Supplementary Table 2). Our data indicate loss of C/EBPa acetylation at K298 and K302 on G-CSF-mediated differentiation of murine 32Dcl3 cells. In addition, C/EBPa acetylation is easily detectable in AML patients compared with partially differentiated human CD34 þ cells. Interestingly, GCN5 is expressed at higher levels in AML patients than in normal CD34 þ cells ( Supplementary Fig. 2d).
Predicting loss of DNA binding for acetylated C/EBPa. On acetylation, lysine side chains are changed in that the acetyl group neutralizes the positive charge. Substitution of lysine with glutamine (neutral side chain) mimics the acetylated form 11 . To examine whether glutamine substitution was consistent with lysine acetylation and to examine the impact of C/EBPa acetylation on its structure (Fig. 3a), we conducted MD simulations on four model systems. In addition to the WT DNA and acetylation mimic (K2Q-DNA) systems, two acetylated models were also included in this study. The H-N-C ¼ O dihedral angle of the acetylated lysine residues was set to 0°and 180°in the two acetylated models (hereafter termed as K2Ac_a-DNA and K2Ac_b-DNA, respectively, Fig. 3b; see Computational methods section for details). The backbone atoms of the four models showed similar fluctuations during MD simulations ( Supplementary Fig. 3). As compared with the C/EBPa WT-DNA complex, lysine acetylation (K2Ac_a-DNA and K2Ac_b-DNA) and acetylation mimic (K2Q-DNA) raised the calculated protein-DNA-binding free energies substantially (Table 1). This indicates a decrease in DNA-binding potency of the models (K2Ac_a-DNA, K2Ac_b-DNA and K2Q-DNA). According to the crystal structure of C/EBPa WT DNA, one of the key interactions is formed between K298 and A -5 (ref. 18). During MD simulations, the three ammonium hydrogen atoms of K302 formed hydrogen bonds with the phosphate oxygen atoms of T -4 ( Fig. 3c, Supplementary Table 3). The electrostatic interaction energy between K298 and K302, and DNA residues T -4 and A -5 is a major component of the total protein-DNA-binding energy. Both acetylated lysine and glutamine formed significantly less hydrogen bonds with DNA during the 3 ns MD simulations (Supplementary Table 3). Furthermore, the electrostatic stabilization between the C/EBPa protein and DNA residues were significantly lowered in the mutant systems (Supplementary Table 4). Acetylation might also destabilize the C/EBPa structure. 15 ms MD simulations have shown higher conformational fluctuations and loss of enzymatic activity both during acetylation and mutation of K104 to glutamine in RAS 23 . Similar structural destabilization might also be responsible for the loss of C/EBPa transcriptional activity on acetylation.

Acetylation mimetic mutants impairs myeloid differentiation.
To investigate the consequence/s of acetylation on C/EBPa function, we generated and tested single and various combinations of K298, K302 and K326 sites in which lysine was replaced with either arginine (hereafter termed as non-acetylated mimetic) or glutamine (acetylation mimetic) in C/EBPa-dependent  reporter assays. Single acetylation mimetics of all three residues did not have any major effect on C/EBPa transactivation ( Supplementary Fig. 4a). Interestingly, K298 and K302Q (C/EBPa-K2Q) and K298, K302 and K326Q (C/EBPa-K3Q) acetylation mimetic mutation combinations completely abolished C/EBPa transactivation ( Supplementary Fig. 4a,b). Western blots of the C/EBPa mutations confirmed similar expression levels, indicating that the loss in reporter activity is not due to reduction in C/EBPa protein levels ( Supplementary Fig. 4a,b). All corresponding lysine-to-arginine mutations had luciferase activity similar to that of C/EBPa-WT, substantiating that mimicking  acetylation results in the inactivation of C/EBPa function.
To eliminate the possibility of altered conformation and affected transactivation by clusters of lysine-to-glutamine mutations, another mutated form (C/EBPa-KIIIQ) was generated to mimic acetylation of a cluster of three lysine residues (K275, K277 and K278) located close to the DBD. These three lysine residues were substituted with glutamine. Although both mutants represented acetylation mimetic forms, unlike the C/EBPa-K3Q form, C/EBPa-KIIIQ indeed retained transcriptional activity ( Supplementary Fig. 4b). This strongly argues that retaining the positive charge is only crucial for the residues in the basic region, and substituting lysine residues for glutamine does not always result in loss of function. The C/EBPa-deficient human cell line K562 was engineered to conditionally express b-estradiol-inducible constructs to study how various C/EBPa mutants affect differentiation 4 . Without induction, fusion proteins are constitutively expressed in the cytoplasm, where they are inactive. On induction with b-estradiol, the fusion protein translocates to the nucleus, where it is active. Consistent with results from MD simulation and luciferase assays, cells expressing C/EBPa WT-oestrogen receptor (ER), C/EBPa K3R-ER and C/EBPa K2R-ER differentiated, whereas cells expressing ER only, C/EBPa K3Q-ER and C/ EBPa K2Q-ER, remained in an undifferentiated state as assessed by surface expression of the myeloid marker CD11b, cell morphology and neutrophilic enzymatic activity (Fig. 4a,b, Supplementary Fig. 4c-f). Expression levels of the individual fusion protein in K562 stable clones were assessed by western blots (Supplementary Fig. 4g). Furthermore, our results showed that K562 C/EBPa-WT-ER was able to inhibit the increase in cell numbers within 5 days of culture, as compared with vector-alone and vehicle (ethanol) control ( Supplementary  Fig. 4h). The K562 C/EBPa-K2Q mutant cell line showed an intermediate inhibition in cell number, which lacked statistically significant difference (P value ¼ 0.148, N ¼ 6 with two-way analysis of variance test; Supplementary Fig. 4h). C/EBPa WT-ER was able to downregulate endogenous c-Myc as previously reported by our group whereas C/EBPa-K3Q and C/EBPa K2Q-ER mutants showed partial downregulation of c-Myc (Supplementary Fig. 4i) 24 .
Using a murine model, we further explored whether the acetylation mimetic form is unable to differentiate progenitors into mature granulocytic cells. As granulocyte differentiation is impaired in Cebpa fl/fl Mx1-cre þ mice, we chose this model to analyse the differentiation potential of the various C/EBPa mutants. Ten-to twelve-week-old Cebpa fl/fl Mx1-cre þ mice and control (Cebpa fl/fl Mx1-cre -) mice were treated with poly(I:C) 1,2,20 . Bone marrow was harvested 17-21 days later, and excision of the Cebpa gene was confirmed by PCR. Multipotent progenitor Lin À Sca-1 þ c-Kit þ (LSK) cells from Cebpa D/D mice were FACS-purified and cultured. Cells were transduced with retroviruses-expressing green fluorescent protein (pMSCV IRES GFP, MIG empty vector), WT (C/EBPa-WT), or mutated forms of C/EBPa (C/EBPa -K3Q and C/EBPa-K3R). Expression levels of MIG C/EBPa-WT and mutants were assessed by western blot (Supplementary Fig. 4j). After 7 days of culture, the percentage of granulocytic differentiation was measured by Gr-1 and Mac-1 surface staining and by Giemsa stain for morphological analysis (Fig. 4c,d). The C/EBPa-K3Q acetylation mimetic mutation was notably unable to form mature granulocytes. These results suggest that acetylation at K298, K302 and K326 antagonizes C/EBPa granulocytic differentiation potential. In addition, acetylation of K298 and K302 alone is sufficient to impede differentiation.
Further testing of the possibility of GCN5 involvement in myeloid differentiation was done by using 10-12 weeks old Gcn5 D/D Vav-iCre þ and control mice (Gcn5 fl/fl ) 25,26 . LSK cells were FACS-purified and placed in myeloid culture conditions containing G-CSF. LSK cells from Gcn5 conditional knockout (Gcn5 D/D ) produced a 1.7-fold increase in the percentage of Gr-1 þ Mac-1 þ cells compared with control mice (Gcn5 fl/fl ; In conclusion, G-CSF induced in vitro myeloid differentiation is enhanced in the absence of GCN5.
Acetylation at K298 and K302 reduces DNA-binding ability.
To understand how acetylation of K298 and K302 leads to C/EBPa inactivation, we assessed changes in homodimerization, cellular sublocalization and dominant negative function of the mutant forms over C/EBPa WT. No difference in homodimerization or localization between C/EBPa WT and mutated forms was detected ( Supplementary Fig. 5a,b). In a C/EBPa-dependent promoter-luciferase reporter assay, co-transfection of the C/EBPa-K2Q mutant together with C/EBPa WT did not show any dominant negative effect by the K2Q mutant, unlike the C/EBPa p30 form ( Supplementary Fig. 5c) 27 . Expression of C/EBPa, C/EBPa p30 form, and C/EBPa-K2Q mutant were confirmed using western blot ( Supplementary  Fig. 5d).
To investigate whether acetylation by GCN5 alters C/EBPa DNA binding in vitro, we performed electrophoretic mobility shift assays (EMSA). C/EBPa was acetylated by an in vitro acetyltransferase assay using GCN5 HAT domain protein. EMSA analysis showed that acetylation of in vitro-translated C/EBPa reduces its ability to bind to the C/EBPa binding site (Fig. 5a). No binding was observed in the control lysate. Western blot analysis of the reaction used for EMSA confirmed C/EBPa acetylation at K298 and K302 residues (Fig. 5b). MD simulations predicted a dramatic reduction in the ability of C/EBPa to bind to its target DNA sequences by the acetylation mimetic C/EBPa-K2Q (Fig. 3c, Table 1). The DNA-binding abilities of WT and mutated forms of C/EBPa were also assessed by EMSA, using nuclear extracts from K562 lines expressing ER empty vector, C/EBPa WT-ER, C/EBPa K2Q-ER and C/EBPa K2R-ER. As expected, C/EBPa K2Q-ER extracts demonstrated weakened DNA binding as compared with C/EBPa WT-ER and C/EBPa K2R-ER. Using anti-C/EBPa antibody for supershift, we demonstrated specificity of the C/EBPa-dependent binding in EMSA ( Fig. 5c and Supplementary Fig. 5f). EMSA performed using 293T cells transfected with C/EBPa K2Q without the ER-tag showed loss of DNA binding similar to K2Q-ER ( Supplementary Fig. 5e,g). The G-CSF receptor is an important downstream target gene of C/EBPa, and its promoter region contains C/EBPa-binding sites 28 . Minimal recruitment of C/EBPa K2Q-ER to the G-CSF receptor promoter by chromatin immunoprecipitation (ChIP) was observed (Fig. 5d). In summary, these results support the hypothesis that acetylation of these two residues leads to the loss of DNA-binding activity and subsequent loss of induction of granulocytic differentiation.

Discussion
In this study, we demonstrated that acetylation on C/EBPa leads to the loss of differentiation potential. We identified three lysine residues (K298, K302 and K326) as the acetylation target sites in C/EBPa by GCN5. With acetylation site-specific antibodies, we detected acetylation at these three sites in leukaemic cell lines expressing endogenous C/EBPa. We provided evidence that C/EBPa acetylation levels and GCN5 expression are reduced on granulocytic differentiation. Furthermore, we demonstrated C/EBPa acetylation in primary leukaemic samples. Using C/EBPa-dependent murine and cellular models, we demonstrated that the acetylation mimetic forms (K2Q and K3Q) are incapable of driving granulocytic differentiation in inducible C/EBPadependent assays and in murine primary cells as a result of loss of DNA binding ability. This is the first report demonstrating an acetyltransferase that has repressive activity on C/EBPa function, consistent with the reciprocal pattern of expression of GCN5 and C/EBPa in HSCs, in which GCN5 is enriched, while C/EBPa is barely detectable 1,29 . Previous studies have described the involvement of multiple PTMs in modulating C/EBPa [30][31][32][33] . In particular, phosphorylation of serine 21 has been reported to be clinically relevant during myeloid differentiation. Hyper-phosphorylation at S21 blocks the maturation of leukaemic blasts and neutrophilic differentiation of CD34 þ progenitors 9,10 . Although both acetylation and phosphorylation impair the ability of C/EBPa to drive differentiation, the mechanisms of inactivation appear to be distinct. Acetylation results in a loss of DNA-binding capacity, while phosphorylation alters conformation of C/EBPa dimers 34 . The differential effects of these PTMs on C/EBPa are attributes to the inherent differences in the chemistry and sites of these modifications.
Our protein domain mapping data for C/EBPa indicates that GCN5 docking interaction occurs at the N terminus, whereas acetylation sites are located at the C terminus in the DBD. Such observations have been reported in C/EBPa and E2F in which deletion of the C/EBPa N terminus or the basic region mutation leads to loss of inhibition of E2F activity 4 . Several studies have reported a number of C/EBPa mutations lacking differentiation potential, some of which are located in the basic region. The BRM2 (I294A, R297A) mutant protein fails to differentiate cells into granulocyte and adipocyte lineages. Unlike C/EBPa-K2Q described in our study, BRM2 does not abrogate DNA binding 17,35 . The BR3 (R297G, K298T, R300G, K302N) mutant contains four substitutions in the basic region. Similar to our results in C/EBPa-K2Q mutations, although BR3 is also incapable of binding DNA, protein retains the ability to dimerize 36 . However, BR3 was designed on a structural basis, in which clusters of amino acids within the basic region exhibit a net positive charge. This differs from C/EBPa-K2Q mutations, which was selected based on the acetylation sites catalysed by GCN5. This is consistent with the idea that the C/EBPa-K2R, which retains a positive charge on these residues, is indistinguishable from WT C/EBPa in its behaviour in DNA binding and transactivation. These data show that the positive charge on K298 and K302 located within the basic region is required for DNA binding. In fact, our molecular simulation data predicted that acetylation disrupts the capacity to bind DNA by neutralizing the positive charge on these lysine residues. It is important to maintain a positively charged cluster of amino acids for the interaction with DNA. In agreement with these data and by using EMSA and ChIP assays, we demonstrated that the acetylation mimetic mutation (K2Q) is sufficient to decrease DNA binding. The C/EBPa-K2Q mutant (or acetylated C/EBPa) lacks DNA binding; however, it does not display dominant negative function over C/EBPa WT.
In comparison with C/EBPa WT, various mutants differ in the ability to drive granulocytic differentiation. We focused on the complete lack of transactivation and/or granulocytic differentiation consistently observed in K2Q and K3Q mutants of C/EBPa. K326 lies in the dimerization domain of C/EBPa, and substitution with glutamine (an acetylation mimetic) also prevented differentiation compared with C/EBPa WT and K326R. Such observations could also be due to the differences in protein-protein interactions, other than the homodimerization, which is required for efficient granulocytic differentiation. Transcription factor Foxo1 acetylation has been linked to decreased ability to bind DNA, and to become phosphorylated 37 . It will be intriguing to examine whether there is crosstalk between different PTMs on C/EBPa. Furthermore, substitution of lysine with arginine creates potential new methylation sites resulting in altered differentiation potential of mutant C/EBPa. At the very least, substitution of lysine with glutamine (acetylation mimetic) eliminates the potential for such self-modification sites.
To investigate the role of GCN5 in hematopoiesis, we crossed Gcn5 fl/fl mice with hematopoietic-specific Vav-iCre. We did not observe any substantial alteration in myeloid lineage commitment and differentiation in Gcn5 excised mice (Gcn5 D/D Vav-iCre þ ). However, by testing in vitro granulocytic terminal maturation in a more refined experiment aimed at analysing the effect of G-CSF on LSK cells from Gcn5 D/D Vav-iCre þ , we observed a 1.7-fold increase in Gr-1 þ Mac-1 þ cells. The possibility exists that this model, in which GCN5 was deleted at the earliest stages of embryonic hematopoietic development, does not represent an ideal model to study C/EBPa-dependent myeloid commitment in adult hematopoiesis. We cannot rule out the possibility that C/EBPa acetylation may be compensated by other HATs. PCAF shares high homology (88%) with GCN5 in the acetyltransferase domain, and is reported to be redundant with . While we did not observe PCAF-mediated repression of C/EBPa transactivation potential in transient transfection assays, given these previous reports of redundancy of GCN5 and PCAF in vivo, future studies involving GCN5 and PCAF double knockout mice could highlight the role of C/EBPa acetylation in hematopoietic cells.
GCN5 has been reported to be highly expressed in non-small cell lung cancer, which correlates with tumour size 41 . GCN5 also represses PGC-1a and b transcriptional activity via acetylation in cultured hepatocytes in mouse liver tissues and in primary skeletal muscle cells. This acetylation has been linked to changes in glucose concentration and controls expression of metabolic genes involved in fatty acid oxidation 16,42 . C/EBPb is also reported as a target of GCN5-mediated acetylation during glucocorticoid stimulated preadipocyte differentiation of 3T3L1 and NIH3T3 cells 43 . Given the role of C/EBPa in lung development, glucose metabolism and adipogenesis, it would be of interest to investigate whether C/EBPa acetylation participates in these cellular processes in addition to its critical role in granulocytic differentiation. C/EBPa is inactivated in leukaemia by diverse molecular mechanisms. In this study, we demonstrated C/EBPa PTM (acetylation) in primary leukaemia cells, which may affect its activity. It will be of great interest to investigate whether the administration of a GCN5 inhibitor represents a novel therapeutic approach for the treatment of leukaemias.
32Dcl3 cells (kindly provided by Motomi Osato, Cancer Science Institute, Singapore) were maintained in RPMI 1640 supplemented with 10% FBS and IL-3 (5 ng ml À 1 ; PeproTech). Induction of granulocytic differentiation was performed with removal of IL-3 by three washes with PBS and addition of recombinant human G-CSF (100 ng ml À 1 ; PeproTech). Cells were harvested for western blot (50 ng ml À 1 ; PeproTech) to induce granulocytic differentiation. IL-6 was replaced with IL-3 (10 ng ml À 1 ) and G-CSF was reduced to 20 ng ml À 1 in the culture media from day 4 onwards. Fresh media was replenished every 3 days and cells were harvested on day 7 for western blot analysis.
Patient AML samples. We obtained frozen bone marrow blasts of AML from patients at the National University Hospital, Singapore. The institutional review board of the National University of Singapore has approved the study. All patients provided written informed consent in accordance with the Declaration of Helsinki Principles. A summary of leukaemia status of the patient samples is described in Supplementary Table 2.
Luciferase reporter assays. The firefly luciferase reporter gene in the pXP2 vector is driven by four C/EBP-binding sites derived from G-CSF receptor promoter, starting from À 57 to À 37 bp upstream of the major transcription start site of the G-CSF receptor gene 28 . As an internal control plasmid for co-transfections, the pRL-null construct encoding a Renilla luciferase gene (Promega) was used 44 . Firefly and Renilla luciferase activities were determined 24 h post transfection with the dual-luciferase reporter assay system (Promega). Firefly luciferase readings were normalized against internal control Renilla luciferase and calculated as fold differences against the activity obtained for the reporter plasmid without C/EBPa transfection. Luciferase assays were done with p(CEBP) 4 TK (200 ng), pRL-null (10 ng), pcDNA6 expression plasmids for C/EBPa (5, 10, 20 or 30 ng), GCN5 or GCN5 (-HAT) (20, 50, 100, 200 ng) and GCN5 or p300 (1, 5, 20 or 50 ng), respectively. For GCN5 knockdown effect on C/EBPa transactivation, 100, 500 pg and 2.5 ng of pcDNA6 C/EBPa plasmid concentrations were used. C/EBPa acetylation mimetic and non-mimetic mutants luciferase assays were performed using 500 pg, 1 and 5 ng concentrations of respective pcDNA6 expression plasmids. Luciferase assays where pcDNA6 C/EBPa K2Q was compared with pcDNA6 C/EBPa K298Q or pcDNA6 C/EBPa K302Q 1, 5 and 50 ng plasmid concentrations were used (Supplementary Fig. 4a).
In vitro acetylation assay. A GST-tagged GCN5 HAT domain expression construct was provided by Axel Imhof (LMU, Munich, Germany). GST-tagged GCN5 protein was produced and purified from Escherichia coli BL21. GST-tagged recombinant GCN5 (Catalogue #K311-381G) was also purchased from SignalChem. In vitro-translated C/EBPa protein was produced in a coupled transcription and translation rabbit reticulocyte lysate (Promega, Catalogue #L1170). Peptides were synthesized by first Base (Singapore; Supplementary Table 1). All peptides were purified by high-performance liquid chromatography to obtain 495% purity. Substrate peptides were incubated with 0.25 m Ci [ 3 H] acetyl coenzyme A (Amersham) and 0.1 mg of purified GCN5-GST acetyltransferase (HAT) domain in a 30 ml acetylation buffer containing 50 mM Tris-HCl (pH 8.0), 50 mM NaCl, 10% glycerol, 0.5 mM EDTA, 1 mM dithiothreitol, 10 mM sodium buytrate and 1 mM phenylmethylsulfonyl fluoride. Reaction mixtures were incubated at 30°C for 45 min. Subsequently, the peptides were spotted onto P81 filter paper (Whatman). The filter paper was air-dried and measured for acetylation activity using liquid scintillation counting.
Mass spectrometric analysis performed on the in vitro acetylated C/EBPa peptides that cover K298, K302 and K326 lysine residues confirmed acetylation of these three residues and uncovered an additional acetylation site at K304 (Supplementary Fig. 1e-g). But as acetylation at K304 was not seen in the in vitro acetyltransferase assay with K304 in overlapping peptides (Supplementary Table 1 for P-7 and P-8; Fig. 1f, Supplementary Fig. 1e-g), no further studies on this lysine residue were pursued.
Liquid chromatography mass spectrometry analysis. Peptide samples were analysed using nano-HPLC coupled to an LTQ Velos (Thermo Fisher Scientific). Peptides were trapped onto a C18 pre-column and separated on an analytical column using a 1 or 2-h gradient ranging from 2% to 8% acetonitrile/0.1% formic acid, followed by a 5 min gradient ranging from 8% to 80% acetonitrile/0.1% formic acid and stayed at 80% acetonitrile/0.1% formic acid for 20 min. MS scans ranged from 310 to 1,400 m/z, AGC target 3E4, and maximum injection time of 50 ms. The 10 most intense ions with an ion intensity above 1,000 and a charge state excluding one were sequentially isolated to a maximum AGC target value of 4E4 for a maximal 100 ms and fragmented by Electron Transfer Dissociation (ETD). The following parameters were employed for ETD: the reagent ion source temperature of 160°C, reagent ion source emission current of 50 mA, reagent ion source electron energy of À 70.00 V, reagent ion source chemical ionisation of 20 p.s.i., reagent vial 1 ion time of 50, reagent vial 1 AGC target of 1E5, reagent vial 2 ion time of 50, reagent vial 2 AGC target of 1E5, supplemental activation energy of 15. A dynamic exclusion list was applied using an exclusion list size of 500, one repeat count, repeat duration of 45 s, exclusion duration of 30 s as well as a mass width of 1.0 low and 1.5 high. Expiration count was disabled.
Data processing and database search. Raw file processing for the peptide samples was carried out using Mascot Daemon (version 2.3.2, Matrix Science). Data import filter for precursor masses was from 700 to 4,000 Da, with a minimum scan per group of 1 and a minimum peak count of 10. Mascot search was performed using the custom made database consisting of three peptide sequences only: RLRKRVEQLSRC, IAVRKSRDKAKQRC and RKSRDKAKQRNVEC. No enzyme cleavage was specified. Acetyl (C), Acetyl (K), Acetyl (N-term), Acetyl (S), Oxidation (M) were set as variable modifications. Tolerance for the precursor masses was 2 Da and for fragments 0.8 Da. The highest score of MS/MS spectrum for each intact unique acetylated peptide (with a minimum mascot score of 25) was manually validated.
Generation of site-specific acetylation antibodies. Acetylation-specific C/EBPa antisera was raised in rabbits against keyhole limpet haemocyanin (KLH)conjugated peptides, which comprised of the following residues of human C/EBPa: IAVRK Ac SRDKAKQR (Ac-K298 antiserum), IAVRKSRDK Ac AKQR(Ac-K302 antiserum) and RLRK Ac RVEQLSR (Ac-K326 antiserum), with a C-terminal cysteine residue added to the peptide sequence. Addition of cysteine was for the conjugation of the synthetic peptide to the carrier protein KLH. Antisera were purified by binding to acetylated peptide-conjugated Sulfolink gel (Pierce) and subsequently passed through an additional non-acetylated peptide-conjugated Sulfolink gel to eliminate non-acetylation-specific antibodies. Purified acetylationspecific antibodies were used at a dilution of 1:1,000 for western blotting.
Immunoblotting and Immunoprecipitation. For whole-cell lysis, 1 Â 10 7 cells were lysed with radioimmunoprecipitation assay buffer. For subcellular fractionation, nuclear and cytosolic fractions were prepared using Nuclear Extract Kit from Active Motif. Protein concentrations were quantitated with Biorad Protein Assay. Proteins were separated on 10% SDS-PAGE gels. Immunoblots were incubated with primary antibody (Supplementary Table 5) overnight at 4°C, followed by a secondary horseradish peroxidase-conjugated antibody at room temperature for 1 h.
For immunoprecipitation of C/EBPa, 800 mg of whole cell lysate was used. radioimmunoprecipitation assay buffer was diluted four times and used as immunoprecipitation buffer with antibody. Lysate was pre-cleared with rabbit IgG antibody (Cell Signaling Technology) and Dynabeads Protein A (Invitrogen), while antibody for immunoprecipitation was bound to Dynabeads Protein A for 1 h at 4°C. Subsequently, the pre-cleared lysate was incubated with the beads-bound antibody for 3 h with constant rotation at 4°C. Beads-bound immunoprecipitates were washed thrice with immunoprecipitation buffer. Co-immunoprecipiation of FLAG-tagged proteins with M2 agarose beads (Sigma) was performed according to manufacturer's instructions. Briefly, 1 mg of whole cell lysate was lysed in lysis buffer (50 mM Tris-HCl, 150 mM NaCl, 1 mM EDTA, 1% Triton X-100) and incubated with anti-FLAG M2 agarose beads overnight at 4°C. Beads-bound immunoprecipitates were washed thrice with TBS.
Proteins were eluted from the beads under acidic conditions with 0.1 M glycine, pH 2.3 at room temperature for 5 min with constant shaking. Elutant was neutralized with 1/10 volume of 0.5 M Tris, pH 7.4 with 1.5 M NaCl and analysed with immunoblotting.
Computational methods. The crystal structure of the C/EBPa-DNA complex (known as C/EBPa WT-DNA) was downloaded from the Protein Data Bank (PDB code: 1NWQ) 18 . To generate the coordinates for the acetylation mimetic C/EBPa-K2Q complexed with DNA (K2Q-DNA), K298 and K302 residues were replaced with glutamine using the UCSF Chimera package 45 . The GLN rotamers from the Dunbrack library were used 46 . The acetylated lysine residues (K2Ac) were built manually with UCSF Chimera. The H-N-C ¼ O dihedral angle of all four acetylated lysine residues were set to 0°in the first acetylated model (hereafter termed as K2Ac_a-DNA) based on the most populated structures derived from computations 47 . However, the acetylated lysine residues in recently published crystal structures were found to have the H-N-C ¼ O dihedral angles with both 0°a nd 180° (Fig. 3b) [48][49][50] . Thus, the H-N-C ¼ O dihedral angle of all four acetylated lysine residues were set to 180°in the second acetylated model (K2Ac_b-DNA). The backbone atoms of the four models showed similar fluctuations during the 3.2 ns long MD simulations ( Supplementary Fig. 3). However, longer MD trajectories are needed to meaningfully discuss the stability of the four models. The scope of the current computational study is to gain insight into how different mutations affect protein-DNA-binding energy and protein-DNA hydrogen bonding. For this purpose, the relatively short MD trajectories presented here should be sufficient 51 . The ff03 molecular mechanics (MM) force field was assigned for all residues 52 . The acetylation of partial charges derived for use with the ff03 force field were taken from literature with corresponding atom names and atom types 47 .The force field parameters were set up using the leap programme of the AmberTools13 package 53 . Na þ counterions were added to neutralize the protein-DNA model systems. The crystallographic water molecules were retained in all four model systems. Each protein-DNA model complex was placed into a pre-equilibrated TIP3P truncated octahedral water box 54 . The crystallographic water molecules were included in the TIP3P solvent boxes which extended at least 12 Å from the protein-DNA complex.
MM minimizations and MD simulations were carried out with the sander programme of the AMBER12 package 53 .The MD simulations were preceded by three minimization steps: (1) solute atoms were held fixed while solvent molecules were relaxed (1000 steps of steepest descent and 1,000 steps of conjugate gradient minimization); (2) solvent atoms were held fixed while solute molecules were relaxed (1,000 steps of steepest descent and 1,000 steps of conjugate gradient minimization); (3) both solvent and solute were optimized without any restraints (1,000 steps of steepest descent and 4000 steps of conjugate gradient minimization). After minimization, a two-step MD simulation was carried out on all systems with a 2 kcal mol À 1 Å 2 restraint on the solute atoms: (1) each system was heated from 0 to 300 K over a period of 50 ps in the canonical (NVT) ensemble; (2) each system was relaxed for 50 ps in the isothermic-isobaric (NPT) ensemble (T ¼ 300 K and P ¼ 1 atm). Finally, production MD simulations were run for 3.2 ns in the NPT ensemble (T ¼ 300 K and P ¼ 1 atm). Force calculations were performed with periodic boundary conditions and a 9 Å cutoff was used for non-bonded interactions. Covalent bonds connecting hydrogen atoms were constrained with SHAKE 55 . A collision frequency of 3 ps -1 was used in the Langevin thermostat to maintain the system temperature 56 . A timestep of 2 fs was used in all MD simulations. Atom coordinates were saved every 10 ps during the MD simulations (100 frames ns À 1 ). The random number generator seed was altered at every restart of the MD simulations.
The polar desolvation free energy was calculated with the MMPBSA.py Generation of stable K562 cell lines. To generate WT or mutant C/EBPa lines, K562 cells were stably transfected by electroporation with the respective pBabe-C/EBPa-ER fusion constructs. Briefly, 2 Â 10 6 cells were electroporated in an Amaxa apparatus with 5 mg of Sca-1 linearized plasmid. At 48 h posttransfection, cells were subjected to selection with 0.5 mg ml À 1 puromycin and 50 cells were seeded per 96-well plate. C/EBPa-ER nuclear translocation was induced by addition of 5 mM of b-estradiol (Sigma, E2758). Cells were assessed 4 days post induction for differentiation by CD11b surface marker expression, morphology and neutrophilic enzymatic activity 4 . Multiple individual clones for each type of mutant were picked and analysed by western blot to confirm expression of the fusion protein. The results from two representative clones are presented in this study.
Assessment of granulocytic differentiation. Nitroblue tetrazolium analysis with 2.5 Â 10 5 cells was done as previously described 4 . The percentage of nitroblue tetrazolium-positive cells was quantified by counting at least 100 cells for each mutant under brightfield microscopy (Nikon). Cell morphology was assessed by Wright-Giemsa staining. Cytospin preparation of 2.5 Â 10 5 cells were done as previously described 60 . For staining of CD11b surface marker, 1 Â 10 6 cells were stained with CD11b antibody (Supplementary Table 6) and analysed using LSRII flow cytometry (BD Biosciences). The resulting data were processed with FlowJo Version 7.6.4 software (TreeStar).
Retroviral transduction of Cebpa D/D LSK cells. Animal work was done at NUS with approval from IACUC. C/EBPa conditional knockout (Cebpa fl/fl ) and Mx1-Cre mice have been described previously 2 .
Chromatin immunoprecipitation. ChIP was performed with 2 mg anti-ER antibodies on K562 C/EBPa-ER cells (2 Â 10 7 cells). Briefly, cells were induced with 5 mM b-estradiol for 45 min, fixed in 1% formaldehyde for 5 min at room temperature and subsequently quenched with addition of glycine to a final concentration of 200 mM. Cells were lysed with Buffer L1 (50 mM Tris pH 8.0, 2 mM EDTA, 0.1% NP-40, 10% glycerol) for 5 min on ice, centrifuged for 10 min at 850 Â g. The nuclear pellet was resuspended in Buffer L2 (50 mM Tris pH 8.0, 5 mM EDTA, 1% SDS). Crosslinked chromatin was sonicated using Bioruptor (Diagenode) at 'high' amplitude, 12 s on/30 s off for 16 cycles, and centrifuged for 10 min at 17,000g. Sonicated material was diluted with 9 volumes of Buffer DB (50 mM Tris, 200 mM NaCl, 5 mM EDTA, 0.5% NP-40) and pre-cleared with salmon sperm DNA (Sigma), rabbit IgG (Santa Cruz) and Dynabeads Protein A (Invitrogen). Pre-cleared chromatin was immunoprecipitated overnight with ERa (HC-20) antibody, followed by addition of Dynabeads Protein A for 30 min at 4°C. Beads were washed thrice with washing buffer (WB) (20 mM Tris pH 8.0, 500 mM NaCl, 2 mM EDTA, 0.1% SDS and 1% NP-40), once with WB plus 500 mM LiCl, thrice with buffer TE. Bound chromatin was eluted from the beads with 100 ml of elution buffer (1x TE, 2% SDS). After reverse-crosslinking with RNase A and proteinase K digestion, DNA was purified by Qiaquick PCR purification kit and eluted with 50 ml of TE buffer. Primers used are indicated in Supplementary  Table 7. qPCR was carried out with GoTaq qPCR master mix (Promega) and analysed with a Corbett Rotorgene system. Relative enrichment is calculated over the inactive involucrin locus.
Dimerization assay. 293T cells were co-transfected with 10 mg of V5-tagged WT or mutant versions of C/EBPa and 10 mg of FLAG-tagged WT C/EBPa, cell lysates were collected 24 h after transfection. Anti-FLAG M2 Affinity gel (Sigma) was used to co-immunoprecipitate FLAG-tagged proteins. Samples were kept under constant rotation for 2 h at 4°C. Beads-bound immunoprecipitates were washed thrice with TBS buffer and eluted in SDS-PAGE sample buffer for immunoblotting analysis.
Gene expression analysis. Raw microarray gene expression data (CEL files) for AML and normal BM CD34 þ samples were taken from GSE13159 (ref. 61) and GSE19429 (ref. 62), respectively. We used the RMA model to extract the gene intensity from multiple probe intensities, while the cross-array data normalization was performed using the Cross-Correlation method 63 .