Genetic Interaction between Mutations in c-Myb and the KIX Domains of CBP and p300 Affects Multiple Blood Cell Lineages and Influences Both Gene Activation and Repression

Adult blood cell production or definitive hematopoiesis requires the transcription factor c-Myb. The closely related KAT3 histone acetyltransferases CBP (CREBBP) and p300 (EP300) bind c-Myb through their KIX domains and mice homozygous for a p300 KIX domain mutation exhibit multiple blood defects. Perplexingly, mice homozygous for the same KIX domain mutation in CBP have normal blood. Here we test the hypothesis that the CBP KIX domain contributes subordinately to hematopoiesis via a genetic interaction with c-Myb. We assessed hematopoiesis in mice bearing compound mutations of c-Myb and/or the KIX domains of CBP and p300, and measured the effect of KIX domain mutations on c-Myb-dependent gene expression. We found that in the context of a p300 KIX mutation, the CBP KIX domain mutation affects platelets, B cells, T cells, and red cells. Gene interaction (epistasis) analysis provides mechanistic evidence that blood defects in KIX mutant mice are consistent with reduced c-Myb and KIX interaction. Lastly, we demonstrated that the CBP and p300 KIX domains contribute to both c-Myb-dependent gene activation and repression. Together these results suggest that the KIX domains of CBP, and especially p300, are principal mediators of c-Myb-dependent gene activation and repression that is required for definitive hematopoiesis.


Introduction
The development of all adult hematopoietic stem cell-derived blood cell lineages requires the transcription factor c-Myb, although how it controls hematopoiesis remains unclear [1,2]. Transcription factors regulate target gene activity through interactions with coactivators and corepressors, and the BIOGRID database (thebiogrid.org, [3]) reports more than fifty c-Myb interacting proteins in mice and humans . Although generally thought of as an activator of gene expression, recent findings show that c-Myb can also directly repress target genes [48]; however, details of how c-Myb-interaction partners affect these opposite transcriptional effects, and hematopoiesis, are not well understood.
Perhaps the most widely studied of c-Myb partners, CBP (CREB binding protein, Crebbp) and p300 (E1A binding protein p300, Ep300) reportedly interact with more than 400 other proteins [49]. Together CBP and p300 form the KAT3 family of acetyltransferases that add acetyl groups to lysines in histones and other proteins, although which of these CBP/p300 substrates are critical for gene regulation remains uncertain [50]. Histone acetylation is often associated with gene activation [51] and CBP and p300 are usually described as coactivators of gene expression; however, paradoxically p300 has been implicated in both direct activation and repression of genes by c-Myb [48].
Normal human development requires CBP and p300 [52,53], and mice lacking either CBP or p300 die before birth [54,55]. This lethality has necessitated studies using either conditional knockout alleles, which allow tissue specific gene inactivation or "knock-in" mice in which CBP and p300 proteins are expressed at normal levels but only specific protein interaction domains are mutated (e.g. CH 1, KIX, Figure 1A) [56][57][58]. The use of domain-specific knock-ins is especially useful in limiting the effect of CBP and p300 mutations to a few specific CBP/ p300-interacting proteins out of the hundreds that have been reported [49].
We previously designed a triple point mutation (Y631A, A635Q and Y639A) in the KIX domain of p300 that alters the  [91]. Nuclear receptor interaction domain (NRID), the CREB-binding domain (KIX), Cys/His-rich region-1 (CH1 [57] or transcriptional-adaptor zincfinger-1, TAZ1), bromodomain (Br), plant homeodomain (PHD), HAT enzymatic region, zinc-binding domain (ZZ), TAZ2 (aka CH3 [92]), and the nuclear receptor binding domain (NCBD). binding surface for c-Myb and CREB, but which should maintain the secondary and tertiary structure of the domain ( Figure 1B) [56]. In this way, these point mutations should specifically block proteins from binding that particular surface of the KIX domain, and not interfere with other surfaces of KIX or other domains of CBP/p300 and their protein interactions. To date, only CREB and c-Myb interactions have been shown to be affected by this mutation [56,59]. We showed previously that mice homozygous for this triple point mutation in the p300 KIX domain exhibit multi-lineage defects in hematopoiesis including severe anemia, B and T cell deficiencies, abnormal megakaryocytes and elevated platelet counts (Table 1) [56]. While blood from mice heterozygous for either a c-Myb null allele or the p300 KIX triple point mutation appears normal, combining these mutations in the same mouse (p300 +/KIX ;c-Myb +/-) synergistically affects megakaryocyte development and increases platelet counts without obvious changes in other peripheral blood cell numbers (Table 1) [56]. This gene interaction (epistasis) experiment highlights the critical nature of the genetic interaction between c-Myb and the p300 KIX domain for normal megakaryocyte development in vivo. It remains unclear, however, whether other blood cell lineages also require the genetic interaction of c-Myb with the p300 or CBP KIX domain.
Although the sequence of the KIX domain of CBP is 90% identical to the p300 KIX domain, hematopoiesis in mice homozygous for a comparable mutation in the KIX domain of CBP (Y650A, A654Q and Y658A, Figure 1B) appeared essentially normal [56]. This disparity may be due to differences in protein expression (i.e. CBP and p300 have the same functions in blood cells) or biochemical properties (i.e. p300 has a different and more critical function in blood cells than CBP). While the KIX domain of p300 seems to be more critical in hematopoiesis than the KIX domain of CBP, multiple studies have established that both holo-CBP and -p300 have important roles in normal blood cell production. Some of these roles seem to be unique to either CBP or p300; for example, analyses of aged mice revealed an increased prevalence of hematopoietic tumors and abnormalities affecting the B cell and myeloid lineages in mice heterozygous for a null allele of CBP, although no such defects were found in p300 null heterozygotes [60]. Similarly, CBP null, but not p300 null hematopoietic stem cells lose self renewal capacity, while p300 null, but not CBP null embryonic stem cells fail in normal blood cell differentiation in vitro [61]. Lineage-specific conditional knockout of CBP and p300 in specific lymphoid compartments have demonstrated that while CBP and p300 are each modestly important for B cells [62], CBP has a unique role in demarcating the development of innate and conventional T cells [63,64]. Importantly, Kimbrel et al. showed that three copies of CBP, the two endogenous alleles plus a CBP cDNA expressed from the p300 promoter, could rescue hematopoiesis in the absence of p300 protein [65]. This suggests that protein expression levels, rather than biochemical properties, may account for the distinct effects of mutating the KIX domains of CBP and p300 on hematopoiesis.
In this study we sought to answer four questions: 1) Does the KIX domain of CBP have a role in hematopoiesis? 2) Is genetic interaction between c-Myb and CBP/p300 KIX necessary for the normal production of other blood cell lineages besides megakaryocytes? 3) Does genetic interaction between c-Myb and KIX largely explain the phenotype of c-Myb and KIX mutant mice? and 4) Is the genetic interaction between c-Myb and KIX necessary for both activation and repression of c-Mybdependent genes? To answer these questions, we assessed blood cell production in mice having compound mutations in c-Myb and/or the KIX domains of CBP and p300 and quantified the effect of KIX domain mutations on c-Myb-dependent changes in gene expression. Our findings suggest that through their KIX domains, CBP, and particularly p300, are the main effectors of the c-Myb-dependent gene activation and repression required for normal hematopoiesis. chimeric mice showed that c-Myb-/-cells did not contribute to megakaryocytes in adult mice [82]. b The c-Myb flox allele is a hypomorph resulting in a reduced c-Myb protein level. c-Myb -/flox mice have 5-10% of normal c-Myb protein expression [83].
c The ENU induced Plt4 mutation equates to c-Myb V384D [90]. d Total lymphocyte count from peripheral blood given, but not split out into B and T cells [90]. e The ENU induced Plt6 mutation equates to p300 Y630N in the KIX domain [81].

Animals
Animal experiments were approved by the St Jude Institutional Animal Care and Use Committee and performed in accordance with IACUC guidelines. Generation of the p300 and CBP KIX mutant alleles (Mouse Genome Informatics IDs 3578128 and 3578129) was described in Kasper et al. [56]. Note that the numbering of the p300 KIX mutations Tyr631Ala, Ala635Gln, and Tyr639Ala now coincide with more recent mouse p300 protein sequences (NCBI NP_808489.4 and Ensembl ENSMUSP00000066789) and are one position different than those originally listed in 2002. The c-Myb knockout allele (MGI ID 2662859) was described in Mucenski et al. [1]. Unless otherwise noted, mice analyzed in this study were C57BL/6J x 129Sv F1 hybrids generated from multiplybackcrossed congenic lines.

Plasmids and transient transfection assays
The Gal-c-Myb plasmid (aa 186 to 325) was described in Parker et al. [66]. The GST-c-Myb 186-325 plasmid was constructed in pGEX 4T-3 (GE Life Sciences) using an EcoRI/ SmaI fragment from Gal-c-Myb containing the c-Myb aa186 to 325 insert. Transient transfection assays using MEFs were performed as previously described using the Promega Dual Luciferase Reporter Assay Kit where reporter gene luciferase activity is normalized to cotransfected Renilla luciferase reporter activity [59]. MSCV c-Myb IRES GFP was a gift of Angelika Hoffmeyer.

Flow cytometry, FACS and hematopoietic analysis
Flow cytometry was performed on BD Biosciences FACS Calibur and FACS LSR instruments. FACS of CD4 + CD8 + double positive thymocytes was performed on a BD Biosciences Aria. All antibodies were from Becton Dickenson, except the IL18R antibody was from R&D Systems. Complete blood counts were generated using a Hemavet Hematology System (Drew Scientific). Hematocrits were determined manually.

Statistical method for calculating genetic interaction (epistasis)
As defined by Jasnos et al., epistasis (E) was considered absent if the multiplicative fitness phenotypes of the extreme genotypes equaled that of the intermediate genotypes [67]; i.e. (WT0 * TH3) = (CM1*CP2), where WT0 = wild type phenotype, CM1= c-Myb +/-, CP2 = CBP +/KIX ;p300 +/KIX , and TH3 = c-Myb +/-;CBP +/KIX ;p300 +/KIX . Blood cell counts were log transformed to stabilize their variance, which coverts the formula to E = (WT0 + TH3) -(CM1 + CP2). There is no epistasis if E = 0, whereas E > 0 defines positive epistasis (higher than expected phenotype), and E < 0 signifies negative epistasis (lower or worse than expected phenotype) [68]. Because the machinederived cell count data were obtained from independent experiments performed on different dates, the experiments were treated as a random effect in our model to account for machine drift over time. Thus, a mixed model was used (SAS 9.1.3 software). Since no significant strain effect was detected between the C57BL/6J and C57BL/6J x 129Sv F1 mice used in this analysis, their data were combined to increase the power of the analysis. Mice used were one to two months old; the effect of age in this cohort was not significant and is excluded from the model. The mean and 95% confidence intervals were calculated for each cell type tested, and no multiple comparison adjustment was performed for this analysis.

Mouse embryonic fibroblasts (MEFs) and retroviral infection
Primary MEFs were generated as described from e14.5 C57BL/6J x 129Sv F1 embryos and were used between passage three and six [69]. Retrovirus was generated from MSCV-IRES-GFP (GFP control virus) and MSCV-c-Myb-IRES-GFP plasmids. MEFs were infected with retrovirus 6 days before RNA harvest.

Both CBP and p300 KIX contribute to c-Myb function in vitro
Although the KIX domain of CBP can directly bind the c-Myb transactivation domain in vitro, and the interaction is sensitive to our KIX mutation ( Figure 1C, [20,56]), only mutation of the p300 KIX domain has previously been shown to affect hematopoiesis in vivo [56]. We also showed that c-Myb transactivation function in both p300 KIX/KIX and CBP KIX/KIX mouse embryonic fibroblasts (MEFs) is reduced, demonstrating that both CBP and p300 contribute to c-Myb activity in an in vitro assay [56]. While a single mutant KIX allele of either CBP or p300 produced no effect on transactivation by a Gal-c-Myb construct, we found that increasing the number of mutant KIX alleles of CBP and p300 in MEFs from two (CBP KIX/KIX , p300 KIX/KIX or p300 +/KIX ;CBP +/KIX ) to three (p300 +/KIX ;CBP KIX/KIX or p300 KIX/KIX ;CBP +/KIX ) further decreased the activity ( Figure 1D). This led us to predict that an additional KIX mutant allele of CBP or p300 would worsen the in vivo c-Myb-dependent phenotype of a p300 KIX/KIX or CBP KIX/KIX mouse, and could reveal a role for the CBP KIX domain in hematopoiesis. To this end, we generated mice bearing mutations in both p300 KIX and CBP KIX and examined the effect on blood cell production.

;CBP KIX/KIX embryos are grossly normal but display defective blood cell formation
While yolk sac-derived primitive (embryonic) hematopoiesis is c-Myb-independent [2], c-Myb-dependent definitive (adult) hematopoiesis initially occurs in the fetal liver. We first looked at the impact of combining p300 KIX and CBP KIX mutant alleles in e15.5 fetal liver, since defective hematopoiesis, including decreased red cell production and increased numbers of megakaryocytes (the cell type that produces platelets) is evident at this stage in p300 KIX/KIX embryos (Figure 2A, B, E and G). We found that adding one CBP KIX mutant allele (p300 KIX/ KIX ;CBP +/KIX ) produced embryos that were grossly normal but had pale fetal livers ( Figure 2F) with reduced cell counts ( Figure 2G) and dramatically reduced red cell production rendering adult survival unlikely ( Figure 2C). The p300 KIX/ KIX ;CBP +/KIX fetal livers had a pallor similar in appearance to that of c-Myb -/-embryos ( Figure 2F), and cell counts in p300 KIX/ KIX ;CBP +/KIX and c-Myb -/-fetal livers were similarly deficient ( Figure 2G and H), suggesting that the major effect of simultaneously mutating the p300 and CBP KIX domains is on c-Myb function. At the cellular level, p300 KIX/KIX ;CBP +/KIX fetal livers also contained even more cells positive for the megakaryocyte marker CD41 than p300 KIX/KIX fetal livers ( Figure  2E). By contrast, embryos with one wild type p300 allele (p300 +/KIX ;CBP KIX/KIX ) appeared to have only a mild deficit in fetal liver erythropoiesis ( Figure 2D, 2F), and their livers contained normal numbers of megakaryocytes ( Figure 2E).

Multiple blood cell lineages require the KIX domains of CBP and p300
Only about 40% of p300 +/KIX ;CBP KIX/KIX (triple-KIX) mice reached weaning age; however, the surviving mice appeared grossly normal (SLL, LHK data not shown). We examined the peripheral blood of 1-4 month old wild type (WT), triple-KIX p300 +/KIX ;CBP KIX/KIX and intermediate genotype control mice (p300 +/KIX single heterozygous, CBP KIX/KIX homozygous, and p300 +/KIX ;CBP +/KIX compound heterozygous) by flow cytometry and complete blood count to determine the effect of mutant KIX alleles on cell numbers. We found that in triple-KIX p300 +/ KIX ;CBP KIX/KIX mice, both platelet ( Figure 3A) and reticulocyte (immature red cell, Figure 3B) counts were significantly elevated above all control mice. By contrast, triple-KIX p300 +/ KIX ;CBP KIX/KIX mice displayed significantly lower hematocrits (percentage of red cells by volume, Figure 3C) and B cell numbers ( Figure 3D) compared with all controls. Counts of CD4 + and CD8 + T cells in peripheral blood showed that these cell types were significantly diminished in triple-KIX p300 +/ KIX ;CBP KIX/KIX mice compared with controls except for p300 +/ KIX ;CBP +/KIX compound heterozygous mice, which showed the same trend, but did not achieve statistical significance ( Figure  3E,F; CD4 + T cells were also moderately reduced in CBP KIX/KIX mice). We also measured two myeloid lineage cells; neutrophil numbers showed a modest increase in triple-KIX p300 +/ KIX ;CBP KIX/KIX mice compared with wild type, but not the other intermediate genotypes ( Figure 3G), and monocyte numbers were not significantly affected ( Figure 3H, p=0.36). These data confirm that in the context of a single KIX mutant allele of p300, the KIX domain of CBP is critical for the production of normal numbers of B and T cells, red cells and platelets. Peripheral blood counts from 1-4 month old C57BL/6J x 129Sv F1 wild type (WT) and KIX mutant mice (N=10-22). Counts from automated Hemavet complete blood count. For B cells and T cells, total lymphocyte counts from complete blood count were parsed using flow cytometry to determine the percentage of lymphocytes that were positive for B220 (B cells), CD3 (T cells), CD4 and CD8 (T cell subsets). Asterisks indicate significant p value by pairwise Tukey post test following one way ANOVA (* p<0.05, ** p<0.01, *** p<0.001).

consistent with KIX and c-Myb acting in the same pathway
We showed previously [56] using the genetic epistasis method of complex haploinsufficient interaction analysis [73] that mice compound heterozygous for p300 KIX and a c-Myb null allele exhibit abnormal megakaryocytes, increased platelet counts and reduced thymocyte numbers. These blood defects are not seen in p300 +/KIX or c-Myb +/-single heterozygotes [56]. Consistent with the physical binding of c-Myb to the KIX domain [20,66], this genetic experiment provided in vivo evidence that a reduced interaction between p300 KIX and c-Myb is responsible for these aspects of the blood defect in p300 KIX/KIX mice (Table 1, p300 KIX/KIX mice display both increased platelet counts and decreased thymocyte numbers compared with wild type littermates) [56]. However, it remained unclear from that experiment whether other blood cell defects seen in p300 KIX/KIX mice, such as anemia and decreased B cell counts, depend on a genetic interaction between c-Myb and p300. Based on the phenotypic overlap between p300 +/ KIX ;CBP KIX/KIX triple-KIX, p300 KIX/KIX and various c-Myb mutant mice (Table 1), we hypothesized that additional blood lineages require the interaction of the KIX domain and c-Myb, and that CBP, as well as p300, participates in a genetic interaction with c-Myb.
We next determined whether a complex haploinsufficiency interaction between CBP KIX , p300 KIX and c-Myb null mutant alleles existed by using a statistical model to calculate epistasis [67]. In this model, epistasis is present when the effect of combined mutations (in this case the triple-het p300 +/KIX ;CBP +/ KIX ;c-Myb +/-) is greater than would be predicted from the phenotype of the intermediate genotypes (p300 +/KIX ;CBP +/KIX and c-Myb +/-). Positive epistasis (a higher than expected phenotype compared to intermediate controls) and negative epistasis (lower than expected phenotype) are represented by positive and negative values respectively; a value of zero indicates no epistasis [68]. Platelet counts in triple-het p300 +/ KIX ;CBP +/KIX ;c-Myb +/-mice showed the effect of positive epistasis (0.41 ± 0.15, mean epistasis value ± 95% confidence interval, Figure 4H), while B cell and both CD4 + and CD8 + T cell counts displayed negative epistasis (-1.35 ± 0.50, -1.08 ± 0.60 and -1.13 ± 0.50, Figure 4H). For hematocrit, the negative epistasis value was small; however, the 95% confidence interval did not overlap zero (-0.11 ± 0.08, Figure 4H). Neutrophils and monocytes in triple-het p300 +/KIX ;CBP +/KIX ;c-Myb +/-mice demonstrated no epistatic effects (-0.013 ± 0.61 and 0.16 ± 0.52, Figure 4H). Overall, the analysis indicates that a genetic interaction between c-Myb, CBP and p300 is limiting for the maintenance of normal numbers of platelets and lymphocytes. The impact of this triple mutation combination on hematocrit was less robust, but there is strong homeostatic pressure to maintain red cell production, so even a small variance in hematocrit may be significant. In fact, increased reticulocyte (immature red cell) counts seen in triple-het p300 +/KIX ;CBP +/ KIX ;c-Myb +/-mice show that red cell production is stressed and support the assertion that compensatory mechanisms are at work ( Figure S1).

Both c-Myb activated and repressed transcription requires the KIX domains of CBP and p300
So far we have shown that KIX and c-Myb mutant mouse phenotypes overlap, and that p300 KIX , CBP KIX , and c-Myb null mutant alleles can synergize. We next investigated whether c-Myb-dependent gene expression requires the KIX domains of CBP and p300. For our initial studies we chose mouse embryonic fibroblasts (MEFs), over hematopoietic cells for three reasons: 1) sufficient numbers of phenotypically uniform primary MEFs with a p300 KIX/KIX ;CBP +/KIX genotype can be obtained; 2) p300 KIX/KIX ;CBP +/KIX MEFs are phenotypically similar to wild-type cells and do not have broad transcriptional defects (see blue probe sets in Figure 5A and B); and 3) MEFs possess little or no endogenous c-Myb allowing us to control c-Myb expression by retroviral transduction ( Figure S2). In this way we could more confidently identify Myb-responsive genes, and eliminate those genes that might depend on KIX via another transcription factor or are altered by developmental defects. Despite the lack of endogenous c-Myb in MEFs, the expression of many genes increased or decreased more than twofold in response to exogenous c-Myb (black probesets in Figure 5), indicating that necessary cofactors for c-Myb activity are present in these cells. Analysis of Affymetrix microarray probe sets induced at least twofold by exogenous c-Myb in wild type MEFs showed that 382 of 509 probe set signals (75%) were at least 1.5 fold higher in wild type + c-Myb MEFs compared with p300 KIX/KIX ;CBP +/KIX + c-Myb MEFs ( Figure 5A, probe sets in black and Table S1). Interestingly, of probe sets repressed at least 50% by c-Myb expression in wild type MEFs, 91 of 148 (61%) were at least 1.5 fold higher in p300 KIX/KIX ;CBP +/KIX + c-Myb MEFs than wild type + c-Myb MEFs ( Figure 5B, probe sets in black and Table S1). Of the 446 genes uniquely identified with probe sets induced or repressed at least twofold by exogenous c-Myb in MEFs, 173 (39%) have previously been shown to recruit c-Myb in their vicinity by ChIP-seq [48]. However, only 26 of these 173 genes were similarly activated or repressed by c-Myb in both MEFs and the ER-inducible c-Myb expressing mouse myeloid precursor cells used in the ChIP-seq study [48] (Table S1) indicating that c-Myb regulation of target genes is highly dependent on cell type. Nonetheless, these data indicate that the KIX domains of CBP and p300 contribute to both c-Myb-dependent activation and repression of transcription.
Next, we sought to verify that the KIX domain has a similar role in endogenous c-Myb-dependent gene activation and repression using a hematopoietic cell type that normally expresses c-Myb. To do this, we utilized data from a study by Yuan et al. that identified genes regulated by T cell specific inactivation of c-Myb in CD4 + CD8 + double positive (DP) thymocytes (c-Myb flox/flox ;CD4-Cre;Tcrα -/-mice compared with c-Myb +/flox ;CD4-Cre;Tcrα -/-) [70]. We compared these c-Myb activated ( Figure 6A, probe sets marked in black, see also Table S2) and repressed ( Figure 6B, probe sets marked in black, see also Table S2) gene data sets to Affymetrix microarray analysis of CD4 + CD8 + DP thymocytes FACS purified from adult wild type (WT) and triple-KIX p300 +/ KIX ;CBP KIX/KIX mice. In general, DP thymocyte gene expression appeared similar between WT and triple-KIX p300 +/KIX ;CBP KIX/KIX mice (see blue probe sets in Figure 6A and B); however, when we looked specifically at c-Myb-dependent gene expression (black probe sets in Figure 6A, B and Table S2), the result resembled the situation we observed in c-Myb expressing MEFs ( Figure 5). Most c-Myb activated genes defined in c-Myb null DP thymocytes decreased in triple-KIX p300 +/KIX ;CBP KIX/KIX DP thymocytes ( Figure 6A, black probe sets, Table S2), while most c-Myb repressed genes increased ( Figure 6B, black probe sets, Table S2), demonstrating that both activation and repression of many c-Myb-dependent genes require the KIX domain.

Both c-Myb activated and repressed gene sets are significantly enriched among genes differentially expressed between triple-KIX p300 +/KIX ;CBP KIX/KIX and wild type CD4 + CD8 + double positive thymocytes
To assess the statistical significance of this correlation, we included the c-Myb-dependent gene set generated from the c-Myb null CD4 + CD8 + double positive (DP) thymocyte Affymetrix data set of Yuan et al. in a Gene Set Enrichment Analysis (GSEA) of our wild type and triple-KIX p300 +/KIX ;CBP KIX/KIX DP thymocyte Affymetrix data [70,75]. GSEA is a computational method used to measure the coordinate regulation of sets of genes defined by biological criteria (e.g. activated or repressed in response to a particular stimulus, dependent on a specific transcription factor or expressed in a certain cell type) even if the changes to individual genes are small [75,76]. Amongst a diverse group of 79 other gene sets we analyzed by GSEA, the two c-Myb-dependent gene sets derived from the data of Yuan  Figure 7A-D, false discovery rate (FDR) q = 0.0, family-wise error rate (FWER) p = 0.0] and followed the expected trend (i.e. wild type DP thymocytes expressed c-Myb activated genes more highly than p300 +/KIX ;CBP KIX/KIX , and c-Myb repressed genes were expressed lower in wild type than p300 +/ KIX ;CBP KIX/KIX DP thymocytes). Of the 283 genes uniquely identified with probe sets induced or repressed at least twofold by c-Myb in Yuan et al., 106 (38%) were shown to recruit c-Myb in their vicinity by ChIP-seq using ER-inducible c-Myb expressing myeloid precursor cells [48]. However, for only 20 of these 106 genes was the direction of regulation by c-Myb the same in both DP thymocytes and the myeloid cells used in the ChIP-seq experiment [48,70] (Table S2). While the agreement in DP thymocyte gene expression between KIX dependence in our study and c-Myb gene dependence in the study of Yuan et al. indicates that an intact KIX domain is required for much of c-Myb dependent gene expression, it appears that c-Myb target genes differ substantially between cell types.
Together, the data suggests that c-Myb and KIX mutations, individually or when combined, result in similar phenotypes and gene expression patterns when considering data from the same cell type (i.e. DP thymocytes). This further supports the idea that c-Myb and CBP/p300 KIX genetically interact in a common pathway to affect many c-Myb-dependent genes and biological processes. Zhao et al. have shown by chromatin immunoprecipitation that p300 is directly recruited to c-Myb activated and repressed target genes [48], but whether the genes affected in this study are also direct targets of c-Myb and CBP/p300 KIX remains a question for future studies.

Multiple blood cell lineages and both c-Myb-dependent gene activation and repression require the genetic interaction between c-Myb and the KIX domains of CBP and p300
We show here for the first time that the KIX domain of CBP contributes to the normal production of multiple blood cell lineages, including red cells, B and T cells and megakaryocytes/platelets (Figures 2 and 3). We also found that genetic interaction between c-Myb and both CBP and p300 controls megakaryocyte/platelet, B and T cell and likely red cell production ( Figure 4). The genetics are consistent with what is known of the physical and functional interaction between c-Myb and the KIX domain, which surprisingly influences both gene activation and repression ( Figure 5-8) [20,56,66]. Although none of the KIX mutant genotypes examined displayed myeloid defects, this may be consistent with findings that certain myeloid cells are c-Myb-independent [74,78]. Together, our findings suggest that the c-Myb interaction with the KIX domain of CBP, and especially p300, is a major pathway contributing to the other hematopoietic lineages.

The KIX domains of both CBP and p300 participate in the normal formation of multiple blood cell lineages
In our original study, the lack of a blood phenotype in CBP KIX/KIX mice compared with that seen in p300 KIX/KIX mice surprised us [56]. We hypothesized that this difference was due to the relative amounts of CBP and p300 protein in a critical cell type rather than functional differences between CBP and p300 [56]. The findings of this study support this hypothesis. Triple-KIX p300 +/KIX ;CBP KIX/KIX mice display multiple blood lineage defects, while p300 +/KIX and CBP KIX/KIX mice have normal blood ( Figure 3); therefore, in this context, mutation of both CBP and p300 KIX domains is necessary to produce a blood phenotype. Our results agree with the finding that in the absence of endogenous p300 protein, CBP produced from a CBP cDNA driven by the p300 promoter can rescue hematopoiesis [65]. In this regard, p300 KIX is more important than CBP KIX for hematopoiesis in mice, but in other organisms CBP might predominate and it may prove fruitful to look for CBP mutations in human and horse hematopoietic syndromes with phenotypic similarities to the p300 KIX/KIX mice [56,79,80].

Multiple blood cell lineages require the genetic interaction between c-Myb and the KIX domains of CBP and p300
Several studies report phenotypic similarities between mice with mutations in the p300 and c-Myb domains through which they physically interact [25,56,81]. The complex haploinsufficiency analysis we performed previously [56] and here, tests for epistasis (gene interaction), which is a method to examine whether c-Myb and KIX might interact in vivo. Indeed, triple-het p300 +/KIX ;CBP +/KIX ;c-Myb +/-mice showed evidence of epistasis, supporting the c-Myb:KIX common pathway hypothesis, most clearly for the lymphoid and megakaryocyte lineages, but also for the tightly-regulated red cell population (Figure 4 and Figure S1).
All of the blood cell lineages affected in p300 KIX/KIX and triple-KIX p300 +/KIX ;CBP KIX/KIX mice show evidence of epistasis in triple-het p300 +/KIX ;CBP +/KIX ;c-Myb +/-mice compared to intermediate genotype controls (Figure 4). This suggests that all blood cell defects caused by mutation of the KIX domain (anemia, increased platelet count and deficiency of peripheral B and T cells) could be explained by decreased physical interaction of the KIX domain with c-Myb. It further suggests that these lineages are not greatly influenced by effects on other KIX-binding factors like CREB [59].
It remains unclear what role the KIX domains of CBP and p300 play in the c-Myb-dependent production of myeloid cells ( Figure 4F,G and H). Previous studies have shown that the total absence of c-Myb does not allow the development of any adult blood cell lineages [1,82]; however, partial reductions in c-Myb protein level or diminutions of the KIX:c-Myb interaction affect blood cell lineages nonuniformly. Low c-Myb protein expression in mice (5-10% of normal) decimates B and T cell numbers and produces fetal liver cells that form few or no red cell or granulocytic myeloid (including neutrophil) lineage colonies, but generate more monocytic myeloid and megakaryocyte lineage colonies than wild type controls (Table  1) [83]. Similarly, mice homozygous for a single amino acid substitution (M303V) in the KIX-binding domain of c-Myb, which results in decreased c-Myb transactivation potential, display decreased lymphocyte counts and hematocrit, elevated numbers of platelets, but normal monocyte and neutrophil numbers (Table 1) [25]. These data suggest a hierarchy: partial c-Myb function impairs B, T and red cell production the most, sometimes spares myeloid cell types, and favors the megakaryocyte/platelet lineage. Indeed, none of the combinations of KIX and c-Myb mutant alleles utilized in this study (p300 KIX/KIX , triple-KIX p300 +/KIX ;CBP KIX/KIX or triple-het p300 +/KIX ;CBP +/KIX ;c-Myb +/-) displayed a deficit in the myeloid lineage cells (neutrophils and monocyte/macrophages) ( Table  1). It is unclear whether normal production of the myeloid lineages requires less wild type CBP/p300 protein than other blood cell lineages (both triple-KIX p300 +/KIX ;CBP KIX/KIX and triple-het p300 +/KIX ;CBP +/KIX ;c-Myb +/-mice have some wild type CBP and/or p300 protein), if c-Myb interacts with CBP/p300 via domains other than KIX or utilizes other cofactors, or if the remaining myeloid cells are c-Myb-independent [74,78].

The KIX domains of CBP and p300 contribute to both c-Myb-dependent activation and repression of transcription
Our data from both c-Myb-expressing mouse embryonic fibroblasts ( Figure 5) and CD4 + CD8 + double positive thymocytes ( Figure 6) show that much of c-Myb-dependent gene expression depends on the KIX domain. Since in both of our model systems one wild type allele of either CBP or p300 remains, we could not determine whether the unaffected minority of c-Myb-dependent genes are truly KIX-independent or if they receive sufficient coactivation from the remaining wild type CBP or p300.
Our observation that c-Myb repressed many genes in a KIXdependent manner (Figures 5B and 6B) was initially unexpected. However, Zhao et al. recently showed using an estrogen inducible ER-c-Myb fusion expressing myeloid cell line that c-Myb and p300 are directly recruited to both c-Myb activated and repressed target genes [48]. They also compared bone marrow cells transduced with a retrovirus expressing wild type or L302A c-Myb (the L302A mutation ablates the interaction of c-Myb with the KIX domain [66]) and showed that most direct c-Myb activated (15 of 19) and repressed (29 of 34) genes tested were sensitive to L302A [48]. Analysis of granuloid/myeloid progenitor (GMP) cells, which express endogenous c-Myb, from wild type and Myb E308G/E308G mutant mice (the E308G mutation decreases c-Myb interaction with the KIX domain) [84] showed a similar mutation sensitivity for both c-Myb activated and repressed gene expression [48]. Our study did not address whether the c-Myb-and CBP/p300 KIXdependent genes we observed were direct or indirect targets of these factors. However, our expression data nicely complement the study of Zhao et al. and demonstrate a clear role for the KIX domain in both c-Myb activated and repressed gene sets (Figures 5-7).

c-Myb may either activate or repress genes in a context dependent manner
In comparing c-Myb-dependent genes from different studies, we were struck by how little overlap exists between the datasets [48,71,[85][86][87][88]. This has been a challenge for those attempting to define c-Myb target genes and it is unclear how much can be explained by differences in cell types used and study design. Interestingly, recent attention to c-Myb repressed genes may offer a clue in this mystery. During GSEA analysis, we found that a gene expression dataset derived from analysis of wild type and Myb Plt4/Plt4 Lin -Sca-1 + c-kit + (LSK) hematopoietic precursor cells [71] was highly enriched in our analysis of wild type and p300 +/KIX ;CBP KIX/KIX CD4 + CD8 + double positive (DP) thymocytes ( Figure 9, FDR q = 0.00079, FWER p = 0.0030). Although the Plt4 mutation (V384D) in Myb does not disrupt the interaction with CBP KIX or full-length p300 [89], it produces a similar phenotype to that seen in p300 KIX/KIX mice [90]. Surprisingly, the GSEA correlation we observed is largely reversed; most genes that are repressed in Myb Plt4/Plt4 LSK cells showed increased expression in p300 +/KIX ;CBP KIX/KIX DP thymocytes compared to wild type ( Figure 9). These data suggest that c-Myb in collaboration with CBP and p300 can either repress or activate the same genes based on as yet unknown cellular cues (e.g. LSK cells vs. DP thymocytes).
In all, our data suggest that while the KIX domain of p300 is more critical for murine hematopoiesis, the KIX domain of CBP also participates. We have shown that the functional interaction between c-Myb and the KIX domains of CBP and p300 impacts all of the blood cell lineages dysregulated by CBP/p300 KIX domain mutations; however, it remains unknown whether c-Myb-dependent myeloid lineage production also requires the c-Myb:KIX interaction. Lastly, while CBP and p300 are primarily known for their roles as coactivators, mutation of the KIX domain impairs both the repression and the activation of c-Myb-dependent genes. These findings provide mechanistic insight into how c-Myb, p300, and CBP influence hematopoiesis, and establish the importance of the interaction between CBP/p300 KIX and c-Myb for multiple blood cell lineages. Figure S1.

Supporting Information
Combined KIX and c-Myb insufficiency produces a genetic interaction that affects reticulocyte number. Peripheral blood counts from 3-12 month old C57BL6x129Sv (F1) background mice. Counts from automated Hemavet complete blood count. Asterisks indicate significant p value by pairwise Tukey post test following one way ANOVA (* p<0.05, ** p<0.01, *** p<0.001). p300 +/KIX and p300 +/KIX ;c-Myb +/data were left out of ANOVA and Tukey post test analyses because these genotypes were represented by a single mouse in this experiment. ANOVA p=0.0002. (TIF)  . c-Myb dependent gene set identified in c-Myb null Lin -Sca-1 + c-kit + (LSK) cells is enriched in KIX mutant DP thymocytes, but displays a largely reversed expression profile from expected. Gene set enrichment analysis (GSEA) using Affymetrix gene expression data from wild type (WT) and p300 +/KIX ; CBP KIX/KIX (Tri-KIX) CD4 + CD8 + double positive (DP) thymocytes found significant (FDR q = 0.00079, FWER p = 0.0030) enrichment of a set of genes dependent on c-Myb in LSK cells as defined by [71]. (A) In the enrichment plot, genes are ranked by signal/noise ratio according to their differential expression between WT and Tri-KIX DP thymocytes. Genes in the LSK gene set are marked with vertical bars, and the enrichment score is shown in green. (B) Relative expression of the LSK gene set in WT and Tri-KIX DP thymocytes (N=4 mice each) are shown in order of their signal/noise ratio rank.