The distinct biological implications of Asxl1 mutation and its roles in leukemogenesis revealed by a knock-in mouse model

We used Lin^- bone marrow cells as a surrogate to identify genome-wide histone modification affected by Asxl1 mutations. Chromatin lysate was harvested and sonicated with a sonicator (Bioruptor®Pico) to shear the DNA into a length ~200 bp, then it was hybridized with anti-H3K27me3 (Millipore, Germany). Immunoprecipitated DNA was sent to the National Center for Genome Medicine and sequenced by Illumina HiSeq 2000 sequencer with 100 × 2 bp paired-end sequencing.

Sequencing reads were aligned to the mm10 mouse reference genome by Burrows-Wheeler Alignment tool (BWA; version 0.7.15). We used the Model-based Analysis of ChIP-Seq tool (MACS2) to detect peaks of reads between sample and input sequences in ASXL1 ^tm/+ and wild-type bone marrow cells. ASXL1 ^tm/+ (or wild-type) specific peaks were called by intersecting the identified peaks with BEDTools v2.17.0 [21]. These condition-specific peaks were annotated by Peak Annotation and Visualization (PAVIS) and analyzed for the enrichments in gene regions [22]. To realize the biological functions governed by the interaction of Asxl1 mutation and H3K27me3, we analyzed peaks-associated genes by The Database for Annotation, Visualization and Integrated Discovery (DAVID) v6.8 with default settings [23, 24]. Furthermore, we performed motif analysis on sequences around the condition-specific peaks (±250 bps from peak center) by MEME-ChIP web tool included in the MEME Suite [25, 26]. Sequencing reads and identified peaks were visualized with Integrative Genome Viewer (IGV) [27, 28].

In vitro and in vivo experiments were performed at least three times independently. Data were processed in Microsoft Excel or GraphPad Prism software. Student’s t test, paired t test, ANOVA or chi-square test were used to compare the differences in values between groups.

For the other experimental procedures, please see the Additional file 1.

In human AML, the most common mutation is c.1934dupG; p.G646WfsX12 (up to 66%) [10]. The mouse and human ASXL1 proteins share 74% identity in amino acid sequence. The changed amino acid G646 is within a stretch of highly conserved region ATTAIGGGGGPGGGG (designated as a bold and underlined G) [10]. An additional guanine was inserted into this 8-G cassette located at mouse Asxl1 exon 13 to mimic this frequent human ASXL1 mutation (Fig. 1a). This insertion causes frame shift in the reading frame and introduces a premature stop codon so that the mutant Asxl1 would be shorter than wild-type form and lack the c-terminal region which contains a plant homeodomain (PHD). Therefore, the cognate mouse mutation c.1925dupG; p.G643WfsX12 is expected to bear similar pathophysiological consequence as human’s. In our mouse model, mutant Asxl1 expression was driven by the endogenous Asxl1 promoter. Therefore, mutant Asxl1 would be expressed identically as the endogenous Asxl1, not restrictive to hematopoietic cells (Fig. 1a).

The additional guanine inserted into the 8-G cassette at exon 13 was confirmed by DNA sequencing (Fig. 1b). Both Asxl1 G643WfsX12 heterozygotes and homozygotes were fertile. The pups’ genotypes fit Mendelian ratio through gestation period till birth (Additional file 1: Figure S1). Homozygous new-born mice suffered from high rate of post-natal death, with only 7% of viability after weaning. Autopsy of the dead new born mice did not show any obvious organ abnormalities (data not shown). Lack of nursing was probably the main reason of these post-natal lethal events, but the exact causes remains to be elucidated. Due to the difficulty in gathering sufficient mice for observation, we hence focused our long-term observation on heterozygous and wild-type mice.

We initially performed several in vitro assays to evaluate the population frequency and differentiation potencies of HSPCs in the bone marrow of Asxl1 ^tm/+ and wild-type mice. We first used cobblestone-area forming cell (CAFC) assays to evaluate the frequencies of hematopoietic precursor cells in bone marrows. The cobblestone areas were counted one week after seeding and we found that Asxl1 ^tm/+ cells formed more cobblestone areas than Asxl1-wild cells (N = 3 each, p = 0.028) (Fig. 2a). Colony-forming cell (CFC) assay were also performed to test the effects of Asxl1 mutant on cell differentiation. Serial plating was performed every 7 days to estimate population frequencies of hematopoietic precursors in the initial plating. In initial plating, Asxl1 ^tm/+ and wild-type bone marrow cells formed similar numbers of each type of colonies. However, total colony number as well as granulocyte colonies (CFU-G) formed by Asxl1 ^tm/+ cells were more than wild-type cells in second plating and this trend last to the third plating (Fig. 2b, Additional file 1: Figure S2A to S2C). These results indicate that mutated Asxl1 confers stronger short-term in vitro proliferation capabilities to hematopoietic precursors than the wild-type Asxl1.

To evaluate the in vivo influence of Asxl1 mutation on hematopoiesis in a long-term basis, we employed bone marrow transplantation assays. Five hundred Lin^-c-Kit⁺Sca-1⁺ (LSK) bone marrow cells sorted from Asxl1 mutant and wild-type mice, respectively, together with 200,000 helper cells, were transplanted into wild-type recipient mice for competitive repopulation unit assays. The peripheral blood of the transplanted mice was sampled monthly and evaluated for reconstitution efficiency in a 4-month period. We found that recipient mice transplanted with LSK bone marrow cells from Asxl1 ^tm/+ donors had less donor-derived cells in peripheral blood and marrow when compared to those receiving wild-type LSK bone marrow cells (Fig. 3a). Interestingly, B cells in the peripheral blood of Asxl1 ^tm/+ mice were particularly reduced (Fig. 3c). These data suggest that Asxl1 mutant LSK cells have reduced in vivo long-term repopulation capacities compared with wild-type LSK cells.

Next, we performed serial bone marrow transplantation assays to rigorously test the potency of in vivo self-renewal ability of the Asxl1 ^tm/+ HSPCs. In this setting, whole bone marrow cells were serially transplanted into recipients without helper cells. To evaluate the reconstitution efficiency, peripheral blood of the recipient mice transplanted with either Asxl1 ^tm/+ bone marrow cells or wild-type bone marrow cells were sampled 2 months after every round of transplantation. We found that the frequencies of total cells and T cells, but not B or myeloid cells, in the recipient mice’s peripheral blood derived from Asxl1 ^tm/+ mice declined faster compared with those derived from wild-type controls (Fig. 4). The results suggest that Asxl1 mutation renders a compromised long-term in vivo self-renewal capability in a variety of lineages compared to wild-type cells in vivo.

The amount of HSPCs in the bone marrow of Asxl1 ^tm/+ and wild-type littermates were analyzed and compared by FACS analysis. We noted that bone marrow LSK cells, long-term (Lin^-Sca-1⁺c-Kit⁺CD150⁺CD48^-) and short-term hematopoietic stem cells (Lin^-Sca-1⁺c-Kit⁺CD150⁺CD48⁺), multipotent progenitors (Lin^-Sca-1⁺c-Kit⁺CD150^-CD48⁺), common myeloid progenitors (CMP, Lin^-Sca-1^-c-Kit⁺CD34⁺FcγR^lo), granulocyte-monocytic progenitors (GMP, Lin^-Sca-1^-c-Kit⁺CD34⁺FcγR^hi), and megakaryocyte-erythroid progenitors (MEP, Lin^-Sca-1^-c-Kit⁺CD34^-FcγR^lo) were all not different between the Asxl1 ^tm/+ heterozygotes and the wild-type mice (Additional file 1: Figure S2D). These data suggest that Asxl1 mutation does not affect the amount of hematopoietic cell components in vivo by surface marker analysis, although both in vitro and in vivo experiments indicate presence of its biological activities in Asxl1 ^tm/+ bone marrow cells as shown above.

A cohort of Asxl1 ^tm/+ and wild-type control mice were collected to observe the influence of Asxl1 mutation on overall health status in an 18- to 24-month period (N = 33 for Asxl1 ^tm/+ mice and N = 38 for wild-type controls). Heterozygous mice were significantly lighter than wild-type mice (Fig. 5a). While there were no significant differences in hemograms in the peripheral blood or marrow hematopoietic components in younger mice, there were subtle hematopoietic phenotypes when the mice were old at 18 months. Old male (18-month age) Asxl1 ^tm/+ mice had higher WBC (p = 0.0091) and RBC counts (p = 0.03893) (Fig. 5b). There were more T cells in bone marrows of Asxl1 ^tm/+ mice (p = 0.049) than wild-type controls, while both mice had similar frequency of HSPCs (Fig. 5c, Additional file 1: Figure S3). Within the Lin^-c-Kit⁺Sca-1^- (LK) bone marrow cells, there was no significant difference in the proportion of CMP, GMP, and MEP between Asxl ^tm/+ and wild-type mice (Fig. 5d). The autopsy also did not show any difference in the incidence of splenomegaly between the two groups. The only six viable Asxl1 homozygous mice did not exhibit obvious abnormalities in hemogram or in autopsy findings at age of 18 months (Additional file 1: Figure S4). During the life span of the mice, Asxl1 G643WfsX12 showed no tendency to induce any kind of blood malignancies; hence, we concluded that Asxl1 G643WfsX12 alone was not sufficient for leukemogenesis in vivo.

Since Asxl1 mutation alone did not produce obvious blood diseases in mice, we sought to determine if this mutation functions as a facilitator for leukemogenesis. In our patients with array data (N = 349, among whom the mutation status of ASXL1 was known in 343) [29‐31], we noted that those bearing ASXL1 mutation tended to have higher MN1 expression (P = 0.056, Fig. 6a). Moreover, among the 225 patients who received standard chemotherapy, those with higher MN1 expression (≥ median) as well as ASXL1 mutation had shorter overall survival compared to those with higher MN1 expression but without ASXL1 mutation (Fig. 6b), suggesting a possible cooperative effect between these two genetic aberrancies in AML patients. MN1 over-expression is a sufficient driving event for mouse leukemogenesis [32]. Therefore, we were interested to know whether Asxl1 mutation facilitates the engraftment of MN1 over-expression in mice. To this end, we overexpressed MN1 in Asxl1 ^tm/+ or wild-type Lin^- bone marrow cells by retroviral transduction. Proliferation of wild-type and Asxl1 ^tm/+ cells were not different as examined by BrdU incorporation assays (Fig. 6c). When MN1 was transduced into Asxl1 ^tm/+ and wild-type bone marrow cells, respectively, we still could not observe significant difference in proliferation rate between these two types of cells (Fig. 6d). However, long-term culture-initiation cell (LTC-IC) assay showed that when MN1 was overexpressed in Asxl1 ^tm/+ bone marrow cells, there was higher percentage of long-term colony forming cells compared to MN1 overexpressed wild-type bone marrow cells (Fig. 6e), implying that Asxl1 mutation promoted stem cell activities of marrow cells in MN1 overexpression background. To test this hypothesis, we transplanted several different doses of MN1-transduced Lin^- bone marrow cells, together with 200,000 helper cells, into lethally irradiated recipients. Bone marrow cells of the recipient mice were harvested between 4 to 5 weeks after transplantation to evaluate the reconstitution efficiency. More than 1% MN1 over-expressing cells in the marrow cells of recipient mice was defined to be successfully reconstituted. At 5000-cell dose, 100% of recipient mice transplanted with either MN1 overexpressed Asxl1 mutant or wild-type bone marrow cells were successfully reconstituted. However, while most recipient mice transplanted with low-dose MN1-transduced cells could be reconstituted in the presence of Asxl1 mutation (9 out of 11 at 1000 test cells and 7 out of 7 at 500 test cells were successfully reconstituted), significantly lower proportion of recipient mice transplanted with the same doses of MN1-transduced cells without Asxl1 mutation were successfully reconstituted (7 out of 12 at 1000 test cells and 1 out of 6 at 500 test cells, p = 0.036 by Chi-square test) (Table 1 and Additional file 1: Figure S5). Our results suggest that Asxl1 G643WfsX12 can lower the threshold of MN1-driven engraftment.

Our mouse model provided an ideal platform to investigate the impacts of Asxl1 mutation per se on global gene expression patterns, since the genetic backgrounds of our mice were far less complicated than those in human leukemia cells. In addition, we were interested in the mechanisms underlying the supportive role of Asxl1 mutation in the engraftment of MN1 overexpressing bone marrow cells. To these ends, we collected Lin^- marrow cells from Asxl1 ^tm/+ and wild-type control mice with or without MN1 transduction (wild-type, Asxl1 mutation, wild-type + MN1 overexpression, and Asxl1 mutation + MN1 overexpression) for microarray analyses to explore differential gene expression patterns as well as molecular functions conferred by Asxl1 mutation per se and/or its interplay with MN1 overexpression. Of note, only 4.6% genes were significantly differentially expressed between Asxl1-mutated and wild-type cells (Fig. 7a; left bar); the number of perturbed functional gene sets was also very modest (3.2%, comparing Asxl1 mutation vs. wild-type cells, Fig. 7b; left bar). However, the differences became obvious when comparing Asxl1-mutated cells overexpressing MN1 vs. wild-type cells overexpressing MN1 cells: up to 12.2% differentially expressed genes among all genes (Fig. 7a, second bar from the left), higher than that achieved by randomly shuffling the microarray dataset for 100 times (empirical p < 0.01), with correspondingly large scale of perturbed gene sets, up to 23.9% (Fig. 7b, second left bar). These data were consistent with our findings that Asxl1 alone did not render obvious blood diseases in the mice but it might play a cooperative role with MN1.

To compare our mouse model with human disease, we profiled gene expression of leukemia cells from a total of 343 AML patients and compared the expression patterns between samples with (N = 50) and without (N = 293) ASXL1 mutation. For 172 AML patients with higher MN1 expression (above the median level), we also compared the expression patterns between those with (N = 29) and without (N = 143) ASXL1 mutation. The disturbance of global gene expression profiles and gene sets related to Asxl1 mutation were quite comparable and obvious in total cohort (Fig. 7a, right two bars) and in the subgroup of patients with higher MN1 expression (Fig. 7b, right two bars).

A deeper look into the lists of significantly differential gene sets derived by GSEA revealed a handful of crucial oncogenic functions perturbed by the interaction between Asxl1 mutation and MN1 overexpression. Hypoxia-related genes were implied to be relevant factors of leukemogenesis [33, 34]. In our data, while the expression of genes of a hypoxia signature did not show an overall change in mice with Asxl1 mutation vs. wild-type littermates (GSEA p = 0.272; Fig. 7c), they were significantly co-upregulated in Asxl1 ^tm/+ , compared to wild-type mice, in the presence of MN1-overexpression (p = 0.007; Fig. 7d). Similar enrichments were seen in signatures representing multipotent progenitors and hematopoietic stem cells, as well as genes related to oncogenic KRAS and MEK, in an MN1-dependent manner (all p values <0.0005 in Asxl1 ^tm/+ vs. wild-type mice transduced with MN1, Fig. 7d; compared with p > 0.05 in Asxl1 mutation vs. wild-type mice without MN1 overexpression, except for genes related to MEK, Fig. 7c, p value 0.012). Such positive enrichment toward Asxl1 mutation was corroborated in AML patients, regardless of the abundance of MN1 (all p values <0.0005; Fig. 7e and f). In aggregate, the logistic relationship between Asxl1 mutation and MN1 overexpression is summarized as (1) Asxl1 mutation promoted engraftment of bone marrow cells in MN1 overexpression background (Fig. 6e and Table 1); (2) AML patients with both ASXL1 mutation and high MN1 expression had inferior survival when compared with ASXL1-wild-type and high MN1 expression (Fig. 6b); (3) In the background of MN1 overexpression, Asxl1 mutation in mice and in human AML patients was associated with upregulation of signatures of hematopoietic stem/progenitor cells and related to hypoxia, KRAS, and MEK pathways.

Several studies have linked functions of Asxl1 mutation to H3K27me3, an inactive mark associated with transcriptional repression [19]. In order to investigate their interactions in our mouse model, we performed histone extraction followed by western hybridization to evaluate the global H3K27me3 levels in bone marrow cells. There was no significant difference in global H3K27me3 levels between Asxl1 ^tm/+ bone marrow cells and wild-type bone marrow cells (Additional file 1: Figure S6). Then, we analyzed if there was different global H3K27me3 pattern between Asxl1 ^tm/+ and wild-type Lin^- bone marrow cells via ChIP-Seq analysis. Comparing sequencing reads of ChIP products and input controls, we identified ~70 k H3K27me3-binding peaks in each of the Asxl1 ^tm/+ and wild-type samples. Of note, considerable proportions of them were Asxl1 ^tm/+ cells-specific (25,695; 37.0%) or wild-type (26,850; 37.7%) cells-specific. These peaks are highly concordant with gene loci in the mouse genome (Fig. 8a). Mn1 harbored three H3K27me3-binding sites, of which one was Asxl1 ^tm/+ -specific (Fig. 8a, left lower panel). We then analyzed the distribution of the condition-specific peaks in gene regions. Significant enrichment of peaks was found in upstream (within 5 k bps; 7.2 and 6.8% of Asxl1 ^tm/+ - and wild-type-specific peaks, respectively; both p < 0.001) and downstream regions of gene bodies (within 1 k bps; 1.4%, p = 0.031; and 1.3%, p = 0.023, respectively) compared to randomly distributed peaks across the genome (Fig. 8b). Other genomic categories, such as 5’ and 3’ untranslated regions (UTRs) and exons, were not enriched (all p values >0.05), suggesting the preference of Asxl1 mutation-associated H3K27me3 occupancy in gene regulatory regions.

To further investigate Asxl1 mutation-modulated targets of H3K27me3, we analyzed enriched motifs on the peaks by the MEME-ChIP web tool, which performs motif discovery, enrichment, and visualization from DNA sequences of interest. Interestingly, while Asxl1 ^tm/+ and wild-type specific peaks do not overlap with each other, they carried very similar motifs: RGRAA, TVTGTR, and TTTAWW (all E values <0.001; Fig. 8c), indicating that the modulation of Asxl1 mutation in H3K27me3 occupancy is independent of the binding motifs.

In order to investigate the effects of such selective targeting on gene expression and biological functions, we linked the peaks with gene expression microarrays. Lists of H3K27me peaks-associated genes with concordant significant downregulation are provided in Additional file 1: Table S1. Notably, the down-regulated genes in Asxl1 wild-type cells were significantly associated with concordant Asxl1 wild-type- specific H3K27m3 peaks (15.69% of the Asxl1 wild-type-specific downregulated genes harbored concordant Asxl1 wild-type-specific H3K27m3 peaks, compared to 11.69% of non-downregulated genes; Fisher’s exact test p = 0.001; Fig. 8d, upper panel). However, Asxl1 ^tm/+ cells-specific peaks were not significantly associated with down-regulated genes in Asxl1 ^tm/+ cells (Fisher’s exact test p = 0.52; Fig. 8d, lower panel), implying that in Asxl1 mutated cells, the association between H3K27m3 and gene downregulation is disrupted when compared with Asxl1 wild-type condition. Taken together, our ChIP-Seq data demonstrated distinct Asxl1 mutation-modulated binding profiles of H3K27me3.