Reduced expression level or loss of Gfi1b promotes progression of MDS to AML in a murine MDS model

To investigate a connection between Gfi1b level and AML, we used different mouse strains and models of human leukemia. We used one strain in which both Gfi1b alleles can be conditionally deleted in the hematopoietic system upon poly(I:C) administration, resembling absence of Gfi1b expression (Gfi1b^fl/flMxCre^tg).⁶ In a second mouse model, one coding allele of Gfi1b is replaced by enhanced green fluorescence protein (EGFP) cDNA (Gfi1b^EGFP/wt),¹³ which leads to a lower expression level of Gfi1b (see below). Finally, wild-type mice were used to model normal/high Gfi1b expression. To study whether reduced Gfi1b expression accelerates MDS to AML progression, we crossed the above-mentioned mouse strains with NUP98/HOXD13^tg mice, which represent a model for human MDS/AML.³⁴

We first used the Gfi1b:EGFP knock-in reporter mouse strain and crossed these mice with NUP98/HOXD13^tg mice (Figure 3A). Loss of one allele of Gfi1b shortened the latency period of AML development (Figure 3B). In BM cells derived from heterozygous leukemic mice, the expression of Gfi1b mRNA and Gfi1b protein levels were reduced to approximately 50% compared to BM cells from Gfi1b^wt/wt leukemic mice (Online Supplementary Figure S2A and B). Furthermore, we found that the EGFP expression level and hence Gfi1b expression level was significantly lower in the myeloid blasts when the disease onset was within the first 250 days compared to Gfi1b expression in blasts from mice that developed overt leukemia more than 250 days after birth (Figure 3C). The leukemic cells from Gfi1b^wt/wt or Gfi1b^EGFP/wt animals showed no significant differences with regard to surface marker expression, spleen size, white blood cell and platelet counts, or cytological appearance, but showed significant differences with regard to hemoglobin and red blood cells (Figure 3D, Online Supplementary Figure S2C-F and data not shown), which might be due to a potential dose-dependent role of Gfi1b in erythropoiesis.^6,9

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Low level or absence of Gfi1b accelerates the progression of myelodysplastic syndrome (MDS) to acute myeloid leukemia (AML) in the NUP98/HOXD13 MDS mouse model. (A) Crossing of the Gfi1b^wt/wt and Gfi1b^EGFP/wt mouse strains with the NUP98/HOXD13 mouse model. (B) Survival of Gfi1b^wt/wt and Gfi1b^EGFP/wtNUP98/HOXD13 transgenic mice; P=0.0039. Number of mice succumbing to AML is indicated. (C) Mean fluorescence intensity (MFI) of the GFP expression level (and hence Gfi1b promoter activity) in Gfi1b^EGFP/wt mice that died of AML before 250 days (n=7) or after 250 days (n=5); P=0.0272. (D) Wright-Giemsa staining of bone marrow (BM) cytospins from representative Gfi1b^EGFP/wt and Gfi1b^wt/wt NUP98/HOXD13 leukemic mice (bar=20 μm). (E) Crossing of the Gfi1b^fl/flMxCre^wt and Gfi1b^fl/flMxCre^tg mouse strains with the NUP98/HOXD13 mouse model. After Cre-mediated deletion of the Gfi1b gene upon poly(I:C) administration, the mice were monitored for signs of leukemia. (F) Survival of Gfi1b^fl/flMxCre^wt(Gfi1b^wt/wt) and Gfi1b^fl/flMxCre^tg (Gfi1b^KO/KO) mice transgenically expressing NUP98/HOXD13 after poly(I:C) administration; P<0.0001. Number of mice succumbing to AML is indicated. (G) Polymerase chain reaction genotyping of DNA from BM cells of poly(I:C)-injected Gfi1b^fl/flMxCre^wt and Gfi1b^fl/flMxCre^tgNUP98/HOXD13 leukemic mice. (H) Wright-Giemsa staining of BM cytospins from representative poly(I:C)-injected Gfi1b^fl/flMxCre^wt and Gfi1b^fl/flMxCre^tgNUP98/HOXD13 leukemic mice (bar=20 μm). (I) The frequency of monocytes (Mac-1hiGr-1int) (left panel, ****P<0.0001), granulocytes (Mac-1hiGr-1hi) (middle panel, *P=0.0206) and CD117⁺ (c-Kit) cells (right panel, ****P<0.0001) in the BM of mice described in (F) (n=15 for Gfi1b^fl/flMxCre^wt mice; n=13 for Gfi1b^fl/flMxCre^tgNUP98/HOXD13 mice).

We next examined how complete absence of Gfi1b influences MDS to AML progression. We used the Gfi1b conditional knockout mouse model (Gfi1b^fl/flMxCre^tg), whereby the expression of Gfi1b can be conditionally abrogated in the hematopoietic system upon poly(I:C) administration⁶ (Figure 3E). The absence of Gfi1b resulted in a substantially earlier onset of AML with a median survival time of approximately 50 days (P<0.0001) (Figure 3F). Cre-mediated excision was verified to be efficient in leukemic Gfi1b^fl/flMxCre^tgNUP98/HOXD13^tg mice after poly(I:C) administration with non-excised Gfi1b alleles below detection levels (Figure 3G), and this was associated with practically no expression of Gfi1b mRNA and protein (Online Supplementary Figure S2A and B).⁶ Leukemic cells from Gfi1b^fl/flMxCre^tg and Gfi1b^fl/flMxCre^tg animals showed no significant differences in spleen size, white blood cells or cytological appearance but significant differences in hemoglobin, red blood cells and platelet counts (Figure 3H and Online Supplementary Figure S2G-J and data not shown), which might be due to a dose-dependent role of Gfi1b in erythropoiesis.^6,9 The absence of Gfi1b led to a reduced frequency of myeloid cells (Figure 3I, left, middle, and Online Supplementary Figure S3A–C). CD117 (c-Kit) was uniformly higher expressed on all Gfi1b-deficient blast cells (derived from Gfi1b^fl/flMxCre^tg) mice compared to Gfi1b expressing blasts (Gfi1b^fl/flMxCre^wt) (Figure 3I, right). Finally, there was no difference with regard to apoptosis in NUP98/HOXD13^tg mice (Online Supplementary Figure S3D). In our murine model of MDS/AML development, we did not observe a positive correlation between Gfi1b and Mllt3 expression nor a negative correlation between Gfi1b and Spi1 expression, which might be disease context-dependent and thus not reproducible in all types of AML (Online Supplementary Figure S3A and B). We also analyzed the expression level of Meis1, since Chowdhury et al. observed a negative correlation between GFI1B and MEIS1.³³ We were able to confirm this finding for this model of AML (Online Supplementary Figure S4C).

Loss of Gfi1b promotes the progression of myeloproliferative disorder in a conditional Kras mouse model

To validate the results above in a second model, we used mice conditionally expressing a mutated form of Kras. RAS mutations are found in 5–10% of AML patients.² These mice harbor a transcriptional stop codon flanked by loxP sites upstream of a mutated Kras allele and, after removal of the stop codon, develop myeloproliferative disorder.³⁵ We crossed these mice with Gfi1b^fl/flMxCre^tg or Gfi1b^wt/wtMxCre^tg mice, and after poly(I:C) administration, we observed mice for the emergence of disease (Figure 4A).

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

Absence of Gfi1b accelerates the progression of myeloproliferative disorder and acute myeloid leukemia (AML). (A) Crossing of the Gfi1b^wt/^wtMxCre^tg and Gfi1b^fl/flMxCre^tg mouse strains with the Kras^+/fl mouse model. (B) Survival of Gfi1b^wt/wtMxCre^tg and Gfi1b^fl/flMxCre^tg mice transgenically expressing Kras after poly (I:C) administration; ****P<0.0015. Numbers indicate the number of mice succumbing to AML. (C) Wright-Giemsa staining of bone marrow (BM) cytospins from representative Gfi1b^wt/wtMxCre^tg and Gfi1b^fl/flMxCre^wt leukemic mice transgenically expressing Kras after poly(I:C) administration (bar=20 μm). (D) Flow cytometric analysis of the BM from the leukemic mice shown in (B) with regard to Gr-1 and Mac-1 expression. (E) Isolation, transduction and transplantation of lineage-negative (Lin-) cells from Gfi1b^fl/flMxCre^wt and Gfi1b^fl/flMxCre^tg mice with MLL-AF9-expressing retrovirus. After Cre-mediated deletion of the Gfi1b gene upon poly(I:C) administration, the mice were monitored for signs of leukemia. (F) Survival of the mice transplanted with Gfi1b^fl/flMxCre^wt and Gfi1b^fl/flMxCre^tg MLL-AF9 transduced cells; ***P=0.0002. Number of mice succumbing to AML is indicated. (G) Wright-Giemsa staining of BM cytospins from the leukemic mice described in (F) (bar=20 μm). (H) Flow cytometric analysis of the BM from the leukemic mice described in (F) with regard to Gr-1 and Mac-1 expression.

While Gfi1b^wt/wtMxCre^tgKras^+/fl mice developed a lethal myeloproliferative disorder with a median survival of approximately 25 days, loss of Gfi1b significantly shortened the latency period of the disease to a median survival of approximately seven days (Figure 4B). There was no difference with regard to cytological appearance, number of myeloid cells or level of apoptosis (Figure 4C and D and Online Supplementary Figure S5A–D). We also did not observe any significant difference with regard to white blood counts, platelet counts or spleen size but a significant difference in hemoglobin and red blood cells between Gfi1b^fl/flMxCre^tg and Gfi1b^wt/wtMxCre^tg animals (Online Supplementary Figure S5E–H), which might be due to the role of Gfi1b in erythropoiesis.^6,9

Loss of Gfi1b promotes the progression of AML initiated by retroviral MLL-AF9 expression

The Mixed Lineage Leukemia (MLL) gene is a common target for chromosomal translocations.² MLL-AF9 is a fusion protein frequently occurring in a subset of AML patients,² and its expression in hematopoietic progenitors has been linked to the induction of AML in mice.³⁶ As a third AML mouse model, we thus used mice that developed AML through the induction of MLL-AF9 expression, the product of the t(9;11)(q22;p23) translocation. Lin- BM cells derived from Gfi1b^wt/wtMxCre^tg or Gfi1b^fl/flMxCre^tg mice were transduced with a retrovirus expressing MLL-AF9 and transplanted into lethally irradiated C57BL/6J mice. For Cre-mediated excision of Gfi1b in the transplanted cells, mice were injected with poly(I:C) three weeks after transplantation (Figure 4E). Poly(I:C)-injected mice with MLL-AF9-transduced Gfi1b^fl/flMxCre^tg (Gfi1b-deficient) cells succumbed faster to leukemia than mice injected with poly(I:C) and transplanted with MLL-AF9-transduced Gfi1b^wt/wtMxCre^tg (Gfi1b-expressing) cells (Figure 4F). However, there were no major qualitative differences concerning cytological findings and or blood parameters (Figure 4G and data not shown). We did not observe a significant change in the number of overall myeloid cells or apoptosis level in the different settings (Figure 4H and Online Supplementary Figure S6A–D).

Loss of Gfi1b increases the number of LSCs

Loss of Gfi1b leads to an expansion in the number of functional HSCs;⁶ therefore we investigated whether the same applies to LSCs. We performed a limiting dilution assay by transplanting MLL-AF9 leukemic BM cells derived from poly(I:C)-treated Gfi1b^fl/flMxCre^wt or Gfi1b^fl/flMxCre^tg leukemic mice into sublethally irradiated congenic mice (Figure 5A). Gfi1b-deficient MLL-AF9 BM cells had a LSC frequency of 1:3500 compared with an LSC frequency of 1:63000 cells in Gfi1b-expressing leukemic cells (Figure 5B). The increased number of functional LSCs in Gfi1b-deficient leukemic cells could explain why loss of Gfi1b accelerated disease progression, as it has already been shown that a higher number of LSCs is associated with a poor prognosis of leukemia patients.³⁷ However, this hypothesis needs to be confirmed in independent experiments.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5.

Loss of Gfi1b increases the stemness of LSCs. (A) In vivo limiting dilution assay for determination of functional LSCs. The indicated numbers of poly (I:C)-treated Gfi1b^fl/flMxCre^wt and Gfi1b^fl/flMxCre^tg MLL-AF9 transduced cells were retransplanted into sublethally irradiated mice. (B) Survival of the mice transplanted with different numbers of poly(I:C)-treated Gfi1b^fl/flMxCre^wt MLL-AF9 and Gfi1b^fl/flMxCre^tg MLL-AF9 transduced cells. (C) Determination of functional LSCs by limiting dilution assay.

Loss of Gfi1b induces gene expression changes supporting AML development

To further study the molecular function of Gfi1b in AML, we performed whole genome gene expression analysis using Gfi1b-expressing and Gfi1b-deficient NUP98/HOXD13^tg leukemic mice (Figure 6A). This model was used since the difference between Gfi1b-deficient and Gfi1b-expressing leukemic cells was most striking in the NUP98/HOXD13^tg mouse model. Using gene set enrichment analysis (GSEA), loss of Gfi1b was associated with a signature showing enrichment of genes involved in AML development as well as regulation of stemness (Figure 6B). This is of interest since we observed an increase in the number of LSCs upon deletion of Gfi1b.

• Download figure
• Open in new tab
• Download powerpoint

Figure 6.

Loss of Gfi1b induces gene expression changes supporting acute myeloid leukemia (AML) development. (A) Microarray analysis was performed with Gfi1b^fl/flMxCre^wtNUP98/HOXD13 and Gfi1b^fl/flMxCre^tgNUP98/HOXD13 leukemic cells. (B) Based on the results of the micro-array analysis, a Gene set enrichment analysis (GSEA) of leukemic cells from Gfi1b^fl/flMxCre^wtNUP98/HOXD13^tgan d Gfi1b^fl/flMxCre^tgNUP98/HOXD13^tg mice w as performed. As a result, Gfi1b deficient leukemic cells showed an enrichment of the gene set of VALK AML cluster 8 with a normalized enrichment score (NES) of 2.1 and false discovery rate (FDR) of q-val=0.00191213. Gfi1b deficient cells showed also an enrichment for RAMALHO STEMNESS with an NES=2.41 and an FDR q-val=0.(C) ChIP and ChIP-Seq analysis was performed with Gfi1b^fl/flMxCre^wtNUP98/HOXD13 and Gfi1b^fl/flMxCre^tgNUP98/HOXD13 leukemic cells. ChIP-Seq analysis for differences in the frequency of H3K9 acetylation of Gfi1b-deficient (Gfi1b^fl/flMxCre^tg) leukemic blasts from NUP98/HOXD13^tg mice compared to leukemic cells with normal Gfi1b expression (Gfi1b^fl/flMxCre^wt). (D) GSEA of genes with an elevated acetylation level in Gfi1b deficient mice are associated with regulation of cell growth NES=1.78 and FDR q-val=0.055857178 and GNF2_MAP2K3 NES=2.02 and FDR q-val=1.7715618E-4. (E) Upon analyzing the differentially acetylated genes and using the MSigDB Pathway approach, we found significant enrichment of different pathways, among them the p38 pathway. P-value was used to rank the enrichment.

Gfi1b recruits different histone-modifying enzymes, among them HDACs,¹⁶ to its target genes. This in turn leads to deacetylation of H3K9, which leads to epigenetic silencing of the particular Gfi1b target genes.¹⁶ We, therefore, analyzed the genome-wide H3K9 acetylation level of leukemic blasts from Gfi1b-expressing and Gfi1b-deficient NUP98/HOXD13^tg leukemic mice. Loss of Gfi1b leads to a genome-wide increase in H3K9 acetylation level (Figure 6C). In a subsequent step, we analyzed those genes, which showed an elevated level of H3K9 acetylation in Gfi1b-deficient leukemic cells compared to the H3K9 acetylation level of the same genes found in Gfi1b-expressing leukemic cells. Using GSEA, we found a significant enrichment of gene sets associated with the regulation of cell growth and MAPK signaling (Figure 6D). We also performed a Kyoto encyclopedia of genes and genomes (KEGG) pathway analysis of those genes, which exhibited differentially H3K9 acetylated promoter areas in Gfi1b-expressing and Gfi1b-deficient leukemic cells. We found a number of processes involved in erythroid regulation (Online Supplementary Figure S7A and B), which is a main function of Gfi1b and hence underscores the validity of our results.^6,9,16 Finally, we analyzed the differentially acetylated genes in Gfi1b-deficient and Gfi1b-expressing leukemic cells and compared these gene sets based on the gene sets provided by the Molecular Signatures Database (MSigDB). Using this approach, we repeatedly found signatures associated with p38 (Figure 6E).

We observed increased H3K9 acetylation of the promoter area of genes involved in stem cell function in Gfi1b-deficient leukemic cells, and these epigenetic changes correlated with the gene-expression changes described above (Figure 6B). As described, gene expression arrays revealed an enrichment of a stem cell/leukemic stem cell gene signature in Gfi1b-deficient leukemic cells. To validate these results, we selected 14 different genes (Abcg2, Gata3, Itga2, Thy1, Cd24a, Pecam1, Prom1, Plaur, Klf4, Mycn, Ptch1, Pecam1, Sav1, and Notch1), which were differentially expressed by more than 2-fold between Gfi1b-expressing and Gfi1b-deficient leukemic mice in the gene expression arrays. We selected these genes based on their diverse role in regulating stem cell function. We then examined these genes and confirmed that these genes were also differentially expressed using RT-PCR (Online Supplementary Figure S8A).

Loss of Gfi1b leads to increased ROS levels and decreased levels of activated p38

To obtain further insight into the molecular mechanism behind our observation we compared the whole genome gene expression pattern in murine Gfi1b-expressing and Gfi1b-deficient leukemic cells based on an AltAnalyze approach (see Online Supplementary Appendix). Using the same algorithm, we compared the gene expression pattern found in AML blasts with low GFI1B expression and high GFI1B expression (data obtained from published studies from Valk et al.²¹ and Verhaak et al.²²). For the analysis of the human dataset, we analyzed the 10% of patients with the lowest and the 20% of patients with the highest GFI1B expression level in order to have enough observations from which to draw any conclusions. Then we compared which pathways were similarly deregulated in the human and murine leukemia sets. ROS and MAPK signaling were among the pathways differentially expressed between both murine and human Gfi1b/GFI1B-deficient/low and Gfi1b/GFI1B high-expressing leukemic cells (Figure 7A). As ROS plays an important role in the pathogenesis of AML,³⁸ we examined the level of ROS. For non-malignant HSCs, it was shown that HSCs with low ROS had a high self-renewal capacity.³⁹ In contrast, HSCs with elevated ROS were mostly located in the vascular niche, had a reduced self-renewal capacity, and were more restricted with regard to their differentiation potential.³⁹ Based on this and our previous report that loss of Gfi1b leads to higher level of ROS in HSCs,⁶ we examined whether ROS level might differ between Gfi1b-deficient and Gfi1b expressing LSCs. Due to the difficulty of defining a distinct LSC population in each set of AML samples, we used CD117 (c-Kit) expression as a surrogate marker to define a population which is enriched for LSCs. CD117 expression has been used to identify a fraction that is enriched for LSCs.⁴⁰ We identified two distinct populations in the CD117⁺ blast cells that differ with regard to their ROS expression (a population with low ROS expression and a population with high ROS expression). Loss of Gfi1b led to an increased level of ROS (defined as the mean fluorescent intensity, MFI) in both ROS-low and ROS-high populations (Figure 7B–D and Online Supplementary Figure S8B).

• Download figure
• Open in new tab
• Download powerpoint

Figure 7.

Loss of Gfi1b deregulates acute myeloid leukemia (AML) signaling pathway. (A) Analysis to see which pathways were enriched in GFI1B low-expressing blast cells in two independent sets of myelodysplastic syndrome (MDS) or AML patients compared to the expression pattern found in GFI1B high-expressing MDS/AML blast cells. The same approach was repeated for Gfi1b-expressing and Gfi1b-non-expressing leukemic cells from our mice experiments. As an overlap, enrichment was observed in pathways of JAK-STAT-, MAPK- and ROS-related signaling. (B) Representative flow cytometric analysis of bone marrow (BM) from Gfi1b^fl/flMxCre^tgNUP98/HOXD13^tg mice compared to Gfi1b^fl/flMxCre^wtNUP98/HOXD13^tg mice showing the gating strategy for determining ROS low and ROS high levels. (C) Mean fluorescence intensity (MFI) for ROS in the ROS-low population of c-Kit⁺ blast cells derived from Gfi1b^fl/flMxCre^wtNUP98/HOXD13^tg (n=6) and Gfi1b^fl/flMxCre^tgNUP98-HOXD13^tg (n=5); *P=0.0488. (D) Mean fluorescence intensity (MFI) for ROS in the ROS-high population of c-Kit⁺ blast cells derived from Gfi1b^fl/flMxCre^wtNUP98-HOXD13^tg (n=6) and Gfi1b^fl/flMxCre^tgNUP98/HOXD13^tg(n=5); *P=0.0191. (E) Flow cytometric analysis of p38 MAPK (pT180/pY182) in CD117⁺ blast cells derived from Gfi1b^fl/flMxCre^wtNUP98/HOXD13^tg (n=5) and Gfi1b^fl/flMxCre^tg NUP98/HOXD13^tg (n=6); *P=0.0144. (F) Flow cytometric analysis of Akt (pS473) in c-Kit⁺ blast cells derived from Gfi1b^fl/flMxCre^wtNUP98/HOXD13^tg (n=4) and Gfi1b^fl/flMxCre^tgNUP98/HOXD13^tg (n=5); **P=0.0040. (G) FoXO3 protein level was detected in nuclear extraction (NER)- and cytoplasmic extraction (CER)-derived BM cells from Gfi1b^fl/flMxCre^tgNUP98/HOXD13^tg and Gfi1b^fl/flMxCre^wtNUP98/HOXD13^tg. (H) Working model hypothesis: normal levels of Gfi1b lead to reduced ROS levels, resulting in normal maturation and differentiation of progenitor cells. Loss of Gfi1b in leukemic cells is associated with higher ROS levels, which have been shown to promote AML development. However, through a still undefined mechanism, this results in lower levels of p38 MAPK and pAkt and higher levels of unphosphorylated FoXO3, which might explain the increased number of functional leukemic stem cells in the Gfi1b-deficient AML population. LSC: leukemic stem cells.

Altered activity of the AKT pathway in Gfi1b-deficient AML

Elevated levels of ROS promote AML development, but ROS also activates various redox-sensitive signaling transduction cascades,⁴¹ including the MAPK pathway, which limits the stemness of the affected cells, at least in a non-malignant setting.⁴² In the presence of ROS, the MAPK pathway component p38 is activated, which subsequently results in an exhaustion of the HSC population.⁴³ It has also been shown that activation of p38 limits oncogenic transformation.⁴⁴ Despite a higher level of ROS, in our models, Gfi1b-deficient leukemic cells have escaped p38 activation, indicated by a decreased level of phosphorylated p38 compared to Gfi1b^fl/flMxCre^wtNUP98/HOXD13^tg (Figure 7E). The fact that Gfi1b might directly or indirectly regulate p38 is also supported by the analysis of differentially acetylated genes in Gfi1b-deficient and Gfi1b-expressing leukemic cells. Because decreased p38 levels are associated with higher oncogenic potential,⁴⁴ this could partially explain the higher number of functional LSCs we observed in the Gfi1b-deficient leukemic cells. Activation of p38 leads to an increased level of Akt^Ser473.45 Akt^Ser473 activity inversely correlates with the number of LSCs in AML.⁴⁶ We thus examined the connection between loss of Gfi1b and Akt^Ser473 and found that the level of phosphorylated Akt^Ser473 is reduced in Gfi1b-deficient leukemia (Figure 7F). Akt represses the function of FoXO3, and since FoXO3 acts as an oncogene in AML,⁴⁶ we determined the FoXO3 protein level. Gfi1b-deficient leukemic cells showed an increased expression of FoXO3 protein in the nucleotide (NER) and cytoplasmatic (CER) cell fraction compared to the expression level of FoXO3 in Gfi1b-expressing leukemic cells (Figure 7G). To obtain an insight into whether this increased level of FoXO3 also has any functional consequences, we re-examined the whole genome expression datasets in Gfi1b-expressing and Gfi1b-deficient leukemic cells and found an enrichment of FoXO3 binding sites among the promoter areas of those genes, which were differentially expressed between cells with absence of Gfi1b expression and cells with intact expression of Gfi1b (Online Supplementary Figure S9), showing that altered level of FoXO3 might be one additional explanation for our observations (Figure 7H).