Initiation of MLL-rearranged AML is dependent on C/EBPα

We hypothesized that transcriptional regulators that collaborate with MLL-fusion proteins during leukemic initiation would generate a transcriptional footprint in MLL-rearranged AML. To identify such a footprint we used gene set enrichment analysis (GSEA) to compare the publically available gene expression profiles of AML blasts from patients with MLL-rearrangements to those of normal healthy GMPs. Interestingly, when we scrutinized the data for transcriptional regulators, we found several signatures for C/EBPα target genes to be up-regulated in MLL-rearranged AML (Fig. 1 A), suggesting that C/EBPα is a potential collaborator in MLL fusion–induced malignant transformation.

To test if C/EBPα is indeed required for MLL-fusion driven transformation, we first generated Cebpa^fl/fl;Mx1Cre mice and used polyinosinic-polycytidylic acid (pIpC) to facilitate ablation of C/EBPα in the hematopoietic compartment (Kühn et al., 1995; Lee et al., 1997). 2 wk after deletion, c-Kit⁺ hematopoietic stem and progenitor cells (HSPCs) from control (Cebpa^fl/fl) and Cebpa-deleted (Cebpa^Δ/Δ) animals were transduced with either the empty pMIG retroviral vector or a pMIG-MLL-ENL construct. Transduced cells were subsequently serially replated in methylcellulose medium (Somervaille and Cleary, 2006) or cultured in liquid medium. Interestingly, whereas Cebpa^fl/fl cells transduced with MLL-ENL had a high replating efficiency, MLL-ENL was unable to provide Cebpa^Δ/Δ cells with an increased colony forming ability compared with Cebpa^fl/fl or Cebpa^Δ/Δ cells transduced with empty vector (EV; Fig. 1 B). Furthermore, whereas MLL-ENL–transduced Cebpa^fl/fl cells selectively accumulated in liquid culture, the MLL-ENL fusion protein did not provide the Cebpa-deficient cells with any proliferative advantage (Fig. 1 C). Analysis of the Cebpa^fl/fl and Cebpa^Δ/Δ HSPCs showed that whereas Cebpa^fl/fl cells lost c-Kit expression and up-regulated Mac-1 expression when transduced with MLL-ENL, the Cebpa^Δ/Δ cells did not alter the expression of these cell surface markers at day 18 after transduction (Fig. 1 D). Collectively, these findings demonstrate that C/EBPα is required for MLL-ENL-induced transformation in vitro.

To test if C/EBPα was also required for MLL-ENL–mediated development of AML in vivo, we next transplanted freshly transduced Cebpa^fl/fl or Cebpa^Δ/Δ HSPCs into irradiated recipients. As expected, mice reconstituted with Cebpa^fl/fl cells expressing MLL-ENL developed a lethal form of myeloid leukemia with a median latency of 10 wk, accumulation of GFP⁺-expressing blast cells in several hematopoietic tissues, pale bones, and splenomegaly (unpublished data). In contrast, MLL-ENL–transduced Cebpa^Δ/Δ HSPCs did not give rise to leukemia or other forms of dysplasia in transplanted mice (Fig. 1 E). Importantly, this was not because of poor reconstitution as we could detect GFP⁺ Cebpa^Δ/Δ cells in the periphery of transplanted mice at 13 wk post-transplant (Fig. 1 F). Moreover, the requirement of C/EBPα for myeloid transformation was not a general feature as the oncogenic fusion protein E2A-HLF was able to provide Cebpa^Δ/Δ HSPCs with a proliferative advantage (Fig. 1 G). Collectively, these results demonstrate that C/EBPα is required for MLL-ENL dependent transformation both in vitro and in vivo.

The aforementioned data could, in principle, be explained by a requirement for C/EBPα during leukemic initiation, leukemic maintenance, or both. To distinguish between these possibilities, we crossed the conditional Cebpa allele into mice expressing a tamoxifen-regulated Cre recombinase estrogen receptor fusion protein (R26-Cre-ER; Ventura et al., 2007) and established MLL-ENL–transformed Cebpa^fl/fl;Cre-ER cells. We next used 4-hydroxytamoxifen (4-OHT) to delete Cebpa and tested the requirement for C/EBPα in sustaining serial replating of the MLL-ENL–expressing cells. Surprisingly, deletion of Cebpa in this context did not affect colony numbers, suggesting that the transformed cells grow independently of C/EBPα (Fig. 1 H). To asses this in an in vivo situation, we generated Cebpa^fl/fl;Mx1Cre and Cebpa^fl/fl MLL-ENL expressing primary AMLs, transplanted these into sublethally irradiated secondary recipients, and treated the resulting mice with pIpC 2 wk later. Intriguingly, we found that ablation of Cebpa in an already established leukemia affected neither the overall survival nor the immunophenotype of the leukemic cells (Fig. 1, I and J). Moreover, transplantation of an equal dose of Cebpa^fl/fl;Mx1Cre and the Cebpa^Δ/Δ leukemic cells into sublethally irradiated tertiary recipients resulted in the development of AMLs with similar latencies, suggesting that C/EBPα is dispensable for the maintenance of leukemic stem cells (Fig. 1 K). Collectively, these results therefore demonstrate that C/EBPα is dispensable for both the in vitro and in vivo maintenance of fully transformed MLL-ENL cells and point to a specific role of C/EBPα during the initial transformation process.

C/EBPα drives the transition from pregranulocytic-monocytic progenitors (preGMs) to GMPs and deletion of Cebpa in the adult murine hematopoietic system results in a complete differentiation block between these two progenitors. In line with the notion that a differentiation block is an initiating step during leukemogenesis, CEBPA is mutated in ∼10% of all cases of AML (Reckzeh and Cammenga, 2010). However, no mutations result in the complete ablation of C/EBPα suggesting that residual C/EBPα activity is required for development of AML and it has therefore been hypothesized that C/EBPα is necessary for the leukemic cells to gain their myeloid identity at the GMP stage (Nerlov, 2004; Wagner et al., 2006). To test if deletion of Cebpa at the GMP stage was able to rescue the C/EBPα dependence of MLL-ENL–induced transformation, we FACS sorted Cebpa^fl/fl;Cre-ER and Cebpa^fl/fl GMP cells and cultured them in presence of 4-OHT to facilitate deletion of Cebpa. Cebpa was fully excised after 2 d of 4-OHT induction (Fig. 2 A) and the cells were further transduced with pMIG-empty or pMIG-MLL-ENL and serially replated in semisolid medium. Cells expressing C/EBPα and MLL-ENL formed increasing numbers of dense colonies and gave rise to leukemia when transplanted into irradiated mice, whereas MLL-ENL–expressing Cebpa^Δ/Δ GMPs were unable to form colonies or give rise to leukemia (Fig. 2, A–D). These results demonstrate that ablation of Cebpa downstream of the preGM to GMP differentiation block also abrogates the transformation process, suggesting that it is not the differentiation state, per se, but rather a specific C/EBPα-driven transcriptional program that is required for the initiation of MLL-rearranged leukemia.

Transformation by MLL-fusion proteins is associated with an up-regulation of the transcription factor Hoxa9 and its cofactor Meis1, and overexpression of these genes in HSPCs efficiently transforms the cells and gives rise to a myeloid leukemia that mimics MLL-rearranged AML (Kroon et al., 1998; Wilhelm et al., 2011). To test if these two key downstream targets were affected by Cebpa deletion, we assessed their expression changes during initial transformation (i.e., 72 h after MLL-ENL–mediated transformation of Cebpa^fl/fl and Cebpa^Δ/Δ preGMs), as well as in established MLL-ENL–transformed cells, or leukemias, after deletion of Cebpa. Loss of Cebpa leads to a pronounced down-regulation of Hoxa9 levels in untransduced preGMs, but leaves the expression of Meis1 unaffected (Fig. 3 A). Moreover, we found that whereas C/EBPα was required for the up-regulation of Hoxa9 and Meis1 during initial transformation (Fig. 3 A; P = 0.06 and P = 0.05 for Meis1 and Hoxa9, respectively), it was completely dispensable for the maintenance of the high expression levels of these genes in both transformed cells and established leukemias (Fig. 3, B and C). In addition, to underline the specific importance of C/EBPα during initial transformation, these results raised the possibility that the resistance of Cebpa-deficient cells to undergo MLL-ENL–mediated transformation might be caused by their failure to up-regulate the expression of Hoxa9/Meis1. This, in turn, suggested that this phenotype may be rescued by overexpression of these two genes. To test this, we retrovirally transduced Cebpa^fl/fl and Cebpa^Δ/Δ c-Kit⁺ HSPCs with Hoxa9-IRES-Meis1-PGK-neo or a control EV selected with G418 for 7 d in semisolid medium and further serially replated cells in methylcellulose medium. In addition, the cells were directly injected into irradiated recipient mice along with support cells. However, whereas Cebpa^fl/fl cells expressing Hoxa9 and Meis1 were efficiently immortalized in semisolid medium (Fig. 3 D) and gave rise to a myeloid leukemia with a mean latency of 7 wk (Fig. 3 E), the Cebpa^Δ/Δ cells were unable to form colonies in vitro and failed to give rise to AML in transplanted mice (Fig. 3, D and E). These findings not only demonstrate that C/EBPα acts upstream of HOXA9/MEIS1 but also imply that C/EBPα plays a role in the activation of a larger number of genes that are responsible for the malignant transformation in MLL-rearranged leukemia. Furthermore, because C/EBPα is dispensable for tumor maintenance, our findings suggest that C/EBPα regulates several genes that are specifically important for initiation of MLL-rearranged leukemia.

To identify such genes, we next performed gene expression analysis of Cebpa^fl/fl and Cebpa^Δ/Δ preGMs transduced with MLL-ENL for 72 h (Table S1). Using GSEA, we find that not only are several MLL leukemia signatures down-regulated in untransduced (GFP⁻) Cebpa^Δ/Δ versus Cebpa^fl/fl preGMs (Fig. 4 A) but also that MLL-ENL (GFP⁺) fail to induce their expression in Cebpa^Δ/Δ preGMs to levels exceeding those of untransduced Cebpa^fl/fl preGMs (Fig. 4 B).

Next, we focused our analysis on the 35 genes that were significantly up-regulated (FC > 1.5; P < 0.05) in Cebpa^fl/fl preGMs as a result of MLL-ENL expression (Fig. 4 C) and, as such, represents genes important for the early MLL-ENL–induced transformation process. Interestingly, 60% (21/35) of these genes were not up-regulated to the same level in Cebpa^Δ/Δ-expressing MLL-ENL preGMs (Fig. 4 D and Table S1) and, moreover, 26% (9/35) of the MLL-ENL–induced genes had a significantly lower basal expression in un-transduced Cebpa^Δ/Δ versus Cebpa^fl/fl preGMs (Table S1). This shows that MLL-ENL is dependent on C/EBPα activity to initiate transcription of genes during early MLL-ENL–induced transformation.

To gain further insights into the role of C/EBPα in this process, we next performed ChIP-seq analysis for C/EBPα binding in WT GMPs as well as for the histone marks H3K4me3 and H3K27me3 in Cebpa^Δ/Δ and Cebpa^fl/fl preGMs (Fig. 4 E; see Materials and methods). This analysis reveals a strong trend toward more C/EBPα binding in the immediate vicinity of MLL-ENL–induced genes (Fig. 4 F; P = 0.06 for the comparison between the white and blue bars at the 1-kb threshold; Table S2). Moreover, when we compare the distances from the TSS to the nearest C/EBPα peak, we found these to be shorter for MLL-ENL–induced genes and the subset of genes that are dependent on C/EBPα for their expression (Fig. 4 G, see Materials and methods for definitions). In line with this, the MLL-ENL–induced genes have significantly more C/EBPα peaks in close proximity to the TSS (Fig. 4 H). We do see minor differences in the abundance of the active chromatin mark, H3K4me3, and the repressive chromatin mark, H3K27me3, in Cebpa^fl/fl and Cebpa^Δ/Δ preGM cells at some of the MLL-ENL–induced loci (Fig. 4 E), but this is not a general feature and we do not observe any overall differences between these epigenetic marks on MLL-ENL–induced genes in Cebpa^fl/fl and Cebpa^Δ/Δ preGM (not depicted). We therefore hypothesized that C/EBPα could be required for the MLL-ENL–dependent recruitment of the methyl transferase DOT1L (Nguyen and Zhang, 2011). This would in turn result in the deposition of the active H3K79me2/me3 mark and in the transcriptional activation of the MLL-ENL–immediate genes. To test this, we performed H3K79me2 ChIP in Cebpa^fl/fl and Cebpa^Δ/Δ preGMs ± MLL-ENL, as well as in established MLL-ENL–expressing Cebpa^fl/fl;Cre-ER cells. Intriguingly, we find that whereas the high levels of the H3K79me2 mark at key Hox genes were unaffected by the C/EBPα status in established leukemic cells (Fig. 4 I), the deposition of H3K79me2 at key MLL-ENL–immediate genes during initial transformation was strongly dependent on C/EBPα (Fig. 4 J; P < 0.05 for all Cebpa^fl/fl vs. Cebpa^fl/fl + MLL-ENL comparisons except for Hoxa9 [P = 0.05] and Meis1 [P = 0.05]). These observations are consistent with C/EBPα being involved in the recruitment of MLL-ENL/DOT1L during the initial phases of leukemic transformation.

On a final note, we examined whether the aforementioned MLL-ENL–induced genes may also play a role in patients with MLL-rearranged leukemia. Here, we found that the majority of these genes displayed higher expression in leukemic cells compared with GMPs (Fig. 4 K), which suggests that the transcriptional network we have identified is highly relevant for patients with MLL-fusion protein leukemia and that C/EBPα plays a similar role in a human setting.

Collectively, we have shown that C/EBPα collaborate with MLL-ENL to activate a group of genes that, together with Hoxa9 and Meis1, are responsible for the transcriptional changes that underlie the transformation of normal hematopoietic cells into leukemic cells. We have demonstrated a critical role for C/EBPα in MLL-ENL–dependent transformation but not in the maintenance of established MLL-ENL–driven tumors. We could show that this was independent of the C/EBPα-mediated differentiation block upstream of GMPs, but rather depended on the ability of C/EBPα to collaborate with MLL-ENL in the induction of a transcriptional transformation program that includes Hoxa9 and Meis1. Thus, our findings place C/EBPα upstream of the MLL-ENL fusion protein and suggest that C/EBPα facilitates the binding of MLL-ENL/DOT1L, and the subsequent deposition of H3K79me2 at MLL-ENL target genes, during the initial phases of leukemic transformation. This promotes the formation of a stable transcriptional network, which depends on the sustained expression of MEIS1 and HOXA9, but is independent of C/EBPα activity.

Importantly, our findings also have implications for human AML. Thus, we find MLL-rearranged AMLs to be correlated both with a general C/EBPα transcriptional signature, as well as with the immediate C/EBPα-dependent MLL-ENL transformation signature identified in the present work. Our work provides a mechanistic explanation for the lack of CEBPA-null mutations in human AML despite its established role as a myeloid tumor suppressor. Thus, C/EBPα impacts in a dichotomous manner on the development and maintenance of human AML.

Animals were maintained at the Department of Experimental Medicine at University of Copenhagen and housed according to institutional guidelines. All animal work was performed under the approval of Danish Animal Ethical Committee. All mouse lines (Cebpa^fl, Mx1-Cre, and R26-Cre-ER) were backcrossed for at least eight generations onto the C57BL/6 background. Excision of the Cebpa allele was achieved by subjecting 10–12-wk-old Cebpa^fl/fl or Cebpa^fl/fl;Mx1Cre mice to 3 injections with 300 µg pIpC, as described previously (Weischenfeldt et al., 2008). Excision of Cebpa was evaluated by competitive PCR using the primers 5′-GTCCTGCAGCCAGGCAGTGTCCCCACTCACCGCCTTGGAAAGTCACA-3′ and 5′-CCGCGGCTCCACCTCGTAGAAGTCG-3′, which give rise to 355-bp and 560-bp products for the floxed and deleted allele, respectively.

All transplantation assays were performed using the Ly-5 congenic mouse system. Generally, 10,000 GFP-expressing Cebpa^fl/fl or Cebpa^Δ/Δ Ly-5.2 (CD45.2) HSPCs or primary leukemic Cebpa^fl/fl;Mx1Cre or Cebpa^fl/fl cells were transplanted by tail vein injection into 10–12-wk-old lethally (900 cGy) or sublethally (500 cGy) irradiated Ly-5.1 (CD45.1) mice. 2.5 × 10⁵ Ly-5.1 whole BM cells/mouse were given as support to lethally irradiated mice. Cebpa excision in leukemic cells was induced by 3 injections of pIpC at 2 wk after transplantation. BMs from leukemic animals were analyzed at the experimental endpoint. Log-rank test was used to evaluate significance of the survival differences.

The MLL-ENL-IRES-GFP, Hoxa9-IRES-Meis1-neo, and E2A-HLF-neo retroviral constructs were provided by D. Bryder (Lund University, Lund, Sweden), G. Sauvageau (University of Montreal, Montreal, Canada), and R. Slany (University of Erlangen-Nuremberg, Erlangen, Germany), respectively. Viral supernatants were produced by transfection of Phoenix-Eco cells. BM cells were enriched for c-Kit-expression using CD117 MicroBeads according to the manual MACS Cell Separation system (Miltenyi Biotec). c-Kit⁺ HSPCs were prestimulated in RPMI 1640 with 20% fetal calf serum, 20% WEHI-conditioned medium, 20 ng/ml SCF (PeproTech), and 10 ng/ml IL-6 (R20/20; PeproTech). Cells were transduced during two consecutive days by incubating retronectin (Takara Bio)-coated wells with viral supernatant for 2 h (3 times, 40 min) and followed by seeding of 5 × 10⁵ cells/ml in R20/20 medium supplemented with 4 µg/ml protamine sulfate. Transduced cells were plated in methylcellulose medium M3231 (Stem Cell Technologies) supplemented with 1% penicillin-streptomycin (PAA Laboratories), 20 ng/ml SCF (PeproTech), 10 ng/ml human IL-6 (PeproTech), 10 ng/ml GM-CSF (PeproTech), and 10 ng/ml IL-3 (PeproTech). Cells expressing the Hoxa9-IRES-Meis1-PGK-neo construct were selected in 750 µg/ml G418 for 7 d. Colonies were counted every seventh day, and GFP expression was quantified by flow cytometry. Student’s two-tailed t test was used to test for significance.

Cebpa^fl/fl;Cre-ER c-Kit⁺ HSPCs were transduced with MLL-ENL-IRES-GFP, plated in methylcellulose medium (M3231, see above), and replated three times. Cebpa excision was induced in the fully transformed cells by supplementing with 0.1 µM 4-OHT to the methylcellulose medium.

Cebpa^fl/fl;Cre-ER and Cebpa^fl/fl GMP (Lin⁻ Sca1⁻ Kit⁺ CD150⁻ CD41⁻ FcγRII/III⁺) were sorted by FACS and prestimulated in R20/20 medium (Somervaille and Cleary, 2006) supplemented with 1 µM 4-OHT. Cebpa was fully excised after 2 d of 4-OHT treatment, and cells were further transduced with pMIG-empty or pMIG-MLL-ENL, serially replated in semisolid medium, and harvested after three passages.

BM and peripheral blood were stained with antibodies and run on a FACSCalibur (BD), LSRII (BD), or FACSAria (BD) and analyzed using the FlowJo software. For analyzing and sorting of hematopoietic progenitors B220, CD3, CD11b, Gr1, Ter119, CD105, FcgRII/III, CD41, Sca-1, c-Kit (eBioscience), and CD150 (BioLegend) were used as described previously (Hasemann et al., 2012). Student’s two-tailed t test was used to test for significance.

Hoxa9, Meis1, and Cebpa expression was quantified relative to β-actin in sorted GFP⁺ Cebpa^fl/fl or Cebpa^Δ/Δ preGM cells transduced with empty vector or MLL-ENL. Each sample was analyzed in triplicates on the same 96-well plate using the LightCycler 480 (Roche). Expression levels were determined by SYBR Green (Roche; Table S3).

c-Kit⁺ Cebpa^fl/fl or Cebpa^Δ/Δ cells were transduced over two consecutive days, and GFP⁺ and GFP-preGM cells (Lin⁻ Sca1⁻ Kit⁺ CD150⁻ CD41⁻ FcgRII/III⁻ CD105⁻) were sorted 72 h after the first transduction. Total RNA was subsequently purified, amplified using the Ovation Pico WTA system (NuGen), labeled, and hybridized to the Mouse Gene 1.0 ST GeneChip Array (Affymetrix). After RMA normalization, the four sample types were subjected to Limma analysis. For each pairwise phenotype comparison, genes were selected using the following criteria: log₂ fold change > 0.58; P < 0.05, moderated t-statistics, corrected for multiple testing (Bonferroni), Limma package (Smyth, 2004). We defined MLL-ENL–induced genes based on differences between the GFP⁺ and GFP⁻ Cebpa^fl/fl samples. For the C/EBPα-dependent subset, we also required these to be significantly down-regulated (P < 0.05) in the GFP⁺ Cebpa^Δ/Δ vs. GFP⁺ Cebpa^fl/fl samples. Raw gene expression data are available at the Gene Expression Omnibus online database under accession no. GSE46534.

Microarray dataset from patients with AML with t(11q23)/MLL were downloaded from GEO (accession nos. GSE13159 and GSE14468; Wouters et al., 2009; Haferlach et al., 2010). Microarrays from normal healthy human GMPs were downloaded from GEO accession no. GSE24006 (Gentles et al., 2010). Datasets were normalized using RMA (Irizarry et al., 2009). The median gene expression for the MLL-ENL–induced genes in cells from leukemic patients was compared with median gene expression in healthy normal GMPs.

We used GSEA (Subramanian et al., 2005) to identify gene signatures that were altered in murine GFP⁻ Cebpa^Δ/Δ preGMs versus GFP⁻ Cebpa^fl/fl preGMs, MLL-ENL–transduced GFP⁺ Cebpa^Δ/Δ preGMs versus GFP⁻ Cebpa^fl/fl preGMs, and human AML-MLL phenotype versus normal healthy GMPs. Gene sets originated from the MSigDB (www.broadinstitute.org/gsea/msigdb).

ChIP-seq was performed in replicates using BM cells from Cebpa^fl/fl or Cebpa^Δ/Δ mice, and ChIP-qPCR was performed in triplicate in sorted preGM cells from Cebpa^fl/fl or Cebpa^Δ/Δ ± MLL-ENL. Chromatin from 100,000 preGMs or 500,000 GMPs was incubated with antibodies for histone marks (H3K79me2,ab3594 [Abcam]; H3K4me3, C42D8 [Cell Signaling Technology]; and H3K27me3, C36B11 [Cell Signaling Technology]) or C/EBPα (14AA ; Santa Cruz Biotechnology, Inc.), respectively. The antibody-bound chromatin was captured with Protein A–Sepharose beads, washed, de-cross-linked, and precipitated. Precipitated DNA was quantified with qPCR (see primers in Table S3) or mixed with 2 ng Escherichia Coli DNA and amplified using NEB Next ChIP-seq sample prep reagent set 1 (New England Biolabs) according to the manufacturer’s protocol. Libraries were sequenced on an Illumina Hiseq2000. Data were deposited in the GEO under accession no. GSE47003.

All reads were mapped using bowtie 0.12.7 (Langmead et al., 2009) using standard parameters. The C/EBPα, H3K4me3, and H3K27me3 ChIP-seqs were performed as biological replicates, and the correlation coefficients, r², were calculated to the following: Cebpa^fl/fl H3K27me3, 0.938; Cebpa^Δ/Δ H3K27me3, 0.901; Cebpa^fl/fl H3K4me3, 0.896; Cebpa^Δ/Δ H3K4me3, 0.922; and Cebpa^+/+ C/EBPα, 0.805. Replicate C/EBPα peaks were called using macs2 with the following settings: macs2 callpeak -t GMP_peak_file.tagAlign -c IgG.tagAlign -f BED -g mm -p 1e-3–to-large. Overlapping peaks (P < 10⁻¹⁴ in both peak callings) were further filtered based on peak height and regions with high enrichment in the IgG samples removed. Genes were defined using the mm9 RefSeq set, taking the longest supported isoform for genes with multiple isoforms. Distances were calculated from TSS to the summit of the closest C/EBPα peaks. P-values were calculated using the Wilcoxon rank sum test.

Table S1 shows gene expression of GFP⁺ and GFP⁻ Cebpa^Δ/Δ and Cebpa^fl/fl preGM cells transduced for 72 h with MLL-ENL-IRES-GFP. Table S2 shows the distance to closest C/EBPa peak of MLL-ENL immediate genes. Table S3 shows the primers used for qPCR.

We thank Anna Fossum for help with cell sorting and Inge Damgaard for technical assistance.

This study was supported by the Danish Cancer Society, the Danish Research Council for Strategic Research, and through a center grant from the NovoNordisk Foundation (The Novo Nordisk Foundation Section for Stem Cell Biology in Human Disease).

The authors have no competing financial interests.