Graphical Abstract

   graphic file with name fx1.jpg
   [33]Open in a new tab

Highlights

     * •
       Deletion of Hif-1α accelerates the progression of
       chemotherapy-treated MLL-AF9-AML
     * •
       Deletion of Hif-1α does not decrease LSC frequency after
       chemotherapy
     * •
       Chemotherapy targets more mature cells indicated by transcriptional
       analysis
     * •
       Hif-1α deletion affects few transcriptional pathways in AML cells
     __________________________________________________________________

   In this article, Bryder, Velasco-Hernandez and colleagues show that the
   combination of chemotherapy and Hif-1α deletion in a murine model of
   MLL-AF9-AML leads to a faster progression of the disease without
   altering LSCs frequency. Single-cell transcriptional analysis indicates
   that Hif-1α deletion leads to few transcriptional modifications.

Introduction

   Acute myeloid leukemia (AML) is a clonal disorder characterized by a
   rapid accumulation of differentiation-arrested myeloid blasts.
   Remission can be achieved in most patients by a combination of
   cytarabine (AraC) and anthracycline therapy. However, few treatment
   improvements have arisen over the past decades ([34]Rowe and Tallman,
   2010) and 40%–60% of patients still relapse ([35]Burnett et al., 2011,
   [36]Yanada et al., 2008), which is associated in general with limited
   treatment options ([37]Dohner et al., 2010).

   Relapse is attributed to a minor subpopulation of cells referred to as
   leukemic stem cells (LSCs). LSCs are resilient to cytotoxic effects of
   chemotherapy, via mechanisms that include resistance to apoptosis,
   increased capacities to efflux drugs, and relative quiescence
   ([38]Thomas and Majeti, 2017). This quiescence in turn underlies
   resistance to compounds that target energy metabolism ([39]Essers and
   Trumpp, 2010).

   Cellular quiescence is typically related to activation of
   hypoxia-signaling pathway, driven by hypoxia-inducible factor (HIF)
   complex, which constitutes a family of 3 heterodimeric transcription
   factors HIF-1, HIF-2, and HIF-3 ([40]Flamme et al., 1997, [41]Makino
   et al., 2002, [42]Semenza and Wang, 1992). The oxygen-dependent
   regulation of HIFs depends on stabilization of an associated α subunit.
   Hence, at oxygen levels above 5%, the α subunit is proteasomally
   degraded. In contrast, under hypoxic conditions the α subunit is
   stabilized, dimerizes with constitutively expressed β subunit, and
   promotes transcription of target genes regulating key cellular
   processes such as angiogenesis, proliferation, metabolism, and
   apoptosis ([43]Semenza, 2003).

   The most primitive hematopoietic stem cells (HSCs) reside in hypoxic
   niches in the bone marrow (BM) ([44]Morrison and Scadden, 2014).
   Thereby, HSCs are kept dormant and free of genetic changes that mostly
   occur during DNA replication. However, while HIF-1α signaling was
   reported to critically regulate HSC maintenance ([45]Takubo et al.,
   2010), other studies showed that depletion of HIF-1α/HIF-2α had no such
   effect ([46]Guitart et al., 2013). The HIF pathway has also been
   proposed to play important roles in AML, a notion supported by the
   preferential localization of chemotherapy-resistant AML cells to the
   hypoxic endosteal niche in the BM ([47]Ishikawa et al., 2007).
   Accordingly, several studies have shown that loss of HIF-1/2 leads to
   abrogation of LSCs ([48]Rouault-Pierre et al., 2013, [49]Wang et al.,
   2011, [50]Zhang et al., 2012). Contrasting these results are studies
   demonstrating that loss of HIF-1/2 does not impact on mouse models of
   AML, or alternatively can give rise to an even more severe leukemic
   phenotype ([51]Velasco-Hernandez et al., 2014, [52]Vukovic et al.,
   2015). Nonetheless, targeting hypoxia and HIFs has been considered a
   key anticancer approach ([53]Frolova et al., 2012, [54]Rouault-Pierre
   et al., 2013, [55]Wang et al., 2011).

   Here, we investigated the possibility to sensitize LSCs from a mouse
   model of AML to a clinically relevant chemotherapy regimen by targeting
   the hypoxia pathway via Hif-1α. Contrary to expectation, our work
   suggests a more aggressive disease outcome upon Hif-1α deletion, which
   challenges the general use of hypoxia targeting for therapeutic benefit
   in AML.

Results

Deletion of Hif-1α Accelerates the Progression of Chemotherapy-Treated
Leukemia

   To study effects of Hif-1α on the sensitivity of LSCs to chemotherapy,
   we bred Hif-1α^fl mice ([56]Ryan et al., 2000) with Rosa26Cre-ER^T2
   mice ([57]Ventura et al., 2007), generating a model that allows for
   Hif-1α deletion upon tamoxifen treatment ([58]Figure S1A). c-Kit^+
   cells from Hif-1α^fl/fl;Rosa26Cre-ER^T2/+ mice were transduced with
   MLL-AF9-GFP retrovirus and transplanted into mice, which led to
   development of MLL-AF9-driven AML. Primary leukemias were next used to
   study AML progression in secondary hosts.

   After comparing two different chemotherapy protocols ([59]Figure S1B),
   we used the one described by [60]Zuber et al. (2009), which resulted in
   severely decreased white blood cells, erythrocytes, and platelets, akin
   to what is observed in human patients. We then titrated the latency of
   the disease using different leukemic cell doses ([61]Figure S1C) and
   chose a dose of 10^4 cells, which ensured a sufficiently long latency
   for posterior procedures. Cells from individual MLL-AF9-leukemic mice
   were transplanted into new recipients and subjected to Hif-1α deletion
   10 days after transplantation, and to a standard chemotherapeutic
   protocol at day 20. As a control, we included mice injected with both
   vehicles but receiving no leukemic cells ([62]Figure 1A).

Figure 1.

   [63]Figure 1
   [64]Open in a new tab

   Deletion of Hif-1α Leads to a Faster Progression of AML after
   Chemotherapy Treatment

   (A–H) Experimental design for the generation of leukemic clones,
   deletion of Hif-1α, and applied chemotherapy protocol (A). At day 28
   after transplantation, mice were sacrificed and disease evolution was
   examined according to different parameters: total white blood cell
   count in PB (B), myeloid cells (Gr1^+-CD11b^+) (C), GFP^+ cells in PB,
   BM, and spleen (D and E), splenomegaly (F), GFP^+ c-Kit^+ cells in BM
   and spleen (G), and GFP^+ Lin^– c-Kit^+ in BM and spleen (H). Results
   are from 1 of 2 experiments and depict results from 4 mice per group (3
   for wild-type control mice). Plots represent mean ± SEM. ^∗p < 0.05,
   ^∗∗p < 0.01, ^∗∗∗p < 0.01. AraC, cytarabine; Dox, doxorubicin; CTX,
   chemotherapy; BM, bone marrow; Ctrl, control.

   See also [65]Figures S1 and [66]S2.

   Three days after the last injection, we started to observe mortality in
   the PBS group ([67]Figure S2A). We collected peripheral blood (PB), BM,
   spleen, and liver samples from half of the cohort and analyzed multiple
   disease parameters ([68]Figures 1B–1F and [69]S2). As expected, disease
   progressed more slowly in the chemotherapy-treated (CTX) mice
   ([70]Figures 1B–1F). More intriguingly, we observed a faster disease
   evolution upon Hif-1α deletion. This was seen in both CTX and PBS
   groups ([71]Figures 1C–1F). When we assessed phenotypically more
   primitive GFP^+ c-Kit^+ or GFP^+ Lin^– c-Kit^+ cells, we observed a
   higher abundance in the Hif-1α-deleted (Hif-1α^–/–) than in the
   Hif-1α-intact (Hif-1α^+/+) group after chemotherapy ([72]Figures 1G,
   1H, and [73]S2). While these populations were very infrequent in BM of
   PBS groups, with no clear differences between Hif-1α^–/– and Hif-1α^+/+
   cells, a higher number of such cells upon Hif-1α deletion were observed
   in the spleens of the same animals. Taken together, these results
   demonstrate an increment of potential LSCs (c-Kit^+) upon Hif-1α^–/–
   deletion, which was particularly significant after chemotherapy.

Deletion of Hif-1α Does Not Decrease the Frequency of LSCs after Chemotherapy
Treatment

   To evaluate the impact of chemotherapy and Hif-1α deletion on LSCs more
   directly, we performed limiting dilution analysis (LDA)
   ([74]Figure 2A).

Figure 2.

   [75]Figure 2
   [76]Open in a new tab

   Deletion of Hif-1α Does Not Decrease LSC Frequency after Chemotherapy

   (A) Experimental design for the analysis of LSC frequency.

   (B) Limiting dilution analysis (LDA) from in vitro data. Variable
   numbers of GFP^+ cells (4, 8, or 12 cells) from the 3 experimental
   groups were cultured in vitro under normoxic or hypoxic conditions and
   proliferative capacity was evaluated after 10 days. LSC frequencies
   were calculated using ELDA software. Plots show 1 representative of 3
   independent experiments (cells from 3 different donors) with 96
   replicates per group and condition.

   (C) Survival curves of mice transplanted with 10 or 100 GFP^+ cells
   from mice treated with the combination of tamoxifen + chemotherapy
   (KO-CTX), tamoxifen + PBS (KO-PBS), or oil + chemotherapy (wt-CTX) (n =
   4–6 mice per group from 1 experiment).

   (D) LDA from in vivo data. Recipient mice were injected intravenously
   with variable numbers of GFP^+ cells from the 3 experimental groups
   (KO-CTX, KO-PBS, and wt-CTX). Mice survival was evaluated 2 months
   after transplantation, and frequency of LSCs was calculated using ELDA
   software.

   (E) Stem cell frequency of LSCs from the different experiments: hypoxia
   in vitro (red, n = 3), normoxia in vitro (blue, n = 3), and in vivo
   (green, n = 1) experiments. ^∗p < 0.05, two-way ANOVA.

   (F) LSC total numbers from the different experiments: hypoxia in vitro
   (red, n = 3), normoxia in vitro (blue, n = 3), and in vivo (green, n =
   1) experiments. Plot shows mean ± SEM. ^∗p < 0.05, ^∗∗p < 0.01, two-way
   ANOVA. CTX, chemotherapy; KO, Hif-1α^–/–; wt, Hif-1α^+/+.

   For an in vitro approach, we sorted 4, 8, or 12 GFP^+ leukemic BM cells
   from each group into individual wells. Proliferation was next evaluated
   in both normoxic (20% oxygen) and hypoxic (1% oxygen) conditions
   ([77]Figures 2B and 2E). We obtained the highest LSC frequency from
   PBS-Hif-1α^–/– cells (1/12.3 and 1/21.7 in hypoxia or normoxia,
   respectively, means from 3 independent experiments). Within CTX-treated
   cells, Hif-1α^–/– samples contained a higher LSC frequency compared
   with Hif-1α^+/+ cells (1/26.8 versus 1/47.1 in hypoxia, and 1/40.4
   versus 1/94.4 in normoxia, respectively, means from 3 independent
   experiments), although this failed to reach statistical significance.

   We next assessed LSC frequencies in vivo, injecting 10, 10^2, 10^3, or
   10^4 GFP^+ BM cells from each group into new recipients
   ([78]Figure 2C). While transplantation of 10^3 and 10^4 cells resulted
   in 100% mortality, lower doses of 10 and 10^2 cells indicated an LSC
   frequency highest in the PBS-Hif-1α^–/– group (1/18.8). Among
   CTX-treated samples, we observed a higher LSC frequency in the
   Hif-1α^–/– group (1/34.5) than in the Hif-1α^+/+ group (1/56.1)
   ([79]Figures 2D and 2E).

   Estimation of total number of LSCs in the BM ([80]Figure 2F)
   demonstrated that deletion of Hif-1α fails to reduce LSC
   frequencies/numbers after chemotherapy.

Hif-1α Deletion Affects Transcriptional Expression of Replication,
Transcription, and Translation-Related Genes

   Our data indicated that Hif-1α deletion contributed to a more rapid
   disease progression ([81]Figure 1). To tease out a potential mechanism,
   we therefore conducted single-cell RNA sequencing on leukemic cells
   from the 4 evaluated settings (Hif-1α^–/–/Hif-1α^+/+ ± chemotherapy)
   ([82]Figure 3A). Aggregated data were subjected to dimensionality
   reduction using the t-distributed stochastic neighbor embedding (t-SNE)
   method ([83]Figure 3B), revealing dramatically different
   transcriptional profiles of cells following chemotherapy, but less
   different according to Hif-1α status. This indicates either a
   modification of transcription upon chemotherapy or, more likely, a
   selective survival regardless of Hif-1α status.

Figure 3.

   [84]Figure 3
   [85]Open in a new tab

   Single-Cell RNA Sequencing of Leukemic Cells after Chemotherapy
   Treatment

   (A) Experimental design for single-cell RNA-sequencing analysis. After
   treatment with tamoxifen/oil and chemotherapy/PBS, GFP^+ cells were
   collected at day 28 after transplantation and subjected to RNA
   processing.

   (B) t-SNE plots representing gene expression profiles of all individual
   cells analyzed: KO-CTX (n = 1,217 cells), KO-PBS (n = 1,696 cells),
   wt-CTX (n = 1,301 cells), and wt-PBS (n = 1,450 cells). Each dot
   represents one cell. K-means clustering groups cells into 10 clusters,
   which are represented by different colors. Differentially expressed
   genes in each cluster were correlated with the main biological function
   indicated at the right side of the plots.

   (C) Heatmap depicting significantly differentially expressed genes in
   single cells. Heatmap with the full set of genes can be found in
   [86]Figure S3. Hif-1α target genes are indicated by black dots.

   (D) Venn diagrams showing the distribution among groups of
   differentially expressed genes (adjusted p value <0.05). Of note, there
   are no overlapping genes between chemotherapy- and PBS-treated groups
   among this set of differentially expressed genes.

   (E) Gene ontology (GO) and KEGG pathway enrichment analysis from
   uniquely differentially expressed genes in each group indicating the
   associated biological process.

   CTX, chemotherapy; IL1/TNF, interleukin-1/tumor necrosis factor; INF,
   interferon; dim, dimension; FDR, false discovery rate.

   See also [87]Figures S3 and [88]S4.

   By K-means clustering, we divided the entire dataset into 10 different
   clusters. We observed that clusters with functions related to more
   differentiated cells (e.g., immune system process, response to
   interleukin-1/tumor necrosis factor, innate immunity or response to
   virus/interferon) were predominant in PBS samples, while others (e.g.,
   mitotic cell cycle, cellular response to DNA damage, and
   transcriptional regulation) were associated with CTX samples. Multiple
   genes related to myeloid differentiation were predominantly expressed
   in PBS samples (data not shown), strongly suggesting effective
   chemotherapy targeting of more mature leukemia-associated cells.

   Pairwise sample comparisons produced a collection of 331 differentially
   expressed genes ([89]Figures S3 and [90]3C). When comparing
   significantly overexpressed genes among groups (adjusted p value
   <0.05), we found a very different distribution of cells between CTX and
   PBS samples, but only a small subset of genes uniquely expressed in
   either Hif-1α^–/– or Hif-1α^+/+ cells ([91]Figures 3D and [92]S4). In
   single-cell analysis, this indicates that there are differences in
   transcriptional regulation or that analyzed cells from each sample are
   different.

   To identify functional protein association networks among
   differentially expressed genes, we used the STRING database
   ([93]Figures 3E and [94]4). For the 4 sets of differentially expressed
   genes (log fold change >0.4, adjusted p value <0.05), we obtained a
   significant protein-protein interaction (PPI) enrichment score,
   indicating that there are more interactions among these proteins than a
   random set of proteins from the genome (PPI enrichment p values:
   Hif-1α^–/–-CTX = 0; Hif-1α^+/+-CTX = 1.03 × 10^−6; Hif-1α^–/–-PBS = 0;
   Hif-1α^+/+-PBS = 0.0251). This indicates that genes in each set are at
   least partially biologically connected. To correlate these genes with a
   biological function, we performed gene ontology and KEGG pathway
   enrichment analyses ([95]Figure 3E). This revealed that mRNA
   processing, DNA replication, and protein folding were significantly
   overrepresented in the Hif-1α^–/–-CTX group, while the Hif-1α^+/+-CTX
   group was associated with the HIF-1α pathway, glycolysis, and other
   carbon metabolism processes. In PBS groups, translation was the most
   represented pathway associated with Hif-1α^–/– cells, while no
   significant pathways could be assigned to Hif-1α^+/+ cells.

Figure 4.

   [96]Figure 4
   [97]Open in a new tab

   Protein Association Networks Affected by Chemotherapy

   Protein association network analysis of differentially expressed set of
   genes (average log fold change >0.4, adjusted p value <0.05). Uniquely
   expressed genes from each group are indicated by its assigned color.
   Common genes among KO and wt samples, within each specific set (CTX- or
   PBS-treated), are shown in dark gray. Associated biological function to
   each gene is indicated, with the color of the border showing the main
   processes regulated by these genes. Interaction enrichment score >0.5.
   CTX, chemotherapy.

Discussion

   Relapse of AML remains a major therapeutic challenge, not least because
   of the few therapeutic options for many of these patients, which in
   most cases is restricted to palliative care.

   One potential strategy for eradicating LSCs, responsible for relapse in
   AML, is to sensitize them to chemotherapy. We here explored whether
   targeting the HIF pathway could be used for this using an MLL-AF9
   murine AML model, which was subjected to a standard and clinically
   relevant chemotherapy protocol. In our work, Hif-1α^–/– cells displayed
   a faster evolution of the disease compared with Hif-1α^+/+ cells
   following chemotherapy. While the results are consistent with our
   previous work ([98]Velasco-Hernandez et al., 2014) and recently
   published data ([99]Vukovic et al., 2015), they appear in sharp
   contrast to previous studies ([100]Wang et al., 2011, [101]Zhang
   et al., 2012) in which Hif-1α was critical for LSCs maintenance.

   In our work, we used the Rosa26Cre-ER^T2 rather than the more commonly
   used Mx1-Cre model, which is associated with significant leakiness
   ([102]Velasco-Hernandez et al., 2016). This allowed us to study the
   effect of Hif-1α inhibition specifically in AML cells of the same
   founder clone, which circumvents the risk that different leukemic
   clones have intrinsically different behavior. Inducing the deletion in
   Rosa26Cre-ER^T2 mice requires injection of tamoxifen that might itself
   affect the initiation of MLL-AF9 leukemia ([103]Sanchez-Aguilera
   et al., 2014). However, as the described effect of tamoxifen is a
   regression of the disease, the isolated effect of knocking out Hif-1α
   could be even more pronounced than the one we report.

   The standard chemotherapy regimen for AML is a combination of AraC and
   an anthracycline (such as doxorubicin). Doxorubicin has itself been
   attributed Hif-inhibiting properties ([104]Lee et al., 2009). Different
   disease progression in CTX-Hif-1α^–/– and Hif-1α^+/+ samples indicates
   that even if doxorubicin is inhibiting HIF function, this effect is not
   complete and can be further increased by alternative means.

   There is evidence that c-Kit deregulation and overexpression could be
   factors contributing to the chemotherapy resistance in AML ([105]Advani
   et al., 2008). The higher numbers of c-Kit^+ cells in the Hif-1α^–/–
   chemotherapy group might therefore be an indicator of higher drug
   resistance of these cells, with c-Kit being a well-described stem cell
   marker of murine MLL-AF9-driven LSCs ([106]Krivtsov et al., 2006).
   However, our c-Kit expression data correlated poorly with LSC
   frequencies measured by LDA. Previously it was thought that
   chemotherapy promoted LSC enrichment ([107]Ishikawa et al., 2007,
   [108]Saito et al., 2010). However, although cytarabine decreases the
   frequency of LSCs in human AML models ([109]Griessinger et al., 2014),
   it fails to promote enrichment either of quiescent cells or more
   primitive CD34^+CD38^– cells, emphasizing that functional assessments
   are critical. We did not observe an increment in LSC frequencies
   measured by LDA but observed a substantial reduction (around 3-fold) of
   the total bulk of leukemic cells, indicating an overall enrichment of
   LSCs following chemotherapy. In addition, we observed an increased
   frequency of c-Kit^+ cells by Hif-1α deletion, which further supports
   the increased aggressive phenotype of Hif-1α^–/– cells.

   Hif-1α is rapidly degraded in normoxia and Hif-1α^+/+ cells should thus
   theoretically behave as Hif-1α^–/– cells, which would suggest similar
   LSC frequencies between Hif-1α^–/– and Hif-1α^+/+ cells in normoxia.
   Somewhat surprisingly, we observed a higher LSC frequency of Hif-1α^–/–
   cells, although not significant (p = 0.1079). One possibility is that
   this indicates the involvement of one or more Hif-1α-oxygen-independent
   mechanisms. For instance, it is well established that Hif-1α can be
   induced by the PTEN/PI3K/AKT axis ([110]Zundel et al., 2000), a pathway
   that appears to extend also to MLL-AF9 leukemia ([111]Hoshii et al.,
   2012). Regardless, even if hypoxia in general promotes a quiescent
   state of cells, such conditions seem to promote AML cell proliferation,
   which is also in line with previous results on human AML
   ([112]Griessinger et al., 2014).

   After chemotherapy, we observed a more aggressive phenotype as a
   consequence of Hif-1α deletion. We hypothesized that this faster
   development of disease could be explained by two scenarios: (1)
   leukemic Hif-1α^–/– cells are less sensitive to the chemotherapy, which
   would be in contrast to our original expectation; and/or (2) leukemic
   Hif-1α^–/– cells expand faster than Hif-1α^+/+ cells after
   chemotherapy. To discriminate between these two options and unveil
   possible mechanisms leading to this effect, we conducted a single-cell
   transcriptome analysis.

   We found that chemotherapy-treated cells displayed a distinct
   transcriptomic profile when compared with PBS-treated cells, but with
   fewer differences when comparing Hif-1α^–/– and Hif-1α^+/+ cells.
   Still, of the significantly overexpressed genes specific for each
   group, we identified an increment in genes involved in translation
   (PBS-Hif-1α^–/– cells) and in mRNA processing, DNA replication, and
   protein folding (CTX-Hif-1α^–/– cells). Overall, this is in line with a
   more active proliferative status of Hif-1α^–/– cells and suggests that
   deletion of Hif-1α is transcriptionally deregulating these
   aforementioned specific pathways. Chemotherapy-depleted cells with an
   immune-related phenotype represent the cells expanded in leukemia. As
   expected, Hif-1α^–/– cells were depleted in expression of Hif-1α
   targets, and Hif-1α deletion leads to an enhanced transcription of
   genes involved mainly in replication, mRNA processing, and translation.

   These results are in accordance with the described effect of Hif-1α,
   inhibiting cell cycle and promoting quiescence, actions attributed to
   blockage of c-MYC and enhancement of cyclin-dependent kinase inhibitors
   such as p21/CDKN1A ([113]Koshiji et al., 2004). Since we are deleting
   only one member of a multigene family, it remains a possibility that
   other members, i.e., Hif-2α, could have a compensatory effect. Future
   work could benefit from an increased understanding of both gene
   redundancy and potential hypoxia-independent effects of Hif-1α.

   Overall, our data suggest that Hif-1α inhibition failed to improve the
   outcome of chemotherapy in MLL-AF9-driven AML and, in contrast, led to
   a faster progression of disease upon withdrawal of the treatment.
   Further investigations are needed to extend these conclusions to other
   genetic subtypes of AML.

   Deletion of Hif-1α in our setting was permanent, which would be a
   different scenario than in a clinical setting where administration of
   Hif-1α inhibitors would be transient. Our results nevertheless
   emphasize that the effects of HIF inhibition have to be further
   investigated before this strategy can be applied in a clinical setting.

Experimental Procedures

Mice

   Hif-1α^lox mice ([114]Ryan et al., 2000) (JAX 007561) were crossed with
   the tamoxifen-inducible Rosa26Cre-ER^T2 mice ([115]Ventura et al.,
   2007) (JAX 008463) to generate a combined conditional knockout (KO)
   model. Mice were maintained at the animal facility of the Biomedical
   Center at Lund University (Sweden) and all experiments were performed
   with consent from a local ethics committee.

Statistical Analysis

   All data are expressed as the mean ± SEM. Differences between groups
   were assessed using unpaired Student's t tests. All analyses were
   performed with Prism software, version 7.0 (GraphPad Software).

   Detailed descriptions of experiments can be found in [116]Supplemental
   Experimental Procedures.

Author Contributions

   T.V.-H. designed the study, performed experiments, analyzed data, and
   wrote the manuscript; S.S. performed bioinformatics analysis; I.H. and
   E.E. performed experiments; and J.C and D.B. designed the study and
   wrote the manuscript. All authors revised the manuscript and approved
   the final version of this study.

Acknowledgments