MYCN and HIF-1 directly regulate TET1 expression to control 5-hmC gains and enhance neuroblastoma cell migration in hypoxia

ABSTRACT

Ten-Eleven-Translocation 5-methylcytosine dioxygenases 1–3 (TET1-3) convert 5-methylcytosine to 5-hydroxymethylcytosine (5-hmC), using oxygen as a co-substrate. Contrary to expectations, hypoxia induces 5-hmC gains in MYCN-amplified neuroblastoma (NB) cells via upregulation of TET1. Here, we show that MYCN directly controls TET1 expression in normoxia, and in hypoxia, HIF-1 augments TET1 expression and TET1 protein stability. Through gene-editing, we identify two MYCN and HIF-1 binding sites within TET1 that regulate gene expression. Bioinformatic analyses of 5-hmC distribution and RNA-sequencing data from hypoxic cells implicate hypoxia-regulated genes important for cell migration, including CXCR4. We show that hypoxic cells lacking the two MYCN/HIF-1 binding sites within TET1 migrate slower than controls. Treatment of MYCN-amplified NB cells with a CXCR4 antagonist results in slower migration under hypoxic conditions, suggesting that inclusion of a CXCR4 antagonist into NB treatment regimens could be beneficial for children with MYCN-amplified NBs.

Key policy highlights

• In MYCN-amplified neuroblastoma cell lines, MYCN directly controls TET1 expression in normoxia.

• In MYCN-amplified neuroblastoma cell lines exposed to hypoxia, HIF-1 augments TET1 expression and TET1 protein stability.

• Hypoxic MYCN-amplified neuroblastoma cell lines have increased cell migration, mediated by genes including CXCR4 that gain 5-hydroxymethylcytosine density.

• Treatment of MYCN-amplified NB cells with a CXCR4 antagonist slows hypoxia-associated migration, suggesting a CXCR4 antagonist could be beneficial in treatment regimens for children with MYCN-amplified neuroblastomas.

KEYWORDS:

• MYCN
• hypoxia
• TET1
• neuroblastoma

Introduction

Most commonly diagnosed in children 5 years and younger, neuroblastoma (NB) tumours originate from the developing sympathetic nervous system, with a predilection for the adrenal gland. NBs exhibit a broad range of clinical behaviours, with some tumours resolving spontaneously, whereas others progress despite intensive medical and surgical intervention [1]. High-risk patients with aggressive tumours are identified based on several diagnostic criteria, including MYCN status, age, stage, and grade [1]. Currently, high-risk patients receive a combination of systemic chemotherapy, surgical resection, tandem autologous transplant, radiation, and immunotherapy as part of established treatment protocols [1]. Although improvements in clinical outcomes have been observed during the past decades, the 5-year survival of high-risk patients remains at less than 60% [2], emphasizing a need for more effective therapies.

Next-generation sequencing has allowed researchers to analyse large datasets of NB genomes and transcriptomes, with the aim of identifying novel genetic biomarkers and therapeutic targets [3,4]. However, results from sequencing studies have shown that NBs have a relatively low mutational burden compared to other cancers [5,6], with only a few recurrent mutations [6]. The paucity of recurrent somatic genetic alterations has driven a search for epigenetic modifications that contribute to NB pathogenesis [5,7].

NBs arise from neural crest tissue, which is known to have high levels of 5-hydroxymethylcytosine (5-hmC) epigenetic marks [8]. When neural crest cells undergo neuronal differentiation, 5-hmC increases and is enriched in the gene bodies of activated genes, implying 5-hmC is an important epigenetic mark associated with transcription [8,9]. The 5-hmC base functions both as a stable and distinct epigenetic modification associated with open chromatin and active gene transcription [10–12] as well as an intermediate in cytosine demethylation [13,14]. 5-hmC is generated from the oxidation of 5-methylcytosine (5-mC) by three TET enzymes, which use oxygen and α-ketoglutarate in the catalytic reaction and Fe(II) as a co-factor [9].

Solid tumours, including NB, are often characterized by a lack of organized vasculature that generates hypoxic regions in the tumour [15]. The hypoxia-inducible factor (HIF) family of transcription factors (TFs), HIF1/2/3, are the master regulators of the hypoxic response [16]. All the family members are alpha-beta heterodimers with both subunits constitutively transcribed and translated [15]. In a normal oxygen environment, oxygen-dependent prolyl hydroxylases bind and hydroxylate two proline residues in the HIF alpha subunit, marking it for ubiquitination and proteasomal degradation [15,16]. In contrast, in a hypoxic environment, the lack of free oxygen lowers prolyl hydroxylase activity, resulting in accumulation of the alpha subunit. Once the alpha-beta heterodimer forms, it localizes to the nucleus and promotes target gene expression that aids in metabolism, proliferation, angiogenesis, and invasion that characterizes the hypoxic tumour phenotype [15]. When NB cell lines are exposed to hypoxia, they develop an aggressive phenotype, consisting of enhanced proliferation, migration, and invasion, all of which are specifically mediated via HIF-1 [17].

Previously, we described how global 5-hmC increases are essential for induction of the hypoxic transcriptional response specifically in MYCN-amplified and tumorigenic NB cell lines [12]. In hypoxia, HIF-1 upregulates TET1 expression, which drives an increase in 5-hmC levels along the gene bodies of hypoxic response genes, augmenting their expression. Here, we show that MYCN and HIF-1 TFs directly upregulate TET1 in normoxia and hypoxia, respectively, describing a novel interaction between HIF-1α and TET1 proteins, and identify a 5-hmC target, CXCR4, as a mediator of the increased migratory phenotype characteristic of hypoxia.

Materials and methods

Cell culture and CRISPR-Cas9 genome editing

NB cell lines were provided by the Cohn Laboratory and were authenticated by STR profiling and tested negative for Mycoplasma. SK-N-BE [2], NBL-WN, LA1-55 n, NBL-S, SH-SY5Y, LA1-5S, NBL-WS, and SHEP cells were cultured in RPMI with 10% foetal bovine serum (FBS). Tet-21/N cells [18] were cultured in RPMI with 10% Tet system FBS (Clontech 631,106). Normoxic culture was done at 37°C under 21% O₂ and 5% CO₂ in a humidified incubator. For hypoxic exposure, the cells were incubated under 1% O₂ and 5% CO₂ in a humidified chamber. Tet-21/N cells were maintained with 1 μg/mL doxycycline, and MYCN induction was induced by the removal of doxycycline from the media.

A CRISPR-Cas9 system was used to perform genome editing. First, gRNAs were designed and evaluated with the programme CRISPOR [19]. For NTC, a guide was used that does not target anywhere in the human genome (Table S1). gRNAs were ligated into plasmid Lenticrispr v2 (Addgene plasmid #52961) following the Zhang laboratory protocol [20]. To produce lentiviral particles, 293 T cells were transfected according to the Addgene protocol adapted from [21]. The same protocol was used to transduce target cells with the lentivirus, but to accommodate the low tolerance of NB cells, 4 μg/mL of polybrene was added to the media [21]. To genotype each clone, DNA was extracted using phenol:chloroform, and PCR using GoTaq (Promega PRM3005) was performed with primers targeting the region surrounding the edited site (Table S1). Sanger sequencing of the amplicon was then used to define the mutation within each clone. For each target, two clones that lacked the TF binding site but had different deletions were selected for subsequent experiments.

Immunoprecipitation and Western blotting

Nuclear protein extraction was performed via high salt fractionation [22], and proteins were separated via SDS-PAGE on 6% acrylamide gels. TET1 immunoprecipitation was performed using the α-TET1 antibody (Santa Cruz sc-293,186). PVDF membranes were probed with primary antibody: α-TET1 (Genetex GT1462) at 1:1000 dilution overnight at 4°C; α-TOP1 (Abcam ab109374) at 1:1000 dilution for 1 hour at room temperature; or α-MYCN (Abcam ab16898) at 1:1000 dilution for 1 hour at room temperature. After primary antibody incubation, the membranes were incubated with secondary antibodies at 1:5000 dilution against the appropriate species: α-Rabbit IgG (Millipore 401,393–2ML) or α-Mouse IgG (Cell Signaling Technology 70,765) for 1 hour at room temperature. Protein degradation assays in SK-N-BE [2] cells were performed following the protocol outlined in [23]. Protein quantification analysis was done with ImageJ [24].

Chromatin-Immunoprecipitation (ChIP)

ChIP was performed following the Covaris truChIP Chromatin Shearing Kit protocol (PN 520154 and PN 520127). Briefly, cells were crosslinked with 1% formaldehyde for 10 minutes at room temperature and quenched with 175 mM Tris pH 7. Extracted nuclei were sonicated with a Covaris S220 Sonolab 7.2 1.0. Sonicated DNA was immunocleared with 2 μg/mL sheared salmon sperm DNA (Thermo Fisher Scientific AM9680), 6 μg/mL normal mouse IgG (Santa Cruz Biotechnology sc-2025), and 30 μL Protein G Dynabeads (50% vol/vol) (Fisher Scientific 10–004-D). Next, antibody pulldown was performed with 5 μg α-MYCN IgG (Santa Cruz, B8.4.B) or 5 μg α-HIF-1α (BD Biosciences, 610,958) overnight at 4°C. As an antibody control, 5 μg of normal mouse IgG was used (Santa Cruz, sc-2025). After immunoprecipitation, 30 μL Protein G Dynabeads and 2 μg/mL sheared salmon sperm DNA were added for an additional 2 hours. Captured protein-DNA complexes were washed, eluted, and reverse-crosslinked. DNA was purified for qPCR or sequencing with Genelute PCR Clean Up Kit (Sigma-Aldrich).

ChIP-qPCR

DNA was amplified for ChIP-qPCR with Power SYBR Green PCR Master Mix. Serially diluted input DNA was used as the standard curve. ChIP-qPCR data were normalized using the percent input method (% input = 100*2^(adjusted input – Ct). Primers for target regions can be found in Table S1.

ChIP-sequencing

ChIP samples were sequenced on an Illumina HiSeq 4000, from a single end for a length of 50 bp. Sequenced reads were aligned to the hg19 genome build with the Burrows-Wheeler Aligner [25]. Peaks were called with MACS2 [26]. Files for visualization on IGV were generated with the Count function of IGVtools [27]. Data from HIF-1α ChIP-sequencing have been deposited in the Gene Expression Omnibus (GEO) under the GEO accession number GSE167477.

Bisulphite Sequencing

Bisulphite sequencing was performed as previously described [11,12], and bisulphite primers can be found in Table S1.

Quantification of 5-hmC by ultra-high performance liquid chromatography coupled with tandem mass spectrometry (UHPLC-MS/MS)

5-hmC quantification was performed as described [28].

Determination of 5-hmC distribution by hMe-SEAL and sequencing

5-hmC selective chemical labelling (hMe-Seal) was performed according to the reported methods [12,29]. DNA libraries were sequenced on an Illumina HiSeq 4000, from a single end for a length of 50 bp. Sequenced reads were aligned to the hg19 genome build with the Burrows-Wheeler Aligner [25]. Peaks were called with MACS2 [26]. Peak files from all time points were merged. HTSeq-count was used to count the number of reads per peak, which was then converted to FPKM [30]. Peaks that were classified as ‘early’ have a FPKM increase greater than 1.2-fold from 0 to 6 hours and remain stable throughout the rest of the time course. Peaks that were classified as ‘late’ were stable from 0 to 24 hours and increased to greater than 1.1-fold between 24 and 48 hours. Following the reported methods, we conducted genomic element analyses by intersecting 5-hmC peaks with files containing the location and annotation of genomic elements [11,12]. Predicted enhancers were generated from data collected by EnhancerAtlas, downloaded in a bedfile format [31]. The HIF-1α binding site regions were generated from HIF-1α ChIP-sequencing peak files, called with MACS2 [26]. Pathway analysis was performed with PANTHER pathway analysis, using adjusted P-values [32]. Data from hMe-Seal sequencing have been deposited in GEO under the accession number GSE167475.

RNA extraction and sequencing

Total RNA was extracted with RNAzol reagent (Sigma-Aldrich, R4533) according to the manufacturer’s protocol. RNA was converted to cDNA with Life Technologies High-Capacity cDNA Reverse Transcription Kit (4,368,814). Quantitative PCR was performed with Power SYBR Green PCR Master Mix (Life Technologies 4,368,706). Primers for qPCR can be found in Table S1.

RNA was sequenced on an Illumina HiSeq 4000, from a single end for a length of 50 bp. Reads from RNA-sequencing in fastq format were aligned to the hg19 reference genome built with Tophat2 [33]. Gene expression was quantified and compared with tools in the Cufflinks package [33]. To detect TET1s, the Cufflinks reference annotation gtf file was modified to include the Gnomon [34] predicted transcript: XM_011540207.2. RNA-sequencing experiment data from our experiments can be accessed under GSE167476.

Transwell and wound healing assays

Transwell assays were performed following reported methods, with PBS control and 10 μg/mL plerixafor added to the media [35]. Wound healing assays began with cells grown to confluency in a 96-well plate. Wells were scratched with an Essen Woundmaker and then placed in IncuCyte and photographed every 4 hours. Alternatively, the wound was created with the Abcam Wound Healing Assay kit (ab242285), and cells were photographed at 0, 6, 12, 24, and 48 hours on an Evos FL Auto 2 microscope. To achieve pseudo-hypoxia, NB cells were treated with FG-4592 at 30 μM and with 10 μg/mL plerixafor at hour 0 and every 12 hours thereafter. Drug optimization experiments were performed with both drugs to determine the optimal concentration for NB cells.

Statistical analysis

Statistical significance was calculated using two-tailed t-tests for most biological experiments between the two groups. When more than one group was compared, a one-way ANOVA test was used. For large datasets for which a normal distribution could not be assumed, a Wilcoxon rank-sum test was used. P-values from all tests were considered significant when less than 0.05. Significance tests were carried out on R, GraphPad Prism, and Microsoft Excel. Graphs were generated in R and GraphPad Prism. Statistical thresholds and exact values for n are given in the Figure Legends. Statistical analysis associated with Figure S9 is described in Supplementary Material.

Results

MYCN binds three potential regulatory regions of TET1

We observed that hypoxia induces 5-hmC gains through up-regulation of TET1 specifically in MYCN-amplified NB cells, but not in NB cells with low MYCN protein levels [12,36]. This implies MYCN regulates TET1 transcription either directly or indirectly. To address this, we examined the correlation between MYCN and TET1 expression using published RNA-sequencing (RNA-seq) data from NB cell lines and primary tumours [37,38]. MYCN-amplified NB lines have a higher baseline level of TET1 expression compared to non-MYCN amplified lines (Figure 1a), and there is a positive correlation between MYCN and TET1 expression levels (r = 0.56, P = 0.01, Figure 1b). We observed a similar positive correlation between TET1 and MYCN expression in NB tumours [37] (r = 0.48, P = 1.50*10⁻³¹, Figure 1c). To test if this observation was specific for TET1, we performed the same analysis with TET2 and TET3 [37,38], and found no correlation between MYCN and that of TET2 or TET3 in NB cell lines (Fig. S1A,C). However, in tumours, there was a modest correlation between MYCN and TET3 (r = 0.31, P = 5.71*10⁻¹², Fig. S1D), but not between MYCN and TET2 (Fig. S1B).

Figure 1. MYCN binds a superenhancer located in the first intron of TET1 (S1), the second intron of TET1 (S2), and a predicted upstream enhancer site near DNA2. (a) TET1 expression data (FPKM) from RNA-seq of 38 NB cell lines [38] graphed according to MYCN-amplified (Amp, red) or non-amplified (Non-Amp, black) status. GEO accession numbers are given in Accession Numbers. (b) TET1 (y-axis) and MYCN (x-axis) expression data from RNA-seq of 38 NB cell lines [38] are expressed as log2 of FPKM. MYCN-amplified cell lines are denoted by red circles, and non-MYCN-amplified lines are denoted by black circles. r represents Pearson correlation coefficient, and P-values have been corrected with the Bonferroni correction method [67]. Cell line names and expression values can be found in Table S2. (c) TET1 (y-axis) and MYCN (x-axis) expression data from RNA-seq of 579 NB tumours [37]. TET1 vs MYCN expression data are plotted as log2 of FPKM. r represents Pearson correlation coefficient, and P-values have been corrected with the Bonferroni correction method [67]. (d) Enhancer regions, MYCN binding motifs, and MYCN ChIP-sequencing data visualized around the TET1 gene. The genomic loci on chromosome 10q21.3 are located at the top according to the hg19 genome build. The Gene Ref track shows the positions of the TET1 and DNA2 genes, and arrows show the directions of gene transcription. S1 and S2 are also indicated. Next, enhancers predicted by ‘EnhancerAtlas.org’ [42] are shown as purple rectangles. An enhancer within TET1 generated from the UCSC LiftOver tool [41] is shown as a purple rectangle. Below, the ‘MYCN Motif’ track, generated with HOMER [41], shows the positions of MYCN binding motifs as vertical bars. The bottom rows display MYCN ChIP sequencing data from SK-N-BE [2]-C, Kelly, NGP, NB1643, COGN415, LAN5, NB69, SHEP, and Tet-21/N cells [39,46] at locus 10q21 visualized with the count function of IGVtools [27]. The scale for each track is listed on the right in brackets. The shaded region corresponding to the first peak, from left to right, is within DNA2. The second shaded peak contains S1, within TET1-intron 1, and the last shaded peak contains S2, within TET1-intron 2. (e) Real time qPCR validation of MYCN ChIP-sequencing performed in SK-N-BE [2]. From left to right, positive control (LARP1), DNA2, S1, S2, and negative control (neg TET1) binding sites are plotted on the x-axis. MYCN pulldown values are plotted in red, and IgG controls are plotted in black. The y-axis corresponds to the amount of DNA pulled down normalized to the amount of input DNA. * and ** represent P < 0.05 and P < 0.01, respectively.

To determine whether reduction of MYCN expression or protein level also results in lowering of TET1 expression, we examined publicly available RNA-seq data from NB cells that had been treated with JQ1, a BET-bromodomain inhibitor that binds and inhibits the MYCN transcriptional regulator BRD4 [39]. Expression data from SK-N-BE [2]-C cells that had been treated with DMSO or JQ1 at 4, 8, or 24 hours [39] demonstrated that MYCN expression decreases by hour 4 and stays at that level for up to 24 hours with a similar trajectory for TET1 (Fig. S2A). To confirm that this decrease was due to loss of MYCN and not a direct result of BRD4 inhibition, we also examined publicly available array data from a mouse NB tumour cell line that had been treated with an Aurora A kinase (AURKA) inhibitor [40]. AURKA stabilizes MYCN protein by preventing its degradation [40]. Upon AURKA inhibition, Tet1 expression decreased (Fig. S2B). Together, these results suggest that MYCN is necessary for maintaining baseline TET1 expression in NB.

We then asked if TET1 was a direct transcriptional target of MYCN. MYCN, like HIF-1, is a TF that recognizes and binds an E-box element that consists of a 5’-CANNTG-3’ sequence [39]. Using HOMER [41], we identified numerous potential MYCN binding motifs within and around TET1 (Figure 1d), suggesting that MYCN could bind and regulate TET1 from several sites. To determine which of these sites is most likely bound by MYCN, we used EnhancerAtlas [42], an enhancer annotation tool based on compiled Chromatin-immunoprecipitation-sequencing (ChIP-seq) and chromatin accessibility data to visualize enhancer sites that could potentially regulate TET1 expression (Figure 1d). EnhancerAtlas provides a database of annotated enhancers from a variety of cell lines, including non-MYCN amplified NB line SK-N-SH [42]. Using data from this cell line, EnhancerAtlas identified two potential enhancers in TET1, in the first and second introns [42] (Figure 1d). The existence of these enhancers has been confirmed experimentally in mouse and human systems. The enhancer located in the first intron is encompassed by a superenhancer originally identified in an intergenic region in mice and remapped to the hg19 human genome [43,44] (Figure 1d). The enhancer in the second intron was identified as a c-MYC target in human embryonic stem cells [45]. Next, we examined the public MYCN ChIP-seq data [39,46] at the TET1 locus from all available MYCN-amplified NB lines to determine the experimentally validated MYCN binding sites (Figure 1d). From data from the six available MYCN-amplified lines, we observed MYCN binding to two sites in TET1, which we will refer to hereafter as Site 1 (S1) and Site 2 (S2) (Figure 1d). S1 is located centrally within the predicted enhancer in the first TET1 intron, and S2 is located within the predicted enhancer in the second TET1 intron. In addition, we considered the possibility that TET1 could be regulated by MYCN binding at distal enhancers. EnhancerAtlas identified a potential TET1-associated distal enhancer located ~88 kb upstream in the first intron of the gene DNA2, which we confirmed was bound by MYCN in MYCN-amplified NB lines (Figure 1d). For comparison, we also visualized MYCN ChIP-seq data from non-MYCN-amplified NB cell lines [39,46]. In NB69 and SHEP cells, there is very little MYCN present and no binding to DNA2, S1, or S2 (Figure 1d). However, one additional non-MYCN-amplified cell line, Tet-21/N, derived from the SHEP line, is engineered to express MYCN under a tetracycline-off system [18]. When we visualized publicly available MYCN ChIP-seq [39] from this cell line in both ‘MYCN on’ and ‘MYCN off’ conditions, we found MYCN bound DNA2, S1, and S2 only in the ‘MYCN on’ condition. We tested the binding of MYCN to each of these three sites in SK-N-BE [2] NB cells with MYCN ChIP-quantitative PCR (qPCR) and found that all three sites were enriched with MYCN binding (Figure 1e). Bisulphite sequencing of the S1 and S2 regions determined that the cytosines are unmodified in normoxia and accessible for TF binding (Fig. S3). These results suggest that MYCN potentially regulates baseline TET1 through any or all of these three sites.

MYCN is associated with full-length TET1 transcript expression

Recent studies have described an alternative ‘short’ TET1 transcript (TET1s) in embryonic development, neural, and cancer model systems [45,47–49]. TET1s is translated into an isoform of TET1 that lacks the CXXC domain and has an intact catalytical domain (Fig. S4A,B). Although the role of TET1s has yet to be fully understood, it is present in differentiated tissues after an isoform switch from full-length TET1 (TET1fl), which is dominant in embryonic tissues [45,47–49]. Because NB is a tumour of embryonic origin, we investigated whether TET1s was present in NB cell lines [38] and whether MYCN preferentially targeted expression of a specific TET1 transcript. We found that both TET1fl and TET1s are expressed in NB lines (Figure 2a). However, MYCN-amplified NB cells expressed higher levels and a higher percentage of TET1fl (Figure 2a,b). Further analysis revealed that TET1fl levels correlate with MYCN expression, but TET1s levels do not (Figure 2c,d). Moreover, we found that TET1fl levels were reduced by about 60% at 4 hours, whereas TET1s was only reduced by about 35% in JQ1-treated SK-N-BE [2]-C cells [39] (Fig. S4C). Together, these results suggest that MYCN preferentially drives TET1fl expression.

Figure 2. Two distinct TET1 transcripts are expressed in NB cell lines. (a) FPKM of TET1fl and TET1s in 38 MYCN-amplified (Amp) and non-MYCN-amplified (Non-Amp) NB cell lines [38]. ** represents P < 0.01; N.S., not significant. (b) Percent of total TET1 expression in MYCN-amplified (Amp) vs non-MYCN-amplified (non-Amp) cells. Percent of TET1fl is plotted in green, and percent of TET1s is plotted in black, with percent calculations displayed in white. * represents P < 0.05. (c) TET1fl (y-axis) and MYCN (x-axis) expression data from RNA-seq of 38 NB cell lines [38] are expressed as log2 of FPKM. r represents Pearson correlation coefficient, and P-values have been corrected with the Bonferroni correction method [67]. Cell line names can be found in Table S2. (d) TET1fl (y-axis) and MYCN (x-axis) expression data from RNA-seq of 38 NB cell lines [38] are expressed as log2 of FPKM. r represents Pearson correlation coefficient, and P-values have been corrected with the Bonferroni correction method [67]. Cell line names can be found in Table S2.

MYCN is sufficient to induce TET1 expression in an inducible MYCN expression system

To test if MYCN is sufficient to modulate the baseline TET1 expression, we used the inducible-MYCN NB line Tet-21/N cells [18], in which we previously showed that, in ‘MYCN on’ condition, MYCN binds S1 and S2 (Figure 1d). MYCN levels were measured daily after induction, with maximal expression reached on Day 4 (Fig. S5A). RNA-seq and qPCR from Day 4 samples showed TET1 and TET3 were induced in MYCN-expressing cells (Fig. S5B,C,D). Analysis of TET1fl and TET1s transcripts revealed that only TET1fl was induced, providing additional support that MYCN targets TET1fl preferentially (Fig. S5E). Using an antibody that specifically detects TET1fl, we found TET1fl protein levels were also elevated on Days 4, 5, and 6 (Fig. S5F,G). Despite the induction of TET1/3, we observed no increase in 5-hmC levels (Fig. S5H), likely due to the very low baseline levels of TET1 expression in comparison to MYCN-amplified lines [50].

MYCN controls TET1 expression directly in normoxia through two binding sites

We hypothesized that one or more of the MYCN binding sites in/near TET1 (Figure 1 and Figure 3a) directly regulate TET1 transcription. In order to test the functional role of each site, we used CRISPR-Cas9 gene editing in SK-N-BE [2] cells to introduce double-strand breaks in the E-box motif and remove each potential binding site from each allele (Figure 3b). From the mixed population of cells, we generated single cell clones and used PCR amplification followed by Sanger sequencing to confirm that each binding site of interest was eliminated (Figure 3b). Deletions generated by this process are often unique in their placement and length, but all lack the CpG contained within the E-box motifs of interest (Figure 3b). Two gene-edited single-cell clones per binding site were used in further experiments. Additionally, we generated SK-N-BE [2] lines lacking both S1 and S2 sites (Figure 3c) through CRISPR-Cas9 targeting of the S2 site in one of the ΔS1 cell lines. Clones were named after their binding site deletion (ΔDNA2, ΔS1, ΔS2, and ΔS1/2) followed by ‘1’ or ‘2’ to distinguish individual clones targeting the same site (Fig. S6A,B and Figure 3a,b,c). We also generated a clone that did not feature any genetic alterations at the binding sites to serve as a non-targeting control (NTC). Using qPCR primers that detected expression of both TET1 transcripts, we found that there was no change in TET1 expression in SK-N-BE [2] cells when either the potential binding site near DNA2 or S2 was deleted, suggesting that MYCN binding at these sites is not essential for maintaining baseline TET1 (Fig. S6C and Figure 3d).

Figure 3. Binding of S1 and S2 by MYCN is important for maintaining normoxic TET1 mRNA expression. (a) Model of MYCN binding around 10q21.3. The two sites are: S1, in a superenhancer (Orange box) in the first intron of TET1, and S2, in the second intron of TET1. (b) Sanger sequencing of the deletions generated in SK-N-BE [2] cells by gRNAs targeting the MYCN binding motifs in TET1: at S1 (ΔS1.1 and ΔS1.2), and at S2 (ΔS2.1 and ΔS2.2). Ref, reference human sequence at this locus. Red font indicates the E-box motif. Slash marks in sequencing from ΔS2 cells represents the loss of 27 base pairs (bps). All clones utilized by this study lacked the CpG that makes up the core of the E-box motif. (c) Sanger sequencing of the deletions generated by gRNAs targeting both MYCN binding motifs at S1 and S2 in TET1 (ΔS1/2.1 and ΔS1/2.2). Ref, reference human sequence at this locus. Red font indicates the E-box motif. Slash marks in sequencing from ΔS1/2 cells represent the loss of 21 bps. (d) Real time qPCR of TET1 mRNA across ΔS1, ΔS2, and ΔS1/2 CRISPR-edited SK-N-BE [2] cell lines (n = 3). P-values were determined with one-tailed t-tests between the value of interest and the parental value; *, **, and *** represent P < 0.05, P < 0.01, and P < 0.001, respectively; N.S., not significant. (e) Western blots for TET1 protein (top panels, cropped at ~280 kDa) in normoxic ΔS1 (left panel); ΔS2 (middle panel), and ΔS1/2 (right panel) CRISPR-edited SK-N-BE [2] cell lines. These blots were stripped and re-probed for TOP1 (bottom panels, cropped at ~110 kDa) as loading controls. Parental cells are non-targeted, and NTC, are cells targeted with a non-specific gRNA to control for off-targets effects.

In contrast, ΔS1 and ΔS1/2 cells exhibited reductions in TET1 expression, of 0.7-fold and 0.25-fold respectively, compared to parental cells (Figure 3d). This indicates that S1 partially moderates baseline TET1 levels, but both sites together maintain a high baseline TET1 expression. To confirm TET1 protein levels reflected TET1 mRNA levels, we used Western blotting to detect the TET1fl protein, which, for simplicity, we will refer to as TET1 in subsequent experiments. Unexpectedly, TET1 protein levels in ΔS1 and ΔS1/2 cells remained comparable to parental and NTC cells (Figure 3e), suggesting that TET1 protein is relatively stable.

S1 and S2 modulate TET1 expression together in hypoxia

We previously reported hypoxic TET1 induction was dependent on the presence of HIF-1α [12] but had not determined the mechanism by which HIF-1 regulates TET1. There are numerous HIF-1 binding motifs in both DNA2 and TET1, including the two that serve as MYCN binding sites in TET1 (Figures 3, 4a). To determine if HIF-1 binds to any of these sites, we performed HIF-1α ChIP-seq to measure HIF-1α binding throughout the genome and compared our results with publicly available HIF-1β ChIP-seq [51]. We found that in hypoxia, both HIF-1 subunits bind S1 and S2 in TET1, but not at the binding motif in DNA2 that is bound by MYCN in normoxia (Figure 4a). To test if HIF-1 alters the binding of MYCN to S1 or S2, we performed MYCN ChIP-qPCR in hypoxia and found that MYCN still binds to S1 and S2 in hypoxic cells (Fig. S7A,B).

Figure 4. Hypoxia affects TET1 mRNA transcription and protein levels. (a) HIF-1 motif and sequencing data aligned with the TET1 promoter, transcriptional start site, and 5’ end of the gene. Top, genomic positions from hg19 at 10q21.3. Gene Ref track indicates locations of the TET1 and DNA2 genes, with arrows representing the direction of transcription of each gene. S1 and S2 are also indicated via shading. HIF-1 motifs are indicated by vertical lines and were identified by HOMER [41]. Next, an enhancer track marks the location of the enhancers found within TET1. HIF-1a and HIF-1b [51] ChIP-seq data from SK-N-BE [2] cells were visualized with IGVtools [27]. The scale for each track is listed on the right in brackets. (b) Real time qPCR quantification of TET1 transcription in normoxia (red) versus hypoxia (dark blue) across ΔS1, ΔS2, and ΔS1/2 clones, normalized to normoxic parental SK-N-BE [2] TET1 expression. P-values were determined with one-tailed t-tests between the parental cell line and the cell line of interest (n = 3). * represents P < 0.05. (c) Representative Western blots for TET1 protein (upper panels) in ΔS1 SK-N-BE [2] cells in normoxia (n) versus hypoxia (h), cropped at ~280 kDa. TOP1 loading control (lower panels), cropped at ~110 kDa. Bottom, mean TET1 protein quantified by ImageJ and normalized to parental normoxic controls in ΔS1 SK-N-BE [2] cells (n = 3). (d) Representative Western blots for TET1 protein (upper panels) in ΔS2 SK-N-BE [2] cells in normoxia (N) versus hypoxia (H), cropped at ~280 kDa. TOP1 loading control (lower panels), cropped at ~110 kDa. Bottom, mean TET1 protein quantified by ImageJ and normalized to parental normoxic controls in ΔS2 SK-N-BE [2] cells (n = 3). (e) Representative Western blots for TET1 protein (upper panels) in ΔS1/2 SK-N-BE [2] cells in normoxia (N) versus hypoxia (H), cropped at ~280 kDa. TOP1 loading control (lower panels), cropped at ~110 kDa. Bottom, mean TET1 protein quantified by ImageJ and normalized to parental normoxic controls in ΔS1/2 SK-N-BE [2] cells (n = 3). (f) Percent 5-hmC levels in Parental, non-targeted control (NTC), ΔS1, ΔS2, and ΔS1/2 CRISPR-edited SK-N-BE [2] cells grown in normoxia (red) versus hypoxia (dark blue). *, **, and *** represent P < 0.05, P < 0.01, and P < 0.001, respectively.

To assess the function of HIF-1 binding at S1 and S2 on TET1 expression, we measured TET1 in ΔS1, ΔS2, and ΔS1/2 cells that were exposed to hypoxia for 48 hours with qPCR (Figure 4b). Although TET1 expression was low in both normoxic and hypoxic ΔS1 cells (Figure 4b), TET1 protein levels were elevated in hypoxic cells (Figure 4c). In contrast, TET1 expression increased as expected in hypoxic ΔS2 cells (Figure 4b) along with its corresponding protein levels (Figure 4d). Despite the very low TET1 mRNA levels in hypoxic ΔS1/2 cells (Figure 4b), TET1 protein was readily detectable (Figure 4e), suggesting that the TET1 protein is stabilized in hypoxic conditions (see below). These results were recapitulated in a second MYCN-amplified NB cell line, NBL-WN, in which both S1 and S2 were deleted (Fig. S8A-E).

To determine the effects of TET1 catalytic activity under normoxic versus hypoxic conditions, we used mass spectrometry to measure global 5-hmC levels [12]. As expected, exposure of MYCN-amplified NB cell lines to hypoxia drove accumulation of 5-hmC levels over cells grown in normoxia (figure 4f and Fig. S8F), consistent with persistent TET activity even under hypoxic conditions [52]. However, global 5-hmC levels in hypoxic ΔS1/2 cells did not reach those measured in parental or NTC cells (figure 4f).

HIF-1 promotes TET1 stability in hypoxia

To determine if elevated TET1 protein levels in hypoxia were a consequence of increased protein stability, we performed protein degradation assays with cycloheximide in normoxic and hypoxic parental SK-N-BE [2] cells. These degradation assays showed that TET1 had a half-life of 20 hours in normoxia and was still detectable even at 30 hours (Figure 5a). In hypoxia, TET1 had an even longer half-life of 40 hours (Figure 5a,c). To determine if HIF-1 is necessary for the increased stability of TET1 in hypoxia, we performed the same degradation assays in normoxia and hypoxia with cells in which HIF1A had been deleted via gene editing (ΔHIF1A) (Figure 5b). Unlike the parental line, the TET1 stability in hypoxia was similar to TET1 stability in normoxia, with TET1 in both conditions having a half-life about 20 hours (Figure 5b,c). This implies HIF-1α is necessary to stabilize TET1 protein in hypoxia.

Figure 5. HIF-1a promotes TET1 protein stability in hypoxia. (a) Western blot of TET1 degradation assay (top) in parental SK-N-BE [2] cells with detection of HIF-1α (middle) and TOP1 (bottom), cropped at ~280 kDa, ~130 kDa, and ~110 kDa, respectively. Samples from normoxia and hypoxia were collected at 0, 3, 6, 12, 24, and 30 hours post treatment with cycloheximide. (b) Western blot of TET1 degradation assay (top) in gene edited ΔHIF1A SK-N-BE [2] cells with detection of HIF-1α (middle) and TOP1 (bottom). Samples from normoxia and hypoxia were collected at 0, 3, 6, 12, 24, and 30 hours post treatment with cycloheximide. (c) Half-life calculations of TET1 protein in normoxia and hypoxia in both parental and ΔHIF1A SK-N-BE [2] cells. * represents P < 0.05; N.S., not significant. (d) Immunoprecipitation (IP) of TET1 protein with detection of HIF-1α under Normoxic (n) and Hypoxic (h) conditions. IP antibodies were a-TET1 (TET1) and IgG (IgG). (e) Fold change of hypoxic 5-hmC levels over normoxic hour 0 5-hmC levels in parental and ΔHIF1A SK-N-BE [2] cells 0 to 36 hours after treatment with cycloheximide. Parental SK-N-BE [2] cells are plotted in red (normoxia) and dark blue (hypoxia). ΔHIF1A SK-N-BE [2] are plotted in coral (normoxia) and sky blue (hypoxia). Percent 5-hmC measured via UHPLC-MS/MS and is calculated relative to guanine. * represents P < 0.05.

Past studies have described TF-TET complexes in which the complex works synergistically to promote TET activity and enhance gene expression from the TF targets [53,54]. We hypothesized that a similar interaction might occur in NB cells between HIF-1 and TET1, which could promote TET1 stability. To determine if HIF-1 and TET1 bound to each other, we performed co-immunoprecipitation experiments in normoxic and hypoxic SK-N-BE [2] cells. Immunoprecipitation of TET1 co-immunoprecipitated HIF-1α, indicating that HIF-1α and TET1 are part of the same protein complex (Figure 5d). Finally, we examined whether the prolonged half-life of TET1, independent of TET1 mRNA regulation, was sufficient to result in higher levels of total 5-hmC in hypoxic cells. We treated ΔHIF1A SK-N-BE [2] cells with cyclophosphamide to inhibit protein synthesis, cultured them under normoxic versus hypoxic conditions, and measured global 5-hmC levels by mass spectrometry. We found that hypoxic 5-hmC levels continued to increase post-cycloheximide introduction in parental cells (Figure 5e), consistent with the increased half-life of TET1 protein demonstrated in hypoxia (Figure 5c), but 5-hmC levels in hypoxic ΔHIF1A cells were lower than those in parental cells (Figure 5e), consistent with our data that a HIF-1/TET1 protein complex increases the stability of TET1 protein (Figure 5a,c).

Hypoxic 5-hmC gains are enriched in and around genes that are important for neuronal morphology, hypoxia adaptation, and cell migration

To further our understanding of the connection between 5-hmC enrichment and initiation of the hypoxic response, we examined the genomic regions where 5-hmC density changes, either increases or decreases, in hypoxic cells over time. To characterize which loci experience altered 5-hmC density over time, we measured 5-hmC distribution changes using hMe-SEAL [29] in MYCN-amplified SK-N-BE [2] cells exposed to hypoxia for 0, 6, 12, 24, 48, or 72 hours. We found that mean FPKM at each time point increased over time (Figure 6a), consistent with our prior observations that global 5-hmC levels increase with time in hypoxia (figure 4f and Figure 5e) and our prior work in these cells [12]. To focus on hypoxic induction of genes that mediate the hypoxic programme, we examined genomic loci that experience 5-hmC gains in hypoxia. We removed all FPKM values that did not increase in hypoxia over time and identified ‘early’ peaks that gained 5-hmC density in the first 6 hours and ‘late’ peaks that gained 5-hmC density after 24 hours (Figure 6b). Analysis of the early gains showed that they were enriched in enhancer and HIF-1 binding regions (Figure 6c). We extracted the gene targets of the enriched enhancers in the early gains, and found over-representation GO-terms [32] in biological processes related to neuronal morphology and development (Figure 6d), consistent with the observation that NB cells de-differentiate and undergo morphological changes in hypoxia [55]. Also, over-represented were processes related to the hypoxic responses, such as artery morphogenesis (Figure 6d). Analysis of the late gains showed that they were enriched in different genomic regions than those of early gains. The late gains were enriched in promoters, 5’ untranslated regions, coding domain sequences, 3’ untranslated regions, CpG islands, and CpG shores (Figure 6e). Among the late gains, we found that many of these over-represented targets were part of neuronal development, hypoxia biology, and cell migration (figure 6f).

Figure 6. Accumulation of 5-hmC in hypoxic SK-N-BE [2] NB cells occurs in genes important for neuronal morphology, hypoxia adaptation, and cell migration. (a) FPKM values of all 5-hmC peaks (n = 415,416) plotted versus time in hypoxic culture: 0, 6, 12 24, 48, and 72 hours. *** represents P < 2.2*10⁻¹⁶. (b) 5-hmC peaks (n = 189,949) that show a positive log fold change from 0 to 72 hours. Shaded regions indicate high fold-change between two timepoints. *** represents P < 2.2*10⁻¹⁶. (c) Log2 enrichment (y-axis) of peaks (n = 26,061) that increase from 0 to 6 hours plotted by genomic element (x-axis): promoters, 5’ untranslated regions (5’ UTR), coding domain sequences (CDS), 3’ untranslated regions (3’ UTR), introns, enhancers, HIF-1a binding regions (generated from our HIF-1a ChIP-sequencing data), CpG islands, CpG shores, and intergenic regions. *** represents P < 2.2*10⁻¹⁶. (d) Biological processes associated with the genes associated with the 5-hmC-enriched enhancers identified in (C) determined with PANTHER statistical overrepresentation test [32,68]. The biological process is shown on the y-axis, and the enrichment factor is plotted on the x-axis. The number at the right of each process gives the enrichment factor. (e) Log2 enrichment (y-axis) of peaks (n = 56,190) that increase from 0 to 6 hours plotted by genomic element (x-axis): promoters, 5’ untranslated regions (5’ UTR), coding domain sequences (CDS), 3’ untranslated regions (3’ UTR), introns, enhancers, HIF-1a binding regions (generated from our HIF-1a ChIP-sequencing data), CpG islands, CpG shores, and intergenic regions. *** represents P < 2.2*10⁻¹⁶. (f) Biological processes associated with 5-hmC-enriched CDS identified in (E) determined with PANTHER statistical overrepresentation test [32,68]. The biological process is shown on the y-axis, and the enrichment factor is plotted on the x-axis. The number at the right of each process gives the enrichment factor.

CXCR4 is critical for the migration of NB cells in hypoxia

We focused on determining which genes are controlled by changes in 5-hmC density and contribute to cell migration as a function of hypoxia, since this phenotype is associated with aggressive cancer cell behaviour and metastasis [56]. To identify candidate genes involved in cell migration, we merged the list of hypoxic 5-hmC-enriched genes from our pathway analysis with our RNA-seq data from normoxic and hypoxic parental SK-N-BE [2] cells, and identified two 5-hmC-enriched genes with hypoxia-augmented expression: CXCR4 and CRKL (Figure 7a).

Figure 7. MYCN and HIF-1 likely bind at the CXCR4 gene. (a) CXCR4 (left) and CRKL (right) expression plotted as FPKM (y-axis) analysed from RNA-seq data from SK-N-BE [2] cells grown under normoxia (red, x-axis) versus hypoxia (dark blue, x-axis). (b) CXCR4 versus HIF1A expression data plotted on the y- and x- axis respectively. r represents Pearson correlation coefficient, and P-values have been corrected with the Bonferroni correction method [67]. (c) CXCR4 RNA-seq expression data plotted as FPKM (y-axis) from Parental, non-targeted control (NTC), and ΔHIF1A SK-N-BE [2] cells (x-axis). Red, normoxia; dark blue, hypoxia. * and ** represent P < 0.05 and P < 0.01, respectively; N.S., not significant. (d) Aligned reads along the CXCR4 gene at locus 2q22.1 visualized with IGVtools [27]. Top track, the Gene Ref track of the CXCR4 gene. Next, HIF-1α ChIP-seq data from SK-N-BE [2] cells at 48 hours hypoxia; followed by tracks presenting the hMe-SEAL data obtained at 0 and 48 hours in hypoxia. The bottom eight tracks display MYCN ChIP-seq data from normoxic NB cell lines: SK-N-BE [2]-C, Kelly, NGP, NB1643, COGN415, LAN5, NB69, and SHEP [39,46]. The scale for each track is listed on the right in brackets.

We focused on CXCR4, since it showed the highest fold induction in hypoxia, 2.1-fold, and prior work had identified CXCR4 as an important hypoxic target in NB cells [17]. CXCR4 encodes a surface receptor that functions in the retention of haematopoietic stem cells in the bone marrow and chemotactic guidance in neural progenitor cells [57]. In NB, CXCR4 expression is correlated with metastatic spread and worse outcome [58]. Consistent with prior work showing that hypoxia induces CXCR4 expression [17,59], we compared CXCR4 and HIF1A expression in NB tumours [37] and found a positive correlation between them (r = 0.26, P = 1.37*10⁻⁷, Figure 7b). However, because HIF-1α is regulated largely by post-translational modifications rather than at the transcriptional level, we tested whether CXCR4 was induced in ΔHIF1A cells that lack HIF-1α protein (Figure 5b). We found that HIF-1 is required for induction of CXCR4 expression in hypoxia (Figure 7c). We examined our HIF-1α ChIP-seq and hMe-SEAL data as well as MYCN ChIP-seq data [39,46] at the CXCR4 locus and identified a HIF-1α peak near the promoter of CXCR4, suggesting that HIF-1 regulates CXCR4 directly (Figure 7d). In addition, we noted that CXCR4 gains 5-hmC density throughout the gene body, consistent with its increased transcription (Figure 7a,c). Additionally, ChIP-seq for MYCN in multiple MYCN-amplified NB cell lines [39,46] shows multiple binding peaks across the gene promoter and first intron, suggesting that MYCN co-regulates CXCR4 in normoxia (Figure 7d).

We extended our analysis of CXCR4 expression to include RNA-seq data from several NB cell lines with varying levels of MYCN gene amplification and corresponding protein levels. CXCR4 was induced in hypoxia in NB lines with abundant MYCN protein and one (SH-SY5Y), with c-MYC [36] (Figure 8a). In NBL-S cells, which express MYCN protein at levels lower than in MYCN-amplified SK-N-BE [2]-C cells [60], CXCR4 was not induced in hypoxia but had relatively high baseline expression (Figure 8a). To test whether CXCR4 expression depends on an intact MYCN/TET1/HIF-1 axis, we measured CXCR4 expression in ΔS1/2 SK-N-BE [2] and NBL-WN cells and observed that CXCR4 expression was virtually eliminated in ΔS1/2 cell lines grown in hypoxia (Figure 8b,c).

Figure 8. CXCR4 controls migration in hypoxic NB cells with high MYCN protein levels. (a) CXCR4 expression plotted as log2 of TPM (y-axis) from RNA-seq data from multiple NB cell lines according to MYCN protein levels (x-axis): NBL-WN, SK-N-BE [2], and LA1-55 n are MYCN-amplified with abundant MYCN protein [36], given as ‘+’ in the MYCN track. SH-SY5Y is a cell line with elevated levels of c-MYC, indicated as a ‘+’ in the c-MYC track. RNA levels were measured from cells grown under normoxic (red) or hypoxic (dark blue) conditions. *** represents P < 0.001. (b) Real time qPCR of CXCR4 expression (y-axis) in Parental, non-targeted control (NTC), and ΔS1/2 SK-N-BE [2] cells (x-axis) after 48 hours of hypoxia exposure (n = 3). P-values were determined with one-tailed t-tests, * represents P < 0.05. (c) Real time qPCR of CXCR4 expression (y-axis) in Parental, non-targeted control (NTC), or ΔS1/2 NBL-WN cells (x-axis) after 48 hours of hypoxia exposure (n = 3). ** represents P < 0.01; N.S., not significant. (d) Transwell migration assays were performed with SK-N-BE [2] ΔS1 versus ΔS1/2 cells in hypoxia. Values of ΔS1, ΔS1/2, and NTC were normalized to Parental cells as a percent. P-values were determined with one-tailed t-tests (n = 3); * and *** represent P < 0.05 and P < 0.001, respectively. (e) Transwell assays performed in hypoxia with parental SK-N-BE [2] (left) and NBL-WN (right) cells and treated with (+) or without (-) 10 mg/mL plerixafor (n = 4). P-values were determined with one-tailed t-tests, * represents P < 0.05. (f) Wound healing assays performed with parental SK-N-BE [2] (left) and NBL-WN (right) cells and treated with 10 mg/mL plerixafor (black squares) or phosphate buffered saline (PBS, open circles). Both control and experimental cells were treated with 30 µM FG-4592 to induce pseudo-hypoxia. P-values were determined with one-tailed t-tests; * and ** represent P < 0.05 and P < 0.01, respectively.

Because prior work has demonstrated that CXCR4 contributes to NB cell migration [58,61,62], we performed transwell assays in hypoxic SK-N-BE [2] ΔS1/2 cells. We observed that ΔS1/2 cells migrated slower in hypoxia compared to controls (Figure 8d). ΔS1 cells, with an intermediate level of CXCR4 induction (Figure 8b) also demonstrated decreased migration in hypoxia, albeit not as much as ΔS1/2 cells (Figure 8d) as well as defective wound healing, another means of measuring cell migration (Fig. S9A,B). The cell migration phenotype was specific to hypoxic cells and was not observed in normoxia (Fig. S9C), suggesting that CXCR4 is involved specifically in hypoxia-directed migration of NB cells. To confirm that this was not the result of altered proliferation, we compared the growth rates of ΔS1 versus control cells and found no difference in growth rate in either normoxia or hypoxia (Fig. S9D,E).

To test if CXCR4 promotes cell migration directly, we treated hypoxic SK-N-BE [2] and NBL-WN cell lines with plerixafor, a CXCR4 antagonist [63]. To measure cell migration, we used two complementary approaches: transwell assays performed in hypoxia and wound healing studies carried out in pseudo-hypoxia. Pseudo-hypoxia was achieved by treating cells with FG-4592, a HIF-α prolyl hydroxylase inhibitor, and confirming HIF-1α stabilization. We found that plerixafor-treated cells migrated slower than their control counterparts in hypoxia and pseudo-hypoxia (Figures 8E,f and Fig. S10), indicating that CXCR4 directs MYCN-amplified NB cell migration in hypoxia.

Discussion

In this study, we describe how the TET1 gene is transcriptionally regulated by MYCN and HIF-1 in MYCN-amplified NBs and identify CXCR4 as a gene induced by hypoxia via MYCN/TET1-induced 5-hmC deposition, which functions as a driver of NB cell migration in hypoxia (Fig. S11). In normoxia, TET1 is bound by MYCN at two sites in the first and second introns, S1 and S2. Moreover, MYCN drives the expression of the TET1fl transcript preferentially. Single deletions of either S1 or S2 lower transcription by 30%, but when both sites are deleted, TET1 expression is reduced by 80%, suggesting that both S1 and S2 sites are needed for full induction of the gene. Similarly, in hypoxia, S1 and S2 serve as docking sites for HIF-1, which further drives TET1 transcription. The MYCN/TET1/HIF-1 axis is further enforced by complex formation by TET1 and HIF-1, which extends the half-life of TET1, a means of maintaining its catalytic activity, which is dependent on oxygen [52].

Several questions remain concerning the role of TET1 isoforms in NB. Studies that have compared the roles of TET1fl versus TET1s have identified TET1fl as the more ‘efficient’ isoform and hypothesized its presence is necessary for large-scale epigenetic reprogramming, such as during cell differentiation [47–49]. Future experiments will determine if TET1 isoforms have distinct roles in NB by examining their impact on the 5-hmC landscape and their relationship with prognosis.

TET1 protein stability is augmented in hypoxia due its binding interaction with HIF-1α. Our previous findings that TET1 modifies DNA at HIF-1 binding sites in canonical HIF-1 targets [12] also suggest that the HIF-1/TET1 interaction may be required to recruit TET1 to specific genomic sites. Future experiments will determine whether the loss of this interaction changes the 5-hmC landscape and whether other proteins participate in the HIF-1α/TET1 complex. Other possible mechanisms of TET1 protein regulation could also be the subject of future work, including identifying whether any post-translational modifications (PTMs) [64] contribute to the stability and/or activity of TET1. Although multiple PTMs of TET proteins have been described [65,66], their functional roles could be defined further in future work. Our prior observation that 5-hmC augments expression from HIF-1 targets in MYCN-amplified NBs prompted us to investigate 5-hmC profiles from both NB tumour and cell-free DNA. We determined 5-hmC profiles can serve as DNA-based biomarkers that are predictive of metastatic burden and prognostic of overall outcome [7]. Here, we determined that 5-hmC accumulation in hypoxic NB cell lines occurs in genes involved in cell migration, identifying CXCR4 as a specific hypoxia target gene that mediates this phenotype and may serve as a useful biomarker for aggressive disease and a therapeutic target in MYCN-amplified NBs. CXCR4 has been previously identified in gene expression studies of primary NB tumour samples, correlating with metastatic spread and worse outcome [58,62], but in other solid tumour cell lines, CXCR4 has been shown to promote or activate MAPK/ERK, JAK/STAT, and PI3K/AKT pathways, resulting in cell migration and survival [63]. Future studies will elucidate the mechanism by which CXCR4 and its downstream targets augment cell migration in hypoxia in NB. Additionally, future studies could assess whether the addition of a CXCR4 antagonist to the NB treatment regimen is for the purpose of slowing cell migration/metastasis, rather than as a haematopoietic stem cell mobilizing agent as it is traditionally given, to improve patient outcomes.

Author contributions

A.E.H. designed and performed experiments, analysed data, and wrote the manuscript. S.U. performed experiments. J.Z.C. analysed the data and helped perform experiments. M.A.A. analysed and contributed sequencing data. H.R.S. helped design experiments. L.A.G. and S.L.C. conceived the study, analysed the data, and edited the manuscript.

Acknowledgments

We thank Dr. Paul Geeleher for performing statistical analysis of wound-healing assay data sets. We would also like to thank Dr. Marsha Rosner and Dr. Chilong Nguyen for their invaluable instruction of several protein biochemistry techniques. Finally, we would like to thank Dr. Julie Losman for her CRISPR plasmids targeting HIF1A.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

Datasets generated for this study can be found at the Gene Expression Omnibus under the SuperSeries accession number GSE167478.

GEO accession numbers for publicly available ChIP-seq datasets used in this study are GSE80151 (39), GSE138315 (46), and GSE71399 (51). Sequencing data for NB cell lines can be found on GEO databank (GSE89413) and sequencing for NB tumours can be accessed through R2: Genomics Analysis and Visualization Platform (https://r2.amc.nl/) (37,38). NB cell line names can be found in Table S2. RNA-seq data from JQ1 experiments can also be accessed under GSE80154 (39). RNA-seq data from MLN8237 studies can be available under GSE57810 (40).

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/15592294.2022.2106078

Additional information

Funding

This work was supported by Alex’s Lemonade Stand and a Northwestern Mutual Young Investigator Grant. A.E.H. was supported by The Molecular and Cellular Biology Training Grant, and the HHMI Med-into-Grad Translational Training Program. M.A.A. was supported by the National Institutes of Health, K08CA226237.