Large-scale mapping of positional changes of hypoxia-responsive genes upon activation

Abstract

Chromosome structure and nuclear organization are important factors in the regulation of gene expression. Transcription of a gene is influenced by local and global chromosome features such as chromatin condensation status. The relationship between the 3D position of a gene in the nucleus and its activity is less clear. Here we used high-throughput imaging to perform a large-scale analysis of the spatial location of nearly 100 hypoxia-responsive genes to determine whether their location and activity state are correlated. Radial distance analysis demonstrated that the majority of Hypoxia-Inducible Factor (HIF)- and CREB-dependent hypoxia-responsive genes are located in the intermediate region of the nucleus, and some of them changed their radial position in hypoxia. Analysis of the relative distances among a subset of HIF target genes revealed that some gene pairs altered their relative location to each other on hypoxic treatment, suggesting higher-order chromatin rearrangements. While these changes in location occurred in response to hypoxic activation of the target genes, they did not correlate with the extent of their activation. These results suggest that induction of the hypoxia-responsive gene expression program is accompanied by spatial alterations of the genome, but that radial and relative gene positions are not directly related to gene activity.

INTRODUCTION

All cells in the human body constantly consume oxygen for efficient energy production, and multiple biochemical reactions are mediated by oxygen-dependent enzymes under normal conditions (Losman et al., 2020). When cells in an organism are exposed to low oxygen levels, the hypoxic response is triggered (Kaelin and Ratcliffe, 2008). The hypoxic response is an active process that involves the induction of a specific gene expression program that counteracts the potentially harmful effects of hypoxia to cells and tissues. Hypoxia-Inducible Factor, HIF, is a master transcription factor of the hypoxic response and plays a critical role by up-regulating a number of hypoxia-responsive genes (Keith et al., 2012). HIF induces a wide range of genes including many involved in vasculo-/angiogenesis, red blood cell production, and regulation of metabolism (Hirota, 2020). HIF-1α expression decreases on prolonged hypoxic treatment, and CREB and NF-κB become activated during this stage (Nakayama, 2013).

The activity of genes is affected by their local chromatin environment. In particular, there are two major types of chromatin: heterochromatin, which has higher density and often contains transcriptionally repressed genome regions (Passarge, 1979), and euchromatin, which is more decondensed and contains transcriptionally active genes (Jost et al., 2012) as well as poised genes, which become transcriptionally active in response to cellular signals. Gene activation often involves changes in chromatin structure, including decondensation of chromatin or association of distal regulatory regions with promoter elements via long-range loop formation (Dekker and Misteli, 2015). The activity of genes has also been linked to their positions relative to nuclear structures, particularly the association of inactive genes with the nuclear envelope and the periphery (Takizawa et al., 2008b; Egecioglu and Brickner, 2011), and of active genes with nuclear splicing speckles (Spector and Lamond, 2011). In addition, it has been suggested that coregulated genes may cluster around shared transcription sites in the cell nucleus (Osborne et al., 2007). These conclusions regarding the relationship of gene activity and nuclear location have relied heavily on analysis of individual or very small sets of genes, often only probing the immediate proximity of genes to the nuclear envelope (Williams et al., 2006; Meaburn and Misteli, 2008; Reddy et al., 2008; Peric-Hupkes et al., 2010; Muck et al., 2012; Gatticchi et al., 2020; Sumner et al., 2021). More recent larger scale studies have mostly related the activity of genes to their location relative to the chromosome they are located on but not with regard to their radial positions within the cell nucleus (Su et al., 2020; Takei et al., 2021). The relationship of radial and relative gene location with activity has not been probed systematically in a large set of coregulated genes.

The hypoxic response requires induction of multiple genes in a coordinated manner in order to adapt to low oxygen and as such offers an opportunity to map the location of a set of coregulated genes in response to a specific milieu. In the present study, we have used the cellular response to hypoxia as a model system to probe the relationship between 3D gene position and activity at a large scale and in an unbiased manner. We performed positional mapping of 104 hypoxia-responsive genes using HIPMap, a High-Throughput Imaging Positioning MAPping method based on fluorescence in situ hybridization (FISH) using barcoded probes, which allow the determination of 3D locations and distances at large scale (Shachar et al., 2015a, b). We find evidence for large-scale reorganization of the genome in response to hypoxia but do not detect any correlation between radial or relative spatial position and gene activity or evidence for spatial clustering of coregulated genes.

RESULTS

Hypoxic treatment in high-throughput format

To probe the relationship between gene position and gene activity, we used high-throughput FISH to comprehensively map the nuclear position of a set of 104 hypoxia-inducible genes in human MDA-MB231 breast cancer cells. Cell culture conditions and hypoxic treatment of 1% oxygen were optimized for use in the 384-well plate format using a customized hypoxia chamber, and standard fixation and imaging methods were applied (Figure 1A; see Materials and Methods). Immunocytochemistry demonstrated the expected induction by 9.3 ± 2.9-fold (p = 3.52 × 10^–26) in expression of HIF-1α after 24 h of hypoxic treatment compared with cells grown at normal oxygen levels (Figure 1B and Supplemental Figure S1). Similarly, phosphorylation of CREB, a factor activated under prolonged hypoxia (Nakayama, 2013), was observed at 48 h of hypoxic treatment (Figure 1B). HIF-1α staining was positive in all wells examined, indicating the absence of position and edge effects within the imaging plate (Supplemental Figure S1). Induction of hypoxia was also confirmed by lactate assays which showed ∼3-fold higher lactate level in hypoxic samples compared with the normoxic samples (Figure 1C; norm 17.3 ± 0.88 μM, hypo 48.6 ± 3.00 μM, p = 3.88 × 10^–7). Since the nuclear size and shape also affect gene position, we compared nuclear morphology by calculating nuclear area and nuclear roundness. No significant differences in nuclear morphology were found. Normoxic- and hypoxic-treated cells both had a nuclear area of 160 μm² on average and similar nuclear roundness, suggesting that nuclear shape and area do not change on hypoxic treatment (Figure 1D).

FIGURE 1:

HIPMap analysis under hypoxic condition. (A) Schematic overview of hypoxic HIPMap analysis. (B) MB231 cells plated in 384-well plate format were treated with normoxia (21% O₂) or hypoxia (1% O₂) for the indicated time. After fixation, cells were stained with anti-HIF1α or anti-phospho CREB (pCREB) antibody. Nuclei were stained with DAPI. Scale bar: 50 μM. (C) Lactate assay of hypoxic-treated MB231 cells in 384-well plates. Cells were maintained under hypoxic condition (1% O₂) for 48 h. Cell culture medium was collected and subjected to lactate assay. Values represent averages of the experiments (n = 6); error bars indicate SD. Significance was analyzed by t test (**p < 0.02). (D) Nuclear area and nuclear roundness were calculated in normoxic- and hypoxic-treated cells using Columbus image analysis software (PerkinElmer). Values represent averages of experiments (n = 6); error bars indicate SD. The difference between normoxic and hypoxic samples was not significant.

Positional mapping of hypoxia-inducible genes in the nucleus

With this experimental setting, we performed a high-throughput FISH analysis by HIPMap (Shachar et al., 2015a); 104 hypoxia-responsive genes were selected based on their dependence on the two major hypoxia transcription factors HIF and CREB. HIF-target (HT) genes were identified based on previous expression studies (Wenger et al., 2005), and putative CREB target genes were identified from RNA-seq analysis comparing WT and CREB-knockdown MB231 cells (Kikuchi et al., 2016). Target genes were divided into three groups: 35 known direct HTs which contain HIF binding sites; 3 direct CREB targets (CDT), which contain CRE motifs in their promoter region; and 64 indirect CREB targets (ICT), which were decreased in CREB-KD cells but do not have CREs (Kikuchi et al. 2016). As expected, most of the target genes were up-regulated in MB231 cells treated with hypoxia for 48 h with typical induction levels ranging from 1.5- to 2-fold (Supplemental Table S1). Genes which did not show any significant difference between normoxic and hypoxic conditions in the RNA-seq analysis were used as controls. Oligo probe-based FISH accurately detected typical gene signals in the nucleus of MB231 breast cancer cells which are mostly triploid (Figure 2A).

FIGURE 2:

Radial Positions of HIF and CREB target genes under hypoxia. (A) Representative images of Oligo paint-based FISH. Three hypoxia-inducible genes on chromosome 9; CA9 (green), DEC1 (red), ALDOB (white). Nuclei were stained with DAPI (blue). Scale bar: 10 μM. (B–D) Radial position of representative genes. Radial positions of genes comparing normoxia and hypoxia were plotted in a histogram (red bar, normoxia; blue bar, hypoxia). Plots represent the data from FISH spots. (B) Genes shifted toward center of the nucleus on hypoxia. (C) Genes shifted toward peripheral of the nucleus on hypoxia. (D) Control genes. (E) Radial positions of hypoxic genes. Distribution of the genes examined were shown in five concentric equidistance and equiarea nuclear shells. Shell 1 is the most peripheral region of the nucleus, and shell 5 is the most central region. HT, HIF-target group; ICT, indirect-CREB target group; CDT, CREB-direct target group; NC, nonresponsive control.

To determine the nuclear position of these genes, we calculated the nuclear radial position of each gene at the allele level, normalizing each measurement to nuclear size as previously described (Gudla et al., 2008). For each experiment, performed with at least two biological replicates per gene, we compared normoxic and hypoxic conditions and calculated 1) the mean radial position in each condition, 2) the variance in radial position in each condition, and 3) the p value to compare induced versus uninduced position using the Wilcoxon rank sum test. Genes were filtered for genes that showed a consistent change in position across biological replicates. To determine the degree of change in gene position, we used inverse variance weighting to calculate the weighted ratio of mean radial distance for each experiment and corresponding 95% confidence intervals (see Materials and Methods; Friedrich et al., 2011). For comparisons between hypoxic to normoxic conditions, multiple p values generated by the Wilcoxon rank sum test on independent experiments were combined using Fisher’s method (Fisher, 1932).

Among 104 genes (Supplemental Table S1), the radial distance distributions showed that each gene has a variable and nonrandom positioning pattern within the nucleus of normoxic MB231 cells (Figure 2, B, C and D). Most genes showed a distinct peak at a particular distance from the nuclear membrane rather than a broad distribution across the interior-exterior axis, suggesting that they have a preferred radial position in the nucleus as previously observed (Shachar et al., 2015a; Meaburn et al., 2009). No characteristic distribution profiles were associated with HT or ICT group genes. Typically, about half of the alleles were located in the most peripheral region and the other half were located in the intermediate region, with no genes enriched in the nuclear interior (Figure 2, B, C and D). Similar distribution patterns were also observed for the control gene group (Figure 2D). We further grouped the genes by equidistant or equiarea shells in the nucleus based on the median of radial distances. We calculated the median radial distance for each gene to determine the corresponding shell location. All genes analyzed were found in the peripheral shells 2 or 3 regardless of the gene group in the equidistance analysis, whereas the distribution showed greater variation in the equiarea analysis which ranged from shells 2 to 5 (Figure 2E).

To ask whether activation of hypoxia-responsive genes leads to a change in their nuclear location, we compared the radial position of 92 genes under normoxia and hypoxia conditions (Supplemental Table S1, bold genes). Among this set, the radial position of 21 genes differed statistically significant between hypoxia and normoxia (Figure 2, B and C and Supplemental Table S2; P < 0.05). Among the 21 repositioning genes, 18 were CREB target genes, and this enrichment was statistically significant by hypergeometric analysis relative to their representation in the gene set (p = 0.031); 14 genes moved toward the interior of the nucleus (ratio of mean [hypoxia/normoxia] > 1), whereas 7 genes moved toward the periphery of the nucleus on hypoxic treatment (ratio of mean [hypoxia/normoxia] < 1; Supplemental Tables S2 and S3). Of the 21 genes that changed radial position, 10 genes showed a positive or negative expression change of more than 1.5-fold. IFIT1B showed the highest up-regulation (3.4-fold), yet its radial position change on hypoxia was 0.986 (ratio of mean [hypoxia/normoxia]) which ranked 16th out of 21. On the other hand, DEC1 was the most down-regulated gene (1.9-fold), and its radial position change (1.027) was ranked 10th (Supplemental Table S2). The largest changes in radial positioning were observed for a potassium voltage-gated channel modifier, KCNS3, whose mean distribution shifted toward the interior of nuclei, and synaptotagmin SYT12, which moved toward the periphery of nuclei. These changes were accompanied by moderate changes in expression level of KCNS3 which decreased 1.1-fold and SYT12 which increased 1.2-fold (for the profile of all 21 genes see Supplemental Table S2). We find no clear correlation between the change in gene activity and the direction of repositioning, with some hypoxia-activated genes becoming more internal and others assuming a more peripheral location (Supplemental Table S3, genes highlighted in green). The same was observed for hypoxia-repressed genes (Supplemental Table S3, genes highlighted in red). Of note, CREB becomes activated at a relatively late time point (24–48 h) and 28.1% (18/64) of CREB target genes exhibited a significant change of radial position in the ICT group, suggesting that transcriptional activity may at least in part contribute to a change in the radial position. Conversely, among the top 10 genes that increased or decreased expression in response to hypoxia, 2 genes in each group (up-regulated or down-regulated genes) showed a statistically significant change in location (Supplemental Table S1). In addition, the 4 genes with the highest levels of change in expression (ANGPTL4, YPEL1, EGLN3, and LOX) did not show a change in location (Supplemental Table S4), and among the 71 genes that did not change position, 38 genes showed a change in expression by more than 1.5-fold (Supplemental Table S4). These results demonstrate a general lack of correlation between changes in radial gene location and gene activity.

The observed changes in radial position were unrelated to the chromosomal location of these genes. For example, there were eight examined genes on chromosome seven and of these only one gene significantly changed its radial position (Supplemental Tables S2 and S5). This result indicates that if changes in gene position occur they are local events occurring at the level of individual loci and not repositioning of entire chromosomes.

Relative distances of hypoxia-inducible genes under hypoxic condition

To examine if coregulated genes form clusters when they become active under hypoxic condition, we next determined the position of well-characterized HIF-responsive genes relative to each other by measuring the pairwise distances between FISH signals for 159 pairs of genes located on different chromosomes (Supplemental Table S6). In this analysis, we focused on the HT group and analyzed the shortest pair-pair distances of each allele (Figure 3A). Given that these genes had a relatively constant radial position, we were interested in their relative positions. Data analysis was performed in the same way as the radial distance analysis by calculating the weighted ratio of means of relative distances (hypoxia/normoxia) and corresponding 95% confidence intervals (Friedrich et al., 2011) and the combined p value on the Wilcoxon rank sum test using Fisher’s method (Fisher, 1932). Of the 159 pairs, 74 pairs showed statistically significant and reproducible differences in the relative distance between normoxia and hypoxia (Figure 3, B and C; Supplemental Table S7; P < 0.05). Based on the ratios of means of the relative distance between normoxia and hypoxia, 30 pairs became more distal (green), whereas 44 pairs became more proximal (red) (Figure 3D and Supplemental Table S7).

FIGURE 3:

Relative distances of HT genes under hypoxic condition. (A) Relative distances of hypoxia-inducible genes in the nucleus. Minimal distance of hypoxia-inducible genes (A, B) were calculated under normoxic and hypoxic conditions (blue solid arrow). (B, C) Distribution of gene distances comparing normoxic and hypoxic conditions. Changes in the distance of gene pairs was measured and plotted. Distance of gene pairs which became more distal (B) and more proximal (C) on hypoxia are shown (red bar, normoxia; blue bar, hypoxia). (D) Number of gene pairs which became distal or proximal. Gene pairs which changed the distance at most in either direction were counted in the integrated data of different experiments and plotted. The x-axis indicate the degree of distance change (hypo/norm). (E) Distance distribution of gene pairs. Mean distance of gene pairs were calculated for the 74 genes which showed significant differences under hypoxic conditions. (F) Genes with frequent movement based on chromosome location. Genes which changed position were grouped based on the chromosome. The numbers of total genes analyzed and repositioned genes on each chromosome are shown. *p < 0.05.

The mean pair-to-pair distances under hypoxic condition ranged from 586 to 1697 nm, and mean pair distances below 500 nm, which would indicate direct physical association, did not exist (Figure 3E). Colocalization within one pixel ranged from 21.8 to 49.5% (average 31.3%) and was strongly correlated with mean distance. Although there was no clear indication of gene clustering, about 60% of the gene pairs analyzed moved apart upon hypoxia (P < 0.05). Pairs of genes which showed altered distances were enriched on chromosomes 1, 3, and 16 (p = 0.02, 0.009, and 0.029, respectively). Among the 74 gene pairs (148 genes), 5 genes (out of 5) were on chromosome 1, 6 genes (out of 6) were on chromosome 3, and 8 genes (out of 10) were on chromosome 16 (Figure 3F). These results demonstrate the absence of clustering of coregulated genes but indicate that some chromosomes reposition in response to hypoxic stimulation.

Gene activity does not lead to clustering

To test the hypothesis that activation of coregulated genes may lead to their clustering in 3D space, we assessed the relationship of gene expression and relative positioning of gene pairs (Figure 4). The expression level of the two genes in the pair was expressed as the sum or product of the induction level of two genes in Figure 4, i.e., a larger product indicates a larger change in the expression of corresponding two genes on hypoxia, induction or repression of both genes will yield a positive value, and opposite responses will yield a negative value. The top 20 gene pairs, which showed the largest change in the expression level, were assessed. There were 10 gene pairs whose distance became closer and 10 pairs which separated, indicating that changes in the distance and expression level have no apparent relationship (Figure 4; P < 0.05). An example is zinc finger E-box binding homeobox (ZEB1) on chromosome 10 and vascular endothelial growth factor (VEGF) on chromosome 6, which on hypoxic treatment moved apart and also showed opposite expression behavior with ZEB1 down-regulated and VEGF up-regulated (Figure 3B and Supplemental Table S7, top line). However, this is not a general pattern since ZEB1 became more proximal to another up-regulated gene, EGLN1 on chromosome 1 (Figure 3C and Supplemental Table S7, bottom line).

FIGURE 4:

Expression profile of genes that change relative distances under hypoxic conditions. mRNA-seq analyses were performed on MB231 cells treated with or without hypoxic condition for 48 h (n = 2). Correlation of gene expression and relative distance was analyzed. The expression level of the two genes in the pair was expressed as the product and sum of the induction level of two genes in the relative distance analysis (left).

Taken together, these results suggest that changes in gene position are not strictly linked to gene expression status and they further point to a lack of clustering of coregulated genes.

DISCUSSION

Genes occupy nonrandom positions in the 3D space of the cell nucleus but the relationship between gene position and gene activity has remained unclear (Takizawa et al., 2008b). Historically, gene position has often been compared with gene expression levels in studies of individual or small sets of genes, potentially introducing biases toward highly expressed genes or reporting of positive correlations (Zink et al., 2004; Takizawa et al., 2008b; Meaburn et al., 2009; Leshner et al., 2016; Forsberg et al., 2019). In addition, while the relationship of direct association of a gene locus with the nuclear periphery and the nuclear lamina is frequently explored (Williams et al., 2006; Reddy et al., 2008; Peric-Hupkes et al., 2010; Muck et al., 2012; Gatticchi et al., 2020; Sumner et al., 2021), more general comparisons of gene location and activity are largely lacking. More recently, large-scale gene position and/or chromatin structure has been examined by multiplexed FISH methods, but these approaches typically do not correlate gene expression with radial location (Moffitt et al., 2016; Gelali et al., 2019; Girelli et al., 2020; Su et al., 2020; Takei et al., 2021). To probe the relationship of radial gene location and gene activity more broadly and in an unbiased manner, we used hypoxia as a model system and high-throughput imaging to map the location of nearly 100 genes and to analyze their spatial positions in the nucleus and their repositioning behavior on hypoxic stimulation (Schito and Semenza, 2016).

We find that in a set of 104 genes, all of them occupy preferred, nonrandom radial positions, yet all genes showed significant variability in their position in individual cells (Supplemental Table S5). These observations are in agreement with mapping of individual or small sets of genes over the years (Zink et al., 2004; Takizawa et al., 2008a; Takizawa et al., 2008b; Meaburn et al., 2009; Leshner et al., 2016; Forsberg et al., 2019) and support the notion that most genes occupy a preferred but probabilistic 3D distribution in the cell nucleus.

After hypoxic treatment, a subset of genes (∼ 20%) changed their radial position. Radial distance analysis demonstrated that 2 out of 35 genes (5.7%) in the HT group but 18 out of 64 genes (28.1%) in the ICT group altered their positions with statistical significance after hypoxic treatment, indicating that CREB target genes are more likely to reposition than HT genes. This difference may indicate that repositioning of genes depends on the signaling pathway which is activated, or they may reflect the fact that CREB is getting activated, whereas HIF-1α is becoming inactivated at 48 h of hypoxic treatment. Based on studies of individual genes, an internal shift of genes has often been associated with gene activation, whereas relocalization to the periphery has been linked to inactivation (Zink et al., 2004; Takizawa et al., 2008b). However, our radial position analysis of a larger gene set indicates that while some hypoxia-up regulated genes became more internal, others became more peripheral. The same was found for hypoxia-repressed genes (Supplemental Table S3). Thus we conclude that the direction of repositioning is not strongly related to a change in gene activity.

Relative gene position analysis was only performed for the HT group in the present study, and it remains unknown for the ICT or CDT groups. Considering the greater changes of radial position for the ICT and CDT group genes, it is possible that these groups also exhibit changes in their relative gene position. Importantly, two genes from the HT group, cMet and DEC1, which repositioned in the radial distance analysis, were also included in six of the gene pairs which repositioned in the relative distance analysis. In addition to the radial position in the 3D nuclear space, there are reports connecting the position of a gene in relation to the chromosome territory and its activity (Mahy et al., 2002; Wiblin et al., 2005; Branco and Pombo, 2006; Torabi et al., 2017). Since our HIPMap experimental condition cannot currently accommodate simultaneous visualization of individual genes by FISH and chromosome territories by chromosome painting, we were unable to explore this relationship.

An attractive idea has been that coregulated genes cluster in 3D space, for example, via association with transcription factories or nuclear bodies which contain the necessary transcription factors for their activation. Correlations between gene expression and gene position, including the suggestion of clustering of genes, have been reported in several systems such as B lymphocyte development, ES cell differentiation, and glial differentiation (Kosak et al., 2002; Williams et al., 2006; Osborne et al., 2007). On the other hand, there are also reports indicating that no clear correlation exists between gene position and expression (Meaburn and Misteli, 2008). Analysis of the pairwise distances of 159 combinations of 35 HT genes showed limited changes in relative positioning, suggesting that coordinated relocation or clustering of coregulated genes is not a pervasive phenomenon. In addition, the degree of change in the relative position of a gene was often modest, because larger movements may be obscured by differential behavior of individual alleles. In this aspect, it would be interesting to deconvolve the behavior of single alleles using allele-specific probes.

The 3D location of genes has been used to monitor cellular states, including to distinguish normal from cancer tissues (Meaburn, 2016). For example, SP100 and TGFB3 localize more peripherally in prostate cancer tissues compared with normal prostate (Meaburn and Misteli, 2019). However, as observed in our study, the location of the repositioning genes was unrelated to their activity status in earlier analyses (Meaburn and Misteli, 2008; Therizols et al., 2014; Shachar et al., 2015a). Similar to breast and prostate cancer, and despite the seemingly limited relationship of gene activity with spatial position, it may be attractive to use gene positioning as a marker for hypoxia in tissues. This approach may overcome the notorious challenge posed by the highly unstable nature of HIF to assess the hypoxic state of tissues. The genes identified here which reposition may be promising candidates to do so, regardless of their activation status. This approach could, for example, be applied to assess the degree of hypoxia in tumor specimens in response to the HIF-2α inhibitor Belzutifan, which is in clinical trials for renal cancers (Courtney et al., 2018; Choueiri et al., 2021) and approved recently by FDA.

Taken together, this study reports the nuclear positioning behavior of the one of the largest set of coregulated genes in response to stimulation. The results point to a lack of correlation between gene activity and radial positioning and 3D clustering of coregulated genes, and they highlight the heterogeneous response of individual genes on changes in gene expression.

MATERIALS AND METHODS

Request a protocol through Bio-protocol.

Cell culture

MDA-MB231cells were obtained from ATCC (Manassas, VA) and cultured in DMEM (high glucose) (Invitrogen, Carlsbad, CA) containing 10% fetal bovine serum and antibiotics.

Hypoxic treatment

Cells were treated under hypoxic conditions (1% O₂ and 5% CO₂, balanced with N₂) in a hypoxia chamber (Billups-Rothenberg, Inc, Del Mar, CA for FISH experiments) or in a hypoxia workstation (Hirasawa Works, Tokyo, Japan for mRNA-seq analysis). An oxygen sensor was used to regulate the oxygen concentration inside the workstation, which was maintained at 1% throughout the experiment (MC-8G-S, Iijima Electrics, Gamagori, Japan).

Oligonucleotide FISH probes

A pool of oligonucleotides for 104 hypoxia-inducible genes (an average of 819 oligonucleotides/gene within a 100 kb region centered around the gene) was synthesized (Twist Bioscience, South San Francisco, CA). Briefly, oligonucleotide probes contained a genomic sequence of the gene which is connected with a 32-bp barcode sequence that hybridizes to secondary probes. The oligo pool was amplified by PCR and used as a template for T7 primer-based in vitro transcription to synthesize primary probes (Beliveau et al., 2014). Fluorescently labeled secondary readout probes (Eurofin Genomics, Louisville, KY), which hybridize with specific barcode sequences on the primary probes, were used for detection (Beliveau et al., 2012).

Oligo paint-based high-throughput FISH in 384-well plates

For high-throughput FISH, cells were plated in 384-well CellCarrier plates (PerkinElmer) at a concentration of 2000 cells/well. After normoxic or hypoxic culture for up to 48 h, cells were fixed in 4% paraformaldehyde in phosphate-buffered saline (PBS) for 15 min. After two washes with PBS, cells were permeabilized in 0.5% Triton X-100, 0.5% Saponin/PBS for 20 min at room temperature (RT) and then incubated in 0.1 N HCl for 15 min at RT. Cells were kept in 50% formamide/2× SSC for at least 30 min at RT. A probe mix containing 60 ng of each library probe and fluorescently labeled readout probe and 40 μg human COT1 DNA (Invitrogen) in 1.1 ml of hybridization buffer (20% dextran sulfate, 50% formamide, 2× SSC, 1× Denhartd’s solution) was used; 15 µl of probe mix was added to the corresponding wells of a 384-well plate prepared using a JANUS liquid handler workstation (PerkinElmer, Waltham, MA). Probes were denatured at 85°C for 7 min, and the plate was incubated at 37°C overnight (16 h) for hybridization. After incubation, excess probe was washed off 3 times with 100 µl 2× SSC, 2× SSC at 42°C, and 2× SSC at 60°C for 5 min each. Cells were stained with DAPI in PBS (5 ng/ml) before imaging.

Image acquisition and analysis

Cells were imaged in 384-well plates on a CV7000 confocal high-throughput imaging system (Yokogawa Inc., Tokyo, Japan) using four solid-state laser lines (405, 488, 561, and 640 nm) for excitation, a 405/488/561/640 nm excitation dichroic mirror, a 40× air objective lens, a 568 emission dichroic mirror, and 2 sCMOS cameras (Andor), matched with appropriate emission bandpass filters (445/45, 525/50, 600/37, and 676/29 nm). Camera pixel binning was set to 2 × 2 for a resulting pixel size of 325 nm. Images were acquired in four channels as z-stacks of a total of 4 microns with images acquired at 1.0 μm steps. At least six randomly sampled fields were imaged per well. All image analysis steps were performed using Konstanz Information Miner (KNIME) software as described (Gudla et al., 2017). First, images from the same field of view and channel were maximally projected. Then, nuclei were segmented using the DAPI channel. The resulting nucleus ROI was used as the search region for the FISH spot detection algorithm in the Alexa 488, ATTO 550, and Cy5 channels, respectively. The normalized radial distance of each nucleus ROI pixel was then measured by dividing each absolute radial distance value by the per-cell maximum radial distance value. The nucleus border assumes a normalized value of 0, whereas the nucleus center has a normalized value of 1. The normalized absolute radial position of the FISH signal was calculated at the spot center pixel (Gudla et al., 2017).

Statistical analysis

For all the experiments, we conducted two to six times of experiments under the experimental condition of hypoxia or normoxia (biological replicates). We measured the nuclear radial position of genes (median of 4409 and 4600 cells imaged in normoxia and hypoxia per biological replicate, respectively) and the relative distance between the two genes (median of 1942 and 1687 cells imaged in normoxia and hypoxia per biological replicate, respectively) in each experiment. Genes or gene pairs whose radial or relative distance were subjected to the Wilcoxon rank sum test, and the genes which were statistically different (p value smaller than a Bonferroni significance threshold) between biological replicates conducted under the same experimental conditions were excluded from the analysis for evaluating the association between the experimental conditions and the radial and relative distances. The Wilcoxon rank sum test was used to compare the radial and relative distance between biological replicates. P values of less than a Bonferroni significance threshold determined by the number of pairwise comparisons between biological replicates for the radial or relative distances were considered to indicate statistical difference. For each gene and gene pair whose radial or relative distances were not statistically different between biological replicates as a result of the above analysis, we performed an experiment-level meta-analysis to evaluate the association between the experimental conditions and the nuclear radial and relative distances. Specifically, in each set of experiments performed under hypoxia and normoxia, we first calculated the ratio of means of radial and relative distances (hypoxia/normoxia) and its variance and the p value on the Wilcoxon rank sum test. Next, we calculated the weighted ratio of means of radial and relative distances and corresponding 95% confidence interval using the inverse variance-weighted average method (Friedrich et al., 2011). Additionally, we used Fisher’s method (Fisher, 1932) to combine multiple raw p values from independent experiments. P values were adjusted for multiple comparisons with the use of the method of Benjamini and Hochberg (Benjamini and Hochberg, 1995) to control the false discovery rate at the 0.05 level. All statistical analyses were performed using SAS version 9.4 (SAS Institute 193 Inc., Cary, NC). Statistical significance was defined as corrected p < 0.05.

mRNA-seq analysis

The mRNA expression profiles of MB231 cells treated with normoxia (21% O₂) or hypoxia (1% O₂) for 48 h were analyzed. RNA sequence library preparation, sequencing, mapping, and gene expression analysis were performed by DNAFORM (Yokohama, Japan). Qualities of total RNA were assessed by Bioanalyzer (Agilent, Santa Clara, CA). After poly (A) + RNA enrichment by NEBNext Poly(A) mRNA Magnetic Isolation Module (New England BioLabs, Ipswich, MA), RNA-seq library was prepared using the SMARTer Stranded Total RNA Sample Prep kit - HI Mammalian (Takara Bio Inc, Shiga, Japan) following the manufacturer’s instructions. Then first-strand synthesis was performed using an N6 primer. Illumina-specific indexed libraries were amplified by PCR. The libraries were sequenced on a NextSeq 500 sequencer (Illumina, San Diego, CA) to generate 50 and 25 nt paired end reads. Obtained reads were mapped to the human GRCh38.p10 genome using STAR (version 2.7.2b). Reads on annotated genes were counted using featureCounts (version 1.6.1). FPKM values were calculated from mapped reads by normalizing to total counts and transcript (Supplemental Table S8). Differentially expressed genes were detected using the DESeq2 package (version 1.20.0). Sequence results were deposited in the DDBJ Sequence Read Archive database (accession number: DRA013760).

Lactate assay

MB231 cells were cultured under normoxic or hypoxic conditions for 48 h, and lactate levels in the medium were measured using a Lactate Assay Kit as per the manufacturer’s instructions (Biovision, Milpitas, CA).

Immunofluorescence

Cells were fixed with 25 μl of 4% paraformaldehyde per well (10 min), permeabilized for 10 min (PBS/0.5% triton-X 100), and washed once with PBS/0.05% Tween-20. Then, cells were incubated for 1 h with primary antibodies (anti-HIF-1α, 1:100, BD Bioscience #610958, anti-phospho CREB, 1:200, CST #9198S) diluted in blocking buffer (PBS/0.05% Tween 20/5% bovine serum albumin). After three washes in wash buffer (PBS/0.5% Tween-20), cells were incubated for 1 h with secondary antibodies (anti-Mouse IgG 488, 1:1000 Invitrogen, #A11029; anti-rabbit IgG 488, 1:1000, Invitrogen, #A11034) followed by DAPI staining (5 μg/ml).

Abbreviations used:

CDT

CREB-direct target

FISH

fluorescence in situ hybridization

HIF

hypoxia-inducible factor

HIPMap

high-throughput imaging position mapping

HT

HIF-target

ICT

indirect CREB target

PBS

phosphate-buffered saline

RT

room temperature

VEGF

vascular endothelial growth factor.