The DEAD-box RNA helicase DDX28 suppresses cell migration and 3D growth and invasion in MDA-MB-231 cells by altering bioenergetics

Abstract

Hypoxia is a common characteristic of the tumor microenvironment leading to aggressive phenotypes. A major response to hypoxia is through the induction of gene programs by the hypoxia-inducible factors (HIF). Previously, we showed that the DEAD-box RNA helicase DDX28 negatively regulates hypoxic eIF4E2-directed translation through its interaction with HIF-2α. We hypothesized that DDX28 is a tumor suppressor that represses the oncogenic HIF-2α axis. Here, we overexpress DDX28 in MDA-MB-231 breast cancer and U87MG glioblastoma cells that have very low and normal endogenous levels of DDX28, respectively, compared with noncancerous HEK293. We show that DDX28 suppresses cell migration, spheroid growth, and invasion in MDA-MB-231, but not U87MG cells. However, suppression is not through the HIF-2α gene program, but through DDX28 impacting cellular bioenergetics. DDX28 levels altered how cells utilized mitochondrial respiration and glycolysis for ATP generation. Furthermore, the pharmacological inhibition of these processes specifically reversed the effects of DDX28 overexpression. This study shows that low endogenous DDX28 levels promote hypoxic migration, and growth/invasion in three-dimensional structures in cells that have a bioenergetic profile that favors glycolysis such as MDA-MB-231.

• New mechanistic insights for DDX28 in mediating the glycolysis and OXPHOS balance that impacts cellular migration and invasion in 2D and 3D contexts.

• Cell types with different bioenergetic profiles have different susceptibilities to inhibitors of OXPHOS and glycolysis based on their DDX28 abundance.

• Identification of DDX28 as a tumor suppressor.

INTRODUCTION

When oxygen is scarce, called hypoxia, cells typically shift their energy production from aerobic respiration that uses oxygen as the final electron acceptor in the electron transport chain (ETC) to the less efficient anaerobic glycolytic pathway. Hypoxia is important for many different aspects of normal human physiology such as embryonic development and stem cell differentiation (Simon and Keith, 2008). However, chronic hypoxia contributes to the pathology of several different diseases such as chronic obstructive pulmonary disease (Sarkar et al., 2003; Raguso and Luthy, 2011), stroke, heart disease, and cancer (Semenza, 2011). Indeed, hypoxia is a feature of the solid tumor microenvironment in many different cancers, developing due to the impaired function and morphology of tumor blood vessels and the rapid growth of tumor cells leading to an imbalance between oxygen supply and demand (Vaupel and Mayer, 2007;Bayer and Vaupel, 2012; Hu et al., 2012). Hypoxia leads to aggressive tumor phenotypes, increased metastasis, and resistance to treatments such as radiotherapy and chemotherapy (Vaupel and Mayer, 2007).

All nucleated cells sense hypoxia (Semenza, 2011) and trigger signaling pathways to increase chances of survival (Chandel et al., 1998). One of the main responses to hypoxia is through hypoxia-inducible factors 1 and 2 (HIF-1 and HIF-2) that induce the expression of over 100 genes that alter energy metabolism, autophagy, and translation (Dunwoodie, 2009). The HIFs are heterodimeric transcription factors that consist of an α and β subunit and bind to hypoxia-response elements (HRE) in the promoters of their target genes to affect their transcription (Keith and Simon, 2007; Klimova and Chandel, 2008). In normoxia, HIF-α is hydroxylated at two conserved proline residues in its oxygen-dependent degradation domain, subsequently allowing the von Hippel–Lindau protein to target it for proteasomal degradation (Keith and Simon, 2007; Semenza, 2011). In hypoxia, however, HIF-α is stabilized and dimerizes with the constitutively expressed HIF-β in the nucleus, where it binds to HREs on their target genes (Carroll and Ashcroft, 2006). HIF-2α also participates in hypoxic cap-dependent translation as member of an initiation complex alongside the cap-binding protein eIF4E2 to translate specific mRNAs (Tee et al., 2004; Uniacke et al., 2012). The gene programs induced by HIF-1α and HIF-2α are oncogenic (Semenza, 2003; Davis et al., 2022). Furthering our understanding of how these proteins are regulated will be an important contribution to cell biology and cancer research.

Our group has shown that DDX28 is a negative regulator of eIF4E2-directed translation through its interaction with HIF-2α in hypoxia (Evagelou et al., 2020). DDX28 is a member of the DEAD-box family of RNA helicases that have roles in RNA metabolism, including transcription, ribosome biogenesis, nucleocytoplasmic transport, and translation (Valgardsdottir and Prydz, 2003; Linder and Jankowsky, 2011). Though it can be detected in the nucleus, cytoplasm, and mitochondria, the primary characterized function of DDX28 is in the mitochondria, where it is involved in the biogenesis of the large subunit of the mitoribosome (Valgardsdottir et al., 2001; Tu and Barrientos, 2015). The mitoribosome is responsible for translating 13 mitochondrial proteins, all of which are subunits that make up ETC complexes (Zhao et al., 2019). We discovered an alternative function of DDX28, where depletion in hypoxia led to increased HIF-2α protein levels, enhanced translation of eIF4E2 target transcripts, and higher growth and proliferation of U87MG glioblastoma cells (Evagelou et al., 2020). Additionally, data retrieved from the Human Protein Atlas indicate little to no DDX28 protein expression across many cancers (Uhlen et al., 2015). We propose that DDX28 is a tumor suppressor, suppressing the oncogenic HIF-2α axis.

Here, we show that DDX28 influences cellular bioenergetics and acts as a tumor suppressor in MDA-MB-231 breast cancer cells by suppressing cell migration, spheroid growth, and invasion. We used two model cancer cell lines that displayed very low levels of DDX28 (MDA-MB-231), and normal levels (U87MG glioblastoma) compared with noncancerous HEK293 cells. We hypothesize that the low levels of DDX28 unleash oncogenic pathways in MDA-MB-231 cells. These two cell lines display different bioenergetic profiles, where MDA-MB-231 rely more on glycolysis for the generation of ATP where U87MG are more plastic in response to hypoxia. When we overexpressed (OE) DDX28, hypoxic cell migration, spheroid growth, and invasion were reduced in MDA-MB-231 breast cancer cells, but not in U87MG glioblastoma cells. This reduction was not achieved through regulating the translational activity of HIF-2α, as HIF-2α protein levels and the translation of hypoxia-specific transcripts were unaffected by DDX28 OE. Rather, endogenous and exogenous DDX28 preferentially localized to mitochondria, and the OE of DDX28 altered how cells utilized both mitochondrial respiration and glycolysis for the generation of ATP. Furthermore, the disruption of these processes through pharmacological inhibition reversed the effects of DDX28 OE on cancer phenotypes. This study highlights DDX28 as a tumor suppressor, where its low levels in select cancers with bioenergetic profiles that favor glycolysis can contribute to cancer progression.

RESULTS

DDX28 transcript and protein levels are lower in cancer cell lines relative to a non-cancerous cell line

Previous research from our group suggested that DDX28 may act as a tumor suppressor in hypoxia because DDX28-depleted cells exhibited increased oncogenic phenotypes such as eIF4E2-directed translation, cell viability, and cell proliferation in U87MG cells (Evagelou et al., 2020). If DDX28 were a tumor suppressor, we expected its levels to be lower in cancer cell lines compared with a noncancerous cell line. Therefore, we surveyed DDX28 mRNA levels in eight cancer cell lines and compared them with the noncancerous cell line HEK293. DDX28 mRNA levels were significantly lower in both normoxia and hypoxia in all tested cancer cell lines relative to HEK293 (Figure 1A). For example, in normoxic and hypoxic U87MG cells, DDX28 mRNA levels decreased by 1.6- and 2.3-fold, respectively, relative to HEK293. Furthermore, in normoxic and hypoxic MDA-MB-231 cells, DDX28 levels decreased by 3.1- and 18-fold, respectively, relative to HEK293. Next, we wanted to determine whether this decrease could be observed at the protein level. Similar to transcript levels, DDX28 protein was significantly decreased by 65.9% and 82.0% in hypoxic U87MG and MDA-MB-231 cells relative to HEK293, respectively (Figure 1B). Additionally, normoxic MDA-MB-231 cells displayed a significant 80.1% decrease in DDX28 protein compared with normoxic HEK293 (Figure 1B). Importantly, for this study, normoxic DDX28 protein levels are significantly lower in MDA-MB-231 cells even compared with U87MG (Figure 1B). Taken together, these data support that DDX28 may be a tumor suppressor because its levels are lower in cancer cell lines compared with a non–cancer cell line.

FIGURE 1:

Expression of DDX28 transcript and protein is significantly decreased in various cancerous cell lines relative to a noncancerous cell line in normoxia and hypoxia. (A) DDX28 transcript levels in 21% O₂ (Normoxia) and 1% O₂ (Hypoxia) were measured via qRT-PCR in U87MG, 786-0, MCF-7, MDA-MB-231, HCT-116, PC3, DU145, and HeLa cancer cell lines, and normalized to noncancerous HEK293 cells. Data (n ≥ 3) are means ± SEM relative to HEK293 cells and normalized to reference genes RPLP0 and RPL13A. The P values (* < 0.05) using a one-sample t test against a hypothetical mean of 0. (B) DDX28 protein levels in normoxia (N) and hypoxia (H) were measured via Western blot in wild-type U87MG and MDA-MB-231 cells and normalized to HEK293 cells. Actin was used as a loading control. Blots were quantified by densitometry. (C) Pathway enrichment analysis of differentially expressed genes in HepG2 cells with DDX28 knockdown. Statistically enriched gene ontology pathways related to biological processes were calculated using the PANTHER overrepresentation test with a significance criterion of P < 0.05. (D) Differentially expressed genes in DDX28 knockdown HepG2 cells compared with control cells with the top 10 up-regulated and top 10 down-regulated genes labeled. Differentially expressed genes had an absolute fold change greater than or equal to 2 and FDR < 0.05.

Additionally, we identified differentially expressed genes in shRNA-induced DDX28 knockdown HepG2 liver carcinoma cells using RNA-sequencing (RNA-seq) data collected for the ENCODE project (ENCSR222CSF) (Figure 1C). The top 10 up- and down-regulated gene names are displayed, where several have been connected to cancer progression. For example, SERPINB9 was found to be up-regulated upon DDX28 depletion. SERPINB9, or serine protease inhibitor of the effector T-cell–derived molecule granzyme B, prevents caspase-mediated tumor apoptosis and is commonly amplified in cancer (Ibanez-Molero et al., 2022). The gene encoding fibrinogen-like protein 1, which is decreased in liver cancer compared with normal tissue, was further down-regulated with DDX28 depletion (Yu et al., 2021). Finally, DDX28 depletion led to decreased MBL2 expression, which correlates with significantly reduced patient survival in hepatocellular carcinoma (Liao et al., 2023). Moreover, we performed pathway enrichment analysis on these differentially expressed genes and identified statistically enriched gene ontology pathways related to biological processes (Figure 1D). Several up- and down-regulated pathways can be directly related to cancer progression and are consistent with the possibility that DDX28 is a tumor suppressor. For example, when DDX28 levels were depleted, “negative regulation of carbohydrate metabolic process” was down-regulated. An increase in carbohydrate metabolism (i.e., glycolysis, glycogen metabolism, and gluconeogenesis) is characteristic of tumor cells (Bose et al., 2021). Furthermore, increased cell growth contributes to the incidence of the cancer hallmark of “sustained proliferative signaling” (Hanahan and Weinberg, 2011), and we observed that the “negative regulation of growth” pathway was down-regulated upon DDX28 depletion.

DDX28 overexpression differentially alters hypoxic cell migration, spheroid growth, and spheroid invasion in U87MG and MDA-MB-231 cells

Because DDX28 levels are low in the aggressive triple-negative MDA-MB-231, we hypothesized that DDX28 OE would suppress hallmarks of this cell line such as migration, and spheroid growth and invasion. Conversely, we hypothesized that DDX28 may be less important in influencing these hallmarks in U87MG as its levels are not much lower than HEK293 (Figure 1). We used previously established U87MG cell lines stably expressing FLAG-DDX28 or an empty-FLAG control vector (Evagelou et al., 2020). We also generated stable cell lines in MDA-MB-231 cells using the same vectors.

We measured the effect of DDX28 OE on cell migration using a Transwell migration assay, where we expected DDX28 OE to decrease cell migration only in MDA-MB-231. Indeed, the relative cell migration of MDA-MB-231 DDX28 OE cells significantly decreased by 10.8-fold only in hypoxia (Figure 2A). Conversely, U87MG DDX28 OE cells displayed a statistically significant 3-fold increase in cell migration only in hypoxia (Figure 2A). These data show that DDX28 OE suppresses cell migration in hypoxic MDA-MB-231.

FIGURE 2:

DDX28 overexpression differentially alters hypoxic cell migration, spheroid growth, and spheroid invasion in U87MG and MDA-MB-231 cells. (A) Transwell migration assays were performed with control (Empty-FLAG vector) and FLAG-DDX28 overexpressing (OE) U87MG and MDA-MB-231 cells. Cells were exposed to 18 h of either 21% O₂ (normoxia) or 1% O₂ (hypoxia). Migrated cells were fixed, stained, imaged, and quantified using ImageJ Fiji. Data (n ≥ 3) represent the number of cells migrated compared with the empty control in normoxia. Error bars represent the SEM. *, P < 0.05 using a one-way ANOVA. (B) U87MG and MDA-MB-231 spheroids stably expressing control or FLAG-DDX28 OE vector. Invasion spheroids were imbedded in Matrigel. Images of spheroids were captured every 24 h for the next 72 h (invasion) or 96 h (growth). Representative images at 72 or 96 h are shown. Scale bar, 500 µm. (C) Surface area of spheroids was measured using ImageJ Fiji software. Data (n ≥ 3) represent the average percentage of change of spheroid size relative to control, calculated by subtracting the spheroid size at 0 h from spheroid size at each timepoint and dividing that over 0 h. (D) Invasion was measured as the daily change in protrusion size measured over time. Error bars represent the SEM. The P values (* < 0.05) using a one-way ANOVA and Tukey's HSD post-hoc test.

We next seeded spheroids, an avascular tumor model containing a hypoxic gradient, to further probe the role of DDX28 in cells in a three-dimensional setting. Spheroids of U87MG or MDA-MB-231 DDX28 OE and control cells were monitored for their growth and invasion over 96 and 72 h, respectively (Figure 2B). U87MG DDX28 OE spheroids displayed a significant 32% reduction in growth by 96 h relative to controls (Figure 2C). However, U87MG DDX28 OE increased invasion by 103% in spheroids compared with controls at 72 h (Figure 2D). Like U87MG OE spheroids, MDA-MB-231 DDX28 OE spheroids had an 88% reduction in growth compared with controls at 96 h (Figure 2C). However, opposite to U87MG OE spheroids, DDX28 OE significantly suppressed invasion by 42% to the controls after 72 h (Figure 2D). Importantly, DDX28 overexpression does not significantly affect cell viability in normoxia and hypoxia relative to controls (Supplemental Figure S1). Our data show that DDX28 OE reduces the growth of spheroids in both cell lines, but has opposite effects for invasion where it is increased in U87MG and suppressed in MDA-MB-231.

HIF-2α protein levels and the polysome association of eIF4E2 target transcripts are unaffected by DDX28 overexpression, but the mRNA abundance of HIF-2α target genes is increased

We showed previously that DDX28 interacted with and reduced HIF-2α protein levels and translational activity to increase cancer hallmarks in U87MG cells through shRNA depletion of DDX28 (Evagelou et al., 2020). However, we found that DDX28 OE in U87MG had no effect on HIF-2α protein levels (Evagelou et al., 2020). We attributed this to a possible equal stoichiometry between HIF-2α and DDX28, whereby only DDX28 reduction produces an observable effect. Here, because MDA-MB-231 cells have much lower endogenous DDX28 levels, we expected that DDX28 OE would decrease HIF-2α levels. However, HIF-2α protein levels were not affected by DDX28 OE in MDA-MB-231 (Figure 3A).

FIGURE 3:

HIF-2α protein levels and the polysome association of eIF4E2 target transcripts are unaffected by DDX28 overexpression. (A) Representative Western blot (n ≥ 4) of HIF-2α protein levels in MDA-MB-231 cells stably overexpressing either FLAG-DDX28 (DDX28 OE) or an empty-FLAG vector (control) after 24 h incubation in either normoxia (21% O₂) or hypoxia (1% O₂). DDX28 indicates endogenous protein levels, while FLAG indicates exogenous FLAG-DDX28 since the DDX28 antibody does not cross-react with the FLAG-tagged protein. Actin was used as a loading control. Blots were quantified by densitometry. (B) RNA was isolated from monosome and polysome fractions that were collected from hypoxic U87MG and MDA-MB-231 control and DDX28 OE cells (n = 3). qRT-PCR was performed to examine the expression levels of eIF4E2 targets (EGFR, IGF1R, and EPAS1) and eIF4E targets (EEF2 and HSP90ab1) in polysomes relative to their monosome. Polysome to monosome expression ratio was calculated using ΔΔC_q values and normalized to reference genes RPLP0 and GAPDH. Significance was determined using an unpaired two-sample t test. The P values (* < 0.05). (C) The mRNA abundance of HIF-2α target genes was measured via qRT-PCR in U87MG and (D) MDA-MB-231 DDX28 OE cells after 24 h incubation in hypoxia. Data (n = 3) are means ± SEM for DDX28 OE cells relative to those for control cells and normalized to values for endogenous control genes RPLP0 and RPL13A. *, P < 0.05 using a one-sample t test against a hypothetical mean of 0. POU5F1, POU Class 5 Homeobox 1; ITPR1, Inositol 1,4,5-trisphosphate receptor type 1; IGFBP3, insulin-like growth factor binding protein 3; LOXL2, lysyl oxidase like 2; CITED2, Cbp/P300 interacting Transactivator with Glu/Asp rich carboxy-terminal domain 2.

We next performed polysome fractionation and qRT-PCR to assess the activity of eIF4E2-mediated translation, which is positively regulated by HIF-2α. The polysome association of eIF4E2 target transcripts EGF receptor (EGFR), insulin-like growth factor 1 receptor (IGF1R), and HIF-2α (EPAS1) did not change in hypoxia with DDX28 OE in either cell line (Figure 3B). This contrasts with our observation of increased eIF4E2-directed translation via higher HIF-2α levels through DDX28 depletion in U87MG (Evagelou et al., 2020). We included two eIF4E transcript targets eukaryotic translation elongation factor 2 (EEF2) and heat shock protein 90 alpha family class B member 1 (HSP90ab1) as unrelated controls, which were not affected by DDX28 OE.

We also assessed the influence of DDX28 OE on the mRNA abundance of HIF-2α target genes in hypoxic DDX28 OE and control cells. Five genes that are preferentially regulated by HIF-2α compared with HIF-1α were chosen (Aprelikova et al., 2006; Elvidge et al., 2006; Pawlus et al., 2012; Stiehl et al., 2012; Isono et al., 2016). In U87MG DDX28 OE cells, only one of the five genes showed a significant (2.8-fold) increase in mRNA abundance compared with controls (Figure 3C). In MDA-MB-231 DDX28 OE cells, however, four of the five genes displayed over 2-fold significant increases in their mRNA abundance (Figure 3D). Our data show that DDX28 OE does not alter HIF-2α protein levels nor its translational activity, but does increase the mRNA abundance of HIF-2α target genes. Because we showed previously that DDX28 only directly impacts the translation activity of HIF-2α, our investigation into the tumor-suppressor function of DDX28 shifted to its major role in mitochondria. Indeed, alterations in mitochondrial energy output can indirectly alter HIF-dependent gene programs (Fuhrmann and Brune, 2017; Nagao et al., 2019).

Exogenous FLAG-DDX28 significantly accumulates in mitochondria in both U87MG and MDA-MB-231 cells in normoxia and hypoxia

Because DDX28 has been found in the nucleus, cytoplasm, and mitochondria (Valgardsdottir and Prydz, 2003), we examined the cellular distribution of both endogenous DDX28 and exogenous FLAG-DDX28 in both cell lines across normoxic and hypoxic conditions. Cell lysates from U87MG (Figure 4A) and MDA-MB-231 (Figure 4B) DDX28 OE and control cells were fractionated into nuclear, cytosolic, and mitochondrial compartments and analyzed by Western blot. Although DDX28 increased in all compartments in OE cells compared with controls, only mitochondria displayed significantly more DDX28 in both cell lines across normoxia and hypoxia (Figure 4, A and B). The only other compartment that gained significantly more DDX28 in OE cells relative to controls was the nucleus in normoxic MDA-MB-231 cells (Figure 4B). Comparing DDX28 levels between compartments would require the fractionation of each compartment to be equally efficient, which is not likely. Considering this caveat, the observed cellular distribution of DDX28 (Figure 4, A and B) follows previously published data where it is most abundant in the mitochondria (Valgardsdottir and Prydz, 2003). However, our data show that DDX28 could have an important nuclear role (Figure 4, A and B). These data suggest that the DDX28-dependent cell migration, spheroid growth, and invasion phenotypes observed in Figure 2 could be due to the mitochondrial role of DDX28.

FIGURE 4:

Exogenous DDX28 significantly accumulates in mitochondria in both U87MG and MDA-MB-231 cells in both oxygen conditions. (A) U87MG and (B) MDA-MB-231 cells stably expressing an empty-FLAG vector (control) or FLAG-DDX28 (OE) were incubated in 21% O₂ (N; normoxia) or 1% O₂ (H; hypoxia) for 24 h. Cells were separated into cytosolic (Cyto), mitochondrial (Mito), and nuclear (Nuc) fractions. Fractions were analyzed by Western blot. GAPDH, Tom20, and Lamin A/C were used as cytosolic, mitochondrial, and nuclear markers, respectively. The arrows indicate bands that are the ∼60 kDa endogenous DDX28 and the ∼63 kDa exogenous FLAG-DDX28. Blots were quantified by densitometry. Data (n = 3), mean ± SEM for endogenous DDX28 levels in control cells or endogenous DDX28 + exogenous FLAG-DDX28 in DDX28 OE cells. DDX28 quantification was normalized to the respective subcellular marker and made relative to the cytosolic fraction in normoxic control cells indicated by the dotted line at 1. *, P < 0.05 using a two-way ANOVA comparing FLAG-DDX28 OE cells with control within each fraction.

DDX28 overexpression differentially alters glycolytic and mitochondrial ATP production in normoxia and hypoxia in U87MG and MDA-MB-231 cells

To shed light on how DDX28 may act as a tumor suppressor through its mitochondrial role, we investigated whether U87MG and MDA-MB-231 cells displayed different DDX28-dependent ATP production efficiencies originating from glycolysis and/or mitochondrial oxidative phosphorylation (OXPHOS). Because DDX28 plays an important role in the translation of many ETC complex subunits (Tu and Barrientos, 2015), we postulated that overexpressing DDX28 may increase OXPHOS efficiency. We utilized the seahorse flux analyzer to measure the basal oxygen consumption rates (OCR) and extracellular acidification rates (ECAR) of U87MG and MDA-MB-231 DDX28 OE and control cells in normoxia and hypoxia. When measured alongside the timed additions of ETC inhibitors oligomycin, rotenone, and antimycin A, the basal ATP production rates from either mitochondrial respiration or glycolysis were calculated. First, in normoxic U87MG cells, the initial level (0 min) of oxygen consumption was 29.8% lower (from 73.6 to 51.7 pmol/min) and the extracellular acidification was 53.0% significantly lower (from 12.0 to 5.6 mpH/min) in DDX28 OE cells compared with control cells (Figure 5A). When plotted as ATP production rates, DDX28 OE cells displayed significantly less total ATP production in normoxia originating from both the mitochondria (from 314.1 to 144.2 pmol/min) and glycolysis (from 129.1 to 56.5 pmol/min) compared with control cells (Figure 5B). These observations were reversed in hypoxia, where DDX28 OE cells displayed a 34.6% higher basal OCR (66.5 to 101.6 pmol/min) and a 21.3% higher basal ECAR (26.4 to 33.6 mpH/min) compared with control cells at 0 min (Figure 5C). Moreover, in hypoxia, U87MG DDX28 OE cells produced significantly more ATP from both glycolysis (from 372.4 to 474.0 pmol/min) and the mitochondria (from 147.8 to 200.1 pmol/min) compared with control cells (Figure 5D).

FIGURE 5:

DDX28 overexpression leads to decreased ATP production in U87MG cells and increased ATP production in MDA-MB-231 cells in normoxia, and is reversed in hypoxia. (A) Basal oxygen consumption, extracellular acidification, and ATP production were measured using the ATP Rate Assay Kit and Seahorse Metabolic Flux Analyzer. Data represent means ± SD OCR (pmol/min), ECAR (mpH/min), and (B) ATP production rate (pmol/min) in normoxic U87MG, (C and D) hypoxic U87MG, (E and F) normoxic MDA-MB-231, and (G and H) hypoxic MDA-MB-231 cells stably overexpressing FLAG-DDX28 (DDX28 OE) compared with control cells expressing an empty-FLAG vector (Control). Error bars represent the SEM. The P values (* < 0.05) using a one-way ANOVA and Tukey's HSD post-hoc test.

In normoxic MDA-MB-231 DDX28 OE cells, the basal (0 min) level of oxygen consumption significantly increased from 51.3 to 85.7 pmol/min (40.2%) and extracellular acidification increased from 19.3 to 25.3 mpH/min (23.9%) compared with control cells (Figure 5E). When calculated into ATP production rates, DDX28 OE cells displayed an increase in ATP produced through glycolysis (from 279.9 to 352.7 pmol/min), a larger increase in ATP produced by the mitochondria (from 147.6 to 386.4 pmol/min), and an overall significant increase in total ATP production from 427.5 to 739.1 pmol/min relative to control cells in normoxia (Figure 5F). Furthermore, the proportion of ATP produced from glycolysis and OXPHOS shifted from 2:1, respectively, in control cells to 1:1 in DDX28 OE cells. The opposite was observed in hypoxia, where the OCR significantly decreased from 127.5 to 88.7 pmol/min (30.4%) in DDX28 OE cells compared with control cells, and ECAR decreased from 31.6 to 23.9 mpH/min (24.4%) at 0 min (Figure 5G). In hypoxia, mitochondrial ATP production decreased from 411.0 to 307.6 pmol/min in MDA-MB-231 DDX28 OE compared with control cells, while glycolytic ATP production decreased from 517.4 to 414.5 pmol/min in DDX28 OE cells (Figure 5H). In summary, DDX28 OE in U87MG reduces OCR, ECAR, and total ATP output in normoxia, while the opposite is observed in MDA-MB-231, along with a shift to an even distribution of ATP generated from glycolysis and OXPHOS. In hypoxia, DDX28 OE increases OCR, ECAR, and total ATP levels in U87MG, but all these decrease in MDA-MB-231. These data show that two cell lines with different basal levels of DDX28 have different energy outputs that have opposite responses to exogenous DDX28.

DDX28 overexpression does not affect mitochondrial area, branching or morphology, but leads to an increase in U87MG mtDNA in normoxia and hypoxia

Next, we investigated whether the observed changes in glycolytic and mitochondrial ATP production could be attributed to alterations in mitochondrial abundance and/or morphology. We quantified mitochondrial area, branching, and morphology in U87MG and MDA-MB-231 DDX28 OE and control cells in normoxia and hypoxia using immunofluorescence (Figure 6A). Overexpressing DDX28 in U87MG cells caused a decrease in normalized mitochondrial area compared with control cells, but only in hypoxia (Figure 6B). In MDA-MB-231 cells, hypoxia decreased mitochondrial area in both control and DDX28 OE cells compared with normoxia (Figure 6C), which could be due to the induction of mitophagy known to be an adaptive response to hypoxic stress (Fuhrmann et al., 2013). Fragmentation leading to decreased branching and connections between mitochondria, as well as rounder versus more rod-shaped mitochondria, can be indicative of stress-induced damage (Karbowski and Youle, 2003; Rafelski, 2013; Miyazono et al., 2018). We did not observe any significant changes to mitochondrial morphology (form factor) or branching in either DDX28 OE cell lines compared with controls (Figure 6, B and C). Mitochondrial DNA (mtDNA) abundance was also measured in U87MG and MDA-MB-231 DDX28 OE and control cells in both normoxia and hypoxia. The mtDNA in a cell not only reflects changes to mitochondrial abundance, but also changes to mitochondrial genome copy number per cell that can indicate mitochondrial dysfunction (Malik and Czajka, 2013). U87MG DDX28 OE cells displayed significant increases in mtDNA compared with control cells in both normoxia and hypoxia (Figure 6D). However, MDA-MB-231 DDX28 OE cells displayed no significant difference in mtDNA copy number relative to control cells (Figure 6E). The protein levels of several ETC complex subunits, including the mitochondrially encoded MTCO2, did not significantly change in either cell line in response to DDX28 OE (Supplemental Figure S2, A and B), except ATP5A1 significantly increased only in U87MG cells (Supplemental Figure S2A). Overall, these data show that DDX28 OE in U87MG decreases mitochondrial abundance and increases mtDNA copy number, while having no effect on these mitochondrial properties in MDA-MB-231.

FIGURE 6:

In only U87MG cells, DDX28 OE increases mtDNA and decreases mitochondrial area, but does not affect branching or morphology. (A) Representative images are shown of mitochondrial area, branching, and morphology that were assessed via immunofluorescence with the marker Tom20 (red) in U87MG and MDA-MB-231 control cells stably expressing an empty-FLAG vector and overexpressing (OE) cells stably expressing FLAG-DDX28 in normoxia and hypoxia. Nuclei stained with Hoechst (blue). (B) Bar graphs display the corrected total cell fluorescence, branches are per mitochondria, and morphology (rounder mitochondria produce a lower form factor value) in U87MG and (C) MDA-MB-231 control and DDX28 OE cells (n = 8). Images were analyzed using the Mitochondrial Analyzer plugin on Fiji ImageJ. Scale bar, 20 µm. The P values (* < 0.05) using a two-way ANOVA. (D) The mean (n = 3) ± SEM mtDNA copy number of DDX28 OE cells and control cells was measured via qPCR of mitochondrial mt-ND1 gene relative to single-copy nuclear gene HBB as a reference in normoxia and hypoxia in U87MG and (E) MDA-MB-231 cells. *, P < 0.05 using two-way ANOVA and Tukey's test.

The DDX28-dependent suppression of cell migration and spheroid growth and invasion is dependent on active OXPHOS and glycolysis in MDA-MB-231

We next investigated whether the DDX28-dependent effects on cell migration and spheroid growth/invasion (Figure 2) could be due to the differences in ATP output observed between these two cell lines (Figure 5) by using inhibitors of glycolysis or OXPHOS. We treated cells with effective and nontoxic doses (Supplemental Figure S3) of the ETC complex I inhibitor rotenone to inhibit OXPHOS (Pamies et al., 2018), and the glucose analogue 2-deoxy-D-glucose (2-DG) to competitively inhibit step two of glycolysis (Fujita et al., 2020). In U87MG, because cell migration was increased by 3-fold in a DDX28-dependent manner only in hypoxia (Figure 2A), we treated hypoxic control and OE cells with rotenone and 2-DG for 18 h in a Transwell assay. Rotenone treatment not only prevented the increase in hypoxic U87MG cell migration but reduced it by 1.5-fold in DDX28 OE cells compared with controls, while 2-DG prevented any increase (Figure 7A). Interestingly, U87MG control cell migration was not affected by either treatment. In MDA-MB-231, hypoxia increased cell migration by 4.7-fold (Figure 2A), which was prevented by rotenone (Figure 7B). On the other hand, DDX28 OE cell migration was not affected by rotenone treatment (Figure 7B). 2-DG treatment decreased MDA-MB-231 control cell migration by more than 2-fold compared with untreated hypoxic controls (Figure 7B). 2-DG treatment in DDX28 OE cells not only prevented the more than 10-fold decrease in hypoxic cell migration compared with control cells (Figure 2A), but caused a 2-fold increase (Figure 7B). These data show that MDA-MB-231 cell migration was much more sensitive to rotenone treatment compared with U87MG. The DDX28-dependent increase in hypoxic cell migration in U87MG relies more so on OXPHOS but also glycolysis. The over 10-fold hypoxic decrease in cell migration mediated by DDX28 in MDA-MB-231 was nullified by both treatments. This suggests that the suppression of hypoxic cell migration by DDX28 in MDA-MB-231 relies on active OXPHOS and glycolysis.

FIGURE 7:

Disrupting glycolysis or OXPHOS in DDX28-overexpressing cells reverses effects of DDX28 overexpression on hypoxic cell migration, spheroid growth, and spheroid invasion and in U87MG and MDA-MB-231 cells. (A) Transwell migration assays were performed with Empty-FLAG (control) and FLAG-DDX28 overexpressing (OE) U87MG and (B) MDA-MB-231 cells. Cells were treated with either rotenone or 2-DG 30 min before their 18 h incubation in 1% O₂. Migrated cells were fixed, stained, imaged, and quantified using ImageJ Fiji. Data (n ≥ 3) represent the number of cells migrated compared with the control in normoxia. Error bars represent the SEM. The P values (* < 0.05) using a one-way ANOVA. (C) U87MG and (D) MDA-MB-231 spheroids stably expressing a control or FLAG-DDX28 OE vector were seeded as single spheroids per well, and grown for 72 h. Images of spheroids were captured every 24 h for the next 72 (invasion) or 96 (growth) h. Spheroids were treated with either rotenone or 2-DG at 0 h. (E and F) Invasion spheroids were also imbedded in Matrigel. Surface area of spheroids was measured using ImageJ Fiji software. Data (n ≥ 3) represent the average percentage of change of spheroid size relative to empty control, calculated by subtracting the spheroid size at 0 h from spheroid size at each timepoint and dividing that over 0 h. Invasion was measured in the same way, with the daily change in protrusion size measured over time. Error bars represent the SEM. The P values (* < 0.05) using a one-way ANOVA and Tukey's HSD post-hoc test.

Next, we tested whether the DDX28-dependent suppression of spheroid growth (Figure 2C) was dependent on OXPHOS and/or glycolysis. Although spheroid growth was repressed by both drugs, the DDX28-dependent decrease in spheroid growth seen in Figure 2C was abolished by rotenone but not 2-DG (Figure 7C). On the other hand, while MDA-MB-231 control spheroids were minimally affected by the drugs, the DDX28 OE spheroids grew over 14-fold larger than control spheroids in the presence of either rotenone or 2-DG (Figure 7D). These data suggest that U87MG spheroid growth is dependent on both OXPHOS and glycolysis, but the DDX28-dependent suppression of growth depends more on OXPHOS. However, MDA-MB-231 spheroid growth is not sensitive to OXPHOS and glycolysis inhibition, but the DDX28-dependent suppression of growth relies on active OXPHOS and glycolysis.

U87MG spheroids invaded into Matrigel only 6% as much as their untreated control counterparts in the presence of rotenone (Figure 2D), but were minimally affected by 2-DG (Figure 7E). The DDX28-dependent 103% increase in U87MG spheroid invasion (Figure 2D) was abolished by both drugs (Figure 7E). MDA-MB-231 spheroid invasion was reduced by 35% by rotenone, but unaffected by 2-DG (Figure 7F) relative to untreated controls (Figure 2D). The DDX28-dependent suppression of spheroid invasion was negated by rotenone and 2-DG where DDX28 OE spheroids invaded better than controls in the presence of rotenone (Figure 7F). These data show that while U87MG spheroid invasion was more susceptible to inhibition of OXPHOS compared with glycolysis, the DDX28-dependent increase in spheroid invasion was suppressed by targeting either glycolysis or OXPHOS. In MDA-MB-231 spheroids, disrupting only OXPHOS led to a reduction in invasion. However, the DDX28-dependent suppression of spheroid invasion was prevented by targeting glycolysis, and even more so by disrupting OXPHOS.

DISCUSSION

We sought to determine whether DDX28 could be acting as a tumor suppressor. Most commonly, tumor-suppressor genes contain loss of function mutations that reduce their expression, preventing their protein products from suppressing oncogenic pathways (Wang et al., 2018). It is important to note that due to differences in the mutational landscape of cancers, a protein can be a tumor suppressor in one cancer but not another (Schneider et al., 2017). Indeed, when we surveyed several cancer cell lines for DDX28 mRNA and protein levels, DDX28 expression was reduced to varying degrees in all cancer cell lines compared with HEK293 (Figure 1, A and B). In addition, surveying databases such as the human protein atlas and performing our own bioinformatics analyses (Figure 1, C and D) demonstrated that DDX28 could be a tumor suppressor in some cancers even though evidence of specific mutations to DDX28 is unknown. We chose MDA-MB-231 as a model cell line, where DDX28 could be a tumor suppressor due to its low DDX28 expression and U87MG as a model where DDX28 may not be a tumor suppressor.

Based on our previous research (Evagelou et al., 2020), we hypothesized that DDX28 OE could suppress the cytoplasmic role of HIF-2α in hypoxic translation, which is oncogenic. However, the levels of HIF-2α and its translational activity remain unchanged with DDX28 OE (Figure 3, A and B). In our previous study, DDX28 shRNA knockdown produced significant increases to hypoxic cell viability and proliferation in U87MG, but OE had no effect (Evagelou et al., 2020). One possibility is that DDX28 protein abundance is close to or equal to HIF-2α levels, where exogenous DDX28 would have minimal impact. In this study, even in MDA-MB-231 cells with lower endogenous DDX28 levels, HIF-2α proteins levels and translation activity could not be suppressed by exogenous DDX28 (Figure 3, A and B). DDX28 levels could be at functional saturation in the cytoplasm. Although the translation activity of HIF-2α was unaffected by DDX28 OE, the mRNA abundance of HIF-2α gene targets was affected. When investigating changes in mRNA abundance to HIF-2α–dependent genes upon DDX28 OE, we observed opposite responses between cell lines (Figure 3, C and D). A fraction of DDX28 is in the nucleus (Figure 4A) and therefore could have a yet uncharacterized role in transcriptional regulation like other DEAD-box proteins (Fuller-Pace and Nicol, 2012). Exploring this possibility would be an exciting avenue for future studies. However, HIFs are known to be responsible for inducing many glycolytic genes and can be regulated by mitochondrial byproducts (Fuhrmann and Brune, 2017; Nagao et al., 2019). The up-regulation of HIF-2α target genes could be a response to the change in bioenergetics due to overexpressing DDX28 (Figure 5) rather than a direct effect of DDX28 on HIF-2α. Therefore, the opposite regulation of the HIF-2α target genes in DDX28 OE cells in both cell lines (Figure 3, C and D) could be a feedback response due to the opposite effects DDX28 OE has on their bioenergetic profiles (Figure 5). The major characterized role of DDX28 is within the mitochondria (Tu and Barrientos, 2015), which is supported by our data (Figure 4). Therefore, because one of the hallmarks of cancer is the deregulation of cellular metabolism (Hanahan, 2022), we turned our attention to the mitochondrion.

In U87MG, DDX28 OE significantly increased hypoxic ATP production (Figure 5D) without affecting mitochondrial morphology and networks (Figure 6, A and B). However, mitochondrial area decreased and mtDNA content increased (Figure 6, B and D). This suggests that although hypoxia leads to a reduction in mitochondrial mass (perhaps through mitophagy-induced clearing), DDX28 OE may increase ETC efficiency. This could be achieved by stabilizing certain ETC complexes or altering their subunit composition to improve electron transfer (Fuhrmann and Brune, 2017). Yet, the subset of ETC subunits that we examined did not generally change their protein levels in response to DDX28 OE (Supplemental Figure S2). This increase in hypoxic ATP levels can be correlated to increases in hypoxic DDX28 OE cell migration (Figure 2A) and spheroid invasion (Figure 2D). However, DDX28 OE spheroid growth was significantly slowed (Figure 2C). A spheroid contains a mixed population of normoxic and hypoxic cells with the hypoxic cells making up most of the core. Cellular metabolites and by-products such as hydrogen ions and lactate from the hypoxic cells can act as signaling molecules to induce the epithelial-mesenchymal transition (EMT) and the transcription of genes to initiate migration and invasion (Parlani et al., 2023). Meanwhile, the repression of spheroid growth upon DDX28 OE could be attributed to the significant decrease in normoxic cell ATP production (Figure 5B).

In MDA-MB-231 DDX28 OE cells, we observed a decrease in hypoxic ATP production (Figure 5H) and a decrease in mitochondrial mass that was due to hypoxia but not DDX28-dependent (Figure 6C). This suggests that mitochondrial ETC efficiency is reduced upon DDX28 OE in this cell line. To understand why glycolysis activity also decreases upon DDX28 OE, one could investigate the translation or activity of glycolytic enzymes. The lower hypoxic ATP production (Figure 5H) is likely responsible for the inability of DDX28 OE MDA-MB-231 cells to increase their migration in response to hypoxia compared with control cells (Figure 2A) considering that migration is an energy costly process (Parlani et al., 2023). MDA-MB-231 cells favor glycolysis (Lunetti et al., 2019) supported by the 2:1 glycolysis:OXPHOS ratio in normoxia that was somewhat maintained in hypoxia (Figure 5F and H). Uniquely, the normoxic ratio of the control was disturbed by DDX28 OE, causing a shift from 2:1 to 1:1 glycolysis:OXPHOS usage (Figure 5F). It is not surprising that DDX28 would shift cells toward OXPHOS due to its role in assembling mitoribosomes that synthesize ETC components (Tu and Barrientos, 2015). However, MDA-MB-231 cells may not respond well to an increase in OXPHOS as this could cause an increase in reactive oxygen species production (An et al., 2024). Because spheroids are composed of both normoxic and hypoxic cells, the concurrent decrease of ATP in hypoxia and increase of ATP in normoxic DDX28 OE cells could be responsible for the decrease in spheroid growth and invasion (Figure 2, C and D). On the other hand, U87MG cells are metabolically plastic (Duraj et al., 2021), evidenced by their ability to switch from a 1:3 glycolysis:OXPHOS ratio in normoxia to a 3:1 ratio in hypoxia, which is unaffected by DDX28 OE (Figure 5, B and D). U87MG OXPHOS usage was not affected by DDX28 OE perhaps because this cell line already has a high baseline glycolysis:OXPHOS ratio of 1:3.

Treating U87MG and MDA-MB-231 cells with OXPHOS or glycolysis inhibitors negated the DDX28-dependent effects observed on cell migration, spheroid growth, and invasion. However, these inhibitors had opposite effects on the cell lines. For example, where DDX28 OE increased cell migration and spheroid invasion in U87MG (Figure 2, A and D), both inhibitors caused DDX28 OE cells to more closely resemble control cells (Figure 7, A and E). Conversely, DDX28 OE suppressed cell migration and spheroid growth and invasion in MDA-MDA-231 cells (Figure 2), but the inhibitors significantly increased all these processes, effectively reversing the effect of DDX28 OE (Figure 7, B, D, and F). Interestingly, although MDA-MB-231 control spheroids were impaired in their growth and invasion by the inhibitors, DDX28 OE was protective and allowed spheroids to thrive in the presence of the inhibitors. This protective effect was not observed in U87MG cells. Therefore, a chemotherapeutic such as rotenone may not be a suitable treatment for cancers with a high glycolysis:OXPHOS ratio and high DDX28 levels.

Although we show that active OXPHOS and glycolysis are required for the DDX28-dependent effects on cell migration, spheroid growth, and invasion (Figure 7), this does not exclude the possibility that DDX28 could produce these effects outside the mitochondria. Like many other DEAD-box family members that possess multiple functions (Linder and Jankowsky, 2011), we have previously characterized functions of DDX28 in translation and have observed it in the nucleus. Moreover, we did not observe global changes to mitochondrial abundance, morphology, or ETC subunit levels in response to DDX28 OE (Figure 6; Supplemental Figure S2). The data that we show in Figure 5 display a DDX28-dependent global effect on ATP production, independent of its source. Perhaps DDX28 has a nuclear signaling role and/or influences an upstream energy pathway such as AMPK to impose effects on both glycolysis and OXPHOS. Indeed, hyperactive AMPK has been shown to cause resistance to 2-DG treatment (Figure 7, B, D and F) (Schmidt and O'Donnell, 2021). Other proteins involved in mitoribosome assembly such as GTPBP5 (Cipullo et al., 2021) or GTPBP10 (Lavdovskaia et al., 2018) may be involved similarly to DDX28 in cancer biology and could be candidates for further evaluation. We show that in cancer cells with a bioenergetic profile that favors glycolysis, high DDX28 causes the suppression of cell migration, spheroid growth (independent of cell viability), and invasion. Tumor suppressors can act as “double agents” in the cell, being able to confer both tumor suppressive and oncogenic effects depending on the context (Datta et al., 2020). Based on our data, it is possible that DDX28 might play pleotropic roles in cancer progression through influencing cellular bioenergetics.

MATERIALS AND METHODS

Request a protocol through Bio-protocol

Cell culture

U87MG, MDA-MB-231, 786-0, MCF-7, HCT116, PC3, DU-145, HeLa, and HEK293 cells were obtained from the ATCC and maintained in either DMEM (U87MG, MDA-MB-231, MCF-7, HCT116, HeLa, and HEK293) or Roswell Park Memorial Institute (786-0, PC3, and DU-145) supplemented with FBS, as suggested. Normoxic cells were maintained at 37°C, 21% O₂ and 5% CO₂ levels. Hypoxic cells were maintained at 37°C and 5% CO₂ and incubated at 1% O₂ for 24 h in a N₂-balanced Whitley H35 HypOxystation.

RNA isolation, reverse transcription, and quantitative PCR

Following 24 h hypoxic treatment, RNA was isolated using a chloroform extraction. Equal concentrations of RNA were reverse transcribed using the high-capacity cDNA reverse transcription kit (Applied Biosystems). Quantitative PCR (qPCR) was performed in triplicate using the SYBR green Supermix (Bio-Rad) with an annealing temperature of 60°C. The primer sequences (5′→3′) used were as follows: DDX28, GCG GGA TGA ACT GCT TTG TG (forward) and TTT CCC TGG GGG CTT AAT GG (reverse); EEF2, TTC AAG TCA TTC TCC GAG A (forward) and AGA CAC GCT TCA CTG ATA (reverse); EGFR, GGA GAA CTG CCA GAA ACT GAC (forward) and GGG GTT CAC ATC CAT CTG (reverse); EPAS1, TCC TCT CCT CAG TTT GCT CT (forward) and GTC CCA TGA ACT TGC TGA TG (reverse); HSP90AB1, TGT CCC TCA TCA TCA ATA CC (forward) and TCT TTA CCA CTG TCC AAC TT (reverse); IGF1R, CCA TTC TCA TGC CTT GGT CT (forward) and TGC AAG TTC TGG TTG TCG AG (reverse); GAPDH, GTC AAG GCT GAG AAC GGG A (forward) and CAA ATG AGC CCC AGC CTT C (reverse); RPLP0, AAC ATC TCC CCC TTC TCC (forward) and CCA GGA AGC GAG AAT GC (reverse); RPL13A, CAT AGG AAG CTG GGA GCA AG (forward) and GCC CTC CAA TCA GTC TTC TG (reverse); CITED2, CCT AAT GGG CGA GCA CAT ACA (forward) and CGT TCG TGG CAT TCA TGT T (reverse); IGFBP3, GCG CCA GGA AAT GCT AGT G (forward) and AAC TTG GGA TCA GAC ACC CG (reverse); ITPR1, CGG AGC AGG GTA TTG GAA CA (forward) and GGT CCA CTG AGG GCT GAA AC (reverse); LOXL2, CCC CCT GGA GAC TAC CTG TT (forward) and GGA ACC ACC TAT GTG GCA GT (reverse); POU5F1, GAT GTG GTC CGA GTG TGG TTC (forward) and TTG ATC GCT TGC CCT TCT G (reverse). Transcript levels were normalized to RPLP0 and RPL13A.

DDX28 knockdown differential gene expression and pathway enrichment analysis

Differential expression analysis was performed on RNA-seq data from shRNA knockdown against DDX28 in HepG2 cells. Data were previously published in the ENCODE Consortium (Luo et al., 2020) and produced from the Brenton Graveley UConn ENCODE production laboratory (ENCSR222CSF). Analyses were performed using DESeq2 (Love et al., 2014) and visualized using subsequent R code. Differentially expressed genes required an absolute fold change greater or equal to two and false discovery rate (FDR) of <0.05. Statistically enriched gene ontology pathways related to biological processes were calculated using the PANTHER overrepresentation test (Thomas et al., 2022) and categories with an FDR of P < 0.05 are shown.

Polysome profiling

After 24 h hypoxic treatments, cells were isolated from four 15-cm² plates at 80 to 100% confluency. To stall active ribosomes on their bound mRNA molecules, cells were treated with 0.1 mg/ml cycloheximide for 10 min. Cells were lysed using an RNA lysis buffer (0.1 mg/ml cycloheximide, 15 mM Tris-HCl [pH 7.4], 15 mM MgCl₂, 0.3 M NaCl, 1% Triton X-100, and 100 U/mL RNase inhibitor) and equal RNA was loaded onto gradients. A total of 7 to 47% sucrose gradients were ultracentrifuged at 39,000 rpm for 90 min at 4°C with a SW-41-Ti Rotor. Next, the Brandel BR-188 density gradient fractionation system was used to fractionate the gradients into 1 ml fractions, while simultaneously being measured for absorbance at 254 nm. For RNA isolation, 18 proteinase K was added to each fraction, and RNA was isolated by phenol:chloroform:isoamyl alcohol (25:24:1) extraction and ethanol precipitation, and pooled into separate monosome and polysome fractions based on absorbance profiles. Equal volumes of total RNA were then reverse transcribed using the high-capacity cDNA reverse transcription kit (Applied Biosystems). Quantitative real-time PCR (qRT-PCR) was performed in triplicate using the SYBR green Supermix (Bio-Rad) with an annealing temperature of 60°C. The primer sequences used were as follows (5′ → 3′). Relative fold change in expression was calculated using the ΔΔC_t method with the data generated by CFX Manager Software.

Western blot analysis

U87MG, MDA-MB-231, and HEK293 wild-type cells were grown to 80 to 90% confluency and lysed in RIPA (10 mM Tris–HCl: pH 8, 1 mM EDTA, 1% NP-40, 0.1% sodium deoxycholate, 0.1% SDS, 150 mM NaCl, 2 µM NaO and 1X protease inhibitor cocktail [New England Biolabs]) following a 24-h normoxic or hypoxic incubation. Standard Western blot protocols were used. Primary antibodies were as follows: DDX28 (Abcam, ab70821), HIF-2α (Novus, NB100-122), FLAG (Sigma, F1804), β-Actin (GeneTex, GT5512), Lamin A/C (Cell Signaling Technology, 2032S), GAPDH (Cell Signaling Technology, 5174S), Tom20 (Santa Cruz Biotechnology, sc-17764), and OXPHOS human WB antibody cocktail (Invitrogen, 45-8199). Blots were quantified through densitometry using Bio-Rad ImageLab.

Generation of stable cell lines

U87MG and MDA-MB-231 cells stably expressing C-terminal 3 × FLAG-tagged DDX28 were generated by transfecting cells (polyethylenimine transfection) with the OmicsLink pEZ-M14 EX-A3144-M14 expression vector (Genecopoeia), encoding the human DDX28 coding sequence. Selection was initiated 48 h posttransfection using 1000 µg/ml G418, and single colonies were picked after ∼7 d.

Cell migration assay

Cells were given serum-reduced (0.5% FBS) media for 24 h then seeded at a density of 75,000 (MDA-MB-231) or 40,000 cells (U87MG) in the upper chamber of 12-well Transwell inserts with 8-µm pores (BD Biosciences). Full media were added to the bottom chamber. Cells were then exposed to 18 h of normoxia or hypoxia. For drug treatments, cells were treated with either 5 µM rotenone or 10 mM 2-DG for 30 min before being added to the upper chamber for an 18-h incubation with each drug still in the medium. Cells that did not migrate were removed from the upper side of the inserts with a cotton swab and migrated cells were fixed in methanol and stained with Hoechst. Membranes were excised, mounted on slides, and imaged on a Nikon Eclipse Ti Microscope. Migrated cells were quantified using Fiji for ImageJ. A threshold was first determined that accurately highlighted migrated cells. A watershed mask was then applied to segregate adjacent cells. Parameters of circularity and size were applied, and number of particles were counted using ImageJ. The number of cells that migrated were then compared with controls.

Spheroid formation and growth assay

Spheroid growth and invasion assays were adapted from a previously described method (Kirsh et al., 2023). Briefly, 10,000 cells were seeded in round-bottom U-shaped 96-well plates (Corning). The plates were then spun in a circular motion to promote aggregation of cells into a single spheroid per well. Spheroids were initially grown for 72 h, after which images of spheroids were captured on a Nikon Eclipse Ti Microscope at 0, 24, 48, 72, and 96 h. For drug treatments, spheroids were treated with 5 µM rotenone or 10 mM 2-DG at 0 h. Surface area of spheroids was measured using ImageJ Fiji software to assess spheroid growth. MDA-MB-231 spheroids do not compact as effectively as U87MG and do benefit from ECM media supplements. However, we omitted any such supplements since we did not want to initiate any invasion for this assay and these were not given to U87MG spheroids.

Spheroid invasion assay

A total of 10,000 cells were plated in flat-bottomed 96-well plates (Thermo Fisher Scientific) coated with 1.5% low-melting agarose (Thermo Fisher Scientific). Spheroids were grown on Spheroid Formation ECM (Cultrex). The plates were spun in a circular motion to promote aggregation of cells in the middle of each well into single spheroids. Spheroids were then grown for an initial 72 h, after which images of spheroids at 0, 24, 48, 72, and 96 h were captured on a Nikon Eclipse Ti Microscope to assess the impact of DDX28 OE on invasion into surrounding ECM. For drug treatments, spheroids were treated with either 5 µM rotenone or 10 mM 2-DG at 0 h. The invasive protrusions into the ECM were measured using ImageJ Fiji software.

Subcellular fractionation

After 24 h of normoxia or hypoxia, U87MG and MDA-MB-231 control (stable expression of non-targeting shRNA) or DDX28 OE cells were harvested and centrifuged at 500 × g for 10 min at 4°C. The pellets were resuspended in 500 µl cold harvest buffer I (50 mM HEPES, 150 mM NaCl, 1 M sucrose, 1 mM EDTA, 25 ug/mL digitonin, 2 µM sodium orthovanadate [NaO], and 1X protease inhibitor cocktail [New England Biolabs]) and incubated on a rotator for 10 min at 4°C. Lysates were centrifuged at 2000 × g for 10 min at 4°C to pellet nuclei and mitochondria. The cytosolic fraction was collected and the pellet was washed (50 mM HEPES, 50 mM KCl, 1 mM EDTA, 1 mM EGTA, 2 µM NaO, and 1X protease inhibitor cocktail). Next, the pellet was resuspended in 500 µl cold harvest buffer II (50 mM HEPES, 150 mM NaCl, 1 M sucrose, 1 mM EDTA, 1% Triton X-100, 2 µM NaO, and 1X protease inhibitor cocktail), rotated at 4°C for 30 min, centrifuged at 7000 × g for 10 min at 4°C, and the mitochondrial fraction collected. After washing the pellet, it was resuspended in 500 µl cold RIPA buffer and rotated for 30 min at 4°C. After centrifugation at 7800 × g for 10 min at 4°C, the nuclear fraction collected. Proteins were then concentrated by TCA precipitation, resuspended in 4X Laemmli sample buffer and SDS Lysis buffer and boiled at 95°C for 90 s before performing Western blot analysis.

Immunofluorescence and microscopy

Cells were seeded onto coverslips in complete medium and cultured for 24 h in normoxia or hypoxia. Cells were washed with 1X PBS, fixed with 4% paraformaldehyde in 1X PBS for 15 min, washed with 1X PBS three times for 5 min each, and then washed in 0.15 M glycine dissolved in 1X PBS for 10 min. After three more 5-min washes in 1X PBS, cell membranes were permeabilized in 0.1% Tween dissolved in 1X PBS for 10 min before blocking (5% BSA, 0.1% Tween, and 1X PBS) for 1 h. Cells were then incubated with Tom20 primary antibody made in blocking solution at 4°C overnight. The next day, cells were washed for 5 min in 1X PBS with 0.1% Tween three times. Then, cells were incubated with anti-mouse secondary antibody Alexa Fluor 555 for 1 h in blocking solution at room temperature. Cells were then washed 5 min in 1X PBS with 0.1% Tween three times. Coverslips were immersed in 1 µg/ml Hoechst in 1X PBS for 10 min and washed 5 min in 1X PBS with 0.1% Tween three times before mounting with Prolong Gold (Thermo Fisher Scientific). Images were analyzed using the Mitochondrial Analyzer plugin on Fiji ImageJ.

mtDNA Copy-number assessment

To assess mtDNA copy number, cells were grown to 80 to 90% confluency and incubated in either normoxia or hypoxia for 24 h. Next, genomic DNA was isolated using the E.Z.N.A. Tissue DNA Kit (Omega BioTek) and qPCR was performed as described above using the single-copy nuclear gene HBB as a reference gene for the mitochondrial mt-ND1 gene expression. The primer sequences (5′→3′) used were as follows: mtND1, GGC TAT ATA CAA CTA CGC AAA GGC (forward) and GGT AGA TGT GGC GGG TTT TAG G (reverse); and HBB, AGA AGT CTG CCG TTA CTG CC (forward) and CCC AGG AGC CTG AAG TTC TC (reverse). Relative copy number was calculated using the formula 2(2^ΔC_t).

Assessment of glycolytic versus mitochondrial ATP generation

Measurement of OCR and ECAR were performed using a Seahorse XFe24 extracellular flux analyzer (Seahorse Bioscience). U87MG and MDA-MB-231 cells overexpressing control empty-FLAG vector and FLAG-DDX28 were plated at 10,000 (normoxic U87MG), 7000 (hypoxic U87MG), 20,000 (normoxic MDA-MB-231), or 30,000 (hypoxic MDA-MB-231) cells per well in XF24 cell culture microplates (Agilent). The ATP Rate Assay protocol was followed as per the manufacturer's recommendations (Agilent). Generation of glycolytic ATP versus mitochondrial ATP was determined by measuring mitochondrial respiration through sequential injection of 1.5 µM (final concentration) oligomycin (an ATP synthase inhibitor) and 0.5 µM rotenone/antimycin A, which inhibits complex III activity. Data were analyzed with XF Wave software (Seahorse Bioscience).

Crystal violet viability assay

For crystal violet assays, at each timepoint, 20,000 U87MG control cells, 25,000 U87MG DDX28 OE cells, 30,000 MDA-MB-231 control cells, and 45,000 MDA-MB-231 DDX28 OE cells per well were seeded in quadruplicate in 24-well plates. The following day (0 h), cells were incubated in either normoxia or hypoxia for 24 h and treated with either 5 µM rotenone, or 10 mM 2-DG (Sigma Aldrich D8375). Following each 24 h increment, cells were washed once with 1X PBS and stained with 400 µl of 1% crystal violet solution prepared in 20% methanol, with gentle rocking for 20 min at room temperature. Cells were washed with water to remove excess stain. After plates were air dried overnight, 400 µl of 10% acetic acid was added to destain cells, and plates were incubated on a shaker for 20 min at room temperature. The absorbance was measured at 595 nm using a microplate reader.

Statistical analyses

Results are expressed as means ± SEM from at least three independent experiments. Experimental data were tested using the one-sample Student's t test when only one mean was of interest, an unpaired two-tailed Student's t test when two means were compared, a one-way analysis of variance (ANOVA) followed by Tukey's honestly significant difference (HSD) test when three or more means were compared, or a two-way ANOVA when three or more means were compared between two independent variables. The Shapiro–Wilk test was used to verify the normal distribution of all data that were compared using t tests. A P value of <0.05 was considered statistically significant using GraphPad Prism.

Abbreviations used:

2-DG

2-deoxy-D-glucose

ANOVA

analysis of variance

ECAR

extracellular acidification rates

EEF2

eukaryotic translation elongation factor 2

EGFR

epidermal growth factor receptor

EMT

epithelial-mesenchymal transition

ETC

Electron transport chain

FDR

false discovery rate

HIF

Hypoxia Inducible Factor

HSD

honestly significant difference

HSP90ab1

heat shock protein 90 alpha family class B member 1

IGF1R

insulin-like growth factor 1 receptor

mtDNA

mitochondrial DNA

NaO

sodium orthovanadate

OCR

oxygen consumption rate

OE

Overexpressed

OXPHOS

oxidative phosphorylation

qRT-PCR

Quantitative real-time PCR

RNA-seq

RNA sequencing.

DATA AND RESOURCE AVAILABILTY

All relevant data and resource can be found within the article and its supplementary information.