Controlling pericellular oxygen tension in cell culture reveals distinct breast cancer responses to low oxygen tensions

Abstract

Oxygen (O₂) tension plays a key role in tissue function and pathophysiology. O₂-controlled cell culture, in which the O₂ concentration in an incubator’s gas phase is controlled, is an indispensable tool to study the role of O₂ in vivo. For this technique, it is presumed that the incubator setpoint is equal to the O₂ tension that cells experience (i.e., pericellular O₂). We discovered that physioxic (5% O₂) and hypoxic (1% O₂) setpoints regularly induce anoxic (0.0% O₂) pericellular tensions in both adherent and suspension cell cultures. Electron transport chain inhibition ablates this effect, indicating that cellular O₂ consumption is the driving factor. RNA-seq revealed that primary human hepatocytes cultured in physioxia experience ischemia-reperfusion injury due to anoxic exposure followed by rapid reoxygenation. To better understand the relationship between incubator gas phase and pericellular O₂ tensions, we developed a reaction-diffusion model that predicts pericellular O₂ tension a priori. This model revealed that the effect of cellular O₂ consumption is greatest in smaller volume culture vessels (e.g., 96-well plate). By controlling pericellular O₂ tension in cell culture, we discovered that MCF7 cells have stronger glycolytic and glutamine metabolism responses in anoxia vs. hypoxia. MCF7 also expressed higher levels of HIF2A, CD73, NDUFA4L2, etc. and lower levels of HIF1A, CA9, VEGFA, etc. in response to hypoxia vs. anoxia. Proteomics revealed that 4T1 cells had an upregulated epithelial-to-mesenchymal transition (EMT) response and downregulated reactive oxygen species (ROS) management, glycolysis, and fatty acid metabolism pathways in hypoxia vs. anoxia. Collectively, these results reveal that breast cancer cells respond non-monotonically to low O₂, suggesting that anoxic cell culture is not suitable to model hypoxia. We demonstrate that controlling atmospheric O₂ tension in cell culture incubators is insufficient to control O₂ in cell culture and introduce the concept of pericellular O₂-controlled cell culture.

Introduction

A cornerstone of biological research, cell culture aims to grow cells in conditions that simulate their native environment as closely as possible. Cell culture models serve as a tool for testing biological hypotheses before validating in vivo. Healthy and diseased tissues are isolated from patients and studied in vitro. In fact, cell culture techniques are used throughout the process of drug development to make “go/no-go” decisions^{1, 2} and to manufacture adoptive cell therapies and regenerative medicines.^{3, 4} Recent advances in this practice include scaffolds that mimic the extracellular matrix^5–7, self-assembly of pluripotent stem cells to form brain organoids⁸, patient-derived organoids that capture tumor heterogeneity in patients and predict therapeutic responses^{9, 10}, etc. Yet, one key discrepancy between in vitro and in vivo remain: the “normoxic” oxygen (O₂) tension in cell culture (141 mmHg) is dramatically higher than the O₂ tension of human tissues (3–100 mmHg) (1% O₂ = 7.7 mmHg O₂ at sea level)).^{11, 12}

Supraphysiological O₂ concentrations (hyperoxia) lead to excessive reactive oxygen species (ROS) production, resulting in cellular damage and dysregulated signaling.^{13, 14} It is therefore not surprising that cells grown in physioxia experience less oxidative stress.^{15, 16} Furthermore, hyperoxia degrades proteins containing a specific iron-sulfur cluster, disrupting diphthamide synthesis, purine metabolism, nucleotide excision repair, and electron transport chain (ETC) function.¹⁷ Many O₂-dependent enzymes require iron and copper metal cofactors, which are susceptible to oxidation in hyperoxia.^{12, 18} Studies culturing cells in normoxia vs. physioxia have found aberrant T-cell activation¹⁹, fibroblast senescence²⁰ and mutation frequency²¹, chondrocyte differentiation²², etc. in normoxia. However, the full impact of culturing cells in normoxia will remain unknown until physioxia becomes common practice.

To address these concerns, tools to perform physiological cell culture have been developed and are commercially available. These products, including portable chambers, tri-gas incubators, and hypoxic workstations, consist of chambers that control O₂ in the atmosphere of cultured cells by adding compressed nitrogen. However, widespread adoption has been hampered by cost, laboratory space requirements, technical challenges, and rapid reoxygenation of cultures.^{23, 24} Reoxygenation, which occurs when cell cultures are removed from portable chambers or tri-gas incubators and exposed to normoxia, makes it difficult to recapitulate physiological O₂ tensions.^{18, 25} One major application for these products are hypoxic cell culture models, conducted at 0.5 – 1% O₂.^26–30 Hypoxia occurs in both physiological (e.g., placenta, renal medulla, intestinal mucosa, germinal centers, bone marrow) and pathophysiological (e.g., infection, inflammation, solid tumors, ischemia) contexts; and is therefore an active area of research.^{31, 32} Hypoxic cell culture was instrumental in the discovery of the prolyl hydroxylase (PHD) / hypoxia-inducible factor (HIF) axis, the mechanism by which cells sense and respond to low O₂.^{33, 34} O₂-controlled cell culture is also performed to mimic physioxia, typically at 5% O₂.^{15, 24}

In O₂-controlled cell culture, it is generally presumed that the atmospheric O₂ tension within incubators is equal to the pericellular O₂ tension, the O₂ concentration that cells experience. The pericellular O₂ tension is dependent on several rates: the O₂ diffusion within the cell culture media, O₂ transfer at the gas–media interface, and the cellular O₂ consumption. Gas–media O₂ transfer is the limiting rate.^{35, 36} Culture vessel geometry, medium volume, and surface area also influence diffusion times. These parameters vary greatly based on user preference and experimental design, yet are not reported. Experimental studies measuring pericellular O₂ tension indicate that confluent normoxic cultures can induce hypoxia.³⁷ However, the impact of O₂ consumption rates in lower O₂ tensions is unclear, since consumption decreases as O₂ becomes limiting.³⁸ We set out to assess how key cell culture parameters (i.e., cell type, cell density, medium volume, and culture vessel geometry) influence the relationship between atmospheric and pericellular O₂ tensions in O₂-controlled cell culture models. After discovering that pericellular O₂ tension is vastly different from atmospheric O₂ tension, we explored how controlling pericellular O₂ tension could be used as a novel tool to study breast cancer cell responses in low O₂.

Results

1% O₂ media conditioning can take over 5 days and is reoxygenated within minutes

For O₂-controlled cell culture experiments, media is typically conditioned to the desired O₂ tension and added to the cells at the start of the experiment. This procedure ensures that cells experience the desired O₂ tension immediately. We investigated how long it would take to condition 25–500 mL of media for hypoxic (1% O₂) experiments, since conditioning times are not reported.^{26, 28, 39} The required time was far longer than anticipated: over 1 day for 25 mL (upright T75 flask) and over 5 days for 500 mL (Figure 1A). The type of culture vessel or tube used did not change conditioning time for 25 or 50 mL of media. To understand media reoxygenation kinetics, 500 mL of 1% O₂ medium was removed from a tri-gas incubator and aliquoted into different culture vessels containing O₂ sensors. By the time the media reached the culture vessels, the O₂ concentration was >6% O₂ (Figure 1B). Depending on the surface area of the medium in different culture vessels, the medium reached 10% O₂ within seconds to 10 minutes. These results indicate that 1% O₂ media conditioning requires surprisingly long periods – more than 5 days for large volumes. Furthermore, 1% O₂ media is rapidly reoxygenated when removed from controlled O₂ atmospheric environments, indicating that portable chambers and tri-gas incubators are not suitable to condition media.

Figure 1: Cell culture parameters influence HIF stabilization kinetics in hypoxic (1% O₂) culture.

(A) O₂ kinetics (left) and time to 1% O₂ (right) of normoxic media (EMEM + 10% FBS + 1% P/S) placed in a 1% O₂ incubator. (B) O₂ kinetics (left) and time to 10% O₂ (right) of 1% O₂ media placed into different culture vessels under normoxia. (C) O₂ kinetics of MCF7 cultures with varying cell densities (left), medium volumes (middle), and culture vessel type (right) placed in a 1% O₂ incubator. (D) Time to 1% O₂ from (C). (E) Average fluorescence (Image-iT^™ Hypoxia) per cell (left) and representative confocal images at 4 hours (right) for MCF7 cultures (60mm dish) with 5 or 15 mL of media placed in a 1% O₂ chamber. Red = Image-iT^™ Hypoxia as an indicator of cellular hypoxia. (F) Percentage of GFP+ cells (left) and representative confocal images at 3 days (right) from MCF7 HIF reporter cells cultured at 7,000 or 29,000 cells/cm² for 4 days in a 1% O₂ incubator. Blue = nuclei stained with DAPI, Green = GFP+ (HIF+) cells. (G) Percentage of GFP+ cells (left), GFP mean fluorescence intensity (MFI) (middle) and representative contour plots with outliers (right) for MDA-MB-231 HIF reporter cells cultured in a 24-well plate or T25 flask for up to 5 days in a 1% O₂ incubator. Data were analyzed with ANOVA and Geisser-Greenhouse (E) or Bonferroni (F-G) corrections. N = 3 – 4 biological replicates per condition.

Cell density, medium volume, and culture vessel type influence HIF stabilization kinetics

We investigated how cell density, medium volume, and culture vessel type influence the time it takes normoxic cultures to reach 1% O₂. Non-invasive optical sensor spots were used to measure pericellular O₂ tension.^{40, 41} For MCF7 breast cancer cultures, all three parameters influenced the time to 1% O₂, ranging from 1–14 hours (Figure 1C–D). Cell density had the largest effect, suggesting that cellular O₂ consumption plays a key role in the induction of hypoxia in vitro.

Next, we explored whether medium volume influenced cellular hypoxia kinetics. MCF7 cultures in 60mm dishes containing either 5 mL or 15 mL of media were placed inside a 1% O₂ incubator. Cellular hypoxia was evaluated using a hypoxia-responsive fluorescent dye (Image-It^™ Hypoxia) for 12h. As expected, the cells in the 5 mL condition reached a maximum cellular fluorescence sooner than the 15 mL cultures: 4 hours vs. 10 hours (Figure 1E, Figure S1).

To understand if cell density or culture vessel type influenced HIF stabilization in 1% O₂ culture, MCF7 and MDA-MB-231 HIF reporter cell lines²⁸ were cultured at two cell densities for 4 days and the percentage of GFP-positive cells was determined using fluorescent microscopy. After 3 days, 2.2% of cells plated at the lower cell density were GFP-positive whereas 22.9% of cells plated at a higher density were GFP-positive (Figure 1F, Figure S2). We then tested whether different culture vessel types would induce a similar effect for MDA-MB-231 reporter cells. Consistent with our findings in Figure 1C, T25 flask cultures had a higher percentage of GFP-positive cells after 4 and 5 days compared to 24-well cultures (Figure 1G). Collectively, we have demonstrated that cell density, medium volume, and culture vessel type, parameters that vary between experiments and are not reported, greatly influence cellular hypoxia and HIF stabilization kinetics. These results suggest that consistent HIF stabilization kinetics in 1% O₂ culture can only be obtained by using a workstation and conditioned media.

Cellular O₂ consumption can induce anoxia in both 5% and 1% O₂ culture

After discovering that cellular O₂ consumption drives the induction of hypoxia in 1% O₂ incubators, we hypothesized that it would also influence the pericellular O₂ tension. To test this, sub-confluent (21,000 cells/cm²) MCF7, MDA-MB-231, and primary human mammary epithelial cells were cultured in physioxic (5% O₂) and hypoxic (1% O₂) conditions for 72 hours and the pericellular O₂ concentration was measured. Cell-free media O₂ tensions matched the incubator setpoints, indicating that the O₂ sensor spots were accurately recording and the incubator O₂ sensors were calibrated (Figure 2A–B). In physioxia, the pericellular O₂ was strikingly low – below 1.5% O₂ for all three lines, and MCF7 cultures were anoxic (anoxia defined as <0.5% O₂⁴²). In hypoxia, the pericellular O₂ of all cultures became anoxic (0.0% O₂) within 5 hours. Next, we investigated whether this effect also occurred in suspension culture. Activated human dendritic cells (DCs) were cultured in physioxia or hypoxia for 72 hours. Both cultures were immediately and sustainably anoxic (0.0% O₂) (Figure 2C).

Figure 2: Cellular O₂ consumption regularly induces anoxia (0.0% O₂) in both physioxic (5% O₂) and hypoxic (1% O₂) culture.

(A–B) O₂ profiles (left) and average O₂ tension values (right) of media, human mammary epithelial, MCF7, or MDA-MB-231 cultures seeded onto a 24-well plate in a 5% O₂ (A) or 1% O₂ (B) incubator for 72h. (C) O₂ profiles (left) and average O₂ tension values (right) of human dendritic cell (DC) 96-well cultures in a 5% or 1% O₂ incubator for 72h. (D) O₂ kinetics of media (green) or MDA-MBA-231 cultures (blue) of the top (dark) or bottom (light) of the well in a 1% O₂ incubator. (E) O₂ profiles of MDA-MB-231 cultures measured at the bottom of the well (spot) or 1mm above the bottom of the well (needle) in a 1% O₂ incubator. (F) Effect of sodium azide (NaN₃) on MCF7 pericellular O₂ tension and cytotoxicity. O₂ profiles (left) and average O₂ tension values (middle) of MCF7 cultures spiked with either 5 mM NaN₃ or PBS in a 1% O₂ incubator. MCF7 cell viability after ± incubation with 5 mM NaN₃ for 6 hours in a 1% O₂ incubator (right). (G) Average O₂ tension values of media and MCF7 cultures with different cell densities and culture vessel types in a 5% O₂ (left) or 1% O₂ incubator (right) for 72h. (H) Live, attached (left) and detached (right) cell counts for MCF7 cultures in a 18.6%, 5% or 1% O₂ incubator for 6 days. Data were analyzed with two-tailed t test (F) or ANOVA and Dunnett’s (A-B) or Tukey’s (C, F, G, H) corrections. N = 3 – 4 biological replicates per condition. Colored * indicates comparison to the control.

If cellular O₂ consumption did indeed affect pericellular O₂ tension, there would be an axial O₂ gradient in these cultures. To test this hypothesis, O₂ tension at the media-gas interface of MDA-MB-231 cultured at 1% O₂ was measured using needle O₂ microsensors (top) and pericellular O₂ tensions were measured using sensor spots (bottom). Figure 2D illustrates that in these cultures, there is a gradient of 0.4 – 0.0% O₂ from the top to the bottom of the well after 36 hours. In wells without cells, both the top and bottom of the well reached 1% O₂ as expected. To validate O₂ sensor spot readings, needle microsensors were placed 1mm above sensor spots in 1% O₂ MDA-MB-231 cultures using a micromanipulator. Sensor spots recorded 0.0% O₂ and needles recorded 0.1% O₂, indicating that the two probes were in good agreement (Figure 2E).

To further validate the role of cellular O₂ consumption in pericellular O₂ tension, we tested whether inhibiting oxidative phosphorylation would ablate axial gradients. Upon addition of sodium azide (NaN₃, complex IV inhibitor), the pericellular O₂ tension of MCF7 cultures rapidly rose from 3% O₂ to the incubator setpoint of 5% O₂, whereas PBS (vehicle) spiked cultures returned to 3% O₂ with continued incubation (Figure 2F, left panel). To confirm that this effect was not due to NaN₃ cytotoxicity, we confirmed that there was no significant change in MCF7 cell viability when incubated with NaN₃ at 5% O₂ for 6h (Figure 2F, right panel).

We next explored how cell density and culture vessel type influenced the gradient between atmospheric and pericellular O₂ tensions. In physioxic MCF7 cultures, cell densities of 7,000, 29,000, and 143,000 cells/cm² in a 12-well plate induced pericellular O₂ tensions of 4.2%, 0.5%, and 0.1% O₂, respectively. Different culture vessels (24-well plate, 12-well plate, and T25 flask) also influenced MCF7 tensions, ranging from 0.4 – 2.5% O₂ (Figure 2G, left panel). In hypoxia, neither cell density nor culture vessel type change pericellular tensions in MCF7 cultures: all were anoxic (<0.2% O₂) (Figure 2G, right panel).

After determining that all hypoxic MCF7 cultures tested were anoxic, we sought to understand how pericellular anoxia affected cell viability. MCF7 cells were cultured in normoxia, physioxia, or hypoxia for 6 days and cell proliferation and viability were evaluated. Cells cultured in normoxia and physioxia proliferated throughout the 6 day period (Figure 2H, left panel). However, cells cultured in hypoxia did not proliferate, and the majority were dead and detached after 4 days (Figure 2H, right panel).

Collectively, these experiments show that cellular O₂ consumption drives pericellular O₂ far below the incubator setpoint, inducing anoxia in both physioxic and hypoxic MCF7, MDA-MBA-231, and human DC cultures. Furthermore, in physioxic MCF7 culture, pericellular O₂ tension is highly dependent on cell culture parameters, ranging from 0.1 – 4.2% O₂.

Setting the incubator to physioxia mimics ischemia-reperfusion injury in hepatocyte culture

We next explored how the difference between the incubator setpoint and pericellular O₂ tension can impact the physiological relevance of cell culture models. Because of their high O₂ consumption rate^{24, 43} and widespread use as an in vitro drug metabolism model⁴⁴, primary human hepatocytes were used for these studies. Hepatocytes were seeded and cultured in either normoxic (18.6% O₂) or physioxic (6% O₂²⁴) conditions for 36 hours (Figure 3A).

Figure 3: Setting the incubator to physiological O₂ conditions mimics ischemia-reperfusion injury in human hepatocyte culture.

(A) Schematic of the primary human hepatocyte culture. RNA-seq was performed on uncultured hepatocytes and cells cultured in normoxic (18.6% O₂) or physioxic (6% O₂) incubator after 36h. (B–C) Average O₂ tension values (B) and detached cells per well (C) during each step of the culturing process. (D) Principal component analysis (PCA) of the RNA-seq results for normoxic and physioxic cultured hepatocytes. (E) Volcano plot indicating upregulated (blue) and downregulated (red) genes for physioxic vs. normoxic cultured hepatocytes (p_adj < 0.05 and |log₂FC| ≥ 1). (F) Hypoxia gene set enrichment analysis (GSEA) from the Hallmark database for normoxic vs. uncultured (left) and physioxic vs. uncultured (right). (G–I) Enriched pathways from the Gene Ontology (GO) database for physioxic vs. normoxic samples associated with oxidative stress (G), sterile inflammatory response (H) and mitochondrial and ribosomal biogenesis (I). Data were analyzed with ANOVA and Tukey’s correction (B-C). N = 3 biological samples per condition for RNA seq analysis.

During the first seeding step, which was conducted in normoxia, hepatocytes were anoxic (0.5% O₂) (Figure 3B). During the first 24 hours of culture, hepatocytes cultured in normoxia were physioxic (5.6% O₂) and hepatocytes cultured in physioxia were anoxic (0.4% O₂). Unexpectedly, when media was exchanged after 24 hours, both types of cultures underwent reoxygenation for the duration of the experiment. Figure 3C illustrates that the pericellular anoxia experienced by hepatocytes cultured in physioxia increased the number of detached cells after 24 hours, the majority of which were dead.

To investigate how the pericellular O₂ tension influenced hepatocyte physiology, we assessed gene expression by RNA-seq of uncultured hepatocytes, and hepatocytes cultured in normoxia or physioxia after 36h. Principal component analysis (PCA) of the transcriptome shows clustering of replicates by O₂ tension, as expected (Figure 3D). Differential gene expression analysis (physioxic vs. normoxic) found 269 upregulated and 332 downregulated genes (Figure 3E). Gene set enrichment analysis (GSEA) revealed an upregulation in hypoxia-associated genes in normoxic and physioxic cultured cells compared to uncultured hepatocytes (Figure 3F), indicating that the 1-hour anoxic seeding step induced a hypoxic response. In addition, GSEA suggests that hepatocytes cultured in physioxia mounted an oxidative stress response (Figure 3G). Corroborating the higher cell death and detachment, physioxic cultured hepatocytes also had an upregulation in IL-1β production, TNF production, TLR signaling, and PRR signaling pathways, indicating a sterile inflammatory response (Figure 3H). Lastly, hepatocytes in physioxia had upregulated mitochondrial and ribosomal biogenesis pathways (Figure 3I). Taken together, these results indicate that setting the incubator to physiological O₂ conditions mimics ischemia-reperfusion injury in hepatocytes due to their cellular O₂ consumption.

Developing a reaction-diffusion model to predict pericellular O₂ tension in cell cultures

Measuring pericellular O₂ tension for every O₂-controlled cell culture experiment would be cumbersome and expensive. We hypothesized that a computational model could predict pericellular tension a priori, given cell density, O₂ consumption rate, culture vessel type and medium volume. Such a tool would reduce the need for experimental measurements.

We first examined whether the unsteady state diffusion equation^{35, 36} could describe O₂ transfer kinetics between cell culture medium and incubator gas phases. Coefficient of determination (R²) values suggested that experimental and numerical values that describe the O₂ transfer between normoxic media and 1% O₂ gas phase in different culture vessels were in good agreement (Figure 4A). The diffusion model also predicted equilibration of normoxic media in a 5% O₂ incubator (Figure S3A). In addition, the diffusion model’s analytical solution was also comparable to numerical solutions for different culture vessel types (Figure S3B).

Figure 4: Developing a reaction-diffusion model to predict pericellular O₂ tension in cell cultures.

(A) Numerical (num) (dashed) and experimental (exp) (solid) O₂ kinetics of normoxic media (EMEM + 10% FBS + 1% P/S) in different culture vessels placed in a 1% O₂ incubator. (B) Diffusion model predictions for different volumes of normoxic media in a 6-well plate (left) or a 24-well plate (right) placed in a 1% O₂ incubator. (C) Numerical and experimental O₂ kinetics of MDA-MB-231 cultures seeded in a 24-well plate placed in a 1%, 4%, 6%, or 8% O₂ incubator for 48h. (D) Reaction-diffusion model predictions of the gradient within MDA-MB-231 cultures seeded in a 24-well plate in a 5% O₂ (left) or 1% O₂ (right) incubator. (E) Reaction-diffusion model predictions of MDA-MB-231 cells cultured at different cell densities and culture vessel types in a 5% O₂ incubator. N = 4 biological replicates per condition for experimental data.

After validating the diffusion model, we applied it to investigate the dependency of O₂ transfer kinetics on medium volume in a 24-well plate and 6-well plate. Figure 4B shows that although commonly used medium volumes for 24-well and 6-well plates modestly impact the time to 1% O₂, there is a substantial difference in kinetics between 24-well and 6-well plate wells. 6-well plate volumes of media require 0.5–12 hours to reach 1% O₂, whereas 24-well plate requires 2.5–40 hours. Importantly, the rate of equilibration for the same medium volumes (1 mL and 2 mL) is substantially lower in 24-well plate than 6-well plate wells. These findings indicate that the gas phase–pericellular O₂ differential increases as culture vessel surface area decreases (Figure 2G) because of a decrease in O₂ transfer rates.

Next, we developed a reaction-diffusion model to describe pericellular O₂ tension in cell cultures within O₂-controlled environments. Michaelis–Menten kinetics were used to model cellular O₂ consumption.³⁸ This model predicts pericellular O₂ tension values for MDA-MB-231 cultured at 1%, 4%, 6%, and 8% O₂ with reasonable Michael–Menten parameters (V_max = 450 amol cell⁻¹ sec⁻¹ and K_m = 1 μM)³⁸ (Figure 4C). For further validation, we tested whether the model predicted axial O₂ gradients like those experimentally determined in Figure 2D. The model predicts gradients of 2.0 – 3.0% O₂ and 0.0 – 0.3% O₂ for 30,000 cells/cm² MDA-MB-231 cultures in physioxia and hypoxia, respectively (Figure 4D), in good agreement with experimental data.

Using the reaction-diffusion model, we next examined the influence of cell density on pericellular O₂ tension in different culture vessel types in physioxia. The model predicts that the highest cell density will have a modest influence on pericellular tension in 6-well plate cultures (3.2% O₂), but will induce anoxia in 24-well (0.4% O₂) and 96-well plate cultures (0.0% O₂) (Figure 4E). Lastly, we investigated the influence of cellular O₂ consumption (V_max) in hypoxia (Figure S3C). Both cell density and O₂ consumption predictions suggest that the smaller the culture vessel surface area, the greater the impact of cellular O₂ consumption on pericellular O₂ tension. This effect is due to the decreasing media surface area to height as culture vessel size decreases (Figure S4D). Taken together, we establish that a reaction-diffusion model can predict pericellular O₂ tension in O₂-controlled cell culture. Furthermore, using the model, we discovered that the effect of cellular consumption on pericellular O₂ tension is highly dependent on the culture vessel type, increasing as culture vessel size decreases.

Pericellular anoxia induces stronger metabolic reprogramming compared to pericellular hypoxia in MCF7 cells

The studies presented thus far demonstrate that standard hypoxic cell culture (1% O₂) routinely induces anoxia due to cellular O₂ consumption. Because anoxia is not physiologically relevant in vivo, we asked whether anoxia is suitable to model hypoxia. To explore this concept, we controlled pericellular O₂ tension to investigate cancer cell responses to pericellular hypoxia (1–2% O₂) vs. pericellular anoxia (0% O₂).

First, we examined MCF7 metabolic reprogramming in response to different pericellular O₂ tensions. Expected metabolic changes in response to hypoxia include an increase in (i) glucose consumption due to increased uptake and glycolytic flux, (ii) extracellular lactate from decreased TCA cycle flux and increased lactate transport, (iii) glutamine uptake to replenish TCA cycle intermediates for lipid metabolism, and (iv) extracellular glutamate secretion, which promotes cancer cell proliferation.^{39, 42, 45–48} MCF7 cells were cultured for 72 hours in 18.6%, 4%, and 1% O₂ incubators, resulting in supraphysiologic (10.8% O₂), hypoxic (1.2% O₂), and anoxic (0.0% O₂) pericellular tensions, respectively (Figure 5A). Consumption (glucose and glutamine) and production (lactate and glutamate) rates trended higher with decreasing pericellular O₂ tension over the 72 hour time course (Figure 5B – 5E). For example, lactate and glutamate production was 2.2- and 1.6-fold higher for cells in anoxia compared to hypoxia, respectively. Only the anoxic MCF7 cultures exhibited increased rates after 24 hours, whereas hypoxic and normoxic cultures maintained constant rates. These findings suggest that for MCF7 cells, anoxia induces a stronger metabolic reprogramming response than hypoxia does.

Figure 5: Pericellular anoxia induces stronger metabolic reprogramming and distinct transcriptional HIFA and HRE controlled gene responses compared to pericellular hypoxia in MCF7 cells.

(A) Average pericellular O₂ tensions in MCF7 cultures placed in 18.6%, 4.5%, and 1% O₂ incubators for 72 hours. (B–E) Extracellular concentrations of glucose (B), glutamine (C), lactate (D), and glutamate (E) in different O₂ tensions for 72 hours. Normalized metabolite concentrations over time (left) and metabolite consumption or production rates for 72 hours (right). (F–I) Gene expression levels of genes associated with the hypoxic response (F), metabolic reprogramming (G), autophagy (H), and immunosuppression (I) in different O₂ tensions for 72 hours. Data were analyzed using ANOVA with Tukey’s correction. N = 4 biological samples per condition. Color code for asterix (*): Red colored * indicate comparison to control (normoxia) and blue colored * indicate comparison between hypoxia and anoxia.

Pericellular hypoxia vs. anoxia induces distinct transcriptional HIFα and HRE gene responses in MCF7 cells

We examined MCF7 transcriptional responses to low O₂ tensions at 24, 48, and 72 hours, including genes associated with (i) the hypoxic response (e.g., hypoxia inducible factor 1 subunit alpha (HIF1A), hypoxia inducible factor 2 subunit alpha (HIF2A), vascular endothelial growth factor A (VEGFA), protein kinase AMP-activated catalytic subunit alpha 2 (PRKAA2), and activating transcription factor 4 (ATF4)); (ii) metabolic reprogramming (e.g., carbonic anhydrase 9 (CA9), lactate dehydrogenase A (LDHA), pyruvate dehydrogenase kinase 1 (PDK1), solute carrier family 2 member 1 (SLC2A1, GLUT1), solute carrier family 1 member 5 (SLC1A5), solute carrier family 7 member 11 (SLC7A11, xCT), and NADH dehydrogenase 1 alpha subcomplex, 4-like 2 (NDUFA4L2)); (iii) mitophagy (e.g., Bcl-2 interacting protein 3 (BNIP3) and Bcl-2 interacting protein 3 like (BNIP3L)); and (iv) the immunosuppressive tumor microenvironment (TME) (e.g., cluster of differentiation 274 (CD274, PD-L1) and 5’-nucleotidase ecto (NT5E, CD73^{49, 50}) (Figure 5F–I). Nearly all of these genes are direct HIF targets (i.e., hypoxia-responsive element (HRE)-controlled genes), excluding PRKAA2, ATF4, and SLC1A5.^{32, 42}

In anoxia, VEGFA, CA9, PDK1, BNIP3, and BNIP3L show elevated expression compared to normoxia throughout the time course. However, in hypoxia, these genes steadily increased expression and peaked after 48 hours, followed by a drop to normoxic expression levels after 72 hours. Interestingly, HIFA expression was different in hypoxia and anoxia: HIF1A had higher expression in anoxia (6- and 3-fold after 24 and 48 hours, respectively) and HIF2A had higher expression in hypoxia (5-fold after 72 hours) (Figure 5F). In addition, MCF7 had higher expression of PRKAA2 (AMPK α2 subunit) in anoxia (2- and 4-fold after 24 and 72 hours, respectively), suggesting lower ATP availability in anoxia.⁵¹ Interestingly, hypoxia induced 2-fold higher expression of ATF4 after 48 hours than did anoxia, indicating a stronger integrated stress response in hypoxia.⁵² Glycolytic genes had higher expression in anoxia (LDHA 7-fold after 72 hours and PDK1 9-fold after 72 hours). SLC2A1 (GLUT1) expression levels also trended higher in anoxia throughout the time course. SLC1A5 expression levels were 2-fold higher in anoxia after 48 hours. Overall, these results support the metabolic profiles in Figure 5B–E.

BNIP3 and BNIP3L expression levels were higher in anoxia after 72 hours, suggesting an upregulation in mitophagy in anoxia.⁴² Lastly, in the context of immunosuppression, hypoxia induced 6-fold higher expression of NT5E (CD73, extracellular AMP to adenosine conversion^{49, 50}) after 24 hours. Anoxia induces 2-fold higher expression of CD274 (PD-L1) after 48 hours. Overall, these results indicate that hypoxia and anoxia induce distinct expression profiles in both HIFA and HRE responsive genes in MCF7.

Proteomic characterization of the temporal differences between pericellular hypoxic and anoxic responses in 4T1 cells

After looking at transcriptional responses to hypoxia and anoxia, we aimed to better understand changes in protein expression in response to low O₂ tensions. To this end, we applied plexDIA⁵³ to understand how the proteome changes in response to hypoxia and anoxia in a murine triple negative breast cancer (TNBC) cell line (4T1). PCA of the proteome shows clustering by O₂ tension and by day (Figure S4A). Furthermore, hypoxic and anoxic samples shift away from normoxic samples along PC1 over time, indicating continuing changes during low O₂ responses.

For the hypoxic response, differential protein abundance analysis indicates no significantly upregulated or downregulated proteins after 1 day of culture, with the maximum response occurring after 3 days. On the other hand, the anoxic response had 50 upregulated and 5 downregulated proteins after day 1, and the response peaked after only 2 days. The number of changing proteins was higher in anoxia than hypoxia for all three days (Figure S4B).

Protein set enrichment analysis (PSEA) was performed to compare each low O₂ response between days. In agreement with the differential protein abundance analysis, PSEA suggests that the anoxic response is faster and peaks by day 2: most of the changes occur for 2 days vs. 1 day (Figure S4C). For both O₂ conditions, hypoxia-associated pathways are upregulated throughout the time course, including hypoxia, glycolysis, cholesterol homeostasis, fatty acid metabolism, and epithelial-to-mesenchymal transition (EMT). Interestingly, Myc targets were downregulated in both tensions for all days, which may contribute to the reduction in proliferation at low O₂ tensions.⁵⁴ Overall, this temporal characterization of the proteome suggests that the response to anoxia is stronger and faster than the response to hypoxia in 4T1 cells.

Characterizing proteomic differences in metabolic reprogramming between pericellular hypoxic and anoxic responses in 4T1 cells

To further characterize low O₂ responses in 4T1 cells, we explored protein synthesis, hypoxic responses, and metabolic reprogramming at the pathway and protein level. As expected, RNA processing and protein translation pathways were downregulated in hypoxia and anoxia compared to nomoxia (Figure 6A). Most of these pathways were upregulated in hypoxia compared to anoxia, suggesting downregulation as a function of O₂ tension.Surprisingly, translation elongation was upregulated in anoxia compared to hypoxia after 1 day of culture, suggesting distinct regulation in the acute anoxic response.

Figure 6: Proteomic characterization of pericellular hypoxic and anoxic metabolic reprogramming in 4T1 cells.

(A) Protein set enrichment analysis (PSEA) using the Hallmark database of mRNA processing and protein translation pathways for Hypoxia (H) vs. Normoxia (N), Anoxia (A) vs. Normoxia, and Hypoxia vs. Anoxia. (B) PSEA using the Hallmark database of hypoxic response pathways (left). Heat maps of selected proteins in Hallmark hypoxia, epithelial to mesenchymal transition (EMT), and reactive O₂ species (ROS) pathways (right) (Hypoxia vs. Anoxia). (C) Heat map of glycolytic and hypoxic response proteins. (D) PSEA using Reactome database for fatty acid (FA) metabolism pathways for Hypoxia vs. Anoxia (top). Heat maps of selected proteins in fatty acid β-oxidation and synthesis Reactome pathways (bottom). (E) Heat maps of selected proteins from Tricarboxylic acid (TCA) cycle and electron transport chain (ETC) processes for Hypoxia vs. Normoxia and Anoxia vs. Normoxia. PSEA using the Reactome database of TCA and ETC pathways for Hypoxia vs. Normoxia, Anoxia vs. Normoxia (left), and Hypoxia vs. Anoxia (right). NES = normalized enrichment score. N = 3 biological replicates and N = 2–3 technical replicates per condition.

Next, we looked at pathways associated with the hypoxic response, and discovered that hypoxia and reactive oxygen species (ROS) pathways were upregulated in anoxia and EMT was upregulated in hypoxia (Figure 6B). Proteins associated with hypoxia-induced stress responses showed higher abundance in anoxia for all three days. In addition, proteins associated with tumor progression, invasion, and metastasis, were upregulated in hypoxia for the duration of the experiment.

Figure 6C illustrates that 9 out of 10 enzymes involved in glycolysis are more abundant in anoxia compared to hypoxia. In addition, proteins involved in hypoxia-mediated metabolic reprogramming show increased abundance in anoxia. We also examined fatty acid β-oxidation and synthesis, which are downregulated and upregulated in hypoxia respectively.^{32, 42} Interestingly, PSEA suggests that both of these processes are downregulated in anoxia compared to hypoxia for the first 2 days of culture (Figure 6D). Most enzymes involved in mitochondrial acyl-CoA to acetyl-CoA conversion (i.e., fatty acid β-oxidation) were upregulated in anoxia.

Finally, we examined oxidative phosphorylation, which is expected to decrease in low O₂ tensions.^{32, 42} Unexpectedly, for hypoxia vs. anoxia, the tricarboxylic acid (TCA) cycle and electron transport chain (ETC) processes were downregulated after 1 day of culture and upregulated after 3 days of culture (Figure 6E). Clustering of TCA and ETC enzymes suggests increasing oxidative phosphorylation activity in hypoxia, yet decreasing activity in anoxia (compared to normoxia). Taken together, these results suggest that 4T1 cells have distinct hypoxic and anoxic metabolic responses (Figure 6F).

Discussion

Despite the widespread use of O₂-controlling chambers in hypoxia-related research, studies quantifying pericellular O₂ concentrations and their impact on cellular response are surpisingly lacking. In the current study, we discovered vast differences between incubator setpoints and pericellular O₂ tensions in every cell type tested. Our results highlight a major challenge with portable chambers and tri-gas incubators: cultures cannot be conditioned to start at the desired O₂ tension due to rapid reoxygenation of media upon exposure to normoxia.

Without media conditioning, MCF7 cultures can take 1–14 hours to reach 1% O₂ depending on experimental set-up. Not only does this time difference introduce significant variability between experiments, it suggests that shorter hypoxic experiments may not even reach hypoxia. We also show that changing cell culture parameters induce a sustained difference of HIF stabilization kinetics for at least 5 days for two different cell lines. Unless a hypoxic workstation with preconditioned media is used, pericellular O₂ tension must be determined to report accurate O₂-controlled incubation times.

Challenging the belief that incubators accurately control O₂ for cell cultures, we discovered that pericellular anoxic tensions are common in both physioxic (5% O₂) and hypoxic (1% O₂) conditions due to cellular O₂ consumption. This effect occurred in both primary human adherent and suspension cultures, with commonly used cell densities, medium volumes, and culture vessel types. Furthermore, our results suggest that physioxic cultures are routinely hypoxic. O₂ tension in physioxia can vary greatly (0.1–4.5% O₂) depending on experimental set-up, engendering reproducibility concerns. These findings are a major concern for physioxic culture of stem cell expansion and differentiation^55–57, and controlling pericellular O₂ tension may improve our understanding of these processes.

Our pericellular O₂ tension results for primary human hepatocyte cultures, a prominent model in drug metabolism, underpin how the incubator setpoint is a poor indicator of the O₂ cells experience. Hepatocytes cultured in physioxia experienced anoxia for 24 hours, followed by a rapid reoxygenation upon media exchange. RNA-seq results indicate that O₂ fluctuations in these conditions induce an upregulation in cellular responses to mitochondrial and NADPH oxidase ROS production (e.g., superoxide, hydrogen peroxide).^{58, 59} Oxidative stress increased cell death via upregulation of TNF (apoptosis) and IL-1β (pyroptosis) production.^{60, 61} Damage-associated molecular patterns (DAMPs) released by dying cells activate toll-like receptor (TLR) and pattern recognition receptor (PRR) pathways, inducing key signatures of a sterile inflammatory response: complement activation, inflammasome complex assembly, MHC class II upregulation, and IL-6 production^{62, 63}. During reoxygenation, hepatocytes increased mitochondria and ribosome biogenesis to meet ATP and protein translation demands as the cells recovered from hypoxic exposure.⁴² Ultimately, these results indicate that physioxic culture of hepatocytes drives a cellular response mimicking liver ischemia-reperfusion injury, a major risk factor in graft dysfunction in liver transplanation.⁶¹

We developed a reaction-diffusion model, which accurately predicted pericellular O₂ tensions of MDA-MB-231 cultures at various incubator setpoints. This novel tool can design O₂-controlled cell culture experiments, modulating parameters like the cell density, culture vessel type, or medium volume to achieve desired pericellular O₂ concentrations. Our finding that the effect of cellular O₂ consumption increases as culture vessel size decreases suggests that smaller vessels (e.g., 96-well plates) should be avoided for O₂-controlled cell culture. This observation carries significant implications for immune cell culture, which are typically done at high cell densities in small culture vessels. Future iterations of the model should consider incorporating cell growth rates as a function of O₂ tension. In addition, experiments to determine O₂ consumption rates (V_max) for different cell types are needed. Such studies may uncover O₂ consumption trends applicable to most cell types.

Using the reaction-diffusion model, we performed the first relation of pericellular O₂ tension to biological response. We investigated the metabolic, transcriptomic, and translational responses to hypoxia (1–2% O₂) and anoxia (0.0% O₂) in two different breast cancer cell lines. Quantification of 14 genes associated with hypoxia (11 are direct HIF targets) and HIF1A/HIF2A suggest distinct transcriptional responses in hypoxia and anoxia. In anoxia, we found higher transcription in HIF1A, hypoxic markers (VEGFA, CA9), metabolism (LDHA, PDK1, SLC1A5), mitophagy (BNIP3, BNIP3L), and CD274 (PD-L1) compared to hypoxia. Conversely, HIF2A, ATF4, NDUFA4L2, and NT5E (CD73) was upregulated in hypoxia. The increase in HIF1A transcription in anoxia may occur through ROS-induced PI3 kinase (PI3K) and protein kinase C (PKC) pathways^{64, 65}, since our proteomic analysis suggests that ROS production is higher in anoxia.

A previous study reported that maximum HIF1 DNA-binding activity occurs at pericellular 0.5% O₂ and sharply decreases as tension approaches 0% O₂.⁶⁶ This suggests that in anoxic culture, HIF is maximally stabilized as cultures approach anoxia (rather than in anoxia per se) inducing a strong HRE transcriptional response. Ultimately, 0.5% pericellular O₂ may be the ideal setpoint for hypoxic cell culture, as long as tensions do not drop to anoxic levels. Future studies will examine the post-translational modification of both HIF1/HIF2 and downstream responses in different low O₂ tensions (0 – 3% O₂).

To further understand low O₂ tension responses, we performed an in-depth characterization of the 4T1 proteome in response to low O₂ tensions. This analysis suggests that the global translational response to anoxia is stronger and faster than the hypoxic response. Yet, the responses are distinct: we found an upregulation of EMT proteins in hypoxia and an increased ROS response in anoxia for 72h of culture. EMT is a critical step in hypoxia-driven metastasis⁶⁷ and metastastic breast cancer represents the most advanced stage of the disease.⁶⁸ Our findings suggest that mitochondrial dysfunction and ROS production are higher in anoxia than in hypoxia.⁶⁹

Ultimately, our exploration of breast cancer responses to low O₂ tensions suggest that anoxia is not suitable to model hypoxia. This is fortified by the fact that the median O₂ tension in breast tumors is 10 mmHg (1.3% O₂).⁷⁰ Equally important, our findings uncover that breast cancer cells respond non-monotonically to low O₂, since many aspects of the low O₂ response are upregulated in hypoxia compared to anoxia. Future work will further explore these responses and relate them to hypoxic environments in vivo.

Conclusion

O₂ is a critical factor for mammalian bioenergetic homeostasis and serves as a substrate for over 200 enzymatic reactions.¹² O₂-controlled cell culture attempts to mimic O₂ tensions that cells are exposed to in vivo and is therefore a critical tool for biological research. Herein, we report the discovery that the metric used to determine O₂ concentration for in vitro cultures, the incubator setpoint, is a poor indicator of the O₂ tension cells actually experience (i.e., pericellular O₂ tension) due to cellular O₂ consumption. Standard physioxic (5% O₂) and hypoxic (1% O₂) protocols routinely induce anoxia (0.0% O₂). Furthermore, in physioxic culture, pericellular O₂ tension is highly dependent on cell culture parameters, making reproducibility difficult. Highlighting the significance of these findings, we demonstrated that a key drug metabolism model, primary human hepatocytes, undergo an effect similar to ischemia-reperfusion injury when cultured in physioxia. To address these challenges, we developed a reaction-diffusion model that predicts pericellular O₂ tension a priori. Using this tool, we controlled pericellular O₂ tension in two breast cancer models to explore transcriptional and translational responses to hypoxia and anoxia. We discovered that breast cancer cells respond non-monotonically to low O₂. Overall, this work calls for a fundamental change to how O₂-controlled cell culture is performed and suggests that pericellular O₂-controlled cell culture is necessary to accurately model O₂ tension.

Methods

In vitro O₂ measurements

Adhesive optical O₂ sensor spots (OXSP5-ADH-STER, PyroScience GmbH) were used to measure the O₂ concentration of media and cell cultures. Sensors were placed on the culture vessel surface and a cable adapter (SPADBAS, PyroScience GmbH) was glued on the opposite side of the culture vessel (lined up with the sensor). Glue was allowed to dry overnight. Optical fiber cables (SPFIB-BARE, PyroScience GmbH) were placed within the adapters and connected to a computer via a meter (FireSting O₂, PyroScience GmbH). The 100% O₂ calibration was performed with aerated Dulbecco’s phosphate-buffered saline (DPBS), and the 0% O₂ calibration was was performed using the factory calibration value. For cell culture experiments, cells were seeded in sensor-containing culture vessels and pericellular O₂ was measured. A temperature probe (TDIP15, PyroScience GmbH) connected to the meter was placed inside the same incubator as the sensor-containing culture vessels. To measure the O₂ concentration at the top of the media or cell culture wells, needle-like probes (OXROB10, PyroScience GmbH) were attached to a micromanipulator (MM33, PyroScience GmbH) and placed at the media–gas interface, such that the probes were submerged at the top layer of the media. Holes were drilled in plate lids to allow the probes to reach the media.

Cell culture

Mycoplasma-free cell lines, MDA-MB-231 (ATCC HTB-26), MCF7 (ATCC HTB-22), 4T1 (ATCC CRL-2539), and primary mammary epithelial (ATCC PCS-600–010) were obtained from the American Type Culture Collection (ATCC). MDA-MB-231, MCF7, and 4T1 were maintained in Leibovitz’s L-15 medium (Cytiva), Eagles’ Minimum Essential Medium (EMEM) with L-glutamine (Quality Biological) and Dulbecco’s Minimum Essential Medium (DMEM) (Corning), respectively, with 10% fetal bovine serum (FBS, Corning) and 1% penicillin/streptomycin (P/S, Invitrogen). Mammary epithelial cells were cultured in basal medium (ATC PCS-600–030) with cell growth kit (ATCC PCS-600–040). MCF7 and MDA-MB-231 hypoxia-inducible factor (HIF) reporter cell lines were transduced and selected as previously described.²⁸ For all cultures, passage number did not exceed 20.

Human dendritic cells (DCs) were differentiated from cryopreserved CD14+ monocytes over a period of 7 days as previously described.⁷¹ In brief, monocytes were seeded into a 6-well plate in ImmunoCult^™-ACF Dendritic Cell Medium (Stemcell Technologies), supplemented with recombinant human granulocyte-macrophage colony-stimulating factor (GM-CSF) (50 ng/mL) (R&D Systems) and recombinant human interleukin-4 (IL-4) (50 ng/mL) (R&D systems).

Primary human hepatocytes (HUCPG, Lonza) were cultured following the manufacturer’s protocol. Briefly, hepatocytes were thawed in thawing medium (MCHT50, Lonza), seeded onto a 24-well plate (BioCoat Collagen I, Corning) with plating medium (MP100 and MP250, Lonza), then cultured with maintenance medium (CC-3198, Lonza), which was exchanged after 24 hours.

For O₂-controlled experiments, cells were incubated within a tri-gas incubator (Heracell VIOS 160i, Thermo Fisher Scientific) which was kept closed during the duration of the experiment.

Cellular hypoxia kinetics

MCF7 cells were seeded onto 60 mm glass dishes (Cellvis) overnight and incubated with Hoechst 33342 (NucBlue Live ReadyProbes Reagent, Invitrogen), 10 μM Image-iT^™ Red Hypoxia Reagent (Invitrogen), and 1 μM Celltracker Orange for 30 minutes at 37°C. Cells were placed within a humidified incubator at 37°C, 5% CO2, and 1% O₂/99% N₂ (O₂ Module S, CO₂ Module S, and Temp Module S, Zeiss) attached to a confocal microscope (LSM 880 with Airyscan, Zeiss). Images were taken every 30 minutes for 12 hours. Average fluorescence per cell was calculated using Fiji. Briefly, cell area multiplied by background flourescence was subtracted from the cell’s integrated density. At least thirty cells were analyzed per image.

Hypoxia-inducible factor (HIF) stabilization kinetics

For cell density experiments, MCF7 reporter cells were seeded into a 12-well plate at different densities (7,000 and 29,000 cells/cm²) overnight. Cells were cultured at 1% O₂ for 4 days, incubated with Hoechst 33342 and confocal images were taken daily. Percentage of GFP+ among MCF7 reporter cells was determined using Fiji (Threshold and Analyze Particles). For culture vessel type experiments, MDA-MB-231 reporter cells were seeded onto a 24-well plate or T25 flask (30,000 cells/cm²) and cultured at 1% O2 for 5 days. Flow cytometry (Cytoflex S, Beckman Coulter) of live singlets was used to determine the GFP positive fraction.

Mitochondrial inhibition

MCF7 were seeded overnight in a 24-well plate and incubated at 5% O₂ for 24h. 10 μL of DPBS or sodium azide (Sigma Aldrich) (final concentration = 5 mM) were added into cultures. For cytotoxicity studies, MCF7 were seeded overnight and incubated with ± 5 mM sodium azide for 6 hours at 5% O₂. Cells were incubated with 1:1000 live/dead dye (LIVE/DEAD Fixable Kit, Thermo Fisher Scientific) for 1 hour. Flow cytometry (Attune NxT, Thermo Fisher Scientific) of live singlets was used to determine the cell viability.

Cell viability time course

MCF7 were seeded into a T25 flask (20,000 cells/cm²) overnight and cultured at 18.6%, 5%, or 1% O₂ for 6 days. Every 2 days, media was harvested, and trypan blue staining and a hemocytometer were used to determine live and dead detached cells. Attached cells were trypsinized (Trypsin-EDTA, Gibco) and counted using the same method.

Library preparation with polyA selection and Illumina sequencing

RNA was extracted from hepatocytes (NucleoSpin RNA, Macherey-Nagel) and quantified using a Qubit 2.0 Fluorometer (Life Technologies). RNA integrity was checked using Agilent TapeStation 4200 (Agilent Technologies). RNA sequencing libraries were prepared using the NEBNext Ultra II RNA Library Prep for Illumina per the manufacturer’s protocol (New England Biolabs). Briefly, mRNAs were enriched with Oligod(T) beads. Enriched mRNAs were fragmented for 15 minutes at 94 °C. First strand and second strand cDNA were subsequently synthesized. cDNA fragments were end repaired and adenylated at 3’ ends, and universal adapters were ligated to cDNA fragments, followed by index addition and library enrichment by PCR with limited cycles. The sequencing libraries were validated on the Agilent TapeStation 4200 and quantified by using Qubit 2.0 Fluorometer as well as by quantitative PCR (KAPA Biosystems). The sequencing libraries were multiplexed and clustered onto a flowcell. After clustering, the flowcell was loaded onto the Illumina instrument (HiSeq 400 or equivalent) according to the manufacturer’s instructions. The samples were sequenced using a 2×150bp Paired End (PE) configuration. Image analysis and base calling were conducted by the HiSeq Control Software (HCS). Raw sequence data (.bcl files) generated from Illumina HiSeq were converted into FASTQ files and de-multiplexed using Illumina bcl2fastq 2.20 software. One mismatch was allowed for index sequence identification. FASTQ files were trimmed with Trimmomatic⁷² and the quality was analyzed with FastQC.⁷³ The human genome (GRCh38.p14) was annotated and reads were aligned using STAR.⁷⁴ Gene counts were determined using FeatureCounts.⁷⁵ Differential gene expression analysis was performed using DESeq2.⁷⁶ Gene set enrichment analysis (GSEA)⁷⁷ was performed using the clusterProfiler package in R.⁷⁸

Reaction-diffusion model

The unsteady state diffusion equation (1) with initial and boundary conditions (2–4) were used to describe O₂ transfer between cell culture media and gas phase, where C is the concentration of O₂, D is the diffusivity coefficient, and k_La is the mass transfer coefficient. x = 0 is the bottom of the well and x = L is the media height. A diffusivity coefficient of 0.09684 cm²/hr was used⁷⁹ and experimental diffusion data were used to determine k_L values. An analytical solution was determined (5–6). Michaelis–Menten kinetics were used for the reaction-diffusion model³⁸ (7), where V_max is the maximum O₂ consumption rate and K_m is the O₂ concentration at which the reaction rate is half of V_max. Numerical values were determined using the MATLAB PDE solver (MathWorks).

$$\frac{dC}{dt}=D\frac{{\partial }^{2}C}{\partial t^2}$$ (1)

$$C(x,t=0)=C_0$$ (2)

$$\frac{dC(x=0,t)}{dx}=0$$ (3)

$$-D\frac{dC(x=L,t)}{dx}=k_{L}a\left(C-C_{gas}\right)$$ (4)

$$\Theta (\zeta ,\tau )=\sum _{m=1}^{\infty }\frac{2\text{sin}\left({\lambda }_m\right)e^{-{\lambda }_m{}^2\tau }\text{cos}\left({\lambda }_m\zeta \right)}{\text{cos}\left({\lambda }_m\right)\text{sin}\left({\lambda }_m\right)+{\lambda }_m}$$ (5)

$$\Theta =\frac{C-C_{\text{gas }}}{C_0-C_{gas}}\zeta =\frac{x}{L}\tau =\frac{Dt}{L^2}$$ (6)

$$\frac{dC}{dt}=D\frac{{\partial }^{2}C}{\partial t^2}-\frac{V_{\mathit{\text{max}}}C}{K_M+C}$$ (7)

Extracellular metabolite quantification

Media was removed from cell cultures and centrifuged at 250g for 5 minutes. Resulting supernatant was stored at −20°C and used for metabolite quantification. Live cells from cultures were determined using trypan blue staining and a hemocytometer. Glucose uptake and lactate secretion were quantified by an Agilent 1260 high performance liquid chromatography (HPLC) Infinity II System equipped with a BioRad Aminex HPX-87H ion exchange column (300mm × 7.8 mm) operated at 60⁰C with a refractive index detector (RID) operated at 50⁰C.^{80, 81} The mobile phase was 14mM sulfuric acid with a flow rate of 0.6 mL/min. The injection volume of each sample was 10 μL. Peak areas for each compound for concentrations ranging from 0.125 g/L to 5 g/L were used to make calibration curves in OpenLab ChemStation (LTS 01.11) and then used to calculate metabolite quantifications. Glutamate and glutamine concentration was determined using the Glutamine/Glutamate-Glo Assay (Promega). Metabolite concentration was normalized by the cell number at each time point.

Reverse transcription and quantitative PCR (RT-qPCR)

RT-qPCR was performed as previously described.^{82, 83} RNA was extracted from MCF7 cultures (NucleoSpin RNA, Macherey-Nagel) and the quality was checked using a NanoDrop One spectrophotometer (Thermo Fisher Scientific). Reverse transcription was conducted using a High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems) on a MyCycler thermal cycler (Bio-Rad). Gene expression was quantified using the following TaqMan Gene Expression Assays (Thermo Fisher Scientific) on an MX3005P QPCR System (Agilent Technologies): VEGF-A (Hs00900055_m1), CA9 (Hs00154208_m1), LDHA (Hs01378790_g1), PDK1 (Hs01561847_m1), NT5E (Hs00159686_m1), PRKAA2 (Hs00178903_m1), BNIP3L (Hs00188949_m1), BNIP3 (Hs00969291_m1), HIF1A (Hs00153153_m1), HIF2A (Hs00909569_g1), SLC2A1 (Hs00892681_m1), SLC1A5 (Hs01056542_m1), SLC7A11 (Hs00921938_m1), NDUFA4L2 (Hs00220041_m1), BNIP3 (Hs00969291_m1), BNIP3L (Hs00188949_m1), NT5E (Hs00159686_m1), CD274 (Hs00204257_m1), and ACTB (Hs01060665_g1).

Bulk proteomic sample preparation

Harvested cells were washed with PBS and resuspended in Mass Spectrometry grade Water (Fisher Scientific, W6500) and frozen at −80°C. Cells were lysed by heating at 90°C for 10 mins⁸⁴, protein concentrations for each lysate were measured using a Nanodrop (A205). Proteins were digested to peptides per the SCoPE2 protocol.^{85, 86} Briefly, 10 μg of protein per sample were digested in a solution containing 100mM triethylammonium bicarbonate at pH 8.5 (TEAB) (Sigma Aldrich, T7408), benzonase nuclease (Millipore Sigma, Cat E1014) and Trypsin Gold (Promega, V5280). The protease was added at a 1:20 enzyme to substrate ratio and LC-MS grade water was added to maintain its concentration at 20 ng/uL. The reaction was carried out for 12 hours at 37°C. Digested peptides were subsequently dried down in a SpeedVac vacuum evaporator and resuspended in 200mM TEAB (pH 8.5). Samples were randomized and labelled using either d0, d4 or d8 of mTRAQ mass tags (SciEx, 4440015, 4427698, and 4427700) in a reaction that maintained 1/3rd organic phase and the manufacturer’s suggested reagent to peptide ratio (1U for 100 μg of peptides). The labelling was carried out for 2 hours at room temperature and excess, unreacted label was quenched by adding hydroxylamine (Sigma Aldrich, 467804) to 0.2% v/v and leaving at room temperature for 1 hour. Two samples from each label were randomly selected and 50 nng analyzed in data-dependent acquisition (DDA) mode to evaluate labelling efficiency.

Samples from each label were combined in equal amounts to make a plexDIA⁵³ set that was dried down and resuspended in 0.1% formic acid (Thermo Fisher, 85178) in MS grade water to a final concentration of 1 μg/μL. Samples within a plexDIA set were randomly paired; a few samples across labels were repeated across multiple sets.

Proteomics data acquisition

The separation was performed at a constant flow rate of 200 nL/min using a Dionex UltiMate 3000 UHPLC, and 1 μL of sample was loaded onto a 25cm × 75uM IonOpticks Odyssey Series column (ODY3–25075C18). The separation gradient was 4% buffer B (80% acetonitrile in 0.1% Formic Acid) for 11.5 minutes, a 30 second ramp up to 12%B followed by a 63 minute linear gradient up to 32%B. Subsequently, buffer B was ramped up to 95% over 2 minutes and maintained as such for 3 additional minutes. Finally, buffer B was dropped to 4% in 0.1 minutes and maintained for 19.9 additional minutes.

The mass spectra were analyzed using a Thermo Scientific Q-Exactive mass spectrometer from minutes 20 to 95 of the LC method. An electrospray voltage of 1700V was applied at the liquid-liquid junction of the analytical column and transfer line. The temperature of the ion transfer tube was 250°C, and the S-lens RF level was set to 30.

Bulk data was collected in Data Independent Acquisition (DIA) mode, the duty cycle consisted of a total of 3 MS1 scans and 30 MS2 scans. All MS1 scans were conducted at 140,000 resolving power with a maximum injection time of 300 milliseconds and a target AGC of 3e6 with a scan range covering 378–1290 m/z. All MS2 scans were conducted at 35,000 resolving power, a maximum inject time of 110 ms, AGC target of 3e6 and normalized collision energy of 27. MS2 scans had variable isolation widths: 10 MS2 scans of 17 m/z isolation width (isolation window) followed the first and second MS1 scan respectively, the third MS1 was followed by 5 windows of 33 m/z, 2 windows of 40 m/z, 1 window of 80 m/z and a final window of 120 m/z.

Proteomics data processing

DIA-NN⁸⁷ (version 1.8.1) was used to search the raw files from each run. A predicted spectral library was made using the swissprot mouse FASTA database and in silico labelled to have mTRAQ as a fixed mod (+140.0949630177) on each trypsin digested peptide.

Peak height was used for quantification with a scan window of 1, mass accuracy of 10 ppm and MS1 accuracy of 5 ppm, MBR was enabled and search outputs were filtered at 1% Q value. The following commands were employed by use of the additional commands dialogue: –fixed-mod mTRAQ 140.0949630177, nK, –channels mTRAQ, 0, nK, 0:0; mTRAQ, 4, nK, 4.0070994:4.0070994; mTRAQ, 8, nK, 8.0141988132:8.0141988132}, – peak-translation, –ms1-isotope-quant, -ms1-base-profile, -ms1-subtract 2.

The report file containing filtered peptide level output was processed using R. First, the peptide level data was collapsed/summarized to a run x protein matrix using the diannmaxlfq function from the ‘diann’ R package.⁸⁷ Subsequently, the matrix was log₂ transformed and the protein levels in each run were normalized for differential loading amounts by adding to each protein value, the median of the difference between the vector of protein levels for that run to the vector of median values across all runs. Relative protein levels were obtained by subtracting away the mean value across runs for each protein.

In order to correct for biases specific to each mTRAQ label, kNN imputation (k = 3) was performed and ComBat⁸⁸ was used with mTRAQ labels as batch covariates. Post batch correction two matrices were used for further analysis, one with imputed values and the other where the imputed values had been set back to NA. Differential protein expression analysis was performed using limma.⁸⁹ Protein set enrichment analysis (PSEA)⁷⁷ was performed using the clusterProfiler⁷⁸ package in R.

Statistical analysis

All data were presented as the mean ± standard error of the mean (SEM). Statistical analyses were performed using Prism 9 software (GraphPad). The statistical tests used are outlined in the figure captions. Values represent the mean ± standard error of the mean. Significance levels are reported as *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.