The goal of this study was to characterize hypoxia-induced metabolic changes in proliferating primary LFs and PASMCs. To accomplish this goal, we used metabolic flux analysis to model how cell metabolism supports cell proliferation. Metabolic flux analysis fits cell proliferation rate, extracellular flux measurements, and ¹³C isotope labeling patterns to a computational model of cell metabolism (Antoniewicz, 2018). This analysis reconstructs comprehensive flux maps that depict the flow of carbon from extracellular substrates through intracellular metabolic pathways into cell biomass and metabolic by-products (Young, 2014). These models assume that cells are at a metabolic pseudo-steady state over the experimental time course (Buescher et al., 2015). Exponential growth phase is thought to reflect metabolic pseudo-steady state as cells in culture steadily divide at their maximal condition-specific rate, provided nutrient supply does not become limiting (Ahn and Antoniewicz, 2011; Buescher et al., 2015). Thus, we first set out to define experimental conditions to capture exponential growth phase in normoxic and hypoxic cultures.

Cells were seeded and placed into hypoxia for 24 hr prior to sample collection to provide adequate time for activation of the hypoxia-dependent transcriptional program (Figure 1A). We selected 0.5% oxygen for hypoxia as this level yielded the most reproducible phenotypic differences compared to 21% oxygen, while being physiologically relevant and above the K_M of cytochrome c oxidase (electron transport chain complex IV) for oxygen (Lee et al., 2020; Wenger et al., 2015). We identified the optimal cell seeding density and time course for exponential cell growth (Figure 1B). Hypoxia decreased cell proliferation rates (Figure 1C), but slower growth was not associated with decreased cell viability (Figure 1—figure supplement 1A). As anticipated, hypoxic cells demonstrated robust stabilization of HIF-1α associated with up-regulation of its downstream targets glucose transporter 1 (GLUT1) and lactate dehydrogenase A (LDHA; Figure 1D–H). These changes persisted for the duration of the experimental time course.

We next determined the extracellular fluxes of glucose (GLC), lactate (LAC), pyruvate (PYR), and amino acids (Figure 1I–J). Flux calculations incorporated cell growth rate, extracellular metabolite concentrations, metabolite degradation rates, and medium evaporation rate (see Materials and methods; Murphy and Young, 2013; Figure 1—figure supplement 1B–C). Interestingly, while we observed a modest increase in glucose uptake, we found that hypoxia actually decreased lactate efflux (Figure 1I). This finding was confirmed by measuring the rate of [U-¹³C₃]-lactate produced from LFs cultured with [U-¹³C₆]-glucose (Figure 1—figure supplement 2). Hypoxia decreased lactate efflux despite activating HIF-1α and increasing glycolytic enzymes expression (Figure 1D–H).

To test if more severe hypoxia would augment glycolysis, we cultured cells in 0.2% ambient oxygen (Figure 1—figure supplement 3). Under these conditions, we observed no change in glucose or lactate fluxes, similar to 0.5% oxygen culture. To test if this unexpected response was unique to LFs, we studied PASMCs under 0.5% oxygen conditions (Figure 1—figure supplement 4). Similar to LFs, we observed no change in glucose uptake and reduced lactate efflux in PASMCs. Together, these data suggest that hypoxia uncouples HIF-1α target gene expression and glycolytic flux in proliferating primary cells.

Since hypoxia did not increase glycolysis in LFs, we wanted to determine how these cells responded to HIF-1α stabilization in normoxia. To activate HIF-1α, LFs were treated with the PHD inhibitor molidustat (BAY, 10 μM) using a similar time course as our hypoxia experiments (Figure 2). Like hypoxia, BAY decreased cell growth rate (Figure 2B–C) and activated the HIF-1α transcriptional program (Figure 2D–H). Unlike hypoxia, HIF-1α stabilization in normoxia markedly increased glucose uptake and lactate efflux (Figure 2I). Although hypoxia and BAY treatments both increased HIF-1α, GLUT1, and LDHA to a similar degree, the glycolytic metabolic response differed markedly between these treatments.

In addition to glucose and lactate, we also determined the extracellular fluxes of pyruvate and amino acids (Figure 1J, Figure 1—figure supplement 3J, Figure 1—figure supplement 4J, Figure 2J). To our knowledge, this is the first comprehensive extracellular flux profiling of key metabolic substrates in primary cells. In LFs, changes in extracellular fluxes were modest overall, with hypoxia generally decreasing the fluxes of all measured metabolites. These findings were similar with 0.2% oxygen (Figure 1—figure supplement 3J).

Notably, we observed a significant decrease in glutamine consumption in hypoxic LFs. This finding contrasts with previous studies of cancer cell metabolism demonstrating increased glutamine uptake as a key feature of the metabolic response to hypoxia (Gameiro et al., 2013; Metallo et al., 2011; Wise et al., 2011). In these systems, glutamine-derived α-ketoglutarate was reductively carboxylated by isocitrate dehydrogenase enzymes to generate citrate for lipid biosynthesis. Glutamine has also been shown to support TCA cycling in hypoxia in a Burkitt lymphoma model (Le et al., 2012). Unlike LFs, PASMCs did exhibit a trend toward increased glutamine uptake (Figure 1—figure supplement 4J). To examine the relative importance of glucose and glutamine to the proliferation of these cells in hypoxia, we measured LF and PASMC growth rates in the absence of either substrate (Figure 3). In LFs, absence of either glucose or glutamine reduced cell proliferation to a similar extent (Figure 3A). In hypoxia, glucose deficiency decreased LF proliferation rate further, while glutamine deficiency had no additional impact. These findings are consistent with extracellular flux measurements demonstrating decreased glutamine consumption by LFs in hypoxia. Interestingly, neither glucose nor glutamine deficiency decreased PASMC proliferation (Figure 3B), suggesting a high degree of metabolic flexibility in these cells.

In LFs, among all of the measured amino acid fluxes, proline consumption uniquely increased (Figure 1J). Hypoxia increases collagen expression in these cells (Liu et al., 2013) and proline constitutes ~10% of the total amino acid content of collagens. Together, these data suggest an important contribution of extracellular proline to collagen production in hypoxic LFs as has been observed in other fibroblast cell lineages (Szoka et al., 2017).

In PASMCs, we observed increased consumption of the branched-chain amino acids (BCAAs) leucine and valine as well as arginine (Figure 1—figure supplement 4J), which was not observed in LFs. BCAAs are transaminated by branch chain amino transferase enzymes to branched chain α-keto acids (BCKAs). BCKAs are further metabolized to yield acyl-CoA derivatives for lipogenesis or oxidation (Crown et al., 2015; Mann et al., 2021). Previous studies have shown that hypoxia up-regulates arginase expression in hypoxic PASMCs (Chen et al., 2009; Xue et al., 2017) to support polyamine and proline synthesis required for cell proliferation (Li et al., 2001). Interestingly, activation of these metabolic pathways in hypoxia was not observed in LFs and suggests distinct metabolic dependencies of these different cell types.

Compared to hypoxia treatment, BAY demonstrated more modest effects on amino acid fluxes generally (Figure 2J). In particular, glutamate efflux was not affected by BAY treatment, while it was reduced by hypoxia. Alanine efflux was increased by BAY treatment, but decreased by hypoxia. In addition to the glucose and lactate fluxes noted above, these findings further highlight fundamental differences in the metabolic consequences of HIF-1α activation in normoxia and hypoxia.

To investigate intracellular metabolic reprogramming in hypoxic cells, we performed ¹³C stable isotope tracing with [U-¹³C₆]-glucose, [1,2-¹³C₂]-glucose, and [U-¹³C₅]-glutamine. Isotopic enrichment of downstream metabolites in glycolysis and the TCA cycle were determined by LC-MS (Figure 4—figure supplement 1, Figure 4—figure supplement 2). Small changes in the patterns of isotope incorporation were observed following hypoxia or BAY treatment. The most substantial differences were observed in pyruvate (PYR), the terminal product of glycolysis, and citrate (CIT), a central metabolic node in TCA and fatty acid metabolism (Figure 4A–C). Both hypoxia and BAY treatments decreased incorporation of glucose-derived carbon into pyruvate (Figure 4A; i.e. the unlabeled, or M0, fraction was greater). This suggests an increased contribution from an unlabeled carbon source, such as extracellular pyruvate, lactate, or alanine to the intracellular pyruvate pool following PHD inhibition.

Total citrate labeling from [U-¹³C₆]-glucose was unchanged across the treatment conditions (Figure 4B). We observed decreased M2 and M4 citrate isotopes, consistent with decreased pyruvate dehydrogenase activity in hypoxia. Interestingly, we observed increased M3 and M5 citrate isotopes. Pyruvate carboxylase catalyzes the carboxylation of pyruvate to oxaloacetate after which all three pyruvate carbons are incorporated into citrate by citrate synthase. Thus, this labeling pattern suggests a more prominent contribution of pyruvate carboxylase to sustain TCA cycle anaplerosis despite pyruvate dehydrogenase inhibition following HIF-1α activation. Compared to glucose, glutamine labeled a smaller fraction of citrate, and this labeling was decreased substantially by hypoxia or BAY treatment (Figure 4C), suggesting a less important contribution of glutamine to TCA anaplerosis under these conditions. In addition, the overall fraction of M5 citrate resulting from reductive carboxylation of glutamine-derived α-ketoglutarate was low (<7%) (Figure 4D). Although a hypoxia-mediated increase in M5 citrate was observed, the overall fraction was much less than the 10–20% levels previously reported in cancer cells (Metallo et al., 2011; Wise et al., 2011).

The stable isotope labeling patterns in PASMCs were generally similar to LFs (Figure 4—figure supplement 2). The most notable differences between LF and PASMC labeling were observed in citrate. Compared with LFs, a much lower fraction of total citrate was labeled by glucose in PASMCs. Less activity of pyruvate carboxylase in these cells was suggested by decreased M3 and M5 citrate isotopes after glucose labeling. Interestingly, the M5 citrate fraction in PASMCs was more consistent with previous reports from the cancer literature (Figure 4D), suggesting activation of glutamine anaplerosis for biomass synthesis in these cells.

The mass isotopomer distribution for a given metabolite is determined by the complex relationship among the rate of isotope incorporation into the metabolic network, the contributions of unlabeled substrates, and fluxes through related pathways. To clarify how these labeling patterns reflect changes in intracellular metabolite fluxes, we next generated metabolic flux models incorporating the extracellular flux measurements and stable isotope tracing data described above. Preliminary labeling time courses indicated that, even after 72 hr of labeling, intracellular metabolites did not reach isotopic steady state (Figure 5—figure supplement 1). Thus, we performed isotopically non-stationary metabolic flux analysis as implemented by Isotopomer Network Compartment Analysis (INCA; Jazmin and Young, 2013; Murphy and Young, 2013; Young et al., 2014).

Overall, LF and PASMC metabolic fluxes were dominated by high rates of glucose uptake and glycolysis (Figure 5—figure supplement 2). In normoxia, approximately 10% of cytoplasmic pyruvate enters the TCA cycle with the balance converted to lactate. Consistent with the extracellular flux measurements and isotope labeling patterns described above, hypoxia significantly decreased glycolysis, the TCA cycle, and amino acid metabolism (Figure 5A). A significant increase in pentose phosphate pathway flux was also observed, although the absolute flux through this pathway is low. By contrast, HIF-1α activation by BAY in 21% oxygen increased glycolysis and lactate fermentation by nearly 50% (Figure 5B), but had a similar effect as hypoxia on decreasing serine and glutamine uptake. Metabolite fluxes in DMSO-treated cells were similar to 21% oxygen controls (Table 1, Table 2).

In normoxia, the magnitude of intracellular metabolite fluxes was similar in LFs and PASMCs (Figure 5—figure supplement 2, Table 1, Table 3). Compared to LFs, PASMCs had slower rates of glycolysis and faster rates of TCA metabolism driven, in part, by increased glutamine uptake (Figure 5—figure supplement 3). In hypoxia, PASMCs exhibited similar decreases in glycolytic flux as LFs but also a marked, and unexpected, increase in TCA flux (Figure 5—figure supplement 4). The increased TCA flux in PASMCs was driven by increased glutamine consumption. This finding is similar to a prior report of glutamine-driven oxidative phosphorylation in hypoxic cancer cells (Fan et al., 2013), where oxidative phosphorylation continued to provide the majority of cellular ATP even at 1% oxygen.

Given the global decrease in bioenergetic metabolic flux in hypoxic LFs, we hypothesized that these differences may be a consequence of decreased growth rate. After normalizing metabolite fluxes in normoxia and hypoxia to the cell growth rate, a modest increase (~10%) in glycolytic flux was observed (Figure 5—figure supplement 5). This finding suggests that, while glycolysis increases relative to growth rate in hypoxic cells, regulators of cell proliferation rate override the anti-proliferative effects of the HIF-1α transcriptional program. Indeed, even after adjusting for cell growth rate, the relative increase in glycolytic flux is modest compared to the marked up-regulation of glycolytic proteins and the glycolytic potential of these cells demonstrated by BAY treatment in normoxia. BAY treatment decreased cell proliferation rate (Figure 2B–C), indicating that, unlike hypoxia, PHD inhibition in normoxia uncouples cell proliferation and metabolic flux.

One important feature of metabolic flux analysis is its ability to determine the individual forward and backward fluxes of bidirectional reactions, or the so-called exchange fluxes. Although the metabolite exchange fluxes for bidirectional reactions tend to be poorly resolved by this method (Wiechert, 2007), two observations from our models are noteworthy (Table 1, Table 2, Table 3). First, consistent with the stable isotope tracing results, the modeled rate of reductive carboxylation through reverse flux by isocitrate dehydrogenase in LFs is low (~4 fmol/cell/h), unchanged by hypoxia, and modestly increased by BAY treatment. By contrast, the rate of reductive carboxylation increases sixfold in PASMCs in hypoxia, highlighting a potentially important role for this pathway in the metabolic response of PASMCs to decreased oxygen availability (Figure 6).

Second, PHD inhibition increases lactate transport exchange flux in LFs from ~0 to 1500 and 700 fmol/cell/h with 0.5% oxygen and BAY treatment, respectively, with similar results in PASMCs (Figure 7A). This observation suggests increased lactate uptake with hypoxia or BAY treatment. To investigate this hypothesis, LFs and PASMCs were treated with [U-¹³C₃]-lactate (2 mM) and ¹³C incorporation into intracellular metabolites was analyzed by LC-MS (Figure 7B, Figure 4—figure supplement 1, Figure 4—figure supplement 2). Lactate labeled ~50% of citrate and ~20% of downstream TCA cycle metabolites (α-ketoglutarate, malate, aspartate) in both LFs and PASMCs, indicating that lactate may be an important respiratory fuel source in these cells even though lactate efflux is high. Although increased labeling of pyruvate was observed in hypoxic PASMCs, this increase did not flow through to TCA metabolites as observed in LFs (Figure 4—figure supplement 2). Lactate has been used less commonly than glucose and glutamine in stable isotope tracing studies. Faubert et al., 2017 demonstrated lactate incorporation in human lung adenocarcinoma in vivo. In this study, lactate incorporation corresponded to regions of high glucose uptake as determined by [¹⁸F]-fluorodeoxyglucose positron emission tomography, suggesting that lactate consumption can occur even in areas of high glucose utilization. Subsequently, investigators have demonstrated the importance of lactate as a metabolic fuel in vivo (Hui et al., 2020; Hui et al., 2017).

In addition to downstream metabolites, we also observed hypoxia- and BAY-dependent increases in lactate incorporation in fructose bisphosphate (FBP) and 3-phosphoglycerate (3PG). This observation is consistent with prior reports describing hypoxia-mediated increases in gluconeogenesis and glycogen synthesis (Favaro et al., 2012; Owczarek et al., 2020; Pelletier et al., 2012). These data suggest that lactate also makes a small (~5% carbon) contribution to glycogen precursors. Together, these findings from stable isotope tracing of lactate reveal its important contribution to primary cell metabolism under standard culture conditions, and also reveal increased lactate uptake in hypoxia.

To reconcile the differential effects of PHD inhibition by hypoxia and BAY, we next addressed whether hypoxia could suppress BAY-stimulated glucose and lactate fluxes (Figure 8). LFs cultured in standard growth medium were treated with BAY and placed in either 21% or 0.5% oxygen. Similar to previous experiments, BAY decreased cell growth rate, increased glucose uptake, and increased lactate efflux in 21% oxygen. However, when combined with 0.5% oxygen, BAY did not enhance lactate efflux. These data demonstrate that hypoxia antagonizes the effects of HIF-1α activation on glycolytic flux.

We observed a similar effect of hypoxia on glycolysis when using RNA interference to silence PHD2 expression and activate HIF-1α gene transcription (Figure 8—figure supplement 1). As with BAY treatment, siPHD2 stabilized HIF-1α in normoxia and increased glycolytic gene expression (Figure 8—figure supplement 1B–E); however, hypoxia inhibited siPHD2-dependent increases in glycolysis (Figure 8—figure supplement 1F). These data indicate that the metabolic consequences of BAY treatment are a direct consequence of PHD inhibition and not simply an off-target effect.

To investigate these metabolic differences further, we performed metabolomic profiling of LFs treated for 72 hr with hypoxia or BAY separately or in combination. Both 0.5% oxygen and BAY treatment caused marked changes in intracellular metabolites (Figure 9—figure supplement 1). Of 133 total metabolites, 98 were differentially regulated by hypoxia and 54 were differentially regulated by BAY. Of these, 44 were modulated by both treatments (Figure 9—figure supplement 1C). Metabolite set enrichment analysis of KEGG biochemical pathways identified increased enrichment of fatty acid biosynthesis pathways with hypoxia (Figure 9—figure supplement 1D). By contrast, BAY treatment was enriched for metabolites involved in pentose/glucuronate interconversions and glycolysis (Figure 9—figure supplement 1E). Aspartate was the most significantly decreased metabolite with both treatments, consistent with prior reports demonstrating an important role for HIF-1 regulation of aspartate biosynthesis in cancer cells (Garcia-Bermudez et al., 2018; Meléndez-Rodríguez et al., 2019).

Principal component analysis revealed greater similarity among both treatment groups cultured in 0.5% oxygen than among the BAY-treatment groups (Figure 9A). Moreover, these hypoxia-treated cells were well-segregated from cells treated with BAY alone. These observations are consistent with our metabolic flux models demonstrating an overriding effect of hypoxia per se on cell metabolism and highlighting important differences between hypoxic and pharmacologic PHD inhibition. To identify the metabolic changes that depend on hypoxia rather than PHD inhibition, we identified differentially regulated metabolites in BAY-treated cells cultured in normoxia and hypoxia (Figure 9B). Of 133 metabolites, 83 were significantly differentially regulated by hypoxia in BAY-treated cells. An enrichment analysis of these metabolites demonstrated up-regulation of arginine and proline metabolism and down-regulation of the TCA cycle as the most impacted by hypoxia in BAY treated cells (Figure 9C). Leading edge analysis highlights negative enrichment scores associated with all of the TCA metabolites detected by our platform (Figure 9D). This result indicates better preservation of TCA cycle flux in normoxic BAY-treated cells than in hypoxic cells, as suggested by our metabolic flux models where hypoxia caused a 1.5- to 2-fold reduction of TCA flux compared to a 1.1–1.5-fold reduction with BAY treatment (Figure 5).

In addition to these differential effects on polar metabolite levels, we reasoned that another critical difference between hypoxia and BAY treatment is the impact of hypoxia on cellular redox state. We measured the impact of these treatments on intracellular NAD(H) and NADP(H) redox couples (Figure 9—figure supplement 2). Hypoxia increased the NADH/NAD⁺ ratio, driven primarily by a decrease in intracellular NAD⁺. Interestingly, while BAY treatment increased the levels of NADH, a concomitant increase in NAD⁺ resulted in preservation of the NADH/NAD⁺ ratio. As NADH accumulation is a putative inhibitor of glycolytic flux (Tilton et al., 1991), this may be one mechanism by which glycolytic flux is decreased in hypoxia but not following BAY treatment. Conversely, hypoxia decreased the NADPH/NADP⁺ ratio in both control and BAY-treated cells, primarily through decreasing NADPH. This finding is consistent with prior studies of bovine coronary artery smooth muscle cells (Gupte and Wolin, 2006). In hypoxia, NADPH is generated primarily by the pentose phosphate pathway (Liu et al., 2016), where it plays an important role in the detoxification of reactive oxygen species (Fessel and Oldham, 2018; Xiao et al., 2018). Although we modeled increased flux through the pentose phosphate pathway in hypoxia (Figure 5A), the overall flux was low and apparently inadequate to maintain the NADPH/NADP⁺ ratio.

To identify the upstream regulators of the observed metabolic changes, we next performed RNA-seq transcriptomic analysis of LFs treated with hypoxia or BAY, separately or together. As anticipated, both hypoxia and BAY treatment induced substantial changes in gene expression (Figure 10—figure supplement 1). Of the 10,686 differentially expressed genes across both conditions, 869 (4%) were unique to BAY treatment in normoxia, 4002 (19%) were shared by BAY and hypoxia, while 5052 (25%) were unique to 0.5% hypoxia (Figure 10—figure supplement 1C). Gene set enrichment analysis of these differentially regulated metabolites was performed using Molecular Signatures Database ‘Hallmark’ gene sets (Liberzon et al., 2015; Figure 10—figure supplement 1D–F). As expected, both treatments were associated with enrichment of the ‘hypoxia’ and ‘glycolysis’ gene sets.

Given the disparate effects of hypoxic and pharmacologic PHD inhibition on cellular metabolism described above, we focused our transcriptomic analyses on the differences between hypoxia and BAY treatments. Principal component analysis again demonstrated clear separation among the four treatment groups (Figure 10A). The first and second principal components correspond to 0.5% oxygen and BAY treatments, respectively. Consistent with our prior observations, the combination of 0.5% oxygen plus BAY was more similar to 0.5% oxygen alone than BAY treatment alone, supporting the hypothesis that hypoxia overrides the effects of BAY treatment. To identify the transcripts driving these differences, we identified genes differentially expressed following hypoxia in BAY-treated cells (Figure 10B). An enrichment analysis of these differentially expressed genes identified pro-proliferative gene sets like ‘E2F targets’, ‘G2/M checkpoint’, and ‘MYC targets’ associated with hypoxia (Figure 10C–E). These findings were further supported by a transcription factor enrichment analysis identifying enrichment of MYC transcription factor targets associated with hypoxia, but not BAY treatment (Figure 10F, Figure 10—figure supplement 1G–H).

As one example of MYC-dependent transcriptional regulation, MYC represses the transcription of CDKN1A, which encodes the cyclin-dependent kinase and cell-cycle inhibitor p21 (García-Gutiérrez et al., 2019). CDKN1A expression increases with BAY treatment in normoxia, but is decreased by hypoxia, consistent with hypoxia-induced MYC activation (Figure 10—figure supplement 2A). While we did not observe increased CDKN1A expression by RT-qPCR in siPHD2-treated cells in normoxia, we did observe hypoxia-mediated down-regulation in both siCTL- and siPHD2-treated cells (Figure 10—figure supplement 2B), consistent with MYC transcriptional activity.

Classically, hypoxia and HIF activation are thought to inhibit cell proliferation by inhibiting pro-proliferative MYC signaling (Koshiji et al., 2004). These results indicate that hypoxia-induced MYC activation may be sustaining proliferation in LFs. We reasoned that these pro-proliferative signals may also account for the unexpected effects of hypoxia on glycolysis that we observed.

To examine the role of hypoxia-induced MYC activation in the metabolic response to hypoxia in proliferating primary cells, we first measured MYC by immunoblot. Consistent with our bioinformatic results, we observed increased MYC in hypoxia-treated cells, but not with BAY-treatment alone, where MYC was decreased (Figure 11A–B). These findings were similar in siPHD2-treated cells (Figure 11—figure supplement 1). Together, these data suggest a model whereby hypoxia activates MYC and inhibits HIF-driven increases in glycolysis (Figure 11C, top). Inhibiting MYC in hypoxia, therefore, should increase glycolysis. Conversely, HIF-dependent glycolysis after BAY treatment in normoxia should be sensitive to inhibition by MYC over-expression (Figure 11C, bottom).

To test the hypothesis that hypoxia-induced MYC expression inhibits glycolysis in primary cells, we first combined MYC knockdown with hypoxia treatment (Figure 11C, top). As expected, MYC-deficient cells proliferated more slowly in normoxia, and MYC was essential for sustaining cell proliferation in hypoxia (Figure 11E). We did not observe any significant differences in HIF-1α stabilization or target gene expression with MYC knockdown in normoxia or hypoxia (Figure 11—figure supplement 2A–E). Consistent with our hypothesis, MYC-knockdown increased lactate efflux in hypoxia, unlike control siRNA-treated cells (Figure 11F).

We next performed the complementary experiment to determine whether MYC over-expression could attenuate the increase in glycolysis observed with BAY treatment in normoxia (Figure 11C, bottom). We did not observe any significant differences in cell growth rate with MYC over-expression (Figure 11H). Interestingly, MYC over-expression potentiated BAY-stimulated increases in HIF-1α stabilization and target gene expression (Figure 11—figure supplement 2F–J). In spite of these transcriptional changes, MYC over-expression attenuated BAY-stimulated lactate efflux in normoxia. Together, these data indicate that hypoxia-induced MYC expression may be one factor that uncouples the HIF transcriptional program from glycolytic flux in proliferating primary cells.

Primary normal human lung fibroblasts (LFs) were purchased from Lonza (CC-2512) and cultured in FGM-2 (Lonza CC-3132). Cells from three donors were used in these studies: #33652 (56 y.o., male), #29132 (19 y.o., female), and #41684 (37 y.o., male). Passages 3–6 were used for experiments. Primary human pulmonary artery smooth muscle cells were purchased from Lonza (CC-2581) and cultured in SmGM-2 (Lonza CC-3182). Cells from three donors were used in these studies: #30020 (64 y.o., male), #27662 (35 y.o., male), #26698 (51 y.o., male), #19828 (51 y.o., male) and #45518 (56 y.o., female). Passages 4–7 were used for experiments. Cell authentication was performed by the vendor. Cells were maintained in a standard tissue culture incubator in 5% CO₂ at 37 °C.

For extracellular flux measurements, cells were seeded in either standard growth medium or MCDB131 medium lacking glucose, glutamine, and phenol red (genDEPOT) which was supplemented with 2% dialyzed fetal bovine serum (Mediatech) and naturally labeled glucose and glutamine (‘light’ labeling medium). For LFs, glucose was supplemented at 8 mM and glutamine was supplemented at 1 mM. For PASMCs, glucose was supplemented at 5.55 mM and glutamine was supplemented at 10 mM. These concentrations match the concentrations of these substrates determined in standard growth medium. Preliminary experiments were performed to identify the optimal cell seeding density, exponential growth phase, and labeling duration consistent with metabolic and isotopic steady state. On Day –1, 25,000 cells were seeded in a 35 mm dish in ‘light’ labeling medium. Hypoxia-treated cells were transferred to a tissue culture glovebox set at 0.5% oxygen and 5% CO₂ (Coy Lab Products). Medium was supplemented with DMSO 0.1% or BAY (10 μM) for these treatment conditions. On Day 0, cells were washed with PBS and the medium was changed to either ‘light’ labeling medium for flux measurements or ‘heavy’ labeling medium containing [1,2-¹³C₂]-glucose, [U-¹³C₆]-glucose, [U-¹³C₅]-glutamine, or [U-¹³C₂]-lactate for tracer experiments. For LFs, samples were collected on Day 0 and every 24 hr for 72 hr. For PASMCs, samples were collected on Day 0 and every 12 hr for 48 hr. Medium and cell lysates were collected at each time point for intra- and extracellular metabolite measurements and total DNA quantification. Dishes without cells were weighed daily to correct for evaporative medium losses and to empirically determine degradation and accumulation rates of metabolites. Medium samples and cell lysates for DNA measurement were stored at –80 °C until analysis. Each individual experiment included triplicate wells for each treatment and time point, and each experiment was repeated 4–8 times.

Direct cell counts of trypsinized cell suspensions in PBS were obtained following staining with propidium iodide and acridine orange using a LUNA-FL fluorescence cell counter (Logos Biosystems). Indirect cell counts for flux measurements were interpolated from total DNA quantified using the Quant-iT PicoGreen dsDNA Assay Kit (Thermo). Cells were washed once with two volumes of PBS, lysed with Tris-EDTA buffer containing 2% Triton X-100, and collected by scraping. Total DNA in 10 μL of lysate was determined by adding 100 μL of 1 X PicoGreen dye in Tris-EDTA buffer and interpolating the fluorescence intensity with a standard curve generated using the λ DNA standard. Cell counts were interpolated from a standard curve of DNA obtained from known cell numbers seeded in basal medium (Figure 12A). No difference in total cellular DNA was identified between normoxia and hypoxia cultures (Figure 12B–C).

Cells were washed with one volume of PBS and collected by scraping in PBS. Cell suspensions were centrifuged at 5000×g for 5 min at 4 °C. Pellets were lysed in buffer containing Tris 10 mM, pH 7.4, NaCl 150 mM, EDTA 1 mM, EGTA 1 mM, Triton X-100 1% v/v, NP-40 0.5% v/v, and Halt Protease Inhibitor Cocktail (Thermo). Protein concentrations were determined by BCA Protein Assay (Thermo). Lysates were normalized for protein concentration and subjected to SDS-PAGE separation on stain-free tris-glycine gels (Bio-Rad), cross-linked and imaged with the Chemidoc system (Bio-Rad), transferred to PVDF membranes with the Trans-Blot Turbo transfer system (Bio-Rad), imaged, blocked in 5% blocking buffer (Bio-Rad), blotted in primary and secondary antibodies, and developed using WesternBright ECL (Advansta). Band signal intensity was normalized to total protein per lane as determined from the stain-free gel or membrane images.

Total RNA was isolated from cells with the RNeasy Mini Kit (Qiagen). cDNA was synthesized from 0.25 to 1.00 ng RNA with the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems). RT-qPCR analysis was performed with an Applied Biosystems 7500 Fast Real Time PCR System with TaqMan Universal PCR Master Mix and pre-designed TaqMan gene expression assays (Life Technologies). Relative expression levels were calculated using the comparative cycle threshold method referenced to ACTB.

Medium samples were diluted 10-fold in PBS. Glucose concentration was determined using the Glucose Colorimetric Assay Kit (Cayman) according to the manufacturer’s protocol. Standards were prepared in PBS.

Medium samples were diluted 10-fold in PBS. Glucose concentration was determined using the ʟ-Lactate Assay Kit (Cayman). Medium samples did not require deproteinization, otherwise the samples were analyzed according to the manufacturer’s protocol. Standards were prepared in PBS.

Pyruvate was measured using either an enzymatic assay (most samples) or an HPLC-based assay (medium from 0.2% oxygen experiments). For the enzymatic assay, medium samples were diluted 20-fold in PBS. Pyruvate concentration was determined using the Pyruvate Assay Kit (Cayman). Medium samples did not require deproteinization, otherwise the samples were analyzed according to the manufacturer’s protocol. Standards were prepared in PBS. For the HPLC assay, 2-oxovaleric acid was added to medium samples as an internal standard. Samples were subsequently deproteinized with 2 volumes of ice-cold acetone. Supernatants were evaporated to <50% of the starting volume at 43 °C in a SpeedVac concentrator (Thermo Savant) and reconstituted to the starting volume with HPLC-grade water prior to derivatization. Samples were derivatized 1:1 by volume with o-phenylenediamine (25 mM in 2 M HCl) for 30 min at 80 °C. Derivatized pyruvate was separated with a Poroshell HPH C-18 column (2.1×100 mm, 2.7 μm) on an Infinity II high-performance liquid chromatography system with fluorescence detection of OPD-derivatized α-keto acids as described previously (Guarino et al., 2019).

Medium amino acid concentrations were determined following the addition of norvaline and sarcosine internal standards and deproteinization with 2 volumes of ice-cold acetone. Supernatants were evaporated to <50% of the starting volume at 43 °C in a SpeedVac concentrator (Thermo Savant) and reconstituted to the starting volume with HPLC-grade water prior to analysis. Amino acids in deproteinized medium were derivatized with o-phthalaldehyde (OPA) and 9-fluorenylmethylchloroformate (FMOC) immediately prior to separation with a Poroshell HPH-C18 column (4.6×100 mm, 2.7 μm) on an Infinity II high-performance liquid chromatography system with ultraviolet and fluorescence detection of OPA- and FMOC-derivatized amino acids, respectively, according to the manufacturer’s protocol (Agilent; Long, 2017).

Approximately 1.25 M LFs were reverse transfected in 6 cm dishes with 40 pmol siRNA or non-targeting siCTL pools (Dharmacon) and 20 μL RNAiMAX (Thermo) in 500 μL OptiMEM (Thermo). After 24 hr, cells were collected by trypsinization and re-seeded as described in Metabolic flux protocol above for growth rate and lactate efflux measurements.

LFs were seeded at 25,000 cells per 35 mm dish on Day –2. On Day –1, cells were transduced with adenovirus for MYC (Vector Biolabs) or YFP overexpression (Oldham et al., 2015). After 24 hr, the medium was changed and samples were collected as described in Metabolic flux protocol above.

The growth rate (μ) and flux (v) for each measured metabolite were defined as follows Murphy and Young, 2013:(1)dXdt=μX(2)dMdt=−kM+vX

where X is the cell density, k is the first-order degradation or accumulation rate, and M is the mass of the metabolite. These equations are solved as follows:(3)X=X0eμt(4)Mekt=vX0μ+k(e(μ+k)t−1)+M0

Growth rate (μ) and cell count at time 0 (X0) were determined by robust linear modeling of the logarithm of cell count as a function of time (t). Metabolite mass was calculated from the measured metabolite concentrations and predicted well volume accounting for evaporative losses (Figure 1—figure supplement 1E). First-order degradation and accumulation rates were obtained from robust linear modeling of metabolite mass v. time in unconditioned culture medium. Rates that significantly differed from 0 by Student’s t-test were incorporated into the flux calculations. Final fluxes were obtained by robust linear modeling of Mekt versus eμ+kt-1 to determine the slope from which v was calculated using Equation 4.

The dry weight of LFs was determined to be 493 pg/cell. The dry weight of PASMCs was determined to be 396 pg/cell. These values were estimated by washing 3×10⁶ cells twice in PBS and thrice in ice-cold acetone prior to drying overnight in a SpeedVac. The composition of the dry cell mass was estimated from the literature (Quek et al., 2010; Sheikh et al., 2005), and stoichiometric coefficients were determined as described (Murphy and Young, 2013; Zamorano et al., 2010).

Metabolic flux analysis was performed using the elementary metabolite unit-based software package INCA (Young, 2014). Inputs to the model include the chemical reactions and atom transitions of central carbon metabolism, extracellular fluxes, the identity and composition of ¹³C-labeled tracers, and the MIDs of labeled intracellular metabolites. The metabolic network was adapted from previously published networks (Murphy and Young, 2013; Vacanti et al., 2014) and comprises 48 reactions representing glycolysis, the pentose phosphate pathway, the tricarboxylic acid cycle, anaplerotic pathways, serine metabolism, and biomass synthesis. The network includes seven extracellular substrates (aspartate, cystine, glucose, glutamine, glycine, pyruvate, serine) and five metabolic products (alanine, biomass, glutamate, lactate, lipid). Models were fit using three ¹³C-labeled tracers, [1,2-¹³C₂] glucose, [U-¹³C₆] glucose, and [U-¹³C₅] glutamine. The MIDs of twelve metabolites (2-oxoglutarate, 3-phosphoglycerate, alanine, aspartate, citrate, fructose bisphosphate, glutamate, glutamine, lactate, malate, pyruvate, serine) were used to constrain intracellular fluxes. The following assumptions were made:

Flux estimation was repeated a minimum of 50 times from random initial values. Results were subjected to a χ² statistical test to assess goodness-of-fit. Accurate 95% confidence intervals were computed for estimated parameters by evaluating the sensitivity of the sum-of-square residuals to parameter variations (Antoniewicz et al., 2006; Murphy and Young, 2013).

Cellular NAD⁺ and NADH were measured using an enzymatic fluorometric cycling assay based on the reduction of NAD⁺ to NADH by alcohol dehydrogenase (ADH) and subsequent electron transfer to generate the fluorescent molecule resorufin (Oldham et al., 2015). Briefly, cells were washed twice with one volume PBS. Pyridine nucleotides were extracted on ice with buffer containing 50% by volume PBS and 50% lysis buffer (100 mM sodium carbonate, 20 mM sodium bicarbonate, 10 mM nicotinamide, 0.05% by volume Triton-X-100, 1% by mass dodecyltrimethylammonium bromide) and collected by scraping. Extracts were divided equally and 0.5 volume of 0.4 N HCl was added to one sample. Both extracts were heated at 65 °C for 15 min to degrade selectively either the oxidized (buffer) or reduced (HCl) nucleotides. The reaction was cooled on ice and quenched by adding 0.5 M Tris-OH to the acid-treated samples or 0.2 N HCl plus 0.25 M Tris-OH to the buffer samples. Samples were then diluted in reaction buffer (50 mM EDTA and 10 mM Tris, pH 7.06). Cell debris was pelleted by centrifugation, and 50 μL was incubated for 2 h with 100 μL reaction buffer containing 0.6 M EtOH, 0.5 mM phenazine methosulfate, 0.05 mM resazurin, and 0.1 mg/mL ADH. Fluorescence intensities were measured with a Spectramax Gemini XPS (Molecular Devices) with excitation 540 nm, emission 588 nm, and 550 nm excitation cut-off filter. Sample intensities were compared to a standard curve generated from known concentrations of NADH. The ratio of fluorescence in buffer-extracted to acid-extracted samples corresponds to the NADH/NAD⁺ ratio. Absolute NADH and NAD⁺ were normalized to estimated cell counts from total DNA quantification as described above.

Samples for NADP⁺ and NADPH quantification were prepared as described above and analyzed using the luminescence-based NADP/NADPH-Glo assay (Promega).

RNA was collected from LFs treated for three days ± hypoxia ± BAY as described above. Four biological replicates were analyzed. Library construction and sequencing was performed by BGI Genomics using 100 bp paired end analysis and a read depth of 50 M reads per sample. Sequences were deposited in the NIH SRA (PRJNA721596). Sequences were mapped to the human GRCh38 primary assembly and counts summarized using Rsubread (Liao et al., 2019). This data is available from the Oldham Lab GitHub repository (https://github.com/oldhamlab/rnaseq.lf.hypoxia.molidustat; Oldham, 2021 ). Differentially expressed transcripts were identified using DESeq2 (Love et al., 2014). Gene set enrichment and transcription factor enrichment was performed using the fgsea and DoRothEA R packages, respectively (Garcia-Alonso et al., 2019; Korotkevich et al., 2021).

The raw data and annotated analysis code necessary to reproduce this manuscript are contained in an R package research compendium available from the Oldham Lab GitHub repository (https://github.com/oldhamlab/Copeland.2023.hypoxia.flux; Oldham, 2023). Data analysis, statistical comparisons, and visualization were performed in R (R Development Core Team, 2023). Experiments included technical and biological replicates as noted above. The number of biological replicates (n) is indicated in the figure legends. Summary data show the mean ± SEM. Outliers were identified using twice the median absolute deviation as a cutoff threshold. Comparisons were performed using linear mixed-effects models with oxygen, treatment, and their interaction as fixed effects and biological replicate as a random effect. Significant differences in estimated marginal means were identified by comparisons to the multivariate t distribution. Metabolomics and RNA-seq data were analyzed as described above. Probability values less than 0.05 were considered statistically significant.