TNFα-induced metabolic reprogramming drives an intrinsic anti-viral state

Cytokines induce an anti-viral state, yet many of the functional determinants responsible for limiting viral infection are poorly understood. Here, we find that TNFα induces significant metabolic remodeling that is critical for its anti-viral activity. Our data demonstrate that TNFα activates glycolysis through the induction of hexokinase 2 (HK2), the isoform predominantly expressed in muscle. Further, we show that glycolysis is broadly important for TNFα-mediated anti-viral defense, as its inhibition attenuates TNFα’s ability to limit the replication of evolutionarily divergent viruses. TNFα was also found to modulate the metabolism of UDP-sugars, which are essential precursor substrates for glycosylation. Our data indicate that TNFα increases the concentration of UDP-glucose, as well as the glucose-derived labeling of UDP-glucose and UDP-N-acetyl-glucosamine in a glycolytically-dependent manner. Glycolysis was also necessary for the TNFα-mediated accumulation of several glycosylated anti-viral proteins. Consistent with the importance of glucose-driven glycosylation, glycosyl-transferase inhibition attenuated TNFα’s ability to promote the anti-viral cell state. Collectively, our data indicate that cytokine-mediated metabolic remodeling is an essential component of the anti-viral response.

Consistent with an important role for glycosylation, we find that inhibition of glycosyltransferases also inhibits TNFα's anti-viral activity. Together, our data indicate that TNFα-induced glycolysis promotes UDP-sugar turnover to support glycosylation, which is required for the expression of intrinsic anti-viral factors and the induction of an anti-viral cellular state.

TNFα induces glycolysis as part of a broadly altered metabolic state
To elucidate how TNFα treatment impacts cellular metabolism, we employed LC-MS/MS to analyze cellular metabolic pools in vehicle-treated non-transformed human foreskin fibroblasts (HFFs) versus those treated with TNFα (S1 Table). Principal-component analysis (PCA) of the resulting data discriminated between vehicle and TNFα-treated samples in the first principal component (Fig 1A). Further, hierarchical clustering of these data segregated TNFα treated samples from control samples (Fig 1B). Together, these data suggest that TNFα treatment induces a distinct metabolic state. The relative abundance of twelve metabolites were significantly altered by TNFα treatment (Fig 1C and S2 Table). Notably, the largest metabolite increase was in kynurenine abundance, which was~20-fold more abundant in TNFα-treated cells relative to control ( Fig 1C). Kynurenine is part of the tryptophan catabolic pathway that supports NAD + biosynthesis [25,26], however, in this study NAD + pools were significantly decreased upon TNFα treatment ( Fig 1C). Intriguingly, a similar response, i.e., increased kynurenine and decreased NAD + levels, was recently shown to occur after inflammatory challenge in macrophages, and subsequently found to be important for proper innate immune responses [27]. In addition, two of the most significantly increased metabolites, ribose-phosphate and sedoheptulose-7-phosphate, are part of the pentose phosphate pathway (Fig 1C and 1D). Hexose-phosphate was also significantly increased by TNFα treatment (Fig 1C and 1D), as was UDP-glucose, a key glycosylation precursor and glycogen building block (Fig 1C and 1D).
Given TNFα's impact on glycolytic and pentose phosphate metabolite abundances, we more thoroughly investigated the impact of TNFα treatment on metabolites from these pathways ( Fig 1D) over a time course. TNFα treatment substantially increased several glycolytic metabolite pools over multiple time points (Fig 2A and S3 Table). Four hours after TNFα treatment, the levels of fructose-1,6-bisphosphate, whose production is one of the rate-determining steps of glycolysis [28], more than doubled. Other glycolytic pools were induced by TNFα treatment at multiple time points including fructose-6-phosphate, dihydroxyacetone phosphate, and glyceraldehyde-3-phosphate (Fig 2A). Similarly, UDP-glucose of the hexosamine pathway was induced at 12 and 24 hours (Fig 2A). For pentose phosphate metabolites: sedoheptulose-7-phosphate was only induced at 24 hours post-TNFα treatment, whereas ribose-phosphate was induced at all time points analyzed (Fig 2A).
The steady-state elevation of several glycolytic intermediates suggested that TNFα-treatment may induce glycolysis. To test this possibility, we analyzed the impact of TNFα on glucose consumption and lactate secretion, which were both substantially induced by TNFα treatment (Fig 2B and 2C). Treatment with 2-deoxyglucose (2DG), a glycolytic inhibitor, diminished basal levels of glycolysis and strongly inhibited TNFα-mediated glycolytic activation as shown by reduced lactate secretion compared to vehicle-treated samples (Fig 2C). Collectively, our data indicate that TNFα treatment promotes significant metabolic alterations and drives glycolytic induction.
PCA indicated that the largest amount of data variance, i.e., 37.1% of the variance associated with PC1, separated all 2DG treated from the non-2DG treated samples regardless of TNFα Hierarchical clustering of metabolite data depicted as Z-scores from min (blue) to max (red). c Metabolites significantly altered by TNFα treatment. d Schematic representing metabolites in pathways of interest. Metabolites depicted in black are experimentally detected, whereas those in grey are not detected. Metabolites in red are significantly more abundant in at least one time point during 24 hours of TNFα treatment (Fig 2A). Solid lines represent a direct metabolic conversion while dashed lines represent an indirect conversion. Single headed arrows represent an irreversible reaction, double headed arrows represent a reversible reaction.  TNFα treatment induces glycolytic flux. a HFFs were treated with TNFα (10 ng/mL, red) or vehicle (black) for 0, 4, 12 and 24 hr prior to metabolite extraction and downstream LC-MS/MS analysis (mean ± SD, n = 3). For each metabolite/time point analyzed, vehicle and TNFα treated groups were compared using student's unpaired two-tailed t-test ( � t<0.1, �� t<0.05, ��� t<0.01). Solid arrows represent direct metabolite conversions while dashed arrows represent indirect conversions. Single-headed arrows represent irreversible conversions while double-headed arrows represent reversible conversions. (b/c) HFFs were treated with TNFα (10 ng/mL, light red or 100 ng/mL, dark red) or vehicle (black) for 24 hr. Media harvested from each sample were analyzed for b glucose consumption and c lactate secretion. (mean ± SD, n = 3, with FDR adjusted ANOVA p-values, � p<0.05, �� p<0.01, ��� p<0.001). https://doi.org/10.1371/journal.ppat.1010722.g002

PLOS PATHOGENS
TNF and metabolic remodeling treatment ( Fig 3A). In the non-2DG treated samples, TNFα treated samples were separated from non-TNFα treated samples along PC2 (Fig 3A). In contrast, co-treatment with 2DG largely collapsed this separation between TNFα treated and non-TNFα treated samples along PC2 (Fig 3A). Similarly, hierarchical clustering of the metabolic data primarily separated samples based on whether they were 2DG treated or not ( Fig 3B). Further, as would be expected, 2DG treatment reversed the TNFα-mediated induction in glycolytic pools' sizes ( Fig 3B, Cluster I, and 3C) and lactate secretion (Fig 2C).
To determine how TNFα-mediated glycolytic activation contributes to its ability to induce the anti-viral state, we assessed how glycolytic inhibition impacted TNFα's ability to limit viral infection of the laboratory strain of HCMV, AD169. For these experiments, cells were pretreated with TNFα alone, or in combination with 2DG, for 24h prior to removing all compounds and infecting cells in medium without any compounds. Pretreatment of cells with TNFα largely blocked the ability of HCMV to initiate infection in HFFs (Fig 3D and 3E). TNFα pretreatment in the presence of 2DG largely restored the ability of HCMV to infect cells (Fig 3D and 3E). We also found that glucose starvation during TNFα pretreatment significantly increased the ability of HCMV to infect cells compared to TNFα pretreatment in glucose-containing medium ( Fig 3D). Pretreatment with TNFα prior to infection at a high multiplicity of infection (MOI = 3.0) substantially reduced the accumulation of HCMV proteins throughout infection ( Fig 3F). Similar to the initiation experiments, TNFα pretreatment in the presence of 2DG partially restored HCMV protein expression ( Fig 3F). These results suggest that TNFα's ability to promote an anti-viral cell state against HCMV relies on glucose availability and glycolytic flux.
To test how broadly TNFα relies on glycolysis for its antiviral activity in different cell types, we assessed AD169 infection in MRC5 lung fibroblasts, and TB40/e infection, a more recently derived clinical isolate of HCMV, in ARPE-19 retinal epithelial cells. Pretreatment with TNFα was sufficient to restrict AD169 and TB40/e infection in MRC5 and ARPE19 cell types, respectively, and restricting glycolysis with 2DG during TNFα pretreatment largely restored the ability of HCMV to initiate infection as in our original model with AD169 and HFFs ( Fig 3G). Collectively, these data indicate that TNFα relies on glycolysis to induce an anti-viral state in multiple different cell types.
Given that TNFα is broadly anti-viral, we next sought to determine if TNFα's effects on glycolysis were specific to limiting HCMV infection, or if a similar phenotype could be observed in the context of evolutionary divergent viruses. To address this question, we analyzed the replication of two β-Coronaviruses, OC43 and SARS-CoV-2 in HFFs or HFFs transduced with the angiotensin-converting enzyme 2 (ACE2), the surface protein required for SARS-CoV-2 entry [29], respectively. TNFα pretreatment was sufficient to restrict the RNA accumulation of both OC43 and SARS-CoV-2 (Fig 3H and 3I). TNFα pretreatment in the presence of 2DG largely restored OC43 and SARS-CoV-2 RNA accumulation (Fig 3H and 3I). These data align with the results from HCMV infection and support that TNFα treatment broadly requires glycolysis for its anti-viral activity.

Hexokinase 2 contributes to TNFα-mediated glycolytic activation
To obtain a complimentary picture of TNFα-induced metabolic changes and how glycolysis affects these changes, we analyzed the impact of TNFα treatment in the presence or absence of 2DG on the proteome. The abundances of 3,780 unique proteins were identified in this study (S5 Table). From this list, we extracted those involved in metabolism to study the differences in the abundance of metabolic enzymes between vehicle and TNFα treated cells. Of the 562 metabolic enzymes detected in our study, 40 were significantly more abundant as a Glycolytic flux is required for TNFα to induce its anti-viral state. (a-c) HFFs treated with TNFα (10 ng/mL, red), 2DG (20 mM, blue), TNFα+2DG (purple), or vehicle (black) for 24 hr (n = 4). Metabolites extracted from cells and analyzed by LC-MS/MS (mean ± SD, n = 4). Metabolite data analyzed via a PCA or b Hierarchical clustering with metabolite values depicted as Z-scores from min (blue) to max (red). c Relative abundance of indicated metabolite. d HFFs treated with TNFα (10 ng/mL, red) in the presence or absence of glucose (Glc ±) or 2-deoxyglucose (2DG, 20 mM) for 24 hr prior to a media change with fresh viral adsorption media containing glucose, but without result of TNFα treatment (S6 Table). Hierarchical clustering ( Fig 4A) and PCA (S1 Fig) of this subset of metabolic enzymes separated TNFα-treated samples from vehicle-treated samples, suggesting that TNFα treatment substantially impacts the expression of metabolically categorized proteins. Major contributors to this shift included solute carrier family 2, facilitated glucose transporter member 1 (SLC2A1, the ubiquitous GLUT1 glucose transporter), and hexokinase-2 (HK2), the muscle predominant form of hexokinase ( Fig 4A and  4B), both of which are proteins involved in glucose metabolism/glycolysis. HK2 protein was induced~2.5 fold in TNFα-treated samples relative to vehicle ( Fig 4C). Likewise, HK2 RNA was also substantially induced upon TNFα treatment ( Fig 4D). TNFα-treatment induced GLUT1 RNA (Fig 4E), as well as GLUT1 protein (Fig 4F and 4G). Notably, 2DG treatment prevented the TNFα-mediated accumulation of GLUT1 bands (Fig 4F). These bands have previously been found to be glycoforms of GLUT1, whose accumulation is dependent on glycosyltransferase activity [30]. Further, GLUT1 RNA levels remained elevated upon cotreatment with TNFα and 2DG (Fig 4E), suggesting the loss of these GLUT1 glycoforms appears to be post-transcriptional. We hypothesized that the loss of these GLUT1 glycoforms might reflect reduced glycosyltransferase activity. To explore this possibility we treated cells with TNFα and either Kifunensine (kif), an inhibitor of N-linked glycosylation [30], or with Benzyl-α-GalNAc (BGNAc), an inhibitor of O-linked glycosylation [31,32]. While the N-linked and O-linked glycosyltransferase inhibitors had slightly different impacts on TNFα-induced GLUT1 glycoform accumulation (Fig 4F), treatment with either BGNAc or kif substantially reduced the expression of glycosylated GLUT1 isoforms ( Fig  4F), consistent with the possibility that 2DG may be inhibiting the glycosylation of GLUT1. Additionally, 2DG treatment did not impact the accumulation of total GLUT1 peptide levels (Fig 4G), suggesting the loss of GLUT1 glycoforms is not a result of decreased GLUT1 protein expression, and consistent with 2DG treatment causing a defect in post-translational GLUT1 glycosylation. While 2DG cotreatment with TNFα reduced the accumulation of GLUT1 glycoforms, neither TNFα treatment alone nor in combination with 2DG, BGNAc, or kif, impacted the total abundance of glycoproteins in the cell (S2 Fig).
In contrast to GLUT1 and HK2, no other glycolytic enzymes were significantly increased as a result of TNFα treatment, however, PFK-M and TIGAR levels were significantly reduced by TNFα treatment (Fig 4B and 4C). We also sought to examine if TNFα could be inducing Hif1α expression given its role as a glycolysis activator [33]. While Hif1α was not detected in the proteomics data set after 24 hours of TNFα treatment, its mRNA levels were induced 2-fold at 8 hours post-TNFα treatment relative to vehicle-treated samples ( Fig 4H).
Given that HK2 is the muscle-predominant hexokinase isoform, and is not thought to be substantially expressed in fibroblasts, we sought to determine the importance of TNFαinduced expression of this isoform for glycolytic activation. We targeted HK2 and HIF1α via CRISPR-Cas9 to assess their contributions to TNFα-mediated glycolytic activation and inhibitors, and HCMV-GFP (MOI = 0.5). Infected cells were identified by GFP expression 24 hr post-infection and quantified (mean ± SD, n = 12). e HFFs pretreated for 24 hr prior to media change with fresh viral adsorption, but without inhibitors, and a known number of HCMV-GFP infectious particles. Viral plaques were quantified (n = 6, relative mean ± SD). f Cells were pre-treated as in (e), infected with HCMV (MOI = 3), and processed for western analysis at the indicated times post-infection. g HFF, MRC5, and ARPE19 cells were treated as described in (e) prior to infection with HCMV-GFP (AD169, MOI = 0.5) or clinical strain of HCMV, TB40, containing mCherry (MOI = 0.02) as indicated. Infected HFFs and MRC5s were identified by GFP expression 24 hr post-infection and infected ARPE-19s were identified by mCherry expression 72 hr post-infection. Percent infection relative to vehicle was calculated for each cell type as described in materials and methods (mean ± SD, n = 6). h Wild-type HFFs or i HFFs transduced with ACE2 were pretreated as in (e) prior to infection with h OC43 (MOI = 3) or i SARS-CoV-2 (MOI = 0.01) and harvested 48 hr post-infection for RT-qPCR analysis (mean ± SD, n = 6 (h) or n = 3 (i)). (c-e,g-i) p-values were calculated using two-way ANOVA and FDR corrected; � p<0.05, �� p<0.01, ��� p<0.001.   (Fig 4I). While Hif1α protein could not be detected by western blot, genomic ablation of the HIF1A gene was confirmed via genomic sequence analysis, with a knockout-score of 89% (S3 Fig). Notably in HIF1A targeted cells, HK2 protein expression was somewhat reduced, raising the possibility that HK2 could be regulated by Hif1α (Fig 4I). We first tested the effect of HK2 and HIF1A knockout (KO) on lactate secretion compared to non-targeting guide (ntg) control cells. HK2 KO cells treated with TNFα did not induce lactate secretion relative to vehicle-treated HK2 KO cells ( Fig 4J), but rather maintained basal lactate secretion observed in vehicle-treated ntg controls cells (Fig 4J). In addition to glycolysis and lactate secretion, the hexokinase-catalyzed product, glucose-6-phosphate, can be converted into glucose-1-phosphate to support the hexosamine pathway and the production of UDP-glucose (Fig 1D), the levels of which are induced by TNFα (Fig 2A). We, therefore, analyzed how HK2 contributed to TNFα-mediated induction of UDP-glucose pools and found that TNFα-induced accumulation of UDP-Glc is not affected by the absence of HK2 ( Fig 4K). Together, these data suggest that HK2 is important for the full glycolytic activation induced by TNFα, but that HK2 is dispensable for basal levels of glycolysis, as well as for some other aspects of TNFα-mediated metabolic alterations, e.g., the elevation of UDP-Glc levels.
We next tested if HK2 or Hif1α are necessary for TNFα to promote an anti-viral state. As expected, control ntg cells pretreated with TNFα in the presence of the glycolytic inhibitor 2DG are more permissive to HCMV infection compared to cells pretreated with TNFα alone ( Fig 4L). However, HK2 KO and HIF1A KO cells pretreated with TNFα fully restricted HCMV initiation of infection ( Fig 4L), indicating these factors are not necessary for TNFα's anti-viral activity. Interestingly, non-TNFα treated HK2 KO and HIF1A KO cells showed an increase in HCMV plaque formation relative to the vehicle pretreated control ntg cells, suggesting that HIF1A and HK2 may play a role in restricting HCMV infection that is independent of TNFα signaling ( Fig 4L).

Glycolytic inhibition attenuates the accumulation of intrinsic anti-viral proteins upon TNFα treatment
To get a more global picture of how glycolytic inhibition affects the TNFα-induced anti-viral state, we reexamined the proteomic data set to compare the impact of TNFα treatment in the presence and absence of 2DG (S5 Table). PCA of the total proteomics data set show that TNFα-treated samples in the presence or absence of 2DG segregated from non-TNFα treated samples largely along PC2, whereas 2DG-treated samples segregated from non-2DG treated samples along PC1 ( Fig 5A). Notably, co-treatment with 2DG and TNFα blunted the TNFαinduced response as indicated by a much-reduced shift along PC2. In contrast, co-treatment with 2DG and TNFα induced a larger leftward shift along PC1 than was observed with 2DG

PLOS PATHOGENS
TNF and metabolic remodeling treatment alone ( Fig 5A). These data suggest that 2DG treatment blunts the impact of TNFα treatment on the proteome, and further that TNFα and 2DG co-treatment accelerates the proteomic impact over 2DG alone.
To further explore the impact of 2DG treatment on TNFα-induced proteomic changes, we analyzed the proteins whose abundances were significantly depleted or induced by co-treatment with 2DG and TNFα relative to TNFα alone (S7 and S8 Tables, respectively). Gene Ontology (GO) analysis indicated that proteins significantly up-regulated in response to 2DG +TNFα co-treatment relative to TNFα treatment alone belonged to a diverse array of pathways including receptor-mediated endocytosis, the unfolded protein response, and various oxidative and metabolic stress responses (S8 Table and S4 Fig). In contrast, the pathways associated with proteins that exhibited substantially lower levels during 2DG and TNFα co-treatment relative to TNFα alone shared a common theme, they are important to the innate immune antiviral response (Fig 5B and S7 Table). These ontological pathways included responses to IFN, defense responses to viruses, and regulation of viral genome replication ( Fig 5B). Approximately 33% of these proteins reduced in 2DG and TNFα co-treated cells included genes within the ontological pathway 'response to cytokine signaling', e.g., IFIs, ISGs, as well as other critical viral defense genes, e.g., MX1, OAS53, SAMHD1, and global innate regulators, e.g., STAT1 (Figs 5C and S5). Additionally, 43% of the substantially down-regulated proteins were known to be glycosylated (Figs 5C and S5). Collectively, these data indicate that restricting glycolysis during TNFα treatment prevents the accumulation of proteins involved in the anti-viral response, a substantial proportion of which are glycosylated.

TNFα treatment induces the accumulation of glycosyl precursors and glycosylation enzymes
Glucose provides the subunits necessary for protein glycosylation, which is critical for the stability of numerous glycoproteins. Given the observation that TNFα activates glycolysis, and that inhibition of glycolysis results in the decreased accumulation of several glycosylated antiviral proteins (Fig 5C) and the apparent reduction of GLUT1 glycosylation (Fig 4F), we sought to explore the impact of TNFα on the labeling of glycosyl precursors such as UDP-Glc and UDP-N-Acetyl Glucosamine (UDP-GlcNAc). The intracellular abundance of UDP-Glc was 3-fold higher in cells treated with TNFα for 24 hours, but UDP-GlcNAc pools remained unchanged ( Fig 6A). To examine the turnover kinetics of these pools, we labeled vehicle or TNFα-treated cells with U-13 C-glucose over time. TNFα treatment induced the accumulation rate of 13 C-UDP-Glc and 13 C -UDP-GlcNAc isotopologues, while increasing the rate of 12 C isotopologue disappearance (Fig 6B and 6C), consistent with the activation of these pathways. Both UDP-Glc and UDP-GlcNAc accumulation depended on glycolysis, as inhibition with 2DG during TNFα treatment blocked the steady-state increases in UDP-Glc pools (Figs 6D and S6) and reduced the accumulation of UDP-GlcNAc (Figs 6E and S6) relative to TNFα treatment alone. Further, the U-13 C-Glc labeling rate of the 13 C-UDP-Glc and 13 C -UDP-Glc-NAc pools was substantially reduced upon co-treatment with TNFα treatment and 2DG ( Fig  6F and 6G). These data support a model wherein TNFα treatment requires glycolysis to induce activation of UDP-sugar metabolism and support protein glycosylation of anti-viral proteins.

Glycosylation is important for the TNFα-induced anti-viral state
To dissect how various metabolic pathways including the hexosamine pathway/glycolysis contribute to TNFα's functional ability to induce the anti-viral cell state, we pretreated cells with a variety of inhibitors that attenuate metabolic pathways adjacent to glycolysis in the presence or absence of TNFα (Fig 7A). Two inhibitors of pyrimidine biosynthesis [N-phosphonacetyl-L-aspartate (PALA) [37] and vidoflumidus (Vid) [38]], and two pentose phosphate pathway inhibitors [6-aminonicotinamide (6-AN) [39] and N3-pyridyl thiamine (N3-PT) [40]] failed to rescue viral replication when co-pretreated with TNFα (Fig 7A and 7B). In contrast, co-pretreatment of TNFα with BGNAc largely rescued the ability of HCMV to initiate infection. The magnitude of BGNAc rescue was indistinguishable from the rescue observed with 2DG copretreatment ( Fig 7B). These results support a model in which protein glycosylation plays a critical role in TNFα's ability to induce the anti-viral cell state.

Discussion
TNFα pretreatment promotes an anti-viral cell state that prevents replication of diverse viral families [10][11][12]. However, many questions remain about the functional requirements and . b/c/f/g HFFs treated with vehicle (black), TNFα (10 ng/mL, red), 2DG (20 mM, blue) or TNFα+2DG (purple) for 19 hours in media containing unlabeled 12 C-glucose prior to media change with media containing U-13 C-glucose, without treatments. Cellular extracts were harvested at t = 0, 1, 2 and 5 hr post-label addition. c/g UDP-Glc or d/h UDP-GlcNAc intracellular isotopologue abundances were quantified by LC-MS/MS (mean ± SD, n = 3). Solid lines represent intracellular abundances of unlabeled 12 C metabolite species, dashed lines represent intracellular abundances of 13 C-labeled metabolite species. M refers to the 12 C unlabeled species and M+n represents detection of a 13 C-labeled species where n represents the number of additional mass units detected by mass spec. f/g represent only M+6 isotopologue abundance, refer to S6 Fig for corresponding M abundances. a/b/e/f FDR-adjusted p-values were determined using 2-way ANOVA followed by two-stage step-up method of Benjamini, Krieger and Yekutieli; ns = not significant, � p<0.05, �� p<0.01, ��� p<0.001. https://doi.org/10.1371/journal.ppat.1010722.g006

PLOS PATHOGENS
TNF and metabolic remodeling cellular activities responsible for limiting viral infection. In this study, we found that TNFα treatment induces significant metabolic changes, specifically activating glycolysis and inducing UDP-sugar metabolism. Further, glycolysis was necessary for the accumulation of several glycosylated anti-viral proteins. TNFα treatment also increased the accumulation of glycosyltransferases such as ALG5 and MGAT1. Inhibiting glycolysis or glycosylation during TNFα pretreatment resulted in the loss of TNFα's ability to attenuate viral replication. Together, these data suggest a mechanism wherein TNFα activates glycolysis and glycosylation as essential components of instituting an anti-viral state.

TNF and metabolic remodeling
A key aspect of TNFα-induced glycolytic activation was the induction of HK2 expression, which was subsequently found to be important for TNFα-mediated glycolytic activation (Fig 4). Previously, TNFα has been reported to induce HK2 in a murine skeletal muscle cell line [41], but little is known about how HK2 contributes to TNFα-associated activities. HK2 is canonically thought of as the muscle-predominant hexokinase isoform, although its expression has also been found to be induced in various tumor types [42]. In tumors, it is thought to be important for cancer-cell survival through modulation of metabolic-associated stress responses [43]. While our results indicate that TNFα induces HK2 expression, which we found to be important for TNFα-induced glycolytic activation (Fig 4J), its deletion did not impact the TNFα-mediated induction of the UDP-glucose pool (Fig 4K), nor did it restore viral infection in the face of TNFα treatment (Fig 4L). This suggests that while induction of HK2 contributes to maximal TNFα-induced glycolysis, it is not essential for all asptects of TNF-induced metabolic alteration nor is it critical for TNFα-mediated inhibition of HCMV plaque formation. However, it was notable that inactivation of HK2 significantly increased HCMV-mediated plaque formation in the absence of TNFα treatment (Fig 4L), suggesting that HK2 could be playing a role in intrinsic anti-viral defense. HK2 has previously been shown to modulate mitochondrial metabolism and cell death pathways [44,45], and these aspects of HK2 metabolism could be playing important roles. Future work should explore these possibilities, as well as the possibility that HK2 induction is contributing to other aspects of TNFα-modulated innate immunity, for example, processing or presentation of viral antigens.
Hif1α is a central regulator of glycolytic gene expression and was implicated in TNFαmediated induction of HK2 in murine skeletal muscle cells [41], raising the possibility that Hif1α is important for TNFα-mediated induction of glycolysis. Our data suggest that Hif1α is dispensable for TNFα-mediated induction of lactate secretion and inhibition of HCMV replication (Fig 4J and 4L). Separately, our data show that cells lacking Hif1α are more permissive to infection (Fig 4L), suggesting Hif1α may play a role in limiting the initiation of infection. Consistent with this finding, Hif1α was recently shown to attenuate HCMV replication in human fibroblasts [46]. The anti-viral phenotype associated with Hif1α-mediated metabolic regulation is intriguing, yet the mechanisms responsible for TNFα-induced glycolysis still require elucidation.
Our data show that TNFα induces changes to several metabolic pools. Our efforts focused primarily on glycolysis, as a number of metabolites upregulated by TNFα were involved in glucose metabolism. However, the metabolite most strongly induced by TNFα treatment was kynurenine (Fig 1C). Kynurenine is a tryptophan-related metabolite that can also be considered part of the NAD + biosynthetic pathway. Kynurenine accumulates in a variety of inflammatory conditions and is induced during Human Immunodeficiency Virus (HIV) infection [47,48]. Others have described pro-viral effects of kynurenine on HCMV replication [46], suggesting a potentially complex relationship with infection that requires further analysis with respect to its roles in inflammation, intrinsic immunity and during viral infection.
In addition to activation of glycolysis, we find that TNFα induces the accumulation UDP--Glucose (UDP-Glc) and stimulates the glucose-mediated labeling of both UDP-Glc and UDP--N-Acetyl-Glucosamine (UDP-GlcNAc) (Fig 6). These nucleotide sugars are molecular substrates for glycosylation reactions that are critical for a variety of cellular activities including the stability of many proteins [49,50], as well as immune cell differentiation and activation [51,52]. Our results indicate that glycolytic inhibition blocks the TNFα-mediated increases in UDP-Glc pools as well as the increases in UDP-Glc and UDP-GlcNAc labeling (Fig 6). Further, 2DG treatment blocked the TNFα-mediated accumulation of GLUT1 glycoforms without impacting the GLUT1 RNA levels or the total amount of GLUT1 protein (Fig 4). Collectively, these data suggest that TNFα induces UDP-sugar metabolism to support its anti-viral activity.
Consistent with this possibility, inhibition of O-linked glycosylation phenocopied glycolytic inhibition in that it prevented TNFα-mediated inhibition of HCMV infection, highlighting the importance of glycosylation to TNFα's anti-viral activity (Fig 7). A number of glycosylated anti-viral effector proteins failed to accumulate upon TNFα treatment in the face of glycolytic inhibition, including MX2, BST2, OSA3 and STAT1, as did various glycosylated components of the adaptive immune response including B2M and HLA-A/B/C/F (Fig 5). These data are consistent with a model in which TNFα treatment drives UDP-sugar production to support the glycosylation and stable expression of anti-viral effector proteins.
That the expression of so many diverse anti-viral effector proteins was impacted by glycolytic inhibition in the face of TNFα treatment likely explains our findings that glycolysis is necessary to attenuate both HCMV and coronavirus replication. HCMV and the coronaviruses tested, OC43 and SARS-CoV-2, possess extremely different viral life cycles, e.g., nuclear DNA replication versus cytoplasmic RNA replication, but the anti-viral effectors downregulated by glycolytic inhibition identified in this manuscript target multiple aspects of various viral infections. These intrinsic anti-viral proteins included MX2, which can block nuclear capsid transport [53]; OAS3, whose activity can result in viral RNA degradation [54]; and BST2, which can tether viruses to membranes for degradation [55,56]. While the mechanisms through which TNFα-induced anti-viral activities rely on specific metabolic activities remain to be elucidated, collectively, our data indicate that TNFα-induced metabolic activities are broadly important for its anti-viral action.
Viruses usurp cellular metabolic resources for their replication (reviewed in [57]). As an example, HCMV induces glycolysis and respiration to support infection [30,[58][59][60][61][62][63]. Similarly, SARS-CoV-2 induces glycolysis, which supports infection [64]. Whereas here, we find that glycolytic activation is important for a cytokine-induced anti-viral state. This raises the question as to whether the TNFα-induced glycolytic program is significantly different from virallyinduced glycolytic programs. Potential differences could be mediated via differential macromolecular assembly of metabolic enzymes that have been found to funnel metabolic molecular flux to diverse downstream enzymes and ultimately different metabolic fates [65][66][67]. Further, despite similar activity increases, differential localization of metabolic enzymes can have a substantial impact on cellular physiology. As mentioned above, HK2 can be post-translationally regulated to localize to mitochondria and regulate mitochondrial metabolism and cell death pathways [44,45]. These observations in other systems suggest that substantial differences in macromolecular assembly and localization could mediate potential differences between antiviral and pro-viral glycolytic activation. In this regard, similar mechanisms could enable antiviral metabolic reprogramming to limit viral access to cellular metabolic resources and thereby limit various points in the viral life cycle, e.g. gene expression or genome replication. Future work in this area can begin to address these questions.
A number of non-glycosylated anti-viral proteins failed to accumulate upon TNFα treatment when glycolysis was inhibited (Fig 5), which could reflect a secondary dependence on the accumulation of a crucial glycosylated protein. STAT1, for example, is glycosylated and fails to accumulate upon TNFα treatment in the presence of glycolytic inhibition (Fig 5). Given STAT1's importance in the transcription of numerous anti-viral effectors [68], it would be predicted that its loss would deplete the expression of glycosylated and non-glycosylated anti-viral effector proteins. Potential secondary effects of losing STAT1 notwithstanding, it is still plausible that in addition to providing UDP-sugar subunits to support glycosylation, glycolysis contributes to TNFα's anti-viral activity via other downstream metabolic activities. In this regard, many questions still remain about the molecular fate of glycolytically-derived carbon and how these pathways could contribute to TNFα's anti-viral activity.
Our results indicate that TNFα-induced metabolic remodeling is important for its ability to promote an anti-viral state. However, many questions remain about other potential metabolic requirements for cytokine-induced intrinsic viral defense. What other cytokine-driven metabolic activities are important for limiting viral infection? What other aspects of innate immunity require specific metabolic activities, e.g., antigen processing and presentation? Is there a common anti-viral metabolic or glycosylation program induced by diverse anti-viral cytokines, e.g., TNFα, IFNγ, IFNα, etc. The answers to these questions will likely shape our understanding of an important host pathogen interaction, that is, the metabolic regulation associated with intrinsic immunity in the face of viral infection and the associated contributions to preventing viral pathogenesis.
The WT strain of HCMV used in experiments with MRC5s and HFFs was BADwt, a Bacterial Artificial Chromosome (BAC) clone of AD169. GFP expressing AD169, referred to as WT-GFP, was BADsubUL21.5 [70]. The clinical isolate of HCMV in this study TB40/Ewt-mCherry (TB40-mCherry) was a gift from Dr. Christine O'Connor [71] and was used in all experiments with ARPE19 cells. OC43 and SARS-CoV-2, Isolate Hong Kong/VM20001061/ 2020 (BEI Resources NR-52282) were cultured as previously described [69]. All experiments involving live SARS-CoV-2 were conducted in a biosafety level 3 facility at the University of Rochester using HFF-ACE2 cells. Experiments involving OC43 were performed in wild-type HFF cells, which are permissive to OC43. Viral stocks were propagated previously described [69,71] and titered using a modified Reed & Muench TCID50 calculator from the Lindenbach lab [72].
For all experiments involving viral infection to determine the anti-viral effects of TNFα in the presence of different inhibitors or treatments, cells were pretreated with the indicated reagent(s) in serum-free DMEM supplemented with 1% pen/strep for 24 hours. Following this pretreatment period, treatment-containing medium was removed and replaced with viruscontaining serum-free DMEM supplemented with 1% pen/strep void of any treatment so as to specifically observe the effect of pretreatment on the cell's ability to promote an anti-viral state.

Reagents and preparation of treatments
For all experiments where cells were treated with TNFα and/or inhibitors, cells were grown to confluence. Twenty-four hours prior to infection, the medium was replaced with serum-free DMEM supplemented with 1% PenStrep and appropriate compounds as indicated.

PLOS PATHOGENS
TNF and metabolic remodeling (2DG) was purchased from Millipore-Sigma (#D8375-5G) and prepared to 20 mM in culture medium for experiments. 6-Aminonicotinamide (MedChem Express, Cat# HY-W01034) was suspended in DMSO to 10 mM and prepared to the indicated concentrations in culture medium. N3-pyridyl thiamine (MedChem Express, Cat# HY-16339, 5 mg) was suspended in DMSO to 8.5 mM and prepared to the indicated concentrations in culture medium. Sparfosic Acid (MedChem Express, Cat# HY-112732B) was suspended in H2O to 10 mM and prepared to the indicated concentrations in culture medium. Vidofludimus (Cayman Chemicals, Item#18377 CAS# 717824-30-1) was suspended in DMSO to 10 mM and prepared to the indicated concentrations in culture medium. Benzyl-2-acetamido-2-deoxy-α-D-galactopyranoside (Benzyl-alpha-GalNAc, BGNAc, MedChem Express Cat# HY-129389) suspended in H2O to 25 mM using ultra sonication and prepared to indicated concentration in culture medium. Kifunensine (Kif, Fisher Scientific, Cat #NC1620501) was suspended in DMSO to 10 mM and prepared to the indicated concentration in cell culture medium. All treatment medium were prepared in serum-free DMEM supplemented with 1% PenStrep unless otherwise indicated. Vehicle or control treatment was serum-free DMEM supplemented with 1% PenStrep.

Plaque efficiency assay
For plaque efficiency assay, cells were grown to confluence in 12 well TC-treated Greiner plates (#82050-930) and treated as described above. Treatment medium was removed and the cell monolayer was infected with a low, known number of plaque forming units (PFUs) of AD169-GFP in 500 uL medium overnight. Following this adsorption period, medium was removed and replaced with 2 mL of a standard agarose gel overlay prepared with 2X concentrated DMEM (Invitrogen #12100046) supplemented with 10% FBS and 1% NuSieve Agarose (Lonza, 12001-722). Plaques were allowed to develop for 10 days and the total number of plaques in each well was counted using a fluorescent microscope. For each indicated experimental condition, the total plaque count of each well was normalized to the average plaque count of the control condition.

Quantification of HCMV infectious units (GFP + cells)
HFFs were grown to confluence in 96-well TC-treated plates and then treated for 24 hours in 100 μL indicated treatment medium. Treatment was removed and replaced with GFP-expressing HCMV viral inoculum (MOI = 0.5, 100 μL) in serum-free DMEM, high glucose, no glutamine, no phenol red medium (Invitrogen #31053036) supplemented with 1x Glutamax and 1% Pen-Strep for 24 hours. The medium was removed and replaced with PBS containing 1:1000 diluted Hoechst fluorescent stain (Thermo Fisher #33342). HFFs were then imaged using a Cytation 5 imaging reader (BioTek). Each well was imaged using a 4X magnification objective lens and predefined DAPI channel with an excitation wavelength of 377 nm and emission wavelength of 447 nm for nuclei count, GFP channel with an excitation wavelength of 469 nm and emission wavelength of 525 nm for cells infected with AD169-GFP, or Texas Red channel with an excitation wavelength of 586 and emission wavelength of 647 for cells infected with TB40-mCherry. Gen5 software (BioTek) was used to determine cell number by gating for objects with a minimum intensity of 3000, a size greater than 5 μm and smaller than 100 μm.
To compare multiple models of infection as a result of pretreatments (Fig 3G), percent infection relative to vehicle was calculated individually for each model (HFF infected with AD169, MRC5 infected with AD169, ARPE19 infected with TB40e). For every model, each biological replicate from each condition was divided by the mean vehicle treated condition then multiplied by 100. Normalized values were plotted at mean +/-SD (Fig 3G).

Analysis of total glycosylated proteins
Cells were treated as previously described. Monolayer was washed 1x with cold PBS and cells scraped into cold NP-40 buffer (500 mM NaCl, 100 mM Tris pH 8.0, 1% NP-40, protease inhibitor tablet (Fisher Scientific #A32955)). Lysates were placed at 4˚C on a rotator for 20 minutes and then sonicated using a sonication probe. Protein samples were processed by SDS-PAGE as indicated above and gel stained for total glycosylated protein using Pro-Q Emerald 300 glycoprotein gel and blot stain kit (Fisher Scientifid #P21857) following manufacturers instructions.

CRISPR knockout
CRISPR knockouts (KO) were performed with the Neon Transfection System 10 uL kit (Ther-moFisher #MPK1025). HFFs were grown to 70% confluence and trypsinized using TrypLE Express (Invitrogen #12605010). HFFs were collected via centrifugation at 600 RPM for 5 minutes and resuspended to a concentration of 1.1x10 7 in resuspension buffer R (ThermoFisher #MPK1025). In a separate tube, 60 pmol sgRNA was combined with 20 pmol Cas9 protein (Synthego) in a volume of 3 uL and incubated at room temperature for 15 minutes. To the prepared sgRNA:Cas9 mixture, 9 uL cell solution (2x10 5 HFFs) was added and gently pipetted up and down to mix. A 10 uL Neon pipette tip was used to extract 10 uL of the cell/sgRNA/Cas9 solution which was then electroporated using the Neon Transfection System (voltage: 1650 V, width: 10 ms, pulses: 3) and transferred to a prepared 6-well dish (Greiner #82050-842) containing growth medium. This process was repeated with the same pipette tip and sgRNA/Cas9 solution and transferred to the same dish for 2 transfections per guide into a single well.

Synthego ICE score [77]
Knockout confirmation of HIF1A using Synthego's Inference of CRISPR Edits (ICE). Genomic DNA was extracted from a 10 cm dish of sub-confluent HFFs transfected with HIF1A sgRNA/Cas9 (described above) or non-target guice control (ntg) sgRNA/Cas9 ribonucleoprotein complex using Lucigen quick extract DNA kit (item # QE09050). HIF1A gene locus surrounding the sgRNA target site was amplified from HIF1A edited DNA extract sample and ntg control DNA extract sample with HIF1A primer set 1 (sequences below) using touchdown PCR method. The resulting DNA fragment was PCR purified using Qiagen QIAquick PCR purification kit (item # 28104) and used as the DNA template in a second PCR method using HIF1A primer set 2 (Sequences below). The resulting DNA fragment was purified as described above and submitted to Genewiz for Sanger Sequencing using Hif1α primer set 3 (sequences below). Sequencing data for HIF1A edited and ntg control samples were uploaded to ice. synthego.com for analysis [77]. Primer sequences: HIF1A Set 1: F-GGGAAGGTTTACAG TTCCATGG; R-GTCTTGCTCTGTCATCCAGG. HIF1A Set 2: F-TCCAGGCTTAATCA GTTGGC; R-CTCAGCTCACCACAACATCC. HIF1A Set 3: F-GCAGCCTAGACTTTA TACGAGG; R-ATCTCCTGACCTCAGATGATCC.

Glucose consumption
HFFs were grown to confluence in a 6-well TC-treated plate (Greiner #82050-842) and treated with 1 mL indicated treatment master mix for 24 hours. The medium was harvested from each well and glucose concentration for each medium sample was quantified using the HemoCue Glucose 201 System (HemoCue).
A standard curve was prepared using virgin medium where the highest dilution was 225 mg/dL glucose. This standard was serially diluted in PBS 1:2 five times for the remaining standards. To detect glucose, 8 μL each standard was loaded into a HemoCue glucose microcuvette (Hemocue #10842-830) and inserted into the glucose meter; mg/dL glucose detected was recorded and plot against the known concentration of glucose in the standards. Samples were diluted 1:4 in PBS and loaded into microcuvettes as described for standards.
To calculate glucose consumption, mg/dL glucose was determined for each sample using the standard curve and then multiplied by its dilution. The amount of glucose detected in each sample was subtracted from the amount of glucose in t0 medium to quantify the total nmol glucose consumed over 24 hr. The total amount of glucose consumed (nmol/hrx1e6 cells) was calculated by dividing the calculated nmol glucose consumed for each sample by 24 hr and the average HFF cell count in a 6-well dish (3.1e5 cells).

Lactate secretion
HFFs were grown to confluence in a 12-well TC-treated plate and treated with 500 μL indicated treatment master mix for 24 hours. The medium was harvested from each well and stored at -80C. To measure lactate in the medium, a standard curve was prepared using a 16 mM lactate standard as the most concentrated standard and serially diluting this most concentrated standard 1:2 8x. Medium samples were diluted 1:64 in PBS. Samples and standards were loaded onto the same MS run and the resulting MS intensity data of the standards was used to generate a standard curve. Standards that began to plateau or fell below the limit of detection (3x the value of the average blank) were discarded from the standard curve. Sample MS intensity values that did not fall within the standard curve were not quantified.
Lactate secretion for each sample was calculated by converting the MS intensity value to mM lactate using the standard curve and multiplying by the dilution factor (1:64) to yield lactate concentration (nmol/μL). Total nmol for each sample was calculated by multiplying the lactate concentration by the total sample volume (500 μL). Lactate secretion (nmol/hr � 1e6 cells) was calculated by dividing the total nmol lactate secreted for each sample by 24 hr and the average HFF cell count in a 12 well plate (1.2e5 cells).

Steady state metabolomics analysis
For steady state metabolomics, HFFs were grown to confluence in 10 cm TC-treated dishes (Greiner #82050-916) and placed in serum-free medium supplemented with 10 mM HEPES for 24 hours. HFFs were treated with 7 mL treatment medium supplemented with 10 mM HEPES for 24 hours. The medium was aspirated and metabolites were extracted from cells immediately by adding 3 mL cold 80% methanol (-80C). Extract was centrifuged at 3,000 RPM and supernatant containing metabolites was decanted into a fresh 50 mL conical on dry ice. Residual metabolites were extracted from the cell pellet with 3x washes of 500 uL 80% cold methanol followed by centrifugation at 3,000 RPM. Supernatants were pooled in the appropriate 50 mL conical and the methanol was evaporated under a gentile stream of nitrogen for 6-8 hours. Samples were suspended in 200 uL 80% cold methanol and analyzed by LC-MS/MS as previously described [78]  For metabolomics data analysis, LC-MS/MS intensity values were imported to Metaboanalyst 5.0 [79]. R-history for metaboanalyst can be found in S1 Data. The data were mean-centered and divided by the standard deviation (SD) of each variable. For principal component analysis, each dot represents metabolomics data of one biologically independent sample. Shaded ellipses represent 95% confidence intervals. PC1 and PC2 refer to the amount of total variation observed between samples that can be attributed to segregation along that principal component. Hierarchical clustering maps follow Euclidian distance and Ward cluster algorithm. Metabolite values are depicted as Z-scores from min (blue) to max (red). Metabolites determined to be significantly different between TNFα and vehicle treatments (Fig 1C) were determined in Metaboanalyst. Log2 fold change (FC) values and FDR-adjusted p-values were calculated using Metaboanalyst. The direction of comparison was TNFα+/Vehicle. FC threshold of 1.5 and FDR-adjusted p-value threshold of 0.05 were applied to yield 12 metabolites significantly altered as a result of TNFα treatment.

Metabolic tracer analysis
HFFs were grown to confluence in 10 cm dishes in growth medium. Medium was replaced with serum-free medium containing 10 mM HEPES for 24 hours. Cells were then treated with 7 mL indicated treatment medium made using glucose-free DMEM (Invotrogen #11966025) supplemented with 10 mM HEPES and 12 C-Glucose (4.5 g/L) for 19 hours. At 19 hours, medium was replaced with 7 mL glucose-free DMEM supplemented with 10 mM HEPES containing U-13 C-Glucose (Cambridge Isotopes #CLM-1396-1) (4.5 g/mL). Metabolites were extracted from cells as described above at 19 hours post treatment (t = 0 hours post label), 20 hours post treatment (t = 1 hours post label), 21 hours post treatment (t = 2 hours post label) and 25 hours post treatment (t = 5 hours post label). Samples were prepared and analyzed by LC-MS/MS as described above [78]. To analyze the intracellular abundance of isotopologues as represented in Fig

Proteomics experiment and data analysis
Cells were grown to confluence in 10 cm dishes (VWR, #82050-916) and placed in serum-free medium for 24 hours. Cells were treated in 7 mL indicated treatment medium at t0. At 24 hr post treatment, the monolayer was washed 3x with cold PBS. All subsequent steps were performed on ice and with chilled 4C PBS. Cells were scraped in 3 mL PBS and transferred to 15 mL conical tubes. Each dish was washed with an additional 3 mL of PBS and combined with its respective cell suspension. For each sample, cells were pelleted via centrifugation at 3,000 RPM for 5 minutes. Supernatant was discarded and the pellet was washed with 500 μL cold PBS and pelleted again at 3,000 RPM for 5 minutes. Supernatant was removed and samples were brought to the Mass Spectrometry Resource Laboratory (MSRL) at University of Rochester Medical Center. Protein extraction and S-trap Digest was performed by the MSRL. Samples were run on Oribtrap Fusion Lumos and analyzed using Data-Independent Acquisition (DIA) to yield relative protein abundance for 3,780 proteins, data shown in S5 Table. Data uploaded to MetaboAnalyst 5.0, mean-centered and divided by the standard deviation (SD) of each variable. For principal component analysis (Fig 5A) each dot represents proteomics data of one biologically independent sample. Shaded ellipses represent 95% confidence intervals. PC1 and PC2 refer to the amount of total variation observed between samples that can be attributed to segregation along that principal component.
To observe the differences in abundance of metabolic enzymes between vehicle and TNFα treated cells, a list of genes involved in metabolism was retrieved from the UniProt Database on August 31 st , 2021 to yield 4,432 human genes involved in metabolism (S6 Table). This list of genes was scanned against our proteomics data set of 3,780 proteins using a custom code in Python v 3.7/PyCharm Community 12.1. The resulting list of intersecting genes (562 genes) were uploaded to MetaboAnalyst 5.0. R-histrory for metaboanalyst can be found in S1 Data. Data were mean-centered and divided by the standard deviation (SD) of each variable. For principal component analysis (S1 Fig) each dot represents proteomics data of one biologically independent sample. Shaded ellipses represent 95% confidence intervals. PC1 and PC2 refer to the amount of total variation observed between samples that can be attributed to segregation along that principal component. Hierarchical clustering maps (Fig 4A) follow Euclidian distance and Ward cluster algorithm. Values are depicted as Z-scores from min (blue) to max (red). Proteins determined to be significantly more or less abundance as a result of TNFαtreatment relative to vehicle were determined in Metaboanalyst. Log2 fold change (FC) values and FDR-adjusted p-values were calculated using Metaboanalyst (S6 Table). The direction of comparison was TNFα+/Vehicle. FC threshold of 1.5 and FDR-adjusted p-value threshold of 0.05 were applied to yield 40 proteins significantly altered as a result of TNFα treatment.
For the analysis of a specific protein of interest, normalized protein abundance was calculated by dividing each sample relative protein abundance by the average of the vehicle-treated relative protein abundance (Fig 4C/4G/4M). Statistics were performed in GraphPad Prism v9.1.0.
To generate a list of proteins more abundant in TNFα treatment compared to TNFα+2DG co-treatment, statistical parameters were applied to the data [Log 2 Fold Change (TNFα+2DG/ TNFα) < -0.6, p-value < 0.05] (S7 Table, 239 proteins). The resulting list of proteins which were uploaded to Gene Ontology (GO) Analysis to identify the biological process most impacted, represented by FDR-values (Fig 5B). The list of proteins more abundant in TNFα treatment compared to TNFα+2DG co-treatment was also scanned against the GO-term [GO:0034097] "Involved in Cytokine Signaling" (54 proteins). These genes were scanned against databases of known glycosylated proteins [80,81] using a customized R-script to identify glycosylated proteins within this subset of proteins more abundant in TNFα treatment that are depleted during TNFα+2DG co-treatment (30 proteins glycosylated, 24 proteins not glycosylated). Missing values were replaced with the minimum detected MS intensity detected by that sample (S7 Table, yellow highlighted cells). Relative protein abundance values for each protein was determined by dividing the sample normalized protein abundance by the average relative protein abundance of vehicle-treated for that protein. The top 15 proteins most strongly induced by TNFα treatment, either glycosylated or not glycosylated, were graphed (Fig 5C).
Similarly, a list of proteins more abundant in TNFα+2DG co-treatment compared to TNFα treatment was generated by applying statistical parameters to the original proteomics list of 3,780 proteins [Log 2 Fold Change (TNFα+2DG/TNFα) < 0.6, p-value < 0.05] (S8 Table, 120 proteins). The resulting list of proteins was uploaded to Gene Ontology (GO) Analysis to identify the biological processes most impacted, represented by FDR-values (S4 Fig)

Statistics
All statistical analysis were carried out using GraphPad Prism v9.1.0 unless otherwise indicated. MetaboAnalyst 5.0 was used to perform PCA analysis and hierarchical clustering from raw LC-MS/MS intensity data mean-centered and divided by the standard deviation (SD) of each variable. R-histrory for metaboanalyst can be found in S1 Data.
Supporting information S1 Fig. HFFs treated with vehicle (black) or TNFα (10 ng/mL, red) for 24 hr. Cells were harvested for proteomics analysis (S5 Table). PCA of a subset of proteins from proteomics analysis involved in metabolism; analysis described in materials and methods (S6 Table). Data shared with Fig 4A. (TIF)  Table). Ontology analysis of TNFα-induced proteins that were significantly more abundant upon co-treatment with 2DG. Bar graph represents the FDR-values from the top 15 GO-terms (S8 Table).  Table). Statistical parameters applied to proteomics data, described in materials and methods, to generate a list of proteins significantly induced by TNFα treatment but depleted in cells co-treated with TNFα and 2DG. Protein list submitted for Gene Ontology (GO) analysis ( Fig 6B) and scanned against databases of known glycosylated proteins (described in materials and methods) or 'Response to Cytokine Signaling' [GO:0034097] (S7 Table). Venn Diagram represents overlap of proteins involved in cytokine signaling (yellow) and glycosylated (blue). (TIF) S6 Fig. HFFs treated with vehicle (black), TNFα (10 ng/mL, red), 2DG (20 mM, blue) or TNFα+2DG (purple) for 19 hours in media containing unlabeled 12 C-glucose prior to media change with media containing U-13 C-glucose, but without treatments. Cellular extracts were harvested at t = 0, 1, 2 and 5 hr post-label addition. UDP-Glc or UDP-GlcNAc intracellular isotopologue abundances were quantified by LC-MS/MS (mean ± SD, n = 3). Solid lines represent intracellular abundances of unlabeled 12 C metabolite species, dashed lines represent intracellular abundances of 13 C-labeled metabolite species. M refers to the 12 C unlabeled species and M+n represents detection of a 13 C-labeled species where n represents the number of additional mass units detected by mass spec. (TIF) S1 Table. Cells treated with TNFα (10 ng/mL) or vehicle for 24 hours (n = 3). Metabolites extracted and analyzed by LC-MS/MS. Table represents the LC-MS/MS intensity raw values. Data uploaded to Metaboanalyst v5.0 for analysis as described in materials and methods ( Fig  1A and 1B).  Fig 1C). (XLSX) S3 Table. Cells treated with TNFα (10 ng/mL) or vehicle. Metabolites harvested at t0, 4, 12 and 24 hours post treatment (hpt, n = 3). Table represents Table. Intersection of proteins from proteomics experiment (S5 Table, Veh & TNFα groups) and entries from Uniprot database involved in 'metabolism' (Column A). Table represents Table. Analysis of proteomics data (S5 Table) to identify proteins in the TNFα+2DG treatment group that were significantly more or significantly less abundant relative to TNFα treatment group. See Materials and methods for details. Orange: subset of proteins from S5 Table significantly induced by TNFα treatment relative to vehicle, and depleted during TNFα treatment in the presence of 2DG relative to TNFα treatment alone. Blue: Gene Ontology analysis output of the 239 proteins significantly induced by TNFα treatment relative to vehicle, and depleted during TNFα treatment in the presence of 2DG relative to TNFα treatment alone. Bold ontology terms are represented in Fig 5B. Yellow = Limit of detection (LOD) data imputation values (see materials & methods) Pale yellow: proteins significantly induced by TNFα treatment relative to vehicle, and depleted during TNFα treatment in the presence of 2DG relative to TNFα. Pale green: proteins significantly induced by TNFα treatment relative to vehicle, and depleted during TNFα treatment in the presence of 2DG relative to TNFα treatment alone that are involved in response to cytokine and are known to be glycosylated (Fig 5B and 5C).