Inhibition of pyruvate dehydrogenase kinase redirects NSCLC cell metabolism and counteracts development of resistance to epidermal growth factor receptor tyrosine kinase inhibitors


 Background Non-small cell lung cancer (NSCLC) patients harboring oncogenic mutations in the epidermal growth factor receptor (EGFR) inevitably develop resistance to targeted therapy. Drug resistance is an example of cancer cell plasticity where cells change according to diverse microenvironmental pressure, which can be described as a way of evolution by natural selection. Metabolic rewiring during cancer development may support several malignant features and facilitate emergence of therapy resistant clones. Results Analysis of transcriptome data from two independent NSLSC patient cohorts identified upregulated markers of glucose metabolism and ROS defense in tumors compared to normal tissue. We show that these alterations were associated with increased expression of pyruvate dehydrogenase kinase 1 (PDK1). We established relevant in vitro models to study metabolic alterations in the context of resistance to EGFR TKIs to determine if targeted metabolic intervention would prevent development of resistance to EGFR tyrosine kinase inhibitors (EGFR TKIs) in NSCLC cells. The PDK inhibitor dichloroacetate (DCA) was shown to reduce cell growth. This mechanism was associated with redirected metabolism towards pyruvate and lactate oxidation, and reduced lactate production, both in EGFR TKI sensitive and resistant NSCLC cells. Conclusion We propose that the intracellular stress created by redirecting pyruvate metabolism can prevent EGFR TKI resistance in NSCLC.

or in combination therapy. Altered metabolic signatures during cancer development was documented as early as 1924 when Otto Warburg described the preference for cancer cells to use glycolysis for ATP production even under aerobic conditions. This phenomenon is referred to as the Warburg effect or aerobic glycolysis and is part of the metabolic reprogramming known to occur in many cancers through various interactions with oncogenes [35][36][37][38][39][40]. This metabolic rearrangement does not only affect ATP generation, but also supports the biosynthesis of cellular building blocks such as carbohydrates, lipids, proteins and nucleic acids, which are crucial in proliferating cancer cells, [38,39,41]. Changes in mitochondrial oxidative phosphorylation (OXPHOS) are often reported in cancer cells, and OXPHOS related mutations are associated with tumorigenesis [28,31,42,43]. To accommodate the increased glucose dependence, increased expression of several transporter proteins occurs in various cancer types, such as glucose transporter proteins (GLUTs) [44][45][46][47]. In addition, changed regulation of monocarboxylate transporters (MCTs) exporting cellular lactate, and mitochondrial pyruvate carriers (MPCs) facilitating mitochondrial pyruvate uptake, have been reported both in cancer and associated stroma [48][49][50]. MPC dysregulation is found to be involved in EMT, resistance to ionizing therapy and metabolic reprogramming [50,51]. Further, lactate dehydrogenase (LDH), which is responsible for the conversion of pyruvate to lactate, and regenerates oxidized NAD + , is found to be upregulated in tumorigenesis [52,53].
A key regulator of cell metabolism is hypoxia inducible factor 1 alpha (HIF1a). This transcription factor is activated due to hypoxia, which induces expression of multiple genes involved in angiogenesis and glucose metabolism [54][55][56]. HIF1a is also known to induce EMT through activation of various EMT related genes [28,[56][57][58]. Cellular responses associated with hypoxia can sometimes be initiated by HIF1a under normoxic conditions, this is true for cancers harboring VHL-mutations, and interestingly a pseudohypoxic state has also been shown to be caused by dysregulation of OXPHOS, or mutations in TCA-related enzymes [28,42,59,60]. One of the direct targets of the transcription factor HIF1a is pyruvate dehydrogenase kinase 1 (PDK1), whose function is to phosphorylate and inhibit the pyruvate dehydrogenase (PDH) enzyme complex [61]. Both PDK1 and PDK3 has been shown to have HIF responsible element (HRE) in their promoter, which allow HIF1a to bind with a high a nity [62]. PDK2, however has a low a nity for HIF1a, and PDK4 has no a nity for HIF1a, however, PDK4 has been shown to enhanced due to high fat diets and diabetes [61][62][63][64][65][66]. The PDH complex catalyzes acetyl-CoA formation from the pyruvate produced by glycolysis. The acetyl-CoA then enters the TCA-cycle, which generates electron donors (NADH and succinate) as fuel for mitochondrial OXPHOS. Inhibitory phosphorylation of PDH by the PDKs leads to increased ux of pyruvate to lactate at the expense of mitochondrial pyruvate oxidation [67,68]. [69,70]. Dichloroacetate (DCA) is a natural occurring pyruvate analog, which acts as a PDK inhibitor, inducing increased mitochondrial pyruvate oxidation [69,70]. By inhibiting PDK, PDH is maintained in a catalytically active state, and the pyruvate pool available from glycolysis is decreased, thus production of lactate and acidi cation of the microenvironment is limited [71]. Several studies have shown that DCA is both inducing apoptosis and sensitizes cancer cells to ionizing therapy, due to hyperpolarized mitochondria [72]. Previous ndings implicate deregulation of PDH and the pyruvate oxidation pathway in mechanisms of tumor initiation and development of cancer [68,[73][74][75].
Increased glycolysis at the expense of mitochondrial respiration, a characteristic feature of aerobic glycolysis, may protect against the production of reactive oxygen species (ROS), as they are by-products of mitochondrial electron transport. Excessive ROS may cause oxidative damage in the cells, and lead to mutations and apoptosis [31,[76][77][78]. It has been suggested that reduced rates of mitochondrial respiration serves a protective function against ROS-induced cell damage, in cancer models of therapy resistance [31,76]. In addition, glucose metabolism generates NADPH through the pentose phosphate pathway, which is important for the antioxidant defense system [40,79]. EGFR TKIs such as ge tinib are known to induce ROS production and mitochondrial dysfunction in NSCLC [29,80], and it is therefore possible that development of resistance involves increased ROS tolerance [30-33, 68, 73-75]. We have investigated the potential of preventing development of TKI resistance in NSCLC in vitro by manipulation of metabolism. By manipulating the pyruvate metabolism through supply of DCA, we show that the cells are forced to oxidize pyruvate. As a result, lactate production decreased drastically. We propose that the cellular stress resulting from this metabolic rerouting inhibits carcinoma cells to develop resistance to additional therapy, including EGFR TKIs.

Results
Increased PDK1 expression is associated with altered expression of genes involved in energy metabolism and ROS protection in NSCLC tumors.
We investigated PDK1 expression levels and correlated the data to a panel of genes involved in energy metabolism and ROS protection in two independent cohort datasets from lung tumors and normal tissue samples. Transcriptomic data was obtained from the lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUSC) from the cancer genome atlas (TCGA LUAD & LUSC, RNA sequencing, n = 1016; adjacent normal, n = 110), and the GSE18842 microarray dataset with paired samples from normal tissue (n = 44) and lung adenocarcinoma and squamous-cell carcinoma (n = 46). In the TCGA data, the mutation status of EGFR is available, whereas in the GSE18842 data, paired data from normal-and cancer tissue from the same patient is available [81]. The mRNA expression of PDK1 was signi cantly elevated in cancer tissue compared to normal tissue in both cohorts. In the TCGA EGFR wild-type samples (EGFRwt, n = 591), the upregulation was higher, compared to the group with EGFR mutations (EGFRmut, n = 76) (Fig. 1A), and in the GSE18842 cohort representing paired samples, the upregulation was prominent (Fig. 1B). To evaluate effects on glucose catabolism in tumors relative to normal tissue, we investigated expression levels of 22 related genes, both regulators and enzymes. The majority of these genes showed signi cant upregulation in cancer tissue compared to normal tissue (Fig. 1C). In particular, the expression of glucose transporter GLUT1 was increased, consistent with increased glucose uptake during tumor development.
Further, multiple glycolytic enzymes were increased, including PKM2 (only available data from the TCGA cohort), G6PD, PFKP, ALDOA, GAPDH, ENO1 and LDHA. For most genes, the ndings were consistent in the two cohorts. These effects were generally similar in EGFR wild type (wt) compared EGFR mutated (mut) tumors; however, some genes displayed difference, such as the lactate importer, MCT1, which had lower expression in EGFRmut compared to EGFRwt tumors. Correlation analysis between PDK1 and the panel of genes involved in glucose metabolism revealed signi cant positive associations for a majority of genes (Rho values > + 0.3) (Fig. 1D). This indicates that PDK1 is co-regulated with the glycolytic machinery at the transcriptional level. Also here, the EGFRwt and EGFRmut groups showed similar results.
The EGFRmut group tended to display lower Rho values, which may be biologically relevant, though it could also result from the smaller sample size for this group. Interestingly, the oncogene c-MYC, known to interact with HIF1a to enhance glycolysis in cancer cells [82], was correlated to the expression of PDK1 in our data. However, only a trend was observed in EGFRmut and the GSE18842 cohort (Supplementary gure S1A). Also, NRAS has been shown to be a regulator of HIF1a and glycolysis [83], and this oncogene is also correlated to the expression of PDK1 in our data (Supplementary gure S1B). We found the majority of genes involved in gluconeogenesis and fatty acid oxidation (FAO) to be downregulated in tumor samples compared to normal tissue samples in both cohorts ( Fig. 1E and 1F). Figure 1E shows data for selected genes encoding key enzymes and regulators of gluconeogenesis [84]. The level of fructose bisphosphatase 1 (FBP1), a key regulator of gluconeogenesis activity was particularly decreased in tumors relative to normal tissue, and interestingly, as seen in Fig. 1D, FBP1 was strongly inversely correlated to the expression of PDK1. The correlation was more prominent in the EGFRwt group compared to the EGFRmut group, which may indicate that the metabolic changes during tumor development are different in the two groups. Genes involved in FAO was generally downregulated in NSCLC compared to normal tissue (Fig. 1F). ACOX4, CYP4A11, PPARG and PDK4 levels indicated that oxidation FAO was downregulated upon NSCLC development. Interestingly, PDK1 was inversely correlated to PDK4 (Supplementary gure S1C), suggesting distinct roles of the two enzyme family members.
Genes re ecting mitochondrial biogenesis and biomass displayed different types of responses (Fig. 1G). Interestingly, PPARGC1α (PGC1α), which is a key regulator of mitochondrial biogenesis, was strongly downregulated in the tumor samples. This was particularly evident in the EGFRmut group. In contrast, genes that are commonly used as markers of mitochondrial biomass, including TFAM, VDAC, TOM20 and ATPB, were mildly upregulated. Genes involved in mitophagy and autophagy also showed heterogeneous responses. PINK1 and PARK2, which are known modulators of mitophagy were signi cantly downregulated in the TCGA cohort. BNIP3 was upregulated, indicating an elevated level of apoptosis in cancer tissues. BNIP3 was also correlated to the level of PDK1 in both cohorts, and even to a higher degree in the EGFRmut group (Supplementary gure S1A). Altogether, we found no indication of loss of mitochondrial biomass, but the effects on regulators of mitochondrial biogenesis and dynamics may suggest that processes of mitochondrial quality control are not operating normally.
Expression of several genes encoding antioxidant enzymes were upregulated in NSCLC compared to normal tissue (Fig. 1H). Three genes of the glutathione peroxidase (GPX) family were induced, particularly GPX2, and to a lesser extent for GPX7 and 8. GPX3 was signi cantly downregulated. The thioredoxins TXN and TXNRD1 were upregulated in the total TCGA cohort and in the EGFRwt group, however less in the EGFRmut group, suggesting differential regulation in cancer cells harboring EGFRwt compared to EGFRmut group. For the peroxidoxin family, 4 out of 6 genes were upregulated in the two cohorts. Further, the tumor mRNA expression of glutathione S-reductase (GSR) and superoxide dismutase 1 (SOD1) showed a trend of increase, whereas catalase (CAT) was signi cantly downregulated. The TCGA data showed mildly reduced expression of the mitochondrial SOD2 in the tumors. Interestingly, the mRNA level of multiple antioxidant genes correlated with PDK1 level (Fig. 1I). For this set of genes, PDK1 predominantly displayed signi cant or trending positive associations, such as for GPX2, GPX7 and TCNRD1. However, there were also negative associations, such as for GPX3 and TXBRD2. These data support the notion that regulation of PDK1 may be associated with contextual mechanisms of ROS defense.
DCA in combination with EGFR TKIs inhibit cell growth in NSCLC cell models.
Based on the nding that NSCLC tumors had upregulated PDK1, we investigated effects of the PDK inhibitor DCA on cells, either alone or in combination with EGFR TKI. We used HCC827 and HCC4006, two NSCLC cell lines which are sensitive to the rst-generation EGFR TKI, erlotinib. We also use H1975 which is resistant to erlotinib, but sensitive to the third generation EGFR TKI rociletinib. Different dosage combinations of DCA and respective EGFR TKIs were tested. Periodic imaging microscopy during a treatment period of 14 days showed a time-dependent inhibition of proliferation and increased cell death in HCC827 cultures treated with both erlotinib and DCA, with apparently only dead cells remaining at the end of the treatment period ( Fig. 2A). Assessment of proliferation was based on image analysis quantifying percent con uency (using the Incucyte software), and statistical analysis after 72 h showed a signi cant, but mild potentiated effect of the erlotinib + 20 mM DCA treatment compared to the agents alone ( Fig. 2B). In cultures only treated with one of the agents, there were also reduced cell number, but the majority of these cells appeared intact and viable. A separate experiment was performed to test different dosage combinations of the agents measuring con uence and resazurin viability ( Fig. 2C and Supplementary gure S2). The three different cell models displayed a similar pattern after 72 h treatment; there was an added antiproliferative effect of 20 or 25 mM DCA in combination with erlotinib/rociletinib. A concentration of 1 µM was required to give a signi cant effect of the EGFR TKIs, but for two of the cell types (HCC827 and H1975) the added effect of DCA was also evident at this concentration of the EGFR TKIs. These data demonstrate that DCA in NSCLC cell inhibits cell proliferation, and further suggest that the added effect in combination with EGFR TKIs may serve to counteract development of cellular EGFR TKI resistance.
Acute effects of DCA on NSCLC cell metabolism.
Treatment with DCA is expected to activate mitochondrial pyruvate oxidation, by releasing PDK-mediated inhibition of the PDH enzyme complex. To investigate how this acutely affects mitochondrial respiration and glycolysis in cultured NSCLC cells, we measured oxygen consumption rate (OCR) and extracellular acidi cation rate (ECAR). Addition of DCA, alone or in combination with EGFR TKI, gave a small acute increase in OCR in HCC827 cells, but not in the two other cell types (H1975 and HCC4006) ( Fig. 3A and 3B, and Supplementary Fig. 3A-D). The minor effect of DCA on OCR may suggest that an internal switch in fueling pathways for oxidative metabolism occurs without major effect on the respiratory rate under these conditions. In contrast, ECAR acutely decreased upon DCA addition in all three cell lines. It was con rmed that the addition of the DCA solution (pH 7.4) did not change the pH of the medium itself, and therefore the most plausible explanation is that DCA caused decreased cellular lactate secretion ( Fig. 3C and 3D). The associated decrease in ECAR supports a metabolic shift with enhanced propensity for mitochondrial oxidation at the expense of lactate production. Of note, these effects of DCA were generally unaffected by the presence of EGFR TKI. Next, to con rm if DCA treatment caused increased mitochondrial oxidation of pyruvate, as well as its precursors lactate and glucose, we incubated cells with 14 C-labeled versions of these substrates, and measured production of 14 CO 2 . Figure 3E positions in which pathway steps the detected 14 CO 2 is produced, depending on 14 C-labeled substrates used in the central energy pathway. The data clearly demonstrated increased pyruvate oxidation after treatment with DCA in all three cell lines (Fig. 3F). DCA caused signi cantly increased rate of lactate oxidation in HCC4006 and H1975, whereas it was decreased in HCC827 (Fig. 3G). All cell lines exhibited signi cant or trending decrease in glucose oxidation upon DCA treatment (Fig. 3H). In support of decreased cellular lactate production, we found mRNA levels of LDHA to be decreased in the three cell lines after treatment with 20 mM DCA for 48 h. Further, the mRNA level of PGC1α was strongly upregulated in DCA treated cells, supporting metabolic adaptations involving mitochondria (Fig. 3I). These functional and regulatory data con rm that DCA promotes mitochondrial pyruvate oxidation in NSCLC cells, and thereby antagonizes several metabolic signatures of cancer cell proliferation and therapy resistance. Noteworthy, these changes in cell metabolism had minor effects on mitochondrial respiratory rate.
Development of resistance to EGFR TKI is associated with EMT and altered metabolic signature. To model the development of erlotinib resistance in NSCLC cells, we treated HCC827 and HCC4006 cells with 1 µM erlotinib for more than 4 weeks, and established sublines that survived (Supplementary gure S4A).
The resistant sublines are referred to as HCC827/BERL and HCC4006/BERL. In addition, we used an established H1975 cell line, with a subline (H1975/COR1-1) resistant to the third generation EGFR TKI rociletinib (CO-1686), provided by Clovis oncology [85,86]. The H1975 cell line is known to be resistant towards erlotinib as they have a T790M mutation in EGFR exon 20, but sensitive towards rociletinib. The resistance phenotype of the three cell models was con rmed by drug response experiments measuring con uence after 72 h treatment with their corresponding EGFR TKI ( Fig. 4A-4C). We further aimed to study the clonogenic potential upon treating cells with rociletinib and erlotinib. As expected, clonogenic potential was absent in the HCC827 and HCC4006 EGFR TKI sensitive cell lines treated with either erlotinib or rociletinib, while H1975 cells were resistant against erlotinib but sensitive to rociletinib (Fig. 4D). All resistant cell lines were resistant to both rociletinib and erlotinib (Fig. 4D). By measuring mRNA ( Fig. 4E-G) and using microscopy and immunocytochemistry, typical features of EMT were documented in EGFR TKI resistant cells, including mesenchymal morphology, low E-cadherin levels and high vimentin levels ( Fig. 4H -J). The effects on E-cadherin and vimentin protein levels were supported by western blot analysis, and c-MET mRNA was reduced (Supplementary gure S4 B-C). This supports previous ndings for the H1975 model [85,86]. The mRNA expression of PDH subunits and the PDKs were evaluated in the resistant compared to the parental cells (Fig. 4K). EGFR TKI resistance was not associated with consistent effects on PDHA and PDHB mRNA levels. PDK1, PDK2 and PDK4 was upregulated in both HCC827/BERL and HCC4006/BERL compared to the parental cells. In H1975/COR1-1 cells compared to the parental, PDK2 was upregulated, whereas PDK3 and PDK4 were downregulated.
Upregulation of PDK1 protein expression in EGFR TKI resistant cells was con rmed in the HCC827 and HCC4006 cell models, and this was consistent with increased phosphorylation of PDH E1-alpha subunit (Fig. 4M). When measuring mRNA of a selection of genes associated with glucose metabolism, we found strong upregulation of one or more MCT and GLUT mRNA in accordance with the patient samples in Mitochondrial respiration and glycolysis in EGFR TKI resistant NSCLC cells.
To determine if development of EGFR TKI resistance affects mitochondrial function in NSCLC cells, we measured respiratory rates following sequential additions of speci c modulators. For two of the models (HCC827, H1975) the resistant cells had signi cantly lower rates of mitochondrial respiration compared to the respective parental cells ( Fig. 5A-E, Supplementary gure S5). This effect was found both regarding basal respiratory activity, and uncoupled maximal respiratory capacity. Further, as the leak respiration upon oligomycin addition was normal or low, the integrity of the inner mitochondrial membrane is intact. In the HCC4006 model, there was an increase in basal respiration, whereas maximal respiratory capacity was reduced, in the resistant cells compared to the sensitive control cells. Characterization of the glycolytic function in EGFR TKI resistant cells was performed in a separate experiment (Fig. 5F-5L). Noteworthy, endogenous ECAR re ects glycolytic activity supported by cellular glucose stores, and other cellular processes in uencing extracellular acidi cation, and these contributes only to a minor part compared to activity induced by addition of extracellular glucose (Supplementary Figure S5F) [87]. Resistant cells of the HCC827 and HCC4006 models had lower basal ECAR after adding glucose, and maximal ECAR in presence oligomycin, compared to the parental cells ( Fig. 5F-I). HCC827/BERL had similar ECAR levels compared to the parental cells. The glucose induced ECAR varied between the three cell lines; however, the spare glycolytic capacity was slightly reduced in resistant cell lines. To sum-up, the resistant cells of the HCC827 and H1975 models, exhibited low rates of mitochondrial respiration, while glycolysis was similar to the parental cells. In contrast, the resistant HCC4006/BERL cells had relatively normal rates of mitochondrial respiration, but low glycolytic rate, compared to the parental cells. To further investigate the metabolic changes upon drug resistance, we measured oxidation rates of pyruvate, lactate and glucose ( Fig. 5J-L). The HCC827/BERL cells had signi cantly lower oxidation of all the three substrates, compared to the parental cells. H1975/COR1-1 only showed reduced glucose oxidation, and HCC4006/BERL had increased pyruvate and lactate oxidation. In summary, it is interesting to see that there was generally an inverse relationship between effects of resistance on basal respiration and glycolysis ECA rates for the three models. The rate of oxygen consumption is further re ected in the substrate oxidation measurements, where HCC827/BERL is oxidizing less pyruvate, lactate and glucose, whereas the HCC4006/BERL which has an increased oxygen consumption rate upon development is oxidizing more of the three substrates compared to parental cells. The changes in substrate oxidation for the H1975/COR1-1 cells are however more modest compared to parental cells, also re ecting the general rate of oxidition, con rming the consistency of metabolic alterations for the three NSCLC cell models.

DCA inhibits cell proliferation in NSCLC cells resistant to EGFR TKI therapy.
Based on our previous ndings, we investigated how DCA treatment in uences metabolism in the resistant clones. In combination with 1 µM EGFR TKI, DCA demonstrated a dose-dependent effect in the EGFR TKI resistant cells (Fig. 6A). An experiment to evaluate colony formation capacity of the EGFR TKI resistant cells showed that upon treating cells with 10 mM DCA, the clonogenic potential was decreased. When combining 10 mM DCA with 1 µM EGFR TKI, the clonogenic potential was even lower in HCC4006/BERL, and at both 20 mM DCA, and 1 µM EGFR TKI combined with 20 mM, DCA the clonogenic potential was totally absent for all subtypes (Fig. 6B). Upon investigating the metabolic effects of DCA ( Fig. 6C-D and Supplementary gure S6), the HCC827/BERL cells showed a signi cant increase in OCR acutely after addition of DCA. H1975/COR1-1 displayed a minor increase in OCR after DCA addition, whereas no effect was seen in HCC4006/BERL cells. However, exposure to DCA reduced ECAR in all the three resistant cell lines ( Fig. 6D and Supplementary gure S6).
Furthermore, the three resistant cell lines showed increased pyruvate and lactate oxidation after treatment with 20 mM DCA for 24 h, accompanied by a trending or signi cant decrease in glucose oxidation ( Fig. 6E-G). These data suggest that even though the three cell lines show different metabolic phenotypes, all increase pyruvate and lactate oxidation upon DCA treatment. DCA also reduced their glycolytic rate, and reduced cell viability upon treatment.
Radiation increases the effect of DCA and EGFR TKI in NSCLC cells.
Our ndings so far suggest that DCA causes a metabolic shift in NSCLC cells, and that this may increase the sensitivity to EGFR TKI treatment. One possible theory is that enforced activation of the pyruvate oxidation axis results in metabolic stress, including increased ROS production. Thus, we investigated if the effects of DCA and erlotinib could be enhanced in combination with ionizing radiation. NSCLC cells were treated with ionizing radiation, in clinically relevant doses, minutes before drugs were added, and the proliferation monitored by periodic microscopy. Furthermore, 8 Gy radiation added to the effect of DCA and erlotinib, both in parental cells and resistant cells (Fig. 7A-B). The proliferation curves showed that the inhibited proliferation rate was more pronounced in cells sensitive to EGFR TKI (Fig. 7C). At 72 h post treatment (Fig. 7D), the cell growth was reduced signi cantly in the combined DCA and erlotinib treated cells compared to control. However not statistically signi cant, there was a trend suggesting that ionizing radiation added to the therapeutic effect of DCA and EGFR.

Discussion
In the present study, PDK1 expression was found to be increased in patients with NSCLC compared to normal tissue. Further, PDK1 expression was correlated with increased expression of genes related to glucose metabolism, including GLUT1 and LDHA. In addition, several genes encoding antioxidants were upregulated and correlated to PDK1 expression, indicating that the redox status in the cells are disturbed during development of NSCLC. We show that manipulation of pyruvate metabolism using the PDK inhibitor DCA inhibits the development of resistance to EGFR TKIs in NSCLC cell models. DCA was shown to induce metabolic stress by increasing oxygen consumption and severely reduce glycolysis in the cancer cells. Indeed, a glycolytic phenotype confers resistance to apoptosis [88]. Apoptosis is thereby induced by mechanisms that does not involve drug targeted mutations that impair receptor tyrosine kinase ATP-binding site mutations such as T790M. By rewiring energy metabolism and thus inducing cellular stress levels in NSCLC cells, we propose we can restrict cellular plasticity and cancer evolution.
We propose that these alterations make cancer cells particularly vulnerable to targeted metabolic manipulations, which can enhance the effect of conventional cancer therapy. DCA is a pyruvate analogue which is known to increase lactate and pyruvate oxidation, and to reduce blood glucose levels [89][90][91].
DCA has primarily been used in patients with lactic acidosis in clinical studies. Wells  We found that 20 mM DCA alone or in combination with EGFR TKIs greatly decreased survival in three different NSCLC cell models (Fig. 2) Further, as seen in the drop in extracellular acidi cation rates and the substrate oxidation assay, DCA severely inhibits glycolytic activity and reduce glucose oxidation (Fig. 3).
The bioenergetic difference between cancer and normal cells may offer selective therapeutic targets, as dependence on glycolysis and lactic fermentation is only found in skeletal muscle cells during strenuous exercise. Side effects from continuous use of DCA has been few, but include reversible peripheral neuropathy [70]. However this effect seems to be limited to patients with mitochondrial diseases [93].
It has been shown that expression of EGFR is associated with increased glucose uptake and glycolysis [94]. Upon adding DCA to EGFR TKIs, the inhibition of glycolysis is fatal for the developing tumor. Interestingly, a recently published study shows that DCA also increases the expression of the key tumor suppressor p53 in colorectal cancer, highlighting new possible anticancer mechanisms induced by DCA [95].
Transportation of glucose across the cell membrane is dependent on GLUTs. We found GLUT1 to be upregulated in the patient cohorts. We also saw an increase in GLUT2, but not as prominent. Both GLUT1 and GLUT2 overexpression has been linked to poor patient outcome [44,96,97]. Interestingly, we found a reduction in GLUT3, as opposed to the ndings in other studies where they have shown upregulation of GLUT3 and its link to poor overall survival [44,96]. Still, research by Younes in 1997 on 289 lung adenocarcinoma samples suggested that GLUT3 overexpression is reported to be mostly found in combination with GLUT1 overexpression as only 21% were positive for both GLUT1 and GLUT3 [98].
Interestingly, in addition to other genes involved in glycolysis, we also found mRNA of GAPDH, frequently used as a reference gene, to be upregulated in tumor samples, indicating that it is not suited as a reference gene for NSCLC. This nding is in concordance with previously published results [99,100].
Supporting the upregulation of glycolysis related gene panel, we also saw a reduction in genes related to gluconeogenesis and mitochondrial biogenesis, with FBP1and PGC1α being the most deregulated genes. Decreased gluconeogenesis is a phenomenon often related to increased glycolytic activity and loss of a key gluconeogenesis regulator FBP1 has previously been shown to induce glycolysis, and lead to hyperglycemia and lactic acidosis [26,101,102]. FBP1 downregulation has also been found to reduce ROS production and to be associated with EMT [26,103,104]. Further, in line with downregulated gluconeogenesis, we nd a panel of genes involved in fatty acid oxidation (FAO) to be decreased in tumor samples compared to NSCLC tissue. One of the genes related to FAO is the PDK4, another member of PDK family. Interestingly, PDK4 is decreased in the tumor samples, and is inversely correlated to PDK1, indicating diverse roles for the two enzymes. We have recently shown that PDK4 expression is a sensitive marker for FAO [65], whereas the role of PDK1, which is a direct target of HIF1α, is correlated to increased glycolysis [61].
As known, most chemotherapies and ionizing radiation increase cellular stress, in parts by stimulation of ROS production. ROS is produced as byproducts in the electron transport chain in the mitochondria, and as DCA stimulates pyruvate and lactate oxidation, we hypothesized that ionizing therapy could add to the effect of DCA and DCA in combination with erlotinib. In this study, we also found that many genes involved in the antioxidant defense system were correlated to the expression of PDK1. This included genes encoding glutathione peroxidases (GPX), as well as thioredoxin-and peroxiredoxin-related genes.
Interestingly, and in accordance t our ndings, reduced expression of GPX3 has been shown to be linked to increased proliferation rate in NSCLC by modulating redox mediated signals [105]. Further, GPX2 is associated with invasion, metastasis and tumor growth, as GPX2 knock down reduced levels of EMT markers such as vimentin, SNAI1 and MMP2 [106].
As development of EGFR TKI resistance remains an important contributing factor to the high mortality rate in NSCLC, resistant cells were developed in vitro to assay the change in metabolic phenotype upon EGFR TKI resistance in three distinct cell lines. These included HCC827 which harbors the EGFR activating deletion mutation E746 -A750 and HCC4006 harboring the L747 -E749 deletion as well as the point mutation A750. H1975 isolated from a primary NSCLC tumor was included. With the additional mutation in the ATP and drug binding domain of EGFR, H1975 cells are resistant to rst generation EGFR inhibitors, and remains sensitive to the third generation EGFR TKI, including rociletinib EGFR TKI resistance was associated with EMT in all sublines [23,24]. Ampli cation of c-MET is recognized as a contributor to EGFR TKI resistance; however, this was not seen in our models. We found reduced glycolytic capacity in the erlotinib resistant cell lines and reduced spare glycolytic capacity in all sublines (Fig. 5L). Although the NSCLC cell lines had different metabolic phenotypes, DCA treatment instantly reduced glycolysis in all of them ( Fig. 2 and Supplementary gure S6), in accordance with [107,108], and increased pyruvate and lactate oxidation in all the EGFR TKI resistant cells ( Fig. 3 and Fig. 6). Although the resistant clones had a more modest response to DCA inhibition, we still found reduced cell growth in these cells. This included both DCA treatment as monotherapy, as well as combination with EGFR TKIs.
This indicates that treatment of cancer which already have developed resistance to EGFR TKI may bene t from DCA treatment. In addition, this effect can be more prominent with the addition of radiation therapy, as shown in Fig. 7.
In the patient cohorts we found PGC1α to be decreased in tumor tissue compared to normal tissue.
Interestingly, upon development of EGFR TKI resistance, we observed an increase in PGC1α in all three sublines (Fig. 4). This is in line with the ndings of Andrzejewski et al., who showed that PGC1α upregulation mediated resistance to metformin in lung cancer cells [109]. PGC1α is a co-transcription factor primarily linked to mitochondrial biogenesis. It has also been shown to be in indirect transcriptional interaction with FOXO1, an important regulator of insulin stimulated gluconeogenesis [110]. In the patient cohorts, the expression of FOXO1 was also decreased in the tumor tissue. For future studies, it would be interesting to investigate further the relation between FOXO1 and PGC1α during development of EGFR TKI resistance.
To avoid intracellular acidi cation and apoptosis as a consequence of increased glycolysis, glycolytic cells must sustain both intracellular and microenvironmental lactate homeostasis. Several transporters are involved in this process, including monocarboxylate transporters (MCT) 1-4 [48,49,111]. We found MCT1 to be upregulated in cancer tissue compared to normal tissue in the patient cohorts, and increased expression of MCT1 and MCT2 in our cell models upon drug resistance (Fig. 4), indicating altered lactate transport pattern. MCT1 and MCT2 have both been found in the plasma membrane and peroxisomal membranes whereas MCT2 is additionally reported to be expressed in the mitochondrial membrane where it is involved in the import of pyruvate following lactate oxidation [48,112]. Mitochondrial pyruvate carriers (MPCs) are located on the mitochondrial inner membrane and are essential for import of pyruvate. We found the mitochondrial pyruvate carrier MPC1 to be downregulated and MPC2 to be upregulated in resistant NSCLC cells. As MPC1 is upregulated, pyruvate shuttling into the mitochondria may be reduced, and lactic fermentation of pyruvate can thus be increased. In accordance with our ndings, MPC1 suppression has previously been found to induce EMT, resistance toward ionizing therapy and to maintain aerobic glycolysis [50,51]. Further, LDHA is increased in patient tumor samples as compared to normal tissue (Fig. 1). Upon lactate fermentation, oxidized NAD + is produced in addition to lactate. NAD + is essential for the conversion of glyceraldehyde-3-phosphate (GAP) to 1,3bisphosphoglycerate (1,3-BPG) by GAPDH in glycolysis [55], providing another rationale for the cancer cell to keep high fermentative activity.
In agreement with a recent tandem mass spectrometry study performed by Zhang and colleagues, we nd that DCA does not increase glucose metabolism, but rather affects pyruvate and lactate metabolism directly [113]. On the level of mRNA, we nd LDHA to be signi cantly downregulated upon 48 h of DCA treatment, further indicating decrease in fermentation. It is known that an acidic microenvironment is inhibiting the innate immune system in patients [114][115][116]. Interestingly, as lactate oxidation increases and lactic fermentation decreases as a consequence of DCA treatment, it represents indications of DCA is a de-toxifying stimulator of the microenvironment allowing both the innate immune system activation as well as the activation of the adaptive immune system. Reducing lactate secretion through inhibition of LDHA has been shown to increase the effect of the immunotherapy including anti PD-L1 through increased activation of in ammatory and anti-tumors responses [114], and increase T-memory cell activation [115,116].
As presented in this study, we nd that by inducing targeted metabolic stress to NSCLC EGFR addicted cells, we prevent development of resistance towards EGFR TKIs. In parallel to the presented experiments, we kept cells in culture asks for more than four weeks to verify that no cells survived and retained proliferation potential upon combining DCA with EGFR TKI´s. Additional in vivo experiments could aid to address the issue of prevention versus delay the onset of resistance. However, when cells have already developed resistance to EGFR TKIs, we experience more modest effect of reversing the sensitivity to EGFR TKIs by co-treating with DCA, although the clonogenic potential is signi cantly reduced, both when treating with 20 mM DCA alone and in combination with EGFR TKI. The therapeutic effect of DCA on EGFR TKI resistant cells is increased upon radiation treatment, however not signi cantly. We propose that manipulation of the pyruvate metabolism is creating intracellular stress leaving the cancer cells vulnerable to additional therapy. As we see a better response in parental NSCLC cells, we suggest that the therapeutic timing may be of importance.

Conclusion
In conclusion, based on our ndings in the NSCLC cell models and the two independent patient cohorts, targeting cancer evolution through DCA treatment or otherwise limiting metabolic plasticity, represents a promising therapeutic option. DCA treatment may modulate the increased PDK1 expression and increased glycolytic addiction found in the NSCLC patient cohorts. Thus, based on our ndings we propose that targeting PDK in combination with EGFR TKI, and possible also other relevant therapies, will increase the therapeutic e cacy and prolong overall survival of patients suffering from NSCLC. Our study supports further exploring PDK targeting in combination with EGFR TKI, or other relevant therapies, as promising options targeting the underlying mechanisms of cancer evolution.

Expression analysis and correlation analysis
We investigated expression levels and association between PDK1 and genes involved in glucose metabolism, gluconeogenesis, fatty acid oxidation and ROS defense. This was done by analyzing RNA sequencing cohort from lung adenocarcinoma obtained from the cancer genome association (LUAD-TCGA, n = 1016), normal lung tissue obtained from TCGA (n = 110) and microarray data obtained from Drug response curves and cell proliferation Cells were plated at a density of 3000 cells/well in 96-well Nunclon Delta Surface plates (Thermo Scienti c). Following overnight incubation, cells were treated with either DCA, erlotinib, rociletinib or DCA in combination with erlotinib or rociletinib. After radiation and/or after treatment the 96w plates were placed in a Incucyte® Live Cell Analysis system (Incucyte) and percent con uency was estimated using the Incucyte Zoom software and normalized to the rst time point of each cell type or DMSO control as described in gure legends. The resazurin cell viability assay was used to measure cell viability by using uorescence to detect cellular metabolic activity. After indicated treatment period (72 h) resazurin was diluted in medium (10% v/v) and added to the cells. The cells were further incubated for 3 hours before uorescence was measured using a Tecan SPARK instrument.
Clonogenic assay 2500 (HCC4006 and H1975) or 7500 (HCC827) cells were plated in 6w plates (Nunc, Thermo sher) and treated alone or in combination with 20 mM DCA and 1 µM erlotinib or 1 µM rociletinib as indicated on the gure. After 14 days, they were xed with a mixture of 6% (v/v) glutaraldehyde and 0.5% (v/v) crystal violet. Cells were imaged with a EPSON Perfection V850 Pro scanner (Epson, Suwa, Japan).

Western blotting
Cells for protein extraction was grown in T75 asks, washed with cold PBS, scraped, pelleted and frozen in -80 o C, before they were lysed using RIPA lysis buffer. 20 µg protein was loaded into 6-12% SDS-PAGE gels from Bio-Rad Laboratories (Hercules, CA). Blotting was performed by using the Bio Rad Trans-Blot Turbo system and blocking was performed for 2 h in 5% milk in TBS-T (0.1% (v/v) Tween). Primary antibodies used were PDK1 (ab207450, Abcam, Cambridge, United Kingdom), PDK4 (ab110336, Abcam) and p-PDH (ab92696, Abcam), E-Cadherin (CDH1, 14472S, Cell Signaling), vimentin (VIM, ab92547, Abcam), FBP1 (72736, Cell signaling), LDHA (3582, Cell signaling). The housekeeping protein a-tubulin or vinculin was used as loading control when total protein staining was not available. Primary antibodies were removed, and the membranes were washed three times for 5 min with TBS-T before incubating them with secondary antibody for 1 h under agitation. The membranes were washed three times with TBS-T before they were examined.

Oxygen consumption
The Seahorse XFe96 Analyzer (Agilent, Santa Clara, CA) was used to measure oxygen consumption rate (OCR) and extracellular acidi cation rate (ECAR). The mitostress and glycostress tests were performed as described in [25,65,119]. Cell number and concentration of compounds was optimized before running the assay. All compounds were from Sigma if otherwise not stated. Throughout the mitostress experiments, we used 3 µM oligomycin, 0.5 µM CCCP (HCC827) or 1.5 µM CCCP (HCC4006 and H1975), 1 µM rotenone and 1 µM antimycin A.
Upon investigating the acute effects of DCA, the assay medium was D5030 supplemented with 10 mM glucose, 2 mM pyruvate and 4 mM glutamine at pH 7.4 (+/-0.05). The pH of the DCA addition was carefully adjusted prior to loading. 20 mM DCA was added to port A, oligomycin in B, CCCP in port C and rotenone and antimycin A in D. Data was rox (antimycin A, non-mitochondrial respiration) corrected.
When measuring glycolysis, the assay medium contained 4 mM glutamine and pH 7.35 (+/-0.05). 10 mM glucose was added in port A, oligomycin port B, CCCP port C and 100 mM 2DG in port D. Data was normalized by using the Pierce BCA Protein Assay Kit (Thermo Fisher) and calculated by using the associated BSA standard curve.

CO 2 -trapping
Substrate oxidation rate of the NSCLC cells was based on [120] and performed as described in [65] without modi cation. The assay medium contained Dulbecco's phosphate-buffered saline (with MgCl2 and CaCl2, #D8662) with 10 µM added fatty acid free BSA as assay medium. The substrates used were For untreated cells, 30 000 cells were plated overnight in a 96w Corning CellBIND Surface Polystyrene plate (Corning NY). For the DCA treated cells, 10 000 (HCC827) and 7500 (HCC4006 and H1975) cells were plated and incubated overnight, before 20 mM DCA or control medium was added to the cells and left for 24 h before trapping. On the day of trapping, the 14 C-labeled medium was added to the cells before a lter plate immersed with 1 M NaOH and a silicon gasket were placed on top and clamped in a sandwich. The "sandwich" was then incubated for 4 hours. After incubation cells were washed twice with PBS before they were lysed with 0.1 M NaOH. Protein was measured using the Pierce BCA Protein Assay Kit as described by manufacturer. 30 µL MicroScint scintillation liquid (Perkin Elmer) was added to each well in the lter plate and incubated for one day before reading the accumulated 14 CO 2 trapped in the lters using top read on the MicroBeta Microplate Counter. All numbers were subtracted from background medium control levels. If substrate levels were negative (below background) they were set to 0.

Radiation
The radiation was performed at Haukeland University Hospital by a Varian Triology (Energy 6 MV, 600 MU / min ~1 .3 Gy / min). Cells were plated in 96w plates (cell proliferation) and incubated over night; cell number was optimized to treatment length. The cells were exposed to 0 Gray or 8 Gray, with gel-covers mimicking the density in the normal human thorax. After radiation the cells were treated alone or in combination of 20 mM DCA and 1 µM erlotinib.

Statistics and gures
Data were analyzed using Graphpad Prism 8 software (Graph-Pad Software; San Diego, CA, US) and gures made by using Adobe Illustrator (San Jose, CA). ANOVA and unpaired two-tailed student's t-test were used to evaluate statistical differences between the samples. Rho values > 0.3 and < -0.3 were considered signi cant. P < .05 was considered statistically signi cant.

Figure 3
Acute effects of DCA on NSCLC cell metabolism. lactate oxidation, and H) glucose oxidation. The data in these experiments represent fold change pooled from two experiments (n = 8-10, data represented as mean +/-SEM). I) Representative relative mRNA expression of LDHA, LDHB and PGC1α from one of three experiments in HCC827, H1975 and HCC4006 cells treated with 20 mM DCA for 48 h. Statistical analysis was performed by two-way ANOVA with Tukey post hoc test for A-D, and an unpaired two-tailed t-test for F-I (* P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001).

Figure 6
Co-treatment with DCA is re-sensitizing NSCLC cells to EGFR TKI therapy. The effect of 1 µM EGFR TKI (erlotinib for HCC827 and HCC4006, and rociletinib for H1975) alone or in combination with DCA on cell proliferation was evaluated in resistant cells compared to the respective parental cells in A). In B) the colony formation capacity of the resistant cell lines is shown upon treatment with 10 mM DCA, 10 mM DCA with 1 µM EGFR TKI, 20 mM DCA, and 1 µM EGFR TKI combined with 20 mM DCA as indicated on the gure. Real-time OCR and ECAR 54 min after addition of DCA to HCC827/BERL, H1975/COR1-1 and HCC4006/BERL was measured. The OCR is shown in C) Representative data of OCR from HCC827, H1975 and HCC4006 at 54 min after addition of drug. Whereas ECAR data is shown in D. All data was adjusted for non-mitochondrial activity and is displayed as percent of initial (third measurement) prior to addition of control medium, 20 mM DCA or 1 µM erlotinib or 20 mM with 1 µM erlotinib. HCC827/BERL, H1975/COR1-1 and HCC4006/BERL cells were treated with 20 mM DCA for 24 h before they were incubated with 14C-labeled substrates for 4 h. Data are displayed as fold change relative to DMSO control as a mean of n = 6 -10 with data from two representative experiments. D) 14C-labeled pyruvate, E) 14C-labeled lactate and F) C14 labelled glucose. Statistical analysis was performed by two-way ANOVA, with Tukey post hoc test in A-D, and an unpaired two-tailed t-test for D-F ( * P < 0.05, ** P < 0.01, *** P < 0.001, **** P < 0.0001).

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. 20200513Suppltable1.xlsx