Ref-1 redox activity alters cancer cell metabolism in pancreatic cancer: exploiting this novel finding as a potential target

Pancreatic cancer is a complex disease with a desmoplastic stroma, extreme hypoxia, and inherent resistance to therapy. Understanding the signaling and adaptive response of such an aggressive cancer is key to making advances in therapeutic efficacy. Redox factor-1 (Ref-1), a redox signaling protein, regulates the conversion of several transcription factors (TFs), including HIF-1α, STAT3 and NFκB from an oxidized to reduced state leading to enhancement of their DNA binding. In our previously published work, knockdown of Ref-1 under normoxia resulted in altered gene expression patterns on pathways including EIF2, protein kinase A, and mTOR. In this study, single cell RNA sequencing (scRNA-seq) and proteomics were used to explore the effects of Ref-1 on metabolic pathways under hypoxia. scRNA-seq comparing pancreatic cancer cells expressing less than 20% of the Ref-1 protein was analyzed using left truncated mixture Gaussian model and validated using proteomics and qRT-PCR. The identified Ref-1’s role in mitochondrial function was confirmed using mitochondrial function assays, qRT-PCR, western blotting and NADP assay. Further, the effect of Ref-1 redox function inhibition against pancreatic cancer metabolism was assayed using 3D co-culture in vitro and xenograft studies in vivo. Distinct transcriptional variation in central metabolism, cell cycle, apoptosis, immune response, and genes downstream of a series of signaling pathways and transcriptional regulatory factors were identified in Ref-1 knockdown vs Scrambled control from the scRNA-seq data. Mitochondrial DEG subsets downregulated with Ref-1 knockdown were significantly reduced following Ref-1 redox inhibition and more dramatically in combination with Devimistat in vitro. Mitochondrial function assays demonstrated that Ref-1 knockdown and Ref-1 redox signaling inhibition decreased utilization of TCA cycle substrates and slowed the growth of pancreatic cancer co-culture spheroids. In Ref-1 knockdown cells, a higher flux rate of NADP + consuming reactions was observed suggesting the less availability of NADP + and a higher level of oxidative stress in these cells. In vivo xenograft studies demonstrated that tumor reduction was potent with Ref-1 redox inhibitor similar to Devimistat. Ref-1 redox signaling inhibition conclusively alters cancer cell metabolism by causing TCA cycle dysfunction while also reducing the pancreatic tumor growth in vitro as well as in vivo.

pathways and transcriptional regulatory factors. In addition to the expected decrease in glycolysis, dramatic decreases in gene expression in TCA cycle and complexes within the electron transport chain were also observed. Using Ref-1 siRNA and speci c Ref-1 redox inhibitors, we validated these ndings using qRT-PCR on a panel of mitochondrial differentially expressed genes (DEGs) and quantitation of mitochondrial substrates. We also extended these studies to cancer-associated broblasts (CAFs) which constitute a crucial component of the PDAC TME. In order to ascertain the clinical relevance of our ndings, we performed in vivo studies using mice with tumor cells co-injected with CAFs treated with Ref-1 inhibitor, APX2009 or the comparator compound, Devimistat, also known as CPI-613. APX2009 is a second-generation Ref-1 redox signaling inhibitor. [33][34][35][36] Devimistat is a rst-in-class dehydrogenase inhibitor targeting the TCA cycle. [37][38][39] In summary, our ndings demonstrate a newly discovered role for Ref-1 in TCA cycle and expression of genes within the mitochondrial complexes even under hypoxia as well as con rm that Ref-1 redox function has a role in regulating PDAC hypoxia signaling pathways as HIF regulated genes and glycolysis were also affected. Additionally, we identi ed a potential PDAC therapeutic approach using APX2009 and Devimistat in multiple 3D co-culture models containing both pancreatic cancer cells as well as CAFs which was also validated in vivo.

Cell culture
Pa03C, Panc10.05, and CAF19 cells were obtained from Dr. Anirban Maitra at The Johns Hopkins University 40 and were then transduced with TdTomato for tumor cells or EGFP for CAFs as previously described. 41 CAF02-hTERT cells were isolated using the outgrowth method as previously described. 42,43 Cells were maintained at 37 °C in 5% CO2 and grown in DMEM (Invitrogen; Carlsbad, CA) with 10% Serum (Hyclone; Logan, UT), or under hypoxic conditions of 1% O 2 / 5% CO 2 using a Ruskinn Invivo 2 200 hypoxia work station. Cell line identity was con rmed by DNA ngerprint analysis (IDEXX BioResearch, Columbia, MO) for species and baseline short-tandem repeat analysis testing. Cell lines were 100% human and a nine-marker short tandem repeat analysis exists on le. They were also con rmed to be mycoplasma free.
Single cell RNA sequencing (scRNA-seq) scRNA-seq was performed as previously reported. 25 Libraries were prepared by The Purdue Genomics Facility (Purdue University, West Lafayette, IN) using a Nextera kit (Illumina, San Diego, CA, USA). Unstranded 2 × 100 bp reads were sequenced using the HiSeq2500 (Illumina, San Diego, CA, USA) on rapid run mode in one lane.

scRNA-seq Data Analysis
FastQC was applied to evaluate the quality of the single cell RNA sequencing data. Counts were called for each cell sample by using STAR alignment pipeline against human GRCh38 reference genome. Cells with less than 250 or more than 10,000 non-zero expressed genes were excluded from the analysis. Cells with more than 15% counts mapped to the mitochondrial genome were excluded as low quality cells, resulting in 40  Unsupervised cell clustering was performed using Seurat v3.0 with the variedly expressed genes identi ed by default parameters. Cell clusters were annotated by the experimental condition and their uniquely highly expressed genes. Differentially expressed genes and gene expression states were identi ed by using the left truncated mixture Gaussian model based test, with FDR < 0.05 as the signi cant cutoff. 44 Gene co-regulation modules were identi ed by using our inhouse developed Boolean matrix decomposition method MEBF. 45 Pathway enrichment of the differentially expressed genes and the genes in each co-regulation module were analyzed by a hypergeometric test against the canonical gene sets and transcriptional regulatory factor targets retrieved from MsigDB v6 46  Detailed method can be found in supplementary information and was adapted from literature reports 47,48 and vendor provided protocols. Brie y, cells were lysed using urea lysis buffer. Protein isolated and trypsin/Lys-C digested for peptides. Peptides were fractionated using Pierce™ High pH reversed-phase peptide fractionation spin columns and then subjected to Nano-LC-MS/MS analysis.
Pa03C cells (2 × 10 5 ) were plated into 6-well plates and allowed to attach overnight. The following day, APX3330, APX2009, or Devimistat in low serum (5%) DMEM media were added to the wells. DMSO was used as the vehicle control. Cells were treated for 24 hours, after which they were collected for analysis.

Mitochondrial Function Assay
S-1 Mitoplates (Biolog, Hayward, CA) were used to investigate mitochondrial function. Assays were performed as per manufacturer's protocol. Brie y, plates were activated by adding the Assay Mix to the wells to dissolve the substrates, for at least 60 minutes at 37 °C. Following siRef-1 transfection or drug treatment, cells were collected, counted, resuspended in provided buffer and plated at 5 × 10 4 cells/well. This resuspension was added to the plate, which was immediately read at 590 nm kinetically at 5 min intervals for 4 hours at 37 °C. Data was analyzed using Graphpad Prism 8, and statistical signi cance was determined using the 2-way ANOVA and p-values < 0.05 were considered statistically signi cant.

Western Blot Analysis
For whole cell lysates, cells were harvested, lysed in RIPA buffer (Santa Cruz Biotechnology, Santa Cruz, CA), and protein was quanti ed and electrophoresed. Immunoblotting was performed using the following antibodies: Ref- qRT-PCR qRT-PCR was used to measure the mRNA expression levels of the various genes identi ed from the scRNA-seq analysis. Following transfection, total RNA was extracted from cells using the Qiagen RNeasy Mini kit (Qiagen, Valencia, CA) according to the manufacturer's instructions. First-strand cDNA was obtained from RNA using random hexamers and MultiScribe reverse transcriptase (Applied Biosystems, Foster City, CA). Quantitative PCR was performed using SYBR Green Real Time PCR master mix (Applied Biosystems, Foster City, CA) in a CFX96 Real Time detection system (Bio-Rad, Hercules, CA). The relative quantitative mRNA level was determined using the comparative Ct method using actin as the reference gene. The primers used for qRT-PCR are detailed in Supplement Table 2. Experiments were performed in at least triplicate for each sample. Statistical analysis performed using the 2 −ΔΔ C T method and analysis of covariance (ANCOVA) models, as previously published. 23 In Vivo Studies All animal studies were conducted under the guidelines of the National Institutes of Health and were approved by the Institutional Animal Care and Use Committee of Indiana University School of Medicine. NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ (NOD/SCIDγ(-/-))mice (or other strains) were obtained from the In Vivo Therapeutics Core of the Indiana University Simon Cancer Center. Animals were maintained under pathogen-free conditions and a 12 h light-dark cycle. NOD scid gamma (NOD.Cg-Prkdc scid Il2rg tm1Wjl /SzJ) or NSG mice were subcutaneously implanted with 2.5 × 10 6 Pa03C cells or with 5 × 10 6 + 15 × 10 6 Panc10.05 + CAF19 cells in the hind ank using a 200 µl volume of 50:50 solution of Matrigel:DMEM medium. When tumor volumes reached ~ 100 mm 3 , the mice were randomized into 4 groups of 8-9 mice before commencing treatment (~ 11 days for Pa03C and ~ 24 days for Panc10.05 + CAF19 post implantation). The treatment regimen consisted of oral administration of either vehicle (Propylene Glycol, Kolliphor HS15, Tween 80 (PKT) as previously reported 34 ) or 35 mg/kg APX2009 or 50 mg/kg Devimistat dosed twice a day for 15 (for Pa03C xenografts) or 20 (for Panc10.05 + CAF19 xenografts) days. Tumor volumes were measured twice a week and mice were weighed once a week. Data was analyzed using Graphpad Prism 8. Statistical signi cance was determined using the one-way ANOVA and p-values < 0.05 were considered statistically signi cant.

3D Co-Culture Assays
Ultra-low attachment 96-well plates (Corning Inc., Life Sciences) were used to generate 3-dimensional tumor spheroids in the presence of CAFs as reported in the Results and as described previously. 41,54,55 Following plating, cells were treated on Days 4, 7, and 10 with media containing 5% serum, 3% growth factor reduced Matrigel, and inhibitors as indicated. On Days 4, 7, 10, and 14, spheroids were analyzed using Thermo ArrayScan high-content imaging system. 56 Images of 3D structures were captured by ArrayScan using a 2.5x objective for TdTomato and EGFP; then 2D projections were processed to quantify differences in total intensity of both CAFs and tumor.

Interstitial tumor-microenvironment-on-chip (iT-MOC) Assay
The interstitial tumor-microenvironment-on-chip (iT-MOC) is a 3D in vitro micro uidic platform having two layers of microchannels interfaced with a porous membrane in between. Details of fabrication and preparation were described previously. [57][58][59] Brie y, pancreatic cancer cells (Panc10.05) and cancerassociated broblasts (CAF19) were mixed at a 1:1 cell ratio into the cell-collagen mixture. The initial cell concentration was 2 × 10 6 cells/mL for each cell type. After loading, the devices were incubated at 37 °C for 1 hr for collagen gelation. Then, the culture medium was perfused by pressurizing the interstitial channel.
In order to assess drug e cacy, the PDAC iT-MOC was cultured for 48 hrs before treatment. On day 2, drugs were perfused through the capillary channel. The drug solutions were prepared as 0 µM (control), 30 µM of APX2009, 25 µM of Devimistat, and the combination of APX2009 (30 µM) and Devimistat (25 µM) in the culture medium. On day 5, the channels were washed with drug-free medium and reperfused with the same drug-containing medium. On day 8, the iT-MOC platforms were washed and cultured in normal culture medium for 24 hrs before the viability assay. Drug e cacy was analyzed in two ways, cell growth and cell survival as detailed in the supplementary methods.

Statistics
All the experiments were performed at least three independent times. The data obtained were expressed as 'Mean + Standard Error'. Signi cance was calculated as per either 2-way ANOVA or unpaired t-test wherever applicable using Graph Pad Prism Version 8. For iT-MOC assays, Drug e cacy differences of each drug control were statistically analyzed by Tukey post hoc multiple comparison test provided in ANOVA. The difference was considered statistically signi cant when p-value < 0.05. For qRT-PCR, analysis of covariance models was performed to test the Ct difference of each target gene value between treatment with APX2009, Devimistat (CPI-613), and vehicle (DMSO) or siRNA and scrambled control after standardization by reference gene (RPL6/Actin) using ANCOVA as previously described 23 . A p-value of at least < 0.05 was considered statistically signi cant.
To test the tumor growth rate following treatment (i.e., the regression slope for a particular treatment) and differences in tumor growth rates between treatments (i.e., the difference in regression slopes between two treatments) in the in vivo tumor model, mixed effect repeated measure regression models with random intercept were used. 60 Tumor weights over time were estimated and compared between treatments from the regression models. To be considered statistically signi cant, a p-value of at least 0.05 was used. All statistical analysis was conducted using SAS 9.4 (SAS, Inc., Cary, NC, 2016). collected under normoxia from our previous work. In total, 18,204 genes with a signi cant none zero expression were detected in this data set. It is noteworthy the major goal of this study was not for cell type identi cation. Instead, we focused on identi cation of the distinct functional changes that are due to the inhibition of Ref-1 and/or perturbed oxygen level, i.e. biologically explainable gene sets that show distinct expression variation in a subset of the sample. Our previous studies demonstrated these cell numbers provide enough statistical power for the analysis of differentially expressed genes and functional modules. 44,63 The expression level of Ref-1 (APEX1) in the single cells demonstrated the gene was successfully knocked down in the siRef-1 group (Fig. 1G). Unsupervised cell clustering analysis was rst conducted on the cells under all conditions (See Methods). Six cell clusters were identi ed, which were all highly associated with the experimental groups (Fig. 1A, B). Genes overly expressed in each cluster were identi ed by using left truncated mixture Gaussian model. The cell clusters were further annotated by the experimental condition and their speci cally expressed genes and pathways. Two clusters within the hypoxia Scr control cells were identi ed. Pathway enrichment analysis of the marker genes of each cluster suggested one group has high expression of HIF-1α-regulated genes while the other has speci cally elevated lactate production. Within the siRef-1 cells under hypoxia, two clusters were identi ed; one cluster corresponded to consistently down-regulated genes that were downstream of HIF-1α which would be expected based on Ref-1 redox regulation of HIF-1α. The other cluster corresponded to genes related to upregulated translation (Fig. 1B). Figure 1C illustrates the heatmap of the top variably expressed genes in all the tumor cells. Distinct clusters of the cells associated with different conditions were observed. Dendrogram derived by a hierarchical clustering analysis demonstrated dramatic changes in gene expression caused by hypoxia.

Transcriptomic variation caused by Ref-1 inhibition under hypoxia
DEGs between siRef-1 and Scr control were identi ed using left truncated mixture Gaussian model. 44  proliferate much slower compared to Scr control and the cell cycle genes that were upregulated were mainly nucleoporins and some proteosome genes, but not cyclins. The downregulated genes mainly (p < 0.001) enrich in central metabolic pathways (glycolysis, TCA cycle, pentose phosphate, oxidative phosphorylation, amino acids, protein and lipid metabolism), with HIF-1α and PDGF signaling pathways also affected (Fig. 1D). For comparison under normoxia, there were also more genes downregulated than upregulated identi ed in siRef-1 vs Scr control (447 vs. 120, FDR < 0.05). Pathway enrichment analysis revealed the upregulated genes signi cantly (p < 0.001) enrich to transcription, HIV infection and immune response pathways, while the downregulated genes signi cantly (p < 0.001) enrich to central metabolic pathways (glycolysis, pentose phosphate, oxidative phosphorylation), immune response (antigen presentation, T cell receptor and IL6 signaling), and multiple signaling pathways (p53, MAPK, MET and HIF-1α). We also integrated the scRNA data analysis with the proteomic analysis and determined that the differentially expressed genes and pathways identi ed from the scRNA-seq data are highly consistent to the signi cant proteins observed in the proteomics data, especially for the upregulated cell cycle (nucleoporins) and transcription pathways and downregulated metabolic, apoptosis and signaling pathways under hypoxia condition (Fig. 1D,F). Figure 1G shows selected genes associated with each condition and speci c cell clusters within the siRef-1 hypoxia group. Complete lists of the differentially expressed pathways were provided in Supplmentary Table S4. Interestingly, the top downregulated pathway under hypoxia was the TCA cycle with glycolysis and OXPHOS also in the top ve.
One of Ref-1's major functions is its redox activity in which Ref-1 converts an oxidized transcription factor into a reduced transcription factor which leads to an increase in its DNA binding and functional activity. We quantitated the activity of various transcription factors after Ref-1 knockdown by identifying the gene co-regulation modules that are associated with different experimental conditions, i.e. hypoxia. 44,64 Speci cally, gene-wise expression states were rst inferred by the left truncated mixture Gaussian model.
Modules of genes that show consistent activated or suppressed expression in a subset of cells were identi ed using a non-negative matrix factorization method namely MEBF. 45 We further evaluated the enrichment of the genes in each module against known targets of transcriptional regulatory factors, and the association of the cells of each module with the experimental conditions. Supplementary Table S5 lists the predicted transcriptional regulatory factors of each identi ed module (see details in Methods). Metabolic shifts following treatment with Ref-1 inhibitor observed from scRNA-seq data Noting the scRNA-seq and proteomics data consistently revealed downregulated central metabolism pathways in siRef-1 vs Scr control under both hypoxia and normoxia, we speci cally focused on characterizing the gene expression alterations in central metabolic pathways. These pathways included glycolysis, pentose phosphate pathway, lactate production, TCA cycle, oxidative phosphorylation, glutaminolysis, and other amino acid metabolism that can fuel the production of key metabolites succinate, fumarate, malate and oxaloacetate in the TCA cycle.
Signi cant downregulation of lactate dehydrogenase A (LDHA) was also observed. The enzymes asparagine synthetase (ASNS), argininosuccinate lyase (ASL), and adenylosuccinate lyase (ADSL) involved in the metabolism of aspartate to fumarate and oxaloacetate were also signi cantly downregulated. Under normoxia, the glycolysis (p = 7.5e-7) and oxidative phosphorylation (p = 6.2e-4) pathways were also signi cantly enriched in downregulated genes in siRef-1 vs Scr control but were not as dramatically reduced compared to hypoxia ( Fig. 2A, B). Although the TCA cycle, lactate production, and aspartate metabolic genes are downregulated under normoxia, the effects are more robust under hypoxia, and no signi cant changes in genes involved in glutaminolysis were observed when Ref-1 levels are reduced under either hypoxia or normoxia.
With the reduction in expression of genes within glycolysis, TCA cycle, and oxidative phosphorylation pathways, genes involved in the mitochondrial respiratory complexes I-IV as well as ATP synthases were also investigated. Figure 2C and 2D show the dramatic downregulation of genes in the mitochondrial respiratory complexes I-IV and ATP synthases. When Ref-1 is knocked down and the cells are exposed to hypoxia, signi cant downregulation of the mitochondrial complex I (p = 0.001), complex III (p = 0.03) and ATP synthase (p = 0.0004) genes were observed. Similarly, statistically signi cant but less downregulated mitochondrial complex I (p = 0.009) genes were observed under normoxia. Figure 2E (Fig. 3B). As 3D spheroid cultures of PDAC cells mimic in vivo tumor hypoxic regions, further evaluation of these genes was done using Pa03C spheroids. Similar downregulation was observed for mitochondrial complex genes as shown in Fig. 3C. In stark contrast, cancer-associated broblast (CAF) cells that would be found within the tumor microenvironment did not demonstrate the same changes in gene expression of these mitochondrial genes as was observed in the tumor cells under either hypoxia or normoxia (Fig. 3D).  Fig. 3E). BIRC5 (Survivin) and CA9 were used as controls and were previously published as markers of Ref-1 redox inhibition (Fig. 3E). 21,41 Again using 3D spheroid cultures of PDAC cells, further evaluation of these Ref-1 regulated genes was done. Similar yet more pronounced downregulation was observed for Ref-1-regulated gene subsets, with the exception of NOTCH3 (Fig. 3F). As seen previously, PPIF gene expression was signi cantly upregulated with Ref-1 redox inhibition using APX2009. Representative pictures of the 3D spheroids following treatment with vehicle (DMSO) or APX2009 are shown in Fig. 3G. The effect of glucose levels in the growth conditions of both Scr control and siRef-1 was tested using 3D spheroid assays incorporating both patient-derived tumor cells as well as CAFs (Supplemental Fig. S1). Pa03C cells were transfected with Ref-1 or Scr siRNA and cultured with or without CAFs in regular or lowglucose (LG) media. In presence and absence of CAFs, the Scr control spheroids grew signi cantly slower in the LG media compared to regular growth media. The cultures containing CAFs were less affected by the low glucose media, suggesting that the CAFs are providing that necessary tumor growth support (Supplemental Fig. S1B). However, when Ref-1 was knocked down, there was no difference in growth regardless of the media, further demonstrating that the main effect on growth of the spheroids was the lack of Ref-1 expression.
To con rm the gene expression and proteomic effects on TCA cycle and also investigate effects on fatty acid oxidation with functional data, a plate-based mitochondrial function assay that measures the electron ow through the electron transport chain was conducted (Supplemental Fig. S2A). This colorimetric assay measured the utilization rate of different metabolic substrates following knockdown of Ref-1 compared to Scr control. Multiple substrates involved in the TCA cycle exhibited signi cantly reduced rates of reaction after transfection with 10 nM siRef-1 (e ciency of Ref-1 knockdown (99%) shown in Fig. 4C) in comparison to 10 nM Scr siRNA (Fig. 4A & B, Supplemental Fig. S2C). Succinic Acid was reduced by 24%, Fumaric Acid reduced by 22%, and Malic Acid was down by 33%. Changes in fatty acid oxidation was not observed under these conditons with these mitochondrial plates (not shown). The kinetic curves at 37ºC for the four TCA cycle substrates within a representative experiment are shown in Figure S2. In order to determine whether Ref-1 redox activity was responsible for the decrease in TCA substrate activity, mitochondrial function assays were performed following APX2009 treatment at 5 and 10 µM.
The four TCA cycle substrates (α-Keto-Glutaric Acid, Succinic Acid, Fumaric Acid, and Malic Acid, Supplemental Fig. S2A, C) displayed signi cantly reduced rates of reaction (30-72%) in response to 5 µM APX2009 treatment, and further dose-dependent reduction was observed when treated with 10 µM APX2009. These results con rm that Ref-1 redox activity plays a role in tumor cells' ability to utilize TCA cycle substrates (Fig. 4D, E and S2C). As a negative control for this redox inhibition, an inactive analog of APX2009, RN7-58, was evaluated for its ability to block mitochondrial function, and no inhibition of TCA substrates was observed (Fig. 4F, G and S2D). 66 We also used a metabolic inhibitor, Devimistat as a positive control to compare and validate the effects on the TCA cycle by APX2009. Devimistat was able to signi cantly reduce two of the TCA cycle substrates (Malic acid by 20-30% and Fumaric Acid by 35% at 50 µM, Fig. 4H and S2F). Similar to changes in mitochondrial complex gene expression in CAFs, APX2009 did not affect mitochondrial function within CAF02 cells (Fig. 4I and S2G 0.0001, Fig. 5A). Growth rate per day of tumor volume among the groups was determined to be signi cantly different using the repeated measure regression model (p < 0.0001). Tumor weights also showed a signi cant decrease compared to the vehicle control group (p < 0.05) (Fig. 5B), with APX2009 and Devimistat showing a ~ 57% reduction in tumor weight. Both treatments were well tolerated as there was no signi cant change in body weights (Fig. 5C).
Due to the importance of the CAFs in the microenvironment and their impact on response to therapy, a tumor model with co-implantation of low passage pancreatic cancer cells, Panc10.05 and CAF19 cells was utilized. Similar to the results in Fig. 5A, treatment with either APX2009 (35 mg/Kg) or Devimistat (50 mg/Kg) resulted in a signi cant decrease in tumor volume (p < 0.0001, Fig. 5D). At the time of harvest, while both treatments showed a decrease in tumor weight, only APX2009 resulted in a statistically signi cant decrease by ~ 61% (p < 0.05, Fig. 5E). No signi cant change in body weights was observed (Fig. 5F). Tumors were collected and stained for H&E and Vimentin. Vimentin staining shows the presence of CAFs even after tumor shrinkage with APX2009 treatment (Fig. 5G and Supplemental Fig.   S3).

Inhibition of Ref-1 redox activity in combination with metabolic inhibitor Devimistat signi cantly reduces tumor growth in two 3D models with simulated PDAC microenvironment
To assess whether the combination of APX2009 and Devimistat would synergize to prevent further tumor growth, two 3D co-culture assays were utilized. These included a 3D co-culture spheroid assay and an interstitial tumor-microenvironment-on-a-chip (iT-MOC) assay. Our group previously published doses of APX compounds that block spheroid growth in the 3D co-culture model. 53 Hence, the dose response of Devimistat as a single agent is shown in Pa03C (Fig. 6B, left panel) and in Panc10.05 cells (Fig. 6E, left  panel). APX2009, parent drug APX3330 as well as additional analog APX2014, were used to inhibit Ref-1 redox activity as single agent and in combination with Devimistat (Fig. 6B, C, E, F and Supplemental Figure S4A). Single agent Devimistat or APX2009 resulted in signi cant reduction in spheroid growth as visualized through uorescence intensity (Fig. 6A-F). However, combination treatment of Ref-1 inhibitor plus Devimistat resulted in further reduction of spheroid growth of the tumor cells (Fig. 6B & 6E). APX2009 + Devimistat had combination index (CI) values ranging from 0.37-0.74 indicating moderate to strong synergy (see Supplemental Table 6 for all values). Importantly, the single agent and combination treatments had little to no effect on CAF19 growth (Fig. 6A-F, with the exception of combination treatment in the 10.05 + CAF19 co-cultures (Fig. 6F, right panel, S4A). Supplemental Figure S4A shows similar results with the Ref-1 redox inhibitors, APX3330 (completed Phase I clinical trial) and APX2014 in combination with Devimistat.
Combination drug experiments were also performed using the iT-MOC assays with the co-culture of Panc10.05 and CAF19 cells. 67 This assay was utilized due to its ability to reconstitute the interstitial transport in a 3D matrix (Fig. 6G). The capillary channel mimics blood-borne drug transport along a capillary vessel. The porous membrane simulates transvascular transport. The interstitial channel where pancreatic cancer cells and CAFs are co-cultured in the 3D matrix aims to reconstitute the interstitial transport. Two side drainage channels correspond to lymphatic drainage. The drug transport is achieved by applying a hydrostatic pressure difference between the capillary channel and lymphatic channel, simulating elevated interstitial uid pressure (IFP) range in PDAC. Transport properties of iT-MOC have been measured and compared with in vivo tumors previously. 57,58 Figure 6H represents the treatment timeline for iT-MOC assay. Representative uorescence micrographs of iT-MOC samples during the experiment are shown in Fig. 6I. In the control group, both cell types proliferated signi cantly over time as indicated by red and green uorescence (Fig. 6I, S4B). In terms of cell growth (Fig. S4B) and survival (Fig. 6J), no notable difference is observed in cancer cell growth among the experimental groups until Day 5. After the second treatment at Day 5, tumor cells treated with a combination of APX2009 and Devimistat grew much slower at 48% of the control group compared to either single agent (p < 0.05, Fig. 6J & S4B). In all experimental groups, no notable difference in the proliferation of CAFs is noted (Fig. 6J & S4C) indicating that the combination treatment was more dramatically affecting the tumor even in the presence of the CAFs and was similar in both 3D co-culture assays.

Inhibition of Ref-1 redox activity in combination with metabolic inhibitor Devimistat reduces mitochondrial gene expression and causes TCA cycle disruption
Based on the 3D data demonstrating e cacy of the combination treatment, we further investigated the effects on gene expression of combination treatment in Pa03C spheroids. Of the six previously tested OXPHOS mitochondrial genes found from the scRNA-seq data, the combination treatment (blue bars) signi cantly reduced the levels of UQCRC1, SURF1, COX15 and NNT compared to either of the single agents used (Fig. 7A). Within the Ref-1-regulated 12 gene panel, combination treatment reduced the levels of NOTCH3 (which was not affected by either single agent) and RAB3D while increasing the levels of PPIF signi cantly compared to either singles. The effects of combination treatment on ABCG2, CIRBP, COMMD7, ISYNA1, PRDX5 and TNFAIP2 were signi cantly down compared to either APX2009 alone or Devimistat alone, however the combination effects were most dramatic in NOTCH3 and RAB3D. BIRC5 and CA9 were used as controls (Fig. 7B).
Following this, we analyzed whether the reduced gene expression with combination treatment had a further functional effect on the TCA cycle. For this, a combination of Devimistat (50 µM) with APX2009 (5 or 10 µM) was used. Devimistat with APX2009 at 10 µM had a greater effect on reducing the rate of reaction of all four TCA substrates compared to APX2009 or Devimistat alone (α-Keto-Glutaric Acid -42% or 58%, Succinic Acid -30% or 64%, Fumaric Acid -25% or 12.5%, and Malic Acid -32% or 46%) (Fig. 7C). Representative picture of the reduction in colorimetric assay shown in Fig. 7D. The effects were also dose-dependent. As the amount of APX2009 increased, the TCA substrate utilization was decreased. Line graphs showing kinetic curves (Supplemental Figures S5) at 37ºC for the four TCA cycle substrates are represented.

Discussion
A hallmark of PDAC is its capability to reprogram metabolism. Pancreatic cancer's ability to adapt to nutrient and oxygen uctuation results in severe therapeutic resistance. Communicating with its tumor microenvironment (TME) through reciprocal upregulation or downregulation of several metabolic pathways like aerobic glycolysis, oxidative phosphorylation, glutaminolysis, lipogenesis and lipolysis, autophagic status, and anti-oxidative stress enables PDAC to thrive under nutrient de ciency and low oxygen conditions, as well as evade therapeutic death. KRAS oncogenic driver mutations are responsible for these tumor cells highly relying on OXPHOS for survival (by supplying ATP) as well as drug resistance (via multidrug transporters). 68,69 Current options for the treatment of PDAC patients are FOLFIRINOX and gemcitabine + nab-paclitaxel. However these multi-agent options are still not adequate to cure pancreatic cancer and are often toxic to the patients. 70 Hence, this study investigates the targeting of metabolic pathways through Ref-1 inhibition as an option for therapy.
Our scRNA-seq analysis of Pa03C cells transfected with Scr or siRef-1 under hypoxia identi ed novel pathways and transcriptional modules regulated by Ref-1, thereby opening up new avenues to treat PDAC. As represented in the Fig. 1D panel 2, pathway enrichment analysis revealed TCA cycle, lipid metabolism, glycolysis, OXPHOS and HIF1α pathways as the top ve to be downregulated with Ref-1 knockdown under hypoxia. Further detailed analysis combining scRNA-seq, proteomics, and functional assays revealed the enzymes involved in those selected pathways to be clustered to TCA cycle and OXPHOS (Fig. 2). With the intent of validating these effects, we selected a set of genes representive of the above pathways (Fig. 3A)  Another interesting gene expression change is the upregulation of PPIF (peptidylprolyl isomerase F or cyclophilin D). PPIF is part of the mitochondrial permeability transition pore (PTP) that is invoved in the induction of cell death. 73 Interestingly, under hypoxia PPIF is signi cantly downregulated, presumably one of the many transcriptional changes that allow cancer cells to survive hypoxic stress. However, with APX2009 or Devimistat treatment or combination of the two, the levels of PPIF are increased about 5-10fold ( Fig. 7B) indicating that inhibition of TCA is likely increasing tumor cell death. The inhibition of TCA cycle activity was con rmed with APX2009, Devimistat, or combination therapy using a functional assay.
Either agent used alone affected the TCA substrates with APX2009 being more potent than Devimistat. These data indicate that Ref-1 redox signaling inhibition reduces the activity of the TCA cycle similar to Devimistat, but at a 10 fold lower concentration, leading to cell death.
CAFs, which are most abundant in the TME, form a critical component of the PDAC milieu. CAFs produce lactate from glycolysis that fuels OXPHOS in the PDAC cells for ATP production via alpha-ketoglutarate, isocitrate dehydrogenase and pyruvate dehydrogenase enzymes. The production of L-lactate, ketones, free fatty acids, and glutamine by CAFs may also drive tumor cell growth through metabolic coupling and enabling the tumor to utilize OXPHOS. HIF1α, NFκB, and loss of Cav-1 have been implicated in the signaling that drives autophagy and catabolism in CAFs that can directly affect tumor growth. Thus, tumors switch between OXPHOS and glycolysis under diverse microenvironmental conditions. 74 Both transcription factors, HIF1α and NFκB are targets of Ref-1 redox signaling activity. [75][76][77] Hence, targeting both OXPHOS in cancer cells as well as glycolysis in CAFs may prove effective, with the caveat that toxicity to non-cancerous cells must be considered. 76,78 Ref-1 redox inhibition showed minimal effects on the expression of mitochondria complex genes and on mitochondria function in the CAFs (Fig. 3D, 4I). Noticeably, the effects of Ref-1 inhibition on mitochondrial function is not the same between tumor and CAFs, pointing again toward a different mechanism of metabolic reprogramming regarding tumor and CAF cells (Fig. 4I). Furthermore, the effects of Devimistat as a single agent or in combination therapy in the 3D models shown in Fig. 6 demonstrate that targeting of the TCA cycle is more cytotoxic to the tumor cells than the CAFs. The 3D co-culture spheroids consisting of both tumor cells and CAFs demonstrate regions of hypoxia and growth factor gradients. 53 3D coculture assays demonstrated that the combination treatment was not only more effective than the single agents, but that the treatments in general targeted the tumor cells more dramatically than the surrounding microenvironment with minimal effects on the CAFs (Fig. 6A-F). In the iT-MOC system, uorescence from both cell types signi cantly increased over time, while in the 3D spheroid based assay the CAFs are less proliferative than the tumors ( Fig. 6 and Supplementary Fig. S4). Likewise, combination of APX2009 with Devimistat proved to signi cantly reduce the expression of all six mitochondrial complex genes when compared to single agents in 3D spheroids consisting of Pa03C cells (Fig. 7A). Devimistat is inhibiting PDH (pyruvate dehydrogenase) and αKGDH (α-ketoglutarate dehydrogenase) and thereby blocking TCA cycle activity, whereby inhibition of Ref-1 appears to dramatically affect the expression of several genes involved in various aspects of metabolism including glycolysis, TCA cycle, and ETC, with TCA cycle being the most downregulated pathway following knockdown in hypoxia (Fig. 1D). Future studies will further de ne the metabolic pathways that are utilized during varying oxygen levels in the presence and absence of Ref-1 and whether there is a shift to TCA cycle at extreme oxygen deprivation as described in 26 . The ROS levels generated during hypoxia as well as Ref-1 knockdown are of interest, as well as activaton of NFκB. 28 Ostensibly, the combination therapy is further shutting down the tumor cells' ability to generate energy and leading to a decrease in 3D spheroid tumor growth. Tumor-selective killing is desirable in PDAC therapeutic approaches, as attempts to modulate the PDAC stroma to improve treatment response have not been well translated and, in fact, have shown to hinder therapeutic approaches. Due to an unanticipated anti-tumor role of select stroma components, the ablation of CAFs or targeting of certain ECM proteins in the stroma led to changes that actually accelerated tumor growth and impaired treatment outcome. [79][80][81] To demonstrate that inhibition of Ref-1 redox activity blocks tumor growth in vivo and also assess the effects on the CAFs, low passage PDAC cells were implanted alongside CAFs into NSG mice. Treatment with either APX2009 or Devimistat which should lead to a decrease in cancer cell metabolism did indeed translate to a signi cant decrease in tumor volume (Fig. 5A & 5D) as well as tumor weight (Fig. 5B & 5E) and did not obliterate the CAFs as indicated by vimentin positivity. These in vivo xenograft results suggest that targeting the metabolic capacity of PDAC tumors will slow their growth rate without killing the associated stroma have strong translational implications.
Several of the transcription factors that interact with Ref-1 have a profound impact on metabolic reprogramming in cancer, including HIF1-α, STAT3, and NRF2 (nuclear factor erythroid 2-related factor 2). 82,83 NRF2 is a transcription factor that can act in a protective mechanism in cancer cells from oxidative and chemical stresses by controlling their redox balance, regulation of antioxidant genes, and metabolic reprogramming. 83  Taken together from our current work, it is clear that more research is needed to fully understand the intersection of Ref-1 signaling and transcriptional regulation of metabolism and its impact on cancer cell growth.

Conclusion
To summarize, this is the rst study to identify Ref-1's role in regulating glycolysis and mitochondrial metabolism through the integration of single cell RNA seq, proteomics, and mitochondrial function under hypoxia and Ref-1 de cient redox signaling conditions. Ref-1's redox signaling / regulatory role plays a crucial function in mitochondrial metabolism, speci cally altering the TCA cycle and gene expression within OXPHOS / ETC thereby enabling the survival of PDAC even under nutrient and oxygen stress. We used 3D spheroids in monoculture as a mechanistic model or as co-culture with CAFs as a means to demonstrate growth regulation by Ref-1 both in vitro as well as in vivo. In these models, we determined that Ref-1 inhibition leads to signi cant reduction in PDAC growth and in regulation of gene expression that leads to the dysregulation of many genes within Complex 1 of the ETC especially under hypoxia (Fig. 2C).
As mitochondria can also generate reactive oxygen species (ROS) that damage nuclear and