Determination of NO as an inhibitor of TET and ALKBH2 demethylase activity in vitro.
In humans, active DNA demethylation is catalyzed by 2 members of the Fe(II)/2-OG-dependent family of oxygenases; TET (there are 3 isoforms: TET1, TET2 & TET3) and ALKBH2. To test whether NO could inhibit TET catalytic activity we incubated the purified catalytic domain of human TET2 enzyme with all cofactors (2-OG, Fe(II), and ascorbate), substrate (synthetic 5mC-DNA oligo), and with NO (the NO-donor Sper/NO, 50 - 500 mM). After 1 hour incubation we measured the relative amounts of the substrate for TET2 (5mC) and its initial oxidation product (5hmC) at each NO (Sper/NO) concentration using a MALDI-TOF-MS-based assay33 (Fig. 1A). Using a broader range of physiologic/pathologic NO concentrations, we determined that NO inhibited TET2 demethylase activity with half-maximal inhibitory concentration (IC50) of 164.5 mM for Sper/NO (~1.5 mM NO) (Fig. 1B). Next, using the IC50 concentration for Sper/NO, we tested whether NO could also inhibit the stepwise TET2-catalyzed oxidation of 5hmC to 5fC and, in a separate reaction, the conversion of 5fC to 5caC. When synthetic DNA oligos containing 5hmC were used as TET2 substrates, the enzymatic conversion of 5hmC to 5fC was inhibited by 23%. Using 5fC as the substrate, conversion to 5caC was inhibited by 20% (Fig. 1C).
To determine how the duration of NO exposure affected TET2 activity and if TET2 inhibition by NO was reversible, we compared the demethylase activity of TET2 alone with the activity of TET2 incubated with NO for either a short or a long exposure time. This was accomplished using DEA/NO, a short-acting NO donor (t½ ≈ 2 min at 37o C), or Sper/NO, a longer-acting NO donor (t½ ≈ 37 min at 37 oC)34. By using the same concentrations of these two NO donors, we were able to expose the enzyme to the same amounts of NO (moles) but for different durations of time. TET2 activity was measured at 1 and 3 hours. When TET2 was not treated with NO, demethylation of 5mC to 5hmC was 100% complete within one hour (“None”, Fig. 1D). When TET2 was incubated for a short period of time with NO, its demethylase activity was initially inhibited (~50% product formation at 1 h), but when NO was no longer present, TET2 was able to completely convert 5mC to 5hmC at 3 hours (“DEA/NO”, Fig. 1D). Conversely, when TET2 was exposed to a continuous steady-state NO concentration for the duration of the experiment, the catalytic activity of TET2 was inhibited (~55%) within 1 hour and no further catalysis occurred during the remaining 2 hours of the experiment (“Sper/NO”, Fig. 1D). These results suggest that TET2 inhibition by NO is reversible. Next, instead of using a synthetic 5mC-DNA oligo substrate, we tested NO-mediated TET2 inhibition using its biological substrate: genomic DNA (Fig. 1E). Again, NO inhibited TET2-catalyzed conversion of 5mC to 5hmC in a concentration-dependent manner.
Having established that NO could inhibit TET2, we tested whether NO could similarly inhibit ALKBH2, another Fe(II)/2-OG-dependent DNA demethylase. ALKBH2 activity was measured in real-time using a fluorescence assay35. This assay used a fluorogenic 1-methyladenine probe that has a >10-fold increase in fluorescent signal intensity when the alkyl group is demethylated by ALKBH2. We incubated recombinant ALKBH2 and all cofactors (2-OG, Fe(II), ascorbate) with the methylated probe substrate and NO (Sper/NO, 100-500 mM) for 60 minutes (Fig. 1F). Under these conditions, NO inhibited ALKBH2 demethylase activity in a concentration-dependent manner with an IC50 for Sper/NO of 165 mM (Fig. 1G). Next, we isolated nuclear and cytosolic extracts containing endogenous ALKBH2 from cultured MDA-MB-231 breast cancer cells to test whether NO could similarly inhibit ALKBH2 derived from biological sources. The combined extracts were exposed to low steady-state concentrations of NO using the NO donor DETA/NO (t½ ≈ 22 h, 50-150 mM) and demethylase activity was monitored for 12 h using the fluorescent probe method (Fig. 1H). Over this range of DETA/NO concentrations we expect that the steady-state concentrations of NO correspond to low nM physiological levels (as we have previously measured36,37). Under these conditions ALKBH2 demethylase activity was inhibited in a dose-dependent manner. Collectively, these results indicate that NO is a potent inhibitor of Fe(II)/2-OG-dependent DNA demethylases TET2 and ALKBH2. For subsequent studies we solely focused on TET enzymes because of their gene regulatory functions (rather than ALKBH2 which is part of the DNA damage response).
Nitric oxide forms a dinitrosyl iron complex at the mononuclear non-heme iron atom in TET2. A critical step in the TET-catalyzed DNA demethylation reaction is binding O2 to the non-heme iron atom38. We hypothesized that because of NO’s structural and bonding similarity to O2 that NO would compete with O2 and inhibit TET by forming the more stable mononitrosyl complex at the iron site. To test whether NO interacts with the iron center, we conducted electron paramagnetic resonance (EPR) studies of TET2 treated with NO in the presence of all cofactors and substrate. We ran two identical reactions in parallel and stopped them at different time points (<1 and 20 minutes). EPR spectra at both time points showed that the enzyme does not form the expected S = 3/2 mononitrosyl complex (Fe(II)-NO), as shown by the absence of its intense g┴ ~ 4.1 signal (Fig. S1A), and instead revealed the characteristic S = ½ signal of a non-heme dinitrosyl iron complex (DNIC, [Fe(NO)2]9), with g┴~2.03, g|| ~ 2.01 (Fig. 2A)37,39,40. The signature EPR spectrum of DNIC was almost undetectable when the TET2 enzyme was omitted from the complete reaction mixture (Fig. S1A & 2A), indicating that the observed DNIC signal is associated with the enzyme, and not with a complex formed in solution. Kinetically, the reaction was ~80% complete by the time all reactants were added, as shown by the small increase in the EPR intensity between samples frozen at 1 min and 20 min.
Density functional computations support the formation of a DNIC at the catalytic iron atom in TET2. Because the DNIC was formed instead of the mononitrosyl a series of density functional theory (DFT) computations were performed to investigate the relative affinity for binding of NO versus O2 and water in a TET2 resting state. An active site model of TET2 was generated akin to that reported by Lu et al.41 The charge of the model DNIC complex was adjusted to yield a neutral {Fe(NO)2}9 electron count, i.e., d6 Fe(II) + 2 NO p* e- + 1 additional e-. Additional simulations were performed to assess NO binding to models in which 2-OG was ligated to the inner coordination sphere, but these were largely inconclusive apart from indicating that NO is bound more weakly to Fe(II) after ligation of 2-OG. The iron in the enzyme is coordinated by two histidine (His) and one aspartate (Asp) residue. To mimic the pertinent amino acid side chains, Asp was modeled by an acetate (OAc-) and His was modeled by N-methyl-imidizaole (Im). It was assumed that Asp and His model ligands would maintain a fac configuration.
The neutral {Fe(NO)2}9 dinitrosyl complex Fe(OAc)(Im)2(NO)2(OH2) was chosen for modeling studies; the geometry is shown (Fig. 2B). Fe(OAc)(Im)2(NO)2(OH2) is predicted to be a triplet with cis nitrosyl ligands. The lowest energy coordination isomer had one NO trans to OAc and the other trans to Im. The OAc that mimics the Asp side chain forms a strong hydrogen bond with the ligated water. Given the experimental spectroscopic observations, it was investigated whether one of the NO ligands of Fe(OAc)(Im)2(NO)2(OH2) could be displaced by either water or dioxygen; thus, computed DFT ground states of Fe(OAc)(Im)2(NO)(OH2)2 and Fe(OAc)(Im)2(NO)(O2)(OH2) were sought. The lowest energy geometries are shown in Fig. 2C. In both cases, geometry optimizations initiated from six-coordinate, pseudo-octahedral starting guesses yielded minima with a weakly bound exogenous ligand. For Fe(OAc)(Im)2(NO)(OH2)2, the NO ligand is ejected from the inner coordination sphere upon geometry optimization, Fe…NO = 4.13 Å. For Fe(OAc)(Im)2(NO)(O2)(OH2), the complex barely maintains an octahedral geometry with the dioxygen very weakly bonded (Fe…O2 ~ 2.95 Å). The tenuous nature of O2 binding in Fe(OAc)(Im)2(NO)(O2)(OH2) is further indicated by the near complete lack of any spin delocalization from the O2 to the complex (rspin(O2) = 1.99 e-) and the computed free energy for NO/H2O exchange, DG = +23.0 kcal/mol, indicating that NO binds much more tightly than water. The NO/O2 exchange free energy is essentially thermoneutral, DG = -0.3 kcal/mol. In conjunction with the weak O2 binding indicated by the long Fe…O2 bond length of the optimized geometry of Fe(OAc)(Im)2(NO)(O2)(OH2), (Fig. 2D), the DFT results suggest that O2 does not readily displace NO, perhaps except at high O2 partial pressures, and rationalizes the presence of only the di-nitrosyl complex as seen in the EPR spectra (Fig. 2A).
Based on the DFT results, further modeling was done for the DNIC core and the crystal structure of TET2 in complex with N-oxalyglycine (OGA), a 2-oxoglutarate analog42 (PDB ID 4NM6, Fig. 2D). In the model, the NO molecules replace OGA and occupy similar positions to the OGA coordinating oxygens. One of these is close to Arg 1261, which may hydrogen bond to the NO and stabilize overall negative charge buildup on the NO ligands in the DNIC (Fig. 2E,F).
Nitric oxide increases 5mC in the DNA of human cancer cells. Having established that NO was a direct and potent inhibitor of TET enzymes under isolated conditions, the next step was to investigate whether NO could inhibit endogenous cellular TET enzymes and to determine if this would regulate nuclear DNA methylation. We selected four human cancer cell lines derived from aggressive tumor types that are known to express NOS2 and synthesize NO in vivo (2 triple negative breast (TNBC), 1 prostate, 1 brain). These cells do not synthesize NO in culture, so we treated them with DETA/NO (100 mM; 24 h), which resulted in low nM concentrations of NO, and measured 5mC-DNA (Fig. 3A). In all cell lines NO significantly increased global 5mC in DNA. Among NO-associated cancers, TNBC patients who harbor NOS2-expressing tumors have significantly worse prognoses. For this reason, we conducted all subsequent experiments using models of TNBC.
Under biological conditions, DNA methylation patterns are faithfully maintained over multiple cell generations as a form of epigenetic inheritance. Therefore, we developed a cell model to study DNA methylation responses to NO after multiple cell generations; this more accurately mimics the microenvironment of NOS2-expressing tumors in vivo where cells are exposed to chronic NO synthesis. We treated two TNBC cell lines with low physiologic steady-state concentrations of NO for 10 days (~12 cell doublings (NO did not alter the doubling rate)) and examined long-term “heritable” DNA methylation patterns. In the NO-treated cells, there was a significant increase in 5mC in DNA (Fig. 3B), and an increase in 5hmC in DNA, the first oxidation product of TET (Fig. 3C). Cells not treated with NO had no change in 5mC/5hmC, suggesting 5mC increases were not attributable to epigenetic drift. To mimic the endogenous NO production observed in tumors, we transfected MDA-MB-231 cells with a human NOS2 gene (or empty vector control (VC)) (Fig. 3D). Accumulative NO synthesis was measured after 24 and 48 hours in both cell lines and NO was only detected in the NOS2-transfected cells, not in the cells transfected with the empty vector plasmid or in the NOS2-transfected cells treated with a pan-NOS inhibitor (L-NMMA) (Fig. 3E). 5mC in DNA was also measured in both cell lines at 24 and 48 hours and 5mC was elevated only in the NO-producing cells but not the control cells or cells treated with L-NMMA (Fig. 3F).
5mC is catalytically installed on DNA by DNA methyltransferase enzymes (DNMT) which, along with TET enzymes, maintain steady-state 5mC levels. NO-dependent increases in 5mC could therefore be due to increased DNA methyltransferase activity rather than inhibition of TET DNA demethylase activity. To test this, we treated MDA-MB-231 cells with NO and either a DNA methyltransferase 1 (DNMT1) inhibitor (5-Azacytidine (AZA)), or a competitive inhibitor of methionine adenosyltransferase (MAT) (cycloleucine (CL)). CL depletes the cell of S-adenosylmethionine (SAM), the substrate for DNA methyltransferases (Fig. 3G). In cells treated with either AZA or CL alone, a significant reduction in global 5mC levels was observed as expected43, but when NO was present during either CL or AZA treatment, 5mC levels remained elevated (Fig. 3H). Another potential explanation for increases in 5mC would be if NO was changing the expression levels of DNA methyl-modifying enzymes (i.e., increasing DNMT or decreasing TET expression). To examine this possibility, we treated two TNBC cell lines with NO for 10 days and measured the protein expression levels of DNA methyltransferases (DNMT1, DNMT3a, DNMT3b) and DNA demethylases (TET1,2,3, and ALKBH2) (Fig. 3I,J). For all enzymes there was almost no change in protein expression in response to NO after 10 days. Together these data further support the hypothesis that NO-mediated increases in 5mC result from inhibition of TET demethylases and are not a result of changes in the expression levels of DNA methyl-modifying enzymes or increased DNA methyltransferase activity.
Nitric oxide increases 5mC in DNA from tumors in vivo. To investigate whether NO could increase 5mC in vivo we used a mouse xenograft model of NOS2-expressing cell-line derived tumors. Mice bearing NOS2-expressing MDA-MB-231 xenograft tumors were divided into two groups; half were treated with aminoguanidine (AG), a selective inhibitor of NOS2, and the other half were treated with saline (control). After 37 days of treatment, the tumors were removed, the DNA extracted, and 5mC-DNA was quantified (Fig. 3K). 5mC in DNA from the NO-producing tumors was significantly greater than in the tumors where NO synthesis was inhibited. As further confirmation that NO regulates 5mC in vivo we measured 5mC-DNA in NOS2-positive patient derived xenograft (PDX) tumors (Fig. 3L). In this experiment the control group received a vehicle saline injection, and the treatment group was administered the pan-NOS inhibitor NG-monomethyl-L-arginine (L-NMMA) daily. After 40 days the tumors were excised and 5mC was measured. Again, in the NO-producing PDX tumors, 5mC was significantly higher than in the tumors where NO synthesis was inhibited (Fig. 3L).
NOS2 expression and NO production drive aggressive cancer phenotypes. Clinically, NOS2 expression in tumors is associated with worse patient outcomes and poor responses to therapy. We used Kaplan-Meier Plotter44 to analyze transcriptomic datasets of metastatic breast cancer patients found in GEO, EGA and TCGA. When we examined NOS2 expression in 3 breast cancer patient groups (all subtypes, basal-like, and ER-/PR-) we found that high NOS2 expression was associated with decreased overall survival (confirming previous reports11) (Fig. 4A). To experimentally determine whether NO would result in a more aggressive cell phenotype in vitro, we measured cell migration and invasion (two hallmarks of cancer) in real time of TNBC cells exposed to NO. In cells exposed to exogenous NO or in cells endogenously synthesizing NO, the rates of cell migration and invasion were increased compared to untreated control cells (Fig. 4B-D), consistent with what we and others have shown previously6,45. Another phenotype associated with tumor aggressiveness is resistance to chemotherapy. Using the ROC plotter platform46, we analyzed the expression levels of NOS2 as a predictive biomarker of efficacy of any chemotherapy in TNBC patients (n = 164; response based on relapse-free survival at 5 years). Patient “non-responders” to therapy had significantly higher NOS2 expression than patient “responders” to therapy (Fig. 4E).
Nitric oxide regulates gene expression in TNBC cells. With the observation that NO directly inhibits TET demethylase activity leading to global increases in 5mC, we sought to decipher whether this was functionally associated with transcriptional changes in NO-regulated genes that may drive aggressive phenotypes. First, we quantified changes in transcription by performing RNA-sequencing on samples from two TNBC cell lines treated chronically (10 days) with NO. In both cell lines, NO significantly up- and down-regulated several hundred genes compared to untreated control cells (Fig. 4F,G). There were 880 significantly differentially expressed genes in the MDA-MB-231 cells with FDR (False Discovery Rate) <0.05 (454 upregulated log2FC>1, 426 downregulated log2FC <-1) and 765 significantly differentially expressed genes in the MDA-MB-468 cells (451 upregulated log2FC>1, 314 downregulated log2FC <-1). Although there was only a 12 - 14% overlap in common genes transcriptionally regulated by NO between the two cell types (90 in total), this may be due to significant differences in their basal transcriptional profiles (control cells not treated with NO). Multidimensional scaling (MDS) analysis illustrated how the expression profiles differed far more between the two cell types than between the NO treated and control cells (Fig. 4H). Despite only modest overlap in specific genes transcriptionally regulated by NO in both cell types, Gene Set Enrichment Analysis (GSEA) of the 90 genes differentially expressed in the same direction identified several KEGG pathways relevant to cancer progression (Table S1).
Nitric oxide increases 5mC and 5hmC differentially at specific genomic features. To determine if NO-mediated changes in 5mC/5hmC were functionally associated with changes in gene expression, we identified the locations of 5mC/5hmC on a genome-wide scale at single-nucleotide resolution by performing oxidative reduced representation bisulfite sequencing (oxRRBS) on samples from two TNBC cell lines that were chronically treated with NO for 10 days. Between the NO-treated cells and the untreated cells we identified differentially methylated positions (DMPs) and differentially hydroxymethylated positions (DhMPs, 5hmC), defined as: p < 0.05 and difference in β-value > 0.1. On a global scale, we found that both 5mC and 5hmC were increasing in both cell types in the NO-treated cells compared to the untreated control cells, consistent with the ELISA data in Figure 3 that NO increases 5mC/5hmC in cells. Although there were net increases in 5mC and 5hmC, these changes were dynamic in that both increases and decreases in 5hmC and 5mC were observed at all annotated regions (Fig. 4I).When we examined annotated CpG sites (islands, shores, shelves, and open seas)47 we found that the majority (>60%) of DMPs and (>45%) DhMPs were occurring at open sea positions (Fig. 4J). We then focused on specific functional elements (Super enhancers (SE), 5’ UTR, 3’ UTR, Typical enhancers (TE), promoters, exons, intergenic regions, and introns) to determine if they also exhibited specific patterns of 5mC/5hmC enrichment. In both cell types, and at all genic annotations, we identified hyper- and hypo-DMPs and DhMPs (Fig. 4K, L), with the majority located at introns. The locations of DMPs and DhMPs were similar in both cell types, but the numbers of differentially methylated positions tended to be greater in the MDA-MB-468 cells (Fig. 4M).
Determination of 5mC- and 5hmC-associated transcriptional changes. Having demonstrated that NO increased 5mC/5hmC in DNA at genomic loci relevant to regulation of gene expression, and that NO produced significant transcriptional changes, we attempted to link changes in DNA methylation to the changes in gene expression. We identified overlaps between significantly expressed genes (RNA-seq) and their b-values (the degree of CpG methylation) at that gene or at gene regulatory loci associated with that gene (i.e. promoters, enhancers, super enhancers). For both control and NO treatment groups there was a clear negative correlation between 5mC b-values at promoters and gene expression in both cell types (Extended data Fig. 2). We next identified specific genes that had significant transcriptional changes in response to NO (upregulated or downregulated) and also had significant changes in methylation (increase or decrease in b-value for 5mC or 5hmC) at specific gene-regulatory genomic loci (Fig. 5 A-C, Extended data Tables 2 & 3). Although the correlations between methylation status (hyper or hypo, 5mC or 5hmC) of specific gene-regulatory regions and the direction of transcriptional changes were not 100%, certain trends did emerge. For example, increases of 5mC/5hmC at promoters was more associated with downregulated genes whereas gene body enrichment of 5mC/5hmC was more associated with upregulated genes. Increased 5mC/5hmC at typical enhancers correlated to downregulation of associated genes. Although links between many of the differentially expressed genes in Fig. 5 A-C and cancer are unknown or have yet to be established, some of them have an experimental or clinical association with breast cancer progression (Fig. 5D, E). On the left sides of Figures 5D & E (“Cellular Responses to NO”) are select genes that are transcriptionally regulated by NO and show significant changes in their b-values at the promoter regions for these genes (Fig. 5D is 5mC, 5E is 5hmC). On the right side of these figures (“Clinical Correlation”) are results from analysis of publicly available data sets of gene expression from tumors of patients with aggressive breast cancers48. Genes that were transcriptionally regulated by NO in our cell culture models also had directionally similar gene expression changes in NOS2-expressing patient tumors. Kaplan Meier plots demonstrate the correlation between the NO-regulated gene and patient survival44. Although a direct relationship between these NO-regulated genes and cancer progression have yet to be documented in the scientific literature, some reports demonstrate that the genes upregulated by NO (GJC2, CPA4, SMG8, COL5A2, POLQ)49-53 and the genes downregulated by NO (SYTL1, PRR15L)54,55 are associated with deleterious outcomes when up- or down-regulated in the same direction in cancer patients. These data demonstrate that genes that exhibit changes in their promoter 5mC/5hmC status and are transcriptionally regulated by NO in breast cancer cells in vitro show similar trends in vivo in patient tumors, suggesting a link with NO-regulated genes and the association between poor patient outcomes in breast cancer.