Alkylation Sensitivity Screens Reveal a Conserved Cross-species Functionome

This article is featured in Highlights of This Issue, p. 1511

Glioblastoma multiforme (GBM) is the most common and aggressive primary brain tumor (1). Temozolomide (TMZ), radiation and surgery are currently used for treatment of GBM, yet median survival is still less than 2 years (2–4). Chemotherapy resistance and difficulties in surgical removal contribute to poor prognosis (1). TMZ methylates several bases in DNA, including the O⁶ position of guanine (O⁶-MeG), the N7 position of guanine (N7-MeG) and the N3 position of adenine (N3-MeA). Two of the lesions (N7-MeG and N3-MeA) account for greater than 80% of the DNA lesions induced by TMZ, yet result in negligible clinical toxicity because of robust repair mechanisms. These lesions are predominantly repaired by the base excision repair (BER) pathway (5), initiated by one of eleven lesion specific DNA glycosylases (6).

Virtually all of TMZ's clinical cytotoxicity is attributable to the O⁶-methylguanine (O⁶-MeG) lesion, which accounts for approximately 5% of TMZ induced lesions (7). The O⁶-MeG lesion is repaired via a direct reversal mechanism by the protein O⁶-methylguanine-DNA methyltransferase (MGMT), which transfers the O⁶-methyl group from the guanine base onto a Cys residue in the MGMT protein (8). If the O⁶-MeG lesion is not removed by MGMT, during cellular replication the mis-pairing of O⁶-MeG with thymine is detected by the mismatch repair enzymes, triggering apoptosis signaling and cytotoxicity (9). However, 5-year survival rates still remain low in TMZ-treated patients (3, 4), and TMZ resistance and/or recurrence with chemotherapy resistant tumors is common. Resistant cells can harbor mutations in mismatch repair proteins such as mutS homolog 6 (MSH6; ref. 10) or have elevated expression of MGMT (as in the T98G cell line; ref. 11). Earlier endeavors to enhance TMZ efficacy by using MGMT inhibitors to prevent the repair of O⁶-MeG lesions have not shown an increase in sensitivity or efficacy in clinical trials (12), especially in TMZ-resistant GBM (13).

Because GBM tumors can be resistant to TMZ therapy and recurring tumors may acquire resistance to the O⁶-MeG lesion (14, 15), we explored novel mechanisms of TMZ resistance in tumor cells independent of the O⁶-MeG lesion. To discover genes that sensitize cells to TMZ, we undertook a synthetic lethal screen using Ambion's Silencer Human Druggable Genome siRNA library, the chemotherapeutic agent TMZ and the T98G cell line, a TMZ-resistant cell line derived from a GBM tumor (16). We determined that several biologic pathways were enriched for TMZ-sensitizing genes, including proteins involved in BER, response to DNA damage, cellular proliferation, and protein modification. Many similar biologic processes were also highly enriched in previous DNA alkylation screens conducted in Saccharomyces cerevisiae and Escherichia coli (17, 18). We combined the selected hits from all 3 species and conducted a functionome analysis to identify significantly enriched biologic processes conserved across all 3 organisms (19). This allowed us to create a cross-species network representing the shared alkylation response from bacteria, yeast, and humans and shows an alkylation functionome that includes many novel proteins not previously thought to impact alkylation resistance. Our identification of evolutionarily conserved mechanisms affecting TMZ sensitivity suggests that although many proteins and processes impact sensitivity to alkylators, several critical survival pathways can be targeted to improve chemotherapy efficacy. Simultaneous inhibition of both DNA repair and protein modification processes in yeast yielded greater sensitivity to alkylating agents than inhibition of either process alone. However, knockdown (KD) of both protein modification and DNA repair genes did not have an increased effect on alkylation-induced toxicity in human cells. This potential epistatic interaction between the ubiquitin protein ligase E3B (UBE3B) and the DNA repair protein uracil-DNA glycosylase (UNG) suggests they may be in the same survival pathway. Importantly, the biologic processes and corresponding genes identified in our functionome analysis represent novel potential drug targets to increase TMZ efficacy independent of O⁶-MeG–mediated toxicity as well as potential epistatic relationships that would provide mechanistic insight into the function of specific activities.

DharmaFECT 2 transfection reagent and the siGENOME Non-Targeting siRNA #1 were from Dharmacon. CellTiter-Blue Cell Viability Assay and CellTiter 96 AQueous One Solution Cell Proliferation Assay were from Promega. The Silencer Druggable Genome siRNA Library (Version 1.1) and 5× siRNA resuspension buffer were from Ambion. Tissue culture–treated 384-well microtiter plates were from Greiner Bio-One (GmbH). OptiMEM, EMEM, phosphate buffered saline (PBS), and Hoechst 33342 were from Invitrogen. The shRNA vectors used for stable KD cell line creation were obtained as glycerol stocks from Sigma-Aldrich and the UPCI lentiviral core facility (http://www.upci.upmc.edu/vcf/lenti.cfm). TMZ was from the National Cancer Institute Developmental Therapeutics Program (Bethesda, MD). Puromycin was from Clontech Laboratories, and gentamycin was obtained from Irvine Scientific. We used the following primary antibodies: UNG antibody (#MBS200056) was from MyBioSource and proliferating cell nuclear antigen (PCNA) antibody (#sc-56) was from Santa Cruz. Secondary antibodies: goat anti-mouse horseradish peroxidase (GAM-HRP) conjugates were from Bio-Rad. Signal generation substrates were from Bio-Rad and Thermo Fisher Scientific. All electrophoresis reagents were from Bio-Rad.

Cell line and culture conditions were as previously described (20–22). Briefly, the T98G glioblastoma cell line was purchased from American Type Culture Collection (ATCC; obtained in 2004) and cultured in EMEM with 10% heat-inactivated FBS, nonessential amino acids, sodium pyruvate, antibiotic/antimycotic, and gentamycin (16). The LN428 cells were cultured in alpha minimum essential media with 10% heat-inactivated FBS, antibiotic/antimycotic, and gentamicin as we have described (23). The T98G cell line was tested for cross species contamination and authenticated by RADIL cell check services as of November 16, 2012 (after the completion of the siRNA screen). The genetic profile provided by RADIL was checked against the ATCC STR loci database to confirm the identity of the T98G cell line.

Several different programs were used to analyze the hit list for pathway enrichment and network visualization, including NIH Database for Annotation, Visualization and Integrated Discovery (DAVID), Princeton GO term finder, and Ingenuity Pathway Analysis (IPA). DAVID and Princeton GO term finder both classify genes based on their gene ontology (GO; refs. 24–26). Each gene is classified based on its molecular function, biologic process, or cellular compartment and are referred to as “GO terms” (25). Enriched GO terms and their associated genes are more likely to be true hits because multiple genes affecting the same process all yield sensitization (24, 27). The gene enrichment calculations used the 5,520 genes screened as the background, due to the relatively small number of genes screened, instead of the entire human genome. The analysis with NIH DAVID was conducted using multiple classification stringencies, gene ontology hierarchies, protein interaction databases, and pathways. The hit list was also analyzed with IPA (Ingenuity Systems, www.ingenuity.com), using a proprietary, manually curated, interaction database. Because IPA uses a different scheme and algorithm to organize proteins into networks not based on GO terms, it was used as a distinct method to analyze the hit list for enriched protein functions and networks. The Functional Analysis identified the biologic functions that were most significant to the data set. Right-tailed Fisher's exact test was used to calculate a P value determining the probability that each biologic function assigned to that data set is because of chance alone. Fisher exact test was used with α = 0.05 to compute the probability of correct functional assignment for the genes in the hit list. The enriched networks created by IPA were visualized with genes serving as nodes and edges representing known interactions. Genes colored green modulate toxicity to alkylating agents as determined in the screen.

The shuttle vectors for expression of shRNA were from Sigma. Lentiviruses were prepared in collaboration with the UPCI Lentiviral facility. Lentiviral particles were generated by cotransfection of 4 plasmids [the shuttle vector plus three packaging plasmids: pMD2.g(VSVG), pVSV-REV, and PMDLg/pRRE] into 293-FT cells (28, 29) using FuGene 6 Transfection Reagent (Roche), as described previously (21). Lentiviral transduction was conducted as described earlier (21). Briefly, 6.0 × 10⁴ cells were seeded into a 6-well plate 24 hours before transduction. Cells were transduced for 18 hours at 32°C and then cultured for 72 hours at 37°C. Cells were then selected by culturing in growth media with 1.0 μg/mL puromycin, as previously described (21). Stable LN428 glycosylase-KD cell lines are available from Trevigen, Inc.

Expression of mRNA for each of 10 glycosylases and 5 protein modification genes (OGG1, SMUG1, MBD4, UNG, MYH, NTHL1, MPG, NEIL1, NEIL2, NEIL3, UBE3B, ICMT, B4GALT7, CHRM3, and PADI1) after shRNA-mediated KD was measured by quantitative RT-PCR (qRT-PCR) using an Applied Biosystems StepOnePlus system as described previously (21). Applied Biosystems TaqMan Gene Expression Assays used are as follows: human OGG1: Hs00213454_m1; human SMUG1: Hs00204820_m1; human MBD4: Hs00187498_m1; human UNG: Hs00422172_m1; human MYH: Hs01014856_m1; human NTHL1: Hs00267385; human MPG: Hs01012594_m1; human NEIL1: Hs0022637_m1; human NEIL2: Hs00376746_m1; human NEIL3: Hs00217387_m1; human UBE3B: Hs00296200_m1; Human ICMT: Hs00202655_m1; Human B4GALT7: Hs01011258_m1; Human CHRM3: Hs00265216_s1; and Human PADI1: Hs00203458_m1. Gene expression of each gene was normalized to the expression of human β-actin (part #4333762T).

Nuclear extracts were prepared and protein concentrations were determined as described previously (21). Fifteen micrograms of protein was loaded on a precast 4% to 20% Tris-Glycine gel (Invitrogen). The following primary antibodies were used in immunoblot assays: anti-human UNG from MyBioSource.com (#MBS200056) and anti-human PCNA (Santa Cruz Biotechnology). Immunoblot membranes were stripped before reprobing for PCNA. The membranes were stripped with Restore PLUS Western Blot Stripping buffer (#46430) from Thermo Fisher Scientific per the manufacturer's instructions.

All oligodeoxyribonucleotides in the molecular beacon assay were purchased from Integrated DNA Technologies including the following: FD-Con, 6-FAM-dGCACTATTGAATTGACACGCCATGTCGATCAATTCAATAGTGC-Dabcyl, where 6-FAM is carboxyfluorescein and Dabcyl is 4-(4′-dimethylaminophenylazo) benzoic acid; FD-UNG1, 6-FAM-dGCACTUAAGAATTGACACGCCATGTCGATCAATTCTTAAGTGC-Dabcyl, where U is 2′-deoxyuridine. All DNA glycosylase molecular beacon activity assay experiments and data analysis were done as previously described (22, 30). A more detailed description of the assay is also available (31).

Briefly, the molecular beacons form a stem-loop structure containing a 13-nucleotide loop and a 15 base pair stem forcing 6-FAM fluorophore at the 5′ end and the Dabcyl quencher at the 3′ end into close proximity. When in a stem-loop structure, the 6-FAM fluorescence is efficiently quenched by Dabcyl in a nonfluorescent manner via Forster Resonance Energy Transfer (FRET; refs. 32, 33). However, if the uracil is removed by UNG and the DNA backbone is hydrolyzed by APE1, the 6-FAM containing oligonucleotide (5 bases in length) will dissociate from the hairpin at 37°C (Fig. 4B). This will decrease Dabcyl quenching of 6-FAM and the increase in fluorescence is proportional to uracil removal. Any increase in fluorescence in a control beacon with no base lesion is the result of nonspecific DNA backbone cleavage.

Dialysis of nuclear lysates for the molecular beacon assay was done as previously described (22, 30). Briefly, nuclear protein extracts were dialyzed twice for 90 minutes at 4°C using the 7,000 molecular weight cut-off Slide-A-Lyzer Dialysis Cassette (Pierce). Molecular beacon reactions were conducted at 37°C using 2 μg of dialyzed protein extract and beacon substrate (final conc. = 40 nmol/L). Fluorescence was measured every 20 seconds for 60 minutes, using a StepOnePlus real-time PCR system and expressed as arbitrary units.

TMZ induced cytotoxicity was determined as described previously (34). Briefly, cells were seeded into 96-well plates at a density of 2,000 cells per well. Cells were treated with TMZ for 48 hours at 37°C before determining the relative amount of metabolically active cells by an MTS assay. Results were the average of three separate experiments and normalized to vehicle treated control cells with error bars representing the standard error of the mean.

Cells were grown until approximately 50% to 75% confluence before being trypsinized and counted using a CASY cell counter per the manufacturer's instructions (Roche). Cells were seeded into 96-well plates at a density of 120 cells/well and incubated at 37°C for 24 hours. The cells were treated with vehicle or TMZ and incubated for 9 days at 37°C. Plates were removed and fluorescence was determined using the CyQuant kit (Invitrogen, #C7026) following the manufacturer's instructions. Results were the average of 2 separate experiments and normalized to vehicle-treated control cells with error bars representing the SE of the mean.

Cells (T98G) were seeded on 35-mm glass bottom dishes (MatTek Corporation) and incubated with the superoxide indicator dihydroethidium (DHE, 5 μmol/L, Invitrogen) for 15 minutes at 37°C following the treatment with TMZ (1 mmol/L) or vehicle [1% dimethyl sulfoxide (DMSO)] for the times indicated in the figures. Cells were washed with PBS, the media replaced and the dish inserted in a closed, thermocontrolled (37°C) stage top incubator (Tokai Hit Co.) atop the motorized stage of an inverted Nikon TiE fluorescent microscope (Nikon Inc.) equipped with a ×40 oil immersion optic (Nikon, CFI PlanFluor, NA 1.3) and NIS Elements Software. DHE was excited using a Lumencor diode-pumped light engine (SpectraX, Lumencor Inc.) and detected using a DsRed longpass filter set (Chroma Technology Corp) and Photometrics CoolSNAP HQ2 camera (Photometrics). Data were collected in 2 to 12 cells per stage position, with 15 stage positions in each of 2 separate experiments.

The enriched biologic process information associated with our TMZ screen was compared with other alkylation screens reported for S. cerevisiae and E. coli. Our cross-species analysis was conducted similar to a reported analysis (19). Briefly, to determine if the same gene ontology biologic processes were enriched in the alkylation sensitive hit lists from all 3 organisms, representing X number of genes, we aligned the GO terms from each organism. To identify GO terms significantly enriched for alkylation sensitive genes from all 3 organisms we randomly sampled X genes from the complete search space for each screen and identified associated GO functional terms. Random sampling was conducted more than 200 iterations and the average number of hits in each GO category and the standard deviation were determined. These values were then compared with the actual values to identify GO terms significantly (P < 0.05) enriched for alkylation sensitive genes from all 3 organisms. The molecular functions and the corresponding genes, which were significantly enriched in all 3 alkylation screen data sets, were then visualized using Cytoscape (35). Genes which modulate alkylator toxicity from E. coli (dark blue),S. cerevisiae (light blue), and human (light green), all served as protein nodes, with enriched GO terms serving as central nodes. The edges between GO terms and genes demarcate which genes are annotated to that biologic process.

Media preparation and other yeast manipulations were conducted using standard methods. Mutants were made using a G418 knock out cassette from the S. cerevisiae Gene Deletion Project and were selected on Yeast Peptone Dextrose (YPD) plates containing G418 (200 mg/mL). A mag1Δ mutant was also made using a URA3 based strategy with selection occurring on Synthetic Defined media lacking uracil (SD-URA). Mutants were confirmed by PCR. Plate-based methyl methanesulfonate (MMS) viability studies were conducted as previously reported (36).

Transient transfection by siRNA was as previously described (20, 37). Briefly, T98G cells were wet-reverse transfected with the Ambion Silencer Druggable Genome siRNA library in 384-well plates using 1.56 μL of 833.3 nmol/L siRNA for each well, as described previously (20). The library consisted of siRNAs targeting 5,520 different genes. Three unique siRNA duplexes targeting the same gene were pooled into a single well with one-gene target per well. The siRNA were prepared with DharmaFECT2 and OptiMEM and split evenly into two 384-well plates. T98G cells were added directly to the siRNA complexes. The plates were incubated for 5 hours at 37°C with 5% CO₂ before replacement with fresh media.

Cells were incubated at 37°C with 5% CO₂ for 48 hours after siRNA transfection to allow for gene silencing before addition of TMZ or vehicle. The media was removed and the cells were treated with media containing either DMSO vehicle (final concentration 1.0%) or the approximate EC₁₀ of TMZ (final concentration 1 mmol/L in 1% DMSO) for 48 hours. Cell viability was measured 96 hours after siRNA transfection with the CellTiter-Blue viability assay according to manufacturer's protocol.

The siRNA screen was conducted 3 times over 6 separate weeks. Fluorescence units from each well were normalized to plate negative controls (scrambled siRNA) enabling cell viability comparisons between different plates. The specific details on the data analysis were described by one of us previously (20). In addition, data were analyzed by creating a high confidence hit list as described previously (38). Briefly, cell viabilities for each targeting gene from the 3 screening replicates were averaged and data were analyzed using 2 statistical analysis methods. For each gene, a 2-sample t test was conducted to determine if there was a sensitization effect on cellular survival between “siRNA plus vehicle” treated cells as compared with “siRNA plus TMZ” treated cells. We selected targeting siRNAs with a P value less than or equal to this threshold.

Also, viability ratios (ratio of survival of “siRNA plus TMZ” to “siRNA plus vehicle”) were calculated for each targeting siRNA to determine the magnitude of response between the 2 screening conditions. The viability ratios were sorted in descending order and targeting siRNAs with a viability ratio in the lowest 5th percentile were selected for further analysis. The targeting siRNAs identified by both methods (P ≤ 0.05 and viability ratio in the lowest 5th percentile) were included in the final high confidence hit list for further analysis and validation.

T98G cells were wet-reverse transfected in 6-well plates using siPORT NeoFX (Applied Biosystems) per the manufacturer's instruction. Briefly, the siRNA was diluted to a working concentration of 2 μmol/L before being used. Cells were plated at a concentration of 300,000 cells per well and a siRNA final concentration of 90 nmol/L. The cells were incubated for 24 hours at 37°C before replacing transfection media with fresh media. Forty-eight hours after transfection cells were trypsinized and seeded for mRNA quantification and MTS cytotoxicity assays as described above. For KD of UNG and UBE3B the following Silencer Select siRNAs from Applied Biosystems were used: UNG: s14679; UBE3B: s40200; Silencer Negative control v2: am4613.

Total RNA was isolated and purified from T98G and LN428 glioblastoma cell lines and normal human astrocytes using Qiazol Lysis Reagent and MiRNeasy spin columns (Qiagen) according to the manufacturer's instructions. RNA quality was assessed using an Agilent Bioanalyzer (Agilent Technologies). RNAs with an RNA integrity number above 9 were used for these studies. Three μg RNA was reverse transcribed into cDNA using the SuperScript III first-Strand synthesis kit (Invitogen, #18080-400) according to the manufacturer's instructions with Oligo(dT)₂₀ primer in a 20 μL reaction. The cDNA library was then diluted 10-fold for qRT-PCR as described above.

The REMBRANDT database was interrogated in February 2012 using the simple search functions for gene expression of UNG and UBE3B with related Kaplan–Meier survival curves (39, 40). The Cancer Genome Atlas GBM database was queried (July 2012) using the gene expression tool for 2-fold greater or less expression of UNG and UBE3B compared with controls. Percentages of GBM overexpressing or underexpressing UNG or UBE3B were recorded.

To identify genes that when silenced confer sensitivity to the alkylating agent TMZ, we conducted a synthetic lethal screen (in triplicate) on a TMZ-resistant GBM cell line (T98G) using TMZ and an siRNA library (Supplementary Fig. S1A). A concentration of 1 mmol/L TMZ was selected because of minimal (∼10%) toxicity measured at 48 hours via an MTS assay. Furthermore, T98G cells were treated with varying concentrations of DMSO to determine the impact of DMSO on cell survival (Supplementary Fig. S1B). No toxicity was observed at 1% DMSO after 48 hours, the% DMSO used for both vehicle and TMZ treatment in the screen.

The transfection conditions used in the screen maximized KD, similar to previous screens using these conditions (20). We first calculated viability ratios (ratio of survival of “siRNA plus TMZ” to “siRNA plus vehicle”) and normalized to vehicle-treated scrambled siRNA, enabling cross plate comparisons. We then statistically analyzed the viability ratios and determined a group of gene targets that sensitized cells to TMZ by conducting a 2-tail sample t test on each gene to determine the effect of siRNA and TMZ on cellular survival. Targeting siRNAs were selected with P ≤ 0.05. The high confidence hit list contained targets that satisfied both criteria: P ≤ 0.05 and a viability ratio in the lowest 5% (Supplementary Fig. S1C). By using these 2 methods, we created a hit list of 172 genes (Supplementary Dataset S1). While most of the viability ratios on the hit list were between 0.4 and 0.65, these genes significantly modulated the toxicity of TMZ, yielding many new potential targets to increase response to alkylation chemotherapy. All hits were analyzed for pathway and network enrichment to determine the potential biologic pathways that modulated alkylation toxicity

The genes contained in the hit list were analyzed with NIH DAVID, Princeton GO term finder and IPA (Ingenuity Systems, www.ingenuity.com) for pathway and network enrichment. The analysis with DAVID and Princeton GO term finder was conducted to search for gene enrichment in molecular functions, cellular compartments, or biologic processes based on the associated gene ontology terms. By focusing on 5,520 druggable genes, we expected a small but presumably more relevant gene enrichment dataset compared with probing the entire human genome as background. Several different thresholds for viability ratio percentiles were initially analyzed for gene enrichment, including 2.5%, 5%, 7.5%, and 10% (Supplementary Datasets S2–S4). The 5% group was selected for a more detailed analysis because it contained many of the gene enrichment groups seen in both the 7.5% and 10% datasets, but was more focused with higher gene enrichment scores.

The enriched genes from the NIH DAVID analysis contained several different groups including DNA repair, response to DNA damage stimulus, cell proliferation, amino acid glycosylation, and biopolymer glycosylation (Table 1). The most enriched groups have specific DNA repair functions including BER activity, hydrolase activity, and DNA N-glycosylase activity. Surprisingly, 4 genes belonging to the 3 most enriched groups were DNA glycosylases that do not recognize alkylation damage (6). Interestingly, these genes (TDG, OGG1, NEIL1, and UNG) are known to recognize and repair multiple types of oxidative DNA damage (Table 2; ref. 6).

There were large overlaps of enriched genes from DAVID and Princeton GO term finder analyses (Supplementary Fig. S2 and Dataset S5). The oxidative DNA glycosylases, which sensitized cells to TMZ, were identified in the most significantly enriched networks determined by both DAVID and Princeton GO term finder (Table 1 and Supplementary Dataset S6). Data were also analyzed through the use of IPA. IPA uses a manually curated interaction database and a right-tailed Fisher's exact test to calculate a P value, determining the probability that each biologic function assigned to that network is because of chance alone. Analysis using IPA identified several of the DNA glycosylases in a statistically significant network, with a network value of 24 (Fig. 1). The network score is the negative logarithm of the P value (P = 10⁻²⁴). This network is very similar to the biologic processes and corresponding genes overrepresented in the DAVID and Princeton GO analysis. Furthermore, IPA returned several highly enriched molecular functions including DNA repair (P = 5.31 × 10⁻⁴) and posttranslational modifications (P = 2.98 × 10⁻³ − 3.12 × 10⁻²). These 2 pathways or nodes were then further analyzed in this study, as described later. However, because of the highly significant enrichment of DNA repair processes in all three programs, genes associated with the DNA Repair pathway category were selected for initial biologic validation.

In our analysis, we discovered that the DNA repair category was enriched for TMZ-sensitizing genes. This category contained four DNA glycosylases (UNG, OGG1, TDG, and NEIL1), each specific for the repair of oxidative DNA damage (6). To determine if other DNA glycosylases also sensitize cells to TMZ and to validate the siRNA synthetic lethal screen results, 7 of the 11 DNA glycosylases were knocked down using a lentiviral system to create stable T98G-derived cell lines. KD of glycosylase mRNA was validated by qRT-PCR (Fig. 2A). The stable cell lines were then tested for sensitization to 1 mmol/L TMZ in a 96-well plate format using a modified MTS assay (21). As confirmation of our siRNA screen analysis, the newly developed UNG-KD cell line was also sensitive to TMZ (Fig. 2B). Furthermore, KD of the DNA glycosylases MYH and MPG also sensitized cells to TMZ (Fig. 2B). Surprisingly, these results suggest that high-dose TMZ treatment may be initiating the formation of reactive oxygen species (ROS), in line with an earlier report in Saccharomyces cerevisiae (41) and mouse embryonic fibroblasts (42). To determine if the high-dose alkylator treatment might be inducing ROS, we used real-time live-cell imaging and fluorescent quantification with the superoxide indicator dihydroethidium (DHE) to determine if the treatment from 1 mmol/L TMZ used in this study yields an increase in superoxide. As shown, the treatment with TMZ induces superoxide within 30 minutes and an observed 12.1 ± 1.8% increase in DHE emission intensity in the first hour that was then stabilized at 16.3 ± 2.1% by 2 hours (Fig. 2C and Fig. 3).

UNG mRNA KD sensitized the T98G cells to TMZ in the validation study, and we next quantitated mRNA KD, UNG protein levels, and DNA repair activity levels to further support this correlation. First, we tested UNG protein expression via immunoblot and determined that UNG protein levels were decreased in T98G/UNG-KD cells as compared with T98G-GFP control cells (Fig. 4A). However, we were interested in ascertaining if the decrease in UNG protein levels affects the DNA repair capacity of the cells, because there is a large functional overlap of UNG with the three DNA glycoslylases SMUG1, TDG, and MBD4 (6). We were concerned that depletion of one of these DNA glycosylases would not have a functional DNA repair defect due to compensation by the other glycosylases. We, therefore, developed a DNA glycosylase molecular beacon assay to quantify the functional loss of uracil removal by UNG KD, essentially as we have described previously for the analysis of MPG activity (22). The assay uses a molecular beacon composed of a single stranded DNA molecule with a 5′ fluorophore (6-FAM) and a 3′ quencher (Dabcyl). The oligonucleotide also contains a uracil lesion, a substrate for UNG. We selected uracil as the lesion because it is removed by UNG, SMUG1, TDG, and MBD4 (6). Therefore, we could determine if UNG-KD impacts uracil removal and if compensation by other DNA-repair proteins occurred. A similar oligonucleotide with a normal base was used as a control substrate. Removal of the uracil lesion by UNG and hydrolysis of the DNA backbone by APE1 results in separation of the 6-FAM fluorophore from the Dabcyl quencher and the increase in 6-FAM fluorescence is proportional to uracil removal (Fig. 4B). The T98G-SCR cell lysate (from cells expressing a scrambled shRNA) incubated with control beacon (Fig. 4C, red circles) had a minimal increase in fluorescence, signifying the control beacon is intact. However, the T98G-SCR lysate incubated with the beacon containing uracil (Fig. 4C, blue squares) exhibited a large increase in fluorescence (17.69-fold at 60 minutes) compared with the control beacon, indicative of robust uracil removal. The T98G/UNG-KD lysates incubated with either the control beacon (Fig. 4C, green diamonds) or the uracil-containing beacon (Fig. 4C, yellow triangles) contained no difference in fluorescence.

These results support our conclusion that the T98G/UNG-KD cells have impaired uracil removal because of UNG KD as compared with the T98G-SCR control, with no evidence for repair compensation by SMUG1, TDG, or MBD4. The scrambled shRNA control was used instead of the GFP control for the activity assay because of the interference of GFP fluorescence with the molecular beacon fluorophore 6-FAM. Although uracil can also be removed by SMUG1, TDG, and MBD4, UNG-KD alone was sufficient to deplete uracil removal activity (6).

After biologic validation, we then compared our T98G siRNA screen results to other alkylation screens conducted in S. cerevisiae and E. coli, using GO molecular function terms (17, 18). This was done by compiling the sensitizing genes from all the screens and linking them to their specific Gene Ontology Biological Process. The list of sensitizing genes from E. coli, S. cerevisiae, and human was then computationally analyzed to identify GO functional categories overrepresented with genes from human and either E. coli or S. cerevisiae (Supplementary Table S1). The molecular functions enriched in all 3 alkylation screen datasets were then visualized using Cytoscape to create a cross species functionome of proteins which modulate toxicity to alkylating agents in S. cerevisiae, E. coli, and human (Fig. 5). This network of proteins has several “GO hubs” of proteins from all 3 species clustered around a conserved biologic process (node) that significantly impacts the survival of correspondingly depleted cells after exposure to alkylating agents. As was observed in previous DNA alkylation screens conducted in S. cerevisiae and E. coli (17, 18), proteins involved in the cellular response to alkylation damage involve many biologic functions, suggesting that synthesis of new DNA, RNA, protein, and fatty acids are all essential for survival following exposure to alkylating agents. Many of the biologic processes identified from our cross-species functionome analysis are conserved, containing multiple proteins from each organism. We have used this approach to define a network of proteins that constitutes a cross-species functionome of evolutionarily important processes essential for survival after alkylation exposure. The functionome analysis affirms that our TMZ screen results from human cells are consistent with previous alkylation studies in other species and allowed us to computationally parse our data to identify rational starting points for more detailed genetic analysis.

On the basis of the functionome network we identified and the validation of the role of DNA repair proteins in response to TMZ, we chose to validate a second node of proteins involved in the response to alkylators. The protein modifications node was chosen due to enrichment in the cross-species analysis, good sensitization in the screening studies and the observation that this node had multiple genes in both human and yeast cells. Human UBE3B and ICMT KD cell lines (UBE3B-KD and ICMT-KD) were generated using lentivirus shRNA vectors as described above. KD levels of mRNA were determined using qRT-PCR (Supplementary Fig. S3A). The UBE3B-KD and ICMT-KD cell lines were then tested for sensitivity to TMZ treatment using a modified MTS assay. The KD of UBE3B and ICMT significantly sensitized the glioblastoma cells to 1 mmol/L TMZ resulting in viability ratios of 0.6 and 0.7, respectively (Fig. 6A), as compared with the control. We then tested each KD cell line for sensitivity to TMZ in a long-term assay similar to a clonogenic cell survival assay. The CyQuant assay uses a highly sensitive fluorescent intercalating DNA dye to determine relative DNA content and cell number nine days after TMZ or vehicle treatment. The UBE3B-KD cell lines were approximately twice as sensitive as the control cells in the CyQuant assay at concentrations as low as 50 μmol/L TMZ (Fig. 6B).

We have previously reported that aim22Δ, lip2Δ, pby1Δ, and stp22Δ cells from S. cerevisiae, which belong to the protein modification node, are sensitive to the alkylating agent MMS (17). In addition, it has been firmly established that mag1Δ cells are sensitive to MMS because of their alkylbase DNA glycosylase deficiency and inability to repair damaged DNA. We assayed all individual knockouts to further validate our previously reported screening results (Fig. 6C) and show decreased growth after MMS treatment for all 5 mutants. On the basis of our functionome results, we reasoned that there would be increased alkylation sensitivity when deficiencies in protein modification and BER were combined. We generated double knockouts in the 4 protein modification associated mutants, using a mag1Δ deletion cassette. We show that there is increased sensitivity in all of the double mutants tested, relative to the individual parent or mag1Δ knockout strains. The protein modification genes correspond to activities involved in the modification of mitochondrial enzymes by the attachment of lipoic acid groups (Lip2), a protein that works with Lip2 (Aim22), a tubulin tyrosine ligase associated with P-bodies (Pby1), and a component of the ESCRT complex that is involved in ubiquitin-dependent sorting of proteins into the endosome (Stp22). We note that Pby1 is homologous to the mouse and human Tsg101 tumor susceptibility genes, which are homologs of ubiquitin-conjugating enzymes implicated in being involved in cell-cycle regulation and genome maintenance. The mag1Δstp22Δ double mutant is the most sensitive to MMS, but there also seems to be a growth defect in these strains on untreated media, suggesting some synthetic lethal interaction outside of alkylation damage. Nonetheless, our results with combined protein modification and BER mutants in budding yeast support the idea that similar dual knockouts in humans may confer increased sensitivity to TMZ.

Although the UNG-KD and UBE3B-KD cell lines were both more sensitive to TMZ when compared with control cells, we were interested in determining if the double KD cells would have an increased TMZ sensitivity, as was seen for the double KO lines in S. cerevisiae (Fig. 6C). To this end, we investigated if simultaneous inhibition of both DNA repair and protein modification genes increased sensitivity compared with either inhibition alone. However, we noticed during long-term passaging of the UBE3B-KD cell line that sensitivity and KD levels diminished, suggesting outgrowth of WT cells in a pooled population.

To prevent the loss of mRNA KD, we returned to using an siRNA transient transfection approach to KD both UNG and UBE3B. We transfected parental T98G cells with a scrambled control siRNA, UNG siRNA, UBE3B siRNA, or UNG plus UBE3B siRNA before determining mRNA levels by qRT-PCR (Supplementary Fig. S3B). Cells were seeded for qRT-PCR and sensitivity determination by MTS assay at the same time. Wells with less than 35% mRNA for either UNG or UBE3B remaining were analyzed for sensitivity to TMZ. We expected the dual KD of UNG and UBE3B to confer at least an additive effect to TMZ sensitivity, suggesting that the sensitivity observed after KD of the proteins resulted from independent mechanisms. However, although UNG and UBE3B KD alone both confer sensitivity to TMZ, when they are simultaneously knocked down there is no significant increase in sensitivity to TMZ below the level of either single KD (Fig. 6D). Thus, TMZ sensitivity to KD of UNG and UBE3B are not independent events, showing a possible epistatic relationship between the UNG and UBE3B genes. Although it is more likely UBE3B has an indirect role in UNG sensitivity, it is interesting to speculate that ubiquitinylation of UNG by UBE3B can modify protein localization, abundance, or activity as it has been suggested that UNG is targeted for ubiquitin-mediated proteolysis during the S-phase of the cell cycle (43, 44).

Expression of UBE3B affected GBM sensitivity to TMZ and may impact response of various cancers to other chemotherapeutic alkylating agents such as cyclophosphamide, chlorabucil, and dacarbazine. With little known on the function or expression of UBE3B in normal and cancerous tissue, we wanted to determine if UBE3B expression levels vary in different cancer cell lines. To determine if UBE3B is differentially expressed in different cancer types we quantified relative expression levels of UBE3B mRNA by qRT-PCR. We discovered that UBE3B mRNA expression varies by approximately 7- to 8-fold in the tumor cell lines tested (not shown). Although UBE3B expression fluctuates in different cancer cell lines, most importantly we sought to determine if UBE3B and UNG levels vary in the LN428 and T98G cell lines when compared with normal human astrocyte controls. We determined relative mRNA levels of UNG and UBE3B in LN428 and T98G cell lines compared with normal human astrocyte controls using qRT-PCR (Fig. 7). UBE3B levels were significantly lower in the LN428 cells than in astrocytes whereas there was a 2-fold increase in UNG expression in both cancer cell lines when compared with human astrocyte controls.

GBMs are the most common and aggressive primary brain tumor; the current standard of care consists of maximum surgical removal, radiation, and TMZ (2–4). Many tumors are refractory to TMZ treatment, resulting in very poor outcomes. Difficulties in treating GBM make it a candidate for drug research, as an increase in therapeutic options is necessary to improve survival rate. Because TMZ clinical efficacy is primarily due to the toxicity of the O⁶-MeG DNA lesion, much of the alkylation damage sensitization research has focused on this lesion. Unfortunately, previous attempts to improve efficacy of O⁶-MeG lesion-mediated cytotoxicity have been futile. Clinical trials to test the effectiveness of MGMT inhibitors have not shown an increase in efficacy or sensitivity (12), particularly in TMZ-resistant GBM (13). Because of the inherent difficulties and resistance of GBM to chemotherapies, any increase in cancer sensitization could greatly impact patient outcomes.

To discover “druggable” targets that contribute to TMZ sensitivity, independent of the O⁶-MeG lesion, we conducted a synthetic lethal siRNA screen against 5,520 genes in a TMZ-resistant cell line that has elevated expression of MGMT (T98G), uncovering enriched biologic processes independent of MGMT- and O⁶-MeG lesion–induced cell death that included 172 genes (Supplementary Dataset S1), many corresponding to similar or overlapping biologic processes as determined by NIH DAVID, Princeton GO term finder, and IPA. After enrichment analysis, increased TMZ-mediated sensitization in cells depleted of UNG (UNG-KD) was validated using a separate shRNA targeting sequence and an independent experimental design. Thus, cell sensitization to alkylating agents with UNG-KD is not because of RNAi off-target effects or artifacts of the screening protocol. We could not create stable KDs of the other DNA glycosylase hits, likely because the KD of OGG1, TDG, and NEIL1 created cells with a growth disadvantage compared with cells expressing normal levels of the corresponding protein.

Although UNG has not been reported to repair alkylation damage, KD of UNG sensitizes T98G cells to TMZ. This is likely the result of the observed TMZ-induced burst of ROS in this study and as reported after treatment with MMS in yeast (41). TMZ may directly alkylate the electron transport chain proteins and mitochondrial DNA, which could impair electron flow through the chain by direct alkylation damage of the protein and decreased functional protein expression due to replication blocking lesions or mutated DNA. It is possible that mitochondrial alkylation damage induces an increase in ROS formation. Mitochondrial ROS production can increase from defects in electron transport chain proteins, such as complex I (45). Cells lacking oxidative DNA glycosylases may be more sensitive to ROS because repair of oxidative lesions is compromised. For example, cells lacking NEIL1 are more sensitive to γ-irradiation (46) and NEIL1^−/− mice have decreased expansion of germinal center B cells (47). Thus, an increase in ROS may create lethal oxidative lesions that UNG (e.g., 5,6-dihydro-2′-deoxyuridine) and other oxidative DNA glycosylases must repair to prevent cell death after alkylation exposure (6). NAD⁺ kinase, which phosphorylates NAD⁺ to form NADP⁺, was also a hit in our screen. Cells keep NADP⁺ in its reduced state, NADPH, to use NADPH's reducing potential in many different pathways including glutathione regeneration for oxidant defense, reduction of RNA to DNA, and synthesis of fatty and amino acids. New synthesis and repair of macromolecules is essential for alkylation survival (18, 36, 48). Support for a potential role of ROS in alkylation sensitivity was also shown by others using mouse embryonic fibroblasts and several human cancer cell lines, which were treated with the alkylating agent N-methyl-N'-nitro-N-nitrosoguanidine (MNNG) and contained greater ROS production from both NADPH oxidase and mitochondria sources (42, 49). In the report by Chiu and colleagues, MNNG cytotoxicity could be completely abrogated by pretreatment with N-acetylcysteine, showing that increased ROS formation may contribute to alkylation toxicity and is a likely source of cytotoxic DNA lesions (42). Cells expressing UNG may be less sensitive to TMZ because the increase in ROS, as we observed after TMZ exposure, can yield cytotoxic oxidative DNA lesions that are substrates for and, therefore, repaired by UNG, promoting cellular survival.

UNG was not the only DNA glycosylase to sensitize cells to alkylation exposure in the validation experiments. AAG (MPG) was included in our analysis because it can be predicted from our work and the work of others that MPG expression levels may modulate TMZ sensitivity (22). Both MPG and MYH KD sensitized T98G cells to TMZ. This was unexpected because MPG did not sensitize in the siRNA screen, perhaps because of lack of KD. However, our results are consistent with previous reports from several labs, as MPG-KD led to either an increase or decrease in cell death after alkylation exposure depending on the system under study (50–53). Furthermore, MPG status alone does not adequately predict response to alkylators, but instead the balance of the entire BER pathway must be investigated to predict sensitivity to DNA damage (22, 30, 54). The ability of MYH-KD to sensitize cells to alkylators was not anticipated, as expression of MYH had been previously shown to promote cell death after alkylator exposure (55). In that study, however, the authors used a clonogenic cell survival assay that interrogates cell death related to the O⁶-MeG lesion because the cells can undergo 2 or more replication cycles (55). In contrast, in our initial validation experiments, cell survival was determined after 48 hours post TMZ treatment, too short for 2 replication cycles to occur (56). Therefore, the role of MYH in O⁶-MeG–mediated cell death is likely very different than its role in cell survival after 48 hours of exposure to TMZ as described in this study. MYH is primarily known for the removal of the mismatched A opposite 8-oxoG lesions (57, 58) and MYH deficiency does not sensitize cells to H₂O₂, IR, or cis-platinum (6). Therefore, the sensitization of MYH depletion to TMZ treatment may also be the result of the observed increase in ROS and the expected accumulation of cytotoxic ROS-induced lesions. Finally, we focused our efforts on UNG, as it is elevated in the glioma lines under study (see Fig. 7) and in clinical samples as measured by the TCGA study (http://www.nextbio.com/b/study/ov.nb?id=38365) and is one of several DNA glycosylases involved in the repair of ROS induced DNA lesions (6).

After biologic validation was completed, the results were compared with similar alkylator sensitivity screens done in S. cerevisiae and E. coli (17, 18). Many of the same biologic processes significantly enriched in the human screen were enriched in both screens from yeast and bacteria. There are a diverse collection of processes, ranging from DNA repair to modulation of transcription, translation, protein modifications, and fatty acid biosynthesis that seem to be essential for survival after alkylation exposure. The many processes necessary for cellular survival after alkylation exposure likely stem from the alkylator's ability to modify all macromolecules in a cell, thereby requiring repair of not only DNA, but also of RNA, proteins and fatty acids. In the human siRNA screen, these data suggests a novel epistatic relationship between UNG and UBE3B. This is likely an indirect interaction with downstream targets of UBE3B modulating TMZ cytotoxicity through an UNG-dependent pathway. A direct functional interaction between UBE3B and UNG may exist, as UNG is a known target of ubiquitinylation (44, 59, 60), although the data described herein would not implicate UBE3B in the cell-cycle–regulated proteolysis of UNG.

The analysis of the 3 screens revealed that many biologic processes modulate survival after alkylation damage. The greatest alkylator toxicity in yeast cells was generated by dual inhibition (KO) of both BER and the protein modification processes, suggesting that simultaneous inhibition may be required to achieve a clinically significant sensitization for chemotherapies, provided this approach can provide tumor selectivity. In that regard, we evaluated the range of UNG expression levels in primary patient tumors by interrogating the Cancer Genome Atlas (TCGA) dataset on GBM (http://www.nextbio.com/b/study/ov.nb?id=38365). UNG expression is 2-fold greater in approximately 14% of the GBMs studied in the TCGA. The TCGA dataset includes 424 glioblastoma primary tumor samples as compared with 10 normal brain samples and analyzed using the Agilent 244K G4502A array platform. Of these samples, 58 had tumor/normal ratio of UNG expression above 2 (14%). In addition, this dataset also indicates that 49% of the samples express UNG at a level that is 1.5-fold greater in the tumor as compared with normal brain. Interestingly, there are no samples with a 2-fold decrease in UNG expression in the TCGA database. We also queried the Rembrandt database for UNG gene expression data and found that greater than 33% of GBMs contained 2-fold or greater expression of UNG (40). These results, from a separate database, reinforced our hypothesis that expression levels of UNG varied significantly in tumors. The directionality of the changes were consistent and it seems that overexpression of UNG is beneficial for GBM tumor growth or survival, with less than 1% of tumors containing a 2-fold decrease in UNG expression in the REMBRANDT database. This trend also extended to other brain tumor types such as astrocytomas, which had greater than 14% of tested tumors containing 2-fold greater expression of UNG and no samples with a 2-fold downregulation of UNG (40).

Knowing that the expression levels of UNG fluctuate in primary tumor samples, we asked if expression levels of UNG and UBE3B correlated with response to therapy or survival. We would predict the increase in UNG expression levels would correlate with reduced survival due to the ability of the cells to repair their DNA more efficiently, especially TMZ-induced lesions. In support of our prediction, we queried the REMBRANDT database for clinical outcomes related to UNG overexpression compared with intermediate expression in all glioma: 2-fold overexpression of UNG correlated with a decrease in the probability of survival compared with intermediate expression of UNG, with a log-rank P value, calculated using the Mantel–Haenszel procedure, of 1.0177 × 10⁻⁴. This correlation matches our prediction based on our cell culture data, suggesting that increased expression of UNG negatively impacts brain tumor patient survival, possibly from increased resistance to chemotherapeutic treatment. There was no significant relationship about UBE3B and clinical outcomes in REMBRANDT but it was observed that the frequency of a 2-fold decrease in UBE3B expression was 1.1% in the REMBRANDT cohort and 19% in the Cancer Genome Database (data not shown). However, it was recently reported that UBE3B may be a candidate autism gene, as a homozygous mutation (R922C) in UBE3B was identified in affected monozygotic twins and heterozygous in the parents and unaffected siblings (61). On the basis of our studies herein, it is interesting to speculate that the role of UBE3B in autism onset may be related to the response to genotoxic stress.

In summary, these studies suggest that alkylation resistance mechanisms are evolutionarily conserved. The collection of conserved biologic processes in E. coli, S. cerevisiae, and humans composes an alkylation functionome that includes many novel proteins not previously thought to impact alkylation resistance (Fig. 8). We can begin to appreciate the multiple processes that are required for cellular survival after alkylation damage and form an unbiased approach to discover targets for adjuvant chemotherapy. This is an important next step, so as to determine if genes that were not tested in the screen, but share conserved biologic pathways, are also possible targets to enhance TMZ response. Because the analysis has also been enriched for the most essential pathways and genes, there may be functional overlap of the conserved genes and pathways. The highly conserved nature of these biologic processes lends mechanistic insight into potential gene and pathway interactions. Determining the relationships between these genes and biologic processes is important to determine pathway redundancy. Because of the plethora of resistance mechanisms, it is possible that depletion or inhibition of one gene in a conserved biologic process cannot overcome the drug resistance because of compensation by another gene or biologic process. By investigating multiple pathways and processes, we can determine important interactions that promote tumor survival to be targeted for improved chemotherapy response.

R.W. Sobol is a consultant/Advisory board member in Trevigen, Inc. No potential conflicts of interest were disclosed by the other authors.

Conception and design: D. Svilar, P.R. McDonald, J.S. Lazo, I.F. Pollack, T.J. Begley, R.W. Sobol

Development of methodology: D. Svilar, M. Dyavaiah, J.-b. Tang, P.R. McDonald, C.M. St. Croix, J.S. Lazo, T.J. Begley, R.W. Sobol

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D. Svilar, M. Dyavaiah, J. Li, P.R. McDonald, A. Braganza, X.-h. Wang, S. Maniar, C.M. St. Croix, J.S. Lazo, T.J. Begley, R.W. Sobol

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D. Svilar, M. Dyavaiah, J. Li, P.R. McDonald, T.Y. Shun, S. Maniar, C.M. St. Croix, J.S. Lazo, T.J. Begley, R.W. Sobol

Writing, review, and/or revision of the manuscript: D. Svilar, M. Dyavaiah, P.R. McDonald, T.Y. Shun, J.S. Lazo, I.F. Pollack, R.W. Sobol

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.R. Brown, P.R. McDonald, S. Maniar, R.W. Sobol

Study supervision: R.W. Sobol

This work was supported by grants from the National Brain Tumor Society and NIH [GM087798; CA148629; and ES019498] to R.W. Sobol and T.J. Begley (ES01225101) and a NYSTAR James Watson Award to T.J. Begley. Support was also provided by the University of Pittsburgh Department of Pharmacology and Chemical Biology with a Molecular Pharmacology Fellowship to D. Svilar. Support for the UPCI Lentiviral Facility was provided by the Cancer Center Support Grant from the NIH (CA047904).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.