Introduction

Cellular redox control and response to oxidative damage is vital for cell survival1,2,3. An intricate and widespread web of proteins are involved in conferring protection to cells against oxidative stress1,2,3. PARP1, PARP2 and PARP3 enzymes that belong to the PARP enzyme family, can readily respond to oxidative-stress induced DNA strand breaks and induce different forms of cell death4,5,6,7. The majority of cellular PARP activity can be attributed to PARP1 and PARP2, of which PARP1 is responsible for the majority8,9,10. PARP1 and PARP2 can introduce poly(ADP-ribose) (PAR) chains onto target proteins10,11,12,13,14,15,16,17 that can subsequently change their biological activity by modifying the biochemical activity or cellular localization18. PARP1 and PARP2 share multiple functions, encompassing their involvement in DNA repair19,20 or transcriptional regulation19,21. PARP2 is related to oxidative stress-associated pathologies and has pathological roles in these scenarios8,22,23,24,25,26. We recently showed that the silencing of PARP2 induces oxidative, but not nitrosative stress in different cellular models27. The source of the reactive oxygen species (ROS) is partly mitochondrial, however there are still unidentified sources27. We set out to identify the source of the remainder of reactive species in PARP2-silenced cells.

Multiple PARP inhibitors were developed for the pharmacological modulation of cellular PARP activity28,29,30. These inhibitors inhibit multiple PARP enzymes31. A subset of these inhibitors is now available in the clinical practice28,29,30 of which olaparib (Lynparza®) was the first to be registered32,33 Olaparib is an anticancer drug registered for the treatment of ovarian cancer, nevertheless, clinical studies are carried out to encompass the treatment of breast, prostate and pancreatic cancer28,29,30. Although in ovarian, breast and pancreatic cancer the use of olaparib is limited to the patients carrying BRCA mutations, it appears that the deficiency of other DNA repair factors may also sensitize cells for pharmacological PARP inhibition.

Oxidative stress may be elicited by a decrease in the activity or expression of antioxidant defense systems. Nuclear factor erythroid 2 (NFE2)-related factor 2 (Nfe2l2, or NRF2) is a transcription factor that has predominant role in regulating the expression of a set of antioxidant genes34,35,36. NRF2 forms a complex with the Kelch Like-ECH-Associated Protein 1 (KEAP1) that co-localizes NRF2 to Cul-3-RBx1-E3 (Cullin-3-ring box-1 E3) ubiquitin ligase complex for degradation37,38,39,40. However, KEAP1 has several redox-sensitive cysteine residues that are oxidized upon oxidative stress41,42 and, hence, KEAP1 releases NRF2, preventing its degradation43,44,45,46. NRF2 then translocates to the nucleus and dimerizes with other transcription factors, for example the members of the Maf family47,48. The heterodimers bind to EpRE/ARE-response elements in gene promoters and drive their transcription47,49. There are over 200 NRF2-dependent genes in humans, the majority of which are enzymes that detoxify cellular oxidative damage35,50,51,52.

PARP1 was shown to regulate NRF2-dependent transcription through binding to Maf proteins and, through that, modulate the DNA-binding capacity of NRF253,54. Furthermore, there are important overlaps in the function of PARP enzymes and NRF2 in multiple organ systems, for example in the cardiovascular system3,52,55,56,57. Given the structural and functional overlap between PARP1 and PARP219,20 we set out to assess whether there are connections between NRF2 and PARP2 and whether the deregulation of NRF2 could be the source of oxidative stress upon PARP2 silencing.

Results

PARP2 PARylates NRF2 and regulates its intracellular localization

PARP2 has important metabolic roles and modulates mitochondrial metabolism in multiple metabolically active tissues19,20,58,59, among these, skeletal muscle23,60,61, therefore, we chose C2C12 cells as cellular model for our studies23. To assess changes to the function of NRF2 upon PARP2 silencing, we applied a C2C12 cell line pair in which PARP2 was downregulated using an shRNA approach (shPARP2 C2C12) and its isogenic counterpartner expressing a sequence that has no target in murine or human cells (scPARP2 C2C12)23,62. We verified the downregulation of PARP2 mRNA and protein expression in the shPARP2 C2C12 cell line as compared to the scPARP2 C2C12 cell line (Fig. 1A,B). Next, we showed that the silencing of PARP2 does not affect the mRNA expression of NRF2 (Fig. 1C). We also assessed the expression of the NRF2 protein using two different antibodies from different sources, as in the literature there is an ambiguity in the apparent molecular mass of NRF2 (68–130 kDa) and different antibodies detect different NRF2 bands63,64,65. We applied antibodies that were validated in a previous study65. Silencing of PARP2 did not change the protein expression of NRF2 significantly (Fig. 1D). Finally, we showed that the mRNA expression of other members of the NRF2 system (KEAP1, MAFG, MAFK) is not different between scPARP and shPARP2 C2C12 cells (Fig. 1E,F). We were unable to detect the mRNA transcript of MAFF.

Figure 1
figure 1

The silecing of PARP2 does not interfere with the expression of the elements of the NRF2 system. (A,B) 200 000 cells from scPARP2 and shPARP2 C2C12 cells were seeded into a 6-well plate and the expression of PARP2 was determined by (A) RT-qPCR (n = 3) and (B) Western blotting (n = 3). (C,D) 200 000 cells from scPARP2 and shPARP2 C2C12 cells were seeded into a 6-well plate and the expression of NRF2 was determined by (C) RT-qPCR (n = 6) and (D) Western blotting (n = 3) using the antibodies indicated. (E,F) 200 000 cells from scPARP2 and shPARP2 C2C12 cells were seeded into a 6-well plate and the expression of (E) KEAP1 and (F) the indicated MAF genes was determined by RT-qPCR (n = 3). Numerical values are presented as the average ± SD. Statistical significance was determined using paired, two-tailed Student’s t-test. ***Represent statistically significant differences between the scPARP2 and shPARP2 C2C12 cells at p < 0.001. n.s.—not significant.

Full size image

Next, we assessed the localization of NRF2. Silencing of PARP2 drastically decreased the nuclear localization of NRF2 (Fig. 2) suggesting that PARP2 is needed for the nuclear translocation of NRF2. PARP enzymes, among them PARP2, can produce large, branched and negatively charged polymers on target proteins10 that can modulate their localization and biochemical activity66. We hypothesized that PARP2 can PARylate NRF2. Indeed, when NRF2 was immunoprecipitated from cellular extracts we were able to detect PAR chains on the large molecular weight form of NRF2 protein (> 130 kDa; Fig. 3A). PAR chains were largely reduced in shPARP2 cells (Fig. 3A) suggesting a dominant role for PARP2 in PARylating NRF2. In good agreement with that, pharmacological PARP inhibition by PJ34 reduced the nuclear localizaiton of NRF2 in control cells, while PARP inhibitors did not affect NRF2 localization in shPARP2 cells (Fig. 3B).

Figure 2
figure 2

Silencing of PARP2 reduces the nuclear fraction of NRF2. 70 000 cells from scPARP2 and shPARP2 C2C12 cells were seeded into a 24-well plate on glass coverslips. The cells were stained with NRF2 specific antibody and nuclei were visualized using DAPI (n = 3). Images were analyzed using ImageJ software (50/50 cells were measured). For each biological replicate an average was generated that is depicted on the graph. Representative immunofluorescence images are presented in the figure. Numerical values are presented as the average ± SD. Statistical significance was determined using paired, two-tailed Student’s t-test. *Or **statistically significant differences between the scPARP2 and shPARP2 C2C12 cells at p < 0.05 or p < 0.01, respectively.

Full size image
Figure 3
figure 3

The PARylation of NRF2 by PARP2 is responsible for the nuclear localization of NRF2. (A) 200 000 cells from scPARP2 and shPARP2 C2C12 cells were seeded into a 6-well plate and PARylation level of NRF2 was determined by immunoprecipitating NRF2 and embranes were probed with NRF2 or PAR antibody in Western blotting (n = 2). (B) 40 000 cells from scPARP2 and shPARP2 C2C12 cells were seeded into a 24-well plate on glass coverslips and treated with 3 µM PJ34 for 24 h. After treatment, the cells were stained with NRF2 specific antibody and nuclei were visualized using DAPI (n = 3). Images were analyzed using ImageJ software (measured cells: 50 from each group). Representative Western blot images are presented in (A) and representative immunofluorescence images are presented in (B). Numerical values are presented as the average ± SD. Statistical significance was determined using ANOVA followed by Tukey’s post-hoc test. **Statistically significant differences between the scPARP2 and shPARP2 C2C12 cells at p < 0.01.

Full size image

To verify these findings with an alternative method, we performed cell fractionation based on the Sanders salt fractionation tachnique67. Fractionation was verified using antibodies against glyceraldehyde-3-phosphate-dehydrogenase (GAPDH) and Lamin A/C. GAPDH was detected only in the cytosol, while Lamin A/C was detected only in the nucleus (Fig. 4), verifying the specificity of the fractionation experiment. NRF2 signal was detected in both the cytosolic and the nuclear fraction, however, the apparent molecular weight of the signal was different in the compartments. The apparent molecular weight was higher in the nucleus (~ 110 kDa and ~ 160 kDa) than in the cytosol (~ 65 kDa and ~ 90 kDa) (Fig. 4). The ~ 110 kDa form was mostly detected in the cytosol, while the smaller isoforms were rather specific for the cytosol (Fig. 4). The cytosolic NRF2 signal increases, while the nuclear NRF2 signal decreases when PARP2 is silenced or olaparib is applied that points towards decreases to the nucleus-to-cytoplasm ratio of NRF2.

Figure 4
figure 4

Lower nuclear NRF2 is detected in shPARP2 C2C12 cells. 1.5 × 106 sc/shPARP2 C2C12 were seeded in Petri dishes; a set of cells were treated with 1 µM olaparib for 24 h. Then cells were harvested by scraping and cells were fractionated. Protein from the indicated fractions were separated by SDS-PAGE and the separated proteins were blotted. Blots were developed with the antibodies indicated.

Full size image

Silencing of PARP2 leads to the repression of NRF2-dependent genes

We turned to decipher the possible consequences of the interaction between PARP2 and NRF2. To that end we reanalyzed a microarray experiment performed on scPARP2 and shPARP2 C2C12 cells60 and another, assessing gene expression changes between scPARP2 and shPARP2 HepG2 cells68. We assessed changes to the expression of genes with antioxidant function as described in the Materials and Methods section. Silencing of PARP2 reduced the expression of a set of antioxidant genes roughly comprising the half of all antioxidant genes assessed in both settings, while the remainder of the antioxidant genes were upregulated (Fig. 5A,B). Changes were in the same magnitude as in the case of the RT-qPCR study (± 1.5–twofold).

Figure 5
figure 5

Silencing of PARP2 reorganizes the expression of antioxidant genes. (A,B) The expression of genes involved in antioxidant defense were assessed by the reanalysis of microarrays assessing (A) scPARP2 and shPARP2 C2C12 and (B) scPARP2 and shPARP2 HepG2 cells. (C) The differentially expressed redox genes identified upon the reanalysis of the microarrays were grouped as a function of the expression changes and were depicted in Venn-diagrams. In black, the NRF2-independent genes, in red the NRF2-dependent genes with similar changes between the microarray experiments. In blue, the NRF2-dependent genes with different expression between the microarray experiments.

Full size image

We grouped the up and downregulated antioxidant genes in both cell lines, assessed which genes are regulated anonimously between the cell lines and identified NRF2-dependent genes through literature search. Those genes were considered NRF2-dependent the expression of which was regulated by NRF2, those that are NRF2 partners or only modulated NRF2 activity be interfering with redox homeostasis (e.g. antioxidants as albumin) were not considered NRF2-dependent. Among the differentially regulated genes catalase (CAT)69, fatty acid binding protein 1 (FABP1)70, Family With Sequence Similarity 213, Member A (FAM213A)71, metallothionein (MT(3))72, peroxiredoxin 1 (PRDX1)35, peroxiredoxin 2 (PRDX2)73, peroxiredoxin 3 (PRDX3)74,75, superoxide dismutase 1 (SOD1)76, superoxide dismutase 2 (SOD2)77, superoxide dismutase 3 (SOD3)78, sulforedoxin 1 (SRXN1)35, S100A879, S100A979 and Tumor protein p53-inducible nuclear protein 1 (TRP53INP1)80 were identified as NRF2-dependent genes. Out of these, CAT, FAM213A, FABP1, SOD1, S100A9 and TRP53INP1 were downregulated both in shPARP2 C2C12 cells and shPARP2 HepG2 cells, while PRDX1, PRDX3 and SRXN1 were downregulated only in C2C12 cells (Fig. 5C). The expression of PRDX2 was differentially regulated between the two cell models (Fig. 5C). S100A8, SOD2 and SOD3 were upregulated in shPARP2 C2C12 cells and MT3 in shPARP2 HepG2 cells (Fig. 5C). Taken together, out of the the identified NRF2-dependent genes 9 were downregulated and 4 were upregulated.

Discussion

PARP2 is a member of the PARP family and has multiple functions overlapping with PARP1, however, clearly has individual functions independent of PARP119,20. In this study we showed that PARP2 activity regulates NRF2 location and changes the pattern of NRF2-dependent gene expression. PARP2 can poly(ADP-ribosyl)ate NRF2 and tether it to the nucleus that is needed for the appropriate activity of NRF2. In contrast, when PARP2-dependent PARylation is absent, the cellular localization pattern of NRF2 changes, a larger fraction of NRF2 can be found in the cytoplasm. This observation suggests that NRF2 can over time translocate to the cytoplasm more easily (Fig. 6).

Figure 6
figure 6

PARP2 PARylates NRF2 and traps it in the nucleus.

Full size image

We observed changes to the expression of antioxidant genes upon the silencing of PARP2 and identified changes to the expression of 13 NRF2-dependent genes. Out of the 13 NRF2-dependent genes, 9 were suppressed, while 4 were upregulated in the absence of PARP2. Changes to gene expression therefore is linked to exhanges to cellular distribution of NRF2, however may not be the only explanation. PARylated NRF2 may have different transactivation properties than the non-PARylated form influencing transcription initiation efficiency and, subsequently, the NRF2-dependent transcriptome.

PARP enzymes were long-known to be involved in regulating gene expression81,82,83,84, and along that, the data on the involvement of PARP2 in the regulation of gene expression is accumulating8,23,60,61,62,68,85,86,87,88,89,90. In this study we identified a new PARP2-mediated transcription factor, NRF2. Prior, PARP2 was known to act as a transcriptional cofactor of nuclear receptors91, to modulate the mRNA expression of transcription factors (e.g. SIRT1 or SREBPs)8,23,60,61,68 or to impact on coenzymes of transcription factors (e.g. NAD + for SIRT1)61. However, this is the first time to show that PARP2 can regulate the activity of a transcription factor by affecting their cytosolic location. As pharmacological PARP inhibition reduces the nuclear fraction of NRF2, we suggest that the activity of PARP2 is necessary for the nuclear translocation or nuclear trapping of NRF2. On a broader scale we can deduct that NRF2 may be responsive to nuclear PARylation and, hence, act as a PARylation sensor. It is of note that PARP1 cannot poly(ADP-ribosyl)ate NRF253, therefore, the pharmacological inhibition of PARP1 is unlikely to explain the cytosolic trapping/translocation of NRF2. The large negative charge of the PAR polymers may be a likely explanation for the restriction of active NRF2 to the nucleus. Currently interactions between PARP1 and NRF2 were shown in breast cancer cells53 and were suggested in inflammatory processes92 and bone formation93.

What could be the physiological relevance of our findings? As PARP2 can be activated by DNA strand breaks7,9,10,94,95,96,97,98,99 that are frequently induced by oxidative stress, hence, PARP2 activation in these situations can enhance and support NRF2 activation and antioxidant defense. Furthermore, We have previously detected increased oxidative stress in PARP2-silenced cells27 a portion of which stems from the mitochondria. Apparently, changes to antioxidant defense and the deregulation of NRF2 activity may also contribute to oxidative stress in the absence of PARP2.

Our findings may have implications in the PARP2-elicited adverse effects of pharmacological PARP inhibition86,87. Furthermore, our previous report27 provided evidence that the production of reactive species is needed for mitochondrial biogenesis in the absence of PARP2. In addition, changes to cellular redox balance play role in inflammation, cardiovascular and neoplastic diseases or aging, in which NRF252,100,101,102,103 and PARP enzymes1,2,3,28,29,104,105,106,107,108,109. These processes are all characterized by PARP activation, more importantly, PARP2 activation was evidenced8,89,110,111, suggesting a broad applicability for our findings.

Methods

Chemicals

All chemicals, including PJ34, were from Sigma-Aldrich (St. Louis, MO, USA) unless stated otherwise.

Cell culture

PARP2-silenced C2C12 cells were described in23. The 5′-AAGATGATGCCCAGAGGAACT-3′ was used as PARP2-specific interfering sequence, while the 5′-TTCGGGGAACAAACGTGCAAC-3′ sequence was used as a non-specific control62. C2C12 cells were cultured in DMEM (cat. no. D6429, Sigma-Aldrich, St. Louis, MO, USA) containing 10% FBS, 1% penicillin/streptomycin and 2 mM L-glutamine at 37 °C with 5% CO2. PARP2 was silenced by specific shRNA that was maintained over extended periods by selection under 2.5 µg/ml of puromycin. Cells containing the control, non-specific sequence are termed scPARP2, while those containing the PARP2-specific shRNA are termed shPARP2 cells.

Immunofluorescence

Cells were seeded on glass coverslips into a 24-well plate and treated with 3 µM PJ34 for 24 h. After treatment cells were washed with PBS, fixed with 4% paraformaldehyde for 10 min at 37 °C and permeabilized with 1% Triton X-100 in PBS for 10 min. Between each steps, cells were washed three times with PBS. Cells were blocked with 1% BSA in PBS for 1 h at room temperature and incubated with primary antibody diluted in blocking buffer for overnight at 4 °C. Coverslips were washed three times with PBS and cells were incubated with the secondary antibody for 1 h at room temperature. Cell nuclei were visualized with DAPI (NucBlue Fixed Cell ReadyProbes Reagent, cat. no. R37606, Invitrogen).

Confocal images were acquired with Leica TCS SP8 confocal microscope (Leica) and LAS X 3.5.5.19976 software (Leica). Nonspecific binding of the secondary antibodies was checked in control experiments (data not shown). Antibodies are listed in Table 1.

Table 1 Antibodies used in immunofluorescence experiments.
Full size table

Processed images were analyzed using ImageJ software, NRF2 signal intensity nucleus/cell was determined.

Total RNA preparation and reverse transcription-coupled quantitative PCR (RT-qPCR)

Total RNA from cells was prepared using TRIzol reagent (cat. no. 15596018, Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s instructions. DNase treatment of total RNA samples was performed using 2U DNaseI (cat.no A2222, Thermo Fischer Scientific, Waltham, MA, USA) per 10 µg RNA. 2 µg of total RNA were used for reverse transcription using High-Capacity cDNA Reverse Transcription Kit (cat. no. 4368814, Applied Biosystems, Foster City, CA, USA) according to the manufacturer’s instructions. The reverse transcription reaction mixtures were supplemented with RNase inhibitor (cat.no 8080119, Applied Biosystems, Foster City, CA, USA). The qPCR reactions were performed with qPCRBIO SyGreen Lo-ROX Supermix (PCR Biosystems Ltd., London, UK) using 20 ng diluted cDNA with the appropriate primers at 500 nM final concentration in 10 µl total volume. Amplification reactions were performed using a Light-Cycler 480 system (Roche Applied Science, Basel, Switzerland). Cycling conditions are given in Table 2. Expression was normalized to the geometric mean of murine 36B4 and cyclophilin values. Primers are listed in Table 3.

Table 2 qPCR cycling conditions.
Full size table
Table 3 Primer sets used in the study. Primers bore no modification.
Full size table

RNA integrity was not verified. Specificity was assessed by assessing the melting curve, those well that contained more than one peak were omitted. Results were assessed using the software of the Lighcycler II instrument (version 1.5.0) Outliers were not assessed and no values were rejected. CT was identified using the second derivative method provided by the Lighcycler II software. Three technical replicates were used. We reported the results according to the MIQE guidelines112.

Immunoprecipitation

Cells were washed with PBS and lysed in lysis buffer (50 mM Tris, pH 8, 150 mM NaCl, 1% Triton X-100, 0.5% sodium deoxycholate, 0.1% SDS, 1 mM EDTA, 1 mM Na3VO4, 1 mM NaF, 1 mM PMSF, protease inhibitor cocktail) on ice. 20–20 µl of Protein A-Protein G Sepharose beads were resuspended in immunoprecipitation buffer (20 mM Tris, 150 mM NaCl, 2 mM EDTA, 1% NP-40, protease inhibitor cocktail). Samples were precleared for 1 h, the primary antibody was added to the precleared samples for 2 h, then beads were added to the samples for overnight at 4 °C with gentle rotation. The resulting immunoprecipitates were washed with TBSTween, boiled with 50 µl 5 × SDS sample buffer (310 mM Tris–HCl, pH 6.8, 50% glycerol, 10% SDS, 100 mM DTT, 0.01% bromophenol blue) and 2-mercaptoethanol and used for Western blotting.

Cell fractionation

1.5 × 106 sc/shPARP2 C2C12 cells were grown on 10 cm Petri dishes to 80% confluency, washed and scraped in ice-cold PBS, then pelleted by centrifugation at 4 °C at 1,500 × g for 3 min. After removing PBS, cells were homogenized in 1800 μl buffer A (10 mM Hepes pH 7.9, 10 mM KCl, 0.1 mM EDTA, 0.1 mM EGTA, 0.5% NP40, 1 mM DTT, 1 × protease inhibitor cocktail, 1 mM NaF, 1 mM PMSF, 1 mM Na3VO4) by suspending with a 26-gauge needle 10 times using a syringe. The lysate was vortexed for 15 s, then centrifuged at 4 °C at 16,000 × g for 1 min. The resulting supernatant was used as cytosolic fraction. The pellet was then washed in 1000 μl buffer A by passing through a 26G needle 10 times. After centrifugation (4 °C at 16,000 × g for 1 min) 1200 μl buffer B (20 mM Hepes pH 7.9, 420 mM NaCl, 0.5 mM EDTA, 0.5 mM EGTA, 1 mM DTT, 1 × protease inhibitor cocktail, 1 mM NaF, 1 mM PMSF, 1 mM Na3VO4) was added to the pellet and it was incubated for 25 min on ice, while being vortexed several times. Then the pellet was sonicated (3 times for 10 s), centrifuged (4 °C at 12 000 × g for 10 min) and used as nuclear fraction.

SDS-PAGE and western blotting

Cells were washed with PBS, collected and lysed in lysis buffer (50 mM Tris, pH 8, 150 mM NaCl, 1% Triton X-100, 0.5% sodium deoxycholate, 0.1% SDS, 1 mM EDTA, 1 mM Na3VO4, 1 mM NaF, 1 mM PMSF, protease inhibitor cocktail) on ice and boiled with 5 × SDS sample buffer (310 mM Tris–HCl, pH 6.8, 50% glycerol, 10% SDS, 100 mM DTT, 0.01% bromophenol blue) and 2-mercaptoethanol. Protein extracts were separated by SDS-PAGE on 8% acrylamide gels and transferred onto nitrocellulose membranes. Membranes were blocked with 5% BSA in TBSTween for 1 h at room temperature and incubated with primary antibodies overnight at 4 °C. Membranes were probed with the respective peroxidase-conjugated secondary antibody. Signals were visualized by enhanced chemiluminescence reaction and captured by ChemiDoc Touch Imaging System (Bio-Rad, Hercules, CA, USA). Bands were quantified by densitometry using the ImageJ software113 and densitometry data were analyzed by statistical methods. Primary and secondary antibodies are listed in Table 4. The original description of the method can be found in114.

Table 4 Antibodies used in Western blot experiments.
Full size table

Microarray reanalysis

The expression patterns were analyzed on samples from two microarray experiments with PARP-2 silencing: one was conducted on human HepG2 cells68 (GEO accession number: GSE43981), the other was on mouse C2C12 cells60 (GEO accession number: GSE108737).

For both mouse and human, four groups of genes have been constructed from the GO database115,116 (www.geneontology.com). To obtain the expression of antioxidant genes, genes were extracted based on their labeling as ‘antioxidant’. The superoxide dismutase genes (SOD1, SOD2), originally labelled as ‘reactive oxygen species biosynthetic process’, were added to the antioxidant lists as the products of these genes are known antioxidants117.

We used the GEOquery R package118 to retrieve and parse the raw data from both experiments. Both experiments had 3–3 replicates of controls and PARP-2 silenced samples. The limma package119 was used to calculate the log2 expression values and to compare the controls and the silenced samples, and calculate the differential expressions. A least squares linear model was fit, followed by an empirical Bayes model and Benjamini & Hochberg correction for multiple comparisons to get the fold change, moderated F and t-statistics, log-odds of differential expression and the corrected p-values for each gene or gene set represented by the Affymetrix IDs. These complete lists were filtered to get only the genes contained in the predefined antioxidant list.

The Affymetrix IDs were converted to gene symbols, taking into consideration that sometimes an Affymetrix probe represents multiple genes, and sometimes a gene is represented by multiple probes. In the former case, the multiple gene symbols separated with the symbol “///” are shown in a single line corresponding to the probe; in the latter case, the identical gene symbols across multiple lines are indexed as “0.1”, “0.2” and so on.

Heatmaps were generated with the limma package as well, using the log2 expression values of each sample.

Statistical analysis

Statistical analyses, except for the microarray reanalysis, were performed with Prism software v8 (GraphPad Software, San Diego, CA, USA). All groups were checked for normal distribution using the D’Agostino-Pearson method. For comparing two groups, two-sided Student’s tests were applied. For multiple comparisons, one-way ANOVA tests were conducted, followed by Tukey’s post hoc test (to compare all possible combinations). The levels of significance are indicated in figure captions. The n number denotes the number of the biological replicates.