*2.4. Transfection with PEBG-GST and PEBG-GST-HO1*

PC3 cells were transiently transfected with the HO-1 expression plasmid (pEBG-GST-HO-1) or the empty vector as control (pEBG-GST). Each 10 cm diameter plate was transfected using 10 μg of plasmid and 20 μL of polyethylene glycol (PEI) (Sigma-Aldrich, Gillingham, UK) in a final volume of 200 μL of RPMI 1640 culture medium. Transfections were performed on plates with 3 mL of RPMI without FBS or antibiotics. After 5 h of transfection, the culture medium was replaced by complete RPMI with 10% *v*/*v* FBS and antibiotics at the previously mentioned concentrations. For the proteomics tests, 40 plates of 10 cm in diameter of each experimental condition (GST-HO-1 vs. GST) were transfected and incubated for 48 h to carry out the different experiments.

#### *2.5. GST Immunoprecipitation Strategy*

48 h after transfection with pEBG-GST-HO-1 or the empty vector pEBG-GST and subsequent treatment with H2O2, proteins were extracted using a low concentration NaCl buffer (20 mM Tris, 150 mM NaCl, 5 mM MgCl2, 0.5% NP40, pH 7.5) to avoid disruption of protein-protein interactions. Protein extracts were incubated for 2 h at 4 ◦C with beads coated with glutathione-S-agarose.

#### *2.6. Separation of Peptides and Mass Spectrometry Analysis*

Recombinant GST-HO-1 protein complexes were reduced (200 mM DTT), alkylated (200 mM iodoacetamide), and digested with trypsin in-solution overnight, using an estimated 1:30 enzyme to substrate ratio. The peptides were desalted and concentrated in a C18 resin (Zip-Tips, Waters Technologies Corporation, Milford, MA, USA) before analysis by LC ESI-MS/MS at the Center for Metabolomics and Mass Spectrometry (The Scripps Research Institute, La Jolla, CA, USA). Peptides were separated by reverse-phase chromatography before mass spectrometry analysis using the following method: nanoelectrospray capillary column tips were made in-house by using a P-100 laser puller (Sutter Instruments, Novato, CA, USA). The columns were packed with Zorbax SB-C18 stationary phase (Agilent Technologies, Santa Clara, CA, USA) purchased in bulk (5 mm particles, with a 15 cm length and a 75 mm inner diameter). The reverse-phase gradient separation was performed by using water and acetonitrile (0.1% formic acid) as the mobile phases. The gradient consisted of 5% acetonitrile for 10 min followed by a gradient to 8% acetonitrile for 5 min, 35% acetonitrile for 113 min, 55% acetonitrile for 12 min, 95% acetonitrile for 15 min, and re-equilibrated with 5% acetonitrile for 15 min.

Data-dependent MS/MS data were obtained with an LTQ linear ion trap mass spectrometer using a home-built nanoelectrospray source at 2 kV at the tip. One MS spectrum was followed by 4 MS/MS scans on the most abundant ions after the application of a dynamic exclusion list. Tandem mass spectra were extracted using the Xcalibur software (Thermo Scientific, Waltham, MA, USA). All MS/MS samples were analyzed by using Mascot (version 2.1.04; Matrix Science, London, UK) with *H. Sapiens* proteins contained in the NCBI protein database (NCBInr2 20080628 (6655203 sequences; 2281585098 residues). Taxonomy: *Homo sapiens* (human) (202147 sequences), assuming the digestion enzyme trypsin with a maximum of 1 miscleavage. Mascot was searched with a fragment ion mass tolerance of 0.80 Da and a parent ion tolerance of 2.0 Da, and fixed modifications: carbamidomethyl (C) and variable modifications: oxidation (M). Identification was carried out at the 95% confidence level with a calculated false-positive rate of <1% as determined by using a reversed concatenated protein database. The minimum score for a nonrandom identification with more than 95% confidence was 45 for the GST control search and 44 for the GST-HO-1 search. At least one unique peptide with MS/MS data was identified for each protein hit. To control for nonspecific binding, we compared GST-HO-1 co-purifying proteins with those immunoprecipitated in cells transfected with a GST empty vector. Only differential GST-HO-1 binding proteins compared with GST-binding proteins were considered further.

#### *2.7. Co-Immunoprecipitation (Co-IP)*

After treatment with H2O2, total proteins were extracted from PC3 cells and quantified using the BCA method (Sigma-Aldrich, Gillingham, UK) (98% BCA + 2% CuSO4). 500 μg of proteins from the cell lysates and 10 μg of the specific rabbit anti-human HO-1 antibody were diluted in a final volume of 500 μL in RIPA buffer (150 μM NaCl, 20 μM EDTA 1% *v*/*v* sodium deoxycholate, 0.1% *v*/*v* sodium dodecyl sulfate (SDS), 1% Triton x-100Tris, pH 7.4). As a specificity control, samples were incubated with 10 μg of nonspecific rabbit anti-human IgG antibody. All samples were supplemented with MPI protease inhibitors (Sigma-Aldrich, Gillingham, UK), and were incubated overnight at 4 ◦C with orbital shaking. Protein A/G PLUS-Agarose beads (Santa Cruz Biotechnology Inc., Santa Cruz, CA, USA) were washed; 20 μL were added to each sample and incubated overnight at 4 ◦C with orbital shaking. Antibody-protein-beads complexes were washed with 500 μL of RIPA buffer and re-suspended in 40 μL of RIPA 20% loading buffer (5% B-Mercaptoethanol). Samples were heated for 5 min at 95 ◦C. 60 μg of the total protein lysate was used as input. The immune complexes were analyzed by Western blot as previously described [21], using mouse anti-human HO-1 or rabbit anti-human 14-3-3ζ/δ antibodies. Three independent experiments were performed.

#### *2.8. Immunofluorescence (IF) Experiment*

60,000 PC3 cells were cultured on coverslips placed in 12-well plates and the respective treatments were carried out. Afterwards, cells were washed three times with PBS and then fixed with 1 mL of 100% methanol for 20 min at −20 ◦C. After removing the methanol, cells were incubated at room temperature in a permeabilizing solution of 1 mL PBS-triton 0.1% *v*/*v* for 10 min. Subsequently, coverslips were placed in a humid and dark chamber and were incubated with a blocking solution (40 μL of PBS-BSA 3% m/v for 1 h). Coverslips were then incubated overnight at 4 ◦C with 40 μL of the corresponding primary antibodies (mouse anti-human HO-1 and/or rabbit anti-human 14-3-3ζ/δ antibodies), prepared in PBS-BSA 3% m/v. After performing three washes of 10 min with PBS, the coverslips were incubated for 1 h at room temperature with the Alexa Fluor 555 and/or Alexa Fluor 647 secondary antibodies in a 1:5000 dilution prepared in PBS-BSA 3% m/v.

After three washes of 10 min with PBS, the coverslips were mounted on a slide with 2 μL of Mowiol (Sigma-Aldrich, Gillingham, UK) and stored at 4 ◦C in the dark until use.

Confocal images were acquired by confocal microscopy (FV1000, Olympus, Tokyo, Japan), using a UPlanSApo 100X oil immersion objective (NA 1/41.35; Olympus), a diode laser of 543 and 635 nm as the excitation sources. Images were obtained with a Qimaging EXI Aqua camera; >20 cells were analyzed for the IF assay for each condition.

#### *2.9. Image Processing for Presentation*

Confocal images were processed for presentation using ImageJ software (NIH, Bethesda, MD, USA). The background of each channel was subtracted. The JACoP plugin was used in order to evaluate co-localization, and Manders and Pearson's coefficients were estimated. GraphPad Prism software (La Jolla, CA, USA) was used to assess statistical significance, which was set at *p* < 0.05.

#### *2.10. Bioinformatics Analysis*

2.10.1. Identification of HO-1 Interactor Proteins with Nuclear Localization and GO Enrichment Analysis

Protein–protein interactions and gene ontology (GO) analyses were performed using the STRING webtool [22], using a minimum combined interaction score of 0.400. Plots were generated using Cytoscape [23] and enrichplot package [24] in R. Statistical significance was set at false discovery rate (FDR) *p* < 0.05.

#### 2.10.2. Information Source and Eligibility Criteria (GEO: Gene Expression Omnibus)

To study the impact of the expression of the selected genes on the survival of patients, we selected the following dataset:

Ross-Adams 2015 (GSE70770) GPL10558 series [25]: a PCa patient's cohort with 206 tumor tissue samples from men with PCa who had undergone radical prostatectomy and a clinical follow-up of 9 years, including biochemical relapse (BCR) information, defined according to European Guidelines as a persistent rise of PSA above 0.2 ng/mL. Tumor sample expression of 31,000 transcripts was measured by 47,000 probes using the Illumina Human HT-12 V4.0 platform. A descriptive table regarding patient characteristics at baseline (start of the follow-up for survival analyses) is depicted in Supplementary Table S1.

### 2.10.3. Gene Correlation

Pairwise gene correlation between *HMOX1* and all genes encoding for the HO-1 interactor proteins of interest was analyzed with the ggcorrplot package in R. Correlation coefficients were classified as weak (|r| ≤ 0.30), intermediate (0.30 < |r| < 0.66), and strong (|r| ≥ 0.66). Statistical significance was set at *p* < 0.05.

#### 2.10.4. Risk Scoring System Analysis

Based on the expression of genes with a strong/intermediate gene correlation with *HMOX1*, a risk score model was created based on the coefficients of a Cox logistic regression analysis. Then, the sum of the product of the coefficient (Coef) values of all genes and their dichotomic expression (Expr) values was calculated as the patient risk score (risk score = ∑*<sup>n</sup> <sup>i</sup>*=1(*Coefi* × *Expri*)). Using Cutoff Finder software [26], patients were divided into high-risk and low-risk groups. Kaplan–Meier survival analysis was used to determine whether relapse-free survival (RFS) was significantly different between high-risk and low-risk patients.

#### 2.10.5. Survival Analyses

Kaplan–Meier curves showing the RFS of patients with PCa were plotted using the survminer package [27] in R. To find the optimal cutoff value to stratify patients into two groups based on the expression levels of each gene, we used the Cutoff Finder tool. Multivariable analysis was performed in Stata (StataCorp LLC, College Station, TX, USA) and plotted in GraphPad Prism software (La Jolla, CA, USA). For univariable and multivariable analyses of prognostic factors, the Log-rank test and Cox proportional hazard model regression were employed. Statistical significance was set at *p* < 0.05.
