The present study consisted of a search for negative regulators of ferroptosis through in silico testing and validation in clinical samples of CRC patients. The study flow is shown in Fig. 1A. Details on the discovery phase and gene identification are described below.

Stage 1 to stage 4 CRC patients (n=166) who underwent surgical resection at the Department of Gastrointestinal and Pediatric Surgery of the Mie University Graduate School of Medicine from January 2012 to December 2015 were enrolled in this study. Patients with incomplete clinical data or inadequate immunohistochemistry (IHC) results were excluded. Patient clinicopathological characteristics are shown in Table SI. Staging was performed on the basis of clinical and histopathological assessment following the International Union Against Cancer TNM staging system (24). All patients were followed up after initial hospital discharge, with physical examination and tumor marker assays (CEA and CA19-9) performed every 1–3 months and computed tomography every 6 months; endoscopic examinations were performed when necessary. Written informed consent was obtained from all patients following the local ethics guidelines, and the Institutional Review Board (IRB) of Medical Ethics Committee of Mie University Graduate School of Medicine approved this study (IRB number: H2023-172).

Several studies have discussed genes associated with negative regulation of ferroptosis (23,25–29). In this study, we confirmed the candidate ferroptosis negative regulator genes using the Gene Ontology Consortium Website (www.geneontology.org, and http://www.informatics.jax.org/vocab/gene_ontology/). We downloaded gene expression from the RNAseq (IlluminaHiSeq) dataset of ‘The Cancer Genome Atlas (TCGA) Colon and Rectal Cancer (COADREAD)’ patients (https://xenabrowser.net/datapages/). We investigated the association between the expression profile of seven ferroptosis negative regulator genes, SLC7A11, AIFM2, NFE2L2 (NRF2), FTH1, GLS2, GPX4, and HSPB1, and overall prognosis in the 367 CRC patients in the in silico cohort with available survival data.

Formalin-fixed, paraffin-embedded sections (5 µm thickness) of specimens from the 166 CRC patients were subjected to immunohistochemical analysis. After deparaffinization by xylene and rehydration in graded concentrations of ethanol, specimens were heated at 121°C for 10 min in citrate buffer (pH 6.0) to unmask antigens. Endogenous peroxidase activity was blocked by incubation with 3% hydrogen peroxide in distilled water. Nonspecific binding sites were blocked with 10% normal goat serum (Vector Laboratories, Burlingame, CA, USA) in Tris-buffered saline with 0.05% Tween 20 (TBST), and samples were incubated with primary antibody overnight at 4°C. Primary GPX4 antibody (Abcam, Waltham, MA, USA) was used at 1:1,000, and primary HSPB1 antibody (same as HSP27, Santa Cruz Biotechnology, Dallas TX, USA) was used at 1:1,000. After washing with TBST, sections were incubated with secondary antibody coupled with peroxidase-conjugated polymers [Universal Immuno-peroxidase Polymer method, Histofine SAB-PO(M) Kit, Nichirei Biosciences, Inc., Tokyo, Japan] for 30 min and detected using the Histofine DAB substrate kit (Nichirei Biosciences, Inc.). Slides were counterstained with hematoxylin, as previously described (30–33).

GPX4 and HSPB1 expression of CRC specimens were analyzed three times separately by two investigators who were not familiar with the clinical or survival data of patients. The investigators first evaluated the entire tissue specimen at low-power magnification (×40) and then focused on tumor hotspots at high-power magnification (×200 and ×400). As previously described (30–33), an immunohistochemical score was determined for each case as follows: Staining intensity score: 0, colorless; 1, weak; 2, moderate; and 3, strong; and staining percentage score: 1, 1–25; 2, 26–50; 3, 51–75 and 4, >75% positivity. The IHC score of each patient was obtained by multiplying the scores for staining intensity and staining percentage. If the difference between the scores obtained from each investigator was >3, the stained slide was reassessed.

CRC cell lines (DLD-1, RKO, SW480, and LOVO) were acquired from the Cell Resource Center of Biomedical Research, Institute of Development, Aging and Cancer (Tohoku University; Sendai, Japan). The CRC cells were maintained in RPMI 1640 (Nacalai Tesque Inc., Kyoto, Japan), supplemented with 10% fetal bovine serum (FBS, Biowest, Nouille, France) and Antibiotic-Antimycotic (Nacalai Tesque Inc.) and maintained at 37°C in a humidified incubator at 5% CO₂. All cell lines were checked and authenticated using a panel of genetic and epigenetic markers and tested for mycoplasma on a regular basis. 5-Fluorouracil (5FU, Sigma-Aldrich, MA, USA) was dissolved in dimethyl sulfoxide and diluted to the appropriate experimental concentrations in cell culture medium before use.

Specific predesigned small interfering RNA (siRNA) for GPX4 and HSPB1 and Negative Control siRNA were purchased from Ambion (Thermo Fisher Scientific, Inc., USA). The siRNA sequences of GPX4 and HSPB1 were as follows: GPX4: sense, 5′-GGCAAGACCGAAGUAAACUtt-3′ and antisense, 5′-AGUUUACUUCGGUCUUGCCtc-3′; and HSPB1: sense, 5′-CGAGAUCACCAUCCCAGUCtt-3′ and antisense, 5′GACUGGGAUGGUGAUCUCGtt-3′ (SiRNA ID number for GPX4 is 10848, and for HSPB1 is 121323. Catalogue number for Negative Control siRNA is 4390843). We used two CRC cell lines (DLD-1 and SW480) with high gene expression of GPX4 and HSPB1 for experiments. DLD-1 and SW480 cells were seeded in six-well culture plates at 2×10⁵ cells per well in 2 ml RPMI 1640. Cells were cultured for 24 h and then incubated with siRNA oligonucleotides using Lipofectamine RNAiMAX Reagent and OptiMEM I (both Invitrogen, Thermo Fisher Scientific, Inc.) following the manufacturer's protocol. The final concentration of siRNA oligonucleotide of GPX4/HSPB1 was 50 nM. After 48 h, transfected cells were harvested and examined by RT-qPCR and western blotting to check siRNA efficiency.

Total RNA was extracted from CRC cell lines (DLD1, RKO, SW480, and LOVO) using miRNeasy (Qiagen, Germany) following the manufacturer's instructions. RNA quality and concentration were determined using a Denovix DS-11+ spectrophotometer (DeNovix, Inc., USA). cDNA was synthesized from 5 µg of total RNA with random hexamer primers, dNTPs, 5X buffer, RNase inhibitor, and ReverTra Ace (Toyobo Co., LTD., Japan).

RT-qPCR analyses were conducted using the SYBR Green PCR Master Mix and QuantStudio3 Real-Time PCR System (Applied Biosystems, Thermo Fisher Scientific, Inc., USA) following the manufacturer's protocol and as previously described (22). The qPCR cycling conditions were as follows: 95°C for 10 min, followed by 45 cycles of 15 sec at 95°C and 60 sec at 60°C. Primers for GPX4, HSPB1 and GAPDH mRNAs were as follows; GPX4: forward, 5′-GAGGCAAGACCGAAGTAAACTAC-3′ and reverse, 5′-CCGAACTGGTTACACGGGAA-3′; HSPB1: forward, 5′-ACGGTCAAGACCAAGGATGG-3′ and reverse, 5′-AGCGTGTATTTCCGCGTGA-3′; and GAPDH: forward, 5′-GGAAGGTGAAGGTCGGAGTC-3′ and reverse, 5′-AATGAAGGGGTCATTCATGG-3′. Relative expression levels of GPX4 and HSPB1 mRNA were calculated by normalization to the levels of endogenous GAPDH mRNA using the 2^−ΔΔCq method (22). RT-qPCR assays were performed in triplicate for each sample and the mean value was calculated.

At 48 h after transfection of GPX4 or HSPB1 siRNA, cells were lysed in RIPA buffer (BioDynamics Laboratory, Inc., Tokyo, Japan) with protease inhibitors, and lysates were centrifuged for 5 min at 15,000 × g at 4°C. The protein concentration was measured using the BCA protein assay kit (Thermo Fisher Scientific, Inc.). Protein samples (20 µg) were separated on AnykD^™ Mini-PROTEAN^® TGX^™ Precast Protein Gels and then transferred onto a polyvinylidene difluoride membrane (Bio-Rad Laboratories, CA, USA). The blots were first blocked with 5% skimmed milk for 1 h at room temperature and then incubated overnight with primary antibodies against GPX4 (1:20,000; Abcam, Cambridge, UK), HSPB1 (1:10,000; Santa Cruz, CA, USA), and β-actin (1:40,000; MP Biomedicals, Illkirch, France). The blots were incubated with secondary antibody (Promega, Madison, WI, USA) for 60 min at room temperature. The protein bands were visualized by chemiluminescent reaction (Immobilon^™ Western; Millipore, MA, USA) coupled with a WSE-6100 LuminoGraph imaging system (ATTO Corporation, Tokyo, Japan).

After transfection of GPX4 or HSPB1 siRNA, DLD-1 and SW480 cells were plated in 96-well tissue culture plates (TPP Techno Plastic Products AG, Switzerland) at a density of 3,000 cells/well in RPMI1640 medium supplemented with 10% FBS and antibiotics. The cells were allowed to adhere to the plate for 24 h and then treated with a series of two-fold dilution of 5FU (0, 6.25, 12.5, 25, 50, 100, 200, 400 µM). After 72 h of treatment, cell proliferation was measured using the WST8 assay (Dojindo Laboratories, Kumamoto, Japan) following the manufacturer's instructions. The absorbance in each well was measured at a wavelength of 450 nm with a Multiskan FC plate reader (Thermo Fisher Scientific, Inc., USA). Each experiment was performed in triplicate. The cytotoxic effect of 5FU was assessed by the IC50 concept (34), and the IC50 value for 5FU was calculated using CompuSyn software (Chou and Martin, 2005, Compusyn, Inc., USA).

Statistical analyses were performed using Medcalc statistical software V.16.2.0 (Medcalc Software bvba, Ostend, Belgium), JMP software 10.0.2 (SAS Institute, Cary, NC, USA), and GraphPad Prism software ver.8.2.0 (GraphPad Software Inc., San Diego, CA). Comparison of IHC score between high- and low-expression groups and various clinicopathological factors in the clinical cohort were performed using Fisher's exact test. Comparisons of differential gene expression between two groups from in vitro experiments were performed using two-tailed unpaired Student's t-test. On the other hand, comparisons of differential gene expression between multiple groups from in vitro experiments were performed using one-way ANOVA with Tukey's multiple comparisons test. Comparisons of IHC scores between various stages were performed using Kruskal-Wallis test and Dunn's multiple comparisons test. Youden's index for dead/alive or recurrence/non-recurrence within the observation period was used to determine the optimal cutoff thresholds to dichotomize patients into high- and low-expression groups of GPX4 and HSPB1 using Medcalc statistical software. OS was defined as the period from the date of CRC diagnosis to the date of last known follow-up; recurrence-free survival (RFS) was defined as the period from the date of CRC diagnosis to the date of recurrence. Kaplan-Meier analyses for OS and RFS were performed by log-rank test. For OS and RFS survival analyses, we dichotomized patients into GPX4 or HSPB1 high- and low-expression groups as determined by receiver operating characteristic analysis (same as Youden's index) for dead/alive or recurrence/non-recurrence. Univariate and multivariate analyses for OS and RFS were performed using the Cox proportional hazards model to determine the factors affecting OS and RFS. P<0.05 was considered to indicate a statistically significant difference.

As described in the Materials and Methods section, we determined SLC7A11, AIFM2, NFE2L2 (NRF2), FTH1, GLS2, GPX4, and HSPB1 genes as candidate negative regulators of ferroptosis. We then evaluated their prognostic importance by analyzing 367 CRC patients with available information on survival outcome in TCGA-COADREAD datasets. We dichotomized patients into high and low gene expression groups as determined by Youden's index for dead/alive. High expression of SLC7A11, AIFM2, NFE2L2, FTH1, and GLS2 genes did not stratify survival outcome of patients (Fig. 1B-F). In contrast, patients with high expression of GPX4 and HSPB1 genes showed significantly worse OS (P<0.01 for both factors) than those with low expression (Fig. 1G and H). Therefore, we selected GPX4 and HSPB1 as candidate targets for further validation study.

To evaluate the prognostic potential of GPX4 and HSPB1 in CRC patients, we first investigated GPX4 expression in a cohort of stage 1–4 CRC patients by IHC analysis. While GPX4 protein was not expressed or weakly expressed in the adjacent normal mucosa (Fig. 2A), it was expressed mainly in the cytoplasm of CRC cells (Fig. 2A). There was varying expression of GPX4 among CRC cases (Fig. 2B). The GPX4 IHC score for 151 CRC patients was 3.5±2.1 (mean ± SD), and median value for GPX4 IHC score was 4; 15 CRC patients with GPX4-negative staining were excluded from the analysis. The GPX4 IHC score of stage 1 CRC patients was significantly lower than those of stage 2–4 patients (Fig. 2C). We analyzed the association between GPX4 expression and clinicopathological parameters and found that high GPX4 expression (GPX4 IHC score ≥4) was significantly associated with larger tumor size (P=0.002), advanced T factor (P=0.003), lymph node metastasis positivity (P=0.002), lymphatic and venous invasion positivity (P=0.02 and P<0.001, respectively), and stage 4 cases (P=0.02) (Table I). These results indicate that high GPX4 protein expression was significantly associated with an aggressive cancer phenotype in CRC patients in the clinical cohort.

Next, we evaluated the prognostic value of GPX4 protein expression by examining OS using Kaplan-Meier analysis. Patients with high expression of GPX4 (GPX4 IHC score ≥4) exhibited significantly worse OS compared with patients with low expression (P=0.001, Fig. 2D). To elucidate whether GPX4 protein expression can be used for recurrence prediction, we investigated the association between GPX4 protein expression and postoperative recurrence by analyzing RFS of 130 stage 1–3 CRC patients receiving treatment with curative intent. In line with the OS data, patients with high expression of GPX4 (GPX4 IHC score ≥5; dichotomized by Youden's index for positive/negative recurrence) showed significantly worse RFS than those with low expression (P=0.0001, Fig. 2E).

To determine whether high GPX4 protein expression was a risk factor for OS or RFS, the Cox proportional hazard model was used to perform univariate and multivariate analysis. In univariate analysis for OS, along with high GPX4 protein expression (P=0.0005), several clinicopathological factors such as rectal cancer (P=0.02), invasive endoscopic type (P=0.04), larger tumor size (P=0.008), diffuse histological type (P=0.009), advanced T stage (P<0.0001), lymph node metastasis (P=0.0002), lymphatic and venous invasion positive cases (P=0.001 for both factors), and stage 4 cases (P<0.0001) were risk factors for poorer OS (Table II). Multivariate analysis showed that diffuse histological type [hazard ratio (HR)=6.48, P=0.02] and stage 4 cases (HR=5.61, P=0.009) were independent risk factors for poorer OS; high GPX4 expression did not show statistical significance (P=0.14, Table II).

With regard to univariate analysis for RFS, primary lesion in rectum (P=0.0007), invasive endoscopic type (P=0.049), advanced T stage and N stage (P=0.01, P=0.002, respectively), lymphatic and venous invasion positive cases (P=0.0007, P=0.002, respectively), and high GPX4 expression (P=0.018) were risk factors for poor RFS (Table II). Multivariate analysis showed that primary lesion in rectum (HR=6.83, P<0.0001), invasive endoscopic type (HR=5.91, P=0.008), lymphatic invasion positivity (HR=4.66, P=0.02), and high GPX4 expression (HR=4.11, P=0.03) were independent risk factors for poor RFS (Table II). Thus, high GPX4 expression may predict poor RFS in CRC patients.

We also investigated the prognostic potential of HSPB1 in the stage 1–4 CRC patients by IHC. While HSPB1 protein was absent or weakly expressed in the adjacent normal mucosa (Fig. 2F), it was expressed mainly in the cytoplasm of CRC cells (Fig. 2F), similar to the expression of GPX4. HSPB1 expression also varied among CRC cases (Fig. 2G). The HSPB1 IHC score for 156 patients was 3.1±2.4, and median value for HSPB1 IHC score was 3; 10 CRC patients with HSPB1-negative staining were eliminated from analysis. Notably, the IHC score trend of HSPB1 showed a more prominent ‘stage-dependent elevated pattern’ in CRC patients (Fig. 2H). Similar to the results with GPX4 expression, high HSPB1 expression (HSPB1 IHC score >4; dichotomized by Youden's index for OS) was significantly associated with aggressive cancer phenotypes such as diffuse histological type (P=0.03), advanced T factor (P=0.005), lymph node metastasis positivity (P<0.0001), lymphatic and venous invasion positivity (P<0.0001 for both factors), and stage 4 cases (P<0.0001) (Table I). Thus, similar to GPX4, high HSPB1 protein expression was significantly associated with aggressive cancer phenotype in CRC patients.

We further assessed the overall prognostic significance of HSPB1 protein expression by analyzing OS using the Kaplan-Meier method. Patients with high expression of HSPB1 (HSPB1 IHC score ≥5) showed significantly worse OS than patients with low expression (P<0.0001, Fig. 2I). Kaplan-Meier analysis for RFS in 132 stage 1–3 CRC patients treated with curative intent revealed that patients with high expression of HSPB1 (HSPB1 IHC score ≥3; dichotomized by Youden's index for positive/negative recurrence) showed significantly worse RFS than patients with low expression (P<0.0001, Fig. 2J).

We performed Cox proportional univariate and multivariate analysis for OS and RFS in the CRC patients. In univariate analysis for OS, along with high HSPB1 expression (P<0.0001), several aggressive phenotypes such as invasive endoscopic type (P=0.002), larger tumor size (P=0.002), diffuse histological type (P=0.002), advanced T stage and N stage (P<0.0001 for both factors), lymphatic and venous invasion positive cases (P=0.002, P=0.001, respectively), and stage 4 cases (P<0.0001) were risk factors for poor OS (Table III). Multivariate analysis identified that diffuse histological type (HR=5.84, P=0.01), stage 4 patients (HR=3.99, P=0.02), and high HSPB1 expression (HR=6.35, P=0.006) were independent risk factors for poor OS (Table III). Regarding univariate analysis for RFS, primary lesion in rectum (P=0.0007), invasive endoscopic type (P=0.049), advanced T stage and N stage (P=0.01, P=0.002, respectively), lymphatic and venous invasion positivity (P=0.0007, P=0.002, respectively), and high HSPB1 expression (P<0.0001) were risk factors for poor RFS (Table III). Multivariate analysis demonstrated that primary lesion in rectum (P=0.0003), advanced T stage (P=0.009), and high HSPB1 expression (P<0.0001) were independent risk factors for poor RFS (Table III). These results indicate that high HSPB1 expression may indicate poor prognosis and high risk of recurrence in CRC patients, and its biomarker potential may be superior to that of GPX4.

We next investigated whether GPX4 and HSPB1 expressions were biomarkers for adjuvant chemotherapy in advanced (stage 2 and 3) CRC patients. We first investigated the association between clinicopathological factors and GPX4 and HSPB1 expressions by Fisher's exact test. We dichotomized cases into high- and low-expression groups by Youden index for RFS. While there was no significant correlation between the clinicopathological factors and GPX4 expression, the frequency of patients with high expression of HSPB1 was significantly higher in cases with rectal cancer and cases with positive lymph node metastasis (Table SII).

We then evaluated the recurrence prediction potential of GPX4 and HSPB1 by Kaplan-Meier analysis in stage 2 CRC patients who were not administered adjuvant chemotherapy. While there was no significant difference in RFS between patients with high and low GPX4 expression (P=0.18), the recurrence rate of patients with high GPX4 expression was worse than those with low expression (HR=4.52) (Fig. S1A). Patients with HSPB1 high expression showed significantly worse RFS than those with low expression (P=0.0002). HR could not be calculated because no patients with HSPB1 low expression showed recurrence (Fig. S1B).

We next evaluated RFS in stage 2 and 3 CRC patients who were treated with capecitabine. While there was no significant difference for RFS between patients with high and low GPX4 expression (P=0.26), the recurrence rate of patients with high expression of GPX4 was worse than that of patients with low expression (HR=3.61, Fig. S2A). Similarly, for HSPB1, while there was no significant difference of RFS between high- and low-expression groups (P=0.25), patients with high expression of HSPB1 showed worse recurrence rates than those with low expression (HR=3.52, Fig. S2B).

Collectively, our findings showed that HSPB1 expression indicated recurrence in stage 2 and 3 CRC patients treated with curative intent. In patients with high expression of HSPB1, we may recommend administration or reinforcement of adjuvant chemotherapy. However, further studies with more patients are required to strengthen our findings.

On the basis of the results of subgroup analysis for GPX4 and HSPB1 in advanced CRC patients treated with adjuvant therapy with curative intent, we further investigated the functional role of GPX4 and HSPB1 in the response to 5-fluorouracil (5FU)-based chemotherapy using CRC cell lines. First, we evaluated the mRNA and protein expression of GPX4 and HSPB1 in four CRC cell lines (DLD-1, RKO, SW480, and LOVO) (Fig. 3A and B). We selected DLD-1 and SW480 for further experiments, as both GPX4 and HSPB1 mRNA and protein were detected at higher levels in these two cell lines. Next, we transfected GPX4 and HSPB1 siRNA in these two CRC cell lines and confirmed effective knockdown of mRNA (Fig. 3C) and protein levels (Fig. 3D) of both factors. We then compared the cytotoxic effect of 5FU between cells transfected with negative control siRNA and cells transfected with siRNA against GPX4 and HSPB1 using WST8 assays. The IC50 values of 5FU were decreased in cells transfected with GPX4 siRNA (DLD-1, from 106.03 to 58.13 µM; SW480, 67.61 to 24.73 µM) and HSPB1 siRNA (DLD-1, from 106.03 to 67.62 µM; SW480, 67.61 to 18.93 µM) (Fig. 3E). Collectively, these in vitro experiments demonstrated that GPX4 and HSPB1 may play crucial roles in attenuating the cytotoxic effect of 5FU-based conventional chemotherapy.