We retrospectively collected a total of 112 consecutive patients with a diagnosis of ICC. Among them, 90 patients underwent laparoscopic hepatic resection with curative intent from January 2006 to May 2021. The remaining 22 patients were patients with liver mass who underwent diagnostic liver biopsy, eventually diagnosed as ICC, who subsequently underwent surgery in a different hospital. Thirty patients included in our cohort were already investigated before, as part of a previous study (22). Exclusion criteria were (i) the administration of any systemic or loco-regional therapy prior to surgery/biopsy, (ii) a survival of less than 3 months after surgery (for patients who underwent resection), to exclude deaths due to surgical complications, and (iii) the absence of available residual tumor tissue for immunohistochemical (IHC) stains and molecular tests. Only surgical specimens were considered for molecular analyzes. The study complies with the ethical guidelines of the 1975 Declaration of Helsinki and obtained the approval from the local Ethics Committee (Ethics Committee for Clinical Research—University Hospital of Padova, Italy; protocol #: 0038038/17). All of the patients gave their appropriate informed consent to any procedure.

Patients’ relevant clinical and laboratory data were retrieved from medical records, including sex, age, serum ferritin levels, the presence of any underlying chronic hepatic or biliary disease, the presence of cirrhosis, and the administration of adjuvant chemotherapy. We also reported whether lymphadenectomy was performed during surgery or not.

All patients who underwent surgery were clinically followed-up, and regularly subjected to ultrasonography and computed tomography to detect any recurrence of the disease. Overall and disease-free survival time was obtained from medical charts.

All of the cases were blindly and contemporarily reviewed by an experienced (MG) and two trainee (SS and DS) liver pathologists, and relevant histological features were recorded, including macroscopic tumor type, histotype, grade of differentiation, T stage (according to the revised 8th edition of the UICC staging system) (23), margin status (for surgical resections), and the presence of vascular and perineural invasion and lymph node metastasis.

Tissue microarrays (TMAs) made of formalin-fixed paraffin-embedded ICC tissue cores (with a diameter of 2 mm) were obtained by selecting two or three representative tumor areas from each liver resection case, depending on tumor dimension. All of the samples were processed by using the TMA Master platform (3DHistech, Budapest, Hungary), a semi-automatic and computer-assisted TMA platform.

Immunostains were performed on TMA and liver biopsy sections by using the following antibodies: anti-STAT3 (clone F-2; Santa Cruz Biotechnology, Dallas, TX, United States; dilution 1:200; mouse monoclonal), anti-GPX4 (clone E-12; Santa Cruz Biotechnology, Dallas, TX, United States; dilution 1:400; mouse monoclonal), and anti-transferrin receptor 1 (TFR1; also known as CD71) (clone 10F11; Leica Biosystems, Newcastle upon Tyne, United Kingdom; dilution 1:100; mouse monoclonal). All IHC stains were conducted according to standard protocols by using the Dako Omnis autostainer (Dako, Glostrup, Denmark), and all of the slides were counterstained with hematoxylin. Appropriate positive and negative controls were used for each run. In evaluating the expression of all markers, only cytoplasmic staining was considered. STAT3 and GPX4 expression was semi-quantitatively scored from 0 to 2 +, as follows: 0 = negative; 1 + = expression in ≤ 50% of tumor cells (TCs); 2 + = expression in > 50% of TCs (Figure 1). Finally, TFR1 positivity was evaluated as follows: 0 = negative or positive in ≤ 10% of TCs; 1 + = expression in 11–50% of TCs; 2 + = expression in > 50% of TCs (Figure 2).

In every case, we also performed histochemical Perls’ stain (Artisan Iron Stain Kit, Dako, Glostrup, Denmark) to detect intratumoral iron deposits, which were recorded as absent/present (Figure 3).

As reported above, only surgical specimens (90 cases) were considered for molecular tests. For each case, DNA extraction was performed on selected representative areas of formalin-fixed paraffin-embedded ICC tissue, with Qiasymphony DSP DNA Mini Kit (Qiagen, Hilden, Germany). DNA quantity was measured spectrophotometrically by using NanoDrop ND 100 Spectrophotometer (Thermo Fisher Scientific, Waltham, MA, United States). Detection of the main IDH1 (codons 105 and 132) and IDH2 (codons 140 and 172) gene mutations was performed by real-time polymerase chain reaction (PCR) analysis (Easy PGX ready IDH1-2, Easy PGX platform, Diatech Pharmacogenetics, Jesi, Italy), according to the manufacturer’s instructions. When necessary, the identified mutations were confirmed by PCR analysis and Sanger sequencing, by using the ABI PRISM 3500 Genetic Analyzer (Applied Biosystems, Waltham, MA, United States).

Continuous variables were expressed as median (range) while categorical variables as frequency and percentage. For clinical-pathological correlations, appropriated tests were used, including Student’s t-test, one-way ANOVA test, Spearman rank correlation test and Fisher exact probability test. The Kaplan–Meier method was used to create survival curves, which were compared by using both the log-rank and the Breslow (generalized Wilcoxon) tests. Multivariate Cox backward stepwise regression analyzes were performed including all of the variables identified as significant on univariate Cox regression analyzes. Hazard ratios (HRs) and their 95% confidence intervals (CIs) were calculated. p values < 0.05 were considered statistically significant. Data analyzes were performed by applying Statistical Package for the Social Science (SPSS, version 25, IBM SPSS Statistics, Chicago, IL, United States) and GraphPad (version 6, GraphPad Software, San Diego, CA, United States) statistical software.

Overall, in our cohort of 112 patients there were 58 males (51.8%) and 54 females (48.2%), with a median age of 68 years (range 34–92 years). Seventeen patients (15.2%) had a cirrhotic liver, and 36 patients (32.1%) received adjuvant chemotherapy after resection/biopsy. Patients had a median follow-up of 1.7 years (range 0.3–8.8 years). Lymphadenectomy was performed in 33/90 patients who underwent surgery. Clinical and laboratory features of all the patients are summarized in Table 1. Age, sex, the presence of underlying chronic liver diseases, and the administration of adjuvant chemotherapy were not related to patients’ prognosis. Furthermore, overall and disease-free survivals were not different between patients with or without cirrhosis, probably because of the low number of cirrhotic patients in our cohort.

Macroscopically, all cases were mass forming, while, histologically, there were 82 cases (73.2%) of small duct type ICC and 30 cases (26.8%) of large duct type ICC. We observed vascular invasion in 78 out of 112 patients (69.6%), while perineural invasion was present in 50/112 cases (44.6%), as reported in Table 2. Among the 33 patients who underwent lymphadenectomy, 15 had lymph node metastases. Forty-seven patients out of the 90 who underwent surgery (52.2%) had a complete surgical excision of the tumor (R0 cases), while the remaining 43 patients (47.8%) showed a microscopic neoplastic infiltration of the resection margin (R1 cases).

As expected, grading and the presence of perineural and vascular neoplastic invasion were all related to worse overall (p = 0.009, p < 0.0001, and p = 0.004, respectively) and disease-free survivals (p = 0.01, p = 0.002, and p = 0.001, respectively). An advanced T stage (i.e., T stage 3 and 4) was related to a reduced overall survival (p = 0.06). We failed to find any correlation between patients’ prognosis and histotype or resection margin status. We did not observe differences in survival times between cases with or without lymph node metastases.

In our cohort, most of the cases showed GPX4 expression, either 1 + (43/112; 38.4%) or 2 + (46/112; 41.1%), while only 23 cases (20.5%) were completely negative. Twenty-two patients out of 112 (19.6%) showed 1 + TFR1 expression, while 7 cases showed a 2 + TFR1 positivity (6.3%). Intratumoral iron deposits were observed in only 2 cases (1.8%); among them, one showed a focal and one a diffuse deposition. Interestingly, the case that showed diffuse iron deposits had both a strong TFR1 expression and a completely negative GPX4 stain (Supplementary Figure 1). Taken together, these findings suggest an inhibition of the ferroptotic cascade in ICC. Most of our cases showed either a 1 + (56/112; 50%) or 2 + (51/112; 45.5%) STAT3 expression, with only 5 cases (4.5%) showing a completely negative reaction, suggesting an activation of the inflammatory pathway in our ICC cohort. Histochemical and immunohistochemical expressions of the different markers are summarized in Table 3.

Statistical analyzes showed a direct correlation between GPX4 expression and the presence of poor prognostic histological parameters, that is grading (p < 0.0001) and perineural (p = 0.03) and vascular invasion (p < 0.0001). We also found a direct association between the presence of the same unfavorable histological factors and STAT3 expression (p = 0.001, p = 0.04, and p = 0.07 for grading, perineural invasion, and vascular invasion, respectively), confirming what previously reported (4, 10). No associations were found between the same histological features and TFR1 expression, neither between them and the presence of iron deposits. No correlations were found between marker expression (and iron deposits) and tumor T stage. We observed a direct association between STAT3 and GPX4 expression (p < 0.0001) (Supplementary Figure 2), suggesting inhibition of ferroptosis in ICCs with an inflammatory background. Accordingly, we found an inverse correlation between STAT3 and TFR1 expression (p = 0.04). We failed to find any correlation between TFR1 and GPX4 expression, as well as between ferroptosis markers and STAT3 expression and the presence of intratumoral iron deposits.

Age, sex, the presence of any underlying chronic liver disease, the presence of cirrhosis, ferritin levels, tumor histotype, and the presence of nodal metastases were not related to the expression of any of the markers, neither to the presence of iron deposits.

As already reported in the literature, STAT3 expression was related to a worse overall (p = 0.02) and disease-free survival (p = 0.001) (Figure 4). We found significantly reduced overall (p = 0.06) and disease-free survivals (p = 0.04) in cases with a higher GPX4 expression, as shown by the Kaplan–Meyer curves (Figure 5). Considering GPX4 1 + and 2 + positive cases together, the association becomes even stronger for both overall and disease-free survival (p = 0.03 and p = 0.01, respectively) (Figure 5), suggesting that mild GPX4 presence is enough to inhibit ferroptosis and reduce survival times. TFR1 expression and the presence of iron deposits were not related to survival times.

Molecular analysis results are reported in Table 4. As shown, they revealed IDH1 point mutations in 15 out of 76 tested cases (19.7%), of which 14 involving codon 132 and one involving codon 105. Only one case showed a point mutation in IDH2 codon 172 (1.3%). In 9 cases (11.8%) molecular tests described a single nucleotide polymorphism (SNP) in codon 105 of the IDH1 gene, causing a change in the nucleotide sequence from GGC [Gly] to GGT [Gly] (C > T). Among them, two cases showed both a point mutation in IDH1 (codon 132) and the IDH1^105GGT SNP. No relationships were found between the presence of IDH1 point mutations and the presence of the IDH1^105GGT SNP.

Statistical analyzes showed an inverse correlation between the presence of IDH1^105GGT SNP and tumor grading (p < 0.0001). On the contrary, a direct relationship was found between tumor grading and the wild type status of IDH1-2 genes (p = 0.004). No associations between tumor grading and the presence of point mutations in IDH1-2 genes were observed. We also failed to find any correlation between IDH1-2 status and age, sex, the presence of any underlying chronic liver disease, the presence of cirrhosis, ferritin levels, tumor histotype, the presence of perineural and vascular invasion, T stage and the presence of nodal metastases.

Furthermore, GPX4 expression was inversely correlated to the presence of IDH1^105GGT SNP (p = 0.001) (Supplementary Figure 3), while it was directly related to the presence of an IDH1-2 point mutation (p = 0.06) or wild type IDH1-2 (p = 0.04). No associations were found between IDH1-2 status and STAT3 and TFR1 expression, neither between IDH1-2 status and intratumoral iron deposits.

Finally, cases with IDH1^105GGT SNP showed a better overall survival than cases with wild type IDH1-2 (1,648 days vs. 887 days; p = 0.04) and cases with IDH1-2 point mutations (1,648 days vs. 1,333 days; p = 0.09) (Figure 6). We did not find any relationship between IDH1-2 status and disease-free survival times. The multivariate Cox regression analysis, including variables that were identified as significant on univariate survival analyzes (grading, perineural and vascular neoplastic invasion, T stage, STAT3 and GPX4 expression, and IDH1^105GGT SNP) showed that only the presence of perineural invasion is an independent predictor of a worse overall survival [p < 0.0001, HR = 3.64 (95% CI: 1.86–7.11)], while the multivariate Cox regression analysis including grading, perineural and vascular neoplastic invasion, and STAT3 and GPX4 expression found that STAT3 expression and vascular invasion are independent predictors of a reduced disease-free survival [p = 0.03, HR = 1.81 (95% CI: 1.05–3.14), and p = 0.05, HR = 1.97 (95% CI: 1.02–3.84), respectively] (Supplementary Table 1).