An inflammation‐associated ferroptosis signature optimizes the diagnosis, prognosis evaluation and immunotherapy options in hepatocellular carcinoma

Abstract Inflammation and ferroptosis crosstalk complexly with immune microenvironment of hepatocellular carcinoma (HCC), thus affecting the efficacy of immunotherapy. Herein, our aim was to identify the inflammation‐associated ferroptosis (IAF) biomarkers for contributing HCC. A total of 224 intersecting DEGs identified from different inflammation‐ and ferroptosis‐subtypes were set as IAF genes. Seven of them including ADH4, APOA5, CFHR3, CXCL8, FTCD, G6PD and PON1 were used for construction of a risk model which classified HCC patients into two groups (high and low risk). HCC patients in the high‐risk group exhibited shorter survival rate and higher immune score, and were predicted to have higher respond rate in immune checkpoint inhibition (ICI) therapy. Levels of the seven genes were significantly changed in HCC tissues in comparison to adjacent tissues. After inserting the gene expression into the risk model, we found that the risk model exhibited the higher diagnostic value for distinguish HCC tissues compared each single gene. Furthermore, HCC tissues from our research group with high‐risk score exhibited more cases of microsatellite instability (MSI), heavier tumour mutational burden (TMB), higher expression level of PDL1 and cells with CD8. Knockdown of APOA5 reduced HCC cell proliferation combining with elevating inflammation and ferroptosis levels. In conclusion, we considered APOA5 maybe a novel target for suppressing HCC via simultaneously elevating inflammation and ferroptosis levels, and signature constructed by seven IAF genes including ADH4, APOA5, CFHR3, CXCL8, FTCD, G6PD and PON1 can act as a biomarker for optimising the diagnosis, prognosis evaluation and immunotherapy options in HCC patients.


| INTRODUC TI ON
Hepatocellular carcinoma (HCC) ranks sixth in morbidity and third in mortality of tumour-related deaths worldwide. 1 Surgical resection is the main therapeutic strategy for early-stage HCC. However, most of the individuals are diagnosed at an advanced stage, thus losing the opportunity for surgical treatment. 2 Moreover, HCC cells have a high resistance to conventional cytotoxic chemotherapeutic agents, limiting the benefit of chemotherapy in patients with HCC. 3 Since 2008, several tyrosine kinase inhibitors, including sorafenib and regorafenib, have been approved for first-and second-line treatments of HCC, whereas the overall survival rate still has no improvement. 4 In recent years, approval of immunotherapy, especially immune checkpoint inhibitor (ICI) therapy, has greatly relieved this embarrassing treatment status, and significantly improves the outcome of part of individuals. Unfortunately, majority of HCC individuals are defeated to respond to ICI. 5 In patients who fail respond to ICI, immune-related adverse reactions including dermatotoxicity, endocrine system toxicity, pneumonia and gastrointestinal toxicity turn into greater challenge during therapy. 6 Therefore, exploration of novel biomarkers for immunotherapy optimisation in HCC individuals was urgent.
Inflammation is a hallmark of cancer, and its link with the progression of cancer is well established. In fact, inflammation plays a double-edged role in tumours. 7 Studies have shown that transient inflammation exhibit anti-tumour effects, while chronic persistent inflammation can affect the plasticity of tumour cells by regulating cell differentiation, immune cell polarisation in microenvironment and other processes, hence facilitating tumour proliferation, metastasis and drug resistance. 8 Similarly, the role of inflammation in immune regulation is complex. Immunosuppressive inflammatory microenvironment is commonly formed in tumour tissues; therefore, tumour cells often escape from the killing of immune cells including T cells. 9 However, during immune checkpoint inhibition (ICI) therapy, the tumour microenvironment is remodelled, and tumour-promoting inflammation is transformed into tumour-suppressive inflammation. 10 Ferroptosis is a specific cell death model that depends on lipid peroxidation process, and it was different from procedural death in morphology and molecular mechanism. 11 Clinical researchers found that ferroptosis commonly exist in cancer cells while patients obtain radiotherapy, chemotherapy and tumour immunotherapy, indicating that ferroptosis activation is the strategy to inhibit cancer. 12 Ferroptosis has crosstalk with inflammation and immunotherapy.
Previous studies indicated that lipid peroxides produced during ferroptosis are identifying signals to accelerate recognition, phagocytosis and processing of tumour antigens by dendritic cells, thus intentinonally activating cytotoxic T lymphocytes to improve the efficacy of tumour immunotherapy. 13 Inflammation is an important factor involve in the activation of ferroptosis, and this phenomenon can be observed in the tissues after ICI therapy. 14 Therefore, inflammation induced the activation of ferroptosis during ICI therapy is generally beneficial to anti-tumour, and can act as biomarker to reflect the response of ICI.
Herein, our aim was to identify inflammation-associated ferroptosis (IAF) biomarkers for diagnosis and therapy of HCC. We demonstrated that the signature constructed by seven IAF genes including ADH4, APOA5, CFHR3, CXCL8, FTCD, G6PD and PON1 can act as a biomarker for optimising the diagnosis, prognosis evaluation and immunotherapy options in patients with HCC. Among the IAF genes, APOA5 was a novel target for suppressing HCC via simultaneously elevating inflammation and ferroptosis levels.

| Subtypes cluster
Inflammation-related genes were referred to the research of Danaher et al., 15 while ferroptosis-related genes were referred to the terms of 'WP_FERROPTOSIS' in Gene Set Enrichment Analysis (GSEA, http:// www.gsea-msigdb.org/gsea/index.jsp). Then, the inflammation-and ferroptosis-associated genes were used to conduct cluster analysis by the R package 'ConsensusClusterPlus', respectively. The cluster number was set as 3. Then, the difference of the overall survival (OS) between the subtypes was determined by Kaplan-Meier method.
p < 0.05 was considered different.

| Identification of differentially expressed genes (DEGs)
The DEGs between the different subtypes of HCC were analysed by the limma package (Version: 3.1.6). Genes with the |Log2 foldchange (LogFC)| ≥ 1 and adjust p value <0.05 were DEGs. Intersecting DEGs K E Y W O R D S diagnosis, hepatocellular carcinoma, immunotherapy optimisation, inflammation-associated ferroptosis gene, prognosis evaluation between inflammation-and ferroptosis-subtypes were set as IAF genes, and used for further analysis.

| Enrichment analysis of IAF genes
KEGG and GO analysis of IAF genes were analysed in the DAVID Bioinformatics Resources (https://david.ncifc rf.gov/; version: v2022q4). Enrichment GO terms contained biological process (BP) and molecular function (MF) terms. Enrichment terms with p < 0.05 were visualized.

| Risk model establish based on significant IAF genes
In order to construct a risk model by IAF genes, we first performed LASSO analysis to remove similarity factors and reduce the scope by adding penalty parameter using the R package 'glmnet' (Version: 4.1-6). Then, the effects on the OS of HCC patients of residual IAF genes was detected by performing univariate and multivariate regression analysis. By multiplying levels of genes and their corresponding regression coefficients deduced from multivariate Cox regression analysis, risk model was constructed. The HCC tissues were then divided into high-and low-risk groups followed by calculating the risk score of each tissue. The OS difference between the two groups was analysed by Kaplan-Meier method, while the effects of risk score on OS of HCC patients were analysed by univariate and multivariate regression analysis. p < 0.05 was cut-off. Cibersort package (version: 1.04) was conducted to measure the levels of 22 immune cells in HCC tissues via matching the reference data. The t-test was performed to analysed the differences via the cut-off as p < 0.05.

| Tumour immune dysfunction and exclusion (TIDE) analysis
For prediction of the respond rate of HCC tissues in the two groups (high and low risk) after ICI treatment, TIDE database (http://tide. dfci.harva rd.edu/) was conducted. The t-test was performed to analysed the differences of dysregulation, exclusion and TIDE score between the two groups, while difference of respond ratio was determined via chi-squared test. p < 0.05 was cut-off.

| RT-qPCR experiments
Total RNA in HCC tissues and corresponding adjacent tissues were
The specimens were conducted antigen-repair in citrate (Beyotime) followed by dewaxing. The tissues were then infiltrated in 3% H 2 O 2 for 10 min and 8% bovine serum albumin for 15 min for prevention of non-specific binding. Diluted anti-PDL1 antibody (1:50,00; Cat No. 66248-1-Ig; Proteintech) and anti-CD8 antibody (1:10,000; Cat No. 66868-1-Ig; Proteintech) were incubated with the sections overnight. Sections were washed by PBS for twice, and second antibodies were incubated with sections for 2 h. DAB solution (Solarbio) and haematoxylin (Solarbio) were used to stain the sections for 30 and 10 s, respectively. Finally, the HCC sections after treatment were dehydrated, transparent and sealed.

| Cell culture and transfection of small interfering RNAs (siRNAs)
HCC cell HepG2 and Huh7 were obtained from ATCC database, and cultured in DMEM medium (Hyclone) with 10% FBS (Hyclone) at 37°C environment. Targeting APOA5 siRNAs and scramble siRNA were constructed by iGeneBio. The sequences of siRNAs for targeting APOA5 was 5′-GAGCA AGA CCU CAA CAA UAUG −3′ and 5′-CGAUG GAU CUG AUG GAG CAGG −3′. The transfection of siRNAs were conducted by lipo2000 (Invitrogen) according to process provided by manufacturer.

| Western blotting
Total protein in HCC cells were extracted by RIPA buffer (Boster) containing with 1% PMSF (Boster). Following by detecting concentration by BCA method, proteins were separated by 12.5%

| Reactive oxygen (ROS) detection
The cells were seeded in 6-well plates in triplicate and cultured for the appropriate period of time until 70% confluence was reached.
To measure ROS levels in whole cells, the cells were cultured for 24 h in fresh DMEM and stained with 10 μM 2′,7′-dichlorodihydro fluorescein diacetate (DCFH-DA; Invitrogen, USA). The ROS level was assessed using flow cytometry and analysed using FlowJo (version: 7.4.1).

| Data statistics
The data results were analysed in SPSS 19.0. Differences between the two groups were determined by t-test, while one-way analysis of variance was used to measure the differences between multigroups. p < 0.05 was significant.

| Inflammation-and ferroptosis-molecular subtypes was associated with HCC patient prognosis
In order to obtain different subtype HCC tissues with inflammation signature, characteristic inflammation-related genes were conducted for consensus clustering analysis ( Figure 1A These evidences indicated that inflammation-and ferroptosis-molecular subtypes was associated with HCC patient prognosis.

| Identification of IAF genes in HCC
We performed DEGs analysis in inflammation clusters (C3 vs. others clusters), total 106 upregulated genes and 146 downregulated genes were found ( Figure 3A

| Construction of risk model based on significant IAF genes
Total 224 IAF genes were enrolled in LASSO Cox analysis, and 201 genes with high similarity were removed ( Figure 5A,B). Then, resid-

| Risk model served as prognostic biomarker in HCC individuals
We checked the effects of risk model on prognosis evaluation in HCC tissues from the train cohort TCGA first. The HCC tissues in TCGA was divided into high-and low-risk groups based on the median of risk score ( Figure 6A). A higher proportion of deaths was found in the HCC individuals with high risk ( Figure 6B Figure 7D). Expression of APOA5, CXCL8 and G6PD was also elevated in HCC tissues from ICGC, whereas expression of ADH4, CFHR3, FTCD and PON1 was decreased ( Figure 7E).
Similarly, enrolling the information of sex, age, tumour stage and risk score into univariate and multivariate Cox regression analysis, risk score both had significance ( Figure 7F,G). All these evidences exhibited that the risk model established by the significant IAF genes can act as prognostic biomarkers for HCC tissues.

| The risk model reflected the immune characteristic and helped to predict the respond of ICI in HCC
As inflammation and ferroptosis crosstalk with immune environment, we then analysed whether the risk model can reflect the immune characteristic in HCC. HCC tissues from train cohort (TCGA) and test cohort (ICGC) was merged, and ESTIMATE algorithm was performed to measure immune parameters in HCC tissues in the highand low-risk groups. Elevated immune score and ESTIMATE score were observed in the HCC tissues in the high-risk group ( Figure 8A).

Moreover, expression levels of 22 immune cells were calculated by
Cibersort ( Figure 8B). Higher levels of CD8 T cell, naïve CD4 T cell, activated memory T cell, M0 macrophages and resting dendritic cell were observed in HCC tissues in the high-risk group, while the level of  Figure 8C). These evidences indicated that HCC tissues in the high-risk group tend to be 'hot tumours', and risk model may have the function of identifying 'hot tumours'. Therefore, we then analysed whether risk model can help to predict the respond of ICI treatment in HCC. TIDE analysis was performed, risk score was positively related to IFNG score (R = 0.32, p < 0.01; Figure 8D) and Merck18 (R = 0.32, p < 0.01; Figure 8E).
Moreover, through TIDE analysis, MSI score was found higher in HCC tissues in the high-risk group compared with those in the low-risk group ( Figure 8F). Elevated dysregulation score combined with reduced exclusion and TIDE score were discovered in HCC tissues with high risk ( Figure 8G). Furthermore, HCC tissues in the high-risk group was predicted to had higher respond rate in ICI treatment ( Figure 8H).
These evidences indicated that the risk model may had function to predict 'hot tumour' and the respond of ICI treatment.

| The risk model served as diagnostic biomarker for distinguish adjacent tissues and HCC tissues
We verified the practicability of risk model in 40 pair HCC tissues and adjacent tissues from our research group. Levels of the seven IAF genes including ADH4, APOA5, CFHR3, CXCL8, FTCD, G6PD and PON1 in each HCC tissue and adjacent tissue were checked by qRT-PCR ( Figure 9A). Levels of ADH4, CFHR3, FTCD and PON1 were decreased in the HCC tissues compared with adjacent tissues, and APOA5, CXCL8 and G6PD expression was elevated ( Figure 9B).
We substituted the expression of these genes into the formula (risk model). Interestingly, the risk score was also highly expressed in HCC tissues ( Figure 9C). Followed by performing ROC analysis,  The corelationship between risk score and IFNG score. (E) The co-relationship between risk score and Merck18 score. (F) MSI score between the high-and low-risk groups. (G) Exclusion score, dysregulation score and TIDE score between HCC tissues in the high-and low-risk score. (H) The responder rate was predicted in HCC tissues between the high-and low-risk groups. *, p < 0.05; **, p < 0.01.  Figure 10A). We divided our HCC patients into high-and low-risk groups based on medium of risk score, and we found that the patients with MSI-H and MSI-L was significantly more in the high-risk group in comparison with the lowrisk group ( Figure 10B). Similarly, we found that patients from our research group in the high-risk group had higher TMB in comparison to the low-risk group. Furthermore, through performing IHC, higher PDL1 expression and more immune cells with CD8 expression were found in the HCC tissues having high risk ( Figure 10C,D). As higher MSI, TMB, PDL1 and more CD8 cells were biomarkers for respond of ICI, 16 we considered that risk model can help us to distinguish patients may respond for ICI, thus optimising the choice of immunotherapy.

| APOA5 was a novel target for inhibiting HCC via upregulating inflammation and ferroptosis
Moreover, after reviewing previous studies, we found that, as a gene in lipid regulation, 17 the effects of APOA5 on HCC progression, inflammation and ferroptosis has not been reported in previous studies, thus catching our attention. Utilising two siR-NAs, APOA5 knockdown cells were constructed ( Figure 11A,B).
Through performing ELISA, we found that HepG2 and Huh7 with APOA5 knockdown had elevating TNFα ( Figure 11C) and IFNγ levels ( Figure 11D). Then, we detected the levels of MDA, a product of lipid peroxidation in HCC cells, and found that suppression of APOA5 in HCC cells increased the MDA levels in HepG2 and Huh7 cells ( Figure 11E). Moreover, we found that ROS levels ( Figure 11F) were significantly increased in HepG2 and Huh7 cells with APOA5 knockdown, while the GSH ( Figure 11G) in cells were reduced. Moreover, we found that the iron levels in HCC cells were increased ( Figure 11H). These results indicated that knockdown of APOA5 may increase inflammation and ferroptosis. Furthermore, we found that knockdown of APOA5 reduced cell proliferation ( Figure 1I) and colony formation ( Figure 1J). Therefore, we considered that APOA5 was a novel target for inhibiting HCC via upregulating inflammation and ferroptosis.

| DISCUSS ION
Recently, immunotherapy, radio-chemotherapy and targeting therapy were used jointly to suppress HCC progression and elevate the prognosis. 18 HCC was highly heterogeneous, and the degree of inflammation and immune cell infiltration in its microenvironment affect the efficacy of treatment of HCC, especially immunotherapy.
However, due to low MSI and immune suppressive microenvironment, patients with HCC commonly had low respond rate of immunotherapy including ICI. 19 More and more recognized that inflammation plays as 'double-  22 Previous studies indicated that ICI based on the data obtaining from high-throughput sequencing is helpful for the precise treatment of HCC. 23,24 Herein, therefore, we first aimed to explore novel biomarker associated with inflammation and ferroptosis in HCC tissues, and its effect on guiding immunotherapy via analysing high-throughput sequencing data.
In the current study, analysis of the gene expression profile of HCC tissues in TCGA, total 224 IAF genes were identified. Then, through selection of lasso Cox, univariate Cox regression analysis F I G U R E 11 Knockdown of APOA5 can significant reduce the proliferation and colony formation combining with elevating inflammation and ferroptosis levels. (A) RT-qPCR was used to detect the mRNA level of APOA5 in HepG2 and Huh7 cells after APOA5 knockdown. (B) Western blotting was used to detect the protein level of APOA5 in HepG2 and Huh7 cells after APOA5 knockdown. (C, D) Elisa was used to detect the levels of TNFα and IFNγ in HepG2 and Huh7 cells after APOA5 knockdown. Levels of MDA (E), ROS (F), GSH (G) and iron (H) were measured in HepG2 and Huh7 cells after APOA5 knockdown. (I) CCK-8 was used to detect the proliferation rate of HepG2 and Huh7 cells after APOA5 suppression. (J) Colony formation assay was used to detect the colony formation ability of HepG2 and Huh7 cells after APOA5 suppression. **, p < 0.01. and multivariate Cox regression analysis, seven IAF genes including ADH4, APOA5, CFHR3, CXCL8, FTCD, G6PD and PON1 were used for construction of a risk model. Following checking in train corhort (TCGA) and test cohort (ICGC), risk model was found to have potential to play as an independent biomarker for prognostic evaluation in HCC. Previous studies indicated that 'hot tumour' with more immune cell infiltration, especially T cell infiltration, had higher responder rate for ICI treatment. 25 Interesting, the high-risk group classifying by the risk model was found to had more immune score and higher levels of CD8 T cell, naïve CD4 T cell, activated memory T cell, which were predicted to have higher respond rate to ICI.
These results indicated that the risk model may help to identify 'hot tumour', and had potential to guide the ICI choice.
ADH4-encoded protein belong to the family of alcohol dehydrogenase. In a physiological state, it oxidizes long chain omegahydroxy fatty acids into 20-oxoarachidonate, and induces the reduction of benzoquinones. 26 APOA5-encoded protein is an apolipoprotein, involved in regulating the plasma triglyceride levels. 27 APOA5-encoded protein is a component of high-density lipoprotein. 28 CFHR3-encoded protein is a secreted protein, which belongs to the complement factor H-related protein family. CFHR3-encoded protein is a key factor for complement-regulated inflammatory responses. 29 CXCL8-encoded protein is also named as interleukin-8 (IL-8), and is a major mediator of the inflammatory response. 30 FTCD-encoded protein exhibits both transferase and deaminase activity, induces channels 1-carbon units from formiminoglutamate to the folate pool. 31 G6PD-encoded protein is the rate-limiting step of the oxidative pentose-phosphate pathway, via providing reducing power (NADPH) and pentose phosphates for fatty acid and nucleic acid synthesis. 32 PON1-encoded protein belongs to the paraoxonase family, possessing lactonase and ester hydrolase activity. 33 The protein is secreted by the kidney and liver, and has potential to bind with high-density lipoprotein, regulating the process of lipid synthesis. 34 Dysregulation of these genes had been observed in atherosis, autoimmune disease and tumours, including HCC. For example, ADH4 level was reduced in HCC, and low level of ADH4 was related to poor prognosis. 35 Elevated G6PD was observed in HCC tissues, while it promotes the progression of HCC via suppressing ferroptosis. 36 Similarly, our previous study indicated that CFHR3 was respond to hypoxia, and reduced in HCC with hypoxic microenvironment. 37 Consistent with previous studies, through qRT-PCR, ADH4, CFHR3, FTCD and PON1 levels were found to reduce in the HCC tissues compared with adjacent tissues, and APOA5, CXCL8 and G6PD expression was elevated. It is worth noting that the risk model constructed by them exhibited higher diagnostic value than each single genes.
This may be one of shining points of the point. Moreover, we found that APOA5 was a novel target in HCC. Knockdown of APOA5 can significant reduce the proliferation and colony formation combining with elevating inflammation and ferroptosis levels.
Previous studies that patients with higher MSI and TMB commonly respond for ICI, and they were set as biomarkers to decide whether to proceed ICI. 38 Therefore, we plugged in the expression of each gene into the formula (risk model), and classifying the patients from our research group into the high-and low-risk groups.
Through reviewing clinical characteristic, it was indicated that HCC individuals in the high-risk group had higher MSI and TMB. Finally, we suggested that tissues in the high-risk group had higher PDL1 expression and immune cells with CD8 using IHC. These results were consistent with above-mentioned bioinformatics analysis, verifying that the risk model can distinguish 'hot tumour'. Therefore, we consider the risk model had potential to optimize the choice of ICI in patients with HCC by reflecting the immune characteristics in HCC tissues.

| CON CLUS IONS
In conclusion, we considered APOA5 was a novel target for suppressing HCC via simultaneously elevating inflammation and ferroptosis levels, and the signature constructed by seven IAF genes including ADH4, APOA5, CFHR3, CXCL8, FTCD, G6PD and PON1 can act as a biomarker for optimising the diagnosis, prognosis evaluation and immunotherapy options in patients with HCC.

CO N FLI C T O F I NTER E S T S TATEM ENT
The authors declare no conflicts of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data used to support the findings of this study are available from the corresponding author upon reasonable request.