Hypoxic-immune Model Based on Lung Squamous Cell Carcinoma Reveals the Effects of In ammation and Hypoxia on Drug Resistance, CD8 Cell Depletion and Prognosis


 Background: Lung squamous cell carcinoma (LUSC) is a malignant tumor with high mortality and poor prognosis. More evidence shows that hypoxia and the immune environment play an essential role in cancer progression, but the specific impact on lung squamous cell carcinoma is unclear. This study mainly establishes immune and hypoxia risk models to predict the prognosis of patients and formulates personalized treatment plans for patients according to the specific conditions of hypoxia regulation and immune invasion in high-risk groups. Results: Based on the combined use of multiple data, 380 hypoxia and immune co-related genes (HMGs) were obtained， to establish the risk model of immune and hypoxia. Through the use of comprehensive analysis methods, the model has a high predictive value. The survival rate of the high-risk group is low, and the CD8-T cell depletion factor is widely distributed in high-risk groups. It has a large number of neutrophils and low CD8 cells. In addition, hypoxia, inflammation, and drug resistance-related pathways are also abundant in high-risk groups. We also found that high-risk patients were generally resistant to chemotherapeutic drugs. Finally, we constructed a competing endogenous RNA (CeRNA) network closely related to risk genes, including 9 mRNAs, 10 MicroRNAs (miRNAs), and 16 long non-coding RNAs (lncRNAs). Conclusions:This study specifically analyzed the effects of hypoxia regulation and immune Infiltration on the prognosis of patients. It provided a new idea for patients to improve the prognosis and personalized treatment.


Introduction
Lung squamous cell carcinoma is a common malignant tumor with high mortality, accounting for about 30% [1][2] of lung cancer. Although signi cant progress has been made in cancer treatment, lung cancer patients' average 5-year survival rate is less than 18% [3]. The reasons may be the limitations of molecular targeted therapy and cytotoxic chemotherapy drug resistance to patients [4][5][6]. However, immune checkpoint programmed death 1 (PD-1) / programmed death-ligand 1 (PD-L1) is blocked by immune checkpoint block (ICB). Therapy has made a signi cant breakthrough in improving the prognosis of patients, but only 20% of patients will obtain the therapeutic effect. Some ICB treatments will also hurt the prognosis of patients [7][8][9][10], including immunerelated in ammatory reactions and cytokines, which can promote tumor progression and immune escape. [11][12][13] Interestingly, hypoxia is closely related to the formation of in ammatory environment, tumor development, and drug resistance [13][14][15][16][17]. Therefore, establishing and screening immune hypoxia co-related risk models and prognostic markers will play an essential role in improving patients' prognoses and personalized treatment.
In this study, the immune and hypoxia-related risk model established by us has a high ability to predict the prognosis of patients. The high-risk group is signi cantly enriched in hypoxia, in ammation, drug resistance signal pathways, many in ammatory cells, the CD8 cell depletion factor, and low CD8 cells.
This phenomenon reveals that hypoxia regulation and immune invasion have an essential impact on patients' prognosis and provide a new thinking scheme for the treatment of patients in the future.

Data acquisition and preprocessing
We downloaded RNA sequencing (RNA-Seq) gene expression data of 49 normal groups and 502 tumor groups from the Cancer Genome Atlas ( TCGA ) (https://portal.gdc.cancer.GOv), a variety of corresponding clinicopathological information, including age, gender, survival time, survival status, T stage, M stage, Nstage, TMNstage, At the same time, from the Gene Expression Omnibus (GEO) (https://www.ncbi.nlm.nih.GOv/GEO). Then, based on the HALLMARK data set in gene set enrichment analysis (GSEA), I downloaded 200 genes related to the hypoxia pathway. From the TIMER (The Tumor Immune Estimation Resource) website (http://cistrome.dfci.harvard.edu/TIMER ), We downloaded 1650 immune function genes.
We used the string (functional protein binding network) tool to construct the PPI network of 200 hypoxia-related genes; its purpose is to nd out the core hypoxia gene.
Screening of the co-related genes of immunity and hypoxia (HMCORS) We used the "limma" package in R language to analyze the differences between normal and tumor groups. The screening criteria were | logFC | > 2 and FDR <0.05. At the same time, the correlation of signi cantly different genes, immune function genes, and hypoxia core genes were analyzed, respectively. The threshold of screening was the correlation coe cient (| cor | < 0.4, p-value < 0.05). The intersection of immune and hypoxia-related genes was obtained through the online Venny tool https://bioinfogp.cnb.csic.es/tools/venny ) to obtain the co-related genes of immunity and hypoxia (HMCORS).

KEGG and GO functional enrichment analysis
The annotation, visualization, and integrated discovery (David) performed KEGG and GO functional enrichment analysis for immune hypoxia co-related genes.
Establishment and validation of immune and hypoxia-related risk models The most signi cant prognostic genome was rst screened by univariate Cox regression analysis and the minor absolute shrinkage and selector operator (lasso). Then the risk coe cient of each sample was calculated according to the risk gene expression and multivariate Cox regression analysis. The risk score was calculated as follows: risk core = risk coe cient 1 * gene expression 1 +.... risk coe cient N * gene expression N. The high and low-risk groups are divided according to the median value of the risk score. The risk groups in the risk model established by the GEO data set are divided according to the optimal detection thresholds. Kaplan Meier (KM) survival analysis was performed for high and low-risk groups, and receiver operating characteristics (ROC) curves were used to evaluate the predictive value of the risk model for three years (1, 3, and 5 years).
Establishment and validation of a nomogram for predicting prognosis Firstly, univariate and multivariate Cox regression analysis evaluated the risk score and multiple pathological and clinical factors. Then, the nomogram was constructed according to the risk score and numerous clinical characteristics (age, gender, T, M, N, and TMNstage) as prognostic factors. The calibration chart was used to evaluate the stability of the predictive ability of the nomogram.
Scoring of Immune cell in ltration in different risk groups CIBERSOFT calculates the composition ratio of 22 immune cells in vivo based on the deconvolution alGOrithm and the gene expression in each sample (18).
We downloaded the immune score of tcga-LUSC samples calculated, based on the CIBERSOFT method, from the TIMER database (tumor immune estimation resource). At the same time, we used the "e1071", "parallel," and "preprocessor" packages in the R tool to calculate the immune cell ratio of LUSC samples in the GEO database. The standard value of screening is p-value < 0.05. At the same time, we used the "e1071", "parallel," and "preprocessor" packages in the R tool to calculate the immune cell ratio of different risk groups of LUSC in the GEO database. The standard value of screening is p-value < 0.05. Finally, we analyzed the immune scores and immune cell in ltration of different risk groups according to the "ggpubr" package in the R language.
Gene set variation analysis (GSVA) The "GSVA" package of the R tool and multiple data sets (GO, KEGG, HALLMARK) in the gene enrichment analysis tool (GSEA) was used to analyze the enrichment pathways of different risk groups. P < 0.05 was considered statistically signi cant.

Prediction of chemosensitivity of patients in different risk groups
We use the "pRRophetic" package in R language to predict the half-maximal antibiotic concentration (IC50) of patients in different risk groups for chemotherapeutic drugs to infer patients' drug resistance in high-risk and low-risk groups. The primary method is to use the genetics of drug sensitivity in Cancer (GDSC) (www.cancerrxgene.org/) cell line expression Spectrum, TCGA and GEO gene expression pro les were used to construct ridge regression model, and then pRRophetic method in R package was used to calculate IC50 of chemotherapy drugs Establishment of CeRNA network the encyclopedia of rna interactomes (ENCORI) and microRNA target prediction database (miRDB) were used to score the corresponding target genes mRNA, miRNA, and lncRNA of risk genes. Then, correlation analysis and survival analysis were carried out for risk genes and predicted target genes. The screening criteria were the default settings in each online website, with correlation coe cient (COR < -0.1) and p-value < 0.05.

Immune and hypoxia co-related genes
Through the difference analysis between normal and tumor groups, 3632 genes with a signi cant difference were obtained. Then, the PPI network of 200 hypoxia pathway genes was constructed by using a string database, and then 49 immune core genes with the highest degree of protein interaction were screened. Cytoscape software (Cytoscape 3.7.1) was used Mapping. Draw the correlation between all hypoxia core genes Matrix heat map. The correlation between immune function genes and hypoxia core genes and differentially expressed genes were analyzed, respectively, and 399 hypoxia-related genes (HPCORS) and 1750 immune-related genes (IMCORS) were obtained. Finally, we used Venny online software (Venny 2.1.0) to get the intersection of immune and hypoxia-related genes and draw the Wayne diagram. Finally, 380 immune and hypoxia-related genes (HMCORS) were obtained.
Go, and KEGG functional enrichment analysis was performed on HMCORS Online David analysis was performed to analyze the go and KEGG function enrichment of 380 HMGs. We found that these HMCORS were mainly enriched in the immune and hypoxia-related pathways of going analysis, such as response to hypoxia, transforming growth factor, beta activated receptor activity, blood vessel matching, regulation of immune response, vasculogenesis, Negative regulation of cytokine secret, immune response, changing growth factor-beta binding, leukotriene production involved in in ammatory response and gluconeogenesis. The KEGG pathways of immunity and hypoxia mainly include glycolysis/gluconeogenesis, Salmonella infection, ECM receptor interaction, and the HIF-1 signaling pathway.
Establish and predict immune and hypoxia-related risk models Univariate Cox regression analysis was performed on 380 HMCORS, and 44 prognostic genes were screened. Then, Lasso regression analysis was used to select the best prognostic gene. Then the risk score of each sample was calculated through multivariate Cox regression analysis, as follows: risk score = (0.143 * PTGIS expression) + (0.413 * C10orf55 expression) + (0.046 * NAPSA) + (0.142 *MYADM), To visualize the difference between the two groups, we drew PCA principal component analysis. Secondly, the prognosis of the high-risk group in TCGA and GEO data sets was signi cantly worse than that of the low-risk group. ROC curve analysis showed that the AUC values in 1, 3, and 5 years of the model were 0.635 respectively, 0.681, 0.631. The correlation between four risk genes and risk scores was also analyzed. It was found that PTGIS, C10orf55, NAPSA, and MYADM were highly distributed in the high-risk group and positively correlated with the risk score.
To further understand the impact of clinicopathological factors and risk models on the prognosis of patients, we found that risk score was the most signi cant independent factor through univariate and multivariate Cox regression analysis. At the same time, In these pathological groups, we found that the prognosis of high-risk group was worse than that of low-risk group, Including age > 65 (P = 0.003), Mal (P < 0.001), M0 (P < 0.001), N0-1 (P = 0.012), N2-3 (P = 0.001), T1-2 (P = 0.001), stage I-II (P < 0.001).

Establishment and evaluation of prognostic nomogram
The risk score and a variety of clinicopathological parameters were used as prognostic factors to establish a nomogram. The calibration map showed that the nomogram showed better prediction accuracy than the ideal model compared with a perfect model. The c-index of the nomogram for OS prediction was 0.652.

Immune cell in ltration landscape in different risk groups
To further explore the potential mechanism of the model in immunity, the CIBERSOFT algorithm is used to analyze the distribution level of immune cells and scores in high-risk and low-risk groups. We found that macrophages M0, CD4 memory resting, neutrophils, immunity, and matrix scoring are mainly distributed in high-risk groups in the TCGA data set, while B cells memory, B cells naive, T cells CD8 are spread primarily on low-risk groups; at the same time, using Rpackage spearman correlation analysis, the risk score was negatively correlated with B cells naive, T cells CD8, T cells folgular helper and dendritic cells resting, which was opposite to macrophages M0, CD4 memory resting, neutrophils, immune and matrix scores. Tox, PD1, CTLA4, BTL, LAG3, TIM3, MYADM, TGFB1 immune-related factors were higher in the high-risk group. In the GEO dataset, high-risk groups were mainly distributed in macrophages M1, neutrophils, T cells gamma delta, immune and matrix scores. Low-risk groups were primarily distributed in plasma cells, T cells CD8, T cells CD4 naïve, and NK cells resting. Risk scores were positively correlated with B cells naive, T cells CD4 memory resting, T cells gamma delta, neutrophils, eosinophil, immune and matrix scores, In contrast to T cells CD4 naive, T cells CD8 and NK cells resting, CTLA4, BTL, Lag3, tim3, MYADM, and TGFB1 immune-related factors were highly expressed in high-risk groups.
GSVA revealed that the high-risk group was closely related to immune response and hypoxia regulation We performed gene set variation analysis (GSA) on tcga-LUSC RNA Seq data and GEO gene expression data, based on various data sets in GSEA (GO, KEGG, HALLMARK). The GSVA results of the two databases showed that 31, 34, and 19 related functional pathway genes were enriched in high-risk groups, and 10, 7, and 9 immune and hypoxia-related signal pathways were distributed in high-risk groups, respectively (Table 1). Finally, based on different data sets (go, KEGG, HALLMARK), we found that these hypoxia and immune-related pathways were highly distributed in the high-risk group and positively correlated with the risk score.

Sensitivity of patients in high and low-risk groups to chemotherapy drugs
To further explore the drug resistance of different risk groups, we used a prrophic algorithm to calculate the 50% inhibitory concentration (IC50) of various conventional chemotherapy drugs (bleomycin, cisplatin, docetaxel, doxorubicin, etoposide, ge tinib, gemcitabine, and paclitaxel) in patients in high-risk and low-risk groups. In the TCGA training set, doxorubicin, etoposide, and ge tinib had higher IC50 in high-risk groups. In the GEO test set, cisplatin, etoposide, and ge tinib had higher IC50 in the high-risk group.

Establishment and validation of risk genes related CeRNA network
We used the online analysis websites ENCORI (the Encyclopedia of RNA interactomes) and m miRDB (microRNA target prediction database) to analyze four risk gene-related target genes (mRNA, miRNA, lncRNA). Then, based on KM survival analysis and spersman correlation analysis, The target genes opposite to the correlation of risk genes and the survival analysis results were screened. The network diagram and Sankey diagram of Cerna of risk genes and target genes were constructed by Cytoscape software and "ggalluvial" in R language.

Immune and hypoxia co-related genes
The difference analysis between normal and tumor groups obtained 3632 genes with a signi cant difference (Fig. 1A-B). Then, the PPI network of 200 hypoxia pathway genes was constructed by using a string database, and then 49 immune core genes with the highest degree of protein interaction were screened (Fig.  1C). Cytoscape software (Cytoscape 3.7.1) was used Mapping. Draw the correlation between all hypoxia core genes Matrix heat map (Fig. 1D). The correlation between immune function genes and hypoxia core genes and differentially expressed genes were analyzed, respectively, and 399 hypoxia-related genes (HPCORS) and 1750 immune-related genes (IMCORS) were obtained. Finally, we used Venny online software (Venny 2.1.0) to get the intersection of immune and hypoxia-related genes and draw the Wayne diagram (Fig. 1E). Finally, 380 immune and hypoxia-related genes (HMCORS) were obtained.
Go, and KEGG functional enrichment analysis was performed on HMCORS Online David analysis was performed to analyze the go and KEGG function enrichment of 380 HMGs. We found that these HMCORS were mainly enriched in the immune and hypoxia-related pathways of going analysis, such as response to hypoxia, transforming growth factor, beta activated receptor activity, blood vessel matching, regulation of immune response, vasculogenesis, Negative regulation of cytokine secret, immune response, changing growth factor-beta binding, leukotriene production involved in in ammatory response and gluconeogenesis. The KEGG pathways of immunity and hypoxia mainly include glycolysis/gluconeogenesis, Salmonella infection, ECM receptor interaction, and the HIF-1 signaling pathway (Fig. 1F).
Establish and predict immune and hypoxia-related risk models Univariate Cox regression analysis was performed on 380 HMCORS, and 44 prognostic genes were screened ( Fig. 2A). Then, Lasso regression analysis was used to select the best prognostic gene (Fig. 2B-C). Then the risk score of each sample was calculated through multivariate Cox regression analysis, as follows: risk score = (0.143 * PTGIS expression) + (0.413 * C10orf55 expression) + (0.046 * NAPSA) + (0.142 *MYADM) (Fig. 2D), To visualize the difference between the two groups, we drew PCA principal component analysis (Fig. 2E). Secondly, the prognosis of the high-risk group in TCGA and GEO data sets was signi cantly worse than that of the low-risk group (Fig. 2F). ROC curve analysis showed that the AUC values in 1, 3, and 5 years of the model were 0.635 (Fig.  2G), respectively, 0.681, 0.631. The correlation between four risk genes and risk Protein-protein interactive (PPI) network analysis scores was also analyzed. It was found that PTGIS, C10orf55, NAPSA, and MYADM were highly distributed in the high-risk group and positively correlated with the risk score (Fig. 2H-J).

Effects of risk model and clinicopathological parameters on the prognosis of patients
To further understand the impact of clinicopathological factors and risk models on the prognosis of patients, we found that risk score was the most signi cant independent factor through univariate and multivariate Cox regression analysis (Fig. 3A-C). At the same time, In these pathological groups, we found that the prognosis of high-risk group was worse than that of low-risk group, Including age > 65 (P = 0.003), Mal (P < 0.001), M0 (P < 0.001), N0-1 (P = 0.012), N2-3 (P = 0.001), T1-2 (P = 0.001), stage I-II (P < 0.001) (Fig. 3H).

Establishment and evaluation of prognostic nomogram
The risk score and various clinicopathological parameters were used as prognostic factors to establish a nomogram (Fig. 3D). The calibration map showed that the nomogram showed better prediction accuracy than the ideal model compared with a perfect model (Fig. 3E-G). The c-index of the nomogram for OS prediction was 0.652.

Immune cell in ltration landscape in different risk groups
To further explore the potential mechanism of the model in immunity, the CIBERSOFT algorithm is used to analyze the distribution level of immune cells and scores in high-risk and low-risk groups. We found that macrophages M0, CD4 memory resting, neutrophils, immunity, and matrix scoring are mainly distributed in high-risk groups in the TCGA data set, while B cells memory, B cells naive, T cells CD8 are spread primarily on low-risk groups; at the same time, using Rpackage spearman correlation analysis, the risk score was negatively correlated with B cells naive, T cells CD8, T cells folgular helper and dendritic cells resting, which was opposite to macrophages M0, CD4 memory resting, neutrophils, immune and matrix scores (Fig. 4A, C). Tox, PD1, CTLA4, BTL, LAG3, TIM3, MYADM, TGFB1 immune-related factors were higher in the high-risk group (Fig. 4B). In the GEO dataset, The prognosis of the high-risk group was worse than that of the low-risk group (Fig. 5A). Then, based on the CIBERSOFT algorithm, high-risk groups were mainly distributed in macrophages M1, neutrophils, T cells gamma delta, immune and matrix scores. Low-risk groups were primarily distributed in plasma cells, T cells CD8, T cells CD4 naïve, and NK cells resting. Risk scores were positively correlated with B cells naive, T cells CD4 memory resting, T cells gamma delta, neutrophils, eosinophil, immune and matrix scores, In contrast to T cells CD4 naive, T cells CD8 and NK cells resting (Fig. 5B-C, E). CTLA4, BTL, Lag3, tim3, MYADM, and TGFB1 immune-related factors were highly expressed in high-risk groups (Fig. 5D).
GSVA revealed that the high-risk group was closely related to immune response and hypoxia regulation We performed gene set variation analysis (GSA) on tcga-LUSC RNA Seq data and GEO gene expression data, based on various data sets in GSEA (GO, KEGG, HALLMARK). The GSVA results of the two databases showed that 31, 34, and 19 related functional pathway genes were enriched in high-risk groups (Fig. 6A, D, G), and 10, 7, and 9 immune and hypoxia-related signal pathways were distributed in high-risk groups, respectively (Table 1). Finally, based on different data sets (go, KEGG, HALLMARK), we found that these hypoxia and immune-related pathways were highly distributed in the high-risk group and positively correlated with the risk score( Fig. 5B-C, E-F, H-I).

Sensitivity of patients in high and low-risk groups to chemotherapy drugs
To further explore the drug resistance of different risk groups, we used a prrophic algorithm to calculate the IC50 of various conventional chemotherapy drugs (bleomycin, cisplatin, docetaxel, doxorubicin, etoposide, ge tinib, gemcitabine, and paclitaxel) in patients in high-risk and low-risk groups. In the TCGA training set, doxorubicin, etoposide, and ge tinib had higher IC50 in high-risk groups. In the GEO test set, cisplatin, etoposide, and ge tinib had higher IC50 in the highrisk group (Fig. 7A-G).

Establishment and validation of risk genes related CeRNA network
We used the online analysis websites ENCORI (the Encyclopedia of RNA interactomes) and m miRDB (microRNA target prediction database) to analyze four risk gene-related target genes (mRNA, miRNA, lncRNA). Then, based on KM survival analysis and spersman correlation analysis, The target genes opposite to the correlation of risk genes and the survival analysis results were screened (Fig. 8C) (Fig. 9A-C). The network diagram and Sankey diagram of Cerna of risk genes and target genes were constructed by Cytoscape software and "ggalluvial" in R language (Fig. 8A-B).

Discussion
Hypoxia is an inherent feature of most solid tumors, not an accident [19][20]. In recent years, more and more reports have clari ed that hypoxia is closely related to tumor generation, angiogenesis, cancer cell metastasis, and drug resistance [21][22][23][24]. At the same time, immune cells in the tumor microenvironment play an essential role in tumorigenesis and progression [25][26][27]. Interestingly, in recent studies, we found that immunity and immune regulation of the tumor microenvironment are closely related [28][29][30]. Still, at present, In lung squamous cell carcinoma, there is an interaction between hypoxia regulation and immune invasion, and its impact on the prognosis of patients is unclear.
In this study, we made the intersection of immune and hypoxia-related genes to obtain immune and hypoxia-related genes (HMCORS). Then, based on a single factor, multifactor and lasso Cox regression analysis, and immune hypoxia risk model was established, including four risk genes (PTGIS, C10orf55, NAPSA, and MYADM). The high expression of PTGIS can promote tumor-associated macrophages (TAMs) and T-regulatory cells (Tregs), and it is unfavorable to the prognosis of patients with lung cancer, gastric cancer, and ovarian cancer [31]. C10orff has few research reports on cancer at present. In this study, the prognosis of patients with lung squamous cell carcinoma with high expression of C10orf55 is poor. In lung cancer, the expression of NAPSA is closely related to the formation of in ammatory tumor lymphocytes (TIL) and in ammatory immune environment [32]. The expression of MYADM may signi cantly impact the prognosis of prostate cancer patients [33]. The prognostic model based on these four genes is the most signi cant independent prognostic factor compared with other clinicopathological factors. Through nomogram, calibration chart, ROC curve, and two data sets (TCGA, GEO), The application of the model show that the model has high, stable, and accurate prognosis prediction ability. The survival difference analysis is made in different clinicopathological groups and high-risk and low-risk groups. It is found that the patients in the high-risk group have poor prognosis under age greater than 65, Male, M0, T1-2, N0-1, N2-3, and stageI-II groups, which provides new thinking for the selection of clinical scheme. Then, the CIBERSOFT algorithm was used to analyze the immune cell invasion of TCGA training and GEO test sets. We found that the distribution of neutrophils in the high-risk group was higher, and the distribution of CD8 cells in the low-risk group was higher.
At the same time, the immune score and matrix score of the high-risk group were higher than those of the low-risk group. Using three data sets in GSEA (GO, KEGG, HALLMARK), Gsva analysis of the high-risk group showed that the high-risk group was mainly enriched in hypoxia, in ammatory response, in ammatory cells, and TGFB related signal pathway. The in ammatory response in acute diseases often promotes the occurrence and transformation of various cancers, and neutrophils play a signi cant role in in ammatory response [34][35]. Interestingly, in chronic in ammatory and tumor diseases, hypoxia and in ammatory reactions often occur together and affect each other [36][37][38][39]. In discussing drug resistance in different risk groups, the prrophetic algorithm calculates the IC50 of chemotherapeutic drugs from LUSC samples of two data sets. We found that the IC50 values of etoposide and ge tinib are generally higher in high-risk groups. Similarly, In recent studies, many reports have shown that hypoxia and in ammatory response can increase the drug resistance of tumors to etoposide and ge tinib [40][41][42][43][44], which further illustrates the stability and accuracy of the model and detection method. While CD8 cell-related depletion factors (MYADM, BTLA, CTLA4, Lag3, and tim3) are generally elevated in high-risk groups [45][46][47][48][49][50][51][52], Other kinds of lite have also reported that hypoxia and chronic in ammatory response can accelerate the depletion of CD8 cells [53][54], TGFB1 can also inhibit the related functions of CD8 cells and promote tumor metastasis [55].

Conclusion
In conclusion, the immune hypoxia risk model established by us can accurately and stably predict the prognosis of patients. The distribution level of hypoxia and in ammatory response in the high-risk group is high, accompanied by increased CD8 cell depletion factor and drug resistance. These reactions may have a signi cant impact on the prognosis of patients and put forward essential thoughts for personalized treatment of patients