Identification of molecular subtypes and prognostic model to reveal immune infiltration and predict prognosis based on immunogenic cell death-related genes in lung adenocarcinoma

ABSTRACT

Immunogenic cell death (ICD) has been increasingly indicated to be related to caners. However, ICD’s role in Lung adenocarcinoma (LUAD) is still not well investigated. Clinical data along with associated mRNA expression profiles from LUAD cases were collected in TCGA and GEO databases. 13 ICD-related genes were identified. Relations of ICD-related genes expression with prognosis of patients, tumor immune microenvironment (TIME) was analyzed. Then, candidate genes were identified and the prognostic signature were constructed. Afterwards, one nomogram incorporating those chosen clinical data together with risk scores were built. Finally, the effect of HSP90AA1, one gene of the prognostic signature, on LUAD cell were analyzed. Two clusters were identified, which were designated as the ICD-high or -low subtype according to ICD-related genes levels. ICD-high subgroup showed good prognosis, high immune cell infiltration degrees, and enhanced immune response signaling activity compared with ICD-low subtype. Moreover, we established and verified the risk signature based on ICD-related genes. High risk group predicted poor prognosis of LUAD independently and presented negative association with immune score and immune status. Furthermore, nomogram contributed to the accurate prediction of LUAD prognostic outcome. Finally, HSP90AA1 levels were remarkably elevated within tumor cells in comparison with healthy pulmonary epithelial cells. HSP90α, HSP90AA1 protein product, promoted growth, migration, and invasion of LUAD cells. Molecular subtypes and prognostic model were identified by incorporating ICD-related genes, and it was related to TIME and might be adopted for the accurate prediction of LUAD prognosis.

1. Introduction

Lung cancer (LC) ranks the top across various cancers globally in terms of its morbidity and mortality, and non-small cell lung cancer (NSCLC) occupies around 85% [1]. Because there is no representative clinical symptom, many LC cases are currently diagnosed at an advanced stage [2]. Immunity has an important effect on LC occurrence because cancer cells escape immunosurveillance by the activation of inhibition checkpoint pathways for inhibiting anticancer T-cell responses [3]. Immunotherapies (including immune checkpoint inhibitors (ICIs), monoclonal antibodies (mAbs), cancer vaccines, and costimulatory agonists) are identified as the most efficient therapeutic modalities in cases with lung adenocarcinoma (LUAD) [4]. Cancer immunotherapy is becoming more developed, T cell responses to targeted immune checkpoint therapies are becoming more understood, and medicines that successfully block immune checkpoints have been identified, which aids in the investigation of precise prognostic biomarkers for immunotherapy within LUAD [5].

ICD, a regulatory cell death (RCD) type, results in adaptive immunity and facilitates anticancer immunity propagation [6]. ICD could be triggered via cellular stressors, such as IR, great hydrostatic pressure, photodynamic therapy (PDT), targeted antitumor drugs, and chemotherapy [7]. It can trigger adaptive immunity depending on CD8⁺ T cells through the emission of damage-associated molecular patterns (DAMPs) [8]. According to preclinical research, candidate ICD inducer treatment in tumor cells can induce excessive production of cytokines and DAMPs associated with adaptive immunity initiation. The above immunostimulatory factors are cellular nucleic acids, ATP, ANXA1, HMGB1, cytokines such as CCL2, IFN-I, CXCL1, and CXCL10, together with ER chaperones such as ERp57, calreticulin, Hsp70, and Hsp90. Such ICD-related cytokines and DAMPs can 1) promote anti-presenting cell (APC) maturation, recruitment, as well as cross-presentation activity, 2) enhance dying cell adsorption via APCs, and 3) enhance recruitment of neutrophils and T-cells. Cancer immunotherapy suggests harnessing immunity to induce anticancer immunity [9–11]. As discovered in more and more articles, ICD has an effect on eliciting antitumor immunity.

Tumor immune microenvironment (TIME) can reflect immune landscapes within tumor microenvironment (TME), and it has an important effect on carcinogenesis and progression [12]. Immune cells are related to cellular reprogramming, and in this process, cancer cells in TME are modified on their own by producing different biological factors, as a result, neighboring cells can determine tumor development and survival [13]. Within TME, tumor-infiltrating immune cells (TIICs) are the major non-tumor component, and they have a critical effect on prognostic prediction in LUAD cases. Therefore, TIME is greatly significant for tumor occurrence and development; besides, it is increasingly suggested to be tightly related to LC pathogenesis [14]. It is important to evaluate TIME in lung cancer for the sake of understanding cancer cell immune status, promoting immunotherapy development as well as LUAD prognostic outcome.

The present work identified ICD-high or -low subtype through analyzing ICD-related genes expression for exploring how ICD affected survival and TIME in LUAD cases. As suggested by survival and immune analyses, ICD occurrence was related to high-immune status as well as favorable prognosis. Moreover, risk model was established for evaluating whether ICD-related genes were of prognostic significance, which could predict LUAD prognosis accurately, at the same time, high-risk cases showing poor prognosis exhibited the low immune status. This study can help to explore LUAD-related molecular mechanisms, provide the novel insights into targeted therapeutic strategy for LUAD, and enhance individualized treatment for lung cancer cases.

2. Material and methods

2.1. Datasets

This study obtained RNA-sequencing (RNA-seq) together with associated clinical data (survival and phenotype) in The Cancer Genome Atlas (TCGA) databases for LUAD cohort. We collected LUAD cases through University of California, Santa Cruz (UCSC) Xena (http://xena.ucsc.edu) [15]. First of all, this study converted RNA-seq data into fragment per kilobase million (FPKM) levels, which included gene expression profiles from 453 cancer and 49 non-carcinoma. Furthermore, survival data included information on survival time and phenotype data included clinicopathological characteristics from 523 LUAD cases.

2.2. Cell lines and animals

This study applied the BEAS-2B, normal pulmonary epithelial cellline, together with several lung cancer cell lines (A549, H460, H1229, H1650). Specifically, these cells were cultivated with 5% CO₂ under 37°C. Additionally, we raised 4–6-week-old BALB/c male nude mice (Shanghai SLAC Laboratory Animal Co., Shanghai, China) under the specific pathogen-free (SPF) conditions. Our animal experiments gained approval from Ethics Committee for Animal Experiments of the Qingdao University Animal Care Committee, Shandong, China (Permit Number: SYXK: 2008–0039).

2.3. Consensus clustering

For identifying molecular subtypes associated with ICD, we completed consensus clustering with R software Concensus Cluster Plus function and determined the best cluster number as k = 2–10 with 1,000 replicates to ensure the consistency of results. Furthermore, R software pheatmap function was utilized for creating the cluster map [16].

2.4. Differentially expressed genes (DEGs) identification

To evaluate DEGs, R software Limma package (version: 3.40.2) was adopted. For correcting false-positive TCGA data, we investigated adjusted p-values. DEGs were chosen based on the adjusted P < 0.05 and |fold change (FC)| >2 thresholds.

2.5. Gene set enrichment analysis (GSEA)

We carried out GSEA to detect significant differences in gene levels of ICD-high compared with ICD-low groups by enrichment of MSigDB Collection (c2.cp.kegg.v7.4.symbols.gmt) based on GSEA software (http://www.broadinstitute.org/gsea/index.jsp).

2.6. Immune landscape characterization

With the Estimation of Stromal and Immune cells in Malignant Tumor tissues using Expression data (ESTIMATE) algorithm, Estimated Score, Stromal score, Immune score, together with Tumor purity was determined. MCP Counter immune infiltration analyses were carried out for determining TIIC enrichment. Single sample gene set enrichment analysis (ssGSEA) was adopted for assessing 23 TIIC levels within cancer samples. Additionally, eight transcripts were chosen as immune-checkpoint-related factors, including CD274, CTLA4, HAVCR2, LAG3, PDCD1, PDCD1LG2, TIGIT, and SIGLEC15, and the corresponding expression was analyzed.

2.7. Survival analysis

By using survminer and survival packages in R, overall survival (OS) was compared between two risk groups based on Kaplan-Meier (KM) analysis. Prognostic biomarkers were identified by univariate Cox regression, while the independent predction of OS in LUAD patients was analyzed by multivariate Cox regression.

2.8. ICD-Related risk signature establishment

Significant immune-related genes (IRGs) discovered using univariate Cox regression were added in least absolute shrinkage and selection operator (LASSO) algorithm to calculate coefficients of discovered correlation. The most commonly used regression method is LASSO, which combines variable screening and regularization, improving the predicting ability and interpretability of the obtained statistical model.

2.9. ICD-Related gene model establishment and verification

The TCGA cohort was conducted univariate regression with “survival” function in R for identifying OS-related ICD genes. A prognosis prediction signature was then constructed for training cohort by LASSO with “glmnet” function in R. Later, risk score for each LUAD case was determined below: risk score = ∑ incoefficient * levels for ICD-related genes [17]. The LUAD cases were classified as low- or high-risk group according to median risk score. KM plot was drawn for comparing OS of two groups. The “time ROC” function in R was adopted to draw ROC curves to determine whether our prognostic model was sensitive and specific.

2.10. Nomogram construction and verification

According to clinical factors and risk score, the prognosis prediction nomogram was constructed for guiding clinical decision-making. Besides, clinical factors associated with survival were selected upon the threshold of P < 0.05 through univariate regression. Multivariate regression was later adopted for nomogram construction. Thereafter, the calibration curve was used for plotting nomogram-predicted probabilities as a function of actual measurements [18]. Decision curve analysis (DCA) was carried out for investigating clinical significance of our nomogram. The “rms” and “rmda” in R were adopted for plotting the calibration and DCAs of our nomogram.

2.11. RNA extraction and RT-qPCR

RT-qPCR was performed in line with prior works. Total RNAs were extracted and synthesized into cDNA by means of reverse transcription. Later, cDNA amplification and detection were conducted using the Light Cycler480II (Roche, Mannheim, Germany) in RT-qPCR. Gene expression was explored using 2^−ΔΔCt method, and GAPDH was an internal control. Table 1 displays primer sequences.

Table 1.

Primers utilized in qRT-PCR.

Gene symbol	Sequence 5′-3′
HSP90AA1	F: 5’- CATTACTATGCCGAGTCACATG- 3’ R: 5’- CCAGCTATAACTCTTCCTAGGC- 3’
GAPDH	F: 5’- CTCCTCCACCTTTGACGC- 3’ R: 5’- CCACCACCCTGTTGCTGT- 3’

qRT-PCR, quantitative real-time reverse transcription PCR; F, forward; R, reverse.

2.12. Western-blotting (WB) assay

Total proteins were extracted from cells and tissues, followed by processing with 10% SDS-PAGE for protein separation and transfer onto PVDF membranes (Millipore, Boston, MA, USA). Later, membranes were subjected to overnight primary antibody incubation at 4°C. Table 2 lists antibodies adopted in the present study. Then, HRP-labeled secondary antibody (1:1000, DingguoBio, Beijing, China) was employed for further membrane incubation. At last, target bands were visualized with the electrochemi-luminescencekit (Thermo, Waltham, MA, USA).

Table 2.

Primary antibodies used in western blot.

Antibody	Concentration(WB)	Company
HSP90AA1	1:1000	Abcam
Ki-67	1:1000	Abcam
Bcl-2	1:1000	Cell Signaling Technology
GAPDH	1:2000	Cell Signaling Technology

2.13. Lentivirus establishment and cell transfection

Shanghai Genechem Company Co. Ltd. (Shanghai, China) was responsible for constructing pGCSIL-shRNA-HSP90AA1 and pGC-FU-HSP90AA1cDNA lentiviral vectors. As for HSP90AA1 knockdown, the sequence of 5ʹ-ATTGCCCATGAGATCGGAT-3ʹ was adopted to be the negative control (NC), while that of non-silenced short hairp in RNA (shRNA) was 5ʹ-TTCTCCGAACGTGTCACGT-3ʹ. RT-qPCR and WB were then carried out for verifying stably transfected clones.

2.14. Cell counting kit-8 (CCK-8) assay

Proliferation of HCC cells was analyzed by CCK-8assay (Dojindo, Kumamoto, Japan). Briefly, each well of the 200-well plate was introduced with 100 µL cell suspension, later 10 µL CCK-8 solution was also introduced in each well after reaching cell adherence, followed by 1-h incubation within an culture incubator. At last, the microplate reader was adopted for measuring cell proliferation at 450 nm. Blank control wells were set by adding medium and CCK-8 solution but not cells.

2.15. Subcutaneous xenograft development of nude mice

Four- to six-week-old SPF male nude mice (no specific pathogen) were obtained and acclimatized for a 1-week period. Cells at logarithmic phase were rinsed twice by serum-free medium, followed by preparation of single cell suspension. Later, cell suspension (0.1 ml) containing 5 × 10⁶ A549 cells was subcutaneously injected into each mouse into upper left flank region. Finally, tumor growth rate was determined by measuring tumor length and width at the incubation site.

2.16. Transwell assays

In scratch assay, we grew A549 cells for forming the tight cell monolayer. The 10 µl tip was employed to scrape cells in the monolayer center for creating a cell-free strip. Microphotography (100×magnification) was later carried out to monitor wound clousure at wounding (0-h) and 24-h post-wounding with the LeicaDMIRE2 microscope (Leica Microsystems Imaging Solutions Ltd, Cambridge, UK). Each assay was conducted in triplicate.

In Transwell invasion assay, we coated 24-well transwell plate (8 µm, Corning Costar Corp, Corning, NY, USA) with Matrigel (BD Biosciences, Franklin Lakes, NJ, USA) under 37°C for a 45-min period. Top chamber was introduced with A549 cells (1 × 10⁵ cells/well) transfected with pGCSIL-shRNA-HSP90α or non-silenced shRNA, whereas medium that contained serum was introduced into bottom chamber. At 48-h post-culture, cotton swabs were employed to eliminate non-invading cells on upper membrane surface. Whereas those invading cells were subjected to fixation within 4% paraformaldehyde (PFA), followed by 15-min staining using 0.1% crystal violet under ambient temperature. After removing excessive dye, invading cell number was determined from 10 fields with the 200× objective lens, while results were presented by mean±SD. Each assay was carried out twice within triplicate wells.

2.17. Statistical analysis

GraphPad Prism 5v5.01 software (LaJolla, CA, USA) was used for statistical analysis. Student’s t-test was utilized for comparing differences of both groups, while one-way ANOVA and Post-HocTest (least significant difference, LSD) for comparing data of both groups. Every assay was carried out in triplicate, and data were represented by mean±SD. *p < 0.05,**p < 0.01,***p < 0.001 stood for statistical significance.

3. Results

3.1. Consensus clustering conducted to identify two ICD-related subtypes

Thirteen ICD-related genes, CALR, FASLG, HMGB1, PDIA3, ANXA1, HSP90AA1, IFNG, CXCL10, TLR4, CASP8, BAX, P2RX7, and TNF, were identified using a large body of literature [19–24]. The STRING database was used to analyze the protein-protein interaction (PPI) network for discovering relationships among the ICD-related genes (Figure 1a). Furthermore, gene expression profiles within LUAD and non-carcinoma samples were compared. About 12 out of 13 ICD genes, such as CALR, ANXA1, HMGB1, PDIA3, HSP90AA1, IFNG, CXCL10, TLR4, CASP8, BAX, P2RX7, and TNF, were found to be overexpressed in lung cancer (Figure 1b, c). Following that, ICD-related genes clusters in LUAD were examined using consensus clustering. Following k-means clustering, we detected two TCGA clusters showing different ICD genes expression profiles Figure 1(d-f). Compared with cluster 2, the expression of 10 genes (FASLG, ANXA1, HMGB1, HSP90AA1, IFNG, CXCL10, TLR4, CASP8, P2RX7, and TNF) increased and 2 genes (CALR, PDIA3) decreased in cluster 1(Figure 1g, h). Therefore, clusters 1 and 2 were designated to be ICD-high and ICD-low subtypes separately. Furthermore, on the basis of survival analyses, 2 ICD-based subtypes showed diverse clinical survival. ICD-high subtype was associated with the favorable prognostic outcome, while ICD-low subtype had the dismal prognostic outcome (Figure 1i).

Figure 1.

Consensus clustering conducted to identify ICD-related subtypes. (a) Protein-protein interactions (PPIs) of ICD-related genes; (b,c) heatmap and histogram displays expression levels of 13 ICD genes in healthy and LUAD samples derived from TCGA database; (d) heatmap exhibits consensus clustering solution (k = 2) of 13 genes for 453LUAD samples; (e, f) delta area curve for consensus clustering suggests relative change of area under cumulative distribution function (CDF) curve at k = 2–10; (g, h) heatmap and histogram for 13 ICD-related gene expressions within diverse subtypes, with red and blue indicating high- and low-expression groups, separately; (i) KM curves for OS in ICD-high and ICD-low subtypes. *P < 0.05, **P < 0.01, and ***P < 0.001.

3.2. DEGs and pathways between diverse ICD subtypes

Based on these findings, ICD-high subtype had good prognosis, while ICD-low subtype had poor prognosis, respectively. Therefore, this study investigated critical DEGs as well as pathways of each subtype to further understand prognosis-regulating mechanism. There were 1676 DEGs found in ICD-high compared with -low subtypes, including 867 showing upregulation in ICD-high subtype and 809 with down-regulation (Figures 2a, b). For identifying related pathways enriched into ICD-high subgroup, GSEA was conducted on both risk subgroups. As a result, gene sets showed differential enrichment into ICD groups due to their close relation with immune pathways, including natural killer cell-mediated cytotoxicity, T-cell receptor pathway, Toll-like receptor pathway, and antigen processing and presentation(Figure 2c). Other signal pathways are displayed in the supplemental Table 1. Taken together, the ICD-high subtype may be associated with an immune-active microenvironment.

Figure 2.

Discovery of DEGs and associated pathways within diverse subtypes. (a) Volcano plot displays DEGs distribution in ICD-high and-low subtypes from TCGA cohort upon |log2 FC| > 1 and P < 0.05 threshold; (b) heatmap displays DEG levels within diverse subtypes; (c) GSEA analyzes associated pathways in both subtypes.

3.3. Tumor microenvironment landscape of both subtypes

ICD is increasingly being recognized as having a significant impact on anticancer immunity activation [25]. The current study looked at TIME components from both subtypes. Overall, ICD-high group had increased stromal score, immune score, and estimate score, whereas decreased tumor purity in comparison with ICD-low group (Figure 3a). Moreover, according to the heatmap (Figure 3b), the ssGSEA algorithm-based immune landscapes revealed a significant difference in both groups, and ICD-low subtype had the low immune status. Furthermore, the MCP Counter algorithm revealed that ICD-high groups had significantly higher proportions of T cells, CD8⁺T cells, NK cells, cytotoxic lymphocytes, myeloid dendritic cells, endothelial cells, monocytic lineage, fibroblasts, and neutrophils (Figure 3c). Furthermore, the ICD-high group showed up-regulation of the majority of immune checkpoint genes and human leukocyte antigen (HLA), whereas the ICD-low group showed the opposite results Figure 3(d,e). In general, ICD-high group had an immune-hot phenotype, whereas ICD-low group had an immune-cold phenotype.

Figure 3.

Immune landscapes of both subtypes. (a) Violin plots displays median together with quartile of stromal score, immune score, estimate score and tumor purity via ESTIMATE algorithm; (b) heatmap represent TIICs levels of both subtypes according to GSEA; (c) histogram shows TIICs with significant differences between two subtypes through MCP Counter;(D, E) histogram presents several differentially expressed immune checkpoints; (d) and HLA genes; (e) in both subtypes.*P < 0.05, **P < 0.01, and ***P < 0.001.

3.4. ICD risk signature establishment and verification

The prognostic model was later built by incorporating ICD-related genes. ANXA1, HSP90AA1, and TLR4 were linked to patient survival in Cox univariate regression (Figure 4a). ANXA1, HSP90AA1, TLR4, HMGB1, and TNF were chosen for the prediction model after being validated using LASSO regression (Figure 4b). In general, the risk score was calculated as Risk score = (0.136835644) * HSP90AA1 level + (0.0378317051077513) * ANXA1 level + (−0.108113539536896) * TLR4 level +.(0.002127759) * HMGB1 level +(−0.03961569) * TNF level. Our developed risk model classified lung adenocarcinoma cases as low- or high-risk (Figure 4c). Following that, relation of risk score with survival status was investigated. Low-risk group showed markedly improved alive status in relative to high-risk group. Furthermore, high-risk group showed HMGB1, ANXA1, and HSP90AA1 upregulation compared with low-risk patients. TLR4 and TNF expression, on the other hand, demonstrated the opposite trend (Figure 4c). Kaplan–Meier plot suggested that low-risk patients outlived their high-risk counterparts (Figure 4d). ROC curve analysis yielded acceptable results for the diagnostic value of the as-built risk model (Figure 4e). According to time-dependent ROC (t-ROC) curves, our as-developed risk model accurately predicted 5-year survival, and their 1-, 3-, and 5-year AUCs were 0.67, 0.56, and 0.57 separately (Figure 4e). TCGA cohort’s high-risk score predicted poor OS, which was validated by GEO cohort results (Figure 4f). Moreover, stratified analysis was performed on ICD risk model as well as clinical parameters including age and TNM stage. When patients was stratified by age, KM plots suggested the significant relation with survival, no matter < 65-year-old (p = 0.04) or ≥ 65-year-old (p = 0.02) (Figure 5a). As revealed in (Figure 5b), our built prognostic risk model precisely classified T1–2 stage LUAD cases as either short- or long-term survivors (p = 0.02), while risk scores were not related to survival among LUAD cases at T3–4 stage (p = 0.29). Furthermore, risk scores were tightly associated with survival in N2–3 group (p = 2.8e-3) but not in the N0–1 stage patients (Figure 5c). Risk score was not related to survival in LUAD cases with stage I-II, while high-risk cases at stage III-IV showed the poor prognostic outcome (p = 8.4e-3) (Figures 5d). Taken together, the risk signature showed independent prediction value.

Figure 4.

ICD risk signature establishment and verification. (a) Univariate regression assessed whether ICD genes could be used to predict OS; (b) Lasso Cox analysis detected 5genes closely related to OS from TCGA dataset; (c) distribution of risk scores, survival status in every case, and expression heatmaps showing prognostic 5gene signature from TCGA database; (d) KM analysis suggested risk model’s role in prognosis prediction of TCGA cohort; (e) ROC curve showing the risk model for TCGA cohort; (f) predictive value of risk model was validated based on GSE 30,219 cohort through Kaplan – Meier analyses.

Figure 5.

KM curve showing stratified analysis on ICD prognosis prediction signature regarding the relations with clinicopathological features. (a) OS curve of cases aged >65 years and those <65 years; (b) OS curves for T1+T2 and T3+T4 stages; (c) OS curves for N0+N1 and N2+N3 stages; (d) OS curves for stages I +II and stages III + IV.

3.5. Relation of ICD risk signature with TIIC levels

Because ICD is important for anticancer immune response, the study looked into relation of ICD risk score with TIME.TIME was calculated for both groups using the ESTIMATE algorithm; as a result, low-risk cases showed markedly increased stromal/immune/estimate scores whereas decreased tumor purity compared with high-risk cases (Figure 6a). Moreover, low-risk cases had significantly higher percentages of T cells, CD8⁺ T cells, B lineage, cytotoxic lymphocytes, monocytic lineage, NK cells, neutrophils, endothelial cells, fibroblasts, and myeloid dendritic cells, as suggested by the MCP Counter algorithm (Figure 6b). According to the heatmap (Figure 6c), ssGSEA algorithm-based immune landscapes revealed a significant difference of both risk groups, and low-risk cases had significantly improved immune status. Furthermore, cases showing higher risk scores displayed decreased T cells, NK cells, and CD8⁺ T cells infiltration levels (Figure 6d).

Figure 6.

Immune landscape showing both risk groups. (a) Violin plots show stromal score/immune score/estimate score and tumor purity of both groups via ESTIMATE algorithm; (b) histogram visualizes significantly different TIIC levels in both risk groups through MCP Counter; (c) heatmap represents TIICs levels of different groups according to GSEA; (d) scatter plots display relation between risk score and T cells, CD8+ T cells, and NK cells infiltration levels.note: *P < 0.05, **P < 0.01, and ***P < 0.001.

3.6. Integrated nomogram establishment and calibration

Finally, a nomogram incorporating a risk model as well as clinical characteristics (Figure 7a) was developed to accurately predict LUAD prognosis. Pathological features and risk score were assigned specific scores in the risk model based on their contributions to LC prognosis. Variables such as risk score, clinical stage, and T stage were combined to evaluate survival rate using our constructed nomogram (Figure 7a). Our nomogram revealed 1-/3-/5-year AUCs of 0.76, 0.73, and 0.76, separately (Figure 7b). Calibration curves of nomogram revealed good consistency in OS at 3 and 5 years (Figure 7c). In general, our prognosis prediction nomogram had better performance in predicting patient outcomes of lung adenocarcinoma.

Figure 7.

Nomogram evaluation based on risk score and clinical features. (a) The nomogram constructed by incorporating risk model along with clinical features; (b) time-dependent ROC in the prediction of OS; (c) nomogram calibration at 3 and5 years in the training cohort.

3.7. HSP90AA1 promotes LUAD cell growth, migration, and invasion

HSP90AA1, one gene involved in prediction model, was investigated in human lung adenocarcinoma cells. HSP90AA1 mRNA levels increased in lung adenocarcinoma cells, including A549, H460, H1229, and H1650, compared to healthy BEAS-2B lung cells, as shown in (Figure 8a). HSP90α is protein product of HSP90AA1. Western blot showed that level of HSP90α upregulated in cancer cells compared with BEAS-2B cells (Figure 8a). We conducted RT-qPCR and WB assays for assessing HSP90AA1 knockdown’s effects on A549 cells (Figure 8b). The CCK-8 assay was used to determine how HSP90AA1 affected A549 cell viability, and HSP90AA1 silencing reduced A549 cell viability (Figure 8c). At the same time, Ki-67 and Bcl-2 (an anti-apoptotic protein), two key executors of apoptosis with important effects on programmed cell death, were found to be down-regulated in A549-shRNA-HSP90AA1 cells (Figure 8d). Additionally, xenograft tumor formation assays in nude mice were performed to assess HSP90AA1’s effect on the tumorigenicity of lung cancer cells. Male naked mice were subcutaneously injected with A549-shRNA-Ctrl and A549-shRNA-HSP90AA1 cells. (Figure 8e) shows that the tumor volume in the A549-shRNA-HSP90AA1 cell group was significantly lower than in the A549-shRNA-Ctrl cell group. The wound closure assay revealed that A549-shRNA-HSP90AA1 cells closed wounds later than A549-shRNA-Ctrl cells (Figure 8f). Furthermore, fewer invading A549-shRNA- HSP90AA1 cells were observed when compared to A549-shRNA-Ctrl cells (Figure 8g).These findings indicated that HSP90AA1 promoted LUAD cell growth, invasion, and migration.

Figure 8.

HSP90AA1 upregulated in lung cancer cells and enhanced LUAD cell viability, invasion, and migration. (a) relative HSP90AA1 gene expression in healthy pulmonary epithelial cells BEAS-2B as well as LC cells was determined through RT-qPCR, and HSP90α protein expression was explore using Western blot; (b) knockdown of HSP90AA1 in A549 cells were analyzed by RT-qPCR and WB; (c) as revealed by CCK-8 analysis, HSP90AA1 promoted A549 cell viability; (d)protein expression ki-67 and bcl-2 was analyzed following alteration of HSP90AA1 level; (e) A549 cell growth following alteration of HSP90AA1 level was analyzed through subcutaneous xenograft growth within nude mice; (f) cell migration analyzed through scratch assay and quantified based on open area percentage. Magnification, 100 ×; (g) Transwell invasion assay following HSP90AA1 knockdown within A549 cells. Magnification, 200 × .note: *P < 0.05, **P < 0.01, and ***P < 0.001.

4. Discussion

Over the past few decades, the prediction of LUAD prognosis is mainly conducted according to clinical features, such as age, TNM stage, together with certain serum tumor biomarkers [26]. Nonetheless, such factors have restricted prediction performance. Consequently, it is significant to identify efficient biomarkers, which assists physicians in judging patient prognosis and making treatment decision. As sequencing technology develops, genomics is effective for identifying cancer diagnostic biomarkers [27–29]. Actually, an individual gene just has unsatisfactory prediction performance of LUAD prognosis. Many multigene models exhibit superior prediction performance to individual genes.

The notion of ICD was depicted to be the special RCD form, which can induce antigen-specific adaptive immunity through producing DAMPs and danger signals [30]. Abhishek et al. summarized ICD parameters [31]. On the basis, combined with extensive reading of literature, 13 ICD-related genes (CALR, FASLG, HMGB1, PDIA3, ANXA1, HSP90AA1, IFNG, CXCL10, TLR4, CASP8, BAX, P2RX7, TNF) were finally determined in this study. Consensus clustering represents the creditable method for classifying samples in diverse subgroups according to gene expression matrix [32]. In line with ICD-related genes expression matrix in LUAD cases, first of all, consensus clustering was conducted to obtain two molecular subgroups that displayed significant difference in OS. Subsequently, functional analysis was carried out of both subgroups. Based on identified DEGs, GSEA analyses were performed. GSEA has been the typical approach to integrate gene expression profiles, which can also be applied in the direct clarification of gene set expression of diverse groups [33]. According to our GESA results, DEGs indicated close relationship to immune pathways such as natural killer cell-mediated cytotoxicity, T-cell receptor pathway, Toll-like receptor pathway, and antigen processing and presentation.

Then immune analysis was completed for exploring ICD-related genes’ functions in LUAD immune landscape. TIME takes great part in predicting patient prognosis, because tumor development is related to neighboring stromal modification, and immune cells are the important tumor stromal components [34]. ESTIMATE algorithm has been developed as the innovative approach for inferring tumor purity, immune/stromal cell proportions within tumor in line with gene levels [35]. ESTIMATE algorithm-derived immune scores can display quantification of immune components within tumor tissues, which reflects TIME [36]. Tumor purity indicates cancer cell percentage within cancer tissue, and it is tightly related to prognosis. As revealed in previous studies, the increased immune score together with decreased tumor purity predicted dismal prognostic outcome of LUAD [37,38]. Consequently, ESTIMATE was adopted for determining TIME in both subgroups. As discovered, cases having favorable prognostic outcome displayed increased immune score whereas decreased tumor purity, conforming to prior studies. Additionally, MCP Counter and ssGSEA, were performed for assessing immune status in both subgroups. MCP Counter has been the online approach for quantifying ten TIIC subsets [39]. According to MCP Counter results, abundances of nine of ten TIICs were markedly higher in ICD-high subgroup, conforming to ESTIMATE results, indicating the up-regulation of immune landscape. ssGSEA results displayed levels of 23 immune-related cell types, as a result, cases of ICD-high subtype were under the high immune status, thus verifying ESTIMATE and MCP Counter results. Collectively, the immune status and LUAD patients’ prognosis were clarified into different groups according to ICD-related genes expression accurately.

For better validating ICD’s role in TIME and exploring ICD-related genes’ effect on predicting LUAD prognosis, the prognosis prediction risk model was built by incorporating ICD-related genes and validated with validation cohort. Five genes, ANXA1, HSP90AA1, TLR4, HMGB1 and TNF, were chosen for the prediction model after being validated using LASSO regression, that are tightly related to LUAD genesis and development. The gene of HSP90AA1 (heat shock protein 90 alpha family class A member 1) is situated on the chromosomes 14q32.2 [40]. HSP90α, the protein product of HSP90AA1, is considered as an important facilitator for “oncogene addiction” and for the maintenance of malignant phenotype on different cancer types [41–43]. HSP90α takes part in various cellular processes like DNA repair, neuronal signaling, immune response and by chaperoning oncogenes in cancer development and progression [44]. In regard to the antitumor immunity, inhibition of HSP90 was shown to result in increased T-cell recognition of tumor cells, prevention of MDSC induction and enhanced effectiveness of immunotherapy [45]. Plasma HSP90α is suggested as a pan-cancer biomarker and a prognostic indicator for immunotherapy. Many client proteins of HSP90α are known oncogenic drivers that can regulate tumor intrinsic pathways, some of which may provide a route of interference with response to immunotherapy [46]. Studies provide evidence for combining HSP90α inhibition with immunotherapy. Preclinical studies using the first generation HSP90α inhibitor 17-DMAG, provided initial evidence of HSP90α as a tumor intrinsic molecule that could be targeted to enhance responses to immunotherapy in a vaccine model [47]. Inhibition of HSP90α using the potent 2nd generation inhibitor ganetespib, enhances T cell-mediated killing of melanoma cells [48]. We also know that ganetespib potentiates responses to anti-CTLA4 and anti-PD-1 immunotherapy in vivo [49]. While HSP90α inhibitors have been exploited for their cytotoxic effects on tumor cells, the role of HSP90α in immunomodulation, as described in in ex vivo and in vivo studies, raises the possibility that prolonged or continuous exposure to HSP90α inhibitor treatments might be detrimental to an anti-tumor immune response [50]. Inducible Hsp90α in plasma was reported to be high in lung tumor patients as compared to normal healthy controls and the expression level was correlated with the development and progression of cancer. HSP90 is widely studied as a therapeutic target since it is overexpressed by tumor cells, including lung tumor cells. In the extracellular space, HSP90α was demonstrated to enhance cancer cell invasiveness and migration, modulate TME, and promote metastasis formation [51]. In this work, we demonstrated that HSP90α had its critical effects on LUAD cell growth, invasion, and migration.

According to survival analysis, our constructed risk model showed strong ability in predicting LUAD survival, either in training or verification cohort. With the rapid development of bioinformatics technology and the popularity of sequencing technology, there is an increasing number of research works on the use of sequencing data to construct prognosis or diagnosis-related signatures. Liu et al. used the data from TCGA and to construct an ICD- related gene signature that can be used to judge the prognosis of IDH wild-type glioblastoma multiforme patients and the AUC of the signature is between 0.65 and 0.76 [52]. Liao et al. also con structed a mRNA expression-based stemness index (mRNAsi) related gene signature and the AUC of the signature was 0.64–0.68 when the patients using the TCGA data base were used for prognosis analysis [53]. In this study, risk model constructed by incorporating ICD-related genes performed well in predicting LUAD prognosis and low risk group has better prognosis no matter the age, tumor stage or metastasis status based on independence analyses in subgroup. ESTIMATE, MCP Counter and ssGSEA showed evidently increased stromal score, immune score, immune status whereas decreased tumor purity in low risk group, suggesting that some immune cells were activated and had increased activity, which may be the cause of the better prognosis for patients with low risk scores. Liao et al. also found that, high tumor immune cell infiltration (ICI) scores characterized by the T cell receptor signaling pathway, B cell receptor signaling pathway, and natural killer cell – mediated cytotoxicity positively related to better prognosis [54]. At last, the nomogram that incorporated clinical factors and risk score was constructed and validated, which performed well in predicting survival. In conclusion, our constructed risk model based on ICD-related genes could be adopted for predicting prognosis of LUAD patients, besides, ICD was related to TIME.

In general, this work focused on the comprehensive analysis on ICD- related genes for exploring how ICD affected survival and TIME in LUAD cases. Additionally, the ICD-related genes-based risk score model was established for evaluating ICD-related genes’ prognostic significance for LUAD. According to our results, ICD was related to TIME alterations and overall survival, and should be paid more attentions during the determination of anti-LUAD therapeutic strategy, and it may be the candidate target for individualized treatment.

Funding Statement

This work was supported by the National Natural Science Foundation of China, grant number 81803058; the Chinese Society of Clinical Oncology, grant numberY-QL2019-0337; Chinese Postdoctoral Science Foundation, grant number 2022M711316; Natural Science Foundation of Shandong Province in China, grant number ZR2023MH102; Mount Tai Scholar Project Special Fund, NO.tsqn202211363; Special Fund for Qilu Health Leading Talent Cultivation Project.

Abbreviation

GAPDH

Glyceraldehyde Phosphate Dehydrogenase

CCK-8

Cell Counting Kit-8

RT-qPCR

Real-Time Quantitative Polymerase Chain Reaction

HMGB1

High Mobility Group Box 1

FASLG

Factor Related Apoptosis Ligand

PDIA3

Protein Disulfide Isomerase A3

ANXA1

Annexin-A1

HSP90AA1

Heat shock protein HSP 90α

IFNG

Interferon γ

TLR4

Toll-Like Receptor 4

CXCL10

C-X-C Motif Chemokine Ligand 10

CASP8

caspase 8

BAX

Bcl2 Associated X Protein

P2RX7

Purinergic Receptor P2X

LGIC7

Ligand Gated Ion Channel 7

TNF

Tumor Necrosing Factor

CCL2

Chemokine Ligand 2

ATP

Adenosine-Triphosphate

Disclosure statement

No potential conflict of interest was reported by the author(s).

Author contributions

Yinying Dong and Haijun Lu contributed to study conception, manuscript reviewing, and supervising. Hao Song, Mingjin Xu, Jian Liu were responsible for collecting and analyzing data. Xiao Yu, Qingfeng Chen, Xiangyin Sun and Qiuxiao Wang were in charge of manuscript drafting, table and figure drawing. Bin Zheng, Xiaomeng Ji and Ruimei Ren performed experiments. The authors agreed the final version for publication.

Data availability statement

All data utilized in this work can be acquired from corresponding authors on request.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/15384101.2023.2300591.