Cuproptosis‐related lncRNAs are correlated with tumour metabolism and immune microenvironment and predict prognosis in pancreatic cancer patients

Abstract

Cuproptosis is a novel cell death pathway, and the regulatory mechanism in pancreatic cancer (PC) is unclear. The authors aimed to figure out whether cuproptosis‐related lncRNAs (CRLs) could predict prognosis in PC and the underlying mechanism. First, the prognostic model based on seven CRLs screened by the least absolute shrinkage and selection operator Cox analysis was constructed. Following this, the risk score was calculated for pancreatic cancer patients and divided patients into high and low‐risk groups. In our prognostic model, PC patients with higher risk scores had poorer outcomes. Based on several prognostic features, a predictive nomogram was established. Furthermore, the functional enrichment analysis of differentially expressed genes between risk groups was performed, indicating that endocrine and metabolic pathways were potential regulatory pathways between risk groups. TP53, KRAS, CDKN2A, and SMAD4 were dominant mutated genes in the high‐risk group and tumour mutational burden was positively correlated with the risk score. Finally, the tumour immune landscape indicated patients in the high‐risk group were more immunosuppressive than that in the low‐risk group, with lower infiltration of CD8+ T cells and higher M2 macrophages. Above all, CRLs can be applied to predict PC prognosis, which is closely correlated with the tumour metabolism and immune microenvironment.

We conducted bioinformatics analysis to construct a cuproptosis‐related lncRNAs prognostic model in pancreatic cancer and verified that cuproptosis‐related lncRNAs were correlated with pancreatic cancer prognosis, tumour metabolism and immune microenvironment. Our study provided a pancreatic cancer prognostic model and might provide some clews of the mechanism by which cuproptosis regulates pancreatic cancer.

1. INTRODUCTION

Pancreatic cancer (PC) is a highly malignant disease with a 5‐year survival rate of less than 10% [1]. Surgical resection is curative, but approximately 80% of patients are already unresectable at diagnosis [2]. Patients with advanced PC benefit a little from chemotherapy and quickly develop resistance to it. Furthermore, a majority of clinical trials in PC failed to achieve clinically meaningful survival benefits. With the development of precision medicine technologies, we could find certain groups of patients benefit from therapies. Consequently, investigating more druggable targets for PC patients is a continuing concern.

Copper ionophore‐induced cell death termed cuproptosis reveals a new cell death pathway that differs from apoptosis, ferroptosis, pyroptosis, and necroptosis [3]. Peter Tsvetkov et al. first reported that copper ionophore‐induced cell death was closely correlated with mitochondrial respiration and protein lipoylation. FDX1 is a key regulator of cuproptosis, and the deletion of FDX1 protects cells from cuproposis. Elesclomol, a potent copper ionophore, displayed a tumour‐killing effect by cuproptosis. Although it had failed in phase 3 clinical trial, elesclomol was verified antitumour activity in melanoma patients with low plasma lactate dehydrogenase levels [4]. This gives us insights that we can identify cuproptosis‐sensitive tumour patients to treat with copper ionophores. With more in‐depth studies of the mechanism of cuproptosis, copper toxicity could be utilised as an antitumour mechanism in specific groups of patients.

Long non‐coding RNAs (lncRNAs) play a role in a series of signalling pathways in tumourigenesis, growth, and metastasis [5]. In pancreatic cancer, lncRNA RGMB‐AS1 and CYTOR promote cancer cell proliferation and migration [6], [7]. LncRNAs can promote gemcitabine resistance in pancreatic cancer, such as PVT1 and HIF1A‐AS1 [8], [9]. PVT1 was reported to activate Wnt/β‐catenin and autophagy pathway. However, HIF1A‐AS1 enhances glycolysis via the AKT/YB1/HIF1α Pathway. Moreover, some lncRNAs were proved to regulate cancer cell apoptosis [10], ferroptosis [11], and pyroptosis [12].

Our study aimed to construct a cuproptosis‐related lncRNAs (CRLs) prognostic model in pancreatic cancer using The Cancer Genome Atlas (TCGA) datasets. Figure 1 presents the workflow of our study. We identified seven prognostic CRLs and constructed and validated our prognostic model using bioinformatics. Additionally, we performed the functional enrichment analysis to figure out potential signalling pathways in two risk groups. At last, we analysed the genomic features and tumour immune microenvironment in two risk groups to identify the underlying mechanisms.

FIGURE 1.

Workflow of our study.

2. MATERIALS AND METHODS

2.1. Datasets

The RNA sequencing and somatic mutation data of PC and clinical characteristics were obtained from the TCGA database (https://portal.gdc.cancer.gov), including 177 PC patients and 4 normal pancreatic tissues (samples without expression matrix or clinical information were excluded). Then the expression of lncRNAs was extracted according to the human gene annotations in GENCODE (https://www.gencodegenes.org/). Cuproptosis regulators were obtained from the previous study (FDX1, LIAS, LIPT1, DLD, DLAT, GLS, PDHA1, PDHB, MTF1, CDKN2A) [3].

2.2. Identification of prognostic cuproptosis‐related lncRNAs and construction of the model

Our bioinformatic analysis was based on the ‘R’ (version 4.1.3) software. We identified cuproptosis‐related lncRNA by the ‘limma’ package in the ‘R’ (|Pearson R| > 0.4, p < 0.001) [13]. Then we obtained prognostic CRLs through the univariate Cox regression analysis (p < 0.05) with the ‘survival’ package. We performed the least absolute shrinkage and selection operator (LASSO) cox regression analysis with the ‘glmnet’ package to construct the prognostic model [14]. The risk score was calculated according to formula (1):

$$\mathrm{R}\mathrm{i}\mathrm{s}\mathrm{k}\mathrm{s}\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{e}=e^{\sum Coef\ast \mathit{Exp}}.$$ (1)

In this formula, Coef is the coefficient and EXP is the expression level of each prognostic cuproptosis‐related lncRNAs. PC patients were grouped based on the value of risk scores. Kaplan–Meier survival curves were drawn via the ‘survminer’ and ‘survival’ package [15].

To assess the predictive power of the risk model, receiver operating characteristic (ROC) curves were drawn by the ‘timeROC’ package [16]. Then we conducted Kaplan–Meier (K–M) survival analysis in different subgroups to validate the predictive capacity. Based on the results of univariate and multivariate Cox regression analysis, a nomogram was constructed using the ‘rms’ package [17].

2.3. Functional enrichment, genomic features, and tumour immune analyses

Differentially expressed genes (DEGs) between risk categories were filtered using the ‘limma’ package (log₂ | FoldChange | > 1.5, p < 0.01). The DEGs were performed Gene Ontology (GO) and Kyoto Encyclopaedia of Genes and Genomes (KEGG) enrichment analyses via ‘clusterProfiler’ package [18]. The KEGG gene set (c2.cp.kegg.v7.5.1.entrez.gmt) was from the Molecular Signatures Database (MSigDB) [19]. We analysed somatic mutation in the risk groups and the results were visualised by the ‘maftools’ package [20]. Then we analysed the correlation between risk scores and tumour mutational burden (TMB). Next, we calculated the fraction of 22 tumour‐infiltrating immune cells in risk groups using CIBERSORT [21]. ESTIMATE scores in tumour tissues were calculated using the ‘estimate’ package [22].

2.4. Statistical analysis

Cuproptosis‐related lncRNAs were identified by the Pearson correlation test. To compare overall survival between subgroups, Kaplan–Meier analysis was used. The difference in risk scores between subgroups was compared using the Kruskal–Wallis test, and categorical variables of groups were analysed by chi‐square test. The correlation among subtypes was calculated using the Pearson correlation test. We explored the independent prognostic value of the risk scores and other clinical features using univariate and multivariate Cox regression analyses. Statistical analysis was conducted by the ‘R’ software. In our study, a p‐value of less than 0.05 was considered statistically significant.

3. RESULTS

3.1. Identification of prognostic CRLs

We downloaded the pancreatic cancer datasets from TCGA database, which included 177 tumour samples and 4 normal samples (samples without expression matrix or clinical information were excluded). Table 1 exhibited the clinical characteristics of pancreatic cancer patients. According to the GENCODE database, we identified 14,084 lncRNAs from TCGA pancreatic cancer dataset. According to the previous study, there are 10 cuproptosis regulators (FDX1, LIAS, LIPT1, DLD, DLAT, GLS, PDHA1, PDHB, MTF1, CDKN2A) [3]. After obtaining the expression matrix of 10 cuproptosis regulators, Pearson correlation analysis was performed (|Pearson R| > 0.4, p < 0.001) and we obtained 40 CRLs. Finally, to identify prognostic CRLs, we used the univariate Cox regression analysis (p < 0.05). The hazard ratio and expression of 30 prognostic CRLs are shown in Figure 2a–c.

TABLE 1.

Clinical characteristics of PC patients in TCGA dataset.

Clinical characteristics	Count
Age
≤65	93
>65	84
Gender
Female	80
Male	97
Grade
G1	31
G2	94
G3	48
G4	2
Gx	2
T stage
T1	7
T2	24
T3	141
T4	3
Tx	2
M stage
M0	79
M1	4
Mx	94
N stage
N0	50
N1	123
Nx	4
Tumour site
Head of pancreas	138
Others	39

Abbreviations: PC, pancreatic cancer; TCGA, The Cancer Genome Atlas.

FIGURE 2.

Prognostic cuproptosis‐related lncRNAs. (a) Forest map of 30 prognostic cuproptosis‐related lncRNAs identified by univariate Cox regression analysis. (b) Heatmap and (c) violin plot of the expression levels of prognostic CRLs in PC and adjacent normal pancreatic tissues. *p < 0.05 and **p < 0.01. CRLs, cuproptosis‐related lncRNAs; PC, pancreatic cancer.

3.2. Construction and validation of the CRLs prognostic model

To construct the prognostic model of cuproptosis‐related lncRNAs in PC, we performed the LASSO Cox regression. PC patients from the dataset were randomly assigned to the training and test cohort. As a result, 7 of 30 prognostic CRLs (p < 0.01) were filtered to build the prognostic model (Figure 3a,b). Based on formula (1), the following formula (2) was used to compute each patient's risk score:

$$\mathrm{R}\mathrm{i}\mathrm{s}\mathrm{k}\mathrm{s}\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{e}=\mathrm{e}ˆ(-0.327099393974077\ast \mathrm{P}\mathrm{A}\mathrm{N}3-\mathrm{A}\mathrm{S}1\mathrm{e}\mathrm{x}\mathrm{p}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{s}\mathrm{i}\mathrm{o}\mathrm{n}–0.29606012121351\ast \mathrm{L}\mathrm{I}\mathrm{N}\mathrm{C}02593\mathrm{e}\mathrm{x}\mathrm{p}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{s}\mathrm{i}\mathrm{o}\mathrm{n}–1.37664328433923\ast \mathrm{A}\mathrm{L}117335.1\mathrm{e}\mathrm{x}\mathrm{p}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{s}\mathrm{i}\mathrm{o}\mathrm{n}–0.498179055188389\ast \mathrm{L}\mathrm{I}\mathrm{N}\mathrm{C}01091\mathrm{e}\mathrm{x}\mathrm{p}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{s}\mathrm{i}\mathrm{o}\mathrm{n}–0.0197617573760847\ast \mathrm{A}\mathrm{C}092171.3\mathrm{e}\mathrm{x}\mathrm{p}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{s}\mathrm{i}\mathrm{o}\mathrm{n}–0.237742487425625\ast \mathrm{A}\mathrm{C}087501.4\mathrm{e}\mathrm{x}\mathrm{p}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{s}\mathrm{i}\mathrm{o}\mathrm{n}–1.04195463033265\ast \mathrm{S}\mathrm{U}\mathrm{G}\mathrm{T}1\mathrm{P}4-\mathrm{S}\mathrm{T}\mathrm{R}\mathrm{A}6\mathrm{L}\mathrm{P}\mathrm{e}\mathrm{x}\mathrm{p}\mathrm{r}\mathrm{e}\mathrm{s}\mathrm{s}\mathrm{i}\mathrm{o}\mathrm{n}).$$ (2)

FIGURE 3.

Construction of the prognostic model. (a, b) LASSO Cox regression and cross‐validation of 30 prognostic CRLs. Distribution of risk scores of PC patients in (c) the training cohort and (d) the test cohort. Distribution of survival status of PC patients in (e) the training cohort and (f) the test cohort. (g) Heatmap of 7 risk signature genes in the training cohort and the test cohort. Kaplan–Meier survival curve of risk groups in (h) the training cohort and (i) the test cohort. CRLs, cuproptosis‐related lncRNAs; LASSO, least absolute shrinkage and selection operator; PC, pancreatic cancer.

Patients were classified as high or low risk based on their median risk score (Figure 3c–f). PAN3‐AS1, LINC02593, AL117335.1, LINC01091, AC092171.3, AC087501.4, and SUGT1P4‐STRA6LP were downregulated in the high‐risk group (Figure 3g). It's apparent from the survival analysis that PC patients with a low‐risk score enjoyed a better prognosis in the training cohort (p < 0.001, Figure 3h) and the test cohort (p = 0.025, Figure 3i).

3.3. Independence analyses of the prognostic model

We conducted the time‐dependent ROC analysis to assess the predictive power of the risk model. The area under curve (AUC) of the ROC greater than 0.5 was considered to have good predictive capacity. In the training group, AUC values at years 1, 2, and 3 were all greater than 0.7 (0.719, 0.786, and 0.803, respectively, Figure 4a). Compared with other clinical characteristics such as age, gender, grade, stage, T, M, and N stage, our risk model also performed well in predictive capacity (Figure 4b). Then we validated our risk model in the test cohort. It is also verified that our risk model had the precise predictive capability, and the AUC values at years 1, 2, and 3 were 0.658, 0.690, and 0.666 respectively (Figure 4c,d).

FIGURE 4.

Prediction capacity and independence analyses of the prognostic signature. (a) Time‐dependent ROC curve of the risk model for 1, 2, and 3 years in the training cohort. (b) Time‐dependent ROC curve of clinical characteristics in the training cohort. (c) Time‐dependent ROC curve of the risk model for 1, 2, and 3 years in the test cohort. (d) Time‐dependent ROC curve of clinical characteristics in the test cohort. (e) Association of risk score and clinical characteristics. (f) Heatmap of seven prognostic CRLs expression levels and clinical characteristics (*p < 0.05) (g–n) Kaplan–Meier survival curves in different clinical subgroups. CRLs, cuproptosis‐related lncRNAs; ROC, receiver operating characteristic.

Following that, we looked into the relationship between risk scores and clinicopathological variables (Figure 4e). T stage was found to be significantly related to risk score (p = 0.0021), with individuals with T3‐4 stage tumours having higher risk scores. The risk score, on the other hand, was adversely linked with the tumour immune score (p = 0.0044). Other clinicopathological factors such as age (p = 0.14), gender (p = 0.41), tumour site (p = 0.14), and tissue grade (p = 0.18) did not correlate with the risk score.

Figure 4f depicts a heatmap showing the relationship between the expression levels of seven prognostic CRLs and clinicopathological characteristics of individuals. The high‐risk group had significantly higher CIBERSORT immune scores than the low‐risk group (p < 0.05). Furthermore, we confirmed that our prognostic model can be used in different subgroups of pancreatic cancer patients (Figure 4g–n). The prognostic model performed well in the female/male subgroup, T1‐2/T3‐4 stage subgroup, N0/N1‐3 stage subgroup, and different tumour site subgroups. Patients in the low‐risk category still enjoyed a superior outcome in these subgroups.

3.4. Construction of the predictive nomogram

Then we conducted univariate and multivariate Cox regression analyses in all PC patients (Figure 5a,b). Risk score (p < 0.001), tumour site (head of pancreas and others, p = 0.015), tumour tissue grade (p = 0.037) and age (p = 0.012) were negatively correlated with pancreatic cancer prognosis in univariate analysis (Figure 5a). The risk score (p < 0.001) and tumour site (head of pancreas and others, p = 0.024) were likewise validated as independent prognostic factors in multivariate analysis (Figure 5b).

FIGURE 5.

Construction and calibration of survival prediction nomogram in pancreatic cancer (PC). (a) Univariate and (b) multivariate Cox regression analyses in pancreatic cancer patients. (c) A nomogram for predicting the survival rates of 1‐, 2‐, and 3‐year based on risk score and clinical features, including age, tumour tissue grade, and tumour site (head of the pancreas and others). (d) Calibration curve of the nomogram.

To predict the overall survival of PC patients more accurately, we created a predictive nomogram based on four prognostic factors, including risk score, age, tumour tissue grade, and tumour site (head of pancreas and others) (Figure 5c). It's easy to know the probabilities of 1‐, 2‐, and 3‐year OS by calculating total points. Finally, the calibration curve revealed that the predicted and actual probabilities were in good agreement (Figure 5d).

3.5. DEGs and enrichment analyses

To find potential regulatory pathways, we identified DEGs between risk groups. As a consequence, the high‐risk group had 31 upregulated genes and 150 downregulated genes when compared to the low‐risk group (log₂ | FoldChange | > 1.5, p < 0.01, Figure 6a). DEGs were significantly enriched in the endocrine and metabolic‐related signalling pathways, including the secretion and transport of insulin and other peptide hormones, according to Gene Ontology (GO) enrichment analysis (Figure 6b–d). Kyoto Encyclopaedia of Genes and Genomes (KEGG) enrichment analysis also revealed DEGs were enriched in metabolic signalling pathways, including calcium signalling pathway, cytokine–cytokine receptor interaction, cAMP signalling pathway, and so on (Figure 6e–f).

FIGURE 6.

Functional enrichment analysis of DEGs. (a) The DEGs between the two risk groups were shown in a Volcano plot. (b) Bar plot shows the BP, CC, and MF by GO analysis. (c) Network and (d) circle plot visualise the biological process by GO analysis. (e) Bar plot and (f) network shows the signalling pathways enriched by KEGG analysis. BP, biological process; CC, cellular component; DEGs, differentially expressed genes; GO, Gene Ontology; KEGG, Kyoto Encyclopaedia of Genes and Genomes; MF, molecular function.

3.6. Somatic mutation and tumour immune microenvironment analyses between risk groups

To figure out the genomic features in risk groups, we analysed the somatic mutation from TCGA datasets (Figure 7a,b). PC patients in the high‐risk group had a higher frequency of somatic mutations. Mutation in TP53 and KRAS were dominant in the high‐risk group. CDKN2A, a cuproptosis regulator, had a high mutation frequency in the high‐risk group. The risk score was positively related to the TMB (Figure 7c) and the high‐risk group had higher TMB than the low‐risk group (Figure 7d).

FIGURE 7.

Somatic mutation and immune analysis in CRLs risk model. Waterfall plots shows the distribution of the top 20 genes with the highest mutation frequency in (a) the high‐risk group and (b) the low‐risk group. (c) Correlation of the risk score and TMB. (d) Differences in TMB between high‐ and low‐risk groups. (e–j) Correlation of immune cells and risk scores. (k) Immune cell infiltration in risk groups (*p < 0.05 and **p < 0.01) (L) Tumour purity, Stroma score, immune score, and ESTIMATE score in high‐ and low‐risk groups calculated by ESTIMATE algorithm. CRLs, cuproptosis‐related lncRNAs; TMB, tumour mutational burden.

Next, the immune status of the risk groups was determined through CIBERSORT immune analysis. We discovered a negative relationship between risk scores and the infiltration of CD8+ T cells (p = 0.0041, Figure 7e), regulatory T cells (p = 0.016, Figure 7f), naïve B cells (p = 0.00,036, Figure 7g), and plasma cells (p = 0.048, Figure 7h). Meanwhile, the risk score was positively correlated with the abundance of M0 macrophages (p = 0.0066, Figure 7i) and M2 macrophages (p = 0.0073, Figure 7j). In terms of naïve B cells, plasma cells, memory B cells, CD8+ T cells, and M0 and M1 macrophages fraction, there were substantial disparities between the two risk groups (Figure 7k). The high‐risk group had higher tumour purity than the low‐risk group, according to our findings (p < 0.05, Figure 7L). The immune scores and ESTIMATE scores were higher in the low‐risk group than in the high‐risk group (p < 0.01 and p < 0.05 respectively).

4. DISCUSSION

Pancreatic cancer is notorious cancer with high mortality rates. Most PC patients have metastases at diagnosis, thus chemotherapy is the most common treatment option. Nab‐paclitaxel–gemcitabine (AG) and fluorouracil, leucovorin, irinotecan, and oxaliplatin (FOLFIRINOX) are advised for patients with good performance status, but the median overall survival remained frustrating at 8.7 and 11.1 months, respectively [23], [24]. KRAS, CDKN2A, TP53, and SMAD4 are the most frequently mutated genes in pancreatic cancer, however, none of them are currently druggable except KRAS G12C [25], [26]. A list of antiangiogenic drugs, such as the vascular endothelial growth factor (VEGF) inhibitors aflibercept and bevacizumab, have failed in clinical trials due to a lack of blood vessels in the stroma around cancer cells [27], [28]. As for immunotherapy, for pancreatic patients with a positive dMMR/MSI‐H, humanised monoclonal anti‐PD1 antibody pembrolizumab has been suggested as the second‐line therapy [29].

A series of copper complexes were reported to induce apoptosis in PC, such as Cu (II) complex of ketoprofen‐salicylhydrazone (FPA‐306) and tolfenamic acid–Cu (II) complex [30], [31]. In addition, copper complex [Cu^II ₂Cu^I(L)₂(Br)₃] kills pancreatic cancer via nonapoptotic cell death pathways, including ferroptosis [32]. Recently, elesclomol was shown to induce cell death via cuproptosis after inhibiting ferroptosis, necroptosis, and oxidative stress [3]. Taken together, we suspect that some copper complexes or ionophores might induce pancreatic cancer cell death via cuproptosis. Cuproptosis is a promising way to induce tumour cell death, and this research aimed to investigate prognostic cuproptosis‐related lncRNAs in PC and the underlying mechanism.

First, we constructed a model of cuproptosis‐related lncRNAs and verified its good predictive capacity. Our study identified 30 prognostic CRLs with univariate Cox regression analysis, and most of them were downregulated in tumour tissues. Then, seven prognostic CRLs filtered by LASSO Cox regression analysis were used to build a prognostic model. In our prognostic model, high‐risk PC patients had a poor prognosis and lower expressions of these prognostic CRLs, which was validated in our test group. The AUC values of the model were greater than 0.7 at 1, 2, and 3 years. Compared with other clinical risk factors, our model showed better prediction performance. Next, we verified our prognostic model can be used in PC groups with specific clinicopathological characteristics, such as female/male groups, T1‐2/T3‐4 stage groups, N0/N1‐3 stage groups, and different tumour site groups. The risk score was closely related to the prognosis of PC patients by univariate and multivariate Cox regression analyses. In addition, we constructed a nomogram that combined clinical prognostic characteristics to predict 1‐, 2‐, and 3‐year survival.

Next, we performed to further explore possible regulatory pathways. Our study indicated that endocrine and metabolism‐related pathways might be related pathways in cuproptosis regulation. DEGs enriched in insulin and other peptide hormone signalling pathways in the GO analysis. Diabetes is verified as a risk factor for pancreatic cancer in previous studies [33]. Insulin/insulin‐like growth factor 1(IGF‐1) receptors and G protein‐coupled receptors (GPCR) signalling systems regulate the proliferation of pancreatic cancer and chemoresistance [34, 35]. Metformin can inhibit the insulin‐GPCR crosstalk and decrease the risk of pancreatic cancer [34]. DEGs enriched in several metabolic signalling pathways in KEGG enrichment analysis. Calcium signalling which is correlated with gene transcription and cell proliferation regulates early pancreatic carcinogenesis [36, 37, 38].

To figure out the genomic differences between risk groups, we analysed the somatic mutation distribution. PC patients in the high‐risk group had a higher frequency of TP53, KRAS, CDKN2A, and SMAD4, which were verified as the most frequent mutated genes in PC in previous studies. PC patients with higher risk scores tend to have higher TMB. Then we investigated the tumour microenvironment in our risk model by CIBERSORT and ESTIMATE. The risk score was found to have a negative correlation with CD8+ T cells, and a positive correlation with M0 and M2 macrophages. Patients in the low‐risk group had a higher fraction of CD8+ T cells, naïve B cells, and plasma cells and a higher immune score than that in the high‐risk group. Consistent with previous research, the microenvironment of pancreatic cancer is immunosuppressive, enriched with myeloid‐derived suppressor cells (MDSC), tumour‐associated macrophages (TAMs), and Tregs [39]. M1 macrophages exhibit pro‐inflammation, while M2 macrophages suppress immunity and promote tumour growth and angiogenesis [40]. Tregs act as anti‐tumour immunity in PC, suppressing dendritic cells and CD8+ T cell [41]. Indeed, low CD8+ T cells and high macrophages and Tregs in pancreatic cancer were correlated with poor survival [42], [43]. In our study, patients with high‐risk scores were more likely to have cold tumours which deficient in T cells but rich in TAMs. Cold tumours are more insensitive to immune checkpoint inhibitors due to low immunogenicity than hot tumours [44], [45]. In our study, there was no statistical difference in the stroma score between the risk groups, but the tumour purity was higher in the high‐risk group than in the low‐risk group.

According to our results, the mutation in TP53, KRAS, and cuproptosis regulator CDKN2A might affect cuproptosis in PC, which in turn affects metabolism‐related pathways and alters the immune microenvironment. Previous studies have reported that cancer metabolism alters the immune microenvironment to immunosuppressive status to promote tumour progression. Increased glycolysis and lactate production in tumour cells leads to immunosuppression of the tumour microenvironment, manifested by increased M2 macrophage polarisation, increased Tregs, and decreased CD8+ T cells [46, 47, 48]. Meanwhile, TAMs secreted CCL18 promoting the Warburg effect in pancreatic cancer [49]. Pancreatic cancer cells are in a relatively hypoxic environment due to dense stroma and low perfusion. KRAS G12D mutation is critical for the regulation of glucose metabolism in pancreatic cancer [50]. However, cuproptosis relies on mitochondrial metabolism and the tricarboxylic acid cycle. The intracellular copper buildup causes mitochondrial lipoylated proteins to aggregate and Fe‐S cluster proteins to destabilise, resulting in cell death. Glycolysis‐dominated tumour cells are less susceptible to cuproptosis [3]. Inhibition of tumour cell glycolysis may increase sensitivity to copper ionophore therapy.

Above all, we recognised seven prognostic cuproptosis‐related lncRNAs and provided a new prognostic model for PC patients that predicted overall survival. We verified that cuproptosis was related to tumour metabolism and the immune microenvironment. Cuproptosis is a newly discovered way of cell death with many unknown mechanisms waiting to be explored. We can induce cuproptosis using copper ionophores in cuproptosis‐sensitive tumours. Even we can alter the tumour microenvironment through cuproptosis, making it easier for anti‐tumour drugs to enter tumour cells.

Our study has some limitations. First, we only used TCGA datasets to construct and validate our prognostic model. We are carrying out experiments to explore the role of cuproptosis regulators in PC in vivo and vitro. Second, in TCGA dataset, it lacked PC patients with metastasis. We are collecting clinical data in our institution to validate our prognostic signature. In addition, more regulators in cuproptosis need to be explored to further investigate their roles in pancreatic cancer.

5. CONCLUSION

In conclusion, this study identified prognostic CRLs and constructed a prognostic model of pancreatic cancer. Furthermore, we elucidated that the CRLs influenced the prognosis of PC via regulating the metabolism and immune microenvironment. Our finding might provide some clews of pancreatic cancer therapy and prognostic prediction.

Abbreviations

AG

nab‐paclitaxel–gemcitabine

AUC

area under curve

CRL

cuproptosis‐related lncRNA

DEG

differentially expressed gene

FOLFIRINOX

fluorouracil, leucovorin, irinotecan, and oxaliplatin

GO

Gene Ontology

GPCR

G protein‐coupled receptor

IGF‐1

insulin‐like growth factor 1

K‐M

Kaplan–Meier

KEGG

Kyoto Encyclopaedia of Genes and Genomes

LASSO

least absolute shrinkage and selection operator

lncRNA

long non‐coding RNA

MDSC

myeloid‐derived suppressor cell

OS

overall survival

PC

pancreatic cancer

ROC

receiver operating characteristic

TAM

tumour‐associated macrophage

TCGA

The Cancer Genome Atlas

TMB

tumour mutational burden

VEGF

vascular endothelial growth factor

AUTHOR CONTRIBUTIONS

Yanling Wang analysed data and write the original manuscript. Weiyu Ge, Shengbai Xue, Tiebo Mao, Haiyan Xu, and Shumin Li validated the analyses. Jiujie Cui, Xiaofei Zhang, Jingyu Ma, Ming Yue and Daiyuan Shentu reviewed and edited the manuscript. Liwei Wang designed and supervised the study.

CONFLICT OF INTEREST STATEMENT

The authors declare that they have no known competing interest that could influence reported in this paper.

ETHICS APPROVAL AND CONSENT TO PARTICIPATE

Not applicable.

Wang, Y. , et al.: Cuproptosis‐related lncRNAs are correlated with tumour metabolism and immune microenvironment and predict prognosis in pancreatic cancer patients. IET Syst. Biol. 17(4), 174–186 (2023). 10.1049/syb2.12068

DATA AVAILABILITY STATEMENT

TCGA datasets are available through the URL: https://portal.gdc.cancer.gov.