Hypoxia-related lncRNA correlates with prognosis and immune microenvironment in uveal melanoma

Abstract

Background

Hypoxia-related genes are linked to the prognosis of various solid malignant tumors. However, the role of hypoxia-related long non-coding RNAs (HRLs) in uveal melanoma (UVM) remains unclear. This study aimed to identify HRLs associated with UVM prognosis and develop a novel risk signature to predict patient outcomes.

Methods

Data from 80 UVM samples were obtained from The Cancer Genome Atlas. Prognostic HRLs were screened using Cox univariate and Pearson correlation analyses. HRL signature were constructed using Lasso analysis, and gene enrichment analysis was performed to explore the association between HRLs and immune features. Cell Counting Kit-8 assay was used to measure the propagation of human uveal melanoma (MuM2B) cells, while tumor invasion and migration were evaluated using Transwell and wound-healing experiments. Inflammatory factors and macrophage polarization were evaluated using quantitative PCR.

Results

In total, 621 prognostic HRLs were screened and constructed in 12 HRLs. The risk score showed a significant correlation with the survival time of patients with UVM. Additionally, HRL correlated with diverse key immune checkpoints, revealing possible targets for immunotherapy. Immune-related pathways were highly enriched in the high-risk group. LINC02367, a protective HRL, was associated with the tumor microenvironment and survival time of patients with UVM. In vitro, LINC02367 significantly influenced MuM2B proliferation and migration. It also modulated macrophage polarization by regulating inflammatory factor levels, thereby affecting the immune microenvironment.

Conclusions

We developed a novel HRL signature to predict prognosis in patients with UVM. HRLs are potential biomarkers and therapeutic targets for the treatment of UVM.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12935-024-03509-9.

Background

Uveal melanoma (UVM) originates from the melanocytes in the choroid of the eye and is highly malignant [1], with a high fatality rate and short median overall survival (OS) [2, 3]. UVM often metastasizes, which promptly leads to death [4, 5]. Very few treatments are effective upon occurrence of metastasis [6]. Therefore, novel markers for predicting the occurrence and progression of UVM must be identified.

A hypoxic environment is thought to be associated with malignant tumor progression [7, 8]. In UVM, hypoxia has been shown to promote tumor invasion and progression by activating the Notch and MAPK pathways [9, 10]. Furthermore, hypoxia-related genes (HRGs) have been associated with the prognosis of UVM [11], underscoring the importance of hypoxia in its progression.

Long non-coding RNAs (lncRNAs) participate in diverse biological processes primarily through regulating mRNA expression [12–14]. In UVM, lncRNAs are involved in processes such as immune infiltration, metabolism, tumorigenesis, proliferation, and apoptosis [15–18].

A previous study showed that HRGs were correlated with the immune microenvironment and survival time in UVM [11]. However, the role of hypoxia-related lncRNAs (HRLs) in UVM remains unclear. This study utilized data from The Cancer Genome Atlas (TCGA) database to validate HRLs and develop a novel HRL signature, with the aim of investigating their impact on UVM prognosis and immune characteristics.

Methods

Data extraction

We analyzed 80 UVM samples in this study. RNA sequencing (RNA-seq) and clinical information were extracted from TCGA database (https://portal.gdc.cancer.gov/). Single-cell RNA-seq data were extracted from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/). Sample characteristics are provided in Table 1.

Table 1.

Clinical information of 80 samples from TCGA dataset

Features	TCGA (n = 80)
Age
≥ 60	44
< 60	36
Gender
Male	45
Female	35
Stage
Stage II	36
Stage III	40
Stage IV	4
N
N0	76
NX	4
M
M0	73
MX	7
T
T2	14
T3	32
T4	34
Metastasis
Yes	26
No	54

Identification of HRLs

Previous studies identified 26 HRGs [19]. From TCGA database, we extracted a total of 16,882 lncRNAs. Among these, HRLs were defined as lncRNAs significantly correlated with the 26 HRGs (|r|> 0.5 and p < 0.001) were regarded as significant.

Single-cell RNA (scRNA) data processing and dimensionality reduction

Eleven samples from the GSE139829 dataset were analyzed. Cells with nFeature_RNA < 2500, nFeature_RNA > 200, and percentage _mt < 5 were initially incorporated. The “harmony” R package was used to consolidate and batch-correct these samples. We dimensionalized and clustered cells in all samples using Uniform Manifold Approximation and Projection (UMAP), and each cluster was annotated based on the 10 most highly expressed genes.

Bioinformatic analyses

Gene set variation analysis (GSVA) was used to evaluate Gene Ontology (GO) biological processes in each UVM sample [20]. Differentially expressed genes (DEGs) between the two risk groups (high- and low-risk groups) were selected, and gene set enrichment analysis (GSEA) analysis was performed. We utilized the “clusterprofiler” R package to perform Kyoto Encyclopedia of Genes and Genomes (KEGG) and GO enrichment analyses. We established the nomogram and calibration curve using “rms” and “regplot” R packages. The Oncopredict” R package was used to determine the sensitivity of UVM samples to 545 common anti-tumor drugs and their association with the risk signature. Single-nucleotide polymorphism mutation analyses were performed to assess differences in mutated genes that saliently influenced the progression of UVM between risk groups. The waterfall plot was constructed using “waterfalls” and “oncoplot” R packages.

Appraisal of immune features

We utilized the "estimate" R package to calculate stromal scores, tumor purity, and immune scores for each patient [21]. Immune cell infiltration was appraised using the single-sample GSEA (ssGSEA) algorithm across risk groups. Furthermore, the CIBERSORT, TIMER, and EPIC algorithms were adopted to evaluate the relative abundance of immune cells [22–25].

Establishment of risk signature

The LASSO algorithm was used to calculate a risk score for each patient:

$$\text{Risk Score}=\sum _{\text{i}}^{\text{n}}(Coefficient\text{i}\ast Gene\text{i})$$

Construction of competing endogenous RNA network

We used the ENCORI web tool (http://starbase.sysu.edu.cn/) to predict the target miRNAs of LINC02367. Further, we used the mirDIP web tool (http://ophid.utoronto.ca/mirDIP/) to predict potential target mRNAs of these miRNAs. mRNAs significantly correlated with LINC02367 (p < 0.01), as calculated by Pearson correlation analysis, were identified as potential mRNA targets of LINC02367. Cytoscape 3.9.1. was used to establish the lncRNA-miRNA-mRNA network.

Cell transfection of siRNA

Human uveal melanoma (MuM2B) cells were cultured in a complete RPMI-1640 medium supplemented with 10% fetal bovine serum (China). Four specific small interfering RNAs (siRNAs) targeting LINC02367 were developed and synthesized by GenePharma. The siRNAs used in our study were as follows: LINC02367-Homo-1387: F, GCACAGACACAGUAUCAUATT; R, UAUGAUACUGUGUCUGUGCTT. LINC02367-Homo-1472: F, GGAAAUAACUUUGUAGCAATT; R, UUGCUACAAAGUUAUUUCCTT. LINC02367-Homo-2391: F, CUUCUUAACCUGAUGUAUTT; R, AUACAUCAGGUUAAGAAGCTT. LINC02367-Homo-2593: F, GCCCUUGAAAGUCGAUGUATT; R, UACAUCGACUUUCAAGGGCTT.

Lipofectamine 3000 (Invitrogen, Carlsbad, CA, USA) was used for siRNA transfection.

RNA extraction and quantitative PCR implementation

TRIzol reagent (Invitrogen, Carlsbad, CA, USA) was used to extract total RNA. The PrimeScript RT Reagent Kit (RR047A, Takara) was used to remove genomic DNA and perform reverse transcription in two steps. TB Green Fast qPCR Mix (RR430S, Takara) was used for quantitative real-time PCR (qRT-PCR) with glyceraldehyde-3-phosphate dehydrogenase (GAPDH) as an endogenous reference. The primers used were as follows: GAPDH: F, GAACGGGAAGCTCACTGG; R, GCCTGCTTCACCACCTTCT. LINC02367: F, TTGCAAGTTCTGGCTTTGTG; R, GCAGTTGAACTTGTGCCAGA. IL6: F, TACCCCCAGGAGAAGATTCC; R, TTTTCTGCCAGTGCCTCTTT. IFN-:α: F, GCAAGCCCAGAAGTATCTGC; R, ACTGGTTGCCATCAAACTCC. IFN-γ: F, TCCCATGGGTTGTGTGTTTA; R, AAGCACCAGGCATGAAATCT. TGF-β, F: GGGACTATCCACCTGCAAGA; R: CCTCCTTGGCGTAGTAGTCG. TNF-α: F, TCCTTCAGACACCCTCAACC; R, AGGCCCCAGTTTGAATTCTT. CD86: F, ACAGATGTCCTACGGGAACG; R, ATCCCACCTTAGAGCCAGGT.

Cell counting kit-8 (CCK-8) assay

After transfection, MuM2B cells were seeded in 96-well plates at a density of 2 × 10³ cells/well and cultured for 24, 48, 72, and 96 h, respectively. After incubation with 10 µL of CCK-8 reagent (DOJINDO, Japan) for 1 h at 37 ℃, absorbance was measured at 450 nm. Based on the results of qPCR and CCK-8 assays, LINC02367-Homo-1387 and LINC02367-Homo-2593 were used for follow-up experiments.

Cell invasion experiment

We performed Transwell experiments to appraise MuM2B cell invasion capability. MuM2B cells were starved for 24 h, then digested 4–6 h after transfection, suspended in serum-free culture, and inoculated into the upper chamber of a Transwell plate at a concentration of 2 × 10^5/mL for 24 h. Six hundred milliliters of RPMI-1640 culture with 20% serum was infused into the lower chamber to attract MuM2B movement through the Transwell plate. Matrigel (basement membrane matrix) was used to precoat the upper chamber for the cell invasion assay, while it was omitted for the migration assay. The proportion of cells that migrated through the membrane was recorded from five independent random scopes.

Wound-healing experiment

We inoculated MuM2B cells into six-well plates until the cell fusion was > 90%. After transfection, the cells were starved for 24 h in a serum-free medium. The cells were then removed with a single wire design using the tip of a 20 µL pipette to form a cell-free gap. Scratch healing at 0 h and 24 h was recorded, and the percentage of migrating cells was recorded from five independently random scopes.

Macrophage polarization

THP1 cells were planted in six-well plates and differentiated into M0 macrophages by culturing with RPMI-1640 medium containing 100 ng/mL phorbol 12-myristate 13-acetate (PMA) for 24 h. We collected culture supernatants from MuM2B cells knocked down with LINC02367 at 1, 2, and 3 days, respectively. For the supernatant collection, we inoculated the same numbers (10^6 cells) of MuM2B cells in each of the six-well plates with complete medium till cells grew to 70–80% fusion, then they were transfected with si-LINC02367 or si-Control. After cultured for 1, 2, and 3 days, the supernatant was gathered and purified using 0.22 μm filters (Millipore). These supernatants were sequentially added to THP1 cell media, and the culture of M0 THP1 cells was continued. After 3 days, we collected THP1 cells, extracted total RNA, and detected the relative expression of CD86 and CD206.

Data statistics and analysis

R version 4.2.1 and GraphPad Prism version 8.2.1 were utilized to process data and plots. Kaplan–Meier (K–M) analysis and log-rank test were used to compare survival times between risk groups. Pearson analysis was applied to validate the correlation of HRLs with other traits over time. Receiver operating characteristic (ROC) curves were constructed to assess the predictive ability of the risk signature. Statistical significance was set at p < 0.05.

Results

ScRNA analysis revealed distribution of HRGs

After dimensionality reduction, UVM cells from 11 samples were clustered into seven groups: tumor cells, photoreceptors, NKT cells, monocytes and macrophages, T cells, plasma cells, and B cells (Fig. 1A). Univariate Cox analysis identified nine HRGs with prognostic significance, distributed across both tumor and immune cell clusters (Fig. 1B, C). This suggests that hypoxia plays a role in tumor progression by influencing both the tumor itself and its immune microenvironment, particularly T cells and macrophages.

Fig. 1.

Distribution of HRGs in UVM revealed by scRNA analysis. A UMAP of 13 UVM samples. B Univariate Cox analysis of 10 prognostic HRGs in UVM. C Feature plot of 10 prognostic HRGs

Validation of HRLs in TCGA dataset

In total, 16,882 lncRNAs and 26 HRGs from TCGA dataset were identified and incorporated into this study (Fig. 2A). Using the Pearson correlation algorithm (|r|> 0.5 and p < 0.01), 3128 HRLs were screened (Table S1). Univariate Cox and LASSO analyses validated 12 HRLs, which were used to construct risk signatures (Fig. 2B). Among these, five lncRNAs (AC006115.2, AC008555.1, AC104825.1, AP000808.1, and LINC02367) were protective factors in UVM, while the other seven were linked with poorer outcomes (p < 0.05) (Fig. 2C). These 12 selected lncRNAs were closely related to the 26 HRGs (Fig. 2D).

Fig. 2.

Validation of HRLs of this study. A Flow diagram of this project. B Twelve HRLs screened by LASSO analysis. C Identification of 12 HRLs through univariate Cox analysis. D Heatmap of the association of HRGs with HRLs. *p < 0.05, **p < 0.01, ***p < 0.001

HRL model predicted UVM prognosis precisely

A HRL signature was constructed to validate the prognostic value of HRLs in UVM. The coefficient of five lncRNAs (AC006115.2, AC008555.1, AC104825.1, AP000808.1, and LINC02367) was negative, while the others were positive (Fig. 3A). Each patient with UVM received a score according to the expression and coefficients of the 12 HRLs. The hypoxic score calculated using ssGSEA differed significantly between risk groups (Fig. 3C). Further, according to the medium of risk scores, UVM samples were distributed within two groups. OS time of patients in the high-risk group was shorter (Fig. 3B). We applied the ‘caret’ R package to randomly allocate 80 samples into two inner validation groups (n = 40). The area under the curve (AUC) for predicting 1-, 2-, 3-, 4- and 5-year OS of patients with UVM were 0.89, 0.92, 0.96, 0.92, and 0.92, respectively (Fig. 3B). For two validation groups, AUC in predicting 1-, 2- and 3-year survival were 0.87, 0.91, 0.97 and 0.92, 0.91, 0.90, respectively. (Fig. 3C, D). The five protective lncRNAs were more highly expressed in the low-risk group, while the others showed an opposite pattern (Fig. 3F, G). Our HRL signature aligned well with modules validated by Robertson et al. [26], and risk scores in samples with metastasis were significantly higher (p < 0.05) (Fig. 3E). These outcomes revealed a strong correlation between HRL risk signature and UVM prognosis.

Fig. 3.

Development of risk signature according to HRLs. A The coefficient of 12 HRLs constituting risk signature. B–D K–M analysis and time-dependent ROC analysis of risk score of B all samples in 1-, 2-, 3-, 4- and 5-year and C inner group1 in 1-, 2- and 3-year and D inner group2. E Comparison of risk score in SCNA Cluster, mRNA Cluster, DNA Methylation Cluster, lncRNA Cluster, and Metastasis group, and hypoxic score in each sample between two risk groups. F Boxplot of comparison of relative expression of 12 HRLs between risk groups. G Heatmap of expression of 12 HRLs in each patient from low-risk to high-risk. *p < 0.05, **p < 0.01, ***p < 0.001

Risk score independently predicted UVM prognosis

A nomogram model and calibration plot were constructed to predict 3-year OS (Fig. 4A, B). UVM samples were divided into subgroups based on factors such as stage, sex, and age. For each subgroup, high-risk patients had shorter OS, while low-risk patients had better prognoses (p < 0.05) (Fig. 4C–F). These findings suggested that the risk score is a potential prognostic marker for patients with UVM.

Fig. 4.

Risk score independently predicts prognosis of UVM samples. A Nomogram model of mRNA_Cluster, lncRNA_Cluster, Paradigm_Cluster, Metastasis and stage and risk score. B Calibration curve model regarding 1-, 2- and 3-year survival aimed at confirming the prognostic value of HRLs. C K–M analysis of patients in two risk groups in age ≥ 60 and < 60. D K–M analysis of patients in two risk groups in females and males. E K–M analysis of patients in two risk groups in stage II and stage III. F K–M analysis of patients in two risk groups in T3 and T4

Risk score associated with the immune checkpoint

Despite UVM’s insensitivity to immunotherapy, we explored the relationship between the HRL signature and immune checkpoint inhibitors (ICIs) [27]. In the high-risk group, several vital immune checkpoints (CTLA-4, IDO1, TIGHT, LAYN, and LAG3) showed relatively higher levels (p < 0.05) (Fig. 5A). Receptor ligands and HLA genes also showed similar trends, suggesting that the effects of immunotherapy may differ between groups (Fig. 5B, D). LAG3 was identified to be most closely correlated with risk score using Pearson analysis, consistent with the findings by Durante et al. [28], suggesting that the most sensitive immune checkpoint for UVM may be LAG3 (Fig. 5C). Patients with lower risk scores and higher LAG3 levels had better prognoses compared to those with higher risk scores and lower LAG3 levels (p < 0.05) (Fig. 5E).

Fig. 5.

Risk score was associated with the effectiveness of ICIs in UVM patients. A The relative level of multiple immune checkpoints between risk groups. B The relative level of several ligand receptors between risk groups. C The relationship of risk score with immune checkpoints. D The relative expression of HLA between risk groups. E K–M algorithm of overall survival in 4 groups constructed by risk score and relative expression of LAG3

Risk groups showed disparate immune peculiarities

To explore the differences between risk groups, DEGs were identified (Fig. S1). GO enrichment analysis highlighted immune pathways such as macrophages, T cells, and antigen-presenting-related pathways (Fig. 6B). DEGs were also associated with neurodegenerative diseases, oxidative phosphorylation, and reactive oxygen species-associated pathways, all of which were associated with cellular and tissue hypoxia (Fig. 6A). Genes with relatively higher levels in the high-risk group were overtly enriched in immune signaling pathways, such as antigen response, as well as B- and T-cell-related pathways, as shown by GSEA. The Hallmark hypoxia pathway also showed significant differences between the two groups (p < 0.05) (Fig. 6C, D). GSVA further confirmed that both hypoxic and immune signaling pathways varied between risk groups (Fig. 6E, F).

Fig. 6.

HRL signature revealed disparate immune traits. A KEGG enrichment analysis. B GO enrichment analysis. C, D GSEA of whole RNA-seq data in C immune signaling pathways between risk groups in GO term and D Hypoxic signaling pathway in HALLMARK term. E, F Immune and hypoxic signaling pathways between two risk groups shown by GSVA in E GO term and F KEGG term

Risk score revealed differences in immune infiltration and tumor purity

Given that the prognosis of UVM is strongly influenced by peripheral immune infiltration [29, 30], we investigated the relationship between the HRL signature and the immune microenvironment in UVM. CD8+ T cells and M1 macrophages were highly enriched in the high-risk group, as calculated by CIBERSORT algorithm (p < 0.05) (Fig. 7A). Additionally, the ssGSEA, EPIC, and TIMER algorithms containing 28, eight, and six immune cells, respectively, were applied. The high-risk group showed significantly higher levels of macrophages and CD8+ T cells (Fig. 7B, C, and E). Tumor purity, which was evaluated according to the ESTIMATE score, was negatively correlated with the risk score, indicating that hypoxia played a role mainly in the tumor microenvironment (p < 0.05) (Fig. 7D). Additionally, the relationship among the risk score, macrophages, and CD8+ T cells was confirmed by Pearson’s analysis (p < 0.05) (Fig. 7F). This is consistent with previous findings that UVM samples representing higher infiltration of CD8+ T cells have a poorer prognosis and that abundant CD8+ T cells are seen in distant metastatic lesions [28, 31]. These findings indicate a strong relationship between HRL and immune infiltration, which partly explains the risk function of HRL.

Fig. 7.

Risk signature reveals differences in tumor purity and immune infiltration A Twenty-two immune cells assessed by CIBERSORT algorithm between risk groups. B Eight immune cells assessed by EIPC algorithm between two risk groups. C Six immune cells assessed by TIMER algorithm between two risk groups. D Association between risk score and stromal score, immune score, estimate score, and tumor purity. E Abundance of 28 immune cells estimated by ssGSEA algorithm between two risk groups. F Association between risk score and nine immune cells assessed by TIMER and CIBERSORT algorithm

Prognostic HRL LINC02367 was correlated with clinical and immune characteristics

We chose LINC02367 for further research because it has never been reported before and its 95% CI was the narrowest among the 12 HRLs, indicating its relatively stable effect. The relative levels of CD8+T cells and M1 macrophages in the low LINC02367 group were significantly higher than those in the high LINC02367 group, while the infiltration of M2 macrophages showed the opposite trend (Fig. 8A). The Pearson algorithm also confirmed that LINC02367 was positively correlated with M2 macrophages (r = 0.29, p < 0.05), negatively correlated with M1 macrophages (r = −0.34, p < 0.05), and negatively correlated with CD8+T cells (r = −0.40, p < 0.05) (Fig. 8B). We also compared the abundance of LINC02367 with modules validated by Robertson AG et al. The pattern of LINC02367 in these modules was completely opposite to the risk characteristics (Fig. 8C). In addition, the median overall survival of the high LINC02367 group was significantly higher than that of the low LINC02367group (Fig. 8D). In addition, the level of LINC02367 in the transferred patients was significantly reduced (Fig. 8E). Therefore, LINC02367, as a prognostic factor for UVM, is significantly correlated with both clinical and immune characteristics.

Fig. 8.

LINC02367 was correlated with clinical and immune characteristics in UVM. A Level of eight immune cells assessed by three algorithms in between high- and low-LINC02367 groups. B Association between LINC02367 and CD4+ T, CD8+ T, and macrophages type 1 and 2 cells. C Comparison of LINC02367 in SCNA Cluster, mRNA Cluster, DNA Methylation Cluster, lncRNA Cluster, miRNA Cluster, miRNA Cluster, and Paradigm Cluster. D K–M analyses of OS between high and low-LINC02367 groups. E Comparison of LINC02367 in metastasis and non-metastasis group

HRLs combined with drug sensitivity analysis revealed potential therapeutic agents for UVM

From the Cancer Therapeutics Response Portal, multiple drugs having statistically significant associations with both the risk score and LINC02367 were screened (|r|> 0.5, p < 0.001) (Fig. 9A and C). A lower predicted value represented higher tumor sensitivity to the drug. BRD.K07442505, a pan-cancer targeting agent, demonstrated greater sensitivity in high-risk samples (r = − 0.59, p < 0.001), and its predicted value was positively associated with LINC02367 (Fig. 9B). K–M analysis showed that the group less sensitive to BRD.K07442505 with higher risk scores had the worst prognosis, whereas the group more sensitive to BRD.K07442505 with low risk scores had the best prognosis (Fig. 9D). These results suggest the latent therapeutic function of BRD.K07442505 in patients with UVM, especially those at high risk.

Fig. 9.

LINC02367 was associated with the drug sensitivity in UVM. A Abundance of 27 anti-tumor drugs saliently associated with risk score between LINC02367 groups. B Association between BRD.K07442505 and risk score and LINC02367. C Abundance of 27 anti-tumor drugs significantly associated with risk scores in two risk groups. D K–M algorithm of overall survival among four sample groups constructed by HRL signature and relative predict value of BRD.K07442505

HRLs related to the mutation rate of key genes of UVM

Genes with the top 20 mutation frequencies in both risk groups were screened. Genes with the highest mutation rates in the groups revealed a high overlap, including GANQ, GAN11, BAP1, SF3B1, and EIF1AX, which are critical in UVM progression [32] (Fig. 10A, B).

Fig. 10.

HRLs signature was associated with the vital single nucleotide polymorphisms (SNPs) in UVM. A, B The frequency and type of mutation of the top 20 genes that mutate most frequently in A high-risk and B low-risk groups, respectively. C Comparison of tumor mutation burden (TMB) between two groups. D–F Risk score of each patient in D BAP1 mutation group, E EIF1AX mutation group, and F SF3B1 mutation group

Perhaps due to UVM's lack of sensitivity to immune checkpoints, no difference in tumor mutation burden (TMB) was observed between the two risk groups (Fig. 10C). Moreover, the BAP1 mutation group showed a higher risk score than the non-mutation group, whereas SF3B1 and EIF1AX showed the opposite trend (p < 0.05) (Fig. 10D–F). This is consistent with findings that patients with poorer prognoses have higher levels of BAP1 mutations, and those with better prognoses have higher levels of SF3B1 and EIF1AX mutations [26].

Bioinformatic analysis of potential molecular mechanisms of LINC02367

LncRNAs may function as competing endogenous RNAs in tumor progression. Thus, we identified the potential miRNA and mRNA targets of LINC02367, and co-expressed mRNAs were screened using Pearson analysis to determine the target mRNA. A lncRNA–miRNA–mRNA network was constructed (Fig. 11A). LINC02367-targeting mRNAs were highly enriched in neuronal connections and synaptic pathways (Fig. 11B). DEGs were sorted according to the median LINC02367 expression. Hypoxia, glycolysis, as well as DNA repair signaling pathways, were highly enriched in the high-LINC02367 group, whereas several inflammatory pathways, including inflammatory response, IFNα, IFNγ, and IL6 signaling pathways, exhibited high enrichment in the low-LINC02367 group (Fig. 11C, D). These results indicate that LINC02367 may affect the progression of UVM cells by regulating the tumor immune microenvironment.

Fig. 11.

Molecular mechanisms underlying LINC02367. A Competing endogenous RNA network of LINC02367. B GO enrichment analysis of targeting mRNAs of LINC02367. C, D GSEA of total RNA-seq data between high and low-LINC02367 groups in C HALLMARK and D GO term

LINC02367’s effect on proliferation and migration of UVM cells through immune microenvironment shift

In vitro trials were conducted to determine the pathogenic function of LINC02367 in UVM cells. Four siRNAs were transfected into MuM2B cells to knockdown LINC02367, with si-LINC02367#1387 and si-LINC02367#2593 selected because of their higher efficacy (p < 0.05) (Fig. 12A). The knockout of LINC02367 significantly prompted the multiplication of UVM cells, as shown by CCK-8 assay (p < 0.05) (Fig. 12B). As shown in the wound healing assay, LINC02367 inhibition promoted the migration of MuM2B cells (Fig. 12C). Moreover, Transwell experiments with or without Matrigel also confirmed that the knockout of LINC02367 saliently promoted the proportion of MuM2B cells migrating or invading through the upper chamber (Fig. 12E). These results indicate that LINC02367 is related to the propagation and migration of UVM, making it a potential therapeutic target.

Fig. 12.

LINC02367 functioned negatively on the propagation of MuM2B. A Relative level of LINC02367 after transfection of four siRNAs. B CCK8 assay of MuM2B cells after knockdown of LINC02367. C Wound-healing assay of MuM2B cells after knockdown of LINC02367. D Transwell assay of MuM2B cells after knockdown of LINC02367. E The expression of five inflammatory factors of MuM2B cells after knockdown of LINC02367. F The expression of Macrophages type 1 marker CD206 of THP1 cells after co-cultured with cell supernatant of MuM2B cells after knockdown of LINC02367

To validate the possible mechanism of action of LINC02367, we conducted supplementary experiments targeting inflammatory factors and macrophages. IL-6 plays a central role in inflammation and immune activation [33], but we observed no statistical change in IL-6 in cells between the si-LINC02367 and si-Control groups. However, compared to the si-Control group, three inflammatory factors, including TNF-α, TGF-β1, and IFN-α in both si-LINC02367 groups, were increased by varying degrees. IFN-γ only increased in the si-LINC02367-2593 group (Fig. 12G). Among these factors, TNF-α, IFN-α, and IFN-γ are thought to function importantly in the polarization of macrophages and the activation and migration of pro-inflammatory cells [34, 35]. Pearson’s correlation analysis revealed a positive correlation between LINC02367 expression and the expression of the four inflammatory factors (Fig. S2). These outcomes suggest that a pro-inflammatory microenvironment functions predominantly in UVM where LINC02367 is inhibited. Further, we focused on macrophage polarization, given that the IFN-related pathway may impact M0-to M1-type polarization. THP1 cells incubated with supernatants from the si-LINC02367-treated group showed significantly higher expression levels of the M1-type macrophage marker CD86 than those from the NC group, suggesting that LINC02367 functions in macrophage polarization (Fig. 12H).

Studies have reported that high PMA treatment fails to upregulate M2-type macrophage markers [36]; however, we failed to detect CT values for CD206 markers in THP1 cells by qPCR, even though the concentration of PMA was downregulated to 10 ng and the cDNA was not diluted. This may be due to the overall pro-inflammatory microenvironment of MuM2B and the inability of macrophages to differentiate into the M1 subtype. Therefore, LINC02367 may influence the immune features of UVM by inhibiting the activation of pro-inflammatory cells and altering macrophage polarization, thereby reducing distant metastasis and affecting prognosis.

Discussion

Hypoxic environments and lncRNAs have both been demonstrated as crucial factors for the progression of UVM [37, 38]. In this study, we identified 620 HRLs affecting the prognosis of UVM and constructed a novel risk signature using the LASSO algorithm, which included 12 HRLs in TCGA database. We found a significant correlation between the risk score and the prognosis of patients with UVM. Additionally, the relative levels of several vital immune checkpoints, especially LAG3, differed between the risk groups, highlighting HRLs as latent indicators of immunotherapy effectiveness. Functional analysis showed substantial enrichment of immune signaling pathways in the high-risk group, while the hypoxia pathway also showed statistical differences, confirming the correlation between HRL and the immune features of UVM.

In vitro experiments showed that LINC02367 knockout significantly facilitated the propagation and migration of UVM cells. The inflammatory response and activation of IFN signaling pathways likely played a crucial role in this process. In this study, we developed a novel signature based on HRL as a predictor of UVM prognosis, suggesting that LINC02367 may be an effective target for clinical interventions.

Hypoxic environments have been shown to have a marked impact on many malignant tumors, such as hepatocellular carcinoma, bladder carcinoma, lung carcinoma, and glioma [39–42]. In bladder cancer, an 8-gene hypoxia signature accurately predicts patient prognosis and is significantly correlated with immune infiltration [43]. In lung cancer, another 6-gene hypoxia signature showed that a hypoxic environment exhibits independent prognostic value [44]. In addition, HRGs are saliently correlated with an undesirable prognosis and high recurrence of liver cancer. Simultaneously, hypoxia-related models can correctly distinguish hepatocellular carcinoma, normal samples, and nodular tissues, and hypoxia-related features can positively modulate immune responses [45]. In this study, 26 hypoxia-related genes (HRGs) were selected to correlate and screen for HRLs. The selected HRLs were strongly correlated with the HRGs, and each had a significant prognostic value. Among the HRGs, TPI1 and FOSL1 correlated with the vast majority of HRLs. The univariate Cox analysis demonstrated the strong relevance of TPI1 and FOSL1 to worse prognosis in patients with UVM, and both TPI1 and FOSL1 were significantly negatively correlated with LINC02367. Subsequently, we developed a risk signature based on the 12 HRLs. Bioinformatics analysis showed that risk scores accurately predict prognostic information in patients with uveal melanoma. Survival analyses of different subgroups also demonstrated the strong relevance of higher risk scores to worse prognosis and shorter OS. Although incorporated with other clinical characteristics, the risk score still represents a standalone value for the prognosis of patients with UVM. Hence, this study revealed a novel HRL signature that may be a potential prognostic marker for UVM.

Recently, the pivotal impact of the immune microenvironment on tumor progression has been confirmed. In contrast to other solid tumors, the pro-inflammatory microenvironment around the UVM is generally related to an undesirable prognosis and distant metastases. A meta-analysis demonstrated that a higher abundance of B cells represented a better prognosis. Others, such as M1 macrophages, NKT, and CD8+ T cells, predicted a worse prognosis for UVM [46]. Moreover, it has been speculated that the activation of the T cell response is necessary for distant metastasis of UVM, owing to the special immune characteristics of the eyes [46]. Macrophages function in the secretion of inflammatory factors and immune microenvironment remodeling, and are associated with a worse prognosis [29]. Elevated expression of inflammatory factors such as IFN-γ has been associated with metastasis of UVM [47]. Our study discovered that elevated risk score was significantly related to IFN-γ and immune response pathways. DEGs were enriched in hypoxia- and immune-related pathways, and higher infiltration of M1 macrophages and CD8+ T cells was observed in the high-risk group, possibly due to hypoxic conditions. However, the relevance of hypoxia to immune characteristics remains unclear And further research is required to determine whether immune microenvironment remodeling is a direct result of hypoxia.

The contradiction between immune infiltration in UVM and other cancers is a critical issue that needs to be resolved. In most cancers, a higher proportion of cytotoxic immune cells, such as M1 macrophages and CD8+ T cells, suppress the development of tumors, an effect not observed in UVM. A pan-cancer analysis of 32 tumor types revealed that gliomas and UVMs showed similar immune infiltration characteristics, with high immune infiltration representing a worse prognosis [48]. Xu et al. reported the strong relevance of activated CD8+ T cells with a high risk in lower-grade glioma (LGG), along with poorer immunoefficacy and elevated tumor immune dysfunction and exclusion scores, suggesting crosstalk between ICIs and immune infiltration [49]. Similar findings were observed in this study, where ssGSEA revealed that CD8+ T cells were significantly enriched in the high-risk group, indicating that the prospects for immunotherapy in UVM remain uncertain.

LncRNAs have been reported as biomarkers or potential therapeutic targets for various diseases [50, 51]. LncRNA-associated signatures have also demonstrated valuable prognostic value in diverse tumors. A 19-HRL signature precisely predicted the prognosis of LGG and identified the vital relevance of a hypoxic environment for immune infiltration [49]. In addition, Chen et al. developed a novel molecular model of lncRNAs related to fatty acid metabolism that showed significant prognostic potential in hepatocellular carcinoma and was closely related to the efficacy of immunotherapy [52]. Several lncRNA signatures have been reported in UVM. Jin et al. constructed a molecular model of ferroptosis-related lncRNAs that exhibited a strong relationship with the prognosis of UVM and the tumor immune microenvironment [53]. A 6-lncRNA signature associated with autophagy accurately predicted the prognosis of UVM, and high enrichment of cytoplasmic component cycling, energy metabolism, and apoptosis signaling pathways was observed in the high-risk group [54]. However, to the best of our knowledge, no study has revealed the latent impact of HRLs on the clinical and immune features of UVM.

In this study, we established a 12-HRL signature that accurately predicted the prognosis of UVM samples. LINC02367, a protective lncRNA, was selected to represent the HRL signature. Subsequent analyses showed that LINC02367 expression was remarkably correlated with prognosis and the immune environment in UVM. We also found that LINC02367 knockout promoted the proliferation, migration, and invasion of UVM cells with the activation of inflammatory factors and a shift in macrophage polarization. The cytoskeleton is related to the progression and migration of UVM and the susceptibility of UVM tumor cells to treatment [55, 56]. Another study showed that antibodies targeting tumor immune synapses may benefit OS in patients with metastatic UVM [57]. After construction of the ceRNA network, LINC02367 was found to function in the cell skeleton and synapses, and inflammatory pathways, such as the complement, interferon response, and IL-6 signaling pathways, were markedly enriched in the high-LINC02367 group. Given that inflammatory responses and cytokines, such as IFN, have been demonstrated to play a profound role in UVM progression [58, 59], decreased LINC02367 expression may activate the inflammatory response by activating IFN or other cytokines and worsen the prognosis of UVM. Thus, LINC02367 is a potential target for UVM treatment.

Conclusion

This study identified 12 HRLs (AC006115.2, AC008555.1, AC023790.2, AC104129.1, AC104825.1, AC135166.1, AP000808.1, LINC01637, LINC01971, LINC02367, PDE2A.AS2 and SOX1.OT) and established a novel HRL signature based on them that predicted UVM prognosis and was significantly correlated with the immune microenvironment. Thus, HRL may be a novel biomarker for predicting UVM progression and a potential target for UVM treatment. LINC02367 may regulate UVM proliferation and invasion by influencing tumor immune microenvironment, especially macrophage polarization.

Abbreviations

UVM

Uveal melanoma

HRG

Hypoxia-related gene

HRL

Hypoxia-related long non-coding RNA

scRNA-seq

Single-cell RNA sequencing

GEO

Gene Expression Omnibus

TCGA

The Cancer Genome Atlas

CIBERSORT

Estimating relative subsets of RNA transcripts

LASSO

Least absolute shrinkage and selection operator

TIME

Tumour immune microenvironment

RNA-seq

RNA sequencing

qRT-PCR

Quantitative real-time PCR

Author contributions

Y.C. completed the bioinformatics analysis and in vitro experiments and wrote the manuscript. S.C. participated in transwell and wound-healing experiments. Z.K.W reviewed and revised the article. Q.C. guided the writing of the manuscript. D.J. funded and supervised the project and edited the manuscript. All authors have read and agreed to the submission of the manuscript.

Funding

This work was funded by grants from the National Natural Science Foundation of China (Grant No. U23A20435, 82271091) and National key research and development program of China (2021YFA1101200 & 2021YFA1101202).

Availability of data and materials

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.