Skip to main content

Liquid biopsy-based identification of prognostic and immunotherapeutically relevant gene signatures in lower grade glioma

Abstract

Background

Recent studies have shown that immunotherapies, including peptide vaccines, remain promising strategies for patients with lower grade glioma (LGG); however new biomarkers need to be developed to identify patients who may benefit from therapy. We aimed to investigate the feasibility of liquid biopsy-based gene signatures in predicting the prognosis of LGG patients, as well as the benefits of immunotherapy.

Methods

We evaluated the association between circulating immune cells and treatment response by analyzing peripheral blood mononuclear cell (PBMC) samples from LGG patients receiving peptide vaccine immunotherapy, identified response-related genes (RRGs), and constructed RRG-related Response Score. In addition, RRG-related RiskScore was constructed in LGG tumor samples based on RRGs; association analysis for RiskScore and characteristics of TME as well as patient prognosis were performed in two LGG tumor datasets. The predictive power of RiskScore for immunotherapy benefits was analyzed in an anti-PD-1 treatment cohort.

Results

This study demonstrated the importance of circulating immune cells, including monocytes, in the immunotherapeutic response and prognosis of patients with LGG. Overall, 43 significant RRGs were identified, and three clusters with different characteristics were identified in PBMC samples based on RRGs. The constructed RRG-related Response Score could identify patients who produced a complete response to peptide vaccine immunotherapy and could predict prognosis. Additionally, three subtypes were identified in LGG tumors based on RRGs, with subtype 2 being an immune “hot” phenotype suitable for immune checkpoint therapy. The constructed RRG-related RiskScore was significantly positively correlated with the level of tumor immune cell infiltration. Patients with high RiskScore had a worse prognosis and were more likely to respond to immune checkpoint therapy. The therapeutic advantage and clinical benefits of patients with a high RiskScore were confirmed in an anti-PD-1 treatment cohort.

Conclusion

This study confirmed the potential of liquid biopsy for individualized treatment selection in LGG patients and determined the feasibility of circulating immune cells as biomarkers for LGG. Scoring systems based on RRGs can predict the benefits of immunotherapy and prognosis in patients with LGG. This work would help to increase our understanding of the clinical significance of liquid biopsy and more effectively guide individualized immunotherapy strategies.

Background

Gliomas are the most common central nervous system tumors, accounting for approximately 75% of primary malignant brain tumors [1]. Of these, World Health Organization (WHO) grade II and III gliomas are considered lower grade gliomas (LGGs), which have a slower course with a better prognosis than grade IV glioblastoma (GBM) [2, 3]. The prognosis of patients with LGG depends on a variety of factors, including IDH mutation status and molecular subtypes, and survival time can range from 1 to 15 years [4]. Despite the current multimodal standard of treatment for maximal safe surgical resection followed by chemotherapy and radiotherapy, tumor recurrence and progression are almost inevitable due to its high malignancy and treatment resistance [5]. Therefore, the development of new therapeutic strategies is crucial for the treatment of LGG. Accurate and convenient individualized assessment methods are indispensable for the prognostic stratification of LGG patients and for personalized treatment selection.

Recently, immunotherapy has attracted widespread attention among oncologists due to its great success in a variety of tumors, including melanoma [6,7,8]. Current immunotherapy approaches include immune checkpoint inhibitors (ICIs), tumor vaccines, and cytokine therapy [9]. Although in previous studies, only a small proportion of glioma patients responded objectively to ICI therapy and struggled to achieve a survival benefit [10, 11], in a recent clinical trial, Zhao et al. showed that a proportion of molecularly and individually screened GBM patients may benefit from the therapy [12]. For LGG, nearly half of the patients in a recent glioma-associated antigens (GAAs) peptide vaccine trial experienced an objective response and some survival benefit [13]. These results suggest that immunotherapy, particularly ICI and tumor vaccine therapies, remains a promising therapy for patients with glioma. However, this requires screening and filtering of patients to identify those who may respond to immunotherapy to avoid unnecessary, or even harmful, treatment.

In the search for effective biomarkers to predict which patients will respond to immunotherapy, several studies have shown that molecular analysis based on pre-treatment tumor tissue can effectively predict patient response to treatment [14,15,16,17]. These approaches have mainly addressed the expression levels of immune checkpoint molecules, tumor mutation burden (TMB), and T cell function in tumor tissues. However, biomarkers based on tumor tissue samples may have certain limitations. For example, in patients with advanced LGG, it may be difficult to obtain sufficient tissue volume for molecular testing, and in addition, there is a risk of biopsy in sensitive areas or performing serial biopsies of tumors [18, 19]. Another limitation is that assays relying on individual biopsies may be affected by intra-tumor heterogeneity or low tissue volume, resulting in unreliable results [20]. Therefore, there is a large need for the development of new individual patient prognostic and treatment sensitivity prediction methods.

Liquid biopsy represents an exciting new frontier in cancer diagnosis and management, as it is convenient and minimally invasive, giving it significant advantages over tissue biopsy. The material for liquid biopsies can be derived from a variety of body fluids such as blood, saliva, and cerebrospinal fluid [21], and covers a wide range of circulating biological indicators, such as circulating tumor cells, circulating immune cells, and circulating nucleic acids [22]. Further, liquid biopsy is applied for clinical use in a variety of solid tumors, including breast and colorectal cancers [23]. Currently, little progress has been made in circulating biomarkers in glioma, with one study of peptide vaccine therapy in LGG patients in which peripheral blood mononuclear cells (PBMCs)-based liquid biopsy responded to the peripheral immune status of LGG patients and identified several possible biomarkers [24]. However, further in-depth studies on the application of liquid biopsies in LGG patients are still needed, and the potential role of liquid biopsy-based identified biomarkers in tumor tissues remains to be elucidated.

The aim of this study was to explore gene signatures used to predict immunotherapeutic response in liquid biopsy-based PBMC samples from LGG patients and to explore the potential application of gene signatures in LGG tumor samples. We explored the predictive role of circulating immune cells in response to peptide vaccine therapy in LGG patients and identified 43 response-related genes (RRGs). Based on these RRGs, we identified three tumor subtypes with different immune features in LGG tumor samples and constructed an RRGs-related RiskScore that can be used for prognosis and immunotherapeutic response prediction.

Methods

Data extraction and preprocessing

Illumina RNA sequencing (RNA-seq) raw data of 54 PBMC samples from 12 LGG patients were retrieved from ArrayExpress repository (accession number: E-MTAB-6270, https://www.ebi.ac.uk/arrayexpress/). RNA-seq data were converted to transcripts per kilobase million (TPM) values, and log2(TPM + 1) transformation was performed. All patients underwent peptide vaccine immunotherapy, and PBMC samples from patients' blood were collected for RNA-seq at various time points (0, 6, 15, 34, 70, and 86 week) before and during immunotherapy. All clinical data, including treatment response, were obtained from previous studies [13, 24]. Normalized RNA-seq data for 510 LGG tumor samples from The Cancer Genome Atlas (TCGA) database and 431 LGG tumor samples from the Chinese Glioma Genome Atlas (CGGA) database, as well as clinical information, were obtained from GlioVis (http://gliovis.bioinfo.cnio.es/) [25]. Normalized RNA-seq data of 14 pre-treatment tumor samples from the GBM anti-PD-1 treatment cohort and clinical information were obtained from a previous study [12], and all patients were treated with PD-1 inhibitors (nivolumab or pembrolizumab). Only samples with full prognostic information were retained in this study, and all relevant data that could not be extracted from previous publications were obtained from the corresponding authors of the publications. Basic information on all samples in this study is summarized in Additional file 1: Table S1.

Inference of circulating immune cells in PBMC samples and infiltrating cells in the tumor microenvironment (TME) of tumor samples

We used CIBERSORTx to quantify the relative abundance of circulating immune cells in PBMC samples [26], which was used by Nabet et al. to analyze the circulating immune cell profile of non-small cell lung cancer [27]. For LGG tumor samples, we used single-sample gene set enrichment analysis (ssGSEA) to estimate the relative abundance of 28 previously reported immune cells in LGG tumor microenvironment (TME) [28].

Identification of immunotherapeutic RRGs in PBMC samples

The empirical Bayesian approach in the “limma” R package was used to calculate the differentially expressed genes (DEGs) between three treatment responses (SD, stable disease; PR, partial response; CR, complete response) at 0, 6, 15, and 34 week separately [29]. Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis was carried out using the R package “clusterProfiler” after taking the union of DEGs between different treatment responses at each time point [30]. Subsequently, response-associated DEGs at different time points were intersected to obtain DEGs whose expression was altered at all time points. We considered these DEGs as immunotherapeutic RRGs in LGG PBMC samples. Statistical significance was set at p < 0.05.

Unsupervised clustering for 43 RRGs

Based on the gene expression profiles of the identified 43 RRGs in 54 PBMC samples, robust LGG PBMC clusters were identified using unsupervised clustering based on the partition around medoids (PAM) algorithm. As we described previously, a total of 1000 bootstraps were executed, and each bootstrap resampled 80% of samples [31, 32]. The maximum number of clusters was 10, and K values were evaluated using a consensus cumulative distribution function and consensus heatmap [33]. Using the same approach, we identified RRG-associated tumor subtypes in LGG tumor samples.

Gene set enrichment analysis (GSEA) of different RRG-related clusters

Using the GSEA software (Version 4.1.0) obtained from the GSEA website (http://software.broadinstitute.org/gsea/index.jsp) and downloading the “c2.cp.kegg.v7.4.symbols” gene set and the HALLMARK gene set from the MSigDB database, we evaluated the different pathways and molecular mechanisms in the different LGG PBMC-related clusters [34]. The minimum gene set was set to 5, the maximum gene set was set to 5000, and resamples were 1000. GSEA was performed using the same approach for high and low RRG-related RiskScore groups in LGG tumor samples. In addition, different immune-related biological processes between different clusters or subtypes were explored using previously published gene sets, including immune activation-relevant gene set [35], immune checkpoint-relevant gene set [36], and epithelial–mesenchymal transition (EMT)-relevant gene set [37]. P value of < 0.05 was considered statistically significant.

Construction of the RRG-related Response Score

To quantify the response characteristics of individual PBMC samples to immunotherapy, the prognostic value of each RRG in pretreatment samples was analyzed based on the 43 RRGs identified using the univariate Cox regression model. The RRGs with prognostic effect were extracted for further analysis, and the expression of each RRG was transformed into a z-score. Principal component analysis (PCA) was then performed, and principal component 1 was selected as the signature score. This approach has the advantage of focusing the scores on the set of gene blocks with the best correlation in the set, while reducing the contribution of genes that do not track with other set members. Finally, the Response Score was defined using the following formula:

$$\text{Response}\;\text{Score}= \sum {PC1}_{i}-\sum {PC1}_{j}$$

where i is the expression of RRG whose Cox coefficient is positive, and j is the expression of RRG whose Cox coefficient is negative.

Construction of the RRG-related RiskScore

To define RRG-related RiskScore for individual LGG tumor samples, we first identified RRGs with a significant prognostic effect in the TCGA cohort using the univariate Cox regression model. Prognosis-related RRGs were subsequently entered into a stepwise regression to determine the final risk model. The RRG-related RiskScore was calculated using the following formula:

$${RRG{\text{-}}related}\;{RiskScore}=\sum {\beta }_{i}\times {Exp}_{i}$$

where βi is the coefficient of each RRG in the final risk model, and Expi is the gene expression value.

Statistical analysis

Differences between two groups were compared using unpaired Student’s t-test or Wilcoxon rank sum test. For comparisons between more than two groups, differences were compared using one-way ANOVA and Kruskal–Wallis test. Correlation coefficients between RRGs and between RiskScore and TME infiltrating immune cells were calculated using spearman correlation analysis. The R packages “survival” and “survminer” were used to calculate the correlation between the variables and survival time and to find the best cutoff value. Survival curves were generated using the Kaplan–Meier (KM) method, and significance was determined using the log-rank test. The R package “pROC” was used to plot receiver operating characteristic (ROC) curves to verify the validity of the model and obtain the area under the curve (AUC). Independent prognostic factors were identified using a multivariate Cox regression model and visualized using the R package “forestplot.” The “oncoplot” function of the R package “maftools” was used to present a waterfall plot of patients with mutations in the TCGA-LGG cohort. All statistical calculations were performed using R software (Version 4.0.3) or GraphPad Prism 5 (La Jolla, CA, USA); p < 0.05 was considered statistically significant, and p values were two-sided.

Results

Association of circulating immune cells with immunotherapeutic response and prognosis in LGG patients

Peripheral immunity is closely related to immunotherapeutic response in tumor patients [24, 27]; therefore, we explored the possibility of circulating immune cells as immunotherapeutic biomarkers in LGG patients. Specifically, we applied CIBERSORTx to quantify the relative abundance of circulating immune cells in 54 PBMC samples, and a total of 15 circulating immune cells were detected. We confirmed that the composition of circulating immune cell components was highly correlated, such that the abundance of activated dendritic cells was highly positively correlated with the abundance of M0 macrophages and follicular helper T cells, whereas the abundance of monocytes was negatively correlated with the abundance of resting memory CD4 T cells (Fig. 1A). We subsequently found that CR patients had significantly higher peripheral monocytes compared to SD and PR patients, whereas SD patients had higher peripheral resting NK cells (Fig. 1B). Notably, CR patients also had higher CD8 T cell compared to PR patients (Fig. 1B).

Fig. 1
figure 1

Association of circulating immune cells with immunotherapeutic response and prognosis in LGG patients. a The interaction of circulating immune cells in LGG. Correlations are calculated by spearman correlation analysis. Positive correlation is indicated in red and negative correlation in blue. b The abundance of each circulating immune cells in SD, PR and CR patients. The upper and lower ends of the boxes represented interquartile range of values. The lines in the boxes represented median value, and dots showed outliers. c Average (± SD) abundance of B cell memory, B cell naive, Monocytes, NK cell activated and T cell CD8 (y axis) from week 0 to 34 (x axis). Response groups are indicated by color. d Kaplan–Meier curves for PFS showing the association between relative abundance of B cell memory, B cell naive, Monocytes and PFS (log-rank test, p < 0.05). *P < 0.05, **P < 0.01, ***P < 0.001

We observed the dynamics of circulating immune cells at different time points, as shown in Fig. 1C, where CR and PR patients had higher monocytes at week 0 (pretreatment), and CR patients maintained high levels of monocytes at different time points after treatment, whereas PR patients experienced a transient decline at week 6 and SD patients consistently had lower levels of monocytes. CR patients also had low levels of naive B cells at week 0 and only a transient rise occurred at week 6 post-treatment, subsequently maintaining lower levels of naive B cells than PR and SD patients (Fig. 1C). In addition, CR patients also had higher levels of CD8 T cells than PR patients at all time points except week 34, whereas PR patients had higher levels of memory B cells at week 0 and week 6 (Fig. 1C). The level of peripheral myeloid-derived suppressor cells (MDSCs) is known to correlate with the malignancy of gliomas [38], and we were interested in understanding the prognostic value of peripheral circulating immune cells in LGG patients treated with immunotherapy. We estimated the prognostic role of immune cells in pre-treatment samples and found that high levels of memory B cells and monocytes were associated with better progression-free survival (PFS), whereas high levels of naive B cells were associated with poorer PFS (Fig. 1D). It was further found that high levels of memory B cells, follicular helper T cells were associated with better overall survival (OS), while the opposite was true for naive B cells and activated NK cells (Additional file 1: Fig. S1). Overall, these results suggest a strong association between peripheral circulating immune cells and both immunotherapeutic response and prognosis in patients with LGG.

Identification of response-related genes

To explore differences in gene expression in patients with different treatment responses, we identified DEGs in patients at different treatment time points separately using the “limma” R package. As shown in Fig. 2A, a total of 1981 DEGs at week 0 were identified by comparing the differences between SD, PR, and CR patients (Additional file 2: Table S2). KEGG pathway analysis showed that these DEGs were not only associated with key oncogenic-related pathways such as cell cycle and apoptosis, but are also closely related to T and B cell receptor signaling pathways, Th1 and Th2 cell differentiation, and immune checkpoint pathway (Fig. 2B). This further suggested that the response to peptide vaccine immunotherapy was associated with peripheral immunity in patients with LGG prior to treatment. Next, we identified 4543, 3211, and 2761 response-related DEGs at week 6, 15, and 34 following initial treatment, respectively (Additional file 2: Tables S3–S5). These DEGs were mainly enriched in metabolic pathways, NOD-like receptor signaling pathways, and RNA transport (Additional file 1: Fig. S2A–C). Interestingly, the number of DEGs after initial treatment was much higher than that before treatment, which seems to imply that greater differences in gene expression emerged between patients who responded differently after treatment. Furthermore, based on Venn diagrams of DEGs at different time points, an overlap analysis was performed. As shown in Fig. 2C, we identified a total of 43 treatment response-related DEGs that were differentially expressed at all time points (Additional file 2: Table S6). These genes may be closely associated with immunotherapeutic responses in LGG patients, and we therefore identified them as RRGs. PCA analysis showed that based on these 43 RRGs, PBMC samples from the three different treatment responses were divided into three distinct clusters (Fig. 2D).

Fig. 2
figure 2

Identification of response-related genes. a 1981 DEGs at week 0 were identified by comparing the differences between SD, PR, and CR patients. b KEGG pathway analysis of 1981 DEGs at week 0. The size of the circle represents the number of genes enriched. c Venn diagram showing the number of crossover genes with altered expression from week 0 to 34. A total of 43 treatment response-related DEGs that were differentially expressed at all time points. d Principal component analysis based on the transcriptome profiles of 43 RRGs, showing a remarkable difference on transcriptome among SD, PR and CR samples

Response-related clusters in PBMC samples

To explore the relationship between RRGs, we calculated pairwise correlations between the expression of 43 RRGs, and a wide range of positive and negative correlations were found between RRGs (Fig. 3A). Next, we performed unsupervised consensus clustering on PBMC samples based on the expression profiles of 43 RRGs to identify clusters with different RRG expression patterns. Based on the cumulative distribution function and the functional delta area (Additional file 1: Fig. S3), we chose k = 3, and the PBMC samples could be stably classified into three clusters: Cluster 1 (C1), Cluster 2 (C2), and Cluster 3 (C3) (Fig. 3B). The clinical characteristics of the different clusters, as well as the expression patterns of RRGs, are presented in heatmap form in Fig. 3C. Notably, all CR samples were classified as C3, whereas all PR samples were classified as C2 (Fig. 3D), suggesting the potential predictive value of RRG-based clusters for response to peptide vaccine immunotherapy. More importantly, the alluvial diagram showed that PBMC samples from the same patient at different time points both pre- and post-treatment were consistently classified in the same cluster (Fig. 3E), indicating that the stability of RRG-related clusters for classification of samples with different responses was not affected by sampling time. Interestingly, we found that C3 samples had the highest RASAL2 and MAPKAPK2 expression, whereas C2 samples had the lowest HLA-DRB1 expression, and C1 samples had low RASAL2 and MAPKAPK2 expression as well as the highest HLA-DRB1 expression (Fig. 3F). The detection of the expression levels of these three genes appears to allow the initial identification of three distinct clusters, suggesting the feasibility of potential clinical applications. Not surprisingly, C3 was associated with a relatively good OS in pre-treatment samples (Fig. 3G).

Fig. 3
figure 3

Response-related clusters in PBMC samples. a The interaction of 43 RRGs in PBMC samples of LGG. Correlations are calculated by spearman correlation analysis. Positive correlation is indicated in red and negative correlation in blue. b Consensus heatmap (k = 3) based on 43 RGGs expression profile, displaying the clustering stability using 1000 iterations of unsupervised clustering. c The heatmap of 43 RRGs expression in the PBMC samples. The response-related cluster, patient ID, treatment response, samples drawn at weeks, vital status and gender were used as patient annotations. Red represented high expression of RRGs and blue represented low expression. d The proportion of response-related clusters in SD, PR and CR samples. Cluster 1, red; Cluster 2, green; Cluster 3, bule. e Alluvial diagram showing response-related clusters of samples from SD, PR and CR patients at different time points. f Differential expression of RASAL2, MAPKAPK2 and HLA-DRB1 in different clusters. The differences between every two groups were compared through the Kruskal–Wallis test. g Kaplan–Meier curve for OS showing the difference in survival between cluster 3 and cluster 1/2 (log-rank test, p = 0.066). P < 0.1, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

Subsequently, we identified different immune-related features among the different clusters. As shown in Fig. 4A, C3 had the highest level of circulating CD8 T cells, with C2 being the next highest and C1 the lowest, while no other cell types were associated with RRG-related clusters. Based on previous studies, we considered that CLDN3, CLDN7, CLDN4, CDH1, VIM, TWIST1, ZEB1, and ZEB2 were EMT-relevant transcripts [37]; CD274, CTLA4, HAVCR2, LAG3, PDCD1, PDCD1LG2, and TIGIT are immune checkpoint-relevant transcripts [36]; CD8A, CXCL10, CXCL9, GZMA, GZMB, IFNG, PRF1, and TBX21 are immune activation-relevant transcripts [35]. C2 had high expression of ZEB1, CD274, LAG3, and CD8A, indicating that C2 may have high peripheral immunosuppression with some immune activation, whereas C3 had high expression of CD8A, representing better peripheral immune activation, and C1 had the lowest immune activation (Fig. 4B–D). Furthermore, through GSEA, we found that C1 was mainly associated with oncogenic pathways such as the KRAS signaling pathway and angiogenesis; C2 was mainly enriched in the ERBB signaling pathway and mismatch repair; and C3 was enriched in antigen processing and presentation and intestinal immune network (Fig. 4E).

Fig. 4
figure 4

Immune-related features of response-related clusters. a The relative abundance of each circulating immune cells in three response-related clusters. The upper and lower ends of the boxes represented interquartile range of values. The lines in the boxes represented median value, and dots showed outliers. b EMT-relevant genes (CLDN7, CLDN4, CDH1, VIM, ZEB1, and ZEB2) expressed in three response-related clusters. c Immune-checkpoint-relevant genes (CD274, CTLA4, HAVCR2, LAG3, PDCD1, PDCD1LG2, and TIGIT) expressed in three response-related clusters. d Immune activation-relevant genes (CD8A, CXCL10, CXCL9, GZMA, GZMB, IFNG, and TBX21) expressed in three response-related clusters. Within each cluster, the upper and lower ends of the boxes represented interquartile range of values. The lines in the boxes represented median value, and black dots showed outliers. e Enrichment plots showing the relatively significantly enriched pathways in clusters 1, 2 and 3. Different pathways are indicated by different colors in each enrichment plot

Generation of RRG-related Response Score

Using the PCA algorithm as well as the RRG signature, we constructed a scoring system to quantify the pattern of treatment response in individual samples, which we termed Response Score. The high Response Score group had higher circulating CD8 T cells, representing better peripheral immune activation (Fig. 5A). At the same time, the high Response Score group also had lower CD274 expression, implying lower immunosuppression (Fig. 5B). It was also found that the C3 samples had the highest Response Scores (Fig. 5C). Consistent with this, the CR samples also had the highest Response Scores (Fig. 5D). By calculating the Response Scores at different time points, it was found that the CR samples consistently had higher Response Scores than the SD and PR samples (Fig. 5E). Furthermore, we explored the predictive power of Response Scores on the prognosis of patients with LGG treated with peptide vaccine immunotherapy. Unsurprisingly, the high Response Score group in the pretreatment samples was found to have significantly better OS and PFS (Fig. 5F).

Fig. 5
figure 5

RRG-related Response Score in LGG. a The relative abundance of each circulating immune cells in high- and low-Response Score groups. The upper and lower ends of the boxes represented interquartile range of values. The lines in the boxes represented median value, and dots showed outliers. b Immune-checkpoint-relevant genes (CD274, CTLA4, HAVCR2, LAG3, PDCD1, PDCD1LG2, and TIGIT) expressed in high- and low-Response Score groups. Within each cluster, the upper and lower ends of the boxes represented interquartile range of values. The lines in the boxes represented median value, and black dots showed outliers. c Differences in Response Score between Cluster 1, Cluster 2, and Cluster 3. The differences between every two groups were compared through the Kruskal–Wallis test. d Differences in Response Score between SD, PR, and CR samples. The differences between every two groups were compared through the Kruskal–Wallis test. e Average (± SD) value of RRG-related Response Score (y axis) from week 0 to 34 (x axis). Response groups are indicated by color. f, g Kaplan–Meier curves for OS (f) and PFS (g) showing the difference in survival between high-Response Score group and low-Response Score group (log-rank test, p < 0.05). High-Response Score group is indicated in red line and low-Response Score group in blue. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

Identification of RRG-related subtypes in LGG tumor samples

Since 43 RRGs were identified in liquid biopsy-based PBMC samples, we were interested in determining whether RRGs play a key role in tumor samples. Based on the expression profiles of RRGs in the TCGA-LGG cohort, we performed unsupervised consensus clustering to identify potential tumor subtypes. By analyzing the cumulative distribution function, function delta area, and consensus heatmap, we identified three robust RRG-associated tumor subtypes (Additional file 1: Fig. S4): subtype 1 (S1), subtype 2 (S2), and subtype 3 (S3). To understand the immune characteristics of different tumor subtypes, we performed ssGSEA to infer the relative abundance of 28 tumor immune cells in different subtypes. As shown in Fig. 6A, B, S2 tumors had the highest infiltration of immune-activated cells, such as activated CD4 T cells, activated CD8 T cells, and natural killer cells, as well as the highest infiltration of immunosuppressive cells, such as MDSCs and regulatory T cells. This suggested that S2 tumors are immune hot phenotypes with both immune-activating and immunosuppressive features. In contrast, most immune-activated cells including activated CD4 and CD8 T cells in S3 tumors had minimal levels of infiltration, as well as low levels of MDSC and regulatory T cells, indicating that S3 belongs to the immune cold phenotype (Fig. 6A, B), whereas S1 tumors had a complex TME. Furthermore, we characterized the expression of immune-related transcripts in the three tumor subtypes. Unsurprisingly, most of the immune activation-related mRNAs, immune checkpoint-related mRNAs, and EMT-related mRNAs were upregulated in S2 tumors, which further confirmed that S2 tumors possessed immune activation and were highly immunosuppressive (Fig. 6C–E). Notably, EMT-relevant genes ZEB1 and ZEB2 were found to be significantly highly expressed in S1 tumors, whereas CLDN3 and CLDN4 expression was elevated in S3 tumors, suggesting some stromal activation in S1 and S3 tumors (Fig. 6C–E). These results indicate that RRGs have an important TME immunomodulatory role in LGG tumors.

Fig. 6
figure 6

Immune-related features of RRG-related subtypes. a Heatmap of the relationship between RRG-related subtypes and 28 immune cell subpopulations identified by a previous study. IDH1 status, Gender, vital status, histologic subtype, grade and RRG-related subtypes are shown as patient annotations. b boxplots of the relationship between RRG-related subtypes and 28 immune cell subpopulations. The upper and lower ends of the boxes represented interquartile range of values. The lines in the boxes represented median value, and black dots showed outliers. c Immune activation-relevant genes (CD8A, CXCL10, CXCL9, GZMA, GZMB, IFNG, PRF1, and TBX21) expressed in three RRG-related subtypes. d Immune-checkpoint-relevant genes (CD274, CTLA4, HAVCR2, LAG3, PDCD1, PDCD1LG2, and TIGIT) expressed in three RRG-related subtypes. e EMT-relevant genes (CLDN3, CLDN7, CLDN4, CDH1, VIM, TWIST1, ZEB1, and ZEB2) expressed in three RRG-related subtypes. Within each subtype, the upper and lower ends of the boxes represented interquartile range of values. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

Construction of RRG-related RiskScore

To further assess the clinical relevance of RRGs in LGG patients, we first calculated the prognostic value of RRGs in the TCGA-LGG cohort using a univariate Cox regression model, and a total of 16 RRGs were identified that were significantly associated with LGG prognosis (Additional file 1: Fig. S5). Based on the survival-related RRGs using stepwise regression, we constructed an RRGs signature; this model was termed as the RRG-related RiskScore (Additional file 2: Table S8). Using the median as the threshold, we divided the samples in TCGA-LGG into RiskScore-high and -low groups, and the KM curve suggested that patients with high RiskScore have significantly worse OS (Fig. 7A). In addition, the ROC curve for RRG-related RiskScore showed that the AUCs of 1-, 3- and 5-years OS were 0.89, 0.83, and 0.71, respectively (Fig. 7C). To examine whether RiskScore is an independent prognostic factor in LGG patients, we performed a multifactorial Cox regression analysis with age, histology, grade, gender, and IDH status as covariates in LGG patients. The RRG-related RiskScore was found to be an independent and robust prognostic marker for LGG and can be used for the prognostic assessment of patients with LGG (Fig. 7D). Importantly, we confirmed the reliability of the RRG-related RiskScore using 431 samples from the CGGA-LGG cohort as an independent validation cohort. Consistent with the TCGA-LGG cohort, patients with high RiskScore had significantly shorter OS times (Fig. 7B), and RRG-related RiskScore was an independent prognostic factor (Fig. 7E).

Fig. 7
figure 7

Construction of RRG-related RiskScore. a, b Kaplan–Meier curves for OS showing the difference in survival between high-RiskScore group and low-RiskScore group in the TCGA (a) and CGGA (b) cohorts (log-rank test, p < 0.0001). High-RiskScore group is indicated in red line and low-RiskScore group in blue. c The predictive value of RiskScore in patients among the TCGA cohort (AUC:0.89, 083, and 0.71; 1, 3, and 5-years overall survival). d, e Multivariate Cox regression model analysis, which included the factors of RiskScore, age, gender, IDH1 status, grade, histology and patient outcomes in the TCGA (d) and CGGA (e) cohorts. The length of the horizontal line represents the 95% confidence interval (CI) for each group. The vertical dotted line represents the hazard ratio (HR) of all patients shown by the forest plot. *P < 0.05, **P < 0.01, ***P < 0.001

Clinical and mutational characteristics of RRG-related RiskScore

LGG mainly includes three different histological subtypes: astrocytoma, oligoastrocytoma, and oligodendroglioma. We compared the RRG-related RiskScore of different LGG histological subtypes and found that astrocytomas had a significantly higher RiskScore than oligoastrocytomas and oligodendrogliomas (Fig. 8A). WHO grade III LGG has a worse prognosis than grade II LGG, and it was found to have a higher RiskScore in grade III tumors than grade II tumors (Fig. 8B). In addition, we found that S2 tumors had higher RiskScores than S1 and S3 tumors (Fig. 8C). It is well known that TMB is closely related to the effects of immunotherapy [39, 40]. Therefore, TMB in RiskScore-high and RiskScore-low groups were analyzed based on mutation data from the varscan-processed mutation dataset of the TCGA cohort. As shown in Fig. 8D, the RiskScore-high group had a significantly higher TMB. In addition, the RiskScore-high group also had significantly higher neoantigen counts (Fig. 8E). This suggests that LGG patients with a high RiskScore are more likely to respond to immunotherapy. Further, Fig. 8F demonstrates the 30 most frequently mutated genes in the high and low RiskScore groups, with IDH1 having the highest mutation frequency. As shown in Fig. 8G, patients with IDH1 mutations have a lower RiskScore compared to IDH1 wild-type patients, indicating a better prognosis.

Fig. 8
figure 8

Clinical and mutational characteristics of RRG-related RiskScore. a Differences in RRG-related RiskScore between astrocytomas, oligoastrocytomas and oligodendrogliomas. The differences between every two groups were compared through the Kruskal–Wallis test. b Differences in RRG-related RiskScore between grade II and grade III tumor. The difference was compared through the Kruskal–Wallis test. c Differences in RRG-related RiskScore between subtype 1, subtype 2 and subtype 3. The differences between every two groups were compared through the Kruskal–Wallis test. d, e Differences in TMB (d) and neoantigen counts (e) between high- and low-RiskScore groups. The difference was compared through the Kruskal–Wallis test. f The top 30 frequently mutated genes in high- and low-RiskScore groups. g Differences in RiskScore between IDH1-mutant and IDH1-wild-type groups. The difference was compared through the Kruskal–Wallis test. **P < 0.01, ***P < 0.001, ****P < 0.0001

Cellular and molecular characterization of RRG-related RiskScore

To explore the role of RiskScore in the regulation of TME in LGG patients, we first compared the relative abundance of 28 immune cells in patients with high and low RiskScore by ssGSEA in the TCGA-LGG cohort. As shown in Fig. 9A, patients with high RiskScore had higher infiltration levels of immune cells, including multiple immunostimulatory cells (e.g., activated CD8T cells) and immunosuppressive cells (e.g., MDSCs), which indicated that high RiskScore patients had a “hotter” TME. More importantly, we found significant positive correlations between RiskScore and infiltration levels of the vast majority of immune cells in LGG patients (Fig. 9B), suggesting that the TME pattern of LGG patients can be quantified by RiskScore, with patients with higher RiskScore having a “hotter” TME and more likely to benefit from ICI treatment. In addition, the results were validated in the CGGA-LGG cohort (Additional file 1: Fig. S6A, B). Furthermore, analysis of the immune-related transcripts in LGG patients showed that patients with high RiskScore had not only higher expression of immune activation-relevant genes, but also higher expression of immune checkpoint-relevant genes and EMT-relevant genes (Fig. 9C–E). This suggests that patients with a high RiskScore have higher immune activation and strong immunosuppression. GSEA based on the HALLMARK gene set showed that the high-RiskScore group was enriched not only for immune activation-related pathways (inflammatory response and IFNγ response), but also for stromal activation-related pathway (EMT) and oncogenic-related pathways (angiogenesis and KRAS signaling pathway) (Fig. 9F). GSEA based on the KEGG gene set likewise showed that the high-RiskScore group was associated with immune activation and stromal activation pathways, including the T-cell receptor signaling pathway and cell adhesion, among others (Fig. 9F).

Fig. 9
figure 9

Cellular and molecular characterization of RRG-related RiskScore. a Heatmap of the relationship between RRG-related RiskScore and 28 immune cell subpopulations identified by a previous study in the TCGA cohort. b Correlation between RiskScore and abundance of 28 immune cells. Correlations are calculated by spearman correlation analysis. Blue represents weak positive correlation, red represents strong positive correlation. c Immune activation-relevant genes (CD8A, CXCL10, CXCL9, GZMA, GZMB, IFNG, PRF1, and TBX21) expressed in high- and low-RiskScore groups. d Immune-checkpoint-relevant genes (CD274, CTLA4, HAVCR2, LAG3, PDCD1, PDCD1LG2, and TIGIT) expressed in high- and low-RiskScore groups. e EMT-relevant genes (CLDN3, CLDN7, CLDN4, CDH1, VIM, TWIST1, ZEB1, and ZEB2) expressed in high- and low-RiskScore groups. f Enrichment plots based on HALLMARK and KEGG gent sets showing the relatively significantly enriched pathways in high RiskScore group. Different pathways are indicated by different colors in each enrichment plot. P ≥ 0.1, P < 0.1, *P < 0.05, **P < 0.01, ***P < 0.001

The RRG-related RiskScore predicts immunotherapeutic benefits

Based on previous results, LGG patients with higher RiskScore are more likely to benefit from ICI immunotherapy, and we hope to find tangible evidence in glioma for this inference to drive clinical application. We extracted the only currently known multi-sample GBM anti-PD-1 treatment cohort with complete RNA-seq and clinical data and calculated the RRG-related RiskScore for pre-treatment samples. Patients with high RiskScore who received anti-PD-1 therapy achieved significant survival benefits, including improved OS and PFS (Fig. 10A, B). Patients with a high RiskScore showed better clinical response to anti-PD-1 therapy compared to RiskScore-low patients (Fig. 10C). Similar to previous results, further studies showed that patients with high RiskScore had higher levels of immune activation cell infiltration, including activated CD4 T cells, CD56 bright natural killer cells, central memory CD8 T cells, and type 1 T helper cells (Fig. 10D, E). In addition, patients with high RiskScore also had higher levels of regulatory T cells and type 2 helper T cells infiltration (Fig. 10D, E), indicating higher immunosuppression. In conclusion, our results strongly suggest that RRG-related RiskScore is significantly associated with response to anti-PD-1 immunotherapy in glioma patients and helps predict response to anti-PD-1 therapy.

Fig. 10
figure 10

The RRG-related RiskScore predicts immunotherapeutic benefits. a, b Kaplan–Meier curves for OS (a) and PFS (b) showing the difference in survival between high-RiskScore group and low-RiskScore group (log-rank test, p < 0.0001). High-RiskScore group is indicated in red line and low-RiskScore group in blue. c Rate of clinical response (Response and Non-response) to anti-PD-1 immunotherapy in high or low RiskScore groups. d Heatmap of 28 previously reported immune cells infiltration in high-RiskScore group and low-RiskScore group. e Boxplots of the relationship between RRG-related RiskScore and 28 immune cell subpopulations. The upper and lower ends of the boxes represented interquartile range of values. The lines in the boxes represented median value, and black dots showed outliers. *P < 0.05, **P < 0.01, ***P < 0.001

Discussion

Immunotherapeutic approaches, including peptide vaccines, offer a safe and effective treatment option for patients with LGG, whose slower disease progression relative to patients with GBM allows for multiple immunizations to gain anti-tumor immunity and survival benefits. Many studies have supported the potential of peptide vaccine immunotherapy in patients with LGG [41,42,43,44]. Thus, new biomarkers are needed to identify patients who may respond therapeutically. Liquid biopsy offers an easy way to perform minimally invasive biopsies for the diagnosis and treatment of cancer patients and is a rapidly expanding area of translational cancer research [45]. Here, we used PBMC samples from patients with LGG to reveal the potential of circulating immune cells as biomarkers in immunotherapy. Recent studies suggest that circulating monocytes play a key role in immunotherapy and may be a prerequisite for a successful response to anti-PD-1 immunotherapy [46, 47]. In addition, a study by Romano et al. suggested that monocytes may maintain an effective antitumor immune response during anti-CTLA-4 immunotherapy [48]. Consistent with these studies, our results suggest that monocytes are important for the antitumor response of peptide vaccines, and that CR patients maintained high levels of circulating monocytes throughout the treatment period. Interestingly, the study by Krieg et al. found that lung cancer patients with lower levels of circulating CD8 T cells prior to treatment were more likely to experience long-lasting benefits from ICI therapy [27], and similar results were found in patients with melanoma [47]. In our results, higher baseline CD8 T-cell levels were found in patients with CR. Mechanistically, patients with high levels of tumor-infiltrating lymphocytes are more likely to benefit from ICI, indicating that peripheral circulating CD8 T cells in this group of patients are raised to the tumor [47, 49]. In contrast, tumor vaccines are better suited to activate the immune system in patients with lower levels of tumor-infiltrating lymphocytes [50, 51]. Therefore, patients with higher baseline circulating CD8 T cell levels may be more suitable for tumor vaccines, including peptide vaccines; however, a combination of consideration of the patient’s level of immunosuppression as well as monocyte levels must be taken into account. Overall, our results suggest that circulating tumor cells, especially monocytes and CD8T cells, can be potential biomarkers for immunotherapy in patients with LGG.

Our study identified 43 significant RRGs, based on which we identified three distinct clusters, and PBMC samples from the same patient were always classified as the same cluster, suggesting that this classification can be performed at any time point of treatment without influence. In addition, all CR samples were classified as C3 and all PR samples were classified as C2, implying that patients classified as C3 in clinical applications would be most likely to have the best response to peptide vaccine immunotherapy. For easier classification, RASAL2, MAPKAPK2, and HLA-DRB1 were identified as signature genes for classification, based on which three clusters can be initially distinguished, but this needs to be validated by further clinical trials. We subsequently found a gradual decrease in the overall level of circulating CD8 T cells from C3 to C1, which is consistent with our expectations, and the better treatment response may have stimulated the proliferation of more CD8 T cells. Similarly, the proliferation of peripheral blood CD8 T cells occurred in responders with lung cancer receiving anti-PD-1 therapy [52]. GSEA further confirmed that C3 was indeed enriched in immune-related pathways, such as antigen processing and presentation, indicating immune activation in C3. To better assess treatment response in individual samples, we developed a scoring system known as the RRG-related Response Score. The high Response Score group had higher levels of CD8 T cells, indicating higher immune activation. At the same time, CR patients had significantly higher Response Score than SD and PR patients at any time point of treatment, confirming the feasibility of identifying CR patients by Response Score. Importantly, Response Score can also assess the prognosis of patients treated with peptide vaccines, and since it is established in liquid biopsy-based PBMC samples, it highlights the promise of its application in clinical work to serve a predictive function in a minimally invasive manner.

It is unclear how important the gene signatures identified based on PBMC samples are in tumor samples. To reveal the role of 43 RRGs in LGG tumor samples and to explore whether the gene signatures identified based on liquid biopsies are intrinsically linked to tumor samples, we performed unsupervised clustering of tumor samples using RRGs and identified three robust tumor subtypes. S2 tumors were characterized as “hot” tumors, characterized by immune activation and stromal activation, making this subtype more suitable for ICI therapy, which exerts anti-tumor immunity by blocking immunosuppression. In contrast, relatively cold phenotypes of S1 and S3 tumors can be attempted to convert them from “cold” to “hot” by new strategies such as tumor vaccines [53], which can then be further combined with ICI therapy. This suggests an important role for RRGs in regulating TME in tumors. Furthermore, we identified an important prognostic role of RRGs in LGG and developed a scoring model called the RRG-related RiskScore. RiskScore predicts the prognosis of LGG patients with good efficiency; it is an independent prognostic factor for LGG; and its predictive power was validated in an independent cohort. The prognostic stratification function of RiskScore is important for long-term management and personalized treatment selection in patients with LGG.

Our study also revealed the relationship between RRG-related RiskScore and TMB, with patients with high RiskScore having significantly higher TMB and neoantigen counts than those with low RiskScore, suggesting that patients with high RS are more likely to benefit from immunotherapy [39, 40]. Importantly, our study also confirmed that RiskScore is significantly positively correlated with tumor immune cell infiltration and that patients with high RiskScore have a “hotter” TME, which means better immune activation and stronger immune suppression. Thus, this not only suggests that RRG-related RiskScore can quantify TME infiltration patterns in individual patients, but also further demonstrates the important modulatory role of RRGs in TME. Based on the properties of RiskScore, patients with higher RiskScore are theoretically more likely to benefit from ICI; however, there is a lack of available LGG ICI treatment cohorts, so we used a GBM anti-PD-1 treatment cohort for validation. Consistent with our expectations, GBM patients with high RiskScore were more likely to respond to anti-PD-1 therapy, and GBM patients with high RiskScore had longer OS and PFS after anti-PD-1 therapy. This makes the predictive power of RiskScore for immunotherapy benefits directly validated by clinical trials in glioma, further confirming the important clinical relevance of RiskScore.

Although our study has some limitations, such as its retrospective nature, further validation of the identified gene signatures and scoring systems in a prospective cohort is needed. Second, the sample size of PBMC is limited, and the results need to be confirmed in a larger cohort of samples. Nevertheless, our results confirmed that RNA-seq of liquid biopsy-based PBMC samples is a powerful and promising approach for individualized treatment selection and that peripheral circulating immune cells may be promising biomarkers. In addition, we identified 43 important RRGs, which not only play an important role in peripheral immunity in LGG patients, but also have a critical regulatory role in the TME of LGG tumors, indicating an intrinsic association between gene signatures identified based on liquid biopsies and those identified by tumor biopsy. In clinical practice, RRG-related Response Score can identify patients who are more likely to develop CR to peptide vaccine immunotherapy in a minimally invasive manner and predict the prognosis of treated patients. The RRG-related RiskScore not only identifies patients more likely to respond to ICI therapy by tumor samples but is also a powerful prognostic prediction tool. Our study thus provides new ideas for individualized immunotherapy in patients with LGG.

Conclusion

This study confirmed the potential of liquid biopsy-based RNA-seq for individualized treatment selection in LGG patients, determined the feasibility of circulating immune cells as biomarkers for LGG, and identified 43 RRGs. Scoring systems based on RRGs can predict the benefits of immunotherapy and prognosis in patients with LGG. This work highlights the critical clinical implications of liquid biopsy and contributes to the development of personalized immunotherapeutic strategies for LGG patients.

Availability of data and materials

All data generated and described in this article are available from the corresponding databases and publications, and are freely available to any scientist wishing to use them for noncommercial purposes, without breaching participant confidentiality. Further information is available from the corresponding author on reasonable request.

Abbreviations

LGG:

Lower grade glioma

GBM:

Glioblastoma

WHO:

World Health Organization

GAA:

Glioma-associated antigen

ICI:

Immune checkpoint inhibitor

TMB:

Tumor mutation burden

PBMC:

Peripheral blood mononuclear cell

RRG:

Response-related gene

RNA-seq:

RNA sequencing

TPM:

Transcripts per kilobase million

TCGA:

The Cancer Genome Atlas

CGGA:

Chinese Glioma Genome Atlas

ssGSEA:

Single-sample gene set enrichment analysis

DEG:

Differentially expressed gene

SD:

Stable disease

PR:

Partial response

CR:

Complete response

KEGG:

Kyoto Encyclopedia of Genes and Genomes

GSEA:

Gene set enrichment analysis

EMT:

Epithelial–mesenchymal transition

PCA:

Principal component analysis

KM:

Kaplan–Meier

ROC:

Receiver operating characteristic

AUC:

Area under the curve

PFS:

Progression-free survival

MDSCs:

Myeloid-derived suppressor cells

OS:

Overall survival

C1/2/3:

Cluster 1/2/3

S1/2/3:

Subtype 1/2/3

TME:

Tumor microenvironment

References

  1. Ostrom QT, Gittleman H, Liao P, Vecchione-Koval T, Wolinsky Y, Kruchko C, et al. CBTRUS statistical report: primary brain and other central nervous system tumors diagnosed in the United States in 2010–2014. Neuro Oncol. 2017;19:v1–88.

    Article  Google Scholar 

  2. Jiang T, Mao Y, Ma W, Mao Q, You Y, Yang X, et al. CGCG clinical practice guidelines for the management of adult diffuse gliomas. Cancer Lett. 2016;375:263–73.

    Article  Google Scholar 

  3. Lapointe S, Perry A, Butowski NA. Primary brain tumours in adults. Lancet. 2018;392:432–46.

    Article  Google Scholar 

  4. Louis DN, Perry A, Reifenberger G, von Deimling A, Figarella-Branger D, Cavenee WK, et al. The 2016 World Health Organization classification of tumors of the central nervous system: a summary. Acta Neuropathol. 2016;131:803–20.

    Article  Google Scholar 

  5. Comprehensive IG. Analysis of diffuse lower-grade gliomas. N Engl J Med. 2015;372:2481–98.

    Article  Google Scholar 

  6. Weber JS, D’Angelo SP, Minor D, Hodi FS, Gutzmer R, Neyns B, et al. Nivolumab versus chemotherapy in patients with advanced melanoma who progressed after anti-CTLA-4 treatment (CheckMate 037): a randomised, controlled, open-label, phase 3 trial. Lancet Oncol. 2015;16:375–84.

    Article  Google Scholar 

  7. Del Paggio JC. Immunotherapy: cancer immunotherapy and the value of cure. Nat Rev Clin Oncol. 2018;15:268–9.

    Article  Google Scholar 

  8. Zappasodi R, Merghoub T, Wolchok JD. Emerging concepts for immune checkpoint blockade-based combination therapies. Cancer Cell. 2018;33:581–98.

    Article  Google Scholar 

  9. Christofi T, Baritaki S, Falzone L, Libra M, Zaravinos A. Current perspectives in cancer immunotherapy. Cancers. 2019;11:1472.

    Article  Google Scholar 

  10. Filley AC, Henriquez M, Dey M. Recurrent glioma clinical trial, CheckMate-143: the game is not over yet. Oncotarget. 2017;8:91779–94.

    Article  Google Scholar 

  11. Sampson JH, Vlahovic G, Sahebjam S, Omuro AMP, Baehring JM, Hafler DA, et al. Preliminary safety and activity of nivolumab and its combination with ipilimumab in recurrent glioblastoma (GBM): CHECKMATE-143. J Clin Oncol. 2015;33(15_suppl):3010–3010.

    Article  Google Scholar 

  12. Zhao J, Chen AX, Gartrell RD, Silverman AM, Aparicio L, Chu T, et al. Immune and genomic correlates of response to anti-PD-1 immunotherapy in glioblastoma. Nat Med. 2019;25:462–9.

    Article  Google Scholar 

  13. Pollack IF, Jakacki RI, Butterfield LH, Hamilton RL, Panigrahy A, Normolle DP, et al. Immune responses and outcome after vaccination with glioma-associated antigen peptides and poly-ICLC in a pilot study for pediatric recurrent low-grade gliomas. Neuro Oncol. 2016;18:1157–68.

    Article  Google Scholar 

  14. Jiang P, Gu S, Pan D, Fu J, Sahu A, Hu X, et al. Signatures of T cell dysfunction and exclusion predict cancer immunotherapy response. Nat Med. 2018;24:1550–8.

    Article  Google Scholar 

  15. Cristescu R, Mogg R, Ayers M, Albright A, Murphy E, Yearley J, et al. Pan-tumor genomic biomarkers for PD-1 checkpoint blockade-based immunotherapy. Science (80-). 2018;362: eaar3593.

    Article  Google Scholar 

  16. Auslander N, Zhang G, Lee JS, Frederick DT, Miao B, Moll T, et al. Robust prediction of response to immune checkpoint blockade therapy in metastatic melanoma. Nat Med. 2018;24:1545–9.

    Article  Google Scholar 

  17. Anagnostou V, Niknafs N, Marrone K, Bruhm DC, White JR, Naidoo J, et al. Multimodal genomic features predict outcome of immune checkpoint blockade in non-small-cell lung cancer. Nat Cancer. 2020;1:99–111.

    Article  Google Scholar 

  18. Broniscer A. Past, present, and future strategies in the treatment of high-grade glioma in children. Cancer Invest. 2006;24:77–81.

    Article  Google Scholar 

  19. Dietz MS, Beach CZ, Barajas R, Parappilly MS, Sengupta SK, Baird LC, et al. Measure twice: promise of liquid biopsy in pediatric high-grade gliomas. Adv Radiat Oncol. 2020;5:152–62.

    Article  Google Scholar 

  20. Camidge DR, Doebele RC, Kerr KM. Comparing and contrasting predictive biomarkers for immunotherapy and targeted therapy of NSCLC. Nat Rev Clin Oncol. 2019;16:341–55.

    Article  Google Scholar 

  21. Klekner Á, Szivos L, Virga J, Árkosy P, Bognár L, Birkó Z, et al. Significance of liquid biopsy in glioblastoma—a review. J Biotechnol. 2019;298:82–7.

    Article  Google Scholar 

  22. Kros JM, Mustafa DM, Dekker LJM, Smitt PAES, Luider TM, Zheng PP. Circulating glioma biomarkers. Neuro Oncol. 2015;17:343–60.

    Google Scholar 

  23. Zhang H, Yuan F, Qi Y, Liu B, Chen Q. Circulating tumor cells for glioma. Front Oncol. 2021;11: 607150.

    Article  Google Scholar 

  24. Müller S, Agnihotri S, Shoger KE, Myers MI, Smith N, Chaparala S, et al. Peptide vaccine immunotherapy biomarkers and response patterns in pediatric gliomas. JCI Insight. 2018;3: e98791.

    Article  Google Scholar 

  25. Bowman RL, Wang Q, Carro A, Verhaak RGW, Squatrito M. GlioVis data portal for visualization and analysis of brain tumor expression datasets. Neuro Oncol. 2017;19:139–41.

    Article  Google Scholar 

  26. Newman AM, Liu CL, Green MR, Gentles AJ, Feng W, Xu Y, et al. Robust enumeration of cell subsets from tissue expression profiles. Nat Methods. 2015;12:453–7.

    Article  Google Scholar 

  27. Nabet BY, Esfahani MS, Moding EJ, Hamilton EG, Chabon JJ, Rizvi H, et al. Noninvasive early identification of therapeutic benefit from immune checkpoint inhibition. Cell. 2020;183:363-376.e13.

    Article  Google Scholar 

  28. Charoentong P, Finotello F, Angelova M, Mayer C, Efremova M, Rieder D, et al. Pan-cancer immunogenomic analyses reveal genotype-immunophenotype relationships and predictors of response to checkpoint blockade. Cell Rep. 2017;18:248–62.

    Article  Google Scholar 

  29. Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W, et al. Limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res. 2015;43: e47.

    Article  Google Scholar 

  30. Yu G, Wang LG, Han Y, He QY. ClusterProfiler: an R package for comparing biological themes among gene clusters. Omics J Integr Biol. 2012;16:284–7.

    Article  Google Scholar 

  31. Wu C, Qin C, Long W, Wang X, Xiao K, Liu Q. Tumor antigens and immune subtypes of glioblastoma: the fundamentals of mRNA vaccine and individualized immunotherapy development. J big data. 2022;9.

  32. Wu C, Tan J, Wang X, Qin C, Long W, Pan Y, et al. Pan-cancer analyses reveal molecular and clinical characteristics of cuproptosis regulators. iMeta. 2022;e68.

  33. Wilkerson MD, Hayes DN. ConsensusClusterPlus: a class discovery tool with confidence assessments and item tracking. Bioinformatics. 2010;26:1572–3.

    Article  Google Scholar 

  34. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci USA. 2005;102:15545–50.

    Article  Google Scholar 

  35. Rosenberg JE, Hoffman-Censits J, Powles T, Van Der Heijden MS, Balar AV, Necchi A, et al. Atezolizumab in patients with locally advanced and metastatic urothelial carcinoma who have progressed following treatment with platinum-based chemotherapy: a single-arm, multicentre, phase 2 trial. Lancet. 2016;387:1909–20.

    Article  Google Scholar 

  36. Mariathasan S, Turley SJ, Nickles D, Castiglioni A, Yuen K, Wang Y, et al. TGFβ attenuates tumour response to PD-L1 blockade by contributing to exclusion of T cells. Nature. 2018;554:544–8.

    Article  Google Scholar 

  37. Damrauer JS, Hoadley KA, Chism DD, Fan C, Tignanelli CJ, Wobker SE, et al. Intrinsic subtypes of high-grade bladder cancer reflect the hallmarks of breast cancer biology. Proc Natl Acad Sci USA. 2014;111:3110–5.

    Article  Google Scholar 

  38. Alban TJ, Alvarado AG, Sorensen MD, Bayik D, Volovetz J, Serbinowski E, et al. Global immune fingerprinting in glioblastoma patient peripheral blood reveals immune-suppression signatures associated with prognosis. JCI Insight. 2018;3: e122264.

    Article  Google Scholar 

  39. Rizvi H, Sanchez-Vega F, La K, Chatila W, Jonsson P, Halpenny D, et al. Molecular determinants of response to anti-programmed cell death (PD)-1 and anti-programmed death-ligand 1 (PD-L1) blockade in patients with non-small-cell lung cancer profiled with targeted next-generation sequencing. J Clin Oncol. 2018;36:633–41.

    Article  Google Scholar 

  40. Rooney MS, Shukla SA, Wu CJ, Getz G, Hacohen N. Molecular and genetic properties of tumors associated with local immune cytolytic activity. Cell. 2015;160:48–61.

    Article  Google Scholar 

  41. Okada H, Kalinski P, Ueda R, Hoji A, Kohanbash G, Donegan TE, et al. Induction of CD8+ T-cell responses against novel glioma-associated antigen peptides and clinical activity by vaccinations with α-type 1 polarized dendritic cells and polyinosinic-polycytidylic acid stabilized by lysine and carboxymethylcellulose in patients with recurrent malignant glioma. J Clin Oncol. 2011;29:330–6.

    Article  Google Scholar 

  42. Okada H, Butterfield LH, Hamilton RL, Hoji A, Sakaki M, Ahn BJ, et al. Induction of robust type-I CD8+ T-cell responses in WHO grade 2 low-grade glioma patients receiving peptide-based vaccines in combination with poly-ICLC. Clin Cancer Res. 2015;21:286–94.

    Article  Google Scholar 

  43. Pollack IF, Jakacki RI, Butterfield LH, Hamilton RL, Panigrahy A, Normolle DP, et al. Antigen-specific immunoreactivity and clinical outcome following vaccination with glioma-associated antigen peptides in children with recurrent high-grade gliomas: results of a pilot study. J Neurooncol. 2016;130:517–27.

    Article  Google Scholar 

  44. Pollack IF, Jakacki RI, Butterfield LH, Hamilton RL, Panigrahy A, Potter DM, et al. Antigen-specific immune responses and clinical outcome after vaccination with glioma-associated antigen peptides and polyinosinic-polycytidylic acid stabilized by lysine and carboxymethylcellulose in children with newly diagnosed malignant brainstem and nonbrainstem gliomas. J Clin Oncol. 2014;32:2050–8.

    Article  Google Scholar 

  45. Pérez-Callejo D, Romero A, Provencio M, Torrente M. Liquid biopsy based biomarkers in non-small cell lung cancer for diagnosis and treatment monitoring. Transl Lung Cancer Res. 2016;5:455–65.

    Article  Google Scholar 

  46. Chavan R, Salvador D, Gustafson MP, Dietz AB, Nevala W, Markovic SN. Untreated stage IV melanoma patients exhibit abnormal monocyte phenotypes and decreased functional capacity. Cancer Immunol Res. 2014;2:241–8.

    Article  Google Scholar 

  47. Krieg C, Nowicka M, Guglietta S, Schindler S, Hartmann FJ, Weber LM, et al. High-dimensional single-cell analysis predicts response to anti-PD-1 immunotherapy. Nat Med. 2018;24:144–53.

    Article  Google Scholar 

  48. Romano E, Kusio-Kobialka M, Foukas PG, Baumgaertner P, Meyer C, Ballabeni P, et al. Ipilimumab-dependent cell-mediated cytotoxicity of regulatory T cells ex vivo by nonclassical monocytes in melanoma patients. Proc Natl Acad Sci USA. 2015;112:6140–5.

    Article  Google Scholar 

  49. Kluger HM, Zito CR, Barr ML, Baine MK, Chiang VLS, Sznol M, et al. Characterization of PD-L1 expression and associated T-cell infiltrates in metastatic melanoma samples from variable anatomic sites. Clin Cancer Res. 2015;21:3052–60.

    Article  Google Scholar 

  50. Huang X, Tang T, Zhang G, Liang T. Identification of tumor antigens and immune subtypes of cholangiocarcinoma for mRNA vaccine development. Mol Cancer. 2021;20:50.

    Article  Google Scholar 

  51. Huang X, Zhang G, Tang T, Liang T. Identification of tumor antigens and immune subtypes of pancreatic adenocarcinoma for mRNA vaccine development. Mol Cancer. 2021;20:44.

    Article  Google Scholar 

  52. Kamphorst AO, Pillai RN, Yang S, Nasti TH, Akondy RS, Wieland A, et al. Proliferation of PD-1+ CD8 T cells in peripheral blood after PD-1-targeted therapy in lung cancer patients. Proc Natl Acad Sci USA. 2017;114:4993–8.

    Article  Google Scholar 

  53. Gerard CL, Delyon J, Wicky A, Homicsko K, Cuendet MA, Michielin O. Turning tumors from cold to inflamed to improve immunotherapy response. Cancer Treat Rev. 2021;101: 102227.

    Article  Google Scholar 

Download references

Acknowledgements

The authors would like to thank Prof. Raul Rabadan for data support. The authors also would like to thank GlioVis (http://gliovis.bioinfo.cnio.es/) for data collection.

Funding

This work was supported by the National Natural Science Foundation of China (Grant number 81802974) and grants from the National Key Technology Research and Development Program of the Ministry of Science and Technology of China (Grant Number 2014BAI04B01). Sponsoring foundations exert no effect on the writing of this manuscript.

Author information

Authors and Affiliations

Authors

Contributions

CW.W conceived the study, performed the literature search and bioinformatics analysis, and prepared the figures and manuscript; CY.Q, YZ.L and WY.L helped with data collection, analysis, and interpretation. Y.L, XY.W and K.X analyzed data and revised the manuscript; Q.L helped conceive this research and revise the manuscript. All authors read and approved the final manuscript.

Corresponding author

Correspondence to Qing Liu.

Ethics declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declared no potential conflicts of interest.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Additional file 1: Table S1.

Summary of clinical characteristics for patients included in this study. Figure S1. Prognostic role of circulating immune cells in pre-treatment samples. ad Kaplan–Meier curves for OS showing the association between relative abundance of B cell memory (a), B cell naive (b), NK cell activated (c), T cells follicular helper (d) and OS (log-rank test, p < 0.05). Figure S2. Identification of DEGs among samples with different responses at different post-treatment time points and KEGG pathways analysis. ac DEGs at week 6 (a), 15 (b) and 34 (c) were identified by comparing the differences between SD, PR, and CR patients and KEGG pathway analysis was performed based on DEGs. The size of the circle represents the number of genes enriched. Figure S3. Consensus clustering analysis of PBMC samples based on the 43 RRGs expression profile. a, b Cumulative distribution function curve (a) and delta area curve (b) of consensus clustering analysis. Figure S4. Consensus clustering analysis of LGG tumor samples based on the 43 RRGs expression profile. ac Cumulative distribution function curve (a), delta area curve (b) and consensus heatmap (c) of consensus clustering analysis in the TCGA cohort. Figure S5. Forest plot showing the prognostic value of RRGs in LGG tumors as determined by univariate cox regression analysis. Figure S6. Cellular characterization of RRG-related RiskScore in the CGGA cohort. a Heatmap of the relationship between RRG-related RiskScore and 28 immune cell subpopulations identified by a previous study. b Correlation between RiskScore and abundance of 28 immune cells. Correlations are calculated by Pearson correlation analysis. Blue represents weak positive correlation, red represents strong positive correlation.

Additional file 2. Table S2.

DEGs at week 0. Table S3. DEGs at week 6. Table S4. DEGs at week 15. Table S5. DEGs at week 34. Table S6. Intersection of DEGs at week 0, 6, 15 ,34. Table S7. PCA of 17 prognosis-related RRGs. Table S8. Coefficients of RRG-related RiskScore.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Wu, C., Long, W., Qin, C. et al. Liquid biopsy-based identification of prognostic and immunotherapeutically relevant gene signatures in lower grade glioma. J Big Data 10, 19 (2023). https://doi.org/10.1186/s40537-023-00686-8

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s40537-023-00686-8

Keywords