Peripheral blood mononuclear cell DNA methylation biomarkers for prognostic stratification in Chinese lung adenocarcinoma: a genome-wide epigenetic profiling study

Abstract

Background

While DNA methylation profiling from peripheral blood mononuclear cells (PBMCs) has demonstrated utility in cancer risk prediction, notably for non-small cell lung cancer (NSCLC), its prognostic value for survival stratification in Chinese lung adenocarcinoma (LUAD) patients remains unestablished. This study addresses whether PBMC-derived methylation signatures can discriminate clinical outcomes in EGFR-mutation LUAD subgroups.

Methods

We performed genome-wide methylation analysis of PBMCs from LUAD patients using the Infinium Methylation EPIC 850 K array. Clinical characteristics were associated with overall survival (OS) through Cox regression. Prognostic differentially methylated positions (DMPs) were identified via Lasso regression, followed by the construction of risk-score models. Functional enrichment (KEGG/GO) and tissue microarray-based immunohistochemistry (IHC) for FKBP4 expression (n = 90 LUAD samples) were conducted. Analyses were conducted in R 4.4.1 with curated Bioconductor packages.

Results

In the retrospective cohort of 174 Chinese LUAD patients (April 2014–September 2019), PBMC analysis of 128 cases revealed 12 hypomethylated DMPs were associated with OS. EGFR-mutant patients (n = 66) showed 325 significant DMPs (|Δβ|≥ 0.06, P ≤ 0.01), with four DMPs (cg05802998, cg19313959, cg00685115, cg15224444) independently predicting OS. The cg19313959 located in the TSS1500 region of FKBP4 gene (Δβ = 0.21) demonstrated the strongest methylation shift. Reduced FKBP4 protein expression was associated with improved survival (HR = 0.42, 95%CI 0.24–0.72). In EGFR-wildtype patients (n = 51), three prognostic DMPs emerged from 2,531 candidates. EGFR mutation-specific prognostic scoring models were established successfully, while pathway analyses revealed divergent biological processes between EGFR subgroups.

Conclusion

In this epigenome-wide study based on PBMCs in Chinese patients with LUAD, methylation signatures dependent on EGFR mutations and predictive of survival were identified.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13148-025-01992-z.

Introduction

Lung cancer remains the leading cause of cancer-related mortality globally, with approximately 2.2 million new cases and 1.8 million deaths annually [1]. In 2022, the highest age-standardized incidence rates (ASRs) of adenocarcinoma in East Asia were observed in both males and females, differing from patterns in Europe and North America [2]. There are still disparities in risk factors and incidence rates between males and females in China [3]. Lung adenocarcinoma (LUAD) exhibits a higher frequency of proto-oncogene mutations in receptor tyrosine kinases such as EGFR and ALK, whereas lung squamous cell carcinoma (LUSC) is characterized by the inactivation of tumor suppressor genes like TP53 and a relatively limited number of driver gene mutations, suggesting a distinct multi-driver mutation profile. High-frequency mutations in LUAD, including EGFR and KRAS, occur at frequencies below 5% in LUSC [4]. These genetic mutations in LUAD significantly impact short-term and long-term clinical outcomes, particularly EGFR mutations in the Chinese population.

Liquid biopsy has emerged as a critical non-invasive tool for tumor prognostication, encompassing analyses such as circulating tumor DNA (ctDNA), enumeration of circulating tumor cells (CTCs), exosome-derived biomarkers, and methylation profiling of peripheral blood mononuclear cells (PBMCs) or blood cell-free DNA [5, 6]. In addition to the above, liquid biopsy was also employed to explore etiological factors in several other diseases. A study investigated gene expression alterations and immune response mechanisms in bacterial meningitis. This was achieved by utilizing RNA extracted from whole blood and Gene Chip Human Gene arrays [7]. Another research effort concentrated on analyzing serum for gene/protein variations associated with recurrent miscarriage [8]. Regarding motor neurone disease, whole blood transcriptome sequencing was carried out to identify biomarkers for clinical diagnosis and monitoring purposes [9]. Genome-wide DNA methylation profiling from PBMCs led to the development of an early LUAD Diagnostic Panel (LDP score), which integrates three differentially methylated CpG positions (DMPs) with age and sex variables. This panel exhibited superior diagnostic performance (AUC = 0.916, 88.17% sensitivity, 80.20% specificity), outperforming conventional markers like CEA (4.69%) and CT (35.21%) in detecting both early-stage and advanced malignancies [10]. Additionally, while DNA methylation profiling of PBMCs has been primarily utilized for early lung cancer detection, its potential for prognostic applications remains largely unexplored [11]. Therefore, we hypothesize that PBMC-derived methylation profiles could effectively differentiate the prognosis of LUAD patients.

This study utilized the Infinium Methylation EPIC 850 K array to profile DNA methylation in PBMCs from LUAD patients and correlated these profiles with overall survival. Stratification by EGFR mutation status identified distinct prognostic DMPs: four DMPs in EGFR-mutant cohorts (n = 66) and three in EGFR wild-type subgroups (n = 51), as determined by LASSO regression analysis. Functional enrichment analyses (KEGG/GO) revealed that these DMPs were associated with insulin signaling pathways and cortisol synthesis and secretion pathways, respectively. A methylation-based risk score exhibited significant prognostic stratification capability. This proof-of-concept study demonstrates that PBMC methylation signatures may serve as potential non-invasive prognostic biomarkers for Chinese LUAD patients, necessitating further multicenter validation in larger cohorts.

Methods

Patient enrollment and PBMC sample isolation

A total of 311 lung cancer patients from the Department of Respiratory Medicine at Henan Cancer Hospital were recruited for this study between April 2014 and September 2019. Participants were included based on the following criteria: (1) Pathologically confirmed diagnosis of lung cancer (details provided in Table 1); (2) absence of other malignancies; (3) no known inflammatory conditions that could influence the characteristics of peripheral blood mononuclear cells (PBMCs), such as bacterial or viral infections, asthma, autoimmune diseases, or active thyroid disorders. Blood samples were collected prior to any pathological diagnosis or clinical intervention retrospectively, with the final follow-up conducted on September 16, 2023.

Table 1.

The clinical characteristics in 311 lung cancer patients

	N	%
Sex
Male	200	64.3
Female	111	35.7
Age
< 65	195	62.7
>  = 65	116	37.3
Smoke
Yes	155	50
No	155	50
Stage
I–III	65	21.2
IV	242	78.8
Histology
Adenocarcinoma	174	55.9
Squamous carcinoma	52	16.7
Small cell	54	17.4
Others	31	10.0
EGFR mutation
Yes	97	31.1
No	118	37.8
NA	97	31.1

For the extraction of PBMCs, we employed Histopaque-1077, a human lymphocyte isolation reagent (Sigma, 10,771–500 ml), following the manufacturer’s protocol. Specifically, 3 ml of Histopaque-1077 was added to a sterile 15-ml conical centrifuge tube, followed by careful layering of 3 ml of peripheral whole blood on top. The resulting gradient was then centrifuged at 400×g for 30 min at room temperature without brake. After centrifugation, the PBMC layer was carefully harvested and transferred to a new sterile 15 ml conical centrifuge tube containing 10 ml of phosphate-buffered saline (PBS). This suspension was subjected to a second centrifugation at 250×g for 10 min at room temperature. The supernatant was aspirated and discarded, and the pellet was resuspended in an additional 5 ml of PBS before undergoing another centrifugation step under the same conditions. Following aspiration and discard of the supernatant, the PBMCs were collected as sediments and stored at − 80 °C for subsequent analyses.

DNA extraction and bisulfite conversion

DNA was extracted from PBMCs using the TIANamp Genomic DNA Kit (Cat# DP304-03) according to the manufacturer’s instructions. The purity and concentration of the extracted DNA were assessed using a Qubit 3.0 fluorometer (Thermo Fisher Scientific), followed by electrophoresis analysis of 2 μL DNA samples on a nucleic acid gel. Samples with low quality, as indicated by A260/280 ratios ≤ 1.6 or ≥ 2.1, the presence of protein or RNA contaminants, or those that could not be accurately quantified, were excluded. The remaining high-quality DNA was stored at − 80 °C for subsequent experiments. For bisulfite conversion, 1 μg of the isolated PBMCs DNA was processed using the Zymo EZ DNA Methylation-Gold Kit in accordance with the manufacturer’s protocol. Specifically, 20 µL of the DNA sample was mixed with 130 µL of Lightning Conversion Reagent and incubated in a thermocycler under the following conditions: 98 °C for 10 min, 64 °C for 150 min, and then held at 4 °C indefinitely. After incubation, the bisulfite-converted DNA was combined with M-Binding Buffer, passed through a Zymo-Spin™ IC Column, desulphonated, washed, and finally eluted in 20 µL of M-Elution Buffer.

Human methylation 850 K BeadChip

The analysis using the Infinium Human Methylation 850 K BeadChip was performed following the manufacturer’s guidelines, and the resulting data were analyzed with the ChAMP 2.36.0 package (The Chip Analysis Methylation Pipeline) (RRIDS: ChAMP (RRID:SCR_012891) in R version 4.4.1[12]. The β value was used to represent the level of DNA methylation, calculated using the formula [β = intensity of the methylated allele (M)/ (intensity of the unmethylated allele (U) + intensity of the methylated allele (M) + 100)], which ranges from 0 (no methylation) to 1 (complete methylation). Initially, we excluded probes with a detection p value greater than 0.01, those with fewer than 3 beads in at least 5% of samples, non-CpG probes, multi-hit probes, probes on chromosome X/Y, and SNP-related probes were excluded. This filtering resulted in a total of 730,036 probes for further analysis. Subsequently, the β value matrix was then normalized using BMIQ to adjust for biases associated with type I and type II probes. Singular value decomposition analysis (SVA) was conducted to evaluate the batch effects from the BeadChip Slide and Array, and ComBat was applied to correct these batch effects. All CpG sites were annotated with EPICanno.ilm10b4.hg19, and differentially methylated CpG positions (DMPs) were identified using the champ.DMP function. The adjusted p values were calculated using the Benjamini–Hochberg method, and CpGs with |Δβ|≥ 0.06 and p value ≤ 0.01 were considered DMPs, noting that the p values were not corrected for multiple testing.

Construction of the risk scoring model

Initially, we conducted univariate Cox regression analysis to identify CpG IDs significantly associated with survival. Subsequently, we utilized Cox LASSO regression through the glmnet package in R to refine our selection of candidate DMPs. LASSO regression is a method that shrinks regression coefficients toward zero via L1 regularization, effectively reducing dimensionality and mitigating multicollinearity among variables. Subsequently, we conducted multivariate Cox regression to further refine our selection of survival-related DMPs. The risk score for the identified DMPs was calculated by multiplying the coefficients from the multivariate Cox regression by their corresponding β values. Patients were then categorized into low-risk and high-risk groups based on the median risk score. To validate our risk model, we generated ROC curves using the pROC package in R. Additionally, we created survival curves and other visualizations using the ggsurvplot function in R version 4.4.1.

Tissue specimens and immunohistochemistry

The tissue microarray (TMA, no. HLugA180Su12) was obtained from Shanghai Outdo Biotechnology Co., Ltd in Shanghai, China. This TMA included 90 specimens of lung adenocarcinoma (LUAD) tissues and 90 specimens of adjacent non-cancerous tissues, along with survival data collected from 2014 to 2021. The specimens were prepared as formalin-fixed paraffin-embedded sections, each 4 µm thick, and were placed on silanized slides. To prepare the slides for analysis, they underwent deparaffinization using xylene followed by rehydration through a series of graded alcohol solutions. The slides were then treated with 10 mM citrate buffer at pH 8.0 and subjected to heat for antigen retrieval. To eliminate any endogenous peroxidase activity, a 3% hydrogen peroxide solution was applied. Subsequently, the sections were incubated overnight at 4 °C with an anti-FKBP4 antibody diluted at 1:3000 (Thermo Fisher Scientific, JG96-30) (RRID: AB_2848709), followed by the application of a biotinylated secondary antibody. Standard DAB staining was employed to visualize the immunohistochemistry (IHC) targets. Scores were obtained by estimating the staining intensity (0 = none; 1 = mild; 2 = intermediate; 3 = intense) and the percentage of positive cells (0 = none; 1 = 1 − 25%; 2 = 26 − 50%; 3 = 51 − 75%; 4 = more than 75%). The final score was calculated by multiplying the intensity score by the percentage score, with a cut-point score set at 3.

Statistical analysis

All statistical analyses were conducted using R version 4.4.1, with a significance threshold set at a P value of less than 0.05. The methodologies employed in this study were based on reference manuals for R packages and Bioconductor packages. The adjusted P values were calculated using the Benjamini–Hochberg method, with criteria of |Δβ|≥ 0.06 and an adjusted P value of less than 0.05 used to define DMPs. Gene Ontology (GO) analysis and Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis were conducted to explore the genes affected by the DMPs. The biological processes for both GO and KEGG pathway enrichment analyses were carried out using the clusterProfiler package in R 4.4.1, with visual representations created using ggplot2.

Results

Clinicopathological characteristics of 311 lung cancer patients

The study included a cohort of 311 lung cancer patients, comprising 200 males (64.1%) and 111 females (35.9%). Among these patients, there were 174 cases of adenocarcinoma (55.9%), 52 cases of squamous cell carcinoma (16.7%), and 54 cases of small cell lung cancer (17.4%), as summarized in Table 1. Methylation analyses were performed on samples derived from PBMCs, with an overview of the study presented in Fig. 1. Of the 174 patients with LUAD, OS data were available for 129 individuals. After excluding one patient due to insufficient DNA quality, univariate survival analysis was conducted on 128 LUAD patients, as detailed in Table 2. This analysis identified gender and tumor stage as statistically significant factors (P < 0.05), which were subsequently incorporated as adjusted variables in subsequent analyses. Cox proportional hazards regression analysis revealed that among the 128 LUAD patients who received first-line chemotherapy combined with targeted therapy (n = 26), the adjusted P value was 0.029, with a hazard ratio (HR) of 0.578 and a 95% confidence interval (CI) of 0.36–0.95. No significant associations were observed for other variables after adjustment.

Fig. 1.

Workflow chart of the study design

Table 2.

Univariate analysis of characteristics of 128 LUAD and their associations with OS

Patients’ characteristics	N	%	Overall Survival (OS)
			Median OS (months)	P	HR	95%CI
Sex
Female	65	50.8	35.9
Male	63	49.2	22.1	0.017	1.60	1.10–2.40
Age
< 65	79	61.7	27.3
>  = 65	49	38.3	30.5	0.250	1.30	0.85–1.90
Smoke
No	82	64.1	32.4
Yes	46	35.9	22.6	0.110	1.40	0.93–2.10
Stage
IV	111	86.7	26.2
I–III	17	13.3	92.7	0.000	0.26	0.12–0.53
EGFR mutation
No	51	43.6	21.7
Yes	66	56.4	34.4	0.072	0.69	0.45–1.00
1st_TKI
No	61	52.1	24.0
Yes	56	47.9	31.7	0.750	0.10	0.62–1.40
1st_Chemotherapy
No	43	35.0	31.3
Yes	80	65.0	26.4	0.640	1.10	0.73–1.70
1st_EGFR-TKI + Chemo
No	97	78.9	26.4
Yes	26	21.1	37.0	0.240	0.75	0.47–1.2

The subgroup of Sex and Stage, the bolditalic P value means achieveing statistical power; HR and 95% confidence intervals in bold also indicates statistical significance

EGFR-TKI as the first-line treatment (1st_TKI), Chemotherapy as the first-line treatment (1st_ Chemotherapy), EGFR-TKI combined with chemotherapy as the first-line treatment (1st_EGFR-TKI + Chemo)

Differentially methylated positions (DMPs) in 128 LUAD patients

Using the median OS as the cutoff, patients were stratified into two groups: a favorable-prognosis group and a poor-prognosis group. The Champ.DMP analysis identified a total of 12 hypomethylated DMPs distributed across 9 distinct genes, as summarized in Table 3.

Table 3.

Features of 12 differentially methylated CpG positions among longer and shorter OS in the 128 LUAD

Pro ID	Δβ value	P Value	Adjust P Value	CHR	UCSC_RefGene	UCSC_RefGene_Group
cg14177084	0.046778998	3.83E-08	0.026125516	7	HIP1	Body-shore
cg20884163	0.00488986	7.16E-08	0.026125516	9	PTCH1	Body-island
cg14466441	0.045521685	2.34E-07	0.040819367	6		IGR-opensea
cg15224444	0.041060078	2.98E-07	0.040819367	7	ELN	TSS1500-opensea
cg08450732	0.054248542	3.61E-07	0.040819367	1	LDLRAP1	Body-opensea
cg15157266	0.007431623	3.80E-07	0.040819367	10		IGR-opensea
cg10524346	0.067340409	4.32E-07	0.040819367	6	PARK2	Body-opensea
cg07100877	0.024910317	4.47E-07	0.040819367	17	BAIAP2	Body-opensea
cg06484360	0.007372024	5.19E-07	0.04210219	19	LSM4	TSS1500-shore
cg09378497	0.004404351	6.55E-07	0.044040035	16	CLCN7	TSS200-island
cg03546163	0.076830179	6.64E-07	0.044040035	6	FKBP5	5’UTR-shore
cg24477906	0.014484446	7.83E-07	0.047605289	3		IGR-opensea

LUAD lung adenocarcinoma, CHR chromosome, TSS1500 1500 bp of the start site of transcription

*The Δβ value is the average β values of the shorter group minus the average β values of the longer group. The differential methylated CpGs position were calculated by champ.DMP function. The adjusted p value were computed using the Benjamini–Hochberg method. Probe ID: identification of the probe (from 850 K Bead Chip Array)

DMP screening in 66 EGFR-mutant LUAD patients

In the subset of 66 patients with EGFR-mutant LUAD, a total of 8,508 significant DMPs were identified following normalization and batch correction. Using specific thresholds (|Δβ|≥ 0.06, P ≤ 0.01), 325 DMPs corresponding to 192 genes were retained, including 76 hypermethylated sites (34.38%) and 249 hypomethylated sites (76.62%), as depicted in the volcano plot (Fig. 2). Lasso regression analysis for binary outcomes (Supplementary Figure S1A-B) identified four prognostic CpG sites: cg05802998, cg19313959, cg00685115, and cg15224444 (Table 4). Survival analysis demonstrated that hypermethylation of these CpGs was associated with improved overall survival (OS), as shown in the survival curves (Fig. 3A) and supported by the violin plots (Fig. 3B). According to the DMPs, the Cox-based prognostic model effectively stratified patients into high- and low-risk groups. High-risk patients exhibited a median overall survival (OS) of 22.5 months, compared to 56.6 months for low-risk patients, with a log-rank P value < 0.0001 and an area under the curve (AUC) of 0.9587 (Fig. 4A–B). After taking smoking status, DNA methylation age (DNAm age) [13] and first-line targeted therapy into account as adjustment factors, the significance of the p values for the four differentially methylated positions (DMPs) and the risk score (P < 0.001, HR = 0.24, 95% CI 0.13–0.44) remained unchanged. Furthermore, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses demonstrated significant enrichment in protein serine/threonine kinase activity and insulin signaling pathways (Fig. 5A–B).

Fig. 2.

The top 5 DMPs methylation and demethylation in LUAD patients with EGFR-mutant

Table 4.

Features of 4 differentially methylated CpG positions among longer and shorter OS in 66 patients with EGFR mutation

Pro ID	Δβ value	P Value	Adjust P Value	CHR	UCSC_RefGene	UCSC_RefGene_Group
cg05802998	− 0.025525	3.08E-06	0.005857518	19	C19orf38	ExonBnd-opensea
cg19313959	− 0.052773349	2.17E-07	0.001881996	12	FKBP4	TSS1500-shore
cg00685115	− 0.004215017	2.21E-08	0.00073296	12	CRY1	Body-shore
cg15224444	− 0.047182884	1.77E-06	0.004451039	7	ELN	TSS1500-opensea

CHR chromosome, TSS1500 1500 bp of the start site of transcription

*The Δβ value is the average β values of the shorter group minus the average β values of the longer group. The differential methylated CpGs position were calculated by champ.DMP function. The adjusted p value were computed using the Benjamini–Hochberg method. Probe ID: identification of the probe (from 850 K Bead Chip Array)

Fig. 3.

A The survival curves for OS by the methylation levels of 4 DMPs in LUAD patients with EGFR-mutant. B. The Violin plot by the methylation levels of 4 DMPs and different prognosis in LUAD patients with EGFR-mutant

Fig. 4.

The survival curve by riskscore group (A) and the ROC curve (B) using the prognostic risk score in 66 LUAD patients with EGFR-mutant

Fig. 5.

A. GO analyzse in 66 LUAD patients with EGFR-mutant. B. KEGG analyzse in 66 LUAD patients with EGFR-mutant

A key finding was that hypermethylation of cg19313959 (Δβ = 0.053, the largest Δβ value in the 4 DMPs and located at the FKBP4 locus) was predictive of a favorable prognosis (log-rank P < 0.0001). Data from The Cancer Genome Atlas (TCGA) showed that lower FKBP4 mRNA expression was linked to better survival outcomes in LUAD (log-rank P = 0.0005) (https://kmplot.com/analysis/). Furthermore, immunohistochemistry analysis of 89 LUAD tissue samples confirmed that patients with low FKBP4 expression had a longer median OS of 69 months compared to 36 months for those with higher expression, with a log-rank P value of 0.016 (Fig. 6A–B).

Fig. 6.

The survival curve by FKBP4 protein expression groups (A), and the forest plot by clinical charactristics and FKBP4 protein expression groups in 90 LUAD tissue microarray (B)

DMP screening in 51 EGFR wild-type LUAD patients

The 51 patients with EGFR wild-type lung adenocarcinoma (LUAD) identified a total of 83,542 significant differentially methylated positions (DMPs). By applying specific thresholds (|Δβ|≥ 0.06, P ≤ 0.01), we narrowed this down to 2531 DMPs associated with 1146 genes, which included 1191 hypermethylated sites (47.06%) and 1340 hypomethylated sites (52.94%), as illustrated in the volcano plot (Fig. 7). Using Lasso regression, three specific CpG sites were identified: cg25670864, cg04490108, and cg17929042 (Table 5). Notably, hypermethylation of cg04490108 was linked to a favorable prognosis, whereas hypermethylation of cg25670864 and cg17929042 was associated with poorer outcomes, as shown in Supplementary Figures S3A-B. A prognostic model developed from these findings successfully stratified patients into high- and low-risk groups, demonstrating a median overall survival (OS) of 13.2 months for the high-risk group compared to 39.4 months for the low-risk group, with a log-rank P value of less than 0.0001 and an area under the curve (AUC) of 0.9803 (Supplementary Figures S4A-B). In the multivariate analyses, the first line of chemotherapy, DNA methylation age [13] and smoking status were also incorporated as adjustment variables. Nevertheless, the p values for the three differentially methylated positions (DMPs) and the risk score (P < 0.001, HR = 0.20, 95% confidence interval: 0.08–0.47) remained significant. Furthermore, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses suggested involvement of cell growth regulation and cortisol synthesis and secretion pathways (Supplementary Figures S5A-B).

Fig. 7.

The top 5 DMPs methylation and demethylation were respectively labeled in LUAD patients with EGFR-wild

Table 5.

Features of 3 differentially methylated CpG positions among longer and shorter OS in 51 patients without EGFR mutation

Pro ID	Δβ value	P Value	Adjust P Value	CHR	UCSC_RefGene	UCSC_RefGene_Group
cg25670864	0.090301829	2.42E-05	0.001691563	7	ELN	Body-opensea
cg04490108	− 0.005225653	1.78E-04	0.00571583	1	RPRD2	5’UTR-island
cg17929042	0.011303092	1.03E-05	0.001006332	7	AUTS2	Body-opensea

CHR chromosome, TSS1500 1500 bp of the start site of transcription

^*The Δβ value is the average β values of the shorter group minus the average β values of the longer group. The differential methylated CpGs position were calculated by champ.DMP function. The adjusted p value were computed using the Benjamini–Hochberg method. Probe ID: identification of the probe (from 850 K Bead Chip Array)

The calculation of DNA methylation age for EGFR mutated and EGFR wild type in LUAD patients

DNA methylation age was calculated for both EGFR mutation statuses in patients with LUAD. The median DNAm age in LUAD patients with EGFR mutations (62.4 years) was younger than that in those with EGFR wild-type (70.0 years). Among EGFR-mutant LUAD patients, the median OS of those with a lower DNAm age was 39.9 months, while that of patients with a higher DNAm age was 31.7 months. In 51 EGFR-wild-type LUAD patients, the median OS of those with a lower DNAm age was 24.1 months, and that of patients with a higher DNAm age was 19.7 months. These findings were consistent with Guidry’s research [13].

The 450 K methylations from lung cancer tissue in TCGA LUAD Cohort and our 850 K methylation data from PBMC

We analyzed 280 LUAD cases from TCGA with OS data, using a 450 K methylation array and median OS as a cutoff. This analysis did not reveal any significant DMPs. However, when stratifying the data by the first tertile of OS, we identified a total of 1,695 DMPs, as shown in Supplementary Figure S6. By applying specific thresholds of |Δβ|≥ 0.05 and an adjusted P value < 0.05, we retained seven significant DMPs, which are detailed in Supplementary Table S1. Furthermore, intersecting these results with our non-mutant cohort revealed 178 overlapping DMPs, as depicted in Supplementary Figure S7. Among these, 10 sites met the criteria of |Δβ|≥ 0.0125 and an adjusted P value < 0.05, which are listed in Supplementary Table S2.

Comparative analysis of DNA methylation patterns and their clinical prognostic implications in EGFR gene subtypes with exon 19 deletion versus exon 21 L858R mutation

There were 32 patients with exon 19 deletion, 29 patients with exon 21 L858R and 5 patients with atypical mutations. After comparative analysis of DNA methylation between exon 19del and exon 21 L858R, a total of 8,580 significant DMPs were identified following normalization and batch correction. Using specific thresholds (|Δβ|≥ 0.06, P ≤ 0.01), 234 DMPs corresponding to 142 genes were retained, including 208 hypermethylated sites (88.9%) and 26 hypomethylated sites (11.1%), as depicted in the volcano plot (Supplemental Fig. 8). In the univariate analysis, patients with exon 21 L858R had a relatively worse OS than patients with exon 19 deletion, HR = 1.7, 95%CI = 0.97–3.00, P = 0.062. However, this difference did not reach statistical significance. In the multi-variate analysis, it was still not statistically significant, HR = 1.71, 95%CI = 0.94–3.10, P = 0.077, which was presented as forest plot in the Supplemental Fig. 9. The GO and KEGG analyses were also conducted and were presented in the Supplementary Fig. 10A and Supplementary Fig. 10B, respectively.

Discussion

Epigenetic modifications play a pivotal role in orchestrating the abnormal expression of tumor-related genes, thereby promoting cellular malignancy and driving cancer progression. Emerging non-invasive DNA methylation biomarkers hold great significance for predicting prognosis and therapeutic response in cancer [14]. In the Chinese population, a substantial proportion of lung adenocarcinoma cases have EGFR gene mutations, resulting in distinct treatment strategies and clinical outcomes compared to EGFR mutation-negative lung adenocarcinoma. The application of non-invasive peripheral blood mononuclear cell methylation detection as a clinical prognostic marker carries considerable practical importance.

This study is the first to use an 850 K methylation chip in Chinese LUAD patients, with median survival time as the cutoff point. We identified 12 DMPs. Patients were divided into EGFR mutation-positive and mutation-negative groups. Among LUAD patients with EGFR mutations, four DMPs with prognostic significance were found: cg05802998, cg19313959, cg00685115, and cg15224444. Higher methylation levels of these DMPs correlated with improved prognosis. Notably, cg19313959 showed the largest Δβ value and is located on the TSS1500-shore of the FKBP4 gene. Xiao H’s research found that the higher methylation level of FKBP4 usually happened in LUAD than normal lung tissue [15]. FKBP4 mRNA expression was associated with prognosis in TCGA LUAD samples. FKBP4 may modulate IKK/NF-κB signaling to promote the progression of lung adenocarcinoma [16]. Immunohistochemistry analysis of FKBP4 in 90 LUAD tissue microarrays revealed that the median overall survival (OS) for low-expression patients was 69 months, compared to 36 months for high-expression patients (Log-rank test P = 0.016). Therefore, the hypothesis could be that, patients with hypermethylation of cg19313959, which is located in the TSS1500-shore, exhibit lower FKBP4 expression levels in tumor tissue. Consequently, these patients may have a better prognosis compared to those with relatively hypomethylated cg19313959. However, the concrete mechanism of FKBP4 and the methylation status of cg19313959 needs to be further clarified in vitro. In LUAD patients without EGFR mutations, three DMPs significantly affecting prognosis were identified: cg25670864, cg04490108, and cg17929042. Hypermethylation of cg04490108 was linked to better outcomes, while hypermethylation of the other two DMPs was associated with poorer outcomes. The directionality of methylation changes (hypermethylation versus hypomethylation) and their association with survival outcomes exhibit inconsistency across DMPs. DNA methylation occurring at the transcription start site or promoter typically represses gene expression, whereas gene body methylation can facilitate enhanced gene expression. However, more intricate mechanisms of gene body methylation were involved in influencing tumor occurrence, progression, treatment, and prognosis [17]. There is a pressing need for a more robust biological rationale to elucidate these patterns, which represents a limitation of our study. KEGG and GO analyses indicated that the insulin signaling pathway is critical in EGFR-mutant LUAD patients, whereas cortisol synthesis and secretion pathways are significant in EGFR wild-type LUAD patients. Using a prognostic model scoring system, we successfully stratified patients into high- and low-risk groups based on methylation levels, effectively predicting OS. Additionally, we conducted DMPs analysis on the 450 K tumor tissues from TCGA and identified eight DMPs. By comparing these DMPs in TCGA with the EGFR gene mutation-negative population in our cohort, we found 178 overlapping DMPs.

PBMCs serve as critical research samples, capable of predicting efficacy and prognosis [18]. Previous studies have demonstrated that DNA methylation in PBMCs can predict cancer prognosis and cancer risk in lung, liver, colon and breast cancer [19–24]. Additionally, anti-PD-1/PD-L1 monotherapy is universally acknowledged to be ineffective in EGFR-mutant NSCLC due to low tumor mutation burden and high intratumoral heterogeneity. Moreover, EGFR-mutant NSCLC is classified as a “cold” tumor type, characterized by fewer tumor-infiltrating lymphocytes and an immunosuppressive tumor microenvironment [25]. Our research demonstrates that DMPs from PBMCs can predict prognosis not only in the total LUAD but also across groups with differing EGFR gene mutation statuses. Notably, hypermethylation of cg19313959, located in the TSS1500-shore region of the FKBP4 gene, predicts better overall survival (OS) in LUAD patients with EGFR mutations. The most significant methylation modifications are observed in transitional areas of CpG islands (CGIs), termed shores. In lung cancer cell lines, promoter-associated CGI 5′-shores exhibit higher methylation levels compared to 3′-shores [26]. Furthermore, with the onset of cancer, promoter hypermethylation might activate the genes expression, while the hypermethylation of gene body permit more robust expression for certain genes [27]. Therefore, it might be speculated that cg19313959 hypermethylation in FKBP4 downregulates its expression in Chinese LUAD with EGFR gene mutation. Further functional experiments are required to validate this hypothesis. Meanwhile, prior studies indicate FKBP4 is significantly upregulated in various cancers, which is associated with trans-methylase expression. Additionally, the FKBP4 gene is closely linked to immune-infiltrating lymphocytes, immune checkpoint genes, immune regulatory factors, tumor mutational burden (TMB), mismatch repair (MMR), and microsatellite instability (MSI), thereby playing an important role in the efficacy of immunotherapy[15]. Similarly, our data demonstrated that low expression of FKBP4 protein was significantly associated with better OS in a cohort of 90 LUAD patients by immunohistochemistry. Consistently, Shuai Zong et al. also reported that FKBP4 expression was upregulated in LUAD samples and predicted significantly shorter OS based on TCGA and another independent LUAD cohort. Furthermore, FKBP4 enhance the transcriptional activity and nuclear translocation of NF-κB in vitro [27]. Based on these contents, it is plausible that hypermethylation of cg19313959, located in the TSS1500-shore region of the FKBP4 gene, may downregulate FKBP4 expression, thereby contributing to longer OS in Chinese LUAD patients harboring EGFR mutations. Future basic research should investigate whether the methylation level of cg19313959 affects FKBP4 gene and protein expression in EGFR-mutated lung adenocarcinoma cell lines. Additionally, larger sample cohorts are warranted to validate our findings in LUAD patients with EGFR mutations.

In the subgroup of LUAD patients without EGFR-mutant, cg25670864 hypomethylation was associated with better OS. It is located in the body-opensea of the ELN gene and exhibited the highest Δβ in the three DMPs. The ELN gene encodes Elastin protein, which is a component of the extracellular matrix and may influence the development and progression of cancer [28]. The dual role of elastin in NSCLC—as both a protective matrix component and a target for protease degradation—makes it a critical entry point for prognosis assessment. Its expression levels, degradation products, and structural changes reflect tumor biology and inform treatment strategies for matrix remodeling [29, 30]. In the TCGA LUAD cohort, the majority of patients had wild-type EGFR. Thereby, we analyzed the intersection of the TCGA LUAD 450 K methylation array and our LUAD 850 K methylation array from patients without EGFR mutations. A total of 178 DMPs were identified in this overlap. Previous studies demonstrated that the role of cancer risk prediction based on DNA methylation profiles derived from PBMCs had significant concordance with that extracted from the corresponding tumor tissue [31, 32]. Furthermore, it is necessary to validate this result in larger sample size cohort in LUAD.

Cigarette smoking is a crucial determinant in the initiation and progression of lung cancer. It affects blood methylation and transcriptome in chronic inflammatory diseases [33]. However, in our study, smoking status had no influence on the associations between DMPs and the median OS of LUAD patients with EGFR mutations or without EGFR mutations. In Chinese patients with lung adenocarcinoma, EGFR mutations are more prone to occur in non-smokers than the other populations [34]. In the Chinese LUAD patients, our findings of younger DNAm age in EGFR-mutant LUAD and worse median OS in younger DNAm age patients with both EGFR-mutant and wild were consistent with Guidry’s research [13]. Nevertheless, DNAm age was not found to affect the methylations profiles and OS in this PBMCs study.

In the GO/KEGG analysis, protein serine/threonine kinase activity and the insulin signaling pathway had significant effects in LUAD with EGFR gene mutations, consistent with previous studies showing that Metformin could restore sensitivity to EGFR-TKIs by inhibiting the IGF-1R pathway [35, 36]. In patient with EGFR mutation, Insulin signaling pathway components, notably IGFBP1 and IGFBP2, have been implicated in the development of EGFR-TKI resistance by modulating drug tolerance in EGFR-mutated cells [37–39].

The analysis also revealed that cell growth regulation and cortisol synthesis and secretion pathways were crucial in the subgroup of LUAD patients with EGFR-wild type. It has been shown that cortisol secretion affects clinical outcomes in malignant neoplasms [40], and lung cancer may regulate the tumor microenvironment by releasing immunoregulatory glucocorticoids, potentially causing immune evasion and drug resistance [41]. This may partly explain why immunotherapy is more effective in lung adenocarcinoma patients without EGFR mutations than in those with EGFR mutation-positive tumors.

In future clinical practice, the integration of several DMPs into assay kit, along with the conventional clinical prognostic characteristics and combination with single nucleotide variants (SNVs), transcriptome RNA-Seq analysis, circulating tumor DNA (ctDNA) or minimal residual disease (MRD) analysis, may further enhance the prognostic value for different EGFR mutation type of lung adenocarcinoma. While the economic efficiency and clinical operability should be comprehensively considered in clinical settings. Furthermore, a challenge emerges regarding how to attain more accurate sampling and detection for a comparative analysis between the future DMPs assay kits and the existing ctDNA assay kits.

Conclusions

In our study, several significant DMPs were identified as being associated with the OS of Chinese LUAD patients, regardless of whether they harbored EGFR mutations, through Lasso analysis. However, there are some limitations in our study that warrant acknowledgment. First, the findings of this study are based on the Chinese lung adenocarcinoma population and should not be straightforwardly extrapolated to other populations. Moreover, the biological functions of specific methylation sites require further validation and larger sample size research are needed to confirm our findings. Besides, our findings were derived from methylation profiles of PBMC without the matched tumor tissue validation and targeted bisulfite sequencing for key DMPs. Although patients with systemic inflammation were excluded, variations in the immune cell composition could still exert a substantial impact on methylation patterns. Furthermore, given the heterogeneity of methylation profiles between tumor tissue and PBMCs, validation through multi-omics approaches is imperative. Additionally, the heterogeneity of methylation profiles between tumor tissue and PBMCs necessitates validation via multi-omics approaches. Other potential confounders, such as comorbidities, treatment adherence, and environmental exposures, may also influence the outcomes. Future studies should prioritize the dynamic monitoring of methylation status, as fluctuations in methylation levels at various time points may exhibit significant correlations with treatment efficacy and disease progression. This represents one of the limitations in our current study. Notably, the methylation level of cg19313959, located in TSS1500-shore of the FKBP4 gene, was associated with OS in LUAD patients with EGFR-mutant. We also confirm that lower FKBP4 expression was associated with better OS in LUAD using immunohistochemistry.

Author contribution

The main manuscript text was written by Peng Li; Data collection and follow up were performed by Cuicui Zhang and Sen Yang; Data analysis and figure preparation were performed by Peng Li, Yingxi Wu, Haiyang Chen and Shuxiang Ma; Patients recruitment and treatment were performed by Yufeng Wu, Zhen He, Lili Wang, Yang Liu and Qiming Wang; Qiming Wang and Yang Liu contributed to the study conception and design. All authors read and approved the manuscript.

Funding

This research project was supported by the following funding sources: 1、Henan Province Health and Youth Subject Leader Training Project, [2020]60; 2. Leading Talent Cultivation Project of Henan Health Science and Technology Innovation Talents, YXKC2020009; 3. ZHONGYUAN QIANREN JIHUA, ZYQR201912118; 4. Henan International Joint Laboratory of drug resistance and reversal of targeted therapy for lung cancer, [2021]10; 5. Henan Medical Key Laboratory of Refractory lung cancer, [2020]27; 6. Henan Refractory Lung Cancer Drug Treatment Engineering Technology Research Center, [2020] 4; 7. Qiming Wang–Key Research and Development Projects of Henan Province in 2023–Key technologies of novel precision immunotherapy for refractory malignant tumors (Project No. 231111313300); 8. Henan Youth and Middle-aged Health Science and Technology Innovation Leaders Training Project (No. YXKC2022004); 9. Henan Health Young and Middle-aged discipline Leader Project (No. HNSWJW-2022011); 10. Henan Medical Science and Technology Tackling Program of Provincial-Ministry Joint Major Project (No. SBGJ202401004); 11.Key Research and Development Projects of Henan Province (No. 251111310100).

Data availability

The datasets used and analyzed during the current study are not publicly available due to the data of this study involve the privacy of patients but can be obtained from the corresponding authors according to reasonable requirements.

Declarations

Ethics approval and consent to participate

The study was conducted in accordance with the principles of the Declaration of Helsinki, and the study protocol was approved by the Institute Review Board of the Cancer Hospital Affiliated to Zhengzhou University. Because of the retrospective nature of the study without any intervention, and the data were anonymous, the requirement for patient consent was waived.

Competing interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.