Comprehensive genomic and clinical analyses identify APOBEC mutational signatures as a brain metastasis risk factor in lung adenocarcinoma patients

Highlights

• •APOBEC mutational signatures- new brain metastasis risk factor in lung adenocarcinoma.
• •Age≤60 years and EGFR mutation screened as the other 2 brain metastasis risk factors.
• •Metastasis-related pathways upregulated in the patients with high APOBEC signatures.

Abstract

Background

Lung adenocarcinoma is the most common source of brain metastasis (BM), resulting in significant morbidity and mortality. We aimed to identify patients with high BM risk who possibly benefit from brain-penetrant drugs, prophylactic cranial irradiation, or close brain magnetic resonance imaging surveillance.

Methods

Metastatic lung adenocarcinoma patients with extracranial tumor samples profiled by a next-generation sequencing panel targeting 425 tumor-related genes were retrospectively enrolled between February 2008 and July 2021. We compared BM and non-BM patients’ genomic and clinical features and studied their associations with BM risk. Two external cohorts were used for result validation and molecular mechanisms investigation, respectively.

Results

We included 174 eligible patients, including 90 having developed BM by the end of follow-up. Age≤60, EGFR activating mutations, and high-level apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like (APOBEC) mutational signatures were associated with elevated BM risk. Similar findings in BM-free survival were obtained by fitting Fine-Gray subdistribution hazard models addressing competing risks. Increased BM risk related to APOBEC mutational signatures was validated in an external cohort (N = 440). RNA sequencing data analyses performed in another external cohort (N = 230) revealed that expressions of metastasis-related pathways such as transforming growth factor (TGF)β and epithelial-mesenchymal transition (EMT) were upregulated in the patients with high-level APOBEC mutational signatures.

Conclusion

APOBEC mutational signatures related to upregulated TGFβ and EMT, could serve as an independent risk factor for BM and BM-free survival in metastatic lung adenocarcinoma patients. Further investigations are warranted to tailor personalized treatments to improve the susceptible patient's outcomes.

Introduction

Lung cancer is the most diagnosed cancer type and the leading cause of cancer death worldwide, and most lung carcinomas are adenocarcinomas. The brain metastasis (BM) incidence of metastatic lung adenocarcinoma is as high as 57 % [1], and BM is adults’ most common intracranial tumor, resulting in significant morbidity and mortality. Lung adenocarcinoma is its most common source, constituting 37 % of all BM cases [2]. Overall survival (OS) of lung cancer patients decreases fairly following BM, with a median OS of 7 months [3]. Drugs with poor penetration through the blood-brain barrier have a limited impact on BM, leaving the brain relatively undertreated. The outcome of patients with BM has improved considerably owing to more drugs with demonstrated activity in the brain. First-line osimertinib reduced the risk of brain progression [4] and improved OS [5] versus first-generation epidermal growth factor receptor (EGFR) tyrosine kinase inhibitors among patients with untreated EGFR-mutated advanced non-small cell lung cancer (NSCLC). Therefore, new proper BM biomarkers are able to help precisely identify patients most likely to gain survival benefits from first-line osimertinib.

Effective drugs invariably encounter tumor resistance. Consequently, oncologists often face drug-insensitive or heavily treated BM. Prophylactic cranial irradiation (PCI) has long been standard care in patients with small-cell lung cancer; however, the role of PCI in NSCLCs is less established due to limited data regarding OS benefits, even though a consistent decrease in the BM risk has been observed [6,7]. In a randomized trial of locally-advanced NSCLC, PCI decreased the BM rates and prolonged disease-free survival but did not improve OS compared with observation alone; whereas within the nonsurgical arm, PCI effectively improved OS and disease-free survival [7]. In another randomized study including 71 Stage IV and ten Stage III NSCLCs without baseline BM and with EGFR mutations, ALK rearrangements, or elevated carcinoma embryonic antigen, PCI significantly decreased the BM incidence and increased progression-free survival and OS [8]. These studies implied that the NSCLC subgroup at high BM risk might gain OS benefits from PCI.

We assembled a cohort of 174 metastatic lung adenocarcinoma patients. Their extracranial tumor samples were profiled using a targeted next-generation sequencing (NGS) panel covering 425 cancer-related genes, which identified somatic mutations, copy number variations (CNVs), tumor mutational burden (TMB), and intratumor heterogeneity (ITH). We comprehensively analyzed the genomic and clinical differences between patients with and without BM to better understand BM biology and distinguish the high-risk subpopulation that might benefit from first-line brain-penetrant drugs, PCI, or close brain magnetic resonance imaging surveillance.

Materials and methods

Patient inclusion and sample collection

This study retrospectively included patients who were initially diagnosed as metastatic lung adenocarcinoma or had ultimately developed metastasis at our institution from February 2008 to July 2021. To be eligible for inclusion, patients should be 18 or older and have good tissue samples of extracranial tumors passing the quality control process for NGS. Formalin-fixed paraffin-embedded (FFPE) or fresh tumor tissues were collected with pathological type, and two pathologists independently assessed the tumor purity. The leukocyte from peripheral blood was used as germline control. Patients without detectable somatic mutations/CNVs, with synchronous primary tumors, with incomplete clinical data to define distant metastases, or missing NGS data were excluded from the study. Medical records and follow-up data were reviewed to obtain clinical information. The stage was classified according to the 8th edition of the TNM classification for NSCLC [9]. The research was carried out according to the principles set out in the Declaration of Helsinki 1964 and all subsequent revisions. The ethics committee of our institution approved the protocol of this study (No. 2017015). Informed consent was obtained from each patient. All patients had brain magnetic resonance imaging with contrast as part of their initial staging. Tumor assessments were performed at initial diagnosis of metastatic disease, every 2 months for 2 years, then every 3 months until disease progression during treatment.

Patient follow-up

After treatment patients were followed up every 3–6 months for 3 years, then twice a year for the following 2 years, and yearly thereafter. All patients were evaluated with a physical examination, complete blood count, serum biochemistry, thoracic CT scans, abdomen B-ultrasound examination, and other necessary examinations based on the patient's symptoms. Follow-up brain magnetic resonance imaging with contrast was performed upon the development of suspicious symptoms or as part of restaging at the time of disease progression or every 3–6 months for 3 years, then twice a year for the following 2 years, and yearly thereafter. Sites of disease progression were determined either radiologically or histologically. Both the initial and subsequent sites of tumors were documented.

DNA sequencing and library construction

Archived FFPE tumor blocks/slides or frozen fresh tissues were used for tumor genotyping. Paired white blood cells prepared from peripheral blood were also sequenced and used to eliminate germline mutations. In brief, Genomic DNA from tissue and blood samples was extracted using QIAamp DNA FFPE Tissue Kit and DNeasy Blood & Tissue Kit (Qiagen), respectively, according to manufacturer instructions. The concentration and quality of extracted DNA were assessed with Qubit 3.0 Fluorometer and Nanodrop 2000 (Thermo Fisher Scientific). After fragmentation, 1 μg genomic DNA samples underwent endrepairing, A-tailing and ligation with indexed adaptors to prepare sequencing libraries using the KAPA Hyper Prep kit (KAPA Biosystems). Libraries were then amplified and purified using Agencourt AMPure XP beads (Beckman Coulter) before target enrichment. XGen Lockdown Hybridization and Wash Reagents Kit (Integrated DNA Technologies) was used for the target enrichment of 425 cancer-related genes (GeneseeqPrime™, Nanjing Geneseeq Technology Inc.). The library was captured by Dynabeads M-270 (Life Technologies), amplified in KAPA HiFi HotStart ReadyMix (KAPA Biosystems), and quantified by qPCR using the KAPA Library quantitative Kit (KAPA Biosystems). The final library was sequenced using the Illumina Hiseq 4000 platform with 2 × 150 bp pair-end reads. The sequencing coverage and quality statistics of each sample were summarized in Table S1.

Mutation calling and analysis of sequencing data

FASTQ file quality filtering was performed with Trimmomatic [10] to remove low-quality bases (<20) or N bases. The measured sequences were aligned to the human reference genome hg19 using Burrows-Wheller Aligner (BWA-mem, v0.7.12) [11] and deduplicated using Sambamba [12]. Local realignment around base indels and recalibration of base quality scores were performed using the Genome Analysis Toolkit (GATK 3.4.0) [13]. Cross-sample contamination was estimated using ContEst (Broad Institute). VarScan2 was then used to detect single nucleotide variants (SNVs) and insertion/deletion mutations with the following removed: (i) germline mutations detected by white blood cell controls, (ii) previously reported list of clonal hematopoietic mutations [14], and (iii) population frequency>1 % in ExAC [15], gnomAD [16] and 1000 G [17] databases. All somatic variants were filtered: 1) for hotspot mutations in the catalog of somatic mutations in cancer v92 (defined by prevalence≥20) [18], the number of supported reads≥3, the depth≥30x, and variant allele frequency (VAF)≥0.5 %; 2) for non-hotspot mutations, the number of supported reads≥6, the depth≥30x, and the VAF≥1 %. Gene fusions were identified by FACTERA [19] with default parameters and manually inspected on Integrative Genomics Viewer. CNV was detected using the default parameters of CNVkit [20]. Somatic CNVs were identified by paired tumor and normal samples with thresholds of normalized log2 depth ratio at below 0.65 and above 2 for copy number loss and gains, respectively. Chromosomal instability score (CIS) was defined as the proportion of segments with copy number alterations (log₂ depth ratio>0.2 or <−0.2) across 22 autosomal arms. TMB of the 425-gene panel was calculated as previously reported [21]. Mutational signatures were extracted using the R package sigminer (v 2.1.4) based on synonymous and nonsynonymous single base pair substitutions using Cosmic version 2 as reference signatures and then categorized into ten groups according to their etiology [22,23] (Table S2).

To define ITH, nonsynonymous SNVs were first classified as clonal and subclonal mutations based on comparing to the highest VAF in the sample as previously described [24]. In brief, if the VAF of a certain locus was more than half of the maximum VAF in a sample, it was considered as ‘clonal’ and otherwise as ‘subclonal’. ITH was then calculated by dividing the number of all subclonal mutations by the total number of mutations for a given sample. Continuous variables, including the proportion of mutational signatures contribution and ITH, were converted into binary variables (high and low groups) using a cutoff at the top one-third percentile.

Result validation and differential gene expression in external cohorts

To validate the main findings in our study cohort, we used an external lung adenocarcinoma cohort (N = 440) [25]. Mutational signatures extraction and high/low apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like (APOBEC) classification were performed as previously mentioned. We further conducted differential gene expression analysis in another external lung adenocarcinoma cohort (N = 230) from The Cancer Genome Atlas (TCGA) Firehose Legacy. The RNA expression matrix and genetic mutation data were obtained from the cBioPortal database (https://www.cbioportal.org/study/summary?id=luad_tcga). Differential expression analysis was performed using the R package DESeq2 based on negative binomial distribution, and differences with |log2FC|>1 and false discovery rate adjusted P-value (P.adjust) ≤0.05 were considered statistically significant. Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis was conducted using the R package clusterProfiler (v4.2.2) with a P.adjust cut-off of 0.1. Gene set enrichment analysis (GSEA) was also performed using the R package clusterProfiler (v4.2.2), and the gene sets “h.all.v7.4.symbols.gmt” were selected as the reference gene set. The enrichment of a pathway with P.adjust ≤0.1 and |normalized enrichment score|>1 was defined as statistically significant.

Statistical analysis

BM-free survival was defined as the time from cancer metastasis diagnosis to BM occurrence, or death/the last follow-up for patients without BM. Covariates mainly included clinical characteristics (such as patient age, sex, smoking history, ECOG performance status, hypertension, coronary disease, diabetes, pulmonary tuberculosis, cerebral infarction, atrial fibrillation, and serum carcinoembryonic antigen level), genetic mutations, and mutational signatures. Only genes altered in more than 5 % of the study cohort were included in analyses. The frequency of categorical variables were compared using Fisher's exact test, and medians of continuous variables were compared using the Mann-Whitney Wilcoxon Test. Cumulative incidence functions (CIFs) were used to estimate the incidence of BM when competing risk of death was taken into consideration, and Fine-Gray subdistribution hazard models were fitted to estimate hazard ratios (HRs) combined with 95 % confidence intervals (CIs), using R packages cmprsk [26]. The proportional hazard assumption was tested using log(-log) survival plots and the cox.zph function. Covariates representing statistically significant associations with outcomes were then included in multivariable analyses. All analyses were performed in R (version 3.6.3). A two-sided P < 0.05 is required to consider statistical significance.

Results

Patient overview

Of 208 consecutive patients with metastatic lung adenocarcinomas, eight patients were excluded due to incomplete clinical data. Fourteen patients without detectable genetic alterations, ten patients using 14 or 139-gene NGS panels, one patient whose sequencing data were from BM, and one patient with synchronous primary tumors were also excluded. Finally, the remaining174 eligible patients were included in this study. The median follow-up time for the 23 surviving patients without BM was 49.5 months (range, 14.0–128.0 months). By the end of follow-up (December 2022), 90 patients (51.7 %) had developed BM, with the remaining patients developing extracranial metastases. Patients’ clinical characteristics are summarized in Table 1. There were significantly more patients aged ≤60 years in the BM than in the non-BM group (61.1% vs. 45.2 %, P = 0.048). However, significant difference in the level of serum carcinoembryonic antigen was not observed between the patients with and without BM (Figure S1). The majority of the tumor samples were collected from lungs (67.8 %), followed by lymph nodes (27.0 %) and livers (3.4 %).

Table 1.

Patient characteristics.

Characteristics	Total (N = 174)	Non-BM (N = 84)	BM (N = 90)	P
Age, No. (%)				0.05*
≤60 years	93 (53.4)	38 (45.2)	55(61.1)
>60 years	81 (46.6)	46 (54.8)	35 (38.9)
Sex, No. (%)				0.07
Male	84 (48.3)	47 (56.0)	37 (41.1)
Female	90 (51.7)	37 (44.0)	53 (58.9)
Smoking history, No. (%)				0.06
Never smoker	104 (59.8)	44 (52.4)	60 (66.7)
Former/current smoker	70 (40.2)	40 (47.6)	30 (33.3)
ECOG performance status, No. (%)				0.23
0	84 (48.3)	45 (53.6)	39 (43.3)
1	90 (51.7)	39 (46.4)	51 (56.7)
Hypertension, No. (%)				0.37
with	40 (23.0)	22 (26.2)	18 (20.0)
without	134 (77.0)	62 (73.8)	72 (80.0)
Coronary disease, No. (%)				0.15
with	13 (7.5)	9 (10.7)	4 (4.4)
without	161 (92.5)	75 (89.3)	86 (95.6)
Diabetes, No. (%)				0.32
with	9 (5.2)	6 (7.1)	3 (3.3)
without	165 (94.8)	78 (92.9)	87 (96.7)
Pulmonary tuberculosis, No. (%)				>0.99
with	4 (2.3)	2 (2.4)	2 (2.2)
without	170 (97.7)	82 (97.6)	88 (97.8)
Cerebral infarction, No. (%)				0.06
with	7 (4.0)	6 (7.1)	1 (1.1)
without	167 (96.0)	78 (92.9)	89 (98.9)
Atrial fibrillation, No. (%)				0.61
with	3 (1.7)	2 (2.4)	1 (1.1)
without	171 (98.3)	82 (97.6)	89 (98.9)

BM: brain metastasis, *: statistically significant.

Mutational landscape and signatures

A total of 1975 somatic mutations were detected in the 174 tissue samples, including 1735 non-synonymous mutations. Mutated EGFR genes were the most prevalent (63.2 %), followed by TP53 (60.9 %), KRAS (15.5 %), and FAT1 (11.5 %) (Fig. 1A). Mutated genes were mainly enriched in canonical RTK-RAS, P53, and PI3K oncogenic pathways (Fig. 1B). The majority of mutations were missense mutations (70.5 %), and 7.0 % were nonsense mutations (Fig. 1C). Of all somatic mutations, 89.7 % were single nucleotide variants (Fig. 1D), including 37.5 % base transversions (C > A, C > G, T > A, or T > G) and 62.5 % transitions (Fig. 1E). The age-related base transition C > T (48.9 %) and smoking-associated base transversions C > A (15.5 %) accounted for relatively high proportions (Fig. 1E). The median TMB and CIS of 174 samples were 5.18 (range: 0–75.09, Fig. 1F) and 0.250 (range: 0.007–0.893, Fig. 1G), respectively. The median level of ITH was 0.5 (range: 0–0.99, Fig. 1H). The exposure and classification of mutational signatures of each sample were presented in Fig. 2.

Fig. 1.

Mutational features of included patients.

(A) Frequently mutated genes of included patients with and without BM. (B) Mutated pathways of included patients with and without BM. (C) (D) Proportions of different variant classifications. (E) The base mutation patterns. (F) Distribution of variation number per sample. (G) (H) Distribution of CIS, and clone and subclone numbers. BM: brain metastasis; CIS: chromosomal instability score; SNV: single nucleotide variant.

Fig. 2.

Landscape of mutational signatures of included patients.

(A) Box plots showing the exposure of mutational signatures corresponding to the mutations identified in patients. (B) The proportion of mutational signature exposure in each patient. APOBEC: apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like; BRCA: breast cancer gene; MMR: mismatch repair; POLE: polymerase epsilon.

APOBEC signature indicates increased BM risk

By comparing mutated genes between the patients with and without BM, our data revealed a strong association of EGFR aberrations with BM (75.6% vs. 50.0 %, P = 0.001, Table S3), especially for EGFR activating mutations, including exon 19 deletion, L858R, G719X, S768I, L861Q, T790M, and exon 20 insertion (73.3% vs. 47.6 %, P = 0.001, Table S3). As expected, mutated RTK-RAS signaling pathway, which involves the EGFR gene was associated with increased risk of BM (97.8% vs. 89.3 %, P = 0.03, Table S4), whereas other cancer relevant signaling pathways were not strongly related to BM risk. Moreover, patients with EGFR activating mutations exhibited poorer BM-free survival than patients without when death due to any causes were defined as the competing risk (P < 0.001, HR: 2.13, 95 % CI: 1.37–3.31, Figure S2). Notably, there was no difference in BM-free survivals between the patients with EGFR exon 19 deletion and L858R.

Next, we evaluated whether biomarkers for genome-wide alterations were able to aid in identify lung cancer patients at relatively high BM risk. Significantly higher exposure of APOBEC mutational signature was observed in the BM than in the non-BM group (P = 0.01, Fig. 3A); however, there was no significant difference in the level of ITH between two groups (P = 0.12, Fig. 3B). When APOBEC mutational signatures and ITH were classified as the high and low groups and further assessed binarily, samples with high APOBEC mutational signatures (P = 0.02) or high ITH (P = 0.04) were more prevalent in BM patients than in non-BM patients (Fig. 3C). Furthermore, CIFs demonstrated that worse BM-free survival was associated with high-level ITH (P = 0.008, HR: 1.67, 95 % CI:1.13–2.47, Fig. 3D) and high-level APOBEC mutational signature (P = 0.02, HR: 1.61, 95 % CI:1.10–2.36, Fig. 3E). By contrast, CIS and TMB levels were comparable between the BM and non-BM groups (Fig. S3).

Fig. 3.

Molecular features associated with BM risk and BM-free survival.

(A) The differences in mutational signatures between the BM and non-BM groups. (B) Comparison of ITH between the BM and non-BM groups. (C) Proportions of high-level APOBEC mutational signatures and high-level ITH in the BM and non-BM groups. (D) BM-free survivals in the patients with low and high-level ITH. (E) BM-free survivals in the patients with low and high-level APOBEC mutational signatures. APOBEC: apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like; BM: brain metastasis; BRCA: breast cancer gene; ITH: intratumor heterogeneity; MMR: mismatch repair; POLE: polymerase epsilon.

In addition to the APOBEC mutational signature and EGFR activating mutation, several molecular features were also associated with worse BM-free survival, including EGFR CNV amplification, mammalian target of rapamycin (mTOR) variations, RB1 mutations, and TP53 nonsense mutations (Table S5). Moreover, the associations of BM risk with clinical characteristics and comorbidities were investigated. Our data demonstrated that patients >60 years exhibited superior BM-free survival, whereas female patients might have inferior BM-free survival (Table S5). Common comorbidities, including hypertension, coronary disease, diabetes, pulmonary tuberculosis, cerebral infarction, and atrial fibrillation, were not strongly associated with BM-free survival (Table S6).

The results of the multivariable logistic regression including age, sex, smoking history, EGFR activating mutations, APOBEC mutational signatures, and ITH levels revealed that age≤60 years, EGFR activating mutations and high-level APOBEC mutational signatures were three independent biomarkers aiding in BM risk prediction (Table 2). Consistently, the multivariable Fine-Gray subdistribution hazard model including all clinical and molecular features with significant association with BM-free survival in univariate analyses revealed that age≤60 years, EGFR activating mutations and high-level APOBEC mutational signatures were related to worse BM-free survivals (Table 3), and the proportional hazard assumption worked well in the fitted Fine-Gray subdistribution hazard model (Table S7).

Table 2.

Molecular and clinical characteristics associated with BM ris.

Characteristics	Univariate logistic regression		Multivariable logistic regression
	OR (95 % CI)	P	OR (95 % CI)	P
Age, y
>60 vs. ≤60	0.53 (0.27–1.00)	0.05*	0.50 (0.26–0.96)	0.04*
Sex
Female vs. Male	1.81 (0.96–3.47)	0.07	1.48 (0.59–3.70)	0.41
Smoking history
Former/current smoker vs. Never smoker	0.55 (0.28–1.06)	0.06	0.91 (0.35–2.33)	0.84
EGFR activating mutations
With vs. Without	3.00 (1.53–6.00)	<0.01*	2.40 (1.21–4.78)	0.01*
APOBEC mutational signatures
High vs. Low	2.21 (1.09–4.54)	0.02*	2.17 (1.08–4.36)	0.03*
ITH
High vs. Low	1.99 (1.00–4.05)	0.04*	1.93 (0.96–3.90)	0.07

BM: brain metastasis, OR: odds ratio, CI: confidence interval, ITH: intratumor heterogeneity, *: statistically significant.

Table 3.

Molecular and clinical characteristics associated with BM-free survival.

Characteristics	Univariate analysis		Multivariable analysis
	HR (95 %CI)	P	HR (95 %CI)	P
APOBEC High vs. Low	1.61 (1.10–2.36)	0.02*	1.66 (1.12–2.46)	0.01*
EGFR activating mutations vs. wild-type	2.13 (1.37–3.31)	<0.01*	1.77 (1.11–2.81)	0.02*
ITH High vs. Low	1.67 (1.13–2.47)	<0.01*	1.45 (0.95–2.21)	0.08
EGFR amplification vs. wild-type	2.10 (1.08–4.11)	<0.01*	1.80 (0.93–3.49)	0.08
RB1 mutations vs. wild-type	1.99 (1.08–3.68)	0.01*	1.63 (0.84–3.19)	0.15
MTOR aberrations vs. wild-type	2.10 (1.13–3.87)	0.03*	1.96 (1.06–3.65)	0.03*
TP53 nonsense vs. wild-type	1.65 (1.01–2.69)	0.04*	1.67 (1.03–2.70)	0.04*
Age >60 vs. ≤60	0.65 (0.44–0.97)	0.04*	0.65 (0.44–0.97)	0.04*
Sex Female vs. Male	1.47 (0.99–2.17)	0.04*	1.26 (0.84–1.90)	0.26

BM: brain metastasis, HR: hazard ratio, CI: confidence interval, ITH: intratumor heterogeneity, *: statistically significant.

Verification of BM-enriched APOBEC signatures and differentially expressed genes

An external lung adenocarcinoma cohort consisting of 143 BM and 297 non-BM patients was used to confirm the correlation between high APOBEC mutational signatures and BM. The exposure of APOBEC mutational signatures was significantly elevated in primary lung adenocarcinomas from patients who had developed BM compared to those who had not (P = 0.048, Fig. 4A), whereas no obvious differences in the exposure of mutational signatures with other etiology were observed. The proportion of patients with high-level APOBEC mutational signatures was also higher in patients with BM than non-BM (39.9% vs. 29.6 %, P = 0.040, Fig. 4B), consistent with our findings. Subsequently, the differential gene expression of mRNA performed in the TCGA Firehose Legacy cohort showed 319 upregulated genes and 167 downregulated genes in the 76 patients with high-level APOBEC mutational signatures, compared to the 154 patients with low-level APOBEC signatures (Fig. 4C). KEGG pathway analysis demonstrated that the upregulated genes were mainly enriched in transforming growth factor (TGF)β and metabolism-related pathways, while we did not observe an enrichment pathway with the downregulated genes (Fig. 4D). Additionally, GSEA enrichment analysis revealed that epithelial-mesenchymal transition (EMT), TGFβ, and tumor necrosis factor alpha-nuclear factor kappa light chain enhancer of activated B cells pathway expressions were upregulated in the patients exhibiting high-level APOBEC signatures (Fig. 4E).

Fig. 4.

Results validation and molecular mechanism investigation in external datasets.

The differences in mutational signatures between the BM and non-BM groups in the external cohort. (B) Proportions of patients with high-level APOBEC mutational signatures in the BM and non-BM groups in the external cohort. (C) Differential gene expression of mRNA between the patients with high and low-level APOBEC mutational signatures in the external cohort. (D) Representative enriched KEGG pathways of differentially expressed genes between the patients with high and low APOBEC mutational signatures. The pathway with false discovery rate-adjusted P ≤ 0.1 was considered significant. (E) GSEA pathway analysis using hallmark gene sets. The gene set under the pathway of |normalized enrichment score|>1 and false discovery rate-adjusted P ≤ 0.1 was considered significant between the two groups. APOBEC: apolipoprotein B mRNA editing enzyme, catalytic polypeptide-like; BM: brain metastasis; BRCA: breast cancer gene; GSEA: Gene set enrichment analysis; KEGG: Kyoto Encyclopedia of Genes and Genomes; MMR: mismatch repair; TNFA-NFKB: tumor necrosis factor alpha-nuclear factor kappa light chain enhancer of activated B cells; POLE: polymerase epsilon; TGF: transforming growth factor.

Discussion

Lung adenocarcinoma has become the most common histology type of lung cancer and the most common source of BM, whereas the mechanism driving BM is poorly understood. We analyzed 174 metastatic lung adenocarcinoma patients in our institution, comprehensively compared the genomic and clinical differences between the BM and non-BM groups. Our data demonstrated that age≤60 years, EGFR activating mutations and high-level APOBEC mutational signatures were associated with increased risk of developing BM and worse BM-free survival.

Some studies uncovered the mechanism of lung adenocarcinoma BM by comparing the patients’ matched primary lung tumor and BM. Mutations of TP53, EGFR, KRAS, and ALK genes were reported to be highly concordant between primary NSCLC and matched BM, whereas discordance of PI3K signaling suggested the unique genomic evolution and oncogenic mechanisms of BM [27]. Jiang et al. also reported that TP53 and EGFR alterations were highly consistent between paired primary and metastatic tumors [28]. However, in another study, the comparison of BM and primary lung adenocarcinoma revealed TP53, ATR, and APC mutations in BM with increased frequency [29]. Inconsistent mutation status of KRAS has been discovered between the primary and metastatic sites of lung adenocarcinoma, while EGFR mutation relatively consistent on the contrary [30]. The rationale is that the brain-tropic clone may be minor in the primary tumor, which becomes enriched in the brain. Besides, Nguyen et al.’s analysis of genomic alterations from unpaired primary and metastatic samples revealed that lung adenocarcinoma patients with BM had a higher frequency of TP53 mutations, TERT CNV amplifications, and EGFR mutations but a lower frequency of RBM10 mutations [31]. Whereas Shih et al. found that the prevalence of MYC, YAP1, and MMP13 CNV amplifications was elevated in BM using case-control analyses [32].

Other researchers explored the mechanism of lung cancer BM by analyzing the extracranial tumor. A recent study built a risk signature based on age≤ 65 years, high expressions of Nestin and Methylmalonate-semialdehyde dehydrogenase for predicting BM in lung adenocarcinoma patients [33]. An immune-related gene expression signature was reported to predict BM in lung adenocarcinoma patients after surgery, so it is expected that tumor cells have a greater influence on gene expression than tumor stroma/microenvironment and that certain genes are highly expressed in the early stages of cancer, leading to BM [34]. Lin et al. unveiled that the mRNA expression profile, including CDKN2A, significantly differed between BM and non-BM lung adenocarcinomas after primary lung tumor resections, but the targeted tumor DNA sequencing did not disclose significant differences between the groups [35]. A five-gene (TYMS, CDK1, HJURP, CEP55, and KIF11) predictive model exhibited potential clinical utility for predicting postoperative BM [36]. The factors obtained from this approach may be more useful for risk stratification of patients before BM occurs since detecting the minor clone in the extracranial tumor may be difficult. Comparing extracranial tumors from metastatic patients with BM and without BM helps identify factors related explicitly to brain tropism.

Notably, we detected high-level APOBEC mutational signatures as an independent BM risk factor in metastatic lung adenocarcinoma patients, which was validated in an external lung adenocarcinoma cohort. In the research where our validation cohort came from, Lengel et al. observed a higher prevalence of APOBEC mutational signatures in primary lung adenocarcinomas from patients who had developed metastases than those who had not, and the APOBEC mutational signatures were more prevalent among metastases than primary lung adenocarcinomas [25]. A recent study compared 3035 NSCLCs with BM to an independent cohort of 7277 primary NSCLC (pNSCLC) specimens and found that immune checkpoint inhibitors-positive predictive biomarkers of TMB and APOBEC mutational signatures were significantly higher in NSCLCs with BM compared to the pNSCLC cohort, and they were both also enriched in the BM adenocarcinoma only cohort, while EGFR exon19 deletions and EGFR^L858R mutations were significantly decreased in adenocarcinomas with BM (N = 2110) when compared to the pNSCLC with adenocarcinoma histology (N = 4396) [37]. Roper et al. performed integrated exome, transcriptome, and quantitative proteomics analyses of metastases obtained through rapid autopsy of patients with thoracic malignancies and discovered APOBEC mutagenesis, driven by germline variants, mutant TP53, and the tumor microenvironment, and late copy number alterations as key mechanisms underlying proteogenomic evolution and heterogeneity [38]. An analysis of 1001 human cancer cell lines and 577 xenografts clarified that mutagenesis associated with the cytidine deaminase APOBEC occurred in episodic bursts in contrast to mutation signatures associated with DNA replication and repair [39]. Using 508 esophageal squamous cell carcinoma genomes, mutational signature clusters associated with metastasis and patients’ outcomes were uncovered, implying APOBEC activity may continuously contribute to mutagenesis during tumor progression [40]. Our results from RNA analyses of another external cohort revealed that the expression levels of genes in multiple pathways were significantly upregulated in the patients with high APOBEC mutational signatures, particularly in EMT and TGFβ, suggesting that lung adenocarcinomas with high APOBEC mutational signatures might have stronger tumor cell migration and metastasis abilities.

Our results illustrated that EGFR mutations was an independent BM risk factor in metastatic lung adenocarcinoma patients, consistent with previous studies with Caucasian or Asian populations [1,41]. Meantime, we detected younger age as an independent BM risk factor in metastatic lung adenocarcinoma patients, consistent with previous studies [33,42]. It has rarely been reported that mTOR variations are associated with BM. The effect of TP53 nonsense on BM remains unclear.

This study has a few limitations that need to be considered. First, although we analyzed the differences between patients with and without BM within a cohort having a relatively large sample size, and comprehensive genomic and clinical data, the timing of the disease progression assessment was not standardized due to the nature of a retrospective real-world cohort study. Second, our main finding about the association between APOBEC mutational signatures and BM-free survival was not validated due to the lack of mutational signatures or survival data in an external cohort. The mechanism of BM is a complicated spatial and temporal network. Some risk factors may prevail at the early stage, some may be minor at the early stage and develop later, and others may not exist at the early stage and come along the course of BM.

Conclusions

Comprehensive genomic and clinical analyses identified high-level APOBEC mutational signatures as a novel independent BM risk factor in metastatic lung adenocarcinoma patients, and APOBEC mutational signatures might be related to upregulated TGFβ and EMT. Our findings help illuminate the mechanism of BM and identify the susceptible subgroup. Further investigations are warranted to tailor personalized treatments and follow-up to improve the subgroup's outcomes.

CRediT authorship contribution statement

Qiang Li: Writing – review & editing, Investigation. Meng Jiang: Writing – review & editing, Writing – original draft, Formal analysis. Shiqiang Hong: Writing – review & editing, Investigation. Jing Yang: Writing – review & editing, Investigation. Xiaoying Wu: Writing – review & editing, Formal analysis. Jiaohui Pang: Writing – review & editing, Formal analysis. Yedan Chen: Writing – review & editing, Writing – original draft. Xiaotian Zhao: Writing – review & editing, Writing – original draft. Xiao Ding: Writing – review & editing, Writing – original draft, Supervision, Funding acquisition, Conceptualization.

Declaration of competing interest

Xiaoying Wu, Jiaohui Pang, Yedan Chen, and Xiaotian Zhao are employees of Nanjing Geneseeq Technology Inc. All remaining 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.

Data availability

• The NGS data was deposited in GSA-HUMAN HRA003500SRA. Any additional information will be available upon reasonable request.