Different next-generation sequencing pipelines based detection of tumor DNA in cerebrospinal fluid of lung adenocarcinoma cancer patients with leptomeningeal metastases

Lung cancer is one of the most common malignant tumor in the world. Leptomeningeal metastasis which predicts poor prognosis occurs in about 5% lung cancer patients [1]. Genomic characterization of tumor is crucial for treatments. Specifically, more than half lung cancer patients were detected to have acquired EGFR T790 M mutation after they were treated with Gefitinib or Erlotinib [2]. It has become a consensus that lung cancer patients with tumor progression should carry out a re-biopsy to adjust the treatment [3]. However, most lung cancer patients with leptomeningeal metastases cannot get leptomeningeal tumor tissue for DNA mutation testing [4]. Liquid biopsy based on the detection of circulating tumor DNA in plasma also played a limited role for patients with intracranial lesions [5, 6].

It was reported that tumor-associated DNA could be detected in the CSF samples of cancer patients with central nervous system (CNS) tumor using NGS sequencing [6, 7]. Previous studies reported that DNA in brain tumors showed more common mutations with those in CSF when compared to plasma DNA in cancer patients [5], and CSF cfDNA could reveal the unique genetic profiles of leptomeningeal metastases in EGFR-mutated non-small cell lung cancer (NSCLC) [8]. These findings suggest that CSF is a potential source of tumor-derived DNA in lung cancer patients with leptomeningeal metastases. Jiang et al. [9] showed that tumor mutations in leptomeningeal metastases of NSCLC could be detected from CSF tumor cells, which demonstrated that we could get the tumor-related genetic information from both supernatant and cell pellets of the CSF samples of lung cancer patients with leptomeningeal metastases.

In this study, different detection pipelines were utilized for CSF cfDNA, cells, plasma and FFPE samples of primary lung cancer to identify sequence mutations. It aimed to find out whether CSF as liquid biopsy samples in a clinical setting can be substantiated and can be considered as an excellent substitute to primary cancer tissue.

In this study we used various DNA mutation detection pipelines for 4 types of biospecimens, and successfully generated NGS libraries from plasma (1~5.5 ml), CSF cfDNA (2.5~12 ml), CSF cells (0.2~2.6 ml) and FFPE sample. The input DNA for cHOPE pipeline was as low as about 5 ng. Both cHOPE and OncoAim pipelines utilized multiple PCR as enrich technology to detect tumor-associated hotspots. However, they were designed for different sample types, of which cHOPE was specialized for cfDNA while OncoAim was for DNA from FFPE sample. The other 2 NGS pipelines, ddCAP and ddCAP-on-Tissue, used target capture as enrichment technology to scan all the exons of 10 tumor-associated genes. The input DNA of ddCAP pipeline for cfDNA could be as low as 5 ng as well, whereas 50 ng DNA was required for the ddCAP-on-Tissue pipeline for DNA from FFPE sample.

In most cases that had discrepancies between CSF and plasma, CSF had more mutations detected [5, 8], which was confirmed in our study. There are several possible reasons: Firstly, because CSF circulates through the CNS and has a large interface with the CNS, it has a strong potential to carry circulating tumor DNA (ctDNA) of CNS metastases. Secondly, because there is blood-brain-barrier, DNA cannot travel between blood and CSF freely, which leads to discrepancies in DNA species in CSF and blood. Thirdly, because our cohort all had both intracranial and extracranial metastasis, spatial heterogeneity in primary lesions and CNS metastases may contribute to the discrepancies too.

Of the 19 patients who provided both plasma and CSF samples, 13 of them had mutations identified in CSF samples (cfDNA, cell pellets or both). Meanwhile, only 4 out of the 19 plasma samples had mutations identified, all of which were also discovered in the matched CSF samples. Previous research indicated that positive detection of EGFR was more sensitive in CSF than in the matched plasma [8, 10]. Our results indicate that NGS detection of CSF samples could provide more accurate and credible mutation information than plasma samples, which were consistent with the previous conclusion [10].

It was reported that more mutations could be detected in CSF cfDNA than in the paired CSF cells [8]. Due to the tumor heterogeneity and the various analysis settings in different pipelines, we did not observe same gene alternation for all the genes detected, particularly for some low AF mutations (such as ALK G689R with an AF of 1.0% and of 1.2%, and KRAS Q61L with an AF of 0.1%) and rare mutations (like FGFR1 R455H, which is not reported in COSMIC). However, similar detection rates were obtained by the sequencing for CSF cfDNA and cell samples, and the largest number of mutations were detected in those two types of sample. Especially for the EGFR gene, same mutations were identified in the cfDNA and cells of CSF sample, demonstrating the potential of using CSF cells to predict the response of EGFR-tyrosine kinase inhibitors (TKIs). This may be due to the different panels and detection methods used in the NGS library generation and sequencing. Four NGS pipelines were performed for CSF cells samples in this study. If more than 50 ng input DNA was used, all 4 panels performed well. When input DNA was less than 20 ng, cHOPE pipeline had the best performance: cHOPE pipeline identified the largest number of mutations from CSF cell samples, and most of these mutations identified by cHOPE pipeline was also discovered in the CSF cfDNA. The ddPCR results of the patients #12 confirmed the reliability of the results by cHOPE pipeline. OncoAim and ddCAP-on-tissue pipelines are designed for tissue DNA, in which fragmentation steps are contained. The fragmentation step would result in a loss of some DNA sample and miss some variations with low AF in the samples with less DNA input. For the two pipelines designed for plasma DNA, ddCAP is based on targeted capture technology for all exons of genes in the panel, whereas cHOPE is based on multiple PCR for hotspots of genes in the panel. Multiple PCR for hotspots is more focused on the variations we interested in. In general, cHOPE is the most appropriate pipeline for CSF cell samples with less than 20 ng DNA extracted.

Most mutations detected in CSF samples could be found in FFPE samples except for ALK G689R in 2 individuals and KRAS Q61L in 1 individual, demonstrating that most tumors in the brain originated from the primary lung cancer. Because these 3 CSF-only mutations had low AFs, it is unlikely that they are highly instructive in the clinical treatment in the LAC patients with LM; additionally, these mutations were likely to have emerged after the metastasis of the tumor. Although the ALK G689R nutation is labeled as a variant of undetermined significance (VUS) at present, alterations in ALK gene are involved in many malignancies including lung cancer and could mediate acquired resistance to some ALK inhibitor [11, 12]. KRAS is known to be mutated in pancreatic, colon and lung cancers [13]. Q61L was one of the hotspots in the KRAS gene and has been confirmed to influence response to EGFR antagonists for tumor patients [14]. Hence, we deduced that mutations in CSF DNA could help illustrate the gene mutation status of brain metastasis, which can further benefit the analysis of resistance mechanisms, and guide the treatment. Mutation detection of CSF cells or cfDNA improves the accuracy in accessing the feasibility of molecular targeted drug therapies, especially for EGFR-TKIs, which have been widely proven to have significantly extended the patients’ survival.

NGS based detection of CSF samples of both cfDNA and cells can provide more gene information than plasma samples from LAC patients with LM. In addition, the cHOPE pipeline performed better than the other three NGS pipelines when low amount of input DNA from CSF cells was used.