Liquid biopsy is an alternative method to detect mutations for patients if tumor samples are not available. For LM patients, CSF and plasma can be easily obtained. In this research, we characterized the genomic alteration of ctDNA in the CSF and compared it to matched plasma samples. In our study, we revealed that CSF genetic profiles had a unique role to identify patients of LUAD with LM and that plasma genetic profile failed to do so. At the time of initial diagnosis of LM, more abundant genotypings were detected in CSF than in plasma in addition to driver mutations. What’s more, we demonstrated that CSF ctDNA has a signifcantly higher detection rate than plasma. This result is consistent with previous study which show that CSF ctDNA exhibite a more comprehensive genetic landscape than plasma of CNS metastases [14, 16, 17, 19]. Previous studies have compared the detection rate of EGFR and found that it is higher in CSF than plasma. It has been found that LM is more common in patients with EGFR mutations compared to wild-type EGFR [4]. Moreover, in EGFR-mutated patients, the detection rate of EGFR was 67.6%-100% in CSF and 36.4%-73.1% in plasma [16]. In our research, EGFR is the most frequently mutated in CSF, acounting for 82.86%, higher than matched plasma. This is also consistent with Li’s findings [20], suggest that EGFR mutation may be more susceptible to BM. In short, available data revealed the heterogeneous genetic profles between CSF and plasma, and which with a good concordance in driver mutations.
In TRACERx research, Subclonal whole genome doubling (WGD) was detected in 29% of lung tumours. These data demonstrate WGD and Copy number heterogeneity was associated with shorter disease-free survival and distant metastases, respectively. WGD and copy number instability are important factors of relapse in NSCLC patients, which guide the evolutionary of clinical cancer [21]. Numerous CNVs were identified in CSF ctDNA, which were significantly higher than that with plasma in our study. Several studies indicate that CNVs were enriched in the CSF of CNS metastases [14, 22]. Therefore, we can speculate that CNV may cause distant metastasis of tumors and is a major type of mutation that causes LM, which is different from plasma.
Studies have demonstrated the genetic heterogeneity between the original tumor and the metastatic lesions in the same patient [23]. Genetically distinct subclones of the primary tumor result from somatic evolution of the tumor genome and thus have distinct biologic properties and therapeutic individualization. When tumor cells metastasize, they escape from the primary site, spread and proliferate in secondary location, and can also evade immune surveillance, eventually form metastatic lesions. This may lead to the introduction of numerous genomic heterogeneity between the final metastatic cell and the primary cancer [24]. It has been unclear to what extent the genotyping of LM differs from genotyping of primary cancers. Previous research demonstrate that overexpression of MYC, MMP13 or YAP1 that are enriched for focal amplification in brain metastases can each contribute to brain metastasis formation. TP53, CDKN2A, and TERT are abundant in a variety of metastatic cancers. While, TP53 mutations are strongly associated with genomice instability, CDK2NA and TERT play a key role in regulating cell proliferation, these both pathways are frequently perturbed in metastatic tumors. In conclusion, these mutated genes may disturb pan-cancer hallmarks of tumorigenesis, hence, improve the aggressiveness of the tumor [25, 26]. To further investigate the significance and evolutionary process of LM genetic alterations, we compare CSF/plasma samples with primary tumor samples. EGFR, TP53, MYC, CDKN2A and CDKN2B genes in CSF were significantly higher than that in LUAD tissue. Therefore, genomic characterization of CSF of LM represents a feasible strategy to find potential method for the detection of metastasis.
This conclusion was reached in the study of Nanjo et al. that the T790M mutation was less frequent in leptomeningeal than in extracranial specimens by biopsy of patients with lung cancer tissue and leptomeningeal metastases [27]. This study had confirmed this idea, after the patients became resistant to 1st- or 2nd-generation EGFR-TKIs, the T790M mutation was identified in plasma not in CSF, which may be due to the differential expression of T790M mutation in CNS and extra-CNS lesions.
In the study of Zheng et al., EGFR C797S mutation, MET dysregulation, and TP53 plus RB1 co-occurrence may be possible resistant mechanisms of LM on progression of osimertinib of CSF in NSCLC [28]. Unfortunately, C797S mutation were not found in our study of Cohort 2, because of lacking of enough samples hinders the further discussion. The mechanisms of resistance to osimertinib progression of LM patients may be found in the CSF, such as EGFR CNV. Furthermore, other study has also revealed that EGFR amplification is the resistance mechanism associated with EGFR-TKIs [29]. From the AURA3 trial, EGFR mutation was one of the most common acquired resistance mechanisms detected, followed by MET amplification [30]. In Cohort 2, we found that EGFR CNV occure in 7 patients and MET mutations in 3 patients after the diagnosis of LM, which partly accounted for the progressive disease of LM. In our study, we found that cell cycle pathway alterations (CDKN2A, CDKN2B, CDK4, CDK6) detected after osimertinib administration. In previous studies, altered cell cycle genes may be involved in the mechanism of resistance to osimertinib as the 1st- or 2nd-line therapy [31]. PIK3CA amplification or mutations promote tumor infiltration and activate the PI3K/AKT/mTOR pathway, which suggested that PI3K/AKT/mTOR pathway activation might be related to the resistance of 3rd-generation EGFR-TKIs [31, 32]. Similar to this report, in our 2 Cohorts, PIK3CA mutation was present in the osimertinib-resistant cohort but not in the osimertinib-naive cohort.
Tissue biopsy of the primary tumor to determine the presence of SCLC transformation is also one of the resistance mechanisms to osimertinib [33]. However, lacking of matched Primary cancer tissue genetic profiles limits further clarification of the results. The discovery of these important mechanisms of acquired resistance to EGFR-TKIs could facilitate precise treatments for such patients after disease progression.
In our study, a multivariate analysis indicated that the presence of EGFR mutation in lung tissue was independent favourable predictors of survival, whereas TERT mutation in CSF was an independent predictor of poor survival after excluding other confounding factors. Patients with advanced LUAD who harbored EGFR mutations had significantly longer OS than those without EGFR mutations after treatment with EGFR-TKIs [34]. Suda et al. reviewed that the better prognosis of patients with EGFR mutations may be related to the use of EGFR-TKIs [35]. This has been confirmed by other studies that EGFR-TKIs after LM diagnosis were independent favourable predictors of survival [36]. There is no doubt that, all 26 patients with EGFR mutations in lung tissue were treated with EGFR-TKIs before the diagnosis of LM. The TERT gene is a ribonucleoprotease that is essential for the replication of chromosome termini and telomere elongation in eukaryotes. The study suggesting that targeting TERT promoter (pTERT) mutations may serve as a viable approach for cancer therapy [37, 38]. Previous studies have suggested that TERT mutations are associated with a poor prognosis in tumors, such as thyroid malignancies, melanoma and gliomas [39–42]. Yang et al. found TERT mutations were detected in 11% of patients with NSCLC, of which TERT mutation carrier status was an independent risk factor for poor prognosis [43]. Likewise, TERT mutations were found in the CSF of 11.43% of patients in LM with LUAD and these patients had worse prognosis in our study. Thus, we believe that TERT mutation may have clinical value as a potential biomarker for disease monitoring [44, 45].
In the previous literature, TP53 mutation and EGFR/TP53 co-mutation are considered a poor prognostic factor in LUAD patients [46]. In our research, the CSF with TP53 mutation showed shorter OS than the others group (P < 0.05). Dual EGFR/TP53 mutation was associated with inferior OS compared with the dual EGFR/TP53 wild, although these results were not statistically significant. In this study, CDKN2A is common in CSF samples, accounting for 28.6%, and regardless of prognosis. This is consistent with the findings of Yang et al. [47].
In the CSF circulation, disseminated cancer cells can float freely within the CSF or attach to the meninges, which can be captured by CSF cytological examination or appear as linear or nodular enhancement on MRI [48]. Remsik et al. found that floating cells were more invasive in vivo compared to adherent or mixed cells in mouse modeling, further manifested by rapid development of neurological symptoms and reduced survival. Remarkably, they found that patients diagnosed with only positive CSF cytology demonstrated substantially diminished survival after LM diagnosis through clinical case collection [49]. As well, in the current study, patients with negative CSF cytology still survived and survived significantly longer than those with positive CSF cytology, whether or not the enhanced brain MRI was positive.
Finally, we demonstrated with a specific case that dynamic changes in CSF ctDNA at different stages could better predict intracranial tumor responses and track clonal evolution in LM patients.
There are several limitations in our study. First, this was a retrospective study with small samples. Second, matched primary lung cancer tissues of the patients were unavailable, the NGS data were obtained from CSF or plasma derived ctDNA without the analysis of matched tumor tissue DNA. Third, there is a lack of observation of csf tumor markers, and we will continue this study in future observations. However, this is still a rare study that explored exactly matched CSF and plasma genetic information and performed survival prognosis analysis in patients with LUAD with LM.