We retrospectively profiled 63 samples (primary lung tumor, brain metastasis, peripheral blood, and cerebrospinal fluid) from 20 newly diagnosed, untreated, histologically confirmed brain metastatic lung adenocarcinoma (LUAD) patients. Of the 20 patients, 15 had a single brain metastasis, whereas the rest of 5 cases developed multiple brain metastases (median: 3, range 2 ~ 4) diagnosed radiographically (Suppl. Figure 1A-B). In addition, 2 patients had extracranial metastases (liver and thyroid). Demographic and clinical characteristics of the 20-patient case series are listed in Suppl. Table 1. The mean sequencing depth of 20 brain metastases and 13 primary lung tumors were 987.65x and 856.15x, respectively (7 primary lung tumors had low-abundance DNA extraction through biopsy samples or surgical resections had been done in other hospitals). Target sequencing for ctDNA was successfully performed in 19 (95.00%) plasma samples and 11 (55.00%) CSF samples. With a function sequencing depth filtering, 18 plasma ctDNA (mean depth: 1878.89x) and 6 CSF ctDNA data (mean depth: 1643.50x) were proceeded to further analysis.
Genetic Divergence Of Brain Metastases And Primary Tumors
Previous studies have demonstrated that BM and their matched primary tumors are clonally related, yet genetically diverse. Each site harbors its own subset of unique mutations31. To confirm this divergence in our cohort, we performed a targeted sequencing on primary lung cancer and matched BM. The mean somatic TMB was 9.92 mutations per megabase (Mb) of DNA in primary lung cancer (range: 3 ~ 26, median: 5) and 9.35 mutations per Mb in BM (range: 2 ~ 27, median: 7.5; Suppl. Figure 2A-B). The top 20 statistically significant mutated genes in BM profiling were highlighted as well as patient characteristics including sex, number, regions of metastatic lesions and Ki67 index (Fig. 1A, mutational landscape of primary lung cancer shown in Suppl. Figure 3). Genetically, BM and primary lung tumor harbored somatic mutations majorly in TP53 (75.00% BM, 69.23% lung), EGFR (30.00% BM, 46.15% lung) and KRAS (25.00% BM, 15.38% lung), and BM diverged from primary lung cancer with more frequent alterations in MLL3, ATRX and RB1 (Fig. 1B). Several significant co-occurring alterations: EPHA5 (ARID1A), ATM (STK11), PIK3CG (INPP4B), ARID1A (LRP1B) and EPHA5 (LRP1B) were identified in BM (Suppl. Figure 2C). In addition, our data showed that the mean MAF was positively associated with Ki67 expression in BM, independent of tumor size (Fig. 1C, Suppl. Figure 2D). To further evaluate the genetic relevancy and divergence between primary lung cancer and BM, we analyzed the shared and independently evolved mutations in each individual. Results demonstrated that 61.53% (8 out of 13) of patients harbored alterations specifically in the BM that were not detected in matched primary cancer (Fig. 1D). The MAFs in BM and primary lung cancer were correlated (Pearson r = 0.54, P ˂ 0.0001; Fig. 1E), however, the mean MAF in BM was significantly higher than that in primary cancer (36.24% vs. 27.11%, P ˂ 0.0001; Fig. 1F). Furthermore, we found that an average of 23.65% (95% confidence interval [CI], 0%-50%; Fig. 1G) somatic alterations in BM were not detected in primary lung cancer, including approved actionable mutations such as EGFR, molecularly targeted mutations in development such as ERBB4, and common tumor suppressor gene abnormalities such as TP53 and RB1. Together, these genetic divergences indicate that profiling primary lung cancer is not sufficient to predict the mutations in BM. However, the accessibility of BM is challenging due to the region and number of brain lesions, the patient’s physical status, and the willingness to undergo another invasive sampling. Hence, it is urgent to develop a minimally invasive biopsy procedure to accurately profile BM and identify actionable mutations to assist in targeted drug selection.
Plasma Ctdna Is Insufficient To Recapitulate Mutations In Brain Metastases
CtDNA-based liquid assays from the plasma of peripheral blood samples have demonstrated high concordance of genomic profiles compared to tissue-based sequencing within and across cancer types13,14. To test the capability of plasma ctDNA in representing gene alterations in BM, we compared plasma ctDNA with matched BM. The data showed that only 27.78% of plasma ctDNA samples (5 out of 18) were successful in detecting all the mutations harbored in matched BM. The unidentified BM mutations by plasma ctDNA was 28.66% (Fig. 2A-B) including actionable gene alterations such as EGFR, KRAS and RET, tumor suppressor gene mutations TP53 and RB1, investigational targeted mutations ERBB4, HGF, and potential biomarker INPP4B for guiding immune checkpoint inhibitors32. Pathway analysis revealed that the undetected mutations were enriched mainly in metabolic related processes, as well as cell cycle processes (Fig. 2C). Although MAF in plasma ctDNA is weakly correlated with that in BM (Pearson r = 0.39, P ˂ 0.0001), the mean MAF in plasma ctDNA is significantly lower compared with that in BM (4.57% vs. 35.16%, P ˂ 0.0001) except in patients with extracranial metastases (BM-LUAD 6 and 11, liver and thyroid metastases, respectively; Fig. 2D, Suppl. Figure 4B-C), which is in line with previous findings where extrathoracic metastasis has a higher MAF detected than no extrathoracic metastasis in NSCLC25. We next analyzed the concordance by calculating the percentage of identified BM mutations also found in plasma ctDNA. The data indicated an average of 67.44% (95% CI: 33.33%-100%) concordance between BM and plasma ctDNA (Fig. 2E). Of note, plasma ctDNA had significant higher concordance with BM in the patients with multiple metastatic lesions (93.01% vs. 50.90% in single metastasis, P = 0.0449), regardless of sex, TMB, metastatic regions, tumor size, distance to ventricles or peritumor edema size (Fig. 2F, Suppl. Figure 4D-E, Suppl. Table 2), suggesting the number of metastatic lesions is the key factor to determine the concordance between BM and plasma ctDNA.
Csf Ctdna Analysis Is Representative Of Brain Metastatic Lesion
CSF is a potential alternative to obtain brain tumor ctDNA for analyzing mutations in BM given its direct contact with the brain. CtDNA in CSF is relatively concentrated since the blood-brain barrier prevents its cycling with the peripheral circulation system10,33. To assess whether the genomic landscape of CSF ctDNA represents mutational profiling in BM, we sequenced the CSF ctDNA obtained from lumbar puncture before brain tumor resection. We found that 83.33% (5 out of 6) CSF ctDNA samples recapitulated all the mutations in matched BM (Suppl. Figure 5A; 1 out of 25 BM alterations was absent in CSF ctDNA sequencing in BM-LUAD 7). Overall, CSF ctDNA could identify 98.31% mutations harbored in BM with only 1.69% mutations undetected (Fig. 3A-B). The mean MAF in CSF ctDNA is equivalent to that in BM (39.93% vs. 39.58%) and significantly correlated with the size of metastatic lesion by MRI volumetric analyses (Pearson r = 0.95, P = 0.0039; Fig. 3C-D). In addition, MAF in CSF ctDNA was highly correlated with that in BM (Pearson r = 0.96, P ˂ 0.0001; Fig. 3E), as well as TMB (Pearson r = 0.97, P = 0.0003; Suppl. Figure 5B-C). To further investigate whether CSF ctDNA profiling is sufficient to represent the alterations in BM, we analyzed the concordance between BM and CSF ctDNA. The result indicated that CSF ctDNA had an average of 99.33% (95% CI: 96%-100%) concordance with BM, which was significantly greater than that of plasma ctDNA (67.44%, P = 0.0125; Fig. 3F). Altogether, these results suggest that CSF ctDNA may prove a more accurate at profiling BM than plasma ctDNA.
CSF ctDNA has a greater capacity to detect BM mutations than plasma ctDNA for single BM
To further assess whether CSF ctDNA is better than plasma ctDNA in comprehensively representing the mutational landscape of BM, we performed a clonal analysis that showed CSF ctDNA had a higher detection rate of clonal mutations than plasma ctDNA (100% vs. 72.73%; Suppl. Figure 6). Comparing the proportion of the shared mutations between BM and plasma ctDNA or CSF ctDNA demonstrated that CSF ctDNA had a significantly higher percentage of shared mutations with BM than plasma ctDNA (mean: 83.61% vs. 54.55%, P = 0.0098; Fig. 4A). We then performed parallel analyses for matched trios (BM, plasma ctDNA, and CSF ctDNA) from 5 patients. The data showed that sampling of brain metastatic lesions detected a total of 52 somatic mutations with 51 identified in CSF ctDNA while only 39 were found in plasma ctDNA (Fisher exact test: P = 0.0009; Fig. 4B). CSF ctDNA had equivalent MAF to BM, which was more abundant than plasma ctDNA (Fig. 4C). MAF in CSF ctDNA was highly correlated with that in BM compared with plasma ctDNA (Pearson r = 0.98, P ˂ 0.0001 vs. 0.52, P = 0.0008; Fig. 4D). Moreover, TMB in CSF ctDNA was significantly correlated with that in BM whereas plasma ctDNA was not (Pearson r = 0.99, P = 0.0022; Fig. 4E). CSF ctDNA had a higher percentage of shared mutations with BM compared with plasma ctDNA in all 5 patients (mean indicated by dotted line: 82.83% vs. 52.06%; Fig. 4F-G). Of interest, concordance analyses revealed that plasma ctDNA had a better performance of representing BM alterations in the patients with multiple metastases than those with single metastasis (Fig. 2F). Therefore, we asked whether multiple brain metastases could be an exceptional condition where plasma ctDNA is comparable to CSF ctDNA to depict BM mutations. The concordance result indicated that plasma ctDNA had a similar power as CSF ctDNA to represent alterations in BM in the patients with multiple brain metastases (mean, 93.01% vs. 99.33%; Fig. 4H). Together, these data suggest that CSF ctDNA has an advantage in detecting mutations in BM superior to plasma ctDNA, and plasma ctDNA could be an alternative liquid biopsy material only in multiple brain metastatic NSCLC.
Csf Ctdna Facilitates The Targeted Treatment Of Brain Metastasis
Identification of actionable mutations in BM is essential for the selection and delivery of targeted therapies to promote favorable outcomes. By comparing profiling results from BM, plasma ctDNA, and CSF ctDNA (diagram: Fig. 5A), we found that CSF ctDNA had a better ability to detect clinically relevant alterations. In BM-LUAD 7, CSF ctDNA identified an established biomarker, RET mutation, which was undetectable in plasma ctDNA (Fig. 5B). In addition, HGF alteration was also detected only in BM and CSF ctDNA (Fig. 5B), which is associated with poor prognosis in cancer patients34,35. In BM-LUAD 12, CSF ctDNA had a greater advantage in detecting a set of somatic alterations that occurred in BM including tumor suppressor gene mutations TP53 and RB1, and investigational targets CDK13 and MED12 (Fig. 5C). CDK13 regulates gene transcription and co-transcriptional processes by phosphorylating the C-terminal domain of RNA polymerase II, and its selective inhibitor shows promising effects on tumor regression36. MED12 is involved in transcription initiation and has been demonstrated to control drug responses in a variety of cancers 37–39. Furthermore, patient BM-LUAD 9 developed 3 brain metastases located in the frontal lobe and cerebellum. Although MAF in plasma ctDNA here was still significantly lower than that in CSF ctDNA and BM, consistent with our data of plasma ctDNA providing equivalent efficacy to CSF ctDNA in representing profiling of multiple brain metastases (Fig. 4H), all BM mutations that were detected in both CSF ctDNA and plasma ctDNA with no significant differences (Fig. 5D). Accordingly, CSF ctDNA is a high-quality diagnostic material facilitating biomarker-based therapies for all BM, while plasma ctDNA could be an alternative liquid biopsy only in advanced NSCLC with multiple brain metastases.