Clinicopathologic and Genomic Landscape of Non-Small Cell Lung Cancer Brain Metastases

Abstract Background In patients with non-small cell lung cancer (NSCLC), 10%-40% will eventually develop brain metastases. We present the clinicopathologic, genomic, and biomarker landscape of a large cohort of NSCLC brain metastases (NSCLC-BM) samples. Materials and Methods We retrospectively analyzed 3035 NSCLC-BM tested with comprehensive genomic profiling (CGP) during routine clinical care. In addition, we compared the NSCLC-BM to a separate cohort of 7277 primary NSCLC (pNSCLC) specimens. Finally, we present data on 67 paired patients with NSCLC-BM and pNSCLC. Results Comprehensive genomic profiling analysis of the 3035 NSCLC-BMs found that the most frequent genomic alterations (GAs) were in the TP53, KRAS, CDKN2A, STK11, CDKN2B, EGFR, NKX2-1, RB1, MYC, and KEAP1 genes. In the NSCLC-BM cohort, there were significantly higher rates of several targetable GAs compared with pNSCLC, including ALK fusions, KRAS G12C mutations, and MET amplifications; and decreased frequency of MET exon14 skipping mutations (all P < .05). In the subset of NSCLC-BM (n = 1063) where concurrent PD-L1 immunohistochemistry (IHC) was performed, 54.7% of the patients with NSCLC-BM were eligible for pembrolizumab based on PD-L1 IHC (TPS ≥ 1), and 56.9% were eligible for pembrolizumab based on TMB-High status. In addition, in a series 67 paired pNSCLC and NSCLC-BM samples, 85.1% (57/67) had at least one additional GA discovered in the NSCLC-BM sample when compared with the pNSCLC sample. Conclusions Herein, we defined the clinicopathologic, genomic, and biomarker landscape of a large cohort of patients with NSCLC-BM which can help inform study design of future clinical studies for patients with NSCLC with BM. In certain clinical situations, metastatic NSCLC brain tissue or cerebral spinal fluid specimens may be needed to fully optimize personalized treatment.


Introduction
Non-small cell lung cancer (NSCLC) remains a leading cause of morbidity and mortality globally. Currently, 5%-10% of patients with NSCLC have concurrent brain metastases (BM) at time of diagnosis, and 20%-40% will eventually develop BM. 1 Importantly, multiple studies have shown that the prevalence of ALK fusions and EGFR mutations is high in patients who eventually develop BM, potentially associated with the extended survival (induced by the advent of the ALK and EGFR inhibitors for patients with NSCLC) allowing for spread of disease to the brain. [2][3][4] In addition, central nervous system (CNS) only progression was common in ALK-driven NSCLC given the poor CNS penetration of early generation ALK inhibitors. 5 Finally, the brain is often the primary site of progression, even in patients in which the systemic disease is controlled by targeted therapies or immunotherapy. [5][6][7] Multiple targeted therapies for genomic alterations (GAs) in tyrosine kinase-associated genes such as EGFR, ALK, ROS1, MET, RET, KRAS, BRAF, and NTRK1-3 with associated companion diagnostics (CDx) have been developed and are now approved by the United States Food and Drug Administration (FDA) to treat NSCLC. 8 In addition, testing for multiple emerging targetable biomarkers using next-generation sequencing is currently recommended by the National Comprehensive Cancer Network (NCCN) Guidelines for patients with NSCLC, including ERBB2 mutations and MET amplifications. 9 Furthermore, multiple immune checkpoint inhibitors (ICPI) have been approved by the FDA for patients with NSCLC with accompanying CDx including pembrolizumab (in patients with tumor mutational burden-high (TMB-High), microsatellite instability-high (MSI-H), or PD-L1 tumor cell expression), atezolizumab (in patients with PD-L1 tumor or immune cell expression), and ipilimumab in combination with nivolumab (in patients with PD-L1 tumor cell expression). [10][11][12][13][14] Finally, emerging genomic biomarkers for ICPI treatment decision include positive predictive biomarkers like CD274 amplification and APOBEC mutational signature and negative predictive biomarkers such as STK11/KEAP1 mutations and MDM2/MDM4 amplifications. [15][16][17][18][19] The preponderance of the initial clinical trials for targeted and immunotherapies in NSCLC were performed on patients without evidence of BM. 1 However, more recent trials have examined patients with NSCLC-BM with promising findings. For example, EGFR and ALK/ROS1 inhibitors show promising results in patients with NSCLC-BM. [20][21][22][23][24] Given this, the poor prognosis of NSCLC-BM, and the numerous approved therapies for metastatic NSCLC, the presence of the associated biomarkers for these approved agents within NSCLC-BM warrants further study. We examined the genomic landscape of NSCLC-BM samples with comprehensive genomic profiling (CGP) in a large cohort of patients to define the potential applicability of therapeutic advances in NSCLC.

Patient Cohort
This study was approved by the Western Institutional Review Board Protocol No. 20152817. We performed a retrospective CGP analysis of 3035 consecutive NSCLC-BM samples between August 2014 and June 2020 and compared the results with 7277 unpaired pNSCLC specimens. In addition, the CGP findings in 67 paired NSCLC-BM and pNSCLC samples in the same patient were separately studied. Age, sex, and site of specimen were extracted from accompanying pathology reports.

Comprehensive Genomic Profiling of Lung Carcinoma Samples
Comprehensive genomic profiling was performed using a hybrid capture-based CGP assay (FoundationOne/ FoundationOne CDx) in a Clinical Laboratory Improvement Amendments (CLIA)-certified and College of American Pathologists (CAP)-accredited laboratory (Foundation Medicine, Cambridge, MA) using previously described methods. 25 FoundationOne/FoundationOne CDx uses a next generation sequencing platform and a hybrid capture methodology that detects base substitutions, insertions/deletions, copy number alterations, and gene rearrangements in up to 324 genes, as well as tumor mutational burden (TMB) and microsatellite instability (MSI). Hematoxylin and eosin (H&E) stained slides from each sample was reviewed by a board-certified pathologist for the presence of adequate tumor (≥20% of nucleated cells are tumor cells) before sequencing. Amplification was defined as ≥ploidy + 4 (copy number 6 for diploid patients) for all genes except for ERBB2 which was defined as ≥ploidy + 3. TMB was determined on up to 1.14 Mb of sequenced DNA and TMB ≥10 mutations/Mb was considered TMB-High per CDx approval. 12,26 Microsatellite instability analysis was performed from DNA sequencing up to 114 loci and MSI-High was considered positive. 27,28 Smoking status was not known for the patients; however, tobacco as well as APOBEC mutational signatures were called as described by Zehir et al 29 Genomic ancestry of patients was determined using a principle component analysis of genomic single-nucleotide polymorphisms trained on data from the 1000 Genomes Project and each patient was classified as belonging to one of the following super populations: African, Central and South American, East Asian, European, and South Asian. 30,31 PD-L1 22C3 Immunohistochemistry Testing For a subset of cases tested with CGP, PD-L1 immunohistochemistry (IHC) was concurrently performed using the DAKO 22C3 CDx assay per manufacturer's instructions in a CLIA-certified and CAP-accredited reference laboratory (Foundation Medicine, Morrisville, NC). 32 Here, DAKO's tumor proportion scoring (TPS) method was used, where TPS =# PD-L1-positive tumor cells/(total # of PD-L1 positive + PD-L1-negative tumor cells). 33 We explored both the TPS ≥1 and TPS ≥50 cutoff in this analysis.

Statistical Analysis
Clinicopathologic differences between different NSCLC cohorts were analyzed using analysis of variance (ANOVA) or Fisher's exact test. To examine the differences in the genomic landscape of the different cohorts, we identified the top 15 genes that have GAs and compared these same genes with a Fisher's exact test. In addition, targetable GAs and emerging biomarkers were also compared. P-values were adjusted for multiple comparisons using the Bonferroni method and adjusted P-values of < 0.05 were considered significant. 34
In the NSCLC-BM cohort, significantly higher frequencies of several targetable GAs were present when compared with pNSCLC, including ALK fusions ( Table 2).
Last, due to the higher frequencies of adenocarcinoma histologic subtype in the NSCLC-BM cohort compared with the pNSCLC cohort, we examined the targetable biomarkers in  the NSCLC-BM with adenocarcinoma histology and compared it with the pNSCLC with adenocarcinoma histology.
Here, the trends in the overall NSCLC-BM and pNSCLC comparisons were the same when we specifically examined the adenocarcinoma histology. The targetable biomarkers that were significantly enriched and decreased in the overall NSCLC-BM cohort were also enriched and decreased in the NSCLC-BM adenocarcinoma only cohort, although significance was not reached for all the comparisons (Supplementary  Table S3). In addition, while not significantly decreased in the overall NSCLC-BM cohort, EGFR exon19 deletions, and EGFR L858R mutations were significantly decreased in the NSCLC-BM with adenocarcinoma histology when compared to the pNSCLC with adenocarcinoma histology (both P = .001).

Paired Primary NSCLC and NSCLC-BM Samples
In our cohort, 67 paired pNSCLC and NSCLC-BM samples were identified (Supplementary Table 4). The time between the sample collection date of the pNSCLC sample to the

NSCLC-BM PD-L1 (TPS ≥1) and TMB-High Correlation
Total NSCLC-BM sample ranged from 2 days to more than 5 years (median = 440 days). In aggregate, the NSCLC-BM samples had an additional 106 amplifications, 39 truncations, 33 deletions, 29 short variant mutations, 4 splice site variants, 1 duplication, and 1 fusion detected compared with the paired pNSCLC samples. In addition, of the 67 paired cases, 85.1% (57/67) had at least one additional GA in the NSCLC-BM sample when compared with the paired pNSCLC sample. Also, 61.2% (41/67) of the NSCLC-BM samples did not have at least one GA that was found on the paired pNSCLC sample (Supplementary Table 4).

Primary NSCLC With and Without Paired Samples
Last, we compared the genomic profiles of the 67 pNSCLC cases with paired BM samples (that eventually metastasized to the brain) with the pNSCLC without paired BM samples (Fig.  4A, B). Here, we saw a significantly higher prevalence of GAs in ALK (

Discussion
This retrospective study of 3035 NSCLC-BM revealed the clinicopathologic, genomic, and biomarker profile of NSCLC-BM in a large cohort of patients. Clinicopathologically, the NSCLC-BM cohort was different when compared with the pNSCLC cohort: namely, they were more often younger, females, smokers, or adenocarcinoma or large cell neuroendocrine carcinoma histology. Compared with pNSCLC, NSCLC-BM cases were enriched for alterations in TP53, KRAS, CDKN2A, STK11, CDKN2B, NKX2-1, RB1, MYC, KEAP1, NFKBIA, SMARCA4, RICTOR, and showed decreased prevalence of PIK3CA GA. The finding of decreased PIK3CA GA in the NSCLC-BM cohort was noteworthy. While PIK3CA mutations are currently targetable in breast cancer, there are no anti-PIK3CA targeted therapies that are FDA approved for NSCLC. 35 This finding also suggests a potential rationale for in vitro and clinical investigations into whether PIK3CA mutations hinder NSCLC to metastasize to the brain. In sum, these results can be used for future clinical trial design for patients with NSCLC-BM.
Importantly, a significant proportion of the patients (74.5%, 2261/3035) in the NSCLC-BM cohort featured GAs in at least one CDx biomarker or an NCCN recommended targetable biomarker. In the NSCLC-BM cohort, enrichment of ALK fusions, KRAS G12C mutations, and MET amplifications; and decrease of MET exon14 skipping mutations were identified. Given that the full clinical history of the patients was not available, it is not possible to confirm whether the enrichment of ALK fusions, KRAS G12C, and MET amplification in the NSCLC-BM samples is due to longer survival of the patients having been treated with systemic therapy targeting these GAs. Moreover, it is not likely that the KRAS G12C mutations and MET amplifications are associated with a propensity of these patients to develop BM associated with longer survival for these patients in that there are no FDA approvals for these GAs during the time frame of this study. Also, while MET exon14 skipping mutations were less frequent in the NSCLC-BM cohort, 31 (1.0%) patients in the NSCLC-BM cohort still harbored this potentially targetable alteration. In sum, these data suggest that if clinically feasible, CGP could be performed from a brain tissue or cerebral spinal fluid (CSF) specimen in patients with NSCLC-BM to identify targetable alterations in the BM. Of note, while the CGP results of blood liquid biopsy has shown considerable concordance with tissue biopsies and has shown clinical utility in many metastatic diseases, several studies have demonstrated that the blood-brain barrier (BBB) may limit the quantity of ctDNA that is shed into the circulating blood plasma, rendering blood liquid biopsy limited in its ability to detect the genomic profile of the NSCLC-BM. 36 In the 67-paired primary pNSCLC and NSCLC-BM samples, we found that there was at least one additional GA in 85.1% (57/67) NSCLC-BM (post-treatment) sample when compared with the paired pNSCLC; and 61.2% (41/67) of the NSCLC-BM lost at a least one GA when compared with the paired pNSCLC sample. This further adds evidence to the different genomic profiles between pNSCLC and NSCLC-BM and illustrates the importance of performing CGP on the NSCLC-BM when clinically feasible. Last, we observed a significantly higher prevalence of ALK fusions in the pNS-CLC with a paired NSCLC-BM (that eventually metastasized to the brain) when compared with pNSCLC without a paired NSCLC-BM. This is consistent with the literature that patients with ALK fusions have a predilection of developing BM, likely due to the longer survival of patients on ALK inhibitors and having time to develop BM. 2,3 The ICPI biomarker landscape in NSCLC is complex with multiple CDx indications as well as exploratory biomarkers. While the ICPI landscape in NSCLC and other tumor types was explored in numerous other studies, the literature is more lacking in NSCLC-BM. 11,[37][38][39] In the NSCLC-BM samples, enrichment of ICPI-positive predictive biomarkers of TMB-High and APOBEC mutational signature were found. However, enrichment of ICPI-negative predictive biomarkers of STK11 and KEAP1 were also discovered, exemplifying the importance of testing with CGP to identify both positive and negative predictive biomarkers for ICPI in these patients. Of note, the observed finding of higher TMB, STK11, and KEAP1 might represent overall adverse prognostic markers and predictive of more aggressive disease, rather than a finding specific to brain metastases. For a subset of patients with NSCLC-BM, we also had concurrent PD-L1 IHC (n = 1063) results. Here, we identified similar prevalence rates of PD-L1 tumor cell expression as previously reported. 40 Of interest, we discovered that when using both CGP and PD-L1 IHC, these CDx biomarkers identified independent subsets of patients that are potentially eligible for pembrolizumab, with some overlap suggesting that it is important to test with both PD-L1 IHC and CGP to identify the NSCLC-BM patients that are eligible for pembrolizumab.
The major strength of this study is the large number (>3000) of NSCLC-BM samples, all of which underwent centralized CGP. However, a primary limitation of this study is the incomplete clinical information available for patient samples and the smaller number of cases with paired pNS-CLC and NSCLC-BM samples. It is likely that some of the pNSCLC samples in our study could also have concurrent NSCLC-BM; however, in the absence of clinical histories, we do not know the extent of their disease. In addition, since most BMs are not sampled and are treated with radiotherapy, there is likely some sampling bias for NSCLC-BMs in relatively younger patients with good performance status (who are able to undergo a procedure for tissue acquisition) and symptomatic BM as opposed to elderly frail patients with NSCLC-BM whose BM were not symptomatic.

Conclusion
Herein, we defined the clinicopathologic, genomic, and biomarker landscape of a large cohort of patients with NSCLC-BM which can help inform study design of future clinical studies for patients with NSCLC with BM. In certain clinical situations, metastatic NSCLC brain tissue or cerebral spinal fluid specimens may be needed to fully optimize personalized treatment.