Genomic landscape and actionable mutations of brain metastases derived from non–small cell lung cancer: A systematic review

Abstract Background Brain metastases derived from non–small cell lung cancer (NSCLC) represent a significant clinical problem. We aim to characterize the genomic landscape of brain metastases derived from NSCLC and assess clinical actionability. Methods We searched Embase, MEDLINE, Web of Science, and BIOSIS from inception to 18/19 May 2022. We extracted information on patient demographics, smoking status, genomic data, matched primary NSCLC, and programmed cell death ligand 1 expression. Results We found 72 included papers and data on 2346 patients. The most frequently mutated genes from our data were EGFR (n = 559), TP53 (n = 331), KRAS (n = 328), CDKN2A (n = 97), and STK11 (n = 72). Common missense mutations included EGFR L858R (n = 80) and KRAS G12C (n = 17). Brain metastases of ever versus never smokers had differing missense mutations in TP53 and EGFR, except for L858R and T790M in EGFR, which were seen in both subgroups. Of the top 10 frequently mutated genes that had primary NSCLC data, we found 37% of the specific mutations assessed to be discordant between the primary NSCLC and brain metastases. Conclusions To our knowledge, this is the first systematic review to describe the genomic landscape of brain metastases derived from NSCLC. These results provide a comprehensive outline of frequently mutated genes and missense mutations that could be clinically actionable. These data also provide evidence of differing genomic landscapes between ever versus never smokers and primary NSCLC compared to the BM. This information could have important consequences for the selection and development of targeted drugs for these patients.


Protocol
We registered a protocol on the International Prospective Register of Systematic Reviews (PROSPERO: https:// w w w .c r d .y o r k .a c .u k / p r o s p e r o / d i s p l a y _ r e c o r d .php?ID=CRD42022321782) and followed the guidelines of Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA). 10,11We did not require ethical approval for this study as all the data used in our analyses were from previously published articles.

Search Strategy and Selection Criteria
We considered studies to be eligible if they (i) included samples/patients clinically diagnosed with a BM derived from NSCLC; (ii) had at least 2 mutations analyzed in the sequencing of BM; (iii) performed sequencing on BM tissue; and (iv) were cohort studies (including randomized trials and other controlled/uncontrolled clinical trials), case series, or case reports.There were no restrictions on language.
We identified records through a systematic literature search of Embase, MEDLINE, Web of Science, and BIOSIS from inception to 18/19 May (Supplementary Tables 1-4), we then uploaded the records to Endnote and de-duplicated. 12Next, we uploaded the remaining articles to Rayyan. 13Two independent reviewers screened records by title and abstract using Rayyan software and records which did not fit eligibility criteria were excluded.Two independent reviewers assessed the eligibility of the full texts for all remaining references.Any discrepancies during the screening process were referred to a third reviewer.
We carried out the data extraction into a Microsoft Excel document.We extracted data on the following: as publication details, patient characteristics, subtype of NSCLC, time to BM, overall survival, and genes mutated in BM.One reviewer extracted the data from each included record and a second reviewer checked this.We did not extract data looking at loss of heterozygosity.In addition to our prespecified data extraction, we extracted information on PD-L1 protein expression from the BM since this has emerged as an important biomarker for

Importance of the Study
Brain metastases (BMs) derived from non-small cell lung cancer (NSCLC) represent a significant clinical problem.We provide a comprehensive systematic review of the genomic landscape of brain metastatic NSCLC to better inform novel precision medicine approaches.This review reports frequently mutated genes in BM derived from NSCLC and the most common missense mutations, with information on drug targets.Differing genomic profiles in NSCLC BM compared to the NSCLC primary and between smoking status are highlighted.Overall, this information could have important consequences for the selection and development of targeted drugs for patients.

Andrews et al.: Genomic landscape of BM-NSCLC
response to immune checkpoint inhibitors.Where the data were available, we also assessed if the primary NSCLC tumor had the same gene mutated as the BM, since this could provide important information regarding whether targeted treatment can be selected without access to BM tissue.

Risk of Bias
One reviewer assessed risk of bias in the included studies using the Hoy et al. risk-of-bias tool. 14We considered studies to be at low risk of bias where all items received a yes response, moderate risk where 1 item received a no response, and high risk where 2 or more items received a no response.

Statistical and Actionability Analysis
We synthesized the data from the included papers using Microsoft Excel, which we also used to create result tables and bar charts.We included a subgroup analysis looking at the genomic profile of BM in ever and never smokers, as defined in the individual publications.For all patients (including never-and ever-smoker subgroups), we also investigated distinct missense mutations present in frequently mutated genes.This analysis only included data that specified the exact type of missense mutation sequenced.
We used OncoKB to look at specific missense mutations found in the top 10 mutated genes in all patients to generate the level of evidence for each biomarker and considered if they could be actionable (https://www.oncokb.org). 15We also used the drug-gene interaction database (DGIdb) to assess the potential druggability of the selected genes (https://www.dgidb.org). 16We searched ClinicalTrials.gov to identify ongoing or completed clinical trials of drugs targeting mutant genes in NSCLC BM (https://clinicaltrials.gov).We refined our search by using the terms "brain metastasis, " "CNS, " "brain metastases, " "Non-small Cell Lung Cancer" and selecting for recruiting, active, not recruiting, completed studies, and only considering adults or older adults.

Mutation Similarity Between Brain Metastasis and NSCLC Primary
We investigated the top 10 most commonly mutated genes in our gene list.We only included gene mutations that specified the distinct mutation in the BM and the primary.Copy number variant and other nonspecific mutations were not included.We identified the mutation in the NSCLC BM and then looked at the same gene in the primary tumor to see if there was the same/different/no mutation.

Study Selection and Characteristics
We carried out a systematic literature search on Embase, MEDLINE, Web of Science, and BIOSIS (number of papers, n = 3266) (Supplementary Tables 1-4).Of these papers, 1109 were duplicates, 1476 were excluded after title and abstract screen, and 609 were removed after full-text screen (Figure 1).A total of 72 distinct studies were included, with data on 2346 patients.Summary data were reported for 1798 of these patients and individual data for 567; some papers reported both (Table 1).We found 28 studies to be at low risk of bias, 31 at moderate risk, and 13 at high risk of bias (Supplementary Figures 1 and 2).We found 31 studies to have a high risk of bias due to not providing an acceptable case definition.For example, when a study stated the presence of a mutation, for example, mutated EGFR, but not the specific type of mutation, for example, L858R missense mutation in EGFR gene.So, we could not include these data in missense mutation analysis and the comparison between the genomic landscape of BM and primary NSCLC.
The majority of patients with individual-level data included in this analysis were histologically diagnosed with adenocarcinoma (n = 387, 67.2%), SCC (n = 35, 6.1%), adenosquamous carcinoma (n = 15, 2.6%), and LCC (n = 11, 1.9%).The rest were unknown, or the data were unavailable.This is similar to the NSCLC population demographic; however, there is a slight overrepresentation of the adenocarcinoma subtype.Overall, from the limited demographic data we have, we expect these to follow the typical NSCLC population demographic (Table 1).
We observed 4 patients who had >600 mutations reported, so we initially did not extract the data.Once we discovered the gene list of >25 mutations in NSCLC BM, we checked to see if these 4 patients had the same mutated gene, and if so, this was added to the analysis.We also identified 3 patients with >5 mutations in a single gene; this was reported as only 5 mutations to avoid outlier bias.
We further looked at the distinct missense mutations of TP53 and EGFR in BM of ever and never smokers.TP53 had no concordant missense mutations between ever versus never smokers (Supplementary Figure 11).For EGFR, ever and never smokers had 7 and 8 L858R mutations, respectively.Both groups were found to have 2 T790M mutations, but no other concordant mutations were found (Supplementary Figure 12).

Clinically Actionable Mutations and Drugs
For our commonly mutated gene list in all patients, DGIdb found 22 clinically actionable genes, 15 genes related to drug resistance, and 13 that have a potentially druggable genome (Supplementary Table 5).Of 91 studies identified in the clinical trial search, 38 were of drugs to target mutated genes (Supplementary Table 6).
Biomarker Evidence and FDA-Approved Drugs L858R, T790M, G719, and L861Q EGFR missense mutations and G12C KRAS missense mutation are FDArecognized biomarkers predictive of response to an FDA-approved drug (level 1) reported in NSCLC (Table 2).For EGFR, afatinib targets L858R, G719, and L861Q, osimertinib targets L858R and T790M, dacomitinib, erlotinib, erlotinib + ramucirumab combination, and gefitinib target L858R.For KRAS, adagrasib and sotorasib target G12C.Osimertinib has been FDA approved for targeting G719 and L861Q, and these are currently standard-of-care biomarkers (level 2).Other drugs have been considered for missense mutations in TP53, EGFR, KRAS, CDKN2A, STK11, and PIK3CA but these are not FDA approved (Table 2).For EGFR, T790M is a standard-of-care biomarker predictive of resistance to erlotinib, gefitinib, and afatinib in NSCLC, D761Y is also considered a biomarker of resistance to gefitinib but this is less well evidenced (Table 2 and Supplementary Figure 12).It is important to note these levels of biomarker evidence have been accepted for systemic therapies (solid tumors and NSCLC), but this is not evidenced in BM (Table 2).

Mutation Similarity Between BM and NSCLC
There were 647 mutations among the top 10 overall mutated genes incorporated in this analysis.We identified 408 mutations (63%) which were the same in both the BM and the primary NSCLC, and 239 mutations (37%) that were discordant.Of this subgroup, TP53 (n = 121), EGFR (n = 94), and KRAS (n = 65) have the most data.We found the mutations that were most often similar between BM and NSCLC were in TP53 (67%), KRAS (66%), and EGFR (58%).

PD-L1 Expression
8][19] We found a total of 28 patients, consisting of 21 lung adenocarcinoma (75%), 6 SCC (21.4%), and 1 with subtype data unavailable (3.6%).Of this subgroup, 25 patients (89.3%) were found to have 0%-49% of   Level 1 = FDA-recognized biomarker predictive of response to an FDA-approved drug in this indication, level 2 = standard care biomarker to an FDA-approved drug in this indication, level 3A = compelling clinical evidence biomarker is predictive of response to drug in this indication, level 3B = standard care or investigational biomarker predictive of response to FDA-approved or investigational drug in another indication, level 4 = compelling biological evidence biomarker is predictive of response to a drug.Level R1 = standard care biomarker predictive of resistance to an FDAapproved drug in this indication, level R2 = compelling clinical evidence biomarker is predictive of resistance to a drug.

Andrews et al.: Genomic landscape of BM-NSCLC
PD-L1 expression.Only 3 patients (10.7%) had PD-L1 expression which was >50% and these patients were all diagnosed with lung adenocarcinoma (Table 3).Patients are classified as having a high PD-L1 expression if a tumor proportion score ≥50%, as this is the FDA-approved level for first-line treatment of primary NSCLC. 20

Discussion
This review included 72 studies with data from 2346 patients with BM derived from NSCLC, of which 567 had individual-level data.These studies provided information on the commonly mutated genes and missense mutations in BM derived from NSCLC, comparison of the genomic landscape between ever versus never smokers and primary NSCLC versus BM, and PD-L1 expression in BM.
In our cohort, over 350 genes were reported to be mutated at least twice, with 22 genes found to have >25 mutations.Twelve of these mutated genes were found to be concordant with a large cohort study of BM from NSCLC: EGFR, TP53, KRAS, CDKN2A, STK11, PIK3CA, MYC, CDKN2B, KEAP1, NKX2-1, SMARCA4, and RB1 (Figure 2A). 6The same study also found NFKBIA, RICTOR, and NF1 to be frequently mutated; these genes were also identified in our cohort but were not in our top mutated genes. 6A meta-analysis found TP53, EGFR, KRAS, STK11, and EML4-ALK to be frequently mutated in NSCLC. 21We identified a similar pattern in our BM derived from NSCLC.However, our study discovered some differences between the mutations present in the primary NSCLC and the BM.Primary NSCLC and BM were found to harbor different mutations in 37% of cases, this evidence is in keeping with previous studies suggesting the NSCLC primary and derived BM suggest genetic differences, thus highlighting the importance of sequencing BM derived from NSCLC due to differing genomic landscapes. 6,22he frequently mutated genes in BM derived from NSCLC included TP53, EGFR, KRAS, CDKN2A, STK11, MET, PIK3CA, MYC, TERT, and CDKN2B, which we considered to be of most interest to target for intervention.Currently, EGFR and ALK have the most well-established actionable genetic alterations for metastases derived from NSCLC.4][25][26][27] These drugs were also identified in our OncoKB database search with varying levels of biomarker evidence depending on the mutation type.ALK also presented many treatment options such as alectinib, although this was less frequently mutated in our gene list. 8,28More recently, drugs have been discovered that target genes that were previously difficult, such as KRAS.Two G12C inhibitors have been approved (sotorasib and adagrasib), with other clinical trials ongoing. 8OncoKB identified a number of drugs that are currently being tested in our frequently mutated gene list, but these are not FDA approved.These drugs included TP53 with PC14586 in all solid tumors, EGFR with patritumab deruxtecan in NSCLC, KRAS with trametinib, cobimetinib, and binimetinib in all solid tumors, CDKN2A with abemaciclib, palbociclib, and ribociclib, STK11 with bemcentinib + pembrolizumab, and PIK3CA with RLY-2608 and LOXO-783 in all solid tumors (Table 2). 15n our smoking subgroup analysis, the genomic profile of BM in never smokers identified more EGFR mutations compared to ever smokers.Likewise, ever smokers had more TP53 mutations.The genomic landscape comparing smoking status in BM seemed to differ, with alternative genes found to be frequently mutated, excluding TP53 and EGFR (Figure 2B and 2C).Distinct missense mutations in TP53 and EGFR between ever smokers were compared with never smokers and were found to differ, with the exception of L858R and T790M which were identified at similar frequencies.Interestingly, only ever smokers were found to have the missense mutations L861Q and G179S in EGFR which are clinically actionable (Table 2 and Supplementary Figure 12).Previous studies investigating the genomic landscape of NSCLC in ever versus never smokers found a similar pattern to our data, with EGFR mutations more frequent in never smokers, and TP53 and KRAS more commonly mutated in ever smokers. 5Our data found high PD-L1 expression (>50%) to be uncommon in our cohort, with 25 patients (89.3%) with 0%-49% of PD-L1 expression and only 3 patients (10.7%) had >50% PD-L1 expression, suggesting that immune checkpoint inhibition may be effective in only a small proportion of these patients.PD-L1 was also found to be infrequently expressed in the BM in a previous study with found seven (21.9%) of patients with PD-L1 ≥5% and 25 (78.1%) of patients with PD-L1 <5%. 29here were some limitations to this review.We only included studies of patients/samples with sequenced tumor tissue rather than circulating tumor DNA as tissue sequencing is still the gold-standard technique for molecular tests. 30However, the consequence of this is that the many studies that sequence circulating tumor DNA were not included in our review.The data are also biased to BM where the brain tumor was resected, making tumor tissue available to sequence, which likely depends on both BM size and location. 31One limitation of the published literature is the lack of granularity on the lineage of the metastatic NSCLC, that is, adenocarcinoma versus SCC, and we would recommend that all subsequent genomic studies include precise diagnosis by lung pathologists, where possible.
Some studies we reviewed reported the presence of a mutation in a gene but did not clarify the specific type of mutation, so we could not include these data in our analysis of distinct missense mutations in top mutated genes in BM from NSCLC.In addition, for many of the studies using NGS and other sequencing platforms, we have no knowledge of genes that were not mutated as we did not have access to the full list of genes that were tested and/ or which of those tests had failed.There also could be publication and reporting bias as candidate genes that are already known to be mutated in the NSCLC primary tumor are more likely to be sequenced, so their mutation status is more likely to be reported compared to lesser-known genes.Considering these limitations, we were not able to generate a prevalence estimate for each gene in the BM derived from NSCLC.The studies included in our review used a wide range of sequencing panels which may lead to some mutations being more represented or identified compared to others, which could have led to bias in our results.There is also a slight overrepresentation of adenocarcinoma in the NSCLC population in our cohort, which may lead to bias with mutations commonly seen in this subtype to be identified more frequently.
The genomic landscape of BM compared to the NSCLC primary should be interpreted with caution as our search criteria identified BM which had a mutation and we then looked to see if the same gene was mutated in the primary NSCLC.Therefore, the data is biased toward BM gene mutations, as we are missing the data where the primary NSCLC has a mutated gene that is not identified in the BM.In this analysis, we were also unable to include mutations that were identified in either primary or BM but which lacked an exact description to define if they matched, that is, when the gene has a missense mutation versus L858R missense mutation, the first option was insufficient.Similarly, we were not able to include CNV variation in this analysis as we were unable to identify the number of copies of each gene that were present.

Figure 1 .
Figure 1.PRISMA diagram of included studies in genomic landscape of NSCLC-derived brain metastasis.
Andrews et al.: Genomic landscape of BM-NSCLC

Figure 2 .
Figure 2. Common mutated genes within the NSCLC BM cohort in decreasing order.(A) All patients, (B) ever smokers, (C) never smokers.

Table 1 .
Study characteristics of included studies.

Table 2 .
Level of evidence for drugs targeting missense mutations in NSCLC and all solid tumors for the missense mutations NSCLC BM cohort found on OncoKB

Table 3 .
PD-L1 expression in patients included in the NSCLC brain metastasis cohort