*Among NSCLC brain metastases, seventy-two genes significantly signaled clinical outcomes.

*Unfavorable outcomes only associated with 14 genes while 36 only favorable.

*Genetic signals strongly indicate the ability to predict oligometastatic disease.

Nearly 170,000 patients a year in the United States are diagnosed with brain metastases (1), and approximately half of brain metastases derive from NSCLC (2). However, even amongst patients with metastases originating from the same primary cancer type, there is a significant diversity of biological and clinical outcomes. A heterogeneity in clinical outcomes derives from a variability in the size and number of metastases (3), along with differences in histology (4) and health status (5, 6) of each patient. The sensitivity of each patient’s cancer to systemic therapy has also been found to be an important variable that contributes to clinical differences between patients (7).

The diversity of the brain metastasis population has made biomarker discovery an important goal. Recent evidence suggests that there may be brain metastasis-specific mutations, and that the genetics of brain metastases may evolve differently from a patient’s primary cancer (8). Several attempts have been made to link genetic signatures found in resected brain metastasis samples to clinical outcomes of brain metastases (9, 10). However, thus far, clinically useful predictive biomarkers for brain metastasis outcomes have been generally elusive.

Over the past several years, several commercial platforms have developed for comprehensive genomic profiling of non-small cell lung cancers (11). Methods for such profiling include genetic sequencing of a biopsied tumor sample, as well as sequencing of circulating tumor DNA (12). Circulating tumor DNA has been shown to correlate with tumor mutations both in serum and cerebrospinal fluid (CSF) (13). Sensitivities have varied, among populations, notably with circulating tumor DNA of the CSF having a higher correlation to brain metastases mutations given the blood brain barrier (13–15). Though plasma samples remain valid and are more easily and commonly obtained prior to known diagnosis of brain metastases. These platforms for genomic profiling of lung cancers has helped to identify potential therapeutic targets in patients that are found to have targetable mutations (16–18), or identify populations that are more likely to respond to immunotherapy (16, 19).

Amongst the brain metastasis population, several potential clinical dilemmas exist for which biomarker discovery could potentially change practice. The ability to predict such outcomes as brain metastasis velocity (20), leptomeningeal disease (21), and which patients benefit from whole brain radiation (22) would give practitioners guidance to select proper therapies for each individual. The goal of the present study is to demonstrate the feasibility of using comprehensive genomic profiling to discover biomarkers that predict brain metastasis outcomes in patients with NSCLC. To this end, we used a single institution retrospective review of NSCLC brain metastasis patients receiving stereotactic radiosurgery (SRS) who underwent comprehensive genomic profiling and analyzed whether the genes assessed in genomic profiling were associated with clinical characteristics and patient outcomes.

NSCLC represents a genetically diverse population. In the late 1990’s, subpopulations of NSCLC patients who were found to be predominantly female non-smokers were identified and found to have cancers that responded to tyrosine kinase inhibitors (29). The mechanism of this response has been determined to be an activating mutation in the EGFR gene (30). This discovery led to a cascade of subsequent discoveries of various subpopulations of the NSCLC population including the ALK (31), ROS-1 (32), RET (33) and BRAF-mutated (34) populations. While the aforementioned mutations represent activating mutations for which targeted agents have been developed to counter, these may not represent the full story for how genomic analysis may ultimately affect care in NSCLC patients.

Biomarker discovery for brain metastasis behavior has thus far been an elusive process (35). This difficulty derives from several reasons including the histologic heterogeneity of the brain metastasis population (3), and the propensity for continued mutation between a primary tumor and the clonogens that ultimately become brain metastases (8). A preliminary study found that assessing circulating DNA in the serum as done in the present study may better detect mutations not found in a primary colorectal tumor though these findings would need to be validated in lung cancer patients (36). The present study also attempted to address the issue of histologic heterogeneity by using a population of purely NSCLC patients. Future investigations will likely include attempts to use tissue acquired from craniotomy samples in order to ensure that mutations from brain metastases are captured in the genomic analyses (9).

In the present analysis, genetic signatures were discovered for factors that have the potential to affect management. For example, patients with a signature predicting a lower SRS dose represent a population in which a dominant brain metastasis developed that was generally large and/or symptomatic. These brain metastases are ones that historically lead to significant morbidity and mortality (37, 38). Such a signature could yield a population for which surveillance imaging even prior to brain metastasis diagnosis may be useful. In addition, patients with signatures for lower BMV or lower number of brain metastases at distant brain failure may ultimately represent populations for which aggressive use of SRS is justified (39, 40), perhaps even in cases when a greater number of metastases are present than are normally offered SRS (41).

Clinical outcomes in the present analysis were scored with regards to whether they were favorable or unfavorable in the clinical setting. For example, larger or more numerous brain metastases or higher BMV were considered adverse, whereas smaller or fewer brain metastases were considered protective. This allowed for assessment of whether single genes could be favorable or unfavorable in multiple outcomes. Several gene mutations were identified within the multiple genetic signatures and these genes were found to be more likely to be either favorable or unfavorable (as opposed to discordant) across those multiple outcomes. Ultimately, genetic mutations identified in multiple unfavorable signatures are candidates, particularly in outcomes that lead to early or more numerous metastases, to be reverse-translated in laboratory studies to determine if they are part of the genetic pathway leading to brain metastasis formation.

The meaningful clinical separation between risk strata determined in the present series was quite large. For example, the survival difference between favorable, neutral, and unfavorable strata was 124 weeks, 65 weeks, and 9 weeks respectively (p=0.002). Patients with life expectancy as low as 2 months of survival may choose to have treatments that can be costly and affect quality of life. Moreover, the corresponding tumor volumes predicted for each risk strata by the present analysis based on lowest GK dose are 1 cc (favorable), 2 cc (neutral) and 6 cc (unfavorable). Tumors of the favorable or neutral scores tend to be good candidates for SRS whereas those with unfavorable scores had tumor volumes that are often best managed with surgery. These findings with regards to presenting tumor volume suggest that there may be volumetric phenotypes for brain metastasis presentation (large symptomatic vs small asymptomatic) which are driven by biology. Such a hypothesis has significant reverse translational potential.

That KPS was the single clinical characteristic assessed for which the identified signature demonstrated a weaker statistical association and minimal clinical impact served to strengthen the argument that the other signatures may be valid. Other clinical factors such as BMV, number of brain metastases and size of brain metastases at presentation are factors for which there is a reasonable assumption of a biological phenotype responsible for the size and rate of seeding of the brain with cancer. KPS on the other hand is a complex variable dependent upon multiple factors such as age, burden of systemic disease, comorbidities, and ability to access care (42, 43). Many of these factors are beyond the scope of the genetics of a patient’s cancer, and thus it would not be expected that a genetic signature could be found for KPS, but it was significant finding that this was the one variable that could not be predicted by a genomic signature in our dataset, functioning essentially as a negative control.

One of the genetic associations found in the present series was the signature for having oligometastatic extracranial disease. It has been hypothesized that a certain subset of cancers are truly oligometastatic, and thus, limited in their burden of tumor spread (44). As such, these cancers may benefit from local therapies directed towards the few sites of disease. There have been several recently published series that have suggested that such local treatment of extracranial oligometastatic disease can be beneficial for patients with regards to endpoints such as progression free survival, overall survival and need for systemic therapy (45). However, while some patients benefit and truly have a limited burden of metastatic disease, there are others that will experience rapid and diffuse failure, essentially rendering the local therapies non-useful. As with other relevant signatures found in the present series, the true clinical spread between favorable and unfavorable was quite large, as patients with favorable predictive score for oligometastatic disease had a 78% likelihood of having true oligometastatic disease. Conversely, those with unfavorable predictive score only had a 12% risk of having oligometastatic disease. If these genomic signatures for oligometastatic disease are validated, then they will have the potential to help dictate which patients may be candidates for local extracranial therapies.

There are several limitations to the present study. The study is a retrospective analysis and is therefore subject to selection bias. There also exists the issue of circulating DNA sampling as cancers are known to continue to mutate and the mutations found in the circulating DNA may or may not be present in the brain metastases. Additionally, only half the population had ctDNA drawn prior to SRS adding the potential for confounding. Circulating DNA from the CSF has been shown to correlate more closely to mutations of CNS involvement such as leptomeningeal disease (46) and brain metastases, though remain a more invasive procedure via lumbar puncture and less standardized than blood draw. If validated, however, the identified genomic signatures in this series represent clinically useful data obtained via non-invasive liquid biopsy to help risk stratify patients to potentially inform treatment or surveillance decisions.

The data analyzed in this study is subject to the following licenses/restrictions: Data will be made available per request to the corresponding author. Requests to access these datasets should be directed to John Hunting, jhunting@wakehealth.edu.

The studies involving humans were approved by Wake Forest School of Medicine IRB. The studies were conducted in accordance with the local legislation and institutional requirements. Written informed consent for participation was not required from the participants or the participants’ legal guardians/next of kin in accordance with the national legislation and institutional requirements.

MA, MC, FX contributed to conception and design of the study. YW, MA, H-WL, KW, UT procured relevant data. RD’A, JS, GW performed the statistical analysis. MA wrote the first draft of the manuscript. All authors contributed to manuscript revision, read, and approved the submitted version.