Abstract
Introduction
Biomarkers that predict neurologic death may allow for personalization of therapy for high-risk brain metastases patients.
Methods
Patients with NSCLC who underwent comprehensive genomic profiling were identified in an institutional database. Neurologic death was determined by medical record review. Proportional hazards regression models considering non-neurologic death as a competing risk were used to identify mutations statistically associated with the occurrence of neurologic death (p < 0.1) and to create a risk scoring system for neurologic death. A competing risk proportional hazards regression model with non-neurologic death as a competing risk was used to assess the association between the risk score and neurologic death, and to calculate hazard ratios predicting neurologic death between risk groups.
Results
307 patients were included in the primary analysis and 213 in a cohort of patients with brain metastases. Risk scores were constructed in both populations. Patients with higher risk scores had an increased risk of neurologic death when compared to those in the low-risk group, with respective HRs of 3.76 for the entire cohort and 2.87 for the brain metastasis cohort per unit increase in the risk score. When dividing the risk score into three groups, the cumulative incidence of neurologic death in high, moderate, and low risk groups was 49.0%, 20.3% and 0% for the entire cohort and 52.4%, 30.4%, and 0% in the brain metastasis cohort.
Conclusion
Development of a genomic signature to predict neurologic death via non-invasive liquid biopsy appears feasible in NSCLC patients.
Keywords: Brain metastases, Next generation sequencing, Non-small cell lung cancer, Genetic biomarkers, Radiotherapy, Brain death
Introduction
Approximately 200,000 patients in the United States are diagnosed with brain metastases each year [1]. In the past, the diagnosis of brain metastasis led to a high mortality rate due to the likelihood of death from neurologic causes [2]. However, over the past two decades, there has been a stage migration from large metastases diagnosed because of symptomatic presentation to a population with small asymptomatic metastases discovered during staging protocols [3]. Because of this change in the characteristics of the brain metastasis population, there has been a shift in whether metastatic cancer patients die of brain metastases or from other causes, such as their primary cancer, distant metastases, or intercurrent disease [4].
The decrease in neurologic death is generally due to lead-time bias from earlier diagnosis, though novel systemic therapies [5, 6] and CNS-directed therapies [7, 8] have also contributed to this improvement in outcomes. Despite improvements and shifts in the brain metastasis population, a significant proportion of patients with brain metastases will still die from neurologic causes. In the modern population, recent trials have shown that nearly 20% of patients with brain metastasis will experience neurologic death [9]. Potential risk factors for higher neurologic death rates include melanoma histology [10], high brain metastasis velocity [11], brainstem location [12] and larger tumors [8].
Recent analyses have demonstrated that genomic profiling can be used to predict brain metastasis behavior, including which patients may develop brain metastases [13] and which patients develop larger or more numerous metastases [14]. As such, it was hypothesized that genetic profiling might also predict the risk of brain metastasis patients dying of neurologic death. As this remains a concern in the modern brain metastasis population, several options could be considered for patients who are at increased risk of neurologic death, including surveillance, treatment intensification, and prophylaxis.
The present study represents a retrospective analysis of a population of patients with metastatic non-small cell lung cancer who underwent comprehensive genomic profiling with a proportion developing brain metastases at some point during their natural history. We hypothesized that we could use genomic profiling to classify patients into distinct risk groupings that were predictive of time to neurologic death, which could have downstream clinical utility such as escalating upfront treatment and increasing surveillance.
Methods
Data acquisition
This study was approved by the Wake Forest School of Medicine Institutional Review Board, IRB00033881. Patients with non-small cell lung cancer were identified from our prospective institutional database of patients with comprehensive genomic profiling. Patients were included in this study if their comprehensive genomic profiling (CGP) was done using the Guardant platform. Patients with CNS metastases at time of diagnosis, those who developed CNS metastases during their lives and those who never developed CNS metastases were all included in the study to determine if genomic factors could differentiate between positive and negative outcomes regardless of initial clinical staging. The electronic medical record (EMR) was then used to determine which of these patients developed brain metastases during their natural history and to identify patients who died of neurologic death, as defined in the publication by McTyre et al. [15] and Patchell et al. [16]. In brief, cause of death was considered neurologic death if a patient died with progressive neurologic deterioration or severe neurologic dysfunction, typically characterized by an uncontrolled increase in intracranial tumor burden, irrespective of their extracranial disease status.
Comprehensive genomic profiling
Comprehensive genomic profiling (CGP) was performed using the Guardant 360 platform (Guardant Health, Palo Alto, CA). Test samples were acquired via liquid biopsy performed on peripheral blood, which were obtained in accordance with National Comprehensive Cancer Network (NCCN) guidelines prior to the administration of systemic therapy, as previously published by Leighl et al.[17] This test uses circulating tumor DNA in patient samples to assess for the presence of known mutations common to NSCLC, including AKT1, ALK, APC, AR, ARAF, ARID1A, ATM, BRAF, BRCA1, BRCA2, CCND1, CCNE1, CDH1, CDK4, CDK6, CDK12, CDKN2A, CHEK2, CTNNB1, DDR2, EGFR, ERBB2, ESR1, EZH2, FANCA, FBXW7, FGFR1, FGFR2, FGFR3, GATA3, GNA11, GNAQ, GNAS, HNF1A, HRAS, IDH1, IDH2, JAK2, JAK3, KEAP1, KIT, KRAS, MAP2K1, MAP2K2, MAPK1, MET, MLH1, MPL, MSH2, MSH6, MTOR, MYC, NF1, NFE2L2, NOTCH1, NPM1, NRAS, NTRK1, NTRK2, NTRK3, PALB2, PDGFRA, PIK3CA, PMS2, PTEN, PTPN11, RAD51D, RAF1, RB1, RET, ROS1, SMAD4, SMO, STK11, TERT, TP53, TSC1 and VHL.
Statistics
In order to create a risk stratification score for the prediction of neurologic death, a series of Cox proportional hazards regression models were fit to identify genetic mutations associated with time to neurologic death in the full cohort (n = 307) of patients. A p-value of < 0.1 was used for this first screening of statistical association. Genes with statistical associations were assigned a score of + 1 if they led to an increased risk of neurologic death (i.e. had a hazard ratio greater than 1 for predicting time to neurologic death), or -1 if they led to a decreased risk of neurologic death (i.e. hazard ratio less than 1). The sum of these values was calculated to define a genetic risk score (with scores ranging from negative to positive values) and a 3-level risk score (high, moderate, and low risk), with respective scores of > 0, 0, and < 0. A competing risk proportional hazards regression model with non-neurologic death as a competing risk was then used to assess the association between the genetic risk score and neurologic death, and to calculate hazard ratios (HR) predicting neurologic death for each unit increase of the genetic risk score. We additionally examined overall survival, using the 3-level genetic risk score and the overall genetic risk score, and compared groups using Log-Rank tests.
Additional exploratory and sensitivity analyses were conducted to evaluate the clinical significance of the genetic risk score. The association between overall survival and genetic risk score was assessed using a log-rank test comparing Kaplan-Meier curves across the three risk score groups. Next, the association of the genetic risk score with neurologic death was examined among subgroups both with and without targetable mutations (EGFR, ROS1, BRAF, NTRK3 and ALK), using competing risks regression. Finally, we constructed a multivariable competing risks proportional hazards model to investigate whether the genetic risk score remained independently associated with time to neurologic death when accounting for patient-level factors, including Karnofsky performance status (KPS), number of brain metastases, overall disease burden (none, oligometastatic, or widespread), and type of brain irradiation, including whole brain radiotherapy (WBRT), Gamma Knife radiosurgery (GKRS), both, or none.
Results
Patient population
A total of 307 patients underwent CGP between 2005 and 2021 and were included in the analysis. The median follow-up was 32.9 months. Of these 307 patients, 213 (69.4%) developed brain metastases over the course of the study, with 127 (41.4%) presenting with CNS metastases at time of initial diagnosis. A total of 246 of 307 patients in the entire cohort and 182 of 213 in the brain metastasis cohort had died at the time of the analysis. Neurologic death occurred in 63 patients: 21% of the entire population and in 29.6% of patients who developed brain metastases. Patient characteristics for all subjects, and for those in the neurologic death and non-neurologic death cohorts are summarized in Table 1. Patients experiencing neurologic death were younger and more likely to have advanced stage disease. At initial diagnosis, these patients had 4.7 versus 1.9 brain metastases, respectively (p = 0.001), and were more likely to receive brain-directed radiotherapy during their disease course. Patient characteristics among those who developed brain metastases are summarized in Table 2. Across the subset of patients who developed brain metastases, patients with KRAS mutations were more likely and those with EGFR mutations were less likely to be in the high-risk groups. There were no differences in the rates of ALK, ROS1, or BRAF mutations across the different risk groups. Additionally, there were no statistically significant differences in KPS, stage at diagnosis, or NSCLC histology in the brain metastasis subset.
Table 1.
Patient characteristics of entire cohort, neurologic death cohort, and non-neurologic death cohort
| Entire Cohort n (%) / median (IQR) | Neurologic Death Cohort n (%) / median (IQR) (n=63) | Non-Neurologic Death Cohort n (%) / median (IQR) (n=244) | p-value | |
|---|---|---|---|---|
| Age | 67.4 | 65.1 | 68 | 0.024 |
| Sex | 0.048 | |||
| Male | 151 | 24 (15.9%) | 127 (84.1%) | |
| Female | 156 | 39 (25.0%) | 117 (75.0%) | |
| Ethnicity | ||||
| White | 266 | 52 (19.5%) | 214 (80.5%) | 0.538 |
| African American/Black | 38 | 10 (26.3%) | 28 (73.7%) | |
| Other | 3 | 1 (33.3%) | 2 (66.7%) | |
| KPS | 0.750 | |||
| 100 | 5 | 0 (0.0%) | 5 (100%) | |
| 90 | 61 | 9 (14.8%) | 52 (85.2%) | |
| 80 | 131 | 31 (23.7%) | 100 (76.3%) | |
| 70 | 59 | 17 (28.8%) | 42 (71.2%) | |
| 60 | 29 | 5 (17.2%) | 24 (82.8%) | |
| 50 | 10 | 1 (10.0%) | 9 (90.0%) | |
| 40 | 9 | 0 (0.0%) | 9 (100%) | |
| Unknown | 2 | |||
| Actionable Mutations | ||||
| EGFR | 0.486 | |||
| Yes | 68 | 16 (23.5%) | 52 (76.5%) | |
| No | 239 | 47 (19.7%) | 192 (80.3%) | |
| KRAS | 0.778 | |||
| Yes | 93 | 20 (21.5%) | 73 (78.5%) | |
| No | 214 | 43 (20.1%) | 171 (79.9%) | |
| ALK | 0.676 | |||
| Yes | 18 | 3 (16.7%) | 15 (83.3%) | |
| No | 289 | 60 (20.8%) | 229 (79.2%) | |
| ROS1 | 0.286 | |||
| Yes | 12 | 1 (8.3%) | 11 (91.7%) | |
| No | 295 | 62 (21.0%) | 233 (79.0%) | |
| BRAF | 0.559 | |||
| Yes | 25 | 4 (16%) | 21 (84%) | |
| No | 282 | 59 (20.9%) | 223 (79.1%) | |
| Smoking Status | 0.921 | |||
| Former | 191 | 39 (20.4%) | 152 (79.6%) | |
| Current | 67 | 13 (19.4%) | 54 (80.6%) | |
| Never | 49 | 11 (22.4%) | 38 (77.6%) | |
| Stage | 0.032 | |||
| I | 5 | 0 (0.0%) | 5 (100%) | |
| II | 4 | 0 (0.0%) | 4 (100%) | |
| III | 37 | 2 (5.4%) | 35 (94.6%) | |
| IV | 261 | 61 (23.4%) | 200 (76.6%) | |
| NSCLC Histology | 0.746 | |||
| Adenocarcinoma | 226 | 49 (21.7%) | 177 (78.3%) | |
| Squamous cell carcinoma | 47 | 9 (19.1%) | 38 (80.9%) | |
| Large cell carcinoma | 2 | 0 (0.0%) | 2 (100%) | |
| NSCLC-NOS | 32 | 5 (15.6%) | 27 (84.4%) | |
| Diagnosed with Brain Metastases | <0.001 | |||
| Yes | 213 | 63 (29.6%) | 150 (70.4%) | |
| No | 94 | 0 (0.0%) | 94 (100%) | |
| Number of brain metastases at initial diagnosis | 2.5 | 4.7 | 1.9 | 0.001 |
| Radiation | <0.001 | |||
| None | 115 | 7 (6.1%) | 108 (93.9%) | |
| GK SRS | 145 | 35 (24.1%) | 110 (75.9%) | |
| WBRT | 26 | 13 (50%) | 13 (50%) | |
| GK SRS+WBRT | 21 | 8 (38.1%) | 13 (61.9%) |
Table 2.
Characteristics of patients diagnosed with brain metastases
| Brain Metastasis Cohort n=213 n (%) / median (IQR) | High Risk n=42 | Moderate Risk n=135 | Low Risk n=36 | p-value | |
|---|---|---|---|---|---|
| n (%) / median (IQR) | n (%) / median (IQR) | n (%) / median (IQR) | |||
| Age | 67.0 (60.0-73.0) | 66.0 (56.0-73.0) | 67.0 (61.0-72.0) | 67.5 (60.5-74.5) | 0.39 |
| Sex | 0.632 | ||||
| Male | 100 (47.0%) | 17 (17.0%) | 65 (65.0%) | 18 (18.0%) | |
| Female | 113 (53.1%) | 25 (25.1%) | 70 (62.0%) | 18 (15.9%) | |
| Ethnicity | 0.636 | ||||
| White | 185 (86.9%) | 37 (20.0%) | 115 (62.2%) | 33 (17.8%) | |
| African American/Black | 26 (12.2%) | 4 (15.38%) | 19 (73.1%) | 3 (11.5%) | |
| Other | 2 (0.9%) | 1 (50.0%) | 1 (50.0%) | 0 (0.0%) | |
| KPS | 0.77 | ||||
| 100 | 3 (1.4%) | 0 (0.0%) | 2 (66.7%) | 1 (33.3%) | |
| 90 | 47 (22.2%) | 9 (19.2%) | 28 (59.6%) | 10 (21.3%) | |
| 80 | 92 (43.4%) | 16 (17.4%) | 61 (66.3%) | 15 (16.3%) | |
| 70 | 51 (24.1%) | 12 (23.5%) | 31 (60.8%) | 8 (15.7%) | |
| 60 | 14 (6.6%) | 5 (35.7%) | 8 (57.1%) | 1 (7.1%) | |
| 50 | 3 (1.4%) | 0 (0.0%) | 3 (100%) | 0 (0.0%) | |
| 40 | 2 (0.9%) | 0 (0.0%) | 2 (100%) | 0 (0.0%) | |
| Actionable Mutations | |||||
| EGFR | 0.024 | ||||
| Yes | 49 (23.0%) | 3 (6.1%) | 36 (73.5%) | 10 (20.4%) | |
| No | 164 (77.0%) | 39 (23.8%) | 99 (60.4%) | 26 (15.9%) | |
| KRAS | 0.022 | ||||
| Yes | 66 (31.0%) | 20 (30.3%) | 34 (51.5%) | 12 (18.2%) | |
| No | 147 (69.0%) | 22 (15.0%) | 101 (68.7%) | 24 (16.3%) | |
| ALK | 0.654 | ||||
| Yes | 13 (6.1%) | 3 (23.1%) | 9 (69.2%) | 1 (7.7%) | |
| No | 200 (93.9%) | 39 (19.5%) | 126 (63.0%) | 35 (17.5%) | |
| ROS1 | 0.223 | ||||
| Yes | 9 (4.2%) | 0 (0.0%) | 8 (88.9%) | 1 (11.1%) | |
| No | 204 (95.8%) | 42 (20.6%) | 127 (62.5%) | 35 (17.2%) | |
| BRAF | 0.576 | ||||
| Yes | 20 (9.4%) | 4 (20.0%) | 11 (55.0%) | 5 (25.0%) | |
| No | 193 (90.6%) | 38 (19.7%) | 124 (64.3%) | 31 (16.1%) | |
| Smoking Status | 0.662 | ||||
| Former | 131 (61.5%) | 26 (19.9%) | 84 (64.1%) | 21 (16.0%) | |
| Current | 50 (23.5%) | 11 (22.0%) | 28 (56.0%) | 11 (22.0%) | |
| Never | 32 (15.0%) | 5 (15.6%) | 23 (71.9%) | 4 (12.5%) | |
| Stage | 0.756 | ||||
| I | 2 (0.9%) | 0 (0.0%) | 2 (100.0%) | 0 (0.0%) | |
| II | 0 (0.0%) | 0 (0.0%) | 0 (0.0%) | 0 (0.0%) | |
| III | 7 (3.3%) | 1 (14.3%) | 4 (57.1%) | 2 (28.6%) | |
| IV | 204 (95.8%) | 41 (20.1%) | 129 (63.2%) | 34 (16.7%) | |
| NSCLC Histology | 0.869 | ||||
| Adenocarcinoma | 166 (77.9%) | 35 (21.1%) | 103 (62.1%) | 28 (16.9%) | |
| Squamous cell carcinoma | 26 (12.2%) | 3 (11.5%) | 19 (69.2%) | 5 (19.2%) | |
| Large cell carcinoma | 2 (0.9%) | 0 (0.0%) | 2 (100.0%) | 0 (0.0%) | |
| NSCLC-NOS | 19 (8.9%) | 4 (21.1%) | 4 (21.1%) | 3 (15.8%) |
Genes associated with neurologic death
Genetic mutations in the NGS panel found to be associated with neurologic death in the entire cohort and in the subset of patients who developed brain metastases are detailed in Table 3. Deleterious genes common across both cohorts were CDK4, PALB2, and GNAQ. Protective genes in common across both cohorts were CDK12, JAK2, TERT, CCND1, CHEK2, EZH2, HRAS, IDH2, JAK3, MAPK1, ARAF, IDH1, NPM1, GNA11, MPL, and MAP2K1. For the entire cohort, an additional protective gene, VHL, was identified and two additional deleterious genes, STK11 and RAF1, were identified. For the subset of patients who developed brain metastases, two additional deleterious genes, PTPN11 and NTRK3, were identified.
Table 3.
Genes associated with neurologic death
| Entire Cohort | Brain Metastases Patients | |||||
|---|---|---|---|---|---|---|
| Gene | Incidence n (%) | Hazard Ratio | P-Value | Incidence n (%) | Hazard Ratio | P-Value |
| CDK4 | 10 (3.26%) | 3.569 | 0.0008 | 9 (4.23%) | 712 | 0.007 |
| PALB2 | 5 (1.63%) | 4.219 | 0.015 | 5 (2.35%) | 2.776 | 0.087 |
| GNAQ | 3 (0.98%) | 5.238 | 0.025 | 3 (1.41%) | 3.465 | 0.097 |
| STK11 | 32 (10.42%) | 1.965 | 0.042 | N/A | N/A | N/A |
| RAF1 | 7 (2.28%) | 2.855 | 0.093 | N/A | N/A | N/A |
| PTPN11 | N/A | N/A | N/A | 1 (0.47%) | 3.073 | < 0.001 |
| NTRK3 | N/A | N/A | N/A | 13 (6.1%) | 2.145 | 0.087 |
| CDK12 | 9 (2.93%) | < 0.001 | < 0.001 | 3 (1.41%) | < 0.001 | < 0.001 |
| JAK2 | 10 (3.26%) | < 0.001 | < 0.001 | 8 (3.76%) | < 0.001 | < 0.001 |
| TERT | 13 (4.23%) | < 0.001 | < 0.001 | 7 (3.29%) | < 0.001 | < 0.001 |
| CCND1 | 8 (2.61%) | < 0.001 | < 0.001 | 6 (2.82%) | < 0.001 | < 0.001 |
| CHEK2 | 7 (2.28%) | < 0.001 | < 0.001 | 6 (2.82%) | < 0.001 | < 0.001 |
| EZH2 | 7 (2.28%) | < 0.001 | < 0.001 | 6 (2.82%) | < 0.001 | < 0.001 |
| HRAS | 5 (1.63%) | < 0.001 | < 0.001 | 2 (0.94%) | < 0.001 | < 0.001 |
| IDH2 | 4 (1.3%) | < 0.001 | < 0.001 | 4 (1.88%) | < 0.001 | < 0.001 |
| JAK3 | 3 (0.98%) | < 0.001 | < 0.001 | 1 (0.47%) | < 0.001 | < 0.001 |
| MAPK1 | 2 (0.65%) | < 0.001 | < 0.001 | 1 (0.47%) | < 0.001 | < 0.001 |
| ARAF | 2 (0.65%) | < 0.001 | < 0.001 | 1 (0.47%) | < 0.001 | < 0.001 |
| IDH1 | 2 (0.65%) | < 0.001 | < 0.001 | 1 (0.47%) | < 0.001 | < 0.001 |
| VHL | 1 (0.33%) | < 0.001 | < 0.001 | 1 (0.47%) | N/A | N/A |
| NPM1 | 1 (0.33%) | < 0.001 | < 0.001 | 1 (0.47%) | < 0.001 | < 0.001 |
| GNA11 | 1 (0.33%) | < 0.001 | < 0.001 | 1 (0.47%) | < 0.001 | < 0.001 |
| MPL | 1 (0.33%) | < 0.001 | < 0.001 | 1 (0.47%) | < 0.001 | < 0.001 |
| MAP2K1 | 1 (0.33%) | < 0.001 | < 0.001 | 1 (0.47%) | < 0.001 | < 0.001 |
Risk stratification scores for neurologic death
The genomic risk score that was constructed using the 22 genes described above had a score that ranged from − 3 to + 2, and this score was significantly associated with the risk of neurologic death with a HR of 3.76 (95% CI 2.72–5.18, p < 0.001), in the competing risk analysis using non-neurologic death as a competing risk. When the genomic risk score was categorized into three groups: high (risk score > 0), moderate (risk score = 0) and low (risk score < 0), we found that 45, 202, and 60 participants fell into these three categories, respectively. When comparing these three risk groups, we found the cumulative incidence of neurologic death in high, moderate, and low risk groups was 49.0%, 20.3% and 0% after 82.5 months (when the last observed neurologic death occurred), p < 0.001 (Fig. 1), with 3-year overall survival (OS) rates of 35.3%, 52.8%, and 59.1%, p = 0.047 (log-rank test), respectively.
Fig. 1.
Probability of neurologic death among all patients by genomic score risk groups, low risk (blue), moderate risk (red) and high risk (green) p < 0.001
Similar to the entire cohort, in the subset of patients who developed brain metastases, the genomic risk score was significantly associated with the risk of neurologic death with a HR of 2.87 (95% CI 2.09–3.94, p < 0.001). The cumulative incidence of neurologic death in the high, moderate, and low risk groups was 52.4%, 30.4%, and 0% after 82.5 months (when the last observed neurologic death occurred), p < 0.001 (Fig. 2), with 3-year OS rates of 30.6%, 36.4%, and 40.4%, p = 0.29, respectively.
Fig. 2.
Probability of neurologic death among patients with brain metastases for genomic score risk groups, low risk (blue), moderate risk (red) and high risk (green) p < 0.001
In additional exploratory analyses, the genetic risk score was associated with OS, with significant differences in survival across the three risk groups, p = 0.047 (log-rank test; Fig. 3). When stratified by patient subgroups both with and without targetable mutations (EGFR, ROS1, BRAF, NTRK3and ALK), the genetic risk score remained significantly associated with neurologic death. While a distinct separation was seen between all three risk groups in the patients with no targetable mutations, this distinction was observed only between the low-risk group and the combined moderate- and high-risk groups in the subgroup with one or more targetable mutations. Finally, the genetic risk score remained an independent predictor of neurologic death when adjusting for patient characteristics, including KPS, number of brain metastases, overall disease burden (none, oligo-metastatic, widespread), and type of brain irradiation (WBRT, GKRS, both, or none), p < 0.001 (Wald test).
Fig. 3.
Kaplan-Meier estimates of overall survival by genomic score risk groups, low risk (blue), moderate risk (red) and high risk (green) p = 0.047
Discussion
Comprehensive genomic profiling has been used recently as a tool for risk stratification of patients with metastatic cancer. In its early forms, CGP was predominantly used to identify actionable mutations for which patients could be triaged to systemic therapies targeting these specific driver mutations. Successes of CGP in NSCLC include treatment of EGFR, ALK, ROS1, RET, KRAS and BRAF-mutated cancers with specific small molecule inhibitors [18].
Glynn et al. published a study demonstrating that CGP done at time of first metastasis in NSCLC could be used to risk-stratify patients based on their risk of developing brain metastases [13]. Abdulhaleem et al. published a study demonstrating that CGP could also be used to identify the risk of brain metastasis phenotypes, such as large symptomatic lesions, small numerous lesions, and metastasis subtypes that would rapidly re-seed the brain [14]. These proof-of-principle studies, demonstrating that brain metastasis phenotype is largely influenced by genetics, led our team to hypothesize that brain metastases leading to neurologic death was likely another phenotype with a genetic foundation.
Neurologic death from brain metastasis can be caused by several events, including local failure of resistant metastases, unrelenting re-seeding of the brain, hemorrhage of metastases, leptomeningeal dissemination, and long-term neurologic sequela from the original presentation of the brain metastasis. Prior preclinical investigations have demonstrated that certain genetic variants are more radioresistant [19] or more likely to lead to leptomeningeal invasion [20]. Some metastases are more likely to become hemorrhagic, strongly suggesting a genetic causation of this phenotype as well [21]. A seminal publication by Brastianos et al. demonstrated that specific mutations were more likely to be found in brain metastases when evaluating multiple primary cancer types, suggesting that these genetic changes were instrumental in the cancer types more readily seeding the brain [22].
In spite of the several mechanisms leading to neurologic death, the implication of likely neurologic death may be similar. Brain metastases can have a heterogeneous presentation depending on how early or late in the natural history they develop. Patients who develop brain metastases as an end-stage event have a low risk of neurologic death due to competing risk of death from other causes; these patients may not benefit from aggressive CNS-directed therapy. The QUARTZ study showed that in NSCLC patients with poor performance status, there was no benefit from treating brain metastases with whole brain radiotherapy (WBRT) [23]. Those who have a higher risk of neurologic death, either due to well-controlled extracranial disease or more poorly controlled intracranial disease, may benefit from more aggressive CNS-directed therapy [24]. Those with a lower risk of neurologic death may benefit from de-escalation, delayed treatment, or no treatment at all to avoid treatment related toxicities [25].
Biomarkers for brain metastasis behavior have traditionally been challenging to find due to the heterogeneity of the brain metastasis population [26]. However, it has long been known that NSCLC patients with driver mutations EGFR and ALK had a greater risk of developing brain metastases and developed them earlier in their natural history than their wild-type counterparts [27]. In this study, neither of these mutations was more common in the high-risk group than in the lower or moderate ones, and in fact the incidence of EGFR mutations was lower in the high-risk group. This may be due to improved outcomes with the escalation of therapy with blood-brain barrier penetrant agents for patients harboring these mutations, in addition to traditional local therapy approaches [28].
A potential use for biomarkers is the optimization of medical decision-making. Radiation oncologists have been among the first to attempt to use CGP for risk stratification in order to triage patients to the most efficient radiation treatment. Choi et al. demonstrated that the likelihood of oligometastatic versus widespread failure could be risk stratified in NSCLC using liquid-biopsy based CGP [29]. The predictive value of CGP was highly statistically significant and the findings were later validated by an independent dataset, thus suggesting that this risk stratification score could be used to recommend patients with risk of oligometastatic failure undergo ablative radiation [30].
Biomarker-based decision making for the radiotherapeutic management of brain metastases may have positive downstream effects on the cost of management. A recent study by Shenker et al. found that the decision that most affected cost of care for brain metastasis patients is the upfront decision of stereotactic radiosurgery vs. WBRT [31]. Prior analyses have demonstrated that nomogram use may decrease the cost of management in brain metastasis patients [32]. Genetic biomarkers are more refined than traditional nomograms, and as such they may offer greater potential in cost reduction and improved outcomes [33]. At the present time, biomarker discovery for brain metastases is focusing on tumor tissue biomarkers [34], circulating tumor cells and DNA, as well as extracellular vesicles [35]. The advantage of CGP acquired by liquid-biopsy in the present study is its non-invasiveness, commercial availability, and CLIA-certified standards.
There are several limitations to the present dataset. As a retrospective analysis, several biases could exist, including patient selection bias. Validation with an independent dataset would be needed prior to this risk stratification system becoming operationalized clinically for triage of patients. The rapid pace of the development of CNS-penetrant systemic therapy agents is also a limitation, as new agents targeted towards mutations in the signature may impact its prognostic value. However, in spite of the hypothesis-generating nature of retrospective datasets, this study serves as an important proof of concept that CGP can risk-stratify NSCLC patients as a whole, and among those with brain metastases with regard to their likelihood of dying from neurologic causes without requiring invasive biopsy.
Conclusions
The development of a liquid-biopsy based genomic signature to predict neurologic death in NSCLC patients appears feasible. If properly validated, the risk stratification system may serve as a basis for clinical trial design for triaging NSCLC patients to escalated versus de-intensified CNS-directed management, such as more frequent surveillance imaging, radiation dose escalation, or use of CNS-penetrating systemic therapies.
Author contributions
Conceptualization, M.C., C.A.H., M.F., J.R., T.L., W.J.P., R.D.J., C.K.C., S.B.T., A.W.L. and J.J.W.; methodology, R.D.J., M.F., Y.W., M.S., F.X., J.S., Y.J., D.R.S.-P., M.C. and C.A.H.; formal analysis, R.D.J., M.C., C.A.H., S.E.G., J.S., C.T.W. and D.R.S.-P.; data curation, S.E.G., C.M.L., A.R.C., M.A., M.C., F.X., R.D.J., P.Y. and C.A.H.; writing—original draft preparation, S.E.G., C.A.H. and M.C.; writing—review and editing, S.E.G., C.A.H., M.C., R.D.J. and S.B.T.; supervision, C.A.H. All authors have read and agreed to the published version of the manuscript.
Funding
Open access funding provided by the Carolinas Consortium. The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.
Data availability
The data presented in this study are available on request from the corresponding author due to privacy and legal restrictions.
Declarations
Human ethics
Institutional Review Board Statement: The study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board of Wake Forest Health Sciences (IRB00033881 approved 1/23/12, renewed 10/28/25).
Consent to participate
Informed Consent Statement: Patient consent was waived due to the retrospective nature of this medical record study because the research posed minimal risk to the subjects, and the findings would not impact their welfare or provide any direct benefit. The study was conducted using de-identified data, to ensure the privacy and confidentiality of all participants.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.Soike MH, Hughes RT, Farris M, McTyre ER, Cramer CK, Bourland JD et al (2019) Does Stereotactic Radiosurgery Have a Role in the Management of Patients Presenting With 4 or More Brain. Metastases? Neurosurg 84(3):558–566 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Patchell RA, Tibbs PA, Walsh JW, Dempsey RJ, Maruyama Y, Kryscio RJ et al (1990) A randomized trial of surgery in the treatment of single metastases to the brain. N Engl J Med 322(8):494–500 [DOI] [PubMed] [Google Scholar]
- 3.Lester SC, Taksler GB, Kuremsky JG, Lucas JT Jr., Ayala-Peacock DN, Randolph DM 2 et al (2014) Clinical and economic outcomes of patients with brain metastases based on symptoms: an argument for routine brain screening of those treated with upfront radiosurgery. Cancer 120(3):433–441 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Lucas JT Jr., HGt C, White L, Fitzgerald N, Isom S, Bourland JD et al (2015) Competing Risk Analysis of Neurologic versus Nonneurologic Death in Patients Undergoing Radiosurgical Salvage After Whole-Brain Radiation Therapy Failure: Who Actually Dies of Their Brain Metastases? Int J Radiat Oncol Biol Phys 92(5):1008–1015 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Lanier CM, Hughes R, Ahmed T, LeCompte M, Masters AH, Petty WJ et al (2019) Immunotherapy is associated with improved survival and decreased neurologic death after SRS for brain metastases from lung and melanoma primaries. Neurooncol Pract 6(5):402–409 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Johnson AG, Ruiz J, Hughes R, Page BR, Isom S, Lucas JT et al (2015) Impact of systemic targeted agents on the clinical outcomes of patients with brain metastases. Oncotarget 6(22):18945–18955 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Chan M, Tatter S, Chiang V, Fecci P, Strowd R, Prabhu S et al (2023) Efficacy of laser interstitial thermal therapy for biopsy-proven radiation necrosis in radiographically recurrent brain metastases. Neurooncol Adv 5(1):vdad031 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Dohm AE, Hughes R, Wheless W, Lecompte M, Lanier C, Ruiz J et al (2018) Surgical resection and postoperative radiosurgery versus staged radiosurgery for large brain metastases. J Neurooncol 140(3):749–756 [DOI] [PubMed] [Google Scholar]
- 9.Kocher M, Soffietti R, Abacioglu U, Villa S, Fauchon F, Baumert BG et al (2011) Adjuvant whole-brain radiotherapy versus observation after radiosurgery or surgical resection of one to three cerebral metastases: results of the EORTC 22952–26001 study. J Clin Oncol 29(2):134–141 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Ayala-Peacock DN, Attia A, Braunstein SE, Ahluwalia MS, Hepel J, Chung C et al (2017) Prediction of new brain metastases after radiosurgery: validation and analysis of performance of a multi-institutional nomogram. J Neuro-Oncol 135(2):403–411 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.McTyre ER, Soike MH, Farris M, Ayala-Peacock DN, Hepel JT, Page BR et al (2020) Multi-institutional validation of brain metastasis velocity, a recently defined predictor of outcomes following stereotactic radiosurgery. Radiother Oncol 142:168–174 [DOI] [PubMed] [Google Scholar]
- 12.Kilburn JM, Ellis TL, Lovato JF, Urbanic JJ, Bourland JD, Munley MT et al (2014) Local control and toxicity outcomes in brainstem metastases treated with single fraction radiosurgery: is there a volume threshold for toxicity? J Neurooncol 117(1):167–174 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Glynn SE, Lanier CM, Choi AR, D’Agostino R Jr., Farris M, Abdulhaleem M et al (2025) Genomic signature for Initial Brain Metastasis Velocity (iBMV) in non-small-cell lung cancer patients: the elusive biomarker to predict the development of brain metastases? Cancers (Basel). 17(6) [DOI] [PMC free article] [PubMed]
- 14.Abdulhaleem M, Hunting JC, Wang Y, Smith MR, Agostino RJ, Lycan T et al (2023) Use of comprehensive genomic profiling for biomarker discovery for the management of non-small cell lung cancer brain metastases. Front Oncol 13:1214126 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.McTyre ER, Johnson AG, Ruiz J, Isom S, Lucas JT, Hinson WH et al (2017) Predictors of neurologic and nonneurologic death in patients with brain metastasis initially treated with upfront stereotactic radiosurgery without whole-brain radiation therapy. Neurooncology 19(4):558–566 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Patchell RA, Tibbs PA, Regine WF, Dempsey RJ, Mohiuddin M, Kryscio RJ et al (1998) Postoperative radiotherapy in the treatment of single metastases to the brain: a randomized trial. JAMA 280(17):1485–1489 [DOI] [PubMed] [Google Scholar]
- 17.Leighl NB, Page RD, Raymond VM, Daniel DB, Divers SG, Reckamp KL et al (2019) Clinical Utility of Comprehensive Cell-free DNA Analysis to Identify Genomic Biomarkers in Patients with Newly Diagnosed Metastatic Non-small Cell Lung Cancer. Clin Cancer Res 25(15):4691–4700 [DOI] [PubMed] [Google Scholar]
- 18.Villalva C, Duranton-Tanneur V, Guilloteau K, Burel-Vandenbos F, Wager M, Doyen J et al (2013) EGFR, KRAS, BRAF, and HER-2 molecular status in brain metastases from 77 NSCLC patients. Cancer Med 2(3):296–304 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Johung KL, Yao X, Li F, Yu JB, Gettinger SN, Goldberg S et al (2013) A clinical model for identifying radiosensitive tumor genotypes in non-small cell lung cancer. Clin Cancer Res 19(19):5523–5532 [DOI] [PubMed] [Google Scholar]
- 20.Abdulhaleem M, Ruiz J, O’Neill S, Hughes RT, Qasem S, Strowd RE et al (2023) Collagen deposition within brain metastases is associated with leptomeningeal failure after cavity-directed radiosurgery. Neurooncol Adv 5(1):vdac186 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Neal MT, Chan MD, Lucas JT Jr., Loganathan A, Dillingham C, Pan E et al (2014) Predictors of survival, neurologic death, local failure, and distant failure after gamma knife radiosurgery for melanoma brain metastases. World Neurosurg 82(6):1250–1255 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Brastianos PK, Carter SL, Santagata S, Cahill DP, Taylor-Weiner A, Jones RT et al (2015) Genomic Characterization of Brain Metastases Reveals Branched Evolution and Potential Therapeutic Targets. Cancer Discov 5(11):1164–1177 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Mulvenna P, Nankivell M, Barton R, Faivre-Finn C, Wilson P, McColl E et al (2016) Dexamethasone and supportive care with or without whole brain radiotherapy in treating patients with non-small cell lung cancer with brain metastases unsuitable for resection or stereotactic radiotherapy (QUARTZ): results from a phase 3, non-inferiority, randomised trial. Lancet 388(10055):2004–2014 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.McTyre E, Ayala-Peacock D, Contessa J, Corso C, Chiang V, Chung C et al (2018) Multi-institutional competing risks analysis of distant brain failure and salvage patterns after upfront radiosurgery without whole brain radiotherapy for brain metastasis. Ann Oncol 29(2):497–503 [DOI] [PubMed] [Google Scholar]
- 25.Cramer CK, McKee N, Case LD, Chan MD, Cummings TL, Lesser GJ et al (2019) Mild cognitive impairment in long-term brain tumor survivors following brain irradiation. J Neurooncol 141(1):235–244 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Su J, Song Q, Qasem S, O’Neill S, Lee J, Furdui CM et al (2020) Multi-omics analysis of brain metastasis outcomes following craniotomy. Front Oncol 10:615472 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Nishino M, Soejima K, Mitsudomi T (2019) Brain metastases in oncogene-driven non-small cell lung cancer. Transl Lung Cancer Res 8(Suppl 3):S298–S307 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Pike LRG, Miao E, Boe LA, Patil T, Imber BS, Myall NJ et al (2024) Tyrosine Kinase Inhibitors With and Without Up-Front Stereotactic Radiosurgery for Brain Metastases From EGFR and ALK Oncogene-Driven Non-Small Cell Lung Cancer (TURBO-NSCLC). J Clin Oncol 42(30):3606–3617 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Choi AR, D’Agostino RB Jr., Farris MK, Abdulhaleem M, Hunting JC, Wang Y et al (2024) Genomic signature for oligometastatic disease in non-small cell lung cancer patients with brain metastases. Front Endocrinol (Lausanne) 15:1364021 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Choi AR, Lanier CM, Glynn SE, D’Agostino R, Farris M, Abdulhaleem M et al (2024) External Validation of Oligometastatic Genomic Signaturein Non-Small Cell Lung Cancer Patients. Int J Radiat Oncol 120(2):S95–S [Google Scholar]
- 31.Shenker RF, McTyre ER, Taksler GB, D’Agostino RB, Cramer CK, Ruiz J et al (2019) Analysis of the drivers of cost of management when patients with brain metastases are treated with upfront radiosurgery. Clin Neurol Neurosur 176:10–14 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Abdulhaleem M, Ruiz J, Cramer C, Xing F, Lo HW, Su J et al (2022) Brain metastasis prognostic nomograms and brain metastasis velocity: a narrative review. Chin Clin Oncol 11(2):10 [DOI] [PubMed] [Google Scholar]
- 33.Signorovitch J, Zhou Z, Ryan J, Anhorn R, Chawla A (2019) Budget impact analysis of comprehensive genomic profiling in patients with advanced non-small cell lung cancer. J Med Econ 22(2):140–150 [DOI] [PubMed] [Google Scholar]
- 34.Song Q, Ruiz J, Xing F, Lo HW, Craddock L, Pullikuth AK et al (2023) Single-cell sequencing reveals the landscape of the human brain metastatic microenvironment. Commun Biol 6(1):760 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Kalita-de Croft P, Joshi V, Saunus JM, Lakhani SR (2022) Emerging biomarkers for diagnosis, prevention and treatment of brain metastases-from biology to clinical utility. Diseases. 10(1) [DOI] [PMC free article] [PubMed]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The data presented in this study are available on request from the corresponding author due to privacy and legal restrictions.