A low percentage of metastases in deep brain and temporal lobe structures

Abstract

Background

Whole-brain radiotherapy (WBRT) in patients with brain metastases (BM) is associated with neurocognitive decline. Given its crucial role in learning and memory, efforts to mitigate this toxicity have mostly focused on sparing radiation to the hippocampus. We hypothesized that BM are not evenly distributed across the brain and that several additional areas may be avoided in WBRT based on a low risk of developing BM.

Methods

We contoured 2757 lesions in a large, single-institution database of patients with newly diagnosed BM. BM centroids were mapped onto a standard brain atlas of 55 anatomic subunits and the observed percentage of BM was compared with what would be expected based on that region’s volume. A region of interest (ROI) analysis was performed in a validation cohort of patients from 2 independent institutions using equivalence and one-sample hypothesis tests.

Results

The brainstem and bilateral thalami, hippocampi, parahippocampal gyri, amygdala, and temporal poles had a cumulative risk of harboring a BM centroid of 4.83% in the initial cohort. This ROI was tested in 157 patients from the validation cohort and was found to have a 4.1% risk of developing BM, which was statistically equivalent between the 2 groups (P < 1 × 10^–6, upper bound).

Conclusion

Several critical brain structures are at a low risk of developing BM. A risk-adapted approach to WBRT is worthy of further investigation and may mitigate the toxicities of conventional radiation.

Key Points.

1. Brain metastases do not present in a random spatial distribution.

2. Several areas might be safely avoided when treating with whole brain irradiation.

Importance of the Study.

How brain metastases are spatially distributed may affect the presence of neurologic symptoms, therapeutic approach, and outcomes of focal treatment. In the current study, we mapped thousands of lesions and found that metastases are not randomly distributed across the brain. This identified a continuous set of structures that have a relatively low incidence of brain metastases: the hippocampus, parahippocampal gyrus, amygdala, thalamus, brainstem, and temporal poles. Several ongoing randomized trials are evaluating selective avoidance of the hippocampus when treating with whole brain irradiation. Our data suggest that, in patients with brain metastases for whom treatment to the whole brain is indicated, a contiguous set of deep and temporal lobe structures might also be safely spared the toxic effects of radiation given the low likelihood for disease failure.

Brain metastases (BM) affect approximately 20% of all cancer patients, and despite historically poor outcomes, survival is improving.^1,2 Treatment for BM typically involves radiation therapy, and although radiosurgery is now a standard of care for patients with a limited number of brain metastases, whole brain radiotherapy (WBRT) remains the treatment of choice in patients with a large number of lesions. Over 200 000 patients are estimated to be diagnosed with BM and over half are estimated to receive WBRT each year.³ Yet, with longer survival there is greater emphasis on minimizing exposure to WBRT and its associated neurocognitive toxicities. In particular, hippocampal-avoidance (HA) WBRT showed promise in Radiation Therapy Oncology Group (RTOG) 0933, a phase II investigation utilizing intensity modulated radiation therapy (IMRT) to selectively spare the bilateral hippocampi.⁴ Results demonstrated that neurocognitive toxicities were improved with IMRT relative to historical results managed with conventional WBRT, in which the entire brain is treated with opposed lateral beams to a uniform dose, and this strategy is now being tested in several multi-institutional phase III randomized studies. One in particular, NRG CC001 (NCT# 02360215), randomizes eligible patients to receive WBRT with memantine hydrochloride either with or without hippocampal avoidance. The trial closed to accrual in early 2018 and preliminary results are in accord with the previously completed phase II study.⁵ Specifically, the time to neurocognitive failure (NCF) was significantly longer with HA-WBRT, and this effect followed a convincing time-dependent trend, with 58% experiencing NCF at 6 months with HA compared with 69.1% with traditional WBRT (P = 0.012).

In sparing the hippocampi from radiation, one might assume that patients would then be at an increased risk of developing metastatic foci in regions receiving a low radiation dose. On the contrary, local control appears to be maintained with IMRT and extensive interrogation of the hippocampus has revealed that it is at very low risk for developing BM (≤8.6% of metastases within 5 mm of the structure).^6–8 The finding that this critical brain structure can be safely spared the damaging effects of high-dose irradiation without compromising treatment efficacy is intriguing and opens the possibility that other areas might be at low risk of developing BM and could also be safely spared the toxicities of radiation with an approach of risk-adapted (RA) WBRT.

Many decades of experience have demonstrated that BM are not randomly distributed, and some studies have sought to quantify the increased and decreased risks of particular areas for harboring disease.^9–13 To our knowledge, no study has sought to identify a contiguous set of anatomic structures that have a low risk of developing BM and might be avoided with WBRT. To this end, we quantified the spatial distribution of BM from a large single institution database and identified areas of the brain at a low risk of metastatic involvement that might feasibly be spared with RA-WBRT. We then gathered a validation dataset from 2 separate institutions and performed an a priori region of interest (ROI) analysis to confirm this low risk.

Materials and Methods

Patient Population

Patients age ≥18 years at a single institution who were treated with conventional WBRT for a new diagnosis of BM between January 1, 2006 and December 31, 2015 were analyzed. All patients had a primary diagnosis of either breast or lung cancer. The finding of BM was either synchronous or metachronous with their primary diagnosis. A separate validation cohort was generated at 2 independent institutions with unmatched patient and tumor characteristics. The validation dataset consisted of patients age ≥18 years evaluated between May 1, 2004 and December 31, 2017 with BM from any primary cancer diagnosis. To be included in either database, patients could not have a prior diagnosis of BM and must have a T1-contrast (T1c) MRI acquisition that was obtained prior to any surgical resection (if one was performed).

Statistical Methods and ROI Selection

Each BM was manually contoured on the T1c MRI sequence. The T1c image was registered to the Montreal Neurological Institute (MNI) standard brain using an affine registration with 12 degrees of freedom. Each registration matrix was then applied to the respective BM contour and the centroid was mapped in MNI space to a common anatomic atlas (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/Atlases) modified to include 55 relevant brain regions. The centroid of every BM was mapped across the 55 brain regions. As described previously,¹³ the expected rate for each region was computed based on the assumption that each voxel within the brain was at an equal risk of involvement (ie, BM are distributed randomly). Therefore, the expected probability of a BM centroid falling within an ROI was directly proportional to the volume of the ROI. The observed rate of BM centroids was computed by determining the number of centroids intersecting within an ROI and dividing by the total number of centroids being analyzed. Therefore, the rates reported are the frequency of BM centroids within a given brain region rather than the rates of patients having a BM centroid within a region.

A 2-tailed, 1-sample t-test was performed for each ROI to compare the observed rate of BM centroids with the expected value. To adjust for multiple comparisons across the 55 regions, a Bonferroni correction was applied to an α = 0.05 such that statistical significance was reached at P ≤ 9.26 × 10⁻⁴. Regions that met this statistical threshold and those with ≤2% of the total number of centroids across all patients were reviewed. From these, a set of regions that were deemed clinically relevant and might be spared WBRT allowing for technical limitations of standard treatment planning techniques were selected for a single ROI to be further evaluated in the validation cohort. Baseline characteristics between the initial and validation cohorts were compared using chi-squared and 2-sample t-tests.

BM centroids in the validation cohort were mapped as described above. To test for equivalence in the frequency of BM centroids in the ROI derived from the validation cohort relative to the initial dataset, two one-sample equivalence testing was employed¹⁴ with an equivalence bound of ±5% (absolute difference). Both the upper and lower bound P-values are reported for completeness. Equal variance between the initial and validation cohorts was not assumed. The sample size to determine equivalence with 80% power at alpha = 0.05 was 135 patients.

Image processing and data analyses were performed using the Functional MRI of the Brain Software Library¹⁵ v5.0.10 and Matlab vR2017a. Additional data visualization was performed with 3D Slicer^16,17 (https://www.slicer.org) v4.8.1.

This study was approved by each institutional review board with consent being waived for retrospectively acquired data and obtained for patients whose data were collected prospectively.

Results

Patient Characteristics

From the initial dataset of 120 patients, 2757 BM were identified, although 84 lesions (3%) did not correlate with any structure in the standard atlas (eg, calvarium,ventricular) and were removed from the analysis. Histologies of both small cell lung cancer (SCLC) and non–small cell lung cancer (NSCLC) were included. There were 64 female patients (53%) with an average age of 59.3 years (range 31–82 y). The majority of patients (65.8%) had a primary diagnosis of NSCLC and one patient had an unknown BM histology due to a prior history of both breast cancer and NSCLC with no histologic confirmation of the intracranial diagnosis. The validation dataset consisted of a distinct population of 157 patients having 372 BM. Of these, 6 metastases were outside the atlas (1.6%) and were removed from the analysis. The majority of patients were again female (60%), with tumors of NSCLC histology (42%). There was no significant difference in sex or age between the 2 cohorts. Patients in the validation dataset tended to have fewer BM (P < 0.001) and have larger lesions (P = 0.002). Table 1 demonstrates baseline patient and tumor characteristics across these datasets.

Table 1.

Patient demographics

	Initial Database	Validation Database	Statistic
Sex			χ 2= 1.09, P = 0.297
Male	54 (45%)	62 (40%)
Female	64 (53%)	95 (60%)
Mean age, y (range)	59.3 (31‒82)	61 (25‒87)	P = 0.368
Mean BM per patient (SD)	22.9 (67.4)	2.4 (3.5)	P < 0.001
Mean BM volume, cc (range)	0.52 (0.003‒61.95)	2.35 (0.008‒61.1)	P = 0.002
Histology
Breast	31 (25.8%)	29 (18.5%)
NSCLC	79 (65.8%)	66 (42%)
SCLC	9 (7.5%)	10 (6.4%)
Melanoma	0	25 (15.9%)
Colorectal	0	8 (5.1%)
Others	0	8 (5.1%)
Unknown	1 (0.8%)	11 (7%)

Patient demographics for the initial (single institution) and validation (2 institutions) datasets.

Identification and Validation of a Low-Risk Region of Interest

The most frequently involved anatomic region associated with BM was the cerebellum (14.9%), and several regions had a very low (≤2%) frequency. The absolute number and frequency of BM centroids identified in each region were evaluated in tabular form (Table 2) and visualized in orthogonal views (Figure 1). In addition, the statistical relationship between the expected frequency (based on the volume of the brain region) and the observed frequency of BM centroids was formulated (Figure 2). A careful review of these low-risk regions revealed several contiguous central and deep structures. In particular, the bilateral hippocampi, parahippocampal gyri, amygdala, temporal poles, thalami, and the brainstem were collectively found to have a 4.83% combined relative risk of a BM centroid and these were combined into a single ROI for validation (Figure 3). Consistent with this low frequency of BM, the ROI analysis in the validation cohort identified a rate of 4.1% of BM centroids within the ROI, which was statistically equivalent relative to the initial dataset by two one-sample equivalence testing (upper bound P = 3.78 × 10⁻⁷, lower bound P = 3.66 × 10⁻³). Despite having less power to distinguish between anatomical regions, the relative rates of BM centroids in the validation cohort were similar those of the initial dataset (Supplementary Table 1 and Supplementary Figure 1). For example, the superior frontal gyrus, middle frontal gyrus, frontal pole, precentral gyrus, lateral occipital cortex, and cerebellum were all among the most frequently involved anatomic sites. In both datasets, the cerebellum was the most likely region in the brain to harbor a BM centroid.

Table 2.

Spatial quantification of BM distribution for 55 standard brain regions

Atlas Name	Volume (cc)	# Observed	Rate Observed	% Volume at Risk	z-Score	P-value
Left amygdala	56.07	0	0	0.0008	–1.4257	0.1540
Left accumbens	60.18	0	0	0.0003	–0.8394	0.4012
Right pallidum	58.46	0	0	0.0011	–1.7482	0.0804
Right hippocampus	135.96	0	0	0.0014	–1.9029	0.0571
Right accumbens	140.89	0	0	0.0002	–0.7429	0.4575
Left hippocampus	31.95	1	0.0004	0.0011	–1.1210	0.2623
Heschl’s gyrus	41.87	3	0.0011	0.0035	–2.0538	0.0400
Left pallidum	43.89	3	0.0011	0.0012	–0.0762	0.9393
Supracalcarine cortex	17.05	4	0.0015	0.0019	–0.4388	0.6608
Right amygdala	95.05	4	0.0015	0.0010	0.8271	0.4082
Frontal medial cortex	10.65	5	0.0019	0.0062	–2.8316	0.0046
Planum polare	25.19	6	0.0022	0.0053	–2.1831	0.0290
Right caudate	60.35	6	0.0022	0.0022	0.0194	0.9845
Subcallosal cortex*	13.10	7	0.0026	0.0087	–3.3716	0.0007
Planum temporale	31.42	8	0.0030	0.0063	–2.1646	0.0304
Left thalamus	28.45	8	0.0030	0.0062	–2.0983	0.0359
Right thalamus	69.76	8	0.0030	0.0058	–1.9350	0.0530
Left caudate	54.01	9	0.0034	0.0021	1.3868	0.1655
Cuneal cortex	6.17	10	0.0037	0.0072	–2.1106	0.0348
Left putamen	32.76	10	0.0037	0.0034	0.3115	0.7555
Frontal operculum cortex	26.34	14	0.0052	0.0046	0.4488	0.6536
Parietal operculum cortex	35.20	14	0.0052	0.0074	–1.3006	0.1934
Right putamen	59.66	14	0.0052	0.0034	1.6473	0.0995
Temporal fusiform cortex*	39.48	17	0.0064	0.0141	–3.4051	0.0007
Intracalcarine cortex	11.24	19	0.0071	0.0097	–1.3725	0.1699
Insular cortex*	0.48	20	0.0075	0.0159	–3.4699	0.0005
Parahippocampal gyrus*	17.69	20	0.0075	0.0217	–5.0394	< 0.0001
Temporal occipital fusiform cortex	19.57	22	0.0082	0.0107	–1.2604	0.2075
Central opercular cortex	47.85	27	0.0101	0.0113	–0.5913	0.5543
Angular gyrus	15.78	28	0.0105	0.0145	–1.7273	0.0841
Supramarginal gyrus*	4.06	30	0.0112	0.0263	–4.8642	< 0.0001
Frontal orbital cortex	25.74	32	0.0120	0.0193	–2.7621	0.0057
Superior temporal gyrus	2.14	33	0.0123	0.0138	–0.6578	0.5107
Occipital pole	42.03	37	0.0138	0.0230	–3.1557	0.0016
Inferior frontal gyrus	0.38	38	0.0142	0.0173	–1.2058	0.2279
Juxtapositional lobule cortex	3.88	38	0.0142	0.0094	2.6073	0.0091
Lingual gyrus	28.90	39	0.0146	0.0231	–2.9224	0.0035
Temporal pole	6.29	41	0.0153	0.0234	–2.7565	0.0058
Occipital fusiform gyrus	20.60	46	0.0172	0.0156	0.6629	0.5074
Brainstem*	69.65	47	0.0176	0.0297	–3.6787	0.0002
Paracingulate gyrus	6.18	48	0.0180	0.0180	–0.0115	0.9909
Inferior temporal gyrus	1.81	52	0.0195	0.0241	–1.5653	0.1175
Cingulate gyrus*	8.46	52	0.0195	0.0382	–5.0646	< 0.0001
Superior parietal lobule	9.68	58	0.0217	0.0175	1.6364	0.1018
Middle temporal gyrus	3.39	66	0.0247	0.0308	–1.8242	0.0681
Right cerebral white matter	60.86	82	0.0307	0.0330	–0.6845	0.4936
Left cerebral white matter	42.62	100	0.0374	0.0334	1.1503	0.2500
Postcentral gyrus	11.21	104	0.0389	0.0383	0.1633	0.8703
Precuneous cortex	13.46	116	0.0434	0.0331	2.9638	0.0030
Superior frontal gyrus*	2.08	126	0.0471	0.0328	4.1766	< 0.0001
Middle frontal gyrus*	2.46	153	0.0572	0.0321	7.3756	< 0.0001
Lateral occipital cortex	11.49	188	0.0703	0.0746	–0.8489	0.3959
Precentral gyrus*	1.98	211	0.0789	0.0522	6.2189	< 0.0001
Frontal pole	1.38	239	0.0894	0.0774	2.3336	0.0196
Cerebellum*	154.10	410	0.1534	0.0846	12.7780	< 0.0001

*Brain regions with a significantly (P ≤ 9.26 × 10⁻⁴) increased or decreased percentage of BM centroid involvement. List of 55 brain regions from a standard anatomic atlas that were evaluated in the initial, single-institution cohort. The regions are sorted by the relative rate of BM centroid involvement. The rate of involvement is compared with the expected rate assuming the risk of BM is dependent only on the total volume of the brain region being tested.

Fig. 1.

Spatial map of 55 brain regions color-coded for the absolute rate of observed BM. The frequency of a BM centroid within each of 55 brain regions from a standard atlas are mapped in orthogonal views. Light blue regions were associated with the lowest rate of BM centroid identification, while the light yellow were associated with the highest. At 14.9%, the cerebellum was found to harbor the highest frequency of BM centroids in the single-institution analysis. Color bar range is from zero to 8%.

Fig.2.

Region-wide statistical comparison of the observed and expected rates of BM based on region volume. The volume-based, expected rate of BM centroids within each of 55 brain regions was compared with the observed rate in the initial patient cohort. A waterfall plot sorted by the Z-statistic from this comparison is illustrated along with the Bonferroni-corrected upper and lower significance bounds (Z = ±3.113, dotted lines).

Fig. 3.

Experimental ROI defined by a low risk of harboring BM. From the initial single-institution dataset, the thalamus, brainstem and bilateral hippocampus, parahippocampal gyrus, amygdala, and temporal pole were found to be at a low (4.83%) risk of developing BM and combined into a single ROI (blue). This low risk of developing BM within the ROI was then validated in an independent dataset.

In light of the results of RTOG 0933 and NRG CC001, we specifically reviewed the estimated rate of failure within the hippocampus plus a 5 mm avoidance region surrounding this structure. This analysis determined that only 0.67% of all BM centroids were identified within this region, in 14 of 277 (5.05%) patients.

Discussion

Adapting the delivery of WBRT based on the risk of developing BM is a novel approach and this study identified a large area of central brain structures that might safely be spared full-dose radiotherapy. The toxic effects of WBRT have been well described^18,19 and have driven an attempt to employ radiosurgery in more cases.³ Still, over half of patients with BM receive WBRT in the US each year,³ which is typically delivered with the same technique and dose that have been employed for over 50 years.^20,21

Modern treatment planning techniques, such as IMRT and volumetric modulated arc therapy, have made it feasible to treat the majority of the brain while sparing individual anatomic structures. Because of the central role of the hippocampus in learning and memory, efforts have centered on sparing this critical structure using conformal techniques.⁸ A phase II study, RTOG 0933, delivered a dose of 3000 cGy over 10 daily fractions to the brain while limiting the dose covering the entire hippocampus bilaterally to no more than 900 cGy and the point dose within this volume could not exceed 1600 cGy.²² Patients completed standardized assessments of cognitive function and daily quality-of-life questionnaires. There was a significant improvement in the primary endpoint of the study, the Hopkins Verbal Learning Test–Revised Delayed Recall score at 4 months, in patients receiving HA-WBRT relative to historical controls who were treated with conventional WBRT.

Based on these findings, the hippocampus is the only region of the brain that is currently considered for avoidance. One proposed model of neurocognitive decline following exposure to radiation proposes distinct sensitivities of anatomic subunits.²³ However, there are currently no clear preclinical or clinical data to suggest that WBRT may damage specific neuroanatomic structures outside the hippocampus and lead to neurocognitive deficits. Without a clear understanding of which neural structures underlie the link between radiation exposure and neurologic impairments, we have taken the approach of defining areas at lowest risk of developing BM and used modern treatment planning techniques to maximize the volume of brain that may be spared. Further, beyond late neurocognitive impairments, which are the focus of several modern trials, conventional WBRT imparts dose to many other structures that are not considered at risk of developing BM. Several structures and their associated toxicities that might be spared with conformal radiotherapy include the globes (eg, scleritis, retinitis), lacrimal glands (eg, conjunctivitis), parotid glands (eg, xerostomia, parotiditis), and middle ear and vestibular apparatus (eg, otitis media, labyrinthitis). In addition to an improvement in late neurocognitive decline, we hypothesize that lower integral doses given with RA-WBRT may further reduce acute/subacute toxicities and improve quality of life.

We did not consider it technically feasible to spare every region of the brain with a low frequency of BM. In particular, the cingulate gyrus is a midline structure with many functions critical to cognition and emotion that was found to have a low rate of BM centroids (1.9%), significantly less than the volume-based prediction (Table 2). However, this region was enveloped by several other brain areas with a higher rate of BM, such as the frontal pole and precentral gyrus. Due to its close proximity to these high-risk areas and the lack of evidence suggesting that 30 Gy of radiation might significantly impair its function, the cingulate gyrus was not included in the experimental ROI.

The brainstem was included in our experimental ROI based on a low risk of a BM centroid and its physical location adjacent to the thalamus and medial temporal lobes, which allows for a conformal radiation dose gradient across these structures. It should be noted that, while the risk is relatively low, a metastasis within the brainstem may be associated with detrimental clinical sequelae and treatment toxicities. Therefore, our inclusion of the brainstem in this study’s avoidance region is based on experimental simulation and technical considerations, but these might not apply in a given clinical scenario.

A strength of this study is the disparate patient and tumor characteristics across the test and validation databases, which suggests the distribution of BM is related to other factors (such as cerebral vasculature) rather than patient selection. The initial dataset was taken from a single institution’s experience of patients where the majority were treated with up-front WBRT. Accordingly, the mean number of BM per patient within this group was high (~23 per patient), and this provided a large number of lesions by which the spatial distribution could be measured. This is also the patient population most likely to benefit from an improvement in WBRT planning, since they were likely ineligible for radiosurgery based on the high intracranial disease burden. Still, the use of a select patient population has the potential to bias conclusions, which is why we chose to validate our findings in an unselected cohort derived from 2 independent institutions. In the unselected validation cohort, the mean number of BM per patient was 2.4, which is a more balanced representation of cancer patients, particularly at centers with specialization in radiosurgery. This lowers the yield in data per patient for analysis, which is why we chose a combined ROI analysis to avoid limitations of power and multiple comparisons. We further demonstrated that the findings in the validation dataset were statistically equivalent to the single institution data.

It is important to note that we modeled the location of each brain metastasis based on its centroid and not the entire volume of the metastasis. This is advantageous because the centroid estimates the origin of each brain metastasis as a single voxel with the assumption that the growth of a malignant deposit approximates a sphere. By modeling the origin of each lesion, the risk of developing BM can be determined for each voxel and then mapped onto a standard brain atlas so the risk can be segmented into meaningful anatomic subunits. This is clinically valuable for patients with BM because treatment may be delivered to all involved areas while sparing the maximum volume of brain that is at a low risk of developing a subsequent metastasis or of harboring subradiographic disease.

In regard to patient selection, our analysis allowed patients with SCLC. We note that SCLC patients were not allowed to enroll on RTOG 0933 owing to a propensity for diffuse intracranial metastatic disease,⁴ although some data suggest a more modest risk of hippocampal involvement in patients with SCLC.²⁴ Our analysis was aimed at describing the distribution of BM in an unselected population, and the inclusion of SCLC patients would hypothetically skew the data toward a higher frequency of lesions within the experimental ROI. Therefore, we chose to include all lung cancer diagnoses, and SCLC comprised approximately 7% of patients within each of the 2 datasets.

In addition to its retrospective nature, a limitation of this study is the lack of neurocognitive assessments in these patients to compare results with dosimetric data. It is our presumption that sparing a larger volume of brain outside the hippocampus will improve neurocognitive outcomes, which may in turn improve quality of life,²⁵ but our data are unable to address this retrospectively. Prospective evaluation of RA-WBRT may confirm the low risk of failure in these regions and determine the effects on neurocognitive endpoints.

Funding

This study did not receive funding from third parties.

Conflicts of interest statement.

None declared.

Authorship statement.

Study concept and design: T.K.Y., R.Y.H.; data analysis, interpretation, implementation: T.K.Y., J.R.M., P.Y.W., H.A., A.A.A., T.J.C.W., R.Y.H.; database generation: J.R.M., N.J.G., A.W.L., V.S., M.E.W., K.T.H., A.S., C.W., D.Y.; manuscript development: all authors.

Prior presentations.

Parts of this work have been presented at the American Society for Radiation Oncology 59th and 60th Annual Meetings.

Financial disclosures:

TKY: Novocure (travel expenses, research funding)

AAA: Varian Medical Systems (research funding)

TJCW: AbbVie (consultant, travel expenses); AstraZeneca (travel expenses); Elekta (consultant, honoraria); Merck (consultant); Doximity (consultant, stock options); Wolters Kluwer (honoraria); Novocure (travel expenses)

RYH: Agios (research funding); Neuro-Oncology (editorial board)

PYW: Agios (research funding); AstraZeneca (research funding); Beigene (research funding); Eli Lilly (research funding); Genentech/Roche (research funding); Karyopharm (research funding); Kazia (research funding); MediciNova (research funding); Merck (research funding, speaker); Neuro-Oncology (editorial board); Novartis (research funding); Oncoceutics (research funding); Sanofi-Aventis (research funding); VBI Vaccines (research funding)