Emerging therapeutics and evolving assessment criteria for intracranial metastases in patients with oncogene-driven non-small-cell lung cancer

Abstract

The improved survival outcomes of patients with non-small-cell lung cancer (NSCLC), largely owing to improved control of systemic disease from immunotherapy and novel targeted therapies, have highlighted the challenges posed by CNS metastases as a devastating yet common complication, with up to 50% of patients developing such lesions during the course of the disease. Early generation tyrosine-kinase inhibitors (TKIs) often provide robust systemic disease control in patients with oncogene-driven NSCLC, although these agents are usually unable to accumulate to therapeutically relevant concentrations in the CNS owing to an inability to cross the blood–brain barrier (BBB). However, the past few years have seen a paradigm shift with the emergence of several novel or later-generation TKIs with improved CNS penetrance. Such agents have promising levels of activity against brain metastases, as demonstrated by data from preclinical and clinical studies. In this Review, we describe current preclinical and clinical evidence of the intracranial activity of TKIs targeting various oncogenic drivers in patients with NSCLC, with a focus on newer agents with enhanced CNS penetration, leptomeningeal disease (LMD), and the need for intrathecal treatment options. We also discuss evolving assessment criteria and regulatory considerations for future clinical investigations.

Introduction

Non-small-cell lung cancer (NSCLC) remains the leading cause of cancer-related mortality in the USA and worldwide. CNS metastases are common in patients with NSCLC, with 10–20% having such lesions at initial diagnosis^1,2 and up to 50% developing brain metastases over the course of the disease^3–5. CNS metastases substantially reduce patients’ quality of life, often owing to seizures, cognitive impairment and/or other forms of neurological dysfunction⁶. The prognosis of patients with brain-metastatic NSCLC is often poor, with median overall survival (OS) durations estimated to be around 17 months for patients with lung adenocarcinomas and 8 months for those with other subtypes of NSCLC^7,8. However, as treatment strategies evolve, the cumulative incidence of brain metastases will probably also change based on the extent of intracranial penetrance of systemic therapies.

The treatment of patients with intracranial metastases from NSCLC is highly multidisciplinary. Radiotherapy, including stereotactic radiosurgery (SRS) and whole-brain radiotherapy (WBRT), as well as surgical resection when feasible, have been the main treatment modalities for the management of brain metastases. Traditionally, systemic chemotherapy is less effective in patients with CNS metastases relative to those with metastases in other organs owing to the inability of most agents to cross the blood–brain barrier (BBB), typically resulting in intracranial response rates ranging from 15–50%^9,10. Immune-checkpoint inhibitors such as anti-PD-(L)1 agents have shown promising initial intracranial activity in patients with other advanced-stage solid tumours that are often associated with CNS metastases¹¹. For example, in melanoma, nivolumab in combination with ipilimumab have demonstrated 57% intracranial response among patients with untreated CNS metastases¹¹. Investigations designed to understand the extent of any CNS benefits of ICIs both with and without radiotherapy in patients with NSCLC are ongoing^12–14.

The number of systemic treatment options [to editor: the systemic options expanded, but CNS penetrating options are still limited] has expanded substantially with the emergence of therapies targeting specific oncogenes that drive different subtypes of NSCLC. Targeted therapies for nine such driver oncogenes have received regulatory approval for patients with oncogene-driven NSCLCs, most of which are small-molecule tyrosine-kinase inhibitors (TKIs)¹⁵. First-generation TKIs such as erlotinib and crizotinib demonstrated poor BBB permeability and limited activity against CNS metastases, although these agents were nonetheless associated with lower rates of CNS progression compared with chemotherapy^16–19. When used in combination with radiotherapy, early generation TKIs are well-tolerated and moderately effective²⁰, although the contribution of many of these TKIs to the overall efficacy of the combination can be difficult to determine. Later-generation TKIs such as osimertinib and lorlatinib have demonstrated marked improvements in CNS penetrance and have now entered clinical practice^21,22.

In this Review, we summarize the available preclinical evidence on the BBB penetrance of targeted therapies approved for use in patients with NSCLC harbouring specific driver alterations, as well as the clinical intracranial response rates of patients receiving these various agents. The evolving assessment criteria and regulatory considerations are also discussed, given that these criteria are increasingly used in trials and susbsequently in clinical practice. These criteria might also be applicable to CNS metastases from other solid tumours.

The blood-brain and blood-tumour barriers

The BBB consists of endothelial cells lining cerebral capillaries that are connected by tight junctions and are able to interact with astrocytes, pericytes and microganglia (FIG. 1a). Tight junctions generally restrict the paracellular diffusion of most molecules into the CNS. Polarized transport proteins expressed on the endothelial cells enable the selective transport of a range of substrates, including signalling proteins, carbohydrates, fatty acids, amino acids and ions, across the BBB, thus protecting the brain from pathogens and toxins while also ensuring access to essential nutrients. Active transport takes place by efflux via various transporters, such as the ATP-binding cassette (ABC), or by movement against the concentration gradients of other molecules²³. Small molecules that lack a specific transport protein must cross the BBB passively via paracellular or transcellular diffusion along their concentration gradient, which is limited by tight junctions for most molecules. Certain molecules are able undergo passive transcellular diffusion through endothelial cells to cross the BBB. Macromolecules (>400 Da) require carrier-mediated transport to cross the BBB. After internalization, molecules are excreted from the brain through active transporters (such as P-glycoproteins) with efflux pumps along the endothelium, entrance into the CSF or lymphatic system, or by enzymatic metabolism²⁴.

Figure 1. Schematic illustration of the (A) blood–brain barrier and (B) the blood–tumour barrier.

(A) The BBB consists of endothelial cells lining cerebral capillaries that are connected by tight junctions and interact with astrocytes, pericytes and microganglia. Transport proteins on the endothelial cells enable the selective transport of various substrates. Active transport takes place by efflux via transporters such as P-glycoprotein.

(B) In the case of brain metastases, the protective BBB becomes disrupted and transforms into a blood–tumour barrier (BTB) with increased permeability due to a modified basement membrane and loss of tight junctions (as a result of tumor-secreted VEGF and angiopoietin-2). Pericytes and astrocytes are often also lost, and transcytosis components are downregulated

In patients with brain metastases, the protective BBB becomes disrupted and transforms into a blood–tumour barrier (BTB) with modified and more-variable permeability²⁵. Loss of BBB tight junctions is typically seen in the endothelium of the BTB owing to increased vascular endothelial growth factor (VEGF) secretion by tumour cells and hypoxia-induced angiogenesis, causing increased fenestration and vascular permeability. Pericytes and astrocytes are often also lost, and transcytosis components are downregulated²⁵ (FIG. 1b). This loss of barrier function enables greater endocytic penetration of certain macromolecules compared to that seen with an intact BBB^25,26, and these macromolecules can include HER2-targeted antibody–drug conjugates (ADCs) and chemotherapies in mouse xenograft models^26,27. Data from such models indicate varying drug concentrations within the brain, across different metastases and even among different areas of the same lesion, which is likely to reflect variations in paracellular permeability²⁵.

The newer, rationally designed, CNS-penetrating drugs are optimized to facilitate uptake and minimize efflux across the BBB with the aim of maximizing intracranial retention. Most early generation TKIs are substrates of P-glycoprotein efflux transporters. Indeed, certain older TKIs such as crizotinib are very readily excreted via P-glycoprotein efflux pumps, resulting in subtherapeutic²⁸ intracranial concentrations and thus limited efficacy. Hence, later-generation TKIs such as osimertinib, alectinib and lorlatinib were specificially designed to bypass recognition by P-glycoproteins in order to limit their elimination from the CNS²⁹. Further understanding of the heterogenous permeability of the BTB is necessary to develop more-effective therapeutic strategies.

Preclinical quantification of CNS drug concentrations

Preclinical quantification of the CNS concentration of a drug is the first step in predicting BBB penetrance. Therefore, pharmacokinetic models specifically for this purpose have been developed in small animals (FIG. 1). Kp_uu, the unbound partition coefficient, is increasingly used to quantify drug accumulation in such studies³⁰. Kp_uu,brain, which indicates the unbound brain-to-plasma partition ratio, is defined as the ratio between the free drug concentration in the brain parenchyma and the free concentration of the same drug in plasma at a steady state^30,31. Similarly, Kp_uu,CSF, the cerebrospinal fluid (CSF)-to-plasma ratio, indicates the ratio of the unbound drug concentration in CSF versus plasma at equilibrium. CSF drug concentrations provide a fairly good estimate of brain extracellular fluid concentrations but are not an exact measure of parenchymal drug accumulation, as Kp_uu,CSFvalues generally tend to be slightly higher than Kp_uu,brain³². This difference is thought to be attributed to enhanced drug elimination from the brain parenchymal through active efflux transporters at the BBB³². While CSF is the most easily sampled CNS specimen in humans, it is worth noting that Kp_uu,CSF is not entirely representative of the intraparenchymal drug concentration. Values >0.3 reflect high levels of diffusion across the BBB^31,33. Transporter proteins expressed at the BBB and blood–CSF barrier, such as P-glycoprotein, which excrete drugs from the CSF via active efflux, have a major role in limiting CNS penetrance³¹. P-glycoprotein seems to have opposing functions when expressed at the BBB versus the blood-CSF barrier³⁴. The available Kp_uu,brain and Kp_uu,CSF data for various CNS-penetrant TKIs discussed throughout this review can be found in Table 1.

Table 1 │.

CNS accumulation and activity of targeted therapies in patients with brain-metastatic NSCLC

Drug	Kpuu brain	Kpuu, CSF	CNS ORR (%)	Median CNS DoR	CNS PFS
EGFR classical mutations
Gefitinib51,197	0.0021a	0.088a	65%	–	–
Erlotinib51,198	0.11a	0.29a	50%	–	–
Icotinib46,199	0.12a	0.61a	67.9%	–	Median 10.0 months
Afatinib51,200	0.0066a	0.27a	67%	–	–
Osimertinib21,48	0.21–0.39b	0.29a	91%	15.2 months	–
Lazertinib46,201	0.087a	–	85.7% in patients with EGFRT790M	15.1 months	Median 26.0 months
Zorifertinib52	0.65b	0.42b	75%	12.4 months	15.2 months
BLU-70157	0.56–0.98a	–	–	–	–
BDTX-153560	0.55–0.58a	–	–	–	–
EGFR exon 20
BLU-45169	0.66a	–	–	–	–
ALK
Crizotinib28	–	0.0026c	23–50%	3.7–10.2 months	–
Ceritinib81	0.15a	–	72.7%	16.6 months	–
Alectinib84,85	0.63–0.94a	–	81%	–	–
Brigatinib78,202	–	–	78%	–	2-year PFS 48%
Ensartinib87	–	–	63.7%	–	–
Lorlatinib92,203	–	0.75c	82% (first-line); 59% (third-line or later)	–	Median 24.6 months (third line or later)
NVL-65594	0.16a	–	–	–	–
ROS1
Entrectinib101	0.43b	–	80%	12.9 months	Median 8.8 months
Repotrectinib104	–	–	100% (TKI-naive); 75% (TKI-pretreated)	–	–
Taletrectinib204	–	–	91.6%	100%	–
Lorlatinib205,206	–	0.75c	66.7%	–	Median 38.8 months
RET
Selpercatinib110,207	0.20b	–	85%	9.4 months	Median 19.4 months
Pralsetinib208	–	–	70%	10.5 months	–
METex14
Capmatinib128	–	0.34a	54%	–	–
Tepotinib129,131	0.25a	–	55–86.7%	–	–
KRAS G12C
Sotorasib139	–	–	DCR 88%	–	–
Adagrasib141–142	–	0.47c	42%	12.7 months	Median 5.4 months
NTRK
Larotrectinib99	–	0.03a	63%	–	–
BRAF V600E
Dabrafenib plus trametinib	–	–	–	–	–
HER2
Trastuzumab deruxtecan	–	–	–	–	–

CSF, cerebrospinal fluid; DCR, disease control rate; DoR, duration of response; NR, not reported; ORR, objective response rate; PFS, progression-free survival; TKI, tyrosine-kinase inhibitor.

^aAs measured in rats.

^bAs measured in mice.

^cAs measured in humans.

CNS activity of targeted therapies

The range of targeted therapies available for patients with NSCLC has expanded substantially over the past 5 years. Targeted therapies are now available for nine different driver oncogenes with 10 actionable alterations, including EGFR aberrations (classical activating mutations and exon 20 insertions), ERBB2 (HER2) mutations, ALK, ROS1, RET or NTRK rearrangements, MET exon 14 skipping mutations, BRAF^V600E, or KRAS^G12C. Mostly small-molecule inhibitors are indicated for all of these alterations, except for HER2 mutations. For HER2-mutant NSCLC, antibody drug conjugate (ADC) trastuzumab deruxtecan was approved, which has recently demonstrated promising CNS activity in patients with both untreated and previously treated baseline brain metastases³⁵. The intracranial efficacy of certain small-molecule TKIs was initially attributed to their low molecular weight; however, larger molecules such as ADCs have also been demonstrated to penetrate the brain parenchyma, most likely owing to disruptions of the BBB when brain metastases occur^36,37. With the newer TKIs having unprecedented levels of intracranial efficacy, very good systemic treatment options are now available for patients with CNS metastasis from certain subtypes of NSCLC. These drugs, which include those targeting classical EGFR mutations (osimertinib), ALK rearrangements (alectinib, brigatinib and lorlatinib), RET rearrangements (selpercatinib), and ROS1 or NTRK rearrangements (entrecitinib), can produce CNS objective response rates (ORRs) ≥80% in each patient population (TABLE 1). However, an unmet need exists to develop small molecules with intracranial activity against other driver alterations and to establish intracranial efficacy of newer next-generation TKIs, such as EGFR exon 20 insertions, HER2 mutations, MET exon 14 skipping, BRAF^V600E and KRAS^G12C.

Classical EGFR mutations

The ability to inhibit EGFR not only set in motion the TKI renaissance that would later define precision medicine for NSCLC, but also provided the first demonstration of the importance of CNS activity. Brain metastases are found in about a quarter (24.4%) of patients with EGFR-mutant NSCLC at the time of a diagnosis of advanced-stage disease³⁸, with a 5-year cumulative incidence as high as 52.9%^38,39. First-generation and second-generation TKIs such as gefitinib, erlotinib and afatinib have poor BBB permeability (Kp_uu,brain <0.11) and thus limited activity in patients with CNS metastases^40,41. Early generation EGFR TKIs in combination with SRS or WBRT have demonstrated improved systemic control and time to CNS progression in patients with brain metastases, possibly as a result of the disrupted BTB⁴². One approach to augment BBB penetrance involves the ‘pulsed’ oral administration of high-dose (1,500 mg, which is 10 times the recommended daily dose) erlotinib once or twice per week, with or without daily continuous dosing. This approach is reportedly well-tolerated and enables improved CNS activity (ORR 67%; median CNS progression-free survival (PFS) 2.7 months)^{43 44}. These pulsed-dose studies illustrate the strategy of generating a higher peak concentration over a shorter duration in an effort to enhance the CNS drug concentration and thus promote intracranial disease control, as supported by clinical evidence and mathematical PK modeling⁴⁵.

Osimertinib is a third-generation EGFR TKI with activity against the T790M variant, as well as classical activating mutations, with a much improved BBB penetrance compared to that of earlier-generation TKIs. Osimertinib is currently the standard-of-care first-line therapy for patients with metastatic EGFR-mutant NSCLC. Osimertinib is a weak substrate for human BBB efflux transporters and, therefore, has a low BBB influx:efflux ratio and thus high CNS penetrance^46,47. Data from preclinical studies support this observation, with a Kp_uu,brain of 0.21–0.39 compared to <0.01 for afatinib and gefitinib (TABLE 1)⁴⁸. The superiority of osimertinib over first-generation TKIs has been established in the first-line setting through the phase III FLAURA trial, with an impressive intracranial ORR of 91% among patients with measurable asymptomatic or stable CNS metastases (included both untreated and pretreated)^21,49. FLAURA included patients with LMD (seen in 7 of 128 patients with CNS metastases) who historically have been excluded from most clinical trials.

Novel BBB-penetrant EGFR TKIs with even higher Kp_uu (brain and CSF) values are in clinical development (TABLE 1). For example, zorifertinib is an EGFR TKI designed to fully penetrate the BBB by evading efflux transporters. Impressive early pharmacokinetic results were achieved with this agent, which demonstrated Kp_uu,brain of 0.65 and Kp_uu,CSF of 0.42 in mice^51,52. Profound suppression of intracranial tumours and improved survival relative to that achieved with erlotinib have also been demonstrated in mouse models^51,53. Combining zorifertinib with radiation has been found to have more-potent antitumour effects in mouse models, by suppressing both cellular proliferation and DNA damage repair⁵⁴. Intracranial tumour regression and CSF tumour cell reductions >50% were observed in 4 of 5 patients with LMD receiving zorifertinib as part of a larger phase I study⁵⁵. Phase III data on the efficacy of this agent recently confirmed improved intracranial PFS of 15.2 months with zorifertinib versus 8.3 months in control group treated with gefitinib or erlotinib (HR 0.47, p<0.0001) (NCT03653546)⁵⁶. Another novel drug, BLU-701, is designed to target EGFR TKI-resistant tumours harbouring the C797S resistance mutation in EGFR and has demonstrated high levels of BBB penetration in rat models (Kp_uu,brain 0.56–0.98)^57,58. BDTX-1535, another developmental highly CNS-penetrant irreversible EGFR TKI, has also been shown to induce CNS tumour regression in preclinical models^59,60. A phase I study of BDTX-1535 in patients with advanced-stage EGFR-mutated glioblastoma and NSCLC is currently recruiting patients (NCT05256290)⁶¹.

EGFR exon 20 insertions

EGFR exon 20 insertions can be found in approximately 2–3% of all NSCLCs⁶² and are resistant to conventional EGFR TKIs⁶³. Two targeted agents, amivantamab and mobocertinib, are currently approved for this patient population^64,65. Amivantamab is a novel bispecific antibody targeting EGFR and MET that has shown good levels of systemic efficacy and tolerability in the phase I CHRYSALIS trial^66,67. However, intracranial activity is unknown owing to the exclusion of patients with untreated and/or active brain metastases⁶⁷. Mobocertinib is a TKI targeting EGFR exon 20 insertions that demonstrated an inferior ORR among patients with baseline brain metastases (25%) compared to those without (56%) in the phase II EXCLAIM study, suggesting limited CNS activity^65,68. Similar to CHRYSALIS, EXCLAIM only allowed the enrolment of patients with stable and treated brain metastases and excluded those with untreated or active lesions^65,67, thus making the CNS efficacy of mobocertinib very difficult to evaluate. On the basis of current data, both amivantimab and mobocertinib are considered to have minimal CNS activity.

Several novel TKIs targeting EGFR exon 20 insertions are currently at various stages of clinical investigation in an attempt to address this unmet need. BLU-451 is a highly CNS-penetrant inhibitor of exon 20 insertion-mutant EGFR (Kp_uu,brain 0.66)⁶⁹ that is currently being tested in a phase I/II study involving patients with EGFR exon 20 insertion-driven NSCLC and untreated brain metastases (NCT05241873)⁷⁰. Sunvozertinib is another new TKI that has received FDA breakthrough therapy designation for patients with NSCLC harbouring EGFR exon 20 insertions. This agent demonstrated an ORR of 48.5% among 33 patients with pretreated stable and asymptomatic brain metastases at study enrolment in a phase II study⁷¹. However, additional evaluations in patients with untreated active brain metastases are needed to better elucidate the intracranial activity of sunvozertinib. Finally, ORIC-114, a CNS-penetrant irreversible TKI with activity against both EGFR and HER2 exon 20 insertions, has shown promising preclinical intracranial activity in mouse models, with a phase I trial underway (NCT05315700)⁷².

HER2 mutations

HER2 mutations occur in about 3% of all NSCLCs, and are newest addition to the list of targetable driver alterations following the approval of T-DXd based on results from the DESTINY-lung trials³⁵. Because T-DXd is an ADC and a macromolecule, the trial allowed patients with irradiated and non-irradiated CNS metastases⁷¹. Nonetheless, T-DXd has demonstrated a CNS ORR of 73% in 15 patients with HER2-positive breast cancer with new or progressing CNS metastases⁷³. Data in patients with HER2-mutant NSCLC and CNS metastases in DESTINY-Lung01 will mature over time.

Various non-selective HER2 TKIs, such as afatinib, dacomitinib and pyrotinib, have demonstrated some systemic activity in patients with HER2-mutant NSCLC, although CNS-specific data are again unavailable. The pan-ErbB TKI poziotinib was evaluated in a cohort of pretreated patients with stable CNS metastases, demonstrating an ORR of 28.6% with a median PFS duration of 7.4 months. However, interpretation of these data is limited by the small sample size, the use of prior CNS radiation in 13 of the 14 patients (the majority within 3 months of study initiation), and the lack of a mandatory baseline brain MRI⁷⁴. BI-1810631 is a promising new HER2-selective TKI that that does not inhibit EGFR, showing a systemic ORR of 42% and disease control rate (DCR) of 95% in a phase I trial involving 19 patients. Further evaluations of the efficacy of this agent, including intracranial activity are ongoing (NCT04886804). Novel HER2-targeting ADCs are also in early clinical development for the treatment of HER2-driven lung cancers, including disitamab vedotin, SBT-6050, GQ-1001 and ZW-49.

ALK rearrangements

ALK rearrangements occur in approximately 3–7% of patients with NSCLC⁷⁵. Around 20–30% of patients with ALK-rearranged NSCLC will have CNS metastases at the time of diagnosis, and up to 50% will develop such lesions throughout the course of the disease^38,76, although this incidence will probably decline following the availability of CNS-penetrant TKIs. Crizotinib, the first FDA approved TKI for patients with ALK-rearranged NSCLC, has limited CNS penetrance (Kp_uu,CSF 0.0026) and, as expected, more than a third of treatment failures among patients receiving crizotinib failures involve CNS progression, thus limiting the extent of clinical benefit⁷⁷. Newer and more-specific ALK TKIs such as alectinib, brigatinib and lorlatinib all have improved levels of CNS penetrance, with 82% of patients having an objective intracranial response to lorlatinib^78–80.

Indeed, the second-generation ALK TKIs ceritinib, alectinib and brigatinib seem to have improved CNS penetration relative to that of crizotinib or systemic chemotherapy. Ceritinib had far superior intracranial efficacy compared to that of chemotherapy in the phase III ASCEND-4 trial (TABLE 1)^81,82. The CNS penetrance of alectinib is additionally enhanced relative to ceritinib (Kp_uu,brain 0.63–0.94), reflecting the fact that this agent is not a P-glycoprotein substrate^83,84. Data from the phase III ALEX trial confirmed the impressive intracranial activity of alectinib in patients with asymptomatic measurable brain metastases or LMD, with an intracranial ORR of 81%^79,85. Brigatinib has similar levels of systemic and intracranial activity to alectinib, among patients with baseline measurable brain metastases and no LMD (TABLE 1). Analyses of long term CNS DoR and CNS PFS of both alectinib and brigatinib remain under investigation^78,86. Ensartinib has also demonstrated superiority over crizotinib; data from the phase III eXalt3 trial demonstrated that fewer patients without baseline brain metastases develop brain metastases at 12 months when treated with ensartinib compared with crizotinib (4.2% versus 23.9%; HR 0.32, 95% CI 0.16–0.63), demonstrating not only regression of pre-exisiting brain metastases, but also prevention of CNS metastases⁸⁷.

Lorlatinib is a third-generation ALK and ROS1 TKI with inhibitory activity against all known acquired crizotinib resistance mutations⁸⁸. The effective BBB penetrance of this agent (Kp_uu,CSF of 0.75–0.77) has been confirmed in both preclinical animal models and in patients^22,89,90. Clinically, lorlatinib has shown the most impressive CNS efficacy of all ALK TKIs, with a 71% intracranial complete response rate and HR of 0.10 (95% CI 0.04–0.27) for time to intracranial progression among treatment-naive patients with baseline CNS metastases in the phase III CROWN study^80,91. Furthermore, lorlatinib is the only TKI that has demonstrated both intracranial and systemic disease control in patients who previously received ≥1 earlier-generation ALK TKI, making it an excellent second-line option. An impressive median CNS PFS duration of 24.6 months was reported in patients who had previously received at least one ALK TKI. However, the inclusion of patients with both parenchymal disease and LMD, the limited sample size of ths study (15 patients), and the history of prior intracranial radiotherapy in 65% of patients make these results more difficult to interpret⁹². Nevertheless, in light of the impressive intracranial data from the CROWN study, lorlatinib should be used in ALK-TKI pre-treated patients with CNS metastases, or as first-line therapy in those with CNS-predominant metastases.

Several 4^th-generation ALK TKIs are currently being tested in clinical trials. NVL-655 is an ALK TKI currently in development that has activity against both single and compound ALK resistance mutations, while also sparing NTRK and thus minimizing the risk of adverse events related to TRK inhibition seen in other ALK TKIs, as well as a Kp_uu,brain of 0.16⁹⁴.

ROS1 rearrangements

ROS1 rearrangements occur in 1–2% of all NSCLCs and are associated with a similar incidence of CNS metastasis to that of patients with ALK-rearranged NSCLC^95,96. Also similar to ALK-rearranged NSCLC, crizotinib confers limited CNS benefit in patients with ROS1-rearranged NSCLC⁹⁷. Newer ROS1-targeting TKIs have more promising levels of CNS activity⁹⁸. Entrectinib is an inhibitor of ALK, ROS1 and NTRK and is approved by the FDA as both a ROS1 and NTRK inhibitor⁹⁶. As a weak P-glycoprotein substrate, entrectinib has good CNS penetrance, as indicated by its low apical efflux ratio and Kp_uu,brain of 0.43^99–101. An updated integrated analysis of data from the ALKA-372–001, STARTRK-1 and STARTRK-2 trials confirmed durable intracranial activity in patients with measurable CNS disease at enrolment (intracranial ORR 55%; median intracranial DoR 12.9 months^102,103. Repotrectinib and taletrectinib are both next-generation ROS1 TKIs with clinical data demonstrating impressive intracranial activity, including CNS ORRs of 100% (specifically in TKI-naive patients) and 91.6%, respectively, albeit in small cohorts of patients (TABLE 1)^104,105. Both drugs have received FDA breakthrough therapy designation for ROS1-rearranged NSCLC. Similar to its excellent intracranial activity in ALK-rearranged NSCLC, lorlatinib has been shown to induce durable responses with a median intracranial PFS duration of 38.8 months in patients with CNS-only progression on crizotinib¹⁰⁶. Overall, data on the intracranial activity of TKIs in patients with ROS1-rearranged NSCLC remains limited owing to the small sample sizes of many of the cohorts and a lack of head-to-head comparison between different TKIs. Nonetheless, a phase III trial comparing the efficacy of entrectinib versus that of crizotinib in patients with advanced-stage NSCLC, including those with brain metastases (NCT04603807), is underway.

RET rearrangements

RET rearrangements are oncogenic drivers in 1–2% of NSCLCs¹⁰⁷. The highest prevalence of brain metastases at diagnosis in NSCLC was seen in RET-rearranged NSCLC (32.2%), along with ALK-translocated NSCLC (34.9%)¹⁰⁸. While older, broad-spectrum kinase inhibitors (such as cabozantinib, vandetanib and lenvatinib) have enabled confirmed systemic responses¹⁰⁷, the limited available data suggest very poor intracranial response rates in patients with RET-rearranged NSCLC¹⁰⁹. However, the highly selective RET TKIs selpercatinib and pralsetinib, both of which have received FDA approval for RET-rearranged NSCLC, have excellent intracranial efficacy.

In an updated analysis of data from the phase I/II LIBRETTO-001 study, selpercatinib yielded impressive and durable intracranial activity, including a median CNS PFS of 19.4 months among patients with baseline measurable CNS metastases independent of previous systemic therapy or radiotherapy. The phase I part of this study did not require brain imaging at baseline for all study participants. However, phase II did mandate both baseline and subsequent brain imaging, regardless of the presence of pre-existing brain metastases. None of the 178 patients without baseline CNS metastases developed intracranial disease progression (the estimated probability of intracranial disease progression at 2 years was 0.7%), suggesting that selpercatinib either prevents or delays the development of CNS metastases¹¹⁰. In a combined analysis of data from LIBRETTO-001 and LIBRETTO-201, all patients with disease progression who lacked baseline CNS metastases had extracranial-only progression¹¹¹. Pralsetinib is similarly efficacious in terms of overall intracranial response (all 9 patients with baseline brain metastases had an objective response, and one response detected on post-baseline assessments, although the duration of CNS response is unknown at this time, and sample size was limited to 10 patients in the phase I/II ARROW cohort¹¹². In a recent update of ARROW in the Chinese cohort, pralsetinib demonstrated CNS ORR of 73.7% (14/19) in chemotherapy-pretreated patients and 82.6% (19/23) in treatment-naïve patients¹¹³.

NTRK rearrangements

NTRK rearrangements occur in <1% of NSCLCs, and the NTRK inhibitors entrectinib and larotrectinib are now FDA-approved for this rare population, based on histology- and tumor-agnostic trials of NTRK-rearranged solid tumors. For entrectinib, data from patients with NTRK-rearranged NSCLC with CNS metastases indicate similar levels of activity to that seen in patients with ROS1-rearranged NSCLC in the previously mentioned integrated analysis of data from ALKA-372–001, STARTRK-1 and STARTRK-2, with 66.7% (4 of 6 patients) having an intracranial response^102,103,114. However, neurological adverse events can occur with TRK inhibition, as TRK receptors and pathway activation have been found to be closely involved in regulating several aspects of CNS function, including appetite, balance and pain sensitivity¹¹⁵. For example, dysgeusia, or disturbed taste sensations, is the most frequently reported grade 1–2 adverse event in patients receiving entrectinib, with most clinically serious events also being neurological in nature (including one incidence of cognitive disorder, cerebellar ataxia and dizziness)¹¹⁴. Larotrectinib is a pan-TRK inhibitor designed to have poor CNS penetration and thus avoid CNS toxicities. In contrast to entrectinib, larotrectinib has a relatively high apical efflux ratio, which translates to a lower Kp_uu,CSF of 0.03⁹⁹. Despite low BBB penetration, larotrectinib has demonstrated activity among 8 patients with baseline CNS metastases (ORR 63%) in an integrated analysis of data from a phase I trial (NCT02122913) and the phase II NAVIGATE trial, although CNS-specific ORR is unknown¹¹⁶. No clinically serious neurological adverse events have been observed in patients receiving larotrectinib¹¹⁶. Several newer ROS1 and NTRK inhibitors, such as taletrectinib, repotrectinib and selirectinib, are under investigation and some have shown good BBB penetrance and promising overall and intracranial activity^117–119.

BRAF^V600E

Oncogenic alterations in BRAF can be identified in approximately 4% of patients with NSCLC, and around half of these (1–2%) are BRAF^{V600E 120,121}. The BRAF–MEK inhibitor combination of dabrafenib plus trametinib has been demonstrated to induce durable responses in patients with metastatic BRAF^V600E-mutant NSCLC, although intracranial efficacy is less well known owing to the exclusion of patients with symptomatic brain metastases from this phase II trial¹²². An intracranial ORR of 58% has been reported among 125 patients with metastatic BRAF^V600E-mutant melanoma¹²³, suggesting that this combination has a high level of CNS activity. Nonetheless, investigations involving patients with BRAF-mutant NSCLC with CNS metastases are needed.

MET exon 14 skipping mutations

MET exon 14 skipping mutations are found in 3–4% of all NSCLCs^124–127. The incidence of CNS metastases at diagnosis in this group is around 17%, with a lifetime incidence of 36%¹²⁶. The TKIs capmatinib and tepotinib are the both approved by the FDA for patients with NSCLC harbouring METexon 14 skipping mutations¹²⁴. Capmatinib is a highly selective and potent inhibitor of exon-14 altered MET¹²⁷, and has a Kp_uu,CSF of 0.34¹²⁸. In the phase II GEOMETRY mono-1 study, capmatinib demonstrated an intracranial ORR of 54%, with 30.8% of patients having a CNS complete response¹²⁸. Patients receiving tepotinib, another agent targeting exon-14 altered MET, had a similar intracranial ORR in the phase II VISION study^129,130, with an updated analysis indicating that 71.4% (5 of 7 patients) had a partial intracranial response with 86.7% (13 of 15) having intracranial disease control according to response assessment in neuro-oncology brain metastases (RANO-BM) criteria¹³¹. The addition of the highly selective MET inhibitor savolitinib to osimertinib has demonstrated efficacy in overcoming resistance owing to acquired secondary MET amplification and/or overexpression in patients with EGFR-mutant NSCLC, though CNS efficacy was not evaluated¹³². Other MET TKIs currently in early clinical trial development include ABN-401 (NCT04052971)¹³³ and APL-101 (NCT03175224), although no CNS data are currently available for these agents.

KRAS^G12C

Despite being the most common oncogenic alteration, including in up to 30% of all patients with NSCLC^134,135, oncogenic KRAS alterations have been deemed ‘undruggable’ for several decades — until the FDA accelerated approval of the KRAS^G12C inhibitor sotorasib in 2021 for patients with advanced-stage KRAS^G12C-mutant NSCLC, an alteration found in 13% of all NSCLCs. Data from a retrospective analysis indicate that up to 40% of patients with KRAS-mutant NSCLC will develop brain metastases at any time during the course of disease, with a 6-month cumulative incidence of CNS metastases since initial diagnosis of 42%¹³⁶.

Sotorasib is a specific and irreversible KRAS^G12C inhibitor that locks the mutated protein in the inactive GDP-bound state, thus inhibiting subsequent phosphorylation of downstream signalling proteins^137,138. In a post-hoc analysis of data from the phase II CodeBreaK100 trial, 14 of 16 patients (88%) with evaluable brain metastases had intracranial disease control per the RANO-BM criteria and 2 (13%) had a complete response¹³⁹. However, only patients with resected CNS lesions or who received brain radiotherapy >4 weeks prior to enrollment were included in this analysis, whereas those with untreated brain metastases were excluded. Thus, the intracranial activity of sotorasib in patients with active, untreated brain metastases remains unknown and will be difficult to evaluate.

Adagrasib is another KRAS^G12C inhibitor that received accelerated approval for patients with advanced-stage KRAS^G12C-mutant NSCLC in December 2022. This agent has demonstrated BBB penetrance and intracranial activity in both preclinical and clinical studies. Preclinically, adagrasib accumulates to adequate concentrations in the CNS partially through the saturation of P-glycoprotein-mediated efflux¹⁴⁰. Adagrasib is the first KRAS^G12C inhibitor with data available on intracranial activity in patients with active, untreated brain metastases, with a CNS ORR of 42%, intracranial PFS of 5.4 months, and median IC DOR of 12.7 months reported in the phase Ib KRYSTAL-1 trial¹⁴¹. The average Kp_uu,CSF was 0.47 among 2 patients enrolled in this study, thus further confirming the CNS penetration of this drug¹⁴².

An unmet need remains for intracranially penetrant treatments for patients with NSCLC harbouring KRAS^G12C, as well as other KRAS alterations without an approved targeted therapy. Various KRAS^G12C inhibitors are in early clinical development for solid tumors including NSCLC, including JDQ-443 (NCT04699188), GDC-6036 (NCT04449874), JNJ-74699157 (NCT04006301), MK-1084 (NCT05067283) and BI-1823911(NCT04973163), with many more agents expected to follow in the next few years. Furthermore, there are also ongoing phase 1/2 studies evaluating therapies targeting KRAS non-G12C alterions, such as MRTX1133 (NCT05737706) and RMC6236 (NCT05379985).

LMD and intrathecal therapies

LMD describes the secondary spread of malignant cells into the CSF and leptomeninges of the brain and spinal cord, a complication occurring in 3–5% of patients with advanced-stage NSCLC. This pattern of dissemination is viewed as the worst type of CNS metastasis, owing to the limited treatment options available and an extremely poor prognosis⁵⁰ (Figure 2).

Figure 2. MRI brain images (axial view, T1 sequence, post-contrast) demonstrating.

(A) diffuse LMD with enhancement within the cerebral sulci (B) normal brain without metastases or LMD

Osimertinib has a Kp_uu,CSF of 0.29 and was therefore evaluated in the phase II BLOOM study as a systemic therapy for patients with EGFR-mutant NSCLC and LMD. In this study, 41 patients of Asian ethnicity with cytologically confirmed LMD and disease progression on an EGFR TKI received 160 mg daily doses of osimertinib, which is double the standard dose⁴⁶. Measurable CNS lesions were assessed using RECIST 1.1, whereas LMDs were evaluated using blinded central independent review according to RANO leptomeningeal metastases (RANO-LM) criteria. Patients receivivng osimertinib had a confirmed leptomeningeal ORR of 41% and a median DoR of 8.3 months^143. Almost half (49%) of the cohort had received prior radiotherapy, most of whom completed treatment ≤6 months prior to enrolment. Hence, ambiguity exists regarding the extent to which radiotherapy might have contributed to the responses of these patients. Furthermore, 160 mg daily dose is not the standard dose of osimertinib used in patients with metastatic EGFR-mutant NSCLC without LMD, and carries additional toxicities compared to the 80 mg daily dose, including 66% grade ≥ 3 adverse events leading to drug discontinuation in 22% of patients, much higher than that observed in the 80mg trials. Nevertheless, as one of very few clinical studies capturing an LMD-only trial that is often excluded from clinical trials, BLOOM is a major stride forward in the highly understudied field of LMD. The efficacy of osimertinib 80 mg once daily has also been evaluated in patients with LMD through retrospective analyses of the AURA and FLAURA trials, with LM ORRs of 55% and 80% reported, respectively, supporting the efficacy of standard-dose osimertinib in this setting but with fewer serious toxicities^21,144.

A phase II study in which patients with ALK-rearranged NSCLC and intracranial disease progression on ≥1 ALK TKI, including 4 of 23 with LMD, received lorlatinib provides some evidence of the activity of this agent in this setting. The intracranial ORR for the entire cohort was 59%, although the best response among patients with LMD was stable disease⁹².

Intrathecal injections, via either the lumbar or intraventricular routes, in which drugs are delivered directly to the CNS either via the intrathecal spaces in the spinal cord or the subarachnoid space near the base of the brain, are an avenue that requires further exploration especially for patients with LMD (Figure 3). This approach has the advantages of bypassing the BBB and also avoiding drug metabolism within any part of the systemic circulation while also minimizing the risk of systemic toxicities^24. However, the increase risk of neurological complications associated with IT chemotherapy such as paresthesia, cranial nerve palsy or paralysis, as well as the expertise required for administration, must also be acknowledged¹⁴⁵. The intrathecal approach is particularly appealing for the administration of cytotoxic chemotherapy and/or macromolecules. A phase I/II trial evaluating the safety and efficacy of intrathecally administered pemetrexed plus dexamethasone in patients with EGFR-mutant NSCLC and disease progression on a TKI yielded an impressive ORR of 84.6% (22 of 26 patients) and a median OS duration of 9.0 months. The intervention was also well tolerated, with most patients having low-grade adverse events only¹⁴⁶.

Figure 3. Intrathecal chemotherapy can be delivered through two different methods.

A) Direct injection into the CSF in the lower lumbar spinal column or B) Injection into an Ommaya reservoir (intraventricular implant that is surgically placed beneath the scalp)

Intrathecal topotecan did not yield impressive survival outcomes (6-month PFS of 19%) among 62 patients with LMDs from advanced-stage solid tumours, including 13 with NSCLC, in a phase II trial conducted in the early 2000s. A larger-cohort single-centre study evaluating a similar approach in a similar setting and including 29 patients with lung cancer demonstrated a median OS duration of 6.5 months^150,151.

Rapid clearance of the administered drug is a common challenge associated with intrathecal therapy^153–155. Intrathecal formulations with encapsulated chambers designed to preserve bioactivity while slowly releasing the active drug provide one method of overcoming this problem. Liposomal cytarabine, which has been withdrawn from the market due to product-specific manufacturing problems, is a slow-release form of cytarabine with the drug remaining within the patient’s CSF for >14 days¹⁵⁶. This formulation improves the response rates of patients with lymphomatous meningitis and offers some benefit for patients with LMD from certain solid tumours^156–158. This example suggests that innovative chemical engineering approaches could be leveraged to develop formulations that are able to overcome rapid drug clearance from the CSF. Various technologies, such as polymer-protein bio-conjugation (PEGylation), polymer complexation, polymer encapsulation, and microparticle-mediated intrathecal protein or gene delivery¹⁵⁵, are expected to greatly accelerate the speed of development of such intrathecal formulations.

Assessment criteria and trial design

The measurement of intracranial response across clinical trials involving patients with solid tumours has been inconsistent owing to the use of different evaluation criteria (TABLE 2). The RANO-BM criteria originate from a multidisciplinary effort to standardize the evaluation of response and progression of brain metastases in clinical trials. Prior to the development of RANO-BM, most trials used RECIST, which has various shortcomings when used to evaluate brain metastases (TABLE 2). For example, under RECIST, the CNS and systemic lesions are considered to be a single compartment, whereas RANO-BM stipulates that intracranial disease status is assessed separately from systemic disease¹⁴⁶. Unlike RECIST, RANO-BM accounts for steroid use, clinical neurological deterioration (related to the tumor) and neurological symptoms, and also allows for the consideration of pseudoprogression from immunotherapy and/or radiosurgery (SRS)¹⁵⁹. Furthermore, RANO-BM establishes separate guidelines and end points for assessing the activity of local and systemic therapies^160,161. Evaluations of intracranial tumour volume need to be more commonly incorporated into response assessments in clinical trials. Despite the lack of a standardized mandate for volumetric response assessment in trials assessing CNS metastases due to insufficient preexisting data, RANO-BM encourages inclusion of volumetric response evaluation (using gadolinium-enhanced MRI) in effort to expand the current knowledge base and potentially substantiate its use in future trial design¹⁵⁹.

Table 2 │.

Key criteria comparison for RECIST 1.1, RANO-BM and RANO-LM^160–163

Domain	RECIST 1.1	RANO-BM	RANO-LM
Imaging modality	CT or MRI	CT or MRI	MRI brain and spine
Measurement technique	Unidimensional	Bidimensional	Bidimensional
Target lesion	Longest diameter 310 mm	Contrast-enhancing lesions with two perpendicular diameters 310 mm	Nodular lesions, leptomeningeal enhancement, cranial nerve enhancement, hydrocephalus or nerve root enhancement
Maximum CNS target lesions	≤2	≤5	≤5
Definition of progressive disease	≥20% increase in sum of lesion diameter, any new lesion	≥25% increase in sum of enhancing lesion diameter, any new lesions	Worsening neurological function owing to LM, worsening neuroradiographical assessments (≥25% increase in sum of lesion diameters), positive CSF cytology (lack consensus)
Shrinkage needed for partial response	≥30%	≥50%	≥50%
Durability of response	Confirmatory scans only needed in non-randomized trials in which response is the primary end point	Confirmatory scans required at ≥4 weeks apart	Confirmatory scans at baseline and follow-up; CSF analysis not needed
T2/FLAIR	Not required	Required	Not required
Corticosteroids use considered	No	Yes	Only in patients with haematological malignancies
Neurological symptoms considered	No	Yes	Yes

BM, brain metastases; CSF, cerebrospinal fluid; LM, leptomeningeal metastases; RECIST, Response Evaluation Criteria in Solid Tumors; RANO, Response Assessment in Neuro-Oncology

Evaluation of response in patients with LMD is even more challenging and lacks standardization. The RANO-LM group recommends that all patients with LMD enrolled in trials undergo a standardized neurological examination, CSF analysis including cytology, complete brain and spinal MRI, as well as radioisotope CSF flow studies in those receiving intrathecal therapy (at baseline and after treatment to assess response)¹⁶². No single disease characteristic is specific or definitive of LMD; therefore, the RANO-LM criteria incorporate three core components to quantify response and/or disease progression. On the basis of a radiological LM response score designed to standardize subjectivity, imaging results are graded as either stable, progressive or improved¹⁶². Progressive disease is characterized by worsening neurological findings and/or worsening neuroradiographic assessments, with the current criteria defining progression as conversion of negative to positive CSF cytology or failure to convert positive cytology to negative cytology after therapy. However, persistently positive CSF cytology with otherwise unchanged clinical and radiographical assessments is insufficient to qualify as disease progression; how best to assess such responses remains unclear¹⁶³.

Regulatory considerations

In 2021, the FDA provided new guidelines for the design of trials evaluating cancer drugs in patients with CNS metastases from solid tumours, recognizing “that treatment of CNS metastases presents several challenges and unique considerations (e.g., circumventing the blood-brain barrier)”¹⁶⁴.This guidance stipulates that either of the standard response criteria used to evaluate CNS disease in prospective studies, such as the modified version of RECIST 1.1 or RANO-BM, can be used in clinical trials. The FDA also recommends concurrent imaging for patients with both intracranial and extracranial disease and provides criteria for including previously irradiated lesions as target lesions. Notably, the FDA guidance also recommends for case report forms to include response assessments, including changes in neurological symptoms, as well as concurrent steroid use and any use of antiseizure medications¹⁶⁴. The guidance also favours the use of the RANO-BM or RANO-LM criteria over RECIST 1.1 owing to the inclusion of neurological symptoms and steroid use only in the former but not the latter¹⁵⁹. These considerations are not limited to trials testing targeted therapies in patients with brain-metastatic NSCLC, but are also generally applicable to any trials involving patients with CNS metastases from solid tumours.

Clinical considerations

When evaluating the intracranial activity of targeted therapies for CNS metastases, several practical and clinical considerations should be taken into account besides FDA guidelines. First, any measurements of clinical benefit need to take into account both clinical and radiological evidence. Neurological function is a crucial outcome for patients with CNS metastases that should be an integral part of any assessment. Furthermore, the duration of CNS benefit, which can be measured using various different parameters, such as CNS PFS or CNS intervention-free survival, is an important clinical parameter, in addition to response rate. Independent of response, the duration of clinical benefit, defined as the period of time in which a change in systemic therapy is not needed while allowing SRS as required, could also be considered as a parameter of clinical benefit given the minimal adverse effects on patients’ quality of life.

Second, receipt of previous CNS radiotherapy and the types of CNS radiotherapy received need to be specified in order to understand the contributions of any systemic therapy to CNS end points. Currently, many trials, including some of the pivotal trials described herein, allow both patients with untreated and previously irradiated CNS lesions to be evaluated. For trials or cohorts dedicated to assessing the intracranial activity of systemic therapies, receipt of previous radiotherapy needs to be distinguished from no previous radiotherapy, with the types of CNS radiotherapy received in relation to the time of initiation of trial drug considered separately. Assessing the contribution of any targeted therapy is particularly challenging in patients who previously received WBRT, as this intervention can provide oncological benefit across the entire brain parenchyma that might last for several months¹⁶⁵. In patients who previously received SRS, responses to systemic therapy among lesions located in the untreated brain parenchyma are likely to reflect control by systemic therapy regardless of the relative timing of SRS. However, any previously SRS-irradiated lesions would not be ideal for clinical response assessments¹⁶⁶.

Finally, adequately powered dedicated trials involving cohorts of patients with LMD or untreated brain metastases with CNS-specific primary end points are needed. Other than the BLOOM trial, which exclusively enrolled patients with LMD, such patients are typically included alongside other patients with CNS metastases, when allowed to enrol¹⁴³. The LMD subgroups from such trials provide preliminary data and a certain level of clinical confidence in the intracranial efficacy of systemic therapies for patients with LMD, although the sizes of such subgroups are generally very small. For example, the large-cohort AURA, FLAURA and CROWN trials allowed patients with LMD to enrol, although these subgroups all contained <10 patients^21,49,80. Going forward, dedicated trials or trial cohorts for patients with LMD or those with untreated brain metastases will enable a more-homogenous population to be evaluated and provide the optimal level of evidence to support the development of new CNS-active therapies in the most appropriate populations. A big challenge with this approach, however, would be the accrual of a sufficiently powered LMD cohort, given its rarity, especially in trials focused on specific molecular alterations.

Future directions

Many areas of active research, beyond the development of novel CNS-penetrant targeted therapies and/or formulations for intrathecal administration, might support translational and clinical research efforts aiming to improve the diagnosis and treatment of patients with brain metastases from solid tumours.

Advanced imaging techniques

Both the detection of CNS metastasis and assessments of their response to therapy primarily relies on MRI. T1-weighted acquisition post-gadolinium contrast imaging is typically used to assess the extent of BBB disruption and delineate the tumour, whereas T2-weighted fluid-attenuated inversion recovery (T2-FLAIR) enables the detection of vasogenic oedema located around the tumour¹⁶⁷. CNS MRI is the preferred imaging method in both the RANO and RECIST criteria. These criteria stipulate the evaluation of treatment activity based on changes in the diameter of CNS metastases, which can be confounded by several other factors when assessed using standard MRI, including peritumoural oedema, treatment-related inflammation (from immunotherapy or radiotherapy) and delayed complications such as radiation necrosis. Therefore, an unmet need exists for advanced imaging techniques capable of delineating a genuine tumour response from other treatment-related changes.

Perfusion-weighted imaging (PWI) and diffusion-weighted imaging (DWI) are both MRI methods that are already used in neuro-oncology clinics. PWI has shown to detect viable tumour specifically in brain metastases^168,169. Dynamic contrast-enhanced (DCE)-MRI is a subtype of PWI that enables modelling of the distribution of contrast between the vascular and interstitial space, enabling the detection of changes in vascular permeability (for example, owing to BBB breakdown at the tumour interface). Dynamic susceptibility contrast (DSC)-MRI, which measures signal loss on a T2-weighted sequence as contrast passes through is another widely used PWI technique¹⁷⁰. These methods enable qualitative assessments of whether changes in tumour diameter and/or the tumour microenvironment reflect the presence of viable tumour material or inflammation and are being increasingly used clinically for this purpose. DWI also provides information on tumour cellularity through measurements of the apparent diffusion coefficient (ADC), which is expected to inversely correlate with cellularity for a progressing metastatic lesion^171,172. In addition to PWI and DWI, magnetic resonance spectroscopy (MRS) is another technique currently in clinical use. MRS enables the measurement of biochemical alterations in tumour tissue and can be used to deduce treatment response based on metabolic alterations in brain metastases. Specifically in brain metastases, MRS has been shown to effectively distinguish benign from malignant tissues and treatment-induced changes in tumour metabolites. These metabolic alterations (involving choline, creatine and N-acetyl-aspartate) have been found to correlate with prognosis and disease recurrence after SRS in patients with NSCLC, and can therefore be used to predict clinical outcomes^173,174.

Several novel imaging techniques have also been developed but are currently not widely implemented clinically, including those involving PET¹⁷⁵, single-photon emission computed tomography (SPECT) and chemical exchange saturation transfer (CEST). ¹⁸F-Fluorodeoxyglucose (FDG)–PET is not routinely used to evaluate brain metastases owing to high levels of background brain metabolic activity; however, delayed FDG-PET has demonstrated some potential in differentiating between disease progression and treatment effects¹⁷⁶. Furthermore, various PET imaging modalities involving alternative tracers, such as radiolabelled nucleotides or amino acids, are being developed specifically for brain tumour imaging. For example, ¹⁸F-flourothymidine (FLT) provides a moderately specific method of detecting cellular proliferation as thymidine does not typically accumulate to high levels in brain cells or in regions of inflammation; however, FLT requires BBB breakdown in order to penetrate the CNS, thus limiting its use in assessing treatment responses in patients with brain metastases. Another PET amino acid tracer, O-(2-[¹⁸F] fluoroethyl)-L-tyrosine (FET), enables the identification of treatment-related metabolic alterations earlier than changes in tumour diameter on MRI in patients with glioblastoma receiving bevacizumab¹⁷⁷. Several natural or synthetic amino acid-based PET probes (including ¹⁸F-FET as well as ¹¹C-MET, ⁸F-Fluciclovine and ¹⁸F-FDOPA) are gradually entering clinical use^178,179 following recommendations from the RANO working group for use in patients with gliomas¹⁸⁰.

Adapting these advanced imaging techniques into routine clinical practice comes with several challenges. Perfusion-weighted and diffusion-weighted MRI can be performed readily without the need for additional equipment at most comprehensive medical centres, although image processing sometimes requires specialized expertise and/or software. Non-FDG-PET is currently not available for clinical use and might require access to a cyclotron if using radioactive tracers that are not commercially available. Delayed PET can be time-intensive and costly without insurance approval. However, the potential advantages of using one imaging method (non-FDG-PET) to assess tumour status in both the intracranial and extracranial compartments are appealing to both patients and clinicians. Tumour-type or treatment-specific radiotracers, such as PET involving a radiolabelled DLL3 tracer in patients with small-cell lung cancer¹⁸¹ and a radiolabelled PD-L1 tracer for patients who might be eligible for anti-PD-1 or anti-PD-L1 antibodies¹⁸², are currently under preclinical investigation.

Molecular profiling

A deeper understanding of the molecular underpinnings of CNS metastases and the extent of genomic concordance between primary tumours and their associated CNS metastases will enabled improved guidance of treatment selection, both at initial diagnosis and at the time of acquired resistance. This area of research will be greatly facilitated by the profiling of cell-free DNA (cfDNA) from CSF samples, which is a much more sensitive method than CSF cytology¹⁸³. Furthermore, CSF-derived cfDNA has demonstrated higher sensitivity compared to plasma-derive cfDNA, particularly in brain metastases and primary brain tumors, presumably owing to the relative volumes of these compartments¹⁸⁴. Studies involving patients with LMD from EGFR-mutant NSCLC have revealed a high level of concordance of EGFR mutations, as well as TP53, CDKN2A and PIK3CA aberrations, between the primary tumour and cfDNA from the CSF, albeit with notable differences in alterations in genes associated with cell cycle regulation and the DNA damage response¹⁸⁵. After disease progression on osimertinib, EGFR^C797S, MET dysregulation and the co-ocurrence of alterations in TP53 plus RB1 were reported as potential mechanisms of resistance detectable in the CSF cfDNA^185,186. Similarly high levels of concordance between ALK rearrangements and TP53 mutations have been reported, although different secondary ALK mutations were observed in CSF samples in a patient who received brigatinib, suggesting the occurrence of parallel clonal evolution in the CNS and systemic compartments¹⁸⁷. However, data on shared and distinct mutations between CNS metastases and the primary tumour are limited and will require validation in larger-cohort studies.

In addition to genetic profiling, analysis of cfDNA in CSF samples can also enable more accurate disease monitoring¹⁸⁸. A prospective study involving 28 patients with EGFR-mutant NSCLC and LMD demonstrated a 100% sensitivity of cfDNA profiling in detecting driver genes in CSF samples, as well as the ability to capture multiple copy number variations (CNVs), concomitant resistance mutations and loss of heterozygosity more frequently than the equivalent analysis of plasma samples¹⁸⁸. Another study demonstrated an association between >50% reduction in CSF cfDNA after treatment and longer median intracranial PFS, while shorter intracranial PFS was associated with a higher residual burden of clonal mutations in CSF cfDNA¹⁸⁹. These studies have laid the foundations for future use of dynamic assessments of CSF cfDNA to monitor intracranial responses and CNS clonal evolution during treatment^188,189.

Obtaining CSF samples for cfDNA analysis is relatively non-invasive compared with tumour biopsy sampling and thus enables longitudinal assessments. Nonetheless, these methods should not replace tissue profiling, especially for characterization of the tumour immune microenvironment (TIME). In one study, tumour PD-L1 levels were comparable between the primary tumour and brain metastases, although the latter had substantially lower levels of all T cell subsets relative to the matched primary tumour¹⁹⁰, highlighting the need to evaluate tissue samples from brain metastases to gain insights into the TIME. Window-of-opportunity trials have the potential to add value in this area, as these trials typically involve short-term treatment exposures followed by surgical resection. The resected post-treatment specimens can provide an extremely valuable resource that can advances our knowledge of treatment-induced histological and molecular alterations, especially when paired with initial pretreatment biopsy samples. In one such trial in which patients with recurrent glioblastoma received pembrolizumab followed by surgical resection, comparisons of pretreatment and post-treatment specimens indicated a lack of response, characterized by a macrophage-dominated TIME with limited effector T cell infiltration, in most patients¹⁹¹. Patients enrolled in this trial had limited levels of clinical benefit (3 partial responses in 15 patients), although molecular profiling of the TIME of those without a response provides clear directions for future research efforts towards overcoming macrophage-related immunosuppression in this setting.

A small subset of patients with brain-metastatic NSCLC will receive upfront palliative surgical resection of one or more dominant CNS lesions. For this population, offering a highly CNS-penetrant TKI in the context of a well-designed window-of-opportunity trial as systemic maintenance therapy should be considered. An ongoing window-of-opportunity trial (NCT05120960) is evaluating tepotinib in patients with MET-altered brain tumours, including patients with brain metastases from EGFR-mutant NSCLCs with acquired MET amplifications after disease progression on EGFR TKIs.

Targeted therapy and radiotherapy

An increasing number of novel targeted therapies have demonstrated robust intracranial activity; however, an emerging question exists regarding how best to incorporate targeted therapies into the current standard-of-care (comprising radiotherapy and surgery) for patients with brain metastases. This consideration is particularly important given that the approach to radiotherapy has moved away from WBRT owing to the substantial incidence of cognitive toxicities and limited OS benefit, as well as advances in more-focussed radiotherapy techniques. SRS has been increasingly used for patients with up to 15 lesions¹⁹², while multi-fraction stereotactic body radiation therapy (SBRT) enables the targeted irradiation of larger brain metastases or lesions located close to crucial structures such as the brainstem or optic chiasm. In patients receiving first-generation EGFR or ALK TKIs (gefitinib, erlotinib or crizotinib), upfront radiotherapy in combination with a TKI demonstrated survival advantages compared to a TKI only¹⁹³. However, the improved CNS activity of newer-generation TKIs, with intracranial ORRs >70–80% observed with certain agents, raises questions as to whether upfront radiotherapy can be delayed or even omitted in this new era¹⁹⁴. At least two randomized trials are recruiting patients in an attempt to examine whether upfront SRS (NCT03497767) or early SRS (NCT05033691) is superior to osimertinib without radiotherapy in patients with newly diagnosed brain metastatic EGFR-mutant NSCLC. Evidence in these studies will be derived from patients with EGFR-mutant NSCLC, although the principal findings will likely inform the treatment approach for other molecular subtypes with highly CNS-penetrant TKIs available.

Before data from such trials become available, we recommend that the treatment approach for each patient is discussed in a multidisciplinary fashion including clinicians specializing in medical oncology, CNS radiation oncology and neurosurgery. In this regard, ASCO/SNO/ASTRO jointly published guidelines for the treatment of patients with brain metastases in 2022¹⁹⁵, and these recommendations should provide a solid multidisciplinary foundation for the individualized treatment of each patient. Currently, early intervention with local therapy (radiotherapy and/or surgery) is generally recommended for patients with symptomatic or rapidly progressing CNS metastases. For those with asymptomatic CNS metastases from NSCLC, the overall recommendation is that local therapy should not be deferred, with a few exceptions when a highly effective CNS-penetrant TKI is available, such as osimertinib for EGFR-mutant NSCLC and lorlatinib for ALK-rearranged NSCLC. Consensus was not reached on the role of immunotherapy.

Treatment decisions become even more complex upon disease progression. For patients with CNS-only disease progression on a CNS-penetrant TKI, local therapy with radiotherapy and continuation of the TKI is often recommended. For those with extracranial-only progression in the context of continued control of CNS disease, many medical oncologists would continue the TKI to ensure CNS protection, while adding chemotherapy for systemic disease control. This strategy has demonstrated a favorable safety profile (with most frequent G3 adverse event being neutropenia, 11%), while delaying intracranial disease progression in 28 of 37 (75.7%) patients with brain metastases¹⁹⁶. For progression in both the CNS and extracranial compartments, a change of systemic therapy while leveraging radiotherapy for CNS control is a reasonable approach given that the tumour cells will most likely have acquired one or more new mechanisms of resistance. Owing to the rapid pace of development of CNS-penetrant TKIs, evidence informing the optimal sequences and combinations of multimodality treatment is still accumulating; therefore, treatment-related decisions should be made in a multidisciplinary fashion that also takes into account other clinical characteristics and the availability of CNS-penetrant TKIs.

Conclusions

In summary, CNS metastases have always been recognized as a devastating complication of advanced-stage solid tumours. Over the past decade, and despite the rapid development of numerous novel therapeutics that provide improved extracranial tumour control, CNS metastases, including LMD, are emerging as an increasing unmet need that can limit long-term survival. CNS metastases are considered a distinct challenge because the tumour cells reside within a physiologically unique compartment protected by the BBB, and thus demand a different treatment strategy. For patients with oncogene-driven NSCLC, a wave of novel TKIs with excellent brain penetrance has already delivered clinical successes in subsets of patients with tumours harbouring certain driver oncogenes. At the same time, advances in radiotherapy technology and surgical techniques are continuing to improve outcomes, while making the management of CNS metastases highly complex and multidisciplinary in nature, requiring close collaboration. Novel diagnostic and intrathecal therapeutic technologies are also expanding the potential to treat patients with CNS metastases and LMD. With the development of improved comprehensive outcome measurement criteria and the availability of regulatory recommendations guiding drug development towards the unmet needs of patients with CNS metastases, we anticipate strong growth in clinical investigations over the next few years in this area, ultimately leading to improved clinical outcomes for patients.

Key points.

• CNS metastasis and leptomeningeal disease are clinical challenges in the treatment of patients with advanced-stage NSCLC that are often associated with inferior outcomes [Au: Details added, OK?]

• Targeted therapies have improved the outcomes of several molecularly defined subgroups of patients with oncogene-driven NSCLC, although certain small-molecule inhibitors confer only limited levels of CNS benefit, often owing to an inability to cross the blood–brain barrier [Au: Details added here also, OK?].

• Preclinical evaluations of Kp_uu are commonly used to assess the CNS penetrance of novel agents during drug development

• Several newer TKIs having improved CNS efficacy, including osimertinib in EGFR-mutant NSCLC and alectinib and lorlatinib in ALK-rearranged NSCLC, with many others in clinical development

• Response Assessment in Neuro-Oncology (RANO) criteria for response assessment in patients with brain or leptomeningeal metastases are different to RECIST; and future clinical trials should incorporate those criteria into the study design

BOX 1. Summary of FDA recommendations for evaluating cancer drugs in patients with CNS metastases.

Patient population

• CNS response rates should be evaluated in the context of the entire disease burden, regardless of the incidence of CNS metastases in the trial population.

• CNS-specific approvals might not be appropriate.

Available therapy

• Therapies with existing approvals for patients with metastatic disease from a given primary tumour type (such as ALK inhibitors in ALK-rearranged NSCLC) will be prioritized.

Prior therapies

• Attempt to collect information on previous interventions, such as any CNS-directed surgery, SRS, and WBRT, including timing and response rates

• Specify the permitted interval length between completion of CNS-directed therapy and study enrolment (typically at least 12 weeks).

• Include one additional stratification factor to minimize bias (such as treated versus untreated CNS metastases).

Assessments of CNS metastases

• Use RECIST or RANO-BM assesment criteria with gadolinium MRI as the preferred imaging modality, including a baseline scan.

• CNS and extra-CNS disease evaluation should occur simultaneously.

• Pre-specify which previously irradiated lesions are included as target lesions across all trial sites, and their time from irradiation.

• Attempt to capture data at baseline and during the study on variables potentially affecting the interpretation of radiographic images including neurological symptoms, steroid use and use of anti-seizure medications.

Study end points

• Time-to-event and all-cause mortality are preferred end points.

• Evaluate the entire burden of disease as end points (including ORR or PFS, and not only CNS-specific forms of these end points), with assessments verified by independent, blinded review by experts in CNS radiology.

• Specify CNS PFS versus systemic PFS progression to prevent censoring.

Leptomeningeal disease

• LMD should be confirmed and monitored using MRI and/or CSF analysis for response evaluation.

Glossary

CSF

cerebrospinal fluid

LMD

leptomeningeal disease

NSCLC

non-small-cell lung cancer

ORR

objective response rate

PFS

progression-free survival

RANO-BM

Response Assessment in Neuro-Oncology – Brain Metastases

RECIST

Response Evaluation Criteria in Solid Tumors

SRS

stereotactic radiosurgery

WBRT

whole-brain radiotherapy