PD-1/PD-L1 Blockers in NSCLC Brain Metastases: Challenging Paradigms and Clinical Practice

Brain metastases (BM) occur in 20% to 32% of patients diagnosed with non–small cell lung cancer (NSCLC) (1). In some particular settings, such as previously treated ALK-driven NSCLC, however, BM prevalence can reach 59% (2).

Patients with untreated BM have traditionally been excluded from clinical trials due to concerns that the particular pharmacokinetic/pharmacodynamic (PK/PD) profile of the investigational agent in the CNS could preclude the correct interpretation of the results (3). The exclusion of patients with active BM was supported by the fact that initial attempts to use systemic drugs against BM were discouraging (see Background on Systemic Treatment of Brain Metastases). However, in recent years, the development of new-generation highly penetrant kinase inhibitors has shown that appropriately designed drugs are active in the CNS (4, 5). Beyond oncogene-driven NSCLC, immunotherapeutic agents have recently shown promising activity in the CNS in patients with NSCLC BM. The results from the latest clinical trials are challenging traditional dogmas that claimed that monoclonal antibodies (mAbs) were not meaningfully active in the CNS. These findings are expected to change the clinical management of patients with NSCLC BM in the near future and warrant a review of the literature. In this review, we provide an overview of the recent results of mAb-based immunotherapy to treat BM, and how these results are challenging previous paradigms and current clinical practice.

Studies were identified from the PubMed database with the search strategy (PD-1[Title/Abstract]) OR PD-L1[Title/Abstract]) OR immunotherapy[Title/Abstract]) AND brain metastases[Title/Abstract]. The last date the search was performed was on the July 17, 2019. Abstracts from the 2019 American Society of Clinical Oncology (ASCO), American Association for Cancer Research (AACR), European Society for Medical Oncology (ESMO), International Association for the Study of Lung Cancer (IASLC) World Congress on Lung Cancer (WCLC) annual meetings were reviewed and considered for inclusion. Reference lists of reviewed articles were also considered for potential inclusion. Studies in the English language were reviewed. Studies were included regardless of date of publication but studies that included only patients with leptomeningeal disease were excluded. Reports that included patients with melanoma and clear cell renal cell carcinoma (ccRCC) BM were considered relevant if exclusively systemic therapeutic approaches were assessed. 408 references were initially identified. After title/abstract assessment, 93 references were selected for full-text evaluation. After assessing reference lists and previously mentioned additional sources, a total of 90 publications were included.

Clinical observations that chemotherapy was less active in BM have been mainly attributed to poor passage across the blood–brain barrier (BBB). This CNS-specific anatomic structure composed of endothelial cells (with tight junctions between them), pericytes, and astrocytes, is intended to protect the brain from inflammation and other circulating noxa. However, it is believed that the BBB also limits the access of drugs to the brain parenchyma (6).

Drug access to the brain parenchyma is regulated by Starling's forces, binding to serum proteins, liposolubility, molecular weight, and local molecular biology (7). Traditionally, it has been thought that molecular size and solubility are critical for antitumor activity in the CNS (8, 9). In tumors, neovessels are thought to disrupt the physiologic characteristics of the BBB (the basis of contrast enhancement in MRI) and, presumably, increase the access of large molecules to the damaged brain parenchyma. However, the fact that CNS disease progression is common in patients under treatment with first-generation EGFR and ALK tyrosine kinase inhibitors (TKI; ref. 10), but patients can be rescued by next-generation highly penetrant inhibitors such as osimertinib (4) or alectinib (5), underscores that pharmacokinetics/pharmacodynamics (PK/PD) are still critical for the activity of drugs in the CNS.

Initial attempts to effectively treat BM from solid tumors tested small conventional cytostatic drugs, such as temozolomide, that had previously shown activity against glioblastoma. Results were discouraging: the response rate (RR) was 7% in melanoma BM and median progression-free survival (PFS) was 1.2 months (11). Slightly better results have been obtained in NSCLC BM with pemetrexed, which is estimated to reach a distribution that is up to 10% of that in plasma (12, 13). When pemetrexed is combined with carboplatin, brain RR can reach 40% (14). Other platinum-based combinations have obtained similar results (15), but response is generally short-lasting.

With the development of EGFR and ALK TKI, the role of systemic treatment for patients with oncogene-driven NSCLC with BM was significantly reinforced. Currently, with the development of third-generation oral TKI such as osimertinib and alectinib, which have shown meaningful activity in the CNS (16, 17), radiotherapy-sparing management has become a common practice in this patient population.

mAbs have traditionally been excluded when systemic treatment of BM was attempted due to their high molecular weight and low penetration through the BBB. Only a few clinical PK/PD studies have evaluated the penetration of mAbs into cerebrospinal fluid (CSF), a commonly accepted surrogate of drug distribution into the brain interstitium. Rituximab levels in the CSF, for instance, are between 0.1 and 0.7% of those in plasma (18, 19). In solid tumors, trastuzumab CSF levels show a CSF:plasma ratio of 1:420 (20). Interestingly, the ratio increased to 1:76 after brain radiotherapy. Considering clinical evidence, a prospective randomized clinical trial that compared the addition of trastuzumab or lapatinib to a taxane therapy observed that, despite PFS being significantly shorter for the lapatinib arm, CNS was the first site of progressive disease (PD) slightly more frequently in the trastuzumab group (21). This suggests that the anatomical peculiarities of the CNS probably influence the local activity of conventional tumor cell–targeted mAbs.

The novel class of mAbs, the immune checkpoint inhibitors (ICI), have revolutionized oncology practice in recent years, progressively permeating the therapeutic algorithms of many malignancies with high epidemiologic impact. Their novel mechanism of action, distinct from other tumor cell–targeted agents, has prompted research to evaluate anti-PD-1/anti-PD-L1 agents' activity in patients with NSCLC with CNS involvement. The mechanism of action theoretically relates to modified immune cell activity rather than direct action of these agents in the brain. Indeed recently, nivolumab CSF levels have also been measured in 5 patients with suspected leptomeningeal metastases from a variety of solid tumors. CSF:plasma ratios ranged from 1:52 to 1:299 (22), suggesting that local access of anti-PD-1 agents to the brain is not significantly different from that of other mAbs. Among them, the approval of anti-PD-1/anti-PD-L1 mAbs for advanced NSCLC is making an unprecedented difference on long-term survival for a significant proportion of patients (23–26). Their novel mechanism of action, distinct from other tumor cell–targeted agents, has prompted research to evaluate anti-PD-(L)1 agents' activity in patients with NSCLC with CNS involvement.

There are many critical biological features that distinguish the brain from other tissues; the composition of the extracellular matrix is distinctive, there are unique tissue-resident cells including microglia, astrocytes, and neurons, and it is physically protected from inflammation and drug delivery by the BBB (6). Therefore, the inflammatory tumor microenvironment of BM exhibits several unique factors compared with that of extracranial lesions. However, there is robust evidence to support that these factors do not exclude the brain parenchyma from active immune surveillance.

Unlike extracranial tumors, in which macrophages are among the most abundant nonmalignant cells in the infiltrate (27), ontogenetically different cells are represented within the infiltrate in tumors in the CNS; both conventional macrophages of myeloid origin and resident microglial cells (which diverge ontogenetically from the former prenatally) are present. According to a recent publication, monocyte-derived macrophages are more abundant than microglia in the NSCLC BM microenvironment (28). Myeloid macrophages can be recruited from peripheral blood under certain inflammatory conditions (29), while tissue resident microglia are thought to be capable of local self-renewal. Human microglia are represented by highly heterogeneous cell populations, with little transcriptomic similarities to peripheral M1/M2 phenotypes and only a subset expressing major histocompatibility complex (MHC) class II or costimulatory molecules CD80 or CD86 (30).

T-cell infiltration is widely accepted as a key component of adaptive cancer immune surveillance. Certainly, almost all relevant therapeutic advances in the field of immunotherapy have been achieved in so-called “hot” (inflamed) tumors, which are naturally infiltrated by tumor-infiltrating lymphocytes (TIL).

In normal brain parenchyma, low densities of lymphocytes (around 1 to 2 per mm²) have been identified mainly in the perivascular Virchow–Robin spaces (31). However, under particular pathologic conditions, such as cancer, antigen-specific lymphocytes can extensively infiltrate the CNS. It is established that mediators of innate immunity such as TNFα, IL1, and IL6 can bind to brain microvascular endothelial cells and weaken the BBB (32). CD4⁺ cells additionally contribute to brain infiltration by other lymphocytes via local IFNγ production (32). In a series including 116 BM specimens from a variety of primary cancers (with lung cancer representing 53% of cases), high densities of CD3⁺, CD8⁺, and CD45RO⁺ cells were associated with improved OS (33). However, it has been recently reported that T-cell densities were significantly lower in NSCLC BM compared to paired primary tumor specimens (28, 34). Comparable results indicating lower adaptive immune responses in brain metastases relative to primary human NSCLCs have been reported by our group using multiplexed immunofluorescence analysis for PD-L1 and major B- and T-cell populations (35). Representative histology preparations and multiplexed immunofluorescence images from a primary lung adenocarcinoma and the corresponding brain metastasis are shown in Fig. 1.

Although imperfectly, PD-L1 expression by immunohistochemistry (IHC) predicts clinical benefit from ICI. In advanced NSCLC, a tumor PD-L1 expression of ≥50% robustly favors first-line pembrolizumab over chemotherapy (26). Clinical trials evaluating anti-PD-L1 drugs confirm a dose–response relationship between PD-L1 tumor expression and clinical benefit from ICI, although the latter is not restricted to the PD-L1–positive (≥1%) population (23, 36). PD-L1 expression has been evaluated in NSCLC BM surgical resection specimens, showing that PD-L1 tumor expression (≥5%) is present in 22% to 33% of the samples (37–39). This proportion is similar to the 31% PD-L1 positivity (≥5%) reported in extracranial NSCLC metastases (36). To date, the largest series in which PD-L1 expression between BM and primary NSCLC was evaluated (73 patients) show a qualitative concordance in the level of PD-L1 expression between BM and primary tumors that is above 80% (39, 40). However, lower concordance rates have been reported in smaller heterogeneous cohorts, with a trend toward lower PD-L1 expression in the CNS than in primary tumors (41, 42).

The mechanism of action of ICI, which is not specifically dependent on close contact with all tumor cell foci, could lead to immune recognition in more accessible extracranial metastatic lesions and secondarily, immune cell trafficking could be responsible for producing an antitumor response in the CNS, as suggested by others (43). This would confirm preclinical findings suggesting T-cell priming in the extracranial compartment is essential for an effective immune response in the CNS (44). Moreover, recent discoveries of the presence of lymphatic vessels in the dura mater, which are potentially capable of allowing CNS antigen presentation in the peripheral lymph nodes (45), are altering our understanding of immune privilege in the CNS. In addition, activated CD4⁺ T cells in the brain can loosen the BBB to circulating antibodies through local IFNγ production (32), a mechanism anti-PD-1/PD-L1 therapy could potentially enhance. These potentially relevant mechanisms of action are summarized in Fig. 2.

The fact that immune surveillance in BM shares some similarities with that in extracranial tumors supports research to evaluate the role of ICI for the treatment of solid tumor BM. The active research that is being conducted in the field is reviewed here. Solid data regarding the activity of immune checkpoint inhibitors in patients with leptomeningeal involvement are lacking because clinical trials systematically exclude patients in this particular CNS involvement situation and only some retrospective studies and case reports have partially addressed this issue. Therefore, the current review focusses mainly on brain metastasis.

Proof-of-concept that ICI could be active against NSCLC BM was obtained from heterogeneous clinical studies that were conducted in patients with previously untreated BM or in patients with brain involvement that have progressed after previous local therapy. These two scenarios have been included under the “active” BM definition herein.

In 2016, Dudnik and colleagues reported a retrospective analysis of patients with NSCLC BM who remained asymptomatic and corticosteroid-free before nivolumab initiation (46). Five patients were included. Two intracranial responses were observed, which were maintained for up to 24 and 28 weeks. Notably, intracranial and systemic responses were largely concordant, except for one patient in whom stable CNS disease was associated with rapid systemic progression. No severe adverse events (AE) were attributed to treatment.

To date, the most robust evidence on the activity of ICI for the treatment of NSCLC BM comes from a phase II trial (NCT02085070) that included patients with melanoma and NSCLC (47). The published report included the first 18 patients with NSCLC and the first 18 patients with melanoma. BM had to be between 5 and 20 mm in diameter and patients had to be steroid-free and neurologically asymptomatic. In the NSCLC arm, inclusion was restricted to patients with PD-L1 positive (≥1%) tumors. PD-L1 expression was evaluated in tumor tissue from any site. Eight of the first 18 patients with NSCLC had received no prior local therapy for BM. Six of 18 patients (33%) achieved intracranial response. Systemic RR was 33%, with only 1 patient who progressed in the CNS while responding systemically. All other responses were concordant. Neurologic AEs in the NSCLC cohort were all grade ≤2. A recent update included data from 34 patients with PD-L1–positive tumors and 5 patients with PD-L1–negative disease. CNS RR was 10 of 34 (29.4%) in PD-L1–positive patients, with a median duration of response of 10.7 months. Discordance was observed between intracranial and systemic responses in 7 patients. Among these, four individuals experienced PD in the brain and partial response (PR) systemically while the remaining 3 patients exhibited the opposite findings. Interestingly, no intracranial responses were observed among patients with PD-L1–negative tumors. Treatment was well tolerated, with no grade ≥2 treatment-related neurologic AEs reported (48). However, no patients with BM greater than 20 mm were included, and potentially worrying lesions were prophylactically treated with local therapy prior to trial entry (49).

After anti-PD-L1 mAbs were approved for the second-line treatment of advanced NSCLC, evidence from real-world data have been published, confirming that ICI show variable activity against untreated or previously treated progressing BM (50–54). Unfortunately, the majority of real-world experiences have been published as abstracts, and limited information is available making thorough analysis difficult.

A recently published pooled analysis of patients from seven European centers included patients with NSCLC treated with ICI in a variety of settings including daily standard practice (55). Among patients with active BM that were evaluable in the CNS (n = 73), the intracranial RR was 27.3%. Among the 23 patients with active BM and available PD-L1 expression status, positive PD-L1 expression (≥1%) was associated with a higher intracranial RR: 35.7% versus 11.1% in PD-L1–negative patients (55). In a multivariate analysis, BM was not significantly associated with PFS or OS. However, the presence of active BM was significantly associated with poorer survival within the BM subgroup (55).

The potential activity of exclusively ICI-based systemic treatment against BM has been confirmed in similarly designed clinical trials conducted in patients with melanoma (56–59). Nivolumab has also been evaluated in patients with ccRCC with active BM (60), with an activity that seems more modest than in melanoma.

Overall, preliminary data from patients with NSCLC and ccRCC, and more mature data from melanoma trials, suggest that intracranial response is more dependent on the immunogenicity of tumor cells than on the location of metastatic disease or BBB penetration (Table 1). There is considerable evidence to claim that in the subset of patients with metastatic NSCLC or melanoma who obtain clinical benefit from ICI's anticancer activity, clinical benefit extends to the intracranial compartment. Although there might be a perception showing that patients with CNS involvement from melanoma may obtain a significantly larger benefit from immunotherapy (Table 1), different clinical trials design and the more frequent combination regimens employed may perfectly explain that observation. In addition, also higher systemic response rates have been reported among patients with melanoma compared with NSCLC individuals.

Although data are still immature, it is expected that in highly selected patients with NSCLC, anti-PD-L1 agents can reach an activity that is sufficient to shift BM management algorithms toward progressively systemic-based approaches. If clinical and biological characteristics suggest ICI can be active in the CNS, patients with an increasing number of BM could be considered for stereotactic radiotherapy (SRT) instead of whole-brain radiotherapy (WBRT) and even radiotherapy-sparing strategies could be evaluated in carefully selected patients with a very favorable predictive profile. A therapeutic approach is proposed in Fig. 3, which integrates the currently available predictors of intracranial activity of anti-PD-(L)1 agents in NSCLC BM. However, until more robust clinical trials are conducted, patients with NSCLC with BM should be individually evaluated by multidisciplinary tumor boards in highly experienced centers.

In spite of the encouraging results reviewed above that challenge traditional notions of how anticancer drugs perform in the brain, the majority of patients will be affected by CNS disease progression. Because of the particular anatomic configuration of the CNS, in which tumor growth can be more detrimental than in any other anatomic region, combining local therapy (mainly radiotherapy) and systemic therapy should be carefully considered to increase the probability of obtaining local disease control and to avoid burdensome neurologic symptoms and clinical deterioration (61), particularly if systemic regimens have a low RR. The safety and activity of ICI in the subgroup of patients with previously irradiated and controlled BM have been reported in a number of pivotal advanced NSCLC clinical trials.

A pooled analysis included data from five clinical trials with advanced NSCLC patients that were treated with atezolizumab. In the safety cohort (n = 1,452), serious neurologic AE incidence was higher in patients with BM compared with those without BM (6 vs. 3%), but none reached grade 4 (62). An efficacy subgroup analysis of patients with BM included in the OAK trial (23) confirmed that atezolizumab was superior to docetaxel in terms of delaying the appearance of new symptomatic BM [HR, 0.38; 95% confidence interval (CI): 0.16–0.91; ref. 63].

Real-world data from a nivolumab expanded access program have been reported in recent years including patients with squamous and nonsquamous NSCLC. A total of 38 patients with squamous NSCLC and asymptomatic and controlled BM were included. With a median follow-up of 4.5 months, 1 patient obtained complete response (CR), 6 reached partial response (PR), and 11 stable disease (SD; ref. 64). The authors did not specify whether responses were intracranial or systemic. Only one patient discontinued treatment due to AEs. Among the patients with nonsquamous NSCLC (n = 409), the authors compared overall response and survival outcomes between patients with BM and all patients included in the program (n = 1,588). Overall RR was 17% in patients with BM and 18% in the entire cohort. Median PFS and OS was 3 and 8.6 months, respectively, in patients with BM, and 3 and 11.3 months, respectively, in the entire cohort. No relevant safety differences were reported between both groups. Previously mentioned reports have included an advanced NSCLC population that predominantly included patients with controlled BM, but their relevance is limited due to their retrospective design, limited sample sizes, and patient heterogeneity (50, 51, 54, 65). Some reports, however, suggest that brain RR in the real-world setting might be lower than in the more selective clinical trial population (41).

The innovative combination of platinum-based chemotherapy and pembrolizumab has also been evaluated in NSCLC patients with BM. The pivotal phase III trial KEYNOTE-189 included patients with previously treated stable and untreated asymptomatic BM (with no lesions larger than 1.5 cm; ref. 66). It has been recently reported that the benefit of the combination in terms of PFS and OS was confirmed in the subgroup of patients with BM: HR (95% CI) of 0.42 (0.27–0.67) and 0.41 (0.24–0.67), respectively. Notably, the magnitude of benefit attributable to the addition of pembrolizumab was higher in patients with BM than in those without CNS involvement (67). A real-world retrospective cohort experience with the same combination of drugs reported similar results (68).

The preliminary reports on the activity of ICI in the CNS are encouraging, and suggest that when patients are given systemic immunotherapy, the brain should not be considered a fully protected biological sanctuary. However, even in the most favorable scenarios such as first-line pembrolizumab for patients with PD-L1 expression ≥50% or its combination with platinum-based chemotherapy, RR lie below 50% (26, 66). In both settings, however, (first-line pembrolizumab monotherapy or first-line pembrolizumab combined with chemotherapy), immunotherapy has shown to be active against CNS involvement, as discussed previously. Nevertheless, until more active immunotherapeutic agents are developed, the brain continues to be an extremely fragile organ in which the best of anticancer treatment modality combinations are often required to avoid symptomatic local PD. Currently, this means combining radiotherapy and systemic treatment for the vast majority of patients with NSCLC BM and no target mutations. In this scenario, the approval of anti-PD-1/PD-L1 mAbs for advanced NSCLC raises questions about the optimal radiation modality and timing relative to ICI administration that should be offered to patients with NSCLC BM.

There is a strong rationale behind the combination of radiotherapy and ICI, both preclinical and clinical. Before the era of clinical cancer immunotherapy, the abscopal effect reported anecdotally suggested that the response of tumors distant from the irradiation field could be immune-mediated (69). This hypothesis was supported by loss- and gain-of-function experiments using T-cell–deficient athymic nude mice and dendritic cell enhancers, respectively (70). Some preliminary reports suggest that immune enhancers can significantly contribute to abscopal responses in patients with a variety of solid cancers (71), but results from confirmatory trials combining modern ICI and radiotherapy in BM patients are awaited. Moreover, additional complexity is conferred to trial design by the fact that preclinical and clinical data suggest that appropriate timing and dosing of irradiation might be critical for the induction of an effective antitumor immune response (72, 73).

Synergism or additivity with ICI and radiation can be assessed in the CNS. Increased PD-L1 expression in surgically resected BM specimens after radiation, for instance, supports this potential interplay (37). Interestingly, a variety of retrospective analyses from observational studies have suggested that the combination of radiotherapy and ICI could have a positive impact on patient survival over exclusively systemic treatment (74–76).

Data regarding the best brain radiotherapy modality and timing relative to ICI administration remain scarce. Initial case reports supported the rationale behind the concern that the peculiar PK/PD of ICI, which are highly dependent on the immune infiltration by cytotoxic T lymphocytes (CTLs), and the anatomic configuration of CNS could lead to neurologic toxicity (77).

Multiple reports suggest that treatment with anti-PD-1 mAbs could exacerbate symptomatic RN (57, 78, 79). Pathologic examination of surgical samples reveals extensive necrosis with residual tumor cells surrounded by an area with signs of active vasculitis (predominantly infiltrated by CD3+ and CD68+ cells), blood vessel hyalinization and astrocytosis (78). Findings are often hardly distinguishable from previously reported radiation-induced vasculitic leukoencephalopathy associated with radiosurgery alone (80). An immune infiltration analysis from a surgically removed BM of a patient who developed RN during treatment with pembrolizumab is provided in Fig. 4. Although clear vasculitis was not described in this case, the observed dense immune infiltration consisting predominantly of T lymphocytes, which is consistent with previous reports (78), suggests ICI could potentially exacerbate RN.

The largest series of NSCLC patients treated with a variety of radiation modalities and timings relative to ICI was recently published. Safety data were reported from patients with NSCLC BM who received brain irradiation therapy and anti-PD-1/PD-L1 antibodies at any time and compared to similar patients who were anti-PD-1/PD-L1-naïve (81). In this retrospective single-center analysis, subgroups were designated by radiotherapy modality (SRS, partial brain irradiation, or WBRT) and timing according to systemic anti-PD-1/PD-L1 antibody administration (previous, concurrent, and subsequent). In addition to common AEs, a category was recorded to specifically evaluate the occurrence of symptomatic image worsening that resolved with no significant changes in treatment or was confirmed malignancy-free on histopathologic evaluation. No statistically significant differences were observed between patients who received ICI and those who did not according to radiation modalities or timing (81). Another retrospective report from a small series has supported that the neurologic toxicity of combining ICI and brain radiotherapy is manageable, with no patients undergoing surgical resection for symptomatic radionecrosis among 17 patients who received SRT and nivolumab or durvalumab (82). However, a recently published retrospective evaluation comparing the incidence of symptomatic radionecrosis in patients (n = 480) treated with brain SRT for metastatic NSCLC (n = 294), melanoma, or RCC who received ICI (including ipilimumab; n = 115) and in those who did not (ICI-naïve), treatment with ICI was significantly associated with an increased risk of symptomatic radionecrosis regardless of tumor histology. Notably, a tendency toward increased symptomatic radionecrosis remained, but statistical significance was not reached when patients who received ipilimumab were excluded from the analysis (83). Of note, the median time of radionecrosis occurrence after SRT in patients treated with immunotherapy has consistently been reported to be above 10 months (57, 79). Therefore, the prolonged survival experienced by patients treated with ICI could have biased the increased radionecrosis incidence reported in these patients compared with other systemic treatment modalities. An innovative clinical trial (NCT02681549) that is currently recruiting patients with melanoma and NSCLC BM will evaluate whether bevacizumab in combination with pembrolizumab is capable of reducing brain edema and radionecrosis incidence while potentially synergizing with immune cell trafficking (57).

Radionecrosis can affect up to 30% of patients who have received ICIs (57), with incidences between 7% and 20% when symptomatic radionecrosis is specifically reported (83, 84). This phenomenon challenges treating physicians to establish a correct diagnosis between treatment-related changes (both pseudoprogression and radionecrosis) and true PD. Some recently published brain response evaluation criteria have incorporated the clinical status of the patient and corticosteroid use into decision algorithms to differentiate pseudoprogression from PD (49, 85). In addition, because tumor growth due to pseudoprogression is expected to be transient, several cooperative groups have suggested repeated imaging after initial radiologic PD to enhance the identification of patients who will ultimately benefit from ICI (86). The immune Response Evaluation Criteria in Solid Tumors (iRECIST) and the immunotherapy Response Assessment for Neuro-Oncology (iRANO) criteria have stressed the importance of repeated imaging, preferably at least 4 weeks or 3 months after first image evaluation that met radiologic PD criteria, respectively, to discriminate between true PD or pseudoprogression (85). The third diagnostic possibility, radionecrosis, has also been described as a transient image worsening that is followed by regression or stability (79, 87), suggesting that active imaging surveillance strategies might also be appropriate when radionecrosis is suspected. Complementarily, imaging techniques such as positron emission tomography and magnetic resonance spectroscopy and perfusion imaging have been evaluated to differentiate PD from radionecrosis, but noninvasive diagnosis remains uncertain (88). Tumor response evaluation criteria that require PD confirmation are encouraged in both clinical trials and daily practice. From a therapeutic perspective, clinical trials are strongly encouraged to develop novel therapeutic strategies for radionecrosis (e.g., bevacizumab) that could contribute to limiting the immunosuppressive effects associated with first-line corticosteroids (89). Interestingly enough, some previous evidence has shown that bevacizumab may actually have the potential to prevent brain metastases in patients with nonsquamous NSCLC (90). Complementarily, some authors have suggested that the minimum diameter of a BM to be considered measurable should be lowered to 5 mm (using high-resolution MRI with a slice thickness of ≤2 mm for a reliable response assessment) to allow trial inclusion of patients with small BM in which ICI have shown promising activity (49).

The unique PK/PD of ICI are challenging the traditional idea that mAbs are marginally active against BM. There is encouraging preliminary evidence to support the development of clinical trials in which this treatment modality is evaluated alone or combined with radiotherapy. On the basis of the currently available evidence, anti-PD-1/PD-L1 agents play a relevant role when long-term disease control is attempted in the CNS of patients with advanced NSCLC. However, predictive selection criteria are required to identify individuals with the greatest probability of response to anti-PD-(L)1 agents. In these carefully selected patients, SRT can be favored over WBRT or radiotherapy-sparing strategies can be considered, although these approaches require confirmation in more robust clinical trials.

Despite the encouraging activity of ICI in the CNS, regimens with greater activity than anti-PD-1/PD-L1 monotherapy are urgently needed, and prospective trials in patients with NSCLC BM are warranted. Improved imaging modalities are needed to differentiate between radionecrosis, pseudoprogression, and tumor regrowth in previously irradiated lesions to identify patients who will ultimately obtain clinical benefit from the systemic delivery of ICI.

S.B. Goldberg is a paid consultant for AstraZeneca, Boehringer Ingelheim, Eli Lilly, Bristol-Myers Squibb, Genentech, Amgen, and Spectrum, and reports receiving commercial research grants from AstraZeneca and Boehringer Ingelheim. H.M. Kluger reports receiving commercial research grants from Apexigen, Bristol-Myers Squibb, and Merck, is an unpaid consultant/advisory board member for Nektar, Biodesix, Roche-Genentech, Iovance, Immunocore, Celldex, Array Biopharma, Merck, and Elevate Bio, and reports receiving other remuneration from Pfizer. J. Corral reports receiving speakers bureau honoraria from Pfizer, Bristol-Myers Squibb, Boehringer-Ingelheim, Eli-Lilly, Roche, MSD, and AstraZeneca, and is an unpaid consultant/advisory board member for AstraZeneca, Pfizer, Bristol-Myers Squibb, Roche, Eli-Lilly, MSD, and Boehringer-Ingelheim. K.A. Schalper is a paid consultant for AstraZeneca, Clinica Alemana de Santiago, Dynamo Therapeutics, EMD Serono, Merck, Moderna, Pierre Fabre, Shattuck Labs, Takeda, Torque Therapeutics, Ono Pharmaceuticals, Agenus, Abbvie, and Celgene, reports receiving commercial research grants from AstraZeneca, Bristol-Myers Squibb, Eli Lilly, Merck, Moderna, Navigate Biopharma, Pierre Fabre, Surface Oncology, Takeda, and Tesaro, and reports receiving speakers bureau honoraria from Bristol Myers Squibb, Merck, Fluidigm, and PeerView. R.S. Herbst is a paid consultant for Abbvie Pharmaceuticals, ARMO Biosciences, AstraZeneca, Biodesix, Bolt Biotherapeutics, Bristol-Myers Squibb, Eli Lilly and Company, EMD Serono, Genentech/Roche, Genmab, Halozyme, Heat Biologics, IMAB Biopharma, Immunocore, Infinity Pharmaceuticals, Loxo Oncology, Merck and Company, Midas Health Analytics, Mirati Therapeutics, Nektar, Neon Therapeutics, NextCure, Novartis, Pfizer, Sanofi, Seattle Genetics, Shire PLC, Symphogen, Takeda, Tesaro, Tocagen, Spectrum Pharmaceuticals, and Junshi Pharmaceuticals, and reports receiving commercial research grants from AstraZeneca, Eli Lilly and Company, Genentech/Roche, and Merck and Company. I. Gil-Bazo reports receiving speakers bureau honoraria from AstraZeneca, Bristol-Myers Squibb, Roche, Guardant Health, MSD, Boehringer Ingelheim, and Eli Lilly. No potential conflicts of interest were disclosed by the other authors.

This work was funded by the Stand Up To Cancer - American Cancer Society Lung Cancer Dream Team Translational Research Grant SU2C-AACR-DT17-15. Stand Up to Cancer is a division of the Entertainment Industry Foundation. Research grants are administered by the American Association for Cancer Research, the scientific partner of SU2C. This study was also supported partially by Yale SPORE grant in skin cancer P50 CA121974 (PIs: Kluger and Bosenberg) and SPORE grant in Lung Cancer PA50 CA196530 (PI: Herbst).