Predictive role of intracranial PD-L1 expression in a real-world cohort of NSCLC patients treated with immune checkpoint inhibition following brain metastasis resection

Patients with histologically proven NSCLC either based on resected or biopsied tissue originating from primary, extracranial metastatic lesion or lymph node metastasis as well as histologically confirmed NSCLC brain metastasis during their disease between January 2016 and April 2023 were included in this retrospective study (Table 1, Supplementary figure 1). All patients underwent brain metastasis resection and were treated inpatient and outpatient at the Department of Neurosurgery and Department of Thoracic Oncology at one of the three sites of the University Medical Center of the Charité. Most patients had synchronous brain metastases at the time of diagnosis and were treatment-naïve at the time of brain metastasis resection, whereas 10 patients had metachronous brain metastasis and received systemic pre-treatment before neurosurgical brain metastasis resection (Table 1). Clinical data, data from imaging studies, histopathological and treatment-related characteristics were retrieved from electronic patient records. Baseline characteristics including graded prognostic assessment (GPA) and Karnofsky performance score (KPS) are summarized and potential confounding factors of intracranial PD-L1 expression including systemic pre-treatment with CTx or ICI or daily treatment with dexamethasone before brain metastasis resection are listed in Table 1. Twenty patients have had no extracranial PD-L1 TPS assessment as these patients did not undergo sampling of extracranial tissue. Missing values were found in the following variables: two missing values for intracranial Ki67, 33 missing values for extracranial Ki67. For six patients daily dexamethasone doses were incomplete. Evaluation of extra- and intracranial disease progression was based on reports from board-certified radiologists. Tumor volumes of the resected lesion were quantified using a semi-automated 3D rendering algorithm in iPlannet (Brainlab, Munich, Germany) using the SmartBrush tool. As for CT staging and MRI follow-up, radiologic tumor assessment was analyzed for treatment efficacy with response evaluation following the discretion of the treating physician according to standardized response evaluation criteria in solid tumours (RECIST) assessment (version 1.1) and immunotherapy response assessment in neuro-oncology (iRANO) criteria (complete response (CR), partial response (PR), stable disease (SD), or progressive disease (PD)) based on retrospective chart reviews without central confirmation. Progression was categorized as either extracranial progression-free survival (ecPFS) or intracranial progression-free survival (icPFS) and separate extracranial and intracranial PFS estimates were calculated. Median time of follow-up was estimated using the reverse Kaplan–Meier method. All outcome parameters including overall OS, icPFS and ecPFS were defined as the time from neurosurgical brain metastasis resection until death or censoring at last known time alive.

Specimens from resected (intracranial) brain metastasis tissue (n = 64) and available matched extracranial tumor tissue (n = 44, comprising 9 resected primary lung tumors, 9 lung biopsies and 26 lymph node biopsies) were considered for analysis. Each case was reviewed by a pathologist or neuropathologist. FFPE tissue sections of 2–3 µm thickness were used for hematoxylin and eosin (H&E) staining and immunohistochemistry (IHC). IHC staining for PD-L1 was performed using a Leica Bond immunostainer (Leica Biosystems) according to manufacturer’s protocols. Briefly, tissue sections were deparaffinized, rehydrated, and heat-induced antigen retrieval was performed by incubation in CC1 mild buffer (Ventana Medical Systems) for 30 min at 100 °C. Subsequently, tissue sections were incubated with the primary monoclonal antibody rabbit anti-PD-L1 (Clone E1L3N, Cell Signaling, #13684) at 1:200 for 60 min followed by incubation with HRP-conjugated secondary antibody (Leica Biosystems) for 32 min, DAB incubation and counterstaining of nuclei with hematoxylin. Tonsil served as positive control. PD-L1 expression was quantified by assessing the tumor proportion score (TPS), i.e. assessing the percentage of tumor cells with positive membranous staining relative to all vital tumor cells. Only cases with at least 100 evaluable tumor cells were included [25].

Descriptive statistics were performed to summarize the presented patient cohort and associated clinical, histopathological, radiological, and treatment-related patient features. Data collection was done with Excel version 14.3.9 (Microsoft). We used R version 1.1.442 (R Foundation) to compute statistics, including frequencies, means, and SDs, to characterize the cohort. For patient baseline characteristics, continuous data were compared across cohorts using the Wilcoxon rank-sum test, while categorical data were compared using Fisher exact or χ2 tests. We used the gtsummary package (version 1.7.2, R Foundation) to describe tabular data of the patient cohort, including categorical and numerical variables. Optimal cut points of PD-L1 TPS for icPFS, ecPFS and OS were determined using the maxstat package (version 0.7–25) displaying maximally selected rank statistics (M) and p-values as well as distribution of optimal cut point values for a given predictor [26]. Median icPFS, ecPFS and OS was estimated by Kaplan–Meier analysis with 95% CI bands being displayed in light color; and log-rank test was used to compare OS and PFS between patient groups; plotting was performed using the survminer package (version 0.4.3, R Foundation). Multivariable Cox regression model for OS, ecPFS and icPFS were computed, including only complete cases, using clinical covariates assumed to be associated with the respective outcomes. Further R packages used for analysis included dplyr (version 1.1.4.), tidyverse (version 2.0.0), and swimplot (version 1.2.0). A p-value < 0.05 was considered significant with p-values being 2-sided. R code and raw data will be made available to researchers on request.

Sixty-four patients who underwent resection of NSCLC brain metastases were included in the study (Supplementary figure 1). Baseline was defined as the time of first brain metastasis resection. Patient characteristics at baseline are summarized in Table 1. Median follow-up time from first brain metastasis resection was 57.1 months [95% CI: 40.0—NA]. Fifty-four patients (84.4%) were diagnosed at UICC stage IV with synchronous brain metastasis, therefore had been treatment-naïve before brain metastasis resection. Ten patients suffered from metachronous brain metastasis and have had systemic pre-treatment with first-line systemic treatment in a palliative (8 patients) or curative (2 patients) setting including platin-based CTx, with two of these patients receiving CTx + ICI and two patients receiving ICI monotherapy before brain metastasis resection. Only one of the pre-treated patients received both first line (1L) and second line (2L) systemic pre-treatment before brain metastasis resection. Two patients within the group of pre-treated patients received a brain metastasis-specific local treatment including SRS before brain metastasis resection in addition to systemic treatment. After brain metastasis resection, 39 patients received a combination of ICI and platin-based CTx followed by maintenance ICI, whereas 25 patients received ICI monotherapy (Table 1). ICI treatment was administered for a median number of 9 cycles (95% CI: 6 – 13) and mean duration of 207 days (95% CI: 138 – 276). At time of brain metastasis resection, 27 patients (42.2%) had additional extracranial organ metastases, whereas 37 patients (57.8%) showed no sign of extracranial metastasis. 49 patients (76.6%) had two or more brain metastases. Median cumulative daily dexamethasone dose during a time period of 13 days before operation was 44.0 mg (0.0 – 120.0, IQR). There was no correlation between intracranial PD-L1 TPS and cumulative pre-operative dexamethasone, or other clinical parameters (Supplementary figure 2). The clinical course of included patients is displayed in Fig. 1. All 64 patients of the study cohort receiving postoperative RTx + ICI showed better OS, ecPFS and icPFS compared to patients undergoing other postoperative therapies after brain metastasis resection, including RTx, RTx + CTx or RTx + targeted therapies (Supplementary figure 3A-C).

The median time between tissue sampling of brain metastasis and extracranial tissue was 15 days (95% CI: 9 – 33). Overall positive rates for intracranial and extracranial PD-L1 TPS (i.e. values ≥ 1%) were 54.7% and 68.2%, respectively (Supplementary table 1). Although we observed a significant correlation between intra- and extracranial PD-L1 TPS (Supplementary figure 3), there was a trend towards higher extracranial PD-L1 TPS with a median of 15.0% compared to a median intracranial PD-L1 TPS of 7.5% (p = 0.113) (Fig. 2A, Supplementary table 1).

In 44 patients with available matched intra- and extracranial tissues, we found a significant discordance towards higher extracranial PD-L1 TPS (p = 0.013) (Fig. 2B). Exemplary PD-L1 IHC images showing concordant and discordant cases are shown in Fig. 2C-F. Among these 44 patients, median PD-L1 TPS discordance was 9.5% (9.0–95.0, IQR), with 16 patients showing no discordance (35.6%), 29 patients (64.4%) having a discordance of ≥ 1%, and 18 of whom (40.0%) exhibiting a discordance of ≥ 10% (Supplementary table 1).

Since we observed discordant PD-L1 TPS in intra- and extracranial tissues, we analyzed the predictive role of both intra- and extracranial PD-L1 TPS in patients treated with RTx + ICI after brain metastasis resection. Results of maximally selected rank statistics, which was used for cut point determination with respect to outcomes of interest, i.e. icPFS, ecPFS or OS, are summarized in Supplementary table 2.

To predict icPFS, an optimal cut point was identified for intracranial PD-L1 TPS at 40%, while the optimal cut point for extracranial PD-L1 TPS was determined to be 70% (Fig. 3A + C). Patients exhibiting a high intracranial PD-L1 TPS of > 40% vs. ≤ 40% demonstrated a significantly improved median icPFS of 54.8 vs. 15.4 months (p = 0.0036) (Fig. 3A-B). Similarly, patients showing a high extracranial PD-L1 TPS of > 70% vs. ≤ 70% had a significantly increased non-estimable median icPFS vs. 9.13 months (p = 0.0037) (Fig. 3C-D).

Regarding ecPFS, we identified particularly lower cut points for intracranial and extracranial PD-L1 TPS, being 5% and 0%, respectively (Fig. 3E + G). High intracranial PD-L1 TPS of > 5% vs. ≤ 5% was associated with an increased non-estimable median ecPFS vs. 20.8 months (p = 0.034) (Fig. 3E-F). Likewise, patients exhibiting an extracranial PD-L1 TPS of > 0% vs. 0% demonstrated a longer non-estimable median ecPFS vs. 15.5 months (p = 0.0068) (Fig. 3G-H).

As for OS, the optimal cut points for intracranial and extracranial PD-L1 TPS were established at 70% and 0%, respectively (Fig. 3I + K). Here, patients with an intracranial PD-L1 TPS of > 70% vs. ≤ 70% had a longer median OS of 74.5 vs. 22.6 months (p = 0.0096) (Fig. 3I-J). Similarly, patients exhibiting an extracranial PD-L1 TPS of > 0% vs. 0% showed an increased OS of 29.5 months vs. 11.0 months (p = 0.011) (Fig. 3K-L).

Similarly, we investigated the predictive value of PD-L1 TPS discordance between intracranial and extracranial PD-L1 TPS. Here, patients exhibiting a PD-L1 TPS discordance of > 20% vs. ≤ 20% showed a better median icPFS of 54.8 months vs. 16.1 months (p = 0.047) (Supplementary figure 3A-B). PD-L1 TPS discordance was not found to be a significant predictor of ecPFS and OS (Supplementary figure 3C-F).

For multivariable analyses, patients were split into two groups according to the previously determined optimal cut point of intracranial PD-L1 TPS to predict icPFS, i.e. 40% (Table 2). Accordingly, patients with a high intracranial PD-L1 TPS of > 40% had a lower risk of intracranial disease progression (HR 0.13, p = 0.008) (Fig. 4A), extracranial disease progression (HR 0.17, p = 0.041) (Fig. 4C), as well as a lower risk of death (HR 0.29, p = 0.016) (Fig. 4E). The positive predictive value of high intracranial PD-L1 TPS could also be observed in a subgroup of patients who were treatment-naïve before brain metastasis resection for icPFS and OS, but not for ecPFS (Supplementary figure 5A, C, E). The optimal cut point of extracranial PD-L1 TPS to predict ecPFS, i.e. 0%, was not independently associated with patient outcome (Fig. 4B, D, F). In the subgroup of treatment-naïve patients, an extracranial PD-L1 TPS of > 0% was associated with a lower risk of extracranial disease progression (HR 0.09, p = 0.034) (Supplementary figure 5B, D, F).