Melanoma-derived brain metastases (MBM) represent an unmet clinical need because central nervous system progression is frequently an end stage of the disease. Immune checkpoint inhibitors (ICI) provide a clinical opportunity against MBM; however, the MBM tumor microenvironment (TME) has not been fully elucidated in the context of ICI. To dissect unique elements of the MBM TME and correlates of MBM response to ICI, we collected 32 fresh MBM and performed single-cell RNA sequencing of the MBM TME and T-cell receptor clonotyping on T cells from MBM and matched blood and extracranial lesions. We observed myeloid phenotypic heterogeneity in the MBM TME, most notably multiple distinct neutrophil states, including an IL8-expressing population that correlated with malignant cell epithelial-to-mesenchymal transition. In addition, we observed significant relationships between intracranial T-cell phenotypes and the distribution of T-cell clonotypes intracranially and peripherally. We found that the phenotype, clonotype, and overall number of MBM-infiltrating T cells were associated with response to ICI, suggesting that ICI-responsive MBMs interact with peripheral blood in a manner similar to extracranial lesions. These data identify unique features of the MBM TME that may represent potential targets to improve clinical outcomes for patients with MBM.

Brain metastases, arising most commonly from lung cancer, breast cancer, and melanoma (1), represent the most common type of intracranial tumor, occurring in 20% to 40% of patients diagnosed with cancer (1). Metastatic tumors disseminating into the central nervous system (CNS) are associated with a poor prognosis and are traditionally treated with either surgical resection and/or radiotherapy. Despite this standard of care, intracranial metastases are associated with significant morbidity and median survival ranges from 3 to 27 months (2). Given their increasing prevalence, limited treatment options, and historical exclusion from clinical trials, brain metastases represent an unmet clinical need.

Immune checkpoint inhibitors (ICI) have revolutionized the treatment of cancer with approval in 19 cancer types and two tissue-agnostic indications (3). With the growing clinical success of immune checkpoint modulation, attention has shifted toward the potential activity of ICI in treating brain metastases. Studies of agents targeting CTLA4 and PD1 have demonstrated promising ICI-mediated intracranial response rates for metastatic melanoma, with combination therapy achieving similar rates of response observed extracranially (4).

Despite these promising clinical results, approximately half of all patients with melanoma progress on ICI as a result of either innate or acquired resistance. Although extracranial disease progression occurs through both tumor-intrinsic (5) and -extrinsic (6) mechanisms, a paucity of data exists investigating determinants of intracranial response and/or resistance to ICI. The CNS features an immune-specialized microenvironment (7, 8) and the interaction of these CNS-unique elements with ICI is not fully understood. Clinical evidence suggests that the presence of extracranial lesions influences immune-based therapies for brain metastases (9). Potential mechanisms for this process include T-cell priming at extracranial sites of disease and T-cell trafficking into the brain where shared intracranial and extracranial tumor antigens are targeted.

Several features of response and resistance to ICI in the extracranial setting have been investigated. At a genomic level, the overall tumor mutational burden (TMB) has been associated with clinical response to anti-PD1 therapy (10) in addition to predicted neoantigen load at pretreatment timepoints (11). In addition, many studies have investigated the role of T-cell infiltration prior to ICI (12), the spatial distribution of CD8+ T cells along tumor margins (12), and the extent of PD-L1 expression within the tumor (13, 14). With the advent of single-cell sequencing techniques to dissect immune cell phenotypes, further insights have been made into phenotypic T-cell states that are enriched within responding tumors (15). To explore these features within the intracranial context, we utilized single-cell RNA sequencing (scRNA-seq) from a cohort of immunotherapy-naïve and post–ICI-treated melanoma-derived brain metastases (MBM) paired with T-cell receptor sequencing (TCR-seq) from patient-matched blood and extracranial lesions to describe the diversity of immune and malignant cellular elements within the tumor microenvironment (TME), to identify associations between those elements, and to suggest TME-related biomarkers of intracranial response to ICI.

Collection of fresh tissue for scRNA-seq

This study was conducted in accordance with recognized ethical guidelines and patient samples were collected under the Dana-Farber Cancer Institute's (DFCI) Institutional Review Board–approved protocol 10-417, Tissue Bank for Neurological Disorders, in accordance with the Declaration of Helsinki and the Belmont Report. Selection criteria for the study included prospectively enrolled patients undergoing a craniotomy for MBM who could provide written, informed consent. Thirty-two sequentially collected MBM from 27 patients were obtained immediately following craniotomy for resection of tumor. Patients undergoing stereotactic brain biopsy were excluded. Tumors were classified according to prior exposure and therapeutic response to ICI. Patient intracranial and extracranial clinical responses were categorized as responder, partial responder, and nonresponder after review of clinical history and imaging with board-certified medical oncologist, neuro-oncologist, and radiologist. An overview of the clinical features of the cohort is shown in Supplementary Table S1. Collection of solid tumor tissue and blood samples was performed to reduce the time between collection and processing. Approximately 20 mL of blood was collected at any timepoint during the patient's surgery with coordination and permission of the anesthesiology team. Blood was collected in two 10 mL EDTA tubes that we provided. Immediately after collection, a portion of the blood was used to separate plasma, and another portion of whole blood was stored at 4°C for eventual genomic DNA (gDNA) extraction. Solid tumor tissue was collected via the coordination of lab technicians with the participant's surgical team. Lab technicians collected tissue directly from the operating room and brought it to the pathology team to approve a portion of the tissue to be used for research. The tissue was then immediately brought back to the lab to begin the dissociation workflow (described below). Extracranial lesions were obtained from pathology where they were stored following routine clinical care or following autopsy.

Tumor dissociation workflow

After collection, the specimen was transferred to a sterile petri dish and mechanically dissociated with a scalpel. The dissociation mixture used 4 mL of preheated buffer X, 40 μL of buffer Y, 50 μL buffer N, and 20 μL of enzyme A from the Miltenyi Biotec Brain Tumor dissociation kit, a papain-based dissociation kit (Miltenyi Biotec, catalog no. 130-095-942). The tumor was combined with the digestion buffer and incubated on a rotator at 37°C for 30 minutes. The tissue was then resuspended thoroughly via pipette, passed through a 100 μm cell strainer and spun down at 188 × g for 3 minutes. The pellet was washed with 5 mL of Hank's Balanced Salt Solution with calcium and magnesium and ultimately resuspended in 90 μL of PBS + 1% BSA (Sigma-Aldrich, catalog no. A8022).

FACS

The dissociated tumor cell mixture was stained with 10 μL of CD45-Vio-blue (Miltenyi Biotec, catalog no. 130-113-122) and 10 μL of CD3-PE (Miltenyi Biotec, catalog no. 130-113-139), then incubated on ice for 20 minutes. Following this incubation,1.5 μL Calcein-AM live cell stain (Thermo Fisher Scientific, catalog no. C1430) and 0.5 μL To-Pro-3 Iodide dead cell stain (Thermo Fisher Scientific, catalog no. T3605) were added to the cell suspension. The same cells were used for unstained controls to adjust gating. Following dissociation and staining, single cells were purified by FACS with the BD FACSAria Fusion instrument, which employs five lasers (405, 488, 640, 355, and 561 nm). Cells were sorted into fully skirted 96-well plates with 10 μL of buffer TCL (Qiagen, catalog no. 1031576) + 1% beta-mercaptoethanol (American Bio, catalog no. AB01340). Set up for each experiment used a 100 μmol/L nozzle at 20 psi and 31 kHz. Prior to fluorescence gating, doublets were gated out using three successive gates: forward scatter (FSC) area versus height, FSC area versus width, and side scatter area versus width. We then gated conservatively and specifically for live cell populations as Calcein AM positive and To-pro-3 negative. Cells were then assessed for CD45 and CD3 staining, with CD45, all CD45+ and CD45+CD3+ cells sorted into individual plates. An example of this gating strategy is provided as in Supplementary Fig. S1. After sorting, plates were immediately spun down at 188 × g for 1 minute, flash frozen at −80°C and stored for future scRNA-seq. Flow sorting analysis was completed with the FACSDiva (v. 8.0.1) and FlowJo (v. 10).

scRNA-seq

scRNA-seq was performed using the Smart-Seq 2 protocol to create an 8-plate (4 CD45, 2 CD45+, and 2 CD3+) cDNA library for each patient tumor sequenced. RNA isolation was first completed by resuspending each well with 22 μL of RNAclean XP beads (Beckman, catalog no. A63987) and transferring the mixture to a demi-skirted 96-well PCR plate. The cells were incubated at room temperature in the bead suspension, then transferred to a magnetic plate and washed twice with 80% ethanol. Reverse transcription was performed using Maxima H minus reverse transcriptase (Thermo Fisher Scientific, catalog no. EP0753) and 10 μmol/L of TSO oligonucleotides (Exiqon), and incubation at 42°C for 90 minutes then 10 cycles of 50°C for 2 minutes, 42°C for 2 minutes, then heat inactivation at 70°C for 15 minutes. Full-length cDNA amplification with Hi-Fi Hotstart Readymix (Roche, catalog no. 07959079001) was completed with a 98°C incubation then 21 cycles of 98°C for 15 seconds, 67°C for 20 seconds, 72°C for 6 minutes, and a final extension at 72°C for 5 minutes. Amplified cDNA was purified with AmpureXP beads (Beckman, catalog. no. A63881) and cleaned with two 80% ethanol washes on the plate magnet. An Agilent BioAnalyzer high sensitivity DNA chip (Agilent Technologies, catalog no. 5067-4626) was used to ensure proper distribution and fragment length of the cDNA library. cDNA library concentration was measured using the Qubit dsDNA HS assay (Thermo Fisher Scientific, catalog no. Q32854) according to the manufacturer's protocol and values were read on a microplate reader. Using the Qubit assay concentration values, all library wells were diluted to 2 ng with water in order to proceed to library preparation. The Nextera XT Library Prep Kit (Illumina, catalog no. FC-131-1096) protocol was used for fragmentation and unique barcoding of the cDNA libraries. For each plate, 1.5 μL per well was pooled and purified with the Ampure XP beads (0.9× the amount of sample) and two 80% ethanol washes. The cDNA library pools were run on BioAnalyzer high-sensitivity DNA chips to calculate bp size, followed by a Qubit assay on the Qubit Fluorometer to determine cDNA concentration; ultimately, these values were used to dilute the samples to 2 nm with water. A total of 5 μL of eight pools were multiplexed together for sequencing. Samples were sequenced using a NextSeq 500/550 instrument (Illumina) with the Nextseq 500 High Output v2.5 75 cycle kit (Illumina, catalog no. 20024906).

gDNA extraction from blood

A total of 500 μL of whole blood was used to extract gDNA using the Qiagen blood and tissue kit (Qiagen, catalog no. 69506). Blood was processed according to the manufacturer's guidelines no later than 2 weeks after collection, during which time it was stored at 4°C. After extraction, gDNA was stored at −20°C until sequencing.

gDNA extractions from fresh frozen tissue

Between 25 and 35 mg of fresh frozen tissue, stored at −80°C, was first mechanically dissociated using an RNAse free homogenizer until the tissue was completely dissociated into solution. gDNA was then extracted using the Qiagen AllPrep DNA/RNAmiRNA Universal kit (Qiagen, catalog no. 80224) according to the manufacturer's guidelines. Samples were stored at −20°C until sequencing.

gDNA extractions from tumor tissue from formalin-fixed paraffin-embedded slides

One slide from each tumor was stained with hematoxylin (Richard-Allan Scientific, catalog no. 7231) and eosin (Sigma-Alrich, catalog no. HT110116; H&E). Slides were first deparaffinized in Xylene (Sigma-Alrich, catalog no. 335940) for 10 minutes, then rehydrated for 5 minutes in increments of 100%, 90%, 70% ethanol, then washed for 5 minutes in PBS. The slides were stained in hematoxylin for 2 minutes, rinsed in distilled water, and then dipped in eosin two to three times. Slides were then dehydrated in increments of 70%, 90%, and 100% ethanol for 5 minutes each and then washed in xylene and coverslipped. All H&E slides were evaluated by a collaborating pathologist to determine the location of tumor tissue on the paraffin slide. Slides were then scraped to strategically collect the tumor tissue into 1.5 mL Eppendorf tubes containing deparaffinization solution (Qiagen, catalog no. 1064343). gDNA was then extracted using the Qiagen QIAmp DNA FFPE Tissue kit (Qiagen, catalog no. 56404) according to the manufacturer's guidelines. Samples were stored at −20°C until sequencing.

Quantification of DNA

The Pico green assay (Invitrogen, catalog no. P11496) was used according to the manufacturer's protocol to quantify the concentration of gDNA extracted from FFPE, fresh frozen tissue, and blood.

Whole-exome sequencing

Whole-exome sequencing (WES) of extracted gDNA was conducted on Illumina HiSeq platforms at the Broad Institute to a medium target coverage of 95×. All samples were quantified and assessed for quality using Quant-iT PicoGreen dsDNA Assay (Invitrogen, catalog no. P7589). A total of 50 ng of gDNA then underwent exome enrichment using the Illumina Content Exome (ICE; Illumina) or TWIST Somatic exome v6 (TWIST Biosciences) platforms. Sequencing for ICE was then performed using 76 bp runs with an 8-base index sequencing read on Illumina HiSeq RTA v1.18.64 or later. Sequencing for TWIST used 151 bp runs on Illumina's NovaSeq S4. Standard Coverage ICE exomes were 80% of targets at 20×, Deep Coverage ICE exomes were 85% of targets at 50×, and TWIST Somatic v6 exomes were 85% of targets at 100×.

DNA-based TCRβ sequencing

The Adaptive Biotechnologies immunoSEQ human T-cell receptor beta v4b kit (Adaptive Biotechnologies, catalog no. ISK10050) was used to identify and quantify the frequency of specific T-cell clones in extracted gDNA. gDNA that was extracted from fresh frozen tissue, FFPE, and blood, as explained above, was quantified with the PICO assay to use for TCR-seq. gDNA from fresh frozen tissue was diluted to 83 ng/μL and run in duplicates, gDNA from blood samples was diluted to 44 ng/μL and also run in duplicates, whereas all extracted gDNA from FFPE samples was divided and run in quadruplicates. Ultrapure distilled water (Invitrogen, catalog no. 10977) was used for a negative control and run in parallel with the samples. After the two PCR amplification steps, performed according to the manufacturer's protocol, the multiplexed samples were run on agilent BioAnalyzer high-sensitivity DNA chips (Agilent Technologies, catalog no. 5067-4626) and a KAPA Library qPCR quantification kit (Roche, catalog no. 07960140001). Sequencing was performed on an Illumina NextSeq 500/550 instrument (Illumina) using the Nextseq 500 High Output v2.5 150 cycle kit (Illumina, catalog no. 20024907). DNA-based TCRβ rearrangement repertoires, as well as sample T-cell fractions, were then obtained via the immunoSEQ Analyzer pipeline (Adaptive Biotechnologies).

Variant calling and TMB determination from WES

Single-nucleotide variants (SNV) were called from tumor/normal WES pairs using the GATK4 Mutect 2 workflow (CITE) with reference hg19, run using the mutect2.wdl workflow available from gatk v4.1.3.0 (https://github.com/broadinstitute/gatk; ref. 16). Outputs from the GATK4 Funcotator tool were then used to determine SNVs, and the sum of unique SNVs for a tumor sample was regarded as that sample's TMB.

Reconstruction of TCR from scRNA-seq data

TCRα and TCRβ chains were reconstructed using TraCeR (https://github.com/Teichlab/tracer; ref. 17).

Analysis of TCR data

The Simpson index was computed without replacement for TraCeR samples according to the formula:

The Simpson index was computed with replacement for immunoSEQ samples according to the formula:

where n is the total number cells in a sample, yi is the number of cells in clonotype i.

The Morisita–Horn overlap index between samples X and Y is computed according to the formula:

For Simpson indices SX, SY for samples X and Y, respectively, computed without replacement; nX, nY are total cell counts for samples X, Y, respectively, with xi, yi representing the number of cells in clonotype i.

For analyses using only TraCeR data, clonotypes were defined by groups of cells that were matched by each TCRα or TCRβ information. For analyses using immunoSEQ data, or jointly using immunoSEQ and TraCeR-based clonotyping, only TCRβ information was used (because only TCRβ information is recovered from the immunoSEQ assay).

Alignment and preprocessing of scRNA-seq data

Illumina sequencing outputs were demultiplexed via the bcl2fastq2 program, v2.20.0.422. Generated fastq files were aligned and corrected for PCR bias using the RSEM program (18) using version 8 of the smartseq2 workflow on Terra (app.terra.bio) provided by Cumulus (19). Transcripts were aligned using Bowtie 2 (20) to human genome GRCh38, with gene annotation generated using human Ensembl 93 GTF.

Unsupervised transcriptomic analysis

All analyses were done using the panopticon package. Resultant count matrices were combined and normalized via panopticon.preprocessing.generate_count_matrices. Principal component analysis was performed via the command panopticon.analysis.generate_incremental_pca. Uniform manifold approximation and projection (UMAP) embedding was performed with the command panopticon.analysis.generate_embedding with default parameters. Clustering was performed via the command panopticon.analysis.generate_clustering, which performs multiple rounds of agglomerative clustering with a correlation metric, with the number of clusters selected via a silhouette score. Feature selection (first 10 principal components) was recomputed at each subsequent round, wherein clustering within previously identified clusters was performed. Source code for the panopticon package can be found at https://github.com/scyrusm/panopticon, with additional documentation at https://panopticon-single-cell.readthedocs.io/en/latest/.

Cell quality control

Cells were initially filtered on the basis of a minimum unique gene count of 1,000. Clustering was then performed as described above (see Unsupervised transcriptomic analysis). These clusters were then manually reviewed to assess whether they showed signs of contamination or of being doublets. Results of this manual review, and justification for the inclusion or removal of cells, are given in Supplementary Data S1.

Detecting malignant cells using inferred copy-number variations

Inferred copy-number profiles were performed within cells from a single patient using all FACS gating categories (CD45, CD45+, CD3+) according to the procedure in Tirosh and colleagues (21). These copy-number profiles were then projected onto their first principal component (within a group of cells from a single patient). The quantile of cells’ loading onto this component was denoted the “malignancy score,” with the loading sign-adjusted such that the CD45 cells (per FACS) had a greater such mean score than the grouped CD45+/CD3+ cells. This score was used as a factor when considering cell quality control (see Cell quality control). Code for this procedure is implemented in panopticon.analysis.generate_malignancy_score.

Differential expression and gene expression plots

DIfferential expression between sets was computed with the Mann–Whitney U test, as implemented in panopticon.analysis.cluster_differential_expression, between groups using log2(TP100k+1) gene expression values. Dot plots were computed via the function panopticon.visualization.plot_dotmap. Swarm/violin plots were computed via the function panopticon.visualization.swarmviolin.

Gene expression signatures

Module scores were as originally used in Tirosh and colleagues (21), implemented in the panopticon package as panopticon.analysis.generate_masked_module_score, over the set of cells as described in the text. We used the following MSigDB v7.4 signatures (22, 23):

We additionally used signatures from Jerby-Arnon and colleagues 2018 [CYTOTOXIC T CELL (SPECIFIC MARKERS), EXHAUSTED T CELL (SPECIFIC MARKERS), NAÏVE T CELL (SPECIFIC MARKERS), TREG (SPECIFIC MARKERS), CELL CYCLE: G1–S, CELL CYCLE: G2–M; ref. 24], and Li and colleagues 2019 (cell stress signatures; ref. 25). These signatures are given in Supplementary Data S2.

Statistical analyses

All calculations were performed using python v3.7.4. The following functions and associated P values were used from the scipy package, v1.5.4: Mann–Whitney U (scipy.stats.mannwhitneyu), Kendall-τ (scipy.stats.kendalltau), Fisher exact test (scipy.stats.fisher_exact), Theil-Sen slopes (scipy.stats.theilslopes). All P value tests were two sided unless otherwise noted. The panopticon package v0.2 was used throughout; Cohen d and φ-coefficient effect sizes were computed via the panopticon.utilities.cohensd and panopticon.utilities.phi_coefficient functions.

Cells were regarded as being clonotyped when TraCeR was able to determine the CDR3 sequence of either the TCRα or TCRβ CDR3 (or both). Cells were defined as being clonally expanded wherein a TCRα or TCRβ CDR3 (per TraCeR) was shared with one or more other T cell from the same patient. The remaining clonotyped cells we refer to as “not detectably expanded.”

Fold changes were computed as the difference in the means of log2(TP100k+1) expressions between two groups. Fisher exact tests were computed according to the following contingency table:

graphic

where a, b represent the number of cells of the phenotype in question [CD4/FOXP3, IFN-responsive, natural killer (NK)/NK T (NKT), cycling, effector, exhausted, naïve/memory], which are detectably expanded/not-expanded, respectively or blood-associated/nonassociated, respectively. Table elements c, d represent the sum of cells belonging to all other phenotypes, which are detectably expanded/not-expanded, respectively or blood-associated/nonassociated, respectively. The Fisher exact test P value is computed in the usual way using the formula:

where |$( {_k^n} )$| is the binomial coefficient.

The same contingency table is used to compute the φ-coefficient via the usual formula:

Throughout, kernel density estimate plots were computed via the seaborn.violinplot package using seaborn v0.11.0, with the argument “cut = 0, inner = ‘quartile’” all other parameters default.

Min-max normalization computed as:

Analysis of putative invariant NK T, mucosal-associated invariant T cells

Putative invariant NK T (iNKT) cells and mucosal-associated invariant T (MAIT) cells were classified according to known TCRα V/J allele combinations, or TCRβ V alleles that have been associated with these cells, according to Mori and colleagues (26).

Assessing CDR3 public/private status

A large cohort of TCRβ repertoires was obtained from the immuneCODE dataset (https://clients.adaptivebiotech.com/pub/covid-2020; ref. 27). These repertoires correspond to samples obtained from over 1,486 individuals infected with or exposed to COVID-19. These were used as a large set of nonmelanoma-specific TCRβ repertoires, so as to distinguish private and public (28) TCRβ in our cohort, wherein a “private” TCRβ sequence is unique to the repertoire of a particular individual, whereas “public” sequence is detected in the repertoires from multiple individuals. TCRβ CDR3s detected via TraCeR in our cohort were classified as being private or public based on whether they were detected at any frequency in the immuneCODE dataset.

Data availability

The genes-by-cells matrix and associated metadata for the current study, including all UMAPs for plots in this article, are available via the Broad single-cell portal: https://singlecell.broadinstitute.org/single_cell/study/SCP1493/microenvironmental-correlates-of-immune-checkpoint-inhibitor-response-in-human-melanoma-brain-metastases-revealed-by-t-cell-receptor-and-single-cell-rna-sequencing. Raw data have been deposited in the dbGaP database under accession code phs002416.v2.p1 (https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs002416.v2.p1). TCR-seq data generated through the immunoSEQ assay are available through the immuneACCESS portal (https://clients.adaptivebiotech.com/pub/alvarez-breckenridge-2022-cir). All other data generated in this study are available within the article and its Supplementary Data files or from the corresponding authors upon reasonable request.

Code availability

Panopticon v0.1.1 has been made publicly available (https://github.com/scyrusm/panopticon/). Python code used to create final versions of main and supplementary figures available upon reasonable request.

Characterization of the TME of MBM using scRNA-seq

To dissect the TME of MBM, we performed scRNA-seq on 32 sequentially collected MBM from 27 unique patients (Fig. 1A and B; Supplementary Table S1). These samples were taken as part of standard clinical care, that is, from symptomatic lesions associated with symptoms or mass effect prompting surgical resection. Two patients underwent simultaneous intracranial resections, and longitudinal samples were obtained from 2 patients (one contributing two samples, the other three). At the time of tumor resection, 23 patients had been previously treated with ICI, whereas 9 were ICI naïve. Among the ICI-naïve individuals, one went on to receive ICI and was a responder both intracranially and extracranially. Although the 23 post-ICI patients ultimately experienced intracranial progression resulting in the need for craniotomy, they were defined as having either nonresponse (8) or partial response (15) based on their clinical history of systemic response to ICI. In addition, 5 patients were treated with targeted therapy prior to resection. Key genetic features of patients’ tumors derived from WES and the SNaPshot system for SNP genotype are provided in Supplementary Fig. S2, with additional clinical information in Supplementary Data S3.

Figure 1.

Study design, cohort overview, high-level cell classification, and populations. A, Schematic representation of the study. B, Patient clinical trajectories relative to initial diagnosis of brain metastasis. C, UMAP of single-cell transcriptomes, colored and circled by cell type. D, Heat map of standardized gene expression of key marker genes for each cluster. E, Fraction of post-QC FACS sorted CD45+ cells in each cluster, for each sample. P value was computed via Mann–Whitney U test using patient-averaged T-cell fraction for post-ICI nonresponders versus post-ICI partial responders. Samples derived from the same patient are grouped, with groups indicated by dashed brackets (n = 7, 11 patients in nonresponding, partial-responding groups, respectively).

Figure 1.

Study design, cohort overview, high-level cell classification, and populations. A, Schematic representation of the study. B, Patient clinical trajectories relative to initial diagnosis of brain metastasis. C, UMAP of single-cell transcriptomes, colored and circled by cell type. D, Heat map of standardized gene expression of key marker genes for each cluster. E, Fraction of post-QC FACS sorted CD45+ cells in each cluster, for each sample. P value was computed via Mann–Whitney U test using patient-averaged T-cell fraction for post-ICI nonresponders versus post-ICI partial responders. Samples derived from the same patient are grouped, with groups indicated by dashed brackets (n = 7, 11 patients in nonresponding, partial-responding groups, respectively).

Close modal

Following MBM resection, cells were immediately dissociated and sorted via FACS with gating chosen to enrich for CD45, CD45+, and CD45+CD3+ populations. In total, we collected and sequenced 19,968 cells using the full transcript length Smart-seq2 protocol (29); 14,021 of these cells passed quality control (QC), wherein cells were subsetted based on a minimum unique gene count of 1,000 and then clustered via agglomerative clustering, with removal of clusters showing evidence of contamination or doublets. A median of 2,662, 2,773, and 6,431 unique genes were detected per cell in CD45+CD3+, CD45+, and CD45 populations, respectively. Post-QC clusters were also assessed for evidence of cellular stress (Supplementary Fig. S3; ref. 25).

Our analysis identified a total of 27 nonimmune clusters. As has previously been reported, malignant cells’ RNA expression was highly patient and sample specific (Fig. 1C; ref. 21). In contrast, the remaining clusters demonstrated cell type–specific gene expression (Fig. 1C and D) profiles including glial cells, lymphoid (T/NK cells, B cells, and plasma cells) and myeloid cells (neutrophils, macrophages/microglia). Relative proportions of CD45+ cells across samples are shown in Fig. 1E. Neither malignant cell PD-L1 expression nor TMB were significantly associated with partial versus nonresponse among posttreatment individuals (Supplementary Fig. S4).

Landscape of myeloid populations in the MBM TME

Utilizing our CD45+ sorted cells, we identified a total of 1,266 neutrophil and 653 monocyte-derived cells (including macrophages and microglia) after QC. Highly expressed genes associated with the macrophages/microglia cluster included CD14, CD163, and CSF1R, whereas the gene expression profile of neutrophils featured S100A8, S100A9, and NCF2 (Fig. 1D). Macrophages/microglia and neutrophils expressed significantly different levels of CSF1R, CSF3R, CD68, CD163, NCF1, NCF2, among other genes (Supplementary Fig. S5A and S5B). Each myeloid population included a cross-section of patients from the cohort (Supplementary Fig. S5C and S5D).

Within monocyte-derived cells, we observe four cell clusters (Fig. 2A and B). The first, characterized by elevated relative expression of S100A8, S100A9, S100A12, MNDA, and low relative expression of CD16 (FCGR3A, FCGR3B) were denoted “monocytes.” A second cluster, characterized by elevated expression of hypoxia-associated genes including ALDOC, andHK2 was denoted “hypoxia-associated monocyte-derived.” One cluster expressed SPP1 highly, a gene associated with a particular subset of macrophages previously observed in liver cirrhosis (30) and neuroinflammation in response to chronic viral infection (31). We additionally observed a cluster of microglia defined by expression of TREM2, APOE, C1QA, C1QC. Canonical microglia markers TMEM19 and P2RY12 were additionally upregulated in this cluster (Supplementary Fig. S6). This latter population of cells has been described elsewhere as reactive microglia, with an elevated presence in the brain parenchyma of mouse models of neuronal injury and aging (32).

Figure 2.

Multiple myeloid phenotypes and their association with malignant phenotypes observed intracranially. A, UMAP of post-QC monocyte-derived cells, including macrophages and microglia. B, Heat map of standardized gene expression of marker genes for clusters of monocyte-derived cells. C, UMAP of post-QC neutrophils (excluding neutrophil-committed progenitors/eosinophils). D, Heat map of standardized gene expression of marker genes for clusters of post-QC neutrophils (excluding neutrophil-committed progenitors/eosinophils). E, Dot plot of key genes associated with N1 and N2 phenotypes in the calprotectin-high, IFN-responsive, and IL8-high neutrophil populations. F, Swarm/violin plots of fraction of nonneutrophil-committed progenitors (of CD45+ cells) across patients; Mann–Whitney U P value for comparison of posttreatment partial versus nonresponders indicated (n = 7, 11 patients in nonresponding, partial-responding groups, respectively), quartiles indicated by dashed lines. G, H&E stain of tumor section from MEL022 with 100 μmol/L scale bar indicated, showing high levels of blood product and neutrophil infiltration. H, Swarm/violin plot of neutrophil fraction (of post-QC CD45+ cells, n = 5, 22 for samples with and without evidence of necrosis, respectively), IL8 fraction (only samples with 10 or more neutrophils considered) in samples with and without evidence of necrosis (n = 12, 5 for samples with and without necrosis, respectively), quartiles indicated by dashed lines. I, Swarm/violin plot of the hallmark angiogenesis module score for the IL8-expressing, calprotectin-high, and IFN-responsive neutrophils, with Mann–Whitney P value of IL8 versus other neutrophils indicated (n = 662, 269, 256 for calprotectin high, IL8 high, and IFN-responsive, respectively), quartiles indicated by dashed lines. J, Fraction of IL8-high neutrophils (of all mature) versus EMT module score calculated in malignant cells across patients; Kendall-τ correlation and associated P values are indicated. Kendall-τ correlations in Fig. 2J were computed only over patients with 10 or more detected neutrophils (15 patients total). Theil-sen line of best fit indicated by dotted line. K, Dot plot of genes associated with NETosis across calprotectin-high, IFN-responsive, and IL8-high neutrophils.

Figure 2.

Multiple myeloid phenotypes and their association with malignant phenotypes observed intracranially. A, UMAP of post-QC monocyte-derived cells, including macrophages and microglia. B, Heat map of standardized gene expression of marker genes for clusters of monocyte-derived cells. C, UMAP of post-QC neutrophils (excluding neutrophil-committed progenitors/eosinophils). D, Heat map of standardized gene expression of marker genes for clusters of post-QC neutrophils (excluding neutrophil-committed progenitors/eosinophils). E, Dot plot of key genes associated with N1 and N2 phenotypes in the calprotectin-high, IFN-responsive, and IL8-high neutrophil populations. F, Swarm/violin plots of fraction of nonneutrophil-committed progenitors (of CD45+ cells) across patients; Mann–Whitney U P value for comparison of posttreatment partial versus nonresponders indicated (n = 7, 11 patients in nonresponding, partial-responding groups, respectively), quartiles indicated by dashed lines. G, H&E stain of tumor section from MEL022 with 100 μmol/L scale bar indicated, showing high levels of blood product and neutrophil infiltration. H, Swarm/violin plot of neutrophil fraction (of post-QC CD45+ cells, n = 5, 22 for samples with and without evidence of necrosis, respectively), IL8 fraction (only samples with 10 or more neutrophils considered) in samples with and without evidence of necrosis (n = 12, 5 for samples with and without necrosis, respectively), quartiles indicated by dashed lines. I, Swarm/violin plot of the hallmark angiogenesis module score for the IL8-expressing, calprotectin-high, and IFN-responsive neutrophils, with Mann–Whitney P value of IL8 versus other neutrophils indicated (n = 662, 269, 256 for calprotectin high, IL8 high, and IFN-responsive, respectively), quartiles indicated by dashed lines. J, Fraction of IL8-high neutrophils (of all mature) versus EMT module score calculated in malignant cells across patients; Kendall-τ correlation and associated P values are indicated. Kendall-τ correlations in Fig. 2J were computed only over patients with 10 or more detected neutrophils (15 patients total). Theil-sen line of best fit indicated by dotted line. K, Dot plot of genes associated with NETosis across calprotectin-high, IFN-responsive, and IL8-high neutrophils.

Close modal

A total of 1,266 neutrophils were identified in 23 of 27 patients’ cells and were associated with one of four subclusters. One of these subclusters, representing 63 cells from 10 patients (Supplementary Fig. S7A), clustered most distantly from the other three neutrophil groups (Supplementary Fig. S7B). Separate clustering of these cells revealed four clusters (Supplementary Fig. S7C), with gene expression characteristic of eosinophils (CCR3), primary/azurophilic (MPO, AZU1, CTSG, DEFA1B, DEFA3, DEFA4), secondary/specific (LTF, CAMP), and tertiary granules (MMP9), respectively (Supplementary Fig. S7D and S7E); the latter three have recently been termed pro-neutrophils, lineage-committed precursors, and immature neutrophils, respectively (33); these three will be hereafter referred to as “neutrophil-committed progenitors.” These eosinophils/neutrophil-committed progenitors were removed prior to reclustering the remaining, mature neutrophils. Among the three remaining neutrophil clusters, one included a “calprotectin-high” group characterized by high expression of S100A8 and S100A9, jointly coding for the calprotectin heterodimer, and representing 52% of all identified neutrophils (Fig. 2C and D). A second cluster was termed “IFN-responsive” neutrophils, which were characterized by high expression of genes associated with IFNγ response, including IFI6, IFIT2, ISG15, and TAP1 (Fig. 2C and D). The final cluster was termed the “IL8-high” neutrophils and was characterized by elevated expression of IL8/CXCL8 and VEGFA (Fig. 2C and D). Calprotectin-high, IFN-responsive, and IL8-high neutrophils were observed in 23, 15, and 15 patients, respectively (Supplementary Fig. S5D).

This neutrophil polarization is partially concordant with what has been seen in murine models and human in vitro studies (34). Previous work on heterogeneity of tumor-associated neutrophils (TAN) has subdivided neutrophils along an “N1” and “N2” polarization axis, corresponding to antitumor and protumor phenotypes, respectively (35). N1 neutrophils were characterized by higher levels of ICAM-1/CD54, FasR/CD95, and TNFα, whereas N2 neutrophils expressed higher levels of CCL5, CCL2, VEGFA, arginase, and IL8/CXCL8 (34). IL8-high neutrophils derived from our MBM cohort corresponded most closely to N2 neutrophils; however, they showed reduced expression of N2 markers MMP9 and ARG1, as well as elevated expression of the N1 markers CCL3 and ICAM1 (Fig. 2E). Therefore, although our data recapitulate the heterogeneous nature of TANs, further studies will be necessary to delineate the precise markers of neutrophil states and their associated plasticity in the human tumoral context.

Across multiple histologies, neutrophil gene signatures are associated with poor prognosis (35), and the neutrophil-to-lymphocyte ratio has been considered as a biomarker of poor survival outcomes in multiple therapies, including ICI (36). This is consistent with our data, wherein the neutrophil fraction (of CD45+ cells) was significantly higher in posttreatment nonresponders when compared with posttreatment partial responders (Fig. 2F); this was further demonstrated via H&E staining from a nonresponding patient (MEL022; Fig. 2G; Supplementary Fig. S8). However, calprotectin-high, IL8-high, and IFN-responsive neutrophils did not display a significantly larger representation in post-ICI nonresponders compared with partial responders (Supplementary Fig. S9). CXCR2 and IL8/CXCL8 have both been linked previously to ICI resistance as well as to the neutrophil N2 phenotype (37); however, as CXCR2 (one of the two receptors for IL8) was more highly expressed in calprotectin-high neutrophils than in IL8-high neutrophils (Fig. 2E), it may be that the protumor roles of neutrophils are not isolated to a single phenotype, but rather to a spectrum whose protumor effects are context dependent.

Multiple mechanisms for the negative prognostic association of neutrophils have been proposed, including an association with tumor necrosis (itself linked to poor prognosis in multiple histologies), as well as the induction of angiogenesis, EMT, and neutrophil extracellular traps (NETosis; ref. 38). Accordingly, we see a significant association (P = 0.0063) between samples’ neutrophil fraction (of CD45+ cells) and evidence of tumor necrosis on pathology reports (Fig. 2H). Our data are also consistent with the hypothesis that angiogenesis in the TME is promoted in part by IL8-producing neutrophils, based on the significantly higher expression of genes in the hallmark angiogenesis gene module compared with the two other primary neutrophil phenotypes (Fig. 2I). We additionally see a significant correlation between patients’ IL8-high neutrophil fraction and the EMT module score of patients’ malignant cells (Fig. 2J). Finally, genes associated with NETosis (39) were most highly expressed in calprotectin-high neutrophils but were also observed, albeit in decreasing amounts, among the IFN-responsive, and IL8-high neutrophils (Fig. 2K). These collective findings suggest that neutrophils’ protumor function is multifaceted involving a spectrum of phenotypic states, and that a given neutrophil may play certain, though not all, protumor roles at a given time.

Intracranial T-cell phenotypic diversity is consistent with that seen extracranially

We assessed the diversity of intracranial T-cell phenotypes in the MBM TME and found that although our data recapitulated T-cell phenotypic diversity observed extracranially, overall T-cell infiltration did not reach statistical significance as a prognostic biomarker for ICI response in our cohort (Fig. 1E). In extracranial melanoma, both T-cell infiltration and specific T-cell phenotypes have been implicated as prognostic factors for response to ICI (40–44). We therefore performed a focused analysis of only the FACS-selected CD45+CD3+ cells within our scRNA-seq cohort. Following a more stringent QC screening via a cell complexity threshold of 2,000, a total of 2,974 cells were ultimately included for further analysis. Following unsupervised analysis, post-QC cells clustered into the following seven phenotypic populations, which we refer to as IFN-responsive, cycling, memory/naïve, effector, exhausted, CD4/FOXP3, and NK/NKT cells (Fig. 3A and B). Cells within the “IFN-responsive” cluster were predominantly CD8+ T cells with upregulation of IFN-response pathways (IFIT1, IFIT2, IFIT3, ISG15). The “cycling” population consisted primarily of CD8+ T cells with expression of canonical cycling genes (MKI67, ZWINT, TOP2A). Cells within the “memory” cluster included both CD8+ T and CD4+ T cells enriched for IL7R, and CCR7. “Effector” cells were marked by cytotoxic genes including GNLY, GZMH, PRF1, and KLRG1, and lower levels of exhaustion-associated genes than the “exhausted” cluster. “Exhausted” cells, meanwhile, expressed high levels HAVCR2, PDCD1, CTLA4, and TIGIT. These designations are consistent with those based on marker signatures applied in a previous scRNA study of extracranial melanoma using the Smart-Seq2 protocol (ref. 24; Fig. 3C; Supplementary Fig. S10).

Figure 3.

T-cell phenotypic, clonotype heterogeneity, and corresponding association with response to ICI. A, UMAP of post-QC CD3+ (isolated via FACS) T cells. B, Heat map standardized expression of top 10 marker genes for each cluster. C, Heat map of effect sizes (left) and negative base-10 logarithm of P values (right) for module scores of selected signatures from Jerby-Arnon and colleagues (24), with NK/NKT cells removed. Effect sizes and P values used were the common language effect size and Mann–Whitney U P value, respectively. These are computed for each post-QC CD3+ T-cell cluster (NK/NKT cells removed) for the in-cluster and not-in-cluster groups, for the module score of the gene signature in question. Signatures with highest effect size for each cluster and P < 10−10 highlighted in yellow (both cycling signatures highlighted for “cycling” cluster). N = 739, 701, 644, 349, 310, and 171 for effector, exhausted, naïve/memory, CD4FOXP3, IFN-responsive, and cycling cells, respectively. D, Distribution of T-cell fraction (per the immunoSEQ assay) and phenotype fraction across patients; pretreatment patients noted in bold, Mann–Whitney U P value comparing distribution of posttreatment partial versus nonresponders indicated (with BM immunoSEQ data: n = 10 posttreatment partial responder, n = 5 posttreatment nonresponder patients; with blood immunoSEQ n = 10 posttreatment partial responder, n = 6 posttreatment nonresponder patients; with post-QC T cells, n = 11 posttreatment partial responder, n = 5 posttreatment nonresponder patients), quartiles indicated by dashed lines. E, Stacked bar plot for size of clonotypes identified via TraCeR. Only samples with 10 or more cells are shown. F, Swarm/violin plot of Simpson indices according to TraCeR and immunoSEQ across patients; pretreatment patients noted in bold, one-sided Mann–Whitney U P value comparing distribution of posttreatment partial versus nonresponders indicated (with BM immunoSEQ data: n = 10 posttreatment partial responder, n = 4 posttreatment nonresponder patients; with extracranial immunoSEQ data: n = 4 posttreatment partial responder, n = 3 posttreatment nonresponder patients; with blood immunoSEQ n = 10 posttreatment partial responder, n = 5 posttreatment nonresponder patients; with post-QC T cell TraCeR, n = 11 posttreatment partial responder, n = 5 posttreatment nonresponder patients), quartiles indicated by dashed lines. Samples are combined when multiple samples from a single patient were present.

Figure 3.

T-cell phenotypic, clonotype heterogeneity, and corresponding association with response to ICI. A, UMAP of post-QC CD3+ (isolated via FACS) T cells. B, Heat map standardized expression of top 10 marker genes for each cluster. C, Heat map of effect sizes (left) and negative base-10 logarithm of P values (right) for module scores of selected signatures from Jerby-Arnon and colleagues (24), with NK/NKT cells removed. Effect sizes and P values used were the common language effect size and Mann–Whitney U P value, respectively. These are computed for each post-QC CD3+ T-cell cluster (NK/NKT cells removed) for the in-cluster and not-in-cluster groups, for the module score of the gene signature in question. Signatures with highest effect size for each cluster and P < 10−10 highlighted in yellow (both cycling signatures highlighted for “cycling” cluster). N = 739, 701, 644, 349, 310, and 171 for effector, exhausted, naïve/memory, CD4FOXP3, IFN-responsive, and cycling cells, respectively. D, Distribution of T-cell fraction (per the immunoSEQ assay) and phenotype fraction across patients; pretreatment patients noted in bold, Mann–Whitney U P value comparing distribution of posttreatment partial versus nonresponders indicated (with BM immunoSEQ data: n = 10 posttreatment partial responder, n = 5 posttreatment nonresponder patients; with blood immunoSEQ n = 10 posttreatment partial responder, n = 6 posttreatment nonresponder patients; with post-QC T cells, n = 11 posttreatment partial responder, n = 5 posttreatment nonresponder patients), quartiles indicated by dashed lines. E, Stacked bar plot for size of clonotypes identified via TraCeR. Only samples with 10 or more cells are shown. F, Swarm/violin plot of Simpson indices according to TraCeR and immunoSEQ across patients; pretreatment patients noted in bold, one-sided Mann–Whitney U P value comparing distribution of posttreatment partial versus nonresponders indicated (with BM immunoSEQ data: n = 10 posttreatment partial responder, n = 4 posttreatment nonresponder patients; with extracranial immunoSEQ data: n = 4 posttreatment partial responder, n = 3 posttreatment nonresponder patients; with blood immunoSEQ n = 10 posttreatment partial responder, n = 5 posttreatment nonresponder patients; with post-QC T cell TraCeR, n = 11 posttreatment partial responder, n = 5 posttreatment nonresponder patients), quartiles indicated by dashed lines. Samples are combined when multiple samples from a single patient were present.

Close modal

Among the post-ICI samples, only the CD4/FOXP3 fraction was significantly associated with partial over nonresponse (P = 0.0087; Fig. 3D). The fraction of exhausted cells, found in Thommen and colleagues (45) to be associated with better prognosis, was associated with partial versus nonresponse among the post-ICI samples, albeit insignificantly (P = 0.32; Fig. 3D). Neither intracranial nor blood T-cell fraction (as measured by the immunoSEQ assay) was associated with partial versus nonresponse posttherapy. We did observe that the “effector” T-cell fraction was highest in the pretreatment responder sample in our cohort (MEL-027; Fig. 3D), which is consistent with previous reports in the extracranial context (43, 44).

Peripheral T-cell clonal expansion associated with response to ICI

Consistent with previous work (12, 46, 47), we observed that T-cell clonal expansion was associated with response to ICI. From our cohort's full-length transcript scRNA of freshly resected metastases, we performed single-cell TCR clonotyping via TraCeR (17) to quantify T-cell clonal expansion within the MBM TME in addition to using the gDNA-based immunoSEQ assay. Using TraCeR, we successfully clonotyped 2,110 pre-QC cells (with 1,371 clonotyped cells in the CD45+CD3+ post-QC group) from 31 unique samples (samples with greater than 10 clonotyped cells shown in Fig. 3E), with clonotypes defined as groups of T cells connected by either a common TCRα or TCRβ, consistent with previous work (full TraCeR output provided in Supplementary Data S4; ref. 48). We additionally assessed the presence of MAIT cells and iNKT cells (Supplementary Fig. S11), finding that putative MAIT cells were not significantly expanded relative to non-MAIT cells. We investigated the relationship between clonal expansion in both peripherally circulating and tumor-infiltrating T cells using the Simpson index (a diversity-based measure of clonal expansion) across each compartment (49), which was computed using both TraCeR- and immunoSEQ-derived TCR repertoires. Using TraCeR results from within MBM, we found that patients who had partial response to ICI with intracranial progression had a nonsignificant increase in their degree of clonal expansion compared with patients who were entirely nonresponsive to ICI (P = 0.241 via TraCeR, P = 0.868 via immunoSEQ; Fig. 3F). In addition, although it did not reach significance, sampling T cells from the blood demonstrated that clonal expansion tracked with partial response to ICI compared with the blood of patients who were nonresponsive to ICI (P = 0.063 via immunoSEQ; Fig. 3F). These findings are consistent with previous reports suggesting that T cells from the peripheral circulation may be sampled as a biomarker of systemic response to ICI (50).

Intracranial clonally expanded T cells enriched for cells with exhausted phenotype

Given the association between clonal expansion and response to ICI, we assessed phenotypic correlates with clonal expansion. Of the 2,110 successfully clonotyped cells, 684 cells shared a TCRα or TCRβ chain with one or more other T cells from the same patient and were therefore regarded as being “detectably expanded” (448 of the 1,371 post-QC CD3+ T cells were detectably expanded). The remaining clonotyped cells we refer to as “not detectably expanded” (Fig. 4A, these analyses with alternate definitions of clonotype and clonal expansion shown in Supplementary Fig. S12). Differential expression across samples of cells that were “detectably expanded” revealed enrichment of exhaustion- and effector-associated genes, including HAVCR2, TIGIT, PRF1, NKG7, and GZMB, in the detectably expanded cells, whereas those cells that were clonotyped but not detectably expanded expressed higher levels of memory/naïve-associated genes, including CCR7 and IL7R (Fig. 4B, full output in Supplementary Data S5). The majority of cells from two of the seven T-cell clusters—“cycling” and “exhausted”—were detectably expanded (P = 8.70e-7, 4.51e-59, respectively, via Fisher exact test; Fig. 4C). In contrast, “memory” cells were most significantly enriched in the nonexpanded set (P = 7.83e-36 via Fisher exact test, φ-coefficient = −0.3268; Fig. 4C). Thus, although clonal expansion is closely linked to response to ICI, we observed that clonally expanded cells at the site of the lesion were more likely to be exhausted.

Figure 4.

Association between T-cell CDR3 and phenotype. A, UMAP (from Fig. 3A) indicating phenotypic distribution of detectably expanded T cells (blue), not detectably expanded T cells (red), and nonclonotyped cells (gray); clone size of T cells belonging to detectably expanded clone is proportional in size to marker area. B, Volcano plot of differentially expressed genes between detectably and not detectably expanded cells; key genes annotated. P values calculated via Mann–Whitney U, n = 421, 894 detectably and not detectably expanded cells, respectively. C, Heat map of fraction of detectably and not detectably expanded cells in each cluster. Colors indicate fraction in each cluster, on-block annotations indicate absolute number of cells in each stratum. P values (Fisher exact test on in-cluster and out-of-cluster counts for detectably/not detectably expanded cells) and effect sizes (φ-coefficient) are annotated. D, UMAP (from Fig. 3A) indicating phenotypic distribution of blood-unassociated T cells (blue), blood-associated T cells (red), and nonclonotyped cells (gray). E, Volcano plot of differentially expressed genes between blood-unassociated and blood-associated T cells; key genes annotated. P values calculated via Mann–Whitney U, n = 150, 163 blood-associated and blood-unassociated cells, respectively. F, Heat map of fraction of blood-associated and blood-unassociated cells in each cluster. Colors indicate fraction in each cluster, on-block annotations indicate absolute number of cells in each stratum. P values (Fisher exact test on in-cluster and out-of-cluster counts for blood-associated/unassociated cells) and effect sizes (φ-coefficient) are annotated. G, Association between patient-averaged post-QC phenotype fraction and MOI. Kendall-τ correlation and P values are indicated (n = 17). Theil-sen line of best fit indicated by dotted line.

Figure 4.

Association between T-cell CDR3 and phenotype. A, UMAP (from Fig. 3A) indicating phenotypic distribution of detectably expanded T cells (blue), not detectably expanded T cells (red), and nonclonotyped cells (gray); clone size of T cells belonging to detectably expanded clone is proportional in size to marker area. B, Volcano plot of differentially expressed genes between detectably and not detectably expanded cells; key genes annotated. P values calculated via Mann–Whitney U, n = 421, 894 detectably and not detectably expanded cells, respectively. C, Heat map of fraction of detectably and not detectably expanded cells in each cluster. Colors indicate fraction in each cluster, on-block annotations indicate absolute number of cells in each stratum. P values (Fisher exact test on in-cluster and out-of-cluster counts for detectably/not detectably expanded cells) and effect sizes (φ-coefficient) are annotated. D, UMAP (from Fig. 3A) indicating phenotypic distribution of blood-unassociated T cells (blue), blood-associated T cells (red), and nonclonotyped cells (gray). E, Volcano plot of differentially expressed genes between blood-unassociated and blood-associated T cells; key genes annotated. P values calculated via Mann–Whitney U, n = 150, 163 blood-associated and blood-unassociated cells, respectively. F, Heat map of fraction of blood-associated and blood-unassociated cells in each cluster. Colors indicate fraction in each cluster, on-block annotations indicate absolute number of cells in each stratum. P values (Fisher exact test on in-cluster and out-of-cluster counts for blood-associated/unassociated cells) and effect sizes (φ-coefficient) are annotated. G, Association between patient-averaged post-QC phenotype fraction and MOI. Kendall-τ correlation and P values are indicated (n = 17). Theil-sen line of best fit indicated by dotted line.

Close modal

Intracranial T cells belonging to clonotypes detected in blood show reduced exhaustion

To investigate the relationship between clonally expanded T cells within the blood and TME, we performed TCR clonotyping of the TCRβ chain via the immunoSEQ assay of gDNA from peripheral blood. T-cell clones from the MBM TME were subsequently matched (based on shared CDR3) against clones found in the blood. MBM T cell CDR3s were referred to as “blood-associated” T-cell clones when they had a TraCeR-detected TCRβ CDR3 that was also detected in the blood via the immunoSEQ assay. Those MBM T cells with a TraCeR-detected TCRβ CDR3 that was not detected in the blood via immunoSEQ were denoted “blood-unassociated”. We observed a significant divergence in the distribution of “blood-associated” and “blood-unassociated” T cells in our CD45+CD3+ UMAP (Fig. 4D). Differential expression of “blood-associated” and “blood-unassociated” T cells revealed upregulation of progenitor/effector-like genes (e.g., TCF7, GNLY, and KLRG1) in the former group, and exhaustion-related genes (CTLA4, TIGIT, HAVCR2) in the latter (Fig. 4E, full output in Supplementary Data S5). Accordingly, we discovered a significant association between the “effector” population and blood association (P = 2.34e-12, Fisher exact test, φ-coefficient = 0.3833) and a corresponding negative association with the “exhausted” population (P = 1.35e-10, Fisher exact test, φ-coefficient = −0.3208; Fig. 4F). Therefore, the Simpson index of the clonal repertoire measured in the periphery may be a more accurate measure of the reservoir of clonally expanding, nonterminally exhausted T cells at the time of sampling. These findings are consistent with findings in the context of extracranial melanoma (51, 52). We validated this finding orthogonally on a patient-level basis by comparing phenotypic fraction with the Morisita overlap index (MOI), a measure of TCR repertoire similarity between paired samples (53). We observed that the fraction of CD45+CD3+ “effector” cells correlated significantly with the MOI between intracranial and blood-derived immunoSEQ TCR repertoires (P = 0.003, Kendall-τ correlation P values; Fig. 4G), suggesting that the MOI of blood and lesion-derived TCR repertoires can be used as an estimate of the effector-like T-cell fraction within the lesion.

T-cell clones with private CDR3s are more likely to be exhausted

To explore the question of TCR specificity for tumor in our cohort of MBM, we utilized a dataset of more than 500,000,000 clonotyped T cells across 1,486 samples from individuals infected with or exposed to COVID-19 as a reference of presumably nontumor-specific clones and compared these against our 2,110 clonotyped MBM-associated T cells (27). A total of 68% of clonotyped post-QC CD45+CD3+ T cells (213/313 cells) from our MBM cohort had CDR3s that were detected in the COVID-19 dataset, suggesting that a substantial fraction of MBM T cells may not be tumor specific. Of those cells with CDR3s detected in the COVID-19 dataset, 10% (21/213) had CDR3s with known specificity according to VDJdb (ref. 54; 13 cells with CDR3 specific to the influenza M1 protein, five to the IE1 cytomegalovirus protein, two to the Epstein-Barr virus, with one cell's CDR3 specific to the human protein PMEL; Supplementary Fig. S13), with none of the cells with CDR3 not detected in the COVID-19 dataset having CDR3s detected in the VDJdb. We subsequently categorized the MBM T-cell CDR3s as public, indicating that they were present in both our MBM cohort and the COVID-19 dataset, or private, indicating that the T-cell clones were solely identified in our cohort of MBM (Fig. 5A). As observed elsewhere (28), these private CDR3s were significantly longer (two-sided Mann–Whitney U P value = 3.26e-11) than the public CDR3s (Fig. 5B). Upon further investigation of the phenotypic distribution among private and public clones, exhausted T-cell clones were significantly associated with private CDR3 (P = 0.00198) whereas an effector phenotype was primarily associated with public clones (P = 0.0324; Fig. 5C, selected gene associations with CDR3 privacy in Supplementary Fig. S14). These findings suggest that public T-cell clones lack tumor specificity and maintain an effector status without evidence of exhaustion. In contrast, private T-cell clones appear tumor specific and may be driven to an exhausted phenotype by persistent antigenic stimulation.

Figure 5.

Association of clonotype privacy with phenotype and ICI response. A, UMAP (from Fig. 3A) indicating phenotypic distribution of T cells with private CDR3 (blue), public CDR3 (red), and nonclonotyped cells (gray). B, Distribution of CDR3 lengths in public and private clonotypes (n = 97, n = 206, respectively). Distribution is normalized kernel density estimate; quartiles indicated by dashed lines. Mann–Whitney P value and Cohen d effect size annotated. C, Heat map of fraction of cells with private and public CDR3 in each cluster. Colors indicate fraction in each cluster, on-block annotations indicate absolute number of cells in each stratum. P values (Fisher exact test on in-cluster and out-of-cluster counts for cells with private/public CDR3s) and effect sizes (φ-coefficient) are annotated. D, Swarm/violin plots of fraction of T cells with private CDR3 across patients; pretreatment patients noted in bold, Mann–Whitney U P value comparing distribution of posttreatment partial versus nonresponders indicated (n = 9, 4, respectively). Dashed lines indicate quartiles. E, T1 post-contrast MRI of the brain and lung from an ICI-responsive individual (MEL027). Representative images of the brain and lung are included at the time of craniotomy for the large, symptomatic cerebellar metastasis (pre-ICI) and after 6 months of ICI administration (11 cycles of pembrolizumab) indicating resolution of intracranial and extracranial disease. Yellow arrows mark regions of enhancement indicative of pretreatment disease. F, H&E staining of tumor section from MEL027, indicating large population of tumor-infiltrating lymphocytes. G, Multiple prognostic metrics across patients (see Materials and Methods); distribution for each metric, with top, bottom quartiles indicated by dotted lines, median by dashed line. Black line connects values for individual responding patient MEL027.

Figure 5.

Association of clonotype privacy with phenotype and ICI response. A, UMAP (from Fig. 3A) indicating phenotypic distribution of T cells with private CDR3 (blue), public CDR3 (red), and nonclonotyped cells (gray). B, Distribution of CDR3 lengths in public and private clonotypes (n = 97, n = 206, respectively). Distribution is normalized kernel density estimate; quartiles indicated by dashed lines. Mann–Whitney P value and Cohen d effect size annotated. C, Heat map of fraction of cells with private and public CDR3 in each cluster. Colors indicate fraction in each cluster, on-block annotations indicate absolute number of cells in each stratum. P values (Fisher exact test on in-cluster and out-of-cluster counts for cells with private/public CDR3s) and effect sizes (φ-coefficient) are annotated. D, Swarm/violin plots of fraction of T cells with private CDR3 across patients; pretreatment patients noted in bold, Mann–Whitney U P value comparing distribution of posttreatment partial versus nonresponders indicated (n = 9, 4, respectively). Dashed lines indicate quartiles. E, T1 post-contrast MRI of the brain and lung from an ICI-responsive individual (MEL027). Representative images of the brain and lung are included at the time of craniotomy for the large, symptomatic cerebellar metastasis (pre-ICI) and after 6 months of ICI administration (11 cycles of pembrolizumab) indicating resolution of intracranial and extracranial disease. Yellow arrows mark regions of enhancement indicative of pretreatment disease. F, H&E staining of tumor section from MEL027, indicating large population of tumor-infiltrating lymphocytes. G, Multiple prognostic metrics across patients (see Materials and Methods); distribution for each metric, with top, bottom quartiles indicated by dotted lines, median by dashed line. Black line connects values for individual responding patient MEL027.

Close modal

To examine the clinical relevance of this finding, we stratified patients by their clinical response to ICI and quantified the fraction of private T-cell clones per patient. Within our cohort, we observed an association between a patient's percentage of private CDR3 clones and overall response (P = 0.054, two-sided Mann–Whitney U test; Fig. 5D).

Although all post-ICI patients ultimately developed intracranial progression on ICI, prompting surgical resection, MEL027 was a treatment-naïve patient who went on to receive ICI and was the single patient within the cohort who was a true intracranial responder (Fig. 5E). With this unique clinical trajectory, we investigated the features of T-cell infiltration and T-cell clonal expansion within the blood and TME. The patient presented with multiple cerebellar metastases in addition to extracranial disease in the lung. Following resection of the dominant cerebellar lesion, pembrolizumab was initiated and resulted in both cranial and extracranial response (Fig. 5E). Histologic examination of the treatment-naïve cerebellar lesion demonstrated abundant tumor-infiltrating lymphocytes (Fig. 5F). This finding was corroborated with T-cell fraction quantification from MBM and blood using the immunoSEQ platform (Fig. 5G; Supplementary Fig. S15). Across both sites, T-cell fraction was above the median for the cohort (Fig. 5G; Supplementary S15). Similarly, among our MBM cohort, MEL027 demonstrated the greatest degree of clonal expansion within the MBM and blood while also harboring the greatest proportion of effector T cells within the MBM (Fig. 5G; Supplementary S15). Finally, we explored the extent of T-cell clonal overlap with clones from the COVID-19 dataset and observed that the fraction of private CDR3 clones for MEL027 in the blood was highest within the cohort and similarly elevated within the MBM (Fig. 5G; Supplementary S15). These findings are consistent with previously published reports (55) that pretreatment T-cell infiltration is predictive of response to ICI while also integrating T-cell phenotype, clonality, and tumor exclusivity.

With the documented success of ICI for extracranial disease and emerging data demonstrating intracranial response rates across multiple histologies, increasing attention has been placed on the tumor intrinsic and microenvironmental factors that portend a favorable response to ICI (6). Here, we utilized scRNA-seq of MBMs combined with TCR-seq of intracranial and extracranial samples to both characterize biomarkers of response to ICI and to elucidate cross-compartmental correlates of malignant and immune phenotypes. We observed relationships between myeloid and malignant phenotypes, as well as relationships between T-cell phenotype, TCR distribution, and TCR diversity. These findings have implications for both the therapy and monitoring of intracranial disease.

Melanoma metastases have been significantly associated with increased levels of tumor-infiltrating lymphocytes. Beyond T-cell infiltration, however, increasing evidence suggests that the spatial distribution within the tumor, T-cell phenotypic plasticity, and T-cell clonality all impact antitumor immune-mediated responses (15, 46, 47, 56). Although scRNA-seq has been used to characterize CD8+ T-cell states in extracranial melanoma metastases, our cohort of MBM provides a unique opportunity to investigate the degree of phenotypic overlap within the MBM TME. We observed T-cell transcriptional signatures that are concordant with those seen in extracranial melanoma (21, 24, 43). Although the fraction of effector-like T cells was elevated in the one pre-ICI responder in our cohort, neither this nor other CD8+ T-cell phenotypes were significantly associated with partial versus nonresponse in the posttreatment samples.

T-cell clonality is a proxy for antigen-driven T-cell expansion and has been associated with clinical benefit across multiple tumor types and therapies (12). Within our cohort, clonal expansion in post-treatment blood, but not in intracranial lesions, was significantly associated with partial response over nonresponse. This may be explained by the strong association between T-cell clone size and exhaustion observed intracranially, wherein intracranial clonally expanded T cells have lost effector capacity due to persistent antigen stimulation, and therefore have reduced prognostic significance. It is also possible that the stronger association of blood T-cell clonality with response is reflective of the more robust extracranial versus intracranial T-cell response observed in a group of patients requiring resection of MBM.

Rather than representing a unique, immune-isolated environment, our findings further support the idea that the CNS TME is immune specialized rather than immune privileged. We observed unique phenotypic differences between detectably expanded and nonexpanded T-cell clones within the intracranial tumors. Although exhausted and cycling T-cell clones were predominantly expanded, CD4+FOXP3+ T cells, memory CD8+ T cells, and IFN-responsive CD8+ T cells were associated with nonclonally expanded T cells. With the finding of phenotypic divergence between blood-overlapping and tumor-exclusive T-cell clones in MBM, this finding suggests a model whereby the CNS maintains features of the periphery with active cross-talk between the peripheral and intratumoral immune compartments, with the blood acting as a reservoir of fresh, pre–antigen-stimulated T cells for MBMs (57). Future investigation is needed to explore the process of intracranial T-cell trafficking and how this process is further modulated by exposure to ICI (9).

In addition to identifying individual T-cell states, deciphering the nature of T-cell tumor reactivity is an increasingly pressing challenge. The MBM TME is populated both by T-cell clones that are reactive to tumor antigens, as well as by bystander clones that are not cancer specific. Among our cohort of MBM, we observed a phenotypic difference between intracranial T cells with CDR3 regions that were patient specific, and therefore assumed to be more likely specific to that patient's tumor. Consistent with this finding, the greatest positive and negative associations with CDR3 privacy were observed in exhausted and effector cells, respectively. The degree of CDR3 privacy was also associated with clinical benefit to ICI within the MBM context. We note that the association between CDR3 tumor specificity and ICI response was not observed by Caushi and colleagues in a clinical trial of neoadjuvant ICI for non–small cell lung cancer (NSCLC), wherein TCR-neoantigen specificity was directly assessed via the MANA-fest assay (58). This difference may be attributable to these cohorts’ different definition of response, wherein responders in Caushi and colleagues frequently had tumor clearance prior to resection, whereas all patients in our study had symptomatic tumor prompting resection. These cohorts also differed in pretreatment disease stage, as many responding patients in Caushi and colleagues had earlier stage disease. Additional differences may be driven by primary histology (melanoma vs. NSCLC) or the intracranial versus extracranial microenvironment. In summary, disease stage, primary histology, sample timing, and the extent of tumor clearance should all be considered when interpreting the prognostic implications of both T-cell clonotype and phenotype in the setting of ICI.

Apart from exploring T-cell features within the MBM TME and the associated relationship with the periphery, our cohort provided a unique opportunity to explore the myeloid population in the MBM context. Dynamic features of myeloid phenotypic heterogeneity have been associated with both tumor-supportive and antitumor properties (59); however, their role in the CNS is not fully understood. Consistent with work from Klemm and colleagues (60) and Friebel and colleagues (61), we observed multiple monocyte phenotypes within MBM, including TREM2/APOE-expressing reactive microglia (62). TREM2 expression has been inversely correlated with overall survival across multiple histologies (63, 64) and linked to ICI resistance (63, 65). Although we observed TREM2+ reactive microglia in too few samples to evaluate whether they play a significant role in the context of MBM response to ICI (Supplementary Fig. S5C), further work is warranted to explore its role in modulating response to ICI blockade for CNS metastatic disease.

Our analysis of the myeloid compartment provided unique insights into neutrophil heterogeneity within the MBM TME. We identified a significant association between an IL8-expressing neutrophil subset and EMT in neutrophils. Although this neutrophil subset has been previously identified and may correspond to the N2 subset described in other contexts, they have not, to our knowledge, been previously observed in MBMs. Serum IL8 levels have previously been reported to be associated with worse prognosis and reduced clinical benefit of ICIs (37, 66). Because of the significantly higher infiltration of neutrophils in intracranial relative to extracranial tumors (67), it is possible that these IL8-expressing neutrophils play an even greater role in determining intracranial prognosis and ICI responsiveness than they do extracranially. We do note certain discrepancies between previously described N1 and N2 markers and markers observed in our scRNA-seq data, suggesting that the precise markers of different neutrophil phenotypes are context dependent. Further work, including protein-level evidence, is needed to link neutrophils' cell state with tumor-supportive or -suppressive roles and study their differential role within intracranila and extracranial metastases.

Within the T-cell compartment, single-cell immune profiling with associated TCR clonotyping provided clues regarding responsiveness in the MBM setting. We found that multiple factors play roles in determining intracranial MBM response to ICI and recommend that future studies jointly consider these factors whenever possible. Our cohort also provided the unique opportunity to explore the TME of a treatment-naïve individual (MEL027) who went on to respond to ICI. From a histologic perspective, the tumor demonstrated robust lymphocyte infiltration throughout the tumor suggestive of an immunologically “hot” tumor. In addition, features of the blood compartment were reflective of a baseline proinflammatory state with elevated T-cell fraction, Simpson index, and fraction of private T-cell clonotypes. These features of the periphery were recapitulated within the brain metastasis with an elevated T-cell fraction, abundance of effector T-cell clones, robust clonal expansion, and concomitant elevation of private T-cell clones within the tumor.

Although our study provides insights into features defining the MBM TME and potential factors that reflect response to ICI, our results have several limitations. Our cohort reflects a relatively small population of patients with both diverse treatment courses and varied responses to therapy across multiple sites of disease. In addition, we were limited by a lack of matched pre- and post-ICI brain metastasis samples, a challenge inherent with this patient population. Although we were able to analyze a single treatment-naïve patient who went on to respond to ICI, a larger cohort of treatment-naïve individuals who subsequently are treated with ICI is needed to extend the applicability of those findings. We were similarly limited by the ability to sample patient matched cranial and extracranial tumors to precisely decipher at a single-cell resolution the intratumoral features that are unique to the MBM TME.

Nevertheless, the implications of this unique cohort of patients provide intracranial context within the broader context of immunotherapy for metastatic melanoma. Our collective results emphasize the critical role of T-cell mediated response in the setting of ICI for MBM; elucidate the relationship between T cells within the blood and intratumoral compartment; and demonstrate that blood provides insights into ICI response not only for extracranial disease, but also disease within the brain. Moreover, the relationship between T-cell clonal expansion and phenotypic states in the blood and brain has the potential of providing critical insight into the intracranial TME, which may be clinically advantageous when acquisition of brain tumor tissue is limited. Lastly, future work using larger cohorts of human specimens and murine models will be needed to fully understand the intracranial features of the TME that attenuate these initially robust intracranial responses.

C. Alvarez-Breckenridge reports grants from NIH/NINDS R25, American Brain Tumor Association, and AANS/CNS Joint Section on Tumors during the conduct of the study. B.C. Miller reports personal fees from Cellarity and Rheos Medicines, Inc. outside the submitted work. N. Wang reports other support from Seattle Genetics and Merck outside the submitted work. B. Carter reports grants from NIH during the conduct of the study; personal fees from koh young outside the submitted work. D.P. Cahill reports personal fees from Lilly, GSK, Boston Pharmaceuticals, Pyramid Biosciences, and other support from Iconovir outside the submitted work. G.M. Boland reports other support from Olink Proteomics, Palleon Pharmaceuticals, InterVenn Biosciences; personal fees from Merck, Novartis, Nektar Therapeutics, and Ankyra Therapeutics outside the submitted work. M.A. Davies reports grants from NCI, Adelson Medical Research Foundation, American Cancer Society, and Melanoma Research Alliance during the conduct of the study; personal fees from BMS, Roche/Genentech, Novartis, Pfizer, Apexigen, Apexigen, and Eisai; grants and personal fees from ABM Therapeutics outside the submitted work. A.H. Sharpe reports personal fees from Surface Oncology, Sqz Biotech, Selecta, Elpiscience, Bicara; other support from Monopteros; grants from Merck, Novartis, Roche, Ipsen, UCB, AbbVie, and Quark Ventures/Iome outside the submitted work; in addition, A.H. Sharpe has a patent for 7,432,059 licensed and with royalties paid from Roche, Merck, Bristol Myers Squibb, EMD Serono, Boehringer Ingelheim, AstraZeneca, Leica, Mayo Clinic, Dako, and Novartis, a patent for 7,722,868 licensed and with royalties paid from Roche, Merck, Bristol Myers Squibb, EMD Serono, Boehringer Ingelheim, AstraZeneca, Leica, Mayo Clinic, Dako, and Novartis, a patent for 8,652,465 licensed to Roche, a patent for 9,457,080 licensed to Roche, a patent for 9,683,048 licensed to Roche, a patent for 9,815,898 licensed to Novartis, a patent for 9,845,356 licensed to Novartis, a patent for 10,202,454 licensed to Novartis, a patent for 10,457,733 licensed to Novartis, a patent for 9,580,684 issued, a patent for 9,988,452 issued, and a patent for 10,370,446 issued; and is on scientific advisory boards for the Massachusetts General Cancer Center, Program in Cellular and Molecular Medicine at Boston Children's Hospital, the Human Oncology and Pathogenesis Program at Memorial Sloan Kettering Cancer Center, and the GlaxoSmithKine Oncology executive advisory board and Janssen Immunology advisory board. R.J. Sullivan reports personal fees from Pfizer, Bristol Myers Squibb, Iovance, Novartis, Roche-Genentech, Eisai, and OncoSec; grants and personal fees from Merck outside the submitted work. P.K. Brastianos reports grants and non-financial support from Merck; grants from Breast Cancer Research Foundation, Damon Runyon Cancer Research Foundation, Ben and Catherine Ivy Foundation, Terry and Jean de Gunzburg MGH Research Scholar Award, and NIH during the conduct of the study; personal fees from Advise Connect Inspire, Dantari, Elevatebio, Pfizer, Sintetica, SK Life Sciences, Voyager Therapeutics; grants from Mirati; grants and non-financial support from Lilly, BMS, Pfizer, AstraZeneca, Merck, Kazia, Genentech/Roche, and GSK outside the submitted work. No disclosures were reported by the other authors.

C. Alvarez-Breckenridge: Conceptualization, supervision, investigation, writing–original draft, project administration, writing–review and editing. S.C. Markson: Conceptualization, software, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. J.H. Stocking: Data curation, investigation. N. Nayyar: Resources, data curation, investigation, writing–review and editing. M. Lastrapes: Data curation, software, writing–review and editing. M.R. Strickland: Investigation, writing–review and editing. A.E. Kim: Investigation, writing–review and editing. M. de Sauvage: Data curation, investigation, writing–review and editing. A. Dahal: Data curation, investigation, writing–original draft. J.M. Larson: Data curation, investigation, writing–original draft. J.L. Mora: Data curation, investigation. A.W. Navia: Resources. R.H. Klein: Data curation. B.M. Kuter: Data curation, investigation. C.M. Gill: Data curation, investigation. M. Bertalan: Data curation, investigation. B. Shaw: Data curation. A. Kaplan: Data curation. M. Subramanian: Data curation, project administration. A. Jain: Data curation. S. Kumar: Writing–review and editing. H. Danish: Data curation, writing–review and editing. M. White: Data curation. O. Shahid: Writing–review and editing. K.E. Pauken: Writing–review and editing. B.C. Miller: Writing–review and editing. D.T. Frederick: Data curation. C. Hebert: Data curation, investigation. M. Shaw: Data curation, investigation. M. Martinez-Lage: Resources, data curation, investigation, writing–review and editing. M. Frosch: Writing–review and editing. N. Wang: Data curation, investigation. E. Gerstner: Data curation. B.V. Nahed: Investigation. W.T. Curry: Writing–review and editing. B. Carter: Investigation. D.P. Cahill: Writing–review and editing. G.M. Boland: Data curation, writing–review and editing. B. Izar: Writing–review and editing. M.A. Davies: Writing–review and editing. A.H. Sharpe: Writing–review and editing. M.L. Suva: Writing–review and editing. R.J. Sullivan: Writing–review and editing. P.K. Brastianos: Conceptualization, supervision, funding acquisition, project administration, writing–review and editing. S.L. Carter: Conceptualization, supervision, funding acquisition, project administration, writing–review and editing.

Funding for this research was provided by Merck Sharp & Dohme Corp., a subsidiary of Merck & Co., Inc.; Damon Runyon Cancer Research Foundation; Melanoma Research Alliance; Breast Cancer Research Foundation; and 1R01CA244975-01.

P.K. Brastianos and S.L. Carter were supported by the NIH (5R01CA227156-02, 1R21CA220253-0A01, and 1R01CA244975-01). P.K. Brastianos was also supported by the Susan G. Komen, the EOK Demetra Fund from the Hellenic Women's Club and the Terry and Jean de Gunzburg MGH Research Scholar Award. S.L. Carter was also supported by the Wong Family Award, the DF/HCC Lung Cancer Program 25Developmental Research Project Award in Lung Cancer Research, and by Dana-Farber Institutional Research Support. C. Alvarez-Breckenridge was supported by an NIH/NINDS R25, the American Brain Tumor Association, and AANS/CNS Joint Section on Tumors. B. Izar was supported by NIH R37CA258829 and R21CA263381, and the Burroughs Wellcome Fund Career Award for Medical Scientists. B.C. Miller is supported by the NCI of the NIH under 1K08CA248960 and the Wong Family Award. M.A. Davies is supported by the Dr. Miriam and Sheldon G. Adelson Medical Research Foundation, the AIM at Melanoma Foundation, the NIH/NCI (1 P50 CA221703-02), the American Cancer Society and the Melanoma Research Alliance, Cancer Fighters of Houston, the Anne and John Mendelsohn Chair for Cancer Research, and philanthropic contributions to the Melanoma Moon Shots Program of MD Anderson.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.

Note: Supplementary data for this article are available at Cancer Immunology Research Online (http://cancerimmunolres.aacrjournals.org/).

1.
Nieblas-Bedolla
E
,
Nayyar
N
,
Singh
M
,
Sullivan
RJ
,
Brastianos
PK
.
Emerging immunotherapies in the treatment of brain metastases
.
Oncologist
2021
;
26
:
231
41
.
2.
Brastianos
PK
,
Carter
SL
,
Santagata
S
,
Cahill
DP
,
Taylor-Weiner
A
,
Jones
RT
, et al
.
Genomic characterization of brain metastases reveals branched evolution and potential therapeutic targets
.
Cancer Discov
2015
;
5
:
1164
77
.
3.
Twomey
JD
,
Zhang
B
.
Cancer immunotherapy update: FDA-approved checkpoint inhibitors and companion diagnostics
.
AAPS J
2021
;
23
:
39
.
4.
Tawbi
HA
,
Forsyth
PA
,
Algazi
A
,
Hamid
O
,
Hodi
FS
,
Moschos
SJ
, et al
.
Combined nivolumab and ipilimumab in melanoma metastatic to the brain
.
N Engl J Med
2018
;
379
:
722
30
.
5.
Kalbasi
A
,
Ribas
A
.
Tumour-intrinsic resistance to immune checkpoint blockade
.
Nat Rev Immunol
2020
;
20
:
25
39
.
6.
Pitt
JM
,
Vétizou
M
,
Daillère
R
,
Roberti
MP
,
Yamazaki
T
,
Routy
B
, et al
.
Resistance mechanisms to immune-checkpoint blockade in cancer: tumor-intrinsic and -extrinsic factors
.
Immunity
2016
;
44
:
1255
69
.
7.
Louveau
A
,
Smirnov
I
,
Keyes
TJ
,
Eccles
JD
,
Rouhani
SJ
,
Peske
JD
, et al
.
Structural and functional features of central nervous system lymphatic vessels
.
Nature
2015
;
523
:
337
41
.
8.
Arvanitis
CD
,
Ferraro
GB
,
Jain
RK
.
The blood–brain barrier and blood–tumour barrier in brain tumours and metastases
.
Nat Rev Cancer
2020
;
20
:
26
41
.
9.
Lorger
M
,
Andreou
T
,
Fife
C
,
James
F
.
Immune checkpoint blockade – how does it work in brain metastases?
Front Mol Neurosci
2019
;
12
:
282
.
10.
Rizvi
NA
,
Hellmann
MD
,
Snyder
A
,
Kvistborg
P
,
Makarov
V
,
Havel
JJ
, et al
.
Mutational landscape determines sensitivity to PD-1 blockade in non–small cell lung cancer
.
Science
2015
;
348
:
124
8
.
11.
Hugo
W
,
Zaretsky
JM
,
Sun
L
,
Song
C
,
Moreno
BH
,
Hu-Lieskovan
S
, et al
.
Genomic and transcriptomic features of response to anti-PD-1 therapy in metastatic melanoma
.
Cell
2016
;
165
:
35
44
.
12.
Tumeh
PC
,
Harview
CL
,
Yearley
JH
,
Shintaku
IP
,
Taylor
EJM
,
Robert
L
, et al
.
PD-1 blockade induces responses by inhibiting adaptive immune resistance
.
Nature
2014
;
515
:
568
71
.
13.
Topalian
SL
,
Hodi
FS
,
Brahmer
JR
,
Gettinger
SN
,
Smith
DC
,
McDermott
DF
, et al
.
Safety, activity, and immune correlates of anti–PD-1 antibody in cancer
.
N Engl J Med
2012
;
366
:
2443
54
.
14.
Garon
EB
,
Rizvi
NA
,
Hui
R
,
Leighl
N
,
Balmanoukian
AS
,
Eder
JP
, et al
.
Pembrolizumab for the treatment of non–small-cell lung cancer
.
N Engl J Med
2015
;
372
:
2018
28
.
15.
van der Leun
AM
,
Thommen
DS
,
Schumacher
TN
.
CD8+ T cell states in human cancer: insights from single-cell analysis
.
Nat Rev Cancer
2020
;
20
:
218
32
.
16.
McKenna
A
,
Hanna
M
,
Banks
E
,
Sivachenko
A
,
Cibulskis
K
,
Kernytsky
A
, et al
.
The Genome Analysis Toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data
.
Genome Res
2010
;
20
:
1297
303
.
17.
Stubbington
MJT
,
Lönnberg
T
,
Proserpio
V
,
Clare
S
,
Speak
AO
,
Dougan
G
, et al
.
T cell fate and clonality inference from single-cell transcriptomes
.
Nat Methods
2016
;
13
:
329
32
.
18.
Li
B
,
Dewey
CN
.
RSEM: accurate transcript quantification from RNA-seq data with or without a reference genome
.
BMC Bioinformatics
2011
;
12
:
323
.
19.
Li
B
,
Gould
J
,
Yang
Y
,
Sarkizova
S
,
Tabaka
M
,
Ashenberg
O
, et al
.
Cumulus provides cloud-based data analysis for large-scale single-cell and single-nucleus RNA-seq
.
Nat Methods
2020
;
17
:
793
8
.
20.
Langmead
B
,
Salzberg
SL
.
Fast gapped-read alignment with Bowtie 2
.
Nat Methods
2012
;
9
:
357
9
.
21.
Tirosh
I
,
Izar
B
,
Prakadan
SM
,
Wadsworth
MH
,
Treacy
D
,
Trombetta
JJ
, et al
.
Dissecting the multicellular ecosystem of metastatic melanoma by single-cell RNA-seq
.
Science
2016
;
352
:
189
96
.
22.
Subramanian
A
,
Tamayo
P
,
Mootha
VK
,
Mukherjee
S
,
Ebert
BL
,
Gillette
MA
, et al
.
Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles
.
Proc Natl Acad Sci U S A
2005
;
102
:
15545
50
.
23.
Liberzon
A
,
Subramanian
A
,
Pinchback
R
,
Thorvaldsdóttir
H
,
Tamayo
P
,
Mesirov
JP
.
Molecular signatures database (MSigDB) 3.0
.
Bioinformatics
2011
;
27
:
1739
40
.
24.
Jerby-Arnon
L
,
Shah
P
,
Cuoco
MS
,
Rodman
C
,
Su
M-J
,
Melms
JC
, et al
.
A cancer cell program promotes T cell exclusion and resistance to checkpoint blockade
.
Cell
2018
;
175
:
984
97
.
25.
Li
H
,
van der Leun
AM
,
Yofe
I
,
Lubling
Y
,
Gelbard-Solodkin
D
,
van Akkooi
ACJ
, et al
.
Dysfunctional CD8 T cells form a proliferative, dynamically regulated compartment within human melanoma
.
Cell
2019
;
176
:
775
89
.
26.
Mori
L
,
Lepore
M
,
De Libero
G
.
The immunology of CD1- and MR1-restricted T cells
.
Annu Rev Immunol
2016
;
34
:
479
510
.
27.
Nolan
S
,
Vignali
M
,
Klinger
M
,
Dines
JN
,
Kaplan
IM
,
Svejnoha
E
, et al
.
A large-scale database of T-cell receptor beta (TCRb) sequences and binding associations from natural and synthetic exposure to SARS-CoV-2 [database on the Internet]
.
c2020 [update 2020 July 7; cited 2022 July 6]. Available from
: https://doi.org/10.21417/ADPT2020COVID.
28.
Venturi
V
,
P
DA
,
Douek
DC
,
Davenport
MP
.
The molecular basis for public T-cell responses?
Nat Rev Immunol
2008
;
8
:
231
8
.
29.
Picelli
S
,
Faridani
OR
,
Björklund
ÅK
,
Winberg
G
,
Sagasser
S
,
Sandberg
R
.
Full-length RNA-seq from single cells using Smart-seq2
.
Nat Protoc
2014
;
9
:
171
81
.
30.
Ramachandran
P
,
Dobie
R
,
Wilson-Kanamori
JR
,
Dora
EF
,
Henderson
BEP
,
Luu
NT
, et al
.
Resolving the fibrotic niche of human liver cirrhosis at single-cell level
.
Nature
2019
;
575
:
512
8
.
31.
Mahmud
FJ
,
Du
Y
,
Greif
E
,
Boucher
T
,
Dannals
RF
,
Mathews
WB
, et al
.
Osteopontin/secreted phosphoprotein-1 behaves as a molecular brake regulating the neuroinflammatory response to chronic viral infection
.
J Neuroinflammation
2020
;
17
:
273
.
32.
Hammond
TR
,
Dufort
C
,
Dissing-Olesen
L
,
Giera
S
,
Young
A
,
Wysoker
A
, et al
.
Single-cell RNA sequencing of microglia throughout the mouse lifespan and in the injured brain reveals complex cell-state changes
.
Immunity
2019
;
50
:
253
71
.
33.
Ng
LG
,
Ostuni
R
,
Hidalgo
A
.
Heterogeneity of neutrophils
.
Nat Rev Immunol
2019
;
19
:
255
65
.
34.
Fridlender
ZG
,
Sun
J
,
Kim
S
,
Kapoor
V
,
Cheng
G
,
Ling
L
, et al
.
Polarization of tumor-associated neutrophil phenotype by TGF-β: “N1” versus “N2” TAN
.
Cancer Cell
2009
;
16
:
183
94
.
35.
Gentles
AJ
,
Newman
AM
,
Liu
CL
,
Bratman
SV
,
Feng
W
,
Kim
D
, et al
.
The prognostic landscape of genes and infiltrating immune cells across human cancers
.
Nat Med
2015
;
21
:
938
45
.
36.
Valero
C
,
Lee
M
,
Hoen
D
,
Weiss
K
,
Kelly
DW
,
Adusumilli
PS
, et al
.
Pretreatment neutrophil-to-lymphocyte ratio and mutational burden as biomarkers of tumor response to immune checkpoint inhibitors
.
Nat Commun
2021
;
12
:
729
.
37.
Schalper
KA
,
Carleton
M
,
Zhou
M
,
Chen
T
,
Feng
Y
,
Huang
S-P
, et al
.
Elevated serum interleukin-8 is associated with enhanced intratumor neutrophils and reduced clinical benefit of immune-checkpoint inhibitors
.
Nat Med
2020
;
26
:
688
92
.
38.
Teijeira
A
,
Garasa
S
,
Ochoa
MC
,
Villalba
M
,
Olivera
I
,
Cirella
A
, et al
.
IL8, neutrophils, and NETs in a collusion against cancer immunity and immunotherapy
.
Clin Cancer Res
2021
;
27
:
2383
93
.
39.
Papayannopoulos
V
.
Neutrophil extracellular traps in immunity and disease
.
Nat Rev Immunol
2018
;
18
:
134
47
.
40.
Joyce
JA
,
Fearon
DT
.
T cell exclusion, immune privilege, and the tumor microenvironment
.
Science
2015
;
348
:
74
80
.
41.
Berghoff
AS
,
Fuchs
E
,
Ricken
G
,
Mlecnik
B
,
Bindea
G
,
Spanberger
T
, et al
.
Density of tumor-infiltrating lymphocytes correlates with extent of brain edema and overall survival time in patients with brain metastases
.
OncoImmunology
2016
;
5
:
e1057388
.
42.
Berghoff
AS
,
Ricken
G
,
Wilhelm
D
,
Rajky
O
,
Widhalm
G
,
Dieckmann
K
, et al
.
Tumor infiltrating lymphocytes and PD-L1 expression in brain metastases of small cell lung cancer (SCLC)
.
J Neurooncol
2016
;
130
:
19
29
.
43.
Sade-Feldman
M
,
Yizhak
K
,
Bjorgaard
SL
,
Ray
JP
,
de Boer
CG
,
Jenkins
RW
, et al
.
Defining T cell states associated with response to checkpoint immunotherapy in melanoma
.
Cell
2018
;
175
:
998
1013
.
44.
Miller
BC
,
Sen
DR
,
Abosy
RA
,
Bi
K
,
Virkud
YV
,
LaFleur
MW
, et al
.
Subsets of exhausted CD8+ T cells differentially mediate tumor control and respond to checkpoint blockade
.
Nat Immunol
2019
;
20
:
326
36
.
45.
Thommen
DS
,
Koelzer
VH
,
Herzig
P
,
Roller
A
,
Trefny
M
,
Dimeloe
S
, et al
.
A transcriptionally and functionally distinct PD-1+ CD8+ T cell pool with predictive potential in non-small-cell lung cancer treated with PD-1 blockade
.
Nat Med
2018
;
24
:
994
1004
.
46.
Yost
KE
,
Satpathy
AT
,
Wells
DK
,
Qi
Y
,
Wang
C
,
Kageyama
R
, et al
.
Clonal replacement of tumor-specific T cells following PD-1 blockade
.
Nat Med
2019
;
25
:
1251
9
.
47.
Valpione
S
,
Galvani
E
,
Tweedy
J
,
Mundra
PA
,
Banyard
A
,
Middlehurst
P
, et al
.
Immune awakening revealed by peripheral T cell dynamics after one cycle of immunotherapy
.
Nat Cancer
2020
;
1
:
210
21
.
48.
Mathewson
ND
,
Ashenberg
O
,
Tirosh
I
,
Gritsch
S
,
Perez
EM
,
Marx
S
, et al
.
Inhibitory CD161 receptor identified in glioma-infiltrating T cells by single-cell analysis
.
Cell
2021
;
184
:
1281
98
.
49.
Mahajan
VS
,
Leskov
IB
,
Chen
JZ
.
Homeostasis of T cell diversity
.
Cell Mol Immunol
2005
;
2
:
1
10
.
50.
Wu
TD
,
Madireddi
S
,
de Almeida
PE
,
Banchereau
R
,
Chen
Y-JJ
,
Chitre
AS
, et al
.
Peripheral T cell expansion predicts tumour infiltration and clinical response
.
Nature
2020
;
579
:
274
8
.
51.
Lucca
LE
,
Axisa
P-P
,
Lu
B
,
Harnett
B
,
Jessel
S
,
Zhang
L
, et al
.
Circulating clonally expanded T cells reflect functions of tumor-infiltrating T cells
.
J Exp Med
2021
;
218
:
e20200921
.
52.
Pauken
KE
,
Shahid
O
,
Lagattuta
KA
,
Mahuron
KM
,
Luber
JM
,
Lowe
MM
, et al
.
Single-cell analyses identify circulating anti-tumor CD8 T cells and markers for their enrichment
.
J Exp Med
2021
;
218
:
e20200920
.
53.
Horn
HS
.
Measurement of “Overlap” in comparative ecological studies
.
Am Nat
1966
;
100
:
419
24
.
54.
Bagaev
DV
,
Vroomans
RMA
,
Samir
J
,
Stervbo
U
,
Rius
C
,
Dolton
G
, et al
.
VDJdb in 2019: database extension, new analysis infrastructure and a T-cell receptor motif compendium
.
Nucleic Acids Res
2020
;
48
:
D1057
62
.
55.
Li
F
,
Li
C
,
Cai
X
,
Xie
Z
,
Zhou
L
,
Cheng
B
, et al
.
The association between CD8+ tumor-infiltrating lymphocytes and the clinical outcome of cancer immunotherapy: a systematic review and meta-analysis
.
EClinicalMedicine
2021
;
41
:
101134
.
56.
Reuben
A
,
Gittelman
R
,
Gao
J
,
Zhang
J
,
Yusko
EC
,
Wu
C-J
, et al
.
TCR repertoire intratumor heterogeneity in localized lung adenocarcinomas: an association with predicted neoantigen heterogeneity and postsurgical recurrence
.
Cancer Discov
2017
;
7
:
1088
97
.
57.
Louveau
A
,
Harris
TH
,
Kipnis
J
.
Revisiting the mechanisms of CNS immune privilege
.
Trends Immunol
2015
;
36
:
569
77
.
58.
Caushi
JX
,
Zhang
J
,
Ji
Z
,
Vaghasia
A
,
Zhang
B
,
Hsiue
EH-C
, et al
.
Transcriptional programs of neoantigen-specific TIL in anti-PD-1-treated lung cancers
.
Nature
2021
;
596
:
126
32
.
59.
Veglia
F
,
Perego
M
,
Gabrilovich
D
.
Myeloid-derived suppressor cells coming of age
.
Nat Immunol
2018
;
19
:
108
19
.
60.
Klemm
F
,
Maas
RR
,
Bowman
RL
,
Kornete
M
,
Soukup
K
,
Nassiri
S
, et al
.
Interrogation of the microenvironmental landscape in brain tumors reveals disease-specific alterations of immune cells
.
Cell
2020
;
181
:
1643
60
.
61.
Friebel
E
,
Kapolou
K
,
Unger
S
,
Núñez
NG
,
Utz
S
,
Rushing
EJ
, et al
.
Single-cell mapping of human brain cancer reveals tumor-specific instruction of tissue-invading leukocytes
.
Cell
2020
;
181
:
1626
42
.
62.
Wang
Y
,
Cella
M
,
Mallinson
K
,
Ulrich
JD
,
Young
KL
,
Robinette
ML
, et al
.
TREM2 lipid sensing sustains the microglial response in an Alzheimer's disease model
.
Cell
2015
;
160
:
1061
71
.
63.
Molgora
M
,
Esaulova
E
,
Vermi
W
,
Hou
J
,
Chen
Y
,
Luo
J
, et al
.
TREM2 modulation remodels the tumor myeloid landscape enhancing anti-PD-1 immunotherapy
.
Cell
2020
;
182
:
886
900
.
64.
Cheng
B
,
Li
X
,
Dai
K
,
Duan
S
,
Rong
Z
,
Chen
Y
, et al
.
Triggering receptor expressed on myeloid cells-2 (TREM2) interacts with colony-stimulating factor 1 receptor (CSF1R) but is not necessary for CSF1/CSF1R-mediated microglial survival
.
Front Immunol
2021
;
12
:
633796
.
65.
Xiong
D
,
Wang
Y
,
You
M
.
A gene expression signature of TREM2hi macrophages and γδ T cells predicts immunotherapy response
.
Nat Commun
2020
;
11
:
5084
.
66.
Yuen
KC
,
Liu
L-F
,
Gupta
V
,
Madireddi
S
,
Keerthivasan
S
,
Li
C
, et al
.
High systemic and tumor-associated IL-8 correlates with reduced clinical benefit of PD-L1 blockade
.
Nat Med
2020
;
26
:
693
8
.
67.
Fischer
GM
,
Jalali
A
,
Kircher
DA
,
Lee
W-C
,
McQuade
JL
,
Haydu
LE
, et al
.
Molecular profiling reveals unique immune and metabolic features of melanoma brain metastases
.
Cancer Discov
2019
;
9
:
628
45
.

Supplementary data