Patients with BRAF-mutant melanoma show substantial responses to combined BRAF and MEK inhibition, but most relapse within 2 years. A major reservoir for drug resistance is minimal residual disease (MRD), comprised of drug-tolerant tumor cells laying in a dormant state. Towards exploiting potential therapeutic vulnerabilities of MRD, we established a genetically engineered mouse model of BrafV600E-driven melanoma MRD wherein genetic BrafV600E extinction leads to strong but incomplete tumor regression. Transcriptional time-course analysis after BrafV600E extinction revealed that after an initial surge of immune activation, tumors later became immunologically “cold” after MRD establishment. Computational analysis identified candidate T-cell recruiting chemokines as strongly upregulated initially and steeply decreasing as the immune response faded. Therefore, we hypothesized that sustaining chemokine signaling could impair MRD maintenance through increased recruitment of effector T cells. We found that intratumoral administration of recombinant Cxcl9 (rCxcl9), either naked or loaded in microparticles, significantly impaired MRD relapse in BRAF-inhibited tumors, including several complete pathologic responses after microparticle-delivered rCxcl9 combined with BRAF and MEK inhibition. Our experiments constitute proof of concept that chemokine-based microparticle delivery systems are a potential strategy to forestall tumor relapse and thus improve the clinical success of first-line treatment methods.

BRAF-mutant melanoma has become an archetype of targeted therapy after the successful advent of BRAF and MEK inhibitors (BRAFi and MEKi) in clinical practice. Melanoma patients carrying the BRAFV600E alteration have impressive responses to BRAFi+MEKi combinations, a therapeutic strategy that extends the lives of thousands of patients every year (1). Nevertheless, most of these patients show signs of relapse within the first 2 years after treatment start (1). The most frequent molecular mechanism of drug resistance is MAPK pathway reactivation through events such as BRAF amplification or the acquisition of additional MAPK-activating alterations (2). The occurrence of drug resistance and the eventual relapse can be explained by the survival of a cadre of tumor cells that eventually leads to tumor regrowth. These drug-tolerant cells constitute the minimal residual disease (MRD), where cancer cells linger in a dormant state (3).

MRD can be fueled by intrinsic features of cancer cells or extrinsic mechanisms of the microenvironment. Among the extrinsic factors, one of the most relevant is the immune system. When targeted therapy is administered, a strong immune activation is triggered (4). BRAFi have been demonstrated to induce the expression of melanoma antigens, favoring the recognition by the immune system and the infiltration of CD8+ effector and CD4+ helper T cells (5). In this regard, the magnitude of the mounted response is key to disease eradication. If a few cancer cells manage to escape the initial immune response, they can fuel the establishment of MRD. Consistently, the most favorable long-term responses to BRAFi+MEKi are associated with a baseline elevated immune infiltrate, a status described as a “hot” tumor bed (6–9). Contrariwise, a “cold” tumor microenvironment is among the adverse prognostic factors of targeted therapy. A cold tumor microenvironment has a high abundance of suppressive cells and/or a low percent of effector and helper T cells (10). In such a microenvironment, MRD has an increased possibility of surviving during treatment and, over time, acquiring the features to relapse to a full-blown tumor. This is true also for other therapeutic approaches, including immune checkpoint inhibitor therapy (10).

Converting a cold tumor microenvironment to a hot one is a potential mechanism to impair the establishment of MRD and prolong the clinical success of targeted therapy. Recruiting more effector cells to the tumor site can be a viable strategy for patients with low baseline immune infiltration. In this work, we present a microparticle-based approach to favor the recruitment of CD8+ T cells into melanoma MRD using the computationally identified chemokine Cxcl9. We demonstrate that the administration of Cxcl9 induced an ingress of CD8+ T cells into the tumors and that, if administered concomitantly with BRAFi, it significantly delayed the occurrence of tumor relapse.

Genetically engineered mouse models, mouse allografts, and tumor treatments

All animal experiments were approved by the Institutional Animal Care and Use Committees of the University of Texas MD Anderson Cancer Center and performed according to the approved protocols. The inducible Braf Ink/Arf Pten (iBIP) model of BRAFi-resistant melanoma has been previously created by our group and published (9). For this study, we further modified the iBIP mice by swapping in a conditional floxed Ink/Arf locus for the constitutive Ink/Arf knockout by standard breeding practices. This prevented Cdkn2a−/−-driven nonmelanoma tumors from forming. All iBIP mice used in this study were this modified version, and were on a 70% FVB and 30% mixed background but intercrossed for > 20 generations to create a recombinant inbred background suitable for immunocompetent syngeneic tumor establishment. To initiate tumors, iBIP mice were treated with 4-hydroxy-tamoxifen (4-OHT; 70% Z-isomer, 30% E-isomer, H6278; Sigma-Aldrich) dissolved in 100% EtOH. One microliter of 100 μmol/L 4-OHT was applied once on the tip of each ear. Mice were then continually administered doxycycline (dox) either through the drinking water (2 mg/mL; D43020; Research Products International) or in the chow (200 mg/kg) (S3888; Bio-Serv). No differences were seen in tumor penetrance or latency between the water and chow vehicles.

For iBIP allografts, iBIP cell lines were established from tumors by cutting the tissues into small pieces (1–3 mm in diameter) and digesting them in collagenase IV (100 U/mL, #17104019, Thermo Fisher) and DNAse I (10 μg/mL, #07469, StemCell Technologies) at 37°C for 2 hours. Digested cell solutions were filtered through a strainer (100 μm) and cultivated in RPMI1640 (HyClone RPMI1640 medium, sterile, pH 7.0 - 7.4, with l-glutamine, #16777–146, VWR), 10% FBS (Avantor Seradigm Select Grade USDA Approved Origin Fetal Bovine Serum, #89510–186 VWR), and 2 mg/mL dox (Doxycycline hyclate ≥ 98%, #89150–324, VWR). Cells were detached with Trypsin-EDTA (Corning, Trypsin EDTA 1 × 0.25% Trypsin/2.21 mmol/L EDTA in HBSS, #25–053-CI, Neta Scientific) when in the exponential growth curve. Cells were then resuspended (1 × 105/100 μL) in HBSS (#14025092, Thermo Scientific) and injected intradermally into NOD/SCID gamma (NSG) immunodeficient (Jackson Lab) or syngeneic immunocompetent iBIP mice.

“BP” and Yumm1.7 tumors were established from mouse cancer cell lines as previously described (11, 12). BP was established from a BRAFV600E;PTEN−/− melanoma, while Yumm1.7 is from a BRAFV600E;PTEN−/−; CDKN2A−/− melanoma. Tumor cell lines were a kind gift of Dr. Marcus Bosenberg (received in 2018, last authentication by STR in 2018; cells are tested for Mycoplasma every 3 months; cells are not used beyond the 15th passage in vitro). Briefly, tumor mouse cells were cultivated in vitro, in RPMI, 10% FBS. Cells were detached with Trypsin-EDTA (0.25%) when in the exponential growth curve. Cells were then resuspended in HBSS (105/100 μL) and injected intradermally in C57BL/6J mice (Jackson Laboratory, Bar Harbor, ME).

For intratumoral injection of chemokines or microparticles (described below), saline (#470302–026, VWR) was used as a vehicle in a maximum volume of 100 μL.

Tumor volumes were calculated using electronic calipers to measure the length (l), width (w), and height (h) and using the formula (l × w × h) × π/6. For all the experiments, when the tumors reached the average volume of 100 mm3, animals were distributed among the treatment groups. PLX4720 chow (417 parts per million, ppm Research Diets Inc.) and its corresponding control diet replaced the standard mouse chow when indicated. MEK162 (#T2508, TargetMol) was resuspended in 1% carboxymethylcellulose/0.5% Tween 80 and administered 7 mg/kg/day by oral gavage using sterile flexible plastic adapters in a volume of 100 μL.

Body mass was measured using an electronic scale. Animals were euthanized when the tumor burden reached > 1,000 mm3 or when the tumor became ulcerated.

RNA extraction, expression microarrays, and bioinformatics analysis

Mouse tumors were collected at the indicated time points and microdissected under a microscope to remove normal skin and cartilage contamination. Tumors were homogenized in TRIzol solution (#15596026, Thermo Scientific). Nucleic acids were extracted using chloroform, precipitated with isopropanol, washed with 70% ethanol, and resuspended in RNAse-free water. DNA was removed using DNAse (#M6101, Promega) in the presence of RNAse inhibitors (#N2111, Promega) by incubating for 40 minutes at 37°C. Samples were processed through the RNEasy kit (#74004, Qiagen) to wash and elute the RNA. Samples were tested for purity by measuring 260/230 and 260/280 values on a Nanodrop machine (Thermo Scientific). Impure samples (<1.8 ratio) were repurified using standard ethanol precipitation. Further quality control was performed by the Dana-Farber Cancer Institute Microarray Core facility (http://chip.dfci.harvard.edu/) using the Bioanalyzer platform (Agilent). A minimum of 1 μg of RNA per sample was run on a Mouse Genome 430 2.0 Array (#900497, Affymetrix). Microarray data are accessible through Gene Expression Omnibus (GEO) under the accession number GSE79972. Gene expression data were processed using the affy R package (13), normalized using RMA (14), and differential expression using the limma R package (15).

The Short Time-series Expression Miner (STEM) algorithm (16, 17) was used to detect significant trends in genes over time (k-means clustering, k = 5). Gene set enrichment using the hypergeometric test and the MSigDB gene set C2.CP.v7.5.1 was performed on all clusters and used to rank gene sets on statistical significance. Gene set enrichment analysis (GSEA) was performed using the preranked program and the MSigDB gene set C5.CP.v7.5.1 gene ontology (GO; refs. 18, 19). Gene Expression Deconvolution Interactive Tool (GEDIT; ref. 20), using the Minimum Entropy Ranking metric and the Mouse Body Atlas (21).

TRAP analysis

The TRAP algorithm has been previously described (22). Here, we have modified the network analysis to focus on cytokines-to-pathway relationships, whereas the original TRAP network focused on transcription factor-to-pathway relationships. Using Biological Process Gene Ontology annotations (GO:0005125, cytokine activity), we first defined a set of 229 putative cytokines. For pathways, we used the canonical pathway gene sets as defined by MSigDB (C2.CP.v7.5.1; ref. 23). With cytokines and pathways defined, we generated the background network model as described in (22). Briefly, using the compendium of mouse gene expression data defined in (22), we calculated pathway expression for each gene set as the average expression of all genes in the gene set (with any of the 229 cytokines removed). Cytokine-to-pathway relationships were calculated using mutual information, and a threshold of 3 standard deviations away from the mean was used to define cytokine-to-pathway relationships. The final TRAP network consisted of 13,629 edges that connected 157 cytokines to 2,079 pathways.

The pathways identified from the GO term enrichment using the STEM clustering approach on the immune cluster were used as input to rank cytokine centrality in the TRAP network. A subnetwork of 748 edges was defined by the selected pathways (n = 68) and the immediate neighbor cytokines (n = 98). A degree centrality measure was used to rank cytokines that had the most edges in the subnetwork compared with the full TRAP network.

IHC

Formalin-fixed, paraffin-embedded (FFPE) mouse tumor specimens (iBIP and Yumm1.7) were sectioned with a microtome (5-μm sections) and put on slides. After deparaffinization, citrate-based antigen retrieval was performed. Blocking was performed using Normal Horse Serum (#MP7801, ImmPRESS Reagent Kit, Vector Labs). Slides were then incubated with one of the following primary antibodies overnight at 4°C in a humid chamber: anti-Cxcl9 (#701117, clone 11H1L14, Invitrogen), anti-CD8 (#14–0808–82, Clone 4SM15, Invitrogen), or anti-CD4 (#14–0041–82, GK1.5, Invitrogen). Slides were then incubated with horseradish peroxidase–conjugated anti-rabbit secondary antibody (#MP7801, ImmPRESS Reagent Kit, Vector Labs). Processed slides were finally added with TSA Plus Fluorescein System (#NEL701A001KT, Perkin-Elmer) to allow fluorophore deposition and counterstained with hematoxylin (#H3401, Vector Laboratories) and DAPI (#5930–0006, SeraCare). Coverslips were mounted after dehydration of the sections using Permanent Mounting Medium (#H5000, Vector Laboratories). Images were digitally acquired with a Nikon Eclipse T2 microscope connected to DS-Ri2 camera (Nikon). Quantification of IHC expressed as the number of positive cells/number of total cells was performed using ImageJ (Rasband, W.S., ImageJ, NIH, Bethesda, MD, https://imagej.nih.gov/ij/, 1997–2016).

Immunofluorescence

FFPE mouse tumor specimens (iBIP and Yumm1.7) were sectioned with a microtome (5-μm sections) and put on slides. After deparaffinization, citrate-based antigen retrieval was performed. Blocking was performed using Normal Horse Serum. Slides were then incubated with the following primary antibodies overnight at 4°C in a humid chamber: anti-Cxcr3 (#LS-B10183, LSBio) and anti-CD8 (#14–0808–82, Clone 4SM15, Invitrogen). Slides were then incubated with fluorescence-conjugated anti-rabbit or anti-rat secondary antibodies (#A11008 Alexa Fluor 488, #A21247 Alexa Fluor 647, ThermoScientific). Processed slides were counterstained with hematoxylin. Slides were processed with TrueVIEW Autofluorescence Quenching Kit (#SP8400, Vector Labs), and coverslips were mounted using VECTASHIELD Vibrance Antifade Mounting Medium with DAPI (#H1800, Vector Labs). Images were digitally acquired with the ImageXpress Pico Imaging system (Molecular Devices). Quantification was performed using the CellReporterXpress Image Acquisition and Analysis Software v2.0 (Molecular Devices).

Patients’ RNA sequencing data

Melanoma patients’ RNA sequencing (RNA-seq) methods and data were previously published and are available from the European Genome-phenome Archive (EGA S00001000992; ref. 9). It consists of 14 pre-BRAFi, 12 on-BRAFi, and 12 post-BRAFi samples. Student's t test was used to identify significance using GraphPad Prism. Patient samples were exhausted during the analysis.

SMM102 model RNA-seq data

RNA-seq methods and data (GEO dataset GSE161430) regarding the experiments on SMM102 tumors were previously published (24). Raw data for BRAFi-treated animals and respective control (Vehicle) were obtained from repository and expressed as fold-change versus Vehicle.

Chemotaxis and activation assays

EasySep Mouse CD8+ T Cell Isolation Kit (#19853, StemCell Technologies) was used according to the manufacturer's instructions to derive CD8+ T cells from the spleens of the animals (C57BL/6J mice). Isolated T cells were then cultured in plates pre-coated with anti-CD3 (0.5 μg/mL, #100339, Clone 145–2C11, BioLegend) and anti-CD28 (5 μg/mL, #102115, clone 37.51, BioLegend), for lymphocyte activation. The CD8+ T cells were cultured in RPMI1640 medium containing 10% heat-inactivated FBS and supplemented with l-glutamine (2 mmol/L, # 25030081, ThermoFisher Scientific), penicillin+streptomycin (50 U/mL+ 50 μg/mL, # 15140163, ThermoFisher Scientific), 2-mercaptoethanol (50 μmol/L), and IL2 (30 U/mL, #212–12, PeproTech).

Following 72 hours of activation, CD8+ lymphocytes were assayed through a chemotaxis experiment in a Boyden chamber (5-μm pores, 24 well plates). Briefly, lymphocytes from C57BL/6J mice. were counted and distributed in the top chambers (100,000 cells/chamber). Then, recombinant chemokines or microparticle-soaked medium were added to the bottom chamber at the desired concentration. Medium and FBS concentration was the same for the bottom and top chamber (RPMI1640 medium + 10% FBS). After 3 hours, bottom and top chamber media were collected, and the number of cells was counted through a Guava Flow Cytometer. The migration index was calculated as the number of cells migrated to the bottom chamber/total number of cells and normalized to the control group.

For T-cell activation assessment, CD8+ T cells were incubated in the presence of the indicated recombinant mouse chemokines (rCxcl9 #250–18, rCxcl10 #250–16, rCxcl11 #250–29, all PeproTech) for 72 hours, collected and analyzed through flow cytometry.

Porous silica nanoparticle synthesis

Cetyltrimethylammonium bromide (CTAB, 141.75 mg) was dissolved in 70 mL dH2O in a 250 mL screw-cap container. Ammonium hydroxide (NH3·H2O, 3.03 mL; 28%–30%) was added dropwise into the reaction flask. The mixture was stirred for 1 hour at 1,200 rpm. Tetraethyl orthosilicate (TEOS, 0.6025 mL) was added to the mixture and stirred vigorously for 4 hours. The mixture was centrifuged at 5,000 rpm for 10 minutes. The particles were washed and dispersed in EtOH and H2O 3 times (ratio 1:1 v/v). The mixture was centrifuged at 5,000 rpm for 10 minutes between each wash. The final solution was biphasic. The particles were dispersed in 20 mL of 1:1 1 N acetic acid:dichloromethane 3 times. The mixture was washed in H2O twice and centrifuged at 5,000 rpm for 10 minutes between each wash. The mixture was resuspended in isopropyl alcohol for storage. When ready to be processed, the mixture was then placed in a vacuum oven at 60°C and −36 Pa to dry before proceeding with loading and encapsulation with poly(DL-lactide-co-glycolide) acid (PLGA) 50:50 (see Porous Silica Nanoparticle encapsulation with PLGA 50:50). A schematic of the process is provided in Supplementary Fig. S1A. Size and polydispersity index (PDI) were measured using dynamic light scattering (DLS). Size and PDI were evaluated as an average of 5 measurements. The physical characteristics of silica particles were evaluated using transmission electron microscopes (TEM) and scanning electron microscopes (SEM; Supplementary Fig. S1B and S1C), revealing the characteristic porous structure. DLS reported an average size of 371 ± 5.7 nm and a PDI of 0.093 ± 0.027 (Supplementary Fig. S1D–S1E), indicating that nanoparticles were homogenous and uniform. We then included the silica core in a PLGA 5% or 10% shell (Supplementary Fig. S1F–S1G), which increased the average particle size to 1,003 ± 47 nm (Supplementary Fig. S1H). The average rCxcl9 encapsulation efficiency was 39.5% ± 0.9% (Supplementary Fig. S1I).

Loading of silica nanoparticles with chemokines

The desired amount of dried silica nanoparticles was placed into a 1.5-mL Eppendorf tube. The loading solution was prepared with 2.5 μg (in vitro experiments) or 50 μg (in vivo experiment) of rCxcl9 (with BSA carrier, #250–18, Preprotech) in 500-μL PBS. The particles in solution were placed on an orbital mixer at 37°C for 2 hours. Particles were centrifuged to separate them from the solution (5,000 rpm for 10 minutes). The loaded particles were dried in a lyophilizer for 3 to 8 hours.

Porous silica nanoparticle encapsulation with PLGA 50:50

A water solution (2.5% and 1% PVA) was prepared as follows: (i) 5% PVA stock solution was prepared by placing 25 g PVA (13.000–23.000 MW), and 500-mL DI water in a screw-top glass container and autoclaved; (ii) 5% stock PVA was diluted to 2.5% PVA and 1% PVA with DI water. An oil solution (10% or 5%wt PLGA 50:50/DCM) was prepared as follows: 100 mg (105 solution) or 50 mg (5% solution) of PLGA were placed (50:50, viscosity range 0.76–0.94 dL/G, lot#A17–065, Durect Corporation) in a scintillation vial with 2-mL DCM.

Emulsion #1 was then performed by placing the oil solution into the homogenizer tube. Silica particles were added to the oil solution. In the homogenizer tube, 3 mL of 2.5% PVA were added, and the solution was homogenized for 5 minutes at 1,200 rpm. Emulsion #2 was then performed by placing Emulsion #1 (5 mL) in water solution (40 mL, 1% PVA) in a 100mL beaker on a stir plate, covering it, and stirring at 600 rpm for a minimum of 6 hours (or overnight). Contents of the beaker were transferred into 50-mL falcon tubes and centrifuged at 5,000 rpm for 10 minutes. The supernatant was removed. Particles were then transferred to clean 1.5-mL lock-cap Eppendorf tubes, frozen for 1 hour at −80°C, then lyophilized for a minimum of 4 hours. Particles were stored at −80°C until ready to be used. Imaging of particles was performed using SEM and TEM. The specific PLGA-silica formulation was selected as the PLGA copolymer is approved by the FDA and European Medicines Agency for drug-delivery applications. DLS and ζ-potential were performed using a Zetasizer ZEN3600 (Malvern). DLS and ζ-potential were evaluated as an average of 5 measurements. Dry samples were prepared for SEM on a stage and sputter coated with a 7-nm thick layer of platinum/palladium using a Sputter Coater 208HR. Microscopy images were taken using Nova NanoSEM 230. High-resolution transmission electron microscopy was performed using a FEI Tecnai electron microscope at a tension of 200 kV. Samples were prepared by placing 1 drop of particles suspension on the 300 mesh carbon-coated copper grids (#3530C, SPI Supplies) and dried. The rCxcl9 encapsulation efficacy was indirectly measured by the quantity of rCxcl9 in the washing waste with ELISA assay (#DY492, R&D).

Flow cytometry

Mouse tumors were collected, cut into small pieces (1–3 mm in diameter), and digested with a mixture of Collagenase IV (100 U/mL, #17104019, Thermo Fisher) and DNAse I (10 μg/mL, #07469, StemCell Technologies). Tumors were incubated for 2 hours at 37°C, centrifuged, and filtered through a 100-μm strainer. Red Blood Lysis solution was applied (#118–156–101, Quality Biological). Cells were then resuspended at a density of 1×106/mL in flow cytometry binding buffer and stained with the following fluorophore-conjugated antibodies (BioLegend): CD8 (clone 53–6.7), Cxcr3 (clone CXCR3–173), Cd69 (clone H1.2F3), and Cd25 (clone PC61). 7-AAD Viability Staining Solution was used following the manufacturer's instructions (#420403, BioLegend). Guava easyCyte 11HT Benchtop Flow Cytometer was used for the acquisition, and results were analyzed through FlowJo v10.8 Software (BD Life Sciences). The gating strategy for mouse samples flow cytometric analysis is reported in Supplementary Fig. S2.

Statistics

The statistical analysis and corresponding graphs were performed using GraphPad Prism v9 (GraphPad Software LLC). A P value < 0.05 calculated by the software was considered significant in t tests and Log-rank tests (Mantel–Cox) reported in the paper.

Data availability

Microarray data are accessible through GEO under the accession number GSE79972. All other data are available in the figures and Supplementary Materials or from the corresponding author upon reasonable request.

A mouse model of melanoma MRD

As an MRD model, we used a modified version of our previously described inducible and conditional genetically engineered mouse model (GEMM) model of BrafV600E mutant melanoma, iBIP (9), in which we swapped in a conditional Cdkn2a knockout allele for the original constitutive one to increase specificity. Briefly, topical 4-OHT and systemic dox administration restrict BrafV600E expression and Pten and Cdkn2a knockout to melanocytes. These develop into fully formed melanomas in 6 to 8 weeks. Dox withdrawal causes BrafV600E extinction (hereafter “Braf extinction”) and significant tumor regression, consistent with the driving oncogenic role of BrafV600E (Fig. 1A). In nearly all cases (>90%), the regressed tumors do not entirely disappear but stall in a state we describe as MRD (Fig. 1A and B), as the tumors became impalpable. Moreover, we verified that the tissue contained residual tumor cells through IHC staining of GFP, a built-in tumor marker (Supplementary Fig. S3). We determined that MRD was established around 30 days after Braf extinction and could be re-triggered with dox-induced Braf reexpression, which caused a rapid tumor relapse, even after prolonged extinction (Fig. 1A). Overall, this model mimics patients with BrafV600E melanoma who initially respond to BRAFi, but who eventually relapse due to BRAF overexpression, BRAF amplification, or the expression of a nontargetable BRAF splice isoform (collectively, >30% of human patient BRAFi resistance mechanisms; refs. 2, 25).

Figure 1.

A GEMM model of melanoma reveals that MRD is a “cold” immune microenvironment. A, Representative tumor growth curve of iBIP mice treated with dox administration, after Braf extinction, and after Braf reinduction. Data are represented as the average ± SEM (n = 5). The experiment has been performed in > 100 mice in at least 10 independent cohorts. B, Representative pictures of tumor-bearing iBIP mice on dox, after Braf extinction, and after Braf reinduction. C, Top 3 enriched pathways in early Braf extinction versus MRD according to GSEA analysis. D, k-means clustering of microarray data of 45 tumors. Three clusters of interest are reported, and the number of genes in the cluster is specified. Indicated days are after BRAF extinction (dox withdrawal). E, Immune deconvolution analysis (GEDIT) predicting the immune infiltrate composition in iBIP tumors on dox, during early BRAFi (< 30 days), during MRD (> 30 days), and in Braf reexpression-induced relapsing tumors (Rel). Boxes on the x-axis represent single samples. Heat map values represent predicted populations' abundance. F, Tumor growth curves of Braf-extinguished iBIP tumor xenografts in immunocompetent mice (iBIP, n = 6) or immune-deficient mice (NSG, n = 4). The indicated P value is calculated in an unpaired t test between groups 30 days after tumor extinction. A P value < 0.05 was considered significant.

Figure 1.

A GEMM model of melanoma reveals that MRD is a “cold” immune microenvironment. A, Representative tumor growth curve of iBIP mice treated with dox administration, after Braf extinction, and after Braf reinduction. Data are represented as the average ± SEM (n = 5). The experiment has been performed in > 100 mice in at least 10 independent cohorts. B, Representative pictures of tumor-bearing iBIP mice on dox, after Braf extinction, and after Braf reinduction. C, Top 3 enriched pathways in early Braf extinction versus MRD according to GSEA analysis. D, k-means clustering of microarray data of 45 tumors. Three clusters of interest are reported, and the number of genes in the cluster is specified. Indicated days are after BRAF extinction (dox withdrawal). E, Immune deconvolution analysis (GEDIT) predicting the immune infiltrate composition in iBIP tumors on dox, during early BRAFi (< 30 days), during MRD (> 30 days), and in Braf reexpression-induced relapsing tumors (Rel). Boxes on the x-axis represent single samples. Heat map values represent predicted populations' abundance. F, Tumor growth curves of Braf-extinguished iBIP tumor xenografts in immunocompetent mice (iBIP, n = 6) or immune-deficient mice (NSG, n = 4). The indicated P value is calculated in an unpaired t test between groups 30 days after tumor extinction. A P value < 0.05 was considered significant.

Close modal

Transcriptomic time-course analysis of MRD establishment

To investigate the molecular mechanisms underlying the establishment of MRD, we performed a time-course transcriptional analysis on iBIP tumors after Braf extinction. We collected 45 tumors from as early as 8 hours to as late as 160 days after dox withdrawal.

Using RNA microarrays, we analyzed differentially regulated genes before and after MRD establishment (30 days after Braf extinction) using the STEM (16) algorithm and detected multiple trends of interest (Fig. 1C and Supplementary Table S1). The top-scoring gene sets downregulated in MRD were ERK and proliferation pathways, reflecting the facts that MRD is quiescent and that Braf extinction impairs the MAPK pathway. We did not identify pathways significantly upregulated in MRD that would obviously contribute to cell survival or to immune suppression/exhaustion (e.g., PD-1, PD-L1, CTLA-4); instead, the top pathways related to neural and skeletomuscular genes, possibly reflecting the larger relative contribution of normal neurons and muscle cells in the small MRD mass. However, we observed a strong enrichment for highly upregulated immune signature gene sets as early as 8 hours after Braf extinction, until 9 days when it then steeply declined to baseline upon MRD establishment (at ∼30 days; Fig. 1C). To dissect the involved immune pathways more finely, we performed GSEA, which showed that the top 25 enriched pathways in the early phase of Braf extinction (before ∼30 days) were related to antigen presentation, lymphocyte homing, and activation (Fig. 1D). Consistent with these data, immune cells deconvolution analysis (GEDIT) predicted an early influx of CD3+ and CD8+ T cells and myeloid antigen-presenting cells, which faded away when MRD was established (Fig. 1E).

To determine whether the immune response to Braf extinction is critical to the antitumoral effect, we inoculated iBIP-derived tumor cells into either iBIP (immunocompetent) or NSG (severely immunodeficient) mice. When dox was withdrawn, NSG mice showed impaired tumor shrinkage and increased spontaneous relapse, consistent with the hypothesis that MRD relapse is at least in part regulated by the immune microenvironment (Fig. 1F).

Cxcl9 is upregulated upon BRAF extinction/inhibition and mirrors the overall immune infiltrate dynamics

To identify central regulators of the immune response to BRAFi, we adapted our published TRAP network analysis algorithm (9), designed to identify genes, in this case cytokines, that are most central to a set of biological processes by querying an extensive compendium of mouse expression datasets. Chemokines are a large category of molecules involved in the chemoattraction, activation, and differentiation of immune cells; thus, we hypothesized that they are central regulators of the immune response to BRAFi. Using the k-means clustering result (Fig. 1D), we identified 68 significantly enriched gene sets in the “immune activation” cluster (Supplementary Table S2). We queried these 68 gene sets in the TRAP network to establish a subnetwork of cytokine-to-immune pathway relationships (Fig. 2A). The output of the TRAP analysis on the subnetwork is a weighted degree centrality score for each cytokine. We reasoned that cytokines that were initially highly upregulated upon BRAF extinction, then highly downregulated in MRD, and with a high degree centrality score, were high-confidence candidate immune regulators. We, therefore, plotted these elements together: high centrality score (Fig. 2B, circle size), significant downregulation from early BRAF extinction to MRD (Fig. 2B, negative log2 fold-change and high statistical significance), and high upregulation from control to early BRAF extinction (Fig. 2B, circle color; Supplementary Table S3). Two of the top candidates, Cxcl9 and Cxcl10, are highly related paralogs whose protein products bind Cxcr3, are IFNγ response genes, and are well described as important for Cxcr3+ T-cell attraction and/or activation, including in melanoma (26). As shown in Fig. 2C, the expression pattern of Cxcl9 and Cxcl10 closely mirrored the overall immune score trend. Upon BRAF reexpression-induced tumor relapse, the immune score and the expression of Cxcr3 ligands remained low (Fig. 2C, “Relapse”). We note that a third paralog of the same family, Cxcl11, showed similar parameters and trends to Cxcl9 and Cxcl10, except for a lower network centrality.

Figure 2.

Cxcr3 ligands are key players in the immune response to BRAF extinction. A, Visual representation of cytokine-to-pathway connections of the TRAP network. Cytokines (ovals) and relevant associated pathways (rectangles) are shown. Color coding for chemokines represents log2 fold change in MRD versus Early Braf extinction. B, Volcano plot representing the log2 fold change from early Braf extinction to MRD of the indicated chemokines (x-axis), and the corresponding log2P value (y-axis). The size of the circles represents centrality calculated by TRAP, and the color of the circle represents the log2 fold change from control to early Braf extinction. Only the 102 highest-centrality genes are shown for clarity. C, Cxcl9, Cxcl10, and Cxcl11 expression over time compared with the overall immune signature. Data are represented as the average ± SEM.

Figure 2.

Cxcr3 ligands are key players in the immune response to BRAF extinction. A, Visual representation of cytokine-to-pathway connections of the TRAP network. Cytokines (ovals) and relevant associated pathways (rectangles) are shown. Color coding for chemokines represents log2 fold change in MRD versus Early Braf extinction. B, Volcano plot representing the log2 fold change from early Braf extinction to MRD of the indicated chemokines (x-axis), and the corresponding log2P value (y-axis). The size of the circles represents centrality calculated by TRAP, and the color of the circle represents the log2 fold change from control to early Braf extinction. Only the 102 highest-centrality genes are shown for clarity. C, Cxcl9, Cxcl10, and Cxcl11 expression over time compared with the overall immune signature. Data are represented as the average ± SEM.

Close modal

We then selected Cxcl9 as a representative Cxcr3 ligand for initial in vivo validation, as it had the largest and most significant fold-change of the 3 top candidates (Fig. 2B). We first validated Cxcl9 protein expression in iBIP tumors using IHC, confirming an increase upon Braf extinction and a subsequent decrease over time (Fig. 3A and B). We also confirmed concurrent CD8+ and CD4+ T-cell infiltration and abatement, with CD8+ T-cell dynamics lagging slightly after Cxcl9, consistent with a potential regulatory association (Fig. 3B).

Figure 3.

CXCL9 is induced by genetic and pharmacologic BRAF inhibition in mouse and human tumors. A, Representative immunofluorescences (IF) for Cxcl9 protein in iBIP sections, before and after dox withdrawal. DAPI nuclear staining and hematoxylin staining are shown. Bars represent 100 μm. These images are representative of the data shown in panel B. B, IF time-course quantification of Cxcl9, Cd4, and Cd8 positive cells in iBIP sections after dox withdrawal (n = 5, per time point). The y-axis is the % of positive cells over total cells counted in the section. Data are represented as the average ± SEM. C, IF quantification of Cxcl9- and Cd8-positive cells in iBIP tumor sections after pharmacologic BRAFi [PLX4720, 417 parts per million (ppm), 1 week of treatment, n = 3]. Data are represented as the average ± SEM. D, CXCL9 and E) CD8 mRNA levels in melanoma patients pretreatment (Pre-Tx), on BRAFi+MEKi, and resistant (Res) to the treatment, as measured by RNA-seq (9). Each point represents a patient sample (Pre-Tx n = 15, On BRAFi+MEKi n = 13, Res n = 12). RPKM, Reads per Kilobase Million. * represents a P value < 0.05 calculated in an unpaired t test compared with control. Whisker plots represent the minimum, the maximum, the sample median, and the first and third quartiles.

Figure 3.

CXCL9 is induced by genetic and pharmacologic BRAF inhibition in mouse and human tumors. A, Representative immunofluorescences (IF) for Cxcl9 protein in iBIP sections, before and after dox withdrawal. DAPI nuclear staining and hematoxylin staining are shown. Bars represent 100 μm. These images are representative of the data shown in panel B. B, IF time-course quantification of Cxcl9, Cd4, and Cd8 positive cells in iBIP sections after dox withdrawal (n = 5, per time point). The y-axis is the % of positive cells over total cells counted in the section. Data are represented as the average ± SEM. C, IF quantification of Cxcl9- and Cd8-positive cells in iBIP tumor sections after pharmacologic BRAFi [PLX4720, 417 parts per million (ppm), 1 week of treatment, n = 3]. Data are represented as the average ± SEM. D, CXCL9 and E) CD8 mRNA levels in melanoma patients pretreatment (Pre-Tx), on BRAFi+MEKi, and resistant (Res) to the treatment, as measured by RNA-seq (9). Each point represents a patient sample (Pre-Tx n = 15, On BRAFi+MEKi n = 13, Res n = 12). RPKM, Reads per Kilobase Million. * represents a P value < 0.05 calculated in an unpaired t test compared with control. Whisker plots represent the minimum, the maximum, the sample median, and the first and third quartiles.

Close modal

We next asked whether Cxcl9 is also induced by pharmacologic BRAFi. In both the GEMM iBIP model (Fig. 3C) and in previously published data from the syngeneic SMM102 model (refs. 12, 24, 27; Supplementary Fig. S4), Cxcl9 and CD8 markers increased in the tumors in response to the BRAFi PLX4720, consistent with our genetic BRAF extinction results. To confirm clinical relevance, we analyzed our previously published human patient sample RNA-seq dataset (9), which includes pretreatment, on-treatment, and BRAFi-resistant biopsies. Similar to our mouse models, both CXCL9 and CD8A expression sharply rose on BRAFi treatment and abated upon the acquisition of BRAFi resistance (Fig. 3D and E).

rCxcl9 administration chemoattracts CD8+ lymphocytes in vitro and in vivo

We next reasoned that one or all three of these Cxcr3 ligands might regulate T-cell tumor infiltration and/or activation in response to BRAF inhibition. First, we verified the ability of rCxcl9 to attract CD8+ mouse lymphocytes in vitro using a transwell assay. As shown in Fig. 4A, Cxcl9 significantly induced CD8+ T-cell chemoattraction in a dose-dependent manner, and similarly to its paralogs Cxcl10 and Cxcl11 (Fig. 4A). Moreover, Cxcl9 and Cxcl11, but not Cxcl10, also slightly induced activation markers in CD8+ T cells (CD69, early and CD25, late activation marker; Supplementary Fig. S5).

Figure 4.

rCxcl9 attracts CD8+ T cells in vitro and in vivo and delays the occurrence of tumor relapse. A, Transwell migration assay of mouse CD8+ T cells using chemokines at the indicated concentrations. The migration index is calculated as the number of cells migrated over the total number of cells. * represents a P value < 0.05 calculated in an unpaired t test compared with control. Data are represented as the average ± SEM and were obtained in three independent experiments. B, Flow cytometric analysis of CD8+Cxcr3+ and CD8+Cxcr3 T-cell abundance in “BP” tumors after intratumoral injection of rCxcl9 (10 μg/administration, for 3 days; n = 3) or vehicle. The percentage of CD8+Cxcr3+ (or CD8+Cxcr3) cells is calculated over the total number of cells in each sample. * represents a P value < 0.05 calculated in an unpaired t test compared with control. Data are represented as the average ± SEM. C, Survival curves of iBIP allografts injected with rCxcl9 or vehicle (Ctrl, n = 5, rCxcl9, n = 7) after dox withdrawal. At MRD establishment (21 days), tumors were rechallenged with dox and observed for relapse. A death event is considered as the first tumor doubling after MRD acquisition. The P values indicated on the Kaplan–Meier curves are calculated through a Log-rank (Mantel–Cox) test. D, Transwell migration assay for CD8+ T cells using BSA- or rCxcl9-loaded PLGA microparticles. The migration index is calculated as the number of cells migrated over the total number of cells. *, P < 0.05, calculated in an unpaired t test compared with control. Data are represented as the average ± SEM and were obtained in three independent experiments. E, Survival curves of Yumm1.7 tumors treated with PLX4720 with or without microparticles that contained BSA or rCxcl9 (n = 10). A death event is considered as the first tumor doubling after MRD acquisition. The vehicle control group with no BRAFi needed to be euthanized before BRAFi and, therefore, cannot be represented on this graph. The P values indicated on the Kaplan–Meier curves are calculated through a Log-rank (Mantel–Cox) test. F, Complete pathologic responses (pCR) versus live tumor tissue detected by a certified pathologist in the indicated groups. G, Immunofluorescence quantification of CD8+Cxcr3+ T-cell abundance in Yumm1.7 tumors treated with PLX4720 with or without microparticles that contained BSA or rCxcl9 (n = 5). We did not perform molecular analysis for the BRAFi+MEKi+mrCxcl9 group because not enough residual tissue was available (4/5 pCR). The percentage of CD8+Cxcr3+ cells is calculated over the total number of cells in the tissues. *, P < 0.05, calculated in an unpaired t test compared with control. Data are represented as the average ± SEM. H, Schematic of the rational treatment approach proposed in the present work. Created with BioRender.com.

Figure 4.

rCxcl9 attracts CD8+ T cells in vitro and in vivo and delays the occurrence of tumor relapse. A, Transwell migration assay of mouse CD8+ T cells using chemokines at the indicated concentrations. The migration index is calculated as the number of cells migrated over the total number of cells. * represents a P value < 0.05 calculated in an unpaired t test compared with control. Data are represented as the average ± SEM and were obtained in three independent experiments. B, Flow cytometric analysis of CD8+Cxcr3+ and CD8+Cxcr3 T-cell abundance in “BP” tumors after intratumoral injection of rCxcl9 (10 μg/administration, for 3 days; n = 3) or vehicle. The percentage of CD8+Cxcr3+ (or CD8+Cxcr3) cells is calculated over the total number of cells in each sample. * represents a P value < 0.05 calculated in an unpaired t test compared with control. Data are represented as the average ± SEM. C, Survival curves of iBIP allografts injected with rCxcl9 or vehicle (Ctrl, n = 5, rCxcl9, n = 7) after dox withdrawal. At MRD establishment (21 days), tumors were rechallenged with dox and observed for relapse. A death event is considered as the first tumor doubling after MRD acquisition. The P values indicated on the Kaplan–Meier curves are calculated through a Log-rank (Mantel–Cox) test. D, Transwell migration assay for CD8+ T cells using BSA- or rCxcl9-loaded PLGA microparticles. The migration index is calculated as the number of cells migrated over the total number of cells. *, P < 0.05, calculated in an unpaired t test compared with control. Data are represented as the average ± SEM and were obtained in three independent experiments. E, Survival curves of Yumm1.7 tumors treated with PLX4720 with or without microparticles that contained BSA or rCxcl9 (n = 10). A death event is considered as the first tumor doubling after MRD acquisition. The vehicle control group with no BRAFi needed to be euthanized before BRAFi and, therefore, cannot be represented on this graph. The P values indicated on the Kaplan–Meier curves are calculated through a Log-rank (Mantel–Cox) test. F, Complete pathologic responses (pCR) versus live tumor tissue detected by a certified pathologist in the indicated groups. G, Immunofluorescence quantification of CD8+Cxcr3+ T-cell abundance in Yumm1.7 tumors treated with PLX4720 with or without microparticles that contained BSA or rCxcl9 (n = 5). We did not perform molecular analysis for the BRAFi+MEKi+mrCxcl9 group because not enough residual tissue was available (4/5 pCR). The percentage of CD8+Cxcr3+ cells is calculated over the total number of cells in the tissues. *, P < 0.05, calculated in an unpaired t test compared with control. Data are represented as the average ± SEM. H, Schematic of the rational treatment approach proposed in the present work. Created with BioRender.com.

Close modal

We then tested Cxcl9 chemoattraction in vivo, using the previously published syngeneic “BP” melanoma model (12). We injected rCxcl9 (10 μg/administration) intratumorally for 3 days and then analyzed the abundance of the infiltrating CD8+ T-cell population. We observed that rCxcl9 specifically recruited CD8+Cxcr3+ cells into the tumors, consistent with Cxcr3 as the known binding receptor of Cxcl9 (Fig. 4B). To validate the therapeutic potential of our findings, we tested the effect of rCxcl9 administration in iBIP allografts. We injected rCxcl9 intratumorally (0.5–1 μg/administration, every day) for 3 weeks after dox withdrawal. After MRD establishment, we re-administered dox and monitored tumors for regrowth. We found that the control group relapsed significantly earlier and in a higher percentage of mice than the rCxcl9-treated group over the observation period (Fig. 4C; Supplementary Fig. S6).

A microparticle-based approach for chemokine delivery

A potential hurdle to translating Cxcl9 into therapy is that chemokines are rapidly degraded in vivo. Nanoparticles for drug-delivery applications have been developed to overcome some of the limitations of free therapeutics, including drug stability and release kinetics (28, 29). Toward this aim, we leveraged an approach in which rCxcl9 was encapsulated in microparticles composed of a silica core and a PLGA outer shell (Supplementary Fig. S1; ref. 29). PLGA-silica particles assure chemokines protection from degradation while providing a steady release and supply.

In vitro experiments confirmed that PLGA microparticles released rCxcl9 efficiently and attracted mouse CD8+ T cells (Fig. 4D) to levels similar to that of naked rCxcl9. Next, we used the well-established Yumm1.7 syngeneic melanoma model (11) to further confirm our initial findings in the iBIP model. Yumm1.7 tumors have the same clinically relevant genetic alterations as iBIP tumors (Pten−/-; Cdkn2a−/−; BrafV600E), sharply respond to pharmacologic BRAFi, and establish MRD with a natural eventual tumor relapse in syngeneic immunocompetent C57BL6/J mice hosts.

To determine whether rCxcl9-loaded PLGA microparticles (hereafter “mrCxcl9”) affect melanoma relapse, we induced Yumm1.7 tumors and divided mice into 4 groups (n = 10 per cohort): Control Vehicle (Ctrl), BRAFi, BRAFi + PLGA microparticles loaded with BSA (BRAFi+mBSA), and BRAFi + PLGA microparticles loaded with rCxcl9 (BRAFi+rCxcl9). Between the PLGA tested concentrations (10% or 5% PLGA 50:50), 10% was chosen for translation in vivo since it induced the highest migration of CD8+ T cells in vitro (Fig. 4D). Starting 3 days after BRAFi, the two PLGA groups were injected intratumorally with microparticles for 3 weeks, once a week, at 50 μg of total rCxcl9 or BSA per administration. All groups were continuously maintained on BRAFi and monitored for tumor relapse, which we defined as the first tumor size doubling after MRD establishment. The vehicle control group with no BRAFi needed to be euthanized before 21 days after tumor cell injection. By contrast, BRAFi and BRAFi+mBSA groups relapsed after a median of 29 and 27.5 days after BRAFi start, respectively (Fig. 4E; Supplementary Fig S7), indicating no effect of the microparticles themselves. The BRAFi+mrCxcl9 group showed a significantly delayed relapse (median 43.5 days, Fig. 4E; Supplementary Fig S7), supporting a positive therapeutic effect of mrCxcl9 and the feasibility of microparticle delivery.

As BRAFi+MEKi is the standard of care for BRAFV600E melanomas, we also tested mrCxcl9 with this inhibitor combination (PLX4720+MEK162) in the Yumm1.7 model. The administration of mrCxcl9 with BRAFi+MEKi induced a complete pathologic response (pCR) in 4/5 tumors (Fig. 4F; Supplementary Fig. S8) and widespread necrosis in the remaining tumor after only 10 days of treatment (7 days of mrCxcl9). By contrast, BRAFi+MEKi alone, or BRAFi+mrCxcl9, while potent, left detectable tumor cells in all mice. As insufficient tumor tissue remained in BRAFi+MEKi+mrCxcl9 treated mice for IHC assessment, we next took advantage of BRAFi-treated mice (without MEKi) at 10 days of treatment to assess CD8 and Cxcr3 by multiplex immunofluorescence. Consistent with enhancing immune infiltration after BRAFi, mrCxcl9 administration in combination with BRAFi caused a significant further increase of both CD8+Cxcr3+ and overall CD8+ T cells (Fig. 4G; Supplementary Figs. S9 and S10).

In this study, we have demonstrated that Cxcl9 is a key immune modulator of the BRAFi-induced MRD state, which in turn manifests as a combination therapeutic modality to forestall tumor relapse (Fig. 4H). We leveraged the protection of PLGA-silica microparticles for rCxcl9 delivery, enabling slow intratumoral release over time, which both reduces the frequency of administration and provides a consistent output. This discovery spurred from the transcriptional and phenotypic characterization of the iBIP GEMM model of melanoma MRD. We used our previously published TRAP algorithm to prioritize key chemokines potentially regulating MRD establishment and maintenance, which was then validated in pharmacologic, immunocompetent BRAF inhibition models.

As we did not detect an enrichment for specific immunosuppressive/evasive signals to target in the MRD, such as PD-1 or CTLA4, we hypothesized instead that sustaining T-cell recruiting signaling throughout BRAFi administration could be a viable strategy to convert MRD to be immunologically “hot.” Indeed, intratumoral rCxcl9 in combination with BRAFi not only increased the influx of Cxcr3+CD8+ T cells into the tumor but also overall CD8+ T cells, suggesting a positive immunologic feedback loop leading to enhanced BRAFi and BRAFi+MEKi efficacy.

While we selected Cxcl9 as our proof-of-principle chemokine, we also identified other chemokines whose temporal expression profiles suggest a similar immune regulatory role; specifically, our evidence suggests their downregulation may help create the immunologically “cold” state of MRD with a low abundance of CD8+ T cells. Of particular interest are the CXCL9 paralogs CXCL10 and CXCL11, all three of which are IFNγ-induced chemokines known to regulate T-cell homing through their CXCR3 cognate receptor (26). Despite a high degree of amino acid identity, Cxcl9, Cxcl10, and Cxcl11 are known to have different roles in inflammation, GVHD, and cancer (30). Consistent with this, we found differing effects of the 3 proteins on CD8+ T-cell phenotypes in vitro, with all three showing chemoattraction but Cxcl10 showing less T-cell activation. Thus, we hypothesize that future testing of combinations of the 3 CXCR3 ligands and/or other high-scoring cytokines may show improved antitumor efficacy. Because of their versatile properties, the PLGA-silica microparticle approach we used is ideally suited for these future directions. Silica nanostructures can be tailored during manufacturing, changing their size, shape, porosity, and pore size to efficiently load a wide range of small and large biomolecules. Also, they can be incorporated into a wide range of synthetic polymers to finely tune the release of cytokines (31). In addition, PLGA-silica microparticle tolerability in vivo has been reported extensively in different model systems, such as cancer and inflammatory diseases (31). Moreover, the polymer used in this work, PLGA, is approved by the FDA for use in drug-delivery systems, positioning PLGA-based approaches as promising candidates for future clinical trials.

Our results showing significantly prolonged treatment efficacy after mrCxcl9 therapy are consistent with the antitumor effect of Cxcr3 ligands seen in previous studies (32–35). However, unlike the previous efforts, which focused on rapidly growing tumors, our study recontextualizes the efficacy of Cxcr3 ligands in preventing or neutralizing drug-induced MRD, expanding their utility towards this critical clinical obstacle. Specifically, this work validates Cxcl9 as a viable in vivo therapy for melanoma in an MRD setting and demonstrates the utility of a microparticle-based delivery.

Furthermore, the combination of the standard-of-care BRAFi+MEKi therapy with mrCxcl9 induced complete pathologic responses in 4 of 5 animals, suggesting—even if not necessarily cures—a very robust synergy. Therefore, we propose that CXCR3 ligands are clinically relevant to targeted therapies, particularly as CXCL9 expression is strongly induced in patients with melanoma treated with BRAFi but returns to baseline in patients showing relapse, mirroring CD8 expression dynamics and the observations in our mouse models. Consistent with this, multiple studies have found that CXCR3 ligands are critical for the action of anti–PD-1/PD-L1 immune checkpoint therapies (36, 37).

Our findings demonstrate that chemokine-based approaches are a feasible strategy to regulate the composition of the immune infiltrate of tumors, particularly if coupled with an efficient delivery system. Immunologically “cold” tumors are, in fact, a widespread clinical problem and especially relevant to current checkpoint inhibitor therapies (4, 10). Indeed, as the combination of targeted and immune approaches is experiencing the beginnings of clinical validation (38, 39), there is evidence showing that baseline immune “cold” tumors can remain refractory (10, 40). Chemokine-based approaches may be ideally suited to solving such issues by converting tumors to a “hot” immune microenvironment for subsequent targeted and immune checkpoint therapies to create synergy. As several studies suggest that BRAFi+MEKi administration may negatively influence subsequent immunotherapy outcomes (41, 42), further mechanistic assessment of how MRD becomes immunologically “cold” may therefore shed further light on how this might longitudinally impact the immune microenvironment. Overall, our results demonstrate that fine-tuning the composition of the immune infiltrate can be a viable adjuvant approach to boosting existing and experimental treatment approaches and eventually improving their therapeutic outcome.

J.A. Wargo reports personal fees from Imedex, Dava Oncology, Omniprex, Illumnia, Gilead, PeerView, Physician Education Resource, MedImmune, Exelixis, Bristol Myers Squibb, Micronoma; and personal fees from OSE Therapeutics outside the submitted work; in addition, J.A. Wargo has a patent for PCT/US17/53.717 issued. J.C. Costello reports being a cofounder of PrecisionProfile and OncoRX Insights. No disclosures were reported by the other authors.

G. Romano: Conceptualization, resources, data curation, formal analysis, supervision, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing. F. Paradiso: Investigation, methodology. P. Li: Formal analysis, validation, investigation. P. Shukla: Formal analysis. L.N. Barger: Data curation, formal analysis, validation, investigation. O. El Naggar: Formal analysis. J.P. Miller: Data curation, formal analysis. R.J. Liang: Validation, investigation. T.L. Helms: Formal analysis. A.J. Lazar: Formal analysis. J.A. Wargo: Resources, formal analysis. F. Taraballi: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–review and editing. J.C. Costello: Conceptualization, resources, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–review and editing. L.N. Kwong: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, project administration, writing–review and editing.

L.N. Kwong is supported by 1R01CA251608–01, 1R01HG011356–01, Melanoma Research Alliance 508743; J.C. Costello is supported by U01CA231978; L.N. Kwong and F. Taraballi are supported by Melanoma Research Foundation 717906.

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

1.
Flaherty
KT
,
Infante
JR
,
Daud
A
,
Gonzalez
R
,
Kefford
RF
,
Sosman
J
, et al
.
Combined BRAF and MEK inhibition in melanoma with BRAF V600 mutations
.
N Engl J Med
2012
;
367
:
1694
703
.
2.
Rizos
H
,
Menzies
AM
,
Pupo
GM
,
Carlino
MS
,
Fung
C
,
Hyman
J
, et al
.
BRAF inhibitor resistance mechanisms in metastatic melanoma: spectrum and clinical impact
.
Clin Cancer Res
2014
;
20
:
1965
77
.
3.
Bivona
TG
,
Doebele
RC
.
A framework for understanding and targeting residual disease in oncogene-driven solid cancers
.
Nat Med
2016
;
22
:
472
8
.
4.
Galon
J
,
Bruni
D
.
Approaches to treat immune hot, altered and cold tumors with combination immunotherapies
.
Nat Rev Drug Discov
2019
;
18
:
197
218
.
5.
Frederick
DT
,
Piris
A
,
Cogdill
AP
,
Cooper
ZA
,
Lezcano
C
,
Ferrone
CR
, et al
.
BRAF inhibition is associated with enhanced melanoma antigen expression and a more favorable tumor microenvironment in patients with metastatic melanoma
.
Clin Cancer Res
2013
;
19
:
1225
31
.
6.
Cooper
ZA
,
Reuben
A
,
Spencer
CN
,
Prieto
PA
,
Austin-Breneman
JL
,
Jiang
H
, et al
.
Distinct clinical patterns and immune infiltrates are observed at time of progression on targeted therapy versus immune checkpoint blockade for melanoma
.
Oncoimmunology
2016
;
5
:
e1136044
.
7.
Massi
D
,
Brusa
D
,
Merelli
B
,
Falcone
C
,
Xue
G
,
Carobbio
A
, et al
.
The status of PD-L1 and tumor-infiltrating immune cells predict resistance and poor prognosis in BRAFi-treated melanoma patients harboring mutant BRAFV600
.
Ann Oncol
2015
;
26
:
1980
7
.
8.
Yan
Y
,
Wongchenko
MJ
,
Robert
C
,
Larkin
J
,
Ascierto
PA
,
Dreno
B
, et al
.
Genomic features of exceptional response in vemurafenib ± cobimetinib-treated patients with BRAF (V600)-mutated metastatic melanoma
.
Clin Cancer Res
2019
;
25
:
3239
46
.
9.
Kwong
LN
,
Boland
GM
,
Frederick
DT
,
Helms
TL
,
Akid
AT
,
Miller
JP
, et al
.
Co-clinical assessment identifies patterns of BRAF inhibitor resistance in melanoma
.
J Clin Invest
2015
;
125
:
1459
70
.
10.
Bonaventura
P
,
Shekarian
T
,
Alcazer
V
,
Valladeau-Guilemond
J
,
Valsesia-Wittmann
S
,
Amigorena
S
, et al
.
Cold tumors: a therapeutic challenge for immunotherapy
. Front Immunol
2019
;
10
:
168
.
11.
Meeth
K
,
Wang
JX
,
Micevic
G
,
Damsky
W
,
Bosenberg
MW
.
The YUMM lines: a series of congenic mouse melanoma cell lines with defined genetic alterations
.
Pigment Cell Melanoma Res
2016
;
29
:
590
7
.
12.
Dankort
D
,
Curley
DP
,
Cartlidge
RA
,
Nelson
B
,
Karnezis
AN
,
Damsky
WE
Jr
, et al
.
Braf(V600E) cooperates with Pten loss to induce metastatic melanoma
.
Nat Genet
2009
;
41
:
544
52
.
13.
Gautier
L
,
Cope
L
,
Bolstad
BM
,
Irizarry
RA
.
affy–analysis of Affymetrix GeneChip data at the probe level
.
Bioinformatics
2004
;
20
:
307
15
.
14.
Irizarry
RA
,
Bolstad
BM
,
Collin
F
,
Cope
LM
,
Hobbs
B
,
Speed
TP
.
Summaries of Affymetrix GeneChip probe level data
.
Nucleic Acids Res
2003
;
31
:
e15
.
15.
Ritchie
ME
,
Phipson
B
,
Wu
D
,
Hu
Y
,
Law
CW
,
Shi
W
, et al
.
limma powers differential expression analyses for RNA-sequencing and microarray studies
.
Nucleic Acids Res
2015
;
43
:
e47
.
16.
Ernst
J
,
Bar-Joseph
Z
.
STEM: a tool for the analysis of short time series gene expression data
.
BMC Bioinf
2006
;
7
:
191
.
17.
Ernst
J
,
Nau
GJ
,
Bar-Joseph
Z
.
Clustering short time series gene expression data
.
Bioinformatics
2005
;
21
Suppl 1
:
i159
68
.
18.
Mootha
VK
,
Lindgren
CM
,
Eriksson
KF
,
Subramanian
A
,
Sihag
S
,
Lehar
J
, et al
.
PGC-1alpha-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes
.
Nat Genet
2003
;
34
:
267
73
.
19.
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 USA
2005
;
102
:
15545
50
.
20.
Nadel
BB
,
Lopez
D
,
Montoya
DJ
,
Ma
F
,
Waddel
H
,
Khan
MM
, et al
.
The Gene Expression Deconvolution Interactive Tool (GEDIT): accurate cell type quantification from gene expression data
.
Gigascience
2021
;
10
:
giab002
.
21.
Lattin
JE
,
Schroder
K
,
Su
AI
,
Walker
JR
,
Zhang
J
,
Wiltshire
T
, et al
.
Expression analysis of G protein-coupled receptors in mouse macrophages
.
Immunome Res
2008
;
4
:
5
.
22.
Kwong
LN
,
Costello
JC
,
Liu
H
,
Jiang
S
,
Helms
TL
,
Langsdorf
AE
, et al
.
Oncogenic NRAS signaling differentially regulates survival and proliferation in melanoma
.
Nat Med
2012
;
18
:
1503
10
.
23.
Liberzon
A
,
Subramanian
A
,
Pinchback
R
,
Thorvaldsdottir
H
,
Tamayo
P
,
Mesirov
JP
.
Molecular signatures database (MSigDB) 3.0
.
Bioinformatics
2011
;
27
:
1739
40
.
24.
Xu
G
,
Luo
Y
,
Wu
W
,
Liu
X
,
Yu
X
,
Bao
Y
, et al
.
The evolution of acquired resistance to BRAFV600E kinase inhibitor is sustained by IGF1-driven tumor vascular remodeling
.
J Invest Dermatol
2022
;
142
:
445
58
.
25.
Van Allen
EM
,
Wagle
N
,
Sucker
A
,
Treacy
DJ
,
Johannessen
CM
,
Goetz
EM
, et al
.
The genetic landscape of clinical resistance to RAF inhibition in metastatic melanoma
.
Cancer Discov
2014
;
4
:
94
109
.
26.
Tokunaga
R
,
Zhang
W
,
Naseem
M
,
Puccini
A
,
Berger
MD
,
Soni
S
, et al
.
CXCL9, CXCL10, CXCL11/CXCR3 axis for immune activation: a target for novel cancer therapy
.
Cancer Treat Rev
2018
;
63
:
40
7
.
27.
Yu
J
,
Wu
X
,
Song
J
,
Zhao
Y
,
Li
H
,
Luo
M
, et al
.
Loss of MHC-I antigen presentation correlated with immune checkpoint blockade tolerance in MAPK inhibitor-resistant melanoma
.
Front Pharmacol
2022
;
13
:
928226
.
28.
Mitchell
MJ
,
Billingsley
MM
,
Haley
RM
,
Wechsler
ME
,
Peppas
NA
,
Langer
R
.
Engineering precision nanoparticles for drug delivery
.
Nat Rev Drug Discov
2021
;
20
:
101
24
.
29.
Minardi
S
,
Pandolfi
L
,
Taraballi
F
,
De Rosa
E
,
Yazdi
IK
,
Liu
X
, et al
.
PLGA-mesoporous silicon microspheres for the in vivo controlled temporospatial delivery of proteins
.
ACS Appl Mater Interfaces
2015
;
7
:
16364
73
.
30.
Groom
JR
,
Luster
AD
.
CXCR3 ligands: redundant, collaborative, and antagonistic functions
.
Immunol Cell Biol
2011
;
89
:
207
15
.
31.
Danhier
F
,
Ansorena
E
,
Silva
JM
,
Coco
R
,
Le Breton
A
,
Preat
V
.
PLGA-based nanoparticles: an overview of biomedical applications
.
J Control Release
2012
;
161
:
505
22
.
32.
Chow
MT
,
Ozga
AJ
,
Servis
RL
,
Frederick
DT
,
Lo
JA
,
Fisher
DE
, et al
.
Intratumoral activity of the CXCR3 chemokine system is required for the efficacy of anti–PD-1 therapy
.
Immunity
2019
;
50
:
1498
512
.
33.
Pan
J
,
Burdick
MD
,
Belperio
JA
,
Xue
YY
,
Gerard
C
,
Sharma
S
, et al
.
CXCR3/CXCR3 ligand biological axis impairs RENCA tumor growth by a mechanism of immunoangiostasis
.
J Immunol
2006
;
176
:
1456
64
.
34.
Ohtani
H
,
Jin
Z
,
Takegawa
S
,
Nakayama
T
,
Yoshie
O
.
Abundant expression of CXCL9 (MIG) by stromal cells that include dendritic cells and accumulation of CXCR3+ T cells in lymphocyte-rich gastric carcinoma
.
J Pathol
2009
;
217
:
21
31
.
35.
Bellmann
L
,
Cappellano
G
,
Schachtl-Riess
JF
,
Prokopi
A
,
Seretis
A
,
Ortner
D
, et al
.
A TLR7 agonist strengthens T- and NK-cell function during BRAF-targeted therapy in a preclinical melanoma model
.
Int J Cancer
2020
;
146
:
1409
20
.
36.
Reschke
R
,
Yu
J
,
Flood
B
,
Higgs
EF
,
Hatogai
K
,
Gajewski
TF
.
Immune cell and tumor cell-derived CXCL10 is indicative of immunotherapy response in metastatic melanoma
.
J Immunother Cancer
2021
;
9
:
e003521
.
37.
House
IG
,
Savas
P
,
Lai
J
,
Chen
AXY
,
Oliver
AJ
,
Teo
ZL
, et al
.
Macrophage-derived CXCL9 and CXCL10 are required for antitumor immune responses following immune checkpoint blockade
.
Clin Cancer Res
2020
;
26
:
487
504
.
38.
Dummer
R
,
Lebbe
C
,
Atkinson
V
,
Mandala
M
,
Nathan
PD
,
Arance
A
, et al
.
Combined PD-1, BRAF and MEK inhibition in advanced BRAF-mutant melanoma: safety run-in and biomarker cohorts of COMBI-i
.
Nat Med
2020
;
26
:
1557
63
.
39.
Ribas
A
,
Lawrence
D
,
Atkinson
V
,
Agarwal
S
,
Miller
WH
Jr
,
Carlino
MS
, et al
.
Combined BRAF and MEK inhibition with PD-1 blockade immunotherapy in BRAF-mutant melanoma
.
Nat Med
2019
;
25
:
936
40
.
40.
Herbst
RS
,
Soria
JC
,
Kowanetz
M
,
Fine
GD
,
Hamid
O
,
Gordon
MS
, et al
.
Predictive correlates of response to the anti–PD-L1 antibody MPDL3280A in cancer patients
.
Nature
2014
;
515
:
563
7
.
41.
Simeone
E
,
Grimaldi
AM
,
Festino
L
,
Giannarelli
D
,
Vanella
V
,
Palla
M
, et al
.
Correlation between previous treatment with BRAF inhibitors and clinical response to pembrolizumab in patients with advanced melanoma
.
Oncoimmunology
2017
;
6
:
e1283462
.
42.
Atkins
MB
,
Lee
SJ
,
Chmielowski
B
,
Tarhini
AA
,
Cohen
GI
,
Truong
T-G
, et al
.
Combination dabrafenib and trametinib versus combination nivolumab and ipilimumab for patients with advanced BRAF-mutant melanoma: The DREAMseq Trial—ECOG-ACRIN EA6134
.
J Clin Oncol
2023
;
41
:
186
97
.