The central nervous system (CNS) antigen-presenting cell (APC) that primes antitumor CD8+ T-cell responses remains undefined. Elsewhere in the body, the conventional dendritic cell 1 (cDC1) performs this role. However, steady-state brain parenchyma cDC1 are extremely rare; cDCs localize to the choroid plexus and dura. Thus, whether the cDC1 play a function in presenting antigen derived from parenchymal sources in the tumor setting remains unknown. Using preclinical glioblastoma (GBM) models and cDC1-deficient mice, we explored the presently unknown role of cDC1 in CNS antitumor immunity. We determined that, in addition to infiltrating the brain tumor parenchyma itself, cDC1 prime neoantigen-specific CD8+ T cells against brain tumors and mediate checkpoint blockade-induced survival benefit. We observed that cDC, including cDC1, isolated from the tumor, the dura, and the CNS-draining cervical lymph nodes harbored a traceable fluorescent tumor antigen. In patient samples, we observed several APC subsets (including the CD141+ cDC1 equivalent) infiltrating glioblastomas, meningiomas, and dura. In these same APC subsets, we identified a tumor-specific fluorescent metabolite of 5-aminolevulinic acid, which fluorescently labeled tumor cells during fluorescence-guided GBM resection. Together, these data elucidate the specialized behavior of cDC1 and suggest that cDC1 play a significant role in CNS antitumor immunity.
Glioblastoma (GBM) is the most common central nervous system (CNS) primary malignancy and remains inexorably lethal. The standard of care, involving surgery, chemotherapy, and radiation treatment, leads to a median survival of only 15 to 20 months (1). Because immunotherapy has revolutionized treatment of other cancers, there has been significant interest in applying immune-based treatments to patients with GBM, especially given a growing understanding that the CNS is not as immunoprivileged as long perceived (2). However, despite numerous clinical trials, there remain no FDA-approved immunotherapies for GBM (3, 4).
Several critical issues underlie the difficulty of harnessing immune-potentiating therapies to treat GBM. First, GBM represents a severe example of cancer immunoediting in patients (5) and exhibits substantial evidence of immunosuppression. A range of immunologic deficits has been described previously in patients with GBM, including lymphopenia (6), intrinsic immune suppression by tumor cells (7, 8), and overexpression of checkpoint molecules such as PD-L1 (9, 10), among others (11, 12), suggesting a “cold” immune phenotype. Second, because GBM tumors are molecularly heterogeneous and are comprised of several subclonal populations (13), T-cell antigen targets are not uniformly expressed in the tumor mass. Our incomplete understanding of the development of CNS immune responses also remains a significant barrier to successful immunotherapy in the brain. Specifically, the mechanistic underpinnings of antigen presentation leading to T-cell priming of brain tumor–specific T cells remain undefined. Unlike other sites in the body, the brain parenchyma does not harbor dendritic cells (DC) in the resting state (14–16). Microglia and border-associated macrophages instead comprise the putative CNS antigen-presenting cells (APC; ref. 14). Furthermore, the brain does not harbor classical lymphatic and secondary lymphoid structures (14, 15, 17). However, the characterization of lymphatic channels within the dura mater of the cerebral meninges has provided new insights into how APCs may encounter antigen originating from the brain parenchyma and potentially traffic that antigen to draining lymph nodes (18–21). Finally, evidence suggesting that conventional DCs (cDC) are important in mounting autoimmune responses in the brain has been reported, with experiments demonstrating that cDC depletion attenuates experimental autoimmune encephalitis (EAE; ref. 22) and that restricting MHC-II expression to cDC subsets is sufficient to permit EAE progression (23). Although these foundational observations highlight the importance of the anatomic basis for lymphatic drainage in the CNS and suggest that cDC contribute to inflammation in the brain, the specific role cDC play in this process remains unclear. Cell migration from the CNS to cervical lymph nodes (CLN) has been reported (24–27), but these experiments used cultured exogenously introduced monocyte-derived DCs. Thus, there remains a lack of understanding regarding the cellular basis for endogenous antigen presentation in CNS antitumor immunity.
Herein, we characterized the role of the cDC1 subset during the endogenous immune response to brain tumors. Elsewhere in the body, cDC1 phagocytize antigen, migrate to LNs, and cross-present antigen to prime CD8+ T-cell responses (28–32). Although cDC1 are required for effective immune responses against a range of tumor types (28, 29, 31, 33–37), it is unclear whether these cells perform a similarly obligate role in the immune response to CNS tumors. In this study, we demonstrate that cDC1 are required both to mount neoantigen-specific immune responses and to respond to checkpoint blockade in wild-type (WT), syngeneic orthotopic models of GBM. We also observed tumor antigen-containing cDC1 within the tumor, the dura, and CLNs. We determined that chemokine receptor 7 (CCR7) was required for DC to traffic tumor antigen from the CNS to CLNs and that clonal expansion of adoptively transferred tumor antigen-specific CD8+ T cells takes place within these LNs before expanding to other anatomic locations, which required both CCR7 and cDC1 for normal kinetics. In patients, we observed cDC, including the analogous cDC1 subset, infiltrating tumors, as well as the dura. Finally, from patients in whom 5-aminolevulinic acid (5-ALA) was used to facilitate fluorescence-guided GBM resection (38–41), we identified the presence of its fluorescent metabolite protoporphyrin-IX (PPIX; refs. 42, 43) within tumor-infiltrating cDC and monocytes, but not in tumor-infiltrating T cells nor in matched peripheral APC counterparts. Together, these findings clarify the unique behavior of cDC1 in the CNS and demonstrate that cDC1 play a critical role in the CNS antitumor immune responses.
Materials and Methods
The objective of this study was to investigate the nature of antigen presentation and the role of cDC1 in CNS antitumor immunity. We designed and performed experiments in cellular immunology, and used preclinical models of GBM, as well as genotypes of mice deficient in cDC1, cell migration, or with lymphatics and/or cDC1 specifically labeled. We used previously characterized neoantigens and model antigens, as discussed, to study antigen-specific responses and kinetics of clonal expansion of adoptively transferred T cells. We also studied APC infiltration and antigen uptake by tumor-infiltrating APCs in patient tumors.
All study participants were neurosurgical patients at Barnes-Jewish Hospital with grade 4 glioblastomas or meningiomas confirmed by pathology of the resected specimens. All meningiomas were resected without 5-ALA. A subset of GBM tumors were resected using 5-ALA. For patients whose tumors were resected with 5-ALA, 5-ALA (Gleolan) was administered in a dose of 20 mg/kg 2 to 4 hours before the induction of anesthesia. Any combination of tumor, dura, and blood were collected when specified. Experiments were performed in accordance with ethical standards set forth in the 1964 Declaration of Helsinki. Written informed consent for specimen inclusion in this study was obtained using Washington University School of Medicine Institutional Review Board-approved protocols (#20111101 and #202107071).
Human tumor and dura preparation
Human tumors and dura were stored on ice after resection. On the same day, specimens were processed by masserating and incubating in 10 to 20 mL of 2 mg/mL collagenase A (Roche) and 2 mg/mL collagenase D (Roche) in 10% heat-inactivated FBS (Gibco) and IMDM (Gibco) overnight at 37°C, with trituration every few hours for dura disaggregation in particular. The next morning, single-cell suspensions were filtered and prepared for flow cytometry as described below.
All animal experiments were approved by the Washington University Animal Studies Committee. Male and female mice 6 to 16 weeks of age were used for all experiments. WT C57BL/6 mice were purchased from Taconic Biosciences. C57BL/6 IRF8+32kb−/− and SNX22GFP/GFP mice were obtained from Dr. Kenneth Murphy (Washington University in St. Louis). SNX22GFP/+ F1 mice were used for experiments, with cDC1-GFP expression confirmed by flow cytometry. For experiments with GFP-labeled cDC1 and Tomato-labeled lymphatic vessels, we crossed SNX22GFP/GFP C57BL/6 mice to Prox1-Cre-tdTomato+/+ C57BL/6 mice, obtained from Dr. Gwendalyn Randolph (Washington University in St. Louis). We treated the F1s with intraperitoneal tamoxifen [50 μg/g, three times a week, for two weeks with a stock solution of 10 mg/mL tamoxifen (Sigma) dissolved in sunflower oil (Sigma)] as described previously (44) to induce Tomato expression in lymphatic vessels, which was confirmed by examining whole dura with 2-photon microscopy. C57BL/6 CCR7−/− (45), OT-I transgenic (46), and CD45.1+ congenic (47) mice were purchased from The Jackson Laboratory. All mice were housed in accordance with IACUC standards.
All cells for experiments were used 1 to 5 passages after thawing. CT2A cells were obtained from Dr. Peter Fecci in 2016 (Duke University), and GL261 cells were obtained from the National Cancer Institute Tumor Repository in 2014. GL261 and CT2A cell lines were subjected to whole-exome and RNA sequencing to profile, characterize, and validate them (48, 49). U343 cells were obtained from The Cancer Cell Line Encyclopedia at the Broad Institute in 2014 (50). All cell lines were expanded and frozen at early aliquots, and each were cultured for less than a total cumulative time of six months from the time of acquisition to the time of each experiment. GL261-OFP cells were generated by transducing parent cells with mOrange2 (obtained from Dr. Robert Schreiber, Washington University in St. Louis). GL261-zsGreen and CT2A-zsGreen cells were generated by transducing parent cells with zsGreen (obtained from Dr. David DeNardo, Washington University in St. Louis). CT2A-mFlt3L cells were generated by transducing parent cells with murine Flt3L (SinoBiological catalog # MG51113-UT). GL261-OVA (ovalbumin) cells were generated by transducing parent cells with ovalbumin cloned from pcDNA3-OVA (Addgene plasmid # 64599). Transduction was performed as described previously (51). Briefly, 293T cells were transfected with pLX304 vector plasmid along with Δ8.9 and VSV-G packaging plasmids (Addgene) using FuGENE transfection reagent (Promega) according to manufacturer's instructions. Two days later, viral supernatant was pushed through a 0.45 μm polyvinylidene difluoride syringe filter (Sigma), combined with polybrene transfection reagent (Sigma) for a final concentration of 8 μg/mL and added to target cells. This was repeated the following day. After two rounds of transduction and recovery, target cells were selected with Blasticidin (Sigma): 3 μg/mL for CT2A, 10 μg/mL for GL261 until untransduced control cells had died. Retroviral transductions were done with the pBABE backbone plasmid, pCL-Ampho packaging plasmid, and 2 μg/mL puromycin selection. All cell lines were tested for Mycoplasma regularly using an ATCC Universal Mycoplasma Detection Kit. All cells were cultured at 37°C, 5% CO2 in D10 [DMEM (Gibco) with 10% heat-inactivated FBS (Gibco), 1% penicillin/streptomycin (Gibco), 1% minimum essential amino acids (Gibco), 1% l-glutamine, and 1% sodium pyruvate (Gibco)]. Cells were harvested at 70% to 90% confluency to inject intracranially.
50,000 tumor cells in a volume of 5 μL PBS (Gibco) were injected 2 mm to the right and 2-mm posterior of bregma, at a depth of 3.5 mm using a Stoelting stereotactic headframe. Tumors were analyzed at 14 days unless stated otherwise. Sham surgeries consisted of the same protocol (including 5-μL injection of PBS) minus the tumor cells. Sham mice were harvested at the same time as their tumor-bearing counterparts.
Age- and sex-matched WT and IRF8+32kb−/− C57BL/6 mice were intracranially injected with 50,000 untransduced GL261 cells as described above. At days 3, 5, 7, and 14, mice were administered by intraperitoneal injection of either PBS-vehicle or anti–PD-L1 diluted in PBS (Leinco Technologies, Inc. Clone 10F.9G2) at a dose of 200 μg/mouse in a volume of 100 μL. Beginning at 10 days after tumor cell injection, mice were monitored every day for signs of illness and euthanized when moribund.
Adoptive transfer model and OT-I T-cell division
CD45.1+ mice were crossed to CD45.2+ OT-I mice. For adoptive transfer experiments, 5×105 GL261 or GL261-OVA cells were injected intracranially. Four days later, CD8+ T cells were isolated from a CD45.1 x OT-I F1 mouse spleen using an EasySep Mouse CD8a Positive Selection Kit II (Stem Cell), and cells were CFSE (BioLegend) labeled (10 minutes at room temperature, 5 μmol/L), recovered in R10 media [RPMI-1640, 1% l-glutamine, 1% penicillin/streptomycin, 1% minimum essential amino acids, 10% heat-inactivated FBS (all Gibco) supplemented with β-mercaptoethanol (55 μmol/L, Sigma)] at 37°C/5%CO2, washed in PBS, and adoptively transferred via tail vein into recipient mice (5×105 OT-I cells per mouse). At days 3 and 6 after adoptive transfer, tissues from mice were harvested and prepared for flow cytometry as indicated below. The CFSE-high gate was based on the CFSE dilution (or absence thereof) of OT-I cells adoptively transferred into mice bearing untransduced GL261 brain tumors that did not express ovalbumin. The CFSE-low gate was based on the CFSE dilution demonstrated by maximally divided OT-I T cells infiltrating tumors at day 6 post transfer, and the CFSE-low gate was constructed to encompass the majority of terminally divided OT-I T cells within tumors.
C57BL/6 mice were injected subcutaneously in the flank with 1×106 CT2A-mFlt3L cells, such that the transgenic CT2A-Flt3L flank tumors themselves were the source of systemic Flt3L, and the dura could be analyzed in the presence or absence of Flt3L produced by the CT2A-Flt3L transgenic flank tumors compared with untransduced CT2A controls, similar to previously described (15). As a negative control, mice were injected subcutaneously with 1×106 untransduced CT2A cells. Dura from mice was harvested 2 to 3 weeks post transplant of tumor cells, when flank tumors had reached 1 to 2 cm in diameter as measured with a caliper every 2 to 3 days, taking the control and Flt3L mice contemporaneously.
Intracranial tumors from mice were analyzed 14 days after injection, unless otherwise specified. For a given study, up to and including the following tissues were harvested: tumors, superficial CLNs, deep CLNs, a nondraining contralateral inguinal lymph node, dura, and/or spleens. Lymph nodes, dura, and tumors were mechanically dissociated between two frosted slides and digested in 1 mg/mL collagenase A (Roche) in a solution of 2% heat-inactivated FBS (Gibco) in RMPI-1640 (Gibco) for 20 minutes at 37°C/5% CO2. Suspensions were washed and red blood cells lysed with ACK buffer (Lonza) as necessary. Mouse and human brain tumors were separated from myelin using a 22.5% Percoll solution (Thermo Fisher Scientific) and centrifuged at room temperature for 15 minutes at 500 x g (acceleration 9, deceleration 5). Mononuclear cells were separated from spleens by first dissociating spleens between two frosted slides and then centrifuged using a Ficoll gradient (Thermo Fisher Scientific) at room temperature for 20 minutes at 400 x g. We retained the cells in the buffy coat for experiments. In preparation for flow cytometry, cells were suspended in MACS buffer [0.5% BSA (Thermo Fisher Scientific), 2 mmol/L EDTA (Thermo Fisher Scientific) in PBS].
Tumor single-cell suspensions harvested 14 days after injection were separated from myelin using a 22.5% Percoll (Thermo Fisher Scientific) solution and subjected to ACK buffer to lyse red blood cells as described above. CD8+ T cells were isolated with an EasySep Mouse CD8a Positive Selection Kit II (Stem Cell), counted, and plated with naive splenocytes. 50,000 CD8+ T cells were plated with 125,000 naive splenocytes with or without mutant-Imp3 (Imp3 D81N mutation; mImp3) peptide at a concentration of 10 μmol/L (peptide sequence: AALLNKLYA, synthesized by Peptide 2.0) overnight on a pre-coated murine IFNγ detection plate (Cellular Technologies Limited) and analyzed with an ImmunoSpot plate reader (Cellular Technologies Limited). As additional controls for IFNγ release, we plated 125,000 naive splenocytes ± mImp3 peptide ± concavalin A at a concentration of 1 μg/well (Invivogen).
Flow cytometry and tetramer detection
Cell suspensions from tissues were filtered, subjected to Fc block (mouse: BioLegend; human: StemCell), and stained with surface antibodies for >20 minutes on ice. Cells were suspended in MACS buffer as indicated above. Data were acquired on a BD LSRFortessa X-20 flow cytometer and analyzed using FlowJo. The full antibody panel is listed in Supplementary Table S1. The mImp3 (AALLNKLYA)/H-2Db tetramer was generated by the Andrew M. and Jane M. Bursky Center for Immunology and Immunotherapy Programs Immune Monitoring lab (Washington University in St. Louis) as described previously (48). For zsGreen experiments, the 488B/FITC channel was used, with zsGreen+ tumor cells as the compensation control. Tumors originating from untransduced tumor cells were used to assess the baseline fluorescence of immune cells isolated from that particular tumor type, and to determine where to draw the positive gate for zsGreen expression by a particular immune cell. Gating strategies are shown in Supplementary Fig. S1A–S1F (mice) and Supplementary Fig. S2A (human) and are defined in Supplementary Table S2.
5-ALA flow cytometry
5-ALA uptake in U343 cells was assessed by flow cytometry. U343 cells were incubated with 1 mmol/L 5-ALA hydrochloride (Sigma) or vehicle (H2O) as a negative control, trypsinized, and subjected to flow cytometry. In patients with GBM whose tumors were resected using 5-ALA, and who had signed written consent to a Washington University School of Medicine Institutional Review Board–approved protocol (#202107071), resected tumors and peripheral blood were studied. Tumor specimens were prepared as described above. Peripheral blood mononuclear cells (PBMC) were isolated by centrifuging blood on a Ficoll gradient, followed by buffy coat harvest. The 5-ALA metabolite, PPIX, was assessed in the BV650/405C channel, and the unstained tumor cell suspension was used as the compensation control for PPIX+ cells. The thresholds for PPIX+ versus PPIX− populations were determined by comparing tumor-infiltrating APC against their matched peripheral counterpart, which were PPIX−.
Mice were overdosed by intraperitoneal injection with avertin (Sigma) at a dose of 250 mg/kg and perfused with approximately 20 mL of ice-cold PBS until the liver blanched over five minutes using a syringe. The cranial cap was removed with dura still attached to skull, which was then fixed in ice cold 4% paraformaldehyde/30% sucrose in PBS and incubated with shaking overnight. In a subset of mice, blood vessels were labeled 5 minutes before perfusion by injecting intravenously with 594-lectin [Lycopersicon Esculentum (Tomato) Lectin (LEL, TL), DyLight 594, Vector Laboratories]. For brain sections, harvested brains were cut using a vibrotome after fixation. Fixed tissues were glued to a cover slip with superglue (Loctite) and immersed in PBS for imaging. Images were collected using a custom Leica SP8 two-photon microscope (Leica Microsystems) equipped with a 25×0.95 NA water immersion objective, and two femtosecond-pulsing tunable Ti:Sapphir lasers (Mai Tai HP DeepSee and InSight DS+), both from Spectra-Physics. GFP, mOrange, and TdTomato were excited at a wavelength of 925 nm, whereas Dylight 594 and Dylight 649 were excited at 830 nm. Fluorescence emission was guided directly to four external detectors in dendritic arrangement (two hybrid and two classical photomultiplier tubes). For signal separation, three dichroic beam splitters (Semrock) were used. To separate GFP, mOrange, Dylight 594, and the SHG (second-harmonic generation), the three cutoff wavelengths were 358, 538, and 593 nm, respectively. The separation of GFP, tdTomato, DyLight 649, and the SHG was obtained with cutoff wavelengths of 458, 560, and 652 nm. Images were processed and rendered with Imaris cell imaging software (Oxford Instruments).
Student t tests were used to analyze differences between groups. Survival statistics were analyzed with log-rank tests. One-tailed t tests with Welch's correction were used to analyze zsGreen data, in which unequal standard deviations were anticipated between groups, and the signal difference between groups was directionally restricted. Unpaired t tests were used when one mouse was compared with another. Paired t tests were used when individual organs within a mouse were compared against one another, or when cell types within an individual patient were simultaneously analyzed and compared. The Grubbs outlier test was used when noted. A P value of <0.05 was considered significant, and statistical analyses were performed with GraphPad Prism 9. Specific statistical tests used for each experiment are outlined in figure legends.
Data and materials availability
All data needed to evaluate the conclusions of this article are present in the main article or in the Supplementary Materials. Any cell lines generated are available upon request.
cDC1 are recruited to the CNS tumor microenvironment and mediate protection by checkpoint blockade
We first investigated whether cDC were recruited to the brain tumor microenvironment in orthotopic, syngeneic murine GBM models. We injected C57BL/6-derived GL261 or CT2A cells into the cerebral hemispheres of WT mice and identified DC subsets (CD45+F4/80−I-Ab+CD11c+), including cDC1 (additionally gated on XCR1+SIRPα−Ly-6C−; Supplementary Fig. S1A). DC infiltration into both types of brain tumors was much greater than in sham controls (Fig. 1A).
To visualize cDC1 in brain tumors, we used the SNX22GFP/+ knock-in mouse, in which cDC1 specifically and constitutively express GFP (52). In GL261 or CT2A brain tumors from SNX22GFP/+ mice, flow cytometry showed that cDC1 and cDC2 subsets infiltrated brain tumors, and that GFP expression was restricted to the XCR1+/cDC1 subset (Fig. 1B). We used 2-photon microscopy to precisely localize GFP-expressing brain tumor-associated cDC1. Whereas cDC1 were scarce within steady-state brain parenchyma, cDC1 infiltrated extravascular spaces in SNX22GFP/+ brain tumors (Fig. 1C). These data demonstrate that DCs, including cDC1, infiltrate orthotopic GBM.
We next tested the hypothesis that cDC1 were required to mount an immune response against GL261 brain tumors. We used cDC1-deficient IRF8+32kb−/− mice (53). Compared with WT mice, the immune infiltrate within GL261 brain tumors in IRF8+32kb−/− brains lacked XCR1+ cDC1 (Fig. 1D). Both the deep and superficial CLNs were additionally deficient in cDC1 in IRF8+32kb−/− compared with WT mice with brain tumors (Supplementary Fig. S3A–S3B). We next determined whether checkpoint blockade protection against brain tumors required cDC1, as anti–PD-1/PD-L1 treatment improves survival against GL261 (21, 49, 54, 55). Consistently, anti–PD-L1 treatment increased the median and overall survival (OS) of WT GL261-bearing mice compared with vehicle controls; however, cDC1-deficient IRF8+32kb−/− mice experienced no treatment benefit compared with WT controls (Fig. 1E). WT and IRF8+32kb−/− vehicle controls exhibited no difference in median or OS. These data show that DCs, including the cDC1 subset, although scarce in steady-state brain parenchyma, are recruited to the brain tumor microenvironment, and that anti–PD-L1 checkpoint blockade responsiveness in this setting requires cDC1.
cDC1 prime CD8+ T-cell responses against glioblastoma
Because cDC1 can present antigen to prime T-cell responses, we determined the effects of cDC1 deficiency on T-cell composition within the GL261 brain tumors. Compared with WT mice, tumor-infiltrating lymphocytes (TIL) of cDC1-deficient IRF8+32kb−/− mice harbored fewer T cells, including both non-Treg CD4+ and CD8+ T cells (Fig. 2A; Supplementary Fig. S1B–S1C). In contrast, TIL from both genotypes harbored similar frequencies of CD4+ regulatory T cells (Fig. 2A; Supplementary Fig. S1C). A smaller proportion of CD8+ T cells from GL261 tumors in IRF8+32kb−/− mice expressed the functional markers granzyme B and PD-1 compared with WT mice (Fig. 2B; Supplementary Fig. S1D). These data show that GL261 tumors from IRF8+32kb−/− mice both harbor fewer T cells and that a smaller proportion of the infiltrating CD8+ T cells exhibit an activated phenotype.
Although studies have demonstrated that cDC1 cross-present antigen to prime CD8+ T-cell responses in tumors outside the CNS (28, 29, 31, 33–37), it remains unclear whether they perform a similar function in brain tumors, particularly given their scarcity in steady-state brain parenchyma. Therefore, we tested whether cDC1 were required to prime neoantigen-specific CD8+ T cells against GL261 brain tumors. Previously, we found that mice harboring intracranial GL261 mount endogenous CD8+ T-cell responses against the H-2Db-restricted neoantigen, mutant-Imp3 (mImp3; ref. 48). In contrast with WT mice, we identified that equal numbers of CD8+ T cells isolated from cDC1-deficient IRF8+32kb−/− GL261 brain tumors had deficient IFNγ production when stimulated ex vivo by APCs presenting mImp3 peptide by ELISPOT (Fig. 2C and D), and further, completely lacked H-2Db-mImp3 tetramer+ CD8+ T cells (Fig. 2E and F). These data show that cDC1 are required to prime effector CD8+ and neoantigen-specific CD8+ T-cell responses against GL261 brain tumors.
Tumor antigen–containing cDC1 infiltrate brain tumors and localize to CLNs
cDC1 can phagocytize antigen in the periphery and home to secondary lymphoid tissues to prime naive T cells. Although the brain parenchyma does not harbor conventional secondary lymphoid tissue, compelling data have implicated the extracranial CLNs as central to priming CNS antigen-specific T cells (19, 21, 24–26, 56). We used flow cytometry to determine whether cDC1 phagocytized, retained, and trafficked fluorescent protein overexpressed by brain tumor cells, similar to previously used approaches in preclinical melanoma models (29–31). We created two transgenic glioma lines: GL261-zsGreen and CT2A-zsGreen. In mice, the tumor-derived zsGreen protein, phagocytosed by infiltrating immune cells, functions as a detectable surrogate for tumor antigen uptake and trafficking (Fig. 3A). We used CT2A-zsGreen for most tumor antigen-tracking experiments because it most consistently retained zsGreen expression at the time of harvest. In intracranial CT2A-zsGreen tumors, we detected not only zsGreen-positive cDC1, but also zsGreen-positive cDC2, monocyte-derived DCs, and plasmacytoid DCs (pDC; Fig. 3B; Supplementary Fig. S1A), which were absent from nonfluorescent CT2A controls, suggesting that these populations phagocytose tumor-derived antigens in vivo. We next examined the ipsilateral CLNs to probe for extracranial zsGreen+ APCs. We identified migratory and resident zsGreen+ cDC1 in both the deep and the superficial CLNs (Fig. 3C and D, Supplementary Fig. S1E). A larger fraction of migratory cDC1 (CD103+CD8α−) were zsGreen+ compared with the resident cDC1 subset (CD8α+CD103−). We also identified zsGreen+ cDC2, monocyte-derived DC, and pDC in the superficial CLNs and monocyte-derived DC in the deep CLNs (Fig. 3C and D; Supplementary Fig. S4A–S4B). Although GL261-zsGreen tumors less reliably retained zsGreen expression in vivo, zsGreen was still detectable within a considerable proportion of the tumor-infiltrating DCs (Supplementary Fig. S5A). We additionally detected a significant proportion zsGreen+ cDC2 in the deep CLNs, as well as cDC2, monocyte-derived DC and pDC in the superficial CLNs compared with nonfluorescent controls (Supplementary Fig. S5B–S5C). This extended our findings with CT2A-zsGreen brain tumors. These data show that cDC1, as well as other APC populations, harbor tumor-derived antigen within the tumor microenvironment and localize to cervical lymphoid tissue across different brain tumor preclinical models.
Given our findings regarding the importance of cDC1 in mounting CNS antitumor immunity, we set out to investigate the requirement of cDC1 for trafficking tumor antigen from brain tumors to draining CLNs. Although cDC1 were deficient IRF8+32kb−/− mice, cDC2, monocyte-derived DC, and pDC phagocytosed and trafficked zsGreen tumor antigen from tumors to superficial and deep CLNs in equivalent proportions between WT and IRF8+32kb−/− mice (Supplementary Fig. S6A–S6B). These data demonstrate that although cDC1 may be required to mount CNS antitumor immunity, they are not required to traffic tumor antigen from brain tumors to draining CLNs.
Having observed zsGreen+ CLN-associated cDC1, we asked how tumor-derived antigen was trafficked from the CNS to CLNs. We hypothesized that brain tumor–associated antigens could be either actively trafficked by cellular transport within APCs via CCR7-mediated chemotaxis (57) or passively transferred from the brain to the CLN via acellular mechanisms. Therefore, we tested whether zsGreen+ cDC1 required CCR7 to localize to CLNs. Because CCR7−/− mice lack migratory DCs in LNs, we did not distinguish between migratory and resident cDC1 in our analysis and assessed total cDC1. We intracranially injected CT2A-zsGreen into both WT and CCR7−/− mice and monitored the immune cell zsGreen signal in brain tumors and CLNs. We observed equivalent zsGreen uptake by tumor-infiltrating cDC1 in both WT and CCR7−/− mice, suggesting that cDC1 were able to infiltrate brain tumors and phagocytose tumor antigen independent of genotype (Fig. 3E, left). However, in CCR7−/− mice we observed a significantly attenuated zsGreen signal in deep and superficial CLN cDC1 and pDC, and in superficial CLN monocyte-derived DC compared with WT controls (Fig. 3E, middle and right). Small but equal fractions of B cells (CD45+F4/80−I-Ab+CD11c−B220+Ly-6C−) were zsGreen+ in CLNs of WT and CCR7−/− mice (Fig. 3E, middle and right), which poses passive cell migration-independent trafficking as a minor contributing mechanism of CNS lymphatic drainage as B cells enter lymph nodes through high endothelial venules rather than migrating from the periphery (58, 59). In addition, the zsGreen signal was incompletely diminished in DCs from CCR7−/− CLNs, further suggesting that CCR7-independent passive drainage functions to traffic a small fraction of tumor antigen to draining CLNs. Together with our observation that cDC1 are required to generate brain tumor-specific T-cell responses, these results suggest that cDC1 perform this function, at least in part, by phagocytizing tumor-associated material within the brain and trafficking it to the LNs in a CCR7/cell migration-dependent manner.
Dura-associated cDC1 undergo dynamic changes in response to glioblastoma
Previous work has identified the dura, the outer meningeal layer covering the brain, as an immunologically dynamic structure that harbors cDC1, cDC2, pDC, macrophages, T cells, and B cells in both mice (14, 23) and humans (60). This contrasts with the resting-state brain parenchyma, which is devoid of all leukocytes except microglia and border-associated macrophages (14). Additional experiments have demonstrated that Flt3L stimulation expands dura-associated DCs (14, 15). Further work has also shown that the dura harbors a network of lymphatic vessels (18–21, 61, 62), which not only drain cerebrospinal fluid and antigens to the deep CLNs but also proliferate in response to VEGF-C stimulation to potentiate CNS antitumor immunity (21). Given these findings, we set out to further characterize how the dura might support CNS antitumor immunity, focusing specifically on DCs and cDC1. We evaluated cDC1 localization, the dynamics of DC populations during antitumor responses, as well as the antigen presentation capacity of dura-associated cDC during brain tumor growth. We first investigated localization of dura-associated GFP+cDC1 in SNX22GFP/+ mice by 2-photon microscopy. In control and GL261-OFP–bearing SNX22GFP/+ mice, the dura harbored extravascular cDC1 (Fig. 4A). We hypothesized that brain tumors would drive increased dura-infiltrating cDC1. Compared with sham-injected mice, whole dura from intracranial GL261-bearing mice had increased numbers of cDC1, cDC2, monocyte-derived DC, and pDC by flow cytometry (Fig. 4B), although the dura harbored considerably fewer DCs than tumors themselves. We resected the dura surrounding the injection site to avoid skewing data by inadvertently attributing tumor populations to the dura.
Given prior studies demonstrating the existence of dura-associated Flt3L-responsive CD11c+I-Ab+ cells (15) or Flt3L-driven cDC1/cDC2 expansion in which the whole brain and surrounding meninges were examined together (14), we specifically examined the dynamics of dura-associated cDC1 in response to Flt3L stimulation. When we administered Flt3L by introduction of CT2A cells expressing mFlt3L into the flank, dura DC subsets (including cDC1) expanded, observed by flow cytometry (Fig. 4C), as well as by 2-photon microscopy (Fig. 4D, Supplementary Fig. S7A). Notably, Flt3L induced greater expansion of dura DCs than intracranial GL261 brain tumors. These data show that dura-resident cDC1 expand in response to brain tumors or systemically administered Flt3L.
We next investigated the spatial relationship between cDC1 and the CNS-draining dura lymphatic vessels, which could presumably support DC migration from the brain parenchyma to CLNs. We crossed the SNX22GFP/GFP mouse with the Prox1-Cre-tdTomato+/+ mouse, which expresses tdTomato specifically in lymphatic vessels after tamoxifen administration (44). Tamoxifen-treated F1 mice have GFP+ cDC1 and tdTomato+ lymphatic vessels. Using 2-photon microscopy, we detected GFP+ cDC1 within the dura lymphatic vessels of mice that harbored GL261-OFP brain tumors (Fig. 4E, Supplementary Fig. S7B) or were sham-injected (Supplementary Fig. S7C). These data show that cDC1 localize to the lumen of dura-lymphatic vessels in both tumor-bearing and steady-states.
Having identified that dura harbors cDC1 that can localize to lymphatic vessels, we tested if dura-associated cDC1 could acquire tumor-derived antigen from intraparenchymal brain tumors. We orthotopically injected CT2A-zsGreen tumors into the brain parenchyma of mice and monitored the zsGreen APC signal 7 days after injection in tumors, dura, and CLNs by flow cytometry. To avoid contamination of the dura with underlying parenchymal zsGreen+ tumor cells and tumor-infiltrating immune cells, we selected an early time point to ensure that tumors were small and did not abut the dura and resected the dura surrounding the prior injection site to avoid potential contamination of the dura samples by adjacent tumor and immune cells. We observed zsGreen+ migratory cDC1 (in addition to cDC2, pDCs, and monocyte-derived DCs) in tumors, dura, superficial and deep CLNs, as well as resident cDC1 in the CLNs (Fig. 4F and G). We did not observe CD45−zsGreen+ cells in the dura, which indicates that our dura samples lacked tumor contaminant. Together, these data show that dura-associated, Flt3L-responsive cDC1 can localize to dura lymphatic vessels and, along with other dura-associated APC, harbor tumor antigen from intraparenchymal tumors.
CD8+ T-cell priming occurs in CLNs and requires cDC1 and CCR7
Having established that tumor antigen-containing DCs can be found in the dura, as well as the superficial and the deep CLNs, we examined where T-cell priming occurred in response to CNS tumors, as this remains incompletely characterized. Several observations support a role for CLNs in CNS pathology: (i) Ligation of deep CLNs attenuates VEGF-C–mediated brain tumor rejection (21); (ii) lymphadenectomy of CLNs ameliorates disease burden in rodent models of EAE (63–65); and (iii) ablation of meningeal lymphatics decreases antigen-specific T-cell/CD11c+ cell interactions in CLNs of mice with EAE (62). Because we detected tumor antigen–containing APCs in several anatomic sites, we hypothesized that CD8+ T-cell priming could occur within tumors, dura, CLNs, or the spleen. To this end, we tracked cell division in vivo of OT-I T cells adoptively transferred into mice bearing intracranial GL261 tumors expressing full-length ovalbumin (GL261-OVA). OT-I T cells recognize the H-2Kb-restricted OVA antigen SIINFEKL (46) and proliferate upon being primed. We injected GL261-OVA into the brains of CD45.2+ C57BL/6 hosts. Four days later, we intravenously injected CFSE-labeled, CD45.1+CD45.2+ OT-I CD8+ T cells. We harvested tumors, dura, ipsilateral superficial and deep CLNs, spleens, and non–CNS-draining contralateral inguinal LNs at days 3 and 6 post transfer to evaluate CFSE dilution in OT-I CD8+ T cells (Fig. 5A and C). We distinguished CFSE-low, CFSE-mid, and CFSE-high OT-I CD8+ T cells, which we envisaged to reflect terminal effector, initially primed, or unprimed OT-I cells, respectively.
We harvested tissues at days 3 and 6 post transfer to capture snapshots of clonal expansion evolution across tissues. At both timepoints, we observed the greatest numbers of CFSE-high/unprimed OT-I T cells (Supplementary Fig. S1F) in LNs, regardless of location, with fewer in the spleen, and almost none in the tumor or dura (Fig. 5B and C, left), commensurate with our understanding that naive T cells circulate between lymphoid organs before activating and dividing. At both 3 and 6 days post transfer, the CNS-draining superficial and deep CLNs harbored significantly more CFSE-mid/initially primed OT-I T cells than other tissues (Fig. 5B and C, middle), consistent with our understanding that clonal expansion first occurs in proximal draining LNs. At day 3 post transfer, CFSE-low/effector OT-I T cells occupied the greatest fraction of CD45+ cells in the deep CLNs, but were also in the superficial CLNs and TIL, albeit at lower numbers (Fig. 5B, right). However, by day 6 post transfer, TIL harbored the greatest fraction CFSE-low/effector OT-I T cells compared with all other sites except the dura (Fig. 5C, right). This underscores our understanding that the tumor is both the source of antigen and effector site for OT-I T cells.
As a control, we compared CFSE dilution of adoptively transferred OT-I cells from the same tissues in mice that harbored parental GL261 or GL261-OVA brain tumors and observed that mice harboring GL261-OVA tumors had significantly more divided OT-I cells in CNS-draining CLNs, as well as in the spleen, and inguinal LNs at day 3 post transfer compared with GL261 controls (Supplementary Fig. S8A–S8B; top and left, respectively). At day 6 post-transfer, these same differences held, but expanded to include tumors (Supplementary Fig. S8A–S8B; bottom and right, respectively). We observed similar patterns of expansion of OT-I cells in CT2A-OVA tumors versus untransduced CT2A tumors, although the expansion was weaker and somewhat delayed compared with that seen in GL261-OVA tumors (Supplementary Fig. S9A–S9B), which may be due to the immunosuppressive effects of CT2A. Together, these data support a model in which CD8+ T cells are primed against brain tumors in the CNS-draining CLNs, and subsequently home to the CNS to mount antitumor responses. In addition, although tumors are a primary site of effector function, the dura also harbors effector T cells.
Given our observations that cDC1 are required to prime neoantigen-specific T-cell responses against CNS tumors, and that CCR7 is required to traffic antigen from brain tumors to draining CLNs, we next investigated their role in clonal expansion of adoptively transferred OT-I T cells in mice with GL261-OVA brain tumors. Compared with WT mice, CCR7−/− and IRF8+32kb−/− mice had decreased clonal expansion of OT-I T cells in CLNs and spleens 3 days post transfer (Supplementary Fig. S10A, 1st, 2nd, and 5th graph from left). In contrast, we observed minimal differences in clonal expansion between WT, CCR7−/−, and IRF8+32kb−/− mice at day 6 post transfer (Supplementary Fig. S10B). These data collectively demonstrate that CD8+ T-cell priming and clonal expansion against brain tumor-specific antigens occurs in the CLNs and that absence of cell migration to lymph nodes and cDC1 deficiency lead to delayed, but not completely absent clonal expansion in this setting.
CD141+ cDC1 are detectable in human dura and brain tumors
Having identified cDC1 and additional cDC subsets in murine brain tumors and matched dura, we investigated whether human tumors and dura also harbored infiltrating DC populations. Although human DC subsets and functions are less well characterized than in mice, the human cDC1 parallel are CD141-expressing cDC. In both mice and humans, CD141+/cDC1 subset produces IL12 (66), cross-presents exogenous proteins to CD8+ T cells (66), and expresses IRF8 (67), a critical regulatory factor required for mouse (53) and human (68) cDC1 development. We explored immune cell populations within the tumor and matched dura of patients. In addition to tumors, adjacent normal, nontumor involved dura was also resected when indicated (Fig. 6A). We performed flow cytometry on six matched tumor/dura specimens (five meningioma, one GBM) and on 12 additional GBM tumor specimens (13 total: 11 primary, 2 recurrent). In both GBM and meningiomas, we detected multiple human DC subsets within tumor and dura specimens, including cDC1-equivalent CD141+ cDC, cDC1-equivalent CD1c+ cDC, as well as CD14+ and CD16+ monocytes (Fig. 6B and D, Supplementary Fig. S2A). We identified cDC1, cDC2, CD14+ monocytes, CD16+ monocytes, CD4+, and CD8+ T cells in most of the GBM tumors that we analyzed (Fig. 6C). Between GBM and meningiomas, the matched dura samples harbored similar fractions of each subset, although our analysis was limited to a single GBM specimen with matched dura resection (Fig. 6E). These findings collectively show that human conventional DC subsets, as well as monocytes, are abundant in dura and tumors and across multiple brain tumor types, which suggests they may play a role in human CNS antitumor immunity.
CD141+ cDC and other APCs phagocytose tumor-specific markers in GBM
We next evaluated cDC and APC subsets for ability to phagocytize GBM-derived material. Analogous to our studies tracking tumor-derived zsGreen, during GBM resection, the FDA-approved compound 5-ALA can be used in “fluorescence-guided surgery” to fluorescently label tumors and distinguish them from normal brain (38–41). After patients ingest 5-ALA preoperatively, the compound crosses the blood–brain barrier and is metabolized by target GBM cells to the fluorescent PPIX (42), which specifically accumulates and is retained by GBM cells (42, 43). Illuminating the tumor with blue light fluorescently excites PPIX, causing tumors to emit pink light (Fig. 7A). We tested via flow cytometry whether APCs, including cDC, from GBMs resected using 5-ALA could acquire and retain tumor-derived PPIX (Fig. 7B). We analyzed specimens both from 5-ALA–treated patients and U343 GBM cells exposed to 5-ALA. We determined that the PPIX signal was brightest in the BV650 channel (Supplementary Fig. S11A). We also observed clearly defined PPIX-low and PPIX-high cell populations, which suggests a range of PPIX uptake and retention among tumor cells (Fig. 7C, left). We thus used the BV650 channel to detect PPIX for further experiments. Compared with untreated U343 cells, 5-ALA–treated U343 cells also fluoresced brightly in the BV650 channel (Fig. 7C, right). We next analyzed via flow cytometry tumor and matched PBMCs from 5-ALA–treated patients with GBM to determine whether APCs phagocytosed tumor-derived PPIX. Specifically, we addressed two central questions: Do APCs phagocytose PPIX only in the tumor microenvironment and not in the periphery? In tumors, do only phagocytic immune cells acquire and retain tumor-derived PPIX? In PBMCs, neither the CD3ε+ nor the CD3ε− fractions had a detectable PPIX signal. In contrast, whereas the CD3ε+ fraction of the tumor infiltrate was PPIX-negative, a majority of cells in the CD3ε− fraction were PPIX-positive (Fig. 7D and E), demonstrating that PPIX-acquisition was both location-specific and immune cell type-specific. Compared with PBMCs, we observed PPIX within different APC subsets infiltrating tumors, including CD141+ cDC, CD1c+ cDC, CD14+ classical monocytes, and CD16+ nonclassical monocytes across eight specimens: Six primary and two recurrent GBMs (Fig. 7F and G). Moreover, the PPIX-signal occurred in APCs but not T cells infiltrating the tumor (Fig. 7H). Thus, multiple tumor-infiltrating APC subsets acquire and retain GBM-derived tracer. Together, these data show that PPIX-acquisition by CD45+ cells depends on both location and cell identity: Only phagocytic immune cells (including DCs) within the tumor microenvironment, not the periphery, acquire PPIX.
Here, we described the functional importance of cDC1 in immune responses to GBM. We showed that cDC1, although devoid from the brain parenchyma in the steady-state, infiltrate GBM in two preclinical models, that they are critically required for the development of functional neoantigen-specific CD8+ T-cell responses against GBM, and that they mediate checkpoint blockade responsiveness. Using a traceable fluorescent tumor antigen system, we observed that multiple DC subsets, including cDC1, acquire tumor-derived material and traffic it to both the superficial and deep CLNs in a CCR7-dependent manner. We also identified cDC1 within the dura, a subset of which localized to within dura lymphatic vessels. We additionally detected tumor antigen within dura-associated cDC, a presumed midpoint of cDC migrating between the brain-tumor parenchyma and CLNs. On the basis of the dynamics of expansion of adoptively transferred T cells specific to an ectopic GBM antigen, CLNs were the predominant location of priming immune responses to GBM, which required intact cell migration and cDC1 for normal kinetics. Finally, we extended our work in mice to human tumors and dura. Both tumors and dura harbored the mouse cDC1-equivalent CD141+ DCs, among other subsets. In GBM surgically resected using 5-ALA, infiltrating APCs acquired the fluorescent PPIX tumor antigen-surrogate.
Previous reports identified the critical role of cDC1 in antitumor immunity for a range of tumor types (28, 29, 31, 33–37). Our findings extend the importance of the cDC1 to the antitumor immune response in the CNS. Recent work demonstrated that Batf3-dependent cDC1 mediated rejection of immunogenic GL261-FGL2−/− brain tumors (69). One important caveat of our respective observations is that the GL261 orthotopic preclinical model is sensitive to checkpoint blockade therapy. This differs from human GBM, whereby checkpoint blockade therapy is only effective in limited settings (70, 71). Nevertheless, this model is useful to define the elements required for effective CNS antitumor immunity. These data collectively suggest that cDC1 are required for broad CNS antitumor immunity, particularly to generate neoantigen-specific CD8+ Τ-cell responses. Further work is required to understand the requirements for antigen presentation to CD4+ T cells in this setting, and further, to characterize the necessity of cDC1 in additional types of GBM preclinical models. Nevertheless, despite the immunologically specialized nature of the CNS—absence of steady-state parenchyma DCs, lack of LNs in the parenchyma, the presence of cerebrospinal fluid, and the blood brain barrier, among other features—the dependence on the cDC1 subset for immune responses to cancer appears to be shared between the CNS and other extracranial tumor types.
CNS antigen presentation has been a topic of substantial investigation; unlike other sites, the only steady-state parenchymal leukocytes are microglia and border-associated macrophages (which localize to blood vessel basement membranes; ref. 14). They both present antigen but lack extracranial migratory capacity. In contrast, cDC can phagocytose antigen, migrate to secondary lymphoid tissues, and prime naive T cells. Although restricted to the choroid plexus and meninges in the steady-state (14, 15), cDC infiltrate the brain parenchyma during inflammation (16, 22, 23, 25, 26, 56, 72), which holds true in tumors. Previously, the Fabry group showed that cultured intracerebrally injected monocyte-derived DCs could migrate to the CLNs (24). Additional studies demonstrate that cDC depletion attenuates EAE disease severity (22), that MHCII restriction to cDC was sufficient to drive disease progression (23), or that partial depletion of cDC attenuates checkpoint blockade responsiveness against GL261 (73). Nevertheless, these observations leave unanswered whether or how endogenous cDC1 drive CNS antitumor immunity. We demonstrate that cDC1 specifically function as a key CNS APC, by acquiring antigen and migrating extracranially to endogenously prime neoantigen-specific CD8+ T cells and also by mediating anti–PD-L1 responsiveness in the GL261 preclinical model.
The relationship between CNS pathophysiology and the role of CLNs has been supported by a growing body of work in different disease models (18, 19, 21, 63–65). Given the importance of CLNs in CNS lymphatic drainage, we explored the role of DCs, with particular attention to antigen trafficking. Researchers have shown with fluorescent tracers that draining LN APCs (including cDC1/2 subsets, monocytes, and macrophages) contain tumor-associated fluorescent antigens in preclinical melanoma models (28–31). Our studies mirrored this: cDC1 (predominantly migratory), cDC2, monocyte-derived DCs, and pDC harbored zsGreen in both the superficial and deep CLNs, underscoring the tumor antigen trafficking capability of multiple DC subsets. We additionally characterized lymphatic drainage in the absence of cDC1 and found that cDC2 and other DC subsets were capable of trafficking tumor antigens, although they did not possess the necessary machinery to prime CD8+ T cells and stimulate CNS antitumor immunity. This could be for a variety of reasons (74), and further work will be necessary to clarify the mechanism(s).
We consistently observed that both the superficial and deep CLNs contained zsGreen+ cDC following intracranial injection with CT2A-zsGreen, consistent with prior studies demonstrating that deep (19), followed by superficial CLNs (18), can accumulate CNS-derived tracer. Herein, we observed more substantial antigen trafficking to deep CLNs at day 7, but no difference by day 14. Our data support a model whereby deep CLNs capture parenchymal antigens before superficial LNs. However, whereas the drainage route from the venous sinuses to the deep CLNs has been well characterized (18, 19), the drainage route to the superficial CLNs is not, meriting further study.
Parenchymal brain tumor antigens may drain to the CLNs via several potential routes, such as via active cell-mediated transport, LN APC capture of passively draining antigens, or via transfer of antigen-bearing exosomes. We used mice lacking CCR7, which cDC uses to home from the periphery to draining LNs (57), to explore whether active cell-mediated migration was required for tumor antigen to be trafficked to the CLNs. Although tumors from both genotypes harbored tumor antigen-containing DCs, WT mice had a much larger fraction zsGreen-containing DCs in CLNs compared with the CCR7−/− mice. This suggests that antigen trafficking is predominantly an active process that requires intact CCR7-mediated cell migration, which has also been observed in cutaneous melanoma (29). However, it is important to note that a small zsGreen signal was still observed in DCs derived from the LNs of CCR7−/− mice, which suggests that either passive drainage mechanisms or cell-mediated transit via other chemokine receptors may also contribute to the trafficking of tumor-derived material from the brain to the CLNs.
Complementing our CNS tumor antigen-trafficking observations, we investigated where T cells were primed using the GL261-OVA/OT-I model antigen system. We observed OT-I CD8+ T-cell clonal expansion in CLNs, in contrast with the tumor and the dura, which harbored only effector T cells at the later timepoint, suggesting they were primed in the CLNs but had homed to the tumor and dura. We further observed that both CCR7- and cDC1-deficient mice had delayed, but not deficient clonal expansion, which suggests two things: that tumor antigen still passively drains to CLNs in sufficient amounts to eventually drive T-cell clonal expansion, despite not being carried by APCs, and that cell types other the cDC1 can drive clonal expansion of adoptively transferred CD8+ T cells if given enough time. However, these data still underscore the importance of cell migration and cDC1 in priming an effective and early T-cell response against CNS tumors. Our observations extend those of previous work demonstrating the importance of CLN for disease progression in EAE models (63–65), and highlight the importance of the CLNs in priming CNS immune responses. Our data support a model in which APCs carry tumor antigens from intracranial tumors to CLNs, where cDC1 specifically prime tumor-specific CD8+ T cells.
We extended our preclinical observations to patients. There has been significant work on the use of autologous cultured monocyte-derived DCs in GBM immunotherapy (75, 76). Although the approaches are promising, instead of focusing on DCs as a therapeutic approach, we instead studied the behavior and function of endogenously arising cDC subsets in patients with GBM. We identified CD141+ cDC1, CD1c+ cDC2, CD14+ monocytes, and CD16+ monocytes, as well as CD4+ and CD8+ T cells, in GBM, extending our observations that cDC infiltrate mouse GBM.
To our knowledge, the phenomenon of direct tumor antigen uptake by cDC has not been observed in human cancers. 5-ALA is selectively up-taken, metabolized into the PPIX fluorophore, and retained by GBM tumor cells (42, 43), which allows the operator to more precisely discriminate GBM from normal brain (38–42). We leveraged this to probe for PPIX in various GBM-infiltrating DC and APC subsets. We observed PPIX specifically within tumor-associated APCs. Tumor-associated CD3ϵ+ T cells were PPIX-deficient. In addition, PPIX+ APC specifically localized to tumors: matched PBMC subsets obtained contemporaneously were PPIX-negative. These important controls established that the PPIX signal was dependent on immune cell identity and location.
Our traceable GBM tumor antigen data from mice (using zsGreen) and humans (using PPIX) pose several potential mechanisms of antigen acquisition by tumor-infiltrating APCs: (i) Phagocytosis of GBM cells, (ii) antigen-transfer to APCs via exosomes, (iii) phagocytosis of debris released from dying tumor cells, or (iv) ingestion from other currently undefined mechanisms. We observed large fractions (>75%) of PPIX+ DCs in human GBM, similar to mice (50%–75% of zsGreen+ DC). The fluorescent signal was not uniform between DC subsets in both settings. This implies that fluorophores with different molecular structures (one a heme-derivative, the other a protein) might be transferred into APCs by similar mechanisms, and that differences in fluorescent percentage between different APC subsets might reflect functional differences. These findings suggest parallel mechanisms of tumor antigen-uptake exist in mice and humans. Further work should use more granular analysis to characterize additional subsets that might acquire tumor antigen and should aim to identify molecular signatures of APC cell states pre- and post-fluorophore ingestion.
Growing work has identified the meninges as an immunologically dynamic structure that contains Flt3L-sensitive cDC and pDC (14, 15, 23, 60) and harbors a network of VEGF-C–responsive lymphatic vessels (18–21, 61, 62) that drain cerebrospinal fluid and antigen to the deep CLNs and facilitate CNS antitumor immunity. We showed that brain tumors stimulate expansion of Flt3L-sensitive DC populations. We identified cDC1 in dura lymphatic vessels by 2-photon microscopy and observed that dura harbors tumor antigen-containing DCs. We translated our preclinical findings to human specimens and observed that the dura harbors both the CD141+ cDC1 and the CD1c+ cDC2 human equivalents. Additional studies are needed to define the functional contributions of dura-associated DCs, the interplay with the dura lymphatic network, and the anatomic pathways by which APC traffic parenchymal tumor antigens to the dura.
Taken together, these data show that CNS antitumor immunity requires cDC1, which infiltrate tumors, acquire antigen, and traffic the antigen they acquire to CLNs, with the dura playing a supportive role. Additional work will be needed to understand the anatomic basis of cDC1-associated antigen trafficking to the dura and LNs and to determine whether appropriately polarized cDC1 can be leveraged therapeutically in patients with GBM.
N.M. Kretzer reports grants from Howard Hughes during the conduct of the study as well as personal fees from GOG Foundation/Immunogen outside the submitted work. E.C. Leuthardt reports other support from Neurolutions outside the submitted work. A.H. Kim reports other support from Monteris Medical, and grants from Stryker and Collagen Matrix outside the submitted work. G.P. Dunn reports personal fees from Ziopharm Oncology and other support from Immunovalent Therapeutics outside the submitted work. No disclosures were reported by the other authors.
J.A. Bowman-Kirigin: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. R. Desai: Data curation. B.T. Saunders: Data curation, validation, visualization, methodology. A.Z. Wang: Data curation. M.O. Schaettler: Data curation. C.J. Liu: Data curation. A.J. Livingstone: Data curation. D.K. Kobayashi: Data curation. V. Durai: Data curation. N.M. Kretzer: Data curation. G.J. Zipfel: Data curation. E.C. Leuthardt: Data curation. J.W. Osbun: Data curation. M.R. Chicoine: Data curation. A.H. Kim: Data curation. K.M. Murphy: Resources, data curation. T.M. Johanns: Resources, data curation. B.H. Zinselmeyer: Resources, data curation, supervision. G.P. Dunn: Conceptualization, data curation, resources, supervision, funding acquisition, investigation, writing–original draft, project administration, writing–review and editing.
We acknowledge the Immune Monitoring Laboratory (particularly Diane Bender) and Washington University Department of Pathology Flow Cytometry Core for flow cytometer and instrumentation support. We acknowledge David DeNardo, Gwendalyn Randolph, Josh Rubin, Morey Blinder, and Robert Schreiber for their incisive input. We thank Matthew Holt for his superb illustrations. This study was supported by the National Institutes of Health NINDS grant R01NS112712 and Cancer Research Institute Lloyd J. Old STAR Award (to G.P. Dunn), National Institutes of Health NCI grant F30CA236454 (to J.A. Bowman-Kirigin).
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/).