Intrinsically de-sialylated CD103⁺ CD8 T cells mediate beneficial anti-glioma immune responses

Abstract

Background

Cancer vaccines reproducibly cure laboratory animals and reveal encouraging trends in brain tumor (glioma) patients. Identifying parameters governing beneficial vaccine-induced responses may lead to the improvement of glioma immunotherapies. CD103⁺ CD8 T cells dominate post-vaccine responses in human glioma patients for unknown reasons, but may be related to recent thymic emigrant (RTE) status. Importantly, CD8 RTE metrics correlated with beneficial immune responses in vaccinated glioma patients.

Methods

We show by flow cytometry that murine and human CD103⁺ CD8 T cells respond better than their CD103⁻ counterparts to tumor peptide-MHC I (pMHC I) stimulation in vitro and to tumor antigens on gliomas in vivo.

Results

Glioma responsive T cells from mice and humans both exhibited intrinsic de-sialylation-affecting CD8 beta. Modulation of CD8 T cell sialic acid with neuraminidase and ST3Gal-II revealed de-sialylation was necessary and sufficient for promiscuous binding to and stimulation by tumor pMHC I. Moreover, de-sialylated status was required for adoptive CD8 T cells and lymphocytes to decrease GL26 glioma invasiveness and increase host survival in vivo. Finally, increased tumor ST3Gal-II expression correlated with clinical vaccine failure in a meta-analysis of high-grade glioma patients.

Conclusions

Taken together, these findings suggest that de-sialylation of CD8 is required for hyper-responsiveness and beneficial anti-glioma activity by CD8 T cells. Because CD8 de-sialylation can be induced with exogenous enzymes (and appears particularly scarce on human T cells), it represents a promising target for clinical glioma vaccine improvement.

Electronic supplementary material

The online version of this article (doi:10.1007/s00262-014-1559-2) contains supplementary material, which is available to authorized users.

Introduction

High-grade gliomas such as glioblastoma multiforme (GBM) have dismal prognosis. The best established therapies for these tumors extend survival by weeks or months, and have significant toxicity [1]. More effective and less toxic therapies for GBM are therefore needed.

DC vaccination has emerged as an attractive experimental therapy for GBM, as it is associated with negligible toxicity, increased survival, and potential synergism with chemotherapy in early clinical studies [2–12]. Despite such promise, there is little consensus on how to improve DC vaccines for human cancers. One reason for this is that there was previously no clear understanding of what types of immune responses induced by vaccination were clinically important. For example, immunological metrics in vaccinated tumor patients (i.e., those with non-CNS tumors) often failed to correlate with clinical benefits after vaccination, although this correlation was clearly seen in rodent tumor models. More recently, levels of vaccine-associated type I cytokine production (IFN-γ or IL-12) were seen to correlate with post-vaccine survival in GBM patients [11, 13]. This suggests that type I immune responses induced by DC vaccines may be clinically beneficial against glioma, but it is unclear whether targetable factors govern such responses.

We previously showed that in vivo post-vaccine IFN-γ responses are dominated by CD103⁺ CD8 T cells, which are enriched for CD8⁺ recent thymic emigrants (CD8 RTEs) [11, 14]. CD103, or alphaE-integrin, is upregulated on CD8 RTEs as well as on mucosa-associated memory T cells [15–17]. In addition, CD103 is expressed on a subset of DCs associated with mucosal surfaces such as lung and intestine [18–20], and on subpopulations of CD4 and CD8 T regulatory cells [21–23]. Finally, CD103 is important for cellular homing to mucosal tissues, as well as retention in brain and brain tumors [24–27]. Thus, while the exact roles of CD103 in multiple aspects of immunity are still being studied, elucidating the mechanisms whereby CD103⁺ CD8 T cells dominate antitumor responses in brain tumor hosts may reveal important ways to enhance vaccine efficacy. Here, we demonstrate that CD103⁺ CD8 T cells from naïve mice and humans are intrinsically hyper-responsive to cell-free antigen stimulation. Furthermore, in a murine glioma model (GL26) as well as in GBM patients, we show that vaccine-responsive CD8 T cells are CD103⁺ and exhibit prominent de-sialylation.

Protein sialylation is important in cross-reactivity, responsiveness, and peripheral tolerance of T cells [28–31]. Specifically, mice deficient for the ST3Gal-I sialyltransferase are deficient in immune receptor sialylation, including that of CD8 on T cells, and exhibit perturbed central tolerance and T cell maintenance [28, 29]. Moreover, intrinsic de-sialylation of galactosyl (β-1,3) N-acetyl galactosamine on CD8 beta (CD8β) of immature (CD4⁺CD8⁺ double-positive, or DP) thymocytes greatly increases CD8 avidity for pMHC I [29, 30]. This can alter peptide responsiveness through TCR [29, 30]. Neuraminidase enzymatically de-sialylates galactosyl (β-1,3) N-acetyl galactosamine on mature peripheral CD8 T cells and broadens their antigen reactivity [29, 30], but the involvement of CD8β in this phenomenon is not known. Interestingly, mature CD8⁺ single-positive thymocytes from some mouse strains exhibit partial de-sialylation [30], revealing the possibility that their CD103⁺ CD8 T cell progeny might also exhibit partial de-sialylation.

We further show here that exogenous neuraminidase treatment preferentially de-sialylates CD8β and is sufficient to render minimally responsive CD103⁻ CD8 T cells hyper-responsive to antigen stimulation. Conversely, preferential sialylation of CD8β with exogenous ST3Gal-II sialotransferase significantly diminished responsiveness of CD103⁺ CD8 T cells to antigen stimulation. Surface de-sialylation and CD103 expression were also necessary for donor CD8 T cells and/or lymphocytes to benefit murine glioma hosts, and increased ST3Gal-II expression correlated with reduced clinical benefits in a meta-analysis of vaccinated glioma patients. These studies indicate that CD103⁺ CD8 T cells directly mediate beneficial anti-glioma responses in vivo due to hyper-responsiveness to tumor antigens and that de-sialylation of immune receptors such as CD8 may be a useful target to improve glioma vaccine therapy.

Materials and methods

Animal and human subjects

Mice

Female C57BL/6 (Jackson Labs) and B6-Foxn1 mice (Harlan, Inc.) were housed in a pathogen-free vivarium and used on approved protocols according to federal guidelines. All animal procedures were approved prior to performance by our Institutional Animal Care and Use Committee. DC vaccination of mice: C57BL/6J mice were vaccinated subcutaneously 3 and 7 days post-tumor implantation with 10⁷ GL26 lysate-pulsed cultured DC2.4 cells as previously described [32]. Unless otherwise specified, flow cytometry was performed using blood/tissue from mice exhibiting terminal tumor symptoms.

Patients

Investigations were performed under institutional review board-approved protocols and in accord with an assurance filed with and approved by the US Department of Health and Human Services. Participating CSMC patients provided informed consent for gene profiling, vaccination (where appropriate), and all associated analyses, prior to surgery. Vaccination was performed as previously described in the literature under phase II [11] tumor lysate/DC or phase I [33] tumor peptide/DC trials (NCI registry #: NCT00576537 and NCT00576641, respectively). Immunological responses were determined using the following formula: [(antigen-stimulated/nonstimulated post-vaccine % CD8⁺IFN-γ⁺ cells)/(antigen-stimulated/nonstimulated pre-vaccine % CD8⁺IFN-γ⁺ cells)]. Pre-vac IFN-γ response = (antigen-stimulated/nonstimulated pre-vaccine % CD8⁺IFN-γ⁺ cells). Vaccine administration, immunological, and clinical responses have been published for these trials [5, 11, 33].

Tumor cell implantation

The murine (C57BL/6) GL26 glioma cell line [34–36] was obtained in 1998 with permission from Dr. Henry Brem (Johns Hopkins, Baltimore, MD, USA). GL26 cells were cultured in RPMI 1640/10 % FBS (RPMI-10; Invitrogen Corp., Grand Island, N.Y. 14072, USA). Cultured GL26 glioma cells were trypsinized, and 50,000 GL26 tumor cells implanted in 2 μl of 1 % methylcellulose using a stereotactic rodent frame, with injection 1 mm posterior and 2.5 mm lateral to the junction of the coronal and sagittal sutures (bregma), at a depth of 3.5 mm. GL26 tumors in C57BL/6J mice typically ranged from 15 to 65 mg. Authenticity of the GL26 line was continually verified in each experiment by tumor formation upon implantation into C57BL/6 female mouse brain (50,000–200,000 cells), with prominent ependymal features on H&E histology (uniformly circumscribed margins, vascular pseudorosettes).

Time course study

Six-week-old mice were used for the time course study. After intracranial implantation of 50,000 GL26 tumors cells (day 0), mice (n = 3 per date) were killed at day 6, 17, and 27 for tissue harvesting. Single cell suspensions of tumor-infiltrating leukocytes (TILs) were obtained from excised GL26 tumors by forcing brain tissue through a 20 µm cell strainer and rinsing twice in PBS. At the same time, blood and spleen were collected from each mouse. PBMC, splenocytes, and TILs were analyzed by flow cytometry for CD8 studies. Experiments were repeated twice.

Flow cytometry and tetramer stimulation

After FcR blocking, the following mAb were used (all from BD Biosciences): CD8 (53-6.7), anti-αEβ7 integrin (M290), CD62L (MEL14), CD69 (H1.2F3) and anti-IFN-γ (XMG1.2); CD8 (G42-8), CD103 (BER-ACT8). 5 × 10⁵ or 1 × 10⁶ cells were stained with mAb to surface proteins in 50 μl with a titrated amount of antibody for 30 min on ice, followed by washing and incubation with secondary 30 min on ice, permeabilization (where applicable) and staining with mAb to IFN-γ and FcR 30 min on ice, or surface staining with pMHC I tetramers 30 min at 25 °C. Mouse pMHC tetramers included the following: PE-human gp100[KVNPRNQDWL]/H-2D^b; PE-Trp-2[SVYDFFVWL]/H-2K^b; PE-LCMV[KAVYNFATC]/H-2D^b; PE-OVA[SIINFEKL]/H-2K^b and PE-empty HLA-A2 (all Beckman Coulter Immunomics) and were used either for staining or stimulation. Human pMHC tetramers included the following: PE-human gp100/HLA-A0201; PE-Her-2Neu/HLA-A0201; PE-MAGE-1/HLA-A2; APC-MART-1/HLA-A2; PE-CD133[YLQWIEFSI]/HLA-A2; and empty HLA-A2 (all Beckman Coulter Immunomics) and were used either for staining or stimulation.

Sialylation and de-sialylation assays

ST3Gal-II enzyme (Calbiochem #566227) or neuraminidase treatment (Type III from Vibrio cholerae, Sigma) was added to 10⁶ CD8 cells in 50 µl of RMPI/5 % FBS (RPMI-5). CMP-sialic acid (5.5 μl of 500 μM stock) was added to every 7.54 µl ST3Gal-II per 10⁶ cells to sialylate; 8.7 µl of neuraminidase was used per 10⁶ cells to de-sialylate. Cells were incubated 30 min 37 °C in 5 % CO₂, rinsed 2× with media and stained with PNA to validate enzyme function, and cells were used for in vitro or in vivo experiments.

PNA staining

Peanut agglutinin (PNA; Vector Laboratories) binds preferentially to a commonly occurring structure, galactosyl (β-1,3) N-acetylgalactosamine. Terminal sialic acid on CD8 molecules prevents PNA ligation. Therefore, PNA staining and flow cytometer analysis have been used to evaluate the glycosylation of the different subsets of CD8 cells (i.e., CD103⁺ CD8⁺ vs. CD103⁻ CD8⁺ cells). Mean PNA and CD103 fluorescence were quantified on electronically gated IFN-γ⁺ and IFN-γ⁻ CD8 T cells from patients exhibiting greater than 50 % increase in IFN-γ production, either in response to antigen stimulation prior to vaccination or in response to antigen stimulation after vaccination, with combined results and significance presented.

Functional assays, mouse

CD103⁺CD8⁺ and CD103⁻CD8⁺ cells were purified by magnetic separation (Myltenyi Biotech, Sunnyvale, CA, USA) from spleens of 5–7-week-old mice, which typically resulted in >85 % purity. CD8 T cells were stimulated overnight by tetramer (0.5 µl tumor peptide/MHC I tetramers plus 1 µg anti-CD28 mAb in RPMI-5 in 5 % CO₂) and expression of IFNγ analyzed on a FACScan II cytometer and Cell Quest software. The same experiments were performed after ST3Gal-II or neuraminidase treatment.

Functional assays, human

CD8 T cells were purified from buffy coats of healthy (nonsmoker) donors, or from pre-vaccine GBM patients by anti-CD8 immunobeads (Miltenyi Biotec, Sunnyvale, CA, USA). 10⁷ PBMCs, or 5 × 10⁴ purified CD8⁺ T cells, with or without enzyme treatment (neuraminidase or ST3Gal-II), were plated at 10⁶ cells/well into 96-well plates with each well containing 50 μl RPMI/2 % FBS media, 3 μl of pHLA tetramer and 1 μl of antihuman CD28 mAb (BD Biosciences Pharmingen). Plates were incubated at 37 °C, 5 % CO₂ for 3 and 7 days prior to IFN-γ intracellular staining.

Western blotting and immunoprecipitation

CD8 T cells were isolated by MACS column separation from two healthy 8-week-old C57BL/6 mice (A, B), a human GBM patient (C), or a healthy young (22 years) human donor (D), and treated with neuraminidase and/or ST3Gal-II. Cells were lysed at 10⁸ cells/ml, and lysates run on 12 % SDS/PAGE gels with and without reduction using 10 % 2-ME (Red and NR, respectively), and subjected to blotting with PNA-biotin in the presence or absence of anti-CD8 mAb as indicated, followed by development with streptavidin-HRP. For immunoprecipitation, nonreduced lysates (10⁷ cell equivalents each) from surface-biotinylated healthy human CD8 T cells treated as above were subjected to pre-clearing with 5 μg anti-CD8α plus anti-CD8β, or isotype-matched control monoclonal antibodies as indicated, prior to PNA immunoprecipitation. Pellets were subjected to gel electrophoresis and developed with streptavidin-HRP. Identical gel loading was verified by Ponceau S staining, and individual lanes on film scans subjected to identical contrast/brightness adjustment.

Adoptive transfer of CD8 T cells

Mouse CD8⁺ T cells were purified from C57BL/6J spleens as above. Three and seven days after tumor implantation, 10⁷ untreated, ST3Gal-II- or Neuraminidase-treated CD8 T cells in 50 µl PBS were transferred into GL26, B6.Foxn1 (inbred nude) or nu/nu (outbred nude) hosts by i.v. tail vein injection (n = 5 per group). Transfer efficiency was validated by donor cell ≥10 % in nontumor-bearing B6.Foxn1 spleens 3 weeks after injection. Single cell suspensions were prepared from tumors excised from moribund mice as above, and flow cytometry performed to detect donor cells. Young splenocytes from newborn mice (≤14 days) were transferred into nu/nu mice for survival experiments (n = 5 per group). Nude mice were reconstituted with 10⁷ spleen cells (nu/nu + spleen) or CD8 T cells (nu/nu + CD8 T) from young (<10 weeks old) C57BL/6 donors, 3 and 7 days post-tumor injection, either as is, or after treatment with ST3Gal-II (-ST3). Similar transfers were performed using CD8 T cells from CD103-deficient mice on C57BL/6 background (CD103^−/−; n ≥ 5 mice/group.

Quantification of in vivo glioma invasion

Frozen brain tumors from moribund mice were sectioned (8 µm) and stained with hematoxylin and eosin (H&E). Tumors were photographed under a microscope at 4× and 40× magnification. For macroscopic evaluation, grossly noncontiguous tumor regions were counted without magnification in four serial tumor sections. Data depicted are derived from a single individual observer, but qualitative consistency of data trends was validated by at least three individual observers.

Compilation of clinical trial data

Survival and immune responder status data from published clinical trials using dendritic cell vaccines to treat glioblastoma were compiled from the following studies: Yu et al. [2, 5], Kikuchi et al. [3], Kikuchi et al. [6], Rutowski et al. [7], Liau et al. [9], Yamanaka et al. [8] and Wheeler et al. [11].

RNA isolation and microarray analysis

Within 15 min of surgical resection and keeping the tissue on ice, tumors were examined by a neuropathologist and dissected into a portion for tissue diagnosis by histopathology, and another for RNA extraction. The RNA extraction portion (typically about 0.5 cm in diameter) was snap-frozen in liquid nitrogen, stored at −80 °C, and RNA extracted and evaluated as previously described [37]. Ten microgram of total RNA was used to synthesize ds cDNA, biotin-labeled antisense cRNA synthesized by in vitro transcription and 20 µg chemically fragmented cRNA hybridized to Affymetrix HG-U133 Plus 2 GeneChip arrays (Affymetrix, Santa Clara, CA, USA). Quality, yield, and size distribution were estimated by spectroscopy and electrophoresis as previously described [32]. Arrays were imaged, and data acquired as MAS5-normalized values as described previously [32]. Other immunosuppression probe-sets analyzed: Adenosine A2ARec (205013_s_at), CD274/B7-H1 (223834_at), CTLA4 (234362_s_at, 234895_at), FASLG (211333_s_at), FOXP3 (221333_at, 221334_at, 224211_at). IDO/INDO (210029_at), IL10 (207433_at), IL2RA/CD25 (211269_s_at), LGALS3/Galectin-3 (208949_s_at), ST3GAL1 (208322_s_at), ST3GAL2 (205346_at, 217650_x_at, 228784_at, 229336_at), TGFB1 (203084_at, 203085_s_at), TGFB2 (209908_s_at, 209909_s_at, 220406_at, 220407_s_at), TGFB3 (1555540_at, 209747_at) and VEGF (210512_s_at, 211527_x_at, 212171_x_at).

Statistical analysis

Survival (days from tumor implantation to acquisition of characteristic spectrum of terminal neurological symptoms) was assessed by two-tailed Mantel–Cox log-rank. All in vitro and animal experiments were repeated at least twice (≥3 total repetitions) with similar results. All statistical methods in patients (ANOVA and/or t test, Pearson’s correlations) are indicated for specific analyses in their respective figure legends.

Results

Beneficial impact of CD103⁺ CD8 T cells on GBM patients

We first examined whether the presence of CD103⁺ CD8 T cells could predict endogenous or vaccine-induced antitumor responses in GBM patients. Levels of CD103⁺ CD8 T cells determined by flow cytometry correlated significantly with post-vaccine IFN-γ enhancement (Fig. 1a). Such enhancement can predict improved survival, as well as improved chemotherapeutic responsiveness in vaccinated GBM patients, and so, it is potentially beneficial to patients [11]. In contrast, CD103⁺ CD8 T cell levels correlated inversely with endogenous antitumor responses (Fig. 1b). Because endogenous antitumor responses correlate inversely with progression-free survival in GBM patients, they cannot be considered potentially beneficial [11]. Taken together, this suggests that CD103 expression correlates directly with beneficial antitumor immune responses and inversely with immune responses of no proven benefit [12, 14]. This is consistent with previous studies showing CD103 expression to be correlated with recent thymic emigrant (RTE) status, which is itself correlated with better overall survival [14, 16, 38].

Fig. 1.

CD103⁺ CD8 T cells account for vaccine responsiveness in human cancer patients. Proportions of CD8 T cells that were CD103⁺ in GBM patients prior to vaccination (% pre-vac CD103⁺) correlated directly with in vitro IFN-γ responses to tumor antigens after vaccination (a), but inversely with pre-vaccine responses (b), with the indicated coefficients. IFN-γ increase was assessed by flow cytometry following in vitro re-stimulation with immunizing peptide antigens. Patients were analyzed during course of published phase I (a) or phase II (b) vaccine studies

Preexisting CD103⁺ CD8 T cells dominate early anti-glioma responses in mice

To determine whether CD103⁺ CD8 T cells also dominate anti-glioma responses in mice, we first asked whether CD103⁺ CD8 T cells preferentially expanded at various times after intracranial implantation of GL26 glioma cells into mice. Vaccination resulted in increased levels of T cells, especially CD8 T cells, at the tumor site (Supplemental Fig. 1A). In addition, and consistent with previous reports, CD103⁺ but not CD103⁻ CD8 T cells increased within brain tumors very early after tumor implantation (Supplemental Fig. 1A) [25]. Both CD103⁺ and CD103⁻ T cells were significantly increased after vaccination, although the increase in CD103⁻ T cells was slightly delayed but caught up quickly with CD103⁺ T cells (Supplemental Fig. 1A). A significantly greater proportion of activated (CD69⁺) CD8 T cells within GL26 tumors also expressed CD103 with or without vaccination (Supplemental Fig. 1B). This is consistent with our previous detection of antitumor CTL activity and tumor-infiltrating T cells in unvaccinated wild-type GL26 hosts, as well as their increased survival relative to T cell-deficient hosts [32]. These data are consistent with either rapid induction of CD103 expression or rapid expansion of preexisting CD103⁺ CD8 T cells, which are inherently more responsive to glioma antigen (Ag).

To discern these two possibilities, we first identified a dominant epitope within the two known tumor antigens expressed by GL26, gp100 and Trp-2 [34]. Significant proportions of splenic but not blood CD8 T cells from GL26 hosts recognized the Trp-2/H-2K^b (SVYDFFVWL/H-2K^b) epitope, and the proportion reactive to this epitope was greatly enriched within tumor-infiltrating CD8 T cells (Fig. 2a, b). This appeared selective for Trp-2, as splenic or tumor-infiltrating CD8 T cells did not bind high levels of gp100 or control tetramers (Fig. 2a, b). Within tumors, roughly twice as many CD103⁺ as CD103⁻ CD8 T cells were reactive against Trp-2 epitope (Fig. 2c), whereas no significant difference in Trp-2/H-2K^b tetramer binding was observed between splenic CD103⁺ and CD103⁻ CD8 T cells (Fig. 2c). This established Trp-2/H-2K^b-reactive T cells as a major component of peripheral and intratumoral anti-GL26 responses.

Fig. 2.

Tumor antigen reactivity on CD103⁺ and CD103⁻ CD8 T cell subsets in mice. a pMHC I tetramer binding on CD8 T cells within TIL, splenocytes and blood from GL26 hosts 17 days after tumor implantation. b Proportions of tetramer + cells within CD8 T cells from TIL, spleen and blood, compiled from multiple GL26 hosts (n = 5 mice/group; *P < 0.01 relative to Ova control on same tissue by two-sided t test). c Typical CD103/CD8 flow cytometric profile of peripheral lymphocytes from GL26 host 17 days after tumor implantation (n = 6 mice/group; *P < 0.01 by two-sided t test). d CD103⁺ CD8 T cells from TIL 17 days after tumor implantation are enriched for Trp-2 epitope reactivity. Asterisk denotes significantly different tetramer staining between indicated subpopulations (n = 6 mice/group; *P < 0.01 by two-sided t test). Similar trends were observed in ≥3 independent experiments

We then stimulated CD103⁺ and CD103⁻ CD8 T cell subpopulations from naïve mouse donors with tumor antigen pMHC I tetramers, including the Trp-2 epitope we identified, in the presence of anti-CD28 mAb in vitro. IFN-γ production 24 h after stimulation was significantly higher in CD103⁺ compared to CD103⁻ subpopulations after stimulation. Intriguingly, this applied to all antigens including Trp-2/H-2K^b (Fig. 2d). This suggests that preexisting CD103⁺ CD8 T cells are broadly more responsive to tumor antigen relative to CD103⁻ CD8 T cells in vitro.

Tumor antigen-reactive CD8 T cells exhibit low surface sialylation

Surface de-sialylation of galactosyl (β-1,3) N-acetyl galactosamine can broadly augment reactivity to tumor antigen [39] and is developmentally regulated on CD8β subunits during T cell development [29, 30]. Although mature CD4⁺ or CD8⁺ single-positive (SP) thymocytes exhibit less de-sialylation than their precursors, they are relatively de-sialylated compared to fully mature peripheral T cells [29, 30]. The same could conceivably hold true for CD103⁺ CD8 T cells, as a large proportion of them are more similar to SP thymocytes than mature peripheral T cells. Hence, we considered whether peripheral CD103⁺ CD8 T cells might also exhibit more CD8β de-sialylation than CD103⁻ CD8 T cells.

PNA lectin binds to non-sialylated galactosyl (β-1,3) N-acetyl galactosamine on CD8, as well as on other molecules [29, 30], and can be quantified by PNA fluorescence intensity (MFI) in flow cytometry. Using PNA MFI, we found that tumor-infiltrating CD8 T cells as well as Trp-2 tetramer-positive peripheral CD8 T cells from GL26-bearing mice exhibited significantly higher PNA binding than tetramer-negative peripheral CD8 T cells (Fig. 3a, b) [29, 30]. Vaccination did not alter this pattern (Fig. 3b), and the PNA^hi phenotype was evident on preexisting CD103⁺ CD8 T cells (Fig. 3c). Similarly, human CD8 T cells producing IFN-γ in response to tumor Ag stimulation from vaccinated GBM patients were exclusively CD103⁺ and PNA^hi (Fig. 3d, e). GBM patients exhibiting responses to tumor antigen also exhibited higher PNA binding by pHLA tetramer-positive cells, whereas binding to a distinct lectin (PHA-L, which binds N-acetylglucosamine β(1 → 2) mannopyranosyl) was unaffected by CD103 expression (Fig. 3f). These data suggest that CD103⁺ glioma antigen-reactive CD8 T cells from mice and humans exhibit relative de-sialylation of surface galactosyl (β-1,3) N-acetyl galactosamine.

Fig. 3.

Low surface sialylation of antigen-reactive T cells in mouse and human glioma hosts. Vaccinated or nonvaccinated C57BL/6 mice were killed at various timepoints after intracranial GL26 implantation, and CD8 T cells within either TIL or spleen, or tetramer⁺ and tetramer⁻ CD8 T cells from spleen were simultaneously analyzed for PNA-FITC binding by flow cytometry. PNA plots from representative individual mice are shown in (a), with mean PNA fluorescence compiled from three animals per indicated time point and treatment, with significance (n = 9; ***P < 0.0005) between antigen-reactive and nonreactive cells above each pair, shown in b. c Purified CD8 T cells from mouse splenocytes stained for CD103 and PNA. CD103⁺ CD8 T cells show higher PNA binding compared to CD103⁻ CD8 T cells. d Preexisting and vaccine-induced antitumor responsiveness was determined in vaccinated human GBM patients by quantifying IFN-γ production before and after vaccination by intracellular flow cytometry after in vitro stimulation, with simultaneous analysis of CD8, CD103 and PNA on stimulated cells. Mean PNA and CD103 fluorescence on IFN-γ⁺ and IFN-γ⁻ CD8 T cells from immunologically responding patients (n = 6/group;***P < 0.0005; *P = 0.026) shown in e. f Lectin binding profile of peripheral lymphocytes from GBM patient exhibiting vaccine-induced response to Her-2 peptide epitope exhibits specifically increased PNA binding on antigen-reactive CD8 T cells. All statistical evaluations were performed by two-sided t test. Similar trends were observed in ≥3 independent experiments, with representative profiles presented in a, c, d and f

Modulation of CD8 coreceptor sialylation on T cells

Neuraminidase and ST3Gal-II enzymes catalyze the removal and addition of sialic acid residues on galactosyl (β-1,3) N-acetyl galactosamine, respectively [29, 30, 40]. These residues exist on terminal sugars of CD8, which could directly impact antigen reactivity by T cells [28, 41]. ST3Gal-II also has substrate specificity identical to that of ST3Gal-I, a related enzyme that reportedly controls CD8β sialylation in vivo [28]. We therefore characterized de-sialylated glycoproteins on CD8 T cells after treatment with neuraminidase or ST3Gal-II to determine whether CD8 was affected. Flow cytometry verified decreased PNA binding after ST3Gal-II and increased PNA binding after neuraminidase treatment, prior to further analysis (Supplemental Fig. 2A). PNA Western blotting revealed a prominent 35 kDa protein under reducing (i.e., disulfide-breaking) conditions, and a 75 kDa protein under nonreducing (i.e., disulfide-retaining) conditions, from CD8 T cell lysates (Supplemental Fig. 2B). This pattern was suggestive of a disulfide-linked heterodimer with identical mobility to CD8αβ. Addition of anti-CD8α mAb, and especially anti-CD8β mAb during PNA blotting, greatly decreased binding to the nonreduced dimer, further suggesting that it was CD8αβ (Supplemental Fig. 2B). Similarly, CD8 T cell lysates from human GBM patients showed prominent 35 kDa (reducing) and 75 kDa (nonreducing) proteins after neuraminidase treatment (Supplemental Fig. 2C, arrows). Immunodepletion with anti-CD8 mAb prior to PNA blotting identified the 35 and 40 kDa species as CD8α and CD8β, respectively (Supplemental Fig. 2D). Most importantly, neuraminidase treatment selectively increased PNA binding, and ST3Gal-II selectively decreased PNA binding to the smaller of these species (35 kDa β chain) in CD103⁺ CD8 T cell lysate. Since CD8β is an obligate component of the CD8αβ coreceptor, these data suggest that CD8αβ is a major de-sialylated glycoprotein on CD8 T cells that can be selectively modulated by ST3Gal-II or neuraminidase treatment.

De-sialylation enhances binding to tumor peptide/MHC I tetramers

CD8β de-sialylation greatly increases binding avidity of immature thymocytes for peptide-MHC I (pMHC I) tetramers [29, 30]. We therefore tested whether de-sialylation on peripheral human CD8 T cells impacted pMHC I tetramer binding in a similar manner. Indeed, intrinsically de-sialylated CD103⁺ CD8 T cells exhibited increased moderate avidity binding to Her-2, gp100 or Mart-1 tumor antigen pMHC I tetramers (Fig. 4, upper left and lower panels). Notably, moderate avidity binding to Her-2 by these cells was not increased by neuraminidase, consistent with maximal de-sialylation (Fig. 4, upper panel, middle), but ST3Gal-II treatment abrogated their binding without altering high avidity binding (Fig. 4, lower panels). Thus, intrinsic de-sialylation was necessary for moderate avidity binding of CD103⁺ CD8 T cells to pMHC I. Similarly, enzymatic de-sialylation with neuraminidase was sufficient to confer such moderate binding to Her-2 pMHC tetramer by CD103⁻ CD8 T cells (Fig. 4, upper panel, right).

Fig. 4.

Natural and enzyme-induced de-sialylation confers moderate avidity pHLA-A tetramer binding by human CD8 T cells. Analysis of pHLA-A tetramer reactivity by relatively pure CD103⁺ CD8 T cells isolated from healthy young donors, with or without enzymatic treatment as indicated. Moderate tetramer binding was dependent on natural or neuraminidase-induced de-sialylation, and was peptide independent. Representative profiles of ≥3 independent individuals are shown, and similar trends were observed in CD8 T cells from naive mice (not shown)

De-sialylation enhances IFN-γ production in response to pMHC I stimulation

To determine whether de-sialylation directly impacts T cell activation, we measured IFN-γ production by human CD103⁺ CD8 T cells or CD103⁻ CD8 T cells in response to stimulation with tumor pMHC I tetramer + anti-CD28 mAb. CD103⁺ CD8 T cells produced consistently higher levels of IFN-γ after 7 days than CD103⁻ CD8 T cells when stimulated by each of 4 tetramers (Fig. 5a, left). This hyper-responsiveness was most often decreased by ST3Gal-II and was more subtly affected by neuraminidase treatment (Fig. 5a, middle & right), paralleling the effects of these enzymes on tetramer binding (Fig. 4). In contrast, CD103⁻ CD8 T cells were generally hypo-responsive to tetramer stimulation in the absence of enzyme, or with ST3Gal-II pretreatment, but were most often induced to respond better to pMHC I by neuraminidase treatment (Fig. 5a, bottom). Moreover, in mixed CD103⁺ and CD103⁻ CD8 T cell stimulation cultures, neuraminidase increased proportions of CD103⁻ cells producing moderate levels of IFN-γ by about tenfold, while increasing high IFN-γ production by CD103⁺ cells more subtly (Fig. 5b). This suggests that CD103⁺ CD8 T cells respond better to pMHC I stimulation due to their intrinsic de-sialylation and that enzymatic de-sialylation improves responsiveness of CD103⁻ CD8 T cells. Thus, de-sialylation may be necessary and sufficient for robust responsiveness of human CD8 T cells to tumor pMHC I.

Fig. 5.

Surface de-sialylation is necessary and sufficient for enhanced reactivity to pMHC I stimulation by CD8 T cells. a Pure CD103⁺ CD8 T cells from young healthy donors or GBM patients, untreated or treated with neuraminidase or ST3Gal-II enzymes, were stimulated by addition of the indicated tumor peptide/HLA tetramers plus anti-CD28 mAb, followed by IFN-γ quantification by flow cytometry on gated CD103⁺ and CD103⁻ subpopulations 3 and 7 days later. Combined data from three independent donors and experiments is shown (*P < 0.5, **P ≤ 0.01, ***P < 0.005 by two-sided t test). Only CD103⁺ CD8 T cell data were available for a MART-1 tetramer (not shown), which exhibited a nonsignificant trend toward higher responses with neuraminidase (P = 0.06). b IFN-γ staining by gated CD8 T cells from GBM patients was plotted against CD103, 7 days after stimulation with gp100/HLA tetramer, demonstrating preferentially increased IFN-γ production in CD103⁻ CD8 T cells. Representative staining from three independent experiments is shown. c Untreated or ST3Gal-II pretreated CD8 T cells from young (6–8 weeks) C57BL/6 mice were stimulated with tumor pMHC I tetramers plus anti-CD28 mAb, and IFN-γ production assessed by gated CD103⁺ and CD103⁻ CD8 T cells 24 h later. (*P < 0.01 by t test; n = 6 with similar trends observed in ≥3 independent experiments)

Since PNA Western blotting indicated a potentially greater proportion of de-sialylated CD8 in mice, it was unclear whether sialylation would similarly limit responsiveness of mouse CD8 T cells to tumor pMHC I. Responsiveness of mouse CD8 T cells to pMHC I stimulation was consistently higher than in humans, but ST3Gal-II treatment significantly decreased the Trp-2 pMHC I tetramer response exclusively (Fig. 5c). Thus, robust responsiveness to the dominant pMHC I epitope of GL26 was also significantly inhibited by ST3Gal-II.

Beneficial impact of de-sialylation on GL26 hosts

To determine whether de-sialylation represents a viable target to enhance glioma immunotherapy, we exploited our novel observation that GL26 brain tumors grow as multiple, widely disseminated macroscopic tumor clusters (≥500 cells) in terminally symptomatic nude mice, but as well-delineated single tumors in wild-type mice (Fig. 6a). Importantly, this more invasive growth pattern in nude mice was significantly decreased by adoptive transfer of wild-type spleen cells and was maximally decreased by transfer of CD8 T cells, provided they were not first treated with ST3Gal-II (Fig. 6a). In addition, invasive GL26 growth comparable to that in nudes was evident in brains of CD103^−/− mice, which harbor an otherwise intact T cell compartment (Fig. 6a). We also assessed PNA mean fluorescence intensity on CD8 T cells of 10–12-week-old wild-type and CD103^−/− mice, which yielded values of 1,693 ± 250 and 2,178 ± 352, respectively (mean ± SEM; n = 6/strain; P = 0.29 by two-sided t test). Thus, CD103 expression itself did not significantly impact relevant sialylation status of CD8. This suggests that both de-sialylation and CD103 expression are required to effectively reduce invasive GL26 growth, and is consistent with CD103’s role in T cell localization to brain tumors [24, 25].

Fig. 6.

CD103⁺ CD8 T cells counteract malignancy in murine GL26 hosts. a Numbers of noncontiguous macroscopic GL26 tumors in the following groups: C57BL/6 or B6; nude (nu/nu); nu/nu + spleen; nu/nu + CD8 T; nude + CD8-ST3; and CD103^−/− (n ≥ 5 mice/group; ***P < 0.00002 relative to nu/nu; two-sided t test; results shown compiled from four independent experiments). b Survival between GL26-implanted C57BL/6 (n = 32), untreated nude mice (n = 28) and young nude mice adoptively transferred with untreated (n = 10) or ST3Gal-II-treated (yng spl-ST3; n = 8) splenocytes was assessed by log-rank statistics. C57BL/6 and nudes injected with untreated young splenocytes exhibited statistically indistinguishable survival, but survived significantly longer than untreated GL26-bearing nude mice, or GL26-bearing nude mice injected with ST3Gal-II-treated splenocytes (P < 0.008; Mantel–Cox log-rank results shown compiled from four independent experiments, each exhibiting qualitatively similar trends). c Meta-analysis of overall survival in glioblastoma patients receiving DC vaccinations from published clinical trials with immune response data. Patient survival curves were subcategorized by age (young = age ≤50 and old = age ≥50) and immune response after vaccination (*P < 0.045, **P < 0.003; Mantel–Cox log-rank). There is no significant difference in survival between either old and young responders, or young responders and nonresponders (both P = 0.07), or between old responders and young nonresponders (P = 0.466). d Relative mRNA expression of ST3Gal-II from young (n = 6) and old (n = 7) patient tumor samples (*P < 0.032, two-sided t test; GEO accession #GSE9166). A similar increase in ST3Gal-II expression using this probeset (205346_at) was observed in younger (n = 20) relative to older (n = 34) high-grade glioma patients from a public database (537 ± 44 vs. 438 ± 30; P = 0.04, two-sided t test; GEO accession #GSE4412; not shown)

We next exploited a previously reported survival delay between wild-type and nude GL26 hosts [32] (Fig. 6b). Similar to the invasion analysis, adoptive transfer of donor spleen cells into nude GL26 hosts significantly increased their survival, provided they were not first treated with ST3Gal-II (Fig. 6b). Thus, the lymphocyte-mediated survival enhancement was dependent on their de-sialylated status.

We reasoned that analysis of a larger number of vaccinated patients than currently exists in single trials would be required to determine the impact of CD103⁺ CD8 T cells and/or de-sialylation on human vaccine benefits. We therefore plotted survival of immune responder and nonresponder patients from all glioma DC vaccine trials whose enrollment was completed prior to 2007 (to exclude the potentially confounding impact of concomitant radiation/temozolamide therapy introduced at that time). We additionally segregated patient survival by age, since this is a dominant factor in glioma outcomes related to CD103⁺ CD8 T cells [11, 12, 42]. Although younger responder patients (age <50) had the longest survival of any group, this was not significantly different than young nonresponders (Fig. 6c). By contrast, survival of older responders was significantly longer than older nonresponders (Fig. 6c). This suggests that younger glioma patients may benefit proportionally less from immunologically successful vaccination than older ones. This prompted us to analyze the expression of 15 known immunosuppressive genes in young and old glioma patients (see “Materials and methods”). In 13 of our patients, only ST3Gal-II was both inversely correlated with age (r = −0.7568; P = 0.003) and significantly higher in younger patients (Fig. 6d). There was also a trend for decreased ST3Gal-II expression in vaccine responders versus nonresponders (129 ± 97 vs. 176 ± 53, respectively), but sample size was not large enough to reveal significant differences (n = 4; P = 0.214, 1-sided T test).

Together, these data suggest that T cell activity benefits glioma hosts and that this beneficial activity is dependent on CD8β de-sialylation coupled with CD103 expression. Thus, augmenting either CD103⁺ CD8 T cell responses or CD8 T cell sialylation status may improve responsiveness and clinical benefits of cancer immunotherapy in the context of glioma vaccines.

Discussion

Previously, we have shown that CD103⁺ CD8 T cells dominate potentially beneficial immune responses in glioblastoma patients [14], but the therapeutic mechanism exerted by CD103⁺ CD8 T cells remained unknown. In this context, it was initially perplexing that the pre-vaccine IFN-γ response was negatively associated, while the post-vaccine response was positively associated, with CD103 status. One possibility is that CD103 expression in patients is expressed on distinct reactive subpopulations of CD8 T cells with and without vaccination. In direct support of this, T cell receptor excision circle (TREC) concentrations, an independent metric of constitutively (as opposed to inducibly) CD103⁺ CD8 T cells (recent thymic emigrants or RTEs), correlated positively only with post-vaccine responses in GBM patients (r = 0.630; n = 15; P = 0.02; also see [14]), and inversely but nonsignificantly with both pre-vaccine IFN-γ responses and survival. In contrast, both endogenous and vaccine-induced antitumor antigen responses increased GL26 host survival in mice and were mediated by TREC-enriched CD8 T cells (6,139 ± 706 for Trp2-reactive CD8 T cells vs. 3,459 ± 488 for other CD8 T cells; n = 7; P = 0.009 by two-sided t test), providing an explanation for activated (CD69⁺) CD103⁺ CD8 T cells in brains of unvaccinated T cell-sufficient GL26 hosts as well (Fig. 1). Thus, pre-vaccine responding CD103⁺ CD8 T cells in GBM patients specifically appears to represent a distinct subpopulation from those responding after vaccine, the latter of which are enriched for RTEs.

In our study, we further confirm that CD103⁺ CD8 T cells dominate in vivo immune responses in the GL26 syngeneic murine glioma model and go on to demonstrate that self-tumor antigen-responsive CD103⁺ CD8 T cells exhibit relative de-sialylation, allowing for a broader and more robust immune response toward tumor antigens. While nonself recall antigen-responsive CD8 T cells were also CD103⁺, they were not dependent on de-sialylation, consistent with their representing a distinct, non-RTE CD103⁺ subpopulation. We also show that enzymatic de-sialylation of CD8β on CD8 T cells is sufficient to increase CD103⁻ CD8 T cell tumor antigen responsiveness.

De-sialylation of the CD8β chain allows for the high levels of PNA binding observed in immature double positive thymocytes [16, 17]. PNA is known to bind quantitatively to de-sialylated galactosyl (β-1,3) N-acetyl galactosamine on cell surface molecules including CD8 [31]. De-sialylated CD8β can lower the threshold for T cell activation and broaden the range of antigen activity [16]. Consistent with this, we show that enzymatic de-sialylation of CD8 T cells with neuraminidase enhances tumor antigen binding and tumor antigen-induced IFNγ production. In contrast, CD8 T cells treated with ST3Gal-II sialotransferase exhibit dampened pMHCI binding. Using PNA Western blotting, we were able to show that CD8β and CD8αβ dimers are the major species of de-sialylated glycoprotein on mouse CD8 T cells and were prominent species de-sialylated by neuraminidase in human CD8 T cells. On human T cells, CD8β was only one of the several glycoproteins weakly bound by PNA. Neuraminidase pretreatment rendered CD8αβ dimers as the prominent PNA-binding glycoprotein on these cells.

In this context, we show that CD103⁺ CD8 T cells have higher PNA binding than CD103⁻ CD8 T cells, suggesting a potential difference in their effector capabilities. A larger proportion of CD103⁺ CD8 T cells consistently produced IFN-γ in response to stimulation by various distinct tumor antigen tetramers than CD103⁻ CD8 T cells, and in addition, CD103⁺ CD8 T cells were reactive to multiple antigens. This suggests greater promiscuity in antigen binding and responsiveness, which may reflect the stronger influence of CD8, the peptide-independent component of the pMHC I-binding complex on T effectors. Interestingly, neuraminidase treatment was able to enhance CD103⁻ CD8 T cell antigen reactivity, while CD103⁺ CD8 T cell reactivity was unaffected. This is consistent with CD8β de-sialylation being the mechanism of enhanced antigen reactivity on CD103⁺ CD8 T cells.

Using the murine syngenic glioma model, GL26, we investigated the in vivo effects of CD103 expression and CD8 T cell sialylation on host survival. Transfer of splenocytes into nude mice increased survival of tumor bearing hosts, while transfer of ST3Gal-II pretreated splenocytes showed no survival advantage. Using this same model, we also show that CD103 expression is necessary for decreasing tumor invasiveness, with CD103^−/− mice having increased numbers of noncontiguous tumors when compared to C57Bl/6 and splenocyte transferred nude mice. This evidence links both sialylation status and CD103 to beneficial anti-glioma immune responses. In keeping with this, we also show from compiled published trials that younger glioma patients benefited proportionally less from immunologically productive vaccination than older patients. This accompanied a specific increase in ST3Gal-II expression in young glioma patients, suggesting that tumor-derived enzymes capable of sialylating CD8 may underlie clinical resistance to vaccine therapy even when objective immune responses are achieved. Although not definitive and prone to confounding influences such as differences in vaccine preparation and immune monitoring, this meta-analysis raises the intriguing possibility that tumor production of a CD8-reactive sialotransferase may contribute to the clinical failure of otherwise successful vaccine therapy.

Our findings clarify several previous observations. First, they provide a mechanistic basis for the reported dominance of CD103⁺ CD8 T cells in immune and clinical responses within glioma patients [14, 25]. Second, our studies reveal a link between the ability of surface de-sialylation to enhance tumor antigen responsiveness by peripheral T cells [39] and developmentally regulated CD8β de-sialylation [29, 30]. Additionally, our studies suggest that CD103⁺ CD8 T cells are hyper-responsive to tumor antigen-MHC I due to CD8β de-sialylation, which allows individual T cells to bind and potentially respond to multiple tumor antigen-MHC I. Since this suggests that de-sialylation may increase CD8 T cell antigen cross-reactivity, further studies are needed to determine whether de-sialylation-dependent responses may be involved in potentially detrimental autoimmune reactions against self-proteins as well.

The potential clinical relevance of our findings is that de-sialylation may prove useful in inducing CD8 T cell responses to antigen epitopes that normally fail to elicit a response. In this context, tumor antigen peptides are notoriously difficult to mount robust type I responses against, particularly in human glioma patients [43]. Thus, de-sialylation may represent a viable means to augment glioma immunotherapy, perhaps in the form of ex vivo enzymatic de-sialylation and reinfusion of CD8 T cells prior to vaccination. The identification of additional mammalian sialidases with potentially higher specificity for CD8 might also prove useful as agents of in vivo de-sialylation. Given our current data on ST3Gal-II expression in human gliomas, the question of whether human gliomas can acquire the capacity to undermine CD8β de-sialylation should also be addressed as a possible mechanism of adaptive immune evasion by gliomas. Conversely, sialylation of patient cells with ST3Gal-II or related enzymes might be useful as a means to inhibit undesirable CD8 T cell responses after transplantation, or as a treatment for autoimmune disorders.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

CNS

Central nervous system

DC

Dendritic cell

GBM

Glioblastoma multiforme

IFN-γ

Interferon-gamma

IL-12

Interleukin-12

MHC I

Major histocompatibility complex class I

PNA

Peanut agglutinin

RTE

Recent thymic emigrant

TCR

T cell receptor