Glycolipid-peptide conjugate vaccines enhance CD8+ T cell responses against human viral proteins.

An important goal of vaccination against viruses and virus-driven cancers is to elicit cytotoxic CD8+ T cells specific for virus-derived peptides. CD8+ T cell responses can be enhanced by engaging help from natural killer T (NKT) cells. We have produced synthetic vaccines that induce strong peptide-specific CD8+ T cell responses in vivo by incorporating an NKT cell-activating glycolipid. Here we examine the effect of a glycolipid-peptide conjugate vaccine incorporating an NKT cell-activating glycolipid linked to an MHC class I-restricted peptide from a viral antigen in human peripheral blood mononuclear cells. The vaccine induces CD1d-dependent activation of human NKT cells following enzymatic cleavage, activates human dendritic cells in an NKT-cell dependent manner, and generates a pool of activated antigen-specific CD8+ T cells with cytotoxic potential. Compared to unconjugated peptide, the vaccine upregulates expression of genes encoding interferon-γ, CD137 and granzyme B. A similar vaccine incorporating a peptide from the clinically-relevant human papilloma virus (HPV) 16 E7 oncoprotein induces cytotoxicity against peptide-expressing targets in vivo, and elicits a better antitumor response in a model of E7-expressing lung cancer than its unconjugated components. Glycolipid-peptide conjugate vaccines may prove useful for the prevention or treatment of viral infections and tumors that express viral antigens.

the same antigen 12,13 , while the latter approach may be constrained by difficulties identifying shared CD4 + and CD8 + epitopes given the extreme polymorphism of the major histocompatibility complex (MHC) in humans 14 .
Natural killer T (NKT) cells are a class of T cells expressing canonical T cell receptors that recognize glycolipid antigens presented by the non-classical MHC molecule, CD1d. NKT cells exist in a semi-activated state capable of responding rapidly to antigenic stimulation 15 . APCs express CD1d and are capable of processing and presenting glycolipids, such as the potent synthetic antigen α-galactosylceramide (α-GalCer), to NKT cells 16 . Similar to classical CD4 + T cell help, NKT cell help promotes dendritic cell (DC) licensing, maturation and the production of pro-inflammatory cytokines, e.g., IFN-γ and IL-12, which enhance CD8 + T cell responses against co-presented peptide antigens 17 . This help depends on peptide and glycolipid presentation by the same APC, and in vivo requires an interaction between CD40 and CD40L 18,19 . Potential advantages of exploiting NKT rather than conventional CD4 + T cell help in a clinical context include avoiding the need to select adjuvants according to MHC class II expression 20 , and eliciting a CD8 + T cell response with a distinct chemokine receptor profile 21,22 . In mouse models, NKT cell activation at the time of vaccination or infection promotes virus-specific CD8 + T cell memory 23,24 . Although there is abundant evidence of NKT cell adjuvant activity in murine models in vivo, evidence for an adjuvant effect in humans, who have much lower NKT cell frequencies, is much more limited.
We have previously synthesized glycolipid-peptide conjugate vaccines comprising an α-GalCer prodrug covalently linked to an antigenic peptide via a self-immolating para-aminobenzyl (PAB) linker. These vaccines were designed to co-deliver CD8 + T cell peptide epitopes and α-GalCer to the same APC and elicit functional peptide-specific CD8 + T cell responses in mouse models of allergic airway inflammation and B.16 melanoma 25,26 . We have also shown that a glycolipid-peptide conjugate vaccine incorporating α-GalCer with an immunodominant HLA-A*02-restricted peptide from the cytomegalovirus (CMV) pp65 protein can stimulate human NKT cells and peptide-specific CD8 + T cells 25 .
Here we investigate the activity and mechanism of a glycolipid-peptide conjugate vaccine that elicits virus-specific CD8 + T cell responses in human PBMCs. We show that this vaccine activates human DCs and CD8 + T cells in a manner dependent on NKT cells, and that the CD8 + T cells elicited by this vaccine are peptide-specific and have cytotoxic potential. Furthermore, we show that a vaccine of the same design but incorporating a human papilloma virus (HPV) E7 peptide is capable of delaying tumor growth in an in vivo mouse model of E6/ E7-expressing lung cancer.

Results
Glycolipid-peptide conjugate vaccine requires cathepsin cleavage and induces CD1d-dependent NKT cell proliferation. The glycolipid-peptide conjugate vaccine α-GalCer-pp65 495-503 (Fig. 1A) consists of a pro-drug form of the glycolipid α-galactosylceramide (α-GalCer), which readily reverts to its more stable N-acyl form under physiological conditions 25 , linked via a cathepsin-B-cleavable linker to the peptide sequence FFRK-NLVPMVATV (here termed pp65 495-503 ), which contains a HLA-A*02-restricted epitope from cytomegalovirus (CMV) pp65 protein. CD8 + T-cells specific for NLVPMVATV can be readily detected in PBMCs from HLA-A*02 + CMV-seropositive healthy donors using loaded MHC class I multimers 27 . The peptide sequence incorporates the cleavage sequence FFRK at the N-terminus to promote proteolytic generation of the NLVPMVATV epitope within APCs 28 .
To show that conjugate vaccine must first be cleaved into its active components in order to stimulate NKT cells, α-GalCer-pp65 495-503 and free α-GalCer were pre-treated with cathepsin-B or PBS control, and loaded onto plate-bound mouse CD1d monomers. Unlike free α-GalCer, α-GalCer-pp65 495-503 required pre-treatment with cathepsin-B in order to stimulate IL-2 production by the mouse hybridoma NKT cell line DN32.3, indicating that the α-GalCer-pp65 495-503 vaccine requires proteolytic processing to produce free α-GalCer capable of activating NKT cells (Fig. 1B).
We have previously shown that α-GalCer-pp65 495-503 is able to induce IFN-γ production and CD137 up-regulation on human NKT cells 25 . To determine whether α-GalCer-pp65 495-503 can also induce proliferation of NKT cells, PBMCs derived from an HLA-A*02-negative donor were cultured in the presence of equimolar concentrations of α-GalCer or α-GalCer-pp65 495-503 conjugate. Quantification of NKT cells (as a % of total CD3 + cells) showed that addition of α-GalCer-pp65 495-503 induced NKT cell expansion in a dose-dependent manner, although overall expansion was lower with the vaccine than with free α-GalCer (Fig. 1C). Similarly, intracellular staining using anti-Ki67 showed proliferation of NKT cells in response to both α-GalCer-pp65 495-503 and to free α-GalCer, which could be abolished by addition of an anti-CD1d antibody (Fig. 1D). As expected, the peptide alone did not trigger NKT cell proliferation above the level of the media-only control. Finally, interferon (IFN)-γ ELISpot demonstrated that α-GalCer-pp65 495-503 induced IFN-γ production, and that this was blocked by anti-CD1d (Fig. 1E). Taken together, these data demonstrate that the glycolipid-peptide conjugate vaccine α-GalCer-pp65 495-503 induces proliferation and activation of human NKT cells in a CD1d-dependent manner.
After seven days in culture NLV-specific CD8 + T cells expanded with α-GalCer-pp65 495-503 showed higher expression of the T cell activation marker CD137 (4-1BB), a surface glycoprotein belonging to the tumor-necrosis factor receptor superfamily (TNFRSF) 32,33 , compared to cells treated with either peptide alone or admixed peptide and α-GalCer (Fig. 2D,E). Binding of CD137 to its ligand 4-1BBL promotes increased T cell proliferation, cytokine production, functional maturation, and prolonged survival of CD8 + T cells [34][35][36] . Interestingly, at earlier time-points CD137 expression was also elevated in response to admixed peptide and α-GalCer as well as peptide alone, however, this elevated expression was only sustained on NLV-specific CD8 + T cells expanded with α-GalCer-pp65 495-503 , suggesting that conjugation modifies the T cell response to peptide or affects the kinetics of peptide presentation (Fig. 2F).
To determine whether NKT cells are required for the CD8 + T cell response to α-GalCer-pp65 495-503 , NKT cells were depleted from whole PBMCs derived from three separate donors. Compared to mock-depleted controls, NKT depletion led to a significant reduction in the proportion of CD137 + NLV-specific T cells in response to α-GalCer-pp65 495-503 (Fig. 2G), indicating that full vaccine-induced activation of peptide-specific CD8 + T cells requires NKT cells.
Glycolipid-peptide conjugate elicits peptide-specific CD8 + T cells with cytotoxic potential. To assess the functional outcome of α-GalCer-pp65 495-503 treatment, intracellular staining for LAMP-1 (CD107a), a marker of degranulation of cytotoxic molecules, and the pro-inflammatory cytokine IFN-γ was performed on NLV-specific CD8 + T cells. Both LAMP-1 and IFN-γ were expressed to a higher degree on NLV-specific CD8 + T cells expanded with α-GalCer-pp65 495-503 , compared to those exposed to peptide alone, or to admixed peptide and α-GalCer (Fig. 3A,B). In contrast, α-GalCer-pp65 495-503 did not induce expression of LAMP-1 or IFN-γ on the dextramer negative CD8 + T cell population within the cultures (data not shown), indicating that the induction of these molecules is restricted to antigen-specific T cells.
Taken together, these data indicate that the α-GalCer-pp65 495-503 glycolipid-peptide conjugate vaccine can elicit peptide-specific T cells with cytotoxic potential in human PBMCs.
To assess functional activity, α-GalCer-E7 49-57 vaccine was administered to animals implanted with cells from the TC-1 tumor cell line, which expresses HPV16-E6 and -E7 proteins. A single dose of α-GalCer-E7 49-57 vaccine eight days after tumor inoculation significantly delayed tumor growth relative to saline-or unconjugated component-treated control animals. This activity was improved further with a second dose of vaccine 7 days after the first. The antitumor activity was dependent on the vaccine design, with antigen and adjuvant covalently linked together, as injection of unconjugated components alone or as an admix did not provide antitumor activity (Fig. 4C).

Discussion
We have previously shown that synthetic vaccines that combine an NKT-activating glycolipid with an immunodominant antigenic peptide are effective against a model antigen (ovalbumin) in vivo, and that a similar vaccine can activate human CMV-specific CD8 + T cells in vitro 25,26 . He we report that the activation of human virus-specific CD8 + T cells induced by these vaccines is NKT cell-dependent and peptide-specific, and that the vaccines activate human antigen-presenting cells in the presence of NKT cells, elicit peptide-specific CD8 + T cells with cytotoxic potential, and induce cytotoxicity against a clinically-relevant viral oncoprotein in vivo.
To-date, most pre-clinical studies using NKT cell agonists to drive peptide-specific T cell or antibody responses have been performed in mice, which have much higher circulating NKT cell frequencies than most humans 37 . Although human NKT cell frequencies typically exceed those of a CD4 + T cell of a given peptide specificity 38,39 , the reduced NKT cell niche has been raised as a barrier to generating effective immunotherapies utilizing these cells 40 . The current study provides "proof of principle" that human NKT cells can be successfully recruited to augment cytotoxic CD8 + T cell responses, despite their low frequency.
An important component of the α-GalCer-pp65 495-503 vaccine is the cathepsin-B-sensitive linker, the removal of which permits an O→N acyl migration to form α-GalCer. As the glycolipid component of the vaccine must reach the late endosomal or lysosomal compartment for loading onto CD1d 41 , the requirement for this extra processing step may explain why α-GalCer-pp65 495-503 was a less effective NKT cell agonist than free α-GalCer in vitro. However, given that repeated systemic α-GalCer injections have been shown to drive long-term NKT cell functional anergy arising from the development of an IL-10-secreting NKT cell subset (NKT 10 ) 42,43 , this reduced activity may be advantageous in vivo at preventing NKT cell over-activation from stymying the downstream CD8 + T cell response or inducing tissue damage in the liver and other tissues where NKT cells are prevalent. Crucially, this study uses 'real-life' viral antigens, rather than a model antigen, such as ovalbumin. The CMV epitope was selected based on the ability to detect CD8 + T cells specific for this virus-derived peptide at a high frequency in a large proportion of human donors 27 . Due to the persistent nature of CMV infection, CMV-reactive T cells have an altered phenotype compared with other virally-reactive or naïve tumor-associated antigen (TAA)-reactive T cells 44 . For example, CMV-specific CD8 + T cells are chronically activated and express the T H 1-associated transcription factors T-bet (TBX21) and eomesodermin (EOMES), as well as IFNG mRNA and IFN-γ-regulated genes 45 . We cannot therefore be certain that the responses generated against the immunodominant CMV antigen will extend to other viral-or tumor-associated peptides, although it is reassuring that we observe CD8 + T cell responses to an immunodominant influenza matrix protein epitope using a similar vaccine.
The human cell in vitro culture system we used showed no difference in the frequency of NLV-specific CD8 + T cells between cultures treated with α-GalCer-pp65 495-503 , its admixed components or peptide alone, and while we demonstrate upregulation of molecules associated with cytotoxic function in response to the conjugate vaccine, in vitro killing of peptide-loaded targets is not demonstrated here. In vitro immunological studies are limited by the lack of a tertiary lymphoid structure that facilitates juxtaposition of the various immune cells involved 46 . Additional challenges to working with primary human cells in vitro, include: (1) the restricted number of professional, cross-presenting APCs and NKT cells present in each well; (2) the enforced proximity inherent in cell culture conditions, which all but ensures co-delivery of antigen and adjuvant to the same APC, regardless of conjugation; and (3) the substantial degree of heterogeneity between humans, which makes it difficult to detect differences within a small donor pool. Differences between peptide and conjugated vaccines may also be understated in in vitro studies due to the role of biodistribution in determining the nature and magnitude of the downstream immune response. For example, α-GalCer is known to bind apolipoprotein E (ApoE), enabling it to be efficiently acquired by DCs, via low-density lipoprotein (LDL) receptors 47 . Notably our data do show that a similar conjugate vaccine can induce cytotoxicity and delay tumour growth in vivo.
Oncogenic viral antigens are likely to be excellent targets for immunotherapeutic vaccine strategies aimed at reinforcing tumor-specific T-cell responses 48 . A recent clinical study in patients with HPV16 + high-grade vulvar intraepithelial neoplasia showed that vaccination with peptides in montanide, an oil emulsion that improves antigen uptake, could induce partial or complete histological regression of lesions in more than 50% of patients treated 49 . Importantly, this trial reported that clinical efficacy was related to the strength of vaccine-induced immune responses. It is conceivable that the outcome could be improved further with a vaccine that is designed to improve the function of APCs, as has been described here. The antitumor response reported here in mice against a tumor expressing the HPV16 E7 protein is notable both for the low dose of vaccine need for activity, and the fact  [49][50][51][52][53][54][55][56][57] peptide-loaded splenocytes was assessed in vivo 10 days after vaccination. ****p < 0.0001; **p < 0.01, one-way ANOVA with Tukey's multiple comparisons (C) 2.5 × 10 5 HPV16-E6 and -E7 protein-expressing TC-1 cells were implanted subcutaneously into C57/BL6 mice (n = 5 per group). On day 8, E7 49-57 peptide, α-GalCer, admix or α-GalCer-E7 49-57 were administered intravenously; one group was left untreated, and one group received a second dose of α-GalCer-E7 49-57 after a further seven days. Tumor size was determined as the product of the two diameters. *p < 0.0001 (difference between treatment group, and each of the four non-conjugate vaccine groups) by one-way ANOVA.
that activity was reliant on conjugation of the glycolipid and peptide components. Studies of this vaccine design with other tumor-associated antigens, and in models of viral infection, are ongoing.
In conclusion, we show that in human PBMCs a glycolipid-peptide conjugate vaccine can harness NKT cells as cellular adjuvants to enhance cytotoxic CD8 + T cell responses against virus-derived peptides. We have demonstrated that a similar conjugate vaccine directed against an oncogenic viral protein is capable of generating a significantly better anti-tumor response than its admixed components in a model of HPV-associated cancer in vivo. Human trials will be necessary to determine whether glycolipid-peptide conjugates can safely enhance peptide-specific CD8 + T cell responses with clinical benefit.

Methods
Conjugate vaccine synthesis. Conjugate vaccines were synthesized per previously reported protocols 25 .

Solubilization and administration of compounds for biological studies. Solubilization of α-GalCer
and vaccines was achieved by freeze-drying the samples in the presence of sucrose, L-histidine, and Tween 20 as previously described 51 . All solubilized compounds were diluted in sterile water before use in vitro.
Isolation of human PBMCs. Venous blood was drawn into heparin-containing tubes, diluted 1:1 in phosphate buffered saline (PBS; Gibco), and PBMCs isolated by density centrifugation (Lymphoprep ™ ; Axis-Shield, Oslo, Norway). PBMCs were washed twice and either used fresh or cryopreserved in 10% dimethyl sulfoxide (Sigma-Aldrich) and 90% fetal bovine serum (FBS; Gibco) and thawed immediately before use. All donors gave written informed consent; this study was approved by the Human Ethics Committee of Victoria University, and all experiments were performed in accordance with relevant guidelines and regulations.
Cell-free CD1d presentation assay. Dependence of NKT cell activation on proteolytic cleavage of α-GalCer-pp65 495-503 was tested in a cell-free assay. Flat-bottomed 96-well tissue culture plates were coated with 5 μg/mL mouse CD1d monomers (NIH Tetramer Core Facility) overnight at 4 °C. Equimolar concentrations of the vaccine α-GalCer-pp65 495-503 and free α-GalCer were pre-treated with cathepsin-B (Sigma-Aldrich) or mock-treated with PBS for 24 h, at 37 °C, before the equivalent of 50 ng/digest was added to the coated plate for 3 h, at 37 °C. Plates were then washed with Iscove's Minimum Essential Medium (IMDM; Gibco). DN32.D3 NKT hybridoma cells 52 were seeded at a density of 3 × 10 4 cells/well and the plate was incubated at 37 °C in IMDM supplemented with 5% fetal bovine serum (FBS) (Sigma-Aldrich), 2 mM glutamax, 100 U/mL penicillin, 100 mg/mL streptomycin and 50 mM 2-mercaptoethanol (all Invitrogen). The following day, supernatants were collected and IL-2 concentration determined by ELISA (BioLegend), according to the manufacturer's instructions.
Generation of monocyte-derived (mo)DC. Immature DC were generated from human PBMCs over 6 days from adherent monocytes using RPMI (Gibco) supplemented with 2% autologous human serum containing 1000 U/mL GM-CSF and 1000 U/mL IL-4. Cytokines were re-added after 3 days in culture. For co-culture experiments to assess DC activation by flow cytometry, 5 × 10 4 DCs/well were cultured in low adherence 96-well plates (Nunc) with 5 × 10 3 flow-sorted human NKT cells/well in the presence of 1 µM of α-GalCer-pp65 495-503 or α-GalCer. For intranuclear staining of NKT cells with anti-Ki67 (Fig. 1D), following live/dead and surface Ab staining, cells were permeabilized and fixed in FoxP3 Perm/Fix buffer (BioLegend) at room temp for 20 min. Cells were washed and re-suspended in FoxP3 Perm buffer (BioLegend) for anti-Ki67 staining for 30 min at room temperature, then washed 2× in flow buffer before analysis.
ELISpot assay. IFN-γ production was quantified after 24 h using a human IFN-γ ELISpot kit (Mabtech), according to manufacturer's instructions. Briefly, Millipore MultiScreen-HA 96-well filter plates (Millipore) were coated with 5 μg/mL IFN-γ mAb in PBS overnight at 4 °C. Cells were seeded at 2 × 10 5 cells/well in cIMDM and incubated with the relevant compounds overnight. After washing with PBS, plates were incubated with biotin-labeled anti-IFN-γ for 1 h at room temperature. Plates were then washed and treated with 1:1000 dilution of streptavidin-alkaline phosphatase (Sigma) for 30 min at room temperature. Plates were developed with nitro-blue tetrazolium and 5-bromo-4-chloro-3-indolyl-phosphate substrate (Mabtech) until all spots were clearly visible. Developed plates were dried and counted on an automated ELISpot reader (Autoimmun Diagnostika, Strassberg, Germany). Cells treated with PHA were used as a positive control.

Depletion of NKT cells.
PBMCs from HLA-A*02 positive CMV-seropositive donors were incubated at 4 °C for 10 min with anti-Vα24 Jα18 TCR biotinylated antibody (clone 6B11; eBioScience). Cells were then washed twice in Würzburger buffer and incubated at 4 °C, gently shaking, with twice the recommended concentration of biotin binding beads (Dynabead; Invitrogen) for 20 min. Unbound cells were isolated using a DynaMag-15 magnet (ThermoFisher). This process was repeated three times to ensure >99% of NKT cells were depleted from the PBMCs.

RNA isolation and Nanostring analysis.
After seven days in culture as detailed above ("Stimulation of human CMV-specific CD8 + T cells in vitro"), RNA was extracted from PBMCs derived from four donors using the RNeasy Micro Kit (Qiagen) according to the manufacturer's instructions. RNA purity and integrity were assessed using a Nanodrop spectrophotometer (ThermoFisher) and gel electrophoresis. Nanostring RNA analysis of 700 immune-related genes was performed using the nCounter GX Human PanCancer Immune profiling Kit (XT) on the nCounter ® Analysis System. Data were analyzed using the nSolver Analysis software package and Microsoft Excel. Default normalization settings were used for all data. The online tool CIMminer (https://discover.nci.nih. gov/cimminer/home.do) was used to generate heat maps.

Analysis of anti-tumor activity in vivo.
The mouse lung cancer cell line TC-1, a mouse lung epithelial cell line transformed with HPV16-E6 and -E7 genes 53 , was cultured in RPMI1640 supplemented with 10% FBS, 100 U/ml penicillin, 100 mg/ml streptomycin (all from Gibco), and 500 mg/ml G418 (Sigma), and resuspended in PBS for subcutaneous injection. Groups of naive C57BL/6 mice (n = 5) were implanted with 2.5 × 10 5 cells in the flank. On day 8, when tumors reached at least 5mm in diameter, mice were given one intravenous treatment with either 1.5 µg (0.47 nmol) of α-GalCer-E7 49-57 conjugate vaccine, an admix of 100 ng (0.09 nmol) HPV16 E7 49-57 peptide and 5 nmol of α-GalCer, HPV16 E7 49-57 peptide alone or α-GalCer alone; the concentrations of α-GalCer and peptide used were selected as the optimal concentrations based on in vivo titrations. Some mice received two doses ScIenTIFIc RePORTS | 7: 14273 | DOI:10.1038/s41598-017-14690-5 of the α-GalCer-E7 49-57 with a week interval. A control group was injected with PBS. Tumor growth was monitored three times per week, with tumor size calculated as the product of the two bisecting diameters. Measurements were stopped for each group when the first mouse developed a tumor exceeding 400 mm 2 . The C57BL/6 mice used were purchased from Animal Production Colonies, Frederick Cancer Research Facility, National Cancer Institute. All animal experiments were approved by the Animal Care and Use Committee of the National Cancer Institute, and all experiments were performed in accordance with relevant guidelines and regulations.
Statistical analysis and data availability. All statistical analysis was performed using GraphPad Prism software (GraphPad Prism Inc). For Fig. 1B,D and 2F, two-way analysis of variance (ANOVA) was performed before the Bonferroni multiple comparisons test. For Fig. 1E, the Mann-Whitney U test was used. For Fig. 2B, D, the Friedman test was used, with Dunn's multiple comparisons. In all experiments, P values ≤ 0.05 were taken to be significant. The data generated during the current study are available from the corresponding author on reasonable request.