Detecting Tumor Antigen-specific T cells via Interaction Dependent Fucosyl-biotinylation

Summary

Re-activation and clonal expansion of tumor-specific antigen (TSA)-reactive T cells are critical to the success of checkpoint blockade and adoptive transfer of tumor-infiltrating lymphocyte (TIL) based therapies. There are no reliable markers to specifically identify the repertoire of TSA-reactive T cells due to their heterogeneous composition. We introduce FucoID as a general platform to detect endogenous antigen-specific T cells for studying their biology. Through this interaction dependent labeling approach, intratumoral TSA-reactive CD4⁺, CD8⁺ T cells and TSA-suppressive CD4⁺ T cells can be detected and separated from bystander T cells based on their cell-surface enzymatic fucosyl-biotinylation. Compared to bystander TILs, TSA-reactive TILs possess a distinct TCR repertoire and unique gene features. Though exhibiting a dysfunctional phenotype, TSA-reactive CD8⁺ TILs possess substantial capabilities of proliferation and tumor-specific killing. Featuring genetic manipulation-free procedures and a quick turnover cycle, FucoID should have the potential of accelerating the pace of personalized cancer treatment.

Graphical Abstract

In brief

FucoID enables identification of endogenous tumor antigen-specific T cells based on an interaction dependent fucosyl-biotinylation.

Introduction

In the past decade, the development of immune checkpoint inhibitors and adoptive cell transfer (ACT)-based therapies has revolutionized cancer treatment (Chowdhury et al., 2018; Guedan et al., 2018; Rosenberg and Restifo, 2015; Sharma and Allison, 2015). The success, however, is limited to a relatively small subset of patients and cancer types (Galon and Bruni, 2020; Sanmamed and Chen, 2018; Yamamoto et al., 2019). Successful anti-tumor immune responses following immunotherapy are believed to require re-activation and clonal expansion of tumor-specific antigen (TSA)-reactive T cells present in the tumor microenvironment (Gubin et al., 2014; McGranahan et al., 2016; Ott et al., 2017; Rizvi et al., 2015; Sahin et al., 2017; Schumacher and Schreiber, 2015). The unpredictability of a patient’s response to immunotherapy is partially attributed to the heterogeneity of the tumor immune microenvironment and phenotypic profiles of TILs within individual tumors (Chevrier et al., 2017; Lavin et al., 2017; Rizvi et al., 2015; Stevanović et al., 2017). Previous studies revealed that TILs consist of not only those T cells specific for TSAs, but also the ones that recognize epitopes unrelated to the tumor, such as virus-specific antigens, known as bystander T cells (Scheper et al., 2019; Simoni et al., 2018). Therefore, identifying TSA-reactive T cells from cancer patients has critical implications for prediction and therapeutic applications. Although tumor-reactive TIL candidates can be roughly enriched from tumor digests through the expression of cell-surface markers (e.g. PD-1) (Gros et al., 2014; 2016; Yossef et al., 2018), bystander T cells expressing such makers also exist in the tumor microenvironment (Duhen et al., 2018; Sade-Feldman et al., 2018; Scheper et al., 2019; Simoni et al., 2018).

The precise therapeutic effect of immunotherapies that boost immunity of endogenous T cells is governed by the T cells’ capability to recognize TSAs (Schumacher and Schreiber, 2015). At the molecular level, this is determined by the interaction of unique T cell receptor (TCRs) with cognate peptide–major histocompatibility complexes (pMHCs) (Dembić et al., 1986). Designed based on this knowledge, peptide–MHC (pMHC) multimers are widely used to profile TCR specificity of a known antigen and to identify TILs specific to a particular TSA (Cohen et al., 2015; Echchakir et al., 2002; Glanville et al., 2017; Newell et al., 2013; Robbins et al., 2013; Yamamoto et al., 2019). To identify TSAs for the pMHC approach, however, hinges on reverse immunology, in which whole exome sequencing (WES) is performed on tumor cells to uncover nonsynonymous mutations (Robbins et al., 2013; Schumacher and Schreiber, 2015). In silico tools are then used to generate peptides harboring epitopes encoded by these identified nonsynonymous mutations. Such peptides are either left unfiltered, filtered through the use of prediction algorithms for MHC-binding, or used as guides for identifying MHC-associated TSAs by combining with mass spectrometry analysis (Abelin et al., 2017; Yamamoto et al., 2019). Such methods have enabled the identification of TSAs and TSAs-reactive TCRs for melanoma patients and a small number of patients with epithelial cancers (Stevanović et al., 2017; Tran et al., 2016; Zacharakis et al., 2018). Despite these acclaimed successes, a large portion of identified TSA candidates is not immunogenic because available computational tools cannot accurately predict T-cell reactivity (Editorial, 2017). Requiring deep sequencing, bioinformatics analysis and machine learning, these methods heavily rely on expertise in computational biology, and, as a result, turnover times are long (Arnaud et al., 2020; Editorial, 2017; Lee et al., 2018; Liu and Mardis, 2017). Therefore, to accelerate the pace of cancer immunotherapy a computation-free method that enables quick TSA-reactive T cell discovery and easy adoptability would be a much-needed advance.

Here we report the development of a method for the identification of antigen-specific T cells based on an interaction dependent fucosyl-biotinylation and the use of this method to isolate endogenous tumor antigen-specific T cells from tumor digests without the previous knowledge of the TSA identities. In this approach, which we termed FucoID, tumor-lysate primed DCs presenting TSAs, serving as “living” tetramers, are equipped with an enzyme that induces proximity-based transfer of fucosylated biotin (Fuc-Bio) tags to the surface of T cells that interact with the DCs. We demonstrate that the tagged T cells are bona fide antigen-specific T cells and based on their cell-surface fucosyl-biotinylation they can be readily separated from bystander T cells. Employing this approach, we are able to isolate intratumoral TSA-reactive CD4⁺, CD8⁺ T cells and TSA-suppressive CD4⁺ T cells. We discover that nearly all TSA-reactive CD8⁺ T cells co-express PD-1, whereas bystander T cells consist of PD-1⁺ and PD-1⁻ subsets. In comparison to bystander TILs, TSA-reactive CD8⁺ TILs possess a distinct TCR repertoire and unique gene features that are characterized by a dysfunction/activation transcript profile and genes upregulated in steroid biosynthesis and related metabolic pathways. By contrast, genes associated with antiviral defense mechanisms are enriched in bystander CD8⁺ T cells. Furthermore, tumor-specific antigen-suppressive and reactive CD4⁺ T cells co-exist in the tumor microenvironment and execute opposite roles in the regulation of antitumor immunity of CD8⁺ T cells. Despite exhibiting a dysfunctional phenotype, the TSA-reactive CD8⁺ TILs (i.e. PD-1⁺Bio⁺) possess substantial capabilities to proliferate. Upon in vitro expansion, this subset of TILs leads to significantly superior tumor killing than the expanded, entire PD-1⁺ TIL population both in vitro and in vivo.

Compared to the techniques that rely on bioinformatics-assisted TSA identification, FucoID features much simpler procedures and a quicker turnover cycle. Importantly, this method creates an avenue for the characterization and manipulation of the entire repertoire of endogenous, TSA-reactive TILs. It is generally applicable to several murine tumor models with detectable T-cell infiltration, and thus has a high potential to be tested in a clinical setting.

Results

Install H. pylori α(1,3)Fucosyltransferase (FT) onto the cell surface for probing cell-cell interactions

Although a number of enzyme-based proximity-labeling systems have been developed to profile protein-protein interactions (Branon et al., 2018; Kim and Roux, 2016; Liu et al., 2018; Long et al., 2016; Slavoff et al., 2011), very few can be applied to probe cell-cell interactions (Ge et al., 2019; Liu et al., 2018; Pasqual et al., 2018). To the best of our knowledge, all of these approaches rely on genetic manipulations. To design an enzymatic approach for probing cell-cell interactions of primary cells including those of the human origin requires an enzyme that can be installed onto the cell surface of bait cells without genetic engineering. For general applicability, the acceptor substrate of this enzyme should be naturally present on the cell surface of most cell types. The challenge is to identify an enzyme to achieve intercellular labeling with high sensitivity when two cells interact and low background in the absence of an interaction.

Previously, we developed a chemoenzymatic method that enables us to conjugate recombinant proteins onto the cell surface (Li et al., 2018) (Figure 1A, top left panel). This method relies on H. pylori α(1,3)fucosyltransferase (FT), a glycosyltransferase possessing remarkable donor substrate tolerance. It enables rapid and quantitative transfer of proteins conjugated to the enzyme’s natural donor substrate GDP-fucose (GDP-Fuc) to LacNAc (and/or α2,3-sialyl LacNAc), a common building block of the glycocalyx, of live cells. Because LacNAc and sialyl LacNAc are abundantly expressed by most cell types, including dendritic cells (DCs) and T cells, this method serves as a general approach for engineering the cell-surface landscape of immune cells. Based on this technique, here, we devised a genetic-engineering free strategy to probe cell-cell interactions using the membrane anchored FT that is introduced to the cell surface via the chemoenzymatic approach (Figure 1A, top right panel).

Figure 1. Illustration of probing cell-cell interactions via interaction dependent fucosyl-biotinylation (FucoID).

(A) Schematic representation of FucoID for the labeling of cell-cell interactions. Conjugation of H. pylori α(1,3)fucosyltransferase (FT) onto the bait cell surface was achieved via the FT-mediated chemoenzymatic glycan labeling using GDP-Fuc-FT as a self-catalyst. (B) Synthesis of GDP-Fuc-FT. (C) Experimental design and representative fluorescence microscopy images of the interaction dependent labeling mediated by FT-functionalized CHO cells (CHO-FT). CHO-FT cells stained with CFSE were mixed (cell ratio 1:5) with unfunctionalized CHO cells, followed by the addition of GDP-Fuc-Biotin (50 μM). Cells were stained with DAPI and streptavidin-APC for fluorescent microscopy imaging. Scale bar: 10 μm.

We synthesized GDP-fucose-conjugated FT (GDP-Fuc-FT), a unique small molecule-protein conjugate bearing both the donor substrate and the glycosyltransferase itself within the same molecule (Figure 1B). When cells of interest were incubated with GDP-Fuc-FT, the donor substrate-modified enzyme served as a self-catalyst to transfer Fuc-FT onto the cell surface LacNAcylated glycans (Figure S1A, S1B and S1D). Through this method, primary mouse CD8⁺ T cells and bone marrow derived DCs, human lymphocytes and DCs were successfully conjugated with FT and the robust modifications were achieved within 20 min (Figure S1C). Importantly, the installed FT remained on the cell surface for approximately 10 hours (Figure S1E) and did not affect the cell viability (Figure S1F) and functions, e.g. DC-mediated CD8⁺ T cell priming (Figure S1G).

Due to its high K_{m (for the acceptor substrate LacNAc)} (1.3 mM) and high k_cat (442 min⁻¹) (Soriano del Amo et al., 2010; Zheng et al., 2011), we hypothesize that the membrane-anchored FT would serve as an ideal tool to enable proximity-dependent labeling of prey cells that interact with bait cells harboring the enzyme (Figure 1A, bottom panel). In the absence of an interaction, the concentration of LacNAc (sLacNAc) acceptors in the proximity of FT is far below the FT-LacNAc K_m. Under such circumstances, the bimolecular reaction rate is governed by k_cat/K_{m (for LacNAc)}. By having a high K_{m (for the acceptor LacNAc)}, the background labeling is minimized. When two cells interact, the local concentration of LacNAc (sLacNAc) in the vicinity of FT is high such that the pseudo-zero-order reaction rate is determined by k_cat. By having a high k_cat, the labeling signal that occurs during bona fide cell-cell interactions is maximized.

To assess if FT-functionalized cells could mediate intercellular labeling by transferring a probe molecule, e.g. fucose-biotin (Fuc-Bio), from the exogenously added GDP-fucose-biotin (GDP-Fuc-Bio, its structure can be found in (Li et al., 2018)) to interacting cells, we incubated FT-functionalized wild-type CHO cells (CHO-FT) with adherent CHO cells to form cell-cell contacts. Upon the addition of GDP-Fuc-Bio, fluorescence microscopy imaging revealed that unfunctionalized CHO cells not in contact with CHO-FT were not labeled, but the CHO cells in contact with CHO-FT were strongly labeled at the cell-cell contacting interface (Figures 1C and S1H). Not surprisingly, CHO-FT cells were also robustly labeled on the membrane via self-fucosyl-biotinylation.

Probe DC-T cell interactions via FucoID

To determine if FucoID could be applied to probe antigen-specific DC-T cell interactions, we first examined FT-modified immature DCs (iDCs) for the selective labeling of CD8⁺ T cells from OT-I transgenic mice that express a transgenic T-cell receptor (TCR) specific for the SIINFEKL peptide (OVA_{257 −264}) of chicken ovalbumin presented on MHC-I (Figure 2A). FT-modified iDCs (CD45.1^+/+) were primed with OVA_257–264 before co-culturing with naïve OT-I CD8⁺ T cells (CD45.2), followed by the addition of GDP-Fuc-Bio to enable the labeling. Robust Fuc-Bio labeling was found on the interacting CD8⁺ T cells with a signal-to-background ratio of 36% versus 1% (Figure 2B). Here, the background is defined as the signal produced on CD8⁺ cells by incubating OVA_257–264-primed iDCs (without membrane-anchored FT) with naïve CD8⁺ for the same period of time. By contrast, the iDC-FT loaded with the lymphocytic choriomeningitis virus (LCMV) GP_33–41 peptide only induced the background level labeling of OT-I CD8⁺ T cells, indicating that the labeling is antigen-specific (Figure 2B). To further assess the sensitivity and specificity of FucoID in a more challenging situation, we mixed naïve OT-I (CD45.1^+/−) and P14 CD8⁺ T cells (Thy1.1^+/−) that recognize LCMV GP_33–41 presented by MHC-I and co-cultured the mixture with iDC-FT pulsed with OVA_257–264 and LCMV GP_33–41, respectively (Figure 2C). As illustrated in Figure 2C, upon addition of GDP-Fuc-Biotin, OT-I and P14 CD8⁺ T cells were selectively labeled by OVA_257–264 and LCMV GP_33–41 primed iDCs-FT, respectively.

Figure 2. Demonstration of the specificity of FucoID in probing DC-T cell interactions.

(A) Workflow for analyzing the FucoID enabled labeling of naïve OT-I CD8⁺ T cells using LCMV GP_33–41 and OVA_257–264 primed iDCs. (B) Flow cytometric analysis showing antigen-specific fucosyl-biotinylation of CD8⁺ T cells under iDC:T ratio of 1:1. (C) Flow cytometric analysis of antigen-specific fucosyl-biotinylation in the mixture of OT-I and P14 CD8⁺ T cells when incubated with iDC-FT primed with LCMV GP_33–41 or OVA_257–264, n=3.

Next, we sought to determine if antigen-specific labeling via FucoID could be achieved in cell mixtures of natural components and complexity. To this end, iDCs-FT primed by OVA_257–264 and LCMV GP_33–41, respectively, were co-cultured with OT-I splenocytes. As expected, OVA-specific fucosyl-biotinylation of OT-I CD8⁺ T cells were detected (Figure 3A and B). Subsequently, we optimized the labeling condition using the iDC–OT-I splenocyte (1:1) co-culturing system, finding that the optimal labeling was achieved within 2 hours using iDC modified with 0.2 mg/mL GDP-Fuc-FT (Figure S2A and B). Under this condition, the ratio of the specifically labeled CD8⁺ T cells was positively correlated with the quantity of OVA_257–264 used for iDC priming (Figure S2C). Increasing the iDCs-FT:T cell ratio further improved the efficiency of CD8⁺ T-cell capture in OT-I splenocytes (Figure S2D). In many clinical samples, the frequencies of antigen-specific T cells are low. To assess if FucoID could be applied to label antigen-specific T cells in such circumstances, we spiked OT-I T cells into C57BL/6J splenocytes to form a cell mixture with a low frequency of OT-I (i.e. a final ratio of OT-I CD8⁺ being ~2%). At the 1:1 iDC-FT:OT-I T cell ratio, iDC-FT primed with OVA_257–264 could capture 80% OT-I CD8⁺ T cells. When the iDC-FT:OT-I T cell ratio was increased to 3:1 in the co-culturing system, almost all OT-I CD8⁺ T cells (94%) could be captured with high sensitivity and specificity (Figure 3C).

Figure 3. Probing antigen-specific DC-T cell interactions via FucoID in splenocytes.

(A) Workflow for analyzing the interactions of antigen-primed iDCs with naïve OT-I T cells in OT-I splenocytes. (B) Flow cytometric analysis of antigen-specific fucosyl-biotinylation of CD8⁺ T cells in OT-I splenocytes by iDC-FT loaded with OVA_257–264 at the iDC:T cell ratio = 1:1. (C) Flow cytometric analysis of OT-I T cells selective labeling in C57BL/6J splenocytes (doped with OT-I splenocytes, OT-I T cells at a final ratio of 2% in total cells) by iDC-FT. (D) Flow cytometric analysis of OT-I CD8⁺ T cells Fuc-biotinylation by iDC-FT loaded with altered peptide ligands (APLs) derived from the original OT-I peptide SIINFEKL N4 (OVA_257–264). (E) Flow cytometric analysis of CD4⁺ T cells specific fucosyl-biotinylation by iDC-FT primed with OVA_323–339 in OT-II splenocytes (iDC:T ratio = 1:1), n=3.

To determine if FucoID can distinguish strong cell-cell interactions from weaker ones, we repeated the OT-I labeling experiment using iDCs primed with altered peptide ligands (APLs) derived from the original OT-I ligand SIINFEKL (N4). These APLs bind equally well to MHC-I H-2Kb as N4 but differ in their potency for interacting with TCR of OT-I CD8⁺ cells (binding strength: SIINFEKL(N4)>SAINFEKL(A2)>SIITFEKL(T4)) (Zehn et al., 2009). As shown in Figure 3D, the magnitude of intercellular labeling matched consistently with the TCR binding strength of the MHC-I bound APLs with the labeling induced by the N4-primed iDCs being the strongest (64.4%), which was followed by the labeling induced by iDC primed with A2 (30.3%), and the T4-primed iDC mediated labeling being the weakest (9.0%).

It was reported that surface molecules on APCs could be transferred to T cells by trogocytosis (Gary et al., 2012). Baltimore et al. reported recently that trogocytosis of TCR proteins occurring during Jurkat-K562 interactions could be used to identify tumor neoantigens. In our settings, we confirmed that negligible amounts of Fuc-Bio and FT were transferred from iDC to T cells via trogocytosis (Figure S3) (Li et al., 2019a).

Importantly, we found that FucoID could also be applied to probe iDC-CD4⁺ interactions despite significantly weaker binding between MHC-II bound peptides and CD4⁺ TCRs (Figure 3E). iDC-FT pulsed with OVA_323–339 specifically biotinylated 11.3% of OT-II CD4⁺ T cells whose TCR was reactive with OVA_323–339 under 2 hours of co-culturing. By contrast, only background labeling (1.68%) was observed in the group using LCMV GP_61–80 primed iDC-FT.

Rapid detection and enrichment of tumor specific antigen (TSA) reactive TILs based on FucoID in a B16-OVA tumor model.

With the validation of FucoID as a reliable technique for probing antigen-specific DC-T cell interactions ex vivo, we assessed the feasibility of using this strategy to detect TSA-reactive TILs from tumor digests. In our workflow, a harvested solid tumor is dissociated to prepare a single cell suspension and tumor lysates, in which the tumor lysates are used to prime autologous iDCs. Through this process, both non-mutant peptides and TSAs are loaded onto MHCs on the iDC surface. The primed iDCs are then subjected to FT conjugation and added to the single cell suspension, followed by the addition of GDP-Fuc-Biotin to initiate the interaction dependent fucosyl-biotinylation. Our hypothesis is that a majority of self-antigen-reactive T cells would have already been eliminated in the thymus via “negative selection” (Klein et al., 2014; Schietinger et al., 2012; Xing and Hogquist, 2012), which would leave T cells that interact with TSA-presenting iDCs being labeled by Fuo-Bio such that the labeled T cells are highly enriched for tumor-reactive T cells and that their reactivities are directed toward TSAs. (Figure 4A).

Figure 4. Identification and characterization of endogenous TSA-reactive CD8⁺ T cells from a B16-OVA melanoma model.

(A) Schematic illustration of the detection and enrichment of TSA-reactive TILs via FucoID. (B) Workflow of using FucoID to identify TSA-reactive and bystander CD8⁺ TILs from a B16-OVA tumor. (C) Representative flow cytometric analysis of TSA-dependent fucosyl-biotinylation of CD8⁺ TILs in B16-OVA tumors, n=3. (D) Experimental scheme for validating the isolated PD-1⁺Bio⁺ TILs as bona fide TSA-reactive T cells. Three isolated TIL subsets (CD45.1^+/+) were transferred to normal C57BL/6 mice (CD45.2), immunized with LM-OVA on the next day. Blood and splenocytes were analyzed after another 7 and 37 days, respectively. The isolated splenic CD8⁺ T cells were re-transferred to C57BL/6 mice (CD45.2), immunized with LM-OVA on the next day, and blood was analyzed after another 7 days. Representative flow cytometry plots are from three biological replicates. See also Figure S4B and C.

The well-established B16-OVA melanoma that expresses chicken ovalbumin as a TSA was used as the first model system to test this hypothesis. Subcutaneous B16-OVA tumors were digested to prepare single cell suspensions for co-culturing with iDC-FT (CD45.2) that were pre-treated with and without B16-OVA tumor lysates, respectively. GDP-Fuc-Bio was then added to initiate the interaction-dependent labeling. In this experiment, iDC-FT primed with OVA_257–264 was used as the positive control (Figure 4B). Whereas unprimed iDC-FT only labeled negligible numbers of CD8⁺ TILs (1.86%), iDC-FT primed by OVA_257–264 or B16-OVA tumor lysates labeled 25.6% and 56.9% autologous CD8⁺ TILs respectively. Among all CD8⁺ TILs, ~76% were PD-1⁺, within which approximately 74% were fucosyl-biotinylated by the iDC-FT primed with tumor lysates and according to our hypothesis this subset consisted of bona fide TSA-reactive T cells. The remaining 26% of PD-1⁺ TILs were not labeled (Figures 4C), suggesting that this fraction may be bystander cells. Interestingly, only ~3% PD-1⁻ T cells were labeled, indicating that almost all TSA-reactive T cell candidates had encountered their cognate antigens.

To determine if the PD-1⁺Bio⁺ subset possesses better killing capabilities toward B16-OVA cells than the PD-1⁻ and PD-1⁺Bio⁻ subsets, the labeled CD8⁺ TILs were FACS isolated and expanded using a reported rapid expansion protocol (Fernandez-Poma et al., 2017). The tumor killing activities were assessed on expansion day 7. As shown in Figure S4B, the expanded PD-1⁺Bio⁺ TILs showed significantly stronger killing of B16-OVA tumor cells than the other two subsets.

If bona fide TSA-reactive T cells were enriched in PD-1⁺Bio⁺ TILs, a sub-population of them should be OVA-specific. To confirm this, PD-1⁺Bio⁻ and PD-1⁺Bio⁺ TILs were cultured for 48 hours allowing the complete decay of the biotinylated molecules from the cell surface, at which point the cultured TILs were stained with the H-2Kb/OVA_257–264 MHC tetramer. Whereas the PD-1⁺Bio⁺ subset was found to contain 50% tetramer⁺ TILs, among which 6% stained tetramer^high, the PD-1⁺ Bio⁻ subset only contained 9.16% tetramer⁺ TILs and none of them were tetramer^high (Figure S4C).

To determine if the isolated PD-1⁺Bio⁺, PD-1⁺Bio⁻ and PD-1⁻ subsets can undergo re-expansion and memory formation upon OVA stimulation, the isolated TILs (CD45.1^+/+) were immediately transferred into antigen-free wild-type (WT) C57BL/6J hosts (CD45.2) individually (Figure 4D). One day later, these secondary recipients were immunized with Listeria expressing OVA (LM-OVA). Blood and/or spleens were isolated and analyzed on day 8 and day 38. Significant expansion of the transferred PD-1⁺Bio⁺ cells in blood were observed on day 8 (>1.1% of total blood CD8⁺) while no expansion of the transferred PD-1⁺Bio⁻ and PD-1⁻ subsets were detected (<0.05% of total blood CD8⁺) (Figure 4D). Most of these expanded CD45.1⁺CD8⁺ T cells were H-2Kb/OVA_257–264 tetramer positive (83%). On day 38, ~0.2% of PD-1⁺Bio⁺ T cells remained detectable in the spleen and a majority of them were H-2Kb/OVA_257–264 tetramer positive (84%). We then isolated the total CD8⁺ T cells from the spleens of these mice and transferred them into healthy C57BL/6J hosts (CD45.2). The recipient mice were challenged with LM-OVA on the next day. On day 7 post infection, blood was collected and analyzed. We observed re-expansion of the transferred CD45.1^+/+ CD8⁺ TILs in all recipient mice (~1% of total blood CD8⁺) and all these expanded cells are OVA-tetramer positive (Figure 4D). Together, these results provide solid support of FucoID as a highly effective approach to detect and enrich for TSA-reactive TILs from tumor cell suspensions for the subsequent adoptive transfer-based applications.

TILs labeled via FucoID in multiple syngeneic murine tumor models are bona fide TSA-reactive TILs

After confirming FucoID as a highly effective approach to detect and enrich for CD8⁺ TSA-reactive TILs from the B16-OVA model, we sought to explore its scope and limitation. Toward this end, tumor cell suspensions prepared from subcutaneous B16 melanoma, E0771 triple negative breast cancer (TNBC) whose human counterparts only showed a 12–19% objective response rate to checkpoint inhibitors (Crosby et al., 2018; Hendrickx et al., 2017), and MC38 colon tumors (a valid model for hypermutated colorectal cancer) (Efremova et al., 2018) were subjected to the FucoID-based labeling (Figure 5A). iDC-FT that were primed by tumor lysates fucosyl-biotinylated 34%, 17% and 25% autologous CD8⁺ TILs from B16 melanoma, E0771 TNBC and MC38 colon cancer, respectively (Figure 5B and S4A). By contrast, iDC-FT treated with the lysates obtained from the corresponding healthy tissues afforded only background labeling (<3%), which confirmed our hypothesis that T cells captured by tumor-lysate-primed iDC-FT are TSA-reactive T cells. Not surprisingly, iDC-FT primed with gp100_25–33, a predominant human and B16 melanoma specific antigen, labeled 10% CD8⁺ TILs from the B16 tumor cell suspension (Figure 5B). Of note, in all three tumor models, almost all biotinylated TILs were PD-1⁺, suggesting that these TILs have encountered their cognate antigens. By contrast, 35%, 50% and 41% of TILs from each model were PD-1⁺Bio⁻, respectively, suggesting that they are irrelevant, bystander T cells, but displaying a different phenotype than PD-1⁻ bystander TILs (Figure 5B).

Figure 5. Identification and characterization of TSA-reactive and bystander CD8⁺ T cells in distinct murine tumor models.

(A) Workflow for the detection of TSA-reactive CD8⁺ T cells from syngeneic murine tumor models via FucoID. (B) Representative flow cytometric analysis of tumor antigen dependent fucosyl-biotinylation of CD8⁺ TILs in B16 melanoma, E0771 TNBC and MC38 colon cancer models, n=3. (C) IFN-γ ELISpot showing distinct reactivity of expanded PD-1⁻, PD-1⁺Bio⁻ and PD-1⁺Bio⁺ CD8⁺ TILs upon tumor antigen re-stimulation, n=3; (D) Comparisons of the expanded total PD-1⁺ CD8⁺ TILs and PD-1⁺Bio⁺ CD8⁺ TILs for inducing specific lysis of the relevant cancer cells; n=3. (E) In vivo antitumor immunity evaluation of PD-1⁺Bio⁺ and PD-1⁺ CD8⁺ TILs in a murine B16 metastasis tumor model. C57BL/6J mice were intravenously injected with 0.5×10⁶ B16-luc cells. TIL transfer was performed as described in Method on day 3 of tumor inoculation. On day 8 of TIL transfer, luciferin was administrated, and light emission was recorded. Representative bioluminescence images are shown; HBSS: 7 mice; total PD-1⁺: 8 mice; PD-1⁺Bio⁺: 6 mice. (F) In vivo antitumor immunity evaluation of PD-1⁺Bio⁺ and PD-1⁺ CD8⁺ TILs in a murine MC38 s.c. tumor model. MC38 cells were s.c. injected into the right flanks of male C57BL/6J mice (0.5×10⁶ cells per mice). Mice were irradiated (5 Gy) on day 2 of tumor inoculation. Then TIL transfer was performed on day 3 of tumor inoculation. Tumor volumes were measured every 2 days. HBSS: 7 mice; anti-PD-1: 7 mice; total PD-1⁺: 7 mice; PD-1⁺Bio⁺: 8 mice.

To characterize the function of these subsets, the isolated PD-1⁻, PD-1⁺Bio⁻ and PD-1⁺Bio⁺ CD8⁺ TILs were cultured and expanded using the rapid expansion protocol in the presence of feeder cells, anti-mCD3 and rhIL-2. On day 9 upon expansion, we performed IFN-γ ELISpot to assess the TSA reactivity of the expanded T cells. T cells were re-stimulated with DCs pulsed with tumor lysates. As the negative controls, T cells were re-stimulated by DCs pulsed with an irrelevant peptide. As expected, when re-stimulated with DCs primed with tumor lysates specific IFN-γ secretion was only detected in the expanded PD-1⁺Bio⁺ TILs, which was blocked by the addition of an anti-MHC-I antibody (Figure 5C).

The differences in TSA-reactivity of PD-1⁺Bio⁺ and PD-1⁺Bio⁻ CD8⁺ TILs would result from the differences of their TCR clonotypic repertories. We characterized TCR clonotypic repertoires of these two TIL subsets isolated from E0771 and MC38 tumor models directly after their FACS isolation without further in vitro expansion. TCRβ deep sequencing was employed for quantifying the frequency of individual T-cell clonotype in each subset. A productive CDR3 sequence that does not contain stop codons or frame shifts represents a unique TCR clonotype, and the total number of unique sequences determines the clonal diversity in each subset, providing each subset has comparable total CDR3 reads. We found that TCRβs in the PD-1⁺Bio⁺ population were significantly more oligoclonal than their counterparts in the PD-1⁺Bio⁻ subset, suggesting the cells in the PD-1⁺Bio⁺ subset have undergone substantial TSA-driven clonal expansion (Figure S4D). Furthermore, there was no overlap of the 10 most abundant TCRβ CDR3s found between these two subsets in either tumor models (Figure S4D). The results from IFN-γ ELISpot assay and TCRβ deep sequencing indicated that PD-1⁺Bio⁺ and PD-1⁺Bio⁻ TILs represent two functionally and clonotypically distinct T cell subsets that co-exist in the same tumors and share a certain degree of phenotypical similarities (e.g. PD-1⁺).

Because PD-1⁺ TILs consisted of not only TSA-reactive T cells but also bystander T cells, upon in vitro expansion, anti-tumor cytotoxicity of the entire PD-1⁺ TIL population should be considerably weaker than that of the PD-1⁺Bio⁺ TIL subset providing that TSA-reactive and bystander T cells share similar expanding rates. To assess this hypothesis, PD-1⁺Bio⁺, PD-1⁺Bio⁻ and total PD-1⁺ TILs isolated from the same tumors were subjected to the rapid expansion protocol. According to the recorded growth curve, PD-1⁺Bio⁻ and total PD-1⁺ TILs exhibited very similar expansion rate, which was significantly faster than that of PD-1⁺Bio⁺ TILs during the entire course of expansion (Figure S4E). These observations suggest that at the end of expansion bystander PD-1⁺Bio⁻ TILs become the dominant cell population within the expanded total PD-1⁺ TILs. On the expansion day 10, tumor killing capabilities of each subset were assessed. At different effector/target ratios PD-1⁺Bio⁺ TILs exhibited remarkably stronger tumor cell killing than the corresponding PD-1⁺Bio⁻ and total PD-1⁺ TILs (Figures 5D and S4F). These results suggest that due to the faster proliferation of bystander T cells within the entire PD-1⁺ TIL population, anti-tumor activities of total PD-1⁺ TILs become significantly weaker than those of the expanded TSA-reactive PD-1⁺Bio⁺ TILs at the end point of rapid expansion.

The expanded PD-1⁺Bio⁺ CD8⁺ TILs exhibit significantly higher anti-tumor activities than the entire PD-1⁺ TILs in vivo

To compare anti-tumor activities of the expanded PD-1⁺Bio⁺ CD8⁺ TILs and the total PD-1⁺ subset in vivo, we first explored the use of the TILs isolated from the B16 melanoma model to control tumor growth in mice with established pulmonary micrometastases. Three days after the intravenous inoculation of B16 tumor cells stably transduced with firefly luciferase (B16-luc) to induce pulmonary metastasis, tumor bearing mice were treated with the expanded total PD-1⁺ TILs and PD-1⁺Bio⁺ TILs, respectively, while the control group was injected with HBSS only. Tumor proliferation was monitored by longitudinal, noninvasive bioluminescence imaging. As shown in Figure 5E, the total PD-1⁺ TILs showed moderate therapeutic potency for preventing tumor proliferation with 60% lower bioluminescence than the HBSS control group on treatment day 8. In comparison, PD-1⁺Bio⁺ TILs significantly inhibited tumor growth, showing 98% lower bioluminescence signal than the HBSS control group. Remarkably, the survival of the tumor bearing mice was significantly elongated upon treatment with the PD-1⁺Bio⁺ TILs. All mice received buffer only and the total PD-1⁺ TILs died on treatment day 22 and 26, respectively. By contrast, all PD-1⁺Bio⁺ TIL recipient mice were alive until treatment day 27 and by the end of the experiment, still 20% mice remained alive.

To assess capabilities of the expanded PD-1⁺Bio⁺ CD8⁺ TILs for suppressing solid tumor growth we sought to use the TILs isolated from the subcutaneous MC38 tumors. Expanded total PD-1⁺ CD8⁺ TILs and PD-1⁺Bio⁺ CD8⁺ TILs (5×10⁶ per mice), respectively, were intravenously injected into mice with established subcutaneous MC38 tumors, followed by anti-PD-1 administration. In the control groups, mice were treated with anti-PD-1 and HBSS, respectively. Although anti-PD-1 and anti-PD-1 + total PD-1⁺ TIL treatments only showed modest tumor control, PD-1⁺Bio⁺ TILs combined with anti-PD-1 significantly slowed down tumor growth with median tumor size being only ¼ of that treated with anti-PD-1⁺ total PD-1⁺ TILs (treatment day 22, Figure 5F). Moreover, whereas no mice in any control group survived up to day 28, 50% mice treated with PD-1⁺Bio⁺ TILs + anti-PD-1 were still alive by day 34. And one mouse was found to be tumor free by day 40 (Figure 5F). These results indicated that the expanded PD-1⁺Bio⁺ CD8⁺ TILs possess markedly higher activities to control tumor growth in vivo than the expanded total PD-1⁺ CD8⁺ TILs that contain a large fraction of bystander T cells.

PD-1⁺Bio⁺ CD8⁺ TILs are distinct to PD-1⁺Bio⁻ TILs and display activation/dysfunction gene signature

To gain an understanding of the genetic programs that underlie the phenotypical and functional features of the TSA-reactive and two different groups of bystander TILs, we characterized transcriptional profiles of PD-1⁺Bio⁺, PD-1⁺Bio⁻ and PD-1⁻ CD8⁺ T cells isolated from MC38 subcutaneous tumors. Principle component analysis (PCA) revealed that the transcript profiles of these three subsets of TILs shared substantial divergence (Figure 6A). As identified by volcano plot messenger RNA (mRNA) comparisons between PD-1⁺Bio⁺ and PD-1⁺Bio⁻ CD8⁺ TILs, 290 transcripts were significantly upregulated or downregulated (Figure 6B).

Figure 6. Gene-expression and functional marker characterizations of CD8⁺ TIL subsets isolated via FucoID from MC38 tumors.

(A) PCA of the transcriptome of PD-1⁺Bio⁺ (red), PD-1⁺Bio⁻ (blue) and PD-1⁻ (green) CD8⁺ TILs isolated from murine MC38 tumors. Dots represent samples of the three different populations (grouped by colors) from a total of three biological replicates (grouped by shapes). PCA 1 and 2 represents the largest source of variation. (B) Volcano plot of up- (red) and down-(blue) regulated genes between PD-1⁺Bio⁺ and PD-1⁺Bio⁻ TILs. Significance was determined as Benjamini–Hochberg FDR (p.adjust) < 0.05 and |log₂(fold change)| ≥ 0.6. (C) Biological processes (GO terms) enriched in the up– and down–regulated genes identified in Figure 6B. (D) Gene set enrichment analysis (GSEA) of activation/dysfunction CD8 gene module (Singer et al., 2016) in the transcriptome of PD-1⁺Bio⁺ vs. that of PD-1⁺Bio⁻ CD8⁺ TILs. See also Figure S5 and Table S1. (E) GSEA of up- and down-regulated tumor specific CD8 gene signature (Schietinger et al., 2016) in the transcriptome of PD-1⁺Bio⁺ vs. that of PD-1⁺Bio⁻ CD8⁺ TILs. See also Figure S5 and Table S1. (F) The comparison of representative gene expression of PD-1⁺Bio⁺, PD-1⁺Bio⁻ and PD-1⁻ CD8⁺ TILs. n=3. (G) The expression of PD-1, CD137, TIM-3, CD39 and CD103 in PD-1⁺Bio⁺, PD-1⁺Bio⁻ and PD-1⁻ CD8⁺ TILs from murine MC38 tumor according to flow cytometric analysis. Data obtained from at least two independent replicates.

We then focused on analyzing the less pronounced transcriptional differences of PD-1⁺Bio⁺ and PD-1⁺Bio⁻ CD8⁺ TILs. An over-representation analysis was conducted to explore the enrichment of the 290 genes in biological processes annotated by the gene ontology database. Compared to PD-1⁺Bio⁻ TILs, several up-regulated genes of PD-1⁺Bio⁺ TILs were significantly enriched in steroid biosynthesis and related metabolic pathways, such as MSMO1 and DHCR7 (Figures 6C and S5A). This is consistent with the previously reported discovery that the cholesterol metabolism of T cells is fully reprogrammed upon cell activation to support cell proliferation (Bensinger et al., 2008; Tuosto and Xu, 2018). An increase in the plasma membrane cholesterol level of CD8⁺ T cells augments T-cell receptor clustering, signaling and the more efficient formation of the immunological synapse, which are essential for the effector function of CD8⁺ T cells (Yang et al., 2016). By contrast, down-regulated genes were enriched in more diverse biological process networks, including those of lymphocyte differentiation, T cell migration and activation, viral response and calcium homeostasis (Figures 6C and S5A). These findings strongly suggest that PD-1⁺Bio⁻ CD8⁺ TILs are bystander T cells that share similar transcriptional signatures with previously reported viral antigen-specific bystander CD8⁺ TILs and have an altered spectrum of core cellular processes compared to PD-1⁺Bio⁺ TILs (Scheper et al., 2019; Simoni et al., 2018).

To further characterize genetic differences of these three subsets of TILs, we performed gene set enrichment analysis (GSEA) using the gene signatures and gene modules established in chronic virus infection induced T cell exhaustion and tumor associated T cell activation/dysfunction models. We initially compared these three subsets TILs for the enrichment of a previously reported naïve/memory-like T cell gene module and found this module was enriched in PD-1⁻ TILs (Figures S5B and Table S1) (Singer et al., 2016). Next, we assessed the three subsets TILs for the enrichment of the exhaustion signature derived from exhausted T cells isolated from chronic LCMV infection TILs (Wherry et al., 2007), finding a similar enrichment of this signature in both PD-1⁺Bio⁺ and PD-1⁺Bio⁻ CD8⁺ subsets compared to PD-1⁻ TILs (Figures S5C and Table S1). We then compared the TSA-reactive PD-1⁺Bio⁺ TILs with the bystander PD-1⁺Bio⁻ and PD-1⁻ TILs for the enrichment of the gene modules shared by T cells infiltrating human or murine tumors. The T cell activation/dysfunction gene module established for B16F10 melanoma (Singer et al., 2016) was significantly enriched in PD-1⁺Bio⁺ vs. PD-1⁺Bio⁻ TILs; notable genes in this module include genes encoding the cytokine IL2 receptor (IL2RA), the T cell activation related glycolysis enzyme GAPDH (GAPDH) and the plasma membrane transporter for monocarboxylates such as lactate and pyruvate (SLC16A3) (Figures 6D, S5D and Table S1). Consistent with this finding, an enrichment of the upregulated cell cycle gene signature that was validated for human melanoma TILs was found in PD-1⁺Bio⁺ TILs in comparison to both PD-1⁺Bio⁻ and PD-1⁻ bystander TILs (Li et al., 2019b) (Figure S5E and Table S1). This finding, combined with the observed clonal expansion of PD-1⁺Bio⁺ TILs, provided strong evidence for ongoing proliferation within this dysfunctional but TSA-reactive T cell subset. This feature had also been observed previously for a subpopulation of CD8⁺ T cells infiltrating human and murine tumors that were believed to possess tumor reactivity (Li et al., 2019b; Miller et al., 2019). By comparing the transcript profiles of monoclonal CD8⁺ T cells specific for Tag epitope I (Tag-I; SAINNYAQKL) infiltrating early and late stage murine tumors to that of D30 exhausted T cells isolated from chronic LCMV infection, Greenberg and coworkers discovered the unique gene signatures of dysfunctional, tumor-specific T cells that are not shared by dysfunctional T cells triggered by chronic viral infection (Schietinger et al., 2016). We compared our polyclonal TSA-reactive PD-1⁺Bio⁺ TILs with the bystander PD-1⁺Bio⁻ and PD-1⁻ TILs for the enrichment of this tumor-specific T cell gene set and found that it was significantly enriched in the PD-1⁺Bio⁺ subset (Figures 6E, S5F and Table S1).

To compare FucoID with functional marker staining in the identification of TSA-reactive TILs, we analyzed the transcript levels of the genes that were previously reported as tumor reactive TILs selection markers in PD-1⁺Bio⁺ and PD-1⁺Bio⁻ TIL subsets isolated from the MC38 colon cancer model. To our surprise, in both TIL subsets similar transcript expression levels were detected for all of these selection markers, including PDCD1 (PD-1), HAVCR2 (TIM-3), LAG3 (LAG-3), ENTPD-1 (CD39), ITAGE (CD103), TNFRSF4 (CD134) and TNFRSF9 (CD137) (Figure 6F). We further analyzed the expression of several of these selection markers on the cell surface by flow cytometry (Figures 6G, S6A and S6B). Although the PD-1⁺Bio⁺ TIL subset was found to express higher levels of TIM-3 and CD137 than the PD-1⁺Bio⁻ TILs, varying levels of LAG-3, CD39, and CD103 expression were found in both subsets (Duhen et al., 2018; Gros et al., 2014; Yossef et al., 2018). We also analyzed TILs isolated from B16 and E0771 tumor models and found similar varying expression levels of these markers (Figures S6C and S6D). Taken together, we conclude that FucoID may be more generally applicable than these previously reported functional markers-based selection approaches to identify TSA-reactive TILs.

Intratumoral antigen-specific CD4⁺ T cells play bidirectional roles in the regulation of antitumor immunity of TSA-reactive CD8⁺ TILs

Intra-tumoral antigen-specific CD4⁺ T cells have been identified in both patients and murine tumor models (Ahmadzadeh et al., 2019; Alspach et al., 2019; Kreiter et al., 2015; Oh et al., 2020). However, their biological functions are just starting to be elucidated partially due to the lag of research-tool development. To explore if FucoID can capture antigen-specific CD4⁺ TILs, we utilized the well-established Pan02 murine pancreatic ductal adenocarcinoma model that has abundant T cell tumor infiltrations. Following similar procedures as described in Figures 4A and B, antigen-specific CD8⁺ and CD4⁺ TILs were successfully labeled via FucoID and isolated by FACS (Figures 7A and 7B). Specifically, approximately 10% and 15% CD8⁺ and CD4⁺ T cells, respectively, were fucosyl-biotinylated and the interaction-dependent labeling was blocked by anti-MHC antibodies.

Figure 7. TSA-suppressive and -reactive CD4⁺ TILs play opposite roles in regulating the antitumor immunity of TSA-reactive CD8⁺ TILs in a Pan02 pancreatic cancer model.

(A) and (B) Flow cytometric analysis of tumor antigen-specific fucosyl-biotinylation of CD8⁺ TILs (A) and CD4⁺ TILs (B) in the Pan02 model. (C) Flow cytometric analysis of functional markers on Bio⁻ CD4⁺ TILs and Bio⁺ CD4⁺ TILs in Pan02 tumors. (D) IFN-γ ELISpot analysis of tumor-suppressive and -reactive functions of the isolated TIL subsets. Five subsets of TILs (Bio⁺ CD8⁺, Bio⁻ CD8⁺, CD25⁻Bio⁺ CD4⁺, CD25⁺Bio⁺ CD4⁺ and Bio⁻ CD4⁺) isolated from Pan02 tumor digests were re-stimulated by DCs primed with tumor lysates. Re-stimulations were conducted using the individual TIL subsets and a combination of two subsets as specified, n=3.

We then analyzed Bio⁺ and Bio⁻ CD4⁺ TIL subsets using functional markers to characterize their phenotypes. Similar to the Bio⁻ CD8⁺ TILs, the Bio⁻ CD4⁺ TILs consisted of both PD-1⁺ and PD-1⁻ subsets. Interestingly, CD137⁺ and CD137⁻ and CD134⁺ and CD134⁻ subsets, respectively, were found in the Bio⁺ CD4⁺ TILs whereas the Bio⁻ CD4⁺ TILs were mainly CD137⁻ and CD134⁻ (Figures 7C and S7A). Within the Bio⁺ CD4⁺ subset approximately 64% cells were FoxP3⁺ and the remaining 36% cells were CD25⁻FoxP3⁻, suggesting the majority of Bio⁺ cells are tumor-interacting regulatory T cells (Treg) (Figure 7C). Worthy of note, among the FoxP3⁺ cells 1/3 were CD25⁻, thus indicating that CD25⁻Bio⁺ CD4⁺ TILs contained a relatively smaller portion of FoxP3⁺ and a larger portion of FoxP3⁻ cells while all CD25⁺Bio⁺ CD4⁺ TILs were FoxP3⁺. By contrast, within the Bio⁻ subset almost all cells were CD25⁻FoxP3⁻. Similar trends were observed when analyzing CD4⁺ TILs isolated from Pan02 tumors grown in B6-Foxp3^EGFP mice: Bio⁺CD4⁺ TILs mainly consisted of Foxp3⁺ T cells that were either CD25⁺ or CD25⁻ whereas Bio⁻ CD4⁺ TILs were primarily CD25⁻FoxP3⁻ (Figure S7B). Consistent with the enrichment of FoxP3⁺ in the Bio⁺ subset, these cells also expressed abundant glucocorticoid-induced tumor necrosis factor receptor (GITR) and Helios, a transcription factor controlling Treg stability and function (Figure 7C) (Schaer et al., 2013; Thornton et al., 2010).

To verify that the isolated Bio⁺ TILs were tumor antigen-specific and to further characterize their biological functions, we performed an IFN-γ ELISpot assay using five isolated TIL subsets (Bio⁻ CD8⁺, Bio⁺ CD8⁺, Bio⁻ CD4⁺, CD25⁺Bio⁺ CD4⁺, CD25⁻Bio⁺ CD4⁺). In this assay, individual TIL subsets and a combination of TILs as specified in Figure 7D, respectively, were re-stimulated with DCs primed with Pan02 tumor lysates. By examining individual TIL subsets, we discovered that when the same numbers of cells were used for re-stimulation Bio⁺ CD8⁺ TILs exhibited the strongest tumor reactivity, which was followed by CD25⁻Bio⁺ CD4⁺ TILs, whereas CD25⁺Bio⁺ CD4⁺ TILs exhibited the weakest reactivity (Figure 7D, S7C). Notably, when CD25⁺Bio⁺ CD4⁺ TILs were stimulated by PMA/Ionomycin, significantly less IFN-γ secretion was detected in comparison to that of the CD25⁻Bio⁺ CD4⁺ subset (Figure 7D, 8 versus 7), strongly suggesting that CD25⁺Bio⁺ CD4⁺ TILs were enriched with Tregs. As expected, although Bio⁻ CD8⁺ and Bio⁻ CD4⁺ subsets had high IFN-γ secretion when stimulated by PMA/Ionomycin, they showed no reactivity when stimulated with tumor-lysate-primed DCs (Figure 7D, 3 versus 2, 10 versus 9), indicating that these cells are non-specific, bystander TILs. Interestingly, adding CD25⁺ and CD25⁻Bio⁺ CD4⁺ cells, respectively, to Bio⁺ CD8⁺ TILs induced opposite effects. Whereas CD25⁻Bio⁺ CD4⁺ TILs showed an additive effect with Bio⁺ CD8⁺ TILs to augment IFN-γ secretion (Figure 7D, 11 versus 1), CD25⁺Bio⁺ CD4⁺ TILs significantly suppressed the tumor reactivity of Bio⁺ CD8⁺ TILs (Figure 7D, 14 versus 1). These additive and suppressive effects were blocked by the addition of anti-MHCII, indicating the immune reactive and suppressive functions of Bio⁺ CD4⁺ are tumor antigen-specific (Figure 7D, 11 versus 13, 14 versus 15). Together, these findings suggest that selectively boosting the TSA-reactive CD4⁺ T cell subset rather than the TSA-suppressive CD4⁺ TILs in the tumor microenvironment may further potentiate anti-tumor immunity of TSA-reactive CD8⁺ TILs.

Discussion

TILs within individual tumors consist of heterogeneous populations including not only the T cells specific for tumor antigens, but also those recognizing a wide range of epitopes unrelated to cancer (e.g. antigens from Epstein–Barr virus, human cytomegalovirus or influenza virus) (Scheper et al., 2019; Simoni et al., 2018). These bystander CD8⁺ TILs have diverse phenotypes that overlap with those of the tumor antigen-specific T cells, but are not tumor reactive (Duhen et al., 2018; Yossef et al., 2018). Although several selection markers (e.g. PD-1, CD39, CD103) have been utilized to exclude bystander CD8⁺ TILs, they are empirical and may generate false positive selection. Moreover, TSA-reactive TILs in less abundant or rare populations could be missed using such indirect selection methods because these TILs may not share the same exhaustion status or phenotypes with the most abundant TSA-reactive TILs. By contrast, the FucoID strategy developed here generates a selection maker, i.e. biotin, based on the direct TCR-pMHC interaction, thus providing an unbiased approach for TSA-reactive TIL identification. Via this approach, we discovered a new subset of bystander T cells (i.e. PD-1⁺Bio⁻ TILs) that possessed distinct transcriptional signatures as compared to the previously reported bystander PD-1⁻CD8⁺ T cells. Importantly, this subset of TILs could not be easily identified by conventional methods (such as tetramer staining and functional markers). Through FucoID, although we are not able to elucidate the identity of TSAs, T cell candidates that are TSA-reactive would be enriched directly for expansion and for rapid isolation of the corresponding TCRs to construct TCR engineered T cells for functional evaluation. As a consequence, research time and cost are dramatically reduced compared to the aforementioned reverse immunology based pMHC tetramer approach (Arnaud et al., 2020). Significantly, once a TSA-reactive TCR is confirmed, it is possible to use other recently developed methods to identify the corresponding antigen (Gee et al., 2018; Kula et al., 2019; Li et al., 2019a).

TSA-reactive CD8⁺ T cells (i.e. PD-1⁺Bio⁺ T cells) isolated by FucoID from murine tumor models in this study exhibited a dysfunctional phenotype, but still possessed significant proliferative and tumor killing capacities. A subpopulation of these TSA-reactive T cells (4–18%) harbored progenitor exhausted T cell characteristics (TCF1⁺TIM-3^–) (Figures S6B, 6C and 6D), which is in line with tetramer-sorted tumor-specific T cells from previous studies (Miller et al., 2019). It has been demonstrated that it is this subset of CD8⁺ T cells that provides the proliferative burst and effector function following anti-PD-1/PD-L1 therapy (Held et al., 2019; Im et al., 2016; Miller et al., 2019; Siddiqui et al., 2019). Therefore, future efforts should be devoted to approaches for enlarging this subset during ex vivo rapid expansion to boost the therapeutic potential of TIL-based adoptive cell transfer. In addition, we also found that steroid biosynthesis genes were selectively upregulated in TSA-reactive CD8⁺ T cells but not in their bystander counterparts, suggesting that boosting cholesterol biosynthesis may potentially enhance the antitumor efficacy of TSA-reactive CD8⁺ T cells.

CD4⁺ T cells are challenging targets to study using conventional approaches partially due to the diversity and length variation (11 to 30 amino acids) of the corresponding MHCII-binding epitopes and their weak interactions with MHCII (Editorial, 2017; Racle et al., 2019). We demonstrated here that FT-modified mouse DCs could induce antigen specific fucosyl-biotinylation of not only CD8⁺ but also CD4⁺ T cells, and the labeling strength was correlated to the binding affinities of pMHC to TCR. Thus, FucoID opens a new door to study endogenous, antigen-specific CD4⁺ T cells in cancer immunity and autoimmune diseases. In the current study, we successfully used FucoID to separate TSA-suppressive and -reactive CD4⁺ T cells from bystander CD4⁺ TILs in a murine pancreatic tumor model. Likewise, one can envisage applying FucoID to separate T cells possessing high affinity TCRs from those with weaker ones for studying their functions in tumors and related infection models.

As the first glycosyltransferase-mediated tagging approach for probing cell-cell interactions, FucoID does not rely on genetic manipulations such that it is readily applicable to primary cells. Importantly, installing FT onto human DCs is simple and easily accomplished (Figure S1C). Therefore, FucoID has a high potential to be translated to a clinical setting for the detection and isolation of TSA-reactive TILs from human patients. Through popularizing FucoID, we expect the pace for the discovery of TSA-reactive TILs and their TCRs would be significantly accelerated, which in turn would pave the way for lowering the cost and accessibility of personalized cancer treatment (Arnaud et al., 2020; Yamamoto et al., 2019).

STAR★METHODS

RESOURCE AVAILABILITY

Lead Contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Peng Wu (pengwu@scripps.edu).

Materials Availability

All unique/stable reagents generated in this study are available from the Lead Contact with a completed Materials Transfer Agreement.

Data and Code Availability

Raw data of TCR sequencing, mRNA sequencing and their processed data has been deposited in the NCBI GEO database under accession GSE154605.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Cell lines

Cell lines were purchased from ATCC unless otherwise specified. CHO cell lines (WT, Lec2 and Lec8, provided by Prof. Pamela Stanley at Albert Einstein College of Medicine) were grown as monolayers in alpha-Minimum Essential medium (α-MEM) (GIBCO) supplemented with 10% fetal bovine serum (FBS) (Omega Scientific, Inc). Cancer cell lines including E0771 (from Dr. Klemke lab, UCSD), MC38 (Kerafast), mouse B16 and B16-OVA (provided by Prof. Gregoire Lauvau lab) are all grown in DMEM (Dulbecco’s modified Eagle’s medium, GlutaMAX, GIBCO) supplemented with 10% FBS. All cells cultures were incubated at 37 °C under 5% CO₂.

Human peripheral blood lymphocytes

Human blood samples were collected from a healthy donor (40yr old, female, white) under the TSRI Normal Blood Donor Services program (#IRB 15–6710). Peripheral blood mononuclear cells (PBMCs) were obtained by Ficoll (Ficoll-Paque Plus, GE) density centrifugation. Primary human T cells preparation: 4 million per mL PBMCs were cultured in T cell culture media with 15 ng/mL rhIL-2 and stimulated with anti-CD3/CD28 for two days. Activated human T cells were kept at 2.5–4×10⁶ cells/mL in T cell culture media (fresh media with cytokine were added every two days). Phenotypes were characterized after expansion for two weeks, at which stage >95% cells were CD3⁺ human T cells. Primary human DC preparation: CD14⁺ monocytes were isolated from PBMCs using the EasySep™ Human CD14 Positive Selection Kit II. Enriched CD14⁺ monocytes were cultured in complete T cell medium with rhIL-4 (10 ng/mL) and rhGM-CSF (20 ng/mL). Immature human DCs were harvest for FT labeling on day 7.

Mice

All mice were bred and housed under specific pathogen free (SPF) conditions. All animal experiments were approved by TSRI Animal Care and Use Committee. CD45.2 (WT), CD45.1^+/+, Thy1.1^+/+ and Foxp3-GFP mice in C57BL/6J genetic background, BALB/c mice, OT-I mice and P14 mice were purchased from the Jackson Laboratory. OT-II mice were from Dr. James C. Paulson, Scripps Research. OT-I^+/−CD45.1^+/− and OT-I^+/−Thy1.1^+/− mice were generated by cross breeding OT-I with CD45.1^+/+ and Thy1.1^+/+ C57BL/6J mice, respectively. Both male and female mice of 6–12 weeks of age were used for all experiments.

Primary OT-I T cells preparation

Naïve OT-I CD8⁺ T cells and splenocytes were isolated from the spleen of OT-I mice. CD8⁺ T cells were isolated using a mouse CD8⁺ isolation kit (StemCell). T cells and splenocytes were cultured in RPMI 1640 (GlutaMAX) with 10% heat-inactivated FBS, 1 mM sodium pyruvate, 50 μM β-ME, 10mM HEPES and 1×MEM NEAA, 100 U/ml penicillin and 100 mg/ml streptomycin (later referred to as complete T cell culture media). Cytokines in T cell culture media were added as indicated. To generate effector CD8⁺ T cells, OT-I splenocytes were cultured in complete T cell media containing 1 nM SIINFEKL (OVA_257–264) peptide for 2 days, followed by the addition of 100 IU/mL rhIL-2. Effector CD8⁺ T cells were ready for use after another 4 days in culture. All cells cultures were incubated at 37 °C under 5% CO₂.

Immature bone marrow dendritic cells (iDCs) and antigen priming

Bone marrow was isolated from CD45.1^+/+ or WT (CD45.2) C57BL/6J mice. Following erythrocyte lysis, the bone marrow cells were resuspended in complete T cell medium with rmGM-CSF (20 ng/mL). The culture medium was changed every 3 days supplied with rmGM-CSF (20 ng/mL). After 7 days culturing, non-adherent cells and loosely adherent cells were harvested and gently washed by DPBS for subsequent labeling experiments. In the iDC-OT-I CD8⁺ T cell interactions experiment, iDCs or FT functionalized iDCs were cultured with or without indicated antigen peptides at 37 °C for 30 min. Then non-adherent cells in the culture supernatant and loosely adherent cells were harvested and gently washed with DPBS for subsequent experiments. In the TSA-reactive TILs FucoID experiment and IFN-γ ELISpot assay, iDCs were cultured with or without indicated antigen peptides or tumor lysates (tumor cell/DC ratio = 10:1) in complete T cell medium containing rmGM-CSF (20 ng/mL) for 16 hours. Then non-adherent cells in the culture supernatant and loosely adherent cells were harvested and gently washed with DPBS for subsequent experiments.

Adoptive cell transfer and tumor immunotherapy experiments

Tumor inoculation for TILs isolation

1×10⁶ B16 and 1×10⁶ B16-OVA were implanted subcutaneously into the flanks of male C57BL/6J mice (CD45.2 if no specific indication). 1×10⁶ MC38 and 0.7×10⁶ Pan02 tumor cells were implanted subcutaneously into the middle of the armpits of male C57BL/6J mice. 1×10⁶ E0771 cells were implanted into the 4^th mammary fat pad of female C57BL/6J mice. 15 days (for B16, B16-OVA, E0771 and MC38) or 21 days (for Pan02) later, tumors were excised for use for TIL isolations.

Adoptive cell transfer and tumor immunotherapy

B16 tumor model B16-luc (B16 stably transduced with firefly luciferase) melanoma cells (0.5×10⁶ per mice) were inoculated into C57BL/6J mice by tail vein injection. Three days later, mice were administered with 200 μL D-luciferin (15mg/mL) through intraperitoneally injection and tumor sizes were calculated according to bioluminescence signals. The mice were then grouped by tumor volumes. Subsequently, each group was transferred with HBSS, total PD-1⁺ TILs and PD-1⁺Bio⁺ TILs, respectively, by tail vein injection (3×10⁶ cells per mice). After treatment, mice were i.p. injected with rhIL-2 (50,000 IU/mice) every 12 hours for 4 days. On day 8 upon TIL transfer, mice were injected with 200 μL D-luciferin (15mg/mL) through i.p. injection. The bioluminescence signal in mice were analyzed by PerkinElmer IVIS system in 10 min. The total photons were quantified by IVIS software. Survival of mice was tracked from day 0 of TIL transfer to day 36. MC38 tumor model MC38 tumor cells (0.5×10⁶ per mice) were subcutaneously inoculated into the right flanks of male C57BL/6J mice. Two days later, all mice were irradiated (5 Gy). On the next day, mice were randomly grouped. One control group was treated with HBSS by tail vein injection. The other three groups were transferred with HBSS, PD-1⁺ TILs and PD-1⁺Bio⁺ TILs, respectively, by tail vein injection (5×10⁶ cells per mice), followed by immediate administration of anti-PD-1 (100 μg per mice) by intraperitoneal injection. After treatment, all mice were i.p. injected with 50,000 IU/mice rhIL-2 every 12 hours for 4 days. On day 7 upon TIL transfer, another dose of anti-PD-1 (i.p., 100 μg per mice) was given to the above three groups of mice except the first HBSS-treated control group. Tumor major axis D and minor axis d were measured every 2 days from day 10 upon TIL transfer. Tumor volumes were calculated using the formula: Tumor volume (mm³) = 1/2 ×D × d². Survival of mice was tracked from day 0 of TIL treatment to day 40. A mouse was considered as dead when its tumor volume exceeded 400 mm³.

METHOD DETAILS

One-pot protocol for producing GDP-Fucose-α(1,3)fucosyltransferase (GDP-Fuc-FT)

Reactions were typically carried out in a 1.5 mL Eppendorf tube. TCO group was first introduced onto α(1,3)fucosyltransferase (see Li. et al. 2018 for expression and purification details) according to the standard labeling protocol of TCO-PEG₄-NHS ester (Li. et al. 2018). Briefly, 50 mM stock of TCO-PEG₄-NHS in DMSO was added to the solution of FT (5 mg/mL, 200 μL) at a final concentration of 2 mM. The reactions were incubated at room temperature for 30 minutes and quenched by adding 1M Tris buffer (pH 8.0, final concentration of 50mM). The quenched reaction mixtures were incubated at room temperature for 5 minutes and then desalted into DPBS using G25 desalting column (PD-10, GE). The concentration of desalted TCO-FT was ~4 mg/mL. After that, 10 mM GDP-Fuc-Tz stock (see Li. et al. 2018 for preparation protocol) were added to the TCO-FT conjugates at a final concentration of 0.15 mM (5 equiv. of TCO-FT). After 2 hours of incubation at room temperature, the one-pot products of GDP-Fuc-FT are ready to use and could be kept at −20 °C for up to 2 months.

General protocol for enzymatic transfer of GDP-Fuc-FT to cell surface

Typically, 1×10⁶ live cells were resuspended in 100 μL HBSS buffer containing 20 mM MgSO₄, 3 mM HEPES and 0.5% FBS. The cells were then treated with 0.2 mg/mL GDP-Fuc-FT (other concentrations were used in titration experiments). After incubated on ice for 20 minutes, cells were washed with DPBS twice and ready for further application or analysis. For confirming the successful labeling in flow cytometry, FT labeled cells were stained with DAPI and PE-anti-His tag (FT, a His-tagged recombinant protein). For the SDS-PAGE fluorescent gel imaging analysis, GDP-Fuc-FT was prelabeled with fluorescent dye AF647 (GDP-Fuc-FT-AF647) using AF647 NHS Ester and used instead of GDP-Fuc-FT. After cells were labeled with GDP-Fuc-FT-AF647 (0.2 mg/mL) on ice for 30 mins, labeled cells or untreated cells were collected and washed three times. After that, 1×10⁵ CHO cells and 1.5×10⁵ DCs, respectively, were labeled with FT-AF647 were lysed in 20 μL SDS loading buffer and loaded to one lane of the SDS-PAGE gel. Quantitative GDP-Fuc-FT-AF647 protein were used as standards for quantification (1 ng, 3 ng and 9 ng GDP-Fuc-FT-AF647 protein). The resolved fluorescent gel was analyzed by ChemiDoc XRS+ (Bio-Rad). According to the semi-quantitative experiment, each CHO cell has incorporated 9×10⁵ FT molecules and each iDCs has incorporated 5×10⁵ FT molecules. (see Li. et al. 2018 for quantitative and calculation details)

Fluorescence microscopy imaging of the FucoID enabled interaction-dependent labeling on CHO cells

CHO cells pre-stained with CFSE were treated with 0.2 mg/mL GDP-Fuc-FT to generate FT functionalized CHO cells (CHO-FT). CHO-FT were mixed with unlabeled CHO cells at the ratio of 1:5 and then seeded in a glass chamber (Thermal Fisher Scientific). Cells were incubated at 37 °C in αMEM (20 mM MgSO₄) for 2 hours followed by a gentle addition of GDP-Fuc-biotin at a final concentration of 50 μM (see Li. et al. 2018 for the GDP-Fuc-biotin preparation protocol). After 20 min incubation, cells were gently washed twice with DPBS and then stained with DAPI and streptavidin-AF647 (SA-AF647). The labled cells were imaged and analyzed using a Keyence BZ-X800 microscope.

Detection of iDC–OT-I CD8⁺ T cell interactions using FT functionalized dendritic cells (iDC-FT)

iDCs (CD45.1^+/+, 1 million) were suspended in 100 μL HBSS buffer (20 mM MgSO₄, 3 mM HEPES and 0.5% FBS) and treated with GDP-Fuc-FT (50 μM). After 20 minutes of incubation on ice, FT labeled iDCs were washed with DPBS twice, resuspended in T cell medium and primed with OVA_257–264 or LCMV GP _33–41 (100 nM or concentrations specified for different experiments as indicated in the related figures) at 37 °C for 30 min. After washed with DPBS twice, iDC-FT with or without antigen priming were co-cultured with purified CD8⁺ T cells or splenocytes from CD45.2 OT-I mice at a ratio of 1:1 (or ratios specified for different experiments as indicated in the related figures). Typically, 20,000 cells were cultured in a single well of a 96-well plate. After a 2-hour incubation (or indicated incubation time periods) at 37 °C, GDP-Fuc-biotin was added gently at a final concentration of 50 μM to initiate the FucoID reaction. The reaction was then quenched with LacNAc (5 mM, final concentration) after 30 min. Cell mixtures were then stained and analyzed by flow cytometry. In the experiments of probing iDC–T cell interactions in splenocytes, iDC-FT primed with OVA_{257 −264} and LCMV GP_33–41 (100 nM), respectively, were co-cultured with WT C57BL/6J mice splenocytes doped with CD45.1^+/+ OT-I T cells (~2% ratio of OT-I T cells in total population) at iDC-FT:T ratios of 1:1 or 3:1 in a 6-well plate. After 2-hour of co-culturing at 37 °C, GDP-Fuc-biotin (50 μM) was added gently with a further 30 min. The reaction was then quenched with LacNAc (5 mM). Cell mixtures were stained and analyzed by flow cytometry.

Detection of antigen-specific dendritic cell-T cell interactions in the mixture of OT-I CD8⁺ T cells and P14 CD8⁺ T cells

iDCs (CD45.2, 1 million) were suspended in 100 μL HBSS buffer (20 mM MgSO₄, 3 mM HEPES and 0.5% FBS) and then treated with GDP-Fuc-FT (0.2 mg/ml). After 20 minutes of incubation on ice, FT functionalized iDCs (iDC-FT) were washed with DPBS twice. iDC-FT were then divided into two groups and primed with OVA_257–264 or LCMV GP_33–41, respectively. After washed with DPBS, iDC-FT were co-cultured with CD8⁺ T cells from CD45.1^+/− OT-I splenocytes and CD8⁺ T cells from Thy1.1^+/− P14 splenocytes at the ratio of 1:1:1 on a 96-well plate (30000 cells per well). After 2-hour incubation at 37 °C, GDP-Fuc-biotin (50 μM) was added gently to initiate the FucoID reaction. The reaction was incubated at 37 °C for 30 min before quenched with LacNAc (5 mM). Cell mixtures were then stained and analyzed by flow cytometry.

Detection of iDC–OT-II CD4⁺ T cell interactions using FT functionalized dendritic cells (DC-FT)

iDCs (CD45.1^+/+, 1 million) were resuspended in 100 μL HBSS buffer (20 mM MgSO₄, 3 mM HEPES and 0.5% FBS) and then treated with GDP-Fuc-FT (50 μM). After 20 minutes of incubation on ice, FT functionalized iDCs (iDC-FT) were washed with DPBS twice. iDC-FT were resuspended in T cell medium and primed with OVA_323–339 or LCMV GP_61–80 (100 nM) at 37 °C for 30 min. After washed with DPBS twice, iDC-FT with or without antigen priming were co-cultured with splenocytes from CD45.2 OT-II mice as a ratio of 1:1 on a 96-well plate (20,000 cells per well). After 4-hour incubation (or indicated times) at 37 °C, GDP-Fuc-biotin (50 μM) was added gently to initiate the FucoID labeling. The reaction was incubated at 37 °C for 30 min before quenched with LacNAc (5 mM). Cell mixtures were then stained and analyzed by flow cytometry.

Identification and enrichment of TSA-reactive TILs via FucoID

Tumor lysates preparation：

A small portion of tumor tissue was mechanically dissociated. After cell counting, the resulting tumor cells suspension (5–10×10⁶/mL) were subjected to three quick frozen-thrown cycles. After centrifuged at 2000×g for 10 min, the supernatant was collected, aliquoted and stored at −80 °C for subsequent antigen priming.

Enrichment of lymphocytes from tumor cell suspension：

Tumors were mechanically dissociated into single cell suspensions and then subjected to a discontinuous Percoll gradient centrifugation for TILs isolation. The resulting lymphocytes layer were washed with DPBS and rested in complete T cell medium containing 100 IU/mL rhIL2 overnight.

FucoID protocol for TSA reactive TILs labeling：

iDCs were primed with the indicated antigens or tumor lysates as described above overnight. After antigen priming, iDCs were labeled with GDP-Fuc-FT (0.2 mg/mL) to generate iDC-FT following the general procedure described above. Enriched TILs were co-cultured with iDC-FT at the ratio of 10:1 in complete T cell medium with 20 mM MgSO₄. After 2 hours of co-culturing (4 hours co-culturing for the labeling of TSA reactive CD4⁺ TILs from Pan02 tumor), GDP-Fuc-biotin (50 μM) was added gently to initiate the FucoID reaction. The cell mixtures were cultured for another 30 mins. At last, LacNAc (5 mM) was added for 20 mins to quench the reaction. The workflow indicating specific congenic markers is summarized in Figure 4B and Figure 5A. Subsequently, labeled TILs were stained before flow cytometric analysis or cell sorting. For CD8⁺ TILs staining, cell mixtures were incubated with mouse Fc-blocker for 10 mins at room temperature, then stained with anti-mCD8a-PE, anti-mCD45.1-PB, anti-mPD-1-FITC and streptavidin-APC. FITC Rat IgG2aκ FITC was used as a PD-1 isotype control. The gating strategy was shown in supplementary figures. For B16-OVA tumor model, PD-1⁻ population: CD8a⁺/CD45.1⁺/PD-1⁻/Biotin⁻; PD-1⁺Bio⁻ population: CD8a⁺/CD45.1⁺/PD-1⁺/Biotin⁻; PD-1⁺Bio⁺ population: CD8a⁺/CD45.1⁺/PD-1⁺/Biotin⁺. For B16, E0771 and MC38 tumor model, PD-1⁻ population: CD8a⁺/CD45.1⁻/PD-1⁻/Biotin⁻; PD-1⁺Bio⁻ population: CD8a⁺/CD45.1⁻/PD-1⁺/Biotin⁻; PD-1⁺Bio⁺ population: CD8a⁺/CD45.1⁻/PD-1⁺/Biotin⁺. In some experiments, all CD8⁺ TILs expressing PD-1 were sorted as the subset of total PD-1⁺. For analyzing the markers of each subsets, cells were divided into three groups after incubating with mouse Fc-blocker. One group of cells were stained with Ghost Dye Violet 510, anti-mCD8a-PB, anti-mPD-1-FITC, streptavidin-APC, anti-mCD39-PE/Cy7, anti-mCD103-PE. Second group of cells were stained with Ghost Dye Violet 510, anti-mCD8a-PE/Cy7, anti-mPD-1-FITC, streptavidin-percp5.5, anti-mCD137-APC, anti-mTIM-3-PE and anti-mTCF1-PB. Third group of cells were stained with Ghost Dye Violet 510, anti-mCD8a-PE/Cy7, anti-mPD-1-FITC, streptavidin-percp5.5, anti-mLAG3-APC. For the labeling/sorting of TILs from Pan02 tumors, labeled cell mixtures were incubated with mouse Fc-blocker for 10 mins at room temperature after FucoID labeling. In CD8⁺ TILs group, cells were stained with anti-mCD8a-PE, anti-mPD-1-FITC and streptavidin-APC. And CD8a⁺Biotin⁺ and CD8a⁺Biotin⁻ populations were sorted by FACS for ELISpot assay. In CD4⁺ TILs group, cells were stained with anti-mCD4-FITC, anti-mCD25-PB and streptavidin-APC. CD25⁻Biotin⁺ CD4⁺, CD25⁺Biotin⁺ CD4⁺ and Biotin⁻ CD4⁺ populations were sorted by FACS for ELISpot assay. For the analysis of functional markers, Fc-blocker treated cells were divided into three groups. The first group was stained with Ghost Dye Violet 510, anti-mCD3-APC/Cy7, anti-mCD4-AF700, anti-mPD-1-FITC, streptavidin-PE/Cy7, anti-mCD137-APC, anti-mCD103-PE, anti-mCD25-PB and anti-mFoxp3-PE. The second group was stained with Ghost Dye Violet 510, anti-mCD3-APC/Cy7, anti-mCD4-AF700, streptavidin-PE/Cy7, anti-mOX40(CD134)-PercpCy5.5 and anti-mCD25-PB. The third group was stained with Ghost Dye Violet 510, anti-mCD3-APC/Cy7, anti-mCD4-AF700, streptavidin-APC, anti-mGITR-PercpCy5.5, anti-Helios-PE/Cy7 and anti-mCD25-PB.

In vivo re-stimulation of different TIL subsets isolated from B16-OVA tumors

iDCs (CD45.2) were used to labeled TILs from male C57BL/6J mice (CD45.1^+/+) under the procedures described above. TILs were sorted into three subsets according to cell surface marker PD-1 and biotinylation: PD-1⁻, PD-1⁺Bio⁻ and PD-1⁺Bio⁺. After cell sorting, three subsets of TILs (50000 TILs per mice) were injected into C57BL/6J mice (CD45.2) through tail vein, respectively. After TILs were rested in vivo for 24 hours, the recipient mice were infected with listeria bacteria expressed OVA antigen (LM-OVA). After another 7 days, blood was collected from each mouse and stained with anti-mCD45.1-FITC, anti-mCD3-PB, anti-mCD8a-PE and H-2Kb/OVA_257–264 MHC tetramer-APC for specific OVA TCR clone expansion analysis. After another 30 days, splenocytes from each mouse were also stained with the same panel of antibodies for TCR clone expansion analysis. At last, antigen specific CD8⁺ T cells were only found from the splenocytes of PD-1⁺Bio⁺ TILs transferred mice. Splenic CD8⁺ T cells from this group were enriched by CD8⁺ isolation kit and were transferred to C57BL/6J mice (CD45.2, 3×10⁶ cells per mice) by tail vein injection. Recipient mice were then infected with LM-OVA on the next day. PBMC from the blood was stained with anti-mCD45.1-FITC, anti-mCD3-PB, anti-mCD8a-PE and H-2Kb/OVA_257–264 MHC tetramer-APC for flow cytometry on day 8 (The workflow is summarized in Figure 4D).

Rapid expansion of TILs

Sorted TILs were rested in complete T cell medium with 100 IU/mL rhIL-2 for 12–18 hours. Then, TILs (6,000 TILs per well) were co-cultured with feeder cells (50 Gy irradiated BALB/c mouse splenocytes, 0.6×10⁶ cells per well) in 300 μL complete T cell medium supplied with 500 ng/mL anti-mCD3 and 1500 IU/mL rhIL-2 in a 48-well plate. On day 5, cells were counted and fresh T cell medium with cytokines were added to maintain cell density at 0.5–1×10⁶ cells/mL. In most experiments, TILs were collected on day 9 or day 10 for analysis.

IFN-γ ELISpot assay for TIL reactivity analysis

ELISpot assays were performed using a mouse IFN-γ ELISpot kit (ab64029, Abcam). DCs were primed with indicated antigens or tumor lysates as described above. Different subsets of TILs were then co-cultured with DCs primed with the indicated antigens respectively (For TILs from B16 tumor: 2500 TILs/10000 DCs; For E0771 tumor: 3000 TILs/12000 DCs; For MC38 tumor: 2500 TILs/10000 DCs; for Pan02 tumor: 1000 CD4⁺ or CD8⁺ TILs/5000 DCs or 1000 CD4⁺/1000 CD8⁺ TILs/10000 DCs (for CD4/CD8 mixture)) in a pre-coated 96-well plate. The final concentrations of GP100_25–33, OVA_257–264 and OVA_323–339 are 100 nM. Anti-MHCI and anti-MHCII were used at a concentration of 200 μg/mL if indicated. After 20 hours of co-culturing, IFN-γ secretion mediated spots were measured according to the kit protocol (https://www.abcam.com/mouse-interferon-gamma-elispot-kit-ab64029.html). Spot pictures of each well were taken using Zeiss KS ELISpot reader and spot numbers were counted by hand using image J (1.50i).

In vitro TIL-mediated tumor killing assays

B16-OVA-luc, B16-luc, E0771-luc or MC38-luc cells (stably transduced with firefly luciferase) were seeded in 96-well plate (10000 cells per well). Expanded TILs were co-cultured with the corresponding cancer cells in complete T cell medium at indicated effector/target ratios for 20 hours. The detection was performed using a luciferase activity kit according to the manufactory’s manual (Bright-Glo, Promega).

QUANTIFICATION AND STATISTICAL ANALYSIS

TCRβ sequencing, mRNA sequencing and analysis

Sorted TILs were cultured in complete T cell medium supplied with 100 IU/mL rhIL2 for 24 hours. RNA samples were then extracted from 10000 cells of each population using Picopure™ RNA isolation kit (Qiagen RNase-Free DNase set were used to digest trace DNA). TCRβ sequencing was performed and analyzed by iRepertoire, Inc. RNA-seq was performed by Next Generation Sequencing Core at Scripps Research. The library for RNA-seq was prepared using SMARTseq HT kit and sequencing was performed using NextSeq500 sequencing platform. The reads were trimmed for the adapter sequences using cutadapt 1.18 with Python 3.6.3 and the trimmed reads were mapped to the reference genome using the STAR aligner 2.5.2a. Gene abundance was estimated with python 2.7.11, and HTSeq 0.11.0. PCA and the differential gene expression analyses between different cell populations was performed using DESeq2 package (Love et al., 2014) in R. Significantly different threshold was Benjamini–Hochberg FDR (p.adjust) <0.05 and |log₂(foldchange)| ≥0.6. Gene ontology over-representation analysis was performed using clusterProfiler package (Yu et al., 2012) in R. GSEA analysis was performed using GSEA 4.0.3 (Broad Institute).

Generation of gene signatures from the literature

For up-regulated exhaustion gene signature, we used the up-regulated gene profiles of exhausted CD8⁺ T cells induced by LCMV chronic infection compared to naïve CD8⁺ T cells, which was published by Wherry et al. (Wherry et al. 2007). For naïve/memory gene module and activation/dysfunction gene signatures, we used the gene modules reported by Singer et al. (Singer et al. 2016). For cell cycle gene signature, we used the cell cycle gene set reported by Li et al. (Li et al. 2019). For up and down-regulated tumor specific antigen gene signatures, we used the gene profiles shared by tumor-specific CD8⁺ T cells from early (day 8) and late stage (day 30) tumors and excluding the gene profiles of exhausted CD8⁺ T cells by chronic infection, reported by Schietinger et al. (see Figure 5B in Schietinger et al. 2016). All gene sets used for GSEA and GSEA results are summarized in Table S1.

Statistical analysis

Statistical analyses were performed using GraphPad Prism software (version 7.0). Comparisons over groups were analyzed using two-way ANOVA followed by Tukey’s multiple comparisons test, and comparisons of multiple samples at one group were analyzed using two tailed t-test or one-way ANOVA followed by Tukey’s multiple comparisons test. Survival data were analyzed by Log-rank (Mantel-Cox) test. In all figures, ns, P > 0.05; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.

Supplementary Material

1

Figure S1. Establishment of the method of conjugating fucosyltransferase (FT) onto the cell surface for probing cell-cell interactions, related to Figure 1.

(A) Conjugating FT onto the cell surface of Chinese hamster ovary (CHO) mutant Lec2 cells. Lec2 cells that express abundant LacNAc were labeled with FT (0.1 mg/mL), GDP-Fuc-FT (0.1 mg/mL) or left untreated (FT is expressed as a His6-tagged recombinant protein). CHO Lec8 cells that do not express LacNAc were used as the control. Robust labeling was only achieved in Lec2 CHO cells treated with GDP-Fuc-FT. (B) Titration of GDP-Fuc-FT concentrations for Lec2 CHO cell labeling. (C) Conjugating FT onto immune cells. Mouse effector CD8⁺ T cells, mouse bone marrow dendritic cells (BMDCs), human CD3⁺ T cells and human DCs (differentiated from monocytes) were successfully labeled by self-catalyzed GDP-Fuc-FT (0.2 mg/mL). (D) Quantification of cell-surface conjugated FT using fluorescent detection on SDS-PAGE gels. GDP-Fuc-FT modified with Alexa Fluor 647 (GDP-Fuc-FT-AF647) were prepared as previously described (Li et al., 2018) and used in the experiments for quantifying the number of FT molecules conjugated onto the cell surface. Approximately 9 × 10⁵ FT molecules were introduced onto the surface of one CHO cell on average and approximately 5 × 10⁵ FT molecules were introduced onto one DC according to the previously reported estimation method. (E) Characterization of the decay of FT on the cell surface. (F) Cell viability analysis of CD8⁺ T cells with or without FT labeling (DAPI staining). (G) Comparison of iDCs before and after FT labeling in priming antigen specific T cells. iDCs derived from CD45.1^+/+ C57BL/6J mice were either labeled with FT then primed with OVA_257–264 or primed with OVA_257–264 then labeled with FT. They were then co-cultured with splenocytes from CD45.2 OT-I mice at cell ratio 1:1. Unlabeled iDCs with or without OVA_257–264 loading were used as control. After 2 hours of co-culturing, cell mixtures were stained with anti-mCD45.1-FITC, anti-mCD8a-PB and anti-mCD69-PE for flow cytometry analysis. Identical upregulation level of CD69 on CD8⁺ T cells of different groups indicates that installation of FT on iDC cell surface did not affect the iDC-mediated T cell activation. (H) Fluorescence imaging of FucoID-enabled intercellular labeling of CHO cells, related to Figure 1C. Scale bar: 10 μm.

2

Figure S2. Optimization of FucoID for probing antigen-specific DC-T cell interactions, related to Figure 2 and 3.

(A) Representative flow cytometric analysis (left) and bar graphs (right) showing the antigen-specific fucosyl-biotinylation of OT-I CD8⁺ T cells by iDCs anchored with different amount of FT on cell surface. Briefly, iDCs (CD45.1^+/+) were labeled with indicated concentrations of GDP-Fuc-FT. Then, iDC-FT loaded with antigen OVA_257–264 or LCMV GP_33–41 (100 nM) were co-cultured with CD45.2 OT-I splenocytes at iDC:T ratio of 1:1 for 2 hours followed the addition of GDP-Fuc-Biotin (50 μM) and another 30 minutes incubation. (B) Representative flow cytometric analysis (left) and bar graphs (right) showing the antigen specific fucosyl-biotinylation of OT-I CD8⁺ T cells by iDC-FT with different co-culturing time before adding GDP-Fuc-biotin. iDCs (CD45.1^+/+) were treated with 0.2 mg/mL GDP-Fuc-FT. Then, iDC-FT loaded with antigen OVA_257–264 or LCMV GP_33–41 (100 nM) were co-cultured with CD45.2 OT-I splenocytes at iDC:T ratio of 1:1 for the indicated time followed by the addition of GDP-Fuc-Biotin (50 μM) and another 30 minutes incubation. (C) Representative flow cytometric analysis (left) and curve graphs (right) showing the antigen concentration-dependent specific fucosyl-biotinylation of OT-I CD8⁺ T cells by iDC-FT in OT-I splenocytes via FucoID system. Experiment procedure is shown in Figure 3A except iDC-FT were primed with indicated amounts of OVA_257–264. (D) Representative flow cytometric analysis (left) and bar graphs (right) showing the antigen-specific fucosyl-biotinylation of OT-I CD8⁺ T cells by iDCs under different iDC-FT:T cell ratios. 100 nM antigens were used to prime iDC-FT. iDC-FT and OT-I splenocytes were co-culture for 2 hours before adding GDP-Fuc-biotin. Data are presented as mean ± SD (n=3); ns, P > 0.05; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; two-way ANOVA followed by Tukey’s multiple comparisons test and one-way ANOVA followed by Tukey’s multiple comparisons test.

3

Figure S3. Interaction dependent fucosyl-biotinylation is not triggered by trogocytosis, related to Figure 3.

Representative flow cytometric analysis and bar graphs showing FT or biotin conjugated on the iDC cell surface was not transferred from iDC to OT-I T cells across cell-cell interface via trogocytosis. The antigen dependent fucosyl-biotinylation of OT-I T cells was catalyzed by FT installed on the interacting iDC cell surface. Briefly, iDCs (CD45.1^+/+) were labeled with 0.2 mg/mL GDP-Fuc-FT to generate iDC-FT or labeled with GDP-Fuc-Biotin (50 μM) using 0.2 mg/mL free FT) to generate iDC-Fuc-Biotin. Then, iDC-FT primed with OVA_257–264 or LCMV_33–41 were co-cultured with OT-I splenocytes (CD45.2) to perform antigen dependent fucosyl-biotinylation as the procedures in Figure 3A. Meanwhile, iDC-Fuc-Biotin primed with OVA_257–264 or LCMV_33–41 were co-cultured with OT-I splenocytes (CD45.2) for 2 hours. Cells were then stained with anti-mCD45.1-FITC, anti-mCD8a-PB, streptavidin-APC and anti-histage-PE for flow cytometry analysis. Data are presented as mean ± SD (n=3); ns, P > 0.05; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; two-tailed t-test.

4

Figure S4. Functional characterization and TCR repertoire profiling of CD8⁺ TIL subsets isolated via FucoID in different murine tumor models, related to Figure 4 and 5.

(A) Representative flow cytometric plots showing the gating strategy of E0771 TILs after FucoID labeling, related to Figure 5B. Similar gating strategies were used for B16-OVA, B16, E0771 and MC38 TILs. (B) Cytotoxic killing effects against B16-OVA cells of the ex vivo expanded PD-1⁻, PD-1⁺Bio⁻ and PD-1⁺Bio⁺ TILs (effector-to-target ratio of 4:1). Labeling and sorting procedures are shown in Figure 4B and C. Sorted TILs from B16-OVA tumor were expanded under a rapid expansion protocol for 7 days before the analysis. Data are presented as mean ± SD (error bars); n=3; ns, P > 0.05; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; one-way ANOVA followed by Tukey’s multiple comparisons test. (C) Representative flow cytometric analysis showing the OVA-antigen specificity of PD-1⁺Bio⁻ and PD-1⁺Bio⁺ TILs isolated from B16-OVA tumors. The isolated TILs were rested in complete T cell medium for 2 days until the complete decay of the biotin tag from the cell surface. The cells were then stained with anti-mCD8a-PE and H-2Kb/OVA MHC Tetramer-APC. (D) Representative TCRβ RNA sequencing results showing the TCRβ diversity of PD-1⁺Bio⁻ and PD-1⁺Bio⁺ TILs from E0771 and MC38 tumors. Each spot in the plot represents a unique clonotype: V-J-CDR3, and the size of a spot denotes the relative frequency. The entire plot area is divided into sub-area according to V usage, which is subdivided according to J usage and then CDR3 frequency, subsequently. The top 10 most abundant CDR3-encoded peptide sequences in each population are shown. Three independent biological replicates were analyzed to confirm the TCR clonotype enrichment in PD-1⁺Bio⁺ TILs. (E) Cell proliferation curves of PD-1⁺, PD-1⁺Bio⁻ and PD-1⁺Bio⁺ TILs under the same rapid expansion protocol. (F) Cytotoxicity evaluation of the ex vivo expanded PD-1⁻, PD-1⁺Bio⁻ and PD-1⁺Bio⁺ TILs that were isolated from B16 and MC38 tumors, related to Figure 5D. Data are presented as mean ± SD (error bars); n=3; ns, P > 0.05; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; two-way ANOVA followed by Tukey’s multiple comparisons test.

5

Figure S5. Bioinformatic analysis of CD8⁺ TIL subsets isolated via FucoID in MC38 murine colon adenocarcinoma model, related to Figure 6 and Table S1.

(A) Gene concept networks showing the enriched biological processes by up- and down-regulated genes in PD-1⁺Bio⁺ vs. PD-1⁺Bio⁻ TILs, related to Figure 6C. The gene concept networks was generated according to gene ontology (GO) over-representation analysis, showing genes involved in each enriched GO term. (B) to (F) Gene set enrichment analysis (GSEA) of the subsets of MC38 TILs using reported gene signatures, related to Figure 6D and Figure 6E and Table S1.

6

Figure S6. Phenotypical characterization of CD8⁺ TIL subsets isolated by FucoID in different murine tumor models using established functional markers, related to Figure 6G.

(A) Gating strategy for defining three TIL subsets (PD-1⁺Bio⁺, PD-1⁺Bio⁻ and PD-1⁻, Biotin signal is generated from the FucoID labeling) from murine MC38 tumor models. Gating strategy for TILs from B16 and E0771 tumor models are the same. (B) Representative flow cytometry analysis showing the functional markers (TIM-3, TCF1, CD39 and CD103) in the three subsets of TILs from MC38 tumor, related to Figure 6G. (C) Representative flow cytometry analysis and bar graphs of functional markers in the three subsets of TILs from B16 melanoma. (C) Representative flow cytometry analysis and bar graphs of functional markers in the three subsets of TILs from E0771 tumor. Data are presented as mean ± SD (n=3); ns, P > 0.05; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; one-way ANOVA followed by Tukey’s multiple comparisons test.

7

Figure S7. Analysis of CD4⁺ TIL subsets isolated via FucoID from Pan02 tumors, related to Figure 7.

(A) Representative flow cytometric analysis and column graphs showing the functional markers (PD-1, CD137 and CD134) of Pan02 CD4⁺ TILs isolated via FucoID, n=3, related to Figure 7B and C. (B) Representative flow cytometric analysis showing tumor antigen-dependent fucosyl-biotinylation of CD4⁺ TILs from Pan02 tumors inoculated in B6-Foxp3^EGFP mice that co-express EGFP and the regulatory T cell-specific transcription factor Foxp3 with the expression of EGFP restricted to the T cell lineage. (C) IFN-γ ELISpot showing tumor reactivities of the individual TIL subsets and a combination of two subsets of TILs isolated from Pan02 tumors. Re-stimulations by tumor lysates primed DC were conducted using the individual TIL subsets and a combination of two subsets as specified. n=3. Average spot numbers of each condition were shown on the heatmap, related to Figure 7D.

8

Table S1. GSEA ranking gene lists

All gene sets used for GSEA and corresponding GSEA results, related to Figure 6 and S5.

Highlights.

Fucosyl-biotinylation enables the detection of endogenous tumor antigen-specific T cells

Dysfunctional tumor antigen-specific CD8⁺ T cells upregulate steroid biosynthesis genes

Tumor antigen-suppressive and reactive CD4⁺ T cells co-exist in tumor microenvironments

Intratumoral bystander T cells are separated into two groups based on PD-1 expression