A Metabolic Gene Survey Pinpoints Fucosylation as a Key Pathway Underlying the Suppressive Function of Regulatory T Cells in Cancer

The physiologic role of regulatory T cells (Treg) in the immune system is to suppress the immune response to self-antigens and prevent excessive immune reactions (1). However, Tregs are increased in the tumor and in the blood of patients with cancer, and a higher number of Tregs has been correlated with poor prognosis in various tumor types (2, 3). The importance of Tregs in both cancer and autoimmunity has led to increased interest in targeting these cells to improve disease outcomes.

Tregs mainly originate from the thymus, but a subset of Tregs also develops in the periphery from CD4⁺ T cells upon exposure to environmental stimuli, such as in the tumor. In vitro–induced Tregs (iTreg) can be generated from naïve CD4⁺ T cells exposed to IL2 and TGFβ. Tregs are characterized by the expression of the lineage-specific transcription factor Forkhead box P3 (Foxp3), which is crucial for their maturation and suppressive function (4). Other molecules are also important for Tregs to exert their suppressive function, such as efficient T-cell receptor (TCR) stimulation, CTLA-4, PD-1, CD39, and CD73 expression, secretion of inhibitory cytokines, and others (5). However, Tregs under certain conditions, can lose their suppressive functions, which can be accompanied by reduction or loss of Foxp3 expression, known as ‘’unstable’’ Tregs, or by maintenance of Foxp3 expression, known as ‘’fragile’’ Tregs (6). Remarkably, Tregs’ suppressive activity has been associated with their metabolism (7). Indeed, intracellular metabolism of Tregs affects Foxp3 expression and Treg stability (8–10). Vice versa, Foxp3 expression can alter Treg metabolism by impairing glycolysis and promoting oxidative phosphorylation (11). Despite these and other significant efforts in understanding the metabolic needs of Tregs in cancer and autoimmune diseases (8–12), a broad study of the metabolic factors regulating Treg functionality is currently missing. Here, to study the connection between the metabolism of Tregs, the maintenance of Foxp3 expression, and their stability, we performed an in vitro CRISPR/Cas9 metabolic screen in Tregs, followed by the validation of the main target, Slc35c1, both in mice and humans.

HEK293 cells were provided from ATCC (catalog no. CRL-1573) in 2018, and MC38 and CT26 cells were provided from Kerafast (catalog no. ENH204-FP) in 2016. For the generation of CD90.1⁺ CT26 cells was used the plasmid LentiGuideThy1.1 (Addgene, 128063) containing the nontargeting sequence (5′ GAACAGTCGCGTTTGCGACT 3′). CT26 cells were transduced with lentivirus and the CD90.1⁺ CT26 cells were FACS sorted 7 days upon transduction and used for the in vivo experiments. HEK293 and MC38 cells were cultured in DMEM (Gibco Thermo Fischer Scientific, 41965039) supplemented with 10% heat-inactivated FBS (Biowest, S1810), 100 U/mL penicillin, 100 μg/mL streptomycin (Gibco Thermo Fischer Scientific, 15140122), and 2 mmol/L glutamine (Gibco Thermo Fischer Scientific, 25030024). CT26 and CD90.1⁺ CT26 cells were cultured in RPMI 1640 (Gibco, Thermo Fischer Scientific, 21875034) supplemented with 10% heat-inactivated FBS, 100 U/mL penicillin, and 100 μg/mL streptomycin. All cells were frozen between passage 4 and 6 after purchase. Cells were cultured at 37°C and 5% CO₂. All the cell lines were passaged in the laboratory for no longer than 10 passages after receipt and tested for Mycoplasma by PlasmoTest-Mycoplasma Detection kit (InvivoGen) every 6 months.

C57BL/6J females were provided from the internal stock of KU Leuven, Balb/c females were obtained from Charles River, and C57BL/6J Rosa26-Cas9 (B6J.129(Cg)-Gt(ROSA)26Sortm1.1(CAG-cas9*,-EGFP)Fezh/J) mice were obtained from the Jackson Laboratory. Rosa26-Cas9 mice were crossed with Ly5.1⁺ mice for use in in vivo suppressive assays as described below. Ly5.1⁺ Foxp3^Thy1.1 and Ly5.1/2⁺ Foxp3^Thy1.1 mice were provided from the lab of Prof. Susan Schlenner (KU Leuven-University of Leuven, Department of Microbiology, Immunology and Transplantation, Leuven B3000, Belgium; ref. 13), and Pdcd1^−/− mice were provided by Prof. Oldenhove (Laboratory of Immunobiology, Université Libre de Bruxelles, Gosselies, Belgium). Rag2^−/− (C57BL/6N-Rag2Tm1/CipheRj) mice were obtained from Janvier labs. Euthanasia was performed by cervical dislocation. Housing and all the experimental animal procedures were approved by the Animal Ethics Committee of the KU Leuven (ECD number 226/2017 and 102/2021).

Total lymphocytes were isolated from spleen, inguinal, and axillary lymph nodes from naïve Rosa26-Cas9 or C57BL/6J mice depending on the experiment. Spleens were processed on a 40-μm pore-sized strainer in sterile PBS, and cells were centrifuged for 5 minutes at 400 × g. Red blood cell lysis was performed by using Hybri-Max (Sigma-Aldrich, R7757) and cells derived from spleens and lymph nodes were mixed. Naïve CD4⁺ T cells were isolated using a naïve CD4⁺ T-cell negative selection kit (Stemcell Technologies, 19765) with the use of EasySep magnets according to the manufacturer's instructions, and the purity was >92% checked by FACS. For naïve CD4⁺ T-cell activation, 24-well plates were precoated with 2 μg/mL anti-CD3 and 5 μg/mL anti-CD28 (BD Biosciences, 557393) for 2h at 37°C and then washed 1X with 1 mL PBS per well. The anti-CD3 was made in-house (clone 125–2C11), purified according to standard procedures for IgG purification, and subsequently treated with LPS removal beads, followed by performing a LAL assay (Cambrex) to confirm it was LPS-free (kindly provided by Prof. Stijlemans; Lab of Cellular and Molecular Immunology, Vrije Universiteit Brussel, Belgium). Naïve CD4⁺ T cells were resuspended and seeded at 0.5 million/mL in T-cell medium [RPMI 1640 supplemented with 10% heat-inactivated FBS, 100 U/mL penicillin, 100 μg/mL streptomycin, 25 μmol/L β-mercapto-ethanol (Gibco, Thermo Fisher Scientific, 60–24–2), 1% minimum essential medium (MEM) nonessential amino acids (NEAA, Gibco, Thermo Fisher Scientific 11140035), and 1 mmol/L sodium pyruvate (Gibco, Thermo Fisher Scientific, 11360070)]. For iTreg induction, naïve CD4⁺ T cells were cultured in the presence of murine recombinant IL2 (Peprotech, 212–12) at final concentration 20 ng/mL and human recombinant TGFβ (R&D systems, 7754-BH-005/CF) at 5 ng/mL for 3 to 5 days.

In vivo Tregs were isolated from spleens and lymph nodes of 6 to 8 weeks old naïve female C57BL6/J, Rosa26-Cas9, Ly5.1⁺ Foxp3^Thy1.1 and Ly5.1/2⁺ Foxp3^Thy1.1 and Pdcd1^−/− mice, depending on the experiment as previously mentioned. When high numbers of Tregs were needed, their number was expanding in vivo upon intraperitoneal injection of the complex: anti-mouse IL2 (clone: JES6–1A12, BioLegend, 503706) with murine IL2. Briefly, for each mouse 20 μg of IL2 was mixed with 10 μg of the anti-mouse IL2, and the complex was incubated for 30 minutes at 37°C and then injected intraperitoneally to the mice. The injection was repeated for 3 constitutive days followed by 2 days of noninjection and then the mice were sacrificed (14). Cells from spleens and lymph nodes were dissociated as stated above. Cells were resuspended in 1X Mojosort buffer (BioLegend, 480017), followed CD4⁺ T-cell negative selection with the use of the Mouse CD4⁺ T-cell isolation kit (Mojosort, 480033) and Mojosort magnets with purity > 95% checked by FACS. Then Tregs were sorted as live CD4⁺CD25^high or CD4⁺Thy1.1⁺ (fixable viability dye efluor 506, CD4 PerCpCy5.5, CD25 PE, CD90.1 APC-Cy7) depending on the mice used, using a BD FACS Aria II or BD FACS Aria Fusion. In vivo–derived sorted Tregs were seeded at 1 million/mL into 24-well plates and activated with anti-CD3/CD28-coated Dynabeads (Thermo Fischer Scientific, 11453D) at a ratio 3:1 beads to cells for 3 days in the presence of human IL2 (Peprotech, 200–02) at final concentration of 2,000 Units/mL.

For perturbing the metabolism of Tregs, we used a library of 10,640 optimized single-guide RNA (sgRNA) targeting 2,078 metabolic genes (5 sgRNAs per gene), including 46 essential and 41 nonessential control genes plus 250 nontargeting control sequences (15, 16). The library was generated by integrating the metabolic genes from the Kyoto Encyclopedia of Genes and Genomes with previously published high-quality reconstructions of mouse metabolism (17). The corresponding sgRNAs were designed according to previously published criteria (CRISPick genetic perturbation platform, Broad Institute; ref. 17), synthesized as 79-mer oligo pools (CustomArray, Bothell, WA), and amplified by PCR. Using a one-step digestion (Esp3I) and ligation reaction, the purified library PCR pool was then cloned into lentiGuide-Thy1.1 (Addgene, 128063), which coexpressed the congenic marker CD90.1 along with a sgRNA. The ligation reactions were purified and transformed into Endura ElectroCompetent cells (Biosearchtech, 602420) ensuring a proper library representation. To confirm library representation and distribution, the pDNA was sequenced by Illumina HiSeq. After mapping of Illumina reads, we calculated the overall fraction of reads that contained intended sgRNAs, which served as a surrogate for the quality of the oligonucleotide synthesis. The pooled plasmid library was then used to produce lentiviral particles according to standard procedures, as described below. The library we produced has been reported by another lab for different purposes (18).

HEK293 cells were seeded in 150-mm dishes at 9 million cells/dish in DMEM supplemented with 10% FBS, 100 U/mL penicillin, 100 μg/mL streptomycin, and 1% GlutaMax (Gibco, Thermo Fisher Scientific, 35050061). One day later, the medium was changed to IMDM (Gibco, Thermo Fisher Scientific, 12440053) supplemented with 10% FBS, 100 U/mL penicillin, 100 μg/mL streptomycin, and 1% GlutaMax, and each dish was transfected with 7 μg of envelop plasmid pMD2 VSV_G (Addgene, 12259), 16.25 μg packaging plasmid psPAX2 (Addgene, 12260), and 32 μg of sgRNA transfer plasmid. A lentiGuide-Thy1.1 vector was used containing the Slc35c1_s_08724 sgRNA (5′ GCAGTGAGGTCACCAGGCAT 3′) or the nontargeting control sgRNA (5′ GAACAGTCGCGTTTGCGACT 3′). Cell culture media was replenished 16 hours after transfection with fresh IMDM complete media supplemented with 10 mmol/L sodium butyrate (Merck, 303410). The lentiviral supernatant was collected 48 and 72 hours posttransfection and was concentrated with ultracentrifugation (ultracentrifuge: OPTIMA, XPN-80 and 50.2 Ti Rotor) at 302000 x g for 30 minutes at 4°C and titrated in HEK293 cells.

HEK293 cells were seeded in 150-mm dishes at 13.5 million cells/dish in RPMI supplemented with 10% FBS. The day after, the medium was replaced, and the cells were transfected using Genejuice Transfection reagent (Merch Millipore, 70967–4) with 10 μg sgRNA transfer plasmid and 10 μg of ecopac packaging plasmid (Addgene, 12371) per dish. The following morning, the medium was refreshed. 48 and 72 hours post transfection, the cell supernatant was collected and concentrated 20X using Sartorius Vivaspin concentrators (VS2012) and centrifugation. For the Pdcd1 reconstitution of Pdcd1^−/− Tregs, the PD-1 GFP wild-type (WT) and PD-1 GFP N49Q/N74Q double-mutant constructs were kindly offered by Prof. Yoshimura (Department of Microbiology and Immunology, Keio University, Japan) and generated as described in a previous publication without any modification (19).

For Treg lentiviral transduction, iTregs or in vivo–derived Tregs were seeded in 24-well plates at 0.5 million cells/well, and lentivirus with 1 μg/mL protamine sulfate (Sigma-Aldrich, P3369) was added to the top of each well. The plate was centrifuged for 1h at 1,000 × g at room temperature and incubated overnight at 37°C. The day after transduction, the cells were collected, and the T-cell medium with IL2 was refreshed. Transduced cells were sorted 3 to 4 days post transduction as live CD90.1⁺ cells as previously described and used for experiments. For Pdcd1^−/− Treg retroviral transduction, the same process was followed, except that 2 μg/mL polybrene (Sigma-Aldrich, H9268) was used instead of protamine sulfate.

Tregs were sorted from spleens and lymph nodes of naïve 6 to 8 weeks old Pdcd1^−/− female mice and cultured as previously described. 24 hours post activation, cells were transduced with the retroviral vector encoding PD-1 fused to GFP (PD-1 WT) or PD-1 N49Q/N74Q double-mutant fused to GFP (PD-1 mutant, referred as PD-1 MUT) construct. The day after transduction, medium was refreshed, and 72 hours posttransduction, PD-1 WT and PD-1-mutant Tregs were cultured with 130 μmol/L 2-peracetyl fucose (2FF, Sigma-Aldrich, 344827) or control DMSO (Sigma-Aldrich, D2438) for 3 days. PD-1–reconstituted Tregs were sorted as live (viability efluor 506, CD4 PerCpCy5.5, CD90.1 APC-Cy7 and endogenous GFP) using a BD FACS Aria II. Sorted GFP⁺ Tregs with purity > 95% and used for in vivo suppression assays as described.

Naïve CD4⁺ T cells were isolated from spleen and lymph nodes of 6 to 8 weeks old Rosa26-Cas9 female or male mice and induced into iTregs as described previously. Five days after polarization, enrichment of the polarized cells was performed using a CD4⁺CD25⁺ Tregs enrichment kit (CD4⁺CD25⁺ Regulatory T cell isolation kit, Miltenyi, 130–091–041), and the purity of the population was confirmed by flow cytometry staining for Foxp3 (Invitrogen, 17–5773–80). iTregs were transduced with the lentiviral library as described, and a multiplicity of infection (MOI) reaching approximately 30% of transduction was used to guarantee that each cell was carrying a sgRNA. The percentage of transduction was checked by flow cytometry with the use of anti-CD90.1 (BioLegend, 202520 or 202524). The number of infected cells was carefully estimated to maintain a library complexity ≥ 500X. Five days after transduction, the cells were stimulated with a stimulation cocktail (Thermo Fischer Scientific, 00497003) and Brefeldin A (BioLegend, 420601) for 4 hours at 37°C. After the incubation, cells were washed with FACS buffer and surface-stained for viability (Thermo Fisher Scientific, 65–0866–14), CD4 (BioLegend, 100540), CD90.1 (BioLegend, 202520), followed by fixation and permeabilization with the eBioscience Foxp3/Transcription Factor Fixation/Permeabilization kit (Thermo Fischer Scientific, 00–5521–00B) according to the manufacturer's instructions. Cells were incubated overnight at 4°C with intracellular antibodies Foxp3 (Invitrogen, 17–5773–80) and IFNγ (Invitrogen, 25–7311–41) in 1X permeabilization buffer. Cells were isolated using FACS by gating as live, CD4⁺, CD90.1⁺, Foxp3^high or Foxp3^low, and IFNγ⁺ or IFNγ⁻. The pellets of sorted cells were frozen at −20°C until the gDNA extraction was performed. The CRISPR/Cas9 screening was performed twice to achieve a sufficient number of reads. For each repetition, 1 million cells from each of the 2 populations, Foxp3^high and Foxp3^low, were analyzed.

For gDNA extraction, the DNeasy Blood & Tissue kit (Qiagen, 69506) was used, following the manufacturer's instructions. The extracted DNA was resuspended in ddH2O and measured by Nanodrop. We then performed PCR of gDNA to attach sequencing adaptors and barcode samples. For the amplification of all the samples was used a common mix of P5 primers and a P7 primer with unique barcode for each sample (Supplementary Table S1). For each sample, the gDNA was split into multiple 25 μL PCR reactions containing a maximum of 1 μg gDNA. PCR mixture per reaction: 12.5 μL KAPA HiFi HotStart ReadyMix PCR kit (Kapa Biosystems, KK2602), 1 μL of P5 stagger primer mix (stock at 10 μmol/L concentration), 1 μL of a uniquely barcoded P7 primer (stock at 10 μmol/L concentration), adding mQ water and gDNA input (max 1 μg per reaction) to 25 μL P5 and P7 primers were synthesized at Integrated DNA Technologies (IDT). The PCR cycling conditions were as follow: 98°C for 2 minutes (initial denaturation); 5 cycles of 98°C for 30 sec (denaturation), 60°C for 30 sec (annealing), and 72°C for 30 sec (extension); 25 cycles of 98°C for 30 sec (denaturation), 65°C for 30 sec (annealing), and 72°C for 30 sec (extension); and a final extension of 72°C for 5 minutes. An average of 15 PCR reactions per sample was performed, and the pooled PCR products were purified with Agencourt AMPure XP SPRI beads according to the manufacturer's instructions (Beckman Coulter, A63881).

The purified PCR fragments were quantified with the QC Qubit DNA High Sensitivity kit, and samples were equimolarly pooled and subjected to Illumina next-generation sequencing (NGS) by an Illumina NextSeq500 instrument following these parameters: High Output v2, 75bp, Single Reads, 1.1 pmol/L + 34.1% PhiX. Mapped read-counts were subsequently used as input for the MAGeCK analysis software package.

MAGeCK-VISPR (version 0.5.3) was used to process CRISPR/Cas9 screen NGS data (20). The MAGeCK ‘count’ module generated a raw count table with sgRNA as rows and samples as columns. This table was normalized by 250 nontargeting control guides, and corrected for batch effects by Combat (21, 22). The MAGeCK ‘test’ module used all Foxp3^high samples as treatment and Foxp3^low as control, and the robust ranking aggregation (RRA) algorithm to perform positive and negative selection. Log₂ fold-change (LFC) for each gene was calculated as the median LFC of the 5 sgRNAs. Gene LFCs were further centered and scaled to LFC z-scores by R (version 3.6.2), which were used together with P values for positive and negative selections to select regulators and to generate plots shown in Fig. 1. Quality control, gene level, and sgRNA level positive and negative selection results can be found in Supplementary Fig. S1. Samples derived from females or males were pooled bioinformatically. Also, Foxp3^highIFNγ⁺, Foxp3^highIFNγ⁻ were bioinformatically pooled as Foxp3^high and Foxp3^lowIFNγ⁺, Foxp3^lowIFNγ⁻ were bioinformatically pooled as Foxp3^low.

The validation of the screening was performed with the use of electroporation (23). CRISPR-Cas9 RNAs (crRNA) and trans-activating crRNAs (tracrRNAs, IDT, 1072534) were synthesized by IDT. Lyophilized RNA was resuspended in Nuclease-Free Duplex Buffer (IDT, 11–01–03–01) in final concentration 100 μmol/L. The 2 RNA oligos (crRNA and tracrRNA) were mixed at equimolar concentrations to a final duplex concentration 50 μmol/L and annealed at 95°C 5 minutes, 90°C 2 minutes, 85°C 2 minutes, 80°C 2 minutes, 75°C 2 minutes, 70°C 2 minutes, 65°C 2 minutes, 60°C 2 minutes, 55°C 2 minutes, 50°C 2 minutes, 45°C 2 minutes, 40°C 2 minutes, 35°C 2 minutes, 30°C 2 minutes, 25°C infinite in a Doppio Gradient Thermocycler (VWR, 31141–01A00037). The crRNA:tracrRNA duplex was mixed with cas9 enzyme (IDT, 1081059) at 3:1 duplex:cas9 ratio and incubated for 20 to 30 minutes at room temperature to form the ribonucleoprotein complex. As negative control was used the CRISPR-Cas9 Negative control crRNA (IDT, 04204645). iTregs were electroporated 3 days after stimulation and polarization from naïve CD4⁺ T cells. Five million Tregs were resuspended in 100 μL of P4 primary Lonza electroporation buffer in P4 primary (Lonza, V4XP-4024) per cuvette on the Lonza 4D Nucleofector System with pulse code CM137. Immediately after electroporation, cells were resuspended in T-cell medium supplemented with 20 ng/mL IL2. The day after, T-cell medium and IL2 were refreshed, and cells were seeded in 24-well plates at 0.5 million cells/mL per well. The gRNA sequences used for electroporation are listed at Supplementary Table S1.

RNA was extracted using the RNeasy Micro kit (Qiagen, 74004) according to the manufacturer's instructions. Reverse transcription was performed with the Superscript III First Strand cDNA Synthesis Kit (Life Technologies, 18080051) in a Doppio Gradient Thermocycler (VWR, 31141–01A00037) according to the manufacturer's instructions. cDNA, primers, and Power Up SYBR Green Master Mix (Thermo Fischer Scientific, A25780) were prepared in a volume of 20 μL. Samples were loaded into an optical 96-well Fast Thermal Cycling Plate, and RT-qPCR was performed using a QuantStudio 12K Flex Real-Time PCR System (Applied Biosystems). Samples were run in technical duplicates. Data was normalized to housekeeping gene expression for β-actin. The probes used for PCR are listed in Supplementary Table S1.

In vivo Tregs were isolated from spleen and lymph nodes of 6 to 8 weeks old naïve female Rosa26-Cas9 mice. Tregs were transduced by lentivirus expressing the control sequence (sgNT) or sgSlc35c1. Four days upon transduction, transduced Tregs (CD90.1⁺) were sorted. cas9⁺CD45.1⁺CD4⁺CD25⁻ cells were isolated from spleen and lymph nodes of Ly5.1⁺ Rosa26-cas9 female mice with the use of the CD4⁺CD25⁺ Treg isolation kit as the CD4⁺CD25⁻ fraction (Miltenyi, 130–091–041). Cas9⁺CD8⁺ T cells were isolated from spleen and lymph nodes of Rosa26-cas9 female mice with the use of the Mouse CD8⁺ T Cell Isolation kit (Mojosort, 480008). 0.5 million Tregs, with 1 million cas9⁺CD45.1⁺CD4⁺CD25⁻ cells and 0.5 million cas9⁺CD8⁺ T cells, were resuspended in final volume 200 μL PBS and transferred by tail vein injection in each Rag2^−/− mouse. The purity of the injected cells after isolation was tested by flow cytometry and was > 95%. In the mixed cell model, the two subtypes of Tregs, sgNT and sgSlc35c1, were mixed in ratio 50:50 to maintain that total amount at 0.5 million Tregs. The day after adoptive cell transfer (ACT), each mouse was injected subcutaneously on the right flank with 0.5 million MC38 cells resuspended in 100 μL PBS. The tumor volume was measured 3 times per week by electronic calipers, and tumor volume was calculated with the formula V = π × (d² × D)/6, where d is the minor tumor axis and D is the major tumor axis. 17 days after tumor injection, mice were sacrificed, and tumors and spleens were weighted and collected for immune phenotyping by flow cytometry. Isolated cells were stimulated with cell stimulation cocktail (Thermo Fischer Scientific, 00497003) and incubated with Brefeldin A (BioLegend, 420601) for 4 hours at 37°C before staining as described below.

To assess the effect of a fucosylation (Fuco) inhibitor on Tregs function, in vivo–derived Tregs from Ly5.2 Foxp3^Thy1.1 mice were sorted as CD4⁺CD90.1⁺ (here used as Foxp3 reporter) and activated with anti-CD3/CD28-coated Dynabeads in the presence of IL2 and 2F-peracetyl-fucose at final concentration of 100 μmol/L or DMSO as control for 6 days in total. At day 3, media was replenished with fresh media containing IL2 and 2FF or DMSO. At day 6, the cells were tested for lectin staining by flow cytometry as previously described. 1 million 2FF- or DMSO-treated Tregs were co-injected with 2 million CD45.1⁺CD4⁺CD25⁻ cells and 1 million CD45.1⁺CD8⁺ T cells resuspended in 200 μL PBS by tail vein injection in each Rag2^−/− mouse. CD45.1⁺CD4⁺CD25⁻ and CD45.1⁺CD8⁺ T-cell were isolated as previously described. The purity of the injected cells after isolation was tested by flow cytometry and was >95%. The following day, each mouse was injected subcutaneously with 0.5 million MC38 cells resuspended in 100 μL PBS, and the tumor growth was monitored as previously described. In the end stage mice were sacrificed and tumors were collected for analysis by flow cytometry.

8 weeks old Balb/c females were injected with 1 million CD90.1⁺ CT26 cells resuspended in 100 μL PBS subcutaneously in the right flank. When tumor size reached 150 mm³, mice were treated intraperitoneally with 1.25 mg/kg 2FF (Synchem, SC36700) or saline control twice per day until the end of the experiment. In the experiments where Tregs were depleted, at day 5 and day 7 after tumor injection, the mice were treated with 10 mg/kg anti-CD25 (Evitria, clone 7D4; ref. 24) or control rat serum IgG (Sigma-Aldrich, I4131) via intraperitoneal injection, and at day 8 mice were treated with 2FF or saline twice per day until sacrifice. At the end stage, tumors were weighted and collected for flow cytometric analyses.

Naïve CD4⁺ T cells were isolated from Ly5.1⁺ Foxp3^Thy1.1 or from Ly5.1/2⁺ Foxp3^Thy1.1 congenic mice and polarized into iTregs for 3 days as described. iTregs were electroporated with the control sequence or with an RNP targeting Slc35c1, and 4 days after electroporation, live, CD90.1⁺ (Foxp3 reporter) Tregs were sorted from each condition as previously indicated. Two million CD45.1⁺ and CD45.1/2⁺ sorted iTregs per condition were mixed in a 50:50 ratio and co-injected via tail vein injection in healthy Rag2^−/− mice. The ratio of the two types of injected iTregs was checked by flow cytometry before injection. After 7 days, blood and spleen of the Rag2^−/− mice were collected and analyzed by flow cytometry. Blood collection was performed in Eppendorf's containing 10% heparin by retro-orbital bleeding. For the competition assay in MC38 tumor-bearing Rag2^−/− mice, when tumors reached 100 mm³, two million sorted iTregs per condition were mixed in a 50:50 ratio and co-injected as described. Seven days after adoptive cell transfer, mice were sacrificed, and the tumors were processed for flow cytometric analysis.

Mice were sacrificed, and depending on the experiment, tumors and/or spleens were collected. Tumors were digested after incubation for 30 minutes at 37°C in a MEM (Lonza BE12–169F) containing 1% penicillin/streptomycin, 5% FBS, 50 μmol/L β-mercaptoethanol, 0.85 mg/mL Collagenase V (Sigma-Aldrich, C9263–1G), 1.25 mg/mL Collagenase D (Roche, 11 088 882 001), 1 mg/mL Dispase (Gibco, Thermo Fisher Scientific, 17105–041), and 5 U/mL DNase I (Roche, 10104159001). The digested tissues were filtered using a 70-μm pore-sized strainer, and cells were centrifuged for 5 minutes at 300 × g. Spleens were processed on a 40-μm pore-sized strainer in sterile PBS, and cells were centrifuged for 5 minutes at 300 × g. Red blood cell lysis was performed by using Hybri-Max (Sigma-Aldrich, R7757). For both tumor and spleen-derived single cells were resuspended in FACS Buffer [PBS containing 2% FBS and 2 nmol/L EDTA (Sigma-Aldrich, 03690)] and incubated for 15 minutes on ice with Fc Block anti-mouse CD16/32 (BD Pharmingen, 553142). Surface staining was performed for 30 minutes on ice. For intracellular staining, cells were fixed and permeabilized with the eBioscience Foxp3/Transcription Factor Fixation/Permeabilization kit (Thermo Fischer Scientific, 00–5521–00) according to the manufacturer's instructions, and cells were incubated overnight at 4°C with the intracellular antibodies in 1X permeabilization buffer. Cells were subsequently washed and resuspended in FACS buffer before flow cytometric analysis on a Fortessa X-20. Data were analyzed by FlowJo (TreeStar, Version 10.7). Fluorescence Minus One (FMO) or isotype controls were utilized to ensure proper gating of positive populations. For lectin staining, cells were fixed with 4% paraformaldehyde (Sigma-Aldrich, 252549) for 20 minutes at room temperature and then stained with biotinylated Aleuria aurantia lectin (AAL, Vector Laboratories, B1395–1, 1:1,000) for 30 minutes on ice. Cells were washed with lectin-binding buffer (PBS with 0.1 mmol/L CaCl₂) and stained with secondary antibody – Avidin, Neutravidin PE conjugate (Thermo Fisher Scientific, A2660). As a negative control for lectin staining, Tregs were incubated for 120 minutes at 37°C with 3 mg/mL fucosidase (house in-made, kindly provided by Prof. Callewaert, Department of Biochemistry and Microbiology, Ghent University, Belgium), followed by lectin staining. For the detection of MC38 tumor-specific CD8⁺ T cells, single-cell suspensions of tumors and spleens were stained with PE-conjugated H-2Kb MuLV p15E tetramer-KSPWFTTL (MBL, TB-M507–1) for 30 minutes, followed by extracellular staining, including anti-CD8 (clone KT15) compatible for co-staining with an MHC tetramer. Absolute number of cells was counted by flow cytometry with the use of Precision Count Beads (BioLegend, 424902). All the antibodies used are listed in Supplementary Table S2.

Naïve CD4⁺ T cells were isolated from Ly5.1⁺ Foxp3^Thy1.1 mice and polarized to iTregs for 3 days as described. iTregs were electroporated with the control sequence or with an RNP targeting Slc35c1, and 4 days after electroporation, live, CD90.1⁺ (Foxp3 reporter) Tregs were sorted from each condition. Sorted iTregs were labeled with either 3.5 μmol/L Cell Tracer Violet (CTV) (Thermo Fisher Scientific, C34557) or 1 μmol/L carboxyfluorescein diacetate succinimidyl ester (CFSE; Thermo Fisher Scientific, C34570). For CFSE labeling, cells were stained in PBS for 8 minutes at room temperature. For CTV labeling, cells were stained for 20 minutes at 37°C. Two million labeled sgNT iTregs were co-injected with 2 million labeled sgSlc35c1 iTregs by tail-vein injection in MC38 tumor-bearing (tumor size 100 mm³) Rag2^−/− mice. After 20 hours spleen, and tumors were collected and analyzed by flow cytometry. Blood was also collected in Eppendorfs containing 10% heparin with capillary pipettes by retro-orbital bleeding. To exclude probe-specific effects, the fluorescent cell tracers were switched accordingly, and the ratio of the 2 types of iTregs injected was checked by flow cytometry before injection.

In vivo–derived Tregs were isolated from Rosa26-Cas9 mice, and cells were activated and transduced with lentivirus targeting sgNT or sgSlc35c1 as previously described. At day 4 after transduction, transduced cells (CD90.1⁺) were sorted and placed in round-bottom 96-well plate for coculture with cas9⁺CD45.1⁺CD4⁺CD25⁻ effector T cells labeled with CTV and Rag2^−/− splenocytes in the presence of 1 μg/mL soluble anti-CD3 for 72 hours. For ratio 1:1 were used 10⁵ Tregs with 10⁵ CD4⁺ effector and 5*10⁴ Rag2^−/− splenocytes per well. The suppressive function of Tregs on CD4⁺ effector T cells was measured by flow cytometry as the serial dilution of CTV, which indicated proliferation of CD4⁺ effector T cells.

For the EdU proliferation assay, in vivo–derived Tregs were activated and transduced with lentivirus targeting sgNT or sgSlc35c1 as previously described. Four days after transduction, Tregs were cultured in the presence of 10 μmol/L EdU for 16 hours. After 16 hours, cells were collected and first extracellularly staining, followed by the Click-iT EdU assay protocol (Thermo Fisher Scientific, C10419), as suggested by the manufacturer, and analyzed by flow cytometry. For the in vivo EdU proliferation assay in MC38 tumor-bearing mice, 0.5 million MC38 cells resuspended in 100 μL PBS were injected subcutaneously in Rag2^−/− mice. 7 days upon tumor injection, 2 million CD45.1 sgNT or CD45.1/2 sgSlc35c1 iTregs were transferred in the tumor-bearing Rag2^−/− mice by tail vein injection. Following, 7 days upon T cell transfer, mice were injected intraperitoneally with 50 mg/kg EdU and sacrificed 16 hours later. Blood, spleen and tumor were collected for analysis by flow cytometry. For the proliferation assay using Incucyte Live-Cell analysis system, in vivo–derived Tregs were transduced and sorted as specified previously. Followed, Tregs were seeded in nontreated 48-well plates, coated with 20 μg/mL RetroNectin (Takara Bio, T100B) and blocked with 2% BSA (Sigma-Aldrich, 9048–46–8) in PBS. Cell growth was monitored with an S3 Incucyte for 72 hours. The percent confluency of the cells was calculated as the confluency of the cells at 72 hours (t72) minus the confluency at 0 hours (t0).

In vivo–derived Tregs 4 days after transduction were collected, washed with 1X Annexin V Binding Buffer (BioLegend, 422201), followed by staining with Annexin V (BioLegend, 640941, 1:100) in the binding buffer. After 15 minutes at room temperature, 1 μL of propidium iodide (Sigma-Aldrich, P4864, stock 1 mg/mL) was added, and the samples were analyzed by flow cytometry.

Colorectal cancer patient (n = 23) derived tumor single-cell expression data (normalized values with 5th and 95th percentile as min and max), accessed from a published dataset (GSE132465), was processed using a standardized and uniformized workflow (BBrowser v.3) as described elsewhere without modification (25, 26). Data was derived for preannotated CD4⁺ Treg that passed default qualitative filters. Thereafter, a fucosylation signature (Fuco-signature) consisting of following genes: FUT1, FUT2, FUT3, FUT4, FUT5, FUT6, FUT7, FUT8, FUT9, FUT10, FUT11, POFUT1, POFUT2, GMDS, TSTA3, FUK, FPGT and SLC35C1, was used to create a sum-based metagene expression spectrum per Treg cell (974 Treg single-cells expressed this signature). The Fuco-metagene expression ranged from 0 to 16.3 units. Herein, we took a middle cutoff of 7.02 expression units to bifurcate total Tregs into Tregs with a high Fuco-signature (metagene expression > 7.02) or Tregs with a low Fuco-signature (metagene expression < 7.02). Thereafter, we carried out differential gene expression analyses between Fuco-signature, high vs. low, single-cell Treg using the built-in differential expression (DE) Dashboard function. This analysis was performed using the DESeq2 statistical method for deriving threshold-bound volcano plots. These gene hits were further used as input for the built-in Enrichment panel that was used for enrichment of REACTOME pathways using gene set enrichment analyses (GSEA) methodology (27).

Survival analysis was carried out using the top 30 most enriched genes within the Fuco-signature low Tregs in the The Cancer Genome Atlas (TCGA) melanoma dataset. We accessed overall survival (OS) and tumor gene expression profiles for skin cutaneous melanoma (SKCM; n = 458) patients using the GEPIA2 work-flow (28), based on the UCSC Xena project (http://xena.ucsc.edu). Briefly, OS analysis was based on log-rank hypothesis test (the Mantel–Cox statistical test) that also estimated the Cox-proportional HR and the 95% confidence intervals, accompanied by a Kaplan–Meier plot. Herein, the expression threshold cutoff at a median signature expression level was used for splitting the patients into high-expression and low-expression sub-cohorts. Finally, the Fuco-signature was correlated (Pearson's correlation) to signatures of various immune cells, fibroblasts, endothelial cells, or melanocytic (cancer) cells derived from an existing study (29).

Patient survival analysis in immuno-oncology clinical trials was carried out using the above top 30 most enriched genes within the Fuco-signature low Tregs. We accessed tumor transcriptomic data from melanoma patients (at post-treatment) profiled after anti–CTLA-4 immunotherapy (n = 15; ref. 30) or before anti–PD-1 immunotherapy (n = 41; ref. 31). These data and subsequent OS estimates were performed via TIDE-workflow as described elsewhere (32). Herein, the prognostic effects were calculated as z-score deduced using the Coxph statistical model. These data were represented as Kaplan–Meier curves (at median expression cutoff).

Raw reads were downloaded from Gene Expression Omnibus (GEO) under accession number GSE89225 (33), which included Tregs from peripheral blood mononuclear cells (treg_PBMC, n = 4), breast tumors (treg_tumor, n = 6), and normal breast parenchyma (treg_NBP, n = 6). Trimmomatic (v0.39) removed low-quality reads and trimmed adapters with default setting. STAR (v2.7.7a) was used to align reads to human genome hg38 and RSEM (v1.3.1) to estimate gene expression. R package DESeq2 (v1.26.0, R version 3.6.2) calculated normalized counts. GSVA (v1.34.0) performed single-sample GSEA analysis. Significant tests were performed by T-test. A Golgi Fuco-signature consisted of the following genes: FUT1, FUT2, FUT3, FUT4, FUT5, FUT6, FUT7, FUT8, FUT9, FUT10, FUT11, and SLC35C1.

Normalised mRNA expression of fucosyltransferases (Fut1, Fut2, Fut4, Fut7, Fut8, Fut9) was determined in Tregs sorted from the blood of 6 healthy human donors, 3 male and 3 female, by querying a previously published transcriptomic dataset (34). The available data were normalized transcript expression values, and no additional data normalization was performed, nor were any additional expression thresholds applied. Data were plotted using the box plot function in R.

Data entry and all analyses were performed in a blinded fashion. All statistical analysis was performed using GraphPad Prism Software (Version 9.4.1). Pairwise comparisons on two experimental conditions were performed using unpaired or paired Student t-test analysis. Detection of mathematical outliers was performed with the use of Grubbs test in GraphPad. Grouped data were assessed by two-way ANOVA Sidak multiple comparisons test. Sample sizes for all experiments were used on the basis of previous experience. Statistical details of each experiment are mention on the legends of each figure. All graphs show mean value ± SEM.

The human RNA sequencing (RNA-seq) bulk dataset was obtained from GEO under accession number GSE89225. The single-cell RNA-seq dataset from patients with colorectal cancer was obtained from GSE132465. All the other data presented in this manuscript are available in the supplementary materials and data files or from the corresponding authors upon request.

To systemically identify metabolic genes affecting Foxp3 expression in iTregs, we implemented a pooled lentiviral CRISPR/Cas9 screening using a sgRNA library targeting 2,005 metabolic genes. The library co-expressed the congenic marker CD90.1, which allowed an optimal tracing and enrichment of the transduced cells (Fig. 1A). Naïve CD4⁺ T cells were isolated from Rosa26-Cas9 mice and polarized to Tregs for 5 days, which showed a sustained expression of Foxp3 and CD25 (Fig. 1B). Thereafter, iTregs were transduced with the lentiviral sgRNA library. A MOI reaching approximately 30% of transduction was used to guarantee that most cells carried a sgRNA (Fig. 1C). Tregs were kept in culture for 5 additional days, a time point where a library-dependent perturbation of Foxp3 expression was evident compared with control iTregs (i.e., transduced with the nontargeting sgRNA, sgNT), thereby benchmarking the screening settings (Fig. 1D).

Transduced cells (CD90.1⁺) were sorted on the basis of Foxp3 expression and IFNγ production into four populations: Foxp3^highIFNγ⁺, Foxp3^highIFNγ⁻, Foxp3^lowIFNγ⁺, Foxp3^lowIFNγ⁻ (Supplementary Fig. S1A). On the basis of the library coverage obtained in each population, we focused on two populations: Foxp3^low (bioinformatic pool of Foxp3^lowIFNγ⁺, Foxp3^lowIFNγ⁻) versus Foxp3^high (bioinformatic pool of Foxp3^highIFNγ⁺ and Foxp3^highIFNγ⁻) iTregs (Supplementary Fig. S1B). Analysis of sgRNA abundance and distribution in each sorted population was performed with the MAGeCK bioinformatic pipeline to evaluate the quality of the screening (Supplementary Fig. S1C–S1F). After selecting a threshold of P value < 0.05 and z-score LFC > 0.5 or LFC < −0.5, we identified 50 genes as negative regulators and 12 genes as positive regulators of Foxp3 expression (Fig. 1E and F).

To validate our screening hits, we independently assessed the effects of knocking out the top regulators of Foxp3 expression via electroporation of the CRISPR/Cas9 RNP complex (Fig. 2A). First, we confirmed the gene-editing efficiency of the electroporation platform in iTregs using a sgRNA targeting Foxp3 (Fig. 2B). Silencing of all the top positive Foxp3 regulators from our screening (i.e., Slc35c1, Dck, Cyp3a13, A4galt, Duox1, Echs1, and Pla2g4c; Supplementary Fig. S2A–S2G), and measurement of Foxp3 expression at the protein (Fig. 2C; Supplementary Fig. S2H) and/or transcript level (Fig. 2D), revealed that Foxp3 protein expression was significantly reduced upon silencing of Slc35c1, Dck, and A4galt only. Slc35c1 targeting was tested using two different crRNAs, and for the following experiments, guide #1 was used (Supplementary Fig. S2I). Among the negative regulators, we validated Pold3, Adss, and Gys2, confirming that genetic inactivation of any of these three targets led to increased Foxp3 expression at the protein level (Fig. 2C; Supplementary Fig. S2J). The top enriched positive regulator, based on the predefined statistical threshold, and the only regulator that significantly reduced Foxp3 expression at both RNA and protein levels was Slc35c1, a transporter that mediates GDP-fucose entry into the Golgi-lumen, providing substrates for fucosyltransferases (35). Analysis of the expression of fucosyltransferases (from Human Protein Atlas; ref. 34), represented in the metabolic library (Fut1, Fut2, Fut4, Fut7, Fut8, Fut9), showed that only Fut2, Fut4, Fut7, and Fut8 were expressed in human Tregs (Fig. 2E). All gRNAs against these fucosyltransferases (Fut2, Fut4, Fut7, and Fut8) were also identified as positive regulators of Foxp3 expression (Fig. 2F), disclosing a previously unknown role of Fuco in Foxp3 maintenance. Comparison of the total levels of Fuco (MFI AAL) in a murine CT26 model showed significantly higher Fuco in tumor-infiltrating Tregs versus those isolated from the nonperfused spleen of the same mouse (Fig. 2G), likely rendering Slc35c1 as a safe target in the context of cancer biology.

We focused on an upstream process of fucosyltransferases, the transport of GDP-fucose in the Golgi compartment, a process mainly controlled by Slc35c1. Despite some knowledge regarding the implications of Slc35c1 mutations in humans and mice (35–37), its role in Treg biology is largely unknown. Gene editing by electroporation or lentiviral vector transduction in Tregs resulted in 50 to 70% reduction of the Slc35c1 transcript (Fig. 2H and I). By using two different guides targeting Slc35c1 in electroporated Tregs and one independent guide targeting Slc35c1 in lentiviral vector-transduced Tregs, we confirmed that Slc35c1 deficiency reduced Foxp3 expression and increased IFNγ (Fig. 3A; Supplementary Fig. S3A and S3B). In addition, sgSlc35c1-Tregs had a significant decrease in CD25 mean fluorescent intensity (MFI), without affecting the expression of ICOS, TGFβ1, and GITR (Fig. 3B). Moreover, surface or total staining of PD-1 revealed that Slc35c1 deficiency significantly reduced surface expression of PD-1 (Fig. 3C), without affecting its mRNA expression (Supplementary Fig. S3C), suggesting a posttranscriptional, Fuco-dependent effect on PD-1 expression, as shown previously (19). Also, Slc35c1 genetic targeting in Tregs reduced their suppressive capacity towards effector CD4⁺ T cells (Fig. 3D).

Mechanistically, Slc35c1 inactivation in Tregs led to reduced Fuco, as indicated by AAL staining (Fig. 3E; Supplementary Fig. S3D). Also, Slc35c1 deletion led to reduced proliferation of iTregs in vitro (Fig. 3F; Supplementary Fig. S3E) without affecting their viability (Fig. 3G and H). Because Fuco of different surface proteins is important for cell trafficking (38), we assessed if there was any difference in the recruitment of sgSlc35c1-iTregs. A 50:50 mix of sgNT- and sgSlc35c1-iTregs, labeled with CFSE or CTV (Supplementary Fig. S3F), was injected in MC38 tumor-bearing Rag2^−/− mice, and the mice were sacrificed 20 hours after iTreg transfer. There were no differences in the number of sgNT- and sgSlc35c1-iTregs in the spleens and tumors (blood was used as experimental control tissue; Fig. 3I), indicating that the deletion of Slc35c1 does not affect the trafficking of iTregs towards these tissue sites. However, when the experiment was performed at a later timepoint in healthy mice (sacrificed 7 days after CD45.1⁺ or CD45.1/2⁺ Foxp3^Thy1.1 iTreg transfer; Supplementary Fig. S3G), fewer sgSlc35c1 cells were detected in the spleen of recipient Rag2^−/− mice (Fig. 3J; Supplementary Fig. S3H and S3I). This suggested that Slc35c1 deletion conferred a proliferation disadvantage, as corroborated by the in vitro proliferation data (Fig. 3F; Supplementary Fig. S3E) and a decrease in the percentage of intratumoral EdU⁺ sgSlc35c1-iTregs at an early stage, i.e., day 7 after ACT (Fig. 3K). This reduction in EdU associated with a reduced total number of Slc35c1-deficient iTregs per tumor (Fig. 3L).

To test whether Tregs suppressive function could also be negated in the tumor microenvironment (TME) by ablating Slc35c1, we performed an in vivo suppression assay as previously described (39). Sorted CD45.2⁺CD90.1⁺Cas9⁺ sgNT- or sgSlc35c1-Tregs were co-transferred with CD45.1⁺Cas9⁺CD4⁺CD25⁻ and CD8⁺ T cells in Rag2^−/− mice, followed by MC38 cell injection. In line with the reduced suppressive function of sgSlc35c1-iTregs in vitro, growth of MC38 tumors was significantly inhibited in mice that received sgSlc35c1-Tregs compared with the mice that received sgNT-Tregs (Fig. 4A and B). Also, CD4⁺ and CD8⁺ T-cell accumulation was higher in mice injected with sgSlc35c1-Tregs, and CD4⁺ T cells were able to produce more IFNγ compared with the sgNT-Treg group (Fig. 4C–F). Because the effect of sgSlc35c1-Tregs on IFNγ production was stronger for CD4⁺ T cells compared with CD8⁺ T cells, we checked the representation of tumor antigen-specific TCRs in CD8⁺ T cells with using a KSPWFTTL-H-2K^b tetramer. Although there was no difference in cell number (Supplementary Fig. S4A), tumor antigen-specific CD8⁺ T cells produced higher IFNγ in the sgSlc35c1 group (Fig. 4G and H). Thus, deletion of Slc35c1 reduced the suppressive function of in vivo–derived Tregs and restored antitumor immunity. Functionally, sgSlc35c1-Tregs exhibited a lower Foxp3 and slightly higher IFNγ expression compared with sgNT-Tregs (Fig. 4I–M), confirming the impaired Treg phenotype observed in vitro. A 50:50 mix of sgNT- and sgSlc35c1-Tregs was injected in Rag2^−/− mice in the same setting as previously described, and tumor growth in these mice was intermediate (between the two conditions) (Fig. 4N), indicating that sgSlc35c1-Tregs are not able to drive fragility in surrounding Tregs, likely due to the mild increase of IFNγ production upon Slc35c1 deletion. To characterize the effect of sgSlc35c1-Tregs on the remodeling of the TME, possibly explaining the reduction in tumor growth, we also analyzed tumor-associated macrophages (TAM). Here, we found a reduced number of TAMs in the sgSlc35c1 versus sgNT condition (Fig. 4O), which indicates that the impaired suppressive function of sgSlc35c1-Tregs leads to tumor growth inhibition both directly and indirectly via macrophage alterations.

To translate our data in a pharmacologic setting, we used the Fuco inhibitor 2-peracetyl fucose (2FF). In an ex vivo preconditioning regimen, Tregs were isolated from Ly5.1⁺ Foxp3^Thy1.1 mice and cultured in the presence of 100 μΜ 2FF or DMSO for 6 days. AAL FACS staining confirmed the reduction in Fuco in 2FF-treated Tregs (Fig. 5A), and we also observed a slight, but significant, reduction in Foxp3 MFI, frequency of PD-1⁺ cells, and PD-1 MFI, a phenotype similar to what we observed in sgSlc35c1-Tregs (Fig. 5B–D). Subsequently, one day before MC38 implantation, CD4⁺CD25⁻ and CD8⁺ T effector cells were adoptively transferred alone or in combination with 2FF- or DMSO-treated Tregs into Rag2^−/− mice (Fig. 5E). Consistent with our hypothesis, mice that received 2FF-treated Tregs exhibited reduced tumor growth (Fig. 5F–G) compared with the DMSO-Treg group, indicating that 2FF ex vivo preconditioning impairs Tregs in vivo suppressive function. This was accompanied by a significant increase of CD4⁺ T-cell infiltration and a slight increase in CD4⁺ T-cell IFNγ production (Fig. 5H–I). Analysis of tumor-derived Tregs at end stage revealed no difference in their numbers (Supplementary Fig. S5A) or phenotype, as determined by frequency of Foxp3⁺ cells and Foxp3 MFI (Supplementary Fig. S5B–S5C). By assessing the AAL MFI in splenic Tregs from tumor-bearing mice at end stage, we observed significantly higher Fuco in the 2FF-pretreated group (Fig 5J). From these data, we can speculate that the effects of 2FF in reducing Fuco, and on reducing suppressive function of Tregs, is maintained in the early stages following ACT, but it is lost, or possibly compensated, at end stage (Fig. 5G and J).

Augmented Fuco of several proteins involved in tumor progression has been reported in several cancer types (40, 41). However, the functional relevance of Fuco inhibition in Tregs is unknown. To assess the effect of Fuco blockade in Tregs in vivo, CT26 tumor-bearing mice were treated with 2FF or saline starting at an average tumor volume of 150 mm³. 2FF systemic administration led to a significant reduction in tumor growth and weight (Fig. 6A and B). The total number of Tregs was significantly reduced at early stage, i.e., after 7 days of 2FF treatment (Fig. 6C) but disappeared at end-stage (Fig. 6D). In 2FF-treated mice, PD-1 expression was significantly reduced on Tregs (Fig. 6E and F; Supplementary Fig. S6A), without altering CD4⁺Foxp3⁻ and CD8⁺ T cells in the tumor (Fig. 6G–J). In spleens from treated mice, although the percentage of Tregs did not change, we observed a higher percentage of CD4⁺Foxp3⁻ and CD8⁺ T cells (Supplementary Fig. S6B–S6D) and a lower percentage of PD-1 on Tregs and CD4⁺Foxp3⁻ T cells (Supplementary Fig. S6E–S6J). By measuring Fuco levels in different intratumoral cell types, we found that 2FF led to significantly reduced Fuco in Tregs but not in other T cells (Fig. 6K; Supplementary Fig. S6K). Although 2FF treatment did not affect the percentage of all cell populations in the tumor (Supplementary Fig. S6L–S6O), it led to a significant reduction in the percentage of F4/80⁺ macrophages, which also expressed higher MHCII (Supplementary Fig. S6P–S6Q), and to a significant increase in the percentage of CD8⁺ T cells (Supplementary Fig. S6S). When bone marrow–derived macrophages were treated with 2FF in vitro there were no differences either in the percentage of MHCII^HIGH cells, other antigen-presentation-related proteins, or viability (Supplementary Fig. S6R). To better understand the main target of 2FF, we performed systemic 2FF treatment upon efficient Tregs’ depletion with anti-CD25 (Supplementary Fig. S6T–S6U; ref. 24). 2FF alone or Tregs’ depletion alone led to a significantly reduced tumor growth, but their combination did not further reduce tumor growth (Fig. 6L–M). Also, 2FF administration upon Tregs’ depletion, did not lead to differences in TAMs or MHCII expression (Supplementary Fig. S6V–S6W). These data, together with a decrease in TAMs upon transfer of sgSlc35c1-Tregs (Fig. 4O), indicate that the effect on macrophages in the TME is mediated through Tregs.

To understand if impaired Fuco of PD-1 could be responsible for the reduced suppressive function of Tregs, Tregs were isolated from Pdcd1^−/− mice and transduced with retroviral vectors encoding a PD-1 WT or mutant form, which has reduced PD-1 Fuco due to the N49Q/N74Q double mutation as has been previously shown (19). The efficiency of the reconstitution is shown in Supplementary Fig. S6X. Three days upon transduction, WT or mutant PD-1 Tregs were treated ex vivo with 2FF (PD-1 WT + 2FF and PD-1 MUT + 2FF Tregs) or DMSO vehicle (PD-1 WT + DMSO and PD-1 MUT + DMSO Tregs) for 3 additional days. The four groups of Tregs were transferred via tail vein injection into Rag2^−/− mice to perform in vivo suppressive assays as described previously. Mice injected with vehicle-treated mutant PD-1 Tregs (PD-1 MUT + DMSO) exhibited smaller MC38 tumors compared with mice injected with vehicle-treated WT PD-1 Tregs (PD-1 WT + DMSO; Fig. 6N and O), indicating the PD-1 Fuco is important for the maintenance of Treg suppressive function in vivo.

We sought to characterize Tregs with different levels of Fuco within the tumors, and determined if these populations could be relevant prognostic factors for patient survival. An already published scRNA-seq data of patients with colorectal cancer (n = 23) was used to investigate the Fuco profile of tumor-infiltrating Tregs (Fig. 7A; ref. 27). Tregs were bifurcated into Fuco^LOW and Fuco^HIGH Tregs based on the expression of genes related to Fuco (Fig. 7B; ref. 40). Of 974 total Tregs identified to proficiently express the Fuco signature, 352 Tregs were Fuco^LOW and 622 Tregs were Fuco^HIGH in patients with colorectal cancer (Fig. 7C). Thereafter, we performed a REACTOME biological pathway enrichment analysis on Fuco^LOW vs. Fuco^HIGH Tregs and found that Fuco^LOW Tregs were enriched for more immunogenic or pro-inflammatory pathways related to TCR activation, IFNγ signaling, and MHCII antigen presentation (Fig. 7D and E).

We next questioned if the Fuco^LOW Treg footprint in tumors could be used as biomarker for predicting patient survival. Analysis of melanoma patient data retrieved from TCGA showed that patients with high expression of the Fuco^LOW signature (extracted from above scRNA-seq analyses) survived significantly longer than patients with low expression of the Fuco^LOW signature (Fig. 7F). Also, analysis of the melanoma tumor transcriptome prior to treatment of patients with anti–PD-1 therapy (31) or post-treatment with anti–CTLA-4 (30) showed that patients with high expression of the Fuco^LOW signature had improved survival compared with patients with low expression of the signature, thereby indicating that the Fuco^LOW footprint could be a useful biomarker for immuno-oncological applications (Fig. 7G and H). Because the data from TCGA includes transcripts from whole tumors, analysis of the correlation between our Fuco^LOW Treg signature with bulk-transcriptome signatures for tumor-associated Tregs, macrophages, fibroblasts, endothelial cells, and melanocytic cancer cells was performed. It was found that a Fuco^LOW Treg signature positively correlated with a bulk-transcriptome Treg signature within the same melanoma tumor, shown to associate with better prognosis (Supplementary Fig. S7A–S7E). This suggests an important contribution of Tregs to patient survival. Also, the expression of Golgi-related Fuco genes (Fut1–11 and Slc35c1) was analyzed in a published RNA-seq dataset from patients with breast cancer (33). The expression of these genes was enriched in breast tumor-derived Tregs and also in Tregs derived from the normal breast parenchyma compared with Tregs derived from PBMCs (Supplementary Fig. S7F). This finding suggests that Fuco is important for Tregs in tumor and adjacent tissues but not Tregs in circulation.

Despite numerous studies on Tregs’ metabolism (8–12) and CRISPR/Cas9 screenings tailored to identify novel regulators of Treg identity (23, 39, 42, 43), an unbiased and holistic discovery of metabolic factors regulating Treg functionality is currently missing.

Here, we performed a loss-of-function, pooled, CRISPR-based metabolic screen and identified functionally relevant positive (Slc35c1, Dck, Cyp3a13, A4galt, Duox1, Echs1, and Pla2g4c) and negative (Adss, Gys2, and Pold3) regulators of Foxp3. Among the positive regulators, we uncovered genes related to Fuco, specifically fucosyltransferases Fut2, Fut4, Fut7, and Fut8 and the transporter Slc35c1. We further uncovered the role of Slc35c1, which has no previous reported Treg functionality. Slc35c1 is a GDP-fucose transporter from the cytosol into the Golgi, important for the synthesis of fucosylated glycans. Defects of this gene have been associated in humans with leucocyte adhesion deficiency II (LAD II; refs. 35, 37). Fucosylation of glycans is very important for many biological processes such as leucocyte cell migration, immune cell interactions, and antibody-dependent cellular cytotoxicity (40, 44). Besides our focus on positive regulators, future work geared towards understanding the functional relevance of the metabolic negative regulators of Foxp3 identified in this study are underway.

Our data indicate that impaired Fuco after Slc35c1 deletion in Tregs weakens Foxp3 expression, slightly enhances IFNγ production, and limits Treg suppressive capacity in vitro, reminiscent of a “unstable,” less suppressive Treg phenotype (6). Mechanistically, Slc35c1 deletion impaired the proliferation of Tregs in vitro and in vivo without affecting their recruitment capacity into tissues. In line with this, the in vivo transfer of sgSlc35c1-Tregs in tumor-bearing mice led to reduced tumor growth and increased numbers and IFNγ production, mainly driven by the CD4⁺ T cells, confirming that sgSlc35c1-Tregs had impaired suppressive activity once in the TME. In addition, tumors in mice injected with sgSlc35c1-Tregs had significantly fewer TAMs, which are known to suppress antitumor immunity and promote tumor progression. However, the limitations related to the transfer of Tregs in Rag2^−/− mice should be considered. Although commonly used, Rag2^−/− mice do not recapitulate a physiologic tumor environment. The creation of Slc35c1-knockout mice and specific deletion of Slc35c1 in Tregs in the future will allow for a deeper characterization of the role of Slc35c1 in Treg biology.

Pharmacologic inhibition of Fuco has been reported to inhibit tumor growth in vivo (45, 46) and to potentiate the effector functions of CD8⁺ T cells adoptively transferred into tumor-bearing mice (19). Of note, enhancing Fuco in human Tregs prior to their transfer leads to prolonged survival of Tregs, a finding that led to a clinical trial in GVHD patients (NCT02423915; ref. 47). In the current study, we showed that adoptive transfer of Tregs preconditioned with a Fuco inhibitor or systemic inhibition of Fuco in tumor-bearing mice led to reduced tumor growth. Systemic administration of 2FF led to impaired Fuco in Tregs but not in other T-cell populations, consistent with the observation that Fuco's antitumor effect was predominantly through an alteration of suppressive Treg function. We also showed a reduced fraction of TAMs, albeit with a higher expression of MHCII (a marker of improved antigen-presentation), which was similar to when sgSlc35c1-Tregs were transferred. This difference in TAMs was not observed with Treg depletion, nor with in vitro treatment of macrophages with 2FF. Overall, our data suggest that Tregs are the main T-cell population in the tumor affected by inhibition of Fuco, and this could impair the density and suppressive phenotype of TAMs in the TME. Further studies are required to describe the exact mechanism by which Fuco in Tregs affects TAM number and function.

Tregs exhibit their suppressive function through a variety of mechanisms, including immune checkpoint molecules (e.g., GITR, CTLA-4, ICOS, TGFβ1, PD-1, etc.). These are generally families of proteins highly decorated with oligosaccharide moieties, and altering this process might affect their folding, stability, intracellular trafficking, cell-surface marker expression, secretion, cell-cell binding, receptor-ligand interactions, and cell trafficking (48). Although it is known that membrane exposure and turnover of PD-1 is regulated by Fuco (19), we showed that inhibition of Fuco significantly reduced surface PD-1 expression on Tregs, which was eventually more prominent in tumor Tregs than in CD8⁺ or CD4⁺ effector T cells. Also, the reconstitution of Pdcd1^−/− Tregs with a PD-1 mutant that had impaired Fuco, led to less suppressive Tregs in vivo, indicating that PD-1 Fuco was partly responsible for the reduced suppressive phenotype of sgSlc35c1-Tregs. We speculate that in Tregs, Slc35c1 deletion through its effect on Fuco, leads to reduced surface expression of PD-1. PD-1 has been shown to be important for Treg stability and Foxp3 expression, thus regulating the suppressive function of Tregs (49–51). However, the role of PD-1 blockade in Tregs has been controversial (52). In addition, our data suggest that Slc35c1 deficiency may affect Treg function through multiple mechanisms, altogether harnessing their suppressive functions and altering the TME towards an antitumor immune response. Despite our focus on PD-1 Fuco, we believe that Slc35c1 controls the Fuco of several other membrane proteins that are important for the suppressive function of Tregs. For example, Fuco of TGFβ receptors is critical for their function (53, 54). Given the broad biological effects of core Fuco, all the pathways affected by Slc35c1 deficiency in Tregs remain to be elucidated in future studies.

We additionally tested the human relevance of our findings and revealed that tumoral Tregs could be distinguished as Fuco^LOW and Fuco^HIGH subpopulations, with Fuco^HIGH Tregs being the most abundant in the tumor compared with the Fuco^LOW. In accordance with our findings in mouse Tregs, Fuco^LOW Tregs exhibited a pro-inflammatory and IFNγ-producing phenotype, and their high number associated with a relatively better disease outcome and positive response to immune checkpoint blockade in melanoma patients. Analysis of the Fuco signature in RNA-seq data from Tregs derived from patients with breast cancer showed that Tregs in tumor and adjacent tissues had higher levels of Fuco compared with circulating Tregs. This was also corroborated by our observation in mice, whereby tumor-infiltrating Tregs exhibited the highest extent of Fuco. These data point to the notion that targeting Slc35c1 might lower the risk of nonspecific targeting and development of autoimmunity compared with pan-Treg depletion. Because Fuco-related genes are involved in a multitude of pathways that are part of glycosylation structure synthesis, further study of the role of each of these genes in Treg biology could lead to the better understanding of their function.

E. De Bousser reports grants from Foundation for Scientific Research Flanders; and grants and personal fees from VIB during the conduct of the study. N. Callewaert reports grants from Foundation for Scientific Research Flanders; and grants from VIB during the conduct of the study. No disclosures were reported by the other authors.

S. Pinioti: Conceptualization, formal analysis, supervision, funding acquisition, visualization, methodology, writing–review and editing. H. Sharma: Validation, investigation, methodology. N.C. Flerin: Conceptualization, supervision. Q. Yu: Software, formal analysis. A. Tzoumpa: Validation, methodology. S. Trusso Cafarello: Methodology. E. De Bousser: Resources, methodology. N. Callewaert: Resources, methodology. G. Oldenhove: Resources. S. Schlenner: Methodology. B. Thienpont: Software, supervision. A.D. Garg: Software, investigation, visualization. M. Di Matteo: Conceptualization, formal analysis, supervision, funding acquisition, visualization, methodology, writing–review and editing. M. Mazzone: Supervision, funding acquisition, visualization, writing–original draft, project administration, writing–review and editing.

S. Pinioti was supported by a MSCA-ITN grant (META-CAN; #766214). Q. Yu was funded by the Stichting tegen Kanker (F/2020/1544). E. De Bousser and N. Callewaert were funded by the Foundation for Scientific Research Flanders and from VIB. Research in A.D. Garg's lab was supported by Research Foundation Flanders (FWO; Fundamental Research Grant, G0B4620N; Excellence of Science/EOS grant, 30837538, for ‘DECODE’ consortium), KU Leuven (C1 grant, C14/19/098, C3 grant, C3/21/037, and POR award funds, POR/16/040), Kom op Tegen Kanker (Stand Up To Cancer, the Flemish Cancer Society; KOTK/2018/11509/1; KOTK/2019/11955/1), and VLIR-UoS (iBOF grant, iBOF/21/048, for ‘MIMICRY’ consortium). M. Mazzone holds an ERC Consolidator Grant for the project ImmunoFit (773208). We would like to thank Jens Serneels, Dana Liu, Elisabeth Vandeputte, and Alicia Jacquemotte for their technical support, Dr. Jochen Lamote for his scientific input in flow cytometry data analysis, Prof. Akihiko Yoshimura for providing the PD-1 WT and mutant plasmids, and Vincent Van Hoef for contributing on the analysis of the NGS data.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.