Human peripheral blood mononuclear cells (PBMC) were isolated from blood donor-derived buffy coats by density gradient centrifugation with Ficoll-Histopaque 1.077 g/mL density (Sigma-Aldrich, Steinheim, Germany) as separation medium. CD8⁺ T lymphocytes were subsequently purified from PBMC by immunomagnetic negative selection using an EasySep™ Human CD8⁺ T cell kit according to the manufacturer’s recommendations (STEMCELL Technologies, Cologne, Germany). Initial assessment of CD8⁺ T lymphocyte purity from several independent experiments confirmed constantly high purity (>95%) of CD3⁺CD8⁺ T lymphocytes after the purification procedure from human PBMC using flow cytometry (data not shown). The studies on human T lymphocytes were performed with buffy-coats from anonymous healthy blood donors of the blood donation center DRK-Blutspendedienst Baden-Württemberg-Hessen, Institut für Transfusionsmedizin und Immunhaümatologie Frankfurt am Main, Frankfurt, Germany. Prior to blood sampling, all participants were informed in full about all aspects of the study and asked to give written informed consent for use of the samples. According to the institutional ethics committee of the Goethe University Hospital, Frankfurt, Germany, and the local legislation, additional ethical approval was not required, since the cells derived from buffy coats were used anonymously for in vitro experiments with no link to personal data of the donors.

After the enrichment from human PBMC, CD8⁺ T lymphocytes were cultivated under serum withdrawal (starvation) or, if otherwise indicated, with 10% autologous and heat-inactivated donor serum in RPMI 1640 + Glutamax supplemented with 50 mM β-mercaptoethanol, 1 mM sodium pyruvate, 100 μg/mL streptomycin, 100 IU/mL penicillin (all from Gibco, Waltham, USA) and 2 nM HEPES (Sigma-Aldrich, Steinheim, Germany). Lymphocytes were seeded at a density of 5 x 10⁵ cells per mL into 12-well plates (Greiner bio-one, Frickenhausen, Germany), if not otherwise stated. CD8⁺ T lymphocytes were treated with the interleukins IL-33 at 20 ng/mL and IL-12p70 (hereafter IL-12) at a concentration of 5 ng/mL after 20 h cultivation. These cytokines were purchased from Peprotech, Hamburg, Germany and dissolved for use in PBS/0.1% BSA. Additional pharmacologic modulation of S1P receptors was achieved by daily treatment during the cultivation of CD8⁺ T lymphocytes. A time schedule for the cultivation of CD8⁺ T lymphocytes is provided within the supplement ( Supplementary Figure 1 ). FTY720-phosphate (FTY720-P; Novartis, Basel, Switzerland) was dissolved in dimethyl sulfoxide (DMSO). For use as S1P receptor modulators in cell culture, S1P and FTY720-P were pre-diluted in fatty acid-free PBS/0.1% BSA solution and diluted to a final concentration of 200 nM. The cells were consistently stimulated with the selective S1P₄ receptor agonist [CYM50308 (33)] and S1P₄ receptor antagonist [CYM50358 (34)] at 200 nM (Tocris, Bristol, UK).

RNA from CD8⁺ T lymphocytes was extracted using the Isolate II RNA Micro Kit (Bioline, Heidelberg, Germany) according to the manufacturer’s instructions. RNA was then transcribed into cDNA by reverse transcriptase with the Precision nanoScript Reverse Transcription Kit (Primerdesign, Southampton, UK) using the RT-PCR program (65°C for 5 min, 55°C for 120 min and 75°C for 15 min). To quantify KLF2 (Hs00360439_g), S1P₁ (Hs00173499_m1) and S1P₄ (HS02330084_s1) mRNA levels in CD8⁺ T cells, relative gene expression was calculated by normalization to the housekeeping genes GAPDH and RPL13A (Primer Design, Southampton, UK) with the 2^−ΔCt method (all probes were obtained from Applied Biosystems, Waltham, USA). Gene expression of S1P1 on CD8^Low and CD8^High T lymphocyte subpopulations was analyzed after sorting the cells. For the isolation of RNA from CD8^Low and CD8^High the RNeasy micro kit (Qiagen, Hilden, Germany) was used according to the manufacturer’s instructions. Subsequent cDNA synthesis was performed using SuperScript™ VILO™ Master Mix (ThermoFisher, MA, USA) according to the manufacturer’s instructions.

For the characterization of surface molecules on CD8⁺ T lymphocytes, cells harvested from culture plates were stained as a single cell suspension with the following antibodies: anti-CD8-V450 (clone RPA-T8, BD Biosciences, Heidelberg, Germany), anti-CD39-BV510 (clone A1), and anti-PD-1/CD279-APC (clone EH12.2H7, both from Biolegend, San Diego, USA), anti-CD38-APC-eFlour780 (clone HIT2), and anti-CXCR4/CD184-APC (clone 12G5, both from eBiosciences, Waltham, USA), anti-CXCR3-Flourescein (clone 49801), and anti-CXCR7-PE (clone 11G8) (both from R&D Systems, Minneapolis, USA). All antibodies were titrated to determine optimal concentrations. In order to avoid non-specific antibody binding, cells were incubated for 10 min at 4°C with 0.1% PBS/FCS containing human F_c block (BD Pharmingen, Heidelberg, Germany). Cells were fixed in 2% PFA FACS buffer (PBS with 1% FBS and 0.1% NaN₃). Samples were acquired with a Canto II flow cytometer (BD Biosciences, Heidelberg, Germany) and analyzed using FlowJo software V7.6.5 (Treestar, Ashland, USA). For gating, fluorescence minus one (FMO) controls and unstained controls were used. The instrument calibration was regularly controlled using Cytometer Setup and Tracking beads (BD Biosciences, Heidelberg, Germany). The general strategy for gating of CD8^Low and CD8^High subpopulations during analysis of flow cytometry data is provided in the supplements ( Supplementary Figure 2 ).

The capacity of starved regulatory CD8⁺ T lymphocytes to suppress the proliferation of responder T cells was assessed using a transwell in vitro suppression assay. To this end, human PBMC were labelled with violet cell trace (10 µM, ThermoFisher, MA, USA) according to the manufacturer’s instructions. For the proliferation of labeled responder T cells, PBMC (6 x 10⁵/mL/96-well) were cultivated in 1% autologous donor serum and stimulated with 25 µL/mL ImmunoCult™ Human CD3/CD28/CD2 T Cell Activator (αCD, Stemcell Technologies, Vancouver Canada). In parallel, 2 x 10⁵ purified CD8⁺ T lymphocytes from the same donor were seeded into inserts of transwell cultivation plates (0.4 µm, Corning Costar) and co-stimulated with IL-33 and IL-12 (for details see section 2.2) but initially separated from proliferating responder cells. On day 2, co-culture was started by transferring the insert plate into the 96-well plate with labeled and proliferating T cell responders (lower compartment). The assay was performed with n = 3 technical replicates. As a control, 50 ng/mL of rapamycin (LC Laboratories, MA, USA) was added to T cell responders in order to assess the maximal inhibition of proliferation at the beginning of co-culture on day 2. The experimental set-up and time schedule is depicted together with the results in Figure 3A . After 4 days of co-culture, proliferation of labeled T cell responders was measured by flow cytometry. Samples were acquired using MACSQuant 10 cell analyzer (Miltenyi Biotech, Bergisch Gladbach, Germany). Modelling of proliferation was done using the analysis tool from FlowJo software V10.7.1 (Treestar, Ashland, USA).

For the transwell migration assay, 2.5 x 10⁵ purified CD8⁺ T lymphocytes were seeded in serum-free RPMI 1640 medium into the upper compartment of Boyden chamber (5 µm, Corning Costar), cultivated for 40 h and stimulated with S1P receptor modulators as described above. Inserts were pre-coated for 1 h with 10 µg of fibronectin (Sigma-Aldrich, Steinheim, Germany). Directly before migration of the lymphocytes, the medium within the lower compartment was replaced with fresh medium containing 20 nM of diluted recombinant chemokine CXCL12/SDF1α (Peprotech, Hamburg, Germany) as chemoattractant. CD8⁺ T lymphocytes were then allowed to transmigrate into the lower compartment for 2 h. The absolute number of transmigrated cells was determined using MACSQuant 10 cell analyzer (Miltenyi Biotech, Bergisch Gladbach, Germany). The assay was performed with n = 3 technical replicates for each stimulation of individual donor samples. In order to exclude dead cells from the cell count, only DAPI-negative events were considered transmigrated lymphocytes.

Statistical analysis and graphical presentation of data were performed using GraphPad Prism 8 software (La Jolla, CA, USA). If not otherwise stated, all data are presented as means ± SD. Statistical differences between groups were calculated with the Wilcoxon matched pairs signed rank test. Tests for multiple comparisons included the RM one-way ANOVA and Friedman tests. Parametric or non-parametric tests were applied upon testing of normal distribution of data. P-values below 0.05 indicated statistical significance (n.s. for p > 0.05, * for p ≤ 0.05, ** for p < 0.01, *** for p < 0.001 and **** for p < 0.0001).

Nutrient restriction within the tumor microenvironment dampens the activity of mTOR, an essential factor for the activity of potentially infiltrating CD8⁺ T lymphocytes. Serum starvation of human CD8⁺ T lymphocytes allowed the differentiation of two subpopulations, the CD8^High- and CD8^Low-expression T lymphocytes ( Figure 1A and Supplementary Figure 1 ). Since the subpopulation of CD8^Low increased during in vitro nutrient deprivation, we decided to further investigate the specific phenotype of these CD8^Low T lymphocytes by determining whether starvation modulates gene expression of the transcription factor krüppel-like factor 2 (KLF2) in CD8⁺ T lymphocytes, eventually resulting in S1P₁ mRNA. We found that KLF2 mRNA expression was significantly induced after 40 h starvation ( Figure 1B ), whereas levels of S1P₁ mRNA significantly decreased even after only 20 h nutrient deprivation ( Figure 1C ). Since other studies showed that KLF2 induction is followed by upregulation of S1P₁ (35), we asked whether this initial opposing observation could be explained when analyzing S1P₁ mRNA in CD8^Low and CD8^High subsets separately. As shown before, KLF2 mRNA and CD8^Low were simultaneously induced upon 40 h of starvation, and indeed, we determined an increased transcription of S1P₁ limited to CD8^Low lymphocytes ( Figure 1E ). While the role of S1P₁ in lymph node egress has previously been extensively discussed, the function of S1P₄ remains largely unknown. Therefore, we examined and compared the transcription of S1P₁ and S1P₄ ( Figure 1D ). Serum starvation had a major impact on mRNA expression levels of S1P₄ . Prolonged starvation of 40 h significantly decreased overall S1P₄ mRNA expression in all CD8⁺ T lymphocytes, an effect comparable to that observed for S1P₁ mRNA expression levels. Thus, nutrient deprivation in vitro, which could be modeled by mTOR inhibition, appeared to strongly affect the expression of S1P receptors 1 and 4. Despite the presence of IL-12 as a factor in driving cytotoxic antitumor response, IL-33 and IL-12 together can co-induce a T_reg-like phenotype in starved CD8⁺ T lymphocytes, as we recently reported (28). To this end, serum starved CD8⁺ T lymphocytes were stimulated with IL-33 and IL-12 after 20 h nutrient deprivation, prior to gene expression analysis of S1P₁ and S1P₄ after 40 h. Interestingly, gene expression levels of S1P₁ were significantly increased after stimulation with IL-33. On the other hand, co-stimulation with IL-33 and IL-12, which has previously been associated with differentiation of a T_reg-like phenotype, reverted the increase in S1P₁ mRNA expression ( Figure 1F ). Since S1P₄ is the least well-explored S1P receptor, no stimuli for the expression of S1P₄ have yet been described. We next aimed to discover whether transcriptional expression of S1P₄ was upregulated when subjected to nutrient deprivation and cytokine stimulation. Therefore, we serum starved CD8⁺ T lymphocytes and applied the combination of IL-33 and IL-12 – known to trigger a T_reg differentiation program in CD8⁺ T lymphocytes during nutrient deprivation (28) – to investigate S1P₄ mRNA expression levels ( Figure 1G ). Stimulation with the pro-inflammatory cytokine IL-12 also caused a slight but significant decrease of the gene expression of S1P_4, similar to that observed for S1P₁ . Co-stimulation of serum starved CD8⁺ T lymphocytes with IL-33 and IL-12 significantly re-established mRNA expression of S1P₄ , which had been downregulated by IL-12 treatment. Hence, we observed, to our knowledge for the first time, that S1P₄ gene expression was re-established in CD8⁺ T lymphocytes by IL-33 during nutrient deprivation.

Since cytokine stimulation of CD8⁺ T lymphocytes impacted transcriptional regulation of the S1P receptors 1 and 4, we aimed to further define the responsiveness of distinct CD8⁺ T lymphocyte subpopulations to IL-33 during nutrient deprivation. To this end, well-defined human CD8⁺ T lymphocyte subpopulations (CD8^Low and CD8^High) were analyzed with regard to ST2L expression ( Figure 2 ). CD8^Low T lymphocytes clearly showed high ST2L expression (CD8^Low ST2L⁺ T lymphocytes) ( Figure 2A ). Of note, overall ST2L expression was lower on CD8^High compared to CD8^Low T lymphocytes. Accordingly, ST2L expression was increased on CD8⁺ T lymphocytes during starvation compared to serum control, while CD8^Low T lymphocytes were not induced ( Figure 2B ). Next, we focused on the potential role of IL-33-responsive CD8^Low and CD8^High T lymphocytes during nutrient deprivation by analyzing cell surface markers such as the adenosine-converting ectoenzymes CD38 and CD39, which have been proposed as targets for cancer immunotherapy. Having observed that nutrient deprivation induced a subpopulation of CD8^Low T lymphocytes, we sought to determine whether these cells express CD38 and CD39. Interestingly, CD8^Low ST2L⁺ T lymphocytes exhibited increased CD38 and CD39 surface protein expression compared to CD8^High and ST2L-negative CD8 T lymphocyte subpopulations ( Figure 2C ). In general, we observed a higher expression of CD39 compared to CD38. Along with the immunosuppressive adenosine-converting enzymes CD38 and CD39, we also analyzed the expression of PD-1 as an additional important regulator of tumor immune escape. In line with our previous findings, nutrient deprivation conferred higher PD-1 expression on CD8⁺ T lymphocytes compared to serum control ( Figure 2D , left). On analyzing PD-1 expression on CD8^Low and CD8^High subpopulations, we found an increase of PD-1 expression by ST2L⁺ cells, especially by CD8^High T lymphocytes ( Figure 2D , right). Taken together, CD8^Low ST2L⁺ T lymphocytes exhibited higher protein expression of CD38 and CD39, whereas PD-1 expression dominated on CD8^High ST2L⁺ T lymphocytes, potentially limiting the cytotoxic effector functions of CD8⁺ T lymphocytes and thereby presumably promoting a tumor-supporting milieu. Having observed that cytokine stimulation with IL-33 and IL-12 resulted in regulatory CD8⁺ T lymphocyte differentiation during in vitro starvation, we decided to perform a functional assessment of cultured cells using a suppression assay. In order to assess the suppressive capacity of regulatory CD8⁺ T lymphocytes, we co-cultured cells with labeled responder T cells from the same donor ( Figure 3A ). Our data indicates that stimulation of starved CD8⁺ T lymphocytes with IL-33 and IL-12 and subsequent co-culture with responder T cells negatively affected the proliferation of responder T cells ( Figures 3B, C ). In addition, we were able to confirm the induction of CD8^Low T lymphocytes after co-culture in contrast to the serum control, where the CD8^High T lymphocyte subpopulation dominated ( Figure 3D ). These data further support the concept that, under nutrient-deprived conditions, regulatory CD8^Low T lymphocytes have the potential to indirectly suppress the effector function of responder T lymphocytes.

CD8^Low T lymphocytes that developed during serum deprivation were characterized by high responsivity to IL-33, showing upregulation of ST2L as well as enhanced expression of CD38 and CD39 as markers of immunosuppression. In contrast, we observed particularly high expression of PD-1 by CD8^High ST2L⁺ T lymphocytes. Another denominator of the pro- or anti-carcinogenic role of T lymphocyte subpopulations is their migratory status, which is regulated by different chemokine receptors. For that reason, our interest was drawn to the homeostatic chemokine receptor CXCR4, whose high expression levels were linked to an immunosuppressive phenotype of lymphocytes. Moreover, we wanted to see whether CD8^Low and CD8^High T lymphocytes express the inflammatory chemokine receptor CXCR3. We found a generally higher frequency of CXCR4⁺ CD8⁺ T lymphocytes within the CD8^Low subpopulation, whereas the percentage of CXCR3⁺ CD8⁺ T lymphocytes was significantly higher in the CD8^High subpopulation ( Figures 4A, B ). High CXCR4 expression was therefore predominantly observed on the cell surface of CD8^Low ST2L⁺ T lymphocytes ( Figure 4C ). These data not only emphasize the differentiating role of nutrient deprivation within a tumor microenvironment, but also support the finding that CD8^Low T lymphocytes tend to have a regulatory immune cell phenotype. In contrast to CD8^Low, CD8^High represent typical cytotoxic T cell effectors that are partly inhibited during nutrient deprivation, as indicated by the correlation of PD-1 expression and ST2L within this subpopulation.

Since S1P₄ mRNA was expressed in CD8⁺ T lymphocytes under stimulation with IL-33, and despite the presence of IL-12 during nutrient deprivation, we considered whether S1P₄ might contribute to the regulatory activity and migratory status of human CD8⁺ T lymphocytes. Besides controlling leukocyte trafficking, chemokine receptors have also been shown to perform non-trafficking functions, for example in lymphocyte differentiation (36), affecting tumor immunity. Having determined that CXCR4 expression was high on CD8^Low ST2L⁺ regulatory T lymphocytes whereas CXCR3 was dominant on CD8^High T lymphocytes, we investigated whether IL-33 and IL-12 stimulation in combination with a pharmacological treatment using a selective S1P₄ receptor agonist (CYM50308) might alter the expression of the homeostatic chemokine receptor CXCR4 and the inflammatory chemokine receptor CXCR3 ( Figure 5 ). Daily treatment with the S1P₄ agonist was integrated into our existing cultivation protocol of 40 h serum starvation and cytokine stimulation after 20 h. We found that the S1P₄ receptor agonist, when combined with IL-33 and IL-12 stimulation, induced CXCR4 expression on nutrient-deprived CD8⁺ T lymphocytes ( Figure 5A ). Likewise, when comparing the mean fluorescence intensities (MFI) of CXCR4 on CD8⁺ T lymphocytes, we saw an accumulative effect of cytokine stimulation (IL-33 and IL-12) and S1P₄ receptor agonist treatment ( Figure 5B ). Of note, S1P₄ receptor agonist-dependent induction was limited to CXCR4 expression, failing to affect the CXCR3 expression predominant on CD8^High T lymphocytes ( Figure 5C ). Moreover, we observed increased proliferation of CD8^Low T lymphocytes after cytokine co-stimulation in combination with S1P₄ receptor agonist treatment ( Figure 5D ). The S1P₄ receptor agonist was seen to have a dose dependent effect on CXCR4 expression ( Supplementary Figure 3 ). Thus, the effects of the S1P₄ receptor agonist seemed to be restricted to the CD8^Low subpopulation. Since S1P₄ signaling acts as a tumor-promoting factor via the proliferation of CD8⁺ T lymphocytes (20), these findings would endorse a potentially tumor-supportive role of CD8^Low ST2L⁺ regulatory T lymphocytes, as observed during in vitro nutrient deprivation.

The S1P₄ receptor agonist treatment effected an upregulation of CXCR4 surface expression, which had previously been found to be linked to the subpopulation of CD8^Low ST2L⁺ T lymphocytes. This led us to ask whether a selective S1P₄ receptor antagonist might show differential effects on the two different chemokine receptors of CD8⁺ T lymphocytes that were examined. Pharmacologic treatment with the selective S1P₄ receptor antagonist induced no changes in CXCR4 and CXCR3 expression of CD8⁺ T lymphocytes, neither when given alone, nor when combined with IL-33 and IL-12 stimulation ( Figure 6A ). Beyond the selective modulation of the S1P₄ receptor, we additionally used fingolimod phosphate (FTY720-P), which acts as a S1P receptor agonist on S1P₁, S1P₃, S1P₄ and S1P₅, but not on S1P₂ (37). Interestingly, application of the treatment protocol of CD8⁺ T lymphocytes described above resulted in a comparable enhancement of CXCR4 expression ( Figures 6B, C ). We assume that the observed effects of FTY720-P are attributable to its agonistic effects on the S1P₄ receptor, since the selective S1P₄ receptor agonist, but not the antagonist, induced expression of CXCR4 on CD8⁺ T lymphocytes, potentially driving an immunosuppressive phenotype of CD8⁺ T lymphocytes.

Nutrient deprivation enhanced the responsiveness of CD8⁺ T lymphocytes to IL-33, which in turn exhibited an induction of S1P₄ mRNA expression when combined with IL-12 stimulation. Consequently, S1P₄ receptor signaling induced the expression of CXCR4, which was linked to a subpopulation of regulatory CD8^Low ST2L⁺ T lymphocytes (in contrast to CD8^High). In order to assess the migratory potential of CD8⁺ T lymphocytes towards CXCL12 (SDF1α), the ligand of CXCR4, we performed a Boyden chamber migration assay. To this end, CD8⁺ T lymphocytes were cultivated under daily stimulation with either S1P₄ receptor agonist (CYM50308) or FTY720-P within inserts (upper compartment) for 40 h and then allowed to transmigrate along a gradient of CXCL12 into the lower compartment. Neither the S1P₄ receptor agonist nor FTY720-P was found to significantly enhance CXCL12-dependent transmigration of CD8⁺ T lymphocytes ( Figure 7A ). In this context, we proceeded to compare CD8 expression of transmigrated as well as non-migrated CD8⁺ T lymphocytes, since CXCR4 expression was dominant on CD8^Low T lymphocytes. When comparing the frequencies of CD8^High- and CD8^Low-expressing T lymphocytes, we clearly found that CD8^High exhibited a higher capacity to transmigrate along the gradient of CXCL12, whereas CD8^Low generally showed a lower capacity for transmigration ( Figure 7B ). It is known that the expression of CXCR7 (ACKR3), an alternative receptor for CXCL12, might be also relevant to CXCL12-dependent migration. CXCR7 has been reported to bind CXCL12 with high affinity and functions as a scavenging or decoy receptor without exerting chemotactic effects (33). We analyzed in detail whether S1P₄ receptor agonist treatment of CD8⁺ T lymphocytes affected expression of CXCR7, thus potentially perturbing migratory effects of CXCL12 on CD8⁺ T lymphocytes despite induction of CXCR4. To our surprise, the expression of CXCR4 and CXCR7 on CD8⁺ T lymphocytes showed a positive correlation independent of the treatment of CD8⁺ T lymphocytes ( Figure 7C ). Moreover, the S1P₄ receptor agonist significantly induced CXCR7 expression, although to lesser extent than CXCR4 expression ( Figure 7D ). Accordingly, CXCR7 expression was higher on CD8^Low compared to CD8^High T lymphocytes ( Figure 7E ). In summary, our data outline a potential role of two distinct CD8⁺ T lymphocyte subsets that occur during nutrient deprivation and have also been described within a tumor microenvironment. These findings further endorse the possibility that CXCR4, which was predominantly found on CD8^Low and is driven by S1P₄, may affect lymphocyte differentiation rather than being essential for lymphocyte trafficking.