4T1 (Cat# CRL-2539, ATCC) and CT26 (Cat# CRL-2638, ATCC) cells were maintained in complete RPMI (RPMI1640 supplemented with 10% Fetal Bovine Serum (FBS), 1% penicillin-streptomycin (10,000 U/mL stock) and 1% of 200mM L-Glutamine). MB49 (Cat# SCC148, Sigma Aldrich) cell lines were cultured in complete DMEM (DMEM supplemented with 10% FBS, 1% penicillin-streptomycin (10,000 U/mL stock) and 1% of 200mM L-Glutamine Cells). Cells were maintained in 75 cm² tissue culture–treated plastic flasks and used within 10 passages of thawing the cells from frozen stocks.

BALB/c and C57BL/6 female mice were purchased from Charles River Laboratories and housed in accordance with Purdue Animal Use and Care Committee guidelines. FRδ knockout mice embryos (Item# 037093-UNC-EMBRYO, MMRRC) were purchased and mice obtained in collaboration with Purdue University’s Transgenic and Genome Editing Facility. FRβ knockout mice were generously provided by Dr. Richard Finnell (Baylor College of Medicine, Houston TX). Mice were first blindly randomized and used at 8–12 weeks of age for tumor implantation studies, and unless otherwise specified, were housed on corn cob bedding, and placed on folate deficient chow (Teklad Envigo) for 2 weeks prior to use in experiments. 4T1 and CT26 cancer cells (1 × 10⁵ cells per mouse) were implanted subcutaneously in BALB/c mice, whereas MB49 cells were similarly implanted in C57BL/6 mice, after which mice were further randomized based on tumor volume. Tumor volume (V) was quantitated using the formula V= LxW²/2, where L is the length of the tumor’s longest dimension and W is the width perpendicular to this dimension. Treatments were initiated when tumors reached 50 mm³ and harvested at 600–800 mm³ for analysis unless otherwise specified.

Human blood was collected from healthy volunteers (IRB approval number: IRB-2021-1246). Ficoll-Paque separation method was used to obtain PBMCs. Briefly, blood was diluted with 1:1 in 2% FBS in PBS. 50 mL SepMate tubes (Cat# 85450, StemCell Technologies) were prefilled with ~15mL Ficoll (Cat# GE17-5442-02, Sigma Aldrich). Diluted blood was then poured slowly into the tubes, after which the tubes were centrifuged at 12,000 G for 10 minutes. PBMCs were collected and washed twice with cold 2% FBS in PBS. Tregs were isolated from PBMCs using EasySep Human CD4+CD127lowCD25+ Regulatory T Cell Isolation Kit (Cat# 18063, StemCell Technologies) according to the manufacturer’s protocol. Tregs were resuspended in 5% human serum in PBS for downstream applications.

For murine Treg isolation, organs and tissues were removed, mechanically disrupted, and digested using Collagenase I (Cat# C0130, Sigma Aldrich) dissolved in complete RPMI (RPMI1640 supplemented with 10% Fetal Bovine Serum (FBS), 1% penicillin–streptomycin (10,000 U/mL stock) and 1% of 200mM L-Glutamine) at a final concentration of 1mg/ml prior to incubation at 37C for one hour with continuous shaking. Digested tissues were resuspended in 1X RBC lysis buffer (Cat# 420301, BioLegend) for 10 minutes, washed and filtered through a 70 um filter to obtain single cell suspensions. Tregs were purified using EasySep mouse CD4+CD25+ Regulatory T Cell Isolation Kit II (Cat# 18783, StemCell Technologies) according to the manufacturer’s protocol, and CD4+CD25- effector T cells were isolated using the same kit. Isolated cells were resuspended in staining buffer for downstream application.

Human Tregs were labeled with antibodies against CD4 (Cat# 317408, BioLegend), CD25 (Cat# 302610, BioLegend) and CD127 (Cat# 351319, BioLegend), and murine Tregs were labeled with antibodies against CD45 (Cat# 103139, BioLegend), CD4 (Cat# 100514, BioLegend), CD25 (Cat# 102049, BioLegend), FRδ (Cat# 125005, BioLegend), and FOXP3 (Cat# 126403, BioLegend). For transcription factors staining, Transcription Factor Staining Buffer Set (Cat #00-5523-00, eBioscience, San Diego, CA) was used according to manufacturer’s protocol. All flow cytometry sample analyses were carried out on Attune NXT analyzer with data analysis performed using FlowJo v10.7.2 analysis software.

Eight-week-old BALB/c mice were implanted with 4T1 tumors (1 × 10⁵ cells per mouse) with one flank tumor on each animal, and mice were randomized. Once tumors reached a size of approximately 600 mm³, mice were injected with 10 nmols of raltitrexed-S0456 either with or without 100-fold excess of competitor (raltitrexed-glucosamine), or 10 nmols of folate-S0456 with or without 100-fold excess competitor (folate-glucosamine). Two hours later, mice and the excised organs were imaged using Spectral Ami Optical Imaging System. The tumors were digested using Mouse Tumor Dissociation Kit (Cat# 130-096-730, Miltenyi Biotech) according to manufacturer’s protocol. Digested tumor cells were resuspended in 1X RBC lysis buffer (Cat# 420301, BioLegend) for 10 minutes, washed and filtered through 70 um filters, and resuspended in 2% FBS in Phosphate Buffered Saline (PBS) to obtain single-cell suspensions. Spleens and other tissues from the same mice were isolated as described previously to obtain single-cell suspensions.

Murine CD4+CD25- effector T cells (Teff) were obtained as described previously. Cells were stained with Carboxyfluorescein succinimidyl ester (CFSE) using CellTrace™ CFSE Cell Proliferation Kit (Cat #C34554, ThermoFisher Scientific). Briefly, cells were suspended at 1 x 10⁶ per milliliter of serum-free PBS, and CFSE was added with agitation at 5uM final concentration. Cells were incubated for 10 minutes at room temperature. CFSE dye was quenched with 3X serum-containing media, which was added with gentle agitation. Quenched media was put on ice for 5 minutes and washed twice. Tregs (either treated or untreated) and Teff cells were cocultured at 1:4 ratio in complete RPMI with or without mouse T-cell activator CD3/CD28 beads (Cat #11456D, ThermoFisher Scientific). After four days, samples were analyzed on flowcytometry to track proliferation of CD4+CD25- Teff cells (28).

Tregs and Teff cells were isolated as described previously. Tregs were incubated with or without treatment for 3 hours, then cells were co-cultured at 1:4 ratio of Tregs : Teff cells in complete RPMI. Cells were then incubated with or without mouse T-cell activator CD3/CD28 beads for 48 hours. Subsequently, supernatant was collected and either used directly for ELISA assay or stored at -80 for later analysis.

BALB/c mice or C57BL/6 (6 to 8-week-old) were implanted with 0.5 × 10⁵ 4T1 tumor (subcutaneously). Treatment was initiated when tumors reached 50 mm³. Treatments were carried out daily for 5 days per week. Tumor growth and animal weight change were monitored during the study.

All reagents were purchased from commercial sources and were used without further purification. All the reagents and chemicals for synthesis were purchased from Sigma-Aldrich (St. Louis). All other cell culture reagents, syringes, and disposable items were purchased from VWR (Chicago, IL). All preparative HPLC was performed with an Agilent 1200 Instrument with a reverse-phase XBridge OBD preparative column (19 × 150 mm, 5 μm) manufactured by Waters (Milford, MA) with UV detection at 254 nm. LC/MS was performed on an Agilent 1220 Infinity LC with a reverse-phase XBridge Shield RP18 column (3.0 × 50 mm, 3.5 μm). The purity of all final compounds was ≥95% as determined by analytical HPLC on a reverse-phase column with the binary system ammonium acetate (20 mM, pH – 7) and acetonitrile as eluent.

Statistical analysis was performed using a Student’s t-test (paired, single-tail analysis) with data represented as mean ± SEM. Prism 7.0 (GraphPad) was used for all statistical analyses and graph generation. Significance is shown as ns, P > 0.05; *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001; ****, P ≤ 0.0001.

As noted in the Introduction, folate receptor delta (FRδ) is only expressed on regulatory T cells (Tregs), a small subset of memory CD4+ T cells and ovum, suggesting that it could constitute a receptor for targeting drugs to Tregs. To explore this hypothesis, we searched for a ligand that would bind to FRδ and deliver an attached drug into Tregs, but display little or no affinity for other cell types. Although FRδ exhibited no affinity for folic acid (26) its strong structural homology with other isoforms of the folate receptor suggested that it might bind a folate-like molecule. To test this possibility, we employed molecular docking software (Glide docking tool in the Schroedinger Suite) to identify a molecule with the folic acid scaffold that might bind FRδ. For this purpose, virtual screening of a library of ~3 million commercially available compounds was performed in silico, and their binding free energies (MMGBSA dGbind) and docking scores (XP GScore) for both human and mouse FRδ were calculated. As seen from the representative examples in Figure 1A , raltitrexed (panel B), a synthetic antifolate that binds to FRα (29) exhibits one of the lowest MM-GBSA binding free energies for both human (-74.29 kcal/mol) and mouse (-67.95 kcal/mol) FRδ, with several other FDA-approved antifolates exhibiting slightly lower predicted binding affinities (see also binding poses of folate and raltitrexed on human and murine FRδ in panels C & D and Supplemental Figure 1 ). Not surprisingly, folic acid displayed the worst predicted binding free energy among the well-known folate analogs, with values of a -42.29 kcal/mol and -30.95 kcal/mol for human and mouse FRδ, respectively. Docking scores, which are generally less reliable predictors of binding constants, also identified raltitrexed as a strong FRδ ligand ( Figure 1A ). Because raltitrexed had already established a good safety profile in humans (30), we elected to explore its possible use for targeting drugs to Tregs further.

To directly test whether raltitrexed might bind FRδ, we first synthesized a fluorescently labeled raltitrexed conjugate (Ral-S0456; see synthesis in Figure S2 ) and then examined its ability to bind human and murine Tregs isolated from either peripheral blood or other healthy tissues. As shown in Figures S3A–D , fluorescent raltitrexed bound neither human nor murine Tregs, nor did similar fluorescent conjugates of either folic acid or 5-methyltetrahydrofolate (synthesis described in ref. 20). That FRδ was indeed expressed on these murine Tregs ( Figure S3D ), however, was confirmed by showing that an anti-FRδ monoclonal antibody strongly labels the Tregs in the same preparations. We therefore concluded that raltitrexed, folate and 5-methyltetrahydrofolate do not bind FRδ on Tregs isolated from peripheral blood or healthy tissues.

Because FRβ, an isoform of FR that is only expressed on myeloid cells (31), is known to exist in both a folate binding and nonbinding state, and since it converts from its nonbinding to folate binding state following macrophage infiltration into solid tumors (or sites of inflammation) (32–34), we wondered whether FRδ might similarly convert from a nonbinding to binding state following Treg migration into a tumor mass. To test this hypothesis, we implanted folate receptor negative 4T1 murine breast cancer cells into BALB/c mice and injected the tumor-bearing mice intravenously with the same fluorescent folate-S0456 or Ral-S0456 conjugate employed above ( Figure S2 ). As shown in Figures 2A, B , both fluorescent conjugates labeled tumors and kidneys (i.e., the site of excretion) but no other tissues, suggesting that the tumor microenvironment must induce an active binding site for both folate and raltitrexed. Moreover, because tumor uptake of the two targeted conjugates could be blocked by co-injection of 100-fold excess nonfluorescent conjugate (i.e., folate- or Ral-glucosamine conjugate), we concluded that their tumor accumulation must be receptor-mediated (see both panels A & B).

In order to identify which cells in the tumor masses took up the fluorescent conjugates, flow cytometry analysis was performed on cells obtained by dissociating the above tumors into single cell suspensions. As shown in Figure 2C , CD45+CD4+CD25+FRδ+ cells (i.e., Tregs) specifically captured the Ral-S0456 in a manner that was readily competed with excess Ral-glucosamine, i.e. confirming that the tumor Tregs could indeed bind raltitrexed. Analogous studies of tumor cells isolated from folate-S0456 injected mice similarly confirmed that Tregs did not bind the folate-targeted dye (compare red and blue histograms), i.e., consistent with previous reports that Tregs do not bind folic acid (26). More importantly, CD45+CD4+CD25+FRδ+ cells (Tregs) from spleens of Ral-S0456 injected mice remained nonfluorescent ( Figure 2C ), confirming that FRδ on Tregs in healthy tissues persists in its nonbinding state. Considered together with data demonstrating that ~80% of the CD45+CD4+CD25+FRδ+ cells are indeed FOXP3+ ( Figure S4 ), we concluded that raltitrexed does indeed bind Tregs in tumors and that it can be exploited to target attached drugs to Tregs in tumors without delivering the same drugs to nonbinding Tregs in normal tissues.

To more thoroughly characterize the specificities of raltitrexed and folate for the various cells in tumor masses, we next examined the uptake of fluorescent conjugates of both raltitrexed and folate by other cells in the tumor cell suspensions. As shown in Figures 2D , S5A , CD45- cells were all nonfluorescent, confirming that 4T1 cancer cells, fibroblasts, and endothelial cells etc. do not bind the targeted dyes. Moreover, as expected, all folate receptor negative white cells (CD45+) also remained nonfluorescent ( Figures S5B , C ); i.e., concordant with their lack of folate receptors. In contrast, folate-S0456 was aggressively captured by CD45+CD11b+F4/80+FRβ+ cells (tumor-associated macrophages and myeloid derived suppressor cells) ( Figure S5D ), but not by CD45+CD4+CD25+FRδ+ Tregs, whereas Ral-S0456 was primarily accumulated by FRδ+ Tregs and to a lesser extent by CD4+CD25-FRδ+ cells (memory T cells, Figure S5C ) and CD45+CD11b+F4/80+FRβ+ myeloid cells. This higher uptake of folate-S0456 than Ral-S0456 by FRβ+ tumor-associated macrophages and myeloid derived suppressors cells was consistent with the relative affinities of folate and raltitrexed for FRβ.

To more firmly establish that uptake of Ral-S0456 by Tregs is mediated by FRδ, we next produced a FRδ knockout mouse on a C57BL/6 background and repeated the above in vivo labeling studies using a syngeneic tumor (MB49) that grows in C57BL/6 mice. As shown in Figure 2E , uptake of Ral-S0456 by tumor-derived Tregs was largely absent in the FRδ knockout mice, confirming that FRδ is responsible for the vast majority of Ral-S0456 accumulation by Tregs. Collectively, the data also established that tumor- but not spleen-associated Tregs bind and internalize raltitrexed- but not folate-dye conjugates via a mechanism that requires a functional FRδ generated in the tumor milieu.

With the ability of raltitrexed to deliver an attached molecule selectively to tumor-infiltrating Tregs established, we next hypothesized that a raltitrexed-linked drug might be exploited to reduce Treg immunosuppression in solid tumors without perturbing immune functions in healthy tissues. To test this hypothesis, we first compared the abilities of several immune modulators to lower production of the immunosuppressive cytokine, IL-10, by tumor-derived Tregs. As shown in Figure S6 , a Toll-like receptor 7 agonist (TLR7-1a) (32, 35) was found to suppress IL-10 biosynthesis better than inhibitors of HDAC8, EZH2, or Dyrk1a (see also Figure 3A ). Importantly, the same TLR7 agonist was also successful in inhibiting release of the immunosuppressive cytokine, TGFβ ( Figure 3B ) and in reversing the ability of Tregs to block proliferation of effector T cells (compare panels 1-4 of Figure 3C ). The proliferation was quantitated by calculating the area under the curve of the dividing cells to the left of the non-dividing T-cell peak, as illustrated in the bar graph ( Figure 3D ). Based on these observations, we concluded that TLR7-1a would constitute an excellent candidate for assessing the capability of raltitrexed to deliver an immune modulating drug specifically to Tregs in vivo.

To explore the ability of Ral-TLR7-1a to modulate the immunosuppressive properties of Tregs in solid tumors, raltitrexed was linked via a self-immolative spacer to TLR7-1a (see synthesis in Figure S7 ) such that upon internalization of the conjugate (hereafter referred to as Ral-TLR7-1a), the consequent reduction of the disulfide bond in the spacer would trigger release of an unmodified TLR7-1a into the Treg. To quantitate the potency of the targeted Ral-TLR7-1a, immune competent mice with 4T1 syngeneic tumors were treated via tail vein injection with 10 nmoles Ral-TLR7-1a and examined for changes in both tumor growth and body weight. As shown in Figure 4A (left panel), Ral-TLR7-1a reduced tumor growth by ~60% relative to untreated controls. Free raltitrexed, in contrast, inhibited tumor growth by only 20%, suggesting that the majority of growth inhibition by Ral-TLR7-1a was contributed by the TLR7-1a (none of these treatments caused significant mice body weight changes, as illustrated in Figure 4B ) (36). That this growth inhibition was receptor-mediated could then be demonstrated by showing that blockade of FRδ with 100-fold excess Ral-glucosamine reversed the inhibition almost quantitatively (green curve). Moreover, the observation that none of the treated animals lost weight during the three-week therapy suggests that Ral-TLR7-1a was not inhibiting the essential functions of Tregs in healthy tissues and was therefore not toxic to the animals. And since analogous results were obtained when mice implanted with CT26 syngeneic colon cancer tumors were similarly treated ( Figures 4C, D ), we conclude that Ral-TLR7-1a inhibition of tumor growth is likely a property of most Treg-containing tumors.

In order to define the mechanism by which Ral-TLR7-1a inhibits tumor growth in greater detail, we dissociated the tumor masses into single cell suspensions and examined their diagnostic markers by flow cytometry. As shown in Figure 5A , levels of FOXP3, CTLA4, PD1, and HELIOS on Tregs all decreased upon treatment with Ral-TLR7-1a, suggesting that the Tregs were shifted towards a less immunosuppressive state (37, 38). Consistent with the known engagement of other immune cells by Tregs, activation markers on CD8+ T cells (i.e. granzyme B, IFNγ, and CD69) increased, while a marker of T cell exhaustion (Lag3) decreased ( Figure 5B ), i.e. demonstrating that mitigation of the immunosuppressive properties of Tregs allows effector T cells to become more cytotoxic and less exhausted. Not surprisingly, the CD86 to CD206 ratio on tumor-associated macrophages and myeloid-derived suppressor cells [i.e. often considered a measure of M1:M2 ratio (32, 39)] also increased ( Figure 5C ), further suggesting that the targeted reprogramming of Tregs also impacts tumor-infiltrating myeloid cells. Finally, and perhaps most importantly, analysis of the same phenotypic markers on Tregs from spleens of the same mice revealed no change in any of the diagnostic markers, consistent with the absence of raltitrexed binding sites on splenic Tregs and supportive of the hypothesis that tumor-derived Tregs can be manipulated without altering the properties of Tregs systemically (representative data shown in Figure S8 ).

It should also be noted that raltitrexed displayed affinity, albeit much weaker, for FRβ on tumor-associated macrophages ( Figure 2D ), suggesting that part of the impact of Ral-TLR7-1a on tumor growth could have derived from its effect on FRβ+ tumor-associated macrophages. Therefore, to obtain a more accurate estimate of the impact of Ral-TLR7-1a solely on Tregs, we repeated the above studies using mice in which we knocked out the gene for FRβ. As shown in Figure 6 , Ral-TLR7-1a was surprisingly even more effective in suppressing tumor growth in the FRβ knockout mice than in wildtype mice, reducing tumor growth by >80% (compare Figures 4A , 6A ). Whether this enhanced potency was a consequence of the specific syngeneic tumor model (i.e., MB49 versus 4T1) or the switch from balb/c to C57Bl/6 mice was not investigated, but the data clearly confirmed that Ral-TLR7-1a mediated reprogramming of tumor Tregs exerts a profound effect on tumor growth even in the absence of any contribution of the drug to suppression of tumor-infiltrating myeloid cells. Mechanistically, the data in panel C also demonstrate that the effect of Ral-TLR7-1a on the MB49 tumor microenvironment is not limited to Tregs; i.e. by reversing the immunosuppressive activities of the Tregs, the anti-inflammatory properties of other immune cells in the 4T1 tumor microenvironment are also improved.

Finally, although the initial purpose of this study was to determine whether tumor-associated Tregs could be manipulated to adopt a less immunosuppressive state, the question naturally arose whether one could also induce Tregs to acquire a more immunosuppressive phenotype. To explore this possibility, we synthesized a dexamethasone conjugate of raltitrexed (Ral-Dex; see Figure S9 for structures and synthetic schemes) and examined its impact on tumor growth. As shown in Figure 7A , targeted dexamethasone actually enhanced tumor growth relative to untreated controls. Moreover, as revealed in Figure 7B , this augmentation of growth was predictably accompanied by an opposite shift in phenotypic markers on the other tumor resident immune cells. Taken together, these data suggest that the immunosuppressive potential of tumor-infiltrating Tregs was not maximal in the untreated tumors and that their immunosuppressive functions could in fact be further enhanced by targeted stimulation with an immunosuppressive steroid.