THP-1 cells (Cat# TIB-202), 22RV1 cells (Cat# CRL-2505), MDA-MB-231 cells (Cat# CRM-HTB-26), Raw264.7 cells (Cat# TIB-71) and 4T1 cells (Cat# CRL-2539) were purchased from ATCC. Unless otherwise specified, cells were cultured in folate-free RPMI 1640 medium (Gibco, Ireland) containing 10% heat-inactivated fetal calf serum and 1% penicillin-streptomycin in 5% CO₂ at 37°C. When cells reached confluence, spent medium was harvested and centrifuged to remove cells and debris. The resulting supernatant was filtered twice through a sterile 0.2μm cell strainer and then used to differentiate M0 into M2 macrophages as described below.

Fresh whole human blood was collected from healthy donors and was either used directly for cytokine stimulation studies or isolation of peripheral blood mononuclear cells (PBMCs) using density gradient media (Ficoll-Paque PLUS, Cat# 17144002) as described previously (18). The isolated monocytes were differentiated into M0-like macrophages by incubation with 20ng/ml human M-CSF for 7 days (18), after which the M0-like macrophages were polarized into M2-like macrophages by further incubation for 2 days in the presence of 20ng/ml IL-4 plus 20ng/ml IL-13 or the above spent cell culture medium. An FR-positive subclone of the murine macrophage-derived RAW264.7 cell line was obtained as previously described (19). Cells were then analyzed for FRβ expression by flow cytometry (see Supplemental Tables S1 , S2 ).

To assess the effect of TLR7-1A and FA-TLR7-1A on cytokine production, FR-positive Raw264.7 cells were seeded to 96-well plate in a density of 0.2 million cells/well. Cells were then incubated for 24 hours with indicated concentration of either FA-TLR7-1A or TLR7-1A in the absence or presence of 100X excess folate-glucosamine (competition). The culture medium was then collected for cytokine analysis using ELISA.Analysis of TLR7-1A and FA-TLR7-1A induced cytokine release in fresh whole human blood was performed similarly, except the whole blood was treated with either a vehicle control, 100nM TLR7-1A, or 100nM FA-TLR7-1A for 4 hours before centrifugation for 10 minutes at 1,000g. The resulting cell pellets were hemolyzed in RBC lysis buffer (BioLegend; Cat# 420301) and the residual white cells were analyzed using quantitative polymerase chain reaction (qPCR) to quantitate cytokine mRNA levels. Plasma supernatants were also analyzed by ELISA for secreted cytokines as described below.

Six- to eight-week-old female Balb/c mice were purchased from Envigo and used for all live animal studies. Mice were housed in accordance with protocols approved by Purdue University Animal Care and Use Committee. Water and folate-deficient chow (Envigo, Cat#TD.00434) were freely available.

Mice were subcutaneously injected with 4T1 cells (50,000 cells/mouse) and tumor sizes were measured every other day. When tumors reached 500-600mm³, mice were sacrificed and perfused intracardially with chilled phosphate buffered saline (PBS). Part of each tumor was then dissociated with tumor dissociation kit (Miltenyi, Cat#130-096-730) and erythrocytes were depleted using RBC lysis buffer. After two washes with cold PBS, the resulting single cell suspensions were analyzed by flow cytometry for the desired phenotypic markers ( Supplemental Tables S1 , S2 ). The remaining portion of each tumor was fixed in 4% formalin, followed by cryoprotection in 30% sucrose solution, then cryosectioned into 15μm sections and stored as free-floating sections at 4°C for subsequent immunofluorescence staining.

For analysis of monocytes from the peripheral blood of healthy mice, peripheral blood was collected by cardiac puncture prior to washing 3x in PBS. Spleens from the same mice were also resected and gently minced before extrusion through a 70 μm cell strainer. Erythrocytes in the resulting cell suspension were then lysed using RBC lysis buffer and residual cells were washed 2x in PBS. The resulting PBMCs and splenocytes were then stained with the desired antibodies and analyzed by flow cytometry ( Supplementary Tables 1 , 2 ). Analysis of total and functional FRβ expression in human PBMCs was performed similarly, except that the PBMCs were isolated from healthy human peripheral blood as described above.

Healthy mice were injected intravenously with 10nmol/mouse of either FA-TLR7-1A or TLR7-1A, and peripheral blood was collected at the indicated timepoints. Blood was then centrifuged at 4°C and 1,000g for 10 minutes, and the resulting plasma was analyzed for cytokines using ELISA as described below.

Single cell suspensions from both mice and human samples were first stained with Zombie Violet (BioLegend, Cat#423114) and then washed 2x with PBS prior to incubation with anti-mouse TruStain FcX™ (BioLegend, Cat#101319) or anti-human TruStain FcX™ (BioLegend, Cat# 422301), respectively, to block nonspecific Fc domain binding. Cells were then stained with the desired antibodies listed in Supplementary Tables 1 , 2 and washed 2x with PBS. For intracellular TLR7 staining, cells were fixed and permeabilized using Cyto-Fast™ Fix/Perm Buffer Set (BioLegend, Cat#426803), followed by staining with the anti-mouse TLR7 monoclonal antibody ( Supplementary Tables 1 , 2 ). To quantitate the functional subset of FRβ expressing cells, a folate receptor-targeted fluorescent dye conjugate (folate-Cy5, synthesized in-house; or folate-fluorescein, MedChemExpress, #910661-33-5) were used. After washing, cells were resuspended in FACS buffer and examined using an Attune NxT flow cytometry prior to data analysis using Attune Cytometric Software or FlowJoucodep™ v10.

Human IL-6, TNFα and IFNγ were quantified using an ELISA MAX™ Deluxe Set (BioLegend) kit, while similar mouse-specific kits (BioLegend) were used for quantitation of mouse IL-6 and TNFα.

4T1 tumors were collected, processed, and stained as described previously, with minor modifications (20). Briefly, sections were immunostained overnight with anti-F4/80 (BioLegend, Cat#123101, 1:1000), anti-TLR7 (Novus Biologicals, Cat#NBP2-27332SS, 1:1500), and anti-FRβ (AF647, F3 IgG29, 1:100) (21) primary antibodies. Secondary antibody staining was performed with Alexa Fluor^® 488 and Alexa Fluor^® 594 (Jackson ImmunoResearch, 1:2000) antibodies. Sections were then mounted on glass slides in ProLong Gold Antifade Mounting media. Prepared slides were shipped to iHisto, Inc. for whole slide scanning.

After erythrocyte depletion, RNA was isolated from total leukocytes, reverse-transcribed to cDNA and analyzed by qPCR using the methods and primer sequences described previously (18). Melting curve analysis was performed to validate specificity. Each sample was analyzed in triplicate for each marker.

The 4T1 tumor model and treatment strategy described in our previous publication were used (17). In brief, 3nmol/mouse of FA-TLR7-1A was administered intravenously 5 times per week for two weeks. Mice were then sacrificed and tumors were dissociated as described above to obtain single cell suspensions. Viable cells were enriched using Dead Cell Removal Microbeads (Miltenyi, Cat# 130-090-101) and an LS Colum (Miltenyi, Cat# 130-042-401) in the magnetic field of a MidiMACS™ Separator, after which cells were diluted to 1 million cells/ml in ice-cold PBS containing 0.04% Ultrapure BSA (50mg/ml, ThermoFisher, Cat#AM2616) and cell viability (>95%) was again validated by flow cytometry. The whole procedure was completed within 1 hour and the resulting viable cells were used for scRNA-seq library construction immediately.

Samples were handled in accordance with 10x Genomics protocols (Next GEM Single Cell 3’ Reagent Kit v3.1). Following 10x loading suggestions, ~16,000 cells were loaded into each well of the chip for a targeted cell recovery of 10,000 cells per lane. Chip G from 10x Genomics was utilized for GEM generation and all steps following GEM generation were carried out as described in the 10x Genomics manual. After GEM processing on the chromium controller, all steps to amplify cDNA and generate libraries were carried out according to the 10x Genomics manual. Quality control procedures were performed by the Purdue University Genomics Core Facility after cDNA amplification and library generation. All sequencing was then performed by the Purdue University Genomics Core Facility. Sequencing was targeted to approximately 50,000 paired end reads per cell.

Sequencing reads from the Chromium system were de-multiplexed and processed using the CellRanger pipeline v6.0.2 (10x Genomics). CellRanger count was then used for alignment, filtering, barcode counting, and unique molecular identifier (UMI) counting. All reads were aligned to the ENSEMBL mouse genome version mm10 using the STAR aligner v2.7.2a (22). CellRanger was run with the number of expected cells set to 5,000. R version 4.0.1 and Bioconductor version 3.11 were used in all statistical analyses. Cells that had fewer than 200 or greater than 6000 observed genes were discarded. Cells were also removed if greater than 15% of all reads mapped to mitochondrial genes or greater than 50% of all reads mapped to ribosomal genes. The metrics indicate high quality data and that the average number of cells captured approached the intended number of 5000 cells.

All scRNA-seq data were deposited to GEO and are available under accession number GSE236443. Seurat version 4.0.5 (23) was used for data normalization and cell clustering based on differential gene expressions. Data were normalized using scTransform v0.3.2 (24) followed by integration using standard Seurat workflow. These “corrected” data, after permutation and selection of the first 20 principal components based on principal component analysis (PCA) scores, were used for downstream analysis. Unsupervised clustering was performed in Seurat, which uses graph-based approaches to first construct K-nearest neighbor graphs (K = 20) and then identifies clusters by iteratively forming communities of cells to optimize the modularity function. The number of clusters were determined using the Louvain algorithm (25) for community detection (as implemented in Seurat), with a resolution of 0.6. The correct resolution to use was determined both visually through plots and heat maps as well as using clustering trees via the Clustree R package v0.4.3 (26), selecting a resolution that provides stable clusters. Biomarkers were considered statistically significant at a 5% false discovery rate (FDR) using the Wilcoxon rank sum test (27). Differentially expressed genes between sample groups were identified using the Wilcoxon rank sum test with an FDR cutoff of 5%. P-values were corrected for multiple testing using the Benjamini-Hochberg method (28) wherever applicable. Sub-clustering of major cell-types (myeloid, T/NK, and fibroblasts), optimal clustering, biomarker identification and differential expression analysis were performed using the same approach as above. Quantitation of the changes in population sizes of the different cell clusters was performed using the permutation test in R-package scProportionTest (29). Changes in population sizes with FDR < 0.05 were denoted as statistically significant.

Further analysis of the correlation between FRβ expression and cancer patient survival was done by extracting FRβ expression and patient survival data from the TCGA database and analyzed using UALCAN (30).

All cells from a tumor mass were first clustered into a total of 19 primary clusters. To identify the cells within these clusters, differentially expressed marker genes specific to each cluster were analyzed. The clusters were identified and labeled accordingly, as provided in Supplementary Table 3 . To further investigate the impact of FA-TLR7-1A treatment on myeloid cells, T cells/NK cells and fibroblasts, these three major cell groups were extracted separately and subjected to sub-clustering as described above. The differentially expressed marker genes, as listed in Supplementary Table 3 , were used to characterize the relevant sub-clusters within these cell types.

Statistical analyses were conducted using GraphPad Prism version 10 software (Graphpad; San Diego, CA). All figures present mean ± s.d. values unless stated otherwise. To compare multiple groups, a one-way ANOVA followed by a Dunnett’s multiple comparison test was utilized when applicable. Significance levels were marked as follows: (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001).

When analyzing scRNA-seq data, we identified statistically significant changes between sample groups in population size and differentially expressed genes by utilizing an FDR threshold of < 0.05, as mentioned above.

As noted in the Introduction, because FRβ is expressed solely on myeloid cells (5, 6) and since its expression in tumor tissues correlates with poor overall survival (5, 30–32) ( Supplementary Figure 1 ), FRβ has become both an important marker for tumor infiltrating TAMs/MDSCs and an intriguing target for receptor-directed therapeutics (4, 6, 7, 9). However, a major question has remained regarding the ability of FRβ to bind folic acid, since some publications report high affinity binding (6, 33, 34) while others report no detectable binding (35, 36). To evaluate whether the FRβ-expressing myeloid cells in a tumor mass might bind folic acid, 4T1 murine breast cancer cells were implanted in immune competent Balb/c mice and resulting tumors were dissociated into component cells prior to incubation with a folic acid conjugate of the fluorescent dye Cy5 (FA-Cy5). As shown in Figure 1A , although anti-FRβ monoclonal antibody stained ~46% of CD11b+F4/80+ cells, FA-Cy5 bound to only 19% of the same TAMs; i.e. suggesting that not quite half of all FRβ+CD11b+F4/80+ cells express a functional folate receptor. Moreover, because FA-Cy5 binding could be blocked by excess folate-linked glucosamine (i.e. a competitor of folate binding to FR), we conclude that FA-Cy5 binding was FRβ-specific. These data demonstrate that less than half of TAMs in the tumor mass express a functional FRβ.

To examine FRβ expression and functionality in nonmalignant tissues, freshly drawn blood from both tumor-bearing and nontumor-bearing mice was incubated with either anti-FRβ antibody or FA-Cy5, and the above flow cytometry experiments were repeated. As shown in Figure 1B and Supplementary Figure 2 , fewer than 1% of circulating monocytes from the blood of healthy mice bound anti-FRβ monoclonal antibody, and essentially none of the monocytes from the same mice were stained with FA-Cy5. Because similar results were also obtained on myeloid cells from the spleens of healthy mice ( Figure 1B ), we conclude that functional FRβ is primarily detected in a tumor mass.

To assess the expression and functionality of FRβ on human peripheral blood mononuclear cells (PBMCs), we isolated fresh PBMCs from healthy donors and incubated them with both a monoclonal antibody to human FRβ and FA-Cy5, approximately as described above. In contrast to murine monocytes, ~40% of human monocytes stained positive for FRβ (i.e. suggesting that the receptor is present on a population of human monocytes). However, only a fraction of these monocytes bound the fluorescent folate conjugate, i.e. indicating that the folate receptors were largely nonfunctional ( Figure 1C and Supplementary Figures 3A–H ).

Because previous studies had suggested that myeloid cells could bind folic acid and its conjugates in solid tumors, we then decided to investigate the nature of the tumor stimuli that might induce folate binding by FRβ. For this purpose, we collected conditioned cell culture media from three different cancer cell lines and examined their abilities to activate FRβ in human PBMC-derived monocytes. As shown in Figure 1D , while 22RV1 human prostate cancer cell medium induced little or no FRβ activation, THP1 human monocytic leukemia culture medium promoted moderate FRβ activation and MDA-MB-231 human breast cancer medium stimulated strong FRβ activation. Importantly, incubation of macrophages from the same healthy donors in the presence of IL-4/IL-13 cytokine combination [commonly used to differentiate human monocytes into M2-like immunosuppressive macrophages (18, 37)] also induced FRβ activation. These data demonstrate that conversion of FRβ from nonfunctional to functional state can derive from factors secreted by cancer cells or induced by a combination of the immunosuppressive cytokines. The fact that these same inducing factors are largely absent from healthy tissues further explains why macrophages with functional FRβ are essentially absent from noninflamed tissues (17, 33).

Motivated by the fact that functional FRβ is almost exclusively expressed in malignant (or inflamed) tissues, we next undertook to determine which immune-activating pathway might be prominent in FRβ+ TAMs/MDSCs, but only activatable by intracellular drugs that could be internalized by FRβ-mediated endocytosis. To determine whether an intracellular receptor like toll-like receptor 7 (TLR7) might co-localize with FRβ in TAMs or MDSCs, 4T1 tumors from the aforementioned mice were dissociated into single cell suspensions and analyzed for both FRβ and TLR7. As shown in Figures 2A, B and Supplementary Figures 4A–E , 4T1 cancer cells themselves expressed no folate receptors, however, ~58% of TAMs/MDSCs and ~14.2% of myeloid-derived dendritic cells expressed FRβ. Moreover, examination of the same cell suspensions for TLR7 revealed that ~67.5% of TAMs/MDSCs expressed TLR7, and ~46.1% of these same cells produced both FRβ and TLR7 ( Figure 2A ). Although TLR7 was also widely expressed in other cell types, including T cells, dendritic cells, and a fraction of cancer cells ( Figure 2C , Supplementary Figures 4B–D ), the only cells that expressed both FRβ and TLR7 resided in the subset of myeloid cells. Not surprisingly, immunofluorescent staining of the same tumor tissues with antibodies to FRβ and TLR7 confirmed that FRβ and TLR7 colocalize only in tumor macrophages ( Figure 2D ). Because a similar colocalization was also found by scRNA-seq in human ovarian, lung and breast cancer specimens ( Figures 2E–G ), we conclude that FRβ is commonly co-expressed in M2-like macrophages with TLR7 receptors ( Supplementary Figures 6A–C ) (4, 6, 7, 38). Because TLR7 receptors are present in intracellular endosomes (39), and since FRβ traffics to intracellular endosomes (40, 41), this colocalization suggests that a folate-targeted TLR7 agonist might activate TAMs/MDSCs in solid tumors without stimulating similar myeloid cells in healthy tissues.

With acquisition of functionality in FRβ restricted to the tumor microenvironment, we decided to evaluate whether a folate-targeted TLR7 agonist might be able to induce a proinflammatory phenotype in tumor myeloid cells without activating immune cells in healthy tissues. For this purpose, we linked folic acid (FA) to a potent TLR7 agonist (TLR7-1A) in a manner that created a conjugate (FA-TLR7-1A) that would be internalized by functional FRβ-expressing cells but impermeable to cells lacking a functional folate receptor (17). To test the potency and specificity of this conjugate in stimulating FRβ+ myeloid cells, we treated an FRβ-expressing clone of murine macrophage-derived RAW264.7 cells ( Supplementary Figure 5 ) with either FA-TLR7-1A or nontargeted TLR7-1A in the absence or presence of 100-fold excess folate-glucosamine (FA-Glu) to block all FRβ. As shown in Figure 3A , both FA-TLR7-1A and TLR7-1A induced production of IL-6 with similar potency in the absence of FA-Glu, but only nontargeted TLR7-1A stimulated IL-6 release in the presence of FA-Glu. These observations demonstrate that FA-TLR7-1A can only induce IL-6 production in functional FRβ-expressing macrophages, and that FA-TLR7-1A is inactive when FRβ-mediated uptake is blocked by excess FA-Glu.

Because the ability of FA-TLR7-1A to stimulate proinflammatory cytokine production in a tumor microenvironment without inducing cytokine release systemically could be therapeutically useful, we elected to confirm this selectivity in humans by incubating fresh whole human blood from healthy donors with either FA-TLR7-1A or free TLR7-1A and measuring secretion of inflammatory cytokines ( Figures 3B, C ). Although free TLR7-1A was found to induce strong TNFα, IL-6, and IFNγ production, FA-TLR7-1A elicited no detectable cytokines in the same samples, i.e. because healthy human PBMCs express no functional folate receptors ( Figure 1C ) and FA-TLR7-1A cannot enter FR-negative cells.

Finally, to examine this same dependence on functional FRβ in live mice, healthy mice were intravenously injected with either free TLR7-1A or FA-TLR7-1A, and serum cytokine levels were measured in the peripheral blood as a function of time. As shown in Figure 3D , although TLR7-1A promoted strong systemic cytokine release, similar injection of FA-TLR7-1A promoted no cytokine discharge. Considered together with the data in Figures 3A–C , these results establish that FA-TLR7-1A will only induce inflammatory cytokine production in cells that express functional FRβ, and because FRβ is only activated in tumor (or inflamed) tissues, these data argue that FA-TLR7-1A should constitute a safe method for reprogramming TAMs and MDSCs in a tumor mass without inducing their toxic cytokine release in healthy tissues. This specificity is obviously important because systemic activation of the immune system has prevented the FDA from approving any systemically administered TLR7 agonists in the past (42–44).

With the mechanism of FA-TLR7-1A’s tumor specificity established, the question finally arose regarding which genes in TAMs and MDSCs might be activated by FA-TLR7-1A. To explore this question, we employed single cell RNA sequencing (scRNA-seq) to compare the gene expression profiles of the major tumor-infiltrating myeloid cell subtypes following intravenous administration of FA-TLR7-1A. 4T1 tumor bearing mice were treated with either FA-TLR7-1A or vehicle control for 2 weeks prior to sacrificing for scRNA-seq analysis ( Supplementary Figure 7 ). Tumor cells were first clustered into 19 cell types ( Supplementary Figures 8A–C ), and myeloid cells, T cells/NK cells, and fibroblasts were further sub-clustered for following analysis ( Supplementary Figures 9A–F ). As shown in Figures 4A, B , tail vein injection of FA-TLR7-1A promoted significant changes in the cellular composition of the 4T1 tumors, with large increases in the abundances of mast cells, basophils, and neutrophils and smaller rises in the abundances of monocyte-derived macrophages and IFIT^high dendritic cells. In contrast, pro-apoptotic hypoxic myeloid cells and fully apoptotic cells were found to significantly decrease in numbers.

Analyses of the FA-TLR7-1A induced changes in myeloid cell gene expression revealed that transcripts of immune-activating and pro-inflammatory genes largely increased in abundance, while immunosuppressive, anti-inflammatory, and tumorigenesis-related transcripts predominantly decreased ( Figures 4C–H ). For example, S100 family genes (S100a8, S100a9, and S100a6) that enhance activation of NLRP3 inflammasomes and in turn promote expression of IL-1β and IL-18 are seen to increase in myeloid subpopulations (45) whereas immunosuppressive genes [Cd47 (46), Cd9 (47), Lgals3 (48), and Junb (49) etc.), tumorigenesis genes (Arpc1b (50), Ier3 (51), and Cxcl1 (52), etc.) and hypoxic genes (Phb2 (53), Rer1 (54), and Arl1 (55)] are observed to consistently downregulate. Not surprisingly, signaling components downstream of the NFκB (e.g. Ifitm1, Ifitm2, Ifitm3 and Ifi2712a) (56) are also found to increase in the TAMs/MDSCs of FA-TLR7-1A treated mice. Taken together, these data argue that FA-TLR7-1A shifts the tumor myeloid population towards a more pro-inflammatory, anti-tumor phenotype.

Because myeloid cells secrete cytokines and other factors that can influence the abundances and gene expression profiles of other cells in the same TME, we finally examined the changes in numbers of non-myeloid cell types in the same tumor masses. As shown in Figures 5A, B , cancer cells, endothelial cells, apoptotic cells and immunosuppressive myeloid cells generally decreased in number, while T/NK cells, fibroblasts, and mast cells increased in abundance, i.e. confirming the general shift towards a more immune-activating TME. Among the T and NK cell populations that were impacted most by FA-TLR7-1A, the exhausted T cells appeared to decline most dramatically, while the γδT17 cells, NK cells, Tregs and memory T cells all increased in number ( Figures 5C, D ). Not surprisingly, the abundance of profibrotic inflammatory cancer associated fibroblasts (iCAF) and collagen^high CAFs also increased, while the antigen-presenting CAF (ApCAF) population showed mixed trends. We speculate that this latter incongruity may have derived from initiation of a tissue repair response that was triggered by the inflammation induced by FA-TLR7-1A.