Cyclophosphamide and Vinorelbine Activate Stem-Like CD8⁺ T Cells and Improve Anti-PD-1 Efficacy in Triple-Negative Breast Cancer

Checkpoint inhibitors (CI) have shown an unprecedented clinical activity in several types of cancer, but only a fraction of patients has benefit from CIs administered as a single therapy. Moreover, in some patients, the clinical responses are limited in time (1). These observations are promoting preclinical and clinical studies to improve CI efficacy by means of combinatorial therapies (2).

The clinical success of cyclophosphamide (C)-based haploidentical stem cell transplants in hematologic malignancies indicates that this drug has the potential to re-orchestrate the immune system against cancer cells (3). Along a similar way, C is administered before CAR-T cell infusion to improve their clinical efficacy (4). We have previously found that low-dose, daily C, in association with vinorelbine (V) was able to improve the preclinical efficacy of CIs anti-PD-1 and anti-PD-L1 in models of breast cancer and lymphoma (5). As randomized clinical trials have recently indicated that the addition of CIs to other chemotherapy drugs such as taxanes (T), doxorubicin (D), or platinum (P) might be beneficial in patients with triple-negative breast cancer (TNBC; refs. 6–10), we designed this study in two preclinical, immunocompetent models of TNBC (4T1 and EMT6) to investigate whether C and V, in different schedules and doses, could be more effective than T, D, or P.

The first series of preclinical studies reported here have clearly indicated that intermittent, medium-dosage C (140 mg/kg, C140), a dosage already known to have potent effects on the immune system (11), was significantly more effective than lower, continuous doses of the same drug (20 mg/kg, C20), and that intermittent C140, in association with V and CIs, were significantly more effective than combinatorial regimens of CIs plus T, D, or P. Thus, we investigated by means of neutralizing mAbs, single-cell transcriptome analysis, and flow cytometry what immune cells were involved in these anti-TNBC mechanisms and how these drugs reshaped the circulating and intratumoral immune cell landscape in cancer-bearing mice.

A recent meta-analysis from clinical trials involving anti-PD-1 and anti-PD-L1 CIs indicated that anti-PD-1 had a favorable survival outcomes and a safety profile comparable to that of anti-PD-L1 (12). For this reason, we are showing here only the results obtained in our clinical models with the anti-PD-1 CI. At the present time, we have no evidence that anti-PD-L1 can be more effective of anti-PD-1 in these models.

The 4T1 and the EMT6 TNBC cell lines were purchased from ATCC, expanded and stored according to the producer's instructions. Cells were tested and authenticated by the StemElite ID System (Promega). Cells were tested every 6 months for Mycoplasma by means of the ATCC Universal Mycoplasma Detection Kit 30-1012, cultured for no more than 2 weeks and used for no longer than 15 passages.

The 4T1 and EMT6 cell lines were stably transduced with lentiviral vector expressing stably luciferase under the control of CMV promoter: pLenti CMV Puro LUC (w168-1; Addgene plasmid, #17477), carrying puromycin resistance as selective marker. To infect 4T1 and EMT6 cell lines, 293T cells were cotransfected with 10 μg of lentiviral vector, 3 μg PMD2G envelope plasmid, 2.5 μg of REV packaging plasmid, and 5 μg of RRE transfer plasmid using calcium phosphate transfection system. After 16 hours, media was removed and replaced with 5 mL of fresh medium. Viral supernatant was then collected at 24 and 48 hours and added directly to 4T1 and EMT6 cell lines plated at 1 × 10⁵ supplemented with 8 μg/mL of polybrene (Merck). Infection was performed for 3 hours at 37°C twice in a day for two separate days. After 48 hours from first cycle of infection, infected cells were selected adding 1.5 μg/mL (for 4T1 cell line) and 2 μg/mL (for EMT6 cell line) puromycin (Merck) to the medium. Selection was carried out for 72 hours, until not infected control cells were dead.

Experiments involving animals were approved by the Italian Ministry of Health and have been done in accordance with the applicable Italian laws (D.L.vo 26/14 and following amendments), the Institutional Animal Care and Use Committee, and the institutional guidelines at the European Institute of Oncology. In vivo studies were carried out in 6 to 8 weeks old immune-competent BALB/cOlaHsd female mice (Envigo) in mouse facilities at the European Institute of Oncology–Italian Foundation for Cancer Research (FIRC) Institute of Molecular Oncology (IEO–IFOM, Milan, Italy) campus. To generate syngeneic models of TNBC in BALB/c, 2 × 10⁵ 4T1-LUC and 5 × 10⁵ EMT6-LUC were injected in the mammary fat pad as we described previously (5). Tumor growth was monitored weekly using In Vivo Imaging System (IVIS; PerkinElmer). Briefly, mice were intraperitoneally injected with 150 mg/kg of XenoLight D-Luciferin–K+ Salt Bioluminescent Substrate (PerkinElmer). After 10 minutes, animals were anaesthetized with isofluorane apparatus and images acquired using Living Image Software (PerkinElmer; Supplementary Figs. S1A and S1B). Tumor dimension was finally assessed using ImageJ software, by converting pixel intensity into centimeters and determining finally tumor mass sphere volume in mm³.

TNBC resection was performed 28 (for all EMT6 experimental groups and for 4T1 control, anti-PD-1, D+ anti-PD-1, T+ anti-PD-1, P+ anti-PD-1 experimental groups) or 70 (for C140-treated experimental 4T1 tumor) days after tumor implant as we described previously (5, 13, 14). Fifteen days after mastectomy, mice were sacrificed by carbon dioxide inhalation and lung tissues were removed. To confirm the presence of metastases, sections were cut and stained with hematoxylin and eosin (H&E), as described previously (5). In brief, lungs were fixed in 4% phosphate-buffered formalin and embedded in paraffin. Five micrometers thick sections of lungs were made, and slides were counterstained with H&E for the detection of metastases. Images were acquired with a ScanScope XT scanner (Leica) and analyzed with Aperio Digital Pathology software (Leica). To assess the metastases in the lungs, ImageJ software was used. Areas occupied by the metastases was divided to total lungs' volume to assess the percentage of metastasis in the lungs.

Tumor-bearing mice (n  =  5 per study arm) were treated with either vehicle or with different drugs used as single agents or in combination. Drug dosages used in this study are the gold-standard chemotherapeutic drugs for TNBC and doses were based on literature data associated with no or acceptable toxicity, as well as no significant changes in mouse weight. C was used at two different doses: 20 mg/kg in drinking water (C20) and 140 mg/kg (C140) in intraperitoneal injection as already shown in refs. 5 and 11. V was used at 9 mg/kg as we have previously shown in ref. 5. P was used at 0.25 mg/kg (15), T was used at 20 mg/kg (16), D was used at 2.9 mg/kg (15). Chemotherapeutic drugs were dissolved in PBS and administrated, once a week, every 6 days, for 3 weeks by intraperitoneal injection. Mouse monoclonal PD-1-targeting antibody was purchased by Bioxcell (clone RMP1-14) and was administered intraperitoneally 0.2 mg/mouse every 2 days for a total of five doses. Monoclonal neutralizing antibodies were purchased by Bioxcell. Anti-CD4 (clone GK1.5), anti-CD8a (clone 2.43), anti-CD19 (clone 1D3), anti-CD11b (clone M1/70), anti-CD103 (clone M290), and anti-CD122 (clone TM-beta1) were administrated 300 μg/mouse twice a week for a total of five doses.

Mice tumors were surgically removed after 28 days from tumor injection to generate a single cell suspension. Briefly, after mechanical dissociation, tumors were placed in culture medium (1:1 of DMEM with high glucose and Ham's F-12 Nutrient Mixture; EuroClone) supplemented by 2 mg/mL collagenase (Merck) and 0.1 mg/mL Dnase I (Qiagen), and digested for 1 to 2 hours at 37°C. A single cell suspension was obtained by sequential dissociation of the fragments by gentle pipetting, to further disintegrate cell clumps, followed by filtration through a 100-μmol/L nylon mesh. After cell suspension preparation, cells were incubated with PE-conjugated murine CD45 antibody (BD Biosciences, clone 30-F11) for 20 minutes at 4°C. Cells were then washed twice with PBS+0.2% BSA and prepared for cell sorting. Before sorting, cells were costained with 0.1 μg/mL 4′,6-diamidino-2-phenylindole (DAPI, Merck). CD45⁺DAPI⁻ cells were sorted using FACS Fusion sorter (BD Biosciences). Post-sorting cell purity was assessed by FACS analysis (Supplementary Figs. S2A and S2B) and purity was assessed to be higher than 90%. Cells were then counted and 5,000 cells/conditions underwent scRNA-seq library preparation following 10X Genomic V2 and V3 protocol.

Mice tumors were surgically removed and processed to obtain cell suspension as explained in the previous paragraph. Gating strategies are summarized in Supplementary Figs. S3A and S3B. Briefly, viable cells (negative for 7-aminoactinomycin, 7AAD; Beckman Coulter) were stained with antibodies cocktail mix containing anti-CD45, anti-CD3e, anti-CD8, anti-CD4 to identify CD3⁺CD8⁺ and CD3⁺CD4⁺T cells. Than cells were stained with anti-CD279 (anti-PD-1) and anti-Tim3 antibodies. Percentage was referred to alive lymphocytes.

After demultiplexing, reads were aligned with Cell Ranger v3.0.2 to the mm10 reference genome using the GENCODE vM16 feature annotation (17). Transcript counts from genes with the same symbol but different ENSEMBL ID were summed up, resulting in a total of 53,278 unique genes.

We merged the raw transcript counts from all six treatments for each cell line using Seurat v3 (18, 19). Then, cells with 500 transcripts or less, 50,000 transcripts or more, fewer than 250 expressed genes, fewer than 0.80 log10 genes per UMI (Supplementary Figs. S4A and S4B), or more than 15% mitochondrial reads were removed from the analysis (Supplementary Figs. S5A and S5B; Supplementary Tables S1 and S2). Similarly, genes expressed in fewer than 5 cells were removed from the analysis, resulting in 22,806 and 23,757 genes on 4T1 and EMT6 cell line, respectively (Supplementary Figs. S6A and S6B). Downstream analyses, if not specifically mentioned otherwise, were carried out using Seurat v3.

Transcript counts were log-normalized using the “NormalizeData()” function. Canonical correlation analysis (CCA) was used to integrate data generated with the v2 chemistry and the v3 chemistry of the 10× Genomics Single Cell 3′, respectively. First, raw transcript counts were normalized with the “SCTransform()” function separately for each data set of the two chemistry version. Then, the two data sets were integrated using the “SelectIntegrationFeatures()” (to select 3000 features for integration in each data set), “PrepSCTIntegration(),” “FindIntegrationAnchors(),” and “IntegrateData().” Subsequently, principal component analysis (PCA) was performed on the integrated data set. The first 30 principal components were used to construct a Shared Nearest Neighbor (SNN) graph, from which clusters were calculated through the “FindClusters()” function and the resolution parameter set to “0.8.” The Uniform Manifold Approximation and Projection (UMAP) method was used for dimensionality reduction and calculated from the first 30 principal components. Cell-cycle analysis was performed using the "CellCycleScoring()” function and the S- and G2M-specific genes downloaded from ensemldb package (Supplementary Figs. S7A and S7B; ref. 20).

Cell type annotation was carried out with the SingleR R package (21) and the “ImmGenData” reference (22), containing 253 fine labels generated from 830 microarray samples of sorted cell populations. We performed an initial assignment to the finest level possible using the fine labels and then manually grouped together the cell populations to reach a reasonable granularity that helps exploration of the dataset. The conversion of the manually assigned groups of cells can be found in Supplementary Table S2. At both levels, cells belonging to labels with less than 20 cells were assigned to the group “Others” (Supplementary Table S2). Chi-square statistics using as expected the T-cell population frequencies of the control was performed, and its strength was tested with the Cramer V test. Chisq.test() and an in-house R functions were used to calculate the two tests respectively (Supplementary Table S3). We measured the fold change of the T-cell populations in each treatment as result of the expression: F_i/F_c, in which F_i is the frequencies of the T-cell population in the specific treatment while the F_c is the frequencies in the control. Correlations between the T-cell fold changes were performed either with Spearman or Pearson tests according to their normality as in ref. 23. Only the significative comparisons were plotted.

Marker genes were identified through the “FindAllMarkers()” Seurat function with a Wilcoxon test. Parameters were set to only identify overexpressed genes that are expressed in at least 70% of the cells of the cell population of interest, for example, cluster, with a log fold-change of at least 0.25 compared with all other cells. Furthermore, only genes that passed statistical testing with a P value of 0.01 or lower were retained. Pathways enriched in the marker genes were retrieved using the “getEnrichedPathways()” function from the cerebroApp R package (24), which queries the Enrichr database (25, 26). Results were filtered for pathways with an adjusted P value of 0.05 or lower.

Expression values of single genes shown in violin plots represent the log-normalized counts from “RNA” assay. To represent the expression of gene sets, we either calculated a module score with the “AddModuleScore()” function, or computed the average expression with the “AverageExpression()” function applied to the “SCT” assay.

To perform the gene set variation analysis (GSVA), we downloaded the 50 hallmark gene sets (27) using “msigdbr” package (28). From those, we selected 26 pathways of interest with a specific interest for the proliferation and immune cells activation groups. Then, we removed duplicated genes within and between pathways following the procedure by Lambrechts and colleagues (29). Finally, for a subset of the data set containing only T cells, we estimated the GSVA score on the raw transcript counts, fit a linear model and estimated the moderate t statistic in default mode for each combination of treatments.

At least 100,000 cells per sample were acquired using a 3-laser, 10-color flow cytometer (Navios; Beckman Coulter). Viable cells (negative for 7-aminoactinomycin, 7AAD; Beckman Coulter) were labelled with a panel of antibodies to analyze immune cell populations. Antibodies list and gating strategies are summarized in the Supplementary Table S4. Lymphocytes and myeloid cells were characterized using state-of-the-art markers. Specifically, cells were gated for size, singlets, and then by positive and negative markers: CD3⁺CD4⁺ and CD3⁺CD8⁺ T cells, CD335⁺ NKs, CD19⁺ B cells, Gr1-CD11b⁺CD11c⁺ monocytes. SSC^highCD11b⁺Gr1⁺ granulocytes and CD11c⁺CD11b⁻Gr1⁻ antigen-presenting cells (APC). APC were also marked with CD80 to determine their activation. For each immune cell subtype, we calculated cell absolute value by multiplying the percentage to the total number of white blood cells/μL.

T-cell receptor (TCR) clonality was analyzed in the tumor infiltrate using FITC Hamster Anti-Mouse Beta TCR Kit (BD Biosciences, clone H57-597), following manufacturer's protocol. Briefly, tumor was dissociated as shown in the previous paragraph and 5 × 10⁵ cells were stained with antibody cocktail mix containing: 7AAD (Beckman Coulter), CD45 APC-Cy7 (BD Biosciences, clone 30 F-11), CD3 APC (BD Biosciences, clone 145-2C11), CD4 PE (BD Biosciences, clone GK1.5), CD8 PE-Cy7(BD Biosciences, clone 53-6.7), and a specific region of VDJ beta chain rearrangment. Viable cells were gated for CD45/CD3 positivity. Then, cells were gated for CD4 and CD8 positivity, and among them we evaluated for each β chain rearrangement the percentage of positivity. Finally, fold increase was determined by dividing the percentage of positivity in the untreated control versus each experimental condition.

Data were expressed as means ± SEM (in case of normal distribution). Normal distribution was assessed using Shapiro Wilk's normality test. To compare two sample groups, either the Student t test or the Mann–Whitney U test was used on the basis of normal or not normal distribution. Statistical analysis was carried out Prism 8.3.1. (GraphPad) and R (URL http://www.R-project.org; ref. 30). Graphical model was created using Biorender Software (https://biorender.com/).

Single-cell transcriptome data of tumor samples and codes used for their analysis are available from the corresponding author upon request.

We injected five mice per experimental group with 4T1 and EMT6 cells and treated them with chemotherapeutic agents, CI and chemotherapeutic agents combined to CI. Chemotherapeutic agents were administered every 6 days, whereas anti-PD-1 every 2 days for a total of five injections, as shown in the therapeutic scheme presented in Supplementary Fig. S8A. We measured tumor growth in both models, and as shown in Fig. 1A and B and Supplementary Figs. S8B and S8C. In vivo studies in local and metastatic models using both 4T1 and EMT6 TNBC cells clearly indicated that intermittent (i.e., every 6 days), C140 was more effective than any other combinatorial regimens including CIs plus daily C20, or T, D, or P at their regular preclinical dosages.

For the EMT6 model, the triple therapy abrogated tumor growth in 28 days (P < 0.05), whereas for 4T1-bearing mice, C140 alone abrogated tumor growth up to day 28. Therefore, for 4T1 model, we prolonged the time schedule to 70 days and we observed that the association with V and CIs (anti-PD-1) further increased the preclinical efficacy of C140 (P < 0.001). Tumor resection was performed for 4T1 tumor on day 28 for control, anti-PD-1, D+ anti-PD-1, T+ anti-PD-1, and P+ anti-PD-1, and on day 70 for all C-treated groups; whereas for EMT-6 tumor, resection was performed at day 28 for all treatment groups. After tumor resection, we observed that the triple therapy abrogated metastatic tumor dissemination in the lungs in the 4T1 model (P < 0.001; Fig. 1C). In the EMT6 model, we observed a reduction in metastases with the triple therapy compared with controls, even though this mouse model is known to generate less metastases (Fig. 1C and D; Supplementary Figs. S8D and S8E; ref. 31).

To further demonstrate that immunity plays a crucial role in tumor eradication, we performed preclinical studies with neutralizing monoclonal antibodies targeting in separate experiments T, B, NK, and myeloid cells as shown in the therapeutic scheme presented in Supplementary Fig. S9A. Neutralizing antibody efficacy was tested by FACS as shown in the Supplementary Figs. S9B to S9D. In TNBC-bearing mice treated with C140, V, and CIs, the combinatorial therapy demonstrated that CD3⁺CD8⁺ (in 4T1-bearing mice), and CD3⁺CD8⁺ plus CD3⁺CD4⁺ T cells (in ETM6-bearing mice) were needed to abrogate TNBC growth (P < 0.05; Fig. 1E and F).

To a much lesser extent, in ETM6-bearing mice also CD11b⁺ myeloid cells (P < 0.01), CD19⁺ B cells (P < 0.01), and CD122⁺ NK cells (P < 0.01) had a role in reducing TNBC growth.

As we previously observed (5), during therapy the circulating landscape and the intratumoral contexture of immune cells were markedly different. Therefore, we decided to analyze the immune cell landscape in the peripheral blood by flow cytometry, as shown in Supplementary Figs. S10A and S10B, and within the tumor by single-cell RNA seq analysis (scRNA-seq).

In the peripheral blood, C140-containing regimens significantly reduced the number of circulating lymphoid and myeloid cells in both mouse models (Supplementary Figs. S10A and S10B). As also reported previously (5), flow cytometry and single-cell transcriptome analyses confirmed that V-containing regimens were associated with significant APC activation, measured by CD80 expression (Fig. 2A and B; Supplementary Fig. S11). Moreover, flow cytometric immunophenotypic assessment of TCR-V(β) repertoire within tumors suggested a significantly larger circulating CD3⁺CD4⁺ and CD3⁺CD8⁺ T cells clonal expansion in mice treated with C140 when compared with mice treated with other drugs (Fig. 2C,–F).

Using scRNA-seq analyses, we observed that 4T1 and ETM6 TNBC tumors were significantly different in the landscape of intratumoral immune cells: 4T1 tumors were more infiltrated by lymphoid cells, whereas ETM6 tumors by myeloid cells (Fig. 3A and B; Supplementary Table S5). In both models, single-cell transcriptome analyses of more than 50,000 intratumoral immune cells indicated that C140, V, and CI treatment was associated with the largest T-cell intratumoral population, with a predominant presence of CD3⁺CD4⁺ T helper cells (T_H1 cells), known to promote antitumor immunity (Fig. 3A and B; ref. 32). To assess cell diversity, chi-square statistics and Cramer V test were applied for both the cell lines using as expected values the T cells frequencies of the control. The number of cells resulted significative different from the control for each treatment in both the cell lines (Supplementary Table S5). Moreover, we measured the increase and the fold change of the T-cell populations for each treatment compared with the control. We identified a strong correlation between the increase of T cells for PD1, Cisplatin + PD1 and triple treatments between the two lineages (Supplementary Figs. S12A–S12F).

As intratumoral T cells were more abundant in mice treated with C140, V, and CIs, we tested the hypothesis of a phenotypic change in these T cells and performed an analysis of hallmark pathway gene signatures as in ref. 28. We compared the pathways observed in control and after C140, V, and CI treatment and observed that this treatment increased the expression of pathways involving proliferation (i.e., myc target, G₂–M checkpoint) and T-cell activation (i.e., IFNα and γ response, Fig. 4A and B). Our data strongly suggest that C140, V, and CI induce T-cell proliferation and activation.

Moreover, we studied the expression of T-cell different activating genes (Cd27, Cd28, Cd69, Cd2, and IFNγ among others, see Fig. 4C and D), and confirmed that after C140, V, and CI these genes were more expressed in intratumoral T cells, strongly suggesting that the therapy may increase T-cell activation against the tumor.

Finally, because we have noticed that T-cell activation pathway seems to be upregulated by triple therapy, we investigated which genes are mostly expressed. For both models, we ended up with a list of 20 genes comprising genes such as Cd3d, Ptprc, Cd69, Lat, Lck, Cd2, and Cd3e among others, suggestive of a change in morphology and behavior resulting from exposure to a mitogen, cytokine, chemokine, cellular ligand, or an antigen for which it is specific, as well as APC-to-T-cell adhesion (Fig. 5A and B).

To further corroborate this hypothesis, we observed which genes are mostly modulated by the therapy. We observed that for 4T1 model 7 genes are differentially expressed (Fig. 5C), whereas for EMT6 model 11 genes (Fig. 5D). EnrichR was then used to test for enrichment of these genes for specific signaling pathways, based on annotations from the KEGG database, demonstrating that they modulated pathways involved in antigen procession and presentation for both models (P < 0.05; Fig. 5E and F).

We asked what was PD-1 contribution to the treatment. It is widely known that anti-PD-1 favors the selection of progenitors exhausted T cells, which may have beneficial antitumoral effects in patients with melanoma (33). Progenitors exhausted T cells are defined to be Tim3⁻ (gene name Havcr2), whereas terminally exhausted T cells, which have poor anticancer function, are defined as Tim3⁺. Therefore, we wondered whether it was the case in our model systems. We checked first by scRNAseq whether this treatment increased the frequency of intratumoral Tcf1⁺ stem-like CD8⁺ T cells (P < 0.001; Fig. 6A). Synergy between C140/V and CI might be due, at least in part, to CI selective activity on progenitor exhausted CD8⁺ tumor-infiltrating lymphocyte (TIL), which can respond to anti-PD-1 and were significantly increased in mice treated with chemotherapy and CI when compared with C140 and C140/V treatments (P < 0.001; Fig. 6B for both 4T1 and EMT6 models). To further corroborate the scRNAseq data, we performed ex vivo validation of our data by FACS analysis. Briefly, after tumor injection and treatments, tumors were digested and co-expression of PD-1 and Tim-3 was assessed in CD8⁺ T cells. Gating strategies are shown in Supplementary Figs. S3A and S3B. Upon chemotherapy and anti-PD1 treatments, co-expression of PD-1 and Tim3 decreased only in chemotherapy-treated mice. This decrease was more evident when anti-PD1 was present in synergy with both C and V (P < 0.05 for EMT6 and P < 0.01 for 4T1 model; Fig. 6C).

Consistent with this observation, Tim3⁺ T cells, previously designated as terminally exhausted cells, were found predominantly in control and chemo-alone groups, whereas the addition of CI to chemotherapy significantly decreased these cells. To further corroborate this hypothesis, we also validated the presence of Entpd1, another marker on terminally exhausted T cell. Consistent with Tim3, also Entpd1 is less expressed in triple treatment, suggesting a selection towards progenitors exhausted T cells in presence of anti-PD-1 (P < 0.001; Fig. 7).

TNBC is among the most aggressive and lethal types of breast cancer, and currently available therapies have an unsatisfactory impact on patients' survival. The association of CIs with chemotherapy has shown some encouraging results in randomized clinical trials enrolling patients with TNBC, but not all the studies have met their end points and there has been no clear evidence of what should be considered the best chemotherapy backbone to be associated with Cis (6–10). We designed this study to define, at the preclinical level, what could be considered the most promising combinatorial regimen of chemotherapy plus CI. We investigated different options including P, D, T, V, and C, that is the most effective chemotherapy drugs for this disease. Capecitabine (or 5-FU) was not considered as our previous data indicated that C and V were more effective when used in combination with Cis (5).

C is known to elicit potent effects over the immune system including Treg reduction, as well as NK, MDSCs, and DC expansion (reviewed in ref. 34). In fact, the clinical use of C after partially-mismatched hematopoietic stem cell transplant has demonstrated that this drug effectively re-orchestrate also T cells that are pivotal in anticancer immunity (3). The use of C as an immunomodulatory drug has been reported both in preclinical models and in clinical trials (11, 35, 36). In addition to effects on the rewiring of T cells, innate immune cell effector functions are enhanced by C in pathways also involving CD11b⁺ cells as in the ETM6 model presented here. Present data clearly indicated that intermittent, medium dosage C (C140), in combination with V, was more effective of all other tested combinations. Interestingly, this triple therapy was similarly effective in two models of TNBC that seem to elicit different immunologic responses, as in 4T1 tumors the immune cell infiltrate was mainly lymphoid, whereas in ETM6 tumors the infiltrate included in a larger part myeloid (as well as B) cells. In this context, it is noteworthy that myeloid cells, and in particular macrophages, are known to mediate tumor resistance to CIs (37, 38). The observed decrease in MHC-II is likely due to the decrease in intratumoral myeloid cells.

To define the mechanisms crucial for C140, V, and CIs preclinical activity, we first identified what immune cells populations were needed to control tumor progression. In ETM6-bearing mice with a predominant myeloid and B-cell infiltrate (but not in 4T1-bearing mice with a predominant lymphoid infiltrate), neutralization by mAbs of CD11b⁺ myeloid cells, CD19⁺ B cells, and/or CD122⁺ NK cells slightly reduced the preclinical efficacy of the C140, V, and CI therapy, thus suggesting that these immune cell populations might participate in TNBC control by this triple therapy when the immune cell infiltrate had a relevant myeloid (and B) cell landscape. However, neutralization studies with mAbs against T cells clearly indicated a central, more relevant role for these cells in the preclinical activity of the C140, V, and CI therapy. Neutralization of CD3⁺CD8⁺ T cells completely abrogated the efficacy of this therapy in 4T1-bearing mice, and neutralization of CD3⁺CD4⁺ or CD3⁺CD8⁺ T cells completely abrogated the efficacy in ETM6-bearing mice. These partially diverging results underline the need for investigating different in vivo models including infiltrates with myeloid versus lymphoid prevalence.

The central role of T cells in the preclinical activity of C140, V, and CI therapy prompted us to investigate the impact of these drugs over different populations of T cells and over their interactions with other players. To this aim, single-cell RNA sequencing provided a robust tool to investigate how seemingly homogenous cell populations may differ in their functional properties when treated with a given drug, and how diverse cell atlases mediate the drug activity (39). Activation of APCs, their processing of tumor antigens, and APC interaction with T cells is mandatory for an efficient control of tumor growth by T cells. In fact, several preclinical studies have demonstrated that APCs are major players in the anticancer properties of Cis (40–43). As we and others (5, 44) have previously shown that V is a potent activator of APCs, we believe that this drug should always be considered and investigated in chemotherapy regimens associated with CIs. The exact mechanism of APC activation by V and the role of B cells are currently investigated in-depth in other companion studies in our laboratory.

Single-cell transcriptome analysis of more than 50,000 intratumoural immune cells indicated that C140, V, and CIs induced a unique gene signature suggestive of a behavior associated with exposure to a mitogen, a ligand, or an antigen, as well as APC-to-T-cell adhesion. Further analyses in T cells indicated that this triple therapy increased the frequency of intratumoural Tcf1⁺ stem-like CD8⁺ T cells (45, 46). Recent evidence suggested that a successful CI therapy should generate new T-cell clones rather than reinvigorate only preexisting TILs (47–49). In this context, it is noteworthy that C140, when compared with P, T, and D, was the drug more efficient in generating new circulating T cells clones (see Graphical Abstract).

We performed a differential gene expression analysis to identify which genes are consistently up- and downregulated after the triple treatment in T cells. We identified 131 significantly upregulated and 50 significantly downregulated genes (Supplementary Table S6). Many of the upregulated genes are predicted/pseudogenes or ribosomal genes, and pathway enrichment analysis for these genes did not return any significant results. The pathway enrichment analysis for the downregulated genes offered more interesting results, which deserve to be further investigated and validated in future studies.

T-cell exhaustion develops in a progressive manner under conditions of prolonged antigen stimulation, leading to a heterogeneous population of exhausted T cells. Recent studies (33, 50, 51) reported that exhausted T cells exhibit phenotypic and functional heterogeneity, which permits to these cells to be separated in progenitor exhausted and terminally exhausted T populations. The transcription factor T-cell factor 1 (Tcf1), encoded by tcf1, is a key regulator of progenitor exhausted T cells. In contrast, terminally exhausted cells express high levels of Tim-3/ENTPD1, and other co-inhibitory receptors (51). Progenitor exhausted T cells (Tcf1⁺Tim3⁻ENTPD1⁻) mediate long-term tumor control and possess stem-like characteristics allowing them to undergo self-renewal and exclusively respond to CI therapy. Terminally exhausted T subsets (Tcf1⁻Tim3⁺ENTPD1⁺) are differentiated from progenitor cells by higher cytotoxicity, but they have reduced long-term survival and are unable to respond to CI.

Our data suggest that the synergy of CIs to chemotherapy is due in part to its selective activity on progenitor exhausted CD8⁺ TILs, which can respond to CIs and are significantly increased in animals treated with chemo+immunotherapy compared with C140 and C140 plus V treatments alone (Figs. 6A,–C and 7 for both 4T1 and EMT6). Consistent with this, Tim-3⁺ T cells, previously designated as terminally exhausted cells, were found predominantly in control and chemo-alone groups, whereas the addition of CIs to chemo leads to a significantly decrease of these cells. The triple combination therapy may alter the balance between the terminally and progenitor exhausted T cells by favoring the latter and improving preclinical efficacy.

Taken together, our observations indicated that the C140, V, and CI therapy had a unique activity over the circulating and intratumoral immune cell landscape. These drugs showed a strong preclinical activity in two TNBC models, one associated with a predominant lymphoid intratumoral infiltrated and the other associated with a predominant myeloid/B-cell infiltrate. As the dosages of this therapy can be transferred into the clinic with a schedule that seems feasible (11), these findings can be used to design novel clinical trials in this neoplastic disease.

R. Hillje reports grants from AIRC (Fondazione AIRC per la Ricerca sul Cancro) during the conduct of the study. No disclosures were reported by the other authors.

P. Falvo: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, writing—original draft, project administration, writing—review and editing. S. Orecchioni: Conceptualization, data curation, formal analysis, validation, investigation, writing—original draft, writing—review and editing. R. Hillje: Conceptualization, data curation, software, formal analysis, validation, investigation, writing—original draft, writing—review and editing. A. Raveane: Software, formal analysis, writing—original draft, writing—review and editing. P. Mancuso: Software, formal analysis, validation, writing—original draft, writing—review and editing. C. Camisaschi: Formal analysis, validation, writing—original draft. L. Luzi: Software, formal analysis, writing—original draft. P. Pelicci: Software, formal analysis, funding acquisition, validation, writing—original draft. F. Bertolini: Conceptualization, resources, supervision, funding acquisition, validation, writing-original draft, project administration, writing—review and editing.

This work was supported by the Italian Association for Cancer Research (AIRC) and the Italian Ministry of Health. F. Bertolini is a scholar of the US National Blood Foundation. R. Hillje is a PhD student at the European School of Molecular Medicine (SEMM). Authors would like to thank Stefania Roma and Giulia Mitola, and the IEO analysis laboratory, in particular, Marco Picozzi, Giulia Ricci, and Giulia Clericuzio for hematologic analysis, IEO imaging facility, in particular Simona Ronzoni and Matteo Gallazzi for sorting experiments, IEO pharmacy, in particular Dr. Emanuela Omodeo Salè and Martina Milani for giving us the chemotherapeutic drugs, IEO Genomic Unit, in particular Luca Rotta, Salvatore Bianchi, Thelma Capra, and Lucia Massimiliano for scRNA seq libraries preparation, IEO cell culture, in particular Cristina Spinelli, Manuela Moia, and Donatella Genovese for cells technical supports. Finally, we are thankful to Federica Pisati for the hematoxylin and eosin lung slides preparation and Molpath unit, in particular Giovanni Bertolot, Giovanna Jodice, and Francesca Mantovani for ScanScope XT scanner acquisition.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.