Tumor sialylation regulates G-CSF stability and promotes neutrophil-mediated immunosuppression in breast cancer

Abstract

Breast cancer is the most prevalent cancer among women, yet immune checkpoint inhibitors remain largely ineffective in the majority of patients. Here we show that tumor sialylation is negatively associated with T cell infiltration across molecular breast cancer subtypes. In preclinical models, both genetic and pharmacologic inhibition of tumor cell sialylation reshape the tumor microenvironment by reducing immunosuppressive neutrophil infiltration, while CD8⁺ effector and Tcf7⁺ memory T cells are increased within the mammary tumors. Mechanistically, sialylation enhances the serum half-life of granulocyte colony-stimulating factor (G-CSF), a key driver of neutrophilic immunosuppression, and dampens tumor cell immunogenicity by limiting MHC-I surface expression. Disruption of sialylation sensitizes resistant mammary tumors to CD8⁺ T cell-mediated killing and anti-PD-1 therapy in multiple breast cancer models. These findings establish tumor sialylation as a key mechanism of immune evasion and a potential therapeutic target in breast cancer.

Tumor sialylation has been associated with different mechanisms of immune evasion. Here the authors show that tumor sialylation regulates G-CSF stability to promote neutrophil-mediated immunosuppression as well dampens tumor cell immunogenicity by limiting MHC-I surface expression, limiting anti-tumor immunity in breast cancer.

Introduction

Immune checkpoint inhibitors (ICIs) have revolutionized cancer therapy by harnessing the host immune system to target malignant cells. In particular, therapies targeting the PD-1/PD-L1 axis have shown remarkable efficacy in several tumor types, including melanoma and non-small cell lung cancer¹. However, in breast cancer, clinical responses to ICIs remain modest and largely limited to triple-negative subtypes, highlighting a pressing need to identify mechanisms of immune evasion and novel immunotherapeutic targets^2–4. Current efforts are ongoing to establish neoadjuvant ICI therapy for luminal B breast cancer⁵. Notably, even in advanced triple-negative breast cancer, where patients are stratified by high PD-L1 expression, only a subset derives a survival benefit from anti-PD-1 therapy⁶. This points toward alternative immunoregulatory mechanisms that may contribute to immune escape in breast cancer and could be therapeutically exploited.

Glycan modifications are increasingly recognized as critical players in shaping the immune landscape of tumors^7–9. Among these, sialylation, the enzymatic addition of sialic acid residues to glycoproteins, glycolipids, or glyco-RNAs, has emerged as a potent modulator of immune responses⁷. Sialic acids are negatively charged nine-carbon sugars that cap glycan structures and mediate diverse biological processes, including immune cell trafficking, receptor masking, and modulation of immune cell signaling¹⁰. The biosynthesis of sialylated glycans begins with the conversion of free sialic acid to CMP-sialic acid via Cytidine Monophosphate N-Acetylneuraminic Acid Synthetase (CMAS), which is subsequently transported into the Golgi apparatus and utilized by sialyltransferases to modify glycoconjugates^11–13. Aberrant sialylation is a hallmark of many cancers, where it has been implicated in tumor progression, metastasis, and immune evasion^14,15. Despite its emerging significance, the mechanistic role of tumor-intrinsic sialylation in shaping the immunological landscape of breast cancer remains poorly defined.

In this study, we investigate the functional role of tumor cell sialylation in immune evasion and ICI resistance in breast cancer. Using genetic and pharmacological perturbation approaches in murine models, breast cancer cell lines, and clinical specimens, we demonstrate that sialylation fosters immune escape by impairing antigen presentation and prolonging the serum half-life of G-CSF. These effects collectively promote a neutrophil-rich, immunosuppressive tumor microenvironment (TME) while limiting cytotoxic T cell infiltration. Disruption of sialylation restores tumor immunogenicity, reprograms the TME, and sensitizes otherwise resistant tumors to PD-1 blockade. Our findings position sialylation as a key immunoregulatory node and a promising therapeutic vulnerability in breast cancer.

Results

Tumor sialylation is associated with reduced T cell infiltration in human breast cancer

To assess sialylation in human breast cancer, we first optimized histochemical staining protocols using the lectins Sambucus nigra lectin (SNA; binds to α2,6-linked sialic acid) and Maackia amurensis lectin 2 (MAL2; binds to α2,3-linked sialic acid) in human breast sections (Fig. 1A, and Supplementary Fig. 1A). Staining specificity was confirmed using neuraminidase-pretreated control sections, which abolished SNA and MAL2 reactivity upon enzymatic de-sialylation (Supplementary Fig. 1B). We then assessed tumor sialylation patterns in two independent clinical cohorts. The first was a Swiss cohort consisting of 50 treatment-naïve patients diagnosed with primary breast cancer, for whom pre-treatment biopsies were available (Supplementary Data 1). In this cohort, 27% and 57% of cases showed an increased SNA or MAL2 positivity, respectively, compared to healthy control breast specimens. When SNA and MAL2 staining data were combined to derive a sialylation score, no significant differences in the sialylation scores were observed among the four molecular breast cancer cell subtypes (Supplementary Fig. 1C). The second cohort was a French cohort consisting of 86 patients with invasive ductal adenocarcinomas, predominantly luminal subtypes, in which 33% and 38% of cases showed an increased SNA or MAL2 positivity, respectively (Supplementary Data 1).

Fig. 1. Breast cancer sialylation determines tumor infiltrating lymphocytes and CD8⁺ T cell mediated tumor control.

A Representative images of Maackia amurensis lectin 2 (MAL2; marks α2,3 sialic acid) and Sambucus nigra lectin (SNA; marks α2,6 sialic acid) staining in healthy human mammary gland epithelium and breast cancers from patient biopsies. Additional information about the representative images is provided in Supplementary Fig. 1. Scale bars: 50 μm. B, C Comparative analysis between sialylation score and CD8 positive area or tumor infiltrating lymphocytes (TILs) in two independent cohorts. In the Swiss cohort (n = 50, B) CD8 positive area was assessed by immunohistochemistry. In the French cohort (n = 86, C) TILs were assessed by histopathologic scoring of hematoxylin and eosin-stained whole tumor sections. Groups were divided in sialylation score of 2 or less (B, n = 22; C, n = 45) and scores above 2 (B, n = 28; C, n = 41). All values are derived from biological replicates. Two-sided Mann–Whitney test. ^*p = 0.0115, ^**p = 0.0088. D. Tumor growth curves of control and Cmas KO 4T1 mammary tumors, orthotopically implanted into immunocompetent syngeneic BALB/c mice. E–G Quantification of immunohistochemical assessment of cell proliferation (Ki67; Ctrl: n = 5 mice; Cmas KO: n = 5 mice), and CD8 (Ctrl: n = 11 mice; Cmas KO: n = 10 mice; ^**p = 0.0017) as well as PD-1 (Ctrl: n = 10 mice; Cmas KO: n = 9 mice; ^**p = 0.0024) expressing tumor expressing TILs in control and Cmas KO 4T1 mammary tumors, determined by immunostaining. H Tumor growth curves of orthotopic control and Cmas KO 4T1 tumors in immunocompromised Rag2^−/− γc^−/− BALB/c mice. I Tumor growth curves of control and Cmas KO 4T1 tumors orthotopically implanted into syngeneic mice depleted for CD4⁺ and/or CD8⁺ T cells. n = 6 mice per group. Tumor growth curves are shown as mean ± SEM and statistical differences were assessed by two-way ANOVA. Tumor volumes were determined using digital caliper. Histological quantifications are shown as mean ± SEM and statistical difference were assessed by two-sided t-test. ^****p < 0.0001. Source data are provided as a Source Data file.

We next evaluated T cell infiltration in both patient cohorts. In the Swiss cohort, this was done using immunostaining for CD3, CD8, and FOXP3 (Supplementary Fig. 1D–G, and Supplementary Data 1), while the French cohort was assessed using standardized TIL scoring according to Salgado et al.¹⁶ (Supplementary Data 1). Notably, when comparing tumors with sialylation scores greater than those of control healthy mammary glands (>2) to those with normal or reduced sialylation (≤2), we observed a significant inverse association between tumor sialylation and the presence of CD3⁺, CD8⁺, and FOXP3⁺ T cells as well as with TILs (Fig. 1B, C and Supplementary Fig. 1H, I). These findings from two independent patient cohorts demonstrate that elevated tumor sialylation is associated with reduced T cell infiltration in breast cancer.

Sialylation deficient mammary tumors are controlled by CD8⁺ T cells

To genetically address the role of sialylation in breast cancer, we knocked out Cmas in the 4T1 cell line, a widely used syngeneic murine model for aggressive breast cancer, thereby abolishing cellular sialylation. We confirmed the successful knockout of Cmas at the transcript and protein levels (Supplementary Fig. 2A, B, and Supplementary Data 2) and substantiated the abrogation of sialylation by SNA staining and glycomic analysis (Supplementary Fig. 2C–E).

In vitro, Cmas KO 4T1 cells exhibited comparable growth rates (Supplementary Fig. 3A) relative to Cmas-expressing control 4T1 cells. However, Cmas KO cells orthotopically implanted into the mammary fat pad of syngeneic BALB/c mice displayed a significant growth disadvantage compared to controls (Fig. 1D, and Supplementary Fig. 3B). While tumor cell proliferation remained similar, as indicated by Ki67 staining, Cmas KO tumors exhibited a marked decrease in exhausted PD-1⁺ T cells and increase in infiltrating CD8⁺ T cells (Fig. 1E–G), recapitulating the negative correlation between sialylation and T cell infiltration observed in patient samples (Fig. 1B). The overall number of Arg1⁺ and iNOS⁺ expressing cells remained unaltered, indicating no major change in pro- and anti-inflammatory macrophages (Supplementary Fig. 3C–E). Of note, Cmas KO 4T1 tumor cells maintained their abrogated sialylation phenotype in vivo, as determined at the end of the experiment (Supplementary Fig. 3F).

To investigate whether the immune system drives the observed impairment of tumor growth, we orthotopically implanted Cmas KO and control 4T1 cells into the mammary fat pad of immunodeficient Rag2^−/−γc^−/− mice, which lack B cells, T cells, and NK cells¹⁷. In these mice, no differences in tumor growth were observed between Cmas-expressing and Cmas-KO 4T1 cells (Fig. 1H, and Supplementary Fig. 3G, H), confirming that the impaired growth of sialylation-deficient tumors is immune-mediated. To further dissect the role of T cells, we selectively depleted CD4⁺, CD8⁺, or both T cell subsets in tumor-bearing mice (Supplementary Fig. 4A, B). The depletion of CD8⁺ T cells, but not the depletion of CD4⁺ cells, restored the growth of Cmas KO tumors to control levels; depletion of both CD8⁺ and CD4⁺ cells phenocopied the depletion of CD8+ T cells (Fig. 1I). This finding establishes CD8⁺ cytotoxic T cells as the primary effector cells responsible for improved tumor control in sialylation-deficient 4T1 tumors.

Loss of tumor sialylation promotes TCF1⁺ memory T cell recruitment and sensitizes breast tumors to PD-1 blockade

CD8⁺ T cell-mediated tumor control is often limited by poor infiltration into the tumor and increased exhaustion (Fig. 1E, F), a phenomenon frequently driven by remodeling of the tumor microenvironment by components of the innate immune system^4,8. To comprehensively characterize the immune landscape and uncover mechanisms that may enhance tumor control in Cmas KO tumors, we performed comparative single-cell RNA sequencing (scRNASeq) of tumor-infiltrating immune cells from Cmas KO and control 4T1 tumors. To achieve this, we isolated CD45⁺ cells from the respective tumors, resulting in a population of more than 95% viable CD45⁺ cells (Supplementary Fig. 5A). Unsupervised clustering of scRNA-seq data identified 15 distinct immune cell clusters and one cluster of residual cancer cells (cluster 11) (Fig. 2A, and Supplementary Fig. 5B–E). We observed striking differences in the abundance of specific immune cell populations between control and Cmas 4T1 KO tumors, including a disproportionally high tumor infiltration of B cells, memory T cells, dendritic cells, and conversely a relative reduction in myeloid cells and polymorphonuclear myeloid-derived suppressor cells (PMN-MDSCs) in the Cmas-deficient tumors (Fig. 2B, and Supplementary Fig. 5F). Of note, myeloid cells (clusters 1, 4, 6, 7, and 10) in the Cmas KO tumors remained largely unaltered in their prototypic marker expression, including Arg1 and Nos2, when compared to the control tumors (Supplementary Fig. 5G).

Fig. 2. Cmas KO 4T1 mammary tumors are sensitive to anti-PD-1 therapy.

A Uniform manifold approximation and projection (UMAP) of combined single-cell transcriptomic data from tumor-infiltrating immune cells of control and Cmas KO 4T1 tumors, orthotopically implanted into immunocompetent syngeneic BALB/c mice. For each condition, five tumors were pooled into a single transcriptomic dataset. Clusters were annotated according to marker expression (see Supplementary Fig. 5). B Individual UMAPs of tumor infiltrating immune cells from control (top) and Cmas KO (bottom) 4T1 tumors. C Detailed UMAP of the T cell clusters 3, 8, 9 and 14 with marker expression comparisons between cells derived from control (grey) and Cmas KO (orange) 4T1 tumors. D Feature plot of Tcf7 expressing cells in T cell clusters 3, 8, 9 and 14 in control (left) and Cmas KO (right) 4T1 mammary tumors. Representative images (E) and quantification (F) of immunohistochemical assessment of anti-TCF1 immunostaining in Control and Cmas KO 4T1 mammary tumors. Images were taken from the invasive front. E Image enlargements of region of interest are shown to the right. Scale bars = 50 μm. F Each dot represents the average quantification of 3 areas, quantified with automated image analysis, within one tumor section (Ctrl: n = 4 mice; Cmas KO: n = 5 mice; mean ±  SEM). Two-sided t-test. ^*p = 0.0242. G, H Tumor growth of control and Cmas KO 4T1 tumors, orthotopically implanted into syngeneic BALB/c mice and treated with anti-PD-1 or isotype control antibodies. Tumor volumes were determined using a digital caliper (n = number of mice, mean ± SEM). Statistical differences were assessed by two-way ANOVA; ^**p = 0.0052; ^****p < 0.0001. Source data are provided as a Source Data file.

Concomitantly, the numbers of defined T cell populations were markedly changed (clusters 3, 8 and 9; Fig. 2C). In particular, Cmas KO tumors exhibited a remarkable increase of Tcf7⁺ Cd69⁺ memory T cells compared to control tumors (from 1.5% to 16.3%; cluster 3 in blue; Fig. 2C). Of note, this cluster included both Cd4⁺ and Cd8⁺ T cells (Fig. 2A, and Supplementary Fig. 5E). Effector T cells, on the other hand, exhibited less pronounced changes in cell numbers while exhibiting a marked shift in their expression profiles (clusters 8 and 9; Fig. 2C). Cd4⁺ effector T cells (cluster 8) in control tumors included T_H2 cells, activated/exhausted lymphocytes, and regulatory T cells characterized by Gata3, Pdcd1 and Foxp3 expression, respectively. However, these transcription or exhaustion hallmarks were reduced in the effector T cells of Cmas KO tumors, which in turn expressed increased levels of the T cell activation marker Cd69 (cluster 8; Fig. 2C). Similarly, Cd8⁺ effector T cells (cluster 9) from Cmas KO tumors showed reduced expression of Pdcd1 as well as increased expression of Cd44 and Cd69 (Fig. 2C). Importantly, among all these T cell clusters, Tcf7 (TCF1) expression was markedly increased in Cmas KO tumors (Fig. 2D). In line with these transcriptomic changes, TCF1⁺ T cells, together with B cells, were found to localize within clusters at the invasive front of Cmas KO tumors, but were sparse in control tumors (Fig. 2E, F, and Supplementary Fig. 6). These findings demonstrate that loss of sialylation reduces PMN-MDSC recruitment and promotes CD8⁺ T cell-mediated tumor control, while simultaneously facilitating the recruitment and expansion of TCF1⁺ memory T cells in the TME.

Augmented numbers of TCF1⁺ T cells have previously been shown to correlate with increased sensitivity to ICI therapies^18–20. The 4T1 tumor model is known to establish a strongly immunosuppressive microenvironment, which impairs CD8⁺ T cell function and renders anti-PD-1 therapy ineffective^21–23. Given the enhanced anti-tumor immune response and accumulation of Tcf7⁺ memory T cells upon sialylation loss, we evaluated whether Cmas KO tumors could be sensitized to PD-1 blockade. Remarkably, abrogation of tumor sialylation rendered previously resistant 4T1 tumors responsive to anti-PD-1 therapy, resulting in prolonged tumor control and significantly improved survival of tumor-bearing mice (Fig. 2G, H, and Supplementary Fig. 7A–C). These data show that ablation of tumor sialylation resulted in reduced PMN-MDSC infiltration and enhanced accumulation of Tcf7⁺ memory T cells, ultimately sensitizing resistant tumors to anti-PD-1 treatment.

Abrogation of sialylation suppresses PMN-MDSC recruitment and restores anti-tumor immunity

Interestingly, sub-clustering of PMN-MDSCs (cluster 2) revealed two transcriptionally distinct subpopulations, a Siglece⁺ subset (Cluster 2A) and a Ccl3⁺ Cd274⁺ subset (Cluster 2B) (Fig. 3A). Notably, the Ccl3⁺ Cd274⁺ subcluster (2B) was reduced by 23% in Cmas KO tumors compared to control tumors, while maintaining a similar expression profile for Ccl3 and Cd274 (PD-L1) (Fig. 3A, and Supplementary Fig. 8A, B). Given the low RNA content of PMN-MDSCs, reliable assessment via scRNA-seq is challenging²⁴. Therefore, we validated the significant reduction of PMN-MDSCs in Cmas KO tumors by flow cytometry (Fig. 3B): whereas the numbers of monocytic MDSCs (M-MDSCs) was not affected, PMN-MDSCs were indeed in average fivefold reduced within the TME (Fig. 3B).

Fig. 3. Reduction in polymorphonuclear myeloid-derived suppressor cells in Cmas KO tumor-bearing mice.

A UMAP (left) and marker expression profiles (right) of the sub-clusters 2A (green) and 2B (red) derived from cluster 2 (polymorphonuclear myeloid derived suppressor cells, PMN-MDSC). The UMAP is based on combined single-cell transcriptomic data from tumor-infiltrating immune cells of control and Cmas KO 4T1 tumors. Percentages indicate the relative reduction in cell number of respective sub-cluster from Cmas KO 4T1 tumors as compared to control 4T1 tumors. B Flow cytometry-based quantification of tumor infiltrating PMN-MDSCs and M-MDSCs in control (n = 8 mice) and Cmas KO (n = 10 mice) 4T1 tumors. C Blood neutrophil counts in control (n = 6) and Cmas KO (n = 6) 4T1 tumor-bearing mice. D From left to right, representative images of spleens, spleen to body weight ratios (Ctrl: n = 6; Cmas KO: n = 6) as well as flow cytometry-based quantification of splenic PMN-MDSCs (Ctrl: n = 4; Cmas KO: n = 4) and M-MDSCs (Ctrl: n = 4; Cmas KO: n = 4) of control and Cmas KO 4T1 tumor-bearing mice. E ELISA-based quantification of G-CSF concentration in serum (Ctrl: n = 7; Cmas KO: n = 7) and in tumor lysates (Ctrl: n = 5; Cmas KO: n = 5) in control and Cmas KO 4T1 tumor-bearing mice. F Growth curves of control and Cmas KO 4T1 tumors orthotopically injected into BALB/c mice treated with anti-Ly6G or isotype control antibodies. All groups: n = 7 mice, mean ± SEM. Tumor volumes were determined using a digital caliper. Statistical difference was assessed by two-way ANOVA (F). ^*p = 0.0139; ^**p = 0.0019. G Percentages of polymorphonuclear myeloid-derived suppressor cells (PMN-MDSCs) in tumors of Ly6G-depleted or undepleted mice carrying control and Cmas KO 4T1 tumors, assessed by flow cytometry. ^**p = 0.0010. Ctrl, undepleted: n = 6 mice; Ctrl, depleted: n = 6 mice; Cmas KO, undepleted: n = 7 mice. H Percentages of CD8⁺ T cells in tumors of Ly6G-depleted or undepleted mice with control and Cmas KO 4T1 tumors, assessed by flow cytometry. ^*p = 0.0477. Ctrl, undepeleted: n = 6 mice; Ctrl, depleted: n = 7 mice; Cmas KO, undepleted: n = 7 mice. Results are shown as mean ± SEM of biologic replicates. Statistical difference was assessed by two-sided t-test (B–E) or one-way ANOVA (G, H). ^****p < 0.0001. Grey bars in C–E represent the normal range, as established by aged and sex-matched healthy BALB/c mice. Source data are provided as a Source Data file.

To explore the mechanisms underlying these immunologic changes, we performed a serum cytokine analysis in control and Cmas KO tumor-bearing mice, revealing alterations in multiple immune factors, including CXCL13, M-CSF, TREM-1 and G-CSF (Supplementary Fig. 8C). Notably, previous studies have demonstrated that an increase in PMN-MDSCs may stem from increased neutrophil production driven by cancer cell-derived G-CSF^25,26. In agreement with this, we observed pronounced neutrophilia and splenomegaly in mice bearing sialylation-proficient tumors. In contrast, Cmas KO tumor-bearing mice displayed several-fold lower numbers of circulating and splenic neutrophils (Fig. 3C, D), concomitant with significantly reduced serum G-CSF concentrations (Fig. 3E). Interestingly, intratumoral G-CSF levels as well as G-CSF concentrations in cell culture supernatants remained unaltered indicating a change in serum half-life rather than differential expression levels (Fig. 3E, and Supplementary Fig. 8D). To evaluate the functional role of PMN-MDSC accumulation in the TME, we targeted Ly6G-expressing neutrophils using Ly6G-depleting antibodies (Supplementary Fig. 8E, F). Depletion of Ly6G⁺ cells resulted in the reduction of immunosuppressive PMN-MDSCs in the TME and to an increase in CD8⁺ T cell infiltration and a reduction in tumor growth (Fig. 3F–H), mirroring the phenotype observed in Cmas KO tumors.

To assess whether neutrophil extracellular trap formation (NETosis) contributes to the improved tumor control, we first analyzed the NETosis marker citrullinated histone H3 (CitH3) in control and Cmas KO tumors. Despite significant differences in neutrophil numbers, the number of CitH3⁺ cells within the non-necrotic tumor areas appeared comparable between the groups (Supplementary Fig. 8G). Prominent CitH3 staining was also detected in necrotic tumor cores, predominantly with an extracellular pattern consistent with NET deposition; again, the differences did not reach statistical significance (Supplementary Fig. 8H). To functionally address the role of NETosis, tumor-bearing mice were treated with the PAD4 inhibitor GSK484 to suppress NET formation²⁷. This in vivo intervention reduced CitH3 positivity in the necrotic core but did not significantly affect tumor growth in both control and Cmas KO tumors (Supplementary Fig. 8H, I), further suggesting that NETosis is unlikely to be a major determinant of the improved immunologic tumor control. Together, these data show that tumor sialylation promotes neutrophilia and the accumulation of immunosuppressive PMN-MDSCs in the TME, thereby contributing to tumor immune evasion and growth.

Sialylation regulates G-CSF stability and tumor immune evasion via PMN-MDSC expansion and MHC-I modulation

One of the candidates that could explain altered PMN-MDSC numbers is G-CSF, which promotes the mobilization of neutrophils from the bone marrow^28–31. Our data showed that whereas intratumoral levels of G-CSF remain unchanged, serum G-CSF was markedly decreased (Fig. 3E). Since asialo-glycoproteins are known to be rapidly cleared from circulation by the liver and kidneys^32,33, we hypothesized that the sialylation status of G-CSF might influence its serum half-life and thereby explain the observed protein levels. To test this, we injected equal amounts of recombinant G-CSF and enzymatically de-sialylated G-CSF into BALB/c mice (Supplementary Fig. 9A). Indeed, asialo-G-CSF exhibited a significantly reduced serum half-life (1.6 h) compared to its sialylated counterpart (2.4 h) (Fig. 4A). Consequently, 24 h post injection, asialo-G-CSF was nearly undetectable in circulation, whereas serum concentrations of native G-CSF remained around 33-fold above baseline (Fig. 4A). Importantly, in line with its shorter half-life, asialo-G-CSF induced significantly lower mobilization of neutrophils into the circulation (Fig. 4B, and Supplementary Fig. 9B).

Fig. 4. Sialylation controls G-CSF stability and MHC class I expression in Cmas KO 4T1 tumors.

A ELISA-based quantification of G-CSF concentration in serum of BALB/c mice 30 min, 1 h, 4 h, 24 or 48 h after the injection of recombinant G-CSF or enzymatic desialylated G-CSF (Asialo-G-CSF) (both groups and each timepoint: n = 2 mice). B Flow cytometry-based quantification of circulating neutrophils 24 h or 48 h post G-CSF injection or asialo-G-CSF injection (both groups and each timepoint: n = 3 mice). ^*p = 0.012662; ^**p = 0.002996. C ELISA-based quantification of G-CSF concentration in control, Csf3 KO or Csf3 Cmas double KO 4T1 tumor-bearing mice (each group: n = 3 mice; mean ± SEM). D Flow cytometry-based quantification of circulating neutrophils in control, Cmas KO, Csf3 KO or Csf3 Cmas double KO 4T1 tumor-bearing mice (each group: n = 5 mice). ^***p = 0.0006. E Spleen to body weight-ratios in control, Cmas KO, Csf3 KO or Csf3 Cmas double KO 4T1 tumor-bearing mice (each group: n = 5 mice). F Tumor growth curves of control, Cmas KO, Csf3 KO or Csf3 Cmas double KO 4T1 tumors orthotopically injected into BALB/c mice. n = 6 mice, mean ± SEM. Tumor volumes were determined using a digital caliper. Statistical difference was assessed by two-way ANOVA (F). ^***p = 0.0009; ^****p < 0.0001. G Schematic of mechanistic consequences of sialylation interference on G-CSF half-life, neutrophil mobilization, splenomegaly, and PMN-MDSC recruitment to the tumor microenvironment. Created in BioRender. Mereiter, S. (2026) https://BioRender.com/jaqjttr. H Changes in MHC class I glycopeptide abundance comparing control and Cmas knockout 4T1 (n = 3, biological replicates) mammary tumor cells. Results are shown as average fold change and p-values of the unique glycopeptides, with sialylated glycopeptides highlighted in red. Most significantly changed glycopeptides from the MHC-I glycoproteins H2-L, H2-Q6, H2-T3 or H2-K1 are labelled. Putative glycan structures (1-15) are shown. Data are detailed in Supplementary Data 3. I MHC class I surface expression of unstimulated and IFNγ stimulated control and Cmas KO 4T1 mammary cancer cells, assessed by flow cytometry detecting H-2Kd (n = 3, biological replicates, mean ± SD). ^**p = 0.001899. J. MHC class I (H-2Kd) surface expression of Ctrl and Cmas KO human BT-474 (n = 6) and MDA-MB-231 (n = 3) breast cancer cells, assessed by flow cytometry (biological replicates; mean ± SD). ^**p = 0.0016. K MHC class I (H-2Kd) surface expression on fixed control and Cmas KO 4T1 mammary cancer cells following enzymatic desialylation with neuraminidase (NA) or left untreated (n = 3; technical replicates). Exact p values are provided as a Source Data file. Statistical difference was assessed by two-sided t-test (B, I, J) or one-way ANOVA (D, E, K). ^****p < 0.0001. Grey areas in B, C and E represents the normal range, as established by age and sex-matched healthy BALB/c mice. Source data are provided as a Source Data file.

To further investigate the contribution of G-CSF, we generated Csf3 (coding for G-CSF) knockout tumors in both wild-type and Cmas KO 4T1 cells. Tumor-bearing mice with either Csf3 KO or Csf3 Cmas double KO tumors showed no significant elevation in serum G-CSF levels compared to healthy mice. This demonstrates that the >30-fold increase in circulating G-CSF observed in control tumor-bearing mice is a indeed a consequence of tumor cell-derived G-CSF (Fig. 4C). In Csf3-deficient tumor-bearing mice, both circulating neutrophil counts and splenomegaly were significantly reduced, accompanied by a marked reduction in tumor growth (Fig. 4D–F). Together, these data indicate that tumor-derived G-CSF drives neutrophil mobilization, resulting in splenomegaly and PMN-MDSC accumulation in the TME; interference with sialylation shortens the serum half-life of G-CSF, thereby limiting neutrophil mobilization (Fig. 4G). Notably, Csf3 KO and Csf3 Cmas dKO tumors initially exhibited similar regression by day 10, highlighting the importance of tumor-derived G-CSF in establishing an immunosuppressive environment. This reduced tumor progression was mediated by the adaptive immune system, likely through CD8 T cells, as determined by unaltered tumor growth in immunodeficient mice (Supplementary Fig. 9C). However, while the growth of Csf3 KO tumors stabilized thereafter, Csf3 Cmas dKO tumors were completely rejected in subsequent days, suggesting the presence of additional sialylation-dependent mechanisms contributing to tumor control.

To explore whether tumor cell-intrinsic changes in glycosylation might further enhance CD8⁺ T cell-mediated cancer control, we performed quantitative glycoproteomics using our SugarQb pipeline³⁴. We thereby mapped 10,552 unique glycopeptides from 844 glycoproteins, encompassing 257 distinct glycan compositions (Supplementary Data 3), identifying the major histocompatibility complex class I (MHC-I) as a main carrier of altered glycans in Cmas KO cells (Fig. 4H). Specifically, glycoforms of four MHC-I haplotype proteins (H2-L, H2-K1, H2-T3, H2-Q6) shifted from sialylated biantennary N-glycans to terminal galactosylated and truncated structures. Interestingly, Cmas KO 4T1 cells showed significantly higher cell surface expression of MHC-I compared to control 4T1 cells, even in the absence of IFNγ exposure (Fig. 4I). Notably, the total levels of different MHC haplotype proteins remained unchanged, as determined by quantitative proteomics of lysed tumor cells (Supplementary Fig. 10A), indicating alterations in surface MHC-I localization rather than overall expression. In addition, we confirmed this cell-intrinsic effect in CMAS knockouts of two human breast cancer cell lines, MDA-MB-231 and BT-474 (Supplementary Fig. 10B, C). The surface expression levels of MHC-I also significantly increased in two human breast cancer cell lines upon CMAS deletion (Fig. 4J). Finally, to rule out that differences in sialylation affected antibody recognition, we enzymatically desialylated control and Cmas KO 4T1 cells after formalin fixation (Supplementary Fig. 10D). This did not alter the observed differences in surface MHC-I (Fig. 4K), excluding sialylation-mediated differences in antibody detection. Together, these findings suggest that tumor cell-intrinsic sialylation modulates the surface expression of MHC-I, which is a critical determinant of CD8⁺ T cell recognition, immune escape and anti-PD-1 therapy resistance.

Pharmacological inhibition of sialylation phenocopies Cmas genetic ablation for mammary cancer immunosurveillance

Having demonstrated the therapeutic potential of interfering with sialylation through genetic ablation in tumor cell lines, we investigated whether similar effects could be achieved through pharmacologic intervention. To accomplish this, we administered 3F_ax-Peracetyl N-acetylneuraminic acid, a membrane-permeable and selective sialyltransferase inhibitor (STI)³⁵, to 4T1 tumor-bearing mice. Indeed, STI partially inhibited tumor sialylation (Supplementary Fig. 11A, B), leading to a significant reduction in tumor growth (Fig. 5A). Additionally, STI treatment enabled tumor-bearing hosts to better respond to anti-PD-1-based immunotherapy, resulting in prolonged tumor control and a doubling of the overall survival time (Fig. 5A, and Supplementary Fig. 11C–F). Of note, in vitro experiments demonstrated that STI did not exhibit any direct cytotoxicity in 4T1 cells (Supplementary Fig. 11G). Importantly, similar to the effect seen with genetic ablation of sialylation, pharmacological interference was dependent on an intact adaptive immune system: neither STI-induced nor immunotherapy-induced tumoricidal activity was observed in Rag2^−/− γc^−/− mice (Fig. 5B).

Fig. 5. Pharmacologic sialyltransferase inhibition of in vivo 4T1 mammary tumors.

Longitudinal volumetric monitoring of orthotopic 4T1 tumor growth in immunocompetent syngeneic mice (A) or in immunocompromised Rag2^−/− γc^−/− mice (B) in vivo treated with the sialyltransferase inhibitor 3F_ax-Peracetyl-Neu5Ac (STI) and/or anti-PD-1 antibodies. Controls included treatment with vector control and/or an antibody isotype controls. Mean ± SEM. C MHC class I surface expression in 4T1 cells treated with STI and/or IFNγ assessed by flow cytometry detecting H-2Kd. n = 3, mean ±  SD. Controls were treated with vector control. ^*p = 0.0111; ^**p = 0.0019. D ELISA-based quantification of G-CSF concentration in 4T1 tumor-bearing mice treated with STI. Controls were treated with vector control. n = 3, mean ± SEM. ^**p = 0.0023. E Percentages of polymorphonuclear myeloid-derived suppressor cells (PMN-MDSCs), CD8⁺ T cells and CD44⁺ CD8⁺ T cells localized within orthotopic 4T1 tumors were assessed by cytometry on day 8 after tumor cell injection (n = 4, mean ± SEM). Day 8 was chosen because at this time point tumor growth was comparable among the STI treated and control groups. Exact p values are provided as a Source Data file. F Representative images of anti-TCF1 immunohistochemistry at the invasive front of 4T1 tumors treated with STI and/or anti-PD-1 antibodies. Images are representative of 4 evaluated tumor cross sections per treatment group. Positive cells are highlighted by red arrows. Scale bars: 50 μm. High-magnification images are shown in Supplementary Fig. 11H. G Representative images of anti-CD8 immunohistochemistry and quantification within 4T1 tumors treated with STI and/or anti-PD-1 antibodies. Immunohistochemistry was assessed at the endpoint when tumor volumes surpassed 1500 mm³ (Ctrl: n = 10; STI: n = 7; anti-PD-1: n = 4; STI + anti-PD-1: n = 5; mean ± SEM). Scale bars: 50 μm. Exact p values are provided as a Source Data file. H Schematic representation summarizing mechanistic alterations observed in a scenario of impaired sialylation in 4T1 tumors. Created in BioRender. Mereiter, S. (2026) https://BioRender.com/91hoxs7. All data are derived from biological replicates. Statistical differences were assessed by two-way ANOVA (A, B), two-sided t-test (C–E) or one-way ANOVA (G); ^****p < 0.0001. Source data are provided as a Source Data file.

Consistent with the results from CMAS-deficient tumors, STI treatment led to significantly increased surface expression of MHC-I on tumor cells and lower serum G-CSF concentrations (Fig. 5C, D), a marked reduction in PMN-MDSC infiltration, and a concomitant increase in tumor-infiltrating memory CD8⁺ T cells, as determined on day 8 post tumor induction (Fig. 5E, and Supplementary Fig. 12A). By day 14, tumors treated with both STI and anti-PD-1 showed significantly fewer PD-1⁺ T cells (Supplementary Fig. 12B). The elevated levels of CD8⁺ and TCF1⁺ T cells persisted until the endpoint of the study (Fig. 5F, G, and Supplementary Figs. 11H and 12C–I). In summary, these findings indicate that both genetic and pharmacologic disruption of tumor cell sialylation enhances MHC-I surface expression and, in parallel with reduced systemic G-CSF levels and diminished PMN-MDSC infiltration, promotes robust CD8⁺ T cell accumulation, improved tumor control, and increased sensitivity to anti-PD-1 therapy in the 4T1 model (Fig. 5H).

Sialylation is dispensable for mammary gland physiology under homeostatic conditions

To assess the therapeutic significance of sialylation interference in a physiologically relevant setting, we generated a preclinical in vivo model of conditional Cmas deletion in the mammary epithelium. To achieve this, we employed K14-Cre-mediated recombination, which targets the common mammary progenitor cells^36,37, to selectively delete Cmas in both basal and luminal epithelial lineages of the mammary gland (Cmas^ΔK14). The loss of sialylation in Cmas^ΔK14 mice was confirmed by the absence of staining with the sialic acid-specific lectins SNA and MAL1 (Supplementary Fig. 13A, B). Additionally, alcian blue staining, which marks negatively charged glycan clusters typically enriched in sialylated mucins, was also significantly reduced compared to control Cmas^f/f mice (Supplementary Fig. 13C). Importantly, Cmas^ΔK14 mice did not exhibit overt developmental or morphological defects. Histological analysis of the mammary epithelium revealed apparently normal architecture (Supplementary Fig. 13D, E), and mammary gland function was preserved, as evidenced by successful nursing and unaltered litter viability in Cmas^ΔK14 dams compared to controls (Supplementary Fig. 13F, G). Thus, conditional Cmas deletion in the mammary epithelium is compatible with normal gland development as well as mammary gland changes in pregnancy and lactation, indicating that sialylation is dispensable for mammary gland physiology under homeostatic conditions.

Genetic disruption of sialylation improves immunosurveillance and sensitizes autochthonous mammary tumors to immune checkpoint blockade

We next utilized Cmas^ΔK14 and control Cmas ^f/f mice to study the role of sialylation in the autochthonous formation of mammary tumors induced by medroxyprogesterone acetate (MPA) and 7,12-Dimethylbenz[a]anthracene (DMBA) (Fig. 6A)³⁸. The combined application of DMBA/MPA induces aggressive mammary tumors that recapitulate key features of human luminal B-type breast cancer, including resistance to immune checkpoint blockade and elevated G-CSF expression^28,39,40. Notably, Cmas^ΔK14 mice, which lack sialylation in mammary epithelial cells, exhibited improved tumor-free survival with a hazard ratio (HR) of 0.234 (95% CI: 0.103 to 0.530) (Fig. 6B). At humane endpoint, Cmas^ΔK14 displayed significantly lower serum G-CSF levels, accompanied by a reduction in tumor-associated neutrophils and prolonged time-to-death (TTD; HR 0.126, 95% CI: 0.0428-0.373) (Fig. 6C–E, and Supplementary Fig. 13H). The tumors that developed in Cmas^ΔK14 mice remained unsialylated, as evaluated by SNA and MAL2 lectin staining (Supplementary Fig. 14A), thus ruling out the possibility of escape variants resulting from failed Cre-recombination. This observation was consistent with our assessment using lox-stop-lox GFP reporter lines, which demonstrated a high efficiency of K14-Cre mediated recombination in mammary epithelial cells and in tumors harboring floxed alleles (Supplementary Fig. 14B). Importantly, these alterations culminated in a significant increase in overall survival (OS) in Cmas^ΔK14 mice (HR 0.085, 95% CI: 0.028–0.260; Fig. 6F), mirroring our findings in the syngeneic tumor models.

Fig. 6. Cmas deficiency affects the growth of autochthonous mammary tumors and sensitizes to anti-PD1 treatment.

A Treatment scheme for the induction of spontaneous mammary tumors in Cmas^f/f and Cmas^ΔK14 mice through the application of medroxyprogesterone acetate (MPA) slow-release pellets and 7,12-dimethylbenz[a]anthracene (DMBA) oral gavages. Created in BioRender. Mereiter, S. (2026) https://BioRender.com/8iplf9w. B Kaplan–Meier curves of tumor free survival in MPA/DMBA treated Cmas^f/f and Cmas^ΔK14 mice. Tumor-free survival was defined as the time that passed from last gavage until the onset of palpable tumor. C ELISA-based quantification of G-CSF concentration in Cmas^f/f and Cmas^ΔK14 mice with MPA/DMBA induced mammary tumors. Assessed at the endpoint when tumor volume surpassed 1500 mm³ (n = 9 mice; mean ± SEM). ^*p = 0.0386. D Representative images of anti-Ly6G immunohistochemistry and quantification within MPA/DMBA-induced mammary tumors of Cmas^f/f (n = 7) and Cmas^ΔK14 (n = 6) mice. Immunohistochemistry was assessed at the endpoint when tumor volume surpassed 1500 mm³. Positive cells are highlighted by red arrows. Mean ± SEM. ^**p = 0.0045. Scale bars = 50 μm. A higher-magnification view is provided in Supplementary Fig. 13H. E Kaplan–Meier curves of time-to-death in MPA/DMBA treated Cmas^f/f and Cmas^ΔK14 mice. Time-to-death was defined as the time that passed from the onset of a palpable tumor until the humane endpoint when tumor volume surpassed 1500 mm³. F Kaplan–Meier curves of overall survival in MPA/DMBA treated Cmas^f/f and Cmas^ΔK14 mice. Overall survival was defined as the time that passed from last gavage until the ethical endpoint. G Representative images of anti-CD8 and alcian blue co-staining in size-matched MPA/DMBA mammary tumors. Scale bars: 40 μm. H Correlation analysis of alcian blue and CD8 positivity in MPA/DMBA induced tumors of Cmas^f/f mice. Markers were assessed by alcian blue/anti-CD8 co-staining and Spearman correlation analysis was performed to assess p and r values. n = 5 mice with each 5 or 6 areas quantified, as detailed in Source Data file. I Spider plot of treatment responses to anti-PD-1 or isotype control treated MPA/DMBA induced tumors in Cmas^f/f and Cmas^ΔK14 mice. J Kaplan–Meier curves of time-to-death in MPA/DMBA induced and anti-PD-1 treated Cmas^f/f and Cmas^ΔK14 mice. ^**p = 0.0061. B, E, F, J p values were assessed using the Log-rank (Two-sided Mantel-Cox) test. C, D Two-sided t-test. Source data are provided as a Source Data file.

To investigate the relationship between tumor sialylation and immune infiltration, we established a co-staining protocol combining alcian blue, used as a surrogate marker of sialic acid, with CD8 immunohistochemistry (Fig. 6G), allowing spatially resolved quantification of sialylation and CD8⁺ T cell infiltration. When applied to the cohort of sialylation-competent Cmas ^f/f mice, we observed a significant negative correlation between alcian blue positivity and CD8⁺ T cell infiltration (Fig. 6H). We also observed an increased abundance of clustered TCF1⁺ T cells and of B cells at the tumor periphery, paralleling the phenotype in the syngeneic 4T1 tumor setting (Supplementary Fig. 14C, D). Finally, to determine whether loss of sialylation sensitizes immunotherapy-resistant mammary tumors to immune checkpoint inhibition, we treated tumor-bearing Cmas ^f/f and Cmas^ΔK14 mice with anti-PD-1 antibodies. Remarkably, only Cmas^ΔK14 tumors exhibited a significant response to anti-PD-1 therapy as evidenced by reduced tumor growth (Fig. 6I) and significantly prolonged survival (HR 0.084, 95% CI: 0.014-0.493; Fig. 6J). These results demonstrate that while sialylation is dispensable for normal mammary gland function, it plays a critical role in immune evasion of luminal B-type mammary tumors. Genetic disruption of sialylation enhances CD8⁺ T cell infiltration and overcomes intrinsic resistance to immunotherapy in a sex hormone-driven mammary tumor model in vivo.

Sialylation gene signatures correlate with neutrophil infiltration and CD8⁺ T cell exclusion in human luminal breast cancer subtypes

To translate the findings of our clinical and preclinical analyzes to a larger patient context, we analyzed large-scale datasets of human breast cancer. First, we examined the expression of sialylation-related genes across molecular breast cancer subtypes. Among the genes involved in sialic acid uptake, biosynthesis, and transfer, six candidates were identified to be commonly upregulated in breast cancer: CMAS, NANS, ST3GAL1, ST3GAL4, ST6GAL2, and ST6GALNAC2 (Fig. 7A). These genes represent both the biosynthetic core and key α2,3- and α2,6-sialyltransferases, consistent with the observed increase in both α2,3 and α2,6 sialylation across breast cancers (Supplementary Data 1). The combined sialylation gene signature encompassing these 7 genes showed a robust upregulation in breast cancer tissues compared to healthy mammary gland tissue (Fig. 7B). Notably, this increase was particularly pronounced in luminal B breast cancers. To determine whether this sialylation signature is linked to immune cell composition in the tumors, we performed correlation analyzes with gene expression profiles indicative of immune infiltrates. The combined expression of the six sialylation genes correlated positively with neutrophil marker gene profiles and negatively with CD8⁺ T cell signatures (Supplementary Fig. 15A, B), mirroring our in vivo observations in murine models.

Fig. 7. Sialylation gene signatures in human breast cancer.

A Heat-map summarizing the differential sialylation gene expression between breast cancer and healthy breast epithelium, extracted from TNMPlot database (~57,000 breast cancer data derived from multiple integrated cohorts across TCGA, GTEx, GEO, and TARGET). Sialylation genes were grouped into α2,6-sialyltransferases (ST6GAL1, ST6GAL2, ST6GALNAC1-4), α2,3-sialyltransferases (ST3GAL1-6), sialic acid uptake and biosynthesis (SLC35A1, SLC17A5, NANS, GNE, CMAS). ERBB2 was included as reference and significantly upregulated genes were highlighted with bold frame. B Significantly upregulated genes (CMAS, NANS, ST3GAL4, ST6GALNAC2, ST3GAL1 and ST6GAL2) were combined to a sialylation gene signature and compared between breast cancer and normal breast tissue as well as across the molecular subtypes (Basal like, HER2, Luminal A and Luminal B) using the platform GEPIA2. Numbers (n) of patients and healthy controls are indicated. Box plots show median expression (center line), interquartile range (box), and whiskers defined as 1.5×IQR. C Systematic correlation analysis of the individual sialylation gene signature genes (CMAS, NANS, ST3GAL4, ST6GALNAC2, ST3GAL1 and ST6GAL2) with neutrophils (N) and CD8⁺ T cells (CD8) across breast cancer molecular subtypes. Circle sizes reflect statistical significance and colors positive (blue) or negative (red) correlation. Selected correlations are detailed in correlation plots for purity and respective immune populations. Scatter plots with fitted trend lines are shown; statistical significance was evaluated using Spearman’s correlation with the grey shaded region represents the 95% confidence interval around the fitted trend line. Data was extracted from Timer. D Comparative analysis between sialylation score and CD8 positive area in the Swiss cohort (top) or tumor infiltrating lymphocytes (TILs) in the French cohort (bottom). Data corresponds to Fig. 1B but was stratified into luminal B (Swiss cohort: n = 20; French cohort: n = 29) or remaining subtypes (Swiss cohort: n = 30; French cohort: n = 57). B, D Two-sided t-test, p* <0.05, ** <0.01. Source data are provided as a Source Data file.

Further dissecting the individual contributions of each gene across molecular subtypes (Fig. 7C), we found that five of the six genes (CMAS, NANS, ST3GAL1, ST3GAL4, and ST6GAL2) positively correlated with neutrophil infiltration across all subtypes. In contrast, ST6GALNAC2 showed a negative correlation with neutrophil markers. Within luminal tumors, ST3GAL1 and ST6GAL2 emerged as the most prominent genes that correlated with neutrophil recruitment. When examining associations with CD8⁺ T cell infiltration, ST3GAL4 and ST6GAL2 expression demonstrated the strongest negative correlations, suggesting a potential role in T cell exclusion (Fig. 7C). Interestingly, while some of the other sialylation-related genes showed either positive or no correlation with CD8⁺ T cells, ST6GAL2 consistently emerged as the most robust predictor of both increased neutrophil infiltration and reduced intratumoral CD8⁺ T cells. This association was particularly pronounced in the luminal subtypes, where ST6GAL2 expression most strongly correlated with an immunosuppressive tumor microenvironment as characterized by high neutrophil and low CD8⁺ T cell infiltration (Fig. 7C). Importantly, this pronounced association between sialylation and immune infiltration in luminal B breast cancer was also supported by our analysis of the Swiss and French clinical cohorts. Upon stratification by molecular subtype, we confirmed that while the same trend was observed across all subtypes (Supplementary Fig. 15C, D), luminal B tumors were the main contributors to the significant association between high sialylation and low T cell infiltration (Fig. 7D). Together, these data in human breast cancer patients support a robust association between tumor sialylation and intratumoral immune cell composition, in particular in patients with luminal B breast cancers.

Discussion

The unique immune privilege of the breast allows profound and repeated morphological and molecular changes during development, menstrual cycles, and pregnancy. Intriguingly, the immunoregulatory Siglec (Sialic acid-binding immunoglobulin-type lectins) family has expanded in mammals together with the evolution of the mammary glands⁴¹. We therefore speculated that among all organs, the mammary epithelium, which is the evolutionary youngest and defining organ of mammals, may particularly capitalize on such sialic acid immunoregulatory circuits.

In this study, we genetically ablated sialylation by targeting the Cmas gene, which encodes an essential enzyme in the sialic acid biosynthesis pathway and which cannot be bypassed through salvage or other pathways, and which has no known other functions¹¹. Genetic knockout of Cmas had only minor effects on the development and function of mammary glands in mice, suggesting that sialic acids are not crucial components of physiological mammary gland biology. This is in line with previous studies that have shown that sialylation is not essential for early embryonic development and organogenesis in mice^42,43. It is likely that sialylation of mammary epithelial cells plays a role in processes such as infection clearance or prevention of autoimmune diseases, which needs to be explored in future experiments. Importantly, our data indicate that the absence of sialylation has beneficial effects in tumorigenesis, as it protects against the spontaneous formation of mammary tumors and prolongs the survival of progestin- and carcinogen-induced mammary tumors. The lack of evident pathology in the Cmas conditional knockout mice further suggests that local interference with mammary gland sialylation may be well-tolerated and non-hazardous to the non-transformed mammary tissue in breast cancer patients.

In the 4T1 mammary cancer cell line, which is widely utilized in murine mammary tumor studies, the impact of sialylation interference appears to be primarily due to immunoregulation. This is also evident from the absence of growth disparities when compared to sialylation-competent tumor cells in immunocompromised mice. Moreover, it is conceivable that the removal of sialylation also prevents metastasis, as previously described⁴⁴.

Using multiple assays such as single-cell transcriptomics or in situ immunodetection, in combination with functional in vivo assays, we report that sialylation in breast tumors drives the recruitment of PMN-MDSCs while hindering the efficient eradication of tumors by CD8⁺ T cells. Consequently, pharmacologic or genetic abrogation of tumor sialylation facilitated CD8⁺ mediated tumor control and the recruitment of Tcf7⁺ CD8⁺ memory T cells within the tumor microenvironment. Tcf7⁺ CD8⁺ T cells represent a stem-like or progenitor subset that sustains long-term anti-tumor immunity by self-renewal and differentiation into effector T cells^19,45,46. These cells are critical for maintaining CD8⁺ T cell responses within the tumor and have been linked to ICI responsiveness^{18–20,47–50}. The observed increase in the intratumoral Tcf7⁺ CD8⁺ T cells suggests that tumor desialylation may expand the progenitor T cell pool, thereby enhancing the potential for sustained CD8⁺ effector activity. The interference with sialylation, therefore, resulted not only in prolonged control of the breast tumors but, most importantly, also in breaking the resistance of breast tumors to anti-PD1 therapy.

Previous studies have shown that other innate immune cells, such as tumor-associated macrophages, can contribute to improved CD8⁺ T cell-mediated tumor control upon partial desialylation of colorectal cancer and melanoma models¹⁴. This phenotype was functionally linked to the actions of Siglecs^14,15. In contrast, complete desialylation interference in a colorectal tumor model resulted in immune-dependent accelerated tumor growth⁵¹. These observations are likely consequences of the intricate sialylation-immune axis and suggest that, similar to other immunotherapies, the outcome might depend on tumor-intrinsic properties. Thus, it is crucial to define key determinants that can serve as predictive markers for treatment outcomes following sialylation interference. For breast cancer, our data, along with findings from other studies^14,52,53, consistently demonstrate that sialylation contributes to a suppressed immune response. Here, we demonstrate that sialylation in breast tumors promotes neutrophilia and facilitates the recruitment of PMN-MDSCs, ultimately creating an environment that impairs effective CD8⁺ T cell-mediated tumor clearance.

How is impaired sialylation mechanistically linked to better tumor control at the molecular level? Our findings show that interfering with sialylation reduces the serum half-life of G-CSF, which has been identified as a key driver of PMN-MDSC recruitment in the 4T1 tumor model^25,54. Sialylation interference may thus prove especially beneficial for patients whose tumors produce high levels of G-CSF, a common feature in breast cancer^25,28. Cytokines such as CHI3L1 or IL-1β have also been implicated in neutrophil recruitment and may also contribute to the observed phenotype^27,55. Importantly, our data suggest that the therapeutic effects of sialylation inhibition extend beyond the modulation of innate immune cell recruitment. We observed increased cell surface expression of MHC-I in desialylated tumor cells. As MHC-I levels are critical determinants of CD8⁺ T cell-mediated cytotoxicity^4,8,56–58, and prior studies have shown enhanced MHC-I retention on the surface of desialylated dendritic cells⁵⁹, these findings highlight an additional, complementary mechanism by which sialylation may influence anti-tumor immunity. The mechanism underlying increased MHC-I expression appeared to be cancer cell-intrinsic and independent of the G-CSF/PMN-MDSC regulatory axis. Nevertheless, these pathways are likely to act synergistically, as MHC-I levels were further enhanced in Cmas KO compared to their respective control cells following IFNγ stimulation (Fig. 4I). PMN-MDSCs are major suppressors of IFNγ production in the TME⁶⁰. It is therefore likely that MHC-I-dependent CD8⁺ T cell-mediated killing is particularly enhanced in sialylation-deficient tumors, supported by the reduced number of PMN-MDSCs. Not excluding other cellular and molecular mechanisms, these data indicate that tumor cell sialylation modulates multiple immunoregulatory pathways, both by supporting the recruitment of immune-suppressive myeloid cells and by directly altering tumor immunogenicity, thereby enabling immune evasion in breast cancer. While these findings provide important mechanistic insights, they are primarily based on mouse models of breast cancer, and further validation in a human context will be important to assess generalizability.

In this study, using two preclinical models of immunologically cold and ICI-unresponsive mammary cancers, we have demonstrated that inhibition of sialylation sensitizes tumors to combination therapy with anti-PD1 immune checkpoint inhibition. This intervention demonstrates a successful preclinical strategy to render aggressive MPA/DMBA mammary tumors amenable to immunotherapy. This data underscores the potential of sialylation interference strategies for future clinical combination therapies, including those currently being assessed in clinical trials (NCT05259696). Finally, the high frequency with which we observed the upregulation of sialylation genes and hyper-sialylation in breast tumors of patients, along with its significant positive correlation with neutrophils/PMN-MDSCs and negative correlation with tumor-infiltrating T lymphocytes, suggests that a substantial number of breast cancer patients could benefit from therapeutic interventions targeting sialylation.

Methods

Ethical consideration

The research presented complies with all relevant ethical regulations. All mouse experiments were performed according to our license approved by the Austrian Ministry (GZ: BMWF-66.009/0201-WF/II/3b/2014; GZ.: 2023-0.517.961 and their amendments). The use of patient-derived material for histological assessment was approved by the Ethics Committee of Eastern Switzerland (Project-ID 2022-00082) or by the Ethics Committee of Gustave-Roussy for the retrospective study “GrandTMA” and the research was conducted in adherence to the Declaration of Helsinki guidelines.

Cell lines

All breast cancer cell lines were obtained from ATCC (Manassas, VA) and cultured in DMEM High Glucose medium supplemented with 10% FCS and penicillin-streptomycin (100 U Pen/ml; 0.1 mg Strep/ml). Cells were authenticated by short tandem repeat analyzes and regularly tested for mycoplasma by PCR.

Cmas knockout cells

Guide RNAs against mouse or human CMAS or mouse Csf3 were designed using the Broad Institute CRISPick tool (https://portals.broadinstitute.org/gppx/crispick/public). Reference genomes were Human GRCh38 or Mouse GRCm38, and the mechanism CRISPRko Enzyme: SpyoCas9. The top 5 crRNA spacer sequences were selected and cloned into the pX459v2 Cas9 Puro Plasmid (Addgene Plasmid #62988) using single-step golden-gate cloning. After plasmid purification, 500 ng of pX459v2:sgRNA DNA was transfected into cells using Lipofectamine2000 following the manufacturer’s instructions. The day after transfection, the media was replaced with growth medium containing 1 µg/ml Puromycin. After 48 h of transient selection, the medium was replaced with standard growth medium. Cells transfected with guides against murine or human CMAS were bulk sorted on Aria III (BD, Franklin Lakes, NJ) using Sambucus nigra lectin (SNA, Vector) for sialylation negative cells (Cmas KO) and cells with unaltered sialylation levels (Ctrl). Of note, prior to genetic editing, no SNA-negative cells were detectable in any of the breast cancer cell lines used (Supplementary Fig. 2C). Selection for SNA-negative cells therefore enabled enrichment of the edited population without the need for clonal isolation, an approach that has been previously applied in similar contexts^51,61. Three different guides were tested each to target human CMAS or mouse Cmas (Supplementary Table 1). The best performing guides assessed by the capacity of inducing SNA negative cells were human guide 1 and mouse guide 3, and were the guides used for all conditions. In every experiment, a more than 95% purity of SNA-negative or positive cells was confirmed for Cmas KO and their respective control cells. Cells transfected with guides against Csf3 were single-cell sorted into 96-well plates. The supernatant of each clone was tested via murine G-CSF ELISA Kit (MCS00, R&D Systems, Minneapolis, MN) and 10 G-CSF negative clones were pooled to avoid clonal effects. Empty vector transfected and sorted clones were used a control.

Animal housing conditions

All mice were bred and housed under specific pathogen-free conditions at a temperature of 22 ± 1 °C and relative humidity of 55 ± 5%, with a 14 h light/10 h dark cycle. Mice had ad libitum access to food and water. Within each experiment, age- and sex-matched groups were used.

Orthotopic tumor experiments

4T1 cells were orthotopically injected into the right inguinoabdominal mammary fat pad of 8 to 12-week-old female BALB/c mice or BALB/c Rag2^−/−γc^−/− mice (both Jackson Laboratory, Bar Harbor, ME). For this purpose, cells were cultured in monolayer, harvested by trypsin, and washed with PBS. 100 μl of a 1:1 PBS and Matrigel (Corning Inc., Corning, NY) cell suspension (250,000 cells) was injected per mouse. Tumors were monitored and measured every second day starting from day 4 after injection. Tumors were measured using digital calipers; the tumor volume was calculated as length × width × height. Tumor-bearing mice were sacrificed following ethical guidelines, at the latest when tumor volume exceeded the maximum tumor volume of 1500 mm³. In some cases, this limit has been exceeded on the last day of measurement, and the mice were immediately euthanized.

In vivo cell depletion experiments

T cells were depleted from tumor-bearing mice by the application of anti-CD4 (Clone GK1.5, in-house produced) and/or anti-CD8 (Clone 2.43, in-house produced) monoclonal antibodies. For this purpose, i.p. injection of 100 μg anti-CD4 and/or 100 μg anti-CD8 or a species-matched (rat) isotype control antibody IgG2b (BioXCell, Lebanon, NH) was administered on days 0, 2, 8, 14 and 18 post tumor cell inoculation. On day 6 the T cell depletion was confirmed via flow cytometry using blood analysis. Neutrophils were depleted using a protocol modified from Boivin et al.⁶². Briefly, tumor-bearing mice were treated every other day by two i.p. injections. First 50 μg anti-Ly6G (1A8, BioXCell) or isotype control antibodies (BioXCell), followed by a second i.p. injection with 50 μg anti-Rat IgG (MAR18.5) 2 h later. On day 7 the neutrophil depletion was confirmed using flow cytometry analysis of blood samples.

Enzymatic desialylation of G-CSF

25 μg of recombinant murine G-CSF (Z03163, GenScript, Piscataway, NJ) reconstituted in 250 μl PBS, was incubated with 250 ng of recombinant Arthrobacter ureafaciens Sialidase (in-house produced, endotoxin-free) in 500 μl 50 mM sodium acetate buffer (pH 5.5) at 37 °C for 2 h. The successful desialylation was validated via migration shift of G-CSF on Coomassie stained (Simply Blue SafeStain, Invitrogen) SDS-PAGE Nu Page 4–12% Bis-Tris gels (Thermo Fisher Scientific, NP0321BOX) as previously described⁶³. 10-week-old female BALB/c mice were once i.p. injected with 150 μl of desialylated G-CSF corresponding to 5 μg, and neutrophil induction was monitored in the blood though submandibular bleeding. The control group was treated with the same concentration of all constituents (G-CSF in PBS with neuraminidase in sodium acetate buffer) mixed immediately before i.p. injection. G-CSF concentration was measured via murine G-CSF ELISA Kit (MCS00, R&D Systems, Minneapolis, MN).

In vivo inhibition experiments

For sialyltransferase inhibition experiments, tumor-bearing mice were treated starting from day 4 post tumor inoculation. The inhibitor (3F_ax-Peracetyl N-Acetylneuraminic acid (Sigma)) was administered at sub-lethal doses determined in pilot studies³⁵. For intra-tumor injections, 600 μg of 3F_ax-Peracetyl N-Acetylneuraminic acid was dissolved in 30% DMSO in PBS and given every second day until day 12 post tumor induction. To avoid kidney damage, the treatment dose was lowered from day 14 post tumor induction to 300 μg of 3F_ax-Peracetyl N-Acetylneuraminic acid every 4 days. The control group received vehicle control injections. For anti-PD-1 experiments, tumor-bearing mice received i.p. injections of 250 μg anti-PD-1 or 250 μg isotype control in InVivoPure pH 7.0 Dilution Buffer (all BioXCell) on days 10, 12 and 14 post tumor induction. Anti-PD-1 antibody injections were thereafter continued every 4 days. PAD4 inhibition experiments were performed as previously described²⁷. Briefly, 20 mg/kg GSK484 (Supplementary Table 1) was administered daily by intraperitoneal injection starting on day 2 after tumor induction and prior to the onset of palpable tumors, and these injections were continued until the experimental endpoint. Control animals received vehicle injections.

Interferon gamma stimulation

Cancer cell lines were treated for 48 h with 10 ng/ml murine interferon gamma (Supplementary Table 1). Cells were subsequently harvested and stained for 1 h at 4 °C with fluorescently labeled anti-MHC class I antibodies and Fixable Viability Dye eFluor^TM 780 (Supplementary Table 1). Surface expression of MHC class I was determined using a BD Fortessa instrument (BD, Franklin Lakes, NJ) and analyzed with FlowJo (BD) and GraphPad Prism 8 (GraphPad Software, Boston, MA).

Enzymatic desialylation of cells

Cells were cultured in monolayers, harvested by trypsinization, and washed with PBS. Cells were then fixed with 4% paraformaldehyde for 20 min at room temperature using gentle agitation and washed twice with PBS. Fixed cell pellets were incubated with 10 µg/ml recombinant Arthrobacter ureafaciens sialidase (in-house produced, endotoxin-free) in 500 μl of 50 mM sodium acetate buffer (pH 5.5) at 37 °C for 2 h. Desialylation was confirmed by fluorescently labeled Sambucus nigra agglutinin as described below (Glycan analysis by flow cytometry) after which cells were used for MHC class I immunostaining.

Autochthonous mammary tumor induction

Mammary tumors were induced by the combined application of medroxyprogesterone acetate (MPA) and 7,12-Dimethylbenz[a]anthracene (DMBA) as previously described³⁸. 6-week-old female mice were subcutaneously implanted (in the neck region) with 90-day slow-release MPA pellets (50 mg, Innovative Research of America, Sarasota, FL). Subsequently, 200 μl of a 5 mg/ml DMBA (Merck) solution in cottonseed oil was administered by oral gavage six times over the following 8 weeks, as outlined in Fig. 6A. The onset of mammary tumors was monitored by palpation. Tumor-bearing mice were sacrificed following ethical guidelines, at the latest when tumor volume exceeded the maximum tumor volume of 1500 mm³. In some cases, this limit has been exceeded on the last day of measurement, and the mice were immediately euthanized.

Immunophenotyping

For flow cytometry analyzes, tumors were resected from euthanized mice and immediately dissociated using a tumor dissociation kit and GentleMACS dissociator (both Miltenyi). Single cell suspensions were first incubated with CD16/32 Fc Block (BD) and Fixable Viability Dye eFluor^TM 780 (eBiosciences) for 20 min at 4 °C, followed by antibody staining for 20 min at 4 °C in PBS supplemented with 2% FCS using the reagents listed in the Supplementary Table 1. Antibody staining was determined using a BD Fortessa instrument (BD, Franklin Lakes, NJ) and analyzed with FlowJo (BD) and GraphPad Prism 8 (GraphPad Software, Boston, MA). The used gating strategies are detailed in Supplementary Fig. 16. For cytokine array analyzes, blood was collected via submandibular bleeding from tumor-bearing mice on day 20 post tumor induction into BD Microtainer® Serum tubes with Separating Gel (BDAM365968, BD). Sera were consequently isolated by centrifugation, and samples were analyzed using the Mouse Cytokine Array Kit (ARY006, R&D Systems) according to the manufacturer’s protocol.

Glycan analysis by flow cytometry

Cancer cells were harvested with 1 mM EDTA in PBS and washed 3 times in TSM buffer (20 mM Tris HCl, 150 mM NaCl, 2 mM CaCl2, 2 mM MgCl₂). Cells were then stained with fluorescently labeled plant lectins for 45 min at 4 °C in TSM buffer using the reagents listed in the Supplementary Table 1. Staining was assessed with a BD Fortessa instrument (BD). All data was analyzed by FlowJo (BD) and GraphPad Prism 8 (GraphPad Software) or R Studio (Posit, Boston, MA).

Human tumor tissue

For the Swiss cohort, human breast cancer tissues were included from 50 female patients collected at the Kantonsspital St. Gallen, Switzerland. In brief, breast cancer biopsies were taken for diagnostic purposes between September 2018 and July 2020 from female patients who had consented to further scientific use of their biological samples. Ten cases were selected from each Luminal A, Luminal B, Luminal B HER2 positive, HER2 positive, and triple negative molecular subtypes, respectively. Additionally, ten control samples were chosen from female patients with breast reduction surgery between August 2018 and December 2021. The study was approved by the local ethics committee (Ethics Committee of Eastern Switzerland, Project-ID 2022-00082) and conducted in adherence to the Declaration of Helsinki guidelines. For the French cohort, formalin-fixed, paraffin-embedded breast cancer tissue MicroArray (TMA) samples and corresponding tumor sections were obtained from the cohort of the retrospective study GrandTMA and consisting of early-invasive breast cancer specimens retrospectively collected from female patients diagnosed and treated at Gustave Roussy, France. All patients had given their written consent, were over 18 years old and underwent surgical resection as their first treatment.

Histologic analyzes

Tumors were resected from euthanized mice and immediately fixed in neutral buffered formalin (HT5012, Sigma-Aldrich, Burlington, MA) for 48 h at room temperature. Following several washes with PBS, fixed tumors were embedded in paraffin, cut at 2 μm thickness, and mounted on glass slides (Permaflex plus adhesive slides, Leica, Wetzlar, DE). Slides were deparaffinized in automatic stainer Gemini AS (Fisher Scientific, Hampton, NH) by sequential application of xylene substitute (Shandon, Thermo Scientific) and decreasing concentrations of ethanol in water. For chromogenic immunohistochemistry with antibodies, dewaxed slides were subjected to antigen retrieval with sodium citrate buffer (pH 6, see Supplementary Table 1), endogenous peroxidase inactivation with 3% H₂O₂ (H1009, Sigma) and were also blocked in 5% BSA in TBST. In case of lectin staining, endogenous peroxidase inactivation with 3% H₂O₂ was performed on dewaxed specimen followed by blocking with avidin-biotin blocking solution (ab64212, Abcam) and 5% BSA in TBST. Slides were washed in TBS between steps and incubated with the respective primary antibodies or lectins (Supplementary Table 1) for 2 h at room temperature. Detection system rabbit, detection system rat or streptavidin-HRP (Supplementary Table 1) were used for rabbit primary antibody, rat primary antibody or lectin staining, respectively. The chromogenic reaction was induced using the DAB substrate kit (ab64238, Abcam). Hematoxylin counter staining or hematoxylin and eosin (H&E) staining and subsequent dehydration were performed using reagents detailed in the Supplementary Table 1, using automatic stainer Gemini AS. Slides were mounted with cover slips by Tissue-TEK GLC (Sakura Finetek, Torrance, CA), air dried and scanned with Slide Scanner Pannoramic 250 (3DHistech Ltd, Budapest, HU). Whole mount staining was prepared as previously described⁶⁴ and stained with carmine alum and mounted using Eukitt Neo mounting medium (O. Kindler, Freiburg, DE). Alcian blue stains were performed by 30 min incubation with alcian blue (pH 2.5), followed by washing in running water and counterstained with Nuclear fast red for 3 min (Supplementary Table 1). Human sections of the Swiss cohort and the TMA of the French cohort were stained with the lectins SNA or MAL2 and were assessed with CaseViewer (3DHistech Ltd) and manually scored from 0 to 4 for the staining intensity of cancer cells as compared to staining intensity obtained in healthy mammary glands (Supplementary Fig. 1). Sialylation scores were defined as the sum of SNA and MAL2 scores. Tumor sections of the Swiss cohort stained with anti-CD3, anti-CD8 and anti-FOXP3 were analyzed using Fiji ImageJ. For this purpose, 3 representative intra-tumoral areas were selected per case and batch quantified using a sequence of color deconvolution (H DAB), intensity threshold, conversion to mask and area fraction measurement, shown as percentages of positively stained tumor cells. Tumor sections of the French cohort, corresponding to the analyzed TMA cases, were stained with haematoxylin-eosin-saffron (HES) and scanned with an Olympus VS120 scanner at 20× magnification. TILs (tumor-infiltrating lymphocytes) scoring was assessed on WSI (whole slide images) as the percentage of tumor stroma surface occupied by immune mononuclear cells, in accordance with international guidelines¹⁶.

Single cell transcriptomics

Tumors were resected at day 15 post injection and immediately dissociated using a tumor dissociation kit and GentleMACS dissociator (both Miltenyi). For each condition, 5 tumor dissociations were pooled and CD45 positive cells were then isolated with anti-CD45 microbeads (Miltenyi) following a previously described protocol⁶⁵. Cells were counted using NucleoCounter NC250 (Chemometec) following the manufacturer's instructions and CD45 positivity and viability measured on a BD Fortessa instrument (BD); more than 90% cells were viable and 95% of the cells were confirmed to express CD45. For each sample, one million cells were fixed for 22 h at 4 °C, quenched and stored at −80 °C according to 10X genomic Fixation of Cells & Nuclei for Chromium Fixed RNA profiling (CG000478) using the Chromium Next GEM Single Cell Fixed RNA Sample preparation kit (PN-1000414, 10X Genomics, Pleasanton, CA). 250,000 cells per sample were used for probe hybridization using the Chromium Fixed RNA Kit, Mouse Transcriptome, 4rxn × 4BC (PN-1000496, 10X Genomics), pooled at equal numbers and washed following the Pooled Wash Workflow following the Chromium Fixed RNA Profiling Reagent kit protocol (CG000527, 10X Genomics). GEMs were generated on Chromium X (10X Genomics) with a target of 10,000 cells recovered and libraries prepared according to the manufacturer instructions (CG000527, 10X Genomics). Sequencing was performed using NovaSeq S4 lane PE150 (Illumina) with a target of 15,000 reads per cell. Alignment of the samples was perfomed with the 10X Genomics Cell Ranger 7.1.0 multi pipeline⁶⁶. The reads were trimmed to a length of 28 bp read1 and 92 bp read2 and aligned to the mouse reference genome (refdata-gex-mm10-2020-A) and probeset (Probe_Set_v1.0.1_mm10-2020-A.csv) with default parameters using cellranger multi. The four per-sample count outputs were combined with cellranger aggr. The cellranger output files for each sample (filtered barcodes, features and count matrix files) were then used for custom analysis in R using the Seurat package (version 4.2.0)^67,68. In brief, Seurat objects for each file were generated by keeping cells with more than 200 features. After data normalization (NormalizeData, default setting) and scaling (ScaleData, default settings), the top 2000 variable genes were computed using FindVariableFeaturesgenes (“vst” method). For each sample, data were then transformed using SCTransform (SCTransform, vst.flavor = “v2”), followed by principal component analysis (RunPCA, npcs=50). Data integration was then performed by first generating a list containing the individual sample object, followed by selection of the top 3000 most variable features (SelectIntegrationFeatures, nfeatures = 3000) and final integration using PrepSCTIntegration (anchor.features = features from SelectIntegrationFeatures). FindIntegrationAnchors (normalization.method = “SCT”) was used to find combined integration anchors, which were then used to performa final integration using IntegrateData (normalization.method = “SCT”). RunPCA was used to find the top principal components of the integrated data object for downstream clustering. The first 40 PCAs were used to find neighbors FindNeighbors (reduction = “pca”, dims = 1:40), followed by clustering (FindClusters, resolution=0.7 or 0.5 or 0.3) and UMAP dimensionalty reduction (RunUMAP reduction = “pca”, dims = 1:40). To assign clusters to cell types, FindAllMarkers was used to find features that are upregulated in individual clusters with a log fold change of at least 0.25, and which are expressed in at least 25% of cells (logfc.threshold = 0.25,min.pct = 0.25). The list was then used to manually assign cell types. R-code, raw sequencing files and Seurat object are available via (NCBI GEO).

Bioinformatics analysis

Expression data for sialylation-related genes (CMAS, GNE, NANS, SLC17A5, SLC35A1, ST3GAL1, ST3GAL2, ST3GAL3, ST3GAL4, ST3GAL5, ST3GAL6, ST6GAL1, ST6GAL2, ST6GALNAC1, ST6GALNAC2, ST6GALNAC3, ST6GALNAC4) and ERBB2 in breast cancer were extracted from the pan-cancer dot matrix of TNMplot (tnmplot.com)⁶⁹ and visualized using Adobe Illustrator. Expression data comparing sialylation gene signatures (CMAS, NANS, ST3GAL4, ST6GALNAC2, ST3GAL1, ST6GAL2) across molecular breast cancer subtypes were obtained from the GEPIA2 Expression DIY tool (gepia2.cancer-pku.cn/)⁷⁰. Correlation data for the same gene set were retrieved from TIMER2.0 (https://compbio.cn/timer2/)⁷¹. Circle plots were generated in RStudio (v2022.07.2) based on systematically extracted adjusted p-values and Spearman’s rho values across BRCA Subtypes (Basal, Her2, LumA, LumB) and immune cell type (T cell CD8+_EPIC and Neutrophil_TIMER). Representative correlation plots were directly obtained from the TIMER2.0 platform.

N-glycomic analyses

Cancer cells were harvested using 1 mM EDTA in PBS, and cell pellets were washed once with TSM buffer (20 mM Tris HCl, 150 mM NaCl, 2 mM CaCl2, 2 mM MgCl2) and further three times with PBS. Cell pellets were then mixed with 100 mM ammonium bicarbonate buffer containing 20 mM dithiothreitol and 2% sodium dodecyl sulphate at a total volume of 500 µl. After homogenization with an Ultra-Turrax T25 disperser, the homogenates were incubated at 56 °C for 30 min. After cooling down, the solution was brought to 40 mM iodoacetamide and incubated at room temperature in the dark for 30 min. Chloroform-methanol extraction of the proteins was carried out using standard protocols⁷². In brief, supernatants were mixed with four volumes of methanol, one volume of chloroform, and three volumes of water in the given order. After centrifugation of the mixture for 3 min at 2500 rpm, the upper phase was removed, 4 ml of methanol was added, and pellets resuspended. The solution was centrifuged, the supernatant removed, and the pellet was again resuspended in 4 ml methanol. The last step was repeated two more times. After the last methanol washing step, the pellet was dried at room temperature. Dried pellets were resuspended in 50 mM ammonium acetate (pH 8.4). For N-glycan release, 2.5 U N-glycosidase F (Roche, DE) was added, and the resulting mixture was incubated overnight at 37 °C. The reaction mixture was acidified with some drops of glacial acetic acid and centrifuged at 4000 × g. The supernatant was loaded onto a Hypersep C18 cartridge (1000 mg; Thermo Scientific, Vienna) that had been primed with 2 ml of methanol and equilibrated with 10 mL water. The sample was applied, and the column was washed with 4 ml water. The flow-through and the wash solution were collected and subjected to centrifugal evaporation. Reduction of the glycans was carried out in 1% sodium borohydride in 50 mM NaOH at room temperature overnight. The reaction was quenched by the addition of two drops of glacial acetic acid. Desalting was performed using HyperSep Hypercarb solid-phase extraction cartridges (25 mg) (Thermo Scientific, Vienna). The cartridges were primed with 450 μl methanol followed by 450 μl 80% acetonitrile in 10 mM ammonium bicarbonate. Equilibration was carried out by the addition of three times 450 μl water. The sample was applied and washed three times with 450 μl water. N-glycans were eluted by the addition of two times 450 μl 80% acetonitrile in 10 mM ammonium bicarbonate. The eluate was subjected to centrifugal evaporation, and the dried N-glycans were resuspended in 20 μl HQ-water. 5 μl of each sample was subjected to LC-MS/MS analysis. LC-MS analysis was performed on a Dionex Ultimate 3000 UHPLC system coupled to an Orbitrap Exploris 480 Mass Spectrometer (Thermo Scientific). The purified glycans were loaded on a Hypercarb column (100 mm × 0.32 mm, 5 μm particle size, Thermo Scientific, Waltham, MA, USA) with 80 mM ammonium formate as the aqueous solvent A and 80% acetonitrile in 80 mM ammonium formate as solvent B. The gradient was as follows: 0–4.5 min 1% solvent B, from 4.5 to 5.5 min 1–9% B, from 5.5 to 30 min 9–20% B, from 30 to 41.5 min 20–35% B, from 41.5 to 45 min 35–65% B, followed by an equilibration period at 1% solvent B from 45 to 55 min. The flow rate was 6 μl/min. MS analysis was performed in data-dependent acquisition (DDA) mode with positive polarity from 500 to 1500 m/z using the following parameters: resolution was set to 120,000 with a normalized AGC target of 300%. The 10 most abundant precursors (charge states 2-6) within an isolation window of 1.4 m/z were selected for fragmentation. Dynamic exclusion was set at 20 s (n = 1) with a mass tolerance of ±10 ppm. Normalized collision energy (NCE) for HCD was set to 20, 25 and 30% and the precursor intensity threshold was set to 8000. MS/MS spectra were recorded with a resolution of 15,000, using a normalized AGC target of 100% and a maximum accumulation time of 100 ms. Every cell line was analyzed in biological triplicates.

Cell lysis for glycoproteomic analysis

All cell pellets were lysed using a buffer containing 50 mM Tris-HCl (Trizma Base: Sigma-Aldrich, Cat#T6066; HCl: Sigma-Aldrich, Cat#258148), 8M urea (VWR, Cat#0568), 150 mM NaCl (Merck, Cat#106404), pH 8, containing phosphatase inhibitors (Cell Signaling Technology, Cat#5870). Samples were kept on ice for 30 min before being sonicated using a water bath sonicator for 1 min at low intensity. Samples were centrifuged at 20,000 × g for 15 min at 4 °C. Resulting supernatants were transferred to a new tube, and protein contents were quantified using Pierce™ BCA Protein Assay Kit (Thermo Fischer, Cat#23225). For each sample, 400 µg of protein were reduced using DTT (Roche, Cat#10708984001; final concentration 10 mM) for 30 min at 37 °C, and subsequently alkylated using IAA (Sigma-Aldrich, Cat#I6125; final concentration 20 mM) for 30 min in the dark at room temperature. Samples were then diluted using a 50 mM Tris-HCl buffer to lower urea concentrations to 4M before adding 8 µg of Lys C (FUJIFILM Wako, Cat#125 05061) to each sample (1 µg Lys C: 50 µg sample protein). Samples were incubated for 2 h at 25 °C, before being further diluted by adding 50 mM Tris-HCl buffer to lower the urea concentration to 2M. 8 µg of Trypsin Gold (Promega, Cat#V5280) were added to each sample (1 µg Trypsin: 50 µg sample protein), and samples were further incubated for 15 h at 32 °C. At this point, samples were acidified using TFA to a final concentration of 0.3% before being cleaned using 200 mg (3cc) Sep-Pak cartridges (Waters, Cat#WAT054945). The resulting elution fraction containing the cleaned peptides was dried under vacuum and stored at 20 °C before proceeding with the TMT labeling protocol.

TMT labelling before glycoproteomic analyses

After quantification, 120 µg of peptides of each sample were prepared in a final volume of 100 µl of 100 mM HEPES buffer pH 7.6. Samples were labeled using TMTpro 18 plex (Thermo Scientific, Cat#A52045) reagents, according to the manufacturer’s instructions. The labeling efficiency was determined by LC-MS/MS on a small aliquot of each sample. After quenching the labeling reaction, samples were mixed in equimolar amounts, evaluated by LC-MS/MS. The mixed sample was acidified to a pH below 2 with 10% TFA and desalted using C18 cartridges (Sep-Pak Vac 1cc (200 mg), Waters, Cat#WAT054945). Peptides were eluted with 2 ×600 μl 80% Acetonitrile (ACN) and 0.1% Formic Acid (FA), followed by freeze-drying.

Glycopeptide enrichment, LC-MS/MS and data analyses

Glycopeptides from the resulting TMTpro mixes were enriched by performing off-line ion pairing (IP) HILIC chromatography, as described previously³⁴. Briefly, the dried samples were taken up in 100 μl 75% acetonitrile containing 0.1% TFA, and subjected to chromatographic separation on a TSKgel Amide-80 column (4.6 ×250 mm, particle size 5μ) using a linear gradient from 0.1% TFA in 80% acetonitrile to 0.1% TFA in 40% acetonitrile over 35 min (Dionex Ultimate 3000, Thermo). The 25 collected fractions were vacuum-dried. Samples were resuspended using 0.1% trifluoroacetic acid (TFA, Thermo Scientific, Cat#28903). The IP-HILIC fractions were individually analyzed by LC–MS/MS. The nano HPLC system used was an UltiMate 3000 HPLC RSLC nano system (Thermo Scientific) coupled to a Q Exactive HF-X mass spectrometer (Thermo Scientific), equipped with a Proxeon nanospray source (Thermo Scientific). Peptides were loaded onto a trap column (Thermo Scientific, PepMap C18, 5 mm × 300 μm ID, 5 μm particles, 100 Å pore size) at a flow rate of 25 μl/min using 0.1% TFA as mobile phase. After 10 min, the trap column was switched in line with an analytical column (Thermo Scientific, PepMap C18, 500 mm × 75 μm ID, 2 μm, 100 Å). Peptides were eluted using a flow rate of 230 nl/min and a binary 180 min gradient. The two steps gradient started with the mobile phases: 98% A solution (water/formic acid, 99.9/0.1, v/v) and 2% B solution (water/acetonitrile/formic acid, 19.92/80/0.08, v/v/v), which was then increased to 35% B over the next 180 min, followed by a gradient of 90% B for 5 min, which was finally, in a 2 min period, decreased to the gradient 95% A and 2% B for equilibration at 30 °C.

The following parameters were used for MS acquisition using an Exploris 480 instrument, operated in data-dependent mode: the instrument was operated in a positive mode; compensation voltages (CVs) used = CV-40, CV-50, CV-60; cycle time (per CV) = 1 s. The monoisotopic precursor selection (MIPS) mode was set to Peptide. Precursor isotopes and single charge state precursors were excluded. MS1 resolution = 60,000; MS1 Normalised AGC target = 300%; MS1 maximum inject time = 50 msec. MS1 restrictions were relaxed when few precursors were present. MS1 scan range = 400–1600 m/z; MS2 resolution = 45,000; MS2 Normalised AGC target = 200%. Maximum inject time = 250 ms. Precursor ions charge states allowed = 2–7; isolation window = 1.4 m/z; fixed first mass = 110 m/z; HCD collision energies = 30,33,36; exclude isotopes = True; dynamic exclusion = 45 s. All MS/MS data were processed and analyzed using Xcalibur v3.1 (Thermo), and Byonic (Protein Metrics) included in Proteome Discoverer v3.0. Two default glycan databases (Mammalian O-glycans and Human N-glycans) were used together with the latest human or mouse UniProt database to generate the glycopeptide search space. We used an in-house developed R script to filter out poorly matching spectra, considering a Byonic score higher than 200.

Statistics and reproducibility

Statistical analysis was performed using GraphPad Prism 8 software (VERSION 8.1.1, GraphPad Software Inc.). The statistical comparison between two or more groups was performed using two-sided Student’s t-test or one-way ANOVA, respectively. For the comparison of longitudinal tumor growth, two-way ANOVA was applied. Survival differences and hazard ratios were assessed using Kaplan–Meier method with log-rank (Mantel–Cox) test and Mantel–Haenszel method. Data consisting of technical or biological replicates are expressed as mean ± standard deviation (SD) or mean ± standard error of the mean (SEM), respectively, unless specified otherwise. Clinical data are expressed as violin plots or by the median. In addition, wherever possible individual values are plotted. Differences between groups were considered significant when p < 0.05. Representative images for patient-derived clinical cohorts were selected following assessment of at least 50 independent breast cancer cases or 10 independent healthy tissue samples (Fig. 1A). Representative images from mouse experiments were selected after evaluation of at least four biological replicates, all of which showed consistent results (Figs. 2E, 5F, G, and 6D, G). No images were selected based on extreme or atypical features.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

S.M. and J.M.P. designed the study and wrote the manuscript; S.M. and G.J. performed the cytometry experiments; S.M., T.O., J.H. and J.S. performed the mass spectrometry analysis and data interpretation; I.J.G., A.K.J., W.J., M.L.T., O.H.A., L.F., G.K. and L.Z. managed patient’s samples, and performed and analyzed immunohistochemistry experiments; S.M., M.F. and V.T. performed the in vivo experiments; M.A., R.G.S. and A.M.K. supported the in vivo experiments and data interpretation; M.J.K, K.W., J.J., D.H., L.E., F.H. and A.C.O. supported the experiments; S.M., L.Z. and J.M.P. discussed the data and interpreted results; All authors reviewed and approved the manuscript.

Peer review

Peer review information

Nature Communications thanks Sophia Karagiannis, Salomé Pinho and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

S.M. received funding from the European Union’s Horizon 2020 research and innovation Programme under the Marie Sklodowska-Curie grant agreement No 841319 and Austrian Science Fund (FWF, Project numbers: ESP 166 and PIN 9259824). G.J. was supported by a DOC fellowship from the Austrian Academy of Sciences. O.H.A. received funding from the Swiss National Science Foundation (SNSF, grant No P400PM_194473) and the Bruno Bloch Foundation. Work in J.M.P.‘s laboratory is supported by the Austrian Federal Ministry of Education, Science and Research, the Austrian Academy of Sciences, the City of Vienna and grants from the European Research Council (ERC Advanced grant 341036), the Austrian Science Fund (FWF Wittgenstein award Z 271-B19), the T. von Zastrow foundation, and a Canada 150 Research Chairs Program (F18-01336).

Data availability

The single cell transcriptomics data generated in this study have been deposited in the GEO database under accession code GSM9217267 and GSM9217268. The glycoproteomics data used in this study are available in the ProteomeXchange Consortium via the PRIDE partner with the dataset identifier PXD067987. The remaining data are available within the Article, Supplementary Information or Source Data file. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-73401-9.