Targeting branched N-glycans and fucosylation sensitizes ovarian tumors to immune checkpoint blockade

Aberrant glycosylation is a crucial strategy employed by cancer cells to evade cellular immunity. However, it’s unclear whether homologous recombination (HR) status-dependent glycosylation can be therapeutically explored. Here, we show that the inhibition of branched N-glycans sensitizes HR-proficient, but not HR-deficient, epithelial ovarian cancers (EOCs) to immune checkpoint blockade (ICB). In contrast to fucosylation whose inhibition sensitizes EOCs to anti-PD-L1 immunotherapy regardless of HR-status, we observe an enrichment of branched N-glycans on HR-proficient compared to HR-deficient EOCs. Mechanistically, BRCA1/2 transcriptionally promotes the expression of MGAT5, the enzyme responsible for catalyzing branched N-glycans. The branched N-glycans on HR-proficient tumors augment their resistance to anti-PD-L1 by enhancing its binding with PD-1 on CD8+ T cells. In orthotopic, syngeneic EOC models in female mice, inhibiting branched N-glycans using 2-Deoxy-D-glucose sensitizes HR-proficient, but not HR-deficient EOCs, to anti-PD-L1. These findings indicate branched N-glycans as promising therapeutic targets whose inhibition sensitizes HR-proficient EOCs to ICB by overcoming immune evasion.


REVIEWER COMMENTS Reviewer #1 (Remarks to the Author): with expertise in glycosylation, cancer immunology
In this interesting piece of work the authors investigate how N-glycan branching specifically in HR-proficient ovarian cancer augments resistance to anti-PD-L1 therapy.Using KO studies and inhibition with 2-DG this resistance could be reverted and sensitize ovarian cancer cells to checkpoint blockade.
The intricate interplay between glycosylation and anti-tumor immunity is a timely topic and holds great potential for future drug therapies.I do have a couple of issues that in my opinion require further evaluation by the authors.1.In figure 1 the authors analyze the glycome of several cancer cell lines upon tumor formation in the mouse.Although this yields clear differences I am missing the t=0 starting point.What is the basal glycosylation of these cell lines?Does that match the tumor or are specific glycan species being induced upon tumor formation?A lectin array of the basal cell lines, grown under 3D in vitro culture conditions, is therefore needed to fully comprehend what is happening in vivo.

2.
In figure 1 indeed an increase in PHA-L and PHA-E binding is observed, which the authors combinedly address as an increase in N-glycan branching.Have the authors done any MGAT3 KO studies to address the contribution of MGAT3 and does that yield similar results as for MGAT5 KO? Does 2-DG impair expression of bisecting GlcNAc and PHA-E binding?
3. I still have a problem with the use of 2-DG as a specific N-glycan modifier.Yes, the authors show that in vitro it has no effect on proliferation and glycolysis, however in vivo this might be different.Moreover, in vivo 2-DG will not only modify the cancer cells, but also affect other cell type within the tumor, such as immune cells and fibroblasts.Therefore, to pinpoint the effects truly to MGAT5 the in vivo control experiment is needed in which the MGAT5 knock down cells are used.If their hypothesis is true this should yield a similar phenotype and convey sensitivity in vivo to anti-PD-L1 treatment.

4.
Have the authors ever tried anti-PD-1 blockade?This would be a very nice control experiment, as anti-PD-1 treatment should not be affected by 2-DG.
Minor points: 1.The glycan scheme included Fig. 2A contains an error that need correction: FUT7 is depicted here as generating the Lewis X structure.This is incorrect as FUT7 will only generate sialyl-Lewis X. and HR-deficient epithelial ovarian cancer (EOC) cells using murine and human cell lines.The authors demonstrated that while fucosylation was induced by immune pressure regardless of HR-status, branched N-glycans were higher on HR-proficient cells than HR-deficient cells.
Then, the authors analyzed the relationship between MGTA5 expression which is an enzyme catalyzes branched N-glycans and BRCA1/2 expression, and demonstrated BRCA1/2 regulates MGTA5 transcription.Finally, they demonstrated 2DG that is a branched N-Glycans inhibitor in combination with anti-PD-L1 enhanced CD8+ cytotoxicity and activation in vitro and in vivo.The results are interesting and worth to further investigations for EOC patients who did not have benefits from anti-PD-L1 trials.However, the manuscript has major concerns to support the conclusion and evidence.

Point-by-point responses
We are very grateful to the reviewers for providing us with very helpful, consistent, and constructive feedback on our work, and for giving us the opportunity to revise and improve our manuscript, now entitled "Targeting Branched N-Glycans and Fucosylation Sensitizes Ovarian Tumors to Immune Checkpoint Blockade." We have conducted several new in vivo and in vitro experiments, performed new analyses, and made several modifications throughout the manuscript and its figures to address all the reviewers' concerns.We have included detailed, point-by-point responses to each of these concerns, describing the corresponding changes in our manuscript.Our responses are highlighted in blue text to facilitate the review process.

Reviewer #1
Major comments: In figure 1  Authors' response: We thank the reviewer for this suggestion and we have updated Figure 1b and 1d to include the glycomic profiles of all cell lines in vitro (t=0 before injecting to the mice).Notably, the in vitro glycomic profiles of these cell lines differ from their in vivo counterparts.This supports the notion that the glycomic profiles of these cells shift upon tumor formation and under immune pressures.These findings further support our experimental approaches where we performed glycomic profiles in vivo.

In figure 1 indeed an increase in PHA-L and PHA-E binding is observed, which the authors combinedly address as an increase in N-glycan branching. Have the authors done any MGAT3 KO studies to address the contribution of MGAT3 and does that yield similar results as for MGAT5 KO? Does 2-DG impair expression of bisecting GlcNAc and PHA-E binding?
Authors response: Following the reviewer's suggestion, we conducted new experiments to knock down MGAT3 in HR-proficient ovarian cancer cells (OVCAR3) (Extended Data Figure 3a).Our new results show that MGAT3 knockdown did not decrease binding to PHA-L lectin (Extended Data Figure 3b) as did MGAT5 knowck-down (as illustrated in Figure 3B).Additionally, we investigated the effects of 2DG on the binding of both PHA-L (Figure 5a) and PHA-E (Figure 5b), finding that 2DG treatment lessened binding of cancer cells to both lectins.

I still have a problem with the use of 2-DG as a specific N-glycan modifier. Yes, the authors show that in vitro it has no effect on proliferation and glycolysis, however in vivo this might be different. Moreover, in vivo 2-DG will not only modify the cancer cells, but also affect other cell type within the tumor, such as immune cells and fibroblasts. Therefore, to pinpoint the effects truly to MGAT5 the in vivo control experiment is needed in which the MGAT5 knock down cells are used. If their hypothesis is true this should yield a similar phenotype and convey sensitivity in vivo to anti-PD-L1 treatment.
Authors response: To isolate the effect of 2DG on other cell types, we performed new in vivo experiments to knock down MGAT5 in HR-proficient mouse ovarian cancer cells (KPCA).Subsequently, we injected both MGAT5 knockdown KPCA and control KPCA to induce tumor formation in immune-competent mice.As depicted in Figure 6a, we observed a phenotype similar to that of 2DG treatment; while MGAT5 knockdown alone did not affect tumor weight, its knockdown also sensitized KPCA tumors to anti-PD-L1 treatment, consistent with our hypothesis.

Have the authors ever tried anti-PD-1 blockade? This would be a very nice control experiment, as anti-PD-1 treatment should not be affected by 2-DG.
Authors response: Following the reviewer's suggestion, we have performed new in vivo experiment using mice injected with the HR-proficient mouse ovarian cancer cells (KPCA) and treated with either anti-PD-1 blockade or anti-PD-L1 blockade, in the presence or absence of 2DG.As shown in the new Figure 7d, while 2DG affected both anti-PD-1 blockade and anti-PD-L1 blockade, the effects were more pronounced with the anti-PD-L1 blockade.Considering anti-PD-1 blockade is not 100% effective in blocking all PD-1/PD-L1 interactions, 2DG could enhance the effects of both anti-PD-1 and anti-PD-L1.However, as predicated by the reviewer, it should enhance anti-PD-L1 more than anti-PD-1 (which is what we observed).This is because, in the case of anti-PD-1, it only affects the residual PD-1/PD-L1 binding not blocked by anti-PD-1.
Whereas with anti-PD-L1, its effects should be more generalized.
Minor comments: 1.The glycan scheme included Fig. 2A contains an error that need correction: FUT7 is depicted here as generating the Lewis X structure.This is incorrect as FUT7 will only generate sialyl-Lewis X.
Authors response: We thank the reviewer for noticing this oversight and have fixed it.
Authors response: We have now added Extended Data Figure 7, where we show the expression of PD-L1 and the binding to PD-1 of multiple HR-deficient and HR-proficient cells upon IFNγ stimulation.As shown, while HR-proficient cells exhibit similar levels of PD-L1 expression to HR-deficient cells (including PEO1 cells), the binding of these cells to PD-1 trends higher.This supports the notion that qualitative differences of these cells (such as glycosylation) impact the binding of PD-L1 to PD-1 and thus can affect their signaling.
3. In Fig. 6F a clear shift in populations/clusters is seen in the 2-DG+PD-L1 group.What are these different populations/clusters and why is there a shift upon 2-DG+PD-L1?How was this UMAP generated?This was unclear to me from the text.
Authors response: To address this question, we are detailing our analysis method for these data below:

"A combination of unsupervised clustering using Seurat's K-nearest neighbour algorithm and cell-type prediction using SingleR was used to assign cell types to the immune cells. UMAPs were used for visualization of clusters (Panel A in the figure below). The T and NKT cell cluster was subset and re-clustered to further resolve T cells into CD4+ T cells (including FoxP3+ T-regs) and CD8+ T cells (memory and effector) (Panel B in the figure below). The control sample had significantly fewer T cells than others. Gene expression
for IFNG was then visualized as a density plot computed using kernel densities across samples and was found to be concentrated in the CD8+ T cell region, particularly enriched in the 2DG+PDL1 sample (Panel C in the figure below)." Reviewer #2 Major comments: 1.Although the authors showed BRCA1/2 directly regulates MGAT5 expression, the correlation of MGAT5 and BRCA1/2 expression analyzed by CCLE (Fig. 2g) and TCGA (Fig. 2h) were very weak (rho < 0.3).Other mechanisms should be discussed.
Authors' response: To address this concern, we conducted new experiments wherein we inserted the MGAT5 promoter into the pGL4.10[luc2]vector.We then utilized the Dual-Luciferase® Reporter Assay System to determine if BRCA1 knockdown affects the expression of the Firefly Luciferase signal, which is regulated by the MGAT5 promoter in this system.We found that BRCA1 knockdown indeed significantly decreases the Firefly Luciferase signal in OVCAR3.The restoration of full-length BRCA1 can reverse this decrease, but the restoration of truncated BRCA1 without BRCT1&2 domains fails to do so.The BRCT1&2 domains are located in the C-terminus of BRCA1, functioning to regulate transcription.This suggests that MGAT5 expression is indeed regulated by BRCA1 at the transcriptional level.These data are in new Figure 4f-g.
Additionally, we investigated whether BRCA1's DNA repair function is involved in promoting MGAT5 expression.To activate BRCA1's DNA repair function, DNA damage was induced by cisplatin in HR-proficient OVCAR3 cells.Indeed, we observed an increase in the formation BRCA1 foci, a marker of its activation in DNA repair.Notably, MGAT5 expression is reduced in cisplatin-treated OVCAR3 cells (Figure 4i).Thus, BRCA1's role in promoting MGAT5 is not due to activation of its role in DNA damage repair.Authors' response: There are several key differences in our focus and approach compared to those of Okada et al., Cell Reports, 2017.In their study, Okada et al. focused on the fucosylation of T cells, treating them in vitro with 100 µM of 2FF for 6 days.Subsequently, these 2FF-treated T cells were injected into tumor-bearing mice, followed by anti-PD-L1 immunotherapy, but the mice were not directly treated with 2FF.In contrast, our study focuses on the glycosylation of cancer cells themselves.We pre-treated cancer cells with 2FF for 3 days, then injected these cells into mice, and also administered 2FF directly to the mice (oral gavage at a dose of 17.5 mg/kg every 2 days).This approach, similar to that used by Yun Huang et al. in Nature Communications, 2021 (PMID: 33976130), where the investigators used this approach to suppress the progression of breast cancer.Our manuscript focuses on the inhibition of fucosylation of cancer cells, as opposed to the T-cell focus in Okada et al.'s study, which could account for the observed differences.Additionally, the variance in cancer models used in these studies could be another key factor, as the microenvironments and the immune inhibition caused by different tumors can significantly differ.In our study, we focused on EOCs, particularly comparing HR-proficient versus HR-deficient EOCs.

Some of the data are insufficient to conclude branched N-glycans regulates immune checkpoint sensitivity on HR-proficient EOC. For example, binding of PHA-L was only shown on HR-proficient cells, but not HRdeficient cells. PD-L1 expression level between HR-proficient and HR-deficient cells are not present.
Authors response: For the PHA-L level, Figure 1D shows that PHA-L binding is higher in two HR-proficient tumors than in two HR-deficient tumors.We have now added Extended Data Figure 7, where we show the expression of PD-L1 and the binding to PD-1 of multiple HR-deficient and HR-proficient cells upon IFNgamma stimulation.As shown, while HR-proficient cells exhibit similar levels of PD-L1 expression to HRdeficient cells (including PEO1 cells), the binding of these cells to PD-1 is higher.This supports the notion that qualitative differences of these cells (such as glycosylation) impact the binding of PD-L1 to PD-1 and thus can affect their signaling.
Minor comments: 1. Tumor volume of control and most of PD-L1 treatment group in Figure 2g-h and Figure 6b-d looks similar.The experiments were simultaneously performed for 2-DG and 2-FF treatment under the same conditions although it was not stated?

2 .Reviewer # 2 (
Fig. 5C is missing the PE01 (HR-deficient) control experiment.3.In Fig. 6F a clear shift in populations/clusters is seen in the 2-DG+PD-L1 group.What are these different populations/clusters and why is there a shift upon 2-DG+PD-L1?How was this UMAP generated?This was unclear to me from the text.Remarks to the Author): with expertise in ovarian cancer, cancer immunology This study by Nie, et al. investigated the difference of glycosylation between HR-proficient

2 .
Fig. 1d, PHA-E expression on HR-proficient ID8 looks similar to HR-deficient cell lines.3. Why cytotoxicity of CAR against PEO1 was not increased by anti-PD-L1 treatment in Fig. 5g?It should behave like PEO4 with MGAT5 shRNA?The base level of cytotoxicity against different cancer cells should be included in the extended Figure.

Reviewer # 3 (
Remarks to the Author): with expertise in ovarian cancer, cancer immunology Nie et al. describe a new method to treatment homologous recombination (HR)-proficient epithelial ovarian cancers (EOCs).Their main finding is that BRCA1/2 transcriptionally promotes the expression of MGAT5, the enzyme responsible for catalyzing branched Nglycans in HR-proficient tumors and augment their resistance to anti-PD-L1 by enhancing the interaction of PD-L1 and PD-1 between HR-proficient tumors and CD8+ T cells.They also show that inhibiting branched N-glycans with 2-Deoxy-D-glucose can sensitize HR-proficient, but not HR-deficient to anti-PD-L1 therapy.The findings are interesting and novel, the major conclusions are supported by their data.Overall, this study elucidates a new resistant mechanism and may bring new insights into the treatment of HR-proficient ovarian tumors.1. Reporter assays using the MGAT5 promoter is required to further evaluate the transcriptional regulation of BRCA1 on MGAT5 in HR-proficient tumor cells.Which domain of BRCA1 is responsible for the regulation?In Figure 4d, the level of Branched N-glycans needs to be detected in both control and BRCA1/2-knockdown cells.2. In figure 6f and 6g, the IFN-γ expression in CD8+ T cells needs to be assessed by flow cytometric analysis.It is better to include PD-L1 and PD-1 expression analysis in the experiments.3. BRCA1 and BRCA2 play an integral role in homologous recombination repair of doublestrand DNA breaks.Is the DNA repair function of BRCA1/2 involved in regulating expression of MGAT5 in HR-proficient ovarian cancer cells?Reviewer #4 (Remarks to the Author): with expertise in glycosylation, cancer immunology Nie and colleagues report on the role of glycosylation in epithelial ovarian cancer cells.By using mouse models of HR-proficient and -deficient EOC in immunodeficient and -sufficient mice, the authors show increased fucosylation in immunocompetent mice and differences in branched N-glycans in HR-proficient models using a lectin array.The authors then show inhibition of fucosylation led to an improved tumor control and inhibition of N-glycan branching with 2-deoxy-d-glucose led to improved efficacy of PD-1 inhibition.This is a very interesting manuscript, although a few questions should be addressed before publication of the manuscript.1.How is fucosylation influencing tumor growth?What are the mechanisms involved?What happens with other glycans upon 2FF treatment?2. Do the authors observe any changes in immune cells in the tumors treated with 2FF? 3. What is effect of 2DG on other glycans?What is the effect on immune cell infiltration?Are there any changes on immune infiltrats? 4. What is the general role of N-glycans in the in vitro models?Could enzymatic treatment be used? 5. 'Targeting glycomes' is a strange title.Could the authors be more specific (fucosylation, Nglycan branching?
the authors analyze the glycome of several cancer cell lines upon tumor formation in the mouse.Although this yields clear differences I am missing the t=0 starting point.What is the basal glycosylation of these cell lines?Does that match the tumor or are specific glycan species being induced upon tumor formation?A lectin array of the basal cell lines, grown under 3D in vitro culture conditions, is therefore needed to fully comprehend what is happening in vivo.