A signature based on glycosyltransferase genes provides a promising tool for the prediction of prognosis and immunotherapy responsiveness in ovarian cancer

Ovarian cancer (OC) is the most lethal malignancy of the female reproductive system, the incidence of which is only secondary to cervical cancer and uterine corpus cancer. According to the latest global cancer statistics of 2020, ovarian cancer accounts for approximately 3.4% of 9.2 million new cancer cases in females and 4.7% of mortality [1]. Due to the insidious onset of ovarian cancer and lack of effective screening tools, nearly 70% of patients are not diagnosed until the advanced stage (stage III or IV) [2, 3]. The 5-year survival rate of advanced patients is lower than that of patients who are diagnosed at an early stage [4]. Although much progress has been achieved in non-traditional treatment, such as immunotherapy and targeted therapy, the reduction in mortality and recurrence still remains limited [5, 6]. Similar to other malignant tumours, heterogeneity exists widely across subtypes or even within a single tumour, which may result in no response to corresponding treatments [7‐9]. Therefore, there is an urgent need to identify novel molecular signatures to predict prognosis and evaluate the sensitive subpopulations of immunotherapy, which contribute to the improvement of treatment success and the development of precision medicine.

Posttranslational modifications (PTMs) of proteins are often dysregulated in cancer. The specific function of a protein is dynamically achieved by the catalytic action of many enzymes involved in PTMs, suggesting that these enzymes may provide clues for cancer research [10, 11]. Glycosylation is one of the most common PTMs of proteins and plays a vital role in many critical biological processes, including cell adhesion, growth, signal transduction and immune response, by affecting the function of modified proteins [12, 13]. The biosynthesis of glycosylation is a complex process orchestrated by several glycosyltransferases and glycosidases. It has been widely recognized that abnormal glycan changes in proteins are involved in many pathological states, such as viral infection, cancer progression and the inflammatory process [14]. Aberrant glycosylation is considered a marker of cancer, the main factor of which is the abnormal expression of glycosyltransferases (GTs) translated from corresponding GT genes [15, 16].

Numerous studies have proven that altered expression levels of GTs can directly affect the malignant phenotypes of cancer, such as proliferation [17], metastasis [18] and drug resistance [16], indicating that targeting GTs may help us understand the role of aberrant glycosylation in cancer pathogenesis [19]. In ovarian cancer, mounting evidence has suggested that GTs played a critical role in its malignant progression. For example, overexpression of sialyltransferase ST3GAL1 was proven to promote progression and paclitaxel resistance in OC [20]. Elevated α1,3-mannosyltransferase 3 (ALG3) was reported to promote peritoneal metastasis of OC through increasing interaction of α1,3-mannosylated uPAR and ADAM8 [21]. Huang et al. demonstrated that glycosyltransferase 8 domain containing 2 confers CDDP (cis-dichlorodiammine-platinum) resistance through the FGFR/PI3K signalling axis [22]. What’s more, a recent study has proved that GALNT14 was significantly upregulated in OC and regulated ferroptosis through the EGFR/mTOR pathway [23]. However, to our knowledge, a systematic analysis of GT genes in OC is still blank. Since GTs could affect the prognosis of OC through induction of malignancy phenotypes, discovering the GT-gene signature for risk stratification is in demand. In addition, it is worth mentioning that glycosylation is not only under the control of epigenetic regulation (DNA methylation, histone acetylation and noncoding RNAs) but also itself is an epigenetic modifier of  histones that participates in cancer progression [24, 25]. Given that cancer cells can hijack various existing epigenetic modifications, including glycosylation, to modulate antitumour immunity and lead to tumour escape, epigenetic signatures appear to be promising candidates for predicting the outcomes of immunotherapy [26]. For example, sialic acid sugars on the surface of cancer cells are recognized as potent immune modulators that contribute to the immunosuppressive microenvironment and tumour immune evasion [27]. Therefore, these studies provide a rationale for the potential ability of GT-signature to predict immunotherapy success, contributing to personalized medicine.

In this study, we performed a systematic analysis to determine the expression pattern of glycosyltransferase genes between normal ovarian and OC tissues and determine whether OC could be stratified by these GT gene expression profiles. The flow chart of this study is shown in Fig. 1. By using bioinformatic analysis of the samples from The Cancer Genome Atlas (TCGA) and Genotype-Tissue Expression (GTEx) projects, we established a risk signature of six GT genes that was able to predict the prognosis of OC and was further validated in the GEO database. We then studied the correlation between this risk signature and the tumour microenvironment and immunotherapy. Moreover, we performed a knockdown assay at the cellular level to verify the function of B4GALT5 in ovarian cancer, which was upregulated in ovarian tissues and associated with poor prognosis. The pathways involved were predicted by corresponding bioinformatic analyses.

Glycosylation, as an essential type of PTM, has aroused increasing interest in the academic community with recent advances in analytical technologies [28]. Although the intimate connection between glycosylation and cancer was detected decades ago [29‐31], it has not been systematically researched in regards to GTs in OC until now. In the context of precision medicine, constructing a risk signature based on GT genes by means of machine learning made up for the insufficiency of one single factor to stratify OC patients. And we discussed the possibility that GT stratification together with different immune microenvironment will elucidate the poor prognosis of OC.

FUT8 is an α-1,6-fucosyltransferase that take part in the core fucosylation of N-glycans and has a substrate specificity toward biantennary complex N-glycan oligosaccharide [32].But the sialylation of the N-glycans could reduce their activity as a substrate of FUT8 [33]. It was reported that there existed an interplay among FUT8, GnT-IV (MGAT4), and GnT-V (MGAT5) in N-linked glycosylation [34]. FUT8 was reported to be highly expressed in a variety of cancers, including lung, colorectal, ovarian, prostate, breast, melanoma and so on, and was associated with the prognosis of lung cancer, colorectal cancer and prostate cancer [15]. Overexpression of FUT8 can suppress the immune response in triple-negative breast cancer by mediating the abnormal N-glycosylation of B7H3, which may account for the lack of response to anti-PD1/PDL1 immunotherapy in triple-negative breast cancer patients [35]. In ovarian cancer, FUT8 activates the hyper core fucosylation of copper transporter 1 to suppress cisplatin uptake into OC cells [36]. However, FUT8 was found to be expressed at lower levels in osteosarcoma, leading to lower core fucosylation levels of TNF receptors. Lower fucosylation of TNF receptors decreased mitochondria-dependent apoptosis by activating NF-κB2 signalling [37]. Interestingly, our study revealed that FUT8 was upregulated in OC tissues but was a good prognostic factor. Similarly, ST6GAL1 was also overexpressed in OC but associated with favourable prognosis. The result was inconsistent with an earlier study where high ST6GAL1 protein expression in OC was significantly associated with poor prognosis [38]. ST6GAL1 is a β-galactoside α-2,6-sialyltransferase that catalyses α2,6-sialylation of N-glycans. It has been proved that FUT8-mediated core fucosylation of IgG limited the B cell-mediated IgG sialylation catalysed by ST6GAL1 [39, 40]. In view of the tight interaction between core fucosylation and α2,6-sialylation, we reasonably assume that the high sialylation level of the N-glycans in OC may affect the oncogenic function of FUT8. It is worth noting that ST6GAL1 was also found to be upregulated in many solid tumours [41‐45] and promote their malignant phenotype by regulating the sialylation of signalling pathways [46]. What’s more, an increasing number of studies about the function of ST6GAL1 in immune enhancement have been published [47‐51]. Thus, we also infer that the positive response to antitumour immunity may be another reason for its protective function in OC prognosis.

ST8SIA3 is another member of sialyltransferase family which is mainly involved in the sialylated glycolipids, while ST6GAL1 preferentially links α-2,6-linked sialic acid to Galβ4GlcNAc chains, usually present in N-linked chains [52]. Unlike ST6GAL1, there are relatively few studies focusing on the relationship between ST8SIA3 and cancer. ST8SIA3 was proved to participate in the synthesis of A2B5 epitope and was critical for A2B5 immunoreactivity in glioblastoma, indicating the potential of neuraminidase treatment [53]. Apart from that, other studies on ST8SIA3 were mainly about nervous system diseases and antiviral immunity [54, 55]. Interestingly, B4GALT5, another risk factor in our signature, was also reported to involve in the biosynthesis of lactosylceramide [56]. These results inspired us that the further study on the function of glycolipids in OC is promising. Except for glycolipids, B4GALT5 also acts with high preference on substrate that contain the GlcNAc beta1– > 6GalNAc structure which is found in mucin type O-linked core 6 glycan [57]. In our risk model, B4GALT5 was the GT gene with the highest coefficient of 0.7 and was the only independent risk factor, indicating that high levels of B4GALT5 might exert a stimulative effect on OC. In fact, many studies have elucidated its important role in many other tumour types. In colorectal cancer, B4GALT5 was considered as a diagnosis/prognostic biomarker with huge application potential which could be detected by electrochemical immunosensor [58]. In 2020, Tang et al. verified that B4GALT5 modulates the stemness of breast cancer through glycosylation modification to stabilize Frizzled-1 and activate Wnt/β-catenin signalling independent of its cell surface location [59]. A bioinformatic analysis concluded that B4GALT5 is one of the most consistently malignancy-associated enzymes using the transcriptomic data of the 21 TCGA cohorts [60]. Another recent bioinformatic analysis has drawn a similar conclusion as us regarding the negative role of B4GALT5 in hepatocellular carcinoma (HCC) prognosis and its vitro experiments also demonstrated that the knockdown of B4GALT5 in HCC cells was able to inhibit proliferation and metastasis [61], which is consistent with our in vitro experiments conducted on OC cells. In fact, about 20 years ago, researchers have found that beta1,4-galactosyltransferase was likely to be a biomarker in the monitoring OC patients when the serum CA 125 level is normal [62]. Taken together, B4GALT5 may be a promising starting point to help the mechanistic research of OC from the view of glycosylation.

GCNT2 is an initiated enzyme for the synthesis of I-branched glycan which is a cancer-associated glycan [63]. Loss of GCNT2 expression would lead to the ineffective I-branch conversion, thus regulating cancer progression. For example, high expression level of GCNT2 and its I-branched glycan product was proved to accelerate epithelial-to-mesenchymal transition in colon cancer [64]. GCNT2 has been widely studied in melanoma, the loss of which brings corresponding loss of I-antigen and thus enhances melanoma growth and metastasis [65]. Importantly, many studies have pointed out that the high level of I-antigen synthesized by GCNT2 could increase the susceptibility of malignant cells against immune cells in leukemia [66, 67]. This kind of immune regulatory may account for why GCNT2 was upregulated in OC but a good prognostic predictor in our study. What’s more, due to the role of β4GalTs in I-antigen capping, it’s really something to see that whether there exists an interaction between GCNT2 and B4GALT5. ALG8 is an alpha-1,3-glucosyltransferase and ALG8-CGD (congenital disorders of glycosylation) is a widely studied monogenic disorder of glycosylation that involves multisystem disorders [68]. In malignancies, ALG8 was a variate of a risk predictive model established for gastric cancer [69] and ovarian cancer [70]. However, consistent with Zhao et al.’s study [70], ALG8 in our study was overexpressed in OC tissues and led to favourable outcomes. In summary, except for ALG8, other five GTs are all key enzymes that involves in the synthesis of cancer associated glycans and may exist a close tie between each other.

Subsequently, the functional enrichment analyses demonstrated that DEGs between the two risk groups were correlated with immune responses and epithelial maintenance, supporting our speculations above. Ultimately, the immune cell infiltration in the low- and high-risk groups was compared, and we found that the low-risk group had increased infiltrating levels of antitumour immune cells in the TME, such as activated DCs [71, 72], B cells [73, 74], activated T cells [27, 75], and Type 17 T helper cells. In OC, T cells and B cells are associated with superior prognosis in ovarian cancer [76, 77]. High infiltration of type 17 T helper cells and DCs can also improve the outcomes of OC patients [78]. Simultaneously, high-risk patients had an immunosuppressive microenvironment with the presence of numerous CAFs and M2 TAMs. There is a popular belief that CAFs and M2 TAMs in the TME of OC are facilitators of carcinogenesis, tumour progression and metastasis, as well as therapeutic resistance and immunosuppression, leading to worse survival of OC patients [79‐81]. Therefore, the GT-based signature established in our study is accurate enough to predict the OS of OC patients due to the consistent results with previous publications. Notably, GTs may provide a link to explore the crosstalk between cells in the TME and OC cells.

The highlight of our study is that the underlying role of this signature allows the prediction of patients who may benefit from ICB therapy. Genomic instability with accumulation of somatic mutations is widely believed to be associated with cancer cell phenotype shaping, thereby leading to different response to immunotherapy [82]. Tumour mutation burden (or the number of somatic mutations), which can give rise to neoantigens as targets of tumour immunity, has been explored as a promising biomarker for ICB therapy [83, 84]. Patients with high tumour mutation burden is more likely to benefit from ICB therapy and have an improved survival [85].Although the recognition of neoantigens is thought to be a random process, each somatic mutation is able to improve the opportunities for the immune system to recognize and attack cancer cells during ICBs [86, 87]. Our results showed that the number of somatic mutations in the low-risk group was higher than that in the high-risk group. Although not statistically significant, the neoantigens in low-risk group was higher, which, according to Lang F et al., may provide tumour-specific neoepitopes for individual therapeutical cancer vaccines [86]. We found that a total of 19 genes have significantly different mutational frequencies between two subgroups, which may provide supplementary targets for patients with poor response to ICB therapy. The high-risk group exhibited a higher TIDE score and dysfunction score than the low-risk group. The greater the TIDE score is, the less benefit patients achieve from ICB due to the more frequent existence of immune escape [88]. Meanwhile, according to the median TIDE score as the threshold of responders, the proportion of ICB responders was significantly higher in the low-risk group than in the high-risk group. Taken together, these results illustrated that the risk signature may be helpful in screening suitable candidates who would benefit more from ICBs.

The current study of GTs is still in its initial stages, especially regarding their mechanism in OC. Although our experiments have proven that B4GALT5 can regulate the progression of OC, the critical molecular mechanism is still unclear. Much effort will be put into probing the intrinsic mechanism, especially in determining whether B4GALT5 regulates ovarian cancer progression through abnormal glycosylation, in our subsequent studies. What is important to note is that although this signature was validated in an external dataset, formalin-fixed paraffin-embedded specimens and a multicentre prospective study are needed to confirm our findings.

In conclusion, we comprehensively studied the prognostic value of GT genes in OC and provided a theoretical foundation for future research. An effective prognostic signature was established based on six GT genes, and the risk score was an independent risk factor for OS prediction. Encouragingly, our results revealed that this signature may be a predictor of ICB response, contributing to the advancement of precision medicine. Moreover, the preliminary experiment revealed the promoting function of B4GALT5 in OC progression, and bioinformatic analysis predicted the likely pathways involved, paving the way for our follow-up study.