The different prognostic significance of polysialic acid and CD56 expression in tumor cells and lymphocytes identified in breast cancer

Protein glycosylation, the attachment of carbohydrates onto proteins, is a fundamental process that alters the biological activity of proteins. Changes to glycosylation states are associated with many forms of cancer including breast cancer. Through immunohistological analysis of breast cancer patient tumors, we have discovered the expression of an atypical glycan—polysialic acid (polySia)—in breast cancer. Notably, we have identified polySia expression in not only tumor cells but also on tumor-infiltrating lymphocytes (TILs) and our study reveals ST8Sia4 as the predominant polysialyltransferase expressed. Evaluation of ST8Sia4 expression in tumor cells identified an association between high expression levels and poor patient outcomes whereas ST8Sia4 expression in infiltrating stromal cells was associated with good patient outcomes. Investigation into CD56, a protein known to be polysialylated, found CD56 and polySia expression on breast tumor cells and TILs. CD56 expression did not positively correlate with polySia expression except in patient tumors which expressed HER2. In these HER2 expressing tumors, CD56 expression was significantly associated with HER2 expression score. Evaluation of CD56 tumor cell expression identified a significant association between CD56 expression and poor patient outcomes. By contrast, CD56 expression on TILs was significantly associated with good clinical outcomes. Tumors with CD56+ TILs were also consistently polySia TIL positive. Interestingly, in tumors where TILs were CD56 low-to-negative, a polySia+ lymphocyte population was still identified and the presence of these lymphocytes was a poor prognostic indicator. Overall, this study provides the first detailed report of polySia and CD56 in breast cancer and demonstrates that the prognostic significance is dependent on the cell type expression within the tumor.


INTRODUCTION
Glycans are an essential part of all living cells, highlighted by the estimate that half of all human proteins are glycosylated. It is well recognized that glycans have major structural roles and mediate an extensive array of intrinsic and extrinsic actions in biological systems 1 . The generation of glycans through cellular glycosylation is a tightly controlled process involving the coordinated actions of specialized enzymes such as glycosyltransferases. Changes to protein glycosylation states are associated with several diseases and altered glycosylation is a notable trait in many forms of cancer including breast cancer [2][3][4][5][6][7] . There is clear significance for glycans in normal physiology and altered expression in cancer. The unveiling of alterations to the glycan content of cancer, including breast cancer, is a rapidly advancing and evolving field with great potential for therapeutic application 5,8,9 . Glycosylation changes associated with oncogenic transformation of cells can arise through transcriptional and metabolic reprogramming resulting in the neosynthesis of specific glycoepitopes. Changes in the glycan composition of a neoplastic cell are associated with multifarious changes in cancer from promoting tumor proliferation to metastasis and immune evasion 4,10,11 . Increased sialylation is a notable trait of many cancer types and promotes tumor progression and, in breast cancer, elevated sialylation promotes tumor progression 12,13 . Polysialylation is a selective and highly regulated glycosylation event generating long glycan chains composed of α2,8-linked sialic acid residues, termed polysialic acid (polySia). Synthesis of polySia is performed by Golgi-localized sialyltransferase ST8Sia2 or ST8Sia4 14 . In healthy adults, polySia expression is restricted to a limited number of cell types and protein carriers 15 . Classically, polySia has been documented on neuronal cells, but it is now recognized that polySia also occurs in the context of the immune system. PolySia has been found on both human and murine leukocytes 16 . Specifically, in human leukocytes polySia has been found on natural killer (NK) cells and CD3+ T cells 16,17 . CD56 (also known as neural cell adhesion molecule, NCAM-1) was identified as a carrier protein for polySia chains on NK and CD3+ T cells. In the context of cancer, neoexpression of polySia has been documented in multiple cancers, such as neuroblastoma, glioma, lung carcinoma, and leukemia 15,18 . Overexpression of polySia in cancer cells is suggested to increase migration and invasion, and in clinical biopsy specimens from glioblastoma and non-small cell lung cancer (NSCLC) polySia expression strongly correlated with metastasis and poor prognosis [19][20][21] . CD56 was identified as the carrier protein of polySia chains in glioblastoma. Interestingly however, in NSCLC, polySia-positive tumors were not always CD56 positive suggesting the presence of additional polysialylated proteins in cancer. PolySia has clear links to cancer progression and was ranked as the second highest priority glycan for investigation in cancer by a National Cancer Institute pilot project 22 . Here, we set out to examine the profile of expression and clinical relevance of polySia, together with CD56, and polysialyltransferases ST8Sia4 and ST8Sia2, in primary breast tumors.

Polysialic acid expression in primary breast tumors
To explore polySia expression in breast cancer, immunohistochemistry was performed on normal and adjacent normal breast tissue and primary breast cancer tumors. PolySia expression was evaluated using one TMA containing n = 65 healthy/normal breast tissue samples and two TMAs containing primary breast cancer tumors. Tumors were grouped into molecular subtypes based on the expression of ER, PR and HER2 (see Supplemental Fig. 1 for breakdown of all breast cancer TMAs, tissue sections, associated clinical data, corresponding experiments used, and molecular subtype grouping). The two breast cancer TMAs contained a total of n = 144 cases from primary breast tumors from which n = 123 contained data on all three (ER, PR, and HER2) receptor expression levels. These 123 cases were categorized further based on receptor expression as Luminal A (n = 60), Luminal B (n = 9), HER2+ (n = 12), and triple-negative breast cancer (TNBC) (n = 42). PolySia expression levels were evaluated for each molecular subtype (n = 123) and healthy/adjacent normal breast tissue (n = 65). In normal breast tissue polySia expression was low whereas polySia expression was significantly increased in all molecular subtypes of breast cancer ( Fig. 1A and B). No significant difference in polySia expression was found between subtypes and all subtypes exhibited a wide-range of polySia expression from low expression to high expression ( Fig. 1A and B). Further analysis based on individual receptor expression (ER, PR, HER2) did not identify any significant differences between polySia levels and receptor expression (Supplementary Data Fig. 2A-C). Interestingly, we also noted polySia expression on tumor infiltrating lymphocytes (TILs) (Fig. 1C). CD3 and polySia immunostaining of serial sections found that cells expressing polySia and CD3 colocalized in the same regions of the tumor microenvironment (Fig. 1D). To further confirm polySia expression by CD3+ lymphocytes dual fluorescence immunostaining was performed which confirmed that CD3+ lymphocytes were positive for polySia (Fig. 1E). Analysis of polySia expression levels on TILs relative to molecular subtype (Luminal A, n = 36; Luminal B, n = 6; HER2+, n = 8; TNBC, n = 31) demonstrated a trend towards increased polySia TIL levels in HER2+ and TNBC but this was not found to be significant (Fig. 1F). Further analysis of TIL polySia expression identified significantly higher levels of polysialylated TILs in ER weak/negative tumors compared to tumors with strong ER expression (Fig. 1G, H and I).
No association was found with HER2 or PR expression (Supplementary Data Fig. 2D and E). PolySia expression on tumor cells and TILs did not significantly associate with tumor Grade, Stage, or patient age (Supplementary Data Fig. 2F-K). PolySia antibody specificity was validated through western blotting and immunohistochemistry on cell line generated tumor xenografts (Supplementary Data Fig. 3). Overall, these results demonstrate that polySia is expressed in breast tumors and present on tumor cells and also tumor-infiltrating lymphocytes.
Sialyltransferase, ST8Sia4 and ST8Sia2, expression in primary breast tumors PolySia is exclusively synthesized by either sialyltransferase (ST) 8Sia4 or 2 (ST8Sia4 and ST8Sia2). To evaluate polysialyltransferase expression levels in breast tumor cells and TILs, in situ hybridization (ISH) was performed on tumor tissue cores and probed for ST8Sia2 and ST8Sia4 ( Fig. 2A). Cores were evaluated for ST8Sia4 and ST8Sia2 expression in tumor cells and in infiltrating stromal cells. For both tumor cells and infiltrating stromal cells, ST8Sia4 was the predominant polysialyltransferase expressed whereas ST8Sia2 expression was significantly lower in tumor cells and infiltrating stromal cells relative to ST8Sia4 (Fig. 2B-D). ST8Sia4 expression in tumor cells and infiltrating stromal cells was not found to associate with any molecular subtype ( Fig. 2E and F). ST8Sia4 tumor cell expression was found to significantly associate with a higher tumor Grade (Fig. 2G). Grade 1 tumors had significantly lower ST8Sia4 expression levels relative to Grades 2 and 3 tumors. Similar results were also found for infiltrating stromal cells as increased ST8Sia4 expression significantly associated with a higher tumor Grade (Fig. 2H). No significant differences were identified between ST8Sia4 expression and Stage and no significant correlation was found with age ( Supplementary  Data Fig. 4). ST8Sia4 RNA probe was validated using cell line generated tumor xenografts (Supplementary Data Fig. 5).
Next, to evaluate if ST8Sia4 expression correlates with polySia expression, we plotted total polySia expression relative to ST8Sia4 expression for each matched tumor core. No significant correlation between ST8Sia4 expression and polySia expression was identified for tumor cells ( Fig. 3A; n = 93) and TILs ( Fig. 3B; n = 70). We found that some tumors had high ST8Sia4 expression but lacked polySia expression (Fig. 3C) whereas other tumors displayed both ST8Sia4 expression and polySia expression (Fig. 3D). A representative image for a ST8Sia4 and polySia low/negative tumor is also shown (Fig. 3E). Potentially, the lack of a correlation between ST8Sia4 and polySia expression is a result of RNA expression not reflecting protein expression, or potentially it may be related to the presence or absence of a polySia carrier protein(s).

Prognostic impact of ST8Sia4 and polysia expression in breast tumor cells
To determine the prognostic potential of ST8Sia4 and polySia in breast tumor cells, ST8Sia4 and polySia expression levels were evaluated and correlated to patient overall survival in a cohort of invasive ductal carcinoma (IDC) cases with 10 years overall survival data (n = 136). Consecutive tissue microarray (TMA) sections were analyzed for polySia and ST8Sia4 expression. For ST8Sia4 analysis n = 93/136 tumor cores were amenable to analysis (exclusion was based on core loss and tissue deformation which prevented Intellesis analysis). Elevated ST8Sia4 expression in tumor cells was significantly associated with poor overall survival compared to tumors with low expression of ST8Sia4 (p = 0.04; HR [High/ Low] = 2.47, 95% CI: 1.03-5.95; High [n = 23], Low [n = 70]) ( Fig. 4A and B). Evaluation of polySia levels in tumor cells found no significant difference in survival outcomes for individuals with moderate to low polySia expression compared to individual with high expression (Fig. 4C). While polySia expression levels did not significantly correlate with patient outcomes over a 10-year period (p = 0.13; HR = 1.74, 95% CI: 0.85-3.60), we did note that individual with high polySia levels had a trend to worse outcomes, particularly within the initial 5-year period (p = 0.060) (5-year HR = 2.56, 95% CI: 0.96-6.42).
Prognostic impact of ST8Sia4 and PolySia expression in tumor-infiltrating lymphocytes Next, we sought to evaluate the prognostic significance of ST8Sia4 and polySia expression in TILs. Breast tumor cores used to evaluate tumor cell ST8Sia4 and polySia levels, were analyzed for ST8Sia4 expression and polySia expression levels in TILs. Consecutive TMA sections were analyzed for polySia and ST8Sia4 RNA expression. Analysis of ST8Sia4 expression in TIL-positive tumors (n = 84) was limited to 48 cores (exclusion was based on core loss, tissue damage, or tissue deformation which prevented Intellesis analysis) and revealed that high levels of ST8Sia4expressing cells in the stroma were significantly associated with better patient outcomes (Fig. 4D) 18]). This indicates that the elevated levels of ST8Sia4 positive-infiltrating stromal cells identifies individuals with good long-term, 10 year, outcomes (Fig. 4E).
Analysis of polySia levels on TILs was also evaluated and correlated to patient outcomes in all 136 tumors (Fig. 4F). TILs were present in 84/136 tumors and in this subset high levels of polysialylated TILs were significantly associated with poor overall survival compared to moderate/low polysialylation which was a good prognostic indicator (p = 0.0016; HR [High/Moderate] = 8.7, 95% CI: 2.27-33.61; High [n = 12], Moderate/Low [n = 72]). Individuals with moderate/low polysialylated TIL had a trend towards improved overall survival compared to those with TILnegative tumors (n = 52; tumors with no lymphocyte infiltration) (p = 0.07). These results are in discordance with our results demonstrating that ST8Sia4 expression was a favorable prognostic factor. Potentially this is a consequence of ST8Sia4 RNA expression not reflecting protein expression, or potentially it is related to the expression of the polySia carrier protein(s) as ST8Sia4 can only generate polySia chains in the presence of a polySia acceptor protein. In addition, as moderate polySia expression was favorable, our assessment of the extreme/high polySia expressers may reflect a unique population.

Evaluation of polySia and CD56 expression in breast cancer
To evaluate polySia expression in the context of a polySia carrier protein we assessed CD56, also known as neural cell adhesion . Whole tumor tissue sections were immunostained for CD56 and polySia (serial sections) and expression levels of CD56 and polySia were analyzed relative to multiple variables: tumor staging, receptor expression, disease progression/recurrence and disease-specific overall survival. An initial evaluation of the correlation between polySia levels and CD56 expression in this cohort was not found to be significant (Pearson R = −0.1091) (Fig. 5D). Next, we evaluated polySia and CD56 expression relative to all clinical parameters and patient outcomes using a Correlation matrix of all clinical data (Fig. 5E). Patients, to date, had a minimum of 3 years clinical follow-up documenting diseasespecific recurrence and disease-specific overall survival status. One patient was lost to follow-up at 14 months and was excluded from the overall survival analysis. Results from the Correlation matrix analysis found, as expected, that clinical parameters such as progression status, 'Progressed', and N stage were positively correlated with a status of 'Deceased' (Fig. 5E). PolySia expression was found to positively correlate with patients whose disease progressed (Pearson R = 0.273) and disease-specific death (Pearson R = 0.402). In this small cohort, polySia expression was found to be significantly higher in patients who died from their disease (p = 0.046) ( Fig. 5F; Alive [n = 19], Deceased [n = 13]). CD56 did not significantly correlate with any clinical parameter or polySia expression but a positive correlation was found for CD56 expression and HER2 status (Pearson R = 0.271) (Fig. 5E). Taken together this data supports the presence of polySia and CD56 expression in breast cancer and identifies a potential association between CD56 and HER2 expression.
Prognostic impact of polySia and CD56 expression in HER2+ breast cancer To further assess the correlation between CD56 expression and HER2 expression we obtained a TMA containing HER2-expressing breast tumors. The TMA contained n = 106 HER2-expressing IDC breast tumor cores with 5-10 years clinical follow-up data reporting on patient outcomes. Adjacent serial sections were immunostained for CD56 and polySia. Analysis of CD56 expression based on HER2 expression (+1, low; +2, moderate; +3, strong) demonstrated that CD56 expression increased with HER2 receptor expression and CD56 expression was significantly higher in HER2 positive (3+) tumors compared to HER2 negative/low (1+) tumors ii ST8Sia2 iv Nuclei  Of the 106 HER2 expressing tumor cores, 56 cores were positive for TILs (lymphocytes ≥10% of total cells/core). An evaluation of polySia and CD56 expression on TILs found that polySia and CD56-positive TILs were readily detectable in tumors. In line with our results (Fig. 4F), when we evaluated total polySia TIL levels relative to patient outcomes by separating out high polySia TILexpressing tumors (n = 29) and comparing these to moderate/ low polySia TIL-expressing tumors (n = 27), we found high polySia TIL expression to be a poor prognostic factor (Fig. 7A). Evaluation of CD56 TIL levels in serial tissue sections found that the majority of tumors containing CD56-positive TILs (n = 36) also contained polySia-positive TILs localized to the same regions (Fig. 7B). Only one tumor was negative for both polySia-and CD56-positive TILs. Interestingly, in the remaining 19 tumors which lacked CD56expressing TILs we found that these tumors still contained polySia-positive TILs (Fig. 7C). To further support these findings, co-localization of polySia with CD56 was assessed using immunofluorescence staining which identified polySia+, CD56+ and polySia+, CD56− TILs ( Fig. 7D and E). This suggests that polysialylated CD56 TILs are present in the tumor microenvironment and potential other, CD56 negative, TILs carry different polysialylated proteins.

DISCUSSION
This study is, to our knowledge, the first comprehensive analysis of polySia, CD56, and polysialyltransferases ST8Sia2 and ST8Sia4, in primary breast tumors. This is also the first identification of polysialylated TILs in the tumor microenvironment. Our work details the prognostic significance of polySia, CD56, and ST8Sia4 in breast cancer adding to the current literature in support of a role for polySia in cancer progression. However, our findings also add complexity to the role of polySia in cancer and suggest that different polysialylated proteins and/or cell type expression may have distinct prognostic capabilities.
Our findings detail polySia and ST8Sia4 tumor cell expression in breast cancer. Neither polySia nor ST8Sia4 was associated with any molecular subtype. Indeed, all molecular subtypes exhibited varying expression levels of polySia and ST8Sia4. ST8Sia4 tumor cell expression was found to significantly associate with tumor grade and poor patient outcomes, and polySia tumor cell expression trended towards worse outcomes. We found that ST8Sia4 expression, as assessed by RNA ISH, did not correlate with polySia expression which could suggest that ST8Sia4 RNA expression does not reflect ST8Sia4 protein levels. Or, potentially, the lack of correlation could be due to the lack of expression of a polySia carrier protein(s) in some ST8Sia4-positive tumors. Our study did not assess ST8Sia4 protein expression primarily due to difficulties validating the specificity of ST8Sia4 antibodies. Another potential limitation of our study is the size of our cohorts. Analysis of additional clinical samples would have strengthened our work and improved our statistical output.
ST8Sia4 was found to be expressed by infiltrating stromal cells and the presence of these cells was found to be a good prognostic indicator. ST8Sia4 is known to be expressed by multiple immune cell subtypes including CD4+ T helper lymphocytes 23 , NK cells 16,17 and macrophages 24 . As our analysis of ST8Sia4 included all cells in the tumor stromal space, it is possible that different populations of ST8Sia4-expressing stromal cells were present and each population(s) may have unique significance for patient outcomes. While ST8Sia4 expression was a favorable prognostic indicator, high levels of polySia expression on TILs was found to be a poor prognostic indicator relative to moderate/low polySia expression. The discordance of these results could potentially be a consequence of unique lymphocyte subpopulation(s) expressing different polysialylated proteins at varying levels, each with their own prognostic significance, or due to ST8Sia4 RNA levels not reflecting protein levels.
Next, we evaluated the expression of CD56, a well-known polySia carrier protein, in breast cancer. Here, we clearly show that CD56 is expressed in a subset of tumors and in approximately half of these tumors, tumor cells are also polySia positive. This strongly suggests the presence of polysialylated CD56 in breast tumor cells. Importantly, we also clearly identify polySia positive, CD56 negative tumor cells indicating that additional proteins are polysialylated in the context of breast cancer. Similar results were found for lymphocytes where polySia+ TILs were found to be CD56+ in some patient tumors, while other tumors had high levels of polySia+ TILs but lacked CD56. Further work is needed to identify the polySia protein carrier(s) in CD56-negative tumor cells and TILs. Potentially, different protein carriers and cell populations have distinct prognostic capacity.
Our evaluation of CD56 in breast cancer found a positive and significant association with HER2 expression. In addition, in this HER2-expressing cohort CD56 expression positively correlated with polySia expression. High levels of CD56 and polySia-CD56 expression on tumor cells was associated with poor patient outcomes. However, while polySia-CD56 trended towards worse patient outcomes, only CD56 expression was found to be significant. A limitation of our study is our cohort size; a larger patient cohort would have strengthened our statistical output. Our evaluation of TIL CD56 expression found that the majority of tumors with CD56+ TILs were also positive for polySia and the presence of these TILs was an extremely favorable prognostic indicator. However, tumors with polySia+ TILs that were CD56-negative were significantly associated poor patient outcome. This demonstrates that there are different populations of polySia-expressing lymphocytes in the tumor microenvironment each with distinct prognostic significance. Likely, this explain why when we performed a general assessment of polySia expression we found high polySia TIL expression associated with poor outcomes and moderate/low expression associated with good outcomes. Separating out the polySia+, CD56+ TIL-expressing tumors from the polySia+, CD56− TIL-expressing tumors revealed polySia as a good prognostic indicator in the context of CD56+ TIL. The role of CD56+, polySia+ lymphocytes in cancer is relatively uncharacterized. Given our results, CD56+, polySia+ TILs may represent an important immune subset with potential implications for improving patient outcomes. CD56 is most often associated with NK cells (part of the innate immune cell repertoire) but it is most unquestionably not limited to NK cells and has been shown to be expressed by some T cell subsets including gamma delta (γδ) T cells and activated CD8+ T cells 25 . The cytotoxic function of NK cells and cytotoxic T cells has an important role in the elimination of tumor cells but the role of CD56 and polySia in immune function is not well characterized. Some studies have shown that CD56+ γδ T cells display potent antitumor activity and CD56+ T cells display enhanced cytotoxicity compared to CD56− T cells 26,27 . NK cells upregulate their CD56 expression upon activation, while in an immunosuppressive microenvironment NK cells lose their CD56 expression and cytotoxic abilities, thus CD56 can be used as an NK activation marker 28  and in the context of lymphocytes, polySia has been documented to occur on NK and CD3+ T cells with CD56 being a carrier protein and, in line with our work, these studies also found ST8Sia4 to be the sialyltransferase responsible for polySia synthesis 16,17,29 . In addition, for NK cells, CD56 and polySia expression were reported to increase following activation with IL-2 suggesting polySia levels are regulated by NK cell activation state. While CD56 and polySia expression clearly fluctuate with activation state, whether or not they have a direct role in immune function is not well described. A small number of studies support an active role for CD56 in immune function and it has been described that CD56 homophilic interactions between immune cells and cancer cells can mediate tumor cell killing 30,31 . How polySia may influence CD56-mediated tumor cell killing is relatively uncharacterized, but given the well described role of polySia in regulating a diverse array of cell and receptor interactions it likely has a role in mediating immune cell function. Identifying the immune subsets expressing polySia and determining the molecular underpinnings of polySia and CD56 function in the context of cancer will be an important next step.

Human tissue and ethics
Breast cancer tissue sections and microarrays were obtained through the Ontario Tumor Bank and from US Biomax (Rockville, MD, USA). Biological materials, provided by the Ontario Tumor Bank (OTB), is supported by the Ontario Institute for Cancer Research through funding provided by the Government of Ontario. Tissue microarrays from US Biomax: Normal adjacent/Cancer adjacent (BRN801c), Staging (TMA #1 and #2: BR10010/ BR20819), Outcomes (TMA #3 HBRe139Su01 and TMA#4 HBRe140Su07). Clinical information was provided from US Biomax and the Ontario Tumor Bank (OTB) on patient age, time to recurrence (OTB only), overall survival status, and tumor characteristics (N Stage, T stage, Grade). Molecular marker (ER, PR, HER2) expression, as evaluated by a pathologist, was provided for all OTB samples. For OTB cases where HER2 expression was equivocal, fluorescence in situ hybridization (FISH) was performed to classify tumors as HER+ or HER2−. For TMA #1 and #2, pathologist scoring of molecular markers (ER, PR, and HER2) was provided on n = 123 of 144 cores. Molecular markers were not provided for cores when tissue was missing or in cases where only fibrofatty tissue and blood vessels were present (i.e. no tumor cells). No molecular markers were provided for outcomes TMA #3. For outcomes TMA #4, ER, PR, HER2, FISH, and Ki67 expression was provided for each tissue core. See Supplemental Fig. 1 for additional details on patient cohorts. Stratified randomization method was applied to the clinical samples and included only individuals diagnosed with Invasive Ductal Carcinoma; tissue cores from individuals diagnosed with Invasive Lobular Carcinoma (Staging, n = 6 and Outcomes, n = 9), Lobular carcinoma in situ (Staging, n = 2 and Outcomes, n = 0), Medullary carcinoma (Staging, n = 6 and Outcomes, n = 0) (n = 6), Glycogen-rich clear cell carcinoma (Staging, n = 2 and Outcomes, n = 0), mucinous carcinoma (Staging, n = 0 and Outcomes n = 6), and intraductal carcinoma (Staging, n = 0 and Outcomes n = 1) were not included in the analysis. Study inclusion was female breast cancer patients, no male breast cancer patient samples were used. The study is compliant with all relevant ethical regulations on the use of human tissue and study approval was obtained from the Institutional Ethics Review Board of UBC (IRB#H17-01442).

Immunohistochemistry and digital image analysis and scoring
Human paraffin-embedded breast cancer tissue microarray sections (TMAs) were incubated overnight at 37°C, followed by deparaffinized with xylene and rehydration using an ethanol gradient followed by a 1Xphosphatebuffered saline (PBS), pH 7.4, wash. For CD56 and CD3 immunostaining heat-induced antigen retrieval was performed at 95°C for 40 min in a 10 mmol/L citrate buffer (pH 6). After heating, slides were cooled to room temperature and washed in 1XPBS. No antigen retrieval was performed for polySia immunostaining. Slides were then incubated with 0.1% Triton X-100 in PBS for 10 min followed by a 20 min incubation in 10% serum from the host of each secondary. TMAs were washed with PBS and incubated with primary antibody (1:500 for polysialic acid, 1:100 for CD56, and 1:200 for CD3 [ab5690]) for 1 h (polysialic acid and CD56) or 0.5 h (CD3) at room temperature. Excess antibodies were removed by washing three times with 1XPBS. TMAs were incubated for 1 h at room temperature with a biotinylated horse anti-mouse IgG or goat anti-rabbit IgG followed by horseradish peroxidase (HRP) streptavidin for 30 min at room temperature and the stained with DAB (3-3'-Diaminobenzidine) (Vector Laboratories). Slides were counterstained with Harris hematoxylin, dehydrated and mounted with Vecta mount medium (Vector Laboratories).
For chromogenic dual staining of ER and polySia, slides were deparaffinized and heated in EDTA antigen retrieval buffer (pH 8) for 20 min, incubated with 0.1% Triton X-100 solution, followed by a 20 min incubated with BLOXALL ® blocking solution. Slides were incubated with anti-estrogen receptor alpha antibody (ab32063) for 1 h at room temperature, followed by incubation with a secondary antibody (biotinylated goat anti-rabbit IgG), then ABC-alkaline phosphatase reagents, and finally ImmPACT Vector Red substrate. Following this, slides were incubated with anti-polySia antibody for 1 h at RT, incubated with secondary antibody followed by HIGHDEF ® yellow IHC chromogen. Slides were counterstained with Hematoxylin and mounted with Vactamount medium.
For CD3 and polySia dual immunofluorescent staining, FFPE slides were deparaffinized, dehydrated and heated in EDTA antigen retrieval buffer (pH 8) for 20 min. After permeabilization and serum blocking, slides were incubated with anti-CD3 antibody (ab16669) at 1:100 and Alexa 488conjugated polysialic acid antibody (ab00240-2.0) at 1:100 for 1 h at RT. Slides were washed with PBS, incubated with Sudan Black solution for 10 min, followed by incubation with Alexa 594-conjugated Rabbit IgG secondary antibody. Nuclei were stained using Hoechst. The slides were mounted with Dako Fluorescent Mounting Medium.
Stained TMA slides were digitalized with the SL801 autoloader and Leica SCN400 scanning system (Leica Microsystems; Concord, Ontario, Canada) at magnification equivalent to ×40. Pathologist evaluation of tumor cells and tumor-infiltrating lymphocytes was performed using the Aperio ImageScope IHC menu (Leica Biosystems) where areas of interest (tumor or TILs) were selected and evaluated using the positive pixel Count Algorithm for each marker. The digital score-based algorithm reports a value between 0 and 1 based on the intensity and percentage in the given area; this value is reported as the H-Score. All scoring was performed blinded to clinical data.

In-situ hybridization and analysis using ZEN Intellesis software
The RNAscope® chromogenic assay (Advanced Cell Diagnostics, Hayward, CA) was used to detect mRNA molecules which are visualized as punctate dots. ST8Sia2 and ST8Sia4 RNAscope probes were designed and provided by Advanced Cell Diagnostics, Hayward, CA. RNAscope probes were used based on manufactures instructions. RNAscope ® Probes used:-Hs-ST8SIA4-C2 Cat# 540401-C2, Hs-ST8SIA2 Cat# 540411. Briefly, TMAs were baked at 60°C for 1 h, deparaffinized using Xylene and 100% ethanol, and air-dried before pretreatments. Slides were then incubated in retrieval buffer at boiling temperature (100°C) for 15 min, rinsed in deionized water, EtoH and immediately treated with protease at 40°C for 30 min in a HybEZhybridization oven (Advanced Cell Diagnostics, Hayward, CA). TMAs were hybridized using both ST8SIA2 and ST8SIA4 probes. Control slides were hybridized with positive control probe Hs PPIB -C1 (PN 321641) and 2 Plex-Negative control probe DapB-C1/DapB-C2 (PN 320751). All probes were incubated at 40°C for 2 h in the HybEZ oven. Slides were washed twice with 1X wash buffer 2, 2 min each at room temperature, and signals were amplified through a 10-step incubation using RNAscope 2.5 HD Reagents Duplex Detection Kit. (ACD, Cat. No. 322435).
Each TMA core was scanned using a Zeiss Axio Observer microscope microscope slide scanner (Carl Zeiss, Germany) at ×20 magnification. Image segmentation and analysis was performed using ZEN Intellesis deepmachine-learning program software (ZEN Intellesis, ZEN 2.6 Blue edition software, Zeiss, Germany). ZEN Intellesis software was used to train and develop a program which could identify nuclei and differentiate between ST8SIA2 and ST8SIA4, or CD56 based on pixel color. The program was then applied to each TMA and image segmentation was performed. Tumor and stromal areas for each tissue section were manually selected and each area was segmented. The digital output reports the total number of chromogenic RNA punctate for each color and total number of nuclei. The results are reported as the total counts of RNA punctate divided by the total nuclei counts. This represents the average mRNA expression per cell. All scoring was performed blinded to clinical data.

Statistics
Data was handled in Excel (Microsoft Excel; RRID:SCR_016137) and analyzed using GraphPad Prism 8.0.1 (GraphPad Prism, RRID:SCR_002798). Column analysis for two data sets was analyzed using a two-tailed unpaired t-test. Welch's correction was applied to any data sets with significant variances (F test to compare variances). Multiple comparisons between greater than two groups used an ordinary one-way ANOVA Tukey's multiple comparisons test. All data sets were analyzed for normality using the Shapiro-Wilk test. If normality failed, data was transformed using either Y = sqrt(Y) or Y = logit(Y). Transformed data was then analyzed for normality using the Shapiro-Wilk test; if normality failed again, data was analyzed using non-parametric Mann-Whitney t-test or Kruskal-Wallis ANOVA. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.