B7-H3 and B7-H4 have different expression patterns in EOC

To evaluate expression of B7-H3 and B7-H4 by different cell types in the ovarian TME, we stained epithelial ovarian tumors (Additional file 1: Table S2) using immunohistochemistry (IHC) and flow cytometry. By IHC, B7-H3 was expressed by both tumor and stromal cells in the epithelial ovarian TME (n = 39) (Fig. 1A). In contrast, B7-H4 expression was restricted to the tumor cell compartment (Fig. 1A). While all EOC cases examined had some level of cytoplasmic B7-H4 expression (n = 28), a minority of tumors analyzed (10/25) exhibited high levels of surface B7-H4 expression by flow cytometry. In contrast, most non-immune cells in EOC exhibited membranous staining for B7-H3 (28/28). This demonstrated that B7-H3 and B7-H4 have different expression patterns by cells of the ovarian TME.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1

B7-H3 expression level differentiates between tumor and stromal compartments in EOC. A: B7-H3 (left) and B7-H4 (right) immunohistochemical staining. B: Linear correlations between Z-scores of protein levels of B7-H3 and epithelial cell markers (E-Cadherin, EpCAM) or stromal cell markers (FAP, PDGFRβ) in the TCGA ovarian serous cystadenocarcinoma dataset. C: Flow cytometry staining of CD45⁻ cells from EOC tumors for B7-H3, epithelial markers (E-Cadherin, EpCAM), and stromal markers (FAP, PDGFRβ, PDPN, CD10). D: B7-H3 surface expression (gMFI) on cell populations positive for epithelial cell markers (E-Cadherin, EpCAM) or stromal cell markers (FAP, PDGFRβ, PDPN, CD10). E: Comparisons between B7-H3 expression on PDGFRβ⁺FAP^low/neg and PDGFRβ⁺FAP^high cells. Paired values are from the same patient. Significance was determined by paired T test (p = 0.0009). F: Linear correlation between T:S defined by histology (H&E staining) and flow cytometry (B7-H3 staining) (R² = 0.59; p < 0.0001). Data from C, D, E are pooled from 2 experiments

To assess whether B7-H3 protein expression was restricted to the TME, we stained 4 non-cancerous tissues for B7-H3. Liver, tonsil, and spleen all exhibited cytoplasmic but not membranous B7-H3 staining. However, extravillous trophoblasts of the placenta stained strongly for membranous B7-H3 expression (Additional file 2: Fig. S1). These results are in accordance with previous reports [22, 29–31] and suggest that B7-H3 expression, while restricted in non-cancerous tissues, may be readily induced in the appropriate environment.

To further investigate the level of B7-H3 and B7-H4 expression on tumor and stromal cells we stained CD45⁻ cells by flow cytometry. We identified two separate populations that expressed different levels of B7-H3. Because B7-H4 expression was restricted to the tumor cell compartment (Fig. 1A), we identified the B7-H3^lowB7-H4⁺ population as tumor cells (red) and the B7-H3^highB7-H4^neg population as stromal cells (blue) (Additional file 2: Fig. S2A). Both tumor and stromal cells were positive for surface B7-H3 expression (n = 28); however, tumor cells consistently expressed lower levels of B7-H3 compared to stromal cells (Additional file 2: Fig. S2B). In contrast, B7-H4 was expressed by tumor cells but not by stromal cells (Additional file 2: Fig. S2C). This demonstrates that B7-H3 and B7-H4 exhibit different patterns of expression by tumor and stromal cells of the EOC TME.

To determine differences in HLA class I expression by tumor and stromal cells in EOC, we examined these populations for HLA-ABC expression by flow cytometry. Tumor and stromal cells expressed different levels of HLA-ABC, with stromal cells (blue) expressing higher levels than tumor cells (red). Notably, surface expression of B7-H3 and HLA-ABC appeared to be positively correlated on stromal cells (Additional file 2: Fig. S2D).

Level of B7-H3 expression can be used to distinguish tumor from stromal cells

To assess which genes are most strongly associated with B7-H3 expression, we performed rank analysis of the 50 genes with the most significant positive correlation with B7-H3 expression across 22 TCGA datasets. Strikingly, half of the top 50 genes (denoted in red) correlated with higher B7-H3 expression had roles associated with stromal cells and extracellular matrix (ECM) remodeling such as collagens, matrix metalloproteinases, and lysyl oxidase cross-linking enzymes (Additional file 2: Fig. S3). These data suggest that high B7-H3 expression results from greater levels of stromal content across multiple tumor types.

To further investigate the link between B7-H3 expression and stromal cell content in ovarian cancer, we interrogated the TCGA dataset of ovarian serous cystadenocarcinoma samples. We found that B7-H3 protein expression was strongly associated with expression of stromal markers fibroblast activation protein alpha (FAP) and PDGFRβ, and negatively associated with epithelial markers EpCAM and E-Cadherin (Fig. 1B), demonstrating that B7-H3 shows strong positive correlations with stromal markers and negative associations with epithelial markers.

To confirm that stromal cells express higher levels of B7-H3, we examined B7-H3 staining intensity by flow cytometry (gMFI) of CD45⁻ cells expressing epithelial (tumor) or stromal markers (Additional file 2: Fig. S4). Cells positive for epithelial markers E-Cadherin or EpCAM expressed low levels of B7-H3 while cells positive for stromal markers FAP, PDGFRβ, podoplanin (PDPN), or CD10 expressed strikingly high levels of B7-H3 (Fig. 1C,D; Additional file 2: Fig. S5A,B). FAP^high and PDGFRβ⁺ cells had average B7-H3 gMFIs above that of EpCAM⁺ cells of 14,147 and 9610, respectively. These data demonstrate that stromal cells in the ovarian TME express higher surface levels of B7-H3 than tumor cells.

To investigate if CAFs differentially express B7-H3, we compared B7-H3 gMFI on cells expressing high levels of the CAF marker FAP to B7-H3 gMFI on cells positive for PDGFRβ but expressing low levels of FAP. While PDGFRβ is also upregulated by CAFs, it is expressed at lower levels by other stromal cells [32]. Consistent with reported expression patterns, we found that PDGFRβ was expressed by a larger proportion of stromal cells than FAP (Additional file 2: Fig. S5C). We compared proportions of PDGFRβ⁺FAP^high to the total proportion of FAP^high cells and found that the proportions were not significantly different (Additional file 2: Fig. S5D) indicating that all FAP^high cells coexpressed PDGFRβ. PDGFRβ⁺FAP^high cells expressed B7-H3 at significantly higher levels than compared to PDGFRβ⁺FAP^low/neg cells (Fig. 1E). This indicates that CAFs express higher levels of B7-H3 than other tumor-associated stromal cells.

Given the wide differential in levels of B7-H3 expression on tumor and stromal cells (Fig. 1D; Additional file 2: Fig. S5A,B), we explored whether it alone could be used to identify the tumor and stromal components of the CD45⁻ fraction by flow cytometry. We compared the proportion of tumor cells calculated from H&E stained slides (Additional file 2: Fig. S6A) and the proportion of tumor cells (B7-H3^low) as assessed by flow cytometry (Additional file 2: Fig. S6B) (n = 23). We found a strong and highly significant correlation (Fig. 1F; R² = 0.59, p < 0.0001) indicating that B7-H3 can be used to assess the T:S by flow cytometry.

Higher stromal content of tumors has been associated with poor overall and disease-free survival [33]. To assess if T:S was significantly associated with differences in disease progression or survival we compared time to recurrence or death for patients with tumors that had low (blue; T:S < 1.5) or high (orange; T:S > 1.5) T:S and found that patients with high T:S trended towards having longer recurrence-free survival (Additional file 2: Fig. S7; p = 0.098) suggesting that stromal content may impact recurrence in epithelial ovarian cancer.

B7-H4 and PD-L1 are differentially expressed by tumor and stromal cells

We used flow cytometry to differentiate tumor and stromal cells by level of B7-H3 expression to gain more insight into the expression of B7 family members PD-L1 and B7-H4 in the ovarian TME. Our analysis showed that PD-L1 and B7-H4 are predominantly expressed by different cell populations in the ovarian TME. PD-L1 was more highly expressed in the stromal rather than tumor compartment (Fig. 2A,B; p = 0.0006) while B7-H4 expression was primarily expressed by the tumor compartment (Fig. 2C,D; p = 0.031). Importantly, we did not observe a B7-H4⁺PD-L1⁺ double positive population (Fig. 2E). These data show that B7 family members are differentially expressed by tumor and stromal cell populations in the ovarian TME.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2

Cells expressing both PD-L1 and B7-H4 are not prominent within the CD45 negative population. Single cell suspensions from freshly isolated tumor cells were stained for CD45, B7-H3, PD-L1 and B7-H4. Tumor and stromal cells were gated from the CD45- population based on level of B7-H3 expression. A: Examples of PD-L1 staining on tumor (left) and stromal (right) cells are shown. B: Level of PD-L1 expression on tumor and stromal populations (Mean ± SEM; p = 0.0006). C: Examples of B7-H4 staining on tumor (left) and stromal (right) cells. D: Level of B7-H4 expression on tumor (left) and stromal (right) populations (Mean ± SEM; p = 0.031). E: Example plot showing frequency of B7-H4⁺PD-L1⁺ positivity of total CD45⁻ cells. Tumor cells shown in red, stromal cells shown in blue, matched FMOs are shown in grey. Statistical significance was determined by Mann Whitney U test

T:S is not associated with frequency of infiltrating immune cells

To determine if the proportion of stromal cells in the ovarian TME impacted the frequency of immune cell infiltration, we compared immune cell frequency in tumors with low (blue; T:S < 1.5) or high (orange; T:S > 1.5) T:S. We did not observe significant differences in the frequency of T cells (n = 24; Fig. 3A), B cells (n = 16; Fig. 3B), or CD14⁺ cells (n = 18; Fig. 3C) between these two groups suggesting that the T:S does not significantly impact the recruitment of immune cells.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3

Higher tumor-to-stroma ratio (T:S) does not significantly correlate with higher proportions of infiltrating immune cells. A, B, C: Frequencies of T cells (A; n = 24), B cells (B; n = 16), and monocytes (C; n = 18) infiltrating EOC tumors. D: Infiltration scores of tumor-infiltrating CD3⁺, CD20⁺, and CD68⁺ cells in the epithelial or stromal cell compartments. T:S was determined by flow cytometry using B7-H3 staining and is represented below bar graphs by color gradient. Whisker and box plots summarize data from tumors with low (T:S < 1.5; blue) or high (T:S > 1.5; orange) T:S. Statistical significance was determined by Mann Whitney U test

To determine if there were differences in the localization of infiltration, histology sections were stained for immune markers and scored for the extent of infiltration in the epithelial or stromal tumor compartments based on the density of positive staining for immune cell populations. T:S was not associated with differences in T cell (CD3⁺), B cell (CD20⁺), or macrophage (CD68⁺) infiltration density in the epithelial tumor compartment; however, T and B cells exhibited significantly higher densities of infiltration in the stromal compartment of tumors with higher stromal content (Fig. 3D). These data suggest that, while detectable differences in immune cell recruitment to tumors is not corelated with T:S, local density of infiltration may be affected by tumor and stromal composition of the ovarian TME.

CD8⁺ T cells from tumors with high T:S express higher levels of PD-1

Further analysis was done to evaluate whether there were differences in T cell phenotype associated with T:S. We did not find differences in proportions of infiltrating CD4⁺ or CD8⁺ T cells in relation to the T:S (n = 24; Fig. 4A). We also did not see differences in the density of infiltration of cytotoxic (CD8⁺) or regulatory (FoxP3⁺) T cells in either the tumor or stromal cell compartments (Fig. 4B).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4

Tumors with a higher tumor-to-stroma ratio (T:S) have an increased proportion of infiltrating CD8⁺ T cells expressing high levels of PD-1. A: Proportions of CD4⁺ and CD8⁺ T cells isolated from tumors (n = 24). B: Infiltration scores of tumor-infiltrating CD8⁺ and FoxP3⁺ cells in tumors with low (T:S < 1.5; blue) or high (T:S > 1.5; orange) T:S. C,D: Proportion of CD4⁺ (C) and CD8⁺ (D) T cells (n = 24) expressing high levels of PD-1. Whisker and box plots summarize data from tumors with low (T:S < 1.5; blue) or high (T:S > 1.5; orange) T:S. T:S of tumor is represented below by color gradient (A,C,D). Statistical significance was determined by Mann Whitney U test

To further interrogate the phenotype of the infiltrating T cells, we analyzed expression of PD-1 on tumor-infiltrating CD4⁺ and CD8⁺ T cells. While some level of PD-1 expression on T cells was observed in all tumors examined, many samples exhibited a distinct PD-1^high population with a bimodal distribution in both CD4⁺ and CD8⁺ T cell populations (n = 24; Fig. 4C,D). No significant differences were seen in the frequency of PD-1^high CD4⁺ cells in relation to T:S (Fig. 4C), but tumors with a higher T:S had significantly higher proportions of infiltrating CD8⁺ T cells exhibiting high levels of PD-1 (Fig. 4D). However, we did note that 2/13 samples with low T:S had high frequencies of CD8⁺PD-1^high cells, and 5/11 samples with high T:S had low frequencies of CD8⁺PD-1^high cells. This could indicate that T cell functionality or exhaustion may be influenced by T:S but additional factors are likely contributing to this T cell phenotype.

To investigate if neoadjuvant chemotherapy (NACT) contributed to differences in PD-1 expression by T cells, we compared frequencies of PD-1^high CD4⁺ and CD8⁺ T cells between patients who were and were not treated with NACT. We did not find any associations between proportion of T cells expressing high levels of PD-1 and chemotherapy treatment indicating that the observed phenotype is independent of chemotherapy treatment (Additional file 2: Fig. S8A).

To investigate other possible differences in the phenotype of T and B cells infiltrating tumors with high or low T:S, we analyzed expression of various markers of activation. We did not observe significant differences in the proportion of CD4⁺ or CD8⁺ T cells co-expressing high levels of PD-1 and exhaustion markers TIM3 or LAG3 (Additional file 2: Fig. S9A,B). Similarly, we did not observe significant differences in expression of activation or cytotoxicity markers examined (ICOS, TIA-1, granzyme B, 2B4, CD107a, GITR) (Additional file 2: Fig. S9C,D).

To further interrogate the phenotype of infiltrating B cells, we stained for several B cell markers associated with naïve (CD20⁺CD22⁺; IgD⁺) or antigen-experienced (BTLA⁺; CD27⁺) status (Additional file 2: Fig. S10A). We did not observe differences in proportion of CD19⁺ B cells expressing BTLA, CD27, CD20 and CD22, or IgD in relation to T:S (Additional file 2: Fig. S10B).

We found that tumors with higher T:S had greater frequencies of CD8⁺ T cells expressing high levels of PD-1; however, we did not find differences in T cells co-expressing inhibitory markers TIM3 or LAG3 with high levels of PD-1. Additionally, we didn’t find differences in proportions of T cells expressing ICOS, TIA-1, granzyme B, 2B4, CD107a, or GITR, nor in proportions of B cells expressing CD20, CD22, BTLA, or CD27. Taken together, this in-depth analysis of surface marker expression revealed that T and B cells isolated from the ovarian TME express variable levels of inhibitory and activating markers, and the frequency of PD-1^high CD8⁺ T cells was greater in tumors with higher T:S.

Tumors with higher T:S are associated with changes in the phenotype of monocytes and APCs

To further characterize the immune infiltrate of ovarian cancer, we examined the phenotypes of infiltrating monocytes (CD14⁺) and mature APCs (CD11c⁺HLA-DR^high). Monocytes from tumors with a high T:S expressed higher levels of CD16 (Fig. 5A,B). Mature APCs isolated from tumors (Fig. 5C) were stained for coinhibitory (B7-H3, B7-H4, PD-L1, PD-L2) and costimulatory (B7-H3, ICOSL, CD40, CD86, CLEC9a) molecules (Fig. 5D). B7-H3 was highly expressed on APCs from all patients, but no differences in level of expression were seen in relation to the T:S. Similarly, no significant differences were seen in B7-H4, ICOSL, CD40, PD-L2 or CD86 expression in relation to T:S, although expression of both PD-L2 and CD86 trended towards being more highly expressed on APCs infiltrating tumors with higher T:S. APCs isolated from tumors with higher T:S expressed significantly higher levels of inhibitory PD-L1 and significantly lower expression of CLEC9a (Fig. 5E). We confirmed that these differences in monocyte and APC phenotype were not associated with NACT treatment (Additional file 2: Fig. S8B,C). These data indicate that when a higher proportion of tumor cells are present, monocytes express higher levels of CD16 and APCs express higher levels of PD-L1.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5

Tumors with a higher tumor-to-stroma ratio (T:S) have increased CD16 expression on infiltrating monocytes (CD14⁺), and mature APCs (CD11c⁺HLA-DR^high) expressing higher levels of PD-L1 and lower levels of CLEC9a. A: Histogram showing representative CD16 staining on monocytes (red) compared to FMO (grey). B: Differences in normalized CD16 expression compared to FMO on monocytes from tumors with low (T:S < 1.5; blue) or high (T:S > 1.5; orange) T:S. C: Flow plot showing representative gating of mature APCs. D, E: Mature APCs were stained for activating (ICOSL, B7-H3, CD40, CD86, CLEC9a) and inhibitory (B7-H4, B7-H3, PD-L1, PD-L2) surface molecules. Histograms show representative staining (red) compared to FMO (grey) (D). E: Normalized surface molecule expression on mature APCs from tumors with low (T:S < 1.5; blue) or high (T:S > 1.5; orange) T:S. Statistical significance was determined by Mann Whitney U test

Heterogeneity in immune cells isolated from different metastatic sites

For 5 patients, we received two tumor samples from different metastatic sites (ovary and omentum [n = 4], or right and left ovary [n = 1]). We compared data from intrapatient tumors to see if immune cell frequency or phenotype was consistent within a patient or governed by tumor-specific parameters. We calculated the linear correlation between levels of expression of select markers isolated from different tumors from the same patient and found that expression of CD86 on APCs (r = 0.989; p = 0.011), PD-L1 on APCs (r = 1; p = 0.00049) (Fig. 6A), and the proportion of PD-1^high CD4⁺ T cells (r = 0.953; p = 0.012) (Fig. 6B) all exhibited strong positive correlations between tumor sites. Other parameters such as expression of CD40 on APCs (r = 0.829; p = 0.17), ICOSL on APCs (r = 0.732; p = 0.27) (Fig. 6A), the proportion of PD-1^high CD8⁺ T cells (r = − 0.308; p = 0.61) (Fig. 6B), the proportion of tumor cells (r = 0.656; p = 0.23), and the proportion of B7-H4-expressing tumor cells (r = 0.665; p = 0.22) (Fig. 6C) were not significantly correlated between tumor sites. These data demonstrate that the level of expression of PD-L1 and CD86 on APCs and the proportion of PD-1^high CD4⁺ T cells are strongly correlated between metastatic sites within a patient but expression of CD40 and ICOSL on APCs, the proportion of PD-1^high CD8⁺ T cells, the proportion of tumor cells, and the proportion of B7-H4-expressing tumor cells were not correlated between metastatic sites.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 6

Tumors from the same patient show strong correlation across expression of some immune features. A: Linear correlations between normalized expression of CD86 (r = 0.989; p = 0.011), PD-L1 (r = 1.0; p = 0.00049), CD40 (r = 0.829; p = 0.17), and ICOSL (r = 0.732; p = 0.27) on APCs (CD11c⁺HLA-DR^high) isolated from two tumors within the same patient. B: Linear correlations between proportion of CD4⁺ (r = 0.953; p = 0.012) or CD8⁺ (r = − 0.308; p = 0.61) T cells isolated from two tumors within the same patient expressing high levels of PD-1. C: Linear correlations between the percentage of non-immune cells that are tumor cells (r = 0.656; p = 0.23) and the proportion of tumor cells expressing B7-H4 (r = 0.665; p = 0.22)

To examine whether the site of metastatic disease influenced immune cell frequency we compared T, B, and monocyte frequencies from ovary and omental tumors. We found that omental tumors had higher frequencies of T and B cells (Additional file 2: Fig. S11A,B). Monocyte frequency trended towards being lower in omental tumors (Additional file 2: Fig. S11C) which could be due to the increase in the T and B cell populations. Despite finding a higher frequency of T cells in omental tumors, we did not observe differences in the proportion of T cells that were CD4⁺ or CD8⁺ (Additional file 2: Fig. S11D).

To interrogate whether the site of metastatic disease influenced immune cell phenotype we compared cells isolated from ovary or omental tumors. We did not find differences in frequency of PD-1^high T cells (Additional file 2: Fig. S11E), level of CD16 expression on monocytes (Additional file 2: Fig. S11F), or expression of activating and inhibitory markers on APCs (Additional file 2: Fig. S11G) between the two metastatic sites. These data indicate that while immune cell frequency is influenced by the metastatic site, the organ site does not influence immune cell phenotype.