Multidimensional imaging provides evidence for down-regulation of T cell effector function by MDSC in human cancer tissue

See allHide authors and affiliations

Science Immunology  18 Oct 2019:
Vol. 4, Issue 40, eaaw9159
DOI: 10.1126/sciimmunol.aaw9159

Interior tumor views

Previous studies indicate that a high frequency of intratumoral neutrophils is associated with a poor clinical prognosis. Si et al. used a variety of microscopy and imaging techniques to examine how intratumoral interactions between tumor-associated neutrophils (TANs) and tumor-infiltrating lymphocytes (TILs) can affect TIL function. They developed 2D and 3D multiparameter immunofluorescence imaging techniques to localize and define functional cell subsets, which were used to identify hotspots of TAN/TIL interactions within tumors. Some of these TANs had a polymorphonuclear myeloid-derived suppressor cell (PMN-MDSC) phenotype, and their physical association with TILs reduced antitumor functions of those TILs. This study identifies areas of MDSC activity in human tumors and provides a more detailed perspective of how intratumoral cell interactions influence cell function.


A high intratumoral frequency of neutrophils is associated with poor clinical outcome in most cancer entities. It is hypothesized that immunosuppressive MDSC (myeloid-derived suppressor cell) activity of neutrophils against tumor-reactive T cells contributes to this effect. However, direct evidence for such activity in situ is lacking. Here, we used whole-mount labeling and clearing, three-dimensional (3D) light sheet microscopy and digital image reconstruction supplemented by 2D multiparameter immunofluorescence, for in situ analyses of potential MDSC–T cell interactions in primary human head and neck cancer tissue. We could identify intratumoral hotspots of high polymorphonuclear (PMN)–MDSC and T cell colocalization. In these areas, the expression of effector molecules Granzyme B and Ki67 in T cells was strongly reduced, in particular for T cells that were in close proximity or physically engaged with PMN-MDSC, which expressed LOX-1 and arginase I. Patients with cancer with evidence for strong down-regulation of T cell function by PMN-MDSC had significantly impaired survival. In summary, our approach identifies areas of clinically relevant functional interaction between MDSC and T cells in human cancer tissue and may help to inform patient selection in future combination immunotherapies.


The tumor microenvironment is a spatially organized landscape with immune cells located in the core of the tumor, its invasive margin or in the surrounding stroma that mainly consists of fibroblastoid mesenchymal cells (1). Different types of infiltrating immune cells have different effects on tumor progression (2), and location-specific effects on disease progression and response to therapy are possible. A strong lymphocytic infiltration is typically associated with good clinical outcome in many different tumor types, including melanoma, breast, bladder, urothelial, ovarian, colorectal, renal, lung, and head and neck cancer (HNC) (26). Accumulating evidence also indicates a critical role of myeloid cells in the pathophysiology of human cancers. Increased numbers of neutrophil granulocytes have been observed both in the peripheral blood and in the tumor tissues of patients with different types of cancer (7). In murine studies, it appears that tumor-associated neutrophils (TANs) can exert both protumor and antitumor effects (8). Numerous studies have shown that neutrophils can promote tumor progression by degrading matrix, stimulating tumor cell proliferation, increasing metastasis, and enhancing angiogenesis (9). In addition, immunosuppressive types of neutrophils, termed PMN-MDSC (polymorphonuclear myeloid-derived suppressor cells), have been described in the circulation, with preferential accumulation and copurification with mononuclear cells in the so-called “low density neutrophil” (LDN) fraction of the blood (10). In tissues, neutrophils might also contribute to immune suppression and, subsequently, tumor progression in a manner similar to that described for PMN-MDSC (11). However, TAN can also exert antitumor functions, such as promoting tumor cell death via their powerful antimicrobial killing machinery or by inducing factors that recruit and activate innate and adaptive immune cells (12, 13).

In recent years, there has been progress in the phenotypic and functional characterization of circulating immunosuppressive neutrophils. Immunosuppressive MDSC activity has been ascribed to activated neutrophils in the LDN fraction of the peripheral blood (14). Lectin-type oxidized low-density lipoprotein receptor-1 (LOX-1) has been suggested as a potential marker for neutrophils with MDSC activity (15). Of note, mature neutrophils contained stronger suppressive MDSC activity as compared with their immature counterparts (16). In contrast to circulating cells, still very little is known about the identity of PMN-MDSC or other subtypes of neutrophils in human cancer tissue. Functional studies on tissue cells depend on the analysis of bulk cell suspensions released from sufficiently large, fresh tissue pieces. Such studies are possible in lung cancer, and Singhal et al. identified a specific subset of TANs in lung cancer of patients with early-stage disease. Those antigen-presenting cell (APC)–like “hybrid neutrophils” exhibited characteristics of both neutrophils and APCs, expressed human leukocyte antigen (HLA)–DR, cross-presented antigens, and stimulated T cell activity (17). Furthermore, Condamine et al. (15) found that 15 to 50% of neutrophils in tumor tissues stained positive for LOX-1. Neutrophils from the peripheral blood and bone marrow, which expressed LOX-1, were more suppressive compared with their LOX-1 counterparts. However, no functional analysis of LOX-1+ TAN from tissue was performed in that study.

The big disadvantage of ex vivo functional analyses of tumor-infiltrating immune cells based on tissue dissociation is the fact that cells can no longer be assigned to their original localization in either tumor core, stroma, or blood vessels. Cells are isolated from only macroscopically evaluated fresh tissue pieces, and mechanical processing or enzymatic treatment further modulate the cells of interest. The purity of the preparation can even be further influenced by the remaining peripheral blood cells. Hence, the spatial context, e.g., inflamed versus less infiltrated tissue is lost, and functionally relevant physical interactions of cells are disrupted. Thus, cellular interactions restricted to specific tissue regions cannot be assessed.

Histopathological analysis combined with quantitative multiparameter immunofluorescence imaging allows mapping the spatial distribution of tumor cells and immune cells (18). It also enables a separate analysis of tumor core, margin, and stromal areas together with the definition of tumor regions highly or less infiltrated by immune cells. Last, potential interactions between immune cells, including conjugate formation, can also be quantified. This spatial context and localization of particular immune cell subsets are crucial for understanding their function as well as their role in disease progression and response to therapy in HNC and other types of cancer (2, 6, 19). Moreover, aided by advanced image analysis technologies, the spatial context of tumor and stromal cells can be analyzed at single-cell resolution using rigorous spatial statistics to investigate locally restricted clustering, dispersion, and interactions in two or three dimensions (3D) (18).

In our study, we focused on the interaction of TAN and TIL (tumor-infiltrating lymphocytes) in the microenvironment of tumors of the head and neck region. Understanding the role of TAN in the tumor immune microenvironment is important, because TANs are one of the most powerful predictors of poor survival in large pan-cancer analyses (7). Our study explored the possibility that TAN or TAN subsets impair the activity of tumor-infiltrating T cells in situ. To this end, we used both 2D and 3D quantitative multiparameter immunofluorescence imaging approaches and analysis. We determined the localization of TAN and TIL in the tumor tissue to further define functional subsets of the cells. By this approach, we identified intratumoral hotspots of TAN/TIL interactions. We found evidence of strongly reduced TIL function in such hotspots, especially in regions where the TAN showed a phenotype of PMN-MDSC and physically engaged with TIL. Because the degree of inhibitory TAN/TIL interactions strongly correlated with patient survival, we propose that one mechanism of PMN-MDSC function is the intratumoral inactivation of TIL by direct physical confrontation.


Density and intratumoral localization of granulocytes and T cells

Tumor-infiltrating immune cells can be localized either in epithelial tumor core areas with direct contact to tumor cells or in stromal regions, which mainly consist of fibroblastoid mesenchymal cells. Segmentation into stromal and tumor core area was performed for each tissue sample based on training of the image analysis software and pan-cytokeratin staining (Fig. 1, A and B). We then used the markers CD66b and CD3 together with tissue segmentation to determine the density of granulocytes and T cells in tumor core and stromal regions, respectively (Fig. 1, B and C). We found high interindividual variation of cell densities, with equal T cell densities in both stroma and tumor regions. In contrast, the density of CD66b+ cells was significantly lower in the tumor core area. 2D tissue analysis alone is inherently spatially limited and cannot faithfully represent the complex 3D tissue architecture. To overcome this, we established a whole-mount labeling and tissue clearing approach based on 3D light sheet fluorescence microscopy (20), which permitted the determination of absolute immune cell numbers as well as their 3D distribution. We segmented and quantified stroma and tumor volumes and determined granulocyte and T cell infiltrate densities in both areas (Fig. 1, D and E, and movie S1). Using this novel technical approach, we could quantify the absolute number of CD66b+ and CD3+ cells in human tissue pieces of up to 1-cm3 size. Quantitative analysis revealed cellular densities of 15.000 to 60.000 granulocytes and T cells per mm3 for most patients analyzed. 3D absolute cell counts confirmed the reduced density of CD66b+ cells in tumor core areas (Fig. 1F).

Fig. 1 Intratumoral spatial distribution of granulocytes and T cells.

Tissue studio composer technology and image-based machine learning (A), pan-cytokeratin staining (B), and the manual tracing wizard of tissue studio were used to define tumor and stroma regions. Multiparameter immunofluorescence was used to determine the density of CD3 and CD66b cells in 2D tissue sections (n = 137 patients) (C) or whole-mount stained 3D material (n = 11) (F) obtained from patients with HNC. (D) and (E) represent still images of videos and depict CD3 cells (blue), CD66b cells (green), and tumor surface area. Movie S1 in (E) is additionally provided as supplementary material. Using the split into surface Spots tool of Imaris, the density of CD3+ and CD66b+ cells within defined distances to the tumor margin was determined (G to J). Paired t test was used to determine statistical significance.

The application of a 3D shell model also allowed grouping of segmented cells according to their distance from the tumor margin (Fig. 1, G and H). Density of CD3+ cells was highest at the tumor margin and decreased with distance (Fig. 1I). In contrast, CD66b+ cell densities were relatively consistent in tumor-surrounding stromal tissue but also decreased with higher proximity to the tumor core (Fig. 1J).

Identification, spatial distribution, and clinical relevance of LOX1+ TAN

The MDSC activity of a subtype of PMN leukocytes in the peripheral blood, now termed PMN-MDSC (21), is well established. For these circulating PMN-MDSC, both CD66b and CD15 have been used as PMN markers, and suppressive subsets have been defined by CD10, CD16, and LOX-1 positivity in three independent studies (15, 16, 22). Although LOX-1 has been suggested as a PMN-MDSC marker, very little is known about the phenotype, frequency, localization, activity, and clinical relevance of potentially T cell–suppressive granulocytic cells in human tumor tissues.

To shed more light on this topic, we first characterized tumor-infiltrating leukocytes by flow cytometry, applying the markers mentioned above. We found that the majority (90 to 99%) of CD66b+ neutrophils in the tumor tissues were positive for CD10, CD11b, and CD16, a phenotype characteristic for mature neutrophils (Fig. 2, A and B). Confirming previously reported observation (15), LOX-1 defined two subsets of neutrophils in our cohort of patients with HNC. Using multicolor immunofluorescence (Fig. 2C), we found the relative proportion of LOX-1+ TAN to vary greatly among patients (Fig. 2D; range 4 to 99% and 1 to 97% in tumor and stroma, respectively). Consistent with the reduced overall CD66b+ cell densities reported above, LOX1+CD66b+ cells also had lower densities within tumor core areas compared with stroma (Fig. 2E).

Fig. 2 Identification and quantification of intratumoral TAN subsets.

TAN [see (A) for gating] isolated from fresh tumor tissue of patients with HNC cancer (n = 8) were subjected to flow cytometry and stained for the expression of LOX-1, CD11b, CD16, and CD10 (B). FSC, forward scatter. Multiparameter immunofluorescence in cryosectioned tissue samples [see (C) and (F) for examples] was used to determine the density of CD66b+ cells expressing LOX-1, MPO, arginase I, or HLA-DR (D, E, and G) in tumor and stroma regions, respectively. The percentage of LOX-1+ and LOX-1 TAN–expressing MPO (H) and arginase I (I) in tumor and stroma was determined by three-color immunofluorescence combined with a DAPI counterstain (F). Kaplan-Meier survival curves for OS were plotted for patients with CD66b+Lox1+ cell density above or below the median in tumor (J) and stroma (K) regions (patients with LSCC, n = 53). Statistical significance was determined by paired t test (D to I) or log-rank regression analysis (J and K).

The expression of myeloperoxidase (MPO), arginase I and HLA-DR in neutrophils and PMN-MDSC has been associated with their tumor-promoting and T cell–suppressive or T cell–stimulatory activity, respectively (14, 17, 23). Multicolor immunofluorescence (Fig. 2F) revealed that the majority of the CD66b+ cells in both tumor and stroma expressed arginase I and MPO, whereas only very few neutrophils (below 20% in most patients) expressed HLA-DR (Fig. 2G). Although the majority of the LOX-1+ TAN were positive for MPO and immunosuppressive arginase I, these molecules were less frequently expressed in LOX-1/CD66b+ cells (Fig. 2, H and I). Nevertheless, HLA-DR–low/negative and HLA-DR–positive TAN both had substantial expression of LOX-1 and arginase I (fig. S1).

Next, we assessed the clinical relevance of TAN density and localization in two subsets of HNC, namely, squamous cell carcinoma (SCC) of the larynx (LSCC) and SCC of the oropharynx (OSCC). A high density (above median) of TAN in tumor tissue in both cancer types was significantly associated with reduced overall survival (OS) (fig. S2, A and C; see tables S2 and S4 for multivariate analysis). In contrast, high TAN densities in stroma regions did not have a significant effect on OS (fig. S2, B and D). A profound prognostic relevance was observed for LOX-1+ TAN density in the tumor core area of patients with LSCC and OSCC (Fig. 2J and fig. S2E; see table S4 for multivariate analyses). This was in contrast to a moderate and statistically not significant relationship between outcome and LOX-1+ TAN density in the stromal tissue areas (Fig. 2K and fig. S2F). The clinical relevance of TAN and LOX-1 was underscored by the observed correlation of TAN density and LOX-1 expression with lymph node metastasis staging (fig. S2, G and H).

Quantitative in situ tissue analysis provides evidence for in vivo PMN-MDSC activity

It has been suggested that LOX-1+/CD66b+ cells represent PMN-MDSC (15). However, experimental data on a potential in situ interaction of LOX-1+/CD66b+ cells and T cells in tumor tissues are lacking. Therefore, we embarked on a systematic analysis of TAN and potential PMN-MDSC with T cells in the tumor tissue of patients with HNC.

Using our established 3D analysis protocol, we first delineated potential hotspots of interaction between CD66b+ cells and CD3 cells in defined tissue areas. After quantification of CD66b+ and CD3+ cell densities in both 2D and 3D approaches (Fig. 3, A to C), we noted a heterogeneous distribution of both cell types. We therefore defined T cell–dominated, TAN-dominated, and mixed regions. TAN-dominated regions were defined as areas with a neutrophil/T cell ratio (NTR) >9 (more than 90% neutrophils), T cell dominated with an NTR <0.1 (more than 90% T cells), and mixed regions with an NTR between 0.1 and 9. Mixed regions were the most frequent areas, making up 49% of the tumor core area and 54% of the stroma (Fig. 3D). Movie S2 and Fig. 3B show examples for the 3D reconstruction and color coding of those regions (TAN region, green; T cell region, red; and mixed region, yellow).

Fig. 3 Spatial distribution and interaction of TAN and T cells in situ.

T cells (CD3, red) and TAN (CD66b, green) were identified and quantified by whole-mount staining (A and B) using Imaris or in cryosections (C) using tissue studio. NTR was determined. Regions with NTR >9 (more than 90% neutrophils, green circle) were defined as “neutrophil dominant,” regions with NTR <0.1 (less than 10% neutrophils, red circle) as “T cell dominant” and regions with NTR between 0.1 and 9 as mixed regions (yellow circle) (see movie S2 for illustration). White outline in (C) defines the tumor core area, and the pie charts in (D) show the relative percentage of the indicated region in tumor and stroma (n = 137). Determination of relative n-n distance from a CSR model (see Supplementary Materials and Methods for details) indicated a nonrandom distribution and clustering of CD3 and CD66b cells (E). Physical conjugation of CD3 cells with CD66b cells (F to H) and percentage of LOX-1 positivity in conjugated versus unconjugated CD66b+ cells (I) was visualized and quantified (n = 53). Movie S3 illustrates conjugate analysis after whole-mount staining, ultramicroscope image acquisition, and 3D image reconstruction; color code indicates nonconjugated (red/green) or conjugated (violet/blue). Relative density of CD3 target cells as a function of the radial distance around CD66b subject cells was determined in tumor (J) and stroma (K). Here, a relative cell density of 1 corresponds to a CSR distribution of cells. The data are indicative of a significantly increased clustering of CD3 cells around LOX-1+ TAN. For (L) and (M), Kaplan-Meier survival curves for OS were plotted for patients with LSCC (n = 53) with the percentage of CD3 conjugated with CD66b+ Lox1+ cells above or below the median. Statistical significance was determined by log-rank regression analysis with the level of significance set at P ≤ 0.05. Cox multivariate regression analysis was performed by adding UICC-TNM stages, and the percentage of CD3 conjugated with CD66b+ Lox1 into a model. Differences in both tumor and stroma region remained significantly associated with OS [tumor: hazard ratio (HR) = 4.543, P = 0.003; stroma: HR = 4.007, P = 0.005; see tables S1 to S4 for details].

Next, we determined the exact position and mean nearest neighbor (n-n) distances of CD3 to CD66b cells in the entire tissue section of 54 patients. In each instance, the n-n distances were normalized by division with the n-n distances for a complete spatial random (CSR) distribution model, representing theoretical random distribution. We found that the actual relative mean n-n distance had values below 1, which is less than that expected for a CSR distribution, and thus indicates greater proximity and potential functional interaction between the two cellular subsets (Fig. 3E). Quantitative analysis of direct CD3-CD66b conjugates revealed, on average, 5 to 10% of CD3 and CD66b cells engaged in conjugation (Fig. 3, F and G). We also measured conjugation with CD66b cells separately for CD4 and CD8 T cell subsets. Both T cell subsets displayed similar conjugation frequencies with neutrophils (mean around 5%, blue symbols in fig. S3, A and B).

Because cellular conjugation in tissues is a 3D process, we validated our data using whole-mount labeling of tissue pieces. Movie S3 and Fig. 3H show the identification of unconjugated and conjugated cells in situ after two-color whole-mount staining, ultramicroscope image acquisition, digital conversion, and 3D image reconstruction. The quantification confirmed that indeed 5 to 10% of CD3 and CD66b cells engaged in conjugates with each other (fig. S4A). Although more than 10% of CD3 and CD66b cells formed conjugates close to the tumor margin, the relative percentage of conjugated cells declined to below 5% at distances greater than 200 μm into the tumor core or stromal area (fig. S4, B and C). Given the potential role of LOX-1+ TAN in T cell modulation, we next developed a four-color staining protocol (fig. S4D) and found an increased LOX-1 positivity in CD66b+ cells forming conjugates with T cells (Fig. 3I). This increased positivity was likewise observed for conjugates with CD4 and CD8 T cell subsets (fig. S3, C and D). A plot of the relative CD3 cell density as a function of radial distance around LOX+ TAN and LOX TAN cells indicates an increased CD3 clustering around TAN cells in general, with a significantly higher degree of clustering around LOX-1+ TAN cells, specifically in the tumor region (Fig. 3, J and K). Overall, these data suggest a higher propensity of LOX-1+ TAN to engage with T cells, although an up-regulation of LOX-1 as a consequence of conjugate formation cannot be ruled out. Last, clinical follow-up data supported the potential functional and pathophysiological relevance of TAN-TIL conjugate formation. Patients with a high percentage of CD3 conjugated to LOX-1+ TAN (Fig. 3, L and M) or overall TAN (fig. S4, E and F) in their tumor tissue displayed significantly impaired survival compared with those patients with lower CD3 conjugation. Together, these data define and quantify intratumoral areas of substantial TAN-TIL interaction and suggest a functional and clinical relevance of these interactions especially for LOX-1+ TAN.

Data from murine and human ex vivo models suggested down-regulation of T cell function by PMN-MDSC. To obtain evidence for this process in human cancer tissue, we performed an in-depth analysis of potential links between CD66b+ cells and expression of GrzB (cytotoxicity, Fig. 4A) and Ki67 (proliferation, Fig. 4B) in tumor-infiltrating T cells. Experiments focused on mixed regions, in which frequent interaction of both cell types occurred. In this region, we determined the relative percentage and the density of different subsets of CD66b+ cells and determined the correlations with the relative percentage and the density of T cells expressing either GrzB or Ki67 (Fig. 4, C and D). High prevalence of CD66b+ subsets expressing either LOX-1, arginase, or MPO negatively correlated with the density or percentage of proliferating and cytotoxic T cells. The strongest negative correlations were found between the density of arginase and LOX-1–expressing TAN and the density of proliferating and cytotoxic T cells in the tumor area (Fig. 4C, left part). Correlations in the stromal area were similar but less pronounced, suggesting a stronger modulation of T cell effector function by immunosuppressive CD66b+ cells in the tumor area. The density of HLA-DR+ TAN strongly correlated with the increased prevalence of proliferating and cytotoxic T cells, both in stroma and tumor regions, suggesting the possibility of T cell stimulation by this TAN subset (17). Evidence for in situ T cell suppressive activity was underscored by a more detailed analysis of CD66b+ subsets. As shown in Fig. 4D, the density of LOX-1+ cells coexpressing either arginase or MPO strongly correlated with a reduced expression of Ki67 and GrzB in T cells. This correlation was significantly stronger in the tumor region as compared with stroma. Last, we divided patients into two groups with LOX-1 expression either above or below the median and determined the level of Ki67/GrzB expression in the mixed region of the tumor tissue (Fig. 4, E and F). Consistent with previously observed data, T cell effector function was higher in patients with low LOX-1 expression. Separate analysis of CD4 and CD8 T cells showed that proliferation of both T cell subsets was reduced, if the percentage of LOX-1+ TAN was high (fig. S3, E and F). As expected, expression of GrzB was mainly restricted to CD8 cells, and only smaller numbers of CD4 cells expressed this molecule (fig. S3, G and H). Of note, in the presence of higher frequencies of LOX-1+ TAN, GrzB expression in CD8 T cells was substantially reduced, and this reduction was more pronounced in the tumor region. Together, these data suggest that LOX-1 could serve as a valuable surface marker to define CD66b+ cells with MDSC activity in human tumor tissues.

Fig. 4 Correlation of TAN phenotypes and TIL effector function.

Identification of TIL (CD3) was combined with GrzB (A) and Ki67 (B) as surrogate markers for T cell functionality. Identification of TAN (CD66b) was combined with LOX-1, arginase I, MPO, and HLA-DR. Density and relative percentage of Ki67 and GrzB expression in T cells was correlated with density and relative percentage of the various TAN subsets (C and D). Heatmaps (C and D) show pairwise Pearson’s correlation coefficients of the density and percentage of different phenotype of TAN with total T cells, cytotoxic T cells, and proliferating T cells in tumor and stroma, respectively. Color codes indicate positive (blue) or negative (red) correlation. Note the positive correlation of HLA-DR+ TAN with T cell effector function (blue color code). Strongest negative correlations are observed for CD66b+/ArgI+/Lox1+ and T cell effector function in the tumor core area. For (E) and (F), patients were divided into groups with relative percentage of LOX-1+ TAN above (high) or below (low) the median. Percentage of Ki67+ and GrzB+ T cells was determined in the mixed regions of tumor and stroma, respectively. Ki67, n = 30 patients; GrzB, n = 137 patients; t test.

It is tempting to speculate that direct contact or close proximity between CD66b+ cells with MDSC activity and T cells is required for the down-regulation of T cell effector function. To test this hypothesis and further substantiate the modulation of T cells by CD66b+ cells in situ, we used four-color immunofluorescence (Fig. 5, A to C) and compared the relative percentage of GrzB+ T cells, both conjugated and nonconjugated with CD66b+ cells. In both 3D (Fig. 5D and movie S4) and 2D analysis (Fig. 5E), GrzB expression was significantly reduced when T cells were engaged in conjugates with CD66b+ cells. This reduction of GrzB positivity was more pronounced in LOX-1+/CD66b+ pairs compared with pairs with LOX-1 TAN (Fig. 5F) and was additionally dependent on relative proximity between T cells and CD66b+ cells (Fig. 5, G and H). If the distance between T cells and the CD66b subject cell reached 100 μm or above, then GrzB expression was no longer influenced. Consistent with this finding, we observed a higher expression of both Ki67 and GrzB in tissue regions devoid of CD66b+ cells (T cell dominated, fig. S5, A to D). In contrast, both T cell effector molecules were down-regulated in regions of T cell to TAN/PMN-MDSC contact (fig. S5, B and D). A separate analysis of CD4 and CD8 T cell showed that Ki67 expression of both T cell subsets was reduced, if they were in conjugation with CD66b+ cells (fig. S3, I and J). In confirmation of total CD3 T cell data, conjugated CD8 cells had markedly reduced GrzB expression, whereas the much lower expression of GrzB in CD4 cells was less strongly influenced (fig. S3, K and L).

Fig. 5 Conjugation and functional modulation of TIL by LOX-1+ TAN provides evidence for PMN-MDSC activity in vivo.

Four-color immunofluorescence was performed to identify LOX-1+/CD66b+ cells and CD3+/GrzB+ cells. Representative stainings are shown for a mixed region (A), a T cell–dominated region (B) and a neutrophil-dominated region (C). The percentage of GrzB+ TIL conjugated or nonconjugated to CD66b+ cells (D and E) or conjugated to LOX-1+ or LOX-1 CD66b+ cells (F) in tumor and stroma are shown for 2D cryosections [(E and F); n = 53 patients] or in 3D whole-mount stained tissue [(D); n = 8 patients; movie S4 illustrates a 3D reconstruction of functional TIL modulation]. The percentage of Grzb+ TIL was also plotted as a function of the radial distance around different CD66b subsets (G and H). The dashed lines indicate the mean percentage of GrzB positivity for the total tissue area of all patients in the cohort. Kaplan-Meier survival curves for OS were plotted for patients with LSCC (n = 53) divided into groups according to density of LOX-1+ TAN and GrzB+ TIL being above or below the median in tumor (I) or stroma (J). Statistical significance was determined by log-rank regression analysis with the level of significance set at P ≤ 0.05. Cox multivariate regression analysis was performed by adding UICC-TNM stages, and the percentage of CD3 conjugated with CD66b+ Lox1 into a model. Low GrzB+ TIL and high LOX-1+ TAN remained significantly associated with OS (tumor, P = 0.001; stroma, P = 0.01).

To lastly test the clinical relevance of the TAN/PMN-MDSC–T cell interaction, we combined data on GrzB/CD3 expression with LOX-1/CD66b data and analyzed the correlation with patient follow-up (Fig. 5, I and J). An extremely poor outcome was recorded for patients with a low frequency of GrzB+ TIL in conjunction with a high frequency of LOX-1+ PMN-MDSC.

Collectively, our data support the concept that tumor-associated CD66b+ cells down-regulate T cell activity in situ. In agreement with a previous report, which suggested LOX-1 as a PMN-MDSC marker, our data support a higher in situ suppressive capacity of LOX-1+ CD66b+ cells compared with LOX-1 cells. In addition, our data provide evidence that this MDSC activity of LOX1+ CD66b+ cells is more pronounced in the tumor core area compared with stromal regions of the tumor tissue.


The impact of the tumor-infiltrating immune cells (the so-called TIME, tumor-immune microenvironment) on the progression of the disease and response to therapy is determined not only by complex spatial and functional interactions of these cells with other immune cells but also with tumor cells and mesenchymal cells of the tumor stroma. Understanding this interaction is important for optimization and further development of cancer immunotherapy. Recent techniques, such as high-resolution single-cell RNA sequencing, cytometry by time of flight (CyTOF), and multiparameter flow cytometry have added to this understanding by providing an unprecedented view of the composition and function of immune cells within the TME from bulk tissues. However, because of the nature of the datasets being used, these studies lack important information related to actual cellular proportions, cellular and tissue heterogeneity, as well as deeper spatial distribution (24). Immune cells freshly isolated from tissue pieces for CyTOF or flow cytometry may originate from different regions of the tumor, including tumor core, stromal, and necrotic areas, and are thus not taking into account the spatial heterogeneity of the malignant tissue (18). Even contamination with circulating blood leukocytes is difficult to exclude, and tissue processing further influences the expression of surface molecules. At the same time, it has become evident that the spatial context of immune cells is critical for tumor development (2, 25). Carstens et al. (26) defined a 20-μm radius around the cancer cells as enhanced probability of cell-cell contact distance. They found that high infiltration of cytotoxic T cells within this radius of cytokeratin 8–positive cancer cells significantly correlated with prolonged patient survival, which indicates that cytotoxic T cells in the direct vicinity of cancer cells may have a more important biological function than more distant cells. Similarly, in an oral squamous cell cancer study, the CD8+ T cell number at the tumor site had a greater effect on OS than stromal T cells (27).

In our study, we have profiled a quantitative in situ 2D and 3D tissue analysis that identifies intratumoral hotspots of immunosuppressive neutrophil activity. We found that the overall density of CD66b+ cells was higher in stroma compared with tumor core. Nevertheless, a profound prognostic relevance was observed for CD66b+ and CD66b+ Lox-1+ cell density in the tumor core area but not in the stromal zone. The strongest negative correlations were found between arginase I– and LOX-1–expressing TAN and proliferating and cytotoxic T cells in the tumor core. These data suggest a higher pathophysiological relevance of TAN-TIL interactions in the tumor core area. We believe that this has important implications for understanding TAN biology, because in most studies that isolate TAN from bulk tissue pieces, the isolated cell suspension is dominated by neutrophils from mesenchymal stromal regions, which appear to have a weaker modulation of T cell effector function.

Recent studies in murine tumor models and patients with cancer provided evidence for an important functional role of PMN or PMN-MDSC during tumor progression (16, 28, 29). These studies suggested that TAN might, at least in part, mediate the effects on disease outcome via the regulation of T cell activity. The absence of TAN correlates with increased tumor infiltration and function of activated T cells and led to a T cell–dependent suppression of tumor growth in pancreatic tumors (30). In contrast, HLA-DR+ TAN, which originate from immature progenitors, could stimulate and facilitate T cell responses (17). Supporting such a concept, we indeed found positive correlations between HLA-DR–expressing TAN and proliferating and cytotoxic T cells in tumor.

Although published studies suggest that TAN with T cell–suppressive function (i.e., PMN-MDSC) exist in human tumor tissue, until now, no evidence for such a T cell modulatory interaction of TAN and TIL in situ existed. Tissue section analyses provide the spatial context of single-cell interactions within the cancer microenvironment. Such spatial data have already aided our identification of clinically relevant features that are more powerful than simple cell counts. For example, it was reported that regulatory T cells (Tregs) more proximal to CD8+ T cells are more effective at suppressing anticancer function (27). These and similar studies prompted us to investigate the clustering and expression of functional molecules in T cells and neutrophils in patients with HNC with single-cell resolution. We used GrzB as a marker of cytotoxic activity and Ki67 to mark proliferating T cells. Although GrzB expression is necessary to demonstrate killing activity, relocalization and mobilization of cytotoxic granules toward the target cells would be required to unequivocally show the active killing process itself (31). However, such studies are mainly limited to in vitro settings and are not feasible in larger series of tissue samples of human tumors. Phosphorylation downstream of T cell receptor signaling is another means to monitor T cell activity and also often used in vitro (32). Again, these events are difficult to assess and quantify in larger series of human tissue samples because of their very transient and time-dependent nature.

Our spatial analyses suggested that clustering effects of TAN and T cells take place in a radius below 100 μm. This range was therefore used for a more rigorous evaluation of clustering behavior. The relative CD3 density was significantly higher within the distance of 25 μm to LOX1+ TAN compared with LOX1 TAN in tumor, which indicates that LOX1+ TAN have a stronger “local” clustering capacity. However, the significance disappeared with increasing distance, which may indicate that T cell inhibition by LOX1+ TAN is only functional in close proximity, similar to mechanisms described before for Tregs. It is currently unclear whether this requires direct physical interaction or whether close apposition within diffusive reach of inhibitory molecules is sufficient. Nevertheless, the lack of substantial functional difference between LOX1+ TAN and LOX1 TAN in stroma may indicate that localization of LOX1+ TAN within the tumor area may affect their function.

LOX-1 is a class E scavenger receptor that is expressed in macrophages and chondrocytes, as well as in endothelial and smooth muscle cells (33). In our study and in a smaller cohort of 10 patients with HNC and non–small cell lung cancer (15), a high variance of LOX-1 positivity in TAN was observed. Because LOX-1 is only expressed in a small proportion of circulating neutrophils (15), up-regulation of the molecule is expected to take place in the tumor microenvironment. Although the full mechanisms of LOX-1 plasticity still have to be elucidated, induction of endoplasmic reticulum stress has been suggested to be involved (15, 34). LOX-1+ PMN-MDSC also show elevated production of reactive oxygen species and arginase I (15), both of which are suggested as major mechanisms of T cell suppression. We also found a high density of TAN double positive for LOX-1 and arginase I, and their presence was most strongly correlated with a reduced expression of cytotoxic GrzB and the proliferation marker Ki67 in closely located T cells. Last, patients with concomitant high density of LOX-1+ TAN and low density of GrzB+ TIL showed extremely poor survival, reaffirming a strong tumor-promoting and T cell–suppressive function for LOX-1+ TAN.

In our study, whole-tissue sections were imaged using an automated microscope operating in tiling mode, followed by automated detection and quantification of CD66b+ cell and TIL in the whole-slide images. Despite substantial variation in tumor morphology and staining conditions, target structures could be automatically identified, resulting in increased objectivity and reproducibility of the analysis.

However, for many approaches, a histological 3D imaging of the entire organ is essential to gain a deep understanding of tissue morphology and physiological processes, also allowing us to address questions that are inaccessible by 2D methods. Therefore, we used a nondestructive imaging technique for whole-organ imaging with precise qualitative and quantitative analysis of functional structures in their natural spatial context (35). A heterogeneous distribution of TAN and TIL was noted in 2D tumor sections and in intact 3D tissue architecture. However, only the 3D analysis revealed the existence of TAN-dominated, T cell–dominated (T cell–inflamed), and mixed regions. Mixed regions were the most frequent areas observed and may also be the regions for the interaction of CD66b+ cells with T cells. The frequency of GrzB+ TIL and Ki67+ was higher in T cell–rich regions compared with the other two regions in which TAN were present.

Depleting or “re-educating” immunosuppressive myeloid populations has proven to be effective at eliciting antitumor T cell responses (3638). In a genetically engineered pancreatic ductal adenocarcinoma model, the systemic depletion of Gr1+ myeloid cells could increase the infiltration of effector T cells and inhibit tumor growth (39). Host CXCR2 inhibition by genetic ablation prevented neutrophil accumulation in pancreatic tumors and led to a T cell–dependent suppression of tumor growth (30). Therefore, inhibiting neutrophil migration into the tumor may be therapeutic in HNC. However, this is a potentially harmful therapy because the systemic depletion or inhibition of neutrophils bears the risk of secondary infections. Because we have shown that LOX-1+ TAN have a strong tumor-promoting and T cell–suppressive function, we propose that specifically LOX-1+ TAN are a promising therapeutic target.

Whole-mount tissue staining also enabled us to assess the density of both cell populations with relative distance to the tumor margin, representing a high-resolution map of tumor-infiltrating cells. The highest T cell density was reached within 20 μm off the tumor margin to both sides, and a significant decrease of T cell density in the region of 20 to 50 μm, followed by a slighter decrease with relative distance from the tumor margin, was observed. A similar distribution pattern was noted for CD66b+ cells inside the tumor region, although increased cellular densities were noted in the tumor core area beyond 200 μm from tumor margin. On the contrary, T cell density was the lowest in the tumor core, which may indicate spatial immunosuppression by CD66b+ cells.

In summary, we present here detailed data on the local spatial distribution and potential functional interaction of CD66b+ cells and T cell in the microenvironment of a solid tumor. On the basis of these data, “mixed” areas in the tumor core region are especially important hotspots of interaction with high clinical relevance. In addition, we provide experimental evidence for a potential down-regulation of T cell effector function by LOX-1+ TAN in situ. Our findings suggest that a combination immunotherapy, which concomitantly prevents TAN influx and augments T cell effector function, can be beneficial in HNC and possibly also in other cancers with a similar TAN-TIL profile.


Study design

It was the aim of this study to obtain evidence for the existence, the phenotype, the localization, and the potential clinical relevance of PMN-MDSC in human cancer tissue. To this end, patients with primary HNC and no prior radio- or chemotherapy were included. Inclusion was based on (i) tumor type (LSCC or OSCC), (ii) provision of informed consent, (iii) availability and quality of obtained tissue, and (iv) availability of clinical record and follow-up data. Survival data were obtained in a retrospective manner, and OS was chosen as the primary endpoint. Pseudonymization was applied, and during experimentation and data analysis, the primary investigator had no access to patient identities. Because of the retrospective and observational nature, no prior sample size calculation was performed. Multicolor immunofluorescence was applied on tumor samples after prior establishment of methods and antibody panels on tonsillar tissue.

Study participants and collection of tissue

Tumor tissues from patients with HNC were collected during routine surgery. Patients with prior radiotherapy or chemotherapy, synchronous carcinoma in another location, or severe concomitant systemic infectious disease were excluded. Union for International Cancer Control–American Joint Committee on Cancer (UICC/AJCC) tumor, node, metastasis (TNM) 7th edition was used to determine the HNC stage. Only material from patients with histopathologically proven HNC was included. Treatment strategies (surgery alone, surgery combined by adjuvant radiotherapy or radiochemotherapy, and primary radiochemotherapy) were developed by the multidisciplinary tumor board for HNCs, and treatments were conducted according to state-of-the art technique and guidelines in place at the West German Cancer Center. Analysis of tissues was approved by the institutional review board of the Medical Faculty of the University of Duisburg-Essen, and written informed consent was obtained from all patients. In total, material from 74 patients with LSCC [14 females and 60 males, ages 37 to 75 years (median 59 years)] and from 112 patients with OSCC [40 females and 72 males, ages 41 to 78 years (median 61 years)] was used in this exploratory and observational study. Survival analysis was performed on patients with a minimum of 36-month follow-up.

Flow cytometric analysis of tumor-infiltrating cells

Unfixed, fresh tumor tissue was digested for 45 min with dispase (200 μg/ml; Roche Applied Science, Mannheim, Germany), collagen IV (200 μg/ml), and deoxyribonuclease (10 μg/ml; both Sigma-Aldrich, Taufkirchen, Germany). Subsequently, digested samples were stained with antibodies CD45-VioGreen (clone 5B1, Miltenyi Biotec, Bergisch Gladbach, Germany), CD66b fluorescein isothiocyanate (FITC) clone 80H3, Beckman Coulter, Krefeld, Germany), HLA-DR APC (clone G46-6), CD14 APC-Cy7 (clone MphiP9), CD16 phycoerythrin (PE)–Cy7 (clone 3G8), CD11b APC-Cy7 (clone Mac1, all BD Biosciences, Heidelberg, Germany), CD10 APC (clone HI10a), and LOX-1 PE (clone 15C4, both BioLegend, Koblenz, Germany), followed by a live/dead staining using the fixable viability dye eFluor 506 (eBioscience/Thermo Fisher Scientific, Darmstadt, Germany). For staining with anti–arginase I APC (polyclonal sheep, Bio-Techne, Wiesbaden, Germany), cells were fixed and permeabilized with BD Cytofix/Cytoperm Solution Kit. Stained cells were analyzed with BD FACSCanto II using DIVA 8.01 software (BD Biosciences) or FlowJo10 (LLC, Ashland, Oregon, USA).

Whole-slide immunofluorescence staining and image acquisition

Surgical material was embedded in Tissue-Tek O.C.T. Compound (Sakura Finetek, Staufen, Germany), frozen in liquid nitrogen and stored at −80°C until use. For fluorescence staining, 5-μm tissue sections were fixed with BD Cytofix/Cytoperm (BD Biosciences) and incubated with the primary antibody overnight at 4°C, followed by secondary antibody. Cell nuclei were stained with 4′,6-diamidino-2-phenylindole (DAPI). Combinations of cell detection antibodies as well as reagents used for immunofluorescence analysis are listed in Supplementary Materials and Methods.

After staining, multichannel images were acquired on an scientific complementary metal-oxide semiconductor camera (Hamamatsu Orca-Flash, 16 bit 4.2 Mp), using an automated Zeiss AxioObserver microscope equipped with an Apotome for optical sectioning, motorized stage, Colibri LED excitation source (excitation lines: 365, 470, 555, and 625 nm), and optical filter blocks for the fluorescent labels, DAPI, or Alexa Fluor 405 (EX G365, BS 395, and EM 445/50), FITC (EX 470/40, BS 495, and EM 525/50), Alexa Fluor 546 (EX 545/25, BS 570, and EM 605/70), and Alexa Fluor 647 (EX 640/30, BS 660, and EM 690/50). All images were acquired using a Carl Zeiss 20×/0.8 NA plan-apochromat objective. For the imaging of whole-tissue sections, the microscope was operated in tiling mode; otherwise, individual images were acquired in selected regions containing both labeled T cells and neutrophils. Typically, 10 to 20 images were randomly acquired from each patient tissue section.

Tissue image segmentation and cell classification

Using Definiens Tissue Studio image analysis software (Definiens AG, Munich, Germany), the multichannel images acquired from whole-tissue sections were automatically segmented into tissue regions, and cell label classes were identified. On the basis of cell morphology and pan-cytokeratin staining, the software was trained to initially segment the tissue into four tissue classes designated tumor, stroma, necrosis/lumen, and background. DAPI staining together with cell detection antibodies were used for nuclei segmentation and cell segmentation and subsequent cell identification.

Spatial mapping of cells

Following the Tissue Studio batch processing of all patient tissue samples, a CSV file containing the patient identification, tissue class, cell label class, cell coordinates, and coexpression parameters for all segmented cells was exported and used in subsequent spatial point pattern analysis. For the analysis of cellular spatial distributions, Spatstat, an R package for spatial point pattern analysis (, was used together with an ImageJ script for masking tumor and stroma regions.

In general, spatial point patterns for the data were normalized to a corresponding model simulating CSR distribution of cells. Further methodological details on the CSR model, cell cluster analysis, and n-n analysis are provided in Supplementary Materials and Methods.

Tissue clearing and whole-mount staining

The clearing of whole HNC tumor tissue was performed according to Klingberg et al. (20) with some modifications for whole-mount labeling. In brief, fresh tumor tissue was first fixed with 4% paraformaldehyde in phosphate-buffered saline (PBS) at 4°C for 4 hours or overnight, dependent on the tumor size (>1 cm3 overnight). All incubations were performed under constant agitation and in the dark. The tissue was then washed with PBS and permeabilized by incubation in 20% dimethyl sulfoxide, 1% Triton X-100, and 2.3% glycine in PBS, for 5 days at room temperature. This was followed by blocking with wash buffer supplemented with 5% DMSO, 0.1% Triton X-100, 0.02% sodium azide, 5% normal goat serum (Dianova, Hamburg, Germany) and sodium heparin (10 μg/ml; Ratiopharm, Ulm, Germany), for 1 day at 37°C. Whole-tissue samples were then incubated in blocking buffer with rabbit anti-human LOX-1 or mouse anti-human Granzyme B at 37°C for 5 days, followed by 6-hour washing done four times. Incubation with secondary antibodies being either donkey anti-rabbit or donkey anti-mouse, both coupled to Alexa Fluor 546, was performed at 37°C for 3 days. The tissue samples were again washed, as described above, and incubated for 5 days at 37°C with mouse anti-human CD3 Alexa Fluor 647 (BioLegend; Koblenz, Germany) and self-conjugated mouse anti-human CD66b (Beckmann Coulter, Krefeld, Germany) Alexa Fluor 790. Fluorophore conjugation was carried out using an Alexa Fluor 790 antibody labeling kit (A20189, Thermo Fisher Scientific, Darmstadt, Germany), following the provider’s instructions. Afterward, the tissue samples were washed, dehydrated in a graded ethanol series of 50 and 70% for 4 hours, followed by two further 100% incubations, and performed overnight and for 4 hours, respectively. Last, the samples were cleared by incubation in ethyl cinnamate for 1 day at room temperature in the dark.

Image acquisition and data processing for 3D tissue analysis

Cleared whole-mount stained tumor samples were imaged using a light sheet ultramicroscope (LaVision BioTec Bielefeld, Germany), and multichannel images were acquired for each of the following antigens and fluorescent labels: Lox-1 Alexa Fluor 546, Granzyme B Alexa Fluor 546, CD3 Alexa Fluor 647, and CD66b Alexa Fluor 790. Image stacks were acquired on the Ultramicroscope at a magnification of ×6.4 (pixel resolution, 1.0156 μm) using a z-stepping size of 2 μm and using the dynamic focus function. The 3D rendering of images and subsequent analysis were then performed using Imaris 9.1.2 software (Bitplane, Switzerland). To this end, the manual tracing wizard was used to determine the surface and volume of the total tissue, tumor area, and stroma area. After background subtraction, the absolute number and 3D location of each cell type were obtained using Imaris “Spots” objects and the “split into surface” Spots tool. Events double positive for two signals were obtained using the Spots colocalization tool. Spot objects from cells of the same type were defined as colocalized, if they were within 8-μm distance of each other. Spot objects of different cell types were defined as conjugated for a separation distance less than 15 μm. To identify the distribution patterns of neutrophils and T cells at different distances from both inside and outside of the tumor margin, new surface objects were created, whose surfaces were respectively set to the distances 20, 50, 100, and 200 μm from the tumor surface in both directions (tumor and stroma). Then, using the split into surface Spots tool, the cell density within the volume of each new surface could be derived.

Statistical methods

Data were analyzed using GraphPad Prism software. Statistical significance of immunohistochemically quantifications were assessed with Student’s t test, paired t test, or analysis of variance, as appropriate. For the correlation analyses between different phenotype of TAN and T cell distributions, the Pearson’s correlation coefficient (r) was calculated. Clinical data were analyzed with the SPSS statistical software version 22. The immune staining results were then correlated with clinical data using univariate and multivariate Cox regression analysis with UICC tumor stage and p16 immunohistochemistry as the confounding factors. Kaplan-Meier survival curves for OS were plotted, and statistical significance was determined by log-rank regression analysis with the level of significance set at P ≤ 0.05. *P < 0.05, **P < 0.005, ***P < 0.0005.


Materials and Methods

Material used for whole-slide immunofluorescence

Antibody combinations used for multicolor immunofluorescence on tissue sections

Experimental details of CSR analysis, n-n analysis, and cell cluster analysis

Fig. S1. Flow cytometric analysis of intratumoral TAN.

Fig. S2. Clinical relevance of TAN density and LOX-1 expression.

Fig. S3. Interaction of T cells subsets with TAN.

Fig. S4. Quantification and clinical relevance of CD3-CD66b conjugation.

Fig. S5. Quantification of T cell effector function in tumor regions dominated by either T cells or CD66b+ cells.

Table S1. Univariate Cox regression analysis of OS for the LSCC cohort.

Table S2. Multivariate Cox proportional Hazard analysis of OS for the LSCC cohort.

Table S3. Univariate Cox regression analysis of OS for the OSCC cohort.

Table S4. Multivariate Cox proportional hazard analysis of OS for the OSCC cohort.

Movie S1. Identification and localization of TAN and TIL in the 3D HNC tissue architecture.

Movie S2. Segmentation of immune cell compartments in the tumor microenvironment.

Movie S3. Spatial interaction of TAN and TIL in tumor tissue in situ.

Movie S4. Spatial deconvolution of GrzB expression in TIL.

Data S1. Raw data (Excel).


Acknowledgments: We thank all clinicians of the ENT Department for supporting the retrieval of patient material for this study. We thank J. Henriksen for help with the multivariate analysis. Funding: Y.S. was supported by the China Scholarship Council and a scholarship from the Medical Faculty of the University of Duisburg-Essen. S.B., K.B., and Y.S. were supported by European Cooperation in Science and Technology (COST) Action Mye-EUNITER ( and J.K. and M.G. were supported by a research grant form LaVision BioTec. Author contributions: Conceptualization: S.B. Supervision and acquisition of funding: M.G., J.K., and S.B. Development of methodology: Y.S., K.B., S.F.M., P.J., and A.S. Performance of research: Y.S., K.B., and P.A. Data curation, analysis, and visualization: Y.S., S.F.M., B.W., A.S., and S.B. Statistical analysis: Y.S., S.F.M., P.J., A.S., and B.W. Provision of resources: S.M. and S.L. Writing of the manuscript: Y.S., A.S., and S.B. Editing of the manuscript: S.F.M., P.J., and M.G. Competing interests: J.K. and M.G. received research funding from LaVision BioTec, the maker of the light sheet microscope used in this study. The other authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions of the paper are present in the paper and/or the Supplementary Materials. Further information and requests for resources and reagents should be directed to the corresponding author (sven.brandau{at} and will be addressed where feasible.

Stay Connected to Science Immunology

Navigate This Article