PD1^hi cells associate with clusters of proliferating B-cells in marginal zone lymphoma

Fifteen MZL biopsy samples (spleen = 3, lymph node = 7, periorbital = 2, parotid, lung, thyroid = 1 each) were obtained from Leicester Royal Infirmary under Research Ethics Committee 14/EM/1176. (female = 11, male = 4; median age 63.5 years (range 48–74 years)). The characteristics of the patients and the treatment they received are shown (Table 1).

Immunofluorescence (IF) staining was performed using Opal 4-color fluorescent IHC kit (NEL800001KT, Perkin-Elmer, Waltham, MA, USA). Following antigen retrieval formalin fixed and paraffin embedded (FFPE) sections were incubated with protein block (RE7102, Novolink™ Polymer Detection System, Leica Biosystems, Wetzlar, Germany) for 10 min. For multiplex staining antibodies were used sequentially, first, anti-CD20 (1:100) (M0755, DAKO, Glostrup, Denmark), second anti-PD1 (used undiluted) (ab52587, Abcam, Cambridge, UK) and lastly anti-Ki67 (1:100) (M7240, DAKO) antibodies were incubated with tissue sections for 30 min at room temperature. Following incubation with HRP conjugated goat anti-mouse-IgG (1:200) (P0447, DAKO) for 1 h at room temperature slides were washed and incubated for 10 min at room temperature in the dark with Opal 520 (dilution 1:50) for anti-CD20, Opal 670 (1:100) for PD1 and Opal 570 (dilution 1:50) for Ki67. After additional washes nuclei were counterstained with DAPI (D1306, ThermoFisher). A negative control slide, incubated with mouse IgG1 (X0931, DAKO) was included in each staining run. Images of the stained slides were recorded using NanoZoomer-XR Digital slide scanner C12000–01 (Hamamatsu Photonics, Hamamatsu, Japan) and the images were pseudo-coloured and merged by using ImageJ analysis software [16].

In order to identify specific T-cell subsets by multiplex immunohistochemistry (workflow illustrated in Fig. 1a) FFPE tissue sections were stained with anti-CD4 (4B12, DAKO) (1:80 dilution for 20 min), anti-PD1 (EuroMabNet, clone NAT105) (used undiluted for 60 min) and anti-FOXP3 (EuroMabNet, clone 236A) (1:100 dilution for 60 min) [17, 18]. For all antibodies antigen retrieval was with Tris-EDTA pH 9. Peroxidase-coupled secondary antibody (MP-7452, Vector Laboratories, Burlingame, CA, USA) was employed with AEC-peroxidase substrate (SK-4205, Vector Laboratories). Tissue sections were processed through sequential staining and destaining (glycine (25 mM), 1% SDS, pH 2) following which image data was acquired and downstream processing carried out. This involved registration of the separately stained images followed by data analysis, which includes both enumeration of stained cells and also spatial information i.e. co-ordinates of each cell (Visiopharm, Hoersholm, Denmark). All images that were taken forward for data analysis were carefully checked for successful registration. Persisting antibody from early rounds of staining could be a confounding factor and we, therefore, confirmed that the elution process reduced residual staining to background levels (Fig. 1b). Visiopharm apps were developed (available on application to the authors) to paint Tfh cells (CD4⁺PD1^hiFOXP3⁻) orange, Treg cells (CD4⁺PD1⁻FOXP3⁺) yellow and Tfr cells (CD4⁺PD1^hiFOXP3⁺) turquoise (Fig. 1c and d) prior to extracting cell numbers and location coordinates. An intensity threshold for the definition of high PD1 expression was set by utilising PD1 expression of Tfh cells within tonsillar germinal centres from healthy subjects.

Tissue sections from 6 cases (Fig. 2b) were stained with anti-CD20 (M0755, DAKO), anti-CD4 (M7310, DAKO), and anti-CD8 (M7103, DAKO). Surface plots generated in ImageJ (https://​imagej.​net/​Fiji) are shown for each stain. Intensity has been normalised so that images can be compared.

Distance from either FOXP3⁺ cells or PD1^hi cells to the nearest Ki-67 cell was plotted for 10⁵ cells sampled from 10 cases. All clustering and spatial analysis was performed using the statistical programme R (Spatstat package [19] and ggplot2 package [20]).

R was also used to determine Ripleys K function, an analysis tool to quantify the spatial distribution observed for each cell subtype as compared to complete spatial randomness (CSR).

Other statistical tests (Mann-Whitney, Kruskal Wallis tests and linear regression) were carried out with Prism 6 software (GraphPad Software, La Jolla, CA). Gene set enrichment analysis [21] (http://​software.​broadinstitute.​org/​gsea/​index.​jsp) was employed to analyse gene expression data from a publicly available database [22] under accession numbers GSE32233 and GSE32231.

To gain an understanding of T-cell distribution in MZL multiplex semi-automated immunohistochemistry was carried out for CD20, CD4 and CD8. Whole tissue heat-maps show the density of infiltrating T-cells (Fig. 2a). These reveal T-cell zones comprising low numbers of CD8⁺ and abundant CD4⁺ T-cells with clearly differing patterns of spatial localisation.

To provide support for the presence of activated T-cells in the MZL TME we carried out gene set enrichment analysis on data from nodal MZL [22]. Two signatures (GSE2770 - genes up-regulated in CD4⁺ lymphocytes induced to differentiate into Th1 and Th2 by treatment with IL-12 and IL-4 respectively in the presence of TGFbeta and GSE26928 - genes up-regulated in effector memory T-cells as compared to CXCR5⁺ T-cells) of polarised CD4⁺ T-cells (Fig. 2b) were obtained (FDR < 0.25). The tissue density maps and GSEA together suggest effector CD4⁺ T-cells are present in the tumour microenvironment.

Next we used image analysis to overcome biased representation due to counting relatively small fields and acquired similar quantitative information about infiltrating CD4⁺ T-cell subsets (Tfh, Tfr and Treg).

We defined Tfh cells as CD4⁺PD1^hiFOXP3⁻, Tfr cells as CD4⁺PD1^hiFOXP3⁺, and Treg cells as CD4⁺PD1⁻FOXP3⁺ (Fig. 1c and d). Collectively these three populations accounted for 16.4 ± 3.6% mean ± SEM) of total CD4⁺ cells (n = 8). Tfr cells represented only a rare population accounting for 4.6 ± 1.3% (mean ± SEM) of CD4⁺PD1^hi and were not anlaysed further. We found strong associations between PD1^hi and Tfh cell numbers (R² = 0.98; P < 0.0001) and FOXP3⁺ and Treg numbers (R² = 0.64; P = 0.017). Therefore, to facilitate this analysis and future translational lymphoma research we followed others in employing single markers, either PD1^hi or FOXP3⁺ [12, 23] as surrogates for Tfh and Treg subsets respectively (Fig. 2c).

Total numbers of PD1^hi and FOXP3⁺ numbers varied over ~ 20-fold with some cases having very much heavier T-cell infiltration than others (Fig. 3a). The ratio (PD1^hi:FOXP3⁺) also varied with activating PD1^hi cells the majority in some cases whereas suppressive FOXP3⁺ cells were dominant in others.

Analysing SMZL, nodal and extranodal MZL separately showed no significant differences between overall density of PD1^hi or FOXP3⁺ cells or PD1^hi:FOXP3⁺ in the three groups (Kruskal Wallis test) (Fig. 3b). There were also no significant differences in T-cell density or PD1^hi:FOXP3⁺ when the cases were analysed by stage (Fig. 3c).

Next, we chose to explore the spatial distribution of infiltrating PD1^hi and FOXP3⁺ cells, having observed various patterns of CD4⁺ T-cell infiltration (Fig. 2a), in order to determine whether these cells are randomly distributed or clustered. Ripley’s K test, was applied to data from whole tissue sections (mean area 50.7mm², range 4.8–112 mm²) (n = 15) (Fig. 4a). Overall we found that PD1^hi cells were significantly more clustered i.e. greater separation of observed curve (solid black lines in Fig. 3a) from CSR (dotted red lines), than FOXP3⁺ cells (median cluster score 1.84 for PD1^hi cells and 0.45 for FOXP3⁺ cells; Mann-Whitney test P = 0.003) (Fig. 4b).

Tfh cells strongly drive B-cell proliferation and even extra-nodal CD4⁺PD1^hi T-cells are able to secrete factors enabling B-cell help [24]. We hypothesised that the PD1^hi cells might associate with proliferating lymphoma B-cells. To support this idea we first carried out immunofluorescence microscopy on a single case (Fig. 4c). This showed that proliferating Ki-67⁺ cells were predominantly B-cells and formed clusters but without the structure of normal or residual germinal centres. There was also an apparent association between areas with a high density of proliferating cells and PD1^hi cells. In order to support the hypothesis that PD1^hi cells associate with Ki-67⁺ lymphoma B-cells we quantified the spatial interaction between Ki67⁺ cells and Ki67⁻PD1^hi T-cells. Cell co-localization was objectively determined by virtually tessellating stained tissue sections into non-overlapping equally sized quadrants prior to cell counting within each defined spatial area [25]. Two measures, Pearson correlation and Morisita-Horn index, an ecological measure of interaction (illustrative examples Fig. 5a), were employed to analyse data from 17,000 200μm² quadrants from 10 patients. Ki67⁺ and PD1^hi cells demonstrated closer interaction than did Ki67⁺ and FOXP3⁺ cells (Pearson correlation 0.4 vs 0.03; Morisita-Horn index 0.46 vs 0.25, Fig. 5b). To further interrogate spatial interaction we compared the distance between PD1^hi or FOXP3⁺ cells and the nearest Ki67⁺ cell in a nearest neighbour analysis. Ki67⁻PD1^hi cells (median distance 11.3 μm) were significantly more likely to be found in close proximity to Ki67⁺ B-cells than were Ki67⁻FOXP3⁺ cells (median distance 35.1 μm) (Mann-Whitney test P = < 0.0001) (Fig. 5c) suggesting for the first time that infiltrating Tfh cells might have a role in driving proliferation, and hence contribute to the varying clinical course of MZL.