The Adaptive and Innate Immune Cell Landscape of Uterine Leiomyosarcomas

Reactivation of the anti-tumor response has shown substantial progress in aggressive tumors such as melanoma and lung cancer. Data on less common histotypes are scanty. Immune checkpoint inhibitor therapy has been applied to few cases of uterine leiomyosarcomas, of which the immune cell composition was not examined in detail. We analyzed the inflammatory infiltrate of 21 such cases in high-dimensional, single cell phenotyping on routinely processed tissue. T-lymphoid cells displayed a composite phenotype common to all tumors, suggestive of antigen-exposure, acute and chronic exhaustion. To the contrary, myelomonocytic cells had case-specific individual combinations of phenotypes and subsets. We identified five distinct monocyte-macrophage cell types, some not described before, bearing immunosuppressive molecules (TIM3, B7H3, VISTA, PD1, PDL1). Detailed in situ analysis of routinely processed tissue yields comprehensive information about the immune status of sarcomas. The method employed provides equivalent information to extractive single-cell technology, with spatial contexture and a modest investment.

www.nature.com/scientificreports www.nature.com/scientificreports/ Leiomyosarcomas originating from myometrium (ULMs), have not been intensively investigated because of their low incidence and because they are perceived as a minor target for immune intervention, given the few tumor TILs on H&E sections and the low mutational burden 20 .
As few as two dozens ULMs have been treated with checkpoint inhibitors, with dismal results [21][22][23] . However, with the exception of one responsive case 22 , in none of them the TILs or the macrophages have been thoroughly studied.
Here we present a comprehensive high-dimensional analysis of the inflammatory infiltrate in 21 sarcoma cases with a panel of 40 markers including lineage specific leukocyte proteins, activation markers and component of the immunological synapse, by using a novel robust method, effective on routinely processed materials, but capable of highly detailed single cell analysis.

Materials and Methods
Patients and case selection. 21 cases of leiomyosarcoma were selected out of 77, based on both full clinical history and tissue block availability. The female patients were aged 51.2 ± 11.8 years (34-69), 52% post-menopausal, 11 FIGO stage IB, 1 stage IIB, 5 stage IIIB/C, 3 IVB, one unknown. 13 were classified TNM pT1, 1 pT2, 6 pT3. The biological parameters are reported in Table 1.
Two random 2 mm cores of non-necrotic, non-sclerotic, full tumor tissue for a total of 6.28 mm 2 per case (equivalent to 40 HPF) were placed in a Tissue Microarray block (TMA model CK4500, Integrates Systems Engineering srl, Milan, Italy).

Multiple iterative labeling by antibody neodeposition (MILAN). Dewaxed, antigen retrieved
3 µm sections were processed for indirect IHC or multiple labeling as previously described in detail 24,25 and Supplementary Data.
Briefly, the sections were incubated overnight with optimally diluted primary antibodies in combination of four, washed and counterstained with four distinct fluorochrome-tagged secondary antibodies 25 . The slides, counterstained with DAPI and mounted, were scanned on an S60 Hamamatsu scanner (Nikon, Italia) at 20x magnification, after which the stains were removed with a beta-mercaptoethanol and sodium dodecyl sulphate mix, extensively washed and re-staind for the subsequent markers 25  www.nature.com/scientificreports www.nature.com/scientificreports/ 30 cycles showed no consistent antigen loss and occasional increased antigenicity ( Supplementary Fig. 1). The list of primary and secondary antibodies is available in the Supplementary Materials.

Preparation of immunofluorescent images for image analysis.
Single.ndpi images for each case were registered via alignment of DAPI nuclear stained images with Fiji, saved as.tiff files and autofluorescence was subtracted 25 .
Two single cell masks were produced for each case with Cell Profiler (2.2.0) 26 by segmentation of thresholded images: a DAPI mask encompassing all the nuclei and a mask obtained by the combined CD43, CD45, CD68 and CD163, henceforth named "targeted mask". Regions of interest composed of small portions of dendritic-shaped cytoplasm were excluded digitally (see Supplementary Methods).
Comparison of the total cell yield and detailed high-dimensional phenotypic content obtained with the DAPI vs the targeted masks showed variable but constantly inferior cell number representation with the former (36%, 65%, 95%of the targeted mask-derived cells) and loss of minor phenotypic subsets by tSNE and Phenograph clustering (not shown). This because macrophages have reduced chromatin DAPI staining and do not provide enough contrast to be detected by threshold algorithms.
High-dimensional analysis. Individual masks and .tiff files from all cases were loaded in HistoCAT 27 , data obtained from the image analysis were processed by dimensional reduction and unsupervised clusterization algorithms, t-SNE and Phenograph plots were generated. Image analysis data were subsequently exported as .csv files. In order to analyze the infiltrate composition of individual cases, we developed an R script (Supplementary Methods) to systematically process all cells of all cases identified by image analysis. Phenoclusters were plotted as heatmaps ( Supplementary Fig. 2) with a custom R script, to allow the identification of cell composition. Each heatmap was inspected for specific lineage-or function-associated markers, with the requirement that each defining marker needed to be expressed at high levels (clearly visualizable by a blue-red divergent color palette) in a third or more of the cells. A nine cell-type classification of inflammatory infiltrate was obtained by the inspection of all generated heatmaps and was based on criteria listed in Supplementary Table 1. The cell content of each phenogroup was used as the numerator to quantify the percentage of a given marker or phenotype.
Neighborhood analysis. An unbiased quantitative analysis of cell-cell interactions was performed using an adaptation of the algorithm described in 27 for neighborhood analysis to systematically identify social networks of cells and to better understand the tissue microenvironment (Supplementary Methods).

PTEN status by FISH.
In situ hybridization for PTEN was performed with the ZytoLight SPEC PTEN/ CEN 10 dual color probe (ZytoVision GmbH, Germany) for the centromeric and the gene-specific regions of chromosome 10.

Results
The clinicopathologic data of the 21 sarcomas are reported in Table 1.

The inflammatory infiltrate.
High-dimensional analysis of all 21 cases showed a majority of independent, non-overlapping clusters of myeloid phenotype, one or two per case, and smaller overlapping clusters, comprising T-cells and endothelial cells (Fig. 1). Only in four instances (cases N. 17,18, 20, 21) myeloid phenoclusters from separate cases did overlap (Fig. 1).
Thus, the majority of the infiltrating inflammatory cells in each case is composed of macrophages whose phenotype reflects the unique biology of each tumor (Fig. 2), and a minor population of T-cells.
In order to understand the composition of the inflammatory infiltrate, each sarcoma case was analyzed separately in high-dimension (Supplementary Figs. 2 and 3).

Lymphoid cells. TILs, almost exclusively T-cells and NK-cells, represents 3%-29% of the inflammatory infil-
trate (0.3%-15.3% of the total sample cellularity), the rest being myelomonocytic cells (Fig. 2 CD8+ T-cells were identified as distinct phenoclusters in about half of the cases, whenever a sufficient number of TILS was present. In those cases, often multiple phenotypically distinct phenoclusters were detected per case, displaying evidence of activation (CD69) and exhaustion (PD1, TIM3, VISTA, CD39). VISTA+ T-cells were observed in 8 cases, largely CD8+ TCF7−. TCF7, a transcription factor linked to resident memory phenotype and reactivation, was contained in 42% ± 18% of CD8+ cells, in an inverse relationship with PD1 ( Figs. 3 and 4).
Lymphocytes expressing GranzymeB and Granulysin, partly CD8+, were identified as a separate phenocluster in some cases or as part of a single cluster of cells in samples with fewer TILs.
NK cells, defined as CD45+ CD3− and expressing GranzymeB and Granulysin, were a minority of lymphoid cells in abundant infiltrates only.

Myelomonocytic cells.
Myelomonocytic cells represents 64% ± 13% of the infiltrate (Fig. 2, Supplementary Table 2 and Supplementary Data) and contained the most diversified cell type, often represented by multiple distinct phenogroups within each case (Fig. 5).
The most distinctive group, named Tumor Associated Macrophages (TAM; 38% ± 13%) expressed CD16 and restricted lineage markers (CD68, CD163) and was consistently present in all tumor cases, but only occasionally www.nature.com/scientificreports www.nature.com/scientificreports/ in the non-neoplastic samples (Case #8 and 3 normal myometria, not shown). TIM3 expression among myelomonocytic cells was restricted to this cell type and to the inflammatory monocytes (see below).
All cases but five contained a small but very distinct population of small monocytic cells, often intravascular and proliferating, with a distinctive LYZ+ VISTA+ phenotype, which we defined "inflammatory" because of the association with the inflammatory infiltrate. These cells were often CD14+, CD16 low or negative, sometimes TIM3+, but devoid of other myelomonocytic markers.

Spatial relationships between lymphoid and myelomonocytic cells. A neighborhood analysis
between all phenotypic subsets revealed a substantial mutual avoidance across the phenotypes and clustering together of similar cell types, both at submicroscopic (<100 µm range) and at microscopic range (>100 µm range) (Supplementary Fig. 4). TILs avoid close contact with checkpoint-bearing macrophages ( Supplementary Fig. 4 and Methods).

Tumor cells.
Tumor cells had histocompatibility antigens staining patterns which could be defined as constitutive and induced; in the first case tumor cells were all positive or negative throughout the sections ( Supplementary Fig. 5). In the second case, tumor cells admixed with the inflammatory infiltrate showed increased staining, compared to non-inflamed portions ( Supplementary Fig. 5). www.nature.com/scientificreports www.nature.com/scientificreports/  www.nature.com/scientificreports www.nature.com/scientificreports/ No HLA-DR was observed on tumor cells. Tumor cells did not expressed most of the markers tested, including PDL1, with the notable exception of MCM5, IDO, B7H3 and Axl (see Table 1 and Supplementary Fig. 5). Occasional PD1+ tumor cells were observed in some cases ( Fig. 5B and Supplementary Data). CD34+ endothelial cells stained for VISTA and B7H3 in about half of the cases, often in a subset 28 ( Supplementary Fig. 5).
The vast majority of the cases tested were either diploid or had loss of one copy of the PTEN gene in 3.1%-22.1% of the cells (Table 1), with only one case (case #7) carrying homozygous PTEN deletion in 13.2% of the cells.

Discussion
Data about the inflammatory infiltrate in untreated primary sarcomas are scarce. We detail here the composition of the innate and adaptive arm of the response of the host to a soft tissue tumor.
We found a variety of individual phenotypic profiles, not unlike other reports 29,30 . In 11/21 cases we found a T-cell phenotype consistent with antigen exposure and acute or chronic stimulation, leading to exhaustion, mostly in tumors HLA Class I+.
The remaining 10 cases that we classified as "deserted" had usually less than 60 lymphocytes per mm 2 and 0.3%-4.2% of total cellularity, with a phenotype suggestive of passer-by. These tumors were all negative or weakly expressing HLA-A,B,C, with two exceptions.
Each of the cases hosting antigen-experienced T-cell phenotypes displayed an unique combination of cell markers, which could be summarized as acutely exhausted CD8+ T cells, chronically exhausted CD8+ and FOXP3+ CD4+ T cells. These phenotypes were reminiscent of T-cells acutely exposed to a persistent antigen such as a virus or a neoantigen, bearing PD1, CD69, CD39 31,32 and TIM3 33 , and induced to anergy (exhaustion) 3 . Others expressed TIM3, lower levels of PD1 and TCF7 to a variable amount (17-73%), resembling exhausted CD8+ T-cells chronically exposed to an antigen, but in a resident memory state, susceptible of reactivation upon re-stimulation or therapeutic checkpoint inhibition reversal 33,34 . A mixture of these two cell types was sometimes present. It has been reported that up to a third of the cases displaying a CD8+ TCF7+ phenotype may benefit from a checkpoint blockade therapy 33 . www.nature.com/scientificreports www.nature.com/scientificreports/ CD4+ T-cells had a FOXP3+ regulatory phenotype, occasionally as the minor population, as shown in other cancer models 35 .
The variability of T-cell phenotypes found between and within cases is suggestive of an ongoing editing of the adaptive immune response 36 , on a case-by-case basis.
The tumors we have examined are full blown malignant tumors of low mutation rate, yet, as shown in a single successful immunotherapy case 22 , this may induce an anti-tumor T-cell response. ULMs are also conspicuously devoid of HLA-DR+ CD14− dendritic cells and tertiary lymphoid structures, thus an adaptive response may occur in tumor-draining lymph nodes.
The failure of single-agent checkpoint inhibitor therapy in several published cases [21][22][23] is at odd with our findings. One hypothesis is that the antigen to which the adaptive immune system responds is not a tumor antigen 37 . In the single ULM case studied 22 , T-cells were tumor antigen-specific and responded. None of our cases except one (Table 1)   www.nature.com/scientificreports www.nature.com/scientificreports/ Alternatively, failure to elicit an anti-tumor response may be caused by an independent immunosuppressive effect brought by the inflammatory infiltrate, chiefly the infiltrating macrophages.
Macrophages, and in particular tumor-associated macrophages (TAMs) have been shown to mediate the suppression of an anti tumor response 38,39 . A dichotomic view of alternatively polarized macrophages has let to a more nuanced picture 40 , where there is a dynamic equilibrium between various defined stages of macrophage polarization.
As published by others 16,41 , we found a range of leukocytes (CD45+, CD43+) with myelomonocytic differentiation. Differently from lymphocytes, whose aggregate phenotype was conserved across all cases, myelomonocytic cells had unique, case-specific populations. A tissue-restricted secretome influencing macrophages has been described 19 ; a similar effect may occur via monoclonal sarcoma cells, yielding a highly diversified, stimulus-driven differentiation reflecting individual tumor-specific microenvironment 30,42,43 . As shown by others 30,38,44 , the tumor harbors multiple subsets, often not found in the normal counterpart neither described before.
Furthermore, our neighborhood analysis of the macrophage subsets show both a submicroscopic and a microscopic local differentiation, producing a checkered pattern, which has to be kept in mind in order to sample the tissue adequately, as we did with larger (2 mm), multiple TMA cores.
Despite the large variety of individual phenotypes, by analyzing single cases with multiple markers we could identify five rather consistent groups present in all cases: histiocytes, phagocytes, TAM, inflammatory monocytes and Monocytes/Macrophages, undefined.
With the exception of TIM3, largely restricted to CD16+ TAMs, all the immuno-modulating markers were expressed in a minority of cells and were distributed across the five subsets.
We observed PD1, previously reported in circulating monocytes in HIV and in M2-type TAMs 45 expressed on macrophages in 18/21 cases, occasionally with activation markers such as HLA-DR, OX40 and/or CD83. Interestingly, non-lymphoid PD1 expression in FFPE material could only be detected with UMAB197; this antibody has a broader reactivity than other PD1 antibodies and detects the molecule on B cells, monocytes and tumor cells, as reported by others 45,46 . Subtle subcellular variations in membrane staining by each antibody we tested ( Supplementary Fig. 6) may has to do with the recognition of glycosylation-dependent, thus cell type restricted epitopes on PD1.
Consistent with the mouse data 45 , PD1 was expressed on 27 phenogroups, only two of them HLA-DR+ and CD83+.
Markers of activation (HLA-DR, CD83, OX40) were found on macrophages in almost all cases in all subsets, except on the inflammatory monocytes. Often activation was matched with proliferation (MCM5+).
More complex to understand was the expression of members of the immunological synapsis (PD1, PDL1, B7H3, VISTA), which, with the exception of TIM3, restricted to TAMs, were displayed by several subsets. PDL1, expressed by myelomonocytic cells in 15/21 cases, is a ligand for PD1 and CD80 on cognate T-lymphocytes 6,7 and a constitutively negative signaling molecules on macrophages 47 . Engagement of PDL1 induces proliferation, survival and upregulation of MHC Class-II, CD86 and cytokine secretion, promoting a proinflammatory phenotype 47 . In our PDL1-negative sarcoma cases, there is little relationship between activation and PDL1 expression in macrophages; of the 24 PDL1+ phenogroups, only seven express HLA-DR and eight CD83. These data hint at a lack of engagement of this molecule on macrophages.
VISTA-expressing macrophages have been described in prostate cancer 48 , particularly after chemotherapy; we do see this phenotype largely restricted to endocapillary LYZ+ VISTA+ monocytes, occasionally proliferating, which could be the seeding population from the bloodstream.
Human macrophages found in tumors may be able to present antigens to T-cells 41 , where the co-inhibitory molecules act to protect the macrophage from cytotoxicity during the encounter. We failed to demonstrate a statistically significative close interaction between T-cell subsets and checkpoint molecules-bearing macrophages, except for the suggestive images provided by tissue snapshots (Fig. 4). This may has to do with the highly motile properties of T-cells, whereby during an asynchronous activation of multiple immune synapsis, the challenged T-cells may distance themselves from the interactor by the time the phenotype has changed to reflect the effect of the challenge 49 .
In summary, we have described a variety of innate and adaptive immune cell phenotypes in ULMs, suggestive of antigen experience, and exploitable for a targeted immune intervention, despite the variability in phenotypes, particularly on the macrophage side.