Spatial clustering of immunosuppressive macrophages in papillary renal cell carcinoma

Abstract

Background

Papillary renal cell carcinoma (pRCC) is the second most common kidney cancer subtype, yet our understanding of its tumor immune microenvironment (TIME) remains limited.

Objective

We utilized multiplex immunofluorescence (mIF) and spatial transcriptomics (ST) to evaluate immune cell architecture in pRCC contrasted with clear cell RCC (ccRCC).

Methods

Localized RCC tumors (16 pRCC, 70 ccRCC) underwent mIF using markers for T cells, B cells, and tumor-associated macrophages (TAMs). Spatial data in both tumor and stromal compartments of the TIME were collected. A post hoc recurrence free survival analysis (RFS) was performed using Cox proportional hazard models. Single-cell ST was performed on a subset of samples, utilizing probes against 960 transcripts. Cell density, cell spatial clustering, and spatially varying gene expression were analyzed.

Results

Immune cell density was statistically lower in pRCC amongst functional CD8T cells, while cell clustering was higher amongst M2-like macrophages. Using ST, two genes (CCL18, GPNMB) were enriched in clustered M2-like macrophages in pRCC (FDR < 0.001) and are known markers of lipid-associated TAMs (LAMs).

Conclusion

Compared to ccRCC, pRCC has greater M2-like macrophage clustering. Using ST, M2-like macrophage clustering corresponds with lipid associated TAMs (LAMs), and therapeutics against this myeloid subset are currently being tested in pRCC.

Introduction

Immune checkpoint blockade (ICB) has revolutionized the treatment of advanced renal cell carcinoma (RCC), but predictive biomarkers for the response to immunotherapy remain a unique challenge. Much of what is known about RCC's response to ICB is derived from studies of the tumor immune microenvironment (TIME), which have been the most well-characterized amongst patients with clear cell histology (ccRCC).^1–6 The treatment-naive ccRCC TIME typically has an inflamed phenotype, with a high degree of T-cell infiltration, while the ICB resistant TIMEs in ccRCC have been defined by exhausted T cells, M2-like macrophage polarization, endothelial enrichment, and subsets of cancer-associated fibroblasts.^5,7–9 These translational findings will hopefully continue to enhance patient selection and personalization of treatment for this highly immune responsive histological subtype of RCC.

In contrast, the impact of ICB has been more disappointing in non-clear cell RCCs. Papillary RCC (pRCC) is the second most common histological subtype, comprising 15% of all kidney cancer diagnoses and the majority of non-clear cell RCCs. Objective response rates in non-clear cell studies of first line ICB only regimens remain very low at 14% compared to 42% in ccRCC.^10,11 This stark difference highlights an urgent clinical need to understand the biological basis for variable therapeutic responses across RCC subtypes. . Distinct features of the TIME may explain some of these clinical observations. Immunogenomic studies, dominated by The Cancer Genome Atlas (TCGA) data, have indicated that multiple immune gene signatures are downregulated in pRCC relative to ccRCC, and this largely ‘immune-cold’ phenotype has been corroborated in immunohistochemistry (IHC) studies as well.^12,13 However, beyond poor immunogenicity and immune activation, additional biological explanations for this discrepancy have not been elucidated.

Recent work has demonstrated that spatial clustering of specific immunosuppressive immune cells portends a worse prognosis in ccRCC. ³ Mechanistically, spatial immune organization may reflect phenomena such as immune cell exclusion, spatial segregation or aggregation of immune phenotypes, or specific cell-ligand interactions between immune and tumor cells, or combinations thereof. To date, spatial profiling of the TIME in pRCC has not been systematically performed. The aim of this study is to characterize the immune cell composition and spatial organization of the tumor immune microenvironment (TIME) in papillary renal cell carcinoma (pRCC) using multiplex immunofluorescence (mIF) and spatial transcriptomics (ST) by directly comparing microarchitectural features to those of the more common kidney cancer histological subtype ccRCC.

Methods

Patient and sample selection

We obtained primary tumor samples from patients with RCC, from the years 2004–2020, through protocols approved by the institutional review board (H. Lee Moffitt Cancer Center and Research Institute's Total Cancer Care protocol; Advarra IRB Pro00014441 and MCC20148). All methods were performed in accordance with the relevant guidelines and regulations stipulated by H. Lee Moffitt Cancer Center and Research Institute's Institutional Review Board (IRB). Written informed consent was obtained from all tissue donors. Samples from primary kidney tumors were obtained via surgical excision. Patients with ccRCC and pRCC were included based on the histological diagnosis at the time of the initial clinical pathology report. Patients that received systemic treatment prior to surgery and those with stage IV RCC were excluded, as cytoreductive nephrectomy in the setting of metastatic pRCC is a rare procedure

Clinical data were collected for each patient by chart review including demographic information, clinical staging, and key pathological variables such as stage, histology, and nuclear grade (either Fuhrman or International Society of Urological Pathology grading).

Sample preparation

Formalin-fixed paraffin-embedded (FFPE) tissue blocks were prepared for mIF (n = 86). An experienced genitourinary pathologist (JD) reviewed each slide and annotated seven spatially distinct regions of interest (ROIs) from three spatially distinct compartments: two ROIs from the tumor core, three from the tumor-stroma interface, and two from the stroma (Figure 1). Full details of this procedure have been described previously. ³

Figure 1.

Tissue analysis workflow and datasets. Panel 2 displayed on top of Panel 1 in WS-mIF workflow. pRCC = papillary renal cell carcinoma, ccRCC = clear cell renal cell carcinoma, FFPE = formalin fixed paraffin embedded, mIF = multiplex immunofluorescence, ST = spatial transcriptomics.

A subset of patient samples (n = 11) was prepared for ST. Tumor-core and adjacent normal tissue (herein described as ‘stroma’) were annotated by JD on FFPE blocks and prepared for 1 mm cores on a tissue microarray (TMA).

Multiplex immunofluorescence

Orthogonal tissue slides were sequentially stained in two panels using antibodies against CD3, CD8, forkhead box P3 (FOXP3), and T-box transcription factor TBX21 (T-Bet; a T-box protein expressed in T cells) on panel 1, and CD20, CD68, CD163, CD206, and programmed death-ligand 1 (PD-L1) on panel 2. For the patient samples that underwent ST, an orthogonal TMA-derived slide was prepared for mIF (TMA-mIF) using a subset of the above markers on one antibody panel (CD3, CD68, and FOXP3). Details of the whole slide (WS) mIF workflow used for all patient samples have been described previously for the patients with ccRCC and additional information is available in the supplementary materials, including in Supplementary Table 6. ³

Spatial transcriptomics

The CosMx Spatial molecular imager (SMI) (Nanostring, Seattle, WA) is an enzyme-free, amplification-free methodology that performs multiple nucleic-acid hybridization cycles of fluorescent molecular barcodes to enable in-situ measurement of RNA on intact biological samples at subcellular resolution. Full details of the CosMx chemistry and workflow can be found in He et al. 2021. ¹⁴ A list of the 960 probe targets can be found in Supplementary Table 1.

The full workflow including cell segmentation, quality assurance, and cell phenotyping has been described previously and additional information is available in the supplementary materials, including in Supplementary Table 7. ⁹

Geospatial analysis

Total cell count, positive cell counts for each IF marker, and Cartesian coordinates of each cell were then recorded. Cell counts for selecting double positive markers were derivative from this dataset. Cell spatial clustering was quantified using a framework based on Ripley's K estimates. Full details of this framework can be found in the supplementary materials. The same global Ripley's K framework and radius were used for ST-phenotyped cells on SMI and calculated for each TMA core. A local single-cell implementation of Ripley's K (neighborhood density function) was also calculated for each M2-like macrophage in the pRCC cores using the localk function from the spatstat package. ¹⁵

Bulk RNA transcriptomic and whole exome sequencing

To profile patients using established molecular definitions, tumor specimens were sent for bulk RNA-seq and whole exome sequencing (WES) according to our institution's Total Cancer Care (TCC) protocol. An RNA-based classifier was built using TCGA data to estimate patient molecular classification described in Ricketts et al.. ¹² The TCC protocol for RNA and WES has been previously described. ¹⁶ Additional information is available in the supplementary material.

Statistical analysis

Cell densities and spatial clustering were compared across RCC histologic subtypes for the four tissue compartments (tumor-core, stroma, the tumoral side of interface, and the stromal side of interface). Between-group comparisons were made using Wilcoxon signed-rank testing.

Global differences in cell density and spatial clustering between pRCC and ccRCC were estimated using mixed-effect modeling. Generalized linear mixed-effect and linear mixed-effect models were used to compare cell density and spatial clustering for pRCC and ccRCC. Full details of the statistical procedures for these models can be found in the supplementary materials.

Multiple p-value adjustment in comparative analyses between pRCC and ccRCC was made using Benjamini-Hochberg (ie. False-discovery rate, FDR) between mIF markers belonging to the same cell class (T-cell, macrophage, or B cell). One-step FDR was used for p-value adjustment of all ST tests. Statistical significance was defined as an FDR <0.1.

A post hoc recurrence free survival analysis was performed on the 16 patients with pRCC and WS mIF data. Ripley's K of markers that were spatially more clustered in pRCC compared to ccRCC were tested using Cox proportional hazard models. Log transformation was applied to Ripley's K prior to analysis. Tumor and stromal compartments were consolidated (tumor and interface-tumor, stroma and interface-stroma) and a random effect was added for repeated measurements. A per patient analysis was also performed using the average log transformed Ripley's K of select markers.

All statistical analyses were conducted using R statistical software v. 4.3.0–4.3.1 (Vienna, Austria).

Results

Baseline patient characteristics

70 ccRCC patient samples and 16 pRCC patient samples underwent whole slide (WS) mIF. Experimental workflow is outlined in Figure 1. Clinical patient characteristics are shown in Figure 2(a). The only statistically significant difference observed was a higher proportion of pN + patients in pRCC vs. ccRCC (p < 0.001). On WES and whole transcriptomic analysis, no patients harbored an FH mutation although one patient was classified as having CpG methylator phenotype by TCGA categories (Figure 2(b)). Additional information in the supplementary results.

Figure 2.

(a) Patient clinical characteristics of patients undergoing whole slide (WS) mIF. (b) Spectrum of pRCC immune, molecular, and clinical phenotypes for 16 patients undergoing WS mIF. Patients arranged according to overall tumor-core immune cell density from left to right. Select values for cell density and Ripley's K are shown. Cell density and Ripley's K are relative to same marker values across ROIs and scaled from 0 to 1. Cell density and Ripley's K markers were manually selected for tumor and interface-tumor compartments only based on findings from global differences in mixed-effect modeling. TCGA classifications derived from bulk RNA expression data from same tumor specimens. pTS = pathologic T stage, pNS = pathologic N stage, SSIGN = stage, size, grade and necrosis score, TCGA = The Cancer Genome Atlas program.

Lower T-cell density and higher M2-like macrophage clustering in pRCC on mif

Overall immune cell density was compared first between pRCC and ccRCC. 1677 WS-mIF ROIs were available for comparison of cell density using linear mixed-effect modeling controlling for ROI compartment and patient level variables. Model details can be found in Supplementary Table 2. There were no single or double positive markers that were globally more abundant in pRCC compared to ccRCC, however, CD8+, Tbet+, CD8 + Tbet+, and CD8 + FOXP3 + cells were less abundant in pRCC (FDR < 0.1) (Figure 3(d) and Supplementary Fig. 1). Wilcoxon rank sum tests were performed on cell density across distinct ROI compartment subgroups. Significant differences in cell density were seen strongly within tumor and interface-tumor ROIs, and to a lesser extent interface-stromal ROIs (Figure 3(a), 3(b), Supplementary Table 3). Given that cell clustering can only be calculated in ROIs that have 3 or more cells, sparsity of immune cell types was also compared between pRCC and ccRCC. FOXP3 + and CD20 + cells were highly sparse in both pRCC and ccRCC, with no more than 2 cells per ROI in >70% of ROIs (Figure 3(b)).

Figure 3.

Comparison of cell density and spatial clustering between ccRCC and pRCC on WS mIF. (a) Single-marker comparison of median cell density across distinct tissue compartments. Single asterisk = FDR < 0.1, double = FDR < 0.001, triple = FDR < 0.0001 on compartment subgroup Wilcoxon rank sum testing. Scale ranges from 0 to the highest cell density for any marker across ROIs (ccRCC CD163 + density in interface-tumor ROIs = 0.13). Red = ccRCC, blue = pRCC. (b) Single-marker comparison of cell sparsity (defined as < 3 cells in an ROI window). Scale from 0 to 100% of ROIs where the marker is sparse. (c) Single-marker comparison of median u Ripley's K. Scale is from the lowest K for any marker-ROI to the highest K for any marker-ROI. (d) Global comparison of cell-density and Ripley's K for single and double-markers in pRCC and ccRCC using mixed-effect modeling. Red indicated statistically significant and higher in pRCC, blue indicates statistically significant and lower in pRCC.

Non-sparse ROIs were available for comparison in spatial clustering using linear mixed-effect modeling controlling for ROI compartment and patient level variables. The percentage of sparse ROIs per marker can be seen in Supplementary Tables 2 and 4. There were no single or double positive markers that were globally less clustered in pRCC compared to ccRCC, however, CD68+, CD163+, CD206+, CD68 + CD163+, CD68 + CD206+, CD68 + PDL1+, and CD163 + PDL1 + cells were more clustered in pRCC (FDR < 0.1) and the percentage of sparseness ranged from 8.6% (for CD68 + cells) to 45.9% (for CD68 + CD206 + cells) (Figure 3(d), Supplementary Table 2). Wilcoxon rank sum tests were performed on cell density across distinct ROI compartment subgroups. Significant differences in cell density were seen most strongly within tumor and interface-tumor ROIs, but also to a lesser-extent stroma (Figure 3(c), Supplementary Table 4).

Higher M2-like macrophage clustering in pRCC associated with lower recurrence free survival

Given that increased spatial clustering of M2-like macrophages was seen in pRCC compared to ccRCC, and that polarization of these macrophages across the tumor-stromal interface resulted in worse recurrence free survival in ccRCC, we performed a post-hoc survival analysis to estimate the clinical impact of this biological finding in pRCC. ³ Median follow up after surgery was 20 months in the pRCC subgroup. Tumor and stromal compartments were analyzed separately. Seven clustering markers that were significantly increased in pRCC compared to ccRCC were selected for RFS analysis (Supplementary Table 5). CD68 + CD163 + clustering in tumor compartments was significantly associated with RFS (p = 0.015). Dichotomized into top and bottom 50% of average CD68 + CD163+ Ripley's K in tumor compartments, high clustering was associated with lower RFS after surgery for localized pRCC (p = 0.043, Supplementary Fig. 2).

Anti-inflammatory genes are spatially enriched in pRCC M2-like macrophages

To explore cell-level correlative findings, we utilized single-cell ST on a subset of the larger cohort. ST was performed on three orthogonal TMA sections with four pRCC cores and 24 ccRCC cores (Figure 4a, 4b). 90,671 cells were categorized into 30 phenotypes from the Kidney Cell Atlas using supervised clustering (Figure 4(c)). Neighborhood density (or univariate local Ripley's K) was calculated for the 5527 cells classified as M2-like macrophages (Figure 4(d)). Neighborhood density of M2-like macrophages was higher in both tumor and stroma cores of pRCC compared to ccRCC by Wilcoxon rank sum test (p = 0.0079 and p = 0.0214, respectively). Normalized and transformed gene expression was then tested against an interaction term between histology and neighborhood density using mixed-effect modeling. Ten genes demonstrated positive and significant differential variation with M2-like macrophage neighborhood density in pRCC amongst tumor cores, and thirteen genes demonstrated positive and significant differential variation amongst stroma cores (FDR < 0.1, Supplementary Table 8 and 9). Two genes had positive and significant effects in both tumor and stroma cores, namely CCL18 and GPNMB (FDR < 0.001) (Figure 4d, 4e, 4f).

Figure 4.

(a) H&E from patient 2 tumor core (high M2-like spatial clustering) demonstrating lipid-laden foamy macrophages. (b) mIF from same TMA core, orthogonal section. (c) ST phenotypes select immune and stromal cell types from same core. M2-like macrophages highlighted in yellow. Other cell phenotypes grouped into classes of phenotypes for visual representation. M2-like macrophage Neighborhood Density Analysis. (d) Scaled neighborhood densities of M2-like macrophages. (e) Transformed-normalized expression of GPNMB (scaled from 0 to 1) in M2-like macrophages in same TMA core as A. Cells with median GPNMB expression of all cores (pRCC and ccRCC) are white, maximum expression (approaching 10) in green. (f) Transformed-normalized expression of CCL18.

Discussion

Immunogenomic analysis from the TCGA suggests two predominant immune profiles in pRCC: a depleted lymphocyte cluster with a more prominent macrophage signature, and a less common inflammatory cluster. ¹⁷ This small proportion of inflammatory pRCC has been corroborated with IHC studies, indicating that only a fifth of advanced pRCC has a relatively high cell density of tumor infiltrating cytotoxic T cells and NK cells. ¹³ Rare inflammatory phenotypes may represent distinct molecular subclassifications, such as fumarate hydratase (FH) deficient RCC, which is known to be immunogenic and in turn, could explain its favorable response to ICB.^18–20 However, apart from FH-deficient RCC, molecular classifications do not guide systemic treatment of pRCC under current kidney cancer guidelines. ²¹ Although the landscape of genomic classifications in pRCC is vast, its predominant immune phenotype is that of low immunogenicity, corresponding with its poor overall response to ICB.^10,22,23 In this study, we compare the spatial characteristics of a molecularly diverse panel of pRCC patients to those of ccRCC patients using whole slide mIF. Using mixed-effect modeling, we found that the cell density of effector T cell markers (CD8 and T-bet) were indeed globally lower in pRCC compared to ccRCC, but also a higher degree of spatial clustering of M2-like macrophages when present.

Single-cell spatial clustering via Ripley's K is a distinct metric that does not necessarily correlate with cell count or density. ²⁴ This clustering in specific tissue compartments of the TIME can have significant prognostic value in ccRCC, which lends credence to the hypothesis that spatial clustering by itself can have a profound biological significance.^3,24 In this study, we found that M2-macrophage clustering in the tumor compartment of localized pRCC was associated with worse RFS after surgery. Given that single-cell spatial clustering is a metric that is directly measurable in single-cell ST technologies, our secondary objective was to investigate gene expression differences in M2-like macrophages within pRCC vs ccRCC in ST. The largest gene expression difference amongst M2-like macrophages in pRCC v ccRCC was seen in CCL18, which was also associated with increased spatial clustering of M2-like macrophages within pRCC samples. CCL18 and GPNMB are known markers of lipid associated TAMs (LAMs) which have risen to prominence in the era of single cell technologies. ²⁵ Current single-cell classifications of macrophages are beginning to incorporate spatial information from the microenvironment architecture, but the spatial profile of these macrophages has not yet been explored in RCC.^9,25,26 Ripley's K and other spatial metrics should be integrated into cell classification schemas to identify rare phenotypes with biologically important roles in the TIME.

LAMs have been shown to have immunosuppressive effects in other cancers. The exact mechanisms are not completely known, but several possible pathways are being investigated. CCL18 positivity itself in TAMs has been shown to be associated with T-cell exhaustion in both breast cancer and glioblastoma.^27,28 GPNMB-mediated T-cell inhibition has been shown in patient derived co-cultures with autologous T cells from multiple cancer types. ²⁹ LAMs are also known to express TREM2 at high levels, a phospholipid receptor expressed on foamy macrophages in a variety of ailments, which is associated with immunosuppressive effects in multiple cancers. ³⁰ Additionally, scRNA-seq signatures of TREM2 high macrophages are associated with a poor response to ICB in a publicly available melanoma dataset. ³¹ CCL18+, GPNMB+, and TREM2 + macrophage modulation has been shown to enhance treatment with ICB in vitro and in vivo. ³² Clinical trials testing therapies that target GPNMB and TREM2 gene products are currently underway. In addition to breast, melanoma, and others, the anti-TREM2 antibody PY314 is also being tested in combination with pembrolizumab in advanced solid tumors, including pRCC. ³³ Thus, such combination therapy trials seeking to enhance ICB efficacy are of great potential importance in the predominantly low-immunogenic immune phenotype in pRCC. Although the current study demonstrates that CCL18 and GPNMB are expressed at high levels in spatially clustered macrophages in pRCC, additional functional studies are required to elucidate their potential role in response to therapy.

This study has several limitations. First, there are a limited number of samples for survival analysis and ST, thus the specific implications of these findings is purely hypothesis-generating. Second, pRCC is a molecularly heterogenous group as defined by the most recent WHO update thus, the primary findings of this study may not apply to all subtypes that have papillary architecture on histology. However, histology remains an extremely important clinical data point in treatment decision making and moreover, using bulk RNA sequencing and WES, we show that our mIF cohort is diverse and representative. Our study also has technical limitations, such as the lack of an established pipeline at present for identification of papillary tumor cells on ST, limiting our analysis to immune and stromal cells only. Limited IF marker panels and RNA-panels were utilized in mIF and ST, respectively, thus not every protein or gene of interest was available for analysis. Finally, unknown confounding remains a limitation for a small cohort or patients. Future studies are needed to determine if important pathological variables, such as stage and nuclear grade correlate with spatial clustering of immune cells, as these could be driving some of the observations made in this study.

In conclusion, compared to ccRCC, pRCC has fewer T cells but greater M2-like macrophage spatial clustering.. Using ST, we found that multiple LAM-associated genes are spatially enriched in pRCC. Additional resources should be dedicated to investigating the impact of myeloid biomarkers and modulating therapeutics in pRCC.

Supplemental Material

sj-docx-1-kca-10.1177_24684562251358595 - Supplemental material for Spatial clustering of immunosuppressive macrophages in papillary renal cell carcinoma

Supplemental material, sj-docx-1-kca-10.1177_24684562251358595 for Spatial clustering of immunosuppressive macrophages in papillary renal cell carcinoma by Mitchell T Hayes, Jose Laborde, Alex Soupir, Anders Berglund and Jamie K Teer, Kirill Sabitov, Nicholas H Chakiryan, Taylor C Peak, Jonathan Nguyen, Carlos Moran-Segura, Daryoush Saeed-Vafa, Neale Lopez-Blanco, Martina Molgora, Paola M Ramos-Echevarria, Christopher Guske, Jodi Balasi, Jasreman Dhillon, Youngchul Kim, James Mulé, Brooke L Fridley, Brandon J Manley in Kidney Cancer

Supplemental Material

sj-pdf-2-kca-10.1177_24684562251358595 - Supplemental material for Spatial clustering of immunosuppressive macrophages in papillary renal cell carcinoma

Supplemental material, sj-pdf-2-kca-10.1177_24684562251358595 for Spatial clustering of immunosuppressive macrophages in papillary renal cell carcinoma by Mitchell T Hayes, Jose Laborde, Alex Soupir, Anders Berglund and Jamie K Teer, Kirill Sabitov, Nicholas H Chakiryan, Taylor C Peak, Jonathan Nguyen, Carlos Moran-Segura, Daryoush Saeed-Vafa, Neale Lopez-Blanco, Martina Molgora, Paola M Ramos-Echevarria, Christopher Guske, Jodi Balasi, Jasreman Dhillon, Youngchul Kim, James Mulé, Brooke L Fridley, Brandon J Manley in Kidney Cancer