Human kidney tissue samples were collected from patients (n=23) with suspected clear cell renalcell carcinoma (ccRCC) at the Department of Urology, University Hospital Regensburg. The diagnosiswas later confirmed at the Department of Pathology, University Hospital Regensburg. Normal tissue was obtained from unaffected regions of the same kidney. Tumor tissue was identified and marked for analysis by an experienced pathologist, subsequently the % of malignant tissue in the tumor sample was determined (Supplementary Table S1). For the analyses of freshly resected samples (n=11), three anatomically definedregions—normal kidney, tumor periphery, and tumor center—were selected based onpathological assessment. Ethical approval was obtained from the local ethics committee (University of Regensburg vote number: 16-355-101) and conducted in accordance with the Declaration of Helsinki. All patients provided informed consent. Clinical details are summarized in Supplementary Table S1.

Frozen tissue sections (5 μm thick) were fixed in 4% paraformaldehyde for 20 minutes. Antigen retrieval was performed using 10 mmol/L sodium citrate buffer (pH 6.0, Zytomed) heated to 92–98 °C for 20 minutes. After PBS washing, endogenous peroxidase activity was blocked using Dako REAL peroxidase blocking solution for 5 minutes at room temperature.

Slides were incubated overnight at 4 °C with the following primary antibodies: Mouse anti-CD163 (1:100, Invitrogen), Mouse anti-CD36 (1:200, StemCell), Mouse anti-CD68 (1:200, Dako), Rabbit anti-CD147 (1:250, GeneTex). After washing, secondary antibody incubation was performed using a universal HRP-conjugated polymer (Histofine Simple Stain Max PO, anti-rabbit and anti-mouse) for 30 minutes in a dark room. Signal detection was carried out with DAB substrate (Vector, SK-4100) for 2 minutes, followed by hematoxylin counterstaining.

Representative images were acquired using an ECHO Rebel light microscope. For each tissue sample, three independent 4 µm sections were analyzed. In each section, six random fields were examined at 40x magnification using Fiji/ImageJ software. Staining intensity was quantified as optical density per unit area, calculated from the mean gray value of each region of interest after grayscale conversion and background subtraction. In addition, a trained observer, blinded to sample identity and experimental grouping, assigned semi-quantitative scores ranging from 1 (lowest) to 9 (highest) based on both staining intensity and the proportion of positive cells. These two methods were performed concurrently and used to validate each other. Correlation analyses were primarily based on observer-derived scores using Pearson´s correlation coefficients with two-tailed p-values, and plots were generated using GraphPad Prism.

Tissue homogenates representing a wet weight of 2 mg were extracted according to the method of Bligh and Dyer (23) in the presence of not naturally occurring lipid species as internal standards.

Lipid species were quantified by direct flow injection analysis (FIA) using a triple quadrupole mass spectrometer (FIA-MS/MS) and a high-resolution hybrid quadrupole-Orbitrap mass spectrometer (FIA-FTMS). FIA-MS/MS was performed in positive ion mode using the analytical setup and strategy described previously (24, 25). A fragment ion of m/z 184 was used for lysophosphatidylcholines (LPC) (26). The following neutral losses were applied: Phosphatidylethanolamine (PE) and lysophosphatidylethanolamine (LPE) 141, phosphatidylserine (PS) 185, phosphatidylglycerol (PG) 189 and phosphatidylinositol (PI) 277. Sphingosine based ceramides (Cer) and hexosylceramides (Hex-Cer) were analyzed using a fragment ion of m/z 264 (27). Glycerophospholipid species annotation was based on the assumption of even numbered carbon chains only.

A detailed description of the FIA-FTMS method was published recently (28, 29). Triglycerides (TG), diglycerides (DG) and cholesterol esters (CE) were recorded in positive ion mode as [M+NH₄]⁺ at a target resolution of 140,000 (at 200 m/z). CE species were corrected for their species-specific response (30). Phosphatidylcholines (PC) and sphingomyelins (SM) were analyzed in negative ion mode as [M+HCOO]^- at the same resolution setting. Multiplexed acquisition (MSX) was applied for the determination of free cholesterol (FC) and the respective internal standard (FC[D7]) (30).

Cryopreserved tumor and matched normal kidney tissue blocks were sectioned using a cryostat and stored at –80°C. Before staining, slides were air-dried at room temperature and washed with PBS. Lipid staining was performed using Oil Red O (ScienCell, #0843-SC) according to the manufacturer’s instructions. Nuclear counterstaining was performed with Hemalaun (Merck, #1.09249.0500), and slides were mounted with Faramount Aqueous Mounting Medium (Dako, #S3025). Representative images were acquired using an ECHO Rebel microscope.

For quantification, slides were thawed for 2.5 hours at room temperature, fixed for 15 minutes, and rinsed with distilled water. Oil Red O working solution (600 µl stock + 400 µl distilled water, filtered through 0.2 µm) was applied for 15 minutes. Slides were washed thoroughly with tap water, counterstained with hematoxylin for 3 minutes, and mounted. Lipid droplets appeared red, and nuclei blue.

Cryosections from three RCC patients (also included in the IHC cohort) were mounted on coated slides and fixed with 4% paraformaldehyde. Multiplex immunofluorescence staining was performed using the MACSima™ Imaging Platform (Miltenyi Biotec), an automated system for cyclic staining and imaging. Reagents and antibodies used included FcR Blocking Reagent (Cat# 130-059-901), CD36 PE (Cat# 130-110-877), CD147 APC (Cat# 130-124-295), CD8a PE (Cat# 130-117-201), CD45 PE (Cat# 130-113-118), CD68 PE (Cat# 130-128-345), CD163 PE (Cat# 130-127-908), and pan-Cytokeratin APC (Cat# 130-123-091), all from Miltenyi Biotec. DAPI (D9542) was purchased from Sigma Aldrich. All steps, including staining and imaging, were performed according to the manufacturer’s protocol.

Tissues were processed as described previously (19). Briefly, samples were minced and enzymatically digested using Collagenase IV and DNase I to generate single-cell suspensions. After filtration through a 70 µm mesh, cells were washed, counted, and stained with fluorochrome-conjugated antibodies for surface marker analysis by flow cytometry. The following antibodies were used: CD3 V450 (BD, Cat# 560365), CD45 PerCP (BioLegend, Cat# 304026), CD8 FITC (BioLegend, Cat# 344704), CD68 PE-Cy7 (BioLegend, Cat# 333816), CD163 BUV496 (BD, Cat# 752878), CD36 PE (Miltenyi Biotec, Cat# 130-110-877), CD147 APC (BioLegend, Cat# 306214). Flow cytometric acquisition was performed on a BD LSRFortessa™ instrument, and data were analyzed using FlowJo v10 (BD Biosciences).

Bulk RNA sequencing data of 533 ccRCC patients were retrieved from The Cancer Genome Atlas (TCGA) (KIRC dataset) (31, 32) using the UCSC Xena platform (33). Gene expression values are reported as log2(fpkm-uq + 1).

First cohort of single-cell RNA sequencing data from 12 RCC patients was obtained from previouslypublished work by Li R. et al. (34), available as anAnnData (.h5ad) object, Dataset (35). The dataset was analyzed using Scanpy v1.9.1. Quality control filtering excluded cells with fewer than 200 detected genes, fewer than 2000 total counts, or more than 30% mitochondrial transcript content. To correct for batch effects and enable downstream analysis, cell transcriptomes were embedded into a low-dimensional latent space using scVI (36, 37), treating each patient sample as a distinct batch. The model was trained on the 2000 most highly variable genes, identified using Scanpy’s pp.highly_variable_genes function with the parameters flavor=“seurat” and batch_key=“orig.ident”. A neighborhood graph and UMAP embedding were generated from the resulting scVI latent space. Cell type annotations were adopted from the original publication (34). For gene set enrichment analysis, over-representation analysis (ORA) of hallmark gene sets was performed using the decoupler-py package v.2.1.1 (38). To quantify gene co-expression, double-positive cells were defined as those expressing ≥1 UMI for both genes of interest, and fraction of positive cells were calculated based on the number of cells with ≥1 UMI per gene. This dataset includes non-malignant tissues from ccRCC patients (Supplementary Figure S2F), the focus on specific sub-populations is described in the corresponding figure legend.

Second cohort of single-cell transcriptomic data from human ccRCC samples were analyzed using a publicly available dataset published by Bi et al. (18). Data visualization and exploration were conducted via the Single Cell Portal (39). Cell types, including immune subpopulations, were annotated as defined by the original authors. Marker expression was examined across these distinct immune compartments to investigate cell type–specific expression profiles relevant to the tumor microenvironment. The cohort consisted of eight ccRCC patients with five patients having received immune checkpoint blockade (ICB), including four who had also been treated with tyrosine kinase inhibitors (TKIs). The remaining three patients were treatment-naïve (18).

Data are expressed as mean and individual values for microscopy data and as violin plots with median and quartiles for the transcriptomic data. Statistical comparisons were performed using one-way ANOVA with post hoc tests, and Student’s t-test, as appropriate, using GraphPad Prism. Correlation analyses between markers were assessed by Pearson´s correlation with two-tailed p-values. Significance levels were set as * p<0.05, ** p<0.01, *** p<0.001,**** p<0.0001.

We studied samples from ccRCC patients using conventional immunohistochemistry, multipleximmunofluorescence imaging (MIFI), flow cytometry (FC) and publicly available ccRCC transcriptome data (Supplementary Figure S1A). Staining with Oil Red O showed an increased accumulation of lipids in the ccRCC tumor tissue, as compared to healthy kidney. The results of the automated approach generating the area staining intensity were confirmed through observer-based histological score (Figure 1A). Next, the expression of CD36, a key lipid-transporting molecule, was assessed. CD36 was increased in ccRCC tumor tissue (Figure 1B) and microscopy-based data were validated with expression analyses using RNA sequencing data from the TCGA database (Figure 1C). The CD36 expression did not correlate with Oil Red O signals (Supplementary Figure S1B). MIFI analyses confirmed CD36 expression in ccRCC tissues (Figure 1D; Supplementary Figure S1C). Next, we studied freshly resected samples from ccRCC tumor periphery, tumor center andadjacent kidney using FC (Supplementary Figure S1D). As opposed to fixed tissues assessed by immunohistochemistry, no robust FC marker for ccRCC cancer cells has been widely established so far, therefore we analyzed the CD36 expression on CD45-negative (CD45neg) cells as a proxy for cancer cells (40, 41). Compared to CD45neg cells from adjacent kidney, tumor samples showed increased expression of CD36 (Figure 1E). Lastly, ccRCC tumors accumulated triglycerides (TG) and, by a trend, cholesterol esters (CE). The percentage of free cholesterol (FC) from all lipids in the tumor was decreased by a trend (Figure 1F). CD36 expression correlated to intratumoral triacylglycerol (TG) levels by a trend (Figure 1G; Supplementary Figure S1E).

Accumulation of lipids and increased expression of CD36 might associate with intratumoral immune cells such as macrophages, which show high lipid metabolic activity and which also react to lipid metabolic changes in tissues. The expression of the macrophage marker CD68 was increased in ccRCC tumor tissues (Figure 2A; Supplementary Figure S2A). Increased CD68 levels in tumors were validated on the transcriptional level, but CD68 expression was not higher in larger tumors (Figure 2B). CD163 which associates with pro-tumorigenic macrophage populations (22) was increased in the ccRCC tumors (Figures 2C, D; Supplementary Figure S2B). Interestingly, CD163 expression levels increased with tumor size, suggesting its role in tumor promotion (Figure 2D).

MIFI analyses revealed a co-expression of CD68 and CD163 in ccRCC tumor tissues (Figure 2E; Supplementary Figures S2D, E). Next, single cell RNA-sequencing (scRNAseq) data by Li et al. (34) were re-analyzed and confirmed the co-expression of CD68 and CD163 in tumor cells in the myeloid cell cluster (Figure 2F). This data set includes non-malignant tissues from ccRCC patients (Supplementary Figure S2F), however, cells from adjacent kidney were almost not present in the myeloid cell cluster(Supplementary Figure S2G). On average, 45.3% of CD68+ cells also co-expressed CD163 (Figure 2G). The expression of TREM2, APOE, C1QA and to a lesser extent MRC1 suggested apro-tumorigenic macrophage phenotype (42, 43) (Supplementary Figure S2G).

Next, we studied ccRCC samples using FC. While the infiltration with CD45+ cells (which include CD68+ macrophages) and the proportion of living CD45+ cells was lower in adjacent kidney (Figure 2H), the CD163 expression on CD68+ cells was not different and showed a strong variation among the individual patients (Figures 2I, J). These data suggest, that CD163+ macrophages are present both in kidney tissues and ccRCC tumors. However, increased numbers of CD45+ cells result in higher absolute numbers of CD163+ macrophages in ccRCC tumors.

Lastly, CCR2 which can mark “M2-like” macrophages positively correlated with CD163and CD14 on ccRCC TAM in FC data (Supplementary Figure S2H) and in transcriptomics (Supplementary Figure S2I). CCR2 ligands CCL7, CCL8 and CCL13 are negatively prognostic and CCL13 correlates with CD163 in the KIRC TCGA cohort (Supplementary Figure S2J).

To study potential links between ccRCC lipid metabolism and its immune infiltrate, we correlated the obtained microscopy data sets. While the CD36 levels did not correlate with CD68 expression, the correlation of CD163 and CD36 was highly significant, suggesting a favoring of pro-tumorigenic macrophage subpopulations in ccRCC tumors with increased CD36 levels (Figure 3A). While the infiltration of CD3 T cells was increased in ccRCC tumors, as compared to normal kidney tissues (Supplementary Figure S2C), the correlation of CD3 and CD36 was not significant (Figure 3A). Oil Red O staining was used in correlations to immune cell markers next. Interestingly, we observed a negative correlation of Oil Red O and CD68. In contrast, no correlation was observed while comparing Oil Red O to CD163 and CD3 expression (Figure 3B).

CD36 was expressed on ccRCC tumor cells (Figure 1). However, CD36 can also be expressed by macrophages (7, 8). Accordingly, MIFI confirmed CD36 expression on CD68+ cells in ccRCC tissues (Figure 3C). Using FC, we assessed the CD36 expression on macrophages, but did not detect a difference in tumor tissues as compared to adjacent kidney (Figure 3D). Furthermore, macrophage CD36 expression did not correlate to CD163 (Figure 3E) and scRNAseq data showed that only a subpopulation of CD163+ macrophages co-expressed CD36 (Figure 3F; Supplementary Figures S3A, B). However, we detected a positive correlation of CD36 on CD45neg cells with CD163 on macrophages in ccRCC tumors. Importantly, this was not the case with the overall frequencies of CD68+ cells (Figures 3G, H). Similar to CD163, CD36 on CD45neg cells correlated positively with the frequency of CCR2+macrophages and by a trend with their CD14 expression (Supplementary Figures S3C, D). In addition, we observed a trend towards a lower infiltration with CD3+CD8+ T cells in tumors with high expression of CD36 in CD45neg cells (Figure 3I).

Lastly, lipidomic analyses revealed a positive correlation between the intratumoral triacylglycerol (TG) levels and the CD163 expression (Figures 3J, K).

Next, the expression of CD147, a transmembrane protein regulating several metabolic transporters was assessed. While its levels were increased in tumor tissues as assessed by the area staining intensity, the observer-based data and transcriptome data did not show a difference (Figures 4A, B). Next, we correlated the expression of CD147 to the lipid transporter CD36 and the macrophage marker CD163. While we observed a positive correlation of CD147 and CD163 in the area staining intensity data, this could not be validated in the observer-based histological score. Conversely, CD147 correlated positively with CD36 in the histological score analysis, but not in the area staining intensity data (Figures 4C, D). Furthermore, CD147 did not correlate with the Oil Red O intensity (Supplementary Figure S4A).

Using FC, we studied the CD147 expression on the single-cell level. While the variation of CD147 expression on CD45neg cells was high, there was a trend toward its increased expression in tumor tissues (Figure 4E). Confirming the data from the histological score analysis, we observed a positive correlation of CD147 and CD36 on CD45neg cells from both peripheral and central tumor tissues (Figure 4F). As opposed to CD36, the expression of CD147 on CD45neg cells did not correlate with macrophage CD163 expression (Supplementary Figure S4B). MIFI confirmed the expression of CD147 and CD36 in ccRCC tissues (Figure 4G; Supplementary Figure S4C). scRNAseq data showed a high basal expression of CD147 across all cell types, with a strong enrichment in RCC tumor cells (Figure 4H).

Lastly, we studied the expression of CD147 on ccRCC macrophages. FC analyses did not detect a different CD147 expression on tumor macrophages, as compared to those from adjacent kidney samples (Figure 4I). However, CD147 expression positively correlated with the expression of CD163 in FC (Figure 4J) and scRNAseq analyses (Figure 4K).

To further analyze the macrophage sub-populations in ccRCC, we studied a second cohort of single cell RNA sequenced ccRCC tumors by Bi et al. (18). Here, single cell RNA sequencing has been performed and individual cells were assigned to several subpopulations. In addition to the macrophage markers CD68 and CD163, we assessed the expression of CD147, CD36, ACAA2 (a prominent gene involved in fatty acid oxidation), SQLE (a major representative of cholesterol biosynthesis), ACSL3 (key regulator of long-chain fatty acid activation). All macrophage sub-populations showed prominent CD163 expression, suggesting their pro-tumorigenic phenotype. Furthermore, we observed a strong expression of CD147 in tumor cells and in macrophages (Figure 5). ACSL3 was mainly expressed in macrophages and SQLE in tumor cells, suggesting a compartment-specific lipid metabolism. ACAA2 was expressed by both tumor cells and macrophages, suggesting a shared fatty acid oxidation pathway.