Intracellular Factor H Drives Tumor Progression Independently of the Complement Cascade

The complement cascade is part of the innate immune system (1, 2). Many complement therapeutics are entering clinical trials (3). Therefore, it is critical to know whether and how complement proteins affect diseases. Tumors are a complement-rich environment (4) in which numerous cell types, including immune cells (5), endothelial cells (6, 7), fibroblasts (8), and the malignant cells themselves (4, 7) contribute to local production of complement components.

Factor H (FH) is the master regulator of C3 convertase, the central enzyme of the complement alternative pathway (AP). Its canonical function is to block C3 convertase formation, enhance C3 convertase dissociation, and exert cofactor activity for factor I (FI), an enzyme cleaving C3b to inactive fragment iC3b (1). FH in a tumor can come from the circulation or in situ production by tumor cells. Although studies in patients are scarce, animal models and in vitro studies reveal that within the tumor microenvironment (TME), FH has context-dependent actions. Overexpression of FH has protumor effects by hampering C3b opsonization and lytic membrane attack complex formation on tumor cells (9–11). Consistent with these observations, injection of a blocking FH antibody reduces lung cancer growth in mice (12). In contrast, in a mouse model of sarcoma, FH has an antitumoral effect by downregulating the generation of anaphylatoxins, which promote an immunosuppressive environment (13). Moreover, spontaneous hepatic tumor formation occurs in aged mice with FH deficiency (14).

Not all complement proteins produced by cancer cells are secreted, rather some partially remain within the cell (15). Intracellular staining of FH (int-FH) is seen in cancer cells in an array of cancer types (10, 16–20), but the function and relevance of this int-FH for tumor progression are poorly understood. In T cells and in pancreatic islets, C3 and/or C5 have intracellular functions unrelated to the complement cascade, these functions are termed noncanonical functions (21–23). FH also can have roles independent of the complement cascade (19, 20, 24, 25), but it is unclear whether these are related to extracellular or intracellular FH. Considering the diverse, context-dependent and often contradictory modes of action of FH described in vitro and in mice, it is critical to assess the role of FH in human cancer if we are to understand its biological functions and shed light on its usefulness as a prognostic biomarker or therapeutic target.

Here, we found that FH was overexpressed by tumor cells in a subset of patients with clear cell renal cell carcinoma (ccRCC) and lung adenocarcinoma (ADC) and was associated with poor prognosis. The protumoral effect of FH resulted from a tumor cell–specific intracellular mode of action that promoted tumor cell proliferation, survival, and migration, independent of complement activation.

The patients included in this study signed an informed consent. The research was approved by the medical ethics board (no. CEPAR 2014-001, CPP Ile-de-France II no. 2016-07-08, no. 2008-133, and no. 201206-12) and conducted according to the recommendations in the Helsinki Declaration. Primary tumor specimens were collected from three retrospective cohorts. The first cohort (cohort 1) was a retrospective cohort of 133 patients with ccRCC who underwent surgery at Institut Mutualiste Montsouris (IMM, Paris, France). The second cohort (cohort 2) included 91 patients with ccRCC whose tumors were collected at the time of surgery at Necker-Enfants Malades Hospital (Paris, France). Cohort 3 consisted in 212 patients with non–small cell lung (NSCLC), 112 with ADC and 100 with squamous cell carcinoma (SCC), collected at the Departments of Thoracic Surgery and Pathology of Cochin Hospital (Paris, France). Plasma samples were collected from a prospective cohort of 61 patients with ccRCC (cohort 4), established at IMM (Paris, France). Histologic subtype of ccRCC or NSCLC, all tumor–node–metastasis (TNM) stages and tissue with good quality for the analysis constituted the inclusion criteria. Histopathologic features such as Fuhrman grade, TNM stages, and size were available for most of the patients (Supplementary Table S1). The correlation between FH staining and clinical, histopathologic features and immune infiltrate in ccRCC and NSCLC are shown in Supplementary Tables S2 and S3, respectively.

C57BL/6J mice were purchased from Charles River Laboratories. HepatoFH^−/− mice were generated (26) and provided by Mathew C. Pickering (Imperial College London, London, United Kingdom). The animal experiments were conducted in accordance with the recommendations for the care and use of laboratory animals and with approval l'APAFIS no. 3764-201601121739330v3 by the French Ministry of Agriculture. 400,000 mouse TC-1 cells were subcutaneously inoculated in the right flank of the mice in 200 μL PBS. Tumor size was measured with calipers every 2 to 3 days for 25 days or until reaching 3,000 mm³. Two independent experiments were performed with 5 wild-type (WT) and 5 hepatoFH^−/− mice ages 8 to 10 weeks in experiment 1 and 10 WT and 5 hepatoFH^−/− mice in experiment 2.

Formalin-fixed paraffin-embedded human ccRCC or NSCLC specimens were stained for FH, cytokeratin (CK), αSMA, CD31, CD163, CD20, CD3, N-cadherin, E-cadherin, vimentin, calnexin, ACBD5, Golga5, Tom20, PLIN2, Lamp1, C3d, C3c, C3a, cathepsin L, IgG, and NaK. The primary and secondary antibodies used, as well as the antigen retrieval and staining conditions, are summarized in Supplementary Table S4. Antigen retrieval was performed with PT-link (Dako PT200) using EnVision FLEX Retrieval Solutions (Dako, pH Low catalog no. K8005, pH High catalog no. K8004) with high or low pH depending on the antibody. Endogenous peroxidases were then blocked using 3% H₂O₂ (Gilbert, catalog no. 6553036) solution, followed by blocking of nonspecific staining with Protein Block (Dako, catalog no. X0909). The staining was revealed using 3-amino-9-ethylcarbazole substrate (Vector Laboratories, catalog no. SK4200). The slides were then counterstained with hematoxylin (Dako, catalog no. CS700), mounted with Glycergel (Dako, catalog no. C0563), and scanned with NanoZoomer (Hamamatsu). For double staining by immunofluorescence (IF), a tyramide system was used to avoid species compatibility problems and to increase the fluorescence signal. This system needed an additional step after incubation with the secondary horseradish peroxidase (HRP) antibody that consisted of an incubation with AF647 tyramide reagent (diluted 1:100 in TBS 1×, H₂O₂ 0.0015%, Life Technologies, catalog no. B40958). The entire staining protocol was then repeated for the second primary antibody but with AF546 tyramide reagent (Life Technologies, catalog no. B40954). The nuclei were stained with DAPI (Thermo Fisher Scientific, catalog no. 62248). The slides were then mounted with Prolong Glass antifade reagent (Invitrogen, catalog no. P36980 and scanned with AxioScan (Zeiss). Colocalization between FH and CK was analyzed using HALO Image Analysis software (Indica Labs).

The antibody Ox24 (which was produced for this study using a hybridoma from ATCC, a kind gift from Pierre Fabre Research Institute) was selected for FH staining. The specificity of Ox24 for FH was validated by staining of liver sections (Geneticist) and inhibition of the signal by preincubation with purified FH (Complement Technology, catalog no. A317; Supplementary Fig. S1A). Because Ox24 recognizes also factor H like 1 (FHL-1), a splice variant of FH containing the first seven domains, the presence of full-length FH in ccRCC tumor sections was validated by additional FH antibodies, the N terminal–specific antibody Quidel 2 (Quidel, catalog no. A254) and antibodies C18 (Abcam, catalog no. ab121055) and L20 (Thermo Fisher Scientific, catalog no. GAU-020-03-02), which recognize C-terminal parts of the molecule absent in FHL-1 (Supplementary Fig. S1B). For patient stratification, the tumors were classified into two groups, high or low, by a semiquantification method to distinguish the different types of FH staining: int-FH or extracellular [membranous FH (mb-FH)]. For FH int-FH or mb-FH staining, the high group corresponded to the number of FH-positive cells ≥30% and the low group corresponded to that ≤30%. Three independent observers (M.V. Daugan, M. Revel, R. Thouenon, or L.T. Roumenina) performed the analysis to avoid any biases (27). Automated quantification was not feasible because it was not possible to train the algorithm to reliably distinguish intracellular staining from membrane deposits.

A total of 6 μm sections of frozen mouse tumor tissues were fixed with cold acetone for 8 minutes and blocked with 5% BSA-TBS for 30 minutes. Tumor sections were then stained with a rabbit polyclonal anti-CD31 (Abcam, catalog no. ab124432) or a goat polyclonal antiserum specific for FH (Quidel, catalog no. A312) as primary antibodies and then goat anti-rabbit coupled to AF647 (Thermo Fisher Scientific, catalog no. A-21245) or donkey anti-goat coupled to AF647 (Thermo Fisher Scientific, catalog no. A-21447) as secondary antibodies. The nuclei were stained with DAPI. The slides were then mounted with Prolong Glass antifade reagent and scanned with AxioScan (Zeiss). Quantification of the FH or CD31 staining was analyzed using HALO Image Analysis software (Indica Labs).

Human ccRCC cell lines (Caki-1 and A498), human lung ADC (A549), and SCC (SK-MES) cell lines were purchased from ATCC, in 2016. Primary human umbilical vein endothelial cells (HUVEC) and primary renal proximal tubule cells (RPTEC) were purchased from Lonza in 2016 and 2019, respectively. Caki-1 cells were cultured in McCoy medium (Gibco, catalog no. 16600082) + 10% heat-inactivated FCS (Biowest, catalog no. S1810-500) + 1× penicillin/streptomycin (Gibco, catalog no. 10378016). A498 cells were cultured in Eagle Minimum Essential Medium (ATCC, catalog no. 30-2003) + 10% heat-inactivated FCS + 1× penicillin/streptomycin. A549 cells were cultured in DMEM nutrient mixture F12 (Gibco, catalog no. 11320033) + 10% heat-inactivated FCS + 1× penicillin/streptomycin + 1× l-glutamine (Gibco, catalog no. 25030081) + 1× HEPES (Gibco, catalog no. 15630080) + 1× nonessential amino acids solution (Gibco, catalog no. 11140050). Mouse lung cancer TC-1 cells were a kind gift from Eric Tartour, INSERM IMRS970, Paris, France and were cultured in RPMI supplemented with 5 mmol/L L-glutamine (Gibco, catalog no. 61870036), 10% heat-inactivated FCS, 1× penicillin/streptomycin (Gibco), and 50 μmol/L 2-mercaptoethanol (Gibco, catalog no. 31350). No specific authentication of the cell lines was performed except comparison of the morphology with the provided images. HUVEC were stained for CD31 and RPTEC for prominin-1 for authentication and found positive. All cell lines were routinely tested for Mycoplasma (Ozyme, catalog no. LT07-710) and used only if negative. Primary HUVEC and RPTEC were used for three passages, while the tumor cell lines were used for up to 20 passages.

The cells were transfected with a preprepared mixture of two siRNAs against CFH/CFHL-1 at a concentration of 50 nmol/L (Qiagen Hs_CFH_3_Flexitube siRNA, catalog no. SI00003983 and Hs_CFH_4_Flexitube siRNA catalog no. SI00003990) or siRNA control 50 nmol/L (Qiagen AllStars Negative Control siRNA, catalog no. SI03650318, an siRNA with no homology to any known gene sequence) in 5 μL of Lipofectamine RNAiMAX Transfection Reagent (Thermo Fisher Scientific, catalog no. 13778030) in Opti-MEM medium (Gibco, catalog no. 31985070). After 24 hours, the transfection was stopped by changing the Opti-MEM medium to the appropriate medium without antibiotics. The cells were ready to use for the functional experiments 72 hours posttransfection. Photos were taken under a microscope 72 hours posttransfection to determine the differences in cell number and morphology. The efficiency of the silencing was determined by qRT-PCR and ELISA for cell lysates and supernatants (see below). WT cells correspond to nontransfected cells; siC cells correspond to cells transfected with the control siRNA; siFH cells correspond to cells transfected with CFH siRNA.

Cells were lysed in RIPA buffer to obtain a total cell lysate. A specific cell fractionation kit (Abcam, catalog no. ab109719) was used to prepare the organelles and cytoplasmic and nuclear fractions from cultured cells according to the manufacturer's instructions.

FH and C3 concentrations in cell supernatants and cell lysates were measured using a homemade sandwich ELISA (28). In brief, native antibody [either polyclonal anti-FH (Quidel, catalog no. A312) or polyclonal C3 (Calbiochem, catalog no. 204869)] was coated on a Nunc MaxiSorp ELISA 96-well and 1% PBS-BSA was used for the blocking. Supernatant or cell lysate was added to the plate and incubated for 1 hour at room temperature. After washing, the plate was incubated with an in-house biotinylated version of the anti-FH or anti-C3 (biotinylation with EZ link NHS Biotin was performed according to the manufacturer's instructions (Thermo Fisher Scientific, catalog no. 20217) for 1 hour at room temperature. After additional washes, streptavidin coupled with HRP (Dako, catalog no. P039701-2) was added for 1 hour at room temperature. Binding was revealed using SureBlue TMB Microwell Peroxidase Substrate (Sera Care, catalog no. 5120-0075), and the reaction was stopped by 2 mol/L sulfuric acid. Multiskan Ex (Thermo Fisher Scientific) was used to read the optical density at 450 nm. The results are expressed in μg/mL according to the standard curve, made using commercial purified FH or C3 (Complement Technology, catalog no. A113).

The supernatants or cell lysates of cells cultured without serum for 48 hours were recovered and concentrated using Amicon Ultracel 3K units (UFC, catalog no. 900324). A 10% Bis-Tris gel (Thermo Fisher Scientific) was used to separate the proteins before transfer onto a nitrocellulose membrane. After 1 hour of blocking using 5% BSA-TBS, the membrane was incubated overnight at 4°C on a rocking platform with the primary antibodies (Supplementary Table S5) diluted in 5% BSA-TBS. Purified C3 (Calbiochem, catalog no. 204885), purified C3b (Calbiochem, catalog no. 204860), and purified iC3b (Calbiochem, catalog no. 204863) were used as controls. An HRP-conjugated anti-IgG (Supplementary Table S5) was added at 1:1,000 dilution for 1 hour at room temperature under agitation. Visualization was performed by chemiluminescence using a substrate for HRP (SuperSignal WestDuraLuminol, Thermo Fisher Scientific, catalog no. 1856145) and detected by iBright Western Blot Imaging System (iBright FL1500, Thermo Fisher Scientific).

RNA was extracted from the cultured cell lines or from sections of mouse tumors using a Maxwell cell 16 LEV simplyRNA Purification Kit according to the manufacturer's protocol (Promega, catalog no. AS1270). The quality and quantity of RNA were determined using the Agilent RNA 6000 Nano Assay kit (catalog no. 5067-1511) and a 2100 Bioanalyzer (Agilent). An Applied Biosystem High-capacity cDNA Reverse Transcription Kit (Applied Biosystem, catalog no. 4368814) was used for reverse transcription. Gene expression of CFH was assessed by Taqman using primers for human FH (Thermo Fisher Scientific, catalog no. Hs00962373_m1) and compared with GAPDH (Thermo Fisher Scientific, catalog no. Hs00266705_g1) and ACTB (Thermo Fisher Scientific, catalog no. Hs01060665_g1) genes for normalization. Murine gene expression was assessed by Taqman using primers for murine VEGFA (Thermo Fisher Scientific, catalog no. Mm00437306_m1), VEGFB (Thermo Fisher Scientific, catalog no. Mm00442102_m1), VEGFC (Thermo Fisher Scientific, catalog no. Mm00437310_m1), PECAM-1 (Thermo Fisher Scientific, catalog no. Mm00478712_m1), FLT4 (Thermo Fisher Scientific, catalog no. Mm01292604_m1), KDR (Thermo Fisher Scientific, catalog no. Mm01222421_m1), PROKR1 (Thermo Fisher Scientific, catalog no. Mm00517546_m1), VWF (Thermo Fisher Scientific, catalog no. Mm00550376_m1) and compared with GAPDH (Thermo Fisher Scientific, catalog no. Mm01180221_g1) and ACTB (Thermo Fisher Scientific, catalog no. Mm02619580_g1) genes for normalization.

The RNA of three independent biological replicates for each condition (siFH vs. siC) was used for the RNA sequencing (RNA-seq) study. The libraries were prepared with NEBNext Ultra II Directional RNA Library Prep Kit (New England BioLab, catalog no. E7760) according to the manufacturer's protocol: polyA purification using poly-T oligo attached magnetic beads, followed by 300-bp fragmentation with divalent cations under elevated temperature, synthesis of double-stranded cDNA, and amplification by PCR. Sequencing was carried out on a paired-end 75 bp Illumina HiSeq 4000 instrument. The quality of FASTQ files was verified using FASTQC and then aligned to a reference genome using STAR tool. Once the data were mapped on the genome, HTSeqCount was used to count and assign reads to a given exon and generate a count table. Before differential analysis, nonexpressed genes with a count equal to 0 were removed. The differential expression analysis between siFH and siC was performed using R software and the DESeq2 package (29). Results were considered statistically significant at an adjusted P value of < 0.05. The top 50 most significantly upregulated or downregulated genes were sorted by the adjusted P value; volcano plots were produced, as was a heatmap using heatmap.2 and EnhancedVolcano packages in R. Gene Ontology (GO) enrichment analysis was performed with WEV-based Gene SeT AnaLysis Toolkit (WebGestalt) with the GO biological process category. The raw data are available in Gene Expression Omnibus database under the accession number PRJNA720902.

Flow cytometry data were acquired using a BD LSRFortessa X-20 (BD Biosciences) and analyzed using FlowJow software.

Normal human serum (Etablissement Français du Sang) diluted to 33% in PBS was added to A498, Caki-1, A549, and SK-MES cells for 30 minutes at 37°C with or without a blocking anti-FH (Ox24). Staining was performed with mouse monoclonal anti-C3c (Quidel, catalog no. A205) or isotype control IgG1 (Dako, catalog no. X0931) followed by a secondary goat anti-mouse AF700 (Invitrogen, catalog no. A-21036).

CFSE reagent 1:1,000 (Invitrogen, catalog no. C34554) was added to A498, Caki-1, A549, and SK-MES cells for 20 minutes at 37°C. Standard culture medium of each cells was added to stop CFSE staining. Cells were seeded in a 6-well plate and cultured for 72 hours in the presence or absence of complete medium supplemented with purified FH at 2 or 20 μg/mL. The proliferation capacity of cells was evaluated after the addition of either 50% complete medium or 50% A498 supernatant after 72 hours of culture. Supernatants containing dead cells and adherent cells were recovered and stained with DAPI before analysis by flow cytometry.

After fixation in cold 70% ethanol, 3 μmol/L propidium iodide (PI, Invitrogen, catalog no. P1304MP) was added to the cells in the presence of RNase (Thermo Fisher Scientific, catalog no. EN0601) for 15 minutes at room temperature, in a staining buffer according to the manufacturer's instruction.

For int-FH staining, after blocking with 5% BSA, cells were stained with DAPI (1:1,000) for 10 minutes at room temperature. Cells were permeabilized with a BD Cytofix/Cytoperm kit (BD, catalog no. 554714) or kept nonpermeabilized. An FH-specific antibody 1:100 (Quidel, catalog no. A224) or mouse isotype control IgG1 was added to cells for 30 minutes at room temperature. A goat anti-mouse AF488 (Thermo Fisher Scientific, catalog no. A28175) diluted 1:100 was added after washing.

Cells were seeded on a round cover glass and cultured for one day. The adherent cells were fixed with 4% PFA for 30 minutes at room temperature. Adherent cells were then stained either for actin or NFκB. A498, Caki-1, HUVEC, RPTEC, and SK-MES cells were cultured in their standard culture medium at a confluence of 50% to 75% and then stained for their actin cytoskeleton with Phalloidin-IFluor 488 reagent (Abcam, catalog no. ab176753) and with DAPI. A498 were cultured in their standard culture medium at a confluence of 50% before staining for NFκB, nonspecific staining was blocked by 30 minutes incubation of the cells at room temperature with Protein Block. Staining was performed with a rabbit monoclonal anti-NFκB p65 (Cell Signaling Technology, catalog no. 8242) and revealed with a goat anti-rabbit coupled with AF647 (Invitrogen, catalog no. A21245). Nuclei were stained with DAPI. All stained slides were mounted with Prolong Glass antifade reagent (Thermo Fisher Scientific, catalog no. P36980) and scanned with AxioScan (Zeiss). The intensity of NFκB in the nucleus was analyzed on whole slides by using HALO Image Analysis software (Indica Labs).

To evaluate migration capacity, a wound-healing assay was performed on cell monolayers of A498, Caki-1, HUVEC, and RPTEC cells silenced or not for FH. The monolayers were obtained when cells reached a confluence of 90%, allowing a good monolayer for the 12 next hours. A scratch was generated with a tip, and photographs were taken under a microscope immediately or after 12 hours. The wound-healing tool (http://dev.mri.cnrs.fr/projects/imagej-macros/wiki/Wound_Healing_Tool) in ImageJ was used to calculate the percentage of recovery.

The package “survival” in R software was used to generate Kaplan–Meier curves, with censoring at 2,500 days. A log-rank test was applied to examine the difference between curves. Association between the distributions of two semiquantitative variables was assessed by the χ² test. The difference between siFH and siC cells with regard to proliferation and survival was assessed by the paired t test; the unpaired t test is used for cell migration. *, P < 0.05; ** P < 0.01; ***, P < 0.001; ****, P < 0.0001.

In the ccRCC cohort in The Cancer Genome Atlas database (KIRC cohort), gene expression of CFH above the median was significantly associated with decreased disease-free survival (DFS; P = 0.03) and overall survival (OS; P = 0.017), suggesting a role of locally produced FH in this type of cancer.

Staining of human ccRCC tumor sections for FH revealed heterogeneous staining patterns by IHC (Fig. 1). There were deposits, defined as mb-FH, surrounding the tumor cells and intracellular staining of the tumor cells (int-FH). The mb-FH colocalized with IgG and the NaK ATPase pump by IF (Supplementary Fig. S1C and S1D), suggesting an extracellular location of the mb-FH, corresponding to deposits, which is associated with complement activation (7). Because of the limitation of the IHC/IF technology, we are not able to firmly conclude that mb-FH is an extracellular deposition of FH at the tumor membrane following complement system activation. Although this seems the most probable explanation, we cannot exclude an intracellular deposit of FH at the internal leaflet of the plasma membrane. FH binds to apoptotic cells, which have a membrane flip-flop, exposing the internal leaflet outside (30, 31). Therefore, we cannot rule out that some FH present in the cytoplasm could bind to the plasma membrane from inside.

Analysis of the intracellular staining revealed a diffuse coloration of the entire cytoplasm and/or intense granular staining, which was confirmed in consecutive sections using four different FH-specific mAbs recognizing different epitopes of the molecule (Supplementary Fig. S1B). Within the tumor, over 85% of the FH staining localized in tumor cells (CK) after automatic quantification (Supplementary Fig. S2A), confirming the visual observation that tumor cells were the main cells staining positive for FH. Stromal cells, especially fibroblasts, endothelial cells, and macrophages were also positive for FH, although to a much lesser extent (Supplementary Fig. S2B–S2D). There was no colocalization of FH with CD20⁺ B cells or CD3⁺ T cells (Supplementary Fig. S2E and S2F). The plasma concentration of FH was not different between patients with ccRCC and healthy donors (Supplementary Fig. S2G), suggesting that FH production by the tumor itself contributes little, if at all, to the plasmatic pool.

To study the specific impact of mb-FH and int-FH, we analyzed tumor tissues obtained from two cohorts of patients with ccRCC by IHC (Supplementary Table S1). Tumor sections showed different combinations of mb-FH and int-FH staining (Supplementary Fig. S3A–S3D). FH staining at the membrane showed interpatient and intratumoral heterogeneity. These staining patterns allowed us to classify the patients into two groups according to the number of tumor cells carrying FH deposits (Fig. 1A). In the two ccRCC cohorts, the presence of high or low numbers of tumor cells with FH deposits did significantly not impact survival [cohort 1, DFS, P = 0.14; cohort 2, progression-free survival (PFS), P = 0.226 and OS P = 0.627; Fig. 1B–D]. The same staining heterogeneity was observed for int-FH, and two groups of patients were classified depending on the number of FH-producing cells (Fig. 1E). Compared with a low density of int-FH⁺ cells, a high density of int-FH⁺ cells had a negative impact on DFS in cohort 1 (P = 0.00432; Fig. 1F) and cohort 2 (PFS: P = 0.0274, OS: P = 0.0727; Fig. 1G and H). The presence of FH inside tumor cells correlated with tumor size, Fuhrman grade and the presence of other complement components, especially intracellular C4 and C3 (Supplementary Table S2). No association was found between int-FH and the immune infiltrate.

Because FH was expressed mainly by tumor cells, we confirmed that it was expressed in the ccRCC tumor cell lines A498 (Fig. 2A–C) and Caki-1 (Supplementary Fig S4A–S4C). Although these cells lines do not fully recapitulate human RCC (32), they allowed us to explore the role of FH in vitro. In both cell lines, FH was detected intracellularly by flow cytometry (Fig. 2A; Supplementary Fig. S4A), and Western blotting of the lysate (Fig. 2B; Supplementary Fig. S4B), and was shown to be secreted into the supernatant by ELISA (Fig. 2C; Supplementary Fig. S4D); it was not detected at the cell surface (Fig. 2A; Supplementary Fig. S4A).

We studied the impact of FH on key characteristics of tumor cells using gene silencing. After silencing in A498 and Caki-1 ccRCC cell lines, secretion of FH into the supernatant was reduced by 90% to 96% and FH in cell lysates was reduced by 85% to 90% in the cell lysates (Fig. 2D–F; Supplementary Fig. S4B and S4D). The siRNA-induced silencing of FH revealed a high impact on the phenotype of the ccRCC cells, which acquired a round shape (Fig. 2G; Supplementary Fig. S4E).

To identify the biological processes impacted by FH silencing, RNA-seq was performed on three biological replicates for silenced and control A498 and Caki-1 cells. We used P < 0.05 and log₂ fold change >1 to select the differentially expressed genes. With these criteria, in A498 cells, 235 genes were differentially regulated after FH silencing (Fig. 2H). Among these genes, 139 were upregulated and 96 were downregulated. For Caki-1 cells, 192 genes were upregulated and 251 were downregulated after FH silencing (Supplementary Fig. S4F). We then identified the 50 most differentially expressed genes for each condition (Fig. 2I; Supplementary Fig. S4G). The functional classification analyses of all significantly upregulated and downregulated genes using the GO “Biological Pathway” categories revealed that these genes were implicated in different biological processes (Fig. 2J; Supplementary Fig. S4H).

RNA-seq analysis revealed an enrichment of genes implicated in cell proliferation in FH-silenced ccRCC cells and a parallel downregulation of genes implicated in the G₁–S transition of the mitotic cycle (Fig. 2H–J). We confirmed that the cell proliferation capacity of FH-silenced A498 cells was significantly reduced in culture (Fig. 3A). Addition of exogenous FH in the culture medium did not restore the proliferative capacity of FH-silenced A498 cells or modify the ability of control A498 cells to proliferate (Fig. 3B). Moreover, the addition of untreated A498 cell supernatant to siFH cells had no effect on their proliferation capacity (Fig. 3C). These results show that exogenous FH, either purified from plasma or produced by the tumor cells themselves (which may differ from the plasma-derived FH), had no effect on tumor cell proliferation. We therefore conclude that int-FH but not mb-FH is essential for the tumor cell phenotype we observed.

Consistent with the RNA-seq data, we also found that FH-silenced A498 cells were predominantly in the G₀–G₁ phase (Fig. 3D). Cell viability was also affected, as indicated by an increase in the number of dead cells when FH was silenced (Fig. 3E). This could be explained by increased p53 expression (Fig. 3F). Moreover, FH-silenced A498 cells had a higher percentage of nuclei that stained for NFκB (Fig. 3G).

Modifications of genes involved in cell adhesion, extracellular matrix organization, and cell morphogenesis were also detected in the RNA-seq analysis (Fig. 2H–J; Supplementary Fig. S4F–H). These transcriptional changes had functional repercussions. FH-silenced A498 cells were rounder with fewer actin filaments and stress fibers (Fig. 3H). There were also some differentially expressed genes that were involved in cell locomotion and cellular movement. In vitro, we confirmed that the migration capacity of FH-silenced A498 cells was decreased, with only 5% scratch recovery at 12 hours compared with 40% scratch area recovery for control A498 cells (Fig. 3I and J). Similar data on proliferation, cell cycle, viability, morphology, and subcellular localization of FH were obtained with the Caki-1 cell line (Supplementary Fig. S5).

In situ, the more motile cells in a tumor are usually the ones undergoing epithelial-to-mesenchymal transition (EMT). On ccRCC sections, FH colocalized with tumor cells that had low levels of the epithelial marker CK (Supplementary Fig. S2A). Increased intensity of CK staining was associated with decreased positivity for FH: cells with low intensity for CK represented 65% of FH⁺ cells, whereas tumor cells with medium and high intensity of CK represented only 15% and 5% of FH⁺ cells, respectively. In addition, in ccRCC, FH-expressing cells expressed the mesenchymal markers N-cadherin and vimentin (Fig. 3K and L), suggesting that int-FH is associated with a mesenchymal and hence more motile tumor cell phenotype.

Previous studies indicate that tumor cells produce FH, which can be used as a potential mechanism to escape complement attack (33). We found that mb-FH colocalized and correlated with the presence of C3d at the membrane of tumor cells, suggesting that FH is located at the outer leaflet and functions as a complement regulator (Fig. 4A and B). Compared with siRNA control cells, when FH-silenced A498 or Caki-1 cells were exposed to serum, an increased amount of C3 fragment deposits was observed (Fig. 4C; Supplementary Fig. S6A). Moreover, in the supernatant of the FH-silenced cells, the α43 band corresponding to the inactive fragment of C3b and C3(H2O) [iC3b, iC3(H2O)], was largely undetectable due to loss of FI cofactor activity (Fig. 4D; Supplementary Fig. S6B). Indeed, FI was secreted in the supernatants of ccRCC tumor cells (Supplementary Fig. S6C). The presence of C3b was detected in only Caki-1 cell supernatant (presence of α' band), but both Caki-1 and A498 cells secreted C3/C3(H2O) (presence of α band). These results indicate that FH silencing results in deregulation of the complement system and an increased level of C3 activation. To decipher whether this overactivation of complement can impact the tumor cell phenotype, we treated A498 cells with the FH blocking antibody Ox24 at 20 μg/mL, which is in excess compared with the amount of FH secreted. The level of C3 deposits increased when cells were treated with this blocking antibody, reflecting its capacity to prevent FH binding to C3b and its regulatory function (Fig. 4E). However, the tumor cell phenotype was not modified, contrary to the condition when cells were silenced for FH (Fig. 4F), suggesting that extracellular complement regulatory functions had no impact in ccRCC tumor cell phenotype in the tested experimental conditions.

To confirm this in vivo, we inoculated TC-1 cells into WT and hepatoFH^−/− mice (26). HepatoFH^−/− mice produce only extrahepatic FH and have dramatically reduced FH levels in the circulation (Supplementary Fig. S7A). TC-1 expressed very low levels of FH (Supplementary Fig. S7B). After hepatoFH^−/− mice were inoculated with TC-1 cells, the level of FH inside the tumor was largely decreased (Supplementary Fig. S7C and S7D), meaning that plasma FH is an important source of FH for the tumor. However, tumor growth in hepatoFH^−/− mice was not modified compared to that in WT mice (Supplementary Fig. S7E). Although FH has been suggested as a neoangiogenesis inhibitor (34), we saw no modification in vascular density or architecture (Supplementary Fig. S7F) or angiogenic gene expression in tumors from hepatoFH^−/− mice compared with those from WT mice (Supplementary Fig. S7G). Thus, plasma FH did not impact tumor growth in this model, which is known to be complement-sensitive (7, 35).

A498 and Caki-1 express C3aR and C5aR1. Therefore, it was possible that the effect of tumor cell–produced FH was related to the control of the autocrine C3a and C5a, generated by the cells themselves. C5a in both untreated and concentrated supernatant was below the detection limit of the ELISA kit (36). C3a was detected at the limit of the ELISA only in the concentrated supernatant of Caki-1, but was undetectable in A498 (36). These results are in line with Western blotting, where the α' band (corresponding to C3b and indicative of C3a generation) was detected only in the Caki-1 supernatant (Fig. 4D for A498 and Supplementary Fig. S6B for Caki-1). Therefore, even if FH affects local concentrations of C3a and C5a in other contexts, we think this is not a likely mechanism in ccRCC.

FH silencing resulted in downregulation of genes related to cargo loading on vesicles (Fig. 2J). In line with this, in situ int-FH was located mainly in intracellular vesicles (Fig. 5A; Supplementary Fig. S1C). Fractionation of the different compartments from tumor cells showed that most int-FH was detected in a fraction containing organelles (mitochondria/lysosomes/endoplasmic reticulum/vesicles); very little FH was found in the cytosolic or nuclear compartment (Fig. 5B and C). Using IF, we identified lysosomes as the organelles containing int-FH (Fig. 5D–I). C3 is the main partner of FH in the extracellular compartment and has a lysosomal localization in T cells (21). By cell fractionation, we showed that C3 and FH were located in the same organelles (Fig. 5J). Furthermore, IF staining with antibodies recognizing epitopes in C3d, C3c, or C3a confirmed that FH and C3 were localized in the lysosomes, suggesting an intra-lysosomal interaction (Fig. 5K–M). Both FH and C3 can be considered full-length proteins as the staining was detected with different antibodies that detect different domains of FH (Fig. 1A) and C3 (Fig. 5K–M). Although C3 can be cleaved by cathepsin L (21) with the help of FH (37), we found that FH and cathepsin L did not colocalize in the ccRCC tumor sections with the limit of our detection method (Fig. 5N).

Because ccRCC is derived from proximal tubules, we studied FH in this cell type. In peritumoral tissue, we detected areas where some of the proximal tubular cells stained positive for int-FH (Fig. 6A). Although normal mouse tubules stain positive for FH, in normal human renal tissue the tubules are FH⁻ and C3⁻ by the same staining technique we used here (38, 39). When FH was detected in proximal tubules in ccRCC sections, it colocalized with lysosomes (Fig. 6B). In primary human RPTECs, we detected FH in the lysate and the supernatant (Fig. 6C). FH was predominant in the same organelle fraction within these cells as within the tumor cells (Fig. 6D). Despite efficient FH silencing (Fig. 6E and F), no modification of the transcriptional program of the cells was observed; only 11 genes were differentially expressed, including CFH, between FH-silenced and control RPTECs (Fig. 6G), compared with 235 genes for A498 cells. Moreover, only 2 differentially expressed genes were common in A498 cells and RPTECs: CFH and NCKAP5. No effects of FH silencing on the RPTEC phenotype were detected in terms of viability, number of cells, or morphology (Fig. 6H). The actin cytoskeleton of the cells remained the same regardless of whether FH was silenced (Fig. 6I), and the migration capacity did not change (Fig. 6J). Even if other parameters may be modified that we did not explore, the effect of FH seems to be specific to tumor cells.

Within the tumor, endothelial cells can produce FH (Supplementary Fig. S8A). In vitro, endothelial cells (HUVECs) also produce and secrete FH (Supplementary Fig. S8B). After efficient silencing (Supplementary Fig. S8C–S8E), no alterations in the phenotype were found in terms of cell number, morphology, actin cytoskeleton, and migration capacity (Supplementary Fig. S8F and S8G).

The presence of int-FH⁺ cells in tumors was not specific for ccRCC because it was also found in NSCLC tumors (Fig. 7A). However, mb-FH was detected only rarely and in small areas of NSCLC tumors (Fig. 7B). Like ccRCC, the main cell type staining positive for FH was tumor cells (Supplementary Fig. S9A), especially mesenchymal tumor cells that expressed high levels of N-cadherin and low levels of E-cadherin (Fig. 7C and D). Stromal cells such as αSMA⁺ fibroblasts and CD31⁺ endothelial cells also contributed to local production of FH (Supplementary Fig. S9B and S9C). CD163⁺ macrophages, CD20⁺ B cells, and CD3⁺ T cells (Supplementary Fig. S7D–S7F) did not stain positive for FH.

To determine whether the protumoral effect of int-FH could be relevant to NSCLC, two cells lines were selected for in vitro experiments: A549, a lung ADC cell line, and SK-MES, a lung SCC cell line. A549 cells expressed and secreted equivalent amounts of FH as ccRCC cell lines (Fig. 7E), whereas SK-MES expressed less (in agreement with ref. 10). Efficient silencing of FH (Fig. 7F and G) strongly modified the A549 phenotype (Fig. 7H) with differences in morphology and proliferation capacity (Fig. 7I), but without an effect on cell cycle (Fig. 7J). In contrast, in SK-MES cells, no effects on the phenotype were detected despite efficient FH silencing (Supplementary Fig. S8).

Efficient silencing of FH hampered the regulatory activity of FI, preventing C3(H2O) inactivation in the supernatant (Fig. 7K). No α' band indicative of C3b was detected on the Western blot analysis and no C3a or C5a were detected in the supernatant. Upon exposure to normal human serum, FH-silenced A549 cells deposited more complement activation fragments, compared with siRNA control cells (Fig. 7L). Therefore, FH produced by the A549 cells was functionally active and controlled complement. Because deposits of mb-FH were found only in few patients, survival analysis was not possible.

In NSCLC, we also detected int-FH. The pattern of int-FH staining was the same as that for ccRCC (Fig. 7A). Patients with NSCLC with either lung ADC or SCC from cohort 3 (Supplementary Table S1), were classified into two groups according to the number of int-FH⁺ cells (Fig. 7M). In lung ADC, int-FH⁺ cells were associated with a negative impact on OS (P = 0.0156; Fig. 7N). In contrast, in lung SCC, int-FH had no impact on OS (P = 0.798; Fig. 7O). The negative impact on lung ADC did not correlate with clinical or histologic features and did not correlate with the immune infiltrate (Supplementary Table S3).

Complement components form a plasma innate immune cascade but also have functions beyond this. Here, we show that FH is locally expressed by multiple types of human tumors. We provide a paradigm shift for the impact of FH on cancer progression, showing what we believe to be a previously unrecognized, likely tumor cell–specific, intracellular function of FH outside the complement cascade (Fig. 7P). Int-FH served as a driver of the proliferation and migration of ccRCC and lung ADC cells but not of normal cells or lung SCC cells. The presence of int-FH staining in tumor cells indicated poor prognosis for ccRCC and lung ADC.

Prior studies indicate ccRCC and lung ADC are cancers in which complement overactivation occurs and coregulated overexpression of the genes encoding the complement components as well as detection of C4d in plasma and in situ confer poor prognosis (4, 7, 13, 36, 40–44). Therefore, we were surprised to observe that membranous localization of C3d and FH staining were not associated with prognosis and that int-FH staining and high CFH gene expression were associated with decreased survival. Our results echo previous data, where a tendency toward poor prognosis was found in patients with FH⁺ tumors in a small cohort of lung ADC (45). As FH is a key negative complement regulator, one might expect it to control complement cascade activation and reduce C5a, thereby preventing the protumor action of the complement system. In ccRCC, we found that the complement cascade did not proceed to a cytotoxic C5b-9 formation. Therefore, the negative prognostic impact of FH cannot be explained by prevention of complement-mediated tumor cell killing.

Contrary to our expectations, we discovered that the association of FH with poor prognosis was not only a potential biomarker, but also reflected what we believe to be a novel function for FH outside of the complement cascade. Silencing of FH in ccRCC and lung ADC cell lines induced a strong modulation of the tumor cell phenotype, including proliferation, cell cycle, morphology, viability, and migration. These results can be explained by the profound modification of the transcriptional program of FH-silenced tumor cell lines, which was marked by alteration in the expression of genes governing the biological processes of adhesion and locomotion. Moreover, within the tumors, int-FH⁺ cells were frequently also positive for mesenchymal markers, further suggesting a motile phenotype.

We found that FH was localized mainly within lysosomes. This localization is consistent with the observation that C3 in T cells is in lysosomes, allowing C3a–C3aR interactions and promoting cell survival (21). FH is stable at low pH (pH 5; ref. 46), suggesting that it can function within this organelle. These findings, together with the alteration of genes involved in cargo loading into vesicles after CFH silencing, suggests that int-FH participates in lysosomal trafficking and function. This subcellular location makes it less likely that it serves directly as a transcription factor or regulator of gene expression. In apoptotic cells, FH plays a role as a cofactor for cathepsin L–mediated cleavage of C3 (37). In ccRCC, int-FH and cathepsin L did not seem to colocalize, suggesting another mode of action. C3 seemed to be in its full length and in its C3(H2O)-like form, due to the low pH in this organelle. The low pH within lysosomes does not seem to be an obstacle for the interaction of C3(H2O) or C3b with int-FH, as they have an optimum at pH 5 (47, 48). The mechanism by which the tumor cell phenotype is altered by intralysosomal int-FH must be further elucidated. This mode of action is likely different from the one of the intracellular C1s, which also has noncanonical, cascade-unrelated functions in ccRCC and lung ADC but is localized mainly in the cytoplasm, colocalizing with C4 and cytoplasmic C3 (36).

The reduced proliferation and viability of FH-silenced cells could be explained by cell-cycle arrest at G₀–G₁. At the molecular level, these biological functions could be a consequence of upregulation and phosphorylation of p53, which is a powerful tumor suppressor. p53 arrests cells in G₀–G₁, reducing survival and controlling proliferation (49). FH silencing also modified the location of NFκB within the tumor cells; it became localized predominantly in the nucleus. NFκB has multiple roles in physiologic and pathologic processes, promoting or inhibiting apoptosis depending on the context and stimulating gene transcription. Therefore, int-FH may contribute to a regulation loop in which p53 and NFκB participate, especially in cancer (50).

In previous studies, the negative impact of FH on NSCLC was attributed to inhibition of complement activation and protection from complement-mediated tumor cell killing (10, 12, 33, 51–53). The findings in mouse models of lung cancer, though, are strongly context dependent and show that complement activation is protumoral (54), antitumoral (12), or irrelevant (6). Therefore, translating the results from mouse models to patients is challenging. A limitation of our study is that we could not show in a mouse model that injection of FH-silenced human tumor cell lines reduced tumor growth because the proliferation rate of these cells was so low that we could not generate sufficient numbers to inject the 5 million cells per immunodeficient mouse needed to observe tumor growth in our experimental groups. Nevertheless, our results allow us to conclude that tumor cell–produced FH regulates the complement system, but this process does not seem to have an impact on ccRCC and lung ADC patient survival. Moreover, complement activation in human patients with NSCLC is associated with poor prognosis (41, 42), which is consistent with our findings for ccRCC (7). In addition, there is no evidence for C5b-9 formation and tumor cell killing in ccRCC and lung ADC in patients. Moreover, in patients with ccRCC, the presence of positive staining of membranous C3b/iC3b/C3d is not associated with survival (7) and in lung ADC, FH and C3d deposits are a very rare event. Finally, C5a in the supernatant of the tested cell lines was below the detection limit of the ELISA, suggesting that it is not responsible for the observed effect. Therefore, the canonical action of mb-FH as a complement regulator seems to be less relevant for human lung ADC and ccRCC than the noncanonical action of int-FH. The int-FH action could explain in part why FH-silenced A549 cells showed delayed tumor growth in mice (33).

Outside cancer, FH can also act as an immune regulator by modulating the recruitment and functionality of various immune cell types, such as B cells, monocytes, neutrophils, or dendritic cells (55–60) and in breast cancer it promotes recruitment of macrophages and acquisition of an immunosuppressive phenotype (20). However, in our ccRCC and lung cancer cohorts, no associations were observed between int-FH or mb-FH and immune infiltrate. Therefore, the function of mb-FH in the TME and as a complement regulator is not a major driver of tumor progression in ccRCC and lung cancer.

The CFH gene encodes two alternative splice forms: FH and FHL-1. In our study, FH silencing also affected the expression of FHL-1, as previously observed in lung cancer (33) and cSCC (19). We cannot completely rule out the hypothesis that FHL-1 plays a role in the protumoral effect that we observed. Nevertheless, we found much higher expression of CFH within the tumor tissues than CFHL1, and absence of the effect of CFHL-1 on ccRCC patient survival and the use of an shRNA against the C-terminal part of FH, which does not affect FHL-1, induced the same phenotypic modifications. These data support the conclusion that FH is the main factor responsible for the biological functions altered after CFH/CFHL-1 silencing and that FH, not FHL-1, drives tumor progression in ccRCC.

The absence of the effects of FH silencing in RPTECs (from which ccRCC originates) and endothelial HUVECs suggests an oncogenic role. These results do not exclude the possibly that int-FH in other types of normal cells could have a biological function. Indeed, loss of FH impairs antioxidant capacity and energy metabolism of human retinal pigment epithelium cells (34). In light of our results, the reported impact of the FH Y402H polymorphism, critical for the development of age-related macular degeneration (AMD), on retinal pigment epithelial cell behavior could be explained by the action of int-FH, together with its functions in the microenvironment (61). This difference between tumor and healthy cells could be explained by differences in the interactome of int-FH or by posttranslational modifications in tumor versus normal cell–derived FH. The int-FH binding partners within the tumor cells remain to be identified. In ovarian cancer, the sialylation level of FH in sera is lower in advanced stage patients than in healthy donors (62). In hepatocellular carcinoma, enhanced fucosylation of FH is an indicator of premalignant liver disease (63). In the plasma of patients with lung cancer (12, 51) and AMD (52), FH is detected as a reduced form that displays a decreased C3b binding capacity and a loss of FI cofactor activity (52). The posttranslational modifications of FH in ccRCC require further investigation.

It is important to emphasize that FH expression can be protumoral or antitumoral according to the cancer type. In hepatocellular carcinoma, high expression of CFH is associated with a lower risk of recurrence (64) and high CFH is associated with improved survival, whereas CFH mutations are associated with worse survival (14). This may be explained by the fact that FH is mainly produced by the liver and that low FH–expressing tumors will be the ones with the most dedifferentiated cells. Moreover, in addition to cancer organ of origin, our data indicate that the histologic subtype is also important to take into account. FH exerted a protumoral effect on lung ADC, but no effects of FH were detected in lung SCC neither patients or in cell lines upon CFH silencing. Taken together, these results suggest that therapeutic targeting of FH will be difficult with the conventional antibodies, because the FH blocker has to enter into the cell to be efficient. Moreover, the effect will be context dependent. Patients with int-FH⁺ lung ADC and ccRCC may benefit from int-FH targeting, if such a drug becomes available.

In conclusion, we show that int-FH exerts what we believe to be previously unrecognized functions, including modulation of the transcriptional activity of renal ccRCC and lung ADC cancer cells and effects on their proliferation, cell cycle, morphology, and motility. Through these noncanonical functions, rather than through control of complement activation, int-FH confers poor prognosis in these types of cancer.

M.V. Daugan reports a patent for method for prognosis and treatment of a patient suffering from cancer (EP21305187.3) pending. M. Revel reports a patent for method for prognosis and treatment of a patient suffering from cancer (EP21305187.3) pending. A. Mejean reports personal fees from Ipsen, Pfizer, Janssen, Ferring, and Bristol Myers Squibb outside the submitted work. M.C. Pickering reports receiving consultancy fees for providing scientific and clinical advice from Alexion, Apellis, Novartis, GlaxoSmithKline, and Gyroscope Pharma. M. Alifano reports personal fees from AstraZeneca outside the submitted work. W.H. Fridman reports a patent for method for prognosis and treatment of a patient suffering from cancer (EP21305187.3) pending. L.T. Roumenina reports grants from Cancer Research for Personalized Medicine (CARPEM), ARC contre le cancer, La Ligue contre le cancer, the Labex Immuno-Oncology Excellence Program, and INSERM during the conduct of the study, as well as a patent for method for prognosis and treatment of a patient suffering from cancer (EP21305187.3) pending. No disclosures were reported by the other authors.

M.V. Daugan: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. M. Revel: Formal analysis, validation, investigation, visualization, methodology, writing–review and editing. R. Thouenon: Investigation. M.-A. Dragon-Durey: Formal analysis, supervision, validation, writing–review and editing. T. Robe-Rybkine: Investigation. C. Torset: Investigation. N.S. Merle: Investigation, writing–review and editing. R. Noé: Investigation. V. Verkarre: Resources. S.M. Oudard: Resources. A. Mejean: Resources. P. Validire: Resources. X. Cathelineau: Resources. R. Sanchez-Salas: Resources. M.C. Pickering: Resources. I. Cremer: Resources, validation, writing–review and editing. A. Mansuet-Lupo: Resources. M. Alifano: Resources. C. Sautès-Fridman: Resources, validation, writing–review and editing. D. Damotte: Resources. W.H. Fridman: Resources, validation, writing–review and editing. L.T. Roumenina: Conceptualization, formal analysis, supervision, funding acquisition, validation, writing–original draft, project administration, writing–review and editing.

The cytometric and microscopy analyses were performed at the Centre d'Histologie, d'Imagerie et de Cytométrie (CHIC) and the Centre de Recherche des Cordeliers UMRS1138 (Paris, France). The authors are grateful to the CHIC team for their excellent technical assistance. CHIC is a member of the Université Pierre et Marie Curie Flow Cytometry network (RECYF). They thank the Centre d'Expérimentations Fonctionnelles team of the Centre de Recherche des Cordeliers for excellent technical assistance and for their support with animal experimentation. The authors acknowledge the help from the team of the GenomIC platform Cochin Institute INSERM U1016, headed by F. Letourner, for the RNA-seq experiments.

This work was supported by grants from the Cancer Research for Personalized Medicine (CARPEM) and ARC contre le cancer and La Ligue contre le cancer to L.T. Roumenina and équipe labellisée Ligue contre le cancer (W.H. Fridman and C. Sautès-Fridman). This work was also supported by INSERM, University of Paris, Sorbonne University, and the Labex Immuno-Oncology Excellence Program. M.V. Daugan received a Ph.D. fellowship from ARC.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.