The signal transducer CD24 suppresses the germ cell program and promotes an ectodermal rather than mesodermal cell fate in embryonal carcinomas

Testicular germ cell tumors (GCTs) are stratified into seminomas and nonseminomas. Seminomas share many histological and molecular features with primordial germ cells, whereas the nonseminoma stem cell population—embryonal carcinoma (EC)—is pluripotent and thus able to differentiate into cells of all three germ layers (teratomas). Furthermore, ECs are capable of differentiating into extra‐embryonic lineages (yolk sac tumors, choriocarcinomas). In this study, we deciphered the molecular and (epi)genetic mechanisms regulating expression of CD24, a highly glycosylated signaling molecule upregulated in many cancers. CD24 is overexpressed in ECs compared with other GCT entities and can be associated with an undifferentiated pluripotent cell fate. We demonstrate that CD24 can be transactivated by the pluripotency factor SOX2, which binds in proximity to the CD24 promoter. In GCTs, CD24 expression is controlled by epigenetic mechanisms, that is, histone acetylation, since CD24 can be induced by the application histone deacetylase inhibitors. Vice versa, CD24 expression is downregulated upon inhibition of histone methyltransferases, E3 ubiquitin ligases, or bromodomain (BRD) proteins. Additionally, three‐dimensional (3D) co‐cultivation of EC cells with microenvironmental cells, such as fibroblasts, and endothelial or immune cells, reduced CD24 expression, suggesting that crosstalk with the somatic microenvironment influences CD24 expression. In a CRISPR/Cas9 deficiency model, we demonstrate that CD24 fulfills a bivalent role in differentiation via regulation of homeobox, and phospho‐ and glycoproteins; that is, it is involved in suppressing the germ cell/spermatogenesis program and mesodermal/endodermal differentiation, while poising the cells for ectodermal differentiation. Finally, blocking CD24 by a monoclonal antibody enhanced sensitivity toward cisplatin in EC cells, including cisplatin‐resistant subclones, highlighting CD24 as a putative target in combination with cisplatin.


Introduction
Type II germ cell tumors (GCTs) are the most common neoplasia in young men of age 14-45 years. It is generally accepted that GCTs arise from a common precursor lesion, called germ cell neoplasia in situ (GCNIS), eventually developing into seminomas or non-seminomas [1]. The non-seminomatous stem celllike embryonal carcinomas (ECs) can further differentiate into teratomas (TE), yolk-sac tumors (YSTs), or choriocarcinomas (CC) [2,3]. Orchiectomy followed by chemo-or radiotherapy is a widely used procedure in the treatment of type II GCTs, leading to high cure rates of up to 90% [3]. Nevertheless, about 10-15% of patients with progressive disease relapse as a result of drug resistance and are condemned for a poor prognosis and a short survival of only a few months [4].
Cluster of differentiation 24 (CD24) is a small, mucinlike glycosylphosphatidylinositol (GPI)-anchored membrane molecule that is discussed to function in both signal transduction and adhesion. This glycoprotein is mainly expressed on the surface of hematopoietic, neural, muscular, and epithelial cells [5]. Moreover, CD24 has been implicated in tumor metastasis, as fucosylated CD24 interacts with P-and E-selectin, allowing invasion of tumor cells to distal sites [6]. High expression or amplifications of CD24 have been described in a variety of solid malignancies, such as non-small-cell lung carcinoma, gliomas, breast cancer, retinoblastoma, hepatocellular carcinoma, renal cell carcinoma, cervical carcinoma, prostate cancer, urothelial carcinoma, pineal parenchymal tumors, and ovarian cancer [7][8][9][10][11][12][13][14][15][16][17][18]. An exception is multiple myeloma, where CD24 was rather downregulated compared with normal B-cell cell lines [19,20]. This can be explained by the fact that CD24 is expressed in pre-B-lymphocytes, remains expressed in mature resting B cells, and becomes downregulated during the maturation process to plasma cells, which are terminally differentiated B cells [21].
Previously, we could identify high CD24 expression in the EC cell lines NCCIT, 2102EP, and GCT27, but only weak expression in JAR cells (CC) [22]. Similar observations were found in a cohort of 24 GCT patient samples, where CD24 was predominantly upregulated in non-seminoma compared with seminoma [23]. However, to our knowledge, no study has investigated the molecular role of CD24 in GCTs. Hence, in this study, we investigated the putative function of CD24 and its interaction partners in (cisplatinresistant) GCT cell lines and further deciphered the molecular consequences of a CRISPR/Cas9-mediated CD24 deficiency in EC cell lines.

Materials and methods
The study methodologies conformed to the standards set by the Declaration of Helsinki and were approved by the local ethics committees.

Cell culture
Germ cell tumors and non-cancerous cell lines were received from and were cultivated as described in Table S1 and as published [24]. Short tandem repeat (STR) profiles of all cell lines are checked on a regular basis and are available upon reasonable request. All cell lines are checked for mycoplasma contamination on a regular basis by a PCR strategy.

Three-dimensional cell co-cultivation and cell sorting
A detailed description has been published previously [25]. Briefly, a total amount of 3 9 10 3 cells per 40-lL drop have been plated onto an inverted lid of a 15-cm cell culture plate. Hanging drops were incubated at 37°C in a 5% CO 2 incubator for 72 h. For cell sorting, an initial ratio of GCT cell lines to microenvironmental cells of 30 : 70 was used. Cell sorting into GFP + (GCTs) and mCherry + /DeepRed + (microenvironmental) cells was performed on a MoFlo XDP (Beckman Coulter, Krefeld, Germany).

Drug applications
See Table S1 for an overview of drugs used in this study, their solvents, concentrations, and application schemes (Table S1). Individual IC 50 concentrations were determined by XTT assays in quadruplicates and calculated by 'GRAPHPAD PRISM 8.0' as described previously [26,27].

XTT viability assay
XTT viability assays were performed as described previously [32]. Briefly, 3 9 10 3 cells were used for each replicate. Drugs were added once 24 h before first XTT measurement. Viability was measured over 96 h in quadruplicates.

Transwell migration assay
Before harvesting the cells, culture media were exchanged the day before to 'starvation medium' containing only 2% FBS and 1% penicillin/streptomycin. Cells were incubated overnight at 37°C. To measure the migratory capacity, transwell culture inserts (8 µm pore size 'ThinCerts', Greiner Bio-One, Frickenhausen, Germany) were used in 24-well plates (Greiner Bio-One). In each insert, 2.5 9 10 4 cells were seeded in triplicates in 200 µL 'starvation medium'. 600 µL standard culture medium was added to the lower chamber. After 24 h, cells were fixed in ice-cold 100% methanol for 10 min at RT and stained with crystal violet (1 g/ 100 mL) at RT for 15 min. Inserts were washed twice with PBS, and the inside of the inserts was wiped with cotton swabs. Fixed cells on the outside of the membrane were lysed in 200 µL 10% acetic acid for 10 min at RT, of which 150 µL was transferred to a 96-well plate and the absorbance was measured at 595 nm (iMark Microplate Absorbance Reader, Bio-Rad, Feldkirchen, Germany).

Adhesion assay
24-well plates (Greiner Bio-One) were blocked with 1.5% bovine serum albumin in PBS for 1 h at RT.
Hereafter, supernatants were removed and cells (2 9 10 5 ) were seeded in triplicates and incubated for 1 h at 37°C. Non-adherent cells were removed by aspiration, and wells were washed twice with PBS. Remaining cells were fixed in ice-cold 100% methanol for 30 min at RT and stained with crystal violet (1 g/ 100 mL) at RT for 15 min. Crystal violet was removed by aspiration, and the wells were washed gently with PBS until no more discoloration appeared. The stained cells were lysed in 200 µL 10% acetic acid for 10 min at RT, of which 150 µL was transferred to a 96-well plate and the absorbance was measured at 595 nm (iMark Microplate Absorbance Reader, Bio-Rad).

RNA interference
Transfection of CD24 siRNA (Santa Cruz Biotechnology, Heidelberg, Germany) or scrambled RNA ('AllStars Negative Control siRNA', Qiagen, Hilden, Germany) has been performed as described previously [33]. Briefly, 1.5 9 10 5 cells were transfected with FuGene HD (Promega, Mannheim, Germany) (5 µL FuGene HD: 1 µg siRNA). Formation of siRNA-FuGene HD complexes was performed at RT for 15 min. Complexes were added slowly to the cells in standard culture medium without antibiotics. 24 h after transfection, cell culture medium was replaced with fresh standard medium with antibiotics.
2.10. DNA, RNA, and protein extraction DNA was isolated by phenol/chloroform/isoamyl alcohol as published previously [34]. For RNA extraction, cells were harvested and RNA was isolated using the RNeasy Mini Kit (Qiagen) according to the manufacturer's protocol. Concentrations and quality of isolate DNA/RNA were measured/checked by NanoDrop measurement (ratios 260/280 nm, 260/230 nm). Proteins were isolated by RIPA buffer as published [27]. Protein concentration was assessed by the 'BCA Protein Assay Reagent Kit' (Thermo Fisher Scientific).

DNA methylation analysis by sodiumbisulfite sequencing
Analyses of DNA methylation patterns by sodium bisulfite sequencing or methylation-specific PCR were performed as described previously [34]. 500 ng of genomic DNA was used for sodium bisulfite conversion ('EZ DNA Methylation-Gold' Kit; Zymo Research, Freiburg, Germany). See Table S2 for oligonucleotides used for PCR (Table S2).

Sanger sequencing
For Sanger sequencing, PCR-amplified DNA was cloned into the pCR2.1 vector using the 'TA cloning kit' (Thermo Fisher Scientific) according to the manual (ratio insert DNA to vector, 3:1). TOP10 E. coli bacteria were transformed according to the 'TA cloning kit' manual. Plasmid DNA was isolated from bacteria mini-preps following an alkaline lysis protocol. An EcoR1-HF restriction enzyme (New England Biolabs, Frankfurt a. M., Deutschland) double digest of 1 ug plasmid DNA for 30 min at 37°C, followed by agarose gel electrophoresis, confirmed insertion of the PCR product into the plasmid. Plasmids were Sangersequenced using M13 primers at Eurofins (Ebersberg, Germany).

Infinium MethylationEPIC BeadChip: sample preparation
Genomic DNA used for the Infinium Methyla-tionEPIC BeadChip array (850 k) was isolated and purified by two rounds of PCI precipitation including a RNaseA treatment. DNA concentrations were measured by the 'Qubit 4 Fluorometer' (Thermo Fisher Scientific). The Infinium MethylationEPIC array Bead-Chip (Illumina, San Diego, CA, USA) was carried out by the Epigenomic Services from Diagenode (Liege, Belgium, Cat. nr. G02090000) following an established in-house analysis pipeline. For each sample, 500 ng of DNA was used for sodium bisulfite conversion using the EZ-96 DNA Methylation Kit (Zymo Research). Bisulfite conversion was controlled by qPCR. One assay targeting a methylated region of DNAJC15 and two assays targeting the GNAS locus (one assay for the unmethylated allele and one assay for the methylated allele) were used for quality control. Deaminated DNA derived from blood was amplified in parallel and served as positive control. A sample passed the quality control when the received ct-value either for the two GNAS loci or for the DNAJC15 locus reaches the threshold not later than five cycles compared with the positive control.
2.14. Infinium MethylationEPIC BeadChip: array performance and data analysis Data and bioinformatic analyses were carried out using 'GENOMESTUDIO' software (version 2011.1; METHY-LATION MODULE v1.9) (Illumina), and a detailed report on quality controls and internal control measurements provided by Diagenode is available upon request. Beta-values were transformed into M-values according to Du et al. [35]. Data analyses were performed as described previously [36][37][38]. Raw and processed DNA methylation data have been uploaded to 'Gene Expression Omnibus' (GEO) (GSE176450). A two-group comparison was performed using the 'QLUCORE OMICS EXPLORER' software version 3.7 by means of a t-test to analyze for significance.

DNA dot blots
DNA dot blots for comparison of DNA methylation levels were performed as described and repeated twice [39]. Briefly, for each dot, 5 µL of a 1 : 2 dilution series of DNA (ranging from 500 to 62.5 ng diluted in H 2 O) was spotted onto a positively charged nylon membrane (Carl Roth GmbH + Co KG, Karlsruhe, Germany) and air-dried for 15 min. Afterward, the membrane was UV-cross-linked (20 s, 1200 mJÁcm À2 ) and blocked in 5% milk powder in PBST for 1 h at RT. The membrane was incubated with 5mC antibody overnight at 4°C and secondary antibody for 2 h at RT. Methylene blue-stained membranes (0.04% methylene blue in 5 M sodium acetate) served as loading control. Signals were detected using the ChemiDoc Imaging Systems, and dot intensity was quantified using the 'Image Lab' software (both from Bio-Rad). See Table S3 for antibody details.

CRISPR/Cas9-mediated genome editing
Genome editing via CRISPR/Cas has been performed as published previously [24]. A PCR strategy was used to confirm successful CRISPR/Cas9 reaction; a primer pair was designed to amplify a DNA fragment of 160 base pairs (bp), which is case of successful deletion of parts of exons 1 and 2 (PCR product 1). A second primer pair was designed to amplify a 'wild-type' sequence to screen for homo-or heterozygous CD24 deficiency (PCR product 2, 234 bp). See Table S2 for guide RNA sequences.

Chromatin immunoprecipitation followed by quantitative PCR (ChIP-qPCR)
ChIP-qPCR was performed as published previously [40]. Briefly, 1 9 10 7 cells were fixed for 30 min at RT using Diagenode 'Crosslink Gold' Kit (Diagenode). Cells were fixed 10 min in 1% formaldehyde (Appli-Chem, Darmstadt, Germany) and processed according to the 'SimpleChIP Plus Enzymatic Chromatin IP' Kit (Cell Signaling Technology, Leiden, The Netherlands). ChIP was carried out using 200 µg chromatin lysate and 5 µg antibody. 2% input (= 2 µg chromatin) and a ChIP using an IgG antibody served as controls. For ChIP-qPCR, 10 µL of IP samples was amplified using the 'Genome Plex Single Cell Whole Genome Amplification Kit' (Sigma-Aldrich) and subjected to qPCR at 1 : 40 dilution. Experiments were performed in triplicates. See Table S2 for used oligonucleotides and  Table S3 for utilized antibodies.

Quantitative RT-PCR
A total amount of 1 µg of RNA was in vitrotranscribed, and qRT-PCR was performed as described previously [27]. Gene expression was determined on the 384-well C1000 cycler (Bio-Rad). Each measurement was performed in triplicates. Utilized oligonucleotides are given in Table S2. GAPDH and ACTB were used as housekeeping genes and for data normalization.

RNA sequencing
RNA samples used for transcriptome analyses were quantified (Qubit RNA HS Assay, Thermo Fisher Scientific), and quality was measured by capillary electrophoresis using the Fragment Analyzer and the 'Total RNA Standard Sensitivity Assay' (Agilent Technologies Inc., Santa Clara, CA, USA). The library preparation was performed according to the manufacturer's protocol ('VAHTS Stranded mRNA-Seq Library Prep Kit'). 300 ng total RNA was used for mRNA capturing, fragmentation, synthesis of cDNA, adapter ligation, and library amplification. Bead purified libraries were normalized and finally sequenced on the HiSeq 3000/4000 System (Illumina, Inc.) with a read setup of 1 9 150 bp. The BCL2-FASTQ tool was used to convert the bcl files to fastq files as well for adapter trimming and demultiplexing. Data analyses on fastq files were conducted with CLC GENOMICS WORKBENCH (version 12.0.2; Qiagen). The reads of all probes were adapter (Illumina TruSeq) and quality-trimmed (using the default parameters: Bases below Q13 were trimmed from the end of the reads, ambiguous nucleotides max. 2). Mapping was done against the Homo sapiens (hg38) (May 25, 2017) genome sequence. Statistical differential expression tests were determined using the 'Differential Expression in TWO GROUPS' tool (version 1.02) (Qiagen). The resulting P-values were corrected for multiple testing by false discovery rate (FDR) and Bonferroni correction. A P-value ≤ 0.05 was considered significant. RNA-seq data are freely available via GEO (GSE168646).

Western blot
20 µg of whole protein lysates isolated by RIPA buffer was used for western blotting as described previously [36]. Beta-ACTIN or VINCULIN was used as housekeeper and for data normalization. Utilized antibodies are listed in Table S3.

Mass spectrometry: sample preparation
Proteins were extracted from frozen cell pellets (NCCIT-CD24 +/+ n = 3; NCCIT-DCD24 n = 5) as described elsewhere [45]. Briefly, cells were lysed and homogenized in urea buffer with a TissueLyser (Qiagen) and supernatants were collected after centrifugation for 15 min at 14 000 g and 4°C. Protein concentration was determined by means of Pierce 660 nm Protein Assay (Thermo Fischer Scientific), and 10 µg protein per sample was loaded on a SDS/PAGE for in-gel digestion. The isolated gel pieces were reduced (50 µL, 10 mM DTT), alkylated (50 µL, 50 mM iodoacetamide), and underwent afterward tryptic digestion (6 µL, 200 ng trypsin in 100 mM ammonium bicarbonate). The peptides were resolved in 0.1% trifluoroacetic acid and subjected to liquid chromatography.

Mass spectrometry: LC-MS analysis
For the LC-MS analysis, a Q Exactive Plus (Thermo Fisher Scientific) connected with an Ultimate 3000 Rapid Separation Liquid Chromatography System (Dionex/Thermo Fisher Scientific) equipped with an Acclaim PepMap 100 C18 column (75 µm inner diameter, 25 cm length, 2 mm particle size from Thermo Fisher Scientific) was applied. The length of the LC gradient was 120 min. The mass spectrometer was operating in positive mode and coupled with a nano-electrospray ionization source. Capillary temperature was set to 250°C and source voltage to 1.5 kV. In the Q Exactive Plus mass spectrometer for the survey scans, a mass range from 350 to 2000 m/z at a resolution of 140 000 was used. The automatic gain control was set to 3,000,000, and the maximum fill time was 80 ms. The ten most intensive peptide ions were isolated and fragmented by high-energy collision dissociation.

Mass spectrometry: data analysis, protein identification, and quantification
Mass spectrometric data were processed using the MAX-QUANT software version 1.6.17.0 (Max Planck Institute for Biochemistry, Planegg, Germany) applying standard parameters for label-free protein identification and quantification. Searches were carried out based on 74811 Homo sapiens protein entries downloaded from the UniProtKB on March 27, 2020, using tryptic cleavage specificity (behind K and R) and a maximum of two missed cleavage sites. Carbamidomethylation at cysteine residues was considered as fixed modification; methionine oxidation and N-terminal acetylation were considered as variable modifications. After an initial search using a precursor mass tolerance of 20 ppm and recalibration, a second search was performed with 4.5 ppm precursor mass tolerance. Tolerances for fragment spectra were 20 ppm. A FDR of 1% was used for both peptide and protein identification.
The intensity and LFQ intensity values received as output of the MAXQUANT software were normalized, respectively, toward zero median log 2 (fold change) in sample pairwise comparisons of the LFQ intensities over all identified proteins. The sample exhibiting the highest LFQ intensities, leading to positive log 2 (fold change) medians in pairwise comparisons to all other samples before normalization, was used as reference sample; the LFQ intensities of the other samples were scaled by the respective factors, that is, 2 (median log2(fold change)) .
The normalized data set was filtered by importing the protein list into the PERSEUS software version 1.6.6.0 (Max Planck Institute for Biochemistry) and removing potential contaminants, decoy hits, and proteins that were identified 'by site'. Additionally, a minimum of three valid values had to be present in at least one group. The 'Significance Analysis of Microarrays' (SAM) method was applied on intensity and LFQ intensity values, separately [46]. LFQ intensities were log 2 -transformed to reach a normal distribution like data structure, and missing values were filled in with random values from downshifted normal distribution (0.3 s. d. width, 1.8 s. d. downshift). Utilizing Student's t-test-based SAM analysis, with a constant S 0 of 0.17 (intensity values) or 0.12 (LFQ intensity values) and a 0.2 FDR-based cut-off, protein groups showing a significantly higher (or lower) abundance in the NCCIT-DCD24 samples with respect to the parental controls were referred to as up(or down)regulated proteins. As this method considers both the differences between the mean values of log 2 (LFQ) intensities and the standard deviation of repeated measurements, proteins were considered as up-or downregulated proteins on the basis of abundance differences between NCCIT-DCD24 and parental cells and low P-values. The final list of up-or downregulated proteins was obtained by combining the lists of the two separate SAM analyses (based on intensities or LFQ intensities) and filtering for the proteins that had Student's t-test P-value < 0.05 in both SAM analyses with the same direction of abundance change (upregulated in both analyses or downregulated in both analyses). The mass spectrometry proteomics data have been deposited to the 'ProteomeXchange Consortium' via the 'PRIDE' partner repository with the data set identifier PXD025110 [47].

Immunofluorescent staining
For immunofluorescent staining (IF) of CD24, 3 9 10 4 cells per well were seeded onto 96-well plates and incubated for 72 h. Cells were fixed by 4% formaldehyde for 10 min. Subsequently, cells were permeabilized for 5 min and blocked for 1 h at RT with 0.1% Triton X-100 in PBS and 1.5% BSA in PBS, respectively. CD24 antibody was incubated at a dilution of 1 : 50 in blocking buffer overnight at 4°C. The secondary antibody was added for 1 h in the dark at RT, and cells were counterstained with 0.5 lgÁmL À1 4 0 ,6-diamidine-2 0 -phenylindole dihydrochloride in PBS (DAPI, Sigma-Aldrich) for 3 min. Antibody details are depicted in Table S3.

Immunohistochemistry
As published previously, immunohistochemical stainings were performed on 4-lm formalin-fixed and paraffin-embedded tissue sections [48]. Antigen retrieval as carried out at 98°C in EDTA buffer (pH 9; 20 min). Afterward, primary antibodies were incubated for 30 min at RT before incubation with a HRP-labeled secondary antibody at RT for 25 min. Visualization of target antigen was performed by using 'DAB+ Chromogen' System (Dako, Hamburg, Germany). The experiments were undertaken with the understanding and written consent of each subject. Utilized antibodies are listed in Table S3.

Flow cytometry
Number of CD24 + cells were determined using a 'MACSQuant' flow cytometer with the 'Flowlogic' software (all from Miltenyi Biotec, Bergisch Gladbach, Germany). Cell suspensions were incubated with CD24-FITC antibody (1 : 100) diluted in 100 lL FACS buffer (phosphate-buffered saline (PBS), pH 7.2, 0.5% bovine serum albumin (BSA), and 2 mM EDTA) for 30 min in the dark at 4°C. Afterward, cells were washed once with FACS buffer before analysis of at least 5 9 10 4 cells. Utilized antibodies are listed in Table S3. Cell cycle analysis and apoptosis assay using PI or PI/Annexin V staining, respectively, with subsequent flow cytometric analysis have been performed as described previously [49].

CD24 antibodies
CD24 is a small and highly post-translationally modified membrane protein, making detection of CD24 by antibody challenging [50]. Thus, in this study, we utilized three different antibodies (SWA11, SN3, and REA832) to detect CD24 on protein level by various methods, guaranteeing reliable results that back up each other [7,50,51].

Online analysis tools and software
Publicly available 'The Cancer Genome Atlas' (TCGA) data sets were analyzed using the 'cBioPortal for Cancer Genomics', the 'UCSC Xena browser', and 'PINA' [53][54][55][56][57]. The STRING algorithm was used to predict protein-protein interaction by confidence [58]. The DAVID algorithm has been used to predict molecular functions of deregulated genes/proteins found in RNA-seq or mass spectrometry analyses based on 'Gene Ontology' (GO), 'Kyoto Encyclopedia of Genes and Genomes' (KEGG), UniProt, and INTEPRO [59]. 'MethPrimer 2' was used to design 'primers' for sodium bisulfite sequencing analyses and to screen for CpG dinucleotide density across the CD24 gene. 'Phyton' was used to generate volcano and violin plots from DNA methylation, RNA-seq, and MS data [60,61]. 'ClustVis' was used to perform principal component analysis (PCA) of RNA-seq data and the 'Qlucore Omics Explorer' to generate 3D-PCA from DNA methylation data [62].

Statistical analyses
Differences between groups were analyzed using twotailed Student's t-test after confirming equality of two variances according to the F-test. Statistically significant differences are highlighted by asterisk (*P < 0.05, *P < 0.05, or ***P < 0.005). Non-significant differences are labeled by 'n. s.'. Error bars are indicated by means of standard deviations (SDs).

Results
In this study, we characterized the expression of the signal transducer CD24 in GCTs and further deciphered its molecular and (epi)genetic features.

Mutational burden of CD24 in germ cell tumors
Next, we screened the TCGA GCT cohort for mutations in the CD24 gene locus (Fig. S1G). The vast majority of analyzed samples was extracted from testes, harbored the 12p gain, a high aneuploidy score, and was in the typical GCT age of 14-44 years at time point of diagnosis (Fig. S1G). No mutations of CD24 were found in GCTs, suggesting that overexpression of CD24 in ECs compared with the other GCT entities is not due to an amplification.
To correlate sample-specific CD24 DNA methylation to CD24 protein levels, we performed IHC of related GCT samples and found that DNA methylation levels did not correlate to CD24 protein levels (Fig. 2B, Fig. S2D). Within each sample, we found the CD24 CpG island either fully methylated or completely demethylated, mimicking an imprinting-like DNA methylation pattern (Fig. S2B).
To test whether demethylation of the CD24 CpG island derepresses CD24 gene expression, we treated CD24and CD24 CpG island hypermethylated JAR, hemimethylated BeWo, and unmethylated JEG3 (and TCam-2) cells over 96 h with 20 or 50 nM of the DNA-demethylating agent 5-aza-2'-deoxycytidine (decitabine) and demonstrated a dose-dependent induction of CD24 expression only in cells harboring CD24 hemi-or hypermethylation (JAR, BeWo) (Fig. 2C). By sodium bisulfite sequencing, we demonstrated that upregulation of CD24 expression coincided with demethylation of the CD24 CpG island in JAR cells 96 h after 20 (to 33%) and 50 nM (to 10%) decitabine treatment (Fig. S2E). These data suggest that at least in CD24 hemi-or hypermethylated CC-like cells, CD24 expression could be influenced by demethylating the corresponding CpG island.
3.6. The role of SOX2 in regulating CD24 expression Next, we aimed at deciphering the transcriptional regulation of CD24 in GCTs. Since CD24 expression can be induced by SOX2 in melanoma cells, we asked whether CD24 might be regulated by SOX2 in ECs as well [78]. Thus, we first analyzed chromatin immunoprecipitation followed by sequencing (ChIP-seq) data of SOX2 in the EC cell line 2102EP (SOX2 + /CD24 + ) and found four SOX binding sites within or in close proximity to the genomic region coding for CD24 (Fig. 2E, regions 1-4) [40]. Only one SOX binding site (region 4, peak intensity > 100) upstream of the CD24 transcription start site (TSS) was bound by SOX2, suggesting that SOX2 transactivates CD24 expression via binding to this region (canonical SOX2 motif: TTTTCAGATGCAAAT) (Fig. 2E). We also checked whether SOX17, which in part shows redundancy to SOX2 in regulating pluripotency in seminomas, binds to CD24 as well [40]. In the SOX17 + /SOX2 -/CD24seminoma cell line TCam-2, only negligible binding intensities of SOX17 to CD24 could be detected (peak intensities < 15) (Fig. 2E). We verified binding of SOX2 to region 4 of CD24 by performing ChIP-qPCR analysis in 2102EP in vitro and in vivo (xenografted into nude mice; six weeks of growth) (Fig. 2F). Again, SOX17 did not bind to CD24 region 4 in TCam-2 cells, but upon in vivo reprogramming of TCam-2 into an EC-like cell fate (xenografted into nude mice; 6 weeks of growth; accompanied by SOX2 up-and SOX17 downregulation [36]), binding of SOX2 to CD24 could be detected (Fig. 2F). Taken together, in GCTs SOX2, but not SOX17, seems to be able to transactivate CD24 expression by binding to a region in proximity to the TSS.
In contrast to data gathered in breast cancer cells, the cell cycle profiles, proliferation rates, and the adhesion abilities were not different between CD24deficient and parental cells, but the migratory capacity trended to be increased (Fig. 3C-F) [87,88].
NANOG and OCT3/4 as found by the RNA-seq analysis ( Fig. S4E; Table S4A). Furthermore, by using liquid chromatography paired with mass spectrometry (LC-MS), we took a proteome snapshot of NCCIT-DCD24 (n = 5) and parental cells (n = 3) (Table S4C). A PCA demonstrated that NCCIT-DCD24 cells clearly clustered apart from parental cells (Fig. S5A). 163 proteins were significantly enriched in the proteome of NCCIT-DCD24 compared with NCCIT-CD24 +/+ cells, and 99 were depleted (> FC 2) (Fig. S5B; Table S4D). By utilizing the STRING and DAVID algorithm again, we demonstrated that these proteins are involved in similar processes as found when analyzing the RNA-seq data (Fig. S5C,D; Table S4D). Additionally, in the sets of enriched and depleted proteins, we found proteins associated with chromatin and protein modifications, such as acetylation, methylation, ubiquitination, and citrullination (Fig. S5C,D, Table S4D).

Correlation of gene expression dynamics to alterations in DNA methylation
Our previous data demonstrated a link between CD24 and the epigenetic landscape. Therefore, we analyzed changes in DNA methylation (5mC) in CD24-deficient NCCIT and 2102EP cells compared with parental controls by Infinium MethylationEPIC bead arrays (Fig. 5, Table S4E). A PCA demonstrated that NCCIT-/2102EP-DCD24 cells clearly clustered apart from parental cells (Fig. 5A). Global DNA methylation levels increased only slightly in NCCIT/2102EP-DCD24 cells compared with parental controls (NCCIT: + 2%; 2102EP: + 3%) (Fig. 5B). We confirmed this increase in 5mC levels by DNA dot blotting using a 5mC antibody (Fig. S4F). We identified 3157 differentially methylated CpG dinucleotides (2227 5mC up; 930 5mC down; DM-value: 1.5) in NCCIT (DCD24 vs. control) and 5300 in 2102EP (3429 5mC up; 1871 5mC down) (Fig. 5C; Table S4E). Changes in DNA methylation of these differentially methylated CpGs occurred mainly in CpG island context and around transcription start sites (TSS220, TSS1500) (Fig. 5D). 163 CpG dinucleotides were commonly altered in CD24-deficient NCCIT and 2102EP cells (104 5mC up; 59 5mC down) (Table S4C). In NCCIT cells, we correlated changes in 5mC to changes in gene expression (GEX) found by the RNA-seq analysis (Fig. 5E,F; Table S4F). For subsequent analysis, we only considered genes linked to a known 'gene symbol' and with significant alterations in DNA methylation (difference in Mvalues > 1.5) in at least three CpG dinucleotides in all three CD24-deficient samples and with a significant difference in gene expression of FC > log 2 1.5 (Table S4F). In the CD24-deficient cells, we found 6 genes that were upregulated in expression and showed a decrease in 5mC (Group 1) and 28 genes that were downregulated in expression and showed an increase in 5mC (Group 2) (Fig. 5E,F). Additionally, we identified genes upregulated in expression and showing an increase in DNA methylation (Group 3), but found no gene showing downregulation in expression and a decrease in DNA methylation (Fig. 5F, Table S4F). Among Group 1 (5mC down, GEX up) were differentiation factors such as MYOD1 and HNF1A and the zinc finger protein basonuclin-1 coding gene BCN1, which is highly abundant in germ cells (Figs 5F and  4D). Thus, we postulate that expression of these genes might be influenced by DNA methylation. Among Group 2 (5mC up, GEX down) were previously identified STRING interactors such as HPS3, IDUA, SVOPL, and TNFSF11/12, as well as two claudins (CLDN3/23) (Figs 5F and 4D).

The effects of CD24 deficiency on differentiation of EC cells
Analyses of the CD24-deficient NCCIT cells pointed at an involvement of CD24 in poising the cells for ectodermal differentiation and blocking mesodermal differentiation.
Thus, we asked how RA-mediated differentiation is altered in pluripotent and RA-sensitive NCCIT-DCD24 cells compared with the parental cells. Therefore, cells were treated with RA (20 µM) for ten days and we analyzed changes in gene expression by qRT-PCR afterward (Fig. 6A). Besides changes in the morphological appearance, a strong upregulation of RARB and downregulation of the pluripotency factors NANOG, OCT3/4, and SOX2 in NCCIT-DCD24 and parental cells indicated a successful RA treatment (Fig. 6A) [41]. Mesodermal (BRACHYURY (T), EOMES, HAND1) and endodermal (AFP, GATA6, SOX17) genes were more strongly upregulated, while ectodermal genes (CDX2, GATA2, PAX6) were more downregulated in NCCIT-DCD24 compared with the NCCIT-CD24 +/+ cells (Fig. 6A). We also included nullipotent 2102EP cells, but as expected, RA application did not induce differentiation, indicated by an unchanged morphology and expression of genes indicative of a successful RA treatment, that is, RARB, PAX6, and HAND1 (Fig. 6A) [41]. These findings strengthen our hypothesis that CD24 plays a role in cellular differentiation processes and is involved in blocking mesodermal and endodermal differentiation and poising pluripotent EC cells toward ectodermal differentiation (Fig. 6A).

Blocking CD24 as a new therapeutic option to render EC cells that are cisplatin-sensitive
Finally, we analyzed whether blocking CD24 by the SWA11 antibody affects efficacy of cisplatin application in EC cell lines including cisplatin-resistant subclones (-R) [51]. By flow cytometry-mediated measurement of apoptosis rates, we demonstrated that blocking CD24 considerably increased efficacy of cisplatin in all tested cell lines, while application of the SWA11 antibody alone had only minor effects (Fig. 6B). In conclusion, targeting CD24 in combination with cisplatin might be a reasonable concept to treat (cisplatin-resistant) CD24 + GCT cells.  . Two-tailed t-tests were performed to test for significance; *P-value < 0.05, **P-value < 0.005, and ***P-value < 0.0005.

Discussion
In this study, we characterized the molecular and epigenetic mechanisms regulating the signal transducer and glycoprotein CD24 in GCTs.
Overexpression of CD24 has been found in various cancers, such as lung, breast, and ovarian [9]. In GCTs, CD24 (isoform ENST00000606017.1) is strongly expressed in ECs and absent to very weakly expressed in germ cells of different developmental stages, seminomas, YSTs, and CCs (Fig. 7A). GCNIS were completely CD24 -, suggesting that induction of CD24 occurs mainly during reprogramming of GCNIS into an EC and is downregulated again upon differentiation into non-seminomatous subentities, which is in line with strong downregulation of CD24 in RAtreated EC cells (Fig. 7B). Thus, in ECs, CD24 expression can be associated with an undifferentiated and pluripotent cell fate.
How is CD24 regulated in GCTs? We demonstrated that SOX2 binds to a SOX motif in proximity to the CD24 transcription start site and transactivates CD24 in ECs (Fig. 7A). In line, Vencken et al. [89] demonstrated that a RNAi-mediated knockdown of SOX2 in 2102EP cells led to a 3.3-fold downregulation of CD24 expression. Additionally, SOX2 has already been shown to induce CD24 expression in a melanoma cell or a murine model [78,79,90,91].
So, SOX2 is a bona fide transactivator of CD24 expression in ECs. In our RNA-seq analysis and related qRT-PCR validation, SOX17 was upregulated 3.7-fold in NCCIT-DCD24 cells compared with the parental clones (Table S4A; Fig. S4D). Additionally, SOX17 became strongly upregulated in RA-treated NCCIT-DCD24 cells compared with RA-treated parental cells (Fig. 6A). Thus, CD24 might be involved in suppressing SOX17 expression in ECs.
Our results suggest an involvement of epigenetic mechanisms in the regulation of CD24. On the one hand, we found sample-dependent and highly variable CD24 DNA methylation patterns, which did not correlate to CD24 mRNA or CD24 protein levels. On the other hand, CD24 expression could be derepressed upon decitabine-mediated DNA demethylation in CC cell lines, suggesting that at least in choriocarcinomas, DNA methylation might be involved in silencing CD24. Transcriptional regulation of CD24 independent of the DNA methylation status has been described in breast cancers [77]. Thus, DNA methylation seems to play a minor role in regulation of CD24 expression in GCTs. CD24 itself influences the DNA methylation landscape only slightly, but induces sitespecific changes in DNA methylation of factors deregulated in CD24-deficient EC cells, such as demethylation of differentiation-and germ cell-related factors such as MYOD1 and BNC1.
Our data point at histone acetylation as an additional regulator of CD24, since application of HDACi led to histone hyperacetylation, resulting in CD24 upregulation in CD24 -GCT cell lines. In CD24 + EC cells, where CD24 expression remained unaffected upon HDACi application, CD24 expression was downregulated upon treatment with inhibitors of histone methyltransferases (HMTi), E3 ubiquitin ligases (E3-Uli), or histone-code readers (BRDi), indicating that histone methylation or the polycomb repressive complex may be required for proper CD24 expression in ECs (Fig. 7B).
Some GCT cell lines (TCam-2, 1411H, GCT72) show CD24 expression on mRNA level, but do not translate CD24 protein. Here, we propose that CD24 mRNA is bound by microRNAs (miR), thereby preventing initiation of translation. Regulation of CD24 by miR34a, miR-1185-1, and miR146a has already been proposed in colon and breast cancer cells and oral squamous cell carcinomas [92][93][94].
We also found that three-dimensional interaction of CD24 + EC cells with fibroblasts, and endothelial or immune cells is able to reduce CD24 expression. Thus, interaction of EC cells with the somatic microenvironment, for example, during metastasis, might lead to CD24 downregulation, putatively sensitizing the cells to differentiation. Our results indicate that mainly the interaction with immune cells (T cells, M2 macrophages) leads to CD24 downregulation. Thus, triggering the immune system by immunemodulating therapy might also lead to CD24 downregulation, which in turn may render the tumor cells more sensitive toward cisplatin (Fig. 6B). In conclusion, combining immunotherapy with CD24 blocking agents could be a promising therapeutic approach to resensitize chemotherapy-resistant tumors toward cisplatin-based therapy.
By establishing CD24-deficient EC cell clones, we demonstrated that CD24 could be involved in blocking the germ cell/spermatogenesis program and mesodermal differentiation in ECs (Fig. 7C). Among the genes upregulated in NCCIT-DCD24 cells are many homeobox or glycoproteins involved in regulation of transcription and protein processing, suggesting that CD24 interacts with homeobox and glycoproteins to fulfill its molecular tasks (Fig. 7C). On the other side, CD24 deficiency led to a decrease in ectodermal marker genes associated with neuronal, olfactory, and eye-related differentiation (Fig. 7C). Furthermore, our mass spectrometric analysis suggested that CD24 is involved in modifying the chromatin and/or proteins by acetylation, methylation, ubiquitination, or citrullination. Thus, CD24 also utilizes/influences epigenetic mechanisms to mediate its functions. In conclusion, CD24 has a bivalent function in regulating developmental and differentiation processes by negatively and positively regulating glycoproteins; that is, it blocks the germ cell program and mesodermal differentiation and is able to poise cells for ectodermal differentiation. In line, CD24-expressing neural precursor cells generated from pluripotent stem cells effectively differentiated along the neuronal lineage in vitro [95].
In our IHC analysis, CD24 was absent in germ cells of different developmental stages, GCNIS cells, and upregulated during reprogramming to an EC, which closely resemble pluripotent embryonal stem cells. Seminomas, which mimic PGC/GCNIS cells, were also CD24 -. Thus, we hypothesize that in GCT entities resembling (primordial) germ cells (GCNIS, seminoma), CD24 is absent, allowing for expression of germ cell factors, while in embryonal stem cell-like ECs, which show the least relationship to germ cells, CD24 is involved in blocking the germ cell program.
Finally, we demonstrated that targeting CD24 might be reasonable approach to enhance efficacy of cisplatin to treat (cisplatin-resistant) GCT cells. So far, several CD24 targeting strategies based on the SWA11 antibody exist, for example an antibody-drug conjugate (CD24-PE38) and 3 rd -generation CD24-CAR-NK cells [96,97]. Furthermore, the SWA11 antibody used in this study has already been shown to reduce tumorigenicity in vivo [51,88]. Thus, in future studies the suitability of targeting CD24 in the treatment of (refractory) GCTs has to be evaluated in more detail.

Conclusion
In summary, CD24 is involved in suppressing the germ cell program during formation of ECs and promotes differentiation rather into ectodermal than mesodermal lineage upon differentiation-inducing stimuli. CD24 expression can be transactivated by the pluripotency factor SOX2 and reactivated in CD24 -GCTs by histone deacetylase inhibitors or DNA-demethylating agents. In conclusion, this study deciphered the molecular function CD24 in ECs and its (epi)genetic mechanisms of regulation, setting the stage for further analyses on the suitability of CD24 as a therapeutic target in ECs.

Ethics statement
The study methodologies conformed to the standards set by the Declaration of Helsinki. The study methodologies were approved by the local ethics committee. The ethics committee of the Heinrich Heine University D€ usseldorf raised no concerns on using analyzed cell lines for in vitro experiments and drug screening (ethics votes 2018-178 and 2019-412 to DN). The ethics committees of the University Bonn and the University Medical Center G€ ottingen raised no concerns on performing analyses on GCT tissues of local biobanks. Experiments were undertaken with the understanding and written consent of each subject. All animal experiments were performed under license of the 'Landesamt f€ ur Natur und Umwelt NRW' (AZ-84-02.04.2013-A430).

Data accessibility
RNA-seq data are freely available via 'Gene Expression Omnibus' (GEO) (GSE168646). Illumina and Affymetrix expression microarray data are available via GEO (GSE71239, GSE71269, GSE79065, and GSE60698). The mass spectrometry proteomics data have been deposited to the 'ProteomeXchange Consortium' via the 'PRIDE' partner repository (PXD025110). Illumina Infinium MethylationEPIC bead array data are available via GEO (GSE176450).

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article. Fig. S1. (A) Illustration of CD24 isoform expression in GCT and testis tissues as well as other cancer entities and corresponding tissues of origin based on the TCGA and GTEX cohorts. Location of 'primers' used in this study for qRT-PCR analysis of various isoforms are indicated by purple arrow heads. (B) RNA sequencing data of CD24 expression throughout various cancer types based on data extracted from TCGA. Data has been illustrated using 'Firebrowse'. (C) Representative western blot analysis (n = 3) of CD24 protein levels (SWA11 antibody) in GCT cell lines and fibroblasts (MPAF). (D) Summary of immunohistochemistry data of indicated proteins in GCT tissues. (E) HE and immunohistochemical staining of CD24 in a mixed GCT composed of EC (OCT3/4 + ) and YST (FOXA2 + ) cells (picture of CD24 staining is also given in Fig. 1 C). (F) Immunohistochemical staining of CD24 in normal testis tissue. (G) Mutational burden of CD24 in GCT tissues based on the TCGA 'testicular germ cell tumor' cohort. Data has been illustrated using 'cBioPortal'.   (F) Representative densitometric analysis of DNA dot blot data (n = 3) using a 5mC antibody in three NCCIT / 2102EP-DCD24 clones and parental cells. Data were normalized against methylene blue staining (MB). Two-tailed t-tests were performed to test for significance; * = p-value < 0.05, ** = p-value < 0.005, *** = p-value < 0.0005.  Table S1. Cell lines used in this study including cultivation conditions. Table S2. Oligonucleotides used in this study. Table S3. Antibodies used in this study. Table S4. RNA seq, 850k DNA methylation array and MSA data.