Surface and Global Proteome Analyses Identify ENPP1 and Other Surface Proteins as Actionable Immunotherapeutic Targets in Ewing Sarcoma

Abstract

Purpose:

Ewing sarcoma is the second most common bone sarcoma in children, with 1 case per 1.5 million in the United States. Although the survival rate of patients diagnosed with localized disease is approximately 70%, this decreases to approximately 30% for patients with metastatic disease and only approximately 10% for treatment-refractory disease, which have not changed for decades. Therefore, new therapeutic strategies are urgently needed for metastatic and refractory Ewing sarcoma.

Experimental Design:

This study analyzed 19 unique Ewing sarcoma patient- or cell line–derived xenografts (from 14 primary and 5 metastatic specimens) using proteomics to identify surface proteins for potential immunotherapeutic targeting. Plasma membranes were enriched using density gradient ultracentrifugation and compared with a reference standard of 12 immortalized non–Ewing sarcoma cell lines prepared in a similar manner. In parallel, global proteome analysis was carried out on each model to complement the surfaceome data. All models were analyzed by Tandem Mass Tags–based mass spectrometry to quantify identified proteins.

Results:

The surfaceome and global proteome analyses identified 1,131 and 1,030 annotated surface proteins, respectively. Among surface proteins identified, both approaches identified known Ewing sarcoma–associated proteins, including IL1RAP, CD99, STEAP1, and ADGRG2, and many new cell surface targets, including ENPP1 and CDH11. Robust staining of ENPP1 was demonstrated in Ewing sarcoma tumors compared with other childhood sarcomas and normal tissues.

Conclusions:

Our comprehensive proteomic characterization of the Ewing sarcoma surfaceome provides a rich resource of surface-expressed proteins in Ewing sarcoma. This dataset provides the preclinical justification for exploration of targets such as ENPP1 for potential immunotherapeutic application in Ewing sarcoma.

See related commentary by Bailey, p. 934

Translational Relevance.

The survival rate for patients with Ewing sarcoma with localized disease is approximately 70%, but decreases to approximately 30% for metastatic disease and approximately 10% for treatment-refractory disease. Although immunotherapy in adult cancers is progressing rapidly, this approach is much less well-developed for pediatric solid malignancies. Targeting Ewing sarcoma with immunotherapy has the potential to transform Ewing sarcoma treatment paradigms. To address this critical clinical need, we have characterized the surface and global proteomes of Ewing sarcoma using patient-derived xenograft models of Ewing sarcoma. In addition to several known Ewing sarcoma surface targets, we identified many new candidate targets with ostensible high Ewing sarcoma specificity and low normal tissue expression. We highlight ENPP1, which displays robust Ewing sarcoma surface expression and low expression in normal pediatric tissue, as a promising immunotherapeutic candidate. Our comprehensive Ewing sarcoma surfaceome characterization provides a broad translational resource for developing novel immunotherapies targeting Ewing sarcoma and is a model for application to other high-risk pediatric cancers.

Introduction

Ewing sarcoma is the second most common primary bone cancer of children and young adults, with an incidence of roughly 1 case per 1.5 million persons in the United States (US; ref. 1). Ewing sarcoma accounts for roughly 2% of all cancers in children, and is primarily found in large proximal bones such as the femur, or contained within the pelvis (1). Histologically, Ewing sarcoma typically presents as a cluster of undifferentiated small, round, blue cells with a large nucleus and limited cytoplasmic regions (1). This disease is characterized by hallmark gene fusions between the FET gene family (typically EWSR1) and a member of the E-Twenty Six (ETS) gene family (typically FLI1 or ERG; refs. 2, 3), with EWSR1–FLI1 or EWSR1–ERG fusion transcripts being detected in about 85% or 10%–15% of all patients, respectively (2, 3). Treatment of Ewing sarcoma involves a combination of surgery, localized radiotherapy, and chemotherapy (1). Dramatic clinical improvements to survival have been demonstrated over the past 50 years, with the 5-year survival of localized/regional disease increasing from just 44% to 70%–82% in recent years (4). Unfortunately, although some improvements to survival have been demonstrated in patients with metastatic disease, the 5-year survival rate is still only approximately 30%. Furthermore, patients who present with refractory disease within 2 years of their initial diagnosis are largely incurable, with a demonstrated 5-year survival rate of <10% (4). Finally, the comorbidities associated with chemotherapy and radiotherapy at such a young age can have life altering effects on these patients, including severe physical, emotional, and psychological side effects (4). Clearly, there is a critical unmet clinical need for developing novel therapeutic opportunities for patients diagnosed with this disease.

Although immunotherapy has been successful against certain adult solid cancers (5), its promise in pediatric cancers lags behind (excluding hematologic malignancies). Examples in patients with pediatric cancer include the PD1 checkpoint inhibitor Pembrolizumab in the treatment of relapsed/refractory Hodgkin lymphoma (6), and the anti-CD19 chimeric antigen receptor (CAR) T-cell therapy, Tisagenlecleucel, with efficacy in patients with B-cell acute lymphoblastic leukemia (7). However, for rarer solid childhood cancers like Ewing sarcoma, no immunotherapeutic (IT) options have been implemented clinically. The use of antibody–drug conjugates (ADC), immune cell engagers, and CAR T cells to target malignant cells is an appealing prospect for use in Ewing sarcoma; however, the success of IT approaches relies on the discovery of strong candidate surface markers specific to the cancer of interest and with minimal expression in normal tissues. A comprehensive understanding of the Ewing sarcoma surfaceome and complementary proteome would be a critical resource for novel IT treatment strategies for the research community.

Efforts to characterize cancer surfaceomes have used cell surface capture techniques, which involve oxidizing and labeling surface glycans before purification via strepdavidin bead capture and enzymatic release for analysis by mass spectrometry (8). This strategy has been implemented to characterize the surfaceome of cancers like multiple myeloma (9). However, this technology requires viable/intact cells for labeling and so is not yet applicable to cancers where frozen tumor tissue must be dissociated. Other efforts to identify surface proteins for immunotherapy in cancers have relied on RNA-sequencing data to infer surface protein expression from detected transcripts (10). This approach has identified several surface protein targets in pediatric cancers, including Ewing sarcoma (10). Although promising targets have been identified in these studies, inferring surface protein expression from RNA data can be problematic as RNA expression often does not reliably correlate with protein expression (11). Furthermore, although proteomics has been used to characterize the surfaceome of osteosarcoma (12), there has been no rigorous evaluation of the Ewing sarcoma–specific surfaceome to identify actionable targets.

Here, we present a comprehensive proteomic cataloguing of the Ewing sarcoma surfaceome comprising density gradient ultracentrifugation membrane enrichment (13) coupled to quantitative mass spectrometry of multiple Ewing sarcoma xenograft models (14). We complement the surfaceome data with matched global proteome data to assist in understanding the phenotypic landscape of the Ewing sarcoma proteome. Proteins from our surfaceome and global-proteome efforts were compared with a reference standard consisting of an identically prepared mixture of 12 common cell lines (termed SuperMix, SM), to prioritize Ewing sarcoma–specific proteins. We identified Ewing sarcoma–associated surface proteins enriched in our data, including IL1RAP, CD99, STEAP1, and ADGRG2, highlighting the validity of our approach. Furthermore, many other promising candidate markers were also discovered that had elevated surface expression when compared with the SM reference, particularly ecto-nucleotide pyrophosphatase/phosphodiesterase 1 (ENPP1) and Cadherin-11 (CDH11), both of which have only limited prior associations with Ewing sarcoma. Target expression was confirmed in a panel of Ewing sarcoma cell lines, and enhanced staining of ENPP1 was demonstrated using an Ewing sarcoma tissue microarray (TMA). To our knowledge, this is not only the first quantitative proteomic characterization of the Ewing sarcoma surfaceome, but also the first study to carry out deep global quantitative proteomic profiling of the disease. We present these data as a resource for the broader community to inform decisions on surface target prioritization for Ewing sarcoma immunotherapy.

Materials and Methods

Cell culture and cell lines

Cell lines (A673, CHLA10, HEK293t, SiHa, TC32, and U2OS) were purchased from the ATCC. A673, HEK293t, SiHa, and U2OS were grown in DMEM (Gibco) supplemented with 10% FBS (Gibco). TC32 were grown in RPMI-1640 (Gibco) supplemented with 10% FBS. CHLA-10 were grown in IMDM supplemented with 20% FBS and 1X ITS (5 μg/mL insulin, 5 μg/mL transferrin, 5 ng/mL selenous acid; Gibco). For subculturing, the spent medium was aspirated, and cells were washed in 1X PBS (Gibco) before detachment with Tryp-LE (Gibco). Following detachment, cells were passaged at the appropriate subcultivation ratio (1:4 – 1:6) and maintained at 37°C in 5% CO₂. For seeding at specific densities, viable cells were stained with Trypan Blue (Gibco) and manually counted using a hemocytometer before seeding. The SM cell lines (Supplementary Table S1; A594, HCT-116, HEK293t, HEPG2, GS578T, JURKAT, K-562, PANC-1, PC3, SiHa, SK-MEL-2, and TOV-21G) were purchased (Cell Culture Company) as a total of 50 × 10⁶ pellets except for the SiHa cell line, which was bulk-passaged in house and collected in pellets of 50 × 10⁶. The SM was created by mixing equal cell numbers of each line before extraction. Supplementary Table S1 lists the cell lines in the SM reference standard. Cells were routinely confirmed to be free of Mycoplasma contamination using the MycoStrip Mycoplasma detection kit (InvivoGen). Cell line authentication was carried out using short tandem repeat testing and experiments were performed with cell lines no more than 20 passages post-thaw.

Immunofluorescence

Cells were counted and seeded on chamber glass slides (Millipore) at a density of 10,000 cells per well. Two days later, the medium was removed, and following a brief wash with PBS, cells were fixed using 4% paraformaldehyde (Sigma-Aldrich) for 10 minutes at room temperature. Cells were next permeabilized and blocked in a solution of 2% BSA containing 0.2% Triton X-100 for 30 minutes at room temperature. Following 3 washes with PBS, cells were incubated with primary antibodies against selected targets of interest (manufacturer's recommended concentration in 2% BSA) overnight at 4°C. The following primary antibodies were used: Anti-ATP1A1 (ATPase; ab7671, Abcam), anti-CDH11 (71–7600, Invitrogen), anti-ENPP1 (ab223268, Abcam), and anti-Tubulin (ab7291, Abcam). The next day, following 3 washes with PBS, cells were incubated with AlexaFluor secondary antibodies (Goat anti-Rabbit IgG Alexa Fluor 488 #A-11008, Goat anti-Mouse IgG Alexa Fluor 555 #A-21422, Thermo Fisher Scientific) raised against the primary antibody species of interest, for 1 hour at room temperature. Finally, cells were washed 3 times with PBS and a coverslip was applied atop the cells with a small amount of DAPI-counterstain mounting medium (Vector Laboratories) and sealed using nail polish.

Western blot analysis

Cell lysates were prepared following extraction in radioimmunoprecipitation buffer (RIPA; 150 mmol/L NaCl, 50 mmol/L Tris-Cl (pH 7.4), 1% (v/v) Nonident (NP)-40, 0.5% (w/v) Sodium deoxycholate, 0.1% (w/v) SDS). Briefly, cells were removed from the incubator, rinsed with PBS, and scraped on the plate in 1-mL RIPA buffer. Cells were incubated on ice for 10 minutes with vortexing every 2 minutes until the lysate was fully fluid. Cells were centrifuged at 18,000 x g for 10 minutes to remove the insoluble fraction, the soluble fraction was collected to a new tube, and protein content was measured using the BCA assay as per the manufacturer's recommended guidelines. Following extraction, loading buffer was added to 10–50 μg of each sample at a concentration of 1:5, boiled at 100°C for 10 minutes, and spun briefly to recover any precipitated material. Samples were loaded on 10% Tris-Glycine SDS-PAGE and resolved. After the material was transferred to nitrocellulose membranes for 2 hours at 300 mA at 4°C, the filters were blocked in 5% Milk-TBS + 0.1% tween 20, and probed overnight with the primary antibody of interest at 4°C with constant agitation. Horseradish peroxidase (HRP)–linked secondary antibodies targeting the species of the primary antibody were used for visualizing protein bands following the addition of electrochemiluminesence prime Western blotting detection reagent (Cytiva) as per the manufacturer's guidelines. Bands were imaged on a Bio-Rad Go Imaging System. The following primary antibodies were used: Anti-ATP1A1 (ATPase; ab7671, Abcam), anti-CDH11 (1:1,000, 71–7600, Invitrogen), anti-ENPP1 (1:1,000, ab223268, Abcam), and anti-Tubulin (1:5,000, ab7291, Abcam). The following HRP-linked secondary antibodies were used: Anti-Goat (sc-2354, Santa Cruz Biotechnology), anti-Rabbit (1:5,000, NA934, Cytiva), anti-Mouse (1:5,000, NA931, Cytiva).

In vitro plasma membrane isolation

Separation of cell lines into cytoplasmic, membrane, and plasma membrane fractions was carried out using the Minute Plasma Membrane Protein Isolation and Cell Fractionation Kit (Invent Biotechnologies INC) as per the manufacturer's guidelines with the following supplemented information; to ensure a sufficient yield was achieved in the plasma membrane fraction, at least 2 × 15 cm plates of cells at 80% confluent were used. Cells were detached by scraping, rather than trypsinization, so ensure any surface epitopes remained intact. Protein from cell fractions was dissolved and extracted in RIPA buffer as mentioned previously. All fractions were normalized using the BCA assay before 1D SDS-PAGE western blot analysis, and equal amounts of material was loaded to each well.

Patient-derived xenograft models

Patient-derived xenograft (PDX) or cell line–derived xenograft (CDX) models were sourced from the Pediatric Preclinical Testing Consortium (PPTC), as previously described (14, 15), or from other sources (16–18). Written informed consent was obtained from patients. Metadata such as sample type, sex, primary/metastasis status, and age is described in Table 1. The primary inclusion criterion was that each xenograft must have come from a confirmed case of Ewing sarcoma. No randomization, blinding, or power analysis was performed before sample selection. PDX samples were used in accordance to recognized ethical guidelines, and approved by the local University of British Columbia human Research Ethics Board under protocol H19–00016. The majority of models were provided as duplicates of PDX or CDX tumor specimens in individual tubes and were processed as independent samples. For models where independent sections could not be obtained, samples were analyzed on the mass spectrometer as two technical replicates dispersed across independent plexes in an effort to obtain a more accurate quantification of identified peptides.

Table 1.

Data for models used in current study.

Model ID	Sample type	Primary or metastatic	Age (y)	Sex	Site of disease	Status	Source of xenograft
CHLA10	CDX	Metastasis	14	Female	Thoracic lymph node	Relapse	El-Naggar AM et al. EMBO Rep. 2019
CHLA258	CDX	Primary	14	Female	Lung	Relapse	Rokita JL et al. Cell Rep 2019
IC-pPDX-3	PDX	Primary	16	Female	Humerus	Diagnosis	El-Naggar AM et al. EMBO Rep. 2019, Aynaud et al. Cell Rep 2020
ES1	CDX	Primary	45	Female	Left thigh	Diagnosis	Rokita JL et al. Cell Rep 2019
ES2	CDX	Metastasis	14	Female	Ileum/bone marrow	Relapse	Del Pozo V et al. iScience. 2021
ES3	CDX	Primary	12	Male	Pubis/bone marrow	Relapse	Del Pozo V et al. iScience. 2021
ES4	CDX	Primary	18	Male	8th rib	Relapse	Rokita JL et al. Cell Rep 2019
ES6	CDX	Primary	17	Male	Right proximal tibia	Relapse	Rokita JL et al. Cell Rep 2019
ES7	CDX	Primary	15	Male	Right fibula/bone marrow	Relapse	Del Pozo V et al. iScience. 2021
ES8	CDX	Primary	10	Male	Left proximal humerus/bone marrow	Relapse	Rokita JL et al. Cell Rep 2019
EW12	PDX	Primary	13	Male	Left foot mass	Diagnosis	Unpublished model sourced from the PPTC
EW13	PDX	Primary	10	Male	Right proximal tibia	Diagnosis	Del Pozo V et al. iScience. 2021
EW5	PDX	Primary	16	Male	Paraspinal	Diagnosis	Rokita JL et al. Cell Rep 2019
EW8	CDX	Primary	17	Male	Abdominal mass	Diagnosis	Rokita JL et al. Cell Rep 2019
EW9	PDX	Metastasis	16	Female	Lung metastasis	Diagnosis	Unpublished model sourced from the PPTC
NCHEWS1	PDX	Metastasis	15	Male	Lung	Relapse	Rokita JL et al. Cell Rep 2019
SKNEP1	CDX	Metastasis	25	Female	Pleural effusion	Relapse	Rokita JL et al. Cell Rep 2019
TC32	CDX	Primary	17	Female	Bone, left ileum	Diagnosis	Zhang et al. Cancer Discovery 2021
TC71	CDX	Primary	22	Male	Humerus	Relapse	Rokita JL et al. Cell Rep 2019

Enrichment of membrane proteins in Ewing sarcoma xenograft models

Enrichment of plasma membrane proteins was carried out similarly to previous studies (13) with some modifications. Tumors of sufficient weight (range, 170–350 mg) were collected in 10-mL Homogenization Buffer [250 mmol/L Sucrose, 10 mmol/L HEPES, 1 mmol/L EDTA + protease inhibitor cocktail (cOmplete EDTA free (Roche), pH 7.4)] and dissociated using a handheld tissue shredder (QiaShredder, Qiagen) into a fully fluid mixture. Next, the samples were passed through an 18G needle 3 times, followed by a further 3 times with a 21G needle, and subject to low-speed centrifugation (1,000 x g, 15 minutes at 4°C) to remove any cellular debris. The supernatant was collected and layered atop a 60% (w/v) Sucrose cushion in a 13.2 mL ultracentrifuge-compatible tube (Beckman Coulter). The samples were centrifuged at 100,000 x g in a SW41 rotor (Beckman) for 1 hour at 4°C, resulting in a crude membrane fraction above the 60% sucrose cushion. The crude membrane fraction was collected, transferred to a fresh 13.2 mL ultracentrifuge tube, and 1.5 mL of sucrose solutions of differing percentages (42.8%, 42.3%, 41.8%, 41%, 39%, and 37%) were carefully layered above the crude membrane fraction. Tubes were filled and balanced with 37% sucrose and centrifuged in a SW41 rotor at 100,000 x g overnight (18 hours minimum) at 4°C. The following day, the floating membrane fraction was collected from the top of the sucrose gradient and transferred to a fresh 13.2 mL ultracentrifuge tube. The tube was filled with HEPES buffer (150 mmol/L NaCl, 20 mmol/L HEPES, 2.4 mmol/L K2HPO4, 1.2 mmol/L CaCl2, 1.2 mmol/L MgCl2, pH 7.4) and centrifuged in a SW41 rotor at 150,000 x g for 3–5 hours at 4°C. Once complete, the supernatant was discarded and the pellet containing enriched membrane proteins was dissolved in 50 mmol/L HEPES + 10% (w/v) Sodium deoxycholate. Enrichment of membrane proteins from the SM reference was carried out almost identically as mentioned above with the following deviations; A cell pellet comprising equal numbers of all cell lines detailed in Supplementary Table S1, totaling of 1 × 10⁷ cells, was used. Rather than being dissociated with a hand-held homogenizer, SM reference samples were homogenized using a handheld dounce homogenizer (50 strokes with a tight-fitting pestle; type-B). There were no other deviations from the published protocol.

Preparation of enriched membrane proteins for mass spectrometry analysis

Enriched membrane proteins were applied to a molecular weight cutoff (10 kDa) filter spin unit (Millipore) and centrifuged at 12,000 x g for 15 minutes at 4°C. The flow-through was discarded, HPLC-grade H2O was applied to the filter and centrifuged as per the parameters mentioned previously. This step was repeated for a total of 2 wash steps. Proteins remaining in the filter column were reduced for 1 hour at room temperature following the addition of 5 mmol/L DTT in 50 mmol/L HEPES. Reduced proteins were further alkylated with the addition of 20 mmol/L Iodoacetimide, left for 30 minutes at room temperature covered from light. The sample was centrifuged at 12,000 x g for 30 minutes at 4°C, the flow through discarded, and 5-μg trypsin/LysC (Promega) mix was added to the sample contained within the filter. Protein digestion was performed at room temperature for 18 hours.

Labeling of peptides using isobaric tandem mass tags for peptide quantification mass spectrometry and design of TMT plex

Tandem mass tags (TMT)-11 multiplexes were designed before labeling peptides to minimize the potential bias of running biological/technical replicates on the same plex. Each TMT-11 plex contained an “SM” channel, as well as a “Pooled Internal Standard” channel. The Pooled Internal Standard consisted of equal amounts of all samples in the cohort (except the SM samples) and was used to normalize signal between different plexes after mass spectrometry analysis. The TMT reagent (Thermo Fisher Scientific) was first reconstituted to a concentration of 10 μg/μL in acetonitrile (Thermo Fisher Scientific) and brought to room temperature. Peptides to be labeled were incubated with an excess of TMT reagent (10 μL of TMT label per sample), incubated at room temperature for 30 minutes, and then repeated for a total of 2 incubations. The labeling reaction was quenched with the addition of 10 mmol/L Glycine to each sample, at which point all samples were placed into a vacuum centrifuge until the sample volume reduced by at least half. Samples were then combined into a single TMT plex and desalted using the STop And Go Extraction (STAGE) tip protocol (19) with one deviation; because of the presence of Sodium Deoxycholate (SDC), samples required a centrifugation of 20,000 x g for 10 minutes following the addition of 0.1% Trifluoroacetic acid to remove the acid-labile SDC. The samples were dried fully in a vacuum centrifuge, and dried plexes were resuspended in 0.1% Formic Acid (Thermo Fisher Scientific) to be prepared for mass spectrometry analysis. A single deviation to the above methods was used for global proteome samples due to the increased concentration of sample. Briefly, following combination of samples into 1 single plex, the samples were desalted using a Sep-Pak C-18 cartridge (50 mg binding capacity; Waters) as per the manufacturer's instructions and dried fully in vacuum centrifuge.

Protein extraction for global proteome analysis

After PDX models were dissociated using a tissue homogenizer and passed through both 18G and 22G needles as described previously, a small amount of material was reserved for global proteome analysis. The fully fluid mixture was mixed with a detergent-based lysis buffer (500 mmol/L Tris-Cl; pH 8.0), 50 mmol/L NaCl, 40 mmol/L 2-Chloroacetamide, 10 mmol/L TCEP, 5 mmol/L EDTA, 2% (v/v) SDS, 1% (v/v) NP-40, 1% (v/v) Triton X-100) at a 1:1 (v/v) ratio. The extraction, reduction, and alkylation of proteins were performed in one step by incubation at 95°C for 90 minutes in a thermomixer with shaking at 1,000 RPM. Following extraction, the samples were cleaned as per the FASP protocol mentioned above to remove reagents that potentially interfere with accurate quantification. Following FASP, the samples were quantified using a BCA assay (Thermo Fisher Scientific) as per the manufacturer's instructions and 100 μg of material was selected for trypsinization.

Preparation of proteins for global proteome analysis

Further cleaning of proteins extracted for global proteome analysis was carried out using the Single-Pot, Solid Phase-enhanced Sample Preparation (SP3) protocol as described previously (20). Briefly, the carboxylate-functionalized magnetic beads (Cytiva) were washed in HPLC-grade H₂O before incubation with 100 μg of each sample in the presence of 100% EtOH (final = 50%) for 5 minutes at 24°C with mixing at 1,000 RPM. After binding, the magnetic beads + proteins were captured to the side of the tube using a magnet, and the supernatant containing contaminants was aspirated. The bead/protein mixture was washed 3 times with 80% EtOH, removing the supernatant between each wash after the beads were captured by said magnet. Finally, the bead/protein mixture was incubated with Trypsin/Lys-C at a 1:50 (trypsin:protein) ratio overnight at 37°C with agitation (1,000 RPM) in 50 mmol/L HEPES (pH 8.0). The next day, the beads:undigested protein mixture was captured to the side of the tub with a magnet and the digested peptides were collected with the supernatant. The resulting peptides were TMT-labeled as described above.

High-pH reverse phase chromatography and offline fractionation of whole-proteome peptides

Offline fractionation was carried out on the Agilent 1100 HPLC system facilitated by a Kintex 1.7 μm EVBO C18 100A column (30 cm × 2.1 mm). The dried global proteome fraction was resuspended in 200 μL 0.1% Formic Acid in HPLC-grade H2O, 100 μL was loaded to the HPLC, and the sample was separated at a flow rate of 250 μL per minute over a 60 minute gradient with varying concentrations of mobile phase A (10 mmol/L Ammonium Bicarbonate in HPLC-grade H2O) to mobile phase B (100% HPLC-grade Acetonitrile). Gradient parameters: 0–5 minutes (5% B), 5–8 minutes (5%–8% B), 8–35 minutes (8%–30% B), 35–45 minutes (30%–40% B), 45–50 minutes (40%–80% B), and 50–60 minutes (80% – 5% B). Fractions were collected sequentially (A1–12, B1–12, C1–12, and D1–12) at a volume of 200 μL per well in a 96-well plate for a total of 48 fractions, then concatenated to 12 fractions of 800 μL through combining A1-D1, A2-D2, A3-D3 etc. Concatenated fractions were dried fully in a vacuum centrifuge before being resuspended in 0.1% formic acid in HPLC-grade H2O. Global proteome samples were loaded for mass spectrometry simultaneously with surface enriched samples to minimize variation between runs.

Nano LC/MS-MS data acquisition

Analysis of labeled peptides was performed on an Orbitrap Eclipse Tribrid mass spectrometer (Thermo Fisher Scientific) equipped with an Easy-nLC1000 (Thermo Fisher Scientific) by nano-electrospray chromatography using acquisition parameters as reported previously (21), which are also provided in the PRIDE dataset PXD043375.

Bioinformatic processing of mass spectrometry data

Raw MS data were searched using Sequest HT algorithm through Proteome Discoverer suite (v2.4, RRID: SCR_014477), against a human reference (2021/07/16 Swissprot; 20,351 sequences). Precursor and fragment mass tolerance were set at 10 ppm and 0.05 Da, respectively. Dynamic modifications included Oxidation (+15.995 Da, M), Acetylation (+42.011 Da, N-Term). Static modification included Carbamidomethyl (+57.021 Da, C), and TMT (+229.163 Da, K, N-Term). Peptide-to-spectrum matches (PSM) were filtered using Percolator by searching the results against a decoy sequence set, only PSMs with FDR < 1% were retained in the downstream analysis. PSMs were further filtered out if they had a signal-to-ratio (S/N) lower than 10 in the PIS channel and if they mapped to more than one unique protein. To normalize input signal, channel total intensity was scaled to 1e08. Each S/N was normalized to PIS channel (ratio) and, for each peptide, the median ratio of the 3 PSMs with the lowest isolation interference was used. Peptides were then median-aggregated to the protein level. Proteins with fewer than two identified peptides were excluded from analysis in the global proteome data. This filtering step was not applied to the surface-enriched (surfaceome) data because of the lower coverage. To estimate protein relative abundance, the S/N sum of the 3 highest abundance PSMs was taken and the average across channel was multiplied to S/N protein ratios obtained previously. To filter out mouse peptide/proteins from PDX samples, MS-MS data were searched again against a combined human (2021/07/16 Swissprot; 20,351 sequences) + mouse (2020/01/20 Swissprot, 17,057 sequences) reference proteome. The human+mouse combined search was used to determine which human proteins were assigned to proteins groups where a mouse protein was identified as the master protein by parsimony analysis performed by the Proteome Discoverer software. Any proteins identified as emanating from a mouse master protein were subsequently removed from the human only search. In the global proteome, the original 6,975 human proteins were filtered (removing mouse assigned master proteins) to 6,867, and 79,350 peptides were filtered down to 78,820. In the surfaceome, 3,394 human proteins were filtered to 3,357, and 17,194 peptides were filtered to 17,105, using the same analysis protocol. Differentially expressed proteins in Ewing sarcoma versus SM were calculated by the Differential Expression analysis of quantitative Mass Spectrometry data (DEqMS) package (22). Statistical analysis and data visualization were carried out in R studio.

Gene ontology analysis

Gene ontology (GO) analysis was carried out using the GO online platform as described previously (23). For comparisons between surfaceome and global proteome data, the global proteome data were set as the reference dataset rather than the entire background. All terms with an FDR-corrected P value of <0.05 were deemed significant.

IHC staining of TMAs

All IHC for ENPP1 were carried out using standard protocols on formalin-fixed, paraffin-embedded TMA sections. Once sections were deparaffinized and rehydrated, antigen retrieval was carried out by incubating the sections in Tris EDTA buffer (CC1 standard) for 1 hour at 95°C. The resulting TMA sections were incubated for 1 hour (Ventana Discovery platform) in the ENPP1 primary antibody (ab223268, Abcam) according to the manufacturer's guidelines. Next, primary antibody-bound tissue sections were incubated with the appropriate secondary antibody (anti-Rabbit, 111–035–003, The Jackson Laboratory) for 1 hour followed by Ultramap HRP and Chromomap DAB detection. H-scores for each core were calculated by multiplying the ENPP1 intensity (0–3, where 0 denotes no staining, 1 denotes low staining, 2 denotes medium staining, and 3 denotes high staining) by the percentage of coverage of the core (0–100) for a total scale of 0–300. H-scores were determined by an experienced pathologist (A. Delaidelli). Where multiple cores existed for the same sample/patient, the average score was taken between them.

Data availability

All raw mass spectrometry data are deposited in the PRoteomics IDEntifications (PRIDE) database (RRID: SCR_003411), a public data repository of mass spectrometry–based proteomics data, accession: PXD043375. For mining of processed proteomics data, see Supplementary Data S1–S4:

Supplementary Data S1: Protein Log₂ FC Ewing sarcoma versus SM for both surfaceome and global proteome (merged).

Supplementary Data S2: Statistical data (DEqMS) for both global proteome and surfaceome.

Supplementary Data S3: Scoring schema of 218 prioritized targets.

Supplementary Data S4: Relative abundance of proteins identified in both surfaceome and global proteome data.

Results

Surface protein enrichment strategy for Ewing sarcoma xenograft models

A comprehensive proteomic evaluation of the Ewing sarcoma surfaceome has yet to be reported. We therefore performed enrichment of plasma membrane proteins from a series of 19 Ewing sarcoma xenograft models using a density-gradient ultracentrifugation approach, based on a previously described method (ref. 13; Fig. 1A). This technique allows for the isolation of membrane-bound proteins from material that had been previously been frozen or sub-optimally preserved. In parallel, we performed global proteome analysis on the same samples as a comparator for the surfaceome protocol. As indicated in Table 1, the 19 Ewing sarcoma xenograft models were mainly sourced from the PPTC (14). Of these models, 13 were CDXs, and 6 were xenografts derived from patients (PDXs). Each sample was analyzed in duplicate, where each replicate was an independent section of a PDX or CDX tumor. For 8 models (Table 1), separate samples were not available and thus were analyzed as technical duplicates. There was a modest gender bias within our samples, with 11 and 8 models derived from male or female patients, respectively, reflecting how the disease presents in the clinic (1). The average age of origin patients was 17.2 years (range, 10–45 years, standard deviation 7.6 years), again reflecting the average age of diagnosis (ref. 1; Table 1). The majority of xenografts (14) were derived from primary tumors, whereas 5 were derived from metastatic tumors.

Figure 1.

Surface enrichment methodology and Ewing sarcoma surfaceome and proteome metrics. A, Membranes were enriched from Ewing sarcoma models using a density gradient ultracentrifugation approach. Global proteome profiling was carried out in tandem according to the schematic. B, Upset plots depicting the number of proteins identified for the global proteome (Global) or surface surfaceome enrichment (Surfaceome). Proteins were classified as surface proteins using the SurfaceGenie dataset. Numbers listed on the right side are the total protein count for that particular group. C, Principal component analysis (PCA) plots of Ewing sarcoma samples, and SuperMix (SM) samples, in both the surfaceome (blue/gray dots) and the global proteome (red/gray dots) datasets.

As there is no standardized approach for surface protein annotation, we compared three methods to identify surface proteins in surfaceome data; SurfaceGenie (24), GO terms used by Glisovic-Aplenc and colleagues (13), and terms from the COMPARTMENTS database recently described by Bosse and colleagues (25). Similar enrichment scores were demonstrated between the SurfaceGenie annotation and that of the COMPARTMENTS annotation of Bosse and colleagues (Supplementary Table S2). An increase in the overall surface enrichment was highlighted using the GO terms of Glisovic-Aplenc and colleagues, likely attributed to the larger number of potential surface proteins extracted from the specific GO terms (Supplementary Table S2). We chose the SurfaceGenie to use for our studies as it is a manually curated dataset from 4 independent datasets, allowing for higher confidence in identifying surface proteins. Furthermore, SurfaceGenie uses a Surface Prediction Consensus (SPC) score-to-score proteins from 1 to 4, based on how many of the 4 datasets a protein of interest is found in (24), allowing for examining the quality of the data.

A total of 3,357 proteins were identified in the surface-enriched (surfaceome) samples, of which 1,131 were classified as being surface proteins (i.e., 34% are surfaceome candidates; Fig. 1B). In contrast, global proteome analysis identified 6,867 proteins, of which 1,030 were characterized as surface proteins (i.e., 15% of the global proteome; Fig. 1B). Mouse-derived proteins were excluded (see Materials and Methods). Offline fractionation of global proteome samples allowed for deeper proteome coverage, as demonstrated by a higher overall number of proteins (Fig. 1B) and peptides observed (Supplementary Fig. S1A and S1B). Of the 1,364 surface proteins identified across both datasets, 797 were common to both datasets, with 334 identified in surface-enriched data only, and 233 identified in global proteome data only (Fig. 1B). To aid in highlighting Ewing sarcoma–specific surface proteins, a mixture of 12 common cell lines (SM) representing human proteins in 12 different tissue types was prepared and treated similarly to the Ewing sarcoma xenografts. Principal Component Analysis highlighted that Ewing sarcoma samples clustered with other Ewing sarcoma samples, and separated from SM samples (Fig. 1C). Hierarchical clustering of surface proteins in the surfaceome data revealed replicates were well correlated, and samples grouped independently from the SM reference (Supplementary Fig. S2A), though some variation could likely be attributed to the decreased complexity of samples. Similar data were obtained in the global proteome data when analysis was performed across all proteins (Supplementary Fig. S2B). Consensus clustering (26) of surfaceome and global proteome proteins highlighted some sub-clustering in the data, particularly in the global proteome data (Fig. 1C; Supplementary Fig. S2B), possibly driven by whether samples represented CDXs or PDXs (Supplementary Fig. S3A and S3B). Samples did not cluster based on whether they were of primary or metastatic origin, site of disease, or any other metric available (Supplementary Fig. S3A and S3B). These data highlight the successful capture of surface proteins using both methods, although the surfaceome and global proteome data reveal some differences in surface annotated Ewing sarcoma proteins.

Correlation between surface proteins identified in Ewing sarcoma surfaceomes and global proteomes

We evaluated SPC scores of surface proteins in both the surfaceome and global proteome data (Fig. 2A). This revealed that most surface proteins identified in the global proteome-only set were lower annotated (average SPC score 1.4), whereas surface proteins identified in the surfaceome-only set were higher scoring (average SPC score 2.5), and proteins identified by both methods had intermediate scores (average SPC score 1.9). GO analysis identified over- and under-represented cellular component terms in the surfaceome data compared with global proteomic data (Fig. 2B), revealing an enrichment of “membrane” and “plasma membrane” associated terms in the Ewing sarcoma surfaceome data. Moreover, cellular component terms such as “nucleus” and “intracellular” were significantly under-represented in the surfaceome versus the global proteome data, confirming the successful depletion of intracellular components and enrichment for membrane-associated proteins (Fig. 2B). These data verify, at least in part, our surfaceome method for enrichment of membrane-associated proteins.

Figure 2.

Surfaceome and Global proteome profiling provides a comprehensive depiction of the Ewing sarcoma surfaceome. A, Surface Protein Consensus (SPC) scores from the SurfaceGenie dataset of surface proteins found in both the Global and Surfaceome datasets, and in each dataset alone. B, Gene ontology (GO) analysis of the Ewing sarcoma surfaceome data versus the global proteome data. Over- and under-represented GO terms are shown. C, Log₂ fold change (Ewing sarcoma vs. SM) of proteins common to surfaceome (x-axis) and global proteome (y-axis) datasets are well correlated (Pearson R = 0.52, P < 0.001). Highlighted are known Ewing sarcoma–associated surface proteins. D and E, Volcano plots (log₂ fold change vs. –log₁₀ adjusted P value) of all proteins enriched in Ewing sarcoma models in both the surfaceome (D) and global proteome (E) data. Bold labeled proteins are surface proteins with –Log₁₀ adjusted P < 0.05 and log₂FC > 0.5. Statistical analysis was performed using the DEqMS test. Limits were applied to the x and y axes in both D and E for visualization purposes (there are 10 significant, non-surface proteins not shown in E). Proteins enriched in the SM samples (left-hand side of the plots) were not displayed.

The fold change of proteins in Ewing sarcoma versus SM for the surfaceome and global data was strongly correlated (Fig. 2C; Pearson R, 0.52; P < 0.0001; Supplementary Data S1). Correlations between both datasets were slightly higher when only surface proteins were considered (Pearson R, 0.59; P < 0.0001; Supplementary Fig. S4A), and slightly reduced for only non-surface proteins (Pearson R, 0.46; P < 0.0001; Supplementary Fig. S4B). Known Ewing sarcoma–associated proteins were enriched in both surfaceome and global proteome datasets, including CD99 (4), IL1RAP (17), and ADGRG2 (10), among others (Fig. 2C). As examples of Ewing sarcoma–specific enrichment in independent datasets, we investigated the protein expression of Adhesion G Protein-Coupled Receptor G2 (ADGRG2), CD99, and IL1RAP in the published Cell Models Passport dataset (ref. 27; Supplementary Fig. S5A), and their mRNA expression in Ewing sarcoma versus other pediatric malignancies in the PPTC dataset (ref. 14; Supplementary Fig. S5B), and all three targets were confirmed as being highly specific for Ewing sarcoma. This demonstrates a good concordance between the surfaceome and global proteome datasets, highlights excellent enrichment of membrane-associated terms in the surfaceome data, and demonstrates that Ewing sarcoma–associated proteins can be detected as enriched in our data versus the SM reference dataset.

Analysis of Ewing sarcoma surface proteome in PDX and CDX models

The log₂-fold change of proteins in Ewing sarcoma models compared with the SM reference were plotted against their respective −log₁₀ adjusted P values for both the Ewing sarcoma surfaceome (Fig. 2D) and global proteome datasets (Fig. 2E; Supplementary Data S1 and S2). Of the 3,357 proteins identified in the surfaceome data, 194 were significantly enriched in Ewing sarcoma models (FDR P < 0.05) with 49 classified as “Surface” and the other 145 classified as “Non-surface.” Of the 6,867 proteins identified in the global proteome data, 1,349 were significantly enriched in Ewing sarcoma models versus SM (FDR P < 0.05), with 120 classified as “Surface” and the remaining 1,229 classified as “Non-surface” (Fig. 2D and E). This demonstrated that approximately 25% of all significantly enriched proteins in the surfaceome set were surface proteins, whereas only 9% of all significantly enriched proteins in the global proteome were surface proteins. Proteins labeled in Fig. 2D and E are surface proteins with an FDR-corrected P value of <0.05 and log₂ fold change (Ewing sarcoma vs. SM) >0.5. We confirmed enrichment of several membrane proteins known to be associated with Ewing sarcoma, including IL1RAP, ADGRG2, NGFR, and CD99. These data highlight the identification of known Ewing sarcoma–specific surface proteins in our data compared with the SM reference, further validating our protocol.

The Ewing sarcoma–associated protein, STEAP1 (28), was identified in these data; however, it was not strongly associated with Ewing sarcoma when compared with SM. This highlights a potential caveat in our approach, whereby some proteins that also have enhanced expression in other cancers (STEAP1 is expressed in prostate and ovarian cancers, among others; ref. 28) may not be highlighted when compared with the SM reference. Notably, although STEAP1 was not highly enriched in Ewing sarcoma versus SM, the closely related family member, STEAP2, was among the top hits in the surfacecome data (Fig. 2D). We also identified proteins with few or no previous associations to Ewing sarcoma, such as ATP11C and SLCO5A1 (Fig. 2D), confirming the effectiveness of the SM comparison. Notably, both ATP11C and SLCO5A1 have been identified as being potentially regulated by EWS–FLI1 (29), but little is known about their functions in Ewing sarcoma. ATP11C, SLCO5A1, and STEAP2 demonstrate robust Ewing sarcoma–specific expression in our proteomics data (Fig. 2D), as well as in publicly available datasets (PPTC; ref. 14); the Cancer Cell Line Encyclopedia (CCLE; ref. 30); and Cell Model Passport (ref. 27; Supplementary Fig. S6). These data present the first proteomics snapshot of the Ewing sarcoma surface proteome, with many known Ewing sarcoma proteins identified alongside several lesser-known proteins worthy of further investigation.

Prioritizing surface proteins as immunotherapy targets in Ewing sarcoma

Data presented in Fig. 2C–E highlight known Ewing sarcoma surface proteins (e.g., IL1RAP, CD99, and ADGRG2), and many newly identified proteins (e.g., ATP11C and STEAP2) that were enriched in our data. However, except for several known targets, whether such candidates might be suitable IT targets is difficult to ascertain on the basis of these data alone. For example, proteins lacking transmembrane domains might be “surface annotated” in the literature because they reside on the internal side of the plasma membrane, but would not be suitable IT targets. Furthermore, surface proteins ubiquitously expressed across healthy tissues are likely not suitable IT targets. We sought to create a scoring system to robustly prioritize proteins as candidate IT targets. An initial prioritization list was generated, that included proteins classified as surface proteins by SurfaceGenie (SPC ≥ 1), and contained transmembrane domains (per UniProt). With these criteria, a total of 81 and 169 proteins were enriched (FDR P < 0.1) in Ewing sarcoma models versus SM in the surfaceome and global proteome sets, respectively (Fig. 3A). Overlapping these lists resulted in 218 proteins, as 38 proteins were common between both datasets (Supplementary Data S3). A scoring schema was generated for this list, whereby a protein was individually scored, as described below, from 218 to 1 (i.e., the total number of proteins in the list) across four different datasets, namely our surfaceome (SFX) and global proteome (GLB) datasets, and two public gene expression datasets (14, 31), and where 218 is the best score in each dataset. Briefly, proteins scored higher based on the number of models and their fold change enrichment in Ewing sarcoma versus SM from our SFX and GLB datasets, higher mRNA fold change in Ewing sarcoma models versus other pediatric cancers in the PPTC dataset (14), and lower mRNA expression (transcripts per million, TPM) in normal tissue from GTEX datasets (ref. 31; Fig. 3A and B). The scores from each dataset were then summed, and all 218 proteins were ranked on the basis of this set of criteria (Fig. 3B). The summed score values were converted to Z-scores and plotted, highlighting a more-or-less even distribution of proteins across four groups based on Z scores (Groups 1–4; Fig. 3C; Supplementary Fig. S7). Using this method, higher scoring proteins satisfy multiple criteria of a candidate Ewing sarcoma–specific IT target (Supplementary Fig. S7). Supporting the validity of our approach, the top hit was IL1RAP (scoring 203, 195, 209, and 206 in the SFX, GLB, PPTC, and GTEX datasets, respectively; total 813/872), ranking first among the 218 proteins (Fig. 3B–D; Supplementary Data S3). Notably, IL1RAP is in preclinical development as a potential IT target in Ewing sarcoma (17). Additional preclinical immunotherapy candidates in Group 1 (Z-scores > 1) include ADGRG2 (formerly known as GPR64; ref. 10) and LINGO1 (ref. 32; Fig. 3C). Other Group 1 proteins include CDH11, ENPP1, SLCO5A1, STEAP1, and STEAP2 (Fig. 3B–D). The Ewing sarcoma surface marker CD99 scored moderately well (Group 2, Z-score 0.7, Rank 54/218), although its overall score was reduced due to its relatively high expression in normal tissues (139 TPM; range, 0.5–374; Supplementary Data S3; Fig. 3C and D). Several non Ewing sarcoma–specific, but promising surface protein targets were also present in the scoring schema such as ROR1 (Group 2, Z-score 0.68, Rank 57/218; Supplementary Data S3), a surface protein IT candidate for some leukemias and lymphomas (33). This highlights that promising IT targets may still be dispersed among the 4 groups, but higher scoring targets are more likely to be Ewing sarcoma–specific. Using this scoring scheme, the 39 surface proteins prioritized in Group 1 are presented in Fig. 3D (Z-scores > 1), which shows, (i) the number of Ewing sarcoma models enriched versus SM reference in both the SFX and GLB data (with corresponding log₂ fold change values); (ii) the FPKM fold change in Ewing sarcoma models versus other pediatric cancers from the PPTC dataset (14), (iii) the average TPM across normal tissues in the GTEX data (31); (iv) the SPC score (24); (v) their potential regulation by EWSR1–ETS fusions from publicly available sources [ChIP and mRNA (refs. 34, 35) or protein (ref. 36)]; and (vi) any known biochemical features of the protein.

Figure 3.

Prioritization of immunotherapy candidates from the Ewing sarcoma surfaceome and global proteome data. A, The union of the surfaceome and global proteome data provided 218 surface protein immunotherapy candidates containing a transmembrane domain and an FDR < 0.1. The workflow depicts the general scoring schema used to generate the ranked list of immunotherapy candidates. SFX, surfaceome; GLB, global proteome. B, An example of how the scoring system works, highlighting IL1RAP, ENPP1, and ENG in these data. C, Summed scores from B were converted to Z-scores and plotted against their respective ranks. Highlighted are known Ewing sarcoma surface candidates, or promising candidates from this study. D, All data are available for proteins in Group 1 (Z-score >1). Tracks used in the scoring schema, top to bottom: SFX and GLB, # Ewing sarcoma models enriched and average fold change versus SM; PPTC_FC, fold change of targets in Ewing sarcoma versus other pediatric cancers in the PPTC dataset; and GTEX_TPM track, TPM of targets across normal tissues in the GTEX data. Other tracks are SurfaceGenie, SPC score (1–4) of protein; EWS–FLI1 target (ChIP + mRNA regulation), depicting if targets displayed EWS–FLI1 binding at their promotor regions (ChIP) and a subsequent increase in mRNA expression from publicly available data; and N cell lines (Protein WT > EWSR1–ETS KD), number of WT Ewing sarcoma cell lines (19 total) where that protein is significantly higher in versus EWSR1–ETS fusion knockdown (KD) models, depicting surface proteins potentially regulated by EWSR1–ETS fusions.

Among the 39 prioritized surface proteins in Group 1 are a number of proteins not previously considered as candidate IT targets in Ewing sarcoma, which have low expression in normal tissue, are potentially regulated by EWSR1–ETS fusions, and some have known associations with metastasis. Using these criteria and purported functions, we selected Ectonucleotide pyrophosphatase/phosphodiesterase 1 (ENPP1) and Cadherin 11 (CDH11) for further analysis (Fig. 3C and D). ENPP1 (Group 1, Z-score 1.6, Rank 18/218) has not been extensively studied in Ewing sarcoma (37), but was recently shown to have potential roles in cancer metastasis and immune evasion (38), highlighting this protein as a promising lead. CDH11 (Group 1, Z-score 1.5, Rank 21/218; Fig. 3C and D) has been previously linked to Ewing sarcoma metastasis (39). Robust Ewing sarcoma expression at the protein and RNA level was demonstrated for ENPP1 (Fig. 4A) and CDH11 (Supplementary Fig. S8) in three independent public datasets (14, 27, 30). Our integrated prioritization schema therefore provides a strategy to select candidate surface proteins for further evaluation as potential IT targets in Ewing sarcoma.

Figure 4.

ENPP1 is a Ewing sarcoma surface protein with robust expression in Ewing sarcoma compared with other childhood sarcomas and normal pediatric tissues. A, ENPP1 is highly expressed in Ewing sarcoma models from publicly available sources. Points displayed on RNA graphs are limited to the 99th percentile to improve data visualization. P values are from the Wilcoxon signed rank test. B, Western blot analysis depicting elevated expression of ENPP1 in Ewing sarcoma cell lines versus negative control lines (MSC, mesenchymal stem cell; OS, osteosarcoma). Tubulin was used as a loading control. C, ENPP1 is highly expressed in the plasma membrane fractions of three Ewing sarcoma cell lines (A673, CHLA10, and TC32). SiHa, a uterine squamous cell carcinoma cell line, and U2OS, an osteosarcoma cell line, were used as negative controls. ATP1A1 was used as a plasma membrane marker expected to be expressed in all cell lines. Vinculin was used to demonstrate equal loading and successful depletion of the cytosolic fraction. D, Immunofluorescence demonstrating surface staining of ENPP1 in Ewing sarcoma cell lines, with limited expression displayed in non-Ewing sarcoma lines. Tubulin was used as a cytoskeleton marker for all lines. DAPI was used as a nuclear counterstain. E, Representative IHC staining of ENPP1 in a childhood sarcoma cohort, demonstrating variable Ewing sarcoma staining. ARMS, alveolar rhabdomyosarcoma; ERMS, embryonal rhabdomyosarcoma; OS, osteosarcoma; UDS, undifferentiated sarcoma. F, Representative IHC staining of ENPP1 in a tissue microarray (TMA) of 19 Ewing sarcoma tumors, 57 other pediatric cancers, and 21 normal pediatric tissues. G, Plotted are H-scores derived from multiplying the ENPP1 intensity value (0–3) by the percentage of coverage (0–100) in the childhood sarcoma and normal pediatric tissue cohorts. For samples where multiple cores were available, the average of these values was used. Samples are ranked on ascending average H-score. A P value was derived from ANOVA.

ENPP1 and CDH11 are newly identified IT candidates in Ewing sarcoma

Because ENPP1 and CDH11 scored strongly in our Ewing sarcoma surface target scoring schema, and each has been linked to Ewing sarcoma, albeit in limited studies (37, 39), we performed additional preliminary validation of these two proteins. ENPP1 has been reported as a driver of metastasis and immune evasion in other cancers (38), but its role in either process is poorly understood. CDH11 has been linked to a potential role in Ewing sarcoma metastasis (39). Excellent peptide coverage and ample extracellular amino acid exposure was demonstrated for both ENPP1 (826 residues) and CDH11 (563 residues; Supplementary Fig. S9A and S9B; ref. 40), an important feature for designing antibody-based therapeutics. We examined ENPP1 and CDH11 expressions in the Ewing sarcoma models in an effort to understand whether either target was expressed at higher or lower levels in samples of primary or metastatic origin. ENPP1 showed a trend toward lower expression in metastatic versus primary samples, and CDH11 demonstrated the opposite, although neither of these correlations was statistically significant (Supplementary Fig. S10A). We investigated the dependency scores for ENPP1 and CDH11 in Ewing sarcoma cells using the DepMap portal (41). High dependency scores were observed for both FLI1 and EWSR1 (Supplementary Fig. S11). Surprisingly, however, both CDH11 and ENPP1 were also found to have high dependency scores in Ewing sarcoma, with CDH11 scoring almost as strongly as EWSR1 (Supplementary Fig. S11). Western blot analysis confirmed robust expression for ENPP1 and CDH11 across a panel of Ewing sarcoma cell lines (Fig. 4B; Supplementary Fig. S12A). ENPP1 was enriched in membranes using a cell compartment fractionation assay (Fig. 4C), and ENPP1 and CDH11 were densely represented at the cell surface as assessed by immunofluorescence (Fig. 4D; Supplementary Fig. S12B and S13). Owing to the poor performance of commercially available anti-CDH11 antibodies in IHC, we were limited to investigating ENPP1 expression by IHC in TMAs of different pediatric solid tumors. This demonstrated robust staining of ENPP1 in Ewing sarcoma compared with other childhood sarcomas (Fig. 4E), though some staining in a subset of patients with osteosarcoma may warrant further investigation. We also examined the expression of ENPP1 in TMAs of normal pediatric tissues compared with Ewing sarcoma (Fig. 4F), which demonstrated limited staining in several normal tissues (Fig. 4F). Low to moderate staining was detected in normal pancreas and liver (Fig. 4F), which while not precluding ENPP1 as an IT target, should be considered before clinical studies. These data were quantified, highlighting variable but strong expression of ENPP1 in Ewing sarcoma (average H-score of 80/300) versus other childhood sarcomas and normal pediatric tissues (average H-score 18/300; Fig. 4G). To address the variable staining, we examined the disease status of each patient and found that 9/19 samples were derived from primary disease, and 9 were from metastatic disease (1 unknown). Notably, ENPP1 demonstrated staining in 6 of the 9 primary cores (average H-score 134/300 across the 9 cores), whereas lower ENPP1 staining was only demonstrated in 2/9 of the metastatic cores (average H-score 36/300 across the 9 cores, including negative cores; Supplementary Fig. S14). These results demonstrate the utility of surface-based proteomics as a first step in identifying novel surface proteins with therapeutic potential in Ewing sarcoma, given the robust expression of both CDH11 and ENPP1 across Ewing sarcoma models, but with minimal expression in normal tissues.

Discussion

This study provides a comprehensive cataloguing of the Ewing sarcoma surfaceome and complementary proteome as a resource for the Ewing sarcoma research community. Identifying robust and specific surface proteins is critical for developing immunotherapies for pediatric malignancies, highlighting the need for a comprehensive cataloguing of disease-specific surfaceomes. Beyond the detection and quantification of surface proteins in discovery cohorts, subsequent validation of candidate IT targets for further development requires the demonstration of disease specific expression and low normal tissue expression in larger independent cohorts. Validation studies generally rely on antibody-based studies using TMAs; however, IHC grade antibodies are often not available. Thus, to minimize costly and labor-intensive validation studies using antibodies that may or may not be suitable to rule out normal tissue expression, target identification should effectively prioritize candidates more likely to succeed for the focused validation efforts. These considerations informed our choice of method and analysis, as well as the prioritization of lead surface targets in Ewing sarcoma.

Our Ewing sarcoma study used 19 xenograft models analyzed as duplicates to robustly characterize the Ewing sarcoma surfaceome. From a proteomics perspective, there are several methods for membrane protein identification and quantification, though most methods rely on viable material (i.e., intact live cells) for processing, limiting applicability to fresh material or cultured cells. We therefore used sucrose gradient density ultracentrifugation (13) to enrich for membrane fragments followed by MS analysis. We used an SM reference standard of 12 common cell lines as means of both identifying Ewing sarcoma–specific surface proteins and as an initial means of identifying surface proteins broadly expressed in many diverse tissues to down select, with the rationale that such proteins are more likely to be expressed in normal tissues. Notably, the CCLE also uses mixture of 11 cell lines as a reference sample and for quantifying relative expression (42). In parallel, we also used global proteome analysis and surface protein annotation to provide additional coverage of surface proteins. Global proteomes also allow for overall proteome variation and associated biological phenotypes to be addressed within the study cohort.

To aid in down-selection of IT candidates from our surfaceome study, we developed a prioritization schema. The schema, in addition to incorporating SurfaceGenie SPC scores and the presence of transmembrane domains, takes into account the number of PDX/CDX models and expression change for each surface protein of interest versus the SM reference in both the surfaceome and global proteome data, and also incorporates public PPTC and GTEX mRNA expression data to score proteins with high Ewing sarcoma expression versus other pediatric cancers, and low normal tissue expression, respectively. Our final schema thus provides prioritized surface protein candidates commonly expressed across multiple Ewing sarcoma PDX/CDX models, which are likely Ewing sarcoma specific and with low normal tissue expression, thus enabling focused antibody-based validation studies with increased likelihood of success. The prioritization generated 39 candidates in Group 1 (Z-scores ≥ 1) and 56 candidates in Group 2 (Z-scores > 0; Fig. 3C and D; Supplementary Data S3). The validity of this approach was demonstrated by the identification of several known Ewing sarcoma surface targets, including IL1RAP, CD99, ADGRG2, LINGO1, and STEAP1. We previously identified IL1RAP as an Ewing sarcoma surface protein in a screen for suppressors of anoikis, and determined that IL1RAP is a key regular of cysteine and glutathione metabolism in Ewing sarcoma cells (17). CD99 is a cell surface glycoprotein and is well established as a diagnostic IHC biomarker for Ewing sarcoma (4). ADGRG2 is a multi-pass G-protein–coupled receptor with poorly understood functions, though has been proposed as a potential therapeutic target for Ewing sarcoma (10). LINGO1 is part of the Nogo receptor signaling complex that inhibits oligodendrocyte differentiation, axonal regeneration, and myelin production, and has been proposed as a candidate Ewing sarcoma immunotherapy target (32). STEAP1 is a Fe³⁺/Cu²⁺ metalloreductase enzyme with normal tissue expression mainly limited to prostate (28), but has previously been detected in Ewing sarcoma, with transgenic CD4⁺ T cells targeting STEAP1 being tested in Ewing sarcoma (43). Notably, our schema prioritized STEAP2 more favorably than STEAP1. STEAP2 is also a Fe³⁺/Cu²⁺ metalloreductase, though is not as well studied as STEAP1.

In addition to STEAP2, our results identify several other surface proteins as candidate Ewing sarcoma immunotherapy targets but with little or no prior association to Ewing sarcoma in the literature, including ATP11C, SLCO5A1, ROR1, CDH11, and ENPP1. ATP11C is a phospholipid flippase enzyme, flipping the orientation of phospholipids in the plasma membrane from the outer to the inner membrane in an ATP-dependent manner (44). It is involved in B-lymphocyte differentiation (44), but its role in cancer is poorly understood. SLCO5A1, a solute carrier anion transporter, has been linked to breast cancer development (45), but has not been well studied in Ewing sarcoma. ROR1 is a tyrosine kinase–like orphan receptor that plays a role in embryogenesis (33) and is under investigation as an ADC target in lymphoid cancers (46).

We chose to perform additional preliminary validation experiments for two of the newly identified candidate surface targets, namely CDH11 and ENPP1. CDH11 (otherwise known as osteoblast–cadherin) was originally identified in mouse osteoblasts, and regulates cell adhesion in a calcium-dependent manner. It has been described in Ewing sarcoma (39), and has been proposed for antibody-based therapy in other cancers (47). ENPP1 is a purine ecto-nucleotide pyrophosphatase that hydrolyzes extracellular cyclic dinucleotides (cGAMP, c-di-AMP, c-di-GMP) to generate PPi to regulate bone mineralization (48). ENPP1-mediated degradation of the immune-stimulatory metabolite, cGAMP, has been shown to promote metastasis of triple-negative breast cancer (38). ENPP1 loss suppressed metastasis and restored tumor immune infiltration in a cGAS STING–dependent manner (38), and high ENPP1 expression correlated with resistance to anti PD-1/PD-L1 treatment (38). Western blot and immunofluorescence experiments showed that both ENPP1 and CDH11 are enriched in Ewing sarcoma cell lines and demonstrated plasma membrane staining, which was confirmed by IHC for ENPP1. Although available commercial antibodies for CDH11 were suboptimal for IHC, we performed ENPP1 IHC analysis of in-house pediatric cancer and normal tissue TMAs containing 21 normal pediatric tissues, 19 Ewing sarcoma, and 57 other pediatric cancer samples (17). The data generated suggest that ENPP1 is mainly expressed in primary Ewing sarcoma, with lower expression in metastatic disease, which may explain the variable staining pattern detected in the Ewing sarcoma TMA, although expanding this analysis to larger cohorts would undoubtedly assist in providing more statistical power to this observation. The diffuse staining for ENPP1 at the plasma membrane as well as extracellularly, is consistent with reports that ENPP1 is expressed at the plasma membrane, but can also be enzymatically cleaved from the outer membrane (48). These points should be considered when developing ENPP1 as an IT target, as has been discussed by others (48). Given the reported role for ENPP1 in metastasis of other cancers (38), our data raise the possibility that ENPP1 has other functions in Ewing sarcoma, which requires more detailed mechanistic analysis. Expression of ENPP1 in Ewing sarcoma but with low expression in normal pediatric tissues nominates this protein as a potential new IT target for Ewing sarcoma, prompting us to generate new humanized anti-ENPP1 antibodies for cancer therapies (49).

There are limitations to this work. For example, expression of each surface protein was determined relative to the SM reference. Thus, proteins that are highly expressed in Ewing sarcoma but are also high in other cancers may not be enriched versus the SM reference. For instance, STEAP1, which is a known Ewing sarcoma–associated protein (28), was only weakly enriched in these data. This is likely a consequence of STEAP1 being enriched in some cancers such as prostate and ovarian cancer (28), both of which are represented in the SM control. The scoring schema for target prioritization was designed to address this limitation, whereby proteins that did not display robust Ewing sarcoma–specific expression in the proteomics data could still be prioritized on the basis of their expression in normal tissues (31) or other Ewing sarcoma databases (14). Confirming the validity of this approach, proteins like STEAP1 and LINGO1, which were only modest surface hits in the proteomics data, were identified as top hits (Group 1, Z-scores ≥1) in the prioritization schema. As a consequence, manual inspection of our surfaceome and global proteome data for integration with external gene expression data is important, an exercise that is necessary for any surfaceome workflow before validation studies. Selection of appropriate negative controls is critical for the success of projects of this nature, and this should be a priority when designing surfaceome workflows. Nevertheless, using the SM reference assisted in filtering out surface proteins ubiquitously expressed across non-Ewing sarcoma tumor types, thus highlighting Ewing sarcoma–specific candidates. The prioritization schema also considered the mRNA expression of targets in Ewing sarcoma and normal tissues, resulting in prioritization of high-confidence Ewing sarcoma surface targets and reducing false positives, which should aid in down selecting targets for subsequent IT studies. Finally, target selection, prioritization, and validation can be highly dependent on the availability of high-quality commercially available reagents, which can be challenging to source when prioritizing novel, often understudied, proteins.

Conclusions

Our Ewing sarcoma surfaceome data are a comprehensive resource for the scientific community to assist in target prioritization for the development of immunotherapeutic strategies in Ewing sarcoma. We highlight the utility of our data through the identification of proteins previously linked to Ewing sarcoma, as well as the identification of new candidates for lesser-studied proteins, including CDH11 and ENPP1. For example, ENPP1 displays strong expression in Ewing sarcoma but with limited normal tissue expression, and thus represents an immunotherapeutic candidate warranting further investigation in Ewing sarcoma.

Supplementary Material

Supplementary Table S1

SuperMix cell lines

Supplementary Table S2

Surface protein annotation based on COMPARTMENTS and GO

Supplementary Figure S1

Peptide coverage of all proteins and surface proteins in the global and surfaceome approaches.

Supplementary Figure S2

Sample clustering of surface proteome data.

Supplementary Figure S3

Consensus clustering of proteomics data.

Supplementary Figure S4

Correlation between surfaceome and global proteome datasets.

Supplementary Figure S5

Expression of known EwS surface proteins in cell line models and publicly available data.

Supplementary Figure S6

Expression of ATP11C, SLCO5A1, and STEAP2 in publicly available datasets.

Supplementary Figure S7

Characteristics of proteins in the scoring schema.

Supplementary Figure S8

CDH11 expression in three publicly available datasets.

Supplementary Figure S9

Domain structure of ENPP1 and CDH11.

Supplementary Figure S10

ENPP1 and CDH11 expression in xenograft models of EwS.

Supplementary Figure S11

ENPP1 and CDH11 have high EwS dependency.

Supplementary Figure S12

CDH11 is a surface protein in EwS

Supplementary Figure S13

ENPP1 immunofluorescence.

Supplementary Figure S14

ENPP1 is expressed at higher levels in non-metastatic EwS vs metastatic EwS.

Supplementary Data S1

Log2 fold change values for proteins in EwS xenografts vs SuperMix.

Supplementary Data S2

Statistical analysis (DEqMS) of proteins, EwS vs SuperMix

Supplementary Data S3

Scoring schema to prioritize candidate immunotherapy targets.

Supplementary Data S4

Relative abundance values for proteins identified by proteomics.

Authors' Disclosures

B. Mooney reports grants from Michael Smith Health Research BC during the conduct of the study. S.J. Diskin reports grants from NIH/NCI during the conduct of the study. G.B. Morin reports grants from NIH during the conduct of the study. P.H. Sorensen reports grants from National Institutes of Health (NIH) U54 Pediatric Immunotherapy Discovery and Development Network (1U54CA232568-01), Stand Up To Cancer/St. Baldrick's Foundation Pediatric Dream Team Translational Research Grant (SU2C-AACR-DT1113), Empowering Pediatric Immunotherapy for Childhood Cancers Team grant from the St. Baldrick's Foundation, and BC Cancer Foundation during the conduct of the study. No disclosures were reported by the other authors.

Authors' Contributions

B. Mooney: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. G.L. Negri: Conceptualization, data curation, software, formal analysis, visualization, writing–review and editing. T. Shyp: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–review and editing. A. Delaidelli: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–review and editing. H.-F. Zhang: Data curation, formal analysis, validation, visualization. S.E. Spencer Miko: Software, supervision, project administration. A.K. Weiner: Resources, methodology, project administration. A.B. Radaoui: Resources, methodology. R. Shraim: Resources, methodology. M.M. Lizardo: Resources. C.S. Hughes: Resources, methodology. A. Li: Resources. A.M. El-Naggar: Resources. M. Rouleau: Resources, project administration, writing–review and editing. W. Li: Resources. D.S. Dimitrov: Resources. R.T. Kurmasheva: Resources. P.J. Houghton: Resources. S.J. Diskin: Resources, supervision. J.M. Maris: Conceptualization, resources, supervision, funding acquisition, project administration. G.B. Morin: Conceptualization, resources, formal analysis, supervision, methodology, project administration, writing–review and editing. P.H. Sorensen: Conceptualization, resources, supervision, funding acquisition, project administration, writing–review and editing.