DLK1 distinguishes subsets of NF1-associated malignant peripheral nerve sheath tumors with divergent molecular signatures

Abstract

Purpose:

MPNST is the leading cause of premature death among individuals with NF1 and the transcriptional aberrations that precede malignant transformation and contribute to MPNST tumorigenesis remain poorly defined. Alterations involving CDKN2A and components of PRC2 have been implicated as early drivers of PNST evolution, but these events do not occur in all MPNST. Accordingly, emerging data has begun to highlight the importance of molecular-based stratification to improve outcomes in patients with NF1-PNST.

Experimental design:

Here we perform an integrated analysis of multiple, independent datasets obtained from human NF1 patients to gain critical insight into PNST evolution and MPNST heterogeneity.

Results:

We show that DLK1 is significantly increased in MPNST and provide evidence that DLK1 overexpression may precede histological changes consistent with malignancy. In complementary analyses, we find that serum levels of DLK1 are significantly higher in both mice and humans harboring MPNST compared to those without malignancy. Importantly, while DLK1 expression is increased in MPNST overall, through the integration of multiple, independent datasets we demonstrate that divergent levels of DLK1 expression distinguish MPNST subsets characterized by unique molecular programs and potential therapeutic vulnerabilities. Specifically, we show that overexpression of DLK1 is associated with the reactivation of embryonic signatures, an immunosuppressive microenvironment and a worse overall survival in patients with NF1-MPNST.

Conclusions:

Collectively, our findings provide critical insight into MPNST tumorigenesis and support prospective studies evaluating the utility of DLK1 tissue and serum levels in augmenting diagnosis, risk assessment and therapeutic stratification in the setting of NF1-PNST.

INTRODUCTION

Neurofibromatosis type 1 (NF1) is one of the most common cancer predisposition syndromes, affecting approximately 1:2500 individuals globally [1]. It is caused by pathogenic variants or deletions of the NF1 tumor suppressor gene resulting in Ras pathway hyperactivation [2, 3]. A hallmark feature of NF1 is the development of histologically benign peripheral nerve sheath tumors (PNST) called plexiform neurofibromas (PNF)[4, 5]. While benign, in persons with NF1, there is an 8-16% lifetime risk of these pre-existing neurofibroma (PNST) undergoing malignant transformation to a highly aggressive form of sarcoma called a malignant peripheral nerve sheath tumor (MPNST) [6]. MPNST, which are often resistant to both chemotherapy and radiation, represents the leading cause of premature death in individuals with NF1 [6, 7]. Wide marginal excision is the only curative therapy, however due to infiltration of adjacent vital structures, adequate resection is often not possible. At present, our ability to accurately predict risk of PNST transformation or MPNST therapeutic response remains limited. Thus, a deeper understanding of MPNST tumorigenesis is critical to improving outcomes in patients with NF1-associated PNST.

The development of MPNST from pre-existing PNF often proceeds through intermediate lesions known as atypical neurofibromatous neoplasms with unknown biological potential (ANNUBP). ANNUBP are characterized by the presence of at least two of the following histopathological features: hypercellularity, cytologic atypia, loss of neurofibroma architecture and a mitotic index >1/50 high power fields (HPF) and <3/10 HPF [8]. However, not all ANNUBP will progress to MPNST and our ability to identify neurofibroma at a high risk of undergoing malignant transformation remains limited [8]. Importantly, recent work by our group and others has demonstrated that even PNST that appear histopathologically benign can exhibit significant molecular heterogeneity and diverse evolutionary trajectories [9, 10]. Translocations or copy number loss of cyclin-dependent kinase inhibitor 2A (CDKN2A) and somatic mutations involving components of the polycomb repressive complex 2 (PRC2) have been implicated as early drivers of PNST evolution, however these events do not occur in all MPNST [10-18]. Mounting evidence suggests that the re-activation of embryogenic programs promotes MPNST initiation, heterogeneity and tumor progression [19]. Accordingly, recent studies have begun to identify discrete subsets of MPNST characterized by the reactivation of embryonic molecular signatures and exhibiting distinct therapeutic vulnerabilities [17, 20]. The mechanisms by which these transcriptional programs of de-differentiation become aberrantly activated in MPNST, however, remain poorly defined.

Dysregulation of developmentally restricted proteins can promote the acquisition of a stem-like phenotype and the development of aggressive disease in multiple human cancers [21]. One such protein, Delta Like Non-canonical Notch Ligand 1 (DLK1), is overexpressed in a range of solid tumors [22]. During physiologic development, DLK1 regulates cell proliferation, differentiation and apoptosis through interactions with members of the Notch and Wnt signaling pathways [23]. In malignancy, emerging data suggests that DLK1 may contribute to the maintenance of cancer stemness and therapeutic resistance[23].

In the present study, we show that aberrant overexpression of DLK1 is characteristic of MPNST and provide evidence that DLK1 overexpression may precede histological changes consistent with malignancy. In complementary analyses, we find that serum levels of DLK1 are significantly higher in both mice and humans harboring MPNST. Importantly, through the integration of multiple, independent datasets, we demonstrate that divergent levels of DLK1 expression distinguish MPNST subsets characterized by unique molecular programs and potential therapeutic vulnerabilities. Specifically, we show that DLK1 overexpression is associated with the reactivation of embryonic signatures, an immunosuppressive microenvironment and worse overall survival in patients with NF1-MPNST. Together these findings provide critical insight into MPNST tumorigenesis and support future, prospective trials evaluating the utility of DLK1 as a biomarker capable of augmenting diagnosis, risk assessment and therapeutic stratification in the setting of NF1-PNST.

MATERIALS AND METHODS

Ethical considerations

Archived samples associated clinical data and imaging were collected under approval by the Indiana University Institutional Review Board under exempt protocols #17332 and #20042. For patients in the institutional cohort whose samples underwent bulk RNA sequencing, whole exome sequencing and spatial transcriptomics profiling, written informed consent was obtained under IRB protocol #1501467439.

Sample selection

Immunohistochemistry.

Institutional cohort samples were selected retrospectively by querying the Indiana University pathology archives with the search terms “neurofibromatosis type 1”, “neurofibroma”, “ANNUBP”, “atypical neurofibroma”, “neurofibroma with atypia”, “plexiform neurofibroma” and “MPNST.” Hematoxylin and eosin (H&E)-stained tissue sections from each tumor were reviewed by a board-certified pathologist, with expertise in diagnosing NF1-associated nerve sheath tumors[8]. Institutional cohort metadata is available in Supplemental Table 1).Validation cohort tissue samples were obtained from the Johns Hopkins NF1 Biospecimen Repository (Supplemental Table 2)[24].

RNA sequencing.

Fresh frozen tumor samples from patients with NF1 (detailed clinical data unavailable) were obtained for bulk RNA and whole exome sequencing under the protocols listed above from the Indiana Pediatric Biobank and the IU Simon Comprehensive Cancer Center Biospecimen Repository.

Overall Survival.

Patients from our institutional cohort with adequate tumor samples for immunohistochemical staining of DLK1 and appropriate clinical data were included in survival analyses. Overall survival was calculated as the number of days between histopathological confirmation of MPNST and date of death. For patients who are still alive, overall survival was calculated using the date of last know contact documented in the medical record. Patients were stratified into DLK1 positive (n=7) or DLK1 negative (n=8) based on whole tumor DLK1 percent positivity as described in the Immunohistochemistry section below. Kaplan-Meier plots were generated in GraphPad Prism and p-values were determined by log-rank Mantel-Cox test.

RNA sequencing dataset acquisition and corresponding analyses

Bulk RNAseq and scRNAdeq analysis of Schwann cell development.

Bar, tSNE and violin plots of Dlk1 and Notch1 expression at respective stages of embryonic and postnatal development were obtained using bulk transcriptome analysis of FACS isolated, P0Cre eYFP reporter labeled Schwann cells from the Sciatic Nerve Atlas (SNAT) web portal (https://www.snat.ethz.ch) developed by Gerber and colleagues [25].

Nf1^−/− GFP, Nf1^−/− Cre+, Nf1^−/−Arf^−/− Dorsal root ganglia (DRG) NeuroSphere Cells (DNSC).

Nf1^−/− GFP, Nf1^−/− Cre+, Nf1^−/−Arf^−/− Cre+ DNSCs were isolated as previously described[9] and total RNA was extracted using Qiagen RNeasy Plus Mini kit with on-column DNase I treatment according to the manufacturer’s protocol. Library construction was performed using KAPA mRNA Hyperprep Kit (Roche) according to the manufacturer’s protocol with 100ng input total RNA and 13 cycles PCR amplification. FASTQC (RRID:SCR_014583) passed reads were aligned to the mouse reference genome (GENECODE M26) and translated to transcriptome coordinates using Salmon (v1.4)[26] using a decoy-aware transcriptome index generated using GENCODE mouse reference assembly (GRCm39, M26 release). Transcript abundance estimates were collated for gene-level quantification using the tximport function in R using DESeq2 (RRID:SCR_000154)[27]. Statistical analyses were performed in GraphPad Prism as detailed below.

Human MPNST cell lines.

Frozen cell pellets of human MPNST cell lines (2 replicates each) were sent to GENEWIZ Azenta Life Sciences (South Plainfield, New Jersey, USA) for RNA extraction and bulk RNA sequencing. Log₂(CPM+4) normalization of raw counts was performed using iDEP .96[28]. PRC2 status has been previously published[13, 29, 30] and confirmed in our laboratory. Statistical analysis was performed in GraphPad Prism as detailed below.

TCGA NF1-MPNST verses GTEx Normal Nerve.

RNA sequencing data was obtained from the TARGET TCGA GTEx dataset using the UCSC Xena browser database (RRID:SCR_018938) [31]. Differential gene expression (DEG) analysis comparing TCGA MPNST (n=6) and GTEx Normal Nerve samples (n=278) was performed using the UCSC Xena browser DEG pipeline. Gene Ontology Biological Process (GOBP) 2023 and CellMarker Augmented 2021 enrichment by DEGs with adjusted p-values ≤ 0.05 and log₂FC of ≥ 1 were determined using Enrichr[32].

Johns Hopkins MPNST verses Benign Neurofibroma.

Raw FASTQ files from bulk RNA sequencing of n=2 cutaneous neurofibromas, n=13 plexiform neurofibromas , n=11 nodular neurofibroma and n=4 MPNST were obtained from the Johns Hopkins NF1 Biospecimen Repository[24] via the NF Data Portal (Synapse, #syn19522967, Supplemental Table 3)[33]. Multiple FASTQ files corresponding each sample were concatenated in Python (RRID:SCR_024202) prior to quantification of transcript abundance in kallisto using a transcriptome index (.idx) file produced from the Ensembl GRCh38 v96 release available for download from the Patcher lab (https://github.com/pachterlab/kallisto-transcriptome-indices/releases). Transcript abundance estimates from kallisto were used to create a gene-level count matrix using the tximport package in R. Normalized counts were used for DEG analysis via DESeq2. Differentially expressed protein-coding genes were generated using iDEP.96[28].

GSE141438.

DEGs were obtained from the Gene Expression Omnibus (RRID:SCR_005012) under the acquisition number GSE141438[34].

TCGA Sarcoma.

SUZ12 and EED gistic2 thresholded copy number alteration and DLK1 expression data were obtained from the TCGA Sarcoma dataset using UCSC Xena browser database[31]. Correlation analyses were performed in GraphPad Prism as detailed below and verified using the UCSC Xena browser scatterplot function.

NF1 TCGA-MPNST:

The RNA-Seq by Expectation-Maximization (RSEM) normalized counts matrix was obtained from the TCGA Sarcoma (Firehose) dataset using cBioPortal (RRID:SCR_014555)[35] (https://www.cbioportal.org/). MPNST samples obtained from patients confirmed to have NF1 were isolated and those with DLK1 log₂(CPM+4) expression greater than the average expression across all MPNST samples (mean=9.03) were classified as DLK1_Hi, those below the mean as DLK1_Lo. DEG analysis comparing DLK1_Hi and DLK1_Lo TCGA-MPNST (n=6) was performed using the UCSC Xena browser[31] DEG pipeline. PRC2 complex pathway activation, methylation data and normalized counts for SOX11, MEG3 and MEG8 were obtained through UCSC Xena browser[31]. Average methylation across the DLK1-DIO3 locus was calculated by averaging beta values (β) across all CpG sites for each sample. Statistical analyses were performed in GraphPad prism as described below. Additional analyses were performed as detailed below in Corresponding Analyses.

Institutional Cohort.

Genomic DNA and RNA isolation.

RNA was isolated using the RNeasy Plus Mini Kit (Qiagen) according to manufacturer's protocol with optional DNase I treatment (Qiagen) for 15 minutes. A minimum of 200 ng of total RNA was used as input for RNAseq library construction. Genomic DNA was isolated using the DNeasy Blood and Tissue Kit (Qiagen) according to manufacturer's protocol with the optional RNase A treatment for 2 min at room temperature prior to the addition of Buffer AL. 400 ng of total DNA was used as input for whole-exome sequencing.

Whole Exome Sequencing.

Whole exome sequencing was conducted by Novogene. BAM files were sorted with Sambamba and duplicate reads were marked using Picard. Following genomic variant detection (SNVs/indels), annotation of variants was performed using ANNOVAR[36] to identify protein coding changes, genomic regions affected by the variants, allele frequency, and predictions regarding deleteriousness. The RefSeq (RRID:SCR_003496) and Gencode (RRID:SCR_014966) databases were used to identify genomic regions affected by variants. SIFT (RRID:SCR_012813), PolyPhen (RRID:SCR_013189), MutationAssessor (RRID:SCR_005762), LRT and CADD scores were used to predict the deleteriousness of mutations. GERP++ scores were used to access the conservation of mutations. 1000 Human Genome, Exome Aggregation Consortium (ExAC), Genome Aggregation Database (gnomAD) and exome sequencing project (ESP), were used to determine alternative allele frequencies in indicated populations. Databases dbSNP, COSMIC (RRID:SCR_002260), OMIM, GWAS Catalog and HGMD were used to obtained reported information of the variant, such as top SNPs in GWAS and cancer/disease associations.

Bulk RNA sequencing.

Messenger RNA was purified from total RNA using poly-T oligo-attached magnetic beads. Bulk RNA sequencing was performed by Novogene. Log₂(CPM+4) normalization and DEG analysis were performed using iDEP .96[28]. Samples with DLK1 Log₂(CPM+4) greater than the average expression (mean=8.93) across all samples were classified as DLK1_Hi , those below the mean as DLK1_Lo. Additional analyses were performed as detailed below in Corresponding Analyses.

GSE178989.

The raw count matrix was obtained from the Gene Expression Omnibus under the acquisition number GSE178989[37]. Log₂(CPM+4)normalization and DEG analysis were performed using iDEP .96[28]. Samples with DLK1 log₂(CPM+4) expression greater than the average expression across all samples (mean=9.92) were classified as DLK1_Hi , and those below the mean as DLK1_Lo. Additional analyses were performed as detailed below in Corresponding Analyses.

Cancer Cell Line Encyclopedia.

Cancer cell line RNA sequencing, DEGs and treatment response data were obtained from the Cancer Cell Line Encyclopedia[38] using cBioPortal[35, 39, 40] (https://www.cbioportal.org/). Query language “DLK: EXP >2” was used to identify samples with DLK1 expression greater than two standard deviations above the mean (DLK1_Hi, n=50). Samples not meeting this criterion were designated “unaltered” (Belinostat n=10 DLK1_Hi and n=560 unaltered, AR-42 n=10 DLK1_{Hi ,} and n=570 unaltered). Gene Ontology Biological Process (GOBP) 2023 enrichment by genes upregulated in DLK1_Hi cell lines with log₂ ratio ≥ 1 and q-value ≤ 0.05 was generated using ShinyGo 0.80 (RRID:SCR_019213) [41]. Reactome 2022 pathway enrichment analysis DLK1_Hi cell lines with log₂ ratio ≤ 1 and q-value ≤ 0.05 was performed in EnrichR [32, 42, 43]. Statistical analysis for treatment response were performed in GraphPad as described below.

Corresponding analyses.

Principal Component Analysis.

PCA plots were generated in R Studio (RRID:SCR_000432) using coordinates acquired from iDEP.96[28] and using log₂ transformed, normalized counts as input under default settings.

Heatmaps.

Heatmaps were generated in Morpheus (https://software.broadinstitute.org/morpheus, RRID:SCR_017386) using matrices acquired from iDEP .96 [28]. Log₂ transformed, z-score normalized counts from the top 1000 most variable genes were clustered by correlation coefficient with average linkage.

Pathway Enrichment.

Lollipop plots reflecting Gene Ontology Biological Process (GOBP) 2023 enrichment by genes upregulated with adjusted p-values ≤ 0.05 and log₂ fold change of ≥ 1 in DLK1_Hi samples from the institutional, TCGA-MPNST and GSE178989 datasets were generated using ShinyGo 0.80[41].Gene Ontology Biological Process (GOBP) 2023 enrichment analyses by genes downregulated with adjusted p-values ≤ 0.05 and log₂ fold change of ≤ 1 in DLK1_Hi samples from the institutional, TCGA-MPNST and GSE178989 datasets were performed using EnrichR[32, 42, 43].

Other analyses.

Geneset enrichment of indicated signatures, cell type mapping with Immunostates[44] as the reference dataset and L1000 activity[45, 46] plots for respective datasets were generated in Omics Playground software v2.7.18 run using a local Docker image as described by Akhmedov and colleagues[47]. Pairwise scatter plot for the co-expression of DLK1 and SOX11 across MPNST from the institutional TCGA and GSE178989 datasets were also generated using Omics Playground. Geneset enrichment tree depicting GO Biological Process enrichment of the genes up-(gold) and down- (teal) regulated in DLK1_Hi cell lines was generated in iDEP.96.

Spatial Profiling.

H&E stained FFPE slides from selected samples were evaluated for application on the Visium slides to provide optimal coverage of 6.5 × 6.5 mm² capture areas using CytAssist technology (10x Genomics) . Transcriptomic probes were hybridized and processed according to the standard Visium workflow. cDNA libraries were sequenced on an Illumina NovaSeq platform. Space Ranger (v.2.1.0) was used to align probe reads to the human refence genome GRCh38. The resulting count matrices and associated H&E images were used for downstream analysis in scanpy (RRID:SCR_018139). Spots with total counts less than <2000 and >35000 and >5% mitochondrial genes were excluded. Genes detected in <10 spots were also removed. Counts were then normalized, log transformed and the top 2000 variable genes selected for manifold embedding and clustering based on transcriptional similarity. Hierarchically clustered heatmaps of z-score scaled expression values of selected genes by cluster were generated in scanpy using the matrixplot function. Surface plots of depicting the expression of selected genes were generated using the SPATA2 (v2.0.4) package in R. Analysis and visualization of inferred gene expression along spatial trajectories was also performed using the SPATA2 (v2.0.4).

Immunohistochemistry

5 μm thick tissue sections were deparaffinized, hydrated and transferred to 0.1M EDTA (pH 9.0) for antigen retrieval in a pressure cooker. Samples were treated with 3% hydrogen peroxide for 10 min and then incubated with primary antibodies overnight at 4 °C (DLK1:10636-1-AP, 1:3000, Proteintech) or for 30 minutes at room temperature (MTAP: 74683S, 1:100, Cell Signaling Technology). Incubation with secondary antibody was performed for 1 hour at room temperature (goat anti-rabbit, ab205718, 1:1000, Abcam) followed by VECTASTAIN DAB, which was applied for 10 minutes and terminated by rinsing in distilled water. Counterstaining was performed with modified Mayer’s hematoxylin (Vector), and the sections were dehydrated, cleared and cover slipped. Slide images were acquired on an Aperio ScanScope CS at 20x magnification. Quantitative immunohistochemical analyses for DLK1 and MTAP were conducted using the Cytonuclear IHC module of HALO Image Analysis software (version 3.0.5, Indica Labs). Cytonuclear analysis settings were optimized and the intensity of nuclear staining was scored as negative (0, blue), weakly positive (1+, yellow), moderately positive (2+, orange), or strongly positive (3+, red). For quantification of DLK1, an average of 5 regions of interested (ROI) were annotated per sample to encompass the entirety of each viable tumor section. All ROIs in a specific tumor were obtained from a single tissue block. The total percent positive cells for PNF (n=9, ROI=48), ANNUBP (n=7, ROI=38) and MPNST (n=18, ROI=92) was used for the statistical analysis performed in GraphPad Prism software as described below. Whole tumor percent positivity for DLK1 in MPNST was determined from a single annotation of whole MPNST sections. Tumors with less than 1% DLK1 positive cells were considered negative. A cutoff of 1% was chosen as this was the threshold below which no positive cells could be detected on visual inspection. MTAP quantification was performed by determining the percent 2+ (moderate) and 3+(strong) MTAP positive cells from MPNST samples (n=12) annotated with an average of 10 ROIs (n=123) encompassing the entirety of the viable tumor sample. Statistical analyses were performed using GraphPad Prism software as described below.

DLK1 Serum Analyses

Human serum samples (MPNST n=11, PNF n=19) were obtained from the Johns Hopkins NF1 Biospecimen Repository[24]. Murine serum samples were obtained from Nf1^flox/flox;Cdkn2a^flox/flox (n=20) and Nf1^flox/flox;Cdkn2a^flox/+ cre+ (n=4) mice with (n=16) and without MPNST (n=8) in accordance with IACUC protocols. Serum concentrations of DLK1 were determined by ELISA specific to human (Invitrogen, Human Pref-1 ELISA Kit, EH379RB) or mouse (Invitrogen, Mouse Pref-1/DLK-1/FA1 ELISA Kit, Invitrogen, EM66RB) per the manufacturer’s instructions. Concentrations were acquired with a Molecular Devices VersaMax colorimetric microplate reader and SOFTmax PRO 4.3.1 Life Sciences Edition software. For both murine and human data, results were validated in repeat experiments in duplicate. Grubbs’ method with an α=0.0001 was used to identify outliers. Statistical analyses were performed using GraphPad Prism software as described below. Human serum sample metadata is presented in Supplemental Table 4.

Culture of Human MPNST cell lines

Human MPNST cell lines JH-2-079c [29, 30], JH-2-002 [24, 48] and JH-2-103 [30] were obtained from the Johns Hopkins NF1 Biospecimen Repository[24]. ST-8814 (RRID:CVCL_8916) and S462 (RRID:CVCL_1Y70) cells [49] were a generous gift from Dr. Andrew Tee (Cardiff University). NF90.8, R-HT172 and R-HT163 cells [50] were obtained from Dr. Melissa Fishel (Indiana University). Human PNF lines (hTERT NF1 ipNF95.6 and hTERT NF1 ipNF05.5) were obtained from the Peggy Wallace lab[51]. Cells were cultured in either DMEM (NF90.8, S462, ST-8814, R-HT172, R-HT163, ipNF95.6, ipNF05.5) or DMEM/F12 (JH-2-002, JH-2-103, JH-2-079c) media supplemented with 10% FBS (Harvest Midsci), 1% glutamine (Gibco), 1% penicillin/streptomycin (Lonza), and prophylactic 5 μg/mL Plasmocin (Invivogen). TrypLE 0.05% (Gibco) was used to dissociate cells for passaging upon reaching confluence. Cultures were tested for mycoplasma and confirmed to be negative prior to experimentation using the MycoAlert^® Mycoplasma Detection Kit (Lonza). Cell lines were authenticated by IDEXX BioAnalytics CellCheck^™. For spheroid forming conditions, cells were grown in standard media on Corning^® Costar^® ultra-low attachment 6-well plates for a minimum of 72 hours (Millipore Sigma, CLS3471). For spheroid transfer experiments, cells were grown in spheroid-promoting conditions for 72 hours at which time whole spheres were transferred by pipette to standard, adherent 6-well plates. Cell lysates were collected at 24-, 48-, 72- and 120-hours post-transfer.

siRNA transfection

Human MPNST cell lines were reverse transfected with 10 nM siRNA against DLK1(Santa Cruz Biotechnology, sc-39669) or scrambled control (Santa Cruz Biotechnolgy, sc-37007) using Lipofectamine 3000 transfection reagent (Invitrogen, L3000075) according to the manufacturer’s instructions. Cells were plated in the transfection complex at a density of 1.5 million cells per 10cm dish. Cell lysates were collected 48 hours post-transfection to confirm protein knockdown by Western blot. Viability was assessed 48 hours post-transfection by determining the number of living cells using an TC20 automated cell counter (BioRad, 1450102). Cell counts were normalized to control for each replicate. One sample t and Wilcoxon tests were performed in GraphPad Prism software as described below.

Generation of DLK1 overexpressing human PNST cell lines

An expression plasmid containing human DLK1 sequence with a C-terminal Flag epitope tag was purchased from VectorBuilder. The DLK1-Flag sequence was sub-cloned into a Gateway compatible vector, pENTR4 no ccDB (686-1), a gift from Eric Campeau & Paul Kaufman (Addgene plasmid # 17424; http://n2t.net/addgene:17424; RRID:Addgene_17424). Gateway LR Clonase II Enzyme Mix (Thermo Fisher Scientific) was used according to the manufacturer’s instructions to recombine DLK1-Flag into pCW57.1, a gift from David Root (Addgene plasmid # 41393 ; http://n2t.net/addgene:41393 ; RRID:Addgene_41393). Equal amounts of pCW57.1-DLK1-Flag, and the lentiviral packaging plasmids, pMD2.G and psPAX2 (gifts from Didier Trono (Addgene plasmid # 12259 ; http://n2t.net/addgene:12259 ; RRID:Addgene_12259 and Addgene plasmid # 12260 ; http://n2t.net/addgene:12260 ; RRID:Addgene_12260) were used with JetPRIME (2:1 ratio reagent:DNA) to transfect HEK293T cells. Forty-eight hours after transfection, lentiviral supernatant was collected, filtered through low-protein binding 0.45-micron filters, and used to transduce target cells for 48 hours in the presence of 8 μg/mL polybrene. Cells were selected using 2 μg/mL (ST-8814) or 1μg/mL (ipNF05.5, ipNF95.6) puromycin in media for at least 3 days prior to experiments. Doxycycline was added to media at a final concentration of 1 μg /mL to induce expression.

Colony Forming Assays

For colony formation assays, 5000 cells were added to each well of a 6-well plate and allowed to proliferate until controls were confluent (~5-7 days). Cells were then stained with methylene blue diluted in 50% methanol and 50% H₂O. For colorimetric measurement, each well was dissolved in 0.5 N HCl and the dissolved fluid was transferred to a 96-well plate to be read at 660 nm. For experiments utilizing human PNST cell lines overexpressing DLK1, puromycin and doxycycline were replaced every 3 days. Spheroid formations present in each well (n=3 per condition, per cell line) were counted and images were acquired using an ECHO Revolve microscope (ECHO, San Diego, CA) at 4X magnification with phase contrast.

Western Blot Analysis

For blots in Figure 3 and Supplemental Figure 5&6, lysis buffer was prepared using cOmplete Mini Protease Inhibitor Cocktail (Roche, 11836153001), PhosSTOP Phosphatase Inhibitor Cocktail (Roche, #4906837001) and Pierce^™ IP lysis buffer (Invitrogen, #87787). Cell lysates were collected, and protein concentrations were determined using Pierce BCA Protein Assay Kit (Thermo Fisher Scientific, #23227). Isolated proteins were fractionated using 4–20% Mini-PROTEAN^® TGX^™ Precast Protein Gels (BioRad, #4561094) and electro-transferred to PVDF. For blots in Figure 6H, lysis buffer consisted of 50 mM HEPES, 150 mM NaCl, 0.5% Triton X-100, 1 mM EDTA, 1 mM EGTA, pH 7.5) supplemented with cOmplete Mini Protease Inhibitor Cocktail, 1% (v/v) Phosphatase Inhibitor Cocktails 2 and 3 (Sigma-Aldrich, #P5726 and #P0044), 10 mM NaF, 2.5 mM NaVO₄. Cell lysates were collected, and protein concentrations were determined using Pierce Coomassie (Bradford) Protein Assay Kit (PI23200, Fisher Scientific). Isolated proteins were fractionated using 4–20% Mini-PROTEAN^® TGX^™ Precast Protein Gels (BioRad, #4561094) and electro-transferred to nitrocellose membranes. Immunoblots were carried out using antibodies specific to GAPDH (2118L, Cell Signaling Technology), Vinculin (CP74, Millipore Sigma), SOX11 (MA5-32395, Invitrogen), Cleaved Notch1 (41475, Cell Signaling Technology), DLK1 (10636-1-AP, Proteintech), phospho-ERK (#9101, Cell Signaling Technology), SUZ12 (#3737, Cell Signaling Technology), HA-Tag (#3724, Cell Signaling Technology), H3K27me3 (9733, Cell Signaling Technology), total Histone H3 (4499, Cell signaling Technology), or GAPDH (sc-365062, Santa Cruz Biotechnology). Following incubation with primary antibody, blots were incubated with appropriate HRP conjugated secondary antibody (anti-rabbit: NA934V, Cytiva, anti-rabbit: 7074V, Cell Signaling Technology, anti-mouse: NA931V, Cytiva, anti-mouse: 7076V, Cell Signaling Technology). Signals were detected using ECL chemiluminescence substrate (SuperSignal^™ West Pico PLUS Chemiluminescent Substrate 34580 and/or SuperSignal^™ West Femto Maximum Sensitivity Substrate 34095) and images were collected on a BioRad ChemiDoc. Band intensities were quantified using ImageJ software (NIH, version 1.53), with signals normalized to the corresponding loading control and expressed relative to control conditions.

Figure 3. DLK1 expression is increased in human MPNST cells characterized by a stem-like phenotype.

(A) Bar plot of Notch1 transcript expression (RPKM) at indicated days of embryonic (E13.5, E17.5) and postnatal (P1, P5, P14, P24, and P60) development obtained from the Sciatic Nerve ATlas (SNAT). (B) Box and whisker plot of SOX11 mRNA expression comparing DLK1_Hi (orange) and DLK1_Lo (teal) MPNST from our institutional cohort generated in R studio. Dots represent individual samples. Error bars represent the 95% confidence interval. The center line represents the median. The box spans the 25^th to 75^th percentiles. Data beyond the whiskers are outliers and are plotted as individual points. P-value represents unpaired, two-tailed t-tests between groups. (C) Pairwise scatter plot for the co-expression of DLK1 and SOX11 across MPNST from our institutional cohort. Dots indicate individual samples with orange indicating DLK1_Hi and teal, DLK1_Lo, samples. The straight line corresponds to the linear regression fit. Pearson r= 0.80. (D) Western blot of DLK1 expression in human MPNST cell lines. GAPDH serves as a loading control. (E,F) Western blots of DLK1 expression in human MPNST cell lines S462 (E) and ST-8814 (F) grown in standard, adherent (Adh) or low-adhesion, spheroid-promoting (LA) conditions. Vinculin serves as the loading control. (G) Bar graph depicting DLK1 expression in human MPNST cell lines grown under spheroid-promoting (LA) conditions (LA) plates normalized to control (Adherent, Adh). Dots represent independent experiments. Error bars represent the standard error of the mean (SEM). P-value reflects one sample Wilcoxon test (theoretical mean=1). Graph represents pooled results of three independent cell lines. (H,I) Western blots of SOX11 expression in human MPNST cell lines S462 (H) and ST-8814 (I) grown in standard, adherent (Adh) or low-adhesion, spheroid promoting (LA) conditions. GAPDH and vinculin serve as loading controls. (J) Bar plot depicting the percent viability of a human MPNST cell line (S462) following siRNA-mediated depletion of DLK1 normalized to control. Dots represent independent experiments. Error bars represent the SEM. P-value reflects one sample Wilcoxon test (theoretical mean=100). Graph reflects pooled results from three independent experiments. (K) Representative images depicting a decrease in cell number of a human MPNST cell line (S462) following siRNA-mediated depletion of DLK1 compared to control. (L) Western blot depicting the expression of SOX11, DLK1, cleaved NOTCH1, and phospho-ERK (pERK) in a human MPNST cell line (S462) following siRNA-mediated depletion of DLK1 compared to control. Vinculin and GAPDH serve as loading controls. (M) Bar graph depicting DLK1 expression in a human MPNST cell line following siRNA-mediated depletion of DLK1 compared to control. Dots represent independent experiments. Error bars represent the SEM. P-value reflects one sample Wilcoxon test (theoretical mean=1). Graph represents pooled results of six independent experiments. (N) Bar graph depicting cleaved NOTCH1 expression in a human MPNST cell line following siRNA-mediated depletion of DLK1 compared to control. Dots represent independent experiments. Error bars represent the SEM. P-value reflects one sample Wilcoxon test (theoretical mean=1). Graph represents pooled results of three independent experiments. (O) Bar graph depicting SOX11 expression in human MPNST cell lines following siRNA-mediated depletion of DLK1 compared to control. Dots represent independent experiments. Error bars represent the SEM. P-value reflects one sample Wilcoxon test (theoretical mean=1). Graph represents pooled results of three independent cell lines. (P) Bar graph depicting phospho-ERK (pERK) expression in a human MPNST cell line following siRNA-mediated depletion of DLK1 compared to control. Dots represent independent experiments. Error bars represent the SEM. P-value reflects one sample Wilcoxon test (theoretical mean=1). Graph represents pooled results of three independent experiments. ****= p-value ≤0.0001, ***=p-value ≤0.001 **= p-value≤0.01 *=p-value ≤ 0.05, ns=not significant.

Figure 6. Overexpression of DLK1 occurs independently of PRC2 loss.

(A) Schematic depicting the imprinted DLK1-DIO3 locus. Red represents genes expressed on the maternal allele and blue represents genes expressed on the paternal allele. Gray represents genes that are repressed. Filled circles represent methylated DMRs, whereas unfilled circles are unmethylated DMRs. (B,C) Scatter plots comparing DLK1 expression in TCGA Sarcoma samples (n=255) based on indicated gistic2-thresholded copy number alterations of PRC2-components SUZ12 (B) and EED (C) obtained through the Xena Browser database. “1” represents heterozygous loss (−) or gain (+), whereas “2” represents homozygous loss (−) or gain (+) and “0” indicates no copy number alteration (Neutral). Pearson correlation analysis conducted in GraphPad Prism revealed no significant correlation between copy number alterations of SUZ12 (Pearson r=−0.0247, 95% CI [−0.1471 to 0.09841], p=0.6943) or EED (Pearson r=−0.0611, 95% CI [−0.1826 to 0.06220], p=0.3311) and expression of DLK1. (D) Bar plot comparing inferred PRC2 complex activation in DLK1_Hi and DLK1_Lo TCGA-MPNST. Dots represent individual samples. Error bars reflect standard error of the mean (SEM). P-value represents unpaired, two-tailed t-tests between groups. (E) Heatmap depicting the methylation (beta value, β) pattern of TCGA-MPNST across indicated CpG sites (n=560) within the DLK1-DIO3 locus. Rows represent individual samples, and their respective β across each CpG site. Red corresponds to a higher, and blue to lower a β value according to the legend. DLK1 expression level is denoted by red and blue bars as indicated. Missing values are depicted in black (F) Bar plot comparing average methylation (beta value, β) across CpG sites within the DLK1-DIO3 locus between DLK1_Hi and DLK1_Lo TCGA-MPNST. Dots represent individual samples. Error bars reflect the SEM. P-value represents unpaired, two-tailed t-test between groups. (G) Bar plot comparing DLK1 mRNA expression across PRC2-deficient human MPNST cell lines. Dots represent individual samples. Error bars reflect the SEM. P-values represent one-way ANOVA using Tukey’s multiple comparisons tests between groups. Only significant comparisons are depicted on the graph. (H) Western blots of DLK1 expression in PRC2-deficient human MPNST cell lines JH-2-002 and NF90.8 following re-expression of HA-SUZ12 and restoration of H3K27me3. Yellow fluorescent protein with a nuclear localization signal (NLS-YFP) was used as an ectopic expression control. GAPDH serve as the loading control. (I) Bar plot comparing Suz12, Eed and Dlk1 mRNA expression in murine Dorsal root ganglia (DRG) NeuroSphere Cells (DNSCs) with (Nf1Arf-Cre) and without (Nf1-Cre, Nf1-GFP) malignant potential. Dots represent individual samples. Error bars reflect the SEM. P-values represent two-way ANOVA using Šídák's multiple comparisons test. ****= p-value ≤0.0001, ***=p-value ≤0.001 **= p-value≤0.01 *=p-value ≤ 0.05, ns=not significant.

Statistical Analysis

Statistical analyses were performed in R studio (RRID:SCR_000432) or using GraphPad Prism 9.5.1 software (GraphPad, La Jolla, CA, RRID:SCR_002798). Specific analyses used to identify statistically significant differences between groups are detailed in the corresponding figure legends. Outliers were identified using Grubbs’ method (α=0.0001) or the Robust regression and Outlier removal (ROUT) method, Q=0.1%[52] where indicated in corresponding figure legends. P-values ≤ 0.05 were considered statistically significant. Pearson coefficients (r) were calculated by correlation analyses in GraphPad Prism using default parameters. Reported p-values correspond to a two-tailed hypothesis test.

Data Availability

Bulk RNA sequencing of human MPNST cell lines and spatial profiling data of the ANNUBP and MPNST specimens are accessible through Synapse (syn63944395). Whole exome sequencing and bulk RNA sequencing data of tumors from the institutional cohort have been deposited in the database of Genotypes and Phenotypes (dbGaP) under accession number phs003835.v1.p1. Bulk RNA sequencing of Nf1^−/− GFP, Nf1^−/− Cre+, Nf1^−/−Arf^−/− Cre+ DNSCs is available on the GEO database (accession #: GSE232451). All other data presented in this manuscript is available from the authors upon request.

RESULTS

Re-activation of embryogenic programs is associated with overexpression of DLK1 in NF1-associated MPNST.

To better understand the molecular alterations distinguishing MPNST from benign and normal tissue, we first performed differential gene expression (DEG) analysis comparing normal nerve and NF1-associated MPNST (TCGA-MPNST) samples utilizing data obtained from the GTEx (n=278) and TCGA datasets (n=6), respectively (Supplemental Table 5)[31]. Only MPNST arising in patients with confirmed sporadic or familial NF1 were included for analysis. Genes upregulated in TCGA-MPNST broadly enriched for processes involved in mitosis, cell cycle regulation and immune system response (Supplemental Table 6). CellMarker Augmented 2021 analysis further revealed an upregulation of genes enriching for signatures consistent with stem and progenitor cell programs in tumor samples (Supplemental Table 7).

Recent work by Sun et al., has shown that cells with stem-like properties promote tumoral heterogeneity and progression of NF1-MPNST [19]. Therefore, we next utilized the Sciatic Nerve Atlas (https://www.snat.ethz.ch) to identify members of the top 20 significantly upregulated protein-coding genes in TCGA-MPNST that are highly expressed in the embryonic, but not the postnatal sciatic nerve (Supplemental Table 5, yellow) [25]. Eight genes met this criterion, including BIRC5 and RRM2, which were excluded for further consideration as previous reports have already detailed their roles in MPNST [9, 53]. To narrow our focus to genes most relevant to human disease, we performed differential gene expression analysis comparing MPNST to benign neurofibroma (Supplemental Table 8) and identified DLK1 as the most significantly upregulated (log₂FC=6.508, adjusted p-value=0.033) of the aforementioned, embryonically restricted genes (n=8). Comparison of MPNST to adjacent normal tissue (GEO Database, accession #: GSE141438) or benign neurofibroma (GEO Database, accession #: GSE178989) using two additional independent datasets, again revealed significant upregulation of DLK1 in MPNST (Supplemental Table 9 & 10) [34].

DLK1 is highly expressed throughout embryogenesis where it plays an essential role in regulating cell proliferation, differentiation and development [54]. Conversely, in adulthood DLK1 expression is largely undetectable with only select tissues retaining basal level expression [23]. Consistent with these results, analysis of data obtained through the Sciatic Nerve Atlas (SNAT) [25] revealed that expression of Dlk1 in the murine peripheral nervous system peaks on embryonic day 17.5 (E17.5) and gradually returns to near zero by postnatal day 60 (P60) (Supplemental Figure 1A). Within the postnatal day 1 (P1) murine sciatic nerve, Dlk1 is broadly expressed, with the highest expression occurring within the epi- (EpC), per- (PnC) and endoneurial (EnC) cells (Supplemental Figure 1B-D). However, by P60, Dlk1 expression is drastically reduced and what remains is largely restricted to the endoneurium (Supplemental Figure 1E-G). Merged t-SNE plot of Schwann cells, specifically, at P1, P5, P14 and P60 reveals expression of Dlk1 is restricted to a few, immature Schwann cells (iSC), suggesting that aberrant overexpression of DLK1 in NF1-MPNST may identify tumor cells characterized by an immature or de-differentiated phenotype (Supplemental Figure 1H).

Overexpression of DLK1 is associated with a worse overall survival in NF1-MPNST

To validate upregulation of DLK1 at the protein level, we performed immunohistochemical staining of PNST tumors (n=34) obtained from patients (n=22) cared for at our institution (Figure 1A). Consistent with our above results, we found DLK1 protein expression to be significantly upregulated in MPNST compared to both PNF and ANNUBP (Figure 1B). Validation of these findings in an additional cohort from an outside institution (n=23) revealed similar results (Supplemental Figure 2A&B). Three lesions in our institutional cohort (PNF: n=2 of 9, ANNUBP: n=1 of 7) were identified as statistical outliers with respect to DLK1 expression (Supplemental Figure 2C, bold). Strikingly, all three of these tumors were associated with the subsequent development of, or exhibited features concerning for, MPNST. Specifically, IU-26-A, which was histopathologically classified as an ANNUBP in our analysis, exhibited clinical characteristics concerning for malignancy including a nodular appearance with contrast enhancement on MRI and increased metabolic activity FDG-PET (SUV 6.67) (Figure 1C-top). Histopathological diagnostic consensus of this tumor could not be reached with independent expert reviews classifying the lesion as ANNUBP versus low grade MPNST (LG-MPNST). In contrast to other tumors histologically consistent with ANNUBP included in this dataset, immunohistochemical analysis of this lesion demonstrated diffuse DLK1 positivity (Figure 1C-bottom). The second tumor, IU-71-P1, arose in a young (<10y) patient with a significant plexiform burden of the posterior mediastinum and subscapular region (Supplemental Figure 2D-left). This patient subsequently developed a DLK1-positive MPNST (IU-71-M) in this region approximately seven years later (Supplemental Figure 2D-right). The final lesion, IU-48-P, was histologically consistent with PNF and was contiguous with an existing ANNUBP and MPNST. Spatial profiling of the contiguous ANNUBP and MPNST revealed a discrete cluster of DLK1 expressing cells within the ANNUBP and more diffuse DLK1 positivity within the contiguous MPNST, suggesting that aberrant overexpression of DLK1 may precede histopathological evidence of malignancy and may identify pre-malignant lesions at an increased risk of undergoing malignant transformation (Figure 1D). Lastly, while we found tissue expression of DLK1 to be increased in MPNST overall, further review of MPNST samples from our institutional cohort revealed two distinct subgroups defined by either greater than (DLK1+) or less than (DLK1-) 1% of cells staining positively for DLK1. Correlation of DLK1 staining with clinical data from our institutional cohort (n=15) revealed DLK1+ tumors to be associated with a significantly worse overall survival (347 vs. 1739 days, p=0.0443) (Figure 1E), suggesting that elevated tissue levels of DLK1 may identify MPNST with more aggressive behavior.

Figure 1. DLK1 overexpression is associated with a worse overall survival in NF1-MPNST.

(A) Representative photomicrographs of human tumor sections across the PNST continuum immunohistochemically stained for DLK1. Magnification is denoted by 100 μm scale bars with inset high-power magnification as shown. The HALO cytonuclear mask used to quantify DLK1 staining is shown in the bottom panel. Blue cells indicate negative staining for DLK1, yellow, weak positive (1+), orange, moderate positive (2+) and red, strong positive (3+) for DLK1 staining. (B) Box and whisker plot depicting DLK1 positive cells as a percentage of total cells per field generated in R studio. Dots represent individual regions of interest (ROI). Error bars represent the 95% confidence interval. The center line represents the median. The box spans the 25^th to 75^th percentiles. Data beyond the whiskers are outliers and are plotted as individual points. PNF (n=9, ROI=48), ANNUBP (n=7, ROI=38) and MPNST (n=18, ROI=92) were analyzed by one-way ANOVA using Tukey’s multiple comparisons tests between groups. Tumors with greater than 50% of ROIs identified as statistical outliers by ROUT method (Q=0.1%) were considered outliers (PNF: n=2 of 9, ANF: n=1 of 7). Graph represents single experiment with outliers included. (C) Subject IU-26, a 17-year-old female with NF1, presented with a nodular, contrast-enhancing, FDG-PET avid, left pelvic mass (top). Histopathological consensus could not be reached, with the lesion (IU-26-A) being classified as either ANNUBP or LGMPNST following independent, expert reviews. Immunohistochemical staining of the resected lesion revealed diffuse positivity for DLK1 (bottom-right). Box and whisker plot generated in R studio comparing DLK1 positive cells as a percentage of total cells per field of IU-26-A to the other lesions classified as ANNUBP from our institutional cohort (bottom-left). Dots represent individual regions of interest (ROI). Error bars represent the 95% confidence interval. The center line represents the median. The box spans the 25^th to 75^th percentiles. Data beyond the whiskers are outliers and are plotted as individual points. P-value represents unpaired, two-tailed t-tests between groups. (D) Surface plots depicting DLK1 expression within the spatially profiled ANNUBP and the contiguous MPNST. Yellow corresponds to highest expression and dark purple to decreased expression as indicated. Grey represents no expression. (E) Kaplan Meier curve depicting overall survival of patients from our institutional cohort whose MPNST were characterized by either greater than (Positive, n=7) or less than (Negative, n=8) 1% of cells staining positively for DLK1 on IHC (347 vs. 1739 days, p=0.0443). P-value represents log-rank Mantel-Cox test. (F) Box and whisker plot generated in R studio depicting concentration (pg/mL) of DLK1 in the serum of Nf1/Cdkn2a Cre+ mice with MPNST (MPNST, n=19) and without MPNST (No Tumor, n=8). Dots represent individual mice. Error bars represent the 95% confidence interval. The center line represents the median. The box spans the 25^th to 75^th percentiles. Data beyond the whiskers are outliers and are plotted as individual points. P-value represents unpaired, two-tailed t-tests between groups. Graph depicts single experiment. Results were validated in a repeat experiment with a subset of samples. (G) Box and whisker plot generated in R studio depicting DLK1 concentration (ng/mL) in serum obtained from human NF1 patients with PNF (n=19) or MPNST (n=11). A single statistical outlier was identified by Grubbs’ method (α=0.0001) and was excluded from further analysis. Graph depicts single experiment with one outlier excluded (p=0.0452). Dots represent individual patient samples. Error bars represent the 95% confidence interval. The center line represents the median. The box spans the 25^th to 75^th percentiles. Data beyond the whiskers are outliers and are plotted as individual points. P-value represents unpaired, two-tailed t-tests between groups. Results were validated in a repeat experiment conducted in duplicate. ****= p-value ≤0.0001, ***=p-value ≤0.001 **= p-value≤0.01 *=p-value ≤ 0.05, ns=not significant.

Serum DLK1 is elevated in mice and human patients with NF1-MPNST

DLK1 is transmembrane protein that can be released into the serum following cleavage by ADAM17/TACE and prior work has suggested a role for serum DLK1 as a biomarker in other conditions[22, 23, 55]. Review of MPNST samples stained for DLK1 revealed evidence of intravascular DLK1 positivity in several tumor sections (Supplemental Figure 2E). To more rigorously evaluate this observation, we compared DLK1 concentrations in the serum of Nf1^flox/flox;Cdkn2a^flox/floxCre+ and Nf1^flox/flox;Cdkn2a^flox/+ Cre+ (Nf1/Cdkn2a) mice harboring MPNST (MPNST, n=19) to those who had not yet developed malignancy (No Tumor, n=8). Importantly, Nf1/Cdkn2a mice recapitulate human disease and develop MPNST through the malignant transformation of pre-existing plexiform and atypical neurofibroma [9, 11]. This analysis revealed a significant increase (12255 vs 4044 pg/mL, p=0.0153) in serum levels of DLK1 in tumor bearing compared to non-tumor bearing Nf1/Cdkn2a mice (Figure 1F). We next compared the concentration of DLK1 in the serum of human NF1 patients with MPNST (n=11) to that from those without malignancy (n=19). Serum DLK1 levels were significantly higher (12.27 vs. 4.575 ng/mL, p=0.0452) in patients with MPNST compared to those without (Figure 1G). A sample obtained from a patient with a PNF was identified as statistical outlier by Grubbs’ method (α=0.0001). Notably, there were several clinical factors that may have contributed to the abnormally elevated serum DLK1 level seen in this patient. This sample was not obtained at the initial presentation, but rather after numerous resections and revisions for the same lesion (n>6), some of which were associated with infection and other complications. For these reasons and the proposed role of DLK1 in wound healing and the inflammatory response, this patient’s sample was excluded from further analysis [56-58]. Collectively, these findings support future, prospective studies to establish the sensitivity and specificity of serum DLK1 in predicting the risk of malignant transformation and early identification of MPNST in patients with NF1.

Overexpression of DLK1 is associated with the reactivation of embryonic gene signatures

Consistent with the findings from our institutional cohort, we noted that, again while TCGA-MPNST expressed significantly higher levels of DLK1 compared to normal nerve overall, TCGA-MPNST samples were also distinguishable by higher or lower levels of DLK1 expression (Figure 2A). Accordingly, principal component (Figure 2B) and unsupervised hierarchical clustering analysis (Supplemental Figure 3) revealed separation of two distinct subclusters defined by DLK1 expression (DLK1_Hi and DLK1_Lo). Differential gene expression (Supplemental Table 11) and GO Biological Process enrichment analysis comparing DLK1_Hi and DLK1_Lo tumors demonstrated significant upregulation of genes involved in developmental and embryonic processes in DLK1_Hi samples (Figure 2C). To further investigate these findings, we performed bulk RNA sequencing on tumors (n=10) obtained from patients cared for at our institution. Congruent with the above results, NF1-MPNST from our institutional cohort were also distinguishable by DLK1 expression (Figure 2D-F). Differential gene expression (Supplemental Table 12) and pathway enrichment analyses (Figure 2G) similarly revealed enrichment of pathways and processes involved in development and differentiation by the genes upregulated in DLK1_Hi tumors. Finally, publicly available bulk RNA sequencing data of MPNST acquired from the Gene Expression Omnibus (accession #: GSE178989) revealed similar results (Supplemental Table 13, Supplemental Figure 4A-D), underscoring the consistent upregulation of DLK1 and its association with the reactivation of embryogenic signatures in a subset of MPNST.

Figure 2. Overexpression of DLK1 is associated with the reactivation of embryonic gene signatures.

(A) Bar plot comparing DLK1 log₂ normalized counts between TCGA-MPNST (n=6) and GTEx normal nerve samples (n=278). Dots represent individual samples. Red dots indicate TCGA-MPNST with high DLK1 expression (DLK1_Hi), whereas blue dots indicate those with low (DLK1_Lo) expression. Error bars reflect the standard error of the mean (SEM). P-value represents unpaired, two-tailed t-tests between groups. (B) Principal component analysis (PCA) demonstrating global variation in gene expression between DLK1_Hi (red) and DLK1_Lo (blue) TCGA-MPNST samples based on principal components 1 and 2 (PC1 and PC2). (C) Lollipop plot generated in ShinyGO 0.80 depicting GO Biological Process 2023 enrichment of developmental and differentiation pathways by genes upregulated (adj. p-value≤0.05, log₂FC≥1) in DLK1_Hi compared to DLK1_Lo TCGA-MPNST. (D) Bar plot of DLK1 log₂ (CPM+4) normalized counts in MPNST obtained from patients cared for at our institution with high (DLK1_Hi, red) or low (DLK1_Lo, blue) expression of DLK1. Error bars reflect the SEM. P-value represents unpaired, two-tailed t-tests between groups. (E) Principal component analysis (PCA) demonstrating global variation in gene expression between DLK1_Hi (red) and DLK1_Lo (blue) MPNST samples from our institutional cohort based on principal components 1 and 2 (PC1 and PC2). (F) Heatmap depicting the expression pattern of the top 1000 variable genes sorted by 2-way, unsupervised hierarchical clustering. Columns represent individual genes, and their respective log₂ transformed, z-score normalized expression values across each sample in the data, with red corresponding to increased expression and blue to decreased expression according to the figure legend. DLK1 expression level is denoted by red and blue bars as indicated. Rows represent individual samples. (G) Lollipop plot generated in ShinyGO 0.80 depicting Go Biological Process 2023 enrichment of protein-coding genes (adj. p-value≤0.05, log₂FC≥1) upregulated in DLK1_Hi compared DLK1_Lo MPNST samples (n=10) from our institutional cohort.****= p-value ≤0.0001, ***=p-value ≤0.001 **= p-value≤0.01 *=p-value ≤ 0.05, ns=not significant.

DLK1 expression is increased in human MPNST cells characterized by a stem-like phenotype

During embryogenesis, DLK1 is known to regulate differentiation and the maintenance of stem cell pools, however, the precise mechanisms by which it does so remain incompletely understood [23]. Initially described as a Notch antagonist, emerging data now suggests that DLK1 may actually activate the Notch pathway in specific settings, and elevated levels of DLK1 have been observed in tissues exhibiting elevated Notch signaling [59-62]. Consistent with this observation, analysis of the Sciatic Nerve Atlas (SNAT) revealed that Notch1 expression exhibits a pattern reminiscent to that of Dlk1 (Supplemental Figure 1A), peaking at embryonic day 17.5 (E17.5) and gradually decreasing to its lowest level by postnatal day 60 (P60) (Figure 3A)[25]. Like DLK1, Notch signaling is required for the maintenance and proliferation of stem and progenitor cells within the nervous system [62]. Specifically, prior work by Yan et al., revealed that Notch signaling regulates key transcription factors involved in neural differentiation, including SOX11 [62]. Accordingly, we found SOX11 expression to be significantly upregulated in DLK1_Hi compared to DLK1_Lo tumors in all three of the previously described datasets (Figure 3B, Supplemental Figure 5A&B). Furthermore, linear regression analysis of SOX11 and DLK1 expression revealed a positive correlation between SOX11 and DLK1 in all three datasets (Pearson r=0.80-0.91) (Figure 3C, Supplemental Figure 5C&D).

Given the above enrichment of developmental gene signatures and the upregulation of SOX11 in DLK1_Hi tumors, we hypothesized that DLK1 would be increased in MPNST cells characterized by a stemness phenotype[63]. To test this hypothesis, we first confirmed expression of DLK1 in a panel of human MPNST cell lines (Figure 3D). MPNST cell lines (n=4) were then grown in low adhesion (LA) plates to allow for the enrichment of cells with a stem-like phenotype through sphere formation. Consistent with our hypothesis, DLK1 was significantly increased in cells grown under spheroid-promoting conditions (LA) in all lines tested (Figure 3E-G, Supplemental Figure 5E&F). Transfer of spheres back to standard, adherent conditions following three days of growth in low adhesion plates (LA) revealed that this elevated expression of DLK1 was maintained for at least 5 days post-transfer (Supplemental Figure 5G&H). The expression of SOX11 was also elevated in cells grown in spheroid promoting conditions (Figure 3H&I), suggesting a role for both DLK1 and SOX11 in the maintenance of MPNST cells with a stem-like phenotype.

Evaluation of the impact of DLK1 inhibition on spheroid forming ability was limited by the striking decrease in cell viability following siRNA-mediated knockdown in human MPNST cell lines (n=4) (Figure 3J&K, Supplemental Figure 5I-J). Importantly, a significant decrease in viability was observed in two MPNST cell lines generated from NF1 patients whose tumors were known to express high levels of DLK1 in vivo (Supplemental Figure 5K-N). This decrease in viability was accompanied by a significant decrease in the expression of both cleaved NOTCH1 and SOX11 (Figure 3L-O, Supplemental Figure 5O&P). We also noted a dramatic decrease in phospho-ERK following siRNA inhibition of DLK1 (Figure 3P). This is consistent with a previous report observing DLK1-mediated activation of ERK signaling prevents cellular differentiation[64]. Overexpression of DLK1 using a doxycycline-inducible lentiviral system resulted in a modest, yet significant increase in the proliferative rate of the human MPNST cell line ST-8814 (Supplemental Figure 6A-C). Alternatively, while exogenous overexpression of DLK1 did not appear to impact the proliferative rate of the human PNF cell lines, ipNF95.6 and ipNF05.5, colony forming assays revealed a significant increase in the number of spheroid colonies following doxycycline-induced overexpression of DLK1 in both cell lines (Supplemental Figure 6D-G). The acquisition of this three-dimensional growth capability was accompanied by a notable increase in the expression of SOX11 (Supplemental Figure 6H). Collectively, these findings validate our RNA sequencing analyses proposing DLK1 as marker for MPNST cells exhibiting a stem-like phenotype and suggest a novel role for DLK1 as a regulator of SOX11 expression and ERK signaling in MPNST.

DLK1_Hi MPNST are characterized by an immunosuppressive microenvironment

The re-activation of embryonic transcriptional programs by tumor cells is associated with the development of an immunosuppressive tumor microenvironment (TME) in other cancers[65]. Accordingly, DLK1_Hi tumors from all three datasets exhibited significant, positive enrichment of a stemness gene signature (Supplemental Table 14) shown to be associated with a poor anti-tumor, immune response (Figure 4A, Supplemental Figure 7A&B)[66]. Likewise, GO Biological Process 2023 enrichment analysis revealed broad suppression of genes involved in immune and inflammatory processes including cytokine production and immune cell activation among DLK1_Hi MPNST from all datasets (Figure 4B, Supplemental Tables 15 &16). Cell type profiling of tumors from our institutional cohort (n=10) using computational deconvolution methods and Immunostates [44] as the reference dataset, highlighted differences in inferred immune cell subtypes between DLK1_Hi and DLK1_Lo tumors. Compared to DLK1_Hi tumors, gene signatures consistent with M1 macrophages, myeloid dendritic cells, CD14+ monocytes and T cells were positively enriched in DLK1_Lo samples (Figure 4C). In line with these findings, we observed a significant reduction in the expression of CD14, CD4, CD86 and major histocompatibility complex class II (MHC-II) molecules in DLK1_Hi tumors from all three datasets (Supplemental Figure 7C-F, Supplemental Tables 11 & 13). Analysis of the spatially profiled MPNST initially presented in Figure 1E, produced similar results and revealed a decrease in the expression of macrophage-associated genes in clusters characterized by elevated DLK1 expression (Figure 4D). Specifically, expression of monocyte/macrophage markers CD68 and CD14 was suppressed in regions characterized by the highest levels of DLK1 expression (Figure 4E-H).

Figure 4. DLK1_Hi MPNST are characterized by an immunosuppressive microenvironment.

(A) Gene set enrichment plot depicting enrichment of a stemness gene signature by MPNST from our institutional cohort with high (DLK1_Hi) and low (DLK1_Lo) DLK1 expression. Black vertical bars indicate the rank of genes comprising each signature. The black curve corresponds to the running enrichment score for the gene set. Normalized enrichment score (NES) and q-value for DLK1_Hi compared to DLK1_Lo samples are as shown. (B) Bar plot generated using EnrichR and depicting Go Biological Process 2023 enrichment of downregulated protein-coding genes (adj. p-value ≤ 0.05, log2FC≤−1) acquired through iDEP.96. (C) Cell type mapping panel generated in Omics Playground representing inferred cell types in DLK1_Hi and DLK1_Lo samples from our institutional cohort. ImmunoStates served as the reference dataset. (D) Heatmap showing suppression of macrophage associated genes in clusters (bold) characterized by elevated levels of DLK1 expression. Rows represent individual genes, and their respective log₂ transformed, z-score normalized expression values across each cluster with red corresponding to increased expression and blue to decreased expression according to the figure legend. (E-H) Surface plots depicting the expression of macrophage/monocyte markers CD68 (G) and CD14 (H). Surface plot depicting cluster designations is shown in (E). DLK1 expression as shown previously in Figure 1D serves as a reference (F). (I,J) Geneset enrichment plot depicting enrichment of a tumor-associated antigen (TAA) (I) and anti-PD1 response (J) gene signature by MPNST from our institutional cohort. Black vertical bars indicate the rank of genes comprising each signature. The black curve corresponds to the running enrichment score for the gene set. NES and q-value for DLK1_Hi compared to DLK1_Lo samples are as shown.

In addition to promoting the development of an immunosuppressive TME, tumor cell acquisition of a stem-like phenotype has been implicated in driving resistance to immunotherapeutic targeting [65, 67, 68]. Geneset signature analysis revealed positive enrichment of tumor associated antigens (TAAs) (Supplemental Table 17) [66] by DLK1_Hi MPNST in all three datasets (Figure 4I, Supplemental Figure 7G&H), suggesting that the DLK1_Hi-associated immunosuppression is not due to decreased expression of TAAs. Prior work in glioma suggests that cancer cells with a stemness phenotype can also evade anti-tumor immune responses through the suppression of MHC-I expression or disruption of other proteins comprising the antigen-processing machinery (APM) [69]. Notably, the loss of MHC-I mediated antigen presentation has been shown to be associated with resistance to immunotherapy [70]. Consistent with these findings, we observed a significant decreased in the expression of genes encoding MHC-I molecules as well as key components of the APM (B2M, ERAP1, TAPBP) in DLK1_Hi MPNST from all three datasets (Supplemental Figure 7I and Supplemental Tables 11-13). In line with our findings, emerging data has begun to identify subsets of patients who may experience superior benefit from immunotherapeutic approaches[48-51]. Notably, geneset enrichment analysis of the Ayers signature, which has been shown to predict response to anti-PD-1 immunotherapy (Supplemental Table 18), revealed significant negative enrichment by DLK1_Hi tumors in all three datasets (Figure 4J, Supplemental Figure 8A&B), suggesting that DLK1_Hi tumors may be less amenable to anti-PD1 blockade than those characterized by low expression of DLK1[49, 52]. Alternatively, LINCS L1000 activity analysis suggested that patients with DLK1_Hi tumors may, instead, benefit from treatment with topoisomerase, histone deacetylase (HDAC), microtubule and cyclin-dependent kinase (CDK) inhibitors (Supplemental Figure 8C-E). Analysis of treatment response data from the Cancer Cell Line Encyclopedia[53] revealed a significantly lower IC₅₀ (mM) of HDAC inhibitors Belinostat and AR-42 in DLK1_Hi compared to DLK1 unaltered cell lines (Supplemental Figure 8F&G) [12]. Importantly, geneset enrichment analysis suggests that these DLK1_Hi cell lines exhibit a transcriptomic phenotype similar to that of DLK1_Hi MPNST as described above (Supplemental Figure 8H&I). Collectively, these results imply that DLK1 expression level may delineate tumor subsets characterized by distinct therapeutic vulnerabilities stemming from disparate tumor autonomous and microenvironmental phenotypes.

Spatial transcriptomic profiling reveals aberrant expression of DLK1 preceding histopathological evidence of malignancy

In addition to identifying therapeutic vulnerabilities in MPNST, the accurate stratification of benign PNST on the basis of risk for malignant transformation is pivotal for reducing morbidity and mortality in patients with NF1. This has remained a significant challenge as even PNST adhering to uniform histopathological criteria can exhibit substantial molecular heterogeneity and disparate biological behavior [8, 9]. To gain a better understanding of the molecular programs contributing to the malignant transformation of MPNST precursors, we performed spatial transcriptomic profiling of an MPNST-contiguous ANNUBP, which revealed a region of elevated DLK1 expression (Figure 1D, Supplemental Figure 9A). Compared to the surrounding DLK1_Lo tissue, this DLK1_Hi cluster (#5) was characterized by the suppression of genes involved in the immune response as well as those known to positively regulate differentiation (Figure 5A, Supplemental Table 19). Likewise, we also observed a parallel upregulation of genes involved in developmental processes within this DLK1_Hi cluster (#5) (Figure 5B, Supplemental Table 20).

Figure 5. Spatial transcriptomic profiling reveals aberrant expression of DLK1 preceding histopathological evidence of malignancy.

(A) Heatmap showing suppression of genes involved in the inflammatory response (GO:0006954) in the region of elevated DLK1 expression (cluster #5) of the MPNST-contiguous ANNUBP initially presented in Figure 1D. Rows represent individual genes, and their respective log₂ transformed, z-score normalized expression values across each cluster with red corresponding to increased expression and blue to decreased expression according to the figure legend. (B) Heatmap showing upregulation of genes involved in the nervous system development (GO:0007399) in the region of elevated DLK1 expression (cluster #5) of an MPNST-contiguous ANNUBP. Rows represent individual genes, and their respective log₂ transformed, z-score normalized expression values across each cluster with red corresponding to increased expression and blue to decreased expression according to the figure legend. (C-E) Surface plots depicting the expression of Schwann cell specific markers, S100B (D) and NES (E) in the MPNST-contiguous ANNUBP. Plot of DLK1 initially shown in Figure 1D serves as a reference (C). Arrow represents trajectory used to generate trajectory plots in K-O.(F-O) Surface and corresponding trajectory plots depicting the expression of genes associated with stem- and progenitor cell phenotypes including POSTN (F,K), GAP43 (G,L), NGFR (H,M), PTN (I,N) and PTPRZ1 (J,O). Yellow corresponds to increased expression and dark purple to decreased expression according to the figure legend. Grey indicates no expression. (P-S) Surface plots depicting the expression of genes involved in the inflammatory response including, IL-33 (P), CCL2 (Q), CCL5 (R), CCL14 (S) in the MPNST-contiguous ANNUBP. Yellow corresponds to increased expression and dark blue to decreased expression according to the figure legend. Grey indicates no expression. (T-W) Surface plots depicting the expression of T cell, CD2 (T), CD3E (U), CD8A (V), and macrophage/monocyte, CD68 (W) specific markers the MPNST-contiguous ANNUBP. Yellow corresponds to increased expression and dark blue to decreased expression according to the figure legend. Grey indicates no expression.

Surface plots of Schwann cell specific markers S100B, and NES revealed elevated, but not restricted, expression within the DLK1_Hi cluster, implicating Schwann cells as the source of DLK1 expression within this region (Figure 5C-E)[71-73]. Genes associated with stem- and progenitor cell phenotypes including POSTN, GAP43, NGFR, PTN and PTPRZ1, also exhibited elevated expression (Figure 5F-J) within the DLK1_Hi cluster and positive correlative trajectories (Figure 5K-O, Supplemental Figure 9B), suggesting that Schwann cells characterized by an immature phenotype predominate within the DLK1_Hi region [74-77]. Differential gene expression analysis comparing the DLK1_Hi cluster (#5) to the surrounding tissue revealed that expression of interleukin-33 (IL-33), was significantly increased within the DLK1_Hi cluster (Figure 5P). Importantly, IL-33 has been shown to promote the acquisition of stem-like characteristics by tumor cells while simultaneously promoting the development of an immunosuppressive tumor microenvironment through the induction of immune exhaustion [78-80]. Conversely, CCL2, CCL5 and CCL14 which drive the recruitment peripheral immune cells, exhibited little to no expression within the region of elevated DLK1 expression (Figure 5Q-S) [81-83]. Consistent with these findings, the expression of T and NK cell markers CD2, CD3E, CD8A was essentially absent within the DLK1_Hi region (Figure 5T-V). Likewise, expression of macrophage/monocyte marker CD68 was also suppressed within the DLK1_Hi cluster (Figure 5W). These findings are consistent with our above results and with previous studies suggesting that the aberrant re-activation of dedifferentiation programs by tumorigenic cells is associated with the development of an immunosuppressive TME [84]. Importantly, recent work by our group found ANNUBP to exhibit significant CD4⁺FOXP3⁻ and CD8⁺FOXP3⁻ T cell infiltration compared to both PNF and MPNST [9]. This effector T cell infiltration was significantly diminished in MPNST, suggesting that the malignant transformation of ANNUBP is associated with a shift from a “hot” to “cold” immune microenvironment [9, 85, 86]. Here, we provide evidence that aberrant overexpression of DLK1 in MPNST precursor lesions is associated with the acquisition of an immature phenotype and suppression of the anti-tumor immune response. Furthermore, these results suggest that overexpression of DLK1 precedes histopathological changes consistent with malignancy.

Overexpression of DLK1 occurs independently of PRC2 loss

DLK1 is encoded by one of three paternally expressed genes located within an imprinted cluster on chromosome 14q32[87]. Regulation of monoallelic DLK1 expression relies on differential DNA methylation and histone modification within an intergenic differentially methylated region (IG-DMR) located upstream of MEG3 (Figure 6A)[87]. Prior reports have suggested a role for PRC2 in regulating the imprinting and expression of genes within the DLK1-DIO3 locus [88, 89]. Specifically, PRC2 is required for the expression of micro- and long non-coding RNAs (MEG3, MEG8, MIRG) on the maternal allele [89]. Given the frequency of PRC2 loss in MPNST, it would be reasonable to attribute an increase in DLK1 expression to a loss of PRC2 however, we did not appreciate an exclusive trend of PRC2 deficiency in DLK1_Hi tumors in the TCGA dataset (Supplemental Table 21) nor in our institutional cohort (Supplemental Table 22) [13, 14, 90]. Additionally, the expression levels of SUZ12 and EED were not significantly different between DLK1_Hi and DLK1_Lo tumors in any of the previously analyzed datasets (Supplemental Tables 11-13). Reduced expression of MEG3 and MEG8, which would be expected in the setting of PRC2 loss, was also not appreciated in DLK1_Hi compared to DLK1_Lo samples in any of the datasets (Supplemental Figure 10A-F). Expansion of our analysis to include all samples in the TCGA Sarcoma dataset did not reveal a correlation between DLK1 expression and copy number alterations or mutations of PRC2 components SUZ12 (Pearson r=−0.0247, 95% CI [−0.1471 to 0.09841], p=0.6943) or EED (Pearson r=−0.0611, 95% CI [−0.1826 to 0.06220], p=0.3311) (Figure 6B&C). Integration of pathway, expression and copy number data via the PARADIGM algorithm failed to reveal a significant difference in PRC2 pathway activation between the two groups in the TCGA-MPNST dataset (Figure 6D). Accordingly, analysis of Illumina Infinium HumanMethylation450 data revealed no significant difference in the average methylation across the DLK1-DIO3 locus between DLK1_Hi and DLK1_Lo TCGA-MPNST (Figure 6E&F). With respect to specific CpG loci within DLK1, significant differences in methylation were noted in only 6 of 45 (13%) CpG loci (Supplemental Figure 10G-L). These alterations occurred irrespective of PRC2 status as they were present in the DLK1_Hi tumor lacking disruption of SUZ12 or EED (TCGA-QQ-A8VH-01), but not in the DLK1_Lo samples exhibiting copy number loss of EED (TCGA-RN-AAAQ-01, TCGA-SI-A71P-01). Consistent with these results, RNAseq analysis of PRC2-deficient human MPNST lines (n=6) revealed significant variability in DLK1 expression (Figure 6G) [89]. Importantly, differential gene expression (Supplemental Figure 10M) and pathway enrichment (Supplemental Figure 10N) analyses comparing PRC2-deficient MPNST cell lines with DLK1 expression above (DLK1_Hi) and below (DLK1_Lo) the mean, revealed molecular signatures consistent with results described above, suggesting that the phenotypic differences between DLK1_Hi and DLK1_Lo samples are not simply the result of PRC2 loss. Re-expression of SUZ12 and subsequent restoration of histone H3 at lysine 27 trimethylation (H3K27me3) in two human PRC2-deficient MPNST cell lines expression had no impact on DLK1 protein levels (Figure 6H). Finally, we also did not observe decreased expression of Suz12 or Eed in the murine MPNST precursor cells with elevated Dlk1 expression (Figure 6I). Collectively, these findings suggest that factors independent of PRC2 loss contribute to the elevated expression of DLK1 observed in MPNST.

While we did not appreciate an exclusive correlation between PRC2 loss and high DLK1 expression, we did observe that homozygous, combined loss of CDKN2A and S-methyl-5'-thioadenosine phosphorylase (MTAP) occurred exclusively in DLK1_Hi tumors from our institutional cohort (Figure 7A). Beyond CDKN2A and MTAP, DLK1_Hi tumors were characterized by loss of larger portions of 9p21.3, extending 2,421-5,072 kilobases. Importantly, in addition to MTAP and CDKN2A, this region of chromosome 9 is home to several genes believed to serve as tumor suppressors as well as the type 1 interferon gene cluster (Figure 7B). Confirmatory IHC of MPNST from our institutional cohort (n=13) revealed a significant decrease in MTAP expression in DLK1_Hi MPNST (Figure 7C&D). This finding is of particular clinical importance as the inhibition of protein arginine N-methyltransferase 5 (PRMT5) with MRTX1719 has demonstrated clinical efficacy in CDKN2A/MTAP deficient tumors [91, 92]. Given these findings, future studies evaluating the association of DLK1 overexpression with the dual loss of CDKN2A/MTAP in NF1-MPNST are warranted.

Figure 7. DLK1_Hi tumors are characterized by homozygous loss of CDKN2A and MTAP.

(A) Table depicting copy number alterations of CDKN2A and MTAP in MPNST (n=10) from our institutional cohort. −2 represents homozygous deletion and 0 represents neutral status as determined based on CNVkit thresholds. (B) Schematic adapted from cBioPortal depicting genes adjacent to CDKN2A and MTAP within the 9p21.3 locus. (C) Representative photomicrographs of serial MPNST sections from the institutional cohort (n=13, DLK1_Lo regions of interest (ROI)=62, DLK1_Hi ROI=78) immunohistochemically stained for DLK1 (top) and MTAP (bottom). Magnification is denoted by 500 μm scale bars with inset high-power magnification as shown. The HALO MTAP Stain 1 Mask is shown in the third row of panels and the cytonuclear mask used to quantify MTAP staining is shown in the bottom panel. (D) Bar plot depicting MTAP strong (3+) and moderate (2+) positive cells as a percentage of total cells per field. Dots represent individual ROIs. Error bars reflect standard error of the mean (SEM). P-value reflects unpaired, two-tailed students t-test between groups. ****= p-value ≤0.0001, ***=p-value ≤0.001 **= p-value≤0.01 *=p-value ≤ 0.05, ns=not significant.

Discussion

MPNST remains the leading cause of premature death among individuals with NF1 and our ability to accurately predict risk of PNST transformation or MPNST therapeutic response remains limited [93]. In the present study, we have shown that DLK1 is significantly increased in MPNST compared to benign neurofibroma and provide evidence that DLK1 overexpression may precede histological changes consistent with malignancy. In complementary analyses, we find that serum levels of DLK1 are significantly higher in both mice and humans harboring MPNST compared to those without malignancy. Importantly, while DLK1 expression is increased in MPNST overall, through the integration of multiple, independent datasets we demonstrate that divergent levels of DLK1 expression distinguish MPNST subsets characterized by unique molecular programs and potential therapeutic vulnerabilities. Specifically, we show that overexpression of DLK1 is associated with the reactivation of embryonic signatures, an immunosuppressive microenvironment and a worse overall survival in patients with NF1-MPNST. Collectively, these findings provide critical insight into MPNST tumorigenesis and support future, prospective trials evaluating the utility of DLK1 tissue and serum levels in augmenting diagnosis, risk assessment and therapeutic stratification in the setting of NF1-PNST.

Loss of CDKN2A/B has been implicated as key driver of PNST transformation to MPNST, but the utility of CDKN2A/B loss in risk assessment and prognostication has proven limited[11, 18, 94]. Prior work has demonstrated that both hetero- and homozygous CDKN2A/B loss can be observed within distinct regions of a single neurofibroma and p16^INK4A positivity has been observed within tumors known to harbor homozygous deletion of CDKN2A/B [10, 18, 94]. In addition to loss of CDKN2A/B, somatic mutations and copy number alterations involving members of the PRC2 complex (e.g. SUZ12 and EED) are also believed to promote the development of MPNST [13, 14]. However, like CDKN2A/B, PRC2 deficiency is also not detected in all MPNST [14]. Accordingly, emerging data has begun to highlight the importance of molecular-based stratification of MPNST into distinct subsets and work by the Genomics of MPNST (GeM) consortium has recently proposed the use of H3K27me3-based stratification [17, 20, 95]. While these findings provide critical insight into MPNST biology, emerging evidence suggests that divergent methylation profiles alone may not account for the unique molecular signatures exhibited by distinct MPNST subsets [17, 20, 95]. Specifically, work by Suppiah et al., found that while mutations and inactivating gene fusions involving members of the PRC2 complex were exclusive to one MPNST subset, the other subset still exhibited global hypomethylation [20]. Here we show that NF1-associated MPNST can be stratified into distinct subsets on the basis of DLK1 expression levels. These MPNST subsets exhibit opposing molecular signatures, with DLK1_Hi tumors characterized by the reactivation of gene programs involved in embryogenesis and development. Notably, we did not observe a direct association between PRC2 loss and DLK1 expression in any of the analyzed datasets and restoration of H3K27me3 via overexpression of SUZ12 in human MPNST cell lines did not suppress DLK1 protein levels. Collectively, our findings suggest that divergent DLK1 expression levels identify MPNST subsets with distinct molecular profiles that are independent of PRC2 and H3K27me3.

While the mechanisms underlying MPNST immunity remain incompletely understood, work by Høland et al., and others suggests that MPNST are distinguishable by divergent immunologic phenotypes [17, 20, 95]. In the present study, we demonstrate that DLK1_Hi MPNST are characterized by features consistent with impaired antigen presentation and diminished immune cell infiltration. Importantly, through spatial transcriptomic profiling of an MPNST-contiguous ANNUBP, we find evidence that aberrant DLK1 overexpression may precede histopathological signs of malignancy and that this is accompanied by the presence of a “cold” immune microenvironment. This agrees with prior work by our group demonstrating that evidence of immunosuppression and developmental regression in pre-cursor lesions signifies the inception of malignant transformation[9]. Future studies will be needed to elucidate the mechanisms by which DLK1 promotes the acquisition of an immunosuppressive microenvironment in MPNST, but early work in other settings suggests that DLK1 may be capable of directly suppressing the pro-inflammatory responses of infiltrating immune cells [58].

If validated prospectively, the findings presented in this study may have significant implications for biomarker-guided therapeutic interventions in the future. Preliminary evidence in gastrointestinal stromal tumors (GIST) suggests that levels of DLK1 tissue expression have prognostic value in predicting therapeutic response to surgery and adjuvant imatinib[96]. Consistent with this finding, we present evidence that contrasting expression levels of DLK1 may identify MPNST with unique therapeutic vulnerabilities. Specifically, DLK1_Hi MPNST may be more susceptible to treatment with drugs known to preferentially target cells with a stem-like phenotype and less amenable to immunotherapeutic agents, including anti-PD1 blockade [97, 98]. While clinical trials evaluating the efficacy of immune checkpoint blockade (ICB) in MPNST are currently underway (NCT04784247, NCT02834013, NCT04465643), these studies do not incorporate biomarker-guided pre-treatment stratification of patients based on MPNST subtype. Notably, DLK1 exhibits several characteristics that could position it as a promising biomarker in the setting of NF1-PNST. Firstly, the limited expression of DLK1 in healthy adult tissues allows for increased diagnostic specificity[23]. Importantly, here we have shown that elevated tissue expression of DLK1 is restricted to MPNST and neurofibroma associated with the development of, or that exhibited features concerning for, malignancy. Through spatial transcriptomic profiling we find Schwann cells to be the source of DLK1 expression, thereby allowing for the precise identification of tumor cells and those possessing malignant potential. Secondly, in addition to its membrane-bound form, cleavage by ADAM17/TACE can release DLK1 as a secreted form into the serum[23]. Accordingly, here we have shown that serum DLK1 is significantly increased in both mice and humans harboring MPNST compared to those without malignancy. While future, prospective studies are needed, our findings provide evidence that serum and tissue levels of DLK1 may hold promise for improving risk assessment, prognostication and therapeutic stratification of patients with NF1-MPNST.

While this study presents many novel findings, we acknowledge several limitations. Firstly, human tissue and serum samples included in this dataset were identified through retrospective review of patients not participating in a formal natural history study. As a result, the amount and type of clinical data available was variable, as were the treatment strategies employed. While NF1 is one of the most common cancer predisposition syndromes and affects approximately 1:2500 individuals worldwide, the sample sizes of the individual analyses presented throughout this study are modest. To address this limitation, we performed integrative analyses of multiple independent datasets, collectively consisting of over 80 tissue and 54 serum (human n=30, mouse n=24) samples, which revealed congruent results. Additionally, due to the inability to predict risk of PNST transformation, preemptive resection of ANNUBP is standard practice at many NF centers. This poses a limitation to the prospective validation of predictive biomarkers in all such studies. Finally, while this study identifies a novel role for DLK1 in NF1-MPNST, a thorough interrogation of the regulation and biological functions of DLK1 in MPNST pathogenesis is beyond the scope of this manuscript. Future studies will further elucidate the functional role of DLK1 and its regulatory mechanisms in NF1-associated PNST. Nonetheless, these data support the future evaluation of DLK1 as a novel biomarker for improved risk and therapeutic stratification in the setting of NF1-associated PNST and associated malignancies.

TRANSLATIONAL RELEVANCE.

In persons with neurofibromatosis type 1 (NF1) the lifetime risk of pre-existing neurofibroma (PNST) undergoing malignant transformation to MPNST is 8-16%. MPNST is the leading cause of premature death among individuals with NF1 and our ability to accurately predict risk of PNST transformation or MPNST therapeutic response remains limited. This study provides critical insight into MPNST tumorigenesis, identifying distinct MPNST subsets characterized by unique molecular programs and potential therapeutic vulnerabilities. Further, our findings provide support for the prospective evaluation of DLK1 tissue and serum levels in augmenting diagnosis, risk assessment and therapeutic stratification in the setting of NF1-PNST.