Whole Exome Sequencing Reveals the Order of Genetic Changes during Malignant Transformation and Metastasis in a Single Patient with NF1-plexiform Neurofibroma

Neurofibromatosis type 1 (NF1) is one of the most common tumor predisposition syndromes, affecting as many as 1 in 2,500 people worldwide (1). Similar to other tumor predisposition syndromes, patients with NF1 are born with a germline mutation in one copy of the NF1 gene. Tumors arise following somatic loss of the other NF1 allele in the appropriate cell of origin. Among the tumor types most frequently encountered in this inherited cancer syndrome are benign and malignant tumors involving the peripheral nervous system, including benign dermal neurofibromas, benign plexiform neurofibromas, and malignant peripheral nerve sheath tumors (MPNST). Dermal neurofibromas are observed in nearly all adults with NF1, while plexiform neurofibromas and MPNSTs are found in 33% to 50% and 10% to 13% of individuals with NF1, respectively (2, 3). In contrast with dermal neurofibromas, which do not progress to cancer, plexiform neurofibromas can transform into MPNSTs (2).

MPNSTs represent an aggressive type of Schwann cell-derived neoplasm associated with poor overall patient survival. Currently, there are limited therapeutic options, and survival remains dismal following surgical removal, radiation therapy, and chemotherapy. Moreover, there are few clinicopathological factors that influence outcome, and no accurate biologic markers for disease progression. In contrast with other soft tissue sarcomas, there are no pathognomonic chromosomal translocations, and conventional cytogenetic analyses typically demonstrate complex karyotypes in MPNSTs.

In NF1-associated MNPST, bi-allelic inactivation of the NF1 gene has been reported in approximately 60% to 90% of tumors (4–6), whereas loss of NF1 protein (neurofibromin) expression is seen in nearly 90% of NF1-associated cases (7). Similarly, approximately 40% to 60% of sporadic MPNST harbor a somatic mutation in the NF1 gene (5, 6, 8) and 43% have loss of neurofibromin expression (7), supporting an important role for the NF1 tumor suppressor gene in the development of both sporadic and NF1-associated MPNSTs. Although NF1 loss is required for MPNST development, it is not sufficient for malignant transformation. As such, genetically engineered mice with conditional Nf1 gene inactivation in Schwann cell precursors do not develop MPNSTs (9–13): MPNST formation requires the acquisition of additional genetic mutations that cooperate with loss of NF1 gene function to promote malignant transformation. In mouse and human tumors, alterations in the TP53, CDKN2A, and EGFR genes contribute to MPNST pathogenesis (14, 15). Similarly, previous studies have used a variety of discovery platforms to identify other genes critical for human plexiform neurofibroma malignant progression (14, 16–23). Unfortunately, few of these genes are common across studies, and none have emerged as accurate predictors of disease pathogenesis.

In an effort to define the genetic alterations important for malignant progression of plexiform neurofibroma to MPNST, we used whole-exome sequencing technologies. Using a single patient with NF1 with a plexiform neurofibroma followed over a 14-year period, we established the sequence of genetic events associated with malignant transformation and progression (metastasis). Additional validation cases were subsequently used to further assess candidate genes involved in malignant progression. One of these mutated genes (β_III-spectrin) was found to be overexpressed in the majority of MPNSTs, and its silencing dramatically attenuated murine MPNST growth in vivo.

Blood, tumor, and other normal tissues were obtained from individuals diagnosed with NF1 according to established criteria (24) and treated at Washington University/St. Louis Children's Hospital NF Clinical Program (St. Louis, MO) or at the University of California San Francisco (San Francisco, CA) under active Human Studies Protocols approved by the Institutional Review Boards at each respective institution in accordance with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. Whole-exome sequencing or sequencing of target genes was performed on tumor and matched normal DNA. Tumor DNA samples were obtained from formalin-fixed paraffin-embedded (FFPE) blocks or frozen tissue (when available) obtained at surgical resection or biopsy. Normal paired DNA was isolated from peripheral blood obtained by phlebotomy or from adjacent normal tissue from paraffin-embedded blocks at surgical resection or biopsy on deceased patients for whom a peripheral blood sample could not be obtained.

Images of hematoxylin and eosin-stained slides from the discovery patient and validation patients were created on an Aperio ScanScope XT. The Aurora mScope viewer was used to view the images and obtain images at ×20 magnification. A pathologist (S.M. Dahiya or K. DeSchryver) assessed a digitized hematoxylin and eosin-stained slide to determine and outline tumor nests, thereby mapping the section, and directing the laser capture microdissection (LCM) to areas of irrefutably high tumor cell content on the adjacent FFPE sections. Five to 10 sections (10 μm thick) from each FFPE block were cut on a microtome (Leica RM 2255), air dried for 2 hours or longer, and placed at 60°C for 30 minutes. The sections were stored at 4°C until use or were immediately stained with a rapid hematoxylin and eosin ethanol-based staining protocol. Within 10 minutes of air-drying, the slide was placed on the LCM stage of Arcturus PixCell instrument for microdissection. The desired cells were microdissected into the cap of 500-μL safe-lock tubes filled with 50 μL ALT buffer (Qiagen). This procedure was repeated on the next slide until a total of 2,000 to 5,000 cells of interest had been captured for each case. The number of (adjacent) slides used per case to reach this 2,000-cell threshold varied from 5 to 10, as tumor proportion varied across samples.

LCM-derived DNA samples were isolated from 2,000 to 5,000 cells using a Qiamp DNA micro kit (Qiagen Cat. #56304) following RNase A treatment, according to the manufacturer's instructions. DNA was quantified using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific).

DNA was fragmented using a Covaris E210 DNA Sonicator (Covaris) with a size range between 100 bp and 400 bp. Libraries were prepared using a SciClone instrument (PerkinElmer). Short (8 bp) index sequences were utilized during library construction to enable sample multiplexing. For this work, two pools were created for capture: one pool with the FFPE samples and one with fresh-frozen sample, since the capture efficiencies between FFPE and frozen libraries are not compatible. Libraries were pooled (300 ng/library) for hybrid capture and individually processed using the Roche NimbleGen V3 Design Exome reagent (Roche Nimblegen) covering approximately 64 Mb of the human genome. The libraries were hybridized for 64 to 72 hours at 47°C followed by stringent washing. Enriched ssDNA library fragments were eluted and amplified prior to sequencing. Capture products were pooled and first run across two lanes of HiSeq 2000. Because of low cluster density on the original flow cell, an additional lane of HiSeq 2000 was also generated to achieve a minimum coverage of all targeted space, with >80% coverage at >20× coverage. Upon completion of the sequencing runs, data were processed through an automated data processing system to yield sample quality metrics as well as sequence variants relative to the human genome reference sequence.

Exome sequence data were aligned to the reference sequence build GRCh37-lite-build37 using bwa version 0.5.9 (ref. 25; params: -t 4 -q 5::), and then merged and duplicated using picard version 1.46 (https://broadinstitute.github.io/picard/). SNVs were detected using the union of three callers including: (i) Samtools version r963 (ref. 26; params: -A -B) filtered by snp-filter version v1 and intersected with Somatic Sniper version 1.0.2 (ref. 27; params: -F vcf -q 1 -Q 15) filtered by false-positive version v1 (params: –bam-readcount-version 0.4 –bam-readcount-min-base-quality 15) following by somatic-score-mapping-quality version v1 (params:–min-mapping-quality 40 –min-somatic-score 40), (ii) VarScan 2.2.6 (28) filtered by varscan-high-confidence version v1 and then false-positive version v1 (params: –bam-readcount-version 0.4 –bam-readcount-min-base-quality 15), and (iii) Strelka version 0.4.6.2 (ref. 29; params: isSkipDepthFilters = 1). Indels were detected using the union of three callers including: (i) Gatk-somatic-indel version 5336 (30) filtered by false-indel version v1 (params: –bam-readcount-version 0.4 –bam-readcount-min-base-quality 15), (ii) Pindel version 0.5 (31) filtered by pindel-somatic-calls version v1 followed by pindel-vaf-filter version v1 (params: –variant-freq-cutoff = 0.08) and pindel-read-support version v1, VarScan 2.2.6 (30) filtered by varscan-high-confidence-indel version v1 and false-indel version v1 (params: –bam-readcount-version 0.4 –bam-readcount-min-base-quality 15), and (iii) Strelka version 0.4.6.2 (ref. 29; params: isSkipDepthFilters = 1).

Of note, 250 ng aliquots were generated for library construction using the KAPA HTP sample prep kit (KAPA Biosystems). DNA was fragmented using a Covaris E210 DNA Sonicator (Covaris) with a size range between 100 bp and 400 bp. Libraries were prepared using a SciClone instrument (PerkinElmer). Short (8 bp) index sequences are utilized during library construction to enable sample multiplexing. A dual index approach is utilized in which unique sample indices are placed at each end of the ligated library fragment, providing a stringent sample match condition ensuring data is associated with the appropriate sample.

Libraries were pooled at equal molar concentrations of library. A total of 5 μg of library pool was hybridized using biotinylated 120 bp biotinylated oligo nucleotides (IDT lockdown probes, Integrated DNA Technologies) specific to the genes of interest based on the discovery phase of the project. Libraries were then hybridized for 72 hours at 47°C followed by stringent washing. Enriched ssDNA library fragments were eluted and amplified prior to sequencing. DNA sequencing was performed using one lane of a flow cell of 2 × 100 bp reads on an Illumina HiSeq 2000 sequencer.

Mapped reads were manually inspected to confirm the automated variant calls and frequency using IGV (Integrative Genomics Viewer). Review criteria required that (i) the variant was supported by at least 5% of the high-quality reads with a minimum of two reads supporting the variant, (ii) the variant was supported by reads in both sequencing directions, (iii) the variant base was high quality (brightly colored in display), (iv) the reads containing the variant had no more than two high-quality mismatches relative to the reference genome (excluding dnbsnps), and (v) the reads supporting the variant mapped uniquely to the reference genome. On the basis of these criteria, variants are either confirmed or denied.

Copy number variation (CNV) calls were made from the exome data using the Copy Change Assessment Tool 2 (CopyCAT2). Briefly, sequence data files (BAMs) for each case were used to calculate coverage for each bait position in the exome reagent. The resulting coverage data were normalized to a set of 12 unrelated normal control exomes sequenced to a similar coverage depth using the same capture reagent and sequencing chemistry using loess statistics. Variant allele frequencies were also calculated for SNPs within each gene. Segments of constant copy number were identified using circular binary segmentation, and confirmed by skewing of the variant allele frequencies. Data are then plotted to provide visualization of CNVs.

IHC staining was performed using rabbit polyclonal β_III-spectrin antibody (H-70; Santa Cruz Biotechnology; dilution 1:200) or goat polyclonal ZNF208 antibody (Santa Cruz Biotechnology; dilution 1:50) with citrate antigen retrieval. Previously generated tissue microarray blocks contained at least two cores of tissue (2 mm in diameter) from the most representative areas of tumor as well as from normal peripheral nerve (32).

Lentiviral Sptbn2 shRNA (sigma) or control LacZ shRNA (Genome Institute at Washington University) containing pLKO.1 plasmids was utilized for in vitro and in vivo MPNST tumor cell line growth experiments. Two Lentiviral Sptbn2 shRNA constructs were screened, but only one construct led to decreased levels of β_III-spectrin as determined by Western blotting and this construct was chosen for further experiments. The FUW-FLG lentiviral vector encoding firefly luciferase and GFP was utilized to generate tumor cell lines that could be followed in vitro and in vivo by bioluminescence imaging (BLI). Viral particles were produced by the Viral Vectors Core Facility of the Hope Center for Neurological Diseases at Washington University School of Medicine. See Supplementary Table S1.

Cell pellets were lysed in buffer containing 1% NP-40 (nonylphenoxypolyethoxylethanol), supplemented with protease inhibitors. Western blotting was performed as previously described (33) and according to manufacturer's instructions regarding antibody use with β_III-spectrin (H-70; Santa Cruz Biotechnology; dilution 1:200) and α-tubulin (Sigma; dilution 1:10000).

Murine MPNST tumor lines were established from male C57BL6/J Nf1+/−;Trp53+/−cis (NPcis) mice that inherited the mutant allele from the father (NPcis-paternal; refs. 34, 35). Briefly, palpable masses were divided into two sections for (i) histologic analysis and (ii) cell line generation. Tumor pieces were triturated in Triple Express (Gibco) and DNAse (Roche) in 24-well plates for 2 to 5 hours at 37°C in 5% CO₂, and successively passaged in DMEM with 10% FBS. Four lines were used in this study, including JW12 (10 × 5 × 8 mm jaw mass from a 153-day-old mouse), JW15 (15 × 9.5 × 8.5 mm leg mass from a 134 day-old mouse), JW18 (9.2 × 6 × 3 mm shoulder mass from a 99-day-old mouse), and JW23 (15 × 22 × 16 mm jaw mass from a 104-day-old mouse). JW18 cells were confirmed as MPNST by immunocytochemical analyses using S100 (1:1000 rabbit anti-cow; Dako), MyoD1 (1:50 mouse clone 5.8A; Dako), and Myf4 (1:30 mouse 1554; Novocastra) antibodies with appropriate secondary Alexa Fluor-499 antibodies (Invitrogen) and DAPI counterstaining. Cells were mounted in Vectashield, and photomicrographs acquired at ×40 magnification using SPOT 3.5.9 for MacOSX with a Nikon Eclipse E6000 camera. The RD cell line, derived from a human rhabdomyosarcoma (a kind gift from Javed Khan, NIH, Bethesda, MD) was used as a positive control for MyoD1 and Myf4 (Supplementary Fig. S1).

Murine MPNST cell lines were then transduced with green fluorescent protein (GFP)-luciferase viral supernatants as previously described (36, 37), and sorted by GFP expression using flow cytometry. Each cell line was plated at serial dilutions to ensure clonal derivation. Several clones from each cell line were screened for luciferase expression. The highest expressing clone from each cell line was expanded (JW18.2, JW12.1, JW15.3, JW23.3) for use.

Bioluminescence imaging (BLI) was performed on an IVIS 50 (PerkinElmer; Living Image 4.1, 1 sec–5 minutes exposures, binning 8, FOV 12 cm, f/stop1, open filter). In vivo bioluminescence was measured on days 5, 9, 12, and 16 posttumor cell injection following i.p. injection of D-luciferin (150 mg/kg; Gold Biotechnology, Inc.). For the in vitro assays, cells were plated in 24-well black-walled plates and imaged 24, 48, and 96 hours later, after the addition of D-luciferin (150 μg/mL). Total photon flux (photons/second) was measured from software-defined regions of interest using Living Image 2.6.

A total of 10⁵ JW18.2-shSptbn2 or JW18.2-shLacZ cells were plated in each of 6 wells of a 24-well plate. Cells were imaged by BLI and then manually counted at 24 hours, 48 hours, and 96 hours post-plating.

A total of 2 × 10⁵ JW18.2-shSptbn2 or JW18.2-shLacZ tumor cells were injected subcutaneously on the dorsal surface of anesthetized 5-week-old female C57BL6 mice (Taconic) and imaged at days 5, 9, 12, and 16 postinjection (n = 5 mice/group). The experiment was repeated independently, and similar results were obtained.

In vitro and in vivo tumor growth data were analyzed on GraphPad Prism Version 5.03. A two-way ANOVA analysis was used to determine statistical significance.

Whole-exome sequencing was performed on tumor samples and matched normal blood from one single individual with NF1 followed for over 14 years since childhood by a single physician (D.H. Gutmann) at Washington University. She was initially diagnosed using established clinical criteria (24) as a young child when she presented with multiple café-au-lait macules, skinfold freckling, and a mother with NF1. At age 12, she was initially diagnosed with an enlarging plexiform neurofibroma in her left hemipelvis (Plexiform 1), which was biopsied and demonstrated areas of nuclear pleomorphism worrisome for malignancy, prompting surgical removal for diagnostic reasons. Her plexiform neurofibroma recurred on three separate occasions: later that same year (at age 12), at age 14, and again at age 18 (Plexiform 2, Plexiform 3, and low-grade MPNST within Plexiform). The final plexiform neurofibroma specimen contained a few foci suspicious for low-grade MPNST based on increased cellularity, increased nuclear pleomorphism as well as presence of more than isolated mitosis per high power field. At age 25, she was diagnosed with a high-grade MPNST arising in the same location, which was surgically removed. One year later, at age 26, she developed widely metastatic disease, with lung metastases as well as retroperitoneal involvement. One right-sided retroperitoneal lesion was biopsied (Met), confirming metastatic disease. Histologic sections were evaluated by a pathologist (S.M. Dahiya), and laser capture microdissection was performed to obtain pure tumor samples. Representative images of these tumors are shown in Fig. 1A–F.

Samples were sequenced to achieve a minimum coverage of all targeted space at a minimum of >80% coverage with >20× coverage. In this individual, the germline NF1 gene mutation was identified (nonsense mutation; R440*), based on its presence in both the blood and tumor specimens (Fig. 1G). In addition, the somatic NF1 gene mutation was found to be a 4-bp deletion, leading to a frameshift mutation (K1661_fs; Fig. 1G). Using whole-exome sequencing, the mutant NF1 variant allele frequencies (VAF) were quantified in each of the tumors. An increase in the percentage of tumor cells harboring this somatic NF1 gene mutation was observed as the tumors evolved from benign plexiform lesions to primary and metastatic MPNST (Fig. 1H). This increase suggests a clonal evolution in the development of this malignancy. Because the DNA was extracted using laser capture microdissection, the samples obtained were relatively devoid of contaminating cells, as evidenced by VAF estimates of approximately 50% in the primary and metastatic MPNST lesions (presence of the somatic NF1 gene mutation in nearly all tumor cells). These findings are consistent with one prior report in which molecular analysis of a single MPNST divided into benign plexiform neurofibroma, atypical plexiform neurofibroma, and MPNST revealed 9%, 42%, and 97% loss of heterozygosity at the NF1 locus, respectively (38).

Next, we evaluated the possibility that large-scale somatic rearrangements in other genes cooperate with NF1 gene loss during the development of MPNSTs. Using copy number alteration analysis of whole-exome read data, few copy number alterations were detected in the benign precursor lesions. In striking contrast, a number of copy number alterations were observed in the primary and metastatic MPNST lesions (Supplementary Fig. S2 and Supplementary Table S2). Although some of these alterations have been observed in other studies (Supplementary Table S3; refs. 16, 39–42), most of these copy number alterations were found only in the primary MPNST from the discovery patient. Moreover, some of these alterations were only observed in the primary MPNST and not in the metastatic tumor (SPTBN2 and DSCC1).

To identify areas of overlap shared between the primary and metastatic MPNST, a high confidence event identified involved loss of an area of chromosome 17p containing the TP53 tumor suppressor gene (Fig. 2). Previous studies also support a causative role for loss of p53 function during malignant progression to MPNST (43). In further support of this notion, 30% to 50% of genetically engineered mice heterozygous for germline Nf1 and Tp53 mutations develop MPNSTs (44, 45). However in people, TP53 haploinsufficiency, rather than bi-allelic TP53 inactivation, may be sufficient for MPNST development (46, 47). As such, bi-allelic TP53 loss was only observed in 3 of 14 MPNSTs (46). It is also possible that other cell-cycle regulators (e.g., p16) are inactivated in cases where bi-allelic TP53 mutations are not observed; however, CDKN2A deletions were not observed in our patient, as reported in some other studies (21, 48–50). Interestingly, there was an area on chromosome 17 amplified in both the MPNST and metastatic lesion containing the BIRC5 and SOX9 genes (Fig. 2), both of which have been previously found in MPNSTs (17, 40, 51, 52). Although BIRC5 and SOX9 are involved in cell survival (53, 54), their role in MPNST pathogenesis remains to be defined.

In addition to TP53 loss, 42 somatic variants (including missense, nonsense, splice site, deletions, insertions, and silent mutations) were identified (Supplementary Table S4). Of these somatic variants, mutations found only in the benign tumors, silent mutations, and those resulting in conservative amino acid changes were excluded from further study. Following this filtering, nine genes were selected for resequencing at a greater read depth (>40× coverage) and after manual review, only seven genes remained (Table 1). Mutations in these seven genes have previously been reported in various malignancies in the cancer genome atlas (TCGA) database (Table 2; refs. 55, 56), underscoring their potential significance in tumorigenesis. Interestingly, we did not observe EGFR amplification in this specific MPNST case, as previously reported for other MPNSTs (21). Similarly, no copy number gains were identified involving chromosomes 12p12.1, 12p13.31, 12p13.33, or 12q14.1 (16). These results raise the intriguing possibility that there are multiple molecular routes to MPNST.

To determine whether these seven genes are mutated in other individuals with NF1-MPNST, directed sequencing was performed in six plexiform neurofibromas, fifteen primary MPNSTs, and five metastatic MPNSTs. While none of these genes were mutated in the plexiform neurofibroma tumors, 5 of the 7 genes were mutated in these independently identified MPNSTs (Fig. 3A). ZNF208 was the only gene uniquely mutated in the metastatic MPNST specimen (Table 1; Fig. 3A and Supplementary Table S5). ZNF208 is a member of a large family of zinc finger proteins containing Kruppel-associated box domains, which serve as transcriptional regulators (57). Unfortunately, there is a paucity of information about ZNF208 function in mammalian cells. It is worth noting that ZNF208 protein expression was detected by IHC in the metastatic tumor in this study as well as in one additional metastatic MPNST (data not shown). The lack of additional matched MPNST/metastatic lesion specimens with sufficient tissue for analysis precluded a more extensive analysis.

The most frequently mutated gene was the SPTBN2 gene encoding β_III-spectrin, a cytoskeletal protein (Fig. 3A and Supplementary Table S5). Different types of mutations were observed, including missense mutations (n = 5), silent mutations (n = 3), one mutation predicted to cause exon-skipping, and one mutation in the 5-prime untranslated region (Fig. 3B). In addition, one MPNST harbored several different mutations within the SPTBN2 gene. Given that this gene was the most frequently mutated gene in the MPNSTs examined (5/16 tumors), albeit at low variant allele frequencies, we next sought to determine whether deregulated β_III-spectrin expression was observed in these cancers. Robust β_III-spectrin immunoreactivity was found in all MPNSTs in the validation set (15/15) regardless of mutational status (Fig. 3F and G). Similarly, examination of additional MPNSTs, both sporadic and NF1-associated, from two previously generated tissue microarray revealed strong β_III-spectrin immunoreactivity in nearly all MPNSTs (45/50 tumors; 90% samples). In striking contrast, β_III-spectrin immunoreactivity was not observed in benign neurofibromas (0/4), schwannomas (0/4), or normal peripheral nerve (0/2; Fig. 3C–E).

Spectrins are cytoskeletal, heterodimeric proteins composed of α- and β subunits, with important roles in maintaining the stability of the cell membrane, promoting adhesion and spreading, regulating Golgi and vesicular transport, and controlling cell-cycle progression (58–61). Moreover, β_III-spectrin can enhance cell survival through maintenance of the Golgi apparatus and regulation of cell surface receptor vesicular transport, including glutamate receptors (60–62). In addition, a related spectrin molecule (β_II-spectrin; fodrin) interacts with the protein product (merlin/schwannomin) mutated in the Neurofibromatosis type II (NF2) Schwann cell cancer predisposition syndrome (63). On the basis of these observations, we speculate that the observed SPTBN2 mutations represent gain-of-function alterations.

To assess the potential functional significance of the increased β_III-spectrin found in the majority of human MPNSTs, we examined β_III-spectrin protein expression by IHC in four Nf1/Tp53-mutant NPcis MPNST cell lines engineered to express luciferase. Interestingly, the MPNST cell line with the greatest β_III-spectrin expression exhibited the fastest growth rate following subcutaneous transplantation into immunocompetent mice by bioluminescence imaging in vivo (JW18.2; data not shown). For this reason, JW18.2 was chosen for further study. Following Sptbn2 shRNA-mediated knockdown, there was reduced β_III-spectrin expression by Western blot analysis relative to control shLacZ virus infection (Fig. 4A). No difference in cell growth was observed following β_III-spectrin knockdown in vitro, as measured by BLI at 24, 48, and 96 hours (Fig. 4B) or direct cell counting (data not shown). However, β_III-spectrin knockdown attenuated tumor growth in vivo (Fig. 4C–E), supporting a role for this spectrin family member in MPNST maintenance. Collectively, these in vivo results coupled with the observation that MPNSTs harbor a high frequency of SPTBN2 mutation and β_III-spectrin immunoreactivity should prompt further mechanistic investigation into the role of this new MPNST-associated gene in cancer pathogenesis.

In summary, this study represents the first use of whole-exome sequencing to capture the molecular events that accompany the progression of a benign plexiform neurofibroma to malignancy and metastasis in a single individual. Although the sequence and nature of genetic alterations might be specific to a given individual, the discovery of β_III-spectrin as a frequently mutated gene and overexpressed protein in other MPNSTs underscores the value of this approach in identifying new targets for future study.

D.H. Gutmann is listed as an inventor on a patent on NF1 genes, which is owned by The University of Michigan, and mTOR, which is owned by Washington University. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A.C. Hirbe, R.S. Fulton, D.H. Gutmann

Development of methodology: A.C. Hirbe, R.S. Fulton, X. Zhang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.C. Hirbe, R.S. Fulton, S. McDonald, J. Walrath, K.M. Reilly, M. Pekmezci, A. Perry, S. Dahiya, K. DeSchryver

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.C. Hirbe, C.A. Miller, T. Li, R.S. Fulton, E.J. Duncavage, H.J. Abel, D.H. Gutmann

Writing, review, and/or revision of the manuscript: A.C. Hirbe, T. Li, R.S. Fulton, S. McDonald, E.J. Duncavage, M. Pekmezci, A. Perry, D.H. Gutmann, S. Dahiya, K. DeSchryver

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): X. Zhang, D.H. Gutmann

Study supervision: T.J. Ley

Other (manage tissue sample, perform LCM and DNA isloation): X. Zhang

The authors thank Drs. Brian A. Van Tine and Anthony J. Apicelli for helpful discussions, Scott Gianino, Tyler Smallman, and Julie Prior for technical assistance, and Heather Schmidt for manual review of predicted mutations.

This work was partly funded by a generous gift from Schnuck Markets Inc. (to D.H. Gutmann). A.C. Hirbe was supported on the T32 HL007088. BLI was performed at the Washington University Molecular Imaging Center, supported by NIH P50 CA094056. Lentivirus was generated by the Washington University Hope Center Viral Vectors Core, which is supported by a Neuroscience Blueprint Core grant NIH P30 NS057105.

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