Preoperative and postoperative prediction of long-term meningioma outcomes.

BACKGROUND: Meningiomas are stratified according to tumor grade and extent of resection, often in isolation of other clinical variables. Here, we use machine learning (ML) to integrate demographic, clinical, radiographic and pathologic data to develop predictive models for meningioma outcomes. METHODS AND FINDINGS: We developed a comprehensive database containing information from 235 patients who underwent surgery for 257 meningiomas at a single institution from 1990 to 2015. The median follow-up was 4.3 years, and resection specimens were re-evaluated according to current diagnostic criteria, revealing 128 WHO grade I, 104 grade II and 25 grade III meningiomas. A series of ML algorithms were trained and tuned by nested resampling to create models based on preoperative features, conventional postoperative features, or both. We compared different algorithms' accuracy as well as the unique insights they offered into the data. Machine learning models restricted to preoperative information, such as patient demographics and radiographic features, had similar accuracy for predicting local failure (AUC = 0.74) or overall survival (AUC = 0.68) as models based on meningioma grade and extent of resection (AUC = 0.73 and AUC = 0.72, respectively). Integrated models incorporating all available demographic, clinical, radiographic and pathologic data provided the most accurate estimates (AUC = 0.78 and AUC = 0.74, respectively). From these models, we developed decision trees and nomograms to estimate the risks of local failure or overall survival for meningioma patients. CONCLUSIONS: Clinical information has been historically underutilized in the prediction of meningioma outcomes. Predictive models trained on preoperative clinical data perform comparably to conventional models trained on meningioma grade and extent of resection. Combination of all available information can help stratify meningioma patients more accurately.

Stringent comparative sequence analysis reveals SOX10 as a putative inhibitor of glial cell differentiation.

BACKGROUND: The transcription factor SOX10 is essential for all stages of Schwann cell development including myelination. SOX10 cooperates with other transcription factors to activate the expression of key myelin genes in Schwann cells and is therefore a context-dependent, pro-myelination transcription factor. As such, the identification of genes regulated by SOX10 will provide insight into Schwann cell biology and related diseases. While genome-wide studies have successfully revealed SOX10 target genes, these efforts mainly focused on myelinating stages of Schwann cell development. We propose that less-biased approaches will reveal novel functions of SOX10 outside of myelination. RESULTS: We developed a stringent, computational-based screen for genome-wide identification of SOX10 response elements. Experimental validation of a pilot set of predicted binding sites in multiple systems revealed that SOX10 directly regulates a previously unreported alternative promoter at SOX6, which encodes a transcription factor that inhibits glial cell differentiation. We further explored the utility of our computational approach by combining it with DNase-seq analysis in cultured Schwann cells and previously published SOX10 ChIP-seq data from rat sciatic nerve. Remarkably, this analysis enriched for genomic segments that map to loci involved in the negative regulation of gliogenesis including SOX5, SOX6, NOTCH1, HMGA2, HES1, MYCN, ID4, and ID2. Functional studies in Schwann cells revealed that: (1) all eight loci are expressed prior to myelination and down-regulated subsequent to myelination; (2) seven of the eight loci harbor validated SOX10 binding sites; and (3) seven of the eight loci are down-regulated upon repressing SOX10 function. CONCLUSIONS: Our computational strategy revealed a putative novel function for SOX10 in Schwann cells, which suggests a model where SOX10 activates the expression of genes that inhibit myelination during non-myelinating stages of Schwann cell development. Importantly, the computational and functional datasets we present here will be valuable for the study of transcriptional regulation, SOX protein function, and glial cell biology.

Tead1 regulates the expression of Peripheral Myelin Protein 22 during Schwann cell development.

Schwann cells are myelinating glia in the peripheral nervous system that form the myelin sheath. A major cause of peripheral neuropathy is a copy number variant involving the Peripheral Myelin Protein 22 (PMP22) gene, which is located within a 1.4-Mb duplication on chromosome 17 associated with the most common form of Charcot-Marie-Tooth Disease (CMT1A). Rodent models of CMT1A have been used to show that reducing Pmp22 overexpression mitigates several aspects of a CMT1A-related phenotype. Mechanistic studies of Pmp22 regulation identified enhancers regulated by the Sox10 (SRY sex determining region Y-box 10) and Egr2/Krox20 (Early growth response protein 2) transcription factors in myelinated nerves. However, relatively little is known regarding how other transcription factors induce Pmp22 expression during Schwann cell development and myelination. Here, we examined Pmp22 enhancers as a function of cell type-specificity, nerve injury and development. While Pmp22 enhancers marked by active histone modifications were lost or remodeled after injury, we found that these enhancers were permissive in early development prior to Pmp22 upregulation. Pmp22 enhancers contain binding motifs for TEA domain (Tead) transcription factors of the Hippo signaling pathway. We discovered that Tead1 and co-activators Yap and Taz are required for Pmp22 expression, as well as for the expression of Egr2 Tead1 directly binds Pmp22 and Egr2 enhancers early in development and Tead1 binding is induced during myelination, correlating with Pmp22 expression. The data identify Tead1 as a novel regulator of Pmp22 expression during development in concert with Sox10 and Egr2.

Haplotype-specific modulation of a SOX10/CREB response element at the Charcot-Marie-Tooth disease type 4C locus SH3TC2.

Loss-of-function mutations in the Src homology 3 (SH3) domain and tetratricopeptide repeats 2 (SH3TC2) gene cause autosomal recessive demyelinating Charcot-Marie-Tooth neuropathy. The SH3TC2 protein has been implicated in promyelination signaling through axonal neuregulin-1 and the ERBB2 Schwann cell receptor. However, little is known about the transcriptional regulation of the SH3TC2 gene. We performed computational and functional analyses that revealed two cis-acting regulatory elements at SH3TC2-one at the promoter and one ∼150 kb downstream of the transcription start site. Both elements direct reporter gene expression in Schwann cells and are responsive to the transcription factor SOX10, which is essential for peripheral nervous system myelination. The downstream enhancer harbors a single-nucleotide polymorphism (SNP) that causes an ∼80% reduction in enhancer activity. The SNP resides directly within a predicted binding site for the transcription factor cAMP response element binding protein (CREB), and we demonstrate that this regulatory element binds to CREB and is activated by CREB expression. Finally, forskolin induces Sh3tc2 expression in rat primary Schwann cells, indicating that SH3TC2 is a CREB target gene. These findings prompted us to determine if SNP genotypes at SH3TC2 are associated with differential phenotypes in the most common demyelinating peripheral neuropathy, CMT1A. Interestingly, this revealed several associations between SNP alleles and disease severity. In summary, our data indicate that SH3TC2 is regulated by the transcription factors CREB and SOX10, define a regulatory SNP at this disease-associated locus and reveal SH3TC2 as a candidate modifier locus of CMT disease phenotypes.

Structural effects of linkage disequilibrium on the transcriptome.

A majority of SNPs (single nucleotide polymorphisms) map to noncoding and intergenic regions of the genome. Noncoding SNPs are often identified in genome-wide association studies (GWAS) as strongly associated with human disease. Two such disease-associated SNPs in the 5' UTR of the human FTL (Ferritin Light Chain) gene are predicted to alter the ensemble of structures adopted by the mRNA. High-accuracy single nucleotide resolution chemical mapping reveals that these SNPs result in substantial changes in the structural ensemble in agreement with the computational prediction. Furthermore six rescue mutations are correctly predicted to restore the mRNA to its wild-type ensemble. Our data confirm that the FTL 5' UTR is a "RiboSNitch," an RNA that changes structure if a particular disease-associated SNP is present. The structural change observed is analogous to that of a bacterial Riboswitch in that it likely regulates translation. These data further suggest that specific pairs of SNPs in high linkage disequilibrium (LD) will form RNA structure-stabilizing haplotypes (SSHs). We identified 484 SNP pairs that form SSHs in UTRs of the human genome, and in eight of the 10 SSH-containing transcripts, SNP pairs stabilize RNA protein binding sites. The ubiquitous nature of SSHs in the transcriptome suggests that certain haplotypes are conserved to avoid RiboSNitch formation.