Classification of Diffuse Glioma Subtype from Clinical-Grade Pathological Images Using Deep Transfer Learning.

Diffuse gliomas are the most common primary brain tumors and they vary considerably in their morphology, location, genetic alterations, and response to therapy. In 2016, the World Health Organization (WHO) provided new guidelines for making an integrated diagnosis that incorporates both morphologic and molecular features to diffuse gliomas. In this study, we demonstrate how deep learning approaches can be used for an automatic classification of glioma subtypes and grading using whole-slide images that were obtained from routine clinical practice. A deep transfer learning method using the ResNet50V2 model was trained to classify subtypes and grades of diffuse gliomas according to the WHO's new 2016 classification. The balanced accuracy of the diffuse glioma subtype classification model with majority voting was 0.8727. These results highlight an emerging role of deep learning in the future practice of pathologic diagnosis.

MicroRNA-29a activates a multi-component growth and invasion program in glioblastoma.

BACKGROUND: Glioblastoma is a malignant brain tumor characterized by rapid growth, diffuse invasion and therapeutic resistance. We recently used microRNA expression profiles to subclassify glioblastoma into five genetically and clinically distinct subclasses, and showed that microRNAs both define and contribute to the phenotypes of these subclasses. Here we show that miR-29a activates a multi-faceted growth and invasion program that promotes glioblastoma aggressiveness. METHODS: microRNA expression profiles from 197 glioblastomas were analyzed to identify the candidate miRNAs that are correlated to glioblastoma aggressiveness. The candidate miRNA, miR-29a, was further studied in vitro and in vivo. RESULTS: Members of the miR-29 subfamily display increased expression in the two glioblastoma subclasses with the worst prognoses (astrocytic and neural). We observed that miR-29a is among the microRNAs that are most positively-correlated with PTEN copy number in glioblastoma, and that miR-29a promotes glioblastoma growth and invasion in part by targeting PTEN. In PTEN-deficient glioblastoma cells, however, miR-29a nevertheless activates AKT by downregulating the metastasis suppressor, EphB3. In addition, miR-29a robustly promotes invasion in PTEN-deficient glioblastoma cells by repressing translation of the Sox4 transcription factor, and this upregulates the invasion-promoting protein, HIC5. Indeed, we identified Sox4 as the most anti-correlated predicted target of miR-29a in glioblastoma. Importantly, inhibition of endogenous miR-29a decreases glioblastoma growth and invasion in vitro and in vivo, and increased miR-29a expression in glioblastoma specimens correlates with decreased patient survival. CONCLUSIONS: Taken together, these data identify miR-29a as a master regulator of glioblastoma growth and invasion.

Hybrid-Aware Model for Senior Wellness Service in Smart Home.

Smart home technology with situation-awareness is important for seniors to improve safety and security. With the development of context-aware computing, wearable sensor technology, and ubiquitous computing, it is easier for seniors to manage their health problem in smart home environment. For monitoring senior activity in smart home, wearable, and motion sensors-such as respiration rate (RR), electrocardiography (ECG), body temperature, and blood pressure (BP)-were used for monitoring movements of seniors. For context-awareness, environmental sensors-such as gas, fire, smoke, dust, temperature, and light sensors-were used for senior location data collection. Based on senior activity, senior health status can be classified into positive and negative. Based on senior location and time, senior safety is classified into safe and emergency. In this paper, we propose a hybrid inspection service middleware for monitoring elderly health risk based on senior activity and location. This hybrid-aware model for the detection of abnormal status of seniors has four steps as follows: (1) data collection from biosensors and environmental sensors; (2) monitoring senior location and time of stay in each location using environmental sensors; (3) monitoring senior activity using biometric data; finally, (4) expectation-maximization based decision-making step recommending proper treatment based on a senior health risk ratio.

Kazinol-E is a specific inhibitor of ERK that suppresses the enrichment of a breast cancer stem-like cell population.

Growing evidence shows that cancer stem-like cells (CSLCs) contribute to breast cancer recurrence and to its resistance to conventional therapies. The extracellular signal-regulated kinase (ERK) signaling pathway is a major determinant in the control of diverse cellular processes, including the maintenance of CSLCs. In this study, we found that Kazinol-E, an antioxidant flavan from Broussonetia kazinoki, decreased the CSLC population of a breast cancer cell line, MCF7. The CSLC population, characterized by CD44 high/CD24 low expression or by high Aldehyde dehydrogenase 1 activity, was decreased by a concentration of Kazinol-E that did not affect the growth of bulk-cultured MCF7 cells. Kazinol-E did not decrease EGF-induced ERK phosphorylation in CSLCs, but did block the phosphorylation of an ERK substrate, p90RSK2, at Thr359/Ser363. We further demonstrated that EGF-induced ERK activity was blocked by Kazinol-E in a wild-type K-Ras-expressing non-small cell lung cancer cell line H226B. An in vitro kinase assay with purified ERK1 and p90RSK2 as its substrate demonstrated a direct inhibition of ERK activity by Kazinol E. Additionally, a the molecular docking study provided putative binding modes of Kazinol-E into the ATP binding pocket of ERK1 Collectively, these results suggest that Kazinol-E is a direct inhibitor of ERK1, and more studies are warranted to develop this reagent for therapeutic breast CSLC targeting.

Evodiamine selectively targets cancer stem-like cells through the p53-p21-Rb pathway.

In spite of the recent improvements, the resistance to chemotherapy/radiotherapy followed by relapse is the main hurdle for the successful treatment of breast cancer, a leading cause of death in women. A small population of breast cancer cells that have stem-like characteristics (cancer stem-like cells; CSLC) may contribute to this resistance and relapse. Here, we report on a component of a traditional Chinese medicine, evodiamine, which selectively targets CSLC of breast cancer cell lines MCF7 and MDAMB 231 at a concentration that does show a little or no cytotoxic effect on bulk cancer cells. While evodiamine caused the accumulation of bulk cancer cells at the G2/M phase, it did not hold CSLC in a specific cell cycle phase but instead, selectively killed CSLC. This was not due to the culture of CSLC in suspension or without FBS. A proteomic analysis and western blotting revealed that evodiamine changed the expression of cell cycle regulating molecules more efficiently in CSLC cells than in bulk cancer cells. Surprisingly, evodiamine selectively activated p53 and p21 and decreased inactive Rb, the master molecules in G1/S checkpoint. These data collectively suggest a novel mechanism involving CSLC-specific targeting by evodiamine and its possible use to the therapy of breast cancer.

Cancer stem cell targeting: are we there yet?

Several decades after the first notion that cancer cells are heterogeneous, not only histologically but also at the level of tumorigenicity, the existence of a tumor initiation population (so-called cancer stem cells) is now widely accepted. Here, we review the historical background leading up to the identification of this special cancer cell population and its role in the resistance of tumors to conventional therapies. In addition, we briefly review the clinical targeting strategy of the specific signaling network for cancer stem cells. Current clinical trials with newly developed drugs or older drugs for other uses are summarized to provide an understanding of the current status of strategies to target cancer stem cells.

Systemic approaches identify a garlic-derived chemical, Z-ajoene, as a glioblastoma multiforme cancer stem cell-specific targeting agent.

Glioblastoma multiforme (GBM) is one of the most common brain malignancies and has a very poor prognosis. Recent evidence suggests that the presence of cancer stem cells (CSC) in GBM and the rare CSC subpopulation that is resistant to chemotherapy may be responsible for the treatment failure and unfavorable prognosis of GBM. A garlic-derived compound, Z-ajoene, has shown a range of biological activities, including anti-proliferative effects on several cancers. Here, we demonstrated for the first time that Z-ajoene specifically inhibits the growth of the GBM CSC population. CSC sphere-forming inhibition was achieved at a concentration that did not exhibit a cytotoxic effect in regular cell culture conditions. The specificity of this inhibitory effect on the CSC population was confirmed by detecting CSC cell surface marker CD133 expression and biochemical marker ALDH activity. In addition, stem cell-related mRNA profiling and real-time PCR revealed the differential expression of CSC-specific genes, including Notch, Wnt, and Hedgehog, upon treatment with Z-ajoene. A proteomic approach, i.e., reverse-phase protein array (RPPA) and Western blot analysis, showed decreased SMAD4, p-AKT, 14.3.3 and FOXO3A expression. The protein interaction map (http://string-db.org/) of the identified molecules suggested that the AKT, ERK/p38 and TGFß signaling pathways are key mediators of Z-ajoene's action, which affects the transcriptional network that includes FOXO3A. These biological and bioinformatic analyses collectively demonstrate that Z-ajoene is a potential candidate for the treatment of GBM by specifically targeting GBM CSCs. We also show how this systemic approach strengthens the identification of new therapeutic agents that target CSCs.

C. elegans RNA-binding protein GLD-1 recognizes its multiple targets using sequence, context, and structural information to repress translation.

Caenorhabditis elegans GLD-1, a maxi-KH motif containing RNA-binding protein, has various functions mainly during female germ cell development, suggesting that it likely controls the expression of a selective group of maternal mRNAs. To gain an insight into how GLD-1 specifically recognizes these mRNA targets, we identified 38 biochemically proven GLD-1 binding regions from multiple mRNA targets that are among over 100 putative targets co-immunoprecipitated with GLD-1. The sequence information of these regions revealed three over-represented and phylogenetically conserved sequence motifs. We found that two of the motifs, one of which is novel, are important for GLD-1 binding in several GLD-1 binding regions but not in other regions. Further analyses indicate that the importance of one of the sequence motifs is dependent on two aspects: (1) surrounding sequence information, likely acting as an accessory feature for GLD-1 to efficiently select the sequence motif and (2) RNA secondary structural environment where the sequence motif resides, which likely provides "binding-site accessibility" for GLD-1 to effectively recognize its targets. Our data suggest some mRNAs recruit GLD-1 by a distinct mechanism, which involves more than one sequence motif that needs to be embedded in the correct context and structural environment.

Numb regulates glioma stem cell fate and growth by altering epidermal growth factor receptor and Skp1-Cullin-F-box ubiquitin ligase activity.

Glioblastoma contains a hierarchy of stem-like cancer cells, but how this hierarchy is established is unclear. Here, we show that asymmetric Numb localization specifies glioblastoma stem-like cell (GSC) fate in a manner that does not require Notch inhibition. Numb is asymmetrically localized to CD133-hi GSCs. The predominant Numb isoform, Numb4, decreases Notch and promotes a CD133-hi, radial glial-like phenotype. However, upregulation of a novel Numb isoform, Numb4 delta 7 (Numb4d7), increases Notch and AKT activation while nevertheless maintaining CD133-hi fate specification. Numb knockdown increases Notch and promotes growth while favoring a CD133-lo, glial progenitor-like phenotype. We report the novel finding that Numb4 (but not Numb4d7) promotes SCF(Fbw7) ubiquitin ligase assembly and activation to increase Notch degradation. However, both Numb isoforms decrease epidermal growth factor receptor (EGFR) expression, thereby regulating GSC fate. Small molecule inhibition of EGFR activity phenocopies the effect of Numb on CD133 and Pax6. Clinically, homozygous NUMB deletions and low Numb mRNA expression occur primarily in a subgroup of proneural glioblastomas. Higher Numb expression is found in classical and mesenchymal glioblastomas and correlates with decreased survival. Thus, decreased Numb promotes glioblastoma growth, but the remaining Numb establishes a phenotypically diverse stem-like cell hierarchy that increases tumor aggressiveness and therapeutic resistance.

DNA methylation patterns of ulcer-healing genes associated with the normal gastric mucosa of gastric cancers.

Recent evidence suggests that gastric mucosal injury induces adaptive changes in DNA methylation. In this study, the methylation status of the key tissue-specific genes in normal gastric mucosa of healthy individuals and cancer patients was evaluated. The methylation-variable sites of 14 genes, including ulcer-healing genes (TFF1, TFF2, CDH1, and PPARG), were chosen from the CpG-island margins or non-island CpGs near the transcription start sites. The healthy individuals as well as the normal gastric mucosa of 23 ulcer, 21 non-invasive cancer, and 53 cancer patients were examined by semiquantitative methylation-specific polymerase chain reaction (PCR) analysis. The ulcer-healing genes were concurrently methylated with other genes depending on the presence or absence of CpG-islands in the normal mucosa of healthy individuals. Both the TFF2 and PPARG genes were frequently undermethylated in ulcer patients. The over- or intermediate-methylated TFF2 and undermethylated PPARG genes was more common in stage-1 cancer patients (71%) than in healthy individuals (10%; odds ratio [OR], 21.9) and non-invasive cancer patients (21%; OR, 8.9). The TFF2-PPARG methylation pattern of cancer patients was stronger in the older-age group (> or =55 yr; OR, 43.6). These results suggest that the combined methylation pattern of ulcer-healing genes serves as a sensitive marker for predicting cancer-prone gastric mucosa.

DNA methylation and expression patterns of key tissue-specific genes in adult stem cells and stomach tissues.

CpG-island margins and non-island-CpG sites round the transcription start sites of CpG-island-positive and -negative genes are methylated to various degrees in a tissue-specific manner. These methylation-variable CpG sites were analyzed to delineate a relationship between the methylation and transcription of the tissue-specific genes. The level of tissue-specific transcription was estimated by counting the number of the total transcripts in the SAGE (serial analysis of gene expression) database. The methylation status of 12 CpG-island margins and 21 non-island CpG sites near the key tissue-specific genes was examined in pluripotent stromal cells obtained from fat and bone marrow samples as well as in lineage-committed cells from marrow bulk, stomach, colon, breast, and thyroid samples. Of the 33 CpG sites examined, 10 non-island-CpG sites, but none of the CpG-island margins were undermethylated concurrent with tissue-specific expression of their nearby genes. The net methylation of the 33 CpG sites and the net amount of non-island-CpG gene transcripts were high in stomach tissues and low in stromal cells. The present findings suggest that the methylation of the non-island-CpG sites is inversely associated with the expression of the nearby genes, and the concert effect of transitional-CpG methylation is linearly associated with the stomach-specific genes lacking CpG-islands.

Clinical implication of recurrent copy number alterations in hepatocellular carcinoma and putative oncogenes in recurrent gains on 1q.

To elucidate the pathogenesis of hepatocellular carcinoma (HCC) and develop useful prognosis predictors, it is necessary to identify biologically relevant genomic alterations in HCC. In our study, we defined recurrently altered regions (RARs) common to many cases of HCCs, which may contain tumor-related genes, using whole-genome array-CGH and explored their associations with the clinicopathologic features. Gene set enrichment analysis was performed to investigate functional implication of RARs. On an average, 23.1% of the total probes were altered per case. Mean numbers of altered probes are significantly higher in high-grade, bigger and microvascular invasion (MVI) positive tumors. In total, 32 RARs (14 gains and 18 losses) were defined and 4 most frequent RARs are gains in 1q21.1-q32.1 (64.5%), 1q32.1-q44 (59.2%), 8q11.21-q24.3 (48.7%) and a loss in 17p13.3-p12 (51.3%). Through focusing on RARs, we identified genes and functional pathways likely to be involved in hepatocarcinogenesis. Among genes in the recurrently gained regions on 1q, expression of KIF14 and TPM3 was significantly increased, suggesting their oncogenic potential in HCC. Some RARs showed the significant associations with the clinical features. Especially, the recurrent loss in 9p24.2-p21.1 and gain in 8q11.21-q24.3 are associated with the high tumor grade and MVI, respectively. Functional analysis showed that cytokine receptor binding and defense response to virus pathways are significantly enriched in high grade-related RARs. Taken together, our results and the strategy of analysis will help to elucidate pathogenesis of HCC and to develop biomarkers for predicting behaviors of HCC.

PathCluster: a framework for gene set-based hierarchical clustering.

MOTIVATION: Gene clustering and gene set-based functional analysis are widely used for the analysis of expression profiles. The development of a comprehensive method jointly combining the two methods would allow for greater biological insights. RESULTS: We developed a software package, PathCluster for gene set-based clustering via an agglomerative hierarchical clustering algorithm. The distances between predefined gene sets are illustrated in a dendrogram in which the relationships between gene sets can be visually assessed. Valuable biological insights can be obtained according to the type of gene sets, e.g. coordinated action of molecular functions (functional gene sets) and putative motif synergy (promoter gene set) in a biological process. The combined use of gene sets further enables the interrogation of different biological themes and their putative relationships, such as function-versus-regulatory motif or drug-versus-function. PathCluster can also be used for knowledge-based sample partitioning or class categorization for clinical purposes. With extended applicability, PathCluster will facilitate the gleaning of meaningful biological insights and testable hypotheses in the contexts of given expression profiles. AVAILABILITY: PathCluster executable files can be freely downloaded at http://www.systemsbiology.co.kr/PathCluster/.

Chromosomal losses are associated with hypomethylation of the gene-control regions in the stomach with a low number of active genes.

Transitional-CpG methylation between unmethylated promoters and nearby methylated retroelements plays a role in the establishment of tissue-specific transcription. This study examined whether chromosomal losses reducing the active genes in cancers can change transitional-CpG methylation and the transcription activity in a cancer-type-dependent manner. The transitional-CpG sites at the CpG-island margins of nine genes and the non-island-CpG sites round the transcription start sites of six genes lacking CpG islands were examined by methylation-specific polymerase chain reaction (PCR) analysis. The number of active genes in normal and cancerous tissues of the stomach, colon, breast, and nasopharynx were analyzed using the public data in silico. The CpG-island margins and non-island CpG sites tended to be hypermethylated and hypomethylated in all cancer types, respectively. The CpG-island margins were hypermethylated and a low number of genes were active in the normal stomach compared with other normal tissues. In gastric cancers, the CpG-island margins and non-island-CpG sites were hypomethylated in association with high-level chromosomal losses, and the number of active genes increased. Colon, breast, and nasopharyngeal cancers showed no significant association between the chromosomal losses and methylation changes. These findings suggest that chromosomal losses in gastric cancers are associated with the hypomethylation of the gene-control regions and the increased number of active genes.

GEAR: genomic enrichment analysis of regional DNA copy number changes.

UNLABELLED: We developed an algorithm named GEAR (genomic enrichment analysis of regional DNA copy number changes) for functional interpretation of genome-wide DNA copy number changes identified by array-based comparative genomic hybridization. GEAR selects two types of chromosomal alterations with potential biological relevance, i.e. recurrent and phenotype-specific alterations. Then it performs functional enrichment analysis using a priori selected functional gene sets to identify primary and clinical genomic signatures. The genomic signatures identified by GEAR represent functionally coordinated genomic changes, which can provide clues on the underlying molecular mechanisms related to the phenotypes of interest. GEAR can help the identification of key molecular functions that are activated or repressed in the tumor genomes leading to the improved understanding on the tumor biology. AVAILABILITY: GEAR software is available with online manual in the website, http://www.systemsbiology.co.kr/GEAR/.

Transitional CpG methylation between promoters and retroelements of tissue-specific genes during human mesenchymal cell differentiation.

In general, methylation of the promoter regions is inversely correlated with gene expression. The transitional CpG area between the promoter-associated CpG islands and the nearby retroelements is often methylated in a tissue-specific manner. This study analyzed the relationship between gene expression and the methylation of the transitional CpGs in two human stromal cells derived from the bone marrow (BMSC) and adipose tissue (ATSC), both of which have a multilineage differentiation potential. The transitional CpGs of the osteoblast-specific (RUNX2 and BGLAP), adipocyte-specific (PPARgamma2), housekeeping (CDKN2A and MLH1), and mesenchyme-unrelated (RUNX3) genes were examined by methylation-specific PCR. The expression of each gene was measured using reverse-transcription PCR analysis. The RUNX2, BGLAP, and CDKN2A genes in the BMSC, and the PPARgamma2 gene in the ATSC exhibited hypomethylation of the transitional CpGs along with the strong expression. The CpG island of RUNX3 gene not expressed in both BMSC and ATSC was hypermethylated. Transitional hypomethylation of the MLH1 gene was accompanied by the higher expression in the BMSC than in the ATSC. The weakly methylated CpGs of the PPARgamma2 gene in the BMSC became hypomethylated along with the strong expression during the osteoblastic differentiation. There were no notable changes in the transitional methylation and expression of the genes other than PPARgamma2 after the differentiation. Therefore, the transitional methylation and gene expression established in mesenchymal cells tend to be consistently preserved under the induction of differentiation. Weak transitional methylation of the PPARgamma2 gene in the BMSC suggests a methylation-dependent mechanism underlying the adiopogenesis of bone marrow.

The 5'-end transitional CpGs between the CpG islands and retroelements are hypomethylated in association with loss of heterozygosity in gastric cancers.

BACKGROUND: A loss of heterozygosity (LOH) represents a unilateral chromosomal loss that reduces the dose of highly repetitive Alu, L1, and LTR retroelements. The aim of this study was to determine if the LOH events can affect the spread of retroelement methylation in the 5'-end transitional area between the CpG islands and their nearest retroelements. METHODS: The 5'-transitional area of all human genes (22,297) was measured according to the nearest retroelements to the transcription start sites. For 50 gastric cancer specimens, the level of LOH events on eight cancer-associated chromosomes was estimated using the microsatellite markers, and the 5'-transitional CpGs of 20 selected genes were examined by methylation analysis using the bisulfite-modified DNA. RESULTS: The extent of the transitional area was significantly shorter with the nearest Alu elements than with the nearest L1 and LTR elements, as well as in the extragenic regions containing a higher density of retroelements than in the intragenic regions. The CpG islands neighbouring a high density of Alu elements were consistently hypomethylated in both normal and tumor tissues. The 5'-transitional methylated CpG sites bordered by a low density of Alu elements or the L1 and LTR elements were hypomethylated more frequently in the high-level LOH cases than in the low-level LOH cases. CONCLUSION: The 5'-transitional methylated CpG sites not completely protected by the Alu elements were hypomethylated in association with LOH events in gastric cancers. This suggests that an irreversible unbalanced decrease in the genomic dose reduces the spread of L1 methylation in the 5'-end regions of genes.

The length of CpG islands is associated with the distribution of Alu and L1 retroelements.

Alu and L1 retroelements have been suggested to initiate the spread of CpG methylation. In this study, the spread of CpG methylation was estimated based on the distance between the CpG islands and the nearest retroelements. All human genes (23,116) were examined and the correlations between the length of the CpG islands and the distance and density of the confronting retroelements were examined using nonoverlapping 5-kb windows. There was a linear relationship between the length of the CpG islands and the density of the Alu elements and an inverse relationship between the CpG islands and the L1 elements located more distantly, suggesting a suppressive effect of the Alu's on the spread of L1 methylation. Methylation analysis of the transitional CpG sites between the CpG islands and the nearest retroelements upstream of 16 genes was then carried out using DNA preparations from 11 different human tissues. Methylation-variable transitional CpGs were observed for the selected genes and the different tissues.

Alu and L1 retroelements are correlated with the tissue extent and peak rate of gene expression, respectively.

We exploited the serial analysis of gene expression (SAGE) libraries and human genome database in silico to correlate the breadth of expression (BOE; housekeeping versus tissue-specific genes) and peak rate of expression (PRE; high versus low expressed genes) with the density distribution of the retroelements. The BOE status is linearly associated with the density of the sense Alus along the 100 kb nucleotides region upstream of a gene, whereas the PRE status is inversely correlated with the density of antisense L1s within a gene and in the up- and downstream regions of the 0-10 kb nucleotides. The radial distance of intranuclear position, which is known to serve as the global domain for transcription regulation, is reciprocally correlated with the fractions of Alu (toward the nuclear center) and L1 (toward the nuclear edge) elements in each chromosome. We propose that the BOE and PRE statuses are related to the reciprocal distribution of Alu and L1 elements that formulate local and global expression domains.