In this paper, we address the scene segmentation task by capturing rich contextual dependencies based on the self-attention mechanism. Unlike previous works that capture contexts by multi-scale feature fusion, we propose a Dual Attention Network (DANet) to adaptively integrate local features with their global dependencies. Specifically, we append two types of attention modules on top of dilated FCN, which model the semantic interdependencies in spatial and channel dimensions respectively. The position attention module selectively aggregates the feature at each position by a weighted sum of the features at all positions. Similar features would be related to each other regardless of their distances. Meanwhile, the channel attention module selectively emphasizes interdependent channel maps by integrating associated features among all channel maps. We sum the outputs of the two attention modules to further improve feature representation which contributes to more precise segmentation results. We achieve new state-of-theart segmentation performance on three challenging scene segmentation datasets, i.e., Cityscapes, PASCAL Context and COCO Stuff dataset. In particular, a Mean IoU score of 81.5% on Cityscapes test set is achieved without using coarse data. 1 .

Tuberous sclerosis (TSC) is an autosomal dominant disorder characterized by the formation of hamartomas in a wide range of human tissues. Mutation in either the TSC1 or TSC2 tumour suppressor gene is responsible for both the familial and sporadic forms of this disease. TSC1 and TSC2 proteins form a physical and functional complex in vivo. Here, we show that TSC1-TSC2 inhibits the p70 ribosomal protein S6 kinase 1 (an activator of translation) and activates the eukaryotic initiation factor 4E binding protein 1 (4E-BP1, an inhibitor of translational initiation). These functions of TSC1-TSC2 are mediated by inhibition of the mammalian target of rapamycin (mTOR). Furthermore, TSC2 is directly phosphorylated by Akt, which is involved in stimulating cell growth and is activated by growth stimulating signals, such as insulin. TSC2 is inactivated by Akt-dependent phosphorylation, which destabilizes TSC2 and disrupts its interaction with TSC1. Our data indicate a molecular mechanism for TSC2 in insulin signalling, tumour suppressor functions and in the inhibition of cell growth.

Several proteins implicated in the pathogenesis of polycystic kidney disease (PKD) localize to cilia. Furthermore, cilia are malformed in mice with PKD with mutations in TgN737Rpw (encoding polaris). It is not known, however, whether ciliary dysfunction occurs or is relevant to cyst formation in PKD. Here, we show that polycystin-1 (PC1) and polycystin-2 (PC2), proteins respectively encoded by Pkd1 and Pkd2, mouse orthologs of genes mutated in human autosomal dominant PKD, co-distribute in the primary cilia of kidney epithelium. Cells isolated from transgenic mice that lack functional PC1 formed cilia but did not increase Ca(2+) influx in response to physiological fluid flow. Blocking antibodies directed against PC2 similarly abolished the flow response in wild-type cells as did inhibitors of the ryanodine receptor, whereas inhibitors of G-proteins, phospholipase C and InsP(3) receptors had no effect. These data suggest that PC1 and PC2 contribute to fluid-flow sensation by the primary cilium in renal epithelium and that they both function in the same mechanotransduction pathway. Loss or dysfunction of PC1 or PC2 may therefore lead to PKD owing to the inability of cells to sense mechanical cues that normally regulate tissue morphogenesis.

Multiple, complex molecular events characterize cancer development and progression1,2. Deciphering the molecular networks that distinguish organ-confined disease from metastatic disease may lead to the identification of critical biomarkers for cancer invasion and disease aggressiveness. Although gene and protein expression have been extensively profiled in human tumors, little is known about the global metabolomic alterations that characterize neoplastic progression. Using a combination of high throughput liquid and gas chromatography-based mass spectrometry, we profiled more than 1126 metabolites across 262 clinical samples related to prostate cancer (42 tissues and 110 each of urine and plasma). These unbiased metabolomic profiles were able to distinguish benign prostate, clinically localized prostate cancer, and metastatic disease. Sarcosine, an N-methyl derivative of the amino acid glycine, was identified as a differential metabolite that was highly elevated during prostate cancer progression to metastasis and can be detected non-invasively in urine. Sarcosine levels were also elevated in invasive prostate cancer cell lines relative to benign prostate epithelial cells. Knockdown of glycine-N-methyl transferase (GNMT), the enzyme that generates sarcosine from glycine, attenuated prostate cancer invasion. Addition of exogenous sarcosine or knockdown of the enzyme that leads to sarcosine degradation, sarcosine dehydrogenase (SARDH), induced an invasive phenotype in benign prostate epithelial cells. Androgen receptor and the ERG gene fusion product coordinately regulate components of the sarcosine pathway. Taken together, we profiled the metabolomic alterations of prostate cancer progression revealing sarcosine as a potentially important metabolic intermediary of cancer cell invasion and aggressivity.