Patient-Specific Semi-Supervised Learning for Postoperative Brain Tumor Segmentation

In contrast to preoperative brain tumor segmentation, the problem of postoperative brain tumor segmentation has been rarely approached so far. We present a fully-automatic segmentation method using multimodal magnetic resonance image data and patient-specific semi-supervised learning. The idea behind our semi-supervised approach is to effectively fuse information from both pre- and postoperative image data of the same patient to improve segmentation of the postoperative image. We pose image segmentation as a classification problem and solve it by adopting a semi-supervised decision forest. The method is evaluated on a cohort of 10 high-grade glioma patients, with segmentation performance and computation time comparable or superior to a state-of-the-art brain tumor segmentation method. Moreover, our results confirm that the inclusion of preoperative MR images lead to a better performance regarding postoperative brain tumor segmentation.

Interactive Segmentation of MR Images from Brain Tumor Patients

Medical doctors often do not trust the result of fully automatic segmentations because they have no possibility to make corrections if necessary. On the other hand, manual corrections can introduce a user bias. In this work, we propose to integrate the possibility for quick manual corrections into a fully automatic segmentation method for brain tumor images. This allows for necessary corrections while maintaining a high objectiveness. The underlying idea is similar to the well-known Grab-Cut algorithm, but here we combine decision forest classification with conditional random field regularization for interactive segmentation of 3D medical images. The approach has been evaluated by two different users on the BraTS2012 dataset. Accuracy and robustness improved compared to a fully automatic method and our interactive approach was ranked among the top performing methods. Time for computation including manual interaction was less than 10 minutes per patient, which makes it attractive for clinical use.

Novel Prodrug-Like Fusion Toxin with Protease-Sensitive Bioorthogonal PEGylation for Tumor Targeting

Highly potent biotoxins like Pseudomonas exotoxin A (ETA) are attractive payloads for tumor targeting. However, despite replacement of the natural cell-binding domain of ETA by tumor-selective antibodies or alternative binding proteins like designed ankyrin repeat proteins (DARPins) the therapeutic window of such fusion toxins is still limited by target-independent cellular uptake, resulting in toxicity in normal tissues. Furthermore, the strong immunogenicity of the bacterial toxin precludes repeated administration in most patients. Site-specific modification to convert ETA into a prodrug-like toxin which is reactivated specifically in the tumor, and at the same time has a longer circulation half-life and is less immunogenic, is therefore appealing. To engineer a prodrug-like fusion toxin consisting of the anti-EpCAM DARPin Ec1 and a domain I-deleted variant of ETA (ETA″), we used strain-promoted azide alkyne cycloaddition for bioorthogonal conjugation of linear or branched polyethylene glycol (PEG) polymers at defined positions within the toxin moiety. Reversibility of the shielding was provided by a designed peptide linker containing the cleavage site for the rhinovirus 3C model protease. We identified two distinct sites, one within the catalytic domain and one close to the C-terminal KDEL sequence of Ec1-ETA″, simultaneous PEGylation of which resulted in up to 1000-fold lower cytotoxicity in EpCAM-positive tumor cells. Importantly, the potency of the fusion toxin was fully restored by proteolytic unveiling. Upon systemic administration in mice, PEGylated Ec1-ETA″ was much better tolerated than Ec1-ETA″; it showed a longer circulation half-life and an almost 10-fold increased area under the curve (AUC). Our strategy of engineering prodrug-like fusion toxins by bioorthogonal veiling opens new possibilities for targeting tumors with more specificity and efficacy.

Cloning, sequencing and partial functional characterization of the 5' region of the human p75 tumor necrosis factor receptor-encoding gene (TNF-R)

A 1887-bp region at the 5' flank of the human p75 tumor necrosis factor receptor (p75 TNF-R)-encoding gene was found to be active in driving expression of the luc (luciferase-encoding) reporter gene, suggesting that it contains the promoter for the receptor. Rather unexpectedly, a 1827-bp region at the 3' end of the first intron of the p75 TNF-R gene also displayed promoter activity. This activity may be artefactual, reflecting only the presence of an enhancer in this region; yet it also raises the possibility that p75 TNF-R is controlled by more than one promoter and that it encodes various forms of the receptor, or even other proteins. We present here the nucleotide sequences of the 5' flanking and intron regions. Possible implications for the transcriptional regulation of the p75 TNF-R gene are discussed.

Improved bioassay for the detection of porcine tumor necrosis factor using a homologous cell line: PK(15)

Close similarities of various physiological parameters makes the pig one of the preferred animal models for the study of human diseases, especially those involving the cardiovascular system. Unfortunately, the use of pig models to study diseases such as viral hemorrhagic fevers and endotoxic shock syndrome have been hampered by the lack of the necessary immunological tools to measure important immunoregulatory cytokines such as tumor necrosis factor (TNF). Here we describe a TNF-bioassay which is based on the porcine kidney cell line PK(15). Compared to the widely used murine fibroblastoid cell line L929, the PK(15) cell line displays a 100-1000-fold higher sensitivity for porcine TNF-alpha, a higher sensitivity for human TNF-alpha, and a slightly lower sensitivity for murine TNF-alpha. Using a PK(15) bioassay we can detect recombinant TNF-alpha as well as cytotoxic activity in the supernatants of lipopolysaccharide (LPS)-activated porcine monocytes at high dilutions. This suggests that the sensitivity of the test should permit the detection of TNF in biological specimens such as pig serum.

Chromatin structure and DNase I hypersensitivity in the transcriptionally active and inactive porcine tumor necrosis factor gene locus

We have analyzed the chromatin structure of the porcine tumor necrosis factor gene locus (TNF-alpha and TNF-beta). Nuclei from porcine peripheral blood mononuclear cells were digested with different nucleases. As assessed with micrococcal nuclease, the two TNF genes displayed slightly faster digestion kinetics than bulk DNA. Studies with DNaseI revealed distinct DNaseI hypersensitive sites (DH-sites) within the porcine TNF locus. Four DH-sites could be observed in the promoter and mRNA leader regions of the TNF-beta gene. Two DH-sites could be observed for the TNF-alpha gene, one located in the promoter region close to the TATA-box and the other site in intron 3. This pattern of DH-sites was present independently of the activation state of the cells. Interestingly in a porcine macrophage-like cell line, we found that the TNF-alpha promoter DH-site disappeared and another DH-site appeared in the region of intron 1. Additionally, the DH-site of intron 3 could be enhanced by PMA-stimulation in these cells. TNF-beta sites were not detected in this cell line. However, DH-sites were totally absent in fibroblasts (freshly isolated from testicles) and in porcine kidney cells (PK15 cell line) both of which do not transcribe the TNF genes. Therefore, the pattern of DH-sites corresponds to the transcriptional activity of analyzed cells.

The porcine tumor necrosis factor-encoding genes: sequence and comparative analysis

We have cloned and sequenced a 10.22-kb fragment of the genomic locus of the porcine tumor necrosis factor-encoding genes, TNF-alpha and TNF-beta. A liver genomic DNA library, partially digested with Sau3AI, was cloned into the phage lambda EMBL4 and screened with a porcine TNF-alpha cDNA probe. Analysis showed that both the TNF-alpha and TNF-beta genes were present on the cloned fragment. In addition, the cloned fragment contained about 2 kb of repetitive sequences 5' to the TNF-beta gene. The TNF genes are arranged in a tandem repeat, as is the case for the human, mouse and rabbit TNF genes. The comparison of both genes with their human homologues displayed a considerable degree of conservation (80%), suggesting an equal evolution rate.

Generation of a murine hepatic angiosarcoma cell line and reproducible mouse tumor model

Hepatic angiosarcoma (AS) is a rare and highly aggressive tumor of endothelial origin with dismal prognosis. Studies of the molecular biology of AS and treatment options are limited as animal models are rare. We have previously shown that inducible knockout of Notch1 in mice leads to spontaneous formation of hepatic AS. The aims of this study were to: (1) establish and characterize a cell line derived from this murine AS, (2) identify molecular pathways involved in the pathogenesis and potential therapeutic targets, and (3) generate a tumor transplantation model. AS cells retained specific endothelial properties such as tube formation activity, as well as expression of CD31 and Von Willebrand factor. However, electron microscopy analysis revealed signs of dedifferentiation with loss of fenestrae and loss of contact inhibition. Microarray and pathway analysis showed substantial changes in gene expression and revealed activation of the Myc pathway. Exposing the AS cells to sorafenib reduced migration, filopodia dynamics, and cell proliferation but did not induce apoptosis. In addition, sorafenib suppressed ERK phosphorylation and expression of cyclin D2. Injection of AS cells into NOD/SCID mice resulted in formation of undifferentiated tumors, confirming the tumorigenic potential of these cells. In summary, we established and characterized a murine model of spontaneous AS formation and hepatic AS cell lines as a useful in vitro tool. Our data demonstrate antitumor activity of sorafenib in AS cells with potent inhibition of migration, filopodia formation, and cell proliferation, supporting further evaluation of sorafenib as a novel treatment strategy. In addition, AS cell transplantation provides a subcutaneous tumor model useful for in vivo preclinical drug testing.Laboratory Investigation advance online publication, 24 November 2014; doi:10.1038/labinvest.2014.141.

TWIST1 and TWIST2 promoter methylation and protein expression in tumor stroma influence the epithelial-mesenchymal transition-like tumor budding phenotype in colorectal cancer.

Tumor budding in colorectal cancer is likened to an epithelial-mesenchymal transition (EMT) characterized predominantly by loss of E-cadherin and up-regulation of E-cadherin repressors like TWIST1 and TWIST2. Here we investigate a possible epigenetic link between TWIST proteins and the tumor budding phenotype. TWIST1 and TWIST2 promoter methylation and protein expression were investigated in six cell lines and further correlated with tumor budding in patient cohort 1 (n = 185). Patient cohort 2 (n = 112) was used to assess prognostic effects. Laser capture microdissection (LCM) of tumor epithelium and stroma from low- and high-grade budding cancers was performed. In colorectal cancers, TWIST1 and TWIST2 expression was essentially restricted to stromal cells. LCM results of a high-grade budding case show positive TWIST1 and TWIST2 stroma and no methylation, while the low-grade budding case was characterized by negative stroma and strong hypermethylation. TWIST1 stromal cell staining was associated with adverse features like more advanced pT (p = 0.0044), lymph node metastasis (p = 0.0301), lymphatic vessel invasion (p = 0.0373), perineural invasion (p = 0.0109) and worse overall survival time (p = 0.0226). Stromal cells may influence tumor budding in colorectal cancers through expression of TWIST1. Hypermethylation of the tumor stroma may represent an alternative mechanism for regulation of TWIST1.

Tumor budding in upper gastrointestinal carcinomas.

The basis of personalized medicine in oncology is the prediction of an individual's risk of relapse and death from disease. The presence of tumor budding (TB) at the tumor-host interface of gastrointestinal cancers has been recognized as a hallmark of unfavorable disease biology. TB is defined as the presence of dedifferentiated cells or small clusters of up to five cells at the tumor invasive front and can be observed in aggressive carcinomas of the esophagus, stomach, pancreas, ampulla, colon, and rectum. Presence of TB reproducibly correlates with advanced tumor stage, frequent lymphovascular invasion, nodal, and distant metastasis. The UICC has officially recognized TB as additional independent prognostic factor in cancers of the colon and rectum. Recent studies have also characterized TB as a promising prognostic indicator for clinical management of esophageal squamous cell carcinoma, adenocarcinoma of the gastro-esophageal junction, and gastric adenocarcinoma. However, several important issues have to be addressed for application in daily diagnostic practice: (1) validation of prognostic scoring systems for TB in large, multi-center studies, (2) consensus on the optimal assessment method, and (3) inter-observer reproducibility. This review provides a comprehensive analysis of TB in cancers of the upper gastrointestinal tract including critical appraisal of perspectives for further study.