Penalized ensemble Kalman filters for high dimensional non-linear systems.

The ensemble Kalman filter (EnKF) is a data assimilation technique that uses an ensemble of models, updated with data, to track the time evolution of a usually non-linear system. It does so by using an empirical approximation to the well-known Kalman filter. However, its performance can suffer when the ensemble size is smaller than the state space, as is often necessary for computationally burdensome models. This scenario means that the empirical estimate of the state covariance is not full rank and possibly quite noisy. To solve this problem in this high dimensional regime, we propose a computationally fast and easy to implement algorithm called the penalized ensemble Kalman filter (PEnKF). Under certain conditions, it can be theoretically proven that the PEnKF will be accurate (the estimation error will converge to zero) despite having fewer ensemble members than state dimensions. Further, as contrasted to localization methods, the proposed approach learns the covariance structure associated with the dynamical system. These theoretical results are supported with simulations of several non-linear and high dimensional systems.

Probabilistic Data-Driven Sampling via Multi-Criteria Importance Analysis.

Although supercomputers are becoming increasingly powerful, their components have thus far not scaled proportionately. Compute power is growing enormously and is enabling finely resolved simulations that produce never-before-seen features. However, I/O capabilities lag by orders of magnitude, which means only a fraction of the simulation data can be stored for post hoc analysis. Prespecified plans for saving features and quantities of interest do not work for features that have not been seen before. Data-driven intelligent sampling schemes are needed to detect and save important parts of the simulation while it is running. Here, we propose a novel sampling scheme that reduces the size of the data by orders-of-magnitude while still preserving important regions. The approach we develop selects points with unusual data values and high gradients. We demonstrate that our approach outperforms traditional sampling schemes on a number of tasks.

Induction of KIAA1199/CEMIP is associated with colon cancer phenotype and poor patient survival.

Genes induced in colon cancer provide novel candidate biomarkers of tumor phenotype and aggressiveness. We originally identified KIAA1199 (now officially called CEMIP) as a transcript highly induced in colon cancer: initially designating the transcript as Colon Cancer Secreted Protein 1. We molecularly characterized CEMIP expression both at the mRNA and protein level and found it is a secreted protein induced an average of 54-fold in colon cancer. Knockout of CEMIPreduced the ability of human colon cancer cells to form xenograft tumors in athymic mice. Tumors that did grow had increased deposition of hyaluronan, linking CEMIP participation in hyaluronan degradation to the modulation of tumor phenotype. We find CEMIP mRNA overexpression correlates with poorer patient survival. In stage III only (n = 31) or in combined stage II plus stage III colon cancer cases (n = 73), 5-year overall survival was significantly better (p = 0.004 and p = 0.0003, respectively) among patients with low CEMIP expressing tumors than those with high CEMIP expressing tumors. These results demonstrate that CEMIP directly facilitates colon tumor growth, and high CEMIP expression correlates with poor outcome in stage III and in stages II+III combined cohorts. We present CEMIP as a candidate prognostic marker for colon cancer and a potential therapeutic target.

Sulindac reversal of 15-PGDH-mediated resistance to colon tumor chemoprevention with NSAIDs.

Non-steroidal anti-inflammatory drugs prevent colorectal cancer by inhibiting cyclooxygenase (COX) enzymes that synthesize tumor-promoting prostaglandins. 15-hydroxyprostaglandin dehydrogenase (15-PGDH) is a tumor suppressor that degrades tumor-promoting prostaglandins. Murine knockout of 15-PGDH increases susceptibility to azoxymethane-induced colon tumors. It also renders these mice resistant to celecoxib, a selective inhibitor of inducible COX-2 during colon neoplasia. Similarly, humans with low colonic 15-PGDH are also resistant to colon adenoma prevention with celecoxib. Here, we used aspirin and sulindac, which inhibit both COX-1 and COX-2, in order to determine if these broader COX inhibitors can prevent colon tumors in 15-PGDH knockout (KO) mice. Unlike celecoxib, sulindac proved highly effective in colon tumor prevention of 15-PGDH KO mice. Significantly, however, aspirin demonstrated no effect on colon tumor incidence in either 15-PGDH wild-type or KO mice, despite a comparable reduction in colonic mucosal Prostaglandin E2 (PGE2) levels by both sulindac and aspirin. Notably, colon tumor prevention activity by sulindac was accompanied by a marked induction of lymphoid aggregates and proximal colonic inflammatory mass lesions, a side effect seen to a lesser degree with celecoxib, but not with aspirin. These findings suggest that sulindac may be the most effective agent for colon cancer prevention in humans with low 15-PGDH, but its use may also be associated with inflammatory lesions in the colon.

Mst1 and Mst2 protein kinases restrain intestinal stem cell proliferation and colonic tumorigenesis by inhibition of Yes-associated protein (Yap) overabundance.

Ablation of the kinases Mst1 and Mst2, orthologs of the Drosophila antiproliferative kinase Hippo, from mouse intestinal epithelium caused marked expansion of an undifferentiated stem cell compartment and loss of secretory cells throughout the small and large intestine. Although median survival of mice lacking intestinal Mst1/Mst2 is 13 wk, adenomas of the distal colon are common by this age. Diminished phosphorylation, enhanced abundance, and nuclear localization of the transcriptional coactivator Yes-associated protein 1 (Yap1) is evident in Mst1/Mst2-deficient intestinal epithelium, as is strong activation of ß-catenin and Notch signaling. Although biallelic deletion of Yap1 from intestinal epithelium has little effect on intestinal development, inactivation of a single Yap1 allele reduces Yap1 polypeptide abundance to nearly wild-type levels and, despite the continued Yap hypophosphorylation and preferential nuclear localization, normalizes epithelial structure. Thus, supraphysiologic Yap polypeptide levels are necessary to drive intestinal stem cell proliferation. Yap is overexpressed in 68 of 71 human colon cancers and in at least 30 of 36 colon cancer-derived cell lines. In colon-derived cell lines where Yap is overabundant, its depletion strongly reduces ß-catenin and Notch signaling and inhibits proliferation and survival. These findings demonstrate that Mst1 and Mst2 actively suppress Yap1 abundance and action in normal intestinal epithelium, an antiproliferative function that frequently is overcome in colon cancer through Yap1 polypeptide overabundance. The dispensability of Yap1 in normal intestinal homeostasis and its potent proliferative and prosurvival actions when overexpressed in colon cancer make it an attractive therapeutic target.

Identification and functional characterization of paxillin as a target of protein tyrosine phosphatase receptor T.

Protein tyrosine phosphatase receptor-type T (PTPRT) is the most frequently mutated tyrosine phosphatase in human cancers. However, the cell signaling pathways regulated by PTPRT largely remain to be elucidated. Here, we show that paxillin is a direct substrate of PTPRT and that PTPRT specifically regulates paxillin phosphorylation at tyrosine residue 88 (Y88) in colorectal cancer (CRC) cells. We engineered CRC cells homozygous for a paxillin Y88F knock-in mutant and found that these cells exhibit significantly reduced cell migration and impaired anchorage-independent growth, fail to form xenograft tumors in nude mice, and have decreased phosphorylation of p130CAS, SHP2, and AKT. PTPRT knockout mice that we generated exhibit increased levels of colonic paxillin phosphorylation at residue Y88 and are highly susceptible to carcinogen azoxymethane-induced colon tumor, providing critical in vivo evidence that PTPRT normally functions as a tumor suppressor. Moreover, similarly increased paxillin pY88 is also found as a common feature of human colon cancers. These studies reveal an important signaling pathway that plays a critical role in colorectal tumorigenesis.

Inactivating germ-line and somatic mutations in polypeptide N-acetylgalactosaminyltransferase 12 in human colon cancers.

Aberrant glycosylation is a pathological alteration that is widespread in colon cancer, and usually accompanies the onset and progression of the disease. To date, the molecular mechanisms underlying aberrant glycosylation remain largely unknown. In this study, we identify somatic and germ-line mutations in the gene encoding for polypeptide N-acetylgalactosaminyltransferase 12 (GALNT12) in individuals with colon cancer. Biochemical analyses demonstrate that each of the 8 GALNT12 mutations identified inactivates the normal function of the GALNT enzyme in initiating mucin type O-linked protein glycosylation. Two of these inactivating GALNT12 mutations were identified as acquired somatic mutations in a set of 30 microsatellite stable colon tumors. Relative to background gene mutation rates, finding these somatic GALNT12 mutations was statistically significant at P < 0.001. Six additional inactivating GALNT12 mutations were detected as germ-line changes carried by patients with colon cancer; however, no inactivating variants were detected among cancer-free controls (P = 0.005). Notably, in 3 of the 6 individuals harboring inactivating germ-line GALNT12 mutations, both a colon cancer and a second independent epithelial cancer had developed. These findings suggest that genetic defects in the O-glycosylation pathway in part underlie aberrant glycosylation in colon cancers, and they contribute to the development of a subset of these malignancies.

15-Hydroxyprostaglandin dehydrogenase inactivation as a mechanism of resistance to celecoxib chemoprevention of colon tumors.

Pharmacologic inhibitors of the prostaglandin-synthesizing COX-2 oncogene prevent the development of premalignant human colon adenomas. However, resistance to treatment is common. In this study, we show that the adenoma prevention activity of the COX-2 inhibitor celecoxib requires the concomitant presence of the 15-hydroxyprostaglandin dehydrogenase (15-PGDH) tumor suppressor gene, and that loss of 15-PGDH expression imparts resistance to celecoxib's anti-tumor effects. We first demonstrate that the adenoma-preventive activity of celecoxib is abrogated in mice genetically lacking 15-PGDH. In FVB mice, celecoxib prevents 85% of azoxymethane-induced tumors >1 mm in size, but is essentially inactive in preventing tumor induction in 15-PGDH-null animals. Indeed, celecoxib treated 15-PGDH null animals develop more tumors than do celecoxib naive WT mice. In parallel with the loss of tumor prevention activity, celecoxib-mediated suppression of colonic PGE(2) levels is also markedly attenuated in 15-PGDH-null versus WT mice. Finally, as predicted by the murine models, humans with low colonic 15-PGDH levels also exhibit celecoxib resistance. Specifically, in a colon adenoma prevention trial, in all cases tested, individuals who developed new adenomas while receiving celecoxib treatment were also found as having low colonic 15-PGDH levels.

15-Hydroxyprostaglandin dehydrogenase is a target of hepatocyte nuclear factor 3beta and a tumor suppressor in lung cancer.

The forkhead transcription factor hepatocyte nuclear factor 3beta (HNF3beta) is essential in foregut development and the regulation of lung-specific genes. HNF3beta expression leads to growth arrest and apoptosis in lung cancer cells and HNF3beta is a candidate tumor suppressor in lung cancer. In a transcriptional profiling study using a conditional cell line system, we now identify 15-PGDH as one of the major genes induced by HNF3beta expression. 15-PGDH is a critical metabolic enzyme of proliferative prostaglandins, an antagonist to cyclooxygenase-2 and a tumor suppressor in colon cancer. We confirmed the regulation of 15-PGDH expression by HNF3beta in a number of systems and showed direct binding of HNF3beta to 15-PGDH promoter elements. Western blotting of lung cancer cell lines and immunohistochemical examination of human lung cancer tissues found loss of 15-PGDH expression in approximately 65% of lung cancers. Further studies using in vitro cell-based assays and in vivo xenograft tumorigenesis assays showed a lack of in vitro but significant in vivo tumor suppressor activity of 15-PGDH via an antiangiogenic mechanism analogous to its role in colon cancer. In summary, we identify 15-PGDH as a direct downstream effector of HNF3beta and show that 15-PGDH acts as a tumor suppressor in lung cancer.

15-Hydroxyprostaglandin dehydrogenase is an in vivo suppressor of colon tumorigenesis.

15-Hydroxyprostaglandin dehydrogenase (15-PGDH) is a prostaglandin-degrading enzyme that is highly expressed in normal colon mucosa but is ubiquitously lost in human colon cancers. Herein, we demonstrate that 15-PGDH is active in vivo as a highly potent suppressor of colon neoplasia development and acts in the colon as a required physiologic antagonist of the prostaglandin-synthesizing activity of the cyclooxygenase 2 (COX-2) oncogene. We first show that 15-PGDH gene knockout induces a marked 7.6-fold increase in colon tumors arising in the Min (multiple intestinal neoplasia) mouse model. Furthermore, 15-PGDH gene knockout abrogates the normal resistance of C57BL/6J mice to colon tumor induction by the carcinogen azoxymethane (AOM), conferring susceptibility to AOM-induced adenomas and carcinomas in situ. Susceptibility to AOM-induced tumorigenesis is mediated by a marked induction of dysplasia, proliferation, and cyclin D1 expression throughout microscopic aberrant crypt foci arising in 15-PGDH null colons and is concomitant with a doubling of prostaglandin E(2) in 15-PGDH null colonic mucosa. A parallel role for 15-PGDH loss in promoting the earliest steps of colon neoplasia in humans is supported by our finding of a universal loss of 15-PGDH expression in microscopic colon adenomas recovered from patients with familial adenomatous polyposis, including adenomas as small as a single crypt. These models thus delineate the in vivo significance of 15-PGDH-mediated negative regulation of the COX-2 pathway and moreover reveal the particular importance of 15-PGDH in opposing the neoplastic progression of colonic aberrant crypt foci.

15-Hydroxyprostaglandin dehydrogenase, a COX-2 oncogene antagonist, is a TGF-beta-induced suppressor of human gastrointestinal cancers.

Marked increased expression of cyclooxygenase 2 (COX-2), a prostaglandin-synthesizing enzyme that is pharmacologically inhibited by nonsteroid anti-inflammatory-type drugs, is a major early oncogenic event in the genesis of human colon neoplasia. We report that, in addition to inducing expression of COX-2, colon cancers further target the prostaglandin biogenesis pathway by ubiquitously abrogating expression of 15-hydroxyprostaglandin dehydrogenase (15-PGDH), a prostaglandin-degrading enzyme that physiologically antagonizes COX-2. We find that 15-PGDH transcript and protein are both highly expressed by normal colonic epithelia but are nearly undetectable in colon cancers. Using gene transfection to restore 15-PGDH expression in colon cancer cells strongly inhibits the ability of these cells to form tumors in immune-deficient mice and demonstrates 15-PGDH to have functional colon cancer tumor suppressor activity. In interrogating the mechanism for 15-PGDH expression loss in colon cancer, we determined that colonic 15-PGDH expression is directly controlled and strongly induced by activation of the TGF-beta tumor suppressor pathway. These findings thus delineate an enzymatic pathway that induces colon cancer suppression, a pathway that is activated by TGF-beta and mediated by 15-PGDH.

PMEPA1, a transforming growth factor-beta-induced marker of terminal colonocyte differentiation whose expression is maintained in primary and metastatic colon cancer.

To identify potential effectors of transforming growth factor (TGF)-beta-mediated suppression of colon cancer, we used GeneChip expression microarrays to identify TGF-beta-induced genes in VACO 330, a nontransformed TGF-beta-sensitive cell line derived from a human adenomatous colon polyp. PMEPA1 was identified as a gene highly up-regulated by TGF-beta treatment of VACO 330. Northern blot analysis confirmed TGF-beta induction of PMEPA1 in VACO 330, as well as a panel of three other TGF-beta-sensitive colon cell lines. PMEPA1 induction could be detected as early as 2 h after TGF-beta treatment and was not inhibited by pretreatment of cells with cycloheximide, suggesting that PMEPA1 is a direct target of TGF-beta signaling. Wild-type PMEPA1 and an alternative splice variant lacking the putative transmembrane domain were encoded by the PMEPA1 locus and were shown by epitope tagging to encode proteins with differing subcellular localization. Both variants were found to be expressed in normal colonic epithelium, and both were shown to be induced by TGF-beta. Consistent with TGF-beta playing a role in terminal differentiation of colonocytes, in situ hybridization of normal colonic epithelium localized PMEPA1 expression to nonproliferating, terminally differentiated epithelium located at the top of colonic crypts. Intriguingly, in situ hybridization and Northern blot analysis showed that the expression of PMEPA1 was well maintained both in colon cancer primary tumors and in colon cancer liver metastases. PMEPA1 is thus a novel TGF-beta-induced marker of a differentiated crypt cell population. Moreover, as PMEPA1 expression is maintained, presumptively in a TGF-beta-independent manner after malignant transformation and metastasis, it demonstrates that even late colon cancers retain a strong capacity to execute many steps of the normal colonic differentiation program.