Dealing with dimensionality: the application of machine learning to multi-omics data.

MOTIVATION: Machine learning (ML) methods are motivated by the need to automate information extraction from large datasets in order to support human users in data-driven tasks. This is an attractive approach for integrative joint analysis of vast amounts of omics data produced in next generation sequencing and other -omics assays. A systematic assessment of the current literature can help to identify key trends and potential gaps in methodology and applications. We surveyed the literature on ML multi-omic data integration and quantitatively explored the goals, techniques and data involved in this field. We were particularly interested in examining how researchers use ML to deal with the volume and complexity of these datasets. RESULTS: Our main finding is that the methods used are those that address the challenges of datasets with few samples and many features. Dimensionality reduction methods are used to reduce the feature count alongside models that can also appropriately handle relatively few samples. Popular techniques include autoencoders, random forests and support vector machines. We also found that the field is heavily influenced by the use of The Cancer Genome Atlas dataset, which is accessible and contains many diverse experiments. AVAILABILITY AND IMPLEMENTATION: All data and processing scripts are available at this GitLab repository: https://gitlab.com/polavieja_lab/ml_multi-omics_review/ or in Zenodo: https://doi.org/10.5281/zenodo.7361807. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MicroRNA-101 Suppresses Tumor Cell Proliferation by Acting as an Endogenous Proteasome Inhibitor via Targeting the Proteasome Assembly Factor POMP.

Proteasome inhibition represents a promising strategy of cancer pharmacotherapy, but resistant tumor cells often emerge. Here we show that the microRNA-101 (miR-101) targets the proteasome maturation protein POMP, leading to impaired proteasome assembly and activity, and resulting in accumulation of p53 and cyclin-dependent kinase inhibitors, cell cycle arrest, and apoptosis. miR-101-resistant POMP restores proper turnover of proteasome substrates and re-enables tumor cell growth. In ERα-positive breast cancers, miR-101 and POMP levels are inversely correlated, and high miR-101 expression or low POMP expression associates with prolonged survival. Mechanistically, miR-101 expression or POMP knockdown attenuated estrogen-driven transcription. Finally, suppressing POMP is sufficient to overcome tumor cell resistance to the proteasome inhibitor bortezomib. Taken together, proteasome activity can not only be manipulated through drugs, but is also subject to endogenous regulation through miR-101, which targets proteasome biogenesis to control overall protein turnover and tumor cell proliferation.

ApoAI-derived peptide increases glucose tolerance and prevents formation of atherosclerosis in mice.

AIMS/HYPOTHESIS: Finding new treatment alternatives for individuals with diabetes with severe insulin resistance is highly desired. To identify novel mechanisms that improve glucose uptake in skeletal muscle, independently from insulin levels and signalling, we have explored the therapeutic potential of a short peptide sequence, RG54, derived from apolipoprotein A-I (ApoA-I). METHODS: INS-1E rat clonal beta cells, C2C12 rat muscle myotubes and J774 mouse macrophages were used to study the impact of RG54 peptide on glucose-stimulated insulin secretion, glucose uptake and cholesterol efflux, respectively. GTTs were carried out on diet-induced insulin-resistant and Leprdb diabetic mouse models treated with RG54 peptide, and the impact of RG54 peptide on atherosclerosis was evaluated in Apoe-/- mice. Control mice received ApoA-I protein, liraglutide or NaCl. RESULTS: The synthetic RG54 peptide induced glucose uptake in cultured muscle myotubes by a similar amount as insulin, and also primed pancreatic beta cells for improved glucose-stimulated insulin secretion. The findings were verified in diet-induced insulin-resistant and Leprdb diabetic mice, jointly confirming the physiological effect. The RG54 peptide also efficiently catalysed cholesterol efflux from macrophages and prevented the formation of atherosclerotic plaques in Apoe-/- mice. CONCLUSIONS/INTERPRETATION: The RG54 peptide exhibits good prospects for providing glucose control and reducing the risk of cardiovascular disease in individuals with severe insulin resistance.

Effects of kiwifruit extracts on colonic gene and protein expression levels in IL-10 gene-deficient mice.

Inflammatory bowel disease (IBD) is a collective term for conditions characterised by chronic inflammation of the gastrointestinal tract involving an inappropriate immune response to commensal micro-organisms in a genetically susceptible host. Previously, aqueous and ethyl acetate extracts of gold kiwifruit (Actinidia chinensis) or green kiwifruit (A. deliciosa) have demonstrated anti-inflammatory activity using in vitro models of IBD. The present study examined whether these kiwifruit extracts (KFE) had immune-modulating effects in vivo against inflammatory processes that are known to be increased in patients with IBD. KFE were used as a dietary intervention in IL-10-gene-deficient (Il10(-/-)) mice (an in vivo model of IBD) and the C57BL/6J background strain in a 3 × 2 factorial design. While all Il10(-/-) mice developed significant colonic inflammation compared with C57BL/6J mice, this was not affected by the inclusion of KFE in the diet. These findings are in direct contrast to our previous study where KFE reduced inflammatory signalling in primary cells isolated from Il10(-/-) and C57BL/6J mice. Whole-genome gene and protein expression level profiling indicated that KFE influenced immune signalling pathways and metabolic processes within the colonic tissue; however, the effects were subtle. In particular, expression levels across gene sets related to adaptive immune pathways were significantly reduced using three of the four KFE in C57BL/6J mice. The present study highlights the importance of investigating food components identified by cell-based assays with appropriate in vivo models before making dietary recommendations, as a food that looks promising in vitro may not be effective in vivo.

Kiwifruit extracts inhibit cytokine production by lipopolysaccharide-activated macrophages, and intestinal epithelial cells isolated from IL10 gene deficient mice.

Inflammatory bowel disease (IBD) is a chronic, inflammatory disorder of the gastrointestinal tract involving an inappropriate immune response to commensal microorganisms in a genetically susceptible host. This study examined the effects of aqueous and ethyl acetate extracts of gold kiwifruit (Actinidia chinensis) or green kiwifruit (Actinidia deliciosa) using in vitro models of IBD. These models comprised primary macrophages and intestinal epithelial cells isolated from C57BL/5J and interleukin-10 gene deficient (Il10(-/-)) mice and RAW 264.7, a murine macrophage-like cell line. All four kiwifruit extracts reduced the activation of these models after lipopolysaccharide stimulation, decreasing nitric oxide and cytokine secretion by both Il10(-/-) and wild-type cells. The ethyl acetate extracts exhibited the highest anti-inflammatory activity, with almost complete suppression of lipopolysaccharide-stimulated macrophage activation. These results suggest that kiwifruit extracts have significant anti-inflammatory activity relevant to IBD. We suggest that the Il10(-/-) mouse is a suitable model for further study of these compounds.

Identification of urinary biomarkers of colon inflammation in IL10-/- mice using Short-Column LCMS metabolomics.

The interleukin-10-deficient (IL10(-/-)) mouse develops colon inflammation in response to normal intestinal microflora and has been used as a model of Crohn's disease. Short-Column LCMS metabolite profiling of urine from IL10(-/-) and wild-type (WT) mice was used, in two independent experiments, to identify mass spectral ions differing in intensity between these two genotypes. Three differential metabolites were identified as xanthurenic acid and as the glucuronides of xanthurenic acid and of α-CEHC (2,5,7,8-tetramethyl-2-(2'-carboxyethyl)-6-hydroxychroman). The significance of several differential metabolites as potential biomarkers of colon inflammation was evaluated in an experiment which compared metabolite concentrations in IL10(-/-) and WT mice housed, either under conventional conditions and dosed with intestinal microflora, or maintained under specific pathogen-free (SPF) conditions. Concentrations of xanthurenic acid, α-CEHC glucuronide, and an unidentified metabolite m/z 495(-)/497(+) were associated with the degree of inflammation in IL10(-/-) mice and may prove useful as biomarkers of colon inflammation.

A short peptide of the C-terminal class Y helices of apolipoprotein A-I has preserved functions in cholesterol efflux and in vivo metabolic control.

Apolipoprotein A-I (ApoA-I) of high-density lipoprotein (HDL) induces glucose uptake by muscle tissues and stimulates pancreatic insulin secretion, and also facilitates cholesterol transport in circulation, and is explored for anti-diabetic and anti-atherosclerotic treatments. As the better alternative to complex protein-lipid formulations it was recently established that the C-terminal region of the ApoA-I protein singly improves the metabolic control and prevents formation of atherosclerotic plaques. Additional investigations of peptides based on the ApoA-I structure may lead to novel anti-diabetic drugs. We here investigate a short peptide (33mer, RG33) that corresponds to the two last helical segments (aa 209-241) of the ApoA-I structure (so-called class Y-helices which forms amphipathic helices) for stability and solubility in serum, for in vitro cholesterol efflux capability, and for providing in vivo glucose control in an insulin resistant mouse model. The RG33 peptide efficiently solubilizes lipid-vesicles, and promotes the efflux of cholesterol from cultured macrophages. The efflux capacity is significantly increased in the presence of lipids compared to non-lipidated RG33. Finally, acute treatment with the RG33 peptide significantly improves the glucose clearance capacity of insulin resistant mice. The impact of the RG33 peptide on glucose control and cholesterol transport, as well as the physicochemical properties, makes it a good candidate for translational exploration of its therapeutic potential in diabetes treatment.

Correction: Strong antitumor synergy between DNA crosslinking and HSP90 inhibition causes massive premitotic DNA fragmentation in ovarian cancer cells.

Since publication of this article, the authors reported that the online version is missing the links to most of the Supplementary data, specifically, Supplementary Figures S1-S9; Supplementary Table S1; all legends to Supplementary Material.

PDLIM2 Is a Marker of Adhesion and ß-Catenin Activity in Triple-Negative Breast Cancer.

The PDLIM2 protein regulates stability of transcription factors including NF-κB and STATs in epithelial and hemopoietic cells. PDLIM2 is strongly expressed in certain cancer cell lines that exhibit an epithelial-to-mesenchymal phenotype, and its suppression is sufficient to reverse this phenotype. PDLIM2 supports the epithelial polarity of nontransformed breast cells, suggesting distinct roles in tumor suppression and oncogenesis. To better understand its overall function, we investigated PDLIM2 expression and activity in breast cancer. PDLIM2 protein was present in 60% of tumors diagnosed as triple-negative breast cancer (TNBC), and only 20% of other breast cancer subtypes. High PDLIM2 expression in TNBC was positively correlated with adhesion signaling and ß-catenin activity. Interestingly, PDLIM2 was restricted to the cytoplasm/membrane of TNBC cells and excluded from the nucleus. In breast cell lines, PDLIM2 retention in the cytoplasm was controlled by cell adhesion, and translocation to the nucleus was stimulated by insulin-like growth factor-1 or TGFß. Cytoplasmic PDLIM2 was associated with active ß-catenin and ectopic expression of PDLIM2 was sufficient to increase ß-catenin levels and its transcriptional activity in reporter assays. Suppression of PDLIM2 inhibited tumor growth in vivo, whereas overexpression of PDLIM2 disrupted growth in 3D cultures. These results suggest that PDLIM2 may serve as a predictive biomarker for a large subset of TNBC whose phenotype depends on adhesion-regulated ß-catenin activity and which may be amenable to therapies that target these pathways. SIGNIFICANCE: This study shows that PDLIM2 expression defines a subset of triple-negative breast cancer that may benefit from targeting the ß-catenin and adhesion signaling pathways. GRAPHICAL ABSTRACT: http://cancerres.aacrjournals.org/content/canres/79/10/2619/F1.large.jpg.

Mechanism of action and preclinical antitumor activity of the novel hypoxia-activated DNA cross-linking agent PR-104.

PURPOSE: Hypoxia is a characteristic of solid tumors and a potentially important therapeutic target. Here, we characterize the mechanism of action and preclinical antitumor activity of a novel hypoxia-activated prodrug, the 3,5-dinitrobenzamide nitrogen mustard PR-104, which has recently entered clinical trials. EXPERIMENTAL DESIGN: Cytotoxicity in vitro was evaluated using 10 human tumor cell lines. SiHa cells were used to characterize metabolism under hypoxia, by liquid chromatography-mass spectrometry, and DNA damage by comet assay and gammaH2AX formation. Antitumor activity was evaluated in multiple xenograft models (PR-104 +/- radiation or chemotherapy) by clonogenic assay 18 h after treatment or by tumor growth delay. RESULTS: The phosphate ester "pre-prodrug" PR-104 was well tolerated in mice and converted rapidly to the corresponding prodrug PR-104A. The cytotoxicity of PR-104A was increased 10- to 100-fold by hypoxia in vitro. Reduction to the major intracellular metabolite, hydroxylamine PR-104H, resulted in DNA cross-linking selectively under hypoxia. Reaction of PR-104H with chloride ion gave lipophilic cytotoxic metabolites potentially able to provide bystander effects. In tumor excision assays, PR-104 provided greater killing of hypoxic (radioresistant) and aerobic cells in xenografts (HT29, SiHa, and H460) than tirapazamine or conventional mustards at equivalent host toxicity. PR-104 showed single-agent activity in six of eight xenograft models and greater than additive antitumor activity in combination with drugs likely to spare hypoxic cells (gemcitabine with Panc-01 pancreatic tumors and docetaxel with 22RV1 prostate tumors). CONCLUSIONS: PR-104 is a novel hypoxia-activated DNA cross-linking agent with marked activity against human tumor xenografts, both as monotherapy and combined with radiotherapy and chemotherapy.

Therapeutic Ablation of Gain-of-Function Mutant p53 in Colorectal Cancer Inhibits Stat3-Mediated Tumor Growth and Invasion.

Over half of colorectal cancers (CRCs) harbor TP53 missense mutations (mutp53). We show that the most common mutp53 allele R248Q (p53Q) exerts gain of function (GOF) and creates tumor dependence in mouse CRC models. mutp53 protein binds Stat3 and enhances activating Stat3 phosphorylation by displacing the phosphatase SHP2. Ablation of the p53Q allele suppressed Jak2/Stat3 signaling, growth, and invasiveness of established, mutp53-driven tumors. Treating tumor-bearing mice with an HSP90 inhibitor suppressed mutp53 levels and tumor growth. Importantly, human CRCs with stabilized mutp53 exhibit enhanced Jak2/Stat3 signaling and are associated with poorer patient survival. Cancers with TP53R248Q/W are associated with a higher patient death risk than are those having nonR248 mutp53. These findings identify GOF mutp53 as a therapeutic target in CRC.

Strong antitumor synergy between DNA crosslinking and HSP90 inhibition causes massive premitotic DNA fragmentation in ovarian cancer cells.

All current regimens for treating ovarian cancer center around carboplatin as standard first line. The HSP90 inhibitor ganetespib is currently being assessed in advanced clinical oncology trials. Thus, we tested the combined effects of ganetespib and carboplatin on a panel of 15 human ovarian cancer lines. Strikingly, the two drugs strongly synergized in cytotoxicity in tumor cells lacking wild-type p53. Mechanistically, ganetespib and carboplatin in combination, but not individually, induced persistent DNA damage causing massive global chromosome fragmentation. Live-cell microscopy revealed chromosome fragmentation occurring to a dramatic degree when cells condensed their chromatin in preparation for mitosis, followed by cell death in mitosis or upon aberrant exit from mitosis. HSP90 inhibition caused the rapid decay of key components of the Fanconi anemia pathway required for repair of carboplatin-induced interstrand crosslinks (ICLs), as well as of cell cycle checkpoint mediators. Overexpressing FancA rescued the DNA damage induced by the drug combination, indicating that FancA is indeed a key client of Hsp90 that enables cancer cell survival in the presence of ICLs. Conversely, depletion of nuclease DNA2 prevented chromosomal fragmentation, pointing to an imbalance of defective repair in the face of uncontrolled nuclease activity as mechanistic basis for the observed premitotic DNA fragmentation. Importantly, the drug combination induced robust antitumor activity in xenograft models, again accompanied with depletion of FancA. In sum, our findings indicate that ganetespib strongly potentiates the antitumor efficacy of carboplatin by causing combined inhibition of DNA repair and cell cycle control mechanisms, thus triggering global chromosome disruption, aberrant mitosis and cell death.

PDLIM2 regulates transcription factor activity in epithelial-to-mesenchymal transition via the COP9 signalosome.

Epithelial cell differentiation and polarized migration associated with epithelial-to-mesenchymal transition (EMT) in cancer requires integration of gene expression with cytoskeletal dynamics. Here we show that the PDZ-LIM domain protein PDLIM2 (Mystique/SLIM), a known cytoskeletal protein and promoter of nuclear nuclear factor κB (NFκB) and signal transducer and activator of transcription (STAT) degradation, regulates transcription factor activity and gene expression through the COP9 signalosome (CSN). Although repressed in certain cancers, PDLIM2 is highly expressed in invasive cancer cells. Here we show that PDLIM2 suppression causes loss of directional migration, inability to polarize the cytoskeleton, and reversal of the EMT phenotype. This is accompanied by altered activity of several transcription factor families, including ß-catenin, Ap-1, NFκB, interferon regulatory factors, STATs, JUN, and p53. We also show that PDLIM2 associates with CSN5, and cells with suppressed PDLIM2 exhibit reduced nuclear accumulation and deneddylation activity of the CSN toward the cullin 1 and cullin 3 subunits of cullin-RING ubiquitin ligases. Thus PDLIM2 integrates cytoskeleton signaling with gene expression in epithelial differentiation by controlling the stability of key transcription factors and CSN activity.

Metabolomic analysis reveals differences in urinary excretion of kiwifruit-derived metabolites in a mouse model of inflammatory bowel disease.

The interleukin-10-deficient (IL-10(-/-)) mouse, a model of inflammatory bowel disease (IBD), develops intestinal inflammation unless raised in germ-free conditions. The metabolic effects of consuming extracts from the fruits of yellow (Actinidia chinensis) or green-fleshed (A. deliciosa) kiwifruit that displayed in vitro anti-inflammatory activity were investigated in IL-10(-/-) mice by metabolomic analysis of urine samples. Kiwifruit-derived metabolites were detected at significantly higher levels in urine of IL-10(-/-) mice relative to those of wild-type mice, indicating that the metabolism of these metabolites was affected by IL-10(-/-)-wild-type genotypic differences. Urinary metabolites previously associated with inflammation were not altered by the kiwifruit extracts. This study demonstrates the use of metabolomic analysis to study dietary effects and the influence of genotype on food metabolism, which may have implications on the development of functional foods for the treatment of IBD.

Nontargeted urinary metabolite profiling of a mouse model of Crohn's disease.

Crohn's disease is an inflammatory disorder of the bowel, believed to arise from the dysregulation of intestinal mucosal immunity. The interleukin-10-deficient (IL10-/-) mouse, which develops intestinal inflammation in the presence of gut microflora, serves as a mouse model of Crohn's disease. Nontargeted urinary metabolite profiling was carried out to identify systemic metabolic changes associated with the development of intestinal inflammation caused by IL10-deficiency. Spot urine samples, collected from IL10-/- and wildtype mice at ages 5.5, 7, 8.5, and 10.5 weeks old were analyzed by gas chromatography-mass spectrometry (GCMS). The data were analyzed using XCMS software, multiple t tests, and ANOVA. Among the key metabolic differences detected were elevated urinary levels of xanthurenic acid and fucose in IL10-/- mice relative to wildtype, indicating upregulation of tryptophan catabolism and perturbed fucosylation in IL10-/- mice. Three short-chain dicarboxylic acid metabolites were decreased in urine of IL10-/- mice relative to wildtype, suggesting the downregulation of fatty acid oxidation in IL10-/- mice. These metabolic differences were reproducible in an independent set of mice. This study demonstrates that nontargeted GCMS metabolite profiling of IL10-/- mice can provide insights into the metabolic effects of IL10-deficiency and identify potential markers of intestinal inflammation.