Ståle Nygård

Predicting survival from microarray data—a comparative study

MOTIVATION Survival prediction from gene expression data and other high-dimensional genomic data has been subject to much research during the last years. These kinds of data are associated with the methodological problem of having many more gene expression values than individuals. In addition, the responses are censored survival times. Most of the proposed methods handle this by using Coxs proportional hazards model and obtain parameter estimates by some dimension reduction or parameter shrinkage estimation technique. Using three well-known microarray gene expression data sets, we compare the prediction performance of seven such methods: univariate selection, forward stepwise selection, principal components regression (PCR), supervised principal components regression, partial least squares regression (PLS), ridge regression and the lasso. RESULTS Statistical learning from subsets should be repeated several times in order to get a fair comparison between methods. Methods using coefficient shrinkage or linear combinations of the gene expression values have much better performance than the simple variable selection methods. For our data sets, ridge regression has the overall best performance. AVAILABILITY Matlab and R code for the prediction methods are available at http://www.med.uio.no/imb/stat/bmms/software/microsurv/.

The gut microbial profile in patients with primary sclerosing cholangitis is distinct from patients with ulcerative colitis without biliary disease and healthy controls.

Objective Gut microbiota could influence gut, as well as hepatic and biliary immune responses. We therefore thoroughly characterised the gut microbiota in primary sclerosing cholangitis (PSC) compared with healthy controls (HC) and patients with ulcerative colitis without liver disease. Design We prospectively collected 543 stool samples. After a stringent exclusion process, bacterial DNA was submitted for 16S rRNA gene sequencing. PSC and HC were randomised to an exploration panel or a validation panel, and only significant results (p<0.05, QFDR<0.20) in both panels were reported, followed by a combined comparison of all samples against UC. Results Patients with PSC (N=85) had markedly reduced bacterial diversity compared with HC (N=263, p<0.0001), and a different global microbial composition compared with both HC (p<0.001) and UC (N=36, p<0.01). The microbiota of patients with PSC with and without IBD was similar. Twelve genera separated PSC and HC, out of which 11 were reduced in PSC. However, the Veillonella genus showed a marked increase in PSC compared with both HC (p<0.0001) and UC (p<0.02). Using receiver operating characteristic analysis, Veillonella abundance yielded an area under the curve (AUC) of 0.64 to discriminate PSC from HC, while a combination of PSC-associated genera yielded an AUC of 0.78. Conclusions Patients with PSC exhibited a gut microbial signature distinct from both HC and UC without liver disease, but similar in PSC with and without IBD. The Veillonella genus, which is also associated with other chronic inflammatory and fibrotic conditions, was enriched in PSC.

Survival prediction from clinico-genomic models - a comparative study

BackgroundSurvival prediction from high-dimensional genomic data is an active field in todays medical research. Most of the proposed prediction methods make use of genomic data alone without considering established clinical covariates that often are available and known to have predictive value. Recent studies suggest that combining clinical and genomic information may improve predictions, but there is a lack of systematic studies on the topic. Also, for the widely used Cox regression model, it is not obvious how to handle such combined models.ResultsWe propose a way to combine classical clinical covariates with genomic data in a clinico-genomic prediction model based on the Cox regression model. The prediction model is obtained by a simultaneous use of both types of covariates, but applying dimension reduction only to the high-dimensional genomic variables. We describe how this can be done for seven well-known prediction methods: variable selection, unsupervised and supervised principal components regression and partial least squares regression, ridge regression, and the lasso. We further perform a systematic comparison of the performance of prediction models using clinical covariates only, genomic data only, or a combination of the two. The comparison is done using three survival data sets containing both clinical information and microarray gene expression data. Matlab code for the clinico-genomic prediction methods is available at http://www.med.uio.no/imb/stat/bmms/software/clinico-genomic/.ConclusionsBased on our three data sets, the comparison shows that established clinical covariates will often lead to better predictions than what can be obtained from genomic data alone. In the cases where the genomic models are better than the clinical, ridge regression is used for dimension reduction. We also find that the clinico-genomic models tend to outperform the models based on only genomic data. Further, clinico-genomic models and the use of ridge regression gives for all three data sets better predictions than models based on the clinical covariates alone.

Cytokine expression profiling of the myocardium reveals a role for CX3CL1 (fractalkine) in heart failure.

Several lines of evidence suggest that inflammatory processes mediated by cytokines are involved in the pathogenesis of heart failure (HF). However, the regulation of cytokine expression and the role of cytokines during HF development are not well understood. To address this issue, we have examined alterations in gene expression during HF progression by microarray technology in non-infarcted left ventricular (LV) murine tissue at various time points after myocardial infarction (MI). The highest number of regulated genes was found five days after MI. In total, we identified 14 regulated genes encoding cytokines with no previous association to HF. The strongest up-regulation was found for the chemokine fractalkine (CX3CL1). In human failing hearts we detected a 3-fold increase in CX3CL1 protein production, and both cardiomyocytes and fibrous tissue revealed immunoreactivity for CX3CL1 and its specific receptor CX3CR1. We also found that the circulating level of CX3CL1 was increased in patients with chronic HF in accordance with disease severity (1.6-fold in NYHA II, 2.2-fold in NYHA III and 2.9-fold in NYHA IV). In vitro experiments demonstrated that CX3CL1 production could be induced by inflammatory cytokines known to be highly expressed in HF. CX3CL1 itself induced the expression of markers of cardiac hypertrophy and protein phosphatases in neonatal cardiomyocytes. Given the increased CX3CL1 production in both an experimental HF model and in patients with chronic HF as well as its direct effects on cardiomyocytes, we suggest a role for CX3CL1 and its receptor CX3CR1 in the pathogenesis of HF.

Syndecan-4 Is Essential for Development of Concentric Myocardial Hypertrophy via Stretch-Induced Activation of the Calcineurin-NFAT Pathway

Sustained pressure overload leads to compensatory myocardial hypertrophy and subsequent heart failure, a leading cause of morbidity and mortality. Further unraveling of the cellular processes involved is essential for development of new treatment strategies. We have investigated the hypothesis that the transmembrane Z-disc proteoglycan syndecan-4, a co-receptor for integrins, connecting extracellular matrix proteins to the cytoskeleton, is an important signal transducer in cardiomyocytes during development of concentric myocardial hypertrophy following pressure overload. Echocardiographic, histochemical and cardiomyocyte size measurements showed that syndecan-4−/− mice did not develop concentric myocardial hypertrophy as found in wild-type mice, but rather left ventricular dilatation and dysfunction following pressure overload. Protein and gene expression analyses revealed diminished activation of the central, pro-hypertrophic calcineurin-nuclear factor of activated T-cell (NFAT) signaling pathway. Cardiomyocytes from syndecan-4−/−-NFAT-luciferase reporter mice subjected to cyclic mechanical stretch, a hypertrophic stimulus, showed minimal activation of NFAT (1.6-fold) compared to 5.8-fold increase in NFAT-luciferase control cardiomyocytes. Accordingly, overexpression of syndecan-4 or introducing a cell-permeable membrane-targeted syndecan-4 polypeptide (gain of function) activated NFATc4 in vitro. Pull-down experiments demonstrated a direct intracellular syndecan-4-calcineurin interaction. This interaction and activation of NFAT were increased by dephosphorylation of serine 179 (pS179) in syndecan-4. During pressure overload, phosphorylation of syndecan-4 was decreased, and association between syndecan-4, calcineurin and its co-activator calmodulin increased. Moreover, calcineurin dephosphorylated pS179, indicating that calcineurin regulates its own binding and activation. Finally, patients with hypertrophic myocardium due to aortic stenosis had increased syndecan-4 levels with decreased pS179 which was associated with increased NFAT activation. In conclusion, our data show that syndecan-4 is essential for compensatory hypertrophy in the pressure overloaded heart. Specifically, syndecan-4 regulates stretch-induced activation of the calcineurin-NFAT pathway in cardiomyocytes. Thus, our data suggest that manipulation of syndecan-4 may provide an option for therapeutic modulation of calcineurin-NFAT signaling.

Cardiac troponin T levels and exercise stress testing in patients with suspected coronary artery disease: the Akershus Cardiac Examination (ACE) 1 study

Whether reversible ischaemia in patients referred for exercise stress testing and MPI (myocardial perfusion imaging) is associated with changes in circulating cTn (cardiac troponin) levels is controversial. We measured cTnT with a sensitive assay before, immediately after peak exercise and 1.5 and 4.5 h after exercise stress testing in 198 patients referred for MPI. In total, 19 patients were classified as having reversible myocardial ischaemia. cTnT levels were significantly higher in patients with reversible myocardial ischaemia on MPI at baseline, at peak exercise and after 1.5 h, but not at 4.5 h post-exercise. In patients with reversible ischaemia on MPI, cTnT levels did not change significantly after exercise stress testing [11.1 (5.2–14.9) ng/l at baseline compared with 10.5 (7.2–16.3) ng/l at 4.5 h post-exercise, P=0.27; values are medians (interquartile range)]. Conversely, cTnT levels increased significantly during testing in patients without reversible myocardial ischaemia [5.4 (3.0–9.0) ng/l at baseline compared with 7.5 (4.6–12.4) ng/l, P<0.001]. In conclusion, baseline cTnT levels are higher in patients with MPI evidence of reversible myocardial ischaemia than those without reversible ischaemia. However, although cTnT levels increase during exercise stress testing in patients without evidence of reversible ischaemia, this response appears to be blunted in patients with evidence of reversible ischaemia. Mechanisms other than reversible myocardial ischaemia may play a role for acute exercise-induced increases in circulating cTnT levels.

Enhanced liver fibrosis score predicts transplant-free survival in primary sclerosing cholangitis.

There is a need to determine biomarkers reflecting disease activity and prognosis in primary sclerosing cholangitis (PSC). We evaluated the prognostic utility of the enhanced liver fibrosis (ELF) score in Norwegian PSC patients. Serum samples were available from 305 well‐characterized large‐duct PSC patients, 96 ulcerative colitis patients, and 100 healthy controls. The PSC patients constituted a derivation panel (recruited 1992‐2006 [n = 167]; median age 41 years, 74% male) and a validation panel (recruited 2008‐2012 [n = 138]; median age 40 years, 78% male). We used commercial kits to analyze serum levels of hyaluronic acid, tissue inhibitor of metalloproteinases‐1, and propeptide of type III procollagen and calculated ELF scores by the previously published algorithm. Results were also validated by analysis of ELF tests using the ADVIA Centaur XP system and its commercially available reagents. We found that PSC patients stratified by ELF score tertiles exhibited significantly different transplant‐free survival in both panels (P < 0.001), with higher scores associated with shorter survival, which was confirmed in the validation panel stratified by ELF test tertiles (P = 0.003). The ELF test distinguished between mild and severe disease defined by clinical outcome (transplantation or death) with an area under the curve of 0.81 (95% confidence interval [CI] 0.73‐0.87) and optimal cutoff of 10.6 (sensitivity 70.2%, specificity 79.1%). In multivariate Cox regression analysis in both panels, ELF score (hazard ratio = 1.9, 95% CI 1.4‐2.5, and 1.5, 95% CI 1.1‐2.1, respectively) was associated with transplant‐free survival independently of the Mayo risk score (hazard ratio = 1.3, 95% CI 1.1‐1.6, and 1.6, 95% CI 1.2‐2.1, respectively). The ELF test correlated with ultrasound elastography in separate assessments. Conclusion: The ELF score is a potent prognostic marker in PSC, independent of the Mayo risk score. (Hepatology 2015;62:188‐197)

Impact of Sex on the Prognostic Value of High-Sensitivity Cardiac Troponin I in the General Population: The HUNT Study

BACKGROUND A new, high-sensitivity assay for cardiac troponin I (hs-cTnI) permits evaluation of the prognostic value of cardiac troponins within the reference interval. Men have higher hs-cTnI concentrations than women, but the underlying pathophysiological mechanisms and prognostic implications are unclear. The aim of this study was to assess the potential impact of sex on the association between hs-cTnI and cardiovascular death. METHODS By use of the Architect STAT High-Sensitive Troponin assay, we measured hs-cTnI in 4431 men and 5281 women aged ≥20 years participating in the prospective observational Nord-Trøndelag Health Study (HUNT). RESULTS hs-cTnI was detectable in 98.5% of men and 94.7% of women. During a mean follow-up period of 13.9 years, 708 cardiovascular deaths were registered. hs-cTnI was associated with the incidence of cardiovascular death [adjusted hazard ratio (HR) per 1 SD in log hs-cTnI 1.23 (95% CI 1.15-1.31)], with higher relative risk in women than men [HR 1.44 (1.31-1.58) vs 1.10 (1.00-1.20); Pinteraction < 0.001]. This finding was mediated by both lower risk associated with low hs-cTnI concentrations in women than in men and higher risk associated with high concentrations of hs-cTnI in women than in men. Male sex was associated with a higher risk of cardiovascular death [HR 1.28 (1.11-1.49)], but after adjustment for hs-cTnI, this association disappeared [HR 0.87 (0.75-1.02)]. CONCLUSIONS The prognostic value of hs-cTnI concentrations in the general population is stronger in women than in men. Subtle impairment of cardiovascular status may contribute to higher hs-cTnI concentrations in men, reflecting sex-dependent differences in cardiovascular risk.

Circulating cytokine levels in mice with heart failure are etiology dependent

OBJECTIVES The aim of this study was to examine whether alterations in circulating cytokine levels are dependent on the etiology of myocardial hypertrophy and heart failure (HF). BACKGROUND Several heart diseases are associated with altered levels of circulating cytokines. Cytokines are regarded as possible therapeutic targets or biomarkers, but such approaches are currently not in clinical use. If alterations in circulating cytokines are etiology dependent, this should be taken into consideration when using cytokines as disease markers and therapeutic targets. METHODS The serum levels of 25 cytokines were quantified with Luminex and/or ELISA in four murine models of heart disease: banding of the ascending aorta (AB) or the pulmonary artery (PB), myocardial infarction (MI), and a cardiomyopathy model with inducible cardiomyocyte-specific knockout of the sarco(endo)plasmatic reticulum Ca2+-ATPase (SERCA2KO). RESULTS No increase in circulating cytokine levels were found in mice 1 wk after AB, although substantial myocardial hypertrophy was present. After 1 wk of MI, only interleukin (IL)-18 was increased. In the SERCA2KO mice with HF, circulating levels of IL-1alpha, IL-2, IL-3, IL-6, IL-9, IL-10, IL-12p40, eotaxin, granulocyte-colony stimulating factor (G-CSF), interferon-gamma, monocyte chemoattractant protein-1, macrophage inflammatory protein-1beta were increased, and in mice with PB, IL-1alpha, IL-6, G-CSF, and monokine induced by gamma-interferon showed elevated levels. CONCLUSIONS Serum levels of cytokines in mice with HF vary depending on the etiology. Increased serum levels of several cytokines were found in models with increased right ventricular afterload, suggesting that the cytokine responses result primarily from systemic congestion.

Multiple inflammatory-, tissue remodelling- and fibrosis genes are differentially transcribed in the livers of Abcb4 (−/ − ) mice harbouring chronic cholangitis

Objective. Abcb4 (−/−) mice secrete phosphatidylcholine-free, cytotoxic bile and develop chronic cholangitis. The aim of this study was to identify differentially transcribed genes whose products contribute to the liver tissue pathology during this disease. Material and methods. Hepatic gene transcription was measured in 3-, 6-, 9- and 20-week-old Abcb4 (−/−) mice (FVB.129P2-abcb4tm1Bor/J) using cDNA microarrays, with FVB/NJ Abcb4 (+/+) mice serving as controls. Focus was on inflammatory-, remodelling- and fibrosis genes. Marked differential transcription of inflammatory-, tissue remodelling- and fibrosis genes found by cDNA microarrays was verified by real-time polymerase chain reaction (PCR). Liver pathology was quantified by histopathology scoring. Results. Transcription of clade A3 Serpin genes showed early, marked down-regulation. The chemokine genes Ccl2, Ccl20 and Cxcl10 were markedly up-regulated. Tissue remodelling- and fibrosis genes exhibiting markedly up-regulated transcription included: Ctgf, Elf3, Lgals3, Mmp12, Mmp15, Spp1, Loxl2, Pdgfa, Pdgfrb, Sparc, Tgfb1, Tgfb2, Tgfbi, Tgfbr2 and Col1a1, Col1a2, Col2a1, Col3a1, Col4a1 genes. Microarray-based recordings of differential gene transcription of the majority of these genes harmonized with the liver histopathology score. Thus, cDNA microarray-based analysis showed increasing differential transcription of several inflammatory-, tissue remodelling- and fibrosis genes during the first 9 weeks of disease and a tendency towards differential transcription to stabilize at an elevated level from 9 to 20 weeks of disease. Conclusions. Multiple genes regulating inflammation, tissue remodelling and fibrosis not previously linked to Abcb4 (−/−) cholangitis are identified as being differentially transcribed in Abcb4 (−/−) livers, where they contribute to the pathogenesis of liver tissue pathology.