Advertisement
Pegylated IFN-α regulates hepatic gene expression through transient Jak/STAT activation
Michael T. Dill,1,2 Zuzanna Makowska,1 Gaia Trincucci,1 Andreas J. Gruber,3 Julia E. Vogt,4 Magdalena Filipowicz,1,2 Diego Calabrese,1 Ilona Krol,1 Daryl T. Lau,5 Luigi Terracciano,6 Erik van Nimwegen,3 Volker Roth,4 and Markus H. Heim1,2
1Department of Biomedicine, Hepatology Laboratory, University of Basel, Basel, Switzerland. 2Division of Gastroenterology and Hepatology, University Hospital Basel, Basel, Switzerland. 3Biozentrum, University of Basel and Swiss Institute of Bioinformatics, Basel, Switzerland. 4Computer Science Department, University of Basel, Basel, Switzerland. 5Liver Center, Department of Medicine, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts, USA. 6Institute of Pathology, University Hospital Basel, Basel, Switzerland.
Address correspondence to: Markus H. Heim, University of Basel, Department of Biomedicine, Zentrum für Lehre und Forschung, Hebelstrasse 20, 4031 Basel, Switzerland. Phone: 41.61.265.33.62; Fax: 41.61.265.38.47; E-mail: markus.heim@unibas.ch.
Find articles by Dill, M. in: JCI | PubMed | Google Scholar
1Department of Biomedicine, Hepatology Laboratory, University of Basel, Basel, Switzerland. 2Division of Gastroenterology and Hepatology, University Hospital Basel, Basel, Switzerland. 3Biozentrum, University of Basel and Swiss Institute of Bioinformatics, Basel, Switzerland. 4Computer Science Department, University of Basel, Basel, Switzerland. 5Liver Center, Department of Medicine, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts, USA. 6Institute of Pathology, University Hospital Basel, Basel, Switzerland.
Address correspondence to: Markus H. Heim, University of Basel, Department of Biomedicine, Zentrum für Lehre und Forschung, Hebelstrasse 20, 4031 Basel, Switzerland. Phone: 41.61.265.33.62; Fax: 41.61.265.38.47; E-mail: markus.heim@unibas.ch.
Find articles by Makowska, Z. in: JCI | PubMed | Google Scholar
1Department of Biomedicine, Hepatology Laboratory, University of Basel, Basel, Switzerland. 2Division of Gastroenterology and Hepatology, University Hospital Basel, Basel, Switzerland. 3Biozentrum, University of Basel and Swiss Institute of Bioinformatics, Basel, Switzerland. 4Computer Science Department, University of Basel, Basel, Switzerland. 5Liver Center, Department of Medicine, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts, USA. 6Institute of Pathology, University Hospital Basel, Basel, Switzerland.
Address correspondence to: Markus H. Heim, University of Basel, Department of Biomedicine, Zentrum für Lehre und Forschung, Hebelstrasse 20, 4031 Basel, Switzerland. Phone: 41.61.265.33.62; Fax: 41.61.265.38.47; E-mail: markus.heim@unibas.ch.
Find articles by Trincucci, G. in: JCI | PubMed | Google Scholar
1Department of Biomedicine, Hepatology Laboratory, University of Basel, Basel, Switzerland. 2Division of Gastroenterology and Hepatology, University Hospital Basel, Basel, Switzerland. 3Biozentrum, University of Basel and Swiss Institute of Bioinformatics, Basel, Switzerland. 4Computer Science Department, University of Basel, Basel, Switzerland. 5Liver Center, Department of Medicine, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts, USA. 6Institute of Pathology, University Hospital Basel, Basel, Switzerland.
Address correspondence to: Markus H. Heim, University of Basel, Department of Biomedicine, Zentrum für Lehre und Forschung, Hebelstrasse 20, 4031 Basel, Switzerland. Phone: 41.61.265.33.62; Fax: 41.61.265.38.47; E-mail: markus.heim@unibas.ch.
Find articles by Gruber, A. in: JCI | PubMed | Google Scholar
1Department of Biomedicine, Hepatology Laboratory, University of Basel, Basel, Switzerland. 2Division of Gastroenterology and Hepatology, University Hospital Basel, Basel, Switzerland. 3Biozentrum, University of Basel and Swiss Institute of Bioinformatics, Basel, Switzerland. 4Computer Science Department, University of Basel, Basel, Switzerland. 5Liver Center, Department of Medicine, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts, USA. 6Institute of Pathology, University Hospital Basel, Basel, Switzerland.
Address correspondence to: Markus H. Heim, University of Basel, Department of Biomedicine, Zentrum für Lehre und Forschung, Hebelstrasse 20, 4031 Basel, Switzerland. Phone: 41.61.265.33.62; Fax: 41.61.265.38.47; E-mail: markus.heim@unibas.ch.
Find articles by Vogt, J. in: JCI | PubMed | Google Scholar
1Department of Biomedicine, Hepatology Laboratory, University of Basel, Basel, Switzerland. 2Division of Gastroenterology and Hepatology, University Hospital Basel, Basel, Switzerland. 3Biozentrum, University of Basel and Swiss Institute of Bioinformatics, Basel, Switzerland. 4Computer Science Department, University of Basel, Basel, Switzerland. 5Liver Center, Department of Medicine, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts, USA. 6Institute of Pathology, University Hospital Basel, Basel, Switzerland.
Address correspondence to: Markus H. Heim, University of Basel, Department of Biomedicine, Zentrum für Lehre und Forschung, Hebelstrasse 20, 4031 Basel, Switzerland. Phone: 41.61.265.33.62; Fax: 41.61.265.38.47; E-mail: markus.heim@unibas.ch.
Find articles by Filipowicz, M. in: JCI | PubMed | Google Scholar
1Department of Biomedicine, Hepatology Laboratory, University of Basel, Basel, Switzerland. 2Division of Gastroenterology and Hepatology, University Hospital Basel, Basel, Switzerland. 3Biozentrum, University of Basel and Swiss Institute of Bioinformatics, Basel, Switzerland. 4Computer Science Department, University of Basel, Basel, Switzerland. 5Liver Center, Department of Medicine, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts, USA. 6Institute of Pathology, University Hospital Basel, Basel, Switzerland.
Address correspondence to: Markus H. Heim, University of Basel, Department of Biomedicine, Zentrum für Lehre und Forschung, Hebelstrasse 20, 4031 Basel, Switzerland. Phone: 41.61.265.33.62; Fax: 41.61.265.38.47; E-mail: markus.heim@unibas.ch.
Find articles by Calabrese, D. in: JCI | PubMed | Google Scholar
1Department of Biomedicine, Hepatology Laboratory, University of Basel, Basel, Switzerland. 2Division of Gastroenterology and Hepatology, University Hospital Basel, Basel, Switzerland. 3Biozentrum, University of Basel and Swiss Institute of Bioinformatics, Basel, Switzerland. 4Computer Science Department, University of Basel, Basel, Switzerland. 5Liver Center, Department of Medicine, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts, USA. 6Institute of Pathology, University Hospital Basel, Basel, Switzerland.
Address correspondence to: Markus H. Heim, University of Basel, Department of Biomedicine, Zentrum für Lehre und Forschung, Hebelstrasse 20, 4031 Basel, Switzerland. Phone: 41.61.265.33.62; Fax: 41.61.265.38.47; E-mail: markus.heim@unibas.ch.
Find articles by Krol, I. in: JCI | PubMed | Google Scholar
1Department of Biomedicine, Hepatology Laboratory, University of Basel, Basel, Switzerland. 2Division of Gastroenterology and Hepatology, University Hospital Basel, Basel, Switzerland. 3Biozentrum, University of Basel and Swiss Institute of Bioinformatics, Basel, Switzerland. 4Computer Science Department, University of Basel, Basel, Switzerland. 5Liver Center, Department of Medicine, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts, USA. 6Institute of Pathology, University Hospital Basel, Basel, Switzerland.
Address correspondence to: Markus H. Heim, University of Basel, Department of Biomedicine, Zentrum für Lehre und Forschung, Hebelstrasse 20, 4031 Basel, Switzerland. Phone: 41.61.265.33.62; Fax: 41.61.265.38.47; E-mail: markus.heim@unibas.ch.
Find articles by Lau, D. in: JCI | PubMed | Google Scholar
1Department of Biomedicine, Hepatology Laboratory, University of Basel, Basel, Switzerland. 2Division of Gastroenterology and Hepatology, University Hospital Basel, Basel, Switzerland. 3Biozentrum, University of Basel and Swiss Institute of Bioinformatics, Basel, Switzerland. 4Computer Science Department, University of Basel, Basel, Switzerland. 5Liver Center, Department of Medicine, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts, USA. 6Institute of Pathology, University Hospital Basel, Basel, Switzerland.
Address correspondence to: Markus H. Heim, University of Basel, Department of Biomedicine, Zentrum für Lehre und Forschung, Hebelstrasse 20, 4031 Basel, Switzerland. Phone: 41.61.265.33.62; Fax: 41.61.265.38.47; E-mail: markus.heim@unibas.ch.
Find articles by Terracciano, L. in: JCI | PubMed | Google Scholar
1Department of Biomedicine, Hepatology Laboratory, University of Basel, Basel, Switzerland. 2Division of Gastroenterology and Hepatology, University Hospital Basel, Basel, Switzerland. 3Biozentrum, University of Basel and Swiss Institute of Bioinformatics, Basel, Switzerland. 4Computer Science Department, University of Basel, Basel, Switzerland. 5Liver Center, Department of Medicine, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts, USA. 6Institute of Pathology, University Hospital Basel, Basel, Switzerland.
Address correspondence to: Markus H. Heim, University of Basel, Department of Biomedicine, Zentrum für Lehre und Forschung, Hebelstrasse 20, 4031 Basel, Switzerland. Phone: 41.61.265.33.62; Fax: 41.61.265.38.47; E-mail: markus.heim@unibas.ch.
Find articles by van Nimwegen, E. in: JCI | PubMed | Google Scholar
1Department of Biomedicine, Hepatology Laboratory, University of Basel, Basel, Switzerland. 2Division of Gastroenterology and Hepatology, University Hospital Basel, Basel, Switzerland. 3Biozentrum, University of Basel and Swiss Institute of Bioinformatics, Basel, Switzerland. 4Computer Science Department, University of Basel, Basel, Switzerland. 5Liver Center, Department of Medicine, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts, USA. 6Institute of Pathology, University Hospital Basel, Basel, Switzerland.
Address correspondence to: Markus H. Heim, University of Basel, Department of Biomedicine, Zentrum für Lehre und Forschung, Hebelstrasse 20, 4031 Basel, Switzerland. Phone: 41.61.265.33.62; Fax: 41.61.265.38.47; E-mail: markus.heim@unibas.ch.
Find articles by Roth, V. in: JCI | PubMed | Google Scholar
1Department of Biomedicine, Hepatology Laboratory, University of Basel, Basel, Switzerland. 2Division of Gastroenterology and Hepatology, University Hospital Basel, Basel, Switzerland. 3Biozentrum, University of Basel and Swiss Institute of Bioinformatics, Basel, Switzerland. 4Computer Science Department, University of Basel, Basel, Switzerland. 5Liver Center, Department of Medicine, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, Massachusetts, USA. 6Institute of Pathology, University Hospital Basel, Basel, Switzerland.
Address correspondence to: Markus H. Heim, University of Basel, Department of Biomedicine, Zentrum für Lehre und Forschung, Hebelstrasse 20, 4031 Basel, Switzerland. Phone: 41.61.265.33.62; Fax: 41.61.265.38.47; E-mail: markus.heim@unibas.ch.
Find articles by Heim, M. in: JCI | PubMed | Google Scholar
Published February 24, 2014 - More info
J Clin Invest. 2014;124(4):1568–1581. https://doi.org/10.1172/JCI70408.
© 2014 The American Society for Clinical Investigation
Received: April 9, 2013; Accepted: December 17, 2013
-
Results
pegIFN-α2b induced STAT1 phosphorylation and ISG expression in the liver. We studied pegIFN-α2b–induced STAT1 phosphorylation and gene regulation in 18 patients who underwent treatment for CHC with pegIFN-α and ribavirin. All patients had a first liver biopsy before treatment during the routine clinical CHC work-up. A second biopsy was taken 4 hours (n = 6), 16 hours (n = 3), 2 days (n = 3), 4 days (n = 3), and 6 days (n = 3) after the first injection of pegIFN-α2b. The 6 patients whose second liver biopsy was performed 4 hours after injection were selected from among the 16 patients who had already been included in the previous study described above (13), because they had no preactivation of the endogenous IFN system in the liver and a normal response to pegIFN-α2b. The patients were selected in a two-step procedure for the later time points. First, liver biopsies from patients with CHC who agreed to donate part of their liver biopsy for research were analyzed with a previously developed and validated four-gene classifier to predict their likelihood of responding to pegIFN-α (17). Patients with a high probability of an unimpaired, normal response to pegIFN-α were then asked to participate in our study and to consent to a second liver biopsy. This two-step selection process was necessary, because in patients with preinduced hepatic ISGs, the Jak/STAT signaling pathway is refractory in liver cells (13), and a second liver biopsy would have been of little use for the study of pegIFN-α pharmacodynamic effects in the liver. Indeed, the selection process with the four-gene classifier was highly accurate in predicting a good response to pegIFN-α: all patients were treatment responders, and apart from 1 patient who relapsed after treatment, all patients were cured of their HCV infection (Table 1).
We analyzed pegIFN-α2b–induced Jak/STAT signaling by immunohistochemistry (IHC) with phosphorylated STAT1–specific (p-STAT–specific) antibodies. Because we selectively included only patients who had no pretreatment induction of the endogenous IFN system, STAT1 was not activated in the biopsies obtained before treatment (Figure 1A). Following the first pegIFN-α2b injection, we observed a rapid and strong activation of STAT1 already 4 hours later, with nuclear p-STAT1 signals detected in more than 60% of hepatocytes (Figure 1). p-STAT1 signals were still strong after 16 hours, but then rapidly declined. In liver biopsies obtained after 2, 4, or 6 days, p-STAT1 signals in hepatocytes were weak and were detected in less than 5% of hepatocytes. In nonparenchymal cells, we detected p-STAT1 signals at all time points.
Figure 1pegIFN-α2b transiently induces the Jak/STAT pathway in the liver. (A) Representative images of IHC analysis of p-STAT1 in liver biopsies obtained before treatment (B1) and at several time points after the first injection of pegIFN-α2b. Strong nuclear p-STAT1 signals were present at the 4- and 16-hour time points, but not at later time points, where the signals were localized in nonparenchymal cells (arrows). Scale bars: 20 μm. (B) Quantitative analysis of the mean percentage of p-STAT1–positive hepatocyte nuclei (5 × 100 cells counted per sample; the number of samples is indicated) per time point. Bars show the mean with SEM.
To further address the kinetics of ISG induction by pegIFN-α2b, we adapted a highly sensitive and specific in situ hybridization (ISH) method (QuantiGene ViewRNA) that allowed the detection of ISG mRNAs in fresh-frozen liver biopsy samples. We detected MX1 mRNA already 4 hours after the injection of pegIFN-α2b and found that it peaked at the 16-hour time point and then rapidly declined (Figure 2A). IFI27 mRNA expression peaked at 16 hours and declined at a much slower rate. Of note, the intensity of the signals declined in all hepatocytes, and at later time points we did not detect hepatocytes with the signal intensities found at the 16-hour point. Together with the absence of strong nuclear p-STAT1 signals in hepatocytes at later time points (Figure 1A), these data do not support the hypothesis that hepatocytes recover asynchronously from the refractory state and that they are, in part, restimulated by pegIFN-α2b circulating at high concentrations during the entire dosing interval (Table 1).
Figure 2ISH reveals distinct expression patterns of ISG mRNAs at different time points. (A) Representative examples of ISH staining (green) in liver biopsies for MX1 and IFI27 mRNA showing that ubiquitous expression gradually declined over time with distinct kinetics. (B) ISH staining (green) in liver biopsies for SOCS1 and PDL1 mRNA revealed expression in hepatocytes and in nonparenchymal cells at 4 hours. At 144 hours, SOCS1 and PDL1 were detected only in nonparenchymal cells (black arrows). Scale bars: 20 μm.
In contrast to hepatocytes, we found that nonparenchymal cells showed strong nuclear p-STAT1 signals also at later time points (Figure 1A, arrows). Accordingly, SOCS1 and PDL1 mRNAs, two ISGs that are only transiently induced in hepatocytes, were also expressed at the 144-hour time point in nonparenchymal cells (Figure 2B).
We conclude that in hepatocytes, pegIFN-α2b induces a transient activation of the Jak/STAT signaling pathway during the first day, but not during the entire 1-week dosing interval, and this despite sustained high serum concentrations of pegIFN-α2b at all time points (Table 1). We found that nonparenchymal cells remained IFN-α sensitive at all time points investigated.
Induction of negative regulators of Jak/STAT signaling. We then assessed the induction of negative regulators of IFN signaling in the liver biopsies. On the mRNA level, SOCS1 was strongly induced at 4 hours and 16 hours, but then returned to pretreatment expression levels (Figure 3A). SOCS3 was also upregulated in the first 16 hours, albeit to a lesser extent (up to 2.5-fold) and remained slightly elevated for up to 4 days. USP18 was also rapidly induced, but unlike SOCS1 and SOCS3, the expression level of USP18 mRNA remained persistently high during the entire week (Figure 3A). Accordingly, USP18 protein was detectable from 16 hours on at all time points by Western blot and IHC analyses (Figure 3, B and C). Presumably for technical reasons, we could not detect SOCS1 or SOCS3 proteins at any time point, despite testing several different antibodies.
Figure 3The negative regulator USP18 is continuously upregulated during the entire week after pegIFN-α2b injection. (A) Bar plot indicating the mRNA expression fold change between the pretreatment biopsy (B1) and the on-treatment biopsy (B2) of SOCS1, SOCS3, and USP18. Data represent the mean with SEM (n = 6 for the 4-hour time point; n = 3 for all other time points). The black line indicates the baseline measured in pretreatment biopsies from the same patients (n = 18). (B) USP18 protein expression by Western blot analysis using whole-cell extracts of liver samples from B1 and B2. Patients are numbered according to Table 1. (C) Representative images of IHC for USP18 of liver biopsies obtained before treatment (B1) and at several time points after the first injection of pegIFN-α2b as indicated. Scale bars: 20 μm.
We conclude that pegIFN-α2b induces transient activation of the Jak/STAT signaling pathway in hepatocytes because of the rapid induction of SOCS1, SOCS3, and USP18 and that the signaling pathway remains refractory to ongoing stimulation by circulating pegIFN-α2b because of the persistent induction of USP18.
pegIFN-α2b–induced genes fall into four robust classes with distinct temporal expression patterns. We assessed pegIFN-α2b–regulated gene expression with transcriptome analysis using Affymetrix U133 Plus 2.0 arrays. Pairwise comparison of pretreatment and on-treatment biopsies revealed a greater than 2-fold induction in two-thirds of samples of hundreds of genes, with a peak at 16 hours (Figure 4A and Supplemental Table1; supplemental material available online with this article; doi: 10.1172/JCI70408DS1). Likewise, up to 200 genes were downregulated (Supplemental Table 2). To gain insight into the temporal expression patterns of ISGs induced by pegIFN-α2b in the human liver, we analyzed the transcriptome data using a Bayesian clustering algorithm. The algorithm produced four robust clusters of upregulated genes, which were termed early (144 genes), intermediate (31 genes), late (299 genes), and very late ISGs (20 genes) (Figure 4B and Supplemental Table 1). For over 95% of all upregulated genes, the peak mRNA levels occurred 4 or 16 hours after injection, followed by a steady decline over the remaining 128 hours of treatment (Figure 4B). Because of the limited amount of tissue obtained by percutaneous liver biopsies, we could not comprehensively analyze ISG protein expression. We therefore measured the protein expression of three exemplary ISGs. USP18 protein expression peaked at the 16-hour time point and then gradually declined, but remained induced up to the 144-hour time point. We found that USP18 mRNA expression peaked already at 4 hours, but was also persistently induced up to the 144-hour time point (Figure 3). STAT1 mRNA was induced up to the 96-hour time point, whereas STAT1 protein expression was still increased at 144 hours (Supplemental Figure 1). We found a very good correlation between IP10 mRNA expression in the liver and IP-10 protein concentration in the serum (Supplemental Figure 1). Taken together, we found a reasonably good correlation between mRNA and protein expression of ISGs in this limited set of exemplary ISGs.
Figure 4pegIFN-α2b–induced genes fall into four robust classes with distinct temporal expression patterns. (A) Number of genes greater than 2-fold up- or downregulated in two-thirds of the patients at each time point. (B) Clustering analysis of the upregulated genes produced four robust clusters (numbers 1–4) composed of early, intermediate, late, and very late ISGs. Boxes represent the quartiles, and whiskers represent 1.5 times the interquartile range. Bold line indicates the median expression value, and the number of genes in each cluster is indicated.
Compared with conventional IFN-α, pegIFN-α2b induces a broader range of genes including many ISGs involved in cellular immune responses. Given the known superior antiviral efficacy of pegIFN-α, we could not treat our study patients with conventional IFN-α. To compare IFN-α and pegIFN-α–induced gene regulation, we therefore made use of previously published transcriptome data obtained 24 hours after the injection of conventional IFN-α (18). Fortunately, the samples were analyzed on the same Affymetrix U133 Plus 2.0 arrays, allowing a direct comparison of the data. The discrepant time points after injection between the pegIFN-α and the IFN-α studies were a potential pitfall, but unsupervised hierarchical clustering of the combined data positioned the IFN-α2a samples properly between the 16-hour time point and the 48-hour time point of the pegIFN-α samples. Importantly, the magnitude of mRNA upregulation in the IFN-α samples was comparable to that in the pegIFN-α samples from the 16- and 48-hour time points. The most striking difference between IFN-α and pegIFN-α was the number of genes that were induced more than 2-fold in two-thirds of the samples (Figure 5A). We found that most of the genes upregulated by IFN-α were also induced by pegIFN-α, but a substantially larger number of genes were induced more than 2-fold exclusively by pegIFN-α. Gene ontological (GO) analysis revealed these to be genes associated with immune cells and adaptive immunity, whereas the genes upregulated by both IFN-α and pegIFN-α fell into the “classical” ISG group (Figure 5, B and C, and Supplemental Table 4). We conclude that while a common subset of ISGs is upregulated within the first 1–2 days independently of the IFN-α formulation, an additional set of genes associated with cellular immune responses is more markedly induced by pegIFN-α.
Figure 5IFN-α2a induces mainly “classical” ISGs, while pegIFN-α2b leads to transcription of additional immune cell–associated genes. (A) Venn diagram of genes identified as being upregulated by more than 2-fold in two-thirds of the patients at 16 or 48 hours after pegIFN-α2b injection (n = 3 each) or 24 hours after conventional IFN-α injection (n = 6). (B and C) Heatmaps show expression patterns (mean log2 fold change compared with paired pretreatment biopsies) of genes upregulated after IFN-α injection. (B) 43 ISGs were upregulated by conventional IFN-α2a at 24 hours as well as by pegIFN-α2b at both 16 and 48 hours. (C) 70 ISGs were upregulated by pegIFN-α2b at both 16 and 48 hours, but not by conventional IFN-α2a at 24 hours.
pegIFN-α2a and pegIFN-α2b induce overlapping sets of genes in the liver 144 hours after injection despite their different pharmacokinetic properties. Two different formulations of pegIFN-α2 with distinct pharmacokinetic properties are approved for the treatment of viral hepatitis: pegIFN-α2b (PegIntron) and pegIFN-α2a (PEGASYS). While the single-chain PEG moiety of pegIFN-α2b is subject to hydrolysis, which leads to release of IFN-α2b into the human body and faster elimination of the drug, pegIFN-α2a is not hydrolyzed, has a lower absorption rate, and is eliminated at a much slower rate (19). pegIFN-α2a achieves maximal serum levels of 7,000 pg/ml about 80 hours after administration, and the peak extends up to 168 hours after injection (20), as opposed to a much earlier peak (15–44 hours) and a more rapid decline of pegIFN-α2b. In order to investigate whether these distinct pharmacokinetic properties result in distinct pharmacodynamic effects, we included 3 additional patients in our study who were treated with pegIFN-α2a and obtained a second liver biopsy at the end of the 1-week dosing interval. As expected, pegIFN-α2a serum concentrations were still high at the end of the first week, whereas pegIFN-α2b concentrations declined in the second half of the dosing interval (Table 1). However, despite the difference in serum concentration between pegIFN-α2a and pegIFN-α2b, we found that the number of genes upregulated by greater than 2-fold in two-thirds of the patients in each group was not significantly different (59 versus 49 genes, respectively). Furthermore, we observed a considerable overlap of the gene sets, with 26 genes being upregulated by both pegIFN-α2a and pegIFN-α2b, and these common genes comprised most of the typical ISGs (Supplemental Table 3). We conclude that the different pharmacokinetic properties of the two pegIFN-α2 formulations do not cause significant differences in ISG expression at the end of a 1-week dosing interval.
pegIFN-α2b–induced gene transcription is mainly driven by IFN-stimulated response element motifs during the entire dosing interval. Among the hundreds of genes induced by IFN-α, one also finds several transcription factors such as IFN regulatory factors (IRFs) and cytokines and chemokines that could directly or indirectly activate additional signal transduction pathways and transcriptional programs (Supplemental Table 1). Such “secondary” transcription factors could be the drivers of gene transcription at later time points when pegIFN-α2b–induced Jak/STAT signaling is refractory. We therefore analyzed the relative contribution of transcription factor–binding motifs to global gene expression at 4 hours, 16 hours, 2 days, 4 days, and 6 days using a recently developed method called motif activity response analysis (MARA) (21). MARA infers the activities of transcription regulators by modeling genome-wide expression profiles in terms of computationally predicted binding sites for a large array of mammalian regulatory motifs such as IFN-stimulated response element (ISRE). Roughly speaking, MARA infers that a regulatory motif increases in activity when its predicted target promoters show an overall increase in expression that cannot be explained by the occurrence of other regulatory motifs in these promoters. In our current application, we used MARA to calculate changes in the activity of motifs across paired samples (pretreatment versus on-treatment). This analysis revealed ISRE as the most substantially changing motif across all time points up to 6 days (Figure 6, A and B). We observed a strong positive ISRE motif activity change in all patients (Figure 6A). MARA identified additional motifs that contribute to gene expression changes such as GAS, DMAP1_NCOR{1,2}_SMARC (DMAP1), PRRX1,2, and ATF6. However, the changes in their activities were relatively minor in comparison with ISRE (Figure 6B). MARA results of the transcription factor–binding site (TFBS) analysis were confirmed by motif discovery analysis using HOMER software (22). In each of the four ISG clusters (Figure 4B), ISRE was by far the most significantly enriched motif (Supplemental Figure 2).
Figure 6MARA reveals ISRE as the most significantly upregulated motif. (A) Activity changes of the ISRE motif in each patient, as inferred by MARA, showed a significant increase in ISRE activity for every patient at every time point. Shown are inferred activity changes (points) ± 1 SD (bars). (B) Motif activity profiles of the top five motifs with the most significant positive activity changes. Shown are the mean activity changes per time point (lines) ± 1 SEM as well as the P values and sequence logos of the motifs. (C) Fold change of IRF1, IRF7, and IRF9 mRNA expression for every patient. Shown are the mean values with SEM at each time point.
ISRE motifs are the binding sites for ISGF3 and also IRFs. ISGF3 is activated by IFN-α–induced phosphorylation of STAT1 and STAT2. IRFs are transcriptionally induced and are also regulated by phosphorylation (23). We therefore measured the expression of IRF mRNAs. Of the nine IRFs, only IRF1, IRF7, and IRF9 were upregulated by pegIFN-α2b in the liver (Figure 6C). IRF1 is transiently induced at the 4- and 16-hour time points. IRF9 is part of the ISGF3 complex, and its transcriptional activity depends on p-STAT1 and p-STAT2. IRF7 is upregulated during the entire dosing interval of pegIFN-α2b and could also be involved in ISRE-mediated gene transcription. However, the transcriptional activity of IRF7 depends on serine phosphorylation by IKK-α (24), a downstream component of cellular sensory pathways that are activated by viral pathogen–associated molecular patterns (PAMPs).
Unphosphorylated STAT1 does not prolong ISG induction. The central IFN-α–induced signal transducer and transcription factor STAT1 is itself one of the most strongly induced ISGs. Indeed, we found STAT1 mRNA strongly induced in the first 4 days after pegIFN-α2b injection (Supplemental Table 1). STAT1 protein was even upregulated during the entire 1-week dosing interval (Supplemental Figure 1). The functional significance of the expression of large amounts of U-STAT1 protein is unclear, but, intriguingly, a recent paper described a role of U-STAT1 as an active transcription factor that prolongs gene transcription after dephosphorylation of p-STAT1 (25). In that work, thirty ISGs were found to be upregulated by U-STAT1–driven transcription (25). We therefore hypothesized that U-STAT1 could be involved in the prolonged ISG induction by pegIFN-α2b. However, when we took the list of U-STAT1–induced genes and investigated their expression during the first week of pegIFN-α2b therapy, we did not find them to be overrepresented in clusters 3 and 4 with late and very late induced ISGs, respectively (data not shown). We therefore decided to address the potential of U-STAT1 to induce gene transcription in a more rigorous way. To that end, STAT1-deficient U3A cells (26) were stably transfected with STAT1 wild-type (STAT1-WT) or a mutant STAT1 with a phospho-tyrosine acceptor site at position 701 mutated to phenylalanine (STAT1-Y701F). For both STAT1-WT and STAT1-Y701F, three clones with different STAT1 expression levels were selected. One clone each expressed the transfected STAT1 at levels usually present in unstimulated parental 2fTGH cells, one clone each expressed the constructs at levels found after maximal STAT1 expression obtained in 2fTGH cells stimulated with IFN-α for 24 hours, and one clone each expressed the transfected constructs at intermediate levels (Figure 7A). As we expected, IFN-α treatment of U3A cells transfected with STAT1-WT induced ISGs. In contrast, we observed no ISG induction in U3A cells transfected with STAT1-Y701F (Figure 7B and Supplemental Figure 3). These results do not support a role for U-STAT1 in prolonged ISG expression.
Figure 7U-STAT1 does not induce ISGs. (A) Three clones with different expression levels of WT (WT cl1, WT cl2, and WT cl3) or mutated STAT1 (Y701F cl1, Y701F cl2, and Y701F cl3) were stimulated with IFN-α. STAT1-deficient U3A cells and STAT1 WT parental 2fTGH cells were used as controls. IFN-α induced STAT1 phosphorylation in 2fTGH and in all three WT clones. Actin is shown as a loading control. The cells were either untreated or treated for 15 minutes with 1,000 U/ml of IFN-α. WT cl1 and Y701F cl1 express STAT1 in an amount similar to that induced by 2fTGH cells treated for 24 hours with 1,000 U/ml of IFN-α. Shown are representative blots each from three independently performed experiments (black lanes separate blots that were derived from the same gel, but were noncontiguous). (B) IFN-α–induced OAS1 mRNA expression relative to GAPDH was assessed by qRT-PCR. Cells were treated with 1,000 U/ml IFN-α for 8 hours. Upregulation of OAS1 was found only in cells with WT STAT1 after IFN-α treatment. Expression of maximal amounts of Y701F-mutated STAT1 in U3A cells did not induce ISG expression. Shown are the mean values with SEM of three replicate experiments.
Ongoing gene transcription and lower mRNA decay rates both contribute to prolonged expression of “late” ISGs. Since ISRE seems to be the main TFBS in all transcription clusters and U-STAT1 was not able to induce ISGs, we next hypothesized that the genes belonging to the late ISG clusters might show prolonged expression due to lower mRNA degradation rates, since such a mechanism was recently proposed to play an important role in temporal expression patterns of genes induced by TNF-α (27). Decay of mRNAs can be regulated by specific microRNA recognition sequences present in the 3′ untranslated regions (UTRs) of mRNAs (28). We therefore analyzed our transcriptome datasets for specific binding sites of microRNAs to test whether the four ISG clusters defined by our unbiased clustering approach (Figure 4) have distinct microRNA binding sites in their 3′UTRs. However, we could not identify biologically meaningful microRNA binding patterns that would predict or explain the differences in decay rates of the four clusters (data not shown). We also analyzed the decay rates of mRNAs experimentally in IFN-α–treated Huh7 cells by inhibition of gene transcription with actinomycin D. Relative to GAPDH mRNA, early ISGs (RSAD2, USP18) showed faster mRNA decay, while the late ISGs (IFI27, LGALS3BP) decayed more slowly than GAPDH (Figure 8A). However, a delayed mRNA decay rate cannot readily explain the expression peaks at later time points such as those observed in cluster 4 genes (Figure 3). We therefore also analyzed the transcription of representative early (RSAD2, USP18) and late (IFI27, LGALS3BP) ISGs using a nuclear run-on assay. Nuclei were isolated from Huh7 cells after 1, 2, 4, 16, and 24 hours of stimulation with 1,000 IU/ml IFN-α and were then incubated with biotin-labeled UTP for 45 minutes. The newly transcribed mRNA was purified on streptavidin beads and quantified by quantitative PCR (qPCR). We found a markedly prolonged transcription of late versus early ISGs (Figure 8B).
Figure 8Late ISGs show a more prolonged transcriptional induction and a slower mRNA degradation rate than early ISGs in vitro. (A) mRNA degradation of early (RSAD2, USP18) and late (IFI27, LGALS3BP) ISGs was assessed in Huh7 after induction with 1,000 U/ml of IFN-α for 6 hours at the indicated time points. Transcription was blocked with actinomycin D, and the mRNA degradation over time was compared with GAPDH by qRT-PCR. Results from two independent experiments run in duplicate are shown. (B) Transcription rates of early (RSAD2, USP18) and late (IFI27, LGALS3BP) ISGs over time in Huh7 cells treated with 1,000 U/ml of IFN-α for the indicated time points. In vitro transcription in isolated nuclei was performed for 45 minutes. Newly transcribed mRNA labeled with biotinylated UTP was isolated and assessed by qRT-PCR. Results depicted as relative transcription compared with untreated baseline are shown from two independent experiments run in duplicate.
We conclude that the temporal expression patterns of ISGs are determined by the duration of gene transcription as well as by different mRNA decay rates.
-
Methods
Patients. The patients were recruited between March 2006 and April 2010 at the Hepatology Outpatient Clinic of the University Hospital Basel. Patients with CHC who underwent a biopsy for diagnostic purposes (B1) and provided written informed consent were screened for hepatic ISG expression. The four-gene classifier was used to assess the probability of an SVR (17). Patients with a high probability of achieving an SVR were asked to participate in the study, which included a second biopsy (B2) taken at a particular time point after the first therapeutic injection of pegIFN-α. We included 3 patients for each of the following time points: 16, 48, 96, and 144 hours, and additionally, the analysis included data on 6 patients from a previous study who had a biopsy at 4 hours (patients 1, 2, 6, 7, 8, and 9) (13). The patients received 1.5 μg/kg body weight pegIFN-α2b (Essex Chemie). Weight-adjusted ribavirin treatment was initiated only after the second biopsy to avoid confounding effects. An additional 3 patients treated with 180 μg of pegIFNα-2a (Roche) were included for the 144-hour time point study. Blood for serum analysis was taken at the time of the first and second biopsies. Serum HCV RNA was quantified using the COBAS AmpliPrep/COBAS TaqMan HCV Test and the COBAS AMPLICOR Monitor (Roche Molecular Systems).
Details regarding the 6 patients treated with IFN-α2a have been described previously (18). The patients included in the present analysis correspond to patients 2–7 in the original publication (18).
IL28B genotyping. DNA extraction and genotyping for the single nucleotide polymorphism rs12979860 near the IL28B gene were performed as described previously (17).
Measurement of serum proteins. Serum was collected before the first injection of pegIFN-α and at the time of the second biopsy. Serum levels of IFN-α2b and pegIFN-α2a were measured with an ELISA kit (Verikine 41100; PBL InterferonSource). Standard curves were prepared separately for pegIFN-α2a and -2b by a serial dilution starting at 12.5 pg/ml. The patient serum samples were diluted 10 times in sample diluent. IP-10 serum levels were measured with an ELISA (BD OptEIA Set Human IP-10, 2732KI; BD Biosciences) according to the manufacturer’s instructions.
IHC. Four-micrometer-thick serial sections were cut from formalin-fixed, paraffin-embedded liver biopsy specimens, rehydrated, pretreated for 20 minutes in ER2 solution, incubated with a monoclonal rabbit antibody against p-STAT1 (dilution 1:200, no. 9167; Cell Signaling Technology) or USP18 (1:100, catalog 4813; Cell Signaling Technology), and counterstained with hematoxylin. Standard indirect immunoperoxidase procedures were used for IHC (ABC-Elite; Vectra Laboratories). The staining procedure was performed with an automated stainer (Bond; Vision BioSystems).
RNA extraction and microarray hybridization. Total RNA was extracted from human liver tissue using QIAzol reagent and the RNeasy Mini Kit (QIAGEN) according to the manufacturer’s instructions. Gene expression was assessed by microarray analysis using Affymetrix Human Genome U133 Plus 2.0 arrays. Total RNA (1 μg) from each sample was reverse transcribed using a Genechip 3’IVT Express Kit (Affymetrix) according to the manufacturer’s instructions. The Hybridization and Wash Kit (Affymetrix) was used to hybridize the samples. All original array data are deposited in the NCBI’s Gene Expression Omnibus (GEO GSE48445).
RNA ISH. For the present study, we adapted a highly sensitive and specific ISH system (QuantiGene ViewRNA; Affymetrix). OCT-embedded and shock-frozen biopsies were cryosectioned (10-μm-thick sections) in a cryostat and mounted on Superfrost Plus Gold glass slides (Thermo Fisher Scientific). Upon fixation (4% formaldehyde, 16–18 hours at 4°C), washing, and dehydration in ethanol, the sections were pretreated by boiling for 1 minute in Pretreatment Solution, followed by a 10-minute digestion in Protease QF (both from Affymetrix). Sections were hybridized for 2 hours at 40°C with QuantiGene ViewRNA probes against MX1, IFI27, SOCS1, and PDL1 (Affymetrix). Bound probes were preamplified and subsequently amplified according to the manufacturer’s instructions. Labeled oligonucleotide probes conjugated with alkaline phosphatase (LP-AP) type 1 or type 6 were added, followed by the addition of fast red or fast blue substrate used to detect ISG mRNAs. Finally, the slides were counterstained with Meyer’s hematoxylin and embedded with DAPI-containing aqueous mounting medium (Roti-Mount FluorCare DAPI; Roth). Random images were acquired using a laser scanning confocal microscope (LSM710; Zeiss) and Zen2 software (Zeiss). All images were acquired with identical settings and saved in the Zeiss confocal file format (.lsm).
MARA. Here, we provide a brief description of MARA and its particular use in this work. For a detailed description of the general approach, the reader is referred to the FANTOM Consortium study (21).
To model the activity Am,s of a motif m in sample s, MARA uses a simple linear model that relates the number of binding sites Np,m in promoter p, for each of a large number of regulatory motifs m, to the expression Ep,s of promoter p in samples:
(Equation 1)
where cs represents the mean expression in sample s, and cp is the basal expression of promoter p. To determine promoter expression levels, we first computed the transcript expression levels by averaging weighted probeset signals (preprocessed as described above) over all probesets that matched a particular transcript as annotated by Affymetrix (Affymetrix annotation, Release 31 [NM accession RefSeqs only]). In this averaging, a probeset’s signal was weighted by the number of transcripts it matches. Subsequently, transcript expression levels were mapped to the human promoterome by averaging the weighted expression levels over all transcripts associated with a particular promoter. In this averaging, each transcript’s signal was weighted by the inverse of the number of promoters that express this particular transcript. To predict the TFBSs in each promoter, we used a curated set of transcription factor–binding motifs from SwissRegulon (50) with minor changes: since the SwissRegulon IRF1,2,7 motif only covered the IRF core consensus sequence, we replaced it with the ISRE (ThioMac-LPS-exp) motif to better cover the ISGF3 and IRF9 binding sites in our analysis. In addition, we replaced the three highly redundant STAT motifs from SwissRegulon with a single, high-quality GAS motif (HelaS3-STAT1-ChIP-Seq, as the GAS motif representative). Both motifs were obtained with HOMER software (22). By applying MARA, we obtained for every motif m in each sample s an expected activity Am,s and a corresponding error δm,s.
Determining donor-specific motif activity changes due to IFN-α treatment. For every donor and every motif m, we calculated the difference between the motif activity before IFN-α treatment (Abm) and the motif activity after IFN-α treatment: (Aam)
(Equation 2)
as well as the corresponding SD:
(Equation 3)
Thus, for every motif m and each donor d, the expression data D imply an expected activity change AΔm,d with corresponding SD σΔm,d. Consequently, the probability of the data D, assuming a true (unobserved) activity change σΔm,d, is a Gaussian with the expected mean AΔm,d and SD σΔm,d:
(Equation 4)
where σΔm,d is the true (unobserved) activity change.
Determining mean motif activity changes due to pegIFN-α treatment at certain time points. To obtain mean activity changes for every group of donors g ε G whose second biopsy was taken at an equal time point after pegIFN-α treatment, we assumed that the activity changes AΔm of motif m were Gaussian distributed (with mean AΔm,g and variance (σΔm,d)2). Accordingly, the probability of an activity change σΔm,d in donor d is:
(Equation 5)
For each donor d ε g, we combine Equations 4 and 5 and integrate out all unknown (true activity changes) σΔm,d so that we can calculate the probability of the data D given the mean activity of the group AΔm,g and the corresponding error σΔm,g:
(Equation 6)
Solving these integrals analytically gives:
(Equation 7)
We then numerically determine the value σΔ*m,g that maximizes Equation 7. Assuming a uniform prior for AΔm,g, we obtain an expression for the posterior probability P(AΔm,g |D), which is a Gaussian with mean
(Equation 8)
and SEM
(Equation 9)
where σΔ*m,g is the maximum likelihood estimate of Equation 7. We call A Δ –m,g the mean activity change for group g and σ Δ –m,g the corresponding SEM. To obtain a measure for the significance of the mean activity change for group g, we calculate a corresponding z value:
(Equation 10)
Finally, a global z value considering all time points is given by:
(Equation 11)
To calculate a P value for the calculated z value
, we used the null hypothesis that the z statistic in each group g, i.e., the ratio between the motif activity change and its standard error was drawn from a Gaussian with mean zero and variance 1. Under this null hypothesis, the distribution of the statistic,
(Equation 12)
with G representing the number of groups (wherein each group represents a time point), is gamma distributed, and we used this distribution to calculate the P value corresponding to the z value of each of our motifs.
TFBS analysis. TFBS analysis was carried out using HOMER software (22) ( http://biowhat.ucsd.edu/homer/motif/index.html). Briefly, promoter regions (2 kbp upstream and 500 bp downstream of the transcription start site) of all genes within each cluster were screened for known TFBS. Enrichment of TFBS in our gene lists relative to all human promoter regions was assessed by hypergeometric tests.
Quantitative real-time RT-PCR. RNA was reverse transcribed by Moloney murine leukemia virus reverse transcriptase (Promega) in the presence of random primers (Promega) and deoxynucleoside triphosphate. The samples were incubated for 5 minutes at 70°C and then for 1 hour at 37°C. The reaction was stopped by heating at 95°C for 5 minutes. SYBR real-time PCR was performed using the SYBR Green PCR Master Mix (Applied Biosystems). Intron-spanning primers for GAPDH, HERC6, IFI27, IFI44L, ISG15, LGALS3BP, MX1, OAS1, OAS2, RSAD2, and USP18 were used (Supplemental Table 5). All reactions were performed in duplicate on an ABI 7500 Real-Time PCR System (Applied Biosystems). mRNA expression levels of the transcripts were normalized to GAPDH using the ΔCt method.
Western blot analysis. Whole-cell extracts and blotting of human liver samples and cells were performed as described (13). The membranes were incubated with primary antibodies against p-STAT1 (1:1,000, catalog 9171; Cell Signaling Technology), STAT1 (1:1,000, catalog 610116; BD Transduction Laboratories), USP18 (1:1,000, no. 4813; Cell Signaling Technology), and β-actin (1:2,000, A5441; Sigma-Aldrich) diluted in Tris-buffered saline containing Tween-20 (TBST) overnight at 4°C. After three washes with TBST, membranes were incubated for 1 hour at room temperature with fluorescent secondary goat anti-mouse (IRDye 680) or anti-rabbit (IRDye 800) antibodies (both from LI-COR Biosciences). Blots were scanned using the Odyssey Infrared Imaging System (LI-COR Biosciences).
Cell culture. Huh7 cells were maintained in DMEM (Gibco) supplemented with 10% FBS. 2fTGH and U3A STAT1–/– cells were maintained in DMEM with 10% FBS and 250 μg/ml of hygromycin B (Sigma-Aldrich). The stably transfected U3A STAT1–/– cells were selected with 800 μg/ml of G418 (catalog 345810; Calbiochem). Cells were treated with 1,000 U/ml human IFN-α (Roferon; Roche) and/or with 5 μg/ml actinomycin D (Sigma-Aldrich).
Site-directed mutagenesis and transfection. The STAT1-flag-pcDNA3 (STAT1-WT) was provided by J.E. Darnell (Rockefeller University, New York City, New York, USA). STAT1 (Y701F)-flag-pcDNA3 was generated from STAT1-flag-pcDNA3 using the method described by Mikaelian and Sergeant (51). Briefly, two consecutive PCR reactions with 30 cycles were performed using 20 ng of template DNA, 200 μM dNTP, 1 U of Pfu DNA polymerase (Promega), and 5 μM of each of the following primers: 5′-CTGGCACCAGAACGAATGA-3′; 5′-ATTTAGGTGACACTATAG-3′; 5′-GGAACTGGATTCATCAAGACTGAG-3′; and 5′-CTCAGTCTTGATGAATCCAGTTC-3′, in a final volume of 25 μl. The amplified products were loaded on a 1.5% agarose gel, excised, digested by BlpI and ApaI, and ligated into STAT1-flag-pcDNA3, previously cut with the same restriction enzymes. Mutation of Tyr701 to Phe was confirmed by sequencing.
U3A STAT1–/– cells were transfected with 1 μg of the respective plasmid using Fugene HD (Roche) according to the manufacturer’s instructions. Cells were selected with 800 μg/ml of hygromycin B (Roche) for 15 days, and single clones were chosen.
Nuclear run-on assay. After IFN-α treatment, cells were washed with 1× PBS, treated with 0.25% trypsin for 4 minutes (Gibco), suspended in 10 ml of ice-cold diethylpyrocarbonate-treated (DEPC-treated) 1× PBS, and concentrated by centrifugation (160 g, 10 minutes). Cells were then washed once with ice-cold buffer 1 containing 10 mM Tris-HCl (pH 7.4), 150 mM KCl, and 8 mM Mg acetate, centrifuged at 530 g for 10 minutes, and subsequently lysed with buffer 1 with the addition of 0.5% Igepal (Sigma-Aldrich) for 10 minutes at 4°C. Nuclei were then isolated by a sucrose gradient (600 mM), then washed and suspended in buffer containing 40% glycerol. Nuclei were immediately used for the run-on assay. Nuclei (5 × 106) were incubated in reaction buffer containing 5 mM Tris-HCl (pH 8.0), 2.5 mM MgCl2, 150 mM KCl, and 2.5 mM each of ATP, GTP, CTP, UTP, and biotin-16-UTP (Roche) for 45 minutes at 30°C. RNA was then isolated with TRIzol according to the manufacturer’s instructions. Subsequently, biotinylated RNA was purified with streptavidin-coupled beads (Dynabeads M-280; Invitrogen) according to the manufacturer’s instructions, and RNA was again isolated with TRIzol.
Statistics. Microarray analysis was performed with Bioconductor packages within the R statistical environment (52). Data were preprocessed using the standard RMA algorithm. Batch effects observed between the human liver samples that were processed and hybridized at different times were corrected using the ComBat algorithm (53). Probesets with very low expression intensities (below 80 in the highest-expressing sample) as well as the control probesets were excluded from the subsequent analyses. The list of significantly regulated probesets was compiled as follows: (a) probesets showing more than a 2-fold difference in levels between the B1 and B2 samples taken from the same patient were selected; (b) for every time point, the probesets that changed in two-thirds of the patients were retained. Probesets fulfilling those criteria were included in the clustering analysis. The expression data were normalized so that total expression levels did not affect the grouping of the probesets. An infinite Gaussian mixture model with a Dirichlet process prior was used to produce the gene clusters. This nonparametric model suggests a growing number of Gaussians to describe the gene expressions. With the special choice of a Dirichlet process prior, the number of clusters need not be fixed in advance, but is adaptively chosen based on the observed data. The results were tested for robustness by moderately changing the hyperparameters that control the Dirichlet process.
Enrichment of Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways and GO biological process terms were assessed using DAVID software, version 6.7. Additional statistical analyses using a 2-tailed Student’s t test were carried out using GraphPad Prism software, version 6.0 (GraphPad Software). A P value of less than 0.05 was considered significant.
Study approval. All patients provided written informed consent to participate in the study, which was approved by the ethics committee of Basel.
Copyright © 2025 American Society for Clinical Investigation
ISSN: 0021-9738 (print), 1558-8238 (online)