Gene Expression Signatures, Pathways and Networks in Carotid Atherosclerosis

Original Article

doi: 10.1111/joim.12448 Gene expression signatures, pathways and networks in carotid atherosclerosis

L. Perisic1, S. Aldi1,*, Y. Sun2,*, L. Folkersen3,4, A. Razuvaev1, J. Roy1, M. Lengquist1,S.Akesson1, C. E. Wheelock5, L. Maegdefessel4, A. Gabrielsen4, J. Odeberg4,6, G. K. Hansson4, G. Paulsson-Berne4 & U. Hedin1

From the 1Department of Molecular Medicine and Surgery, Karolinska Institute; 2Translational Science Center, Personalized Healthcare and Biomarkers, R&D, Astra Zeneca, Stockholm, Sweden; 3Department of Molecular Genetics, Novo Nordisk, Copenhagen, Denmark; 4Department of Medicine, Karolinska Institute; 5Division of Physiological Chemistry 2, Department of Medical Biochemistry and Biophysics, Karolinska Institute; and 6Science for Life Laboratory, Department of Proteomics, School of Biotechnology, Royal Institute of Technology, Stockholm, Sweden

Abstract. Perisic L, Aldi S, Sun Y, Folkersen L, patients on statins, as well as inflammation and Razuvaev A, Roy J, Lengquist M, Akesson S, apoptosis in plaques removed >1 month compared Wheelock CE, Maegdefessel L, Gabrielsen A, to <2 weeks after symptom. By prediction profiling, Odeberg J, Hansson GK, Paulsson-Berne G, a panel of 30 genes, mostly transcription factors, Hedin U (Karolinska Institute, Stockholm; R&D, discriminated between plaques from S versus AS Astra Zeneca, Stockholm, Sweden; Novo Nordisk, patients with 78% accuracy. By meta-analysis, Copenhagen, Denmark; Karolinska Institute, common gene networks associated with Stockholm; Karolinska Institute, Stockholm; atherosclerosis mapped to hypoxia, chemokines, Royal Institute of Technology, Stockholm, calcification, actin cytoskeleton and extracellular Sweden). Gene expression signatures, pathways matrix. A set of dysregulated genes (LMOD1, and networks in carotid atherosclerosis. J Intern SYNPO2, PLIN2 and PPBP) previously not described Med 2015; doi: 10.1111/joim.12448. in atherosclerosis were identified from microarrays and validated by quantitative PCR and immuno- Background. Embolism from unstable atheromas in histochemistry. the carotid bifurcation is a major cause of stroke. Here, we analysed gene expression in endarterec- Conclusions. Our findings confirmed a central role tomies from patients with symptomatic (S) and for inflammation and proteases in plaque insta- asymptomatic (AS) carotid stenosis to identify bility, and highlighted haemoglobin metabolism pathways linked to plaque instability. and bone resorption as important pathways. Subgroup analysis suggested prolonged inflam- Methods. Microarrays were prepared from plaques mation following the symptoms of plaque insta- (n = 127) and peripheral blood samples (n = 96) of bility and calcification as a possible stabilizing S and AS patients. Gene set enrichment, pathway mechanism by statins. In addition, transcrip- mapping and network analyses of differentially tional regulation may play an important role in expressed genes were performed. the determination of plaque phenotype. The results from this study will serve as a basis for Results. These studies revealed upregulation of further exploration of molecular signatures in haemoglobin metabolism (P = 2.20E-05) and bone carotid atherosclerosis. resorption (P = 9.63E-04) in S patients. Analysis of subgroups of patients indicated enrichment of Keywords: atherosclerosis, carotid plaques, gene calcification and osteoblast differentiation in S expression, microarrays.

cardiovascular disease. In order to improve identi- Introduction ﬁcation of individuals at risk and to facilitate the Atherosclerotic plaque rupture followed by myocar- development of future preventive therapeutic dial infarction or stroke is the major cause of strategies, there is a need for better understanding morbidity and mortality in patients with of the pathophysiology underlying the transition of atherosclerosis from clinically asymptomatic (AS) *These authors contributed equally.

ª 2015 The Association for the Publication of the Journal of Internal Medicine 1 L. Perisic et al. Atherosclerosis transcriptome

and stable lesions to symptomatic (S) and unstable clinical categories. Hence, factors determining disease [1]. plaque vulnerability and healing remain largely unknown. The unstable atheroma has previously been characterized by enhanced activity of proinflammatory The Biobank of Karolinska Endarterectomy (BiKE) T cells and macrophages, increased cytokine was established in 2002 for prospective collection release and secretion of matrix metalloproteinases of atherosclerotic plaque tissue and blood from (MMPs), thinning of the fibrous cap due to impaired patients undergoing carotid endarterectomy for collagen fibre formation, collagenolysis and cell high-grade carotid artery stenosis, with a database apoptosis. Together with haemodynamic forces, of clinical parameters including risk factors, med- these processes lead to fibrous cap rupture, expo- ication, symptoms, time of surgery, preoperative sure of tissue factor, atherothrombosis and clinical imaging and anthropometric and laboratory mea- manifestations [1, 2]. surements. Transcriptomic and genomic profiles have been generated by microarray and genotyping Embolization of ruptured plaques in the carotid chips, enabling multivariate analysis of gene bifurcation to the cerebral circulation can occur, expression patterns in relation to clinical parame- causing transient or permanent ischaemia. ters and genotype. Data from this biobank have Patients with high-grade carotid stenosis are been utilized previously, both in hypothesis-driven treated with stroke preventive interventions, but and hypothesis-independent studies [16–31]. In neither laboratory tests nor imaging have been this study, we postulated that phenotypes of clinically established for precise detection of unsta- patients with carotid atherosclerosis would corre- ble individuals or lesions [3]. A number of clinical late with specific gene expression patterns and we parameters influence the risk of stroke in patients analysed the global transcriptome in relationship with carotid stenosis. AS carotid disease is associ- to clinical features stratifying stroke risk. We have ated with an annual stroke risk of 1–3%, whereas S been able to identify novel molecular signatures patients with either transient ischaemic attack and pathways of plaque instability, validate (TIA) or minor stroke (MS) have a more than 10- established processes featuring late-stage fold increased risk [4]. In addition, it has been atherosclerosis as well as to highlight the temporal shown that TIA and MS confer a higher risk than dynamics and modulation of atherosclerosis by ocular symptoms [retinal TIA, amaurosis fugax; statin therapy. (AF)] [3]. Recurrent embolization and symptoms are most pronounced within the first 2 weeks after Materials and methods the index event [5, 6] and are thereafter gradually Human material reduced. Medical therapy with HMG CoA reductase inhibitors (statins) significantly reduces stroke Patients undergoing surgery for S or AS, high- risk, probably by improving plaque stability [7, 8]. grade (>50% North American Symptomatic Carotid Endarterectomy Trial, NASCET) [32] carotid steno- Large-scale transcriptomic analyses using sis at the Department of Vascular Surgery, Karolin- microarrays enable studies of expression patterns ska University Hospital, were consecutively to be performed for thousands of genes simultane- enrolled in the study. Informed consent was ously [9]. Although alternative technologies such obtained, and clinical data were recorded on as RNA sequencing are becoming available, admission. The following clinical parameters were microarrays remain powerful tools for investiga- selected for investigation in this study: presence tions of disease-related gene expression changes, and type of symptoms; statin treatment; and time particularly when combined with open-source soft- between last symptom and surgery. Symptoms of ware for high-throughput data exchange (e.g. Gene plaque instability were defined as TIA, MS and AF. Expression Omnibus (GEO) data sets and ArrayEx- TIA was defined as a transient episode of neuro- press) and a growing number of web-based pro- logical dysfunction, and MS as symptoms persist- grams for functional analysis. Previous attempts to ing >24 h with complete or nearly complete investigate global gene expression profiles in recovery. human atherosclerotic plaques [10–15] have been restricted by small sample sizes, the lack of Patients without qualifying symptoms within undiseased (nonatherosclerotic) vascular control 6 months prior to surgery were categorized as AS tissues or the inability to classify patients based on and indication for carotid endarterectomy based on

2 ª 2015 The Association for the Publication of the Journal of Internal Medicine Journal of Internal Medicine L. Perisic et al. Atherosclerosis transcriptome

results from the Asymptomatic Carotid Surgery (a) Trial (ACST) [33]. Patients with atrial ﬁbrillation and those who had suffered a major stroke were excluded from the study. Carotid endarterectomies (carotid plaques, CP) and peripheral blood samples were collected at surgery and retained within the BiKE. Plaques were divided transversally by the surgeon at the most stenotic part, and the proximal half of the lesion used for RNA preparation and the distal half ﬁxed in 4% Zn-formaldehyde and processed for histology. RNA from tissues and peripheral blood mononuclear cells (PBMCs) was analysed by gene expression microarrays, and gene expression in plaques was validated by quantitative PCR (qPCR). A summary of the samples used in the study is provided in Fig. 1a, and demographics of the cohort are given in Table S1. Nonatherosclerotic, further referred to as normal (b) artery (NA), control samples were obtained from nine macroscopically disease-free iliac arteries and one aorta from organ donors without a history of cardiovascular disease. The vessels were dissected and the intima and media used for RNA isolation (c) and arrays. NAs for immunohistochemistry (IHC) were radial arteries obtained at coronary bypass surgery and one control internal carotid artery from a 61-year-old man undergoing surgical exci- sion of a neck tumour. All samples were collected with informed consent from patients and organ donors or their guardians. The study was approved by the regional ethics committee.

Antibodies Fig. 1 The Biobank of Karolinska Endarterectomy (BiKE) A summary of the primary antibodies kindly pro- data sets, workflow for microarray comparisons and vided by the Human Protein Atlas (HPA) project is distribution of samples. (a) Outline of the subcohorts from shown in Table S2 with information about the the BiKE used for microarray profiling. (b) Schematic representation of the comparisons performed from the corresponding gene function and tissue expression discovery data set within clinically defined groups. (c) pattern obtained from www.proteinatlas.org. For Principal components analysis of discovery data set from colocalization, the following antibodies for cell plaque tissues for symptomatic and asymptomatic type-specific markers were used as recommended patients based on expression values shows random by the manufacturers: anti-alpha smooth muscle scattering of the samples in groups and no clustering. actin (aSMA; #M0851; DAKO, Glostrup, Denmark), The first two principal components (PC1 and PC2) explain CD61 (#M0753; DAKO) and CD68 (#NCL-L-CD68; 23% and 22%, respectively, of variance amongst samples Novocastra, Newcastle, UK). in this comparison. PBMC, peripheral blood mononuclear cell; S, symptomatic; AS, asymptomatic; qPCR, quantitative PCR; TIA, transient ischaemic attack; MS, minor RNA extraction stroke; AF, amaurosis fugax. Plaques were frozen at À80 °C immediately after surgery, and peripheral blood was processed through Ficoll-Paque preparation tubes (Vacu- according to standard procedures for separation tainer CPT, Becton-Dickinson, Franklin Lakes, of the mononuclear cell fraction and RNA isolation. NJ, USA). PBMCs were resuspended using RLT Briefly, whole blood was collected at surgery and buffer (Qiagen, Hilden, Germany) before freezing at subjected to density gradient centrifugation À80 °C. RNA was prepared using Qiazol Lysis

ª 2015 The Association for the Publication of the Journal of Internal Medicine 3 Journal of Internal Medicine L. Perisic et al. Atherosclerosis transcriptome

Reagent (Qiagen) and purified using the RNeasy Cytoscape for protein–protein interactome network Mini kit (74106, Qiagen), including DNase diges- analyses; MSigDB (Broad Institute) for mapping of tion. The concentration of RNA was measured transcription factors and microRNA motif binding using Nanodrop ND-1000 (Thermo Scientific, sites (FDR set to 1%). Information about individual Waltham, MA, USA) and quality estimated using genes and proteins was extracted from www.- a Bioanalyser capillary electrophoresis system genecards.org, www.noncode.org and www.unipro- (Agilent Technologies, Santa Clara, CA, USA). For t.org. Prediction modelling was performed based on microarrays, only RNA of good integrity (RNA sparse partial least square-discriminant analysis Integrity Number >7), A260/A280 ratio 1.8–2.1, (sPLS-DA) [39]. Statistical resampling was used to A260/230 ratio 0.7–1.5 and concentration about balance the number of S and AS patients in the À 50–500 ng lL 1 was used, according to recom- analysis. In order to optimize the prediction per- mended standards for whole-transcript arrays. formance, a nested cross-validation machine learning method was applied: the samples were randomly divided into one training set and one test Microarray analysis set as an independent hold-out set for validation. Gene expression analyses of 127 endarterectomy Within the training set, the samples were randomly samples were performed in two batches using divided into subtraining sets and subtest sets. The Affymetrix HG-U133 plus 2.0 Genechip arrays sPLS-DA model was built on the training set. The (Santa Clara, CA, USA). A separate analysis was prediction error rate was estimated by classifying a performed on an additional 50 samples using subset of test samples. The prediction accuracy Affymetrix HG-U133a Genechip arrays. Robust was determined by performing a 10-fold cross- multi-array average normalization, filtering of validation. New sPLS-DA models were built probesets based on signal intensity and batch sequentially by including different numbers of effect correction were performed, and processed variables in the model, and prediction error rates gene expression data were recorded on a log2 scale were determined. Then, the final sPLS-DA model as previously described [16, 23]. The full data set is was constructed with the minimum number of available from Gene Expression Omnibus (acces- variables selected for achieving the smallest pre- sion number GSE21545). diction error rate. One dimension of sPLS-DA was used in the model. The independent hold-out test set was predicted based on this final model. Data analyses Data set analyses were performed with GraphPad qPCR Prism 6 and Bioconductor using a linear regression model adjusted for age and gender and a two-sided For qPCR, total RNA was reverse-transcribed Student’s t-test assuming nonequal deviation, with using the High-capacity RNA-to-cDNA Kit correction for multiple comparisons performed (4387406; Applied Biosystems, Life Technologies according to the method of Sidak–Bonferroni [34]. Corporation, Carlsbad, CA, USA). PCR amplifica- Spearman correlations with Benjamini–Hoechberg tion was performed in 96-well plates using the correction were calculated to determine the asso- 7900 HT real-time PCR system (Applied Biosys- ciation between mRNA expression levels from tems), TaqMan Universal PCR Master Mix (Applied microarrays. In all analyses, P-values <0.05 were Biosystems) and TaqMan Gene Expression Assays considered statistically significant, although raw P- (LMOD1 Hs00201704_m1, SYNPO2 Hs003264 values are also reported. The following publically 93_m1, PPBP Hs00234077_m1, PLIN2 Hs00605 available programs were used for functional anal- 340_m1; Applied Biosystems). All samples were yses of the microarray data sets: ArrayMining [35], measured in duplicate. Results were normalized GOrilla and DAVID for gene set enrichment anal- relative to the equal mass of total RNA as well as yses (GSEAs) based on functional grouping of the Ct values of the RPLPO housekeeping control genes defined in gene ontology (GO) categories gene (probe Hs99999902_m1). The relative [raw P-values reported with false discovery rate amount of target gene mRNA was calculated by (FDR) correction; categories with more than one the 2 (ÀDDCt) method and presented as fold gene and nonoverlapping gene sets considered]; change compared to the average baseline expres- INMEX for meta-analysis using Fisher’s weight-free sion in normal arteries. The qPCR results were method for combined P-values [36]; GeneMania for evaluated using the nonparametric Mann–Whitney functional network construction; PINA [37, 38] and U-test.

4 ª 2015 The Association for the Publication of the Journal of Internal Medicine Journal of Internal Medicine L. Perisic et al. Atherosclerosis transcriptome

groups roughly cluster together with 23.6% and IHC 22.2% variability explained by the first two princi- All IHC reagents were purchased from Biocare pal components. These results emphasize subtle Medical (Concord, CA, USA). Isotype rabbit and biological differences that reflect the overall gene mouse IgG were used as negative controls. In brief, expression patterns in plaques from S and AS 5-lm sections were deparaffinized in Tissue Clear patients. and rehydrated in a graded series of ethanol concentrations. For antigen retrieval, slides were Hb metabolism and bone resorption are enriched in plaques from S subjected to high-pressure boiling in DIVA buffer patients (pH 6.0). After blocking with Background Sniper, primary antibodies diluted in Da Vinci Green Analyses of plaques isolated from S and AS solution were applied and incubated at room patients revealed small differences in gene expres- temperature for 1 h. A double-stain probe–polymer sion levels (up to 2-fold change). The fraction of system containing alkaline phosphatase and differentially regulated probesets was <5%. In this horseradish peroxidase was applied, with subse- analysis, many genes were identified that have quent detection using Warp Red and Vina Green. previously not been reported in similar compar- Slides were counterstained with Haematoxylin QS isons, for example ADM, PPBP, MARCO and MREG (Vector Laboratories, Burlingame, CA, USA), (Table 1; for full list see Table S3), in addition to dehydrated and mounted in Pertex (Histolab, those already established in atherosclerosis, such Gothenburg, Sweden). Images were captured using as matrix metalloproteinases (MMPs), markers of a Nikon OPTIPHOT-2 microscope equipped with a macrophages, chemokines, cathepsins and digital camera and processed with NIS-Elements apolipoproteins that were upregulated and mark- software. ers of differentiated smooth muscle cells (SMCs) that were downregulated [10, 14, 23, 40]. GSEA on Results GO categories for the differentially expressed genes in plaques from S patients showed enrichment of Characterization of gene expression differences in patient subgroups low-density lipoprotein particle clearance, inflam- Plaques from 127 consecutively enrolled patients matory response, macrophage-derived foam cell (40 AS and 87 S) were analysed by gene expression differentiation, cell proliferation, apoptosis, angio- microarrays (discovery data set). An overlapping genesis and extracellular matrix (ECM) disassem- subset of 96 PBMC preparations (30 AS and 66 S bly. It is interesting that enrichment was also patients) was separately analysed by microarray observed for positive regulation of bone resorption profiling. A nonoverlapping set of 50 plaques (10 (Tables 2 and S5). There were also small differ- AS and 40 S patients) was analysed with a different ences in gene expression (up to 1.5-fold change; microarray platform and used as an independent overall about 1% differentially regulated probesets; validation cohort. In addition, RNA from 77 Tables 1 and S4) in PBMCs from S versus AS nonoverlapping plaques (21 AS and 56 S patients) patients. Here, we observed upregulation of a was used for qPCR validation of the microarray number of genes including HBB, CA1, IFNG, F8, results. A summary of these cohorts and the study ALAS2, ARG2, ARSD and SELENBP1. Apart from workflow is shown in Fig. 1a. HBB, several genes implicated in erythropoiesis were also upregulated, particularly GATA2, ERAF Pairwise comparisons of microarray profiles from and KLF1. Even genes associated with Hb catabo- the discovery data set were assessed as outlined in lism were upregulated, including iron-binding PIR, Fig. 1b, by grouping the patients into clinically OGFOD1 and several members of the cytochrome defined categories. The majority of the study cohort P450 family such as CYP2B6 and CYP26A1. The was male (78%); S patients were generally older most enriched pathways were response to nitrosa- than AS patients (mean age 72.5 vs. 66.4, tive stress, positive regulation of nitric oxide P = 0.0002) and had significantly lower mean biosynthesis, IL1B secretion, lymphocyte À haemoglobin (Hb) levels (131.55 vs. 141.62 g L 1, activation and Hb metabolism (Tables 2 and S5). P = 0.0077). The demographic characteristics are shown in detail in Table S1. Principal components We also assessed gene expression differences in analysis, performed to give an overview of the subgroups of clinical phenotypes of S patients, expression data distribution in S and AS tissue which consequently reduced sample size and samples (Fig. 1c), showed that these patient statistical power. Gene expression profiles in

ª 2015 The Association for the Publication of the Journal of Internal Medicine 5 Journal of Internal Medicine L. Perisic et al. Atherosclerosis transcriptome

Table 1 Most signiﬁcantly upregulated genes in microarray comparisons

Affymetrix ID Gene Gene product P- value Fold change Plaques S vs. AS (n = 87 + 40) 210072_at CCL19 Chemokine (C-C motif) ligand 19 0.002 1.871 209396_s_at CHI3L1 Chitinase 3-like 1 (cartilage glycoprotein-39) 0.002 1.817 203980_at FABP4 Fatty acid binding protein 4, adipocyte 0.046 1.759 212657_s_at IL1RN Interleukin 1 receptor antagonist 0.021 1.712 202859_x_at IL8 Interleukin 8 0.006 1.705 223484_at C15ORF48 Chromosome 15 open reading frame 48 0.010 1.702 204580_at MMP12 Matrix metallopeptidase 12 (macrophage elastase) 0.026 1.694 203936_s_at MMP9 Matrix metallopeptidase 9 (gelatinase B, 92-kDa 0.016 1.621 gelatinase, 92-kDa type IV collagenase) 203665_at HMOX1 Haeme oxygenase (decycling) 1 0.013 1.620 205819_at MARCO Macrophage receptor with collagenous structure 0.006 1.602 PBMCs S vs. AS (n = 66 + 30) 1562981_at HBB Haemoglobin, beta 0.027 1.507 205949_at CA1 Carbonic anhydrase I 0.023 1.421 235867_at GSTM3 Glutathione S-transferase mu 3 (brain) 0.027 1.392 238813_at ALAS2 Aminolevulinate, delta-, synthase 2 0.043 1.380 221690_s_at NLRP2 NLR family, pyrin domain containing 2 0.021 1.333 219672_at ERAF Erythroid-associated factor 0.022 1.331 210354_at IFNG Interferon-gamma 0.030 1.310 229437_at MIR155HG MIR155 host gene (nonprotein coding) 0.034 1.296 222793_at DDX58 DEAD (Asp-Glu-Ala-Asp) box polypeptide 58 0.035 1.289 219385_at SLAMF8 SLAM family member 8 0.012 1.273

Top 10 genes sorted by P-value and fold change for each comparison are shown. P-values are adjusted for multiple comparisons; expanded lists of genes are shown in the Tables S3 and S4. PBMC, peripheral blood mononuclear cell; S, symptomatic; AS, asymptomatic. patients with TIA and MS were compared to those comparisons (Table S7). Of note, prolonged in patients with AF because it has been shown that immune system activation and apoptosis were patients with TIA and MS have an increased stroke ongoing in plaques extracted 1 month or later after risk [41]. Small differences in expression levels (up symptom as compared to those removed within the to 1.5-fold change) and in the number of differen- first 2 weeks (Table S9). We also investigated tially regulated probesets (about 3%; Table S6) differences between S patients treated with statins were observed. Here, several genes associated with compared to those who did not receive this therapy. platelet activation, such as TUBB1, PTAFR, As expected, serum cholesterol and LDL choles- TREML1 and platelet-specific GP6, were strongly terol levels were significantly decreased in the upregulated in patients with TIA and MS. GSEA treated group (mean difference Æ SD À0.7204 Æ À highlighted angiogenesis, oxidoreductase activity, 0.3310 mmol L 1, P = 0.03 and À0.8782 Æ À secretion, chemotaxis and inflammation as pro- 0.3321 mmol L 1, P = 0.01, respectively). Compar- cesses enriched in plaques of S patients with TIA ison of gene expression in lesions from S patients and MS (Table S9). Categorization of patients with with and without statin treatment showed modest respect to time from last qualifying symptom to differences in fold change (up to 2.5-fold), but an surgery surprisingly yielded marked differences in abundance of differentially expressed probesets gene expression. A higher relative fold change (up (about 15%; Table S8). Notably, regulation of to 10-fold) and more differentially expressed probe- calcium-mediated signalling, osteoblast differenti- sets (about 20%) were found in this than in other ation and negative regulation of angiogenesis and

6 ª 2015 The Association for the Publication of the Journal of Internal Medicine Journal of Internal Medicine L. Perisic et al. Atherosclerosis transcriptome

Table 2 Gene set enrichment analysis (GSEA) based on gene ontology (GO) processes enriched in microarray comparisons

GO term Description P-value Enrichment Plaques S vs. AS (n = 87 + 40) GO:0006935 Chemotaxis 9.68E-08 1.66 GO:0007155 Cell adhesion 8.30E-07 1.43 GO:0002523 Leucocyte migration involved in inﬂammatory response 3.22E-06 26.79 GO:0034383 Low-density lipoprotein particle clearance 9.66E-06 422.67 GO:0030593 Neutrophil chemotaxis 4.99E-05 2.23 GO:0008284 Positive regulation of cell proliferation 1.26E-04 1.39 GO:0002544 Chronic inﬂammatory response 1.34E-04 14.92 GO:0043277 Apoptotic cell clearance 2.65E-04 25.47 GO:0033005 Positive regulation of mast cell activation 3.01E-04 9.65 GO:0010743 Macrophage-derived foam cell differentiation 7.12E-04 8.89 GO:0048514 Blood vessel morphogenesis 7.61E-04 9.63 GO:0050708 Regulation of protein secretion 7.81E-04 1.85 GO:0045780 Positive regulation of bone resorption 9.63E-04 5.14 GO:0022617 Extracellular matrix disassembly 9.91E-04 5.97 PBMCs S vs. AS (n = 66 + 30) GO:0051409 Response to nitrosative stress 1.44E-05 38.56 GO:0020027 Haemoglobin metabolic process 2.20E-05 47.46 GO:0045429 Positive regulation of nitric oxide biosynthetic process 2.95E-04 77.12 GO:0009636 Response to toxic substance 4.24E-04 23.14 GO:0050718 Positive regulation of interleukin-1 beta secretion 7.23E-04 51.42 GO:0046649 Lymphocyte activation 9.88E-04 3.3

LDL particle clearance was the most enriched category in comparisons between plaques from symptomatic (S) and asymptomatic (AS) patients. In peripheral blood mononuclear cells (PBMCs), nitric oxide biosynthesis, haemoglobin metabolism, lymphocyte activation and IL1B secretion were dominant in cells from S versus AS patients. Adjusted P- values are shown; expanded pathway analyses are shown in the Table S5. Enrichment was calculated as (b/n)/(B/N), where N is the total number of genes, B is the total number of genes associated with a speciﬁc GO term, n is the number of genes in the target set and b is the number of genes in the intersection. inﬂammatory response were enriched in S patients analysed in a separate microarray data set of receiving statin treatment (Table S9). n = 50 plaques (Fig. S1d).

For comparison, gene expression alterations in MALAT1, MIR155HG and TP53TG are involved in the regulation of gene atherosclerotic CP tissue compared with NA tissue expression in carotid atherosclerosis were considerable (up to 274-fold change for MMP12) with a larger portion of dysregulated Gene expression analyses in plaques and PBMCs probesets (25%), reflecting the profound biological from patients with carotid atherosclerosis also differences in these tissues. In plaques, pathways revealed dysregulation of many noncoding (nc) of cell proliferation, nitric oxide signalling, lipopro- transcripts (selected ncRNAs are presented in tein and apoptotic particle clearance, activation of Fig. 2a; for full list see Table S10). MIR155HG, immune cells, chemokine secretion, blood coagu- found in activated B and T cells and macrophages, lation and ECM disassembly were all enriched was significantly upregulated in comparisons (Fig. S1a–c) and, as in PBMCs, the Hb–haptoglobin between S and AS plaques and PBMCs. The long (Hp) complex was one of the most dominant GO noncoding (lnc) RNA MALAT1 associated with cell categories. In confirmation, generally similar proliferation, apoptosis [42], neovascularization results were obtained when pathways were [43] and prostate cancer [44] was substantially

ª 2015 The Association for the Publication of the Journal of Internal Medicine 7 Journal of Internal Medicine L. Perisic et al. Atherosclerosis transcriptome

(a)

(b)

Fig. 2 Regulation of gene expression in carotid atherosclerosis. (a) MIR155HG was strongly upregulated in plaques compared with normal arteries (mean difference Æ SD 1.627 Æ 0.2513; P < 0.0001), whilst MALAT1 was the most upregulated gene in plaques extracted more than 1 month after symptom compared to those removed within <2 weeks (mean difference 1.018 Æ 0.3451; P = 0.0045). TP53TG1 was downregulated in the same comparison (mean difference À0.3072 Æ 0.1324; P = 0.0238). (b) Examples of transcription factors regulating genes associated with immune response, response to cytokines, angiogenesis and muscle expansion. Binding motifs are shown on the right-hand side. downregulated in S compared with AS plaque patients. This approach conﬁrmed enrichment of tissue, and in comparison TIA+MS versus AF factors involved in regulation of the immune patients. By contrast, MALAT1 was upregulated response, response to cytokines, expression by in plaques extracted >1 month after the symptoms liver, angiogenesis and muscle expansion (Fig. 2b; compared to those removed within 2 weeks. for full list see Table S11). TP53TG lncRNA regulating p53 expression [45] was repressed in TIA+MS compared with AF A panel of 30 genes discriminates between plaques from S and AS patients. patients To investigate the regulation through transcription Prediction modelling was performed to determine factors and miRNAs, we performed reverse analysis molecular signatures for classiﬁcation of plaques by mapping for the corresponding binding motifs in from S versus AS patients. A panel of 30 genes was genes differentially expressed between S and AS found to be capable of predicting S and AS lesions

8 ª 2015 The Association for the Publication of the Journal of Internal Medicine Journal of Internal Medicine L. Perisic et al. Atherosclerosis transcriptome

with 78% accuracy, and 47% and 7% variability 3.46E-03) and response to hypoxia (FDR explained by the first two principal components 1.52E-02). Overall, 87% of these genes exhibited respectively (Fig. 3). Of interest, many of these strong positive expression cross-correlations in the genes, such as HCG11, DNAJC19, ZNF41, ZNF493 BiKE, many with Spearman r > 0.90 and P < 10E- and RPL9, were involved in transcriptional regula- 30 (e.g. MMP12 and MMP9 with CD36 and CD163 tion (Table S12). The panel also included genes markers of macrophages; see representative heat associated with lysosomes (PIK3C3 and CNO), plot in Fig. 4b and for full data see Table S14). cytokine signalling (ASB1), cholesterol distribution When meta-genes were connected into an extended (TSPO2) and the ECM (CRTAC1). protein–protein interaction network, key interaction nodes were centred around Hif1a, Actn2, Cxcr4, S100A9, S100A10, Casp4, Hspg1, Ubc, Functional gene networks in carotid atherosclerosis Pdlim7, Nos1, Hspa6 and Egfr (Fig. 4d). A meta-analysis was performed using a similar published data set [14] of CP samples from a LMOD1, SYNPO2, PPBP and PLIN2 are dysregulated in carotid smaller cohort of S and AS patients (12 S and 9 AS) atherosclerosis with the same microarray platform and technical requirements as the BiKE. This analysis showed Many genes were identified in this study that had about 20% overlap in significantly differentially previously not or only poorly been characterized in expressed probesets (Fig. 4a; for full list see atherosclerosis. Several strongly dysregulated Table S13), such as CCL18, MMP12, CTSZ, CD36, genes, representative of different cell types based CD163, IGFBP4, FABP5, GLUL, MYH11, SMTN, on publically available information (Table S2), were MYOCD and CNN1. From the full list of meta- chosen for validation by IHC and qPCR. LMOD1 genes, 85 of the most commonly reported and well and SYNPO2 were downregulated in CP versus NA characterized in atherosclerosis [23, 40] were microarrays (mean difference Æ SD À2.405 Æ selected for functional network construction 0.2162, P < 0.0001 and À4.256 Æ 0.2827, P < (Fig. 4c), and GSEA showed that they were mainly 0.0001, respectively) and in microarrays compar- involved in ECM organization (FDR 1.53E-09), ing S and AS patients (mean difference À1.431 Æ inflammation (FDR 3.71E-03), angiogenesis (FDR 0.5172, P < 0.0001 and À1.709 Æ 0.3711,

(a) (b)

Fig. 3 Prediction modelling for genes discriminating between plaques from symptomatic (S) and asymptomatic (AS) patients. Heat plot showing expression proﬁle for a panel of 30 genes (a; patients presented on the x-axis) found to discriminate between plaques from S and AS patients with high accuracy as shown by the principal components analysis (b).

ª 2015 The Association for the Publication of the Journal of Internal Medicine 9 Journal of Internal Medicine L. Perisic et al. Atherosclerosis transcriptome

(a) (b)

Fig. 4 Network analyses of plaques from symptomatic (S) and asymptomatic (AS) patients. (a) Meta-analyses with another similar publically available data set confirmed an overlap in significantly differentially expressed (DE) probesets comparing plaques from S and AS patients. (b) Many of the meta-genes showed strong positive expression correlations as depicted by the heat plot representation of the large correlation Supplementary Table (white, correlation not significant; green, Spearman r ≤ 0.3; red, Spearman r > 0.3). (c) In the functional coupling network generated from meta-genes, response to hypoxia, regulation of immune response, angiogenesis, chemokine production and extracellular matrix organization were the most significant functional categories. (d) When meta-genes were connected in a protein–protein interaction network, several nodes were distinctly populated with interactions.

P < 0.0001, respectively). PPBP and PLIN2 were respectively) and PPBP and PLIN2 were upregu- upregulated in CP versus NA tissues (mean differ- lated (mean difference 32.25 Æ 12.40 and ence 3.560 Æ 0.3138, P < 0.0001 and 8.701 Æ 1.156, respectively). The corresponding 2.640 Æ 0.2852, P < 0.0001, respectively) as well proteins were localized in plaques and normal as in S versus AS patients (mean difference internal carotid artery by IHC (Fig. 5c; for addi- 1.439 Æ 0.5253, P < 0.0001 and 1.564 Æ 0.6460, tional images and higher magniﬁcations see P < 0.0001, respectively; Fig. 5a). The same trends Fig. S2). Ppbp (CXCL7) co-localized with the plate- for these genes were replicated in the microarray let marker CD61 in areas of intraplaque haemor- validation data set (n = 50 patients; data not rhage (IPH). Plin2 (perilipin 2) was co-localized with shown) and qPCR analysis in a separate subcohort the macrophage receptor CD68 mainly in foam of patients (n = 77, Fig. 5b): LMOD1 and SYNPO2 cells. Lmod1 (leiomodin 1) and Synpo2 (synap- were signiﬁcantly downregulated in CP compared topodin 2) were expressed strictly by aSMA-positive with NA tissues (mean difference SMCs in the NA media, and no staining for these À0.5654 Æ 0.2078 and À0.9424 Æ 0.1668, two proteins was observed in CPs. Expression

10 ª 2015 The Association for the Publication of the Journal of Internal Medicine Journal of Internal Medicine L. Perisic et al. Atherosclerosis transcriptome

(a) (b) (c)

Fig. 5 Validation of novel genes from microarrays. LMOD1, SYNPO2, PPBP and PLIN2 were some of the most dysregulated genes in microarrays (a) as confirmed by quantitative PCR (qPCR) (b). Data are presented in graphs as fold change Æ SD. (c) Immunohistochemistry (IHC) showed that Lmod1 and Synpo2 localized to alpha smooth muscle actin (aSMA)-positive smooth muscle cells in normal carotid artery (left) whereas no staining was detected in the fibrous cap of plaques (right, arrowheads). Insets show the peripheral area of a plaque (with remains of the media) where Lmod1 and Synpo2 are not detected either. Ppbp was detected in CD61-positive platelets at sites of intraplaque haemorrhage, whereas Plin2 colocalized with CD68-positive foam cells in plaques (right, arrowheads). Ppbp and Plin2 were also expressed in normal artery adventitia and media, respectively (left, arrowheads). correlations for LMOD1 and SYNPO2 in plaque rapid tissue repair followed by prolonged inflam- arrays were positive solely with markers of differ- mation after plaque rupture were new observations entiated SMCs whereas negative correlations were that may be related to plaque stabilization, whilst, observed with cytokines, ECM degrading enzymes that is, processes associated with intraplaque and inflammation markers (Table S15). haemorrhage persisted in connection with vulnerable lesions. Coordination of gene expression and transcription-regulating elements was shown to be Discussion essential in controlling the subtle overall differ- Here, we present a comprehensive transcriptomic ences between S and AS disease. Importantly, analysis of human end-stage carotid atherosclero- many genes and candidate transcriptomic sis in relationship to clinical parameters of plaque biomarkers previously not characterized in unsta- instability. We confirm established features of ble atheroma were identified in our study. vulnerable plaques and patients, but also reveal novel pathways and gene signatures determining Modest alterations in gene expression levels were disease phenotype. Data suggesting enhanced found in plaques and PBMCs from S and AS calcification in patients on statin therapy and patients, probably because these two groups

ª 2015 The Association for the Publication of the Journal of Internal Medicine 11 Journal of Internal Medicine L. Perisic et al. Atherosclerosis transcriptome

represent entities of the same underlying disease a rapid initiation of tissue repair and SMC [14, 46]. Despite a reduction in sample size and proliferation followed by a prolonged immune statistical power, we also compared gene expres- response in the resolution phase towards restored sion between different subgroups of patient plaque stability [23]. phenotypes. It is interesting that more significant gene expression differences were observed when Intraplaque haemorrhage, involving release from comparisons were based on either the time red blood cells of Hb that is subsequently bound by between index symptoms and surgery or whether Hp to prevent oxidation [55], has previously been or not patients received statin therapy. In S recognized in association with plaque instability. In patients treated with statins, pathways previously line with this, we found significantly lower Hb associated with plaque stability such as suppres- levels in S patients in our cohort. Furthermore, the sion of angiogenesis and MMPs were enriched. Hb–Hp complex was one of the most enriched GO Notably, dysregulation of genes associated with the categories in plaques and Hb metabolism was HMG CoA reductase pathway, such as AMFR [47], enriched in PBMCs. A common genetic variant of indicated direct effects of statins on plaques. With Hp has recently been associated with cardiovascu- respect to the pleiotropic influence of statins, the lar disease risk in multiple independent cohorts, documented decrease in cholesterol and LDL levels both in plaques and in peripheral blood [56–58], also indicates effects related to lipid lowering in and Hp genotyping has been recommended for risk these patients [48]. Enrichment of pathways identification and treatment optimization [56]. associated with plaque calcification, such as regulation of calcium-mediated signalling and IL-1B secretion appeared as a major pathway in osteoblast differentiation, is a novel finding with PBMCs and in CP tissue. In support of this this methodological approach. This finding sug- observation, activation of IL-1B by cholesterol has gests that statins promote plaque stability as been reported in coronary atherosclerosis, and calcified plaques have been associated with AS neutralization of IL-1B by a human monoclonal carotid disease [49] and statins have also been antibody canakinumab has been shown to reduce shown to increase vascular calcification [50, 51]. inflammation; this antibody is being investigated Interestingly, in the entire group of plaques from S as an attractive therapy for atherosclerosis [59]. patients, bone resorption emerged as one of the enriched pathways. Taken together, these results In patients with increased stroke risk (TIA/MS), suggest that statins mediate plaque stabilization compared with those with AF [60], processes pre- through calcification and by counteracting resorp- viously associated with unstable lesions such as tion of calcified tissue. Mechanistically, these angiogenesis [61] were enriched and the most effects may be related to the lipid-lowering effects upregulated gene was TUBB1, a platelet-specific of the treatment as hyperlipidaemia has been tubulin variant linked to cardiovascular disease shown to enhance osteoclast activity [52]. risk [62]. PTAFR (platelet-activating factor receptor), GP6 (platelet-specific glycoprotein VI) and Because the risk of stroke in S patients with carotid genes from the TREML family involved in platelet stenosis decreases over time [4], plaques removed aggregation, inflammation and activation were also at different intervals after symptoms may represent upregulated. These findings support the notion time-dependent changes in plaque phenotype. that plaque rupture with platelet activation is more Comparison of gene expression in plaques from S prevalent in patients with TIA/MS than in those patients who underwent surgery >1 month versus with ocular symptoms [63]. <2 weeks after symptom indicated altered plaque biology following plaque rupture [23, 24, 53], with Here, we have shown that the profile of transcrip- recurrent upregulation of inflammatory processes tion factors in S compared with AS plaques reflects in the later phases of the healing response and the same main pathways as in the gene expression downregulation of pathways associated with tissue analysis, and prediction modelling further empha- regeneration. One of the most repressed genes was sized a central role for transcriptional regulation in MSTN, encoding a protein that regulates muscle determination of patient phenotype. Previously, we growth [54]. These findings are in contrast to those have reported that the predictive value of plaques of some earlier studies [24, 53], but support for risk profiling is superior to that of blood previous conclusions from this cohort, and suggest samples [16]. The common gene expression pat- that the initial period after plaque rupture involves terns in S and AS plaques, especially those related

12 ª 2015 The Association for the Publication of the Journal of Internal Medicine Journal of Internal Medicine L. Perisic et al. Atherosclerosis transcriptome

to chemokine signalling, apoptosis and from AS patients may vary with regard to morpho- calcification, were highlighted in meta-analyses logical features of plaque instability [46]. Unfortu- using another publicly available data set. Together, nately, histological plaque classification [70] was these results suggest that the fine determination of incompatible with the sampling protocol used here. atherosclerosis state requires coordinated In analyses of time-dependent gene expression regulation by multiple factors and plaque-specific changes in plaques, two separate time-points were networks. It is conceivable that end-stage defined, as too few patients underwent surgery in atherosclerosis consists of an array of processes the period between 2 weeks and 1 month after where minor alterations in plaque biology shift the symptoms. In addition, comparisons based on balance from a stable AS lesion to atherothrombo- statin medication, time from symptom to surgery sis and clinical symptoms. Similar observations and type of symptoms may be underpowered have recently been reported in cancers [44]. regarding limited group sizes, but nevertheless may provide novel insights into processes related Many studies of novel target genes in atherosclero- to these clinical categories. Of note, even the sis using data from the BiKE have already been smallest subgroups of patients compared in the reported [21, 23, 24, 29]. Here, the analysis revealed present study were larger than in most published additional genes that had not previously been human atherosclerosis microarray studies. associated with atherosclerosis, of which several were selected for replication in a nonoverlapping set Consensus is lacking regarding the selection of of patients. LMOD1 (Leiomodin 1) has been pre- appropriate control tissues, and arteries of various dicted to be an SMC-specific gene [64] and an SRF/ embryonic origins have been used for this purpose MYOCD-dependent transcript [65] and SYNPO2 as well as adjacent macroscopically intact parts of (Synaptopodin 2) was predicted to be associated lesions [40, 71]. In this study, control vessels with SMC cytoskeletal organization [66]. Because contained outer media that is not included in the LMOD1 and SYNPO2 were strongly downregulated endarterectomy intimal samples. From a technical in plaques and correlated with all markers of differ- viewpoint, it is noteworthy that the gene array entiated SMCs, they may represent novel markers of platform used here does not provide a complete SMC phenotype and should be explored in the analysis of ncRNAs. From a conceptual viewpoint, context of SMC function in atherosclerosis. The it should be noted that, even if the complexity of megakaryocyte PPBP gene product CXCL7 is a large-scale gene expression analyses is reduced by platelet chemokine involved in vessel repair after construction of functional networks, this method is vascular injury [67] and was found in the present limited in its ability to identify novel pathways as it study in plaques with IPH. Plin2 was detected in is based on existing literature and databases. We foam cells and cholesterol clefts within the necrotic obtained similar overall results in our analyses of core of plaques. Perilipin has previously been independent S versus AS array data sets of differ- described as an adipocyte protein that coats intra- ent sizes, suggesting that increasing the number of cellular lipid droplets. It was proposed to contribute patients does not necessarily improve the discovery to inhibition of lipolysis [68, 69] and thus could of new pathways. promote plaque vulnerability through increased lipid accumulation. Because of their association In conclusion, extensive analysis of the transcrip- with IPH and lipid content, PPBP and Plin2 may be tomic landscape in human atherosclerosis, in interesting new candidate transcriptomic biomark- combination with patient phenotyping, facilitated ers of unstable atheroma. novel discoveries related to overall plaque instability, time-dependent gene expression changes A limitation of this study is that the BiKE cohort and medication. Although expression analyses comprises only advanced lesions (American Heart from clinical samples cannot prove a causal Association grades IV and V), and therefore cannot relationship between increased gene expression provide information about disease progression. and disease state, our findings provide a powerful This restricts the generalization of the present platform for further functional and mechanistic results to high-risk populations with end-stage exploration of novel molecular signatures in disease. It is most likely that the phenotyping of carotid atherosclerosis, which may aid in the patients based on the presence or absence of development of clinical biomarkers and targeted cerebral symptoms was insufficient to completely molecular imaging for identification of vulnerable exclude overlap between the groups, as lesions patients and lesions.

ª 2015 The Association for the Publication of the Journal of Internal Medicine 13 Journal of Internal Medicine L. Perisic et al. Atherosclerosis transcriptome

8 Kunte H, Amberger N, Busch MA, Ruckert RI, Meiners S, Acknowledgements Harms L. Markers of instability in high-risk carotid plaques – The authors would like to thank Malin Kronquist, are reduced by statins. J Vasc Surg 2008; 47: 513 22. € 9 Lockhart DJ, Winzeler EA. Genomics, gene expression and Ingrid Tornberg, Anneli Olsson and Linda Haglund DNA arrays. Nature 2000; 405: 827–36. (Karolinska Institute, Stockholm, Sweden) for 10 Hiltunen MO, Tuomisto TT, Niemi M et al. Changes in gene skilful technical assistance and Siw Frebelius expression in atherosclerotic plaques analyzed using DNA (Karolinska Institute) for long-standing adminis- array. Atherosclerosis 2002; 165: 23–32. trative support. 11 Ijas P, Nuotio K, Saksi J et al. Microarray analysis reveals overexpression of CD163 and HO-1 in symptomatic carotid plaques. Arterioscler Thromb Vasc Biol 2007; 27: 154–60. Funding sources 12 Randi AM, Biguzzi E, Falciani F et al. Identification of differentially expressed genes in coronary atherosclerotic This study was supported by the Swedish Heart plaques from patients with stable or unstable angina by and Lung Foundation, the Swedish Research cDNA array analysis. J Thromb Haemost 2003; 1: 829–35. Council (K2009-65X-2233-01-3, K2013-65X- 13 Woodside KJ, Hernandez A, Smith FW et al. Differential gene 06816-30-4 and 349-2007-8703), Uppdrag expression in primary and recurrent carotid stenosis. – Besegra Stroke (P581/2011-123), the Strategic Biochem Biophys Res Commun 2003; 302: 509 14. 14 Saksi J, Ijas P, Nuotio K et al. Gene expression differences Cardiovascular Programs of Karolinska Institutet between stroke-associated and asymptomatic carotid pla- and Stockholm County Council, the Stockholm ques. J Mol Med 2011; 89: 1015–26. County Council (ALF-2011-0260 and ALF-2011- 15 Sluimer JC, Kisters N, Cleutjens KB et al. Dead or alive: gene 0279), the Foundation for Strategic Research and expression profiles of advanced atherosclerotic plaques from the European Commission (CarTarDis, AtheroR- autopsy and surgery. Physiol Genomics 2007; 30: 335–41. emo, VIA and AtheroFlux projects). 16 Folkersen L, Persson J, Ekstrand J et al. Prediction of ischemic events on the basis of transcriptomic and genomic profiling in patients undergoing carotid endarterectomy. Mol Conflict of interest statement Med 2012; 18: 669–75. 17 Miller CL, Haas U, Diaz R et al. Coronary heart disease- No conflicts of interest to declare. associated variation in TCF21 disrupts a miR-224 binding site and miRNA-mediated regulation. PLoS Genet 2014; 10: e1004263. References 18 Kojima Y, Downing K, Kundu R et al. Cyclin-dependent kinase inhibitor 2B regulates efferocytosis and atherosclero- 1 Libby P, Ridker PM, Hansson GK. Progress and challenges in sis. J Clin Investig 2014; 124: 1083–97. translating the biology of atherosclerosis. Nature 2011; 473: 19 Smedbakken LM, Halvorsen B, Daissormont I et al. Increased 317–25. levels of the homeostatic chemokine CXCL13 in human 2 Hansson GK. Inflammation, atherosclerosis, and coronary atherosclerosis - Potential role in plaque stabilization. artery disease. N Engl J Med 2005; 352: 1685–95. Atherosclerosis 2012; 224: 266–73. 3 Rothwell PM, Eliasziw M, Gutnikov SA, Warlow CP, Barnett 20 Salagianni M, Galani IE, Lundberg AM et al. Toll-like receptor HJ; Carotid Endarterectomy Trialists C. Endarterectomy for 7 protects from atherosclerosis by constraining “inflamma- symptomatic carotid stenosis in relation to clinical subgroups tory” macrophage activation. Circulation 2012; 126: 952–62. and timing of surgery. Lancet 2004; 363: 915–24. 21 Perisic L, Hedin E, Razuvaev A et al. Profiling of atheroscle- 4 Giles MF, Rothwell PM. Risk of stroke early after transient rotic lesions by gene and tissue microarrays reveals PCSK6 as ischaemic attack: a systematic review and meta-analysis. a novel protease in unstable carotid atherosclerosis. Arte- Lancet Neurol 2007; 6: 1063–72. rioscler Thromb Vasc Biol 2013; 33: 2432–43. 5 Rothwell PM. Prediction and prevention of stroke in patients 22 Wang J, Razuvaev A, Folkersen L et al. The expression of IGFs with symptomatic carotid stenosis: the high-risk period and and IGF binding proteins in human carotid atherosclerosis, the high-risk patient. Eur J Vasc Endovasc Surg 2008; 35: and the possible role of IGF binding protein-1 in the regula- 255–63. tion of smooth muscle cell proliferation. Atherosclerosis 2012; 6 Redgrave JN, Lovett JK, Gallagher PJ, Rothwell PM. 220: 102–9. Histological assessment of 526 symptomatic carotid plaques 23 Razuvaev A, Ekstrand J, Folkersen L et al. Correlations in relation to the nature and timing of ischemic symp- between clinical variables and gene-expression profiles in toms: the Oxford plaque study. Circulation 2006; 113: carotid plaque instability. Eur J Vasc Endovasc Surg 2011; 2320–8. 42: 722–30. 7 Crisby M, Nordin-Fredriksson G, Shah PK, Yano J, Zhu J, 24 Agardh HE, Folkersen L, Ekstrand J et al. Expression of fatty Nilsson J. Pravastatin treatment increases collagen content acid-binding protein 4/aP2 is correlated with plaque and decreases lipid content, inflammation, instability in carotid atherosclerosis. J Intern Med 2011; metalloproteinases, and cell death in human carotid 269: 200–10. plaques: implications for plaque stabilization. Circulation 25 Gertow K, Nobili E, Folkersen L et al. 12- and 15-lipoxy- 2001; 103: 926–33. genases in human carotid atherosclerotic lesions:

14 ª 2015 The Association for the Publication of the Journal of Internal Medicine Journal of Internal Medicine L. Perisic et al. Atherosclerosis transcriptome

associations with cerebrovascular symptoms. Atherosclerosis 44 Ren S, Peng Z, Mao JH et al. RNA-seq analysis of prostate 2011; 215: 411–6. cancer in the Chinese population identifies recurrent gene 26 Gistera A, Robertson AK, Andersson J et al. Transforming fusions, cancer-associated long noncoding RNAs and aber- growth factor-beta signaling in T cells promotes stabilization rant alternative splicings. Cell Res 2012; 22: 806–21. of atherosclerotic plaques through an interleukin-17-depen- 45 Takei Y, Ishikawa S, Tokino T, Muto T, Nakamura Y. Isolation dent pathway. Sci Transl Med 2013; 5: 196ra00. of a novel TP53 target gene from a colon cancer cell line 27 Gabrielsen A, Qiu H, Back M et al. Thromboxane synthase carrying a highly regulated wild-type TP53 expression sys- expression and thromboxane A2 production in the tem. Genes Chromosom Cancer 1998; 23: 1–9. atherosclerotic lesion. J Mol Med 2010; 88: 795–806. 46 Salem MK, Vijaynagar B, Sayers RD et al. Histologically 28 Olofsson PS, Soderstrom LA, Jern C et al. Genetic variants of unstable asymptomatic carotid plaques have altered expres- TNFSF4 and risk for carotid artery disease and stroke. J Mol sion of genes involved in chemokine signalling leading to Med 2009; 87: 337–46. localised plaque inflammation and rupture. Eur J Vasc 29 Osterholm C, Folkersen L, Lengquist M et al. Increased Endovasc Surg 2013; 45: 121–7. expression of heparanase in symptomatic carotid atheroscle- 47 Song BL, Sever N, DeBose-Boyd RA. Gp78, a membrane- rosis. Atherosclerosis 2013; 226: 67–73. anchored ubiquitin ligase, associates with Insig-1 and cou- 30 Ovchinnikova OA, Folkersen L, Persson J et al. The collagen ples sterol-regulated ubiquitination to degradation of HMG cross-linking enzyme lysyl oxidase is associated with the CoA reductase. Mol Cell 2005; 19: 829–40. healing of human atherosclerotic lesions. J Intern Med 2014; 48 Farmer JA. Pleiotropic effects of statins. Curr Atheroscler Rep 276: 525–36. 2000; 2: 208–17. 31 Persson J, Folkersen L, Ekstrand J et al. High plasma 49 Kwee RM. Systematic review on the association between adiponectin concentration is associated with all-cause mor- calcification in carotid plaques and clinical ischemic symp- tality in patients with carotid atherosclerosis. Atherosclerosis toms. J Vasc Surg 2010; 51: 1015–25. 2012; 225: 491–6. 50 Saremi A, Bahn G, Reaven PD, Investigators V. Progression of 32 Rothwell PM, Pendlebury ST, Wardlaw J, Warlow CP. Critical vascular calcification is increased with statin use in the appraisal of the design and reporting of studies of imaging and Veterans Affairs Diabetes Trial (VADT). Diabetes Care 2012; measurement of carotid stenosis. Stroke 2000; 31: 1444–50. 35: 2390–2. 33 Halliday A, Harrison M, Hayter E et al. 10-year stroke 51 Grasser WA, Baumann AP, Petras SF et al. Regulation of prevention after successful carotid endarterectomy for osteoclast differentiation by statins. J Musculoskelet Neuronal asymptomatic stenosis (ACST-1): a multicentre randomised Interact 2003; 3: 53–62. trial. Lancet 2010; 376: 1074–84. 52 Tintut Y, Morony S, Demer LL. Hyperlipidemia promotes 34 Lesack K, Naugler C. An open-source software program for osteoclastic potential of bone marrow cells ex vivo. Arte- performing Bonferroni and related corrections for multiple rioscler Thromb Vasc Biol 2004; 24: e6–10. comparisons. J Pathol Inform 2011; 2: 52. 53 Peeters W, Hellings WE, de Kleijn DP et al. Carotid atheroscle- 35 Glaab E, Garibaldi JM, Krasnogor N. ArrayMining: a modular rotic plaques stabilize after stroke: insights into the natural web-application for microarray analysis combining ensemble process of atherosclerotic plaque stabilization. Arterioscler and consensus methods with cross-study normalization. Thromb Vasc Biol 2009; 29: 128–33. BMC Bioinformatics 2009; 10: 358. 54 Kostek MA, Angelopoulos TJ, Clarkson PM et al. Myostatin 36 Xia J, Fjell CD, Mayer ML, Pena OM, Wishart DS, Hancock and follistatin polymorphisms interact with muscle pheno- RE. INMEX–a web-based tool for integrative meta-analysis of types and ethnicity. Med Sci Sports Exerc 2009; 41: 1063–71. expression data. Nucleic Acids Res 2013; 41: W63–70. 55 Levy AP, Moreno PR. Intraplaque hemorrhage. Curr Mol Med 37 Wu J, Vallenius T, Ovaska K, Westermarck J, Makela TP, 2006; 6: 479–88. Hautaniemi S. Integrated network analysis platform for 56 Ijas P, Saksi J, Soinne L et al. Haptoglobin 2 allele associates protein-protein interactions. Nat Methods 2009; 6: 75–7. with unstable carotid plaque and major cardiovascular 38 Cowley MJ, Pinese M, Kassahn KS et al. PINA v2.0: mining events. Atherosclerosis 2013; 230: 228–34. interactome modules. Nucleic Acids Res 2012; 40: D862–5. 57 Degraba TJ, Hoehn GT, Nyquist PA et al. Biomarker discovery 39 Le Cao KA, Boitard S, Besse P. Sparse PLS discriminant in serum from patients with carotid atherosclerosis. Cere- analysis: biologically relevant feature selection and graphical brovasc Dis Extra 2011; 1: 115–29. displays for multiclass problems. BMC Bioinformatics 2011; 58 Wingrove JA, Daniels SE, Sehnert AJ et al. Correlation of 12: 253. peripheral-blood gene expression with the extent of coronary 40 Papaspyridonos M, Smith A, Burnand KG et al. Novel candi- artery stenosis. Circ Cardiovasc Genet 2008; 1: 31–8. date genes in unstable areas of human atherosclerotic 59 Ridker PM, Howard CP, Walter V et al. Effects of interleukin- plaques. Arterioscler Thromb Vasc Biol 2006; 26: 1837–44. 1beta inhibition with canakinumab on hemoglobin A1c, 41 Rothwell PM. ACST: which subgroups will benefit most from lipids, C-reactive protein, interleukin-6, and fibrinogen: a carotid endarterectomy? Lancet 2004; 364: 1122–3; author phase IIb randomized, placebo-controlled trial. Circulation reply 5-6. 2012; 126: 2739–48. 42 Tripathi V, Ellis JD, Shen Z et al. The nuclear-retained 60 Benavente O, Eliasziw M, Streifler JY, Fox AJ, Barnett HJ, noncoding RNA MALAT1 regulates alternative splicing by Meldrum H; North American Symptomatic Carotid modulating SR splicing factor phosphorylation. Mol Cell Endarterectomy Trial C. Prognosis after transient monocular 2010; 39: 925–38. blindness associated with carotid-artery stenosis. N Engl J 43 Michalik KM, You X, Manavski Y et al. Long noncoding RNA Med 2001; 345: 1084–90. MALAT1 regulates endothelial cell function and vessel 61 McCarthy MJ, Loftus IM, Thompson MM et al. Angiogenesis growth. Circ Res 2014; 114: 1389–97. and the atherosclerotic carotid plaque: an association

ª 2015 The Association for the Publication of the Journal of Internal Medicine 15 Journal of Internal Medicine L. Perisic et al. Atherosclerosis transcriptome

between symptomatology and plaque morphology. J Vasc Table S1. Demographics of the patient cohorts Surg 1999; 30: 261–8. (*P < 0.05, ns-nonsignificant). 62 Freson K, De Vos R, Wittevrongel C et al. The TUBB1 Q43P functional polymorphism reduces the risk of cardiovascular disease in men by modulating platelet function and structure. Table S2 Summary of antibodies used in valdation Blood 2005; 106: 2356–62. of microarray results. 63 Howard DP, van Lammeren GW, Redgrave JN et al. Histological features of carotid plaque in patients with Table S3. Most highly significantly dysregulated ocular ischemia versus cerebral events. Stroke 2013; 44: probes in microarray comparison plaque tissue – 734 9. symptomatic vs. asymptomatic. 64 Nelander S, Mostad P, Lindahl P. Prediction of cell type- specific gene modules: identification and initial characterization of a core set of smooth muscle-specific genes. Genome Table S4. Most highly significantly upregulated Res 2003; 13: 1838–54. probes in microarray comparison PBMCs symp- 65 Nanda V, Miano JM. Leiomodin 1, a new serum response tomatic vs. asymptomatic. factor-dependent target gene expressed preferentially in differentiated smooth muscle cells. J Biol Chem 2012; 287: Table S5. GO terms enriched in comparisons – 2459 67. between symptomatic and asymptomatic patients. 66 Schroeter MM, Beall B, Heid HW, Chalovich JM. In vitro characterization of native mammalian smooth-muscle protein synaptopodin 2. Biosci Rep 2008; 28: 195–203. Table S6. Most significantly upregulated probes in 67 Gleissner CA, von Hundelshausen P, Ley K. Platelet chemoki- microarray comparison symptomatic plaque nes in vascular disease. Arterioscler Thromb Vasc Biol 2008; TIA+MS vs AF. 28: 1920–7. 68 Martinez-Botas J, Anderson JB, Tessier D et al. Absence of Table S7. Most significantly upregulated probes in perilipin results in leanness and reverses obesity in Lepr(db/ > db) mice. Nat Genet 2000; 26: 474–9. microarray comparison symptomatic plaque 1 < 69 Tansey JT, Sztalryd C, Gruia-Gray J et al. Perilipin ablation month vs 2 weeks. results in a lean mouse with aberrant adipocyte lipolysis, enhanced leptin production, and resistance to diet-induced Table S8. Most highly significantly upregulated obesity. Proc Natl Acad Sci USA 2001; 98: 6494–9. probes in microarray comparison symptomatic 70 Lovett JK, Redgrave JN, Rothwell PM. A critical appraisal of plaque statins yes vs no. the performance, reporting, and interpretation of studies comparing carotid plaque imaging with histology. Stroke 2005; 36: 1091–7. Table S9. GO terms enriched in comparisons sorted 71 Turpeinen H, Raitoharju E, Oksanen A et al. Proprotein according to clinical parameters of instability. convertases in human atherosclerotic plaques: the overexpression of FURIN and its substrate cytokines BAFF and Table S10. ncRNAs APRIL. Atherosclerosis 2011; 219: 799–806. differentially regulated in different atherosclerosis profiles. Correspondence: Ljubica Perisic, PhD, Department of Molecular Medicine and Surgery, Division of Vascular Surgery, L8:03, Karolinska Institute, SE-171 76 Stockholm, Sweden. Table S11. Transcription factors and miRs regu- (fax: +468339309; e-mail: [email protected]). lating the genes differentially expressed in symp- and tomatic vs. asymptomatic plaques. Prof Ulf Hedin, MD, PhD, Department of Vascular Surgery, Karolinska University Hospital A2:01 SE-171 76 Stockholm, Table S12. Panel of genes obtained by predictive Sweden. (fax: +468339309; e-mail: [email protected]). modelling in S vs. As patient’s plaques.

Supporting Information Table S13. Genes with signiﬁcant combined p- values from meta-analysis. Additional Supporting Information may be found in the online version of this article: Table S14. Expression correlation matrix for genes from meta-analyses (Spearman r values given, Figure S1. Pathway analysis of genes differentially p < 0.05 after correction for multiple testing). expressed in plaques vs. normal arteries. Table S15. Expression correlations for LMOD1 Figure S2. Additional photomicrographs showing and SYNPO2 with markers from plaque arrays. cellular localization of novel genes from microarrays.

16 ª 2015 The Association for the Publication of the Journal of Internal Medicine Journal of Internal Medicine