Basic Research in Cardiology

, Volume 106, Issue 1, pp 13–23

MicroRNA signatures in total peripheral blood as novel biomarkers for acute myocardial infarction

Authors

  • Benjamin Meder
    • Department of Internal Medicine IIIUniversity of Heidelberg
  • Andreas Keller
    • Biomarker Discovery Center Heidelberg
  • Britta Vogel
    • Department of Internal Medicine IIIUniversity of Heidelberg
  • Jan Haas
    • Department of Internal Medicine IIIUniversity of Heidelberg
  • Farbod Sedaghat-Hamedani
    • Department of Internal Medicine IIIUniversity of Heidelberg
  • Elham Kayvanpour
    • Department of Internal Medicine IIIUniversity of Heidelberg
  • Steffen Just
    • Department of Internal Medicine IIIUniversity of Heidelberg
  • Anne Borries
    • Biomarker Discovery Center Heidelberg
  • Jessica Rudloff
    • Department of Internal Medicine IIIUniversity of Heidelberg
  • Petra Leidinger
    • Department of Human Genetics, Medical SchoolSaarland University
  • Eckart Meese
    • Department of Human Genetics, Medical SchoolSaarland University
  • Hugo A. Katus
    • Department of Internal Medicine IIIUniversity of Heidelberg
    • Department of Internal Medicine IIIUniversity of Heidelberg
    • Department of Medicine IIUniversity of Ulm
Original Contribution

DOI: 10.1007/s00395-010-0123-2

Cite this article as:
Meder, B., Keller, A., Vogel, B. et al. Basic Res Cardiol (2011) 106: 13. doi:10.1007/s00395-010-0123-2

Abstract

MicroRNAs (miRNAs) are important regulators of adaptive and maladaptive responses in cardiovascular diseases and hence are considered to be potential therapeutical targets. However, their role as novel biomarkers for the diagnosis of cardiovascular diseases still needs to be systematically evaluated. We assessed here for the first time whole-genome miRNA expression in peripheral total blood samples of patients with acute myocardial infarction (AMI). We identified 121 miRNAs, which are significantly dysregulated in AMI patients in comparison to healthy controls. Among these, miR-1291 and miR-663b show the highest sensitivity and specificity for the discrimination of cases from controls. Using a novel self-learning pattern recognition algorithm, we identified a unique signature of 20 miRNAs that predicts AMI with even higher power (specificity 96%, sensitivity 90%, and accuracy 93%). In addition, we show that miR-30c and miR-145 levels correlate with infarct sizes estimated by Troponin T release. The here presented study shows that single miRNAs and especially miRNA signatures derived from peripheral blood, could be valuable novel biomarkers for cardiovascular diseases.

Keywords

Cardiovascular diseasesDiagnosisMyocardial infarctionBiomarkerMicroRNA

Introduction

MicroRNAs (miRNAs) represent a group of regulatory elements that enable cells to fine-tune complex gene expression cascades in a wide range of biological processes, such as proliferation, differentiation, apoptosis, and stress-response [9, 12, 26, 42, 43]. In the cardiovascular system, miRNAs are not only important for heart and vascular development but also play an essential role in cardiac pathophysiology, such as hypertrophy, arrhythmia, and ischemia [7, 11]. However, their potential role as biomarkers for the diagnosis of cardiovascular diseases has not been systematically evaluated yet.

Today, biomarkers play a key role in early diagnosis, risk stratification, and therapeutic management of cardiac diseases such as acute myocardial infarction (AMI) and heart failure [22, 25, 28, 32]. Established biomarkers such as the cardiac troponins and b-type natriuretic peptides were mainly discovered by candidate approach [22, 31]. In contrast, the recent development of high-throughput molecular technologies that allow with a reasonable effort the analysis of whole transcriptomes, proteomes, and metabolomes of individuals at risk may lead to the discovery of novel biomarkers in an unbiased approach [15, 16].

In the present study we aimed to identify a miRNA signature for the diagnosis of AMI, which is not solely based on the release of miRNAs from necrotic myocardium but also on active processes involved in the pathogenesis of AMI, like inflammation, plaque rupture, and vascular injury. Therefore, we assessed for the first time the expression of miRNAs on a whole-genome level in total peripheral blood of patients with AMI and identified a unique miRNA signature that predicts myocardial infarction with high specificity and sensitivity, implicating that both single miRNAs and complex miRNA signatures could be used as novel biomarkers for the diagnosis of cardiovascular diseases.

Materials and methods

Study population

According to a priori power analyses (power of 0.95), we included 20 patients with acute ST elevation myocardial infarction (STEMI) and 20 controls in the present study. Acute myocardial infarction (AMI) was diagnosed according to the ESC/AHA redefined guidelines. Vessel occlusion by an underlying thrombotic event as cause of AMI was confirmed in all patients by early coronary angiography. miRNA profiles from 20 control subjects without acute coronary syndrome, who underwent routine coronary angiography, served as control (see Table 1 for detailed clinical characteristics). MiRNA profiles from 20 healthy volunteers recruited during the “Biomarker Discovery Center” pilot study served as an internal control. The analysis of blood from patients and controls has been approved by local ethics committees, and participants have given written informed consent. Serial Troponin T levels were assessed using the Elecsys highly sensitive Troponin T assay (Roche, Germany) [17].
Table 1

Patient characteristics

Characteristics

Patients with AMI (n = 20)

Patients without AMI (n = 20)

p value

Age (years)

59.3 ± 14

63.3 ± 14.8

0.38

Male/female (n/n)

16/4

14/6

0.72

Current smoking, n (%)

6 (30)

4 (20)

0.72

DM, n (%)

3 (15)

4 (20)

1.00

Hypertension, n (%)

11 (55)

13 (65)

0.75

Hyperlipidaemia, n (%)

9 (45)

6 (30)

0.52

SBP (mmHg)

134 ± 27

128 ± 13

0.45

DBP (mmHg)

80 ± 15

73 ± 9

0.08

TG (mg/dL)

170.4 ± 49.9

139.4 ± 77.5

0.44

HDL (mg/dL)

38.1 ± 12.6

44.4 ± 28.3

0.65

LDL (mg/dL)

103.5 ± 38.9

104.2 ± 34.7

0.97

WBC/nL

10.28 ± 3.6

9.39 ± 3.2

0.47

Creatinine (mg/dL)

1.12 ± 0.68

0.98 ± 0.24

0.43

Urea (mg/dL)

39.2 ± 21.4

36.4 ± 10.1

0.62

DM diabetes mellitus, SBP systolic blood pressure, DBP diastolic blood pressure, TG triglycerides, HDL high-density lipoprotein, LDL low-density lipoprotein, WBC white blood cell count

MiRNA expression profiling from whole peripheral blood samples

Five milliliters of blood was collected from study subjects in PAXgene Blood RNA tubes (BD, USA) and stored at 4°C until total RNA was extracted from blood cells using the miRNeasy Mini Kit (Qiagen, Germany) [27]. MiRNA expression profiling was performed by personnel blinded to patient characteristics. Samples were analyzed with the Geniom Real-time Analyzer (febit, Germany) using the Geniom Biochip miRNA homo sapiens. Each array contains seven replicates of 866 miRNAs and miRNA star sequences as annotated in the Sanger miRBase 12.0 [18]. Sample labeling with biotin was carried out by microfluidic-based enzymatic on-chip labeling of miRNAs (MPEA) as described before [39].

Following hybridization for 16 h at 42°C the biochip was washed, and signals were measured. The resulting images were evaluated using the Geniom Wizard Software (febit, Germany). For each array, the median signal intensity of all features was extracted from the raw data file such that for each miRNA seven intensity values were calculated corresponding to each replicate on the array. Following background correction, the seven replicate intensity values of each miRNA were summarized by their median value. To normalize the data across different arrays, quantile normalization was applied, and all further analyses were carried out using the normalized and background subtracted intensity values. To this end, we removed those miRNAs having an overall median signal intensity of less than 10. The final data set contained expression levels of 697 miRNAs.

To verify the accuracy of the microarray-based miRNA measurements, expression levels of miR-145, -30c, -455-3p, -10b*, -216a, -1291, and -223 were assessed using quantitative real-time PCR (measured in triplets) according to manufacture’s instructions (ABI, USA). The small nuclear RNA RNU6B-2 served as reference.

Statistical analysis

After ensuring approximate normal distribution of the measured intensity values using Shapiro–Wilk test, we carried out parametric t test and limma test (unpaired, two-tailed) for each miRNA separately to detect miRNAs with differential expression between the patient and control groups. The resulting p values were adjusted for multiple testing by Benjamini–Hochberg equation [5]. Correlation coefficients between Troponin T and miRNA levels were calculated using Pearson’s product–moment coefficient. Outlier testing was performed using the tests of Grubbs and Dixon. Clustering of miRNAs has been carried out using complete linkage hierarchical clustering. As distance measure, the Euclidian distance has been applied, and miRNAs and samples have been clustered independently of each other. To assess significance of the clustering, Fishers exact test on the contingency table has been applied.

In addition to the single biomarker analysis, classification of samples using miRNA patterns was carried out using support vector machines (SVM) as implemented in the R e1071 package (http://www.r-project.org) [35]. In particular, different kernel SVMs (linear, polynomial, sigmoid, and radial basis function) were evaluated, where the cost parameter was sampled from 0.01 to 10 in decimal powers. The measured miRNA profiles were classified using 20 repetitions of standard tenfold cross-validation. The classification has been carried out using equal class sizes. To detect the most suitable set of miRNAs that achieves the best discriminatory performance, a feature extraction (subset selection) method relying on t test p values was used. In detail, the following filter technique was applied: The s miRNAs with lowest p values in the t test were computed on the training set in each fold of the cross-validation, where s was sampled from 1 to 697 in regular intervals. The respective subset was used to train the SVM and to carry out the prediction of the test samples. The mean accuracy, specificity, and sensitivity were calculated for each subset size. We additionally computed the area under the curve value (AUC), which is more discriminative than calculating the accuracy alone [4]. To check for overtraining we applied permutation tests. Here, we sampled the class labels randomly and carried out classifications using the permuted class labels. All statistical analyses were performed using R.

Results

Acute ST elevation myocardial infarction leads to dysregulation of specific miRNAs

In the present pilot study, we analyzed the expression of 866 miRNAs and miRNA star sequences in blood cells of 20 patients with acute myocardial infarction (AMI) and 20 control subjects. Mean age in AMI patients was 59.25 ± 14 years and in controls 63.30 ± 14.8 years (p = 0.38). In the AMI group 4/20 and in the controls 6/20 patients were females (p = 0.72). According to ECG criteria, 15 patients showed signs of inferior/posterior wall infarction and 5 of anterior wall infarction. The blood for miRNA and Troponin T measurements was drawn in the mean 3.0 ± 2.3 h after the reported onset of symptoms. The mean hsTnT on admission was 345.8 ± 562.8 pg/ml. Ten patients (50%) with AMI had Troponin T levels of <50 pg/ml, and three of them were below 14 pg/ml (15%). White blood cell counts showed no significant differences between both groups.

To test for the overall correlation of miRNA expression between cases and controls, we computed the median expression of all miRNAs in both groups on a logarithmic scale. As shown in the scatter plot in Fig. 1a, both groups had high correlation of 0.891 and data variance as low as 0.002. To cross-validate the whole-genome miRNA microarray, we exemplarily measured the expression levels of up- and downregulated miRNAs by real-time PCR. As shown in Supplemental figure 1, we find a high reproducibility between both methods. Ensuring an approximate normal distribution using Shapiro–Wilk test, we performed two-tailed unpaired t tests and limma test for each miRNA. The respective p values were adjusted for multiple testing by the Benjamini–Hochberg approach [19]. In total, we detected 121 miRNAs to be significantly dysregulated in blood cells from AMI patients in comparison to controls at an adjusted significance level of p < 0.05. Histogram plots of the logarithm of fold ratios, the raw t test and limma p values, and the adjusted p values are presented in Fig. 1b. Notably, from 121 dysregulated miRNAs only 45 (37%) were upregulated, while 76 (63%) were downregulated in AMI patients. The expression values of the 20 most significantly down- and upregulated miRNAs are shown in Fig. 2a, b. Interestingly, miR-21, which is known to be upregulated in cardiomyocytes after myocardial infarction and heart failure, was significantly upregulated in blood of AMI patients in comparison to healthy volunteers; however, it was not significantly dysregulated in comparison to the control group (Supplemental figure 2).
https://static-content.springer.com/image/art%3A10.1007%2Fs00395-010-0123-2/MediaObjects/395_2010_123_Fig1_HTML.gif
Fig. 1

MiRNA expression profiling in peripheral blood cells of AMI patients and controls. a Matrix plot showing high linear correlation of the mean expression ratios of miRNAs in AMI patients (n = 20) versus controls (n = 20). b Histograms of the logarithm of fold changes, AUC, raw, and adjusted p values for all screened miRNAs. The logqmedian diagram shows the logarithms of miRNA expression changes in AMI patients in comparison to controls. The expression changes are distributed between −4 and +4. The AUC for all expression values is given; the vertical blue line denotes the threshold of 0.1 and 0.9. The middle and bottom parts of the figure represent histograms of adjusted p values for the limma test and t test. The vertical blue line denotes the significance threshold of 0.05

https://static-content.springer.com/image/art%3A10.1007%2Fs00395-010-0123-2/MediaObjects/395_2010_123_Fig2_HTML.gif
Fig. 2

Specific miRNAs are dysregulated in acute myocardial infarction. Bar graphs showing mean expression values of a the 20 most significantly downregulated and b upregulated miRNAs in patients with acute myocardial infarction (n = 20; red bars) in comparison to controls (n = 20; blue bars)

To now evaluate the predictive value of dysregulated miRNA for AMI, we calculated receiver operator characteristic curves (ROC) for each of the best miRNAs together with the area under the curve value (AUC). For the most predictive miRNAs, miR-1291, and miR-663b, we obtained AUC values of up to 0.94 (Fig. 3a, b). Using miR-1291, we classified 17 of the 20 disease samples and 17 of 20 controls correctly resulting in a specificity of 85% and a sensitivity of 85% with test accuracy of 85%. miR-663b shows even better predictive values, allowing us to classify 19 of 20 cases and 18 of 20 controls correctly, resulting in specificity, sensitivity, and accuracy of 95, 90, and 92.5%, respectively.
https://static-content.springer.com/image/art%3A10.1007%2Fs00395-010-0123-2/MediaObjects/395_2010_123_Fig3_HTML.gif
Fig. 3

Single miRNAs predict myocardial infarction. Receiver operating characteristic (ROC) analysis of miRNA-1291 and miRNA-663b to predict AMI in the study population. a MiRNA-1291 is able to predict the presence of AMI with a specificity of 85% and a sensitivity of 85%, while miRNA-663b shows a specificity of 95% and a sensitivity of 90%. TP true positives, FP false positives

MiR-145 and MiR-30c levels significantly correlate with Troponin T levels

With the ability of single miRNAs to predict AMI, we next wondered if the magnitude of miRNA dysregulation correlates with infarct sizes estimated by Troponin T release (hsTnT). To this end, we computed for all dysregulated miRNAs the correlation between miRNA expression and hsTnT at days 3 and 4. Furthermore, we calculated correlations to peak hsTnT levels. We identified two upregulated miRNAs in AMI patients that show high correlation to Troponin T levels, with correlation coefficients up to 0.71. Figure 4 represents the two miRNAs with highest linear correlation to Troponin T, miR-145 (A, C), and miR-30c (B, C) (see also Table 2).
https://static-content.springer.com/image/art%3A10.1007%2Fs00395-010-0123-2/MediaObjects/395_2010_123_Fig4_HTML.gif
Fig. 4

MiR-145 and miR-30c levels correlate with Troponin T release. Matrix-plots showing correlation of miR-145 (a) and miR-30c (b) with peak Troponin T serum levels (left column) and Troponin T levels at day 3 (right column). c Absolute expression values (±SEM) of miR-145 and miR-30c in AMI patients and controls. hsTnT highly sensitive Troponin T

Table 2

Correlation of miRNA levels and Troponin T release

 

Day 3

Peak

miRNA

Correlation coefficient to hsTnT

Confidence interval

p value

Correlation coefficient to hsTnT

Confidence interval

p value

miR-30c

0.667

0.31–0.86

0.0018

0.713

0.39–0.88

0.0004

miR-145

0.665

0.30–0.86

0.0019

0.710

0.39–0.88

0.0005

hsTnT highly sensitive Troponin T

MiRNA signatures enhance diagnostic discrimination of AMI patients from controls

Although single miRNAs may predict the presence of AMI with good sensitivity and specificity, we tested whether complex miRNA signatures derived from unsupervised hierarchical clustering and supervised classification may improve the sensitivity and specificity. First, to test for common patterns in AMI patients and controls, we applied hierarchical clustering using the expression levels of the 20 miRNAs with highest data variance and applied the Euclidian distance measure. As shown in Fig. 5, the algorithm clusters all patients and controls separately, with only three outliers. The significance of the clustering has been computed by Fishers test with p value of <0.001. Based on the ability to differentially cluster miRNA patterns in controls and AMI subjects and in order to improve the predictive power of a miRNA-based biomarker, we next combined the information content of multiple miRNAs by using statistical learning techniques. In detail, we applied SVM with different kernels as described in “Materials and methods”. The best results were obtained using radial basis function SVM and a subset of 20 miRNAs. The cross-validation procedure has been carried out 20 times to gain additional statistical significance. As shown in Fig. 6a, the values for sensitivity, specificity, and accuracy increased with the number of selected miRNAs in the diagnostic signature. 20 miRNAs allowed discrimination of AMI patients and controls with an accuracy of 93%, a specificity of 96%, and a sensitivity of 90% (Fig. 6a, b). The AUC for the AMI signature comprising 20 miRNAs (miR-142-5p, -498, -492, -1281, -497*, -151-5p, -802, -23b*, -455-3p, -1250, -380*, -135b*, -345, -566, -631, -1254, -139-5p, -892b, and -146b-3p) is 0.99, representing a significant improvement to single miRNA markers. As shown in the classification result, this signature allows discrimination of patients and controls except for three outliers (Fig. 6c). In permutation tests significantly decreased accuracy, specificity, and sensitivity rates were computed, resembling random guessing.
https://static-content.springer.com/image/art%3A10.1007%2Fs00395-010-0123-2/MediaObjects/395_2010_123_Fig5_HTML.gif
Fig. 5

Clusters of circulating miRNAs in patients with AMI versus controls. Unsupervised hierarchical clustering of expression levels of the 20 most dysregulated miRNAs using the Euclidian distance measure. AMI patients (n = 20) and control subjects (n = 20) cluster separately with only three outliers, showing that distinct miRNA patterns are unique for cases and controls, respectively

https://static-content.springer.com/image/art%3A10.1007%2Fs00395-010-0123-2/MediaObjects/395_2010_123_Fig6_HTML.gif
Fig. 6

Complex miRNA signatures predict AMI. a Classification plot demonstrating that a multi-marker signature increases test accuracy, specificity, and sensitivity depending upon the number of miRNAs that compose the diagnostic signature. b ROC analysis of the complex miRNA expression signature used to predict AMI in the study population. c Representative example of a classification result using a trained SVM. The logarithm of the quotient of the probability to be an AMI sample and the probability to be a control sample for each study sample is given on the y-axis. Of all 40 samples, only three were misclassified, leading to an accuracy of 93%, a specificity of 96%, and a sensitivity of 90% (AUC = 0.99)

Discussion

Today, biomarkers play a fundamental role in the diagnosis of cardiovascular diseases such as acute myocardial infarction and heart failure. However, many of these markers have shortcomings such as reduced sensitivity, not sufficient specificity or do not allow timely diagnosis. A multiple biomarker strategy may circumvent these limitations by adding accuracy and predictive power. To evaluate for the first time if complex miRNA signatures could potentially serve as a new class of biomarkers, we assessed here whole-genome miRNA expression levels in total peripheral blood from patients with acute myocardial infarction (AMI). We find that a unique miRNA signature predicts AMI even at a stage where some patients are still Troponin T negative.

In recent studies, miRNAs were identified as novel regulators and modifiers of cardiac development, function, and disease [29]. For instance, miRNA-21 not only controls cardiac fibrosis in response to cardiac overload but also is upregulated in the myocardium during the early phase of infarction [11, 36]. We measured increased levels of miR-21 in AMI patients in comparison to healthy individuals, but not in comparison to our control cohort, which includes a high number of patients with stable coronary artery disease (CAD). Hence, the high levels of miR-21 in AMI patients might be due to the underlying CAD, which goes along with elevated miR-21 levels in circulating angiogenic progenitor cells [14]. Likewise, most miRNAs found to be dysregulated in AMI patients are not heart-specific. However, this is not surprising since our miRNA profiles are derived from whole peripheral blood and not myocardial tissue. Hence, dysregulated miRNAs in AMI might equally be derived from other cellular population that play an active role in AMI pathophysiology, such as endothelial and smooth muscle cells involved in plaque rupture and vessel injury, thrombocytes in aggregation, and inflammatory cells that are recruited to the ischemic area. For instance, the vascular smooth muscle-enriched miR-145 is involved in neointima repair in response to vascular injury, regulating cytoskeletal components and migratory activity of smooth muscle cells (SMC) [41]. Thus, elevated miR-145 levels in the AMI group might be a result of vessel injury during plaque rupture. Similarly, miR-27b, which directly targets and destabilizes peroxisome proliferator-activated receptor-gamma (PPARγ) mRNA [21], is also upregulated in the AMI group (24.8-fold; p < 0.05). Consequently, elevated miR-27b might result in decreased PPARγ expression in SMCs, enhancing their ability to proliferate, migrate, and participate vascular remodeling processes [30]. Peripheral blood monocytes (PBMC) are critically involved in plaque destabilization and rupture as well as early inflammatory responses during myocardial infarction [6, 23]. MiR-134, which is strongly upregulated in our AMI cohort, was very recently identified as PBMC-based biomarker that is able to identify CAD patients at risk for acute coronary syndromes (ACS) [20]. Thus, studying the time-course of miR-134 in PBMCs might reveal its potential as a very early marker for ACS and AMI. In addition to miRNAs transcribed in peripheral blood cells, miRNAs released from damaged tissue can be taken up by leukocytes, resulting in the so-called convergence of serum and cellular miRNAs [8]. A recent study identified miRNAs specifically modulated in peripheral blood cells of heart failure patients. Intriguingly, many of the identified miRNAs were previously reported to be misexpressed in failing hearts of humans and mice [38]. Accordingly, next to the non-cardiac miRNAs, which significantly contribute to the AMI-specific signature, we also found important, cardiac-enriched miRNAs such as miR-30c to be upregulated in blood from AMI patients, implicating that those might have been released from ischemic myocardium.

Cardiac Troponins are currently the best validated biomarkers for the diagnosis of AMI. However, measurable amounts of Troponin proteins are usually not released from damaged myocardium before 4 to 8 h after onset of symptoms, making an early biomarker-based diagnosis of AMI rather difficult. Accordingly, 50% of AMI patients included in this study were still Troponin-negative (TnT < 50 pg/ml) when entering the catheter laboratory and blood was drawn for miRNA analyses. However, miRNA profiles of these patients also showed the AMI characteristic signature, implicating that miRNA signatures might improve biomarker-based early diagnosis of myocardial infarction.

Next to their diagnostic potential, cardiac biomarkers can provide information on disease severity and prognosis [1, 33, 34]. For instance, cardiac Troponin T levels directly correlate with the size of myocardial infarction [24, 37]. Accordingly, we find that early expression levels of miR-30c and miR-145 significantly correlate with Troponin T levels. Interestingly, miR-30c is highly expressed in the heart and hence might be directly derived from damaged myocardium [13].

In a recently published candidate approach, miRNA-1 levels were found to be elevated in serum of AMI patients [3]. However, with this single biomarker approach, AMI patients could be distinguished from controls with only a moderate sensitivity and specificity (AUC = 0.774), indicating that miRNA-1 might not be the optimal marker. In subsequent candidate studies using different cardiac miRNAs, such as miR-208a, -133b, and -499-5p, the predictive values were already considerably better [2, 10, 40]. To further increase the predictive power of miRNA-based markers, we combined here whole-genome miRNA profiling with pattern-recognition algorithms. As shown, our multi-marker approach might further increase sensitivity, specificity, and accuracy. However, the number of samples used in our analysis is too small to definitely proof the diagnostic power of microRNA signatures and their value for clinical testing of AMI patients. Hence, future prospective trials on large patient cohorts are needed to establish miRNAs as a novel biomarker class for acute myocardial infarction.

Acknowledgments

This work was supported in parts by funding of the German Ministry of Research Education (BMBF 01EX0806) and by grants from the Postdoc-Fellowship of the medical faculty of the University of Heidelberg.

Conflict of interest

A. Keller and A. Borries are employed by febit biomed GmbH. All other authors declare that they have no conflict of interest.

Supplementary material

395_2010_123_MOESM1_ESM.ppt (283 kb)
Supplementary material 1 (PPT 283 kb)

Copyright information

© Springer-Verlag 2010