Transcript PowerPoint-Präsentation

Prediction for successful treatment of
methotrexate in rheumatoid arthritis with
mRNA and miRNA microarraydata
Pascal Schendel, Marc Bonin, Karsten Mans, Florian Heyl, Jekaterina Kokatjuhha, Sascha Johannes, Irene Ziska,
Biljana Smiljanovic, Till Sörensen, Bruno Stuhlmüller, Thomas Häupl
Department of Rheumatology and Clinical Immunology, Charité University Hospital, Berlin, Germany
Background and Objective:
Materials and Methods:
Treatment of chronic arthritis is challenged by the need to adapt dosis or exchange
therapeutic agents without prior knowlegde of the individual response characteristics.
With genomewide screening of transcriptional activities in whole blood, we hope to
identify molcular patterns that help to distinguish responders from non-responders
prior to treatment and thus may support therapeutic decisions.
Samples were collected before Methotrexate treatment. The transcriptomes were
determined by Affymetrix technology and the therapeutic response level by clinical
follow-up in an observational study. To select potential molecular predictors and to
compare the molecular classifiers between different clinical response groups, various
R-packages were applied.
Figure 1
number of transcripts
Results:
Good to very good predictors could be identified by using the Limma and the Lasso
algorithms in the mRNA transcriptoms, which enabled to classify nearly without an
error by linear discrimination analysis (LDA). Between groups of genes determined by
different selection methods an overlap up to 40% could be reached and hierachical
clustering generated nearly perfect grouping. Nevertheless among mRNAs the
heatmap patterns seemed to be in part heterogeneous. The analysis was repeated
after splitting the samples into two groups with respect to the expression level of the
gene HLA-DRB4 (Fig.3), a gene locus, which is genetically important for risk prediction
in rheumatoid arthritis. Between the molecular predictors of response for the two
groups of HLA-DRB4 positive and negative patients no overlap could be found. Also
the overlap with the predictors of the combined group decreased notable. Similar
results were observed when analysing the microRNA transcriptoms. Finally the
samples were seperated into a test and a training set for an independent validation.
Only the investigation of the groups spitted by HLA criteria showed adequate
reproducibility whereas the combined group obviously generated unstable predictors.
transcript
selection
classification
Conclusion:
2 - 38
Figure 2
In summary, combining the advantages of different algorithms like Limma, Lasso and
LDA for selecting and testing molecular predictors for clinical response increases the
dignostic power of biomarkers. Nevertheless, appropriate characterization and
splitting into distinct subgroups is essential to increase reproducibility and validity in
biomarker development.
Figure 5
miRNA
biomarker
without grouping by
HLA-DRB4 positive
and negative.
Figure 3
HLA-DRB4- positive Group with 5 miRNAs,
after Kruskal-Wallis-Test & SVM.
Reciever Operation Characteristic
HLA-DRB4- negative Group with 4
miRNAs, after Kruskal-Wallis-Test.
Reciever Operation Characteristic
HLA-DRB4 Groups (mRNA): HLA-DRB4 negative: min. intensity: 4, max. intensity: 66
HLA-DRB4 positive: min. intensity: 1464, max. intensity: 5354
Figure 4
HLA-DRB4- positive Group with 19
miRNAs, after Kruskal-Wallis-Test.
HLA-DRB4- negative Group with 17
miRNAs, after Kruskal-Wallis-Test.
Contacts:
Acknowledgement: BTCure IMI grant agreement no. 115142
ArthroMark grant no 01EC1009A
ROC of cross validation: HLA-DRB4- positive Group
with 5 miRNAs, after Kruskal-Wallis-Test & SVM.
ROC of cross validation: HLA-DRB4- negative Group
with 4 miRNAs, after Kruskal-Wallis-Test.
HLA-DRB4- positive Group with 3 miRNAs,
after Lasso. (100 % overlap with KruskalWallis-Test & SVM)
HLA-DRB4- negative Group with 4
miRNAs, after Lasso. (75 % overlap with
Kruskal-Wallis-Test & SVM)
Marc Bonin
Department of Rheumatology
and Clinical Immunology
Charité University Hospital
Charitéplatz 1
D-10117 Berlin Germany
Tel: +49(0) 30 450 513 296
Fax: +49(0) 30 450 513 968
E-Mail: [email protected]
Web: www.charite-bioinformatik.de
www.charite-bioinformatik.de