Background It has been reported that the specificity for discriminating polymyalgia rheumatica (PMR) from rheumatoid arthritis (RA) was up to 70% with 2012 EULAR/ACR provisional classification criteria. There are many similarities between PMR and RA, especially late-onset RA, and it may lead to significant diagnostic difficulties. Microarrays for gene expression analysis developed in recent years had made it possible to reveal the expression profile of non-coding RNAs together with protein-coding RNAs. So far, the differences of transcription profiles between PMR and RA have not been elucidated.
Objectives To identify the gene expression signature that may distinguish PMR from RA.
Methods The study included 71 RA (age 60.9±12.4), 17 PMR (age 70.4±10.8) and 11 osteoarthritis (OA: age 73.9±10.4) patients. Peripheral blood was drawn from the patients who were newly diagnosed and exposed neither steroids nor anti-rheumatic drugs. The samples were prepared and subjected to RNA extraction with QIAGEN PAXgene system. Messenger RNA levels were then measured using Agilent whole human genome 60K (single-color procedure) and the log-transformed raw intensity data were normalized with a quantile algorithm. Based on the differences in gene expression among RA, PMR and OA by ANOVA, 556 genes related to gene ontology terms were selected from top1000 significant genes, and then subjected to a hierarchical clustering with assessment of the statistical robustness. In order to discriminate RA from PMR, we also performed discriminant analysis with using differentially expressed genes (DEGs) selected by t-test.
Results The hierarchical clustering showed major 3 clusters. All of OA and 82.4% of PMR samples were segregated into 1st and 3rd clusters respectively. OA samples were aggregated in the edge of 1st cluster, while PMR samples were distributed among RA samples. In the top 100 DEGs between RA vs. OA, OA vs. PMR and PMR vs. RA, the long intergenic non-coding RNAs (lincRNAs) accounted for 7%, 9% and 26% respectively. In the comparison of PMR with RA, 86% of the significant lincRNAs were upregulated in PMR. With using top 49 DEGs containing both protein-coding RNA and lincRNA, PMR is differentiated from RA with retrospective accuracy of 98.9% by discriminant analysis. Meanwhile, it was calculated as 92.0% with using only 14 lincRNAs. We also calculated the accuracy with leave-one-out crossvalidation. The accuracy was calculated as 83.0% if the top 14 lincRNAs were used, while the accuracy was 78.4% with using the top 14 protein-coding RNAs.
Conclusions The result of hierarchical clustering shows the similarity of gene expression profile between PMR and RA in the peripheral blood. It may give us the explanation of the rather low specificity for discriminating PMR from RA by 2012 EULAR/ACR provisional classification criteria. However, if genes are selected properly, evaluation of expression profile might be a promising tool for increasing an accuracy of the differential diagnosis. We are only beginning to understand the nature and extent of the involvement of lincRNAs in diseases. It is quite interesting that the expression of lincRNA in PMR seems to be distinctive compared with that in RA.
Disclosure of Interest None Declared