Using biomarkers to predict progression from clinically isolated syndrome to multiple sclerosis
© Tossberg et al.; licensee BioMed Central Ltd. 2013
Received: 21 August 2013
Accepted: 30 September 2013
Published: 3 October 2013
Detection of brain lesions disseminated in space and time by magnetic resonance imaging remains a cornerstone for the diagnosis of clinically definite multiple sclerosis. We have sought to determine if gene expression biomarkers could contribute to the clinical diagnosis of multiple sclerosis.
We employed expression levels of 30 genes in blood from 199 subjects with multiple sclerosis, 203 subjects with other neurologic disorders, and 114 healthy control subjects to train ratioscore and support vector machine algorithms. Blood samples were obtained from 46 subjects coincident with clinically isolated syndrome who progressed to clinically definite multiple sclerosis determined by conventional methods. Gene expression levels from these subjects were inputted into ratioscore and support vector machine algorithms to determine if these methods also predicted that these subjects would develop multiple sclerosis. Standard calculations of sensitivity and specificity were employed to determine accuracy of these predictions.
Our results demonstrate that ratioscore and support vector machine methods employing input gene transcript levels in blood can accurately identify subjects with clinically isolated syndrome that will progress to multiple sclerosis.
We conclude these approaches may be useful to predict progression from clinically isolated syndrome to multiple sclerosis.
KeywordsGenomics Multiple sclerosis Disease prediction Diagnosis
Diagnosis of multiple sclerosis [MS] rests on clinical symptoms and examination as outlined in the revised McDonald’s criteria supported by appropriate magnetic resonance imaging findings or other laboratory tests such as detection of oligoclonal bands in cerebrospinal fluid and evoked potential testing [1–7]. Clinically isolated syndrome (CIS) is a first neurologic episode lasting at least 24 hours possibly caused by focal inflammation or demyelination [8, 9]. Approximately 10,000-15,000 new diagnoses of MS are made in the United States each year . Approximately 2–3 times that number experience a CIS each year indicating that a far greater number of subjects experience a CIS than develop MS [11–14]. Costs to healthcare of determining if a subject with a CIS will develop MS are significant considering the cost of MRI and additional testing performed and the fact that many more subjects develop CIS than MS.
Presence of abnormal MRI findings and detection of oligoclonal bands in the cerebrospinal fluid in an individual at the time of CIS increase the likelihood of an eventual diagnosis of MS. However, these findings do not guarantee an eventual diagnosis of MS nor do their absence preclude a diagnosis of MS. We have considered that measuring gene transcript patterns in blood may provide a means to develop tests with the ability to exclude the diagnosis of a given disease, such as MS, or to establish a diagnosis of MS, and have performed studies to identify gene expression patterns that distinguish subjects with MS from a) healthy control subjects, b) subjects with inflammatory neurologic conditions distinct from MS (other inflammatory neurologic conditions, OND-I), e.g. transverse myelitis , neuromyelitis optica (NMO) and c) subjects with other non-inflammatory neurologic conditions (OND-NI) [15, 16]. We have also applied this approach to gastro-intestinal diseases and have found it possible to discriminate between irritable bowel syndrome and inflammatory bowel disease, two conditions with similar clinical presentations, and to discriminate between the two most frequent and related forms of inflammatory bowel disease, ulcerative colitis and Crohn’s disease, thus demonstrating the general utility of our approach .
A limitation to these studies is that subjects included in these analyses do not completely represent patients in the general population in whom these tests may be performed. Presumably, tests would be performed on subjects who do not yet have a clinical diagnosis of a given disease. To address this limitation, we decided to examine subjects at the time they experience CIS who acquire a diagnosis of MS in the future using established criteria. We applied two independent analytic methods, a ratioscore algorithm we previously developed and support vector machines. Our results demonstrate that these methods predict future conversion to MS with a high degree of specificity.
Blood samples in PAXgene tubes were obtained from CTRL, MS, OND-I and OND-NI subjects. Samples were also obtained from subjects with CIS at the time of the blood draw. All of these subjects have gone on to develop MS according to the McDonald’s criteria for the diagnosis of MS. Age, race and gender were not statistically different among the different study groups. Time of blood draw, for example, morning/afternoon clinics, was also not statistically significant among the different study groups. Relevant institutional review board approval was obtained from all participating sites.
Total RNA purification, cDNA synthesis, and analysis using a 384-well Taqman Low Density Array (TLDA) were as previously described (Additional file 1: Figure S1) [16, 17]. Patient diagnosis was blinded for all experimental procedures. Relative expression levels were determined directly from the observed threshold cycle (CΤ). Linear expression levels were determined using the formula, 2(40-CΤ).
Ratioscore and support vector machine algorithms
The identification of the gene expression ratios and permutation testing strategy employed to identify discriminatory combinations of ratios to create the ratioscore have been previously described.16 Briefly, all possible gene-expression ratios were computed. Ratios in which the greatest number of subjects in case groups possessed a ratio value greater than the highest ratio value in the control group were saved. We performed permutation testing by randomly selecting 80% of the control group to compare with the case group and repeating this process 200 times producing 200 subsets of ratios. From these subsets of ratios, we identified the smallest number of ratios to identify the ratioscore with maximum separation between case groups and control groups. For example, we compared MS versus CTRL, MS versus OND, etc. Each comparison produced a unique set of ratios that were used to define the ratioscore algorithm for that pairing of the case–control groups.
A support vector machine (SVM) was created from each set of ratioscores using LS-SVMLab software (http://www.esat.kuleuven.be/sista/lssvmab). For example, the gene-expression ratios from the MS versus CTRL were used to create a SVM for this type of comparison. The SVM was trained with L-fold cross-validation using 60% of the data. In this type of training a certain fraction of the training set was omitted from training and the remaining portion of the partial training set was used to estimate the parameters in the SVM. Once the SVM was trained, the SVM was applied to the total data set. Numbers of correct and incorrect classifications were tabulated for total sets (training and validation), training sets and validation sets. As expected, the overall accuracy in the training sets was greater than overall accuracy of the validation sets.
Analysis of CIS➔MS subject data
Gene expression ratio data obtained from CIS➔MS cohort samples were determined and applied to the ratioscore or SVM defined by the independent training cross-comparisons, e.g. CTRL versus MS, OND versus MS. New subjects were classified into their respective category based upon their profile of gene expression ratios.
Results and discussion
Demographic characteristics of the different subject populations
43 ± 10
46 ± 10
46 ± 10
41 ± 11
35 ± 6
Our rationale for performing this two-tier analysis rather than combining the CTRL and OND subjects into one cohort was that previous studies demonstrated that accuracy was severely compromised. To confirm that this was the case in this analysis we compared the MS cohort to the combined CTRL plus OND cohort and inputted these data into the ratioscore algorithm. As expected, overall ability to discriminate MS from this combined cohort was compromised. Only 58% of MS subjects were assigned to the MS category while 100% of subjects in the combined CTRL plus OND cohort were excluded from the MS category (Additional file 2: Figure S2A). When we input data from the CIS➔MS cohort, only 28 of 46 subjects (61%) were categorized as MS (Additional file 2: Figure S2B). Thus, overall accuracy of the ratioscore method was much improved by performing two tiers of analyses, first MS versus CTRL, then MS versus OND.
We also sub-divided the OND cohort into OND-I and OND-NI (Table 1) and repeated the ratioscore algorithm to assess how well these sub-groups could be distinguished from MS (Additional file 3: Figure S3A & B). In the OND-I versus MS comparison, 90% of MS subjects were assigned to the MS class and 100% of OND-I subjects were excluded from the MS class. When we input data from the CIS➔MS cohort, 46 of 46 subjects (100%) were categorized as MS. In the OND-NI versus MS comparison, 86% of MS subjects were assigned to the MS class and 100% of OND-NI subjects were excluded from the MS class. When we input data from the CIS➔MS cohort, 46 of 46 subjects (100%) were categorized as MS. We conclude that this further subdivision of OND subjects produces only limited improvement in overall accuracy.
Accuracy of ratioscore and SVM methods
Sensitivity and specificity of ratioscore and SVM methods
CIS ➔ MS
CIS ➔ MS
CIS ➔ MS
CIS ➔ MS
To summarize, overall transcript profiles in the CIS➔MS, MS-naïve, and MS-established were markedly different and we suggest that these dynamic transitions may reflect different pathogenic states of MS or progression of MS. Thus, we suggest that this gene expression analysis could also be used to classify different stages of MS in an individual. In addition, studying the molecular origins of the robust transcript signature in CIS➔MS subjects may produce insights into the origins of MS. In spite of the differences in overall transcript profiles in these three subject groups, ratioscore and SVM methods were able to assign CIS➔MS subjects to the MS category with a high degree of accuracy. This is due, in part, to the fact that the ratioscore method does not require that all subjects within these three cohorts representing three distinct stages of disease progression possess identical gene expression signatures. In contrast, many other standard methods of analysis of gene expression signatures are dependent upon identification of overall differences between or among groups.
A limitation to this study is that we did not include subjects with an initial CIS that did not develop MS. Our rationale for not including this parameter is three-fold. First, there is not a uniform clinical definition of CIS. Second, subjects with a CIS may or may not have MRI findings indicating inflammation or demyelination and the probability that a subject with CIS will develop MS is greater if MRI lesions are also detected. Third, with our current knowledge, it is uncertain if it is experimentally possible to absolutely conclude that a person with CIS will not develop MS. In fact, the period of time between an initial CIS and diagnosis of clinically definite MS is quite variable and can exceed 5 years.
Clinical isolated syndrome
Magnetic resonance imaging
Other neurologic disorders
Other inflammatory neurologic disorders
Other non-inflammatory neurologic disorders
Support vector machines
Taqman low density array
This work was supported by the US National Institutes of Health grants AI053984, AI044924 and ULITR000445 and National Multiple Sclerosis Society grant RG4576A2/1.
- Swanton JK, Rovira A, Tintore M, Altmann DR, Barkhof F, Filippi M, Huerga E, Miszkiel KA, Plant GT, Polman C, et al: MRI criteria for multiple sclerosis in patients presenting with clinically isolated syndromes: a multicentre retrospective study. Lancet Neurol. 2007, 6: 664-665. 10.1016/S1474-4422(07)70181-3.View ArticleGoogle Scholar
- Polman CH, Reingold SC, Edan G, Filippi M, Hartung HP, Kappos L, Lublin FD, Metz LM, McFarland HF, O’Connor PW, et al: Diagnostic criteria for multiple sclerosis: 2005 revisions to the "McDonald criteria". Ann Neurol. 2005, 58: 840-846. 10.1002/ana.20703.View ArticlePubMedGoogle Scholar
- McDonald WI, Compston A, Edan G, Goodkin D, Hartung HP, Lublin FD, McFarland HF, Paty DW, Polman CH, Reingold SC, et al: Recommended diagnostic criteria for multiple sclerosis: guidelines from the international panel on the diagnosis of multiple sclerosis. Ann Neurol. 2001, 50: 121-127. 10.1002/ana.1032.View ArticlePubMedGoogle Scholar
- Awad A, Hemmer B, Hartung HP, Kieseier B, Bennett JL, Stuve O: Analyses of cerebrospinal fluid in the diagnosis and monitoring of multiple sclerosis. J Neuroimmunol. 2009, 219: 1-7.View ArticlePubMedGoogle Scholar
- Link H, Huang YM: Oligoclonal bands in multile sclerosis cerebrospinal fluid: an update on methodology and clinical usefulness. J Neuroimmunol. 2006, 180: 17-28. 10.1016/j.jneuroim.2006.07.006.View ArticlePubMedGoogle Scholar
- Chiappa KH, Ropper AH: Evoked potentials in clinical medicine (second of two parts). N Engl J Med. 1982, 306: 1205-1211. 10.1056/NEJM198205203062004.View ArticlePubMedGoogle Scholar
- Chiappa KH, Ropper AH: Evoked potentials in clinical medicine (first of two parts). N Engl J Med. 1982, 306: 1140-1150. 10.1056/NEJM198205133061904.View ArticlePubMedGoogle Scholar
- O'Riordan JI, Thompson AJ, Kingsley DP, MacManus DG, Kendall BE, Rudge P, McDonald WI, Miller DH: The prognostic value of brain MRI in clinically isolated syndromes of the CNS. A 10-year follow-up. Brain. 1998, Pt 3: 495-503.View ArticleGoogle Scholar
- Comi G, Martinelli V, Rodegher M, Moiola L, Bajenaru O, Carra A, Elovaara I, Fazekas F, Hartung HP, Hillert J, et al: Effect of glatiramer acetate on conversion to clinically definite multiple sclerosis in patients with clinically isolated syndrome (PreCISe study): a randomised, double-blind, placebo-controlled trial. Lancet. 2009, 374: 1503-1511. 10.1016/S0140-6736(09)61259-9.View ArticlePubMedGoogle Scholar
- Jacobson DL, Gange SJ, Rose NR, Graham NM: Epidemiology and estimated population burden of selected autoimmune diseases in the United States. Clin Immunol Immunopathol. 1997, 84: 223-243. 10.1006/clin.1997.4412.View ArticlePubMedGoogle Scholar
- Comi G, Filippi M, Barkhof F, Durelli L, Edan G, Fernandez O, Hartung H, Seeldrayers P, Sorensen PS, Rovaris M, et al: Effect of early interferon treatment on conversion to definite multiple sclerosis: a randomised study. Lancet. 2001, 357: 1576-1582. 10.1016/S0140-6736(00)04725-5.View ArticlePubMedGoogle Scholar
- Koch MW, Mostert JP, de Vries JJ, De Keyser J: Treatment with interferon beta-1b delays conversion to clinically definite and McDonald MS in patients with clinically isolated syndromes. Neurology. 2007, 68: 1163-1164. 10.1212/01.wnl.0000261131.23708.1c. 1163; author replyView ArticlePubMedGoogle Scholar
- Kappos L, Polman CH, Freedman MS, Edan G, Hartung HP, Miller DH, Montalban X, Barkhof F, Bauer L, Jakobs P, et al: Treatment with interferon beta-1b delays conversion to clinically definite and McDonald MS in patients with clinically isolated syndromes. Neurology. 2006, 67: 1242-1249. 10.1212/01.wnl.0000237641.33768.8d.View ArticlePubMedGoogle Scholar
- Dalton CM, Brex PA, Miszkiel KA, Hickman SJ, MacManus DG, Plant GT, Thompson AJ, Miller DH: Application of the new McDonald criteria to patients with clinically isolated syndromes suggestive of multiple sclerosis. Ann Neurol. 2002, 52: 47-53. 10.1002/ana.10240.View ArticlePubMedGoogle Scholar
- Fossey SC, Vnencak-Jones CL, Olsen NJ, Sriram S, Garrison G, Deng X, Crooke P, Aune TM: Identification of molecular biomarkers for multiple sclerosis. J Mol Diagn. 2007, 9: 197-204. 10.2353/jmoldx.2007.060147.PubMed CentralView ArticlePubMedGoogle Scholar
- Tossberg JT, Crooke PS, Henderson MA, Sriram S, Mrelashvili D, Chitnis S, Polman C, Vosslamber S, Verweij CL, Olsen NJ, Aune TM: Gene-expression signatures: biomarkers toward diagnosing multiple sclerosis. Genes Immun. 2012, 13: 146-154. 10.1038/gene.2011.66.PubMed CentralView ArticlePubMedGoogle Scholar
- Crooke PS, Tossberg JT, Horst SN, Tauscher JL, Henderson MA, Beaulieu DB, Schwartz DA, Olsen NJ, Aune TM: Using gene expression data to identify certain gastro-intestinal diseases. J Clin Bioinformatics. 2012, 2: 20-10.1186/2043-9113-2-20.View ArticleGoogle Scholar
- Ray S, Britschgi M, Herbert C, Takeda-Uchimura Y, Boxer A, Blennow K, Friedman LF, Galasko DR, Jutel M, Karydas A, et al: Classification and prediction of clinical Alzheimer's diagnosis based on plasma signaling proteins. Nat Med. 2007, 13: 1359-1362. 10.1038/nm1653.View ArticlePubMedGoogle Scholar
- Liu Z, Maas K, Aune TM: Identification of gene expression signatures in autoimmune disease without the influence of familial resemblance. Hum Mol Genet. 2006, 15: 501-509.View ArticlePubMedGoogle Scholar
- Maas K, Chan S, Parker J, Slater A, Moore J, Olsen N, Aune TM: Cutting edge: molecular portrait of human autoimmune disease. J Immunol. 2002, 169: 5-9.View ArticlePubMedGoogle Scholar
- Maas K, Chen H, Shyr Y, Olsen NJ, Aune T: Shared gene expression profiles in individuals with autoimmune disease and unaffected first-degree relatives of individuals with autoimmune disease. Hum Mol Genet. 2005, 14: 1305-1314. 10.1093/hmg/ddi141.View ArticlePubMedGoogle Scholar
- Gandhi KS, McKay FC, Cox M, Riveros C, Armstrong N, Heard RN, Vucic S, Williams DW, Stankovich J, Brown M, et al: The multiple sclerosis whole blood mRNA transcriptome and genetic associations indicate dysregulation of specific T cell pathways in pathogenesis. Hum Mol Genet. 2010, 19: 2134-2143. 10.1093/hmg/ddq090.View ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.