Chronic fatiguing illness remains a poorly understood syndrome of unknown pathogenesis. We attempted to identify biomarkers for chronic fatiguing illness using microarrays to query the transcriptome in peripheral blood leukocytes.
Cases were 44 individuals who were clinically evaluated and found to meet standard international criteria for chronic fatigue syndrome or idiopathic chronic fatigue, and controls were their monozygotic co-twins who were clinically evaluated and never had even one month of impairing fatigue. Biological sampling conditions were standardized and RNA stabilizing media were used. These methodological features provide rigorous control for bias resulting from case-control mismatched ancestry and experimental error. Individual gene expression profiles were assessed using Affymetrix Human Genome U133 Plus 2.0 arrays.
There were no significant differences in gene expression for any transcript.
Contrary to our expectations, we were unable to identify a biomarker for chronic fatiguing illness in the transcriptome of peripheral blood leukocytes suggesting that positive findings in prior studies may have resulted from experimental bias.
Citation: Byrnes A, Jacks A, Dahlman-Wright K, Evengard B, Wright FA, Pedersen NL, et al. (2009) Gene Expression in Peripheral Blood Leukocytes in Monozygotic Twins Discordant for Chronic Fatigue: No Evidence of a Biomarker. PLoS ONE 4(6): e5805. doi:10.1371/journal.pone.0005805
Editor: Etienne Joly, Université de Toulouse, France
Received: March 2, 2009; Accepted: April 21, 2009; Published: June 5, 2009
Copyright: © 2009 Byrnes et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This project was funded by R01 AI056014 (PFS) from the National Institute of Allergy and Infectious Diseases of the US National Institutes of Health. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: In the interests of full disclosure, Dr. Sullivan reports receiving unrestricted research funding from Eli Lilly for genetic research in schizophrenia. The other authors report no conflicts.
The etiology of chronic fatigue syndrome (CFS) is unknown –. Many theories of the pathophysiology of CFS have been suggested –, often based on suspicions of the role of an acute viral illness or immune dysfunction. The availability of a biomarker for CFS would be of particular benefit for clinical and basic research.
Gene expression studies of peripheral blood leukocytes (PBLs) are a potentially promising source of biomarkers for CFS . PBLs are both accessible and salient for CFS given the prominence of immune and infectious theories of its etiology. Moreover, gene expression patterns in human PBLs are not unrelated to less accessible tissues like brain . We are aware of four published non-overlapping studies that compared gene expression in PBLs in cases with CFS in comparison to controls –. These small studies included a total of only 45 cases (5, 7, 8, and 25) and, of the 108 transcripts reported to have altered expression, only one transcript was altered in more than one study (MSN, moesin). , 
Given the lack of clarity in existing studies of CFS, we undertook an “unbiased”, transcriptome-wide search for gene expression changes associated with CFS in PBLs. Our study had two notable design features. First, we used a control group optimal for detecting state-related gene expression changes and minimized false-positive findings due to genetic mismatching between cases and controls as we contrasted 44 individuals with clinically-evaluated chronic fatiguing illness with their 44 unaffected monozygotic co-twins. Use of rigorously discordant monozygotic twins provides the best control for genetic background currently possible in humans and allows use of paired statistics with greater statistical power. Second, we carefully standardized sampling conditions so that PBL samples were drawn into RNA-stabilizing media and taken from both members of a twin pair at the same time and place.
Materials and Methods
The protocol was approved in advance by the ethical review board at UNC-CH and the Karolinska Institutet and all subjects provided written informed consent.
We screened ∼61,000 individual twins from the Swedish Twin Registry for the symptoms of fatiguing illness –. All twins were born in Sweden of Scandinavian ancestry. Of 5,597 monozygotic twin pairs where both were alive and had provided usable responses to CFS screening questions, we identified 140 pairs of twins who met preliminary inclusion criteria: born 1935–1985, classified as a monozygotic twin based on questionnaire responses , and discordant for chronic fatiguing illness (i.e., one twin reported substantial fatigue and the other twin was evidently well). A telephone interview using a standardized script was used to assess eligibility for participation. Twins who remained eligible attended a half-day clinical assessment by a specially trained physician at the Karolinska Institutet in Stockholm. At this visit, a CFS-focused medical assessment was conducted that included standardized medical history, physical examination, and screening biochemical, hormonal, and hematological studies in accordance with international recommendations .
Of 140 monozygotic and preliminarily discordant twin pairs, one or both twins declined participation in 23 pairs, 25 pairs were concordant for CFS-like illness, and inclusion criteria were not met in 35 pairs (e.g., chronic fatigue had resolved or an illness that could explain fatiguing symptoms such as neoplasia had emerged). After excluding these 83 pairs, 57 pairs of twins attended the clinical evaluation sessions, and 10 pairs were found not to meet inclusion criteria (9 pairs were concordant for the presence or absence of chronic fatigue or a medical explanation was detected – e.g., newly diagnosed type 2 diabetes mellitus – and 1 pair was dizygotic). Zygosity was confirmed by genotyping 46 single nucleotide polymorphisms using two Sequenom iPlex panels. In 3 pairs, microarray data of satisfactory quality could not be obtained.
The analysis sample consisted of 44 pairs of rigorously discordant and genetically proven monozygotic twins. Discordance was defined as one twin meeting criteria for either idiopathic chronic fatigue (ICF, 12 pairs) or CFS (32 pairs) ,  and the co-twin was required never to have experienced impairing unusual fatigue or tiredness lasting more than one month. Thus, all affected twins were required to have current, long-standing (≥6 months), medically unexplained fatigue associated with substantial impairment in social and occupational functioning and the unaffected co-twins were effectively well. A diagnosis of CFS adds a requirement for ≥4 of 8 specific symptoms (e.g., unrefreshing sleep, muscle pain) to that of ICF. We explain elsewhere the rationale for including ICF along with CFS based on phenotypic  and twin analyses .
Transient/situational factors can influence gene expression measurements. Biological sampling was standardized by having samples drawn from both members of a twin pair at the same place and time (∼0900) after an overnight fast. We required that all subjects be in their usual state of health on the day of sampling (i.e., no acute illness or recent exacerbation of a chronic illness). It was neither practical nor ethical to study subjects medication-free, but we delayed assessment if there had been a recent significant dosage change.
Peripheral venous blood was drawn using sterile technique into PAXgene tubes manufactured in the same batch (Qiagen, to protect RNA from degradation and to minimize ex vivo gene expression). Total RNA was purified using the PAXgene blood RNA kit following the manufacturer's instructions (Qiagen). RNA quality was determined using the Agilent 2100 Bioanalyzer. Total RNA (5.0 µg) was labeled with the one-cycle cDNA synthesis kit (Invitrogen) and spiked with eukaryotic Poly-A RNA controls to check the target labeling process (Affymetrix). Synthesized cDNA was transcribed in vitro using the GeneChip IVT labeling kit (Affymetrix). The biotin labeled cRNA product (20 µg) was purified with a sample cleanup module (Qiagen) and samples were fragmented with the fragmentation buffer from Affymetrix at 94°C for 35 minutes. Fragmented and labeled targets (together with hybridization and oligo B2 controls) were hybridized to Affymetrix Human Genome U133 Plus 2.0 arrays at 45°C for 16 hours. Washing and staining of the arrays were performed on the Affymetrix fluidics station using the EukGE-WS2v5_450 protocol. Imaging of the arrays and signal quantification were performed with the Affymetrix GeneChip Scanner 3000 and GeneChip Operating Software. For verification, we used qRT-PCR. RNA was converted to cDNA with Superscript III (Invitrogen) and qRT-PCR was run with ABI's Taqman gene expression assays (with 18S rRNA as control). The ΔΔCt method was used for the calculations.
Array images were manually checked for defects using DChip ,  and then normalized using the RMA algorithm in Affymetrix Expression Console (v1.0). After normalization, the Bioconductor  Significance Analysis of Microarrays package  was used to compute modified paired t-tests that contrasted an affected twin with the unaffected co-twin for each transcript using R . To adjust for multiple comparisons, the nominal permutation-based p-values from SAM were used to compute false discovery rate q-values –. Pathway analyses for KEGG pathways , GO keywords  (biological process, cellular component, and molecular function), and PFAM protein family groupings  were conducted using SAFE which performs array permutation to account for transcript correlation , . These expression data are available from GEO (http://www.ncbi.nlm.nih.gov/geo) under accession number GSE16059 and were prepared in accordance with MIAME 2.0 standards.
The analytic data set consisted of microarray data from 44 pairs of monozygotic twins discordant for clinically-evaluated chronic fatiguing illness (Table 1). Most pairs were female (89%), and the median age at evaluation was 51 years. Of the affected twins, 32 met criteria for CFS and 12 for ICF with a median duration of chronic fatigue of 8 years with no significant difference between affected twins with CFS and ICF (paired t43 = 0.32, p = 0.75). Body mass index was similar between the affected and unaffected twins. Two affected individuals (4.5%) reported sudden onset of fatigue. Affected twins had significantly worse physical and mental functioning on the SF-36  and reported significantly greater current fatigue. The mean functioning of affected twins was over a standard deviation worse than Swedish norms whereas the unaffected twins were similar to Swedish norms (http://www.sf-36.org/nbscalc/index.shtml, accessed 12 December 2008).
The main analyses contrasted gene expression in PBLs in 44 pairs of monozygotic twins affected with CFS or ICF to that of their unaffected co-twins. As inclusion of males could increase noise, the second planned analysis compared 39 female pairs with CFS or ICF and the third planned analysis compared the 28 pairs of female twins with CFS. For each of these three sets of statistical comparisons, the observed results did not deviate from those expected by chance (Figure S1). When we compared our findings to a list of 108 transcripts reported as differentially expressed in CFS –, 107 were studied in our experiment, 101/107 had p>0.1 in our study, and only two had p<0.05 (CXCR4 p = 0.03 and RAP2C p = 0.04), a degree of overlap that does not depart from chance expectations. At the single transcript level, there was no biological evidence of altered gene expression in PBLs that correlated with chronic, impairing, and medically unexplained fatiguing illness. For verification, we used qRT-PCR to assess the expression of seven genes selected from the CFS gene expression literature and our empirical findings (ANKLE2, BLKE, BRD1, CPA3, DCTN1, ICAM, and ORC). All p-values from paired t-tests contrasting affected and unaffected monozygotic twins were ≥0.26.
It is possible that functionally-related genes might have important set-wise gene expression changes with no individual transcript meeting criteria for significance. As a hypothesis-generating analysis, we used SAFE  to conduct analyses of gene groupings defined by KEGG pathways , GO keywords  (biological process, cellular component, and molecular function), and PFAM protein family groupings . Broadly, these analyses revealed significant differences in cell replication processes and amino acid and lipid metabolic pathways (Table S1). These results do not map directly onto current major theories of CFS pathogenesis and should be regarded as hypothesis-generating.
The overarching goal in this study was to attempt to identify one or more biomarkers for chronic fatiguing illness via a comprehensive search of the “transcriptome” in an accessible tissue (peripheral blood leukocytes, PBLs) plausibly involved in the pathophysiology of this idiopathic syndrome. We attempted to correct methodological issues in prior reports by careful control of sources of bias (e.g., by studying discordant monozygotic twin pairs, use of RNA stabilizing media, and standardized sampling conditions). We found no evidence of differential PBL gene expression that characterized the presence or absence of CFS or ICF. Therefore, unlike most prior published studies, we did not find evidence of a gene expression biomarker for chronic fatiguing illness.
Methodological Issue: genetic matching in gene expression studies
These results may hold a lesson for case-control gene expression studies in humans. There are certainly examples where transcriptomic studies have yielded results that have the potential to improve disease prognosis and management (e.g., breast cancer) ; however, gene expression studies have the potential to yield false positive findings if the ancestry of cases and controls are not appropriately matched. Genetic background is usually not taken into consideration although gene expression is can be both heritable and under strong genetic control . Relatively low-resolution studies in immortalized PBLs suggest that hundreds of human genes are under relatively strong genetic control by common genetic variants (e.g., the single nucleotide polymorphism “rs407257” is strongly associated (p∼10−66) with the expression level of glutathione S-transferase theta 1, GSTT1) . The genetic variant rs407257 is variable in human populations (allele frequencies of 0.72, 0.64, and 0.39 in African, East Asian, and European samples) . If case and control subjects are not extremely well-matched for genetic background (including for location within Europe) highly significant differences could occur because of bias from inappropriate case-control matching. This concern is particularly important for studies of PBLs as genes whose expression is under strong genetic control  are highly enriched for genes expressed in lymphoid tissue and lymphocyte cell populations (analyses using DAVID , data not shown).
Use of discordant monozygotic twins represents the best control for genetic background currently possible in humans. Assuming identify at the DNA level and control for experimental bias, gene expression differences in discordant monozygotic twins can be cleanly attributed to disease state. It is reasonable to consider if use of discordant monozygotic twins represents “over-matching”. In comparisons of unrelated cases and controls, gene expression differences are an amalgam of disease state, the RNA-level impact of genetic loci causal to the trait, and the effects of case-control genetic mismatching (i.e., non-causal loci that differ in frequency between cases and controls and which have strong control on gene expression). Use of discordant monozygotic twins yields more interpretable results particularly as there are large numbers of non-causal loci under genetic control. We would also argue that a gene expression study is a poor way to identify genetic loci causal to a disease when an alternative study design (the genome-wide association study) has been so successful .
We were unable to identify a biomarker for chronic fatiguing illness in the transcriptome of peripheral blood leukocytes suggesting that positive findings in prior studies may have resulted from experimental bias.
Shown are results from pathway analyses using SAFE to investigate KEGG pathways, GO keywords (BP = biological process, CC = cellular component, and MF = molecular function), and PFAM protein family groupings. These all had an empirical p-value (from permutation) <0.005 and were composed of ≥10 transcripts.
(0.11 MB DOC)
Quantile-quantile (QQ) plots from the planned paired analyses contrasting monozygotic (MZ) twins affected with chronic fatiguing illness versus their unaffected co-twins. CFS = chronic fatigue syndrome, ICF = idiopathic chronic fatigue. The observed distribution of statistical results conform to chance expectations.
(0.11 MB DOC)
Conceived and designed the experiments: FW NLP PFS. Performed the experiments: AJ KDW BE. Analyzed the data: AB KDW FW PFS. Contributed reagents/materials/analysis tools: FW PFS. Wrote the paper: AB NLP PFS.
- 1. Fukuda K, Strauss SE, Hickie I, Sharpe MC, Dobbins JG, et al. (1994) The chronic fatigue syndrome: a comprehensive approach to its definition and study. Annals of Internal Medicine 121: 953–959.
- 2. Reeves WC, Lloyd A, Vernon SD, Klimas N, Jason LA, et al. (2003) Identification of ambiguities in the 1994 chronic fatigue syndrome research case definition and recommendations for resolution. BMC Health Serv Res 3: 25.
- 3. Baker R, Shaw EJ (2007) Diagnosis and management of chronic fatigue syndrome or myalgic encephalomyelitis (or encephalopathy): summary of NICE guidance. Bmj 335: 446–448.
- 4. Komaroff AL, Buchwald DS (1998) Chronic fatigue syndrome: an update. Annual Review of Medicine 49: 1–13.
- 5. Sharpe M (1996) Chronic fatigue syndrome. Psychiatric Clinics of North America 19: 549–573.
- 6. Cho HJ, Skowera A, Cleare A, Wessely S (2006) Chronic fatigue syndrome: an update focusing on phenomenology and pathophysiology. Curr Opin Psychiatry 19: 67–73.
- 7. Devanur LD, Kerr JR (2006) Chronic fatigue syndrome. J Clin Virol 37: 139–150.
- 8. Klimas NG, Koneru AO (2007) Chronic fatigue syndrome: inflammation, immune function, and neuroendocrine interactions. Curr Rheumatol Rep 9: 482–487.
- 9. Chen R, Liang FX, Moriya J, Yamakawa J, Sumino H, et al. (2008) Chronic fatigue syndrome and the central nervous system. J Int Med Res 36: 867–874.
- 10. Kerr JR, Christian P, Hodgetts A, Langford PR, Devanur LD, et al. (2007) Current research priorities in chronic fatigue syndrome/myalgic encephalomyelitis: disease mechanisms, a diagnostic test and specific treatments. J Clin Pathol 60: 113–116.
- 11. Sullivan PF, Fan C, Perou CM (2006) Evaluating the comparability of gene expression in blood and brain. American Journal of Medical Genetics (Neuropsychiatric Genetics) 141: 261–268.
- 12. Vernon SD, Unger ER, Dimulescu IM, Rajeevan M, Reeves WC (2002) Utility of the blood for gene expression profiling and biomarker discovery in chronic fatigue syndrome. Dis Markers 18: 193–199.
- 13. Powell R, Ren J, Lewith G, Barclay W, Holgate S, et al. (2003) Identification of novel expressed sequences, up-regulated in the leucocytes of chronic fatigue syndrome patients. Clin Exp Allergy 33: 1450–1456.
- 14. Grans H, Nilsson P, Evengard B (2005) Gene expression profiling in the chronic fatigue syndrome. J Intern Med 258: 388–390.
- 15. Kerr JR, Petty R, Burke B, Gough J, Fear D, et al. (2008) Gene expression subtypes in patients with chronic fatigue syndrome/myalgic encephalomyelitis. J Infect Dis 197: 1171–1184.
- 16. Evengard B, Jacks A, Pedersen NL, Sullivan PF (2005) The epidemiology of chronic fatigue in the Swedish Twin Registry. Psychological Medicine 35: 1317–1326.
- 17. Sullivan PF, Jacks A, Pedersen NL, Evengard B (2005) Chronic fatigue in a population sample: definitions & heterogeneity. Psychologal Medicine 35: 1337–1348.
- 18. Sullivan PF, Evengard B, Jacks A, Pedersen NL (2005) Twin analyses of chronic fatigue in a Swedish national sample. Psychol Med 35: 1327–1336.
- 19. Lichtenstein P, Sullivan P, Cnattingius S, Gatz M, Johansson S, et al. (2006) The Swedish Twin Registry in the Third Millennium – an update. Twin Res Hum Genet 9: 875–882.
- 20. Li C, Wong WH (2001) Model-based analysis of oligonucleotide arrays: expression index computation and outlier detection. Proc Natl Acad Sci U S A 98: 31–36.
- 21. Lin M, Wei LJ, Sellers WR, Lieberfarb M, Wong WH, et al. (2004) dChipSNP: significance curve and clustering of SNP-array-based loss-of-heterozygosity data. Bioinformatics 20: 1233–1240.
- 22. Gentleman R, Rossini AJ, Sudoit S (2006) Bioconductor
- 23. Tusher VG, Tibshirani R, Chu G (2001) Significance analysis of microarrays applied to the ionizing radiation response. Proc Natl Acad Sci U S A 98: 5116–5121.
- 24. R Development Core Team (2007) R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing.
- 25. Storey JD, Tibshirani R (2003) Statistical significance for genomewide studies. Proc Natl Acad Sci U S A 100: 9440–9445.
- 26. Storey JD (2003) The positive false discovery rate: a Bayesian interpretation and the q-value. Annals of Statistics 31: 2013–2035.
- 27. Storey JD, Taylor JE, Sigmund D (2004) Strong control, conservative point estimation, and simultaneous conservative consistency of false discovery rates: A unified approach. Journal of the Royal Statistical Society (Series B) 66: 187–205.
- 28. Kanehisa M, Goto S, Hattori M, Aoki-Kinoshita KF, Itoh M, et al. (2006) From genomics to chemical genomics: new developments in KEGG. Nucleic Acids Res 34: D354–357.
- 29. Harris MA, Clark J, Ireland A, Lomax J, Ashburner M, et al. (2004) The Gene Ontology (GO) database and informatics resource. Nucleic Acids Res 32 Database issue D258–261.
- 30. Coggill P, Finn RD, Bateman A (2008) Identifying protein domains with the Pfam database. Curr Protoc Bioinformatics. Chapter 2: Unit 2 5.
- 31. Barry WT, Nobel AB, Wright FA (2005) Significance analysis of functional categories in gene expression studies: a structured permutation approach. Bioinformatics 21: 1943–1949.
- 32. Barry WT, Nobel AB, Wright FA (2008) A Statistical Framework for Testing Functional Categories in Microarray Data. Annals of Applied Statistics 2: 286–315.
- 33. Ware JE Jr, Sherbourne CD (1992) The MOS 36-item short-form health survey (SF-36). I. Conceptual framework and item selection. Med Care 30: 473–483.
- 34. Peppercorn J, Perou CM, Carey LA (2008) Molecular subtypes in breast cancer evaluation and management: divide and conquer. Cancer Invest 26: 1–10.
- 35. Cheung VG, Spielman RS, Ewens KG, Weber TM, Morley M, et al. (2005) Mapping determinants of human gene expression by regional and genome-wide association. Nature 437: 1365–1369.
- 36. Stranger BE, Nica AC, Forrest MS, Dimas A, Bird CP, et al. (2007) Population genomics of human gene expression. Nat Genet 39: 1217–1224.
- 37. Altshuler D, Brooks LD, Chakravarti A, Collins FS, Daly MJ, et al. (2005) A haplotype map of the human genome. Nature 437: 1299–1320.
- 38. Huang da W, Sherman BT, Tan Q, Kir J, Liu D, et al. (2007) DAVID Bioinformatics Resources: expanded annotation database and novel algorithms to better extract biology from large gene lists. Nucleic Acids Res 35: W169–175.
- 39. Altshuler D, Daly M (2007) Guilt beyond a reasonable doubt. Nat Genet 39: 813–815.