Chronic Fatigue Syndrome (CFS) is a complex and poorly understood illness that affects between 1 and 4 million individuals [1, 2] and costs an estimated $35 billion per year in lost productivity and health care [3, 4]. Persistent and unexplained fatigue has been observed as a sequela of acute infection since the first half of the 20th century and the period’s notable outbreaks of infectious disease (e.g., polio epidemics, influenza pandemics). From 1934 to 1984, there were several reports of infectious disease outbreaks simulating poliomyelitis [5, 6] though no causal pathogen was isolated from the majority of the cases. Clinical investigation of individuals who did not recover several years after initial infection showed persistent symptoms of fatigue, sleep disturbance and cognitive impairment [7, 8]. Following an outbreak investigation in 1984 by the Centers for Disease Control and Prevention (CDC) in Incline Village, Nevada [9], an international group of medical experts defined this protracted post-infection illness as chronic fatigue syndrome (CFS) [10]. Since then, post-infection illness that lasts for more than six months is a central component of CFS, and studies have shown that the incidence of CFS following infection with a number of different pathogens is about 10% [11–14]. The pathogens that most commonly lead to CFS in prospective studies are those that cause infectious mononucleosis (IM)/glandular fever (mainly Epstein-Barr virus (EBV); occasionally cytomegalovirus (CMV), human herpesvirus 6, hepatitis A, and adenovirus [11], Q fever and Ross River virus (mainly in Australia [13]), enteroviruses [15] and parvovirus B19 [16, 17]. Indeed in recent work by Kerr and colleagues, antibody testing for EBV, enterovirus, Coxiella burnetii (causative agent of Q fever) and parvovirus B19 revealed evidence of subtype-specific relationships for EBV and enterovirus, two of the most common infectious triggers of CFS/ME [18].

There remains therefore an overwhelming body of evidence reinforcing the link to an infectious etiology in at least a subset of CFS patients, despite the fact that specific viral serology are not required for diagnosing CFS [19, 20]. Specifically, there is a notable and long-standing association of CFS with EBV infection. In our previously published analyses of the adolescent cohort presented here [14, 21–24], we reported a diagnosis of CFS at 6, 12 and 24 months following IM, in 13%, 7% and 4%, of the subjects respectively. Ninety percent of CFS subjects at 6 months were female as were all cases at 12 (n = 22) and 24 (n = 13) months following IM. This is consistent with the rates observed by Buchwald and colleagues (2000) [12] who found 12% of subjects had not recovered and were reporting symptoms of fatigue and impaired functioning 6 months following onset of IM, 76% of whom were female. Nearly identical results were also reported in an Australian cohort where 11% and 9% of subjects satisfied the case definition for CFS at 6 and 12 months following onset of IM, Q fever and Ross River fever (RRF) [13].

Investigators have sought to understand the underlying causes of these persistent symptoms by surveying immune signaling in the CFS patient population. Early measurements of cytokine concentrations in blood samples from this broader population produced widely discordant results due mainly to differences in case definition and laboratory protocols [25]. Limiting factors also include the focus on a varied and narrow selection of cytokines [26] as well as a conventional univariate analysis that does not account for the context-specific expression of cytokines [27]. Only one previous study has prospectively evaluated cytokine production in CFS patient populations where a uniform infectious trigger was supported by serology. Hickie et al. (2006) [13] measured the concentrations of 8 cytokines: IL-1b, IL-2, IL-4, IL-6, IL-10, IL-12, TNF-α, and IFN-γ in serum and culture supernatants of peripheral blood mononuclear cells collected 1, 2, 3, 6 and 12 months following the acute illness (IM, RRF or Q fever). The participants were 22 subjects with confirmed PI-CFS (11 EBV, 6 RRV, and 5 Q fever) and 42 control subjects (17 EBV, 14 RRV, and 11 Q fever) matched for age from the same cohort who had recovered within 6 weeks of symptom onset. The analyses were performed on all PI-CFS subjects as a single group, combining the 11 EBV patients with those infected by a virus uncommon outside Australia. They found no significant differences between PI-CFS cases and recovered controls in serum or culture supernatants [28]. However, serum cytokine levels were almost exclusively below the assay detection limit (8–15 pg/ml) and supernatants were cultured from cryopreserved PBMCs. As a result several unanswered questions remain. For example, do key CFS-specific cytokines remain unmeasured? Perhaps most importantly, could PI-CFS cytokine signatures reflect pathogen-specific (and gender-specific) variations in immune response? If so these may be lost when pooling subjects with different infectious triggers (and different genders).

In an attempt to address some of these questions, this secondary analysis has focused specifically on a group of female adolescents diagnosed PI-CFS as the result of a uniform and known pathogen, namely EBV. An extended survey of 16 cytokines was conducted in plasma using a modified chemiluminescent assay [27] and identified significant differences in IL-8 and IL-23 concentrations in the patient group at 24 months post-infection. Because cytokines are expressed in a coordinated fashion we also extended the analysis beyond the traditional univariate analysis to investigate combinatorial effects across multiple cytokines. Taken together, a classification of subjects based on levels of IL-2, 6, 8 and 23 supported assignment into the patient or control group with an accuracy exceeding 80% when applied relative to interferon gamma (IFN-γ) concentration. Interestingly the latter did not differ significantly in expression level across subject groups.

In this work we studied a population of adolescents that developed CFS as a sequela of monospot-positive IM and compared them to control subjects who recovered normally from the same infection. Approximately 4% of subjects (or 13 individuals) diagnosed with IM fit the case definition for CFS at 24 months, a number consistent with other prospective studies of infectious onset CFS [12, 13]. In comparing the concentrations of 16 cytokines between cases and recovered controls we found significantly different levels of IL-8, IL-23 and possibly IL-2 and IL-5 in plasma samples from CFS patients. Increased IL-8 and decreased IL-5 is a pattern also seen asthma [34] and in B cell chronic lymphocytic leukemia (B-CLL) [35]. Most striking were differences in IL-23, a cytokine expressed by dendritic cells and macrophages. Though not required for the generation of Th17 cells from naïve CD4+ T cells, IL-23 is nonetheless essential for the full and sustained differentiation of the Th17 cell subset [36]. Median concentration of this cytokine was significantly lower in patients with CFS than in recovered controls. IL-23 expression is highly inducible in PHA-stimulated CD4+ cells, in particular when primed with IFN-γ, suggesting a potential role for IL-23 in Th1 response [37]. We found decreased expression of IL-23 in CFS patients despite comparable levels of IFN-γ. In addition we found levels of IL-17 to be similar despite elevated levels of known antagonist IL-2 [38]. Consistent with this, our previous analysis of cytokine expression in a population of adult subjects with CFS of unknown etiology [27] indicated an abnormally subdued IL-23/Th17/IL-17 response to elevated levels of known inducers IL-1b and IL-6 [39].

The above-mentioned discrepancies emphasize the importance of immune context. Indeed it was necessary to use the combined expression of 5 cytokines to support a robust separation of patients from recovered controls. Ultimately the context linking the expression of these cytokines arises from one or several basic immune processes acting in concert. Here the choice of cytokines in both the all-possible-subsets and stepwise regression models is interesting in that the majority of these are either directly or indirectly related to Th17 response [39]. In this analysis IL-6, 8 and 23 were selected by two different statistical approaches, with IL-1a, IL-2 and IFN-γ also selected in one model or the other. It is interesting to note that neither IL-1a, nor IL-6 or IFN-γ were differentially expressed across groups even though they contributed significantly to the classification of these subjects. IL-1, 6 and 23 are known inducers of Th17 response, which in turn is a producer of IFN-γ [39]. Conversely IL-2 is generally known as a Th17 antagonist [38]. Liu et al. (2007) [40] describe a mechanism by which IL-1 induces the production of IL-23 via NF-kappa B activation, which in turn promotes the production of IL-6 and 8 in human fibroblast-like synoviocytes from rheumatoid arthritis patients. Though this is still poorly understood, recent evidence in animal models supports a role for EBV and EBV-inducible genes in the pathogenic modulation of Th17 response [41, 42]. This may involve TLR9 which has been shown recently to play a significant role in the recognition of EBV by primary dendritic cells (DC), as indicated by a marked inhibitory effect on their synthesis of IFN-α, IL-6, and IL-8 [43]. Collectively the current analysis, as well as results from our previous work [27], suggests that illness-specific differences in the regulation of Th17 response may be a shared component in a significant subset of CFS cases.

When comparing this work with results from other studies it is important to remember the type of sample and the specific patient population studied. For example, recent work by Brenu et al. (2011) [44] demonstrated higher levels of IL-10, IFN-γ and TNF-α in CFS. However this was measured in mitogen-stimulated CD4+ cell cultures not in plasma. Similar in vitro protocols were used in earlier work by Amel Kashipaz et al. (2003) [45] and Skowera et al. (2004) [46]. Using the same sample type and laboratory protocol applied here, we recently reported higher levels of IL-5 and lower levels of IL-8 in CFS subjects compared to healthy controls, while levels of IL-23 showed little difference [27]. However both the work of Broderick et al. (2010) [27] and Brenu et al. (2011) [44] were conducted in a much broader CFS patient population where the illness trigger was not uniform. In addition the control population consisted of otherwise healthy individuals not individuals recovering from acute infection. Estimates of the number of patients with CFS who can date their illness to a specific (presumably infectious) acute illness range from 20- 90%, with the highest percentages being recorded in adolescent populations [8, 47–53]. Accordingly our current results align closely with those of another study conducted specifically in a post-infectious patient population by Vollmer-Conna et al. (2007) [28]. The authors found no differences in the concentrations of IL-1b, IL-2, IL-4, IL-6, IL-10, IL-12, TNF-α, and IFN-γ analyzed in serum. With the possible exception of elevated IL-2, we also did not find differences in the levels of these cytokines. Levels of IL-5, 8 and 23 were not measured in this earlier study.

Limitations of our current study include the fact that it reports on a small cohort. This is important when considering the general applicability of our statistical classification model in particular. The hope is that this subset of subjects is somewhat representative of a larger PI-CFS patient population. CFS is well known for its heterogeneous presentation; however it can be argued that much of this is the result of varied etiology and unknown triggers. Our results are derived from a prospective study and constitute an important first look at the molecular phenotyping of persistent fatigue in a cohort with a uniform and known infectious trigger; namely infectious mononucleosis. While this trigger does not explain all presentations of CFS, it does represent a significant subset of patients, in particular in the pediatric population. Moreover though specific to the EBV pathogen, elements presented here may also be of relevance to other forms of CFS with an infectious etiology. Constraints such as these further emphasize the importance of prospective studies and the need for much larger initiatives of this type in this and related patient populations.

In conclusion, the co-expression profile of 5 of 16 cytokines measured in the plasma of subjects recruited in a 2-year prospective study of mononucleosis-induced chronic fatigue syndrome readily discriminates cases from recovered control subjects and describes an atypical immune response in these individuals. In particular, irregular patterns of co-expression in IL-6, 8 and 23 were identified using two distinct statistical methodologies. These differences were especially evident when cast in the context of IL-1a, IL-2 and IFN-γ. This evidence points to a possible dysregulation of the innate system’s priming of the appropriate adaptive Th17 response in these subjects.