Exercise – induced changes in cerebrospinal fluid miRNAs in Gulf War Illness, Chronic Fatigue Syndrome and sedentary control subjects

Gulf War Illness (GWI) and Chronic Fatigue Syndrome (CFS) have similar profiles of pain, fatigue, cognitive dysfunction and exertional exhaustion. Post-exertional malaise suggests exercise alters central nervous system functions. Lumbar punctures were performed in GWI, CFS and control subjects after (i) overnight rest (nonexercise) or (ii) submaximal bicycle exercise. Exercise induced postural tachycardia in one third of GWI subjects (Stress Test Activated Reversible Tachycardia, START). The remainder were Stress Test Originated Phantom Perception (STOPP) subjects. MicroRNAs (miRNA) in cerebrospinal fluid were amplified by quantitative PCR. Levels were equivalent between nonexercise GWI (n = 22), CFS (n = 43) and control (n = 22) groups. After exercise, START (n = 22) had significantly lower miR-22-3p than control (n = 15) and STOPP (n = 42), but higher miR-9-3p than STOPP. All post-exercise groups had significantly reduced miR-328 and miR-608 compared to nonexercise groups; these may be markers of exercise effects on the brain. Six miRNAs were significantly elevated and 12 diminished in post-exercise START, STOPP and control compared to nonexercise groups. CFS had 12 diminished miRNAs after exercise. Despite symptom overlap of CFS, GWI and other illnesses in their differential diagnosis, exercise-induced miRNA patterns in cerebrospinal fluid indicated distinct mechanisms for post-exertional malaise in CFS and START and STOPP phenotypes of GWI.

. Overlap of diagnostic criteria for CFS 1 , GWI 7 and major depressive disorder 78 . Diagnostic protocols for CFS, GWI and depression selected different sets of primary and ancillary symptoms. CFS requires fatigue, then confirmation with ≥4 of 8 ancillary criteria. Active depression and other psychiatric diseases are exclusionary diagnoses for CFS. GWI requires 3 of 7 categories of symptoms. Depression requires depressed affect and anhedonia, then sufficient supporting complaints. personal pesticides 19 . Objective findings that distinguish GWI subjects from their unaffected deployed and nondeployed peers and civilians include cerebrospinal fluid proteomics 20 , low activity butyrylcholinesterase alleles in GWI cases who used pyridostigmine (odds ratio = 40.0) 21 , brain white matter dysfunction with increased axial diffusivity by diffusion tensor imaging 22,23 , and mitochondrial dysfunction 18,24 . A logical hypothesis is that Gulf War era exposures to cholinesterase inhibitors 19 caused acute acetylcholine neurotoxicity in persons with genetically reduced levels of acetylcholinesterase activity 21 , followed by chronic progression of the initial lesions 18,20,[22][23][24] .
We have reported that GWI veterans can be divided into two phenotypes based on responses to the physiological stressor of submaximal exercise testing 25 . One third of subjects developed new postural tachycardia after exercise and were positively selected as the START (Stress Test Activated Reversible Tachycardia) group. START subjects had: (i) exercise -induced postural tachycardia, (ii) increased blood oxygenation level dependent (BOLD) signal in the cerebellar vermis during a cognitive task before exercise, (iii) reduced BOLD signals during a working memory task after exercise, and (iv) reduced brainstem volumes suggesting atrophy. The remainder formed the Stress Test Originated Phantom Perception (STOPP) group because they had significantly greater BOLD activation of basal ganglia and anterior insula during cognitive testing than sedentary controls (SC) and START 25 . That pattern was similar to phantom limb pain 26 . The two phenotypes suggest there were two mechanisms of initial injury or on-going progression that will require different diagnostic and treatment approaches.
Cerebrospinal fluid was extensively assayed for micro-RNAs (miRNA), proteomics 20 , metabolomics, and other analytes to interrogate the central neurotoxic pathologies proposed in GWI 27 and CFS 16,17 . miRNAs are ~22 nucleotide long, single-stranded RNAs transcribed from genomic DNA 28 . They form the RNA-induced silencing complex (RISC) and bind to complementary sequences in the 3′ untranslated region of mRNAs to repress translation or promote mRNA degradation. miRNAs dynamically fine-tune the expression of most cellular proteins. Quantitative polymerase chain reaction (QPCR) was used to measure cerebrospinal fluid miRNAs.
First, we hypothesized that CFS (cfs0), GWI (gwi0) and sedentary control (sc0) subjects at rest (nonexercise, lower case italics with 0) would have significant differences in cerebrospinal fluid biomarkers from each other. The nonexercise groups rested overnight and had no exercise before their lumbar punctures. Nonexercise miRNA patterns were predicted to be different from other conditions in the differential diagnosis such as depression and fibromyalgia. Second, differences would be magnified in post-exercise SC, CFS, and the exercise-defined START and STOPP phenotypes of GWI subjects when compared to each other (upper case italics to denote post-exercise). Third, differences between the post-exercise groups and their appropriate nonexercise comparison groups (SC vs. sc0, CFS vs. cfs0, START vs. gwi0, STOPP vs. gwi0) would model the effects of exercise on the central nervous system and the pathology of exertional exhaustion.  32 , and Center for Epidemiologic Studies Depression Scale 33 scores. Fibromyalgia was assessed by pain plus tenderness (1990 criteria) 34 . Clinical and methodological details were published previously 25 . All subjects had submaximal bicycle exercise stress tests on 2 consecutive days with magnetic resonance imaging before and afterwards, followed by a lumbar puncture 25 . Subjects cycled at 70% of age predicted maximum heart rate for 25 min 35 followed by stepwise increases in bicycle resistance to reach 85% predicted heart rate 25 . Exercise was required to induce postural tachycardia that defined the START phenotype. Cerebrospinal fluid total protein, albumin and IgG were measured 36 and aliquots were frozen at −80 °C until thawed for miRNA extraction.

Methods
Quantitative PCR. miRNA analysis was completed in blinded fashion by N.S. without knowledge of subject diagnosis. Total RNA was isolated by mixing 0.5 ml cerebrospinal fluid 1:10 with QIAzol ™ lysis reagent (Qiagen) and 0.1 ml CHCl 3 before vortexing for 1 min [37][38][39] . miRNA was extracted from the upper phase using miRNAeasy Mini Kits (Qiagen). miRNA levels were estimated by optical density of cDNA after reverse transcription with miScriptII RT kits (Qiagen). miRNA expression profiling used miScript PCR arrays for 380 miRNAs and miScript SYBRgreen PCR kits (Qiagen) on an ABI 7900 HT Real-Time PCR system (Applied Biosystems) and manufacturer's protocol. miRNA selection. The first level of constraint required miRNAs to be detectable with PCR cycle threshold (Ct) ≤ 35. miRNAs with Ct > 35 were designated as "undetectable" and 35 was imputed as their Ct. Second, in order to be considered a viable biomarker, miRNA had to detectable with Ct ≤ 35 in at least two thirds of subjects in a group. miRNA normalization and ΔΔCt. Four normalization strategies and ΔΔCt computations were compared.
The N0 (no normalizer) analysis used the entire dataset with 35 imputed for all Ct > 35. The average ΔCt was calculated for each group, then ΔΔCt determined between each pair of groups 40 . Significantly different miRNAs were detected by one-way ANOVA followed by Tukey's Honest Significant Difference (HSD; p < 0.05). Student's 2-tailed unpaired t-tests were computed for all miRNAs and pairs of groups, and False Discovery Rates (FDR) calculated to correct for multiple comparisons 41 . FDR ≤ 0.10 was used as the next constraint to detect significant differences in ΔΔCt. The other normalizers used 2 (N2), 3 (N3) and 6 (N6) miRNAs. For each individual, ΔCt was calculated as the difference of the N2, N3 or N6 normalizer minus the Ct for each of the other miRNAs. Average ΔCt and ΔΔCt were calculated. Outcomes were: (i) differences between non-exercise groups (sc0, cfs0 and gwi0), (ii) differences between groups after exercise (SC, START, STOPP, CFS), and (iii) exercise-induced differences between each post-exercise group and its appropriate non-exercise control group. ΔΔCt data were reported as mean ± SD to allow calculation of Cohen's d (mean difference/SD pooled ) and to predict future sample sizes 42 . Receiver operating characteristics were calculated for each significant ΔΔCt in SPSS 24.

Results
Demographics. CFS groups (cfs0, CFS) had more females, and GWI groups (gwi0, START, STOPP) more males ( Table 2). Quality of life 31 , fatigue, cognitive, sleep, pain and interoceptive symptoms 29,30 were significantly impaired in GWI and CFS groups compared to sedentary control subjects. Fibromyalgia (1990 criteria) was more prevalent in CFS and GWI than controls 34 . The case designation criteria of GWI, CFS, generalized anxiety and depression share fatigue, sleep, cognition, and sympathetic nervous system symptoms 32 . The Center for Epidemiology -Depression questionnaire found depression in 78.3% of GWI, 64.0% of CFS and 25.0% of control subjects 33 . Generalized Anxiety Disorder 7 scores were significantly higher for gwi0 and START compared to sc0 (FDR < 0.01) and SC (FDR < 0.01), respectively 33 . Cerebrospinal fluid total protein, albumin, IgG and their ratios were equivalent between groups 36 .
Normalizers. The raw data were placed in Supplementary Table S1. The statistical constraints reduced the number of miRNAs that were candidates for biomarkers and normalizers down to 88 (Supplementary Table S2).
The N0 normalizer selected 31 miRNAs that met the significance criteria of (i) being detected with Ct ≤ 35 in more than two thirds of subjects in at least 1 of the 7 groups, (ii) HSD ≤ 0.05, and (iii) FDR ≤ 0.10. N2 identified 21, N3 had 24, and N6 found 23 significant miRNAs. The intersection of the 4 normalizers identified 18 miRNAs with at least 1 significant difference between groups (Fig. 5). One was added by the intersection of 3 normalizers. There was excellent agreement for the magnitudes of ΔΔCt between the 4 normalizers (Supplementary Tables S4  and S5). In contrast, N0 selected 9 additional miRNAs that were not found with N2, N3 and N6 (Fig. 5). These were considered false positive results.
Nonexercise groups. None of the miRNAs were significantly different between nonexercise groups using our stringent criteria. miR-22-3p ΔCt values were higher in cfs0 than sc0, but the differences were not significant after ANOVA and Tukey tests.
Post-exercise groups. miR-22-3p and miR-9-3p were the only miRNAs to be significantly different between the post-exercise groups (Fig. 6). miR-22-3p was an anomaly by having wide ranges of Ct in all groups. miR-22-3p was virtually not detectable in START, and so levels in START were significantly diminished compared to SC and STOPP (Supplementary Table S6). The reduction in START, but relative increase in STOPP, supported the presence of 2 phenotypes of GWI. Specificities and sensitivities were 0.76 for START versus SC (Ct threshold of 29) and START versus STOPP (Ct threshold of 33) ( Table 2).
miR-9-3p demonstrated a different trend. Only the START group had detectable levels (Ct ≤ 35) in more than two thirds of subjects. The difference between START and STOPP was small (Fig. 6, Supplementary Table S6) but significant (HSD < 0.05, FDR < 0.05). The low specificity and sensitivity of 65% at a threshold of Ct = 33 reflected the low levels of miR-9-3p in cerebrospinal fluid. miRNAs elevated after exercise compared to nonexercise groups. Exercise elevated the levels of several miRNAs compared to appropriate nonexercise groups. SC had higher levels than sc0 for miR-22-3p (Fig. 6), miR-30d-5p, miR-204-5p, miR-425-3p, and miR-99b-5p (Fig. 7). Specificities and sensitivities for miR-204-5p and miR-22-3p were 0.80 at thresholds of 31 and 32, respectively. START had elevated levels of miR-425-3p and miR-99b-5p when compared to gwi0. miR-370 was detected in almost all cerebrospinal fluid samples, but only START had a significant elevation compared to nonexercise (ΔΔCt = 1.7 ± 2.1 versus gwi0, mean ± SD). STOPP shared the exercise -induced elevation of miR-99-5p with SC and START. Specificities and sensitivities for miR-99b-5p were about 0.75 at Ct thresholds of 33 for SC, START and STOPP compared to their nonexercise controls.
CFS did not have any elevations of miRNA levels compared to cfs0. miRNAs diminished after exercise compared to nonexercise groups. miR-328 and miR-608 were significantly diminished by exercise in SC, CFS, START and STOPP compared to the nonexercise sc0, cfs0, and gwi0 groups (Fig. 8). These miRNAs were detectable in almost all cerebrospinal fluid specimens in this study.   . Diminished miR-328 and miR-608 may be a consequence of exercise that affected all subjects regardless of their disease status. miR-let-7i-5p, miR-200a-5p and miR-93-3p were significantly reduced in START, STOPP and CFS compared to their gwi0 and cfs0 nonexercise controls (Fig. 9). They were unchanged between SC and sc0 groups.
Gender. Cerebrospinal fluid miRNA levels for females and males in the nonexercise and the post-exercise groups were equivalent except for a significantly higher level of miR-9-3p in START than STOPP males (ΔΔCt = 1.7 ± 1.4) 43 .
These differences were consistent with the overall group effects

Discussion
This is the first description of the effects of exercise on cerebrospinal fluid miRNA expression in healthy subjects. Exercise diminished miR-328 and miR-608 in all subjects suggesting a general effect on the brain (Fig. 8, Supplementary Table S5). Exercise caused distinct patterns of miRNA changes in CFS and the START and STOPP phenotypes of GWI indicating significant pathophysiological differences between conditions. Unlike our starting hypothesis, there were no differences in miRNA levels between the nonexercise groups of control, CFS and GWI subjects. Therefore, baseline levels of cerebrospinal fluid miRNAs may not be useful for diagnosis of CFS or GWI.
The only significant differences between groups after exercise were diminished miR-22-3p in START compared to SC and STOPP, and elevated miR-9-3p in START compared to STOPP (Fig. 6). These differences between START and STOPP support our 2 phenotypes of GWI 25 .
The most striking findings were the changes between post-exercise groups and their appropriate nonexercise comparison groups. SC had 5 elevated miRNAs after exercise, compared to 3 for START, 1 for STOPP, and none in CFS (Fig. 7, Supplementary Table S4).
The reduction of miR-608 after exercise has implications for the cholinergic hypothesis of GWI pathophysiology because it targets acetylcholinesterase and interleukin-6 (IL6) mRNAs. miR-608 binds weakly to the single-nucleotide polymorphism rs17228616 allele in the 3′-untranslated region of acetylcholinesterase mRNA 44 . Homozygotes for rs17228616 have reduced affinity for miR-608. This promotes mRNA stability and increases acetylcholinesterase protein translation. As a consequence, more miR-608 is available to bind to IL6 mRNA and reduces its translation. This allele also contributes to reduced cortisol and elevated blood pressure. Because rs17228616 promotes higher acetylcholinestase activity, it may be relatively protective against nerve agent and pyridostimine bromide exposure.
miR-let-7i-5p, miR-93-3p and miR-200a-5p were significantly diminished after exercise in START, STOPP and CFS, but not SC (Fig. 9). This was consistent with a cardinal finding in CFS and GWI: function may appear normal when rested, but will deteriorate after a physiological stressor 1-4,7,25,45,46 . miR-let-7i was reduced in plasma after exercise in athletes 47 . miR-let-7i has decreased expression in the prefrontal cortex of FSL rats in a model of Figure 10. Decreased miRNAs after exercise in CFS group. CFS had significant reductions in -(a) 92a-3p, (b) miR-323b-5p, (c) miR-126-5p, (d) miR-19b-3p, (e) miR-505-3p, (f) miR-532-5p, and (g) miR-186-3p, compared to its nonexercise cfs0 comparison group (bars above the groups). depression 48 . IL6 is a target of miR-let-7i, and, as predicted, this cytokine was significantly elevated in the brains of these rats. When FSL rats were given access to running wheels, their miR-let-7i expression was increased and IL6 reduced. Modulation of miR-let-7i and IL-6 may contribute to exercise-induced benefits in "inflammatory" depression 48 . miR-let-7i also contributes to the regulation of acetylcholine's muscarinic and α4β2 nicotinic receptors and epigenetic regulation of acetylcholinesterase. These animal models may not be appropriate for CFS or GWI because human subjects develop exertional exhaustion after exercise, and are unlikely to significantly increase spontaneous exercise levels when provided with a treadmill 45 .
The CFS group had 12 miRNAs reduced after exercise. miR-186-3p was decreased in aging mice 49 where it targets β-site amyloid precursor protein-cleaving enzyme 1 (BACE1) mRNA. Diminished miR-186-3p allows increased BACE1 mRNA translation and cleavage of amyloid peptides that increase the risk for brain disease. miR-19b-3p was reduced in serum from Alzheimer's patients, and targeted signal transduction and activator of transcription 3 (STAT3) mRNA in a murine model 50 . miR-92a-3p was increased in glioblastoma and targeted BCL2L11 to reduce tumor apoptosis 51 . Its reduction after exercise may promote apoptosis in CFS. miR-126-5p was highly expressed in endothelial cells where it targets vascular (VCAM), intercellular (ICAM) and activated leukocyte (ALCAM) cell adhesion molecule mRNAs and so reduces transendothelial migration 52,53 . This is relevant for immune cell influx into the brain, and hypotheses of neuroinflammation in CFS pathogenesis.
The choroid plexus epithelium may be a primary source of miRNAs in cerebrospinal fluid [55][56][57] . Epithelial cells form a monolayer linked by tight junctions that creates the "shrink wrapped" cellular barrier around fenestrated capillaries 58 . Interferon-gamma and other mediators generated by exercise, inflammation, and other stressors act directly on choroid plexus to modulate barrier permeability, plasma protein transport, protein synthesis and secretion of nutrients into cerebrospinal fluid [59][60][61] . miR-328, which was present in all subjects and reduced after exercise (Fig. 8), binds to the 3′-untranslated regions of CD44 and collagen type 1α1 mRNAs to modulate extracellular barrier functions 62 . Choroid plexus miRNAs 55 are packaged into extracellular vesicles and released into cerebrospinal fluid 56,57,63,64 . Downstream targets include subventricular neural stem cells, mature neurons, astrocytes, oligodendrocytes, microglia, meningeal and central immune cells 56,57,63,[65][66][67] . Blockade of extracellular vesicle secretion from choroid plexus cells decreased brain inflammation in a mouse model of lipopolysaccharide-induced inflammation 56 . Choroid plexus miRNAs may be novel drug targets to modulate acute illness behaviours, fever, and chronic pain in systemic illnesses.
Choroid plexus is dysfunctional in Alzheimer's disease 68 . This provides the rationale to consider the role of the blood -cerebrospinal fluid barrier in the cognitive dysfunction of CFS and GWI. There are numerous reports of elevated and diminished miRNAs in cerebrospinal fluid in Alzheimer's disease 36,[69][70][71][72][73][74] , but none matched the patterns of our groups. miR-let-7i-5p 36 was elevated in Alzheimer's, but levels were equivalent in nonexercise groups (Fig. 8).
Depression is in the differential diagnosis because of the shared ancillary diagnostic criteria (Fig. 1) [75][76][77][78] . Major depressive disorder is defined by affective dysfunction with sadness, flat affect and anhedonia as essential features, followed by secondary criteria including fatigue, cognitive, sleep, and somatic dysfunction 78 . However, screening questionnaires for depression emphasize somatic symptoms 79,80 . Complaints of fatigue, sleep, and cognitive dysfunction will inflate total questionnaire scores, and may lead to false positive inference of major depressive disorder even if anhedonia or affective complaints are absent [81][82][83] . This is particularly problematic in CFS and GWI where these features are diagnostic criteria (Fig. 1). As a result, Center for Epidemiology -Depression (CESD) 33 scores were significantly elevated in GWI (78.3%), CFS (64.0%) and control (25.0%) groups (Table 1).
Quantitative PCR of miRNAs offers a more objective solution 84 . miR-16 in cerebrospinal fluid was significantly lower in major depressive disorder patients than control subjects 85 . However, this was not confirmed in an independent group who had a different pattern of 11 significantly elevated and 5 reduced miRNAs 86 . Our data did not confirm either of these findings because only 3 of the miRNAs were detected with Ct ≤ 35 in more than two thirds of our nonexercise group. miR-425-3p was significantly reduced in depression patients [84][85][86] , and was detected in about half of all nonexercise subjects. It was increased after exercise in SC, START and STOPP but not CFS (Fig. 7). This lack of reproducibility highlights the need to independently verify miRNA findings, and supports our rationale for strict statistical criteria to define potential miRNA biomarkers.
The pain and tenderness of GWI subjects (Table 1) indicated systemic hyperalgesia and suggested parallels with fibromyalgia 34 . Nine miRNAs were virtually undetectable in 10 fibromyalgia women compared to 8 healthy control women 87 . miR-99b-5p and miR-29a-3p were absent in fibromyalgia, but were detected in more than two thirds of our participants. miR-99b-5p was significantly increased after exercise in SC, START and STOPP (Fig. 7). The other 7 miRNAs were detected in less than half of our specimens. This suggested that GWI and CFS were distinct from fibromyalgia.
Limitations to the diagnostic use of quantitative miRNA analysis in cerebrospinal fluid include the remarkable lack of consensus about miRNA levels in control subjects. This can be remedied by standardization of reagents and protocols 72 , open source sharing of study outcomes, and meta-analysis of the raw data. The yield of extracted miRNA 88 and detectability were improved with 0.5 ml instead of 0.2 ml of cerebrospinal fluid 36,87 . QPCR with Ct cut-offs ≤ 35 cycles reduced amplification artifacts 73,87 . The wide range of miR-22-3p Ct values (Fig. 6) may be due to commercial changes in reagents designed to improve miRNA detection. Highly abundant miRNAs that were detected with Ct < 35 in all subjects were used as normalizers (Fig. 4) rather than the global average level, miR-423-5p, miR-124-3p or U6 36,71,73,86,89 . Constraints included (i) significant ANOVA and Tukey HSD between groups, (ii) significant FDR to correct for multiple comparisons, and (iii) focusing on miRNAs that were detected in more than two thirds of subjects per group that may be viable biomarker candidates for use in the general population 36,71,86 . Ages were comparable between groups 74 (Table 1) and there were no differences in expression between males and females 43,87 . Next generation sequencing is an excellent discovery tool but needs careful internal standardization to be as sensitive as QPCR for quantification [90][91][92] . Adequate sample sizes were essential because our initial findings with about a dozen subjects per group showed differences between START and STOPP after exercise 93 , but these differences eventually regressed to the mean as more subjects were analyzed. This is especially pertinent to smaller studies examining the differential diagnosis of CFS and GWI [85][86][87]94,95 .
Limitations of the testing paradigm include the intensive nature of the exercise and magnetic resonance imaging characterization of GWI subjects to determine their phenotypes. Lumbar puncture was required to obtain the cerebrospinal fluid miRNA biomarkers, but this procedure is not a contraindication to making an objective diagnosis of GWI. On the contrary, magnetic resonance imaging with cerebrospinal fluid QPCR miRNA profiling may be complementary tools for diagnosis of CFS, GWI and their subtypes.

Conclusions
Cerebrospinal fluid miRNA levels were equivalent between SC, CFS and GWI subjects who had rested before exercise (nonexercise groups). miRNA levels were different from the ones that are altered in depression, fibromyalgia, and Alzheimer's disease suggesting that these are all distinct diseases, or that the data from those smaller studies could not be replicated in this larger study. miRNA levels were equivalent between the post-exercise SC, CFS and GWI phenotypes of START and STOPP with the exception of miR-22-3p and miR-9-3p that significantly distinguished START from STOPP. This adds another layer of evidence to support neurotoxic pathology in GWI 27 and these 2 phenotypes of GWI veterans 25 . Post-exercise levels were significantly elevated (n = 6) or diminished (n = 12) compared to the nonexercise comparison groups. miR-328 and miR-608 were elevated in SC, CFS, START and STOPP and may be a global marker of the exercise stressor on the choroid plexus and brain. CFS had 12 diminished and zero elevated miRNAs after exercise indicating its pathophysiology and responses to exercise were unique compared to GWI and control subjects. Despite the symptom overlap of CFS, GWI and other illnesses in the differential diagnosis ( Fig. 1), the distinct exercise-induced miRNA patterns in cerebrospinal fluid imply separate mechanisms for post-exertional malaise in these diseases.