Multi-omics examination of Q fever fatigue syndrome identifies similarities with chronic fatigue syndrome

Background Q fever fatigue syndrome (QFS) is characterised by a state of prolonged fatigue that is seen in 20% of acute Q fever infections and has major health-related consequences. The molecular mechanisms underlying QFS are largely unclear. In order to better understand its pathogenesis, we applied a multi-omics approach to study the patterns of the gut microbiome, blood metabolome, and inflammatory proteome of QFS patients, and compared these with those of chronic fatigue syndrome (CFS) patients and healthy controls (HC). Methods The study population consisted of 31 QFS patients, 50 CFS patients, and 72 HC. All subjects were matched for age, gender, and general geographical region (South-East part of the Netherlands). The gut microbiome composition was assessed by Metagenomic sequencing using the Illumina HiSeq platform. A total of 92 circulating inflammatory markers were measured using Proximity Extension Essay and 1607 metabolic features were assessed with a high-throughput non-targeted metabolomics approach. Results Inflammatory markers, including 4E-BP1 (P = 9.60–16 and 1.41–7) and MMP-1 (P = 7.09–9 and 3.51–9), are significantly more expressed in both QFS and CFS patients compared to HC. Blood metabolite profiles show significant differences when comparing QFS (319 metabolites) and CFS (441 metabolites) patients to HC, and are significantly enriched in pathways like sphingolipid (P = 0.0256 and 0.0033) metabolism. When comparing QFS to CFS patients, almost no significant differences in metabolome were found. Comparison of microbiome taxonomy of QFS and CFS patients with that of HC, shows both in- and decreases in abundancies in Bacteroidetes (with emphasis on Bacteroides and Alistiples spp.), and Firmicutes and Actinobacteria (with emphasis on Ruminococcus and Bifidobacterium spp.). When we compare QFS patients to CFS patients, there is a striking resemblance and hardly any significant differences in microbiome taxonomy are found. Conclusions We show that QFS and CFS patients are similar across three different omics layers and 4E-BP1 and MMP-1 have the potential to distinguish QFS and CFS patients from HC.

Background Q fever fatigue syndrome (QFS) is characterised as a state of prolonged fatigue following acute Q fever infection (1). The fatigue lasts for at least six months and is usually associated with musculoskeletal complaints, neurocognitive problems, sleeping problems, headache, respiratory tract problems and mood disorders (1). QFS was rst described by Shannon et al. in 1993 and occurs worldwide (2,3). In many ways, complaints of QFS are similar to those of chronic fatigue syndrome (CFS) (4,5), and the pathogenesis of these syndromes remains largely unclear. An important distinction between QFS and CFS is the fact that QFS has a known aetiology, being precipitated by an acute Q fever infection, and is therefore considered to be a postinfectious fatigue syndrome.
Symptomatic infection with Coxiella burnetii is called acute Q fever and constitutes around 40% of human primary Coxiella infections: the other 60% remain asymptomatic (6)(7)(8). Acute Q fever often is a self-limiting disease that usually presents as a u-like illness that may be accompanied by pneumonia or hepatitis (7). Of all those who become infected with C. burnetii, both symptomatic and asymptomatic, around 1-5% develop a persistent infection with the bacterium, called chronic Q fever or persistent focalised infection (9). Of all those who develop acute Q fever, i.e., symptomatic, infection, around 20% will develop QFS (1).
Like CFS (10), some studies of QFS suggest that there is a low-grade in ammatory component. First reports supporting this notion came from Pentilla et al. in 1998, who showed that peripheral blood mononuclear cells (PBMCs) of QFS patients produce more IL-6 when stimulated with Q fever antigen than controls (11). During the Dutch Q fever outbreak (2009 -2012), our group demonstrated that QFS patients exhibit signs of altered immunity through the monocyte-derived cytokines Tumor Necrosis Factor (TNF)α, interleukin (IL)-1β, and especially IL-6, together with the interferon (IFN)γ-axis (12)(13)(14). In addition, we found that monocytes of both QFS patients and CFS patients show decreased expression of mitochondrial derived peptide (MDP)-coding genes MT-RNR1 and MT-RNR2, resulting in a decreased production of humanin (MT-RNR2) (15).
To better understand the various molecular aspects of QFS aetiology and its place in the chronic fatigue syndrome spectrum, we applied a multi-omics approach and investigated the in ammatory proteome, metabolome, and composition of gut microbiome of QFS patients, CFS patients, and healthy controls (HC), matched for age, gender, use of medication, and general geographic region.

Study population
The study population consisted of QFS patients (n = 31), CFS patients (n =50), and HC (n = 72). All subjects were matched for age (± 10 years), gender (non-pregnant females), and general geographical region (south-eastern part of the Netherlands). QFS patients were actively recruited for this study while CFS patients and HC participated in previous studies. As all CFS patients included in this study are nonpregnant females, so were recruited QFS patients and selected HC. All subjects were between 18 and 59 years of age and did not use medication, other than paracetamol or oral contraceptives, and were not vaccinated, in the last 6 months. Both QFS and CFS patients were diagnosed according to similar guidelines as described previously (27,28) with the only difference on whether complaints were precipitated by an acute Q fever infection or not.
All QFS patients were diagnosed at the Radboud Expert Center for Q fever, Nijmegen, the Netherlands, after a uniform work-up according to the Dutch guideline on QFS diagnosis (29). All QFS patients met the following diagnostic criteria: i. fatigue that lasted ≥ 6 months; ii. sudden onset of severe fatigue (de ned as a score ≥ 35 on the subscale fatigue severity of the Checklist Individual Strength (CIS) questionnaire), or signi cant increase in fatigue, both related to a symptomatic acute Q fever infection; iii. chronic Q fever and other somatic or psychiatric causes of fatigue were excluded; and iv. fatigue resulted in signi cant functional impairment (de ned as a total score ≥ 450 on the Sickness Impact Pro le-8 (SIP-8) questionnaire). All QFS patients were fatigued for less than 10 years.
All CFS patients were diagnosed with CFS at the Department of Internal Medicine and Expert Center for Chronic Fatigue (ECCF) of the Radboud University Medical Center, Nijmegen, the Netherlands, after a uniform work-up according to the Centers for Disease Control (CDC) criteria for CFS. All CFS patients participated in a randomized trial on cytokine inhibition in CFS and samples were collected at baseline, prior to the intervention (30). All CFS patients had a score ≥ 35 on the subscale fatigue severity of the CIS questionnaire and a score ≥ 450 on the SIP-8 questionnaire. All CFS patients were fatigued for less than 10 years.
All HC reported having no complaints of fatigue and participated in the Human Functional Genomics Project (www.humanfunctionalgenomics.org) (31). Samples were collected at the Radboud University Medical Center, Nijmegen, the Netherlands. All HC were selected based on age, gender and general geographical region that matched with both QFS and CFS patient groups. For proteomics and metagenomics, a larger HC group (n = 50) was used than for metabolomics data (n = 22). For evaluating the correlation patterns between metabolome and in ammatory proteome, an independent populationbased cohort from the same general geographical region (n = 318) was used for comparison.

Sample handling and omics measurements
Faecal and plasma samples were collected in 2016 and 2017 as previously described (30,32). Venous blood was collected in EDTA tubes, and kept on ice until centrifugation, which was performed within 2 -3 hours. Next, samples were centrifuged at 2960xG for 10 min at 4 °C. Plasma aliquots were then frozen at − 80 °C for a maximal duration of 2 years. Metabolome and in ammatory proteome analyses for all patients and controls were run at the same time. Protein levels were measured in 131 participants and determined using Olink in ammation panel, including 92 in ammation-related protein biomarkers (https://www.olink.com/). During the quality control step, in ammatory markers with > 20% of measurements below the detection limit were excluded for further analysis, leaving 65 proteins in total.
Serum metabolite levels were measured in General Metabolomics platforms (https://generalmetabolics.com) for all 131 individuals. Metabolome was measured and annotated by General Metabolomics (Boston, MA) using ow injection-time-of-ight mass ( ow-injection TOF-M) spectrometry. Faecal samples were collected within 24 hours before processing and cooled at 4 °C before processing. Samples were aliquoted and then frozen at − 80 °C for a maximal duration of 2 years. The gut metagenomic sequencing was performed at Novogene, China, using the Illumina HiSeq platform. The metagenome pro ling was measured according to a previously described protocol (33). KneadData tools (v0.5.1) (34) were used to remove the adapters, trim the sequencing reads to PHRED quality 30, and remove reads aligned to human genome (GRCh37/hg19).
PBMC stimulation and cytokine assay PBMC isolation was performed by dilution of blood in PBS (1:1) and fractions were separated by density centrifugation over Ficoll-Paque (Ficoll-Paque Plus; GE healthcare). Cells were washed three times with cold PBS and resuspended in Roswell Park Memorial Institute (RPMI) 1640 Dutch modi cation culture medium (Life Technologies/ Invitrogen) supplemented with 50 mg/mL gentamicin, 2 mM Glutamax TM , and 1 mM pyruvate (Life Technologies). PBMCs were then plated in 96-well round-bottom plates (Corning) at a concentration of 5 × 10 5 /mL in a total volume of 200 µL. Cells were exposed to RPMI, as a negative control, 0.5mM l-cysteine, and 25mM l-cysteinylglycine for 24 hours at 37°C with 5% CO 2 . After stimulation, supernatants were collected and MCP-1 and TGF-β were measured using enzyme-linked immune sorbent assay (ELISA) according to the manufacturer's protocol (R&D Systems).

Statistical analysis
Patient characteristics data were analysed using Graphpad Prism (Graphpad Software Inc., version 5.03). ANOVA was used to determine differences between groups. For the correlation analyses, Spearman's Rank-Order correlation coe cients were used followed by hierarchical clustering. R package 'corrplot' was used for visualization. Cytokine production data were analysed using the Mann-Whitney U test in GraphPad Prism (Graphpad Software Inc., version 5.03). The differential proteome and metabolome analyses were conducted using robust linear regression (39) with age effect corrected. For prediction model, Least Absolute Shrinkage and Selection Operator (LASSO) model (40) was utilized. Repeated Cross validation (CV) approach was used for building prediction models: 2/3 of samples were randomly selected for training while the rest of samples were used for prediction. The procedure was repeated 1000 times, and the Area under the curve (AUC) was calculated to evaluate the predictive power of the model (Supplementary gure 1). The metabolic pathway enrichment analysis was performed online by using MetaboAnalyst 4.0 (41). A principal coordinates analysis (PCoA) was performed on gut microbiome taxonomy. All statistical analyses were performed using the computing environment R (version 3.5.3). Statistical signi cance was obtained if P ≤ 0.05. To account for multiple testing, we assessed signi cance using Benjamini-Hochberg false discovery rate (FDR < 0.05).

Ethical statement
All participants provided written informed consent and the study, including studies from which CFS patients and HC were protracted (30,32), was approved by the Medical Ethical Review Committee of the Arnhem-Nijmegen region.

Subject characteristics
All QFS and CFS patients were severely fatigued and functionally impaired at the time of sample collection. Mean fatigue severity scores were signi cantly higher for CFS patients compared to QFS patients (Student's t test, P = 0.0034). No signi cant differences in mean functional impairment scores were observed when comparing QFS patients with CFS patients (Student's t test, P = 0.3055) ( Table 1) Differential association patterns between in ammatory protein and metabolites in disease and health There are 319, 441, and 12 signi cantly in-and decreased metabolites when comparing QFS patients to HC, CFS patients to HC, and QFS patients to CFS patients, respectively (FDR < 0.05, Figure 2, Supplementary gure 5, and Supplementary table 1). When comparing QFS to HC, the identi ed metabolites are enriched in primary bile acid biosynthesis (P = 0.0116), sphingolipid metabolism (P = 0.0256), nitrogen metabolism (P = 0.0394), and D-glutamine and D-glutamate metabolism (P = 0.0394) pathways. When comparing CFS to HC, the sphingolipid metabolism (P = 0.0033) pathway is enriched. When comparing QFS patients to CFS patients, the nitrogen (P = 0.0154), D-Glutamine and D-Glutamate metabolism (P = 0.0154), arginine (P = 0.0357), butanoate (P = 0.387), and histidine metabolism (P = 0.0407) are enriched.
Next, we investigated in which way the in ammatory proteins are associated with the metabolites in patients and healthy individuals, respectively. We illustrate the correlation between the differentially expressed proteins (FDR < 0.05) and the top 20 differentially expressed metabolites (with similar number of proteins) in QFS + CFS patients versus HC. This clustering pattern was then used as a reference for the same type of data from a population-based cohort of 318 individuals (www.humanfunctionalgenomics.org) ( Figure 3). As shown in Figure 3, metabolites acetohexamide, sphingosine 1-phosphate, l-cysteinylglycine, l-cysteine, and 2-(2,4-dihydroxy-5-m are of particular interest as they positively correlate with in ammatory proteins. Validation experiments with PBMCs of HC showed that stimulation with 25 mM l-cysteinylglycine resulted in a signi cantly higher MCP-1 production compared to RPMI as a negative control (Mann-Whitney U test, P = 0.0238). No signi cant differences were observed for TGF-β, or MCP-1 when stimulating with lower concentrations, i.e., 0.5 mM and 5 mM, of l-cysteinylglycine, l-cysteine, or acetohexamide ( Figure 4).

QFS and CFS show a microbiome composition distinct from HC
A PCoA on gut microbiome taxonomy of QFS, CFS, and HC was performed, showing a clear-cut difference between QFS and CFS, and HC ( Figure 5). There are 36, 44, and 2 features showing signi cant differences in gut microbiome taxonomy when comparing QFS to HC, CFS to HC, and QFS to CFS, respectively (Supplementary gure 3, Supplementary gure 5, and Supplementary table 2). When comparing QFS patients to HC there is an increase in abundance of Bacteroidetes with Bacteroides and Alistiples spp., and a decrease in abundance of Firmicutes and Actinobacteria with Ruminococcus and Bi dobacterium spp., respectively. When comparing CFS patients to HC, we nd an increase in abundance of Firmicutes and Actinobacteria with Ruminococcus and Bi dobacterium spp., respectively, and a decrease in abundance of Bacteroidetes with Alistiples and Bacteroides spp. When comparing QFS patients to CFS patients, we nd a slight increase in abundance of Firmicutes with Eubacterium and Faecalibacterium spp. in the former. Supplementary table 3 depicts signi cantly in-and decreased gut microbiome functional pathways when comparing QFS to HC, CFS to HC, and QFS to CFS.
Next, we investigated in which way the gut microbiome is associated with metabolites in fatigued patients, i.e., QFS and CFS, as HC hardly show any overlap. Only two signi cant correlations were found; Bi dobacterium_adolescentis and N-Docosahexaenoyl GABA, and Subdoligranulum_unclassi ed and Arbekacin (Supplementary gure 4).

Discussion
This study showed that in ammatory and metabolomic pro les, together with gut microbiome taxonomy, of QFS patients and CFS patients are quite similar, and both groups clearly differ from HC (with CFS patients showing a larger difference than QFS patients). These ndings are important, as they indicate that QFS patients and CFS patients show a common denominator in the long term, i.e., alterations in in ammatory and metabolomic pro les, together with gut microbiome taxonomy, regardless of the precipitating event that started the complaints.
Although important characteristics such as blood in ammatory pro le, gut microbiome and blood metabolome are very similar in QFS and CFS, subtle differences are still observed. It was previously shown that QFS patients tend to exhibit more of an in ammatory pro le than CFS patients (5,12). A similar trend is observed in our study. One could speculate that the microbial origin of QFS plays a role in this subtle persistent in ammation. Together with previous ndings on differences in fatigueperpetuating factors and response to cognitive behavioural therapy (CBT) (42)(43)(44), one could advocate that QFS should be seen as a separate, more in ammatory, fatigue syndrome entity that requires a different diagnostic (27,28) and therapeutic (44,45) approach. These ndings argue for a 'splitting' rather than a 'lumping' approach to chronic fatigue (46).
In ammatory markers 4E-BP1, AXIN1, and MMP-1 showed the potential to differentiate both QFS and CFS patients from HC and might therefore be associated with fatigue in general as this is the common denominator between these groups. We further elaborated on these ndings by using a machine-learning approach showing that both 4E-BP1 and AXIN1 are good candidate biomarkers for predicting/diagnosing chronic fatigue. The eukaryotic translation initiation factor 4E binding protein 1 (4E-BP1) represses mRNA translation downstream of the mammalian target of rapamycin (mTOR). The latter is known to phosphorylate and inactivate 4E-BP1 (47). Several upstream stimuli, e.g., growth factors and cytokines, can regulate downstream processes, e.g., cell growth, cell proliferation, and cell plasticity, through mTOR (47). Dennis et al. showed that the 4E-BP1 phosphorylation was inhibited when intracellular adenosine triphosphate (ATP) levels were lowered (48). Interestingly, chronic fatigue has previously been associated with a decrease in cell metabolism (15,18,49,50), and PBMCs of CFS patients showed a decrease in mitochondrial function compared to PBMCs of HC when stressed (51)(52)(53). Axis inhibition protein (AXIN1), negatively regulates the Wnt signalling pathway by downregulation of β-catenin (54), but has also been identi ed as a scaffold protein that activates TGF-β signalling (55). Especially the latter nding is of interest as elevated levels of TGF-β have frequently been associated with CFS (10). However, it should be noted that results on TGF-β levels must be interpreted with great caution as measuring TGF-β in plasma has some noteworthy, pre-analytic, pitfalls (56). Matrix metalloproteinase 1 (MMP-1) is a collagen cleaving protease that has been associated with in ammation in infections such as HIV (57,58), but has also shown to have a negative association with the risk of being a CFS patient (59). Exactly how, and how strong, 4E-BP1, AXIN1, and MMP-1 relate to chronic fatigue warrants further investigation in independent cohorts.
Comparing CFS patients to HC, studies on metabolomic pro les consistently found differences between these groups (17)(18)(19)(20). Armstrong et al. found that CFS patients show lower levels of glutamine and ornithine compared to HC (20). Germain et al. found pathway abnormalities in taurine, glycerophospholipid, primary bile acid, glyoxylate, dicarboxylate, and fatty acid metabolism (19). Naviaux et al. suggested that CFS patients exhibit a hypometabolic state and found pathway abnormalities in sphingolipid, phospholipid, purine, cholesterol, microbiome, pyrroline-5-carboxylate, ribo avin, branch chain amino acid, peroxisomal, and mitochondrial metabolism (18). Our study shows enrichment similarities in sphingolipid and primary bile acid biosynthesis pathways. As the sphingolipid pathway is altered in both QFS and CFS, these pathway alterations might be speci c for chronic fatigue in general, whereas the primary bile acid biosynthesis pathway might be more speci c for QFS.
Additionally, several of these metabolites, e.g., l-cysteine and l-cysteinylglycine, appear to positively correlate with various in ammatory proteins, e.g., MCP-1, but also 4E-BP1 and MMP-1. PBMCs stimulated with l-cysteinylglycine produced signi cantly more MCP-1 compared to PBMCs that are stimulated with the negative control RPMI. A similar trend was observed for l-cysteine. This shows us that some of these metabolites might have the potential to initiate a more (anti-)in ammatory environment. One could speculate that such a mechanism contributes to changes in in ammation in QFS patients and CFS patients, and that the observed in ammation is secondary to metabolic alterations. Further investigation and validation of these results is warranted, with additional cytokines and chemokines, e.g., 4E-BP1, AXIN1, and MMP-1, in which the metabolite sphingosine 1-phosphate is of particular interest as it is part of the sphingolipid pathway (enriched in both QFS and CFS compared to HC). Furthermore, as our group recently showed that monocytes of QFS patients and CFS patients exhibit a decreased expression of MDP-coding genes MT-RNR1 and MT-RNR2 compared to HC (15), it would also be interesting to investigate the role of these MDP-coding peptides in these metabolic and in ammatory alterations.
Previous studies on gut microbiome composition compared CFS patients to HC and found differences between these groups. Unfortunately, many of the differences are inconsistent. Giloteaux et al., showed that the gut microbiome of CFS patients has less bacterial diversity with the balance shifting towards more pro-in ammatory species (22). Sheedy et al., showed that CFS patients have more aerobic microbial ora, with more Gram-positives, and an abundance of E. faecalis and S. sanguinis compared to HC (60). Armstrong et al., found an increase in Clostridium spp. and a decrease in total bacteria, total anaerobic bacteria, and Bacteroides spp. In CFS patients compared to HC (61). Fremont et al., found that both Belgian and Norwegian CFS patients had an increase in Lactinofacter compared to HC (62). Shukla et al., found a decreased mean relative abundance of Actinobacteria in CFS patients compared to HC (63). Our study found a similar decrease in Bacteroides spp. when comparing CFS patients to HC. Interestingly, this genus appears to be increased when comparing QFS patients to HC. Furthermore, we con ictingly nd an increase in Actinobacteria when comparing CFS patients, but a decrease when comparing QFS patients, to HC. Our most important observation, however, is that the taxonomy of QFS patients and CFS patients is quite similar, while both groups appear to differ quite profoundly from HC (with CFS patients showing a larger difference than QFS patients). This is similar to our ndings in in ammatory and metabolomic pro les and functionally re ected by highly signi cant upregulation of pathways, like urate biosynthesis/inosine 5'-phosphate degradation and CMP-3-deoxy-D-manno-octulosonate biosynthesis, when comparing QFS and CFS patients to HC. When one compares QFS patients to CFS patients, less signi cant upregulation of pathways, like L-lysine biosynthesis III and VI, is found. Exactly how gut microbiome dysbiosis plays part in the pathophysiology of chronic fatigue remains unclear but likely involve the microbiome-brain-axis, and/or subsequent systemic low-grade in ammation. A recent systematic review con rmed that even though independent studies do report differences, these differences are inconsistent (23). Such inconsistencies are likely to occur if control groups are not representative and/or in-and exclusion criteria for patients are not strictly adhered to. Further investigation of the gut microbiome, using strict in-and exclusion criteria together with adequate and representative control groups (64), in patients with chronic fatigue is de nitely of interest.
Although our study lacks a replication cohort, the observed differential patterns among QFS, CFS and HC are consistent across three omics layers. A batch effect across different (control) groups is unlikely but should be kept in mind when interpreting these data. Because systematic assessment of multi-omics data is still limited, our detailed datasets are an important reference for improving our understanding of the molecular processes leading to a state of chronic fatigue.

Conclusion
In conclusion, this study shows that QFS and CFS patients are similar based on their in ammatory and metabolomic pro les, together with gut microbiome taxonomy, while both QFS and CFS patients differ from HC (with CFS patients showing a larger difference than QFS patients). These data suggest that QFS and CFS are similar across three omics layers, indicating cross validation. Furthermore, correlation between metabolomic and proteomic data was validated with laboratory experiments, and a prediction analysis was performed on proteomic data, exposing 4E-BP1 and MMP-1 as potential biomarkers for chronic fatigue. However, while similarities between QFS and CFS are seen and could be associated with chronic fatigue in general, subtle differences, e.g., in in ammatory pro les, should be considered when further investigating its pathogenic mechanisms.