A new approach to find biomarkers in chronic fatigue syndrome/myalgic encephalomyelitis (CFS/ME) by single-cell Raman micro-spectroscopy

Jiabao Xu a, Michelle Potter b, Cara Tomas c, Joanna L. Elson c, Karl J. Morten b, Joanna Poulton b, Ning Wang d, Hanqing Jin d, Zhaoxu Hou d and Wei E. Huang *a
aDepartment of Engineering Science, University of Oxford, Begbroke Science Park, Woodstock Road, Oxford, OX5 1PF, UK. E-mail: wei.huang@eng.ox.ac.uk; Fax: +44 (0)18653749; Tel: +44 (0)1865 283786
bNuffield Department of Women's and Reproductive Health, University of Oxford, the Women Centre, John Radcliffe Hospital, Headley Way, Headington, Oxford, OX3 9DU, UK
cInstitute of Genetic Medicine, Newcastle University, International Centre for Life, Central Parkway, Newcastle upon Tyne, NE1 3BZ, UK
dMathematical Institute, University of Oxford, Andrew Wiles Building, Radcliffe Observatory Quarter, Woodstock Road, Oxford, OX2 6GG, UK

Received 30th July 2018 , Accepted 21st August 2018

First published on 22nd August 2018

Chronic fatigue syndrome (CFS), also called myalgic encephalomyelitis (ME), is a debilitating disorder characterized by physical and mental exhaustion. Mitochondrial and energetic dysfunction has been investigated in CFS patients due to a hallmark relationship with fatigue; however, no consistent conclusion has yet been achieved. Single-cell Raman spectra (SCRS) are label-free biochemical profiles, indicating phenotypic fingerprints of single cells. In this study, we applied a new approach using single-cell Raman microspectroscopy (SCRM) to examine ρ0 cells that lack mitochondrial DNA (mtDNA), and peripheral blood mononuclear cells (PBMCs) from CFS patients and healthy controls. The experimental results show that Raman bands associated with phenylalanine in ρ0 cells and CFS patient PBMCs were significantly higher than those of the wild-type model and healthy controls. As similar changes were observed in the ρ0 cell model with a known deficiency in the mitochondrial respiratory chain as well as in CFS patients, our results suggest that the increase in cellular phenylalanine may be related to mitochondrial/energetic dysfunction in both systems. Interestingly, phenylalanine can be used as a potential biomarker for the diagnosis of CFS by SCRM. A machine learning classification model achieved an accuracy rate of 98% correctly assigning Raman spectra to either the CFS group or the control group. SCRM combined with a machine learning algorithm therefore has the potential to become a diagnostic tool for CFS.


Chronic fatigue syndrome (CFS), also called myalgic encephalomyelitis (ME), is a debilitating complicated disorder characterized by extreme fatigue that is not relieved by rest and other disabling symptoms including neurocognitive impairment and post-exertional malaise.1 CFS affects a population prevalence of at least 0.2% in the UK and represents an extensive burden on the patients, their families and carers, and hence on the society.2 However, it is a challenge to physicians and researchers as it remains to be an incompletely characterized illness, in part due to its controversial definition, pathogenesis and diagnosis.1 Therefore, finding potential biomarkers is of great importance for understanding the disease and in developing targeted treatment.3

Mitochondria have been of great interest to CFS research due to the emerging evidence of mitochondrial dysfunction as a putative biological mechanism for fatigue.4 The underlying hypothesis is that fatigue and other accompanying symptoms are in part due to impaired energy metabolism at the cellular level which is largely controlled by mitochondria as an energy power plant and its ATP production.5 The ATP profile of peripheral blood mononuclear cells (PBMCs) has been used as a diagnostic tool for CFS; however, controversial results were obtained. Several studies have shown a decreased level of ATP in patients’ cohorts,6–8 while others failed to detect any differences or have surprisingly found an elevated energy production in CFS patients.9,10

Here, in an attempt of linking mitochondrial dysfunction and CFS pathogenesis, we sought to analyze single-cell Raman spectra (SCRS) and identify universal biomarkers in cells. A SCRS can be regarded as an intrinsic chemical fingerprint of a cell containing highly resolved Raman bands for major cellular building blocks such as proteins, amino acids, lipids and phospholipids, and carbohydrates.11 Hence, SCRS are label-free biochemical profiles of individual cells reflecting physiological states and metabolic changes.12–14 With rich and semi-quantitative metabolic information of single cells, SCRS can be used to indicate cellular metabolism and display disease-related biomarkers.

In this study, firstly, we used single-cell Raman microspectroscopy (SCRM) to obtain the phenotypic profiles of human ρ0 cell lines completely depleted of mitochondrial DNA (mtDNA) to study the effect of severe mitochondrial dysfunctions. Secondly, we compared the SCRS of PBMCs from 5 CFS patients and 5 healthy controls as a pilot study to find potential biomarkers for CFS diagnosis. The comparison of two results provides insights into the association of mitochondrial dysfunction with the pathophysiology of CFS and demonstrates a coherent and consistent approach for the diagnosis of the illness.


Cell culture, WT cybrid generation and PicoGreen staining

The 143B ρ0 cell line lacking mtDNA was originally generated by King and Attardi.15 It has been extensively characterized and used in many studies investigating the pathogenic effect of mtDNA mutations.16 The ρ0 line was generated by treating the original 143B osteosarcoma line with low concentrations of ethidium bromide (EtBr) for several months. The EtBr intercalated with mtDNA resulted in its removal from the cell. The high glucose concentration in tissue culture media with uridine and pyruvate supplementation allowed ρ0 mtDNA-free cells to be propagated indefinitely. The wild-type (WT) line used in this study was generated by fusing 143B ρ0 cells with normal mtDNA from human platelets. Repopulating the ρ0 line is a better control system than the original 143B wild-type cells as the long-term treatment with EtBr to generate the ρ0 line could have introduced genetic changes not found in the original 143B ρ0 cell line. Platelets do not contain nuclei and provide a simple system for the polyethylene glycol (PEG) induced fusion of mitochondria into the ρ0 cells. The protocol for the PEG induced fusion was carried out as previously described.17 The ρ0 cells were removed by culturing on Dulbecco's Modified Eagle's Medium (DMEM) (25 mM) containing dialyzed serum lacking uridine and pyruvate. Under these conditions, the ρ0 cells died, and the culture contained only new cybrids (cytoplasmic-hybrids) with WT mtDNA. Clones were generated by ringing growing colonies, removing cells with trypsin and re-plating.

Once cultures were established the presence of mtDNA was confirmed by PicoGreen staining. The PicoGreen staining was carried out on live cells at 37 °C for 30 min in standard culture medium using a 3 μL mL−1 dilution of PicoGreen. The cells were examined using a FITC filter and a Leica DM IRE2 fluorescence microscope (Leica Ltd, UK). The mtDNA copy number of the ρ0 cells and the WT cells was also assessed by real-time PCR comparing nuclear and mtDNA markers. The results from the WT line being repopulated with mtDNA18 were consistent with the results from the original 143B line (data not shown).

ATP measurements of WT and ρ0 cells

The cells were added with 50[thin space (1/6-em)]000 or 75[thin space (1/6-em)]000 cells per well in a 96-well white clear bottom plate, in which each well contained a high concentration of glucose (25 mM) in DMEM. The cells were incubated at 37 °C for 8 h to allow adherence. Once the cells had attached, the medium was removed and replaced with 100 μL of fresh media containing various concentrations of glucose from 0 mM to 25 mM and incubated overnight at 37 °C and 5% CO2. ATP was measured using an Abcam ATP assay kit (Ab113849). Before use, 1 vial of lyophilized substrate was reconstituted in 5 mL of substrate buffer. The reconstituted substrate and the detergent were equilibrated to room temperature. Once equilibrated, 50 μL of detergent was added to the wells and placed on a shaker for 5 min at 700 rpm for cell lysis and ATP stabilization. Next, 50 μL of substrate was added and the plate was agitated for a further 5 min at 700 rpm. The plate was dark-adapted for at least 10 min before luminescence was read. ATP levels were calculated from the standard curve that was run alongside the samples.


This work involving human subjects was done under ethics number MS-IDREC-C1-2015-173 and 12/NE/0146.

Isolation of PBMCs from blood samples

PBMCs were separated from the whole blood using the Histopaque method as described by Tomas et al.19 Briefly, whole blood was centrifuged at 700g for 10 min. Plasma was removed and blood was made up to its original volume with sterile phosphate buffered saline (PBS). Histopaque 1.077 was carefully layered on top of a layer of Histopaque 1.119 and the blood layer was added on top of the Histopaque gradient. The tube was centrifuged at 700g for 30 min with the break off. The PBMC layer was collected and washed with fresh PBS. Two types of cell morphologies in the PBMC fraction were observed under the microscope with different cell sizes (5–7 μm and 15–20 μm). Only cells with larger sizes were selected in the Raman experiments after the comparison of the results to minimize heterogeneity generated from different cell types (representative Raman spectra of smaller cells shown in Fig. S1). Further experiments should be done to confirm the identity of the cells, which is now believed to be monocytes according to the cell population and cell size.

Single-cell Raman spectra (SCRS) measurements

The cells were fixed in 4% paraformaldehyde in PBS (v/v) at room temperature for 15 min. The fixed cells were washed twice with Milli-Q water and dropped onto a specifically-coated microscopic slide (which gives no background Raman signal) to be air-dried. SCRS were acquired using an HR Evolution confocal Raman microscope (Horiba Jobin–Yvon, UK Ltd) equipped with a 532 nm neodymium–yttrium aluminum garnet laser. The laser power on the cells was 5 mW, attenuated using neutral density filters. An objective with a magnification of 50× was used to focus single cells with a laser spot size of ∼1 μm2, and Raman scattering was detected by a charge coupled device (CCD) cooled at −70 °C. The spectra were acquired in the range from 330 to 1900 cm−1 with a 600 grooves per mm diffraction grating. A mapping mode was used to characterize the single cells and the acquisition parameters were 5 s per spectrum, 10 spectra per cell and 20–30 single cells per condition.

Raman data pre-processing and analysis

All spectra were pre-processed by comic ray correction and polyline baseline fitting using LabSpec 6 (Horiba Scientific, France). Spectral normalization was done by vector normalization of the entire spectral region. Data analysis, statistics and visualization were done under an R environment using in-house scripts. Quantification of intracellular biomolecules was done by integrating the corresponding Raman bands in SCRS. Bands associated with phenylalanine were integrated in the range of 993–1013 cm−1 and 1022–1036 cm−1 to quantify its intracellular concentration. The quantification results were represented as box plots. The rectangle in the box plots represents the second and third quartiles with a line inside representing the median. The lower and upper quartiles are drawn as lines outside the box. Sample means were compared by using Welch's two sample t-test for unequal variance.

Machine learning classification model

A machine learning classification model was built based on the SCRS of 5 CFS patients and 5 controls (80 and 126 SCRS, respectively). A non-linear support vector machine (SVM) algorithm was used to build the model due to its best performance among all the models tested (data not shown). In the SVM, non-linear hyperplanes (Radial Basis Function kernel) were used to separate the data in high-dimensional space. Leave-one-out-cross-validation (LOOCV) was used to evaluate the performance of the model, and the performance measurements were computed as sensitivity for each condition as well as an overall accuracy rate.

Extracellular flux analysis

Oxidative phosphorylation (OXPHOS) of PBMCs was determined using a Seahorse XFe96 extracellular flux analyzer as described by Tomas et al.19 Briefly, PBMCs were seeded in quadruplicate on a 96-well microplate (Agilent Technologies), coated with poly-D-lysine, at a density of 500[thin space (1/6-em)]000 cells per well. The plate was incubated overnight at 37 °C and 5% CO2. The oxygen consumption rate (OCR) was recorded using the Seahorse XFe96 extracellular flux analyzer following the sequential addition of 1 μM oligomycin, 3 μM carbonyl cyanide-4-(trifluoromethoxy) phenylhydrazone (FCCP), and a combination of 0.5 μM rotenone and 0.5 μM antimycin A. Three basal measurements of the OCR were made before the addition of the first injection (oligomycin) and three measurements of the OCR were made after the injection of each of the compounds. The OCR values were normalized for protein concentration using a bicinchoninic acid assay. Analysis was conducted using Wave software version and Microsoft Excel 2013. Graphs were drawn using GraphPad Prism 7 software.

Results and discussion

Cells lacking mtDNA have distinct single-cell Raman spectra (SCRS)

PicoGreen can be used specifically to stain mitochondrial DNA (mtDNA) in living cells.20 After incubation with PicoGreen, the ρ0 cells repopulated with wild-type mtDNA (WT cells) showed brightly stained nuclei surrounded by numerous bright cytoplasmic speckles (Fig. 1A), whilst the ρ0 cells only displayed brightly stained nuclei without surrounding speckling (Fig. 1B), confirming the depletion of mtDNA in the ρ0 cells.
image file: c8an01437j-f1.tif
Fig. 1 (A) Live WT cells stained with PicoGreen show bright nuclear and cytoplasmic staining. (B) Live ρ0 cells only show apparent nuclear staining (scale bar 20 μm).

As mitochondria are the primary energy power plant of most eukaryotic cells and supply the cells with metabolic energy in the form of ATP,21 we measured the ATP production of WT cells and ρ0 cells with either 50[thin space (1/6-em)]000 or 75[thin space (1/6-em)]000 cells (Fig. 2). Surprisingly, at both cell loadings, the ρ0 cells produced a similar amount of ATP compared to the WT cells, when the supplemented glucose concentrations were high (11 mM and 25 mM). On the other hand, the ATP concentration in the ρ0 cells was significantly lower than that in the WT cells, when the glucose concentrations were low (0 mM to 5 mM). Our results suggest that, when glucose is sufficient, the ρ0 cells can adapt to stimulate ATP production via non-mitochondrial glycolysis to compensate for having a poor mitochondrial respirational chain. When the glucose is low and ATP production from glycolysis is restricted, the WT cell line is able to switch to a mitochondrial mode of ATP production using the respiratory chain via electron transport coupled phosphorylation (ETCP). However, the ρ0 cells are unable to use ETCP due to defective mitochondria.22

image file: c8an01437j-f2.tif
Fig. 2 Intracellular ATP concentrations of the ρ0 and WT cells were measured in (A) 50[thin space (1/6-em)]000 cells and (B) 75[thin space (1/6-em)]000 cells in the presence of different concentrations of glucose. Values are the mean ± S.D. of three independent measurements.

Here we applied single-cell Raman microspectroscopy (SCRM) to examine the WT cells and the ρ0 cells on high glucose and explain the differences in the bioenergetic pathway despite a similar ATP production in two lines. Fig. 3A shows the SCRS of the ρ0 and WT cells with 25 mM glucose, averaged from 30 single cells each. Variations from the single-cell measurements at each Raman peak position were represented by the shallow shade. Relatively low standard deviation was seen due to the small heterogeneity in the in vitro study of the cell lines (Fig. 3A). Compared with the WT cells, the ρ0 cells have a distinct spectral pattern in the fingerprint region of their SCRS (600–1800 cm−1), which typically can be used as a phenotypic fingerprint of cells that contains the most important biochemical information.

image file: c8an01437j-f3.tif
Fig. 3 (A) Raman spectra of the 143B WT cells and ρ0 cells, averaged from 30 single cells each, show distinct spectral patterns. The shaded area represents the standard deviation of the spread in single-cell measurements. (B) Unsupervised PCA separates the SCRS of the WT cells and ρ0 cells in two clusters. (C) PCA loading plot shows the most significant Raman wavenumbers contributing to PC1 of the PCA.

Unsupervised principal component analysis (PCA) was then used to reduce the high dimensionality of SCRS due to the presence of over 1500 Raman bands. A PCA plot along PC1 and PC2 illustrates two clearly separable clusters representing the WT cells and the ρ0 cells (Fig. 3B). As the differences between the two cell types are most significant along PC1, a loading plot was generated for feature extraction of important Raman bands (Fig. 3C). The most prominent feature was observed at 1003 cm−1, which can be assigned to the aromatic ring vibrations of phenylalanine.23 Other bands related to phenylalanine were also found to have high contribution including 1609 (phenyl ring bond-stretching vibrations of phenylalanine), 1030 (C–H/C–C in-phase motion of phenylalanine) and 618 cm−1 (phenyl ring breathing vibrations of phenylalanine).23 All bands related to phenylalanine were higher in the ρ0 cells compared with the WT cells (Fig. 3A), which indicates an important role of the aromatic amino acid, phenylalanine, in the metabolism of cells with mitochondrial dysfunction.

Other differences include bands centered at 1658 (amide I of proteins)24 and 1440 cm−1 (CH2 and CH3 deformation vibrations of lipids)25 (Fig. 3C), both of which are higher in the WT cells than in the ρ0 cells (Fig. 3A). This suggests that despite a higher accumulation of cellular phenylalanine, the ρ0 cells have an overall reduced intracellular concentration of proteins and lipids due to impaired metabolism. As phenylalanine alone might not be sufficient to characterize the pathomechanism, further biomarkers should be identified to simultaneously and more reliably identify mitochondrial dysfunction.

PBMCs of CFS patients can be distinguished by a single Raman marker

Fig. 4A and B elucidate the SCRS of the peripheral blood mononuclear cells (PBMCs) from 5 CFS patients and 5 healthy controls, averaged either for each of the 10 individuals (20–30 SCRS per individual) (Fig. 4A) or for the 2 groups with 80 SCRS for the CFS group and 126 SCRS for the control group (Fig. 4B). As phenylalanine has been found to be a potential biomarker in the ρ0 cells, it is hypothesized that it could be a suitable candidate for investigation in CFS patients that are believed to have similar bioenergetic dysfunction. Out of 5 patients, 4 of them demonstrated Raman spectra with an elevated phenylalanine band at 1003 cm−1 (CFS 1–4) while the other one (CFS 5) exhibited a similar intensity to the controls (Fig. 4A). By averaging 80 SCRS from the patients and 126 spectra from the controls, it is more visible that the phenylalanine band shows a marked increase in patients’ cells (Fig. 4B).
image file: c8an01437j-f4.tif
Fig. 4 (A) Averaged SCRS of the PBMCs from 5 CFS patients and 5 healthy controls, grouped by individuals. (B) Averaged SCRS of the PBMCs from 5 CFS patients (n = 80) and 5 healthy controls (n = 126). The shaded area represents standard deviation from single-cell measurements. (C) PCA plot of the SCRS of PBMCs from 5 CFS patients (n = 80) and 5 healthy controls (n = 126), grouped by individuals. (D) Raman wavenumber loading plots against the contributions of relevant Raman bands to PC4 and PC6 of the PCA.

The PCA of SCRS from 10 samples showed a considerable separation of the CFS group and the control group (Fig. 4C). While CFS 1–4 have smaller ellipsoids with little overlap with the controls, CFS 5 has a higher single-cell scattering and largely overlaps with the control ellipsoids, which correlates with the observation of phenylalanine in Fig. 4A. In order to verify that the separation between groups in the PCA clustering was attributable to phenylalanine, we plotted the Raman wavenumber loadings along PC4 and PC6 which showed the largest separation between the groups. Raman bands centered at 1003 and 1030 cm−1 were identified for describing the maximum variances along both dimensions (Fig. 4D). The intracellular concentration of phenylalanine was semi-quantified by integrating the Raman bands at 1003 and 1030 cm−1, respectively (Fig. 5A and B). Both the signature bands of phenylalanine were found to be significantly higher in the patients compared to the controls (p < 0.0001).

image file: c8an01437j-f5.tif
Fig. 5 Semi-quantification of intracellular phenylalanine in the PBMCs of the CFS group and control group by integrating phenylalanine Raman bands centered at (A) 1003 cm−1 and (B) 1030 cm−1.

Previous research supports the possibility that the pathomechanism of CFS is linked with changes in amino acids. Reductions in the concentrations of certain amino acids including phenylalanine were reported in the serum and urine of CFS patients.26–29 These findings support a possible metabolic defect in CFS patients related to amino acid metabolism by analyzing the metabolomics of the biofluids from the patients. Our work, to the best of our knowledge, is the first to report the changes in the single peripheral blood cells of CFS patients. Our results suggest that patients’ cells could be accumulating more amino acids (e.g. phenylalanine) which results in a reduction in the fluids observed in other studies. This may be due to an induction of secondary rescue mechanisms to intracellularly accumulate more amino acids and maintain a normal ATP production in metabolically dysfunctional patients’ cells.28 Furthermore, our finding demonstrates the impact of phenylalanine on both ρ0 cells lacking mtDNA and blood cells from CFS patients, which could also link mitochondrial defects with CFS. Nevertheless, how the raised phenylalanine levels in PBMCs relate to other tissues in the patients is still unclear. Raman approaches which could provide phenotypic spectra from deeper penetration into tissues would be very useful in future investigations.

Extracellular flux analysis of four CFS patients who were most disparate from the controls in the SCRS results (CFS 1–4) was performed in order to investigate whether a correlation exists between the abnormalities in oxidative phosphorylation (OXPHOS) and the SCRS analysis when PBMCs were incubated under low and high glucose conditions (Fig. 6). Two of the patients showed very low OXPHOS under both the low and high conditions (CFS 2 and 4) and the other two samples had much higher OXPHOS profiles (CFS 1 and 3). While mitochondrial deficiency was markedly detected in two of the four patients by OXPHOS analysis, the SCRS of the PBMC cells were able to find out the abnormalities in all of them. Nevertheless, the sample size of the patients and controls is small considering the possibly high heterogeneity in CFS individuals. Our results serve as a pilot study to explore the potential of SCRM as a tool to identify biomarkers in CFS. A larger-scale study is currently underway to consolidate the results from the current work.

image file: c8an01437j-f6.tif
Fig. 6 Traces from mitochondrial stress tests performed using PBMCs from 4 CFS patients incubated with (A) low glucose (1 mM) and (B) high glucose (10 mM).

Machine learning models can successfully classify CFS patients with a high accuracy rate

In Raman approaches, besides feature extraction to find informative biomarkers, classification based on samples’ spectra is often desirable for diagnostic purposes. Machine learning approaches are usually suitable for resolving the complicated Raman dataset.30 A non-linear support vector machine (SVM) is a supervised learning algorithm using higher dimensional spaces to separate different classes which are non-separable in linear classification. A classification model was established using a non-linear SVM to distinguish between the CFS group and the control group based on their SCRS (80 SCRS for the CFS group and 126 SCRS for the control group). Leave-one-out-cross-validation was used to evaluate the model by using any 205 spectra to establish the model and then testing if the model can correctly classify the one spectrum being left out. The results are summarized in Table 1. This model successfully classified the CFS group with a 96.3% sensitivity (77 out of 80) and the control group with a 99.2% sensitivity (125 out of 126). The lower sensitivity observed in the patient group can be explained by the presence of high heterogeneity in the CFS patients. A total accuracy of 98.1% was achieved based on 206 Raman spectra. Notably, with an increasing sample size and number of Raman spectra to construct the reference database, classification models with better robustness can be achieved. As one Raman spectrum can be obtained within seconds, one patient sample that consists of multiple spectra and multiple cells can be characterized and classified within a few minutes, which implies enormous potential and feasibility in clinical practice.
Table 1 Evaluation of machine learning classification based on SCRS of PBMCs from CFS patients (n = 80) and healthy controls (n = 126)
  Successful classification Sensitivity
CFS group 77/80 96.3%
Control group 125/126 99.2%
Overall accuracy 98.1%


This study is to evaluate the feasibility of single-cell Raman analysis for the detection of biomarkers related to mitochondrial dysfunction and CFS. Accordingly, we identified that the aromatic amino acid, phenylalanine, has an elevated intracellular concentration and can be used as a potential biomarker in ρ0 cells lacking mitochondrial DNA, as well as in the peripheral blood mononuclear cells of CFS patients. Moreover, a machine learning model achieved an accuracy of 98.1% correctly classifying patients and controls based on their Raman spectra. The combination of Raman biomarkers and classification models might lead to improvements in our understanding of CFS pathogenesis and have the potential to be used as a diagnostic tool of CFS.


All experiments were performed in accordance with the guidelines of Newcastle and Oxford University for carrying out clinical studies. Blood samples were collected from patients and controls following approval by the National Research Ethics Committee North East – Newcastle & North Tyneside. Informed consent was obtained from human participants of this study.

Conflicts of interest

There are no conflicts to declare.


MP and CT were supported by the ME association. KM was supported by Diabetes UK. CT was supported by the Medical Research Council. WEH acknowledges support from the EPSRC (EP/M002403/1 and EP/M02833X/1) and the NERC (NE/M002934/1) in the UK.


  1. D. B. Fischer, A. H. William, A. C. Strauss, E. R. Unger, L. Jason, G. D. Marshall Jr. and J. D. Dimitrakoff, Fatigue, 2014, 2, 93–109 Search PubMed.
  2. L. C. Nacul, E. M. Lacerda, D. Pheby, P. Campion, M. Molokhia, S. Fayyaz, J. C. Leite, F. Poland, A. Howe and M. L. Drachler, BMC Med., 2011, 9, 91 CrossRef PubMed.
  3. J. C. Edwards, S. McGrath, A. Baldwin, M. Livingstone and A. Kewley, Fatigue, 2016, 4, 63–69 Search PubMed.
  4. K. Filler, D. Lyon, J. Bennett, N. McCain, R. Elswick, N. Lukkahatai and L. N. Saligan, BBA Clin., 2014, 1, 12–23 CrossRef PubMed.
  5. D. Mitochondrial Medicine Society's Committee on, R. H. Haas, S. Parikh, M. J. Falk, R. P. Saneto, N. I. Wolf, N. Darin, L. J. Wong, B. H. Cohen and R. K. Naviaux, Mol. Genet. Metab., 2008, 94, 16–37 CrossRef PubMed.
  6. S. Myhill, N. E. Booth and J. McLaren-Howard, Int. J. Clin. Exp. Med., 2013, 6, 1–15 Search PubMed.
  7. N. E. Booth, S. Myhill and J. McLaren-Howard, Int. J. Clin. Exp. Med., 2012, 5, 208–220 CAS.
  8. J. Castro-Marrero, M. D. Cordero, N. Saez-Francas, C. Jimenez-Gutierrez, F. J. Aguilar-Montilla, L. Aliste and J. Alegre-Martin, Antioxid. Redox Signaling, 2013, 19, 1855–1860 CrossRef CAS PubMed.
  9. N. Lawson, C. H. Hsieh, D. March and X. Wang, J. Nat. Sci., 2016, 2, e221 Search PubMed.
  10. R. C. Vermeulen, R. M. Kurk, F. C. Visser, W. Sluiter and H. R. Scholte, J. Transl. Med., 2010, 8, 93 CrossRef PubMed.
  11. J. Xu, I. Webb, P. Poole and W. E. Huang, Anal. Chem., 2017, 89, 6336–6340 CrossRef CAS PubMed.
  12. W. E. Huang, R. I. Griffiths, I. P. Thompson, M. J. Bailey and A. S. Whiteley, Anal. Chem., 2004, 76, 4452–4458 CrossRef CAS PubMed.
  13. W. E. Huang, S. Ude and A. J. Spiers, Microb. Ecol., 2007, 53, 471–474 CrossRef CAS PubMed.
  14. Y. Wang, W. E. Huang, L. Cui and M. Wagner, Curr. Opin. Biotechnol., 2016, 41, 34–42 CrossRef CAS PubMed.
  15. M. P. King and G. Attardi, Science, 1989, 246, 500–503 CrossRef CAS PubMed.
  16. H. M. Wilkins, S. M. Carl and R. H. Swerdlow, Redox Biol., 2014, 2, 619–631 CrossRef CAS PubMed.
  17. A. Chomyn, S. T. Lai, R. Shakeley, N. Bresolin, G. Scarlato and G. Attardi, Am. J. Hum. Genet., 1994, 54, 966–974 CAS.
  18. K. J. Morten, N. Ashley, F. Wijburg, N. Hadzic, J. Parr, S. Jayawant, S. Adams, L. Bindoff, H. D. Bakker, G. Mieli-Vergani, M. Zeviani and J. Poulton, Mitochondrion, 2007, 7, 386–395 CrossRef CAS PubMed.
  19. C. Tomas, A. Brown, V. Strassheim, J. L. Elson, J. Newton and P. Manning, PLoS One, 2017, 12, e0186802 CrossRef PubMed.
  20. N. Ashley, D. Harris and J. Poulton, Exp. Cell Res., 2005, 303, 432–446 CrossRef CAS PubMed.
  21. L. D. Osellame, T. S. Blacker and M. R. Duchen, Best Pract. Res., Clin. Endocrinol. Metab., 2012, 26, 711–723 CrossRef CAS PubMed.
  22. M. Potter, E. Newport and K. J. Morten, Biochem. Soc. Trans., 2016, 44, 1499–1505 CrossRef CAS PubMed.
  23. B. Hernández, F. Pflüger, S. G. Kruglik and M. Ghomi, J. Raman Spectrosc., 2013, 44, 827–833 CrossRef.
  24. A. Rygula, K. Majzner, K. M. Marzec, A. Kaczor, M. Pilarczyk and M. Baranska, J. Raman Spectrosc., 2013, 44, 1061–1076 CrossRef CAS.
  25. K. Czamara, K. Majzner, M. Z. Pacia, K. Kochan, A. Kaczor and M. Baranska, J. Raman Spectrosc., 2015, 46, 4–20 CrossRef CAS.
  26. C. W. Armstrong, N. R. McGregor, J. R. Sheedy, I. Buttfield, H. L. Butt and P. R. Gooley, Clin. Chim. Acta, 2012, 413, 1525–1531 CrossRef CAS PubMed.
  27. C. W. Armstrong, N. R. McGregor, D. P. Lewis, H. L. Butt and P. R. Gooley, Metabolomics, 2015, 11, 1626–1639 CrossRef CAS.
  28. O. Fluge, O. Mella, O. Bruland, K. Risa, S. E. Dyrstad, K. Alme, I. G. Rekeland, D. Sapkota, G. V. Rosland, A. Fossa, I. Ktoridou-Valen, S. Lunde, K. Sorland, K. Lien, I. Herder, H. Thurmer, M. E. Gotaas, K. A. Baranowska, L. M. Bohnen, C. Schafer, A. McCann, K. Sommerfelt, L. Helgeland, P. M. Ueland, O. Dahl and K. J. Tronstad, JCI Insight, 2016, 1, e89376 CrossRef PubMed.
  29. S. H. Niblett, K. E. King, R. H. Dunstan, P. Clifton-Bligh, L. A. Hoskin, T. K. Roberts, G. R. Fulcher, N. R. McGregor, J. C. Dunsmore, H. L. Butt, I. Klineberg and T. B. Rothkirch, Exp. Biol. Med., 2016, 232, 1041–1049 CrossRef PubMed.
  30. H. J. Butler, L. Ashton, B. Bird, G. Cinque, K. Curtis, J. Dorney, K. Esmonde-White, N. J. Fullwood, B. Gardner, P. L. Martin-Hirsch, M. J. Walsh, M. R. McAinsh, N. Stone and F. L. Martin, Nat. Protoc., 2016, 11, 664–687 CrossRef CAS PubMed.


Electronic supplementary information (ESI) available. See DOI: 10.1039/c8an01437j

This journal is © The Royal Society of Chemistry 2019