Biochimica et Biophysica Acta 1863 (2016) 56–63

Contents lists available at ScienceDirect

Biochimica et Biophysica Acta

j ourna l homepage: www.e lsev ie r .com/ locate /bbamcr

Functional hyperspectral imaging captures subtle details of cellmetabolism in olfactory neurosphere cells, disease-specific models ofneurodegenerative disorders

Martin E. Gosnell a,c, Ayad G. Anwer a, Juan C. Cassano b, Carolyn M. Sue b, Ewa M. Goldys a,⁎a ARC Centre of Excellence for Nanoscale BioPhotonics, Macquarie University, Sydney 2109, New South Wales, Australiab Department of Neurogenetics, Kolling Institute, Royal North Shore Hospital, Northern Clinical School, University of Sydney, New South Wales, Australiac Quantitative Pty Ltd, ABN 17 165 684 186, Australia

⁎ Corresponding author.E-mail address: ewa.goldys@mq.edu.au (E.M. Goldys)URL: http://www.quantitative.net.au (M.E. Gosnell).

http://dx.doi.org/10.1016/j.bbamcr.2015.09.0300167-4889/© 2015 The Authors. Published by Elsevier B.V

a b s t r a c t

a r t i c l e i n f o

Article history:Received 24 June 2015Received in revised form 17 August 2015Accepted 25 September 2015Available online 26 October 2015

Keywords:Molecular analysisNeurodegenerative disordersMELASMetabolic imagingHyperspectral

Hyperspectral imaging uses spectral and spatial image information for target detection and classification. In thiswork hyperspectral autofluorescence imaging was applied to patient olfactory neurosphere-derived cells, a cellmodel of a human metabolic disease MELAS (mitochondrial myopathy, encephalomyopathy, lactic acidosis,stroke-like syndrome). By using an endogenous source of contrast subtle metabolic variations have been detect-ed between living cells in their full morphological context which made it possible to distinguish healthy fromdiseased cells before and after therapy. Cellular maps of native fluorophores, flavins, bound and free NADH andretinoids unveiled subtle metabolic signatures and helped uncover significant cell subpopulations, in particulara subpopulation with compromised mitochondrial function. Taken together, our results demonstrate that multi-spectral spectral imaging provides a new non-invasive method to investigate neurodegenerative and otherdisease models, and it paves the way for novel cellular characterisation in health, disease and during treatment,with proper account of intrinsic cellular heterogeneity.

© 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license(http://creativecommons.org/licenses/by/4.0/).

1. Introduction

Olfactory neurosphere (ONS) cells are easily accessible, patient-derived models of neurological disease [1]. They are harvested fromthe human olfactory mucosa, the organ of smell in the nose, whichregenerates throughout life. This neural tissue is accessible in humanadults and it demonstrates disease-dependent cell biology alterationsin Alzheimer's and Parkinson's diseases, Rett syndrome, fragile X syn-drome, schizophrenia, MELAS and other diseases [2–8]. The analysis ofsuch cells provides new routes for the understanding of the pathogene-sis of complex neurodegenerative conditions. Relative to other tissueneurons exhibit intense metabolic demands, therefore impairment ofcellular metabolism accompanies many neurodegenerative diseases.Thus new methods are required to characterise and quantify metabo-lism of neural and other cells and tissue on a single cell level. Suchmethods are important for accurate early diagnosis, treatmentmonitor-ing and the development of therapies.

.

. This is an open access article under

Imaging of endogenous cell fluorophores including nicotinamideadenine dinucleotide (NADH), nicotinamide adenine dinucleotidephosphate (NADPH), flavin adenine dinucleotide (FAD) and flavinmononucleotide (FMN) to investigate cell metabolism has beenpioneered by Chance et al.[9]. By using optical microscopy techniques,these autofluorescent compounds can be easily observed providinginsights into metabolic activity in cells without altering them withexogenous labels [10–27]. Multiphoton and fluorescence lifetime imag-ing microscopy (FLIM) have been the leading methods of cellular auto-fluorescence imaging with pioneering works by Skala et al. [10–12],Gratton et al.[15–20], and others. Multiphoton and FLIM imagingmake it possible to determine the relative amounts of FAD and reducedNADH and to distinguish their free and protein-bound forms. This hasbeen used to document increased glycolysis in cancer cells [10–12].Although quantitative analysis of autofluorescence has been previouslyexplored in multiphoton and FLIM modalities, there has been noprevious in-depth work concerned with single photon excited fluo-rescence. This simple approach provides clinically relevant information,and it can be carried out by using unsophisticated, low costinstrumentation.

Here we report the application of hyperspectral autofluorescenceanalysis to functional, metabolic imaging of patient-derived ONS

the CC BY license (http://creativecommons.org/licenses/by/4.0/).

57M.E. Gosnell et al. / Biochimica et Biophysica Acta 1863 (2016) 56–63

cells suffering from themitochondriopathy MELAS. MELAS syndrome iscommonly attributed to the m.3243A N Gmitochondrial DNA (mtDNA)point mutation within the MT-TL1 gene [28–31]. This mutation isthought to disrupt respiratory chain complex assembly as reflected byimpaired mitochondrial protein synthesis [28] and reduced mitochon-drial respiratory chain enzyme activities [29]. Mitochondrial dysfunc-tion follows, with studies reporting reduced mitochondrial membranepotential, with a parallel increase in reactive oxygen species productionleading to reduced ATP production [32] and induction of mitochondrialpermeability [29]. Cellular damage is apparent too, with increased gly-colytic rate, impaired NADH response, decreased glucose oxidationand increased lactate production [31]. This defective oxidativemetabol-ic state is thought to influence the clinical expression of disease. Becausehighly metabolic cells such as myocytes and neurons are highly depen-dent on efficient mitochondrial function, the percentage of mutantmtDNA relative to wild-type mtDNA known as the “mutational load”is important for determining whether the cell will be affected by themutation or not. This poorly understood phenomenon known asheteroplasmy, the existence of both wild-type and mutant mtDNAwithin a cell or tissue is unique to mitochondriopathies. Earlier studieshave demonstrated that highmutational load is detrimental tometabo-lism. Thus the analysis method that is able to yield quantitative insightsconcerning metabolism of individual live cells as well as their popula-tions is a valuable tool to improve the understanding of this and othermitochondrial diseases. This work demonstrates that a simplehyperspectral imaging system coupled to a conventional wide-fieldmicroscope and high content analysis of the images provides novelinsights into cell metabolic signatures, their heterogeneity and MELASdisease biology.

This study examines the effect of MELAS mutational load andthe shift in heteroplasmydue to galactose treatment on cellmetabolism.To this aim we have cultured ONS cells from both control andMELAS patients with greatly varying low (11%) and high (44%) muta-tional loads. We have subsequently exposed them to galactose-supplemented medium which has been shown to cause a reductionin mutational load [33,34]. The cells were imaged to observe theirautofluorescence characteristics. Throughout this work, hyperspectralautofluorescence images of live cells have been obtained at a numberof excitation wavelength ranges between 334 and 495 nm (eachapproximately 10 nm wide). The emission is detected in the range570 nm–605 nm. More details are described in the Methods sectionand in Supplementary Material S1–S5.

Fig. 1. Spectral projections of autofluorescent properties of ONS cells. Controls 1 and2, 11%MELAcells form statistically separate clusters with respect to autofluorescence properties. a) Scatter pspectra for each cell; the colour of these small markers identifies cell populations from diffe44% MELAS); cyan triangles: 11% MELAS; dark green crosses: 11% MELAS galactose treated; pinwithin the single standarddeviation from themean values for each cell population; (b–g) histogNumbers in brackets give statistical distance; p-values for class separation are less than 10−12

green, control 1 — green (1.97); d) 11% MELAS — cyan, 11% MELAS galactose treated — darkMELAS galactose treated — blue, control 2 — red (1.23); g) 44% MELAS — pink, control 2 — reddata have been projected (see Supplementary Section S4.1).

2. Results

2.1. Hyperspectral imaging can distinguish MELAS patients from healthycontrols and sheds light on the effectiveness of galactose treatment

We present the results of hyperspectral autofluorescence character-isation of the ONS cells with low (11%) and high (44%)mutational loadsof the m.3243A N G MELAS mutation (“MELAS ONS cells”) and corre-sponding control cells (control 1 for 11% and control 2 for 44% MELASmutational load, respectively). Hyperspectral data correlations havebeen removed by using PCA as described in Supplementary MaterialSection S4. Subsequently, average cell spectra (represented here bysingle data points) have been plotted in a two-dimensional spectralspace produced by the linear discriminant analysis (LDA, see Supple-mentary Material S4). These are presented in Fig. 1 a where the axesrepresent the directions onto which the cellular data have beenprojected by LDA. The LDA optimally separates the three groups ofcells: healthy cells from both controls 1 and 2, 11% mutant MELAS and44% mutant MELAS cells.

Clear separation of clusters observed in Fig. 1 a indicates that thecells from MELAS patients can be distinguished from control cells.Moreover, the cells from two different MELAS patients also form differ-ent clusters, thus all three cell classes are well separated. To statisticallyanalyse the cluster separations, a second LDA projection of cell datawas carried out [35]. For this new projection, two classes of cells werechosen at a time, and their average spectra were projected onto a com-mon line. This approach visualises cell distributions by histograms(Fig. 1 b–g). We found that every pair of classes selected among control,MELAS, MELAS galactose treated cell classes gave p-values in theKolmogorov–Smirnov test of less than 10−12, consistently with Fig. 1a [36,37]. The Kolmogorov–Smirnov test if passed with p b 0.05 allowsus to reject the null hypothesis that the samples are drawn from thesame distribution.

Themultispectral autofluorescence analysis of cells that have under-gone galactose treatment to reducemutational load is shown in Fig. 1 a.In thisfigure, the clusters ofMELAS treated cells are located between thecontrol population and the MELAS population. The histogramsproduced by pairwise LDA projections are significantly different asshown in Fig. 1 d–g. Thus multispectral imaging makes it possible todistinguish MELAS ONS cells from control cells and also quantifies theresponse of such cells to pharmacological interventions.

S and 44%MELAS aswell as 11%MELAS galactose treated and 44%MELAS galactose treatedlot of average cell spectra in optimised spectral space. Smallmarkers represent average cellrent patients. Green circles: control1 cells (for 11% MELAS); red circles: (control 2 fork triangles: 44% MELAS; blue crosses: 44% MELAS galactose treated cells. Ovals — regionsrams (Parzen estimates, see SupplementaryMaterial S4) of the selected pairs of cell classes.; b) 11% MELAS — cyan, control 1 — green (1.87); c) 11% MELAS galactose treated— darkgreen (1.15); e) 44% MELAS — pink, 44% MELAS galactose treated — blue (8.33); f) 44%(2.08). Spectral variables are distance measures along the directions onto which the cell

58 M.E. Gosnell et al. / Biochimica et Biophysica Acta 1863 (2016) 56–63

2.2. Mapping of key fluorophores in cells

In order to shed light on biochemistry of the ONS cells wedecomposed the spectral images of the cells under investigation intofive most significant spectral components representing themost prom-inent fluorophore groups. This analysis uses an unsupervised unmixingapproach (see Methods and Supplementary Material Section S4.2 fordetails). The resulting component (endmember) spectra are shown inFig. 2 a–e (44% MELAS). The corresponding data for 11% MELAS areshown in Sup. Fig S2. The endmember spectra are found to closelycorrespond to the spectra of bound (red shifted) flavins, FAD and/orspectrally similar FMN, A2E, bound NADH, free NADH, and lipofuscin,the most abundant cellular fluorophores according to the literature[15]. All groups of cells had significant contributions of free and boundNADH as well as A2E and lipofuscin. The fifth spectral component wasdifferent in 11% MELAS and 44% MELAS cells. We found FAD/FMN inthe group of 11% MELAS, 11% treated and control 1 ONS cells butbound flavins were prominent in the group of 44% MELAS, 44% MELAStreated and control 2 ONS cells.

To verify our assignment we carried out colocalisation analysis ofautofluorescence endmembers with organelle stained images (Sup.Figs. S3, S4). An adaptive algorithm was used to correct for cell motionoccurring between autofluorescence imaging and subsequent organelle(mitochondria and lysosome) staining. The image correlations betweenthe endmembers and individual stains are displayed in Sup. Fig. S4

Fig. 2. Label-free imaging of biochemistry in 44%MELAS cells. a–e: Comparison of the excitationin our images byunsupervised unmixing (black). Our endmembers agreewellwith (a) bound (rimages of cells showing the abundance of the five key fluorophores in control 2 (f–j), 44%MELAflavins; g, l,q—A2E;h,m, r— boundNADH; i,n, s— freeNADH; j,o, t— lipofuscin. Cellswere randclarity. The colour scale reflects fluorophore concentration (abundance).

indicating (a) high degree of colocalisation between (bound) flavinsand stained mitochondria; (b) the absence of colocalisation betweenfree NADH bound NADH (correlation coefficient r of 0.17); (c) strongcolocalisation of bound NADH and the mitochondria, with the valuesof r between 0.62 and 0.9; and (d) co-localisation of retinoids with thelysosomes and mitochondria.

Finally, we explored whether the identified fluorophores appropri-ately respond to chemical interventions. To this aim we applied severalknown fluorescence quenchers including (a) sodium borohydridewhich reduces free NAD+ to fluorescent NADH and reduces fluorescentflavins thus quenching their fluorescence; (b) acrylamide whichquenches flavins; and (c) FCCP and rotenone, both specificallyquenching NADH [38,39]. Their effect on the fluorescence intensity isshown in Sup. Fig. S5. Fig. S6 shows a comparison of hyperspectralunmixing with fluorescence lifetime imaging (FLIM) in cells treatedwith H2O2 where we observed similar trends.

Our approach produced cellular maps of each of the five keyfluorophore groups within the examined cells. These are shown inFig. 2 f–t where we present images of the 44% MELAS ONS cells beforeand after galactose treatment as well as the control ONS cells (seeFig. S2 for 11% MELAS). We note that flavins in MELAS ONS cells(Fig. 2 k and Fig. S2 k) are concentrated in filamentous structuressurrounding the nucleus. Retinoids in control ONS cells (Fig. 2 g, j andFig. S2 g, j) form perinuclear rings consistent with their mitochondriallocalisation [14,40] with some contributions from the lysosomes as

spectra offive keyfluorophores (red) in 44%MELASONS cellswith endmembers identifieded shifted)flavins, (b) A2E, (c) boundNADH, (d) freeNADH and (e) lipofuscin.We presentS (k–o) and 44%MELAS treated ONS cells (p–t). Columns from left to right: f, k, p— boundomly selected and their outlines determinedusing top 3 PCA imageshave beenmarked for

59M.E. Gosnell et al. / Biochimica et Biophysica Acta 1863 (2016) 56–63

well. Statistically significant trends in average fluorophore intensitiesin all examined cells are shown in Sup. Fig. S7. These data indicatethat MELAS ONS cells present a fluorophore profile that reflects alteredmetabolism, different to control ONS cells, while galactose treatmentappears to modify the fluorophore profile of the treated MELAS cells.

2.3. Identification of cell subpopulations

We used our automated algorithm described in Methods to searchfor cell subpopulations by looking for clustering of fluorophore abun-dances. All combinations of features have been tested and scored interms of howwell they clustered indicating subpopulation coincidence.This was done for all features shown to have definite one-dimensionalsubpopulations.

Interestingly, we found two uneven well-separated subpopulationsfor 11% MELAS (red and blue symbols, Fig. 3 a), and two similar sizeequally well-separated subpopulations for 44% MELAS (yellow andgreen symbols, Fig. 3 b). Statistical analysis of these and other subpopu-lations before and after galactose treatment is shown in Sup. Fig. S8. Wealso draw attention to the behaviour of the optical redox ratio, in partic-ular the ratio of flavins to boundNADH (Sup. Fig. S8) showing clear sub-populations in MELAS ONS cells. These subpopulations disappear aftergalactose treatment. The ratio of free to boundNADHwhich is a relativemeasure of glycolysis and several other features also shows clearsubpopulations (Sup. Fig. S8).

2.4. Biochemical differences between the examined cell groups

Sup. Fig. S9 summarises the results of biochemical characterisationof the examined cell groups. It shows that the biochemistry of MELAScells is significantly altered compared to healthy controls. For examplethe ATP production andMTMP inMELAS cells are lower than in healthycontrols, while the levels of lactate and SOX are elevated. These valuesreturn closer to those in control cells after treatment. These results areconsistentwith significant differences observed in the autofluorescenceproperties of these cell groups.

3. Discussion

3.1. Hyperspectral imaging enables differentiation of cell groups

Hyperspectral analysis to differentiate cell groups enables to statisti-cally establish that MELAS ONS cells differ from control cells. The ONScells from each MELAS patient before and after galactose treatmentcan also be distinguished (Fig. 1 a). This is consistentwith earlier studiesshowing that galactose treatment enhances mitochondrial metabolism.We and others have shown improved mitochondrial function withexposure of diseased cells to galactose by shifting the mutational loadof diseased cells [33,34]. The mechanism by which cells shift in muta-tional load is unknown, however, selective apoptosis in combinationwith cell proliferation and mitophagy for surviving cells may reducethe mutational load of cells.

The method presented here produces p-values for the hypothesisthat the two cell groups are different. It is also robust with respect toreproducibility and the same results have been obtained in repetitivecell experiments from different cell passages carried out four monthsapart (data not shown). These results also show that hyperspectralanalysis of cellular autofluorescence is sensitive enough to monitorfunctional treatment response of ONS cells from single individuals.

3.2. Hyperspectral imaging enables biochemically sensitive functionalimaging of cells

In addition to cell classifications, our hyperspectral dataset providesdetailed insights into cell biochemistry. The fluorescence of unlabelledcells is produced by endogenous fluorophores including NADH

and NAD(P)H, riboflavin and flavin–coenzymes including FAD andFMN, retinol and other retinoids including components of lipofuscin,pyridoxine, vitamin B12, vitamin D, ceroid, cytochromes, vitamin K,tryptophan, tyrosine, phenylalanine, kynurenine and porphyrins, aswell as the complexes of these fluorescent molecules with proteins[19,20,26,41–47]. Among these fluorophores protein-bound and freeNAD(P)H produces a significant part of the fluorescence signalresponding to excitation in the 330–390 nmwavelength range. Flavinsare prominent in the longer wavelength range (excitation peaks at360 nm and at 440–450 nm, emission in the range 450–600 nm) withparts of their spectra well separated from the NAD(P)H group. Reti-noids, especially retinol can also be observed, in particular lipofuscinand A2E tend to be highly visible [19,20,41]. Previous studies of autoflu-orescence in cells indicate the absence of fluorophore quenching; thismeans that fluorophore concentration is proportional to fluorescenceintensity. In particular, a linear correlation has been shown betweentotal NADH content determined biochemically and NADH fluorescence,[42,43]. The total NADH concentration is typically in 100 μM range [44].Other fluorophores listed here are less abundant, for example flavins arepresent in cells at ~50 μM [45], and the most abundant retinoid, retinolat 1–2 μMconcentrations. Retinoic acid, anhydroretinol and retinal havelower concentrations; all are below the onset of concentrationquenching [26], while lipofuscin can form highly concentrated granules.

Unlike in other works [14] our spectra have sufficient spectral reso-lution to differentiate between unbound and protein-bound forms ofNADH, whose spectrum is blue-shifted by 20 nm compared to freeNADH [46,47]. However we are unable to differentiate between NADHand NADPH, thus the NADH components in this work (free or bound)represents both NADH and NADPH fluorescence [26]. The bound flavincomponent represents the majority of the cellular flavins' fluorescenceproduced by the mitochondrial lipoamide dehydrogenase (LipDH) andelectron transfer flavoprotein (ETF), but also any other bound flavins.The shape of the bound flavin excitation spectrum observed here issimilar to that of aqueous solution of riboflavin, with a red shift by50 nm by the effect of protein binding [38,48–50]. We have been ableto separately identify FAD, but it could not be separated from spectrallysimilar FMN [14]. In the retinoid group, the A2E has also been separatelyidentified, while the spectrally similar retinoids [51,52] (retinol, retinoicacid, cellular retinol-binding proteins (CRBPs), and cellular retinoicacid-binding proteins (CRABPs)) have been merged with lipofuscin.

3.3. Trends in fluorophore content

Here, we discuss the trends observed in various cellular characteris-tics between control, MELAS and MELAS treated cells. We first discussthe fluorophore content presented in Fig. S7 which shows thatthe 44% MELAS cells show significantly more bound flavin fluores-cence than the controls. The 11% MELAS cells seem to contain mostlyFAD/FMN instead of bound flavins and the increase of FAD/FMN inMELAS cells is much less pronounced. Mitochondriopathies often leadto accumulation of reduced forms of FAD, NADH and FMN, as they can-not bind to the respiratory chain proteins [53]. In mitochondriopathiessuch asMELASwhich affects assembly of respiratory chain proteins, it isexpected that at high mutational loads (such as 44%), we would findreduced forms of FAD and FMN (all flavins) present at higher levelswhen compared to control cells. However, at lower mutational loads(such as 11%) this effect is expected to be less pronounced, becausethe assembly of proteins may not be as disrupted as it is at higher mu-tational loads. The build-up of reactive oxygen species also confirmedin our experiment (Fig. S9)may change the flavin redox balance and in-crease flavin fluorescence.

The trends in free and bound NADH content in the examined cellsare complicated, with 44% MELAS cells showing a slight increase ofbound NADH compared to controls (Fig. S7), while 11% MELAS cellsshow slightly less bound NADH and slightly more free NADH than therespective control cells. These complex trends need to be reconciled

60 M.E. Gosnell et al. / Biochimica et Biophysica Acta 1863 (2016) 56–63

with widely reported increase in NADH fluorescence in more glycolyticcells [11]. However this increase has been reported only in cells withintact, functioning mitochondria. While MELAS ONS cells do exhibithigher levels of glycolysis than control cells (Fig. S9), they may havedecreased mitochondrial numbers at certain mutational loads. In suchcells, the amount of NADH present under glycolysis conditions maydepend upon the overall cytosolic pool size of NAD+ and NADH, andmay bear no simple relation to the amount of NADH in the controlcells. The relatively higher levels of bound NADH in 44% than in 11%MELAS cells (Fig. S9) could be consistent with a higher mitochondrialactivity combined with a smaller mitochondrial mass, resulting in asimilar ATP production, with a slightly reduced production of lactate,and a much higher superoxide generation.

3.4. Discussion of specific cell subpopulations

Here we discuss cell populations visualised by scatterplots in Fig. 3and those statistically analysed in Fig. S8. For 44% MELAS, Fig. 3 bshows two distinctive populations with different metabolic profiles(yellow and green symbols) where the “blue” cells have a generallyhigher ratio of free to bound NADH, and they also have a higher averagelipofuscin. These cells constitute ~41% of the total, consistent with44% mutational loading, and they form a completely separated cluster.Fig. 3 a displays subpopulations uncovered in 11% MELAS ONS cells(red and blue symbols). These two populations have similar character-istics as those in 44% MELAS. Cells in the small cluster comprising7% cells (red symbols) has a high ratio of free NADH to bound NADHand a majority of cells (blue symbols) have a much lower free tobound NADH ratio. The red cells have comparatively high levels oflipofuscin. We do not know at this point whether these metabolicallydifferent “red” and “yellow” cells are carrying themutations; this shouldbe investigated further.

The behaviour of subpopulations before and after treatment isshown in Fig. S8. This figure shows the plots of bound flavin versusbound NADH content for control, 44% MELAS and 44% MELAS treatedcells. We can see here that control cells form two fairly tight popula-tions, with bound flavins' abundance varying in the range of 0.2 to 2.2.In contrast, the MELAS cells form very distinctive populations, onewith low bound flavins (between 0 and 3) and a group of cells whichcould be still mutated with an unusually high bound flavin content(between 5 and 20). After treatment, the high bound flavins' groupcompletely disappears and the remaining cells form a single populationspreading over the same region of abundances as the control cells. Sup.Fig. S8 also shows the populations with respect to the free and boundNADH content. Again, we observe two well-defined subpopulationsin control cells. The MELAS cells also show two populations, one of

Fig. 3. Scatterplots showing the most pronounced correlations between fluorophore abundancare indicated on the axes. Ovals are guide for the eye only, to highlight cell populations.

which has a much higher free NADH content that the control cells.After treatment only a single population is observed. Sup. Fig. S8shows similar trends in the fluorophore ratios, again displaying clearsubpopulations in most cases. These subpopulations are particularlypronounced in MELAS cells for bound flavins to bound NADH ratio.These subpopulations are associated with varying levels of metabolicactivity.

The data in Fig. 3 and Sup. Fig. S8 are consistent with the view thatMELAS cells in this patient belong to two distinctive classes. Thus ourdata shed light on the hypothesis that MELAS is a disease which affectscells to a different degree as opposed to the alternative hypothesis thatall cells are similar and characterised by a certain proportion of defectivemitochondria in each. Distinguishing between these two hypotheses issignificant for the design of patients' treatment regimens.

4. Conclusions

Originally developed for remote sensing, this work expands thehyperspectral approach to the fields of medicine and cell biology inareas utilising fluorescence microscopy. It yields quantitative andstraightforward interpretation of physiological processes in livingtissues similar to multiphoton fluorescence and/or FLIM but withoutthe complexity and cost of a multiphoton FLIM imaging system. Theuse of single photon fluorescence at comparatively low excitationirradiances (in the order of 103 W/cm2) from light emitting diodesmakes this method compatible with in-vivo use.

We demonstrated that hyperspectral or multispectral imagingprovides richly detailed information about living cells which allows totest hypotheses about similarity or otherwise of cell distributions andabout the effects of chemical interventions, in this case galactosetreatment.

We have been able to unambiguously decompose cellular autofluo-rescence into contributions from several most abundant fluorophorespresent in those cells, generate cellular maps and provide statistical in-formation about cellular abundances of these fluorophores. Althoughwe have not included here detailed analysis cellular distribution ofthese fluorophores, this method can be naturally extended to providecombined morphological and chemical cellular features.

The information about average content of the most abundantfluorophores in each individual cell obtained in our method bearssome similarity to flow cytometry measurements. Hundreds of cellscan be analysed in each experiment mostly automatically and there isscope for full automation. This approach has allowed to identify biolog-ically and functionally distinguishable cell subpopulations of potentialbiological significance.

es in cells and cell subpopulations (a) in 11% MELAS cells, (b) in 44%MELAS. Fluorophores

61M.E. Gosnell et al. / Biochimica et Biophysica Acta 1863 (2016) 56–63

Overall, our results indicate that hyperspectral imaging providesnew windows into biological functions of individual living cells, and itcan detect and quantify their metabolic activity. Our approachcombined with high content data mining of cell images has providednovel insights into the MELAS disease which is characterised by genet-ically determined mitochondrial dysfunction. We have shown that thenew method can help assess the disease status and the effectivenessof therapy. Olfactory neurosphere cells investigated here are regardedas disease-specific models of neuronal disorders [1]. Using this methodit will be possible to analyse impairment of metabolic function in thesediseases. Thus hyperspectral analysis presented here may help to betterunderstand a broad range of diseases affecting metabolism includingneuronal degeneration.

5. Methods

See Supplementary Material for expanded description of methods.

5.1. Primary cell culture

In this study we used cells from the MELAS patient-derivedolfactory-derived neurospheres (ONS)with varying levels ofmutationalload (11% and 44%) generated from olfactory primary cultures. Toprepare these ONS, nasal biopsies were carried out and harvested cellswere expanded and subcultured as described in more details in Supple-mentary Material S1. All biopsy tissues were obtained with informedwritten consent of the patients and the study carried out under a proto-col that was approved by the ethics committees of the Royal NorthShore Hospital and theUniversity of Sydney, according to the guidelinesof the National Health and Medical Research Council of Australia. Cellsfrom control subjects were harvested and treated in the same fashion.The MELAS cells were subjected to galactose-supplemented, glucose-free media (starvation) over a period of 4 days. We present data fromthree different cell passages.

5.2. Microscopic imaging

Imaging was carried out on cells having an approximately constantcell density. Each cell sample was prepared and imaged in duplicate.Altogether, over 1500 cells were imaged in this study, between350 and 400 for each patient (more details in Supplementary MaterialS2 and S3).

5.3. Mitochondrial parameter assays

Cells were carefully characterised biologically by a variety ofmethods. In particular we measured mutational load by standardradiolabelled (“hot-finish”) PCR/RFLP analysis, carried out the citratesynthase assay for the determination of mitochondrial mass, and mea-sured the level of lactate, ATP, mitochondrial superoxide generation(SOX), and mitochondrial transmembrane potential by flow cytometry.All cells were also imaged by laser scanning microscopy, with GreenMitotracker, Red Lysotracker and Blue ER-Tracker staining (see Supple-mentary Material S2 for more details).

5.4. Hyperspectral measurements technique

Our method uses fluorescence of native fluorophores commonlyfound in cells. In our approach the images of live cells are obtained byan Andor IXON camera under illumination at a number of selectedbands of excitation wavelengths (here, centred at 334, 365, 385, 395,405, 415, 425, 435, 455, 475, 495 nm, each about 10 nm wide).The emission is measured with a 532 nm long pass dichroic mirrortogether with a 587 nm bandpass filter (35 nm bandwidth), in therange 570 nm–605 nm. The spectra obtained in this way are excitationspectra. This selection of excitation–emission channels with optimised

exposure times enables to capture images with sufficient signal tonoise ratio for accurate unmixing of multiple cellular fluorophores.This includes free and bound NADH whose spectra have tails in the570–605 nm range, however these compounds produce a significantproportion of the autofluorescence signal at 334 nm excitation wave-length. Optical powers at the objective ranged from 0.1 μW at 334 nmexcitation to 102 μWat 475 nm. Thesemeasurements yieldfluorescenceexcitation spectra measured at each pixel for all images of cells comingfrom all patients. These spectra are further corrected for instrumentalresponse. A “background” reference image of a culture dish with amedium is also taken and subtracted from all images with cells.

5.5. Data analysis method for differentiation of cell groups

The original pixel spectra containing intensity andwavelength infor-mation are represented as vectors in an n-dimensional spectral spacewhose coordinates are intensities at each of the n excitationwavelength(here, n = 11). This dataset contains significant correlations, as cellimages at adjacent wavelengths are very similar; this is first removedby using a covariance matrix calculated from the data (see Supplemen-taryMaterial S4 formathematical details). Further, we calculate averagecell spectra. The method of linear discriminant analysis (LDA)[35] isthen employed, in order to establish whether these average spectrafrom biologically identifiable classes of cells form well separated clus-ters. In this method we first choose the m cell classes and by usingLDA we project the original n-dimensional spectral space and the datapoints representing the average spectra of our cells onto a new, lower-dimensional space. Its dimension is given by the number of groups ofcell classes to be distinguished less1. For convenience of presentation,in this work we have been using three classes of cells, so that afterLDA the spectra of these cells can be depicted as points on two dimen-sional plots. This two-dimensional spectral space produced by LDA isreferred to as the “canonical spectral space”. Its basis vectors are orthog-onal, and aligned with the axes in our relevant figures.

The LDAmethod ensures that the new space is optimised to providethe best degree of separation between selected cell classes (such as, forexample, cells from different patients). Finally, to quantify the distinc-tiveness between selected pairs of cell clusters we perform the LDAanalysis again, on each pair of cell cluster data projecting them onto aone-dimensional line. The Kolmogorov–Smirnov statistical test is thenapplied to gauge and compare the similarity of the pair of clusters. Wealso calculate the maximum Fisher statistical distance, a measure ofcluster closeness which is sensitive to cluster means and takes accountof the data dispersion.

We add that the data for additional cells and/or patients or data pro-duced by other authorsmay beplotted togetherwith our data. Althoughthere is no mathematical certainty that optimum separation will beachieved for such blended datasets, a clear separation may often beachieved in the case when class distinction results are statisticallystrong with small p-values.

5.6. Data analysis method to determine fluorophore abundance

We are assuming that the observed spectra are a linear combinationof fundamental spectra originating from particular substances, calledendmembers. ThisN number of endmember spectra Xis weighted byan abundance coefficient vector f such that an observed spectrumZpwhich is given by Eq. (1):

Zp ¼ Xf þ e ¼XN

i¼1Xi f i þ e ð1Þ

where e is the residual, or noise, not explained by the mixture ofendmembers.

The problem of finding the endmembers has a simple geometricalinterpretation. The spectra of all individual pixels in cells represent a

62 M.E. Gosnell et al. / Biochimica et Biophysica Acta 1863 (2016) 56–63

certain cluster in anN dimensional space. The cluster is contained with-in a convex hull called a simplex and each pixel spectrum point withinthis simplex represents a linear combination of the spectra representedby the extreme points (vertices) of that simplex. These vertex spectraare chosen to be endmembers. See Supplementary MaterialSection S4.2 for more details about identifying endmembers.

In order to find the best simplex representing the data points twoconditions need to be satisfied: that the coefficients of the abundancevector are all positive and that they add to unity. To identify that bestsimplex we minimised the least squares error between the targetspectra and the product of our abundance vector multiplied by theendmember spectra, for the entire set of data. The correspondingendmembers spectra are shown in Fig. 2 and Sup. Fig. S1. These areidentified to measured reference spectra of known fluorescentcompounds in cells and/or to their spectra reported in the literature.Fit errors are typically being between 5 and 10%, due to the presenceof other less abundant and unidentified residue spectra.

5.7. Data analysis method to detect cell subpopulations

In order to characterise subpopulations in the examined cells wefirst specify K, the number of subpopulations we wish to find in eachdata group (K=2 in the presented analysis). Then for all groupswithinall variables, the data undergo an unsupervised mixture model. Thenon-deterministic algorithm is used to break the data into potentialsubpopulations. Each produced solution goes through selection criteriaso that only the highest scoring solution per data group is selected. Thecriteria are that each of the subpopulations must be N30% of the wholedataset, the subpopulations must have a statistical separation greaterthan 1, and they must pass a Kolmogorov–Smirnov test with p b 0.05.The mixture model returns only the subpopulation means and covari-ance, so in order to classify the data points in one of the K subpopula-tions we used a Naive Bayes classifier. The method reproducibly findssubpopulations that agree with intuition when looking at the histo-grams. Boxplots are used to give an overview of data and a visualisationhas been created to display groups of data with subpopulations. Thewidth of the subpopulation boxes is proportional to their size, to allowsome inference power across groups.

Contributions

E.G., C.S. and M.G. designed the experiments. M.G., J.C. and A.A.performed the experiments. M.G. and E.G. analysed the data. E.G., C.S.,M.G., J.C. and A.A. wrote the manuscript. All authors have reviewedthe manuscript.

Competing financial interests

Authors have no competing financial interests.

Transparency document

The Transparency document associated with this article can befound, in the version.

Acknowledgements

This work was partially supported by the Australian ResearchCouncil (ARC) through its Centre of Excellence scheme (CE140100003).

Appendix A. Supplementary data

Supplementary data to this article can be found online at http://dx.doi.org/10.1016/j.bbamcr.2015.09.030.

References

[1] N. Matigian, G. Abrahamsen, R. Sutharsan, A.L. Cook, A.M. Vitale, A. Nouwens, B.Bellette, J. An, M. Anderson, A.G. Beckhouse, M. Bennebroek, R. Cecil, A.M. Chalk, J.Cochrane, Y. Fan, F. Féron, R. McCurdy, J.J. McGrath, W. Murrell, C. Perry, J. Raju, S.Ravishankar, P.A. Silburn, G.T. Sutherland, S. Mahler, G.D. Mellick, S.A. Wood, C.M.Sue, C.A. Wells, A. Mackay, Disease-specific, neurosphere-derived cells as modelsfor brain disorders, Dis. Model. Mech. 3 (2010) 785–798.

[2] B. Wolozin, T. Sunderland, B.-b. Zheng, J. Resau, B. Dufy, J. Barker, R. Swerdlow, H.Coon, Continuous culture of neuronal cells from adult human olfactory epithelium,J. Mol. Neurosci. 3 (3) (1992) 137–146.

[3] M.T. Abrams, W.E. Kaufmann, F. Rousseau, B.A. Oostra, B. Wolozin, C.V. Taylor, N.Lishaa, M.-L. Morel, A. Hoogeveen, A.L. Reiss, FMR1 gene expression in olfactoryneuroblasts from two males with fragile X syndrome, Am. J. Med. Genet. 82 (1)(1999) 25–30.

[4] F. Feron, C. Perry, M.H. Hirning, J. McGrath, A. Mackay-Sim, Altered adhesion, prolif-eration and death in neural cultures from adults with schizophrenia, Schizophr. Res.40 (3) (1999) 211–218.

[5] S.E. Arnold, L.-Y. Han, P.J. Moberg, B.I. Turetsky, R.E. Gur, J.Q. Trojanowski, C.-G.Hahn, Dysregulation of olfactory receptor neuron lineage in schizophrenia, Arch.Gen. Psychiatry 58 (9) (2001) 829–835.

[6] G.V. Ronnett, D. Leopold, X. Cai, K.C. Hoffbuhr, L. Moses, E.P. Hoffman, S. Naidu,Olfactory biopsies demonstrate a defect in neuronal development in Rett'ssyndrome, Ann. Neurol. 54 (2) (2003) 206–218.

[7] R.D. McCurdy, F. Féron, C. Perry, D.C. Chant, D. McLean, N. Matigian, N.K. Hayward,J.J. McGrath, A. Mackay-Sim, Cell cycle alterations in biopsied olfactoryneuroepithelium in schizophrenia and bipolar I disorder using cell culture andgene expression analyses, Schizophr. Res. 82 (2–3) (2006) 163–173.

[8] A. Haehner, S. Boesveldt, H.W. Berendse, A.Mackay-Sim, J. Fleischmann, P.A. Silburn,A.N. Johnston, G.D. Mellick, B. Herting, H. Reichmann, T. Hummel, Prevalence ofsmell loss in Parkinson's disease—a multicenter study, Parkinsonism Relat. Disord.15 (7) (2009) 490–494 (2009).

[9] B. Chance, J.R. Williamson, D. Jamieson, B. Schoener, Properties and kinetics ofreduced pyridine nucleotide fluorescence of the isolated and in vivo rat heart,Biochem. Z. 341 (1965) 357–377.

[10] M.C. Skala, J.M. Squirrell, K.M. Vrotsos, J.C. Eickhoff, A.G. -Fitzpatrick, K.W. Eliceiri, N.Ramanujam, Multiphoton microscopy of endogenous fluorescence differentiatesnormal, precancerous, and cancerous squamous epithelial tissues, Cancer Res. 65(2005) 1180–1186.

[11] M.C. Skala, K.M. Riching, D.K. Bird, A.G. -Fitzpatrick, J. Eickhoff, K.W. Eliceiri, P.J.Keely, N. Ramanujam, In vivo multiphoton fluorescence lifetime imaging ofprotein-bound and free nicotinamide adenine dinucleotide in normal and precan-cerous epithelia, J. Biomed. Opt. 12 (2) (2007) 024014–024114.

[12] M.C. Skala, K.M. Riching, A.G. -Fitzpatrick, J. Eickhoff, K.W. Eliceiri, J.G. White, N.Ramanujam, In vivo multiphoton microscopy of NADH and FAD redox states,fluorescence lifetimes, and cellular morphology in precancerous epithelia, Proc.Natl. Acad. Sci. U. S. A. 2007 104 (49) (2007) 19494–19499.

[13] A. Uchugonova, K. Konig, Two-photon autofluorescence and second-harmonicimaging of adult stem cells, J. Biomed. Opt., 2008 13 (5) (2008) 054068-054068.

[14] W.L. Rice, D.L. Kaplan, I. Georgakoudi, Two-photon microscopy for non-invasive,quantitative monitoring of stem cell differentiation, PLoS ONE 5 (4) (2010),e10075.

[15] C. Stringari, A. Cinquin, O. Cinquin, M.A. Digman, P.J. Donovan, E. Gratton, Phasorapproach to fluorescence lifetime microscopy distinguishes different metabolicstates of germ cells in a live tissue, Proc. Natl. Acad. Sci. U. S. A. 108 (33) (2001)13582–13587.

[16] C. Stringari, R.A. Edwards, K.T. Pate, M.L. Waterman, P.J. Donovan, E. Gratton, Meta-bolic trajectory of cellular differentiation in small intestine by Phasor FluorescenceLifetime Microscopy of NADH, Sci. Rep. 2 (2012) 00568.

[17] C. Stringari, L. Jamison, N. Lisa, A. Flanagan, E. Gratton, Phasor fluorescence lifetimemicroscopy of free and protein-bound NADH reveals neural stem cell differentiationpotential, PLoS ONE 7 (11) (2012), e48014.

[18] C. Stringari, R. Sierra, P.J. Donovan, E. Gratton, Label-free separation of humanembryonic stem cells and their differentiating progenies by phasor fluorescencelifetime microscopy, J. Biomed. Opt. 17 (4) (2012) 0460121–04601211.

[19] B.K. Wright, L.M. Andrews, M.R. Jones, C. Stringari, M.A. Digman, E. Gratton, Phasor-FLIM analysis of NADH distribution and localization in the nucleus of live progenitormyoblast cells, Microsc. Res. Tech. 75 (12) (2012) 1717–1722.

[20] B.K. Wright, L.M. Andrews, J. Markham, M.R. Jones, C. Stringari, M.A. Digman, E.Gratton, NADH distribution in live progenitor stem cells by phasor-fluorescencelifetime image microscopy, Biophys. J. 103 (1) (2012) L7–L9.

[21] W.R. Zipfel, R.M. Williams, R. Christie, A.Y. Nikitin, B.T. Hyman,W.W. Webb, Live tis-sue intrinsic emission microscopy using multiphoton-excited native fluorescenceand second harmonic generation, Proc. Natl. Acad. Sci. U. S. A. 100 (12) (2003)7075–7080.

[22] R.M. Rodrigues, P. Macko, T. Palosaari, M.P. Whelan, Autofluorescence microscopy: anon-destructive tool to monitor mitochondrial toxicity, Toxicol. Lett. 206 (3) (2011)281–288.

[23] M.G. Field, D. Yang, Z.M. Bian, H.R. Petty, V.M. Elner, Retinal flavoprotein fluores-cence correlates with mitochondrial stress, apoptosis, and chemokine expression,Exp. Eye Res. 93 (4) (2011) 548–555.

[24] R. Orzekowsky-Schroeder, A. Klinger, B. Martensen, M. Blessenohl, A. Gebert, A.Vogel, G. Hüttmann, In vivo spectral imaging of different cell types in the smallintestine by two-photon excited autofluorescence, J. Biomed. Opt. 16 (11) (2011)116025–11602513.

63M.E. Gosnell et al. / Biochimica et Biophysica Acta 1863 (2016) 56–63

[25] K.C. Reinert, W. Gao, G. Chen, X. Wang, Y.-P. Peng, T.J. Ebner, Cellular and metabolicorigins of flavoprotein autofluorescence in the cerebellar cortex in vivo, Cerebellum10 (3) (2011) 585–599.

[26] S. Huang, A.A. Heikal, W.W. Webb, Two-photon fluorescence spectroscopy andmicroscopy of NAD(P)H and flavoprotein, Biophys. J. 82 (5) (2002) 2811–2825.

[27] K.A. Kasischke, H.D. Vishwasrao, P.J. Fisher, W.R. Zipfel, W.W. Webb, Neural activitytriggers neuronal oxidative metabolism followed by astrocytic glycolysis, Science305 (5680) (2004) 99–103.

[28] A. Chomyn, A. Martinuzzi, M. Yoneda, A. Daga, O. Hurko, D. Johns, S.T. Lai, I. Nonaka,C. Angelini, G. Attardi, MELAS mutation in mtDNA binding site for transcriptiontermination factor causes defects in protein synthesis and in respiration but nochange in levels of upstream and downstream mature transcripts, Proc. Natl. Acad.Sci. U. S. A. 89 (10) (1992) 4221–4225.

[29] D. Cotán, M.D. Cordero, J.G. -Maraver, M.O. -Ávila, Á.R. -Hernández, L.G. Izquierdo,M.D.I. Mata, M.D. Miguel, J.B. Lorite, E.R. Infante, S. Jackson, P. Navas, J.A.S. -Alcázar, Secondary coenzyme Q10 deficiency triggers mitochondria degradationby mitophagy in MELAS fibroblasts, FASEB J. 25 (8) (2011) 2669–2687.

[30] D.M. Sproule, P. Kaufmann, Mitochondrial encephalopathy, lactic acidosis, andstrokelike episodes: basic concepts, clinical phenotype, and therapeutic manage-ment of MELAS syndrome, Ann. N. Y. Acad. Sci. 1142 (2008) 133–158.

[31] P.B.M. De Andrade, B. Rubi, F. Frigerio, J.M.W. Van den Ouweland, J.A. Maassen, P.Maechler, Diabetes-associated mitochondrial DNA mutation A3243G impairscellular metabolic pathways necessary for beta cell function, Diabetologia 49 (8)(2006) 1816–1826.

[32] R.K. Shepherd, N. Checcarelli, A. Naini, D.C. De Vivo, S. DiMauro, C.M. Sue, Measure-ment of ATP production in mitochondrial disorders, J. Inherit. Metab. Dis. 29 (2006)86–91.

[33] K. Luangtrakool, F.Y. Tate, R. Shepherd, S. Campbell, C.M. Sue, P. Lertrit, Characteriza-tion of 2-deoxy-D-glucose uptake in fibroblast cultures derived from patients withA3243G mitochondrial DNA mutation, Southeast Asian J. Trop. Med. Public Health39 (2008) 745–752.

[34] V. Desquiret-Dumas, N. Gueguen, M. Barth, A. Chevrollier, S. Hancock, D.C. Wallace,P. Amati-Bonneau, D. Henrion, D. Bonneau, P. Reynier, V. Procaccio, Metabolicallyinduced heteroplasmy shifting and l-arginine treatment reduce the energetic in aneuronal-like model of MELAS, Biochim. Biophys. Acta 1822 (2012) 1019–1029.

[35] J. Ye, Characterisation of a family of algorithms for generalized discriminant analysison undersampled problems, J. Mach. Learn. Res. 6 (2005) 483–502.

[36] A.W. Bowman, A. Azzalini, Applied smoothing techniques for data analysis, OxfordStatistical Science Series, 1997.

[37] F.J. Massey Jr., The Kolmogorov–Smirnov test for goodness of fit, J. Am. Stat. Assoc.46 (253) (1951) 68–78.

[38] S.K. Chapman, G.A. Reid, Flavoprotein protocols,series, Methods in MolecularBiology, Humana Press, New Jersey, 1999.

[39] C. Frezza, L. Zheng, D.A. Tennant, D.B. Papkovsky, B.A. Hedley, G. Kalna, D.G. Watson,E. Gottlieb, Metabolic profiling of hypoxic cells revealed a catabolic signaturerequired for cell survival, PLoS ONE 6 (9) (2011), e24411.

[40] S.J. Ruff, D.E. Ong, Cellular retinoic acid binding protein is associated with mitochon-dria, FEBS Lett. 487 (2) (2001) 282–286.

[41] L.E. Lamb, T. Ye, N.M. Haralampus-Grynaviski, T.R. Williams, A. Pawlak, T. Sarna, J.D.Simon, Primary photophysical properties of A2E in solution, J. Phys. Chem. B 105(46) (2001) 11507–11512.

[42] J. Eng, R.M. Lynch, R.S. Balaban, Nicotinamide adenine dinucleotide fluorescencespectroscopy and imaging of isolated cardiac myocytes, Biophys. J. 55 (1989)621–630.

[43] E.M. Nuutinen, Subcellular origin of the surface fluorescence of reduced nicotin-amide nucleotides in the isolated perfused rat heart, Basic Res. Cardiol. 79 (1984)49–58.

[44] Q. Yu, A.A. Heikal, Two-photon autofluorescence dynamics imaging reveals sensitiv-ity of intracellular NADH concentration and conformation to cell physiology at thesingle-cell level, J Photochem. Photobiol. B. 95 (1) (2009) 46–57.

[45] T. Pérez-Ruiz, C. Martínez-Lozano, A. Sanz, E. Bravo, Determination of riboflavin,flavin mononucleotide and flavin adenine dinucleotide in biological tissues bycapillary zone electrophoresis and laser-induced fluorescence detection, Electro-phoresis 22 (6) (2001) 1170–1174.

[46] B. Chance, M. Baltscheffsky, Spectroscopic effects of adenosine diphosphate uponthe respiratory pigments of rat-heart-muscle sarcosomes, Biochem. J. 68 (2)(1958) 283–295.

[47] S.R. Piersma, A.J. Visser, S. de Vries, J.A. Duine, Optical spectroscopy ofnicotinoprotein alcohol dehydrogenase from amycolatopsismethanolica: a compar-ison with horse liver alcohol dehydrogenase and UDP-galactose epimerase,Biochemistry 37 (9) (1998) 3068–3077.

[48] J. Lakowicz, Principles of Fluorescent Spectroscopy, 3rd Edition Springer, Berlin,2006.

[49] C.C. Fjeld, W.T. Birdsong, R.H. Goodman, Differential binding of NAD+ and NADHallows the transcriptional co-repressor carboxyl-terminal binding protein to serveas a metabolic sensor, Proc. Natl. Acad. Sci. U. S. A. 100 (16) (2003) 9202–9207.

[50] M.R. Kasimova, J. Grigiene, K. Krab, P.H. Hagedorn, H. Flyvbjerg, P.E. Andersen, I.M.Møller, The free NADH concentration is kept constant in plant mitochondria underdifferent metabolic conditions, Plant Cell 18 (3) (2006) 688–698.

[51] T.E. Gundersen, Methods for detecting and identifying retinoids in tissue, J.Neurobiol. 66 (7) (2006) 631–644.

[52] N.M. Haralampus-Grynaviski, L.E. Lamb, C.M.R. Clancy, C. Skumatz, J.M. Burke, T.Sarna, J.D. Simon, Spectroscopic and morphological studies of human retinallipofuscin granules, Proc. Natl. Acad. Sci. U. S. A. 100 (6) (2003) 3179–3184.

[53] K. Majamaa, H. Rusanen, A. Remes, I.E. Hassinen, Metabolic interventions againstcomplex I deficiency in MELAS syndrome, Mol. Cell. Biochem. 174 (1997) 291–296.