Brain cytochrome-c-oxidase as a marker of mitochondrial function: a pilot study in major depression using NIRS

Abstract

Background:

Brain mitochondrial dysfunction is implicated in the pathophysiology of mood disorders. Brain cytochrome-c-oxidase (COX) activity is associated with mitochondrial function. Near-infrared spectroscopy (NIRS) non-invasively measures oxidized COX (oxCOX) and tissue oxygenation index (TOI) reflecting cerebral blood flow and oxygenation.

Methods:

oxCOX and TOI were assessed in prefrontal cortex (Fp1/2, Brodmann area 10) in patients in a major depressive episode (N=13) with major depressive disorder (MDD; N=7) and bipolar disorder (BD; N=6) compared to controls (N = 10). One MDD patient and all BD patients were taking medications. Computational modelling estimated oxCOX and TOI related indices of mitochondrial function and cerebral blood flow, respectively.

Results:

oxCOX was lower in patients than controls (p=0.014) correlating inversely with depression severity (r=−0.72, p = 0.006), driven primarily by lower oxCOX in BD compared with controls. Computationally modelled mitochondrial parameters of the electron transport chain, such as the nicotinamide adenine dinucleotide ratio (NAD⁺/NADH p=0.001) and the proton leak rate across the inner mitochondrial membrane (k_lk2 p=0.008), were also lower in patients and correlated inversely with depression severity. No such effects were found for TOI.

Conclusions:

In this pilot study, oxCOX and related mitochondrial parameters assessed by NIRS indicate an abnormal cerebral metabolic state in mood disorders proportional to depression severity, potentially providing a biomarker of antidepressant effect. Since the effect was driven by the medicated BD group, findings need to be evaluated in a larger, medication-free population.

1. Introduction

More than 300 million people suffer from mood disorders globally. Mood disorders are projected to be the leading cause of disability-adjusted life years lost by 2030 (NIH 2018). Heterogeneity of depression is indicated by pleomorphism of its symptom and the observations that up to 60% of patients do not respond adequately to medications (Nemeroff 2007; Wiles, Thomas, and Abel A 2014). Multiple neurobiological and psychosocial factors are thought to contribute to mood disorders, yet biomarkers that can guide treatment selection have proven limited (Strawbridge, Young, and Cleare 2017).

Mitochondrial dysfunction has been proposed as a contributing factor to the pathophysiology of mood disorders (Bansal and Kuhad 2016; Kato 2017). Mitochondria are intracellular organelles that produce adenosine triphosphate (ATP), the main source of cellular energy (Picard, Wallace, and Burelle 2016). Impaired mitochondrial function results in decreased ATP production, and more apoptosis and oxidative stress (Fišar and Hroudová 2010; Hroudová and Fišar 2011). ATP deficiency and the resulting alterations in energy metabolism may explain some features of mood disorders, e.g., physiological (sleep disturbances, fatigue), psychological (depressed mood, lack of motivation) and neurocognitive (impaired concentration and memory) abnormalities (E. Streck et al. 2014; Karabatsiakis et al. 2014). Primary mitochondrial disorders are associated with higher rates of depression (Anglin et al. 2012).

Prior to generation of ATP, oxidative phosphorylation diverts electrons extracted from nutrients into a transmembrane proton gradient, a process mediated by the electron transport chain (ETC) (Li et al. 2006). Cytochrome-c-oxidase (COX) is the final enzyme in the ETC, which is responsible for the maintenance of the transmembrane proton gradient that is used for ATP synthesis (Wikstrom 1977). Based on its rate-limiting role in this oxidative process (Arnold 2012), COX is considered a key marker of mitochondrial function (Srinivasan and Avadhani 2012).

Lower COX levels have been previously reported in major depressive disorder (MDD) and bipolar disorder (BD) based on peripheral blood measures, muscle biopsies and postmortem brain samples (for review see (L. Holper, Ben-Shachar, and Mann 2018)). In MDD, lower COX enzyme activity has been reported in peripheral platelets (Ben-Shachar et al. 1999) and muscle (Gardner et al. 2003, 2008) but greater COX mRNA expression has been found in serotonergic neurons of postmortem nucleus raphe in depressed suicide decedents (Sanchez-Bahillo et al. 2008). In BD, low COX enzyme activity, mRNA and protein expression have been reported in peripheral blood (Ben-Shachar et al. 1999; de Sousa et al. 2015; Naydenov AV et al. 2007; Washizuka et al. 2005; Beech et al. 2010), postmortem hippocampus (Altar et al. 2005; Konradi C et al. 2004) and postmortem frontal cortex (Iwamoto, Bundo, and Kato 2005; H. K. Kim et al. 2016; X. Sun et al. 2006). Further, there is evidence of abnormal mitochondrial function in vivo in the brain using neuroimaging methods. Specifically, 18F-fluorodeoxyglucose positron emission tomography (18F-FDG-PET) reveals metabolic changes in glucose utilization primarily in MDD that generally involves lower glucose uptake in dorsolateral prefrontal cortical and increased uptake in subcortical and ventral prefrontal cortex (De Crescenzo et al. 2017; Su et al. 2014; Schöll, Damián, and Engler 2014). Phosphorus (31)P magnetic resonance spectroscopy (MRS) has been used to assess decreased phosphocreatine as an indicator of mitochondrial oxidative phosphorylation, for example in frontal lobe, primarily in BD (for reviews see (Kato 2017; Quiroz et al. 2008; Clay, Sillivan, and Konradi 2011; Stork and Renshaw 2005; Moretti, Gorini, and Villa 2003)). However, while these techniques can assess some aspects of mitochondrial functions, they cannot measure COX activity directly.

To investigate COX activity in the present study, we used near-infrared spectroscopy (NIRS), an optical neuroimaging method. Compared to PET and MRS, NIRS is non-invasive, relatively low-cost, does not involve radiation, and can be applied in a physically minimally demanding setting (Obrig and Villringer 2003; Ferrari and Quaresima 2012; Wolf, Ferrari, and Quaresima 2007; Ferrari, Mottola, and Quaresima 2004). In our previous work, we showed that the COX signal has adequate test-retest reliability (Lisa Holper and Mann 2018). Consequently, NIRS permits highly feasible assessment of brain COX as a marker of mitochondrial function.

NIRS measures the oxidative state of COX’s chromophore (its binuclear copper Cu_A center), termed oxCOX signal, as well as hemodynamic signals in terms of oxygenated (HbO₂) and deoxygenated (HHb) hemoglobin. Compared to the hemodynamic signals routinely assessed using NIRS (Ferrari and Quaresima 2012), the oxCOX signal has hitherto not been routinely assessed. While HbO₂ and HHb provide information on blood flow and oxygenation, oxCOX is considered a key indicator of cellular oxygen metabolism (Ilias Tachtsidis et al. 2009; Highton et al. 2014; Kolyva et al. 2013; Ghosh et al. 2013). The concentration of oxCOX is approximately 10% that of hemoglobin, which makes it harder to detect. On the other hand, oxCOX is a stable (Bale, Elwell, and Tachtsidis 2016) and reliable signal (Lisa Holper and Mann 2018) with high brain specificity (de Roever et al. 2016). Thus, alterations in oxCOX indicate a change in mitochondrial metabolic balance (Bainbridge et al. 2014).

Previously, it has been shown that the oxCOX signal is a potential marker of mitochondrial function (Caldwell, Hapuarachchi, et al. 2015). Specifically, lower oxCOX activity, as hypothesized in major depressive episodes (MDE), can be related to reductions in mitochondrial parameters derived from computational modelling (Caldwell, Hapuarachchi, et al. 2015). In the present pilot study, we hypothesized that the oxCOX signal would be lower in MDE compared with control subjects.

2. Methods and materials

2.1. Study population

Patients with either DSM-IV MDD or BD were recruited at the New York State Psychiatric Institute (NYSPI)/Columbia University Medical Center. Patients were compared with physically and psychiatrically healthy controls (HC). All participants were assessed diagnostically using the Structured Clinical Interview for Axis I disorders (SCID-I/P) (First et al. 2002), the 24-item Hamilton Depression Rating Scale (HDRS) (Hamilton 1960) and the Young Mania Rating Scale (YMRS) (Young et al. 1978). Exclusion criteria for all participants included any uncontrolled medical condition, history of head trauma, any other major psychiatric disorder such as schizophrenia, schizoaffective disorder, current psychotic depression, substance abuse within the previous three months or substance dependence within the previous six months, a history of intolerable side effects from both citalopram and fluoxetine, a first-degree family history of schizophrenia, current serious suicidal ideation or behavior, pregnancy or currently lactating women. The Institutional Review Board of NYSPI approved the study protocol, and all participants gave written informed consent. All patients were taking part in other study protocols at NYSPI, explaining the variation in medication regimens, and why the MDD group was older than the other groups.

2.2. Respiratory challenges

Quantification of the oxCOX signal demands significant changes in arterial oxygen saturation to be reliably detected in the cortex. We therefore chose two respiratory challenges to systemically manipulate cerebral oxygenation via hypo- and hypercapnia, i.e., reduced and increased carbon dioxide in the blood, with the order of the challenges counterbalanced between participants (L. Holper, Scholkmann, and Seifritz 2015; Kolyva et al. 2014). Hypocapnia was induced by hyperventilation consisting of five repetitions of alternate periods of rapidly breathing in and out with constant respiratory volume (20 seconds) and normal breathing (40 seconds). Hypercapnia was induced by breath-holding consisting of five repetitions of alternate periods of breath holding (20 seconds) and normal breathing (40 seconds). Prior to recording, participants were trained to perform the inspirational breath-holding a volume of air, similar to a normal breath cycle, in order to avoid inhaling larger volumes of air than that of a normal breath cycle. Before and after each challenge, resting-states of five minutes of data were collected, during which participants were asked to sit still in order to allow the hemodynamic cortical system to normalize. The resting-state periods were not included in the data analysis.

2.3. NIRS instrumentation

A NIRO 300 instrument (Hamamatsu Photonics) was used to measure the oxCOX, HbO₂ and HHb signals based in four discrete wavelengths (775, 810, 850, and 910 nm) (Ilias Tachtsidis et al. 2009, 2004; I Tachtsidis et al. 2007). Although, more recent broadband NIRS systems may obtain a more robust oxCOX signal (Bainbridge et al. 2014; Bale et al. 2014; Arifler et al. 2015), we have reported adequate signal detection and test-retest reliability with NIRO 300 (Lisa Holper and Mann 2018). The NIRO 300 measures concentration changes of oxCOX, HbO₂ and HHb based on conventional differential spectroscopy with the use of the modified Beer-Lambert law (Patterson, Chance, and Wilson 1989; Delpy et al. 1988a). In addition, the system provides the tissue oxygenation index (TOI) based on the principle of spatially resolved reflectance spectroscopy (Delpy et al. 1988b; Matcher et al. 1995), which is in essence the proportion of HbO₂ relative to total hemoglobin (HbO₂ + HHb) concentration (Susumu Suzuki et al. 1999; Ilias Tachtsidis et al. 2008). The system uses four light sources (pulsed laser diodes) and three detectors (photodiodes) held in a light-proof holder, set at a distance of 4.5 cm, fixed in a rubber probe to allow for direct skin contact. Two channels were positioned to cover parts of the prefrontal cortex (i.e., Fp1/2, Brodmann area 10). Data were detrended, zero-meaned, and bandpass-filtered in the range 0.005 – 0.3 Hz, using a 5^th order Butterworth filter, to remove physiological noise, based on functions of the HOMER2 processing package (Huppert et al. 2009). Peripheral physiological measures were estimated using a capnometer (LifeSense LS1–9R, Nonin Medical) by means of the arterial tissue oxygen saturation (SpO₂) and the partial end-tidal carbon dioxide (PetCO₂) of the exhaled air. Participants wore a nasal cannula through which tidal gases were sampled at 1 Hz.

2.4. Computational modelling

The computational procedures using the BS model outlined below regarding sensitivity analysis, model fit and model parameters are described in detail in previous work (Caldwell, Hapuarachchi, et al. 2015). The modelling of the oxCOX signal was our a priori primary metabolic signal of interest, because it provides important information about mitochondrial metabolic state related to reduced COX activity (Caldwell, Hapuarachchi, et al. 2015). The modelling of the TOI signal was our a priori secondary hemodynamic signal of interest (Caldwell, Hapuarachchi, et al. 2015). All modelling was performed separately for hypo-and hypercapnia, but results are presented averaged across both conditions, since no significant differences were found, as expected.

2.4.1. BS model

The BS model, described elsewhere (Caldwell, Hapuarachchi, et al. 2015), models metabolism and cerebral blood flow in the human adult brain. Simulations were carried out using the Brain/Circulation Model Developer (Banaji et al. 2008) implemented in Python (Python Software Foundation). The model was applied with the following variables: 1) Input signals. The two model inputs were the experimentally measured PetCO₂ and SpO₂. 2) Output signals. The two model outputs were oxCOX and TOI. 3) Model parameters. Based on the input signals, the model simulates the behavior of the output signal based on a large parameter space (N = 139) reflecting cerebral metabolism. Simulations were run individually for each patient and control participant.

2.4.2. Sensitivity analysis and optimization

The 20 most influential model parameters for the oxCOX and TOI signals (Table 2) were selected for an exploratory sensitivity analysis using the extended Fourier Amplitude Sensitivity Test (eFAST) (Cukier et al. 1973; Schaibly and Shuler 1973). eFAST calculates the unique contribution of each model parameter in determining the model output based on variance decomposition, i.e., the first-order sensitivity index (S1) of a given parameter i, and a summed sensitivity index of the entire complementary set of parameters (all parameters except i), i.e., the total-order sensitivity indices (ST). The larger the S1, the more influential a parameter is on the output, while larger ST indicates a greater degree of interaction of a parameter i with other parameters.

Table 2. Optimized model parameters.

The 20 most influential model parameters on oxCOX and TOI included in the optimization are shown with initial values and ranges before optimization. Parameters are ordered from top to bottom according to their sensitivity to oxCOX and TOI, respectively.

	oxCOX parameters	Unit	Initial value	Range
oxCOX signal	Oxidized cytochrome-c-oxidase assessed by NIRS
CuA,o,n	Normal concentration of oxidized CuA	mM	0.066	0.046–0.085
Volmit	Fraction of brain tissue volume that is mitochondria	-	0.067	0.054–0.080
klk2	Constant controlling depending of leak rate Llkon Δp	mV−1	0.038	0.030–0.456
Δpn	Normal value of Δp (proton-motive force)	mV−1	168.611	118.028–219.194
NADn/NADHn	Normal NAD+/NADH ratio	-	9	0.1–20
k1,n	Forward reaction rate for CuAreduction at normal Δp/NADH	s−1	8.3	5.810–10.790
ck1	Parameter controlling sensitivity of k1to Δp	mV−1	0.01	0.008–0.012
Tmax,0	Maximal muscular tension under normal regulatory stimulus	mmHg cm	2.11	1.477–2.743
vCO2,n	Normal filtered carbon dioxide partial pressure	mmHg	40	28–52
ϕ	Oxygen concentration at half-maximal saturation	mM	0.036	0.029–0.043
rm	Vessel radius at which muscular tension is maximal	cm	0.027	0.022–0.032
kunc	Parameter simulating effect of uncouplers to the system	-	1	0.8–1.2
COXtis	Concentration of COX in tissue	mM	0.006	0.004–0.007
Llk0	Constant controlling the depending of the leak rate Llk on Δp	mM s−1	0.004	0.003–0.006
ptot	Total protons moved out of mitochondrial matrix	-	20	18–20
Kσ	Parameter controlling sensitivity of σe to vessel radius	-	10	7–13
CVinh	Control parameter representing action of CCO inhibitors	-	1	0.8–1.2
vu,n	Normal filtered demand	-	1	0.7–1.3
h0	Thickness of blood vessel walls at which radius is r0	cm	0.003	0.002–0.004
Pic	Intracranial pressure	mmHg	9.5	7–11
	TOI parameters	Unit	Initial value	Range
TOI signal	Tissue oxygenation index assessed by NIRS
Tmax,0	Maximal muscular tension under normal regulatory stimulus	mmHg cm	2.109	1.477–2.743
D02	Diffusion rate for oxygen between capillaries and mitochondria	s−1	0.842	0.589–1.094
rt	Radius in the muscular tension relationship	cm	0.018	0.014–0.022
ϕ	Oxygen concentration at half-maximal saturation	mM	0.036	0.029–0.043
vCO2,n	Normal filtered carbon dioxide partial pressure	mmHg	40	28–52
rn	Normal effective blood vessel radius	cm	0.019	0.015–0.022
Kσ	Parameter controlling sensitivity of σe to vessel radius	-	10	7–13
Volv	Relative venous blood volume	-	0.75	0.525–0.975
klk2	Constant controlling depending of leak rate Llkon Δp	mV−1	0.038	0.030–0.456
vO2,n	Normal filtered capillary oxygen concentration	mM	0.064	0.045–0.084
kaut	Overall functioning of autoregulatory response	-	1	0–1
NADn/NADHn	Normal NAD+/NADH ratio	-	9	0.1–20
vu,n	Normal filtered demand	-	1	0.7–1.3
σe,0	Parameter in elastic tension relationship	mmHg	0.143	0.114–0.171
RCO2	Autoregulatory reactivity to carbon dioxide	-	2.2	1.5–2.9
ε0NADH	Standard redox potential for NADH	mV	−320	−384−-256
Δpn	Normal value of Δp (proton-motive force)	mV−1	168.611	118.028–219.194
COXtis	Concentration of COX in tissue	mM	0.006	0.004–0.007
Llk0	Constant controlling depending of leak rate Llk on Δp	mM s−1	0.004	0.003–0.006
p2	Proton cost of reaction reducing a3	−	4	2.8–5.2

Optimization of all 20 parameters was carried out using a Global Problem (GLP) solver (solver: de, maximum iterations 1000) to minimize the difference (Euclidean distance) between the measured and modelled signals on which the corresponding parameters were influential. Parameters for each patient and control participant were optimized at once, in order to reduce the dimensionality of the optimizations (Caldwell, Moroz, et al. 2015). The initial model values (before optimization) and their ranges used for optimization are given in Table 2.

2.4.3. Statistical analysis

Statistical analysis was carried out using Matlab (Mathworks, Verision 2017b).

Preconditions for parametric testing, such as normality using the Lilliefors test and homogeneity of variance using O’Brien’s modification of Levene’s test, were checked for both oxCOX and TOI signals and their related parameter space.

Changes in the absolute block-averaged oxCOX and TOI signals were computed from baseline, i.e., time-locked two seconds before onset of hypo-/hypercapnia, to challenge, i.e., time-locked two seconds after offset of hypo-and hypercapnia.

Changes in the model parameters were expressed as percentage change (%) of the optimized values (after optimization) for each patient and control relative to the initial model values (before optimization).

$$\text{Percentage}\text{Change}\left(\text{\%}\right)=\frac{OptimizedValue-InitialValue}{\left|InitialValue\right|}\text{*}100$$ (1)

Linear mixed-effect regression was used to statistically assess the effect of patient versus control group (fixed effects) on the signal changes and corresponding optimized model parameters (responses), while controlling for age, PaCO₂ and SaO₂ (random effects) to account for normal aging and systemic effects; note that repeating the ANOVA without controlling for those covariates did not change results. ANOVA with Bonferroni correction was applied to account for multiple comparisons.

Spearman correlation coefficients were calculated to assess the relation of the signal changes and corresponding optimized model parameters to depression severity as assessed by the HDRS.

3. Results

3.1. Study population

Demographics of patients and controls are listed in Table 1. Higher HDRS scores (main effect F₂ = 26.43, p < 0.0001) were seen in MDD (22.6 ± 11.6) and BD (23.5 ± 3.6) patients compared to controls (2.3 ± 2.2). In the MDD group, 6/7 of patients had late-life depression (LLD) and were thus older (64.4 ± 17.2) than the control (32.3 ± 10.06) and BD groups (40.8 ± 12.4) (main effect F₂ = 16.57, p < 0.0001; post-hoc MDD vs. HC p < 0.0001; BD vs. HC p = 0.199, MDD vs. BD p = 0.007). Medication status also differed between groups. While 6/7 MDD subjects were medication-free, one MDD patient had started taking citalopram (20 mg/d) less than two weeks before scanning. All BD patients were on a consistent dose of a mood stabilizer at scanning (1 × lithium 1350mg/d, 4 × valproic acid 750–1000mg/d, 1 × lamotrigine 250mg/d) and 5/6 had started fluoxetine (20mg/d) less than two weeks prior to scanning (range 1 – 13 days).

Table 1. Demographics.

Patients with major depression disorder (MDD, N = 7) and bipolar disorder (BD, N = 6) compared to controls (N = 10). m = male, f = female. HDRS = Hamilton Depression Rating Scale. Medication: escitalopram (20mg/d), fluoxetine (20mg/d), lamotrigine (250mg/d), lithium (1350mg/d), valproic acid (750–1000mg/d).

Subjects	Sex	Age	HDRS	Previous episodes	Medication	Race	Education (years)
MDD	f	30	1	0	none	More than one race	17
MDD	f	63	39	3	none	Black or African American	21
MDD	f	63	23	1	none	Black or African American	16
MDD	f	63	31	1	none	Black or African American	13
MDD	f	79	19	1	escitalopram	White	18
MDD	f	70	26	1	none	White	17
MDD	m	83	18	0	none	White	12
BD	f	44	21	0	valproic acid, fluoxetine	White	16
BD	f	29	23	0	lamotrigine, fluoxetine	White	15
BD	m	56	19	0	lithium	White	16
BD	m	45	23	1	valproic acid, fluoxetine	Black or African American	17
BD	f	23	29	3	valproic acid, fluoxetine	White	16
BD	m	48	26	0	valproic acid, fluoxetine	White	17
Control	m	52	0	0	none	Black or African American	13
Control	f	28	5	0	none	White	14
Control	m	41	0	0	none	White	16
Control	f	26	1	0	none	Black or African American	19
Control	m	23	3	0	none	Asian	17
Control	f	24	6	0	none	White	16
Control	f	31	1	0	none	Black or African American	16
Control	f	28	5	0	none	White	-
Control	f	45	0	0	none	White	16
Control	f	25	3	0	none	Black or African American	16

3.2. Preconditions for parametric testing

The preconditions for parametric testing were fulfilled for both oxCOX and TOI signals and their related parameter space (not reported), including normality as assessed using the Lilliefors test (oxCOX signal: BD k-stats = 0.25, p = 0.287; MDD k-stats = 0.40, p = 0.001; HC k-stats = 0.17, p = 0.500; TOI signal: BD k-stats = 0.22, p = 0.468; MDD k-stats = 0.30, p = 0.168; HC k-stats = 0.22, p = 0.195) and homogeneity of variance between groups as assessed using O’Brien’s modification of Levene’s test (oxCOX signal F = 1.62, p = 0.222; TOI signal F = 0.04, p = 0.961).

3.3. Sensitivity analysis and optimization

In line with previous work (Caldwell, Hapuarachchi, et al. 2015), only a small part of the optimized model parameter space dominated the influence on the oxCOX and TOI signals, which space we selected for optimization. All parameters optimized reflected aspects of mitochondrial function and blood flow. The names and definitions of the parameters as well as the order of their sensitivity to the corresponding signals are given in Table 2.

3.4. Signal changes in oxCOX and TOI

Results revealed typical patterns of response to the respiratory challenges with hypocapnia producing a decrease in oxCOX and TOI and hypercapnia producing an increase in oxCOX and TOI (Bale, Elwell, and Tachtsidis 2016). The absolute block-averaged signal changes across both challenges revealed smaller oxCOX responses in the MDE groups (ANOVA main effect MDE vs. HC F_2,20 = 7.22, p = 0.014), indicating less efficient COX function compared to controls (Figure 1, Top). This effect correlated strongly with depression severity across the MDE groups (Spearman, r = −0.72, p = 0.006) (Figure 2). By contrast, the hemodynamic signal TOI revealed no significant differences between groups (MDE vs. HC F_2,20 = 0.50, p = 0.487) and no correlation with depression severity (Table 3).

Figure 1. Signal and parameter changes.

Boxplots illustrating the oxCOX and TOI signal changes and the parameters optimized for the oxCOX signals being significantly different between patients and controls as assessed using ANOVA. Note that the parameters values, and thus the percentage changes (%), are estimates and not true physical quantities. Note that the oxCOX and TOI signals on top assume an arbitrary 50% change in controls. Parameters are defined in Table 2 and statistics are reported in Table 3.

Figure 2. Correlation oxCOX parameters and HDRS.

Scatter plots of the correlation between depression severity and selected parameters optimized for the oxCOX signal assessed using Spearman correlation, i.e., Cu_A,o,n (r = −0.72, p = 0.006), k_lk2 (r = −0.63, p = 0.022) and COX_tis (r = −0.58, p = 0.039). These parameters correlated inversely with increasing depression severity indicating lower oxidative properties with higher symptom severity. Note that the parameters values, and thus the percentage changes (%), are estimates and not true physical quantities. Parameters are defined in Table 2 and statistics are reported in Table 3.

Table 3.

ANOVA for oxCOX and TOI and their related parameters. Main effects are shown between diagnoses within depressive episodes (MDE) versus controls, with post-hoc comparisons between major depressive disorder (MDD) and bipolar disorder (BD) versus controls; there were no significant differences between MDD and BD. A separate ANOVA tested medication effects for which main effects are shown between subjects receiving medication versus no medication, with post-hoc comparisons between mood stabilizers (MS) and antidepressants (AD) versus no medication (none). The covariate age had no significant effect. (Last two columns) Spearman correlation for oxCOX and TOI and their related parameters across MDD and BD patients with HDRS. Significant effects are highlighted (bold). Definitions of parameters can be found in Table 2.

	ANOVA diagnoses				ANOVA medication				Correlation
	Main effect		Post-hoc		Main effect		Post-hoc
	MDE vs controls		MDD vs controls	BD vs controls	Medication vs none		MS vs none	AD vs none	MDD/BD vs HDRS
	F2,20	p-value	p-value	p-value	F2,20	p-value	p-value	p-value	r	p-value
oxCOX signal	7.22	0.014	1.000	0.028	2.35	0.121	1.000	1.000	−0.72	0.006
CuA,o,n	9.86	0.005	0.435	0.012	1.36	0.280	1.000	1.000	−0.22	0.473
Volmit	1.66	0.212	0.678	0.459	0.98	0.391	0.352	0.508	0.42	0.155
klk2	8.72	0.008	0.760	0.017	1.62	0.224	1.000	0.685	−0.63	0.022
Δpn	0.57	0.457	1.000	1.000	0.70	0.509	1.000	1.000	−0.30	0.325
NADn/NADHn	14.18	0.001	0.627	0.002	3.68	0.044	1.000	0.732	−0.19	0.526
k1,n	3.47	0.076	1.000	0.156	3.34	0.056	0.137	0.054	−0.45	0.121
ck1	5.09	0.035	1.000	0.055	3.17	0.064	1.000	0.586	−0.23	0.444
Tmax,0	1.12	0.302	0.265	0.774	3.47	0.051	1.000	0.364	0.13	0.665
vCO2,n	0.09	0.771	1.000	1.000	0.88	0.429	0.398	0.554	0.11	0.719
ϕ	6.03	0.023	1.000	0.107	3.75	0.041	0.708	1.000	−0.01	0.971
rm	0.15	0.702	0.264	1.000	1.80	0.191	0.485	1.000	−0.35	0.246
kunc	0.64	0.434	0.688	0.838	0.37	0.697	0.854	1.000	−0.31	0.302
COXtis	3.96	0.060	1.000	0.101	2.43	0.113	1.000	0.243	−0.58	0.039
Llk0	0.69	0.417	0.352	0.895	1.06	0.364	0.496	1.000	−0.02	0.957
ptot	0.82	0.375	1.000	0.718	1.81	0.189	0.540	1.000	0.49	0.089
Kσ	0.01	0.907	1.000	1.000	0.52	0.604	0.677	0.687	0.12	0.692
CVinh	0.14	0.712	0.871	1.000	0.20	0.818	1.000	1.000	0.09	0.766
vu,n	1.28	0.271	1.000	0.563	1.82	0.188	0.283	1.000	−0.32	0.280
h0	0.00	0.986	1.000	1.000	0.23	0.799	1.000	1.000	0.23	0.444
Pic	2.04	0.168	0.614	0.369	0.52	0.605	0.803	1.000	−0.16	0.606
TOI signal	0.50	0.487	1.000	0.996	1.50	0.248	0.213	0.472	−0.27	0.369
Tmax,0	2.32	0.143	1.000	0.430	1.99	0.163	1.000	0.960	−0.29	0.331
D02	0.15	0.699	1.000	1.000	0.37	0.694	0.816	1.000	−0.04	0.901
rt	0.31	0.582	1.000	1.000	2.49	0.108	0.108	0.460	0.16	0.613
ϕ	0.06	0.811	1.000	1.000	0.20	0.824	1.000	1.000	0.18	0.681
vCO2,n	1.10	0.306	0.406	0.661	0.04	0.960	1.000	1.000	0.25	0.403
rn	2.53	0.126	0.756	0.181	3.08	0.068	1.000	0.216	−0.13	0.667
Kσ	5.71	0.026	1.000	0.074	2.06	0.154	1.000	1.000	−0.28	0.349
Volv	2.29	0.145	1.000	0.519	1.62	0.222	1.000	1.000	−0.09	0.779
klk2	2.20	0.153	0.703	0.287	1.46	0.257	0.316	0.857	0.04	0.685
vO2,n	0.07	0.791	0.777	1.000	0.31	0.739	1.000	0.885	0.24	0.424
kaut	1.00	0.329	0.205	0.515	0.12	0.891	1.000	1.000	−0.29	0.334
NADn/NADHn	0.52	0.480	1.000	0.943	3.11	0.067	0.091	0.578	−0.29	0.336
vu,n	0.44	0.513	0.669	0.916	1.29	0.297	1.000	0.493	0.34	0.255
σe,0	0.55	0.468	0.978	0.861	0.98	0.391	1.000	0.739	0.31	0.305
RCO2	0.08	0.780	0.445	1.000	1.17	0.330	0.967	0.393	0.30	0.318
ε0NADH	0.00	0.987	1.000	1.000	0.77	0.478	0.652	1.000	0.20	0.508
Δpn	0.02	0.876	0.925	1.000	0.49	0.619	1.000	0.701	0.49	0.086
COXtis	0.04	0.846	1.000	1.000	0.65	0.534	0.538	0.701	−0.11	0.729
Llk0	0.30	0.592	0.828	1.000	0.89	0.427	1.000	0.655	0.41	0.163
p2	3.00	0.098	1.000	0.163	3.19	0.063	0.177	0.045	0.40	0.173

Secondary ANOVA analyses assessing the effects of medication or age of the participants revealed no significant effects on either oxCOX or TOI (Table 3). Although the study was not powered to detect differences between MDD and BD groups, secondary analysis found no significant difference in oxCOX or TOI between MDD and BD groups. Nevertheless, the reduction in oxCOX in MDE groups compared to controls was driven primarily by lower oxCOX in the BD group, (Table 3).

3.5. Optimized model parameters for oxCOX and TOI

To provide a realistic picture of the differences in the parameters between patients and controls, the optimized values were expressed as percentage changes (%) relative to the initial value (Figure 1, Bottom). Statistical assessment using linear mixed-effect regression revealed that four model parameters optimized for the oxCOX signal were sensitive to the effects of depression. In particular, the parameters Cu_A,o,n (ANOVA main effect MDE vs. HC F_2,20 = 9.86, p = 0.005), k_lk2 (F_2,20 = 8.72, p = 0.007), NAD_n/NADH_n (F_2,20 = 14.18, p = 0.001), and c_k1 (F_2,20 = 5.09, p = 0.035) decreased in the MDE groups, whereas ϕ (F_2,20 = 6.03, p = 0.023) increased in the MDE groups compared to controls, all primarily accounted for by the differences between BD and controls (Table 3). There were no significant differences detected between MDD versus BD. A separate ANOVA assessing medication effects revealed modest significant effects indicating decreasing NAD_n/NADH_n (F_2,20 = 3.68, p = 0.044) and increasing ϕ (F_2,20 = 3.74, p = 0.041) with medication, however without significant differences post-hoc comparisons, which again were not powered adequately given the small number of cases (Table 3).

Some parameters also showed a moderate correlation with depression severity after adjusting for age (Figure 2). In particular, k_lk2 (r = −0.63, p = 0.022) and COX_tis (r = −0.58, p = 0.039) were inversely correlated with increasing depression severity indicating lower oxidative properties with higher symptom severity (Table 3).

Only one of the parameters optimized for the TOI signal, Kσ, showed a difference between MDE and controls (main effect MDE vs. HC F_2,20 = 5.71, p = 0.026) but without correlation with depression severity. None of the other parameters optimized for TOI showed group differences (all p > 0.05) (Figure 1) or correlations with HDRS score (all p > 0.05) (Figure 2, Table 3). ANOVA assessing medication effects revealed no significant effects on TOI parameters (Table 3).

3.6. Power analysis

Due to limitations of sample size, we performed a post-hoc power analysis using the pwr package (Champely 2018) in R (R Development Core Team 2008). Given the observed mean COX differences between HC and MDE groups of d = 1.13 (between HC and MDD d = 0.81, between HC and BD d = 1.28) with N = 23 subjects, we calculated a power of 0.73 for the comparison between HC versus MDE (0.34 for the comparison HC versus MDD, 0.64 for the comparison HC versus BD) at a significance level of 0.05 using t-test for independent samples. We asked what minimum difference between HC and MDE groups would have been detectable with our N = 23 subjects if we would have had a power of 0.8, and calculated a change of d = 1.24. Lastly, we ccomputed the number of participants that would be needed to detect a change similar in size of our observed effect (d = 1.13) with a power of 0.8, yielding at least N = 21 participants.

4. Discussion

This is the first study to evaluate oxCOX signal using NIRS in subjects experiencing major depressive episodes. The oxCOX signal is a proposed marker of mitochondrial health (Bale, Elwell, and Tachtsidis 2016). Our results indicate that both experimentally measured oxCOX signal and computationally modelled mitochondrial parameters are altered in mood disorders. There were two main findings. Firstly, experimentally measured oxCOX activity was lower in the MDE groups and correlated with depression severity (Figure 1 & 2). Secondly, optimized mitochondrial parameters for oxCOX relating to the electron transport chain detected lower activity in MDE, largely in the BD disorder group, when compared with HC, with lower activity correlating with depression severity (Figures 1 & 2). These parameters are potential mitochondrial contributors to an oxCOX marker. Combining oxCOX measurements derived from NIRS with computational modelling may therefore potentially enhance our understanding of which factors drive mitochondrial dysfunction in MDEs. Our power analysis indicated that the study was relatively underpowered, so these data should therefore be considered pilot results that require replication with a larger sample size. Confidence in the findings is thus tempered by the small sample size limiting statistical power for subgroup comparisons, such as between MDD and BD. Although the results appear to be largely driven by the BD subjects, despite the lack of a significant difference between MDD versus BD, we observed differences between MDE and controls and lack of a difference between BD and controls. Together, our results suggest that mitochondrial abnormalities that have been previously reported in the periphery (L. Holper, Ben-Shachar, and Mann 2018) may be present in the brain in mood disorders.

The computational model used here cannot prove dysfunction of the modelled mitochondrial mechanisms, because the optimized parameters are estimates and not true physical quantities. The model can, however, approximate alterations in the mitochondrial mechanisms that in principle can produce the reduced oxCOX response observed in depressed patients (Caldwell, Hapuarachchi, et al. 2015). These mechanisms may reflect disease-specific information that may together be used as a marker of mitochondrial health in MDEs. We briefly highlight the main functions of the parameters analysed here, and discuss their relevance to current literature in mood disorders.

COX (also known as complex IV) is the last enzyme of the ETC within the process of oxidative phosphorylation at the inner mitochondrial membrane. Cu_A,o,n represents the normal concentration of its oxidized Cu_A center (i.e., after the loss of electrons), indicating the portion of COX already used for electron transfer, in contrast to the concentration of reduced Cu_A (Cu_A,r,n, i.e., after the gain of electrons), whereas COX_tis indicates its overall concentration in tissue. In the case of COX deficiency as part of the pathophysiology of depression, both Cu_A,o,n and COX_tis would be expected to decrease because of less functional COX availability, as observed in the present study (Figure 1). The correlation with depression severity corroborates our results suggesting that Cu_A,o,n and COX_tis can be related to physical quantities (Figure 2).

The NAD⁺/NADH ratio (nicotinamide adenine dinucleotide, NAD) is the product of complex I (NADH dehydrogenase) of the ETC. The two enzymes, complexes I and IV, are functionally closely related in that complex IV is required for the assembly and stability of complex I (Li et al. 2007; Diaz et al. 2006; Schäfer et al. 2007). Although, our model does not directly estimate complex I activity, we can use its product, the NAD⁺/NADH ratio, as an indicator of complex I functioning. The NAD⁺/NADH ratio defines the balance between oxidized NAD⁺ biosynthetic and reduced NADH consuming pathways, which plays a critical role in energy production and antioxidant defences. Due to its proton pumping function at the beginning of the ETC (Schäfer et al. 2007; Rich 2003; Belenky, Bogan, and Brenner 2007), the NAD⁺/NADH ratio can be thought of as the fuel of the ETC (Belenky, Bogan, and Brenner 2007; Houtkooper et al. 2010; Lin and Guarente 2003). A decrease of the NAD⁺/NADH ratio can lead to the formation of reactive oxygen species that potentially contribute to mitochondrial damage (Lin and Guarente 2003; Murphy 2009; Ben-Shachar and Karry 2008). In the case of deficient COX as part of the pathophysiology of depression, the NAD⁺/NADH ratio would be expected to decrease, as observed in the present study (Figure 1). The correlation of the NAD⁺/NADH ratio with depression severity indicates that the scenario modelled here, could be related to physical quantities (Figure 2).

The potential importance of the interplay between NADH-dehydrogenase (complex I) and COX (complex IV) (Li et al. 2007) in depression is also supported by previous studies reporting less complex I and IV activity in MDD (Gardner et al. 2003, 2008; Ben-Shachar and Karry 2008) and BD (Washizuka et al. 2005; X. Sun et al. 2006; de Sousa et al. 2015; Beech et al. 2010; H. K.Kim et al. 2016; Konradi C et al. 2004; Altar et al. 2005; Ben-Shachar and Karry 2008; Iwamoto, Bundo, and Kato 2005; Naydenov AV et al. 2007), a reduced complex I/IV enzyme-ratio in muscle mitochondria of MDD patients (Gardner et al. 2003) as well as a reduction of the NAD⁺/NADH ratio in patients with first-episode BD as assessed using MRS (S.-Y. Kim et al. 2017). ¹

The fundamental proton pumping role of both complexes I and IV (Barney 2017) builds and maintains the transmembrane proton gradient used by ATP synthase to synthesize most of the cell’s ATP (Wikstrom 1977). The proton-motive force, Δp, resulting from that transmembrane proton gradient, promotes proton movement down the electrochemical potential across the mitochondrial membranes in two directions. The protons pumped out of the mitochondria are used by COX for electron transfer, and thus, reduction of O₂ to H₂O. Protons re-enter the mitochondria via two processes, either associated with ATP production or through endogenous proton leak channels (Nicholls 1977; Korzeniewski and Zoladz 2001). These proton leaks serve the important purpose of preventing dielectric membrane breakdown (in limiting Δp), and restricting leakage of single electrons from the ETC to form superoxide (Brand et al. 1994; Rolfe and Brand 1997). The parameters k_lk2 (the parameter controlling the sensitivity of the leak current to changes in Δp) and c_k1 (a parameter controlling the sensitivity of k₁ to Δp), correlated with depression severity or differed between patients and controls (Figures 1 & 2). Hence, a major endpoint of all of these processes is the maintenance of Δp, as it is key to mitochondrial survival (Storey 2004). Under conditions of deficient COX activity with insufficient proton pumping, the resultant collapse of Δp can lead to cells becoming ATP users instead of being ATP producers (Wikström et al. 2015). This ATP imbalance may be related to some of the pathophysiological manifestations of mood disorders such as fatigue, lack of motivation or working memory deficits (E. Streck et al. 2014; Karabatsiakis et al. 2014).

The current literature provides several hypotheses on the connection between mitochondrial dysfunction an mood disorders (McCann and Ross 2018), based on glucocorticoid-induced stress (Wallace 2005), oxidative stress (Bakunina, Pariante, and Zunszain 2015), and susceptibility to genetic alterations (Pei and Wallace 2018). The assumption that COX activity might be more deficient in BD compared with MDD, as our findings might suggest (Fig. 1–2), needs to be replicated in a future study. A study in a larger and medication-free sample of MDD and BD subjects would address this question. It also remains to be determined in BD whether this deficient COX activity is absent when the patient becomes euthymic and how it is impacted by a switch into mania.

Several methodological limitations should be considered when interpreting our findings. The sample size was small and the MDE groups were heterogenous with respect to diagnosis, age, and medication status. The majority (6/7) of the MDD patients were older than 60 years of age and met criteria for late-life depression (LLD). LLD differs from younger patients with MDD in both pathophysiology and clinical presentation (Glover and Srinivasan 2013). Since there is an age-dependent decline of mitochondrial functioning (N. Sun, Youle, and Finkel 2016; L. Holper, Ben-Shachar, and Mann 2018), the present results may not generalize to a younger MDD population. However, our findings of lower activity in MDE survived statistical age-adjustment and we therefore suggest that age may not be the main contributing factor for the observed reduced COX activity in MDE. The lack of effect of age on differences between MDE and HC groups may be explained by the fact that the group differences were largely driven by the BD group, and that BD did not have significant age differences from HCs. Of note, most of the patients were medicated. One of the MDD subjects was taking escitalopram, a selective serotonin reuptake inhibitor (SSRI). We note that this patient’s data were not outliers in the analysis. Escitalopram has not been shown to affect mitochondrial function in general or COX activity in particular (Gonçalves et al. 2012; Shetty et al. 2015), although other SSRIs may decrease mitochondrial function (Adzic et al. 2016; Hroudová and Fisar 2010). Further, 5/6 of BD patients had commenced antidepressant treatment with fluoxetine, another SSRI, within two weeks of scanning. Fluoxetine has so far not been reported to have a clear effect on COX but that may depend on sex (Adzic et al. 2013; Adzic, Mitic, and Radojcic 2017) and be brain region-specific (Shumake et al. 2010; Freo et al. 2000; Padilla et al. 2011). Lastly, all BD patients were receiving mood stabilizers at the time of scanning. Mood stabilizers, such as lithium or valproic acid, when given as long-term treatment have not been found to affect mitochondrial function (Kato 2017; Clay, Sillivan, and Konradi 2011), instead, they have been suggested to protect against mitochondria-targeted neurotoxicity (Bachmann et al. 2009) thereby preserving or even enhancing mitochondrial function in general (X. Sun et al. 2006) and COX activity in particular (Hroudová and Fisar 2010; Bachmann et al. 2009; E. L. Streck et al. 2015; Cancelier et al. 2017). Our results showed that these medications may had modest effects on oxCOX or related parameters (Table 3). Based on the literature and these findings, it is unlikely that the larger effects in BD are solely driven by an inhibition of COX by the small effects of SSRIs; nevertheless, our findings require further evaluation in medication-free depressed patients.

In conclusion, we found in this pilot study that the brain oxCOX signal was lower in MDE, and computational modelling identified alterations in key parameters of mitochondrial metabolism based on the impaired oxCOX signal. While this preliminary finding is consistent with measures of mitochondrial function in blood cells and muscle biopsies, and reports of less FDG uptake in dorsolateral prefrontal cortex in MDE, we cannot rule out an effect of antidepressants, or be certain that the finding is present in both MDD and BD. The small number of subjects as well as the intergroup heterogeneity is not sufficient for a robust conclusion. Future research therefore needs to replicate our findings in a larger, unmedicated mood disorder sample. Given the reported implication of mitochondrial abnormalities in BD and other psychiatric illnesses reported in blood, muscle and now here in brain, the study of mitochondrial metabolism using NIRS is an exciting, novel approach to disease etiology. Its correlation with depression severity suggests that oxCOX may be a potential biomarker of clinical response, and perhaps even a novel target of treatment.