Whole blood mitochondrial copy number in clinical populations with mood disorders: a meta-analysis

Abstract

Major depressive disorder (MDD) and bipolar disorder (BD), are globally prevalent, contributing to significant disease burden and adverse health outcomes. These mood disorders are associated with changes in many aspects of brain reward pathways, yet cellular and molecular changes in the brain are not readily available in clinical populations. Therefore, the use of biomarkers as proxies for changes in the brain are necessary. The proliferation of mitochondria in blood has emerged as a potentially useful biomarker, yet a clear consensus on how these mood disorders impact mitochondrial DNA copy number (mtDNAcn) has not been reached. To determine the current available consensus on the relationship of mood disorder diagnosis and blood mtDNcn, we performed a meta-analysis of available literature measuring this biomarker. Following PRISMA guidelines for a systematic search, 22 papers met inclusion criteria for meta-analysis (10 MDD, 10 BD, 2 both MDD and BD). We extracted demographic, disorder, and methodological information with mtDNAcn. Using the metafor package for R, calculated effect sizes were used in random effects or meta regression models for MDD and BD. Overall, our data suggest blood mtDNAcn may be a useful biomarker for mood disorders, with MDD and BD Type II associated with higher mtDNAcn, and BD Type I associated with lower mtDNAcn. Interestingly we observed a trending increase in mtDNAcn in patients with MDD, which reaches significance when one study with outlying demographic characteristics is excluded. Further, subgroup and meta-regression analysis indicated the relationship between mtDNAcn and diagnosis in patients with BD is dependent on BD type, while no relationship is detectable when BD types are mixed. Further study of blood mtDNAcn could predict downstream health outcomes or treatment responsivity in individuals with mood disorders.

Introduction

Human depressive disorders, including both major depressive disorder (MDD) and bipolar disorder (BD), affect millions across the globe every year, yet there are no currently accepted, reliable biomarkers for these illnesses. Recent studies have more intensively been investigating the energetic and oxidative changes associated with mood disorder across the brain and body, finding significant differences in mitochondrial functioning across altered mood states in response to chronic stress, and in the context of mood disorder diagnoses (Allen et al., 2018; Büttiker et al., 2023; Marazziti et al., 2012; Picard and McEwen, 2018a; Ridout et al., 2020). Mitochondria have emerged as both a tracker and an indicator of altered energetics and overall health across tissues, with changes signaling increased energetic demand, oxidative damage, or increased mitophagy (Garabadu et al., 2019; Picard et al., 2018; Sebastián and Zorzano, 2018). In the context of depressive disorders, both MDD and BD have been linked to mitochondrial adaptations in the brain and in the periphery (Allen et al., 2018; Büttiker et al., 2023; Guo et al., 2022; Marazziti et al., 2012; Scaini et al., 2016; Skokou et al., 2023; Verhoeven et al., 2018; Weger et al., 2020; Xie et al., 2017). In MDD patients depressive episodes are linked to higher levels of oxidative damage markers and lower levels of antioxidant compounds in serum or red blood cell samples (Liu et al., 2015). Altered mitochondrial gene expression has been observed in postmortem dorsolateral prefrontal cortex (Wang and Dwivedi, 2017), and increased reactive oxygen species, reduced antioxidant levels and inflammatory markers have been found across tissue types (Büttiker et al., 2023; Czarny et al., 2018; Klinedinst and Regenold, 2015). In postmortem tissue from BD patients, mitochondrial DNA copy number (mtDNAcn) is increased in the dorsolateral prefrontal cortex, where there is also an increase in deletions of mtDNA and a marked reduction in electron transport chain complex I activity (Das et al., 2022). In peripheral blood samples from first episode manic patients, mitochondrial complex I-related genes were increased at a transcriptional level compared to healthy control individuals (Akarsu et al., 2015). Heat shock proteins, including HSP60, HSP70 and manganese superoxide dismutase (MnSOD) are elevated in blood of patients with BD (Stertz et al., 2015; Wang et al., 2020), while glutathione peroxidase activity is reduced (Wang et al., 2020).

In recent years blood mtDNAcn has emerged as a potentially useful biomarker across disease classes (Castellani et al., 2020; Hubens et al., 2022; Picard, 2021; Yang et al., 2021). Mitochondrial DNA can be used both as a measure of mitochondrial content as well as an indicator of mitochondrial damage or disturbance (Clay Montier et al., 2009; Picard, 2021). Each cell contains multiple mitochondria, and within each mitochondrion are one or multiple copies of the mitochondrial genome. Mitochondrial DNA is particularly vulnerable to disruption, as it is not protected by protein histones like genomic DNA. Measuring the relative amount of mitochondrial DNA to nuclear DNA provides an approximate measure of the number of mitochondria per cell that can be compared across disease or treatment groups. Generally, a higher relative copy number is taken to indicate more mitochondria per cell or per nucleus. Changes in mtDNAcn have been demonstrated in diseases as disparate as various cancers (Afrifa et al., 2018), neurodegenerative disorders (Coppedè, 2024; Klein et al., 2021; Yang et al., 2021), psychosis (Das et al., 2022; Kumar et al., 2018), and substance use disorders (Caspani et al., 2021; Feng et al., 2013; Wang et al., 2021), indicating diagnostically useful relationships between this measure and disease states that may also exist in the context of mood disorders. Blood is readily accessible in human populations, and amenable to resampling after disease progression or treatment, making blood mtDNAcn specifically a robust candidate for a mood disorder biomarker. Both diagnoses of MDD and BD involve periods of time of low mood, increased anhedonia, and disruption of daily life. BD also includes periods of elevated mood, hyperactivity, and reduced need for sleep in either manic (Type I) or hypomanic (Type II) periods. While sharing many mood features, treatment and pharmacological interventions for each disorder are unique, and considering how the dynamics of a new biomarker vary between diagnoses with overlapping mood profiles is critical in understanding what disorder feature may be influencing the biomarker, here mtDNAcn.

For both MDD and BD, some studies have found increases in mtDNAcn, while others have found decreases or no change. As the field moves towards using mtDNAcn as a biomarker, the current meta-analysis seeks to compile recent data across available studies for each disorder to determine consensus in the literature regarding the correlative relationship between mood disorders and mtDNAcn. One such meta-analysis on mtDNAcn in BD was conducted in 2018 (Yamaki et al., 2018), however, at the time only five studies were available. We have now identified 22 papers examining mood disorders and blood mtDNAcn, with most having been published since 2018. Further, other metanalyses examining MDD patients have focused on circulating cell free mitochondrial DNA (Park et al., 2022), rather than whole blood or cell-based mtDNAcn. Therefore, the current analysis provides a comprehensive systematic search of available literature to determine if whole blood/cell-based mtDNAcn is a successful biomarker for specific mood disorders, and whether it is consistently elevated, depressed, or unchanged by these disorders, and how this compares to findings with cell-free mitochondrial copy number. This compilation of available findings along with our analysis serves as a guide to the field in determining the utility of mtDNAcn to accurately track mood disorder diagnosis, disorder progression, treatment responsivity, or other concurring biophysiological changes in the body and brain.

Methods and Materials

Systematic literature search

Meta-analysis procedures were conducted in accordance with PRISMA guidelines. To systematically search the available literature, a medical librarian created searches using PubMed (1809-present), Embase (embase.com; 1974-present), and Scopus (Elsevier; 1960-present). Each search was tailored to the database and contained both controlled vocabulary and keyword terms. Searches were required to include a term for each of three categories: (i) mood disorders; (ii) mitochondrial DNA copy number; and (iii) blood. See Appendix A for exact search terms. 87 unique references were retrieved on July 9, 2021 and imported into Covidence systematic review software (Veritas Health Innovation, Melbourne, Australia. Available at www.covidence.org) for screening. An additional search using the same parameters was conducted on June 9, 2022, and 20 unique citations were added to Covidence for screening. Four references that were not identified in either search but matched significant key words and procedures (Cai et al., 2015; Czarny et al., 2020; Kim et al., 2011; Ryan et al., 2023) were imported manually into Covidence for screening. Within Covidence, a title/abstract screen was conducted independently by two screeners. Titles/abstracts that received two ‘yes’ votes subsequently had their full text reviewed, and studies that received two ‘no’ votes were excluded from further analysis. Ties were broken by discussion between both screeners. Studies were included based on the inclusion and exclusion criteria described in Supplementary Table 1. Grey literature such as conference abstracts, systematic reviews, literature reviews, pilot studies, and any other sources that are relevant to the search but do not contain data extractable for meta-analysis quantification were excluded. After independent full text review, 22 studies met criteria for inclusion in the study and subsequently used for data extraction and analysis (Figure 1).

Figure 1:

Study selection flow chart. 172 studies were identified from database searches and duplicate studies were removed. After independent screening, 60 studies were deemed irrelevant, and 28 studies were excluded based on the displayed reasons. 4 additional studies were added to the analysis due to relevance to the experimental question. 22 studies were included for data extraction and final analysis.

Data extraction

A template for data extraction was created through Google Forms, and a team member populated the form with the key data points mentioned for each study. Data was then transferred to a separate spreadsheet that facilitated data visualization for the team statistician. The following key data points were extracted from the studies that met inclusion criteria: the specific mood disorder addressed in the paper (MDD, BD), geographical location of the study, demographics of the study participants (sex, age, race), the sample size of both the patient and control populations, tissue collected for blood analysis (ex: whole blood, leukocytes, or peripheral blood mononuclear cells (PBMCs)), methodology for determining mtDNAcn (genes tested) (Table 1). Information on participant recruitment, diagnostic evaluation tools used, and diagnostic inclusion criteria for defining the mood disorders examined was also extracted (Table 2). The corresponding author contact information from each study was extracted in the case that authors needed to be contacted for missing data or clarification; sample demographic information, and regression information was retrieved from email contact with Yamaki, et al. (Yamaki et al., 2018) to recover their supplemental information document.

Table 1:

Study demographic and experimental details

First Author Year (Ref)	Country	Disorder/Type	Sex M, F (total)	Age (mean ± SD)	Race	Tissue Source	mtDNAcn Measurement	Data Source (paper/transformation)
Papers including MDD patients
Cai 2015 European Sample (Cai et al., 2015)	UK	HC DeCC	108 ea. HC and MDD sex unsp.	42.24 DeCC	White European	Whole Blood	mt-CYB RNaseP	Q1, median, and Q3 of data were extracted from Figure 1 Converted to mean and SD
		MDD Unsp. DeCC and GENDEP		47.17 DeCC 42.5 ± 11.8 GENDEP
Chang 2015 (Chang et al., 2015)	Taiwan	HC	28, 42 (70)	38 ± 16.5a	Unspecified Assumed Asian	Leukocytes	mt-ND1 HBB	Reported n and m values t calculated from p value in Table 2
		MDD	12, 28 (40)	42 ± 18.75a
Chung 2019 (Chung et al., 2019)	South Korea	HC	46, 70 (116)	47.6 ± 16.7	Unspecified Assumed Asian	Unspecified Assumed Whole Blood	mt- CYB PK	Reported Mean and SD
		MDD	46, 72 (118)	47.7 ± 16.8
Czarny 2020 (Czarny et al., 2020)	Poland	HC	24, 36 (60)	45 ± 15.1	Unspecified Assumed Caucasian	PBMCs Isopycnic centrifugation	mt-ND1 mt-ND5 SLCO2B1 SERPINA1	Reported Mean and SD
		MDD	11, 14 (25)	49.44 ± 10.86
Fernström 2021 (Fernström et al., 2021)	USA CA	HC	5, 6 (11)	35.1 ± 11.1	Unspecified	PBMCs Ficoll centrifugation	mt-ND1 B2M mt-COX1 RnasP	Reported uncorrected mean and SD
		MDD	21, 26 (47)	35.0 ± 10.7
He 2014 (He et al., 2014)	China	HC	109, 108 (217)	30.8 ± 7.05	Asian Chinese	Whole Blood	mtDNA HBB	Reported mean and SD
		MDD	100, 110 (210)	30.2 ± 8.10
Kim 2011 (Kim et al., 2011)	South Korea	HC	86 F only	74.6 ± 6.46	Asian Korean	Whole Blood	mt - ND1 HBB	Reported Q1, median, Q3 Calculated mean and SD
		MDD	56 F only	74.8 ± 4.9
Lindqvist 2018 (Lindqvist et al., 2018)	USA CA	HC	22, 33 (45)	37.6 ± 13.9	Unspecified	PBMCs Ficoll centrifugation	mt-ND1 Ribonuclease P	Reported mean and SD
		MDD	23, 27 (50)	39.6 ± 14.7
Otsuka 2017 (Otsuka et al., 2017)	Japan	HC	248, 287 (535)	50.8 ± 17.5 (M) 51.0 ± 16.8 (F)	Asian Japanese	Whole Blood	mt-ND1 HBB	Reported n, m, and t values
		MDD	332, 176 (508)	49.9 ± 17.2 (M) 50.6 ±18.3 (F)
Tyrka 2016 (Tyrka et al., 2016)	USA RI	HC	50, 63 (113)	28.5 ± 9.2	241 – White 26 – Black 9 – Asian 10 - Other	Whole Blood	mtDNA HBB	Extracted adjusted mean and SD from Figure 2
		“Lifetime MDD” subset	(48)	Not Specified for Subset
Papers including both MDD and BD patients
Ryan 2023 (Ryan et al., 2023)	Ireland	HC	30, 62 (89)	53.42 ± 10.39	White/ Caucasian	Whole Blood	mt-ND1 mt-ND5 SLCO2B1 SERPINA1	Reported mean and SD
		Unipolar Depression	38, 62 (79 UD, 21 BD)	56.36 ± 14.28
		Bipolar Depression
Tsuji 2019	Japan	HC	29, 29	48.2 ±	Unspecified	Whole Blood	mt-ND1	Reported n, m, and t
(Tsujii et al., 2019)			(58)	13.1	Assumed Asian		HBB	values
		MDD	31, 13 (44)	42.9 ± 10.8
		BD	40, 39 (79)	43.4 ± 12.4
Papers including BD patients
Angrand 2021 (Angrand et al., 2021)	France	HC	100, 80 (180)	35.6 ± 13.2	Unspecified	Whole Blood	mt-ND1 EIF2C1	Reported mean and SD
		BD	164, 148 (312)	40.8 ± 14.2
Chang 2014 (Chang et al., 2014)	Taiwan	HC	28, 42 (70)	38 ± 16.5a	Unspecified Assumed Asian	Leukocytes	mt-ND-1 HBB	Extracted Q1, median and Q3 from Figure 1 Computed mean and SD
		BD1, euthymic	18, 22 (40)	41.5 ± 19a
Chang 2022 (Chang et al., 2022)	Taiwan	HC	27, 39 (66)	35.45 ± 10.99	Unspecified Assumed Asian	Leukocytes	mt-ND1 HBB	Reported mean and SD separated by sex
		BD unspecified	24, 36 (60)	37.9 ± 12.98
Chung 2022 (Chung et al., 2022)	South Korea	HC [BD1]	91, 84 (175)	35.5 ± 13.0	Unspecified Assumed Asian	Whole Blood	mt-CYB PK	Mean and SD extracted from Figure 1
		HC [BD2]	43, 53 (96)	29.2 ± 10.7
		BD1	98, 88 (186)	35.6 ± 13.0
		BD2	44, 51 (95)	29.2 ± 10.7
De Sousa 2014 (de Sousa et al., 2014)	Brazil	HC	14, 10 (24)	28.3 ± 7.3	Unspecified	Leukocytes	mt-DNA HBB	Reported mean and SD Reported BD1 and BD2 separately
		BD1/2	6, 17 (23: 7 BD1, 16 BD2)	28.5 ± 6.1
Singh 2019 (Singh et al., 2019)	UK	HC	14, 15 (29)	60.14 ± 10.2	Unspecified	Unspecified Assumed Whole Blood	mt-CYB, RNaseP	Mean and SE extracted from Figure 4 SE converted to SD
		BD (early onset)	10, 16 (26: 14 BD1, 12 BD2)	56.3 ± 9.26
		BD (late onset)	5, 15 (20: 14 BD1, 6 BD2)	57.31 ± 11.87
Spano 2022 (Spano et al., 2022)	France	HC	31, 47 (78)	40.8 ± 15.1	Caucasian	Unspecified Assumed Whole Blood	mt-DNA HBB	Reported n, m and t values
		BD unspecified	54, 76 (130)	44.3 ± 12.9
Wang 2018 (Wang et al., 2018)	China	HC	19, 15 (34)	24.5 ± 3.82	Unspecified Assumed Asian	Unspecified Assumed Whole Blood	mt-ND1 HBB	Reported mean and SD of transformed data
		BD1 Manic	26, 23 (55)	26.67 ± 7.15
		BD1 Dep	23, 24 (47)	26.94 ± 8.91
		BD1 Euth	9, 20 (29)	23.76 ± 3.94
Wang 2020 (Wang et al., 2020)	China	HC	16, 15 (31)	24.51 ± 4.10	Unspecified Assumed Asian	Unspecified Assumed Whole Blood	mt-ND1 HBB	Extracted mean and SD controlled for covariates from Figure 1D
		BD Manic	22, 29 (51)	26.90 ± 7.26
		BD Dep	26, 20 (46)	27.07 ± 9.12
Yamaki 2018 (Yamaki et al., 2018)	Japan	HC	27, 27 (54)	48.7 ± 13.6	Asian	Whole Blood	mt-ND1 HBB	Reported n, m, and t values
		BD	32, 37 (69)	47.6 ± 13.5

^aMedian and IQR

BD = Bipolar Disorder

CYB – Cytochrome b

HBB – Hemoglobin subunit beta (beta-globin, beta-hemoglobin)

HC = healthy controls

HGB – Beta hemoglobin

MDD = Major Depressive Disorder

ND1 – NADH dehydrogenase, subunit 1

PBMC – Peripheral blood mononuclear cells

PK – Pyruvate Kinase

Table 2:

Study participant recruitment, diagnostic criteria, and inclusion criteria details

Author, Year		Recruitment	Diagnostic evaluation	Diagnostic criteria
Cai 2015	HC DeCC MDD from both pops	United Kingdom Depression Case-Control (DeCC) study (Cohen-Woods et al., 2009) Three clinical UK sites: London, Cardiff and Birmingham. Identified from psychiatric clinics, hospitals and general medical practices and from volunteers responding to media advertisements	Modified version of the Past History Schedule Beck Depression Inventory	Excluded if they, or a first-degree relative, ever fulfilled criteria for depression or any other psychiatric disorder. Excluded 10 or > Beck Depression Inventory
			Schedules for Clinical Assessment in Neuropsychiatry (SCAN) Beck Depression Inventory	≥ 2 episodes of unipolar depression of at least moderate severity separated by ≥ 2 months of remission DSM-IV ICD-10
		Genome-Based Therapeutic Drugs for Depression (GENDEP) study (Uher et al., 2009) Recruited by generalist and specialist referrals and advertisement	Semi-structured SCAN v 2.1 interview	diagnosis of major depressive episode of at least moderate severity ICD-10 DSM-IV
Chang 2015	HC	Health Screening Clinic	Patient Interview
	MDD	Psychiatric Outpatient Clinic	Chart Record Review; patient interview with Mini-International Neuropsychiatric Interview	DSM-IV Clinical stability by CGI Unchanged meds for 2 months
Chung 2019	HC	Age and sex matched controls pulled from lab-maintained pool of control subjects
	MDD	Psychiatric outpatient clinic	Medial record review with consensus by 2 psychiatrists with additional information from patient interview conducted by research nurse	DSM-IV
Czarny 2020	HC	Selected from a healthy population
	MDD	Hospitalized at the Department of Adult Psychiatry of the Medical University of Lodz	Psychiatrist evaluation; SCID; HAMD	ICD-10
Fernström 2021	HC	Flyers, bulletin board notices, Craigslist postings, newspaper ads		Excluded any Axis I disorder and acute illness
	MDD	Flyers, bulletin board notices, Craigslist postings, newspaper ads, and clinical referrals	SCID; Verified in a separate diagnostic eval with a Board-Certified psychiatrist	DSM-V HAMD >= 17 or >= 20 HDRS
He 2014	HC	Hospital outpatients		Do not meet criteria for an Axis I disorder; No family history of psychiatric disorders
	MDD	Hospital outpatients	Psychiatrist diagnosis; HDRS-17; GAF	DSM-IV-TR
Kim 2011	HC	Yonsei Aging Study (YAS) participants recruited through public health centers located in the Yangpyung and Ilsan districts of South Korea		GDS 8 or below
	MDD		Short form GDS (15Q) conducted by two experienced medical staff fully educated on the GDS	GDS above 8
Lindqvist 2018	HC	Flyers, bulletin board notices, Craigslist postings, newspaper ads	SCID verified by board-certified psychiatrist
	MDD	Flyers, bulletin board notices, Craigslist postings, newspaper ads, and clinical referrals	SCID verified by board-certified psychiatrist HDRS	DSM-IV; HDRS > 17
Otsuka 2017	HC	Recruited from the main islands of Japan, including medical students, hospital workers, and the general population	Unstructured interviews conducted by two psychiatrists Excluded for personal and/or familial history of psychiatric disorders/suicidal behaviors	DSM-IV
	MDD	Autopsies on suicide victims	The verdict of “completed suicide” was made through discussion with the Medical Examiner’s Office of Hyogo Prefecture and the Division of Legal Medicine in the Kobe University Graduate School of Medicine
Tyrka 2016	HC	Local, newspaper, and internet advertisements directed toward healthy adults and individuals with depression, childhood parental loss, or a history of early life stress.
	MDD		SCID for DSM-IV; ISD-SR; STAI; PSS; CD-RISC	DSM-IV
Ryan 2023	HC	Recruited in three batches through advertisement in local newspapers and social media from 2008 to 2016		No history of psychiatric illness
	Uni and Bipolar dep.	Recruited into the EFFECT-Dep Trial (Enhancing the Effectiveness of Electroconvulsive Therapy in Severe Depression; NCT01907217) from 2008 to 2012 in St. Patrick’s Mental Health Services, Ireland	Referred for ECT for treatment of a major depressive episode as diagnosed by the SCID HAM–D24	DSM-IV Ham-D 24 >= 21
Tusji 2019	HC	Recruited from the Osaka, Kobe, and Tokushima city areas of Japan (Hondo) and participated via outpatient consultations		no lifetime, or first-degree relative history of psychiatric disorders, or any current serious medical disorder.
	MDD		Previous clinical diagnosis	DSM-IV or DSM-5
	BD		Previous clinical diagnosis	DSM-IV or DSM-5; excluded BD NOS (all YMRS < 8)
Angrand 2021	HC	Recruited in the clinical investigation center (CIC) of the Henri Mondor Hospital (Créteil).	French version DIGS; Interviewed by trained psychiatrists or psychologists.
	BD	Assessed in the university affiliated psychiatric department of the Henri Mondor Hospital (Creteil, France) Included under the framework of the cohort I-GIVE (Immuno-Genetics, Inflammation, retro-Virus, Environment).	French version SCID YMRS MADRS PANSS Interviewed by trained psychiatrists or psychologists.	DSM-IV MADRS above 17 YMRS above 8 or PANSS above 60 were considered as being in an acute phase.
Chang 2014	HC	Health screening clinic of Changhua Christian Hospital	Chart review	No history of major psych disorders
	BD1, euthymic	Psychiatric outpatient clinic at Changua Christian Hospital, Taiwan	Chart review and board-certified psychiatrist interview Mini International Neuropsychiatric Interview; CGI	DSM-IV CGI of 1 or 2 (clinically stable) No med change in last 2 mo
Chang 2022	HC	Healthy controls from the community		exclusion of individuals with mental illnesses by a senior psychiatrist using the Chinese version of the Mini International Neuropsychiatry Interview.
	BD unspecified	Outpatients at the National Cheng Kung University Hospital	Interview by an attending psychiatrist Chinese Version of the Modified Schedule of Affective Disorder and Schizophrenia–Life Time HDRS-17 YMRS-11	DSM-V On valproate
Chung 2022	HC	Drawn from a pool of control subjects previously established using a matching process
	BD	Psychiatric outpatient departments of Eulji General Hospital and other university hospitals in Seoul, Korea	Consensus of at least two psychiatrists YMRS HAM-D	DSM-IV
De Sousa 2014	HC	Advertisement in the community	SCID	excluded if they had a lifetime history mental disorder or first degree relative with mood or psychotic disorder
	BD	Newspaper, internet and radio advertisement Outpatient diagnosis of BD	Assessments performed by experienced psychiatrists SCID HAM-D - 21	DSM-IV-TR currently experiencing a depressive: HAM-D ≥ 18
Singh 2014	HC	Recruited within the same geographical area using print and online posters	SCID	No history of axis I disorders
	BD	Recruited from OXTEXT (an Oxford mood monitoring research programme) database	SCID with diagnosis confirmed by an experienced psychiatrist	DSM- IV
Spano 2022	HC		SCID	no personal or familial history of mood disorders nor suicidal
				behavior, were also recruited in this study.
	BD	University-affiliated psychiatric department in Paris	SCID MADRS YMRS	DSM-IV; ≤ 8 MADRS and YMRS
Wang 2018	HC	Advertisements in local communities		Excluded for history of substance use, mental or neurological disorders, serious medical disorders
	BD1	Inpatient and outpatient departments of the Second Xiangya Hospital in Hunan Province of China Euthymic patients primarily recruited from the outpatient clinic.	Diagnosed by two or more trained psychiatrists using SCID with psychotic screen (SCID-I/P). YMRS HAM-D CGI-BD-S	DSM IV
Wang 2020	HC	Enrolled by advertisement.	SCID non-patients edition (NP)
	BD	Recruited from the Suzhou Guangji Hospital (Jiangsu Province, China)	Diagnosed by more than two psychiatrists with expertise according to the SCID	DSM-IV No less than two mood episodes of BD.
Yamaki 2018	HC	Recruited from the Kobe, Osaka, and Tokushima city areas of Japan between 2005–2016		no present, past, or family history of psychiatric disorders, or any current serious medical disorder
	BD		All patients were currently diagnosed by two psychiatrists	DSM-IV or DSM-V

BD – bipolar disorder

CGI - Clinical Global Impression of Illness (BD-S: Bipolar Disorder-Severity of Illness Scale)

CD RISC - Connor-Davidson Resilience Scale

DIGS - Diagnostic Interview for Genetic Studies

GAF - Global Assessment of Function Scale

GDS – Geriatric Depression Scale

HAMD/ HDRS - Hamilton Depression Rating Scale

HC = healthy controls

ICD-10 - International Statistical Classification of Diseases 10^th Edition

IDS-SR - Inventory for Depressive Symptoms, Self-Report

MADRS - Montgomery-Asberg Depression Rating Scale

MDD – Major Depressive Disorder

PANSS - Positive and Negative Syndrome Scale

PSS - Perceived stress scale

SCID - Standardized Composite International Diagnostic Interview

STAI - State trait anxiety inventory

YMRS - Young Mania Rating Scale

Table 1 also indicates the mitochondrial copy number data extracted or calculated for each study. Where possible, reported mean and standard deviations of patient and control mtDNAcn were directly extracted (10 papers). This also includes papers that performed a transformation on their raw data (e.g. log-transform) and reported these means. Where means and standard deviations were not reported in text, data was directly extracted from available figures using a web-based plot digitizer tool, WebPlotDigitizer (Version 4.6). A screenshot of the relevant figure was uploaded to the site, and under the 2D Bar Plot setting, the X and Y axes were calibrated with known points. Key points on the graph were selected and reviewed as acquired data including the graphed mean and the top error bar position. To obtain the standard deviation, the mean was subtracted from the standard deviation bar upper limit. Three papers required the extraction of mean/standard deviation in this way. One additional paper graphed standard error rather than standard deviation, which was extracted in the same manner. The standard deviation was calculated from standard error using sample size in the equation SD = SE(√N). Three studies did not use the mean/standard deviation to report their data, but rather reported either in text or in figures the median and interquartile range of skewed data. One paper reported the medians and interquartile ranges, while for two papers quartile 1, median, and quartile 3 data points were extracted using the plot digitizer tool as described above using the bottom, middle, and top lines of box plots as the key points for extraction. Reported or extracted data were transformed to generate an approximate mean and standard deviation for further analysis using estimator formulas found in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins et al., 2022). Mean was estimated using the formula X = (Q1 + median + Q3)/3. Standard deviation was estimated using the formula SD = (Q3-Q1)/1.35 for studies where the n was greater than 50 (2 studies), and SD = (Q3 - Q1)/ η for studies where n was less than 50 (1 study), where η values were defined as previously described (Wan et al., 2014).

Alternately, some papers conducted linear regression analysis to control for factors such as sex or medication use in their mtDNAcn analysis. We believe these adjusted analyses provide a better representation of the original study data, and where these analyses were conducted, we extracted the number of factors used in the analysis (m) along with the reported t-statistic (t) (4 papers). One paper reported the associated p-value, but not the t-statistic, which we calculated from a look up table based on the reported p-value. The factors used to create these linear models are catalogued in Supplementary Table 2.

Meta-analysis and evaluation of publication bias

The patient and control population means, standard deviations, and group sizes (n) were used to calculate the standard mean difference (SMD) for papers where mean and standard deviation were reported or calculated. For papers where linear regression models were used, the number of factors in the model (m) along with the t-score and total n were used to calculate Fisher’s r-to-z transformed partial correlation coefficient. The resulting effect size (yi) and sample variance (vi) were used for a random effects model to assess the observed relationship between mood disorder status and blood mtDNAcn. A random effects model was chosen due to the small number of studies in this analysis. Studies including MDD and BD were analyzed separately. Two methods of sensitivity analysis were used to assess the impact of each study in our models, influence analysis and leave-one-out analysis.

Subgroup analysis was used to further probe causes of heterogeneity in our data sets. Factors for analysis in separate models included geographic location of the study, as well as the type of tissue used for analysis including whole blood, PBMCs, or leukocytes. BD data was analyzed using the type of BD (BD1, BD2, or unspecified/data not collected or reported) as a modifier. Meta-regression was used to examine the impact of sample population age and % sex composition. Age and % female were averaged between the patient and control populations to yield one value per study. To assess the risk of publication bias in our meta-analysis models, we used Egger’s regression test for MDD and BD paper sets, using the rma model built for the meta-analysis as input. Funnel plots, which depict the effect size of each study with respect to each study’s variance, were also produced for MDD and BD samples, and visual inspection to judge asymmetry was also conducted as a secondary check for bias.

Statistics

Statistical analysis was performed using R Version 4.2.2 (R Foundation for Statistical Computing, Vienna, Austria https://www.R-project.org/) including the package metafor 4.0 (Viechtbauer, 2010). Statistical significance was defined as p ≤ 0.05.

Results

MDD and mtDNAcn

The pooled effect sizes across all MD studies yielded a trending increase in mtDNAcn in MDD patients with an estimated effect size of 0.2393 and a 95% confidence interval of −0.0508 – 0.5294 (p = 0.106) (Figure 2A). The analysis revealed a statistically significant degree of heterogeneity (I² = 95.015; τ² = 0.235, Q₍₁₁₎ = 133.647, p < 0.0001). Sensitivity analysis showed no paper had a significant influence on the model, and a leave one out analysis revealed a range of effect sizes from −0.3034 – 0.1548 (p = −0.038 – 0.1324). When examining potential factors that could explain the high degree of heterogeneity, we used subtype and meta-regression models using geographic area (America, Asia, or Europe), type of blood sample used (whole blood, PBMCs, or leukocytes), participant age, and sample % female in independent models. These models yielded no significant results, however, when we examined age and sex of the study participants, one study stood out as an outlier from the rest of the studies. While many study participants were young to middle-aged adults ~ 25–60 years old, Kim, et al. (2011) specifically examined MDD in an elderly population as part of an aging study. The average age of its study participants was ~75 years old. Further, this study was also the only one to solely include women in its sample populations. All other studies included at least 30% women, with most studies being at or near 50% female. When the model was repeated excluding this study, the summary statistic for mtDNAcn was 0.303, with a confidence interval of 0.017–0.5897 (p = 0.0379), indicating an increase in mtDNAcn with MDD status (Figure 2B). This analysis did not substantially reduce the heterogeneity observed in the model (I² = 94.60; τ² = 0.207; Q₍₁₀₎ = 120.3651, p < 0.0001), however leave-one-out meta-analysis yielded a range of effect sizes from 0.3645 – 0.2166 (p = 0.012 – 0.0937) indicating a much more consistent picture of the role of MDD on mtDNAcn, specifically in increasing mtDNAcn. No single study showed significant influence on the model. Subtype and meta-regression testing for the impact of geographic area, type of blood sample used, age, and sex did not yield significant results when including the Kim et al. (2011) study. Meta-regression models excluding Kim et al. (2011) continued to show no effect of geographic area, age, and sex. However, blood sample type now showed a significant impact on the relationship between diagnosis and mtDNAcn (Q₍₃₎ = 12.441, p = 0.006). Studies using whole blood showing a significant effect size 0.848 (95% CI 0.0316 – 1.664; p = 0.0418), while studies using leukocytes (−0.337; 95% CI = −1.099 – 0.425; p = 0.386) and PBMCs (0.3024; 95% CI = −0.616 – 1.221; p = 0.519) did not (Supplemental Figure 1A). This adjusted analysis also did not substantially reduce heterogeneity in the model (I² = 92.09; τ² = 0.141; Q₍₈₎ = 91.429, p < 0.001). For random effects models including and excluding the Kim, et al. (2011) study showed no publication bias, as measured with an Egger’s regression test (z = −0.317, p = 0.752; z = −0.183, p = 0.855 respectively), and yielded symmetrical funnel plots (Figure 3). Meta-regression models also showed no significant publication bias, including the models built with blood sample as the regression factor (Supplemental Figure 1B; z = 1.541, p = 0.123).

Figure 2:

MDD Diagnosis and mtDNAcn. A. Forrest plot showing the results of meta-analysis of the relationship between MDD and mtDNAcn compared to controls including all identified studies. Heterogeneity: I² = 95.015; τ² = 0.235, Q₍₁₁₎ = 133.647, p < 0.0001 B. The same meta-analysis adjusted excluding Kim, et al., 2011. Effect sizes (yi) and 95% confidence intervals are represented graphically, with exact values in the righthand column. The size of the square represents the weight of the study in the model. Heterogeneity: I² = 94.60; τ² = 0.207; Q₍₁₀₎ = 120.3651, p < 0.0001

Figure 3:

MDD model funnel plots. A. Funnel plot including all MDD studies and B. all studies excluding Kim, et al., 2011. Effect size for each study is graphed with its corresponding standard error. Both with and without Kim, et al., 2011, studies are comparable in size and symmetrically distributed indicating unlikely publication bias.

BD and mtDNAcn

Initial random effects model for all the BD studies indicated no significant effect on mtDNAcn with an effect size of −0.115 (95% CI = −0.316 – 0.0931; p = 0.285) (Supplemental Figure 2). As with the MDD model, there was significant heterogeneity (I² = 88.06; τ² = 0.1712, Q₍₁₉₎ =131.05, p < 0.001). Here the influence sensitivity analysis revealed that the Chung, et al. (2022) BD2 sample had a significant influence on the model. Leave one out analysis yielded effect sizes ranging from −0.094 - −0.200 (p = 0.3781 – 0.0081), with the lowest range value representing the exclusion of the Chung et al. (2022) BD2 sample. Further subtype and meta-regression testing for the impact of geographic area, type of blood sample used, age, and sex did not yield significant results.

An additional subtype analysis model was generated examining the impact of BD type on effect size. For this analysis, studies were categorized by their patient population and whether they contained patients diagnosed with BD type 1, BD type 2 or “unspecified” if the authors either did not collect this information or did not separate their participants for analysis. This yielded eight studies with BD Type 1 patients, two studies with BD type 2 patients, including the Chung, et al. (2022) study, and 10 studies with unspecified patient populations (Figure 4). In this model, there was a significant effect of the diagnostic type moderator (Q₍₃₎ = 26.358, p < 0.001). BD type 1 patients had lower copy number than controls with a significant effect size of −0.374 (95% CI = −0.603 - −0.145, p = 0.0014). Conversely, BD type 2 patients had a significantly higher copy number with an effect size of 1.254 (95% CI = 0.743 – 1.766, p < 0.001). Comparable to the original random effects model, in the studies where BD type was unspecified the effect size was not significantly different from zero at 0.245 (95% CI = −0.050 – 0.593, p = 0.103). While still significant, heterogeneity was much lower in this model than in the original model (I² = 70.6; τ² = 0.0574, Q₍₁₇₎ =53.21, p < 0.001). In a sensitivity analysis for this model, the Chung et al. (2022) BD2 sample again showed significant influence on the model. As there are only two studies with selective BD type 2 patients and in light of evidence that BD type may be differentially correlated with copy number based on diagnostic type, further studies on BD type 2 patients would be necessary to determine if this result is due to the large effect of one outlying study, or if the results shown here are an accurate representation of BD type 2 mtDNAcn. Egger’s regression test of this model showed no evidence of bias (z = 0.684, p = 0.494), and the funnel plot was symmetrical (Figure 5).

Figure 4:

BD Diagnosis and mtDNAcn. A. Forrest plot showing the results of subtype meta-analysis of the relationship between BD type and mtDNAcn compared to controls. Effect sizes (yi) and 95% confidence intervals are represented graphically, with exact values in the righthand column. The size of the square represents the weight of the study in the model. Grey diamonds represent the expected effect size for the given type of BD. Heterogeneity: I² = 70.6; τ² = 0.0574, Q₍₁₇₎ =53.21, p < 0.001. MRM = meta-regression model.

Figure 5:

BD model funnel plot. A. Funnel plot including all BD studies for the subtype meta-regression model using BD diagnosis type as a factor. The residual term for each study is plotted against its corresponding standard error. Studies are roughly symmetrically distributed indicating unlikely publication bias.

Discussion

The present study used a comprehensive search strategy and explored the relationship between the mood disorders MDD and BD and whole blood/cell-based mtDNAcn. MDD diagnosis is associated with elevated blood mtDNAcn across studies in mixed male and female populations in adulthood. For BD, when all studies are considered regardless of BD subtype diagnosis, there is no predictable relationship between diagnosis and mtDNAcn, however, when subtype diagnosis is considered, BD type 1 patients show reduced mtDNAcn across studies, and BD type 2 patients show an increase in mtDNAcn. Across mood disorders, diagnoses involving depression in the absence of true mania are associated with higher mtDNAcn, while patients experiencing manic episodes show decreased mtDNAcn.

With respect to the BD type 1 studies, both studies by Wang et al. tracked the active mood state of their participants, differentiating between manic, depressive, and in their 2018 paper, euthymic state (Wang et al., 2020, 2018). Interestingly, mtDNAcn was consistently reduced in these patients during both depressed and manic states, however, patients in a euthymic state did not show a comparable decrease. Chang, et al., (2014) (Chang et al., 2014) specifically recruited individuals with BD type 1 who were clinically stable (Clinical Global Impression of Illness; CGI), and did observe decreased mtDNAcn, in contrast to the euthymic patients in the Wang, et al. study. Both BD type 1 and type 2 patients in the de Sousa, et al. (2014) (de Sousa et al., 2014) study were in active depressive episodes, but their mtDNAcn followed the trends of decreased mtDNAcn for type 1 and increased for type 2. The rest of the studies did not track or report mood state at the time of testing, therefore we could not fully evaluate whether current mood state or broader diagnostic phenotype plays a more significant role in mtDNAcn. A significant limitation of our model with respect to BD type is that only two studies specifically examined mtDNAcn in BD type 2 patients, with the findings of Chung, et al. (2022) (Chung et al., 2022) having a significant effect on the model. The eight studies selectively examining BD type 1 patients provide strong evidence of reduced mtDNAcn, however, the results relating to BD type 2 could be a result of small sample size. When type is not specified, there is no pattern in mtDNAcn change, indicating diagnostic type increases the variability in the correlative relationship between diagnosis and mtDNAcn, and more diagnostic specificity should be used. Future work on BD and mtDNAcn should strive to both differentiate between BD diagnosis type, as well as to examine how depressive, manic, or euthymic states influence mtDNAcn, potentially dynamically with repeated sampling in individuals over time.

Active mood state could also impact mtDNAcn in MDD patients. Half of the MDD studies specified current mood state, with most using Hamilton Depression Rating Scale (HAMD), Montgomery-Asberg Depression Rating Scale (MADRS), Geriatric Depression Scale (GDS), or recruiting from an actively hospitalized population to identify patients actively experiencing depressive symptoms. One study noted recruitment for clinically stable patients by the CGI (Chang et al., 2015). While no discernable pattern emerges when examining the data based on symptom severity status, it is important to keep in mind for interpreting results. Additionally, we included the study by Otsuka, et al. (2017) (Otsuka et al., 2017) in our analysis because of the close link between depression and suicidality, however it is important to note that MDD is not the only diagnosis that could lead to suicide attempts or completion.

In our initial analysis, age, sex and geographic region/race did not impact the relationship of either mood disorder with blood mtDNAcn. Age has been shown before to impact mtDNAcn on its own (Knez et al., 2016; Mengel-From et al., 2014; Yang et al., 2021; Zhang et al., 2017), and some studies have found no effect of sex on mtDNAcn (Mengel-From et al., 2014), while others have found higher mtDNAcn in women (Yang et al., 2021), with variance based on hormone use (Knez et al., 2016). These effects of age and sex on mtDNAcn influenced our exclusion of the Kim, et al. (2011) paper from our MDD analysis due to its older, female-only patient population. Few studies have examined the effect of race/geographic region on mtDNAcn explicitly. In the previous meta-analysis by Yamaki et al. in 2018 on BD mtDNAcn, mtDNAcn was significantly decreased in studies from Asian countries, but not North and South American samples, leading them to conclude race was impacting the relationship of mood disorder and blood mtDNAcn. With the addition of multiple subsequent studies, this relationship does not hold up, and was perhaps an artifact of the coincidence that two of the studies included in their Asian-population subset were BD type 1 specific studies. Further, their initial analysis also included Fries et al., (2017) (Fries et al., 2017) which we excluded from our analysis. The BD patient population in this study did not exclude comorbid anxiety disorders, including PTSD, which has been shown to have its own potentially contrasting impacts on mtDNcn, and was outside the scope of the current analysis (Bersani et al., 2016; Hummel et al., 2023). Other important work has linked changes in mtDNAcn to early life adversity, which can have significant comorbidity with MDD or BD (Picard and McEwen, 2018b; Tyrka et al., 2016, 2015).

Sample type, either whole blood, leukocytes, or PBMCs did not initially impact the relationship between diagnosis and mtDNAcn. However, MDD analysis after exclusion of the Kim et al. (2011) study showed the significant positive effect was restricted to whole blood samples, and did not hold up in leukocyte or PBMC samples, although these are both underpowered (1 and 3 studies respectively) compared to 9 whole blood studies. Five BD studies used leukocytes, and no BD studies used PBMCs as there sample preparation. Both leukocyte and PBMC preparations selectively isolate cellular blood fractions, however whole blood samples also contain circulating cell-free mtDNA (ccf-mtDNA). Unlike tissue/cell-based mtDNAcn, which roughly approximates tissue respiratory capacity, ccf-mtDNA is found extracellularly. Levels of ccf-mtDNA are uncoupled from respiratory capacity or copy number and can be measured in blood serum, plasma, saliva, cerebrospinal fluid, or other tissues (Lindqvist et al., 2018; Park et al., 2022; Trumpff et al., 2021). Fewer studies have examined this biomarker, and both a recent systemic review (Trumpff et al., 2021) and meta-analysis (Park et al., 2022) of ccf-mtDNA in psychiatric illness have described varying relationships between this measure and psychiatric diagnosis, however, ccf-mtDNA is lower across studies of patients with non-psychiatric neurological conditions including neurodegenerative disorders (Park et al., 2022). While whole blood contains ccf-mtDNA, the majority of measured mtDNA and all of the nuclear DNA comes from PBMCs, of which leukocytes are a subtype. Platelets also contain mitochondria, but no nucleus, and would contribute to whole blood mtDNAcn. While reduced in more selective PBMC or leukocyte fractions, platelets can still contribute to overall mtDNAcn (Picard, 2021). Nevertheless, whole blood or cell-based mtDNAcn reflects mitochondrial changes in immune tissues, which is important to keep in mind with respect to the high comorbidity of chronic pain and inflammatory disorders with mood disorders (Bauer and Teixeira, 2021; Benedetti et al., 2020; Jones et al., 2020). Understanding what these changes in mtDNAcn reflect will greatly inform how the utility of this biomarker is interpreted.

Finally, it is important to note that across all our models for MDD and BD, heterogeneity measures remained high despite attempts to explain variability with subtype and meta-regression analysis, reflecting high heterogeneity between studies and relatively low variance within studies. This may signal further relevant underlying variables that may be influencing blood mtDNAcn that are not apparent currently or are not sufficiently powered in the current analysis. Moving forward, the consensus described here indicates that blood mtDNAcn is altered in different ways by MDD and BD. In this context, whole blood, cell-based mtDNAcn may serve as a broad marker for changes in mood disorder patients, with diagnostic segregation. Future work should examine how mtDNAcn may correlate with less accessible central physiological changes associated with mood disorders or aspects of treatment response, potentially serving as a guide to predict medication responses based on individual differences in neuropathology.

Highlights.

• MDD is associated with an increase in blood mitochondrial copy number

• The effect on copy number in BD patients is dependent on diagnostic type

• BD Type 1 is associated with lower blood mitochondrial copy number

• BD type 2 is associated with higher mitochondrial copy number

Abbreviations:

MDD

Major Depressive Disorder

BD

Bipolar Disorder

mtDNAcn

mitochondrial DNA copy number

Data Statement

Data and R code for analysis is available upon request.