INTRODUCTION
Major depressive disorder (MDD) is a common debilitating mental disorder [1] that affects approximately 280 million people worldwide [2]. Per the criteria outlined by the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition, MDD is characterised by at least two weeks of pervasive low mood, low self-esteem, and loss of interest [3]. Additionally, MDD is ranked by the World Health Organisation as the single leading cause of disability in the world [4]. Despite its prevalence, MDD is still underdiagnosed and undertreated in the clinical setting [5].
It has been suggested that the pathogenesis of MDD involves biological, psychological and genetic factors [6]. In practice, psychiatrists diagnose MDD based on patient-reported symptoms, clinical judgment, and other subjective psychometric or behavioural tests. Examples of psychometric tests include the Patient Health Questionnaire-9 [7], Hamilton Depression Rating Scale [8], and Beck Depression Inventory [9]. This heavy reliance on subjective measures may lead to discrepancies in the diagnosis of MDD. Furthermore, the clinical assessment cannot differentiate endogenous depression and reactive depression to life events. The inherently subjective nature of diagnosing MDD has, therefore, given rise to a need for biomarkers, which offer a conceivable target for assisting in the diagnosis of MDD [10].
To date, there has not been an overarching review that quantifies the wide variety of objective measures used to diagnose MDD in adults. Recent systematic reviews include studies on the efficacy of electroencephalogram (EEG) [11], inflammatory markers [12] and metabolomics-based biomarkers [13]. However, these existing systematic reviews restricted their literature search exclusively to studies that investigated only one modality. Therefore, this systematic review aims to provide an overview of the available objective measures (i.e., candidate biomarkers) across different modalities that may be used to diagnose MDD reported between 2013-2023. The various measures will be categorised based on their nature and specific biomarkers investigated. Its underlying theoretical concepts, strengths, weaknesses, and potential for further research will be explored.
METHODS
A systematic search was conducted using PubMed and PsycINFO from January 2013 to December 2023. A combination of search terms was used to ensure that all relevant papers referring to objective biomarkers available for diagnosing MDD were found (Supplementary Material). The protocol of this systematic review was accepted on PROSPERO on the 18th of December 2023: CRD42023488051.
Key inclusion criteria for our systematic review included articles that were published in a peer-reviewed journal in English, primary publications that introduced biomarkers used to diagnosed MDD in an adult population (above 18 years), and participants with a formal diagnosis of MDD. Papers were excluded if they were published in a non-peer-reviewed journal and/or not published in English; featured non-primary study designs (e.g., systematic review, metaanalysis, literature review); included children or adolescents in the study population; featured participants without a clinical diagnosis of MDD; featured participants with a diagnosis of other forms of MDD such as treatment resistant depression, vascular depression, remitted depression. The decision to exclude non-primary publications was made due to the nature of a systematic review. The decision to restrict the search to the last 10 years allowed his systematic review to focus on the most recent technology.
Following de-duplication, abstracts and titles of articles identified in our literature search were independently screened by two authors (A.S.H., R.H.) to determine which studies potentially met the inclusion criteria. Both authors then carried out full-text screening of potentially relevant studies. Data extraction of included articles was performed on Excel (Microsoft). Any discrepancies throughout the process were reviewed by both authors and resolved through discussion.
Data extraction consisted of study identification, study aims, country where the study was conducted, study design, sample size, participant demographics, and outcomes of interest. In this study, outcomes of interest included comparison method, biomarkers, medication-naive state, area under the curve (AUC) values, and sensitivity/specificity values.
RESULTS
The systematic search resulted in 4,972 articles after removing duplicates, of which 225 were accepted for full-text review. Of the 255 articles, 43 articles that reviewed 42 studies were deemed relevant and included in the review. One article reported on an updated version of a previous study (Figure 1).
Of the 42 studies, 18 studies reported on the clinical utility of blood biomarkers to diagnose MDD; 13 studies on the clinical utility of neuroimaging/neurophysiology biomarkers; four studies on the clinical utility of urine biomarkers; one study each on the utility of auditory, vocal, dermatological and cerebrospinal fluid (CSF) biomarkers; three studies on the utility of combined biomarkers.
Most of the studies were conducted in China (18), Japan (4), and Brazil (3). Other countries include the USA, Germany, South Korea, Taiwan, Australia (2 each), Canada, Iran, Singapore, Croatia, Poland, Ukraine and Lithuania (1 each). All studies included were designed in the case-control format, with 15 studies incorporating a discovery and validation cohort within the same study.
Blood biomarkers
The 18 papers that reported on blood biomarkers for the diagnosis of MDD were broadly categorised into transcriptomics (4) [14-17], metabolomics (8) [16,18-24], genomics (4) [25-28], and combined (2) [29,30]. A summary of the following studies is shown in Table 1. For this review, relevant outcomes that were considered include sensitivity/specificity and AUC (Table 1).
Four studies reported information on transcriptome-based blood biomarkers. Three studies reported on the utility of messenger ribonucleic acid (mRNA) [14-16], while one study reported on the utility of complementary RNA (cRNA) [17]. mRNA proved inconsistent in diagnosing MDD, with only one study [16] showing high AUC (0.889) and sensitivity (96.0%), while Gecys et al. [15] and Fan et al. [14] exhibited moderate performance under the receiver-operating characteristic (ROC) (0.60 and 0.631, respectively). Individual cRNA molecules showed promising AUC and sensitivity values in diagnosing MDD, with the best performing molecules exhibiting high AUC (0.862) and sensitivity/specificity (78.6%/86.7%) [17].
Eight studies reported information on metabolomics-based blood biomarkers [16,18-24]. Of the eight studies, two reported on the utility of lipid biomarkers [23,31] and two reported on the utility of inflammatory blood markers [19,20]. Zhang et al. [23] presented changes in 20 lipid profiles among the male population, while Liu et al. [31] studied 17 lipid profiles in both sexes. Both studies exhibited excellent performance of 0.897 and 0.863 under the ROC curve.
As for inflammatory markers, Min et al. [20] investigated the diagnostic value of interleukin (IL)-6 and tumor necrosis factor (TNF)-α, while Kim et al. [19] investigated the diagnostic value of IL-6, IL-17, IL-17B and C-reactive protein in people with MDD. Kim et al. [19] indicated that lowered IL-6 was a more accurate predictor of MDD than TNF-α. However, Min et al. [20] revealed that there was no significant difference in IL-6 between healthy controls and people with MDD.
Four other studies reported information on neurotransmitters, insulin-like growth factor (IGF)-1, nervonic acid, and Peripheral Blood Mononuclear Cells (PBMC) [18,21,22,24]. Pan et al. [22] reported excellent sensitivity for a combination panel of dopamine, GABA (gamma-aminobutyric acid), tyramine and kyneuramine in the diagnosis of MDD, while Levada et al. [21] indicated that increased IGF-1 in MDD exhibited good performance under the ROC curve (0.820). Zheng et al. [24] featured a combination panel of 17 discriminants found in PBMC, including neurotransmitters and metabolites in glucose metabolism, which exhibited a strong AUC value of 0.926.
Four studies reported information on genomics-based blood biomarkers. Two studies reported on the diagnostic value of the BDNF gene [25,27], while the other two studies reported on the MYT1 gene and DNA methylation of peripheral leukocytes [26,28]. Lowered BDNF levels in people with MDD on medications exhibited high ROC performance (0.892) [27], but this observation was not replicated in the non-medicated population (AUC=0.652) [25]. In people with MDD, downregulated MYT1, a gene that promotes myelin growth in neurogenesis, exhibited high AUC and sensitivity/specificity [26].
Two studies reported on blood biomarkers that spanned across different categories [29,30]. Bilello et al. [29] reported that a combination panel of genomics (e.g., BDNF) and metabolomics (e.g., apolipoprotein CIII, alpha-1 antitrypsin, cortisol, resistin, prolactin, myeloperoxidase, TNF-α) displayed excellent AUC and sensitivity/specificity in people with MDD who received antidepressants. For medication-naive populations, Jiang et al. [30] reported that a combination panel of genomics (e.g., BDNF, proBDNF), metabolomics (tPA) and neurogenic receptors (p75NTR, TrKB) also exhibited similar performance in AUC and sensitivity/specificity.
Neuroimaging/neurophysiology biomarkers
Thirteen studies reported on neuroimaging and neurophysiology biomarkers for the diagnosis of MDD. Most studies (10) used magnetic resonance imaging (MRI) as their imaging modality [32-41], with five studies reporting on functional MRI (fMRI) [32,34,37,40,41]. Other imaging modalities used include functional near-infrared spectroscopy (fNIRS) [42], EEG [43], and positron emission tomography (PET) [44] (Table 2).
Four of the five studies utilising fMRI investigated the usage of resting state fMRI (rs-fMRI) [32,34,40,41]. Of the four rs-fMRI studies, two studies reported on the abnormal functional connectivity between brain hemispheres in people with MDD: Guo et al. [32] demonstrated that decreased voxel-mirrored homotopic connectivity in the posterior cingulate cortex (AUC=0.922) and cuneus (AUC=0.911) reliably identified MDD in participants, while Zhong et al. [41] identified that abnormal functional connectivity in the affective network (amygdala, pallidum, superior temporal lobe), salience network (anterior insula, dorsal anterior cingulate cortex), default mode network (precuneus/post cingulate cortex, inferior parietal gyrus, angular, thalamus), visual cortex and cerebellum was highly sensitive and specific for MDD (89.60%/93.90%).
The other two rs-fMRI studies reported on the diagnostic value of differences in spontaneous brain signal fluctuations [37,40]. Wei et al. [40] demonstrated that differences in Hurst exponent—a measure of regularity and persistence in brain signals—in the salience network, ventromedial prefrontal network, and lateral prefrontal network demonstrated high sensitivity and specificity for MDD (95.0%/85.0%). Liu et al. [34] demonstrated that lowered amplitude of low-frequency oscillations in the left parahippocampal gyrus (0.930) and right middle frontal gyrus (0.854) had excellent performance under the ROC curve.
Five other studies reported information using MRI-based biomarkers, with three studies utilising MRI [33,35,36] and two studies using arterial spin labelling-MRI (ASL-MRI) [38] and diffusion tensor imaging-MRI, respectively [39]. All three MRI studies reported the diagnostic value of changes in grey matter volume (GMV). Niida et al. [36] reported that increased GMV in the subgenual anterior cingulate cortex (SgACC) exhibited high AUC (0.911) and sensitivity/specificity (95.7%/96.0%), while Hellewell et al. [33] reported that decreased GMV in the midline, medial temporal lobe and prefrontal cortex exhibited moderate diagnostic value (AUC=0.750). Liu et al. [35] reported on the structural asymmetry of GMV and showed that reduced asymmetry of GMV in the dorsolateral prefrontal cortex of participants with MDD exhibited moderate sensitivity and specificity (61%/59%).
Ramasubbu et al. [38] utilised ASL-MRI to illustrate differences in cerebral blood flow in people with MDD. Following that, a combination panel of differences in cerebral blood flow and sex of participants revealed moderate sensitivity/specificity (80.0%/75.0%).
Three separate studies utilised EEG [43], fNIRS [42], and PET [44] to identify neuroimaging biomarkers in medication-naive cohorts. In Lubiński et al. [43], following a standardized visual stimulus, EEG was used to identify changes in pattern visual evoked potentials (PVEP) between people with MDD and healthy controls. A combination panel incorporating amplitude and peak time for PVEP exhibited excellent performance under the ROC curve (0.841). fNIRS was used to quantify variants in regional oxy-haemoglobin following a standardised verbal fluency test. Reduced oxy-haemoglobin in the frontal (0.760) and temporal (0.820) regions of people with MDD exhibited moderate AUC values [42]. Pillai et al. [44] utilised PET to investigate the diagnostic value of 5-HT1A autoreceptor binding in the dorsal and median raphe nuclei. Higher autoreceptor binding in the dorsal and median raphe nuclei demonstrated excellent AUC (0.934) and sensitivity/specificity (94.0%/84.0%) for MDD.
Urine biomarkers
Four papers reported on the diagnostic value of urinary biomarkers for the diagnosis of MDD [45-48]. Three studies investigated metabolomics-based biomarkers [46-48], while one study investigated proteomics-based biomarkers [45] (Table 3).
Of the three metabolomics studies, two employed a combination panel of metabolites for diagnosis [47,48], with one study focussed on a singular metabolite.46 Wu et al. [46] reported that the downregulation of argininosuccinate synthetase in plasma was moderately sensitive (83.7%) and specific (76.7%) in medication-naive MDD participants. In the first two separate studies, Zheng et al. [47] investigated a panel of urinary metabolites related to gut intestinal flora, glucose and tryptophan-nicotinic acid metabolism. Reduced malonate (glycolysis), m-hydroxyphenylacetate (gut intestinal flora) and increased Nmethylnicotinamide had good diagnostic values in a nonmedication naive population (AUC=0.810). The second study by Zheng et al. [48] involved a similar panel investigating the kynurenine, tyrosine-phenylalanine and oxidative stress pathways in addition to glucose and tryptophan-nicotinic acid metabolism. Increased sorbitol and decreased quinolinic acid, tyrosine, azelaic acid, hippuric acid, and uric acid showed strong performance under the ROC curve (0.905) in people with MDD who were medication-naive [48].
For the single proteomics study, Wang et al. [45] reported that a combination panel of upregulated peptides related to albumin, apolipoprotein, basement membrane peptides, and downregulated immune-related peptides exhibited excellent performance under the ROC curve (0.914) and high sensitivity/specificity (90.5%/92.9%).
Other biomarkers
Four separate studies reported on dermatological [49], voice [50], CSF [51] and auditory biomarkers [52]. Three studies utilised a combination panel of biomarkers that spanned different categories. None of the studies involved people with MDD who were medication-naïve (Table 4).
Kim et al. [49] reported on the diagnostic value of skin conductance in people with MDD. Participants were subjected to a standardized mental arithmetic task and a relaxation task while their skin conductance was measured. Differences in the skin conductance responses (amplitude, slope, number of non-specific skin conductance responses, power spectral density) demonstrated moderate sensitivity/specificity (70.0%/71.0%) in differentiating people with MDD from healthy controls.
Taguchi et al. [50] reported on the diagnostic value of voice quality in people with MDD. Participants were subjected to a standardised vocal recording exercise, following which their fundamental frequency, root mean square of energy, zero crossing rate, harmonics to noise ratio and 12-dimensional mel frequency cepstral coefficient (MFCC-a set of coefficients representing the distribution of sound energy across different frequencies) was calculated. The study showed that the second order coefficient, i.e., MFCC-2, exhibited good sensitivity (77.8%) and specificity in distinguishing MDD from healthy controls.
Schmidt et al. [51] separately investigated the diagnostic value of neuron specific enolase (NSE) and glial cell protein S100beta (S100B)-markers of neuronal metabolism and glial cell activity, respectively. While S100B did not possess diagnostic value, decreased NSE in people with MDD demonstrated good performance under the ROC curve (AUC=0.820) and had good sensitivity (81.0%) and specificity (75.0%).
A panel of six sensory oto-acoustic features derived from electrovestibulography was used to distinguish between people with MDD and healthy controls [52]. Though AUC or sensitivity/specificity values were unavailable, the combination panel achieved 87.0% accuracy.
Koo et al. [53] utilised a combinatory panel of biomarkers pertaining to executive dysfunction (Trail Making Test A and B, Stroop Task), motor activity (actigraphy) and neurophysiology (EEG, alpha power asymmetry). Results revealed that reduced psychomotor speed, increased divided attention lowered motor activity and increased asymmetry in upper alpha power was sensitive (94.7%) but not specific (66.7%) for diagnosing MDD.
Two other panels investigated the diagnostic value of combinatory panels that involved blood biomarkers. Wan et al. [54] reported on the diagnostic value of changes in blood and CSF microRNA (miRNA) expression, with the best performing molecules exhibiting excellent ROC performance (0.98) and high sensitivity/specificity (96.9%/95.2%). Bortolasci et al. [55] reported on the collective diagnostic value of PON1 enzyme, high density lipoprotein (HDL) in blood and lifestyle smoking habit. Lower levels of PON1 enzyme, HDL and smoking habit were associated with high specificity (93.8%), but not sensitivity (53.8%) for diagnosing MDD.
DISCUSSION
In summary, 42 studies published between 2013 and 2023 were included in this systematic review. Most studies reported on the clinical utility of blood (18) and neuroimaging/neurophysiology (13) markers associated with the diagnosis of MDD. Four studies reported findings based on urine biomarkers, three studies on the utility of combined biomarkers, and one study each on auditory, vocal, dermatological and CSF biomarkers. mRNA (3), BDNF (2), and inflammatory markers (2) featured repeatedly in the analysis of blood biomarkers, while GMV (3) was similarly highlighted for neuroimaging/neurophysiology biomarkers.
Of all blood biomarkers, the MYT1 [26] and the combination panel of dopamine, GABA, tyramine, and kynuramine were the most sensitive in detecting MDD [22]. Increased 5-HT1A autoreceptor binding in the raphe nuclei of the brain exhibited the best sensitivity among all neuroimaging/neurophysiology biomarkers [44], while a combination panel of peptides that included albumin, AMBP, HSPG, and APOA1 exhibited the best sensitivity among all urinary biomarkers [45]. Other biomarkers with notable performance include NSE in CSF [51], and miRNAs of peripheral mononuclear blood cells in blood and CSF [54] (Table 5).
The performance of these biomarkers can be attributed to the pathophysiology of MDD. The monoamine theory of depression attributes depressive symptoms to the reduced function of monoamine transmitters such as serotonin and norepinephrine. This explains why increased 5-HT1A autoreceptor expression by the dorsal and median raphe nuclei, which depletes the availability of serotonin in the body, is associated with MDD [56]. The neurotrophic model of depression also establishes the correlation between decreased BDNF (a neurogenesis-promoting protein), and higher frequency of depression, neuronal loss and cortical atrophy [57]. This explains why expression of MYT1-a neurogenesis-promoting genes-is decreased in people with MDD [26]. Similarly, increased expression of CSF NSE (which is released from neuronal cytoplasm during neuronal cell death) could reflect the increased levels of neuronal loss found in the MDD population [51].
As for the discriminant miRNAs-single-stranded noncoding RNAs that negatively regulate gene expression at the post-transcriptional level [14]-identified, a total of 734 genes related to neurotransmission, learning and memory were regulated by these 5 miRNAs [54]. These miRNA were found to be involved in BDNF down-regulation, thereby explaining its potential link to the depressive state. Additionally, the finding of increased levels of urinary APOA1 (a peptide involved in cholesterol transportation) in people with MDD aligns with the established relationship between dysregulated cholesterol metabolism and MDD symptoms, which does so via disrupting hypothalamic-pituitary-adrenal axis regulation and neurotransmitter synaptic formation [58].
However, discordance in sensitivity and AUC values surfaced between studies that analysed biomarkers of similar nature. For genetic biomarkers, mRNA proved inconsistent in diagnosing MDD in a non-medication naive population, with only Lin et al. [16] showing high AUC (0.889) and sensitivity (0.960), while Gecys et al. [15] and Fan et al. [14] exhibited moderate performance under the ROC (0.600 and 0.631, respectively). Similarly, Karlović et al. [27] revealed that serum BDNF had high AUC (0.892) in both male and female participants with MDD, whereas Chiou and Huang [25] revealed that serum BDNF had moderate AUC (0.652) in only males and was not diagnostic for females. For neuroimaging, all three studies that investigated reduction in GMV identified different regions as potential biomarkers, with Niida et al. [36] exhibiting excellent sensitivity, while Hellewell et al. [33] and Liu et al. [35] reflected only modest sensitivity.
For the mRNA studies, the mixed results could be attributed to differences in the specific type of biomarker investigated. Each mRNA study identified different molecules, which coded for different proteins and functions. Lin et al. [16] identified mRNA segments that code for the COMT gene, which degrades catecholamines [59]. This aligns with the dopamine theory of depression, which reflects the association between deficiency in dopaminergic transmission and the symptoms of depression [60]. On the other hand, Gecys et al. [15] identified mRNA segments relating to the toll-like receptor signaling pathway, while Fan et al. [14] identified mRNA segments relating to cancer signaling pathways. These pathways are not known to be associated with the pathophysiology of MDD, which could explain the reduced AUC performance.
Similarly for neuroimaging/neurophysiology biomarkers, the SgACC region—as investigated by Niida et al. [36]-possesses a higher density of serotonin transporters than other cortical areas [61]. It follows that a reduction in SgACC volume would be associated with depression. However, the midline, temporal lobe and prefrontal regions of the brain investigated by Hellewell et al. [33] and Liu et al. [35] have no known association with MDD, which could explain the relatively weaker AUC performance observed in these brain regions.
Differences in the medication status of the cohort investigated may have also affected the consistency of the results. For the studies investigating BDNF, Chiou and Huang [25] featured a medication-naive cohort, whereas Karlović et al. [27] did not. The heterogeneity in study design may have contributed to differences in the observed magnitude of diagnostic sensitivity of serum BDNF in persons with MDD.
Limitations
This inaugural review has comprehensively synthesised available literature reporting on biomarkers for the diagnosis of MDD. This includes both conventional biomarkers (e.g., blood, neuroimaging) and novel biomarkers, such as measures of auditory, vocal and skin characteristics. This review also delineates the accuracy and sensitivity of these measures and provides an update on the evidence supporting select biomarkers as candidates for diagnostic markers in MDD.
However, this study has several limitations. Given the small sample size of most of the included studies, the results may not be generalizable to a larger population. The casecontrol design of studies included in our review allows associations to be drawn between the biomarkers and MDD but is insufficient in establishing a causal relationship between both factors. Additionally, the decision to include both medication-naive and non-medication-naive cohorts was made due to the limited number of articles. Thirty-eight percent of the studies, and only one of 6 (17%) best performing biomarkers selected featured a medication-naïve population. This limits direct comparison between medication-naïve and non-medication-naïve cohorts. Geographical diversity of the studies was also skewed towards the Asian population, with a vast majority of studies originating from China (18/42). This over-representation may reduce the generalisability of the results. Due to the language capability of the researchers, the search was limited to only papers that were available in English and thus may have excluded other relevant papers reported in a different language. Lastly, the search strategy was restricted to the past 10 years (i.e., 2013-2023), owing to the rise in novel biomarkers used to detect MDD. This restriction may lead to the exclusion of certain biomarkers investigated more than 10 years ago.
Implications for future practice and research
This review may aid psychiatrists, psychologists, allied health professionals and service providers in implementing biological tests to assess MDD. While our review herein cannot confirm the causal association between a given biomarker and diagnosis of MDD, the available evidence supports that there are multiple biomarkers closely related to the hypothesised pathophysiology of MDD, which provides candidate markers for further investigation. Future research should focus on the best performing biomarkers and evaluate their combined sensitivity/specificity when utilised together. This can be conducted through machine learning methods, which have been increasingly incorporated into the classification of biomarkers for psychiatric conditions [62]. For example, machine learning methods extract common characteristics from combined data sets and incorporate them into supervised classifiers, allowing for the evaluation of similar data from different studies [63]. Recent examples of machine learning methods include the use of support vector machine and k-nearest neighbours in the evaluation of fNIRS signals as a neuroimaging marker for MDD [64].
Future research could also restrict study inclusion based on its medication-naive status to address this confounding effect. Future studies may also translate to a wider demographic by investigating the accuracy of diagnostic biomarkers in non-adult populations, such as adolescent or geriatric depression.
Conclusion
This systematic review identifies the variety of objective biomarkers used to diagnose MDD in the adult population in the past 10 years. The variety of biomarkers available range from blood, neuroimaging/neurophysiology, urine, dermatological, voice, CSF, hearing, executive dysfunction to motor activity, with varying levels of effectiveness. The most accurate biomarkers identified-MYT1 gene in blood [26]; 5-HT1A auto receptor binding in the dorsal and median raphe in PET neuroimaging/neurophysiology [22]; urinary albumin, AMBP, HSPB, APOA1 [45]; NSE in CSF [51]; microRNA in blood and CSF [54]-were intrinsically related to the pathophysiology of MDD.
It should be noted that clinical research pertaining to MDD biomarkers is in its infancy. Most of the leading hypotheses are based on results from cross-sectional research, treatment studies or symptomatology studies, which cannot determine causality [13]. To this date, there are also no biomarkers that have been approved for diagnostic purposes [65]. Therefore, the utility of this systematic review lies in highlighting biomarkers of significant performance and the theoretical models of depression that justifies its efficacy. Potential for future research lies in investigating the joint sensitivity of the best performing biomarkers identified via machine learning methods, establishing the causal effect between the biomarkers and MDD, adjusting for the medication-naïve status of the investigated population, and widening the inclusion criteria to increase the generalisability of the results.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation
Supplement
Print
