Metabolic phenotyping of opioid and psychostimulant addiction: A novel approach for biomarker discovery and biochemical understanding of the disorder

Despite the progress in characterising the pharmacological profile of drugs of abuse, their precise biochemical impact remains unclear. The metabolome reflects the multifaceted biochemical processes occurring within a biological system. This includes those encoded in the genome but also those arising from environmental/exogenous exposures and interactions between the two. Using metabolomics, the biochemical derangements associated with substance abuse can be determined as the individual transitions from recreational drug to chronic use (dependence). By understanding the biomolecular perturbations along this time course and how they vary across individuals, metabolomics can elucidate biochemical mechanisms of the addiction cycle (dependence/withdrawal/relapse) and predict prognosis (recovery/relapse). In this review, we summarise human and animal metabolomic studies in the field of opioid and psychostimulant addiction. We highlight the importance of metabolomics as a powerful approach for biomarker discovery and its potential to guide personalised pharmacotherapeutic strategies for addiction targeted towards the individual's metabolome.

mesolimbic pathway elicited by drug use. However, a detailed and refined understanding of the biochemical processes underlying addiction and the nature of the profound interpersonal variability in drug responses and progression of the disorder is still unclear.
The observation that addicted individuals present a range of metabolic abnormalities led to the notion of drug addiction as a 'metabolic disease' (Dole & Nyswander, 1967). Thus, a global investigation of the biochemical perturbations characteristic of the disease may be successful in providing mechanistic insights into disease states and progression. Metabolomics uses high-resolution analytical chemistry techniques to simultaneously measure a large number of low MW molecules in a biological sample. This results in large datasets where the variables (i.e. metabolites) largely outnumber the observations (i.e. mice or human participants). As such, appropriate dimensionality reduction techniques are necessary to analyse the entire metabolic profile in relation to an outcome of interest (Worley & Powers, 2013).
Multivariate models can be constructed to predict class membership (e.g.,disease state) or a continuous response variable (e.g. behavioural data) from linear combinations of the original variables (Saccenti et al., 2014). The predictive ability and significance of the model can then be assessed through cross-validation and permutation testing, respectively. While acknowledging the interrelation of metabolites, multivariate methods allow for assessing the 'weight' or contribution of each metabolite to the overall predictive model (Saccenti et al., 2014), aiding the identification of potential biomarkers. This unique biochemical fingerprint, referred to as the metabolome, reflects the metabolic processes occurring in the biological system at the time of analysis and its overall metabolic status (Kosmides et al., 2013). In addition, the metabolome contains exogenous molecules entering the system (e.g. dietary factors and xenobiotics) and products resulting from their breakdown. Drug addiction is a unique disorder in that it arises from the combination of genetic risk factors and exposure to an exogenous substance (i.e. drug of abuse). By capturing the 'metabolic phenotype' originating from both endogenous processes and the interaction with exogenous molecules (Figure 1), metabolomics provides a unique technique to investigate the biochemical basis of addiction.
Metabolomic approaches are also a powerful tool for biomarker discovery. As the fourth most costly mental disorder in the European Union (Gustavsson et al., 2011), addiction is a major public health issue with serious socio-economic implications and efforts continue to be made to improve the diagnosis and management of this disorder. The diagnosis of drug addiction is based solely on identifying characteristic symptoms and behaviours in accordance with the Diagnostic and Statistical Manual of Mental Disorders Fifth Edition (American Psychiatric Association, 2000). No reliable diagnostic test currently exists for primordially predicting drug addiction vulnerability and for identifying individuals at risk of relapse or at risk of co-morbidity. When pharmacological options for managing addiction symptomology are available (e.g. methadone/buprenorphine for opioid withdrawal and naloxone for craving), efficacy is often limited and responses are highly variable. It also remains impossible to predict the efficacy and potential side effects of pharmacotherapy on an individual basis. These challenges demonstrate the need for quantitative biomarkers to predict an individual's addiction risk, disease progression, relapse vulnerability and response to interventional strategies. Pharmacometabolomics is a branch of metabolomics whereby an individual's baseline metabolic phenotype is used to predict their handling and response to a pharmacological intervention (Kaddurah-Daouk & Weinshilboum, 2014). As our understanding in F I G U R E 1 Each level of the 'omics' cascade gives a different level of insight into the phenotype. The interaction of each component (genome, transcriptome, proteome and metabolome) with the environment (nutrition, stress and drugs) influences the resulting phenotype and can contribute to disease. The size of the green arrows indicates the influence of the environment increases on each level of the omics cascade and is highly reflected by the metabolome. Although each omics technique can reveal important diagnostic and prognostic biomarkers of disease, the ability of metabolomics to capture both endogenous (i.e. genetic) and exogenous (i.e. environmental and drug-related) influences on the observed disease phenotype, its chemical diversity and dynamic nature, suggests that metabolic biomarkers may better represent the resulting phenotype of drug abuse this area grows, a personalised approach to care delivery and clinical decision making in the management of substance abuse disorders may become possible.
This review summarises the findings of published studies in humans and rodent models investigating the biomolecular perturbations elicited by opioids and psychostimulants on the brain (target organ) and peripheral tissues/biofluids with translational value (e.g., blood, urine and hair). As our understanding in this area grows, a personalised approach to care delivery and clinical decision making in the management of substance abuse disorders may become possible.

| METABOLIC PHENOTYPING OF OPIOID ADDICTION
Opioids, such as morphine and heroin, are highly addictive substances. Their rewarding effects are mediated by their ability to induce dopamine transmission in the nucleus accumbens by relieving the inhibition of GABAergic interneurons on mesolimbic dopaminereleasing neurons in the ventral tegmental area (Spanagel & Weiss, 1999). The following section provides an overview of animal and human studies investigating the metabolic changes associated with the distinct stages of opioid (morphine and heroin) addiction.
Although the analysis of brain samples can help unravel biochemical pathways affected by repeated opioid administration Gao et al., 2007;Hu et al., 2012;Li et al., 2017) or involved in the reinforcing effects of the drugs (Meng et al., 2012), sequential sampling of plasma and urine samples allows for the identification of biomarkers of the different addiction states (i.e. euphoria, tolerance, abstinence and withdrawal, Liu et al., 2015;Zaitsu et al., 2014) and predictors of treatment outcome (i.e. response and relapse, Ning et al., 2018;Zheng et al., 2013). Human studies have also been conducted to understand the perturbations in the hair metabolome driven by heroin (Xie et al., 2016) and to investigate the metabolic changes induced by withdrawal from opioids . These studies are summarised in Table 1. 2.1 | Metabolic signature of morphine addiction 1 H-NMR spectroscopy-based studies of the metabolic abnormalities induced by repeated morphine administration on brain samples were conducted in rhesus monkeys  and rodents (Gao et al., 2007;Hu et al., 2012). Significant disturbances in the glutamine-glutamate-GABA (Gln-Glu-GABA) axis, which are markers of oxidative stress and involved in neurotransmission, were a common finding. The specific changes reported varied depending on the species and the brain region considered and are reviewed in Table 2. Some discrepancies regarding the direction of change of these metabolites may also be underpinned by differences in experimental design (e.g. dose and length of drug administration; see Table 1). Disturbances in the equilibrium state between GABA, Glu and Gln, with a general increase of GABA and decrease in Glu, were consistent with microdialysis studies showing increased GABA and decreased Glu within the medial prefrontal cortex (Ramshini et al., 2019), nucleus accumbens (Sun, Yang, et al., 2011) and hippocampus (Kang et al., 2006) in response to morphine. Such changes were ascribed to alterations in tricarboxylic acid (TCA) cycle activity and enhanced conversion of Glu into GABA by GAD. A decrease in GABA degradation has also been suggested (Gao et al., 2007). The shift in the Gln-Glu-GABA equilibrium state observed in these studies may also be the result of neuroadaptations caused by the effect of morphine on GABAergic neurotransmission, which is known to mediate the rewarding properties of opioids. This is supported by evidence suggesting that GABA and activation of the GABAergic system attenuates the reinforcing effects of drugs of abuse via its modulatory effect on the mesolimbic dopaminergic pathway (Tsuji et al., 1996;Westerink et al., 1996) and that its disruption is involved in the development of tolerance and dependence to opioids Sepúlveda et al., 2004). These metabolic changes were not evident until Day 10 of morphine administration, suggesting that long-term adaptive mechanisms underlie these alterations (Gao et al., 2007). The membrane constituent phosphocholine and the phosphoinositol precursor myo-inositol were dysregulated in response to morphine exposure, indicating an effect of morphine on membrane integrity.
Myo-inositol is also an osmolyte highly expressed in glial cells.
Changes in the abundance of myo-inositol and the neuronal marker Nacetylaspartate (NAA) may reflect glial hypertrophy and altered neuronal morphology and activity. Glial cells are emerging as an important player in addiction pathophysiology due to their role in supporting neurotransmission and brain energy metabolism (Miguel-Hidalgo, 2009). Morphine has been shown to affect neuronal maturation in vitro by modulating astrocytic proliferation (Stiene-Martin et al., 1991). Irrespective of species and brain regions considered, an increase in lactic acid was observed in all studies Hu et al., 2012). Lactic acid is the end product of anaerobic cellular metabolism and is produced when energy demand exceeds the rate of oxidative metabolism. Elevated lactic acid indicates tissue damage and impaired pyruvate oxidation (Veech, 1991). Given that lactate can only be completely oxidised in mitochondria, increased lactic acid observed in these studies is likely to be an indication of mitochondrial dysfunction, energy metabolism impairment, oxidative stress and/or up-regulation of the enzyme LDH. Consistently, metabolites related to oxidative stress such as the antioxidants glutathione, taurine and creatine also showed a large deviation from controls following repeated morphine administration. Feng et al. (2013) suggested hippocampal mitochondrial damage and decreased mitochondrial DNA copy number as a hallmark of addiction. Using cultured rat pheochromocytoma cells and mouse neurons treated with morphine, the authors showed that oxidative stress caused by morphine administration led to mitochondrial damage and autophagy. Given the involvement of mitochondria in synaptic remodelling, mitochondrial dysfunction is likely to have downstream effects on synaptic plasticity and neurotransmission, with an inevitable downstream impact on the addiction cycle ( Figure 2). In support of this hypothesis, the upregulation of ROS in the hippocampus, as a result of morphine   administration, was shown to elicit increased inhibitory and decreased excitatory synapses, whereas the antioxidant compound plateletderived growth factor (PDGF) reverses the synaptic effects of morphine (Cai et al., 2016). Similarly, the antioxidative compound thioredoxin-1 (Trx-1) inhibits morphine-induced conditioned-place  Table 1). '"' indicates increase, '#' indicates decrease, '□' indicates rhesus monkeys , '□' indicates rats , '□' indicates rats (Gao et al., 2007) and '*' indicates disagreement between Hu et al. (2012) and Gao et al. (2007). NMR-based metabolic profiling was used to study various brain regions from morphine-treated rhesus monkeys and rats undergoing detoxification with the long-acting opioid methadone or the α 2 adrenoceptor agonist clonidine Hu et al., 2012).
Upon detoxification, the majority of morphine-induced metabolic variation was normalised to baseline Hu et al., 2012), although several metabolites remained altered in specific brain regions (Table 3). Although clonidine was generally more effective than methadone in reversing the biochemical effects of morphine in both species Hu et al., 2012), rats showed a more profound reduction in withdrawal symptoms in response to methadone compared with clonidine . Advanced correlation analysis is warranted to investigate causality between the behavioural effect of pharmacotherapeutic interventions and their effectiveness in restoring predose biochemical profiles.
The effect of 6 days of methadone-aided detoxification on the plasma metabolome was investigated in a human study ). An liquid chromatography electrochemical array platform was used to measure purine, monoaminergic and redox metabolites in 14 opioid-dependent individuals undergoing methadone detoxification and 10 nondrug users. Opioid-dependent participants were given methadone orally at 9:00-10:00 AM daily, and blood samples were collected on Days 2 and 3 at 10:00-11:00 AM. The ratio of glutathione (GSH)/oxidised GSH and the antioxidants α-tocopherol and F I G U R E 2 Diagram of metabolic alterations caused by drugs of abuse. Metabolites that are altered in response to drug exposure provide information on the underlying cascade of events leading to addiction. As most of the changes leading to an established addiction state are likely to involve adaptive mechanisms, it is important to investigate and discriminate between the short-term effects of acute drug exposure and the long-term, compensatory changes resulting from chronic drug administration (i.e. dependence) on the metabolome. A general mechanism of action of both opioids and psychostimulants seems to involve a short-term increase in energy demand, which leads to long-term energy depletion, mitochondrial dysfunction and oxidative stress. These long-term effects contribute to a cascade of events that feed the cycle of metabolic and pathophysiological derangements characterising addiction, as shown with the backwards arrows. 3-HB, 3-hydroxybutyric acid; Gln, glutamine; Glu, glutamate; NAA, N-acetylaspartate; TCA, tricarboxylic acid

| Metabolic signature of heroin addiction
Note: Studies conducted in rhesus monkeys  and rats ) (see protocol in Table 1 (Galli et al., 1993). A 2-day withdrawal from heroin was sufficient for catecholamines to return to baseline levels, suggestive of an autonomic readjustment during abstinence. In contrast, histidine was observed to decrease upon withdrawal, whereas phenylalanine, tryptophan and N-acetyl-5-HT increased, pointing to adaptive mechanisms involving the 5-HT system.
These findings are supported by the results of a human study that investigated the hair metabolome of heroin abusers (Xie et al., 2016).
The heroin group showed increased concentrations of sorbitol and cortisol and decreased concentrations of arachidonic acid, GSH, linoleic acid and myristic acid (Xie et al., 2016). The impact of heroin on the HPA axis via opioid signalling may underlie the variation noted in cortisol. A decrease in the free fatty acids, arachidonic acid, linoleic acid and myristic acid is consistent with increased energy production, as seen in the mice exposed to heroin (Zheng et al., 2013).

| METABOLIC PHENOTYPING OF PSYCHOSTIMULANT ADDICTION
Psychostimulants increase striatal dopamine concentrations by increasing dopamine levels in the nucleus accumbens. Cocaine inhibits the reuptake of dopamine in mesolimbic dopaminergic neurons projecting from the ventral tegmental area to the nucleus accumbens by blocking dopamine transporters located presynaptically, whereas amphetamine and methamphetamine facilitate presynaptic dopamine release. Psychostimulants also stimulate the release of other monoamines such as 5-HT and noradrenaline (Kim et al., 2019). Rodent metabolomic studies using brain tissue have explored the abnormalities in central metabolic processes driven by acute (Kaplan et al., 2013;Li et al., 2012;Olesti et al., 2019) and repeated psychostimulant administration Bu et al., 2013;Kong et al., 2018;Li et al., 2014Li et al., , 2012Lin et al., 2019;McClay et al., 2013).

| Metabolic signature of cocaine addiction
The central metabolic signature of rats undergoing both acute and repeated cocaine administration was investigated using NMR spectroscopy-based metabolomics . Seven-day cocaine-conditioned-place preference increased Glu and GABA in the nucleus accumbens and enhanced GAD activity, implying altered neurotransmission along the Gln-Glu-GABA axis. The reported ability of cocaine to elicit Gln production by glial cells may explain these observations (S a Santos et al., 2011). These alterations were evident after repeated administration but not after a single dose.
Such changes are consistent with what observed with morphine (Gao et al., 2007) indicating a common adaptive, long-term response to chronic drug exposure ( Figure 2). Indeed, adaptive functional changes are known to occur at glutamatergic synapses in the nucleus accumbens in response to repeated cocaine administration (Maze et al., 2010). As these changes mirror behavioural sensitisation (Russo et al., 2009;Thomas & Malenka, 2003;Ungless et al., 2001), they represent a key molecular component of the addictive properties of cocaine. Lactate, which is produced via anaerobic metabolism and can be metabolised through the TCA cycle, was decreased after a single dose but increased after chronic administration, indicating a long-term compensatory change in energy metabolism in response to cocaine exposure. An NMR-based study by Kong et al. (2018) suggested that disturbances in energy metabolism may be explained by epigenetic mechanisms. Cocaine-conditioned mice exhibited significantly higher concentrations of nicotinamide mononucleotide and nicotinamide adenine dinucleotide in ventral tegmental area and nucleus accumbens. Nicotinamide mononucleotide and nicotinamide adenine dinucleotide are produced from nicotinamide by the enzyme nicotinamide phosphoribosyltransferase (NAMPT) and play a role in energy metabolism. Nicotinamide phosphoribosyltransferas was shown to be up-regulated in cocaine-conditioned mice via an epigenetic mechanism involving nicotinamide adenine dinucleotidedependent histone deacetylase sirtuin 1 (SIRT1), thus pointing towards a role for SIRT1 in epigenetic regulation of genes, such as nicotinamide phosphoribosyltransferase, that control energy metabolism (Kong et al., 2018). Altered creatine levels may also be indicative of a shift in normal energy metabolism . The observed dysregulation of creatine (increased in nucleus accumbens and decreased in striatum) and taurine (increased in both areas) has been proposed to reflect oxidative damage. Although an increase in taurine and decrease in its metabolic precursor cysteine could indicate the induction of a brain protective mechanism following cocaine administration, a single dose of cocaine induced a reduction in taurine in nucleus accumbens, suggesting a short-term depletion of its antioxidant capacity before the long-term adaptive increase. The concentrations of N-acetylaspartate, a marker of neuronal density synthesised in mitochondria, were increased in nucleus accumbens and striatum after both acute and chronic cocaine administration, pointing to an immediate effect of the drug on mitochondrial dysfunction. Finally, membrane damage is indicated by alterations in myo-inositol, glycine and choline concentration, which were affected by a single dose of cocaine .
A quantitative evaluation of the global neurobiochemical profile of cocaine-treated rats was achieved by ion mobility mass spectrometry (Kaplan et al., 2013). Acute cocaine administration significantly reduced thalamic and striatal glucose, with the greatest decrease seen in the thalami. In the frontal cortex, cocaine exposure increased glucose content, indicating region-specific shifts in glucose metabolism following cocaine treatment. The availability of 5-HT, noradrenaline, glucose, dopamine, 3,4-dihydroxyphenylacetic acid (DOPAC) and 5-HIAA in the thalamus, striatum and prefrontal cortex was also altered as a result of cocaine exposure (Kaplan et al., 2013), indicating abnormalities in neurotransmission induced by the drug. This is consistent with the reported ability of psychostimulants to promote the release of other monoamines such as 5-HT and dopamine (Kim et al., 2019), which is thought to be part of the underlying mechanism of cocaine reward (Sora et al., 2001).
Abnormalities in neurotransmitter metabolism were also reported in brain and blood samples of rats administered with a single dose of cocaine (Olesti et al., 2019) and in the serum of rats undergoing a cocaine self-administration protocol (Goodwin et al., 2014). Using a targeted liquid chromatography-mass spectrometry approach, significant elevations were noted in acetylcholine (ACh) in the prefrontal cortex; valine, leucine, GABA, Glu, choline, ACh, carnitine, acetylcarnitine, creatine, creatinine and adenosine in the hippocampus; and choline and adenosine in the striatum. In the cerebellum, Glu, choline, , carnitine and creatinine were increased. In plasma, choline and creatine were increased, whereas creatinine was decreased. Some of these alterations are likely to reflect pharmacological effects of cocaine on the muscle and the brain. For example, cocaine-induced rhabdomyolysis (muscle injury) can lead to altered creatine and creatinine metabolism, with downstream consequences on the brain highenergy phosphate system (Lyoo et al., 2003), whereas the reported inhibition of ChAT by cocaine (Wilson et al., 1994)  increase in ROS and nitric oxide production, two factors that contribute to liver injury in cocaine-dependent individuals (Aoki et al., 1997).
In contrast, N-ε-acetyl-L-lysine could provide acetylated lysine residues for the epigenetic changes underlying cocaine-based reinforcement.
Biochemical modulations in plasma and urine were assessed in rats undergoing cocaine-conditioned-place preference (Zaitsu et al., 2014). Although significant metabolic changes were found in plasma of cocaine-treated rats (higher L-threonine and n-propylamine; lower cysteine and spermidine), no metabolic variation was identified in urine following treatment relative to controls. In a separate rat study by Yao et al. (2013), clear differences were observed in the urine between cocaine and a control group. Several factors may account for these differences, including differences in animal strain, cocaine dose, length and means of drug administration as well as in chromatographic techniques (liquid chromatography-mass spectrometry n Yao et al., 2013, vs. gas chromatography-mass spectrometry in Zaitsu et al., 2014). Additionally, there is ample evidence that cocaine metabolism differs profoundly between mice and rats, possibly explaining the more marked behavioural and hepatotoxic phenotype in response to cocaine in mice relative to rats (Thompson et al., 1979).
The persistence of biochemical modulations following acute withdrawal from cocaine was studied in the nucleus accumbens of rats 2, 24 and 48 h following a single exposure . and to decrease cocaine self-administration (Zhang, Xue, et al., 2016).
Importantly, although GABA was not dysregulated after a single cocaine exposure nor after repeated administration, it was significantly altered in mice exposed to cocaine-conditioned-place preference, highlighting that contextual conditioning of a drug is able to induce metabolite changes in the brain, which are independent of the effect of the drug administration per se. Therefore, the inclusion of a group of animals treated with the drug but not undergoing conditioned-place preference should be considered when investigating cocaine-conditioned-place preference effects on the metabolome.
In a separate study, the metabolic perturbations in energy supply (creatine, creatinine and adenosine), oxidative stress (GSH and spermidine), neurotransmission (pyroglutamic acid, Gln, Glu and GABA), mitochondrial function (carnosine) and membrane integrity (choline) induced by cocaine self-administration were still evident after 1-day abstinence in prefrontal cortex, striatum and nucleus accumbens but normalised at Week 3 in all brain areas except the striatum (Zhang, Chiu, et al., 2016). The finding that cocaine use leads to long-term metabolic abnormalities in the striatum may explain the presence of drug craving long after withdrawal from the drug (Volkow et al., 2006).
Understanding the metabolic changes occurring upon drug withdrawal is important to shine light on the biochemical mechanisms underlying recovery from addiction. A human study involving 18 cocaine-dependent individuals investigated the effect of 2 weeks of abstinence on their plasma metabolic profile ).
This study found significant alterations in purine and tryptophan metabolism, as reported in opioid-dependent individuals upon detoxification ), but no changes in oxidative stressrelated metabolites. It is generally believed that oxidative stress might be applicable for acute rather than more prolonged intoxication, which may explain these findings. Plasma metabolic profiles from cocainedependent individuals were correlated with their addiction severity index (ASI) drug scores. Specifically, N-methyl-5-HT accounted for 62% of variance in severity of drug abuse based on addiction severity index drug score,and combined with xanthine it accounted for 73%.
These findings implicate plasma N-methyl-5-HT and xanthine as good candidate biomarkers for assessing and predicting addiction severity.
In accordance with Mannelli et al. (2009), no significant changes in 5-HT metabolism were observed, indicating that the biosynthesis of 5-HT from tryptophan was unaffected by cocaine. Instead, the increase in N-methyl-5-HT suggests a dysregulation of the enzyme that metabolises 5-HT to N-methyl-5-HT following chronic cocaine exposure. This hypothesis requires further investigation and future studies with larger sample sizes and more appropriate controls are warranted.

| Metabolic signature of methamphetamine addiction
The animal studies investigating the metabolic effects of methamphetamines have reported contradicting findings. Similar to what was observed for cocaine, several studies reported disrupted energy metabolism as a consequence of chronic methamphetamines intake (Kim et al., 2019;Shima et al., 2011;Zheng et al., 2014). A significant depletion of TCA cycle intermediates (Shima et al., 2011;Zheng et al., 2014) and branched-chain amino acids (Kim et al., 2019;Zhenget al., 2014) was observed in the blood and urine of rats repeatedly exposed to methamphetamines. On the other hand, no change in TCA intermediates was detected in plasma and urine after methamphetamine-conditioned-place preference training (Zaitsu et al., 2014). As glycolysis is down-regulated upon drug deprivation (Muneer et al., 2011), it can be speculated that the metabolic effect of methamphetamine on the TCA cycle is due to acute withdrawal and cannot be elicited by a chronic conditioned-place preference paradigm. Collectively, these results suggest that different drug administration protocols have distinct effects on plasma and urine metabolic signatures, possibly as a result of adaptive mechanisms to repeated drug use and/or to the presence or absence of contextual learning (i.e. conditioned-place preference). A common finding to these studies was altered lipid metabolism. Reduced plasma lauric acid and increased urinary stearic acid were induced by methamphetamineconditioned-place preference (Zaitsu et al., 2014). In the absence of contextual learning, methamphetamine elicited changes in the β-oxidation of free fatty acids and the formation of 3-hydroxybutyric acid, indicative of altered lipid turnover, as well as changes in glycerophospholipids and sphingolipids, suggestive of membrane breakdown, in both urine and blood, although the direction of change is unclear (Shima et al., 2011;Zheng et al., 2014). Changes to lipid metabolism may represent a compensatory mechanism to meet the increased energy demand induced by methamphetamine exposure.
Methamphetamine was seen to elicit different effects on some lipids compared with heroin. For example, serum myo-inositol and myo-inositol-1-phosphate was increased in response to heroin (Zheng et al., 2013) but reduced in response to methamphetamine (Zheng et al., 2014). Further evidence of altered energy metabolism in response to methamphetamine comes from a study performed on Drosophila melanogaster (Sun, Li, et al., 2011). Flies fed on a methamphetamine-supplemented diet had lower circulating trehalose, the major blood sugar in the Drosophila, indicating higher metabolic rates and/or increased glycolysis. Interestingly, trehalose supplementation increased the flies' lifespan, indicating that methamphetamine toxicity is linked to a depletion of energy cofactors and that replenishing these cofactors may attenuate the negative effects of the drug (Sun, Li, et al., 2011).  (Moszczynska et al., 2004;Thomas et al., 2008). The decrease of Gln and Glu mirrored the decrease of GABA and 2-oxoglutarate, which is partly due to decreased TCA cycle activity and, possibly, increased Glu uptake. Succinic acid semialdehyde levels increased, consistent with its role as an intermediate of GABA catabolism. A decrease in Glu and Gln was also observed in the nucleus accumbens and dorsal hippocampus by Lin et al. (2019), although these metabolites, along with the amino acid and excitatory neurotransmitter aspartate, were down-regulated in the ventral hippocampus. Together, these findings suggest that the disturbance to Gln-Glu-GABA axis in the brain may be involved in the behavioural sensitisation to methamphetamine. Together with a general increase in nucleotides like ADP, GMP and AMP in nucleus accumbens and dorsal hippocampus (but a decrease in the ventral hippocampus), these findings point towards an alteration in energy homeostasis in a brain region-dependent manner. A reduction in the antioxidant GSH was reported in both studies (Bu et al., 2013;Lin et al., 2019). Moreover, a reduction in N-acetylaspartate and an increase of phosphocholine were observed in brain regions of chronically treated rats, indicating that oxidative damage was present alongside neuronal and mitochondrial dysfunction (Bu et al., 2013).
The increase in homocysteine, an amino acid and precursor of methionine, could be regarded as an indicator of apoptosis and neuronal hypersensitivity to excitation as well as DNA damage (Kruman et al., 2000). Moreover, the increase in homocysteine may be caused by the inhibition of methionine synthesis by methamphetamine (Chandra et al., 2006), leading to changes in DNA methylation.
Increased myo-inositol and phosphocholine are consistent with membrane disruption (Bu et al., 2013) and may reflect cell death due to the severely neurotoxic properties of methamphetamine (Zheng et al., 2014). Phospholipids were generally down-regulated in nucleus accumbens and dorsal hippocampus but up-regulated in the ventral hippocampus (Lin et al., 2019), suggesting region-specific effects of methamphetamine treatment. Finally, Bu et al. (2013) found no significant correlation between metabolic disruptions and locomotor sensitisation behaviour. Given that locomotor sensitisation is linked to increased craving and vulnerability to relapse (Robinson & Berridge, 1993;Vanderschuren & Pierce, 2020), this finding suggests that these metabolites may have poor predictive potential. Future studies are needed to determine whether a conditioned-place preference design and the resulting contextual learning are needed in order to find a significant correlation with locomotor sensitisation.
Mass spectrometric analysis of the metabolic effects of repeated methamphetamine self-administration in rat urine and hair revealed abnormalities in the metabolism of mineralocorticoid, fatty acid amides and mitochondrial fatty acid oxidation (Choi et al., 2017). In the urine, ion features corresponding to potential urine markers of methamphetamine addiction were detected but only putatively assigned. In the hair, a decrease in deoxycorticosterone suggests altered central production of neurosteroids (Mellon & Griffin, 2002), whereas an increase in carnitine and acylcarnitines is indicative of elevated metabolic capacity by mitochondrial oxidation of fatty acids.
The reduction observed in the fatty acid amides oleamide and stearamide, known to interact with endocannabinoid, glutamatergic and GABAergic signalling, may indicate a modulation of central neurotransmission. The authors concluded that methamphetamine evoked more dramatic metabolic changes in the hair than in the urine. This may reflect the greater stability of hair and longer accrual of biochemical information related to drug-induced metabolic perturbations compared with urine samples. As such, hair metabolomics should be considered as a non-invasive way to profile the addiction status of an individual.
A study conducted on the hair metabolome of methamphetamine users revealed additional abnormalities in the abundance of amino acids and lipids (Kim et al., 2020). The amino acids arginine and methionine, both known ROS scavengers (Liang et al., 2018;Luo & Levine, 2009), were down-regulated in the hair of drug abusers, which may indicate higher susceptibility to oxidative stress. Lower phosphatidylcholines, but higher lysophosphatidylcholines and sphingomyelin, amino acids and other biomolecules (Lin et al., 2019). An increase in excitatory amino acids (Gln, asparagine and Glu) and a decrease in inhibitory amino acids (glycine and alanine) point to an increase in brain activity induced by the drug. ACh was also elevated (whereas choline was reduced) in the plasma of methamphetamine users, collected at fasting (Lin et al., 2019). Similar findings associated with altered energy metabolism, steroid biosynthesis, amino acid and fatty acid metabolism were reported in response to other types of psychostimulants, such as MDMA, amphetamine and mephedrone in human plasma (Nielsen et al., 2016;Steuer et al., 2020), suggesting a common mechanism of action.
Despite the lack of human studies investigating the metabolic consequences of abstinence from methamphetamines, three of the rodent studies presented above examined the effects of withdrawal on the metabolic phenotype. Shima et al. (2011) reported that the methamphetamine-induced plasma and urinary alterations (mainly associated with altered TCA intermediates, amino acids and fatty acids) persisted 24 h after the last drug administration but were not evident at 96 h. In a separate study, 2 days of detoxification was generally sufficient to restore the serum and urinary metabolic profile of methamphetamine-treated rats to pretreatment levels (Zheng et al., 2014). In the serum, several amino acids including Gln, Glu and aspartate (involved in neurotransmission and energy metabolism) were completely restored, whereas fatty acids like arachidonic acid, decanedioic acid, stearic acid and glycerol-3-phosphate (involved in membrane stability and energy metabolism) were only partially recovered. Isoleucine, palmitic acid, creatinine, citrate and 2-oxoglutarate did not return to pretreatment values. In urine, only lactate was persistently altered after withdrawal (Zheng et al., 2014).
Similarly, Kim et al. (2019) reported that 12-or 24-h abstinence resulted in the reversal of the metabolic abnormalities in glycerophospholipids, sphingolipids and most amino acids elicited by methamphetamine self-administration. However, the concentrations of aspartate, Glu and glycine remained significantly lower than controls 12 and 24 h after the last methamphetamine administration.
Acetylcarnitines and biogenic amines were further altered during abstinence, suggesting that the dynamic response of the metabolome to methamphetamine exposure spans at least 24 h following the last administration. Moreover, 5-HT exhibited a shortterm increase (immediately after self-administration) but a long-term decrease after methamphetamine exposure (12 and 24 h after the last exposure). Similarly, pathway analysis demonstrated a decrease in the phenylalanine, tyrosine and tryptophan biosynthesis and in the valine, leucine and isoleucine biosynthesis pathways immediately after self-administration but an increase after 12 and 24 h relative to controls. These results highlight the importance of investigating the time profile of metabolic responses to drugs of abuse, which may provide biomarkers relevant for discriminating addiction states.
Moreover, the time when the metabolic change occurs may inform on whether the change is the result of a short-term molecular mechanism or of a longer term compensatory mechanism (e.g. transcriptional/epigenetics). Exploring these molecular mechanisms may help to guide the development of interventional strategies for clinical practice.

| FUTURE PERSPECTIVES
In recent years, findings from metabolomic studies have provided a metabolic perspective to the addiction cycle that complements our provides an objective measure that can inform personalised approaches to addiction to maximise the chances of recovery. Wellcharacterised longitudinal studies where the metabolic profile of an individual is measured at baseline, before the initiation of pharmacological detoxification therapies, will allow to identify early predictive markers and personalised treatment strategies tailored to the metabolome of the individual patient. The prediction of addiction outcomes based on neuroimaging data has already been successful (Reske & Paulus, 2008). The use of metabolic markers quantified from non-invasive samples (e.g. urine and hair) could offer a more feasible and cost-effective method to bring precision medicine to clinical practice.

| CONCLUSION
In recent years, the misuse of prescription opioids in the United States has led to what is currently known as the 'opioid epidemic', a public health crisis costing $26 billion to the US healthcare system and 16,000 deaths in 2013 alone (Florence et al., 2016). Identifying

CONFLICT OF INTEREST
The authors declare no conflicts of interest.