Dysfunction in the monoamine systems of serotonin (5-HT), norepinephrine (NE) and dopamine (DA) may causally be related to major depressive disorder (MDD). Monoamine depletion studies investigate the direct effects of monoamines on mood. Acute tryptophan depletion (ATD) or para-chlorophenylalanine (PCPA) deplete 5-HT, acute phenylalanine/tyrosine depletion (APTD) or alpha-methyl-para-tyrosine (AMPT) deplete NE/DA. Available depletion studies found conflicting results in heterogeneous populations: healthy controls, patients with previous MDD in remission and patients suffering from MDD. The decrease in mood after 5-HT and NE/DA depletion in humans is reviewed and quantified. Systematic search of MEDLINE and EMBASE (1966–October 2006) and cross-references was carried out. Randomized studies applying ATD, PCPA, APTD or AMPT vs control depletion were included. Pooling of results by meta-analyses was stratified for studied population and design of the study (within or between subjects). Seventy-three ATD, 2 PCPA, 10 APTD and 8 AMPT studies were identified of which 45 ATD and 8 APTD studies could be meta-analyzed. 5-HT or NE/DA depletion did not decrease mood in healthy controls. 5-HT or NE/DA depletion slightly lowered mood in healthy controls with a family history of MDD. In drug-free patients with MDD in remission, a moderate mood decrease was found for ATD, without an effect of APTD. ATD induced relapse in patients with MDD in remission who used serotonergic antidepressants. In conclusion, monoamine depletion studies demonstrate decreased mood in subjects with a family history of MDD and in drug-free patients with MDD in remission, but do not decrease mood in healthy humans. Although depletion studies usefully investigate the etiological link of 5-HT and NE with MDD, they fail to demonstrate a causal relation. They presumably clarify a vulnerability trait to become depressed. Directions for further investigation of this vulnerability trait are proposed.
Major depressive disorder (MDD) is characterized by a lowered mood. MDD is a disabling disease which affects 20% of the world's population.1 MDD is often treated with antidepressants (ADs), mostly selective serotonin reuptake inhibitors (SSRIs), serotonin norepinephrin reuptake inhibitors (SNRIs), norepineprine reuptake inhbitors (NERIs) or tricyclic antidepressants (TCAs).2, 3, 4, 5
The working mechanism of AD is believed to be either by (1) increased neurotransmission by increased synaptic levels of serotonin, norepinephrine (NE) and dopamine (DA) (monoamines) or (2) specific agonistic effects on serotonin or NE (sub-)receptors. The increased levels of monoamines were discovered in the late fifties, when the TCAs and Monoamine-oxidase A inhibitors (MAO-I) appeared to effectively treat MDD. This discovery led to the monoamine hypothesis: MDD might etiologically be explained by a deficiency in monoamine neurotransmitters: serotonin (5-HT), NE or (to a lesser degree) DA. The monoamine systems in the brain have complex interactions. Therefore, the current, less pertinent view is that the monoamine hypothesis only partially explains MDD and the response to AD.6, 7, 8, 9, 10
Depletion of the available 5-HT, NE and/or DA is used as a model to test the involvement of monoaminergic systems in MDD. Two reviews recently described and reviewed the techniques of monoamine precursor depletion and enzyme-blocking methods.11, 12 In brief, 5-HT depletion can be achieved by rapidly lowering the levels of the essential amino-acid tryptophan, which cannot be synthesized by the body and must be ingested to enable formation of 5-HT. To achieve depletion, a tryptophan free amino-acid mixture is administered (acute tryptophan depletion (ATD)).13 Depletion of NE and DA uses the same concept (acute depletion of the essential amino acids phenylalanine/tyrosine (APTD)).14 As an alternative to induce a state of depletion, enzyme-blocking agents decrease the production of the monoamines. Para-chlorophenylalanine (PCPA) blocks 5-HT synthesis,15, 16 and alpha-methyl-para-tyrosine (AMPT) NE and DA synthesis.17 Since 1975 an increasing number of depletion studies have been conducted. Monoamine depletion showed differential effects in different study populations: healthy controls, healthy controls with a family-history of MDD, patients with MDD in remission using AD or after cessation of AD, and patients who had a current episode of MDD.
Previous reviews summarized the methodology of depletion tests.18 Others reviewed the findings of specific depletion tests across different psychiatric illnesses,11, 18, 19, 20 or the prediction of response to depletion.21, 22 Booij et al.21 presented a study that pooled results across studies individual subject data of six ATD studies (‘mega-analysis’) in order to investigate the mediating role of clinical, demographic and biochemical characteristics in the mood response to ATD in remitted subjects who previously had MDD. However, a clear summary of the mood effects of monoamine depletion across different populations is lacking. Except from the study of Booij et al., we are unaware of any attempt of pooling studies. Pooling is important because the small-sized depletion studies might not have detected small differences by a lack of power. Finally – as in drug research – pooling will quantify the weight of positive vs negative studies.
Therefore, we aimed to review the evidence for mood lowering properties of monoamine depletion studies as a model for MDD. Our main question was: does the depletion of monoamine (5-HT and NE/DA) systems lower mood in humans? A secondary question was: is the lowering of mood different across different populations? This paper reports our systematic review with a stratified meta-analysis of the mood effects of monoamine depletion studies.
Methods and materials
Design of the study
We included all available randomized prospective monoamine depletion studies (tryptophan and phenylalanine/tyrosine) and enzyme blocking studies (PCPA or AMPT) in humans. Included studies measured a change in mood after depletion. Two study designs were included. First, randomized within-subjects studies, where each subject was exposed to a true depletion vs a sham intervention at a second occasion. This way, each subject served as his/her own control. Second, randomized between-subjects studies, where two groups of subjects were compared, with the intervention applied to one group and a control intervention to the other group. We excluded animal studies and studies which selected patients with apparent co-morbidity (e.g. substance abuse or dependence, studies with only smokers, psychotic disorders, anxiety disorders). Additionally, we excluded studies in depression subtypes with a supposed different etiology (e.g. seasonal affective disorder, bipolar disorder). We also excluded studies that combined depletion with other interventions (e.g. sleep deprivation) or studies that did not report mood effects. Finally, we excluded studies recruiting a patient group with both unipolar and bipolar depression when >25% of subjects had bipolar depression and when no separate data for the unipolar group were provided.
Literature searches and selection
We searched PubMed from 1966 to 1 October 2006 and EMBASE from 1980 to 1 October 2006 using a comprehensive search strategy (Table 1). We retrieved additional references identified in previous reviews11, 12, 18, 19, 20, 23, 24, 25 and cross-references from identified studies. Two reviewers (HGR and NSM) independently selected articles based on the a priori inclusion and exclusion criteria as stated above. In case of doubt an article was retrieved and the full content was considered. In 85.4% the reviewers agreed on selection instantly. The initial kappa for agreement was 0.61 (95% confidence interval (CI): 0.52–0.71). Agreement was mainly influenced by a differential inclusion of research letters, studies in depression subtypes with a supposed different etiology and studies without apparent mood-scores in the abstract. Discrepancies in included studies were discussed between both reviewers until full consensus was reached.
Data extraction and assessment of the studies
Two authors (HGR and NSM) abstracted the retrieved studies as follows: population studied, study design (within-subjects or between subjects), applied mood scale(s), number of subjects (and – if applicable – number of subjects for each sequence of intervention and control depletion), male/female ratio, the intervention and its control condition (including dosage and duration). Furthermore, we extracted plasma levels of tryptophan, tyrosine or relevant levels of monoamine-metabolites before and after intervention and control condition and other relevant data that could influence the outcome of the study (e.g. different consistency of control drink, unclearly reported data, combined interventions). Primary outcome variables were changes in mood-scale scores before and after intervention and control, and the number of relapses in the intervention and control group. If more than one mood-scale was used, we primarily used the Profile of Mood States (POMS).26 Otherwise, we used the Multiple Affect Adjective Checklist (MAACL),27 Visual Analogue (Mood) Scales (VA[M]S),28 Hamilton-depression rating-scale (HAMD),29 and Montgomery-Åsberg Depression Rating Scale (MADRS).30 However, in studies in patients with previous MDD in remission we primarily chose the HAMD or MADRS followed by the POMS, because most of these studies used depression rating scales to measure effects of depletion. If subscales were used, we took the subscale representing depressed mood.
We validated the abstracted studies (criteria available on request). We based criteria on previous reviews and the Cochrane handbook.11, 12, 18, 31 We assessed studies as poor, moderate or good in the context of this meta-analysis. We particularly assessed quality as good when studies applied a randomized double-blind design, when the achieved depletion was judged to be sufficient, and when mood-scale ratings were provided.11, 18 If more than one of these items was rated inadequate, we assessed the study as moderate. If studies had one of these items missing and at least one other aspect of the study was not well reported, we also assessed the study as moderate. If crucial data (e.g. scores and SDs) were missing the study was assessed as poor.
We qualitatively summarized all included studies in Tables 2, 3, 4 and 5 (ATD, PCPA, APTD and AMPT respectively), irrespective whether the studies were pooled in the meta-analysis. We acknowledged three clinically heterogeneous study populations a priori: (A) healthy controls (A1 negative/A2 positive family history of MDD); (B) patients with a previous MDD currently in remission (B1 currently not using AD/B2 currently using AD); and (C) patients with a current episode of MDD (with or without AD). These populations were considered and analyzed separately.
Mostly, continuous scores for changes in mood rating scales were reported. Some studies in patients with MDD in remission presented relapse rates. In order to pool continuous effect estimates from different scales, we applied a standardization (providing Hedges’ g). The effect estimates (one per study) with the corresponding standard error (SE) were entered in the inverse-variance statistics for pooling in Review Manager 4.2.125, 126 Appendix gives the formulas used to determine Hedge's g, the difference in relapse rates and corresponding SEs for different study designs.
Different study designs were not combined in pooling. Especially the within-subject design needs attention in meta-analysis. In this design, differences between the experimental and control condition are statistically paired. As paired data improve power, calculations with data from a within subjects (or cross-over) design in a weighted mean differences model would not be justified.126
Heterogeneity, effect modification, sensitivity-analysis and assessment of publication bias
We first performed the meta-analyses with fixed effects models. We assumed more homogeneous results after our a priori attempt to reduce clinical heterogeneity by stratification. However, if effect-estimates and 95% CI for the individual studies showed graphical poor overlap or consistency, we interpreted this an indication of statistical heterogeneity. We used the χ2 test and I2 in addition. I2 represents a χ2 statistic relative to its degree of freedom. An I2 value >50% is indicative of heterogeneity.127 We applied a conservative random effects model128 when we suspected statistical heterogeneity.
We investigated effect modification by gender, and the influence of a positive family history of MDD in healthy controls. Furthermore, we investigated whether different mood scales caused differences in outcomes. Therefore, we stratified analyses for these variables, and presented stratified Hedges’ g, with 95% CI. Differences between strata were tested by subtracting the χ2 heterogeneity statistic per stratum from the total χ2 heterogeneity statistic. This residual (Qres) has a χ2 distribution, with the total number of strata-1 degrees of freedom.
We imputed R=0.5 to calculate missing pooled SDs within each intervention/control and for the changes in mood-scores between interventions (Appendix). We checked the impact of this imputation on the calculated effect estimates. Therefore, we increased R to 0.8 (less conservative) and decreased R to 0.2 (more conservative) to compare the new effect estimates with the original findings for R=0.5.
We assessed publication bias graphically with a funnel-plot, plotting Hedges’ g vs the precision of the study (the inverse of the standard error of Hedges’ g). Additionally, we tested publication bias with Galbraith's radial plot, which regresses the standard normal deviate (Hedges’ g divided by its standard error) against the precision. For a set of studies not distorted by selection bias the intercept of the regression model will be close to zero.129
Our systematic searches identified 392 articles. In total, we selected 90 studies. Three studies applied both ATD and APTD and a control.48, 49, 58, 59 Therefore, 73 studies with ATD, two with PCPA, 10 with APTD and eight with AMPT as monoamine depletion method were identified (summarized in Tables 2, 3, 4 and 5 respectively). Several studies investigated contrasts in different populations.36, 50, 55, 56, 65, 66, 67, 81, 90, 92, 103 A list of excluded studies is available on request. Most studies had a within-subjects design (n=74). The majority of studies investigated healthy controls (n=64).
The majority of the 90 studies also investigated other effects of monoamine depletion: 24 studies measured effects on cognitive functions,13, 33, 34, 44, 45, 46, 47, 49, 50, 52, 56, 60, 64, 69, 75, 77, 78, 82, 95, 112, 113, 114, 115, 116, 117 11 measured effects on other behavioral measures (pain, impulsivity, panic attacks, appetite, noise-stress, aggression),32, 38, 57, 61, 68, 73, 76, 86, 96, 111, 120 eight measured effects on neuroendocrine parameters,39, 48, 71, 108, 109, 112, 118, 121 11 measured electric encephalogram (EEG) alterations and/or sleep effects,37, 40, 51, 53, 60, 72, 84, 91, 99, 100, 101 14 measured changes in brain function with positron emission tomography (PET),66, 67, 70, 87, 97, 103, 104, 123 single photon emission computed tomography (SPECT),83 or magnetic resonance imaging (MRI).33, 42, 43, 44, 45, 46, 74 Two studies stratified mood response to ATD by genetic polymorphism of the 5-HT transporter promoter region.65, 67
We judged the overall methodological quality of the identified studies as good, with appropriate application of the ATD, APTD or AMPT depletion. Two studies with PCPA reported case series only and were rated ‘poor’.15, 110 In eight studies no data on the adequacy of the achieved depletion was reported.15, 32, 40, 110, 111, 118, 119, 123 In two studies insufficient depletion was attained.94, 115 In 10 studies no clear SDs or SEs were given for the observed mood effects.13, 61, 71, 72, 79, 85, 98, 118, 119, 123, 124 Six studies included patients (<25% of total) with Bipolar I or II disorder.15, 98, 99, 108, 117, 124 In total, this were 14 patients. The reason that several apparently high-quality studies were rated as moderate in this review was due to the lack of presentation of the mood effects, which in those studies was not the primary outcome.
In 34 of 90 studies the POMS was used, in two studies the MAACL, in 34 studies a visual analogue scale, in 29 studies a version of the HAMD, and in five of the 90 studies the MADRS. In five studies only other or undefined mood-scales were used.40, 51, 76, 85, 86 Especially for the POMS and the VAS the direction of a decrease of mood in subjects was not always clearly reported.52, 68, 77, 85 In our meta-analysis no differential effect of the applied mood-scale was observed (Qres=3.89, df=2, P>0.05; data not shown).
Quantitative summary (pooling)
Fifty-eight percent (52/90) of the studies supplied enough data to be included in the meta-analysis. Meta-analysis was possible for ATD and APTD studies only. Table 6 gives an overview of the number of identified studies for each pre-defined population and eligibility for meta-analysis. Due to remaining heterogeneity we used random effects models for all meta-analyses.
In Figure 1 the pooled results of ATD in healthy controls in studies with a within subjects design are presented. Overall Hedges’g (95% CI) was −0.27 (−0.45 to −0.09). We stratified results by family history for MDD (negative, positive or not reported in the studies). Pooled Hedges’ g for healthy controls with a negative family history (−0.19 (−0.43 to 0.05)) was significantly different (Qres=6.59, df=1, P=0.01) compared to controls with a positive family history (−0.56 (−1.00 to −0.13)). The pooled result in studies that did not report the family history status resembled the studies with a negative family history (−0.28 (−0.57 to 0.00) Qres=1.36, df=1, P=0.24). In between subjects studies similar results were found. Pooled Hedges’ g was −0.63 (−1.95 to 0.70) for controls with a negative family history and −0.06 (−0.57 to 0.45) in one study that did not report family history status (Qres=0.14, df=1, P=0.71). The large Hedges’ g in controls with a negative family history was largely determined by one study.71 Leaving this study out reduced Hedges’ g to 0.16 (−0.43 to 0.76) (data not shown).
Figure 2a and b show the modification of Hedges’ g by gender for healthy controls (within subjects design) with a negative or positive family history. The difference in Hedges’ g between males and females was most prominent in healthy controls with a negative family history (0.23 (−0.10 to 0.57) vs −0.44 (−0.81 to −0.06) respectively; Qres=11.92, df=1, P<0.001). In controls with a positive family history, males experienced a larger decrease in mood after ATD in only one study (Hedges’ g −0.98 (−1.53 to −0.42) Qres=11.92, df=1, P<0.001). In contrast, females with a positive family history only had a slightly larger Hedges’ g (−0.56 (−1.43 to 0.31)) than females with a negative family history (Qres=0.62, df=1, P=0.43).
Figure 3a and b present the pooled Hedges’ g for patients with MDD in remission without or with current AD (within subjects design). In the remitted patients without current AD, we stratified results by length of time without AD. Only two studies with 3–6 months without AD66, 67 largely differed in Hedges’ g (−4.35 (−7.39 to −1.31) compared to one study directly after successful electroconvulsive therapy (Hedges’ g 0.04 (−0.85 to 0.94),89 and three studies with at least 6 months without AD (pooled Hedges’ g −0.60 (−1.38 to 0.18) (Qres=29.34, df=2, P<0.0001).91, 92, 93 Leaving one possible outlier66 out diminished the overall pooled Hedges’ g for remitted patients without AD from −1.90 (−3.02 to −0.78) to −1.06 (−1.83 to −0.29), but the observed effect modification by length of time without AD remained highly significant (P<0.001; data not shown).
In remitted patients previously depressed and currently using AD, ATD caused a decrease in mood (pooled Hedges’ g −0.49 (−0.89 to −0.10). Hedges’ g varied slightly depending on type of AD (Qres=0.92, df=3, P=0.82). Surprisingly, the pooled Hedges’ g for SSRIs showed a moderate point estimate, which did not reach significance (−0.60 (−1.28 to 0.08). Especially for bupropion treatment Hedges’ g was small and not significant (−0.25 (−0.96 to 0.47). No ATD studies with other ADs without a 5-HT mechanism of action were available for this comparison.
We stratified the results of ATD studies by length of remission, which revealed significant effect modification. In remitted patients using AD decreased mood after ATD was especially seen in the first 5 months after the achievement of remission (pooled Hedges’ g −0.55 (−0.90 to −0.21) Qres=47.18, df=2, P<0.0001).92, 95, 99, 100, 101, 105 In contrast, remitted patients without AD showed more decrease in mood after 2 months of remission (Hedges’ g −1.65 (−2.60 to −0.69) Qres=13.81, df=2, P=0.001)67, 92, 93 (figure available on request).
Relapse rates in remitted patients with AD were increased after ATD compared to control depletion (pooled difference in relapse rate 47% (28–66%); Figure 4). This increase in relapse rate was especially seen in patients using SSRIs (47% (27–67%)) or an SNRI (35% (14–56%)). The NE acting drug desipramine showed no significant difference in relapse rate (7% (−6 to 19%)) after ATD. This effect modification by drug was statistically significant (Qres=18.02, df=2, P<0.001).
In patients who were depressed at the time of ATD we found two studies for meta-analysis. These studies included patients who used,106 or did not use109 AD (Figure 5). The effects of ATD were opposed: Hedges’ g was 0.32 (−0.22 to 0.86) for patients using different types of ADs, and −0.12 (−0.45 to 0.21) for patients without AD. Two studies in depressed patients without AD were not suitable for meta-analysis. ATD did not decrease mood during depletion in these studies.107, 108 Contra-intuitively, in one study mood increased the day after depletion in 16/43 patients,107 a result which was also found by one study in the meta-analysis.106
Figures 6a and b show results from APTD studies in healthy controls (within and between subjects). APTD did not decrease mood: pooled Hedges’ g were 0.10 (−0.23 to 0.43) in within subjects, and 0.12 (−0.43 to 0.68) in between subjects studies. However, in one study with healthy controls with a positive family history for MDD, a moderate but nonsignificant effect on mood was found (Hedges’ g −0.49 (−1.17 to 0.19)).115 In patients with MDD in remission without AD (Figure 7), no effect of APTD was observed in two studies (pooled Hedges’ g −0.02 (−0.50 to 0.47).116, 117 No studies in patients with current MDD were identified.
Sensitivity analysis and publication bias
We examined our assumptions for intercorrelation of before–after mood ratings per condition in ATD studies in healthy controls without a family history of MDD. At the same examination, we also examined the assumed intercorrelation between two test conditions (ATD vs CONT). As expected, increasing the assumed R to 0.8 (less conservative) increased the pooled Hedges’ g from −0.19 (−0.43 to 0.05) to −0.31 (−0.64 to 0.02). Reducing R to 0.2 (more conservative) decreased the pooled Hedges’ g to −0.14 (−0.34 to 0.06). The calculated R's were higher than 0.5 in 80% of the studies that reported all relevant data. Therefore, we judged the imputed value of 0.5 for R as acceptable.
Inspection of the funnel-plot of the within subjects ATD studies in healthy controls without a family history of MDD revealed asymmetry (figure available on request). More studies with a decrease in mood after ATD vs control depletion were published. In a Galbraith plot of studies, the intercept in the regression equation was −2.08 (SE 0.90; P=0.030). Therefore, we concluded that publication bias probably distorted our findings. We did not inspect funnel-plots in other populations of our review due to the limited number of studies.
In this systematic review we performed the first meta-analysis of the mood effects in ATD and APTD studies. The depletion of monoamine systems (both 5-HT and NE/DA) does not decrease mood in healthy controls. However, in healthy controls with a family history of MDD the results suggest that mood is slightly decreased, both by ATD and APTD. Additionally, healthy female subjects are more affected by ATD than healthy male subjects, especially in controls without a family history of MDD. In patients who were previously depressed but in remission without AD, ATD moderately decreases mood, whereas APTD does not significantly decrease mood. ATD has comparable mood lowering effects in patients with MDD in remission who are still using ADs. The site of action of these ADs (5-HT or NE/DA) predicts the occurrence of a lowering of mood or a short relapse in MDD after depletion of the corresponding monoamine. Our findings are in line with the summaries in previous reviews.11, 18, 19, 20, 24
The most consistent finding from this review is the decrease of mood and relapse into a depressed state after ATD and APTD in remitted MDD patients who use AD. In remitted patients without AD relapses are less prominent. Because after remission medication is often continued, the difference in mood responses after depletion might be related to the duration of the achieved remission. Previous reviews discussed the relationship between the duration of the remission and the effect of monoamine depletion, with opposite conclusions. Bell et al.130 concluded that effects of ATD were more pronounced early in recovery. Contrarily, Booij et al.21 concluded that duration of remission was not associated with mood response to ATD.
Booij et al.21 investigated predictors of relapse in a pooled analysis of the individual patient data of some of the studies included in this review. This is often referred to as a ‘mega-analysis’. They found that recurrent depressive episodes (2), female gender, and a history of suicidal thoughts/attempts predicted relapse. Duration of remission did not contribute to this prediction when confounding was considered. We found a modest relationship between relapse and the duration of the remission after ATD. We defined the duration of remission as the reported average duration, or – if not stated – the minimum duration of remission used as inclusion criterion for the studies. We found that especially in the first 5 months after remission, ATD caused lowering of mood in remitted patients still using AD. However, the problem with our and Bell's comparison is the intraindividual spread in duration within the studies. This spread is not considered at the same level of detail as in a ‘mega-analysis’. In addition, confounding can only be considered in ‘mega-analysis’. Therefore – although their study did not include all available studies – we think that Booij et al. provide the best available indication of risk factors for mood lowering effects by ATD.
Do depletion studies elucidate the pathogenesis of MDD?
The absence of robust mood effects in healthy controls indicates that mood is not a direct correlate of 5-HT or NE levels in the brain. The only healthy controls who are modestly affected by monoamine depletion studies are healthy controls with a positive family history for MDD. This might be indicative of a biological vulnerability, which is revealed by depletion studies. Of interest are the findings of recent studies that combined ATD with neuroimaging or genetic sampling,33, 42, 43, 44, 45, 46, 65, 66, 67, 70, 74, 83, 87, 97, 103, 104, 123 reviewed by Fusar-Poli et al.25 The intelligent approach of combining depletion with imaging or genotyping appears very promising. A summary of the complex results of these studies goes beyond the scope of our review. However, these neuroimaging and genotyping studies also suggest that monoamine depletion discloses rather a ‘trait’ vulnerability than a pure ‘state’ dependent change due to depletion.
Additionally, a depressive relapse after monoamine depletion in remitted patients who use AD, occurs only if the target of the depletion (5-HT, NE) coincides with the working mechanism of the AD used. This emphasizes that AD indeed specifically affect their supposed target systems. However, we may only conclude that an undepleted 5-HT system is required for serotonergic AD. The same holds for the NE system and norepinephrinergic AD. Delgado proposed an alternative explanation for the decrease in mood after monoamine depletion in patients: depletion of, for example, 5-HT may give the same effect as abrupt discontinuation of SSRIs.131 Rapid discontinuation is also associated with mood effects, which are considered to be different from a depressive relapse.132
What certainly cannot be concluded is that MDD is caused by low levels of 5-HT and/or NE/DA. This simplification, which is often used to promote the use of AD specifically affecting 5-HT or NE or both systems, represents a Catch-22 argument, and ignores the notion that serotonergic and norepinephrinergic AD presumably act by a final common pathway. In this pathway postsynaptic changes at the cellular level are supposed to be responsible for the remission of MDD.133 Cellular changes include up or downregulated receptors, increased neuronal interconnections and sprouting and changes in levels of neuropeptides (e.g. corticotrophin releasing hormone). The clinical question of how to distinguish a patient who will respond to an AD with, for example, NE effects has not been solved. Descriptive variables at the symptom level have not yet sufficiently predicted the response to any selective agent. Therefore, a pragmatic approach to affect this final common pathway might be to prescribe ADs which target both 5-HT and NE. However, the effectiveness of this approach is still equivocal.134, 135, 136
In patients who are depressed at the time of monoamine depletion, no further decrease in mood is observed. A ceiling effect could be responsible for this finding. But, a more straightforward conclusion is that there is no simple relation between 5-HT or NE deficiency and mood or MDD. Nevertheless, this finding is complicated by the finding of some authors106, 107 that mood is lowered or elevated the day after ATD. A delayed decrease of mood was indicative for treatment refractoriness,107 a finding that was not yet replicated by others.106, 108, 109 The number of comparable studies to date is limited. Therefore, we think no clear conclusions or explanations can be made, except that even in MDD patients mood is not a correlate of 5-HT or NE levels in the brain.
We agree with the conclusion of Booij et al.11, 21 and Bell et al.130 that a relapse of depressive symptoms in remitted patients after depletion probably reflects a biological vulnerability of the 5-HT system in remitted patients. This vulnerability increases their risk to become depressed. This increased risk was further demonstrated in two prospective studies, which used the response to ATD in remitted patients to predict later relapse/recurrence.63, 137 However, also these results need replication. Moreover, the cause of an increased vulnerability remains uncertain. Probably the cause is a combination of genetic, environmental and other determinants (e.g. ‘scarring’ the brain after multiple depressive episodes).130
In conclusion, monoamine depletion by ATD and APTD does not elucidate a causal factor in the pathogenesis of MDD. However, ATD and APTD remain useful models to safely and directly manipulate 5-HT, NE and DA function in living humans, and to study the behavioral consequences of this manipulation, especially in subgroups of humans with an apparent vulnerability.11, 12, 24
Still, several methodological issues need to be addressed. First, because of the competition of amino acids to pass the blood–brain barrier, ATD might unwillingly result in an intracerebral increase of tyrosine/phenylalanine.24, 138 Vice-versa, APTD might increase intracerebral tryptophan availability. Levels of other amino-acids than those depleted are not provided in the studies. Second, ATD provides a specific net lowering of 5-HT. Contrarily, depletion of tyrosine and phenylalanine lowers both NE and DA, which are synthesized in the same cascade (with DA thereafter transformed into NE by DA beta-hydroxylase). Also AMPT interferes early in this cascade. Although evidence from animal studies points to more DA depletion by APTD and more NE depletion by AMPT,139 it is impossible to truly distinguish between net NE and DA depletion. Third, test re-test reliability for monoamine depletion paradigms was only tested for ATD and was rather limited, which limits the robustness of the method.41, 88 Fourth, also in healthy controls subtle cognitive effects of monoamine depletion in the brain occur: deficits in learning and memory consolidation and improvement in focused attention and executive function.140 These effects show similarity with symptoms of MDD. The question remains whether these effects might represent mild first symptoms of MDD or the starting symptoms of a cascade of altered brain functions leading to MDD. Fifth, in line with the fourth issue, MDD does not develop within one day. Therefore, the changes by experimental monoamine depletion by ATD and APTD may be too short to really induce a complex biological and psychological deregulation which is recognized as MDD. Patients suffering from gastrointestinal carcinoid tumors – 5-HT producing tumors with expected prolonged states of secondary tryptophan depletion – are generally not depressed but do show improved focused attention.141 However, carcinoid findings were not yet related to tryptophan levels over time. Sixth, the single depletion of one monoamine system by ATD or APTD/AMPT may be too simplistic, especially because of the complex interaction of monoamine systems. Five dual depletion studies were not included in this review.49, 142, 143, 144, 145 In healthy controls contrasting results were found. Hughes et al.143 found some decrease of mood on 3 VAMS-subscales, which was also found in another open study.144 However, no effects were found in two other studies.49, 145 In unmedicated patients with MDD no significant increase of MDD was found after dual depletion.142 It would be interesting to investigate the effects of simultaneous depletion of 5-HT and NE/DA in other populations.
Limitations of the studies and the review
Several limitations should be mentioned. First, female gender is a risk-factor for MDD, and was also found to be a predictor of relapse after ATD in remitted patients. Based on several studies, gender differences in 5-HT metabolism are probable, and hormonal factors may play a role in 5-HT function.21 In our results we examined the effect of gender in healthy controls. The difference between male and female subjects was most prominent in healthy controls without a family history of MDD. In the included studies hormonal status (pre- or postmenopausal state) was not distinguished in the results nor used as inclusion criterion. Second, many small differences between the studies existed: different composition of depletion and control drinks, different presentation of tryptophan/tyrosine/phenylalanine or their ratios to other neutral amino acids, different presentation of free vs total tryptophan/tyrosine/phenylalanine values, different measurements, different scales. Differences between studies undoubtedly introduced heterogeneity between the studies, which may bias the results of this review. Therefore, we recommend an international consensus protocol. Third, limited presentation of the data forced us to make assumptions, that may have influenced our findings. However, the assumptions appeared to have little influence on the results in a sensitivity analysis. Future studies need to address clear presentation of their data, preferably including a description of the directions of the effects, SDs, in cross-over designs the numbers of subjects having the control or sham first, and preferably include an SD of the pooled difference between the depletion and the control condition. For example, only 10 studies adequately reported the number of patients treated in each sequence in the within subjects design.33, 44, 46, 64, 77, 88, 100, 116, 117, 122 Fourth, six studies included Bipolar Patients. Three of these studies were also included in the meta-analysis.98, 99, 117 This involved 6 patients of the total of 265 patients (2.2%) in the concerning meta analytical comparisons. Therefore, we consider the possibility of bias by the inclusion of this etiologically different disorder unlikely. Fifth, the rate of agreement in the selection of studies still may rise questions about the clarity of our selection criteria. However, discrepancies in agreement could easily be solved between the two reviewers. Therefore, we do not think our sensitive searches missed relevant studies. Finally, an indication of publication bias was found. If publication bias truly exists, the studies which found no effect or an increase in mood after ATD would not have been published. Indeed, the mood effects of studies that could not be included in the meta-analysis were mostly very small and nonsignificant. Furthermore, the exact information in studies required for meta-analysis forced us to exclude many studies. It seems natural that a non-significant result will not be given this level of detailed attention (change-scores with SDs), especially when mood effects are not the primary outcome in these studies. Without this publication bias the pooled effect of ATD in healthy controls would probably have been lower. Because our conclusion already is that there is no apparent mood effect of ATD in healthy controls, we conclude that the observed publication bias will not severely distort the general conclusion. Therefore, despite these limitations, we consider the results of our review after our strict methods as valid.
Conclusion and future studies
We conclude that although ATD and APTD are important in the investigation of the monoamine systems, monoamine depletion does not directly decrease mood. Although previously the monoamine systems were considered to be responsible for the development of MDD, the available evidence to date does not support a direct causal relationship with MDD. There is no simple direct correlation of 5-HT or NE levels in the brain and mood. The depletion of 5-HT by ATD and NE/DA by APTD most clearly decreases mood in vulnerable patients who are in remission from their MDD, while still using AD. Furthermore, depletion affects mood in unmedicated patients in remission or healthy controls with a family history positive for MDD. Therefore, the monoamine systems are probably important systems in the vulnerability to become depressed. The changes in brain metabolism in remitted patients who relapse after ATD or AMPT suggest that 5-HT and NE systems give input to a final common pathway, which needs further research to be clarified.
We suggest some lines for future research. First, three or four-armed depletion studies comparing ATD, APTD (and – if possible – their combination) vs sham depletion to investigate the differential effects of the 5-HT and NE systems on mood in remitted patients or controls with positive family history for MDD. Second, a further exploration of the relation between known genetic polymorphisms of the 5-HT, NE and DA systems (e.g. 5-HTTPR) and biological cerebral responses to depletion paradigms, as measured by PET/fMRI (e.g. like Neumeister et al.65, 67). Third, the relations between monoamine depletion indicating biological vulnerability and psychological vulnerability for MDD (see Booij et al.11) Fourth, replication, further validation and standardization of the properties of ATD and other depletion paradigms as a diagnostic test for recurrences in remitted patients (see Moreno et al.63 and Neumeister et al.137). New studies will increase our knowledge of the 5-HT and NE systems, which are important targets in the current treatments of MDD. This knowledge will finally improve the treatment for MDD.
We thank Mrs Natasha Wiebe, MMath PStat, Research Associate at the University of Alberta, Canada and especially Dr Rob JPM Scholten, MD, PhD, epidemiologist and director of the Dutch Cochrane Centre at the Academic Medical Centre, Amsterdam for their indispensable help with the statistic methods used in this review. This study was partially funded by the program Opleiding Onderzoekers GGZ (OOG) (project number 100-002-002) by ZonMw, the Hague, the Netherlands.
Differences between intervention and control measurements
In depletion studies changes in mood scores typically represented mean mood-scores both before (pre) and after (post) the depletion/challenge (experimental intervention) and the placebo/sham/control-intervention. Because the mood scores were not necessarily identical at the start of the experiment and the control, we first calculated the mean change in mood-score (pooled difference) for the experimental and control condition separately per study. Some studies also provided the standard deviation (SD) of the pooled differences. When the standard deviation was not reported, we calculated the standard deviation of the pooled difference for paired data:
In this formula, the correlation coefficient R was calculable in four studies only36, 59, 81, 105 and ranged between 0.42 and 1.00 for the experimental and between 0.34 and 0.95 for the control condition. To be able to calculate the standard deviation of the change between pre- and post-test mood scores for the rest of the studies we imputed a correlation coefficient R of 0.5. This value was considered to be a conservative assumption.
Statistics for studies with a within subjects design
The difference in the changes of mood scores between intervention and control were expressed as difference of change scores:
For this difference the SD of the difference was calculated by again applying formula (A1), with an assumed R of 0.5.
To acknowledge the different mood scales to measure change in mood, the difference in changes between experimental and control condition were standardized by calculating Hedges’ adjusted g, which is similar to Cohen's d, but includes an adjustment for small sample bias:126
In this formula nAB and nBA represent the number of subjects randomized to intervention or control as first test in the study. If the numbers for nAB and nBA were not reported, we assumed that the sample was split half for the two sequences. For Hedges’ g an SE was calculated as follows:146, 147
Statistics for studies with a between subjects design
For between subject studies comparable statistics were used to calculate the mean change in mood-scores. Because the between-subjects design is a parallel group design, Hedges’ g was calculated with formula (A3) in which for nAB and nBA nINT and nCONT were substituted. The formula for the SE was slightly different to acknowledge the absence of paired data:
Statistics for relapse rates
For relapse rates of MDD after depletion provided in a within subjects design the difference in relapse rates was calculated as in which N is the total number of patients included. The standard error then is in which b represents the number of patients with a relapse after the intervention but not the control condition and c the number of patients with a relapse after the control but not the intervention.148 If the numbers of ‘pairs’ were not extractable from the paper, a conservative approach was used assuming the minimal number of patients relapsing both after the intervention and the control condition (maximal c), resulting in the largest SE.
About this article
Conflict of interests