Prediction of Lithium Treatment Response in Bipolar Depression using 5-HTT and 5-HT_1A PET

Abstract

Background:

Lithium, one of the few effective treatments for bipolar depression (BPD), has been hypothesized to work by enhancing serotonergic transmission. Despite preclinical evidence, it is unknown whether lithium acts via the serotonergic system. Here we examined the potential of serotonin transporter (5-HTT) or serotonin 1A receptor (5-HT_1A) pre-treatment binding to predict lithium treatment response and remission. We hypothesized that lower pretreatment 5-HTT and higher pretreatment 5-HT_1A binding would predict better clinical response. Additional analyses investigated group differences between BPD and healthy controls and the relationship between change in binding pre- to post-treatment and clinical response. 27 medication-free patients with BPD currently in a depressive episode received positron emission tomography (PET) scans using 5-HTT tracer [¹¹C]DASB, a subset also received a PET scan using 5-HT_1A tracer [¹¹C]-CUMI-101 before and after eight weeks of lithium monotherapy. Metabolite-corrected arterial input functions were used to estimate binding potential, proportional to receptor availability. 14 patients with BPD with both [¹¹C]DASB and [¹¹C]-CUMI-101 pre-treatment scans and eight weeks of post treatment clinical scores were included in the prediction analysis examining the potential of either pre-treatment 5-HTT or 5-HT1A or the combination of both to predict post-treatment clinical scores.

Results:

We found lower pre-treatment 5-HTT binding (p=0.003) and lower 5-HT_1A binding (p=0.035) were both significantly associated with improved clinical response. Pre-treatment 5-HTT predicted remission with 71% accuracy (77% specificity, 60% sensitivity), while 5-HT_1A binding was able to predict remission with 85% accuracy (87% sensitivity, 80% specificity). The combined prediction analysis using both 5-HTT and 5-HT_1A was able to predict remission with 84.6% accuracy (87.5% specificity, 60% sensitivity). Additional analyses BPD and controls pre- or post-treatment, and the change in binding were not significant and unrelated to treatment response (p>0.05).

Conclusions:

Our findings suggest that while lithium may not act directly via 5-HTT or 5-HT_1A to ameliorate depressive symptoms, pre-treatment binding may be a potential biomarker for successful treatment of BPD with lithium.

Clinical Trial Registration:

PET and MRI Brain Imaging of Bipolar Disorder Identifier: NCT01880957; URL: https://clinicaltrials.gov/ct2/show/NCT01880957

Background:

Bipolar disorder, the 6^th leading cause of disease burden worldwide, is a debilitating psychiatric condition characterized by periods of mania and depression. Bipolar depression (BPD), which clinically resembles major depression, poses a high risk for suicide. Two major obstacles in the treatment of BPD are our lack of understanding of its underlying neurocircuity, and the mechanism of action for existing treatments.

Lithium remains one of the first lines of therapy for treating both mania and depression in bipolar disorder(1). Lithium is used for long-term, prophylactic treatment of bipolar disorder to prevent further occurrence of both mania and depression(2). A meta-analysis of five controlled trials found that lithium was 20% more effective in preventing recurrence of bipolar disorder than placebo(3). Interestingly, it has also proven extremely effective in reducing suicide attempts and mortality(4, 5). While many find near total resolution of symptoms following lithium treatment, more than 25% of patients show no clinical response(3). What is needed is a better understanding of lithium’s mood stabilizing mechanisms to identify a potential biomarker for successful treatment response.

Little is known about the cellular mechanisms underlying lithium’s effects on mood. A leading hypothesis is that lithium works by enhancing serotonergic transmission(6, 7). Previous studies have provided evidence for lithium’s effects on serotonergic synthesis(8), uptake(6, 7), receptor interactions(9), and perhaps most importantly serotonin (5-HT) release(10). Considering the serotonergic system is the site of action of some of the most successful antidepressants (serotonin reuptake inhibitors, SSRIs), it is plausible that lithium too works in concert with or directly upon the serotonergic system to exert its anti-depressant effects. Of interest is lithium’s effect on 5-HT uptake at the serotonin transporter (5-HTT). 5-HTT is the protein responsible for the termination of 5-HT in the synapse, and the target of SSRIs. Preclinical studies reveal increases in 5-HTT density (B_max) in cortical regions following chronic lithium treatment(7), but not in subcortical regions known to be altered in BPD and major depression, including midbrain and amygdala(11, 12). Additionally, it has been suggested that a functional polymorphism in the 5-HTT gene increases the susceptibility to bipolar disorder(13), and can modulate lithium’s effectiveness as a mood stabilizer(14).

Of equal interest is the serotonin 1A receptor (5-HT_1A) of which there are two types: 1) autoreceptors which reside in the raphe nucleus upon serotonergic cell bodies and dendrites, and 2) post-synaptic receptors which exist within terminal fields throughout the brain(15). SSRIs are thought to contribute to the desensitization of raphe 5-HT_1A autoreceptors, thereby increasing 5-HT in the synapse, and eventually leading to changes in post-synaptic 5-HT_1A, and an efficacious response. It is plausible, then, that lithium also affects 5-HT_1A either pre- or post-synaptically to mediate treatment response. Indeed, preclinical studies reveal decreases in post-synaptic 5-HT_1A the frontal cortex and hippocampus following lithium monotherapy(16, 17), while raphe 5-HT_1A levels remain unchanged(17).

Neuroimaging is a promising tool for in vivo investigation of serotonergic proteins and their relationship with treatment response in BPD. Using Positron Emission Tomography (PET), our group has found higher 5-HT_1A binding was associated with improved clinical response in BPD following 3 months of treatment with SSRIs (18). Further, we have found that that decreased 5-HTT binding in the amygdala was predictive of future remission in MDD (19), and that suicide attempters had lower midbrain 5-HTT binding as compared to nonattempters and controls(11). Here we examine the potential of pre-treatment 5-HTT (assessed with carbon 11–labeled 3-amino-4-(2-dimethylaminomethyl-phenylsulfanyl)- benzonitrile ([¹¹C]DASB)) or 5-HT_1A (assessed with partial agonist tracer carbon 11-labeled [O-methyl-(11)C]2-(4-(4-(2-methoxyphenyl)piperazin-1-yl)butyl)-4-methyl-1,2,4-triazine-3,5(2H,4H)dione ([¹¹C]-CUMI101)) binding to predict treatment response following eight weeks of standardized lithium monotherapy. We use cross-validated LASSO regression for robust estimation of brain regions with 5-HTT and 5-HT_1A binding that predict treatment response and classify the accuracy, sensitivity, and specificity of these predictions.

As per the proposed aims of the NIH Grant: R01MH090276 our goals were to investigate the potential of 5-HTT or 5-HT_1A binding to predict treatment response following 8 weeks of lithium treatment. Additional grant aims sought to determine pre-treatment differences in 5-HTT or 5-HT_1A binding between groups and assess whether change in binding pre-to-post treatment related to clinical response. We predicted pretreatment 5-HTT and 5-HT_1A binding would be predictive of clinical response following lithium treatment. We theorized that lithium would alleviate depressive symptoms by exerting its effects in the synapse, by increasing 5-HT release, leading to homeostatic upregulation of cortical (anterior cingulate cortex) but not subcortical (midbrain and amygdala) 5-HTT, and downregulation of post-synaptic 5-HT_1A but not in the somatodendritic raphe region. We hypothesized that these lithium-induced alterations would associate with better clinical response.

Methods:

Subjects/Treatment

This study was approved by the Institutional Review Boards of Columbia University Medical Center (CUMC), Brookhaven National Laboratory (BNL), Yale University Medical Center (Yale), and Stony Brook University (SBU). All participants provided written, informed consent. Recruitment occurred from March 2008 through June 2017.

BPD group: 27 adults met DSM-IV criteria for bipolar depression, (assessed with the Structured Clinical Interview for DSM-IV (SCID-I)(20) and psychiatrist interview). All subjects were medication-free for a minimum of 3 weeks prior to treatment initiation; approximately 30% of the sample was medication naïve upon enrollment. Exclusion criteria for patients with BPD were: 1. other major psychiatric disorders (not including eating disorders or phobias), 2. substance abuse for the past two months, 3. recent dependence within the past six months (except cannabis use disorders), 4. IV drug use in the past five years, 5. MDMA use more than 15 times in the past 10 years or any use in the past month(21), 6. satisfactory response to psychiatric medications or lack of response to a trial of adequate dose/duration of lithium, 7. active suicidality, 8. electroconvulsive therapy within the past six months, 9. pregnancy, currently lactating, planning to conceive during the course of the study, or abortion in the past two months, 10. significant physical illness, 11. neurological disease/prolonged loss of consciousness, 12. any magnetic resonance imaging (MRI) contraindications, 13. medicinal patch that cannot be removed, and 14. lacking capacity to consent. Inclusion criteria for patients with BPD were: 1. a score of ≥15 on the 17-item Hamilton Depression Rating Scale (HDRS-17; inclusion only)(22), 2. the ability to stop all drugs that would likely interact with the serotonergic system for three weeks prior to scanning, and 3. being between 18 and 70 years of age.

In addition to patients with BPD, 28 healthy adult volunteers (HV) were enrolled. Primary inclusion criteria for HV were no lifetime history of Axis I disorders (other than specific phobias or adjustment disorder) and being between 18 and 70 years of age. Additional exclusion criteria not overlapping with patients with BPD include: having a first-degree relative with a history of major depression (if participant was < 44 years old), history of suicide attempt, and two or more first degree relatives with a history of substance dependence (if participant was < 27 years old).

Clinical Procedures

Following baseline PET and MRI scans and assessment with the HDRS-24-item (HDRS-24), treatment was initiated with lithium monotherapy as previously described(23). In brief: days 1–3: 300mg 2x/day, days 4–7: 300mg every morning and 600mg every night; titration to a therapeutic plasma level of 0.8–1.2 mEq/l. Following eight weeks of treatment, patients were rescanned with PET and MRI and reassessed with the HDRS-24.

PET Scanning Protocol

Preparation and scanning of [¹¹C]DASB (24) [¹¹C]-CUMI-101(25) was performed as previously described. In brief, PET images were acquired on an ECAT EXACT HR+ scanner (Siemens/CTI, Knoxville, Tennessee) at either CUMC, BNL or Yale. Following a 10-minute transmission scan, [¹¹C]-CUMI-101 was injected as an intravenous bolus and emission data was collected for 120 minutes after which time subjects exited the scanner for a short break (approximately two hours, six additional [¹¹C] half-lives, approximately twelve half-lives in total). Following a short break, a 10 minute transmission scan was collected and [¹¹C]-DASB was injected as an intravenous bolus; emission data was collected for 100 minutes.

Image Processing

[¹¹C]DASB(11, 26) and [¹¹C]-CUMI-101(25) image analysis was performed as previously described. In brief, frame-wise rigid body registration to a reference frame was performed to correct the PET data for motion using the FMRIB linear image registration tool (FLIRT). We check this motion correction algorithm using movies that compare the original PET frames to the motion corrected PET frames. The mean motion-corrected frame was then co-registered to the MRI. Finally, probabilistic regions of interest (ROIs) were generated for each subject and applied to the PET data.

Quantitative Analysis

Automated arterial blood sampling was performed for the first 7 minutes of the PET acquisition, followed by manual arterial sampling thereafter. A metabolite-corrected input function was generated from the full sampling arterial data (linear fit to the peak and a sum of three exponentials thereafter) or a validated simultaneous estimation algorithm was used to compute the metabolite-corrected arterial input function using a single venous or arterial blood sample(23, 27). Time activity curves were fit with likelihood estimation in graphical analysis (LEGA) as previously described(28) to obtain estimates of optimal binding potential measures: V_T/f_P (V_T= total distribution volume, f_P = tracer plasma free fraction) for [¹¹C]-DASB(29) and BP_f (B_avail/K_d, where B_avail is receptor availability and 1/K_d is the affinity) for [¹¹C]-CUMI-101(25). Standard errors (SE), which assess the goodness-of-fit of the binding data, were computed using an in-house implementation of bootstrapping via resampled residuals from the fit PET brain data, as previously described(30), The SEs were used to weight the subsequent linear mixed effects models. That is, the higher the confidence in the fit, the lower the SE, and the more heavily the data point is weighted.

Statistical Analysis

T-tests were used to examine demographic characteristics. Binding outcomes were log-transformed to stabilize region-wise variance and to fulfill normality assumptions of subsequent models, as previously described(11). For 5-HTT binding data, a model was fit for the a priori regions of interest: midbrain, amygdala, and anterior cingulate cortex. For 5-HT_1A binding data, separate models were fit for: 1. the a priori somatodendritic region (raphe nucleus) and 2. the a priori post-synaptic regions (12 previously validated ROIs: anterior cingulate, amygdala, cingulate, dorsolateral prefrontal cortex, hippocampus, insula, medial prefrontal cortex, occipital cortex, orbitofrontal cortex, parietal cortex, parahippocampal gyrus, and temporal cortex)(31). All statistical analyses were conducted in R version 3.3.3 (http://www.R-project.org).

Prediction of Treatment Response

To investigate whether pre-treatment binding predicted treatment response, a model was fit with the subject-wise pre-treatment aggregate score of 5-HTT binding in the 3 a priori regions and pre-treatment HDRS-24 as fixed effects. Post-treatment HDRS-24 was the model outcome and subject was fit as a random effect. The model was repeated with each of the three a priori ROIs in separate models as well.

To investigate whether pre-treatment 5-HT_1A binding predicted treatment response, separate models were fit with (1) pre-treatment a priori raphe nucleus 5-HT_1A binding as a fixed effect, and (2) a subject-wise pre-treatment aggregate score of 5-HT_1A binding in the 12 a priori post-synaptic regions (together) as a fixed effect. In addition to the subject-level aggregate score for pre-treatment 5-HT_1A binding, each of the post-synaptic a priori ROIs were fit in separate models to see if specific regions were driving the results.

All models were also repeated using the dichotomous remission status variable (remitter vs. non-remitter, where remitter is defined a priori by HDRS-24 <10 post-treatment and a reduction of greater than or equal to 50% in HDRS-24 pre-to-post treatment(26)) to determine if pre-treatment binding predicted those who would remit following 8 weeks of lithium treatment.

Next, LASSO linear regression (Glmnet package (ORIGINAL CITE); R Core Team, 2016. Vienna, Austria; ORIGINAL LINK) was used to predict the continuous post-treatment HDRS-24 scores with the independent variables: HDRS-24 pre-treatment score and binding data in each of the a priori regions (amy, acn, mid for 5-HTT and the raphe, as well as the 12 post-synaptic regions, for 5-HT_1A). LASSO regression is a method that uses penalized maximum likelihood estimation to fit generalized linear models(32) and is commonly used in psychiatric research(33–35). LASSO regression is particularly beneficial when a dataset contains multiple, potentially inter-correlated predictors, which is common with brain imaging data, because the approach enforces model sparsity by selecting the most relevant predictors and shrinking the other coefficients to zero⁵. The tuning parameter lambda within the LASSO regression framework was selected via 10 repetitions of 5-fold cross-validation(36). The final models retained all variables with coefficients greater than zero. Prediction of the dichotomous remission status variable was not performed here due to sample size limitations in the 5-fold cross-validation step. Therefore, the HDRS-24 post-treatment scores for each participant were extracted from the selected LASSO regression prediction models and were used to classify each participant as a remitter or non-remitter (using the criteria described above). Post-hoc predictive accuracy, sensitivity, and specificity metrics were computed based on the predicted and true HDRS-24 post-treatment scores as previously defined(37).

The above models were then repeated to determine if combining the pre-treatment 5-HTT aggregate score with the 5-HT_1A post-synaptic aggregate score and 5-HT_1A raphe nucleus binding metrics into one model would explain more of the variance in treatment response than any measure alone.

Pre- or Post-Treatment Group Differences

To investigate differences in binding between HV and patients with BPD before treatment, or patients with BPD post-lithium treatment, linear mixed models were fit with [¹¹C]DASB or [¹¹C]-CUMI-101 binding as the model outcome and group (HV, BPD pre-treatment; or HV, BPD post-treatment), site, region, and sex (for [¹¹C]-CUMI-101 models only) as fixed effects, and subject as a random effect.

Lithium Treatment Effects

To investigate the effect of lithium on binding in patients with BPD, linear mixed models were fit with either [¹¹C]DASB or [¹¹C]-CUMI-101 binding as the model outcome, with pre- or post-treatment status, region, site, or sex (for [¹¹C]-CUMI-101 models only) as fixed effects. Random effects were subject and scan nested within subject.

Associations with Clinical Response

To investigate the relationship between clinical response to lithium and alterations in binding in patients with BPD, linear mixed models were fit with post-treatment HDRS-24 as the model outcome with binding and pre-treatment HDRS-24 as fixed effects. For [¹¹C]DASB data, a subject-level aggregate score for pre-to-post change in imaging was created across all three a priori regions and included as a fixed effect. For [¹¹C]-CUMI-101 analyses, raphe binding was fit in a separate model, and a subject-level aggregate score for pre-to-post change in imaging was created for the post-synaptic regions and included as a fixed effect.

Results:

Sample Characteristics

Summary statistics of clinical variables for the [¹¹C]DASB and [¹¹C]-CUMI-101 clinical sample is presented in Table 1. Of the 27 patients with BPD scanned with [¹¹C]DASB at pre-treatment, 19 went on to complete the lithium trial, and received post-treatment scans. Each subject’s pre- and post-scans were conducted at the same site. HDRS-24 scores significantly decreased following lithium treatment in the [¹¹C]DASB subsample from 27.1 (±4.9) to 15.4 (±9.0) (p<0.001), where 27.8% achieved remission. We found no significant differences in f_P (free tracer in plasma) between pre- and post-lithium treatment. Prediction analyses for the [¹¹C]DASB sample were conducted on the 14 patients with BPD that had eight weeks of post-treatment HDRS-24 data. Mean injected dose across participants scanned with [¹¹C]DASB was 522.2 MBq (Range 158.0 MBq – 754.4 MBq).

Table 1:

Sample Characteristics

	[11C]-DASB (N=19)			[11C]-CUMI-101 (N=13)
	BPD Pre-Lithium	BPD Post-Lithium	p-value	BPD Pre-Lithium	BPD Post-Lithium	p-value
Age (years)	34.0 ± 12.3	--	--	30.5 ± 10.5	--	--
Sex (% female)	47.4%	--	--	46.2%	--	--
HDRS-24	27.3 ± 4.9	15.5 ± 8.7	<0.001	26.8 ± 4.4	15.9 ± 9.2	0.001
HDRS-17 Baseline (Inclusion Criteria)	21.1 ± 4.1	--	--	21.9 ± 3.8	--	--
% Remitted*	--	26.3%	--	--	23.1%	--
Free Fraction (fp)	0.1 ± 0.03	0.1 ± 0.03	0.35	0.4 ± 0.1	0.4 ± 0.1	0.27
Reference Region Binding**	--	--	--	6.2 ± 1.4	6.5 ± 1.7	0.61

Abbreviations: HDRS – Hamilton Depression Rating Scale, BPD = bipolar depression

^*remission defined as <10 on HDRS-24 post-treatment and >= 50% reduction in HDRS-24 from pre to post-treatment

^**Region: cerebellar grey matter V_T (volume of distribution)

Of the 27 patients with BPD scanned with [¹¹C]DASB, 21 of these patients with BPD also received [¹¹C]-CUMI-101 pretreatment scans, and 13 of those went on to receive post-treatment [¹¹C]-CUMI-101 scans. Each subject’s pre- and post-scans were conducted at the same site. HDRS-24 scores also significantly decreased in the [¹¹C]-CUMI-101 subsample following lithium treatment in the whole group from 26.8 (±4.4) to 15.9 (±9.2) (p=0.001) on average, where 23.1% subjects achieved remission. We found no significant differences in f_P (free tracer in plasma) between pre- and post-lithium treatment in either sample, and no significant difference in reference region binding (grey matter of the cerebellum) for the [¹¹C]-CUMI-101 sample between pre- and post-lithium treatment groups (p>0.05) (this was not analyzed for the [¹¹C]DASB sample as no ideal reference exists for this region). Prediction analyses were conducted on the 14 patients with BPD that had eight weeks of post-treatment HDRS-24 data. Mean injected dose across participants scanned with [¹¹C]CUMI was 547.4 MBq (Range 127.8 MBq – 751.5 MBq).

Prediction of Treatment Response

The pre-treatment 5-HTT aggregate score (composite score of the three a priori regions) was found to be a significant predictor of treatment response (F=14.83, p=0.003) (Fig 2A). Anterior cingulate, amygdala, and midbrain pre-treatment 5-HTT binding were each found separately to be significant predictors of lithium treatment response (anterior cingulate: p=0.005; amygdala: p=0.015; midbrain: p=0.003; Fig 1A/B). With 10 repetitions of 5-fold cross-validation, the LASSO linear regression analysis using pre-treatment 5-HTT binding across the a priori regions and pre-treatment HDRS-24 indicated that all entered independent variables (amygdala, midbrain, anterior cingulate and pre-treatment HDRS-24 score) were statistical predictors of post-treatment HDRS-24, agreeing with the previous linear models. Extracting predicted HDRS-24 estimates from the model (using the aggregate score across all three a priori regions), remission status was predicted with 71.4% accuracy, 77.8% specificity, and 60% sensitivity.

Figure 2A:

Regression plots of model adjusted pre-treatment V_T/f_P (X-axis) by HDRS-24 post-treatment score (Y-axis) for the brain regions selected as significant predictors of treatment response with penalized logistic regression.

2B. Regression plots of model adjusted BP_F (X-axis) by HDRS-24 post-treatment score (Y-axis) for the brain regions selected as significant predictors of treatment response with penalized logistic regression (elasticnet).

Figure1A:

Regression plots of model adjusted V_T/f_P (X-axis) by HDRS-24 post-treatment score (Y-axis) in the brain regions selected as significant predictors of treatment response with penalized logistic regression (elasticnet). Model β estimate shown in solid orange lines, with the model 95% confidence intervals in dashed orange lines (n=14).

1B: V_T /f_P [¹¹C]DASB voxel maps from single subjects representing range of clinical response.

1C: Regression plots showing model adjusted BP_F [¹¹C]CUMI-101 binding potential (X-axis) by HDRS-24 post-treatment score (Y-axis) in the regions selected as significant predictors of treatment response with penalized logistic regression (elastic-net). Model β estimate shown in solid orange lines, with the model 95% confidence intervals in dashed orange lines (n=14).

1D: Single subject voxel-wise BP_F [¹¹C]CUMI-101 voxel maps representing range of clinical response.

Raphe nucleus pre-treatment 5-HT_1A binding did not significantly predict treatment response (p=0.18; ). The pre-treatment 5-HT_1A aggregate score (across the 12 post-synaptic regions) significantly predicted treatment response (F=5.74, p=0.035; Fig 2B). A subset of the 12 post-synaptic regions were found to be significant predictors of treatment response when fit in separate models (amygdala: p=0.016, hippocampus: p=0.015, parahippocampal gyrus: p=0.013, temporal cortex: p=0.039; (Fig 1C/D)). With 10 repetitions of 5-fold cross-validation, the LASSO linear regression analysis using 5-HT_1A binding across the a priori regions and pre-treatment HDRS-24 selected amygdala, hippocampus, and parahippocampal gyrus as statistical predictors of post-treatment HDRS-24 (coefficients of all other entered independent variables were shrunk to zero), agreeing with the previous linear models. Extracting predicted HDRS-24 estimates from this model (using the aggregate score across all terminalpost-synaptic field regions), remission status was predicted with 85.7% accuracy, 87.5% sensitivity, and 80% specificity.

Combined logistic regression analyses using pre-treatment 5-HTT and 5-HT_1A binding(38) across the a priori regions and pre-treatment HDRS-24 selected a subset of regions (amygdala, midbrain, and parahippocampal gyrus) and pre-treatment HDRS-24 score as significant predictors of post-treatment HDRS-24, partially agreeing with the previous linear models. Extracting predicted HDRS-24 post-treatment estimates from the model, remission status was predicted with 84.6% accuracy, 87.5% specificity, and 60% sensitivity, slightly lowering the predictive accuracy of 85.7% obtained using 5-HT_1A binding alone.

Group Differences

No significant differences emerged in binding between controls and pre-treatment ([¹¹C]-DASB: F=0.35, p=0.56; [¹¹C]-CUMI raphe: F=2.24, p=0.14, post-synaptic: F=0.60, p=0.44); or control and post-treatment BPD groups in any region ([¹¹C]-DASB: F=1.69, p=0.20; [¹¹C]-CUMI raphe: F=2.11, p=0.15, post-synaptic: F=0.78, p =0.38).

Associations with Clinical Response

Post-lithium-treatment [¹¹C]DASB or [¹¹C]-CUMI-101 binding did not significantly differ from binding in pre-treatment scans in the a priori [¹¹C]DASB regions (F=1.41, p=0.25; Fig 3A) or in the a priori [¹¹C]-CUMI-101 regions (raphe: F=0.43, p=0.52; post-synaptic regions: F=0.26, p=0.62; Fig 3B), respectively.

Figure 3A:

Group-wise averaged V_T /f_P voxel maps displaying lithium effect on [¹¹C]DASB 5-HTT binding, where pre-treatment is in top pane and post-treatment is in bottom pane.

3B: Group-wise averaged BPF voxel maps displaying lithium effect on [¹¹C]-CUMI-101 5-HT1A binding, where pre-treatment is in left pane and post-treatment is in right pane.

There was no significant relationship between the change in [¹¹C]DASB binding pre- to post treatment and lithium treatment response (F=1.27 p=0.28; Fig 4A) or the change in [¹¹C]-CUMI-101 binding pre- to post-treatment (raphe: F=1.26, p=0.28; post-synaptic regions: F=1.25, p=0.30 for aggregate score; Fig 4B).

Figure 4A:

Regression plots of model adjusted V_T/f_P (X-axis) by the change in HDRS-24 score (Y-Axis) in a representative a priori region (n=19).

4B: Regression plots showing model adjusted BP_F (X-axis) and the change in HDRS-24 score (Y-Axis) in representative a priori regions: Raphe Nucleus (somatodendritic), Hippocampus (post-synaptic) (n=13).

Discussion

Prediction of Treatment Response

Here we find that 5-HT_1A binding before lithium treatment, both as an aggregate score across all 12 a priori post-synaptic regions together, and temporal cortex regions independently (amygdala, hippocampus, parahippocampal grys), was predictive of post-treatment clinical response following 8-weeks of lithium treatment with over 85% accuracy. The temporal lobes (including amygdala and hippocampal formation) have been previously implicated in mood disorders such as unipolar and/or bipolar depression(39, 40), and have been found to be differentially responsive to antidepressant treatment(18, 41, 42). Additionally, previous studies from our lab have found that higher 5-HT_1A binding (including in temporal regions) were associated with improved clinical response following 3 months of treatment with SSRIs (18). We also found that 5-HTT binding before lithium treatment, both as an aggregate score across all a priori regions, and each region independently, was able to predict post-treatment clinical response with over 70% accuracy. This accuracy is improved to almost 85% when both 5-HTT and 5-HT_1A were included in the predictive model; however, the predictive accuracy of the combined serotonergic metrics was not better than 5-HT_1A alone.

Contrary to our previous findings and our hypothesis, the relationship we found was between lower 5-HT_1A pre-treatment binding and post-treatment HDRS-24 score as well (the hypothesized) lower 5-HTT pre-treatment binding and post-treatment HDRS-24 score. That is, those subjects with the lowest 5-HTT levels (greatest pre-synaptic serotonergic abnormality, but smallest post-synaptic abnormality) find the greatest benefit from lithium treatment. Our data, then, is consistent with the idea that lithium may act in the pre-synapse to ameliorate depressive symptoms, though potentially not via the candidate proteins examined here.

Interestingly, in both 5-HT_1A, 5-HTT, and the combination analysis, the amygdala was found to be a region of significant potential. This is noteworthy, as the amygdala is known to be integral to emotional regulation, especially in relation to BPD(43). Moreover, amygdalar abnormalities including structural and functional changes have been seen in patients with BPD as compared to controls(43). Our group has previously found that decreased 5-HTT binding in the amygdala was predictive of future remission in MDD(19). It is possible the amygdala represents a niche for serotonergic abnormalities present in depression, despite the heterogeneity of the condition. Future work will be necessary to further test this hypothesis.

Taken together with our previous findings(18), we find 5-HT_1A levels before treatment could be useful in determining successful treatment course (either via SSRIs or lithium) in BPD. If alterations to 5-HT_1A binding do prove to be predictive of treatment outcome, this could have great clinical utility, aiding in treatment selection. Data such as those presented here are the first steps toward personalized treatment plans that will increase the likelihood of treatment success and eventual remission.

Lithium Treatment Response

Despite preclinical evidence suggesting elevations in 5-HTT and downregulation of 5HT_1A following lithium exposure(6, 16, 44) we found, in vivo, that lithium monotherapy does not significantly alter 5-HTT or 5-HT_1A binding either cortically or subcortically post-treatment. As there was no change in post-treatment binding, it is not surprising that no relationship was found between the change in binding pre-to-post lithium treatment and clinical response. Given that previous studies did not report a relationship between 5-HTT binding and polymorphisms of the 5-HTT gene(12, 45), we did not look at genotype of study participants, and think that this is unlikely to drive this finding.

5-HTT and 5-HT1A Binding in Bipolar Depression: Comparison to Previous Findings

As expected, and recapitulating our previous findings(46), here we find no differences in 5-HTT [¹¹C]-DASB binding between unmedicated patients with BPD and healthy controls. This is inconsistent with a large body of evidence showing dysfunction in 5-HTT as a marker of depression(12, 47). It is possible that a difference in 5-HTT [¹¹C]-DASB binding between depressed patients with BPD and controls is small. If this is true, then the outcome is not likely to be clinically relevant to BPD as a whole. Further, our findings indicate that in this sample of patients with BPD, 5-HT_1A binding was not significantly different from healthy controls, contrary to our previous findings 5-HT_1A antagonist, [¹¹C]-WAY100635(31). There are several differences between this sample and the previous sample that could contribute to discrepant findings: 1. the previous study found a sex-depending interaction (in males), however the smaller sample size in the current study precluded this type of analysis; 2. the previous study used an antagonist tracer [¹¹C]-WAY100635, whereas we used a partial agonist tracer [¹¹C]-CUMI-101; differences in the two findings could represent a difference in low affinity 5-HT_1A (perhaps 5-HT_1A availability) that we do not see when specifically examining the high-affinity state. 3. the previous sample was predominantly medication exposed (only 5% medication naïve) whereas we have a higher proportion of medication naïve subjects (38%), though not enough to test their hypothesis that medication naïve patients with BPD would demonstrate exaggerated 5-HT_1A differences from controls. We posit that the decreased 5-HTT and elevated 5-HT_1A, from previous studies may not be trait markers for BPD, but could be state marker of a specific subset of patients (e.g. males) and a receptor subgroup (e.g. low affinity 5-HT_1A). Future studies with larger, and more heterogeneous patients cohorts can test these hypotheses.

Lithium’s Mechanism of Action: An Update

How, then, does lithium work to stabilize mood? It is possible that lithium modulates treatment response via alternative serotonergic mechanisms. For example, it is possible lithium acts indirectly to inhibit 5-HT_1B synaptic autoreceptors (via inhibition of GSK3β), thereby increasing serotonin release(48). This model is further complicated by the fact that 5-HT_1A receptors have been shown to mediate the relationship between serotonin release and GSK3β inhibition(49), and 5-HTT (serotonin transporter) levels are regulated by 5-HT_1B receptor expression. This is consistent with our findings that lithium treatment most benefits subjects with pre-synaptic serotonergic abnormalities (low 5-HTT would allow for prolonged 5-HT in the synapse if 5-HT release is increased). PET studies with 5-HT_1B radiotracers, such as [¹¹C]AZ10419369, would help determine whether lithium acts through 5-HT_1B to mediate treatment response. Alternatively, it is possible that lithium’s effects are much more complex, leading to changes in a number of systems(50). In this hypothesis, lithium would alter the overall excitation-inhibition balance(51), thereby stabilizing mood. Future studies would be needed to test these and other mechanistic models of lithium’s mood stabilizing effects.

Despite our finding that lithium may not act directly by changing 5-HTT or 5-HT_1A density, it is still possible that pretreatment 5-HT_1A levels could provide valuable information about a subset of participants that will find symptom relief following Lithium treatment. This leaves us with two interesting avenues to follow-up with: 1) Is there some follow-up time post Lithium treatment where 5-HT_1A or 5-HTT density is in fact changed – will we find pre-post differences following a longer follow-up period; 2) What is the upstream target of Lithium, how does it affect clinical symptoms, and how does this target interact/mediate serotonergic tone.

Study Strengths, Limitations, & Future Directions

The strengths of this study include a longitudinal treatment and PET imaging design, a comprehensive PET protocol that takes into account pre- and post-synaptic measures of serotonergic function in each subject, and full quantification PET methods that are not biased by reference region estimates. Limitations include: 1. a small sample size, which precludes investigation into interaction factors, such as sex or BPD subtype; 2. a short washout duration (limited to 3 weeks for ethical reasons), which may not be sufficient to abate treatment effects, though the length of time required for this is unknown and many studies are confounded by heterogenous medication use; 3. scans for both patients and controls conducted across multiple sites, though, importantly, all pre- and post-treatment scans within-participant were conducted at the same site and multiple sites provide a more representative sample.

Conclusion

Understanding the mechanism of one of the few effective treatments for BPD is of great importance and will help to determine a biological marker for treatment response. We find, though lithium may not act directly modulate 5-HTT or 5-HT_1A levels, pretreatment 5-HT_1A binding is predictive of post-treatment remission status. Future studies should further evaluate 5-HT_1A PET as a potential biomarker for lithium treatment response.

List of abbreviations:

BPD

Bipolar Depression

PET

Positron Emission Tomography

MRI

Magnetic Resonance Imaging

5-HTT

Serotonin Transporter

5-HT_1A

Serotonin-1A receptor

HV

healthy volunteer

SSRI

Selective Serotonin Reuptake Inhibitor

CUMC

Columbia University Medical Center

BNL

Brookhaven National Laboratory

Yale

Yale University Medical Center

SBU

Stony Brook University

DSM-IV

Diagnostic and Statistical Manual of Mental Disorders – version IV

HDRS-24

Hamilton Depression Rating Scale – 24 item

LEGA

Likelihood Estimation in Graphical Analysis

V_T

Distribution Volume

f_P

Free Fraction in Plasma

BP_F

Binding Potential

SCID

Structured Clinical Interview for DSM-IV