Abstract
Response to drug treatment of major depression is variable and biomarkers of response are needed. Cyclic AMP response element binding protein (CREB) is considered a key mediator of antidepressant drug effect. We studied CREB in T-lymphocytes as a potential predictor of response to a selective serotonin reuptake inhibitor (SSRI) in 69 Korean depressed patients. We determined total CREB (tCREB), phosphorylated CREB (pCREB) and CRE-DNA binding using immunoblot and electrophoretic mobility shift assays, at baseline and after 6 wk treatment. Thirty-four healthy controls were also studied. The rate of response was 36 of 69 cases (52%). Baseline levels of tCREB and pCREB were lower in the total depressed group compared to controls (p = 0.044 and p<0.001, respectively). Baseline tCREB values in responders were significantly reduced in comparison to non-responders and to controls. After 6 wk treatment, median values of change of all CREB measures were greater in responders (36) than in non-responders (33; p<0.001 for tCREB, p = 0.003 for pCREB, and p=0.072 for CRE-DNA binding). Similar but less robust changes in CREB variables distinguished remitters from non-remitters. The optimum value of baseline tCREB predicted response with a positive predicted value of 0.778 [21/27; 95% confidence intervals (CI) 0.621–0.935], negative predictive value of 0.643 (27/42; 95% CI 0.498–0.788) and accuracy of 0.695 (48/69; 95% CI 0.586–0.804). Patients with low baseline tCREB had a significantly greater rate of response (78%) than patients with high baseline tCREB (36%), p < 0.001. Moreover, the greatest changes in tCREB with treatment were observed in subjects who did respond. This preliminary study suggests that T-lymphocytic CREB biomarkers are reduced in depressed patients and may assist in the prediction of response to SSRI drugs in depression.
Introduction
Response to antidepressant drug treatment of depression is variable and often delayed by 2–6 wk (Nierenberg & Amsterdam, 1990). The time lag suggests that several cellular and molecular mechanisms are involved in effecting a response. Candidate mediators include post-synaptic signal transduction processes and gene transcriptions following antidepressant actions on pre-synaptic monoamine neurotransmitter transporters (D'Sa & Duman, 2002). Moreover, 30–50% of patients do not respond to an initial course of antidepressant drugs (Trivedi et al.2006). Thus, discovery of predictive biomarkers is a current research priority (Papakostas & Fava, 2008).
Most antidepressants act initially by blocking the presynaptic reuptake sites and increasing the concentrations of one or more monoamine neurotransmitters in the synaptic cleft (Bianchi et al.2002). The neurotransmitter signal on post-synaptic receptors is increased, and is followed by a series of well characterized post-synaptic signaling steps, which include cyclic AMP increase, protein kinase A (PKA) activation and translocation to the nucleus, then activation of the transcription factor cyclic AMP response element binding protein (CREB) through phosphorylation. In turn, CREB initiates transcription by recruiting polymerase and co-factors. An important target of CREB is brain-derived neurotrophic factor (BDNF), which has been implicated in neuronal plasticity and antidepressant response (Duman et al.1999; Imaki et al.1994).
CREB is a ubiquitous 43 kDa transcriptional factor belonging to the leucine zipper family (Duman et al.1999) and is a final convergent molecular target of several signalling pathways such as c-AMP, calcium or calmodulin, and Ras-dependent kinases (Gonzalez & Montminy, 1989; Tiraboschi et al.2004) and is relevant to the mechanism of antidepressant action (Blendy, 2006). CREB exists in active (phosphorylated) or inactive (non-phosphorylated) forms, which are determined by phosphorylation of Ser133 (Gonzalez & Montminy, 1989). The phosphorylated CREB (pCREB) binds to the CRE-consensus sequence on the promoter region of DNA, which enters the serial transcription process (Montminy & Bilezikjian, 1987; Yamamoto et al.1988).
We investigated whether the CREB variables – active and non-active forms of CREB, and functional activity of CRE-DNA binding – expressed in peripheral T-lymphocytes are altered in depressed patients and whether these measures are related to antidepressant drug response during treatment with a selective serotonin reuptake inhibitor (SSRI). We also evaluated whether the CREB biomarkers have potential for predicting antidepressant response. As an in vitro model system, T-lymphocytes possess the desirable feature of expressing functional serotonin transporter (5-HTT) activity with similar pharmacological properties to 5-HTT in brain (Barkan et al.2004). The primary hypotheses were that baseline CREB variables and also change in CREB variables during treatment would differ among groups – healthy controls, responding patients and non-responding patients.
Method
Subjects
We studied 69 clinically referred out-patients with major depressive disorder, treated in the Department of Psychiatry, Samsung Medical Center. The age range was 54–86 yr. Entry criteria were a diagnosis of a unipolar major depressive episode made by an experienced psychiatrist, using Diagnostic and Statistical Manual of Mental Disorders-IV (DSM-IV) criteria, confirmation of these clinical diagnoses by the Structured Clinical Interview for DSM-IV and informed consent for a 6-wk trial of antidepressant monotherapy. A minimum baseline 17-item Hamilton depression scale (HAMD) score of 15 was required (Hamilton, 1967). Exclusion criteria were significant medical conditions, abnormal laboratory baseline values, unstable psychiatric features (e.g. suicidal attempt), history of alcohol or drug dependence, seizure, neurological illness or concomitant Axis I psychiatric disorder, or Axis II disorders. A total of 34 healthy volunteers were recruited through an advertisement and potential subjects with a history of psychiatric illness or significant neurological or medical illness were excluded. They were then screened by the Minnesota Multiphasic Personality Inventory (MMPI), on which they had validity and clinical scale scores ranging from 35 to 65 (Colligan et al.1984). The protocol was approved by the ethics review board of Samsung Medical Center, Seoul, Korea. Signed informed consent was obtained from all participants. The study is registered (NCT00817011) in ClinicalTrials.gov.
Clinical procedures
All patients received open label monotherapy with one of three SSRI drugs: fluoxetine; sertraline; or paroxetine. No patient had received psychotropic medication within 2 wk of the study or fluoxetine within 4 wk. Choice of drug was driven by anticipated side-effects in at-risk individuals (Karasu et al.2000). We restricted the study to SSRI drugs because of pharmacodynamic differences of signal transduction among various classes of antidepressants (Kim et al.2006; Thome et al.2000). Altogether, 73 patients commenced the study (40 fluoxetine, 19 sertraline, 14 paroxetine). A dose of 0.5–1 mg lorazepam was allowed at bedtime for insomnia. Other psychotropic medications including neuroleptics were not allowed. Dose titration was completed within 2 wk. Doses were titrated into the usual range based on tolerability and side-effects. Plasma samples to measure trough drug levels and to document adherence were drawn at the end of week 6. Mean doses of antidepressant were: fluoxetine 29.3 mg/d (range 20–40 mg/d); paroxetine 20.9 mg/d (range 10–40 mg/d); sertraline 71.1 mg/d (range 50–100 mg/d). Patients were seen by a psychiatrist, who monitored their adverse events by the UKU scale (Lingjaerde et al.1987) at weeks 0, 1, 2, 4 and 6. HAMD-17 was administered by a single trained rater at 0 and 6 wk. The rater and all laboratory technicians were blinded to the hypotheses of the study. To maintain the blindness, a trained research coordinator managed the data and schedules. HAMD and experimental data were not disclosed to the psychiatrist and the rater was blinded to the experimental data. Response to medication was defined as ⩾50% decrease of HAMD at 6 wk relative to baseline and remission as a HAMD score ≤7 at 6 wk. Four patients with insufficient T-lymphocyte numbers for laboratory analyses were excluded. Thus, the final sample comprised 69 patients (39 fluoxetine, 18 sertraline, 12 paroxetine).
Laboratory procedures
In order to minimize variance due to circadian variation (Pietraszek et al.1992), we collected peripheral venous blood from all subjects between 09:00 hours and 11:00 hours. T-lymphocytes were purified using a Ficoll– Histopaque density gradient and MACS column (magnetic cell sorting, CD3 + ; Miltenyi Biotec, Germany). Nuclear extraction was conducted by methods reported previously (Fujioka et al.2004; Itoh et al.2004). The nuclear extracts were stored at −80 °C until use.
Nuclear extracts (10 µg) were evaluated by immunoblots using anti-tCREB (Cell Signaling Technology, USA), pCREB (Upstate, USA), or Lamin A/C (Santa Cruz Biotechnology, USA). The CRE-DNA binding activity was quantified in the nuclear extracts by binding CRE-oligonucleotide (Promega, USA) labelled with [32P]ATP-γ (PerkinElmer NEN, USA, 10 000 cpm). To confirm CRE-DNA binding, shift bands were determined by an electophoretic mobility shift assay (EMSA) after incubation at 4 °C for 18 h with 1:5 diluted anti-CREB antibody, and non-specific binding was determined by incubation for 1 min with cold-3 µm CRE oligonucleotide before adding labelled CRE-oligonucleotide. The optical density of the bands was quantified through image analysis software (Image Lab, version 2.2.4.0, MCM design, Denmark). The lower detection limit of the assay for baseline CREB was 28.6 optical density (OD). Determinations were made for total CREB (tCREB), which represents both the activated and non-activated forms of CREB, for phosphorylated CREB (pCREB), which is the activated form, and for CRE-DNA binding. The average between-assay coefficients of variation (CV) were 2.92% for tCREB, 1.35% for pCREB, and 2.65% for CRE-DNA binding.
Statistical analyses
Changes in the CREB measures were calculated as the difference between week 0 and week 6. Continuous variables were summarized as median and interquartile range, and compared using the Mann–Whitney test between the two groups due to non-normality. Categorical variables are presented with frequency and proportion. Fisher's exact test was performed to compare the distributions of categorical variables between groups. To compare continuous variables among three groups, we applied the Kruskal–Wallis test followed by Tukey's test using ranks for multiple comparisons. The possible relationships between continuous variables were tested by Pearson's correlation analysis or by Spearman's correlation analysis depending on normality of distribution. For the analysis of relationships adjusting for other variables, partial correlation analysis was applied. Using multiple logistic regression analysis, receiver operating characteristic (ROC) curves were derived to determine the optimum cut-offs of baseline CREB variables to maximize the sum of sensitivity and specificity in predicting response. Areas under the ROC curves (AUC) and prediction of response by baseline CREB variables were then calculated. With the optimum cut-off, we calculated overall accuracy, positive predictive value (PPV), negative predictive value (NPV), sensitivity and specificity with 95% confidence intervals (CI). Predictors for binary outcomes were investigated adjusting for covariates using multiple logistic regression analysis. Probability values were adjusted with Bonferroni's correction for multiple testing. Two-tailed significance testing was applied to all analyses. The criterion for significance of all tests was set at p < 0.05. For power analyses we assumed effect sizes of 0.8 for Cohen's d, which resulted in 80% power at a significance level of 0.05 (Cohen, 1992). All statistical analyses were conducted using SAS version 9.1.
Results
Characteristics of subjects
A total of 69 depressed patients and 34 healthy controls were compared. There were no significant differences in age and gender between the two groups (Table 1). Rate of response to SSRI drugs was 36 (52.2%) of 69 patients. There were no significant differences between responders and non-responders in age, gender, family history, age of onset, number of past depressive episodes, duration of current episode and HAMD score at week 0 (Table 1). Twenty-five patients achieved remission: 36.2% of all 69 patients; 69.4% of all 36 responders. No significant differences between remitters and non-remitters were seen in age, gender, family history, age of onset, number of past depressive episodes and duration of illness. Remitters had lower initial HAMD scores than non-remitters: 17.0 (15.0–20.0) for remitters; 20.0 (18.0–22.0) for non-remitters [medians (interquartile ranges); p=0.016]. Absolute rates and ratio of remission to response did not differ among the SSRI drugs (data not shown). The trough antidepressant drug concentrations obtained at 6 wk confirmed that all patients were treatment adherent.
Baseline characteristics of SSRI responders and non-responders
| Variables§ | Healthy controls (n = 34) | Responders (n = 36) | Non-responders (n = 33) | p |
|---|---|---|---|---|
| Age (yr) | 64.0 (61.0–67.0) | 64.5 (61.5–69.0) | 66.0 (59.0–71.0) | 1.000$ |
| Gender (M/F) | 15/19 | 10/26 | 10/23 | 0.910& |
| Family history (P/A) | 4/32 | 5/28 | 0.728# | |
| Onset age (yr) | 58.5 (51.0–63.0) | 57.0 (53.0–67.0) | 0.847* | |
| Episode no. | 2.00 (1.00–3.00) | 2.00 (1.00–2.00) | 0.858* | |
| Duration of current episode (months) | 3.50 (1.00–11.0) | 4.00 (2.00–8.00) | 0.933* | |
| Baseline HAMD score | 18.5 (16.0–22.0) | 20.0 (17.0–21.0) | 0.422* |
| Variables§ | Healthy controls (n = 34) | Responders (n = 36) | Non-responders (n = 33) | p |
|---|---|---|---|---|
| Age (yr) | 64.0 (61.0–67.0) | 64.5 (61.5–69.0) | 66.0 (59.0–71.0) | 1.000$ |
| Gender (M/F) | 15/19 | 10/26 | 10/23 | 0.910& |
| Family history (P/A) | 4/32 | 5/28 | 0.728# | |
| Onset age (yr) | 58.5 (51.0–63.0) | 57.0 (53.0–67.0) | 0.847* | |
| Episode no. | 2.00 (1.00–3.00) | 2.00 (1.00–2.00) | 0.858* | |
| Duration of current episode (months) | 3.50 (1.00–11.0) | 4.00 (2.00–8.00) | 0.933* | |
| Baseline HAMD score | 18.5 (16.0–22.0) | 20.0 (17.0–21.0) | 0.422* |
SSRI, Selective serotonin reuptake inhibitor; M/F, Males/Females; P/A, presence/absence; Episode no., lifetime number including current episode; Duration of current episode, duration months of current episode; HAMD, Hamilton Rating Scale for Depression.
Values are shown as median (25–75% interquartile range).
Kruskal–Wallis test, corrected p.
χ2 test, corrected p.
Fisher's exact test.
Mann–Whitney test.
Baseline characteristics of SSRI responders and non-responders
| Variables§ | Healthy controls (n = 34) | Responders (n = 36) | Non-responders (n = 33) | p |
|---|---|---|---|---|
| Age (yr) | 64.0 (61.0–67.0) | 64.5 (61.5–69.0) | 66.0 (59.0–71.0) | 1.000$ |
| Gender (M/F) | 15/19 | 10/26 | 10/23 | 0.910& |
| Family history (P/A) | 4/32 | 5/28 | 0.728# | |
| Onset age (yr) | 58.5 (51.0–63.0) | 57.0 (53.0–67.0) | 0.847* | |
| Episode no. | 2.00 (1.00–3.00) | 2.00 (1.00–2.00) | 0.858* | |
| Duration of current episode (months) | 3.50 (1.00–11.0) | 4.00 (2.00–8.00) | 0.933* | |
| Baseline HAMD score | 18.5 (16.0–22.0) | 20.0 (17.0–21.0) | 0.422* |
| Variables§ | Healthy controls (n = 34) | Responders (n = 36) | Non-responders (n = 33) | p |
|---|---|---|---|---|
| Age (yr) | 64.0 (61.0–67.0) | 64.5 (61.5–69.0) | 66.0 (59.0–71.0) | 1.000$ |
| Gender (M/F) | 15/19 | 10/26 | 10/23 | 0.910& |
| Family history (P/A) | 4/32 | 5/28 | 0.728# | |
| Onset age (yr) | 58.5 (51.0–63.0) | 57.0 (53.0–67.0) | 0.847* | |
| Episode no. | 2.00 (1.00–3.00) | 2.00 (1.00–2.00) | 0.858* | |
| Duration of current episode (months) | 3.50 (1.00–11.0) | 4.00 (2.00–8.00) | 0.933* | |
| Baseline HAMD score | 18.5 (16.0–22.0) | 20.0 (17.0–21.0) | 0.422* |
SSRI, Selective serotonin reuptake inhibitor; M/F, Males/Females; P/A, presence/absence; Episode no., lifetime number including current episode; Duration of current episode, duration months of current episode; HAMD, Hamilton Rating Scale for Depression.
Values are shown as median (25–75% interquartile range).
Kruskal–Wallis test, corrected p.
χ2 test, corrected p.
Fisher's exact test.
Mann–Whitney test.
Baseline CREB
Baseline tCREB immunoreactivity in depressed patients [60.5 (53.0–74.2)] was significantly lower than in controls [69.9 (61.7–89.9); p = 0.044]. Baseline pCREB immunoreactivity in patients [98.6 (86.6–112.3)] also was significantly lower than in controls [122.0 (113.9–127.8); p < 0.001]. CRE-DNA binding did not differ between depressed patients [72.6 (60.3–81.8)] and controls [66.8 (60.7–80.8); p=0.568].
Comparison among SSRI responders, non-responders, and controls demonstrated significant overall differences for baseline tCREB and pCREB (p=0.000 for tCREB, p=0.000 for pCREB; Table 2). The level of baseline tCREB proteins was significantly lower in drug responders than in non-responders and also lower than in controls (responders vs. non-responders, p = 0.004; responders vs. controls, p=0.001; Table 2). The baseline pCREB protein level was also significantly lower in responders compared to healthy controls (p = 0.000), but was not significantly different between non-responders and responders (p=0.546; Table 2). However, baseline CRE-DNA binding level was not different among the three groups (p=0.064; Table 2).
Baseline level of transcription factor, CREB
| Variables§ | Con (n = 34) | Responders (n = 36) | Non-responders (n = 33) | P$ | p# Con vs. R | p# R vs. NR | p# Con vs. NR |
|---|---|---|---|---|---|---|---|
| tCREB | 69.9 (61.7–89.9) | 56.7 (50.6–66.5) | 68.0 (58.7–77.6) | 0.000& | 0.001* | 0.004* | 0.833 |
| pCREB | 122.0 (113.9–127.8) | 93.8 (85.3–110.3) | 101.5 (90.1–114.4) | 0.000& | 0.000* | 0.546 | 0.000* |
| CRE-DNA binding | 66.8 (60.7–80.8) | 64.4 (58.2–77.3) | 75.9 (66.6–85.4) | 0.064 | 0.819 | 0.058 | 0.213 |
| Variables§ | Con (n = 34) | Responders (n = 36) | Non-responders (n = 33) | P$ | p# Con vs. R | p# R vs. NR | p# Con vs. NR |
|---|---|---|---|---|---|---|---|
| tCREB | 69.9 (61.7–89.9) | 56.7 (50.6–66.5) | 68.0 (58.7–77.6) | 0.000& | 0.001* | 0.004* | 0.833 |
| pCREB | 122.0 (113.9–127.8) | 93.8 (85.3–110.3) | 101.5 (90.1–114.4) | 0.000& | 0.000* | 0.546 | 0.000* |
| CRE-DNA binding | 66.8 (60.7–80.8) | 64.4 (58.2–77.3) | 75.9 (66.6–85.4) | 0.064 | 0.819 | 0.058 | 0.213 |
CREB, Cyclic AMP response element binding protein; Con, healthy controls; R, responders; NR, non-responders.
Expressed as optical density, arbitrary units, median (25–75% interquartile range); tCREB represents the immunoreactivity level of total CREB proteins, comprising both activated and non-activated forms; pCREB denotes active phosphorylated CREB; CRE-DNA binding denotes the binding level of the radioisotope-labelled CRE-consensus DNA sequence to pCREB.
Kruskal–Wallis test, not corrected p.
Significant differences among healthy controls, responders and non-responders (p < 0.05).
Multiple post hoc pairwise comparisons, Tukey's method.
Significant differences between two groups.
Baseline level of transcription factor, CREB
| Variables§ | Con (n = 34) | Responders (n = 36) | Non-responders (n = 33) | P$ | p# Con vs. R | p# R vs. NR | p# Con vs. NR |
|---|---|---|---|---|---|---|---|
| tCREB | 69.9 (61.7–89.9) | 56.7 (50.6–66.5) | 68.0 (58.7–77.6) | 0.000& | 0.001* | 0.004* | 0.833 |
| pCREB | 122.0 (113.9–127.8) | 93.8 (85.3–110.3) | 101.5 (90.1–114.4) | 0.000& | 0.000* | 0.546 | 0.000* |
| CRE-DNA binding | 66.8 (60.7–80.8) | 64.4 (58.2–77.3) | 75.9 (66.6–85.4) | 0.064 | 0.819 | 0.058 | 0.213 |
| Variables§ | Con (n = 34) | Responders (n = 36) | Non-responders (n = 33) | P$ | p# Con vs. R | p# R vs. NR | p# Con vs. NR |
|---|---|---|---|---|---|---|---|
| tCREB | 69.9 (61.7–89.9) | 56.7 (50.6–66.5) | 68.0 (58.7–77.6) | 0.000& | 0.001* | 0.004* | 0.833 |
| pCREB | 122.0 (113.9–127.8) | 93.8 (85.3–110.3) | 101.5 (90.1–114.4) | 0.000& | 0.000* | 0.546 | 0.000* |
| CRE-DNA binding | 66.8 (60.7–80.8) | 64.4 (58.2–77.3) | 75.9 (66.6–85.4) | 0.064 | 0.819 | 0.058 | 0.213 |
CREB, Cyclic AMP response element binding protein; Con, healthy controls; R, responders; NR, non-responders.
Expressed as optical density, arbitrary units, median (25–75% interquartile range); tCREB represents the immunoreactivity level of total CREB proteins, comprising both activated and non-activated forms; pCREB denotes active phosphorylated CREB; CRE-DNA binding denotes the binding level of the radioisotope-labelled CRE-consensus DNA sequence to pCREB.
Kruskal–Wallis test, not corrected p.
Significant differences among healthy controls, responders and non-responders (p < 0.05).
Multiple post hoc pairwise comparisons, Tukey's method.
Significant differences between two groups.
By ROC analysis, the AUC of baseline tCREB for response was 0.731 [95% CI (0.613–0.850), p=0.001]. The optimal cut-off of baseline tCREB for response was determined as 58.2 OD. The response rates differed significantly between the group with baseline tCREB OD ≤58.2 (78% or 21 of 27) and the group with baseline tCREB OD > 58.2 (36% or 15 of 42; p = 0.001). Predicting response for baseline tCREB ≤ 58.2 OD, and non-response for baseline tCREB > 58.2 OD provided sensitivity of 0.583 [21/36; 95% CI (0.422–0.744)], specificity of 0.818 [27/33; 95% CI (0.686–0.950)], PPV 0.778 [21/27; 95% CI (0.62–0.935)], NPV 0.643 [27/42; 95% CI (0.498–0.788)] and accuracy of 0.695 [48/69; 95% CI (0.586–0.804)].
The prior probability of response (52%) in the absence of knowledge of the tCREB OD biomarker increased to a posterior probability of 78% when the tCREB OD was ≤58.2. The prior probability of nonresponse (48%) increased to a posterior probability of 64% when the CREB OD exceeded 58.2.
Change of CREB and treatment response to SSRI drugs
We examined the change of CREB variables during 6 wk SSRI treatment. All the pre-post treatment changes of CREB variables were significantly different between responders and non-responders. Overall, median change (25–75% interquartile range) values of tCREB and pCREB immunoreactivity at 6 wk were positive in SSRI responders [tCREB: 2.50 (−2.55 to 9.90), pCREB: 5.50 (−5.55 to 14.75)], while they were negative in non-responders [tCREB: −5.80 (−12.50 to −1.70), pCREB: −9.20 (−18.40 to 2.20); p<0.001 for tCREB, p = 0.003 for pCREB; Fig. 1a). The change of binding level of radioisotope-labelled CRE-consensus DNA sequence to pCREB for 6 wk was not significantly different between SSRI responders [2.10 (−7.05 to 16.30)] and non-responders [−4.20 (−12.60 to 3.90); p = 0.072; Fig. 1a].
Cyclic AMP response element binding protein (CREB) change in responders and non-responders after 6 wk selective serotonin reuptake inhibitor treatment. (a) Changes of CREB variables in responders and non-responders: CREB change denotes change of CREB immunoreactivity, expressed as optical density (OD) change from week 0 (baseline) to week 6. tCREB denotes the immunoreactivity level of total CREB proteins, comprising both activated and non-activated forms. pCREB denotes active phosphorylated CREB. CRE-DNA binding denotes the binding level of radioisotope-labelled CRE-consensus DNA sequence to pCREB. Each marker indicates individual patient's CREB change. a Significant tCREB changes between responders and non-responders (Pcorrected = 0.000). b Significant pCREB changes between responders and non-responders (Pcorrected = 0.003). c CRE-DNA binding changes between responders and non-responders (Pcorrected = 0.072), All analysed by Mann–Whitney test with Bonferroni's correction. (b) Correlation between change of tCREB and change of CRE-DNA binding: tCREB represents the immunoreactivity level of total CREB proteins, comprising both activated and non-activated forms. CRE-DNA binding denotes the binding level of radioisotope-labelled CRE-consensus DNA sequence to pCREB. Change of CREB means OD change of total CREB immunoreactivity, calculated as the absolute difference between baseline and 6 wk. Change of CRE-DNA likewise is the difference between week 0 and week 6. Each point represents one subject (ρ = 0.613, Pcorrected = 0.000 by Spearman's ρ correlation).
Also, we found that baseline CREB values were in negative correlation with change values at 6 wk for all three CREB measures (tCREB: ρ = −0.432, p < 0.001, pCREB: ρ = −0.434, p<0.001, CRE-DNA binding: ρ = −0.325, p = 0.006).
As displayed in Fig. 1b, we observed that for the total patient group (n = 69), there were moderately positive correlations of changes among the CREB variables (r = 0.613, p < 0.001 for tCREB change and CRE-DNA binding change; r = 0.695, p < 0.001 for tCREB change and pCREB change; r = 0.566, p < 0.001 for pCREB change and CRE-DNA binding change).
CREB variables and remission
We also analysed the CREB variables in relation to remission. CREB variables were associated with remission, which is a more stringent outcome than response. The baseline level of tCREB in remitters [57.3 (50.8–66.1)] was marginally lower than the baseline level of tCREB in non-remitters [65.9 (56.4–77.1); p=0.050]. Baseline CRE-DNA binding also was lower in remitters [60.5 (57.9–74.3)] than non-remitters [75.8 (65.7–84.1); p=0.021. However, baseline pCREB did not differ between remitters [92.2 (86.4–112.3)] and non-remitters [100.7 (87.3–112.6); p=1.000].
As was found for response, the median changes of tCREB and pCREB from baseline to week 6 were positive in remitters [tCREB: 2.40 (−3.30 to 10.20), pCREB: 3.10 (−4.50 to 8.10)] but not in non-remitters [tCREB: −3.95 (−10.85 to 1.25), pCREB: −4.40 (−16.50 to 6.55)]. The tCREB difference was significant (tCREB: p=0.009; pCREB: p=0.144). CRE-DNA binding rose in both remitters [6.20 (−5.10 to 18.40)] and non-remitters [3.90 (−12.15 to 4.05); p=0.042; Fig. 2)].
Cyclic AMP response element binding protein (CREB) changes in remitters and non-remitters after 6 wk selective serotonin reuptake inhibitor treatment. tCREB denotes the immunoreactivity level of total CREB proteins, comprising both activated and non-activated forms. pCREB denotes the active, phosphorylated form of CREB. CRE-DNA binding means the binding level of radioisotope-labelled CRE-consensus DNA sequence to pCREB. Change of CREB means change of total CREB immunoreactivity, expressed as optical density change from week 0 (baseline) to week 6. Each marker indicates an individual patient. a Significant tCREB changes between remitters and non-remitters (Pcorrected = 0.009). b pCREB changes between remitters and non-remitters (Pcorrected = 0.144). c Significant CRE-DNA binding changes between remitters and non-remitters (Pcorrected = 0.042), All analysed by Mann–Whitney test with Bonferroni's correction.
Because baseline HAMD scores differed between remitters and non-remitters, we controlled on the baseline HAMD score as a possible confounding factor. By Spearman's partial correlation analyses, remission was related to tCREB change (p=0.007) and to CRE-DNA binding change (p=0.027) but not to pCREB change (p=0.084).
Discussion
Human peripheral lymphocytes express 5-HTT, the target of SSRI antidepressants, in the cell membrane, with similar pharmacological properties to 5-HTT in brain (Barkan et al.2004). These cells also express many components of the neuronal signal transduction cascade in their cytosol (Stieler et al.2001). Thus, human peripheral T-lymphocytes provide a means of studying CRE-mediated gene expression after exposure to 5-HTT reuptake inhibitors (SSRI drugs) and these cells have been used as a peripheral model of brain in several studies (Leuner et al.2007; Zhang et al.2008).
We evaluated CREB variables in peripheral T-lymphocytes as possible markers of response to SSRI drugs in major depression. Compared to healthy controls, baseline tCREB and pCREB were significantly lower in T-lymphocytes of depressed patients but CRE-DNA binding was not (Results). In pre-treatment comparisons of tCREB, values in responders were significantly lower than in non-responders or in controls (Table 2). The changes of both tCREB and pCREB at 6 wk significantly differed between responders and non-responders: responders showed greater increases (Fig. 1).
Two preliminary studies by Kochet al. in a small number of patients reported that peripheral T-lymphocytic pCREB change increased largely in patients who responded well to antidepressants or interpersonal treatment (IPT; Koch et al.2002; Koch et al.2009). In the present study, we investigated whether the several CREB variables (tCREB, pCREB, and CRE-DNA binding) in peripheral T-lymphocytes showed a different change depending on antidepressant response.
We confirmed that the median change values of all CREB variables were positive with treatment in SSRI responders but negative in non-responders (Fig. 1a). This difference suggests that the increased CREB signalling in responders is not simply a function of exposure to the drugs. As mentioned earlier, SSRI drugs target neurotransmitter transporters in brain and activate several post-synaptic signal transduction systems that converge upon CREB (Trivedi et al.2006). The therapeutic action of antidepressant drugs appears at about 2–6 wk after drug initiation (Papakostas & Fava, 2008) and during this time lag it is known that activation and transcription of CREB, and CRE-mediated expressions of several genes related to neuroplastic change occur (Bianchi et al.2002). In this study, increase of tCREB, pCREB and binding activity of CRE and consensus DNA sequence suggests the possibility that CRE-mediated gene expression was increased by the drugs primarily in patients who responded. The significant negative correlations between baseline values and change values of all CREB measures support this suggestion (Results).
From in vivo studies it is known that CREB is a transcription factor that results in the induction of gene expression for BDNF and trkB, which are believed to contribute to antidepressant response through their neurotrophic effect (Duman et al.1999). In the learned helplessness depression model, an antidepressant effect is seen when CREB is overexpressed by the herpes simplex virus-mediated gene transfer technique (Tiraboschi et al.2004). Further convergent evidence associating CREB with antidepressant response comes from reports that long-term antidepressant treatment causes increased mRNA and/or protein expression of CREB in rat hippocampus (Gonzalez & Montminy, 1989) and increased CREB protein expression in human post-mortem temporal cortex (Blendy, 2006). Also, BDNF protein expression was increased in post-mortem hippocampus in antidepressant treated-depressed patients compared to non-treated patients (Chen et al.2001). Taken together, these results found in the peripheral lymphocyte model suggest that responders have a pre-existing deficit of CREB signalling, which is reversed by the SSRI drugs.
We found significant correlations among changes of tCREB, pCREB, and CRE-DNA binding (Fig. 1b; Results). Our interpretation is that the three dependent variables constitute a sequential process wherein the transcription factor CREB is activated by the signal cascade initiated by the SSRI, and the activated pCREB can then bind to the cis-regulatory element, CRE, of DNA, which initiates transcription. The results also indicate that the experimental measurement values ‘CREB variables’ really represent the actual CREB-mediated transcription and gene expression. However, we found that the relationship between the change of CRE-DNA binding and tCREB (r = 0.613) or pCREB (r = 0.566) change was slightly weaker than between protein changes (tCREB and pCREB, r = 0.695). The results suggest that CRE-DNA binding is a transient step, which is known to occur within several dozen minutes after drug targeting, so it might be influenced by the period variations between drug administration time and venous sampling time among individual patients (West et al.2002).
We observed significant differences in baseline tCREB and CRE-DNA binding between responders and non-responders (Table 2). Interestingly, in depressed patients, responders showed lower baseline CREB variables than non-responders, suggesting an intrinsic under-activity of CREB in subjects who are destined to be responders. Moreover, such differences between responders and non-responders cannot be ascribed to baseline clinical or demographic differences (Table 1). We also found significant negative correlations between baseline values and change values after 6 wk for all CREB variables, such that a larger change after 6 wk was associated with lower baseline levels (Results). Moreover, each of these baseline measures was associated with response rather than non-response.
We performed ROC analysis to examine whether the baseline level of transcription factor, tCREB, might predict SSRI response (Results). The optimal cut-off value for discriminating responders from non-responders was a baseline tCREB OD of 58.2. With this criterion, we obtained an overall accuracy of 70% for predicting response or non-response. Based on this criterion, 81.8% of actual non-responders (27 of 33) would be predicted (specificity 81.8%), while 58.3% of actual responders (21 of 36) would be recognized (sensitivity 58.3%). Of 27 patients 21 predicted as responders would actually be responders (PPV 77.8%), while 64.3% of the patients predicted as non-responders (27 of 42) would actually be non-responders (NPV 64.3%). The group with tCREB OD < 58.2 showed a significantly higher response rate (78% or 21 of 27) than the group with tCREB OD > 58.2 (36% or 15 of 42; p=0.001, odds ratio = 6.3; Results). The prior probability of response and non-response in the absence of knowledge of the low tCREB OD biomarker (52 and 48%, respectively) increased to posterior probabilities of 78 and 64% with knowledge of the baseline tCREB OD. Thus, baseline tCREB OD may allow the clinician to predict outcome, especially non-response, to SSRI treatment with greater confidence than is currently possible. The superiority of specificity (0.82) compared to sensitivity (0.58) in these results presumably reflects the fact that the responder group comprises both specific drug responders and so-called placebo responders. These results add to the increasing evidence that biomarkers may provide useful information for clinical decision making (Biomarkers Definitions Working Group, 2001).
We found the baseline level of tCREB and pCREB in depressed patients was significantly lower than that of healthy controls (Results). This finding indicates that a decrease of baseline protein expression tCREB and pCREB in T-lymphocytes of patients was associated with depression, which is in line with previous results in post-mortem orbitofrontal and temporal cortex tissue of depressed patients (Dowlatshahi et al.1998, Yamada et al.2003). CREB targets many genes implicated in neuronal plasticity, such as BDNF (Duman et al.1999; Imaki et al.1994), so our finding implies that a decreased expression of CREB may represent a risk factor for depression due to reduced capacity for neuronal/synaptic plasticity. A previous animal study has demonstrated that glucocorticoid elevation in corticotropic-releasing hormone containing neurons in rat brain inhibited the phosphorylation and activation of CREB (Légrádi et al.1997). A recent study in peripheral neutrophils of depressed patients also reported a decreased tCREB protein expression compared to healthy controls (Ren et al.2011), which is similar to our results. Thus, our results give a measure of convergent validation to the CREB-neurotrophin theory of antidepressant drug action.
A potential limitation of our study is the reliance on a single tertiary care site for recruitment, which might result in selection bias. However, the advantage of the single-site design is strict quality control of the protocol. Another limitation of our study is that it includes a large variation in distribution of CREB variables in a small population (Figs. 1a and 2). These promising initial data will need to be examined in larger populations and in other centres.
This study demonstrates: (1) baseline tCREB protein concentrations in T-lymphocytes of depressed patients were lower than healthy controls and allowed incremental prediction of antidepressant response or non-response to SSRI drugs at 6 wk; (2) Increase of CREB variables after 6 wk SSRI treatment was associated with lower baseline values, and with antidepressant drug response and remission. Thus, CREB measures in peripheral T-lymphocytes may be candidate biomarkers of treatment response to SSRI antidepressants. Further, large-scaled studies, including younger patients, are needed to evaluate the robustness of our findings.
Acknowledgements
This study was supported by grants from the Korea Science and Engineering Foundation (KOSEF) NRL Program (Grant R0A-2007-000-20129-0), and the Korean Health Technology R&D Project (A110339) and the National Project for Personalized Genomic Medicine (A111218-PG02) from Ministry for Health & Welfare, Republic of Korea.
Statement of Interest
None.
References
