Does the 5-HT 1A rs6295 polymorphism influence the safety and efficacy of citalopram therapy in the oldest old?

Patients and recruitment

Study recruitment took place between January 2010 and January 2013 at Brighton and Sussex University Hospitals NHS Trust, UK. To be eligible for participation in the study, participants had to meet the following inclusion/exclusion criteria and provide written informed consent:

Inclusion criteria: ⩾80-years old and admitted as an inpatient under the care of the elderly team; have a new clinical diagnosis of depression on admission or during inpatient stay that required treatment with the local hospital formulary SSRI of choice, citalopram 20 mg once daily.

Exclusion criteria: a current prescription for an antidepressant regardless of indication; patients lacking capacity, as determined by a score of ⩽7/10 on a routine Abbreviated Mental Test Score (AMTS) conducted on admission.

Ethical approval

Ethics approval was granted by both NHS Research Ethics Committee South-East Coast, Brighton Sussex Research Ethics Committee Reference 09/H1107/116 and the Medicines and Healthcare products Regulatatory Agency, alongside the University of Brighton Research Ethics Committees. The study is listed on the EU Clinical Trials Register [EudraCT number 2009-016716-20]. All experiemental procedures were conducted in accordance with Human Tissue Authority and Good Medical Practice regulations and guidelines.

Measurement of depression and clinical status

A full clinical and biochemical assessment took place at baseline, 1 week, and 4 weeks after starting SSRI treatment to determine the efficacy, tolerability and safety of citalopram therapy. The duration of 4 weeks was chosen as the temporal endpoint, as current UK guidelines suggest that if no repsonse is observed at this point, an increase in dose or switch to an alternative medication is required. Measurements included: the Geriatric Depression Scale (GDS; 30/30 long form); Mini Mental State Examination Score (MMSE); Hunter Serotonin Toxicity Criteria (HSTC); urea and electrolytes, and full blood count. Remission was defined as a GDS score of ⩽11 at week 4 of the study. ¹⁵

Assessment of plasma citalopram levels

Plasma citalopram levels were determined at 4 weeks through enzyme-linked immunosorbent assay (Neogen^®, Ayr, Scotland, UK).

Assessment of platelet serotonin levels

Serotonin concentrations in platelet pellets were measured using a high-performance liquid chromatography (HPLC) system which consisted of a Jasco HPLC pump (model: PU-980, Great Dunmow, United Kingdom) and Rheodyne (Schenkon, Switzerland) manual injector equipped with a 20 µl loop. A Kinetic^® (Macclesfield, United Kingdom) ODS (octadecylsilyl) 2.6 µm 150 mm × 4.6 mm i.d. (internal diameter) analytical column with a guard column (Phenomenex^®, Macclesfield, UK) was employed. The HPLC system was run at a flow rate of 100 µl/min. A CHI630B potentiostat (CH Instruments, Austin, TX, USA) was used to control the detector voltage and record the current. A 3 mm glassy carbon electrode (flow cell, BASi^®, West Lafayette, IN, USA) served as the working electrode and was used with a silver|silver chloride (Ag|AgCl) reference electrode and a stainless steel block as the auxiliary electrode. Amperometric recordings were carried out, where the working electrode was set at a potential of +950 mV versus Ag|AgCl reference electrode. Control and data collection/processing were handled through the CHI630B software. Briefly, 500 µl of ice cold 0.1 mol/l perchloric acid was added to the platelet pellet and samples were sonicated and vortexed for 2 min and then centrifuged at 14,600g for 10 min prior to chromatographic analysis. The supernatant was removed and filtered through a 0.2 µm filter and the resulting solution analysed using HPLC with electrochemical detection.

Characterization of 5-HT_1A polymorphisms

Deoxyribonucleic acid (DNA) was extracted from 200 µl of whole blood using DNA extraction columns (DNeasy Blood and Tissue Mini-kit, Qiagen, Hilden, Germany). The 5-HT_1A receptor gene promotor has an SNP, rs6295, at position −1019C/G, which was typed by amplification using the following primers: forward 5’ TGTCGT-CGTTGTTCGTTTGT 3’ and reverse 5’ GGT-GAACAGTCCTGGGTCAG 3’. ¹⁶ Amplifications were carried out in 25 µl reactions containing approximately 5 ng of DNA template and final concentrations of 15 nmol/l 10× reaction buffer, 1.5 mmol/l MgCl₂ (magnesium chloride), 0.2 mmol/l for each deoxynucleotide (dNTP), 10 pmol/l forward and reverse primers, 1 unit Platinum®Taq DNA polymerase (Invitrogen, Darmstadt, Germany). Cycling was performed in a Techne TC-4000 (Cambridge, United Kingdom) thermal cycler employing 40 cycles (30 s at 94°C, 30 s at 56°C and 60 s at 72°C), with a final extension at 72°C for 10 min. Sequencing (Sanger sequencing) procedures were performed by Source-Bioscience (Nottingham, England) to determine the nucleotide at position −1019.

Frequentist analysis

Chi-square tests were performed to determine the significance of any allele/genotype associations with the incidence of serotonergic side effects or the efficacy of citalopram therapy (remission, defined as a GDS score of ⩽11 at week 4). General linear regression models were fitted to estimate the relative influence of demographics and genotypes on variation in response to treatment with citalopram, and the incidence of side effects. A one-way analysis of variance with Fisher’s least significant difference (LSD) post hoc test was used to assess statistical significance of differences in the effects of citalopram according to genotype. Effect size η² was calculated as the sum of squares between groups/total sum of squares. To assess the statistical significance of differences in platelet [5-HT], MMSE, and GDS over time, either a repeated-measures one-way analysis of variance (ANOVA) or Friedman test was used according to the distribution profile of the data. Normality of data was tested using the Shapiro–Wilk test, where we accepted the null hypothesis that the data were normally distributed if p > 0.05. Data analysis and graphical representation were performed in R (R Core Team) ¹⁷ and GraphPad Prism v6, Graphpad software, California, USA. Statistical significance was assumed if p < 0.05. Mean ± standard error of the mean are presented unless otherwise stated.

Bayesian forecasting

Following the method outlined in Mould and colleagues’ work, ¹⁸ we framed our research question in the following way: what is the conditional probability that individuals will not present in remission at week 4, that is, a GDS of ⩽11, given they are known to have a specific genotype and that they have embarked on a course of citalopram? (The details of this approach can be found in the supplementary material.)

We took the odds form of Bayes’ rule: the probability of not being in remission at week 4, relative to being in remission. The appropriate prior probability distribution to employ with a likelihood that takes this form is a beta distribution. ¹⁹ This distribution has two parameters, the values of which can be drawn from previously observed data relating to the unconditional probability that an older individual will go into remission following a course of citalopram. For example, the combined remission rates in six studies that investigated the use of citalopram in late-life depression is calculated at around 47% (n = 567). ²⁰ In this scenario, the distribution is broadly described as symmetrically bell shaped around a mean of 0.5 (corresponding to a probability of remission of 0.5). For comparison, we also considered the more conservative position that all possibilities for the probability of not responding to citalopram are equally likely, which is a specific case of the beta distribution and defines a non-informative uniform distribution.

To obtain the posterior distribution of the relative risk of not being in remission at week 4 relative to being in remission, conditional on the genotype, we applied Gibbs sampling in order to sample from the two distributions without having to explicitly calculate the integrals. This was performed in R (R 3.2.2) using the script employed in ‘OPTIMISE trial in a Bayesian framework’, ²¹ which incorporates the scripts ‘openGraphSaveGraph.R’ and ‘plotPost.R’. ¹⁹

Data availability

The datasets generated during and analysed during the current study are available from the corresponding author on reasonable request.

Baseline demographics

A total of 29 patients were enrolled into the study over a 3-year period. Recruitment was more difficult than expected due to the population being acutely unwell. Ten patients missed either the final, or both follow-up visits and were therefore excluded from the analysis. The mean age of the study group was 88 ± 4 years and the mean baseline GDS for the 19 patients remaining in the study was 15/30 ± 5. The genotype frequencies of the 5-HT_1A receptor was found to be 7:6:6 for CC, GC and GG respectively. By comparing the observed genotype frequencies with the expected frequencies (5:9:4 for CC, GC and GG respectively), we were able to confirm that our sample population does not deviate, by any large extent, from Hardy–Weinberg equilibrium (χ² = 2.55; p = 0.53, β-1 = 0.59). A summary of the baseline demographics is shown in Table 1.

OPEN IN VIEWER

Table 1.

Demographic and clinical data of study recruitments.

Baseline characteristics (n = 19)
Age (years)	88.16 (±3.8)
Sex (male:female)	5:14
Genotype (CC:GC:GG)	7:6:6
CrCl (ml/min)	41.01 (±30.8)
Weight (kg)	64.32 (±13.1)
Baseline MMSE	27 (±2)
Baseline GDS	15 (±5)
Hb (g/dl)	11.97 (±2.0)
WCC (×109/l)	9.91 (±4.7)
Plt (×109/l)	267.16 (±91.4)
Na+ (mmol/l)	138 (±6.0)
K+ (mmol/l)	3.9 (±0.6)
Urea (mmol/l)	9.2 (±3.6)
Ethnicity	White British 17 Mixed White and Black African 2

Values represent mean ± standard deviation (with the exception of sex and genotype).

C, cytosine; CrCl, creatinine clearance; G, guanine; Hb, haemoglobin; MMSE, Mini Mental State Examination Score; GDS, Geriatric Depression Scale; WCC, white cell count; Plt, platelet count; Na⁺, sodium; K⁺, potassium.

Response to citalopram

To measure the therapeutic effect of citalopram we compared GDS at baseline, 7 days and 4 weeks for all the patients. There was a trend towards a reduction in GDS over the 4 weeks of the study, although this was not found to be statistically significant (p = 0.11, repeated measures one-way ANOVA, n = 18 [1 patient did not attend the day 7 visit and so was excluded from this analysis], Figure 1a). A total of 8/19 patients reached remission by week 4 (that is, a GDS score of ⩽11). ¹⁵ Mean platelet [5-HT] reduced significantly over the duration of the study, indicating the pharmacological activity of citalopram on platelet 5-HT transporters, although we were only able to successfully measure concentrations in 7/19 participants (p < 0.001, Friedman test; p < 0.01 between baseline and 1 month, Dunn’s post hoc; n = 7; Figure 1b). Mean [citalopram]_plasma was 0.75 ± 0.17 mg/l.

OPEN IN VIEWER

Figure 1.

Effect of citalopram on clinical and psychological measures.

A clinical improvement in mood has previously been associated with improved cognition in older people. ²² We did show a significant correlation between change is GDS (ΔGDS) and change in MMSE (ΔMMSE) over the course of the study (R² = 0.27, p = 0.02, Pearson’s correlation; Figure 1c), although we were unable to detect a significant change in mean MMSE (p = 0.88, Friedman test, n = 19; Figure 1d) over the three visits.

Treatment with citalopram has been associated with several side effects which could be particularly problematic to older patients. Recognized complications of citalopram therapy include hyponatraemia, and gastrointestinal bleeding as a result of depletion of platelet 5-HT. We found a small, clinically, and statistically insignificant change in plasma Na⁺ concentration over the duration of the study (mean plasma [Na⁺] = 137.6 ± 1.5 versus 135.6 ± 1.2 versus 134.7 ± 1.4 mmol/l at baseline, 1 week and 4 weeks, respectively; p = 0.10, repeated measures one-way ANOVA, n = 16, Figure 1e). Plasma Hb concentration, which may fall due to gastrointestinal bleeding as a consequence of SSRI treatment did not change significantly over the course of the study (Figure 1f). Another rare, but serious reaction to SSRIs is serotonin syndrome. Using the Hunter Serotonin Toxicity Criteria, one of our 19 patients was identified as positive for serotonin syndrome over the 4-week period.

Factors that determine response to citalopram

Despite the overall small decrease in mean GDS score over time, it is clear from the individual data that some patients responded well to citalopram, whilst others showed a continued deterioration in mood (Figure 2a). To explore this further, we constructed a generalized linear model to determine whether certain baseline characteristics could predict response to citalopram (i.e. absolute change in 4-week GDS). We chose a generalized linear model due to the multinomial distribution of the data. The first iteration of the model included four predictor variables (5-HT_1A genotype, sex, age, and weight). We did not include ethnicity as a variable due to only two of our sample being nonwhite. In further iterations of the model, predictably, given the small sample size, individual variables that did not reach statistical significance were excluded until a minimum effective model was produced. ²³ The final model contained only 5-HT_1A genotype as a statistically significant predictor variable, explaining 32% of the deviance in GDS score (p = 0.005).

OPEN IN VIEWER

Figure 2.

The relationship between 5-HT_1A genotype and response to citalopram.

The effect of the 5-HT_1A genotype on response to citalopram

To probe the role of the 5-HT_1A genotype in response to citalopram, we looked at the mean change in GDS over the 4-week study according to the three genotypes (Figure 2b). The mean change in GDS for the CC, GC, and GG genotypes were: –5.0 ± 1.1, 1.8 ± 1.3, –4.0 ± 2.1 (p = 0.02, one-way ANOVA with Fisher’s LSD post hoc test; n = 7, 6, 6, respectively). These data show that individuals with the GC genotype demonstrate a significantly different response to citalopram at 4 weeks compared with either CC or GG [although only individuals with a CC genotype are significantly different from 0 (single sample t test, p < 0.05)]. Interestingly, individuals with a GC genotype showed a mean increase in GDS (i.e. worsening of depression), although this is not significantly different from 0. Genotype was found to be associated with the incidence of remission at week 4 (remission/total: CC = 5/7, CG = 0/6, GG = 3/6; χ² test, p = 0.031).

We also observed a significant reduction in absolute MMSE score in patients with a GC genotype, compared with the CC group (0.4/30 ± 0.7 versus –2.3/30 ± 1.0 versus 0.3/30 ± 0.5; p < 0.05, one-way ANOVA, Fisher’s LSD post hoc test, Figure 2c). The role of genotype on platelet [5-HT] is inconclusive due to small numbers in each group (Figure 2d). We found no relationship between changes in plasma Na⁺ and Hb concentrations, or the presence of serotonin syndrome and genotype over the study period (data not shown).

Relative risk of not responding to citalopram, conditional on genotype

The traditional frequentist treatment of our data presented above is indicative of an effect and, being the standard approach, is potentially of value for interstudy comparisons; however, this approach does not explicitly address the question of clinical interest. When making a decision about starting SSRI therapy, it may be more useful for a clinician to know the probability that an individual will respond poorly to citalopram, in the knowledge of their 5-HT_1A genotype. Bayesian inference provides a means of addressing the former question, and also determines the probability of our model (i.e. that there is a relationship between rs6295 genotype and response to citalopram). To perform this analysis, we first categorized our participants as either responders or nonresponders; where responders are those who reached remission (GDS score ⩽ 11) at week 4 and nonresponders are those who did not. Figure 3 shows the influence of the prior distributions (first column) on the likelihood of each genotype (first row) in arriving at the posterior relative risk of not responding (θ₁) to responding (θ₂). When an informative beta prior is considered, an individual that has the CG genotype is, on average, 1.79 times more likely to not remit at 4 weeks than to go into remission. The highest-density interval (HDI) spans between 0.8 and 3; however, the probability that the CG individuals have a higher chance of not remitting than remitting is 99.3% (proportion of the posterior distribution > θ₁ / θ₂ = 1). Figure 4 shows the relative risk of responding (θ₁) to not responding (θ₂). In this analysis, when an informative prior is used, an individual that has the CC genotype is, on average, 1.9 times more likely to go into remission at 4 weeks than to not go into remission (Figure 4). The HDI spans between 0.7 and 3.5, where the probability that the CC individuals have a higher chance of going into remission than not is 93.5%.

OPEN IN VIEWER

Figure 3.

Influence of prior and likelihood on posterior relative risk of not-responding to responding to citalopram. The histograms show the posterior relative risk of not responding (θ₁), relative to responding (θ₂), conditional upon genotype using a non-informative, uniform prior [beta(1,1)], and a prior that corresponds to the literature but with a reasonable degree of uncertainty [beta(4,4)] (first column). The first row gives the likelihoods of the not-responding group for each genotype. Black horizontal bars represent 95% highest-density intervals (95% chance that the true relative risk falls within this interval). Note the variable scale on the x axis.

OPEN IN VIEWER

Figure 4.

Influence of prior and likelihood on posterior relative risk of responding to not-responding to citalopram. The histograms show the posterior relative risk of responding (θ₁), relative to not responding (θ₂), conditional upon genotype using a non-informative, uniform prior (beta(1,1)), and an informative prior [beta(4,4)] (first column). The first row gives the likelihoods of the not-responding group for each genotype. Black horizontal bars represent 95% highest-density intervals (95% chance that the true relative risk falls within this interval). Note the variable scale on the x axis.

Response to citalopram in the oldest old is related to 5-HT_1A receptor genotype

Our study set out to explore an important question relating to the value of 5-HT_1A receptor genotypes in predicting response to citalopram in the oldest old. Given the nature of our target population, recruitment was suboptimal and our sample size is likely to be considered too small for any meaningful statistical analysis. In light of this, we wish to emphasize two key points: (a) our data are drawn from a difficult-to-obtain population and are likely to be of value to future studies; and (b) we propose that for studies such as these, a simple Bayesian analysis is more transparent in conveying the influence of the evidence (proportional to the sample size) on any one hypothesis.

The current study has shown that for a population of depressed older patients (aged > 80 years), genotype explains approximately 32% of the variation in response to citalopram. We were able to exclude age, ethnicity and sex as confounders, although we acknowledge that these may have small effects that we were unable to detect with our small sample size. Other factors, which we were unable to control for may also contribute to the observed changes. Interestingly, we showed that both homozygote groups (CC and GG) displayed a mean reduction in GDS over the study period, whilst the heterozygote group’s score increased, indicating a worsening of depression. The calculated effect size for genotype equates to approximately 0.37 (η²), which is considered large. The effect size of genotype on remission rates is also large (Cramer’s V = 0.61), compared with similar studies conducted in younger patients (Cramer’s V = 0.34), which raises the intriguing possibility that the effect of genotype is more prominent in this older population. It should be noted, however, that whilst it appears the results of the GC group reflect an absence of response at 4 weeks, it may equally be the case that the response is delayed. Conducting the study over a longer period could resolve this question, but may prove difficult due to retention issues.

This nonlinear relationship between the addition of a G allele on ΔGDS was not expected, yet this pattern was also observed in our analysis of genotype on MMSE following citalopram treatment (the cognition of homozygotes both improved slightly over the course of the study, whilst the heterozygote group showed a decline). Indeed, post hoc analysis revealed that the GC genotype behaves differently in response to citalopram compared with both homozygotes. Nonetheless, it should be noted that only the CC ΔGDS showed a statistically significant difference from 0 (p < 0.05, one-sample student’s t test). A similar observation was noted by Koto and colleagues, who showed that the rs6295 heterozygote has significantly lower response and remission rates at 2 weeks compared with both homozygotes. ²⁴

One possible reason for this observation could be an interaction between age and phenotype. Evidence suggests that both 5-HT_1A receptor and SERT expression decline with increasing age.^10,12 The effect of the latter would be to increase synaptic 5-HT, further reducing 5-HT_1A surface expression through desensitization (internalization and transcriptional repression). ²⁵ This may perhaps improve SSRI efficacy in CC individuals (as they would be susceptible to Deaf-1 repression), whilst having little effect on those carrying a G allele. Our results were consistent with our hypothesis in terms of CC individuals responding positively to citalopram, and GC individuals not so; but there is an obvious conflict in the positive response observed within our GG group. A possible explanation for this finding is summarized in Figure 5. Briefly, based on previous studies, we speculate that individuals with a GG genotype in the population fall into two categories: those who have a compensatory mechanism that restores 5-HT signalling, and those who do not. Those individuals who do not compensate will be more prone to depression, presenting with symptoms throughout life. ²⁶ The idea of a compensatory mechanism in this genotype has been postulated previously and may involve a reduction in the expression of SERT, and a restoration of 5-HT_1A expression to levels not dissimilar to other genotypes. It is these individuals that we believe may present late in life with depression, and, due to the downregulation of 5-HT_1A receptors may show an uncharacteristic (based on genotype) positive response to SSRI therapy. The situation may be complicated further by the fact that there are recognized sex differences in age-related changes to 5-HT_1A expression, with females showing less presynaptic receptor decline with age compared with men. ²⁷ This may potentially make aged men more sensitive to SSRIs than females. Indeed, all males in our study (n = 5) showed a reduction in GDS over the course of the study. At this stage, however, we must caution that this interpretation is speculative, but this offers a possible hypothesis that can be tested in future studies.

OPEN IN VIEWER

Figure 5.

Proposed mechanism to explain unexpected positive response of GG individuals to citalopram.

Despite demonstrating a relationship between 5-HT_1A genotype and SSRI efficacy in old age, we did not find a role for genotype in susceptibility to common adverse reactions associated with SSRIs. Neither a reduction in plasma Na⁺ or Hb concentration, or the presence of serotonin syndrome were significantly different between genotype populations, however, this does not necessarily rule out a relationship, as they may take a longer period to manifest than our data collection period allowed, and it may not have been possible to detect them because of our small sample size.

Bayesian analysis

So far in our discussions, we have considered data analysed from a frequentist’s viewpoint and asked the statistical question: what is the probability of observing our data if there were no underlying genotype effects. For data where p < 0.05, it is standard to consider the probability of observing our data by chance to be so small that it is more likely that there are real genotype effects. However, with an n = 19, it would be prudent to exercise some caution in drawing any firm conclusions. Furthermore, when making clinical decisions about which treatment should be initiated in an individual, this type of analysis offers little help. Instead, it may be more useful to ask the following question: what is probability of observing treatment failure in an individual, given a particular genotype? This approach is termed Bayesian forecasting and is arguably of more value to a clinician than frequentist data. Importantly for situations such as ours, the simple Bayesian approach we adopt at least offers a transparent indication of the contribution our evidence offers in support of any one hypothesis.

The Bayesian analysis we performed incorporated data from previous studies investigating response to citalopram in older patients as the prior. Limited as these data are for these difficult-to-obtain samples, our Bayesian forecasting indicates that the CG and CC genotypes may influence response to citalopram in different directions and highlights the need for more studies in this area. Importantly, by adopting the methodology presented here, further studies can readily incorporate our posterior distributions as future priors, thereby fully benefitting from the strength Bayesian analysis brings to studies of this nature.