This study demonstrates the influence of genetic variation in the OPRM1 gene on antidepressant response and remission from major depressive disorder during citalopram treatment. The most statistically significant finding was for a single SNP for the specific response phenotype, which met a correction threshold for the number of markers tested. This SNP, rs540825, was predicted to lead to the exchange of a glutamine residue for a histidine residue in the carboxyl terminal region of the alternatively spliced μ-opioid receptor-1X transcript. Given the correlation between markers in the region, we cannot rule out variation in the final exon of the μ-opioid receptor-1 transcript or in the regions flanking these exons. All of the SNPs in reside within a 44-kb region encompassing the final exons of the μ-opioid receptor-1 and -1X splice variants. Interestingly, we did not detect significant findings for the most commonly studied OPRM1 SNP, the nonsynonymous rs1799971, for any of the analyzed phenotypes. For the most associated SNP in the ancestry-corrected principal components analysis model, rs540825, we were not able to detect an effect in either the African American or Hispanic Caucasian subjects, but there was evidence of association in the Non-Hispanic Caucasian subjects. This may have been the result of lack of power resulting from the smaller sample size of these ethnic groups in our study sample. However, the race interaction term was significant, indicating a difference between the ancestral groups.
Association with the OPRM1
gene was nominal in our genome-wide association of citalopram response, using the same sample. Although the rs540825 SNP was not directly genotyped on the arrays used, several SNPs in modest linkage disequilibrium with rs540825 showed p values ranging from 0.02 to 0.06 for response and remission phenotypes (26
). This highlights the potential utility of targeted genotyping at higher density in genes of interest or potentially for whole-genome sequencing. In the context of studying an entire genome of 2.8 billion base pairs, the findings in our study would not meet genome-wide significance for association (i.e., p<1.76×10−11
), but such an extremely conservative statistical view has not been generally adopted in association studies such as this.
Association studies are plagued with conflicting results for various reasons, the main one being phenotypic heterogeneity. This may come about as a result of differential syndromal definitions, varying instrumentation used for assessment, and ethnic classification between studies. Our efforts to decrease this common problem in our study consisted of treatment in one condition only (citalopram), a rigorous phenotype of response to citalopram, and correction for population stratification. The results were strengthened by identification of responders who were more likely to be true drug responders; that is, specific responders.
In vitro experiments comparing the nonsynonymous-associated SNP rs540825 showed that both the A and T alleles encode functional opioid receptors but failed to reveal any significant difference between alleles in receptor signaling or regulation. However we cannot exclude functional distinctions between alleles that might occur in physiologically relevant neurons but not in the heterologous cell model examined, since HEK-293 cells do not fully recapitulate the cellular environment of neural tissue. This possibility may merit investigation in future studies. Although rs540825 is an appealing candidate for functional studies, it is also possible that it is in linkage disequilibrium with a nontyped variant that may be functionally relevant.
The odds ratios observed for the rs540825 variant ranged from 1.4 to 1.6 for general response, remission, and specific response. We can compare these odds ratios with those seen in other candidate gene studies using the STAR*D sample, namely for the HTR2A
). For the HTR2A
SNP, we found an odds ratio of 1.52 for remission and 1.43 for response (27
). We analyzed our data for interaction between the previously reported HTR2A
SNP (rs7997012) and our OPRM1
SNPs. No significant interaction was observed.
In an attempt to analyze our data in the manner of previous reports, we removed from analysis individuals who, on the Quick Inventory of Depressive Symptomatology–Self Report, had a 41%–49% reduction in scores from baseline to the last visit or a final score between 6 and 9 for our response and remission phenotypes, respectively. This removed subjects whose response phenotypes may have been equivocal. The odds ratios remained constant, and, as such, provide further support for our findings.
We detected several SNPs to have statistically significant race interaction terms (see Table 8 in the online data supplement
). Moreover, many of these SNPs showed association with the analyzed phenotypes when tested using principal components analysis-corrected models. Therefore, it is important to note that the association was potentially driven by a subset of the tested sample. These results are reported in Table 9 of the online data supplement
, in which all SNPs listed in of this article are further detailed for race interaction term as well as p values for principal components and race-stratified analyses. These results are further dissected in Tables 2–4 of the online data supplement
, which report all of the p values for principal components analysis-corrected models as well as the findings when each ethnicity was tested separately. These data allow for a more complete understanding of the association values, specifically the subsample that was potentially driving the signal detected in the principal components analysis-corrected model.
Our p value threshold was determined using SNP spectral decomposition, which estimates the number of collin-ear tests actually being performed. Thus, our initial analyses for 48 SNPs decreased to 23.4 tests and resulted in a Bonferroni-corrected p value of 0.0007. This was a conservative correction for the three phenotypes as a result of the nested nature of these phenotypes; namely, that the specific response and remission groups were subsamples of the responder group.
This genetic association with the specific response phenotype might suggest a likely association with the pharmacologic action of citalopram. However, an alternative possibility is that this association actually represents an association with placebo response or the nonspecific response that can occur even with an active antidepressant medication. This possibility should be addressed in studies of OPRM1
variation in placebo-controlled trials. Previous studies showing that placebo-induced analgesia is associated with activation of the endogenous opioid system might suggest the latter interpretation (28
). Although our study did not include subjects taking placebo, we observed that the strongest association findings were in the specific response phenotype, which was based on attempts to clinically differentiate placebo responders from drug responders.
The STAR*D trial was a large clinical trial designed to assess the clinical utility and effectiveness of antidepressant response with different treatment alternatives for major depressive disorder. Although it has proven useful for obtaining large numbers of subjects with the same clinical disorder and treated with the same medication (in level 1), it was not well designed to control for all possible confounding factors in genetic studies.
In addition to the lack of a placebo, there are several limitations to our study design, which have been discussed previously (20
). First, there was a lack of information about other medications the participants were taking aside from citalopram. It is possible that responders were augmenting with an unknown medication and artificially inflating the response rate, thus masking the true effects of citalopram response, or that nonresponders were taking concomitant medications and experiencing drug-drug interactions, leading to presumed intolerance to citalopram. It is also possible that subjects were taking μ-opioid receptor agonists or antagonists, which may have interfered with the genotype-driven effect that we observed. We were able to assess concomitant narcotic usage and provide evidence that the SNPs associated with antidepressant response were not confounded by concomitant narcotic usage. Although this does not rule out other possible concomitant medications influencing the response rates, it does reflect the drugs most likely to play a role with the μ-opioid receptor (i.e., narcotics). Second, there was incomplete data on adherence to citalopram treatment in level 1 of the STAR*D study. Adherence was only estimated by the treating clinician for participants who exited the study at level 1 (N=208) and was not measured in those participants who entered follow-up evaluation or the next level of the trial. Thus, we cannot be sure that the nonresponders took their medication as diligently as the responders, thereby artificially inflating the response rate for the responders.
Third, DNA samples were not a requirement. DNA collection began well into the STAR*D trial, but not all participants agreed to provide blood samples. The key differences between the study participants who did and did not provide blood samples have been previously described (18
) and include ethnicity, marital status, age, education level, primary versus specialty care, recurrent major depressive disorder, and number and length of major depressive episodes. Although these differences exist, the role they play in contributing to differential response rates is unclear. However, it is notable that we may not be able to generalize our findings to the entire STAR*D study population.
Fourth, medication dosage in the STAR*D study population was designed to model what would typically be prescribed in a clinician’s office, and thus there was not a fixed dosage for all study participants. However, the variability in dosage did not have an effect on response rates, since the responders and nonresponders were receiving 30.42 mg and 29.88 mg of citalopram, respectively, at their final visit.
Another limitation is that drug intolerance was a potential confound for our response phenotypes. We found that 97% of responders were tolerant to citalopram, while only 80% of nonresponders were tolerant to the drug (p<0.0001). Similar findings were observed for the specific response and remission phenotypes. It is possible, then, that the observed genetic associations were actually with a tolerance phenotype, which itself may have determined the response phenotype. However, we do not believe this is the case, since none of the SNPs were associated with a tolerance phenotype.
Although the STAR*D study was designed to assess clinical effectiveness of medications, we cannot suggest that our data contain useful information for driving clinical decision making as of yet because of our lack of replication, assessment of only one antidepressant (citalopram), and modest size of association. These data do provide insights into mechanisms of antidepressant response and pave new roads for investigations of pharmacotherapy.
In summary, we reported significant findings for association between variants in the OPRM1 region and antidepressant response and remission from major depressive disorder. The most significantly associated SNP was in an exon of an alternatively spliced isoform of OPRM1, which we showed to be present and functional at the plasma membrane. These data suggest that response to citalopram therapy is influenced by the OPRM1 gene and endogenous opioidergic neurotransmission may be involved in major depressive disorder and its response to antidepressant pharmacotherapy.