This paper 10.1016/j.neuron.2007.11.017 has to be de-emphasized on this and other (rs2235067, rs4148740, rs2032583, rs4148739, rs11983225, rs2235040, rs12720067, rs7787082, rs102484207) SNP pages for several reasons. One of them - statistical dirty tricks:
- Out of 9 SNPs, the authors chose only two to illustrate the effects, and they give no response frequency numbers for other SNPs.
- Further, out of three possible endpoints (4 weeks, 5 weeks, 6 weeks) the authors arbitrarily chose one - 4 weeks. When you look into the supplement, the differences at 5 and 6 weeks are not significant or less significant.
- most of studies by groups other than Uhr,conclude that ABCB1 has little effect on response to antidepressants. The later review,(2016- ABCB1 genotyping in the treatment of depression, one of the authors is Uhr) admits, "The clear majority of studies examining the relationship between ABCB1 and antidepressant response have focused on three other SNPs that lie in the coding region of ABCB1 (rs1128503, rs2032582 and rs1045642), but did not have a highly significant effect in the Uhr et al. study  or in any other study that could confirm the effect of one of the intronic ABCB1 SNPs [46,61].)
A recent meta analysis [PMID 25847751] failed to reproduce Uhr et al (10.1016/j.neuron.2007.11.017) results. We have to change the summaries of the involved SNPs to NO effect of ABCB1 on antidepressant response/remission.
Major importance of [PMID 25847751] is that they obtained unpublished results of a huge (900 subjects), well-run, double-blind study of a well-researched ABCB1 substrate citalopram in depression (STAR-D study). These negative results dramatically change the overall outcome of the meta-analysis.
- rs2032583. A significant association of antidepressant treatment outcome among all studies with this SNP becomes insignificant after Bonferroni correction for multiple comparisons. The result is nominally positive for a subgroup of inpatients, but it is mainly driven by Uhr et al study, which is an outlier among the rest (figure 2 in the meta-analysis). After excluding Uhr and applying Bonferroni correction, the association is insignificant.
- rs2235015. No significant association of antidepressant treatment outcome with this SNP among all studies. The result is significant in inpatients. (Again, I believe due to the problematic Uhr study).
- rs2235040. The same story. No significant association of antidepressant treatment outcome with this SNP among all studies. "There was a significant association with treatment outcome in the inpatients subsample but data regarding rs2235040 and treatment outcome among inpatients entirely stemmed from the study by Uhr et al."
- rs1045642. No significant association.
- rs2032582. No significant association overall. A nominally significant result in the meta-analysis stratified for co-medication in a subsample.
- rs1128503. No significant association
Further problems with Uhr et al study 10.1016/j.neuron.2007.11.017: quality of the study, data manipulation: unexplained exclusion of patients, unexplained change of the remission threshold , internal inconsistency.
- Quality of the study: Uhr et al used the data from the Munich Antidepressant Response Signature (MARS) study - an open label study of hospital inpatients [PMID 18586274]. Open label studies are generally deprecated in psychiatry because the expectations of the patients and doctors have a significant influence on the outcome. To the contrary, the main study with negative genetic association results - the StAR-D study - was double blind, and a study of outpatients.
- Unexplained exclusion of subjects: the main text of the Uhr paper claims "443 inpatients with depressive disorder receiving antidepressants". However, their analysis of genetic association for rs2032583 and rs2235015 was limited to 228(or 231) patients (figure 3B-F and Supplements). Why did they exclude 50% of the subjects? - there is no explanation. Data manipulation is likely the reason.
- Unexplained change of the outcome definition: Traditionally, in psychiatry remission is defined as a score of less than 7 on the 17-item HAM-D. This was the definition in the MARS study [PMID 18586274] whose data Uhr used. However, in their paper, Uhr et al used as threshold "achieving a remission with a score of <10 on the HAM-D-scale." Why? - probably because they could get better results at this threshold.
- Internal contradiction. According to HapMap data of Caucasian ethnicities, SNPs rs2032583,rs2235015, rs2235040, rs2032582, rs1128503 are in strong LD (D' = 1.0).[PMID 25847751] Uhr at al obtained positive association of remission with rs2032583, rs2235015, and rs2235040 but no association with rs2032582 and rs1128503 (see Supplement). This makes no sense because these SNPs are strongly dependent on each other. Positive association with one means positive associations with the others. This contradiction casts a very strong doubt on validity of all Uhr's results.