Unmet needs in precision psychiatry

Abstract

A recent meta-analysis has suggested a 5-HTR1A promoter variant may predict antidepressant response. The present review comments on the claims made in view of sensitivity issues and issues pertaining to genetic exposure. We also alert to errors in the original data that had been carried over. Specifically, primers meant to amplify the HTR1A gene aligned to the BDNF gene sequence. Alleles had been confounded owing to DNA strand ambiguities and demographic information proved inaccurate. In the light of these findings, adherence to PRISMA guidelines and use of the Newcastle-Ottawa Scale did not safeguard against bias. More after action reviews are encouraged to identify factors likely to interfere with estimates of genetic risk in large data sets. These may result from pooling of ethnic groups, the use of binary data or other formats that are not human-readable, the introduction of surrogate identifiers and a failure to reverse-engineer previously published experimental protocols. Unless the above challenges are met, sequence variants are unlikely to inform personalized medicine strategies in psychiatry.

Key Words: Major depressive disorder; Antidepressant efficacy; Meta-analysis; 5-HTR1A gene; C-1019G polymorphism

Core Tip: A recent meta-analysis has suggested a 5-HTR1A promoter variant may predict antidepressant response. However, re-analysis of the original data unveiled numerous shortcomings pertaining to genetic exposure and interpretation of genetic findings. We alert to ambiguities that are likely to have introduced bias in the calculation of overall effects by muddling of alleles or entire genes, and by demographic inaccuracies. In part, these may be explained by specific features of genetic information that are not encoded in the DNA sequence, such as transcriptional orientation.

TO THE EDITOR

We read with interest the recent article by Wu et al[1] on the role of the 5-HTR1A gene in predicting the response to antidepressants. From a voluminous literature, and amidst a snowstorm of conflicting data, the authors have sought to distill the overall effects and thereby enhance the robustness of pharmacogenetic predictions. Upon second glance, however, the conclusions warrant further scrutiny. Reference is made to PRISMA and the Newcastle-Ottawa Scale (NOS), but the actual genotype counts obtained from pooling primary research are not listed. The data shown indicate discrepancies for: (1) The number of cases and controls in 5/11 studies[2-6]; (2) Male and female gender in 4/11 studies[2-4,6]; (3) Diagnostic criteria in 4/11 studies[2,4,7,8]; and (4) Demographic information in 3/11 studies[6,9,10].

From a clinical point of view, an approach that indiscriminately pools all currently available antidepressants and multiple definitions of response to these agents, is already at odds with precision medicine. Spurious variability is further enhanced by multicentre recruitment[5], a 6-fold difference in duration of treatment[3,11], ethnic admixture[7] and naturalistic settings[11]. The inclusion/exclusion of lithium and other active co-medications in a subset of populations is also liable to have impacted on effect size. Numerous eligible investigations were excluded[12-23] reducing statistical power and lowering sensitivity of the meta-analysis (N_examinedvs N_total = 1436/3359).

From a basic science point of view, roughly 15% of human single nucleotide polymorphisms, including C-1019G, are population-specific, and require separate analyses for different ethnic groups. The study by Wu et al[1] did not conduct a separate analysis despite a bimodal allelic distribution in the total sample (781 Asian participants of whom 25% carry the C allele vs 635 Caucasian participants of whom 50% carry the C allele). Moreover, at least one study never even addressed the 5-HTR1A promoter variant of interest[2]. Instead of rs6295 on chromosome 5, the authors appear to have assessed an alternative polymorphism, rs6265 in the BDNF gene on chromosome 11. Wu et al[11] are invited to consult the methods section for details and may consider reranking this publication, rated 9/9 on the NOS.

Which brings us to measures of bias. Unlike ratings of standard clinical trials, the assessment of genetic association analyses calls for a thorough understanding of potential pitfalls specific to DNA code, to the selection, arrangement and operation of alleles, the modes of inheritance etc. To meet some of these additional requirements, an adaptation of the NOS to genetic studies has been proposed[24]. Still, many key issues remain unresolved, e.g., a minimum requirement is to ensure the allele frequencies obtained match those of reference populations. When we apply this criterion to studies by Chang et al[3] or by Kato et al[11], that both served for the above meta-analysis, we note major inconsistencies. In sharp contrast to the remaining studies[10], and to other Asian reference populations in public databases, the reported C allele frequency is in the range of 0.75, rather than 0.25. Allele frequencies that fail to align with expected values strongly suggest the presence of bias. It is worth noting that multiple alerts have been published following incorrect readout of 5-HTR1A alleles[25-27]. These artefacts persist in 3/10 of current 5-HTR1A genotyping protocols[3,6,11] and have been missed (Figure 1).

Figure 1 Genomic organization of 5-HTR1A and neighbouring genes. Unlike LLRC70 and RNF180, the 5-HTR1A gene is transcribed on the chromosome 5 reverse strand (in black). Depending on the template used for primer design, some experimental protocols have mistakenly amplified the forward DNA strand (in blue). Restriction patterns then suggest C-1019 is G-1019.

CONCLUSION

In summary, the present meta-analysis illustrates the need to ascertain exposure, to narrow the phenotype, and to take into account additional confounders. We propose that the cumulative effects of C-1019G on response to antidepressants be reassessed. Subgroup analyses by ethnicity are advocated, as are standardized antidepressant response criteria across studies and formal power calculations to assess sample sufficiency.