In this study, the BD group had significantly lower concentrations of LA and AA based on the analysis that took into account the QTL effect of SNPs. This indicates the independent diagnostic effect of specific PUFAs on BD susceptibility. In this study, we replicated the QTL effect on the concentrations of AA and GLA by the FADS SNP (rs174550), which was reported in various ancestries (Table 3) [15, 16, 17], although previous studies have shown that SNPs were associated with the concentrations of most PUFAs consisting of n-3 and n-6.
Table 3
Comparison of QTL analysis results for rs174550 between this study and previous studies
PUFA/study | This study N = 645 | Dorajoo et al. 2015 N = 1361 | Guan et al. 2014 N = 8631 |
β (%) | p | β (%) | p | β (%) | p |
LA | NS | NS | 0.048 | 1.5 × 10− 16 | 1.47 | 4.37 × 10− 274 |
GLA | −0.079 | < 2 × 10− 16 | −0.52 | 9.4 × 10− 171 | −0.02 | 5.95 × 10− 72 |
DGLA | NS | NS | -0.18 | 1.2 × 10− 29 | 0.36 | 7.44 × 10− 152 |
AA | −0.36 | < 2 × 10− 16 | −0.15 | 2.1 × 10− 66 | −1.69 | 0 |
ALA | NS | NS | 0.38 | 1.1 × 10− 16 * | NA | NA |
EPA | NS | NS | −0.045 | 2.2 × 10− 16 ** | NA | NA |
DHA | NS | NS | NS | NS | NA | NA |
N: The number of samples, β: Effect of C-allele on PUFA concentration ratio in rs174550(T > C Effect allele: C-allele), BD: bipolar disorder, LA: linoleic acid, GLA: γ-linoleic acid, DGLA: |
dihomo-γ-linolenic acid, AA: arachidonic acid, ALA: α-linoleic acid, EPA: eicosapentaenoic |
acid, DHA: docosahexaenoic acid, NS = no significant difference, NA = not applicable |
* rs174547 data (rs174547 is in linkage disequilibrium with rs174550, R2 = 1.00) |
** rs174538 data (rs174538 is in linkage disequilibrium with rs174550, R2 = 0.95) |
Significant p-values are indicated with bold cases |
In the previous GWAS targeting BD [13], the C-allele of rs174550 was the risk allele for BD (OR = 1.17, p = 1.2 × 10− 8). This finding supported the result that low GLA and AA concentrations in individuals harboring the C-allele were associated with BD, although other PUFAs, such as DHA and EPA, were not risk factors for BD. This also clarified the finding that BD risk by low AA concentration was explained both by the QTL effect of the SNP (decreased AA by genotype) and the diagnostic effect, which were independent of each other (AA, β = 0.0703, p = 0.451; Fig. 1). By contrast, the low LA concentration was not influenced by the QTL effect but is mainly explained by the diagnostic effect. Such “non-significant” QTL effect is reasonable because LA concentration is mainly determined by diet (i.e., LA cannot be synthesized in the body), is in the upstream of the n-6 pathway, and is metabolized into GLA (significant QTL effect), DGLA, and AA (significant QTL effect) by FADS.
Compared with previous lipidomics studies of BD (Table 4), the sample size in the present study is one of the largest and thus provided the most powerful results. However, we could not obtain concordant and conclusive results similar to other studies because the sample size was not large enough to deny type I and II errors. In addition, previous studies did not consider the QTL effect of the FADS SNP, which is also a susceptible SNP to BD. Such a small sample size induces variable allele frequencies either in cases or controls, which may influence the QTL effect on PUFAs. Therefore, the QTL effect is one of the most important confounding factors when interpreting the lipidomics for BD.
Table 4
Comparison of the results of this study with previously reported lipidomics results
Study | Chiu et al, 2003 | Sublette et al,2007 | Clayton et al,2008 | McNamara et al,2010 | Evans et al,2013 | Pomponi et al,2013 | Saunders et al,2015 | Koga et al,2019 | This study |
Sample | Erythrocyte | Plasma | Erythrocyte | Erythrocyte | Plasma | Plasma | Plasma | Plasma | Plasma |
Unit | % | nmol/mL | mg/day | % | nmol/mL | nmol/mL | nmol/mL | µg/mL | % |
Mood state (N) | manic (20) control (20) | manic (10) control (10) | BD (15) control (15) | BD (20) MDD (20) control (20) | BD (40) control (18) | depressed (21) manic (9) euthymic (12) control (57) | BD (27) control (31) | BD (83) control (217) | BD (535) control (107) |
PUFA | LA | NS | NA | NA | NS | NS | ↑ | NA | ↑ | ↓ |
GLA | NS | NA | NA | NA | NS | NA | NA | ↑ | NS |
DGLA | NA | NA | NA | NS | ↑ | NA | NA | ↑ | NS |
AA | ↓ | NS | NA | NS | NS | ↑ | NS | ↑ | ↓ |
ALA | NA | NA | NA | NA | NS | ↑ | NS | ↑ | NS |
EPA | NS | NS | NS | NS | NS | ↑ | NS | ↓ | NS |
DHA | ↓ | NS | NS | ↓ | NS | ↓ | NS | ↓ | NS |
N: number, NS: no significant difference, NA: not applicable, ↑/↓: Each PUFA concentration increases/decreases in the BD group compared to the control group |
The strength of this study is attributed to the largest sample size, and the “abnormality” of AA concentration is a good reason to explore the causal effect on BD susceptibility because the QTL effect was concordant with those in other QTL studies [15, 16, 17]. Because QTLs in other PUFAs, except GLA and AA, were not replicated in this study, we speculated the following reasons: (1) the sample size was small to detect the QTL effect and (2) previous QTL results showed that the direction of the effect on DGLA in rs174550 was contrasting between Caucasians and Han Chinese, which suggest population difference in the effect of rs174550 on DGLA. However, as mentioned above, we replicated the QTL effect of the SNP on AA and GLA, concordant with other large QTL analyses, and thus concluded concrete results at least for AA.
We believe that AA is important in elucidating the pathogenesis of BD because several associations between AA metabolism and BD were reported so far [18]. According to reports, mood stabilizers such as lithium, valproate, carbamazepine, and lamotrigine affected AA metabolism in the brain via N-methyl-D-aspartate receptors and D2 receptors, which the authors called the “arachidonic cascade hypothesis.” In this hypothesis, plasma AA is consumed in BD because of increased AA metabolism in the brain, and mood stabilizers decrease the turnover of AA in several brain phospholipids. Even in accordance with this hypothesis, as our study revealed compatible results that the plasma AA concentrations decreased, it supported the finding that impaired AA metabolism affects the pathogenesis of BD. Therefore, further studies investigating AA metabolism in the brain are essential and may elucidate the pathogenesis of BD.
To interpret this result, we stress important limitations. We could not control the effect of diet. Although we covariated the time from the last diet in the model and applied the ratio (e.g., AA/all PUFAs) to control the diet, we could not follow the contents of the diet, and such is associated with PUFAs located in the upstream of this pathway (such as LA).
In conclusion, the BD group has significantly lower AA and LA concentration ratios than the control group. In addition, a strong QTL effect of the FADS SNP was found on AA and GLA; however, such an effect was independent of BD susceptibility. Therefore, AA and LA are good candidates for susceptibility to BD, and such QTL effect of the SNPs should be covariated to interpret the lipidomics for BD.