Fasting state assessment
Analysis of opportunistically collected samples brings a challenge of the unknown fasting state. The control cohort contained samples collected in the fasted state per ADRC and EHBS protocols, but the AD cohort included many who had no collected fasting state information and consist of samples collected in both fasted and non-fasted states. Therefore, to allow a direct comparison of the control and AD groups, we assessed the estimated subject fasting state using our previously published predictive model (20). As expected, control group was predicted to contain mostly fasted subjects (Figure S1). Out of 133 control subjects, 105 (i.e. 79%) were predicted as fasted, while 17 (i.e. 12%) as non-fasted, with a probability of > 60% and 11 (i.e. 8%) had a fasting state probability of < 60%. Out of 148 AD subjects, 60 (i.e. 40%) were predicted as fasted and 81 (i.e. 55%) as non-fasted, with a probability of > 60% and 7 (5%) had a fasting state probability of < 60%. Fifty percent of detected metabolites manifested difference between predicted fasted and non-fasted AD subjects in plasma and only minimal differences were observed in CSF (Table S2).
Cytochrome P450/soluble epoxide hydrolase metabolism is elevated in AD subjects
We compared plasma and CSF lipid mediator concentrations between the control and the AD groups, using only estimated fasted subjects with probability > 60% for plasma. In plasma, we detected 42 oxylipins (85 measured), 5 PUFAs (5 measured), 17 endocannabinoids (22 measured), 3 NSAIDs (4 measured), 19 bile acids (23 measured) and 8 steroid hormones (8 measured). The mean values and p-values for t-tests and two-way ANOVA interactions for all detected metabolites are provided in Table S3. Plasma group-fold differences in the oxylipin, endocannabinoids and PUFAs, projected onto their metabolic pathway, are presented in Fig. 1. The largest differences were observed in the long chain omega-3 PUFA metabolism. Both EPA and DHA enzyme derived mono-alcohols (5-LOX-derived 5-HEPE and 4-HDoHE and 12-LOX-derived 12-HEPE and 14-HDoHE) were lower (1.5-fold in average) in the AD group, when compared to the control. On the other hand, the sEH EPA metabolite 17,18-DiHETE, was 3-fold higher in the AD group. In the AD group, the AA pathway manifested lower levels of the COX derived prostaglandins PGF2α and PGD2 (1.6-fold average). Additionally, the AD group showed lower levels of acylethanolamides (1.5-fold in average) derived from dihomo-gamma-linolenic acid (DGLEA), AA (AEA), docosatetraenoic acid (DEA), DHA (DHEA) and oleic acid (OEA). Notable are also lower levels of autooxidation markers, particularly the EPA-derived 9-HEPE (2-fold), linoleic acid (LA) -derived TriHOMEs (1.65-fold) and AA–derived isoprostanes (1.3-fold) in AD group.
Fewer lipid mediators were detected in CSF than in plasma. Detected CSF lipid mediators included 17 oxylipins, 5 PUFAs, 3 endocannabinoids, 14 bile acids and 6 steroids. The mean values and p-values for ttests and two-way ANOVA interactions are provided in Table S4. CSF significant group-fold differences in the level of oxylipin, endocannabinoids and PUFAs, projected onto their metabolic pathway, are presented in Fig. 2. In this matrix, the largest differences were observed in the LA CYP metabolic pathway, where both epoxy and dihydroxy FA, products of CYP and subsequent sEH metabolism, were higher in the AD group when compared to the control: epoxide average 1.5-fold; diol average 1.3-fold. All PUFAs from both omega-3 and omega-6 pathway were lower in the AD group, although the difference was only 1.2-fold on average. Additionally, AD group manifested 1.5-folds lower level of OEA and 1.3-fold lower level of the EPA-derived 14,15-DiHETE.
While few differences were observed in plasma and CSF bile acid levels between control and AD subjects, numerous differences were present in the specific bile acid ratios (Table 1). Figure S2 shows bile acids metabolic pathway together with their median plasma levels to help understand the biological aspects of specific bile acid ratios. In plasma, the AD group was characterized by lower levels of cholic acid (CA), a product of the neutral bile acids synthesis pathway, while chenodeoxycholic acid (CDCA), a product of the acidic pathway was unchanged. This difference becomes even more pronounced when looking at the CA/CDCA ratio. On the other hand, the difference between neutral and acidic pathway was not present in downstream metabolites, when comparing the secondary unconjugated bile acids ratio, like deoxycholic acid/(lithocholic acid + ursodeoxycholic acid) (DCA/(LCA + UDCA)) or the most abundant primary conjugated derivatives glycocholic acid/glycochenodeoxycholic acid (GCA/GCDCA); Table S3. Of note, small differences between the neutral and acidic pathway were observed in the low abundance taurine conjugates of the secondary bile acids taurodeoxycholic/taurolithocholic acid (TDCA/TLCA). Difference in conjugation ratio (more conjugates than the substrate) was observed in the neutral synthesis pathway (GDCA/DCA and GCA/CA) but not in the acidic synthesis pathway. Differences between the neutral and acidic synthesis pathway were also observed in the conversion of the primary to secondary bile acids. The ratio of downstream products to their precursor in the neutral pathway was higher in the AD group in the case of DCA/CA, TDCA/CA and GDCA/CA, but not in parallel acidic pathway metabolites (i.e. LCA/CDCA, UDCA/CDCA).
Table 1
Differences in bile acids metabolites and their specific rations between control and AD group. Means are expressed in nM or as a ratio of the concentrations. Metabolites and their ratio are stratified by the metabolic affiliations. All tested bile acids and their ratios are presented in the Table S3.
Metabolite | P value | Mean [95%CI] |
---|
Control | AD |
---|
Neutral vs acidic synthesis pathway |
CA | 0.0321 | 27.4 [20.8–36.2] | 19.3 [13.6–27.4] |
CDCA | 0.978 | 38.7 [29.2–51.5] | 46.7 [32.6–66.9] |
CA/CDCA | 0.0008 | 0.707 [0.571–0.876] | 0.363 [0.271–0.487] |
GCDCA/GDCA | 0.5 | 2.08 [1.66–2.61] | 1.97 [1.51–2.57] |
TDCA/TLCA | 0.0114 | 1.33 [1.1–1.62] | 1.81 [1.49–2.21] |
Conjugation, neutral synthesis pathway |
GDCA/DCA | 0.0031 | 0.719 [0.604–0.856] | 1.05 [0.836–1.33] |
TDCA/DCA | 0.1 | 0.0605 [0.0467–0.0783] | 0.0777 [0.0556–0.109] |
GCA/CA | 0.0104 | 2.52 [1.86–3.39] | 4.13 [2.84–6.01] |
TCA/CA | 0.076 | 0.506 [0.358–0.716] | 0.711 [0.434–1.16] |
Conjugation, acidic synthesis pathway |
GUDCA/UDCA | 0.7 | 0.546 [0.341–0.876] | 0.515 [0.334–0.793] |
TUDCA/UDCA | 0.746 | 2.83 [1.93–4.17] | 2.04 [1.33–3.13] |
GCDCA/CDCA | 0.491 | 8.15 [6.21–10.7] | 8.36 [5.68–12.3] |
TCDCA/CDCA | 0.644 | 0.734 [0.53–1.01] | 0.687 [0.433–1.09] |
TLCA/LCA | 0.961 | 0.282 [0.2–0.396] | 0.252 [0.182–0.347] |
GLCA/LCA | 0.721 | 0.711 [0.53–0.955] | 0.749 [0.552–1.02] |
Conversion of primary to secondary. Gut metagenome activity |
DCA/CA | 0.0459 | 8.01 [5.8–11.1] | 12 [7.89–18.1] |
TDCA/CA | 0.0056 | 0.469 [0.309–0.711] | 0.906 [0.517–1.59] |
GDCA/CA | 0.0007 | 5.76 [4.01–8.27] | 12.5 [7.71–20.2] |
LCA/CDCA | 0.451 | 0.66 [0.419–1.04] | 0.589 [0.42–0.825] |
UDCA/CDCA | 0.691 | 1.39 [0.815–2.36] | 1.94 [1.23–3.07] |
TUDCA | 0.0275 | 1.07 [0.78–1.48] | 1.92 [1.53–2.41] |
CSF manifested few differences in bile acids and their ratios. The AD group had 1.3-fold higher levels of GLCA and 1.4-fold higher level of T-a-MCA. Additionally, AD group had lower ratio of GCDCA/GLCA (1.3-folds, Table S4).
Of those measured, only a few steroid hormones showed different levels between AD and the control. In plasma, dehydroepiandrosterone sulfate (DHEAS) and progesterone were lower in AD group (1.9 and 1.7-folds respectively). Additionally, testosterone and the testosterone/progesterone ratio showed significant gender x group interaction. Females AD subjects showed 1.4-fold lower testosterone, when compared to females controls, but no differences were observed in males. On the other hand, the testosterone/progesterone ratio was 2-fold higher in AD male subjects compared male controls. Testosterone/progesterone ratio differences were not observed in females.
In CSF, only corticosterone showed a significant difference between AD and the control group, however, the magnitude of the-fold difference only ~ 1.1.
Relation between CSF and plasma AD markers.
In the current study, matched plasma and CSF samples were collected, allowing an assessment of the relationships between metabolites in these pools. Spearman’s ρ rank order correlation between plasma and CSF lipid mediator levels are shown in Table 2. The associations were distinct by metabolite classes, with oxylipins showing only 2 of 15 significant correlations, while bile acids and steroids showing 14 of 18 significant correlations. Correlations within PUFA and PUFA ethanolamide were also apparent for the long chain omega 3 species (DHA EPA and DHA ethanolamide) but not others.
Table 2
Spearman’s rank order correlation between plasma and CSF metabolites. Significant p values are bolded.
Metabolite class | Metabolite | Spearman's ρ | P value |
---|
Oxylipins | PGF2a | -0.2 | 0.0746 |
F2-IsoP | 0.34 | 0.0019 |
12_13-DiHOME | 0.26 | 0.0223 |
9_10-DiHOME | 0.15 | 0.1810 |
15_16-DiHODE | 0.21 | 0.0625 |
14_15-DiHETrE | 0.062 | 0.5830 |
11_12-DiHETrE | 0.11 | 0.3340 |
17_18-DiHETE | 0.33 | 0.0027 |
19_20-DiHDoPE | 0.16 | 0.1450 |
13-HODE | 0.029 | 0.7970 |
9-HODE | -0.054 | 0.6380 |
13-HOTE | -0.014 | 0.9010 |
9-HOTE | -0.051 | 0.6520 |
12(13)-EpOME | 0.0089 | 0.9380 |
9(10)-EpOME | -0.028 | 0.8070 |
Acyl-EA | OEA | -0.031 | 0.7820 |
LEA | 0.062 | 0.5860 |
DHEA | 0.51 | 0.0001 |
PUFA | LA | 0.15 | 0.1860 |
ALA | 0.2 | 0.0826 |
AA | -0.049 | 0.6670 |
EPA | 0.45 | 0.0001 |
DHA | 0.42 | 0.0001 |
Bile acids | CDCA | 0.51 | 0.0001 |
UDCA | 0.78 | 0.0001 |
DCA | 0.71 | 0.0001 |
TCA | 0.62 | 0.0001 |
TCDCA | 0.43 | 0.0001 |
TUDCA | 0.16 | 0.1570 |
TDCA | 0.54 | 0.0001 |
GCA | 0.36 | 0.0012 |
GCDCA | 0.19 | 0.0925 |
GUDCA | 0.47 | 0.0001 |
GDCA | 0.54 | 0.0001 |
GLCA | -0.083 | 0.4670 |
Steroids | 17OH-PROG | 0.58 | 0.0001 |
Cortisol | 0.3 | 0.0071 |
Cortexolone | 0.089 | 0.4340 |
corticosterone | 0.51 | 0.0001 |
Testosterone | 0.81 | 0.0001 |
Next, we used partial least square discriminant analysis (PLS-DA) to illustrate the relationship between plasma and CSF AD markers (Fig. 3). The analysis showed that discrimination between control and AD was dominated by the plasma metabolites. Fifteen plasma metabolites (and their ratios) manifested variable importance in projection (VIP) score > 1.4 compared to only 4 CSF metabolites. The discrimination between AD and the control group was characterized by higher plasma 17,18-DiHETE (VIP = 2.16) and CSF EpOMEs (VIP = 1.95 and 1.58 for the 12(13) and 9(10) isoforms respectively) and lower levels of the acylethanolamide ratios including both DHEA/LEA and DEA/LEA in plasma and both plasma and CSF OEA/LEA. Plasma and CSF OEA/LEA manifest similar discriminatory power based on their proximity on the loading plot. On the other hand, plasma 17,18-DiHETE and CSF EpOMEs occupied distinct parts of the loading plot, suggesting distinct discriminatory properties. The VIPs for each metabolite are provided in the Table S5.
Fatty acid ethanolamides and CYP/sEH metabolites are strong AD predictors in both plasma and CSF. We used predictive modeling to investigate how well plasma and CSF metabolites can report AD status. Plasma lipid mediators generated stronger models than those in CSF with area under the receiver operator characteristic curves (ROC AUC) of 0.924 vs. 0.824, with the two models consisting of distinct metabolites (Fig. 4). However, in both matrices, the strongest predictors belonged to the same two metabolic pathways, the acyl ethanolamides and CYP/sEH pathway. Plasma predictors included ethanolamides (OEA and DEA normalized to the LEA level), the 12,13-DiHOME/EpOME an indicator of sEH activity (28) and sEH metabolite of AA (14,15-DiHETrE). In CSF, the strongest predictors included OEA/LEA and the linoleate-derived epoxides 12(13)-EpOME and 9(10)-EpOME. When plasma and CSF markers were combined in predictive model efforts, the resulting model consisted uniquely of ethanolamides, including plasma long chain PUFA ethanolamides (DEA/LEA and DHEA/LEA) and CSF OEA/LEA. This model resulted in the ROC AUC of 0.889.
For all 3 models, ethanolamides OEA, DEA and DHEA were stronger predictors when used as a ratio to LA derived ethanolamide – LEA. LEA itself was not different between AD and the control group in either plasma or CSF (Fig. 1 and Fig. 2) unlike OEA, DEA and DHEA. Therefore, LEA likely serves as a surrogate for the general acyl-ethanolamides level and adjustment of other ethanolamides by LEA lowers intra-individual variability.
Lipid mediator–cognitive score associations in AD. The AD cohort is characterized by a high log(t-Tau/Αβ42) ratio and MoCA scores ranging from normal cognitive function to severe cognitive impairment (Figure S3). Taking advantage of the broad MoCA range, we investigated lipid mediator associations with cognitive function in this group pathological levels of t-Tau/Aβ42. Additionally, since the AD cohort was represented by subjects in both fasted and non-fasted states, we stratified the analysis by fasting state for plasma samples (Table 3). In the fasting state, PUFA oxidation markers, 5,15-DiHETE and 9-HETE were negatively associated with the MoCA score (although only 5,15-DiHETE passed FDR correction). 5,15-DiHETE can have enzymatic or autooxidative origin, whereas 9-HETE is a strictly an autooxidative product. 5,15-DiIHETE correlated with 9-HETE in fasted subjects with an R2 = 0.415 (n = 60; p < 0.001). In non-fasted AD subjects, a strong positive association between the MoCA score and EPA-derived ethanolamide (EPEA) as well as the levels of EPA and DHA were observed. Additionally, a positive correlation was detected between MoCA and the EPA-derived 17,18-DiHETE, the DHA-derived 14-HDoHE and the 18 carbon PUFAs (LA and ALA), however, these did not pass FDR correction.
Table 3
Spearman’s ρ rank order correlation between MoCA score and plasma lipid mediators in AD patients. Analysis stratified by predicted fasted state. Only correlations with the p < 0.05 are shown. P-values that passed FDR correction at q = 0.2 are bolded.
Fasted | | Non-fasted |
---|
Metabolite | Spearman ρ | P value | | Metabolite | Spearman ρ | P value |
5,15-DiHETE | -0.448 | 0.0005 | | EPEA | 0.424 | 0.0003 |
9-HETE | -0.338 | 0.0102 | | EPA | 0.386 | 0.001 |
13-KODE | -0.299 | 0.0238 | | DHA | 0.338 | 0.0043 |
DCA | -0.273 | 0.0398 | | 17,18-DiHETE | 0.3 | 0.0117 |
| | | | 4-HDoHE | 0.269 | 0.0246 |
| | | | LA | 0.267 | 0.0254 |
| | | | ALA | 0.249 | 0.0373 |
| | | | 9-HETE | -0.246 | 0.0405 |
| | | | 8-HETE | -0.24 | 0.0459 |
In CSF, the linoleic acid derived epoxides 12(13)- and 9(10) -EpOMEs showed weak but significant positive correlations with MoCA (ρ > 0.2, p < 0.005; Table 4). Additionally, positive associations were observed between MoCA and DHA and DHA derived diol (19,20-DiHDoPE) and conjugated bile acids GCA (and the ratio of GCA to GDCA and GCDCA), TCDCA and the conjugated to unconjugated ratio for DCA and CDCA (GCA/GCDCA, GCDCA/CDCA and TCDCA/CDCA). However, only linoleic acid epoxides passed the FDR correction.
Table 4
Spearman’s ρ rank order correlation between MoCA score and CSF lipid mediators in AD patients. Only correlations with the p < 0.05 are shown. P-values that passed FDR correction at q = 0.2 are bolded.
Metabolite | Spearman ρ | P value |
---|
12(13)-EpOME | 0.279 | 0.0009 |
9(10)-EpOME | 0.24 | 0.0047 |
19,20-DiHDoPE | 0.21 | 0.0138 |
GCA | 0.207 | 0.0152 |
DHA | 0.203 | 0.0174 |
GCA/GDCA | 0.202 | 0.0182 |
GCDCA/CDCA | 0.19 | 0.026 |
TDCA/DCA | 0.19 | 0.0261 |
GCA/GCDCA | 0.18 | 0.0352 |
TCDCA/CDCA | 0.172 | 0.0441 |
TCDCA | 0.17 | 0.0468 |