This section provides information about the study participants, the results of the univariate analysis of the metabolites, the multi-variate analyses for the 4 subsets of metabolites discussed in the methodology section, and lastly a correlation analysis to investigate the grouping of metabolites into five primary groups.
Univariate analysis
Participants. The medical histories and characteristics of the patients are shown in tables 1–3. The hypothesis testing in the tables was done using either Chi-squared test or the Student’s t-test. Table 1 lists basic characteristics of the mothers and their children, the medical histories of the mothers, and the developmental history of their children. Table 2 lists the medications that were being taken by the mothers at the time of the study. Table 3 lists mental and physical symptoms of the mothers. Each table lists n.s. for the p-value or FDR when the result was greater than 0.05 for the p-value or 0.1 for FDR, marking that the measurement showed no statistically significant differences between the two groups. All ASD-M participants enrolled in this study had a child that met full criteria for ASD based on ADI-R scores. The average age of mothers in the ASD-M group and TD-Group were similar (35.4 years and 34.9 years, respectively), since they were matched for maternal age. The average ages of their children was slightly older for the ASD group (4.71 vs. 3.87 years), as we allowed any ages between 2 and 5 years, and the ASD group was skewed towards the end of that range since it takes time for children with ASD to be diagnosed and for us to have contact with them..
|
ASD (n=30)
|
TD (n=29)
|
p-Value of t-test (T) or Chi-Squared (C)
|
FDR
|
Maternal age
|
35.4 (5.2)
|
35.2 (5.8)
|
n.s. (T)
|
|
Child gender
|
22 m, 8 f (73% male)
|
14 m, 14 f (50% male)
|
0.05 (C)
|
n.s.
|
Child age
|
4.71 (1.0)
|
3.87 (1.3)
|
0.0091 (T)
|
0.00
|
Child birthweight
|
7.21 (4.2)
|
6.20 (4.7)
|
n.s. (T)
|
|
Pregnancy complications
|
43%
(18% mild, 18% moderate, 7% severe
|
39%
(25% mild, 14% moderate, 0% severe)
|
n.s. (C)
|
|
Birth complications
|
50%
(36% mild, 11% moderate, 4% severe)
|
32%
(21% mild, 7% moderate, 4% severe)
|
n.s. (C)
|
|
C-section
|
43%
|
29%
|
n.s. (C)
|
|
Months of Breastfeeding without formula
|
9.1 (11)
|
8.4 (9.6)
|
n.s. (T)
|
|
Months of Breastfeeding with formula
|
2.9 (4.8)
|
4.3 (7.1)
|
n.s. (T)
|
|
Months of formula only
|
4.4 (5.7)
|
3.1 (4.2)
|
n.s. (T)
|
|
Solids introduced
|
6.4 (1.3)
|
6.2 (2.6)
|
n.s. (T)
|
|
Prenatal usage
|
89%
|
93%
|
n.s. (C)
|
|
% used prenatal prior to conception
|
37%
|
37%
|
|
|
Week started prenatal (with preconception use scored as week zero)
|
2.8 (3)
|
3.8 (4)
|
n.s. (T)
|
|
Pesticide exposure
|
14%
|
11%
|
n.s. (C)
|
|
% organic food
|
24% (26%)
|
24% (25%)
|
n.s. (T)
|
|
Child’s Antibiotic usage (Rounds, where 1 round=10 days)
|
0-6 months
|
0.22 (0.6)
|
0.29 (0.5)
|
n.s. (T)
|
|
6-12 months
|
0.73 (1.3)
|
0.79 (1.0)
|
n.s. (T)
|
|
12-24 months
|
1.08 (2.0)
|
1.25 (1.7)
|
n.s. (T)
|
|
24-36 months
|
1.12 (1.1)
|
0.74 (1.1)
|
n.s. (T)
|
|
36-48 months
|
0.77 (1.1)
|
0.36 (0.6)
|
n.s. (T)
|
|
Total antibiotic usage 0-48 months
|
3.29 (4.1)
|
3.24 (3.0)
|
n.s. (T)
|
|
Child’s Asthma severity
|
3%
1 mild
|
18%
3 mild, 1 moderate, 1 severe
|
n.s. (C)
|
|
Child’s Food allergies/sensitivities (severity based on most severe allergy)
|
32%
(7% mild
14% moderate
11% severe)
|
11%
(4% mild, 7% moderate
0% severe)
|
n.s. (C)
|
|
Other allergies
(severity based on most severe allergy)
|
14%
(7% mild, 4% moderate, 4% unknown)
|
33%
(19% mild, 15% moderate, 0% severe)
|
n.s. (C)
|
|
Child’s Developmental history of ASD
|
Early Onset
|
14%
|
|
|
|
Normal development, then regression (age of regression)
|
46%
19 (5) months
|
|
|
|
Normal development, then plateau (age of plateau)
|
39%
19 (9) months
|
|
|
|
Table 1
Characteristics and medical histories of participants. The means are shown with the standard deviations in parentheses.
The results in Table 1 indicate that besides the child’s age, there were no significant differences between the two groups of mothers/children in their medical histories.
| ASD (n = 30) | NT (n = 29) | Chi-Squared | FDR |
Use of Any Medications | 43% (13) | 30% (9) | n.s | |
Psychiatric | 17% (5) | 7% (2) | n.s. | |
Allergy | 10% (3) | 14% (4) | n.s. | |
Birth control | 33% (10) | 10% (3) | 0.03 | n.s. |
Inhaler | 0% (0) | 3% (1) | n.s. | |
Blood pressure | 7% (2) | 0% (0) | n.s. | |
Thyroid | 10% (3) | 7% (2) | n.s. | |
GI | 0% (0) | 3% (1) | n.s. | |
Pain | 0% (0) | 3% (1) | n.s. | |
Epi-pen as needed | 3% (1) | 0% (0) | n.s. | |
Use of Nutritional Supplements* | 7% (2) | 7% (2) | n.s. | |
*None of these nutritional supplements contained folate or vitamin B12. |
Table 2
Current maternal medication use at 2-5 years after pregnancy. The percentage of participants that used these medications is listed with the actual number in parentheses.
| ASD (n = 30) | TD (n = 29) | Ratio of ASD/TD | t-test | FDR |
Fatigue | 1.15 (1.0) | 0.79 (0.8) | 1.46 | n.s. | |
Fibromyalgia | 0.11 (0.6) | 0 (0) | | n.s. | |
Depression | 0.67 (0.8) | 0.48 (0.7) | 1.38 | n.s. | |
Irritability | 0.78 (0.9) | 0.71 (0.8) | 1.09 | n.s. | |
Anxiety | 1.11 (0.9) | 0.75 (1.0) | 1.48 | 0.08 | n.s. |
Cognition | 0.56 (1.0) | 0.32 (0.6) | 1.73 | n.s. | |
Attention | 0.85 (1.0) | 0.32 (0.55) | 2.65 | 0.056 | n.s. |
Sensory Sensitivity | 0.44 (0.7) | 0.18 (0.4) | 2.49 | n.s. | |
Stool/GI Problems | 0.44 (0.8) | 0.71 (1.0) | 0.62 | n.s. | |
Sleep Problems | 0.74 (0.9) | 0.46 (0.74) | 1.60 | n.s. | |
Headaches | 0.56 (0.8) | 0.64 (0.7) | 0.86 | n.s. | |
Chemical Sensitivity | 0.19 (0.5) | 0.14 (0.5) | 1.30 | n.s. | |
Allergies | 0.78 (1.0) | 0.71 (0.7) | 1.09 | n.s. | |
Expressive Language | 0.30 (0.9) | 0.11 (0.6) | 2.77 | n.s. | |
Receptive Language | 0.31 (0.7) | 0.11 (0.4) | 2.77 | n.s. | |
Other symptoms | Multiple sclerosis; very sensitive to alcohol; frequent boils; | Nausea/pain from ovarian cyst; chronic pain; sensitive to loud noises; touch aversive | | | |
Average | 0.60 (0.5) | 0.44 (0.4) | 1.37 | n.s. | |
Table 3
Current mental and physical symptoms of mothers. The severity scale was 0=none, 1=mild, 2=moderate, 3=severe. The mean is listed with the standard deviation in parentheses.
The p-values and FDR results in Tables 2 and 3 show that there were no significant differences in the medication use listed and the symptoms experienced between the two groups of mothers during the study period. The slightly higher use of birth control in the ASD group may be due to less desire to having additional children after one child is diagnosed with ASD.
FOCM/TS Metabolites. The univariate results for the FOCM/TS metabolites are shown in Table 4. Levels of vitamin B12 and the SAM/SAH ratio are significantly lower in the ASD-M group compared to the TD-M group, (p 0.05, FDR 0.1). Also, levels of Glu-Cys, fCysteine, and fCystine are significantly higher in the ASD-M group compared to the TD-M group (p 0.05, FDR 0.1).
Metabolite
|
Test
|
ASD-M (mean std)
N= 30
|
TD-M
(mean std)
N= 29
|
p-Values
|
FDR
|
AUC
|
Ratio (ASD-M/TD-M)
|
B12
|
MW
|
355 196
|
473 173
|
2.40E-03
|
0.00
|
0.73
|
0.75
|
fCysteine
|
t=
|
23.8 1.88
|
22.6 1.94
|
0.01
|
0.00
|
0.70
|
1.06
|
Glu-Cys
|
t=
|
1.89 0.22
|
1.72 0.24
|
0.01
|
0.00
|
0.69
|
1.10
|
SAM/SAH
|
t=
|
1.94 0.25
|
2.09 0.20
|
0.01
|
0.00
|
0.67
|
0.93
|
fCystine
|
t=
|
24.1 2.78
|
22.4 2.43
|
0.02
|
0.01
|
0.67
|
1.07
|
tCysteine
|
t=
|
248 23.1
|
234 28.3
|
0.04
|
0.40
|
0.64
|
1.06
|
tGSH
|
t=
|
6.23 0.98
|
5.85 1.07
|
0.15
|
1.00
|
0.63
|
1.07
|
SAM
|
t=
|
47.2 5.37
|
49.3 5.76
|
0.15
|
1.00
|
0.62
|
0.96
|
Methionine
|
t=
|
19.9 2.56
|
20.8 2.98
|
0.19
|
1.00
|
0.61
|
0.95
|
MTHFR mut. (A1298C)
|
|
|
|
0.15
|
1.00
|
0.60
|
|
tGSH/GSSG
|
t=
|
29.5 6.21
|
27.3 6.06
|
0.18
|
1.00
|
0.60
|
1.08
|
Folate
|
MW
|
17.6 6.19
|
21.1 9.17
|
0.20
|
1.00
|
0.60
|
0.83
|
Homocysteine
|
MW
|
8.63 0.98
|
8.27 1.18
|
0.21
|
1.00
|
0.60
|
1.04
|
tGSH/GSSG
|
t=
|
29.5 6.21
|
27.3 6.06
|
0.18
|
1.00
|
0.60
|
1.08
|
SAH
|
t=
|
24.5 2.66
|
23.6 2.04
|
0.15
|
1.00
|
0.59
|
1.04
|
Vitamin D3*
|
t=
|
27.1 9.10
|
24.9 6.08
|
0.27
|
1.00
|
0.57
|
1.09
|
Ferritin
|
MW
|
35.1 31.0
|
29.5 26.1
|
0.36
|
1.00
|
0.57
|
1.19
|
Cys-Gly
|
t=
|
38.7 5.06
|
37.6 6.38
|
0.46
|
1.00
|
0.57
|
1.03
|
Adenosine
|
t=
|
0.22 0.03
|
0.21 0.03
|
0.28
|
1.00
|
0.55
|
1.04
|
fGSH/GSSG
|
MW
|
8.73 2.08
|
8.81 1.84
|
0.49
|
1.00
|
0.55
|
0.99
|
% Oxidized Glutathione
|
MW
|
0.19 0.03
|
0.19 0.04
|
0.49
|
1.00
|
0.55
|
1.01
|
Chlorotyrosine
|
t=
|
26.8 4.27
|
27.6 4.23
|
0.51
|
1.00
|
0.55
|
0.97
|
Nitrotyrosine
|
t
|
32.9 6.28
|
33.7 4.67
|
0.59
|
1.00
|
0.55
|
0.98
|
fGSH
|
t=
|
1.85 0.32
|
1.89 0.35
|
0.62
|
1.00
|
0.55
|
0.98
|
fCystine/fCysteine
|
t=
|
1.01 ± 0.11
|
1.00 0.10
|
0.59
|
1.00
|
0.54
|
1.02
|
Vitamin E
|
t=
|
9.23 2.53
|
9.82 3.22
|
0.75
|
1.00
|
0.54
|
0.94
|
Isoprostane (U)*
|
MW
|
0.15 0.10
|
0.18 0.14
|
0.70
|
1.00
|
0.53
|
0.83
|
MTHFR mut. (C677T)
|
|
|
|
0.90
|
1.00
|
0.53
|
|
GSSG
|
t=
|
0.22 0.04
|
0.22 0.03
|
0.93
|
1.00
|
0.52
|
1.00
|
MMA
|
MW
|
0.15 0.06
|
0.15 0.08
|
0.64
|
1.00
|
0.51
|
0.96
|
Note. The measurements are ordered by decreasing AUC. Statistically significant metabolites with p-value ≤ 0.05 and FDR ≤ 0.1 are marked with Δ and * indicates measurements that were left out of the classification procedure as the measurements were not collected from all mothers. Specifically, these were vitamin D with 28 mothers in ASD-M and 28 TD-M mothers and Isoprostane with 28 participants that were ASD-M and 25 mothers in TD-M.
|
Table 4
Univariate results for FOCM/TS metabolites and vitamin E, folate, ferritin, B12, MMA, and MTHFR status.
Global metabolic profile- Metabolon. 622 metabolites were measured in whole blood. The univariate analysis for the 50 metabolites from broad metabolomics with the highest AUC values are shown in the Table 5. They are ordered starting with those with the highest AUC. Note that these are semi-quantitative measurements (no absolute values), so only the ratio of ASD-M/TD-M is shown. In almost every case the ASD-M group had lower levels of metabolites than the TD-M group, with the levels of 4-vinylphenol sulfate, NAD+, and three glycine-containing metabolites (gamma-glutamylglycine, cinnamoylglycine, propionylglycine) being especially low (ASD-M/TD-M ratio < 0.50). Four metabolites were higher in the ASD-M group (histidylglutamate, asparaginylalanine, dimethyl sulfone, and mannose). Note that dimethyl sulfone was unusually high in the ASD-M group (ASD-M/TD-M ratio = 18.7, p = 0.01, but the FDR was not significant). 80% of the TD-M measurements of dimethyl sulfone and 47% of the ASD-M measurements of dimethyl sulfone were below the detection limit, and the distribution of the data for is skewed.
Metabolite
|
Test
|
p-Value
|
FDR
|
AUC
|
Ratio (ASD-M/TD-M)
|
Fructose
|
t *
|
6.88E-04
|
0.00
|
0.81
|
0.60
|
Histidylglutamate
|
t
|
2.67E-05
|
0.00
|
0.80
|
1.50
|
Decanoylcarnitine (C10)
|
t
|
1.55E-04
|
0.00
|
0.78
|
0.66
|
S-1-pyrroline-5-carboxylate
|
MW
|
3.09E-04
|
0.00
|
0.77
|
0.63
|
Octanoylcarnitine (C8)
|
MW
|
4.00E-04
|
0.00
|
0.77
|
0.67
|
4-vinylphenol sulfate
|
t *
|
1.30E-03
|
0.00
|
0.77
|
0.31
|
Cis-4-decenoylcarnitine (C10:1)
|
t=
|
5.82E-04
|
0.00
|
0.74
|
0.75
|
N-formylanthranilic acid
|
t=
|
1.20E-03
|
0.00
|
0.74
|
0.69
|
N-acetylasparagine
|
t *
|
1.90E-03
|
0.00
|
0.73
|
0.78
|
Arachidoylcarnitine (C20)*
|
MW
|
3.00E-03
|
0.00
|
0.73
|
0.85
|
N-palmitoylglycine
|
t=
|
2.20E-03
|
0.00
|
0.72
|
0.81
|
Citrulline
|
t=
|
2.60E-03
|
0.00
|
0.72
|
0.91
|
6-hydroxyindole sulfate
|
t
|
1.20E-03
|
0.00
|
0.72
|
0.59
|
N-palmitoylserine
|
MW
|
4.00E-03
|
0.00
|
0.72
|
0.77
|
Myristoylcarnitine (C14)
|
MW
|
4.10E-03
|
0.00
|
0.72
|
0.82
|
Laurylcarnitine (C12)
|
MW
|
4.30E-03
|
0.00
|
0.72
|
0.74
|
Stearoylcarnitine (C18)
|
MW
|
0.01
|
0.00
|
0.71
|
0.85
|
Gamma-glutamylglycine
|
t *
|
2.20E-03
|
0.00
|
0.71
|
0.27
|
5-oxoproline
|
t=
|
0.01
|
0.00
|
0.70
|
0.94
|
Asparaginylalanine
|
t=
|
3.90E-03
|
0.00
|
0.70
|
1.32
|
Glutamine
|
t=
|
0.02
|
0.00
|
0.70
|
0.95
|
Catechol sulfate
|
MW
|
0.01
|
0.00
|
0.70
|
0.67
|
3-indoxyl sulfate
|
t
|
2.70E-03
|
0.00
|
0.70
|
0.67
|
7-methylxanthine
|
MW
|
0.01
|
0.00
|
0.70
|
0.58
|
Phenol sulfate
|
MW
|
0.01
|
0.00
|
0.70
|
0.67
|
Cinnamoylglycine
|
t *
|
0.01
|
0.00
|
0.70
|
0.46
|
Alpha-ketoglutaramate*
|
t=
|
0.02
|
0.00
|
0.70
|
0.73
|
Isovalerylglycine
|
MW
|
0.01
|
0.00
|
0.69
|
0.78
|
Propionylglycine
|
MW
|
0.01
|
0.00
|
0.69
|
0.48
|
Docosapentaenoylcarnitine (C22:5n3)*
|
MW
|
0.01
|
0.00
|
0.69
|
0.72
|
N-acetyl-2-aminooctanoate*
|
t
|
3.20E-03
|
0.00
|
0.69
|
0.54
|
S-methylglutathione
|
t=
|
0.02
|
0.01
|
0.69
|
0.86
|
Gamma-glutamyltyrosine
|
MW
|
0.02
|
0.00
|
0.68
|
0.64
|
Succinylcarnitine (C4-DC)
|
t=
|
0.03
|
0.07
|
0.68
|
0.87
|
Arachidonoylcarnitine (C20:4)
|
MW
|
0.02
|
0.00
|
0.68
|
0.77
|
Glycine
|
t=
|
0.01
|
0.00
|
0.68
|
0.87
|
N-acetylvaline
|
t=
|
0.04
|
0.28
|
0.68
|
0.80
|
Lignoceroylcarnitine (C24)*
|
t=
|
0.02
|
0.01
|
0.68
|
0.84
|
Guaiacol sulfate
|
MW
|
0.02
|
0.01
|
0.68
|
0.81
|
5-methylthioadenosine (MTA)
|
MW
|
0.02
|
0.00
|
0.68
|
0.86
|
Proline
|
MW
|
0.02
|
0.00
|
0.68
|
0.90
|
Pyridoxate
|
MW
|
0.02
|
0.00
|
0.68
|
0.75
|
Palmitoylcarnitine (C16)
|
MW
|
0.02
|
0.03
|
0.67
|
0.83
|
Eicosenoylcarnitine (C20:1)*
|
MW
|
0.02
|
0.05
|
0.67
|
0.83
|
Nicotinamide adenine dinucleotide (NAD+)
|
t *
|
0.03
|
0.05
|
0.67
|
0.41
|
Dimethyl sulfone
|
MW
|
0.01
|
0.23
|
0.67
|
18.7
|
Tiglylcarnitine (C5:1-DC)
|
MW
|
0.02
|
0.02
|
0.67
|
0.63
|
Adrenoylcarnitine (C22:4)*
|
MW
|
0.03
|
0.11
|
0.67
|
0.74
|
3-methylxanthine
|
MW
|
0.03
|
0.13
|
0.67
|
0.74
|
Mannose
|
MW
|
0.03
|
0.19
|
0.67
|
1.21
|
Note. The metabolites listed here are the 50 metabolites measured by Metabolon from broad metabolomics with the highest area under the receiver operating characteristic (ROC) curve (AUC). Metabolites with p-value ≤ 0.05 and FDR≤0.1 marked with Δ.
|
Table 5
Univariate results of the metabolites from broad metabolomics.
Hypothesis testing was also done on the entire Metabolon dataset and revealed that 48 of these metabolites had significant differences between the two groups of mothers. 3 of these metabolites were not included in the top 50 used for analysis. These were not included in the multivariate analysis because they had lower AUC values than the metabolites included (see Table S-1).
Table 6 contains more information about the metabolites in Table 5. Table 6 lists the many metabolic pathways which had significant differences between the ASD-M and TD-M groups, including amino acids (15 metabolites), carbohydrates (1 metabolite), vitamins (2 metabolites), energy (1 metabolite), lipids (16 metabolites), peptides (4 metabolites), and xenobiotics (7 metabolites). When considering sub-pathways, there were differences in alanine/aspartate metabolism (1 metabolite), glutamate metabolism (3 metabolites), glutathione metabolism (2 metabolites), glycine (1 metabolites), leucine/isoleucine/valine (3 metabolites), polyamine (1 metabolite), tryptophan (2 metabolites), tyrosine (1 metabolite), urea cycle (2 metabolites), fructose/mannose (2 metabolites), nicotinamide (1 metabolite), vitamin B6 (1 metabolite), vitamin B12 (1 metabolite), TCA cycle (1 metabolite), endocannabinoid (1 metabolite), carnitine/fatty acid metabolism (12 metabolites), other fatty acid metabolism (3 metabolites), dipeptides (2 metabolites), gamma-glutamyl (2 metabolites), benzoate (3 metabolites), chemical/xenobiotics (2 metabolites), cinnamoylglycine (1 metabolite), and xanthine metabolism (2 metabolites).
Metabolite
|
Pathway
|
Sub-Pathway
|
Higher/lower in ASD-M group
|
N-acetylasparagineΔ
|
Amino Acid
|
Alanine and Aspartate Metabolism
|
↓
|
S-1-pyrroline-5-carboxylateΔ
|
Amino Acid
|
Glutamate Metabolism
|
↓
|
Glutamine
|
Amino Acid
|
Glutamate Metabolism
|
↓
|
Alpha-ketoglutaramate*Δ
|
Amino Acid
|
Glutamate Metabolism
|
↓
|
5-oxoprolineΔ
|
Amino Acid
|
Glutathione Metabolism
|
↓
|
S-methylglutathioneΔ
|
Amino Acid
|
Glutathione Metabolism
|
↓
|
GlycineΔ
|
Amino Acid
|
Glycine, Serine and Threonine Metabolism
|
↓
|
IsovalerylglycineΔ
|
Amino Acid
|
Leucine, Isoleucine and Valine Metabolism
|
↓
|
N-acetylvalineΔ
|
Amino Acid
|
Leucine, Isoleucine and Valine Metabolism
|
↓
|
Tiglylcarnitine (C5:1-DC)Δ
|
Amino Acid
|
Leucine, Isoleucine and Valine Metabolism
|
↓
|
5-methylthioadenosine (MTA)Δ
|
Amino Acid
|
Polyamine Metabolism
|
↓
|
N-formylanthranilic acidΔ
|
Amino Acid
|
Tryptophan Metabolism
|
↓
|
3-indoxyl sulfateΔ
|
Amino Acid
|
Tryptophan Metabolism
|
↓
|
Phenol sulfateΔ
|
Amino Acid
|
Tyrosine Metabolism
|
↓
|
CitrullineΔ
|
Amino Acid
|
Urea cycle; Arginine and Proline Metabolism
|
↓
|
ProlineΔ
|
Amino Acid
|
Urea cycle; Arginine Proline Metabolism
|
↓
|
MannoseΔ
|
Carbohydrate
|
Fructose, Mannose and Galactose Metabolism
|
↑
|
FructoseΔ
|
Carbohydrate
|
Fructose, Mannose, and Galactose Metabolism
|
↓
|
Nicotinamide adenine dinucleotide (NAD+)Δ
|
Cofactors and Vitamins
|
Nicotinate and Nicotinamide Metabolism
|
↓
|
PyridoxateΔ
|
Cofactors and Vitamins
|
Vitamin B6 Metabolism
|
↓
|
Succinylcarnitine (C4-DC)Δ
|
Energy
|
TCA Cycle
|
↓
|
N-palmitoylserineΔ
|
Lipid
|
Endocannabinoid
|
↓
|
Decanoylcarnitine (C10)Δ
|
Lipid
|
Fatty Acid Metabolism (Acyl Carnitine)
|
↓
|
Octanoylcarnitine (C8)Δ
|
Lipid
|
Fatty Acid Metabolism (Acyl Carnitine)
|
↓
|
Cis-4-decenoylcarnitine (C10:1)Δ
|
Lipid
|
Fatty Acid Metabolism (Acyl Carnitine)
|
↓
|
Arachidoylcarnitine (C20)*Δ
|
Lipid
|
Fatty Acid Metabolism (Acyl Carnitine)
|
↓
|
Myristoylcarnitine (C14)Δ
|
Lipid
|
Fatty Acid Metabolism (Acyl Carnitine)
|
↓
|
Laurylcarnitine (C12)Δ
|
Lipid
|
Fatty Acid Metabolism (Acyl Carnitine)
|
↓
|
Stearoylcarnitine (C18)Δ
|
Lipid
|
Fatty Acid Metabolism (Acyl Carnitine)
|
↓
|
Docosapentaenoylcarnitine (C22:5n3)*Δ
|
Lipid
|
Fatty Acid Metabolism (Acyl Carnitine)
|
↓
|
Arachidonoylcarnitine (C20:4)Δ
|
Lipid
|
Fatty Acid Metabolism (Acyl Carnitine)
|
↓
|
Lignoceroylcarnitine (C24)*Δ
|
Lipid
|
Fatty Acid Metabolism (Acyl Carnitine)
|
↓
|
Palmitoylcarnitine (C16)Δ
|
Lipid
|
Fatty Acid Metabolism (Acyl Carnitine)
|
↓
|
Eicosenoylcarnitine (C20:1)*Δ
|
Lipid
|
Fatty Acid Metabolism (Acyl Carnitine)
|
↓
|
Adrenoylcarnitine (C22:4)*Δ
|
Lipid
|
Fatty Acid Metabolism (Acyl Carnitine)
|
↓
|
N-palmitoylglycineΔ
|
Lipid
|
Fatty Acid Metabolism (Acyl Glycine)
|
↓
|
PropionylglycineΔ
|
Lipid
|
Fatty Acid Metabolism (also BCAA Metabolism)
|
↓
|
N-acetyl-2-aminooctanoate*Δ
|
Lipid
|
Fatty Acid, Amino
|
↓
|
HistidylglutamateΔ
|
Peptide
|
Dipeptide
|
↑
|
AsparaginylalanineΔ
|
Peptide
|
Dipeptide
|
↑
|
Gamma-glutamylglycineΔ
|
Peptide
|
Gamma-glutamyl Amino Acid
|
↓
|
Gamma-glutamyltyrosine Δ
|
Peptide
|
Gamma-glutamyl Amino Acid
|
↓
|
4-vinylphenol sulfateΔ
|
Xenobiotics
|
Benzoate Metabolism
|
↓
|
Catechol sulfateΔ
|
Xenobiotics
|
Benzoate Metabolism
|
↓
|
Guaiacol sulfateΔ
|
Xenobiotics
|
Benzoate Metabolism
|
↓
|
6-hydroxyindole sulfateΔ
|
Xenobiotics
|
Chemical
|
↓
|
Dimethyl sulfoneΔ
|
Xenobiotics
|
Chemical
|
↑
|
CinnamoylglycineΔ
|
Xenobiotics
|
Food Component/Plant
|
↓
|
7-methylxanthineΔ
|
Xenobiotics
|
Xanthine Metabolism
|
↓
|
3-methylxanthine
|
Xenobiotics
|
Xanthine Metabolism
|
↓
|
Note. The metabolites listed here are the 50 metabolites measured by Metabolon from broad metabolomics with the highest area under the receiver operating characteristic (ROC) curve (AUC). The metabolites are sorted alphabetically by pathway and then subpathway. A fourth column lists whether the metabolites were higher or lower in the ASD-M group. Metabolites that had a p-value ≤0.05 and FDR ≤ 0.1 (FDR-values listed in Table 3) are marked by Δ.
|
Table 6
Pathways and subpathways of the metabolites from the broad metabolomics.
Carnitine. As shown in Table 6, several carnitine-conjugated metabolites are significantly different in the two groups of mothers. Table 7 below highlights the univariate hypothesis testing results for the carnitine-conjugated metabolites specifically in order of increasing size, from 4-carbon to 24-carbon chains. The ratio of ASD/TD for carnitine-conjugated metabolites was consistently low, ranging from 0.63 to 0.87, with an average of 0.77. There were 33 additional carnitine metabolites in the 600 metabolites measured by untargeted metabolomics. Of these 33, only three had ratios indicating levels of the carnitine were higher in the ASD-M group than in the TD-M group. Also, eight of these metabolites showed significant difference in mean/median between the two groups using hypothesis testing. All of the eight carnitine metabolites had ratios indicating that the levels of carnitine-conjugated molecules in the ASD-M group were less than in the TD-M group.
In contrast, carnitine and two of its precursors (lysine and trimethyllysine) are very similar in the ASD-M and TD-M groups (within 1%), suggesting that the low levels of carnitine-conjugated metabolites is not due to a carnitine deficiency.
Carnitine | Test | p-Value | FDR | AUC | Ratio (ASD-M/TD-M |
Succinylcarnitine (C4-DC) | t= | 0.03 | 0.07 | 0.68 | 0.87 |
Tiglylcarnitine (C5:1-DC) | MW | 0.02 | 0.02 | 0.67 | 0.63 |
Octanoylcarnitine (C8) | MW | 4.00E-04 | 0.00 | 0.77 | 0.67 |
Decanoylcarnitine (C10) | t | 1.55E-04 | 0.00 | 0.78 | 0.66 |
Cis-4-decanoylcarnitine (C10:1) | t= | 5.82E-04 | 0.00 | 0.74 | 0.75 |
Laurylcarnitine (C12) | MW | 4.30E-03 | 0.00 | 0.72 | 0.74 |
Myristoylcarnitine (C14) | MW | 4.10E-03 | 0.00 | 0.72 | 0.82 |
Palmitoylcarnitine (C16) | MW | 0.02 | 0.03 | 0.67 | 0.83 |
Arachidoylcarnitine (C20)* | MW | 3.00E-03 | 0.00 | 0.73 | 0.85 |
Eicosenoylcarnitine (C20:1)* | MW | 0.02 | 0.05 | 0.67 | 0.83 |
Arachidonoylcarnitine (C20:4) | MW | 0.02 | 0.00 | 0.68 | 0.77 |
Adrenoylcarnitine (C22:4)* | MW | 0.03 | 0.11 | 0.67 | 0.74 |
Docosapentaenoylcarnitine (C22:5n3)* | MW | 0.01 | 0.00 | 0.69 | 0.72 |
Lignoceroylcarnitine (C24)* | t= | 0.02 | 0.01 | 0.68 | 0.84 |
Note. Statistically significant metabolites with a p-value ≤ 0.05 and FDR ≤ 0.1 are shown in gray.
|
Table 7
Univariate hypothesis testing results for the carnitine-conjugated metabolites.
Multivariate Analysis
The multivariate analysis was performed using multiple subsets of data. The subsets included the twenty metabolites from the FOCM/TS pathways (i), the FOCM/TS metabolites plus some additional nutritional information (ii), the FOCM/TS metabolites plus the additional nutritional information and the MTHFR gene information (iii), and subset (iii) plus fifty metabolites from the broad metabolomics analysis (iv). The first two subsets were analyzed using FDA because all of the variables were continuous and the last two subsets were analyzed using logistic regression because the variables included both continuous and binary data. Each multivariate analysis was combined with leave-one-out cross-validation in order to analyze the success of the model on classification. The best combinations of metabolites from each of the first three subsets had errors ranging from 20-27% which shows only a very modest ability to predict which group of mothers the sample came from. Table 8 below details the type I/type II errors using these metabolites.
Subset
|
Combination
|
Type I Error
|
Type II Error
|
(i): FOCM/TS Metabolites
|
tCysteine, Glu-Cys, fCysteine, fCystine/fCystiene, Nitrotyrosine
|
24%
|
27%
|
(ii): FOCM/TS metabolites plus nutritional information
|
SAM/SAH, Glu-Cys, GSSG, fCysteine, B12
|
24%
|
27%
|
(iii): FOCM/TS metabolites, nutritional information, and MTHFR gene information
|
SAM/SAH, tCysteine, Glu-Cys, B12, MTHFR mut. (A1298C)
|
24%
|
20%
|
Table 8
Results for the combination of metabolites from the first three subsets (i-iii) with lowest errors.
In order to visually demonstrate the separation between the two groups, a probability density function (PDF) was created for each of the best combinations analyzed using FDA. These PDFs are shown in figures 1 and 2.
Figure 1: PDFs of the combination of metabolites from the FOCM/TS metabolites (i) that resulted in the respective errors shown in table 8.
Figure 2: PDFs of the combination of metabolites from the FOCM/TS metabolites and additional measurements (ii) that resulted in the respective errors shown in table 8.
In order to visually demonstrate the classification accuracy between the two groups when using the logistic regression classification model, a scatter plot was created showing the probabilities of each sample being classified as one group or another. The scatter plot representing the combination of metabolites from the FOCM/TS metabolites plus additional information and the MTHFR gene information (iii) is shown in the figure 3 below.
Figure 3: Scatter plot of the probabilities of being classified into one group or the other using a combination of variables from the FOCM/TS pathways, the additional measurements, and the MTHFR gene information (iii) that resulted in the errors listed in table 8.
The highest accuracies were found when analyzing the fourth, and largest, subset of metabolites. The best combinations for 2, 3, 4, and 5 metabolites for subset iv are shown in Table 9; combinations that contained more than 5 variables resulted in a decrease in accuracy due to overfitting of the classification model. It is important to note that many other combinations of metabolites yielded similar results and the top combinations of five metabolites are listed in Table 10. The results for using even only two metabolites resulted in lower Type 1 and Type 2 errors than the analysis using the other subsets described above (Table 5) and including more than two metabolites for classification further improved accuracy.
Metabolites
|
Type I Error (FPR)
|
Type II Error (FNR)
|
2 metabolites:
Histidylglutamate, 6-hydroxyindoel sulfate
|
17%
|
13%
|
3 metabolites:
Histidylglutamate, N-formylanthranilic acid, palmitoylcarnitine (C16)
|
7%
|
7%
|
4 metabolites: Histidylglutamate, S-1-pyrroline-5-carboxylate, N-acetyl-2-aminooctanoate*, 5-methylthioadenosine (MTA)
|
3%
|
7%
|
5 Metabolites:
Glu-Cys, histidylglutamate, cinnamoylglycine, proline, adrenoylcarnitine (C22:4)*
|
3%
|
3%
|
Table 9
Multivariate results using top combinations of 2-5 variables from subset (iv).
Metabolites
|
Type I Error (FPR)
|
Type II Error (FNR)
|
SAM/SAH, percent oxidized, histidylglutamate, cis-4-decenoylcarnitine (C10:1), 3-indoxyl sulfate
|
3%
|
7%
|
fGSH/GSSG, histidylglutamate, 4-vinylphenol sulfate, 3-indroxyl sulfate, palmitoylcarnitine (C16)
|
3%
|
7%
|
Histidylglutamate, 4-vinylphenol sulfate, cinnamoylglycine, N-acetylvaline, palmitoylcarnitine (C16)
|
3%
|
7%
|
Glu-Cys, histidylglutamate, catechol sulfate, phenol sulfate, N-acetyl-2-aminooctanoate*
|
3%
|
7%
|
tGSH, 4-vinylphenol sulfate, 5-oxoproline, asparaginylalanine, tiglylcarnitine (C5:1-DC)
|
7%
|
3%
|
Table 10
Multivariate results using the top combinations of 5 variables from subset (iv).
To further illustrate classification accuracy, the 5-metabolite model from Table 9 was used and the probabilities that the samples would be classified by the model in each of the two groups are shown in Figure 1. The metabolites of this 5-metabolite model consisting of Glu-Cys, histidylglutamate, cinnamoylglycine, proline, adrenoylcarnitine (C22:4)* are hereafter referred to as the core metabolites.
Figure 4: Scatter plot of the probabilities of being classified into one group or the other using a combination of variables from the FOCM/TS pathways, the additional measurements, and the top 50 metabolites from the metabolon.
The plots show that the ASD-M samples have a high probability of being classified as ASD-M and the TD-M samples have a high probability of being classified as TD-M. The results from this figure coupled with the low misclassification errors from Table 9 show that there are significant metabolic differences between the two groups of mothers and that these differences are sufficiently large to allow for accurate classification in the vast majority of cases.
In order to further investigate the differences between the two groups, we calculated the correlation coefficients between the 5 metabolites from the best classification model (Table 9) and the rest of the metabolites considered in the analysis for the combined set of ASD-M and TD-M samples. The metabolites that had the highest correlation coefficients with these metabolites are listed in Table 11. We also calculated the correlation of the top 5 metabolites with one another, and, as expected, found very little correlation among these five (see Table 12); this is not unexpected as the classification algorithms tries to identify metabolites that provide new information that can be used for classification as redundant information will not increase classification accuracy. This suggests that there are five general areas of metabolic differences in mothers of children with/without ASD involving 9 or more metabolites for each area.
Metabolite
|
Correlation Coefficient
|
p-Value
|
Glu-Cys
|
|
|
tGSH
|
0.55
|
5.71E-06
|
tGSH/GSSG
|
0.35
|
0.01
|
6-hydroxyindole sulfate
|
-0.25
|
0.05
|
SAM/SAH
|
-0.26
|
0.04
|
N-formylanthranilic acid
|
-0.28
|
0.03
|
5-methylthioadenosine (MTA)
|
-0.28
|
0.03
|
Pyridoxate
|
-0.31
|
0.02
|
Folate
|
-0.38
|
3.40E-03
|
Histidylglutamate
|
|
|
Asparaginylalanine
|
0.55
|
6.74E-06
|
Mannose
|
0.40
|
1.70E-03
|
fCystine
|
0.30
|
0.02
|
Succinylcarntine (C4-DC)
|
-0.27
|
0.04
|
Citrulline
|
-0.27
|
0.04
|
Fructose
|
-0.28
|
0.03
|
Octanoylcarnitine (C8)
|
-0.29
|
0.02
|
Gamma-glutamylglycine
|
-0.30
|
0.02
|
Isovaleryglycine
|
-0.32
|
0.01
|
Decanoylcarnitine (C10)
|
-0.33
|
0.01
|
Cinnamoylglycine
|
|
|
N-acetyl-2-aminooctanoate*
|
0.45
|
4.13E-04
|
N-formylanthranilic acid
|
0.44
|
4.30E-04
|
3-indoxyl sulfate
|
0.35
|
0.01
|
Citrulline
|
0.33
|
0.01
|
6-hydroxyindole sulfate
|
0.32
|
0.01
|
Chlorotyrosine
|
0.29
|
0.02
|
Alpha-ketoglutaramate*
|
0.28
|
0.03
|
Nicotinamide adenine dinucleotide (NAD+)
|
0.28
|
0.03
|
Pyridoxate
|
0.27
|
0.04
|
Guaiacol sulfate
|
0.26
|
0.04
|
S-methylglutathione
|
0.26
|
0.04
|
Methionine
|
-0.29
|
0.02
|
fCysteine
|
-0.33
|
0.01
|
Proline
|
|
|
S-1-pyrroline-5-carboxylate
|
0.59
|
1.22E-06
|
Gamma-glutamyltyrosine
|
0.45
|
3.73E-04
|
3-indoxyl sulfate
|
0.44
|
5.33E-04
|
6-hydroxyindole sulfate
|
0.43
|
6.01E-04
|
Phenol sulfate
|
0.41
|
1.20E-03
|
Glutamine
|
0.36
|
0.01
|
Propionylglycine
|
0.35
|
0.01
|
Glycine
|
0.35
|
0.01
|
Gamma-glutamylglycine
|
0.32
|
0.01
|
5-oxoproline
|
0.30
|
0.02
|
Alpha-ketoglutaramate*
|
0.28
|
0.03
|
Folate
|
0.28
|
0.03
|
N-formylanthranilic acid
|
0.27
|
0.04
|
Adenosine
|
-0.30
|
0.02
|
Adrenoylcarnitine (C22:4)*
|
|
|
Arachidonoylcarnitine (C20:4)
|
0.93
|
8.00E-26
|
Docosapentaenoylcarnitine (C22:5n3)*
|
0.85
|
8.64E-18
|
Eicosenoylcarnitine (C20:1)*
|
0.74
|
2.35E-11
|
Palmitoylcarnitine (C16)
|
0.70
|
8.13E-10
|
Myristoylcarnitine (C14)
|
0.69
|
2.17E-09
|
Laurylcarnitine (C12)
|
0.49
|
8.88E-05
|
Fructose
|
0.41
|
1.20E-03
|
N-acetylasparagine
|
0.38
|
2.60E-03
|
Stearoylcarnitine (C18)
|
0.31
|
0.02
|
Methionine
|
0.26
|
0.04
|
Cys-Gly
|
0.26
|
0.05
|
Arachidoylcarnitine (C20)*
|
0.26
|
0.05
|
N-palmitoylserine
|
0.25
|
0.05
|
fCysteine/fCystine
|
-0.29
|
0.03
|
fCystine
|
-0.30
|
0.02
|
Table 11
Correlations between the core metabolites and the other 71 analyzed metabolites.
Metabolites
|
Correlation Coefficient
|
P-value
|
Glu-Cys x Histidylglutamate
|
-0.01
|
0.92
|
Glu-Cys x Cinnamoylglycine
|
-0.06
|
0.63
|
Glu-Cys x Proline
|
-0.09
|
0.51
|
Glu-Cys x Adrenoylcarnitine (C22:4)*
|
0.01
|
0.95
|
Histidylglutamate x Cinnamoylglycine
|
-0.03
|
0.81
|
Histidylglutamate x Proline
|
-0.07
|
0.59
|
Histidylglutamate x Adrenoylcarnitine (C22:4)*
|
-0.06
|
0.65
|
Cinnamoylglycine x Proline
|
1.80E-03
|
0.99
|
Cinnamoylglycine x Adrenoylglycine (C22:4)*
|
-0.13
|
0.31
|
Proline x Adrenoylcarnitine (C22:4)*
|
0.04
|
0.74
|
Table 12
Correlation coefficients between the five core metabolite models from Table 9 that provide the highest accuracy.
Most of the metabolites listed in Tables 4 and 5 that were significantly different between the ASD and TD groups were found to be significantly correlated with the 5 core metabolites. However, there were 5 metabolites that were significantly different between the ASD-M and TD-M groups that did not significantly correlate with the 5 core metabolites. These five metabolites were B12, cis-4-decenoylcarnitine (C10:1), catechol sulfate, 7-methylxanthine, and tiglylcarnitine (C5:1-DC). A correlation analysis was conducted to determine if any of the 5 metabolites were correlated with one another, possibly forming a 6th group of correlated metabolites. However, none of the 5 metabolites were significantly correlated with one another. So, it appears that there are 5 primary sets of metabolites, and 5 additional metabolites that are not part of those 5 groups, which are significantly different between the ASD-M and TD-M groups.
Carnitine and Beef
Since the levels of carnitine-conjugated molecules were lower in the ASD-M group (see Table 7), and since beef is the primary dietary source of carnitine (some can also be made by the body), hypothesis testing was performed on the beef quantity and beef frequency in the mother’s diets to see if there was a difference between the two groups of mothers. The results are shown below.
Variable
|
Test
|
p-Value
|
FDR
|
AUC
|
Ratio (ASD-M/TD-M)
|
Beef Frequency
|
MW
|
0.73
|
1.00
|
0.53
|
1.12
|
Beef Quantity
|
MW
|
0.83
|
1.00
|
0.51
|
1.41
|
Table 13
Univariate hypothesis testing results for beef intake of mothers during pregnancy.
There was no significant difference found in the mean/median of the beef consumption frequency and quantity between the two groups. Also, the beef consumption frequency and quantity measurements did not significantly correlate with carnitine levels, except for a slight negative correlation of beef frequency and lignoceroylcarnitine (C24) (r= -0.26, p=0.05, unadjusted).
Overall, many of the metabolites measured in this study are significantly different between the two groups of mothers, ASD-M and TD-M. The subset of metabolites that worked the best for classification was a subset of five metabolites which were each correlated with many others.