Population structure analysis
Quality control of the genotypic data was performed by PLINK according to the specified criteria, and a total of 458,188 SNPs and 5,072 individuals were passed in the quality control and used for the final analysis. The genotypic data set was analyzed using PCA and ADMIXTURE [18]. PCA analysis showed that the top three PC explain 86.4% of total genetic variation (Fig. 1a, b). PC1 separated the EA (blue) and CA (green) groups (40.2% of the variation), and the PC2 separates the AA (red) and HA (purple) (34.4%), while PC3 separates CA from other subpopulations (11.8%) (Fig. 1a, b). We have used ADMIXTURE with estimated population size (K) as five, and the results are reasonably straightforward that K = 5 is a sensible modeling choice for MESA data (Fig. 1c, d).
Genetic and ethnicity-specific effects
The interaction network plot of quantitative trait SNPs (QTSs) is presented in Fig. 2, in which there were a total of five QTSs controlling calibrated FVIII would be detected by six different models. Five QTSs could be detected by directly analyzing FVIII by five models (0, 2, 3, 4, and 5), among them Q2-34 (rs13428770) and Q7-5 (rs1468382) on chromosome 2 and 7, respectively involving epistasis (additive × additive aa, dominance × additive da, and dominance × dominance dd), Q4-40 (rs17326624) and Q7-5 (rs1468382) on chromosome 4 and 7, respectively involving epistasis (additive × additive aa, ethnicity-specific additive × dominance ade1, and ade3). However, only Q2-34 (rs13428770), Q4-29 (rs12641227), Q4-40 (rs17326624), and Q7-5 (rs1468382) could be detectable under all six models, indicating that they are the key QTSs for FVIII (Fig. 2). One QTS Q7-44 (rs17864995) on chromosome 7 will be lost after considering fat as a cofactor in model 1. It is indicated that the genetic effects of this QTS are contributed by fat. The effects of additive × additive epistasis (aa) under models 2, 3, 4, and 5 and ethnicity-specific additive × dominance (ade3) effect for models 1-5 were observed between QTS Q4-40 and Q7-5. It is indicated that the nutrition cofactors mask the positive additive × dominance epistatic effects in AA ethnic group.
The QTS effects and QTS × ethnicity interaction effects of five SNPs are presented in the G × E plot (Fig. 3). For Q2-34, negative additive effects were observed for all models, whereas dominance effects were positive for all models; an ethnicity-specific additive effect was identified in CA and AA populations for models 0, 1, and 4. Only the ethnicity-specific additive effect was identified in the CA population for models 3 and 5, but no ethnicity effect was observed for model 2. For Q4-29, no additive effect was observed in model 1 but detected negative effects in the other five models (0, 2, 3, 4, and 5). All models in two ethnicities identified an ethnicity-specific additive effect, including a positive effect for the CA population and a negative for the AA population. All six models in three ethnicities identified the ethnicity-specific dominance effect: the positive effects for EA and HA populations and the negative effects for the AA population. For Q4-40, only positive dominance effects were identified for models 2, 3, 4, and 5 without ethnicity-specific effects (Fig. 3 and Additional file 2). For Q7-5, the positive additive effect for models 2, 3, 4, and 5 was identified, and ethnicity-specific negative additive effects were identified by five models (0, 1, 2, 3, and 5) in the EA population (Additional file 2). For Q7-44, positive additive effect but ethnicity-specific negative additive effect are identified by five models (0, 2, 3, 4, and 5) for the EA population (Additional file 2).
All models identified significant epistasis effects for Q2-34 × Q7-5 (positive effect for aa and dd, negative effect for da) without ethnicity-specific effect (Additional file 2). Negative epistatic effects (aa) were identified when setting animal protein, vegetable protein, alcohol, and energy as a cofactor in the model for epistasis Q4-40 × Q7-5 (Additional file 2). Also, ethnicity-specific additive × dominance negative effects have been identified for the EA population by all models. Interestingly, ethnicity-specific additive × dominance positive effect was observed when only considered nutrition as a cofactor in Q4-40 × Q7-5 for the AA population (Additional file 2).
The predicted effects of QTS and QTS × ethnicity interaction with high significance (−log10 (P-value) > 5.0) are presented in Table 1 for four individual QTSs and two pairs of epistatic QTSs. Negative additive effects but dominant positive effects were highly significant for Q2-34 for all the models (Table 1). Highly significant ethnicity-specific dominant effects of Q4-29 could be detected for two ethnicities (EA and AA populations) but in opposite patterns. Positive EA-specific dominant effect but negative AA-specific dominant were observed for all models (Table 1). Among all the significant QTSs, only the positive dominance effects of Q4-40 and positive additive effects if Q7-5 were highly significant and predictable for models 2, 3, 4, and 5 (Table 1). All models for Q2-34 × Q7-5 identified highly significant epistasis effects (positive for aa, negative for da). Also, highly significant negative epistatic effects (aa) between Q4-40 and Q7-5 were identified for models 2, 3, 4, and 5 (Table 1).
Table 1 Predicted QTS effects and QTS × ethnicity interaction effects with high significance and heritability for FVIII under the different model setting
ID
|
Chr_QTS_SNP
|
Gene
|
Effect
|
Model
|
Predict
|
-log10(P)
|
h2 (%)
|
2-34
|
2_rs13428770_C_G
|
24 kb 3' of STK25
|
a
|
0
|
-12.86
|
90.09
|
13.67
|
1
|
-12.74
|
88.40
|
13.64
|
2
|
-12.80
|
89.30
|
10.07
|
3
|
-12.86
|
90.17
|
10.09
|
4
|
-12.91
|
91.44
|
10.26
|
5
|
-12.75
|
88.71
|
9.9
|
d
|
0
|
8.60
|
33.47
|
3.06
|
1
|
8.54
|
32.96
|
3.06
|
2
|
8.58
|
33.33
|
2.26
|
3
|
8.50
|
32.75
|
2.21
|
4
|
8.38
|
32.01
|
2.16
|
5
|
8.54
|
33.02
|
2.22
|
4-29
|
4_rs12641227_G_T
|
LOC105377567
|
de1
|
0
|
5.55
|
6.29
|
1.28
|
1
|
5.18
|
5.53
|
1.12
|
2
|
5.62
|
6.43
|
0.97
|
3
|
5.70
|
6.59
|
0.99
|
4
|
6.55
|
8.53
|
1.32
|
5
|
5.78
|
6.77
|
1.02
|
de3
|
0
|
-10.78
|
7.63
|
4.81
|
1
|
-10.94
|
7.83
|
5.03
|
2
|
-10.71
|
7.54
|
3.52
|
3
|
-10.76
|
7.60
|
3.53
|
4
|
-11.17
|
8.18
|
3.84
|
5
|
-10.77
|
7.61
|
3.53
|
4-40
|
4_rs17326624_T_C
|
SCRG1
|
d
|
2|0
|
22.86
|
83.41
|
16.06
|
3|0
|
22.73
|
82.46
|
15.75
|
4|0
|
22.35
|
80.30
|
15.36
|
5|0
|
22.79
|
82.97
|
15.82
|
7-5
|
7_rs1468382_G_A
|
LOC107986841
|
a
|
2|0
|
7.86
|
43.80
|
3.8
|
3|0
|
7.98
|
45.07
|
3.88
|
4|0
|
7.93
|
44.79
|
3.86
|
5|0
|
8.09
|
46.33
|
3.99
|
2-34/7-5
|
2_rs13428770_C_G/
7_rs1468382_G_A
|
STK25/ LOC107986841
|
aa
|
0
|
11.06
|
56.88
|
20.24
|
1
|
10.89
|
55.13
|
19.94
|
2
|
11.00
|
56.26
|
14.87
|
3
|
11.06
|
56.85
|
14.92
|
4
|
10.63
|
52.97
|
13.9
|
5
|
10.97
|
56.00
|
14.66
|
da
|
0
|
-10.51
|
27.03
|
9.14
|
1
|
-10.33
|
26.11
|
8.96
|
2
|
-10.51
|
27.03
|
6.79
|
3
|
-10.41
|
26.51
|
6.6
|
4
|
-10.17
|
25.51
|
6.36
|
5
|
-10.39
|
26.45
|
6.58
|
4-40/7-5
|
4_rs17326624_T_C/
7_rs1468382_G_A
|
SCRG1/ LOC107986841
|
aa
|
2|0
|
-7.18
|
30.37
|
6.34
|
3|0
|
-7.31
|
31.42
|
6.52
|
4|0
|
-7.69
|
34.87
|
7.27
|
5|0
|
-7.52
|
33.20
|
6.89
|
Notes: Environment (Ethnicity) defined as in Fig. 2. Genetic effect: a = additive effect, d = dominance effect, aa = additive × additive epistasis effect, dd = dominance × dominance epistasis effect, da = dominance × additive epistasis effect; ae1 = EA-specific additive effect, ae3 = AA-specific additive effect, ae4 = HA-specific additive effect, aae3 = AA-specific additive × additive epistasis effect, dae2 = CA-specific dominance × additive epistasis effect. −log10 (P) = minus log10 (P−value). h2 (%) = heritability (%). Model: 0 = detectable with no cofactor, 1 = detectable with fat as cofactor, 2 = detectable with animal protein as cofactor, 3 = detectable with vegetable protein as cofactor, 4 = detectable with alcohol as cofactor; 5 = detectable with energy as cofactor; x|0 = detectable only with x as cofactor, and not detectable with no cofactor.
Positive AA-specific additive effect of Q2-34 was observable without setting cofactor, and this QTS effect was absent when considering the nutrition of animal protein, vegetable protein, and energy (Additional file 3). It is indicated that the AA-specific additive effect could be due to animal protein, vegetable protein, and energy. A negative additive effect was observable for the Q4-29, and this QTS effect was absent when including the fat as a cofactor, indicating that this effect could be due to the fat (Additional file 3). The EA-specific additive effect of Q7-5 was negative, and this QTS effect was absent when including the alcohol as a cofactor in the model, indicating that the effect of Q7-5 might be due to the alcohol. Moreover, a positive additive effect (a = 2.00) and EA-specific additive effect was identified, and the QTS Q7-44 effect was absent when the model considered fat as a cofactor. This effect might be happening due to the fat (Additional file 3). Some kinds of nutrition could also suppress the genetic effects of QTSs on the FVIII level. A positive dominant effect of Q4-40 and the positive additive effect of Q7-5 could be inhibited by animal protein, vegetable protein, alcohol, and energy (Additional file 4). Also, animal protein, vegetable protein, alcohol, and energy can suppress the epistasis effects of Q4-40 × Q7-5 (negative effect for aa and positive effect for ade3).
Heritability results by all models
Estimated heritability was listed in Table 2 for six models. For model 0, the total heritability was primarily due to additive effects and epistasis effects. When including fat as a cofactor in model 1, the total heritability was only decreased indicating that some gene expression will be lost when removing the effect of fat as a cofactor. But the total heritability was increased in models 2, 3, 4 and 5 as compared with the model with no cofactor. It could be due to the exposure of significant dominant effect of Q4-40 (rs17326624), an additive effect of Q7-5 (rs1468382), epistatic effects (aa), and ethnicity-specific additive × dominance epistatic effect (ade3) between Q4-40 and Q7-5 after accounting for fat, animal protein, vegetable protein, alcohol and energy effects in model 2, 3, 4 and 5, respectively.
Table 2 Estimated heritability (%) for FVIII under different model settings
Model
|
|
|
|
|
|
|
|
0
|
14.33
|
3.06
|
30.01
|
3.06
|
2.29
|
0.95
|
53.70
|
1
|
13.64
|
3.06
|
29.55
|
2.60
|
2.31
|
1.75
|
52.91
|
2
|
14.32
|
18.32
|
28.46
|
1.59
|
1.71
|
1.19
|
65.59
|
3
|
14.42
|
17.95
|
28.50
|
2.01
|
1.71
|
1.26
|
65.85
|
4
|
14.50
|
17.52
|
28.00
|
2.57
|
1.96
|
1.23
|
65.78
|
5
|
14.34
|
18.04
|
28.60
|
1.94
|
1.72
|
1.28
|
65.92
|
Notes: Heritability: = heritability for additive effects; = heritability for dominance effects; = heritability for epistasis effects including AA, AD, DA, DD; = heritability for ethnicity-specific additive interaction effects; = heritability for ethnicity-specific dominance effects; = heritability for ethnicity-specific epistasis effects; = total heritability. Model: 0 = with no cofactor, 1 = with fat as cofactor, 2 = with animal protein as cofactor, 3 = with vegetable protein as cofactor, 4 = with alcohol as cofactor; and 5 = with energy as cofactor.
Bioinformatics and network analysis of the targeted genes of the QTSs
NetworkAnalyst [21] showed that 22 protein-coding genes were associated with STK25 (Fig. 4a). Gene ontology (GO) of these genes was carried out using the PANTHER database (http://pantherdb.org). PPI network generated by the STRING database (http://string-db.org) for the identified genes, and the results showed that these genes were associated with binding (GO: 0005488), molecular function regulator (GO: 0098772), and catalytic activity (GO: 0003824) in the molecular function category (Additional file 7a). Most of the genes were engaged in the cellular process (GO: 0009987), cellular component organization or biogenesis (GO: 0071840), biological regulation (GO: 0065007), and metabolic process (GO: 0008152) in the biological process category (Additional file 7b). In the cellular component category, most of the protein-coding genes were involved in a cell (GO: 0005623) and cell part (GO: 0044464) (Additional file 7c). Pathway analysis showed that most genes are involved in the FGF signaling pathway (P00021) (Additional file 7d). These results indicated that diverse GO terms and pathways were associated with FVIII and nutrients.
One of the candidate genes, GRM8, was associated with several diseases such as alcohol withdrawal delirium, alcoholic intoxication, anxiety disorders and small cell carcinoma of the lung, which might have a relationship between different nutrients and FVIII (Fig. 4b). The gene-TFs network analysis showed that one gene, STK25, interacted with 28 TFs, most involved in diverse pathways. Notably, Zinc finger protein (ZFP64, ZNF423, ZNF341), DNA methyltransferase 1-associated protein 1 (DMAP1), DNA-binding protein (RFXANK), cell division cycle 5-like protein (CDC5L), E3 ubiquitin-protein ligase (TRIM22), hypermethylated in cancer 1 protein (HIC1), and transcriptional repressor p66-alpha (GATAD2A) were some of them (Fig. 4c). Results of the gene-miRNA show that the 52 miRNA are involved with SCRG1 (Fig. 4d). We used the miRDB database (http://mirdb.org/) to extrapolate these miRNA functions [22]. We found that 15 miRNAs were in association with different genes. These miRNAs were involved in immunity-related GTPase Q, transmembrane protein 30B, serine protease 16, glutamate receptor-interacting protein, and SWI/SNF-related, matrix-associated actin-dependent regulator of chromatin, subfamily a, containing DEAD/H box 1 (Additional file 5).
We found only STK25 and GRM8 genes associated with different TF-miRNA, where a noticeable number of TFs are on the basic helix-loop-helix transcription factor and C2H2 zinc finger transcription factor (Fig. 4d and Additional file 6a). Besides, 14 miRNAs were associated with these genes. Importantly, LIM zinc finger domain containing 1, ubiquitin-conjugating enzyme E2 H, methyl-CpG binding domain protein 2, forkhead box N3, cytoplasmic linker associated protein 1, F-box protein 34, apolipoprotein O like, polypeptide N-acetylgalactosaminyltransferase 3 were involved with FVIII and nutrients in this studies (Fig. 4e and Additional file 6b). Moreover, protein-chemicals interaction showed that GRM8 and SCRG1 were associated with different chemicals, playing essential roles in regulating FVIII and nutrients (Fig. 4f).
We also investigated the expression of the identified genes using the human protein atlas database (https://www.proteinatlas.org/). Blood cell-specific expression analysis showed that only STK25 and SCRG1 gene expression were found in different blood cell type lineages (Additional file 8). Expression results showed that both genes were expressed in diverse blood cells, including NK-cell, T-reg, Basophil, Eosinophil, and Neutrophil. Notably, the highest or lowest expression was found in NK-cell and neutrophil cells, respectively, for the STK25 gene. Interestingly, the highest expression level was found in Neutrophil cells for the SCRG1 gene (Additional file 8).