Skip to content

S-LDSC DBP Analysis

I applied Stratified Linkage Disequilibrium Score Regression (S-LDSC)1 to summary statistics from Keaton et al.'s meta-GWAS of blood pressure2. Here I focused on diastolic blood pressure (DBP).

Reference Data Sources

I used the standard reference datasets recommended and preprocessed by the authors of the S-LDSC method.

Results

GTEx and Franke lab tissue expression data

Surprisingly, no cell types were significant using this reference dataset under a false discovery rate of 0.01.

The cell types with the lowest p values are given below:

Name Coefficient Coefficient_P_value Reject Null
A05.360.319.679.690.Myometrium 8.9244e-09 0.000945108 False
A05.810.453.Kidney 8.58043e-09 0.00118444 False
A05.360.319.679.Uterus 1.0349e-08 0.00173672 False
Heart_Atrial_Appendage 6.83951e-09 0.0030787 False
A03.556.875.875.Stomach 1.05254e-08 0.00345681 False
Heart_Left_Ventricle 7.51812e-09 0.00353501 False
Artery_Tibial 8.10254e-09 0.00454901 False
A03.556.124.684.Intestine..Small 7.02051e-09 0.0076707 False

The failure to find a significant cell type despite the very large size of the Keaton et al. meta-GWAS is surprising. I discuss this more below.

Looking at the top non-significant cell types, the presence of arterial and heart-related cell types as contributors to blood pressure is unsurprising. What is more surprising is the presence of reproductive tissue at low p values: myometrium and uterus. There are at least two potential explanations:

  1. Siricilla et al.3 state that "RNA sequencing also revealed that the myometrial and vascular SMCs [smooth muscle cells] were more than 90% similar in their transcriptome". Thus, one plausible explanation is that S-LDSC is picking up on associations of blood pressure with transcriptional programs that are active in vascular smooth muscle, and these transcriptional programs are similar to those of myometrial smooth muscle. This is biologically plausible: the importance of vascular smooth muscle to blood pressure is well known.
  2. S-LDSC may be picking up on a sex hormone effect on blood pressure. Blood pressure is well known to be sexually dimorphic.

Roadmap Chromatin data

I next applied S-LDSC to the DBP GWAS using reference data generated by Finucane et al.1 from the Roadmap Epigenetics Project.

The following graph and table show the results:

Name Coefficient Coefficient_P_value Reject Null
Right_Atrium__H3K4me1 1.35454e-07 1.40926e-07 True
Right_Ventricle__H3K4me3 4.75769e-07 1.47786e-07 True
Right_Atrium__H3K27ac 2.27715e-07 2.29194e-07 True
Aorta__H3K27ac 1.73281e-07 1.31198e-06 True
Left_Ventricle__H3K27ac 1.44065e-07 1.37252e-06 True
Fetal_Kidney__DNase 2.02354e-07 3.77236e-06 True
Right_Atrium__H3K4me3 4.0257e-07 4.91503e-06 True
Left_Ventricle__H3K4me1 1.20989e-07 4.97588e-06 True
Fetal_Lung__H3K4me1 1.0149e-07 6.22595e-06 True
Lung__H3K27ac 1.73477e-07 7.23294e-06 True
Fetal_Adrenal_Gland__H3K4me3 5.20417e-07 9.76305e-06 True
Adipose_Nuclei__H3K4me1 9.18524e-08 9.86468e-06 True
Colonic_Mucosa__H3K4me1 1.73559e-07 1.00295e-05 True
Fetal_Stomach__DNase 2.56155e-07 1.10398e-05 True
Lung__H3K4me1 1.51356e-07 1.18822e-05 True
Right_Ventricle__H3K4me1 1.22013e-07 1.3042e-05 True
Fetal_Stomach__H3K27ac 1.08069e-07 1.57813e-05 True
Adipose_Nuclei__H3K27ac 8.34657e-08 3.44702e-05 True
Aorta__H3K4me1 2.50867e-07 3.54758e-05 True
Right_Ventricle__H3K27ac 1.70317e-07 5.12232e-05 True
Ovary__H3K4me3 3.5071e-07 5.38539e-05 True
Stomach_Smooth_Muscle__H3K4me3 2.211e-07 6.47863e-05 True
Left_Ventricle__H3K4me3 4.85369e-07 7.96825e-05 True
Fetal_Lung__DNase 1.62032e-07 8.89937e-05 True
Fetal_Stomach__H3K4me3 4.01725e-07 9.7844e-05 True
Fetal_Stomach__H3K4me1 7.13326e-08 0.00010137 True
Duodenum_Smooth_Muscle__H3K4me1 1.41764e-07 0.00012871 True
Fetal_Kidney__H3K9ac 4.31301e-07 0.000141293 True
Skeletal_Muscle_Male__H3K4me1 7.46959e-08 0.000153192 True
Fetal_Heart__H3K9ac 1.47374e-07 0.000180017 True
Skeletal_Muscle_Male__H3K9ac 1.40079e-07 0.000193951 True
Spleen__H3K4me1 1.03075e-07 0.000282654 True
Aorta__H3K4me3 5.18442e-07 0.000283657 True
Skeletal_Muscle_Male__H3K4me3 2.48575e-07 0.000312901 True
Fetal_Muscle_Leg__H3K4me3 4.9079e-07 0.000318039 True
Fetal_Adrenal_Gland__H3K36me3 9.39596e-08 0.0006646 True
Small_Intestine__H3K27ac 1.42644e-07 0.000743932 True
Fetal_Adrenal_Gland__H3K4me1 1.42379e-07 0.00108434 False
Colonic_Mucosa__H3K4me3 2.74248e-07 0.00125257 False
Psoas_Muscle__H3K4me3 2.92573e-07 0.00125564 False
Fetal_Muscle_Leg__H3K27ac 8.85931e-08 0.00157292 False
Fetal_Adrenal_Gland__H3K27ac 1.40667e-07 0.00162927 False
Foreskin_Fibroblast_Primary_Cells_skin02__H3K4me3 1.35138e-07 0.00165603 False
Ovary__H3K27ac 1.13578e-07 0.0017744 False
Skeletal_Muscle_Female__H3K4me3 1.65886e-07 0.00189594 False
Rectal_Smooth_Muscle__H3K4me3 3.193e-07 0.00209098 False
Adipose_Nuclei__H3K9ac 1.45823e-07 0.00211634 False
Fetal_Stomach__H3K36me3 4.77451e-08 0.00236169 False

chromatin-dbp-sldsc

The contrast with the previous section is striking. Whereas using gene expression reference data produced no significant cell types, using epigenetic reference data produces a wide range of significant cell types across multiple categories.

The most significant cell types are heart, blood vessel, or kidney related, which is biologically plausible.

ImmGen data

No cell types were significant from the ImmGen data, consistent with diastolic blood pressure not being a primarily immune condition.

Name Coefficient Coefficient_P_value Reject Null
DN.SLN.CFA.d6.v2 9.12756e-09 0.000235668 False
DN.SLN.v2 8.84128e-09 0.000593548 False
BEC.SLN.OT 1.09635e-08 0.000885477 False
BEC.SLN 1.02689e-08 0.00217312 False
B.T1.Sp 7.36352e-09 0.00286334 False
B.Mem.Sp.v2 8.22723e-09 0.00413403 False
BEC.MLN 9.61318e-09 0.00437598 False
B.FrE.BM 7.62703e-09 0.00465458 False
Fi.MTS15+.Th 6.49713e-09 0.00470782 False
St.31-38-44-.SLN 7.1822e-09 0.00508416 False
LEC.SLN.CFA.d6.v2 6.93733e-09 0.00588418 False
SC.LT34F.BM 9.02238e-09 0.00704497 False
B.Fo.LN 7.54089e-09 0.00963479 False
B.FrE.FL 7.46381e-09 0.014395 False
SC.LTSL.FL 8.5851e-09 0.0175241 False
T.4int8+.Th 4.59416e-09 0.0265722 False

Corces et al. ATAC-seq data

Again, as expected, there were no significant hematopoietic-related cell types in the Corces ATAC-seq dataset.

Name Coefficient Coefficient_P_value Reject Null
HSC 4.74203e-08 0.10657 False
Erythro 6.07373e-08 0.109014 False
MPP 3.33285e-08 0.173504 False
CMP 2.92056e-08 0.200355 False
MEP 3.31256e-08 0.24754 False
CLP -1.67977e-09 0.516239 False
LMPP -1.89293e-09 0.518985 False
Bcell -6.86252e-09 0.579055 False
GMP -7.54583e-09 0.592731 False
NK -4.80328e-08 0.938961 False
Mono -1.13595e-07 0.985252 False
CD8 -1.15941e-07 0.996904 False
CD4 -1.27186e-07 0.999554 False

Cahoy and GTEx-Brain data

There were no significant neurological cell types. One might expect to find a significant neurological cell type, given how blood pressure is tightly controlled by the nervous system. The failure to find this may reflect a limitation of the present reference dataset, or of the S-LDSC method itself.

Name Coefficient Coefficient_P_value Reject Null
Neuron 3.72999e-09 0.067606 False
Oligodendrocyte -2.21628e-09 0.815887 False
Astrocyte -3.53281e-09 0.933489 False
Name Coefficient Coefficient_P_value Reject Null
Brain_Cerebellum 2.85144e-09 0.13386 False
Brain_Cerebellar_Hemisphere 2.4761e-09 0.156651 False
Brain_Putamen_(basal_ganglia) -2.62188e-11 0.504228 False
Brain_Frontal_Cortex_(BA9) -1.16454e-10 0.523032 False
Brain_Substantia_nigra -3.23304e-10 0.55205 False
Brain_Amygdala -4.74749e-10 0.594472 False
Brain_Anterior_cingulate_cortex_(BA24) -1.66269e-09 0.809254 False
Brain_Caudate_(basal_ganglia) -2.25646e-09 0.819346 False
Brain_Nucleus_accumbens_(basal_ganglia) -2.07221e-09 0.839876 False
Brain_Hippocampus -2.18755e-09 0.855084 False
Brain_Cortex -2.1191e-09 0.876525 False
Brain_Hypothalamus -2.26791e-09 0.904322 False
Brain_Spinal_cord_(cervical_c-1) -5.35052e-09 0.994565 False

Comment on contrast between GTEx/Franke Lab dataset and Roadmap dataset results

As noted above, it is striking that using the GTEx/Franke lab dataset produces no S-LDSC hits for diastolic blood pressure, while using the Roadmap epigenetics dataset produces many hits.

This contrast implies that:

  • A) The DBP GWAS identified signal in regions of the genome that are epigenetically differentially regulated in the heart, blood-vessels, kidney, etc.
  • B) These GWAS-signal-enriched regions of the genome do not lie close to genes that are strongly differentially expressed in the heart, blood-vessels, kidney, etc.

I see two interpretations:

  • The genetic regulation of blood pressure is sufficiently complex that the genetic variants involved in BP control cannot be identified by simply taking a 100-kb window around key genes (which is how the GTEx/Franke lab dataset is generated). The use of epigenetics allows a much more precise identification of the key regulatory regions.
  • Bulk RNAseq is too crude a tool: the differences in bulk gene expression levels in the key cell types may be too small or too variable for S-LDSC to pick up. It would be interesting to test this interpretation by using a single-cell RNAseq dataset with S-LDSC.

How to reproduce

Use this script to reproduce the above analysis.

  1. Hilary K Finucane, Yakir A Reshef, Verneri Anttila, Kamil Slowikowski, Alexander Gusev, Andrea Byrnes, Steven Gazal, Po-Ru Loh, Caleb Lareau, Noam Shoresh, and others. Heritability enrichment of specifically expressed genes identifies disease-relevant tissues and cell types. Nature Genetics, 50(4):621–629, 2018. URL: https://pmc.ncbi.nlm.nih.gov/articles/PMC5896795/. ↩↩

  2. Jacob M Keaton, Zoha Kamali, Tian Xie, Ahmad Vaez, Ariel Williams, Slavina B Goleva, Alireza Ani, Evangelos Evangelou, Jacklyn N Hellwege, Loic Yengo, and others. Genome-wide analysis in over 1 million individuals of European ancestry yields improved polygenic risk scores for blood pressure traits. Nature Genetics, 56(5):778–791, 2024. URL: https://pmc.ncbi.nlm.nih.gov/articles/PMC11096100/. ↩

  3. Shajila Siricilla, Kelsi M Knapp, Jackson H Rogers, Courtney Berger, Elaine L Shelton, Dehui Mi, Paige Vinson, Jennifer Condon, Bibhash C Paria, Jeff Reese, and others. Comparative analysis of myometrial and vascular smooth muscle cells to determine optimal cells for use in drug discovery. Pharmacological Research, 146:104268, 2019. URL: https://pmc.ncbi.nlm.nih.gov/articles/PMC6889064/. ↩