Identification of an Allele-Specific Transcription Factor Binding Interaction that May Regulate PLA2G2A Gene Expression

Abstract

The secreted phospholipase A₂ (sPLA₂) isoform, sPLA₂-IIA, has been implicated in a variety of diseases and conditions, including bacteremia, cardiovascular disease, COVID-19, sepsis, adult respiratory distress syndrome, and certain cancers. Given its significant role in these conditions, understanding the regulatory mechanisms impacting its levels is crucial. Genome-wide association studies (GWAS) have identified several single nucleotide polymorphisms (SNPs), including rs11573156, that are associated with circulating levels of sPLA₂-IIA. The work in the manuscript leveraged 4 publicly available datasets to investigate the mechanism by which rs11573156 influences sPLA₂-IIA levels via bioinformatics and modeling analysis. Through genotype-tissue expression (GTEx), 234 expression quantitative trait loci (eQTLs) were identified for the gene that encodes for sPLA₂-IIA, PLA2G2A. SNP2TFBS was used to ascertain the binding affinities between transcription factors (TFs) to both the reference and alternative alleles of identified eQTL SNPs. Subsequently, candidate TF-SNP interactions were cross-referenced with the ChIP-seq results in matched tissues from ENCODE. SP1-rs11573156 emerged as the significant TF-SNP pair in the liver. Further analysis revealed that the upregulation of PLA2G2A transcript levels through the rs11573156 variant was likely affected by tissue SP1 protein levels. Using an ordinary differential equation based on Michaelis-Menten kinetic assumptions, we modeled the dependence of PLA2G2A transcription on SP1 protein levels, incorporating the SNP influence. Collectively, our analysis strongly suggests that the difference in the binding dynamics of SP1 to different rs11573156 alleles may underlie the allele-specific PLA2G2A expression in different tissues, a mechanistic model that awaits future direct experimental validation. This mechanism likely contributes to the variation in circulating sPLA₂-IIA protein levels in the human population, with implications for a wide range of human diseases.

Introduction

Members of the secreted phospholipase A₂ (sPLA₂) family of enzymes play important roles in numerous biological processes, chiefly by cleaving the sn-2 position of glycerophospholipids. This hydrolysis results in an unsaturated fatty acid (UFA) and a 2-lyso-phospholipid. In humans, the 11 isoforms (PLA₂-IB, PLA₂-IIA, PLA₂-IIC, PLA₂-IID, PLA₂-IIE, PLA₂-IIF, PLA₂-III, PLA₂-V, PLA₂-X, PLA₂-XIIA, and PLA₂-XIIB) have distinct cellular distributions, enzymatic specificities, and preferences for different polar head groups and UFAs at the sn-3 and sn-2 positions, respectively.^1-4 Collectively, sPLA₂ isoforms participate in multiple physiological and pathophysiological activities, generating proinflammatory and anti-inflammatory signaling molecules. Their varied functions also encompass remodeling of membrane phospholipids, food phospholipid degradation, and the hydrolysis of bacterial membranes. Notably, the sPLA₂-IIA isoform, encoded by the PLA2G2A gene, efficiently hydrolyzes bacterial membrane phospholipids. This is especially the case for Gram-positive bacteria, where sPLA₂-IIA, even at minimal concentrations, plays an essential antimicrobial role in innate immunity.^5-8

Furthermore, circulating sPLA₂-IIA levels can be reliably measured and have been identified as key biomarkers indicative of the severity or susceptibility linked with cardiovascular disease, ⁹ rheumatoid arthritis,^10,11 cancer,^12-15 sepsis, ¹⁶ and viral infections including human hepatitis virus B ¹⁷ and severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). ¹⁸ Notably, experimental and clinical evidence show that sPLA₂-IIA plays a critical role in hypotension and multiple organ failure observed in sepsis, COVID-19, and respiratory distress syndrome.^19-21 While numerous genetic variants influencing mRNA levels or blood sPLA₂-IIA protein levels have been identified, the precise molecular mechanism by which these variants impact sPLA₂-IIA levels remains to be elucidated. Therefore, gaining insight into the genetic and molecular mechanisms that regulate circulating and tissue sPLA₂-IIA levels is crucial, as this knowledge can identify populations and individuals vulnerable to either reduced or excessive production of this pivotal enzyme.

The single nucleotide polymorphism (SNP) rs11573156, where G is defined as the reference allele, and C is defined to be an alternative allele in the coding strand sequence, has been strongly associated with circulating levels of sPLA₂-IIA. This association is supported by 5 studies that reveal a dose-dependent allele association between circulating levels and activity. ²² For instance, sPLA₂-IIA levels increased by 57% to 62% in those with the homozygous alternative allele (CC) compared to the homozygous reference (GG).

Recognizing the significance of SNPs like rs11573156, our study aimed to understand the genetic and molecular foundations of PLA2G2A expression. This research developed an analysis pipeline (outlined in Supplementary Figure 1), by using data from 4 publicly available resources (genotype-tissue expression [GTEx], SNP2TFBS, ENCODE, and Human Protein Atlas [HPA]) and mathematical modeling to explore how rs11573156 modulates sPLA₂-IIA levels. More specifically, we used these datasets to examine transcription factor (TF) binding sites where PLA2G2A-eQTLs are located and how different alleles at rs11573156 influence TF-binding affinity. A simple ODE model was constructed, assuming Michaelis-Menten kinetics, to simulate how varying SP1-rs11573156 binding affects allele-specific PLA2G2A transcription. This model was tested with data of SP1 protein level and PLA2G2A expression across various tissues, as obtained from HPA and GTEx, respectively. Our findings suggest that rs11573156 affects PLA2G2A transcription via the differential binding affinities of the 2 alleles to SP1, which likely constitutes an important genetic/molecular mechanism affecting circulating sPLA₂-IIA levels.

Materials and Methods

Collection of PLA2G2A eQTLs in different tissues

The GTEx Portal V8 (https://gtexportal.org) ²³ was used to determine PLA2G2A expression quantitative trait loci (eQTLs) across 18 tissues including thyroid, testis, stomach, skin-sun exposed (lower leg), skin-not sun exposed (suprapubic), prostate, ovary, nerve-tibial, muscle-skeletal, lung, liver, heart-left ventricle, heart-atrial appendage, esophagus-muscularis, esophagus-gastroesophageal junction, artery-aorta, adipose-visceral (omentum), and adipose-subcutaneous. The GTEx includes data measured in tissues collected within 24 hours of death for 948 individuals (32.9% female in total; 84.6% white with 34.3% of them female, 12.9% African American with 30.5% female, 1.3% Asian with 41.7% female, and 1.1% unknown with 16.7% female). The genotype data based on whole genome sequencing from 838 donors were applied to eQTL analyses in the GTEx V8, where only variants with MAF ⩾1% across all 838 samples were included. A significant eQTL was determined by a false discovery rate (FDR) threshold of ⩽0.05. The normalized effect size (NES) of the eQTLs was defined as the slope of the linear regression model between genotype and expression. The NES was calculated as the effect of the alternative allele relative to the reference allele in the human genome reference GRCh38/ hg38. A more detailed description is on the GTEs Portal site (https://gtexportal.org/home/methods). A positive NES indicates upregulation, while a negative NES indicates downregulation of the affected gene.

TF binding to eQTL alleles

The SNP2TFBS tool (http://ccg.vital-it.ch/snp2tfbs/) was used to predict TF-eQTL SNP binding variations. SNP2TFBS is a comprehensive database of regulatory SNPs affecting predicted transcription factor binding site (TFBS) affinity. ²⁴ An SNP’s effect on TF binding is estimated based on a position weight matrix (PWM) model for the binding specificity of the corresponding factor, which are fixed length TFBS models represented by a matrix of probabilities reflecting the occurrence frequencies of bases at binding site positions. SNP2TFBS uses the JASPAR database (http://jaspar.genereg.net) as a source of the TF-binding profiles, which provides PWMs as base frequency matrices. SNP2TFBS reports raw binding scores, which are computed as the sum of the position-specific weights over all bases of the binding site. It uses SNP data from the 1000 Genomes Project (minor allele frequency > 0.001), human genome assembly GRCh37/hg19, and RefSeq gene annotations provided by ANNOVAR. ²⁵ In this study, eQTLs affecting PLA2G2A gene expression were queried with SNP2TFBS to obtain a list of predicted TFs that differentially bind to those SNPs.

Verification of the predicted TF-eQTL binding by ChIP-seq dataset

ChIP-seq Peaks (340 TFs in 129 cell and tissue types) from ENCODE 3, downloaded from the UCSC Genome Browser (https://genome.ucsc.edu/) alignment to GRCh38/hg38, were used to examine whether the candidate TFs bind to the predicted eQTL regions. This dataset represents a large-scale TF-binding profile based on ChIP-seq experiments generated by the ENCODE Consortium between February 2011 and November 2018.^26,27 If candidate TFs had SNP2TFBS-predicted differential binding to PLA2G2A SNPs, we checked whether the predicted TF binding occurred with the eQTL-corresponding regions with the PLA2G2A gene locus (chromosome 1; 18990581-20971970) in the matched tissue.

Gene expression and genotyping data

The RNA-sequencing (RNA-seq) data (TPMs) were obtained from the GTEx portal open access database (https://gtexportal.org) on April 21, 2022. The genotyping data were obtained per dbGaP approval and downloaded as dbGaP accession number phg001219.v1 on October 02, 2022.

$\mathrm{Delta}$ log TPM values used in Figure 4 were computed by (1) applying a log (TPM + 1) transformation to all raw TPM values retrieved from GTEx, (2) computing the mean of transformed values from step 1, grouping on both tissue and allele, and (3) subtracting the GG/GC from the CC values from step 2, grouping on tissue.

Figure 4.

The impact of SP1 protein levels on allele-specific PLA2G2A gene expression across tissues. (a) Modeled PLA2G2A allele expression. In the upper panel, the ODE across an arbitrary range of SP1 level ([SP1]) (x-axis) was simulated to get the steady-state PLA2G2A expression level ([G2A]_ss) corresponding to the C and G alleles of rs11573156 (K_c and K_G, respectively). In the lower panel, $\Delta {\left[G2A\right]}_{ss}$ (y-axis) indicates the difference in the C and G allele PLA2G2A expression levels, which is box-plotted against [SP1] (x-axis) spanning 3 categories (with 5 bins in each category, as indicated in the top panel). The model parameters are $range\left(\left[SP1\right]\right)=\left[0,20\right],$ $k=1;d_{G2A}=0.01;n=2;K_C=1;K_G=2.$ (b) Empirical delta PLA2G2A allele expression. The differences (delta logTPM) of PLA2G2A transcript levels by genotype (C vs G allele at rs11573156) were calculated in each tissue based on the GTEx dataset and box-plotted against the 3 categories of SP1 protein levels (low, medium, and high) as in a. (c) Normalized effect size of rs11573156 in each tissue box-plotted against the 3 categories of SP1 protein levels as in a.

SP1 protein level in different tissues

SP1 protein expression data by tissue are publicly available from the HPA (https://www.proteinatlas.org/about/download). Expression values are semiquantitative, derived from immunohistochemistry of tissue microarrays, and take on 4 possible values: “Not Detected,” “Low,” “Medium,” and “High.” Tissues not present in GTEx were omitted. When a tissue had multiple recorded cell types, the protein expression profile in the cell type that exhibits the largest cellular population within the tissue was selected.

Modeling SP1 level-dependent allele-specific PLA2G2A gene expression in tissues

A simple ordinary differential equation (ODE) model was developed to simulate and test the correlation between SP1-eQTL binding and PLA2G2A expression in different tissues:

$$\begin{array}{c}\frac{d\left[G2A\right]}{dt}=\frac{k{\left[SP1\right]}^n}{K^n+{\left[SP1\right]}^n}-d_{G2A}\left[G2A\right]\#\end{array}$$ (1)

The structure of the equation was based on standard Hill function kinetics. $\frac{d[G2A]}{dt}$ describes the rate of the change of PLA2G2A transcript concentration over time. $k$ is a rate constant describing SP1-dependent PLA2G2A transcription. $K,$ with the specific instances of K_C and K_G for CC and GG/GC alleles, respectively, is a Hill constant inversely correlated with the eQTL allele-specific binding affinity to SP1. $n$ is a Hill coefficient. $d_{G2A}$ is the degradation rate constant of PLA2G2A transcripts. The steady-state concentration of PLA2G2A expression is given by the following equation:

$$\begin{array}{c}{\left[G2A\right]}_{ss}=\frac{k{\left[SP1\right]}^n}{d_{G2A}\left(K^n+{\left[SP1\right]}^n\right)}\#\end{array}$$ (2)

To explore qualitative insights into the system’s general behavior, simple, generic parameter values were selected as follows: $k$  = 1, $d_{\left[G2A\right]}=$ 0.01, $n$  = 2, $K_C$  = 1. With this base parameter set, and with K_G > K_C, reflecting the higher binding affinity of SP1 to the CC versus GG/GC allele of the PLA2G2A promoter, consistent qualitative patterns are observed (Supplementary Figure 2). PLA2G2A steady-state concentrations were simulated over a range of SP1 protein concentrations from 0 to 20 arbitrary units.

Results

eQTLs of PLA2G2A in human tissues

To investigate the potential genetic basis for variations in PLA2G2A expression in humans, the relationship between the SNPs in the human genome and PLA2G2A expression levels were examined. Specifically, we leveraged data from the GTEx project, which correlates genome-wide SNPs with variations in gene expression across multiple human tissues. Our analysis identified 234 eQTLs significantly associated with PLA2G2A expression (with an FDR ⩽ 0.05) in GTEx (V8) (Figure 1a). Located on chromosome 1 (hg38; chr1: 19975431-19980416), these eQTLs primarily occupy noncoding regions proximal to and flanking the PLA2G2A gene locus. Of the 234 eQTLs, 171 are associated with increased PLA2G2A mRNA expression, while 63 are linked with its downregulation. Interestingly, 37 of these eQTLs were observed in 2 or more tissues, with each displaying a consistent effect (upregulating or downregulating) on PLA2G2A expression across these tissues (Figure 1b, Supplementary Table 1). This consistency pointed to the recurring roles these 37 eQTLs play in modulating PLA2G2A expression in the human body. Consequently, our subsequent analysis was centered on these 37 eQTLs.

Figure 1.

Genome-wide SNPs that Impact PLA2G2A Gene Expression in Multiple Tissues. (a) The location of 234 eQTLs for PLA2G2A. X-axis shows the chromosomal position, and Y-axis shows tissues. The marker size indicates the magnitude of the effect size. The marker color indicates the effect direction in alternative alleles compared to reference alleles. (b) The location of 37 eQTLs shared among more than one tissue. X-axis shows the chromosomal position, and Y-axis shows tissues. The marker size indicates the magnitude of the negative logarithm of the P value. The color intensity indicates the magnitude of normalized effect size. The color indicates the effect direction in alternative alleles compared to reference alleles. Red: up-regulated, Blue: down-regulated. The genome-wide SNPs were examined using data from the GTEx project. The normalized effect size (NES) of the eQTLs is defined as the slope of the linear regression and is computed as the effect of the alternative allele (ALT) relative to the reference allele (REF) in the human genome reference GRCh38/hg38, as described in GTEx Portal (https://gtexportal.org/home/methods).

Identification of eQTL-bound TFs

Expression quantitative trait loci have the potential to modulate gene expression by various mechanisms, including transcription, RNA processing, RNA stability, and translation.^28-30 One of the predominant mechanisms involves the alteration of TF-binding sites on the targeted gene promoters.^31-33 To identify the TF-binding sites that may be affected by PLA2G2A eQTLs, we examined the 37 eQTLs (Figure 1b, Supplementary Table 1) using SNP2TFBS, a computer program that predicts how an SNP may affect TF-binding site sequence and binding affinity. ²⁴ From the SNP2TFBS analysis, 9 of the 37 eQTLs were associated with known TF-binding sites. Notably, 7 of these 9 eQTLs had the potential to influence TF-binding affinities due to sequence variations between the reference and alternative SNP alleles (Table 1). In particular, 7 eQTLs were predicted to affect the binding of 8 TFs: EBF1, ZNF263, SP1, PPARG_RXRA, SREBF1, SREBF2, Gata1, and Foxq1 (Table 1). Normalized effect sizes of significant tissue-eQTLs were obtained for these 7 eQTLs from the GTEx Portal (Table 1).

Table 1.

Candidate TFs associated with PLA2G2A gene expression determined by SNP2TFBS.

SNP (eQTL) rsID	Ref	Alt	Effect	eQTL tissues	Effect size	Matched TFs
rs12733637	A	G	down	Liver	−0.39	EBF1
				Prostate	−0.34
rs1891320	C	T	up	Liver	0.45	ZNF263
				Prostate	0.50
rs11573156	G	C	up	Liver	0.66	SP1
				Prostate	0.68
				Ovary	0.25
				Adipose-Visceral (Omentum)	0.17
				Stomach	0.19
				Heart-Atrial Appendage	0.15
				Heart-Left Ventricle	0.17
rs1796923	T	G	down	Skin-Sun Exposed (Lower leg)	−0.80	PPARG_RXRA
				Heart-Atrial Appendage	−0.51
				Skin-Not Sun Exposed (Suprapubic)	−0.65
				Esophagus-Muscularis	−0.62
				Thyroid	−0.40
				Heart-Left Ventricle	−0.48
				Adipose-Subcutaneous	−0.48
				Esophagus-Gastroesophageal Junction	−0.54
rs114156017	C	T	up	Liver	0.42	SREBF1 SREBF2
				Prostate	0.41
				Adipose-Visceral (Omentum)	0.18
				Heart-Atrial Appendage	0.16
rs149238894	G	A	up	Liver	0.40	GATA1
				Prostate	0.42
				Adipose-Visceral (Omentum)	0.18
				Heart-Atrial Appendage	0.15
rs139466428	TG	T	up	Prostate	0.38	FOXQ1
				Adipose-Visceral (Omentum)	0.15
				Esophagus-Muscularis	0.13

Ref , reference allele; Alt , alternative allele; Effect , the direction of the eQTL’s effect on PLA2G2A gene expression; “up” means the alternative allele upregulates PLA2G2A gene expression compared to the reference allele, and “down” means the alternative allele downregulates PLA2G2A gene expression compared to the reference allele.; eQTL tissues , tissue distribution of each eQTL; Effect size , the normalized effect size of each significant tissue-eQTL obtained from GTEx Portal, which was computed as the effect of the alternative allele relative to the reference allele in the human genome reference GRCh38/hg38; TF , transcription factor.

Verification of TF-eQTL binding

The relationships between eQTLs and TFs in the tissues they influence were next examined. We posited that if an eQTL modulates gene expression (in this case PLA2G2A) by affecting TF-binding affinity, there would be TF binding to the eQTL site in the tissues where the eQTL has an effect. To address this question, we used the ENCODE chromatin immunoprecipitation (ChIP)-seq dataset, which maps the chromatin-binding patterns of 340 TFs across 129 cell lines or tissue types. From the 8 TFs hypothesized to be influenced by PLA2G2A eQTLs, 6 TFs were evaluated in the ENCODE dataset (Table 2). However, only one TF, SP1, had been studied in an eQTL-matched tissue, specifically, the liver.

Table 2.

The list of candidate TFs examined in the ENCODE ChIP-seq dataset.

TF	Cell line	Origin of cell line
EBF1	GM12878	B-lymphocyte
ZNF263	HEK293	Kidney
SP1	A549 H1-hESC HEK293T HepG2 K562 Liver MCF-7	Lung Embryo Kidney Liver Bone marrow Liver Breast
GATA1	Erythroblast K562	Erythroblast Bone marrow
SREBF1	A549 K562 MCF-7	Lung Bone marrow Breast
SREBF2	A549 HeLa-S3	Lung Uterus cervix

The ENCODE data revealed that SP1 indeed binds to the chromatin region of the eQTL rs11573156 in the liver (Figure 2). Consistently, among all tissues examined in GTEx, the liver displayed the most robust eQTL effect (P < 5 × 10⁻²¹, NES = 0.66) of rs11573156 on PLA2G2A expression (Supplementary Table 2). Therefore, the subsequent analysis focused on the SP1-rs11573156 binding in modulating PLA2G2A expression.

Figure 2.

Display of confirmed SP1 binding regions in the liver. The chromatin regions of SP1 binding in the liver are shown around the rs11573156 eQTL site in the UCSC Genome Browser. The SP1 binding was determined by ChIP-seq (ENCODE 3 dataset). The gray and black rectangles indicate the respective SP1-binding regions (ChIP-seq peaks) in liver tissues derived from separate individuals. A vertical light blue line indicates the location of rs11573156.

SP1 bound to rs11573156 as a transcriptional activator of PLA2G2A

If SP1 binds to the eQTL rs11573156 in the liver and thereby influences PLA2G2A expression, a discernible correlation between liver SP1 and PLA2G2A would be anticipated. Specifically, considering the predictive higher binding affinity of SP1 to the alternative allele (C) of rs11573156 over the reference allele (G) (as shown in Table 1) coupled with the positive effect of the alternative allele on PLA2G2A expression, we postulated that there would be a positive correlation between SP1 and PLA2G2A levels in the liver. In the GTEx dataset containing 208 liver donor samples, SP1 was positively associated with PLA2G2A expression (correlation coefficient, r = 0.31; Figure 3a). This result supported the role of SP1 as a transcriptional activator of the PLA2G2A gene in the liver.

Figure 3.

Associations between PLA2G2A and SP1 gene expression in GTEx. (a) Correlation analysis of the gene expression between PLA2G2A and SP1 in the liver (r = 0.31). X-axis shows transcripts per million (TPM) of SP1 and Y-axis shows TPM of PLA2G2A. (b) Box plots of the correlation coefficients between SP1 and PLA2G2A TPMs across tissues. Top 11: Top 11 tissues by multitissue posterior probability of effect (m-value) of the eQTL rs11573156. Bottom 11: bottom 11 tissues by m-value.

To examine how the correlation between the gene expression of SP1 and PLA2G2A changes in various tissues with different effect sizes of rs11573156, we next focused on the top 11 and bottom 11 tissues or tissue subtypes by probability assigned by the GTEx project (out of a total of 48) where the effect of eQTL rs11573156 on PLA2G2A expression was most and least pronounced. The top 11 tissues were defined by a single-tissue P value <.001 and the multitissue posterior probability m-value = 1 (Supplementary Table 2), indicating the highest level of confidence in the eQTL effect within those tissues. We hypothesized that if SP1 binds to the eQTL rs11573156 and elevates PLA2G2A expression, then the correlation between SP1 with PLA2G2A would be stronger in these 11 tissues compared to others. This expectation was validated in our analysis. As illustrated in Figure 3b, the correlation coefficients between SP1 and PLA2G2A in these top 11 tissues (with an m-value of 1) were significantly higher (P = .03) than in the control tissues (bottom 11 with m-value <.21; Supplementary Table 2). This result suggested that tissues with larger allele effects of rs11573156 exhibited stronger correlations between SP1 and PLA2G2A, consistent with rs11573156 exerting its impact on PLA2G2A expression through SP1.

Model differential SP1-rs11573156 binding-mediated PLA2G2A expression

Next, we asked why the allele effect of rs11573156 varied depending on the tissue type. In the top 11 tissues in Figure 3b, there was a striking variation in the effect size of the eQTL rs11573156, spanning over a 5-fold range from 0.12 to 0.68 (Supplementary Table 2). To explore the potential molecular mechanism explaining how SP1-rs11573156 binding may account for the varying effect size of rs11573156 on PLA2G2A expression, we constructed a simple ODE model based on the premise of Hill function kinetics:

$$\begin{array}{c}\frac{d\left[G2A\right]}{dt}=\frac{k{\left[SP1\right]}^n}{K^n+{\left[SP1\right]}^n}-d_{G2A}\left[G2A\right]\#\end{array}$$ (3)

In this model, the change of PLA2G2A transcript level over time ( $\frac{d[G2A]}{dt}$ ) is a function of SP1 (TF) protein level (with $k,K,n$ and $d_{G2A}$ being constants, see Method for details). The difference between the alternative allele (C) and the reference allele (G) of rs11573156 is reflected in the value of the Hill constant $K$ that is inversely correlated with SP1 binding affinity representing a higher SP1 binding affinity to the alternative allele C than the reference allele G).

This minimalist model suggests that because of the difference in SP1- rs11573156 binding affinity $(K_G>K_C),$ a general trend regarding the steady-state level of PLA2G2A transcripts, [G2A]_ss, can be observed (Figure 4a and Supplementary Figure 2): (1) [G2A]_ss is higher with the alternative allele (C) than with the reference allele (G), (2) the discrepancy in [G2A]_ss between the 2 alleles, represented as Δ[G2A]_ss, becomes more pronounced at lower concentrations of SP1 protein levels, and (3) this divergence, Δ[G2A]_ss, gradually diminishes as SP1 concentrations rise. This result can largely be explained by the “saturating” effect at high SP1 levels, where the excess amount of SP1 protein has the capacity to fully transactivate PLA2G2A even to the lower binding affinity allele of rs11573156.

Validation of model predictions across tissues

To test the model predictions, we obtained the SP1 protein levels across diverse tissues from the HPA. SP1 protein levels were emphasized for a couple of reasons: (1) It is the protein level (rather than the mRNA level) of the TF SP1 that impacts PLA2G2A expression and (2) the SP1 mRNA levels across different tissues may not always correlate well with SP1 protein levels, due to tissue-specific differences in posttranscriptional modifications, translation, and protein stability, and so on.

Human Protein Atlas estimates protein levels in different human tissues in 3 categories, low, medium, and high. The categorical SP1 protein levels in the top 11 tissues are shown in Suppl. Table 2. When the PLA2G2A transcript level differential (ΔlogTPM) between the C and G alleles in each of the 11 tissues was plotted against their corresponding SP1 protein categories, a clear trend emerged. This trend was consistent with the model’s predictions: the difference in PLA2G2A transcript levels between alleles C and G was most significant at the lowest SP1 protein level, tapering off as SP1 levels ramped up to medium and then high (Figure 4b). Similarly, the effect size of the eQTL rs11573156, which is positively correlated with the PLA2G2A level difference between the C and G alleles, also exhibits a similar trend that is inversely correlated with SP1 protein level in the tissue (Figure 4c). Cumulative, these results strongly support the hypothesis that variations in PLA2G2A gene expression associated with the eQTL rs11573156 alleles are remarkably, and inversely, influenced by the tissue-specific SP1 protein levels.

Discussion

Much of the early work with the sPLA₂-IIA isoform centered on its bactericidal role where it was shown to increase 100 to 1000-fold (up to 5 μg/ml) in serum/plasma during infectious sepsis. ³⁴ However, studies beginning in the mid-1980s, also identified its capacity at persistently high concentrations to induce lethal multiple organ failure associated with sepsis.^18,20,21 In fact, Nyman and colleagues demonstrated that sPLA₂-IIA levels could effectively predicted lethal multiple organ failure when measured at later time points (>3 days in the intensive care unit [ICU]) in sepsis patients. ²⁰

More recently, our group discovered that deceased COVID-19 patients, primarily those succumbing to lethal multiple organ failure, also exhibited considerably elevated levels of circulating, catalytically active sPLA₂-IIA. ¹⁸ Elevated sPLA₂-IIA levels paralleled several indices of COVID-19 disease severity such as kidney dysfunction, hypoxia, and multiple organ dysfunction. Moreover, 3 machine learning models identified sPLA₂-IIA levels as the pivotal factor in differentiating patients who died from sepsis due to COVID-19. A pressing question emerging from these studies and other clinical conditions, such as cardiovascular disease, ⁹ rheumatoid arthritis, ¹⁰ and cancer,^12-15 is: Why do only certain patients produce and release high concentrations of sPLA₂-IIA from tissues, especially the liver?

Several GWASs have demonstrated the impact of genetic variation on sPLA₂-IIA levels and further revealed that rs11573156 G > C polymorphism is highly associated with circulating level of sPLA₂-IIA.^35-38 The mean allele frequencies across all global populations are 77% for the G reference allele and 23% for the C alternative allele. Frequencies of the C allele varying from 5% (Asian) to 36% (Estonian) in different populations.

This study demonstrated that among the 234 eQTLs for PLA2G2A, rs11573156 in the liver showed the most robust effect. This SNP exclusively affects PLA2G2A gene expression, and it is not identified as an eQTL with any of the other 11 sPLA₂ genes, including the 6 isoforms (PLA2G2E, PLA2GA, PLA2G5, PLA2G2D, PLA2GF, and PLA2G2C) that occur as a gene cluster (hg38 chr1:19 920 009-20 186 518) on chromosome 1 (data not shown). Furthermore, the pQTL (protein quantitative trait locus) analysis also indicated that rs11573156 is strongly associated with circulating protein levels of sPLA₂-IIA.^36-39

The rs11573156 SNP is positioned immediately upstream of the 5′ untranslated region (5′-UTR) in the consensus transcript of the PLA2G2A gene, as designated by the MANE Select project, ⁴⁰ acting at a TF-binding site that regulates PLA2G2A expression. This study, through our developed analysis pipeline (as outlined in Supplementary Figure 1), suggests that this region containing rs11573156 is a crucial SP1 binding site that modulates PLA2G2A transcript levels. Specifically, in tissues such as the liver where the SP1 protein levels are low, the difference in the binding affinity of SP1 to the G versus C allele leads to a significant disparity in SP1 binding dynamics and occupancy at the rs11573156 site, resulting in a notable difference in allele-specific PLA2G2A expression. In contrast, in tissues with high SP1 protein levels (e.g. skin and lung), SP1 saturates both the high and low affinity binding sites (i.e, both the C and G allele), thereby diminishing the allele-specific difference in PLA2G2A expression. That is, as summarized in Supplementary Figure 3, our study suggests that the PLA2G2A transcription depended on both (1) the SP1 binding affinity affected by the alleles of rs11573156 and (2) the concentration of SP1 protein in a given tissue, which could explain tissue-specific rs11573156 effect size. Moreover, there appears to be a direct relationship between PLA2G2A transcription and sPLA₂-IIA protein level, as indicated by the fact that rs11573156 also serves as a pQTL for sPLA₂-IIA. Notably, the liver is believed to be the primary organ to produce sPLA₂-IIA released into circulation.^41-43 This aligns with the observation that rs11573156 alleles via their differential binding affinity to SP1 modulate allele-specific PLA2G2A expression in the liver, and correspondingly, sPLA₂-IIA protein level variations in circulation.

This study has important limitations. First, the ENCODE ChIP-seq dataset does not include all tissues where the eQTL rs11573156 exhibits significant effect on PLA2G2A gene expression, including the liver, prostate, ovary, stomach, adipose-visceral, heart-left ventricle, and heart-atrial appendage. Among these tissues, only the liver has the ChIP-seq data of SP1. It remains unclear whether SP1-mediated allele-specific PLA2G2A expression also occurs in the other tissues. In addition, there were several other TF-eQTL combinations possibly showing allele-dependent binding affinity influencing the expression of other sPLA₂ isoforms. These include RFX1-rs71647137 for PLA2G5, CTCF-rs2070873 for PLA2G1B, NRF1-rs762521 and CTCF-rs11407099 for PLA2G10, ZNF143-rs9621185 and NFYB-rs5997969 for PLA2G3. Further analysis of such associations is currently limited by the unavailability of ChIP-seq datasets in the matched tissue types. Second, this study uses datasets from diverse publicly available resources, including GTEx, SNP2TFBS, ENCODE, and HPA. Comparing data from different sources presents inherent challenges and uncertainties. To mitigate these concerns, we employed complementary measures, by (1) focusing our main analysis on the liver, using multiple datasets from the same tissue to ensure context consistency, (2) adopting a categorical analysis approach, grouping tissues based on SP1 protein levels, enabling identification of broader trends and robust patterns across datasets (GTEx and HPA) and tissues (Figure 4b and c) independently validating our findings through a simple ODE model that successfully accounts for allele-specific PLA2G2A expression across 11 tissue types (Figure 4a). Third, this study focused on investigating the effect of TF-SNP binding dynamics on PLA2G2A expression. It is worth noting that some SNPs could also impact gene expression via a variety of other mechanisms, including mRNA stability, subcellular localization, micro RNAs (miRNAs) modulation, and epigenetic modulation.^44,45 Future studies with a wider range of experimental datasets and more sophisticated models are needed to expand our understanding of allele-specific PLA2G2A expression. Finally, it is important to note that our findings in this study, based on computational and modeling analysis, need to be carefully tested and validated in future experimental investigations involving cell models derived from various tissues and genotypes.

Given the diverse functions and diseases where sPLA ₂ -IIA plays a modulatory role, it is imperative to determine how levels of this pivotal enzyme are controlled and whether its synthesis can be inhibited. This article describes a TF-binding interaction between SP1 and rs11573156 (the SNP that has been most associated sPLA₂-IIA levels), which likely accounts for the allele-specific PLA2G2A expression in the liver. Unraveling a genetic molecular mechanism that controls sPLA₂-IIA levels in the primary organ of its generation represents an exciting finding in the field. Studies like this pave the way for future precision medicine approaches that may modulate PLA2G2A expression by altering the dynamics of SP1 binding to SNPs, such as rs11573156, especially in potentially fatal diseases like infectious sepsis.