Transcriptome-Wide Analysis Reveals Modulation of Human Macrophage Inflammatory Phenotype Through Alternative Splicing

Abstract

Objective

Human macrophages can shift phenotype across the inflammatory M1 and reparative M2 spectrum in response to environmental challenges, but the mechanisms promoting inflammatory and cardiometabolic disease-associated M1 phenotypes remain incompletely understood. Alternative splicing (AS) is emerging as an important regulator of cellular function, yet its role in macrophage activation is largely unknown. We investigated the extent to which AS occurs in M1 activation within the cardiometabolic disease context and validated a functional-genomic cell model for studying human macrophage-related AS events.

Approach and Results

From deep RNA-seq of resting, M1, and M2 primary human monocyte derived macrophages, we found 3860 differentially expressed genes in M1 activation and detected 233 M1-induced AS events; the majority of AS events were cell- and M1-specific with enrichment for pathways relevant to macrophage inflammation. Using genetic variant data for 10 cardiometabolic traits, we identified in M1 activation 28 trait-associated variants within the genomic loci of 21 alternatively spliced genes and 15 variants within seven differentially expressed regulatory splicing factors. Knockdown of one such splicing factor, CELF1, in primary human macrophages led to increased inflammatory response to M1 stimulation, demonstrating CELF1's potential modulation of the M1 phenotype. Finally, we demonstrated that an induced pluripotent stem cell derived macrophage system recapitulates M1-associated AS events and provides a high-fidelity macrophage AS model.

Conclusions

AS plays a role in defining macrophage phenotype in a cell- and stimulus-specific fashion. Alternatively spliced genes and splicing factors with trait-associated variants may reveal novel pathways and targets in cardiometabolic diseases.

Graphical abstract

Introduction

As the phagocytic cells of the innate immune system, macrophages reside in almost every organ of the human body and mediate homeostasis through surveillance for tissue injury and invading pathogens.^1-3 The most plastic cells of the hematopoietic system, they can rapidly evolve function in response to their microenvironment,^1,4-8 adopting phenotypes across a spectrum that consists of pro-inflammatory M1 and reparative M2 states at the extremes. Emerging evidence has demonstrated that systemic chronic inflammation seen in cardiometabolic diseases can induce macrophages to shift toward the M1 phenotype in disease relevant tissues, e.g., the arterial wall in atherosclerosis and adipose in obesity,^5,8-14 thereby exacerbating the disease-promoting inflammatory milieu and highlighting the potential of therapeutic strategies in mitigating M1 macrophage activation.

Although a few transcription factors have been implicated in the divergent gene expression profiles seen in M1 vs. M2 macrophages,^1,2,8,15,16 the mechanisms underlying the promotion of these specific phenotypes remain incompletely understood. A missing link is whether post-transcriptional regulation also contributes to the function of key proteins that modulate macrophage activation. As such, the study of alternative splicing (AS), or differential inclusion of exons into multiple mature mRNA transcripts,^17-19 may provide new insights into this process given the potential for functional differences in the translated protein products. Such study may be of particular importance in the human context given the rapid evolution of AS in primate species.^20,21

AS occurs in >95% of human multi-exon genes, enhancing the proteomic diversity encoded by the genome. Often, AS profiles are tissue- or cell type-specific^22-26 with variable interspecies conservation,^20,21,27 and these patterns are coordinated by a balance of the expression or activity of splicing factors, which are RNA binding proteins that bind to cis-regulatory elements near alternative exons. This binding affects recruitment of the spliceosome to pre-mRNA splice sites and influences whether an alternative exon is incorporated into the final transcript.²⁸ In addition, AS patterns can be disrupted by genetic variants affecting the “splicing code” of these cis-regulatory elements as well as the actual splice sites recognized during exon inclusion.^26,29,30

As a regulatory process, differential AS on a transcriptome-wide scale has been shown to occur during cellular phenotypic changes, e.g., during epithelial to mesenchymal transition, somatic cell reprogramming for induced pluripotent stem cell (iPSC) generation, neuronal and cardiomyocyte development, and T-cell activation.^31-35 Therefore, we hypothesized that aberrant splicing of a network of genes may contribute to pathogenic M1 activation in cardiometabolic diseases. Although a few studies have used RNA-sequencing (RNA-seq) to profile the transcriptomes of M1 and M2 macrophages,^36-38 differential AS during macrophage activation and its relationship to cardiometabolic diseases have not yet been explored.

To address this gap in knowledge, we sought to determine the extent to which AS and its regulation modulate human macrophage phenotype, with a focus on the cardiometabolic disease-associated M1 macrophage activation paradigm. The aims of our study were three-fold: (1) to characterize the AS landscapes of macrophage activation in primary human monocyte derived macrophages (HMDMs), (2) to identify splicing targets and splicing factors as novel targets for cardiometabolic traits using genetic variant data, and (3) to validate iPSC-derived macrophages (iPSDMs) as a cellular model system for investigating macrophage-relevant AS events. Here, we report the first comprehensive assessment of AS in human macrophage activation with translational context for the study of human cardiometabolic diseases.

Materials and Methods

Materials and Methods are available in the online-only Data Supplement.

Results

Macrophage Activation Leads to Distinct Transcriptomic Landscapes

To study the full spectrum of transcriptomic changes occurring during macrophage activation, we performed deep RNA-seq (∼130 million reads per sample) of resting M0-HMDMs, M1-HMDMs (LPS / IFN-γ), and M2-HMDMs (IL-4) from six healthy human participants as previously described, analyzing differences between resting and activated HMDMs (Figure 1A, GSE55536).¹⁶ We characterized elsewhere the divergent transcript-level profiles of M1-HMDM and M2-HMDM activation, including comprehensive differential gene expression of M1 and M2 mRNA markers as well as enrichment of different gene ontology pathways for each condition (pathways are summarized in Supplemental Table I).¹⁶ In this previous analysis, we had found 3860 genes to be differentially expressed in M1 activation while only 357 genes were differentially expressed in M2 activation.¹⁶ To explore whether macrophage activation also involved changes in AS patterns, we analyzed our RNA-seq data using the PennDiff, DEXseq, and IUTA^39-41 software programs to detect differential AS between M0-HMDMs and M1-HMDMs, our primary analysis, and between M0-HMDMs and M2-HMDMs in our secondary analysis. Three different software programs were used to maximize the number of detected AS events and to analyze AS on both the exon-level (DEXseq and PennDiff) and whole-gene level (degree of AS occurring at a locus, IUTA and PennDiff). We extended this approach to RNA-seq of PBMCs collected before and after intravenous bolus of 1 ng LPS per kg body mass exposure in eight healthy participants in a human experimental endotoxemia trial (GSE50792).^42-44 The PBMC transcriptomic profiles served as a basis to probe whether M1-HMDM AS changes occurred uniquely within the macrophage system or more broadly in other inflamed hematopoietic cells and contexts. Given that M1-HMDM activation induced differential expression of genes enriched for pathways distinct from those seen in M2-HMDM activation and in post-LPS PBMCs (Supplemental Table I)¹⁶, we hypothesized that AS patterns would also manifest in a cell-specific and stimulation-specific manner.

Figure 1. RNA-seq differential AS analysis of HMDMs, iPSDMs, and PBMCs.

A, Overview of cell types studied. PBMCs were isolated from human donors and were differentiated into primary M0-HMDMs, reprogrammed into iPSCs with subsequent macrophage differentiation, or stimulated with LPS in vivo in an evoked endotoxemia experiment. M0-HMDMs and M0-iPSDMs were then activated to M1 or M2 states. B, Distribution of differential AS and alternative promoter events for each experimental condition and cell type. C, Venn diagrams of single cassette exon events for each type of cell stimulation shows distinct differential AS patterns for each condition. HMDMs = human monocyte derived macrophages; iPSDMs = induced pluripotent stem cell derived macrophages; PBMCs = peripheral blood mononuclear cells; AS = alternative splicing.

Differential AS events in each experimental condition (M0- vs. M1-HMDMs, M0- vs. M2-HMDMs, or pre- vs. post-LPS PBMCs; Figure 1A) were considered significant if they were called by either DEXseq or PennDiff to have > 5% Δ exon inclusion level at a false discovery rate (FDR) P-value < 0.05 as determined by the Benjamini-Hochberg method.^39,40 We found that out of the three experimental conditions, M1-HMDM activation was accompanied by the most dynamic landscape for AS differences: the 233 AS events for M1-HMDM activation, which were predominately single cassette exon events, amounted to (1) more than 20 times the number of events occurring in M2-HMDM activation and (2) more 2 times the number of events occurring in LPS-stimulation of PBMCs (Figure 1B, Supplemental Tables II-IV). Although on a different scale, this pattern echoes what we had seen in differential expression of genes for M1 vs. M2 activation.

The computational programs we used also allowed us to detect alternative promoter usage in addition to AS events, and we found that the majority of isoform-switching across all experimental contexts involved alternative promoter sites (Figure 1B, Supplemental Tables II-IV). Previous studies have reported widespread usage of alternative promoters in the human genome, including differential alternative promoter events in M1-HMDMs,^36,45 and within genes using alternative promoters, concurrent AS events have been seen.^46,47 Indeed, for M1-HMDM activation one-sided Fisher's exact tests revealed robust enrichment of alternative promoter events not only in genes that were differentially expressed (P = 2.8 × 10^-6) but even more so in genes undergoing differential AS (P < 2.2 × 10^-16). Enrichment of AS events in differentially expressed genes was less robust (P = 0.02, Supplemental Figure I). Alternative promoter sites in M1-HMDM activation, however, had significant overlap with post-LPS PBMCs (Supplemental Table VI), suggesting that unlike AS (see below), alternative promoter events may not be driving most of the macrophage-specific changes in M1-HMDM activation.

Differential Alternative Splicing Patterns Are Specific to Cell Type and Stimulus

Differentially Spliced Genes in M1-HMDM Activation Are Enriched for Inflammatory Pathways

After establishing that M1-HMDM activation resulted in its own pattern of transcriptomic changes, we evaluated whether differential AS occurred in genes enriched for particular pathways specific to M1-HMDM activation. We interrogated differentially spliced genes for each type of cell stimulation using Ingenuity Pathway Analysis and found distinct pathway enrichment patterns for each experimental condition. Top pathways in M1-HMDM activation were consistent with intracellular responses to inflammation and included apoptosis as well as signaling in the AKT, mTOR, CDK5, ERK / MAPK, and NF-κB pathways (Table 1). These pathways contrasted with those involved in PBMC inflammation and M2-HMDM activation, which both included metabolic pathways such as lipoate modification and glycogen synthesis, respectively (Table 1). Top pathways in genes undergoing differential AS for each condition also differed from those for genes differentially expressed in the corresponding condition; as an example, in M1-HMDM activation the top pathways for alternatively spliced genes including apoptosis and mTOR signaling differ from top pathways for differentially expressed genes (Table 1, Supplemental Table I). Furthermore, the top functions enriched in networks of differentially spliced genes for each condition also differed. For example, angiogenesis and atherosclerosis predominated networks in M1-HMDM activation, while fatty acid metabolism and wound healing predominated M2-HMDM networks, and cellular migration and adhesion were top functions of post-LPS PBMC networks (Supplemental Figures II A-C).

Table 1. Top Enriched Pathways Associated with Differential Alternative Splicing Events.

Pathway	-log(P-value)
M0-HMDM vs M1-HMDM
Apoptosis Signaling	6.03
PI3K/AKT Signaling	4.15
mTOR Signaling	4.08
CDK5 Signaling	4.02
ERK/MAPK Signaling	3.94
NF-κB Signaling	2.04
M0-HMDM vs M2-HMDM
TWEAK Signaling	3.47
TNFR1 Signaling	3.03
PEDF / PPARγ Signaling	2.54
Interferon Signaling	2.11
PKR in inflammation	1.95
Glycogen Biosynthesis	1.70
Pre- vs Post-LPS PBMCs
IGF-1 Signaling	2.10
Lipoate Salvage and Modification	1.74
Cell Cycle Checkpoint Regulation	1.57
Hypoxia Signaling in the Cardiovascular System	1.57
Insulin Receptor Signaling	1.51
Nitric Oxide Signaling in the Cardiovascular System	1.51

Pathways identified by Ingenuity Pathway Analysis for genes that are differentially spliced in each experimental condition. LPS = lipopolysaccharide, PBMC = peripheral blood mononuclear cells, HMDM =human monocyte derived macrophage.

Cassette Exon Events Occur in Cell- and Stimulus-Specific Manner

To determine whether specific differential AS patterns are unique to cell type and condition, we focused on single cassette exons, the most common type of differential AS we found in our experiments (Figure 1B). M1-HMDM activation had only two out of its 123 cassette exon events overlapping with those occurring in M2-HMDM activation and only 14 overlapping with those in LPS-stimulated PBMCs (Figure 1C). We selected a subset of cassette exon events observed in M1-HMDM activation for RT-PCR validation prioritized based on biological relevance (e.g., PLD1 and GSK3B for macrophage metabolism; ARRB1, CD74, and TLR4 for immune response). This confirmed that these splicing changes were specifically seen in M1-HMDMs but not in M2-HMDMs (Figure 2A-B, Figure 3A). Differential AS of the alternative exons for these genes may have biological relevance as they encode a portion of the functional domains of the corresponding proteins such as the protein kinase domain of GSK3B and the TRAF6 interacting domain of ARRB1 (Supplemental Figures IV, V). These AS patterns differed not only from those for reparative M2-HMDMs but also from patterns observed during inflammatory stimulation of several other cells and tissues; for example, differential AS events seen in M1-HMDMs were not seen in visualization of our RNA-seq data for inflamed PBMCs, adipose, or renal epithelial cells (Supplemental Figure III).

Figure 2. RT-PCR validation of identified differential AS events.

A, Genome browser view of M0-HMDM vs. M1-HMDM RNA-seq wiggle tracks showing differential exon 16 inclusion in PLD1 (gray shading with arrow) and RT-PCR validation of AS of exon 16 in PLD1 in M1-HMDMs in 6 human donors independent of the RNA-seq experiments. % exon inclusion and SD calculated from gel densitometry of (inclusion PCR product) / (total PCR product). L denotes the DNA ladder in bp. B, Genome browser view and RT-PCR validation of AS of GSK3B's exon 9. SD = standard deviation; AS = alternative splicing; bp = base pairs.

Figure 3. iPSDMs recapitulate AS events in M1 activation.

A, RT-PCR validation of bioinformatically identified differential AS events in HMDMs and iPSDMs. % exon inclusion and SD calculated from gel densitometry of (inclusion PCR product) / (total PCR product). B, Correlation plot of AS events predicted for each gene (calculated to be value θ) in M1-HMDM vs. M1-iPSDM. Spearman's rho = 0.77, P < 0.001. C, Correlation plot of differential expression of splicing factors on the log(2) scale. Spearman's rho = 0.78, P < 0.001. iPSDMs = induced pluripotent stem cell derived macrophages; RT-PCR = reverse transcriptase polymerase chain reaction; SD = standard deviation; AS = alternative splicing; HMDM = human monocyte derived macrophage.

Because these differential AS events occur specifically in M1-HMDM activation, we hypothesized that they play a functional role during this phenotypic transition. CD74 has macrophage-relevant roles in antigen presentation,⁴⁸ and increased inclusion of CD74's exon 7, which encodes the proteolytic thyroglobulin domain of the CD74 protein (Supplemental Figure V), is observed during M1-HMDM (Figure 3A, Supplemental Table II). To test if exon 7 is involved in the inflammatory M1 response, we used anti-sense locked nucleic acid oligonucleotides to target exon 7 and found that compared to non-targeting control, knockdown of the exon 7-containing isoform induced even higher IL6 and CCL5 mRNA expression after 6 hours of 100 ng/mL LPS with 20 ng/mL IFN-γ stimulation of HMDMs (Supplemental Figure VI B) and almost complete macrophage detachment by 18 hours of stimulation (data not shown), suggesting that the inclusion isoform plays a role in modulating the M1-HMDM response.

Splicing Factors in M1-HMDM Activation

After determining that M1-HMDMs have a distinct AS pattern, we focused on potential splicing factors, which regulate tissue-specific splicing,^23,26,49 that might contribute to these differences. Performing RNA-binding motif enrichment analysis for simple cassette exon events in M1-HMDM activation, we found that regions within 300 base-pairs (bp) of a differentially spliced exon were enriched for binding sites of 12 splicing factors listed in Table 2. Enrichment for these splicing factor binding sites was not found near differentially spliced exons in M2-HMDM activation and LPS-stimulation of PBMCs. Because splicing factors that regulate smaller subsets of transcripts may not be identified through enrichment testing for all differential AS events, we examined the RNA-seq data of M0-HMDMs vs. M1-HMDMs for splicing factor gene expression patterns. Out of 220 known and putative splicing factors,⁵⁰ 37 were differentially expressed (>1.5 fold change, FDR < 0.05), with 14 upregulated in M1-HMDM activation (Supplemental Table VII) whereas no splicing factors were differentially expressed in M2-HMDM activation, consistent with the more dynamic and diverse changes in AS in M1-HMDMs. Although splicing factors do not necessarily need to be differentially expressed to have differential functionality in one condition over another, we used differential expression as a criterion for selecting M1-HMDM modulated splicing factor CELF1 for further study, as described below.

Table 2. Splicing Factor Binding Motif Enrichment Flanking Alternatively Spliced Exons in M1-HMDM Activation.

Splicing Factor	IUPAC for Motif	Unadjusted P	FDR –Adjusted P	Bonferroni-Adjusted P
CELF1	UGUUUGUUUGU	4.54E-04	4.5E-04	4.90E-02
HNRNPC	HUUUUUK	2.11E-04	2.5E-04	2.13E-02
HuR	UUDUUUU	9.91E-07	4.4E-06	1.01E-04
PABPC1	ARAAAAM	2.77E-06	5.5E-06	2.83E-04
PTBP1	HYUUUYU	1.10E-06	4.4E-06	1.12E-04
RBM3	RADACKA	3.68E-04	4.0E-04	3.69E-02
SART3	ARAAAAM	2.73E-06	5.5E-06	2.78E-04
SRSF10	AGAGAVM	2.30E-08	2.8E-07	2.34E-06
SRSF2	GGAGWD	1.93E-04	2.5E-04	1.95E-02
SRSF9	AKGAVMR	3.89E-05	6.7E-05	3.95E-03
TIA1	WUUUUUB	7.03E-05	1.1E-04	7.14E-03
U2AF2	UUUUUYC	1.51E-06	4.5E-06	1.54E-04

RNA-binding motif enrichment identified for splicing factors in the 300 bp regions flanking differentially spliced exons using the Analysis for Motif Enrichment tool of MEME-Suite.⁷⁹ Adjusted P-values reflect false discovery rate (FDR) and Bonferroni corrections for multiple testing. FDR was determined using the Benjamini-Hochberg method,⁸⁰ and Bonferroni correction entailed dividing the unadjusted P-value for the number of tests performed.

Common Genetic Variants Associated with Cardiometabolic Traits Highlight Potential Modulators of M1 Macrophage Activation Through Alternative Splicing

Because previous studies have shown that common genetic variants associated with cardiometabolic traits can have functional effects in disease pathophysiology,^51,52 we probed the relationship between trait-associated variants and differential AS in each cell condition.

Variants in Differentially Spliced Genes during M1-HMDM Activation

To determine which of the AS events occurring in macrophage activation and PBMC stimulation might be implicated in cardiometabolic diseases, we interrogated the genomic loci of these differentially spliced genes for cardiometabolic trait-associated SNPs in publicly available GWAS databases.^53-56 Our primary hypothesis for this analysis was that (1) trait-associated SNPs may identify genes differentially spliced in M1-HMDM activation that contribute to pathogenic M1-activation, and (2) a subset of these SNPs may directly cause disease-promoting splicing events by altering the splicing code. As a complementary study, we also interrogated SNPs in genes differentially spliced in M2-HMDM activation and in post-LPS PBMCs. After Bonferroni correction for the number of SNPs tested, we identified 28 tag SNPs that fall within the loci of genes differentially spliced in M1-HMDM activation and are associated with cardiometabolic traits including lipid phenotypes, obesity, and coronary artery disease (Table 3); the tag SNPs also fall within loci of genes with alternative promoter usage. These SNPs, which are mostly intronic and more than 300 bp away from splice sites, have very low LD with each other and are located throughout the genome at metabolism-related genes such as PLD1 and GSK3B or inflammation-related genes such as TAX1BP1, FBXL19, and ALOX5. Interestingly, trait-associated SNPs were not found in genes differentially spliced in M2-HMDM activation, and although GSK3B, GUCY1A3, and SYNE1 have tag SNPs for both M1-HMDM AS and post-LPS PBMC AS, the tag SNPs for genes alternatively spliced in both inflamed cell populations mostly differed (Table 3 and Supplemental Table X).

Table 3. Cardiometabolic Trait-Associated Genetic Variants in Differentially Spliced M1-HMDM Genes.

SNP	Chr	Position	Gene	Effect Allele	Trait	Consortium	Unadjusted P	FDR –Adjusted P	Bonferroni-Adjusted P
rs12765320	10	45930805	ALOX5	T	HDL	GLGC	2.8E-07	1.7E-04	2.8E-04
rs12325278	16	28848818	ATXN2L	G	BMI	GIANT	3.5E-10	8.1E-07	2.8E-07
rs12448482	16	28838073	ATXN2L	G	BMI	GIANT	8.8E-08	7.3E-05	7.1E-05
rs10838663	11	47184971	C11orf49	T	HDL	GLGC	2.8E-16	2.1E-12	2.8E-13
rs2291443	11	47179861	C11orf49	A	HDL	GLGC	3.1E-07	1.7E-04	3.0E-04
rs7117404	11	47127153	C11orf49	A	HDL	GLGC	3.5E-05	2.9E-03	3.4E-02
rs2156499	18	47007234	C18orf32	A	HDL	GLGC	2.5E-08	2.0E-05	2.5E-05
rs3094379	9	136334910	CACFD1	T	CAD	CARDIoGRAM	2.6E-05	1.5E-02	2.2E-02
rs739468	9	136326248	CACFD1	T	LDL, TC	GLGC	5.5E-06	8.8E-03	5.5E-03
rs11086985	20	44532255	CTSA	A	HDL	GLGC	6.6E-08	4.4E-05	6.5E-05
rs3817731	20	44518843	CTSA	A	HDL	GLGC	4.8E-06	7.8E-03	4.7E-03
rs10982200	9	117168551	DFNB31	A	HDL	GLGC	4.8E-05	3.7E-03	4.7E-02
rs9319588	16	30930983	FBXL19	T	TG	GLGC	7.9E-06	1.7E-02	7.7E-03
rs6781942*	3	119758049	GSK3B	A	TC	GLGC	3.8E-05	2.2E-03	2.4E-02
rs2306556*	4	156638573	GUCY1A3	A	CAD	CARDIoGRAM	3.4E-06	9.2E-04	2.2E-03
rs5763662	22	30378703	MTMR3	T	TC	GLGC	3.6E-05	1.5E-02	3.5E-02
rs881803	16	15802334	MYH11	T	CAD	CARDIoGRAM	5.5E-05	2.1E-02	4.7E-02
rs4687614	3	52492085	NISCH	A	WHR_BMI	GIANT	2.0E-07	3.3E-04	1.5E-04
rs360399	3	171362977	PLD1	A	HbA1c	MAGIC	8.2E-06	2.2E-02	6.8E-03
rs1265151	8	103213460	RRM2B	A	CAD	CARDIoGRAM	3.2E-05	1.5E-02	2.6E-02
rs13332660	16	29905882	SEZ6L2	T	TC	GLGC	1.5E-05	9.4E-03	1.4E-02
rs13107325	4	103188709	SLC39A8	T	BMI	GIANT	1.4E-07	5.4E-04	7.8E-05
rs13107325	4	103188709	SLC39A8	T	HDL	GLGC	7.2E-11	6.6E-08	7.1E-08
rs9939417	16	31053467	STX4	T	TG	GLGC	2.7E-07	1.5E-03	2.7E-04
rs594522*	6	152581079	SYNE1	A	TG	GLGC	4.3E-05	4.0E-02	2.8E-02
rs17611937	7	27775894	TAX1BP1	T	CAD	CARDIoGRAM	3.0E-05	1.E-02	2.5E-02
rs2655260	3	12563640	TSEN2	T	LDL	GLGC	7.3E-06	8.8E-03	7.3E-03
rs1618545	3	12560763	TSEN2	A	TC	GLGC	1.3E-06	9.4E-03	1.3E-03

Adjusted P values are corrected for multiple testing. HDL = high density lipoprotein, Fasting Glu = fasting glucose, T2DM = type 2 diabetes mellitus, BMI = body mass index, CAD = coronary artery disease, TG = triglycerides, FDR = false discovery rate.

^*Also present in post-LPS PBMCs

The effects of these SNPs on protein abundance, isoform expression levels, and function remain unknown, but they may exert effects on AS patterns. For example, ALOX5 is identified by the PennDiff software to have 7% more inclusion of exon 13 in M1-HMDM activation (P = 5.2 × 10^-4, Supplemental Table II, wiggle file in Figure 4A) and hosts rs12765320, a SNP associated with circulating high-density lipoprotein (HDL-C). Because the vast majority of tag SNPs in our analyses do not fall within the splicing regulatory regions of 300 bp flanking an alternative splice site, we performed additional LD analysis using 1000 Genomes SNP data to identify potential SNPs that could affect the splicing code. The intronic tag SNP rs12765320 is in high LD (R² ≥ 0.8) with rs56145077, an intronic SNP falling within 300 bp of alternative exon 13 which encodes a portion of the enzyme's lipoxygenase domain (Supplemental Figure V). rs56145077 lies within bioinformatically identified binding sites of splicing factors SRSF1 (P = 1.8 × 10^-3), RBM45 (P = 3.6 × 10^-2), and RBM5 (4.4 × 10^-2), thus potentially affecting the splicing code as the minor allele is predicted to decrease exon 13 inclusion by 32% (Figure 4A). Additional tag SNPs in LD with 1000 Genomes SNPs (European data) are listed in Supplemental Table XI.

Figure 4. Cardiometabolic trait-associated variants within genomic loci of alternatively spliced genes and splicing factors.

A, Genome browser view of ALOX5 RNA-seq wiggle tracks showing more exon 13 (purple) inclusion in M1-HMDMs, with a SNP, in LD with a tag SNP for HDL-C, predicted to disrupt the splicing code to decrease exon inclusion by 32%. The LD SNP falls within the binding sites for splicing factors SRSF1, RBM45, and RBM5. B, Genome browser view of splicing factor gene CELF1: RefSeq annotation in blue, C/EBPβ ChIP-seq peaks,⁵⁸ and histone tail modification marks on ChIP-seq.⁸¹ Yellow highlights indicate positions of SNPs in LD with tag SNPs for T2DM (green diamond) and fasting glucose (teal circle), notably overlapping C/EBPβ binding sites. HMDMs = human monocyte derived macrophages; SNP = single nucleotide polymorphism; LD = linkage disequilibrium; ChIP-seq = chromatin immunoprecipitation sequencing; T2DM = type 2 diabetes mellitus.

Variants in Differentially Expressed Splicing Factor Genes during M1-HMDM Activation

In addition to exerting effects on the splicing code of cis-regulatory elements, cardiometabolic trait-associated genetic variants may also affect AS if they are located within the loci encoding relevant splicing factors. Among the 37 differentially expressed splicing factor genes during M1-HMDM activation, seven have SNPs associated with cardiometabolic traits after Bonferroni correction (Table 4). CELF1, a splicing factor implicated in other human diseases,^34,57 has the broadest and most statistically significant associations with cardiometabolic traits including HDL-C, body mass index, fasting glucose, and type 2 diabetes mellitus (T2DM) (Table 4). We chose to study CELF1 further for the following reasons: (1) its robust associations with multiple cardiometabolic traits, (2) the enrichment of its RNA-binding motif near differentially spliced exons in M1-HMDM activation (Table 2), and (3) its differential expression on both the mRNA and protein levels during M1 activation (Figure 5A and 5B).

Table 4. Cardiometabolic Trait-Associated Genetic Variants in M1-HMDM Regulated Splicing Factors.

SNP	Chr	Position	Gene	Effect Allele	Trait	Consortium	Unadjusted P	FDR –Adjusted P	Bonferroni-Adjusted P
rs1044269*	11	47487740	CELF1	A	HDL	GLGC	4.9E-12	2.8E-10	1.3E-10
rs1317149†	11	47486885	CELF1	T	HDL	GLGC	4.1E-10	1.3E-08	1.1E-08
rs7933019‡	11	47509137	CELF1	C	Fasting Glu	MAGIC	4.4E-05	7.5E-03	1.8E-03
rs11607518‡	11	47504935	CELF1	A	T2DM	DIAGRAM	1.2E-03	1.7E-02	2.6E-02
rs2242081‡	11	47500267	CELF1	T	BMI	GIANT	1.5E-09	1.7E-07	4.3E-08
rs2232965‡	19	18426950	LSM4	T	CAD	CARDIoGRAM	7.9E-04	1.9E-02	2.2E-02
rs3891584‡	11	114269266	RBM7	C	T2DM	DIAGRAM	1.1E-01	1.7E-02	2.4E-02
rs7304944‡	12	56958447	RBMS2	T	TG	GLGC	1.7E-04	1.2E-02	5.0E-03
rs2657888‡	12	56938383	RBMS2	T	TG	GLGC	2.9E-04	1.2E-02	8.7E-03
rs11171894‡	12	56951219	RBMS2	A	Fasting Glu	MAGIC	2.7E-04	1.2E-02	1.1E-02
rs12232026‡	12	56960766	RBMS2	A	Fasting Glu	MAGIC	7.0E-04	1.2E-02	2.8E-02
rs17465651†	1	38418496	SF3A3	T	CAD	CARDIoGRAM	4.7E-04	1.9E-02	1.3E-02
rs17465826*	1	38423521	SF3A3	T	TG	GLGC	1.0E-03	1.2E-02	3.1E-02
rs3766287‡	1	31765685	SNRNP40	T	LDL	GLGC	9.9E-05	1.5E-02	7.7E-03
rs2302593‡	19	46196634	SNRPD2	C	Fasting Glu	MAGIC	4.8E-04	1.2E-02	1.9E-02

Adjusted P-values are corrected for multiple testing. HDL = high-density lipoprotein, Fasting Glu = fasting glucose, T2DM = type 2 diabetes mellitus, BMI = body mass index, CAD = coronary artery disease, TG = triglycerides, FDR = false discovery rate.

^*3′ UTR,

^†intergenic,

^‡intronic

Figure 5. Functional studies for CELF1 reveal role in modulating M1-HMDM activation.

A, Relative CELF1 mRNA expression in M1-HMDMs vs. M0-HMDMs by QPCR, normalized to ACTB. N=6. P < 0.01. B, CELF1 protein expression in M1-HMDMs vs. M0-HMDMs by Western blot. N=1. C, siRNA knockdown efficiency of CELF1 compared to non-targeting siRNA control, data normalized to ACTB. N = 3. D, Measured differences on dot blot cytokine array of secreted proteins in the cell supernatant of HMDMs with CELF1 siRNA knockdown vs. non-targeting siRNA control. PAI-1 (P < 0.05) and IL-1β (P < 0.01) were quantitated on densitometry normalized to reference spots. N = 3. N represents number of individuals and does not include triplicate replication. HMDMs = human monocyte derived macrophages.

LD analysis of CELF1's trait-associated tag SNPs revealed that rs11607518 (T2DM, adjusted P = 2.6 × 10^-2) and rs7933019 (fasting glucose, adjusted P = 1.8 × 10^-3) are in high LD (R²> 0.8) with rs77204316, rs11604233, and rs11039290—SNPs that fall within ChIP-seq binding sites for macrophage-relevant transcription factor C/EBPβ (GSE31621).⁵⁸ The location of these trait-associated SNPs at the CELF1 locus within binding sites for an essential macrophage transcription factor as well as evidence that M1 macrophages are involved in T2DM^59,60 prompted us to ask whether CELF1 plays a role in modulating M1 activation. First, we confirmed in samples of independent healthy human donors, that CELF1 mRNA (Figure 5A) and protein (Figure 5B) expression increased during M1-HMDM activation. Next, we performed siRNA-mediated knockdown of CELF1 (Figure 5C) and evaluated inflammatory response after HMDM stimulation with LPS and IFN-γ. CELF1 knockdown led to increased expression of a subset of inflammatory markers, including secreted IL-1β on cytokine array (Figure 5D, P < 0.01; Supplemental Figure VII B); mRNA expression of IL6 (P = 0.04), IRF1 (P = 0.02), and CXCL10 (P < 0.01) (Supplemental Figure VII A); and secreted IL-6 and CCL5 (Supplemental Figure VII C, P = 0.04 and P = 0.03, respectively). These results suggest that CELF1 may modulate a portion of the anti-inflammatory response during M1 activation.

Induced Pluripotent Stem Cell Derived Macrophages Demonstrate High Fidelity of Differential Alternative Splicing Events

With genetic variation emerging as a significant basis of complex cardiometabolic diseases,^54,56,61,62 the use of differentiated iPSCs, including iPSDMs, to model human diseases has become increasingly important in understanding functional impact and translational implications.^16,36,63-65 Recent studies have demonstrated that genetic variants may be causal in human diseases by inducing aberrant splicing through altering the splicing code (see above),^29,30,66 so we evaluated whether the iPSDM system can be a model for studying AS events. We analyzed RNA-seq data for M1-iPSDM and M2-iPSDM activation (derived from the same donors contributing to HMDM RNA-seq, Figure 1A) to determine if iPSDM cellular models recapitulated the AS profiles for M1-HMDMs and M2-HMDMs. On a transcriptome-wide scale, a strong correlation in gene-level differential AS events (defined as degree of differential isoform expression θ, using identical PennDiff and IUTA criteria as for HMDMs) is seen between M1-HMDMs and M1-iPSDMs (ρ = 0.77, P < 0.001; Figure 3B). RT-PCR assays also confirmed fidelity of predicted differences in AS patterns for both M1-HMDMs and M1-iPSDMs (examples from three sequenced donors shown in Figure 3A). Furthermore, a strong correlation in differential expression of splicing factor genes was seen between the two cell systems (ρ = 0.78, P < 0.001; Figure 3C), indicating that splicing regulation is largely similar between iPSDMs and HMDMs. These results demonstrate that the iPSDM system can be a useful tool in functional genomic interrogation AS events.

Discussion

To our knowledge, this the first study to characterize comprehensively the AS landscape of human macrophage activation, particularly with respect to cell- and phenotypic specificity and cardiometabolic diseases. Although one previous study demonstrated that a common genetic variant associated with low density lipoprotein cholesterol can alter AS,⁶⁶ ours is the first to integrate summary-level data from multiple GWAS consortia to explore the relationship between cardiometabolic variants and AS in M1 activation. We present novel evidence that the splicing factor CELF1, mostly studied in the context of muscular dystrophy and cardiomyopathy, plays a role in modulating M1 activation of human macrophages and also harbors genetic variation for multiple cardiometabolic traits. Finally, our iPSDM system provides a reliable cellular model for functional interrogation of human macrophage splicing events facilitating translational study of disease associated genetic variation in both the splicing code and splicing factors.

Transcriptional programs have been recognized as regulators of macrophage activation,^1,2,8,15,16 yet little is known about whether isoform-level expression profiles are also distinct for M1 and M2 macrophages. Because knockdown of whole genes may lead to unanticipated off-target effects, identifying disease-relevant isoform-level differences in a tissue- or cell-specific manner may open the door for more selective therapeutic targeting in re-programming macrophage phenotype, a focus of increasing interest in clinical translation.^8,67,68 Here we demonstrated that M1-HMDM activation was indeed associated with many changes on the isoform level, including more than 200 differential AS events. Although the number of AS events was far fewer than the number of differentially expressed genes in M1 activation, these AS events were enriched for a network of biologically relevant genes involved in macrophage functions such as autophagy (e.g., PLD1)⁶⁹ as well as immune processes (e.g., CD74 and TLR4),^70-73 all of which are linked to the modulation of macrophage inflammation. M2 activation also confers a recognized pattern of differential gene expression,¹⁶ but it induces relatively few differential AS events, suggesting that the inflammatory stress response of M1 activation may create more transcriptomic diversity through AS than in M2 activation. Differential AS events in inflamed PBMCs overlapped minimally with those in M1-HMDMs, reinforcing the concept that stress-induced splicing changes are mostly cell- and context-specific.

After establishing that AS contributes to the transcriptomic response during macrophage activation, the next mechanistic steps to advance clinical translation are to study (1) specific splicing targets and (2) the larger regulatory mechanisms underlying differential AS in macrophage activation. Here we showed through our proof-of-principle experiment with CD74 isoform-specific knockdown that transcript variants can have major effects on cellular function and survival. Whether transcriptome-wide differential AS events in M1-HMDM activation have systematic functional and pathophysiological significance requires further investigation. It is more likely that genes differentially spliced in M1-HMDM activation can modulate cardiometabolic pathophysiology if they house SNPs associated with relevant traits. The combination of structural mRNA changes in response to environmental challenges and the existence of trait-associated variants within a locus make a case for the potential importance of a gene in macrophage phenotype. We found 28 such tag SNPs within the loci of 21 genes differentially spliced in macrophage M1 activation. A subset of these SNPs are in LD (identifiable via 1000 Genomes⁷⁴ data, Supplemental Table XI) with specific variants that fall within regulation sites of alternative splicing. They may alter the splicing code and subsequently induce aberrant splicing to generate specific isoforms favoring a disease phenotype, a phenomenon documented for rare and common genetic variants for a variety of human diseases.^29,66 For example, here we showed that ALOX5, a gene involved in pro-inflammatory functions and leukotriene synthesis,⁷⁵ is predicted to have a decrease in its alternative exon inclusion level in the presence of a minor allele SNP associated with HDL-C. Further functional validation of this variant in terms of its effect on ALOX5's alternative splicing and downstream pathways is warranted. Because the majority of genes alternatively spliced in macrophage activation do not house cardiometabolic variants at their loci, while only a subset of those with trait-associated variants have sites where SNPs may affect splicing efficiency, trait-associated variants likely are not the primary driver of alternative splicing in this process. However, tag SNPs, serving as guideposts to identify genes functionally relevant in cardiometabolic traits, may still help identify and prioritize novel targets in studies of macrophage activation within the human disease context.

Beyond specific AS events and specific spliced genes, the expression and function of splicing factors determining isoform switch patterns in a tissue- and cell-specific manner holds promise for novel mechanistic insights and therapeutic translation. We found that M1-associated glucometabolic traits⁷⁶ are statistically associated with SNPs at the CELF1 locus overlying binding sites for the macrophage-relevant transcription factor C/EBPβ, suggesting that regulation of CELF1 gene expression in macrophages may impact human cardiometabolic diseases. Indeed, knockdown of CELF1 levels modulated expression of a subset of M1 markers, suggesting that CELF1 splicing targets are involved in M1 activation. Because AS is a critical process during development,^31,34,57 the increased splicing switches that occur in M1 activation may indicate that inflammatory stress induces a dedifferentiation profile. This concept is consistent with our observation of increased M1-HMDM expression of CELF1, which is more highly expressed in the embryonic state.^50,57 Remarkably, 36% of our bioinformatically identified AS events in M1-HMDM activation overlap with transcripts found to interact with CELF1 in CLIP-seq of murine hearts (GSE61892-61893),⁵⁷ although CLIP-seq within the human macrophage system will need to be performed in order to confirm experimentally that CELF1 is a regulator of AS in M1 activation.

Beyond splicing factors, additional regulatory mechanisms could be at play in influencing splicing patterns in macrophage biology. As expected, M1-HMDM activation was accompanied by a wealth of alternative promoter usage given that transcriptional regulation has been established as a driving force in determining macrophage phenotype. However, our data show that alternative promoter usage was highly correlated with AS events in the same gene, a poorly studied and understood phenomenon.⁴⁶ Future studies are needed to identify whether specific pro-inflammatory transcription factors interact with splicing factors to orchestrate M1-HMDM isoform switches and whether these switches can be explained by chromatin changes during M1 activation. In this respect, future studies on epigenetics, which have also been implicated in AS,⁷⁷ may be informative.

This study has several strengths. First, we took an unbiased approach to discover M1-HMDM differential AS events not previously identified in the literature. Second, our use of primary human macrophages for our primary analysis and for RT-PCR validation strengthens the applicability of our findings to human physiology and thus future clinical translational potential given the species-specific nature of many AS events.^18,20,21,78 Third, we leverage GWAS data for cardiometabolic traits to identify potentially novel modulators of M1 activation in complex human diseases. Fourth, our exploration of whether differential AS events can be recapitulated in the iPSDM system validates the use of this cutting-edge genomic technology for future disease variant-oriented studies of AS in macrophages.

This study also has limitations that suggest further avenues of research. First, while we identified many differential AS events and validated a few for M1-HMDM activation, demonstrating that specific isoforms are necessary for M1 activation was beyond the scope of this work. Second, although we were able to validate splicing changes on the mRNA level, we did not present protein-level data as our attempts to resolve predicted small molecular weight changes and modest isoform shifts using Western blot methods did not yield detectable differences. These differences would be more accurately detected and quantitated by mass spectrometry, which will be employed in future targeted follow-up. Also, our use of HMDMs from healthy individuals as a model for M1 and M2 activation may not capture the full spectrum of macrophage-specific transcriptomic changes in chronic cardiometabolic diseases (e.g., atherosclerosis) that develop and progress from factors including but not limited to age, genetic background, and other co-morbidities. In addition, our analysis of splicing factors lacked human macrophage CLIP-seq data, which can help prove that CELF1's effects on the M1 phenotype are via AS rather than other RNA binding regulatory mechanisms. Finally, our GWAS translation can be considered a starting point to generate experimental data to demonstrate whether the disease-associated SNPs are truly causal in altering splicing patterns and in modulating M1 activation. Future studies on the molecular mechanisms, RNA-binding protein interactions, causal splicing variants, and novel pathways of alternative splicing changes in macrophage activation are warranted.

In summary, we show that isoform-level differences help define macrophage phenotype and that the splicing mechanisms for producing isoform switches occur in a cell- and stimulus-specific fashion. Aberrant splicing that leads to deleterious traits may occur in the presence of trait-associated genetic variants that provide a causal link between splicing event and disease-related phenotype. These findings lay the groundwork for future macrophage isoform-specific studies and for evaluating novel causal mechanisms and therapeutic targets in cardiometabolic diseases.

Highlights.

• Alternative splicing (AS), the generation of multiple isoforms from a single gene, is emerging as a key mechanism for proteomic diversity that contributes to cellular reprogramming and cellular phenotypes. For complex cardiometabolic diseases, AS is an understudied cell-specific process that may contribute to inflammatory macrophage activation and disease pathophysiology. This is the first comprehensive assessment of AS and its regulation in macrophage activation, with a translational focus on the cardiometabolic disease-associated, pro-inflammatory M1 phenotype.

• We characterized in primary human macrophages cell-specific, phenotype-specific, and context-specific AS landscapes of macrophage activation.

• Using human genetic data, we identified alternatively spliced genes and splicing factors relevant to M1 activation as novel candidates and targets for cardiometabolic pathophysiologies.

• iPSDMs were validated as a functional genomic cell model system for investigating human AS events.

• This work lays the foundation for further isoform-specific functional studies of potential novel causal mechanisms and therapeutic targets for cardiometabolic diseases and advances the study of AS in the context of human genetics toward future translation in precision medicine.

Nonstandard Abbreviations and Acronyms

AS

Alternative splicing

BMI

Body mass index

CAD

Coronary artery disease

CLIP-seq

Cross-linking and immunoprecipitation sequencing

GWAS

Genome-wide association study

HDL

High density lipoprotein

HMDM

Human monocyte derived macrophages

iPSC

Induced pluripotent stem cell

iPSDM

Induced pluripotent stem cell derived macrophages

LD

Linkage disequilibrium

LDL

Low density lipoprotein

LPS

Lipopolysaccharide

PBMC

Peripheral blood mononuclear cell

RNA-seq

RNA sequencing

RT-PCR

Reverse transcriptase polymerase chain reaction

SNP

Single nucleotide polymorphism

T2DM

Type 2 diabetes mellitus

TG

Triglyceride