Altered Enhancer and Promoter Usage Leads to Differential Gene Expression in the Normal and Failed Human Heart

October 29, 2020 Editor

[*][**][***]

What is New?

We identified genomic regulatory regions used in the human left ventricle from normal and failed hearts using cap analysis of gene expression sequencing.
We found >120 genes using alternative promoters in heart failure where the resulting protein changed its amino terminus because of this promoter shift.
The PRKAG2 gene shifts its promoter usage in failed hearts, which could change the action or location of the AMPK (AMP-activated protein kinase) enzyme.

What are the Clinical Implications?

The failed heart expresses different genes compared with the normal heart; some of these gene expression changes are adaptive and others are maladaptive.
These gene expression changes are directed by distinct regulatory genomic regions.
These regulatory regions are targets for genetic manipulation to potentially alter or treat heart failure.

Introduction

Heart failure with reduced ejection fraction is characterized by altered metabolism and function, each which contribute to an inability to meet demands for normal activity. Heart failure is associated with global changes in gene expression, and some of these gene expression changes directly drive pathological and adaptive remodeling.¹ For example, the failed heart shifts its metabolism towards glycolysis, driven, in part, by gene expression changes.² Within the failed human heart, a distinct form of myosin heavy chain is expressed,^3,4 as are alternatively spliced forms of TNNT2 and TTN, encoding troponin T and titin, respectively, and these changes directly modify contractility and compliance.^5–7 In addition, mutations in many of these genes directly lead to cardiomyopathy and heart failure.^8,9

Gene expression is regulated by promoters, which locate immediately upstream from the target gene, and enhancers, which can reside much further away from the target gene. Transcription factors are proteins that bind promoter and enhancer sequences, which interact in 3-dimensional space to regulate gene expression. The genomic regions responsible for regulating normal and pathological gene expression in the heart are incompletely understood. A few specific genetic regulatory regions have been characterized,^10,11 but comparatively few genome-wide analyses have been conducted. Surveys of the cardiac epigenome have been performed using developing mouse hearts and embryonic stem cell derived-cardiomyocytes.^12,13 One study of mouse hearts subjected to pressure overload, examined chromatin conformation to indicate potential active regulatory regions.¹⁴ However, the mature human heart has its own gene expression program that differs from mouse hearts and stem cell-derived cardiomyocytes.

Estimates of number of active heart enhancers vary from several thousand to tens of thousands depending on the approach used.^15,16 An assessment of active cardiac enhancers relied on detecting p300/CBP binding sites from one human fetal and one adult failed heart.¹⁵ This analysis identified ≈5000 active enhancers in fetal tissue and ≈2000 active enhancers in adult tissue, with approximately half of adult heart enhancers also active in fetal heart.¹⁵ A similar approach used normal human and mouse hearts integrating p300/CBP binding sites with active histone (H3K27ac [histone H3, acetylated on lysine 27]) marks.¹⁶ This combined analysis suggested >80 000 potential heart enhancers.¹⁶ Although these studies provide valuable data sets, the genomic alterations seen in human heart failure require a better definition.

Genes can be expressed from more than one promoter. Alternative promoters are estimated to affect 30% to 50% of human genes.¹⁷ Alternative promoters may alter the 5′ untranslated regions of transcripts or may also affect the amino terminus of the encoded protein. Both of these alterations can have functional consequences. Additionally, alternative promoters can also influence the effect of genetic variants on protein function.¹⁸ Despite the potential of broad proteome differences due to alternative promoter usage, a genome-wide view of promoter shifts in human heart failure is lacking.

Cap analysis of gene expression (CAGE) determines RNA transcriptional start sites at high resolution.¹⁹ Enhancer RNAs (eRNAs) are short RNAs produced from enhancer sites. CAGE sequencing (CAGE-seq) detects eRNAs from enhancers, seen as short bidirectional transcripts, as well as the unidirectional sequences from the capped ends of longer transcripts, reflecting promoters.^20,21 To define alternative promoter and enhancer use in heart failure, we generated CAGE-seq data from healthy and failed human left ventricles (LV). We relied on a no-amplification nontagging CAGE-seq protocol, to permit more robust and less biased detection of transcriptional start sites.¹⁹ We combined CAGE-seq with RNA-seq from the same hearts to better compare healthy and failed hearts. In aggregate, we identified >23 000 unidirectional signals and >5500 bidirectional signals in the heart. The promoters used by highly expressed, tissue-specific genes, like those encoding the sarcomere proteins, had distinct molecular signatures. We found enhancers in tissue-specific genes often mapped to first introns, and the features of these enhancers suggest they are highly critical for driving cardiomyocyte-specific gene expression. We compared regulatory regions between failed and healthy hearts and identified 129 genes with differential promoter usage in heart failure. These alternative promoters have the potential to encode proteins with unique aminotermini, highlighting potential protein composition shifts in the failed heart which may contribute to heart failure.

Methods

For detailed methods, see Expanded Methods in the Data Supplement.

Approvals

Subjects provided informed consent under Institutional Review Board approval from the University of Chicago or Northwestern University.

Transparency and Data Availability

All scripts/code used in this analysis is available upon request. Raw and processed sequence data are publicly available at NCBI-GEO under accession number GSE147236.

RNA-Extraction, Library Preparation, and Sequencing

Healthy and failed LV samples were obtained from failed transplants or as discarded tissue, respectively. Living subjects provided consent. Healthy LV samples were obtained from hearts provided by the Gift of Hope of Illinois and were found to be unsuitable for transplant due to age or prior cardiac surgeries. All patients were declared to be brain dead as the result of cerebral hemorrhage and familial consent was obtained for organ use in research. Custom nAnT-iCAGE-seq (no-Amplification-no-Tagging Illumina Cap Analysis of Gene Expression libraries) libraries were prepared by DNAFORM (Japan) following a previously described protocol.¹⁹ Approximately 50 pmol/L of pooled libraries was sequenced using a NextSeq 500 (Illumina) to yield ≈400 million total 75 bp single-end reads (Table I in the Data Supplement). RNA-seq libraries were prepared using the TruSeq mRNA-seq library preparation kit (Illumina). Libraries were pooled in equimolar amounts and loaded on the HiSeq 4000 (Illumina) to generate ≈40 million 150 bp paired-end reads/sample. CAGE-seq analysis was conducted using adapted methods.^22–29

Epigenetic data sets of interest were downloaded from their respective locations (Table II in the Data Supplement) an compared to CAGE-seq clusters using HOMER.³⁰

Results

Identifying Genetic Regulatory Regions of the LV

RNA-seq assays gene expression by looking at the entire length of RNA transcripts. CAGE-seq identifies RNAs originating from promoters and enhancer regions by targeting just the 5′ capped ends of transcripts. Since gene expression patterns differ between normal and failing hearts, we generated CAGE-seq data sets from LV from 3 healthy and 4 failed hearts. Healthy LV samples were those acquired but not used for transplant due to age or other incompatibility. Failed hearts were obtained at the time of transplant from patients with a range of ages (Table). The small sample size did not allow the consideration of age and primary gene mutation in the analysis. Each library was sequenced to high depth with comparable alignment rates (Table I in the Data Supplement). To generate a comprehensive list of all potential promoters and enhancers, the initial analysis combined the results from healthy and failed hearts. CAGE-seq analysis identified 23 676 promoter regions, defined as unidirectional sequence clusters, and 5647 enhancers, defined as bidirectional sequence clusters.

**Table.** Left Ventricle Tissue Source Demographics, Phenotypes, and Mutations
Sample	Primary Phenotype	Additional Phenotypes	Sex	Race	Age	Primary Gene Mutation(s)
Healthy 1	Healthy	…	M	White	62	NA
Healthy 2	Healthy	…	M	White	47	NA
Healthy 3	Healthy	…	F	White	76	NA
Heart failure 1	Cardiomyopathy	Ventricular tachycardia	M	White	20	TPM1 D230N
Heart failure 2	Cardiomyopathy	…	M	Hispanic	16	TTN c.42521-5 C>G, TNNT2 K210del
Heart failure 3	Cardiomyopathy	Becker muscular dystrophy	M	White	54	DMD IVS +1 G>T
Heart failure 4	Cardiomyopathy	Limb girdle muscular dystrophy	F	White	26	LMNA c.1142-1157+1del17

The promoter sequences, seen as unidirectional CAGE-seq clusters, were mapped relative to known genes (example shown in Figure I in the Data Supplement and pipeline shown in Figure II in the Data Supplement). A total of 70.1% of these promoter sequences mapped near transcriptional start sites (±100 bp), as would be expected for promoter function (Figure 1A). An additional 8.1% of these unidirectional clusters, mapped between 100 and 1000 bp upstream of transcriptional start sites. The remaining 21.8% of sequence clusters mapped to untranslated regions, coding and noncoding exons, introns or intergenic regions.

**Figure 1.** **The genomic position of promoters and enhancers used by human left ventricles (LV)**. A, Using cap analysis of gene expression sequencing (CAGE-seq) of LVs, >23 000 promoters were found across the genome, and 78.2% mapped near the transcriptional start sites (TSS) of genes (orange and blue combined, 1–1000 bp from TSS). B, The promoter sites were analyzed for transcription factor binding sites. Elk, Sp1, and GFY-Staf transcription factor binding sites were enriched in promoters near transcriptional start sites (orange bars). C, The promoters near transcriptional start sites showed epigenetic marks of open chromatin and active histone marks (orange lines in assay for transposase-accessible chromatin sequencing [ATAC-seq] and H3K4me3 [histone H3, tri-methylated on lysine 4] chromatin immunoprecipitation sequencing [ChIP-seq]), indicating these regions are open and active in the heart. D, More than 5500 enhancer signals were found across the genome, and >30% of these enhancers mapped into intronic and intergenic regions (green and brown combined). E, Intronic and intergenic enhancers were enriched for cardiac transcription factor binding sites like GATA, GRE, and Mef2, which are transcription factors known to be important for heart function (green and brown bars). F, Intronic and intergenic enhancer sites also had signals of active enhancer function, seen as ATAC-seq and H3K4me1 (histone H3, mono-methylated on lysine 4) and H3K27Ac (histone H3, acetylated on lysine 27) histone marks (green and brown lines in each). Dashed lines in C and F represent negative controls.

We next analyzed the promoter clusters for the presence of transcription factor binding motifs. The 70.1% of clusters mapping within 100 bp of transcriptional start sites were highly enriched for GFY-Staf, Sp1, and Elk/ETS binding motifs, which are transcription factors known to bind promoters.³¹ Clusters mapping into other regions showed minimal enrichment of these motifs (Figure 1B). To provide additional support for the promoter-enriched sequence clusters, assay for transposase-accessible chromatin sequencing (ATAC-seq) and H3K4me3 (histone H3, tri-methylated on lysine 4) chromatin immunoprecipitation sequencing (ChIP-seq) data sets from human LV were evaluated because these data indicate open and active chromatin; Table II in the Data Supplement provides the source information. The promoter clusters overlapped considerably with ATAC-seq and H3K4me3 ChIP-seq signals, providing strong support for these CAGE-seq clusters as identifying promoters (Figure 1C). These clusters also showed high CTCF (CCTCT binding factor) and Pol2A (RNA polymerase II) binding, as well as a reduction of H3K4me1 (histone H3, mono-methylated on lysine 4) histone modifications, which is consistent with these regions being promoters and not enhancers (Figure III in the Data Supplement). Taken together, the unidirectional CAGE-seq clusters bear the genomic signatures of active promoters.

Bidirectional eRNA clusters indicate likely enhancer regions. We similarly annotated bidirectional CAGE clusters for their position relative to genes. Only 44.5% of bidirectional clusters mapped ±100 bp within transcriptional start sites. In contrast to unidirectional clusters, 24.3% of clusters mapped to gene introns and 7.6% were intergenic (Figure 1D). These intergenic and intronic enhancer clusters showed enrichment of GATA, GRE, and MEF2 transcription factor binding motifs. Each of these transcription factors is essential for cardiomyocyte specification and maintenance (Figure 1E).³² Intergenic bidirectional clusters showed enrichment of open chromatin signals (ATAC-seq), H3K4me1, and H3K27ac histone modifications in human LV. Intronic clusters also showed a similar pattern but with a lower magnitude (Figure 1F). The intergenic and intronic bidirectional clusters showed enrichment of CTCF and Pol2A binding as well as reduced H3K4me3 modifications (Figure III in the Data Supplement). These patterns are highly consistent with the role of intronic and intergenic bidirectional CAGE-seq clusters as being enhancers, rather than promoters. Furthermore, these patterns represent multiple, independently derived sources of evidence that bidirectional eRNA transcription signify functional enhancers.

Two Types of Promoters

Mammalian promoters can initiate transcription across broad or narrow genomic regions, and these promoter shapes, broad or narrow (sharp), correlate with distinct transcriptional regulatory mechanisms.³³ We evaluated cardiac promoters predicted from CAGE clusters for these 2 major types of promoters by calculating the interquartile range of promoter CAGE clusters by determining the base pair distance between 10% and 90% of a promoter’s total signal. We observed the expected 2 distinct populations, defined as sharp (interquartile range <10 bp) and broad promoters (interquartile range ≥10 bp; Figure 2A). Broad promoters were those associated many different cellular functions, including housekeeping functions. In contrast, genes with sharp (narrow) promoters were those encoding proteins critical for tissue-specific functions seen by the presence myofilament, muscle tissue, and muscle contractions genes (Figure 2B). Thus, tissue-specific genes important for LV specification and function were more likely to have sharp promoters.

Figure 2. Cardiac promoters have sharp or broad transcriptional start sites. There are 2 types of promoters, those that initiate transcription over a narrow region (<10 bp, sharp promoter) or over a broad range (10–100 bp). A, The >17 000 CAGE promoters signals near transcriptional start sites can be divided into these 2 groupings, sharp and broad. B, Genes with broad promoters (red bars) encoding proteins with housekeeping functions (cellular localization, gene expression, organelle organization, and cellular metabolic processes). In contrast, genes encoding proteins of the muscle filament and muscle contraction had sharp promoters (blue bars). C, Sharp promoters have TATA regulatory motifs, while broad promoters, including those for tissue-specific genes, are defined by CpG islands. D, Genes with sharp promoters, which include tissue-specific genes, are expressed at higher levels than genes with broad promoters. The left shows that sharp promoters have higher normalized transcripts per million (TPM) levels (80.5 vs 73.5). The right panel indicates that sharp promoters span a shorter genomic distance (39 vs 98 bp). E, On average, failed hearts have slightly wider promoters compared with healthy hearts, which likely reflects differences in gene expression that occur during heart failure. Significance determined by 2-tailed nonparametric Mann-Whitney test (P≤0.05 [*], ≤0.0005 [***]). IQR indicates interquartile range; and TSS, transcriptional start site.

Sharp and broad promoters also displayed differential enrichment of upstream sequence DNA-binding motifs. Sharp promoters had TATA motifs at positions 30–33 upstream of the predominant transcriptional start site, representing canonical TATA boxes (Figure 2C). Broad promoters were devoid of TATA motifs, and instead showed enrichment of GC nucleotides consistent with CpG islands.³⁴ Sharp, tissue-specific promoters were also more highly expressed compared with broad promoters, and this observation was driven by a smaller population of very highly expressed sharp promoters, for example, those encoding sarcomere genes (Figure 2D). We compared promoter shape between healthy and failed hearts and found a small but significant genomewide increase in promoter interquartile range in failed hearts, consistent with a slight widening of transcriptional start sites in heart failure (Figure 2E).

Predicted Enhancers Map Within the First Intron

A large proportion of the predicted enhancers mapped to introns. We observed that the majority (69%) of intronic enhancers in this data set mapped to the first intron (Figure 3A). First introns are more conserved than other introns and correlate with higher levels of gene expression.³⁵ In the LV CAGE-seq data, these first intron enhancers generated more eRNA than enhancers in other introns but were not wider and did not differ in their balance of bidirectionality (Figure 3B). Notably, these intronic enhancers mapped within genes enriched for cardiac and muscle gene ontology terms (Figure 3C). This finding identifies that highly expressed, tissue-specific genes, like those encoding myofilament proteins, often have enhancers in their first introns.

Figure 3. The majority of intronic cardiac enhancers localize to the first intron. A, Of the 1371 enhancers that fell within introns, ≈70% of intronic enhancers mapped to the first intron of a gene. B, First intron enhancers were expressed more highly than other intronic enhancers, while other features like the width and expression from both strands were similar between first intronic enhancers and other intronic enhancer. Thus, first intronic enhancers were similar to other enhancers except that they were expressed more highly. C, Intronic enhancers were more likely to be found in tissue-specific genes, like those that encode components of the sarcomere. Significance determined by 2-tailed nonparametric Mann-Whitney test (P≤0.05 [*]).

Correlation of CAGE-Seq and RNA-Seq Data

RNA-seq, which evaluates the transcripts along their entire length, was performed on the same LV samples and was compared to CAGE-seq. CAGE-seq, which assays just the 5′ end of capped transcripts can assay gene expression, but is more sensitive to RNAs derived from promoters and enhancer regions. Since the assays were performed on the same tissue, a tight correlation between expression values was observed. Consistent with this, there was a tight correlation between CAGE-seq and RNA-seq results for any given gene (Figure 4A). Additionally, we assessed correlations between pairs of samples. In general, healthy hearts correlated best with other healthy hearts and failed hearts compared best with failed hearts. The RNA-seq and CAGE-seq expression estimates were most correlated for matched samples except for failed heart 4, which likely reflects the lower CAGE-seq read depth in this sample (Figure 4B). We next compared gene expression differences between failed and nonfailed hearts using both CAGE- and RNA-seq data sets. Overall, RNA-seq was more sensitive and, therefore, identified more upregulated and downregulated genes; approximately half of the genes identified by CAGE-seq were also identified by RNA-seq (Figure 4C). The types of genes differentially expressed were similar in both sequence data sets. Genes associated with developmental pathways and extracellular matrix organization were upregulated in heart failure while genes associated with catabolism were downregulated in heart failure (Figure 4D), consistent with prior gene expression profiling of failed hearts.^1,2

Figure 4. RNA-seq and CAGE-seq gene expression levels correlate. RNA sequencing (RNA-seq) measures all RNAs while cap analysis of gene expression sequencing (CAGE-seq) measures expression capped transcripts including the short enhancer RNAs (eRNAs). A, Overall, there was a good correlation between a gene’s expression level measured by RNA-seq and measured by CAGE-seq. Each dot represents a single gene, and genes with high expression by RNA-seq also had high expression by CAGE-seq analysis. This pattern was seen in both healthy (teal) and failed (red) hearts. B, The correlation coefficient of genomewide gene expression levels showed that expression in healthy hearts correlated best with other healthy hearts and failed hearts compared best with failed hearts. C, Comparing gene expression between healthy and failed heart identified differentially expressed genes and eRNAs. RNA-seq identified more upregulated and downregulated genes, and approximately half of the genes identified by CAGE-seq were also identified by RNA-seq. D, Genes associated with developmental pathways and extracellular matrix organization were upregulated in heart failure while genes associated with mitochondrial transport and catabolism were downregulated in heart failure.

CAGE-Seq-Defined Enhancer Regions Are Validated by Other Enhancer Data Sets

Of the ≈1800 candidate enhancer regions identified by CAGE-seq, data were available from 45 of these in the Vista Enhancer Browser, a list of enhancers tested in an in vivo reporter assay using transgenic mouse embryos.³⁶ Of the 45 present in Vista, 31 (70%) demonstrated enhancer activity in the developing mouse heart (Figure IVA in the Data Supplement). We also compared CAGE-seq-predicted enhancers to those predicted by H3K27Ac and p300 ChIP-seq from developing and adult human and mouse tissues.¹⁶ CAGE-seq-defined enhancers showed significantly higher overlap to H3K27Ac/p300 ChIP regions compared to length-matched scrambled control regions (Figure IVB in the Data Supplement). One study with H3K27Ac ChiP-seq data from healthy and failed human hearts were similarly compared and showed significant overlap (Figure IVC in the Data Supplement).³⁷ Finally, we compared CAGE-seq-defined enhancer predictions from the Functional Annotation of Mouse consortium, which used CAGE-seq across many nondiseased tissues to define enhancers.³⁸ We observed significant overlap with FANTOM predictions (Figure IVD in the Data Supplement), but we found many additional enhancers beyond the FANTOM predictions because we used a higher depth of sequencing (Table I in the Data Supplement). The intersection of these multiple data sets corroborates the cardiac enhancers now identified by CAGE-seq in both healthy hearts and failed hearts.

Alternative Promoter Usage in Heart Failure

In the LV, 3032 (23%) expressed genes had evidence for more than one promoter (Figure 5A). For these multipromoter genes, we compared the average percent usage of each promoter in healthy and failed hearts and found 609 promoters in 325 genes with a shift ≥10% (Figure 5B). Of these, 149 promoters in 124 genes occurred after the exon containing the start codon, indicating the potential to alter the aminoterminal amino acid sequence of the resulting protein (Figure 5C). Of the 124 genes identified as having alternative promoters that occur after start codons in heart failure, many are associated with sarcomere regulation or muscle structure development, including TNNT, MYOT, and SPEG. This indicates the heart failure can result in alternative proteins due to promoter shifts. We annotated a significant promoter switch in PRKAG2, a gene linked to hypertrophic cardiomyopathy and critical to heart metabolism.³⁹ Three major PRKAG2 promoters were identified, encoding 3 different isoforms- γ2a, γ2-3b, and γ2b (Figure 5D). In healthy hearts, the relative expression of these 3 transcripts is 53% γ2b, 28% γ2-3b, and 17% γ2a. In heart failure, these percentages significantly shift with 29% γ2b, 59% γ2-3b, and 10% γ2a isoform (Figure 5E). Notably, the γ2-3b isoform encodes a unique 32 amino acid sequence at the aminoterminus (Figure 5F). Total expression of PRKAG2 was not different between healthy and failed hearts. We interrogated the 30 kb upstream of the γ2b and γ2-3b isoforms for transcription binding motifs and found an enrichment of Smad and GRE motifs upstream of the γ2-3b isoform, suggesting a role for these transcription factors (Table III in the Data Supplement).

Figure 5. Heart failure was associated with significant changes in promoter usage. A, Approximately 20% of genes active in the heart use more than one promoter. B, For genes with more than one promoter, promoter usage is altered between failed and healthy. The x-axis represents the difference between a promoter’s average percent usage in the healthy or failed left ventricles (LVs). Left-shifted promoters make up a greater percentage of gene expression in failed ventricles and right-shifted promoters are a greater percentage in healthy ventricles. C, Approximately 25% of promoter shifts result in using a promoter 3′ downstream of the start codon, potentially adding new sequences to the 5′ end of the mRNA or amino terminus of the protein. D, A total of 124 genes were identified as potentially changing protein structure by using an alternative promoter in heart failure. An example is shown for the PRKAG2 gene. Three known isoforms of PRKAG2 are represented at the bottom (γ2a, γ2-3b, and γ2b). Above the promoter of each transcript, the cap analysis of gene expression sequencing (CAGE-seq) signal for healthy (blue) and failed (red) hearts is shown and the scales of each representation are indicated in black. E, The promoter percent usage indicates the γ2-3b form is upregulated and the γ2b form is downregulated in failed heart. Significance determined by a 2-tailed Student t test. F, The predicted amino acid sequences translated from each PRKAG2 isoform are shown (P≤0.05 [*], ≤0.005 [**]). PRKAG2 indicates protein kinase AMP-activated non-catalytic subunit gamma.

Enhancer Usage Shifts in Heart Failure

The CAGE-seq analysis identified ≈1800 LV enhancer regions in intergenic and intronic regions (Figure 1A). Clustering analysis of the normalized expression levels showed an overall similar profile of enhancer usage across the healthy LVs but disparate enhancer usage across the 4 failed LVs (Figure 6A). Comparing enhancer usage across healthy and failed LV revealed 264 enhancers that changed significantly in heart failure (raw P≤0.05). To assess whether differential enhancer transcription was associated with differential transcription factor binding site profiles, we compared transcription factor motif instances across enhancers in healthy and failed LVs. We found SMAD2, NFIX, NFAT, TCF7L2, ZNF740, and AR binding motifs enriched in enhancers that changed in heart failure. SMAD2, NFIX, TCF7L2, and AR binding motifs were found more often in downregulated enhancers. While NFAT and ZNF740 binding motifs were found more often in upregulated enhancers. RNA-seq demonstrated that the SMAD2 and NFAT5 genes were significantly upregulated in heart failure (Figure 6C).

Figure 6. Differential enhancer usage in heart failure. A, Enhancer expression clustered more tightly in the healthy hearts compared with failed hearts, which may reflect more dysregulated gene expression in heart failure. B, Some enhancers are increased while other are decreased in healthy vs failed hearts. C, Differentially used enhancers preferentially have SMAD2, NFIX, NFAT, TCF7L2, ZNF740, and AR transcription factor binding motifs. SMAD2 and NFAT were themselves differentially expressed in the healthy vs failed hearts (purple asterisks). D, An example of a differentially used enhancer is shown in the first intron of the TRPM7 gene. A reduction of enhancer RNA (eRNA) signal is seen in the failed hearts (red) compared with the healthy hearts (blue). E, eRNA expression was significantly reduced from the TRPM7 enhancer. F, Overall TRPM7 gene expression was correspondingly reduced. Additional failed hearts were added to increase power (P≤0.05 [*], ≤0.0005 [***]). FC indicates fold change; and TRPM7, transient receptor potential cation channel subfamily M member 7.

Figure 6D illustrates alternative enhancer use within the first intron of TRPM7, which encodes the transient receptor potential cation channel subfamily M member, a gene implicated in ischemic cardiomyopathy and cardiac rhythm.^40,41 This intronic enhancer showed significantly lower eRNA expression in heart failure (Figure 6E), concomitant with a significantly lower expression of TRPM7 in failed hearts (Figure 6F). We also identified an enhancer cluster upstream of the NPPA/NPPB gene loci (Figure 7). NPPA/NPPB encode natriuretic peptides, which serve as important clinical biomarkers of volume overload in heart failure. One enhancer of this cluster was detected in all samples and showed higher expression of eRNA in heart failure (Figure 7C), consistent with the regulatory region responsible for the upregulation of natriuretic proteins in heart failure.

Figure 7. An enhancer for the NPPA/NPPB genes carries the signature of activation in heart failure. A, The NPPA/NPPB locus, which encodes ANP and BNP, and is known to be upregulated in heart failure. Three upstream enhancers are shown in purple, and these were identified by cap analysis of gene expression sequencing (CAGE-seq). These same regions overlap with marks of active and open chromatin (H3K27Ac [histone H3, acetylated on lysine 27] in blue and assay for transposase-accessible chromatin sequencing [ATAC-seq] in red). B, The CAGE-seq signals of the enhancers shown in A demonstrate increased signal in failed (red) compared with healthy (blue) hearts. C, Enhancer RNA (eRNA) expression from enhancer 1 shows increased expression in failed compared with healthy hearts (P≤0.005 [**]).

Discussion

Defining the Promoterome of the Human LV in Health and Disease

Heart failure is known to be accompanied by shifts in gene expression, including the re-expression of developmentally expressed genes. We now describe genome-wide promoter usage in LV, and further highlight how promoter usage shifts in the failed LV. In total, we report ≈17 000 high likelihood promoters active in the adult human heart. We observed 2 major promoter types, the sharp TATA-box-associated and the broad CpG island-associated.³⁴ Sharp promoters had single or a few transcriptional start sites and were linked to highly expressed, tissue-specific genes like MYH7, TTN, and MYL2. Broad promoters had a wider distribution of transcriptional start sites and included both housekeeping and some tissue-specific genes. The slight but significant increase in genomewide promoter width in failed LV may suggest a loss of tight regulation of transcriptional start sites, which could derive from epigenetic modifications or transcriptional factor profile differences.

Alternative Promoter Usage in Heart Failure

Approximately 20% of genes active in the human LV have more than one active promoter, similar to what has been reported for other cell types.^42,43 Promoter switches that alter the noncoding regions can affect translational efficiency, imparting developmental and tissue specificity.⁴⁴ Some promoter switches can directly alter the aminoterminus of the resulting protein. We provide as an example a failure-linked promoter shift in the PRKAG2 gene. Mutations in PRKAG2 cause hypertrophic cardiomyopathy and arrhythmias.³⁹ In healthy LV, we found that ≈55% of transcripts originated from the γ2b promoter and ≈35% originate from the γ2-3b promoter. In failed LV ≈60% of the PRKAG2 transcripts represent the γ2b-3b isoform. The γ2b-3b isoform includes a unique 32 amino-acids that may affect the ability of the AMPK (AMP-activated protein kinase) complex to interact with troponin I and regulate contraction dynamics.⁴⁵ In the UK Biobank, a polymorphism, rs10224210, which is in the first intron of the y2-3b isoform, associates with cardiovascular disease. This specific sequence could alter γ2b-3b isoform expression through a first intron enhancer, and this association provides additional evidence that the γ2b-3b isoform may be an important mediator of heart failure. Upregulation of the γ2b-3b isoform in failed hearts may also influence how mutations in PRKAG2 are expressed. The γ2b isoform is expressed in healthy LV and in cultured fibroblasts, as seen in the GTEx data set, indicating isoform shifts might affect multiple cell types in the heart.

Differential Enhancer Usage in Heart Failure

Heart failure is associated with transcriptional changes.¹ We found that enhancer usage was more variable in failed ventricles, which may indicate genome-wide dysregulation of gene expression. SMAD2 binding motifs were enriched in differentially used heart failure enhancers, and this is highly consistent with the known upregulation of TGF (transforming growth factor)-β signaling in failing hearts.^46,47 The enrichment of this motif in differential enhancers may reflect increased TGF-β/SMAD activation in cardiomyocytes and a larger proportion of cardiac fibroblasts in the failed LV tissues. We highlighted a specific differential enhancer located within the first intron of the TRPM7 gene. TRPM7 encodes kinase domain-containing cation channel. Deletion of Trpm7 in mice disrupts cardiac automaticity and causes cardiac hypertrophy and fibrosis.⁴¹ In ischemic cardiomyopathy, TRPM7 was significantly downregulated in the left atria and ventricle.⁴⁰ Together, these findings support a reduction in TRPM7 in the setting of end-stage heart failure. Sequences upstream of the NPPA/NPPB gene loci were also identified as differentially activated in heart failure. The comparable region in the mouse genome has been shown to regulate expression of these genes, validating its function in vivo.⁴⁸ As natriuretic peptide elevation serves as a biomarker for heart failure, this enhancer region may be an attractive target to modulate natriuretic factor expression in heart failure.

Conclusions and Study Limitations

This study used CAGE-seq to define a broad spectrum of predicted cardiac promoters and enhancers, with focus on their differential use in heart failure. We identified 129 genes that had alternative 5′ sequences between failed and healthy hearts. We also identified 264 enhancers that were differentially used between the failed and healthy hearts. Due to the small cohort size, age, sex, and race could not be considered in the analysis. The majority of samples were of European descent, and controlling for age was not possible as age correlates with heart failure status. Therefore, this analysis was uncorrected for these covariates and thus may limit the broader applicability of this data. The use of bidirectional eRNA transcription as a genomewide mark of enhancer function is relatively new and the exact role of eRNAs is unknown. To address this, we compared the CAGE-seq findings to multiple independent approaches which have been used to predict enhancers. Despite this, these approaches may still over or underestimated the true number active enhancer regions. We observed variability in differential promoter and enhancer usage in failed heart, as the normal control hearts showed tighter correlations to each other. This variability may reflect the end-stage process of heart failure. While a larger data set may be more revealing, the diversity of response in the failed hearts mirrors what has been observed when RNA-seq was used to define transcripts produced for TTN, a large gene that has been examined in multiple failed hearts.^49,50 The wide array of transcripts produced from even this single gene may underscore that a lack of uniform response itself could contribute to heart failure.

Nonstandard Abbreviations and Acronyms

AMPK
AMP-activated protein kinase

ATAC-seq
assay for transposase-accessible chromatin sequencing

CAGE-seq
Capped Analysis of Gene Expression sequencing

ChIP
chromatin immunoprecipitation

CTCF
CCTCT binding factor

eRNA
enhancer RNA

H3K27ac
histone H3, acetylated on lysine 27

H3K4me1
histone H3, mono-methylated on lysine 4

H3K4me3
histone H3, tri-methylated on lysine 4

LV
left ventricle

Pol2A
RNA polymerase II

TGF
transforming growth factor

TRPM7
transient receptor potential cation channel subfamily M member

Acknowledgments

We thank the patients for their participation. We thank Dr Xinkun Wang from NUSeq and Yujiro Takegami from DNAFORM for their excellent technical support. A.M. Gacita conducted the analysis and drafted the article. L. Dellefave-Castillo secured patient consent and genotype information. P.G.T. Page assisted with genotyping. Dr Wasserstrom provided access to control samples. Drs Barefield and Puckelwartz provided helpful advice and commentary and assisted with interpretation. Drs Nobrega and McNally assisted with analysis, writing, and editing the article.

Sources of Funding

This work was supported by the National Institutes of Health (NIH) HL128075, NIH HL142187, NIH HL141698, and American Heart Association (AHA) 18CDA34110460.

Footnotes

References

1. Heinig M, Adriaens ME, Schafer S, van Deutekom HWM, Lodder EM, Ware JS, Schneider V, Felkin LE, Creemers EE, Meder B, et al.. Natural genetic variation of the cardiac transcriptome in non-diseased donors and patients with dilated cardiomyopathy.Genome Biol. 2017; 18:170. doi: 10.1186/s13059-017-1286-zCrossref Medline Google Scholar

2. Razeghi P, Young ME, Alcorn JL, Moravec CS, Frazier OH, Taegtmeyer H. Metabolic gene expression in fetal and failing human heart.Circulation. 2001; 104:2923–2931. doi: 10.1161/hc4901.100526Crossref Medline Google Scholar

3. Miyata S, Minobe W, Bristow MR, Leinwand LA. Myosin heavy chain isoform expression in the failing and nonfailing human heart.Circ Res. 2000; 86:386–390. doi: 10.1161/01.res.86.4.386Crossref Medline Google Scholar

4. Yin Z, Ren J, Guo W. Sarcomeric protein isoform transitions in cardiac muscle: a journey to heart failure.Biochim Biophys Acta. 2015; 1852:47–52. doi: 10.1016/j.bbadis.2014.11.003Crossref Medline Google Scholar

5. Anderson PA, Malouf NN, Oakeley AE, Pagani ED, Allen PD. Troponin T isoform expression in humans. A comparison among normal and failing adult heart, fetal heart, and adult and fetal skeletal muscle.Circ Res. 1991; 69:1226–1233. doi: 10.1161/01.res.69.5.1226Crossref Medline Google Scholar

6. Makarenko I, Opitz CA, Leake MC, Neagoe C, Kulke M, Gwathmey JK, del Monte F, Hajjar RJ, Linke WA. Passive stiffness changes caused by upregulation of compliant titin isoforms in human dilated cardiomyopathy hearts.Circ Res. 2004; 95:708–716. doi: 10.1161/01.RES.0000143901.37063.2fLink Google Scholar

7. Beqqali A. Alternative splicing in cardiomyopathy.Biophys Rev. 2018; 10:1061–1071. doi: 10.1007/s12551-018-0439-yCrossref Medline Google Scholar

8. Mestroni L, Brun F, Spezzacatene A, Sinagra G, Taylor MR. Genetic causes of dilated cardiomyopathy.Prog Pediatr Cardiol. 2014; 37:13–18. doi: 10.1016/j.ppedcard.2014.10.003Crossref Medline Google Scholar

9. Konno T, Chang S, Seidman JG, Seidman CE. Genetics of hypertrophic cardiomyopathy.Curr Opin Cardiol. 2010; 25:205–209. doi: 10.1097/HCO.0b013e3283375698Crossref Medline Google Scholar

10. Akerberg BN, Gu F, VanDusen NJ, Zhang X, Dong R, Li K, Zhang B, Zhou B, Sethi I, Ma Q, et al.. A reference map of murine cardiac transcription factor chromatin occupancy identifies dynamic and conserved enhancers.Nat Commun. 2019; 10:4907. doi: 10.1038/s41467-019-12812-3Crossref Medline Google Scholar

11. He A, Gu F, Hu Y, Ma Q, Ye LY, Akiyama JA, Visel A, Pennacchio LA, Pu WT. Dynamic GATA4 enhancers shape the chromatin landscape central to heart development and disease.Nat Commun. 2014; 5:4907. doi: 10.1038/ncomms5907Crossref Medline Google Scholar

12. Wamstad JA, Alexander JM, Truty RM, Shrikumar A, Li F, Eilertson KE, Ding H, Wylie JN, Pico AR, Capra JA, et al.. Dynamic and coordinated epigenetic regulation of developmental transitions in the cardiac lineage.Cell. 2012; 151:206–220. doi: 10.1016/j.cell.2012.07.035Crossref Medline Google Scholar

13. Paige SL, Thomas S, Stoick-Cooper CL, Wang H, Maves L, Sandstrom R, Pabon L, Reinecke H, Pratt G, Keller G, et al.. A temporal chromatin signature in human embryonic stem cells identifies regulators of cardiac development.Cell. 2012; 151:221–232. doi: 10.1016/j.cell.2012.08.027Crossref Medline Google Scholar

14. Rosa-Garrido M, Chapski DJ, Schmitt AD, Kimball TH, Karbassi E, Monte E, Balderas E, Pellegrini M, Shih TT, Soehalim E, et al.. High-resolution mapping of chromatin conformation in cardiac myocytes reveals structural remodeling of the epigenome in heart failure.Circulation. 2017; 136:1613–1625. doi: 10.1161/CIRCULATIONAHA.117.029430Link Google Scholar

15. May D, Blow MJ, Kaplan T, McCulley DJ, Jensen BC, Akiyama JA, Holt A, Plajzer-Frick I, Shoukry M, Wright C, et al.. Large-scale discovery of enhancers from human heart tissue.Nat Genet. 2011; 44:89–93. doi: 10.1038/ng.1006Crossref Medline Google Scholar

16. Dickel DE, Barozzi I, Zhu Y, Fukuda-Yuzawa Y, Osterwalder M, Mannion BJ, May D, Spurrell CH, Plajzer-Frick I, Pickle CS, et al.. Genome-wide compendium and functional assessment of in vivo heart enhancers.Nat Commun. 2016; 7:12923. doi: 10.1038/ncomms12923Crossref Medline Google Scholar

17. Kimura K, Wakamatsu A, Suzuki Y, Ota T, Nishikawa T, Yamashita R, Yamamoto J, Sekine M, Tsuritani K, Wakaguri H, et al.. Diversification of transcriptional modulation: large-scale identification and characterization of putative alternative promoters of human genes.Genome Res. 2006; 16:55–65. doi: 10.1101/gr.4039406Crossref Medline Google Scholar

18. Zou J, Tran D, Baalbaki M, Tang LF, Poon A, Pelonero A, Titus EW, Yuan C, Shi C, Patchava S, et al.. An internal promoter underlies the difference in disease severity between N- and C-terminal truncation mutations of Titin in zebrafish.Elife. 2015; 4:e09406. doi: 10.7554/eLife.09406Crossref Medline Google Scholar

19. Murata M, Nishiyori-Sueki H, Kojima-Ishiyama M, Carninci P, Hayashizaki Y, Itoh M. Detecting expressed genes using CAGE.Methods Mol Biol. 2014; 1164:67–85. doi: 10.1007/978-1-4939-0805-9_7Crossref Medline Google Scholar

20. De Santa F, Barozzi I, Mietton F, Ghisletti S, Polletti S, Tusi BK, Muller H, Ragoussis J, Wei CL, Natoli G. A large fraction of extragenic RNA pol II transcription sites overlap enhancers.PLoS Biol. 2010; 8:e1000384. doi: 10.1371/journal.pbio.1000384Crossref Medline Google Scholar

21. Kim TK, Hemberg M, Gray JM, Costa AM, Bear DM, Wu J, Harmin DA, Laptewicz M, Barbara-Haley K, Kuersten S, et al.. Widespread transcription at neuronal activity-regulated enhancers.Nature. 2010; 465:182–187. doi: 10.1038/nature09033Crossref Medline Google Scholar

22. Quinlan AR, Hall IM. BEDTools: a flexible suite of utilities for comparing genomic features.Bioinformatics. 2010; 26:841–842. doi: 10.1093/bioinformatics/btq033Crossref Medline Google Scholar

23. Crooks GE, Hon G, Chandonia JM, Brenner SE. WebLogo: a sequence logo generator.Genome Res. 2004; 14:1188–1190. doi: 10.1101/gr.849004Crossref Medline Google Scholar

24. Thomas PD, Campbell MJ, Kejariwal A, Mi H, Karlak B, Daverman R, Diemer K, Muruganujan A, Narechania A. PANTHER: a library of protein families and subfamilies indexed by function.Genome Res. 2003; 13:2129–2141. doi: 10.1101/gr.772403Crossref Medline Google Scholar

25. Anders S, Pyl PT, Huber W. HTSeq–a Python framework to work with high-throughput sequencing data.Bioinformatics. 2015; 31:166–169. doi: 10.1093/bioinformatics/btu638Crossref Medline Google Scholar

26. Robinson MD, McCarthy DJ, Smyth GK. edgeR: a Bioconductor package for differential expression analysis of digital gene expression data.Bioinformatics. 2010; 26:139–140. doi: 10.1093/bioinformatics/btp616Crossref Medline Google Scholar

27. Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, Batut P, Chaisson M, Gingeras TR. STAR: ultrafast universal RNA-seq aligner.Bioinformatics. 2013; 29:15–21. doi: 10.1093/bioinformatics/bts635Crossref Medline Google Scholar

28. Haberle V, Forrest AR, Hayashizaki Y, Carninci P, Lenhard B. CAGEr: precise TSS data retrieval and high-resolution promoterome mining for integrative analyses.Nucleic Acids Res. 2015; 43:e51. doi: 10.1093/nar/gkv054Crossref Medline Google Scholar

29. Thodberg M, Thieffry A, Vitting-Seerup K, Andersson R, Sandelin A. CAGEfightR: cap analysis of gene expression (CAGE) in R/Bioconductor. bioRxiv Preprint posted online April 28, 2018. doi: 10.1101/310623Google Scholar

30. Thodberg M, Thieffry A, Vitting-Seerup K, Andersson R, Sandelin A. CAGEfightR: analysis of 5′-end data using R/Bioconductor.BMC Bioinformatics. 2019; 20:487. doi: 10.1186/s12859-019-3029-5Crossref Medline Google Scholar

31. Benner C, Konovalov S, Mackintosh C, Hutt KR, Stunnenberg R, Garcia-Bassets I. Decoding a signature-based model of transcription cofactor recruitment dictated by cardinal cis-regulatory elements in proximal promoter regions.PLoS Genet. 2013; 9:e1003906. doi: 10.1371/journal.pgen.1003906Crossref Medline Google Scholar

32. Schlesinger J, Schueler M, Grunert M, Fischer JJ, Zhang Q, Krueger T, Lange M, Tönjes M, Dunkel I, Sperling SR. The cardiac transcription network modulated by Gata4, Mef2a, Nkx2.5, Srf, histone modifications, and microRNAs.PLoS Genet. 2011; 7:e1001313. doi: 10.1371/journal.pgen.1001313Crossref Medline Google Scholar

33. FANTOM Consortium and the RIKEN PMI and CLST (DGT); Riken Forrest AR, Kawaji H, Rehli M, Baillie JK, de Hoon MJ, Haberle V, Lassmann T, Kulakovskiy IV, Lizio M, Itoh M, et al.. A promoter-level mammalian expression atlas.Nature. 2014; 507:462–470. doi: 10.1038/nature13182Crossref Medline Google Scholar

34. Carninci P, Sandelin A, Lenhard B, Katayama S, Shimokawa K, Ponjavic J, Semple CA, Taylor MS, Engström PG, Frith MC, et al.. Genome-wide analysis of mammalian promoter architecture and evolution.Nat Genet. 2006; 38:626–635. doi: 10.1038/ng1789Crossref Medline Google Scholar

35. Park SG, Hannenhalli S, Choi SS. Conservation in first introns is positively associated with the number of exons within genes and the presence of regulatory epigenetic signals.BMC Genomics. 2014; 15:526. doi: 10.1186/1471-2164-15-526Crossref Medline Google Scholar

36. Visel A, Minovitsky S, Dubchak I, Pennacchio LA. VISTA enhancer browser–a database of tissue-specific human enhancers.Nucleic Acids Res. 2007; 35(database issue):D88–D92. doi: 10.1093/nar/gkl822Crossref Medline Google Scholar

37. Spurrell CH, Barozzi I, Mannion BJ, Blow MJ, Fukuda-Yuzawa Y, Afzal SY, Akiyama JA, Afzal V, Tran S, Plajzer-Frick I, et al.. Genome-wide fetalization of enhancer architecture in heart disease.bioRxiv. 2019:591362. doi: 10.1101/591362Google Scholar

38. Andersson R, Gebhard C, Miguel-Escalada I, Hoof I, Bornholdt J, Boyd M, Chen Y, Zhao X, Schmidl C, Suzuki T, et al.. An atlas of active enhancers across human cell types and tissues.Nature. 2014; 507:455–461. doi: 10.1038/nature12787Crossref Medline Google Scholar

39. Porto AG, Brun F, Severini GM, Losurdo P, Fabris E, Taylor MRG, Mestroni L, Sinagra G. Clinical spectrum of PRKAG2 syndrome.Circ Arrhythm Electrophysiol. 2016; 9:e003121. doi: 10.1161/CIRCEP.115.003121Link Google Scholar

40. Ortega A, Roselló-Lletí E, Tarazón E, Gil-Cayuela C, Lago F, González-Juanatey JR, Martinez-Dolz L, Portolés M, Rivera M. TRPM7 is down-regulated in both left atria and left ventricle of ischaemic cardiomyopathy patients and highly related to changes in ventricular function.ESC Heart Fail. 2016; 3:220–224. doi: 10.1002/ehf2.12085Crossref Medline Google Scholar

41. Sah R, Mesirca P, Van den Boogert M, Rosen J, Mably J, Mangoni ME, Clapham DE. Ion channel-kinase TRPM7 is required for maintaining cardiac automaticity.Proc Natl Acad Sci U S A. 2013; 110:e3037–e3046. doi: 10.1073/pnas.1311865110Crossref Medline Google Scholar

42. Kim TH, Barrera LO, Zheng M, Qu C, Singer MA, Richmond TA, Wu Y, Green RD, Ren B. A high-resolution map of active promoters in the human genome.Nature. 2005; 436:876–880. doi: 10.1038/nature03877Crossref Medline Google Scholar

43. Cooper SJ, Trinklein ND, Anton ED, Nguyen L, Myers RM. Comprehensive analysis of transcriptional promoter structure and function in 1% of the human genome.Genome Res. 2006; 16:1–10. doi: 10.1101/gr.4222606Crossref Medline Google Scholar

44. Davuluri RV, Suzuki Y, Sugano S, Plass C, Huang TH. The functional consequences of alternative promoter use in mammalian genomes.Trends Genet. 2008; 24:167–177. doi: 10.1016/j.tig.2008.01.008Crossref Medline Google Scholar

45. Oliveira SM, Zhang YH, Solis RS, Isackson H, Bellahcene M, Yavari A, Pinter K, Davies JK, Ge Y, Ashrafian H, et al.. AMP-activated protein kinase phosphorylates cardiac troponin I and alters contractility of murine ventricular myocytes.Circ Res. 2012; 110:1192–1201. doi: 10.1161/CIRCRESAHA.111.259952Link Google Scholar

46. Khalil H, Kanisicak O, Prasad V, Correll RN, Fu X, Schips T, Vagnozzi RJ, Liu R, Huynh T, Lee SJ, et al.. Fibroblast-specific TGF-β-Smad2/3 signaling underlies cardiac fibrosis.J Clin Invest. 2017; 127:3770–3783. doi: 10.1172/JCI94753Crossref Medline Google Scholar

47. Chen H, Moreno-Moral A, Pesce F, Devapragash N, Mancini M, Heng EL, Rotival M, Srivastava PK, Harmston N, Shkura K, et al.. WWP2 regulates pathological cardiac fibrosis by modulating SMAD2 signaling.Nat Commun. 2019; 10:3616. doi: 10.1038/s41467-019-11551-9Crossref Medline Google Scholar

48. Sergeeva IA, Hooijkaas IB, Ruijter JM, van der Made I, de Groot NE, van de Werken HJ, Creemers EE, Christoffels VM. Identification of a regulatory domain controlling the Nppa-Nppb gene cluster during heart development and stress.Development. 2016; 143:2135–2146. doi: 10.1242/dev.132019Crossref Medline Google Scholar

49. Haggerty CM, Damrauer SM, Levin MG, Birtwell D, Carey DJ, Golden AM, Hartzel DN, Hu Y, Judy R, Kelly MA, et al.. Genomics-first evaluation of Heart Disease Associated with titin-truncating variants.Circulation. 2019; 140:42–54. doi: 10.1161/CIRCULATIONAHA.119.039573Link Google Scholar

50. Roberts AM, Ware JS, Herman DS, Schafer S, Baksi J, Bick AG, Buchan RJ, Walsh R, John S, Wilkinson S, et al.. Integrated allelic, transcriptional, and phenomic dissection of the cardiac effects of titin truncations in health and disease.Sci Transl Med. 2015; 7:270ra6. doi: 10.1126/scitranslmed.3010134Crossref Medline Google Scholar