Transposable Element Dysregulation in Systemic Lupus Erythematosus and Regulation by Histone Conformation and Hsp90

Abstract

Systemic lupus erythematosus (SLE) represents an autoimmune disease in which activation of the type I interferon pathway leads to dysregulation of tolerance and the generation of autoantibodies directed against nuclear constituents. The mechanisms driving the activation of the interferon pathway in SLE have been the subject of intense investigation but are still incompletely understood. Transposable elements represent an enormous source of RNA that could potentially stimulate the cell intrinsic RNA-recognition pathway, leading to upregulation of interferons. We used RNA-seq to define transposable element families and subfamilies in three cell types in SLE and found diverse effects on transposable element expression in the three cell types and even within a given family of transposable elements. When potential mechanisms were examined, we found that Hsp90 inhibition could drive increased expression of multiple type of transposable elements. Both direct inhibition and the delivery of a heat shock itself, which redirects heat shock regulators (including Hsp90) off of basal expression promoters and onto heat shock-responsive promoters, led to increased transposable element expression. This effect was amplified by the concurrent delivery of a histone deacetylase inhibitor. We conclude that transposable elements are dysregulated in SLE and there are tissue-specific effects and locus-specific effects. The magnitude of RNAs attributable to transposable elements makes their dysregulation of critical interest in SLE where transposable element RNA complexed with proteins has been shown to drive interferon expression.

1. Introduction

Systemic lupus erythematosus (SLE) can involve any organ of the body and multiple cells types have been implicated in disease pathogenesis. The interferon signature has been identified as a consistent hallmark of peripheral blood gene expression across multiple laboratories [1–3] and can even precede the onset of disease [4]. The interferon signature is defined as expression of a group of genes that are induced by type I interferons. Multiple potential mechanisms have been posited to explain the induction of type I interferons in autoimmune diseases. Central to proposed mechanisms is the recognition of nucleic acids by toll-like receptors (TLR) or cytoplasmic nucleic acid sensors, yet there are many biologic controls to limit access of self-nucleic acids to immune stimulatory pathways. This study was undertaken to investigate the largest category of self-nucleic acids - that of transposable elements.

The largest category of self-nucleic acids is encoded by ancient remnants of viruses. Nearly half of the human genome is derived from ancient transposable elements [5, 6]. DNA transposons comprise nearly 3% of the genome and retroelements comprise over 40% of the human genome [7]. DNA transposons amplify via a cut and paste strategy whereas retroelements require an RNA intermediate transcribed by reverse transcriptase. There are both LTR and non-LTR retroelements. Short interspersed elements (SINE), which are predominantly Alu repeats, are the major non-LTR retroelement along with long interspersed elements (LINE) [8]. The LTR class comprises 8% of the human genome and is comprised largely of endogenous retroviruses (ERVs) and residual LTRs from retroviruses. ERVs can assemble into virus-like particles but are not infectious. Thus, transposable elements comprise a heterogeneous source of nucleic acids capable of stimulating interferon (Figure 1).

Figure 1. Repetitive elements in SLE.

A) the schematic diagrams represent the major families of transposable elements in the human genome. Primers used in qRT-PCR experiments are shown in the figure as asterisks. B) The Venn diagrams display the numbers of repetitive elements had significant increase or decrease (p<0.05) of expression level in SLE across the three cell types. The overlap of these elements demonstrated little concordance across cell types. These data represent RNA-seq data from three SLE patients and three controls. Overlap was defined as significant change in the same direction. Abbreviations: TIR: terminal inverted repeat, LTR: long terminal repeat, UTR: untranslated region, ORF: open reading frame,

Alu elements, ancestrally derived from 7SL RNA and transcribed by RNA polymerase III, are the most abundant of the retroelements [9, 10]. There are over a million copies and they are enriched in gene-rich regions [11]. Although Alu elements are incapable of mediating their own transposition, they can collaborate with a LINE element protein, Orf2, to achieve transposition [12, 13]. One in 20 births exhibits a new insertion of an Alu element [14]. LINE elements are also extremely common with 500,000 individual insertions, however, only 3000 of the insertions are full-length and only approximately 40 are active in somatic tissues [11, 15]. Unlike Alu elements, LINE elements are enriched in gene-poor regions. There is evidence that control of the only active LINE family, LINE-1 or L1, is not complete in somatic tissues. Brain has higher LINE-1 copy numbers than other tissues [16]. While expression is not sufficient for transposition, it is a necessary precondition and increased expression of LINE elements has been observed in various human diseases such as malignancy, illicit drug use, and in the MRL/lpr murine model of lupus [17–20].

Expression of Alu elements is difficult to analyze because 5 to 10% of messenger RNAs have an Alu element in the 3’ untranslated region [21, 22]. Alu elements are also common in introns and thus their expression can be modified by changes in expression of the encompassing messenger RNAs. Nevertheless, Alu elements are extremely important because of their abundance. They can serve as a sink for RNA-binding proteins, miRNAs, and chromatin modification enzymes through their sheer abundance [23, 24]. Recently, Alu transcripts were identified as the major nucleic acid species bound by the classic SLE autoantigen, Ro60 [25]. Ro60-Alu RNA immune complexes induce robust interferon and cytokine expression in human blood cells, dependent on TLR7/9. This pivotal study mandates a better understanding of variables governing the expression of transposable elements in SLE.

Expression of transposable elements is known to be upregulated by female hormones, pro-inflammatory cytokines, and during embryogenesis [26–30]. A central mechanism regulating expression of all transposable elements is epigenetics. Inhibition of DNA methylation or histone deacetylation de-represses transposable elements leading to expression of RNA and up-regulation of interferons as a cellular response to viral gene expression [31, 32]. The repressive histone mark, H3K9me2, is required for repression of transposable elements in embryogenesis, a particularly vulnerable time [33, 34]. Additional layers of regulation are provided by RNA interference and restriction factors such as the APOBEC family of proteins [35]. Patients with certain types of Aicardi Goutieres lose control of retrotransposition and accumulate single-stranded DNA products which then drive the induction of interferon [36]. Although there is an older literature demonstrating the accumulation of intracisternal viral particles, antibodies to ERV proteins, and retroelement RNA in patients with SLE and murine models of SLE where the interferon signature has been identified as a consistent feature [37–43], there have been very few analyses of expression in SLE using modern techniques. A recent study identified elevated expression of LINE RNAs in kidneys of patients with lupus nephritis [44]. Alu elements were found to have decreased DNA methylation in peripheral blood mononuclear cells in Behcet’s syndrome while LINE elements were normal [45]. We had previously analyzed RNA-seq of SLE monocytes and found lower expression of SINE and LINE however, we did not analyze subfamilies of transposable elements [46]. We therefore examined the expression of different transposable element families and subfamilies in three cell types from patients with SLE. We found tissue-specific effects and family type-specific dysregulated expression in SLE.

2. Methods

2.1. Samples and cell purification

The healthy donors were obtained from the Center for AIDS Research (University of Pennsylvania Perelman School of Medicine Institutional Review Board approved). The SLE samples were from the Johns Hopkins Lupus Cohort [47, 48] under a separate Institutional Review Board-approved protocol. All donors signed informed consent. Six SLE patients and six controls were used (Table 1). Peripheral blood mononuclear cells (PBMC) were purified using Ficoll-Paque. Primary monocytes were purified using adherence as previously described [46]. The purity of monocytes was >90% by flow cytometry for CD14 staining. Dynal beads (Dynabeads® CD3 and CD19, Invitrogen, Oslo, Norway) were used to separate CD3 T cells and CD19 B cells according to the manufacturer’s protocol. In most experiments, purified cells were utilized. In the studies of cell stimulation effects, both PBMCs and monocytes were used, as indicated in the figures.

Table 1.

Clinical characteristics of patients

Patient	Assays	Autoantibodies	SLEDAI	PGA	Medications
1	RNA-seq, ChIP-seq	dsDNA, Sm, RNP	0	1.4	MMF, ASA
2	RNA-seq, ChIP-seq	dsDNA	2	0.5	HCQ, NSAID
3	RNA-seq, ChIP-seq	dsDNA	0	0.5	HCQ, NSAID, ASA
4	ChIP-seq	dsDNA, Sm, Ro	6	0.5	2.5mg prednisone, MMF, HCQ
5	ChIP-seq	dsDNA	3	0.5	HCQ
6	ChIP-seq	dsDNA, Sm, Ro, RNP	16	2.5	HCQ

HCQ= hydroxychloroquine

NSAID= nonsteroidal anti-inflammatory

MMF= mycophenylate mofetil

2.2. Cell stimulation and inhibitor use

PBMCs were plated and rested for 1hr before proceeding with experiments. 17-AAG (10μM) was added to cells 20 minutes before stimulation with α2-interferon (IFNA) (500U/ml), Poly I:C (25μg/ml), PHA (3μg/mlL) and LPS (1μg/ml). IFNA was used for 24hrs alone or in combination with 6hrs of Poly I:C. Poly I:C was also used alone for 6hrs. PHA was used for 24hrs. LPS was used for 6hrs. For the heat shock experiments using monocytes, cells were rested overnight before the experiment. Trichostatin A (TSA) (330nM) was added to monocytes for 1hr followed by 1hr at 42 degrees. A549 cells and the knockout cells (KO) were plated 24hrs before the experiments. Cells were stimulated with IFNA (500U/ml) and Poly I:C (25μg/ml). IFNA was used for 24hrs alone or in combination with 6hrs of Poly IC. Poly I:C was also used for 6hrs alone.

LINE-1 overexpression was achieved using transient transfection into MonoMac6 cells. The LINE-1 expression plasmid was pBS-L1PA1-CH-mneo [49]. For controls, GFP and empty vector were transfected in parallel. Transfection was achieved using the Amaxa instrument with Nucleofector kit V (Lonza Biosystems, Morristown, NJ) and transfection protocol V001.

2.3. RNA isolation and RNA-seq

Qiagen RNeasy Kit (Valencia, CA) was used for RNA purification and DNA was removed by DNase digestion. The RNA quality was defined by RIN and OD260/280. The RNA-seq libraries were made with the Ovation® Ultralow Library Systems and sequenced on an Illumina HiSeq at BGI@CHOP. Human reference genome (hg38) was used for alignment using STAR. Aligned reads were mapped to known genes. The samples were compared in terms of their total reads mapped to genes; distribution of reads across genes; and global correlation with each other.

2.4. ChIP-seq

The antibodies for H3K4me3 and H3K27ac were from Active Motif (Carlsbad, CA). The Illumina TruSeq ChIP library preparation kit (San Diego, California) was used. Sequencing was performed on an Illumina HiSeq at BGI@CHOP. The reference genome (hg38) was indexed by the novoindex function of the NovoAlign package. Aligned reads were loaded into R. Loaded reads were filtered by SAM fields such as mapq, cigar, and flag. Loaded and filtered reads were extended to 200bp long from their end, and converted to sequencing depth at each base.

2.5. Quantitative RT-PCR

We performed qRT-PCR using primers (listed below) designed to amplify specific families of transposable elements. Interferon-inducible genes and interferon genes were similarly detected. The location of the primers is given in Figure 1. The Clontech Advantage RT for PCR kit (Clontech, Mountain View, CA) was used to generate cDNA. Gene expression was detected by real-time PCR using the TaqMan 7900. Transcript levels were normalized to the 18S signal. The equation: 2^-(ΔΔCt)= fold change. ΔΔCt is defined by (Ct gene of interest- Ct 18S)- ΔCt calibrator.

PRIMERS:
L1-ORF1 Forward: 5’ CTCGGCAGAAACCCTACAAG 3’
L1-ORF1 Reverse: 5’ CCATGTTTAGCGCTTCCTTC 3’
HERV Forward: 5’ CCAACCCCGTGCTCTCTGAA 3’
HERV Reverse: 5’ TTGTGGGGAGAGGGTCAGCA 3’
ALU-RM Forward: 5’ GCCTGTAGTCCCAGCTACT 3’
ALU-RM Reverse: 5’ GCAGTGGCGCGATCATA 3’
L1-ORF2 Forward: 5’ TGACAAACCCACAGCCAATA 3’
L1-ORF2 Reverse: 5’ CCCTGTCTTGTGCCAGTTTT 3’
L1–5’UTR Forward: 5’ TTCCGAGTCAAAGAAAGG 3’
L1–5’UTR Reverse: 5’ AGGTGTGGGATATAGTCTC 3’
ALU Forward: 5’ GCTCACGCCTGTAATCCCA 3’
ALU Reverse: 5’ GTCTCGAACTCCTGACCTCA 3’
ALU-LM Forward: 5’ AGTTCGAGACCAGCCTGG 3’
ALU-LM Reverse: 5’ CGGGTTCAAGCGATTCTCC 3’
18S –Qiagen primer QT00199367
B-Actin Forward: 5’ AAGGGACTTCCTGTAACAATGCA 3’
B-Actin Reverse: 5’ CTGGAACGGTGAAGGTGACA 3’
IFNL Forward: 5’ CGCCTTGGAAGAGTCACTCA 3’
IFNL Reverse: 5’ GAAGCCTCAGGTCCCAATTC 3’
RIGI –Qiagen primer QT00040509
NFKB1 –Qiagen primer QT00063791
IFIH1 –Qiagen primer QT00033789
IFNA2 – Qiagen primer QT00212527
IFNB1–ThermoFisher primer Hs01077958_s1
18S –ThermoFisher primer 4310893E

2.6. Bioinformatics

Location, annotation, and classification of repetitive DNA elements were downloaded from the Dfam (http://dfam.org) database. Repetitive elements were grouped into superfamilies, classes, and then types. RNA-seq and ChIP-seq reads were aligned to the GRCh38 human reference genome by STAR (Spliced Transcripts Alignment to a Reference) and Novoalign respectively [50, 51]. Reads mapped to the locations of the same repetitive elements were summed to obtain a read count of each repetitive element from each RNA-seq or ChIP-seq library. Position-specific sequencing depth of ChIP-seq libraries were calculated after extending all reads to 200bp long at the 3’ end. Read counts were adjusted to total number of aligned reads of each library and log2-transformed. Comparison between control and SLE groups was performed using the limma method [52]

2.7. Ethics

This study was conducted in compliance with the ethical principles originating in, or derived from, the Declaration of Helsinki, and in compliance with all International Conference on Harmonization Good Clinical Practice guidelines. The study and all relevant documentation were reviewed and approved by the Johns Hopkins University School of Medicine Institutional Review Board. All subjects gave written informed consent.

3. Results

3.1. Transposable elements are dysregulated in SLE

The regulation of expression of transposable elements is incompletely understood and there have been few studies in human disease states. Transposable elements can exhibit somatic diversity and therefore purified cell types were utilized to understand whether the expression of transposable elements is altered in SLE [16]. RNA-seq was performed in purified CD3 T cells, CD19 B cells, and monocytes in three female patients with SLE and three female disease-free controls. We examined classes of elements that could be discerned through sequence analysis. Categories of simple repeats, including satellite DNA were used as controls. We defined classes of transposable elements according to standard nomenclature. Figure 1A displays the structures of the elements. Of 1266 defined elements, 202,117, and 109 had altered expression in B cells, monocytes, and T cells, respectively in SLE (Figure 1B). No repetitive element was increased in all three cell types and only 10 were decreased in all three cell types, all of which belong to the LTR class. Their direction of change differed and the affected cell type differed but this initial analysis revealed that repetitive elements are clearly impacted by the disease state. These data suggest that the dysregulation was not due to a shared mechanism conserved across all three cell types.

We reasoned that we could define central mechanisms of dysregulation through a more detailed analysis of the types of transposable elements affected in SLE. We defined individual sub-families of DNA transposons and retrotransposons for each cell type comparing SLE to controls (Figure 2). Among DNA transposons, the cut-and-paste families were generally up-regulated in SLE, with the smallest sub-family of Tip100 having decreased expression. ERVs and SINE elements were generally downregulated in SLE again with some sub-families exhibiting increased expression in one or more cell types. SINE (Alu) expression is linked to that of the mRNA in which they are often embedded [21, 22]. In our previous work, we found a global down-regulation of mRNAs in SLE and therefore the finding of decreased Alu expression is concordant with the previous finding [46, 53]. In contrast, LINE elements were generally upregulated in monocytes and B cells but not T cells. The LINE-2 subfamily is the exception with global downregulation that was highly variable from locus to locus. When individual LINE elements were examined, B cells exhibited >1.5 log fold change of just one LINE element, X1-LINE (p=7 X 10⁻⁷), a CR1-type LINE element. In monocytes, one LINE element exhibited >1.5 fold log change in SLE: X2-LINE (p=0.0003), also a CR1-type LINE element. T cells did not express any LINE element to a greater extent in SLE than in controls. T cells exhibited increased expression of three ERV elements (MER548, MER98, LTR59) and three cut-and-paste DNA transposons (MARE4, UCON14, Charlie12). The CR1 family of LINE elements appeared to have overall the highest expression in SLE. We therefore examined this family in detail. Among 25 CR1 LINE family members, expression ranged from undetectable in all three cell types (X7D and X5B) to many thousands of reads in all three cell types (L3). While expression of the CR1 family was generally increased in SLE in B cells and monocytes, each individual LINE element generally had a unique pattern of expression in the three cell types (Table 2). We found two transposable elements with markedly decreased expression across all three cell types: ERVK and ERV1. These two endogenous retroviruses are among the most recently acquired retroelements [54, 55]. Therefore, this analysis of transposable elements demonstrated family-specific effects but with highly specific disease effects at the level of individual elements. This is somewhat surprising since sequence conservation is high within a given family.

Figure 2. Expression of sub-families of transposable elements.

Each bar represents the average percentage change and its standard error of a superfamily in one cell type. Only superfamilies including more than 10 elements were plotted. Each title includes class name, superfamily name, and the number elements.The effect is not uniform across all repetitive elements nor across all three cell types. B cells (red), monocytes (blue), and T cells (green) are represented as individual bars for each sub-family of transposable element. The error bars represent standard error. The sample size is derived from the RNA-seq data from three SLE patients and three controls. Signficant change from controls was seen in B cells for hAT-Charlie (p=0.004), LINE-1 (p=0.003), ERV1 (p=0.0002), ERVL-MaLR (p=0.0008), Alu (p=10⁻¹¹). In monocytes, Mar-Tigger (p=0.009), LINE-1 (p=0.002), ERV1 (p=10⁻¹⁰), and Alu (p=0.0001) were significant. In T cells, Alu (p=10⁻¹¹) was significant.

Table 2.

CR1 LINE element expression

	B mean control	B mean SLE	B log FC	B p value	M mean control	M mean SLE	M log FC	M p value	T mean control	T mean SLE	T log FC	T p value
X21_LINE	0.5051	0.6854	0.2889	1	0.28	1.3287	1.3837	0.42	0.503	1.6257	1.637	0.25
CR1–11_Crp	0.0026	0.4614	2.1308	1	0.6363	0.3625	-0.975	1	1.1574	2.5652	0.9438	0.27
CR1-L3B_Croc	1.5804	1.8319	0.0011	1	1.285	3.1569	1.8579	0.34	0.8777	1.7309	0.9311	0.74
CR1–13_AMi	6.0575	10.7723	0.8209	0.15	3.1842	2.6442	−0.2452	1	4.8557	8.3516	0.74	0.24
X7C_LINE	33.4833	35.9951	0.1005	0.77	21.6312	31.7686	0.5923	0.28	22.0492	35.7954	0.7037	0.042
CR1–3_Croc	23.8647	30.6034	0.3644	0.3	21.3718	20.9643	−0.0591	1	25.169	33.0745	0.3999	0.27
X8_LINE	12.4148	5.1701	−1.2179	0.035	8.1038	10.3687	0.381	0.81	7.3419	9.2806	0.3735	0.63
CR1–1_Amn	19.3762	38.1052	0.9714	0.0035	24.1073	14.9565	−0.6795	0.32	19.452	24.1999	0.3245	0.47
X6A_LINE	17.2241	24.8258	0.5267	0.2	12.9183	20.0214	0.6676	0.28	22.6936	26.056	0.2138	0.61
X1_LINE	19.6699	65.1097	1.7097	7.20E-07	12.7504	10.1151	−0.3783	0.63	13.2609	15.2398	0.2087	0.7
L3b_3end	734.2814	814.5706	0.1525	0.3	667.2304	811.4538	0.275	0.26	852.5902	910.0707	0.102	0.53
CR1–16_AMi	61.9315	93.8623	0.5998	0.019	54.9236	87.9612	0.6629	0.076	79.166	84.0009	0.0922	0.78
X20_LINE	2.0987	0.6807	−1.424	0.31	1.3487	4.9627	1.6101	0.074	1.5494	1.7442	0.0864	1
X17_LINE	21.0925	26.4039	0.3218	0.4	19.3269	15.0094	−0.3479	0.44	20.1407	20.812	0.0587	0.95
L3	8858.1464	9093.7943	0.0417	0.74	8879.3724	9867.6675	0.1447	0.49	10332.4362	10633.0316	0.0483	0.74
CR1_Mam	715.5417	704.6769	−0.0188	0.91	662.2284	823.8186	0.3076	0.22	838.715	834.4505	0.0005	1
X5B_LINE	0	0	0	1	0	0	0	1	0	0	0	1
X7D_LINE	0	0	0	1	0	0	0	1	0	0	0	1
X7B_LINE	15.0906	15.1183	0.0152	1	15.583	21.3983	0.4931	0.44	22.5276	21.4562	−0.0718	0.91
Plat_L3	775.8389	780.9666	0.0118	0.94	844.5409	989.1516	0.2232	0.41	1002.1257	923.6114	−0.1108	0.5
X2_LINE	23.9422	11.5563	−1.0336	0.027	26.4463	77.9676	1.5772	0.00028	25.8591	23.6143	−0.1259	0.8
X7A_LINE	80.1888	63.2915	−0.3377	0.22	63.7113	73.6497	0.2109	0.66	68.8062	56.6391	−0.2793	0.4
CR1–12_AMi	9.288	10.6024	0.1922	0.75	7.2773	9.1757	0.2266	0.67	13.9201	10.2878	−0.4532	0.42
X6B_LINE	6.1049	8.5474	0.4882	0.36	7.2186	9.3907	0.3126	0.57	12.2151	8.3959	−0.5359	0.32
CR1-L3A_Croc	4.6525	5.0763	0.1043	1	4.0712	4.2879	0.1524	1	14.4635	7.7513	−0.9209	0.13

3.2. Non-coding RNA expression in SLE

As a relevant comparison, we examined other non-coding RNAs in SLE. Satellite RNA and 5S RNA expression levels were clearly different in SLE but not in a pattern similar to any of the families of transposable elements. Nor were they similar to each other (Figure 3). Splicing RNAs, snRNAs, and snoRNAs, also had no uniform direction of change in SLE (not shown). We had previously demonstrated an association between the direction of long non-coding RNAs (lncRNAs) and the expression of nearby mRNAs [46]. In this current study, we confirmed that the lncRNAs located on the opposite strand upstream of coding genes generally had the same direction of differential expression in SLE as the corresponding coding genes (Figure 4). The positive correlation was extended until the distance between coding genes and lncRNAs was about 250kb. These might represent promoter proximal enhancer RNAs regulating the adjacent mRNAs. Therefore, the effect on transposable elements was itself diverse and did not parallel that of any other no-coding RNA that we could identify, suggesting that the effect was specific to transposable elements. In spite of high sequence homology within families of transposable elements, dysregulated expression in SLE was highly diverse. We hypothesized that local epigenetic control could account for locus specificity.

Figure 3. Differential expression of Satellite and 5S RNAs in SLE.

Each bar represents the average fold change of repetitive elements in the Satellite class (n = 26) and 5S superfamily (n =1 ) and its standard error. The three cell types: B cells (red), monocytes (blue), and T cells (green) are represented as individual bars. The differences found in SLE are not consistent with those found for transposable elements nor with each other. None of the changes are statistically significant.

Figure 4. Long non-coding RNAs (lncRNAs) near coding genes.

lncRNAs on the opposite strand and upstream of differentially expressed coding genes were grouped by their distance to the TSS of the corresponding coding genes. Bars are average change of lncRNAs in SLE. The greatest change in lncRNA expression near differentially expressed coding genes was close to the TSS for down-regulated genes and 1–2kb upstream for upregulated genes. The error bars represent standard error. The sample size is derived from the RNA-seq data from three SLE patients and six controls. P values are not reported.

3.3. Chromatin features at transposable elements

We examined H3K4me3 and H3K27ac Chip-seq libraries from a previously published analysis [53], focusing on the repetitive elements previously filtered out. We aligned the subfamilies of transposable elements according to their start site and asked if there were chromatin features consistently observed near transposable elements (Figure 5A). Nearly all subfamilies examined exhibited a discontinuity between the surrounding chromatin (to the left of the TSS) and the element itself. The CR1 LINE family, the most ancient family, stood out as having less distinct chromatin at the site of the repeat. The finding of element-specific chromatin suggests that most transposable element families are uniquely regulated and not simply subject to the control of the surrounding genes. We then defined differences between control and SLE cells at these elements (Figure 5B). The differences shown represent change in SLE and all subfamilies demonstrate change that renders them more like the surrounding chromatin. We then asked whether the change in H3K4me3 was associated with concordant change in RNA levels. H3K4me3 is often taken as a mark of transcriptional activation at promoters. In no case was there any statistical association between change in RNA and H3K4me3 in SLE. Two representative examples are shown in Figure 5C. We then asked if H3K27ac was associated with transposable elements. We had data on T cells and B cells in SLE patients and controls. H3K27ac is present at promoters but is believed to be more critical for enhancer function. H3K27ac and H3K4me3 changes in SLE were inversely related and H3K27ac was generally decreased in SLE compared to controls in both B cells and T cells (Figure 6A). Whereas Alu (SINE) H3K4me3 increased in SLE, H3K27ac decreased in both cell types (Figure 6B). These data implicated chromatin as a layer that contributes to the regulation of transposable element expression, but the lack of association with expression suggested there were other contributors that were more impactful. We therefore attempted to replicate the dysregulated expression by modulating known pathways involved in the regulation of expression of transposable elements.

Figure 5. H3K4me3 at transposable elements.

A) Selected transposable element subfamilies were examined to define differences in H3K4me3 at the 5’ end of the element compared to the surrounding chromatin. The y-axis is the average sequencing depth of six control samples. Many elements have undergone deletion and therefore the elements are aligned by the 5’ ends at the TSS. Discontinuity from surrounding upstream chromatin is nearly universal. The CR1 subfamily of LINE element is the exception. The color code for the lines is: B cells (red), monocytes (blue), and T cells (green). Only superfamilies with more than 10 elements were plotted. B) The change in H3K4me3 for each cell type displayed according to position. The y-axis is the ratio of average sequencing depth of SLE samples over average of control samples. All changes reduce the difference between the element and the upstream chromatin. C) RNA changes (red) and H3K4me3 changes (blue) related to SLE are displayed for two elements as examples. In no case was there an association across a subfamily or within a cell type. The error bars represent standard error. The sample size is derived from the RNA-seq data from three SLE patients and three controls and the ChIP-seq data were derived from six SLE patients and six controls. Significant differences were seen in B cells for Mar-Tc2 (p=0.007), LINE1 (p=10⁻²⁷), ERVL (p=10⁻³¹), ERVL MaLR (p=10⁻²⁹), Gypsy (p=10⁻⁹) and Alu (p=10⁻¹⁴). In monocytes, the significant differences were hAT-Charlie (p=10⁻⁹), Mar-Tigger (p=10⁻⁶), ERVK (p=10⁻⁷), ERVL (p=10⁻¹²), ERVL MaLR (p=10⁻¹³), and Alu (p=10⁻¹⁸). In T cells, significant differences were seen for hAT-Charlie (p=0.003), LINE-1 (p=0.004), ERVL (p=10⁻¹⁶), and ERVL MaLR (p=10⁻¹⁹).

Figure 6. H3K27ac at transposable elements.

A) H3K4me3 and H3K27ac were plotted as the change in SLE to define concordant changes. Unexpectedly, they are anti-correlated with H3K4me3 generally increased in T cells and B cells and H3K27ac generally decreased in SLE compared to controls at transposable elements. B) The transposable element families were defined with H3K27ac change in SLE plotted for T cells and B cells separately. Monocyte data was not available. The error bars represent standard error. The sample size is derived from the ChIP-seq data from six SLE patients and six controls. Signficant change from controls was seen in B cells for Mar-Tigger (p=10⁻²¹), CR1 (p=10⁻⁵), ERV1 (p=10⁻¹⁶), ERVL (p=10⁻¹⁴), ERV-MaLR (p=10⁻¹⁵). In T cells, hAT-Charlie (p=10⁻¹⁵), Tip100 (p=10⁻⁷), Mar-Tigger (10⁻⁹), ERV1 (p=10⁻²⁸), ERVL-MaLR (p=10⁻¹⁶), and Alu (p=10⁻⁵).

3.4. Restriction of expression of transposable elements

The regulation of expression of transposable elements is not fully understood. Retrotransposition is inhibited by proteins implicated in Aicardi Goutieres syndrome such as SAMHD1, ADAR, TREX1, and RNASEH2A, however, these proteins do not inhibit RNA production from transposable elements [56–59]. We therefore undertook an analysis to understand whether endogenous restriction elements control the expression of transposable elements. We utilized a panel of cell lines where OAS1, OAS2, OAS3, p150 ADAR1, RNASEL, and RNASEL+ADAR1 of the cytoplasmic nucleic acid pathway were knocked out in A549 cells using CRISPR-Cas9 [60]. The experimental design allowed us to define effects on transposable elements, genes downstream of type I interferon (IFIH1, NFKB1, RIGI), IFNB1 as an example of a type I interferon and IFNL as an example of a type III interferon. ACTB was included as a cellular mRNA that could be cleaved by the RNAaseL pathway for comparison. The OAS proteins bind dsRNA and in the presence of ATP, generate 2’−5’ oligoadenylate structures which activate the latent RNaseL monomers, leading to cleavage of viral and host RNAs. ADAR1 blocks the recognition of RNA by RIGI and MDA5, the main recognition receptors for dsRNA and complex structures of RNA [61]. Therefore, this panel of KO cells allowed us to probe both the endogenous recognition and cleavage pathway as well as the induction of interferons and downstream genes.

We stimulated the cells with α2-interferon (IFNA) and poly I:C, a protocol utilized in neonatal mice to activate retrotransposition [62–64]. We used the ribosomal RNA 18S to calibrate our qRT-PCR results (Figure 7). Many RNAs can be targeted by RNaseL and ribosomal RNAs have been described as both targets and non-targets [65, 66]. To understand the effects of stimulation on a well-characterized cellular mRNA, we first analyzed the ACTB mRNA, encoding actin (Figure 7A). Poly I:C either with or without α2-interferon (IFNA) pretreatment was associated with significant diminution in ACTB transcript levels that was dependent on the RNAseL system. Analysis of the transposable elements, using primers to various regions of the elements, revealed far less of an effect on RNA abundance compared to ACTB (Figure 7B). Poly I:C treatment was associated with increased levels of nearly all transposable element transcripts in the parent A549 cell line. Importantly, unlike ACTB, there was no clear diminution related to poly I:C treatment that was dependent on RNaseL. Instead, there appeared to be a component of control by p150 Adar for the retroelements but not ERV in poly I:C stimulated cells. This was not observed in cells treated with both IFNA and poly I:C. In cells pretreated with IFNA, OAS1 appeared to have a more significant effect, likely because OAS1 is expressed at low levels in resting cells and is highly upregulated by type I interferons [67]. Our underlying model is that upregulation of expression of transposable elements could contribute to the interferon signature. We therefore utilized this model to examine effects on interferon expression and interferon-inducible genes.

Figure 7. Transposable element expression in knockout cells.

A) Actin expression in the parent cell line, A549 is reduced after poly I:C treatment with our without IFNA pretreatment. This is due to activation of the RNaseL pathway as RNASEL KO mitigates the observed reduction. B) The transposable elements are graphed similarly according to KO cell and stimulus. The non-LTR retroelements exhibit a peak in the p150 Adar KO cells that is not seen with ERV. C) Interferon effects were examined in the same KO cell lines treated with the same stimuli. Both interferon and genes downstream of type I interferon signaling are increased in stimulated cells. The effects of the different KO are diverse. D) The data from parts A-C are represented as a heat map for visual comparison. N=3. For visualization, error bars are not indicated in this figure. Significant differences are reported in the text. Labelling: Knockout (KO) cells were all generated in A549 cells. OAS1, OAS2, OAS3 are as indicated. P150 Adar KO represents specific KO of the p150 isoform. RL+A= RNaseL and p150 Adar KO. RNaseL KO is as indicated. The stimuli are abbreviated: No stim= mock stimulated, IFNA=α2-interferon for 24 hours, Poly IC= poly I:C for 6 hours, AP=α2-interferon for 18 hours with poly I:C added for another 6 hours.

3.5. Interferon expression in cells with dysregulated expression of transposable elements

Uncleaved transposable element RNA can induce interferons [25, 68, 69]. Based on this, we examined the levels of a type I interferon (IFNB1) and a type III interferon (IFNL) in the same cells. There was a striking discordance between the two with IFNB1 transcripts strongly induced by poly I:C with or without pretreatment with IFNA (Figure 7C). This induction was markedly less in the ADAR and RNASEL KO cells. In contrast, IFNL induction was augmented in the cells lacking RNaseL. When we evaluated a set of genes that are part of the interferon signature seen in SLE and other autoimmune diseases (IFIH1, NFKB1, IRF1), the pattern more closely resembled that of λ-interferon than β-interferon. Focusing on the transposable element RNAs, it is clear that they are far less inducible than the interferons or the genes downstream of type I interferons (Figure 7D). Nevertheless, they are inducible and appear to have some control imposed by p150 Adar. We concluded from this that circulating RNAs, observed in patients with SLE and similar to poly I:C, could provide a stimulus to induce expression of transposable elements [70, 71]. The RNA produced by transposable elements could represent a key element in a feed-forward loop.

3.6. LINE1 over-expression induces type I interferon.

Previous reports had suggested that LINE-1 over-expression could induce the expression of type I interferons. We tested that directly by transfecting in a full length LINE-1 expressing plasmid. We performed qRT-PCR for two targets, IFNB and IFNA2, two highly expressed type I interferons. A monocyte cell line was utilized and interferon targets were assayed at 6 and 16 hours post-transfection. Both 6 and 16 hours gave comparable results and the 6 hour results is displayed in Figure 8A. The LINE-1 expressing plasmid induced far more type I interferon than the empty vector or the GFP controls. Thus, increased LINE-1 RNA can contribute to the type I interferon signature seen in SLE.

Figure 8. Gene regulation.

A) MonoMac6 cells are a mature monocyte cell line. They were transfected with expression plasmids for GFP or LINE-1 and 6 hours later, RNA was purified from the cells and qRT-PCR performed to detect production of type I interferon mRNAs. Both IFNB and IFNA2 transcripts were increased at 6 hours post-transfection, demonstrating the ability of LINE-1 RNA to induce expression of type I interferons. B) 17AAG was used as an Hsp90 inhibitor in healthy PBMCs. Immune stimuli were used with or without 17AAG. As a class, the 17AAG-treated cells had higher expression than the cells stimulated without 17AAG. Error bars represent standard deviation. The p value represents ANOVA for 17AAG effect against stimuli without 17AAG. C) Heat shock, TSA or both were used in primary human monocytes and transposable element RNA was measured by qRT-PCR. The combination was effective at inducing expression. The p value represents ANOVA for the combination effect against unstimulated. Error bars represent standard deviation.

3.7. Heat shock in the control of expression of transposable elements

Recent studies have identified a role for heat shock proteins, in particular Hsp90, in the regulation of transposable element expression and we recently published an analysis of transcriptional changes in SLE where diminished heat shock gene expression was one of the most consistent features across different cell types [53, 72]. Hsp90 represses both RNA expression and retrotransposition [72, 73]. Key aspects of Hsp90 repression of transposable element expression include synergy with other cellular stressors and locus-specific effects [72, 73]. We tested whether stimuli designed to mimic immune activation in SLE could induce expression of transposable elements and found a very modest effect. When partnered with 17AAG, an Hsp90 inhibitor, the stimuli significantly augmented expression of both LINE-1 and ALU elements, defined by qRT-PCR (17AAG effect ANOVA p=0.02; Figure 8B). In the original observation in SLE sample, LINE-1 elements were over-expressed in monocytes in SLE and ALU was under-expressed compared to controls. Thus, it is significant that both were increased after dual treatment with 17AAG and stimulation. These data are significant in linking our observation of diminished heat shock gene expression in SLE with the increased expression of transposable elements.

In a model Candida system, heat shock caused 18% of all cellular genes to be up-regulated and most were dependent on Hsp90 [74]. We therefore hypothesized that heat shock might lead to disruption of the normal control of transposable elements by Hsp90 as it drove re-localization of Hsp90 from transposable elements to heat shock-inducible genes. We focused on monocytes as these cells had the greatest decrement in HSP90 in SLE in our previous RNA-seq study (−1.96 log fold change, p value 2.6E-06 compared to the control monocytes). Heat shock on its own had a minimal effect on transposable element expression (Figure 8C). We paired it with a strong inhibitor of histone de-acetylases, Trichostatin A (TSA). With a demonstrated effect of histone acetylation in murine ERV regulation [72], we reasoned that the chromatin may limit the inducibility of the transposable elements. Indeed, neither TSA nor heat shock had a significant effect on transposable element expression, however, when used in combination, ANOVA demonstrated a significant effect with p=0.0005. Therefore, biological stressors such as heat shock could drive increased expression of transposable elements. This effect required histone deacetylate inhibitor treatment supporting a role for chromatin in the locus specific effects of Hsp90 inhibition.

4. Discussion

One theme to emerge from this study is the dysregulated expression of a set of repetitive elements, as has been previously described in renal tissue from patients with lupus nephritis and minor salivary gland tissue from patients with the closely related Sjogren’s syndrome [44]. Repetitive elements comprise nearly half of the human genome [75] and therefore represent an enormous source of self-nucleic acids. We found dysregulated expression of repetitive elements across all three cell types from patients with SLE but with tissue-specific effects and strongly dependent on the type of transposable element. RNA-seq provided a very detailed analysis of the transcripts related to repetitive elements. Although it is difficult to assign all reads to a specific locus, the families and sub-families are sufficiently distinct that abundance can be assigned categorically. The CR1 subfamily of LINE elements was the most significantly upregulated in monocytes and B cells from SLE patients. This family of LINE elements is the most ancient of the retroelements, arising before the divergence of birds and reptiles [76]. Regulation of expression of the CR1 family of LINE elements is not known to be distinct from that of other LINE elements, however, little is understood of the transcriptional regulation. We noted that the local chromatin was least distinct around the CR1 LINE elements, perhaps suggesting loss of chromatin-mediated supression of this family over years of evolution. Control of CR1 retrotransposons in chicken cells is independent of Dicer while other LINE families require Dicer for control of transcription and retrotransposition, supporting the concept of significant differences between this ancient CR1 family and more recently evolved LINE families [77]. In our study, LINE-1 and LINE-2 were less dysregulated than the CR1 family and were discordantly regulated with LINE-1 slightly increased in SLE and LINE-2 slightly decreased in SLE. Thus, it is clear that there are family differences and the effects of SLE are not identical in all LINE families nor in the three cell types. For LINE elements, B cells and monocytes appeared to have more dysregulated expression than T cells. This pattern was not seen in the other transposable element families. SINEs accumulate in gene-rich regions with high GC content, while LINEs insert into AT-rich regions [78–80], thus suggesting that their mechanisms of regulation may be distinct. Each ERV subfamily exhibited unique patterns of expression. This pattern of dysregulation is different than that seen in cancer which mimics the expression in stem cells with a more consistent picture of upregulated expression for SINEs. As cells differentiate, SINE expression becomes more restricted in a tissue-specific manner [81]. An examination of LINE expression in prostate cancer cells also found heterogeneous effects within the family [82]. Therefore, it appears as though expression of individual elements is heterogeneous outside of the setting of stem cells and cancer cells.

The motivation for examination of transposable elements was our desire to understand whether large quantitative changes in transcription might contribute to the interferon signature. Host RNAs can participate and drive the expression of type I interferons and transposable element RNA may have unique structural features which contribute to recognition and stimulation of type I interferons [25]. Upon transfection of LINE-1 expressing plasmids, type I interferon was induced in a cell line, supporting the concept that LINE-1 RNA exhibits unique structural features related to interferon induction.

Regulation of expression of transposable elements is not well understood. Ultraviolet light and aryl hydrocarbons are known to upregulate LINE expression in general [83, 84] and lymphocyte activation and microbial products are two other recognized stimuli [85–87]. To understand what mechanisms might underlie the dysregulated expression, particularly up-regulation of RNA, we stimulated cells from healthy donors. The effects of the stimuli were individually modest but were increased with Hsp90 inhibition. The rationale for this approach was a finding in our previous study that heat shock protein mRNAs were globally decreased in SLE [53]. The reason for this is not understood, however, it suggested a pathway that could be central to the dysregulated expression of transposable elements as Hsp90 has been reported to control LINE elements [72]. Indeed, 17AAG synergized with the other stimuli and significantly increased expression across all stimuli when considered as an independent variable in ANOVA analysis.

As an independent approach, we also used heat shock in human primary monocytes to induce expression of transposable elements. Heat shock leads to redistribution of transcriptional regulators for heat shock to massively up-regulate the expression of heat shock genes, required to limit cell damage from the heat shock event. Heat shock itself caused somewhat increased expression across the retroelements, however, the histone deacetylase inhibitor, TSA, synergized with heat shock to increase expression dramatically. Hsp90 inhibition has been demonstrated to improve disease progression in a murine model of lupus, seemingly contrary to our study, however, it has also been demonstrated to up-regulate interferon response genes, consistent with our analysis [88, 89]. The complex in vivo effects of Hsp90 inhibition include immune modulatory effects on MHC expression, recruitment of lymphocytes, and co-stimulatory molecule effects [90]. This makes it difficult to align in vivo responses with those seen in pure cell populations. Histone acetylation was targeted specifically due to our finding of overall decreased H3K27ac at transposable elements in SLE. These data support a strong link between our findings of decreased heat shock gene expression and upregulation of some families of transposable elements and supports our hypothesis that for families or subfamilies of transposable elements unaffected by SLE or heat shock, that chromatin may provide another layer of regulation that restricts consistent, universal dysregulation. Chromatin is known to have altered characteristics in SLE. Murine models treated with histone deacetylase inhibitors have demonstrated improvement in disease parameters which appears contrary to our data [91]. The effects of histone deacetylase inhibitors, however, are known to be immune suppressive and have effects on acetylation of many signaling molecules [92, 93]. Similarly to Hsp90 inhibition, histone deacetylase inhibitor effects in intact organisms may be complex and reflect off-target phenomenon. Our data demonstrate a novel mechanism regulating expression of transposable elements in SLE and support additional studies defining potential drug targets directed at this specific effect.

This study is among the first to comprehensively define transposable element expression in a human disease and we found broad but diverse dysregulation in SLE. We also found that transposable element transcripts are less susceptible to RNaseL targeting than ACTB transcripts. Heterogeneity of effects has been previously described [66], however, transposable elements are thought to be regulated by the RNaseL/Adar pathway [56, 94, 95]. Thus, our finding was unexpected and may relate to tissue-specific effects [96]. This is the first study in SLE using next generation sequencing to demonstrate discordant dysregulation between cell types and between families of transposable elements. The lesson is that while these elements have traditionally been treated as a unified category of the human genome, they behave in ways as diverse as that of mRNAs. Although the use of next generation sequencing provided unbiased reporting of transcriptional features and gave us unparalleled detail regarding families and subfamilies, there are still some important limitations to this study. The sample size was small, dictated by our pairing of RNA-seq and ChIP-seq. The patients in this study had low disease activity to minimize the effect of medications and therefore it is unknown whether these findings would be more or less extreme in patients with very active disease. This study is also limited to family and subfamily descriptions. Table 2 and the error bars in Figures 2–5 hint at the diversity at an individual locus level. Our data strongly support a chromatin mechanism for the regulation of expression of individual transposable elements.

The importance of this study is the finding that there is a strong biological role for Hsp90 in the regulation of human transposable elements. Inhibition of Hsp90 and heat shock where Hsp90 is redirected from its usual genomic locations to heat shock protein promoters both led to de-repression of retroelements. HSP90 mRNA production is regulated primarily by the heat shock family of transcription factors, HSF1, 2, 3, 4, however, HSF family member mRNA levels were not significantly different in SLE compared to controls. Thus, the mechanism of altered heat shock protein expression in SLE is thus not yet understood. Nevertheless, Hsp90 effects have been demonstrated to be locus-specific and impacted by cellular stressors [72]. Our findings provide critical new information and offer a framework for understanding recognized epidemiologic contributors to the etiology of SLE, all recognized cellular stressors. Opportunities to modulate Hsp90 may be able to reset the transposable element expression pattern in SLE.

5. Conclusions

Transposable elements are expressed in a disease-specific fashion in SLE with unique patterns in sorted cells. We identified three critical pathways regulating transposable element expression that may be affected in SLE: histone acetylation, endogenous restriction pathways, particularly p150 Adar, and Hsp90. Expression could be upregulated in vitro with nucleic acid stimulation, heat shock, and with an Hsp90 inhibitor. The impact of these findings relates to the role of nucleic acids in driving the characteristic interferon signature seen in SLE.

Highlights.

• Transposable element subfamilies have dysregulated expression in SLE

• Transposable elements have distinctive chromatin

• Restriction of transposable element expression, particularly Alu elements, was provided by p150 Adar and Oas1

• Expression of transposable elements could be induced through inhibition of Hsp90 or by heat shock

• Histone deacetylase inhibition led to increased expression of transposable elements in the setting of heat shock

¹Abbreviations

17-AAG

17-Demethoxy-17-(2-propenylamino) geldanamycin

APOBEC

Apolipoprotein B mRNA Editing Enzyme Catalytic Polypeptide-Like

ERV

Endogenous retrovirus

H3K4me3

Histone 3, lysine 4 trimethylated

H3K27ac

Histone 3, lysine 27 acetylated

Hsp90

Heat shock protein 90

IFNA

∝-Interferon

LINE

Long interspersed element

lncRNA

Long non-coding RNA

LPS

Lipopolysaccharide

LTR

Long terminal repeat

PBMC

Peripheral blood mononuclear cells

PHA

Phytohemagglutinin

SINE

Short interspersed element

SLE

Systemic lupus Erythematosus

snRNA

Small nuclear RNA

snoRNA

Small nucleolar RNA

TLR

Toll-like receptor