Catching RNAs on chromatin using hybridization capture methods

Abstract

The growing appreciation of the importance of long noncoding RNAs (lncRNAs), together with the awareness that some of these RNAs are associated with chromatin, has inspired the development of methods to detect their sites of interaction on a genome-wide scale at high resolution. Hybridization capture methods combine antisense oligonucleotide hybridization with enrichment of RNA from cross-linked chromatin extracts. These techniques have provided insight into lncRNA localization and the interactions of lncRNAs with protein to better understand biological roles of lncRNAs. Here, we review the core principles of hybridization capture methods, focusing on the three most commonly used protocols: capture hybridization analysis of RNA targets (CHART), chromatin isolation by RNA purification (ChIRP) and RNA affinity purification (RAP). We highlight the general principles of these techniques and discuss how differences in experimental procedures present distinct challenges to help researchers using these protocols or, more generally, interpreting the results of hybridization capture experiments.

Introduction

The predictable base pairing of oligonucleotides to RNA has led to a multitude of biochemical enrichment strategies to study the functional roles of RNA in a cell. These techniques were inspired by the success of in situ hybridization, in which RNA is detected through hybridization with a radioactive or fluorescently labeled antisense oligonucleotide [1, 2]. Beyond visualizing RNA, antisense oligonucleotides have proven to be powerful affinity reagents for biochemically enriching cellular RNAs under native conditions together with their associated proteins. The discovery of noncoding RNAs that bind to chromatin, such as Xist in mammals [3, 4], and roX1/roX2 in Drosophila [5], raised the need for a new class of RNA-centric methods to isolate chromatin associated with an RNA of interest, an approach analogous to chromatin immunoprecipitation (ChIP). Pioneering work demonstrating the feasibility of enriching RNAs from cross-linked chromatin preparations focused on Alu noncoding RNAs with a quantitative polymerase chain reaction (qPCR) output [6]. Further optimizations led to two related techniques to achieve genome-wide localization of RNA on chromatin: capture hybridization analysis of RNA targets (CHART) [7] and chromatin isolation by RNA purification (ChIRP) [8]. These in turn led to additional methods (e.g. RNA affinity purification, RAP [9]). We will hereafter refer to these collectively as hybridization capture methods [10].

Overview of hybridization capture methods

Hybridization capture methods to study chromatin-bound RNAs are similar to approaches for investigating RNAs under native conditions, but instead require cross-linked chromatin extracts. Techniques to study protein–chromatin interactions through the enrichment of DNA or proteins such as proteomics of intact chromatin [11], and the well-established ChIP protocol provide important precedent for these methods. In general, cells and tissues expressing the RNA of interest are cross-linked and solubilized by shearing. In contrast to ChIP, which relies on a high-affinity antibody raised against a specific protein or modification, hybridization capture methods enrich RNAs using rationally designed biotin tagged antisense oligonucleotides. Captured RNA complexes are then rinsed and eluted. The eluted material is then analyzed for the molecules of interest. The associated DNA and RNA are commonly analyzed with qPCR or high-throughput sequencing, and the recovered proteins are analyzed with western blots or mass spectrometry (Figure 1). Here, we focus on three hybridization capture methods (CHART, ChIRP and RAP) and explore their similarities and differences to illuminate the opportunities they provide (Table 1). We also present tips to help those performing hybridization capture experiments and highlight challenges that could lead to misinterpretation of the results. For detailed protocols for each approach, we refer the reader to recently published protocols [12–14].

Figure 1.

Overview of hybridization capture methods workflow. In hybridization capture methods, cells expressing an RNA of interest are cross-linked to stabilize interaction within the RNA complexes. The content of the nuclei is solubilized by shearing, and the RNA is hybridized to biotinylated antisense oligonucleotides. While CHART uses a handful of short capture oligonucleotides, ChIRP and RAP use a pool of oligonucleotides that tile to full length of the RNA. The RNA complexes are then enriched on magnetic streptavidin beads, washed under stringent conditions and eluted. Isolated DNA and RNA can be then analyzed using qPCR or high-throughput sequencing, while recovered proteins could be identified with mass spectrometry.

Table 1.

Comparison of hybridization capture methods

Parameter	CHART	ChIRP	RAP
Cross-linking conditions	1% formaldehyde for 10 min 3% formaldehyde for 30 min	1% glutaraldehyde for 10 min (later 3% formaldehyde for 30 min)	2 mM DSG for 45 min 3% formaldehyde for 10 min (later UV or 0.5 mg/ml AMT)
Chromatin shearing	3 kb median size	100–500 bp	100-300 bp
Capture oligonucleotide number	Handful of oligos (3–12)	Two sets of tiled nonoverlapping oligos	One set of tiled overlapping oligos
Capture oligonucleotide length	∼25 nt DNA	20 nt DNA	120 nt RNA (later 120 nt DNA or 90 nt DNA)
Capture oligonucleotide design	RNase H mapping	Full-length RNA tiling	Full-length RNA tiling
Hybridization conditions	1.9 M urea, 817 mM NaCl 20 °C	10% formamid, 500 mM NaCl 37 °C	3 M GuSCN, 150 mM LiCl, 45 °C (later 4 M urea, 500 mM LiCl, 67 °C)
Wash conditions	250 mM NaCl, detergents 20 °C	2× SSC, detergents 37 °C	3 M GuSCN, 150 mM LiCl, 45 °C (later 4 M urea, 500 mM LiCl, 67 °C)
Elution conditions	RNase H	RNase A+H	Proteinase K or benzonase

AMT = 4’-Aminomethyltrioxalen hydrochloride

Biological system

Using current technology, capture hybridization methods work best using millions to tens of millions of cells. This scale allows the enrichment from relatively precious samples such as embryonic stem cells [9, 15, 16] and primary tissues (e.g. limb buds) [17]. On the other hand, when it comes to identifying the protein interactors, hundreds to thousands of millions of cells might be needed [14, 18, 19]. Therefore, with notable exceptions [17], these experiments are performed using cell lines that are relatively easy to grow. In addition to selecting a cell line that reflects the appropriate biology (e.g. cells with an inactive X-chromosome to study Xist), it is important to choose a cell type that transcribes the RNA of interest at reasonably high levels. There is no simple rule to define how much RNA expression is required, in part because experiments to determine this have not been reported, but also because each RNA is likely to bind its targets with different stoichiometry and lead to different profiles (e.g. discrete peaks, which are easier to detect than broad signals). For cases in which the endogenous RNA is expressed at low levels, some approaches rely on overexpression of the target RNA [8]. Induced expression can also be useful when tracking changes over time (e.g. the spreading of Xist) [9]. In these cases, proper characterization is needed to confirm that the overexpressed RNA is functional and its properties have not been changed.

Preparation of cross-linked cells

Although some RNA complexes can be isolated under native conditions, extension of hybridization capture methods to chromatin and isolation of genomic targets have been performed with the help of cross-linking. The advantage of cross-linking is that the covalent attachment stabilizes weak interactions and permits the use of more stringent wash conditions to remove spuriously associated molecules. Ideally, cross-linking would provide a covalent record that offers a snapshot of where the RNA was in the cell at the moment of cross-linking. Despite cross-linking being neither instantaneous nor uniform, it can be leveraged to capture the state of the cell, as in ChIP, RNA immunoprecipitation (RIP) and crosslinking immunoprecipitation. Several cross-linking reagents have been used in hybridization capture methods, including formaldehyde, glutaraldehyde, disuccinimidyl glutarate (DSG), ultraviolet (UV) light, psoralen and in some cases combinations of more than one of these reagents [10]. Each cross-linking reagent has its own specificity and biases that influence the type of interactions that can be preserved, and therefore, which molecules will be enriched. Naturally, this substantially influences the success of an experiment, and also the interpretation of the results. Here, we provide a brief overview of common cross-linking conditions used in hybridization capture experiments. In addition to the cross-linker-specific considerations discussed below, it is worth noting that the presence of a biomolecule after enrichment does not necessarily imply that it was cross-linked to the RNA of interest. In many cases, there are noncovalently bound contaminants (such as precipitants) that can bind to streptavidin beads or RNA complexes through ionic or hydrophobic interactions.

Formaldehyde

Formaldehyde is a cell-permeable and fast-acting chemical that reacts with a wide range of nucleophilic functional groups found in DNA, RNA and proteins. The small size of formaldehyde constrains the distance between two functional groups to be within 2 Å to create a direct linkage. Formaldehyde is a specific cross-linker well suited to capturing interactions between macromolecules that are in close proximity [20]. Most formaldehyde cross-links are reversible under relatively mild heating in an appropriate buffer, which allows for retrieval of both proteins and nucleic acids after cross-link reversal. These properties made formaldehyde cross-linking the method of choice in many protocols, including ChIP and chromatin conformation capture methods. To enrich RNA using hybridization capture methods, extensive treatment with 3% formaldehyde was originally introduced in CHART [7], and was later adopted by domain ChIRP (dChIRP) [21] and RAP protocols [9].

Despite the wide reactivity of formaldehyde with nucleophilic groups in biomolecules, not all amino acid side chains produce cross-links. Under the cross-linking conditions used in ChIP and hybridization capture experiments, reaction products have been detected involving lysine, tryptophan and cysteine side chains as well as the peptide N-terminus [22]. Analysis of linkage formation with nucleotides revealed that the highest yields are obtained between guanosine and lysine, which is about 10 times higher than the other tested combinations [23]. Therefore, even though formaldehyde is a versatile cross-linker, one must use caution when assuming that distinct RNA–protein complexes will display similar levels of formaldehyde cross-linking.

Glutaraldehyde

Glutaraldehyde is a linear five-carbon dialdehyde that reacts with similar nucleophilic groups as formaldehyde, and is widely used in stabilizing biomolecular interactions for electron microscopy [24]. It is cell-permeable and can cross-link proteins, nucleic acids and lipids. Owing to the spacer between the two aldehyde groups, glutaraldehyde can create cross-links at longer distances and therefore provides additional opportunities to stabilize the RNA-associated complexes through nucleophilic groups that would otherwise be out of range for other short-distance cross-linkers such as formaldehyde. One key disadvantage is that glutaraldehyde cross-links are largely irreversible at pH 7–9, which can negatively influence recovery of both proteins and nucleic acids [25]. Accordingly, although glutaraldehyde was used in the original ChIRP protocol, it was switched to formaldehyde in dChIRP and later iterations of the protocol because of better signal-to-noise properties of formaldehyde [15, 21]. Despite this limitation, a 2% glutaraldehyde cross-linking step was used in a recent modification of CHART and ChIRP protocols [26].

DSG

DSG is a cell-permeable cross-linker containing two N-hydroxysuccinimide ester groups that react with primary amines found in proteins. DSG is approximately the same length as glutaraldehyde and therefore can cross-link amines between biomolecules that are more distant than formaldehyde. However, the amino groups in RNA or DNA are largely unreactive with DSG, which is why in a typical RAP experiment DSG is accompanied by formaldehyde to strengthen the interactions with nucleic acids [9, 27]. DSG cross-links are amide bonds that are effectively irreversible, which complicates analysis of RNA-associated proteins.

UV light

UV light (wavelength: 260 nm) can cross-link nucleic acids to other biomolecules. Because proteins do not substantially absorb light at this wavelength, UV treatment does not cross-link proteins to other proteins. Moreover, UV-induced cross-links form only at short distances, making this cross-linking method specific and ideal for discovery of direct RNA–protein interactions. UV cross-linking has been used in conjunction with hybridization capture enrichment to identify Xist protein partners [18, 28]. Nevertheless, cross-linking with UV light has several disadvantages, including variable efficiency and high bias toward uracil, which yields about 30 times more cross-linking than other bases [29, 30]. Furthermore, not all amino acid residues can cross-link with the same efficiency. For example, in the case of uracil, only 11 amino acids were observed to cross-link with RNA (with the most reactive being phenylalanine, tyrosine and cysteine) [31]. The efficiency of cross-linking may depend on the specific stereochemical orientations of the protein and RNA as well, thus further complicating the cross-linking bias. Therefore, not all RNA-binding proteins cross-link with equal efficiency to RNA, resulting in a high false-negative rate and complicating quantitative interpretations. Because UV light damages nucleic acids, UV cross-linking also complicated downstream analysis of DNA and RNA.

Psoralen

Psoralen and its derivatives are functional photoreagents that intercalate into double-stranded regions of nucleic acids and form an interstrand cross-link on exposure to lower energy (365 nm) UV light. The advantage of using light at 365 nm is that it does not create pyrimidine dimers that would otherwise interfere with reverse transcription and sequencing. The cross-link is also reversible, making psoralen an ideal tool for probing direct RNA-RNA interactions. For example, psoralen was successfully used to analyze interactions between U1 small nuclear RNA (snRNA) and pre-mRNAs [27]. However, a significant limitation of its utility is the disruptive nature of treating cells with an intercalating reagent, as well as a strong reactivity bias toward uracil (thymine), thus making it primarily useful for sequences UA/UA dinucleotides [32].

Extracting chromatin

After cross-linking the cells, the complexes need to be solubilized to make a cross-linked chromatin extract. When cells are cross-linked with a chemical cross-linker such as formaldehyde, solubilization is typically achieved by fragmentation of the chromatin. This is generally accomplished with high-energy sonication, similar to ChIP, except that the higher levels of cross-linking used in hybridization capture experiments tend to require more extensive shearing. In hybridization capture methods, the median fragment sizes have ranged from 150 bp to 3 kb (Table 1). In theory, shorter DNA fragments are preferable, as they should produce higher resolution high-throughput sequencing information. However, direct comparison of published data from different Xist or roX2 long noncoding RNA (lncRNA) chromatin interaction studies have revealed little difference in overall signal profile [33].

In the process of DNA shearing, RNA strands are also inevitably fragmented. In some cases, this can be a useful property. For example, dChIRP used this fragmentation to demonstrate long-distance interactions between distant regions of the roX1 lncRNA complex [21]. However, RNA fragmentation can also be a limitation, as it brings a potential risk of losing RNA fragments with important proteins or chromatin interactions. This RNA fragmentation can be minimized by implementing enzymatic digestion of DNA (and not RNA) with DNAse I, thereby minimizing the need for mechanical shearing (used in RAP [9]). Given the extensive handling of the cell extract during hybridization capture experiments, fragmentation of RNAs, especially long ones, cannot be completely avoided.

Capture oligonucleotide design and hybridization

Central to hybridization capture methods are the biotinylated antisense capture oligonucleotides used for enrichment, which are made synthetically from DNA (CHART, ChIRP) or from transcribing DNA into RNA (RAP). There are two general strategies to design such oligonucleotides: design small numbers of capture oligonucleotides to specific sites on the lncRNA, or tile the entire RNA with many capture oligonucleotides. To develop a small number of targeted capture oligonucleotides, CHART uses RNase H mapping to identify oligonucleotides that can hybridize to RNA in cross-linked chromatin extracts. These can be directed across the entire RNA (for small RNAs) or to targeted regions of larger RNAs. Using a smaller number of oligonucleotides allows for more precise design of the melting temperature and off-target binding of each oligonucleotide. On the other hand, RNase H mapping can be time-consuming. Using only a few oligonucleotides also increases the risk that the capture sites on the RNA will fragment away from the RNA–chromatin or RNA–protein cross-links, resulting in poor yield of associated targets, especially if the RNA is highly sheared (see above). ChIRP and RAP approach this challenge by using a large pool of overlapping or nonoverlapping capture oligonucleotides that tile the complete length of the RNA (Figure 1, Table 1). This reduces concerns about hybridization accessibility of any given site, and provides more even coverage of RNA fragments, but has its own disadvantages. Synthesizing a large number of biotinylated oligonucleotides can be expensive. Further, as the amount of off-target signal retrieved is largely determined by the least specific oligonucleotide, using more oligonucleotides increases the likelihood of problems with enrichment (see below).

The hybridization reaction is the heart of the hybridization capture methods. Its efficiency and specificity will substantially influence the outcome of the experiment. Therefore, the sheared extract must be adjusted to conditions that allow specific annealing of capture oligonucleotides to the desired target RNA with a reasonable yield. This is done by introducing detergents, salts and chemical denaturants into the cell extract. High concentrations of salts enhance base pair stability by increasing the melting temperature of nucleic acids, while disrupting nonspecific ionic interactions of macromolecules with charged surfaces (e.g. beads). On the other hand, denaturants such as urea, guanidinium thiocyanate and formamide act to decrease the melting temperature. Therefore, both additives must be balanced to provide appropriate conditions for RNA hybridization, while keeping the chromatin soluble, as precipitation of components of the extract onto beads can be a major source of background. Further, if the hybridization conditions are not properly tuned, unwanted off-target RNA and DNA sequences could be recovered through either direct or indirect hybridization [10].

Direct off-target

Direct off-target hybridization occurs when capture oligonucleotides spuriously hybridize directly to RNAs or DNAs that contain a similar sequence to the target RNA (Figure 2A and B) [10]. The first step to avoid such background signal, and to achieve high specificity, is to carefully design the sequence of capture oligonucleotides. Specifically, sequences in the target RNA that are too similar to the other parts of the genome and transcriptome should be avoided. Such sequences can be identified through a simple BLAST search [8], and more sophisticated pipelines have also been reported [9]. Direct off-target artifacts can be further reduced by using a small number of antisense oligonucleotides with similar melting temperatures. Melting temperature is a property of the capture oligonucleotide sequence and also the buffer composition (such as ionic strengths and concentration of denaturants). The maximum hybridization specificity of nucleic acids is reached near their melting temperature, where the free energy is maximally different between the desired duplex and off-target complexes with lower complementarity. Even similar sequences may be distinguished from the real target based on their binding affinities to a complement (Figure 3A). However, it can be difficult to design many sequences with highly tuned melting temperatures, especially with the nonstandard buffers used in hybridization capture experiments. When oligonucleotides bind with low specificity, this results in increased background (Figure 3B). Therefore, many approaches have been successful using a small number of carefully designed oligonucleotides [21, 26].

Figure 2.

Types of hybridization artifacts. In hybridization capture experiment, several hybridization-induced artifacts can occur that increase the amount of background signal and complicate downstream analysis. In case the hybridization conditions are not well tuned or the sequence is poorly designed, the capture oligonucleotide (arrow) can directly hybridize to (A) off-target RNA (string) or (B) denatured regions of DNA (ribbons). Even when the capture oligonucleotide hybridizes to RNA target (hairpin), the single-stranded parts of the RNA can form base pairing with (C) indirect off-target RNA or (D) DNA.

Figure 3.

Hybridization specificity. (A) When the sequence of an off-target nucleic acid differs significantly (far off-target) from the real RNA target, the window of usable hybridization conditions and temperatures that can efficiently discriminate between the two is relatively wide (grey area). For more similar sequences (near off-target), the range of usable temperatures converges around the melting temperature (Tm) of target-oligonucleotide hybrid (dark grey). (B) In case of multiplexed hybridization, balancing melting temperatures of many oligonucleotides can be challenging. At given temperature, it is likely that only a subset of oligonucleotides will be within conditions that are optimal for target discrimination.

Direct off-target signal is relatively easy to control by using different sets of capture oligonucleotides. For example, in roX2 ChIRP, two different enrichment experiments were conducted using two independent oligonucleotide pools that tile the RNA. Each individual pool retrieved high amounts of off-target signal, which in principle complicates data interpretation (Figure 4A–C). The challenge of the direct off-target binding in ChIRP was largely mitigated by data post-processing by filtering for signal that is only present in both oligonucleotide pools (Figure 4D) [8, 21]. This postprocessing led to accurate identification of roX2 localization and agreed well with other biochemical expectations (e.g. enrichment on the X-chromosome and colocalization with a roX2 binding protein, MSL3). Another useful method to identify off-target signal, particularly for DNA, is to include controls in which hybridization is conducted on samples that have been treated with RNase. Signals that remain can generally be attributed to unwanted direct hybridization of capture oligonucleotides to DNA. The best control for direct off-targets of hybridization would be to perform the hybridization capture in cells that do not contain the target RNA, as any signal is necessarily spurious. Although this type of experiment can help, direct binding to capture oligonucleotides is not the only hybridization-induced artifact.

Figure 4.

Comparison of roX2 CHART and ChIRP signal distribution. Comparing the coverage of sequencing reads over 1 kb bins on Drosophila X chromosome shows that (A–C) ChIRP performed with ‘odd’ or ‘even’ oligonucleotide pools recover a distinct sets of noise patterns (light gray) that are of similar intensity as true roX2 signal (dark gray). (D) After removal of signal that was not found reproducibly in both ‘odd’ and ‘even’ data sets, the ChIRP data reache quality that is similar to CHART data set.

Indirect off-target

Indirect off-target hybridization occurs when capture oligonucleotides hybridize to their intended target RNA that further hybridizes to denatured DNA or another RNA species (Figure 2C and D). For example, if the target RNA has long strings of C-nucleotides, it is likely to artifactually enrich DNA and RNA with complementary strings of G-nucleotides. These artifacts are extremely difficult to distinguish from true signal using the controls discussed above. For example, removing the target RNA (e.g. knockout cells) would lead to loss of both the specific signal and also all the indirect off-targets. One approach to overcome this problem is to attempt to remove off-targets using stringent wash conditions (e.g. high temperature and denaturant concentration) that would disrupt their hybridization contacts, which are presumably weaker than the exact match of the capture oligonucleotide to the target RNA. The increase in stringency, however, must be matched with an increase in capture oligonucleotide affinity to preserve interaction with the target RNA. RAP originally addressed this challenge by using long (∼120 nt) RNA oligonucleotides, which have significantly higher affinity compared with the shorter DNA capture oligonucleotides used in CHART and ChIRP (Table 1). In a more recent protocol [13], RAP has converged with the other methods by using DNA rather than DNA oligonucleotides, but still uses longer (90 nt) capture oligonucleotides than CHART or ChIRP. The specificity and affinity of nucleic acid hybridization are strongly anti-correlated, which means that the increase in affinity is generally associated with a loss in specificity [34]. This occurs because the difference of free energy caused by a single-nucleotide mismatch is constant and relatively small; therefore, the thermodynamic benefit from many correctly paired bases can override the penalty from a few mismatches. Therefore, although longer oligonucleotides bring higher affinities and allow for more stringent washes, they also increase the risk of direct off-target binding.

While the specificity of hybridization capture (and ensuing levels of both direct and indirect artifacts) is initially determined by the hybridization reaction, once the RNA-capture oligonucleotide hybrids are enriched on streptavidin beads, specificity is further improved by stringent washes. These washes are used to remove nonspecifically bound or weakly hybridized background. Then, the desired complexes must be eluted from the solid support.

Elution

To release RNA complexes from beads, hybridization capture methods use several established elution strategies (Table 1). For example, proteinase K can be used to isolate nucleic acids while destroying all the proteins including the streptavidin on the beads. The disadvantage is that RNA or DNA unspecifically associated with streptavidin will be eluted as well. Alternatively, nucleases such as benzonase or RNase A can be used to degrade nucleic acids and release protein material. However, complete nucleic acid degradation can also release proteins that were associated with the beads through nucleic acids in a nonspecific manner. When using DNA oligonucleotides, specific elution can be accomplished using RNase H, which will only digest the portion of the RNA that was hybridized to a complementary DNA sequence. This approach was shown to dramatically decrease background signal in CHART sequencing data [7]. RNase H elution can therefore be used in the isolation of proteins, DNA and also RNA. Including other enzymes such as RNase A can increase yields but is less specific and compromises later analyses.

Although it is common to analyze the eluted material for enriched RNA and DNA sequences with qPCR, a more global view of the RNA targets is achieved using high-throughput sequencing. In addition to providing a more complete picture of the signal, sequencing also provides insight into the extent of enrichment. As hybridization capture methods are analogous to ChIP, the preparation of sequencing libraries and subsequent data analysis take into account similar considerations and follow the same established protocols [35]. Once the sequencing data are acquired, the reads are aligned to the reference genome and normalized to the input sample consisting of randomly sheared genomic DNA. The most significantly enriched genomic regions are then considered as potential RNA associated sites. Owing to tendency of hybridization capture methods to recover off-target sequences (see above), it is recommended to compare these regions with enrichment experiments using capture oligonucleotides without expected targets. Similarly, because artifacts are most likely when analyzing sequences with complementarity to the capture oligonucleotides or the target RNA, it is good practice to search for regions that have high sequence similarity to the targeted RNA to identify possible false-positive signals.

In summary, hybridization capture methods each have their own technological advantages and disadvantages, which must be carefully weighed when designing an experiment. Moreover, technological limitations and biases resulting from these approaches must be considered, both when interpreting one’s own results and when comparing across different studies. Erroneous conclusions arising from hybridization-induced artifacts can also be limited by well-designed positive and negative controls, and by comparing the resulting data with known biological functions (e.g. chromatin marks, FISH localization).

New insights into lncRNA functions

Since their introduction, hybridization capture methods have been widely adopted by the RNA community and applied to several lncRNAs. These studies have contributed to our understanding of how lncRNAs adopt diverse mechanisms to perform their biological functions. For example, the binding sites of the fly dosage compensation complex on the male X-chromosome were originally studied using ChIP (to measure localization of the protein components, for example, see [36, 37]). Hybridization capture methods including ChIRP and CHART have made it possible to perform similar studies on the RNA components of this complex [7, 8, 21]. Hybridization capture methods have been used to investigate lncRNAs across a range of systems, including the genomic loci of viral EBER2 lncRNA and the localization of FMR1 mRNA to its own promoter [38, 39]. Additional studies have revealed that RNAs can display diverse patterns of localization, as either well-defined peaks as in case of roX1 and roX2 lncRNAs or broad distributions such as spreading of Xist on the X-chromosome [7–9, 16, 21, 40]. These lncRNAs localize to distal sites on the same chromosome, but other lncRNAs were reported to reach sites on different chromosomes such as HOTAIR or 116HG lncRNA [8, 41]. In addition, several recent reports describe lncRNAs that interact with transcription factors and enable their binding to certain promoters [42–45]. As these results are largely based on qPCR analysis, an important next step is the analysis of global localization patterns, which will also the quality of the enrichment in these studies and the broader levels of colocalization between lncRNAs and specific transcription factors. How can lncRNAs reach theirs sites of interaction is still not completely understood, but a correlation of hybridization capture and chromosome conformation capture data suggests that lncRNAs, like other elements of chromatin, can interact at a distance through higher-level folding of the chromatin [9, 21]. Despite these many examples, the extent of lncRNAs acting on chromatin in trans remains an open question.

In this review, we have focused on the use of hybridization capture techniques to probe the genomic localization of RNAs. In addition, these methods have proven successful in enriching proteins that interact with lncRNAs. For example, CHART was used to study the proteins associated with MALAT1 and NEAT1 lncRNAs [19], and several different hybridization capture methods were used to identify dozens of novel Xist binding proteins [15, 18, 28], albeit with limited overlap between studies. In addition, hybridization capture methods can be used to study RNA-RNA interactions. For example, dChIRP helped to uncover intramolecular interactions between core regions of the roX1 lncRNA [21], while providing evidence that roX1 domains responsible for interactions with chromatin appear to be devoid of intramolecular contacts. On other hand, RAP used psoralen cross-linking to detect direct intermolecular interactions of snRNAs with unspliced pre-mRNAs, as well as other RNA-RNA interactions [27]. These examples anticipate the more widespread use of hybridization capture approaches for detecting RNA–protein and RNA-RNA interactions.

Conclusions and future directions

Hybridization capture methods have contributed substantially to our understanding of lncRNAs and how they interact with chromatin. Although this family of methods have demonstrated notable successes, we highlighted the challenges of obtaining high-quality results. Such challenges make hybridization capture experiments difficult to perform and interpret relative to better-established methods such as ChIP. In future work, these challenges are likely to be overcome through the development of complementary approaches to facilitate interpretation of hybridization capture results, and through the further improvement of these protocols. To avoid artifacts in hybridization capture experiments, in principle, it should be possible to create a sufficiently stringent approach that would dramatically decrease artifactual signal. Such an approach would likely require increasing the hybridization specificity (to decrease off-target capture) and simultaneously increasing the stability of the desired captured RNA hybrid (enabling the complexes to withstand even more stringent rinses to denature and remove residual off-target signals). The true targets of the RNA would survive these stringent conditions by virtue of the covalent nature of the cross-links formed when the lncRNA was in its biological context in the cell. We anticipate that the current hybridization capture protocols will be further improved and become increasingly powerful means to reveal the biomolecular interactions that underlie lncRNA function.

Key Points

• Uncovering lncRNAs’ interactions with chromatin, proteins and other RNAs is crucial for understanding their biological role.

• Three main hybridization capture methods (CHART, ChIRP and RAP) can enrich specific RNAs through hybridization to rationally designed antisense oligonucleotides.

• Cross-linking reagents each have distinct sets of strengths and biases.

• Suboptimal hybridization conditions and poor oligonucleotide design can lead to increased direct and indirect off-target hybridization, complicating interpretation of results.

• Further improvements of hybridization capture methods that would simultaneously increase both specificity and affinity are needed.

Funding

This work was supported by a fellowship from the Charles H. Revson Foundation (M.M.), NIH New Innovator Award DP2 HD083992-01 (M.D.S.), and a Searle scholarship (M.D.S.).

Martin Machyna finished his PhD at MPI-CBG and is currently a Charles H Revson fellow and a postdoctoral researcher at Yale University where he is investigating RNA interactions with chromatin.

Matthew D. Simon is an associate professor of Molecular Biophysics and Biochemistry and a member of the Chemical Biology Institute at Yale, developing chemical tools to study RNA and chromatin biology.