Discrete RNA–DNA hybrid cleavage by the EXD2 exonuclease pinpoints two rate‐limiting steps

Abstract

EXD2 is a recently identified exonuclease that cleaves RNA and DNA in double‐stranded (ds) forms. It thus serves as a model system for investigating the similarities and discrepancies between exoribonuclease and exodeoxyribonuclease activities and for understanding the nucleic acid (NA) unwinding‐degradation coordination of an exonuclease. Here, using a single‐molecule fluorescence resonance energy transfer (smFRET) approach, we show that despite stable binding to both substrates, EXD2 barely cleaves dsDNA and yet displays both exoribonuclease and exodeoxyribonuclease activities toward RNA–DNA hybrids with a cleavage preference for RNA. Unexpectedly, EXD2‐mediated hybrid cleavage proceeds in a discrete stepwise pattern, wherein a sudden 4‐bp duplex unwinding increment and the subsequent dwell constitute a complete hydrolysis cycle. The relatively weak exodeoxyribonuclease activity of EXD2 partially originates from frequent hybrid rewinding. Importantly, kinetic analysis and comparison of the dwell times under varied conditions reveal two rate‐limiting steps of hybrid unwinding and nucleotide excision. Overall, our findings help better understand the cellular functions of EXD2, and the cyclic coupling between duplex unwinding and exonucleolytic degradation may be generalizable to other exonucleases.

EXD2 preferentially degrades RNA over DNA within hybrid duplexes, unwinding four basepairs before excising single nucleotides at a time.

Introduction

Exonucleases are a class of structurally and biochemically characterized enzymes that catalyze the excision of nucleoside monophosphates from 3′‐ or 5′‐NA termini (Shevelev & Hubscher, 2002; Mimitou & Symington, 2009; Yang, 2011). Exonuclease activities are integral to many nucleic acid metabolic processes, such as DNA repair and RNA splicing (Shevelev & Hubscher, 2002; Mimitou & Symington, 2009; Yang, 2011; Cannavo et al, 2013; Tsutakawa et al, 2014; Keijzers et al, 2016; Ranjha et al, 2018; Stroik et al, 2020; Lee et al, 2022a). EXD2 is a newly identified 3′‐to‐5′ exonuclease that has proved to facilitate DNA end resection during homologous recombination (HR) and counteract fork regression in the process of DNA replication (Broderick et al, 2016; Nieminuszczy et al, 2019). However, in contrast to DNA‐related nucleic metabolism, different groups recently provided several lines of evidence that EXD2 is primarily associated with the mitochondrial membrane or matrix and interacts with RNA (Hensen et al, 2018; Silva et al, 2018; Park et al, 2019). EXD2 was thus proposed to be a mitochondrial exoribonuclease that possibly functions in respiration, ATP production, and mitochondrial translation (Silva et al, 2018). Meanwhile, in vitro biochemical assays demonstrated both exoribonuclease and exodeoxyribonuclease activities of EXD2 toward single‐stranded (ss) and double‐stranded (ds) nucleic acids (NAs) (Broderick et al, 2016; Silva et al, 2018). Interestingly, as revealed by a recent structural study, the exonuclease domain of EXD2 (EXD2‐exo) complexed with Mn²⁺/Mg²⁺ forms a chimeric dimer that appears to be strict ionic requirements for different NA hydrolysis (Park et al, 2019). To be precise, the EXD2‐exo dimer preferentially utilizes Mn²⁺ as a cofactor to degrade both RNA and DNA, whereas the presence of Mg²⁺ is only permissive for RNA degradation (Silva et al, 2018; Park et al, 2019). Moreover, EXD2 removes deoxyribonucleotides one at a time from the 3′ end of ssDNA, though its RNA cleavage product remains unknown (Broderick et al, 2016). Identifying the similarities and differences between the exoribonuclease and exodeoxyribonuclease activities of EXD2 would help clarify its cellular functions and further our understanding of its substrate discrimination.

An integral enzymatic cycle of an exonuclease degrading ssNA is typically comprised of nucleotide excision, product release, and directional translocation along the substrate (Yang, 2011; Zhang et al, 2011). Compared with ssNA degradation, dsNA cleavage by an exonuclease sometimes involves an additional step in its hydrolysis cycle, namely, base‐pair unwinding, which is often closely coupled with ssNA translocation and typically occurs before nucleotide excision (Zhang et al, 2011). During unwinding, helicases mainly use NTP hydrolysis as an energy source. In contrast, exonucleases are usually powered by the chemical energy released from the hydrolysis of the NA chain (Lee et al, 2011; Myler et al, 2016; Sun & Wang, 2016; Qin et al, 2020; Spinks et al, 2021). Therefore, it is not surprising that several exonucleases, such as λ exonuclease and Rrp44, are sequence‐dependent motor proteins and have a major rate‐limiting step of duplex unwinding during dsNA cleavage (Perkins et al, 2003; Lee et al, 2012). As a dsNA exonuclease, whether EXD2 involves duplex unwinding in the process of dsNA cleavage and, if yes, how it periodically coordinates dsNA unwinding with nucleolytic degradation are still unclear (Broderick et al, 2016; Silva et al, 2018; Park et al, 2019).

In this work, we examined and compared the detailed dynamics of the exonuclease activity of EXD2 toward RNA and DNA in double‐stranded forms. We found that EXD2 exhibited extremely poor exonuclease activity on dsDNA, although it demonstrated stable binding to it. In comparison, EXD2 cleaves both RNA and DNA within a hybrid duplex. Taking advantage of the high spatial–temporal resolution of the smFRET technique, we also revealed that the hybrid unwinding and ssNA degradation of EXD2 coordinate discretely. In each hydrolysis cycle, four hybrid base pairs were unwound in a burst, followed by a dwell time for nucleotide excision and preparation for the next round of 4‐bp duplex unwinding. The two reactions during the dwell consecutively occurred at rates on the same order of magnitude. Consequently, both reactions are rate‐limiting steps restraining the overall turnover frequency of EXD2. Our findings provide new insight into the mechanism underlying the unwinding‐degradation coordination of an exonuclease.

Results

EXD2 catalyzes a continuous excision of tens of ribonucleotides within the hybrids

To improve solubility, we constructed and purified a truncated version of human EXD2 (residues 61–621, hereafter referred to as EXD2) (Silva et al, 2018). After confirming that the purified EXD2 can digest ssNAs and stably bind to an RNA–DNA hybrid substrate (Fig EV1A and Appendix Fig S1), we first examined the exoribonuclease activity of EXD2 toward the hybrid substrate. To this aim, we employed a previously developed smFRET approach (Fig 1A; Roy et al, 2008; Lee et al, 2011, 2022a; Hwang et al, 2018). In this assay, the DNA in the hybrid was labeled with a donor (Cy3) at the 5′ end and an acceptor (Cy5) at 20 nt from the 5′ end. Meanwhile, five deoxyribonucleotides at the 3′ end of the same strand were modified by the phosphorothioate (PS) to prevent degradation (Fig EV1B). Additionally, the 5′ end of the paired RNA was labeled with biotin, and the hybrid substrate can thus be immobilized on a PEG‐streptavidin‐coated quartz surface. The 3′–5′ degradation of the biotin‐labeled RNA strand would generate a 5′‐overhang of the fluorescently labeled ssDNA strand. The intrinsic flexibility of the remaining ssDNA would cause a decrease in the distance between the two fluorophores (Murphy et al, 2004; Liu et al, 2008) and thus an increase in FRET efficiency (E) (Fig 1A). Hereafter, this RNA–DNA hybrid substrate was referred to as RDH, with the first letter “R” representing the degrading strand of RNA.

Figure EV1. Gel‐based bulk assays of NA cleavage by EXD2.

• A
  A representative gel showing EXD2 (15 nM) cleaving the Cy5‐labeled 53‐nt ssRNA and ssDNA (10 nM) substrates under various MnCl₂ concentrations.
• B
  A representative gel showing EXD2 (30 nM) cleaving the RDH substrate with the phosphorothioate (PS) modification on the five nucleotides at the 3′ end. These PS modifications prevent EXD2 from ssDNA cleavage.
• C
  A representative gel showing EXD2 (30 nM) cleaving the Cy5‐labeled RDH (10 nM) substrate with the presence of 3×, 6×, and 9× unlabeled RDH traps. After a 5‐min incubation of the labeled RDH substrate with EXD2, unlabeled traps were introduced with Mn²⁺ to initiate the cleavage. They barely interfere with the labeled RDH substrate cleavage, suggesting a single turnover event.
• D, E
  A representative gel showing EXD2 (30 nM) cleaving the fluorescently labeled DRH (D) and dsDNA (E) (10 nM) substrates. The blue arrows indicate the degradation direction. The cleavage products were resolved by an 18% denaturing polyacrylamide gel electrophoresis and visualized by phosphorimaging.

Source data are available online for this figure.

Figure 1. Exoribonuclease activity of EXD2 toward the RNA–DNA hybrid substrate.

• A
  A schematic showing the design of the smFRET assay. The 3′–5′ degradation of the biotin‐labeled RNA by EXD2 gives rise to the gradual generation of the Cy3‐ and Cy5‐labeled DNA, causing an increase in FRET efficiency. PS represents phosphorothioate.
• B–G
  (B, D, and F) E histograms of the RDH substrate in the absence of EXD2 (gray) and after 0‐ (black), 100‐ (purple), and 200‐s (blue) incubation with 200 (B), 100 (D), and 5 (F) nM EXD2. (C, E, and G) Representative smFRET trajectories showing EXD2 at 200 (C), 100 (E), and 5 (G) nM cleaving RDH in real time. The fluorescence intensities of the donor and acceptor are colored green and red, respectively. The corresponding FRET efficiencies are shown below in black. The duration between two Cambridge blue lines represents the time for the FRET efficiency to increase from ~0.2 to ~0.9.
• H
  The FRET trajectories' percentages show the FRET values ending above 0.9 and between 0.2 and 0.9. The word “Surface” represents the experiment with EXD2 anchored on the glass surface.
• I
  Statistics of the time for the FRET efficiency to increase from ~0.2 to ~0.9 under indicated experimental conditions. In violin plots, light‐blue dots represent mean, and black bars show the interquartile range (IQR) (thick bars) and 1.5 times IQR (thin bars) (n = 122, 91, 78, and 101 from left to right).

According to our bulk ssNA degradation assays and previous studies (Fig EV1A; Silva et al, 2018; Park et al, 2019), we chose to supply EXD2 with 1 mM MnCl₂ in the following FRET experiments to achieve optimal performance unless otherwise stated. A control experiment showed that in the absence of Mn²⁺ (0 s), EXD2 at 200 nM only caused a minor increase in E from ~0.13 to ~0.22, which is possibly due to its binding to the substrate (Fig 1B). In contrast, upon the addition of Mn²⁺, a 100‐s incubation gave rise to an apparent shift in the FRET histogram, in which a majority of the E efficiency increased to ~0.96, and a small portion was centered at ~0.45 (Fig 1B). An additional 100‐s incubation resulted in a similar FRET histogram, indicating that the reaction was adequate (Fig 1B). In accordance with the FRET histograms, real‐time trajectories also exhibited an increase in FRET efficiency within tens of seconds (Fig 1C). According to the literature (Broderick et al, 2016; Park et al, 2019), we constructed and purified two EXD2 mutants: a nuclease‐deficient mutant (EXD2‐D108A‐E110A) and a dimerization‐deficient mutant (EXD2‐N198P). Experiments with these two mutants showed minor changes in the FRET histograms, supporting that the detected increase in FRET efficiency indeed arises from EXD2 processing the hybrids and that the active form of EXD2 is possibly a dimer (Appendix Figs S2 and S3).

The increase in E from 0.22 to 0.96 corresponds to tens of ribonucleotides processed by EXD2. To answer whether it is a result of a single EXD2 effector, we repeated the smFRET assay under a relatively low EXD2 concentration of 100 nM (Fig 1D and E) or 5 nM to minimize multi‐protein events (Fig 1F and G). After a 100‐s or 200‐s incubation with Mn²⁺, the FRET histograms displayed right shifts, similar to those with 200 nM EXD2 (Fig 1D and F). Statistical analysis of the real‐time trajectories revealed that over 75% of them ended up with E states over 0.9 within 100 s even under 5 nM EXD2 (deemed as a thorough reaction) and the rest stalled at E states between 0.2 and 0.9 (Fig 1H). Notably, the reaction times (defined as the duration for E to increase from 0.2 to 0.9) under different EXD2 concentrations are comparable, indicating a single protein event (Fig 1I). To ensure a single turnover condition, we further performed the smFRET assay with EXD2 anchored on the glass surface and the fluorescently labeled hybrid substrates bound to the anchored proteins. In this assay, free proteins in the chamber were removed, followed by adding Mn²⁺ to initiate the cleavage. We found that up to 66% of real‐time FRET trajectories can reach up to 0.9 within a comparable reaction time (Figs 1H and I, and EV2). Note that in this experimental configuration, both substrate dissociation and Cy3 photobleaching could cause the sudden disappearance of the fluorescence signal, which was considered incomplete cleavage. The incapability to differentiate the two scenarios partially accounts for the slightly increased fraction of incomplete‐reaction trajectories detected in the protein‐anchored assay (34%). Furthermore, we demonstrated by a gel assay that increasing unlabeled RDH traps would not cause apparent perturbation on the degradation of a labeled RDH substrate by EXD2 (Fig EV1C). Based on these findings, we concluded that most of the detected increases in FRET originate from a single turnover event.

Figure EV2. The stepwise RDH cleavage by surface‐anchored EXD2.

1. Schematic of the smFRET experiment to monitor RDH cleavage by surface‐anchored EXD2.
2. Representative smFRET trajectories showing surface‐anchored EXD2 cleaving RDH in real time. The fluorescence intensities of the donor and acceptor are colored green and red, respectively. The corresponding FRET efficiencies are shown below in black. The duration between two Cambridge blue lines highlights the time for the FRET efficiency to increase from ~0.2 to ~0.9.
3. A representative smFRET trajectory showing a four‐step FRET increase. The solid red line is the fitting result of an automated step‐finding algorithm.
4. The histogram showing the distribution of visited FRET states during EXD2 cleaving RDH.
5. The corresponding TDP shows stepwise increases in FRET.

Our data suggest that EXD2, possibly in a dimeric form, can continuously process tens of ribonucleotides in the hybrids, although some may stall at specific positions.

EXD2 unwinds RNA–DNA hybrids before excising ribonucleotides one at a time

Regarding dsNA cleavage, exonucleases often involve dsNA unwinding in their catalytic cycles (Zhang et al, 2009, 2011; Lee et al, 2012). We next asked whether EXD2‐mediated hybrid cleavage also necessitates dsNA unwinding. Note that the increase in E as detected in the smFRET assay reflects the generation of ssDNA, which could result from either hybrid unwinding or RNA excision by EXD2. However, the two activities differ in reversibility: the duplex unwinding could be conditionally reversed by rewinding, whereas the nucleotide excision could not. Therefore, if EXD2 requires dsNA unwinding for hybrid cleavage, this step can be reversed under some conditions, such as perturbed nucleotide excision. With this notion in mind, we constructed another RDH substrate with four ribonucleotides (the 5^th–8^th nucleotides from the 3′ end) modified with PS (P‐RDH) that were supposed to prevent nucleotide excision of EXD2 but not dsNA unwinding if there is any (Fig 2A). In the smFRET assay with this substrate, FRET histograms and real‐time trajectories expectedly reported a stall of E states at ~0.34 due to the incapability of EXD2 to degrade PS‐modified ribonucleotides (Appendix Fig S4). However, notably, sudden bursts during the stalled E state were also noticed in 56% of the examined trajectories, indicating that the hybrids were transiently unwound (Fig 2A). The infrequent unwinding events could be attributable to EXD2 using thermal fluctuations instead of the hydrolysis of the NA chain to separate the hybrid. Similarly, smFRET experiments with the deficient EXD2 mutant (EXD2‐D108A‐E110A) on the RDH substrate also showed sudden bursts in 48% of 189 trajectories (Fig 2B). These bursts were based on a stable protein binding E state of ~0.22 (Fig 2B). These findings point to a model of EXD2 unwinding a few base pairs of the RNA–DNA hybrids during dsNA cleavage. In support of this model, the smFRET assay with EXD2 on an AU‐rich RDH substrate (the 3^rd–20^th ribonucleotides from the 3′ end of the RNA are A and U only, AU‐RDH) with reduced stability demonstrated frequent unwinding‐rewinding transitions in 85% of the examined FRET trajectories before reaching the final E state of ~0.9 (Fig 2C). The reversibility in E detected in these three sets of experiments favors the duplex unwinding prior to nucleotide excision in the process of EXD2‐mediated hybrid cleavage.

Figure 2. EXD2 unwinds the hybrid before excising ribonucleotides one at a time.

• A–C
  The cartoons on the top represent experiments using P‐RDH (A), EXD2‐D108A‐E110A (B), and AU‐RDH (C), respectively. Representative smFRET trajectories showing EXD2 processing P‐RDH (A), EXD2‐D108A‐E110A processing RDH (B), and EXD2 processing AU‐RDH (C) in real time. The red arrows highlight the unwinding signal. The fluorescence intensities of the donor and acceptor are colored green and red, respectively. The corresponding FRET efficiencies are shown below in black. The solid purple lines are the fitting results of an automated step‐finding algorithm. PS represents phosphorothioate.
• D
  A representative gel showing that EXD2 (500 nM) cleaves the Cy5‐RNA and Cy5‐RDH substrates.
• E
  The model of EXD2 unwinding RNA–DNA hybrid before excising nucleotides one at a time.

Source data are available online for this figure.

After sudden unwinding a few base pairs of RNA–DNA hybrids, how many ribonucleotides does EXD2 digest at one time? To answer that, we performed bulk RNA cleavage assays with Cy5‐labeled ssRNA (100 nt) and RDH substrates (100 bp) and examined the cleavage products of EXD2 toward them. Like ssDNA cleavage (Broderick et al, 2016), EXD2 also digests RNA within both ssRNA and hybrids one ribonucleotide at a time, albeit unwinding a few base pairs of the hybrid duplex first (Fig 2D and E).

Stepwise RNA degradation by EXD2 reveals two rate‐limiting steps

Next, we further examined the dynamics of the RDH cleavage by EXD2. Through a careful analysis of the real‐time FRET trajectories, we were surprised to find a discrete stepwise increase in FRET efficiency (Fig 3A). A circular dichroism (CD) spectrum measurement of the ssDNA sequence paired with the degrading RNA excludes the formation of DNA secondary structure (Appendix Fig S5; Bishop & Chaires, 2003; Kim & Lee, 2021). Additionally, experiments with EXD2 anchored on the glass surface represented a similar stepwise pattern in FRET changes, ruling out that this observation originates from multiple EXD2 effectors (Fig EV2C–E). We employed an automated step‐finding algorithm to unbiasedly quantify the stepwise behavior (Fig 3A; Juette et al, 2016). This analysis revealed that there were commonly five peaks in FRET states unevenly distributed in the range from 0.2 to 0.9 (Fig 3B and Appendix Fig S6A and B). A transition density plot (TDP) was employed to demonstrate the two‐dimensional histogram for pairs of FRET values before and after each transition. This plot presents the four stepwise increments in FRET efficiency between the five states (Fig 3C). It is also noteworthy that 10% of the trajectories exhibited six states with an additional E state of ~0.22 (Appendix Fig S6C and D). This E state agrees well with the status of the EXD2‐bound substrate in the absence of Mn²⁺ and possibly represents EXD2 binding to the substrate prior to cleavage (Fig 1B). It is plausible that this RDH binding step is too transient to be resolved in the commonly detected five‐state trajectories.

Figure 3. Stepwise RNA cleavage by EXD2 and kinetic analysis.

1. Representative smFRET trajectory showing a stepwise FRET increase during EXD2 cleavage of the RDH. The solid line in red is the fitting result from an automated step‐finding algorithm.
2. The E histogram showing five E states.
3. The TDP representing FRET efficiency changes before and after transition for the EXD2 degradation of RDH.
4. A cartoon depicting one binding and four pausing intermediates during RDH cleavage by EXD2 and the sudden stepping transitions between the intermediate states.
5. The dwell‐time histograms of Pauses 1–3 and their combination (Total). The histogram fittings to a double exponential function are shown in red. The R ² of the double exponential function are 0.84, 0.82, 0.87, and 0.92, respectively.
6. The kinetic rates k ₁ and k ₂ obtained from the double exponential fitting to the dwell‐time histograms of Pauses 1–3 and their combination (Total). Fit parameters and 95% confidence intervals are shown.

We further asked how many base pairs of the hybrid duplex were unwound in each burst. To address this question, we carried out the smFRET experiment with a tailed RDH substrate containing a 6‐nt ssDNA at the 5′ end (6t‐RDH) to mimic a hydrolytic intermediate of the RDH substrate (Fig EV3A–C). Besides protein binding, three steps were monitored with this 6t‐RDH substrate (6t‐RDH) (Fig EV3A–C). However, compared to the blunt substrate, real‐time trajectories with 6t‐RDH displayed utterly different E states. The missing one unwinding step suggests an unwinding step size smaller than 6 bp. Meanwhile, the distinct E states exclude an unwinding step size of 1, 2, or 3 bp. We next carried out the smFRET with an 8‐nt ssDNA‐tailed RDH (8t‐RDH). In contrast to the 6t‐RHD substrate, experiments with 8t‐RDH revealed three FRET states and two unwinding steps that coincide with those with the blunt RDH substrate (Figs 3B and EV3D–F). The missing two unwinding steps with the 8t‐RDH substrate and the coincidence of the three E states suggest a 4‐bp (8/2) hybrid unwinding step of EXD2 (Fig 3D).

Figure EV3. Stepwise cleavage of tailed RDH substrates by EXD2.

• A–F
  (A and D) Representative smFRET trajectories show a stepwise FRET increase during the cleavage of the 6t‐RDH (A) and 8t‐RDH (D) substrates by EXD2. The solid red lines are the fitting result of an automated step‐finding algorithm. (B and E) The E histogram exhibit five E states and four steps during the cleavage of the 6t‐RDH substrate (B) and four E states and three steps during the cleavage of the 8t‐RDH substrate (E). (C and F) The TDP representing FRET efficiency changes before and after transition for the EXD2 degradation of 6t‐RDH (C) and 8t‐RDH (F) substrates.
• G, H
  Note that the first step detected in the FRET increase stems from EXD2 binding to the substrates, as confirmed by the control experiments with the 6t‐RDH (G) and 8t‐RDH (H) substrates.

The iteration of pausing and stepping in the hydrolysis process motivated us to propose an unwinding‐degradation coordination model for EXD2. As shown in Fig 3D, the sudden increments in FRET likely represent a quick 4‐bp hybrid duplex unwinding. The following pauses possibly reflect the times required for degrading ssNA and recycling the released energy for the next round of 4‐bp unwinding. To better understand the kinetics of EXD2 in each unwinding‐degradation cycle, we performed a statistical analysis of the dwell times. Consistently, individual (Pauses 1–3) and combined (Total) dwell‐time histograms all contained a rising phase and a decaying phase (Fig 3E), indicating that more than one rate‐limiting step dominated the pauses. It is well acknowledged that the dwell‐time histogram follows a gamma distribution if there are m equivalent steps (Lu et al, 1998; Xie, 2001; Chemla et al, 2008). However, a gamma distribution fitting to these histograms yields a noninteger step number m of ~1.7, suggesting nonequivalent rate‐limiting steps (Appendix Fig S7). We first speculated that there are two rate‐limiting steps. From the kinetic analysis (Xue et al, 2006), the dwell‐time histograms of a two‐rate‐limiting‐step process can be described by the following equation:

$$f\left(t\right)=\frac{k_1{k}_2}{k_2-k_1}\left(\mathit{exp}\left(-k_{1}t\right)-\mathit{exp}\left(-k_{2}t\right)\right),$$

where k ₁ and k ₂ are the kinetic rates of the two rate‐limiting steps. Indeed, four histograms are all well fitted by the equation, yielding a slow rate of ~0.28 s⁻¹ (k ₁) and a fast rate of ~1.36 s⁻¹ (k ₂) (Fig 3E and F). We also conducted a kinetic fitting to the histograms assuming that the pauses constituted three rate‐limiting steps. This fitting yielded two rates (k ₁ and k ₂) similar to those from the two‐rate‐limiting‐step fitting and an extremely fast rate k ₃, further supporting the existence of only two rate‐limiting steps (Appendix Fig S7). Altogether, we conclude that EXD2 carries out cyclic 4‐bp stepwise unwinding, with each cycle containing two nonequivalent rate‐limiting steps.

The rate‐limiting steps are dsNA unwinding and nucleotide excision

We next aimed to identify the two rate‐limiting steps within individual hydrolysis cycles of EXD2 (Fig 4A–F). As evident previously, many exonucleases showed a characteristic of sequence‐dependence in dsNA cleavage (Perkins et al, 2003; van Oijen et al, 2003; Lee et al, 2012). This characteristic is due to the differences in the energy of base stacking and hydrogen bonding of different NA sequences (van Oijen et al, 2003; Mourgues et al, 2005; Matsumoto et al, 2020). Compared with the RDH substrate, EXD2 distinctively behaves on the AU‐RDH substrate (Figs 1C and 2C). Moreover, EXD2 displayed a relatively slower dsNA cleavage than ssNA in the bulk assays (Fig 2D), in agreement with previous studies (Silva et al, 2018). Combined with the finding that EXD2 unwinds the RNA–DNA hybrids before nucleotide excision (Fig 2A–C), we speculated that one of the rate‐limiting steps of EXD2 could be duplex unwinding. To test this hypothesis, we performed the smFRET cleavage assay under 1 mM Mn²⁺ with a GC‐rich (88%) RDH substrate (GC‐RDH) to attenuate the duplex unwinding by EXD2. Real‐time trajectories and the TDP obtained from the cleavage on this substrate consistently revealed five centered E states and four unwinding steps (Fig 4A and B). Notably, compared with the RDH substrate containing random sequence (50% GC), the dwell times with this substrate were significantly prolonged. After confirming the absence of additional rate‐limiting steps (Appendix Fig S8), we used the two‐rate‐limiting‐step scheme to describe the combined dwell‐time histogram (Total) (Fig 4C). Compared to the RDH substrate, the two‐rate‐limiting‐step fitting to the histogram yielded significantly reduced k ₁ and moderately decreased k ₂ (Fig 4G). Therefore, duplex unwinding is one of the two rate‐limiting steps in EXD2‐mediated hybrid cleavage.

Figure 4. Kinetic analysis of EXD2‐mediated RDH cleavage under various conditions.

• A–F
  (A and D) Representative smFRET trajectories showing a stepwise FRET increase during EXD2 cleavage of the GC‐RDH substrate under 1 mM MnCl₂ (A) or the RDH substrate under 0.2 mM MnCl₂ (D). The solid red lines are the step‐fitting results from an automated step‐finding algorithm. (B and E) The corresponding TDP of EXD2 cleavage of GC‐RDH under 1 mM MnCl₂ (B) or RDH under 0.2 mM MnCl₂ (E). (C and F) The dwell‐time histograms of combined pauses (Total) in the process of EXD2 cleavage of GC‐RDH under 1 mM MnCl₂ (C) or RDH under 0.2 mM MnCl₂ (F). The histogram fittings to a double exponential function are shown in red. The R ² of the double exponential function are 0.94 and 0.95, respectively.
• G
  Two characteristic rates of EXD2 degradation of RDH under the indicated experimental conditions. Fit parameters and 95% confidence intervals are shown.
• H
  The kinetic pathway of EXD2 cleavage of RDH constitutes two rate‐limiting steps: nucleotide excision and duplex unwinding.

As demonstrated by our gel assays and previous studies, EXD2 is not a fast exonuclease in terms of ssNA degradation (Fig EV1A; Broderick et al, 2016; Silva et al, 2018; Park et al, 2019). Therefore, ssNA degradation could be the other rate‐limiting step during dsNA cleavage by EXD2, the rate of which could be tuned by adjusting the Mn²⁺ concentration (Fig EV1A; Park et al, 2019). To examine this hypothesis, we revisited the smFRET cleavage assays with the Mn²⁺ concentration reduced from 1 to 0.2 mM to retard EXD2 in ssNA degradation. An optical tweezer‐based hybrid unzipping assay confirmed that the influence of the Mn²⁺ change on the RNA–DNA stability is negligible (Appendix Fig S9; Huguet et al, 2017). Under this condition, both real‐time FRET trajectories and their corresponding TDP with the RDH substrate demonstrated a consistent four‐step unwinding process of EXD2 (Fig 4D and E). As expected, the dwell times between sudden unwinding steps elongated (Appendix Fig S8B). We also used the two‐rate‐limiting‐step scheme to describe the histogram of the combined dwell times (Total) (Fig 4F). Compared with those under 1 mM Mn²⁺, this fitting generated a similar k ₁ but a nearly halved k ₂ (Fig 4G). A three‐rate‐limiting‐step fitting excluded the possibility that the decrease in Mn²⁺ incorporates additional rate‐liming steps (Appendix Fig S8B). Therefore, it is highly likely that k ₂ reflects one of the rate‐limiting steps, namely, ssNA degradation. It is also noteworthy from both real‐time trajectories and the TDP that RDH rewinding occasionally and transiently occurred under this condition (Fig 4D and E). These rewinding events could be due to delayed ssDNA degradation by EXD2 after RDH unwinding.

The ssNA degradation of an exonuclease typically consists of two reaction steps: nucleotide excision and product release (Yang, 2011; Zhang et al, 2011). To further determine which one serves as the rate‐limiting step, we carried out additional smFRET experiments with the presence of GMP. If product release limits the nucleotide excision rate of EXD2, the involvement of GMP would cause a reduced rate, and two rates with a shortened k ₂ from the dwell‐time histogram fitting would be expected. However, GMP introduction led to the appearance of a third rate‐limiting step instead (Fig EV4A). Furthermore, we performed the smFRET assay with the RDH substrate under a high Mn²⁺ concentration of 10 mM. The high divalent metal concentration was also expected to perturb the product release of a two‐metal‐ion exonuclease (Hwang et al, 2018). According to the dwell‐time histogram fitting, the increase in Mn²⁺ concentration also introduced a third rate‐limiting step (Fig EV4B). Thus, instead of product release, nucleotide excision is likely to be another rate‐limiting step in the process of EXD2‐mediated RDH cleavage.

Figure EV4. The RDH cleavage by EXD2 in the presence of 0.5 mM GMP or 10 mM MnCl₂ .

• A–F
  (A and D) Representative smFRET trajectories showing a stepwise FRET increase during RDH cleavage by EXD2 in the presence of 0.5 mM GMP (A) and under 10 mM MnCl₂ (D). The solid red line is the fitting result of an automated step‐finding algorithm. (B and E) The TDPs exhibit the stepwise cleavage of RDH by EXD2 in the presence of 0.5 mM GMP (B) and under 10 mM MnCl₂ (E). (C and F) The dwell‐time histograms of combined pauses (Total) in the process of EXD2 cleavage of RDH in the presence of 0.5 mM GMP (C) and under 10 mM MnCl₂ (F). The histogram fittings to a triple exponential function are shown in red. The R ² of the triple exponential function is 0.95 (C) and 0.86 (F), respectively.

Based on these findings, we attribute the two rate‐limiting steps of EXD2 in RDH cleavage to duplex unwinding and nucleotide excision with respective kinetic rates k ₁ and k ₂ (Fig 4G and H).

DNA degradation by EXD2 is determined by its paired strand

Next, we constructed a 53‐bp DNA–RNA hybrid substrate with RNA labeled with fluorophores and DNA labeled with biotin (hereafter referred to as DRH with the first letter “D” representing the degrading strand of DNA). Using this substrate, we further examined the EXD2‐mediated degradation of DNA in the hybrid (Fig 5A). Compared with a minor increase in E from EXD2 binding without Mn²⁺ (0 s), a 100‐s incubation of this substrate with EXD2 supplied with 1 mM Mn²⁺ caused a moderate shift of the FRET efficiency from ~0.18 to higher values (Fig 5B). In comparison, a 200‐s incubation led to a further right shift of the overall E values, indicating that the DNA within the hybrids was inefficiently cleaved by EXD2 in 100 s. Real‐time trajectories showed that the increases in E were often interrupted by long pauses, which could not be resumed within our experimental time (100 s) (Fig 5C).

Figure 5. DNA degradation by EXD2 and its interruption by NA rewinding.

• A
  A schematic showing the smFRET cleavage assay. The 3′–5′ degradation of the biotin‐labeled DNA by EXD2 gives rise to the gradual generation of the fluorescently labeled ssRNA, causing an increase in FRET efficiency. PS represents phosphorothioate.
• B
  Histograms of the DRH substrate in the absence of EXD2 (gray) and after 0‐ (black), 100‐ (purple), and 200‐s (blue) incubation with 200 nM EXD2.
• C
  Representative smFRET trajectory showing EXD2 cleaving ssDNA in DRH in real time.
• D
  Representative smFRET trajectory showing a stepwise FRET increase during DRH cleavage by EXD2. The solid red line is the step‐fitting result from an automated step‐finding algorithm.
• E, F
  The E histogram (E) and TDP (F) showing five E states and four unwinding steps in the process of DRH cleavage by EXD2.
• G
  The dwell‐time histogram of pause 1 in the process of DRH cleavage by EXD2. The histogram fitting to a double exponential function (red) yields two kinetic rates. The R ² of the double exponential function is 0.85.
• H
  The percentage of the FRET trajectories stalling at the indicated E states.
• I
  The percentage of the FRET trajectories showing DRH rewinding prior to stall at the indicated E states.

We further investigated the similarities and discrepancies between the exoribonuclease and exodeoxyribonuclease activities of EXD2 toward the hybrids. A close examination of the real‐time FRET trajectories with DRH also demonstrated a stepwise increase in FRET efficiency (Fig 5D). The E distribution and TDP analysis led to the conclusion that five E states and four sudden unwinding steps also exist (Fig 5E and F). The dwell‐time histogram from Pause 1 can also be well described by the two‐rate‐limiting‐step scheme (Fig 5G). However, the E distribution reflects that compared with RDH, the highest E state with DRH centered 0.87 was significantly less visited (Figs 3B and 5E). Moreover, sudden high‐to‐low transitions appeared in the corresponding TDP, indicating DRH rewinding (Fig 5F). The statistics showed that a majority of trajectories stalled at three intermediate states (Pauses 1–3), and less than 30% of them completed the fourth unwinding step (Fig 5H). As demonstrated by the bulk ssDNA degradation assay (Fig EV1A), one of the explanations for this observation could be the inefficient ssDNA degradation by EXD2 that delays the subsequent DRH unwinding, thus in favor of the appearance of duplex rewinding. In support of this notion, we found that rewinding events prior to a longstanding stall indeed occurred in up to half of the FRET trajectories (Fig 5I). In comparison, rewinding before a longstanding stall occurred in less than 6% of the trajectories with the RDH substrate (Appendix Fig S10). Therefore, the inefficient ssDNA degradation partially contributes to the non‐processive DRH cleavage by EXD2.

We also examined the exodeoxyribonuclease activity of EXD2 toward dsDNA. The dsDNA substrate exhibited a consistent characteristic of slow dissociation from EXD2 with a binding rate comparable to that of RDH (Appendix Fig S1 and Fig EV5A–E). However, unlike the DRH substrate, FRET histograms and real‐time trajectories showed extremely poor exodeoxyribonuclease activity of EXD2 toward dsDNA (Fig EV5F and G). In line with the FRET results, bulk assays presented similar exonuclease activities of EXD2 toward these substrates (Fig EV5D and E). In conclusion, EXD2 displayed promiscuous exonuclease activities toward dsNAs, which depend not only on the degrading strand but also on the type of paired strand.

Figure EV5. Binding and exonuclease activities of EXD2 on dsDNA.

1. Representative fluorescence images showing Cy3‐labeled dsDNA bound by surfaced‐anchored EXD2 at indicated times and 10 min after the washout.
2. Normalized number of dsDNA molecules bound by EXD2 as a function of time. The solid lines are the single exponential fitting to the increase in molecule numbers under different substrate concentrations.
3. Normalized fluorescence signals of the Cy3‐labeled dsDNA substrates in the presence and absence of anti‐His and His₆‐EXD2 as a function of time.
4. k _binding under different dsDNA concentrations. The solid lines are the linear fits to the data. Error bars represent SD (n = 3).
5. Normalized number of dsDNA molecules at the indicated times (1, 5, 10, and 15 min) after substrate washout. Each data point is the average of three experiments. Error bars represent SD.
6. E histograms of the dsDNA substrates in the absence of EXD2 (gray) and after 0‐ (black), 100‐ (purple), and 200‐s (blue) incubation with EXD2.
7. Representative smFRET trajectories showing EXD2 cleaving dsDNA in real time. The fluorescence intensities of the donor and acceptor are colored green and red, respectively. The corresponding FRET efficiencies are shown below in black. Collectively, these findings support that EXD2 barely cleaves dsDNA, though it promptly and stably binds to it.

Discussion

This work examined the dynamics of EXD2 cleaving dsNA substrates at the single‐molecule level (Figs 1, 5 and EV5). Unlike other exonucleases that smoothly unwind and degrade dsNA (Yoo & Lee, 2015; Lee et al, 2022a, 2022b), EXD2 was found to iterate a 4‐bp duplex unwinding step during RNA–DNA hybrid cleavage. This distinct behavior allowed us to explore the kinetics of EXD2‐mediated dsNA cleavage (Fig 3). The dwell‐time analysis points to two rate‐limiting steps, which were further designated dsNA unwinding and nucleotide excision (Fig 4). The stepwise duplex unwinding by EXD2 resembles the behavior of Rrp44, which was also found to open four base pairs of dsRNA in each hydrolysis cycle (Lee et al, 2012). A spring‐loaded mechanism was proposed for Rrp44. Briefly, the chemical energy released from a series of hydrolysis of the RNA chain was accumulated and later converted into elastic energy for unwinding several base pairs in bursts (Lee et al, 2012). It is plausible that EXD2 employs a similar unwinding‐excision coordination mechanism. A structural study proved that EXD2 preferentially exists in a dimeric form (Park et al, 2019). Examining whether and how dimeric EXD2 regulates coordination warrants further investigation. For Rrp44, dsRNA unwinding is the only rate‐limiting step in its catalytic cycles. EXD2 differs from Rrp44 in that both nucleotide excision and duplex unwinding govern its overall reaction rate (Fig 4). This is possible because EXD2, compared with Rrp44, is a slow exonuclease, holding a nucleotide excision rate comparable to that of dsNA unwinding (Fig 4).

We also compared the exoribonuclease and exodeoxyribonuclease activities of EXD2 (Fig 5). Intriguingly, despite stable binding, EXD2 shows promiscuous exonuclease activities toward different dsNAs (Figs 1, 5 and EV5). In general, EXD2 preferentially cleaves RNA over DNA, and the exodeoxyribonuclease activity of EXD2 on dsNAs is regulated by the sugar type (ribose vs. deoxyribose) of the complementary strand. The inefficiency of EXD2 in cleaving DNA is not due to unstable binding to the substrate. Moreover, both RNA and DNA cleavages of the hybrids by EXD2 proceed stepwise. The similar four‐step unwinding of the examined duplex DRH implies that the step size of DRH unwinding by EXD2 may also be 4 bp. However, compared with RDH, rewinding occurred more frequently during DRH cleavage by EXD2, possibly because of inefficient ssDNA degradation. A mechanistic model is illustrated in Fig 6 to summarize these findings and to highlight the differences between EXD2‐mediated RNA and DNA cleavage.

Figure 6. A mechanistic model of EXD2 cleaving RNA–DNA hybrids.

EXD2 efficiently binds to RNA–DNA hybrids, inducing the initial unwinding of the duplex (gray box). EXD2 then enters the cyclic unwinding and excision stage, wherein a 4‐bp duplex is unwound and excised periodically (green box). The two reactions have comparable rates, and both are rate‐limiting steps. Compared with RNA cleavage, EXD2 is not efficient in excising DNA (indicated by smaller gray arrows). After a few unwinding‐excision cycles, it is often interrupted by duplex rewinding events and enters a longstanding stall stage (red box).

Our work also has important implications for the potential cellular functions of EXD2. Based on our findings, it is conceivable that the preferential degradation substrate of EXD2 in cells involves RNA (Silva et al, 2018). EXD2 was previously demonstrated to be a DNase, playing a role in HR‐mediated double‐strand break (DSB) repair (Broderick et al, 2016). However, EXD2 is not an efficient exonuclease regarding dsDNA cleavage (Fig EV5). Therefore, it is pretty unlikely that EXD2 alone digests dsDNA during HR. This incapability does not necessarily exclude EXD2 from HR. It is well appreciated that RNA–DNA hybrids exist as intermediates in HR‐mediated DSB repair (Ohle et al, 2016; Liu et al, 2021). EXD2 may participate in regulating these intermediates and help the removal of RNA for the generation of ssDNA in HR. Additionally, given the characteristic of stable binding to dsDNA (Fig EV5), EXD2 may act as a scaffold to recruit other partner proteins in HR. This characteristic may also assist EXD2 in protecting stressed replication forks, as demonstrated previously (Nieminuszczy et al, 2019).

Materials and Methods

EXD2 expression and purification

Recombinant EXD2 (61–621) was used for protein expression because of its improved solubility and yield (Broderick et al, 2016). The EXD2 gene was amplified by PCR using Homo sapiens cDNA as a template and inserted between the SacI and XhoI sites of the plasmid pET28a. In the resulting plasmid, pET28a‐EXD2, the N‐terminus of EXD2 was fused to a His₆‐tag which has proved not to interfere with the exonuclease activity of EXD2 (Appendix Fig S11). Two EXD2 mutants, the nuclease‐deficient EXD2‐D108A‐E110A and the dimerization‐deficient EXD2‐N198P, were produced by site‐directed mutagenesis (TransGen). The expression plasmid was transformed into BL21 (DE3) (TransGen). This strain was then cultured and induced by 1 mM isopropyl‐1‐thio‐D‐galactopyranoside (IPTG) at 18°C for 16 h. Cells were collected by centrifugation and resuspended in the lysis buffer containing 50 mM HEPES (pH 7.5), 300 mM NaCl, 10 mM imidazole, 5% glycerol, and 1 mM phenylmethylsulfonyl fluoride, and passed through a homogenizer three times at ~1,000 bar. The lysed dilution was then ultracentrifuged at 11,000 × g for 30 min. The supernatant was applied to a Ni‐Sepharose resin (TransGen) and was washed extensively with 50 mM HEPES (pH 7.5), 300 mM NaCl, 20 mM imidazole, and 5% glycerol. Then, the bound protein was eluted in a single step with the elution buffer containing 50 mM HEPES (pH 7.5), 300 mM NaCl, 250 mM imidazole, and 5% glycerol. Finally, the protein sample was buffer exchanged into the storage buffer containing 50 mM HEPES (pH 7.5), 500 mM NaCl, and 5% glycerol and stored at −80°C before use.

NA substrates preparation

For substrates of smFRET experiments and gel‐based bulk NA cleavage assays, HPLC‐purified DNA and RNA oligonucleotides with or without labels used to construct the templates were purchased from Sangon Biotech (Shanghai, China) and GenScript (Nanjing, China), respectively. The sequences of the oligonucleotides are listed in Appendix Table S1. The dsNA substrates for the smFRET experiments were produced by mixing fluorophore‐labeled DNA or RNA with complementary DNA or RNA at a molar ratio of 1:1.2 in the annealing buffer containing 10 mM Tris–HCl (pH 8.0) and 1 mM EDTA. The mixtures were incubated at 95°C for 3 min and then slowly cooled down to room temperature within 3 h. Annealed templates were stored at −80°C before use.

To construct the 100‐nt Cy5‐ssRNA substrate, a 100‐bp dsDNA including a 5′ T7 promoter (T7 DNA) was PCR amplified from plasmid pET28a. The 100‐nt Cy5‐ssRNA substrate was generated by transcribing from T7 DNA using 4 mM ATP, 4 mM CTP, 4 mM GTP, 4 mM UTP, and 0.5 mM Cy5‐UTP (ENZO Life Sciences). Then, the Cy5‐labeled ssRNA was purified using RNA Purification Kit (TransGen). The 100‐bp RNA–DNA hybrid substrate was constructed by heating the Cy5‐ssRNA and an HPLC‐purified 100‐nt DNA oligonucleotide (GenScript) at a molar ratio of 1:5 in annealing buffer containing 10 mM Tris–HCl (pH 8.0) and 1 mM EDTA at 95°C for 5 min and then slowly cooled down to room temperature within 3 h.

Gel‐based bulk RNA cleavage assays

For the bulk RNA cleavage assay, EXD2 (200 nM) was incubated with the Cy5‐ssRNA or Cy5‐RDH substrate (10 nM) in the reaction buffer containing 20 mM HEPES‐KOH (pH 7.5), 50 mM KCl, and 1 mM Mn²⁺ for the indicated time at room temperature. The reaction was quenched by proteinase K (TIANGEN) (Wu et al, 1999). The reaction products were resolved by 18% denaturing polyacrylamide gel electrophoresis (7 M urea PAGE) and visualized by phosphorimaging (GE Health Care). Cy5‐UTP was treated with Apyrase (NEB) at 25°C for 20 min to generate Cy5‐UMP as a marker. Each experiment was repeated in triplicate.

Single‐molecule FRET assay and data analysis

Single‐molecule FRET experiments were performed using TIRF microscopy (Ye et al, 2021). Utilizing an Andor EMCCD camera, fluorescence images were obtained with an integration time of 100 ms (10‐Hz frame rate). Quartz slides and coverslips (Fisher Scientific, USA) for the construction of the flow chamber were first coated with polyethylene glycol (PEG). Next, the coverslip was treated with aminosilane and coated with a mixture of 99% mPEG (m‐PEG‐5000, Laysan Bio, Inc.) and 1% biotin‐PEG (biotin‐PEG‐5000, Laysan Bio, Inc.). Next, the chamber was incubated with streptavidin (10 μg/ml) for 5 min. The hybrid or dsDNA substrate at a concentration of 10 or 20 pM was introduced into the chamber and immobilized on the quartz surface. After immobilization, free substrates were washed out of the chamber using 200 μl of the reaction buffer (20 mM HEPES‐KOH (pH 7.5), 50 mM KCl, 0.8% D‐glucose, 1 mg/ml glucose oxidase, 0.4 mg/ml catalase, and 4 mM Trolox). Then, 200 nM EXD2 and Mn²⁺ at the indicated concentrations were injected into the chamber prior to data acquisition.

The imaging was performed using a homemade two‐color TIRF microscope (Zhang et al, 2021). All images were recorded with an exposure time of 100 ms for 1,000 frames. Each frame was further processed to extract single‐molecule fluorescence intensities. Only fluorescence spots in the acceptor channel were used for analysis to avoid missing or inactivating the acceptors. The FRET efficiency of a single molecule was approximated as E = I _A/(I _D + I _A), where I _D and I _A are the donor and acceptor's background and leakage‐corrected emission intensities, respectively. Each experiment was performed at least three times to ensure reproducibility. Single‐molecule trajectories were collected using SPARTAN 3.7.0 (Juette et al, 2016). Hidden Markov modeling was performed on smFRET trajectories (Myong et al, 2007; Juette et al, 2016). The Baum–Welch algorithm embedded in SPARTAN (Juette et al, 2016) was used to determine the most likely number of unwinding steps, and the TDPs were correspondingly plotted. Over 120 molecules were analyzed for each experimental condition. The fitting results were used to construct the E histograms and determine the dwell times.

Author contributions

Xinshuo Jia: Conceptualization; data curation; software; formal analysis; validation; investigation; methodology; writing – original draft. Yanan Li: Conceptualization; data curation; software; formal analysis; validation; investigation; methodology; writing – original draft. Teng Wang: Data curation; methodology. Lulu Bi: Formal analysis; validation; methodology. Lijuan Guo: Validation; methodology. Ziting Chen: Validation; methodology. Xia Zhang: Methodology. Shasha Ye: Methodology. Jia Chen: Conceptualization; resources; methodology. Bei Yang: Conceptualization; resources; methodology. Bo Sun: Conceptualization; resources; formal analysis; supervision; funding acquisition; validation; investigation; methodology; writing – original draft; project administration; writing – review and editing.

Disclosure and competing interests statement

The authors declare that they have no conflict of interest.

Supporting information

Appendix

Expanded View Figures PDF

Source Data for Expanded View and Appendix

PDF+

Source Data for Figure 2

The EMBO Journal (2023) 41: e111703

Data availability

All data generated or analyzed during this study are included in the manuscript. This study includes no data deposited in external repositories.