Abasic sites and strand breaks in DNA cause transcriptional mutagenesis in Escherichia coli

Abstract

DNA damage occurs continuously, and faithful replication and transcription are essential for maintaining cell viability. Cells in nature are not dividing and replicating DNA often; therefore it is important to consider the outcome of RNA polymerase (RNAP) encounters with DNA damage. Base damage in the DNA can affect transcriptional fidelity, leading to production of mutant mRNA and protein in a process termed transcriptional mutagenesis (TM). Abasic (AP) sites and strand breaks are frequently occurring, spontaneous damages that are also base excision repair (BER) intermediates. In vitro studies have demonstrated that these lesions can be bypassed by RNAP; however this has never been assessed in vivo. This study demonstrates that RNAP is capable of bypassing AP sites and strand breaks in Escherichia coli and results in TM through adenine incorporation in nascent mRNA. Elimination of the enzymes that process these lesions further increases TM; however, such mutants can still complete repair by other downstream pathways. These results show that AP sites and strand breaks can result in mutagenic RNAP bypass and have important implications for the biologic endpoints of DNA damage.

Most cells in nature are in a state of limited growth and are therefore not often engaged in replication (1, 2). Thus, the functional viability of most types of cells likely depends more on faithful transcription and translation than on faithful replication. Several base damages are bypassed in vivo by RNA polymerase (RNAP) in a mutagenic manner, resulting in transcriptional mutagenesis (TM) (3 –7). In nondividing cell populations, TM could have a large contribution to the mutant protein burden and resulting cellular phenotype, compared with replication-based mutagenesis events.

Base excision repair (BER) is responsible for the processing of many small lesions, and in particular, those that do not cause significant helix distortion. During BER these lesions are removed by specialized DNA N-glycosylases. Monofunctional glycosylases remove the damaged base, leaving an abasic (AP) site. Bifunctional glycosylase/AP lyase enzymes remove the lesion and nick the DNA strand via β- or β,δ-elimination, resulting in a 3′ blocking group (8). Next, AP endonucleases process the AP sites and 3′ blocking groups to produce 3′-termini compatible with subsequent DNA repair synthesis (8, 9). In vitro, RNAP can bypass several types of repair intermediates, including AP sites and various strand breaks (10 –14). Whether RNAP is capable of bypassing abasic sites and strand breaks in live organisms is an important issue for defining the spectrum of DNA damages that can cause TM, as well as for understanding the relationships between RNAP and DNA repair processes. Here we have demonstrated that RNAP is capable of bypassing AP sites and strand breaks in Escherichia coli.

Results

We investigated TM mediated by AP sites and strand breaks by using damage-tailored plasmid constructs and using a transcriptional mutagenesis luciferase assay system (TM-LAS) (3, 15) (Fig. S1) in isogenic E. coli strains with different DNA repair backgrounds (Table S1). In this system, the configuration of the damage-containing constructs places the lesion across from a Stop codon sequence, enabling a measurement of TM through production of a full-length luciferase protein product only when an unrepaired lesion is bypassed by RNAP (Fig. S1) (3). Therefore, two factors can affect the total amount of active luciferase produced: (i) the rate of repair of the lesion by the available repair proteins in a cell, which will convert the lesion site to a Stop codon during repair synthesis, and (ii) the level of bypass by the RNAP for the lesion, which affects the total amount of full-length RNA produced from the luciferase gene. In addition to examining the functional protein product of the luciferase gene, cDNA sequencing was done to determine which nucleotide is inserted opposite the lesion during bypass, making this system particularly useful for examining TM resulting from a variety of DNA damages in vivo.

All experiments were carried out in the presence of the DNA gyrase inhibitor novobiocin to prevent DNA replication and hold the cells in a nongrowth state while allowing transcription to occur (5). By using cells under nongrowth conditions, we were able to study TM and DNA repair in a more natural state (16).

We used three AP site constructs: (i) URA/ST, containing a uracil to be converted to an AP site in vivo; (ii) AP^THF/ST, containing the AP site analog tetrahydrofuran that cannot be processed by AP lyases (17); and (iii) AP/ST, which contains a bona fide AP site created in vitro (Materials and Methods). The strand break that we investigated is that which is created through the β,δ-elimination activity of AP lyases, containing a one-nucleotide gap with phosphate groups on both the 3′ and 5′ ends. We used two strand break constructs: (i) 8OG/ST, which harbors an 8-oxoguanine to be converted to a strand break in vivo; and (ii) SSB/ST, which contains a strand break created in vitro (Materials and Methods).

AP Sites Cause TM in E. coli Deficient in AP Endonuclease Activity.

To examine TM caused by AP sites, we transformed the constructs into repair deficient E. coli. By comparing a single lesion (AP site) between strains with different DNA repair backgrounds, we were able to determine the relative roles of DNA repair enzymes in the removal of this lesion. First, we examined strains deficient in a single AP endonuclease, either Nfo or Xth, and determined the level of TM. In the nfo, but not the xth strain, there was a significant increase in TM at 120 min following transformation (Table 1), demonstrating that Nfo, but not Xth, is a major component for prevention of TM caused by AP sites. Although nfo is an inducible gene whereas xth is not, we were able to show using a β-galactosidase assay (18) with a strain containing an nfo′-lacZ fusion (Table S1) that nfo is not induced by treatment with novobiocin (Miller units for media with novobiocin, 766; media without novobiocin, 747). These results indicate that under conditions where nfo is induced, the effect on TM could be greatly enhanced. The removal of both E. coli AP endonucleases (xth nfo mutant) resulted in a significant elevation in TM at all time points examined (Table 1 and Fig. 1A), confirming that TM is further elevated when repair is substantially compromised. These results indicate that mutagenic RNAP bypass of AP sites occurs in vivo.

Table 1.

TM caused by abasic sites and strand breaks in repair-proficient and -deficient E. coli

Construct, Strain*	RLU†/106 cells ± SEM‡	Mutant/Wild-type§	P¶
URA/ST	TS** 3′ ATG UTT CCT 5′
	NTS†† 5′ TAC TAA GGA 3′
AB1157 (WT)	193 ± 32‡‡
AB1157 xth	207 ± 47	1.07	0.8057
AB1157 nfo	395 ± 40	2.04	0.00081
AB1157 xth nfo	1946 ± 302	10.07	0.0000031
AB1157 xth nfo nfi	641 ± 58	3.32	0.0000014
APTHF/ST	TS** 3′ ATG FTT CCT 5′
	NTS†† 5′ TAC TAA GGA 3′
AB1157 (WT)	585 ± 63
AB1157 xth	363 ± 113	0.62	0.0890
AB1157 nfo	1,999 ± 285	3.42	0.00014
AB1157 xth nfo	2,733 ± 378	4.67	0.000012
AB1157 xth nfo nfi	1,409 ± 173	2.41	0.00014
AP/ST	TS** 3′ ATG _TT CCT 5′
	NTS†† 5′ TAC TAA GGA 3′
AB1157 (WT)	371 ± 243
AB1157 xth	346 ± 100	0.93	0.9337
AB1157 nfo	1,104 ± 222	2.98	0.0473
AB1157 xth nfo	2,774 ± 355	7.48	0.00032
AB1157 xth nfo nfi	874 ± 217	2.36	0.1662
8OG/ST	TS** 3′ ATG 8TT CCT 5′
	NTS†† 5′ TAC TAA GGA 3′
AB1157 (WT)	410 ± 135‡‡	1.00
AB1157 xth	194 ± 94	0.47	0.2742
AB1157 nfo	3015 ± 575	7.35	0.0013
AB1157 xth nfo	21160 ± 3261	51.56	0.000083
AB1157 xth nfo nfi	7773 ± 1292	18.94	0.00021
SSB/ST	TS** 3′ ATG //TT CCT 5′
	NTS†† 5′ TAC TAA GGA 3′
AB1157 (WT)	87 ± 22
AB1157 xth	78 ± 6	0.90	0.6946
AB1157 nfo	1523 ± 697	17.50	0.0847
AB1157 xth nfo	3351 ± 510	38.51	0.000057
AB1157 xth nfo nfi	770 ± 344	8.85	0.000048

*Full strain descriptions in Table S1.

^†Relative light units measured 120 min post IPTG induction.

^‡Each value is the average of at least three replicate samples ± SEM.

^§Ratio of mutant luciferase activity over repair-proficient luciferase activity for the same construct.

^¶ P values for Student's t- test comparison between mutant and repair-proficient luciferase activity for the same construct. Distributions were considered to be significantly different when P < 0.01.

**Transcribed strand. U = uracil, F = tetrahydrofuran, _=abasic site, 8 = 8-oxoguanine, //=strand break.

^††Nontranscribed strand.

^‡‡These data included in a previously submitted manuscript.

Fig. 1.

AP site-mediated transcriptional mutagenesis in vivo. (A) AB1157 xth nfo cells were transformed with URA/ST (open circles), AP^THF/ST (closed triangles), AP/ST (open triangles), or ST/ST (closed circles) construct, allowed a 30-min recovery from electroporation, and TM level (expressed as relative light units or RLUs) was assayed by TM-LAS (Fig. S1) at 0, 45, 120, and 240 min following IPTG initiation of transcription. Each point represents the mean of at least three replicates ± SEM. (B) AB1157 xth nfo nfi cells were transformed with designated damage-containing constructs, followed by a 30-min recovery from electroporation; then, at indicated times following luciferase induction, RNA was extracted to determine AP site-driven TM events. *In the AP/ST 5-min experiment, cells were immediately resuspended in medium containing IPTG following electroporation, and RNA was isolated 5 min later. RT-PCR and subsequent cDNA sequencing allowed determination of the ribonucleotide incorporated during RNAP bypass. The number of insertion events detected for each ribonucleotide is indicated for either adenine or uracil, with the percentage of the total in parentheses. Each number represents combined sequencing results from two independent transformation events with two independent preparations of construct. Uracil at the first position of codon 445 results from transcription of repaired molecules and yields a Stop codon (producing inactive luciferase), except in the case of AP^THF/WT, in which repaired constructs would direct adenine insertion at the first position of codon 445. Adenine insertion results from RNAP bypass over the AP site and yields a Lys codon (producing active luciferase).

There is not a substantial increase in TM following IPTG induction, however, it is known that the promoter used in these studies is leaky, allowing some degree of transcription to occur before IPTG in introduced to the culture media. This leakiness is particularly evident for the Norm/Norm control construct (Fig. S2A), where a large amount of luciferase activity is measured at t = 0. Therefore, it is likely that the majority of the TM caused by AP sites is occurring during the 30-min recovery time, but the lesions are also repaired during this time. This would mean that an increase in transcription at t = 0 would be an increase of transcription of repaired constructs, that would not yield active luciferase, and, therefore, the total amount of active luciferase would remain unchanged by IPTG induction. Although it might appear that at t = 0 the cells transformed with URA/ST construct displayed a lower level of TM compared with cells transformed with AP/ST or AP^THF/ST, these differences were not statistically significant (Fig. 1A).

Interestingly, despite in vitro evidence that an alternative repair enzyme, Nfi, can recognize AP sites (19 –21), here we observed a suppressive effect on the levels of TM caused by AP sites when Nfi was deleted (Table 1 and Table S2). Although we currently do not have an explanation for the observed suppressive effect, the same pattern is observed in the several different repair backgrounds examined.

RNAP Bypass of AP Sites Results in Production of In-Frame Mutant Transcripts.

Next, we determined the bases inserted by RNAP opposite the AP site. By using an xth nfo nfi strain we found that, for AP sites and AP^THF, RNAP inserted exclusively adenine with no deletion events observed (Fig. 1B). An AP^THF/WT construct was used to distinguish between uracil that is incorporated opposite a repaired lesion or that which results from RNAP insertion of both adenine and uracil opposite AP sites. In the AP^THF/WT construct, repair of the AP^THF would lead to thymine on the template strand during repair synthesis, subsequently coding for adenine during transcription. As we observed only adenine incorporation, we conclude that RNAP preferentially inserts adenine opposite AP sites, and all uracil incorporation observed for the Lesion/ST constructs results from transcription events across repaired templates.

cDNA sequencing not only reflected RNAP insertion events but also revealed rates of DNA repair and provided information about the DNA repair pathways involved in the repair of AP sites. It is difficult to infer rates of conversion of uracil to an AP site in vivo, as the sequencing results of transcription past the URA/ST construct revealed an unusual pattern of repair, which we cannot explain at this time. However, when we compare AP/ST and AP^THF/ST, we observe that AP^THF was repaired more slowly than AP (Fig. 1B). Because AP^THF cannot be processed by AP lyases, this indicates a role for AP lyases in the repair of AP sites in vivo (17). In addition, the use of the triple mutant xth nfo nfi strain for cDNA sequencing allowed us to simultaneously exclude three important DNA repair enzymes that could be involved in the repair of AP sites. It is interesting, therefore, that the transcript sequencing results demonstrate that repair of AP^THF occurs when AP endonuclease and Nfi-mediated repair is compromised under conditions when AP lyase cleavage can be excluded as a repair mechanism. Also, eliminating a nucleotide excision repair (NER) protein, UvrA, did not have any effect on TM caused by AP sites (Table S2). As indicated in the previous section, these results support the idea that the majority of TM caused by these lesions results from TM occurring before IPTG induction.

Strand Breaks Cause TM in E. coli Deficient in AP Endonuclease Activity.

We were also interested in strand breaks resulting from AP lyase–mediated β,δ elimination. As above, we compared the TM that resulted from bypass of a single lesion (strand break) between strains of different DNA repair backgrounds and determined the role of different repair enzymes in the repair of that lesion. We found that strand breaks do cause TM, with nfo mutants displaying increased TM compared to xth mutants (Table 1). These results demonstrate that the 3′-phosphatase activity of Nfo exerts a greater effect in prevention of strand break–mediated TM than that of Xth. Simultaneous removal of both AP endonucleases resulted in a further increase in TM for all time points examined (Table 1 and Fig. 2A), and, similar to TM caused by AP sites, the majority of TM caused by strand breaks appears to be occurring during the 30-min recovery, as there was not a substantial increase in TM following IPTG induction (Fig. 2A and Fig. S2B). We observed that Nfi cannot prevent strand break-mediated TM (Table 1 and Table S2). Collectively, these results demonstrate DNA strand break bypass by RNAP, resulting in TM.

Fig. 2.

Strand break–mediated transcriptional mutagenesis in vivo. (A) AB1157 xth nfo cells were transformed with 8OG/ST (open circles), SSB/ST (closed triangles) or ST/ST (closed circles) construct, allowed a 30-min recovery from electroporation, and TM level (expressed in relative light units or RLUs) for these transformed cells was measured as before (Fig. 1). (B) AB1157 xth nfo nfi cells were transformed with indicated damage-containing constructs, and RNA extracted to determine strand break-driven TM as before (Fig. 1).

RNAP Bypass of Strand Breaks Results in Production of In-Frame Mutant Transcripts.

The above findings revealed a significant in vivo role for AP endonucleases in processing 3′ blocking groups and preventing TM caused by strand breaks. Nevertheless, the observation of high levels of luciferase activity was unexpected, as this would require the production of high levels of full-length, in-frame luciferase mRNA. Previous in vitro studies demonstrated that strand breaks with 3′- and 5′-phosphate termini are not bypassed efficiently by prokaryotic RNAPs compared to other types of strand breaks (10 –14). In addition, the transcripts resulting from in vitro break/gap bypass contain deletions equal to the size of the gap (10 –14). Our cDNA sequencing revealed that RNAP bypass of strand breaks containing phosphate termini in vivo resulted in adenine incorporation, but no deletions, in the resulting transcripts (Fig. 2B). This result was completely unexpected and indicates that there are factors that aid RNAP in the bypass of strand breaks and gaps in vivo. However, the level of bypass is likely to be low compared with certain types of base damage. Accordingly, the 8OG/ST construct yielded higher TM levels than the SSB/ST construct (Fig. 2A and Table 1), most likely due to RNAP bypass of 8-oxoguanine occurring before conversion in vivo to a SSB via BER. This is supported by the results of 8OG/ST cDNA sequencing in xth nfo nfi cells, which revealed cytosine incorporation. Adenine and cytosine insertion events were previously observed using the 8OG/ST construct in glycosylase (MutM)–deficient strains (3).

Discussion

This study provides an in vivo demonstration of RNAP bypass of AP sites and strand breaks. In addition, through the use of a panel of E. coli strains deficient in various DNA repair enzymes, we were able to elucidate the relative roles of various repair enzymes in nondividing cells.

We demonstrated that Nfo plays a greater repair role than Xth for the processing of both AP sites and 3′ blocking groups in cells in a nongrowth state. Xth has generally been believed to be the major AP endonuclease in E. coli, whereas Nfo plays a more minor role; but here we show that Nfo is more important for the prevention of TM caused by BER intermediates. However, Xth is still important, as simultaneous removal of both Xth and Nfo resulted in an increase in TM compared with Nfo alone.

Once it was known that both of the lesions examined here could be bypassed by RNAP, it was important to determine the nucleotide that RNAP incorporated opposite the lesions and to determine whether there were any deletions in the transcript as had been observed in vitro. RNAP appears to follow the same “A-rule” of the DNA polymerases, wherein the polymerase will preferentially incorporate adenine opposite noninformative sites on the template strand (22).

It was surprising that when RNAP bypassed the gapped, strand break structure used in this study, there were never any deletions observed in the transcript population. In vitro, for all combinations of strand break ends observed, the resulting transcripts contained a deletion in the mRNA (10 –14). Our results demonstrate that there are factors that facilitate RNAP bypass of broken/gapped templates in vivo to prevent production of mRNA containing frameshift mutations, and that strand breaks are an additional noninformative lesion that follows the “A-rule” for polymerase bypass. In addition, these data demonstrate the importance of performing TM studies in live cells, as there are clearly many factors that can affect the bypass of lesions. It will be interesting to examine strand break bypass in eukaryotic cells, as human RNAP II can also bypass gapped strand breaks with various ends in vitro, albeit at a low efficiency (23).

Although RNAP can bypass strand breaks without introducing a deletion into the transcript, it is likely that the rate of bypass is low compared with other types of base damage. The levels of luciferase, and therefore TM, observed are mediated by several factors, including the rates of repair and the level of bypass. Although the TM-LAS does not allow us to independently measure these two factors, the transcript sequencing results allow several inferences to be made. Transcript sequencing revealed that the processing of strand breaks containing a 3′-phosphate terminus is slow compared with AP site processing and should therefore lead to higher levels of TM caused by strand breaks than that observed from AP sites. However, TM is approximately equal for these two lesions (Figs. 1A and 2A and Table 1), indicating that the rates of RNAP bypass of strand breaks must be low compared with those in the AP sites, effectively lowering the total amount of luciferase transcript that can be produced and lowering TM to the level observed for AP sites. Unfortunately, E. coli possess robust activities to remove truncated mRNAs from the transcript pool (24), preventing a direct measurement of the in vivo RNAP bypass and stalling at this time.

Transcript sequencing information allowed us to make several conclusions about repair of AP sites and strand breaks in vivo. Sequencing was carried out in xth nfo nfi DNA repair mutant strains. The difference in repair of AP sites versus AP^THF indicates a role for AP lyases in the repair of AP sites, even if this activity is uncoupled from the glycosylase activity. Most surprising was the observation that repair of AP^THF can occur even in the absence of AP lyase activity, AP endonucleases, and Nfi with nearly complete repair observed soon after the 30-min recovery period following transformation. We have also shown that NER is not involved in the repair of AP^THF, indicating that an additional mechanism of repair exists. It is interesting to note that, although it would have been especially informative to have conducted TM studies with AP^THF in a strain that is deficient in xth nfo nfi and uvrA concurrently, thus compromising all known AP site repair pathways simultaneously, we were unable to construct such a quadruple mutant strain, even when using a P1 stock that produced viable mutants in other backgrounds (Table S1). Together, these results support recent observations that repair of AP sites can occur through another, unknown mechanism in addition to those mediated by AP endonucleases, AP lyases, NER, or Nfi (25). Similarly, the strand breaks are processed in the absence of AP endonucleases, Nfi, and NER. It is known that nucleoside diphosphate kinase (NDK) possesses 3′-phosphatase activity, but a role in BER has not been demonstrated (26). A comparison of the transcript sequencing results of AP^THF and strand breaks demonstrates that the repair activity for strand breaks in these strains occurs at a somewhat lower level compared with that of the AP sites.

In the present study, we demonstrate that RNAP bypass of both AP sites and strand breaks can cause TM in vivo. As depicted in Fig. 3, both nonbulky base lesions, and the intermediates that result from BER, are bypassed by RNAP resulting in TM. Therefore, perturbation of any one of several steps of BER can increase the level of TM in a cell. There are several important biological consequences for TM (7). TM could be a mechanism for adaptive mutagenesis wherein the generation of a large mutant mRNA pool and subsequent accumulation of mutant proteins could alter the cellular phenotype. If such mutant proteins conferred a growth advantage, the altered phenotype could then become permanent during the ensuing round of DNA replication past the lesion in a process termed retromutagenesis (27). In bacteria and other unicellular organisms, retromutagenesis could result in escape from growth restricted environments, and/or the acquisition of antibiotic resistance.

Fig. 3.

RNAP bypass of AP sites and strand breaks reveal the relationship between the transcriptional machinery and DNA repair processes. Nonbulky lesions (small green box) in DNA cause transcriptional mutagenesis (TM) and are primarily repaired by enzymes in the base excision repair (BER) pathway. DNA N-glycosylases remove base lesions from the DNA, generating an abasic site (monofunctional glycosylase) or a strand break (bifunctional glycosylase/AP-lyase). Such BER intermediates also cause TM, where RNAP bypass results in transcripts containing adenine inserted opposite the site of damage. Repair of these BER intermediates can be mediated by AP endonucleases, and the absence of AP endonucleases will significantly increase TM caused by BER intermediates.

Materials and Methods

Strains and Media Conditions.

The genotype and sources of E. coli strains used in this study are listed in Table S1. New strains were constructed using generalized transduction with phage P1 Δdam rev6, as described previously (28), with the donor and recipient designated “P1(donor) x recipient.” Ultraviolet sensitivity of uvrA strain was confirmed for several tetracycline-resistant transductants. The nfi strain was confirmed for chloramphenicol-resistant transductants using PCR and primers Nfi1 and Nfi2 (Table S1) as previously described (20). Cell growth media was liquid LB supplemented with the following antibiotics, when appropriate: kanamycin (50 μg/mL), tetracycline (15 μg/mL), and ampicillin (50 μg/mL). Competent cell preparation was as described previously (3).

Preparation of Constructs.

Preparation of control constructs and constructs containing 8OG, URA, and AP^THF was described previously (3). SSB/ST construct was prepared by treating 8OG/ST construct with commercially available Fpg (New England Biolabs) as per the manufacturer's instructions. The quantitative conversion to a construct containing a SSB/ST was confirmed by agarose gel electrophoresis in the presence of ethidium bromide, wherein the amounts of covalently closed and nicked constructs can be distinguished and assessed. AP/ST construct was prepared by treating URA/ST construct with commercially available UDG (New England Biolabs) as per the manufacturer's instructions. Presence of the AP site is confirmed through the digestion of the AP construct with Nfo (a generous gift from Yoke Wah Kow, Emory University) and analysis with agarose gel electrophoresis in the presence of ethidium bromide. In vitro digested constructs were prepared on the day of transformation and filtered on Millipore membrane filters before electroporation.

Transcriptional Mutagenesis Luciferase Assay System (TM-LAS).

Luciferase assays were performed as described previously (3). The assay methodology is illustrated in Fig. S1.

β-Galactosidase assay for detection of nfo induction.

β-Galactosidase activity was measured in suspensions of cells grown in the presence or absence of novobiocin and were treated with CHCl₃ and SDS (18).

RNA Preparation and RT-PCR.

AB1157 xth nfo nfi cells were transformed with construct as previously described (3) and grown at 25°C; then RNA was isolated from the bacteria at appropriate time points after IPTG induction. RNA was prepared using PerfectPure RNA Cultured Cell Kit (5Prime), subjected to a 2 h DNase I digestion (Baseline-ZERO, Epicentre), and further processed using a DNA-Free RNA Kit (Zymo Research). An additional DNaseI digestion (Promega) containing all components of the subsequent RT-PCR (except random hexamers and reverse transcriptase) was performed for 6 h. Approximately 500 ng RNA was reverse transcribed using random hexamers (Applied Biosystems) and then PCR amplified using the primers LBRT1, LBRT2 (Table S1) to prime. Removal of contaminating DNA was confirmed by Taq PCR without a preceding reverse transcriptase step.

Transcript Sequencing.

Subcloning of cDNA was carried out by ligating the Sau3AI/HincII fragment of the RT-PCR product between BamHI and HincII sites of pUC18 as previously described (3), using X-Gal and IPTG on LB-Amp plates for blue-white screening. White ampicillin-resistant colonies were subjected to PCR amplification using Clo18U and Clo18L (Table S1). Sequencing was carried out by Macrogen or Agencourt using Clo18U to prime sequencing reactions.