Endogenous Mutagenesis in Recombinant Sulfolobus Plasmids

Abstract

Low rates of replication errors in chromosomal genes of Sulfolobus spp. demonstrate that these extreme thermoacidophiles can maintain genome integrity in environments with high temperature and low pH. In contrast to this genetic stability, we observed unusually frequent mutation of the β-d-glycosidase gene (lacS) of a shuttle plasmid (pJlacS) propagated in Sulfolobus acidocaldarius. The resulting Lac⁻ mutants also grew faster than the Lac⁺ parent, thereby amplifying the impact of the frequent lacS mutations on the population. We developed a mutant accumulation assay and corrections for the effects of copy number and differential growth for this system; the resulting measurements and calculations yielded a corrected rate of 5.1 × 10⁻⁴ mutational events at the lacS gene per plasmid replication. Analysis of independent lacS mutants revealed three types of mutations: (i) G·C-to-A·T transitions, (ii) slipped-strand events, and (iii) deletions. These mutations were frequent in plasmid-borne lacS expressed at a high level but not in single-copy lacS in the chromosome or at lower levels of expression in a plasmid. Substitution mutations arose at only two of 12 potential priming sites of the DNA primase of the pRN1 replicon, but nearly all these mutations created nonsense (chain termination) codons. The spontaneous mutation rate of plasmid-borne lacS was 175-fold higher under high-expression than under low-expression conditions. The results suggest that important DNA repair or replication fidelity functions are impaired or overwhelmed in pJlacS, with results analogous to those of the “transcription-associated mutagenesis” seen in bacteria and eukaryotes.

INTRODUCTION

The universal need of organisms to replicate genomes accurately, and in coordination with cell division, has special significance for “extreme” or “hyper-” thermophiles, which grow optimally at temperatures that kill mesophilic organisms and denature their macromolecules (1). The hyperthermophilic archaea which colonize terrestrial and marine hydrothermal systems represent diverse and ancient lineages, and all grow optimally at temperatures that accelerate DNA decomposition reactions by orders of magnitude relative to those in mesophiles (2). The idea that these archaea should have effective strategies of genome protection which compensate for diverse damaging effects of high temperature thus seems logical and is supported by the low rates of replication errors measured in several Sulfolobus isolates (3, 4). The extreme phylogenetic divergence separating archaea from bacteria and eukaryotes (5) further suggests that the strategies which preserve genome integrity in hyperthermophilic archaea may include molecular features not identified in model organisms. This idea remains consistent with the unusual gene inventories and certain biochemical features of DNA metabolism in these archaea (6). The mechanisms that maintain genomic integrity in hyperthermophilic archaea remain challenging to identify in vivo, however. Recent studies, for example, found that several putative homologous recombination and DNA repair genes are essential in Thermococcus, whereas other such genes were deleted without yielding an obvious damage sensitivity phenotype (7).

Basic genetic tools developed for hyperthermophilic archaea include shuttle vectors for Sulfolobus species constructed from plasmids pRN1 and pRN2 (8, 9), which originate in S. islandicus strain REN1H1 (10). Several features of these plasmids have intrinsic interest for molecular biology. For example, the large protein with DNA primase/polymerase activity encoded by pRN1 ORF904 (11) is essential for replication of pRN1 derivatives in Sulfolobus (8). This suggests that pRN1 uses DNA primers for a critical stage of its replication, making it and related plasmids potentially unique among known DNA replication systems. Here, we demonstrate that a Sulfolobus β-d-glycosidase gene, placed in pRN1 constructs as a readily scored marker, can mutate spontaneously at extremely high rates. The properties of this mutagenesis do not suggest a specific role for DNA primers but instead implicate one or more processes associated with high-level expression of the plasmid-borne gene.

MATERIALS AND METHODS

Strains, plasmids, and growth conditions.

The Sulfolobus acidocaldarius uracil auxotroph MR31 served as the host strain for transformations by electroporation as previously described (12). This strain contains an 18-bp deletion in the orotate phosphoribosyl transferase gene, pyrE, which provides selection for a functional pyrE gene carried on a plasmid or incorporated elsewhere in the genome (Table 1). S. acidocaldarius MR31 cells were transformed by Sulfolobus/Escherichia coli shuttle vectors pJlacS and pClacS_3712 (referred to here as pClacS), each of which carries an S. solfataricus pyrE gene used for selection of Pyr⁺ transformants (see Fig. S1 in the supplemental material). To protect the plasmid DNA from the S. acidocaldarius restriction-modification (R-M) system, SuaI sites (GGCC) were methylated in vivo prior to transformation (13, 14). Both vectors express the S. solfataricus β-d-glycosidase lacS reporter gene, but they use different promoters, i.e., the tf55α gene promoter in the case of pJlacS (15) and a maltose-inducible promoter in pClacS (16).

Table 1.

Strains, plasmids, and primers

Strain, plasmid, or primer	Descriptiona	Source or reference
S. acidocaldarius strains
DG185	Wild type	ATCC 33909
MR31	pyrE131 (18-bp internal deletion)	46
ESF	Genome-integrated Sso lacS	C. Joshua
SaES−F 5-11	lacS C1172T	Lac− ESF
pJlacS 1	lacS C286T	Lac− pJlacS transformant
pJlacS 17	lacS Δ(GATTG)938	Lac− pJlacS transformant
pJlacS 38	lacS +A212	Lac− pJlacS transformant
pClacS 5-4	lacS Δ(GATTG)938	Lac− pClacS_3712 transformant
pClacS 27-3	lacS C286T	Lac− pClacS_3712 transformant
pClacS 8-4	lacS T1032A	Lac− pClacS_3712 transformant
pClacS 23-4	lacS G522A	Lac− pClacS_3712 transformant
S. islandicus strains
Ren1H1	Wild type	10
Ren1H1 10-0	lacS +(CTGGT)1272	Lac− Ren1H1
Ren1H1 19-15	lacS +A1012	Lac− Ren1H1
Plasmids
pJlacS	Sulfolobus sp./E. coli shuttle vector	8
pClacS_3712	Sulfolobus sp./E. coli shuttle vector	16
PCR primers
SsoLacS2191r	CTGTGGATAACCGTATTACC
SsoLacS356f	ATCTTCTTCCTCCTACTACG
pClacS3712f	GATATCTGATAGTTGGAGAAATGC
pClacS3712reverse	CTGTCTCTTATACACATCTGGATC
EndpyrEfor	GTATATAAGGTAAGCGAAATACTG
BeginpyrFrev	TATATGGGTAGTGAAATCCTTTAG
LacSmidf	GATGTGACAGAGGTTGAGATAAA
Sisland_lacS_for	GAAAATGAGGTCAGGGTGAG
Sisland_lacS_rev	TCTCCCCTATTCTTGACGAT
Sis_lacSmidf	GTTGGAATGAAGAATTTGCG
1aF	CTAGATGAATTGTTGAAGAGTGA
1aR	CTGCATAATTCTCTTACTGTCAT
1bF	TTTAAAGTTCTGCTATGTGG
1bR	ACTTTCTCAAGTCTCACTAT
2aF	GCTAAACCGAGAAGTAATTT
2aR	TATGTGAAAGAGTAAGCAGT
2bF	TACTGTATACGTCTATCGTTT
2bR	CACATCACTTAACGTAAGATT

^alacS mutations are indicated by the local sequence change and position within the gene (in bp).

Unless otherwise noted, all growth media included a mineral base (17) supplemented with 0.1% Bacto tryptone; 0.2% Dextrin-10 (Fluka) was added for S. islandicus strains, and 0.2% d-xylose was added for S. acidocaldarius strains. In order to induce high levels of lacS expression in pClacS transformants, 0.2% maltose was substituted for d-xylose. Analyses of genome-integrated lacS used S. islandicus isolate Ren1H1 and S. acidocaldarius strain ESF (Table 1). The former is the natural source of pRN1 and pRN2, and the latter is an S. acidocaldarius construct in which the S. solfataricus β-d-glycosidase lacS gene is inserted between the chromosomal pyrE and pyrF genes by homologous recombination and expressed from the pyrEF common promoter (C. Joshua, personal communication).

Colony staining and enzymatic assays.

The activity of β-d-glycosidase in Sulfolobus clones was assessed first qualitatively on plates by colony staining and then quantitatively in liquid cultures by spectrophotometric assays. For colony staining, plates containing isolated colonies were sprayed with an warmed overlay mixture containing 200 μg/ml 5-bromo-4-chloro-3-indolyl-β-d-galactoside (X-Gal), 0.1 M KCl, 0.1 M, β-alanine, 10% ethanol (vol/vol) (EtOH), and 0.2% gellan gum (Gelrite). After the applied mixture had gelled (about 5 min at room temperature), the plates were incubated at 55°C for up to18 h. Colonies appearing very light blue or white in color compared to a stained Lac⁺ colony (dark blue) were analyzed further.

For quantitative measurements of β-d-glycosidase activity, approximately 10⁶ cells were incubated at 70°C for 10 to 240 min, depending on the strain, in 0.5 ml of an assay reagent (50 mM KCl, 50 mM sodium citrate [pH 5], 0.1% N-lauroyl sarcosine, 0.5 mM 4-nitrophenyl-β-d-galactopyranoside). The reaction was stopped by the addition of 0.3 ml of 1 M Na₂CO₃. The relative rate of substrate hydrolysis (ΔOD₄₂₀ per min per ml) was normalized by the cell density (optical density at 600 nm [OD₆₀₀] of cell suspension) to yield a relative specific activity analogous to the “Miller unit” of lacZ expression in E. coli strains (18).

PCR and sequencing.

Strains having <50% of the β-d-glycosidase specific activity of parental Lac⁺ cells were scored as Lac⁻. The lacS mutations were verified by PCR of the lacS gene followed by sequencing. The PCR program used initial denaturation at 95°C for 2 min, followed by 28 cycles of 95°C for 22 s, 48°C for 22 s, and 72°C for 3 min and final extension at 72°C for 7 min. The primers used to amplify lacS in (i) pJlacS transformants, (ii) pClacS transformants, (iii) Ren1H1, and (iv) S. acidocaldarius ESF were as follows, respectively: (i) SsoLacS219r and SsolacS356f, (ii) pClacS3712f and pClacS3712reverse, (iii) land_lacS_for and Sisland_lacS_rev, and (iv) EndpyrEfor and BeginpyrFrev (Table 1). The lacS gene was sequenced in three sections using the forward and reverse primers used to amplify the gene as well as a middle primer. Middle primers included lacSmidf for sequencing S. solfataricus lacS and Sis_lacSmidf for sequencing S. islandicus lacS (Table 1).

Nonselectable regions of pJlacS were similarly amplified and sequenced in two overlapping parts. Each of these regions was amplified with Velocity DNA polymerase (Bioline). The PCR program used initial denaturation at 97°C for 2 min, followed by 28 cycles of 95°C for 22 s, 48°C for 22 s, and 72°C for 1.5 min and final extension at 72°C for 7 min.

Competition assays.

Differential growth of Lac⁻ and Lac⁺ strains was quantified by competition assays. Clonally pure liquid cultures grown to a density of about 10⁸ cells/ml were pelleted and resuspended to an OD of 0.2 and then diluted and plated to determine cell viability. Initial ratios of inocula were chosen to obtain accurate colony counts for each Lac phenotype after growth; the Lac⁻/Lac⁺ ratio was 1:50 for the pJlacS strain and either 1:5 or 1:1 for all other strains. The cell mixtures were then cultured for 1 to 2 days, followed by subculturing (1:30) to fresh medium for a total of two or three transfers. Upon each transfer, cells were plated and stained to determine ratios of Lac⁻ to Lac⁺ cells. Ratios of specific growth rate constants (k^m/k^w) were then calculated as in previous studies (19) using the relationship k^m/k^w = [ln(M) − ln(M₀)]/[ln(W) − ln(W₀)], where M is the number of mutant cells in the culture, W is the number of wild-type cells, and subscript “0” refers to the beginning of the growth period.

Mutant accumulation assays.

To determine spontaneous lacS mutation rates, a number of cell lines were initiated from independently isolated Lac⁺ colonies by inoculation of liquid medium. Each of the resulting cultures of approximately 10⁸ cells was further divided into multiple identical aliquots and cryopreserved in 9% (vol/vol) dimethyl sulfoxide at −80°C.

Each independent culture in a mutation rate assay was started by inoculating fresh medium (3.0 ml) with a thawed aliquot (about 5 × 10⁶ viable cells) of the corresponding clone described above. Each culture was grown to a size of about 5 × 10⁸ cells, and 1/30 of the culture was transferred to fresh medium; the growth between transfers thus averaged about 5 generations. Depending on the strain, the independent cell lines were propagated for 1 to 25 transfers, and at each transfer, a sample of each mature culture was diluted quantitatively, plated, and incubated at 70°C for 6 to 7 days. The resulting colonies were then stained with X-Gal (see above) to identify Lac⁻ mutants, and parental and mutant colonies were counted to determine total cells per culture.

Calculations.

Calculations of mutation rates from mutation accumulation data (20) used the equation μ = f/ln(Nμ), where μ is the mutation rate, f is the frequency (proportion) of Lac⁻ cells in the population, and N is the theoretical population size, i.e., the size that would have resulted if no cells had been discarded at transfers (equal to about 2^5T, where T is the number of transfers). This equation was solved for μ by iteration, using an initial μ estimate in the right-hand term equal to f/5; values generally converged to about 1% agreement within 3 iterations. Unless otherwise noted, the median value of μ for a set of independent cultures was taken as the representative value for the set. For conditions under which most of the cultures revealed no Lac⁻ clones in plated samples, f was estimated from the Poisson term P₀ = (e^−x)(x⁰)/0! = e^−x by substituting the proportion of independent populations that did not reveal detectable Lac⁻ cells for P₀. Solving for x yielded the average number of Lac⁻ mutants per culture sample, and dividing x by the average number of colonies per culture sample yielded f for the set.

To correct for the empirically measured growth advantage of Lac⁻ mutants, mutation rates were calculated by the equation of Koch and Drake (21), i.e., μ = sf/[1 − e^{−sln(N/N₀)}], where N₀ and N represent initial and final population sizes, respectively, f is mutant frequency, and s is the selection coefficient, defined as the fractional growth penalty per generation (a negative value under our conditions). The value of s for Lac⁻ lineages was calculated for each Sulfolobus strain based on the competition experiments described above. The initial and final numbers of mutant (M₀ and M, respectively) and of wild-type (W₀ and W) cells in each competition were substituted into the relationship (1 + s)^log2(N/N₀) = MW₀/M₀W (21), and the equation was solved for s.

RESULTS

Instability of the lacS reporter gene in pJlacS.

During attempts to analyze chemically induced mutations in Sulfolobus, we transformed S. acidocaldarius with damaged (cisplatin-treated) pJlacS DNA and scored lacS function by colony staining. The control (undamaged) DNAs in these experiments yielded an unexpectedly large proportion of Lac⁻ colonies (about 0.3%). This frequency seemed too high to be explained by mutation during normal plasmid propagation in E. coli, which was used to produce DNA of sufficient amount and purity for the manipulations. However, the SuaI-cognate R-M system used to methylate the plasmids in E. coli (14) had not been evaluated for a mutagenic effect; in addition, electroporation had been reported to generate mutations in Sulfolobus (4). It was, accordingly, difficult to exclude these or other potential sources of mutagenesis a priori.

Therefore, to determine whether the observed Lac⁻ forms of pJlacS arose during propagation in S. acidocaldarius, we picked several pJlacS-transformed colonies, purified each clonally, and then monitored the resulting Lac⁺ populations as they expanded. In most cases, the cultures became predominantly Lac⁻ after a few transfers. This shift in composition of clonally purified cultures confirmed that the pJlacS mutation occurred during propagation in S. acidocaldarius and further suggested that mutants could outcompete the parental strain after their formation.

Although this phenomenon seemed to represent a dramatic example of genetic instability, the nonselectable nature of the mutants precluded mutation rate measurements by standard fluctuation assays. We therefore developed an alternative assay that monitored several independent clones of pJlacS transformants for the accumulation of mutants, starting at an early stage of growth. When Lac⁻ clones reached a measurable proportion of the culture (about 1%), the population size N and the frequency of the mutant cells f at that point were used to calculate the mutation rate μ (see Materials and Methods). When applied to 49 independent cultures, this method yielded extremely high f values and correspondingly high rates of spontaneous mutation; the median value of μ for lacS mutation across the set was 5.15 × 10⁻³ events per cell division (see Table S1 in the supplemental material). This is orders of magnitude higher than typical genic mutation rates, including those of chromosomal genes of S. acidocaldarius (3), suggesting that the accuracy of replicating the lacS gene was severely compromised.

Corrections to mutation rate measurements.

Given the extremely high mutation rate indicated by this analysis, we considered properties of the experimental system that could have overestimated μ. One relates to the copy number of shuttle plasmids derived from pRN1, which varies slightly among constructs and growth stages but averages about four per cell (8). This average copy number implies that a gene carried on the pRN1 construct is replicated four times per cell division. Applying this correction to the assay data yields 5.15 × 10⁻³/4 = 1.29 × 10⁻³ mutational events per gene replication, which is about 4,000 times the corresponding rate measured for the S. acidocaldarius pyrE gene (3).

Another factor that can distort mutation rate measurements is a difference in growth rate between mutant and parental strains. The fact that Lac⁻ clones inevitably dominated liquid cultures grown for many generations suggested a growth advantage of the mutant, and this was reinforced by the observation that Lac⁻ colonies identified in such cultures by colony staining were generally larger than Lac⁺ colonies. (The latter result was not an effect of X-Gal toxicity, since X-Gal was applied to plates only after colony formation.) We therefore measured the fitness difference between Lac⁻ and Lac⁺ clones experimentally, so that its impact on mutation rate calculations could be corrected. Competition assays were performed as reconstruction experiments (see Materials and Methods), and the resulting ratio of specific growth rate constants, k^m/k^w, averaged over several distinct lacS mutant alleles was 1.23 (see Table S2 in the supplemental material). This ratio represents a significant growth advantage; each 10-fold increase in the Lac⁺ cell number, for example, is matched by a 17-fold increase in Lac⁻ cells, making corresponding corrections important for the accuracy of mutation rate calculations (see below).

A third correction we considered relates to phenotypic lag. In our assay, a mutational event does not create a phenotypically Lac⁻ cell until further cell division produces “homozygous” daughter cells in which all copies of lacS are mutant. Because our assays used nonlethal scoring of fully grown colonies, the potential impact of this lag on mutant detection was confined to its effect on colony staining. Based on Lac⁻ mutants exhibiting residual enzyme activity (see Table S2 in the supplemental material), we estimated that colonies with fewer than about 10% Lac⁺ cells would be scored reliably as Lac⁻ under our conditions. Therefore, the question of impact on the mutation assays reduces to the following: if a cell containing three Lac⁺ copies of pJlacS and one Lac⁻ copy was plated, would the resulting colony contain at least 90% Lac⁻ cells when stained?

The calculations were based on the empirically measured growth advantages (k^m/k^w ratios) and the assumption that Lac⁺ and Lac⁻ pRN1 constructs segregate randomly into daughter cells; random segregation is commonly assumed for simple plasmids and multicopy prokaryotic chromosomes, although we note that it rarely has been confirmed experimentally. When k^m/k^w equals 1.23, the colony arising from the hypothetical cell that contains one mutant plasmid reaches the threshold of 90% Lac⁻ at a size of 4 × 10⁷cells, which corresponds to a colony about twice as large as those we routinely scored. This suggests that colony staining would generally miss only those mutations occurring within the last generation of culture growth before plating. Corresponding calculations also determined that cultures in which Lac⁻ mutants have much less, or no, growth advantage would similarly have a negligible impact on mutant detection. This reflects the fact that although four Lac⁻ copies per cell plated would be required for a colony to score as Lac⁻ under these conditions, the corresponding mutation accumulation assays require many more generations of growth (see Table S1 in the supplemental material). As a result, heterozygous cells formed by recent mutational events represent a much smaller proportion of the population at plating than in the other assays. Thus, under a model of random plasmid segregation, the resulting numerical dominance of “old” (and fully segregated) mutants compensates for the longer phenotypic lag.

In view of their relative impacts, we therefore neglected phenotypic lag but corrected for the copy number of pRN1 derivatives (as described above) and the higher growth rates measured for Lac⁻ mutants, as follows. The effect of differential growth of parental versus mutant cells on mutant frequency in a population had been determined previously by Koch and Drake (21) in terms of a parameter s, the selection coefficient. We calculated s for the pJlacS strain (see Materials and Methods) and substituted it, along with independently determined N and f values for each culture, into the mutation rate expression derived by Koch and Drake (21) (Table 2). These calculations, accounting for both relative growth rates and plasmid copy number, yielded a mutation rate of 5.1 × 10⁻⁴ mutational events per gene replication for lacS in pJlacS, i.e., about 1,500-fold higher than the corresponding rate previously measured for the chromosomal pyrEF locus (3). Thus, the growth advantage of mutants over the parental strain, though influential, merely exaggerated an underlying genetic instability of lacS in this plasmid when replicating in S. acidocaldarius.

Table 2.

Mutation parameters of lacS in five contexts

Contexta	km/kwb	sb	Nc	Mutant frequency		Uncorrected mutation ratef		s-corrected mutation rateh
				Mediand	Poissone	Median (relg)	Poisson (relg)	Median (relg)	Poisson (relg)
pJlacS	1.231	−0.181	4.1 × 109	0.077		1.3 × 10−3 (460)		2.6 × 10−4 (175)
pClacS/xyl	1.007	−0.0053	1.0 × 1034		7.7 × 10−4		2.8 × 10−6 (1)		2.9 × 10−6 (1)
pClacS/mal	1.024	−0.0175	9.4 × 1025	0.0042	0.0051	2.4 × 10−5 (8.6)	2.3 × 10−5 (8.4)	9.3 × 10−5 (32)	8.5 × 10−5 (29)
S. acidocaldarius chromosome	1.195	−0.160	8.1 × 1030		2.2 × 10−4		3.6 × 10−6 (1.3)		7.2 × 10−11 (9.8 × 10−5)
S. islandicus chromosome	1.034	−0.0316	8.6 × 1030		0.0022		3.5 × 10−5 (12)		7.5 × 10−6 (2.6)

^aData are for mutation of the same reporter gene in a pRN1 construct, on the S. acidocaldarius chromosome, or in the native location of the S. islandicus chromosome. The strain bearing pClacS was grown in medium containing 0.2% d-xylose or 0.2% maltose to give low or high levels of expression, respectively (see Table 5).

^bGrowth advantage measured under assay conditions, expressed as the ratio of specific growth rate constants and the selection coefficient; a negative s value means that the mutant strain grows faster than the parental strain (see Materials and Methods).

^cCumulative final size of all cultures in the set.

^dMedian mutant frequency of the set.

^eMutant frequency calculated from the P₀ term of the Poisson distribution of Lac⁻ clones among sampled cultures (see Materials and Methods).

^fCalculated by solving μ = f/ln(Nμ), iteratively and normalizing for copy number of the lacS gene.

^gRelative to pClacS during growth on xylose, which is the condition giving the lowest mutation rate among the five tested.

^hCalculated by the equation μ = sf/[1 − e^{−sln(N/N₀)}] and normalizing for copy number of the lacS gene.

Molecular nature of the lacS mutations.

The extremely high rates of lacS mutation suggested that DNA replication or repair of this region of pJlacS is somehow perturbed and that the molecular properties of the mutations may give insight into the molecular nature of the perturbation. We sampled the lacS mutational spectrum by sequencing the lacS genes of 72 independent, randomly picked Lac⁻ mutants. As shown in Fig. 1A, the mutations were concentrated in the 5′ portion of lacS and encompassed various molecular changes, including large deletions, slipped-strand events, and base pair substitutions (BPSs). As shown in Table 3, the latter two classes of mutations were also associated with particular motifs within the target. For example, short repeats generated insertion/deletion (“indel”) hot spots in the pJlacS spectrum. At one, a pentanucleotide is repeated three times, and one copy of this repeat was deleted in 16 (22%) of the 72 lacS mutants; at another hot spot (positions 212 to 217), a tract of six As was expanded or contracted by 1 bp in 15 mutants (21%) (Fig. 1A; Table 3). This behavior mimics that of spontaneous mutations in both the S. solfataricus and S. acidocaldarius pyrE genes, where similar motifs occur and account for similar slipped-strand events in the corresponding mutational spectra (3, 22, 23), albeit at much lower rates.

Fig 1.

Spectra of spontaneous lacS mutations The identities of independent spontaneous mutations of the S. solfataricus lacS coding region of plasmids propagating in S. acidocaldarius strain MR31 are indicated. Base pair substitutions are shown by placement of an A, T, C, or G directly above a base. Deletion mutations (>5 bp) are represented by an underline or overline, with the number of bases deleted indicated at the beginning of each mutation. Tandem duplications are indicated by gray italic text. A “+” or a “△” sign above a mononucleotide run denotes a single-base insertion or deletion, respectively, of the base below it. Positions within homonucleotide runs are genetically equivalent, and therefore the positions of symbols above a run are arbitrary. Bracketed text indicates a deletion mutation occurring in a single mutant. Single-base-pair insertions not located at mononucleotide runs are shown by a ‘v’ sign with the insertion above it. (A) Seventy-two independent mutations in pJlacS cultures grown on xylose. (B) Thirty-five independent mutations in pClacS cultures grown on maltose (inducing conditions).

Table 3.

Comparison of spontaneous lacS mutations and Sulfolobus genomic pyrE mutations^a

Class of mutations	Fractionb of total mutations in:
	pJlacS	Induced pClacS	Genomic pyrEc
BPS	0.39	0.37	0.26
Transitions	0.38	0.34	0.19
Transversions	0.01	0.03	0.07
Frameshifts	0.24	0.28	0.55
+1 bp	0.13	0.17	0.36
−1 bp	0.08	0.11	0.19
+2 bp	<0.01	<0.03	<0.006
−2 bp	0.03	<0.03	<0.006
Other indels	0.38	0.34	0.19
Expansions or contractions of short repeats	0.22	0.31	0.09
Deletions > 5 bp	0.14	0.03	0.02
Duplications > 5 bp	0.02	<0.03	0.09

^aBased on the following numbers of independent mutants: pJlacS, 72; induced pClacS, 35; chromosomal pyrE, 161.

^bValues with “<” represent mutations not observed in the indicated sample.

^cThe spontaneous genomic pyrE mutations were analyzed previously and occurred in the native S. acidocaldarius gene and in S. solfataricus pyrE sequences integrated into the S. acidocaldarius chromosome (21).

Deletions were also prominent among the lacS mutations, occurring in 10 (14%) of the 72 mutants analyzed. Deletions with similar molecular properties (i.e., ranging from about 20 to 300 bp and lacking direct or inverted repeats associated with endpoints) were seen previously in chromosomal pyrE genes, but they comprised only about 2% of the independent pyrE mutations analyzed (Table 3). Thus, the lacS deletions seemed to be more frequent than the deletions seen in other Sulfolobus mutation spectra. However, despite the frameshift hot spots and high proportion of deletions, BPSs still comprised the largest class of lacS mutations (Table 3). About 96% of these (27/28) were G·C-to-A·T transitions, while the only transversion in the set also occurred at G·C (forming T·A). It was also striking to note that all 28 BPSs created chain termination (stop) codons, yet the 11 corresponding sites represented all three codon positions and produced all three stop codons (UAG, UAA, and UGA).

To examine whether the observed lacS mutations may have formed as part of a “shower” of coincident mutations, perhaps reflecting a transient, error-prone state of the cell or plasmid (24), we sequenced two nonessential regions of 36 Lac⁻ pJlacS clones encompassing 2.8 kbp between lacS and pyrEF and 2.6 kbp between lacS and ORF904 (see Fig. S1 in the supplemental material). No mutations were found in these regions of any plasmid, indicating that lacS mutations were not frequently accompanied by mutations elsewhere. To evaluate the mutagenic potential of the DNA primase/polymerase encoded by ORF904, which is essential for replication of pRN1 constructs, we compared the positions of BPSs with the priming sites identified in vitro, GTG/CAC (11). This trinucleotide occurs at 12 sites in the first half of lacS, but BPSs that inactivated it occurred at only two (CAC at positions 211 and 286 of the sense strand). These two sites nevertheless accounted for 43% (11/28) of the BPSs in the set of Lac⁻ mutants and thus contributed significantly to the mutation spectrum.

Factors affecting the lacS mutation rate.

To shed light on the basis of mutagenesis, we evaluated various aspects of pJlacS or the host strain for their contribution to the observed mutagenesis of lacS. The approach involved analyzing the lacS gene in four additional contexts: (i) in its native location on the S. islandicus REN1H1 chromosome, (ii) inserted between the pyrE and pyrF genes on the S. acidocaldarius chromosome, (iii) in another pRN1 construct that expressed it at high levels from an inducible promoter, and (iv) in that construct at a lower level of gene transcription. These contexts thereby tested the relevance of several variables, including gene location, native versus heterologous host species, plasmid versus chromosomal replication, and level of gene expression. As was done for the pJlacS strain, spontaneous mutations were identified by monitoring multiple, independent cell lines propagated over many generations. Only one Lac⁻ mutant from a given cell line was retained for sequence analysis, and several mutants were used in competition experiments to measure the growth advantage, if any, of Lac⁻ mutants, as had been done in the analyses of plasmid pJlacS (see Materials and Methods).

The competition assays revealed that fitness effects of lacS mutations, indicated by the s or k^m/k^w values, varied considerably across these various situations (Table 2), even when the lacS allele was the same (Table 1; see Table S2 in the supplemental material). Furthermore, compared to plasmid pJlacS, the lacS gene placed in the other contexts generally yielded lower β-d-glycosidase levels and fewer Lac⁻ mutants. In several cases, mutants were never detected in a large proportion of the mutant accumulation cultures, despite allowing many more generations of growth for these strains than for the pJlacS cultures (see Table S1 in the supplemental material).

The differences among the five conditions, summarized in Table 2, show that pJlacS promoted a much higher rate of spontaneous lacS mutations than any other context tested. Taking the s-corrected estimates as generally most reliable, lacS mutated at least five times more frequently in pJlacS than in the next most mutagenic setting, maltose-induced pClacS (Table 2). Second, the s-corrected mutation rates increased with the copy-number-normalized β-galactosidase levels across all the various contexts evaluated (Table 4), suggesting an association between gene expression and lacS mutation. Third, the host cell, chromosomal location, or a factor correlated with these variables influenced the level of mutagenesis, as suggested by the difference in rates between single-copy lacS in its native chromosomal position in the pRN1 host S. islandicus REN1H1 versus integrated artificially into the S. acidocaldarius chromosome. In particular, the higher genetic stability of single-copy lacS in S. acidocaldarius than in S. islandicus (Table 2) argued that the frequent mutations observed in pJlacS are not due to an inability of S. acidocaldarius to replicate heterologous (pRN1-derived) constructs or the lacS sequence accurately.

Table 4.

Comparison of gene expression levels and spontaneous mutation rates^a

lacS context	Enzyme level normalized to:		Mutation rate (rel)	Mutation/expression ratio (rel)
	Cells	Gene copies (rel)
pClacS/xyl	0.74	0.185 (1)	2.91 × 10−6 (1)	1.6 × 10−5 (1)
S. islandicus chromosome	0.20	0.20 (1.08)	7.50 × 10−6 (2.58)	3.8 × 10−5 (2.4)
pClacS/mal	3.51	0.88 (4.74)	9.34 × 10−5 (32.1)	1.1 × 10−4 (6.8)
pJlacS	11.4	2.85 (15.4)	5.10 × 10−4 (175)	1.8 × 10−4 (11.4)

^aThe selection-corrected mutation rate for the S. acidocaldarius chromosome (Table 2) was considered too low to be reliable for this comparison.

As done previously for pJlacS, we sampled the spectrum of lacS mutations arising in plasmid pClacS under inducing conditions by sequencing 35 independent Lac⁻ mutants from maltose-supplemented cultures (Fig. 1B). Comparison to the corresponding mutations of pJlacS indicated similar proportions of the major categories of mutations (Table 3), which was supported by χ² analysis (P = 0.6491 for BPSs versus frameshifts, P = 0.9287 for BPSs versus indels, and P = 0.5953 for indels versus frameshifts). In addition, hotspots for slipped-strand events and G·C-to-A·T transitions occurred at the same sites under both conditions, and both yielded a strong preference for nonsense mutations among BPSs (12/13) (Fig. 1). Apparent differences in the frequency of G·C-to-A·T transitions at lacS bp 453 (Table 5) or large deletions (Fig. 1) were not significant by χ² analysis.

Table 5.

Specificity of base pair substitutions in plasmid-borne lacS

Plasmid and positiona	BPS	Codon		Coding change	No. of mutants
		Initial	Final
pJlacS
40	C to T	CAG	TAG	Gln to amber	1
52	C to T	CAA	TAA	Gln to ochre	1
107	G to A	TGG	TAG	Trp to amber	3
211	C to T	CAA	TAA	Gln to ochre	3
286	C to T	CAA	TAA	Gln to ochre	8
429	C to A	TAC	TAA	Tyr to ochre	1
453	G to A	TGG	TGA	Trp to opal	6
468	G to A	TGG	TGA	Trp to opal	1
521	G to A	TGG	TAG	Trp to amber	1
522	G to A	TGG	TGA	Trp to opal	2
721	C to T	CAA	TAA	Gln to ochre	1
pClacS/mal
32	G to A	GGT	GAT	Gln to Asp	1
34	C to T	CAA	TAA	Trp to ochre	1
98	G to A	TGG	TAG	Trp to amber	1
107	G to A	TGG	TGA	Trp to opal	1
179	G to A	TGG	TAG	Trp to amber	1
211	C to T	CAA	TAA	Gln to ochre	1
286	C to T	CAA	TAA	Gln to ochre	4
453	G to A	TGG	TGA	Trp to opal	1
631	G to T	GGA	TGA	Gly to opal	1
721	C to T	CAA	TAA	Gln to ochre	1

^aPosition relative to the first nucleotide in the nontranscribed (i.e., sense) strand.

DISCUSSION

An archaeal model of conditional, endogenous mutagenesis.

The molecular strategies used by hyperthermophilic archaea to preserve genome integrity have interesting implications for the diversity and evolution of DNA repair mechanisms but remain largely undefined. The present study reveals, for the first time, extremely high error rates for replication of a plasmid reporter gene in a hyperthermophilic archaeon, despite the host's apparently accurate replication of its chromosome (3). The effects of certain genetic and cellular factors on the observed mutation rate, as well as the molecular nature of the mutations, support several conclusions but also raise certain questions that remain to be addressed by future study.

Our results establish that spontaneous mutations inactivate the lacS gene of plasmid pJlacS at an extremely high rate in S. acidocaldarius. Even after correcting for plasmid copy number and differential growth of the mutant form, the lacS mutation rate is more than 300 times higher in pJlacS than in the S. acidocaldarius chromosome and about 1,500 times higher than for the chromosomal pyrE and pyrF genes. These results indicate that some aspect of lacS or its plasmid context impairs the fidelity of replicating this gene, raising the possibility that the basis of this phenomenon, once identified, could be harnessed to manipulate genetic stability experimentally in hyperthermophilic archaea.

Second, our results argue that residual DNA primers contribute little to the mutagenesis observed in pRN1 constructs. Only a minority of lacS mutations recovered in pJlacS and pClacS occurred near GTG/CAC sites; thus, most of the potential priming sites did not participate in mutagenesis. Also, the rates of mutation at the GTG/CAC sites appeared to vary in parallel with those of two other classes of mutations, indicating a conditional nature not predicted for persistent DNA primers. Although their pattern of parallel variation suggests a common or related origin, the three predominant classes of mutation seem mechanistically dissimilar; an alternative explanation for the formation of the BPSs that does not involve DNA priming is discussed below.

Third, the mutagenesis we observed appears to result in part from some aspect of gene expression, as indicated by a positive relationship between the relative gene-normalized enzyme level and the mutation rate across the conditions we varied (Table 4). The relationship is not linear, however, as the ratio of mutation rate to gene-normalized enzyme level generally increases with enzyme level (Table 4). Although the available data do not define the basis of this nonlinearity, they remain consistent with the hypothesis that mechanisms which normally suppress mutagenesis become increasingly overwhelmed as gene expression exceeds a certain threshold value.

Fourth, the results also demonstrate that the β-d-glycosidase of S. solfataricus can retard growth of S. acidocaldarius. Although its basis remains to be identified, this retardation varies among Sulfolobus strains and can be rather large. Because mutant accumulation assays are intrinsically sensitive to differential fitness of parent and mutants, this highlights the importance of measuring the growth advantage for each strain and applying appropriate corrections to the mutation rate calculations.

Mechanistic implications.

The endogenous, conditional mutagensis of lacS that we observed resembles various examples of transcription-associated mutagenesis (TAM) documented in bacteria and eukaryotic cells (25). The effects seen in S. acidocaldarius are as large as TAM effects reported for E. coli (26–28) or yeast (29). Specifically, maltose-induced pClacS versus noninduced pClacS, or pJlacS versus maltose-induced-pClacS, represented 5-fold increases in enzyme level and 2-fold increases in mutation rate, whereas pJlacS versus noninduced pClacS represented a 15-fold increase in enzyme level and more than a 100-fold increase in mutation rate. For comparison, a 4-fold increase in E. coli leuB gene transcription increased mutation 7-fold (27), and a 50-fold increase in Saccharomyces cerevisiae LYS2 gene transcription increased mutation 10-fold (29).

Mechanistically, TAM is thought to result primarily from single-stranded DNA (ssDNA), which is prone to base deamination, oxidation, and the formation of replication-perturbing secondary structures. During DNA replication, these processes can promote transitions, transversions, and frameshifts, respectively (28, 30, 31). Another mechanism potentially contributing to TAM in bacteria and eukaryotes is transcription-coupled repair (TCR), in which the RNA polymerase complex directs nucleotide excision repair to DNA lesions on the transcribed strand, leaving similar lesions in the nontranscribed strand unrepaired, where they promote mutation (32, 33). However, most of the TAM evident in E. coli genomes has been attributed to a bias in deamination of dC rather than a bias in TCR-promoted dU repair (34). Furthermore, Sulfolobus appears to lack any functional TCR mechanism, as UV photoproducts are removed at the same rate from both strands of several genes in S. solfataricus (35).

In addition to generating these hypotheses, our data also raise certain questions which remain unresolved. For example, we note that the extremely frequent lacS mutations in pJlacS, though not especially diverse compared to other mutation spectra, nevertheless seem to represent three distinct mechanisms of formation. Nearly all the BPSs recovered are G·C-to-A·T transitions, which suggests deamination of dC to dU (36). In contrast, the molecular nature of the frameshift mutations implicates a distinct process, i.e., strand slippage during replication stabilized by short repeats (37), whereas the large deletions imply a mechanism that causes replication to skip relatively long intervals unassisted by sequence repeats (Fig. 1) (38). In this context, the similarity between the spontaneous deletions within lacS and those within the chromosomal pyrE gene (38) seems remarkable, given that the lacS deletions were about 10,000 times more frequent (Table 3).

We note further that the BPSs share three distinctive properties: (i) all occurred within the first half of the lacS coding sequence, (ii) nearly all created a chain-terminating (stop) codon, and (iii) nearly all were G·C-to-A·T transitions. Each of these biases has implications for the formation or detection of BPSs in this system. For example, all 41 of the BPSs we recovered from plasmid-borne lacS (28 from pJlacS and 13 from pClacS) map to 16 sites in the first 241 of the 490 lacS codons (Table 5) and thus affect only the amino-terminal half of the 60-kDa protein. Although the available data remain limited, three observations suggest that this positional bias primarily reflects formation of the mutations rather than their recovery. The first concerns the fact that the 5′ and 3′ halves of lacS have similar numbers of potential mutation sites (15 and 11, respectively), which we identify as G·C base pairs capable of generating nonsense mutations by transition to A·T (see below). The second comes from a study of spontaneous pyrE and pyrF mutations selected by 5-fluoroorotic acid (FOA); about 90% of these mutations mapped to the promoter-proximal pyrE gene, despite the similar base composition, mutation-prone motifs, and mutant phenotype of the adjacent pyrF gene (3). The behavior of both these mutational targets therefore remains consistent with a process that promotes spontaneous mutation but extends only a limited distance from the promoter. The third observation is that the β-d-glycosidase subunit encoded by lacS does not have functionally distinct N- and C-terminal domains. Instead, both halves of the polypeptide contribute about equally to the subunit's 8(β-α) barrel structure and the loops which define subunit interactions and the active site. Consistent with this, the carboxyl terminus has at least one catalytically essential residue, Glu389 (39). Thus, it remains difficult to postulate a priori the basis of a selective filter that prevents BPSs in the 3′ half of lacS from being recovered in our mutation assays.

A second bias concerns the fact that nearly 98% (40/41) of the BPSs obtained from pJlacS and maltose-grown pClacS create stop (nonsense) codons in lacS. On balance, this bias seems likely to result primarily from a selective process in which nonsense mutations are recovered preferentially. The other two major classes of lacS mutation recovered under these conditions are frameshifts and large deletions; both of these classes should severely disrupt the protein structure and abolish activity. Among BPSs, however, only nonsense mutations would consistently match this severe impact on enzyme function. Perhaps more significantly, the BPSs occur in all three codon positions and create all three nonsense codons. The pattern thus argues that chain termination, rather than a particular sequence change, represents the critical feature of the BPSs that we recovered.

The third bias concerns the fact that 95% of the BPSs (39/41) are G·C-to-A·T transitions. This has conceptually simple mechanistic explanations, the most obvious of which is spontaneous deamination of dC residues to form dU, which is predicted to have several consequences in hyperthermophilic archaea. DNA polymerases of these organisms stall specifically 4 nucleotides (nt) ahead of template dU, but assays in vitro indicate that this lesion is eventually bypassed with insertion of dA (40, 41). Alternatively, if dU becomes excised by a corresponding DNA N-glycosylase but is not repaired before replication, dA is likely to be inserted opposite the resulting noninstructional abasic site, based on assays performed in vitro (42, 43) and in vivo (D. W. Grogan and C. J. Sakofsky, unpublished data). Accordingly, either route leads to a G·C-to-A·T transition.

To determine whether a requirement for nonsense mutations would impose a corresponding enrichment of G·C-to-A·T transitions, we identified the 260 BPSs within lacS that can generate a stop codon by one BPS and found that G·C-to-A·T transitions represent only 43 (16.5%) of these. Thus, the transitions that dominated our BPSs represent neither the only route to stop codons in lacS nor the most probable one. We also note, however, that no A·T-to-G·C transition creates a stop codon; as a result, the possibility that A·T-to-G·C transitions also occurred but went undetected because of a selection bias that required formation of a nonsense mutation cannot be excluded (see above).

On balance, therefore, the weight of available data suggests a working hypothesis in which some aspect of transcribing a plasmid-borne gene in Sulfolobus promotes dC deamination in the first 600 to 700 bp of the transcriptional unit. These features have precedent in eukaryotic or bacterial TAM. For example, an excess of BPSs in transcribed regions of E. coli genomes is attributed to dC deamination in the nontranscribed strand, not to transcription-coupled repair of the transcribed strand (34). In the human genome, transcription induces mutation via dC deamination, but only in the first 1 to 2 kbp of the transcriptional units (44), and a similar localized, transcription-dependent dC deamination occurs in somatic hypermutation of immunoglobulin genes (36). It must be noted, however, that the BPSs we recovered in lacS do not exhibit the strand bias seen in bacterial and eukaryotic TAM. In lacS, the dC partner of the mutated G·C occurred about as often on the transcribed strand as on the nontranscribed strand (Fig. 1), whereas in bacterial and eukaryotic TAM, these dC residues occur predominantly on the nontranscribed strand (34, 44). A role of dC deamination in promoting the repeat expansion/contractions and deletions in Sulfolobus also does not seem obvious, yet these mutations seem to covary in parallel with G·C-to-A·T transitions among the conditions we tested and thus may share an origin with the BPSs. We note in this regard that formation of dU could promote slipped-strand events and deletions indirectly as a consequence, for example, of the steps needed to restart replication forks stalled at these lesions (45).

Finally, we note that if nonsense mutations were indeed the only BPSs that were detected with high efficiency under our conditions (as we propose), then our data underestimate the true rate of mutagenesis of plasmid-borne lacS. Only about 21% (120/582) of the G·C base pairs in lacS can generate a stop codon via BPS, and of those, only about 36% (43/120) result from G·C-to-A·T transitions. A requirement for chain termination accordingly predicts a maximal detection efficiency for G·C-to-A·T transitions in lacS of about 7%. Applying this efficiency to the growth rate-corrected mutation rate in pJlacS yields a predicted (0.39/0.074)(2.6 × 10⁻⁴) = 1.4 × 10⁻³ total BPS events in lacS per plasmid replication.