The effect of supercoil and temperature on the recognition of palindromic and non-palindromic regions in phi X174 replicative form DNA by S1 and Bal31.

The effect of supercoil and temperature on the topology of phi X174 replicative form (RF) DNA was studied using single-strand specific endonucleases S1 and Bal31 as probes for cruciform extrusion and other structural perturbations of the B-helix. Both enzymes were found to recognize specifically and reproducibly over 30 sites, most of which were cleaved by both enzymes independent of the superhelicity of the genome. A negative superhelical density exceeding 0.06 stabilized a transition in the DNA conformation that generated several new cleavage sites for Bal31. The underlying structures appeared to be only transiently stable and were lost from in vitro supercoiled DNA during brief incubation at 65 degrees C. They were generally absent from in vivo supercoiled RF DNA of equal superhelicity as a consequence of the extraction and storage procedure. Mapping of the cleavage sites suggested that they were preferentially located near the beginnings and ends of genes and that the structural basis for at least some of them was the extrusion of relatively small palindromes into the cruciform state. Insertion of a short synthetic palindromic sequence into the phi X174 genome generated a supercoil-dependent, temperature-sensitive secondary structure that was cleaved in the Bal31 but not the S1 reaction, further supporting the hypothesis that even small cruciforms with stem size of 7 or less base pairs may be transiently stable. Subjecting supercoiled RF DNA to the typical S1 reaction conditions induced a topological shift that diminished all but one of the supercoil-induced Bal31 recognition sites and promoted the formation of one major new site.

The Effect of Supercoil and Temperature on the Recognition of Palindromic and Non-palindromic Regions in 4x1 74 Replicative Form DNA by S1 and BuZ31* (Received for publication, July 31, 1986) Uwe R. Muller and Carole L. Wilson The effect of supercoil and temperature on the topology of 4x174 replicative form (RF) DNA was studied using single-strand specific endonucleases S1 and Ba131 as probes for cruciform extrusion and other structural perturbations of the B-helix. Both enzymes were found to recognize specifically and reproducibly over 30 sites, most of which were cleaved by both enzymes independent of the superhelicity of the genome. A negative superhelical density exceeding 0.06 stabilized a transition in the DNA conformation that generated several new cleavage sites for Ba131. The underlying structures appeared to be only transiently stable and were lost from in vitro supercoiled DNA during brief incubation at 65 "C. They were generally absent from in vivo supercoiled RF DNA of equal superhelicity as a consequence of the extraction and storage procedure. Mapping of the cleavage sites suggested that they were preferentially located near the beginnings and ends of genes and that the structural basis for at least some of them was the extrusion of relatively small palindromes into the cruciform state.
Insertion of a short synthetic palindromic sequence into the 4x174 genome generated a supercoil-dependent, temperature-sensitive secondary structure that was cleaved in the Ba131 but not the S1 reaction, further supporting the hypothesis that even small cruciforms with stem size of 7 or less base pairs may be transiently stable.
Subjecting supercoiled RF DNA to the typical S1 reaction conditions induced a topological shift that diminished all but one of the supercoil-induced Bd31 recognition sites and promoted the formation of one major new site.
Physical (1,2), chemical (3-6), and enzymatic (7-10) probes of DNA secondary structure have provided evidence that deviations from the regular B-helix exist in prokaryotic and eukaryotic genomes and have lent credence to the speculations that such conformations fulfill significant roles in the regulation of gene expression (11). There appear to be three types of non-right-handed secondary structures: a variety of left-handed helices (4, 12, 13), cruciforms (14), and loop-out regions which are formed for reasons other than breathing (see Ref. 24 for additional references). Since formation of all of these alternate structures is energetically disfavored, they form generally only under the torsional stress of supercoiling (16)(17)(18), and even then their formation may be kinetically hindered (19,20). In addition, the pH and ionic environment have been found to be crucial factors that influence these structural transitions (9,13,21,22). Once these structures are formed, the unpaired bases at their center and/or at their junctions with the regular B-helix can be specifically cleaved with a variety of single-strand specific nucleases (4,13,(23)(24)(25). Nuclease S1 has been used most extensively (4,13,(26)(27)(28), but several other enzymes with similar properties have also been employed (9,18,29), giving considerable insight into the variety of structures that may coexist or compete in natural DNA molecules. Yet, the results obtained sometimes varied between different laboratories depending on the nuclease probe used (9,14), the reaction conditions employed (23,29), or for no obvious reason (7, 30), making further investigation of these probes and the structures they recognize desirable.
We have used the biophysically and genetically well-characterized bacteriophage 4x174 genome to study the distribution of cruciforms and other nuclease-sensitive sites in an attempt to correlate structure with function. For this and other genomes we have previously noted a statistical overrepresentation, specifically in their intercistronic regions, of perfectly base-paired hairpins with loop sizes between 3 and 20 bases and stem sizes of 2 6 base-pairs (31-33). This, we speculated, may imply an evolutionary selection and thus a function. For most genomes analyzed the largest of these theoretical structures had a stem size of 8 or 9 base-pairs, while most structures were shorter (32). Experimental data (21) and theoretical considerations (18,34) would thereby eliminate all but a few of these as candidates for cruciform formation at physiological superhelical densities, and secondary structure analysis of several of these and other genomes have supported this view (14,24,26,35). On the other hand, a recent analysis of supercoiled pBR322 DNA suggested cleavage by Mung bean nuclease of a potential hairpin with 6 bp' in the stem (9), which had not been recognized by S l (14). This allows for the possibility that at least transiently very small cruciforms may be extruded in supercoiled genomes but are generally not recognized because they escape detection by the techniques employed.
We have sought to test this hypothesis by carefully analyzing the topology of supercoiled 4x174 RF DNA, and that of a mutant strain with a synthetic palindromic insert, using two different endonuclease probes. We show herein that small cruciforms with 7 and possibly less bp in the stem can apparently be extruded since they are recognized specifically by Ba131, but not S1. However, these structures are very unstable and are lost due to a topological shift in the RF DNA that can be induced by heat or by the special conditions of the S1 reaction. In addition to these supercoil-dependent sites, a variety of other specific cleavage sites exist that are present even in relaxed DNA and are shared by both enzymes.

MATERIALS AND METHODS
Bacterial and Phuge Strains-Bacteriophage 4x174 strain am3 is referred to herein as wild type. Strain de13 (4x174 J-F de13) has a deletion of 27 bp in the J-F intercistronic region (position 979 on the standard 4 X map (36)), and ins6 (4x174 J-F ins6) has an insertion of 117 bp at the same position. Both strains have been described previously (37, 38). Strain ins6430 was constructed by inserting the palindromic BamHI linker sequence CCGGATCCGG into the PvuII site of strain ins6 (39). Escherichia coli HF4714 was the permissive host (40), and E. coli C was the nonpermissive host.
Enzymes-Calf thymus topoisomerase I and nucleases S1 and Ba131 were purchased from Bethesda Research Laboratories (BRL), T4 polynucleotide kinase was from Pharmacia P-L Biochemicals or U. S. Biochemical Corporation, and calf intestinal alkaline phosphatase was from Boehringer-Mannheim. Restriction enzymes were obtained from either International Biotechnologies Inc., New England Biolabs, or BRL and used in buffers supplied by the manufacturer or made according to the manufacturer's specifications.
Isolation and Purification ofRFZ DNA-E. coli C cells were infected with phage at a multiplicity of infection of 10, and chloramphenicol (40 pg/ml) was added 7 min after infection to prevent the shift to single-stranded DNA synthesis and thereby increase the yield of RF DNA. Infected cells were incubated for 2 h before harvesting. RF DNA was purified by two CsCl gradients in the presence of ethidium bromide by standard procedures and was generally 295% in the supercoiled RFI form. After final dialysis against Tris-HC1 (10 mM, pH 7.8,O.l mM EDTA) DNAs were stored at 4 "C for several months without significant loss of supercoiling.
Zn Vitro Supercoiling of RF DNA-The procedure of Singleton and Wells (41) was followed exactly, with the exception that we used topoisomerase from calf thymus instead of wheat germ.
Linearization with BaZ.31 and SI-In a typical reaction 1 pg of DNA in 20-30 p1 of buffer (12 mM MgCl,, 12 mM CaCl,, 1 mM EDTA, 20 mM Tris-HC1, pH 8.1) was treated with Ba131 (0.1 unit/pg) for 15 min in high-salt buffer (600 mM NaCl) at 15 'C or with 0.001 unit/ pg for 15-20 min in low-salt buffer (50 mM NaC1) at temperatures ranging from 7 to 30 'C (usually 15 "C). Units of Ba131 were as defined by BRL. This amount of enzyme was sufficient to linearize between 70 and 90% of the RF DNA, but no efforts were made to determine whether and to what extent nucleotides were removed from either end of the linearized molecules. For linearization with endonuclease S1, 1 pg of RFI DNA in 20-30 pl of buffer (50 mM NaC1,30 mM sodium acetate, pH 4.6, 1 mM ZnS04, and 5% glycerol) was incubated with 10 units of S1 (units according to Vogt (42)) for 30 min to 2.5 h, depending on the particular DNA preparation used. Following inactivation of the nuclease by phenol extraction and ethanol precipitation, DNAs were cleaved with either AuaI or AuaII, using an excess of enzyme to ensure complete digestion.
Dephosphorylation and End-labeling-DNA fragments were treated with 0.2-0.4 units of calf intestinal alkaline phosphatase in either AuaI or AuaII restriction buffer. The phosphatase was administered in 2 separate aliquots with each reaction proceeding for 30 min at 37 "C. DNAs were phenol-extracted once, ether-extracted three times, dried, and resuspended in HZ0 to restore the original volume. The fragments were end-labeled with T4 polynucleotide kinase and [32P]ATP according to Maniatis et al. (43), with the following modifications. Dephosphorylated DNAs (0.5-2 pg) were incubated with 5 units of kinase and 10 pCi of [32P]ATP (approximately 3000 Ci/mmol; New England Nuclear), for 40 min at 37 "C in restriction buffer containing 15 mM dithiothreitol and 0.1 mM spermidine.
Gel Electrophoresis and Autoradiography-Fragments generated by nuclease and restriction enzyme treatment were separated on horizontal 1.4% agarose (Sigma) slab gels (40.18.0.5 cm) by electrophoresing for approximately 44 h at 60 V in 40 mM Tris-acetate, pH 8.2, 5 mM sodium acetate, and 1.15 mM EDTA. Unlabeled fragments were visualized under uv light after staining with ethidium bromide. Gels containing end-labeled fragments were briefly soaked in 7% trichloroacetic acid and dried under vacuum at 80 "C. Autoradiography was for 2-16 h using Kodak XAR-2 film. For band counting in the topoisomerase studies, the gels were photographed with Polaroid type-55 film, and the negatives were scanned with an LKB Laser Densitometer linked to an IBM 9000 computer. The fragment sizes were determined by computer (Sizer Program of BIONET) after measuring their migrational distance from the origin on the autoradiograph.

RESULTS
Probing the Topology of Supercoiled 4x1 74 RF DNA with Single-strand-specific Nucleuses Bal31 and SI -Previous analyses of the distribution of cruciforms in the supercoiled 6x174 RF DNA using S1 nuclease have led to the conclusion that there were either no (7), or only a single major S1 cleavage site identifiable, which coincided with the 28 base-pair long palindrome in the F-G intercistronic region of this genome (14, 30). Since single-strand-specific endonuclease Ba131 was reported to cleave supercoiled ColEl DNA under more physiological conditions but with similar specificity as S1 (18), we have used this enzyme in addition to S1 to probe for the presence of cruciforms and other non-right-handed secondary structures in supercoiled wild type 4x174 RF DNA. The strategy employed was to treat supercoiled DNA with these nucleases under conditions where most of the DNA was converted from the supercoiled to the linear form. The nucleases were then removed by phenol extraction to avoid "nibbling" at the ends of the linear molecules, and the number of S1or Bal31-sensitive sites was determined by introducing into each molecule a second but unique cut with restriction endonuclease AuaII. Hence, each Sl/Bal31 cleavage site should be represented by two unique fragments, one larger than half-genome size, the other smaller, unless the cleavage occurred precisely opposite from the AuaII site. Since AuaII leaves 5"protruding single-stranded ends, these fragments could be efficiently end-labeled with [32P]ATP. Fig. 1 shows an autoradiograph of Ba131-and S1-generated fragments separated on high resolution agarose gels. The typical products of a Ba131 reaction carried out under our standard conditions were run in lane A. Approximately 30 bands with different intensities can be identified, most of which have been assigned a number that has been used consistently throughout this paper. In the original autoradiograph the bands are very sharp and their intensity and position is reproducible between different experiments and DNA preparations. Thus, Ba131 recognizes specifically, but with different frequency, at least 30 sites in in vivo supercoiled 6x174 RF DNA.
An S1 analysis of this DNA is shown in lane D. In general the same bands were generated as with Ba131 but with two significant differences. Band 22, a minor band only in the Ba131 digest (and generally absent), appeared as the most predominant band. Mapping of the corresponding cleavage site shows it to correspond to the "major cruciform," as identified by Lilley (14,30), which is located almost exactly half a genome length away from the AuaII site. S1 cleavage at that position generates, therefore, two fragments of almost equal size, which comigrate in the same position on this gel. The identity of this cleavage site was confirmed as discussed below. The second major cleavage site (6a) was also only a minor site in the Ba131 digest and will be discussed later.
While the Bal31 pattern was obtained reproducibly with different DNA preparations, the S1 reaction varied, but only with regard to the major band. This is shown in lane C, where the same S1 reaction was carried out with a different RF DNA preparation. With the exception of band 22, the typical S1 pattern was obtained.
Since it is known that the formation of at least fairly large Cleavage of supercoiled 6x174 RF DNA with Bd31 and S1. 6x174 wild type RFI DNA (3 pg) was incubated at room temperature or 65 "C for 30-40 min, cooled slowly to 15 "C, and then treated either with 30 units of S1 nuclease (1.5-2.5 h at 15 "C in 60 pl of S1 buffer), or with 0.003 units of Bal31 (60 pl buffer containing 50 mM NaC1, 20 min at 15 "C). After removal of the nucleases with phenol and ether, the DNA samples were digested for 16 h with 3 units of AuaII (37 "C), end-labeled, and electrophoresed as described under "Materials and Methods." The gel was soaked in trichloroacetic acid, dried, and autoradiographed. Lane A (unheated, BaZ31-treated DNA), lane B (heated, Bal31-treated DNA), lune D (unheated, S1treated DNA), lane E (heated, S1-treated DNA), lune C (unheated, S1-treated DNA from a different preparation). The major bands in the Bal31 digest have been labeled 1-22. As the figure was assembled from three separate gels, the thin lines have been drawn to identify the same bands between different digests. The large arrow points at band 22. The marker lane contained a mixture of end-labeled fragments generated by separate digests of 4x174 RF DNA with BstNI, NarI, HpaII, and Sau96I (incomplete digest).
cruciforms may be kinetically hindered (19, 20), we tested whether kinetic barriers were responsible for the absence of the F-G cruciform in the DNA used in lane A. The RF DNA used in lanes A and D was therefore heated to 65 "C and then cooled to 15 "C before treatment with either Ba131 ( l a n e B ) or S1 (lane E ) under otherwise identical conditions. Virtually the identical banding pattern was obtained with either enzyme with the exception of band 22, which was strongly diminished. This suggests that the cleavage sites recognized here by both enzymes were insensitive to partial melting and renaturing of the DNA, except for the F-G cruciform, suggesting that it is thermodynamically the least stable structure recognized.
As will be shown below, the difference between the DNA preparations in lanes C and D was apparently a result of different superhelical densities to which extrusion of the F-G cruciform is exquisitely sensitive.

Extrusion of the F-G Cruciform Is Promoted by Low pH and
SI-The striking difference between the S1 and Ba131 reaction with regard to recognition of the F-G cruciform (as well as band 6a), but not the other cleavage sites, suggested a sensitivity of this structure to the special reaction conditions, i.e. the ionic strength or the low pH of the S1 buffer. Thus, the reactions were repeated by exchanging the reaction buffers and adjusting for the reduced efficiencies of the enzymes with increased enzyme concentrations. Fig. 2A shows that under physiological supercoil S1 recognizes the F-G palindrome only at low pH but not under the Ba131 reaction conditions. Ba131 on the other hand does not cleave this palindrome under either condition. This suggests that low pH as well as the action of S1 itself are required for recognition of the F-G cruciform. Varying the conditions of the Ba131 reaction to include a temperature range of 7-37 "C did not affect the number and type of fragments obtained (data not shown), but increasing the NaCl concentration from 50 to 600 mM changed the results somewhat (Fig. 2B). In general the same number and types of fragments were obtained, but band 12 became the major cleavage site, while it was only a minor band at the lower ionic strength. Such a change in cleavage specificity in response to increased ionic strength is not unexpected as it has been observed by Kowalski (23) for other single-strandspecific enzymes. Effect of Superhelical Density on the Bal31-specific Cleavage of 6x174 RF DNA-Extrusion of cruciforms is known to be supercoil-dependent (16,34,44). Hence, we have investigated the effect of changing the superhelical density on the specificity of Ba131 cleavage. The method of Singleton and Wells (41) was used to either decrease or increase the average superhelical density of purified covalently closed RF DNA preparations. This technique consists of treatment of RF DNA with topoisomerase I in the presence of ethidium bromide and subsequent removal of the enzyme and the dye with phenol. The resulting superhelical density is characteristic for the ethidium bromide concentration used. Using the electrophoresis conditions described in the "Materials and Methods" section, but with the addition of ethidium bromide (45) or chloroquine in the gel (46), individual topoisomers were resolved, and the mean superhelical density of a given DNA preparation was determined by band counting. Since the Ba131 or S1 reactions were carried out at 15 "C, but the gels were run at 37 "C, the temperature effect (47) and the relaxation effect of the gel itself were also determined in control experiments (not shown), and 6.5 supercoils were added to the measured mean number of supercoils. With this method the mean superhelical density of in vivo supercoiled 4x174 RF DNA, when stripped of all proteins, was generally found to be in the order of u = -0.07, but preparations with significantly lower values were periodically obtained.
The effect of superhelical density on the cleavage of 4x174 wild type RF DNA with endonuclease Ba131 is shown in Fig.  3. Scrutiny of this autoradiograph shows that most of the Ba131 bands observed in Fig. 1 were obtained even with completely relaxed RF DNA (lune 0) and continued to be present over the whole range of superhelical densities from u = 0 to -0.15, which is best observed in the lower part of the gel (values for u were estimated based on the concentration of ethidium bromide used). In fact, when the DNA was A, 6x174 wild type RFI DNA was treated with either BaZ31 (lanes A ) or S1 (lanes B ) under standard reaction conditions. The concentration of BaZ31 was raised 100-fold (0.1 unitslpg DNA), and the reaction time was doubled (30 min) when the reaction was carried out in S1 buffer (lane C ) . For the S1 reaction in BaZ31 buffer, the enzyme concentration was raised 20-fold ( Markers were the same as in Fig. 1.

2747-40-
2419--r'*linearized with AuaII prior to the Ba131 treatment the same cleavage pattern was observed as shown in lane 0 (data not shown). However, increase of a beyond -0.06 (lanes 15-30) resulted in the appearance of at least five major new recognition sites (bands 6, 9, 11, and 22). Several additional bands (15, 15a, 19, and 20) became more intense, but it is not clear whether that was due to generation of new bands of similar size or increase in intensity of existing bands. This suggests that Ba131 recognizes two types of structural perturbations in the closed circular RF DNA. One type is supercoil-independent and is cleaved with relatively low frequency, the other type is supercoil-dependent and, once formed, is recognized with high frequency, leading to very prominent bands in the gel. Formation of this type of structure requires a negative superhelical density of approximately 0.06-0.07. We have estimated this based on two other experiments, where the characteristic bands appeared after treatment of RF DNA with topoisomerase in the presence of 10 PM ethidium bromide (see Fig. 4), which generates a mixture of topoisomers with 33 f 5 supercoils (not shown).
BaL31 but Not Sl Cleaves Small Cruciforms in Supercoiled RF DNA-The supercoil-dependence of several of the Ba131 cleavage sites, and particularly the appearance of band 22 in Fig. 3, suggests that they may be cruciforms. For mapping of the cleavage sites, a larger amount of RF DNA was treated with topoisomerase in the presence of 10 p~ ethidium bromide before Bal31/AuaII cleavage and subsequent end-labeling. After electrophoresis and autoradiography (Fig. 4) the most prominent bands (6,11,19, and 22) were cut from the agarose gel using the autoradiograph as a guide. Both the large and corresponding small fragments were electroeluted and further restricted with either HhaI, HaeIII, or Hinfl and electrophoresed on 4% polyacrylamide gels (not shown). All fragment sizes were determined graphically as described under "Materials and Methods." The cleavage sites were thereby mapped with an error of approximately +30 bp to position 3969 (site  (Fig. 7). However, at least for site 19 the underlying cruciform would only have a stem size of 6 bp, which is thought to be thermodynamically unstable (18). Thus, we have tested more directly whether Ba131 can recognize and cleave small palindromes. The recently constructed mutant strain ins6430 (39) contains a synthetic palindrome with the sequence GCAGCCGGATCCGGCTGC, potentially capable of forming a hairpin/cruciform with 7 bp in the stem and 4 bases in the loop. A Ba131 analysis of ins6430 RF DNA and that of the parental strain ins6 is shown in Fig. 4, lanes E and

D. A new band is clearly visible in lane E but not lane D,
which was mapped by the above described procedure and corresponds to the inserted palindrome. This presents strong evidence that at least certain small palindromes can be extruded from the helical DNA backbone under sufficient torsional stress and can then be recognized and cleaved by Ba131.
When heating these DNAs prior to the Ba131 treatment (lanes B and F) the major bands disappeared, including the one for the inserted palindrome, resulting in the "typical" cleavage pattern shown in Fig. 1. This suggests that the secondary and tertiary structure that 4x174 RF DNA assumes during supercoiling is not the thermodynamically most stable configuration. Apparently, the cruciforms corresponding to bands 6, 11, 19, and 22, as well as the inserted palindrome, are the most predominant structural features in unheated in vitro supercoiled (a = -0.06 to -0.07) RF DNA, but become only minor sites, or disappear completely, after partial denaturation and renaturation. The sites that remain appear to be those that are recognized by Ba131 even in the absence of supercoil (Fig. 3). This topological shift can also be induced in the absence of heat by the S1 reaction conditions. In Fig. 1  shown that S1 recognizes basically the same sites as BaZ31 but cleaves in vivo supercoiled RF DNA predominantly at sites 22 and 6a (if sufficiently supercoiled). The same cleavage pattern was obtained with these unheated in vitro supercoiled DNAs (Fig. 4, lunes C and G). Thus, the low pH of the S1 reaction most likely causes partial denaturation of the DNA and loss of the extruded cruciforms, with the exception of the F-G cruciform (site 22), which may be stabilized by the action of S1 itself (Fig. 2).
Loss of Cruciforms during DNA Purification and Storuge-The finding that in vitro supercoiled RF DNA is topologically different from in vivo supercoiled DNA of similar superhelical density raises the question whether the topoisomerase reaction introduces topological artifacts or whether such artifacts are introduced into the in vivo supercoiled DNA during the extraction. We have fortuitous evidence suggesting the latter to be true. Fig. 5 shows a Ba131 analysis of several RF DNA preparations that were made at the onset of this study. The digest of wild type RF DNA (lune A ) gave a pattern very similar to that of in vitro supercoiled RF DNA (Fig. 3, lune   15), except that band 22 was very weak and band 17 fairly strong. A similar pattern was seen in the mutant digests, but no trace of the band characteristic for the inserted palindrome was observed in the ins6430 digest. Heating these DNAs (lunes B and E ) , or storage at 4 "C for several weeks (not shown), caused them to shift into a topological configuration whereby

FIG. 4.
Mapping of major supercoil-dependent B d 3 1 cleavage sites. RFI DNA (3-10 pg) from @X174 wild type, the palindrome insertion mutant ins6430, and its parental strain ins6 were treated with topoisomerase in the presence of 10 PM ethidium bromide. After removal of the enzyme and dye, aliquots were tested with Ba131 before and after heating to 65 "C, or with S1, as described for Fig. 1. band 17 displaced band 19 as the major Ba131 cleavage site, band 21 emerged, bands 6,9,11, and 19 became minor bands, and bands 22 and 15a disappeared completely. This pattern is virtually identical to that in Fig. 1, lanes A and B, and is what we have observed in all subsequent RF DNA preparations as the typical Ba131 cleavage pattern, even though they were not exposed to temperatures above ambient throughout extraction.
We conclude that supercoiling +X174 RF DNA in vitro or in vivo causes the DNA to assume a very specific but unstable configuration, which converts into a more stable but different topological form due to thermal fluctuations. The kinetic  (lanes A and B), ins6 ( l a n e C ) , and ins6430 (lanes D and E ) were isolated from infected E. coli cells by the standard procedure and shortly thereafter subjected to the Ba131 analysis before (lanes A, C, and D) and after heating (lanes B, and E). Several bands in lanes C-E migrate slower than their counterparts in lane A or B due to the insertion of 117 bp into strain ins& barriers that hinder this conversion are rapidly overcome at elevated temperatures. These different topological isomers can be differentiated with Ba131, but not with S1, since the reaction conditions (and possibly the enzyme itself) apparently lower the above-mentioned kinetic barriers and favor a topological fine structure in which site 6a and site 22 are predominantly recognized. On the other hand S1 recognizes a difference in the structure between heated and unheated DNA that can apparently not be distinguished by Ba131 (Fig.   1).

A B C D E
RF DNA Structure at Maximum Supercoil-The above data show that at physiological superhelical density the conformation of RF DNA with extruded cruciforms is relatively unstable and that an alternate conformation without these structures is apparently a thermodynamically more stable sink for the available supercoil energy. We have wondered whether this alternate conformation was able to absorb additional supercoil energy, or whether a superhelical density could be reached at which cruciforms remained stable structures. Hence, the superhelicity of wild type and mutant RF DNAs was increased to a maximum value of u = -0.16 (41) before probing the structure with Ba131 and S1 (Fig. 6). As expected from the data in Fig. 3, bands 8,15,15a, 16, and 20 in addition to 6, 19, and 22 were the most prominent bands in the Ba131 reaction (lane A), suggesting that some of them may also result from cruciform extrusion. Apparently several of these survived the melting and renaturation process (lane B), or the destabilization effect of the S1 reaction (lane C), suggesting that sufficient supercoil energy was left for their stabilization. It is also interesting to note that the inserted palindrome as well as some other bands that were dominant near physiological supercoil (bands 9, 11, 22; Fig. 3, lane 15) became very weak or were absent at high superhelicity, apparently being replaced by other competing structures (bands 8 and 20). Thus, stabilization of a given cruciform appears to be limited not only by a minimum superhelical density, which provides the energy for extrusion, but also by a maximum superhelicity, which when exceeded may cause a conformational transition that favors other structures.
Mapping of the Major SI and Bal31 Cleavage Sites on the 4x1 74 Genome-The reproducibility of the size and intensity of almost all bands obtained by cleavage of in vivo supercoiled RF DNA with either Ba131 or S1 allowed a preliminary mapping of the corresponding cleavage sites on the 4x174 genome. Two mutant 4x174 strains, ins6 with an insertion of 117 bp and de13 with a deletion of 27 bp (37, 48) were compared to the wild type strain to aid in the mapping. The Bal31/AuaII or Sl/AuaII cleavages were carried out as described for Fig. l , and the typical cleavage pattern was ob-tained for all strains (not shown). Any differences between these strains in the migration of bands were due to the insertion or deletion of nucleotides at position 979 on the Sanger map (36). The length of each clearly identifiable band of size greater than half-genome length (2693 bp) was determined graphically (''Materials and M e t h~s " ) , and plotted clockwise and counter-clockwise from the AuuII restriction site (position 5042) around the circular #X174 map. One of these sites was eliminated based on comparison of the wild type to the mutant maps. For an example, comparison of lanes A (wild type) and D (ins6) in Fig. 4 shows migration of one of the two fragments corresponding to cleavage site 19, and separation of dimer band 22, due to the insertion of 117 bp. This clearly indicates which of the fragments spans position 979.
This approach was useless, however, for fragments exceeding 4060 bp in length, and a second analysis, using the unique AuaI site (position 162) instead of AuaII was necessary to assign the cleavage position of larger fragments and to confirm the location of other sites (not shown). Nevertheless, because of the multitude of bands, the relative inaccuracy of an agarose gel in the high molecular weight range (230 bp), and comigration of some bands, not all of the cleavage sites could be mapped with certainty. Fig. 7 shows the location of the major and several minor Bal31ISl cleavage sites on the standard 4x174 map. At least four more cleavage sites between bp 4500 and 300 have been determined with the AuaI digest but could not be confirmed by the AuaII digest. The locations of statistically significant palindromes (see legend) have also been included in this figure to determine whether they might be the basis for S1 or Bd31 cleavage. Clearly, several arrows point directly at or very near positions of palindromes, but the inherent inaccuracy of the mapping technique precludes an assignment with any level of confidence based on this technique alone. Yet, the dependence on supercoil for the formation of bands 6,9,11,15a,19,20,and 22 (Fig. 3) support the theory that they are cruciforms cleaved by Ba131 in the non-base-paired loop region. At least for the F-G cruciform (site 22) this has been well established (14,30), and theoretical calculations predict also the extrusion of the two largest perfect palindromes under sufficient torsional stress (49,but see also Ref. 34). Their positions coincide with sites 6 and 15a respectively.
Several other sites (2, 3, 7, 10, and 17b) that map near palindromes appear to be only marginally or not at all favored by supercoiling, suggesting that this overlap may be coincidental or that cruciform extrusion is not required for their recognition by the nuclease.
On the other hand it can be said that several cleavage sites (1, 4 , 5 , 8, 12, 14a,b, 16, 17a,c, 18, and 21) are sufficiently far removed from the indicated palindromic regions to preclude them as the basis for recognition, even though cleavage by Ba131 of some of these was promoted by supercoiling. However, several of these sites map at or very near palindromes with loop sizes between 11 and 20 bases (not shown).
Comparison of cleavage positions with the base-pair opening probability of the underlying DNA sequence shows that there is little correlation, and apparently increased breathing of DNA at certain positions in the genome is not the reason for cleavage by S1 or Bal31.
It is interesting to note that with the exception of site 11 all supercoil-dependent sites are located near the start sites or ends of genes.

DISCUSSION
Computer studies of the 4x174 genome have revealed that it contains many more palindromic sequences than mere chance would predict (31-33), particularly in intercistronic regions, and several of these have been speculated to have a biological function (reviewed in Ref. 50). If this putative function requires formation of a cruciform structure in the RFI DNA, rather than a hairpin in mRNA or the singlestranded genomic DNA, one would assume that this structure would be present in at least one or two of the roughly 20 RF molecules in the infected cell. However, most of these palindromes are relatively small, and the idea of small cruciforms being stabilized at physiological supercoil has been generally dismissed, based on some experimental evidence (21), but mostly on theoretical grounds (51). Based on the experimental findings in this report and the following consideration, we argue that cruciforms with stem sizes of 6 to 7 bp may exist at least transiently in a fraction of the DNA molecules at physiological supercoil. Purified 4x174 RFI DNA isolated from infected E. coli cells generally has a superhelical density near -0.065, which provides a substantial amount of free energy for the formation of non-right-handed secondary structures. The free energy cost for cruciform formation has been estimated by various laboratories and was found to range between 17 and 25 kcal/mol (19,(51)(52)(53)(54). Based on this one can estimate that the minimum stem length of a cruciform that would be stable in the 4X genome must be around 9 bp (44,55), which was also estimated to be the minimum stable cruciform size in the plasmid CoiEl (18). These calculations assume a mean supercoil of 33.5 negative superhelical turns for #X174, but it is known that a distribution of topoisomers around this mean exists. We have found that a measurable fraction of molecules have 40 and more supercoils, which would decrease the minimum stem size for stable cruciforms to 7 or 8 bp in those topoisomers {loop size assumed to be 4 bases).
Using a mutant #X174 strain from our collection with a palindromic insert, theoretically capable of forming a cruciform with 7-bp stem size and 4 bases in the loop, we have shown that a measurable fraction of mutant RF DNA molecules were cleaved by Ba131 specifically at the site of insertion, when such molecules were supercoiled in vitro to a negative superhelical density of -0.06 & 0.01. This supports the theoretical argument made above that small cruciforms can be stable in a fraction of RF DNA molecules at physiological superhelicity. Furthermore, we have indirect evidence that the smallest cruciform that can be detected in the 4x174 genome with the technique employed here has 6 bp in the stem (sites 9 and 19, Fig. 7). This is based on our mapping studies, which place each of the supercoil-dependent cleavage sites at the location of a potential cruciform with at least 6 bp in the stem and not more than 10 bases in the loop. Cleavage of a cruciform with 6-bp stem length by a nuclease was also reported by Sheflin and Kowalski (9).
It is unlikely that all such palindromes in a genome can be extruded into cruciforms, since in addition to stem length there are clearly other factors that determine this transition. This is borne out by the Ending that no consistent correlation can be made between band intensity and the stem size of the putative underlying cruciform structure. There i s also at least one palindrome (position 1700) for which we have not detected any evidence of cleavage either by Ba131 or S1. We assume that the intensity of a band is directly proportional to the frequency with which a given site is cleaved by the nuclease, i.e. the frequency with which that structure occurs among the population of DNA molecules. This is known to be a function of stem and loop size, as well as sequence (27, 56).
If small cruciforms can exist at physiological supercoil, why are they generally not detected by the variety of techniques that have been employed in the analysis of DNA conformation? Our data strongly suggest that most, if not all, cruciforms in 4x1'74 RF DNA are only transient structures. They apparently serve as temporary sinks of supercoil energy, but this energy can be stored more permanently in an alternate DNA conformation. We have shown that a substantial difference exists between the conformation of in vitro supercoiled RF DNA and that isolated from infected cells, even though both had similar superhelical densities. Heating the in. vitro supercoiled DNA changed its conformation to where it gave the same 3~131 digestion pattern as unheated in vivo supercoiled DNA, while heating the latter was without effect. Thus, partial melting of the DNA allowed the transition into the stable conformation, which is devoid of cruciforms. The loss of small cruciform structures due to heating has been observed by other investigators (22). Apparently the activation energy required for this transition is very low, since in the case where the supercoil-dependent Ba131 cleavage sites were preserved during the DNA extraction,.they were subsequently tost upon storage of the DNA for an extended period of time at 4 "C.
Just as important as the effect of temperature and isolation protocol on the structural transitions in DNA is the probe employed for its analysis. We have found a substantial difference between nuclease Ba131 and S1 with regard to cleavage of the supercoil-dependent sites. The cloned palindromic sequence and most other supercoil-dependent BaZ31 cleavage sites were not recognized by Sf. to any detectable degree, with the exception of the major palindrome (site 22). This structure was the predominant S1 cleavage site in sufficiently supercoiled RF DNA but was recognized only at low pH. The second most frequently cleaved SI site (6a) was only a minor site in the 3~131 reaction and is apparently not a cruciform structure. We believe that the low pH at which the S1 reaction is generally performed, and possibly the enzyme itself, induces a structural transition in the DNA which destabilizes some structures (cruciforms) and stabilizes others. Recognition sites other than cruciforms have already been described for S1 (4, 13), and differences between this enzyme and other nucleases in their specificity for alternate DNA secondary structures have been observed (9). In genera1 our data suggest that BaL31 is the better of these two probes when testing for the presence of small cruciforms.
While differences exist between S1 and Ea131 regarding the recognition of supercoil-dependent sites, they both recognize consistently and specifically a series of sites on the 6x174 genome independent on superhelical density, which appeared to be present even in linear DNA. Similar observations have been reported by Kowalski for PM2 DNA (23). Regions with a high base-pair opening probability can be excluded as candidates for recognition as our mapping studies have shown, but there are many supercoil-independent deviations from the B-helix known that are due to sequence and can be recognized by a variety of nucleases (28, 57,58).
We have discussed our results of the supercoil-dependent 3~131 cleavage sites with a strong bias toward cruciforms as the underlying non-B secondary structure, based on our mapping data. Yet, we can not completely exclude that the overlap of cleavage sites with the location of palindromes is coincidental in at least some cases, and that the structure recognized by the nuclease is actually a 2-DNA to B-DNA junction or a slippage site. But independent on the actual conformation that leads to recognition by 3~131, it is important to note that the supercoil-dependent sites are preferentially located within intercistronic regions or near start or stop sites of genes, while the supercoil-independent sites appear to be randomly distributed in the genome. This supports the notion that these alternate secondary structures are involved in the regulation of gene expression. The labile nature of particularly small palindromes, which are highly sensitive to the superhelical density and any other factors that might redistribute the