Regulatory circuit design and evolution using phage λ

Abstract

Bistable gene regulatory circuits can adopt more than one stable epigenetic state. To understand how natural circuits have this and other systems properties, several groups have designed regulatory circuits de novo. Here we describe an alternative approach. We have modified an existing bistable circuit, that of phage λ. With this approach, we used powerful genetic selections to identify functional circuits and selected for variants with altered behavior. The λ circuit involves two antagonistic repressors, CI and Cro. We replaced λ Cro with a module that included Lac repressor and several lac operators. Using a combinatorial approach, we isolated variants with different types of regulatory behavior. Several resembled wild-type λ—they could grow lytically, could form highly stable lysogens, and carried out prophage induction. Another variant could form stable lysogens in the presence of a ligand for Lac repressor but switched to the lytic state when the ligand was removed. Several isolates evolved toward a desired behavior under selective pressure. These results strongly support the idea that complex circuits can arise during the course of evolution by a combination of simpler regulatory modules. They also underscore the advantages of modifying a natural circuit as an approach to understanding circuit design, systems behavior, and circuit evolution.

Complex gene regulatory circuits involve interactions among many components, including specific DNA-binding proteins and the cis-acting sites to which they bind. Work over the past decades has led to a solid understanding of the operations of many important biological circuits and switches, including mechanistic detail and qualitative circuit diagrams of the connections (Wall et al. 2004). This work has generally taken place in a reductionist framework, focusing ever more finely on the mechanistic details by which individual components of the circuit operate and interact.

At the same time, this work has revealed that many natural circuits involve factors such as nonlinearity and feedback that complicate their behavior and make it difficult to predict. For instance, in phage λ, the system we use here, nonlinearity in repressor binding to DNA arises from weak dimerization by CI and Cro and from cooperative DNA binding by CI repressor; expression of cI is also subject to both positive and negative feedback loops (Ptashne 1992, 2004). The field of systems biology is concerned in part with developing approaches to understanding and predicting the “emergent” behavior of systems—behavior that results from factors like nonlinearity and feedback—using a combination of computational and experimental approaches (Kitano 2002). Such endeavors ideally will complement the reductionist approach; together, they should provide far more complete descriptions of a system as a whole than either approach could do alone.

One approach to these questions is systematic experimentation on natural circuits, focusing on questions of systems behavior. In an early example of this, the chemotaxis system of Escherichia coli was found to be highly robust to changes in certain parameters—that is, its behavior does not change very much relative to the magnitude of the change in parameter (Alon et al. 1999). With λ, we have found that the qualitative operation of the phage λ circuitry is robust to changes that affected the affinities of two repressors for their operators (Little et al. 1999). More recently, we have characterized mutants that alter the set point of the threshold behavior in prophage induction (J.W. Little, unpubl.). Although clearly feasible, however, this general approach has not yet enjoyed wide use.

A contrasting experimental approach to understanding emergent behavior is that of de novo design of gene regulatory circuits, using well-characterized components (Elowitz and Leibler 2000; Gardner et al. 2000; Guet et al. 2002). In a relatively simple example of this approach, the “toggle switch” (Gardner et al. 2000; see also Atkinson et al. 2003), two repressors whose activities could each be modulated by the experimenter were arranged such that each repressed the other. When one repressor was inactivated, the other came to predominate and continued to do so even under conditions that allowed the first repressor to be active if it was present. The converse was also true. Hence, the system was “bistable”; two alternative stable states could exist. Significantly, however, in addition to several successful examples of this design, it was found that many other combinations were not bistable, in that one repressor could take over from the other when allowed to function. In these cases, the two regulatory states were not in balance with one another; one state was too “strong” for the other to be maintained.

In the present work, we have devised an alternative approach to systems design that combines the simplicity of de novo design with the interesting complexity shown by a natural circuit. We have taken a well-characterized natural circuit, the bistable switch of phage λ, and replaced one of the critical regulatory proteins, Cro repressor, with another well-studied regulatory protein, Lac repressor. We also included cis-acting sites for Lac repressor to allow it to execute the functions of Cro.

λ offers several advantages for this “hybrid” approach. First, as detailed following, the λ circuitry exhibits a range of interesting regulatory behavior (Ptashne 1992, 2004). Second, the powerful genetics available in the λ system allows us to identify functional circuits, select rare variants with desired characteristics, and evolve the behavior of these variants. Finally, the great depth of knowledge about λ allows interpretation at the mechanistic level.

The λ circuit can exist in two alternative regulatory states, and it can switch from one state to the other under control of the experimenter (Ptashne 1992, 2004). The circuitry also involves an initial choice between the two alternative states, a choice made soon after a cell is infected and later stabilized by well-known regulatory interactions. A cell infected with λ can follow either of two exclusive pathways. It can follow the lytic pathway, in which the viral DNA replicates and is packaged into ∼100 new virions, which are then released by cell lysis. Alternatively, the infected cell can follow the lysogenic pathway, leading to a stable association with the host; the phage DNA is integrated into the host genome and its lytic genes are repressed by CI repressor. In addition, although the lysogenic state is highly stable, it can switch to the lytic state in a “genetic switch” called prophage induction. When the host SOS system is induced by DNA damage (such as from ultraviolet [UV] irradiation), RecA protein becomes activated to a form that can mediate specific cleavage of CI, inactivating it and leading to expression of lytic genes and the lytic pathway (Little and Mount 1982; Roberts and Devoret 1983).

At the mechanistic level, the choice between lysis and lysogeny after infection is controlled mainly by the levels of the viral CII protein. CII activates the P_RE promoter, leading to high levels of cI expression and favoring lysogeny. CII levels are controlled by the physiological state of the cell, and by the level of another viral protein, CIII (Herskowitz and Hagen 1980). Once a regulatory state has been chosen, it is maintained by the actions of two proteins, CI and Cro, acting mostly at a complex regulatory region termed the O_R region (Fig. 1). Both proteins bind to three sites in O_R (Fig. 1B) but with differing affinities and different consequences (Ptashne 1992). Each protein maintains its own expression while blocking that of the other. If CI predominates, it represses the lytic promoter P_R and activates the lysogenic promoter P_RM, stabilizing the lysogenic state; it is generally believed that if Cro predominates, it stabilizes the lytic state by repressing P_RM.

Figure 1.

Design of λlacI. (A) Logical representation of the phage λ lysis–lysogeny circuit. Plus (+) and minus (-) signs indicate activation and repression of promoter, respectively (Lac repressor could repress P_RM when a lacO variant was in the position of O_R3). CIII and CII (not shown) are expressed from P_L and P_R, respectively. Expression of P_L is controlled in λ by CI binding to two operators, O_L1 and O_L2, not shown here, which lie to the right of P_L in positions analogous to O_R1 and O_R2, respectively. (B) Control of regulatory state by the pattern of occupancy by CI or Cro. The circuit is maintained by the actions of CI and Cro at the O_R region. CI stabilizes the lysogenic state, and Cro stabilizes the lytic state. (C) Design of λlacI. Cro was replaced with lacI. LacO alleles were installed in two positions (P_L and P_R); at the position of O_R3, some variants had a lacO allele, and some had O_R3. The same set of lacO alleles was used at all three sites. (D) LacO alleles and the Shine-Dalgarno sequences used in combinatorial approach. Allele A is wild-type lacO; mutations in other lacO alleles are in bold type; binding is progressively weaker going from B to E (Betz et al. 1986). W is λ O_R3. In box SD, Shine-Dalgarno sequences and start codons are in bold type; SD alleles allow progressively less-efficient translation going from A to F (Gardner et al. 2000). (E) O_R3 and lacO are in bold type. O_R1 and O_R2 are italicized. Locations of -35 and -10 regions for the promoters are shown. The base substitutions at the ends of the lacO at O_R3 should not affect LacI binding (Betz et al. 1986).

In addition to testing the feasibility of our approach, we had several specific goals in the present work. The first was to test whether a natural gene regulatory circuit retains a modular organization (Hartwell et al. 1999; Wolf and Arkin 2003). Success would provide support for the idea that circuits could arise initially in evolution by proper combinations of functional modules. Second, it should allow us to explore issues of balance. We expected that only certain circuits would be in balance well enough to approximate the behavior of natural λ? Moreover, the use of a repressor whose activity we could manipulate offered the possibility of affecting the balance at will. Third, the use of genetics might allow evolution of circuits with novel behavior. We describe experiments that have met all of these goals.

Results

Approach

We chose to replace λ Cro rather than CI, because Cro is a much simpler protein than CI. CI has several features for which it would be difficult to substitute. In addition to its repressor activity, CI acts as an activator, stimulating its own expression from P_RM about 10-fold. It also binds cooperatively to two adjacent operators, O_R1 and O_R2, and also has higher-order cooperativity, mediating looping between the O_R and O_L regions. In contrast, Cro is only known to act as a repressor, and has little or no cooperative DNA binding.

Cro partially represses the early lytic promoters P_L and P_R, particularly at a late stage of the lytic cycle; failure of this function leads to overproduction of toxic gene products and to other poorly understood effects that interfere with lytic growth (Sergueev et al. 2001). As a consequence, cro mutations are almost lethal; λ cro forms tiny plaques. In addition, it is believed that Cro operates during the switch from the lysogenic state to the lytic state, binding to O_R3, thereby repressing P_RM and making the switch irreversible. Accordingly, in replacing Cro we needed to provide the ability to repress P_R and P_L, and we also provided in some constructs the ability to repress P_RM.

We needed to provide cis-acting sites at which another repressor with different DNA-binding specificity could carry out the functions of Cro in the λ life cycle. An additional consideration in designing our system was that, as described earlier, both Cro and CI bind to the same sites, although with different affinities. This meant that we could not remove the Cro-binding sites without also affecting the functions of CI.

We chose Lac repressor to replace Cro for several reasons (Gilbert and Müller-Hill 1966). First, it has been extensively characterized. Second, its function can be modulated by adding isopropyl-β-D-galactopyranoside (IPTG) to the growth medium. IPTG weakens the affinity of Lac repressor for its operator (Riggs et al. 1970). Importantly, titration with IPTG should allow us to shift the balance of the circuitry. Finally, a feature of the lac operon that is fortunate for our purposes is that the lac operator lies after the start point of transcription, rather than being within the promoter as in most repressors. Hence, we could install lac operators at P_L and P_R without at the same time removing the binding sites for Cro and CI (Fig. 1).

Lac repressor does differ from Cro in at least two ways that are likely to bear on the behavior of the λ circuitry. First, Cro dimerizes weakly, providing a source of nonlinearity, whereas oligomers of Lac repressor are highly stable. Cro does not form higher-order oligomers. Second, in contrast, Lac repressor is a tetramer; moreover, it can mediate looping between lacO sites (Schleif 1992). With supercoiled DNA, this looping can occur over distances comparable to those separating P_L and P_R (Whitson et al. 1987). Hence, in the present design we have likely made a qualitative change in the circuitry, because repression and derepression of these two promoters during lytic growth is likely to be coupled to some extent, unlike the case in wild-type λ. In addition, looping allows highly cooperative binding, providing another source of nonlinearity.

As described in the introduction, for a circuit to be bistable, the two stable states must be in balance, such that one cannot readily take over once the other has been established. This balance can in principle be affected by the values of parameters such as the amount of a regulatory protein or its affinity for its operators. Because it is difficult to predict the proper choices of parameters to balance a new circuit, we tested many combinations using a combinatorial approach (Guet et al. 2002; Yokobayashi et al. 2002), in which each of four cis-acting sites had several alleles with different properties. These sites included the lacO sites at P_L and P_R; O_R3, at which either the native λ O_R3 site or a lacO was allowed; and the Shine-Dalgarno sequence in the message for Lac repressor (see Fig. 1D; Supplementary Fig. S1; Supplementary Table S1). At lacO sites, we provided the wild-type operator or one of four lacO mutants that weakened binding to varying extents; the same five lacO alleles were possible at each of the three sites. The spacing between lacO located at O_R3 and P_R, 72 bp, did not support highly efficient looping (Muller et al. 1996). At the Shine-Dalgarno sequence, we used a series of sequences that show a wide range of translation efficiencies in a different context (Gardner et al. 2000); we have not tested directly their efficiencies in the present context.

Isolation of λlacI variants

Pools of variant O_R regions were cloned into a new λ vector that allows cloning of the O_R region (Michalowski et al. 2004). Regulation in this vector, λJL351, differs somewhat from that of wild-type in two respects. It contains a silent XhoI site early in cI, which weakens by about half the strength of P_RM (probably by reducing translation efficiency of the leaderless P_RM message), presumably leading to lower levels of CI. Probably as a consequence of this, lysogens of λJL351 are induced by lower levels of UV light than is wild-type λ. This phage carries a kanamycin marker, allowing us to select for the ability to form stable lysogens. We cloned the variant O_R regions, plus the lacI gene in place of cro (see Supplemental Material) using as vector pooled DNA from five variants with different lacO alleles at P_L. Ligation mixtures were packaged, resulting in a pool of phage (Supplementary Table S2).

We then applied an initial genetic selection for the capability to grow lytically. The pool was plated for plaques in the absence of IPTG or at various levels of IPTG. Thirty-six plaque-forming phage, which we term λlacI, were isolated and characterized (Table 1; data for all mutants are in Supplementary Table S3). Sequence analysis showed that only three of these isolates were duplicates, implying that we have not recovered all of the isolates that can form plaques. We symbolize these isolates by the alleles they carry at each of the four variable sites; for instance, AWDF has allele A at P_L, W at O_R3, D at P_R and F in the Shine-Dalgarno sequence of lacI. For lacO alleles, binding is progressively weaker going from A to E (Betz et al. 1986). W is λ O_R3. SD alleles allow progressively less efficient translation going from A to F (Gardner et al. 2000; in this work, their translational efficiency was measured using GFP expression). As an initial characterization, variants were tested for their ability to grow lytically and to lysogenize under a range of IPTG concentrations. These tests revealed a wide variety in the properties of the recombinants.

Table 1.

Behavior of selected λlacI variants

	Plaque formation and morphologya				Lysogenizationc
[IPTG]	0	10-5 M	10-4 M	10-4 M	0
MitCb	-	-	-	+	-
Name
WT	L	L	L	L	S
ABDA	S	M	No	No	S&U
ADDD	M	M	No	S	U
AEAF	No	Tiny	No	S	No
AECF	S	M	No	No	No
AEDA	M	M	No	No	U
AWBF	M	M	No	No	S
AWCF	L	L	No	L	S
AWDF	M	S	No	M	S
CECA	S	No	No	No	No
CWCD	M	No	No	No	S

Variants were symbolized by the alleles they carry at each of the four variable sites (P_L-O_R3-P_R-SD; see Fig. 1 and text).

^a“No” indicates no plaque formation; “S”, “M”, or “L” indicates the size of plaques, small, medium, or large, respectively

^bMitC indicates presence of 1.0 μg/mL mitomycin C in top and bottom agar

^c“S”, “U”, or “No” indicates stable, unstable, or no lyosogen, respectively, in the absence of IPTG; “unstable” means that colonies of lysogens were of heterogeneous size and that they did not grow to high density on inoculation into liquid media

Lytic growth of all isolates was sensitive to the presence or absence of IPTG (Table 1; Supplementary Table S3). Many variants could establish and maintain the lysogenic state (Table 1; Supplementary Table S3). Finally, we tested whether lysogenizing variants could carry out prophage induction (Fig. 2; Supplementary Table S3). For this process to occur, CI levels in the lysogen must be reasonably close to those of wild type, and the lytic state must be able to take over when CI is removed. We found that several variants could execute this switch (Fig. 2); the behavior of AWCF was closest to that of wild type.

Figure 2.

UV dose responses for prophage induction. Exponentially growing cultures of lysogens were centrifuged, suspended in TMG, and irradiated (see Materials and Methods) at the indicated doses; aliquots were diluted 10-fold in LBGM and shaken for 2 h at 37 °C in the dark, treated with CHCl₃, and titered. (Crosses) Wild type; (diamonds) AWCF; (stars) AWBF; (squares) AWCE; (triangles) AWBE.

From these data we conclude, first, that it is possible to isolate λlacI variants with regulatory and phenotypic behavior similar to those of wild-type λ? Importantly, these findings validate our approach of replacing one crucial regulatory protein with another. Second, we conclude that many variants are defective in one or more aspects of λ behavior, confirming the utility of a combinatorial approach. We next discuss which combinations are acceptable, then turn to a more detailed analysis of selected variants.

Acceptable combinations of cis-acting sites

It is difficult to infer general rules for acceptable combinations of cis-acting sites (Table 1; Supplementary Table S3). At P_L, the strongest lacO allele, “A”, was preferred (28 of 33). The P_L operon encodes several proteins that are toxic when overproduced (Sergueev et al. 2001). This pattern suggests that repression of P_L is the most important function for Lac repressor during lytic growth (see also following). At O_R3, both the native O_R3 site and lacO variants were allowed for lytic growth. However, many variants with a lacO at this site could not lysogenize, even in the presence of IPTG. We do not fully understand the mechanistic basis for this pattern. Although Lac repressor might block P_RM by binding to the lacO site, it should not bind tightly in the presence of IPTG (see also Discussion). It is possible that the substitutions between the -10 and -35 regions of P_RM alter the properties of this promoter. At P_R, all five lacOs were found, but “A” was rare; presumably this would cause too low a level of P_R expression. In the lacI SD sequence, none of the variants with “W” in O_R3 had the strongest SD “A”; in contrast, half of those with lacO alleles at O_R3 had “A” in SD.

A cluster of variants (AWBE to AWEE; Supplementary Table S3) had roughly similar behavior. This group had “AW” in the first two positions and a relatively weak SD sequence at lacI; none had the strongest lacO (A) at P_R. These did differ from one another, mostly in the phage yield after prophage induction, but all gave good lytic growth and had the ability to lysogenize, indicating that there is a range of parameters for which the λlacI system can maintain a stable regulatory state and switch to an alternative state.

Balance in the λlacI system

The properties of many variants gave insight into the important systems property of balance. We found that the balance of the λ system could be perturbed in our isolates by adding or removing IPTG, which weakens specific DNA binding by Lac repressor. Balance could be affected in each of two important aspects of λ regulation, in the lysis–lysogeny decision and in the stability of the lysogenic state. We describe these in turn.

Lytic growth of almost all variants was responsive to IPTG. All but three isolates could not form plaques in the presence of 10^-4 M IPTG, where the activity of Lac repressor is very weak. Failure of plaque formation might result from one or both of two causes—there might be overexpression of toxic genes or the infected cell might usually follow the lysogenic pathway. Cells might be unable to complete the lytic pathway because weak repression by Lac repressor leads to overexpression of toxic phage functions, such as kil, during lytic growth (Sergueev et al. 2001). This is likely for variants that could not form stable lysogens. Conversely, variants that could form stable lysogens might usually or always choose the lysogenic pathway after infection. That is, IPTG would upset the balance between these two choices.

We tested these possibilities by assaying plaque formation in the presence of Mitomycin C, an SOS-inducing agent that leads to cleavage of CI and thereby blocks the lysogenic pathway. We expected that a variant that was capable of lytic growth but usually followed the lysogenic pathway would form a plaque under these conditions. Among variants unable to form stable lysogens, 9/12 did not form plaques in the presence of Mitomycin C and 10^-4 M IPTG (Supplementary Table S3), suggesting that these could not complete the lytic cycle. Among stably lysogenizing variants, 10 could also not form plaques under these conditions. In contrast, seven others could do so, suggesting that in these cases IPTG conferred a bias toward the lysogenic response. A direct test of lysogenization frequency after single infection in the absence or presence of IPTG confirmed this prediction for two isolates (ADDD and AWCF; Table 2), and showed that the choice was also biased by IPTG in two variants (ABDA and AWBF) that could not complete the lytic pathway. At the mechanistic level, we surmise that poor repression by Lac repressor led to overproduction of CII and CIII and expression of P_RE (Herskowitz and Hagen 1980).

Table 2.

Effect of IPTG on lysogenization frequency

	-IPTG	+IPTG
WT	1.2	1.2
ADDD	2.3	57
AWCF	0.9	38
ABDA	0.2	18
AWBF	0.3	26

Each number shows lysogen efficiency (%). Cells were grown, singly infected, and plated on kanamycin plates without or with 10^-4 M IPTG (see Materials and Methods). Lysogenization by wild-type λ after single infection is very inefficient (Kourilsky 1973); it is much more efficient after multiple infection.

In another case, removal of IPTG also upset the balance, but in this case it disrupted the balance between two stable states of the system. The CECA variant formed lysogens at 10^-4 M IPTG, and these were almost as stable at 10^-4 M IPTG as those of wild-type λ, as judged by the levels of free phage in the culture (in the wild type, these probably arise primarily from sporadic activation of the SOS system in a small fraction of cells, resulting in a switch to the lytic state). Several generations after removal of IPTG, lysogens of CECA began to grow slowly and eventually lysed (Fig. 3A), releasing high levels of free phage (Fig. 3B). We conclude that in this variant the regulatory circuitry was in balance when Lac repressor function was weak, but that Lac repressor could eventually disrupt the balance if it was allowed to function by removal of IPTG.

Figure 3.

Time course for cell growth and free phage release by lysogens of the wild type and CECA, with or without IPTG. Lysogens of wild type and CECA in strain JL7223 were isolated by kanamycin selection, and single lysogens were identified by a PCR test (Powell et al. 1994); those of CECA were propagated with 10^-4 M IPTG. For both strains, overnight cultures were grown in LBGM with 10^-4 M IPTG and grown to exponential phase in the same medium; after washing by centrifugation, portions were diluted and grown with or without IPTG. Aliquots were removed at the indicated times; cell number was estimated by O.D.₅₉₀ (A), and free phage titers (B) were measured. (Open circles) Wild type without IPTG; (closed circles) wild type with IPTG; (open squares) CECA without IPTG; (closed squares) CECA with IPTG. In A, phage/cell is not shown for 300 or 330 min, because the O.D.₅₉₀ does not give an accurate cell count in the presence of cell debris from lysis; phage titers at 270, 300, and 330 min were 4 × 10⁸, 9 × 10⁸, and 2 × 10⁹/mL, respectively.

At the mechanistic level, we infer that the variant could make Lac repressor, at least occasionally, while in the lysogenic state. The long delay in switching probably resulted from the need to dilute out existing CI after repression of P_RM by Lac repressor, until CI levels fell enough to permit derepression of P_L and P_R. In addition, perhaps Lac repressor was not present in all cells at the time IPTG was removed; in such cells, switching would be further delayed by the length of time it took for the first molecules of Lac repressor to be made.

Evolution of λlacI variants

We tested whether variants could evolve under selective pressure. Because most had defects in at least one aspect of λ regulatory circuitry, we could select for secondary variants with altered phenotypes. Two examples follow. First, λlacI variants whose lacO allele at P_L was “C” could not make plaques in the presence of 10^-5 M IPTG. We isolated two independent variants of CWCD that could do so. Both contained mutations in P_L (-10 GATACT to GATGCT, or -35 TTGACA to TTGATA) that probably weakened P_L. We infer that the growth defect of CWCD with 10^-5 M IPTG resulted from weak repression of P_L, giving overexpression of toxic gene products. Second, we identified many pairs of variants in which a single site change of lacO or SD caused an altered phenotype (Supplementary Table S4). For example, AEAF and AECF differed only in the lacO sites in P_R. AECF could form plaques without IPTG, but AEAF could not. Three independent secondary variants of AEAF that could do so were isolated. All three had changes in lacO in P_R. The changes probably weakened the affinity of Lac repressor for this site, implying that the growth defect of AEAF results from too-strong repression of P_R, which restricted early lytic gene expression more than in wild type.

Discussion

Our success at replacing Cro with Lac repressor in the context of the intact phage circuitry has numerous implications for regulatory mechanisms, systems properties, and evolution of gene regulatory circuitry, as we discuss following. Our work also illustrates the advantages of a genetic approach to analyzing circuit design and systems behavior. The use of a combinatorial approach allowed us to identify acceptable circuits from a broad range of possibilities; combinatorial approaches have been similarly used in de novo design to identify toggle switches that are in balance and circuits with a variety of connectivities.

Mechanisms

In the Results, we discussed several examples in which the behavior of variants, or comparisons between pairs, appear to be consistent with current understanding of λ circuitry and the properties of Lac repressor. Conversely, there are many cases in which we do not understand the behavior. A major advantage of using λ and Lac repressor is that our depth of knowledge helps us to recognize the limits to current understanding and departures from it.

For instance, we found that variants ACDF and CDBA were able to form stable lysogens, despite the fact that both have a 1-bp deletion in O_R1. This mutation would probably weaken the strength of P_R, but should completely abolish binding of CI to this site. It is unclear how the lysogenic state could be maintained in such phage.

With Lac repressor, our results highlight a limitation in our current understanding. Careful studies have not been done to assess the interaction of Lac repressor with mutant operators as a function of IPTG concentration. Accordingly, we do not know how various levels of IPTG affect these interactions.

An additional complication in the case of our constructs is that Lac repressor probably supports looping between lacO sites at P_L and P_R; in addition, in phages with a lacO site at O_R3, three different loops are possible. Looping between the sites at O_R3 and at P_R may be weaker than the long-range looping between either site and that at P_L. In any case, in vitro studies on supercoiled DNA have shown that looped complexes with DNAs containing two operator sites are far more stable than complexes with DNA as a single operator, presumably because of the local concentration effects and juxtaposition due to supercoiling (Hsieh et al. 1987; Whitson et al. 1987; Vologodskii et al. 1992). Dissociation at each of the individual operator sites may mirror release of LacI from a single-operator DNA, but the high local concentration of a second high-affinity site would favor reformation of the looped structure and ensure a long half-life for the protein–DNA complex.

The consequences of looping and the uncertain energetics for all of the possible looped forms make it difficult to predict in detail the behavior of λlacI phages. However, we suggest that looping could account for the finding that most phages with a lacO at the position of O_R3 could not form stable lysogens, even with 10^-4 M IPTG. Possibly looping increased the residual affinity of Lac repressor enough to allow substantial occupancy of this site, repressing P_RM and preventing maintenance of lysogeny.

Another unexpected finding was that the AACA variant showed a “reverse” response to IPTG—it could lysogenize in the absence of IPTG but not in its presence. That is, Lac repressor function was necessary for stable lysogenization. We speculate that two different factors may help account for this unexpected behavior. First, looping may again be involved. Loops might form between Lac repressor bound at lacO at P_L and with either of two sites in the O_R region—the site at the lacO at P_R or the one at O_R3. By itself, the latter loop would be more stable (because the lacO at O_R3 is wild type). Second, it is known that CI can also mediate cooperative looping between CI dimers bound to O_R1 and O_R2 and those at O_L1 and O_L2 (Dodd et al. 2001, 2004). Possibly a looped protein–DNA complex forms involving both CI (at O_R1 and O_R2) and Lac repressor (at P_R). The cumulative free energy gain might favor CI binding to the mutated O_R2 site, stimulating expression of CI, and thereby stabilizing the lysogenic state. Addition of IPTG might destabilize this complex so that P_RM was no longer stimulated, disrupting the lysogenic state.

Behavior and evolution of regulatory circuitry

The likelihood that Lac repressor-mediated looping is taking place implies that the connectivity of the circuits studied here differ somewhat from that of normal λ, because expression of P_L and P_R is likely to be controlled in parallel. The consequence of this for lytic growth is uncertain. The only essential protein expressed from P_L is the N protein. If P_L were repressed for several minutes, N levels would drop substantially, because this protein is very unstable, but that might have little effect on lytic growth if some transcripts from P_R had already been antiterminated by the action of N (Roberts 1993).

The λ regulatory circuitry is often termed “bistable,” because both of its regulatory states are stable. The lytic state does not need to be stable during normal λ growth, because it is transient. When lytic functions are blocked by mutation, however, the circuitry can exist in a stable “anti-immune” state in which Cro predominates (Eisen et al. 1970; Calef et al. 1971). We have not shown that our λlacI variants can adopt this anti-immune state. Hence, it is possible that the circuits studied here are not bistable in this sense.

It is becoming increasingly clear that many biological systems have a modular organization. Theoretical approaches suggest that modularity of circuits is needed for robustness and facilitating the generation of variation (Wolf and Arkin 2003). Although definitions of a module vary widely, we use the term here to refer to a regulatory protein, the cis-acting sites to which it binds, and its actions while bound. A plausible model for creation of complex regulatory circuits is that they arise by random associations between functional modules, followed by selection for those with adaptive properties. However, it is possible in principle that during continued selection and refinement of a circuit, the modules would lose their identity by extensive interlocking with other elements of the system. Our ability to replace Cro with Lac repressor and its cis-acting sites provides a straightforward counterexample. Clearly the modularity of the λ circuit has been retained sufficiently to allow this substitution to take place.

On the basis of this work, we suggest a plausible pathway for creation of a complex regulatory circuit from simpler parts (Ptashne and Gann 1998). A regulatory circuit like that in AWCF could arise by nonhomologous recombination between one module containing CI and its operators and another with Lac repressor and its operator at O_R. This would require only that a promoter like P_R be present in the λ module. Because it is likely that looping occurs in our variants and strengthens the binding of Lac repressor, it would be helpful if another lacO site was nearby to allow looping. In this case, a module carrying two lacO sites and the lacI gene would undergo insertion of the P_R-P_RM-cI module between the two lacO sites, creating a functional circuit.

It is widely believed that changes in cis-acting sites more often underlie the evolution of morphological diversity than do changes in gene number or protein functions (Carroll 2000). Our findings support this general model. We identified 38 pairs of variants that differed in only one cis-acting site (Supplementary Table S4). Half of these pairs were in the group of variants including AWCF, and these usually had quantitative differences in phage yield. In most of the other pairs, the two variants exhibited distinctly different phenotypes. Moreover, in evolution experiments, almost all mutations were in the lacO and promoter regions, not in protein-coding regions, even though the target sizes of protein-coding regions are much larger than those of cis-acting sites. Continued analysis of these variants, and extensions of our approach, should provide further insight into ways that evolution of cis-acting sites modifies the behavior of gene regulatory circuits.

Materials and methods

Media

Tryptone broth and LB were as described (Miller 1972) and were supplemented with antibiotics as appropriate, except that kanamycin was at 10 μg/mL. LBGM and LBMM were LB supplemented with 1 mM MgSO₄ and 0.2% glucose or 0.2% maltose, respectively. TMG was 10 mM Tris-HCl at pH 8.0, 10 mM MgSO₄, and 10 μg/mL gelatin.

Phage and bacterial strains

Phage strain λJL351 was used as wild type (Michalowski et al. 2004). It differs slightly from wild-type λ in that its set point for UV induction is ∼6–7 J/m² instead of 10–12 for wild type. It contains a unique XhoI site early in cI. In λJL351, the Tn5 kan marker, followed by the strong trp a terminator, replaced λ DNA from position 23901 to 26818 in the b2 region. Its phenotype differs slightly from wild type, in that it is somewhat easier to induce with UV. λJL387 was λJL351 with v1 and v3 mutations in O_R2 and O_R1, respectively, and was used as the cloning vector for the O_R region (Michalowski et al. 2004) as described in the Supplemental Material. Derivatives of λJL387 with various lacO alleles at P_L were made by two sequential phage-by-plasmid crosses. First, λJL387 was crossed with a plasmid that contained P_L and the v2 mutation in O_L1. Cross progeny were plated on strain JL5892 (JL2497 carrying pLR1 [Whipple et al. 1994], which provides a moderate level of CI), on which λvir (v2 v1 v3), but not λJL387, could plate. The resulting phage (λJL1000) formed small clear plaques; presence of v2 was verified by DNA sequencing. Second, λJL1000 was crossed with each of five plasmids carrying P_L and one of five lacO alleles. To distinguish recombinants carrying lacO in the P_L region from λJL1000, we plated phages on a 4:1 mixture of JL2497 and JL5895 (a derivative of JL2497 with F′::Tn3 replacing F′::Tn9, and carrying a plasmid, pA3B2 [Whipple et al. 1994], providing a low level of CI). On this mixed indicator, λJL1000 formed clear plaques, whereas v2⁺ recombinants formed turbid plaques. Presence of the lacO site was verified by DNA sequencing. DNA was made from each of five recombinants, using the Qiagen lambda DNA kit; equal amounts of DNA from each recombinant were pooled to provide the vector DNA. From this pooled DNA, left and right arms were purified as described (Michalowski et al. 2004), except that we used NsiI, which cuts just distal to cro, for the right arm; in addition, the left arm preparation was treated a second time with streptavidin beads, as described, to recover right arms, which were pooled with the right-arm preparation. Left arms were treated with shrimp alkaline phosphatase after right arms were removed for the second time.

Bacterial strain JL6142 Δ(lacIPOZYA) was derived in two steps from strain JL2497 (Little et al. 1999). First, a derivative lacking the F′::Tn9 was isolated by growth in the absence of chloramphenicol and screening among chloramphenicol-sensitive colonies for those not sensitive to M13. Second, lacI and the lac operon were deleted as described (Datsenko and Wanner 2000). Strain JL7223 was an hfl^- version of JL6142 made by transduction with JL6039 hflA150 (linked to Tn10) as donor, selecting for tetracycline resistance and screening for the Hfl phenotype as described (Michalowski et al. 2004).

Screening of λ lacI

Phages of this library (made as described in Supplemental Material) were plated on JL6142 on tryptone plates without or with IPTG at 10^-5 M or 10^-4 M. Turbid or small plaques were purified and analyzed by PCR; isolates with a fragment of the expected size were further characterized. Segments of the λlacIs were amplified by PCR of plaques. PCR products (P_L-O_L and CI-P_RM-O_R-P_R-lacI-λ NsiI) were sequenced by automated cycle sequencing at the Division of Biotechnology, University of Arizona. To test whether recombinants existed that formed clear plaques, we analyzed 20 clear plaques by PCR, and none had an insert; an additional 200 clear plaques were tested on a mixture of JL2497 and JL5434 (Michalowski et al. 2004), a mixture on which recombinants and vector form turbid and clear plaques, respectively; all isolates formed clear plaques, suggesting that recombinants forming clear plaques on JL6142, if present, were very rare.

UV dose responses for prophage induction

This was carried out as described (Michalowski et al. 2004). Stably lysogenizing variants were first screened at a dose of 40 J/m² (Supplementary Table S3). For selected variants, single lysogens were identified by a PCR test (Powell et al. 1994) and were further analyzed by irradiating with various doses at 0.5 J/m²/sec (Fig. 2).

Lysogenization tests

To test for ability to lysogenize, we centrifuged JL6142 grown in LBMM to 2 × 10⁸/mL, concentrated it 10-fold in TMG, and mixed it with phage at an m.o.i of ∼5 (a condition that favors the lysogenic pathway); after 10 min at room temperature, aliquots were diluted into LBGM containing 0, 10^-5, or 10^-4 M IPTG, shaken 30 min at 37° C, and spread onto with tryptone plates containing 10 μg/mL kanamycin and the corresponding amount of IPTG (results given in Table 1; Supplementary Table S3). To test for lysogenization frequency after single infection (Table 2), we grew cells as described earlier, infected them at a multiplicity of 0.01, diluted and incubated them as described earlier, and plated them in top agar with or without IPTG in top and bottom agar as indicated.

Isolation of secondary variants with altered phenotypes

Multiple plate stocks of λlacI variants were made on JL6142 from single plaques, using fresh tryptone plates and 6 mL of 0.35% top agar. Stocks were plated, and variants of particular phenotypes were chosen. Segments of these variants were amplified by PCR of plaques. PCR products (P_L-O_L and CI-P_RM-O_R-P_R-lacI-λ NsiI) were sequenced.