Several basic features of gene regulation have been established in the Molecular Biology Department of the Pasteur Institute (Jacob et al., 1960). The E. coli lactose operon has provided a model for these studies. The three structural genes of the lac operon synthesize the enzymes in charge of lactose utilization in the bacterium (Figure 1A). The beta-galactosidase catalyses the breakdown of lactose into glucose and galactose. It also catalyses the isomerization of lactose into allolactose. The permease is needed for the transport of lactose from the growth medium to the cell. The transacetylase transfers an acetyl group to some beta-galactosides, a function which remains obscure for lactose catabolism. The three genes are expressed and controlled by the same elements from a unique promoter region. In the presence of glucose in the cell environment, the repressor protein acting from an operator site prevents gene expression. In 1984, only the O1 operator located on the promoter region is thought to be functional. The two other operators, O2 and O3, with sequence homology to O1, are regarded as evolution left-overs. Lactose is only degraded if glucose is lacking in the growth medium. In this case, lactose derivatives (allolactose or isopropyl-thio-beta-D-galactoside, IPTG) weaken repressor binding to DNA. The CAP is a sensor of the glucose level in the medium. When repression is removed, the expression of the three genes can be switched on by the standard E. coli RNA polymerase (sigma70-RNA polymerase) and the adjacent CAP activator from the lac operon promoter.

A remaining question in 1984 is the question of how RNA polymerase is precisely activated by the adjacent CAP activator at the lac promoter. For some groups, CAP contiguity provides direct and specific protein-protein contacts for RNA polymerase activation (Blazy et al., 1980). More frequently, these contacts are thought to be fortuitous and unspecific. (1) In fact, extensive CAP mutagenesis to disrupt any contact with RNA polymerase is unsuccessful. One has to wait for the ulterior finding that a unique contact between CAP and polymerase is incredibly responsible for activation (Irwin and Ptashne, 1987). (2) DNA is rightly considered to be rigid when the activator and polymerase are close to contiguity. Because two CAP molecules are often present at the E. coli promoters, two molecules might relay activation (Irwin and Ptashne, 1987). (3) CAP has been shown by Crothers’ group to bend DNA by the circular permutation assay of gel electrophoretic mobility (Wu and Crothers, 1984). This local and protein-induced curvature might indirectly increase the affinity of RNA polymerase for its promoter downstream (see also Miyada et al., 1984). (4) In case of the lac control region specifically, CAP is also supposed to act indirectly by preventing RNA polymerase from binding at an unproductive promoter which overlaps the productive one (see Amouyal and Buc, 1987 for example).

The CAP activator locally bends DNA, as evidenced by electrophoretic gel retardation assays with DNA fragments carrying the CAP site at different places (Wu and Crothers, 1984). To confirm this induced curvature, DNA cyclization experiments are planned with the same fragments in the presence of CAP (Crothers et al., 1992).

Crothers’ planned study is based on various works between 1981 and 1984 dealing with spontaneous DNA cyclization (see for example Shore and Baldwin, 1983; Horowitz and Wang, 1984, Figures 2A and 3). These studies have shown that as a large chain of polymerized nucleotides, DNA shares similar properties with its chemical analogs (Jacobsen and Stockmayer, 1950). In this respect, DNA is characterized by a bending flexibility, a torsional rigidity and a persistence length (Figure 3).

In 1984, starting molecular biology (see short biography in supplementary material), I am immediately interested in proving a specific contact between RNA polymerase and CAP on the model of Crothers’ planned work with the sole CAP activator. If the contact for activation is specific instead of being incidental, DNA cyclization of a fragment carrying the CAP site and the promoter at each end of the fragment must be accelerated (Figure 2B).

On the other hand, due to DNA bending flexibility, the spontaneous cyclization of a DNA fragment is the most probable over the DNA persistence length, when its size lies between 150 and 2000 pb, and is optimal for 400 bp (Figure 3A, Shore and Baldwin, 1983; Horowitz and Wang, 1984 for example). The concentrations of DNA must be weak enough to prevent DNA aggregation. These conditions could also describe the best conditions for cyclization of a DNA fragment induced by RNA polymerase and CAP interaction, as probed by DNA ligation (Figure 2B).

In this light, the activation by CAP through local DNA bending is turned into an activation by a protein-protein contact, a naturally contiguous contact of activation is replaced by an artificially remote contact from two clearly separated CAP and polymerase sites (Figure 2B).

The physical interaction was later shown by means of these same DNA cyclization/ligation assays without separating the CAP and promoter sites and by placing them in the middle of the DNA fragment (Crothers et al., 1992), as well as genetically (Irwin and Ptashne, 1987). A genetic system was also later built to artificially generate activation by the CAP activator at some distance of a lac promoter and gene (work by Koop and Bourgeois, 1991, personal communication by S. Bourgeois). These constructs were not productive: CAP only activates E. coli RNA polymerase from its natural adjacent locations on the promoter. In fact, proximal activators are not suited for remote activation because they do not make the same contacts of activation with the transcriptional complex of initiation (Seipel et al., 1992). Thus, when E. coli cells are deprived of nitrogen, the paired NtrC activator is phosphorylated and RNA polymerase replaces the standard sigma-70 component by the sigma-54 component to make enhancer-like contacts.

In 1984, these elements have been newly unraveled by two groups (Banerji et al., 1981; Moreau et al., 1981). In other words, eukaryotic genes are found to be regulated by sequences and proteins located far from the gene on the chromosome. They activate gene expression independently of their orientation and distance with respect to the gene, which will become their definition (Yaniv, 1984).

In 1985, another set of data broadens the prospect of the initial project. Muller-Hill’s group has just found that the lac repressor is able to repress beta-galactosidase from two synthetic operator sites which surround the promoter and are separated by 220 bp in an artificial system (see next section, Besse et al., 1986). The nucleoprotein complex which is formed between the fragment carrying the 220 bp-spaced operators and the lac repressor, remains to be identified.

This artificial lac repression system has several advantages. (1) The genetic constructs carry high affinity sites, the so-called “ideal” lac operators, which avoids T4 DNA ligation. (2) lac Repressor is made of the aggregation of two dimers. This permanent association may advantageously replace the transient protein–protein association of CAP and RNA polymerase in the assays (Figure 2C).

Gel retardation is the technique which is selected in the first place. The assays are expected to reveal an abnormal electrophoretic mobility of any species different from one operator simply bound to the repressor on the same fragment. Consequently, all species possibly formed between a fragment carrying two lac operators and the lac repressor are expected to be discriminated from this simple complex, as well as between them in the gel: repressor-induced DNA loops, simple repressor binding to one operator, tandem repressor binding to two operators, DNA aggregates as well.

Indeed, this very first step was sufficient to show that the lac repressor could easily induce DNA looping when the two operators were separated by distances close to the 150 bp DNA persistence length and up to 220 bp. This finding was made directly readable in electrophoretic gels by Krämer et al. (1986, 1987).

Nevertheless, enhancers function at much larger distances, whereas large loops are too much retarded in the gels to be easily detected by this technique. Electron microscopy (em) is in fact better suited for the observation of large loops. The optimal size of a DNA fragment for its spontaneous cyclization is 400 bp (Figure 3A). When applied to the genomic cyclization induced by the lac repressor, this property implies a 400 bp spacing between the two operator sites. In this case, em grids are quantitatively covered by a field of loops when the preparation is that of the gel retardation assays^∗. The first tested spacing between operators was actually larger, 535 bp. This spacing was unproductive in the Besse et al. (1986) work in vivo because of concentration effects (next section), but is closer to the theoretical 400 bp value than the in vivo productive 220 bp distance. For this optimal spacing, as standard DNA spontaneously cyclizes, it is important to check for artifacts. For example, internal and spontaneous 400 bp coils are naturally observed on em grids within long DNA fragments in the absence of protein. In an ulterior set of assays, larger operator–operator distances were also successful, up to 2000 bp. em also confirmed the electrophoretic DNA loop formation below 220 bp.

This very first work with linear bacterial DNA (Krämer et al., 1986, 1987) and the studies that immediately followed, with natural operators as well as artificial ones, showed that a transcription factor (depicted by the E. coli lac repressor) could simultaneously interact with two distant regions responsible for gene regulation (depicted by the two E. coli lac operators), generating genome folding. This finding was reproducible, quantitative and consistent with the properties initially defining eukaryotic enhancers.

The DNA cyclization model revealed a supplementary condition of feasibility which was not expected from the comparison with eukaryotic enhancers: the concentrations of transcription factor and genomic fragment had to be low enough in vitro to avoid saturating the genomic sites with protein or the formation of DNA-protein aggregates (Amouyal, 2007a).

^∗The 400 bp optimal distance for enhancer-induced genomic cyclization was also used to evidence DNA looping by em with the mammalian estrogen receptors, as well as with the NtrC regulatory component of the E. coli ammonia deprivation system (Su et al., 1990), as both suggested by the author.

The in vitro studies had shown that repressor-induced DNA looping was favored by low concentrations of operator sites and repressor. Operator and repressor concentrations in the cell are also low. In the wild-type situation, there is only one copy of lac operon when it is carried by the chromosome and only 10 copies of lac repressor per cell. When the genetic studies were performed as close as possible to these natural conditions, the two operators O2 and O3 which are not located on the promoter, were found to strengthen repression issued from the proximal operator O1 by 72-fold. When strains overproducing the repressor were used, the contribution of the distant sites was nearly erased in the cells (Oehler et al., 1990, 1994; Amouyal and von Wilcken-Bergmann, 1992). These conditions were frequently used in the past with the rationale behind that in order to see the contribution of distant sites, one had to make sure that the repressor is effectively bound to the sites. However, these concentrations saturate the sites and preclude DNA looping. As a result, the contribution of the non-promoter lac operator sites O2 and O3 was skipped or underestimated (see Besse et al., 1986, for example). For technical reasons of commodity, multi-copy plasmids are also often used with the same risk of masking the contribution of distant sites to regulation.

With O3, the lac operon shares a common feature with eukaryotic gene regulation. The affinity of this sequence for the repressor is so weak that repressor binding cannot be detected by classical techniques in vitro. Though O3 had been shown to interact in vivo, it was not expected to be an auxiliary of repression (Borowiec et al., 1987). In fact, the contribution of O3 to beta-galactosidase repression is far from being modest. O3 contributes to repression through DNA looping. (1) Repressor binding in vivo only occurs in the presence of the proximal operator O1 (Borowiec et al., 1987). (2) A dimeric repressor unable to tetramerize has the same effect as a tetrameric repressor at O1 alone (Oehler et al., 1990). (3) A high level of cooperative repression is obtained from two O3 or even weaker sites (Amouyal and von Wilcken-Bergmann, 1992; Oehler et al., 1994). (4) The CAP activator, absent from some artificial constructs, cannot be responsible for this repression by interacting with repressor at (or near) O3 (Amouyal and von Wilcken-Bergmann, 1992; Perros and Steitz, 1996 and the whole discussion related to repressor-operator crystallization in Science 1996). (4) Repressor sliding along genomic DNA is excluded (Amouyal and von Wilcken-Bergmann, 1992).

In the wild-type situation, O3 is nearly as efficient as O2 in strengthening O1 repression (25-fold for O3, 40-fold for O2, Oehler et al., 1990). In an artificial system and in cooperation with itself (two O3 operators), O3 is able to repress beta-galactosidase as efficiently as O1, up to 35-fold (Amouyal and von Wilcken-Bergmann, 1992). So that whether O1 interacts with O3, or with O2, beta-galactosidase is very efficiently repressed through DNA looping (440 and 700-fold, respectively, Oehler et al., 1990).

If the strong operator O1 replaces the weakest operators O2 or O3, repression is even more efficient, but inducibility is totally or partially lost (Amouyal and von Wilcken-Bergmann, 1992; Oehler et al., 1994). The level of repression has to be determined by comparison with an identical strain which lacks the repressor, instead of the inducer. Modulation of DNA binding by gene-specific transcription factors is a crucial element and some aspects still have to be cleared (Schleif, 2013). Replacing the weak operators by the strong one has qualitatively the same effect (high level of repression) as replacing the repressor by a mutant superrepressor which binds tightly DNA (Willson et al., 1964; Bourgeois and Jobe, 1970). Yet, these superrepressors have not been naturally selected, nor a set of strong operators. Instead, a system of weak operators has been naturally selected. Then, if DNA looping is necessary to have efficient repression, low affinity sites are required to keep the operon inducible.

In the wild-type situation, alternative DNA looping involving either the O1/O2 or O1/O3 pair of operators, increases the probability of obtaining a high level of repression from the proximal operator O1 in each cell (Oehler et al., 1990). However, if the O1–O2 loop strengthens repression at the promoter, it also consolidates beta-galactosidase repression from O2 by blocking any RNA elongation at O2 (Flashner and Gralla, 1988). DNA loop formation is required for this process, since O2 alone is unable to stop RNA elongation. The O1–O3 loop may play a similar role.

The O3 operator is located at the extreme end of the laci gene (Figures 1A,C). The view that the laci gene is not regulated, is anterior to the finding that O3, in spite of its low affinity for the repressor, is involved in repression of the lac operon. However, the O1–O3 loop might contribute to self-repression of the repressor simultaneously to repression of the operon. This would not be surprising since several E. coli repressors are autorepressed, as reviewed by Collado-Vides et al. (1991). Furthermore, Selliti et al. (1987) have noticed that several transcripts are synthesized from the laci gene and that RNA elongation is blocked within the control region of the lac operon, or before. In particular, one of these transcripts ends 5 bp upstream of the lacO3 operator, within the laci gene and before the lac control region, in the presence of repressor. This truncated transcript is less stable than the transcript giving rise to the lac repressor, according to Selliti et al. (1987).

If the repressor simultaneously bound to O3 and O1 (since O3 alone is unable to do so) stops RNA elongation, like O2 through O1–O2 DNA looping, this would not be the main process of limiting lac repressor production. This production is already limited, at the overall level of 10 molecules per cell, by a weak promoter (Müller-Hill et al., 1968; Figure 1B). This is confirmed by mutations in the promoter region that increase repressor production. For example, a particular laci^Q mutant produces about 10-fold more repressor (Müller-Hill et al., 1968) thanks to a single CG →TA base pair mutation located at -35 of the laci promoter (Calos, 1978). A 15 bp deletion in the promoter region leads to a 50–100-fold increase of lac repressor synthesis, creating an improved –35 homology region and a strong promoter (Calos and Miller, 1981).

Another element has to be taken into account: the gene producing the repressor is tied to the lac operon (Figure 1A). In most other E. coli operons, the gal operon or the deo operon for example, respectively, responsible for galactose and for deoxynucleoside catabolism, the repressor is not produced in close proximity to the promoter. Thus the apparent concentration of lac repressor available for gene regulation is higher than the 10 molecules actually measured. In this context, if a weak promoter is necessary to produce the first molecules required for repression and to limit this production, the O1–O3 loop is necessary to stop it after the initial burst of production.

Genes inserted into host genomes are deregulated, are not expressed in the desired cell type, or can deregulate host genes because of chromosomal position and undesired enhancer activity. Some sequences and proteins seem to partially protect them, naturally or artificially, from the neighboring environment. This is the case for the HS4 sequences, CTCF and cohesin proteins of the chicken beta-globin locus or of the scs/scs’ sequences of the heat-shock HSP70 locus in Drosophila.

The HS4 or scs/scs’ sequences delimit a transcriptional unit on the chromosome. Two operator sequences also surround the lac repressor-operator DNA loop and delimit a topological domain (Krämer et al., 1988). By analogy, the lac operator sequences might also act as gene insulators through DNA looping in eukaryotic cells producing the lac repressor (D. Bienvenu, B. Pineau and M. Amouyal, unpublished results, 1999–2000; Amouyal, 2010b). This was a way to postulate that gene insulators also operate through genomic looping. It now appears that insulators often provide an architectural prop for enhancer action through DNA looping when they do not simultaneously possess the capacity for both DNA looping and the enzymatic enhancer activity (Mongelard and Corces, 2001; Amouyal, 2010a,b, Amouyal et al., 2011 on Gene insulation, Amouyal et al., 2013). In some instances, they also stop RNA elongation, indirectly blocking heterochromatin propagation, like DNA looping to which it can be associated. Again, this dual mechanism is initially a property of E. coli transcription factors involved in DNA looping, such as the E. coli lac repressor.