CONTRIBUTIONS TO DNA and RNA POLYMERASE KINETICS
( included in the web site http://www.lps.ens.fr/~ninio )
TOPICS DISCUSSED HERE:
SEE IN OTHER WEB CHAPTERS
More on the next
effect (Contributions to the kinetic theory of accuracy).
The peelback effect (Contributions to mathematical biology).
Mutation strategies (in molecular evolution section).
Non-enzymatic replication (Contributions to the origins of life).
The existence of a proofreading activity associated with E. coli or phage T4 DNA polymerase was revealed in 1972 [1, 2]. I was not immediately aware of this work, and was mostly concerned with ribosomal accuracy, trying to interpret Gorini’s observations on high and low accuracy ribosomes in terms of changes in kinetic parameters of the amino acid incorporation reaction. When this work was sufficiently advanced, and I had worked out the (rather elementary) mathematics needed to understand how changes in the kinetic parameters of an enzyme reaction impinged on its accuracy , I turned to DNA polymerases. In the beginning, I spent much time on two articles from Kornberg’s laboratory. The first  gave the affinities of Kornberg’s polymerase for the four deoxynucleotide triphosphates in the absence of template. The natural preferences were ranked as follows : G > A > T > C . I was thinking of using these results in relation to base pair energetics (the DNA polymerase was perhaps using these differential binding affinities for counteracting some potential errors) or in relation to the nucleotide substitution tables derived from phylogenetic studies. I failed to derive any useful insight from this work. The second was a very astute, historically important article, in which Josse, Kaiser and Kornberg  made an analysis of the nearest neighbours, in the DNA strand being replicated, of each of the four incorporated nucleotides. I hoped to use these nearest-neighbour data to have a better understanding of replication energetics, but here also, I failed to derive any useful insight. When the first papers on the proofreading function appeared, I remember that there was some disagreement between Hershfield  who was claiming that the DNA polymerase fidelity mutants were essentially altered at the incorporation level and Bessman  who spoke of the ratio between the incorporation and the excision activties of the DNA polymerase. From my own experience with ribosome kinetics  I knew that accuracy was governed by ratio of rates, and that distinguishing between forward and backward effects was not an easy task, so I did not have a preference for incorporation or excision effects.
Another article fell into my hands, in this period (I was in Orgel’s laboratory at the Salk Institute), and this one did inspire me. In this article, Hall and Lehman described puzzling observations . The authors were studying T/G misincorporation errors made by a standard and a mutagenic DNA polymerase from phage T4 in the in vitro replication of Poly(dC). Surprisingly, they found that they could not dilute the T misincorporations by increasing the concentration of dGTP at constant TTP. Instead, the T/G ratio seemed to reach an asymptotic value. This odd result was not even mentioned in the rather inocuous summary of their article. The article had been published in 1968, at a time in which the proofreading function was still unknown. Reading this article in 1973, I almost immediately saw that Hall and Lehman’s results had to do with the balance between nucleotide incorporation and excision at various nucleotide concentrations. It was childplay to transpose my ribosome kinetic analysis  to the incorporation/excision balance.
I thus derived the « next nucleotide effect » according to which incorporation errors at any position would increase with the concentration of the nucleotide to be incorporated next. I did not hurry to publish it because on one hand it seemed to be a modest step forward with respect to , and on the other hand I had the hope that my colleagues in Chapeville’s laboratory in Paris would perform the experiments, and prove the phenomenon. I also found in a work by Howard and Tessman  indirect evidence for the next nucleotide effect in vivo. (With modern technologies, the Howard and Tessman kind of experiment may produce superb results). I wrote a note, in French, entitled « commentaires sur une expérience de Hall et Lehman » in which I made a mathematical analysis of Hall and Lehman’s data, and sent the note to Gilbert Brun, in Chapeville’s laboratory. The note also contained kinetic calculations on what I expected when misincorpotation was due to a tautomery. (In this case, proofreading becomes useful only after the return to the standard conformation of the nucleotides facing each other, giving rise to a quadratic effect in error correction). However, either my colleagues did not understand what was at stake, or they were not interested enough in the subject of DNA polymerase accuracy, so nothing came out, experimentally, from their hands.
Ultimately, I mentioned the next-nucleotide effect in various places. For instance, it is mentioned as follows, in , page 593 :
« Discrimination being related to the time which separates two incorporations, it should depend on the concentration of the dNTP’s. More precisely, the errors of incorporation at any position along the chain would increase with increasing concentrations of that dNTP which is to be incorporated at the next position. Such an effect can in principle be revealed in classically designed experiments [29, 30], »
References 29 and 30, in the above paragraph were to Hall and Lehman’s and to Howard and Tessman’s papers [7, 8]. I did not say more, because this effect was quite obvious, within the context of the more sophisticated preceding developments in the same article. On this topic, see also the website chapter on « Contributions to the kinetic theory of accuracy ».
At some point during the Sept 72 to March 74 period at the Salk Institute, Leslie Orgel was contacted by David Galas, who was then working with Elbert Branscomb on translation error propagation in E. coli. I do not remember whether or not he visited the lab in person, but this was the beginning of a correspondence between David and I, on theoretical treatments of error-feedback, and later on DNA polymerase kinetics (see the website chapter « contributions to the kinetic theory of accuracy »).
My article on ribosome kinetics  was submitted in July 1973 and after that, I must have started trying to go one step further and construct kinetic amplification mechanisms. After the sparkle, the very last days of December 1973, I must have tried to extend the concept of kinetic amplification to the workings of a classical proofreading function. As I mentioned above, I had already in mind the notion that it was useful to wait, in the case of tautomeries, before exerting the proofreading activity. I soon found that in the standard case, the shape of the time-window allocated to proofreading was extremeyl important, and that what really mattered was how it ended. More precisely, I found that the substrate (the base-pair being proofread) must not be allowed to escape from the proofreading function in a single elementary step, but must be held under attack for at least two steps. This was my « delayed-escape » mechanism, which has not yet received much attention, although I think that there is much evidence in its favour in the case of aminoacyl-tRNA ligases. The experiments made in my group on polymerization/excision kinetics were for some homopolymers consistent with a delayed escape mechanism, and for others they did not show the kinetic complexity expected from these mechanisms (see the section below on polymerization/excision kinetics)
Upon my return in France, in April 74, I tried to initiate work on catalysis by randomly generated peptides (see the website chapter « contributions to the origins of life »). At the end of 1974, I read John Hopfield’s article on kinetic proofreading  and saw immediately that when applied to DNA or RNA polymerases, his scheme made predictions which were radically different from my expectations, based on either the next nucleotide effect, or the delayed escape mechanisms. I must stress again the point that although there is a striking convergence between Hopfield’s proposals  and mine, there are also differences, and these become rather important, when one tries to confront the predictions with experimental reality. (See for instance my recent analysis of the GTPase reactions on the ribosome ). These differences have been obscured, in part because John Hopfield chose to make his arguments general, and incorporated classical proofreading in his conception of kinetic proofreading. On my side, I chose to make a sharp distinction between classical and kinetic proofreading. Classical proofreading is what we learnt from the studies of Kornberg, Bessman and others, e.g., [1, 2, 6]. After making the internucleotide bond, a DNA polymerase may excise the last nucleotide it had incorporated. Now, there was a new idea in Hopfield’s article which was not in my article, namely that in the incorporation pathway, the dNTP could be cut into dNMP and pyrophosphate, and the dNMP could be rejected BEFORE having a chance to be linked to the preceding nucleotide. Then, dNMP would be produced in proportion to the corresponding dNTP, and would not vary with the concentration of the next nucleotide.
This time, I convinced my colleagues in Franćois Chapeville’s laboratory to look seriously at the polymerization and excision kinetics. Kinetic experiments were performed. Thin layer chromatography was used to separate the various products of the reactions. It turned out that nucleoside diphosphates were produced, in direct proportion to the nucleoside triphosphates. So, I thought, we could have here a variant of Hopfield’s scheme, in which the incoming dNTP would be split, prior to incorporation, into dNDP and phosphate . The work was received with skepticism. I now believe that these cleavage reactions are biologically relevant companions of the polymerase activities, and that they have a role in accuracy, but not the one I had imagined (See section below on the NTP or dNTP cleavage reactions).
After the NDP and dNDP episode, Franćoise Bernardi and I returned to the quantification of the dNMP produced during polymerization reactions, and publlished in fact the first kinetic experiments showing how the polymerization/excision ratio varied with the concentration of the correct nucleotide . (See further down the section on polymerization/excision kinetics). The work was pursued by Mohammed Saghi and Mireille Dorizzi, then took a new direction with the arrival of Catherine Papanicolaou. We had found strange variations in the kinetics with the concentrations of the reactants, and also with the concentration of the catalyst, the DNA polymerase itself. This led to a series of articles, initiated in  in which we demonstrated that E. coli DNA polymerase I had a mnemonic behaviour. This was both interesting and frustrating, because the mnemonic behaviour made it very difficult to establish the mechanism of the reaction at the level of detail which would have satisfied me. (See section below on mnemonic effects and exonuclease warm-up).
After her thesis, Catherine Papanicolaou went to the USA and worked with Lynn Ripley. Philippe Lecomte, who had collaborated with her on the mnemonic effects was progressively discouraged, and our last experimental paper on the topic was published in 1988. Nevertheless, Tom Kunkel and Roel Schaaper were kind enough to invite me to a meeting they were organizing in North Carolina in September 1989, on « The fidelity of DNA synthesis : structural and mechanistic perspectives ». By then, DNA polymerase crystals had been obtained and interpreted in the laboratory of Tom Steitz , and fast kinetic experiments were being performed in Stephen Benkovic’s laboratory . There were two major findings which I had not anticipated. First, until then, I had conceptualized the bifunctional polymerases as having a head which incorporates nucleotides, and a proofreading function located behind. However, the crystal structures demonstrated that the exonuclease activity was located ahead of the incorporation function. Accordingly, after the formation of the internucleotide link, the growing terminus and the portion of DNA being copied had to slide forward to reach the exonuclease site. These mechanistic details did not threaten the kinetic treatments leading for instance to the next-nucleotide effect, since these treatments were rather general, and were not tied to precise mechanistic details. On the other hand, it was also the case that I hade never taken into consideration this arrangement with an exonuclease activity ahead of the polymerase activity. I still blame myself for this lack of lucidity.
There was an enormous difference of scale, between my meagre group, and the American teams. Not only they had budgets which were orders of magnitude above mine, but also they had an almost unlimited access to student’s manpower, which was unthinkable under the constraints of the French system. Although I had some confidence in my capacity to produce astute work, I felt at this congress that we were entering an era in which scientific truths in biology could be negociated between big laboratories. So, there was no room for independent work. After publishing as a swan song my « transient mutator » hypothesis , I resigned from my position of group leader in molecular biology, and joined a team of physicists interested in cognitive sciences at Ecole Normale Supérieure.
NTP or dNTP cleavage reactions
In 1974, several members of Franćois Chapeville’s laboratory were working on DNA polymerases. The most active in the field, among them was Gilbert Brun who was mainly working on eukaryotic DNA polymerases. Franćois Rougeon was also there, and was working on reverse transcription and its biological implications. Franćoise Bernardi joined the lab and teamed with me in the summer, and she started a project on looking for catalytic activities in randomly synthesized peptides (see the website chapter of « contributions to the origins of life »). The publication of John Hopfield’s paper (which we received quite late, due to a postal strike) incited me to have a close look at DNA polymerase kinetics, and in particular, see whether the deoxnucleoside monophosphate was produced, in a classical way, through exonucleolytic proofreading, or as envisioned by Hopfield, through a splitting of the dNTP into dNMP and pyrophosphate, prior to the formation of the internucleotide linkage.
Franćoise Bernardi initiated the experiments, possibly in January 1975 (see the website chapter on « contributions to the kinetic theory of accuracy). We followed the kinetics by spotting each extract from the reaction mix on a thin-layer chromatography sheet, and separating the various products of the reaction. It turned out that, in addition to the appearance of dNMP, we found that substantial amounts of dNDP (nucleoside diphosphates) were produced during the reaction. I was very excited, because I thought that the dNDP could be produced according to a variant of Hopfield’s scheme, in which the dNTP was split into dNDP and phosphate prior to the formation of the internucleotide bond. The production of dNDP was observed with E . coli DNA polymerase I (and, I think, with one or two eukaryotic DNA polymerases), and the production of NDP was observed with E. coli RNA polymerase. The kinetics of appearance of these nucleoside diphosphates were in remarkable agreement with the predictions I had made on the basis of the modified Hopfield scheme. However Chapeville was not convinced by the arguments. He had in mind an enzymologist’s golden rule stating that enzymes which produced posphate from nucleotides were quite distinct from those which produced pyrophophate. A same enzyme could not produce both, as implied by my reaction scheme. Nevertheless, I was enthousiastic and convincing, and by March 21st, 1975 we submitted to Franćois Chambon acting as editor of FEBS Letters, a manuscript describing our work. The paper was accepted, after making small modifications. It appeared in the September 1975 issue of FEBS Letters .
The least I can say is that it did not convince other workers in the field. Soon afterwards, a Russian team described the separation of a NDP generating activity from the polymerising activity in RNA polymerase. This was a blow, but it left the door open for different reaction schemes in which NDP could have been a side product. Note that in eukaryotic DNA polymerases, the polymerising activity and the exonucleolytic proofreading activities are on different subunits of the enzme (e.g., ) so the separability of the two activities does not imply that they should be functionally unrelated. There was however another serious problem with our paper. I did not know, at the time of the work, about radioactivity quenching problems. The quantitative arguments which I had derived, in the case of RNA polymerase, from comparative measurements of tritium and P32 radioactivity did not take into account the differential quenching effects of the two radioactive labels. Once they were taken into account, much of the strength of the argument was removed, and alternative, less interesting interpretations could be found. I acknowledged this state of affairs, writing for instance in  :
« In an earlier work, we reported that both DNA and RNA polymerases were able to degrade dNTPs into dNDPs and phosphate (Ninio et al., 1975). Because of a counting artefact, the quantitative relationships given in the paper were not correct, so that the relationship between these activities and kinetic amplification is doubtful. With the progress in the chemical synthesis of DNAs of defined sequences, we hope to clarify the point in the near future. »
Remaining with RNA polymerases, the NDP producing activiy came to the front scene again when a low fidelity mutant of RNA polymerase was isolated . Jonathan Gallant and Randell Libby found that the genetic defect coincided with a loss of the activity converting NTP into NDP [21-23]. They assigned this activity to the b subunit of RNA polymerase. Libby came for a short period in my lab, in Paris, to perform some kinetic experiments, taking advantage of the competence of Philippe Lecomte in this domain. At that time, I was paying great attention to diffusion reactions, and to the implications of in vivo viscosity. Assume that an incorrect nucleotide has been rejected by a polymerase. Will it diffuse away immediately, or should we expect the nucleotide to hang around the binding site, and have several additional opportunities to be handled again by the polymerase at the same site ? I thus thought that one way to reduce the danger would be to « label » the expelled nucleotide, so that it could not be re-used immediately. My concern for the implications of viscosity was expressed in  :
« The passage from the highly viscous conditions of the cell’s interior to the liquid conditions of in vitro experiments may affect quite differently the various kinetic constants of a reaction. A polymerase will wait longer for the incoming nucleotide, but if it dissociates from the template, it might fail to diffuse away, and might reassociate readily, the more so as imprinting effects on the template may last longer. Side products of the reaction like pyrophosphate or deoxynucleoside monophosphate might hang around and be in abnormally high local concentrations, with respect to usual in vitro assays. Is this the reason why RNA polymerase carries a companion activity which eliminates pyrophosphate by an unusual method (Volloch, Rits and Tumerman, 1979) ? ».
The reaction described by Volloch et al. [24, 25] is the production of trimetaphosphate (PPPi) and phosphate from two pyrophosphate (PPi) molecules. There was also a report, by Hinrichsen et al.  of the synthesis of dinucleoside tetraphosphates by RNA polymerase B from calf thymus, according to the reaction pppN + pppN’ ą pppNpN’ + pp.
When Randell Libby came, I proposed to him to study the effect of viscosity. I believed that NDP formation had to do wih a labelling of the nucleotides which had been rejected. Unfortunately, there were technical problems with the preparations or the assays, which could not be solved within the time allocated for Libby’s visit, and since Libby left the subject afterwards, I do not know what to think about all that. The proposals of Libby and Gallant were challenged by Butzow et al.  who wrote :
« The greater cleavage with noncomplementary DNA occurs because of the lack of incorporation by the polymerase, which then does not compete with the NTPase for the substrate pool. Thus, these findings indicate that the cleavage preference of the NTPase for noncomplementary NTPs is not part of a mechanism for error prevention during transcription ».
The same interpretation applied to my early work , once the radioactivity quenching artefact was removed. On the other hand, while such interpretations are compatible with in vitro kinetics, they relegate a number of biological observations to the rank of mere oddities : (i) why should there be NTPase activities at all ? (ii) why these activities should be so hard to separate from polymerising activities ? (iii) what is the biochemical basis for the RNA polymerase accuracy mutants studied by Libby and Gallant ?
I consider that there might well be a link between the above mentioned « parasitic » activities and DNA or RNA polymerase fidelity, although the link is probably different from that considered originally . It could have to do, as I mentioned, with the clearing of the active centre of the polymerase from the undesired NTPs. I note in this respect that hepatitis B virus DNA synthesis occurs within a core from which PPi and other side products of the polymerization reaction may have difficulties to escape . The « parasitic » degradative reactions may also intervene in vivo, when a polymerase is stalled as a results of its previous actions, so the work must be undone [25, 29]. Stalling may also be due to nucleotide damage on the template, so it could be that some polymerases carry with them repair activities, which, in some cases, are known to generate nucleoside diphosphates .
I came out of the previous work on NDP and dNDP forming reactions with the feeling that these activities may after all be contaminants, and that they were not part of the standard incorporation pathway. So I focused on the less controversial topic of the laws of dNMP generation. Most of the work was carried out at the beginning, by Franćoise Bernardi, with E. coli DNA polymerase I or, more precisely, its Klenow fragment. This time, the radioactivity quenching corrections were carefully made. Several homopolymers were used as templates, and we followed the ratio of incorporated versus excised nucleotides, as the reaction mixture was being depeleted in substrate. In at least two systems, the polymerization/excision ratio varied as a function of the remaining substrate exactly as expected on the basis of the next-nucleotide effect . In at least one other system, the ratio was as expected on the basis of the « delayed-escape » mechanism. The results were briefly given in a review on DNA polymerase accuracy , then submitted in August 1977 in more complete form, with full mathematics to the Journal of Molecular Biology, where it was handled by Sydney Brenner. There was a favourable report, but accompanied with requests for experimental controls. At that time, Franćoise Bernardi was no longer in the laboratory, and the controls were carried out by Mohammed Saghi and Mireiille Dorizzi, who had taken her succession in the DNA polymerase work. In July 1978, I attended a FEBS meeting in Dresden. There, I met Alan Fersht. He told me that he was a reviewer, and that he found the manuscript « very nice ». However, our revised version, submitted later in July 1978 received a negative report from a reviewer who did not understand our mathematics (he was still living at the stone age of Michaelis kinetics). Fortunately, Sydney Brenner trusted me, and the article was ultimately published .
My proof of the next-nucleotide effect was complete. It showed how the misincorporation/excision ratio varied with the concentration of the correct nucleotide, when Poly(dC) was the template, dGTP the correct nucleotide, and dATP or dCTP the competing incorrect nucleotides (Figures 2 and 3 in , and Fig. 4, Fig. 5 in ). The next-nucleotide effect was subsequently demonstrated with more sophisticated experiments, using natural DNA templates and several DNA polymerases [32-34]. The effect is now considered as obvious in the field of DNA polymerase mechanisms. Rather than measuring polymerization/excision ratios, people usually just measure misincorporations as a function of the next nucleotide concentration. This is now being used as a test for detecting the existence of an exonuclease activity coupled to the incorporation activity of a polymerase (e.g., ). The work by Clayton et al.  is interesting, it shows so to speak how to transform a standard polymerase into an antimutator polymerase, by lowering the dNTP levels. Maurice Bessman could have shown this earlier , but failed to appreciate the next-nucleotide effect, due to an incorrect mathematical analysis of his data. The mathematical article  following  introduced though the interesting concept of the random walk of a DNA polymerase over a template, due to series of forward and backward steps.The solution of the simplest kinetic model was given in [9, 31, 38]. Solutions for more complex cases were given in [39-41].
After the departure of Franćoise Bernardi to follow her husband in Marseille, the DNA polymerase kinetics studies were interrupted until Mireille Dorizzi, an experienced enzymologist and Mohammed Saghi, a pre-doctoral fellow from Morocco took the subject again. The previous work had provided evidence for two families of kinetic curves. In all likelihood, there was a single kinetic mechanism, of which we were observing different limiting cases. So, it seemed to me important at this stage to establish the polymerization/excision kinetics under a wider range of substrate concentrations, with the hope that the extended kinetics would display the two types of behaviour in different concentration ranges. Mireille Dorizzi, on her side, wished to go down to earth and make better connections with classical enzymology. So, Saghi and Dorizzi made pH dependence and temperature dependence studies . They reported that « temperature influences excision more than inorporation » the ratio of the two activities increasing linearly with temperature. The pH dependency curves possibly reflected a ionization of the template in one case, and of the substrate in another. Although this was not stated in the summary, the polymerization/excision kinetics were now established with improved precision. A type of deviation from the simplest polymerization/excision kinetics, already noticed in  was furher substantiated. Here there was perhaps a clue on the mechanism of the reaction, but it did not fit with any known scheme.
Having completed and defended his thesis, Mohammed Saghi returned to his native country where he teaches molecular biology at the university of Rabat. The elucidation of the origin of the kinetic anomaly occupied me and my co-workers for quite a while. The work, described in the next section, was initiated by Catherine Papanicolaou.
The next nucleotide effect had one important practical consequence : replication is more accurate at low substrate concentration. The enzymologists were taking as an article of faith that all enzymes worked correctly only when the substrate concentration was above the Km (whatever Km means). Thus, all studies with substrates below Km were discarded as being unphysiological and irrelevant. The next-nucleotide effect forced the DNA polymerase enzymologists to abandon the stupid prejudice about the necessity of using substrates above Km. This was rather important at that time because so much DNA sequencing work was done with the PCR reactions, using too high nucleotide concentrations. Research scientists are often asked to prove that their work has practical applications. In my case, I can thus claim two contributions to biotechnology : the next-nucleotide effect, and the fast algorithm for searching homologies between sequences , now known under the name of BLAST and often attributed to others (See the web chapter of « Contributions to bio-informatics »).
Excellent work has been performed by others on the contributions to DNA replication accuracy of incorporation, excision, or pyrophosphorolysis (e.g., [44, 45], review in ). The kinetic mechanisms of DNA polymerases has also been dissected with great sophistication using fast kinetic methods, and taking advantage of the crystallographic breakthrough (e.g., review in ). I would like to add here a special mention to a very informative and lucid review by Beckman and Loeb  touching on polymerization and excision kinetics.
While elegant, probabilistic mathematical treatments go directly to the point in problems of accuracy, the field of accuracy is often spoiled by irrelevant stone age enzymological treatments which use Km and Vmax reasonings.
The Beckman and Loeb’s review, however, uses correct mathematics. As a matter of fact, Robert Beckman detected errors in my J. Mol. Biol. article . One was a printing confusion : (k3 + k4) in Eq(17) should have been (k-3 + k4). More importantly, Robert Beckman noted that my statement about the effect of pyrophosphorolysis (page 101, after Eq. (25)) was incorrect.
The more general issue there was about polymerization/excision kinetics that did not agree with the simplest next nucleotide effect. As observed by several authors (see review in ) the exonuclease/incorporation ratio reaches an asymptote at increasingly high dNTP concentrations, it does not decrease to zero as predicted by the simplest next nucleotide effect. The observed behaviour, on the other hand, was predicted by the « delayed-escape » model which proposes that the exonuclease activity extends over two distinct kinetic steps. A similar behaviour can be predicted if there is some production of dNMP prior to nucleotide incorporation, according to Hopfield’s kinetic proofreading model PLUS the classical exonuclease. Kunkel Beckman and Loeb studied the effect of pyrophospahte . Putting all the evidence together, Beckman and Loeb concluded  :
« Although five proofreading models predict residual proofreading after the next nucleotide effect is saturated, only one, the modified two-step proofreading model (Bernardi et al., 1979), is in accord with the powerful functional constraints provided by the study of combination of proofreading probes (Kunkel et al., 1986). ».
In this paragraph, Bernardi et al. corresponds here to , Kunkel et al. corresponds here to  and the « modified two-step proofreading model » is the combined Hopfield’s kinetic proofreading + classical exonuclease model. I would be more cautious on the implications of the PPi data. Pyrophosphate may act as a standard product of the incorporation reaction, and it may also act as a modifier of the enzyme, and it may also be the substrate responsible for a delayed-escape effect as proposed by Kuchta et al. . To settle matters, it would be useful to add to the PPi experiments, studies which test more directly possible successive stages in dNMP production. In particular, it would be useful to develop a system in which a nucleotide is misincorporated due to a transient conformation (e.g., a tautomeric shift) and proofread efficiently after switching to the standard conformation.
Mnemonic effects and exonuclease warm-up
In addition to the kinetic anomaly at high dNTP (non-saturation of the exonuclease), we had also observed a systematic kinetic anomaly in the low substrate concentration range which could not be simply explained by experimental uncertainties. There seemed to be a deficit in the exonuclease function at low concentration of substrate [31, 42].
The place left vacant in my team by Mohammed Saghi was soon taken by a pre-doctoral student, Catherine Papanicolaou. She worked in parallel on two themes : (i) energy models for the prediction of RNA secondary structure [49, 50] – see the web chapter « contributions to bio-informatics » and (ii) the DNA polymerase kinetic anomaly at low substrate.
The anomaly could be interpreted as reflecting an allosteric effect, but the DNA polymerase is a monomeric effect. It had been realized quite early [51, 52] that apparent allosteric behaviour could in fact be the reflection of hysteretic effects. The idea is that a monomeric enzyme may exist under two conformations, one with high and one with low activity towards the substrate. Right after a productive interaction with the substrate, the enzyme is in the state in which it is competent for a further interaction with another molecule of substrate. However, at low concentration of substrate, as time elapses, the enzyme has increasing chances of switching to the state in which it is incompetent for substrate binding. This is how an apparent « positive cooperativity » is created (e.g., reviews in [53-55]).
In the case of E. coli DNA polymerase I, all our experiments turned out to be consistent with a mnemonic mechanism. More precisely, the crucial factor was the interval of time between a dissociation of the DNA polymerase from the template, and its subsequent binding to another molecule of template [14, 56]. After her thesis, Catherine Papanicolaou moved to Lynn Ripley’s laboratory in the USA, and the work was pursued by the late Philippe Lecomte. If our interpretation of mnemonic kinetics was correct, a DNA polymerase could bind to the template with its exonuclease in a relatively inactive state, and the exonuclease would become more and more active as the DNA polymerase would progress along the template. We thus undertook the task of determining how the nucleotide misincorporations varied according to the number of elongation steps. This was a rather high-tech experiment with respect to the state of the art in the field. We still worked, it is true, with homopolymers. But the DNA chains which were elongated were separated by gel electrophoresis, allowing us to determine the misincorporation ratio at each position. Philipe Lecomte was doing all the wet biochemistry, and I took the responsibility of the quantification of the radioactivity in the gel electrophoresis bands. Here, I will relate an instructive episode.
In order to quantify the error levels in each band, we were making autoradiograms of the gels. We had bought an expensive densitometer which was scanning the autoradiograms through a narrow linear slit, giving a profile of the density accross the bands. It was also supposed to make background subtraction and provide the integrated values for the denisty of each band. However, I soon realized that the numerical figures given by the high tech densitometer could not be correct. Instead of separating the bands in terms of slightly overlapping gaussian curves, the algorithm used by the densitometer consisted of going from one local minimum to the next, and take the surface of the peaks above the segment joining two consecutive local minima. There followed a long and disagreable correspondence with the manufacturer of the densiometer. There seemed to be no way to have access to the original (uncorrected) data. It would have been an easy task for me to retrieve the correct densities from the original data. At last, the manufactuer stated that the method which they had implemented to integrate the band densities had been demanded by the physicians, because this was the way the physicians thought it should be done !!! In any event there was nothing they could do at this stage (but it seems that in later versions of the densitometer, they proposed as an option, the access to the raw data). So I decomposed manually all the density profiles into gaussians, and mesaured all surfaces manually on the rolls of paper coming out of the densitometer. We thus showed very clearly, with the Poly(dA).oligo(dT) system that the misincorporation level of A was rather high at the +1 position, and it decreased rapidly, from the 1st to the 25th position . The work was summarized as follows :
A.A mismatch errors occurring during poly(dA) replication with the Klenow fragment of E. coli DNA polymerase I have been quantified. The A/T ratio measured for chains extended by 1-25 nucleotides decreases by a factor of at least 15 from beginning to end. The deduced true error rate may decrease by a factor 2.5 at each successive nucleotide addition. When ddATP is used instead of dATP, the ddA/T ratio indicates little variation of the misincorporation probability with position. Thus, the accuracy improvement in the first case is due to warm-up of the proofreading function.
On this account it seemed that the exonuclease activity of E. coli DNA polymerase I was nearly inactive following the first incorporation act. However, strangely enough, it seemed that a substantial fraction of the initial oligomers had been trimmed. This feature was elucidated in a subsequent work. At the end, we reached an understanding of the system described as follows in the summary of our last paper  :
Escherichia coli DNA polymerase I exists in at least two distinct kinetic forms. When it binds to a template, the proofreading activity is usually switched off. As the enzyme progresses along the template, it becomes more and more competent for excision. This phenomenon introduces a link between fidelity and processivity. Processivity is best studied when the chain length distributions of synthesized polymers are stationary. Even then, however, one cannot avoid multiple initiations on a given template by a same molecule of the enzyme. When synthesis is initiated with primer lengths 15 or 20, a strange phenomenon is observed. It seems that the polymerase starts by hydroyzing the primer down to a length of 7-10 nucleotides, and only then starts to add nucleotides. It does so in low-accuracy mode, suggesting that while the exonuclease is clearly active, it does not contribute to proofreading. The warm-up of the proofreading function is therefore reinterpreted as a switch between two modes of behaviour : a mode 1 of low accuracy in which the 3’ ą 5’ exonuclease, while active, is uncoupled from the polymerase and does not contribute to proofreading, and a mode 2 of high accuracy in which the exonuclease is kinetically linked to the polymerase activity.
Our work was not taken very seriously. In particular, Stephen Benkowic and co-workers, working on short heteropolymers containing the four nucleotides - which looked more physiological models than our homopolymers - could not reproduce our observations. Stephen was kind enough to send us some of his template and we did not observe with this template anything like our mnemonic behaviour. Last, we did not see what could be the biological rationale for such a property. We wrote that « Allowing the enzyme to excise efficiently only when the cellular state permits active incorporation minimizes the risk of accidental degradation of DNA », but were not fully convinced by this explanation. Dominique Job and his team in Marseille accumulated evidence in favour of a mnemonic behaviour for wheat germ RNA polymerase . However, the work did not catch the attention of the RNA polymerase community.
In retrospect, I think that our work was quite good, and physiologically relevant. Possibly the failure to observe mnemonic effects with Benkovic’s template could be due to the possibility that the template was not sufficiently long to facilitate the renaturation of the enzyme, and that the exonuclease was therefore always working at low efficiency with this oligomer. Concerning the biological relevance of an « exonuclease warm-up » during elongation, a possible biological rationale emerged nearly twelve years later . I was then paying attention to the optimum accuracy of DNA repair enzymes, and found, quite naturally, that the shorter a repair patch, the higher the tolerable level of repair errors. So an enzme could start at a rather high level of errors, and switch to lower levels further along on the repair patch. I also explained why some inaccuracy was useful in the case of DNA repair .
On the whole, I find that this DNA polymerase work was not a loss of time. In some aspects, it was not as high-tech as other works (we worked on homopolymers, we used E. coli Pol I which is now thought to be a repair enzyme, and we did not perform fast kinetic experiments). It was however for some other aspects at a quite good technical level (the separation, for each time in the reaction of the products of the reaction, either by thin-layer chromatography, to quantify the dNMPs, the dNDPs and the remaining dNTPS, or later, the determination of misincorporation levels position by position using sequencing gels).
The article in BBA with Philippe Lecomte  was my last published experimental work in molecular biology. Philippe Lecomte made some further experiments with a periodic polymer, Poly(T-C-A)30, raising progressively the pyrophosphate concentration to act selectively on a single type of base pair then on another one. I find in my notes from that time that at moderate PPi, chain extension paused after the C sites as expected. At higher PPi, there was no polymerization, but polymerization occurred again at still higher PPi, for reasons which were not elucidated (or perhaps this result was an artefact).
 Brutlag, D. and Kornberg, A. (1972) Enzymatic synthesis of deoxyribonucleic acid. XXXVI. A proofreading function for the 3’ ą 5’ exonuclease activity in deoxyribonucleic acid polymerases. J. Biol. Chem. 247, 241-248.
 Muzyczka, N., Poland, R.L. and Bessman, M J. (1972) Studies on the biochemical basis of spontaneous mutation. 1. A comparison of the deoxribonucleic acid polymerases of mutator, antimutator and wild type strains of bacteriophage T4. J. Biol. Chem. 247, 7116-7122.
 Ninio, J. (1974) A semi-quantitative treatment of missense and nonsense suppression in the strA and ram ribosomal mutants of Escherichia coli . Evaluation of some molecular parameters of translation in vivo. J. Mol. Biol. 84, 297-313.
 Englund, P. T., Huberman, J. A., Jovin, T. M. and Kornberg, A. (1969) Enzymatic synthesis of deoxyribonucleic acid. XXX. Binding of triphosphates to deoxyribonucleic acid polymerase. J. Biol. Chem. 244, 3038-3044.
 Josse, J., Kaiser, A. D. and Kornberg, A. (1961) Enzymatic synthesis of deoxyribonucleic acid. VIII. Frequencies of nearest neighbour base sequences in deoxyribonucleic acid. J. Biol. Chem. 236, 864-875.
 Hershfield, M.S. (1973) On the role of deoxyribonucleic acid polymerase in determining mutation rates. J. Biol. Chem. 248, 1417-1423.
 Hall, Z.W. and Lehman, I.R. (1968) An in vitro transversion by a mutationally altered T4-induced DNA polymerase. J. Mol. Biol. 36, 321-333.
 Howard, B. D. and Tessman, I. (1964) Identification of the altered bases in mutated single-stranded DNA. II. In vivo mutagenesis by 5-bromodeoxyuridine and 2-aminopurine. J. Mol. Biol. 9, 364-371.
 Ninio, J.(1975) Kinetic amplification of enzyme discrimination. Biochimie 57, 587-595.
 Hopfield, J.J. (1974) Kinetic proofreading : A new mechanism for reducing errors in biosynthetic processes requiring high specificity. Proc. Nat. Acad. Sci. USA 71, 4135-4139.
 Ninio, J. (2006) Multiple stages in codon-anticodon recognition : double-trigger mehanisms and geometric constraints. Bochimie 88, 963-992.
 Ninio, J., Bernardi, F., Brun, G., Assairi, L., Lauber, M. and Chapeville, F. (1975) On the mechanism of nucleotide incorporation into DNA and RNA. FEBS Letters 57, 139-144.
 Bernardi, F. and Ninio, J. (1978) The accuracy of DNA replication. Biochimie 60, 1083-1095.
 Papanicolaou, C., Dorizzi, M. and Ninio, J. (1984) A memory effect in DNA replication. Biochimie 66, 43-48.
 Joyce, C. M. and Steitz, T. A. (1987) DNA polymerase I : from crystal structure to function via genetics. Trends in Biochem. Sci. 12, 288-292.
 Kuchta, R. D., Benkovic, P. and Benkovic, S. J. (1988) Kinetic mechanism whereby DNA polymerase I (Klenow) replicates DNA with high fidelity. Biochemistry 27, 6716-6725.
 Ninio, J. (1991) Transient mutators: a semiquantitative analysis of the influence of translation and transcription errors on mutation rates. Genetics 129, 957-962.
 Scheuermann, R. H. and Echols, H. (1984) A separate editing exonuclease for DNA replication : the e subunit of Escherichia coli DNA polymerase III holoenzyme. Proc. Nat. Acad. Sci. USA 81, 7747-7751.
 Ninio, J. (1986) Kinetic and probabilistic thinking in accuracy. In Accuracy in Molecular Processes (Kirkwood. T.B.L., Rosenberger, R. and Galas, D.J., eds) Chapman & Hall, London, pp. 291-328.
 Blank, A., Gallant, J.A., Burgess, R. R. and Loeb, L. A. (1986) An RNA polymerase mutant with reduced accuracy of chain elongation. Biochemistry 25, 5920-5928.
 Libby, R. T., Nelson, J. L., Calvo, J. M. and Gallant, J. A. (1989) Transcriptional proofreading in Escherichia coli. EMBO J. 8, 3153-3158.
 Libby, R. T. and Gallant, J. A. (1991) The role of RNA polymerase in transcriprional fidelity. Molecular Microbiology 5, 999-1004.
 Libby, R. T. and Gallant, J. A. (1994) Phosphorolytic error correction during transcription. Molecular Microbiology 12, 121-129.
 Volloch, V.Z., Rits, S. and Tumerman, L. (1979) Pyrophosphate-condensing activity linked to nucleic acid synthesis. Nucleic Acid Res. 6, 1521-1534.
[25 ] Volloch, V.Z., Rits, S. and Tumerman, L. (1979) A possible mechanism for the correction of transcription errors. Nucleic Acids Res. 6, 1535-1546.
 Hinrichsen, A. I., Ortner, I. M. and Hartmann, G. R. (1985) Synthesis of dinucleoside tetraphosphates by RNA polymerase B (II) from calf thymus. FEBS Lett. 193, 199-202.
 Butzow, J. J., Garland, C. Van Lee, L. and Eichhorn, G. L. (1997) Specificity of an RNA polymerase-associated NTPase. Biochemistry 36, 14794-14798.
 Urban, S., Urban, S., Fischer, K. P. and Tyrell, D. L. (2001) Efficient pyrophosphorolysis by a hepatitis B virus polymerase may be a primer-unblocking mechanism. Proc. Nat. Acad. Sci. USA 98, 4984-4989.
 Rozovskaya, T. A., Chenchik, A. A. and Beabealashvilli, R. S. (1982) Processive pyrophosphorolysis of RNA by Escherichia coli RNA polymerase. FEBS Lett. 137, 100-104.
 Postel, E. H. and Abramczyk, B. M. (2003) Escherichia coli diphosphate kinase is a uracil-processing DNA repair nuclease. Proc. Nat. Acad. Sci. USA 100, 13247-13252.
 Bernardi, F., Saghi, M., Dorizzi, M. and Ninio, J. (1979) A new approach to DNA polymerase kinetics. J. Mol. Biol. 129, 93-112.
 Fersht, A. R. (1979) Fidelity of replication of phage ĮX174 DNA by DNA polymerase III holoenzyme : Spontaneous mutation by misincorporation. Proc. Nat. Acad. Sci. USA 76, 4946-4950.
 Clayton, L. K., Goodman, M. F., Branscomb, E. W. and Galas, D. J. (1979) Error induction and correction by mutant and wildt type T4 DNA polymerases. Kinetic error discrimination mechanisms. J. Biol. Chem. 254, 1902-1912.
 Kunkel, T. A., Schaaper, R. M., Beckman, R. A. and Loeb, L. A. (1981) On the fidelity of DNA replication. Effect of the next nucleotide on proofreading. J. Biol. Chem. 256, 9883-9889.
 Phear, G., Nalbartoglou, J. and Meuth, M. (1987) Next-nucleotide effects in mutations driven by DNA precursor pool imbalances at the aprt locus of Chinese hamster ovary cells. Proc. Nat. Acad. Sci. USA 84, 4450-4454.
 Bessman, M.J., Myzyczka, N., Goodman, M.F. Schnaar, R.L. (1974) Studies on the biochemical basis of spontaneous mutation II. The incorporation of a base and its analogue into DNA by wild-type, mutator and antimutator DNA polymerases. J. Mol. Biol. 88, 409-421 ou 422.
 Goodman, M.F., Gore, W.C., Muzyczka, N. and Bessman, M.J. (1974) Studies on the biochemical basis of spontaneous mutation. III. Rate model for DNA-polymerase-effected nucleotide misincorporation. J. Mol. Biol. 88, 423-435.
 Galas, D.J. and Branscomb, E.W .(1978) Enzymatic determinants of DNA polymerase accuracy. J. Mol. Biol. 124, 653-687.
 Herbomel, P. and Ninio, J. (1980) Fidélité d'une réaction de polymérisation selon la proximité de l'équilibre. Comptes-Rendus Acad. Sci. Paris, Série D, 291, 881-884.
 Malygin, E.G. and Yashina, L.N. (1980) Kinetic description of the error-correcting mechanism of bifunctional DNA polymerases (in Russian). Doklad. Akad. Nauk. SSSR 250, 246-250.
 Durup, J. (1982) On the relation between error rates in DNA replication and elementary chemical rate constants. J. Theoret. Biol. 94, 607-632.
 Saghi, M. and Dorizzi, M. (1982) Polymerization/excision kinetics of Escherichia coli DNA polymerase I. Stability in kinetic behaviour and variation of the rate constants with temperature and pH. Eur. J. Biochem. 123, 191-199.
 Dumas, J.-P. and Ninio, J. (1982) Efficient algorithms for folding and comparing nucleic acid sequences. Nucleic Acids Res. 10, 197-206.
 Kunkel, T.A., Beckman, R. A. and Loeb, L. A. (1986) On the fidelity of DNA synthesis. Pyrophosphate-induced misincorporation allows detection of two proofreading mechanisms. J. Biol. Chem. 261, 13610-13616.
 Schaaper, R. M. (1993) Base selection, proofreading and mismatch repair during DNA replication in Escherichia coli. J. Biol. Chem. 268, 23762-23765.
 Kunkel, T. A. and Bebenek, K. (2000) DNA replication fidelity. Annual Rev. Biochem. 69, 497-529.
 Johnson, K. A. (1993) Conformational coupling in DNA-polymerase fidelity. Annu. Rev. Biochem. 62, 685-713.
 Beckman, R.A. and Loeb, L.A. (1993) Multi-stage proofreading in DNA replication. Quarterly Reviews of Biophysics 26, 225-311.
 Papanicolaou, C., Gouy, M. and Ninio, J. (1984) An energy model that predicts the correct folding of both the tRNA and the 5S RNA molecules. Nucleic Acids Res. 12, 31-44.
 Gouy, M., MarliŹre, P., Papanicolaou, C. and Ninio, J. (1985) Prediction des structures secondaires dans les acides nucléēques: aspects algorithmiques et physiques. Biochimie 67, 523-531.
 Witzel, H. (1967) Zur Interpretation kinetischer Daten bei der Reaktion der Pankreas-Ribonuclease. Hoppe-Syler’s Z. Physiol. Chem. 348, 1249-XXXX.
 Rabin, B. R. (1967) Co-operative effects in enzyme catalysis – a possible model based on substrate-induced conformation isomerization. Biochem. J. 102, 22C – 23C.
 Neet, K. R. and Ainslie, G. R. Jr (1980) Hysteretic enzymes. Methods in Enzymology 64B, 192-226.
 Ricard J, Meunier J. C. and Buc, J. (1974) Regulatory behaviour of monomeric enzymes. 1. Mnemonical enzyme concept. Eur. J. Biochem. 49, 195-208.
 Cornish-Bowden, A. and Cardenas, M.L. (1987) Co-operativity in monomeric enzymes. J. Theoret. Biol. 124, 1-23.
 Papanicolaou, C., Lecomte, P. and Ninio, J. (1986) Mnemonic aspects of E. coli DNA polymerase I. Interaction with one template influences the next interaction with another template. J. Mol. Biol. 189, 435-448.
 Lecomte, P.J. and Ninio, J. (1988). Nucleotide excision by E. coli DNA polymerase I in proofreading and non-proofreading modes. Biochim. et Biophys. Acta 951, 255-260.
 Lecomte, P.J. and Ninio, J. (1987) Variations with position of replication errors due to exonuclease warm-up. FEBS Lett. 221, 194-198.
 Job, D., Soulié, J. M., Job, C. and Shire, D. (1988) Potential memory and hysteretic effects in transcription. J. Theoret. Biol. 134, 273-289.
 Ninio, J. (2000) Illusory defects and mimatches - or why must DNA repair always be (slightly) error-prone? BioEssays, 22, 396-401.