One of the main purposes of this part of analysis was to verify if E. coli would be able to employ Met as a sole common donor of atoms for DNA synthesis and methylation, as HeLa cells do [47] [50] ; another purpose was to verify whether, on bacterial DNA, the timing for optimal yield of m⁵C would coincide with that for optimal yield of m⁶A [48] [49] .

It emerged that, when percolated through reverse phase HPLC [51] , control nitrogen bases, mixed in equimolar proportions, were eluted in the following time-order: C, m⁵C, G, T, A and m⁶A (Figure 1(a)). But there was a striking quantitative difference between the eluted unmethylated vs. methylated bases, when these were constituents of an E. coli DNA hydrolyzate: C, G, T, and A resulted in equimolar proportions, as expected; m⁵C and m⁶A were present in very small amounts (Figure 1(b)). Figure 1(c) illustrated the profile of an E. coli DNA hydrolyzate labelled with [¹⁴C]methyl-L-methionine ([¹⁴C]Met) in mid-culture growth cycle (CGC): C was the only base not incorporating the labelled carbon coming from the precursor; by contrast, m⁵C was labelled to a notable extent comparable to that of m⁶A. The radioactivity found in the two modified bases meant that post- synthetic DNA modification took place, as for HeLa cells [47] [50] , by virtue of transfer, via S-adenosyl-L-me- thionine (AdoMet), of the whole radioactive ?CH₃ from [¹⁴C]Met to the DNA Cs and As. In addition, in Figure 1(c), the radioactivity found in the A, G and T bases revealed that, once again, as for humans [47] [50] , DNA synthesis emerged because there was, via C₁-chain oxidation, an insertion of the labelled carbon coming from the ?CH₃ of [¹⁴C]Met into the purine heterocycle of A and G and into the ?CH₃ of T.

The quantitative analysis of the experiments performed through the CGC of E. coli (in the same conditions corresponding to those of Figure 1(c)), indicated that, while the curve of bacterial proliferation was S-shaped [48] , the synthetic and methylase pathways of DNA developed in a differential fashion (Figure 2(a)).

i. The specific labelling accounting for synthesis of A, G and T appeared to be much lower than that accounting for construction of m⁵C and m⁶A (this meant that the methylase pathway evolved faster than the synthetic pathway).

ii. The specific labelling of T was always lower than that of G and A since, one thought, the exit of the C₁- chain, oriented towards synthesis of G and A, was highly facilitated in comparison with that oriented towards synthesis of T.

iii. The specific labelling of m⁵C was much lower than that of m⁶A, probably because the dam enzyme system worked faster than the dcm enzyme system.

iv. In harmony with the development of the specific labelling of all nitrogen bases taken together and of the specific labelling of single A, G and T, the specific labelling of m⁶A gradually decreased during the CGC, when that of m⁵C rapidly decreased after the inoculum to rise again, with a sharp peak, in mid-CGC (an interpretation of these patterns was that the dam and dcm DNAmet activities are uncoupled) [48] [49] .

The data reported above demonstrated that the two metabolic pathways, leading to synthesis and methylation of DNA, originated in the prokaryotic world by exploiting Met as donor of carbon atoms and ?CH₃ groups (Figure 2(a)). This crucial property was preserved fairly well by the mammalian world (Figure 3(a) and Figure 3(b)), even if at the level of the DNA modifying enzyme bifurcation one of the two branches―the one yielding m⁶A on DNA (Figure 2(a))―was erased in HeLa cell DNA (Figure 2(b)).

The abandonment of m⁶A by DNA of the majority of the highly evoluted cells was a consequence of the substitution, in them, of the obsolete bacterial antiphage restriction-modification defense [6] with the innovative immune reaction. Actually, with the extinction of the system of dam and dcm DNAmets [9] and of the system of restriction-modification endonucleases [11] -[13] , both characterizing the world of bacteria [52] , molecular selection led in mammals to the loss of m⁶A not only in nuclear [46] but also in mitochondrial [53] [54] and chloroplastic [55] DNAs. In these different nucleic acids, the m⁵C base was conserved by a novel DNAmet family that likely evolved through an alternate splicing [56] -[58] , as in mammals the fundamental defense against antigens was exerted by antibodies.

In such a structure, the question concerning the speed of reactions in the pathway leading to DNA synthesis vs. the speed of reactions in the pathway leading to post-synthetic DNA modification assumed great interest since, during the bacterial CGC, the rate of formation of T was always lower than that of A and G. Namely, the time required by the carbon, on the way from the ?CH₃ of Met to the ?CH₃ of T, was longer when compared to the time required for its insertion into the purine heterocycle of A and G [48] [49] . This differential rate was in harmony with the situation seen in HeLa cells: quantitatively, the biosynthetic rate of T was of the same order of that of A, even though a notable delay of its optimum with respect to the optimum of synthesis of A and G was observed (Figure 3(a)).

This kind of correspondence was expected, since either in bacteria (Figure 1) or mammals (Figure 3(a)) the process leading to synthesis of T had to join into a pyrimidine ring the carbamylphosphate and aspartate molecules before accepting, at position 5 of the newly formed U, the ?CH₃ constructed with the carbon coming from formic acid. This carbon entered directly the A and G purine heterocycle (Figure 2(a)).

Biosynthesis of A, G and T transpired in comparable loci (in liquid matrix of a bacterial cell or in cytosol of a mammalian cell), whereas post-synthetic modification of DNA occurred in different compartments: in E. coli, it yielded m⁵C and m⁶A in the same liquid matrix (Figure 1(b) and Figure 1(c)); in HeLa cells, it yielded the sole m⁵C in the nucleus (Figure 3(b)). As follows, the longer time required by the nucleoside AdoMet to enter the nucleus would justify the advantage of the bacterial cell over the mammalian cell in exploiting the ?CH₃ of Met: if in E. coli the specific building of A, G and T was much lower than that of m⁵C and m⁶A [48] [49] , in HeLa cells the specific building of A, G and T turned out to be either similar or only a little lower than that of m⁵C [47] [50] .

Summarizing, the large percentage of DNA methylation, necessary for antiphage defense in bacteria, became unnecessary in mammals.

Was this a non-turning point in evolution?

To make a long reasoning short: the prokaryotic world used Met to promote both the pathways leading to synthesis and methylation of DNA (Figure 1(b) Figure 1(c) and Figure 2(a)); this capacity was inherited almost completely by the mammalian world, with the exception that, in HeLa cells, the reactions yielding m⁶A on DNA were suppressed (Figure 3(a) and Figure 3(b)); the diversification of the speed of single reactions (in the pathways leading to synthesis and methylation of DNA) depended on the evolution of cellular organization, of course, since the eukaryotic nucleus concentrated in itself important genetic functions that were diffused earlier in the bacterial liquid matrix [47] -[49] .

It was suggested that DNA methylation would continue for some hours after DNA synthesis is completed [68] [69] . In this research, by incubating synchronized HeLa cells with [¹⁴C]Met used as common tracer for both DNA synthesis and methylation [47] , it was possible to verify that DNA methylation accompanied DNA synthesis: the labelled carbon atom, coming from the ?CH₃ of Met, did not enter the pyrimidine ring but pierced via C₁-chain the purine ring of A and G and the ?CH₃ of T, while the entire ?CH₃ of [¹⁴C]Met was transferred via AdoMet to DNA C (Figure 2(b)). Truly, the maximal labelling emerging from the ¹⁴C constituting the ?CH₃ of [¹⁴C]Met was always found in S in four hydrolyzed bases: in A, G and T, implying synthesis, and in m⁵C, signifying methylation (Figure 3(a)).

It was known that in isolated nuclei, in the absence of triphosphonucleosides (supplied by cytosol), DNA could not be synthesized. On this basis, by employing nuclei labelled with [¹⁴C-methyl]AdoMet, one could verify whether or not the absence of DNA replication would influence DNA methylation. The result was unexpected: in nuclei isolated from HeLa cells during S, among other DNA bases, only C continued to be methylated in the absence of DNA synthesis (Figure 3(b)). This showed that the two pathways of DNA replication and DNA methylation were separated from one another (Figure 2(b)): in whole cells, DNA methylation followed DNA replication during S (Figure 3(a)); in isolated nuclei, DNA methylation proceeded during S in the absence of DNA synthesis (Figure 3(b)). By itself the occurrence of DNA methylation in S-phase nuclei was not an absolute demonstration that it involved new chains formed just before their isolation; but the correlations between in vivo DNA synthesis and in vivo and in vitro DNA methylation strongly suggested this possibility [47] :

These equations established that the cells entering S would carry old chains inherited by parental cells, while targets of DNA met both in vivo and in vitro should be nothing but nascent chains: several of them still remained in the nuclei to be isolated [47] . A similar conclusion was reached thinking that, if old and new DNAs were methylated in isolated nuclei [70] , new DNAs would be formed without m⁵C in whole cells [71] .

Whatever the case would be, the semiconservativity of DNA methylation (Figure 3(c)), revealed by treating synchronized cells with the [¹⁴C]Met tracer [47] , was confirmed not only by the analysis that used restriction endonucleases [72] but also by the direct separation in alkaline CsCl of heavy bromodeoxyuridine (BrdUrd)- containing new chains and light non-containing BrdUrd old chains (Figure 3(d)).

It is worth re-visiting in some more details the basic dispute into which we get involved. Georgyev and Mantieva found the polyribonucleotide called giant nuclear RNA in mouse [73] , as also evidenced by us in nerve cells [37] [38] . Penman, Vesco and Penman claimed that their polydisperse RNA had nothing to do with the giant nuclear RNA [74] . In return, Georgyev assumed that the polydisperse RNA would likely correspond to a product of maturation of the giant nuclear RNA [75] .

Accepting this assumption, Darnell, Jelinek and Molloy considered the giant nuclear RNA as pre-processed mRNA (measuring the size of a gene) and the polydisperse RNA as processed mRNA (much shorter than a gene): an agreement stating that a transcriptase would copy the whole eukaryotic gene to form a large pre- processed pre-mRNA, namely the heterogeneous nuclear RNA [76] .

This conclusion helped indeed to elucidate the structure of the monogenic transcriptional unit in mammals, since we judged such a monogenic unit as an extraordinary innovation exerted by evolution with respect to the complex bacterial polygenic transcriptional unit known as operon.

Therefore, the information reported above on semiconservative methylation of newly replicating DNA chains (Figure 3(c) and Figure 3(d)) served as a key for a crucial biochemical analysis [46] : segments of these chains, obtained through shearing, were employed as probes, labelled in m⁵C, in order to look for particular sequences located inside the eukaryotic gene (Figure 4(A)).

Filaments of semiconservatively newly made methylated DNA, previously separated during the S-phase in alkaline CsCl gradients (Figure 3(d)), were shortened through sonication into fragments of about 1 × 10⁶ Daltons (Figure 4(A), left top) and hybridized with pre-mRNA (Figure 4(Ab)) extracted from HeLa cell nuclei [46] [66] [67] or with mRNA (Figure 4(Ac)) extracted from polysomes of these same cells [46] [77] : in the first case, the modified Cs were found in about 50 per cent of the hybrids “pre-mRNA/DNA probe” (signed with “plus” vs. “minus” in Figure 4(A)); in the second, the modified Cs were essentially absent in 100 percent of the hybrids “mRNA/DNA probe” (marked with “minus” in Figure 4(A)).

For clarity’s sake, the result of the former experiment signified that the m⁵Cs were located along the promoter and along the non-coding regions inside the gene (all uncoding sequences, including the promoter, being complementary to regions to be removed from the unprocessed pre-mRNA); by contrast, the result of the latter experiment suggested that the joined translatable sequences of processed mRNA were complementary to the unmethylated coding sequences scattered in the gene.

Such a design of human transcriptional unit-structured in alternating non-informative and informative sequences, starting from the promoter (Figure 4(A), right top)―was illustrated at the 9th Meeting of the Federation of the European Biochemical Societies held in Budapest in 1974: the oral presentation was preceded by an invited paper that opened the ritual Special Issue of FEBS-Letters [46] .

Three years later, in 1977, the novelty brought by this paper was confirmed in France by Breathnach, Mandel and Chambon [78] : in their electron microscopic observations, the ovalbumin mRNA repeatedly did not hybridize sequences of the ovalbumin gene, forming characteristic loops (these were nothing else but the uncoding regions found earlier by us); differently, the sequences of the ovalbumin gene, that did become hybridized to the ovalbumin mRNA, were nothing else but our coding regions.

Soon, a number of other laboratories, mostly in USA, observed that the eukaryotic gene looked split: the pre-mRNA coding sequences were interrupted by uncoding ones [79] -[82] ; uncoding sequences were seen even in the adenovirus mRNA [83] [84] .

Then, when an intervening sequence was also described in the mouse beta-globin gene [85] , Gilbert [86] and Crick [87] proposed to call exons and introns, respectively, the coding and non-coding regions found in Naples [46] .

Nevertheless, while two years earlier the proposal of these new appellations Maclean and Hilder [88] , from United Kingdom, in the International Review of Cytology, attributed to us the priority of discovery of the “preferential methylation of bases in the promoter proximal region of the transcriptional unit”, two years later the introduction of these new appellations, Tentravhi, Guntaka, Erlanger and Miller [89] , from United States, in the Proceedings of the National Academy of Sciences of USA, attributed to us also the priority of discovery of the “preferential methylation of the early replicating regulatory sequences that do not code for mRNAs”.

In sum, if in HeLa cells half of the population of hybrids between genomic DNA fragments (used as probes) and pre-processed mRNAs (purified from nuclei) contained a large number of m⁵Cs (Figure 4(Ab)), the whole population of hybrids between genomic DNA fragments (used as probes) and processed mRNAs (purified from polysomes) contained few, if any, m⁵Cs (Figure 4(Ac)). Since polysomal mRNAs were known to be much shorter than nuclear pre-mRNAs, these results suggested that, in comparison with the bacterial operon structure (containing cistrons essentially representing informative sequences), the eukaryotic gene had to be thought of as a repetition, after the promoter, of alternate coding and uncoding regions (Figure 4(A), top right): methylation did not significantly involve the coding regions but involved the non-coding ones, complementary to parts of pre-mRNAs to be removed during processing [46] .

Evidence indicated that the S-phase should be subdivided into two parts with respect to the characteristics of DNA, since early replicating euchromatic DNA tended to be CG-rich, while late replicating heterochromatic DNA tended to be AT-rich [103] .

Furthermore, the DNA fraction extracted from Chinese hamster cells in early S was methylated to a greater extent when compared to the DNA fraction extracted from the same cells in late S [104] .

This difference was reproduced in HeLa cells [46] . On its basis, we supposed that, while in newly replicating DNA chains there would be sequences not uniformly methylated in the course of S, the CG-rich sequences would be preferentially methylated [46] .

Since DNA did not appear to be uniformly methylated in S, investigations continued to focus on nitrogen bases specifically adjacent to m⁵C [46] rather than considering the rough subdivision of hypermethylated euchromatic vs. hypomethylated heterochromatic replicons [104] .

With this purpose in mind, methylated DNA coming from HeLa cells was fractioned in alkaline Ag⁺/Cs₂SO₄ gradients: a small heavy CG-rich fraction and a large light AT-rich fraction were obtained (the first―containing genes for rRNA―was mainly expressed in early S and, the second, was mainly expressed in late S); as for the m⁵C concentration, if in early S it was found to increase on the heavier peak, in late S it was found to increase on the lighter peak [105] -[107] .

All this suggested that, at the different stages of S, one dealt with CG- or AT-rich regions probably polymerized in correspondence of specific hypo- or hypermethylated templates: diversely, it seemed that the genes would be methylated according to a given order and intensity along the newly replicating chains [105] .

This idea was definitely demonstrated when we observed that the foldback DNAs were subjected to hypermethylation: if in HeLa cells―through the S-phase―a methylation wave chronologically involved palindromic, highly repeated and moderately repeated sequences, the unique sequences were characterized by a minimal late methylation [108] .

A circumstantial analysis of the two genomic DNA fractions, separated in alkaline Ag⁺/Cs₂SO₄ gradients during S, confirmed that m⁵C had a differential distribution in them: the highest concentration of methylated sequences was found in the denser side of the heavy fraction; in the light DNA the highest concentration of methylated sequences was found in the lighter side [105] -[107] .

This implied two generalizations: hypermethylation on CG-rich sequences would offer no surprise because of its rather statistical character (the CG-rich rRNA genes would be hypermethylated according to the size of their CG targets); hypermethylation on AT-rich sequences would seem to work, instead, against the principle of CG target size, since a decrease in C residues would correspond to an increase in their methylation. Hence, when the extra-S time was characterized by a minimal, probably de novo DNA methylation, S exhibited maintenance DNA methylation [47] .

For this reason, one assumed, the mechanisms discussed above should function in S [90] .

But it was obvious that, with consecutive cell cycles (Figure 6(a)), the methylated sequences could not increase without limits [109] , also because changes in the amount of methylated DNA during development [100]

and in differentiated tissues [29] [30] were not conspicuous.

So, whatever the mechanism of DNA methylation was in S, statistical or specific, it was reasonable to imagine that, mostly during the extra-S time, each small wave of de novo DNA methylation should be followed by a corresponding small wave of DNA de-methylating repair [53] .

In HeLa cells, replication of genomic DNA [65] [67] [101] was followed by its methylation [47] in S, whereas transcription [110] and translation [111] mainly resulted in G₁ and G₂. This was the first concrete suggestion as to a macroscopic inverse correlation observable between gene methylation and expression (Figure 6(a)). As for the enzyme components: DNA polymerase (DNApol) alpha, responsible for replication of genomic DNA, was induced in S; DNApol gamma, responsible for replication of mtDNA, was made in S and G₂; DNApol beta, involved in rebuilding, was produced at a rather low level through the whole mitotic cycle with some increase in early S (the increased DNApol beta repair activity in early S signifying an effort by the cell to highly reconstruct the genomic DNA just before its replication) [121] . Moreover, in conformity with the sharp peak of maintenance genomic DNA methylation in S (Figure 6(a)), the DNAmet system [63] was induced in S as well (Figure 3(b)), while a moderate presence of DNAmet characterized the rest of the cell cycle mainly accounting for de novo genomic DNA methylation (Figure 3(b)). During S, a coupling―definable as repair-modification―took place between the activities of DNApol beta (involved in excision-repair) and DNAmet (making re-methylation), whereas during the extra-S time an uncoupling occurred between these two enzyme systems [53] [64] .

Thus, using synchronized HeLa cells [65] [67] , we took a look at the differential reconstruction of the UV- damaged unmethylated coding and methylated uncoding gene sequences (Figure 6(b)): in a repair patch (RP) located along given promoters or introns (1), it could happen that a symmetrically dimethylated 5’-m⁵CG- 3’/3’-Gm⁵C-5’ dinucleotide pair would flank a UV-induced TT-dimer (2); after digestion of the damaged nucleotide sequence (3), the excision-repair (being the methylated dCTP absent in the soluble pool of triphosphonucleosides) would replace―in the “incompletely” reconstructed sequence―the previous m⁵C nucleotide with a normal unmethylated cytosine nucleotide (4). Even if the existence of DNA de-methylating proteins should not be excluded in general [113] , this “non-enzymatic de-methylation” certainly would occur through the whole cell cycle. Anyhow, during S―via AdoMet―DNAmet would add to the incompletely reconstructed (unmethylated) DNA sequence (4) the lost ?CH₃ group, rebuilding the previous symmetrically di-methylated 5’-m⁵CG-3’/3’- Gm⁵C-5’ dinucleotide pair (5). This “complete” rebuilding of a previously methylated RP would require, in any situation, a “re-methylation” catalyzed by DNAmet [53] [95] [114] , as demonstrated by the fact that the labelled carbon atom coming from the ?CH₃ of [¹⁴C]Met (Figure 3 and Figure 2(b)) was found in the T, G, A and m⁵C nitrogen bases carried by the re-methylated RPs (Figure 7(c)) distributed along the light parental strands (Figure 7(b)). Yet, to what kind of regions―along the eukaryotic transcriptional unit―did such RPs belong?

Sedimentation of the semiconservatively newly replicating chains, made heavier in alkaline CsCl by BrdUrd incorporation (Figure 7(a)), was confronted with that of the labelled RPs appearing in the light parental chains (Figure 7(b)). These light parental chains, after an extensive shearing by sonication [46] , were hybridized with polyA-containing pre-mRNA extracted from nuclei (Figure 7(c)) or with polyA-containing mRNA extracted

from polysomes (Figure 7(d)). In the first combination, the hybrids “DNA/pre-mRNA” showed an appreciable amount of m⁵Cs; in the second, the hybrids “DNA/mRNA” did not appear to be methylated. These results signified that re-methylation involved the uncoding parts of the repaired genes and that the coding parts of these genes only underwent a rebuilding of their primary base sequence, without modified bases.

In summary, we considered one of the possible mechanisms―repair-modification―through which the presence of m⁵C in eukaryotic genes might have exerted a key role in the evolution of their unmethylated coding vs. methylated uncoding regions: after radioinduced damage, the RPs formed in exons (where the modified bases were practically absent) would be completely reconstructed by the sole excision-repair, whereas the RPs formed in promoters and introns (where the modified bases were normally present) would be incompletely reconstructed by excision-repair. Only later, in S, the unmethylated RPs would become completely reconstructed by re-me- thylation.

We assumed, therefore, that the repair-modification machinery exerts a continuous differential pressure on evolution of given gene pieces, when methylated to a different extent (Figure 7(c) and Figure 7(d)).

In particular, according to Darwinian concepts, repair-modification would lead to a very high variability of introns (if hypermethylated), whereas on the basis of this wide variability selection would favor their transposition from one gene to another. In them, the conservative nature of exons (if unmethylated) would be ensured by excision-repair.

Besides, it is worth stressing the possible influence that the incomplete repair of RPs (see the de-methylation shown in Figure 6(b4)) could exert on resumption of transcription by virtue of the cell cycle-dependent inverse correlation seen between DNA methylation and RNA synthesis (Figure 6(a)).

In this case, in harmony with the azacytidine effect [94] observed in HeLa cells (Figure 8), certain previously silent (Figure 6(b1)) and then re-activated (Figure 6(b4)) genes would lead even to neoplasy.

After all, the idea suggesting that the radio-induced DNA de-methylation could be considered as a feasible molecular basis for cancerogenesis [53] conformed with the known fact denouncing that at least 30 per cent of tumors are really induced in humans by radiations [148] [149] .

In bacterial DNA the m⁶A and m⁵C modified bases were related not only to the restriction-modification anti-phage defense [6] : complementing the induction or repression operon functions [39] , they were further associated with an auxiliary mechanism capable of partially regulating transcription in the same phage-infected microorganisms [115] . This “embryonic” stage of control evolved in the major part of eukaryotes: in them the higher concentration of m⁵C, present in the promoter and introns, was suggested to be responsible for modulation of gene expression [46] , as supported by others [60] [116] [117] and reviewed [89] [118] . Such an involvement found its first clear demonstration in the fact that the maximal rate of DNA methylation [47] [119] followed the maximal rate of DNA replication [101] during S, while the maximal rates of transcription [110] and translation [77] [111] appeared during G₁ and G₂ (Figure 6(a)). Further results supporting the idea about the role of m⁵C in regulation of gene expression were soon detected [120] -[123] , afresh to those on the methylation-de- pendent regulation of integrated viral genes [92] -[109] [117] [124] [125] . Moreover, maintenance DNAmet [47] [72] was shown to recognize hemimethylated 5’-CpG-3’/3’-Gpm⁵C-5’ dinucleotide pairs to yield dimethylated 5’-m⁵CpG-3’/3’-Gpm⁵C-5’ dinucleotide pairs (Figure 3(c)), whilst an inverse correlation between methylation and expression was established for a large number of sequenced HK and TS genes [98] [99] . In fact, an inverse correlation between gene methylation and transcription did occur in animals signifying, at the same time, that it could be caused most likely by the higher concentration of m⁵C in promoter and intron sequences: one theory justifies the other, and vice versa.

After establishing that in eukaryotes the ?CH₃ groups are preferentially attached to the CpG dinucleotides [106] [107] [126] , as in bacteria [127] , clearcut evidence about the existence of two classes of targets for DNAmet emerged, as mentioned already, from analysis of HeLa cell double-helical DNA in alkaline Ag⁺/Cs₂SO₄ [106] .

This type of gradient separated a heavier fraction (representing about 20% of the total) from a lighter one (representing about 80% of the total) [128] .

However, the consequent ultracentrifugation of these fractions in CsCl [106] further showed that the heavier, banding at 1.715 g/cm³, contained 53% CG (10% of the total CG), whereas the lighter, banding at 1.703 g/cm³, contained 40% CG (32% of the total CG).

Generalizing, four possible triplets as sites for DNAmet recognition were suggested: GCG and CCG, in CG-rich sequences; ACA and ACG, in AT-rich sequences [107] .

The evidence that the specific methylation of CG-rich sequences (as those for rRNAs) was maximal in early S and that the specific methylation of AT-rich sequences was maximal in late S not only demonstrated that the DNA sequences were replicated and methylated following an order during S [107] [108] but also pointed that the DNAmet activity exerted the role of maintenance modification along CG-rich sequences and the role of de novo modification along AT-rich sequences [107] .

This important deduction was in accordance with the fact that the extra-S methylation (about 10% of the total) involved some AT-rich sequences either along the old or the new chains, when the S phase methylation (about 90% of the total) almost exclusively involved newly replicating chains [47] [50] .

In S, the maintenance DNAmet function [63] followed at a distance of about 30 min [47] the DNApol alpha activity [112] .

The reasoning made sense because, in S, the maintenance modification―involving CpG targets in newly replicating chains―would be induced by the ?CH₃ of the m⁵C present in the complementary strand as a genetically encoded signal.

Otherwise, a CpG dinucleotide (Figure 3(c)) would be methylated on the new chain in the presence of a complementary already methylated GpC dinucleotide on the old chain [47] [72] ; differently, during the extra-S time, a de novo modification occurring on an ApC dinucleotide, for instance, could not be induced by a signal coming from the complementary TpG dinucleotide since this cannot be methylated at all.

For the sake of clarity, the above-established sedimentation of CG- and AT-rich sequences in alkaline Ag⁺/ Cs₂SO₄ suggested the existence of at least four targets for methylation [106] :

GCG, CCG, ACA, and ACG.

In one of the trinucleotides there was a methylatable CpA; in three of them, the methylatable CpG was repeated. For this reason, a direct isolation of such restricted targets was required (Figure 9).

This was achieved by exploiting DNAase I digests [126] [129] of DNA samples methylated in isolated nuclei of previously synchronized cells with [³H-methyl]AdoMet [47] [109] . Chromatographic analysis of the material contained in these digests [102] yielded a number of obviously unmethylated dinucleotides

ApA, ApG, ApT, CpT, and TpT

and, for the rest, four dinucleotides methylated to a different extent:

m⁵CpT, Cpm⁵C, m⁵CpA, and m⁵CpG.

Their methylation level was cell-cycle dependent, in agreement with [47] : in M, few CpGs were methylated; in G₁ and G₂, methylation of CpGs slightly increased, in comparison with that seen in M, and there appeared some methylation on CpTs, CpCs, and CpAs; in S, again, a methylation of four dinucleotides

m⁵CpTs, Cpm⁵Cs, m⁵CpAs, and m⁵CpGs

was detected. In this phase, methylation of CpGs became intensive (90% of the total), accounting for its maintenance character along newly made chains (Figure 3(c)). The occurrence of methylation on CpTs, CpCs, and CpAs accounted instead for a de novo phenomenon, which characterized especially the extra-S part of the interphase [47] . Considering the two possible directions of the separated dinucleotides, due to the double helix antiparallelism (the 5’-3’ and 3’-5’ directions could not be distinguished chromatographically), the analysis actually regarded the identification of eight words written in the genome in terms of dinucleotide pairs:

5’-m⁵CpG-3’ 5’-Tpm⁵C-3’ 5’-Cpm⁵C-3’ 5’-m⁵CpA-3’

5’-Gpm⁵C-3’ 3’-ApG-5’ 3’-GpG-5’ 3’-GpT-5’

+ + + +

3’-m⁵CpG-5’ 5’-m⁵CpT-3’ 5’-m⁵CpC-3’ 5’-Apm⁵C-3’

3’-Gpm⁵C-5’ 3’-GpA-5’ 3’-GpG-5’ 3’-TpG-5’

Briefly, if the CpG target only allowed a final dimethylation of dinucleotide pairs, as expected from semiconservative methylation [47] [72] [102] [130] , the other combinations always led to a monomethylation of dinucleotide pairs [102] . Still, what could be demonstrated about the diversity of the methylation code at the level of the promoter and at that of the intervening uncoding sequences?

Using the model of human transglutaminase (hTGc) gene [131] and the method based on bisulfite conversion of C to T residues along a DNA filament [132] , the following analysis showed that m⁵C, behaving as bisulfite-in- dependent base [133] , could be sequenced directly despite its low molar proportion [134] .

The hTGc gene was chosen principally because it was among those that are regulated by methylation [135] . In its 1665 bp long promoter [131] , three CpG-rich domains were characterized (Figure 10): the first domain (330 bp), close to 5’, contained 12 CpGs corresponding to an average frequency of 3.63% with respect to the total number of nucleotides; the second (227 bp), roughly located in the middle of the promoter, contained eight CpGs corresponding to an average frequency of 3.52%; the third (264 bp), close to 3’ (including 70 bp of the 73 bp long 5’-UTR), contained 31 CpGs corresponding to an average frequency of 11.74% [136] .

In leukocytes and lymphocytes, where hTGc is silent [135] , out of the three CpG-rich domains, only the first two were found to be methylated, while the third, on the 3’-side, did not present any m⁵Cs: the 11 CpGs of domain 1 and the 7 CpGs of domain 2 were methylated 100%; the 31 CpGs of domain 3 were on the contrary unmethylated 100% (in this domain the 5’-UTR, also resulting 100% unmethylated, was almost entirely included) [136] .

The lack of m⁵C in domain 3 was expected, in agreement with the conventional definition of unmethylated CpG-rich “island” (where the intrafilament CpG/GpC ratio has to be higher than 0.6); but the occurrence of methylation in all CpGs of domains 1 and 2 was unexpected. Domain 1 contained 79 Cs and 11 CpGs out of 330 bases; domain 2 contained 44 Cs and 7 CpGs out of 227 bases; domain 3 contained 116 Cs and 31 CpGs out of 264 bases. Compared with the average 5%-methylation of human genomic DNA [137] , methylation of domains 1 and 2 corresponded to 15.18% and 18.18%, respectively [136] . This confirmed the idea that the promoter of an inactive gene should be characterized by hypermethylation [46] [98] [99] .

Certainly, HUVEC cells joined their hTGc gene activity with a loss of m⁵C along the hTGc promoter domain 1, at least at the −1380, −1349, −1338, and −1320 CpG sites [138] . This is an appropriate example of direct

Methylated Methylated Unmethylated

base-sequence analysis showing the inverse correlation between the activity of a gene and the methylation of its promoter (of course, in harmony with this correlation, the on-going research is aimed at clarifying the regulatory role of m⁵Cs along the non-coding sequences).