(Received for publication, January 24, 1997, and in revised form, April 16, 1997)
From the Institut de Génétique et de Biologie Moléculaire et Cellulaire (CNRS/INSERM/ULP), F-67404 Illkirch Cedex C.U. de Strasbourg, France
As an initial approach to characterizing the molecular structure of the human RNA polymerase II (hRPB), we systematically investigated the protein-protein contacts that the subunits of this enzyme may establish with each other. To this end, we applied a glutathione S-transferase-pulldown assay to extracts from Sf9 insect cells, which were coinfected with all possible combinations of recombinant baculoviruses expressing hRPB subunits, either as untagged polypeptides or as glutathione S-transferase fusion proteins. This is the first comprehensive study of interactions between eukaryotic RNA polymerase subunits; among the 116 combinations of hRPB subunits tested, 56 showed significant to strong interactions, whereas 60 were negative. Within the intricate network of interactions, subunits hRPB3 and hRPB5 play a central role in polymerase organization. These subunits, which are able to homodimerize and to interact, may constitute the nucleation center for polymerase assembly, by providing a large interface to most of the other subunits.
In eukaryotic cells transcription of protein coding genes is carried out by RNA polymerase II (RPB),1 in combination with a number of cofactors that control the selectivity and/or efficiency of transcription initiation, elongation and termination (for recent reviews, see Refs. 1-3). The initiation process involves the formation of a multi-protein complex, referred to as the preinitiation complex (PIC), on the template DNA. The PIC may be assembled stepwise on the promoter by sequential binding of RPB and general transcription factors (3, 4), as suggested by in vitro DNA-binding experiments (3, 4). Recent observations also suggest that formation of the PIC may result from the direct promoter association of a large preformed complex (referred to as the holoenzyme) comprising, in addition to the above mentioned components, several polypeptides, some of which are involved in the chromatin clearing of the target promoter region (5). The description, at the molecular level, of the spatial organization of this complex edifice necessitates a detailed knowledge of the structure of its central component, RPB.
Depending on its origin, RPB consists of 10-14 polypeptides ranging from 220 to 10 kDa (6-8). To date, most studies have focused on the yeast polymerases; much less is known about the human (h) RPB, although the sequences encoding the subunits homologous to the yeast RPB have now been identified (see Table I).
|
Electron microscopy image processing and two-dimensional crystal analyses have been applied to yeast RNA polymerase I (RPA) and RPB (9, 10) and have suggested a rather complex structure. Several subunit-specific epitopes have recently been localized within the yeast RPA, using immunoelectron microscopy (11). Potential interactions between the various subunits of yeast RNA polymerase III (RPC) have been tested by the two-hybrid system. However, few contacts could be detected by this method and comprised only RPC-specific subunits (12).
Functional domains could be assigned to specific subunits, on the basis
of primary sequence conservations between subunits of
DNA-dependent RNA polymerases from all living organisms and experimental approaches such as affinity labeling, protein and DNA
blotting, cross-linking, protease-susceptibility measurements, and
functional inhibition by subunit-specific antibodies (11). Both largest
subunits (so-called and
-like subunits, respectively) have been
found to interact with the DNA template and RNA, while only the second
largest binds the initiator nucleotide. Together, these observations
strongly suggest that these large subunits contain most of the
catalytic activities. On the other hand, the phenotypes associated with
the deletion of six yeast subunit genes could be independently
compensated for, at least to some extent, by their human counterparts
(13-16). These results further emphasize the role of conserved
sequence elements, but also indicate that the polymerases can
accommodate, at least to some extent, variations in primary structure
between subunits.
In the present study, we have undertaken a systematic analysis of the interactions that may exist between the various hRPB subunits. Using a baculovirus expression system, we were able to overexpress all the hRPB subunits (except hRPB4, which was originally not available) in a soluble form, either as untagged subunits or as glutathione S-transferase (GST) fusions. The interactions between pairs of subunits were then tested in a GST-pulldown assay for the available subunits. While not all subunits interact under these conditions, the results reveal a number of reciprocal bindings whose relevance is discussed in the context of the complete enzyme. Detailed knowledge of the interaction network between the subunits will ultimately help in generating functional complexes from recombinant polypeptides.
For the sake of simplicity, the subunits of
hRPB (initially based on their molecular mass in kDa) have been
renamed, following the genetic nomenclature of yeast RPB subunits (7,
17). Accordingly, each hRPB subunit was attributed a number
(n = 1-9, 10, 10
, and 11), as indicated in Table
I.
Cloning of sequences encoding hRPB1, 2, 3, 5, 6, 7, 8, 9, 10, and 10
subunits has been described previously (in Refs.
18, 19, 20, 21, 22, 15, 13, 23, 13, and 13, respectively). Recently,
the sequence of a subunit named hRPB14 was described (24). Our attempts
to reclone this cDNA from our HeLa cell cDNA library failed.
Instead, upon screening data bases with the hRPB14 sequence, we
identified a human expressed sequence tag (EST) (HUMGS00602; GenBankTM
accession no. [GenBank]) exhibiting a strong homology with the 5
, but not
the 3
portion of hRPB14. Using primers derived from the 3
portion of
the EST sequence and from the 5
portion of the hRPB14 sequence, we
PCR-amplified a fragment of the expected size (420 bp) from our HeLa
cDNA library. The resulting fragment was cloned and its sequence
established (accession no. [GenBank]). This cDNA, which represents the
major detectable species in our library, will be referred to henceforth
as hRPB11. Independently, a human cDNA was recently published (25)
(accession no. [GenBank]), which encodes a polypeptide identical to
hRPB11. The original hRPB14 (24) may therefore represent either a rare
related protein or a cloning artifact consisting of a chimeric
cDNA.
PCR-mediated mutagenesis was used to insert unique restriction sites (NheI, SpeI, or XbaI, depending on the target cDNA) at the boundaries of each hRPB open reading frame. These sites have been chosen because they generate cleavage products which are compatible with NheI, SpeI, XbaI, and AvrII sites, and may be inserted into any one of them.
The DNA fragments containing the open reading frames of all subunits
except hRPB1 and hRPB2 were inserted into the XbaI site of
the pVL1393 baculovirus transfer vector (PharMingen) (Fig. 1, left). The hRPB1 and hRPB2 sequences were
subcloned together into the XbaI and SpeI cloning
sites of the pAcAB4 transfer vector (PharMingen), respectively. The
hRPB1 NheI fragment lacking its proper termination codon was
inserted, after filling-in of the ends, into the repaired
XbaI site, creating an artificial stop codon (underlined) at
the resulting junction: 5-GCT AGC TAG A-3
. The
corresponding hRPB1 peptide consisted of 1972 amino acids including the
C-terminal extra Ala-Ser residues. The XbaI fragment spanning the hRPB2 coding sequence was inserted into the
SpeI cloning site of the same vector. The resulting plasmid
enabled the creation of a baculovirus that simultaneously expressed
hRPB1 and hRPB2 (Fig. 1, middle).
In another series of constructions, the pVL1393 vector was modified by insertion of an EcoRI fragment spanning the GST coding sequence (excised from the pGEX-3X plasmid; Pharmacia Biotech Inc.) into the unique EcoRI site of pVL1393. In addition, the polylinker was altered so as to include unique NheI and NdeI cloning sites that allowed insertion of the subunit sequences and subsequent expression of the corresponding GST fusion proteins, with the GST at the N-terminal position (Fig. 1, right). In the case of the GST-hRPB1 construct, six His residues were added to the natural hRPB1 before reaching a translation termination site. In all other cases, translation terminated at the natural position.
The resulting transfer vectors were then recombined with linearized baculovirus DNA (BaculoGold DNA, PharMingen) in Sf9 cells, using procedures recommended by the manufacturer. The Sf9 cells were grown in the TNM-FH medium as described (26) at 28 °C. The recombinant viruses were plaque-purified, and correct expression of the corresponding hRPB subunits was verified by Western blot analysis. Viral stocks were prepared by a three-step growth amplification.
Baculovirus Infection and Metabolic LabelingAbout 2 × 105 Sf9 cells, grown in 24-well dishes (Falcon), were
coinfected with pairs of the appropriate baculoviruses. Multiplicities of infection (between 2 and 10 plaque-forming units/cell) were adjusted, so as to balance the amount of recombinant proteins simultaneously expressed from each virus. At 44 h postinfection, the medium was replaced by a Met- or Met/Cys-free medium and
35S-labeled Met (hRPB1, 2, 6, 7, and 8) or a mixture of Met
and Cys (hRPB5, 9, 10, 10
, and 11) (25 µCi each in 500 µl of
medium) were added. The cells were collected 4 h later and
disrupted by three cycles of freeze-thawing in 500 µl of buffer A (20 mM Tris-HCl, pH 7.5, 20% glycerol, 0.1% Nonidet P-40, 150 mM NaCl, 5 mM
-mercaptoethanol) containing
2.5 ng/ml each of leupeptin, pepstatin, aprotinin, antipain, and
chymostatin (Sigma). The cell lysate was centrifuged (10 min at
14,000 × g), and the cleared supernatant was used for labeled protein analysis.
The 35S-labeled cell
extracts (400 µl) were first incubated with 15 µl of protein
A-Sepharose (Sigma) to adsorb nonspecific binding proteins. The treated
extracts were then incubated with 15 µl of glutathione
(GSH)-Sepharose beads (Pharmacia) that were preblocked in
phosphate-buffered saline containing 0.3% bovine serum albumin. After
2 h at 4 °C, the beads were washed with buffer A. The final
concentrations of salt and detergent were adjusted to minimize
nonspecific interactions; NaCl was 300 (hRPB9), 500 (hRPB1, 2, 3, 6, 7, 8, 10, 10
, and 11) or 1000 mM (hRPB5), and Nonidet
P-40 was 0.1 (hRPB9) or 1% (all the other subunits). After washing,
the beads were boiled in the electrophoresis sample buffer (20 µl)
and the retained proteins were analyzed by SDS-polyacrylamide gel
electrophoresis (SDS-PAGE) (27). After soaking the gel in a 10%
solution of diphenyloxazole in acetic acid, the labeled proteins
were revealed by autoradiography.
A monoclonal antibody (anti-hRPB6) against the purified bacterially synthesized recombinant hRPB6 subunit was produced, as described previously for the anti-hRPB5 antibody (28). Polyclonal antibodies against hRPB3 were raised in rabbits by injecting a mixture of ovalbumin-coupled synthetic peptides spanning residues 131-141 and 207-217. Proteins that were separated by SDS-PAGE and electroblotted onto nitrocellulose filters were reacted with specific antibodies and revealed by a chemiluminescence detection system (NEN Life Science Products), as described (29).
The capacity of two proteins to interact with each other may be examined by a number of different techniques, including the yeast two-hybrid system (30), the co-immunoprecipitation test (31), far Western detection (31), or the GST-pulldown assay (32). The two-hybrid assay has been quite informative in many instances, but, for reasons that are not clear, did not give satisfactory results in the present case; no positive results were obtained, even with combinations of recombinant hRPB subunits that, according to genetic evidence in yeast, were expected to interact. Co-immunoprecipitation experiments have also been used with success to identify interacting protein(s) with the help of antibodies directed against one of the partners (31). Here, an extensive study would require a complete set of antibodies capable of precipitating each hRPB subunit. Furthermore, some antibodies may destabilize or even disrupt the interaction that one is looking for, especially in the case of the smallest subunits. Far Western assays, which involve the probing of a protein blot with a protein of interest, are critically dependent on the efficiency of SDS removal and protein renaturation after SDS-PAGE; less stable interactions may be missed in such a system. For these reasons, we used the GST-pulldown assay in which any interacting subunit would be retained on GSH-agarose beads, together with the GST-tagged partner subunit.
Why Baculovirus as an Expression System?We first tried to
overexpress each of these subunits, as untagged polypeptides or GST
fusions, in bacterial
isopropyl-1-thio--D-galactopyranoside-inducible expression vectors. However, most of the proteins were either produced
as insoluble inclusion bodies or found partially degraded in the
bacterial extracts (data not shown). Therefore, we opted for the
baculovirus expression system; each hRPB cDNA, with the exception
of hRPB4, was cloned into appropriate recombinogenic vectors (under the
control of the baculovirus P10 or polyhedrin promoter) and introduced
into the baculovirus genome by in vivo recombination.
The resulting viruses, upon infection of Sf9 insect cells, allowed efficient synthesis of all subunits, whether free or fused to GST, as both soluble and intact proteins. The only subunit that remained poorly soluble and accumulated in reduced amounts was hRPB2, despite care being taken to coexpress it with hRPB1 (or hRPB3; data not shown), from the same vector. Coexpression was indeed found to improve the yields of proteins otherwise weakly soluble.2 The baculovirus expression system had the additional advantage of providing an eukaryotic environment for protein synthesis and assembly, as opposed to bacterial or in vitro systems.
To facilitate detection of the virally expressed proteins, the infected cells were radiolabeled by growing them in the presence of 35S-labeled methionine or methionine/cysteine, at a late time (44 h) postinfection. Since only a limited number of viral or cellular proteins continue to be synthesized at this period after infection, the labeling pattern is reasonably clean and allows easy visualization of the extraneous polypeptides. Efficient incorporation occurred in all recombinant subunits, allowing their easy identification by SDS-PAGE analysis of a crude infected Sf9 cell extract, even in the case of hRPB11, which contains only one of each of these residues. The overall content of each subunit in Met and Cys residues, whether fused to GST or not (Table II), was taken into account for a rough estimation of the relative amounts of proteins visualized.
|
Sf9
cells were coinfected with pairs of baculovirus recombinants expressing
the various RPB polypeptides, one as a GST fusion and the other as an
untagged protein. Potential interactions between each combination of
tested subunits were scored by the coretention of the untagged subunit
onto the GSH-agarose beads, as revealed by autoradiography after
SDS-PAGE (Fig. 2). In case of ambiguous identification
of a particular band, due to overlapping electrophoretic mobilities of
the tagged and untagged molecules, the proteins were blotted onto
nitrocellulose and probed with specific antibodies (Fig. 2,
B and D, and data not shown). The strength of the
interactions between the subunits was deduced from the relative
intensities of the bands corresponding to the GST-fused subunits and
their non-fused counterparts (see legend to Table
III).
|
Each of the GST-hRPB fusion protein was coexpressed pairwise with every available non-fused hRPB subunit, using this baculovirus system. This allowed the examination of the reciprocal interactions between any tagged subunit with any non-tagged one and thereby allowed us to take into account possible influences of the GST tag on subunit interactions. The results of GST-pulldown experiments such as those shown in Fig. 2 and others (data not shown) are summarized in Table III.
In each series of GST-pulldown experiments, the specificity of the binding reactions was systematically controlled by coexpressing the non-fused GST protein and verifying the lack of interaction with any of the hRPB proteins. In cases where some nonspecific interaction was detected with the GST protein alone (as in Fig. 2A, for example), the amount of nonspecifically bound protein was subtracted from the GST-hRPB-pulldown quantitations.
hRPB3 and hRPB5 Are the Major Anchoring Sites for the Other hRPB SubunitsThree hRPB subunits were repeatedly found in this assay to interact with themselves (hRPB3, hRPB5, and hRPB7), suggesting that they may be present at two (or more) copies within the whole enzyme molecule. Strikingly, together hRPB3 and hRPB5 were able to contact all of the other hRPB subunits, with perhaps the only exception being hRPB9.
We did not analyze separately the interaction pattern of the non-fused
hRPB1 and hRPB2 subunits with the GST fusions. To improve the
solubility of the untagged hRPB2 subunit, it was coexpressed with
hRPB1, from the same recombinant baculovirus. The behavior of both
proteins together was examined in the GST-pulldown assay (Fig.
2A). Because of the likelihood of direct interactions
between these two large subunits, unambiguous interpretation of the
observed results was difficult. However, combining them with those
obtained with the GST-hRPB1 and hRPB2 fusion proteins (panels
B-I), suggested that both hRPB1 and hRPB2 efficiently interact
with hRPB3, whereas only hRPB1 strongly binds to hRPB5. This conclusion
was further corroborated by immunoprecipitation experiments in which
antibodies against hRPB1 effectively coprecipitated hRPB3 or hRPB5,
from extracts of cells coinfected with the corresponding recombinant baculoviruses (data not shown). The hRPB1 subunit also contacts hRPB7,
hRPB10, and, more weakly, hRPB8. By contrast, it is noteworthy that
GST-hRPB8 is the fusion protein that interacts most efficiently with
hRPB1 (Fig. 2A).
The reciprocal interactions analyzed in this study were generally not
dependent on which of the two subunits was fused to GST, indicating
that the GST tag did not critically interfere with the recognition of
the partner subunit. However, a few subunits did seem to be affected in
their interaction capacities by the GST moiety. hRPB5, when fused to
GST, was unable to contact hRPB6 (Fig. 2D) and hRPB10
(Fig. 2G); similarly, fusion of GST to hRPB7 impeded
interaction with hRPB1 (Fig. 2A and data not shown). These observations suggest that the contacts established by the hRPB5 and
hRPB7 subunits may involve their N-terminal portion and that this
domain must remain free to allow efficient interactions. Alternatively,
the binding sites are localized elsewhere on these proteins, but they
may be masked, either directly or allosterically, by the N-terminally
fused GST.
Subunits hRPB6 and hRPB11 exhibit only a few contacts with other subunits, essentially with hRPB3 and hRPB5. Subunit hRPB9 did not show any binding capacity in our GST-pulldown assay, whether tested as a GST fusion (Fig. 2) or as an untagged protein (data not shown). This result suggests that hRPB9 may be assembled into the RPB complex through multiple interactions, each one alone being too weak to be detected by our binding assay. If true, hRPB9 binding may only be observed with larger preassembled complexes. Alternatively, the possibility exists that this subunit, which contains a putative zinc-ribbon motif (33, 34), is not properly folded under the present expression conditions.
The hRPB subunits have been tested pairwise by the GST-pulldown
assay, for their capacity to interact with each other. The observed
interactions have been schematically represented on a two-dimensional
representation of the hRPB (Fig. 3). Interestingly, interactions that were readily detected in extracts from cells that had
been coinfected with two recombinant baculovirus vectors (like
GST-hRPB3 and hRPB5) could not be reproduced after mixing extracts from
cells infected separately with each virus. Although the mixing assay
has not been systematically repeated with each subunit combination,
this result suggests that coexpression of the tested polypeptides is
required for proper folding and optimal interaction between present
subunits. Since subunit coexpression is likely to occur during native
polymerase assembly, this result may reflect a greater specificity of
the interactions as tested by coinfection of cells compared with mixing
of extracts.
The Pattern of hRPB Subunit Interactions Deduced from the Present Study Finds Confirmation in Other Species
Some of the interactions detected in this system between hRPB subunits have previously been documented with their yeast homologues: these results, which were based either on genetic analyses or two-hybrid assays, further reinforce the present conclusions. A mutation (rpoB1) of the gene for RPB1 has been found that facilitates the release of the RPB4·RPB7 heterodimer from the yeast polymerase (35). Our observation that hRPB1 and hRPB7 strongly interact in the GST-pulldown assay may be related to this finding. On the other hand, RPB4 has also been shown to participate to the recruitment of RPB4·RPB7 on the yeast enzyme (36); it will therefore be worth examining also the ability of hRPB4 to interact with hRPB1. The growth defect of another RPB1 mutant (rpo21-4) has been shown to be compensated by overexpression of the RPB6 subunit (35), a result suggesting that both RPB1 and RPB6 subunits are in close contact. The direct, albeit weak interaction that we see between hRPB1 and hRPB6 further supports this conclusion, particularly if one assumes that such contacts may be stabilized in the context of the complete enzyme.
From the study of mutants altered in each of the three largest subunits of yeast RNA polymerase II (RPB1, RPB2, and RPB3), it has been suggested that RPB2 and RPB3 form a complex that subsequently interacts with RPB1 during RPB assembly (37). In agreement with this conclusion, we detect strong interactions between the hRPB1 and hRPB3 as well as between the hRPB2 and hRPB3 subunits, suggesting that human and yeast RPB assembly may follow similar pathways.
Subunits AC40 and AC19 (both of which are shared by yeast RPA and RPC) are capable of interacting in a two-hybrid assay (12), consistent with the direct and strong interaction between hRPB3 (an AC40 homologue) and hRPB11 (an AC19 homologue) detected here. The homologous subunits from Arabidopsis thaliana (RPB3 and RPB11) have recently also been shown to interact both in double-hybrid and co-immunoprecipitation assays (38). That these interactions occur in vivo is suggested by the observation that the effects of yeast mutations affecting either AC40 or AC19 can be compensated by overexpression of AC19 and AC40, respectively (12). Finally, the reality of these interactions was best illustrated by the colocalization of both subunits on the surface of the yeast RPA, as revealed by immunoelectron microscopy (11).
Overexpression in yeast of ABC10, a subunit shared by all three
polymerases (RPA, RPB, and RPC), has been shown to efficiently suppress
a mutation in AC40 (12). Together with our results showing strong
interactions between hRPB10
and hRPB3 (the human homologues of
ABC10
and AC40, respectively), this effect suggests that these
subunits directly contact each other in the natural polymerase
molecule. By contrast, overexpression of ABC10
only partially
compensates for a mutation affecting AC19 (12), suggesting that this
effect might be indirect. Our results indicating that hRPB10
and
hRPB11 do not contact each other directly, but through their common
target hRPB3 (see Fig. 3), are in agreement with such a conclusion.
Two subunits (hRPB3 and hRPB5)
were found to establish nearly 60% (33 over 56) of all contacts
detected. Some properties of these subunits are reminiscent of those of
the Escherichia coli subunit, which, as a dimer,
recruits the two largest subunits,
and
, during prokaryotic RNA
polymerase assembly. Our observation that each of these subunits may
homodimerize further supports the conclusion that they play a central
role in the architecture of the native hRPB complex; clearly, a
stoichiometry of 2 within the whole enzyme may facilitate multiple
contacts with other subunits by providing a larger molecular interface.
Interestingly, RPB3 and RPB5, their yeast homologues, are also
represented with a stoichiometry of 2 in the yeast enzyme (39). A short
stretch of sequence homology, the so called "
-motif," has in
fact been noticed upon comparison of subunits
, RPB3 and AC40 (12).
Mutagenesis of this motif within AC40 confirmed that its integrity is
essential for yeast survival (12). Finally, the ability of hRPB5 to
directly interact with a transcriptional activator, the hepatitis B
virus HBx factor (40), is also reminiscent the activator-binding
property of the prokaryotic
subunit.
The hRPB7 subunit also revealed the capacity to interact with itself, although its yeast homologue, RPB7, is present in the enzyme of cells with a variable stoichiometry ranging from 0.2 to 1, depending on whether the cells are in exponentially growing or stationary phases (41, 42). It will be of interest to examine if hRPB7 interacts with hRPB4 and whether these subunits, like their yeast counterparts RPB7 and RPB4, can readily dissociate from the polymerase without affecting basal transcriptional activity.3
It is usually, and perhaps naively, accepted that the strongest interactions are the most physiologically relevant. This implicit assumption might have to be revised, particularly in the case of multi-subunit complexes. In such complexes the stability of the whole edifice most likely results from the multiplicity rather than strength of contacts established between each component. Thus, in complement to the present pairwise binding measurements, it will be important to examine the binding potential of hRPB subcomplexes of increasing complexity. Together, both approaches should help elucidating the order of subunit assembly during the polymerase biosynthetic pathway.
Strikingly, the two smallest subunits, hRPB10 and hRPB10
, were
found to bind a number of different subunits: given their size (7 and
7.6 kDa, respectively), it is difficult to imagine how these relatively
short polypeptides might establish contacts with such an array of
subunits, in the native enzyme. It is possible, therefore, that some of
the interactions detected with these small subunits are nonspecific.
Alternatively, the detected contacts may represent all possible
interactions that these small subunits may establish with other ones
during polymerase assembly. Some of the interactions occurring at an
early step of this complicated process may actually be excluded later
on, as higher order complexes are being elaborated. In this respect, it
may be relevant that the strong interaction seen between hRPB5 and
hRPB8, when coexpressed alone, is disrupted in the presence of a third
subunit, hRPB6, which establishes tight contacts with hRPB5, but
excludes hRPB8.2
Altogether, these results clearly provide additional confirmation of the interactions observed with the GST-pulldown assay. Furthermore, this opens up avenues toward the reconstitution of an enzymatically active polymerase. The success of such an enterprise will first of all depend on the availability of baculovirus vectors capable of expressing several hRPB subunits, so as to limit as much as possible the number of required recombinant viruses. The construction of such vectors, which has recently been achieved,4 should facilitate coordinate expression of the various subunits. Reconstitution of an entirely recombinant polymerase will allow the detailed functional study of each subunit by use of the powerful in vitro genetics.
The nucleotide sequence(s) reported in this paper has been submitted to the GenBankTM/EMBL Data Bank with accession number(s) X98433[GenBank].
We thank Marguerite Wintzerith for her contribution to the initial steps of this work, R. Gopalkrishnan for critical reading of the manuscript, C. Hauss for expert technical assistance, the IGBMC chemistry staff for oligonucleotides, the antibody staff and S. Vicaire for automated DNA sequencing, and H. Boeuf, B. Chatton, R. Gopalkrishnan, and P. Lutz for helpful discussions.