Synthesis and monitored selection of 5′-nucleobase-capped oligodeoxyribonucleotides

Abstract

Oligodeoxynucleotides bearing 5′-appendages consisting of a nucleobase and an amide linkage were prepared from 5′-amino-5′-deoxyoligonucleotides, amino acid building blocks and thymine or uracil derivatives. Small chemical libraries of 5′-modified oligonucleotides bearing the nucleobase moieties via five, three or two atom linkages were subjected to spectrometrically monitored nuclease selections to identify members with high affinity for target strands. The smallest of the appendages tested, a uracil acetic acid substituent, was found to convey the greatest duplex stabilizing effect on the octamer 5′-T*GGTTGAC-3′, where T* denotes the 5′-amino-5′-deoxythymidine residue. Compared to 5′-TTGGTTGAC-3′, the modified sequence 5′-u-T*GGTTGAC-3′ gives a duplex with 5′-GTCAACCAA-3′ that melts 4°C higher. The duplex-stabilizing effect of this 5′-substituent does not require a specific residue at the 3′-terminus of the complement and the available data suggest that the uracil moiety is located in the major groove of the duplex.

INTRODUCTION

Oligodeoxyribonucleotides are used as primers and hybridization probes in molecular biology, genomics and bioorganic chemistry (1). While they have many favorable properties, such as their ability to predictably engage in Watson–Crick base pairing, they also have properties that are undesirable for most applications. Among these are their vulnerability to enzymatic degradation and the sequence dependence of their affinity for target strands. Therefore, a number of analogs of oligonucleotides have been synthesized, hoping to improve on the properties of their natural counterparts with respect to at least one of these counts (2). Lately, oligonucleotides bearing modifications at their termini have been of particular interest, partly because the main nuclease activity in blood is exonuclease degradation from the terminus (3), partly because the so-called gap technology (4) in the antisense field called for compounds whose interiors are unmodified, so as to recruit RNase H, and partly because molecular beacons bearing chromophores at both termini were discovered (5).

This laboratory has an interest in ‘molecular caps’ for oligonucleotides. These caps are envisioned to be not just modifications that render the termini of oligonucleotides resistant to nuclease attack via steric hindrance, but also contribute to specific recognition of the target strand. So far, this program has focused on appendages consisting of non-nucleic acid moieties, including amino acids and dipeptides (6), aromatic acids (7) and steroids (8). Here we present results from a study aimed at identifying a terminus bearing a nucleobase linked via a non-nucleic acid linker. Thymine and uracil were employed as nucleobases. These pyrimidines were appended via an amide-containing linker to an amino-terminal oligonucleotide, leading to the general structure shown in Scheme 1. An optimized linker was sought via combinatorial synthesis and mass spectrometrically monitored nuclease selections, a technique that had recently been developed in these laboratories for identifying terminally modified oligonucleotides with increased affinity for target strands (7). Such selection experiments employ small chemical libraries or chemsets of up to 10 members and require sub-nanomolar quantities of crude oligonucleotides. The initial structural search performed here led to a molecular cap with a short linker that can enhance the affinity for a target strand without requiring a specific nucleobase at the 3′-terminus of the target.

Scheme 1. Schematic representation of the terminal portion of a duplex of a 5′-nucleobase-capped oligonucleotide with its target strand. The linker portion to be optimized in this work is shown as a box.

MATERIALS AND METHODS

General

DMF, N-ethyldiisopropylamine and 1-hydroxybenzotriazole hydrate were from Fluka; pyridine, thymine, uracil, ethyl bromoacetate, ethyl 3-bromopropionate, 2,4,6-trihydroxyacetophenone monohydrate (THAP) and diammonium citrate were from Aldrich; dichloromethane was from VWR; THF and triphenylphosphine were from Riedel-deHaen. All reagents were used as received. Reagents for DNA synthesis were from Perseptive Biosystems, except for dG^dmf, which was from ABI/Perkin Elmer. Amino acid building blocks and HBTU were from Perseptive Biosystems. Fmoc-Lys(Alloc)-OH was from Novabiochem and cholic acid, phosphodiesterase I (EC 3.1.4.1, type II, from Crotalus adamanteus venom) and phosphodiesterase II (EC 3.1.16.1, type I-SA, from bovine spleen) were from Sigma. The 5′-monomethoxytritylamino-5′-deoxythymidine phosphoroamidite was synthesized as previously described (9,10). MALDI TOF spectra were recorded on Bruker Biflex spectrometers in negative, linear mode, with matrices made up of THAP (0.3 M in ethanol) and diammonium citrate (0.1 M in water) (2:1 v/v). EI MS were recorded on a Varian MAT 312. HPLC was on a 250 × 4 mm Vydak Protein C4 column. Gradients of CH₃CN (solvent B) in 0.1 M triethylammonium acetate buffer, pH 7, were used. NMR spectra were recorded on Bruker DPX 300 or Bruker AC 250 spectrometers. The following abbreviations are used in sequences of modified oligonucleotides: T*, 5′-amino-5′-deoxythymidine residue; t, thymine acetic acid residue; t*, thymine propionic acid residue; u, uracil acetic acid residue.

(2,4-Dioxo-3,4-dihydro-2H-pyrimidin-1-yl)-acetic acid (8)

Uracil (100 mg, 0.89 mmol) and NaOH (107 mg, 2.67 mmol) were dissolved in water (1 ml) and stirred for 30 min at 22°C. Ethyl bromoacetate (99 µl, 0.89 mmol) was added in one portion and the resulting slurry was vigorously stirred at room temperature for 24 h. The resulting clear solution was acidified with a 3 M aqueous HCl solution to pH 3. After 1 h, the resulting colorless precipitate was filtered off, washed three times with ice-cold water (5 ml) and dried under vacuum overnight. The title compound (8) was obtained after re-crystallization from water. Yield 103 mg (68%). ¹H NMR (DMSO-d₆), p.p.m.: δ 13.25 (s, 1H), 11.30 (s, 1H), 7.60 (d, ³J = 7.8 Hz, 1H), 5.57 (d, ³J = 7.8 Hz, 1H), 4.39 (s, 2H); ¹³C NMR (DMSO-d₆), p.p.m.: δ 169.5, 163.7, 151.0, 146.0, 100.8, 48.6; MS (EI, 70 eV) m/z 171 (M⁺, 9%), 127 ([M-CO₂]⁺, 25%).

Ethyl 3-(5-methyl-2,4-dioxo-3,4-dihydro-2H-pyrimidin-1-yl)-propionate (9)

Thymine (50 mg, 0.40 mmol) and DBU (0.12 ml, 0.80 mmol) were mixed in pyridine (1 ml) and stirred at room temperature until all thymine was dissolved. Ethyl 3-bromopropionate (51 µl, 0.40 mmol) was added. The solution was stirred for 2 h and then diluted with water (20 ml) and neutralized with ice-cold AcOH. The solution was saturated with NaCl and the product extracted with ethyl acetate (3 × 10 ml). The combined organic phases were washed twice with a solution of HCl (0.3 M in water) and then twice with water. After evaporation of the solvent, a white solid was obtained. Flash chromatography (silica, 6–10% MeOH in CHCl₃) afforded 9. Yield 84 mg (94%), R_f = 0.7 (silica, 15% MeOH/2% AcOH/CHCl₃). ¹H NMR (CDCl₃), p.p.m.: δ 9.32 (s, 1H), 7.13 (s, 1H), 4.08 (q, ³J = 7.1 Hz, 2H), 3.89 (t, ³J = 5.7 Hz, 2H), 2.70 (t, ³J = 5.7 Hz, 2H), 1.84 (s, 3H); ¹³C NMR (CDCl₃), p.p.m.: δ 171.35, 164.36, 150.88, 141.57, 110.21, 61.05, 44.98, 33.12, 14.08, 12.21; MS (EI, 70 eV) m/z 226 (M⁺, 100%), 181 ([M-C₂H₅O]⁺, 50%), 152 ([M-H₂O-CO-C₂H₅O]⁺, 31%).

3-(5-Methyl-2,4-dioxo-3,4-dihydro-2H-pyrimidin-1-yl)-propionic acid (10)

Compound 9 (84 mg, 0.37 mmol) was suspended in a solution of NaOH (1 ml, 1 M) at 22°C. The suspension was stirred until the entire solid had dissolved (∼1 h). The solution was then neutralized with 3 M aqueous HCl, concentrated to 0.5 ml and cooled to 4°C. After 5 h, colorless crystalline 10 was filtered off in the cold, washed with ice water and dried under high vacuum (0.01 Torr). Yield 71 mg (97%). ¹H NMR (DMSO-d₆), p.p.m.: δ 12.41 (s, 1H), 11.23 (s, 1H), 7.50 (s, 1H), 3.80 (t, ³J = 7.0 Hz, 2H), 2.58 (t, ³J = 7.0 Hz, 2H), 1.72 (s, 3H); ¹³C NMR (DMSO-d₆), p.p.m.: δ 172.3, 164.3, 150.8, 141.9, 108.1, 43.9, 32.8, 11.9; MS (EI, 70 eV) m/z 198 (M⁺, 48%), 152 ([M-H₂O-CO]⁺, 40%), 139 ([M-C₂H₃O₂]⁺, 17%), 126 ([M-C₃H₆O₂]⁺, 32%).

Assembly of the DNA portion of oligonucleotides

Non-modified oligodeoxyribonucleotides were synthesized on an ABI 381 DNA synthesizer and purified by reverse phase HPLC on an Alltech C4 column. 5′-Amino-5′-oligodeoxyribonucleotides were synthesized in an analogous fashion, except that the last coupling step employed the aminothymidine building block. The yields of all modified DNA strands are based on integration of the product peak in HPLC traces of crude products.

Coupling of carboxylic acids to amino-terminal oligonucleotides

The following procedure for the preparation of 5i is representative. A mixture of Fmoc-Lys(Alloc)-OH (6.4 mg, 15 µmol), HBTU (6.8 mg, 13.5 µmol) and HOBT (2.6 mg, 15 µmol) was dried under vacuum, dissolved in DMF (0.2 ml) and treated with DIEA (5.8 µl, 33 µmol). After vortexing, the slightly darkened solution was added to a polypropylene reaction vessel containing DNA-bearing controlled pore glass (CPG) 4 (5.0 mg, 0.15 µmol loading for the first 3′-terminal nucleoside of the DNA). The slurry was vortexed every 5 min during the coupling reaction. After 30 min, the solution was decanted, the CPG washed three times with DMF (1 ml each) and dried in vacuo. An analytical sample of the aminoacylated DNA was prepared by treatment with NH₄OH (50 µl) overnight, followed by dilution with water (150 µl), lyophilization and DE-MALDI analysis in negative mode. According to semi-quantitative analysis of the resulting MALDI TOF spectrum, the coupling efficiency was ≥90%. MALDI TOF MS for H-Lys(Alloc)-T*GCGCAA: calculated 2316.9 ([M-H]^–), found 2316.1.

Mixed coupling reactions

Chemsets 1, 2 and 4 were created in mixed coupling reactions using procedures similar to the one given above. Ratios of amino acid building blocks were adjusted according to the signal intensities observed in the crude products of exploratory coupling reactions performed with equimolar mixtures of Fmoc-amino acid building blocks. For example, a mixture of Fmoc-Gly-OH (1.9 mg, 6.4 µmol), Fmoc-Pro-OH (4.2 mg, 12.4 µmol), Fmoc-Asp(OBzl)-OH (4.2 mg, 9.4 µmol), Fmoc-Phe-OH (3.2 mg, 8.3 µmol), Fmoc-Trp-OH (5.3 mg, 12.4 µmol), HBTU (16.7 mg, 44 µmol), HOBT (7.5 mg, 49 µmol), DIEA (18 µl, 108 µmol) and DMF (1.2 ml) was used in the synthesis of chemset 1.

Removal of Fmoc protecting groups

DNA-bearing CPG (∼5.0 mg, 0.15 µmol loading of 3′-terminal nucleoside) was treated with a solution of piperidine in DMF (1:4 v/v, 0.5 ml) for 30 min in a polypropylene reaction vessel. The solution was decanted, the CPG washed three times with DMF (1 ml each) and dried in vacuo.

Removal of Alloc protecting groups

Pd(PPh₃)₄ (2.5 mg, 2.2 µmol), PPh₃ (0.5 mg, 2.0 µmol) and [Et₂NH₂]⁺[HCO₃]^– (2.5 mg, 18.5 µmol) in CH₂Cl₂ (0.4 ml) were mixed, the solution purged with Ar and added to a sample of solid support 6i (5 mg, ∼15 µmol 3′-terminal nucleoside). The slurry was vortexed every 5 min and the solution decanted after 1 h. The support was thoroughly washed with DMF (1.5 ml), 1% TEA solution in DMF (1.5 ml), 0.5% ethyl dithiocarbamate solution in DMF (1.5 ml) and DMF (2 × 1.5 ml) to remove residual Pd salts. After drying, the support was either used directly for coupling or deprotected with NH₄OH to produce 7i.

Full deprotection and release from support

All oligonucleotides reported were fully deprotected and cleaved from the support with NH₄OH at room temperature for 16 h.

Compound 7a (t-Gly-T*GCGCAA)

Yield 38%; HPLC, 0% B for 5 min, in 45 min to 20% B, in 5 min 20–50% B, 10 min 50–100% B, R_t = 40.4 min. MALDI TOF MS for C₇₇H₉₅N₃₂O₄₂P₆ [M–H]^–: calculated 2326.6, found 2325.0.

Compound 7d (t-Phe-T*GCGCAA)

Yield 47%; HPLC, 0% B for 5 min, in 45 min to 20% B, in 8 min to 100% B, R_t = 47.4 min. MALDI TOF MS for C₈₄H₁₀₁N₃₂O₄₂P₆ [M–H]^–: calculated 2416.8, found 2415.0.

Compound 7i (t-Lys-T*GCGCAA)

Yield 31%; HPLC, 0% B for 5 min, in 45 min to 25% B, in 30 min 25–100% B, R_t = 53.4 min. MALDI TOF MS for C₈₁H₁₀₄N₃₃O₄₂P₆ [M–H]^–: calculated 2397.8, found 2396.7.

Compound 15a (t-T*GGTTGAC)

Yield 54%; HPLC, 0% B for 5 min, in 35 min to 20% B, in 19 min 20–100% B, R_t = 34.0 min. MALDI TOF MS for C₈₆H₁₀₆N₃₂O₅₀P₇ [M–H]^–: calculated 2604.8, found 2604.3.

Compound 15g (t*-T*GGTTGAC)

Yield 50%; HPLC, 0% B for 5 min, in 40 min 0–30% B, in 10 min 20–100% B, R_t = 29.1 min. MALDI TOF MS for C₈₇H₁₀₈N₃₂O₅₀P₇ [M–H]^–: calculated 2618.8, found 2618.8.

Compound 15h (u-T*GGTTGAC)

Yield 51%; HPLC, 0% B for 5 min, in 35 min to 20% B, in 19 min 20–100% B, R_t = 32.1 min. MALDI TOF MS for C₈₅H₁₀₄N₃₂O₅₀P₇ [M–H]^–: calculated 2590.8, found 2594.2.

Compound 17b (Ac-T*GCGCAA)

Yield 52%; HPLC, 0% B for 5 min, in 65 min to 20% B, in 50 min 20–50% B, in 3 min 50–100% B, R_t = 47.4 min. MALDI TOF MS for C₇₀H₈₈N₂₉O₃₉P₆ [M–H]^–: calculated 2145.5, found 2146.8.

Compound 17c (Ac-Phe-T*GCGCAA)

Yield 49%; HPLC, 0% B for 5 min, in 35 min to 20% B, in 10 min 20–100% B, R_t = 40.7 min. MALDI TOF MS for C₇₉H₉₇N₃₀O₄₀P₆ [M–H]^–: calculated 2292.7, found 2292.9.

Compound 17d (u-T*GCGCAA)

Yield 56%; HPLC, 0% B for 5 min, in 65 min to 20% B, in 10 min to 100% B, R_t = 26.4 min. MALDI TOF MS for C₇₄H₉₀N₃₁O₄₁P₆ [M–H]^–: calculated 2255.6, found 2255.9.

NMR spectroscopy

Hybrid 17a (t-T*GCGCAA) was re-lyophilized three times from 10% ammonia in water (1 ml each) to remove residual triethylamine from HPLC, followed by lyophilization from D₂O. It was dissolved in D₂O (250 µl total volume) containing 10 mM phosphate buffer (KH₂PO₄/Na₂HPO₄), pH 7, uncorrected for the deuterium effect and 150 mM NaCl. Spectra were acquired at 0.4 mM strand concentration in a microtube (Shigemi Co., Tokyo, Japan). NOESY and COSY spectra were acquired at 22°C with suppression of the residual solvent peak by presaturation for 1.0 s during the recycle delay. The NOESY spectrum was acquired with 250 ms mixing time at 3000 Hz spectral width with 2048 data points in t2, 128 increments in t1 and 256 scans per increment. Data were zero filled to 1024 data points in t1, followed by apodization with Gaussian line broadening and exponential line narrowing and were analyzed using XWINNMR.

Compound 17a (t-T*GCGCAA)

Yield 43%; HPLC, 0% B for 5 min, in 65 min to 25% B, in 10 min 20–100% B, R_t = 53.2 min. MALDI TOF MS for C₇₅H₉₂N₃₁O₄₁P₆ ([M–H]^–): calculated 2269.6, found 2269.4; ¹H NMR (selected resonances, 300 MHz, D₂O), p.p.m.: δ 8.308 (H8-A8), 8.223 (H8-G3), 8.130 (H8-A7), 7.991 (H8-G5, H2-A8), 7.941 (H2-A7), 7.439 (H6-C4), 7.415 H6-C6), 7.389 (H6-t1), 7.307 (H6-T2), 6.360 (H1′-A8), 6.208 (H1′-G3), 6.186 (H1′-A7), 5.895 (H1′-T2), 5.880 (H1′-G5), 5.842 (H1′-C4), 5.823 (H1′-C6), 5.538 (H5-C4), 5.512 (H5-C6), 5.132 (H3′-G3), 5.049 (H3′-G5), 4.966 (H3′-A7/A8), 4.956 (H3′-C4), 4.766 (H3′-C6), 4.752 (H5′-G3), 4.747 (H3′-T2), 4.625 (H4′-G3), 4.570 (CH₂-t1), 4.443 (H4′-G5), 4.366 (H5′/H5′′-A8), 4.327 (CH₂-t1), 4.323 (H5′/H5′′-A7), 4.288 (H5′/H5′′-C6), 4.284 (H5′/H5′′-C4), 4.280 (H4′-T2), 4.200 (H5′/H5′′-G5), 4.190 (H5′′-G3), 3.489 (H5′/H5′′-T2), 2.972/2.939 (H2/H2′′-G3), 2.815/2.623 (H2′/2′′-A8), 2.678 (H2′/H2′′-G5), 2.650 (H2′/H2′′-A7), 2.481/2.096 (H2′/2′′-C4), 2.273/1.708 (H2′/2′′-T2), 2.250/2.029 (H2′/H2′′-C6), 1.805 (CH₃-t1), 1.559 (CH₃-T2).

Molecular modeling

Force field minimizations were performed with Macromodel (11) using the Amber force field (12) and default parameter settings, starting from a B-form conformation for the core DNA duplex T*GCGCAA:TTGCGCA and either a base paired state for t1 and A8 (structure shown in Fig. 2a) or B-form conformation for A8 and a random unbound conformation for t1 (structure shown in Fig. 2b). For the latter, interproton distance constraints of r_ij = 4.5–5.0 were employed, based on the weak NOESY cross-peaks observed. The structure in Figure 2c was obtained by replacing t in (t-T*GCGCAA)₂ with u, restraining the T2:A7 base pair to torsion angles for B-form DNA and minimizing without other restraints.

Figure 2.

Terminal regions of force field minimized structures of the duplexes 5′-X-T*GCGCAA-3′:5′-TTGCGCAA-3′. In (a) and (b) X is a thymine acetic acid residue and in (c) X is a uracil acetic acid residue. Structure (a) was obtained using Watson–Crick base pair geometry between the terminal nucleobases and the default B-form conformation from Macromodel for the core duplex as starting conformation. Structure (b) was obtained similarly to (a), except that the thymine acetic acid residue was unbound in the starting conformation and four strategic NMR-based distance restraints were used for this residue during the calculation. Structure (c) was obtained by replacing the thymine in (b) with uracil and minimizing the resulting duplex. Color code: yellow, 5′-appended nucleobase acetic acid residues; orange, A residues; dark green, T residues; gray, G residues; silver, C residues. The figure was created with VMD (35) using the Raster3D module.

Nuclease selection from chemsets 1–3

The following protocol for selection from chemset 1 is representative. A solution (11 µl total volume) of chemset 1 (∼200 pmol of each library member) in 80 mM NH₄OAc was sampled (1 µl) for the control spectrum and a solution (1 µl, 8 × 10^–5 U) of phosphodiesterase I was added at 22°C. Samples (1 µl each) were taken at chosen time points and immediately thoroughly mixed with the matrix mixture of THAP (5 µl, 0.3 M in ethanol), diammonium citrate (2 µl, 0.1 M in water) and internal standard 5′-CCGTGGTTGAC-3′ (0.5 µl solution in water, 10 pmol). An aliquot of the resulting mixture (1 µl) was placed on the MALDI target, the solvents evaporated and spectra acquired in negative, linear mode at 20 kV total extraction voltage with delayed extraction 60 ns after the instrument response time. Between four and five spectra from 100–150 laser shots of ∼50–60 µJ/shot were acquired per sample. Spectra in which the internal standard yielded less than 1000 ion counts were discarded. Data analysis was performed with the computer routines DAS and Automaton, available from the authors’ web page (currently at http://microvirus. chem.tufts.edu/downloads.html or http://134.34.110.31/htdocs/downloads.html ) (7).

Selections from chemsets of non-self-complementary sequences were similar, except that an equimolar mixture of the components of the chemset (∼50 pmol each) and one equivalent of target strand 18 (50 pmol) were annealed by heating to 90°C, cooling to 22°C in 40 min and then to 0°C, at which temperature the assay was performed. A solution with 4 × 10^–4 U phosphodiesterase I was used.

Nuclease accessibility of the 3′-terminal residue of the target

Three individual solutions containing either TTGGTTGAC (16) and 18, u-T*GGTTGAC (15h) and 18 or TGGTTGAC (20) and 18 (100 pmol of each per solution) containing NH₄OAc (80 mM) were prepared (13.3 µl total volume each). The solutions were annealed as described above and assayed at 0°C using phosphodiesterase I (3 × 10^–4 U). The degradation kinetics of 18 in each solution were measured, based on the decrease in its relative signal intensity (18/internal standard).

Resistance to nuclease attack at the 5′-terminus

A solution of u-T*GGTTGAC (15h), t-T*GGTTGAC (15a), t*-T*GGTTGAC (15g) and TTGGTTGAC (16) (50 pmol of each component) in NH₄OAc buffer (80 mM, 11 µl total volume) was brought to 22°C, sampled (0.5 µl) for a control spectrum and then treated with a solution of phosphodiesterase II (1 µl, 2 × 10^–3 U) as described above.

UV melting experiments

Melting and cooling curves were recorded on a Perkin-Elmer Lambda 10 UV-visible spectrometer by measuring absorbance at 260 nm in a 1 cm path length at a heating or cooling rate of 1°C/min. Prior to acquisition of the melting curves, samples were annealed by heating to 85°C for 15 min and cooling to 5°C at a rate of 2°C/min. Melting temperatures were determined with UV TempLab 1.2 as the extrema of the first derivative of the smoothed melting curves. Melting points are the average ± SD from two heating and two cooling curves. Thermodynamic data were calculated using Meltwin (13).

RESULTS

The preparation of 5′-modified oligonucleotides started from commercially available CPG bearing the 3′-terminal nucleoside (1, Scheme 2). DNA synthesis followed phosphoramidite protocols, including coupling of the 5′-amino-5′-deoxythymidine building block 3 (9,10) to intermediate 2 to produce, after detritylation, amino-terminal oligomer 4 required for coupling the 5′-terminal appendages. The initial design of the appendages used PNAs (14) as a model, in that it contained an amino acid (glycine in unmodified PNAs) as one of the backbone units (15–17) and an acetyl linker between backbone and nucleobase. A series of Fmoc-amino acid building blocks was coupled to 4 under conditions similar to those previously employed for other hybrids (18) to give aminoacylated intermediates 5a–5j. These were deprotected with piperidine in DMF and coupled to thymine acetic acid with the ‘uronium salt’-containing activation mixture HBTU, HOBT and DIEA. Fmoc-Lys(Alloc)-bearing oligonucleotide 6i was Alloc deprotected with [Pd(PPh₃)₄] and diethylammonium hydrogen carbonate in CH₂Cl₂ (19), followed either by full deprotection with NH₄OH or coupling of the ɛ-amino group to another molecule of thymine acetic acid to yield, after deprotection, compound 7j. Following this general procedure, the oligonucleotide hybrids shown in Scheme 2 were obtained. Chemsets 1 and 2 were produced via mixed couplings, whereas crude products from parallel syntheses were combined to give chemset 3.

Scheme 2. (a) Standard phosphoramidite protocol; (b) coupling cycle with 3; (c) Fmoc-Aa-OH, HOBT, HBTU, DIEA, DMF; (d) (1) piperidine, DMF; (2) thymine acetic acid, HOBT, HBTU, DIEA, DMF; (e) NH₄OH; (f) (1) [Pd(PPh₃)₄], PPh₃, [H₂NEt₂]⁺[HCO₃]^–, CH₂Cl₂; (2) NH₄OH; (g) (1) [Pd(PPh₃)₄], PPh₃, [H₂NEt₂]⁺[HCO₃]^–, CH₂Cl₂; (2) thymine acetic acid, HOBT, HBTU, DIEA, DMF; (3) NH₄OH.

The chemsets were subjected to MALDI TOF-monitored nuclease selection assays (7) to identify components with high affinity for complementary strands. Such selections rely on the greater resistance of duplex DNA to single strand-specific nucleases compared to single-stranded DNA, so that increased duplex stability can be detected in the prolonged survival of high affinity compounds (7). Exposing chemset 1 to snake venom phosphodiesterase (phosphodiesterase I) led to degradation of its members at similar rates. Kinetics of degradation of the full-length oligomer (Fig. 1) revealed that glycine-containing 7a was slightly more nuclease-exposed than its side chain-bearing analogs 7b–7e, while phenylalanine-containing 7d was the component most protected from nuclease attack. Even though chemset 1 was considered diverse, including members with flexible (Gly), conformationally constricted (Pro), polar (Asp) and aromatic residues (Phe and Trp) (20), no clear hit was found. Numerically, the relative affinity of these compounds, as determined in the nuclease assays, can be represented in the form of ‘protection factors’, as discussed in detail by Altman et al. (7). Comparing the protection factors obtained (Table 1) with those found in a previous study employing the nuclease selection methodology (7) suggested that there should be no more than a few degrees difference in the melting points of the duplexes formed by the components of chemset 1. This was confirmed in melting point experiments with the ‘loser’ (7a) and ‘winner’ (7d) of the assay. Table 2 shows the UV melting points obtained for HPLC purified compounds, together with that of the unmodified control octamer 5′-TTGCGCAA-3′ (21). The duplex of phenylalanine-containing 7d melts higher than that of glycine-containing 7a at all salt concentrations employed and the melting points of (7d)₂ are similar or slightly lower than that of (21)₂.

Figure 1.

Nuclease selection from chemset 1. Kinetics of disappearance of the full-length oligonucleotides from solution upon treatment with phosphodiesterase I, as determined by quantitative MALDI TOF mass spectrometry. Data points are averages ± SD of the relative signal intensities (analyte/internal standard) from four spectra per data point, normalized to the signal ratio in the control spectra of the undigested library.

Table 1. Protection factors obtained from MALDI-monitored nuclease selection with snake venom phosphodiesterase.

Compound	5′-Residue	Target stranda	PFb,c	PFco+truncd
Chemset 1
7a	t-Gly	s.c.	1.0	1.0
7b	t-Pro	s.c.	1.3	1.6
7c	t-Asp	s.c.	1.2	1.4
7d	t-Phe	s.c.	1.4	1.8
7e	t-Trp	s.c.	1.3	1.6
Chemset 2
7a	t-Gly	s.c.	1.3	1.5
7f	t-Ala	s.c.	1.5	1.8
7g	t-Val	s.c.	1.0	1.0
7h	t-Tyr	s.c.	1.4	1.6
7e	t-Trp	s.c.	1.8	2.3
Chemset 3
7f	t-Ala	s.c.	1.0	n.d.e
7i	t-Lys	s.c.	1.1	n.d.
7j	t-Lys(t)	s.c.	1.0	n.d.
Chemset 4
15a	t	18	1.4	1.6
15b	t-Gly	18	1.0	1.0
15c	t-Pro	18	1.3	1.4
15d	t-Asp	18	1.1	1.1
15e	t-Phe	18	1.1	1.2
15f	t-Trp	18	1.2	1.3
Chemset 5
16	T	18	1.6	1.7
15a	t	18	1.2	1.2
15g	t*	18	1.0	1.0
15h	u	18	1.4	1.5

^as.c., self-complementary sequence.

^bSelections with snake venom phosphodiesterase at 22°C.

^cProtection factor.

^dProtection factor after cut-off and truncation correction; cut-off values were 26% (chemset 1), 19% (chemset 2) and 8% (chemsets 4 and 5). For a discussion of protection factors and their correction in case of incomplete degradation see Altman et al. (7).

^eNot determined.

Table 2. Melting temperatures of duplexes of self-complementary oligonucleotides in phosphate buffer^a.

Oligonucleotide	NaCl concentration (M)
	0	0.15	1.0
5′-TTGCGCAA-3′ (21)	21.1 ± 0.5	35.4 ± 0.6	40.8 ± 0.6
5′-t-Gly-T*GCGCAA-3′ (7a)	17.6 ± 0.7	30.0 ± 0.7	35.7 ± 0.5
5′-t-Phe-T*GCGCAA-3′ (7d)b	19.9 ± 0.1	32.0 ± 0.7	40.7 ± 0.3
5′-t-Lys-T*GCGCAA-3′ (7i)	20.1 ± 0.9	29.0 ± 1.0	34.4 ± 0.5
5′-t-T*GCGCAA-3′ (17a)	19.7 ± 0.8	35.0 ± 1.3	39.5 ± 0.5
5′-Ac-T*GCGCAA-3′ (17b)	15.9 ± 1.2	28.4 ± 0.8	33.5 ± 0.8
5′-Ac-Phe-T*GCGCAA-3′ (17c)	15.7 ± 0.9	32.0 ± 0.5	37.3 ± 1.2
5′-u-T*GCGCAA-3′ (17d)	23.4 ± 0.8	34.0 ± 0.5	37.8 ± 1.1

^aMelting temperatures are the avarage ± SD of four individual curves obtained at 1.1 µM strand concentration and 10 mM phosphate buffer, pH 7.

^bThis compound showed a second transition at higher temperature, indicating that a higher order structure was melting at low temperature.

Nuclease selection among the members of chemset 2, containing two reference compounds from chemset 1 together with others containing alanine (7f), β-branched valine (7g) and tyrosine (7h), gave results similar to those obtained with chemset 1 (Table 1). Only because the protection factors obtained were now based on the valine-containing 7g with a poor performance were their absolute values higher than those of the first assay. Similar results were obtained in selection experiments for five or six member libraries containing other amino acids (results not shown). In fact, both glutamine- and arginine-containing oligonucleotide derivatives were more rapidly degraded than competitors containing aliphatic amino acid side chains. Chemset 3, including 7i with a cationic residue and 7j with a lysine residue with a second thymine acetic acid residue attached on its side chain, again did not provide a clear winner in the selection (Table 1).

Since modifications of the backbone side chains did not appear to have a major effect on the stability of duplexes, changes in the linker and nucleobase structures were made. In particular, derivatives bearing the terminal nucleobase on a shorter linker were prepared to avoid large entropic penalties for engaging the 5′-appendage in binding. The structural changes included removal of the amino acid moiety and including a propionic acid in addition to the acetic acid subunit. Further, a uracil derivative was included in the building block collection (8, Scheme 3), based on the results of molecular modeling. When a duplex including a terminal base pair between the 5′-terminal thymine acetic acid residue and a 3′-terminal adenosine residue was force field minimized, a steric conflict was detected between the methyl groups of the appended thymine and that of the penultimate aminodeoxythymidine residue of the same strand (Fig. 2a).

Scheme 3. (a) NaOH, H₂O; (b) DBU, pyridine; (c) NaOH, H₂O.

The preparation of a library containing members with both amino acid-linked and directly linked nucleobase–acid building blocks is shown in Scheme 4. The DNA portion of the sequences was chosen to be non-self-complementary to facilitate studies on mismatch discrimination at the terminus (8), which would otherwise require the synthesis of a series of modified oligonucleotides. Synthesis started from the cytidine-bearing CPG 11, whose extension via 12 and 13 yielded the protected oligonucleotide derivatives 14a–14f, obtained from 13 in one mixed coupling with five Fmoc-amino acid building blocks. Chemset 4, containing the directly linked thymine acetic acid derivative 15a together with a number of amino acid residue-containing members, was the first of the new libraries to be subjected to nuclease selection. The results from this assay (Fig. 3 and Table 1) demonstrated that oligonucleotide 15a with its directly linked thymine acetic acid residue did indeed fare slightly better than the phenylalanine-containing hybrid 15e.

Scheme 4. (a) Standard phosphoramidite protocol; (b) coupling cycle with 3; (c) Fmoc-Aa-OH or nucleobase building block, HOBT, HBTU, DIEA, DMF; (d) NH₄OH; (e) (1) piperidine, DMF; (2) thymine acetic acid, HOBT, HBTU, DIEA, DMF; (3) NH₄OH.

Figure 3.

Nuclease selection from chemset 4. Kinetics of disappearance of the full-length oligonucleotides from solution upon treatment with phosphodiesterase I, as determined by quantitative MALDI TOF mass spectrometry. Data points are averages ± SD of the relative signal intensities (analyte/internal standard) from four spectra per data point, normalized to the signal ratio in the control spectra of the undigested library.

To validate the small protection factor differences from the nuclease selection involving chemset 4, control UV melting curves were acquired. In order to compare the melting points of this series with those obtained with components of chemsets 1 and 3, thymine acetic acid-bearing self-complementary oligonucleotide 17a was synthesized (Scheme 5). Its UV melting points were indeed as high as those of the highest affinity compound among the amino acid-containing oligonucleotides, phenylalanine derivative 7d (Table 2). Derivatives lacking either the thymine acetic acid residue (17c, Scheme 5) or both the thymine acetic acid and the phenylalanine residue (17b) on average fared more poorly than 7d, indicating that both residues contributed to duplex stability (Table 2). Compound 7i, with its lysine-linked thymine acetic acid residue, also gave the lower melting points expected based on the nuclease selection results, except at very low salt concentration, where ion pairs between the cationic side chain functionality and the phosphodiester anions of the DNA are strong.

Scheme 5. (a) Thymine acetic acid or NaOAc or Fmoc-Phe-OH or uracil acetic acid, HOBT, HBTU, DIEA, DMF; (b) NH₄OH; (c) (1) piperidine, DMF; (2) NaOAc, HOBT, HBTU, DIEA, DMF; (3) NH₄OH. Control DNA oligomer 21 was synthesized via a standard phosphoramidite DNA synthesis protocol.

Given that thymine acetic acid-bearing 17a gives almost the same duplex melting points as the control DNA octamer duplex (TTGCGCAA)₂, (21)₂, even though it has a different number of backbone atoms between the 5′-terminal and penultimate nucleobase than DNA, RNA or PNA (PNA has previously been shown to be largely intolerant of changes in the contour length of its backbone; see for example ref. 21), we decided to study 17a in greater detail. The ¹H NMR spectrum of a 0.4 mM solution in D₂O was partially assigned, (Fig. S11, Supplementary Material), allowing monitoring of duplex melting at the level of individual residues. Figure 4 shows chemical shift increases upon heating for protons from the terminal and penultimate bases. The downfield shifts upon duplex dissociation are considerably smaller for the protons from the terminal base pair than those for the penultimate base pair, which in turn are only as large as those in the terminal T:A base pair of the aminoacylated DNA duplex (Trp-T*GCGCAC)₂, whose NMR structure was determined previously in these laboratories (20). The melting point determined from the chemical shift data of all four protons in 17a were similar, however, suggesting that all four nucleobases are engaged in formation of a complex that melts cooperatively.

Figure 4.

Duplex dissociation, as monitored by chemical shift changes for selected protons from the nucleobases at the penultimate positions of (t-T*GCGCAA)₂, (17a)₂. Shift changes are noted relative to their position at 22°C. Data are from spectra recorded at 300 MHz (0.4 mM solution in D₂O, 10 mM phosphate buffer, 150 mM NaCl).

NOESY cross-peaks obtained from a spectrum acquired at 300 MHz included one between the two methyl groups of t1 and T2, one between H6 of t1 and CH₃7 of T2 and one between the methylene group of the t1 acetyl linker and H2′/2′′ (no stereoselective assignment of the diastereotopic protons was attempted) of T2. Further, a cross-peak could be discerned between CH₃7 of t1 and H5 of C6 (Fig. S13, Supplementary Material). The latter is a proton located in the major groove of the core duplex, next to the T2:A7 base pair. Though this cross-peak was not very strong, it did appear on both sides of the diagonal and was clearly above the noise. A force field minimization starting from B-form DNA employing distance constraints from the above cross-peaks led to the structure shown in Figure 2b. In this structure the nucleobase of the t1 residue is located in the major groove, forming what would usually be called a base triplet. In this arrangement the thymine contacts the G3:C6 base pair with its H6/CH₃7 edge rather than with the ‘Watson–Crick edge’ capable of Hoogsten-type binding. Though the results from these force field minimizations have to be treated with great caution, since they assume an otherwise undisturbed B-DNA structure, they did provide a plausible picture of how the terminal residues may stabilize the duplex without engaging in Watson–Crick base pairing. Based on this picture, two predictions could be made: (i) that poor mismatch discrimination between nucleobases presented to t1 at the 3′-terminus of the opposite strand should be found; (ii) that substituting uracil for thymine would lead to a larger contact area between residue 1 and G3:C6 and thus to tighter binding. The latter was based on the observation that the packing of the methyl group of t1 against the major groove leaves a hole between H6 of t1 and H8/N7 of G3 (Fig. 2b).

Nuclease selection from chemset 5 was performed next. This library contains one compound with the acetyl-linked thymine residue (15a), one with its propionyl derivative (15g), one uracil derivative (15h) and control DNA 16. The selection yielded 15h as the ‘winner’ among the modified oligonucleotides (Fig. 5 and Table 1). Unmodified 16 did persist slightly longer in the nuclease-containing solution, but it was known from previous studies that unmodified DNA does not fit well into protection factor–melting point correlations for modified oligonucleotides (7). The order of duplex stability seen in the protection factors of the modified oligonucleotides from the assay was confirmed in UV melting experiments, in which 15h gave higher melting points than 15a, which in turn melted at a higher temperature than 15g (Fig. 6). The melting point for the uracil-bearing 15h with 5′-GTCAACCAA-3′ was 4°C higher than that of unmodified control strand 16. For the self-complementary sequence u-T*GCGCAA (17d) the melting point increase was +2.3°C over the melting point of DNA (21)₂. Compared to this octamer duplex, the stabilizing effect disappeared at high salt, as expected for an uncharged backbone that fails to benefit from reduced charge repulsion (22). Compared to compound 17b, bearing an acetyl group instead of the cap, the melting point increase for 17d was +7.5°C at low salt and +4.3°C at 1 M NaCl. When a hydrogen was substituted for the methyl group of thymine t1 in the structure shown in Figure 2b to generate a uracil moiety and then force field minimized with B-form constraints for the canonical terminal base pair, the structure shown in Figure 2c was obtained. In this structure both H5 and H6 of the uracil moiety pack tightly against G3.

Figure 5.

Selection from chemset 5 via phosphodiesterase I. Kinetics of disappearance of the full-length oligonucleotides from solution, as determined by quantitative MALDI TOF mass spectrometry. Data points are averages ± SD of the relative signal intensities (analyte/internal standard) from four spectra per data point, normalized to the signal ratio in the control spectra of the undigested library.

Figure 6.

Overlay of UV melting curves of the duplexes of 15a, 15g, 15h and 16 with target strand 18, as monitored at 260 nm (11 µM strand concentration, 10 mM phosphate buffer, pH 7).

Next, a series of melting curve experiments was performed in which the 3′-terminal residue of the target strand for nonamer 16 was varied, so that either the Watson–Crick match A or the mismatched nucleobases C, G or T were presented to T1. For control compound 16 the mismatched targets gave duplexes with melting points 2.6–4.0°C lower than that of the fully complementary duplex (Table 3). This level of mismatch discrimination at the terminus is similar to that observed for the octamer TGGTTGAC studied earlier in these laboratories (8). Compounds 15a, 15g and 15h did not show a preference for an adenine residue at the terminal position of the target nor did they show a clear selectivity for any other of the nucleobases. This suggested that their 5′-terminal residues did not form base pairs with the 3′-terminal residue of the opposite strand.

Table 3. Melting temperatures of oligonucleotides with target sequences containing different 3′-terminal residues^a.

Oligonucleotide	Target strand (5′-GTCAACCAX-3′)				Target strand 5′-GTCAACCA-3′ (22)
	X = A (18, match)	X = C (19a)	X = G (19b)	X = T (19c)
5′-TTGGTTGAC-3′ (16)	22.9 ± 0.3b	18.9 ± 0.2	19.6 ± 0.3	20.3 ± 0.3	19.7 ± 0.5
5′-t-T*GGTTGAC-3′ (15a)	21.9 ± 0.3	22.3 ± 0.3	21.4 ± 0.3	22.2 ± 0.3
5′-t*-T*GGTTGAC-3′ (15g)	18.6 ± 0.2	17.7 ± 0.3	19.1 ± 0.3	19.0 ± 0.1
5′-u-T*GGTTGAC-3′ (15h)	26.9 ± 1.1b	27.6 ± 1.1	26.6 ± 1.1	27.3 ± 1.4	27.3 ± 0.5

^aMelting temperatures are the avarage ± SD of four individual curves obtained at 11 µM strand concentration and 10 mM phosphate buffer, pH 7.

^bAt 1 M NaCl concentration, the melting points are 44.0 ± 1.1°C for 16:18 and 43.3 ± 0.4°C for 15h:18.

Finally, two sets of nuclease experiments were performed. The first of these probed to what extent the 3′-terminal residue of the target strand of 15h was exposed to attack by snake venom phosphodiesterase. If the uracil residue did indeed fail to engage in Watson–Crick base pairing, the 3′-terminal residue of the target strand (A9) should be a dangling residue and thus more readily attacked by the single strand-specific nuclease [the vulnerability of a single dangling residue at the 3′-terminus to attack by snake venom phosphodiesterase was also observed earlier, e.g. compound 15 in chemset 3 (7)]. Two control experiments were performed as well: one with the fully base paired TTGGTTGAC:GTCAACCAA (16:18) and one with the octamer/nonamer duplex TGGTTGAC:GTCAACCAA (20:18), in which a dangling residue exists at the 3′-terminus of the nonamer. Unlike the selection experiments, these nuclease assays were conducted in three separate solutions, each with just one duplex. The nuclease resistance of the nonamer in the duplex with 15h was between that of the two controls (Fig. 7). This result is not trivial to interpret because of the 5′-modification introduced, but it demonstrates an enhanced vulnerability of A9 over the base-paired control.

Figure 7.

Probing for a 3′-dangling residue. Kinetics of removal of the 3′-terminal adenosine residue A9 from target strand 18 by phosphodiesterase I in the presence of oligonucleotide 15h, 16 or 20. Residue T1 of nonamer 16 can form a base pair with A9, whereas octamer 20 lacks a residue to do so, leaving a dangling residue A9 in 18. Data points are averages ± SD of the relative signal intensities (analyte/internal standard) from four spectra per data point acquired with the internal standard to allow for quantitation and are normalized to the signal ratio in the control spectrum of the undigested duplexes. Unlike in the selection experiments, the nuclease reactions of each duplex were run in individual solutions and the graph is an overlay of the individual kinetics obtained at the same enzyme concentration. Note that in the duplex with 15h, residue A9 is more rapidly degraded than in the duplex with 16, but more slowly than in the duplex with 20.

Lastly, the resistance of oligonucleotides bearing the acyl-linked nucleobases to nuclease degradation from the 5′-terminus was tested. This test was performed to evaluate their usefulness for applications where nuclease resistance is an advantage. For this, a mixture of 15a, 15g, 15h and unmodified control nonamer 16 was treated with bovine spleen phosphodiesterase and their degradation monitored by MALDI TOF mass spectrometry (Fig. S17, Supplementary Material). Whereas the unmodified 16 was readily hydrolyzed by the nuclease, all three modified oligonucleotides largely withstood nuclease attack at their 5′-terminus.

DISCUSSION

These results are interesting in several respects. First, it was unexpected that among the nucleobase-containing 5′-appendages tested the smallest would provide the greatest duplex stabilizing effect. Uracil acetic acid with its residue weight of 153 g/mol is, in fact, smaller than a single tryptophan or tyrosine residue and approximately half the residue weight of a thymidine monophosphate. Still, it constitutes a clear minimum in structure space, as introducing a methylene unit in the 5 position of its nucleobase does lead to a substantial decrease in its duplex-stabilizing effect. Even 5′-appendages such as those of 7j or 15f, with their much larger intrinsic binding energy (23), or 7i, with its ability to form ion pairs, did not provide as much of a duplex stabilizing effect as the uracil acetic acid residue in 15h. This was not because the former could not engage their nucleobase in base pairing, as they do have a backbone structure that makes binding conformations accessible. Possibly, the amino acid-containing appendages failed to bind with high affinity due to too high an entropic cost for freezing out rotations in their backbone. They probably behave like short peptides, which are known to adopt ‘random coil’ conformations in dilute aqueous solutions. This, in turn, highlights the preorganizing effect of the ribose–phosphate backbone in natural DNA, where the ribose rings probably have a proline-like rigidifying effect [such a proline-like effect was also found in molecular modeling studies on cyclic peptide–DNA hybrids (24)] and the like charges of the phosphodiesters counteract backfolding of the oligonucleotide chains.

Secondly, the structural picture that emerged for the acetyl-linked nucleobases from the available data is, in itself, surprising. Several lines of evidence indicate that the thymine and uracil moieties in 15a, 15h and 17a bind in the major groove of the core duplex to which they are appended. One strong indication is that despite their duplex stabilizing effect, they do not show any mismatch discrimination for the residue at the 3′-terminus of the target strand (Table 3). Further, the 3′-terminal residue of the target strand is more nuclease-accessible in the complex with the uracil acetic acid-bearing strand than in the control DNA duplex, in which it is engaged in Watson–Crick base pairing. Additionally, there is at least one unambiguous NOESY cross-peak between a major groove proton of the core duplex and H5 of t1 in (t-T*GCGCAA)₂. The modeled structure obtained with the constraints based on this and other cross-peaks, though only a very early attempt at the full 3-dimensional structure, does explain why the uracil-containing duplex is more stable than its thymine-containing counterpart. The packing of the C5/C6 side of the uracil against the major groove edge of the penultimate base pair is not very unusual, if one considers that this buries the more hydrophobic edge of the pyrimidine ring and presents the more polar Watson–Crick edge to the water. Burying the most hydrophobic surfaces is, of course, a major driving force in the folding of proteins in aqueous solution. The increased stability of the phenylalanine-containing duplexes (7d) compared to those containing other amino acid residues may be explained based, at least in part, on the same argument, though structural details are currently unknown. Phenylalanine is one of the most hydrophobic amino acid residues (+2.8 on the hydropathy scale of Kyte and Doolitle) (25). The hydrophobic effect has also been cited as the major stabilizing force in the high melting duplexes of alanyl-PNA, in which phenylalanine residues face cytosine residues (26).

Thirdly, with a melting point up to 4°C higher than for an additional T:A base pair, the extent to which the small uracil acetic acid residue stabilizes the duplex is surprising. Its contact surface with the target strand is small, indicating that its stabilizing effect may result both from direct interactions with this target strand and from a rigidifying effect that its packing against the T2 of its own strand may have on the duplex. In this, packing of the pyrimidine ring against the methyl group of T2 may be a particularly favorable interaction (27). If uracil does pack against this T2 residue, which is part of the exposed, terminal base pair, it may reduce ‘fraying’ at this terminus, thus indirectly stabilizing the duplex. The thermodynamic parameters of duplex dissociation (Table S1, Supplementary Material), obtained via analysis of melting curves with Meltwin (13), indicate that the greater stability of 15h:18 is due to a decreased entropic penalty for duplex formation. Apparently, the 3′-dangling residue of the target strand does not provide a substantial additional duplex stabilizing effect, since deleting it (duplex 15h:22, Table 3) does not change the melting point to a measurable extent.

Ironically, modification of the acetic acid-linked thymine residue to a uracil-containing one was conceived when viewing a Watson–Crick paired structure (Fig. 2a) and not a structure in which the uracil was located in what is now believed to be a more accurate model of the structure (see Fig. 2c). This suggests that, at least in these laboratories, rational design of oligonucleotide analogs is still in its infancy and that the synthesis of combinatorial libraries followed by selection via nuclease survival can successfully complement it. Nuclease selection, with its low demand on compound quantities and purities, did provide the correct ranking of compounds, as seen for all cases in which both protection factors and melting point data were obtained. Still, with compounds that differ only moderately in their target affinity differences in protection factors are small. As expected, the correlation is best when the melting curve data at the salt concentration closest to that of the nuclease experiments (160 mM) are considered.

Finally, the 5′-appendages identified as the most duplex stabilizing seem to have some properties that could make them useful for practical applications. Since they are either commercially available or trivial to synthesize from inexpensive starting materials, they complement those molecular caps that are strongly discriminating between nucleobases at the 3′-terminus of the target strand (8). All three directly linked nucleobases proved to convey to the 5′-terminus the nuclease resistance sought. Thus, they may help to make oligonucleotides more biostable, particularly those immobilized at their 3′-terminus, such as the probe strands in some ‘DNA chips’. The uracil-containing appendage may also be useful for increasing the affinity of oligonucleotide probes for target strands at low salt. Finally, the structural information presented here may help to design improved PNA–DNA hybrids (28–34), since the known linkers in PNA–DNA chimeras destabilize duplexes with DNA target strands, particularly so if a 5′-amino-5′-deoxythymidine residue is linked to an unmodified PNA unit (22).

SUPPLEMENTARY MATERIAL

MALDI spectra of modified oligonucleotides, NMR spectra of 8, 9, 10 and 17a, plots showing the nuclease resistance of 15a, 15g and 15h, a correlation between protection factors and melting points for chemset 5 and a table with thermodynamic parameters for duplex dissociation are available at NAR Online