Liquid-Phase Synthesis of 2′-Methyl-RNA on a Homostar Support through Organic-Solvent Nanofiltration

Due to the discovery of RNAi, oligonucleotides (oligos) have re-emerged as a major pharmaceutical target that may soon be required in ton quantities. However, it is questionable whether solid-phase oligo synthesis (SPOS) methods can provide a scalable synthesis. Liquid-phase oligo synthesis (LPOS) is intrinsically scalable and amenable to standard industrial batch synthesis techniques. However, most reported LPOS strategies rely upon at least one precipitation per chain extension cycle to separate the growing oligonucleotide from reaction debris. Precipitation can be difficult to develop and control on an industrial scale and, because many precipitations would be required to prepare a therapeutic oligonucleotide, we contend that this approach is not viable for large-scale industrial preparation. We are developing an LPOS synthetic strategy for 2′-methyl RNA phosphorothioate that is more amenable to standard batch production techniques, using organic solvent nanofiltration (OSN) as the critical scalable separation technology. We report the first LPOS-OSN preparation of a 2′-Me RNA phosphorothioate 9-mer, using commercial phosphoramidite monomers, and monitoring all reactions by HPLC, 31P NMR spectroscopy and MS.


Introduction
Oligonucleotides (oligos)h ave re-emerged as am ajor pharmaceuticalt arget due to the unprecedented opportunity for controlling protein expression mediated by short RNA oligomers (ca. 20 nucleotides long) through RNA interference (RNAi)w ith small interfering RNA (siRNA) or micro-RNA (miRNA), and these have in turn re-invigorated research in the field of anti-sense oligonucleotides (ASO/AS-ON). [1][2][3] Excitement rose with the recent demonstration of safe and effective delivery of oligos in humans. [4] This imperative has underlined the need for scalable methods of RNA synthesis. Today the overwhelming majority of oligos are prepared using solid-phase oligo synthesis (SPOS), but this is very challenging to scale up. [5] We are developing al iquid-phase oligo synthesis (LPOS) synthetic strategy that will be more amenable to standard batch production techniques than SPOS, [6] using organic-solvent nanofiltration (OSN) as the critical scalable technology for separatingt he growingo ligo from all other reagents. [7] We now report the LPOS-OSN preparation of a2 ' -methyl RNA phosphorothioate 9mer,m onitoring all reactions by HPLC, 31 PNMR spectroscopy and MS.
The defining characteristic of SPOS is the ease of separation of the growing oligo from excess reagents:t he solid synthesis support bed/column is simply washed with solvent to remove any molecular speciesn ot covalently attached to it. SPOS has been scaled up to 1-2 kg per batch, [5] and the largestt rial of the new generation of RNAi therapies requiredafew kg of oligo. [5] Thus it is expected that 100s of kilograms of oligo might be required annually to treat rare diseases,a nd possibly tons for major ones. The leading companies in the field have claimed that SPOS can be extended to yet largerscales. [8] However,t he specialized equipment is demanding and expensive to use in an industrial setting, and we believe that SPOS, even with major advances, is incapable of approachingt he 100 kg scale per batch, because of the challenge of completely and reproducibly washing large beds of synthesis support. [9a] Therefore av ery serious gap is expected to open between oligo supply and demand that will restrictt his otherwise promising new mode of therapy.C onsequently,anew method of oligo production is urgently required.
Early on scalability was identified as the Achilles' heel of SPOS, and LPOS has long been proposed to overcome this problem. [10] However,t he critical question that must be addressedi na ny LPOS strategy is how to separatet he growing oligo from excess reagents and byproducts. So far,a mongst the alternative strategies reported for the synthesis of oligos, chromatography has been dismissed as too time-consuming, solventi ntensive, and inefficient. [11] Whilsta pproaches including size-exclusion chromatography [12] and extraction [13] have been proposed to overcome this separation problem, precipitation of polymer-supported oligo has been explored much more widely.I nitiallyD NA oligos supported on poly(styrene) (the same solid-phase support as had recently been used by Merryfield for peptides ynthesis) [14] were assembled by means of Khorana's phosphodiester approach. [10,15] Subsequenta uthors explored poly(vinyl alcohol) (PVA) [16,17] and cellulose [18] as supports, but poly(ethylene glycol) came to dominate this strategyw herein the oligonucleotidyl-PEG was usuallyp recipitated with diethyl ether. [17,[19][20][21] Recently ad iscrete, non-polymeric synthesis support was developed in which four oligo chains were grown simultaneously around ap entaerythritol core, and the productsw ere then precipitated from methanol. [22,23] Whilst approaches to oligo synthesis based upon precipitation or crystallisation are in theory scalable, it is questionable whether this would be truly practical. Process development of industrial precipitation is often time-consuming and labour-intensive, with the conditions being unique to each compound.
[9b] During Bonora's preparation of 10 mg DNA 20mer, [20] 79 separate precipitationsa nd crystallisations were required. Furthermore, it is inevitable that materialw ill alwaysb e lost to incomplete separation:o nt he PEG support it was found that losses, starting around 1% per cycle, became more significant as the increasing solubility of the growing oligo began to overwhelm the polymer-driven phase separation; [20] during the preparation of an RNA 5-mer on the small pentaerythritol support, the coupling cycle yield only averaged 85 %f or a5 4% overall yield, whichw ould be unacceptablef or commercial production. [23] The use of membrane-based technologies for the separation of synthetic biopolymers has been little explored, despite their evident potential to realise scalable liquid-phasea pproaches to valuablet argets. In the first such synthesis, Bayer and Mutter bound peptides to mono-methyl PEG-10 000 (mPEG) that was purified by ultrafiltration. [24] This approachw as repeated for oligos by the same laboratory,u sing the now obsolete phosphodiester coupling strategy and either PVAo rP EG-10 000 polymerics upports. [17] However,t here wasn of urtherd evelopment of this approach for either biopolymer.A lternation between chain extension in organic solvent and diafiltration in water after each chain extension cycle probablym akes this strategyi mpractical.
We postulated that organic-solvent nanofiltration (OSN) could fulfil the critical separation role in an LPOS strategy conducted entirely in organic solution,F igure 1. Furthermore, since organic-solvent stable nanofiltration(as opposed to ultrafiltration)m embranes are now available, we proposed that as maller,d iscrete synthesis support could be used. During OSN, solutes are separated by size exclusion and geometric selection as they pass through am embrane possessing nanometer-scale permeation pathways. Solutest hat cannot pass through the membrane are said to be rejected and remain in the upstream retentate.T his scalablet echnology is fully compatible with ap harmaceutical industry batch reactor.W ef ur-ther postulated that LPOS-OSN would provide an excellent platform for monitoring the ongoing oligo synthesis, by means of sampling using as imple liquid draw-off. Although in principle it is also possible to monitorc hain extension progress with SPOS, we are unaware of any report of such ap rocedure. This is most likely due to the difficulty of engineering repeated access to beds of solid support in large diameter,p ressurized steel columns (columns for preparing 1-2 kg oligo by SPOS have diameters 50-100 cm and pressure ratings of 15-20bar). Ready access to samples, although providing the opportunity to optimize reactions and to rescue failed steps, is of marginal value on the small synthetic scales regularly produced today. However,i nt he future, during the preparationo ft ens of kilograms to tons of oligos, it would be economically unacceptable to risk the complete loss of such large batches of very expensive building blocks withoutcritical quality control.

Selection of LPOS-OSNmaterials
Buildingo no ur earlier experience with membrane-enhanced peptides ynthesis (MEPS)i no rganic solvent(DMF), [25] we initially explored OSN separation of mPEG-5000-supported dinucleotides. [26] However,e ven a,w-bis(dinucleotidyl)-PEG-10 000 had too low ar ejection for practical LPOS-OSN. Therefore we in- www.chemeurj.org stead adopted monodisperse tris(octagol) homostar 1 (a homostar is as tar polymer in which all the arms are identical) as ab ranched LPOS support, see Scheme 1. [27] We hypothesised that this would have three advantages:1 )branching shouldi nhibit threading of the supported oligo into the membrane permeationp athways,a nd therefore increase membrane rejection of the construct; [28] 2) with three oligos growing around one hub, the molecular weight will rise by three nucleotides per cycle, rapidlyi ncreasing the overall size, and hence the rejection of the tris(oligonucleotidyl) support with oligo length;a nd 3) since the tris(oligonucleotidyl) support is ad iscrete species, HPLC and mass spectral (MS) analyses of real-time synthetic quality should be feasible. We next required an OSN membrane compatible with acetonitrile, the solvent in which phosphoramidite couplings are typicallyc onducted. The membrane shoulda lso be compatible with feed mixtures containing typical oligo coupling, oxidation/thioylation, capping and 5'-O-unblocking reagents. Furthermore, the membrane must be able to permeate nucleotide monomer debris after oligo chain extension;t hese species are the largestm olecular weight debris generated during the synthesis cycle.T om eet these challenges we developed an ew class of OSN membrane (PBI-17DBX), prepared from poly(benzimidazole) and cross-linked with para-dibromoxylene. PBI-17DBX is very resistantt oc hemical degradation and gave highly reproducible performance in CH 3 CN, whilst being open enough to allow species of similars ize to nucleotide monomers to permeate. [29] Second-and third-generation therapeutic oligos most commonly contain either 2'-deoxyo r2 ' -modified nucleosides (e.g.; CH 3 O-, CH 3 OCH 2 CH 2 O-, F-), as well as more complex locked/ bridged ribose analogues. [30][31][32] For this reason we elected to focus on nucleic acid analogues for our LPOS-OSN test sequence, instead of native RNA. Adoption of LPOS-OSN by other groups would be encouraged if this technology was compatible with commercial building blocks and common protective group combinations. Therefore we selected readily available2 ' -methoxy nucleosides,a ctivated as their 2-cyanoethyl (Cne) N,N-diisopropylphosphoramidites, carrying 5'-O-(4,4'-dimethoxytriphenylmethyl) (Dmtr) temporary protection and with various amides blocking the exocyclic amino groups of the nucleobases. [5] We also selected the widely used, firstgeneration phosphorothioate modification as at arget for this pilot project because the debris from thioylation reagents is likely to be am ore severe test of LPOS-OSN purification than common oxidants (e.g. iodine-pyridine-water,o rtert-butyl hydroperoxide).
All previous LPOS studies, except for that of Lonnberg, [23] have concerned the synthesis of DNA oligos, which are easier to prepare than RNA. Furthermore, with the exception of Bonora's DNA 20-mer prepared using numerousp recipitations and crystallisations, [19] the largest oligo prepared by LPOS to date using iterative synthesis is aD NA 10-mer,t hrough Hphosphonate coupling; [21] aD NA phosphorothioate 15-mer has also been reported, but this was constructed using dimer buildingb locks. [20] As ac hallenging target for this new LPOS-OSN technology,w es et out to synthesise a2 ' -methyl RNA phosphorothioate 9-mer section of the M23D ASO. [33] To assess the performance of the support and phosphoramidite chemistry in this new environment, we planned to undertake global deprotection at both the 5-mer (four chain extensions) and 9mer (eight chain extensions) stages.

Results and Discussion
Homostar 1 was first condensed with 4.5equivalents 5'-Dmtr-2'-methyl-3'-succinyl uridine (2,D mtr-mU-Suc-OH), see Scheme 1. Classicala ctivation with 8equivalents N,N'-diisopropylcarbodiimide, in addition to catalytic 4-(dimethylamino)pyridine (DMAP,0 .2 equiv) in THF,w as incomplete with excess uridine succinate being consumed as the acyl urea. To maximize the analytical potential of LPOS-OSN it is highly desirable to drive loading of the synthesis support to completion to give ah omogeneous product, as well as to avoid waste of expensive excessn ucleoside on al arge scale. Thus, condensation of 4.5 equivalents uridine succinate (2)w ith homostar 1 was initiated with more reactive 2,6-dichlorobenzoyl chloride (DcbCl) and N-methyl imidazole (NMI), [34] after which no PEG-terminal hydroxylsr emained. The resultant tris-Dmtr-ether (3)w as then detritylated with dichloroacetic acid (DCA), using pyrrole as ac ation scavenger, [35] to provide fully loaded homostar 4 in 80 %y ield over the two steps, ready to commence the chain extension cycle. At this stage as mall amount of Dcb-ester (5) was separated chromatographically from 4;a lthough this contaminantw ould not affect oligo synthesis at all, in this study it was removed to simplify HPLC analysis of chain extension.
The loaded synthesis support 4 (1.24 g) was next chain extended with 5'-Dmtr-2'-methyl N-acetylcytidine (Dmtr-mC) phosphoramidite 6 C (1.5 equiv per OH) to mUmC homostar 7 under typical conditions, see Scheme 1: ethylthiotetrazole (ETT, 3equiv per OH) in CH 3 CN, 35 min, then phenylacetyl disulfide (PADS) in pyridine, 30 min, monitoring by HPLC (see Scheme 1 and Supporting Information). For this pilot study unusually long times were used for both coupling and thioylation so that the reactions could be sampled and monitored in real time before moving on to the next process. For this reason, the widely used ETT (pK a 4.3, 0.25 m in CH 3 CN) was selected as the activator, firstly because it is ac ompromise that provides higher activity than classical tetrazole (pK a 4.8), but less than 4nitrophenyl tetrazole (NPT,p K a 3.7). [36] Secondly,i nl arger scale couplings with 2'-methyl phosphoramidites,0 .5 m ETT has provede ffective over 5-15 min reaction times, [37] so 0.25 m ETT is commensurate with our longerr eactions. Furthermore, al-thoughN PT has been reported to give very high coupling yields with 2'-methyl phosphoramidites, [38] we were concerned that the greater acidity of this activatorthan ETT would exacerbate contamination from double coupling duringt he long reaction times used here.
During SPOS mass transfer occurs between the bulk solution and the solid support. For fast reactions, such as phosphoramidite coupling, mass transfer is the rate-limiting step. [9a, 24] Thus, if the same chemistry is used in both cases, yields in LPOS are expected to be highert han in SPOS. Consequently,i na nL POS strategyc apping should not be as criticala si nS POS, and this step was omitted simplifying the process during pilot study development. Indeed, it is hoped that assay of the chain extension reactionw ill in future permiti dentification of otherwise economically catastrophicf ailed couplings on very large scales, and provide the opportunity to repeat the reaction. However, if capping were implemented before the assay,i tw ould then be impossible to recover af ailed coupling.
Once chain extension and thioylation were complete, the crude mixture was then diluted with CH 3 CN, and poured directly into the OSN apparatus (see Supporting Information). The rig was pressurized with nitrogen to force solvent ands olutes through the PBI-17DBX membrane, aprocess termed "diafiltration". Ac onstant volumeo fr etentate was maintained throughout diafiltration. Thus, the efficiency of OSN can be related to how many retentate system volumes, or "diavolumes", must be permeated to achieve ag iven degree of purification of the retentate.
After 12 diavolumes, all small molecules had been removed from the retentate. However,a long with the desired Dmtr-dinucleotidyl homostar (7), most of the building block related species, consisting of am ixture of amidates (8)a nd thioate salts (9), were retained (corresponding to N n -OP in Figure 1). Notably,t he proportion of phosphoryl species 8b and 9b compared to thioyl derivatives 8a and 9a (P=Ov s. P=S), determined by 31 PNMR spectroscopy of the mixture (see Figure 2a), increased substantially when the amount of PADS was reduced from 10 to 3equivalents per 5'-OH. Indeed, thioamidate 8a was almost undetectable when the intermediate phosphite was thioylated with 3equivalents PADS, although amidate 8b rose to between 15 and 25 %o ft he 31 PNMR signal integral intensity of product 7.
The crude tris(mUmC-Dmtr) homostar 7 was washed from the OSN rig and re-dissolved in CH 2 Cl 2 .T ot his were added pyrrole then DCA, and after 30 min the detritylation was complete by HPLC. Unlike in SPOS, in which the detritylation equilibrium is driven to completion by flushing the Dmtr + cation away from 5'-OH oligo bound to the solid support, in solution phase as cavenger (here pyrrole [35] )i sn ecessary to ensure total unblocking. Otherwise even small amounts of mono-tritylated homostar would be carried through to the next cycle where (even after capping) subsequentd etritylation would lead to nÀ1s hort-mers. It had been anticipated that at this stage the smaller fragments from the excess buildingb lock (Dmtr-pyrrole, and 5'-OH amidates 11 and thioates 12,n ow corresponding to P and N n -OH in Figure 1) would then permeate, but www.chemeurj.org they did not. Suspecting that ion exchange could occurb etween the protonatableP BI membrane surface and thioate salts 12,1vol %D CA wasa dded to the first five diavolumes. After at otal of 15 diavolumes had permeated the product tris-(mUmC-OH) homostar 10 was then of as imilarp urity to that achieved by flash chromatography.H owever,t he detritylated amidates (11)w ere slower to permeate than the thioate salts 12,a nd thioamidate 11 a exhibited substantially greater rejection than amidate 11 b.T hus by reducing the excesso fP ADS from 10 to 3equivalents, when very little or no thioamidate 11 a formed, the purity of dinucleotidyl homostar 10 was maximised ( Figure 2b). Dmtr-pyrrole was the only contaminant significantly rejected by PBI 17DBX (see HPLC in Supporting Information). Thus, although Dmtr-pyrrole probablyd oes not interfere with subsequent couplings, this was removed by precipitation of tris(mUmC-OH) homostar 10 in diethyl ether so that an accurate mass recovery could be determined; apart from Dmtr-pyrrole no other species could be detected in the supernatant by 1 Ho r 31 PNMR spectroscopy.A fter the first chain extension cycleamoderate 75 %y ield of tris(mUmC-OH) homostar 10 was isolated. The only detectable impurity was al ow level of cytosine N-deacetylation, identified by LC-MS (see Supporting Information) and believed to be the minor peaks in Figure 2c.
Despite the need to remove Dmtr-pyrrole by precipitation, the above cycle was repeated on homostar 10,s ee Scheme 2; from this point on, all chain extension cycles start with 1.2-1.4 gt ris(5'-HO-oligo) homostar.T hus, after chain extension, Dmtr-3-mer homostar 13 wasp artially purified by OSN (12 diavolumes) then detritylated, after which all the nucleotidyl debris was separated by OSN, andp recipitation was again used to remove residual Dmtr-pyrrole. This time during the second diafiltration,t he first fived iavolumes contained only 0.1 %D CA to minimize N-deacetylation. The 85 %y ield of tris-(mUmCmC-OH) homostar 14 was significantly highert han that of dinucleotidyl homostar 10 at the same stage (see Scheme 2, inset graph), indicating that as expected the homostar rejection had risen with oligo length. It should be noted that chain extension cycle yields are calculated assuming 100 %p urity of the product homostar.H owever,a sl ow levelso fs ide-reactions accumulate on the growing oligo, the purity cannot be 100 %, so the molecular weight cannot be precisely defined, and the yields are more correctly referred to as apparent yields.
Both the tritylated (13)a nd detritylated (14)t ris(trinucleotidyl) homostars were less soluble in CH 3 CN than the shorter species 4, 7 and 10-a trend that continued with increasing length. Noting that all the oligonucleotidyl homostars (14-18 and 20-27)w ere highly soluble in DMF, all subsequent phosphoramidite couplings were conducted in CH 3 CN-DMF (ca. 9:1);t his solubility of ab ranched 2'-Me-RNA 24-mer oligonucleotidyl homostar (18)m ay be favorably contrasted with the previously reported poor solubility of 5'-OH DNA 8-mers in CH 3 CN. [39] The solvent was also changed during OSN from neat CH 3 CN, in which tris(tetranucleotidyl) homostar 16 is almost insoluble, to CH 3 OH-CH 3 CN (1:4 or 1:3v /v) in which all the oligonucleotidyl homostars are soluble up to at least tris(9-mer) homostar 27 (0.4 wt % 27 during final diafiltration;t he saturation conc. was not determined). Finally,t he DCA in the second diafiltration was replaced by 1% pyridinium dichloroacetate (Py·DCA) which promoted permeation of thioate salts 12 just as effectively as un-buffered DCA. This protocolwas used on 4mer 16 and for all later chain extension cycles,f ollowing each reactionb yH PLC, and assaying the productsb y 31 PNMR spectroscopya nd MALDIM S, both before and after detritylation (see Supporting Information).
Twof urtherr ounds of chain extension were conducted, with the apparent yield continuing to rise (82 % 16,9 4% 18,s ee inset graph, Scheme 2). Although HPLC usefully exhibited retention times lengtheningi nr elation to the number of 5'-Dmtr ethers per homostar,b oth during chain extension and detritylation, by 5-mers 17 and 18 the peaks were too broad to be of www.chemeurj.org furthera nalytical use (see Supporting Information), presumably due to the exponentially growing number of diastereoisomers at the P-centers of the oligo backbone. However, 31 PNMR spectroscopy continued to demonstrate acceptably low levels of amidate contamination after each cycle.F urthermore, MS confirmed that full-length tris(mUmCmCmAmU-OH) homostar 18 was the principal product. By contrast, MS of an oligo conjugated to ap olydisperse support would be spread over too many polymeric homologues to provides ufficiently intense peaks for analysis.
Pentanucleotidyl homostar 18 was deprotected first with diethylamine, then overnight in aqueous ammonia at 55 8C. The following day,t rituration with CH 3 CN removed protective group debris to give crude 2'-methyl RNA phosphorothioate 5mer 19.H PLC assay of this material (Figure 3a)e xhibited am oderate purity of 74 %, with both short-mer and long-mer contaminants. Although these short-mers could be explained by lack of capping, we believe that they actually derive from chain extension of residual amidate buildingb lock 11 after OSN. Examinationo ft he mass spectra of the 5'-OH tris(oligonucleotidyl) homostars 10, 14, 16 and 18 (from 2-mer to 5mer) exhibit no detectable ions corresponding to incomplete chain extension. Since our synthesis support possesses three arms, if 1% incomplete chain extension had occurred, this would afford approximately 3% homostar having one arm bearing the nÀ1s hort-mer.T hus, assuming that MALDIi onization of full-length oligohomostars and their singlyt runcated oligohomostar contaminants are similar,m ass spectral analysis of homostarss upported oligonucleotides should usefully amplify sequence errors to detectable amounts. The long-mers probablya rise from two sources:1 )relatively long coupling times compared to SPOS (35 min vs. 6-12 min) were used here to allow time for HPLC confirmationo fc omplete coupling. This favors double coupling due to ETT induced detritylation. [40] 2) N-Deacetylation of cytosine residues (as observed at the dimer stage) could provide sites forb ranching, although we suspectt hat the switching from DCA to Py·DCAi nt he second diafiltration of each cycle largely suppressed this.

Conclusion
In this report we have demonstrated for the first time an ew liquid-phase synthesis and separation paradigm for oligonucleotides:l iquid-phase oligonucleotide synthesis/organic-solvent nanofiltration, LPOS-OSN.T his was used to prepare 2'-methyl RNA phosphorothioate9 -mer 28.T his promising technology has yet to equal the speed andp urity of SPOS, requiring aroundt wo days per chain extension cycle with the limited area available from current laboratory-scale flat membrane cells. However,t he fact that we were able to perform eight chain extension and detritylation cycles, with intermediate purifications, all in the liquid phase demonstrates that it has high potential. Compared to competing precipitation strategies, LPOS-OSN is more amenable to industrial exploitation because liquid-phase handling is intrinsically scalable. LPOS-OSN also has the major advantage over SPOS that it is straightforward to sample and monitor every step of the process. Apart from 31 PNMR spectroscopy and HPLC, the choice of am onodisperse support also allowed characterization of the growing oligonucleotidyl homostars by MS.
From the above experience, several modifications can be suggested to improve future protocols:S hortening the coupling time, and analysing only after thioylation, will reduce long-mer formation. [40] Minimising 5'-unblocking time, and therefore acid exposure, will minimise cytosine deacetylation, and again possible long-mer formation. Biasingb uildingb lock debris away from amidates (8/11)i nto thioates (9/12)t hat are more easily removed by diafiltration would suppress short-mer formation.T he overall yield of fully protected 5'-OH tris(oligonucleotidyl) homostar 27 from loaded uridine homostar 4 is only about 39 %, mainly due to poor recovery from the early cycles of OSN;t he first three couplings (4!4-mer 14)g ive ac umulative yield of only 52 %, but the next five couplings (4- Figure 3. HPLC of deprotectedo ligos:a )crude 5-mer 19,7 5% purity; b) crude 9-mer 28,4 9%,containing 8-mer 29,1 8%;c )purified 9-mer 28, 94 %, from LPOS-OSN;d )9-mer 28,9 5%,f rom SPOS. mer 18!9-mer 27)h ave ac ombined yield of about 76 %. This will be much improved using our recently developed 2-stage diafiltration;w ew ould expect early stage recovery to be > 95 %a nd from 5-mer onwards > 99 %. [41] As with SPOS, LPOS-OSN consumes alot of solvent. We have recently demonstrated that an additional stage of diafiltrationw ith al ow-molecular-weight cut-off membrane can be used to recycle the permeate solvent, greatly reducing the potentialc ost on an industrial scale. [42] As the scale of LPOS-OSN increases, an alternative analysist od irectH PLC of the retentate will be required; we believe that rapid ammonia-methylamine (AMA) global deprotection,followed by HPLC of the crude unblocked oligo will provide as uitable method to assay forc omplete chain extension. [43] Finally,i dentifying am embrane that permeates Dmtr derivatives, or using as maller 5'-protecting group, such as the methoxyisopropylidene acetal, [23] would maket he process even more efficient.

Experimental Section
General experimental details 1 Ha nd 13 CNMR spectra were recorded on Brüker AV-400 or Brüker AV-500 spectrometers. Chemical shifts in ppm are referenced with respect to residual solvent signals: d H (CHCl 3 )7 .25 ppm, d H (CHD 2 OD) 3.31 ppm, d H (CD 3 COCHD 2 )2 .05 ppm; d C (CDCl 3 ) 77.50 ppm, d C (CD 3 OD) 49.15 ppm, d C (CD 3 COCHD 2 )2 9.92 ppm. The splitting patterns for 1 HNMR spectra are denoted as follows; s( singlet), d( doublet), t( triplet), q( quartet), quin (quintet), m (multiplet), br (broad) and combinations thereof. Coupling constants (J)a re in Hertz (Hz). 13 CNMR assignments (C, CH, CH 2 and CH 3 )a nd 1 HNMR assignments were established with the aid of DEPT-135, HSQC and COSY experiments. Molecular fragments not abbreviated in main text are denoted as follows:U ,u racil;C ,c ytosine;R i, ribose;S uc, succinate;H ub, C 6 H 3 (CH 2 OR) 3 .C DCl 3 was purchased from VWR, and CD 3 OD and CD 3 COCD 3 from Merck. NMR spectrscopy of small Dmtr derivatives was conducted in the presence of either as mall amount of Et 3 No ro fp yridine. Mass spectra were recorded on Micromass MALDI micro MX, or Micromass LCT Premier (ESI) mass spectrometers. Phosphoramidites and 2'-Omethyl uridine were purchased from Fisher Scientific Ltd.,U Ko r ChemGenes Corp.,U SA. Other reagents were purchased from Sigma-Aldrich Ltd. and used as supplied, except where specified. Reactions were carried out under anhydrous conditions under an itrogen atmosphere. Dichloromethane, acetonitrile, THF and DMF were dried and stored over baked 4 molecular sieves. Triethylamine, diethyl ether,m ethanol, isopropanol and N-methyl imidazole were used as supplied. Flash chromatography was conducted in a9cm diameter,p orosity 3g lass sinter funnel:G eduran (Si 60) from Merck was used for normal phase columns, and Merck silanised silica for reverse phase columns. Thin-layer chromatography was carried out using Merck silica gel 60 F 254 aluminium-backed plates;c ompounds were visualised using UV light or KMnO 4 stain. Solid phase oligonucleotide synthesis (SPOS) was carried out on aG EA KTAO ligopilot 10, using preloaded 2'OMe UP rimer Support 200 and manufacturer's standard protocols on a3 0m mol scale. Cleavage from solid support and deprotection of nucleobases was carried out in 0.88 aqueous ammonia at 55 8Cf or 16 h.
Partially purified tris(DmtrO-6-mer) homostar 20 (1.708 g) was placed in CH 2 Cl 2 (28 mL), to which was added pyrrole (0.48 mL) then DCA (0.28 mL). After 45 min the reaction was complete by TLC and pyridine (0.28 mL) was added. The mixture was diluted with CH 3 CN (100 mL) and the liquid concentrated until all the CH 2 Cl 2 had evaporated. To this solution was then added CH 3 OH (20 mL), the solution was diluted with further CH 3 OH-CH 3 CN (3:17 v/v) containing pyridinium.DCA (0.5 vol %) and 5d iavolumes were permeated. The flux was observed to drop significantly,s ot his was followed by 10 diavolumes CH 3 OH-CH 3 CN (1:4 v/v) when the flux improved. The retentate was evaporated to dryness, and the residual glass was re-dissolved in CH 2 Cl 2 -CH 3 OH (10 mL). The solution was added dropwise to briskly stirred diethyl ether (300 mL), and the precipitate collected to afford tris(HO-6-mer) homostar 21 (1.394 g, 98 %) as ab rown powder. 31