Scalable convergent synthesis of therapeutic oligonucleotides

. The recent explosive growth in the number of oligonucleotide clinical development programs and drug approvals underscores their ability to treat diseases via mRNA regulation. Currently, solid-supported synthesis is limited to ≤ 5 kg (~ 1 mole) batch sizes and the feasibility of liquid-phase syntheses to supply materials of sufficient purity and amount for clinical trials has not been proven. Herein we describe the first convergent synthesis of a full-length 18-mer mixed backbone (PO/PS) 2’ -MOE gapmer oligonucleotide by the phosphoramidite approach suitable for use in clinical trials. Techniques described to control impurities during its synthesis can be implemented in common active pharmaceutical ingredient (API) manufacturing facilities and should enable a >10-fold increase in production batch scaling.


2
Messenger RNA (mRNA) has increasingly become a therapeutic target. 1, 2 Therapeutic oligonucleotides regulate the composition and quantity of mRNAs through splicing regulation, RNase H-mediated degradation, or RNA interference mechanisms [3][4][5] with downstream effects on protein expression. Eleven chemically synthesized oligonucleotide drugs have been approved to treat rare diseases, 2, 6 which include antisense oligonucleotides (ASOs) and small interfering RNAs (siRNAs) made via solid-phase oligonucleotide synthesis (SPOS). In current clinical research, new oligonucleotides are showing great promise to treat diseases afflicting millions of people, including cardiovascular disease, and yearly demand for each drug could reach over a metric ton. 2 The current SPOS process at ~ 5 kg scale 7 would require ~200 batches to deliver 1000 kg. While it is conceivable that economies of scale may be realized in massive, specialized facilities, this approach has significant engineering and cost challenges 8 as well as a negative environmental impact. 9 An alternative approach is the development of an economical liquid -phase oligonucleotide synthesis (LPOS) that can be carried out in existing synthetic API manufacturing plants. The most advanced efforts reported in the literature use linear synthesis strategy similar to SPOS (Fig. 1A) by building the oligonucleotide chain on soluble polymeric anchors 10 that enable the simplification of intermediate isolations by using precipitation and extraction extractions. Separation of the intermediates via membrane filtration based on an homostar anchor has also been reported (Fig.   1B). 11 To date, feasibility of LPOS of a typical therapeutic oligonucleotide has not been 3 demonstrated beyond a few grams at a time, although linear synthesis of morpholino oligonucleotides has been reported at a 10-kg scale. 10  4 Only a few literature examples describe convergent oligonucleotide syntheses. Am ong them are the synthesis of a 6-mer assembled from phosphoramidites, 12 a 37-mer from phosphotriesters, 13 a 21-mer employing H-phosphonates, 14 a 20-mer by a template-guided enzymatic assembly ( Figure   1C), 15 and double-stranded siRNAs by enzymatic synthesis. 16 Unlike-phosphonates and phosphotriesters, many phosphoramidites suitable for the synthesis of therapeutic oligonucleotides are commercially available. To date, enzymatic reactions have only worked efficiently for the synthesis of natural phosphodiester (PO) bonds and broadening the technology to phosphorothioate (PS) linkages that are dominant in ASO and siRNA therapeutics 17 has had limited success. Furthermore, product analyses of the alternative approaches to SPOS reported rely heavily on liquid chromatography with UV detection (LC-UV). 18,19 This technique is insufficient to resolve and quantify all of the product related impurities. 20 Therefore, it is difficult to know whether they will provide oligonucleotides with purities high enough to support clinical trials. 21,22 Quantitation of certain types of product related impurities can currently only be achieved by LC-UV coupled to mass spectrometry (LC-UV-MS). 23 Herein, we report the chemical learnings and characterization details from the development of a scalable convergent LPOS (Fig. 1D) of an 18-mer oligonucleotide ON-A by the phosphoramidite approach that provides a product with purity comparable to that made from the current state-of-the-art SPOS. The sequence of ON-A is (MOE) nucleotides, and black = 2'-deoxy ribose nucleotides; o indicates a phosphate (PO) linkage and the remaining are phosphorothioate (PS) linkages; Me indicates 5-methyl in C and U.

Results and discussion
5 Synthetic Challenges and Strategic Approaches. A single strand of an ASO or siRNA typically contains 18 to 25 nucleotides and requires up to 100 separate chemical reactions to synthesize, as installation of each nucleotide proceeds via a four-reaction cycle. 7,23,24 To scale up the synthesis of therapeutic oligonucleotides, impurities must be controlled to very low levels in each isolated intermediate to avoid accumulation of impurities that cannot be separated from the full-length product (FLP) during purification. The majority of existing LPOS strategies require extensive purification of intermediates by precipitation, 25 extractions, 26 or chromatography. 27 Time consuming isolations with accompanying product losses and large solvent requirements can be reduced by telescoping the coupling, sulfurization, and detritylation reactions in each cycle. 28,29 The full-length ON-A may be assembled convergently (Fig. 1D) from 2 tetramer (II and III) and 2 pentamer (I and IV) fragments. This approach can be expected to provide significant operational, economic, and quality advantages over the linear syntheses including: 1. shorter production cycle time with parallel processing of fragments 2. reduced product isolation losses 3. reduction in impurities-as nucleotides are exposed to fewer reaction cycles and are less likely to degrade and or participate in undesired side reactions 4. more effective impurity purging from smaller intermediate fragments

limited impact of single batch failure with independent fragment synthesis
For the full-length ON-A synthesis, tetramers and pentamers of deoxy and MOE fragments I-IV ( Fig. 2A) were made separately. The free 3'-OH in fragments I-III was converted into the corresponding phosphoramidites prior to coupling with the 5'-OH of the fragment IV, 10-mer 6 (VII), or 14-mer (VIII). Synthesis of the Solubility Determining Group (SDG)-containing fragment IV can be accomplished by growing the full fragment from SDG-nucleotide V or by coupling of V with fragment VIa. SDG (MW~ 1000 Da) 19 is a protecting group that imparts the overall solubility of each fragment.
Attempts to control the impurity burden across the entire process focused on 1) using extremely dry reaction conditions to avoid deamination impurity formation during detritylations; 2) implementing use of a novel thiol DMTr cation scavenger to achieve > 99.95% detritylation with simultaneous suppression of retritylation during work-up; 3) achieving >99.95% coupling reaction completion with minimal amidite excess by rigorous drying of reaction solutions; 4) complete removal of the difficult to remove byproduct amidite derivatives, by exploitation of the relative solubility differences between impurities and intermediates or fragments; 5) separation of the fragments I-III from VII-IX during full-length assembly by use of the SDG installed in fragment IV; 6) use of a 3'-OH protecting group that is stable during synthesis and removed quantitatively following fragment assembly. 7

Figure 2. Fragments for convergent synthesis of ON-A and synthesis of fragment I and IV
8 Synthesis of I-IV. Syntheses of I-III and VI started with 3'-OH protection of the first nucleoside with TBDPS-Cl followed by detritylation to provide the 3'-OTBDPS nucleosides (for example, see compound I-2 in the synthesis of I (Fig. 2B)). A DCI mediated coupling of I-2 with deoxy Me C amidite followed by sulfurization with xanthane hydride (XH) and detritylation with TFA in one pot provided dimer I-3. Fragment I was synthesized by repeating this procedure with required amidites, except that in the last cycle 3'-OTBDPS deprotection, rather than the 5'-ODMTr detritylation, was performed. Synthesis of II-III and VI followed the same reaction cycle under similar conditions, except for two cycles where oxidation rather than sulfurization was performed to install the PO linkages. The coupling reactions produce chirality at the phosphorus P(III) stereocenter. The stereoselectivity of the coupling reaction is determined mainly by the amidites, their coupling partners and the activator. [30][31][32] DCI was used as an activator to maintain similar stereochemistry in both the LPOS and SPOS processes. Solvents were chosen to en able telescoping with high reaction efficiency and work-up simplicity. Rigorous residual water control (<100 ppm) and reaction monitoring (HPLC) enabled > 99.95% reaction completions using 1.05 (for the first cycle) to 1.4 (for the last cycle) equivalents of the amidite starting materials, which is less half of the amidites when compared to prior literature examples. 25,28 XH was used for sulfurizations for all but the last cycle in I-III and VI syntheses since its byproduct was found to promote detritylation. Consequently, DDTT was used in the last cycle. In the trimer synthesis for III and VI, I 2 /pyridine was used to introduce the PO linkages. Isolations were performed after the oxidation since the large quantity of pyridine used made the one-pot reaction impractical. For PS/PO mixed backbone fragments, a one-pot oxidation reaction using m- 9 Cl-PhCO 2 H 28 resulted in high PO impurities; alternatively, use of t-BuOOH led to difficult to remove coupling products of amidites and t-BuOH.
Dichloroacetic acid (DCA) is commonly used for SPOS detritylations; however, trifluoroacetic acid (TFA) was found to be more effective in LPOS and produced fewer impurities. In addition, use of thiols, novel DMTr cation scavengers, significantly reduced the equivalents of TFA (~7) and reaction times (~30 min at 0 o C). Without added thiols, reactions stalled even with 20 equivalents of TFA. Retritylation observed during work-ups was eliminated when thiol use was increased to 3.0 equivalents. In the last cycle of the synthesis of I-III and VI the detritylation was replaced with a 3'-OTBDPS deprotection reaction, which was accomplished with a novel HFimidazole combination. This was the only deprotection condition that gave clean and facile reactions, while the typical literature conditions afforded complex mixtures at reaction completion.
After the detritylation or desilylation (last cycle) reaction of the one-pot synthesis was complete, a work-up was carried out, which generally included quenching with aqueous NaHCO 3 solution, extraction, washing, and precipitation to remove impurities based on their solubilities. In the dimer syntheses using Me C, the amidite derivatives were removed by filtration through a SiO 2 pad, which retained the amidite derivatives.
The synthesis of IV was accomplished by two routes (Fig. 2C). The coupling of V with tetramer amidite VIa produced significantly lower n-1 impurities than the one-pot linear elongation 29  Impurities in fragments are considered critical if they lead to impurities in FLP that are not effectively removed by purification; all other impurities are non-critical. 33 Fig 3A shows that the critical impurities, including PO, n-1, and deamination, as determined by LC-UV-MS analysis, were controlled to low levels in all fragments. Most non-critical impurities are more polar and elute earlier than the fragments, as shown in the LC-UV chromatogram of fragment III (Fig 3B). 11 Loss of the cyanoethyl (CE) protecting group forms the Des-CE impurities (~4.6%), which are eventually converted into the desired FLP after convergent assembly and global deprotection. The other non-critical impurity observed at a relatively high level (~2%) is the DMTr-MOE Me C nucleoside originating from MOE Me C amidite used in the last cycle of fragment III synthesis.

Control of Impurities.
Minimizing impurity generation is a key focus during development of a good pharmaceutical manufacturing process. It is even more critical in the synthesis of oligonucleotides due to their accumulation through a greater number of reactions. Several classes of impurities are formed during synthesis and each class contains multiple distinct impurities that can occur at various positions in the sequence (Fig. 4A). 21 The families of impurities formed 12

Figure 4. Main impurities and pathways for formation
during synthesis are the result of incomplete reactions leading to truncated n-1 impurities, over reactions generating extended n+1 impurities, and degradation reactions including hydrolytic replacement of the N 4 -NHBz in the nucleobase C (deamination) and the nucleobase A and G from 13 the ribose (abasic), deprotection of amino (NH 2 ) groups contained in A and C that may lead to branched impurities, and sulfur substitution with oxygen (PO) or vice versa (thioate). The following discussion focuses on n-1 and deamination impurities, which are both easy to produce and difficult to control and remove during purification of the DMTr-on ON-A (X, Fig. 5). With an incomplete coupling reaction, the remaining 5'-OH participates in the coupling of the next cycle and gives rise to an n-1 impurity. Capping unreacted 5'-OH is employed in SPOS 23 and some LPOS. 11 Alternatively, in LPOS syntheses coupling reactions are driven to completion by adding 3.0 equivalents of amidites 25,28 to avoid the need for capping. Since amidites are expensive and their byproducts are challenging to remove from the intermediates, minimizing their use has an impact not only on process economics, but also on the process performance. We minimized n-1 impurities by achieving on average ≥ 99.95% coupling completion with 1.05 (cycle one) to 1.3 (cycle four) equivalents of amidites by drying the reaction mixture to < 100 ppm residual water using 3 Å molecular sieves. More amidite was required for later cycles as it proved difficult to reduce equivalents of residual water from the reaction solutions. No significant coupling reactivity difference was noticed in fragments of different lengths. 14 Incomplete sulfurization or oxidation reactions leave the newly attached nucleoside linked through an acid labile P(III) phosphite triester that can be cleaved during detritylation in the following cycle resulting in an n-1 impurity. Complete sulfurization and oxidation reactions are accomplished with XH or DDTT and I 2 as described previously.
Incomplete detritylation or its reverse reaction, retritylation, also leads to n -1 impurities. Primary thiols were found to be excellent DMTr cation scavengers that drive the deprotection to completion and prevent retritylation. In the presence of 3.0 equivalents of C 12 H 25 SH, the detritylation reaction reached >99.95% completion using 5 (cycle one) to 12 equivalents (in later cycles) of TFA with no detectable retritylation observed during the work-up. These conditions contrast with the 25-70 equivalents of Cl 3 CCO 2 H and 15-25 equivalents of Et 3 SiH as a cation scavenger that were used to achieve reaction completion in Walther's procedure. 28 The n-1 impurities can also arise from 3'-H-phosphonates, the hydrolysis products of excess amidites, present in the dimer intermediates. The H-phosphonate group acts as a 3'-OH protecting group and allows the chain to grow from the 5'-OH during fragment synthesis. Its removal from 3'-OH in the TBDPS deprotection step produces an n-1 fragment (Fig. 4B). This n-1 fragment eventually propagates into full-length n-1 impurity after the convergent assembly. The n-1 impurities from this pathway were controlled by limiting amidite excesses, adding aqueous washes, filtering through silica, or product precipitation.

Deamination impurities.
A surprisingly significant side reaction of LPOS that is not usually observed in SPOS is the substitution of the N 4 -NHBz of cytidines by -OH (Fig. 4A)  Repeating the coupling, sulfurization and precipitation, and detritylation (step excluded in the last 16 cycle) and precipitation, VII was converted into VIII, and then to fully protected IX. Phosphorus deprotection followed by global deprotection by heating with NH 4 OH provided crude X.

Figure 5. Convergent assembly of ON-A
Several side reactions were observed in the synthesis of amidites Ia-IIIa and VIa and assembly steps. Use of ≤2 equivalents of the phosphitylation reagent produced dimeric 8 -mer and 10-mer derivatives of I-III and VI. Residual phosphitylation reagent in Ia-IIIa was also found to react with fragment IV, VII, and VIII in the coupling steps (Fig. 5) to generate the corresponding amidites that in turn generate dimeric impurities of IV, VII and VIII. In the coupling reactions > 99.0% conversions of IV, VII and VIII were achieved with 1.2, 1.3, and 1.4 equivalents of Ia-IIIa, respectively. Incomplete coupling, detritylation and sulfurization resulted in n -4 and n-5 impurities. Although these shortmer impurities could not be removed from the desired VII, VIII, and IX by precipitations, they were removed during purification and downstream processing, which used the same process as SPOS including hydrophobic interaction chromatography (HIC), detritylation, and AEX chromatography. purity than SPOS due to the presence of additional shorter and longer retention time impurities ( Fig 6A). Among the critical impurities coeluting with X, PO (~ 4%) was the only one that was higher in the LPOS material (Fig 6B). Since there are 15 total PS linkages in X, the 4.0% PO impurities in the LPOS may consist of 15 isomeric impurities each at ~ 0.26%. PO impurities in LPOS can come from starting material fragments, be formed when linking fragments during convergent assembly and from side reactions during ammonolysis. 39 The impurity profiles of ON-A from LPOS and SPOS after HIC and AEX purification are comparable ( Fig. 6C and 6D), demonstrating that the impurities originating from the convergent synthesis were effectively removed by the downstream purification process. The purity of ON-A from both SPOS (92.4%) and LPOS (91.4%) is high. The PO impurities from the LPOS were reduced from 4.0% to 2.2% by the AEX chromatography. Total n-1 impurities were found to be lower in the LPOS (1.2%) vs. SPOS (2.9%). This can be attributed to impurity control at the fragment stage. The ions labeled as "adduct" in Fig. 6B and 6D are common artifacts of the ion-pair reversed phase LC-MS analysis and do not represent covalent modification of the ON-A. Except for PO, the sum of the levels of the other critical impurities in the four fragments is roughly the same as the total critical impurities observed in crude X.

Conclusion
The first convergent synthesis of a therapeutic 18-mer mixed backbone (PO/PS) 2'-MOE gapmer oligonucleotide drug candidate was completed. Control of the hydrolytic substitutions of N 4 -NHBz of Me C nucleotides and essentially quantitative chain extension reactions gave ind ividual tetramer and pentamer fragments in high yields and purities. The total synthesis, with the longest linear path including 9 isolations and 14 reactions is practical and scalable. Syntheses of II and VI were completed at ~3 kg scale and details of the optimization and scale-up will be reported in the future.
Building upon this strategy the late stages of completing the convergent synthesis of a second 18mer oligonucleotide is in progress and future efforts are focused on exploiting economies of scale as well as decreasing the environmental impact of oligonucleotide manufacturing by streamlining reaction conditions, reducing reagent use and simplifying downstream purification.

Methods
Detailed experimental and analytical procedures and data are available in the supplementary materials.

Data and materials availability
Detailed experimental and analytical procedures and data are available in the supplementary materials.