Article Category: Oligo Purification

@article{weldon2022galnac,
  title   = {Purification of a GalNAc-cluster-conjugated oligonucleotide by
             reversed-phase twin-column continuous chromatography},
  author  = {Weldon, Richard and Lill, J{\"o}rg and Olbrich, Martin
             and Schmidt, Pascal and M{\"u}ller-Sp{\"a}th, Thomas},
  journal = {Journal of Chromatography A},
  volume  = {1663},
  pages   = {462734},
  year    = {2022},
  doi     = {10.1016/j.chroma.2021.462734}
}

The problem has a familiar shape. Widen the collection window and yield goes up while purity falls; narrow it and purity rises while yield falls. For an oligonucleotide, impurities differing from the target by a single nucleotide tend to co-elute with it, which makes that trade-off an expensive one. The study reports a clinical-grade purity target of more than 90% as typically costing between 20% and 50% of the product, followed by expensive re-chromatography to claw some of it back.

That loss does not stay in the chromatography suite. Solid-phase synthesis is the major cost driver in oligonucleotide production, so whatever the purification gives up has to be made in the first place.

It is described as the first evaluation of MCSGP for a GalNAc-cluster-conjugated gapmer in reversed-phase mode, and it carries the result through to a scale-up estimate: what that yield is worth in kilograms of synthesis per day.

1. Why This Study Matters: Synthesis Is Where the Cost Is

Oligonucleotide production runs in four major steps — solid-phase synthesis, chromatographic purification, desalting by diafiltration or precipitation, and isolation by freeze drying. The study is direct about where the money sits: oligonucleotide synthesis is currently the major cost driver, and any loss of product in the downstream steps has a major impact on overall manufacturing costs.

It also frames the scale problem. Demand for production capacity is expanding rapidly, but oligonucleotide therapeutics are yet to be manufactured at multi-ton scale and no compound-dedicated production facilities exist — which is why pharmaceutical companies and CMOs are actively evaluating platform technologies that will optimize production and reduce costs.

Why the separation is hard

Synthesis delivers roughly 60% to 80% desired product alongside a mixture of impurities, and they fall into three families:

• Shortmers Sequences missing one or more nucleotides (N-x), left behind by incomplete chemical coupling reactions during elongation.
• Full-length product impurities Molecules of the right length carrying a defect — cyanoethyl modification, phosphate impurity, depurination — produced by modifying side reactions.
• Longmers Sequences with extra material, such as lambdamers and branchmers, arising from impurities present in the starting raw materials.

N-1 and N+1 are singled out as particularly challenging: one nucleotide shorter or longer than the target, chemically almost identical, and so tending to co-elute. Even after considerable optimization, a preparative batch method leaves impurities significantly co-eluting with the product, especially for highly phosphorothiolated sequences. Meeting specification therefore means fractionating carefully and accepting that the side fractions are either discarded or re-chromatographed at cost — lower productivity, more buffer, intermediate storage, extra quality control.

Recycling, but not the usual kind

MCSGP is not the only way to recycle. Steady-state recycling and closed-loop recycling do it too, and the paper draws a specific distinction. In those single-column processes the chromatographic profile passes through a pump every cycle, which is possibly detrimental to product stability and limits the process to isocratic operation. In MCSGP the recycled side fractions are fully adsorbed onto the second column and then subjected to renewed gradient elution, so product and impurities pass through a pump only during the load step. The authors argue this makes MCSGP well suited to complex molecules such as oligonucleotides.

2. The Study at a Glance

The molecule is a DNA-LNA gapmer conjugated to a bislysin GalNAc cluster — the sugar ligand that transports synthetic oligonucleotides into hepatocytes by endocytosis, allowing the development of treatments targeting diseases of the liver. It was assembled by standard phosphoramidite chemistry on solid phase at a 2.0 mmol scale, then conjugated to the GalNAc cluster in solution phase. The crude entered purification at 77.5 area% by RP-HPLC.

What was run, and on what

Two deliberate simplifications distinguish the continuous method from its batch parent. The wash after loading was dropped, because the strong recycling step effectively substitutes for it. And regeneration, which the batch process performs after every elution, was applied only at the end of the MCSGP run — with no evident negative impact on process performance.

3. What the Yield Buys Upstream

At a matched purity of roughly 94 area%, the batch process recovered 52.70% of the product loaded and MCSGP recovered 91.58%. The interesting question is what that difference is worth once the process is scaled, and the study answers it with a simplified estimate built around a single 60 cm batch column producing 1.3 kg of product a day.

The study is even-handed about the first scenario. Two 60 cm columns hold 70 L of resin between them against 42 L for the single batch column, so the extra output is not free. The authors point out that a batch process using the same total resin volume could deliver 2,614 g/day — but would need twice the feed, 6,400 g/day, to do it. The comparison that isolates the yield advantage is the third: same product out, same total resin, 42.5% less crude in.

4. Designing the Continuous Process

MCSGP is designed by dividing a batch gradient chromatogram into four phases that make up one switch. The batch run is not merely a benchmark in this study; it is the design template, which is why the paper notes that the batch process should already reach the target purity with good productivity and acceptable buffer consumption before conversion begins. A batch process with a yield between 20% and 80% is described as a good starting point.

• P1 — columns in parallel Weakly adsorbing impurities elute from column 1 to waste, while column 2 is regenerated ready for loading.
• P2 — columns interconnected The mixture of weakly adsorbing impurities and product elutes from column 1 and is recycled, with in-line dilution, directly onto column 2.
• P3 — columns in parallel Pure product is collected from column 1 while fresh feed is loaded onto column 2. To hold a switch-to-switch steady state, the quantity of new feed applied is in equilibrium with the product removed — a load the software calculates from the batch chromatogram and fraction analysis.
• P4 — columns interconnected The mixture of product and strongly adsorbing impurities is recycled onto column 2 in the same way as P2. The columns then swap positions and the sequence repeats.

Setting the boundaries is where the judgment sits. P3 was positioned to achieve the target purity of more than 94 area%, calculated by the wizard, and P2 and P4 were positioned to fully recycle the lower-purity side fractions. Because N-1 and N+1 are structurally so close to the target and correspondingly hard to remove, their abundance across the peak was checked as well, so that the product collection window could be placed to minimize them.

The wizard then calculated the in-line dilution flow rates needed to dilute the modifier concentration of the eluate back to initial loading conditions — essential if product recycled from one column is to fully re-adsorb on the other — along with the loading parameters required to establish a cyclic steady state. It produced three methods: a startup that preloads the upstream column as a batch process would, a main method of six MCSGP cycles, and a shutdown consisting of one switch with a final product elution but no feeding.

5. What Six Cycles Showed

Six cycles were run, each comprising two switches, for twelve product elutions in total. The question a continuous process has to answer is whether it settles, and whether anything accumulates while it does.

The height of the UV profile inside the P3 product collection window stayed consistent from cycle 1 to cycle 6, which indicates that material entering the system was in equilibrium with material leaving it. The visible changes across early cycles were in the P2 and P4 recycling windows, where impurities accumulate and take several cycles to reach steady state.

All six cycles were therefore analyzed, by HPLC for purity and Nanodrop for concentration. Concentration held at around 2.7 g/L and total purity stayed consistently above 94 area% across the run, and neither N-1 nor N+1 showed accumulation. On that basis the authors conclude that the process parameters are well designed with respect to product quality. Comparing the final pooled product from all six cycles against the input material shows enrichment from 77.5 area% to 94.2 area%.

Why the reported figures are steady-state values

One thing to read carefully. Startup and shutdown both cost performance, because the load applied during startup contains more product than is recovered in the final shutdown, and the shutdown does not recycle its side fractions — making it analogous to a batch run in terms of product loss. With only six cycles run, the authors restricted the performance comparison to steady state, meaning process performance if the number of cycles were infinite. They describe this as the best-case scenario, and state that in practice a run of more than 20 cycles would be very similar.

6. Full Results

Swipe the table sideways to see the feed, batch and MCSGP columns.

Values from Table 3 of the study, rearranged. Weldon, Lill, Olbrich, Schmidt and Müller-Späth, J. Chromatogr. A 2022, 1663, 462734, published under CC BY-NC-ND 4.0.

Where the two processes differ, and where they do not

Yield is the change. Almost everything else is close to unchanged: purity is matched by design at roughly 94 area%, productivity is 1.33 against 1.30 g/L_resin/h, and buffer consumption is 9.22 against 9.59 L per gram of product. N-1 came out marginally lower in MCSGP and N+1 marginally higher.

Two figures moved the wrong way, and both are acknowledged. Product concentration was 17% lower in MCSGP, which the authors flag as leaving scope to reduce eluate volumes. And the MCSGP load was lower than the batch load, 1.83 against 2.52 g per liter of resin — not because that was optimal but because load optimization was not part of the study. Batch experiments qualitatively indicated that small increases in load led to further yield loss, which suggests the binding capacity of the resin was already being used to a great extent.

Scale-up scenarios

Swipe the table sideways to see all three scenarios.

Values from Table 4 of the study, rearranged. These are simplified scale-up estimates, not demonstrated production runs.

7. What It Means for Manufacturing

The scale-up estimate is where the yield figure becomes an operational argument. Producing the same 1.3 kg of product a day from 1,841 g of crude instead of 3,200 g means the preceding steps, including oligonucleotide synthesis, can be downscaled by 42.5% — or, read the other way, the same amount of product can be made 42.5% faster.

The equipment picture is mixed. In the same-output scenario the two MCSGP columns hold the same total resin as the single batch column, at a smaller inner diameter, and the smaller diameter allows smaller pumps on the skid to reach the same linear flow rate. Because MCSGP uses similar equipment with a comparable footprint to batch chromatography, the authors consider it feasible to equip existing facilities.

Beyond yield, the paper lists three further advantages. Re-chromatography of side fractions is no longer required, which removes a step that improves batch yields but decreases overall productivity, increases buffer consumption and demands manual handling and storage space. The analytical burden can fall: a stable batch process requires a minimum of three pools to be collected and analyzed, more when product collection is less robust and more again after re-chromatography, whereas MCSGP typically generates a single product pool per cycle and it may be sufficient to pool and analyze every few cycles. And less human intervention is needed during production, so fewer personnel are required.

Against that, the paper names its own trade-off plainly: elevated equipment and validation complexity, since an MCSGP setup uses two columns and approximately doubles the amount of hardware.

An equivalent conversion in anion exchange mode, on a conjugated siRNA sense strand, is reported in a companion study in this series. For the platform view of where MCSGP fits in an oligonucleotide workflow, see oligonucleotide purification with MCSGP.

8. Key Terms in This Study (Glossary)

9. Frequently Asked Questions

@article{weldon2025continuous,
  title   = {Continuous Purification of a Conjugated Short Interfering RNA
             Therapeutic Using Anion Exchange Twin-Column Chromatography (MCSGP)},
  author  = {Weldon, Richard and Vandermeersch, Tobias
             and M{\"u}ller-Sp{\"a}th, Thomas},
  journal = {Organic Process Research \& Development},
  volume  = {29},
  number  = {6},
  pages   = {1400--1410},
  year    = {2025},
  doi     = {10.1021/acs.oprd.4c00513}
}

In oligonucleotide manufacturing, the expensive step comes before the chromatography. Solid-phase synthesis assembles the sequence one coupling at a time, and every incomplete coupling leaves a chemically similar impurity behind — shortmers and longmers that the purification then has to separate from a target they closely resemble. Product lost at the polishing step has to be replaced by synthesizing more, which is where the cost sits.

That makes purification yield unusually valuable here. A polishing step that recovers more of what was loaded lets the synthesis batch shrink proportionally, taking reagents, solvents, waste and column volume down with it.

It also sets out how the conversion was done — a software-aided procedure for taking a batch chromatogram and turning it into a running continuous method — and projects what the same process would deliver on production-scale equipment.

1. Why This Study Matters: Yield Is the Expensive Variable

Oligonucleotide drugs have moved from promise to product, with 15 approvals to date and hundreds of candidates in clinical trials. Making them at commercial scale remains challenging and expensive. At the time of writing, the maximum commercial GMP scale is limited to a 2,000 mmol synthesis scale using solid-phase oligonucleotide synthesis, and losses during downstream purification limit the maximum output per batch further still.

What batch polishing costs

The standard polishing step is single-column ion exchange or reversed-phase chromatography, with the pure center cut of the chromatogram collected and the rest set aside. Because the target does not fully separate from chemically similar impurities, the lower-purity side cuts have to be reprocessed or discarded.

• Fractionation and mock pooling Yields in batch mode are maximized by fractionating and analyzing the elution peak to find the optimal center-cut boundaries, then generating multiple mock product pools for analytical testing and QC. The paper describes these tasks as essential for batch release and a significant proportion of the overall manufacturing burden in a GMP environment.
• Rechromatography of side cuts The common alternative is to pool out-of-specification side fractions from several batches and re-run them. Side cuts are transferred into custom storage bags, held in a dedicated walk-in freezer, pooled and diafiltered to adjust buffer composition before reloading. It improves overall yield, but reduces the productivity of the process.
• Synthesis has to make up the difference Whatever the polishing step loses must be replaced upstream. Oligonucleotide synthesis is scaled up to compensate for product lost during purification, which the paper notes is very costly.

MCSGP addresses this by recycling the low-purity side cuts internally to a second column, with in-line dilution to aid readsorption, so that rechromatography becomes unnecessary. The technique has been demonstrated across a range of biomolecules; what this study adds is a case in anion exchange mode, on a conjugated siRNA, taken from an existing batch process.

2. The Study at a Glance

The molecule is the sense strand of a therapeutic siRNA, synthesized by standard phosphoramidite chemistry on the solid phase with the sugar conjugation performed during synthesis. The crude entered purification at 85 area% product purity, with the critical N-1 shortmer impurity at 0.93 area%, against a target purity specification of more than 95%.

What was run, and on what

Both eluents were the same in kind: 20 mM sodium phosphate with 10 vol% acetonitrile at pH 8.0, with sodium chloride providing the gradient. The pilot method used 0.9 M NaCl in mobile phase B and the downscale runs 1.1 M, which produced a narrower peak in the downscaled runs — a difference the authors expected the internal recycling to compensate for.

3. Three Ways to Run the Same Separation

The study reports three processes on the same chemistry: the pilot batch method, the same method downscaled, and MCSGP built from that downscaled method. Comparing all three is what separates the effect of the operating mode from the effect of column geometry.

4. From Batch Method to Continuous Process

MCSGP is more complex to set up than single-column chromatography, and the paper notes that it requires specialized process development tools to optimize and implement. Its practical contribution is a procedure that starts from something a process chemist already has — a fractionated batch chromatogram with purity data — and turns it into a running twin-column method.

• Run a batch design method The downscaled batch run was repeated at an elevated elution flow rate of 200 cm/h to maximize MCSGP performance, the eluate fractionated, and each fraction analyzed by HPLC. The chromatogram was then overlaid with the measured fraction purities.
• Set the phase boundaries on screen That chromatogram and its purity data were uploaded to MCSGP Wizard v8.1. The boundaries between the recycling and collection phases were set by drag-and-drop against the purity profile, to optimize the purity of the product collection window.
• Let the software size the rest Gradient parameters, the load per switch and the in-line dilution factors were calculated automatically from the imported chromatogram and the relative position of the phase boundaries. Washing, cleaning and regeneration steps were entered into a template table.
• Run the generated methods The wizard produced a set of startup, main and shutdown methods ready to run on the Contichrom CUBE hardware.

What the four phases do

Each MCSGP switch has four phases, and the two that define the technique are the recycling steps. In phase 1 a gradient starts on the first column while the second is cleaned and regenerated in parallel. In phase 2 the columns are interconnected and the early-eluting, low-purity side cut is transferred from the upstream column to the downstream one, with in-line dilution lowering the modifier content so the product readsorbs. In phase 3 the columns run in parallel again: pure product is collected from the first while the second receives fresh crude feed. In phase 4 the late-eluting, low-purity side cut is recycled the same way as in phase 2. The columns then swap positions and the sequence repeats.

5. What the 13-Cycle Run Showed

A 13-cycle MCSGP run was carried out, giving 27 product elutions including startup and shutdown, over a total run time of 17 hours. The cycle number was limited only by how much crude was available for the study.

Convergence

A continuous process has to settle before it can be judged, and the study tracks that settling three ways. The area under the UV curve for the product collection phase reached a constant value by cycle 5, indicating that the mass of product entering and leaving the system was in steady state. Product purity followed the same pattern: it starts at 95.6 area%, declines and stabilizes at around 95.1 area%. The critical N-1 impurity starts at 0.5 area%, rises and stabilizes at around 0.7 area% — inside the specification of less than 1.0%.

Per-cycle yield was steady at roughly 94–95%, with product recovery holding between 127 and 130 mg per cycle. The shutdown step returned a much lower 62%, which is a normal feature of MCSGP: it is calculated against the larger load applied during startup, and its effect on the overall figure shrinks as the number of cycles rises. Including startup and shutdown, the overall yield was 93%.

Impurity accumulation, watched rather than assumed

Some accumulation was detectable in the region undergoing recycling of strongly adsorbing impurities. The paper treats this as expected rather than alarming: impurity accumulation in the recycling regions is typical in MCSGP, and only becomes a problem if it extends into the product pool and causes the specification to be missed. It did not here, and the stated remedy if it had would be to narrow the recycling windows during process design and send more impurities to waste.

The conductivity traces overlaid reproducibly across the whole run, indicating stable system performance, and the authors note that monitoring area under the curve cycle to cycle gives an operator a live signal: deviation from predefined limits can trigger a run to be adjusted, paused or aborted before time and material are wasted.

6. Full Results and Scale-Up

Swipe the table sideways to see all three processes.

Source: Table 4, Weldon, Vandermeersch and Müller-Späth, Org. Process Res. Dev. 2025, 29, 1400–1410 (CC BY). * The downscaled batch was evaluated by HPLC-UV only, not by mass spectrometry, so those two values are not strictly comparable to the others.

Reading the productivity figure carefully

Productivity per liter of resin was 9.88 g/L_resin/h for MCSGP against 10.09 for the downscaled batch — effectively the same. The gain over the pilot batch process, at 5.28, comes from bed height: the downscaled column is half as tall, so residence time during elution halved and productivity roughly doubled. In practice, half the bed height with twice the productivity gives similar overall throughput, since throughput is productivity multiplied by column volume.

The 87% throughput advantage the paper reports is a different comparison: an MCSGP system runs two columns rather than one, at the same column diameter. Put another way, one elution in MCSGP completed in 38 minutes against 90 minutes for one batch elution from the same 10 cm column. Three things contribute: the higher yield, the higher elution flow rate used in MCSGP (200 against 150 cm/h, which carries less risk because recycling safeguards yield), and the fact that regeneration, cleaning and feeding run in parallel with elution on the other column.

Scale-up scenarios

MCSGP processes developed on the Contichrom CUBE are directly scalable to the Contichrom TWIN, a production system capable of operating under GMP. The study projects four configurations.

Swipe the table sideways to see all four configurations.

Source: Table 5 of the study. These are calculated scale-up scenarios rather than demonstrated runs. The paper states that the largest configuration, assuming 24 hours of continuous operation, enables multiton-scale production per year.

7. What This Means for Oligonucleotide Manufacturing

The headline number is a 13 percentage point gain in yield, but the argument the paper makes is about what that gain does upstream. Higher chromatographic yield allows the chemical synthesis batch to be scaled down proportionally for the same output of product.

That matters because synthesis is where the raw material cost, the custom intermediates and the solvent burden sit. The authors report that for equivalent product output, MCSGP carries 14% lower cost of goods, and they attribute the saving to fewer raw materials and consumables, fewer custom oligonucleotide intermediates, smaller synthesis batches and smaller volumes of chromatography resin. On resin specifically, they calculate that a production-scale single-column batch system requires 86% more resin than MCSGP for equivalent output — and note that this figure may understate the difference, because batch chromatography is not typically run continuously.

Alongside the measured process improvements, the paper sets out a series of operational benefits it expects from automation: fewer in-process control samples and so a reduced GMP analytical burden, the elimination of rechromatography with its storage bags, freezer space and diafiltration, less operator interaction and less scope for user error, and smaller columns that are easier to pack. These are argued rather than measured in this study.

One methodological note worth carrying forward: the anion exchange method here reached above 95% purity in both batch and MCSGP modes, where the authors say NaOH-based AIEX methods in their experience typically give 88–92%. They suggest the phosphate-buffered mobile phase with 10% acetonitrile may have improved selectivity by reducing unwanted hydrophobic interactions, but also note the crude entered at a high starting purity of 85%.

The same conversion was carried out in reversed-phase mode on a GalNAc-conjugated gapmer, reported in a companion study in this series. For the platform view of where MCSGP fits in an oligonucleotide workflow, see oligonucleotide purification with MCSGP.

8. Key Terms in This Study (Glossary)

9. Frequently Asked Questions