Residual solvent compliance in pharmaceutical liquid-liquid extraction is not a single post-crystallisation drying requirement; it is embedded in solvent selection, phase equilibria, and the number of theoretical stages allocated to each separation. The regulatory framework in ICH Q3C(R8) categorises solvents into Class 1, Class 2, and Class 3, with Class 1 solvents such as benzene and carbon tetrachloride controlled at single-digit ppm limits and Class 2 solvents assigned both permitted daily exposure values and concentration limits normalised to a 10 g/day daily dose. For a solvent with a PDE of 8.9 mg/day, toluene, the corresponding Option 1 limit is 890 ppm; for dichloromethane, a PDE of 6.0 mg/day yields a limit of 600 ppm. A liquid-liquid extraction train for a weakly basic API may use toluene or dichloromethane as the organic phase, dilute hydrochloric acid at pH 2.0–3.0 as the selective extraction phase, and sodium hydroxide solution at pH 9.0–10.0 for re-extraction of the free base. In such a train, the residual solvent content of the isolated intermediate is determined by the partition coefficient of the solvent between the organic and aqueous phases, the mutual solubility of the two phases, the volumetric phase ratio, and the mechanical entrainment generated by the mixing device. Changing solvent polarity alters all four factors simultaneously, because a solvent with a higher polarity index may improve solute extraction efficiency but also reduce interfacial tension, increase mutual solubility, and create smaller droplets that require longer residence time for coalescence. This interaction means that process development for residual solvent limit compliance requires a thermodynamic screening step followed by a hydrodynamic verification step on production-scale extraction equipment.

Can Solvent Polarity Index Values Provide an Adequate Rank Order for Residual Solvent Control in Multi-Stage Pharmaceutical Extraction?

Polarity ranking alone cannot establish a residual solvent control strategy because solvent polarity indices are derived from normal-phase chromatographic strength and do not reflect the aqueous solubility or regulatory limit that governs final carryover. The Snyder polarity index places n-hexane at 0.1, toluene at 2.4, dichloromethane at 3.1, tetrahydrofuran at 4.0, chloroform at 4.1, ethyl acetate at 4.4, acetone at 5.1, acetonitrile at 5.8, methanol at 5.1, and water at 10.2. Despite the similarity in polarity between dichloromethane and tetrahydrofuran, the ICH limits differ: dichloromethane has a limit of 600 ppm, while tetrahydrofuran has a limit of 720 ppm, and tetrahydrofuran is freely miscible with water whereas dichloromethane has a water solubility of approximately 13 g/L at 20°C. Toluene has a low water solubility of approximately 0.52 g/L at 20°C and a limit of 890 ppm, while acetonitrile has a limit of 410 ppm and complete water miscibility. These differences illustrate that a higher polarity solvent may be removed more easily during a water wash if it is water-miscible, but the same miscibility can reduce phase recovery and raise aqueous waste residual levels. The distribution ratio of a monobasic amine API can vary by several orders of magnitude over a pH interval of 2–3 units, so pH adjustment is often a stronger lever than solvent polarity adjustment for extraction completeness. The table below summarises selected solvent parameters relevant to residual solvent control in pharmaceutical extraction.

Independent of the thermodynamic partition coefficient, residual solvent carryover in pharmaceutical liquid-liquid extraction is often dominated by microdroplet entrainment, particularly when the process is scaled from a laboratory separatory funnel to a continuous mixer-settler. The terminal settling velocity of a dispersed organic droplet in a continuous aqueous phase can be estimated from Stokes settling when the droplet Reynolds number is below 0.1. For a dichloromethane droplet with diameter 25 μm, a density difference of approximately 300 kg/m³, and a continuous-phase viscosity of 1.0 mPa·s, the terminal velocity is below 0.1 mm/s. A decanter with a residence time of 15–30 min and a liquid depth of 0.5 m cannot completely separate such droplets, so the aqueous raffinate leaves with both dissolved dichloromethane and suspended organic droplets. During downstream pH adjustment, evaporation, or crystallisation, these droplets coalesce or become incorporated into the product, and the residual solvent level then exceeds the equilibrium solubility prediction. The problem is made more severe by solvent polarity modification: replacing toluene with ethyl acetate increases mutual water solubility and lowers interfacial tension from approximately 36 mN/m to 6 mN/m against aqueous solutions, producing smaller droplets under identical impeller tip speed. Published data for this exact continuous extraction train is limited, but the trend is consistent with standard scale-up correlations for agitated liquid-liquid dispersion, which predict mean droplet diameter decreasing with interfacial tension divided by power dissipation per unit mass. The equipment configuration must therefore be matched to the solvent polarity target; increasing polarity to improve extraction selectivity may require reducing impeller speed, increasing settling area, or adding a coalescer cartridge to maintain residual solvent limits.

When a Class 2 Toluene Extraction Is Followed by an Aqueous Stripping Stage at Low pH

Consider a process in which a weakly basic API is taken into toluene at pH 9.0 and back-extracted into 0.5 M hydrochloric acid at pH 2.5. The protonation of a base with a pKa of 8.0 at pH 2.5 produces an ionised fraction greater than 99.99%, so the distribution ratio between organic and aqueous phases becomes very low and the extraction yield is controlled primarily by phase contact efficiency, not by solvent polarity. Toluene has a water solubility of approximately 0.52 g/L at 20°C, and the ionic strength of the acidic aqueous phase can further reduce this solubility by a salting-out effect; however, the same ionic strength may increase emulsion formation if the solute behaves as a surface-active ion-pair. Residual toluene in the aqueous strip is then carried forward to the final re-extraction stage. For a product with a daily dose of 10 g/day, toluene is limited to 890 ppm; for a product with a daily dose of 2 g/day, the Option 2 calculation gives a permissible concentration of 4450 ppm, demonstrating that the same extraction train can pass or fail depending on the dose-normalised limit. The critical processing window is narrow because a pH below 1.5 may hydrolyse acid-labile ester or acetal protecting groups, while a pH above 6.0 reduces ionisation of a pKa 8.0 base below 99% and causes yield loss. Therefore the extraction is bounded on one side by degradation kinetics and on the other side by extraction equilibrium. Residual solvent limits are not independent of these pH boundaries: a lower pH may allow a longer aqueous wash but increases the risk of acid-catalysed degradation, and a higher pH may require additional organic extraction stages with increased solvent inventory. Continuous mixer-settler trains have exhibited production-scale failure modes in which interface-level drift caused by rag-layer accumulation and pH probe fouling in the aqueous strip leads to local phase-ratio excursions and residual solvent values above the predicted equilibrium. These failure modes are more common when the organic phase contains suspended API particles or when the aqueous phase contains buffering salts that precipitate on the pH probe surface.

At manufacturing scale, the choice of extraction equipment imposes additional constraints on solvent polarity control. Annular centrifugal contactors, for example, can handle phase ratios down to 1:10 and residence times as low as 5 s, but they generate high shear and may produce droplets below 10 μm when the organic phase has a low interfacial tension. A Podbielniak or Robatel centrifugal extractor can provide several theoretical stages in one unit, but the seal material must be compatible with chlorinated solvents and acidic brines; polytetrafluoroethylene and perfluoroelastomer seals are typically specified. In contrast, a mixer-settler train with separate decanters allows longer residence times of 15–30 min per stage, which is beneficial for coalescence but increases solvent inventory and may require nitrogen inertisation for peroxide-forming solvents such as tetrahydrofuran. Production-scale batch-to-batch variation in residual solvent levels is frequently associated with changes in feedstock particle size, which alters the rate of solid dissolution during the initial solvent extraction and changes the concentration of surface-active impurities that stabilise emulsions. A batch with a higher fines fraction can increase the rag layer volume in the settler and reduce the effective organic phase recovery. Under these conditions, the residual solvent content of the final isolated API may vary by a factor of 2–3 despite unchanged operating setpoints. This variability must be addressed through feed conditioning, such as filtration before extraction, or through a coalescer installed after the primary decanter, rather than by solvent polarity adjustment alone.

Residual Solvent Partitioning During Phase Splitting in Mixer-Settler Trains

The phase-splitting step in a mixer-settler train is the point at which thermodynamic residual solvent and mechanical carryover diverge. The organic phase leaves the settler as a clarified supernatant, while the aqueous phase leaves through a bottom nozzle or a weir. The measured residual solvent in the aqueous product stream can be expressed as the sum of the equilibrium solubility term, which is governed by temperature and ionic strength, and an entrainment term, which is governed by droplet size distribution and residence time. For a dichloromethane-water system at 20°C, the equilibrium aqueous solubility is approximately 13 g/L, but a well-separated aqueous raffinate should contain far less than this value if the organic phase is not saturated with water and the entrainment term is controlled. The entrainment term is not a fixed percentage; it depends on the rag-layer thickness at the interface, the specific gravity difference between phases, and the presence of surface-active impurities. A rag layer with a thickness greater than 20% of the settler height can reduce the free disengaging area and cause local velocities that drag small droplets into the outlet nozzles. Temperature control is also a compliance variable because dichloromethane solubility in water decreases with increasing temperature, while toluene solubility is relatively flat over a 10–30°C window. Heating the settler to 30°C can therefore reduce dissolved dichloromethane in the aqueous phase, but it also increases vapour pressure and requires condenser recovery to maintain workplace exposure limits. The selection of a solvent with a higher polarity index does not eliminate this balance; a water-miscible solvent such as acetonitrile may not present a phase-splitting problem when the feed is fully miscible, but it may require a subsequent aqueous extraction or solvent swap to move the API into a Class 2 or Class 3 solvent for isolation. Published data for this specific configuration is limited, and the operating window must be established on the actual production train.

Verification of residual solvent limit compliance after liquid-liquid extraction requires a sampling strategy that accounts for phase-separated liquid and for the isolated solid API. The pharmacopoeial method described in USP <467> assigns Procedure A to water-insoluble samples and Procedure B to water-soluble samples, with headspace gas chromatography as the principal separation technique. For an extraction train using dichloromethane and toluene, the analytical method must resolve both solvents from matrix peaks and from the process-derived degradation products. A headspace column with a stationary phase of 6% cyanopropylphenyl and 94% dimethylpolysiloxane can provide adequate separation, but the exact oven program and split ratio must be optimised for each matrix. The limit of quantitation for Class 1 benzene should be below 2 ppm when a daily dose of 10 g/day is used, and a method validation according to ICH Q2(R1) should establish specificity, linearity, precision, and accuracy across the expected range. Process control samples taken from the aqueous strip, the re-extraction organic phase, and the final wet cake provide early warning of carryover before drying. The residual solvent level in the dried API is then compared against the Option 1 concentration limit or the Option 2 PDE-based limit. The table below summarises the compliance matrix.