This study focuses on a step-by-step optimization of commercial scale preparative HPLC purification of cannabinoids, specifically cannabidiol, starting with stationary reversed-phase screening, followed by method optimization.
The downstream processing of natural cannabinoids for pharmaceutical, nutraceutical, and cosmetic applications is increasingly challenging when targeting purities as high as >98% for different cannabinoids. Such a high purity can only be achieved with chromatography, which is a highly effective, but costly separation process. Therefore, the high efficiency of this chromatography step is crucial for the overall production costs. This study focuses on a step-by-step optimization of commercial scale preparative high performance liquid chromatography (HPLC) purification of cannabinoids, specifically cannabidiol (CBD), starting with stationary reversed-phase screening, followed by method optimization. Comparison of various reversed-phase stationary phases characterized by standard Tanaka-testing reveals a strong dependency of CBD on hydrophobic interactions in contrast to cannabichromene (CBC) and cannabinol (CBN). The optimized method gives high purity CBD (>99%) in a reasonable yield (77%) and is capable of tetrahydrocannabinol-isomer (THC-isomer) isolation on the preparative scale. This study also aims to answer: How to interpret the characteristic dimensionless chromatography parameters, such as selectivity? When is a change of the value an actual optimization and not just a deviation?
The recent surge of cannabinoids in medical, cosmetic, and nutritional applications requires high purity product and good manufacturing practices. The most challenging aspect of cannabinoid purification from plant-based extraction is maintaining the integrity of the natural products while still complying with the strict regulations. This specifically relates to levels of the cannabinoid tetrahydrocannabinol (THC), which is well known for its psychotropic effects. Recent developments in processing technology have increased the number of possible techniques to extract and purify cannabinoids, but chromatographic purification has proven to be the most cost-effective process that yields the highest purity product. The chromatographic purification process can be optimized depending on the specific objective by using a basic set of parameters. In particular, the downstream purification steps are the key in obtaining a white, crystalline product that is required for pharmaceutical or nutraceutical applications. The US Food and Drug Administration (FDA) recently started introducing guidelines for standardized tests (1), because slowly but surely the use of cannabidiol (CBD) and other cannabinoids are finding their way into clinical trials and pharmaceutical research (2).
There are a wide range of publications and application notes on analysis and the chromatographic separation of cannabinoids focusing on CBD (3), cannabidiol acid (CBDA) (4), and other cannabinoids (5), most of which focus on either analytical purification or only result in 80% purity (3-5). Alternatively, the large-scale purification systems show high yield and purity but use methanol or acetonitrile and formic acid as additives to achieve a better performance (6). In scale-up the use of toxic solvents, such as acetonitrile may result in additional challenges, such as the removal of acetic acid requiring extra processing steps. Faced with large-scale processing constraints the preparative method should only use ethanol as mobile phase without any additive in the final chromatographic purification step. A robust, scalable chromatographic purification method is developed using standard comparison methods. The use of ethanol as mobile phase at a concentration above 50% (V/V) restricts to reversed-phase chromatography. In this study, first we tested the hydrophobicity of different C8 and C18 stationary phases characterized by standard Tanaka-testing(7). The best performing column is further used for general chromatography method optimization, while respecting the production process constraints (8,9).
All methods were performed and optimized with respect to the downstream processing constraints: the mobile phase used was pure ethanol (96% EtOH and water) without any buffer salts or additives. Cannabinoid oil is only soluble in an ethanol concentration larger than 50% (V/V) with respect to water.
For preparative scale HPLC with columns of the size 80 x 500 mm a Labomatic HD5000 with DAD 5600 UV-vis detector was used at a flowrate of 100 mL/min unless stated otherwise. For analytical HPLC (4.6 x 250 mm) a Dionex Ultimate 3000 HPLC with MWD detector was used at a flowrate of 1 mL/min unless stated otherwise. The CBD crude was provided by Schibano Pharma AG, diluted in ethanol (96%) with a ratio of 1:1 (V/V) and spiked with uracil as dead-time marker with a final concentration of 0.1 mg/mL. For standard tests, the wavelength 254 nm was used and for the analysis of cannabinoids additionally the wavelength of 284 nm. Standard injection amounts of the crude were 2 mL or 6 µL in preparative and analytical scale, respectively.
The standard gradient method used was a standard preparative run starting at 70% A (96% ethanol), increased up to 90% A within 90 min, and held at 90% A for 10 min before going back to 70% A. Equilibration time between runs is at least 30 min at preparative scales.
The standard analytical method for quantification of cannabinoids used was isocratic methanol-water (60:40) with 0.1% acetic acid on 100 Å C8 5 µm 4.6 x 250 mm column.
The standard reversed phase test used was at isocratic conditions of 74% A (96% ethanol) a mixture of 0.15 mg/mL uracil, 1.5 mg/mL phenol, 4.0 mg/mL diethyltoluamide (DEET), and 10 mg/mL toluene. Flowrates were 1 mL/min in analytical HPLC and 100 mL/min in the preparative setup. Tanaka tests (phenylbenzene, butylbenzene, caffeine, and phenol) were performed according to the original literature (7) in 80% and 30% methanol in water, respectively, only in analytical conditions.
The stationary phases used were Nucleosil 100 Å C8 in 4.6 x 250 mm and 80 x 500 mm format columns and ZEOsphere 100 Å C18 / 10 µm 4. 6 x 250 mm and 80 x 500 mm format columns. All materials are silica based spherical functionalized beads with an average pore size of 100 Å and 10 µm average particle size.
Retention factors k’ stated always refer to uracil as dead time marker and are calculated according to the following formula (Equation 1):
Selectivity of two compounds is the ratio between their retention times and was calculated using their retention factors (Equation 2):
Packing quality was determined with a standard reversed-phase test in conditions close in line with the separation conditions in 74% A. There is a substantial difference between the porosity of the C8 phase in analytical and preparative scale (> 30%) and the reversed-phase test results in prep-scale indicate the poor packing of this column (N < 3000). Therefore, further comparison of this C8 phase is only considered from the analytical column.
For separation efficiency of the stationary phase, the selectivity between CBD and THC based on their retention factors was evaluated(10). THC and its derivatives (∆9- and ∆8-THC) are indeed the critical impurities in this preparative separation process. Other selectivity’s, such as CBD-CBN, correlate linearly with the CBD-THC selectivity (see Figures A1 and A2).
Looking at the selectivity of CBD and THC (Figure 2) and the dependency on the hydrophobic selectivity of the stationary phase, it is clear that the high hydrophobicity of the end-capped C18 phase is required in this case and within the given constraints C18 is the phase to go for in all scales.
The poor retention factor of the C8 80 x 500 mm column is another proof that the packing of this larger column is not satisfactory. The high dead volume (Table I) of 0.5 L more than the standard packing shows that the preparative C8 column has collapsed for unknown reasons. This finding emphasizes the importance of regular standard tests to determine packing quality, dead volume, and the general state of the column before the application testing.
In an extensive screening sequence, we compared different C8 and C18 reversed phases under the given conditions (Figure 3) and tested how the cannabinoid selectivity behaves with standard Tanaka-test based characteristics(7), such as hydrophobic selectivity, hydrogen bonding capacity, and shape selectivity. The highest variation (steepest parameter increase) is visible with the hydrophobicity of the stationary phase and the selectivity between CBD and THC, which is shown in Figure 3. The phase with the highest hydrophobic selectivity also shows the highest selectivity between all cannabinoids. Thus, the highly hydrophobic stationary phase is best suitable using ethanol-water as mobile phase.
Additionally, this helps us choose the selectivity between CBD and THC, even though they are not neighboring elutes, as parameter for optimization, because it indicates the highest sensitivity during the testing.
The process constraints in the larger scale are use of ethanol as mobile phase without additives at a concentration of at least 50% ethanol (V/V). At the lower ethanol concentrations, the cannabinoid-oil is not soluble anymore and separates as liquid phase.
Since the dimension of the preparative column and the choice of mobile phase are fixed, what can be optimized further?
In the following section, the different chromatographic tools for selectivity improvement are investigated and optimum separation conditions are determined.
Since we will be comparing dimensionless factors, such as selectivity and retention factors, it is important to define when the change is significant and when it is just variation within standard deviation of the system. This is best done while we investigate the reproducibility of the chromatographic systems. All our experiments have been performed multiple times and the collected data gave the following standard deviations (absolute) as shown in Table II.
Table II shows the average retention factors and selectivities and their standard deviations. We chose to show the factors with the highest absolute standard deviations in this table. Any change in selectivity α larger than 0.15 is significant and is not caused by the preparative HPLC system. In this simple summary we can additionally show that both system scales, analytical and preparative, are well comparable, since the selectivities are within the defined
Optimization of the Gradient
The gradient change of the eluent during the separation has a big impact on selectivity and efficiency (runtime) of the process. Using the gradient can potentially increase selectivity and peak purity. Different gradients have been tested within the same runtime by variation of the ethanol starting concentration and the selectivities and the retention factor of CBD are shown in Figure 3. The isocratic conditions at 70% ethanol show the best selectivity between CBD and THC. With the least steep gradient starting at 75% ethanol the runtime (k’ CBD) can be decreased, but the selectivity is still significantly lower (αisocratic = 2.21 versus α75-90%EtOH= 1.93) than isocratic conditions.
In both scales (Figure 3), the selectivity increases with higher ethanol starting concentrations and less steep gradients. This leads to the conclusion that isocratic conditions may be indeed beneficial for this separation. With higher ethanol concentrations the retention factor of CBD decreases. Theoretically, in an efficient separation process the retention factors of target molecule should be between k’ >= 2 <= 5. A k’>2 guarantees a stable process (11,12). Thus having the higher ethanol starting concentration than 75% would further decrease k’[CBD] and increase the risk of an instable process, since the molecule is not properly retained on the stationary phase anymore.
Why is the impact of the gradient on the selectivity with the C18 much bigger than the C8? This may be because the interaction of the C18 stationary phase with the cannabinoids is much greater which leads to higher retention times. Consequently, the impact of concentration change in the eluent is also greater.
Optimization of Injection
The separation selectivity with a gradient method can be improved by compressing the sample onto the column or the opposite, so-called injection equilibration. During injection equilibration the starting concentration of the mobile phase (in this case 70% pure ethanol) is held constant for a certain amount of time before the gradient starts.
The first option, compression, would increase locally the sample concentration and lead to liquid-liquid phase separation, thus this is not applicable. The latter—injection equilibration—is tested in dependence of a factor of the column volume as shown in Table III. In these experiments, the ethanol concentration is isocratic at the beginning and followed by the standard gradient method to 90% A within 90 min.
In both scales—analytical and preparative—the selectivity is not much influenced by an equilibration time larger than the column volume. All changes of selectivity are within the reproducibility tolerances (Table II).
Optimization of the Flowrate
From previous tests we concluded that isocratic conditions show the beneficial effect on the selectivity, we can base the future optimization on this isocratic separation. Now the question arises if we are running at the optimal flowrate. In the following, the isocratic separation method (74% A) varied in flowrates from 0.05 mL/min–3 mL/min and 60–220 mL/min in the analytic and preparative setup, respectively. In a chromatographic separation, mass transfer effects, diffusion, and intermolecular interactions are dependent on the flowrate v (13). The Van Deemter Equation summarizes the following dependencies of non-retained species (Equation 3):
where A is the constant eddy diffusion; B is the longitudinal diffusion alongside the column (impact decreases with flowrate); C is the mass transfer between stationary and mobile phase—increases with flowrate and pressure; HETP is the height equivalent to a theoretical plate should be minimal for the best separation efficiency and is calculated by the number of theoretical plates N and taking the column length into account.
HETP in each case is calculated with the total number of plates and the length of the column(13,14). HETP, however, only takes mass transfer due to diffusion and flow into account and no chemical interactions such as repulsion or absorption.
The resolution parameter R—in contrast to HETP—is additionally taking into account the chemical interaction of the cannabinoids with the stationary phase. The peak resolution Rs is an estimate for the separation efficiency of the system and calculated according to the following Equation 4:
tRi being the retention time and wb being the peak width at 10% peak height.
The calculation only using HETP indicates an optimum flowrate with minimal plate height for toluene at a flowrate of 0.1–0.2 mL/min. The resolution (Figure 4) was determined between the neighboring cannabinoids CBD and CBN, which is eluting directly after CBD. All major cannabinoids of this crude (CBD, CBN, ∆9-THC, and CBC), however, behave similarly. Appendix Figures A1 and A2 show the trends in selectivity and plate height of all of them.
Figure 4 shows the dependence of resolution on the linear flowrate, which is comparable for both preparative and analytical systems. The linear flowrate is independent on the column diameter. The analytical column with an inner diameter of 4.6 mm was tested between 3 and 0.01 mL/min, which translates into 18 to 0.3 cm/min linear velocity. The preparative system was tested from 60 to 220 mL/min, which is in between 1.2 and 4.4 cm/min, thus well within the same range.
There is obviously no maximum in resolution as the resolution increases drastically with lower flowrates. Both scales, analytical and preparative, show the typical pay-off between productivity (flowrate) and purity (resolution). The optimal flowrate is therefore not determined by a particular maximum of resolution, but by the best productivity comprising runs per hour/day, loading of the crude onto the column, and recovery of solvent.
Increasing the Sample Load
Since we are interested in preparative purification of CBD the next step is to see how the purification performs at higher concentrations and ultimately how much sample can be loaded per separation run, which is called loadability. In the previous experiments in preparative setup, we injected 2 mL CBD oil in ethanol. Figure 5 shows the chromatogram of the higher injection: 50 mL of crude oil sample been separated and the collected fractions have been analyzed and quantified with standardized analytical method. Injecting 50 mL of the ethanol diluted crude oil represents a loading of 2wt.% onto the HPLC column (80 x 500 mm). Table IV gives the purity and size of the major fractions.
At higher loading there is no baseline separation anymore, however, >77% of the fractions (each 100 mL) contain pure CBD. First fraction still contains cannabigerol (CBG). We state purity >99%, since only known impurities are included in this analysis due to unknown response factors of unknown impurities. The usual yield-purity payoff starts to show at this loading: Higher loading may increase throughput, but at the cost of number of fractions that achieve the required purity. For crystallization of the pure cannabinoid only fractions with >99% purity is used.
In this study, we compared and optimized chromatographic reversed-phase separation of cannabinoids and investigated separation efficiency of CBD and THC. The first screenings showed superiority of the C18 compared to C8. Both the analytical system (with a standard sized 4.6 x 250 mm column) and the preparative HPLC system (with an 80 x 500 mm column) shows good reproducibility and comparability.
After choosing the best performing stationary phase, we optimize the preparative chromatography process using pure ethanol as mobile phase with a minimum concentration of 50% (V/V) in water. These process constraints result from the upstream and downstream processing of pure cannabidiol at large scale. We looked at different ethanol-water gradients, the injection method and how the flowrate impacts peak resolution. Best selectivity between THC and CBD was obtained using standard isocratic conditions at 70% ethanol. This method even achieved baseline separation between the isomers of ∆9-THC and ∆8-THC. With this optimized separation process a higher load of crude (50 mL) on the C18 phase yielded in seven out of nine (77% yield) with a CBD purity greater than 99% and without any known impurities. This purity criteria is mandatory for pharmaceutical use of CBD. In the next step, we would determine the loadability with even higher loading than 50 mL of crude oil (2wt.%), which then defines the economics of this separation process.
This online version of this article includes Appendix Table AI and Figures A1 and A2. Those graphics did not appear in the print edition.
(1) FDA, “Cannabis and Cannabis-Derived Compounds: Quality Considerations for Clinical Research Guidance for Industry”, draft guidance, 1st. June 2020, FDA-2020-D-1079.
(2) K.M. Nelson, J. Bisson, G. Singh, J.G. Graham, S. Chen, J.B. Friesen, J.L. Dahlin, M. Niemitz, M.A. Walters, and G.F. Pauli, J. Med. Chem. 63(21), 12137–12155 (2020).
(3) S. Marzorati, D.Friscione, E.Picchi, and L.Verotta, Industrial Crops and Products155, 112816 (2020).
(4) Cannabinoid purification, Application Note 21-001, Zeochem AG, (2021).
(5) F. Leyva-Gutierrez, J. Munafo Jr., and T. Wang, J. Agric. Food Chem. 68(29), 7648 (2020).
(6) A. Wohlfarth, H. Mahler, and V. Auwärter, J.Chrom. B, 879(28), 3059-3064 (2011).
(7) K. Kimata, K. Iwaguchi, S. Onishim K. Jinno, R. Eksteen, K. Hosoya, M. Araki, and N. Tanaka, J. Chrom. Sci. 27, 721-728 (1989).
(8) A. Rathor and A.Velayudhan, BioPharm International, 34-42 (2003).
(9) H. Schmidt-Traub, Preparative Chromatography of fine chemicals and Pharmaceutical Agents Chapter 4, Criteria for Choice of Chromatographic Systems, Wiley-VCH, 157ff (2005).
(10) G. Barth, LCGC North America 36(7), 472-473 (2018).
(11) Back to Basics - Gradient Retention Factor, K*, LC Technical Team, Schimadzu, 2020.
(12) European Pharmacopeia 7.0, Ch 2.2.46., Chromatographic separation techniques, 70-77 (2015).
(13) JJ. van Deemter, F.J. Zuiderweg, and A. Klinkenberg, Chem. Eng. Sci. 5(6), 271-289 (1956).
(14) O. Kaltenbrunner and P.Walter, Progess in Biotechnology16, 201-206 (2000).
Dr. Urs Lengweiler is with Schibano Pharma AG in Switzerland. Dr. Ana Stojanovic and Dr. Victoria B.F. Custodis are with Zeochem AG in Switzerland. Direct correspondence to: firstname.lastname@example.org