Pilot-Scale Preparation of Broad-Spectrum CBD: Extraction Optimization and Purification using Centrifugal Partition Chromatography Open Access

Cannabis sativa L., widely known as marijuana or hemp, is a botanically and chemically diverse plant with significant pharmacological and commercial importance. The plant’s therapeutic potential is primarily attributed to its cannabinoid constituents, with Δ⁹-tetrahydrocannabinol (Δ⁹-THC) and cannabidiol (CBD) being the most studied. Δ⁹-THC, the primary psychoactive compound abundant in marijuana but present at much lower concentrations in hemp has been extensively researched for its analgesic, muscle relaxant, mood-enhancing, appetite-stimulating, sedative, and antiemetic properties [1‒3]. However, due to its psychoactive effects and regulatory restrictions, interest has shifted toward CBD, a nonpsychoactive cannabinoid with significant therapeutic potential, including anticonvulsant, anti-inflammatory, neuroprotective, anxiolytic, and analgesic properties [3‒6]. Additionally, some preclinical studies have explored CBD’s effects in oncology, investigating its potential influence on cancer-related pathways in cultured cell lines and mouse tumor models [7]. However, further research is required to determine its clinical relevance. An extract primarily containing CBD with a limited THC content of less than 0.2–0.3% by weight, called “broad-spectrum CBD,” has surged across multiple industries, including pharmaceuticals, nutraceuticals, cosmetics, and functional foods [4‒6]. This growing market necessitates the development of efficient, scalable, and regulatory-compliant extraction and purification methods to obtain high-purity CBD while minimizing unwanted components, such as Δ⁹-THC, pigments, and waxes [8‒10]. Despite various extraction technologies, challenges remain in optimizing methods that maximize CBD yield while efficiently removing undesirable compounds.

Several extraction and isolation techniques have been explored and reported for cannabinoid recovery, including solvent-based extraction, supercritical carbon dioxide extraction, microwave-assisted extraction, molecular distillation, and chromatographic techniques [9‒22]. Organic solvent extraction is one of the most widely used methods due to its efficiency, scalability, and cost-effectiveness. However, factors such as temperature, extraction duration, and sample-to-solvent ratio significantly influence the extraction efficiency as well as the cannabinoid composition of the extract [8, 14, 15, 23, 24]. Among various solvents, ethanol is preferred in industrial applications because it is generally recognized as safe, inexpensive, and capable of extracting a broad range of cannabinoids [16, 21, 22]. Nevertheless, ethanol also co-extracts undesired chlorophyll and pigments, resulting in dark-colored extracts that require additional purification steps [13‒16]. To refine extraction efficiency and improve cannabinoid recovery, this study employed the Box-Behnken design (BBD), a response surface methodology (RSM)-based statistical approach, to optimize key variables such as temperature, extraction time, and solvent ratios. BBD is advantageous because it enables efficient evaluation of interactive effects among multiple variables while minimizing the number of experimental runs.

While molecular distillation is commonly used for semi-purification of cannabinoids, it requires expensive, high-performance equipment and operates at high temperatures, which can degrade thermosensitive cannabinoids [25, 26]. Chromatographic techniques, particularly centrifugal partition chromatography (CPC), have emerged as a more effective and scalable alternative for cannabinoid purification. CPC is a promising liquid-liquid chromatography (LLC) technique that offers key advantages over traditional solid-liquid chromatographic methods, such as preparative high-performance liquid chromatography (HPLC) and flash chromatography. Unlike conventional chromatography, CPC does not require a solid stationary phase (SP), eliminating concerns about adsorption losses, SP degradation, and high solvent consumption [27‒31]. Instead, CPC operates using two immiscible liquid phases, where one phase serves as the SP retained in the rotor by centrifugal force, while the other acts as the mobile phase (MP). CPC’s advantages, including high recovery rates, fast processing times, and cost efficiency, make it an attractive alternative to solid-phase chromatography and molecular distillation for large-scale cannabinoid processing. The separation process is governed by the partition coefficients (K_D) of the target compounds, enabling selective isolation based on differential solubility [27‒35]. The process relies on selecting an appropriate biphasic solvent system to ensure effective partitioning of cannabinoids while minimizing impurities. Several studies have demonstrated the efficiency of CPC for cannabinoid purification, particularly in selectively removing Δ⁹-THC while retaining CBD and other minor cannabinoids [36‒40]. However, the system optimization – including the selection of a suitable biphasic solvent system, balancing partition coefficients and other factors – is critical for achieving high-purity CBD extracts.

Based on these considerations, this study aimed to develop an optimized and scalable process for the preparation of broad-spectrum CBD with minimal Δ⁹-THC content. Specifically, ethanol-based extraction was optimized using BBD-RSM to maximize cannabinoid yield while minimizing co-extraction of undesirable compounds. In parallel, a CPC-based purification system was developed and optimized to efficiently reduce Δ⁹-THC level, ensuring compliance with regulatory limits and enhancing extract purity. By integrating RSM-guided extraction with CPC purification, the study presents a practical and scalable approach for producing high-quality, broad-spectrum CBD extracts suitable for industrial applications.

CBDA and THCA cannabinoid standards were purchased from Cayman Chemical Co. (Ann Arbor, MI, USA) with CAS numbers (CBDA: 1244-58-2) and (THCA: 23978-85-0) respectively. CBD was purchased from THC Pharm GmbH (Frankfurt, Germany) (CAS: 13956-29-1). Δ⁹-THC was purchased from Lipomed (Arlesheim, Switzerland) (CAS: 1972-08-3). The solvents used for HPLC analysis and CPC, including acetonitrile (CAS: 75-05-8), methanol (CAS: 67-56-1), absolute ethanol (CAS: 64-17-5), isopropanol (CAS: 67-63-0), and n-hexane (CAS: 110-54-3), were purchased from RCI Labscan Limited (Thailand). Analytical grade 98% v/v formic acid (CAS: 64-18-6) was obtained from Fisher Scientific and 97% w/w ammonium formate (CAS: 540-69-2) was obtained from Sigma-Aldrich. Deionized water was obtained from the Milli-Q^® Reference Water Purification Merck System from a representative company in Thailand.

The cannabis samples in this study were divided into four groups. Sample 1 was a mixture of RPF3 cultivar inflorescences from the Highland Research and Development Institute (Public Organization, Thailand) (HRDI) (sample 2, high-fiber type) and the samples from the Narcotics Suppression Bureau (NSB) (sample 3, THC-dominant) in a ratio of 3:1, while sample 4 (CBD-dominant) was a CD1 cultivar from the HRDI. The cannabinoid content of cannabis samples analyzed by HPLC is shown in online supplementary Table S1 (for all online suppl. material, see https://doi.org/10.1159/000546263). The cannabis material was individually dried in a hot-air oven at 50–60°C for 2 h. The dried cannabis materials were ground to a fine powder and sieved through 250-micron particle sieves. Cannabis powder was homogenized by hand mixing and shaking for 10 min. The powder was then weighed, placed in a vacuum bag, sealed tightly, and stored at −20°C before use.

For the experimental cannabis ethanolic extraction, three variables, that is, extraction temperature (−40 to 60°C, variable A), extraction time (15–75 min, variable B), and sample-to-solvent ratio (1:3% w/v to 1:11% w/v, variable C), were studied. Six parameters (% yield [%w/w], % cannabinoid [CBD, CBDA, Δ⁹-THC, THCA] content [%w/w], and lightness [L*]) were monitored. RSM analysis was performed using the BBD in Design-Expert^® software version 12. A BBD consisting of 12 experiments with five center point replicates was created to evaluate the aforementioned responses. Atmospheric pressure was maintained constant during the extraction process. Partial least squares regression was applied to evaluate the fit of the model and the response surface. The F test was used to estimate the significance of the regression coefficients. Modeling was performed using a quadratic model, linear, 2FI, and interaction terms.

Samples 2, 3, and 4 (3 kg each) were separately weighed and extracted under optimized conditions in a pilot-scale triple-wall glass reactor model GR-50TCE (Siam Intercrop Co., Ltd, Thailand) with a stirring reaction bath model DHJF-4030 and a water-circulating vacuum pump model SHB-B95T. This process was performed with constant stirring at 250 rpm. The extracted solution was filtered through Whatman^® No. 1 filter paper and the filtered extracts were evaporated to dryness under reduced pressure. The dried extracts were stored at −20°C until use.

CPC experiments were performed using Centrifugal Partition Chromatography Model CPC240 (EverSeiko Corporation Ltd., Tokyo, Japan). The column was a circular partition disk with 2136 channels, and the total column capacity was approximately 240 mL. A 4-way mode switching valve can operate in either ascending or descending mode. The CPC system was connected to a Flash 100 HPLC pump (LabAlliance, Scientific Systems, Inc., State College, PA, USA) and a UV-Vis detector (Spectrasystem UV1000 (Thermo Scientific, Waltham, MA, USA). Fractions were collected using a fraction collector model Foxy R1 (Teledyne ISCO, Lincoln, NE, USA). The sample was injected manually through a 10 mL loop.

The two immiscible liquids were mixed in a separating funnel, agitated by manual shaking for 15 min, and allowed to equilibrate at room temperature for at least 30 min to ensure that each phase was well-saturated with the other phase. Each liquid was then separately filtered through a 47 mm diameter 0.45-micron nylon membrane filter and transferred to a borosilicate glass bottle. The ascending or descending modes were chosen. In descending mode, the SP is an upper and lighter liquid. The apparatus was first filled with SP. Only then was the denser MP pumped through the SP from the top to bottom of the centrifuge. In the ascending mode, SP is a lower and denser liquid. The lighter MP was pumped through the SP from the bottom to the top of the rotor cartridges. The SP was maintained in each channel by the centrifugal field created by the rotor spin.

Five proposed CPC systems were chosen from preliminary K_D studies and evaluated using a CPC device in ascending mode, where the organic phase served as the MP and the aqueous phase as the SP. Initially, 0.5 g of sample 4 was dissolved in 10 mL of MP in a centrifuge tube and agitated with a vortex shaker for 2 min or until fully dissolved. The prepared solution was carefully loaded into the CPC loop using a glass syringe to prevent air bubbles, ensuring an uninterrupted separation process. Fractions were collected at 1-min intervals and analyzed using thin-layer chromatography (TLC), with confirmation by high-performance liquid chromatography (HPLC). The most suitable solvent system was subsequently reevaluated in duplicate to confirm reproducibility and then applied in a scale-up experiment. The sample load was incrementally increased from 0.5 g by 1 g per run until a noticeable change in separation efficiency was observed.

The TLC analysis was adapted following the system described in the American Herbal Pharmacopoeia [43] and performed to screen the promising biphasic system. The procedure utilized C18 (UV 254) TLC plates (150 µm, 10 cm × 10 cm) and a MP composed of methanol/water (75:25, v/v) containing 0.1% glacial acetic acid. Sample and standard solutions were applied as 5 mm bands, spaced 2 mm apart, positioned 8 mm from the lower edge and at least 15 mm from the sides. A flat-bottom chamber (14 cm × 14 cm × 8 cm) lined with filter paper was saturated with ∼25 mL of the MP for 15 min before development. The plate was developed to a distance of 60 mm, then removed and dried at 70°C for 2 min. Visualize the plates under UV 254 nm, then spray the plates with the vanilin/H₂SO₄ reagent, followed by visualization under white light.

HPLC was conducted using an Agilent 1260 HPLC/DAD system (Agilent Technologies, CA, USA), equipped with a G7129A 1260 vial sampler, a G7115A 1260 Quat Pump VL, and a G7115A 1260 DAD WR. Cannabinoid separation was performed using an Agilent Infinity Lab Poroshell 120 EC-C18 column (4.6 × 5 mm, 2.7 µm) with a Phenomenex security guard cartridge C18 (4 × 3.0 mm ID). The MP consisted of 20 mm ammonium formate (pH 3.6) as MP A and 0.1% (v/v) formic acid in acetonitrile as MP B, with a gradient elution from A:B = 35:65 to 18:82 (v/v). The flow rate was maintained at 1.0 mL/min, with an injection volume of 5 µL. The column temperature was set to 40°C, and detection was carried out at 220 nm. Peak integration was performed using Agilent LC Open LAB (offline) software.

Dried cannabis plant material was processed according to Jaidee et al. [44]. The plant material was finely ground and sieved, then 20 mg of the sieved powder was extracted with 5 mL of 80% (v/v) methanol in an ultrasonic bath at 25°C for 15 min. The extract was centrifuged at 3,500 rpm at 4°C for 10 min, and the supernatant was filtered through a 0.45-µm nylon syringe filter before being transferred to an amber glass vial for HPLC analysis. Additionally, homogenized extracts (10 mg) were dissolved in HPLC-grade methanol to achieve a final concentration of 0.1 mg/mL, filtered through a 0.45-µm nylon syringe filter, and analyzed for cannabinoid content.

Descriptive and inferential statistics were used to summarize the data in terms of mean with standard deviation and coefficients of regression (R²) of the linear graph. The results are expressed as the mean ± SD. Statistical significance was assessed by one-way analysis of variance (ANOVA) using SPSS 20 (SPSS Inc., Chicago, IL, USA), followed by Tukey’s test. Results with p < 0.05 were considered statistically significant when p < 0.05. The Pearson correlation coefficients (r) of the results were computed to measure the linear relationship between variables.

In order to optimize the extraction conditions, different samples were used to ensure that the final method was robust, reproducible, and applicable to various cannabis samples. Initially, sample 1 representing a cannabis sample containing both THC and CBD was used in the BBD-RSM experiment to optimize critical parameters related to the target responses, including maximizing extraction yield, increasing cannabinoid content, and improving the color of the extracted product. Once the optimal conditions were established, samples 2 (high-fiber type), 3 (THC-dominant), and 4 (CBD-dominant) were individually used to evaluate the method’s effectiveness in a realistic extraction scenario at both laboratory and pilot-scales. The crude extract from sample 4, obtained under optimized extraction conditions, was selected for CPC separation due to its high CBD content ensuring effective purification and separation efficiency.

BBD-RSM was performed using the extraction temperature, extraction time, and sample-to-solvent ratio as independent variables to determine the ideal conditions that provided the maximum extraction yield, high cannabinoid content, and light color of the extracts. The selection of parameter ranges was based on the study’s objectives, equipment capacity, literature reviews, and practical considerations for industrial applications. Preliminary experiments were conducted to evaluate the behavior of target analytes under various conditions. For example, the reported temperature range for cannabis extraction varies widely, from below −20°C [14] to above 40°C [17]. Similarly, extraction durations in previous studies range from 45 to 60 min [18, 22], while the sample-to-solvent ratio can vary between 1:5 and 1:15% w/v [13]. Each factor influences the extraction outcome depending on the target product characteristics and the technique employed. Therefore, the temperature, extraction time, and sample-to-solvent ratio were systematically optimized using RSM to identify the most efficient extraction conditions.

The BBD-RSM suggested 17 runs, comprising 12 experimental runs with five center point replicates. Two additional runs under optimal conditions (X1 and X2) were conducted at the end of the study to determine whether optimal conditions could provide the expected results. The data for the actual and predicted responses are listed in Table 1. The response was defined as the % yield, % CBD, % CBDA, % Δ⁹-THC, % THCA content, and lightness. Most of the observed and predicted data were in agreement, with no more than a 10% difference, implying that optimized conditions can be used. The results from BBD-RSM suggested that the optimal conditions were −31°C, 33 min, and 1:8% w/v sample-to-solvent ratio, with a desirability of 0.576. Although this value of 0.576 may not appear particularly high, it reflects a realistic compromise among multiple optimization goals. The optimization aimed to maximize the extraction yield and cannabinoid content while also improving the lightness of the extract to ensure acceptable appearance and quality for further processing. These responses often have competing tendencies – higher yields may co-extract pigments – so a moderate desirability score was acceptable. In this study, the primary response of interest – cannabinoid content – was prioritized, while yield and lightness were also considered to ensure practical applicability.

A summary of the ANOVA statistical data for all the responses observed in the BBD-RSM is shown in online supplementary Table S2. The equations for the prediction of % yield, % cannabinoid content (CBD, CBDA, Δ⁹-THC, and THCA), and lightness created by the program are listed in online supplementary Table S3. All model parameters fell within an acceptable range, indicating good model fit and predictive reliability. Therefore, all models can be used to predict the trends of all responses.

Regarding to Equation 3 in online supplementary Table S3, the coefficient of extraction temperature (3.64) was greater than that of both the extraction time (0.2502) and sample-to-solvent ratio (0.1473), indicating that temperature had the most significant influence on extraction yield. The interaction between temperature and time also showed a considerable impact on the responses, whereas the sample-to-solvent ratio had a lesser effect.

The predicted equations reveal that temperature and extraction time showed significant effect on the extraction yield, but a contrasting effect on the cannabinoid content and lightness. As shown in Figure 1, increasing temperature and time led to increased yield; however, these conditions also resulted in low cannabinoid content and darker extract color. Furthermore, temperature had a greater effect on extraction yield than extraction time. This is consistent with previous findings that temperature enhances extraction efficiency for many phytochemicals [13, 14, 19‒24]. For instance, Addo et al. [13] and Chang et al. [14] reported that heat treatment can enhance cannabinoid extraction efficiency.

However, cold ethanol extraction is a more promising technique for cannabis extraction, as it effectively minimizes the co-extraction of undesirable compounds such as chlorophylls and waxes. This approach is particularly advantageous for cannabinoid-rich extracts, allowing for selective extraction of lipophilic cannabinoids while minimizing pigments and polar impurities. At ultralow temperatures, cannabinoids can be selectively extracted with ethanol, whereas impurities like chlorophyll, waxes, and sugars typically require higher temperatures. As a result, low-temperature extraction may produce lower yield but higher cannabinoid purity and better extract appearance. Therefore, careful optimization of the temperature is essential to balance cannabinoid concentration with extract purity.

r represents the relationship between the independent variables and responses indicates that there is a significant relationship between the two responses (r > 0.75 and r < −0.75, p < 0.001). The correlations between three independent variables, including temperature, extraction time, and sample-to-solvent ratio, and responses including % yield, % cannabinoid content, lightness, and cannabis ethanolic extraction are shown in Table 2. When r is close to 1 or close to −1 means the two parameters are correlated. The analysis further demonstrated the trade-offs between extraction yield, cannabinoid content, and lightness. Specifically, when optimizing for maximum extraction yield, the model predicted an optimal temperature of 60°C, with a desirability value of 1.000. However, this condition led to a dark-colored crude extract and minimal cannabinoid content. In contrast, prioritizing cannabinoid content resulted in an optimal temperature of −40°C, achieving a desirability of 0.892 and the highest lightness value, but at the cost of lower extraction yield. These trade-offs illustrate how competing responses influenced the overall desirability score and support the rationale for selecting a balanced, intermediate condition.

The extraction was scaled up to ensure that the optimized conditions could be effectively applied at a larger volume without compromising the extraction efficiency or the cannabinoid profile. The results from the laboratory-scale and pilot-scale extractions of three different cannabis samples with different levels of THC and CBD contents (samples 2–4) are presented in Table 3. This suggests that the extraction conditions can be effectively applied at both scales, making them suitable for the extraction of CBD- and THC-dominant cannabis. The crude extract from sample 4 was chosen for cannabinoid separation using the CPC technique because it had the highest CBD content.

To find the suitable CPC two-phase system, 9 potential systems were selected from the literature [31‒35] based on polarity of target compounds. The two-phase systems were prepared and mixed with the cannabis extracts. The concentrations of cannabinoids in each phase were analyzed using HPLC to determine the K_D, as shown in Table 4. The data show that THC exhibited a greater K_D than CBD across all two-phase systems, indicating that THC was eluted before CBD in the ascending mode and that THC was retained longer in the descending mode. For two-phase system numbers 1, 2, 8, and 9, THC and CBD had similar K_D values, with a K_DTHC/K_DCBD ratio less than 2, implying that these systems were not suitable for the separation of the target cannabinoids. In addition, individual K_D values of CBD and THC lower than 1 were deemed less suitable, owing to the extended separation times. Based on these findings, two-phase systems 3–7 were selected for the CPC condition study to separate the THC from CBD.

Two-phase systems 3–7 were employed in an ascending-mode CPC device under the conditions described in online supplementary Table S4 for the separation of cannabinoids. The fractions from two-phase system 6 provided satisfactory separation of THC and CBD, making this condition suitable for further optimization (online suppl. Table S5). The combination of hexane, 0.1% formic acid in acetonitrile, and 20 mm ammonium formate at ratio of 10:6.5:3.5% v/v/v created a favorable environment for selectively retaining CBD, while allowing THC to elute more rapidly. This separation dynamic is attributed to the differing polarities and solubility properties of cannabinoids; THC, which is more hydrophobic, interacts more favorably with the nonpolar components of the solvent system, while the higher polarity of CBD allows it to remain in the SP longer. In the descending mode, THC exhibited a stronger interaction with the SP, causing it to elute after CBD. However, this mode led to prolonged retention of both THC and CBD within the SP, extending the separation time and increasing solvent consumption. Thus, the chosen conditions not only facilitate effective separation but also enhance the purity of the final broad-spectrum CBD product.

Experimental two-phase system 6 was tested in triplicate using 0.5 g extract of sample 4 to observe the chromatographic pattern, demonstrating the repeatability of the conditions. To explore scalability, the sample load was increased from 0.5 to 1 g, with subsequent increments of 1 g until a change in the chromatographic pattern was observed. Figure 2 shows that the chromatograms of the 3-, 4-, and 5-gram loads exhibited similar patterns (a–c). However, as the sample load was increased to 6 g (d) and 7 g (e), a noticeable change in the chromatographic pattern occurred. The cannabinoid content in the fractions was analyzed and quantified on a weight-by-weight basis. For sample loads ranging from 3 to 5 g, Δ⁹-THC and Cannabichromene (CBC) constituted the first elution group, followed by THCA and CBD. Δ⁹-THC was completely eluted from the rotor after fraction 40, as shown in Figure 2. When the sample load ranged from 6 g to 7 g, the first highest peak appeared after a longer time. Δ⁹-THC was completely eluted from the rotor after fraction 60, implying prolonged separation time. Therefore, the maximum capacity of this CPC system is 5 g sample load per run.

The experiment was repeated for two additional runs at the same load to confirm the separation results. The data are presented in Table 5. The total yield recovery after combining fractions 1–40 and 41–100 was 4,531.1 ± 38.94 mg, corresponding to a percentage recovery of 89.3 ± 0.21% w/w. Fractions 1–40 were pooled together as THC-rich fractions. Fractions 41–100 and the descending run were pooled together as broad-spectrum CBD. The total CBD yield recovery was 61.74 ± 1.59% w/w calculated relative to the initial CBD content in the cannabis extract. Notably, 73.3 ± 0.37% w/w of the total CBD content was present in the broad-spectrum CBD product. Additionally, the total THC in the broad-spectrum extract was very low, at 0.2 ± 0.00% w/w, showing a 92% reduction from the crude extract. These results are consistent with the overall levels of CBDA and CBD in sample 4. The chromatogram pattern and recovery of cannabinoid content in the fractions are plotted in online supplementary Figure S1.

The substantial presence of CBD in the broad-spectrum CBD indicates the successful isolation and enrichment of this compound. However, scaling up the CPC system beyond the initial 5 g sample load may present challenges that could affect both technical performance and economic feasibility. Variations in flow dynamics can lead to inefficient separation or a decline in product quality, particularly if the increased load exceeds the capacity of the system to maintain optimal partitioning conditions. Previous studies have demonstrated the effectiveness of CPC in isolating cannabinoids, which aligns with our findings [36‒40]. Hazekamp et al. [36] successfully demonstrated the preparative isolation of seven major cannabinoids from C. sativa using CPC, primarily aiming to obtain highly pure cannabinoid isolates for medical applications. However, the study identified several challenges, including decarboxylation stability, contamination, large-scale loading capacity, and the limited availability of certain cannabinoids, all of which require further investigation. Similarly, Maly et al. [37] employed fast CPC (FCPC) to effectively isolate CBD and CBDA while achieving complete removal of Δ⁹-THC and Δ⁹-THCA-A. Despite its efficiency, the authors acknowledged potential limitations in scaling FCPC for industrial applications. Popp et al. [38] further demonstrated the effectiveness of CPC in separating cannabinoids, particularly CBDA and cannabidivarinic acid, achieving high purity and recovery rates. However, significant research gaps persist regarding the scalability and economic feasibility of this technique for large-scale production. Additionally, Ohtsuki et al. [39] provided valuable insights into cannabinoid analysis, highlighting the issue of “residual complexity” in cannabinoid mixtures. Their findings suggest that the presence of multiple structurally diverse cannabinoids complicates analytical processes, necessitating further research to improve the characterization and understanding of cannabinoid mixtures in various cannabis products. Luca et al. [40] evaluated different LLC methods for THC remediation, demonstrating the effectiveness of advanced techniques in producing a THC-free hemp extract. While the study successfully validated THC remediation, the scalability and cost-effectiveness of LLC processes for industrial applications require further assessment. The authors also noted that while Trapping multiple dual mode (tCPC) shows potential, its current implementation may be inefficient for practical THC remediation. Specifically, the extended process time associated with tCPC presents a significant limitation, underscoring the need for further optimization to improve its suitability for large-scale applications.

While each of these studies had distinct objectives and yielded different findings and limitations, several overarching challenges remain. Key issues include the need for pre-fractionation of the extract, limited loading capacity, and modifications to separation conditions to enhance efficiency. Additionally, the selection of an optimal solvent system is inherently dependent on the desired final product. The solvent ratio plays a crucial role in determining the solubility of active compounds, directly impacting both the efficiency and effectiveness of the separation process. Future research should focus on optimizing these parameters to enhance the scalability and economic feasibility of cannabinoid purification techniques.

Although the CPC system in our study was successfully optimized for high sample load per run and high recovery, certain limitations remain, particularly the loss of CBD during THC removal due to overlapping elution. Further refinement of CPC conditions is necessary to enhance separation efficiency and yield. Future research should explore alternative operational parameters and system configurations to optimize CPC performance further. Despite these challenges, the selected CPC system demonstrated efficiency by retaining CBDA in the rotor for a longer duration compared to CBD. A key consideration in our approach is the timing of decarboxylation. While decarboxylation is commonly used to convert acidic cannabinoids into their neutral forms [45], our method intentionally preserves CBDA within the SP for an extended period. This strategy minimizes potential CBD loss that could occur if CBDA were fully converted to CBD before separation. Additionally, this system can be adapted for CBDA separation, further expanding its application. Therefore, post-purification decarboxylation is recommended to maintain the integrity of broad-spectrum CBD while optimizing cannabinoid content. These findings highlight the potential for further refinement of CPC conditions to improve cannabinoid isolation, ensuring a more efficient and scalable purification process. Moreover, this approach bypassed the sand filtration step [36] and successfully eliminated THC in a single CPC run, making it more practical for scaling up. The minimal Δ⁹-THC content in the final product highlights the effectiveness of this separation method for producing broad-spectrum CBD formulations that comply with regulatory standards and meet consumer demands. Additionally, using cold ethanol as the solvent for the initial extraction improved the separation process by isolating cannabinoids while eliminating chlorophyll. This approach not only enhances product quality, but also minimizes the risks associated with nonpolar solvent toxicity, making it a safer option for researchers. Overall, these findings demonstrate the optimal conditions for pilot-scale cannabis extraction and underscore the potential of CPC as an effective technique for cannabinoid processing and production. This approach enables high-yield recovery and selective purification of broad-spectrum CBD, ensuring both efficiency and scalability.

This study presents a significant advancement in the preparation of broad-spectrum CBD by integrating an BBD-RSM optimized ethanolic extraction process with CPC. Key extraction parameters – including temperature, extraction time, and solvent ratio – were optimized to achieve high cannabinoid yield, improved purity, and desirable extract appearance. The optimized method was successfully scaled from laboratory to pilot-scale, confirming its reproducibility and practical feasibility.

CPC effectively removed Δ⁹-THC while retaining CBDA, reducing processing time and solvent use. Although partial CBD loss due to co-elution with THC remains a limitation, this approach offers a scalable, cost-effective, and regulatory-compliant solution for high-purity CBD formulations. The findings hold promise for applications in pharmaceuticals, nutraceuticals, and cosmetics, where standardized, broad-spectrum cannabinoid extracts are increasingly in demand.

We thank the Highland Research and Development Institute (Public Organization, Thailand) for providing cannabis biomass. We also thank Mr. Roy I. Morien of the Naresuan University Graduate School, Thailand for his assistance in editing the grammar, syntax, and general English expressions used in this paper. OpenAI’s ChatGPT (March 2025 version) was used for assistance in improving the clarity and grammar of the manuscript. All content was critically reviewed and approved by the authors.

An ethics statement was not required for this study type since no human or animal subjects or materials were used.

The authors declare that they have no conflicts of interest.

This research study was supported by grants from the Research and Researchers for Industries (RRI) PhD program National Research Council of Thailand (NRCT), NRCT5-RRI63010-P11 R2564A034; a co-contributor supported by GRD Tech Co., Ltd., Thailand; the Agricultural Research Development Agency (ARDA), Naresuan University, and the Center of Excellence for Innovation in Chemistry (PERCH-CIC) from the Ministry of Higher Education, Science, Research, and Innovation, and NSRF via the Program Management Unit for Human Resources and Institutional Development, Research, and Innovation (grant No. B16F640099).

V. Wongumpornpinit and K. Ingkaninan conceived and designed the project; V. Wongumpornpinit, P. Temkitthawon, N. Khumpirapang, S. Paenkaew, T. Saesong, P. Boonnoun, E. Wongwad, N. Waranuch, and K. Ingkaninan planned and conducted the experiments; V. Wongumpornpinit, S. Paenkaew, T. Saesong, and P. Temkitthawon performed the CPC and HPLC data analysis; W. Wongumpornpinit wrote the draft of the manuscript; all authors revised and edited the manuscript; K. Ingkaninan administered the project and is the corresponding author. All the authors have read and agreed to the published version of the manuscript.

For legal reasons, the dataset is not available for public viewing, but can be made available on request for confirmation of statistics if requested by an accredited peer-reviewed committee or government agency. Further inquiries can be directed to the corresponding authors.