Structure-Dependent Incorporation of Terpenes into Amphiphilic Poly(2-oxazoline) Micelles

Abstract

Poly(2-oxazoline) (POx) amphiphilic block copolymers are highly effective carriers for poorly water-soluble small molecules, yet the structural features governing drug incorporation remain incompletely understood. Motivated by the unusually high loading of diterpene taxanes in POx micelles, we systematically examined whether shared terpene motifs support micellization with POx using natural compounds with established safety profiles. Representative mono-, di-, and triterpenes - α-pinene, carnosic acid (CA), carnosol (CARN), squalene, lupeol, betulin, and ursolic acid - were evaluated with a well-defined POx triblock copolymer. Only the diterpenes CA and CARN showed efficient solubilization, forming small, stable, and uniform micelles, whereas α-pinene produced polydisperse assemblies and all triterpenes were insoluble. CA exhibited the best performance, forming reproducible 22–29 nm micelles at a 2/10 (w/w) CA/polymer feed ratio with ~17% loading capacity and quantitative efficiency; higher feed ratios generated large, polydisperse aggregates by DLS and TEM. CA-loaded micelles were colloidally stable for at least 48 h at room temperature, although CA chemical stability varied across batches. Lyophilization markedly improved stability, enabling 67–89% recovery upon reconstitution. Experimental solubilization data were further used to benchmark existing computer-aided prediction models, which failed to capture the behavior of several terpene classes, underscoring the need to expand their chemical space and incorporate environmental effects. Overall, a rigid diterpene scaffold combined with ortho-hydroxyl functionality emerges as a key determinant of POx incorporation, identifying CA as a robust and reproducible formulation candidate for further POx-based delivery and pharmaceutical development.

Graphical Abstract

Introduction

The development of universal drug delivery systems capable of overcoming physiological barriers and resisting biochemical degradation remains a central challenge in modern pharmaceutics ^1–2. Such systems must accommodate a wide range of therapeutics, from poorly water-soluble small molecules to biologics, including nucleic acids, proteins, peptides, and antibodies. However, most current existing platforms are constrained by limited drug-loading cargo-loading capacity, immunogenicity, off-target distribution, and dependence on cold-chain infrastructure³. Our previous work⁴ demonstrated that amphiphilic block copolymers from the poly(2-oxazoline) (POx) family can function as highly efficient carriers for diverse small-molecule therapeutics. POx-based micelles loaded with antitumor agents show markedly enhanced tumor accumulation in vivo and favorable pharmacokinetic properties⁵. The safety of several POx micellar formulations, including paclitaxel (PTX)-loaded micelles, has been established in both rodent and non-human primate (NHP) models⁶. Notably, unlike polyethylene glycol (PEG), which can elicit pre-existing anti-PEG antibodies in a subset of patients⁷, POx has not been associated with detectable pre-existing antibody to date.

The objective of this study is to assess the feasibility of using POx micelles for terpene delivery. This direction is motivated by the observation that taxanes, which are diterpenes, exhibit exceptionally high incorporation and solubilization within POx micelles⁸, suggesting that other terpenoid structures may similarly benefit from POx-based encapsulation.

Terpenes constitute a broad class of natural products defined by a repeating isoprene (C₅H₈) structural motif and encompass numerous bioactive compounds with anti-inflammatory, anticancer, antimicrobial, analgesic, and anxiolytic properties⁹. Many terpenes are also classified as Generally Recognized as Safe (GRAS). In this study, we screened and characterized several mono-, di-, and triterpenes for their ability to load POx micelles.

α-Pinene, a monoterpene, was selected based on its strong safety record and GRAS designation by the U.S. Food and Drug Administration (FDA) for use as a synthetic flavoring substance and adjuvant under the Code of Federal Regulations (CFR) (21 CFR 172.515). Structurally, α-pinene contains a single carbon–carbon double bond and no additional functional groups, making it highly hydrophobic and practically insoluble in water¹⁰.

In contrast, pure carnosic acid (CA) is not listed by the FDA as a standalone regulated food additive; however, rosemary extracts standardized to CA content fall under the FDA’s food additive and GRAS framework (e.g., 21 CFR 182.20). CA is a phenolic diterpene featuring hydroxyl groups at C-11 and C-12 and a carboxylic acid at C-20. Despite these functionalities, its predominantly hydrophobic backbone renders it sparingly soluble in water and physiological buffers¹¹. Carnosol (CARN), another diterpene abundant in rosemary extract, shares CA’s core scaffold, including hydroxyl groups at C-11 and C-12, but contains a lactone moiety in the B ring. CARN also forms through oxidative degradation of CA¹².

The cyclic triterpenes evaluated lupeol, ursolic acid, and botulin, are not currently GRAS or listed in the CFR. Nonetheless, their structural features, including hydroxyl and/or carboxylic acid groups combined with strong hydrophobicity¹³, made them compelling candidates for assessment in POx micelles. Squalene, a linear polyunsaturated triterpene, in regulated as a component of biologic products (e.g., vaccines) rather than as an independent food or cosmetic additive. Its six conjugated double bonds (all-E configuration) and pronounced hydrophobicity¹⁴, motivated its inclusion in the panel.

Overall, the goal of this work was to identify structural determinants that enable or enhance micellization with POx by systematically evaluating natural organic compounds with established safety profiles. Among the terpenes examined, only the diterpenes CA and CARN exhibited excellent solubilization in POx micelles, forming small, stable, and uniform particles. The monoterpene α-pinene could also be incorporated but yielded polydisperse micelles, whereas all other terpenes were insoluble. We relate the observed solubilization behavior of these compounds to their structural features and use the experimental results to demonstrate that existing computer-aided solubilization models require further refinement to adequately capture new structural classes. Given the broad pharmaceutical potential of CA, particularly as an antioxidant, we further evaluated the reproducibility of CA micellar formulations across multiple batches and compared their performance with PTX formulations under varying pH and ionic strength. These studies demonstrate robust and reproducible formulation behavior, supporting further development of POx-based CA delivery systems.

Materials

Reagents for micelles formulations and characterization.

Amphiphilic triblock copolymers of methyl oxazoline (MeOx) and butyl oxazoline (BuOx), p(MeOx_x​-b-BuOx_y​-b-MeOx_z) were synthesized as previously described¹⁵ or procured from delAQUA (Supplementary Table S1). CA was purchased from three different vendors (Supplementary Table S2). CARN was purchased from MedChemExpress. Alpha-pinene, Betulin, Lupeol, Ursolic Acid, Squalene were purchased from Sigma Aldrich. HPLC Water and Trifluoracetic acid (TFA) were purchased from Sigma Aldrich. HPLC Acetonitrile was purchased from Avantor Sciences. Sterile water was purchased from Stericare solutions. Ethanol 200 Poof was purchased from Decon Labs Inc.

Reagents for cytotoxicity asessment.

4T1 mouse breast cancer cell line was obtained from frozen and stored in lab aliquot. RPMI medium 1640 was purchased from Gibco. Fetal bovine serum (FBS) was purchased from GeminiBio. Penicilin/Streptomycin solution was purchased from Life Technologies Inc. Cell Counting Kit-8 (CCK-8) was purchased from ApexBio.

Instrumentation.

Mettler Toledo XS105 Dual Range Analytical Balance. Malvern OMNISEC GPC/SEC. Bruker AVANCE NEO 400 MHz Spectrometer 1HNMR. Malvern Zetasizer Nano ZS (Malvern Panalytical Ltd., UK). Thermo Scientific Talos F200X scanning/transmission electron microscope (Thermo Fisher Scientific, CZ). Labconco FreeZone^® Triad^™ Benchtop Freeze Dryer (Labconco, US). Agilent Technologies 1200 High-Pressure Liquid Chromatograph (Agilent, US) equipped with Nucleosil C18 column (4.6 mm × 250 mm, 5 μm) (Macherey-Nagel, Germany). Oakton Ion 700 benchtop pH meter (OAKTON Instruments, US). SpectraMax M5 microplate reader (Molecular Devices, US).

Methods

Small-scale thin film preparation of polymeric micelles for screening.

For small-scale thin-film preparation, stock ethanol solutions of the tested terpenes were prepared at their maximally ethanol-soluble concentrations, determined by serial dilution and visual confirmation of no precipitation (CA, CARN, squalene, and α-pinene at 10 mg/mL; lupeol, betulin, and ursolic acid at 2.5 mg/mL). The calculated volumes of each terpene stock were combined with 125 μL of polymer stock solution in ethanol (20 mg/mL). The mixtures were transferred into 1.5 mL microtubes at the desired polymer-to-terpene ratios, mixed thoroughly, and placed in a 65 °C heating block. A gentle nitrogen stream was applied to accelerate ethanol evaporation. After complete solvent removal (~15–20 min), the resulting thin films were hydrated with saline and incubated at 65 °C for an additional 10 min to facilitate micelle formation.

Large-batch thin-film preparation and lyophilization of polymeric micelles.

200 μL of an ethanol solution containing 2 mg of CA (from different vendors) and 500 μL of an ethanol solution containing 10 mg of P2 polymer (JR12) were combined in a pre-treated 10 mL round-bottom flask. Prior to use, the flask was heated in an oven at 220 °C and allowed to cool to room temperature for 30 min. Thin films were formed using a rotary evaporator with the water bath set to 65 °C, and ethanol was removed under carefully reduced vacuum (gradually lowered to prevent bumping or “boiling”) until complete solvent evaporation was achieved. The dried thin film was then hydrated with 1 ml saline to form a micellar solution. The flask was returned to the 65 °C water bath and rotated to ensure complete removal of the solvent and homogeneous dispersion. The resulting micellar solution was aliquoted into 15 ml glass vials (250 μl per vial), sealed with rubber stoppers, and placed into the freeze-dryer chamber (Labconco FreeZone^® Triad^™ Benchtop). The lyophilization cycle (Supplementary Table S3) was optimized to yield an amorphous solid dispersion (ASD). Upon completion, vials were hermetically sealed and wrapped in aluminum foil to protect CA from light.

Particle size determination.

The size distribution of POx micelles was measured using a Nano-ZS dynamic light-scattering (DLS) instrument. Prior to analysis, each terpene/POx micelle formulation was diluted 1:6 in sterile 0.9% saline. The hydrodynamic (effective) diameter and polydispersity index (PDI) were obtained using cumulant analysis of the intensity-weighted z-average. Measurements were performed on three independent micelle samples, and results are reported as average values.

Transmission electron microscopy (TEM).

The morphology of CA/POx micelles was examined using a Thermo Scientific Talos F200X scanning/transmission electron microscope. A drop of each diluted CA/POx micelle solution (500- or 1000-fold dilution in DI water) was deposited onto a copper grid with a carbon support film and allowed to adsorb for 5 min. Excess liquid was gently wicked off using fine filter paper. A drop of 1% (w/v) uranyl acetate staining solution was then applied, allowed to contact the grid for 1 min prior to TEM imaging (Supplementary Figures S1, S2).

Drug loading characterization.

Drug concentrations in POx micelles were quantified by reverse-phase HPLC using an Agilent 1200 system equipped with a Nucleosil C18 (5 μm, 250 mm× 4.6 mm) column. Samples exhibiting large particle sizes by DLS were first centrifuged for 1 min to remove large aggregates prior to analysis. Each micelle sample was diluted 1:10 in the mobile phase (acetonitrile with 0.01% TFA/water with 0.01% TFA; 60/40, v/v). A 10 μL aliquot of the diluted sample was injected. Chromatographic separation was performed at a flow rate of 1.0 mL/min, column temperature of 40°C, and detection wavelength of 240 nm. The analyte was eluted over a 15-min run time. Quantification of CA was performed using a calibration curve (Supplementary Table S4, Figure S3).

The drug loading capacity ($\text{LC}$) and loading efficiency ($\text{LE}$) were calculated according to Equations (1) and (2):

$$LC=\frac{{\left[Drug\right]}_{solution}}{{\left[Drug\right]}_{solution}+\left[Polymer\right]}\mathrm{*}100\mathrm{\%}$$ (1)

$$LE=\frac{{\left[Drug\right]}_{solution}}{{\left[Drug\right]}_{feed}}\mathrm{*}100\mathrm{\%}$$ (2)

where ${\left[Drug\right]}_{feed}$ and ${\left[Drug\right]}_{solution}$ are the concentrations of the drug initially added to the mixture and quantified in the final micellar solution, respectively, and $\left[Polymer\right]$ is the concentration of the polymer excipient.

pH stability test.

A 0.1 M citrate buffer (pH 3) was prepared using citric acid, trisodium citrate dihydrate, and deionized (DI) water. Acetate buffers (0.1 M, pH 4 and pH 5) were prepared from glacial acetic acid, sodium acetate trihydrate, and DI water. Phosphate buffers (0.1 M, pH 6 and pH 7.4) were prepared using sodium phosphate monobasic and dibasic salts in DI water. A 0.1 M borate buffer (pH 9) was prepared using boric acid and sodium borate in DI water. A 0.1 M carbonate buffer (pH 10) was prepared using sodium carbonate and sodium bicarbonate in DI water. The pH of each buffer was verified using an Oakton Ion 700 benchtop pH meter. CA/POx and PTX/POx ASDs were prepared using a large-batch thin-film method and a modified large-batch thin-film method, in which the thin films were dispersed in DI water rather than saline, aliquoted, and lyophilized. Aliquots (100 μL) of each buffer were then added to the corresponding CA/POx or PTX/POx samples. pH-dependent micelle stability was evaluated at room temperature by monitoring changes in hydrodynamic diameter and PDI using DLS.

In vitro cytotoxicity assays.

In vitro cytotoxicity of micelle-encapsulated CA and CARN was evaluated using a CCK-8 assay in the 4T1 triple negative breast cancer (TNBC) cell line. Cells were seeded in 96-well plates at a density of 10,000 cells per well and allowed to adhere for 24 h prior to treatment. Each set of six wells was treated with a serial dilutions of the compounds prepared in complete culture medium, with a maximum concentration of 250 μg/mL. After 24 h of incubation, 10 μL of CCK-8 reagent was added to each well and the plates were incubated for an additional 2 h at 37 °C. Absorbance was subsequently measured at 450 nm using a microplate reader. Cell viability was calculated relative to untreated control wells. The half-maximal inhibitory concentration (IC₅₀) values for the terpenes were determined using GraphPad Prism (version 10.4.1).

Modeling of terpenes solubilization in POx micelles.

The models developed by Alves et al.¹⁶ were rebuilt using the data available in their supplemental files, including the precalculated SiRMS descriptors¹⁷. We followed the same protocols and reproduced the models using scikit-learn 1.3.2 implemented in Python 3.8.19.¹⁸ The new drug-polymer combinations, to be predicted by our models, were prepared following the same protocols for data curation and descriptor calculation described by Alves et al.¹⁶. All the models, datasets, and predictions are available in a repository (https://github.com/cleber-mf/polymeric_micelle_models).

Search for the nearest neighbor in the modeling set.

To further investigate the limitations of the chemical space covered by the models and examine mismatched predictions, we searched the modeling set for the nearest neighbor of each predicted compound. We calculated each compound’s five nearest neighbors with the bulk Tanimoto similarity function available in RDKit 2024.03.5 (DOI: 10.5281/zenodo.12782092), using Morgan fingerprints (2048 bits, radius 2).

Experimental Results

Diterpenes demonstrated the ability to incorporate into POx micelles.

Using a POx triblock copolymer, p(MeOx_x​-b-BuOx_y​-b-MeOx_z) (Supplementary Table S1), with ethanol as the solvent for stock solutions and saline for thin-film hydration, we evaluated the solubilization of several affordable and readily available pure terpenes in POx micelles. Among the monoterpenes, α-pinene was evaluated; among the diterpenes, carnosic acid (CA) and carnosol (CARN); and among the triterpenes, linear squalene and the cyclic compounds lupeol, betulin, and ursolic acid (Figure 1). Of these compounds, CA and CARN exhibited the most favorable incorporation into POx micelles (Table 1). In contrast, although α-pinene could be solubilized at mass ratios as high as 10/2, it formed highly polydisperse particles. None of the evaluated triterpenes produced uniform or stable micellar formulations under these conditions (Table 1). Following thin-film hydration, lupeol, betulin, and ursolic acid precipitated, whereas squalene formed an emulsion-like system, which phase separated over time.

Figure 1.

Chemical structures of the POx block copolymer and tested terpenes.

Table 1.

DLS characterization of POx (JR12) micelles loaded with different terpenes at 10/2 POx/drug wt. ratio.

Terpene class	Chemical compound	Z average diameter (nm)*	PDI*
Monoterpene (C10)	α–pinene	84.0 ±7.7	0.64 ± 0.03
Diterpenes (C20)	CARN	23 ± 0.1	0.03 ± 0.01
	CA	23.6 ± 2.92***	0.1 ± 0.1***
Triterpenes (C30)	Betulin	1183.0 ± 11.5**	0.28 ± 0.04**
	Lupeol	1004.3 ± 389.9**	0.76 ± 0.04**
	Ursolic Acid	2377.0 ± 602.4**	n/a**
	Squalene	431.8 ± 22.3	0.42 ± 0.03

^*Unless stated differently data are mean ± SD values of n = 3.

^**The particle size exceeded the reliable detection range of the DLS instrument (1–1000 nm).

^***The mean ± SD values were determined using five independent P2/CA (10/2) batches (Table 2).

Building on its distinct formulation behavior, we examined the reproducibility of CA-loaded micelles across multiple batches and benchmarked their performance against PTX under varying acidity and ionic strength. We first optimized loading ratios for CA. Our previous work^{5–6, 8} showed exceptionally high loading for taxanes - also diterpenes - which informed this strategy. At a CA/polymer feed ratio of 2/10 (w/w), we consistently obtained small, uniform, and reproducible micelles with diameters of 22–29 nm, confirmed across two independent polymer batches and three CA batches from different vendors (Table 2). In subsequent experiments unless it is indicated differently the CA V2 and JR12 batch are used. Although CARN also forms highly uniform micelles with POx, we chose to focus on the CA/POx formulation due to their structural similarity and the potential oxidative conversion of CA to CARN¹⁹.

Table 2.

DLS characterization of CA-loaded POx micelles across CA sources and copolymer batches.

POx batch code	CA vendor code	CA/P2 wt. ratio	Z average diameter (nm)*	PDI*
JR12	V1	8/10	690.1 ± 446.3	0.76 ± 0.23
		4/10	292.3 ± 65.7	0.36 ± 0.09
		2/10	22.5 ± 0.7	0.06 ± 0.03
	V2	8/10	156.7 ± 50.0	0.57 ± 0.38
		4/10	163.6 ±14.5	0.52 ± 0.06
		2/10	22.1 ±0.1	0.05 ± 0.01
	V3	8/10	548.6 ± 427.3	0.52 ± 0.38
		4/10	153.5 ± 65.7	0.55 ± 0.18
		2/10	28.8 ± 0.7	0.27 ± 0.01
1021	V1	8/10	181.4 ± 5.1	0.28 ± 0.03
		4/10	288.0 ± 32.6	0.64 ± 0.09
		2/10	22.6 ± 0.5	0.06 ± 0.04
	V2	8/10	504.8 ± 302.0	0.48 ± 0.20
		4/10	169.9 ± 105.1	0.35 ± 0.08
		2/10	22.0 ± 1.0	0.06 ± 0.01

^*Unless stated differently data are mean ± SD values of n = 3.

An optimized HPLC method²⁰ was used to quantify CA concentration and loading in the micelles. At the 2/10 CA/polymer ratio, the loading capacity (LC) was approximately 17%, and the loading efficiency (LE) remained quantitative (Table 3, Supplementary Figure S4). (The relatively high variability in chromatographic peak area likely led to overestimation of LE values, resulting in calculated values exceeding 100%.)

Table 3.

Characteristics of polymeric micelles, loaded with CA.

Feeding ratio CA/POx (g/L)	LC (%)	LE (%)
2/10	17.5	106
4/10	n.a.	9.5
8/10	n.a.	0.8

In contrast, increasing the CA/polymer ratios to 4/10 or 8/10 resulted in a pronounced increase in particle size and polydispersity, accompanied by substantial reductions in LE. Notably, our HPLC procedure included a centrifugation step to remove large aggregates prior to column injection. We therefore concluded that, at higher feed ratios, CA was largely incorporated into the oversized aggregates detected by DLS; however, these aggregates were removed during centrifugation, and thus the associated CA was separated and not detected in the micellar fraction analyzed by HPLC.

The TEM analysis further confirmed the particle morphology and the presence of aggregates in the studied dispersions (Figure 2 and Supplementary Figure S1, S2). At the 2/10 CA/polymer ratio, the micelles appeared as small, relatively uniform and nearly spherical particles with diameters of approximately 20–30 nm, which occasionally assembled into bead-like structures on the grid. These observations are generally consistent with the DLS results (Table 2). Differences between sizes measured by DLS and TEM are expected, as the relationship depends on particle rigidity and interactions with the TEM substrate. Soft particles may spread or interact strongly with the substrate and each other, appearing larger in TEM, while in many cases TEM-measured sizes are smaller than the hydrodynamic diameters obtained by DLS. At higher CA/polymer feed ratios of 4/10 and 8/10, along with individual spherical micelles the micrographs revealed substantially larger, micron-scale aggregates, often exhibiting worm-like structures suggestive of multiple micelles clustered together or fused (Figure 2). These observations were generally consistent with much larger particle size and polydispersity measured by DLS (Table 2). It is noteworthy that TEM imaging was performed ~24 hours after micelle preparation; therefore, storage time may have contributed to the observed morphological changes, especially in aggregates systems. As a result, direct comparisons between DLS and TEM data should be interpreted with caution.

Figure 2.

TEM images of CA-loaded polymeric micelles prepared at different CA/polymer weight ratios: (a) 2/10, (b) 4/10, and (c) 8/10. Scale bar: 500 nm.

Stability of the CA-containing micelles.

To assess the stability of CA-loaded polymeric micelles at 2/10 CA/polymer ratio, particle size distributions were measured by DLS under three conditions: (i) within 1 h after thin-film hydration, (ii) after 48 h of storage of the micellar solution at room temperature, and (iii) after reconstitution of lyophilized micelles obtained by freeze-drying the micellar solution and storing the sealed powder at room temperature. Under all three conditions, the micellar solutions exhibited narrow size distributions and consistently small particle sizes of approximately 24–27 nm across two different CA batches (Figure 3).

Figure 3.

DLS intensity weighted size distributions of P2 POx micelles loaded with CA from different sources at a 2/10 CA/polymer weight ratio. Particle sizes were measured at room temperature under three conditions: (a) within 1 h after thin film hydration with saline to a polymer concentration of 10 mg/mL (large batch preparation); (b) after 2 days of storage of the micellar solution at room temperature (aliquots protected from light by foil wrapping); and (c) within 1 h after reconstitution in DI water of the lyophilized powder, which was produced by freeze drying the micellar solution (a), hermetically sealed, and stored for 1 day at room temperature.

In contrast, HPLC analysis revealed a substantial loss of solubilized CA following storage of the micellar solution, with CA levels decreasing to 25% or 64% of the initial concentration, depending on the CA batch (Table 4, Supplementary Figure S5, S6). This loss, observed after 48 h of storage at room temperature, was accompanied by the appearance of a CARN peak in the CA chromatograms (Supplementary Figure S7), suggesting partial oxidation of CA to CARN during storage in solution. Significantly improved stability of CA was observed when the micellar solutions were first lyophilized, stored at room temperature for 48 h, and then reconstituted. In this case, 67% or 89% of the initial CA content was recovered in the reconstituted solutions depending on the batch used. These results indicate that although CA stability may vary depending on the source of the compound, lyophilizing the formulation and storing it as ASD, with subsequent reconstitution, provides a superior storage strategy for POx micelle formulations containing CA, when properly optimized.

Table 4.

CA concentration (mg/mL) measured in micellar solutions: within 1 hour after thin-film hydration, after 48 hours of storage at room temperature, and after reconstitution of the lyophilized powder stored for 48 hours at room temperature.

CA vendor code	After film hydration	After 48h in solution	After reconstitution
V1	2.12	0.55	1.42
V2	1.75	1.12	1.56

Effect of pH on the size of CA-loaded micelles.

CA is a monocarboxylic acid bearing an additional acidic hydroxyl at the C12 position of the C-ring. With pKₐ values of 3.8 for the carboxyl group, 3.8–4.3 for the C12 phenolic hydroxyl, and 14.1 for the second C11 phenolic hydroxyl, CA exists predominantly in its deprotonated (negatively charged) form at physiological pH but becomes largely protonated under acidic conditions²¹. This pH-dependent ionization can modify CA’s localization within polymeric micelles by altering coulombic interactions and may also influence hydrogen bonding between the hydroxyl group of the protonated carboxylic acid (–COOH) and the C11 hydroxyl of CA and the carbonyl (C=O) groups in the POx backbone. Together, these effects may influence micelle size. To investigate the effect of pH on CA-loaded micelle size, particle size was measured by DLS over a pH range from 3 to 10. pH was adjusted using a series of buffers listed in Supplementary Table S5. As a comparator, POx micelles loaded with PTX, a diterpene lacking ionizable groups within this pH range, were evaluated at the same drug/polymer ratio (2/10).

CA/POx and PTX/POx micelles (terpene/polymer ratio 2/10) were prepared using a large-batch thin-film method. To evaluate the effect of ionic strength, the thin films were hydrated either with distilled water or with saline, aliquoted, and lyophilized. The resulting micelles were reconstituted in the same respective media and analyzed by DLS (Figures 4–5). For both CA and PTX, micelles prepared in distilled water showed higher polydispersity than those prepared in saline. This effect was most pronounced for PTX: PTX/POx micelles formed in distilled water exhibited two distinct size populations (25.9 nm and 203.4 nm, with peak areas of 52.8% and 47.2%), which persisted across the entire pH range, yielding consistently high PDI (Figures 4b and 5a). In contrast, PTX micelles prepared in saline remained small and uniformly sized throughout (Figures 4c,d and 5c).

Figure 4.

Dependence of the (a,c) DLS effective hydrodynamic diameter and (b,d) intensity weighted PDI of CA or PTX loaded polymeric micelles on pH after reconstitution, with formulations prepared using a thin film method and thin film hydrated in (a,b) deionized (DI) water or (c,d) 0.9% NaCl solution at a 2/10 compound/polymer feed ratio (w/w). Linear regression analysis was used to evaluate the relationship between PDI and pH: in DI water (b), CA/POx exhibited Y = 0.048•X – 0.026 (R² = 0.864) and PTX/POx exhibited Y = – 0.002•X + 0.495 (R² = 0.1292); in 0.9% NaCl (d), CA/POx exhibited Y = 0.021•X – 0.005 (R² = 0.512) and PTX/POx exhibited Y = –0.002•X + 0.17 (R² = 0.13).

Figure 5.

DLS intensity weighted size distributions of POx micelles at a 2/10 compound to polymer feed ratio (w/w): (a, b) micelles hydrated with DI H₂O, and (c, d) micelles hydrated with 0.9% NaCl solution. Micelles were loaded with (a, c) PTX and (b, d) CA and analyzed as a function of pH after reconstitution.

For CA, micelles formed without added salt also showed distinct size populations, particularly at neutral and alkaline pH; however, micelle uniformity improved and PDI decreased as pH was lowered (Figures 4b and 5b). Similar to PTX, CA micelles prepared in saline remained small and uniform across the entire pH range, including at pH 3, where CA is expected to be negatively charged (Figures 4c,d and 5d). Therefore, these results show that the presence of NaCl during formulation CA and PTX, into POx micelles enhances micellar stability and improves their resilience to pH variations.

Cytotoxicity of the CA and CARN loaded micelles.

The IC₅₀ values of CA/POx and CARN/POx micelles in TNBC 4T1 cells were 26.6 μM and 33.5 μM, respectively (Supplementary Figure S8), under the conditions described in the Methods. Previously reported IC₅₀ values for free CA are 29.3 μM in MDA-MB-231 (TNBC) cells and 47.5 μM in MCF-7 cells²², while IC₅₀ values of approximately 40 μM have been reported for free CARN in both MDA-MB-231 and MCF-7 cells²³. Overall, the cytotoxicity of the micellar formulations is comparable to that of the corresponding free compounds, consistent with effective drug release from the micelles upon cellular exposure.

Discussion

This study evaluated the ability of an amphiphilic POx block copolymer to form micelles capable of solubilizing mono-, di-, and triterpenes. Among all compounds tested, the most stable and uniform micelles were obtained with diterpenes, including CA and CARN, as well as with the taxane PTX, as previously reported¹⁵. Limited micellar solubilization was observed for the monoterpene α-pinene. In contrast, no micellar incorporation was detected for the triterpenes botulin, lupeol, ursolic acid, or squalene.

It has been noted previously that the solubilization of chemical compounds in polymeric micelles formed by POx cannot be rationalized by single molecular descriptors such as molecular mass, logP, or aqueous solubility alone^{16, 24}. Consistent with these observations, neither molecular mass nor logP of the terpenes examined in this study correlates with their solubilization efficiency in POx micelles (Supplementary Table S6). At the same time, the three diterpene compounds that were solubilized efficiently share several structural features. Specifically, they all contain (i) at least one aromatic ring, (ii) hydrogen bond donor functionality, (iii) hydrogen bond acceptors, and (iv) a varying number of rotatable bonds. In contrast, compounds exhibiting low or no solubilization in POx micelles lack one or more of these features. Most notably, α-pinene and squalene lack hydrogen bond donors and acceptors, while all poorly solubilized compounds are devoid of aromatic rings.

We propose that the cyclized isoprene framework, the presence of multiple hydroxyl groups, and aromaticity are key structural features that facilitate incorporation of the studied diterpenes into POx micelles. Specifically, the ortho-phenolic hydroxyl groups at the C11 and C12 positions in CA and CARN likely enable strong hydrogen-bonding interactions with the carbonyl (C=O) groups of the POx backbone. In addition, the aromatic rings present in these molecules serve as particularly effective interaction partners for the amide groups of POx. Their planar, highly polarizable π-electron systems, combined with strong quadrupole moments, enable favorable directional electrostatic matching, weak but persistent hydrogen bonding, significant dispersion contributions, and cooperative binding. It should also be noted that, in addition to PTX, aromatic structural motifs and hydrogen-bond-donating functionalities are present in multiple taxanes that have previously been shown to solubilize efficiently in POx micelles²⁵.

To further substantiate our hypothesis, we analyzed the structural features of thirteen additional compounds that were previously reported to incorporate efficiently into POx micelles²⁶. All of these compounds possess hydrogen-bond-donating functionalities, and all but one – rapamycin - contain aromatic rings (Supplementary Table S7). Rapamycin represents a notable exception. While lacking aromatic moieties, it features a large 29-membered macrolide ring containing a conjugated triene system, together with multiple hydroxyl, methoxy, and ketone groups. This combination of structural features facilitates extensive intermolecular interactions, collectively contributing to its exceptionally high binding affinity for protein receptors^{27, 28} and potentially to the interactions with POx micelles.

These structure-based considerations for solubilized molecules are noteworthy and warrant further evaluation using a broader set of compounds to determine whether they could support intuitive, chemistry-based predictions of drug solubilization in POx micelles. However, given the complexity of chemical space and the large number of potential drug-polymer interactions within micelles, a more comprehensive approach to predicting drug solubilization is required.

Previously, a computer-aided strategy for predicting solubilization of poorly soluble drugs in polymeric micelles was developed by Alves et al.¹⁶ and later updated by Kehrein et al.²⁴. These models predict drug loading efficiency (LE) and loading capacity (LC) using novel descriptors of drug-polymer combinations. We applied the Kehrein et al.²⁴ model, which incorporates the original Alves et al. dataset but extends it to a broader range of POx polymers and drug combinations. Using the publicly available online tool (https://kehrein.xyz/poxload/), solubilization parameters were estimated for the terpenes studied here (Supplementary Table S7). Paclitaxel was included in the training set of all models and is therefore not discussed further.

In contrast, for the newly evaluated terpenes, the model performed poorly, predicting identical LE values of 30% and similar LC values ranging from 5–6% for CA and CARN to 9–11% for α-pinene and triterpenes, despite the latter being experimentally insoluble in these micelles. Moreover, the model predicted complete solubilization of squalene (LE = 100%, LC = 29%), which is inconsistent with the experimental observations.

Therefore, we reverted to the original model developed by Alves et al.¹⁶, whose database is more narrowly weighted toward p(MeOx_x​-b-BuOx_y​-b-MeOx_z)-type copolymers similar to those used in this study. However, the predictions of this model were also unsatisfactory (Supplementary Table S7). For all terpenes, it yielded inconclusive predictions (LE < 80%). Although the model appeared to more reasonably rank the LC values for CA, CARN, and α-pinene, it also predicted appreciable solubilization of triterpenes, in contradiction to the experimental results. Specifically, betulin, lupeol, ursolic acid, and squalene were all predicted to be soluble by the models, whereas experimentally none of these compounds were solubilized in the micelles. These triterpenes are likely outside the chemical space represented in the training datasets, which may explain the observed mismatch between predictions and experimental outcomes.

To further investigate this divergence, we examined the nearest neighbors of each triterpene within the modeling set. As expected, ergocalciferol - the nearest neighbor to betulin and lupeol - exhibited low Tanimoto coefficients of 0.14 and 0.16, respectively, indicating substantial structural dissimilarity. Consistent with prior experimental reports¹⁶, ergocalciferol is soluble in polymeric micelles across a range of drug/polymer ratios (LE = 18–89%; LC = 8–31%) and lacks the steroid-type tetracyclic backbone present in betulin and lupeol, underscoring the absence of representative structures containing this core ring system in the training dataset.

Bexarotene was identified as the nearest neighbor of ursolic acid (Tanimoto coefficient = 0.14) and similarly lacks the triterpene core ring system. Notably, bexarotene exhibits substantially lower micellar solubility than ergocalciferol (LE = 3–4%; LC = 1–2%)¹⁶ and contains a cyclohexyl moiety that only loosely resembles elements of the tetracyclic backbone found in the insoluble triterpenes. Simplex-driven searches of bounded and non-bounded QSAR fragments ^{29, 30} did not identify any statistically significant molecular fragments differentiating these outliers.

Overall, the Alves et al.¹⁶ model developed seven years ago shows partial agreement with experimental results, capturing LC trends with moderate accuracy for three out of seven evaluated compounds (excluding PTX, which was included in the training sets of all models). However, the pronounced discrepancies observed for the triterpenes and poor prediction of LE for all compounds highlight clear opportunities for model improvement and underscore the continued importance of experimental validation to enhance predictive accuracy and broaden the chemical space addressed by future models.

While computational models require further development to provide a robust framework for preliminary screening of drug solubilization in block copolymer micelles, current approaches do not account for the environmental conditions present in experimental systems. Therefore, for terpenes that exhibited good solubilization, such as PTX and CA (which contains ionizable groups) we investigated the effects of pH and ionic strength on micelle properties.

Notably, pH had only a minimal effect: the polydispersity of CA loaded micelles showed a modest decrease with decreasing pH. In contrast, micelle size distribution for both CA and PTX was strongly influenced by ionic strength. Hydration in water produced polydisperse micelles, whereas hydration in 0.9% NaCl yielded micelles with a narrow size distribution. These findings suggest that Na⁺ and Cl⁻ ions at isotonic concentration stabilize drug loaded micelles, potentially by altering the local hydrogen bonding network of water³¹, screening electrostatic repulsion between charged molecules (such as ionized CA), and promoting hydrophobic association of the polymer side chains and the loaded drug via salt induced dehydration (“salting out” effects). It is well established that salts such as NaCl can affect micelle size, stability, and micellization behavior of poloxamers via salting-out effects, including polymer dehydration—an effect analogous to increasing temperature³². However, comparable salt effects in POx-based systems have received far less attention. In our view, systematic studies of ionic strength effects on POx micelles are warranted, both to optimize POx-based drug formulations and to advance fundamental understanding of this important class of block copolymers.

A final important aspect of this study concerns CA, a terpene with broad potential across multiple pharmaceutical contexts. The well-defined physicochemical properties of CA/POx micelles, together with the enhanced stability of CA in its ASD form, provide a strong platform for further investigation of its antioxidant and free-radical–scavenging activities in aqueous media. We demonstrate that following solubilization in POx micelles, both CA and CARN retain cytotoxic activity in TNBC cells. While robust antioxidant nanoplatforms based on gel systems have been reported previously³³, incorporation of CA/POx micelles into gel matrices could further enhance activity retention and enable sustained long-term performance. These properties may also facilitate studies of CA-mediated modulation of oxidative signaling and mitigation of oxidative damage through metal-ion chelation.

Improved aqueous solubility of CA afforded by POx micelles additionally enables advanced in vivo investigations of CA-mediated inhibition of globus pallidus externus parvalbumin-positive (GPe-PV) neurons and its effects on cocaine and opioid reward, sensitization, and self-administration³⁴. Beyond therapeutic applications, CA/POx formulations may also be useful as protective coatings for biomedical devices or biomaterials, where the antioxidant activity of CA could mitigate oxidative stress and improve biocompatibility³⁵. Finally, the observed tendency of CA to form aggregated structures with POx - assemblies that resist hydration while preventing the release of precipitated free CA - may be exploited for the development of pH-responsive, co-loaded drug delivery systems.