Drug release testing of long-acting intrauterine systems

Abstract

Performance evaluation of polydimethylsiloxane (PDMS) based long-acting (e.g. 3–5 years) levonorgestrel (LNG) intrauterine systems (IUSs), such as Mirena^®, is challenging due to their complex formulation, locally-acting feature, and extremely long duration of drug release. To achieve such long-term release, a large amount of drug (up to 52 mg in Mirena^®) must be incorporated as a drug reservoir in the IUS. Consequently, dose dumping or unanticipated changes in the LNG-IUS in vivo release characteristics may give rise to adverse product safety and efficacy. Therefore, it is crucial to understand, and have appropriate control over, the physicochemical properties and in vitro release characteristics of these products. This requires an understanding of the LNG-IUSs drug release mechanism and the development of a sensitive yet robust in vitro release testing method. There have been no previous reports on in vitro drug release and the release mechanism from LNG-IUSs. This is probably a consequence of the extremely slow drug release rate of LNG-IUSs under real-time in-use conditions (e.g., 3–5 years) and therefore it is impractical to obtain complete release profiles (e.g. there is only 60% release in 5 years for Mirena^®). Therefore, the development of appropriate accelerated in vitro release methods is imperative. Following preparation of LNG-IUSs, similar to Mirena^®, real-time release was tested in (0.9% w/v NaCl) media in a water shaker bath at 37 °C for over 2 years. Addition of surfactant (sodium dodecyl sulfate (SDS)), elevation of temperature, addition of organic solvents (ethanol (EtOH), isopropanol (IPA), tert-butanol (TBA) and tetrahydrofuran (THF)) and a combination thereof were utilized as release media to accelerate drug release for LNG-IUSs. Complete drug release was achieved in 32 and 672 days in THF and TBA hydro-organic media, respectively. The release profile in THF was considered too fast as it may result in change of release mechanism, whereas the release profile in TBA was deemed suitable following model fitting. Model fitting was performed to understand the release characteristics as well as the release mechanisms. The release rate in the hydro-alcoholic media was linearly proportional to the swelling ratio of the PDMS in the corresponding organic solvents. Zero-order, first-order and two-phase models were utilized to fit the release profiles obtained under the different release conditions. The data analysis was comparable using the parameters from different models given the high R² values. However, the two-phase model was better in terms of the release mechanism of the LNG-IUSs considering the full drug release profile. The present study will facilitate the process of granting of biowaivers through an in vitro approach, thus reducing the necessity for clinical studies. In addition, it will help reduce the regulatory burden without sacrificing product quality of LNG-IUS products.

1. Introduction

Controlled drug delivery systems can be traced back to the 1950s when Folkman and Long developed a silicone rubber-based membrane diffusion device for the release of anesthetic and cardiovascular drugs [1,2]. Drug delivery systems can be designed to achieve desired release profiles and release characteristics. There are several advantages of controlled release: 1) desired drug plasma and/or tissue concentrations; 2) efficiency in the total amount of drug administered to the patient; 3) less frequent drug administration; 4) easy realization of local drug delivery (e.g., intrauterine devices, ocular insert and transdermal patches); and 5) reduced systemic toxicity. Based on the approaches to achieve controlled release, there are two major types of delivery systems [3] (Fig. 1): reservoir systems and monolithic systems. The most common release profiles of controlled drug delivery follow three kinetic models: zero-order (constant release rate); first-order (exponentially decreasing release rate); and Higuchi (or square root of time) release. Among all controlled release systems, the long-acting levonorgestrel (LNG) intrauterine system (IUS) or intrauterine device (IUD) is one of the most successful controlled release systems, releasing LNG locally in the uterus for 3–5 years in a controlled manner. According to Centers for Disease Control and Prevention, LNG-IUS is one of the most effective (< 1% failure rate) approaches to contraception [4]. In addition to being approved as a birth control, Mirena^® (Bayer) is the first and only IUD that has been FDA-approved to treat heavy menstrual bleeding [5]. Since the approval of Mirena^® in 2000 [6], three additional commercial LNG-IUS products have been approved by the U.S. FDA.

Fig. 1.

Schematic of two major types of primarily diffusion-controlled drug delivery systems: A) reservoir systems; and B) monolithic systems. (Modified from [3]).

All the commercially available LNG-IUS products are formulated using a polydimethylsiloxane (PDMS) matrix base. The macroscopic structure (T-shaped frame device) of the LNG-IUSs (such as Mirena^®) (refer to graphical abstract) is described elsewhere [7,8]. Briefly, the drug device includes a T-shaped polyethylene stem, a hollow cylindrical reservoir composed of crystalline drug particles evenly distributed in a crosslinked PDMS matrix, a PDMS outer membrane for release rate control and a retrieving thread for removal of the device, as necessary. The drug release rate of these LNG-IUS products gradually declines over time. Release rates of LNG-IUSs are listed in Table 1 for comparison. Although the release rates appear relatively constant if comparing month-to-month, the release rates change significantly over the device lifetime (for example, the rate for Mirena^® is reduced by 50% at 5 years and rate for Skyla^® is reduced by 64% at 3 years).

Table 1.

In vivo release rate change of the LNG-IUS products approved by the U.S. FDA. (The unit in the table is μg/day).

Product (total amount of LNG)	< 1 year	At 1 year	End of approved year for use	Average	Reference
Mirena® (52 mg)	20.0 (3 months)	18.0	10.0 (5 years)	17.1a	[7,9,10]
Kyleena® (19.5 mg)	17.5 (24 days)	9.8	7.4 (5 years)	9.0	[11]
Skyla® (13.5mg)	14.0 (24 days)	6.0	5.0 (3 years)	6.0	[12]
Liletta® (52 mg)	19.5 (initial)	17.0	9.8 (5 years)	14.1a	[13,14]

^aThe average release rate was calculated based on the amount of drug released at the end of the approved 5-year usage (60% cumulative release for Mirena^® and 49.4% cumulative release for Liletta^® over a 5-year duration). These data were obtained based on the amount remaining in the device after retrieval from patients at the specified amount of time.

It is impractical to develop an IVIVC for LNG-IUSs due to: 1) they are locally-acting complex drug products and release rate is not directly related to drug plasma levels; and 2) the release duration is extremely long for these products. Consequently, to ensure product quality, it is crucial to understand the physicochemical properties and in vitro release characteristics of such products. Therefore, it is important to develop reliable in vitro release testing methods as well as to understand the release mechanism for the LNG-IUSs. The in vitro release testing and the release mechanism of LNG-IUSs have rarely been reported [7] except in the patent [15] despite the fact that LNG-IUSs have been studied for several decades. This may be largely due to: 1) the difficulties in manufacturing reproducible devices since they require extensive mold design and material optimization; 2) extremely long duration (years) of release testing; and 3) lack of regulatory guidance on in vitro drug release testing methods. There have been a few clinical studies reporting in vivo drug release [14,16,17]. However, this data is incomplete due to sampling difficulties with human subjects and inaccuracies in analysis as a result of the low drug release rate compared to the total drug content. Consequently, the release profiles may not be reflective of the true release profiles due to the limited number of samples collected. The available in vivo release data of these products is for ~ 3 to 5 years, however the total amount of drug released by the end of the in vivo lifetime of the product is ~ 40% to 60%. Accordingly, the release mechanism is not well understood since the available data only covers ~ 40% to 60% of drug release.

Since the LNG-IUS product is designed to control the release of levonorgestrel over a long period (years), a large amount of drug must be incorporated into the reservoir. Consequently, if the drug product quality is not controlled, dose dumping could occur which may pose a great threat to the health of the patients. To ensure the quality of drug products, a reliable and robust in vitro release testing method is necessary [18]. Due to the extremely long period (up to 5 years) of real-time (at 37 °C) drug release of LNG-IUS, the development of accelerated drug release testing methods is critical to speed up product development and to reduce the cost associated with long-term studies, as well as assist the review process through a better understanding of the drug release mechanism. There are several commonly used methods [19–21] to accelerate in vitro drug release of controlled drug delivery systems (e.g., PLGA microspheres): 1) elevation of temperature; 2) addition of surfactants; 3) addition of organic solvents. These methods can be applied to other delivery systems as long as the drug release mechanism remains the same [19].

It has been previously reported that hollow cylindrical LNG-IUSs have been successfully manufactured and characterized using Mirena^® as the reference listed drug (RLD) [6]. The current research investigated in vitro release testing methods for the prepared LNG-IUSs in real-time and accelerated conditions for over 2 years. Conditions such as elevation of temperature, addition of surfactants, addition of organic solvents and combinations thereof were utilized to study accelerated release of the prepared LNG-IUSs. Release data analysis (model fitting) and release mechanism are reported. The impact of the swelling ability of PDMS in different organic solvents, as well as temperature on the accelerated release were also extensively investigated.

2. Materials and methods

2.1. Materials

Levonorgestrel with a particle size of 16 μm was purchased from Tecoland Corporation (Irvine, CA, USA). Liquid silicone rubber (MED-4840 Part A and Part B) was purchased from NuSil^™ (Carpinteria, CA, USA). Sodium chloride and sodium dodecyl sulfate (SDS) were purchased from Sigma-Aldrich (St. Louis, MO, USA). Ethanol (EtOH, HPLC grade), Isopropanol (IPA, HPLC grade), tert-Butanol (TBA, certified), and tetrahydrofuran (THF, HPLC grade) were purchased from Fisher Scientific (Fair Lawn, NJ, USA). The PDMS silicone tubing (Silastic^®) was purchased from Dow Corning (Midland, MI, USA). Unless otherwise specified, all materials were of analytical grade.

2.2. Preparation of levonorgestrel intrauterine systems

2.2.1. Preparation of levonorgestrel intrauterine reservoir

The LNG reservoir was prepared as previously described [6]. Briefly, a customized designed extrusion device composed of one polycarbonate mold, two 3-ml plastic syringes with the needle part removed, a metal rod with a centering device A and a centering device B (Fig. 2) were used. Before preparation, all the devices and items were cleaned with ethanol. The batch size of the LNG-PDMS reservoir was 300 mg. The drug reservoir was prepared using liquid silicone rubber and a certain amount of LNG according to the drug loading required. After blending, the matrix was left under vacuum to remove any air bubbles. The mixture of the polymer and LNG was transferred into the mold to form a tubular shaped drug reservoir and cured at 80 °C for 20 h. The cured reservoir was then peeled off the mold and then the steel rod was removed for further fabrication.

Fig. 2.

Pictorial demonstration of LNG-IUS manufacturing: A) devices used for preparation of LNG-IUS drug reservoirs; B) the processes of extrusion of PDMS polymer and LNG mixture to form the drug reservoirs; C) the resultant LNG drug reservoir with steel rod; D) 100 mg standard LNG reservoir with plastic rod; and E) LNG-IUS (with release controlling membrane) (with permission from [6]).

2.2.2. Preparation of the levonorgestrel intrauterine systems

The LNG-IUSs were prepared as previously described [6]. Briefly, the resultant LNG intrauterine reservoirs were cut into 100 mg pieces and enveloped onto plastic rods. Silastic^® PDMS tubing was used as the rate controlling membranes and was cut into 24 mm-long pieces. The PDMS tubing segments were soaked in hexane at room temperature for 5 min to swell. The swollen tubing was then coated onto the drug reservoirs in a symmetrical way. The coated LNG intrauterine units were placed into a desiccator under vacuum at room temperature overnight to remove the residual hexane solvent. The resultant LNG-IUSs are shown in Fig. 2E.

2.3. HPLC analytical methods

The concentration of LNG was determined using an Agilent 1290 Infinity HPLC system with a UV detector set at 240 nm. The mobile phase was a mixture of acetonitrile and water (60/40, v/v). A C18 column (Kinetex^®, 250 × 4.6 mm, 5 μm, Phenomenex^®) was used with a flow rate at 1.5 ml/min. The column temperature was set at 30 °C and the injection volume was 50 μl. The chromatographs were analyzed using Agilent OpenLAB CDS ChemStation.

2.4. Drug loading and content uniformity of the reservoir

The drug loading of the prepared reservoirs was determined using a THF extraction method as described before [6]. In brief, the drug reservoirs were cut into small pieces (with a weight of approximately 8 mg) and each piece was soaked in a 25-ml volumetric flask with THF. The flasks were kept sealed and left at room temperature for 2 days. An appropriate volume of THF was added to the flasks to achieve a total volume of 25 ml. The samples were vortexed and mixed well and then diluted with the mobile phase (acetonitrile/water = 60/40 (v/v)) prior to being injected into the HPLC for analysis. For each batch, three samples were cut from different regions of the reservoirs and tested for drug loading and content uniformity.

2.5. Swelling of PDMS membranes in different media

Swelling of PDMS membranes in different pure organic solvents (including EtOH, IPA, TBA and THF) as well as water was determined at room temperature. In brief, 2.4 cm-lengths of the PDMS tubing were soaked in the above-mentioned media for several days until swelling reached equilibrium, as determined by three consistent tubing dimensions over a period of 3 weeks. The pieces of tubing were then withdrawn from the media and the surface residual liquid was removed before each measurement. The length of the tubing was measured before and after the swelling test using a caliper (0.1 mm accuracy). All swelling tests in each of the media were performed in triplicate. The swelling ability was calculated using the following equation:

$$\mathit{\text{Swelling}}\mathit{\text{ratio}}=L_{\text{equ }}/L_0$$

Where, L_equ represents the length of the tubing at equilibrium and L₀ represents the initial length of the tubing.

2.6. In vitro drug release testing of LNG-IUSs

In vitro drug release from the LNG-IUSs was tested using a water shaker bath agitated at 100 rpm. Glass bottles (PYREX^®, Corning Inc.) with screw caps (GL45) were used to perform the release tests. All of the release testing of the LNG-IUSs was performed under sink conditions. Unless otherwise specified, the sampling plan for all the release testing was as follows: 1 ml samples were withdrawn on day 1, 2, 3, 4 and 7 in the first week, and replenished with fresh media. Following the day 7 sampling, all the media in the bottles were drained and replenished with fresh media. Thereafter, samples were withdrawn weekly and all the media in the bottles were drained and replenished with fresh media following sampling. All of the experiments were performed in triplicate.

The real-time release testing of LNG-IUSs was performed in 300 ml of 0.9% w/v NaCl at 37 °C. To investigate the impact of surfactants and temperature of the release media on the real-time drug release rate, 0.25% w/w of SDS in pH 7.4 PBS was used for release testing of LNG-IUSs at 37 °C and 45 °C.

Accelerated drug release testing of the prepared LNG-IUSs was performed in hydro-organic media containing 20% v/v of organic solvent (either EtOH, IPA, TBA or THF), 80% v/v of pH 7.4 PBS and 0.25% w/v of SDS. The accelerated drug release of the LNG-IUSs was tested at 45 °C. Different release media volumes were used in different accelerated conditions (50 ml media was used for EtOH hydroalcoholic condition and 100 ml media was used for the rest of the hydro-organic conditions). For THF hydro-organic release condition, the sampling plan was as follows: 1 ml samples were withdrawn at 2, 4, 6, 8 and 24 h every day and fresh media was replenished after sampling; following sampling at 24 h, the media were drained and replenished with fresh media.

In addition, to understand the effect of temperature on the drug release rate in TBA hydroalcoholic release media, release testing of LNG-IUSs was also performed at 37 °C, 55 °C and 65 °C. Due to the increased release rate at higher temperatures, the sampling plan was adjusted to meet sink conditions. For the release testing at 55 °C, 1 ml samples were withdrawn on days (1, 2, 3 and 4) in the first week, and replenished with fresh media. Following the day 7 sampling, all the media in the bottles were drained and replenished with fresh media. Thereafter, samples were withdrawn every subsequent fourth and seventh day, and all the media in the bottles were drained and replenished with fresh media following sampling. For the release testing at 65 °C, 1 ml samples were withdrawn every day and replenished with fresh media and full media was exchanged every 4 days.

2.7. Particle size and morphological studies

The particle size of the API used to prepare the LNG-IUS drug reservoirs was characterized using an Olympus BX51 polarized light microscope (PLM) (Olympus America Inc., New York, USA) as previously described [22]. Briefly, the API powder was dispersed using a drop of mineral oil on glass slides. Cover slips were placed on top of the dispersed samples. Three images were acquired at 200-fold magnification, followed by image analysis using an ImageJ software (National Institutes of Health, Bethesda, MD) to determine the particle size of the API. The particle size of the API was also characterized using an AccuSizer autodiluter particle sizing system (Nicomp, Santa Barbara, CA) by dispersing the powder in 0.5% poly(vinyl alcohol). The morphology of the LNG-IUS drug reservoirs before and after the accelerated drug release (in TBA hydroalcoholic media at 45 °C) was determined using a scanning electron microscope (a FEI Nova NanoSEM 450 unit). Samples were mounted on carbon taped aluminum stubs and sputter coated with gold for 3 min at 6 mA with a deposition rate of 6 nm/min before imaging.

2.8. Statistical analysis

The data analysis, linear regression and fitting were performed using OriginPro 2017 software (OriginLab Corporation, Northampton, MA, USA) and Excel 2016 (Microsoft Corporation, Redmond, WA, USA). All data were presented as average ± standard deviation (SD). Significant differences were accepted when p value < .05.

3. Results

3.1. Drug loading and content uniformity

The drug loading of the LNG-IUS drug reservoirs was determined using the THF swelling and extraction method. The results showed all of the reservoirs were within the target 50% w/w drug loading with < 5% RSD, indicating good content uniformity.

3.2. Swelling ability in organic solvents

The swelling ability of the PDMS membrane in different solvent systems is listed in Table 2. The swelling ability of the PDMS in the solvents is in the following order: THF > TBA > IPA > EtOH > water. The data is consistent with the results from a literature report [23].

Table 2.

Swelling ratio of the PDMS in different solvents (n = 3).

Solvents	Swelling ratio
Water	1.02 ± 0.01
EtOH	1.05 ± 0.01
IPA	1.09 ± 0.01
TBA	1.21 ± 0.01
THF	1.42 ± 0.02

3.3. Real time drug release testing

The in vitro drug release profiles of the IUSs in real-time conditions (0.9% w/v sodium chloride at 37 °C), and accelerated conditions (pH 7.4 PBS with 0.25% w/v SDS at 37 °C, and pH 7.4 PBS with 0.25% w/v SDS at 45 °C) are shown in Fig. 3. It is apparent that the drug release rate can be accelerated by addition of surfactants as well as elevation of temperature. However, the extent of the release rate increase under these accelerated conditions was less than a 3-fold increase which is insufficient to reduce the duration of release testing.

Fig. 3.

In vitro drug release profiles of LNG-IUSs in real-time conditions (0.9% w/v sodium chloride at 37 °C), and accelerated conditions (pH 7.4 PBS with 0.25% w/v SDS at 37 °C, and pH 7.4 PBS with 0.25% w/v SDS at 45 °C) conducted in a water shaker bath agitated at 100 rpm (n = 3).

To further accelerate drug release from the LNG-IUSs, the addition of different organic solvents to the release media at 45 °C was used (Fig. 4.). Organic solvents significantly accelerated drug release compared to the real-time and accelerated (45 °C) drug release in aqueous-based media (Fig. 3). THF hydro-organic media showed the strongest ability to accelerate the drug release with 100% release in approximately 1 month. The drug release rate of the IUSs obtained from the hydro-alcoholic release media (TBA, IPA and EtOH) had the following rank order: TBA > IPA > EtOH. Since the drug release from aqueous-based media is extremely slow, the release rates were compared using zero-order fitting for the first 0 to 7% of drug release. This resulted in release rates of 8.93, 49.56, 80.03, 177.09 and 3117.45 μg/day for the aqueous (real-time), EtOH, IPA, TBA and THF hydro-organic media, respectively. Compared with the real-time release rate (8.93 μg/day, zero-order fitting ranging from 0–7% drug release) in 0.9% w/v NaCl, release rates increased approximately 6-, 9-, 20- and 349-fold in EtOH, IPA, TBA and THF hydro-organic media, respectively. (See Fig. 4.)

Fig. 4.

In vitro drug release profiles of IUSs in accelerated release media containing 20% v/v organic solvents (EtOH, IPA, TBA or THF), 80% v/v of pH 7.4 PBS and 0.25% w/v of SDS. The release testing was conducted at 45 °C in a water shaker bath agitated at 100 rpm (n = 3). (EtOH-SDS-PBS represents the release media of EtOH containing 20% v/v of EtOH, 0.25% w/v SDS and 80% v/v pH 7.4 PBS. The same name rule applied to the other organic solvents.)

3.4. Accelerated drug release testing at different temperatures

Accelerated drug release in TBA media was tested at 37 °C, 45 °C, 55 °C and 65 °C to investigate the impact of temperature on drug release from the LNG-IUSs. The release profiles showed that increasing temperature increased the drug release rate. (see Fig. 5) Release rates were 56.96, 164.76, 391.41 and 1030.10 μg/day (zero-order model fitting ranging from 0 to 20% release) in the TBA aqueous media at 37 °C, 45 °C, 55 °C and 65 °C, respectively.

Fig. 5.

In vitro drug release profiles of IUSs in accelerated release media containing 20% v/v TBA, 80% v/v of pH 7.4 PBS and 0.25% w/v of SDS at 37 °C, 45 °C, 55 °C and 65 °C. The in vitro release testing was conducted in a water shaker bath agitated at 100 rpm (n = 3).

3.5. Particle size and morphology of the drug in LNG-IUSs

API particle size used in the LNG-IUSs is 12.46 ± 2.61 μm (D90, number weighted) and 10.48 ± 3.83 μm (D50, volume weighted) using the polarized light microscopy and Accusizer, respectively. The particle size obtained from both methods was smaller than that reported by the supplier (16 μm). SEM images showed that the drug particle size of LNG-IUS is approximately 5 μm (Fig. 6A), which was significantly smaller than the raw API. After manufacturing, the measured particle size of the API appeared smaller, probably due to the shear force applied to the drug particles during the processing. The SEM image (Fig. 6A) showed that the drug particles were evenly distributed in the PDMS matrix. After completion of the drug release testing from the LNG-IUS under accelerated release conditions (TBA hydroalcoholic media), pores were left behind and the pore size was close to the drug particle size (Fig. 6B, C and D).

Fig. 6.

A) SEM image of the prepared LNG-IUS drug reservoirs before release testing at 5000× magnification; and SEM images of LNG-IUS drug reservoirs after completion of release testing under accelerated conditions (tert-butanol hydroalcoholic media): B) 25,000× magnification; C) 5000× magnification; and D) 500× magnification.

4. Discussion

4.1. Release kinetics modeling of LNG-IUSs

Model fitting of the release profiles is beneficial for two purposes: 1) to allow statistical comparison among different release profiles via the fitting parameters; and 2) to understand underlying release mechanisms of the drug delivery system. This, in turn, provides robust approaches for drug product quality control, better in vitro-in vivo correlation to support bioequivalence evaluation, and insight to design optimum drug products.

For the first purpose of model fitting, simple models (fewer parameters) are preferred to make easy comparison between different release profiles based on comparable R² coefficient of goodness-of-fit. However, the models may be meaningless in the sense of understanding the drug release mechanisms.

4.2. Release percentage range for modeling and data analysis

There has been no previous report on drug release from Mirena^®. According to the Mirena^® package insert [9], the initial release rate of levonorgestrel (LNG) in the commercial Mirena^® drug product is initially 20 μg/day and reduces progressively over its 5 year in vivo use to approximately 10 μg/day. Approximately 60% of the total drug was released over 5 years [7,10]. Creinin et.al [14] investigated the potential duration of action for Liletta^®, a similar PDMS based LNG-IUS (52 mg) to Mirena^® and showed that the in vivo release of Liletta^® followed first-order kinetics. The in vivo drug release rate of Liletta^® decreased with time and was fitted using exponential regression with an exponential constant (k) = 3.75 × 10⁻⁴ per day:

$$\mathit{\text{Remaining}}\mathit{\text{content}}(mg)=52e^{-3.75E-04t\text{ (}day\text{) }}$$

Despite the similarity in release rates of Liletta^® and Mirena^® in the first year (18 μg/day vs. 17 μg/day) and at the end of use (5 years) (10 μg/day vs. 9.8 μg/day) (Table 1), the cumulative drug release at the end of use is different, approximately 49.4% and 60% for Liletta^® and Mirena^®, respectively. Additionally, the release profiles as well as the release mechanisms of these two products appear to be different. From the standpoint of products designed for 5 years, the release range of 0–60% for model fitting may be suitable. However, for the purpose of understanding the release mechanisms of the PDMS based LNG-IUSs, complete release profiles of the LNG-IUSs should be considered.

Different release models (e.g., zero-order, first-order, Higuchi, and a combination thereof such as two-phase fitting) were performed to fit the release profiles of the LNG-IUSs obtained from different release media. Based on the real-time data and part of the accelerated release fitting parameters, both zero-order and first-order showed better goodness of fit (R² value > 0.94). Especially at lower release percentage (< 30%), the drug release showed a perfect linear relationship (R² value > 0.99) with time. However, the goodness of fit is expected to have an increased variance with time for zero-order fitting and therefore the parameters (release rate) need to be compared at the same drug release percentages. The goodness of fit remains comparatively constant for first-order fitting and the rate constant did not show higher fluctuation with time. For example, complete release profiles (> 85% release) of the LNG-IUSs in TBA-aqueous media was fitted to zero and first-order models. The R² values change from 0.9994 at 7% cumulative release to 0.9476 at complete (~97%) release when using the zero-order model. Whereas, the R² values remains at a high level throughout the release profiles (0.9989 at 7.5% release, 0.9928 at full release) using the first-order model. Although both the models showed goodness of fit, the first-order model has a better fit for the prepared LNG-IUSs.

It is difficult to obtain complete release profiles of LNG-IUSs (which is estimated to be over 10 years based on the daily release rate) since the real-time release data was extremely slow (< 6% release in a 1-year period). Drug release rate from the accelerated media showed different extents of increase in release rate. Drug release was complete in 32 days and 672 days in THF and TBA hydro-organic media, respectively. Drug release from EtOH and IPA hydroalcoholic media were significantly slower and not complete by the time of this publication. Cumulative release is approximately 50% (826 days) and 70% (791 days) in the EtOH and IPA hydroalcoholic media, respectively. Since all of the release profiles were at different percentages, zero-order fitting was performed based on the same release percentage to allow fair comparison of the release rate (μg/day) under different release conditions. Additionally, the release rate constant (k₀) as well as the goodness-of-fit (R² values) of zero-order modeling gradually decreased with time and therefore data comparison should be made at the same release percentages. While using first order model fitting, the rate constant (k₁) did not change regardless of the release percentage in all of the release conditions except the THF aqueous media. The rate constant (k₁) varied to a significant extent even with first-order fitting at different cumulative release percentages (611.07E-4 at 7.5% vs 781.00E-4 at complete release, p value < .05). The potential reasons of this phenomena have been discussed in Section 4.2. The model fitting parameters (rate constant, fitting coefficient and T_50%) for the release profiles of LNG-IUSs obtained from different release conditions are listed in Table 3. For data comparison purposes, both zero-order and first-order were suitable. However, the release rates obtained from zero-order modeling were dependent on the selected time range. Zero-order modeling allows better understanding of the daily drug release. On the contrary, release rates obtained from first-order modeling were independent of the time range. The fitting parameters are useful to predict the time required for different percent of drug release (e.g. T_50% is the time required to achieve 50% of drug release) (Table 3).

Table 3.

Model fitting parameters for the release profiles of LNG-IUSs obtained from different release conditions (Q represents cumulative release percentage in the table).

Equation	Zero-order modela		First-order modelb
	Q (%) = k0 t(day)		$\text{Q}(\text{\%})=100\left(1-{\text{e}}_1^{-\text{kt}}\text{ (day) }\right)$
Release conditions	k0 × 102	R2	K1 × 104	R2	T50% (day)
0.9% w/v NaCl-37 °C	1.79 ± 0.03	0.9994 ± 0.0001	1.82 ± 0.03	0.9992 ± 0.0002	3808.50
SDS-PBS-37 °C	3.12 ± 0.03	0.9995 ± 0.0003	3.14 ± 0.06	0.9972 ± 0.0006	2207.48
SDS-PBS-45 °C	4.23 ± 0.05	0.9990 ± 0.0004	4.46 ± 0.07	0.9989 ± 0.0005	1554.14
EtOH-SDS-PBS-45 °C	9.84 ± 0.29	0.9994 ± 0.0000	8.23 ± 0.00	0.9967 ± 0.0004	842.22
IPA-SDS-PBS-45 °C	15.82 ± 0.37	0.9995 ± 0.0003	14.90 ± 0.00	0.9992 ± 0.0002	465.20
TBA-SDS-PBS-45 °C	35.45 ± 0.82	0.9994 ± 0.0003	36.00 ± 0.02	0.9928 ± 0.0019	192.54
THF-SDS-PBS-45 °C	589.15 ± 71.63	0.9995 ± 0.0002	781.00 ± 56.40	0.9746 ± 0.0012	8.88

^aThe fitting was made from 0–7% cumulative release to make a fair comparison among all the release profiles from different release conditions.

^bThe fitting was made up to the release presented in the release profiles.

4.3. Suitable conditions for accelerated release of LNG-IUSs

Based on the first-order model, the T_Q% were obtained to predict the time required for Q% drug release under different release conditions. The representative T_50% (or half-lives of the LNG-IUSs) are listed in Table 3. The predicted time required for different drug release percentage were close to the experimental results. Since the real-time release only reached 12.64% and it followed zero-order release (R² = 0.9989), the accelerated release from all of the accelerated conditions also showed zero-order release kinetics, and therefore it can correlate well to the real-time release profiles. However, compared with the real-time drug release, release profiles obtained from THF hydro-organic media was very fast (T_50%: 3808.05 days vs. 8.88 days) and therefore may not be representative of the real-time release profiles, as some critical release characteristics may be obscured at such a fast release rate. All of the accelerated conditions (addition of surfactants, addition of organic solvents or temperature elevation) may be suitable for quality control purposes.

4.4. Release mechanism of the LNG-IUS

Based on the above analysis, the first-order model resulted in the best fit for the in vitro release profiles of the LNG-IUS, which indicates that the drug release mechanism of LNG-IUS is diffusion controlled. There are two types (Fig. 1A) of drug reservoir systems (constant activity source and non-constant activity source) in the classification system for primarily diffusion-controlled drug delivery systems [3]. If the drug reservoir contains both saturated solution and substantial excess of solid drug particles, and the saturation state at the inner surface of the outer membrane can be maintained by dissolution of the drug particles in the reservoir, the release rate is constant (Eq. (1)), or it has zero-order release kinetics. On the contrary, in the case of a reservoir with an initial drug concentration below the drug solubility, released drug molecules are not replaced and the drug concentration at the inner membrane surface continuously decreases with time. The release rate exponentially decreases with time, following first-order kinetics (Eq. (2)). Both of the Eqs. (1) and (2) were based on the following assumptions: 1) the membrane does not swell or dissolve; 2) perfect sink conditions are maintained throughout the release period; and 3) drug permeability through the membrane remains constant.

$$\frac{\text{d}{M}_t}{\text{dt}}=\frac{A{J}_{lim}}{l}=\frac{ADK{C}_s}{l}$$ (1)

$$\frac{\text{d}{M}_t}{\text{dt}}=\frac{ADK}{l}{C}_t=\frac{ADK}{l}\frac{M_0-M_t}{V}$$ (2)

where M_t represents the cumulative drug released at time t and dM_t/dt denotes the steady state release rate at time t. A is the total surface area of the device (edge effects being ignored) and l the thickness of the rate controlling membrane. J_lim represents the membrane-limiting flux. K represents the partition coefficient of the drug between the membrane and the reservoir and D is the diffusion coefficient of the drug within the membrane. C_s is the drug solubility in the reservoir and C_t is the concentration of the drug in the release medium at time t. M₀ is the initial amount of drug within the dosage form and V is the volume of the drug reservoir.

The structure of LNG-IUS is a combination of a monolithic and a reservoir system (Fig. 7A), rendering a complicated drug release mechanism (Fig. 7B). Initially, the drug diffuses through the outer membrane layer from the monolithic PDMS drug reservoir. As time progresses, the drug is depleted from the surface of the monolithic reservoir and therefore the drug must then diffuse through the monolithic PDMS matrix (Phase 1) and the outer PDMS membrane (Phase 2) sequentially. Accordingly, the release kinetics can be expressed by the following equation [1]:

$$\frac{d{M}_t}{dt}=\frac{A{D}_1{D}_2{C}_{1(s)}}{x{D}_2-K\lambda D_1}$$ (3)

where M_t represents the cumulative drug released at time t and dM_t/dt denotes the steady state release rate at time t. A is the total surface area of the device and λ is the thickness of the membrane. K represents the partition coefficient of the drug between the membrane and the reservoir. D₁ and D₂ is the diffusion coefficient of the drug within Phase 1 and Phase 2, respectively. At time t, a drug depleting layer (Phase 2) of thickness x has formed immediately adjacent to the outer membrane of thickness λ. C_1(s), C₁₍₀₎ and C₂₍₀₎ represent the solubility of levonorgestrel in the matrix phase, and the initial drug concentration at the membrane-matrix interface in Phase 1 and Phase 2, respectively. When xD₂ < < KλD₁, the release rate is characteristic of a simple reservoir system (zero-order release kinetics) and Eq. (3) reduces to Eq. (1). After a sufficient amount of time, the effect of diffusion through the matrix layer becomes more prominent and xD₂ > > KλD₁, where the equation reduces to the Higuchi equation (refer to [1] [24,25] for deduction details) for release from a monolithic matrix system. Based on the two-phase model fitting of the complete release profiles obtained from TBA aqueous release media, zero-order (Fig. 8A) and Higuchi model fitting (Fig. 8B) showed high R² values (> 0.99) in the range of 0–30% release and 30% to complete release, respectively. From the analysis, zero-order (lower release percentage), first-order and two-phase models for LNG-IUS release showed comparable goodness-of-fit. For quality control purposes or other data comparison, the release rate of LNG-IUSs can be obtained mathematically from either model given the R² values are sufficiently high. However, when considering the complete release profiles (from 0 to 100% release), the release of the LNG-IUSs may follow the two-phase mechanism since the drug releases sequentially from the monolithic matrix (Phase 1) and through the outer membrane (Phase 2) as discussed above. Drug release from the commercial IUSs are incomplete by the time of removal, and therefore the zero-order release mechanism may be misleading.

Fig. 7.

A) Depiction of LNG-IUS device structure combining monolithic and reservoir system; and B) Concentration profile in LNG-IUS. At time t, a drug depleting layer (Phase 2) of thickness x has formed immediately adjacent to the outer membrane of thickness λ. The C_1(s), C₁₍₀₎, C₂₍₀₎ and C_t represent the solubility of levonorgestrel in the matrix phase, drug concentration at the membrane-matrix interface in Phase 1, Phase 2, and in the release media, respectively. (The drug reservoir PDMS matrix is designated as Phase 1 and the outer PDMS membrane is Phase 2.)

Fig. 8.

Two-phase model profiles for the LNG-IUS: A) zero-order model ranging from 0–30% release; and B) Higuchi model ranging from 30% to complete release of the LNG-IUSs obtained from TBA aqueous release media in a 45 °C water shaker bath with an agitation speed at 100 rpm.

4.5. Swelling of the PDMS

The drug release rate of LNG-IUS from different solvents and water media system was significantly different (p value < .05). A linear correlation was obtained between the drug release rates and the swelling ratios of PDMS in the different solvents (excluding THF) (Fig. 9). At the time of publication, the drug release from the aqueous media was around 7%. To make a fair comparison, the release rate of LNG-IUS in different release media was obtained using zero-order modeling over the release range of 0–7%. Considering only the release rates from the three hydro-alcoholic release media, there was a linear relationship with the swelling ratios of the PDMS in the corresponding organic solvents over the release range investigated to date (0–50% for the slowest one). These results demonstrated that the swelling ability of the organic solvent greatly impacted the accelerated drug release rate. To confirm the correlation between the swelling ratio and drug release rate, the rate constant (k₁) fit using first-order model was correlated to the swelling ratio of the PDMS in different solvents (Fig. 9B). Release rate from both zero-order and first-order models showed similar correlation with the swelling ratio.

Fig. 9.

Linear regression of release rate of LNG-IUS versus the swelling ratio of PDMS in different solvents: A) daily release (μg) from the LNG-IUS (daily release data was obtained through zero-order release modeling ranging from 0–7% drug release); and B) rate constant (day⁻¹) of LNG-IUSs release profiles using first order modeling.

The release rate of LNG-IUS was extremely high (> 3000 μg initial daily release) in THF release media and it did not follow a linear relationship as had occurred with the hydro-alcoholic release media. This may be attributed to 2-fold higher drug solubility (approximately 100 μg/ml) in the THF media compared with the hydro-alcoholic media (40–50 μg/ml) (Fig. 10). The high release rate may also be due to the higher solvent mobility of THF compared to water and the alcoholic solvents. The boiling points of water, EtOH, IPA and TBA are 100.0 °C, 78.4 °C, 82.50 °C and 82.0 °C, respectively, which are far from the release testing temperature (45 °C). However, the boiling point of THF is 66.0 °C, which is close to the release testing temperature. Therefore, the mobility of THF molecules during the release test is significantly higher than water and alcoholic molecules. The viscosity of THF is the lowest among the organic solvents, and this may also contribute to the higher drug release rate.

Fig. 10.

Solubility of levonorgestrel in different release media at 37 °C (n = 3).

4.6. Impact of temperature on drug release rate

Since the THF aqueous media showed an extremely high increase in drug release rate (over 300-fold), it may not be suitable for accelerated drug release testing for the LNG-IUSs as such high increase in the drug release rate may be a result of change in the drug release mechanism. TBA aqueous media showed a reasonably high increase (approximately 20-fold) compared with the real-time release in 0.9% w/v NaCl at 37 °C and accordingly it was decided to investigate this medium at different temperatures (37 °C, 45 °C, 55 °C and 65 °C). The release rate (either from zero-order or first-order) obtained from the TBA aqueous media under different temperatures followed the Arrhenius equation (Fig. 11):

$$lnk=lnA-\frac{E_a}{R}\frac{1}{T}$$ (4)

where k denotes the release rate of the LNG-IUS and, E_a represents the activation energy. R is gas constant (8.314 J mol⁻¹ K⁻¹)) and T is absolute temperature (in Kelvin). The activation energy E_a was calculated to be 88.29 kJ mol⁻¹ K⁻¹ and 89.21 kJ mol⁻¹ K⁻¹ based on the slope obtained from the Arrhenius plot using zero-order and first-order release rate, respectively. The results were comparable regardless of the release models used. As discussed earlier, drug release from the LNG-IUS is directly related to the product of the drug diffusion coefficient (D) and the drug solubility (C_s) in the system (see Eq. (1)). D is a function of diffusion activation energy (E), R (gas constant (8.314 J mol⁻¹ K⁻¹)) and the absolute temperature (T) [26]. Likewise, the drug solubility (Cs) is a function of the fusion energy (ΔH), R and T (Van’t Hoff equation) [27].

$$k\propto D\times C_s$$ (5)

$$D\propto \text{exp}\left(\frac{-E}{RT}\right)$$ (6)

$$Cs\propto \text{exp}\left(\frac{-\Delta H}{RT}\right)$$ (7)

Fig. 11.

Arrhenius plot for release rate of LNG-IUS release profiles obtained in TBA aqueous media at different temperatures. A) daily release (k₀, μg) from the LNG-IUS (daily release data was obtained through zero-order release modeling ranging from 0–20% drug release); and B) rate constant (k₁, day⁻¹) of LNG-IUSs release profiles using first order model.

Combining Eq. (5), (6) and (7) and taking the natural logarithm of both sides,

$$lnk\propto -\frac{1}{RT}(\text{E}+\Delta H)$$ (8)

The activation E_a is a sum of all forms of energy (primarily diffusion activation energy and fusion energy of the crystalline levonorgestrel) associated with the drug release. According to a previous report [6], ΔH of crystalline levonorgestrel is 139.8 J/g which is 43.68 kJ mol⁻¹ (molecular weight of LNG: 312.45 g/mol). Assuming diffusion and solubility largely contributed to the drug release of LNG-IUS, then the diffusion activation energy is 44.61 kJ mol⁻¹.

5. Conclusions

This is the first report on in vitro drug release method development and discussion on the release mechanism of LNG-IUSs. The time required for in vitro drug release testing of LNG-IUSs can be greatly reduced (e.g., 20-fold in TBA aqueous media) if using appropriate organic solvents, surfactants and elevation of temperature. In addition, accelerated testing may be appropriate for modeling/correlating to facilitate understanding of the expected real-use conditions. Accordingly, the developed accelerated drug release testing method will enable the development and assessment of new and potential generic LNG-IUS products. Drug release rates of the LNG-IUSs under hydro-alcoholic conditions correlated with the PDMS swelling ratio in the corresponding organic solvents. In addition, the drug release rates from accelerated conditions obeyed the Arrhenius law. Therefore, release rate timing for LNG-IUS products can be tuned via the selected organic solvent release media and temperature, whilst still being able to appropriately model and compare results. A two-phase model (zero-order followed by the Higuchi model) is a good model to understand the release mechanism of LNG-IUSs despite that zero-order and first-order models also showed sufficiently high goodness-of-fit. In addition, it was shown that THF significantly accelerates the drug release (> 300-fold) and therefore may not be appropriate for release mechanism studies but may be useful for batch release of LNG-IUS products. This study has provided an in-depth understanding of the drug release mechanism from LNG-IUSs and therefore will facilitate bioequivalence evaluation of the LNG-IUSs through in vitro approaches.

Abbreviations:

LNG

levonorgestrel

IUD

intrauterine device

IUS

intrauterine system

LNG-IUS

levonorgestrel intrauterine system

PDMS

polydimethylsiloxane

RLD

reference listed drug